11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * linux/fs/namei.c 31da177e4SLinus Torvalds * 41da177e4SLinus Torvalds * Copyright (C) 1991, 1992 Linus Torvalds 51da177e4SLinus Torvalds */ 61da177e4SLinus Torvalds 71da177e4SLinus Torvalds /* 81da177e4SLinus Torvalds * Some corrections by tytso. 91da177e4SLinus Torvalds */ 101da177e4SLinus Torvalds 111da177e4SLinus Torvalds /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname 121da177e4SLinus Torvalds * lookup logic. 131da177e4SLinus Torvalds */ 141da177e4SLinus Torvalds /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture. 151da177e4SLinus Torvalds */ 161da177e4SLinus Torvalds 171da177e4SLinus Torvalds #include <linux/init.h> 18630d9c47SPaul Gortmaker #include <linux/export.h> 1944696908SDavid S. Miller #include <linux/kernel.h> 201da177e4SLinus Torvalds #include <linux/slab.h> 211da177e4SLinus Torvalds #include <linux/fs.h> 221da177e4SLinus Torvalds #include <linux/namei.h> 231da177e4SLinus Torvalds #include <linux/pagemap.h> 240eeca283SRobert Love #include <linux/fsnotify.h> 251da177e4SLinus Torvalds #include <linux/personality.h> 261da177e4SLinus Torvalds #include <linux/security.h> 276146f0d5SMimi Zohar #include <linux/ima.h> 281da177e4SLinus Torvalds #include <linux/syscalls.h> 291da177e4SLinus Torvalds #include <linux/mount.h> 301da177e4SLinus Torvalds #include <linux/audit.h> 3116f7e0feSRandy Dunlap #include <linux/capability.h> 32834f2a4aSTrond Myklebust #include <linux/file.h> 335590ff0dSUlrich Drepper #include <linux/fcntl.h> 3408ce5f16SSerge E. Hallyn #include <linux/device_cgroup.h> 355ad4e53bSAl Viro #include <linux/fs_struct.h> 36e77819e5SLinus Torvalds #include <linux/posix_acl.h> 3799d263d4SLinus Torvalds #include <linux/hash.h> 381da177e4SLinus Torvalds #include <asm/uaccess.h> 391da177e4SLinus Torvalds 40e81e3f4dSEric Paris #include "internal.h" 41c7105365SAl Viro #include "mount.h" 42e81e3f4dSEric Paris 431da177e4SLinus Torvalds /* [Feb-1997 T. Schoebel-Theuer] 441da177e4SLinus Torvalds * Fundamental changes in the pathname lookup mechanisms (namei) 451da177e4SLinus Torvalds * were necessary because of omirr. The reason is that omirr needs 461da177e4SLinus Torvalds * to know the _real_ pathname, not the user-supplied one, in case 471da177e4SLinus Torvalds * of symlinks (and also when transname replacements occur). 481da177e4SLinus Torvalds * 491da177e4SLinus Torvalds * The new code replaces the old recursive symlink resolution with 501da177e4SLinus Torvalds * an iterative one (in case of non-nested symlink chains). It does 511da177e4SLinus Torvalds * this with calls to <fs>_follow_link(). 521da177e4SLinus Torvalds * As a side effect, dir_namei(), _namei() and follow_link() are now 531da177e4SLinus Torvalds * replaced with a single function lookup_dentry() that can handle all 541da177e4SLinus Torvalds * the special cases of the former code. 551da177e4SLinus Torvalds * 561da177e4SLinus Torvalds * With the new dcache, the pathname is stored at each inode, at least as 571da177e4SLinus Torvalds * long as the refcount of the inode is positive. As a side effect, the 581da177e4SLinus Torvalds * size of the dcache depends on the inode cache and thus is dynamic. 591da177e4SLinus Torvalds * 601da177e4SLinus Torvalds * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink 611da177e4SLinus Torvalds * resolution to correspond with current state of the code. 621da177e4SLinus Torvalds * 631da177e4SLinus Torvalds * Note that the symlink resolution is not *completely* iterative. 641da177e4SLinus Torvalds * There is still a significant amount of tail- and mid- recursion in 651da177e4SLinus Torvalds * the algorithm. Also, note that <fs>_readlink() is not used in 661da177e4SLinus Torvalds * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink() 671da177e4SLinus Torvalds * may return different results than <fs>_follow_link(). Many virtual 681da177e4SLinus Torvalds * filesystems (including /proc) exhibit this behavior. 691da177e4SLinus Torvalds */ 701da177e4SLinus Torvalds 711da177e4SLinus Torvalds /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation: 721da177e4SLinus Torvalds * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL 731da177e4SLinus Torvalds * and the name already exists in form of a symlink, try to create the new 741da177e4SLinus Torvalds * name indicated by the symlink. The old code always complained that the 751da177e4SLinus Torvalds * name already exists, due to not following the symlink even if its target 761da177e4SLinus Torvalds * is nonexistent. The new semantics affects also mknod() and link() when 7725985edcSLucas De Marchi * the name is a symlink pointing to a non-existent name. 781da177e4SLinus Torvalds * 791da177e4SLinus Torvalds * I don't know which semantics is the right one, since I have no access 801da177e4SLinus Torvalds * to standards. But I found by trial that HP-UX 9.0 has the full "new" 811da177e4SLinus Torvalds * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the 821da177e4SLinus Torvalds * "old" one. Personally, I think the new semantics is much more logical. 831da177e4SLinus Torvalds * Note that "ln old new" where "new" is a symlink pointing to a non-existing 841da177e4SLinus Torvalds * file does succeed in both HP-UX and SunOs, but not in Solaris 851da177e4SLinus Torvalds * and in the old Linux semantics. 861da177e4SLinus Torvalds */ 871da177e4SLinus Torvalds 881da177e4SLinus Torvalds /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink 891da177e4SLinus Torvalds * semantics. See the comments in "open_namei" and "do_link" below. 901da177e4SLinus Torvalds * 911da177e4SLinus Torvalds * [10-Sep-98 Alan Modra] Another symlink change. 921da177e4SLinus Torvalds */ 931da177e4SLinus Torvalds 941da177e4SLinus Torvalds /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks: 951da177e4SLinus Torvalds * inside the path - always follow. 961da177e4SLinus Torvalds * in the last component in creation/removal/renaming - never follow. 971da177e4SLinus Torvalds * if LOOKUP_FOLLOW passed - follow. 981da177e4SLinus Torvalds * if the pathname has trailing slashes - follow. 991da177e4SLinus Torvalds * otherwise - don't follow. 1001da177e4SLinus Torvalds * (applied in that order). 1011da177e4SLinus Torvalds * 1021da177e4SLinus Torvalds * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT 1031da177e4SLinus Torvalds * restored for 2.4. This is the last surviving part of old 4.2BSD bug. 1041da177e4SLinus Torvalds * During the 2.4 we need to fix the userland stuff depending on it - 1051da177e4SLinus Torvalds * hopefully we will be able to get rid of that wart in 2.5. So far only 1061da177e4SLinus Torvalds * XEmacs seems to be relying on it... 1071da177e4SLinus Torvalds */ 1081da177e4SLinus Torvalds /* 1091da177e4SLinus Torvalds * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland) 110a11f3a05SArjan van de Ven * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives 1111da177e4SLinus Torvalds * any extra contention... 1121da177e4SLinus Torvalds */ 1131da177e4SLinus Torvalds 1141da177e4SLinus Torvalds /* In order to reduce some races, while at the same time doing additional 1151da177e4SLinus Torvalds * checking and hopefully speeding things up, we copy filenames to the 1161da177e4SLinus Torvalds * kernel data space before using them.. 1171da177e4SLinus Torvalds * 1181da177e4SLinus Torvalds * POSIX.1 2.4: an empty pathname is invalid (ENOENT). 1191da177e4SLinus Torvalds * PATH_MAX includes the nul terminator --RR. 1201da177e4SLinus Torvalds */ 1217950e385SJeff Layton 122fd2f7cb5SAl Viro #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname)) 12391a27b2aSJeff Layton 12451f39a1fSDavid Drysdale struct filename * 12591a27b2aSJeff Layton getname_flags(const char __user *filename, int flags, int *empty) 12691a27b2aSJeff Layton { 12794b5d262SAl Viro struct filename *result; 1287950e385SJeff Layton char *kname; 12994b5d262SAl Viro int len; 1301da177e4SLinus Torvalds 1317ac86265SJeff Layton result = audit_reusename(filename); 1327ac86265SJeff Layton if (result) 1337ac86265SJeff Layton return result; 1347ac86265SJeff Layton 1357950e385SJeff Layton result = __getname(); 1363f9f0aa6SLinus Torvalds if (unlikely(!result)) 1374043cde8SEric Paris return ERR_PTR(-ENOMEM); 1381da177e4SLinus Torvalds 1397950e385SJeff Layton /* 1407950e385SJeff Layton * First, try to embed the struct filename inside the names_cache 1417950e385SJeff Layton * allocation 1427950e385SJeff Layton */ 143fd2f7cb5SAl Viro kname = (char *)result->iname; 14491a27b2aSJeff Layton result->name = kname; 1457950e385SJeff Layton 14694b5d262SAl Viro len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX); 14791a27b2aSJeff Layton if (unlikely(len < 0)) { 14894b5d262SAl Viro __putname(result); 14994b5d262SAl Viro return ERR_PTR(len); 15091a27b2aSJeff Layton } 1513f9f0aa6SLinus Torvalds 1527950e385SJeff Layton /* 1537950e385SJeff Layton * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a 1547950e385SJeff Layton * separate struct filename so we can dedicate the entire 1557950e385SJeff Layton * names_cache allocation for the pathname, and re-do the copy from 1567950e385SJeff Layton * userland. 1577950e385SJeff Layton */ 15894b5d262SAl Viro if (unlikely(len == EMBEDDED_NAME_MAX)) { 159fd2f7cb5SAl Viro const size_t size = offsetof(struct filename, iname[1]); 1607950e385SJeff Layton kname = (char *)result; 1617950e385SJeff Layton 162fd2f7cb5SAl Viro /* 163fd2f7cb5SAl Viro * size is chosen that way we to guarantee that 164fd2f7cb5SAl Viro * result->iname[0] is within the same object and that 165fd2f7cb5SAl Viro * kname can't be equal to result->iname, no matter what. 166fd2f7cb5SAl Viro */ 167fd2f7cb5SAl Viro result = kzalloc(size, GFP_KERNEL); 16894b5d262SAl Viro if (unlikely(!result)) { 16994b5d262SAl Viro __putname(kname); 17094b5d262SAl Viro return ERR_PTR(-ENOMEM); 1717950e385SJeff Layton } 1727950e385SJeff Layton result->name = kname; 17394b5d262SAl Viro len = strncpy_from_user(kname, filename, PATH_MAX); 17494b5d262SAl Viro if (unlikely(len < 0)) { 17594b5d262SAl Viro __putname(kname); 17694b5d262SAl Viro kfree(result); 17794b5d262SAl Viro return ERR_PTR(len); 17894b5d262SAl Viro } 17994b5d262SAl Viro if (unlikely(len == PATH_MAX)) { 18094b5d262SAl Viro __putname(kname); 18194b5d262SAl Viro kfree(result); 18294b5d262SAl Viro return ERR_PTR(-ENAMETOOLONG); 18394b5d262SAl Viro } 1847950e385SJeff Layton } 1857950e385SJeff Layton 18694b5d262SAl Viro result->refcnt = 1; 1873f9f0aa6SLinus Torvalds /* The empty path is special. */ 1883f9f0aa6SLinus Torvalds if (unlikely(!len)) { 1893f9f0aa6SLinus Torvalds if (empty) 1901fa1e7f6SAndy Whitcroft *empty = 1; 19194b5d262SAl Viro if (!(flags & LOOKUP_EMPTY)) { 19294b5d262SAl Viro putname(result); 19394b5d262SAl Viro return ERR_PTR(-ENOENT); 1941da177e4SLinus Torvalds } 19594b5d262SAl Viro } 1967950e385SJeff Layton 1977950e385SJeff Layton result->uptr = filename; 198c4ad8f98SLinus Torvalds result->aname = NULL; 1991da177e4SLinus Torvalds audit_getname(result); 2001da177e4SLinus Torvalds return result; 2013f9f0aa6SLinus Torvalds } 2023f9f0aa6SLinus Torvalds 20391a27b2aSJeff Layton struct filename * 20491a27b2aSJeff Layton getname(const char __user * filename) 205f52e0c11SAl Viro { 206f7493e5dSLinus Torvalds return getname_flags(filename, 0, NULL); 207f52e0c11SAl Viro } 208f52e0c11SAl Viro 209c4ad8f98SLinus Torvalds struct filename * 210c4ad8f98SLinus Torvalds getname_kernel(const char * filename) 211c4ad8f98SLinus Torvalds { 212c4ad8f98SLinus Torvalds struct filename *result; 21308518549SPaul Moore int len = strlen(filename) + 1; 214c4ad8f98SLinus Torvalds 215c4ad8f98SLinus Torvalds result = __getname(); 216c4ad8f98SLinus Torvalds if (unlikely(!result)) 217c4ad8f98SLinus Torvalds return ERR_PTR(-ENOMEM); 218c4ad8f98SLinus Torvalds 21908518549SPaul Moore if (len <= EMBEDDED_NAME_MAX) { 220fd2f7cb5SAl Viro result->name = (char *)result->iname; 22108518549SPaul Moore } else if (len <= PATH_MAX) { 22208518549SPaul Moore struct filename *tmp; 22308518549SPaul Moore 22408518549SPaul Moore tmp = kmalloc(sizeof(*tmp), GFP_KERNEL); 22508518549SPaul Moore if (unlikely(!tmp)) { 22608518549SPaul Moore __putname(result); 22708518549SPaul Moore return ERR_PTR(-ENOMEM); 22808518549SPaul Moore } 22908518549SPaul Moore tmp->name = (char *)result; 23008518549SPaul Moore result = tmp; 23108518549SPaul Moore } else { 23208518549SPaul Moore __putname(result); 23308518549SPaul Moore return ERR_PTR(-ENAMETOOLONG); 23408518549SPaul Moore } 23508518549SPaul Moore memcpy((char *)result->name, filename, len); 236c4ad8f98SLinus Torvalds result->uptr = NULL; 237c4ad8f98SLinus Torvalds result->aname = NULL; 23855422d0bSPaul Moore result->refcnt = 1; 239fd3522fdSPaul Moore audit_getname(result); 240c4ad8f98SLinus Torvalds 241c4ad8f98SLinus Torvalds return result; 242c4ad8f98SLinus Torvalds } 243c4ad8f98SLinus Torvalds 24491a27b2aSJeff Layton void putname(struct filename *name) 2451da177e4SLinus Torvalds { 24655422d0bSPaul Moore BUG_ON(name->refcnt <= 0); 24755422d0bSPaul Moore 24855422d0bSPaul Moore if (--name->refcnt > 0) 24955422d0bSPaul Moore return; 25055422d0bSPaul Moore 251fd2f7cb5SAl Viro if (name->name != name->iname) { 25255422d0bSPaul Moore __putname(name->name); 25355422d0bSPaul Moore kfree(name); 25455422d0bSPaul Moore } else 25555422d0bSPaul Moore __putname(name); 2561da177e4SLinus Torvalds } 2571da177e4SLinus Torvalds 258e77819e5SLinus Torvalds static int check_acl(struct inode *inode, int mask) 259e77819e5SLinus Torvalds { 26084635d68SLinus Torvalds #ifdef CONFIG_FS_POSIX_ACL 261e77819e5SLinus Torvalds struct posix_acl *acl; 262e77819e5SLinus Torvalds 263e77819e5SLinus Torvalds if (mask & MAY_NOT_BLOCK) { 2643567866bSAl Viro acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS); 2653567866bSAl Viro if (!acl) 266e77819e5SLinus Torvalds return -EAGAIN; 2673567866bSAl Viro /* no ->get_acl() calls in RCU mode... */ 2683567866bSAl Viro if (acl == ACL_NOT_CACHED) 269e77819e5SLinus Torvalds return -ECHILD; 270206b1d09SAri Savolainen return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK); 271e77819e5SLinus Torvalds } 272e77819e5SLinus Torvalds 2732982baa2SChristoph Hellwig acl = get_acl(inode, ACL_TYPE_ACCESS); 2744e34e719SChristoph Hellwig if (IS_ERR(acl)) 2754e34e719SChristoph Hellwig return PTR_ERR(acl); 276e77819e5SLinus Torvalds if (acl) { 277e77819e5SLinus Torvalds int error = posix_acl_permission(inode, acl, mask); 278e77819e5SLinus Torvalds posix_acl_release(acl); 279e77819e5SLinus Torvalds return error; 280e77819e5SLinus Torvalds } 28184635d68SLinus Torvalds #endif 282e77819e5SLinus Torvalds 283e77819e5SLinus Torvalds return -EAGAIN; 284e77819e5SLinus Torvalds } 285e77819e5SLinus Torvalds 2865909ccaaSLinus Torvalds /* 287948409c7SAndreas Gruenbacher * This does the basic permission checking 2885909ccaaSLinus Torvalds */ 2897e40145eSAl Viro static int acl_permission_check(struct inode *inode, int mask) 2905909ccaaSLinus Torvalds { 29126cf46beSLinus Torvalds unsigned int mode = inode->i_mode; 2925909ccaaSLinus Torvalds 2938e96e3b7SEric W. Biederman if (likely(uid_eq(current_fsuid(), inode->i_uid))) 2945909ccaaSLinus Torvalds mode >>= 6; 2955909ccaaSLinus Torvalds else { 296e77819e5SLinus Torvalds if (IS_POSIXACL(inode) && (mode & S_IRWXG)) { 2977e40145eSAl Viro int error = check_acl(inode, mask); 2985909ccaaSLinus Torvalds if (error != -EAGAIN) 2995909ccaaSLinus Torvalds return error; 3005909ccaaSLinus Torvalds } 3015909ccaaSLinus Torvalds 3025909ccaaSLinus Torvalds if (in_group_p(inode->i_gid)) 3035909ccaaSLinus Torvalds mode >>= 3; 3045909ccaaSLinus Torvalds } 3055909ccaaSLinus Torvalds 3065909ccaaSLinus Torvalds /* 3075909ccaaSLinus Torvalds * If the DACs are ok we don't need any capability check. 3085909ccaaSLinus Torvalds */ 3099c2c7039SAl Viro if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) 3105909ccaaSLinus Torvalds return 0; 3115909ccaaSLinus Torvalds return -EACCES; 3125909ccaaSLinus Torvalds } 3131da177e4SLinus Torvalds 3141da177e4SLinus Torvalds /** 3151da177e4SLinus Torvalds * generic_permission - check for access rights on a Posix-like filesystem 3161da177e4SLinus Torvalds * @inode: inode to check access rights for 3178fd90c8dSAndreas Gruenbacher * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...) 3181da177e4SLinus Torvalds * 3191da177e4SLinus Torvalds * Used to check for read/write/execute permissions on a file. 3201da177e4SLinus Torvalds * We use "fsuid" for this, letting us set arbitrary permissions 3211da177e4SLinus Torvalds * for filesystem access without changing the "normal" uids which 322b74c79e9SNick Piggin * are used for other things. 323b74c79e9SNick Piggin * 324b74c79e9SNick Piggin * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk 325b74c79e9SNick Piggin * request cannot be satisfied (eg. requires blocking or too much complexity). 326b74c79e9SNick Piggin * It would then be called again in ref-walk mode. 3271da177e4SLinus Torvalds */ 3282830ba7fSAl Viro int generic_permission(struct inode *inode, int mask) 3291da177e4SLinus Torvalds { 3305909ccaaSLinus Torvalds int ret; 3311da177e4SLinus Torvalds 3321da177e4SLinus Torvalds /* 333948409c7SAndreas Gruenbacher * Do the basic permission checks. 3341da177e4SLinus Torvalds */ 3357e40145eSAl Viro ret = acl_permission_check(inode, mask); 3365909ccaaSLinus Torvalds if (ret != -EACCES) 3375909ccaaSLinus Torvalds return ret; 3381da177e4SLinus Torvalds 339d594e7ecSAl Viro if (S_ISDIR(inode->i_mode)) { 340d594e7ecSAl Viro /* DACs are overridable for directories */ 34123adbe12SAndy Lutomirski if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE)) 342d594e7ecSAl Viro return 0; 343d594e7ecSAl Viro if (!(mask & MAY_WRITE)) 34423adbe12SAndy Lutomirski if (capable_wrt_inode_uidgid(inode, 34523adbe12SAndy Lutomirski CAP_DAC_READ_SEARCH)) 346d594e7ecSAl Viro return 0; 347d594e7ecSAl Viro return -EACCES; 348d594e7ecSAl Viro } 3491da177e4SLinus Torvalds /* 3501da177e4SLinus Torvalds * Read/write DACs are always overridable. 351d594e7ecSAl Viro * Executable DACs are overridable when there is 352d594e7ecSAl Viro * at least one exec bit set. 3531da177e4SLinus Torvalds */ 354d594e7ecSAl Viro if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO)) 35523adbe12SAndy Lutomirski if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE)) 3561da177e4SLinus Torvalds return 0; 3571da177e4SLinus Torvalds 3581da177e4SLinus Torvalds /* 3591da177e4SLinus Torvalds * Searching includes executable on directories, else just read. 3601da177e4SLinus Torvalds */ 3617ea66001SSerge E. Hallyn mask &= MAY_READ | MAY_WRITE | MAY_EXEC; 362d594e7ecSAl Viro if (mask == MAY_READ) 36323adbe12SAndy Lutomirski if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH)) 3641da177e4SLinus Torvalds return 0; 3651da177e4SLinus Torvalds 3661da177e4SLinus Torvalds return -EACCES; 3671da177e4SLinus Torvalds } 3684d359507SAl Viro EXPORT_SYMBOL(generic_permission); 3691da177e4SLinus Torvalds 3703ddcd056SLinus Torvalds /* 3713ddcd056SLinus Torvalds * We _really_ want to just do "generic_permission()" without 3723ddcd056SLinus Torvalds * even looking at the inode->i_op values. So we keep a cache 3733ddcd056SLinus Torvalds * flag in inode->i_opflags, that says "this has not special 3743ddcd056SLinus Torvalds * permission function, use the fast case". 3753ddcd056SLinus Torvalds */ 3763ddcd056SLinus Torvalds static inline int do_inode_permission(struct inode *inode, int mask) 3773ddcd056SLinus Torvalds { 3783ddcd056SLinus Torvalds if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) { 3793ddcd056SLinus Torvalds if (likely(inode->i_op->permission)) 3803ddcd056SLinus Torvalds return inode->i_op->permission(inode, mask); 3813ddcd056SLinus Torvalds 3823ddcd056SLinus Torvalds /* This gets set once for the inode lifetime */ 3833ddcd056SLinus Torvalds spin_lock(&inode->i_lock); 3843ddcd056SLinus Torvalds inode->i_opflags |= IOP_FASTPERM; 3853ddcd056SLinus Torvalds spin_unlock(&inode->i_lock); 3863ddcd056SLinus Torvalds } 3873ddcd056SLinus Torvalds return generic_permission(inode, mask); 3883ddcd056SLinus Torvalds } 3893ddcd056SLinus Torvalds 390cb23beb5SChristoph Hellwig /** 3910bdaea90SDavid Howells * __inode_permission - Check for access rights to a given inode 3920bdaea90SDavid Howells * @inode: Inode to check permission on 3930bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 394cb23beb5SChristoph Hellwig * 3950bdaea90SDavid Howells * Check for read/write/execute permissions on an inode. 396948409c7SAndreas Gruenbacher * 397948409c7SAndreas Gruenbacher * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. 3980bdaea90SDavid Howells * 3990bdaea90SDavid Howells * This does not check for a read-only file system. You probably want 4000bdaea90SDavid Howells * inode_permission(). 401cb23beb5SChristoph Hellwig */ 4020bdaea90SDavid Howells int __inode_permission(struct inode *inode, int mask) 4031da177e4SLinus Torvalds { 404e6305c43SAl Viro int retval; 4051da177e4SLinus Torvalds 4063ddcd056SLinus Torvalds if (unlikely(mask & MAY_WRITE)) { 4071da177e4SLinus Torvalds /* 4081da177e4SLinus Torvalds * Nobody gets write access to an immutable file. 4091da177e4SLinus Torvalds */ 4101da177e4SLinus Torvalds if (IS_IMMUTABLE(inode)) 4111da177e4SLinus Torvalds return -EACCES; 4121da177e4SLinus Torvalds } 4131da177e4SLinus Torvalds 4143ddcd056SLinus Torvalds retval = do_inode_permission(inode, mask); 4151da177e4SLinus Torvalds if (retval) 4161da177e4SLinus Torvalds return retval; 4171da177e4SLinus Torvalds 41808ce5f16SSerge E. Hallyn retval = devcgroup_inode_permission(inode, mask); 41908ce5f16SSerge E. Hallyn if (retval) 42008ce5f16SSerge E. Hallyn return retval; 42108ce5f16SSerge E. Hallyn 422d09ca739SEric Paris return security_inode_permission(inode, mask); 4231da177e4SLinus Torvalds } 424bd5d0856SMiklos Szeredi EXPORT_SYMBOL(__inode_permission); 4251da177e4SLinus Torvalds 426f4d6ff89SAl Viro /** 4270bdaea90SDavid Howells * sb_permission - Check superblock-level permissions 4280bdaea90SDavid Howells * @sb: Superblock of inode to check permission on 42955852635SRandy Dunlap * @inode: Inode to check permission on 4300bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 4310bdaea90SDavid Howells * 4320bdaea90SDavid Howells * Separate out file-system wide checks from inode-specific permission checks. 4330bdaea90SDavid Howells */ 4340bdaea90SDavid Howells static int sb_permission(struct super_block *sb, struct inode *inode, int mask) 4350bdaea90SDavid Howells { 4360bdaea90SDavid Howells if (unlikely(mask & MAY_WRITE)) { 4370bdaea90SDavid Howells umode_t mode = inode->i_mode; 4380bdaea90SDavid Howells 4390bdaea90SDavid Howells /* Nobody gets write access to a read-only fs. */ 4400bdaea90SDavid Howells if ((sb->s_flags & MS_RDONLY) && 4410bdaea90SDavid Howells (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) 4420bdaea90SDavid Howells return -EROFS; 4430bdaea90SDavid Howells } 4440bdaea90SDavid Howells return 0; 4450bdaea90SDavid Howells } 4460bdaea90SDavid Howells 4470bdaea90SDavid Howells /** 4480bdaea90SDavid Howells * inode_permission - Check for access rights to a given inode 4490bdaea90SDavid Howells * @inode: Inode to check permission on 4500bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 4510bdaea90SDavid Howells * 4520bdaea90SDavid Howells * Check for read/write/execute permissions on an inode. We use fs[ug]id for 4530bdaea90SDavid Howells * this, letting us set arbitrary permissions for filesystem access without 4540bdaea90SDavid Howells * changing the "normal" UIDs which are used for other things. 4550bdaea90SDavid Howells * 4560bdaea90SDavid Howells * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. 4570bdaea90SDavid Howells */ 4580bdaea90SDavid Howells int inode_permission(struct inode *inode, int mask) 4590bdaea90SDavid Howells { 4600bdaea90SDavid Howells int retval; 4610bdaea90SDavid Howells 4620bdaea90SDavid Howells retval = sb_permission(inode->i_sb, inode, mask); 4630bdaea90SDavid Howells if (retval) 4640bdaea90SDavid Howells return retval; 4650bdaea90SDavid Howells return __inode_permission(inode, mask); 4660bdaea90SDavid Howells } 4674d359507SAl Viro EXPORT_SYMBOL(inode_permission); 4680bdaea90SDavid Howells 4690bdaea90SDavid Howells /** 4705dd784d0SJan Blunck * path_get - get a reference to a path 4715dd784d0SJan Blunck * @path: path to get the reference to 4725dd784d0SJan Blunck * 4735dd784d0SJan Blunck * Given a path increment the reference count to the dentry and the vfsmount. 4745dd784d0SJan Blunck */ 475dcf787f3SAl Viro void path_get(const struct path *path) 4765dd784d0SJan Blunck { 4775dd784d0SJan Blunck mntget(path->mnt); 4785dd784d0SJan Blunck dget(path->dentry); 4795dd784d0SJan Blunck } 4805dd784d0SJan Blunck EXPORT_SYMBOL(path_get); 4815dd784d0SJan Blunck 4825dd784d0SJan Blunck /** 4831d957f9bSJan Blunck * path_put - put a reference to a path 4841d957f9bSJan Blunck * @path: path to put the reference to 4851d957f9bSJan Blunck * 4861d957f9bSJan Blunck * Given a path decrement the reference count to the dentry and the vfsmount. 4871d957f9bSJan Blunck */ 488dcf787f3SAl Viro void path_put(const struct path *path) 4891da177e4SLinus Torvalds { 4901d957f9bSJan Blunck dput(path->dentry); 4911d957f9bSJan Blunck mntput(path->mnt); 4921da177e4SLinus Torvalds } 4931d957f9bSJan Blunck EXPORT_SYMBOL(path_put); 4941da177e4SLinus Torvalds 495894bc8c4SAl Viro #define EMBEDDED_LEVELS 2 4961f55a6ecSAl Viro struct nameidata { 4971f55a6ecSAl Viro struct path path; 4981f55a6ecSAl Viro struct qstr last; 4991f55a6ecSAl Viro struct path root; 5001f55a6ecSAl Viro struct inode *inode; /* path.dentry.d_inode */ 5011f55a6ecSAl Viro unsigned int flags; 5028f47a016SAl Viro unsigned seq, m_seq, root_seq; 5031f55a6ecSAl Viro int last_type; 5041f55a6ecSAl Viro unsigned depth; 505756daf26SNeilBrown int total_link_count; 506697fc6caSAl Viro struct saved { 507697fc6caSAl Viro struct path link; 508697fc6caSAl Viro void *cookie; 509697fc6caSAl Viro const char *name; 510237d8b32SAl Viro struct inode *inode; 5110450b2d1SAl Viro unsigned seq; 512894bc8c4SAl Viro } *stack, internal[EMBEDDED_LEVELS]; 5131f55a6ecSAl Viro }; 5141f55a6ecSAl Viro 515756daf26SNeilBrown static struct nameidata *set_nameidata(struct nameidata *p) 516894bc8c4SAl Viro { 517756daf26SNeilBrown struct nameidata *old = current->nameidata; 518756daf26SNeilBrown p->stack = p->internal; 519756daf26SNeilBrown p->total_link_count = old ? old->total_link_count : 0; 520756daf26SNeilBrown current->nameidata = p; 521756daf26SNeilBrown return old; 522894bc8c4SAl Viro } 523894bc8c4SAl Viro 524756daf26SNeilBrown static void restore_nameidata(struct nameidata *old) 525894bc8c4SAl Viro { 526756daf26SNeilBrown struct nameidata *now = current->nameidata; 527756daf26SNeilBrown 528756daf26SNeilBrown current->nameidata = old; 529756daf26SNeilBrown if (old) 530756daf26SNeilBrown old->total_link_count = now->total_link_count; 531756daf26SNeilBrown if (now->stack != now->internal) { 532756daf26SNeilBrown kfree(now->stack); 533756daf26SNeilBrown now->stack = now->internal; 534894bc8c4SAl Viro } 535894bc8c4SAl Viro } 536894bc8c4SAl Viro 537894bc8c4SAl Viro static int __nd_alloc_stack(struct nameidata *nd) 538894bc8c4SAl Viro { 539bc40aee0SAl Viro struct saved *p; 540bc40aee0SAl Viro 541bc40aee0SAl Viro if (nd->flags & LOOKUP_RCU) { 542bc40aee0SAl Viro p= kmalloc(MAXSYMLINKS * sizeof(struct saved), 543bc40aee0SAl Viro GFP_ATOMIC); 544bc40aee0SAl Viro if (unlikely(!p)) 545bc40aee0SAl Viro return -ECHILD; 546bc40aee0SAl Viro } else { 547bc40aee0SAl Viro p= kmalloc(MAXSYMLINKS * sizeof(struct saved), 548894bc8c4SAl Viro GFP_KERNEL); 549894bc8c4SAl Viro if (unlikely(!p)) 550894bc8c4SAl Viro return -ENOMEM; 551bc40aee0SAl Viro } 552894bc8c4SAl Viro memcpy(p, nd->internal, sizeof(nd->internal)); 553894bc8c4SAl Viro nd->stack = p; 554894bc8c4SAl Viro return 0; 555894bc8c4SAl Viro } 556894bc8c4SAl Viro 557894bc8c4SAl Viro static inline int nd_alloc_stack(struct nameidata *nd) 558894bc8c4SAl Viro { 559da4e0be0SAl Viro if (likely(nd->depth != EMBEDDED_LEVELS)) 560894bc8c4SAl Viro return 0; 561894bc8c4SAl Viro if (likely(nd->stack != nd->internal)) 562894bc8c4SAl Viro return 0; 563894bc8c4SAl Viro return __nd_alloc_stack(nd); 564894bc8c4SAl Viro } 565894bc8c4SAl Viro 5667973387aSAl Viro static void drop_links(struct nameidata *nd) 5677973387aSAl Viro { 5687973387aSAl Viro int i = nd->depth; 5697973387aSAl Viro while (i--) { 5707973387aSAl Viro struct saved *last = nd->stack + i; 5717973387aSAl Viro struct inode *inode = last->inode; 5727973387aSAl Viro if (last->cookie && inode->i_op->put_link) { 5737973387aSAl Viro inode->i_op->put_link(inode, last->cookie); 5747973387aSAl Viro last->cookie = NULL; 5757973387aSAl Viro } 5767973387aSAl Viro } 5777973387aSAl Viro } 5787973387aSAl Viro 5797973387aSAl Viro static void terminate_walk(struct nameidata *nd) 5807973387aSAl Viro { 5817973387aSAl Viro drop_links(nd); 5827973387aSAl Viro if (!(nd->flags & LOOKUP_RCU)) { 5837973387aSAl Viro int i; 5847973387aSAl Viro path_put(&nd->path); 5857973387aSAl Viro for (i = 0; i < nd->depth; i++) 5867973387aSAl Viro path_put(&nd->stack[i].link); 5877973387aSAl Viro } else { 5887973387aSAl Viro nd->flags &= ~LOOKUP_RCU; 5897973387aSAl Viro if (!(nd->flags & LOOKUP_ROOT)) 5907973387aSAl Viro nd->root.mnt = NULL; 5917973387aSAl Viro rcu_read_unlock(); 5927973387aSAl Viro } 5937973387aSAl Viro nd->depth = 0; 5947973387aSAl Viro } 5957973387aSAl Viro 5967973387aSAl Viro /* path_put is needed afterwards regardless of success or failure */ 5977973387aSAl Viro static bool legitimize_path(struct nameidata *nd, 5987973387aSAl Viro struct path *path, unsigned seq) 5997973387aSAl Viro { 6007973387aSAl Viro int res = __legitimize_mnt(path->mnt, nd->m_seq); 6017973387aSAl Viro if (unlikely(res)) { 6027973387aSAl Viro if (res > 0) 6037973387aSAl Viro path->mnt = NULL; 6047973387aSAl Viro path->dentry = NULL; 6057973387aSAl Viro return false; 6067973387aSAl Viro } 6077973387aSAl Viro if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) { 6087973387aSAl Viro path->dentry = NULL; 6097973387aSAl Viro return false; 6107973387aSAl Viro } 6117973387aSAl Viro return !read_seqcount_retry(&path->dentry->d_seq, seq); 6127973387aSAl Viro } 6137973387aSAl Viro 6147973387aSAl Viro static bool legitimize_links(struct nameidata *nd) 6157973387aSAl Viro { 6167973387aSAl Viro int i; 6177973387aSAl Viro for (i = 0; i < nd->depth; i++) { 6187973387aSAl Viro struct saved *last = nd->stack + i; 6197973387aSAl Viro if (unlikely(!legitimize_path(nd, &last->link, last->seq))) { 6207973387aSAl Viro drop_links(nd); 6217973387aSAl Viro nd->depth = i + 1; 6227973387aSAl Viro return false; 6237973387aSAl Viro } 6247973387aSAl Viro } 6257973387aSAl Viro return true; 6267973387aSAl Viro } 6277973387aSAl Viro 62819660af7SAl Viro /* 62931e6b01fSNick Piggin * Path walking has 2 modes, rcu-walk and ref-walk (see 63019660af7SAl Viro * Documentation/filesystems/path-lookup.txt). In situations when we can't 63119660af7SAl Viro * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab 63219660af7SAl Viro * normal reference counts on dentries and vfsmounts to transition to rcu-walk 63319660af7SAl Viro * mode. Refcounts are grabbed at the last known good point before rcu-walk 63419660af7SAl Viro * got stuck, so ref-walk may continue from there. If this is not successful 63519660af7SAl Viro * (eg. a seqcount has changed), then failure is returned and it's up to caller 63619660af7SAl Viro * to restart the path walk from the beginning in ref-walk mode. 63731e6b01fSNick Piggin */ 63831e6b01fSNick Piggin 63931e6b01fSNick Piggin /** 64019660af7SAl Viro * unlazy_walk - try to switch to ref-walk mode. 64119660af7SAl Viro * @nd: nameidata pathwalk data 64219660af7SAl Viro * @dentry: child of nd->path.dentry or NULL 6436e9918b7SAl Viro * @seq: seq number to check dentry against 64439191628SRandy Dunlap * Returns: 0 on success, -ECHILD on failure 64531e6b01fSNick Piggin * 64619660af7SAl Viro * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry 64719660af7SAl Viro * for ref-walk mode. @dentry must be a path found by a do_lookup call on 64819660af7SAl Viro * @nd or NULL. Must be called from rcu-walk context. 6497973387aSAl Viro * Nothing should touch nameidata between unlazy_walk() failure and 6507973387aSAl Viro * terminate_walk(). 65131e6b01fSNick Piggin */ 6526e9918b7SAl Viro static int unlazy_walk(struct nameidata *nd, struct dentry *dentry, unsigned seq) 65331e6b01fSNick Piggin { 65431e6b01fSNick Piggin struct dentry *parent = nd->path.dentry; 65531e6b01fSNick Piggin 65631e6b01fSNick Piggin BUG_ON(!(nd->flags & LOOKUP_RCU)); 657e5c832d5SLinus Torvalds 658e5c832d5SLinus Torvalds nd->flags &= ~LOOKUP_RCU; 6597973387aSAl Viro if (unlikely(!legitimize_links(nd))) 6607973387aSAl Viro goto out2; 6617973387aSAl Viro if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq))) 6627973387aSAl Viro goto out2; 6637973387aSAl Viro if (unlikely(!lockref_get_not_dead(&parent->d_lockref))) 6647973387aSAl Viro goto out1; 66548a066e7SAl Viro 66615570086SLinus Torvalds /* 66715570086SLinus Torvalds * For a negative lookup, the lookup sequence point is the parents 66815570086SLinus Torvalds * sequence point, and it only needs to revalidate the parent dentry. 66915570086SLinus Torvalds * 67015570086SLinus Torvalds * For a positive lookup, we need to move both the parent and the 67115570086SLinus Torvalds * dentry from the RCU domain to be properly refcounted. And the 67215570086SLinus Torvalds * sequence number in the dentry validates *both* dentry counters, 67315570086SLinus Torvalds * since we checked the sequence number of the parent after we got 67415570086SLinus Torvalds * the child sequence number. So we know the parent must still 67515570086SLinus Torvalds * be valid if the child sequence number is still valid. 67615570086SLinus Torvalds */ 67719660af7SAl Viro if (!dentry) { 678e5c832d5SLinus Torvalds if (read_seqcount_retry(&parent->d_seq, nd->seq)) 679e5c832d5SLinus Torvalds goto out; 68019660af7SAl Viro BUG_ON(nd->inode != parent->d_inode); 68119660af7SAl Viro } else { 682e5c832d5SLinus Torvalds if (!lockref_get_not_dead(&dentry->d_lockref)) 683e5c832d5SLinus Torvalds goto out; 6846e9918b7SAl Viro if (read_seqcount_retry(&dentry->d_seq, seq)) 685e5c832d5SLinus Torvalds goto drop_dentry; 68619660af7SAl Viro } 687e5c832d5SLinus Torvalds 688e5c832d5SLinus Torvalds /* 689e5c832d5SLinus Torvalds * Sequence counts matched. Now make sure that the root is 690e5c832d5SLinus Torvalds * still valid and get it if required. 691e5c832d5SLinus Torvalds */ 692e5c832d5SLinus Torvalds if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 6935a8d87e8SAl Viro if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) { 6945a8d87e8SAl Viro rcu_read_unlock(); 6955a8d87e8SAl Viro dput(dentry); 6965a8d87e8SAl Viro return -ECHILD; 6977973387aSAl Viro } 69831e6b01fSNick Piggin } 69931e6b01fSNick Piggin 7008b61e74fSAl Viro rcu_read_unlock(); 70131e6b01fSNick Piggin return 0; 70219660af7SAl Viro 703e5c832d5SLinus Torvalds drop_dentry: 7048b61e74fSAl Viro rcu_read_unlock(); 705e5c832d5SLinus Torvalds dput(dentry); 706d0d27277SLinus Torvalds goto drop_root_mnt; 7077973387aSAl Viro out2: 7087973387aSAl Viro nd->path.mnt = NULL; 7097973387aSAl Viro out1: 7107973387aSAl Viro nd->path.dentry = NULL; 711e5c832d5SLinus Torvalds out: 7128b61e74fSAl Viro rcu_read_unlock(); 713d0d27277SLinus Torvalds drop_root_mnt: 714d0d27277SLinus Torvalds if (!(nd->flags & LOOKUP_ROOT)) 715d0d27277SLinus Torvalds nd->root.mnt = NULL; 71631e6b01fSNick Piggin return -ECHILD; 71731e6b01fSNick Piggin } 71831e6b01fSNick Piggin 7197973387aSAl Viro static int unlazy_link(struct nameidata *nd, struct path *link, unsigned seq) 7207973387aSAl Viro { 7217973387aSAl Viro if (unlikely(!legitimize_path(nd, link, seq))) { 7227973387aSAl Viro drop_links(nd); 7237973387aSAl Viro nd->depth = 0; 7247973387aSAl Viro nd->flags &= ~LOOKUP_RCU; 7257973387aSAl Viro nd->path.mnt = NULL; 7267973387aSAl Viro nd->path.dentry = NULL; 7277973387aSAl Viro if (!(nd->flags & LOOKUP_ROOT)) 7287973387aSAl Viro nd->root.mnt = NULL; 7297973387aSAl Viro rcu_read_unlock(); 7307973387aSAl Viro } else if (likely(unlazy_walk(nd, NULL, 0)) == 0) { 7317973387aSAl Viro return 0; 7327973387aSAl Viro } 7337973387aSAl Viro path_put(link); 7347973387aSAl Viro return -ECHILD; 7357973387aSAl Viro } 7367973387aSAl Viro 7374ce16ef3SAl Viro static inline int d_revalidate(struct dentry *dentry, unsigned int flags) 73834286d66SNick Piggin { 7394ce16ef3SAl Viro return dentry->d_op->d_revalidate(dentry, flags); 74034286d66SNick Piggin } 74134286d66SNick Piggin 7429f1fafeeSAl Viro /** 7439f1fafeeSAl Viro * complete_walk - successful completion of path walk 7449f1fafeeSAl Viro * @nd: pointer nameidata 74539159de2SJeff Layton * 7469f1fafeeSAl Viro * If we had been in RCU mode, drop out of it and legitimize nd->path. 7479f1fafeeSAl Viro * Revalidate the final result, unless we'd already done that during 7489f1fafeeSAl Viro * the path walk or the filesystem doesn't ask for it. Return 0 on 7499f1fafeeSAl Viro * success, -error on failure. In case of failure caller does not 7509f1fafeeSAl Viro * need to drop nd->path. 75139159de2SJeff Layton */ 7529f1fafeeSAl Viro static int complete_walk(struct nameidata *nd) 75339159de2SJeff Layton { 75416c2cd71SAl Viro struct dentry *dentry = nd->path.dentry; 75539159de2SJeff Layton int status; 75639159de2SJeff Layton 7579f1fafeeSAl Viro if (nd->flags & LOOKUP_RCU) { 7589f1fafeeSAl Viro if (!(nd->flags & LOOKUP_ROOT)) 7599f1fafeeSAl Viro nd->root.mnt = NULL; 7606e9918b7SAl Viro if (unlikely(unlazy_walk(nd, NULL, 0))) 76148a066e7SAl Viro return -ECHILD; 76248a066e7SAl Viro } 7639f1fafeeSAl Viro 76416c2cd71SAl Viro if (likely(!(nd->flags & LOOKUP_JUMPED))) 76539159de2SJeff Layton return 0; 76639159de2SJeff Layton 767ecf3d1f1SJeff Layton if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) 76816c2cd71SAl Viro return 0; 76916c2cd71SAl Viro 770ecf3d1f1SJeff Layton status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); 77139159de2SJeff Layton if (status > 0) 77239159de2SJeff Layton return 0; 77339159de2SJeff Layton 77416c2cd71SAl Viro if (!status) 77539159de2SJeff Layton status = -ESTALE; 77616c2cd71SAl Viro 77739159de2SJeff Layton return status; 77839159de2SJeff Layton } 77939159de2SJeff Layton 78018d8c860SAl Viro static void set_root(struct nameidata *nd) 7812a737871SAl Viro { 782f7ad3c6bSMiklos Szeredi get_fs_root(current->fs, &nd->root); 7832a737871SAl Viro } 7842a737871SAl Viro 78518d8c860SAl Viro static unsigned set_root_rcu(struct nameidata *nd) 78631e6b01fSNick Piggin { 78731e6b01fSNick Piggin struct fs_struct *fs = current->fs; 7888f47a016SAl Viro unsigned seq; 789c28cc364SNick Piggin 790c28cc364SNick Piggin do { 791c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 79231e6b01fSNick Piggin nd->root = fs->root; 7938f47a016SAl Viro nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq); 794c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 7958f47a016SAl Viro return nd->root_seq; 79631e6b01fSNick Piggin } 79731e6b01fSNick Piggin 7981d957f9bSJan Blunck static void path_put_conditional(struct path *path, struct nameidata *nd) 799051d3812SIan Kent { 800051d3812SIan Kent dput(path->dentry); 8014ac91378SJan Blunck if (path->mnt != nd->path.mnt) 802051d3812SIan Kent mntput(path->mnt); 803051d3812SIan Kent } 804051d3812SIan Kent 8057b9337aaSNick Piggin static inline void path_to_nameidata(const struct path *path, 8067b9337aaSNick Piggin struct nameidata *nd) 807051d3812SIan Kent { 80831e6b01fSNick Piggin if (!(nd->flags & LOOKUP_RCU)) { 8094ac91378SJan Blunck dput(nd->path.dentry); 81031e6b01fSNick Piggin if (nd->path.mnt != path->mnt) 8114ac91378SJan Blunck mntput(nd->path.mnt); 8129a229683SHuang Shijie } 81331e6b01fSNick Piggin nd->path.mnt = path->mnt; 8144ac91378SJan Blunck nd->path.dentry = path->dentry; 815051d3812SIan Kent } 816051d3812SIan Kent 817b5fb63c1SChristoph Hellwig /* 818b5fb63c1SChristoph Hellwig * Helper to directly jump to a known parsed path from ->follow_link, 819b5fb63c1SChristoph Hellwig * caller must have taken a reference to path beforehand. 820b5fb63c1SChristoph Hellwig */ 8216e77137bSAl Viro void nd_jump_link(struct path *path) 822b5fb63c1SChristoph Hellwig { 8236e77137bSAl Viro struct nameidata *nd = current->nameidata; 824b5fb63c1SChristoph Hellwig path_put(&nd->path); 825b5fb63c1SChristoph Hellwig 826b5fb63c1SChristoph Hellwig nd->path = *path; 827b5fb63c1SChristoph Hellwig nd->inode = nd->path.dentry->d_inode; 828b5fb63c1SChristoph Hellwig nd->flags |= LOOKUP_JUMPED; 829b5fb63c1SChristoph Hellwig } 830b5fb63c1SChristoph Hellwig 831b9ff4429SAl Viro static inline void put_link(struct nameidata *nd) 832574197e0SAl Viro { 83321c3003dSAl Viro struct saved *last = nd->stack + --nd->depth; 834237d8b32SAl Viro struct inode *inode = last->inode; 835b9ff4429SAl Viro if (last->cookie && inode->i_op->put_link) 8365f2c4179SAl Viro inode->i_op->put_link(inode, last->cookie); 8376548fae2SAl Viro if (!(nd->flags & LOOKUP_RCU)) 838b9ff4429SAl Viro path_put(&last->link); 839574197e0SAl Viro } 840574197e0SAl Viro 841561ec64aSLinus Torvalds int sysctl_protected_symlinks __read_mostly = 0; 842561ec64aSLinus Torvalds int sysctl_protected_hardlinks __read_mostly = 0; 843800179c9SKees Cook 844800179c9SKees Cook /** 845800179c9SKees Cook * may_follow_link - Check symlink following for unsafe situations 84655852635SRandy Dunlap * @nd: nameidata pathwalk data 847800179c9SKees Cook * 848800179c9SKees Cook * In the case of the sysctl_protected_symlinks sysctl being enabled, 849800179c9SKees Cook * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is 850800179c9SKees Cook * in a sticky world-writable directory. This is to protect privileged 851800179c9SKees Cook * processes from failing races against path names that may change out 852800179c9SKees Cook * from under them by way of other users creating malicious symlinks. 853800179c9SKees Cook * It will permit symlinks to be followed only when outside a sticky 854800179c9SKees Cook * world-writable directory, or when the uid of the symlink and follower 855800179c9SKees Cook * match, or when the directory owner matches the symlink's owner. 856800179c9SKees Cook * 857800179c9SKees Cook * Returns 0 if following the symlink is allowed, -ve on error. 858800179c9SKees Cook */ 859fec2fa24SAl Viro static inline int may_follow_link(struct nameidata *nd) 860800179c9SKees Cook { 861800179c9SKees Cook const struct inode *inode; 862800179c9SKees Cook const struct inode *parent; 863800179c9SKees Cook 864800179c9SKees Cook if (!sysctl_protected_symlinks) 865800179c9SKees Cook return 0; 866800179c9SKees Cook 867800179c9SKees Cook /* Allowed if owner and follower match. */ 868237d8b32SAl Viro inode = nd->stack[0].inode; 86981abe27bSEric W. Biederman if (uid_eq(current_cred()->fsuid, inode->i_uid)) 870800179c9SKees Cook return 0; 871800179c9SKees Cook 872800179c9SKees Cook /* Allowed if parent directory not sticky and world-writable. */ 873800179c9SKees Cook parent = nd->path.dentry->d_inode; 874800179c9SKees Cook if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) 875800179c9SKees Cook return 0; 876800179c9SKees Cook 877800179c9SKees Cook /* Allowed if parent directory and link owner match. */ 87881abe27bSEric W. Biederman if (uid_eq(parent->i_uid, inode->i_uid)) 879800179c9SKees Cook return 0; 880800179c9SKees Cook 88131956502SAl Viro if (nd->flags & LOOKUP_RCU) 88231956502SAl Viro return -ECHILD; 88331956502SAl Viro 8841cf2665bSAl Viro audit_log_link_denied("follow_link", &nd->stack[0].link); 885800179c9SKees Cook return -EACCES; 886800179c9SKees Cook } 887800179c9SKees Cook 888800179c9SKees Cook /** 889800179c9SKees Cook * safe_hardlink_source - Check for safe hardlink conditions 890800179c9SKees Cook * @inode: the source inode to hardlink from 891800179c9SKees Cook * 892800179c9SKees Cook * Return false if at least one of the following conditions: 893800179c9SKees Cook * - inode is not a regular file 894800179c9SKees Cook * - inode is setuid 895800179c9SKees Cook * - inode is setgid and group-exec 896800179c9SKees Cook * - access failure for read and write 897800179c9SKees Cook * 898800179c9SKees Cook * Otherwise returns true. 899800179c9SKees Cook */ 900800179c9SKees Cook static bool safe_hardlink_source(struct inode *inode) 901800179c9SKees Cook { 902800179c9SKees Cook umode_t mode = inode->i_mode; 903800179c9SKees Cook 904800179c9SKees Cook /* Special files should not get pinned to the filesystem. */ 905800179c9SKees Cook if (!S_ISREG(mode)) 906800179c9SKees Cook return false; 907800179c9SKees Cook 908800179c9SKees Cook /* Setuid files should not get pinned to the filesystem. */ 909800179c9SKees Cook if (mode & S_ISUID) 910800179c9SKees Cook return false; 911800179c9SKees Cook 912800179c9SKees Cook /* Executable setgid files should not get pinned to the filesystem. */ 913800179c9SKees Cook if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) 914800179c9SKees Cook return false; 915800179c9SKees Cook 916800179c9SKees Cook /* Hardlinking to unreadable or unwritable sources is dangerous. */ 917800179c9SKees Cook if (inode_permission(inode, MAY_READ | MAY_WRITE)) 918800179c9SKees Cook return false; 919800179c9SKees Cook 920800179c9SKees Cook return true; 921800179c9SKees Cook } 922800179c9SKees Cook 923800179c9SKees Cook /** 924800179c9SKees Cook * may_linkat - Check permissions for creating a hardlink 925800179c9SKees Cook * @link: the source to hardlink from 926800179c9SKees Cook * 927800179c9SKees Cook * Block hardlink when all of: 928800179c9SKees Cook * - sysctl_protected_hardlinks enabled 929800179c9SKees Cook * - fsuid does not match inode 930800179c9SKees Cook * - hardlink source is unsafe (see safe_hardlink_source() above) 931800179c9SKees Cook * - not CAP_FOWNER 932800179c9SKees Cook * 933800179c9SKees Cook * Returns 0 if successful, -ve on error. 934800179c9SKees Cook */ 935800179c9SKees Cook static int may_linkat(struct path *link) 936800179c9SKees Cook { 937800179c9SKees Cook const struct cred *cred; 938800179c9SKees Cook struct inode *inode; 939800179c9SKees Cook 940800179c9SKees Cook if (!sysctl_protected_hardlinks) 941800179c9SKees Cook return 0; 942800179c9SKees Cook 943800179c9SKees Cook cred = current_cred(); 944800179c9SKees Cook inode = link->dentry->d_inode; 945800179c9SKees Cook 946800179c9SKees Cook /* Source inode owner (or CAP_FOWNER) can hardlink all they like, 947800179c9SKees Cook * otherwise, it must be a safe source. 948800179c9SKees Cook */ 94981abe27bSEric W. Biederman if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) || 950800179c9SKees Cook capable(CAP_FOWNER)) 951800179c9SKees Cook return 0; 952800179c9SKees Cook 953a51d9eaaSKees Cook audit_log_link_denied("linkat", link); 954800179c9SKees Cook return -EPERM; 955800179c9SKees Cook } 956800179c9SKees Cook 9573b2e7f75SAl Viro static __always_inline 9583b2e7f75SAl Viro const char *get_link(struct nameidata *nd) 9591da177e4SLinus Torvalds { 960ab104923SAl Viro struct saved *last = nd->stack + nd->depth - 1; 9611cf2665bSAl Viro struct dentry *dentry = last->link.dentry; 962237d8b32SAl Viro struct inode *inode = last->inode; 9636d7b5aaeSAl Viro int error; 9640a959df5SAl Viro const char *res; 9651da177e4SLinus Torvalds 9668fa9dd24SNeilBrown if (!(nd->flags & LOOKUP_RCU)) { 9678fa9dd24SNeilBrown touch_atime(&last->link); 9688fa9dd24SNeilBrown cond_resched(); 9698fa9dd24SNeilBrown } else if (atime_needs_update(&last->link, inode)) { 9708fa9dd24SNeilBrown if (unlikely(unlazy_walk(nd, NULL, 0))) 9718fa9dd24SNeilBrown return ERR_PTR(-ECHILD); 9728fa9dd24SNeilBrown touch_atime(&last->link); 9738fa9dd24SNeilBrown } 9748fa9dd24SNeilBrown 975bda0be7aSNeilBrown error = security_inode_follow_link(dentry, inode, 976bda0be7aSNeilBrown nd->flags & LOOKUP_RCU); 977bda0be7aSNeilBrown if (unlikely(error)) 9786920a440SAl Viro return ERR_PTR(error); 97936f3b4f6SAl Viro 98086acdca1SAl Viro nd->last_type = LAST_BIND; 981d4dee48bSAl Viro res = inode->i_link; 982d4dee48bSAl Viro if (!res) { 9838c1b4566SAl Viro if (nd->flags & LOOKUP_RCU) { 9848c1b4566SAl Viro if (unlikely(unlazy_walk(nd, NULL, 0))) 9858c1b4566SAl Viro return ERR_PTR(-ECHILD); 9868c1b4566SAl Viro } 9876e77137bSAl Viro res = inode->i_op->follow_link(dentry, &last->cookie); 988fab51e8aSAl Viro if (IS_ERR_OR_NULL(res)) { 9896920a440SAl Viro last->cookie = NULL; 990fab51e8aSAl Viro return res; 9910a959df5SAl Viro } 992fab51e8aSAl Viro } 993fab51e8aSAl Viro if (*res == '/') { 9948c1b4566SAl Viro if (nd->flags & LOOKUP_RCU) { 9958f47a016SAl Viro struct dentry *d; 9968f47a016SAl Viro if (!nd->root.mnt) 9978f47a016SAl Viro set_root_rcu(nd); 9988f47a016SAl Viro nd->path = nd->root; 9998f47a016SAl Viro d = nd->path.dentry; 10008f47a016SAl Viro nd->inode = d->d_inode; 10018f47a016SAl Viro nd->seq = nd->root_seq; 10028f47a016SAl Viro if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq))) 10038c1b4566SAl Viro return ERR_PTR(-ECHILD); 10048f47a016SAl Viro } else { 1005fab51e8aSAl Viro if (!nd->root.mnt) 1006fab51e8aSAl Viro set_root(nd); 1007fab51e8aSAl Viro path_put(&nd->path); 1008fab51e8aSAl Viro nd->path = nd->root; 1009fab51e8aSAl Viro path_get(&nd->root); 1010fab51e8aSAl Viro nd->inode = nd->path.dentry->d_inode; 10118f47a016SAl Viro } 1012fab51e8aSAl Viro nd->flags |= LOOKUP_JUMPED; 1013fab51e8aSAl Viro while (unlikely(*++res == '/')) 1014fab51e8aSAl Viro ; 1015fab51e8aSAl Viro } 1016fab51e8aSAl Viro if (!*res) 1017fab51e8aSAl Viro res = NULL; 10180a959df5SAl Viro return res; 10190a959df5SAl Viro } 10206d7b5aaeSAl Viro 1021f015f126SDavid Howells /* 1022f015f126SDavid Howells * follow_up - Find the mountpoint of path's vfsmount 1023f015f126SDavid Howells * 1024f015f126SDavid Howells * Given a path, find the mountpoint of its source file system. 1025f015f126SDavid Howells * Replace @path with the path of the mountpoint in the parent mount. 1026f015f126SDavid Howells * Up is towards /. 1027f015f126SDavid Howells * 1028f015f126SDavid Howells * Return 1 if we went up a level and 0 if we were already at the 1029f015f126SDavid Howells * root. 1030f015f126SDavid Howells */ 1031bab77ebfSAl Viro int follow_up(struct path *path) 10321da177e4SLinus Torvalds { 10330714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 10340714a533SAl Viro struct mount *parent; 10351da177e4SLinus Torvalds struct dentry *mountpoint; 103699b7db7bSNick Piggin 103748a066e7SAl Viro read_seqlock_excl(&mount_lock); 10380714a533SAl Viro parent = mnt->mnt_parent; 10393c0a6163SAl Viro if (parent == mnt) { 104048a066e7SAl Viro read_sequnlock_excl(&mount_lock); 10411da177e4SLinus Torvalds return 0; 10421da177e4SLinus Torvalds } 10430714a533SAl Viro mntget(&parent->mnt); 1044a73324daSAl Viro mountpoint = dget(mnt->mnt_mountpoint); 104548a066e7SAl Viro read_sequnlock_excl(&mount_lock); 1046bab77ebfSAl Viro dput(path->dentry); 1047bab77ebfSAl Viro path->dentry = mountpoint; 1048bab77ebfSAl Viro mntput(path->mnt); 10490714a533SAl Viro path->mnt = &parent->mnt; 10501da177e4SLinus Torvalds return 1; 10511da177e4SLinus Torvalds } 10524d359507SAl Viro EXPORT_SYMBOL(follow_up); 10531da177e4SLinus Torvalds 1054b5c84bf6SNick Piggin /* 10559875cf80SDavid Howells * Perform an automount 10569875cf80SDavid Howells * - return -EISDIR to tell follow_managed() to stop and return the path we 10579875cf80SDavid Howells * were called with. 10581da177e4SLinus Torvalds */ 1059756daf26SNeilBrown static int follow_automount(struct path *path, struct nameidata *nd, 10609875cf80SDavid Howells bool *need_mntput) 106131e6b01fSNick Piggin { 10629875cf80SDavid Howells struct vfsmount *mnt; 1063ea5b778aSDavid Howells int err; 10649875cf80SDavid Howells 10659875cf80SDavid Howells if (!path->dentry->d_op || !path->dentry->d_op->d_automount) 10669875cf80SDavid Howells return -EREMOTE; 10679875cf80SDavid Howells 10680ec26fd0SMiklos Szeredi /* We don't want to mount if someone's just doing a stat - 10690ec26fd0SMiklos Szeredi * unless they're stat'ing a directory and appended a '/' to 10700ec26fd0SMiklos Szeredi * the name. 10710ec26fd0SMiklos Szeredi * 10720ec26fd0SMiklos Szeredi * We do, however, want to mount if someone wants to open or 10735a30d8a2SDavid Howells * create a file of any type under the mountpoint, wants to 10740ec26fd0SMiklos Szeredi * traverse through the mountpoint or wants to open the 10750ec26fd0SMiklos Szeredi * mounted directory. Also, autofs may mark negative dentries 10760ec26fd0SMiklos Szeredi * as being automount points. These will need the attentions 10770ec26fd0SMiklos Szeredi * of the daemon to instantiate them before they can be used. 10785a30d8a2SDavid Howells */ 1079756daf26SNeilBrown if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | 1080d94c177bSLinus Torvalds LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && 10815a30d8a2SDavid Howells path->dentry->d_inode) 10829875cf80SDavid Howells return -EISDIR; 10830ec26fd0SMiklos Szeredi 1084756daf26SNeilBrown nd->total_link_count++; 1085756daf26SNeilBrown if (nd->total_link_count >= 40) 10869875cf80SDavid Howells return -ELOOP; 10879875cf80SDavid Howells 10889875cf80SDavid Howells mnt = path->dentry->d_op->d_automount(path); 10899875cf80SDavid Howells if (IS_ERR(mnt)) { 10909875cf80SDavid Howells /* 10919875cf80SDavid Howells * The filesystem is allowed to return -EISDIR here to indicate 10929875cf80SDavid Howells * it doesn't want to automount. For instance, autofs would do 10939875cf80SDavid Howells * this so that its userspace daemon can mount on this dentry. 10949875cf80SDavid Howells * 10959875cf80SDavid Howells * However, we can only permit this if it's a terminal point in 10969875cf80SDavid Howells * the path being looked up; if it wasn't then the remainder of 10979875cf80SDavid Howells * the path is inaccessible and we should say so. 10989875cf80SDavid Howells */ 1099756daf26SNeilBrown if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT)) 11009875cf80SDavid Howells return -EREMOTE; 11019875cf80SDavid Howells return PTR_ERR(mnt); 110231e6b01fSNick Piggin } 1103ea5b778aSDavid Howells 11049875cf80SDavid Howells if (!mnt) /* mount collision */ 11059875cf80SDavid Howells return 0; 11069875cf80SDavid Howells 11078aef1884SAl Viro if (!*need_mntput) { 11088aef1884SAl Viro /* lock_mount() may release path->mnt on error */ 11098aef1884SAl Viro mntget(path->mnt); 11108aef1884SAl Viro *need_mntput = true; 11118aef1884SAl Viro } 111219a167afSAl Viro err = finish_automount(mnt, path); 1113ea5b778aSDavid Howells 1114ea5b778aSDavid Howells switch (err) { 1115ea5b778aSDavid Howells case -EBUSY: 1116ea5b778aSDavid Howells /* Someone else made a mount here whilst we were busy */ 111719a167afSAl Viro return 0; 1118ea5b778aSDavid Howells case 0: 11198aef1884SAl Viro path_put(path); 11209875cf80SDavid Howells path->mnt = mnt; 11219875cf80SDavid Howells path->dentry = dget(mnt->mnt_root); 11229875cf80SDavid Howells return 0; 112319a167afSAl Viro default: 112419a167afSAl Viro return err; 11259875cf80SDavid Howells } 112619a167afSAl Viro 1127ea5b778aSDavid Howells } 11289875cf80SDavid Howells 11299875cf80SDavid Howells /* 11309875cf80SDavid Howells * Handle a dentry that is managed in some way. 1131cc53ce53SDavid Howells * - Flagged for transit management (autofs) 11329875cf80SDavid Howells * - Flagged as mountpoint 11339875cf80SDavid Howells * - Flagged as automount point 11349875cf80SDavid Howells * 11359875cf80SDavid Howells * This may only be called in refwalk mode. 11369875cf80SDavid Howells * 11379875cf80SDavid Howells * Serialization is taken care of in namespace.c 11389875cf80SDavid Howells */ 1139756daf26SNeilBrown static int follow_managed(struct path *path, struct nameidata *nd) 11409875cf80SDavid Howells { 11418aef1884SAl Viro struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */ 11429875cf80SDavid Howells unsigned managed; 11439875cf80SDavid Howells bool need_mntput = false; 11448aef1884SAl Viro int ret = 0; 11459875cf80SDavid Howells 11469875cf80SDavid Howells /* Given that we're not holding a lock here, we retain the value in a 11479875cf80SDavid Howells * local variable for each dentry as we look at it so that we don't see 11489875cf80SDavid Howells * the components of that value change under us */ 11499875cf80SDavid Howells while (managed = ACCESS_ONCE(path->dentry->d_flags), 11509875cf80SDavid Howells managed &= DCACHE_MANAGED_DENTRY, 11519875cf80SDavid Howells unlikely(managed != 0)) { 1152cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1153cc53ce53SDavid Howells * being held. */ 1154cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 1155cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1156cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 11571aed3e42SAl Viro ret = path->dentry->d_op->d_manage(path->dentry, false); 1158cc53ce53SDavid Howells if (ret < 0) 11598aef1884SAl Viro break; 1160cc53ce53SDavid Howells } 1161cc53ce53SDavid Howells 11629875cf80SDavid Howells /* Transit to a mounted filesystem. */ 11639875cf80SDavid Howells if (managed & DCACHE_MOUNTED) { 11649875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 11659875cf80SDavid Howells if (mounted) { 11669875cf80SDavid Howells dput(path->dentry); 11679875cf80SDavid Howells if (need_mntput) 1168463ffb2eSAl Viro mntput(path->mnt); 1169463ffb2eSAl Viro path->mnt = mounted; 1170463ffb2eSAl Viro path->dentry = dget(mounted->mnt_root); 11719875cf80SDavid Howells need_mntput = true; 11729875cf80SDavid Howells continue; 1173463ffb2eSAl Viro } 1174463ffb2eSAl Viro 11759875cf80SDavid Howells /* Something is mounted on this dentry in another 11769875cf80SDavid Howells * namespace and/or whatever was mounted there in this 117748a066e7SAl Viro * namespace got unmounted before lookup_mnt() could 117848a066e7SAl Viro * get it */ 11791da177e4SLinus Torvalds } 11809875cf80SDavid Howells 11819875cf80SDavid Howells /* Handle an automount point */ 11829875cf80SDavid Howells if (managed & DCACHE_NEED_AUTOMOUNT) { 1183756daf26SNeilBrown ret = follow_automount(path, nd, &need_mntput); 11849875cf80SDavid Howells if (ret < 0) 11858aef1884SAl Viro break; 11869875cf80SDavid Howells continue; 11879875cf80SDavid Howells } 11889875cf80SDavid Howells 11899875cf80SDavid Howells /* We didn't change the current path point */ 11909875cf80SDavid Howells break; 11919875cf80SDavid Howells } 11928aef1884SAl Viro 11938aef1884SAl Viro if (need_mntput && path->mnt == mnt) 11948aef1884SAl Viro mntput(path->mnt); 11958aef1884SAl Viro if (ret == -EISDIR) 11968aef1884SAl Viro ret = 0; 11978402752eSAl Viro if (need_mntput) 11988402752eSAl Viro nd->flags |= LOOKUP_JUMPED; 11998402752eSAl Viro if (unlikely(ret < 0)) 12008402752eSAl Viro path_put_conditional(path, nd); 12018402752eSAl Viro return ret; 12021da177e4SLinus Torvalds } 12031da177e4SLinus Torvalds 1204cc53ce53SDavid Howells int follow_down_one(struct path *path) 12051da177e4SLinus Torvalds { 12061da177e4SLinus Torvalds struct vfsmount *mounted; 12071da177e4SLinus Torvalds 12081c755af4SAl Viro mounted = lookup_mnt(path); 12091da177e4SLinus Torvalds if (mounted) { 12109393bd07SAl Viro dput(path->dentry); 12119393bd07SAl Viro mntput(path->mnt); 12129393bd07SAl Viro path->mnt = mounted; 12139393bd07SAl Viro path->dentry = dget(mounted->mnt_root); 12141da177e4SLinus Torvalds return 1; 12151da177e4SLinus Torvalds } 12161da177e4SLinus Torvalds return 0; 12171da177e4SLinus Torvalds } 12184d359507SAl Viro EXPORT_SYMBOL(follow_down_one); 12191da177e4SLinus Torvalds 1220b8faf035SNeilBrown static inline int managed_dentry_rcu(struct dentry *dentry) 122162a7375eSIan Kent { 1222b8faf035SNeilBrown return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ? 1223b8faf035SNeilBrown dentry->d_op->d_manage(dentry, true) : 0; 122462a7375eSIan Kent } 122562a7375eSIan Kent 12269875cf80SDavid Howells /* 1227287548e4SAl Viro * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if 1228287548e4SAl Viro * we meet a managed dentry that would need blocking. 12299875cf80SDavid Howells */ 12309875cf80SDavid Howells static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, 1231254cf582SAl Viro struct inode **inode, unsigned *seqp) 12329875cf80SDavid Howells { 123362a7375eSIan Kent for (;;) { 1234c7105365SAl Viro struct mount *mounted; 123562a7375eSIan Kent /* 123662a7375eSIan Kent * Don't forget we might have a non-mountpoint managed dentry 123762a7375eSIan Kent * that wants to block transit. 123862a7375eSIan Kent */ 1239b8faf035SNeilBrown switch (managed_dentry_rcu(path->dentry)) { 1240b8faf035SNeilBrown case -ECHILD: 1241b8faf035SNeilBrown default: 1242ab90911fSDavid Howells return false; 1243b8faf035SNeilBrown case -EISDIR: 1244b8faf035SNeilBrown return true; 1245b8faf035SNeilBrown case 0: 1246b8faf035SNeilBrown break; 1247b8faf035SNeilBrown } 124862a7375eSIan Kent 124962a7375eSIan Kent if (!d_mountpoint(path->dentry)) 1250b8faf035SNeilBrown return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); 125162a7375eSIan Kent 1252474279dcSAl Viro mounted = __lookup_mnt(path->mnt, path->dentry); 12539875cf80SDavid Howells if (!mounted) 12549875cf80SDavid Howells break; 1255c7105365SAl Viro path->mnt = &mounted->mnt; 1256c7105365SAl Viro path->dentry = mounted->mnt.mnt_root; 1257a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 1258254cf582SAl Viro *seqp = read_seqcount_begin(&path->dentry->d_seq); 125959430262SLinus Torvalds /* 126059430262SLinus Torvalds * Update the inode too. We don't need to re-check the 126159430262SLinus Torvalds * dentry sequence number here after this d_inode read, 126259430262SLinus Torvalds * because a mount-point is always pinned. 126359430262SLinus Torvalds */ 126459430262SLinus Torvalds *inode = path->dentry->d_inode; 12659875cf80SDavid Howells } 1266f5be3e29SAl Viro return !read_seqretry(&mount_lock, nd->m_seq) && 1267b8faf035SNeilBrown !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); 1268287548e4SAl Viro } 1269287548e4SAl Viro 127031e6b01fSNick Piggin static int follow_dotdot_rcu(struct nameidata *nd) 127131e6b01fSNick Piggin { 12724023bfc9SAl Viro struct inode *inode = nd->inode; 12737bd88377SAl Viro if (!nd->root.mnt) 127431e6b01fSNick Piggin set_root_rcu(nd); 127531e6b01fSNick Piggin 127631e6b01fSNick Piggin while (1) { 1277aed434adSAl Viro if (path_equal(&nd->path, &nd->root)) 127831e6b01fSNick Piggin break; 127931e6b01fSNick Piggin if (nd->path.dentry != nd->path.mnt->mnt_root) { 128031e6b01fSNick Piggin struct dentry *old = nd->path.dentry; 128131e6b01fSNick Piggin struct dentry *parent = old->d_parent; 128231e6b01fSNick Piggin unsigned seq; 128331e6b01fSNick Piggin 12844023bfc9SAl Viro inode = parent->d_inode; 128531e6b01fSNick Piggin seq = read_seqcount_begin(&parent->d_seq); 1286aed434adSAl Viro if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq))) 1287aed434adSAl Viro return -ECHILD; 128831e6b01fSNick Piggin nd->path.dentry = parent; 128931e6b01fSNick Piggin nd->seq = seq; 129031e6b01fSNick Piggin break; 1291aed434adSAl Viro } else { 1292aed434adSAl Viro struct mount *mnt = real_mount(nd->path.mnt); 1293aed434adSAl Viro struct mount *mparent = mnt->mnt_parent; 1294aed434adSAl Viro struct dentry *mountpoint = mnt->mnt_mountpoint; 1295aed434adSAl Viro struct inode *inode2 = mountpoint->d_inode; 1296aed434adSAl Viro unsigned seq = read_seqcount_begin(&mountpoint->d_seq); 1297aed434adSAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1298aed434adSAl Viro return -ECHILD; 1299aed434adSAl Viro if (&mparent->mnt == nd->path.mnt) 130031e6b01fSNick Piggin break; 1301aed434adSAl Viro /* we know that mountpoint was pinned */ 1302aed434adSAl Viro nd->path.dentry = mountpoint; 1303aed434adSAl Viro nd->path.mnt = &mparent->mnt; 1304aed434adSAl Viro inode = inode2; 1305aed434adSAl Viro nd->seq = seq; 130631e6b01fSNick Piggin } 1307aed434adSAl Viro } 1308aed434adSAl Viro while (unlikely(d_mountpoint(nd->path.dentry))) { 1309b37199e6SAl Viro struct mount *mounted; 1310b37199e6SAl Viro mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry); 1311aed434adSAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1312aed434adSAl Viro return -ECHILD; 1313b37199e6SAl Viro if (!mounted) 1314b37199e6SAl Viro break; 1315b37199e6SAl Viro nd->path.mnt = &mounted->mnt; 1316b37199e6SAl Viro nd->path.dentry = mounted->mnt.mnt_root; 13174023bfc9SAl Viro inode = nd->path.dentry->d_inode; 1318b37199e6SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 1319b37199e6SAl Viro } 13204023bfc9SAl Viro nd->inode = inode; 132131e6b01fSNick Piggin return 0; 132231e6b01fSNick Piggin } 132331e6b01fSNick Piggin 13249875cf80SDavid Howells /* 1325cc53ce53SDavid Howells * Follow down to the covering mount currently visible to userspace. At each 1326cc53ce53SDavid Howells * point, the filesystem owning that dentry may be queried as to whether the 1327cc53ce53SDavid Howells * caller is permitted to proceed or not. 1328cc53ce53SDavid Howells */ 13297cc90cc3SAl Viro int follow_down(struct path *path) 1330cc53ce53SDavid Howells { 1331cc53ce53SDavid Howells unsigned managed; 1332cc53ce53SDavid Howells int ret; 1333cc53ce53SDavid Howells 1334cc53ce53SDavid Howells while (managed = ACCESS_ONCE(path->dentry->d_flags), 1335cc53ce53SDavid Howells unlikely(managed & DCACHE_MANAGED_DENTRY)) { 1336cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1337cc53ce53SDavid Howells * being held. 1338cc53ce53SDavid Howells * 1339cc53ce53SDavid Howells * We indicate to the filesystem if someone is trying to mount 1340cc53ce53SDavid Howells * something here. This gives autofs the chance to deny anyone 1341cc53ce53SDavid Howells * other than its daemon the right to mount on its 1342cc53ce53SDavid Howells * superstructure. 1343cc53ce53SDavid Howells * 1344cc53ce53SDavid Howells * The filesystem may sleep at this point. 1345cc53ce53SDavid Howells */ 1346cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 1347cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1348cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 1349ab90911fSDavid Howells ret = path->dentry->d_op->d_manage( 13501aed3e42SAl Viro path->dentry, false); 1351cc53ce53SDavid Howells if (ret < 0) 1352cc53ce53SDavid Howells return ret == -EISDIR ? 0 : ret; 1353cc53ce53SDavid Howells } 1354cc53ce53SDavid Howells 1355cc53ce53SDavid Howells /* Transit to a mounted filesystem. */ 1356cc53ce53SDavid Howells if (managed & DCACHE_MOUNTED) { 1357cc53ce53SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 1358cc53ce53SDavid Howells if (!mounted) 1359cc53ce53SDavid Howells break; 1360cc53ce53SDavid Howells dput(path->dentry); 1361cc53ce53SDavid Howells mntput(path->mnt); 1362cc53ce53SDavid Howells path->mnt = mounted; 1363cc53ce53SDavid Howells path->dentry = dget(mounted->mnt_root); 1364cc53ce53SDavid Howells continue; 1365cc53ce53SDavid Howells } 1366cc53ce53SDavid Howells 1367cc53ce53SDavid Howells /* Don't handle automount points here */ 1368cc53ce53SDavid Howells break; 1369cc53ce53SDavid Howells } 1370cc53ce53SDavid Howells return 0; 1371cc53ce53SDavid Howells } 13724d359507SAl Viro EXPORT_SYMBOL(follow_down); 1373cc53ce53SDavid Howells 1374cc53ce53SDavid Howells /* 13759875cf80SDavid Howells * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() 13769875cf80SDavid Howells */ 13779875cf80SDavid Howells static void follow_mount(struct path *path) 13789875cf80SDavid Howells { 13799875cf80SDavid Howells while (d_mountpoint(path->dentry)) { 13809875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 13819875cf80SDavid Howells if (!mounted) 13829875cf80SDavid Howells break; 13839875cf80SDavid Howells dput(path->dentry); 13849875cf80SDavid Howells mntput(path->mnt); 13859875cf80SDavid Howells path->mnt = mounted; 13869875cf80SDavid Howells path->dentry = dget(mounted->mnt_root); 13879875cf80SDavid Howells } 13889875cf80SDavid Howells } 13899875cf80SDavid Howells 139031e6b01fSNick Piggin static void follow_dotdot(struct nameidata *nd) 13911da177e4SLinus Torvalds { 13927bd88377SAl Viro if (!nd->root.mnt) 13932a737871SAl Viro set_root(nd); 1394e518ddb7SAndreas Mohr 13951da177e4SLinus Torvalds while(1) { 13964ac91378SJan Blunck struct dentry *old = nd->path.dentry; 13971da177e4SLinus Torvalds 13982a737871SAl Viro if (nd->path.dentry == nd->root.dentry && 13992a737871SAl Viro nd->path.mnt == nd->root.mnt) { 14001da177e4SLinus Torvalds break; 14011da177e4SLinus Torvalds } 14024ac91378SJan Blunck if (nd->path.dentry != nd->path.mnt->mnt_root) { 14033088dd70SAl Viro /* rare case of legitimate dget_parent()... */ 14043088dd70SAl Viro nd->path.dentry = dget_parent(nd->path.dentry); 14051da177e4SLinus Torvalds dput(old); 14061da177e4SLinus Torvalds break; 14071da177e4SLinus Torvalds } 14083088dd70SAl Viro if (!follow_up(&nd->path)) 14091da177e4SLinus Torvalds break; 14101da177e4SLinus Torvalds } 141179ed0226SAl Viro follow_mount(&nd->path); 141231e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 14131da177e4SLinus Torvalds } 14141da177e4SLinus Torvalds 14151da177e4SLinus Torvalds /* 1416bad61189SMiklos Szeredi * This looks up the name in dcache, possibly revalidates the old dentry and 1417bad61189SMiklos Szeredi * allocates a new one if not found or not valid. In the need_lookup argument 1418bad61189SMiklos Szeredi * returns whether i_op->lookup is necessary. 1419bad61189SMiklos Szeredi * 1420bad61189SMiklos Szeredi * dir->d_inode->i_mutex must be held 1421baa03890SNick Piggin */ 1422bad61189SMiklos Szeredi static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir, 1423201f956eSAl Viro unsigned int flags, bool *need_lookup) 1424baa03890SNick Piggin { 1425baa03890SNick Piggin struct dentry *dentry; 1426bad61189SMiklos Szeredi int error; 1427baa03890SNick Piggin 1428bad61189SMiklos Szeredi *need_lookup = false; 1429bad61189SMiklos Szeredi dentry = d_lookup(dir, name); 1430bad61189SMiklos Szeredi if (dentry) { 143139e3c955SJeff Layton if (dentry->d_flags & DCACHE_OP_REVALIDATE) { 1432201f956eSAl Viro error = d_revalidate(dentry, flags); 1433bad61189SMiklos Szeredi if (unlikely(error <= 0)) { 1434bad61189SMiklos Szeredi if (error < 0) { 1435bad61189SMiklos Szeredi dput(dentry); 1436bad61189SMiklos Szeredi return ERR_PTR(error); 14375542aa2fSEric W. Biederman } else { 14385542aa2fSEric W. Biederman d_invalidate(dentry); 1439bad61189SMiklos Szeredi dput(dentry); 1440bad61189SMiklos Szeredi dentry = NULL; 1441bad61189SMiklos Szeredi } 1442bad61189SMiklos Szeredi } 1443bad61189SMiklos Szeredi } 1444bad61189SMiklos Szeredi } 1445baa03890SNick Piggin 1446bad61189SMiklos Szeredi if (!dentry) { 1447bad61189SMiklos Szeredi dentry = d_alloc(dir, name); 1448baa03890SNick Piggin if (unlikely(!dentry)) 1449baa03890SNick Piggin return ERR_PTR(-ENOMEM); 1450baa03890SNick Piggin 1451bad61189SMiklos Szeredi *need_lookup = true; 1452baa03890SNick Piggin } 1453baa03890SNick Piggin return dentry; 1454baa03890SNick Piggin } 1455baa03890SNick Piggin 1456baa03890SNick Piggin /* 145713a2c3beSJ. Bruce Fields * Call i_op->lookup on the dentry. The dentry must be negative and 145813a2c3beSJ. Bruce Fields * unhashed. 1459bad61189SMiklos Szeredi * 1460bad61189SMiklos Szeredi * dir->d_inode->i_mutex must be held 146144396f4bSJosef Bacik */ 1462bad61189SMiklos Szeredi static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry, 146372bd866aSAl Viro unsigned int flags) 146444396f4bSJosef Bacik { 146544396f4bSJosef Bacik struct dentry *old; 146644396f4bSJosef Bacik 146744396f4bSJosef Bacik /* Don't create child dentry for a dead directory. */ 1468bad61189SMiklos Szeredi if (unlikely(IS_DEADDIR(dir))) { 1469e188dc02SMiklos Szeredi dput(dentry); 147044396f4bSJosef Bacik return ERR_PTR(-ENOENT); 1471e188dc02SMiklos Szeredi } 147244396f4bSJosef Bacik 147372bd866aSAl Viro old = dir->i_op->lookup(dir, dentry, flags); 147444396f4bSJosef Bacik if (unlikely(old)) { 147544396f4bSJosef Bacik dput(dentry); 147644396f4bSJosef Bacik dentry = old; 147744396f4bSJosef Bacik } 147844396f4bSJosef Bacik return dentry; 147944396f4bSJosef Bacik } 148044396f4bSJosef Bacik 1481a3255546SAl Viro static struct dentry *__lookup_hash(struct qstr *name, 148272bd866aSAl Viro struct dentry *base, unsigned int flags) 1483a3255546SAl Viro { 1484bad61189SMiklos Szeredi bool need_lookup; 1485a3255546SAl Viro struct dentry *dentry; 1486a3255546SAl Viro 148772bd866aSAl Viro dentry = lookup_dcache(name, base, flags, &need_lookup); 1488bad61189SMiklos Szeredi if (!need_lookup) 1489a3255546SAl Viro return dentry; 1490bad61189SMiklos Szeredi 149172bd866aSAl Viro return lookup_real(base->d_inode, dentry, flags); 1492a3255546SAl Viro } 1493a3255546SAl Viro 149444396f4bSJosef Bacik /* 14951da177e4SLinus Torvalds * It's more convoluted than I'd like it to be, but... it's still fairly 14961da177e4SLinus Torvalds * small and for now I'd prefer to have fast path as straight as possible. 14971da177e4SLinus Torvalds * It _is_ time-critical. 14981da177e4SLinus Torvalds */ 1499e97cdc87SAl Viro static int lookup_fast(struct nameidata *nd, 1500254cf582SAl Viro struct path *path, struct inode **inode, 1501254cf582SAl Viro unsigned *seqp) 15021da177e4SLinus Torvalds { 15034ac91378SJan Blunck struct vfsmount *mnt = nd->path.mnt; 150431e6b01fSNick Piggin struct dentry *dentry, *parent = nd->path.dentry; 15055a18fff2SAl Viro int need_reval = 1; 15065a18fff2SAl Viro int status = 1; 15079875cf80SDavid Howells int err; 15089875cf80SDavid Howells 15093cac260aSAl Viro /* 1510b04f784eSNick Piggin * Rename seqlock is not required here because in the off chance 1511b04f784eSNick Piggin * of a false negative due to a concurrent rename, we're going to 1512b04f784eSNick Piggin * do the non-racy lookup, below. 1513b04f784eSNick Piggin */ 151431e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) { 151531e6b01fSNick Piggin unsigned seq; 1516766c4cbfSAl Viro bool negative; 1517da53be12SLinus Torvalds dentry = __d_lookup_rcu(parent, &nd->last, &seq); 15185a18fff2SAl Viro if (!dentry) 15195a18fff2SAl Viro goto unlazy; 15205a18fff2SAl Viro 152112f8ad4bSLinus Torvalds /* 152212f8ad4bSLinus Torvalds * This sequence count validates that the inode matches 152312f8ad4bSLinus Torvalds * the dentry name information from lookup. 152412f8ad4bSLinus Torvalds */ 152563afdfc7SDavid Howells *inode = d_backing_inode(dentry); 1526766c4cbfSAl Viro negative = d_is_negative(dentry); 152712f8ad4bSLinus Torvalds if (read_seqcount_retry(&dentry->d_seq, seq)) 152812f8ad4bSLinus Torvalds return -ECHILD; 1529766c4cbfSAl Viro if (negative) 1530766c4cbfSAl Viro return -ENOENT; 153112f8ad4bSLinus Torvalds 153212f8ad4bSLinus Torvalds /* 153312f8ad4bSLinus Torvalds * This sequence count validates that the parent had no 153412f8ad4bSLinus Torvalds * changes while we did the lookup of the dentry above. 153512f8ad4bSLinus Torvalds * 153612f8ad4bSLinus Torvalds * The memory barrier in read_seqcount_begin of child is 153712f8ad4bSLinus Torvalds * enough, we can use __read_seqcount_retry here. 153812f8ad4bSLinus Torvalds */ 153931e6b01fSNick Piggin if (__read_seqcount_retry(&parent->d_seq, nd->seq)) 154031e6b01fSNick Piggin return -ECHILD; 15415a18fff2SAl Viro 1542254cf582SAl Viro *seqp = seq; 154324643087SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) { 15444ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 15455a18fff2SAl Viro if (unlikely(status <= 0)) { 15465a18fff2SAl Viro if (status != -ECHILD) 15475a18fff2SAl Viro need_reval = 0; 15485a18fff2SAl Viro goto unlazy; 15495a18fff2SAl Viro } 155024643087SAl Viro } 155131e6b01fSNick Piggin path->mnt = mnt; 155231e6b01fSNick Piggin path->dentry = dentry; 1553254cf582SAl Viro if (likely(__follow_mount_rcu(nd, path, inode, seqp))) 15549875cf80SDavid Howells return 0; 15555a18fff2SAl Viro unlazy: 1556254cf582SAl Viro if (unlazy_walk(nd, dentry, seq)) 15575a18fff2SAl Viro return -ECHILD; 15585a18fff2SAl Viro } else { 1559e97cdc87SAl Viro dentry = __d_lookup(parent, &nd->last); 156024643087SAl Viro } 15615a18fff2SAl Viro 156281e6f520SAl Viro if (unlikely(!dentry)) 156381e6f520SAl Viro goto need_lookup; 15645a18fff2SAl Viro 15655a18fff2SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval) 15664ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 15675a18fff2SAl Viro if (unlikely(status <= 0)) { 15685a18fff2SAl Viro if (status < 0) { 15695a18fff2SAl Viro dput(dentry); 15705a18fff2SAl Viro return status; 15715a18fff2SAl Viro } 15725542aa2fSEric W. Biederman d_invalidate(dentry); 15735a18fff2SAl Viro dput(dentry); 157481e6f520SAl Viro goto need_lookup; 15755a18fff2SAl Viro } 1576697f514dSMiklos Szeredi 1577766c4cbfSAl Viro if (unlikely(d_is_negative(dentry))) { 1578766c4cbfSAl Viro dput(dentry); 1579766c4cbfSAl Viro return -ENOENT; 1580766c4cbfSAl Viro } 15811da177e4SLinus Torvalds path->mnt = mnt; 15821da177e4SLinus Torvalds path->dentry = dentry; 1583756daf26SNeilBrown err = follow_managed(path, nd); 15848402752eSAl Viro if (likely(!err)) 158563afdfc7SDavid Howells *inode = d_backing_inode(path->dentry); 15868402752eSAl Viro return err; 158781e6f520SAl Viro 158881e6f520SAl Viro need_lookup: 1589697f514dSMiklos Szeredi return 1; 1590697f514dSMiklos Szeredi } 1591697f514dSMiklos Szeredi 1592697f514dSMiklos Szeredi /* Fast lookup failed, do it the slow way */ 1593cc2a5271SAl Viro static int lookup_slow(struct nameidata *nd, struct path *path) 1594697f514dSMiklos Szeredi { 1595697f514dSMiklos Szeredi struct dentry *dentry, *parent; 1596697f514dSMiklos Szeredi 1597697f514dSMiklos Szeredi parent = nd->path.dentry; 159881e6f520SAl Viro BUG_ON(nd->inode != parent->d_inode); 159981e6f520SAl Viro 160081e6f520SAl Viro mutex_lock(&parent->d_inode->i_mutex); 1601cc2a5271SAl Viro dentry = __lookup_hash(&nd->last, parent, nd->flags); 160281e6f520SAl Viro mutex_unlock(&parent->d_inode->i_mutex); 160381e6f520SAl Viro if (IS_ERR(dentry)) 160481e6f520SAl Viro return PTR_ERR(dentry); 1605697f514dSMiklos Szeredi path->mnt = nd->path.mnt; 1606697f514dSMiklos Szeredi path->dentry = dentry; 16078402752eSAl Viro return follow_managed(path, nd); 16081da177e4SLinus Torvalds } 16091da177e4SLinus Torvalds 161052094c8aSAl Viro static inline int may_lookup(struct nameidata *nd) 161152094c8aSAl Viro { 161252094c8aSAl Viro if (nd->flags & LOOKUP_RCU) { 16134ad5abb3SAl Viro int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); 161452094c8aSAl Viro if (err != -ECHILD) 161552094c8aSAl Viro return err; 16166e9918b7SAl Viro if (unlazy_walk(nd, NULL, 0)) 161752094c8aSAl Viro return -ECHILD; 161852094c8aSAl Viro } 16194ad5abb3SAl Viro return inode_permission(nd->inode, MAY_EXEC); 162052094c8aSAl Viro } 162152094c8aSAl Viro 16229856fa1bSAl Viro static inline int handle_dots(struct nameidata *nd, int type) 16239856fa1bSAl Viro { 16249856fa1bSAl Viro if (type == LAST_DOTDOT) { 16259856fa1bSAl Viro if (nd->flags & LOOKUP_RCU) { 162670291aecSAl Viro return follow_dotdot_rcu(nd); 16279856fa1bSAl Viro } else 16289856fa1bSAl Viro follow_dotdot(nd); 16299856fa1bSAl Viro } 16309856fa1bSAl Viro return 0; 16319856fa1bSAl Viro } 16329856fa1bSAl Viro 1633181548c0SAl Viro static int pick_link(struct nameidata *nd, struct path *link, 1634181548c0SAl Viro struct inode *inode, unsigned seq) 1635d63ff28fSAl Viro { 1636626de996SAl Viro int error; 16371cf2665bSAl Viro struct saved *last; 1638756daf26SNeilBrown if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) { 1639626de996SAl Viro path_to_nameidata(link, nd); 1640626de996SAl Viro return -ELOOP; 1641626de996SAl Viro } 1642bc40aee0SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1643cd179f44SAl Viro if (link->mnt == nd->path.mnt) 1644cd179f44SAl Viro mntget(link->mnt); 16457973387aSAl Viro } 1646626de996SAl Viro error = nd_alloc_stack(nd); 1647626de996SAl Viro if (unlikely(error)) { 1648bc40aee0SAl Viro if (error == -ECHILD) { 1649bc40aee0SAl Viro if (unlikely(unlazy_link(nd, link, seq))) 1650bc40aee0SAl Viro return -ECHILD; 1651bc40aee0SAl Viro error = nd_alloc_stack(nd); 1652bc40aee0SAl Viro } 1653bc40aee0SAl Viro if (error) { 1654cd179f44SAl Viro path_put(link); 1655626de996SAl Viro return error; 1656626de996SAl Viro } 1657bc40aee0SAl Viro } 1658626de996SAl Viro 1659ab104923SAl Viro last = nd->stack + nd->depth++; 16601cf2665bSAl Viro last->link = *link; 1661ab104923SAl Viro last->cookie = NULL; 1662181548c0SAl Viro last->inode = inode; 16630450b2d1SAl Viro last->seq = seq; 1664d63ff28fSAl Viro return 1; 1665d63ff28fSAl Viro } 1666d63ff28fSAl Viro 16673ddcd056SLinus Torvalds /* 16683ddcd056SLinus Torvalds * Do we need to follow links? We _really_ want to be able 16693ddcd056SLinus Torvalds * to do this check without having to look at inode->i_op, 16703ddcd056SLinus Torvalds * so we keep a cache of "no, this doesn't need follow_link" 16713ddcd056SLinus Torvalds * for the common case. 16723ddcd056SLinus Torvalds */ 1673254cf582SAl Viro static inline int should_follow_link(struct nameidata *nd, struct path *link, 1674181548c0SAl Viro int follow, 1675181548c0SAl Viro struct inode *inode, unsigned seq) 16763ddcd056SLinus Torvalds { 1677d63ff28fSAl Viro if (likely(!d_is_symlink(link->dentry))) 1678d63ff28fSAl Viro return 0; 1679d63ff28fSAl Viro if (!follow) 1680d63ff28fSAl Viro return 0; 1681181548c0SAl Viro return pick_link(nd, link, inode, seq); 16823ddcd056SLinus Torvalds } 16833ddcd056SLinus Torvalds 16844693a547SAl Viro enum {WALK_GET = 1, WALK_PUT = 2}; 16854693a547SAl Viro 16864693a547SAl Viro static int walk_component(struct nameidata *nd, int flags) 1687ce57dfc1SAl Viro { 1688caa85634SAl Viro struct path path; 1689ce57dfc1SAl Viro struct inode *inode; 1690254cf582SAl Viro unsigned seq; 1691ce57dfc1SAl Viro int err; 1692ce57dfc1SAl Viro /* 1693ce57dfc1SAl Viro * "." and ".." are special - ".." especially so because it has 1694ce57dfc1SAl Viro * to be able to know about the current root directory and 1695ce57dfc1SAl Viro * parent relationships. 1696ce57dfc1SAl Viro */ 16974693a547SAl Viro if (unlikely(nd->last_type != LAST_NORM)) { 16984693a547SAl Viro err = handle_dots(nd, nd->last_type); 16994693a547SAl Viro if (flags & WALK_PUT) 17004693a547SAl Viro put_link(nd); 17014693a547SAl Viro return err; 17024693a547SAl Viro } 1703254cf582SAl Viro err = lookup_fast(nd, &path, &inode, &seq); 1704ce57dfc1SAl Viro if (unlikely(err)) { 1705697f514dSMiklos Szeredi if (err < 0) 1706f0a9ba70SAl Viro return err; 1707697f514dSMiklos Szeredi 1708caa85634SAl Viro err = lookup_slow(nd, &path); 1709697f514dSMiklos Szeredi if (err < 0) 1710f0a9ba70SAl Viro return err; 1711697f514dSMiklos Szeredi 171263afdfc7SDavid Howells inode = d_backing_inode(path.dentry); 1713254cf582SAl Viro seq = 0; /* we are already out of RCU mode */ 1714697f514dSMiklos Szeredi err = -ENOENT; 1715caa85634SAl Viro if (d_is_negative(path.dentry)) 1716697f514dSMiklos Szeredi goto out_path_put; 1717766c4cbfSAl Viro } 1718697f514dSMiklos Szeredi 17194693a547SAl Viro if (flags & WALK_PUT) 17204693a547SAl Viro put_link(nd); 1721181548c0SAl Viro err = should_follow_link(nd, &path, flags & WALK_GET, inode, seq); 1722d63ff28fSAl Viro if (unlikely(err)) 1723d63ff28fSAl Viro return err; 1724caa85634SAl Viro path_to_nameidata(&path, nd); 1725ce57dfc1SAl Viro nd->inode = inode; 1726254cf582SAl Viro nd->seq = seq; 1727ce57dfc1SAl Viro return 0; 1728697f514dSMiklos Szeredi 1729697f514dSMiklos Szeredi out_path_put: 1730caa85634SAl Viro path_to_nameidata(&path, nd); 1731697f514dSMiklos Szeredi return err; 1732ce57dfc1SAl Viro } 1733ce57dfc1SAl Viro 17341da177e4SLinus Torvalds /* 1735bfcfaa77SLinus Torvalds * We can do the critical dentry name comparison and hashing 1736bfcfaa77SLinus Torvalds * operations one word at a time, but we are limited to: 1737bfcfaa77SLinus Torvalds * 1738bfcfaa77SLinus Torvalds * - Architectures with fast unaligned word accesses. We could 1739bfcfaa77SLinus Torvalds * do a "get_unaligned()" if this helps and is sufficiently 1740bfcfaa77SLinus Torvalds * fast. 1741bfcfaa77SLinus Torvalds * 1742bfcfaa77SLinus Torvalds * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we 1743bfcfaa77SLinus Torvalds * do not trap on the (extremely unlikely) case of a page 1744bfcfaa77SLinus Torvalds * crossing operation. 1745bfcfaa77SLinus Torvalds * 1746bfcfaa77SLinus Torvalds * - Furthermore, we need an efficient 64-bit compile for the 1747bfcfaa77SLinus Torvalds * 64-bit case in order to generate the "number of bytes in 1748bfcfaa77SLinus Torvalds * the final mask". Again, that could be replaced with a 1749bfcfaa77SLinus Torvalds * efficient population count instruction or similar. 1750bfcfaa77SLinus Torvalds */ 1751bfcfaa77SLinus Torvalds #ifdef CONFIG_DCACHE_WORD_ACCESS 1752bfcfaa77SLinus Torvalds 1753f68e556eSLinus Torvalds #include <asm/word-at-a-time.h> 1754bfcfaa77SLinus Torvalds 1755f68e556eSLinus Torvalds #ifdef CONFIG_64BIT 1756bfcfaa77SLinus Torvalds 1757bfcfaa77SLinus Torvalds static inline unsigned int fold_hash(unsigned long hash) 1758bfcfaa77SLinus Torvalds { 175999d263d4SLinus Torvalds return hash_64(hash, 32); 1760bfcfaa77SLinus Torvalds } 1761bfcfaa77SLinus Torvalds 1762bfcfaa77SLinus Torvalds #else /* 32-bit case */ 1763bfcfaa77SLinus Torvalds 1764bfcfaa77SLinus Torvalds #define fold_hash(x) (x) 1765bfcfaa77SLinus Torvalds 1766bfcfaa77SLinus Torvalds #endif 1767bfcfaa77SLinus Torvalds 1768bfcfaa77SLinus Torvalds unsigned int full_name_hash(const unsigned char *name, unsigned int len) 1769bfcfaa77SLinus Torvalds { 1770bfcfaa77SLinus Torvalds unsigned long a, mask; 1771bfcfaa77SLinus Torvalds unsigned long hash = 0; 1772bfcfaa77SLinus Torvalds 1773bfcfaa77SLinus Torvalds for (;;) { 1774e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name); 1775bfcfaa77SLinus Torvalds if (len < sizeof(unsigned long)) 1776bfcfaa77SLinus Torvalds break; 1777bfcfaa77SLinus Torvalds hash += a; 1778f132c5beSAl Viro hash *= 9; 1779bfcfaa77SLinus Torvalds name += sizeof(unsigned long); 1780bfcfaa77SLinus Torvalds len -= sizeof(unsigned long); 1781bfcfaa77SLinus Torvalds if (!len) 1782bfcfaa77SLinus Torvalds goto done; 1783bfcfaa77SLinus Torvalds } 1784a5c21dceSWill Deacon mask = bytemask_from_count(len); 1785bfcfaa77SLinus Torvalds hash += mask & a; 1786bfcfaa77SLinus Torvalds done: 1787bfcfaa77SLinus Torvalds return fold_hash(hash); 1788bfcfaa77SLinus Torvalds } 1789bfcfaa77SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 1790bfcfaa77SLinus Torvalds 1791bfcfaa77SLinus Torvalds /* 1792bfcfaa77SLinus Torvalds * Calculate the length and hash of the path component, and 1793d6bb3e90SLinus Torvalds * return the "hash_len" as the result. 1794bfcfaa77SLinus Torvalds */ 1795d6bb3e90SLinus Torvalds static inline u64 hash_name(const char *name) 1796bfcfaa77SLinus Torvalds { 179736126f8fSLinus Torvalds unsigned long a, b, adata, bdata, mask, hash, len; 179836126f8fSLinus Torvalds const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 1799bfcfaa77SLinus Torvalds 1800bfcfaa77SLinus Torvalds hash = a = 0; 1801bfcfaa77SLinus Torvalds len = -sizeof(unsigned long); 1802bfcfaa77SLinus Torvalds do { 1803bfcfaa77SLinus Torvalds hash = (hash + a) * 9; 1804bfcfaa77SLinus Torvalds len += sizeof(unsigned long); 1805e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name+len); 180636126f8fSLinus Torvalds b = a ^ REPEAT_BYTE('/'); 180736126f8fSLinus Torvalds } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); 1808bfcfaa77SLinus Torvalds 180936126f8fSLinus Torvalds adata = prep_zero_mask(a, adata, &constants); 181036126f8fSLinus Torvalds bdata = prep_zero_mask(b, bdata, &constants); 181136126f8fSLinus Torvalds 181236126f8fSLinus Torvalds mask = create_zero_mask(adata | bdata); 181336126f8fSLinus Torvalds 181436126f8fSLinus Torvalds hash += a & zero_bytemask(mask); 18159226b5b4SLinus Torvalds len += find_zero(mask); 1816d6bb3e90SLinus Torvalds return hashlen_create(fold_hash(hash), len); 1817bfcfaa77SLinus Torvalds } 1818bfcfaa77SLinus Torvalds 1819bfcfaa77SLinus Torvalds #else 1820bfcfaa77SLinus Torvalds 18210145acc2SLinus Torvalds unsigned int full_name_hash(const unsigned char *name, unsigned int len) 18220145acc2SLinus Torvalds { 18230145acc2SLinus Torvalds unsigned long hash = init_name_hash(); 18240145acc2SLinus Torvalds while (len--) 18250145acc2SLinus Torvalds hash = partial_name_hash(*name++, hash); 18260145acc2SLinus Torvalds return end_name_hash(hash); 18270145acc2SLinus Torvalds } 1828ae942ae7SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 18290145acc2SLinus Torvalds 18303ddcd056SLinus Torvalds /* 1831200e9ef7SLinus Torvalds * We know there's a real path component here of at least 1832200e9ef7SLinus Torvalds * one character. 1833200e9ef7SLinus Torvalds */ 1834d6bb3e90SLinus Torvalds static inline u64 hash_name(const char *name) 1835200e9ef7SLinus Torvalds { 1836200e9ef7SLinus Torvalds unsigned long hash = init_name_hash(); 1837200e9ef7SLinus Torvalds unsigned long len = 0, c; 1838200e9ef7SLinus Torvalds 1839200e9ef7SLinus Torvalds c = (unsigned char)*name; 1840200e9ef7SLinus Torvalds do { 1841200e9ef7SLinus Torvalds len++; 1842200e9ef7SLinus Torvalds hash = partial_name_hash(c, hash); 1843200e9ef7SLinus Torvalds c = (unsigned char)name[len]; 1844200e9ef7SLinus Torvalds } while (c && c != '/'); 1845d6bb3e90SLinus Torvalds return hashlen_create(end_name_hash(hash), len); 1846200e9ef7SLinus Torvalds } 1847200e9ef7SLinus Torvalds 1848bfcfaa77SLinus Torvalds #endif 1849bfcfaa77SLinus Torvalds 1850200e9ef7SLinus Torvalds /* 18511da177e4SLinus Torvalds * Name resolution. 1852ea3834d9SPrasanna Meda * This is the basic name resolution function, turning a pathname into 1853ea3834d9SPrasanna Meda * the final dentry. We expect 'base' to be positive and a directory. 18541da177e4SLinus Torvalds * 1855ea3834d9SPrasanna Meda * Returns 0 and nd will have valid dentry and mnt on success. 1856ea3834d9SPrasanna Meda * Returns error and drops reference to input namei data on failure. 18571da177e4SLinus Torvalds */ 18586de88d72SAl Viro static int link_path_walk(const char *name, struct nameidata *nd) 18591da177e4SLinus Torvalds { 18601da177e4SLinus Torvalds int err; 18611da177e4SLinus Torvalds 18621da177e4SLinus Torvalds while (*name=='/') 18631da177e4SLinus Torvalds name++; 18641da177e4SLinus Torvalds if (!*name) 18659e18f10aSAl Viro return 0; 18661da177e4SLinus Torvalds 18671da177e4SLinus Torvalds /* At this point we know we have a real path component. */ 18681da177e4SLinus Torvalds for(;;) { 1869d6bb3e90SLinus Torvalds u64 hash_len; 1870fe479a58SAl Viro int type; 18711da177e4SLinus Torvalds 187252094c8aSAl Viro err = may_lookup(nd); 18731da177e4SLinus Torvalds if (err) 18743595e234SAl Viro return err; 18751da177e4SLinus Torvalds 1876d6bb3e90SLinus Torvalds hash_len = hash_name(name); 18771da177e4SLinus Torvalds 1878fe479a58SAl Viro type = LAST_NORM; 1879d6bb3e90SLinus Torvalds if (name[0] == '.') switch (hashlen_len(hash_len)) { 1880fe479a58SAl Viro case 2: 1881200e9ef7SLinus Torvalds if (name[1] == '.') { 1882fe479a58SAl Viro type = LAST_DOTDOT; 188316c2cd71SAl Viro nd->flags |= LOOKUP_JUMPED; 188416c2cd71SAl Viro } 1885fe479a58SAl Viro break; 1886fe479a58SAl Viro case 1: 1887fe479a58SAl Viro type = LAST_DOT; 1888fe479a58SAl Viro } 18895a202bcdSAl Viro if (likely(type == LAST_NORM)) { 18905a202bcdSAl Viro struct dentry *parent = nd->path.dentry; 189116c2cd71SAl Viro nd->flags &= ~LOOKUP_JUMPED; 18925a202bcdSAl Viro if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { 1893a060dc50SJames Hogan struct qstr this = { { .hash_len = hash_len }, .name = name }; 1894da53be12SLinus Torvalds err = parent->d_op->d_hash(parent, &this); 18955a202bcdSAl Viro if (err < 0) 18963595e234SAl Viro return err; 1897d6bb3e90SLinus Torvalds hash_len = this.hash_len; 1898d6bb3e90SLinus Torvalds name = this.name; 18995a202bcdSAl Viro } 19005a202bcdSAl Viro } 1901fe479a58SAl Viro 1902d6bb3e90SLinus Torvalds nd->last.hash_len = hash_len; 1903d6bb3e90SLinus Torvalds nd->last.name = name; 19045f4a6a69SAl Viro nd->last_type = type; 19055f4a6a69SAl Viro 1906d6bb3e90SLinus Torvalds name += hashlen_len(hash_len); 1907d6bb3e90SLinus Torvalds if (!*name) 1908bdf6cbf1SAl Viro goto OK; 1909200e9ef7SLinus Torvalds /* 1910200e9ef7SLinus Torvalds * If it wasn't NUL, we know it was '/'. Skip that 1911200e9ef7SLinus Torvalds * slash, and continue until no more slashes. 1912200e9ef7SLinus Torvalds */ 1913200e9ef7SLinus Torvalds do { 1914d6bb3e90SLinus Torvalds name++; 1915d6bb3e90SLinus Torvalds } while (unlikely(*name == '/')); 19168620c238SAl Viro if (unlikely(!*name)) { 19178620c238SAl Viro OK: 1918368ee9baSAl Viro /* pathname body, done */ 19198620c238SAl Viro if (!nd->depth) 19208620c238SAl Viro return 0; 19218620c238SAl Viro name = nd->stack[nd->depth - 1].name; 1922368ee9baSAl Viro /* trailing symlink, done */ 19238620c238SAl Viro if (!name) 19248620c238SAl Viro return 0; 19258620c238SAl Viro /* last component of nested symlink */ 19264693a547SAl Viro err = walk_component(nd, WALK_GET | WALK_PUT); 19278620c238SAl Viro } else { 19284693a547SAl Viro err = walk_component(nd, WALK_GET); 19298620c238SAl Viro } 1930ce57dfc1SAl Viro if (err < 0) 19313595e234SAl Viro return err; 1932fe479a58SAl Viro 1933ce57dfc1SAl Viro if (err) { 1934626de996SAl Viro const char *s = get_link(nd); 19355a460275SAl Viro 19363595e234SAl Viro if (unlikely(IS_ERR(s))) 19373595e234SAl Viro return PTR_ERR(s); 1938172a39a0SAl Viro err = 0; 193912b09578SAl Viro if (unlikely(!s)) { 194012b09578SAl Viro /* jumped */ 1941b9ff4429SAl Viro put_link(nd); 194212b09578SAl Viro } else { 1943071bf501SAl Viro nd->stack[nd->depth - 1].name = name; 194432cd7468SAl Viro name = s; 19459e18f10aSAl Viro continue; 194648c8b0c5SAl Viro } 194731e6b01fSNick Piggin } 19483595e234SAl Viro if (unlikely(!d_can_lookup(nd->path.dentry))) 19493595e234SAl Viro return -ENOTDIR; 19505f4a6a69SAl Viro } 1951ce57dfc1SAl Viro } 19521da177e4SLinus Torvalds 1953368ee9baSAl Viro static const char *path_init(int dfd, const struct filename *name, 1954368ee9baSAl Viro unsigned int flags, struct nameidata *nd) 195531e6b01fSNick Piggin { 195631e6b01fSNick Piggin int retval = 0; 1957fd2f7cb5SAl Viro const char *s = name->name; 195831e6b01fSNick Piggin 195931e6b01fSNick Piggin nd->last_type = LAST_ROOT; /* if there are only slashes... */ 1960980f3ea2SAl Viro nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT; 196131e6b01fSNick Piggin nd->depth = 0; 1962368ee9baSAl Viro nd->total_link_count = 0; 19635b6ca027SAl Viro if (flags & LOOKUP_ROOT) { 1964b18825a7SDavid Howells struct dentry *root = nd->root.dentry; 1965b18825a7SDavid Howells struct inode *inode = root->d_inode; 1966fd2f7cb5SAl Viro if (*s) { 196744b1d530SMiklos Szeredi if (!d_can_lookup(root)) 1968368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 19695b6ca027SAl Viro retval = inode_permission(inode, MAY_EXEC); 19705b6ca027SAl Viro if (retval) 1971368ee9baSAl Viro return ERR_PTR(retval); 197273d049a4SAl Viro } 19735b6ca027SAl Viro nd->path = nd->root; 19745b6ca027SAl Viro nd->inode = inode; 19755b6ca027SAl Viro if (flags & LOOKUP_RCU) { 19768b61e74fSAl Viro rcu_read_lock(); 19775b6ca027SAl Viro nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 19788f47a016SAl Viro nd->root_seq = nd->seq; 197948a066e7SAl Viro nd->m_seq = read_seqbegin(&mount_lock); 19805b6ca027SAl Viro } else { 19815b6ca027SAl Viro path_get(&nd->path); 19825b6ca027SAl Viro } 1983368ee9baSAl Viro return s; 19845b6ca027SAl Viro } 19855b6ca027SAl Viro 198631e6b01fSNick Piggin nd->root.mnt = NULL; 198731e6b01fSNick Piggin 198848a066e7SAl Viro nd->m_seq = read_seqbegin(&mount_lock); 1989fd2f7cb5SAl Viro if (*s == '/') { 1990e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 19918b61e74fSAl Viro rcu_read_lock(); 19927bd88377SAl Viro nd->seq = set_root_rcu(nd); 1993e41f7d4eSAl Viro } else { 1994e41f7d4eSAl Viro set_root(nd); 1995e41f7d4eSAl Viro path_get(&nd->root); 1996e41f7d4eSAl Viro } 199731e6b01fSNick Piggin nd->path = nd->root; 199831e6b01fSNick Piggin } else if (dfd == AT_FDCWD) { 1999e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 200031e6b01fSNick Piggin struct fs_struct *fs = current->fs; 2001c28cc364SNick Piggin unsigned seq; 200231e6b01fSNick Piggin 20038b61e74fSAl Viro rcu_read_lock(); 200431e6b01fSNick Piggin 2005c28cc364SNick Piggin do { 2006c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 200731e6b01fSNick Piggin nd->path = fs->pwd; 2008c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 2009c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 2010e41f7d4eSAl Viro } else { 2011e41f7d4eSAl Viro get_fs_pwd(current->fs, &nd->path); 2012e41f7d4eSAl Viro } 201331e6b01fSNick Piggin } else { 2014582aa64aSJeff Layton /* Caller must check execute permissions on the starting path component */ 20152903ff01SAl Viro struct fd f = fdget_raw(dfd); 201631e6b01fSNick Piggin struct dentry *dentry; 201731e6b01fSNick Piggin 20182903ff01SAl Viro if (!f.file) 2019368ee9baSAl Viro return ERR_PTR(-EBADF); 202031e6b01fSNick Piggin 20212903ff01SAl Viro dentry = f.file->f_path.dentry; 202231e6b01fSNick Piggin 2023fd2f7cb5SAl Viro if (*s) { 202444b1d530SMiklos Szeredi if (!d_can_lookup(dentry)) { 20252903ff01SAl Viro fdput(f); 2026368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 2027f52e0c11SAl Viro } 20282903ff01SAl Viro } 20292903ff01SAl Viro 20302903ff01SAl Viro nd->path = f.file->f_path; 2031e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 20328b61e74fSAl Viro rcu_read_lock(); 203334a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 203434a26b99SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 20355590ff0dSUlrich Drepper } else { 20362903ff01SAl Viro path_get(&nd->path); 203734a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 20381da177e4SLinus Torvalds } 203934a26b99SAl Viro fdput(f); 2040368ee9baSAl Viro return s; 2041e41f7d4eSAl Viro } 2042e41f7d4eSAl Viro 204331e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 20444023bfc9SAl Viro if (!(flags & LOOKUP_RCU)) 2045368ee9baSAl Viro return s; 20464023bfc9SAl Viro if (likely(!read_seqcount_retry(&nd->path.dentry->d_seq, nd->seq))) 2047368ee9baSAl Viro return s; 20484023bfc9SAl Viro if (!(nd->flags & LOOKUP_ROOT)) 20494023bfc9SAl Viro nd->root.mnt = NULL; 20504023bfc9SAl Viro rcu_read_unlock(); 2051368ee9baSAl Viro return ERR_PTR(-ECHILD); 20529b4a9b14SAl Viro } 20539b4a9b14SAl Viro 2054893b7775SAl Viro static void path_cleanup(struct nameidata *nd) 2055893b7775SAl Viro { 2056893b7775SAl Viro if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 2057893b7775SAl Viro path_put(&nd->root); 2058893b7775SAl Viro nd->root.mnt = NULL; 2059893b7775SAl Viro } 20609b4a9b14SAl Viro } 20619b4a9b14SAl Viro 20623bdba28bSAl Viro static const char *trailing_symlink(struct nameidata *nd) 206395fa25d9SAl Viro { 206495fa25d9SAl Viro const char *s; 2065fec2fa24SAl Viro int error = may_follow_link(nd); 2066deb106c6SAl Viro if (unlikely(error)) 20673bdba28bSAl Viro return ERR_PTR(error); 206895fa25d9SAl Viro nd->flags |= LOOKUP_PARENT; 2069fab51e8aSAl Viro nd->stack[0].name = NULL; 20703b2e7f75SAl Viro s = get_link(nd); 2071deb106c6SAl Viro return s ? s : ""; 207295fa25d9SAl Viro } 207395fa25d9SAl Viro 2074caa85634SAl Viro static inline int lookup_last(struct nameidata *nd) 2075bd92d7feSAl Viro { 2076bd92d7feSAl Viro if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) 2077bd92d7feSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 2078bd92d7feSAl Viro 2079bd92d7feSAl Viro nd->flags &= ~LOOKUP_PARENT; 2080deb106c6SAl Viro return walk_component(nd, 20814693a547SAl Viro nd->flags & LOOKUP_FOLLOW 20824693a547SAl Viro ? nd->depth 20834693a547SAl Viro ? WALK_PUT | WALK_GET 20844693a547SAl Viro : WALK_GET 20854693a547SAl Viro : 0); 2086bd92d7feSAl Viro } 2087bd92d7feSAl Viro 20889b4a9b14SAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2089625b6d10SAl Viro static int path_lookupat(int dfd, const struct filename *name, unsigned flags, 2090625b6d10SAl Viro struct nameidata *nd, struct path *path) 20919b4a9b14SAl Viro { 2092368ee9baSAl Viro const char *s = path_init(dfd, name, flags, nd); 2093bd92d7feSAl Viro int err; 209431e6b01fSNick Piggin 2095368ee9baSAl Viro if (IS_ERR(s)) 2096368ee9baSAl Viro return PTR_ERR(s); 20973bdba28bSAl Viro while (!(err = link_path_walk(s, nd)) 20983bdba28bSAl Viro && ((err = lookup_last(nd)) > 0)) { 20993bdba28bSAl Viro s = trailing_symlink(nd); 21003bdba28bSAl Viro if (IS_ERR(s)) { 21013bdba28bSAl Viro err = PTR_ERR(s); 21026d7b5aaeSAl Viro break; 2103bd92d7feSAl Viro } 2104bd92d7feSAl Viro } 21059f1fafeeSAl Viro if (!err) 21069f1fafeeSAl Viro err = complete_walk(nd); 2107bd92d7feSAl Viro 2108deb106c6SAl Viro if (!err && nd->flags & LOOKUP_DIRECTORY) 2109deb106c6SAl Viro if (!d_can_lookup(nd->path.dentry)) 2110bd23a539SAl Viro err = -ENOTDIR; 2111625b6d10SAl Viro if (!err) { 2112625b6d10SAl Viro *path = nd->path; 2113625b6d10SAl Viro nd->path.mnt = NULL; 2114625b6d10SAl Viro nd->path.dentry = NULL; 2115625b6d10SAl Viro } 2116deb106c6SAl Viro terminate_walk(nd); 2117893b7775SAl Viro path_cleanup(nd); 2118bd92d7feSAl Viro return err; 211931e6b01fSNick Piggin } 212031e6b01fSNick Piggin 2121625b6d10SAl Viro static int filename_lookup(int dfd, struct filename *name, unsigned flags, 21229ad1aaa6SAl Viro struct path *path, struct path *root) 2123873f1eedSJeff Layton { 2124894bc8c4SAl Viro int retval; 21259ad1aaa6SAl Viro struct nameidata nd, *saved_nd = set_nameidata(&nd); 21269ad1aaa6SAl Viro if (unlikely(root)) { 21279ad1aaa6SAl Viro nd.root = *root; 21289ad1aaa6SAl Viro flags |= LOOKUP_ROOT; 21299ad1aaa6SAl Viro } 21309ad1aaa6SAl Viro retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, &nd, path); 2131873f1eedSJeff Layton if (unlikely(retval == -ECHILD)) 21329ad1aaa6SAl Viro retval = path_lookupat(dfd, name, flags, &nd, path); 2133873f1eedSJeff Layton if (unlikely(retval == -ESTALE)) 2134625b6d10SAl Viro retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, 21359ad1aaa6SAl Viro &nd, path); 2136873f1eedSJeff Layton 2137873f1eedSJeff Layton if (likely(!retval)) 2138625b6d10SAl Viro audit_inode(name, path->dentry, flags & LOOKUP_PARENT); 2139756daf26SNeilBrown restore_nameidata(saved_nd); 2140e4bd1c1aSAl Viro putname(name); 2141873f1eedSJeff Layton return retval; 2142873f1eedSJeff Layton } 2143873f1eedSJeff Layton 21448bcb77faSAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 21458bcb77faSAl Viro static int path_parentat(int dfd, const struct filename *name, 21468bcb77faSAl Viro unsigned int flags, struct nameidata *nd) 21478bcb77faSAl Viro { 2148368ee9baSAl Viro const char *s = path_init(dfd, name, flags, nd); 2149368ee9baSAl Viro int err; 2150368ee9baSAl Viro if (IS_ERR(s)) 2151368ee9baSAl Viro return PTR_ERR(s); 2152368ee9baSAl Viro err = link_path_walk(s, nd); 21538bcb77faSAl Viro if (!err) 21548bcb77faSAl Viro err = complete_walk(nd); 2155deb106c6SAl Viro if (err) 2156deb106c6SAl Viro terminate_walk(nd); 21578bcb77faSAl Viro path_cleanup(nd); 21588bcb77faSAl Viro return err; 21598bcb77faSAl Viro } 21608bcb77faSAl Viro 21618bcb77faSAl Viro static int filename_parentat(int dfd, struct filename *name, 21628bcb77faSAl Viro unsigned int flags, struct nameidata *nd) 21638bcb77faSAl Viro { 21648bcb77faSAl Viro int retval; 21658bcb77faSAl Viro struct nameidata *saved_nd = set_nameidata(nd); 21668bcb77faSAl Viro 21678bcb77faSAl Viro retval = path_parentat(dfd, name, flags | LOOKUP_RCU, nd); 21688bcb77faSAl Viro if (unlikely(retval == -ECHILD)) 21698bcb77faSAl Viro retval = path_parentat(dfd, name, flags, nd); 21708bcb77faSAl Viro if (unlikely(retval == -ESTALE)) 21718bcb77faSAl Viro retval = path_parentat(dfd, name, flags | LOOKUP_REVAL, nd); 21728bcb77faSAl Viro 21738bcb77faSAl Viro if (likely(!retval)) 21748bcb77faSAl Viro audit_inode(name, nd->path.dentry, LOOKUP_PARENT); 21758bcb77faSAl Viro restore_nameidata(saved_nd); 21768bcb77faSAl Viro return retval; 21778bcb77faSAl Viro } 21788bcb77faSAl Viro 217979714f72SAl Viro /* does lookup, returns the object with parent locked */ 218079714f72SAl Viro struct dentry *kern_path_locked(const char *name, struct path *path) 21815590ff0dSUlrich Drepper { 218251689104SPaul Moore struct filename *filename = getname_kernel(name); 218379714f72SAl Viro struct nameidata nd; 218479714f72SAl Viro struct dentry *d; 218551689104SPaul Moore int err; 218651689104SPaul Moore 218751689104SPaul Moore if (IS_ERR(filename)) 218851689104SPaul Moore return ERR_CAST(filename); 218951689104SPaul Moore 21908bcb77faSAl Viro err = filename_parentat(AT_FDCWD, filename, 0, &nd); 219151689104SPaul Moore if (err) { 219251689104SPaul Moore d = ERR_PTR(err); 219351689104SPaul Moore goto out; 219451689104SPaul Moore } 219579714f72SAl Viro if (nd.last_type != LAST_NORM) { 219679714f72SAl Viro path_put(&nd.path); 219751689104SPaul Moore d = ERR_PTR(-EINVAL); 219851689104SPaul Moore goto out; 219979714f72SAl Viro } 220079714f72SAl Viro mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 22011e0ea001SAl Viro d = __lookup_hash(&nd.last, nd.path.dentry, 0); 220279714f72SAl Viro if (IS_ERR(d)) { 220379714f72SAl Viro mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 220479714f72SAl Viro path_put(&nd.path); 220551689104SPaul Moore goto out; 220679714f72SAl Viro } 220779714f72SAl Viro *path = nd.path; 220851689104SPaul Moore out: 220951689104SPaul Moore putname(filename); 221079714f72SAl Viro return d; 22115590ff0dSUlrich Drepper } 22125590ff0dSUlrich Drepper 2213d1811465SAl Viro int kern_path(const char *name, unsigned int flags, struct path *path) 2214d1811465SAl Viro { 221574eb8cc5SAl Viro struct filename *filename = getname_kernel(name); 2216e4bd1c1aSAl Viro if (IS_ERR(filename)) 2217e4bd1c1aSAl Viro return PTR_ERR(filename); 22189ad1aaa6SAl Viro return filename_lookup(AT_FDCWD, filename, flags, path, NULL); 2219d1811465SAl Viro } 22204d359507SAl Viro EXPORT_SYMBOL(kern_path); 2221d1811465SAl Viro 222216f18200SJosef 'Jeff' Sipek /** 222316f18200SJosef 'Jeff' Sipek * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair 222416f18200SJosef 'Jeff' Sipek * @dentry: pointer to dentry of the base directory 222516f18200SJosef 'Jeff' Sipek * @mnt: pointer to vfs mount of the base directory 222616f18200SJosef 'Jeff' Sipek * @name: pointer to file name 222716f18200SJosef 'Jeff' Sipek * @flags: lookup flags 2228e0a01249SAl Viro * @path: pointer to struct path to fill 222916f18200SJosef 'Jeff' Sipek */ 223016f18200SJosef 'Jeff' Sipek int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, 223116f18200SJosef 'Jeff' Sipek const char *name, unsigned int flags, 2232e0a01249SAl Viro struct path *path) 223316f18200SJosef 'Jeff' Sipek { 22349ad1aaa6SAl Viro struct path root = {.mnt = mnt, .dentry = dentry}; 223574eb8cc5SAl Viro struct filename *filename = getname_kernel(name); 223674eb8cc5SAl Viro 223774eb8cc5SAl Viro BUG_ON(flags & LOOKUP_PARENT); 223874eb8cc5SAl Viro 2239e4bd1c1aSAl Viro if (IS_ERR(filename)) 2240e4bd1c1aSAl Viro return PTR_ERR(filename); 2241e4bd1c1aSAl Viro 22429ad1aaa6SAl Viro /* the first argument of filename_lookup() is ignored with root */ 22439ad1aaa6SAl Viro return filename_lookup(AT_FDCWD, filename, flags , path, &root); 224416f18200SJosef 'Jeff' Sipek } 22454d359507SAl Viro EXPORT_SYMBOL(vfs_path_lookup); 224616f18200SJosef 'Jeff' Sipek 2247eead1911SChristoph Hellwig /** 2248a6b91919SRandy Dunlap * lookup_one_len - filesystem helper to lookup single pathname component 2249eead1911SChristoph Hellwig * @name: pathname component to lookup 2250eead1911SChristoph Hellwig * @base: base directory to lookup from 2251eead1911SChristoph Hellwig * @len: maximum length @len should be interpreted to 2252eead1911SChristoph Hellwig * 2253a6b91919SRandy Dunlap * Note that this routine is purely a helper for filesystem usage and should 22549e7543e9SAl Viro * not be called by generic code. 2255eead1911SChristoph Hellwig */ 2256057f6c01SJames Morris struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 2257057f6c01SJames Morris { 2258057f6c01SJames Morris struct qstr this; 22596a96ba54SAl Viro unsigned int c; 2260cda309deSMiklos Szeredi int err; 2261057f6c01SJames Morris 22622f9092e1SDavid Woodhouse WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex)); 22632f9092e1SDavid Woodhouse 22646a96ba54SAl Viro this.name = name; 22656a96ba54SAl Viro this.len = len; 22660145acc2SLinus Torvalds this.hash = full_name_hash(name, len); 22676a96ba54SAl Viro if (!len) 22686a96ba54SAl Viro return ERR_PTR(-EACCES); 22696a96ba54SAl Viro 227021d8a15aSAl Viro if (unlikely(name[0] == '.')) { 227121d8a15aSAl Viro if (len < 2 || (len == 2 && name[1] == '.')) 227221d8a15aSAl Viro return ERR_PTR(-EACCES); 227321d8a15aSAl Viro } 227421d8a15aSAl Viro 22756a96ba54SAl Viro while (len--) { 22766a96ba54SAl Viro c = *(const unsigned char *)name++; 22776a96ba54SAl Viro if (c == '/' || c == '\0') 22786a96ba54SAl Viro return ERR_PTR(-EACCES); 22796a96ba54SAl Viro } 22805a202bcdSAl Viro /* 22815a202bcdSAl Viro * See if the low-level filesystem might want 22825a202bcdSAl Viro * to use its own hash.. 22835a202bcdSAl Viro */ 22845a202bcdSAl Viro if (base->d_flags & DCACHE_OP_HASH) { 2285da53be12SLinus Torvalds int err = base->d_op->d_hash(base, &this); 22865a202bcdSAl Viro if (err < 0) 22875a202bcdSAl Viro return ERR_PTR(err); 22885a202bcdSAl Viro } 2289eead1911SChristoph Hellwig 2290cda309deSMiklos Szeredi err = inode_permission(base->d_inode, MAY_EXEC); 2291cda309deSMiklos Szeredi if (err) 2292cda309deSMiklos Szeredi return ERR_PTR(err); 2293cda309deSMiklos Szeredi 229472bd866aSAl Viro return __lookup_hash(&this, base, 0); 2295057f6c01SJames Morris } 22964d359507SAl Viro EXPORT_SYMBOL(lookup_one_len); 2297057f6c01SJames Morris 22981fa1e7f6SAndy Whitcroft int user_path_at_empty(int dfd, const char __user *name, unsigned flags, 22991fa1e7f6SAndy Whitcroft struct path *path, int *empty) 23001da177e4SLinus Torvalds { 230191a27b2aSJeff Layton struct filename *tmp = getname_flags(name, flags, empty); 2302e4bd1c1aSAl Viro if (IS_ERR(tmp)) 2303e4bd1c1aSAl Viro return PTR_ERR(tmp); 23042d8f3038SAl Viro 23052d8f3038SAl Viro BUG_ON(flags & LOOKUP_PARENT); 23062d8f3038SAl Viro 23079ad1aaa6SAl Viro return filename_lookup(dfd, tmp, flags, path, NULL); 23081da177e4SLinus Torvalds } 23091da177e4SLinus Torvalds 23101fa1e7f6SAndy Whitcroft int user_path_at(int dfd, const char __user *name, unsigned flags, 23111fa1e7f6SAndy Whitcroft struct path *path) 23121fa1e7f6SAndy Whitcroft { 2313f7493e5dSLinus Torvalds return user_path_at_empty(dfd, name, flags, path, NULL); 23141fa1e7f6SAndy Whitcroft } 23154d359507SAl Viro EXPORT_SYMBOL(user_path_at); 23161fa1e7f6SAndy Whitcroft 2317873f1eedSJeff Layton /* 2318873f1eedSJeff Layton * NB: most callers don't do anything directly with the reference to the 2319873f1eedSJeff Layton * to struct filename, but the nd->last pointer points into the name string 2320873f1eedSJeff Layton * allocated by getname. So we must hold the reference to it until all 2321873f1eedSJeff Layton * path-walking is complete. 2322873f1eedSJeff Layton */ 232391a27b2aSJeff Layton static struct filename * 2324f5beed75SAl Viro user_path_parent(int dfd, const char __user *path, 2325f5beed75SAl Viro struct path *parent, 2326f5beed75SAl Viro struct qstr *last, 2327f5beed75SAl Viro int *type, 23289e790bd6SJeff Layton unsigned int flags) 23292ad94ae6SAl Viro { 2330f5beed75SAl Viro struct nameidata nd; 233191a27b2aSJeff Layton struct filename *s = getname(path); 23322ad94ae6SAl Viro int error; 23332ad94ae6SAl Viro 23349e790bd6SJeff Layton /* only LOOKUP_REVAL is allowed in extra flags */ 23359e790bd6SJeff Layton flags &= LOOKUP_REVAL; 23369e790bd6SJeff Layton 23372ad94ae6SAl Viro if (IS_ERR(s)) 233891a27b2aSJeff Layton return s; 23392ad94ae6SAl Viro 23408bcb77faSAl Viro error = filename_parentat(dfd, s, flags, &nd); 234191a27b2aSJeff Layton if (error) { 23422ad94ae6SAl Viro putname(s); 234391a27b2aSJeff Layton return ERR_PTR(error); 234491a27b2aSJeff Layton } 2345f5beed75SAl Viro *parent = nd.path; 2346f5beed75SAl Viro *last = nd.last; 2347f5beed75SAl Viro *type = nd.last_type; 23482ad94ae6SAl Viro 234991a27b2aSJeff Layton return s; 23502ad94ae6SAl Viro } 23512ad94ae6SAl Viro 23528033426eSJeff Layton /** 2353197df04cSAl Viro * mountpoint_last - look up last component for umount 23548033426eSJeff Layton * @nd: pathwalk nameidata - currently pointing at parent directory of "last" 23558033426eSJeff Layton * @path: pointer to container for result 23568033426eSJeff Layton * 23578033426eSJeff Layton * This is a special lookup_last function just for umount. In this case, we 23588033426eSJeff Layton * need to resolve the path without doing any revalidation. 23598033426eSJeff Layton * 23608033426eSJeff Layton * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since 23618033426eSJeff Layton * mountpoints are always pinned in the dcache, their ancestors are too. Thus, 23628033426eSJeff Layton * in almost all cases, this lookup will be served out of the dcache. The only 23638033426eSJeff Layton * cases where it won't are if nd->last refers to a symlink or the path is 23648033426eSJeff Layton * bogus and it doesn't exist. 23658033426eSJeff Layton * 23668033426eSJeff Layton * Returns: 23678033426eSJeff Layton * -error: if there was an error during lookup. This includes -ENOENT if the 23688033426eSJeff Layton * lookup found a negative dentry. The nd->path reference will also be 23698033426eSJeff Layton * put in this case. 23708033426eSJeff Layton * 23718033426eSJeff Layton * 0: if we successfully resolved nd->path and found it to not to be a 23728033426eSJeff Layton * symlink that needs to be followed. "path" will also be populated. 23738033426eSJeff Layton * The nd->path reference will also be put. 23748033426eSJeff Layton * 23758033426eSJeff Layton * 1: if we successfully resolved nd->last and found it to be a symlink 23768033426eSJeff Layton * that needs to be followed. "path" will be populated with the path 23778033426eSJeff Layton * to the link, and nd->path will *not* be put. 23788033426eSJeff Layton */ 23798033426eSJeff Layton static int 2380197df04cSAl Viro mountpoint_last(struct nameidata *nd, struct path *path) 23818033426eSJeff Layton { 23828033426eSJeff Layton int error = 0; 23838033426eSJeff Layton struct dentry *dentry; 23848033426eSJeff Layton struct dentry *dir = nd->path.dentry; 23858033426eSJeff Layton 238635759521SAl Viro /* If we're in rcuwalk, drop out of it to handle last component */ 238735759521SAl Viro if (nd->flags & LOOKUP_RCU) { 23886e9918b7SAl Viro if (unlazy_walk(nd, NULL, 0)) 2389deb106c6SAl Viro return -ECHILD; 23908033426eSJeff Layton } 23918033426eSJeff Layton 23928033426eSJeff Layton nd->flags &= ~LOOKUP_PARENT; 23938033426eSJeff Layton 23948033426eSJeff Layton if (unlikely(nd->last_type != LAST_NORM)) { 23958033426eSJeff Layton error = handle_dots(nd, nd->last_type); 239635759521SAl Viro if (error) 2397deb106c6SAl Viro return error; 23988033426eSJeff Layton dentry = dget(nd->path.dentry); 239935759521SAl Viro goto done; 24008033426eSJeff Layton } 24018033426eSJeff Layton 24028033426eSJeff Layton mutex_lock(&dir->d_inode->i_mutex); 24038033426eSJeff Layton dentry = d_lookup(dir, &nd->last); 24048033426eSJeff Layton if (!dentry) { 24058033426eSJeff Layton /* 24068033426eSJeff Layton * No cached dentry. Mounted dentries are pinned in the cache, 24078033426eSJeff Layton * so that means that this dentry is probably a symlink or the 24088033426eSJeff Layton * path doesn't actually point to a mounted dentry. 24098033426eSJeff Layton */ 24108033426eSJeff Layton dentry = d_alloc(dir, &nd->last); 24118033426eSJeff Layton if (!dentry) { 2412bcceeebaSDave Jones mutex_unlock(&dir->d_inode->i_mutex); 2413deb106c6SAl Viro return -ENOMEM; 24148033426eSJeff Layton } 241535759521SAl Viro dentry = lookup_real(dir->d_inode, dentry, nd->flags); 2416bcceeebaSDave Jones if (IS_ERR(dentry)) { 2417bcceeebaSDave Jones mutex_unlock(&dir->d_inode->i_mutex); 2418deb106c6SAl Viro return PTR_ERR(dentry); 24198033426eSJeff Layton } 2420bcceeebaSDave Jones } 24218033426eSJeff Layton mutex_unlock(&dir->d_inode->i_mutex); 24228033426eSJeff Layton 242335759521SAl Viro done: 2424698934dfSDavid Howells if (d_is_negative(dentry)) { 24258033426eSJeff Layton dput(dentry); 2426deb106c6SAl Viro return -ENOENT; 242735759521SAl Viro } 2428191d7f73SAl Viro if (nd->depth) 2429191d7f73SAl Viro put_link(nd); 24308033426eSJeff Layton path->dentry = dentry; 2431295dc39dSVasily Averin path->mnt = nd->path.mnt; 2432181548c0SAl Viro error = should_follow_link(nd, path, nd->flags & LOOKUP_FOLLOW, 2433181548c0SAl Viro d_backing_inode(dentry), 0); 2434deb106c6SAl Viro if (unlikely(error)) 2435d63ff28fSAl Viro return error; 2436295dc39dSVasily Averin mntget(path->mnt); 24378033426eSJeff Layton follow_mount(path); 2438deb106c6SAl Viro return 0; 24398033426eSJeff Layton } 24408033426eSJeff Layton 24418033426eSJeff Layton /** 2442197df04cSAl Viro * path_mountpoint - look up a path to be umounted 24438033426eSJeff Layton * @dfd: directory file descriptor to start walk from 24448033426eSJeff Layton * @name: full pathname to walk 2445606d6fe3SRandy Dunlap * @path: pointer to container for result 24468033426eSJeff Layton * @flags: lookup flags 24478033426eSJeff Layton * 24488033426eSJeff Layton * Look up the given name, but don't attempt to revalidate the last component. 2449606d6fe3SRandy Dunlap * Returns 0 and "path" will be valid on success; Returns error otherwise. 24508033426eSJeff Layton */ 24518033426eSJeff Layton static int 2452668696dcSAl Viro path_mountpoint(int dfd, const struct filename *name, struct path *path, 245346afd6f6SAl Viro struct nameidata *nd, unsigned int flags) 24548033426eSJeff Layton { 2455368ee9baSAl Viro const char *s = path_init(dfd, name, flags, nd); 2456368ee9baSAl Viro int err; 2457368ee9baSAl Viro if (IS_ERR(s)) 2458368ee9baSAl Viro return PTR_ERR(s); 24593bdba28bSAl Viro while (!(err = link_path_walk(s, nd)) && 24603bdba28bSAl Viro (err = mountpoint_last(nd, path)) > 0) { 24613bdba28bSAl Viro s = trailing_symlink(nd); 24623bdba28bSAl Viro if (IS_ERR(s)) { 24633bdba28bSAl Viro err = PTR_ERR(s); 24648033426eSJeff Layton break; 24658033426eSJeff Layton } 24663bdba28bSAl Viro } 2467deb106c6SAl Viro terminate_walk(nd); 246846afd6f6SAl Viro path_cleanup(nd); 24698033426eSJeff Layton return err; 24708033426eSJeff Layton } 24718033426eSJeff Layton 24722d864651SAl Viro static int 2473668696dcSAl Viro filename_mountpoint(int dfd, struct filename *name, struct path *path, 24742d864651SAl Viro unsigned int flags) 24752d864651SAl Viro { 2476756daf26SNeilBrown struct nameidata nd, *saved; 2477cbaab2dbSAl Viro int error; 2478668696dcSAl Viro if (IS_ERR(name)) 2479668696dcSAl Viro return PTR_ERR(name); 2480756daf26SNeilBrown saved = set_nameidata(&nd); 248146afd6f6SAl Viro error = path_mountpoint(dfd, name, path, &nd, flags | LOOKUP_RCU); 24822d864651SAl Viro if (unlikely(error == -ECHILD)) 248346afd6f6SAl Viro error = path_mountpoint(dfd, name, path, &nd, flags); 24842d864651SAl Viro if (unlikely(error == -ESTALE)) 248546afd6f6SAl Viro error = path_mountpoint(dfd, name, path, &nd, flags | LOOKUP_REVAL); 24862d864651SAl Viro if (likely(!error)) 2487668696dcSAl Viro audit_inode(name, path->dentry, 0); 2488756daf26SNeilBrown restore_nameidata(saved); 2489668696dcSAl Viro putname(name); 24902d864651SAl Viro return error; 24912d864651SAl Viro } 24922d864651SAl Viro 24938033426eSJeff Layton /** 2494197df04cSAl Viro * user_path_mountpoint_at - lookup a path from userland in order to umount it 24958033426eSJeff Layton * @dfd: directory file descriptor 24968033426eSJeff Layton * @name: pathname from userland 24978033426eSJeff Layton * @flags: lookup flags 24988033426eSJeff Layton * @path: pointer to container to hold result 24998033426eSJeff Layton * 25008033426eSJeff Layton * A umount is a special case for path walking. We're not actually interested 25018033426eSJeff Layton * in the inode in this situation, and ESTALE errors can be a problem. We 25028033426eSJeff Layton * simply want track down the dentry and vfsmount attached at the mountpoint 25038033426eSJeff Layton * and avoid revalidating the last component. 25048033426eSJeff Layton * 25058033426eSJeff Layton * Returns 0 and populates "path" on success. 25068033426eSJeff Layton */ 25078033426eSJeff Layton int 2508197df04cSAl Viro user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags, 25098033426eSJeff Layton struct path *path) 25108033426eSJeff Layton { 2511cbaab2dbSAl Viro return filename_mountpoint(dfd, getname(name), path, flags); 25128033426eSJeff Layton } 25138033426eSJeff Layton 25142d864651SAl Viro int 25152d864651SAl Viro kern_path_mountpoint(int dfd, const char *name, struct path *path, 25162d864651SAl Viro unsigned int flags) 25172d864651SAl Viro { 2518cbaab2dbSAl Viro return filename_mountpoint(dfd, getname_kernel(name), path, flags); 25192d864651SAl Viro } 25202d864651SAl Viro EXPORT_SYMBOL(kern_path_mountpoint); 25212d864651SAl Viro 2522cbdf35bcSMiklos Szeredi int __check_sticky(struct inode *dir, struct inode *inode) 25231da177e4SLinus Torvalds { 25248e96e3b7SEric W. Biederman kuid_t fsuid = current_fsuid(); 2525da9592edSDavid Howells 25268e96e3b7SEric W. Biederman if (uid_eq(inode->i_uid, fsuid)) 25271da177e4SLinus Torvalds return 0; 25288e96e3b7SEric W. Biederman if (uid_eq(dir->i_uid, fsuid)) 25291da177e4SLinus Torvalds return 0; 253023adbe12SAndy Lutomirski return !capable_wrt_inode_uidgid(inode, CAP_FOWNER); 25311da177e4SLinus Torvalds } 2532cbdf35bcSMiklos Szeredi EXPORT_SYMBOL(__check_sticky); 25331da177e4SLinus Torvalds 25341da177e4SLinus Torvalds /* 25351da177e4SLinus Torvalds * Check whether we can remove a link victim from directory dir, check 25361da177e4SLinus Torvalds * whether the type of victim is right. 25371da177e4SLinus Torvalds * 1. We can't do it if dir is read-only (done in permission()) 25381da177e4SLinus Torvalds * 2. We should have write and exec permissions on dir 25391da177e4SLinus Torvalds * 3. We can't remove anything from append-only dir 25401da177e4SLinus Torvalds * 4. We can't do anything with immutable dir (done in permission()) 25411da177e4SLinus Torvalds * 5. If the sticky bit on dir is set we should either 25421da177e4SLinus Torvalds * a. be owner of dir, or 25431da177e4SLinus Torvalds * b. be owner of victim, or 25441da177e4SLinus Torvalds * c. have CAP_FOWNER capability 25451da177e4SLinus Torvalds * 6. If the victim is append-only or immutable we can't do antyhing with 25461da177e4SLinus Torvalds * links pointing to it. 25471da177e4SLinus Torvalds * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. 25481da177e4SLinus Torvalds * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. 25491da177e4SLinus Torvalds * 9. We can't remove a root or mountpoint. 25501da177e4SLinus Torvalds * 10. We don't allow removal of NFS sillyrenamed files; it's handled by 25511da177e4SLinus Torvalds * nfs_async_unlink(). 25521da177e4SLinus Torvalds */ 2553b18825a7SDavid Howells static int may_delete(struct inode *dir, struct dentry *victim, bool isdir) 25541da177e4SLinus Torvalds { 255563afdfc7SDavid Howells struct inode *inode = d_backing_inode(victim); 25561da177e4SLinus Torvalds int error; 25571da177e4SLinus Torvalds 2558b18825a7SDavid Howells if (d_is_negative(victim)) 25591da177e4SLinus Torvalds return -ENOENT; 2560b18825a7SDavid Howells BUG_ON(!inode); 25611da177e4SLinus Torvalds 25621da177e4SLinus Torvalds BUG_ON(victim->d_parent->d_inode != dir); 25634fa6b5ecSJeff Layton audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); 25641da177e4SLinus Torvalds 2565f419a2e3SAl Viro error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 25661da177e4SLinus Torvalds if (error) 25671da177e4SLinus Torvalds return error; 25681da177e4SLinus Torvalds if (IS_APPEND(dir)) 25691da177e4SLinus Torvalds return -EPERM; 2570b18825a7SDavid Howells 2571b18825a7SDavid Howells if (check_sticky(dir, inode) || IS_APPEND(inode) || 2572b18825a7SDavid Howells IS_IMMUTABLE(inode) || IS_SWAPFILE(inode)) 25731da177e4SLinus Torvalds return -EPERM; 25741da177e4SLinus Torvalds if (isdir) { 257544b1d530SMiklos Szeredi if (!d_is_dir(victim)) 25761da177e4SLinus Torvalds return -ENOTDIR; 25771da177e4SLinus Torvalds if (IS_ROOT(victim)) 25781da177e4SLinus Torvalds return -EBUSY; 257944b1d530SMiklos Szeredi } else if (d_is_dir(victim)) 25801da177e4SLinus Torvalds return -EISDIR; 25811da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 25821da177e4SLinus Torvalds return -ENOENT; 25831da177e4SLinus Torvalds if (victim->d_flags & DCACHE_NFSFS_RENAMED) 25841da177e4SLinus Torvalds return -EBUSY; 25851da177e4SLinus Torvalds return 0; 25861da177e4SLinus Torvalds } 25871da177e4SLinus Torvalds 25881da177e4SLinus Torvalds /* Check whether we can create an object with dentry child in directory 25891da177e4SLinus Torvalds * dir. 25901da177e4SLinus Torvalds * 1. We can't do it if child already exists (open has special treatment for 25911da177e4SLinus Torvalds * this case, but since we are inlined it's OK) 25921da177e4SLinus Torvalds * 2. We can't do it if dir is read-only (done in permission()) 25931da177e4SLinus Torvalds * 3. We should have write and exec permissions on dir 25941da177e4SLinus Torvalds * 4. We can't do it if dir is immutable (done in permission()) 25951da177e4SLinus Torvalds */ 2596a95164d9SMiklos Szeredi static inline int may_create(struct inode *dir, struct dentry *child) 25971da177e4SLinus Torvalds { 259814e972b4SJeff Layton audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); 25991da177e4SLinus Torvalds if (child->d_inode) 26001da177e4SLinus Torvalds return -EEXIST; 26011da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 26021da177e4SLinus Torvalds return -ENOENT; 2603f419a2e3SAl Viro return inode_permission(dir, MAY_WRITE | MAY_EXEC); 26041da177e4SLinus Torvalds } 26051da177e4SLinus Torvalds 26061da177e4SLinus Torvalds /* 26071da177e4SLinus Torvalds * p1 and p2 should be directories on the same fs. 26081da177e4SLinus Torvalds */ 26091da177e4SLinus Torvalds struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 26101da177e4SLinus Torvalds { 26111da177e4SLinus Torvalds struct dentry *p; 26121da177e4SLinus Torvalds 26131da177e4SLinus Torvalds if (p1 == p2) { 2614f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 26151da177e4SLinus Torvalds return NULL; 26161da177e4SLinus Torvalds } 26171da177e4SLinus Torvalds 2618a11f3a05SArjan van de Ven mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 26191da177e4SLinus Torvalds 2620e2761a11SOGAWA Hirofumi p = d_ancestor(p2, p1); 2621e2761a11SOGAWA Hirofumi if (p) { 2622f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT); 2623f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD); 26241da177e4SLinus Torvalds return p; 26251da177e4SLinus Torvalds } 26261da177e4SLinus Torvalds 2627e2761a11SOGAWA Hirofumi p = d_ancestor(p1, p2); 2628e2761a11SOGAWA Hirofumi if (p) { 2629f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 2630f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); 26311da177e4SLinus Torvalds return p; 26321da177e4SLinus Torvalds } 26331da177e4SLinus Torvalds 2634f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 2635d1b72cc6SMiklos Szeredi mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT2); 26361da177e4SLinus Torvalds return NULL; 26371da177e4SLinus Torvalds } 26384d359507SAl Viro EXPORT_SYMBOL(lock_rename); 26391da177e4SLinus Torvalds 26401da177e4SLinus Torvalds void unlock_rename(struct dentry *p1, struct dentry *p2) 26411da177e4SLinus Torvalds { 26421b1dcc1bSJes Sorensen mutex_unlock(&p1->d_inode->i_mutex); 26431da177e4SLinus Torvalds if (p1 != p2) { 26441b1dcc1bSJes Sorensen mutex_unlock(&p2->d_inode->i_mutex); 2645a11f3a05SArjan van de Ven mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 26461da177e4SLinus Torvalds } 26471da177e4SLinus Torvalds } 26484d359507SAl Viro EXPORT_SYMBOL(unlock_rename); 26491da177e4SLinus Torvalds 26504acdaf27SAl Viro int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, 2651312b63fbSAl Viro bool want_excl) 26521da177e4SLinus Torvalds { 2653a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 26541da177e4SLinus Torvalds if (error) 26551da177e4SLinus Torvalds return error; 26561da177e4SLinus Torvalds 2657acfa4380SAl Viro if (!dir->i_op->create) 26581da177e4SLinus Torvalds return -EACCES; /* shouldn't it be ENOSYS? */ 26591da177e4SLinus Torvalds mode &= S_IALLUGO; 26601da177e4SLinus Torvalds mode |= S_IFREG; 26611da177e4SLinus Torvalds error = security_inode_create(dir, dentry, mode); 26621da177e4SLinus Torvalds if (error) 26631da177e4SLinus Torvalds return error; 2664312b63fbSAl Viro error = dir->i_op->create(dir, dentry, mode, want_excl); 2665a74574aaSStephen Smalley if (!error) 2666f38aa942SAmy Griffis fsnotify_create(dir, dentry); 26671da177e4SLinus Torvalds return error; 26681da177e4SLinus Torvalds } 26694d359507SAl Viro EXPORT_SYMBOL(vfs_create); 26701da177e4SLinus Torvalds 267173d049a4SAl Viro static int may_open(struct path *path, int acc_mode, int flag) 26721da177e4SLinus Torvalds { 26733fb64190SChristoph Hellwig struct dentry *dentry = path->dentry; 26741da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 26751da177e4SLinus Torvalds int error; 26761da177e4SLinus Torvalds 2677bcda7652SAl Viro /* O_PATH? */ 2678bcda7652SAl Viro if (!acc_mode) 2679bcda7652SAl Viro return 0; 2680bcda7652SAl Viro 26811da177e4SLinus Torvalds if (!inode) 26821da177e4SLinus Torvalds return -ENOENT; 26831da177e4SLinus Torvalds 2684c8fe8f30SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 2685c8fe8f30SChristoph Hellwig case S_IFLNK: 26861da177e4SLinus Torvalds return -ELOOP; 2687c8fe8f30SChristoph Hellwig case S_IFDIR: 2688c8fe8f30SChristoph Hellwig if (acc_mode & MAY_WRITE) 26891da177e4SLinus Torvalds return -EISDIR; 2690c8fe8f30SChristoph Hellwig break; 2691c8fe8f30SChristoph Hellwig case S_IFBLK: 2692c8fe8f30SChristoph Hellwig case S_IFCHR: 26933fb64190SChristoph Hellwig if (path->mnt->mnt_flags & MNT_NODEV) 26941da177e4SLinus Torvalds return -EACCES; 2695c8fe8f30SChristoph Hellwig /*FALLTHRU*/ 2696c8fe8f30SChristoph Hellwig case S_IFIFO: 2697c8fe8f30SChristoph Hellwig case S_IFSOCK: 26981da177e4SLinus Torvalds flag &= ~O_TRUNC; 2699c8fe8f30SChristoph Hellwig break; 27004a3fd211SDave Hansen } 2701b41572e9SDave Hansen 27023fb64190SChristoph Hellwig error = inode_permission(inode, acc_mode); 2703b41572e9SDave Hansen if (error) 2704b41572e9SDave Hansen return error; 27056146f0d5SMimi Zohar 27061da177e4SLinus Torvalds /* 27071da177e4SLinus Torvalds * An append-only file must be opened in append mode for writing. 27081da177e4SLinus Torvalds */ 27091da177e4SLinus Torvalds if (IS_APPEND(inode)) { 27108737c930SAl Viro if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) 27117715b521SAl Viro return -EPERM; 27121da177e4SLinus Torvalds if (flag & O_TRUNC) 27137715b521SAl Viro return -EPERM; 27141da177e4SLinus Torvalds } 27151da177e4SLinus Torvalds 27161da177e4SLinus Torvalds /* O_NOATIME can only be set by the owner or superuser */ 27172e149670SSerge E. Hallyn if (flag & O_NOATIME && !inode_owner_or_capable(inode)) 27187715b521SAl Viro return -EPERM; 27191da177e4SLinus Torvalds 2720f3c7691eSJ. Bruce Fields return 0; 27217715b521SAl Viro } 27227715b521SAl Viro 2723e1181ee6SJeff Layton static int handle_truncate(struct file *filp) 27247715b521SAl Viro { 2725e1181ee6SJeff Layton struct path *path = &filp->f_path; 27267715b521SAl Viro struct inode *inode = path->dentry->d_inode; 27277715b521SAl Viro int error = get_write_access(inode); 27281da177e4SLinus Torvalds if (error) 27297715b521SAl Viro return error; 27301da177e4SLinus Torvalds /* 27311da177e4SLinus Torvalds * Refuse to truncate files with mandatory locks held on them. 27321da177e4SLinus Torvalds */ 2733d7a06983SJeff Layton error = locks_verify_locked(filp); 2734be6d3e56SKentaro Takeda if (!error) 2735ea0d3ab2STetsuo Handa error = security_path_truncate(path); 27361da177e4SLinus Torvalds if (!error) { 27377715b521SAl Viro error = do_truncate(path->dentry, 0, 2738d139d7ffSMiklos Szeredi ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, 2739e1181ee6SJeff Layton filp); 27401da177e4SLinus Torvalds } 27411da177e4SLinus Torvalds put_write_access(inode); 2742acd0c935SMimi Zohar return error; 27431da177e4SLinus Torvalds } 27441da177e4SLinus Torvalds 2745d57999e1SDave Hansen static inline int open_to_namei_flags(int flag) 2746d57999e1SDave Hansen { 27478a5e929dSAl Viro if ((flag & O_ACCMODE) == 3) 27488a5e929dSAl Viro flag--; 2749d57999e1SDave Hansen return flag; 2750d57999e1SDave Hansen } 2751d57999e1SDave Hansen 2752d18e9008SMiklos Szeredi static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode) 2753d18e9008SMiklos Szeredi { 2754d18e9008SMiklos Szeredi int error = security_path_mknod(dir, dentry, mode, 0); 2755d18e9008SMiklos Szeredi if (error) 2756d18e9008SMiklos Szeredi return error; 2757d18e9008SMiklos Szeredi 2758d18e9008SMiklos Szeredi error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC); 2759d18e9008SMiklos Szeredi if (error) 2760d18e9008SMiklos Szeredi return error; 2761d18e9008SMiklos Szeredi 2762d18e9008SMiklos Szeredi return security_inode_create(dir->dentry->d_inode, dentry, mode); 2763d18e9008SMiklos Szeredi } 2764d18e9008SMiklos Szeredi 27651acf0af9SDavid Howells /* 27661acf0af9SDavid Howells * Attempt to atomically look up, create and open a file from a negative 27671acf0af9SDavid Howells * dentry. 27681acf0af9SDavid Howells * 27691acf0af9SDavid Howells * Returns 0 if successful. The file will have been created and attached to 27701acf0af9SDavid Howells * @file by the filesystem calling finish_open(). 27711acf0af9SDavid Howells * 27721acf0af9SDavid Howells * Returns 1 if the file was looked up only or didn't need creating. The 27731acf0af9SDavid Howells * caller will need to perform the open themselves. @path will have been 27741acf0af9SDavid Howells * updated to point to the new dentry. This may be negative. 27751acf0af9SDavid Howells * 27761acf0af9SDavid Howells * Returns an error code otherwise. 27771acf0af9SDavid Howells */ 27782675a4ebSAl Viro static int atomic_open(struct nameidata *nd, struct dentry *dentry, 277930d90494SAl Viro struct path *path, struct file *file, 2780015c3bbcSMiklos Szeredi const struct open_flags *op, 278164894cf8SAl Viro bool got_write, bool need_lookup, 278247237687SAl Viro int *opened) 2783d18e9008SMiklos Szeredi { 2784d18e9008SMiklos Szeredi struct inode *dir = nd->path.dentry->d_inode; 2785d18e9008SMiklos Szeredi unsigned open_flag = open_to_namei_flags(op->open_flag); 2786d18e9008SMiklos Szeredi umode_t mode; 2787d18e9008SMiklos Szeredi int error; 2788d18e9008SMiklos Szeredi int acc_mode; 2789d18e9008SMiklos Szeredi int create_error = 0; 2790d18e9008SMiklos Szeredi struct dentry *const DENTRY_NOT_SET = (void *) -1UL; 2791116cc022SMiklos Szeredi bool excl; 2792d18e9008SMiklos Szeredi 2793d18e9008SMiklos Szeredi BUG_ON(dentry->d_inode); 2794d18e9008SMiklos Szeredi 2795d18e9008SMiklos Szeredi /* Don't create child dentry for a dead directory. */ 2796d18e9008SMiklos Szeredi if (unlikely(IS_DEADDIR(dir))) { 27972675a4ebSAl Viro error = -ENOENT; 2798d18e9008SMiklos Szeredi goto out; 2799d18e9008SMiklos Szeredi } 2800d18e9008SMiklos Szeredi 280162b259d8SMiklos Szeredi mode = op->mode; 2802d18e9008SMiklos Szeredi if ((open_flag & O_CREAT) && !IS_POSIXACL(dir)) 2803d18e9008SMiklos Szeredi mode &= ~current_umask(); 2804d18e9008SMiklos Szeredi 2805116cc022SMiklos Szeredi excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT); 2806116cc022SMiklos Szeredi if (excl) 2807d18e9008SMiklos Szeredi open_flag &= ~O_TRUNC; 2808d18e9008SMiklos Szeredi 2809d18e9008SMiklos Szeredi /* 2810d18e9008SMiklos Szeredi * Checking write permission is tricky, bacuse we don't know if we are 2811d18e9008SMiklos Szeredi * going to actually need it: O_CREAT opens should work as long as the 2812d18e9008SMiklos Szeredi * file exists. But checking existence breaks atomicity. The trick is 2813d18e9008SMiklos Szeredi * to check access and if not granted clear O_CREAT from the flags. 2814d18e9008SMiklos Szeredi * 2815d18e9008SMiklos Szeredi * Another problem is returing the "right" error value (e.g. for an 2816d18e9008SMiklos Szeredi * O_EXCL open we want to return EEXIST not EROFS). 2817d18e9008SMiklos Szeredi */ 281864894cf8SAl Viro if (((open_flag & (O_CREAT | O_TRUNC)) || 281964894cf8SAl Viro (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) { 282064894cf8SAl Viro if (!(open_flag & O_CREAT)) { 2821d18e9008SMiklos Szeredi /* 2822d18e9008SMiklos Szeredi * No O_CREATE -> atomicity not a requirement -> fall 2823d18e9008SMiklos Szeredi * back to lookup + open 2824d18e9008SMiklos Szeredi */ 2825d18e9008SMiklos Szeredi goto no_open; 2826d18e9008SMiklos Szeredi } else if (open_flag & (O_EXCL | O_TRUNC)) { 2827d18e9008SMiklos Szeredi /* Fall back and fail with the right error */ 282864894cf8SAl Viro create_error = -EROFS; 2829d18e9008SMiklos Szeredi goto no_open; 2830d18e9008SMiklos Szeredi } else { 2831d18e9008SMiklos Szeredi /* No side effects, safe to clear O_CREAT */ 283264894cf8SAl Viro create_error = -EROFS; 2833d18e9008SMiklos Szeredi open_flag &= ~O_CREAT; 2834d18e9008SMiklos Szeredi } 2835d18e9008SMiklos Szeredi } 2836d18e9008SMiklos Szeredi 2837d18e9008SMiklos Szeredi if (open_flag & O_CREAT) { 283838227f78SMiklos Szeredi error = may_o_create(&nd->path, dentry, mode); 2839d18e9008SMiklos Szeredi if (error) { 2840d18e9008SMiklos Szeredi create_error = error; 2841d18e9008SMiklos Szeredi if (open_flag & O_EXCL) 2842d18e9008SMiklos Szeredi goto no_open; 2843d18e9008SMiklos Szeredi open_flag &= ~O_CREAT; 2844d18e9008SMiklos Szeredi } 2845d18e9008SMiklos Szeredi } 2846d18e9008SMiklos Szeredi 2847d18e9008SMiklos Szeredi if (nd->flags & LOOKUP_DIRECTORY) 2848d18e9008SMiklos Szeredi open_flag |= O_DIRECTORY; 2849d18e9008SMiklos Szeredi 285030d90494SAl Viro file->f_path.dentry = DENTRY_NOT_SET; 285130d90494SAl Viro file->f_path.mnt = nd->path.mnt; 285230d90494SAl Viro error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode, 285347237687SAl Viro opened); 2854d9585277SAl Viro if (error < 0) { 2855d9585277SAl Viro if (create_error && error == -ENOENT) 2856d9585277SAl Viro error = create_error; 2857d18e9008SMiklos Szeredi goto out; 2858d18e9008SMiklos Szeredi } 2859d18e9008SMiklos Szeredi 2860d9585277SAl Viro if (error) { /* returned 1, that is */ 286130d90494SAl Viro if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { 28622675a4ebSAl Viro error = -EIO; 2863d18e9008SMiklos Szeredi goto out; 2864d18e9008SMiklos Szeredi } 286530d90494SAl Viro if (file->f_path.dentry) { 2866d18e9008SMiklos Szeredi dput(dentry); 286730d90494SAl Viro dentry = file->f_path.dentry; 2868d18e9008SMiklos Szeredi } 286903da633aSAl Viro if (*opened & FILE_CREATED) 287003da633aSAl Viro fsnotify_create(dir, dentry); 287103da633aSAl Viro if (!dentry->d_inode) { 287203da633aSAl Viro WARN_ON(*opened & FILE_CREATED); 287303da633aSAl Viro if (create_error) { 287462b2ce96SSage Weil error = create_error; 287562b2ce96SSage Weil goto out; 287662b2ce96SSage Weil } 287703da633aSAl Viro } else { 287803da633aSAl Viro if (excl && !(*opened & FILE_CREATED)) { 287903da633aSAl Viro error = -EEXIST; 288003da633aSAl Viro goto out; 288103da633aSAl Viro } 288203da633aSAl Viro } 2883d18e9008SMiklos Szeredi goto looked_up; 2884d18e9008SMiklos Szeredi } 2885d18e9008SMiklos Szeredi 2886d18e9008SMiklos Szeredi /* 2887d18e9008SMiklos Szeredi * We didn't have the inode before the open, so check open permission 2888d18e9008SMiklos Szeredi * here. 2889d18e9008SMiklos Szeredi */ 289003da633aSAl Viro acc_mode = op->acc_mode; 289103da633aSAl Viro if (*opened & FILE_CREATED) { 289203da633aSAl Viro WARN_ON(!(open_flag & O_CREAT)); 289303da633aSAl Viro fsnotify_create(dir, dentry); 289403da633aSAl Viro acc_mode = MAY_OPEN; 289503da633aSAl Viro } 28962675a4ebSAl Viro error = may_open(&file->f_path, acc_mode, open_flag); 28972675a4ebSAl Viro if (error) 28982675a4ebSAl Viro fput(file); 2899d18e9008SMiklos Szeredi 2900d18e9008SMiklos Szeredi out: 2901d18e9008SMiklos Szeredi dput(dentry); 29022675a4ebSAl Viro return error; 2903d18e9008SMiklos Szeredi 2904d18e9008SMiklos Szeredi no_open: 2905d18e9008SMiklos Szeredi if (need_lookup) { 290672bd866aSAl Viro dentry = lookup_real(dir, dentry, nd->flags); 2907d18e9008SMiklos Szeredi if (IS_ERR(dentry)) 29082675a4ebSAl Viro return PTR_ERR(dentry); 2909d18e9008SMiklos Szeredi 2910d18e9008SMiklos Szeredi if (create_error) { 2911d18e9008SMiklos Szeredi int open_flag = op->open_flag; 2912d18e9008SMiklos Szeredi 29132675a4ebSAl Viro error = create_error; 2914d18e9008SMiklos Szeredi if ((open_flag & O_EXCL)) { 2915d18e9008SMiklos Szeredi if (!dentry->d_inode) 2916d18e9008SMiklos Szeredi goto out; 2917d18e9008SMiklos Szeredi } else if (!dentry->d_inode) { 2918d18e9008SMiklos Szeredi goto out; 2919d18e9008SMiklos Szeredi } else if ((open_flag & O_TRUNC) && 2920e36cb0b8SDavid Howells d_is_reg(dentry)) { 2921d18e9008SMiklos Szeredi goto out; 2922d18e9008SMiklos Szeredi } 2923d18e9008SMiklos Szeredi /* will fail later, go on to get the right error */ 2924d18e9008SMiklos Szeredi } 2925d18e9008SMiklos Szeredi } 2926d18e9008SMiklos Szeredi looked_up: 2927d18e9008SMiklos Szeredi path->dentry = dentry; 2928d18e9008SMiklos Szeredi path->mnt = nd->path.mnt; 29292675a4ebSAl Viro return 1; 2930d18e9008SMiklos Szeredi } 2931d18e9008SMiklos Szeredi 293231e6b01fSNick Piggin /* 29331acf0af9SDavid Howells * Look up and maybe create and open the last component. 2934d58ffd35SMiklos Szeredi * 2935d58ffd35SMiklos Szeredi * Must be called with i_mutex held on parent. 2936d58ffd35SMiklos Szeredi * 29371acf0af9SDavid Howells * Returns 0 if the file was successfully atomically created (if necessary) and 29381acf0af9SDavid Howells * opened. In this case the file will be returned attached to @file. 29391acf0af9SDavid Howells * 29401acf0af9SDavid Howells * Returns 1 if the file was not completely opened at this time, though lookups 29411acf0af9SDavid Howells * and creations will have been performed and the dentry returned in @path will 29421acf0af9SDavid Howells * be positive upon return if O_CREAT was specified. If O_CREAT wasn't 29431acf0af9SDavid Howells * specified then a negative dentry may be returned. 29441acf0af9SDavid Howells * 29451acf0af9SDavid Howells * An error code is returned otherwise. 29461acf0af9SDavid Howells * 29471acf0af9SDavid Howells * FILE_CREATE will be set in @*opened if the dentry was created and will be 29481acf0af9SDavid Howells * cleared otherwise prior to returning. 2949d58ffd35SMiklos Szeredi */ 29502675a4ebSAl Viro static int lookup_open(struct nameidata *nd, struct path *path, 295130d90494SAl Viro struct file *file, 2952d58ffd35SMiklos Szeredi const struct open_flags *op, 295364894cf8SAl Viro bool got_write, int *opened) 2954d58ffd35SMiklos Szeredi { 2955d58ffd35SMiklos Szeredi struct dentry *dir = nd->path.dentry; 295654ef4872SMiklos Szeredi struct inode *dir_inode = dir->d_inode; 2957d58ffd35SMiklos Szeredi struct dentry *dentry; 2958d58ffd35SMiklos Szeredi int error; 295954ef4872SMiklos Szeredi bool need_lookup; 2960d58ffd35SMiklos Szeredi 296147237687SAl Viro *opened &= ~FILE_CREATED; 2962201f956eSAl Viro dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup); 2963d58ffd35SMiklos Szeredi if (IS_ERR(dentry)) 29642675a4ebSAl Viro return PTR_ERR(dentry); 2965d58ffd35SMiklos Szeredi 2966d18e9008SMiklos Szeredi /* Cached positive dentry: will open in f_op->open */ 2967d18e9008SMiklos Szeredi if (!need_lookup && dentry->d_inode) 2968d18e9008SMiklos Szeredi goto out_no_open; 2969d18e9008SMiklos Szeredi 2970d18e9008SMiklos Szeredi if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) { 297164894cf8SAl Viro return atomic_open(nd, dentry, path, file, op, got_write, 297247237687SAl Viro need_lookup, opened); 2973d18e9008SMiklos Szeredi } 2974d18e9008SMiklos Szeredi 297554ef4872SMiklos Szeredi if (need_lookup) { 297654ef4872SMiklos Szeredi BUG_ON(dentry->d_inode); 297754ef4872SMiklos Szeredi 297872bd866aSAl Viro dentry = lookup_real(dir_inode, dentry, nd->flags); 297954ef4872SMiklos Szeredi if (IS_ERR(dentry)) 29802675a4ebSAl Viro return PTR_ERR(dentry); 298154ef4872SMiklos Szeredi } 298254ef4872SMiklos Szeredi 2983d58ffd35SMiklos Szeredi /* Negative dentry, just create the file */ 2984d58ffd35SMiklos Szeredi if (!dentry->d_inode && (op->open_flag & O_CREAT)) { 2985d58ffd35SMiklos Szeredi umode_t mode = op->mode; 2986d58ffd35SMiklos Szeredi if (!IS_POSIXACL(dir->d_inode)) 2987d58ffd35SMiklos Szeredi mode &= ~current_umask(); 2988d58ffd35SMiklos Szeredi /* 2989d58ffd35SMiklos Szeredi * This write is needed to ensure that a 2990d58ffd35SMiklos Szeredi * rw->ro transition does not occur between 2991d58ffd35SMiklos Szeredi * the time when the file is created and when 2992d58ffd35SMiklos Szeredi * a permanent write count is taken through 2993015c3bbcSMiklos Szeredi * the 'struct file' in finish_open(). 2994d58ffd35SMiklos Szeredi */ 299564894cf8SAl Viro if (!got_write) { 299664894cf8SAl Viro error = -EROFS; 2997d58ffd35SMiklos Szeredi goto out_dput; 299864894cf8SAl Viro } 299947237687SAl Viro *opened |= FILE_CREATED; 3000d58ffd35SMiklos Szeredi error = security_path_mknod(&nd->path, dentry, mode, 0); 3001d58ffd35SMiklos Szeredi if (error) 3002d58ffd35SMiklos Szeredi goto out_dput; 3003312b63fbSAl Viro error = vfs_create(dir->d_inode, dentry, mode, 3004312b63fbSAl Viro nd->flags & LOOKUP_EXCL); 3005d58ffd35SMiklos Szeredi if (error) 3006d58ffd35SMiklos Szeredi goto out_dput; 3007d58ffd35SMiklos Szeredi } 3008d18e9008SMiklos Szeredi out_no_open: 3009d58ffd35SMiklos Szeredi path->dentry = dentry; 3010d58ffd35SMiklos Szeredi path->mnt = nd->path.mnt; 30112675a4ebSAl Viro return 1; 3012d58ffd35SMiklos Szeredi 3013d58ffd35SMiklos Szeredi out_dput: 3014d58ffd35SMiklos Szeredi dput(dentry); 30152675a4ebSAl Viro return error; 3016d58ffd35SMiklos Szeredi } 3017d58ffd35SMiklos Szeredi 3018d58ffd35SMiklos Szeredi /* 3019fe2d35ffSAl Viro * Handle the last step of open() 302031e6b01fSNick Piggin */ 3021896475d5SAl Viro static int do_last(struct nameidata *nd, 302230d90494SAl Viro struct file *file, const struct open_flags *op, 3023669abf4eSJeff Layton int *opened, struct filename *name) 3024fb1cc555SAl Viro { 3025a1e28038SAl Viro struct dentry *dir = nd->path.dentry; 3026ca344a89SAl Viro int open_flag = op->open_flag; 302777d660a8SMiklos Szeredi bool will_truncate = (open_flag & O_TRUNC) != 0; 302864894cf8SAl Viro bool got_write = false; 3029bcda7652SAl Viro int acc_mode = op->acc_mode; 3030254cf582SAl Viro unsigned seq; 3031a1eb3315SMiklos Szeredi struct inode *inode; 303216b1c1cdSMiklos Szeredi struct path save_parent = { .dentry = NULL, .mnt = NULL }; 3033896475d5SAl Viro struct path path; 303416b1c1cdSMiklos Szeredi bool retried = false; 303516c2cd71SAl Viro int error; 3036fb1cc555SAl Viro 3037c3e380b0SAl Viro nd->flags &= ~LOOKUP_PARENT; 3038c3e380b0SAl Viro nd->flags |= op->intent; 3039c3e380b0SAl Viro 3040bc77daa7SAl Viro if (nd->last_type != LAST_NORM) { 3041fe2d35ffSAl Viro error = handle_dots(nd, nd->last_type); 3042deb106c6SAl Viro if (unlikely(error)) 30432675a4ebSAl Viro return error; 3044e83db167SMiklos Szeredi goto finish_open; 30451f36f774SAl Viro } 3046a2c36b45SAl Viro 3047ca344a89SAl Viro if (!(open_flag & O_CREAT)) { 3048fe2d35ffSAl Viro if (nd->last.name[nd->last.len]) 3049fe2d35ffSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 3050fe2d35ffSAl Viro /* we _can_ be in RCU mode here */ 3051254cf582SAl Viro error = lookup_fast(nd, &path, &inode, &seq); 305271574865SMiklos Szeredi if (likely(!error)) 305371574865SMiklos Szeredi goto finish_lookup; 305471574865SMiklos Szeredi 3055ce57dfc1SAl Viro if (error < 0) 3056deb106c6SAl Viro return error; 3057a1eb3315SMiklos Szeredi 305837d7fffcSMiklos Szeredi BUG_ON(nd->inode != dir->d_inode); 3059b6183df7SMiklos Szeredi } else { 3060fe2d35ffSAl Viro /* create side of things */ 3061a3fbbde7SAl Viro /* 3062b6183df7SMiklos Szeredi * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED 3063b6183df7SMiklos Szeredi * has been cleared when we got to the last component we are 3064b6183df7SMiklos Szeredi * about to look up 3065a3fbbde7SAl Viro */ 30669f1fafeeSAl Viro error = complete_walk(nd); 3067e8bb73dfSAl Viro if (error) 30682675a4ebSAl Viro return error; 3069fe2d35ffSAl Viro 307033e2208aSJeff Layton audit_inode(name, dir, LOOKUP_PARENT); 30711f36f774SAl Viro /* trailing slashes? */ 3072deb106c6SAl Viro if (unlikely(nd->last.name[nd->last.len])) 3073deb106c6SAl Viro return -EISDIR; 3074b6183df7SMiklos Szeredi } 30751f36f774SAl Viro 307616b1c1cdSMiklos Szeredi retry_lookup: 307764894cf8SAl Viro if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { 307864894cf8SAl Viro error = mnt_want_write(nd->path.mnt); 307964894cf8SAl Viro if (!error) 308064894cf8SAl Viro got_write = true; 308164894cf8SAl Viro /* 308264894cf8SAl Viro * do _not_ fail yet - we might not need that or fail with 308364894cf8SAl Viro * a different error; let lookup_open() decide; we'll be 308464894cf8SAl Viro * dropping this one anyway. 308564894cf8SAl Viro */ 308664894cf8SAl Viro } 3087a1e28038SAl Viro mutex_lock(&dir->d_inode->i_mutex); 3088896475d5SAl Viro error = lookup_open(nd, &path, file, op, got_write, opened); 3089fb1cc555SAl Viro mutex_unlock(&dir->d_inode->i_mutex); 3090fb1cc555SAl Viro 30912675a4ebSAl Viro if (error <= 0) { 30922675a4ebSAl Viro if (error) 3093d58ffd35SMiklos Szeredi goto out; 30946c0d46c4SAl Viro 309547237687SAl Viro if ((*opened & FILE_CREATED) || 3096496ad9aaSAl Viro !S_ISREG(file_inode(file)->i_mode)) 309777d660a8SMiklos Szeredi will_truncate = false; 3098d18e9008SMiklos Szeredi 3099adb5c247SJeff Layton audit_inode(name, file->f_path.dentry, 0); 3100d18e9008SMiklos Szeredi goto opened; 3101d18e9008SMiklos Szeredi } 3102d18e9008SMiklos Szeredi 310347237687SAl Viro if (*opened & FILE_CREATED) { 31049b44f1b3SAl Viro /* Don't check for write permission, don't truncate */ 3105ca344a89SAl Viro open_flag &= ~O_TRUNC; 310677d660a8SMiklos Szeredi will_truncate = false; 3107bcda7652SAl Viro acc_mode = MAY_OPEN; 3108896475d5SAl Viro path_to_nameidata(&path, nd); 3109e83db167SMiklos Szeredi goto finish_open_created; 3110fb1cc555SAl Viro } 3111fb1cc555SAl Viro 3112fb1cc555SAl Viro /* 31133134f37eSJeff Layton * create/update audit record if it already exists. 3114fb1cc555SAl Viro */ 3115896475d5SAl Viro if (d_is_positive(path.dentry)) 3116896475d5SAl Viro audit_inode(name, path.dentry, 0); 3117fb1cc555SAl Viro 3118d18e9008SMiklos Szeredi /* 3119d18e9008SMiklos Szeredi * If atomic_open() acquired write access it is dropped now due to 3120d18e9008SMiklos Szeredi * possible mount and symlink following (this might be optimized away if 3121d18e9008SMiklos Szeredi * necessary...) 3122d18e9008SMiklos Szeredi */ 312364894cf8SAl Viro if (got_write) { 3124d18e9008SMiklos Szeredi mnt_drop_write(nd->path.mnt); 312564894cf8SAl Viro got_write = false; 3126d18e9008SMiklos Szeredi } 3127d18e9008SMiklos Szeredi 3128deb106c6SAl Viro if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) { 3129deb106c6SAl Viro path_to_nameidata(&path, nd); 3130deb106c6SAl Viro return -EEXIST; 3131deb106c6SAl Viro } 3132fb1cc555SAl Viro 3133756daf26SNeilBrown error = follow_managed(&path, nd); 3134deb106c6SAl Viro if (unlikely(error < 0)) 3135deb106c6SAl Viro return error; 3136a3fbbde7SAl Viro 3137decf3400SMiklos Szeredi BUG_ON(nd->flags & LOOKUP_RCU); 313863afdfc7SDavid Howells inode = d_backing_inode(path.dentry); 3139254cf582SAl Viro seq = 0; /* out of RCU mode, so the value doesn't matter */ 3140deb106c6SAl Viro if (unlikely(d_is_negative(path.dentry))) { 3141896475d5SAl Viro path_to_nameidata(&path, nd); 3142deb106c6SAl Viro return -ENOENT; 314354c33e7fSMiklos Szeredi } 3144766c4cbfSAl Viro finish_lookup: 3145191d7f73SAl Viro if (nd->depth) 3146191d7f73SAl Viro put_link(nd); 3147181548c0SAl Viro error = should_follow_link(nd, &path, nd->flags & LOOKUP_FOLLOW, 3148181548c0SAl Viro inode, seq); 3149deb106c6SAl Viro if (unlikely(error)) 3150d63ff28fSAl Viro return error; 3151fb1cc555SAl Viro 3152896475d5SAl Viro if (unlikely(d_is_symlink(path.dentry)) && !(open_flag & O_PATH)) { 3153896475d5SAl Viro path_to_nameidata(&path, nd); 3154deb106c6SAl Viro return -ELOOP; 3155a5cfe2d5SAl Viro } 3156a5cfe2d5SAl Viro 3157896475d5SAl Viro if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path.mnt) { 3158896475d5SAl Viro path_to_nameidata(&path, nd); 315916b1c1cdSMiklos Szeredi } else { 316016b1c1cdSMiklos Szeredi save_parent.dentry = nd->path.dentry; 3161896475d5SAl Viro save_parent.mnt = mntget(path.mnt); 3162896475d5SAl Viro nd->path.dentry = path.dentry; 316316b1c1cdSMiklos Szeredi 316416b1c1cdSMiklos Szeredi } 3165decf3400SMiklos Szeredi nd->inode = inode; 3166254cf582SAl Viro nd->seq = seq; 3167a3fbbde7SAl Viro /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ 3168bc77daa7SAl Viro finish_open: 3169a3fbbde7SAl Viro error = complete_walk(nd); 317016b1c1cdSMiklos Szeredi if (error) { 317116b1c1cdSMiklos Szeredi path_put(&save_parent); 31722675a4ebSAl Viro return error; 317316b1c1cdSMiklos Szeredi } 3174bc77daa7SAl Viro audit_inode(name, nd->path.dentry, 0); 3175fb1cc555SAl Viro error = -EISDIR; 317644b1d530SMiklos Szeredi if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry)) 31772675a4ebSAl Viro goto out; 3178af2f5542SMiklos Szeredi error = -ENOTDIR; 317944b1d530SMiklos Szeredi if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry)) 31802675a4ebSAl Viro goto out; 31814bbcbd3bSDavid Howells if (!d_is_reg(nd->path.dentry)) 318277d660a8SMiklos Szeredi will_truncate = false; 31836c0d46c4SAl Viro 31840f9d1a10SAl Viro if (will_truncate) { 31850f9d1a10SAl Viro error = mnt_want_write(nd->path.mnt); 31860f9d1a10SAl Viro if (error) 31872675a4ebSAl Viro goto out; 318864894cf8SAl Viro got_write = true; 31890f9d1a10SAl Viro } 3190e83db167SMiklos Szeredi finish_open_created: 3191bcda7652SAl Viro error = may_open(&nd->path, acc_mode, open_flag); 3192ca344a89SAl Viro if (error) 31932675a4ebSAl Viro goto out; 31944aa7c634SMiklos Szeredi 31954aa7c634SMiklos Szeredi BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */ 31964aa7c634SMiklos Szeredi error = vfs_open(&nd->path, file, current_cred()); 31974aa7c634SMiklos Szeredi if (!error) { 31984aa7c634SMiklos Szeredi *opened |= FILE_OPENED; 31994aa7c634SMiklos Szeredi } else { 320030d90494SAl Viro if (error == -EOPENSTALE) 3201f60dc3dbSMiklos Szeredi goto stale_open; 3202015c3bbcSMiklos Szeredi goto out; 3203f60dc3dbSMiklos Szeredi } 3204a8277b9bSMiklos Szeredi opened: 32052675a4ebSAl Viro error = open_check_o_direct(file); 3206015c3bbcSMiklos Szeredi if (error) 3207015c3bbcSMiklos Szeredi goto exit_fput; 32083034a146SDmitry Kasatkin error = ima_file_check(file, op->acc_mode, *opened); 3209aa4caadbSMiklos Szeredi if (error) 3210aa4caadbSMiklos Szeredi goto exit_fput; 3211aa4caadbSMiklos Szeredi 32120f9d1a10SAl Viro if (will_truncate) { 32132675a4ebSAl Viro error = handle_truncate(file); 3214aa4caadbSMiklos Szeredi if (error) 3215aa4caadbSMiklos Szeredi goto exit_fput; 32160f9d1a10SAl Viro } 3217ca344a89SAl Viro out: 321864894cf8SAl Viro if (got_write) 32190f9d1a10SAl Viro mnt_drop_write(nd->path.mnt); 322016b1c1cdSMiklos Szeredi path_put(&save_parent); 32212675a4ebSAl Viro return error; 3222fb1cc555SAl Viro 3223015c3bbcSMiklos Szeredi exit_fput: 32242675a4ebSAl Viro fput(file); 32252675a4ebSAl Viro goto out; 3226015c3bbcSMiklos Szeredi 3227f60dc3dbSMiklos Szeredi stale_open: 3228f60dc3dbSMiklos Szeredi /* If no saved parent or already retried then can't retry */ 3229f60dc3dbSMiklos Szeredi if (!save_parent.dentry || retried) 3230f60dc3dbSMiklos Szeredi goto out; 3231f60dc3dbSMiklos Szeredi 3232f60dc3dbSMiklos Szeredi BUG_ON(save_parent.dentry != dir); 3233f60dc3dbSMiklos Szeredi path_put(&nd->path); 3234f60dc3dbSMiklos Szeredi nd->path = save_parent; 3235f60dc3dbSMiklos Szeredi nd->inode = dir->d_inode; 3236f60dc3dbSMiklos Szeredi save_parent.mnt = NULL; 3237f60dc3dbSMiklos Szeredi save_parent.dentry = NULL; 323864894cf8SAl Viro if (got_write) { 3239f60dc3dbSMiklos Szeredi mnt_drop_write(nd->path.mnt); 324064894cf8SAl Viro got_write = false; 3241f60dc3dbSMiklos Szeredi } 3242f60dc3dbSMiklos Szeredi retried = true; 3243f60dc3dbSMiklos Szeredi goto retry_lookup; 3244fb1cc555SAl Viro } 3245fb1cc555SAl Viro 324660545d0dSAl Viro static int do_tmpfile(int dfd, struct filename *pathname, 324760545d0dSAl Viro struct nameidata *nd, int flags, 324860545d0dSAl Viro const struct open_flags *op, 324960545d0dSAl Viro struct file *file, int *opened) 325060545d0dSAl Viro { 325160545d0dSAl Viro static const struct qstr name = QSTR_INIT("/", 1); 3252625b6d10SAl Viro struct dentry *child; 325360545d0dSAl Viro struct inode *dir; 3254625b6d10SAl Viro struct path path; 32555eb6b495SAl Viro int error = path_lookupat(dfd, pathname, 3256625b6d10SAl Viro flags | LOOKUP_DIRECTORY, nd, &path); 325760545d0dSAl Viro if (unlikely(error)) 325860545d0dSAl Viro return error; 3259625b6d10SAl Viro error = mnt_want_write(path.mnt); 326060545d0dSAl Viro if (unlikely(error)) 326160545d0dSAl Viro goto out; 3262625b6d10SAl Viro dir = path.dentry->d_inode; 326360545d0dSAl Viro /* we want directory to be writable */ 3264625b6d10SAl Viro error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 326560545d0dSAl Viro if (error) 326660545d0dSAl Viro goto out2; 326760545d0dSAl Viro if (!dir->i_op->tmpfile) { 326860545d0dSAl Viro error = -EOPNOTSUPP; 326960545d0dSAl Viro goto out2; 327060545d0dSAl Viro } 3271625b6d10SAl Viro child = d_alloc(path.dentry, &name); 327260545d0dSAl Viro if (unlikely(!child)) { 327360545d0dSAl Viro error = -ENOMEM; 327460545d0dSAl Viro goto out2; 327560545d0dSAl Viro } 3276625b6d10SAl Viro dput(path.dentry); 3277625b6d10SAl Viro path.dentry = child; 3278625b6d10SAl Viro error = dir->i_op->tmpfile(dir, child, op->mode); 327960545d0dSAl Viro if (error) 328060545d0dSAl Viro goto out2; 3281625b6d10SAl Viro audit_inode(pathname, child, 0); 328269a91c23SEric Rannaud /* Don't check for other permissions, the inode was just created */ 3283625b6d10SAl Viro error = may_open(&path, MAY_OPEN, op->open_flag); 328460545d0dSAl Viro if (error) 328560545d0dSAl Viro goto out2; 3286625b6d10SAl Viro file->f_path.mnt = path.mnt; 3287625b6d10SAl Viro error = finish_open(file, child, NULL, opened); 328860545d0dSAl Viro if (error) 328960545d0dSAl Viro goto out2; 329060545d0dSAl Viro error = open_check_o_direct(file); 3291f4e0c30cSAl Viro if (error) { 329260545d0dSAl Viro fput(file); 3293f4e0c30cSAl Viro } else if (!(op->open_flag & O_EXCL)) { 3294f4e0c30cSAl Viro struct inode *inode = file_inode(file); 3295f4e0c30cSAl Viro spin_lock(&inode->i_lock); 3296f4e0c30cSAl Viro inode->i_state |= I_LINKABLE; 3297f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 3298f4e0c30cSAl Viro } 329960545d0dSAl Viro out2: 3300625b6d10SAl Viro mnt_drop_write(path.mnt); 330160545d0dSAl Viro out: 3302625b6d10SAl Viro path_put(&path); 330360545d0dSAl Viro return error; 330460545d0dSAl Viro } 330560545d0dSAl Viro 3306669abf4eSJeff Layton static struct file *path_openat(int dfd, struct filename *pathname, 330773d049a4SAl Viro struct nameidata *nd, const struct open_flags *op, int flags) 33081da177e4SLinus Torvalds { 3309368ee9baSAl Viro const char *s; 331030d90494SAl Viro struct file *file; 331147237687SAl Viro int opened = 0; 331213aab428SAl Viro int error; 331331e6b01fSNick Piggin 331430d90494SAl Viro file = get_empty_filp(); 33151afc99beSAl Viro if (IS_ERR(file)) 33161afc99beSAl Viro return file; 331731e6b01fSNick Piggin 331830d90494SAl Viro file->f_flags = op->open_flag; 331931e6b01fSNick Piggin 3320bb458c64SAl Viro if (unlikely(file->f_flags & __O_TMPFILE)) { 332160545d0dSAl Viro error = do_tmpfile(dfd, pathname, nd, flags, op, file, &opened); 3322f15133dfSAl Viro goto out2; 332360545d0dSAl Viro } 332460545d0dSAl Viro 3325368ee9baSAl Viro s = path_init(dfd, pathname, flags, nd); 3326368ee9baSAl Viro if (IS_ERR(s)) { 3327368ee9baSAl Viro put_filp(file); 3328368ee9baSAl Viro return ERR_CAST(s); 3329368ee9baSAl Viro } 33303bdba28bSAl Viro while (!(error = link_path_walk(s, nd)) && 33313bdba28bSAl Viro (error = do_last(nd, file, op, &opened, pathname)) > 0) { 333273d049a4SAl Viro nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); 33333bdba28bSAl Viro s = trailing_symlink(nd); 33343bdba28bSAl Viro if (IS_ERR(s)) { 33353bdba28bSAl Viro error = PTR_ERR(s); 33362675a4ebSAl Viro break; 3337806b681cSAl Viro } 33383bdba28bSAl Viro } 3339deb106c6SAl Viro terminate_walk(nd); 3340893b7775SAl Viro path_cleanup(nd); 3341f15133dfSAl Viro out2: 33422675a4ebSAl Viro if (!(opened & FILE_OPENED)) { 33432675a4ebSAl Viro BUG_ON(!error); 334430d90494SAl Viro put_filp(file); 3345015c3bbcSMiklos Szeredi } 33462675a4ebSAl Viro if (unlikely(error)) { 33472675a4ebSAl Viro if (error == -EOPENSTALE) { 33482675a4ebSAl Viro if (flags & LOOKUP_RCU) 33492675a4ebSAl Viro error = -ECHILD; 33502675a4ebSAl Viro else 33512675a4ebSAl Viro error = -ESTALE; 33522675a4ebSAl Viro } 33532675a4ebSAl Viro file = ERR_PTR(error); 33542675a4ebSAl Viro } 33552675a4ebSAl Viro return file; 3356de459215SKirill Korotaev } 33571da177e4SLinus Torvalds 3358669abf4eSJeff Layton struct file *do_filp_open(int dfd, struct filename *pathname, 3359f9652e10SAl Viro const struct open_flags *op) 336013aab428SAl Viro { 3361756daf26SNeilBrown struct nameidata nd, *saved_nd = set_nameidata(&nd); 3362f9652e10SAl Viro int flags = op->lookup_flags; 336313aab428SAl Viro struct file *filp; 336413aab428SAl Viro 336573d049a4SAl Viro filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU); 336613aab428SAl Viro if (unlikely(filp == ERR_PTR(-ECHILD))) 336773d049a4SAl Viro filp = path_openat(dfd, pathname, &nd, op, flags); 336813aab428SAl Viro if (unlikely(filp == ERR_PTR(-ESTALE))) 336973d049a4SAl Viro filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL); 3370756daf26SNeilBrown restore_nameidata(saved_nd); 337113aab428SAl Viro return filp; 337213aab428SAl Viro } 337313aab428SAl Viro 337473d049a4SAl Viro struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, 3375f9652e10SAl Viro const char *name, const struct open_flags *op) 337673d049a4SAl Viro { 3377756daf26SNeilBrown struct nameidata nd, *saved_nd; 337873d049a4SAl Viro struct file *file; 337951689104SPaul Moore struct filename *filename; 3380f9652e10SAl Viro int flags = op->lookup_flags | LOOKUP_ROOT; 338173d049a4SAl Viro 338273d049a4SAl Viro nd.root.mnt = mnt; 338373d049a4SAl Viro nd.root.dentry = dentry; 338473d049a4SAl Viro 3385b18825a7SDavid Howells if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN) 338673d049a4SAl Viro return ERR_PTR(-ELOOP); 338773d049a4SAl Viro 338851689104SPaul Moore filename = getname_kernel(name); 338951689104SPaul Moore if (unlikely(IS_ERR(filename))) 339051689104SPaul Moore return ERR_CAST(filename); 339151689104SPaul Moore 3392756daf26SNeilBrown saved_nd = set_nameidata(&nd); 339351689104SPaul Moore file = path_openat(-1, filename, &nd, op, flags | LOOKUP_RCU); 339473d049a4SAl Viro if (unlikely(file == ERR_PTR(-ECHILD))) 339551689104SPaul Moore file = path_openat(-1, filename, &nd, op, flags); 339673d049a4SAl Viro if (unlikely(file == ERR_PTR(-ESTALE))) 339751689104SPaul Moore file = path_openat(-1, filename, &nd, op, flags | LOOKUP_REVAL); 3398756daf26SNeilBrown restore_nameidata(saved_nd); 339951689104SPaul Moore putname(filename); 340073d049a4SAl Viro return file; 340173d049a4SAl Viro } 340273d049a4SAl Viro 3403fa14a0b8SAl Viro static struct dentry *filename_create(int dfd, struct filename *name, 34041ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 34051da177e4SLinus Torvalds { 3406c663e5d8SChristoph Hellwig struct dentry *dentry = ERR_PTR(-EEXIST); 3407ed75e95dSAl Viro struct nameidata nd; 3408c30dabfeSJan Kara int err2; 34091ac12b4bSJeff Layton int error; 34101ac12b4bSJeff Layton bool is_dir = (lookup_flags & LOOKUP_DIRECTORY); 34111ac12b4bSJeff Layton 34121ac12b4bSJeff Layton /* 34131ac12b4bSJeff Layton * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any 34141ac12b4bSJeff Layton * other flags passed in are ignored! 34151ac12b4bSJeff Layton */ 34161ac12b4bSJeff Layton lookup_flags &= LOOKUP_REVAL; 34171ac12b4bSJeff Layton 34188bcb77faSAl Viro error = filename_parentat(dfd, name, lookup_flags, &nd); 3419ed75e95dSAl Viro if (error) 3420ed75e95dSAl Viro return ERR_PTR(error); 34211da177e4SLinus Torvalds 3422c663e5d8SChristoph Hellwig /* 3423c663e5d8SChristoph Hellwig * Yucky last component or no last component at all? 3424c663e5d8SChristoph Hellwig * (foo/., foo/.., /////) 3425c663e5d8SChristoph Hellwig */ 3426ed75e95dSAl Viro if (nd.last_type != LAST_NORM) 3427ed75e95dSAl Viro goto out; 3428ed75e95dSAl Viro nd.flags &= ~LOOKUP_PARENT; 3429ed75e95dSAl Viro nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL; 3430c663e5d8SChristoph Hellwig 3431c30dabfeSJan Kara /* don't fail immediately if it's r/o, at least try to report other errors */ 3432c30dabfeSJan Kara err2 = mnt_want_write(nd.path.mnt); 3433c663e5d8SChristoph Hellwig /* 3434c663e5d8SChristoph Hellwig * Do the final lookup. 3435c663e5d8SChristoph Hellwig */ 3436ed75e95dSAl Viro mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 34376a9f40d6SAl Viro dentry = __lookup_hash(&nd.last, nd.path.dentry, nd.flags); 34381da177e4SLinus Torvalds if (IS_ERR(dentry)) 3439a8104a9fSAl Viro goto unlock; 3440c663e5d8SChristoph Hellwig 3441a8104a9fSAl Viro error = -EEXIST; 3442b18825a7SDavid Howells if (d_is_positive(dentry)) 3443a8104a9fSAl Viro goto fail; 3444b18825a7SDavid Howells 3445c663e5d8SChristoph Hellwig /* 3446c663e5d8SChristoph Hellwig * Special case - lookup gave negative, but... we had foo/bar/ 3447c663e5d8SChristoph Hellwig * From the vfs_mknod() POV we just have a negative dentry - 3448c663e5d8SChristoph Hellwig * all is fine. Let's be bastards - you had / on the end, you've 3449c663e5d8SChristoph Hellwig * been asking for (non-existent) directory. -ENOENT for you. 3450c663e5d8SChristoph Hellwig */ 3451ed75e95dSAl Viro if (unlikely(!is_dir && nd.last.name[nd.last.len])) { 3452a8104a9fSAl Viro error = -ENOENT; 3453ed75e95dSAl Viro goto fail; 3454e9baf6e5SAl Viro } 3455c30dabfeSJan Kara if (unlikely(err2)) { 3456c30dabfeSJan Kara error = err2; 3457a8104a9fSAl Viro goto fail; 3458c30dabfeSJan Kara } 3459ed75e95dSAl Viro *path = nd.path; 3460e9baf6e5SAl Viro return dentry; 34611da177e4SLinus Torvalds fail: 3462a8104a9fSAl Viro dput(dentry); 3463a8104a9fSAl Viro dentry = ERR_PTR(error); 3464a8104a9fSAl Viro unlock: 3465dae6ad8fSAl Viro mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 3466c30dabfeSJan Kara if (!err2) 3467c30dabfeSJan Kara mnt_drop_write(nd.path.mnt); 3468ed75e95dSAl Viro out: 3469dae6ad8fSAl Viro path_put(&nd.path); 3470ed75e95dSAl Viro return dentry; 3471dae6ad8fSAl Viro } 3472fa14a0b8SAl Viro 3473fa14a0b8SAl Viro struct dentry *kern_path_create(int dfd, const char *pathname, 3474fa14a0b8SAl Viro struct path *path, unsigned int lookup_flags) 3475fa14a0b8SAl Viro { 347651689104SPaul Moore struct filename *filename = getname_kernel(pathname); 347751689104SPaul Moore struct dentry *res; 347851689104SPaul Moore 347951689104SPaul Moore if (IS_ERR(filename)) 348051689104SPaul Moore return ERR_CAST(filename); 348151689104SPaul Moore res = filename_create(dfd, filename, path, lookup_flags); 348251689104SPaul Moore putname(filename); 348351689104SPaul Moore return res; 3484fa14a0b8SAl Viro } 3485dae6ad8fSAl Viro EXPORT_SYMBOL(kern_path_create); 3486dae6ad8fSAl Viro 3487921a1650SAl Viro void done_path_create(struct path *path, struct dentry *dentry) 3488921a1650SAl Viro { 3489921a1650SAl Viro dput(dentry); 3490921a1650SAl Viro mutex_unlock(&path->dentry->d_inode->i_mutex); 3491a8104a9fSAl Viro mnt_drop_write(path->mnt); 3492921a1650SAl Viro path_put(path); 3493921a1650SAl Viro } 3494921a1650SAl Viro EXPORT_SYMBOL(done_path_create); 3495921a1650SAl Viro 34961ac12b4bSJeff Layton struct dentry *user_path_create(int dfd, const char __user *pathname, 34971ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 3498dae6ad8fSAl Viro { 349991a27b2aSJeff Layton struct filename *tmp = getname(pathname); 3500dae6ad8fSAl Viro struct dentry *res; 3501dae6ad8fSAl Viro if (IS_ERR(tmp)) 3502dae6ad8fSAl Viro return ERR_CAST(tmp); 3503fa14a0b8SAl Viro res = filename_create(dfd, tmp, path, lookup_flags); 3504dae6ad8fSAl Viro putname(tmp); 3505dae6ad8fSAl Viro return res; 3506dae6ad8fSAl Viro } 3507dae6ad8fSAl Viro EXPORT_SYMBOL(user_path_create); 3508dae6ad8fSAl Viro 35091a67aafbSAl Viro int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) 35101da177e4SLinus Torvalds { 3511a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 35121da177e4SLinus Torvalds 35131da177e4SLinus Torvalds if (error) 35141da177e4SLinus Torvalds return error; 35151da177e4SLinus Torvalds 3516975d6b39SEric W. Biederman if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) 35171da177e4SLinus Torvalds return -EPERM; 35181da177e4SLinus Torvalds 3519acfa4380SAl Viro if (!dir->i_op->mknod) 35201da177e4SLinus Torvalds return -EPERM; 35211da177e4SLinus Torvalds 352208ce5f16SSerge E. Hallyn error = devcgroup_inode_mknod(mode, dev); 352308ce5f16SSerge E. Hallyn if (error) 352408ce5f16SSerge E. Hallyn return error; 352508ce5f16SSerge E. Hallyn 35261da177e4SLinus Torvalds error = security_inode_mknod(dir, dentry, mode, dev); 35271da177e4SLinus Torvalds if (error) 35281da177e4SLinus Torvalds return error; 35291da177e4SLinus Torvalds 35301da177e4SLinus Torvalds error = dir->i_op->mknod(dir, dentry, mode, dev); 3531a74574aaSStephen Smalley if (!error) 3532f38aa942SAmy Griffis fsnotify_create(dir, dentry); 35331da177e4SLinus Torvalds return error; 35341da177e4SLinus Torvalds } 35354d359507SAl Viro EXPORT_SYMBOL(vfs_mknod); 35361da177e4SLinus Torvalds 3537f69aac00SAl Viro static int may_mknod(umode_t mode) 3538463c3197SDave Hansen { 3539463c3197SDave Hansen switch (mode & S_IFMT) { 3540463c3197SDave Hansen case S_IFREG: 3541463c3197SDave Hansen case S_IFCHR: 3542463c3197SDave Hansen case S_IFBLK: 3543463c3197SDave Hansen case S_IFIFO: 3544463c3197SDave Hansen case S_IFSOCK: 3545463c3197SDave Hansen case 0: /* zero mode translates to S_IFREG */ 3546463c3197SDave Hansen return 0; 3547463c3197SDave Hansen case S_IFDIR: 3548463c3197SDave Hansen return -EPERM; 3549463c3197SDave Hansen default: 3550463c3197SDave Hansen return -EINVAL; 3551463c3197SDave Hansen } 3552463c3197SDave Hansen } 3553463c3197SDave Hansen 35548208a22bSAl Viro SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, 35552e4d0924SHeiko Carstens unsigned, dev) 35561da177e4SLinus Torvalds { 35571da177e4SLinus Torvalds struct dentry *dentry; 3558dae6ad8fSAl Viro struct path path; 3559dae6ad8fSAl Viro int error; 3560972567f1SJeff Layton unsigned int lookup_flags = 0; 35611da177e4SLinus Torvalds 35628e4bfca1SAl Viro error = may_mknod(mode); 35638e4bfca1SAl Viro if (error) 35648e4bfca1SAl Viro return error; 3565972567f1SJeff Layton retry: 3566972567f1SJeff Layton dentry = user_path_create(dfd, filename, &path, lookup_flags); 3567dae6ad8fSAl Viro if (IS_ERR(dentry)) 3568dae6ad8fSAl Viro return PTR_ERR(dentry); 35692ad94ae6SAl Viro 3570dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3571ce3b0f8dSAl Viro mode &= ~current_umask(); 3572dae6ad8fSAl Viro error = security_path_mknod(&path, dentry, mode, dev); 3573be6d3e56SKentaro Takeda if (error) 3574a8104a9fSAl Viro goto out; 35751da177e4SLinus Torvalds switch (mode & S_IFMT) { 35761da177e4SLinus Torvalds case 0: case S_IFREG: 3577312b63fbSAl Viro error = vfs_create(path.dentry->d_inode,dentry,mode,true); 35781da177e4SLinus Torvalds break; 35791da177e4SLinus Torvalds case S_IFCHR: case S_IFBLK: 3580dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode, 35811da177e4SLinus Torvalds new_decode_dev(dev)); 35821da177e4SLinus Torvalds break; 35831da177e4SLinus Torvalds case S_IFIFO: case S_IFSOCK: 3584dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); 35851da177e4SLinus Torvalds break; 35861da177e4SLinus Torvalds } 3587a8104a9fSAl Viro out: 3588921a1650SAl Viro done_path_create(&path, dentry); 3589972567f1SJeff Layton if (retry_estale(error, lookup_flags)) { 3590972567f1SJeff Layton lookup_flags |= LOOKUP_REVAL; 3591972567f1SJeff Layton goto retry; 3592972567f1SJeff Layton } 35931da177e4SLinus Torvalds return error; 35941da177e4SLinus Torvalds } 35951da177e4SLinus Torvalds 35968208a22bSAl Viro SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) 35975590ff0dSUlrich Drepper { 35985590ff0dSUlrich Drepper return sys_mknodat(AT_FDCWD, filename, mode, dev); 35995590ff0dSUlrich Drepper } 36005590ff0dSUlrich Drepper 360118bb1db3SAl Viro int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 36021da177e4SLinus Torvalds { 3603a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 36048de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 36051da177e4SLinus Torvalds 36061da177e4SLinus Torvalds if (error) 36071da177e4SLinus Torvalds return error; 36081da177e4SLinus Torvalds 3609acfa4380SAl Viro if (!dir->i_op->mkdir) 36101da177e4SLinus Torvalds return -EPERM; 36111da177e4SLinus Torvalds 36121da177e4SLinus Torvalds mode &= (S_IRWXUGO|S_ISVTX); 36131da177e4SLinus Torvalds error = security_inode_mkdir(dir, dentry, mode); 36141da177e4SLinus Torvalds if (error) 36151da177e4SLinus Torvalds return error; 36161da177e4SLinus Torvalds 36178de52778SAl Viro if (max_links && dir->i_nlink >= max_links) 36188de52778SAl Viro return -EMLINK; 36198de52778SAl Viro 36201da177e4SLinus Torvalds error = dir->i_op->mkdir(dir, dentry, mode); 3621a74574aaSStephen Smalley if (!error) 3622f38aa942SAmy Griffis fsnotify_mkdir(dir, dentry); 36231da177e4SLinus Torvalds return error; 36241da177e4SLinus Torvalds } 36254d359507SAl Viro EXPORT_SYMBOL(vfs_mkdir); 36261da177e4SLinus Torvalds 3627a218d0fdSAl Viro SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) 36281da177e4SLinus Torvalds { 36296902d925SDave Hansen struct dentry *dentry; 3630dae6ad8fSAl Viro struct path path; 3631dae6ad8fSAl Viro int error; 3632b76d8b82SJeff Layton unsigned int lookup_flags = LOOKUP_DIRECTORY; 36331da177e4SLinus Torvalds 3634b76d8b82SJeff Layton retry: 3635b76d8b82SJeff Layton dentry = user_path_create(dfd, pathname, &path, lookup_flags); 36366902d925SDave Hansen if (IS_ERR(dentry)) 3637dae6ad8fSAl Viro return PTR_ERR(dentry); 36386902d925SDave Hansen 3639dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3640ce3b0f8dSAl Viro mode &= ~current_umask(); 3641dae6ad8fSAl Viro error = security_path_mkdir(&path, dentry, mode); 3642a8104a9fSAl Viro if (!error) 3643dae6ad8fSAl Viro error = vfs_mkdir(path.dentry->d_inode, dentry, mode); 3644921a1650SAl Viro done_path_create(&path, dentry); 3645b76d8b82SJeff Layton if (retry_estale(error, lookup_flags)) { 3646b76d8b82SJeff Layton lookup_flags |= LOOKUP_REVAL; 3647b76d8b82SJeff Layton goto retry; 3648b76d8b82SJeff Layton } 36491da177e4SLinus Torvalds return error; 36501da177e4SLinus Torvalds } 36511da177e4SLinus Torvalds 3652a218d0fdSAl Viro SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) 36535590ff0dSUlrich Drepper { 36545590ff0dSUlrich Drepper return sys_mkdirat(AT_FDCWD, pathname, mode); 36555590ff0dSUlrich Drepper } 36565590ff0dSUlrich Drepper 36571da177e4SLinus Torvalds /* 3658a71905f0SSage Weil * The dentry_unhash() helper will try to drop the dentry early: we 3659c0d02594SJ. Bruce Fields * should have a usage count of 1 if we're the only user of this 3660a71905f0SSage Weil * dentry, and if that is true (possibly after pruning the dcache), 3661a71905f0SSage Weil * then we drop the dentry now. 36621da177e4SLinus Torvalds * 36631da177e4SLinus Torvalds * A low-level filesystem can, if it choses, legally 36641da177e4SLinus Torvalds * do a 36651da177e4SLinus Torvalds * 36661da177e4SLinus Torvalds * if (!d_unhashed(dentry)) 36671da177e4SLinus Torvalds * return -EBUSY; 36681da177e4SLinus Torvalds * 36691da177e4SLinus Torvalds * if it cannot handle the case of removing a directory 36701da177e4SLinus Torvalds * that is still in use by something else.. 36711da177e4SLinus Torvalds */ 36721da177e4SLinus Torvalds void dentry_unhash(struct dentry *dentry) 36731da177e4SLinus Torvalds { 36741da177e4SLinus Torvalds shrink_dcache_parent(dentry); 36751da177e4SLinus Torvalds spin_lock(&dentry->d_lock); 367698474236SWaiman Long if (dentry->d_lockref.count == 1) 36771da177e4SLinus Torvalds __d_drop(dentry); 36781da177e4SLinus Torvalds spin_unlock(&dentry->d_lock); 36791da177e4SLinus Torvalds } 36804d359507SAl Viro EXPORT_SYMBOL(dentry_unhash); 36811da177e4SLinus Torvalds 36821da177e4SLinus Torvalds int vfs_rmdir(struct inode *dir, struct dentry *dentry) 36831da177e4SLinus Torvalds { 36841da177e4SLinus Torvalds int error = may_delete(dir, dentry, 1); 36851da177e4SLinus Torvalds 36861da177e4SLinus Torvalds if (error) 36871da177e4SLinus Torvalds return error; 36881da177e4SLinus Torvalds 3689acfa4380SAl Viro if (!dir->i_op->rmdir) 36901da177e4SLinus Torvalds return -EPERM; 36911da177e4SLinus Torvalds 36921d2ef590SAl Viro dget(dentry); 36931b1dcc1bSJes Sorensen mutex_lock(&dentry->d_inode->i_mutex); 3694912dbc15SSage Weil 36951da177e4SLinus Torvalds error = -EBUSY; 36967af1364fSEric W. Biederman if (is_local_mountpoint(dentry)) 3697912dbc15SSage Weil goto out; 3698912dbc15SSage Weil 36991da177e4SLinus Torvalds error = security_inode_rmdir(dir, dentry); 3700912dbc15SSage Weil if (error) 3701912dbc15SSage Weil goto out; 3702912dbc15SSage Weil 37033cebde24SSage Weil shrink_dcache_parent(dentry); 37041da177e4SLinus Torvalds error = dir->i_op->rmdir(dir, dentry); 3705912dbc15SSage Weil if (error) 3706912dbc15SSage Weil goto out; 3707912dbc15SSage Weil 37081da177e4SLinus Torvalds dentry->d_inode->i_flags |= S_DEAD; 3709d83c49f3SAl Viro dont_mount(dentry); 37108ed936b5SEric W. Biederman detach_mounts(dentry); 37111da177e4SLinus Torvalds 3712912dbc15SSage Weil out: 3713912dbc15SSage Weil mutex_unlock(&dentry->d_inode->i_mutex); 37141d2ef590SAl Viro dput(dentry); 3715912dbc15SSage Weil if (!error) 3716912dbc15SSage Weil d_delete(dentry); 37171da177e4SLinus Torvalds return error; 37181da177e4SLinus Torvalds } 37194d359507SAl Viro EXPORT_SYMBOL(vfs_rmdir); 37201da177e4SLinus Torvalds 37215590ff0dSUlrich Drepper static long do_rmdir(int dfd, const char __user *pathname) 37221da177e4SLinus Torvalds { 37231da177e4SLinus Torvalds int error = 0; 372491a27b2aSJeff Layton struct filename *name; 37251da177e4SLinus Torvalds struct dentry *dentry; 3726f5beed75SAl Viro struct path path; 3727f5beed75SAl Viro struct qstr last; 3728f5beed75SAl Viro int type; 3729c6ee9206SJeff Layton unsigned int lookup_flags = 0; 3730c6ee9206SJeff Layton retry: 3731f5beed75SAl Viro name = user_path_parent(dfd, pathname, 3732f5beed75SAl Viro &path, &last, &type, lookup_flags); 373391a27b2aSJeff Layton if (IS_ERR(name)) 373491a27b2aSJeff Layton return PTR_ERR(name); 37351da177e4SLinus Torvalds 3736f5beed75SAl Viro switch (type) { 37371da177e4SLinus Torvalds case LAST_DOTDOT: 37381da177e4SLinus Torvalds error = -ENOTEMPTY; 37391da177e4SLinus Torvalds goto exit1; 37401da177e4SLinus Torvalds case LAST_DOT: 37411da177e4SLinus Torvalds error = -EINVAL; 37421da177e4SLinus Torvalds goto exit1; 37431da177e4SLinus Torvalds case LAST_ROOT: 37441da177e4SLinus Torvalds error = -EBUSY; 37451da177e4SLinus Torvalds goto exit1; 37461da177e4SLinus Torvalds } 37470612d9fbSOGAWA Hirofumi 3748f5beed75SAl Viro error = mnt_want_write(path.mnt); 3749c30dabfeSJan Kara if (error) 3750c30dabfeSJan Kara goto exit1; 37510612d9fbSOGAWA Hirofumi 3752f5beed75SAl Viro mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 3753f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 37541da177e4SLinus Torvalds error = PTR_ERR(dentry); 37556902d925SDave Hansen if (IS_ERR(dentry)) 37566902d925SDave Hansen goto exit2; 3757e6bc45d6STheodore Ts'o if (!dentry->d_inode) { 3758e6bc45d6STheodore Ts'o error = -ENOENT; 3759e6bc45d6STheodore Ts'o goto exit3; 3760e6bc45d6STheodore Ts'o } 3761f5beed75SAl Viro error = security_path_rmdir(&path, dentry); 3762be6d3e56SKentaro Takeda if (error) 3763c30dabfeSJan Kara goto exit3; 3764f5beed75SAl Viro error = vfs_rmdir(path.dentry->d_inode, dentry); 37650622753bSDave Hansen exit3: 37661da177e4SLinus Torvalds dput(dentry); 37676902d925SDave Hansen exit2: 3768f5beed75SAl Viro mutex_unlock(&path.dentry->d_inode->i_mutex); 3769f5beed75SAl Viro mnt_drop_write(path.mnt); 37701da177e4SLinus Torvalds exit1: 3771f5beed75SAl Viro path_put(&path); 37721da177e4SLinus Torvalds putname(name); 3773c6ee9206SJeff Layton if (retry_estale(error, lookup_flags)) { 3774c6ee9206SJeff Layton lookup_flags |= LOOKUP_REVAL; 3775c6ee9206SJeff Layton goto retry; 3776c6ee9206SJeff Layton } 37771da177e4SLinus Torvalds return error; 37781da177e4SLinus Torvalds } 37791da177e4SLinus Torvalds 37803cdad428SHeiko Carstens SYSCALL_DEFINE1(rmdir, const char __user *, pathname) 37815590ff0dSUlrich Drepper { 37825590ff0dSUlrich Drepper return do_rmdir(AT_FDCWD, pathname); 37835590ff0dSUlrich Drepper } 37845590ff0dSUlrich Drepper 3785b21996e3SJ. Bruce Fields /** 3786b21996e3SJ. Bruce Fields * vfs_unlink - unlink a filesystem object 3787b21996e3SJ. Bruce Fields * @dir: parent directory 3788b21996e3SJ. Bruce Fields * @dentry: victim 3789b21996e3SJ. Bruce Fields * @delegated_inode: returns victim inode, if the inode is delegated. 3790b21996e3SJ. Bruce Fields * 3791b21996e3SJ. Bruce Fields * The caller must hold dir->i_mutex. 3792b21996e3SJ. Bruce Fields * 3793b21996e3SJ. Bruce Fields * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and 3794b21996e3SJ. Bruce Fields * return a reference to the inode in delegated_inode. The caller 3795b21996e3SJ. Bruce Fields * should then break the delegation on that inode and retry. Because 3796b21996e3SJ. Bruce Fields * breaking a delegation may take a long time, the caller should drop 3797b21996e3SJ. Bruce Fields * dir->i_mutex before doing so. 3798b21996e3SJ. Bruce Fields * 3799b21996e3SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 3800b21996e3SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 3801b21996e3SJ. Bruce Fields * to be NFS exported. 3802b21996e3SJ. Bruce Fields */ 3803b21996e3SJ. Bruce Fields int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode) 38041da177e4SLinus Torvalds { 38059accbb97SJ. Bruce Fields struct inode *target = dentry->d_inode; 38061da177e4SLinus Torvalds int error = may_delete(dir, dentry, 0); 38071da177e4SLinus Torvalds 38081da177e4SLinus Torvalds if (error) 38091da177e4SLinus Torvalds return error; 38101da177e4SLinus Torvalds 3811acfa4380SAl Viro if (!dir->i_op->unlink) 38121da177e4SLinus Torvalds return -EPERM; 38131da177e4SLinus Torvalds 38149accbb97SJ. Bruce Fields mutex_lock(&target->i_mutex); 38158ed936b5SEric W. Biederman if (is_local_mountpoint(dentry)) 38161da177e4SLinus Torvalds error = -EBUSY; 38171da177e4SLinus Torvalds else { 38181da177e4SLinus Torvalds error = security_inode_unlink(dir, dentry); 3819bec1052eSAl Viro if (!error) { 38205a14696cSJ. Bruce Fields error = try_break_deleg(target, delegated_inode); 38215a14696cSJ. Bruce Fields if (error) 3822b21996e3SJ. Bruce Fields goto out; 38231da177e4SLinus Torvalds error = dir->i_op->unlink(dir, dentry); 38248ed936b5SEric W. Biederman if (!error) { 3825d83c49f3SAl Viro dont_mount(dentry); 38268ed936b5SEric W. Biederman detach_mounts(dentry); 38278ed936b5SEric W. Biederman } 3828bec1052eSAl Viro } 38291da177e4SLinus Torvalds } 3830b21996e3SJ. Bruce Fields out: 38319accbb97SJ. Bruce Fields mutex_unlock(&target->i_mutex); 38321da177e4SLinus Torvalds 38331da177e4SLinus Torvalds /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 38341da177e4SLinus Torvalds if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { 38359accbb97SJ. Bruce Fields fsnotify_link_count(target); 38361da177e4SLinus Torvalds d_delete(dentry); 38371da177e4SLinus Torvalds } 38380eeca283SRobert Love 38391da177e4SLinus Torvalds return error; 38401da177e4SLinus Torvalds } 38414d359507SAl Viro EXPORT_SYMBOL(vfs_unlink); 38421da177e4SLinus Torvalds 38431da177e4SLinus Torvalds /* 38441da177e4SLinus Torvalds * Make sure that the actual truncation of the file will occur outside its 38451b1dcc1bSJes Sorensen * directory's i_mutex. Truncate can take a long time if there is a lot of 38461da177e4SLinus Torvalds * writeout happening, and we don't want to prevent access to the directory 38471da177e4SLinus Torvalds * while waiting on the I/O. 38481da177e4SLinus Torvalds */ 38495590ff0dSUlrich Drepper static long do_unlinkat(int dfd, const char __user *pathname) 38501da177e4SLinus Torvalds { 38512ad94ae6SAl Viro int error; 385291a27b2aSJeff Layton struct filename *name; 38531da177e4SLinus Torvalds struct dentry *dentry; 3854f5beed75SAl Viro struct path path; 3855f5beed75SAl Viro struct qstr last; 3856f5beed75SAl Viro int type; 38571da177e4SLinus Torvalds struct inode *inode = NULL; 3858b21996e3SJ. Bruce Fields struct inode *delegated_inode = NULL; 38595d18f813SJeff Layton unsigned int lookup_flags = 0; 38605d18f813SJeff Layton retry: 3861f5beed75SAl Viro name = user_path_parent(dfd, pathname, 3862f5beed75SAl Viro &path, &last, &type, lookup_flags); 386391a27b2aSJeff Layton if (IS_ERR(name)) 386491a27b2aSJeff Layton return PTR_ERR(name); 38652ad94ae6SAl Viro 38661da177e4SLinus Torvalds error = -EISDIR; 3867f5beed75SAl Viro if (type != LAST_NORM) 38681da177e4SLinus Torvalds goto exit1; 38690612d9fbSOGAWA Hirofumi 3870f5beed75SAl Viro error = mnt_want_write(path.mnt); 3871c30dabfeSJan Kara if (error) 3872c30dabfeSJan Kara goto exit1; 3873b21996e3SJ. Bruce Fields retry_deleg: 3874f5beed75SAl Viro mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 3875f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 38761da177e4SLinus Torvalds error = PTR_ERR(dentry); 38771da177e4SLinus Torvalds if (!IS_ERR(dentry)) { 38781da177e4SLinus Torvalds /* Why not before? Because we want correct error value */ 3879f5beed75SAl Viro if (last.name[last.len]) 388050338b88STörök Edwin goto slashes; 38811da177e4SLinus Torvalds inode = dentry->d_inode; 3882b18825a7SDavid Howells if (d_is_negative(dentry)) 3883e6bc45d6STheodore Ts'o goto slashes; 38847de9c6eeSAl Viro ihold(inode); 3885f5beed75SAl Viro error = security_path_unlink(&path, dentry); 3886be6d3e56SKentaro Takeda if (error) 3887c30dabfeSJan Kara goto exit2; 3888f5beed75SAl Viro error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode); 38891da177e4SLinus Torvalds exit2: 38901da177e4SLinus Torvalds dput(dentry); 38911da177e4SLinus Torvalds } 3892f5beed75SAl Viro mutex_unlock(&path.dentry->d_inode->i_mutex); 38931da177e4SLinus Torvalds if (inode) 38941da177e4SLinus Torvalds iput(inode); /* truncate the inode here */ 3895b21996e3SJ. Bruce Fields inode = NULL; 3896b21996e3SJ. Bruce Fields if (delegated_inode) { 38975a14696cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 3898b21996e3SJ. Bruce Fields if (!error) 3899b21996e3SJ. Bruce Fields goto retry_deleg; 3900b21996e3SJ. Bruce Fields } 3901f5beed75SAl Viro mnt_drop_write(path.mnt); 39021da177e4SLinus Torvalds exit1: 3903f5beed75SAl Viro path_put(&path); 39041da177e4SLinus Torvalds putname(name); 39055d18f813SJeff Layton if (retry_estale(error, lookup_flags)) { 39065d18f813SJeff Layton lookup_flags |= LOOKUP_REVAL; 39075d18f813SJeff Layton inode = NULL; 39085d18f813SJeff Layton goto retry; 39095d18f813SJeff Layton } 39101da177e4SLinus Torvalds return error; 39111da177e4SLinus Torvalds 39121da177e4SLinus Torvalds slashes: 3913b18825a7SDavid Howells if (d_is_negative(dentry)) 3914b18825a7SDavid Howells error = -ENOENT; 391544b1d530SMiklos Szeredi else if (d_is_dir(dentry)) 3916b18825a7SDavid Howells error = -EISDIR; 3917b18825a7SDavid Howells else 3918b18825a7SDavid Howells error = -ENOTDIR; 39191da177e4SLinus Torvalds goto exit2; 39201da177e4SLinus Torvalds } 39211da177e4SLinus Torvalds 39222e4d0924SHeiko Carstens SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) 39235590ff0dSUlrich Drepper { 39245590ff0dSUlrich Drepper if ((flag & ~AT_REMOVEDIR) != 0) 39255590ff0dSUlrich Drepper return -EINVAL; 39265590ff0dSUlrich Drepper 39275590ff0dSUlrich Drepper if (flag & AT_REMOVEDIR) 39285590ff0dSUlrich Drepper return do_rmdir(dfd, pathname); 39295590ff0dSUlrich Drepper 39305590ff0dSUlrich Drepper return do_unlinkat(dfd, pathname); 39315590ff0dSUlrich Drepper } 39325590ff0dSUlrich Drepper 39333480b257SHeiko Carstens SYSCALL_DEFINE1(unlink, const char __user *, pathname) 39345590ff0dSUlrich Drepper { 39355590ff0dSUlrich Drepper return do_unlinkat(AT_FDCWD, pathname); 39365590ff0dSUlrich Drepper } 39375590ff0dSUlrich Drepper 3938db2e747bSMiklos Szeredi int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) 39391da177e4SLinus Torvalds { 3940a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 39411da177e4SLinus Torvalds 39421da177e4SLinus Torvalds if (error) 39431da177e4SLinus Torvalds return error; 39441da177e4SLinus Torvalds 3945acfa4380SAl Viro if (!dir->i_op->symlink) 39461da177e4SLinus Torvalds return -EPERM; 39471da177e4SLinus Torvalds 39481da177e4SLinus Torvalds error = security_inode_symlink(dir, dentry, oldname); 39491da177e4SLinus Torvalds if (error) 39501da177e4SLinus Torvalds return error; 39511da177e4SLinus Torvalds 39521da177e4SLinus Torvalds error = dir->i_op->symlink(dir, dentry, oldname); 3953a74574aaSStephen Smalley if (!error) 3954f38aa942SAmy Griffis fsnotify_create(dir, dentry); 39551da177e4SLinus Torvalds return error; 39561da177e4SLinus Torvalds } 39574d359507SAl Viro EXPORT_SYMBOL(vfs_symlink); 39581da177e4SLinus Torvalds 39592e4d0924SHeiko Carstens SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, 39602e4d0924SHeiko Carstens int, newdfd, const char __user *, newname) 39611da177e4SLinus Torvalds { 39622ad94ae6SAl Viro int error; 396391a27b2aSJeff Layton struct filename *from; 39646902d925SDave Hansen struct dentry *dentry; 3965dae6ad8fSAl Viro struct path path; 3966f46d3567SJeff Layton unsigned int lookup_flags = 0; 39671da177e4SLinus Torvalds 39681da177e4SLinus Torvalds from = getname(oldname); 39691da177e4SLinus Torvalds if (IS_ERR(from)) 39701da177e4SLinus Torvalds return PTR_ERR(from); 3971f46d3567SJeff Layton retry: 3972f46d3567SJeff Layton dentry = user_path_create(newdfd, newname, &path, lookup_flags); 39731da177e4SLinus Torvalds error = PTR_ERR(dentry); 39746902d925SDave Hansen if (IS_ERR(dentry)) 3975dae6ad8fSAl Viro goto out_putname; 39766902d925SDave Hansen 397791a27b2aSJeff Layton error = security_path_symlink(&path, dentry, from->name); 3978a8104a9fSAl Viro if (!error) 397991a27b2aSJeff Layton error = vfs_symlink(path.dentry->d_inode, dentry, from->name); 3980921a1650SAl Viro done_path_create(&path, dentry); 3981f46d3567SJeff Layton if (retry_estale(error, lookup_flags)) { 3982f46d3567SJeff Layton lookup_flags |= LOOKUP_REVAL; 3983f46d3567SJeff Layton goto retry; 3984f46d3567SJeff Layton } 39856902d925SDave Hansen out_putname: 39861da177e4SLinus Torvalds putname(from); 39871da177e4SLinus Torvalds return error; 39881da177e4SLinus Torvalds } 39891da177e4SLinus Torvalds 39903480b257SHeiko Carstens SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) 39915590ff0dSUlrich Drepper { 39925590ff0dSUlrich Drepper return sys_symlinkat(oldname, AT_FDCWD, newname); 39935590ff0dSUlrich Drepper } 39945590ff0dSUlrich Drepper 3995146a8595SJ. Bruce Fields /** 3996146a8595SJ. Bruce Fields * vfs_link - create a new link 3997146a8595SJ. Bruce Fields * @old_dentry: object to be linked 3998146a8595SJ. Bruce Fields * @dir: new parent 3999146a8595SJ. Bruce Fields * @new_dentry: where to create the new link 4000146a8595SJ. Bruce Fields * @delegated_inode: returns inode needing a delegation break 4001146a8595SJ. Bruce Fields * 4002146a8595SJ. Bruce Fields * The caller must hold dir->i_mutex 4003146a8595SJ. Bruce Fields * 4004146a8595SJ. Bruce Fields * If vfs_link discovers a delegation on the to-be-linked file in need 4005146a8595SJ. Bruce Fields * of breaking, it will return -EWOULDBLOCK and return a reference to the 4006146a8595SJ. Bruce Fields * inode in delegated_inode. The caller should then break the delegation 4007146a8595SJ. Bruce Fields * and retry. Because breaking a delegation may take a long time, the 4008146a8595SJ. Bruce Fields * caller should drop the i_mutex before doing so. 4009146a8595SJ. Bruce Fields * 4010146a8595SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 4011146a8595SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 4012146a8595SJ. Bruce Fields * to be NFS exported. 4013146a8595SJ. Bruce Fields */ 4014146a8595SJ. Bruce Fields int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode) 40151da177e4SLinus Torvalds { 40161da177e4SLinus Torvalds struct inode *inode = old_dentry->d_inode; 40178de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 40181da177e4SLinus Torvalds int error; 40191da177e4SLinus Torvalds 40201da177e4SLinus Torvalds if (!inode) 40211da177e4SLinus Torvalds return -ENOENT; 40221da177e4SLinus Torvalds 4023a95164d9SMiklos Szeredi error = may_create(dir, new_dentry); 40241da177e4SLinus Torvalds if (error) 40251da177e4SLinus Torvalds return error; 40261da177e4SLinus Torvalds 40271da177e4SLinus Torvalds if (dir->i_sb != inode->i_sb) 40281da177e4SLinus Torvalds return -EXDEV; 40291da177e4SLinus Torvalds 40301da177e4SLinus Torvalds /* 40311da177e4SLinus Torvalds * A link to an append-only or immutable file cannot be created. 40321da177e4SLinus Torvalds */ 40331da177e4SLinus Torvalds if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 40341da177e4SLinus Torvalds return -EPERM; 4035acfa4380SAl Viro if (!dir->i_op->link) 40361da177e4SLinus Torvalds return -EPERM; 40377e79eedbSTetsuo Handa if (S_ISDIR(inode->i_mode)) 40381da177e4SLinus Torvalds return -EPERM; 40391da177e4SLinus Torvalds 40401da177e4SLinus Torvalds error = security_inode_link(old_dentry, dir, new_dentry); 40411da177e4SLinus Torvalds if (error) 40421da177e4SLinus Torvalds return error; 40431da177e4SLinus Torvalds 40447e79eedbSTetsuo Handa mutex_lock(&inode->i_mutex); 4045aae8a97dSAneesh Kumar K.V /* Make sure we don't allow creating hardlink to an unlinked file */ 4046f4e0c30cSAl Viro if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) 4047aae8a97dSAneesh Kumar K.V error = -ENOENT; 40488de52778SAl Viro else if (max_links && inode->i_nlink >= max_links) 40498de52778SAl Viro error = -EMLINK; 4050146a8595SJ. Bruce Fields else { 4051146a8595SJ. Bruce Fields error = try_break_deleg(inode, delegated_inode); 4052146a8595SJ. Bruce Fields if (!error) 40531da177e4SLinus Torvalds error = dir->i_op->link(old_dentry, dir, new_dentry); 4054146a8595SJ. Bruce Fields } 4055f4e0c30cSAl Viro 4056f4e0c30cSAl Viro if (!error && (inode->i_state & I_LINKABLE)) { 4057f4e0c30cSAl Viro spin_lock(&inode->i_lock); 4058f4e0c30cSAl Viro inode->i_state &= ~I_LINKABLE; 4059f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 4060f4e0c30cSAl Viro } 40617e79eedbSTetsuo Handa mutex_unlock(&inode->i_mutex); 4062e31e14ecSStephen Smalley if (!error) 40637e79eedbSTetsuo Handa fsnotify_link(dir, inode, new_dentry); 40641da177e4SLinus Torvalds return error; 40651da177e4SLinus Torvalds } 40664d359507SAl Viro EXPORT_SYMBOL(vfs_link); 40671da177e4SLinus Torvalds 40681da177e4SLinus Torvalds /* 40691da177e4SLinus Torvalds * Hardlinks are often used in delicate situations. We avoid 40701da177e4SLinus Torvalds * security-related surprises by not following symlinks on the 40711da177e4SLinus Torvalds * newname. --KAB 40721da177e4SLinus Torvalds * 40731da177e4SLinus Torvalds * We don't follow them on the oldname either to be compatible 40741da177e4SLinus Torvalds * with linux 2.0, and to avoid hard-linking to directories 40751da177e4SLinus Torvalds * and other special files. --ADM 40761da177e4SLinus Torvalds */ 40772e4d0924SHeiko Carstens SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, 40782e4d0924SHeiko Carstens int, newdfd, const char __user *, newname, int, flags) 40791da177e4SLinus Torvalds { 40801da177e4SLinus Torvalds struct dentry *new_dentry; 4081dae6ad8fSAl Viro struct path old_path, new_path; 4082146a8595SJ. Bruce Fields struct inode *delegated_inode = NULL; 408311a7b371SAneesh Kumar K.V int how = 0; 40841da177e4SLinus Torvalds int error; 40851da177e4SLinus Torvalds 408611a7b371SAneesh Kumar K.V if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) 4087c04030e1SUlrich Drepper return -EINVAL; 408811a7b371SAneesh Kumar K.V /* 4089f0cc6ffbSLinus Torvalds * To use null names we require CAP_DAC_READ_SEARCH 4090f0cc6ffbSLinus Torvalds * This ensures that not everyone will be able to create 4091f0cc6ffbSLinus Torvalds * handlink using the passed filedescriptor. 409211a7b371SAneesh Kumar K.V */ 4093f0cc6ffbSLinus Torvalds if (flags & AT_EMPTY_PATH) { 4094f0cc6ffbSLinus Torvalds if (!capable(CAP_DAC_READ_SEARCH)) 4095f0cc6ffbSLinus Torvalds return -ENOENT; 409611a7b371SAneesh Kumar K.V how = LOOKUP_EMPTY; 4097f0cc6ffbSLinus Torvalds } 4098c04030e1SUlrich Drepper 409911a7b371SAneesh Kumar K.V if (flags & AT_SYMLINK_FOLLOW) 410011a7b371SAneesh Kumar K.V how |= LOOKUP_FOLLOW; 4101442e31caSJeff Layton retry: 410211a7b371SAneesh Kumar K.V error = user_path_at(olddfd, oldname, how, &old_path); 41031da177e4SLinus Torvalds if (error) 41042ad94ae6SAl Viro return error; 41052ad94ae6SAl Viro 4106442e31caSJeff Layton new_dentry = user_path_create(newdfd, newname, &new_path, 4107442e31caSJeff Layton (how & LOOKUP_REVAL)); 41081da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 41096902d925SDave Hansen if (IS_ERR(new_dentry)) 4110dae6ad8fSAl Viro goto out; 4111dae6ad8fSAl Viro 4112dae6ad8fSAl Viro error = -EXDEV; 4113dae6ad8fSAl Viro if (old_path.mnt != new_path.mnt) 4114dae6ad8fSAl Viro goto out_dput; 4115800179c9SKees Cook error = may_linkat(&old_path); 4116800179c9SKees Cook if (unlikely(error)) 4117800179c9SKees Cook goto out_dput; 4118dae6ad8fSAl Viro error = security_path_link(old_path.dentry, &new_path, new_dentry); 4119be6d3e56SKentaro Takeda if (error) 4120a8104a9fSAl Viro goto out_dput; 4121146a8595SJ. Bruce Fields error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode); 412275c3f29dSDave Hansen out_dput: 4123921a1650SAl Viro done_path_create(&new_path, new_dentry); 4124146a8595SJ. Bruce Fields if (delegated_inode) { 4125146a8595SJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4126d22e6338SOleg Drokin if (!error) { 4127d22e6338SOleg Drokin path_put(&old_path); 4128146a8595SJ. Bruce Fields goto retry; 4129146a8595SJ. Bruce Fields } 4130d22e6338SOleg Drokin } 4131442e31caSJeff Layton if (retry_estale(error, how)) { 4132d22e6338SOleg Drokin path_put(&old_path); 4133442e31caSJeff Layton how |= LOOKUP_REVAL; 4134442e31caSJeff Layton goto retry; 4135442e31caSJeff Layton } 41361da177e4SLinus Torvalds out: 41372d8f3038SAl Viro path_put(&old_path); 41381da177e4SLinus Torvalds 41391da177e4SLinus Torvalds return error; 41401da177e4SLinus Torvalds } 41411da177e4SLinus Torvalds 41423480b257SHeiko Carstens SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) 41435590ff0dSUlrich Drepper { 4144c04030e1SUlrich Drepper return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 41455590ff0dSUlrich Drepper } 41465590ff0dSUlrich Drepper 4147bc27027aSMiklos Szeredi /** 4148bc27027aSMiklos Szeredi * vfs_rename - rename a filesystem object 4149bc27027aSMiklos Szeredi * @old_dir: parent of source 4150bc27027aSMiklos Szeredi * @old_dentry: source 4151bc27027aSMiklos Szeredi * @new_dir: parent of destination 4152bc27027aSMiklos Szeredi * @new_dentry: destination 4153bc27027aSMiklos Szeredi * @delegated_inode: returns an inode needing a delegation break 4154520c8b16SMiklos Szeredi * @flags: rename flags 4155bc27027aSMiklos Szeredi * 4156bc27027aSMiklos Szeredi * The caller must hold multiple mutexes--see lock_rename()). 4157bc27027aSMiklos Szeredi * 4158bc27027aSMiklos Szeredi * If vfs_rename discovers a delegation in need of breaking at either 4159bc27027aSMiklos Szeredi * the source or destination, it will return -EWOULDBLOCK and return a 4160bc27027aSMiklos Szeredi * reference to the inode in delegated_inode. The caller should then 4161bc27027aSMiklos Szeredi * break the delegation and retry. Because breaking a delegation may 4162bc27027aSMiklos Szeredi * take a long time, the caller should drop all locks before doing 4163bc27027aSMiklos Szeredi * so. 4164bc27027aSMiklos Szeredi * 4165bc27027aSMiklos Szeredi * Alternatively, a caller may pass NULL for delegated_inode. This may 4166bc27027aSMiklos Szeredi * be appropriate for callers that expect the underlying filesystem not 4167bc27027aSMiklos Szeredi * to be NFS exported. 4168bc27027aSMiklos Szeredi * 41691da177e4SLinus Torvalds * The worst of all namespace operations - renaming directory. "Perverted" 41701da177e4SLinus Torvalds * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 41711da177e4SLinus Torvalds * Problems: 4172d03b29a2SJ. Bruce Fields * a) we can get into loop creation. 41731da177e4SLinus Torvalds * b) race potential - two innocent renames can create a loop together. 41741da177e4SLinus Torvalds * That's where 4.4 screws up. Current fix: serialization on 4175a11f3a05SArjan van de Ven * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 41761da177e4SLinus Torvalds * story. 41776cedba89SJ. Bruce Fields * c) we have to lock _four_ objects - parents and victim (if it exists), 41786cedba89SJ. Bruce Fields * and source (if it is not a directory). 41791b1dcc1bSJes Sorensen * And that - after we got ->i_mutex on parents (until then we don't know 41801da177e4SLinus Torvalds * whether the target exists). Solution: try to be smart with locking 41811da177e4SLinus Torvalds * order for inodes. We rely on the fact that tree topology may change 4182a11f3a05SArjan van de Ven * only under ->s_vfs_rename_mutex _and_ that parent of the object we 41831da177e4SLinus Torvalds * move will be locked. Thus we can rank directories by the tree 41841da177e4SLinus Torvalds * (ancestors first) and rank all non-directories after them. 41851da177e4SLinus Torvalds * That works since everybody except rename does "lock parent, lookup, 4186a11f3a05SArjan van de Ven * lock child" and rename is under ->s_vfs_rename_mutex. 41871da177e4SLinus Torvalds * HOWEVER, it relies on the assumption that any object with ->lookup() 41881da177e4SLinus Torvalds * has no more than 1 dentry. If "hybrid" objects will ever appear, 41891da177e4SLinus Torvalds * we'd better make sure that there's no link(2) for them. 4190e4eaac06SSage Weil * d) conversion from fhandle to dentry may come in the wrong moment - when 41911b1dcc1bSJes Sorensen * we are removing the target. Solution: we will have to grab ->i_mutex 41921da177e4SLinus Torvalds * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 4193c41b20e7SAdam Buchbinder * ->i_mutex on parents, which works but leads to some truly excessive 41941da177e4SLinus Torvalds * locking]. 41951da177e4SLinus Torvalds */ 41961da177e4SLinus Torvalds int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, 41978e6d782cSJ. Bruce Fields struct inode *new_dir, struct dentry *new_dentry, 4198520c8b16SMiklos Szeredi struct inode **delegated_inode, unsigned int flags) 41991da177e4SLinus Torvalds { 42001da177e4SLinus Torvalds int error; 4201bc27027aSMiklos Szeredi bool is_dir = d_is_dir(old_dentry); 420259b0df21SEric Paris const unsigned char *old_name; 4203bc27027aSMiklos Szeredi struct inode *source = old_dentry->d_inode; 4204bc27027aSMiklos Szeredi struct inode *target = new_dentry->d_inode; 4205da1ce067SMiklos Szeredi bool new_is_dir = false; 4206da1ce067SMiklos Szeredi unsigned max_links = new_dir->i_sb->s_max_links; 42071da177e4SLinus Torvalds 4208bc27027aSMiklos Szeredi if (source == target) 42091da177e4SLinus Torvalds return 0; 42101da177e4SLinus Torvalds 42111da177e4SLinus Torvalds error = may_delete(old_dir, old_dentry, is_dir); 42121da177e4SLinus Torvalds if (error) 42131da177e4SLinus Torvalds return error; 42141da177e4SLinus Torvalds 4215da1ce067SMiklos Szeredi if (!target) { 4216a95164d9SMiklos Szeredi error = may_create(new_dir, new_dentry); 4217da1ce067SMiklos Szeredi } else { 4218da1ce067SMiklos Szeredi new_is_dir = d_is_dir(new_dentry); 4219da1ce067SMiklos Szeredi 4220da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 42211da177e4SLinus Torvalds error = may_delete(new_dir, new_dentry, is_dir); 4222da1ce067SMiklos Szeredi else 4223da1ce067SMiklos Szeredi error = may_delete(new_dir, new_dentry, new_is_dir); 4224da1ce067SMiklos Szeredi } 42251da177e4SLinus Torvalds if (error) 42261da177e4SLinus Torvalds return error; 42271da177e4SLinus Torvalds 42287177a9c4SMiklos Szeredi if (!old_dir->i_op->rename && !old_dir->i_op->rename2) 42291da177e4SLinus Torvalds return -EPERM; 42301da177e4SLinus Torvalds 4231520c8b16SMiklos Szeredi if (flags && !old_dir->i_op->rename2) 4232520c8b16SMiklos Szeredi return -EINVAL; 4233520c8b16SMiklos Szeredi 4234bc27027aSMiklos Szeredi /* 4235bc27027aSMiklos Szeredi * If we are going to change the parent - check write permissions, 4236bc27027aSMiklos Szeredi * we'll need to flip '..'. 4237bc27027aSMiklos Szeredi */ 4238da1ce067SMiklos Szeredi if (new_dir != old_dir) { 4239da1ce067SMiklos Szeredi if (is_dir) { 4240bc27027aSMiklos Szeredi error = inode_permission(source, MAY_WRITE); 4241bc27027aSMiklos Szeredi if (error) 4242bc27027aSMiklos Szeredi return error; 4243bc27027aSMiklos Szeredi } 4244da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && new_is_dir) { 4245da1ce067SMiklos Szeredi error = inode_permission(target, MAY_WRITE); 4246da1ce067SMiklos Szeredi if (error) 4247da1ce067SMiklos Szeredi return error; 4248da1ce067SMiklos Szeredi } 4249da1ce067SMiklos Szeredi } 42500eeca283SRobert Love 42510b3974ebSMiklos Szeredi error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry, 42520b3974ebSMiklos Szeredi flags); 4253bc27027aSMiklos Szeredi if (error) 4254bc27027aSMiklos Szeredi return error; 4255bc27027aSMiklos Szeredi 4256bc27027aSMiklos Szeredi old_name = fsnotify_oldname_init(old_dentry->d_name.name); 4257bc27027aSMiklos Szeredi dget(new_dentry); 4258da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4259bc27027aSMiklos Szeredi lock_two_nondirectories(source, target); 4260bc27027aSMiklos Szeredi else if (target) 4261bc27027aSMiklos Szeredi mutex_lock(&target->i_mutex); 4262bc27027aSMiklos Szeredi 4263bc27027aSMiklos Szeredi error = -EBUSY; 42647af1364fSEric W. Biederman if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry)) 4265bc27027aSMiklos Szeredi goto out; 4266bc27027aSMiklos Szeredi 4267da1ce067SMiklos Szeredi if (max_links && new_dir != old_dir) { 4268bc27027aSMiklos Szeredi error = -EMLINK; 4269da1ce067SMiklos Szeredi if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links) 4270bc27027aSMiklos Szeredi goto out; 4271da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir && 4272da1ce067SMiklos Szeredi old_dir->i_nlink >= max_links) 4273da1ce067SMiklos Szeredi goto out; 4274da1ce067SMiklos Szeredi } 4275da1ce067SMiklos Szeredi if (is_dir && !(flags & RENAME_EXCHANGE) && target) 4276bc27027aSMiklos Szeredi shrink_dcache_parent(new_dentry); 4277da1ce067SMiklos Szeredi if (!is_dir) { 4278bc27027aSMiklos Szeredi error = try_break_deleg(source, delegated_inode); 4279bc27027aSMiklos Szeredi if (error) 4280bc27027aSMiklos Szeredi goto out; 4281da1ce067SMiklos Szeredi } 4282da1ce067SMiklos Szeredi if (target && !new_is_dir) { 4283bc27027aSMiklos Szeredi error = try_break_deleg(target, delegated_inode); 4284bc27027aSMiklos Szeredi if (error) 4285bc27027aSMiklos Szeredi goto out; 4286bc27027aSMiklos Szeredi } 42877177a9c4SMiklos Szeredi if (!old_dir->i_op->rename2) { 4288520c8b16SMiklos Szeredi error = old_dir->i_op->rename(old_dir, old_dentry, 4289520c8b16SMiklos Szeredi new_dir, new_dentry); 4290520c8b16SMiklos Szeredi } else { 42917177a9c4SMiklos Szeredi WARN_ON(old_dir->i_op->rename != NULL); 4292520c8b16SMiklos Szeredi error = old_dir->i_op->rename2(old_dir, old_dentry, 4293520c8b16SMiklos Szeredi new_dir, new_dentry, flags); 4294520c8b16SMiklos Szeredi } 4295bc27027aSMiklos Szeredi if (error) 4296bc27027aSMiklos Szeredi goto out; 4297bc27027aSMiklos Szeredi 4298da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE) && target) { 42991da177e4SLinus Torvalds if (is_dir) 4300bc27027aSMiklos Szeredi target->i_flags |= S_DEAD; 4301bc27027aSMiklos Szeredi dont_mount(new_dentry); 43028ed936b5SEric W. Biederman detach_mounts(new_dentry); 4303bc27027aSMiklos Szeredi } 4304da1ce067SMiklos Szeredi if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) { 4305da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 4306bc27027aSMiklos Szeredi d_move(old_dentry, new_dentry); 4307da1ce067SMiklos Szeredi else 4308da1ce067SMiklos Szeredi d_exchange(old_dentry, new_dentry); 4309da1ce067SMiklos Szeredi } 4310bc27027aSMiklos Szeredi out: 4311da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4312bc27027aSMiklos Szeredi unlock_two_nondirectories(source, target); 4313bc27027aSMiklos Szeredi else if (target) 4314bc27027aSMiklos Szeredi mutex_unlock(&target->i_mutex); 4315bc27027aSMiklos Szeredi dput(new_dentry); 4316da1ce067SMiklos Szeredi if (!error) { 4317123df294SAl Viro fsnotify_move(old_dir, new_dir, old_name, is_dir, 4318da1ce067SMiklos Szeredi !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry); 4319da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4320da1ce067SMiklos Szeredi fsnotify_move(new_dir, old_dir, old_dentry->d_name.name, 4321da1ce067SMiklos Szeredi new_is_dir, NULL, new_dentry); 4322da1ce067SMiklos Szeredi } 4323da1ce067SMiklos Szeredi } 43240eeca283SRobert Love fsnotify_oldname_free(old_name); 43250eeca283SRobert Love 43261da177e4SLinus Torvalds return error; 43271da177e4SLinus Torvalds } 43284d359507SAl Viro EXPORT_SYMBOL(vfs_rename); 43291da177e4SLinus Torvalds 4330520c8b16SMiklos Szeredi SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname, 4331520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname, unsigned int, flags) 43321da177e4SLinus Torvalds { 43331da177e4SLinus Torvalds struct dentry *old_dentry, *new_dentry; 43341da177e4SLinus Torvalds struct dentry *trap; 4335f5beed75SAl Viro struct path old_path, new_path; 4336f5beed75SAl Viro struct qstr old_last, new_last; 4337f5beed75SAl Viro int old_type, new_type; 43388e6d782cSJ. Bruce Fields struct inode *delegated_inode = NULL; 433991a27b2aSJeff Layton struct filename *from; 434091a27b2aSJeff Layton struct filename *to; 4341f5beed75SAl Viro unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET; 4342c6a94284SJeff Layton bool should_retry = false; 43432ad94ae6SAl Viro int error; 4344520c8b16SMiklos Szeredi 43450d7a8555SMiklos Szeredi if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) 4346da1ce067SMiklos Szeredi return -EINVAL; 4347da1ce067SMiklos Szeredi 43480d7a8555SMiklos Szeredi if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) && 43490d7a8555SMiklos Szeredi (flags & RENAME_EXCHANGE)) 4350520c8b16SMiklos Szeredi return -EINVAL; 4351520c8b16SMiklos Szeredi 43520d7a8555SMiklos Szeredi if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD)) 43530d7a8555SMiklos Szeredi return -EPERM; 43540d7a8555SMiklos Szeredi 4355f5beed75SAl Viro if (flags & RENAME_EXCHANGE) 4356f5beed75SAl Viro target_flags = 0; 4357f5beed75SAl Viro 4358c6a94284SJeff Layton retry: 4359f5beed75SAl Viro from = user_path_parent(olddfd, oldname, 4360f5beed75SAl Viro &old_path, &old_last, &old_type, lookup_flags); 436191a27b2aSJeff Layton if (IS_ERR(from)) { 436291a27b2aSJeff Layton error = PTR_ERR(from); 43631da177e4SLinus Torvalds goto exit; 436491a27b2aSJeff Layton } 43651da177e4SLinus Torvalds 4366f5beed75SAl Viro to = user_path_parent(newdfd, newname, 4367f5beed75SAl Viro &new_path, &new_last, &new_type, lookup_flags); 436891a27b2aSJeff Layton if (IS_ERR(to)) { 436991a27b2aSJeff Layton error = PTR_ERR(to); 43701da177e4SLinus Torvalds goto exit1; 437191a27b2aSJeff Layton } 43721da177e4SLinus Torvalds 43731da177e4SLinus Torvalds error = -EXDEV; 4374f5beed75SAl Viro if (old_path.mnt != new_path.mnt) 43751da177e4SLinus Torvalds goto exit2; 43761da177e4SLinus Torvalds 43771da177e4SLinus Torvalds error = -EBUSY; 4378f5beed75SAl Viro if (old_type != LAST_NORM) 43791da177e4SLinus Torvalds goto exit2; 43801da177e4SLinus Torvalds 43810a7c3937SMiklos Szeredi if (flags & RENAME_NOREPLACE) 43820a7c3937SMiklos Szeredi error = -EEXIST; 4383f5beed75SAl Viro if (new_type != LAST_NORM) 43841da177e4SLinus Torvalds goto exit2; 43851da177e4SLinus Torvalds 4386f5beed75SAl Viro error = mnt_want_write(old_path.mnt); 4387c30dabfeSJan Kara if (error) 4388c30dabfeSJan Kara goto exit2; 4389c30dabfeSJan Kara 43908e6d782cSJ. Bruce Fields retry_deleg: 4391f5beed75SAl Viro trap = lock_rename(new_path.dentry, old_path.dentry); 43921da177e4SLinus Torvalds 4393f5beed75SAl Viro old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags); 43941da177e4SLinus Torvalds error = PTR_ERR(old_dentry); 43951da177e4SLinus Torvalds if (IS_ERR(old_dentry)) 43961da177e4SLinus Torvalds goto exit3; 43971da177e4SLinus Torvalds /* source must exist */ 43981da177e4SLinus Torvalds error = -ENOENT; 4399b18825a7SDavid Howells if (d_is_negative(old_dentry)) 44001da177e4SLinus Torvalds goto exit4; 4401f5beed75SAl Viro new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags); 44021da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 44031da177e4SLinus Torvalds if (IS_ERR(new_dentry)) 44041da177e4SLinus Torvalds goto exit4; 44050a7c3937SMiklos Szeredi error = -EEXIST; 44060a7c3937SMiklos Szeredi if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry)) 44070a7c3937SMiklos Szeredi goto exit5; 4408da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4409da1ce067SMiklos Szeredi error = -ENOENT; 4410da1ce067SMiklos Szeredi if (d_is_negative(new_dentry)) 4411da1ce067SMiklos Szeredi goto exit5; 4412da1ce067SMiklos Szeredi 4413da1ce067SMiklos Szeredi if (!d_is_dir(new_dentry)) { 4414da1ce067SMiklos Szeredi error = -ENOTDIR; 4415f5beed75SAl Viro if (new_last.name[new_last.len]) 4416da1ce067SMiklos Szeredi goto exit5; 4417da1ce067SMiklos Szeredi } 4418da1ce067SMiklos Szeredi } 44190a7c3937SMiklos Szeredi /* unless the source is a directory trailing slashes give -ENOTDIR */ 44200a7c3937SMiklos Szeredi if (!d_is_dir(old_dentry)) { 44210a7c3937SMiklos Szeredi error = -ENOTDIR; 4422f5beed75SAl Viro if (old_last.name[old_last.len]) 44230a7c3937SMiklos Szeredi goto exit5; 4424f5beed75SAl Viro if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len]) 44250a7c3937SMiklos Szeredi goto exit5; 44260a7c3937SMiklos Szeredi } 44270a7c3937SMiklos Szeredi /* source should not be ancestor of target */ 44280a7c3937SMiklos Szeredi error = -EINVAL; 44290a7c3937SMiklos Szeredi if (old_dentry == trap) 44300a7c3937SMiklos Szeredi goto exit5; 44311da177e4SLinus Torvalds /* target should not be an ancestor of source */ 4432da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 44331da177e4SLinus Torvalds error = -ENOTEMPTY; 44341da177e4SLinus Torvalds if (new_dentry == trap) 44351da177e4SLinus Torvalds goto exit5; 44361da177e4SLinus Torvalds 4437f5beed75SAl Viro error = security_path_rename(&old_path, old_dentry, 4438f5beed75SAl Viro &new_path, new_dentry, flags); 4439be6d3e56SKentaro Takeda if (error) 4440c30dabfeSJan Kara goto exit5; 4441f5beed75SAl Viro error = vfs_rename(old_path.dentry->d_inode, old_dentry, 4442f5beed75SAl Viro new_path.dentry->d_inode, new_dentry, 4443520c8b16SMiklos Szeredi &delegated_inode, flags); 44441da177e4SLinus Torvalds exit5: 44451da177e4SLinus Torvalds dput(new_dentry); 44461da177e4SLinus Torvalds exit4: 44471da177e4SLinus Torvalds dput(old_dentry); 44481da177e4SLinus Torvalds exit3: 4449f5beed75SAl Viro unlock_rename(new_path.dentry, old_path.dentry); 44508e6d782cSJ. Bruce Fields if (delegated_inode) { 44518e6d782cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 44528e6d782cSJ. Bruce Fields if (!error) 44538e6d782cSJ. Bruce Fields goto retry_deleg; 44548e6d782cSJ. Bruce Fields } 4455f5beed75SAl Viro mnt_drop_write(old_path.mnt); 44561da177e4SLinus Torvalds exit2: 4457c6a94284SJeff Layton if (retry_estale(error, lookup_flags)) 4458c6a94284SJeff Layton should_retry = true; 4459f5beed75SAl Viro path_put(&new_path); 44602ad94ae6SAl Viro putname(to); 44611da177e4SLinus Torvalds exit1: 4462f5beed75SAl Viro path_put(&old_path); 44631da177e4SLinus Torvalds putname(from); 4464c6a94284SJeff Layton if (should_retry) { 4465c6a94284SJeff Layton should_retry = false; 4466c6a94284SJeff Layton lookup_flags |= LOOKUP_REVAL; 4467c6a94284SJeff Layton goto retry; 4468c6a94284SJeff Layton } 44692ad94ae6SAl Viro exit: 44701da177e4SLinus Torvalds return error; 44711da177e4SLinus Torvalds } 44721da177e4SLinus Torvalds 4473520c8b16SMiklos Szeredi SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, 4474520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname) 4475520c8b16SMiklos Szeredi { 4476520c8b16SMiklos Szeredi return sys_renameat2(olddfd, oldname, newdfd, newname, 0); 4477520c8b16SMiklos Szeredi } 4478520c8b16SMiklos Szeredi 4479a26eab24SHeiko Carstens SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) 44805590ff0dSUlrich Drepper { 4481520c8b16SMiklos Szeredi return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 44825590ff0dSUlrich Drepper } 44835590ff0dSUlrich Drepper 4484787fb6bcSMiklos Szeredi int vfs_whiteout(struct inode *dir, struct dentry *dentry) 4485787fb6bcSMiklos Szeredi { 4486787fb6bcSMiklos Szeredi int error = may_create(dir, dentry); 4487787fb6bcSMiklos Szeredi if (error) 4488787fb6bcSMiklos Szeredi return error; 4489787fb6bcSMiklos Szeredi 4490787fb6bcSMiklos Szeredi if (!dir->i_op->mknod) 4491787fb6bcSMiklos Szeredi return -EPERM; 4492787fb6bcSMiklos Szeredi 4493787fb6bcSMiklos Szeredi return dir->i_op->mknod(dir, dentry, 4494787fb6bcSMiklos Szeredi S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV); 4495787fb6bcSMiklos Szeredi } 4496787fb6bcSMiklos Szeredi EXPORT_SYMBOL(vfs_whiteout); 4497787fb6bcSMiklos Szeredi 44985d826c84SAl Viro int readlink_copy(char __user *buffer, int buflen, const char *link) 44991da177e4SLinus Torvalds { 45005d826c84SAl Viro int len = PTR_ERR(link); 45011da177e4SLinus Torvalds if (IS_ERR(link)) 45021da177e4SLinus Torvalds goto out; 45031da177e4SLinus Torvalds 45041da177e4SLinus Torvalds len = strlen(link); 45051da177e4SLinus Torvalds if (len > (unsigned) buflen) 45061da177e4SLinus Torvalds len = buflen; 45071da177e4SLinus Torvalds if (copy_to_user(buffer, link, len)) 45081da177e4SLinus Torvalds len = -EFAULT; 45091da177e4SLinus Torvalds out: 45101da177e4SLinus Torvalds return len; 45111da177e4SLinus Torvalds } 45125d826c84SAl Viro EXPORT_SYMBOL(readlink_copy); 45131da177e4SLinus Torvalds 45141da177e4SLinus Torvalds /* 45151da177e4SLinus Torvalds * A helper for ->readlink(). This should be used *ONLY* for symlinks that 45161da177e4SLinus Torvalds * have ->follow_link() touching nd only in nd_set_link(). Using (or not 45171da177e4SLinus Torvalds * using) it for any given inode is up to filesystem. 45181da177e4SLinus Torvalds */ 45191da177e4SLinus Torvalds int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen) 45201da177e4SLinus Torvalds { 4521cc314eefSLinus Torvalds void *cookie; 45225f2c4179SAl Viro struct inode *inode = d_inode(dentry); 45235f2c4179SAl Viro const char *link = inode->i_link; 4524694a1764SMarcin Slusarz int res; 4525cc314eefSLinus Torvalds 4526d4dee48bSAl Viro if (!link) { 45275f2c4179SAl Viro link = inode->i_op->follow_link(dentry, &cookie); 4528680baacbSAl Viro if (IS_ERR(link)) 4529680baacbSAl Viro return PTR_ERR(link); 4530d4dee48bSAl Viro } 4531680baacbSAl Viro res = readlink_copy(buffer, buflen, link); 45325f2c4179SAl Viro if (inode->i_op->put_link) 45335f2c4179SAl Viro inode->i_op->put_link(inode, cookie); 4534694a1764SMarcin Slusarz return res; 45351da177e4SLinus Torvalds } 45364d359507SAl Viro EXPORT_SYMBOL(generic_readlink); 45371da177e4SLinus Torvalds 45381da177e4SLinus Torvalds /* get the link contents into pagecache */ 45391da177e4SLinus Torvalds static char *page_getlink(struct dentry * dentry, struct page **ppage) 45401da177e4SLinus Torvalds { 4541ebd09abbSDuane Griffin char *kaddr; 45421da177e4SLinus Torvalds struct page *page; 45431da177e4SLinus Torvalds struct address_space *mapping = dentry->d_inode->i_mapping; 4544090d2b18SPekka Enberg page = read_mapping_page(mapping, 0, NULL); 45451da177e4SLinus Torvalds if (IS_ERR(page)) 45466fe6900eSNick Piggin return (char*)page; 45471da177e4SLinus Torvalds *ppage = page; 4548ebd09abbSDuane Griffin kaddr = kmap(page); 4549ebd09abbSDuane Griffin nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1); 4550ebd09abbSDuane Griffin return kaddr; 45511da177e4SLinus Torvalds } 45521da177e4SLinus Torvalds 45531da177e4SLinus Torvalds int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 45541da177e4SLinus Torvalds { 45551da177e4SLinus Torvalds struct page *page = NULL; 45565d826c84SAl Viro int res = readlink_copy(buffer, buflen, page_getlink(dentry, &page)); 45571da177e4SLinus Torvalds if (page) { 45581da177e4SLinus Torvalds kunmap(page); 45591da177e4SLinus Torvalds page_cache_release(page); 45601da177e4SLinus Torvalds } 45611da177e4SLinus Torvalds return res; 45621da177e4SLinus Torvalds } 45634d359507SAl Viro EXPORT_SYMBOL(page_readlink); 45641da177e4SLinus Torvalds 45656e77137bSAl Viro const char *page_follow_link_light(struct dentry *dentry, void **cookie) 45661da177e4SLinus Torvalds { 4567cc314eefSLinus Torvalds struct page *page = NULL; 4568680baacbSAl Viro char *res = page_getlink(dentry, &page); 4569680baacbSAl Viro if (!IS_ERR(res)) 4570680baacbSAl Viro *cookie = page; 4571680baacbSAl Viro return res; 45721da177e4SLinus Torvalds } 45734d359507SAl Viro EXPORT_SYMBOL(page_follow_link_light); 45741da177e4SLinus Torvalds 45755f2c4179SAl Viro void page_put_link(struct inode *unused, void *cookie) 45761da177e4SLinus Torvalds { 4577cc314eefSLinus Torvalds struct page *page = cookie; 45781da177e4SLinus Torvalds kunmap(page); 45791da177e4SLinus Torvalds page_cache_release(page); 45801da177e4SLinus Torvalds } 45814d359507SAl Viro EXPORT_SYMBOL(page_put_link); 45821da177e4SLinus Torvalds 458354566b2cSNick Piggin /* 458454566b2cSNick Piggin * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS 458554566b2cSNick Piggin */ 458654566b2cSNick Piggin int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) 45871da177e4SLinus Torvalds { 45881da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 45890adb25d2SKirill Korotaev struct page *page; 4590afddba49SNick Piggin void *fsdata; 4591beb497abSDmitriy Monakhov int err; 45921da177e4SLinus Torvalds char *kaddr; 459354566b2cSNick Piggin unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE; 459454566b2cSNick Piggin if (nofs) 459554566b2cSNick Piggin flags |= AOP_FLAG_NOFS; 45961da177e4SLinus Torvalds 45977e53cac4SNeilBrown retry: 4598afddba49SNick Piggin err = pagecache_write_begin(NULL, mapping, 0, len-1, 459954566b2cSNick Piggin flags, &page, &fsdata); 46001da177e4SLinus Torvalds if (err) 4601afddba49SNick Piggin goto fail; 4602afddba49SNick Piggin 4603e8e3c3d6SCong Wang kaddr = kmap_atomic(page); 46041da177e4SLinus Torvalds memcpy(kaddr, symname, len-1); 4605e8e3c3d6SCong Wang kunmap_atomic(kaddr); 4606afddba49SNick Piggin 4607afddba49SNick Piggin err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, 4608afddba49SNick Piggin page, fsdata); 46091da177e4SLinus Torvalds if (err < 0) 46101da177e4SLinus Torvalds goto fail; 4611afddba49SNick Piggin if (err < len-1) 4612afddba49SNick Piggin goto retry; 4613afddba49SNick Piggin 46141da177e4SLinus Torvalds mark_inode_dirty(inode); 46151da177e4SLinus Torvalds return 0; 46161da177e4SLinus Torvalds fail: 46171da177e4SLinus Torvalds return err; 46181da177e4SLinus Torvalds } 46194d359507SAl Viro EXPORT_SYMBOL(__page_symlink); 46201da177e4SLinus Torvalds 46210adb25d2SKirill Korotaev int page_symlink(struct inode *inode, const char *symname, int len) 46220adb25d2SKirill Korotaev { 46230adb25d2SKirill Korotaev return __page_symlink(inode, symname, len, 462454566b2cSNick Piggin !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS)); 46250adb25d2SKirill Korotaev } 46264d359507SAl Viro EXPORT_SYMBOL(page_symlink); 46270adb25d2SKirill Korotaev 462892e1d5beSArjan van de Ven const struct inode_operations page_symlink_inode_operations = { 46291da177e4SLinus Torvalds .readlink = generic_readlink, 46301da177e4SLinus Torvalds .follow_link = page_follow_link_light, 46311da177e4SLinus Torvalds .put_link = page_put_link, 46321da177e4SLinus Torvalds }; 46331da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink_inode_operations); 4634