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; 5029883d185SAl Viro unsigned seq, m_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]; 5139883d185SAl Viro struct filename *name; 5149883d185SAl Viro struct nameidata *saved; 5159883d185SAl Viro unsigned root_seq; 5169883d185SAl Viro int dfd; 5171f55a6ecSAl Viro }; 5181f55a6ecSAl Viro 5199883d185SAl Viro static void set_nameidata(struct nameidata *p, int dfd, struct filename *name) 520894bc8c4SAl Viro { 521756daf26SNeilBrown struct nameidata *old = current->nameidata; 522756daf26SNeilBrown p->stack = p->internal; 523c8a53ee5SAl Viro p->dfd = dfd; 524c8a53ee5SAl Viro p->name = name; 525756daf26SNeilBrown p->total_link_count = old ? old->total_link_count : 0; 5269883d185SAl Viro p->saved = old; 527756daf26SNeilBrown current->nameidata = p; 528894bc8c4SAl Viro } 529894bc8c4SAl Viro 5309883d185SAl Viro static void restore_nameidata(void) 531894bc8c4SAl Viro { 5329883d185SAl Viro struct nameidata *now = current->nameidata, *old = now->saved; 533756daf26SNeilBrown 534756daf26SNeilBrown current->nameidata = old; 535756daf26SNeilBrown if (old) 536756daf26SNeilBrown old->total_link_count = now->total_link_count; 537756daf26SNeilBrown if (now->stack != now->internal) { 538756daf26SNeilBrown kfree(now->stack); 539756daf26SNeilBrown now->stack = now->internal; 540894bc8c4SAl Viro } 541894bc8c4SAl Viro } 542894bc8c4SAl Viro 543894bc8c4SAl Viro static int __nd_alloc_stack(struct nameidata *nd) 544894bc8c4SAl Viro { 545bc40aee0SAl Viro struct saved *p; 546bc40aee0SAl Viro 547bc40aee0SAl Viro if (nd->flags & LOOKUP_RCU) { 548bc40aee0SAl Viro p= kmalloc(MAXSYMLINKS * sizeof(struct saved), 549bc40aee0SAl Viro GFP_ATOMIC); 550bc40aee0SAl Viro if (unlikely(!p)) 551bc40aee0SAl Viro return -ECHILD; 552bc40aee0SAl Viro } else { 553bc40aee0SAl Viro p= kmalloc(MAXSYMLINKS * sizeof(struct saved), 554894bc8c4SAl Viro GFP_KERNEL); 555894bc8c4SAl Viro if (unlikely(!p)) 556894bc8c4SAl Viro return -ENOMEM; 557bc40aee0SAl Viro } 558894bc8c4SAl Viro memcpy(p, nd->internal, sizeof(nd->internal)); 559894bc8c4SAl Viro nd->stack = p; 560894bc8c4SAl Viro return 0; 561894bc8c4SAl Viro } 562894bc8c4SAl Viro 563397d425dSEric W. Biederman /** 564397d425dSEric W. Biederman * path_connected - Verify that a path->dentry is below path->mnt.mnt_root 565397d425dSEric W. Biederman * @path: nameidate to verify 566397d425dSEric W. Biederman * 567397d425dSEric W. Biederman * Rename can sometimes move a file or directory outside of a bind 568397d425dSEric W. Biederman * mount, path_connected allows those cases to be detected. 569397d425dSEric W. Biederman */ 570397d425dSEric W. Biederman static bool path_connected(const struct path *path) 571397d425dSEric W. Biederman { 572397d425dSEric W. Biederman struct vfsmount *mnt = path->mnt; 573397d425dSEric W. Biederman 574397d425dSEric W. Biederman /* Only bind mounts can have disconnected paths */ 575397d425dSEric W. Biederman if (mnt->mnt_root == mnt->mnt_sb->s_root) 576397d425dSEric W. Biederman return true; 577397d425dSEric W. Biederman 578397d425dSEric W. Biederman return is_subdir(path->dentry, mnt->mnt_root); 579397d425dSEric W. Biederman } 580397d425dSEric W. Biederman 581894bc8c4SAl Viro static inline int nd_alloc_stack(struct nameidata *nd) 582894bc8c4SAl Viro { 583da4e0be0SAl Viro if (likely(nd->depth != EMBEDDED_LEVELS)) 584894bc8c4SAl Viro return 0; 585894bc8c4SAl Viro if (likely(nd->stack != nd->internal)) 586894bc8c4SAl Viro return 0; 587894bc8c4SAl Viro return __nd_alloc_stack(nd); 588894bc8c4SAl Viro } 589894bc8c4SAl Viro 5907973387aSAl Viro static void drop_links(struct nameidata *nd) 5917973387aSAl Viro { 5927973387aSAl Viro int i = nd->depth; 5937973387aSAl Viro while (i--) { 5947973387aSAl Viro struct saved *last = nd->stack + i; 5957973387aSAl Viro struct inode *inode = last->inode; 5967973387aSAl Viro if (last->cookie && inode->i_op->put_link) { 5977973387aSAl Viro inode->i_op->put_link(inode, last->cookie); 5987973387aSAl Viro last->cookie = NULL; 5997973387aSAl Viro } 6007973387aSAl Viro } 6017973387aSAl Viro } 6027973387aSAl Viro 6037973387aSAl Viro static void terminate_walk(struct nameidata *nd) 6047973387aSAl Viro { 6057973387aSAl Viro drop_links(nd); 6067973387aSAl Viro if (!(nd->flags & LOOKUP_RCU)) { 6077973387aSAl Viro int i; 6087973387aSAl Viro path_put(&nd->path); 6097973387aSAl Viro for (i = 0; i < nd->depth; i++) 6107973387aSAl Viro path_put(&nd->stack[i].link); 611102b8af2SAl Viro if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 612102b8af2SAl Viro path_put(&nd->root); 613102b8af2SAl Viro nd->root.mnt = NULL; 614102b8af2SAl Viro } 6157973387aSAl Viro } else { 6167973387aSAl Viro nd->flags &= ~LOOKUP_RCU; 6177973387aSAl Viro if (!(nd->flags & LOOKUP_ROOT)) 6187973387aSAl Viro nd->root.mnt = NULL; 6197973387aSAl Viro rcu_read_unlock(); 6207973387aSAl Viro } 6217973387aSAl Viro nd->depth = 0; 6227973387aSAl Viro } 6237973387aSAl Viro 6247973387aSAl Viro /* path_put is needed afterwards regardless of success or failure */ 6257973387aSAl Viro static bool legitimize_path(struct nameidata *nd, 6267973387aSAl Viro struct path *path, unsigned seq) 6277973387aSAl Viro { 6287973387aSAl Viro int res = __legitimize_mnt(path->mnt, nd->m_seq); 6297973387aSAl Viro if (unlikely(res)) { 6307973387aSAl Viro if (res > 0) 6317973387aSAl Viro path->mnt = NULL; 6327973387aSAl Viro path->dentry = NULL; 6337973387aSAl Viro return false; 6347973387aSAl Viro } 6357973387aSAl Viro if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) { 6367973387aSAl Viro path->dentry = NULL; 6377973387aSAl Viro return false; 6387973387aSAl Viro } 6397973387aSAl Viro return !read_seqcount_retry(&path->dentry->d_seq, seq); 6407973387aSAl Viro } 6417973387aSAl Viro 6427973387aSAl Viro static bool legitimize_links(struct nameidata *nd) 6437973387aSAl Viro { 6447973387aSAl Viro int i; 6457973387aSAl Viro for (i = 0; i < nd->depth; i++) { 6467973387aSAl Viro struct saved *last = nd->stack + i; 6477973387aSAl Viro if (unlikely(!legitimize_path(nd, &last->link, last->seq))) { 6487973387aSAl Viro drop_links(nd); 6497973387aSAl Viro nd->depth = i + 1; 6507973387aSAl Viro return false; 6517973387aSAl Viro } 6527973387aSAl Viro } 6537973387aSAl Viro return true; 6547973387aSAl Viro } 6557973387aSAl Viro 65619660af7SAl Viro /* 65731e6b01fSNick Piggin * Path walking has 2 modes, rcu-walk and ref-walk (see 65819660af7SAl Viro * Documentation/filesystems/path-lookup.txt). In situations when we can't 65919660af7SAl Viro * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab 66057e3715cSMike Marshall * normal reference counts on dentries and vfsmounts to transition to ref-walk 66119660af7SAl Viro * mode. Refcounts are grabbed at the last known good point before rcu-walk 66219660af7SAl Viro * got stuck, so ref-walk may continue from there. If this is not successful 66319660af7SAl Viro * (eg. a seqcount has changed), then failure is returned and it's up to caller 66419660af7SAl Viro * to restart the path walk from the beginning in ref-walk mode. 66531e6b01fSNick Piggin */ 66631e6b01fSNick Piggin 66731e6b01fSNick Piggin /** 66819660af7SAl Viro * unlazy_walk - try to switch to ref-walk mode. 66919660af7SAl Viro * @nd: nameidata pathwalk data 67019660af7SAl Viro * @dentry: child of nd->path.dentry or NULL 6716e9918b7SAl Viro * @seq: seq number to check dentry against 67239191628SRandy Dunlap * Returns: 0 on success, -ECHILD on failure 67331e6b01fSNick Piggin * 67419660af7SAl Viro * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry 67519660af7SAl Viro * for ref-walk mode. @dentry must be a path found by a do_lookup call on 67619660af7SAl Viro * @nd or NULL. Must be called from rcu-walk context. 6777973387aSAl Viro * Nothing should touch nameidata between unlazy_walk() failure and 6787973387aSAl Viro * terminate_walk(). 67931e6b01fSNick Piggin */ 6806e9918b7SAl Viro static int unlazy_walk(struct nameidata *nd, struct dentry *dentry, unsigned seq) 68131e6b01fSNick Piggin { 68231e6b01fSNick Piggin struct dentry *parent = nd->path.dentry; 68331e6b01fSNick Piggin 68431e6b01fSNick Piggin BUG_ON(!(nd->flags & LOOKUP_RCU)); 685e5c832d5SLinus Torvalds 686e5c832d5SLinus Torvalds nd->flags &= ~LOOKUP_RCU; 6877973387aSAl Viro if (unlikely(!legitimize_links(nd))) 6887973387aSAl Viro goto out2; 6897973387aSAl Viro if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq))) 6907973387aSAl Viro goto out2; 6917973387aSAl Viro if (unlikely(!lockref_get_not_dead(&parent->d_lockref))) 6927973387aSAl Viro goto out1; 69348a066e7SAl Viro 69415570086SLinus Torvalds /* 69515570086SLinus Torvalds * For a negative lookup, the lookup sequence point is the parents 69615570086SLinus Torvalds * sequence point, and it only needs to revalidate the parent dentry. 69715570086SLinus Torvalds * 69815570086SLinus Torvalds * For a positive lookup, we need to move both the parent and the 69915570086SLinus Torvalds * dentry from the RCU domain to be properly refcounted. And the 70015570086SLinus Torvalds * sequence number in the dentry validates *both* dentry counters, 70115570086SLinus Torvalds * since we checked the sequence number of the parent after we got 70215570086SLinus Torvalds * the child sequence number. So we know the parent must still 70315570086SLinus Torvalds * be valid if the child sequence number is still valid. 70415570086SLinus Torvalds */ 70519660af7SAl Viro if (!dentry) { 706e5c832d5SLinus Torvalds if (read_seqcount_retry(&parent->d_seq, nd->seq)) 707e5c832d5SLinus Torvalds goto out; 70819660af7SAl Viro BUG_ON(nd->inode != parent->d_inode); 70919660af7SAl Viro } else { 710e5c832d5SLinus Torvalds if (!lockref_get_not_dead(&dentry->d_lockref)) 711e5c832d5SLinus Torvalds goto out; 7126e9918b7SAl Viro if (read_seqcount_retry(&dentry->d_seq, seq)) 713e5c832d5SLinus Torvalds goto drop_dentry; 71419660af7SAl Viro } 715e5c832d5SLinus Torvalds 716e5c832d5SLinus Torvalds /* 717e5c832d5SLinus Torvalds * Sequence counts matched. Now make sure that the root is 718e5c832d5SLinus Torvalds * still valid and get it if required. 719e5c832d5SLinus Torvalds */ 720e5c832d5SLinus Torvalds if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 7215a8d87e8SAl Viro if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) { 7225a8d87e8SAl Viro rcu_read_unlock(); 7235a8d87e8SAl Viro dput(dentry); 7245a8d87e8SAl Viro return -ECHILD; 7257973387aSAl Viro } 72631e6b01fSNick Piggin } 72731e6b01fSNick Piggin 7288b61e74fSAl Viro rcu_read_unlock(); 72931e6b01fSNick Piggin return 0; 73019660af7SAl Viro 731e5c832d5SLinus Torvalds drop_dentry: 7328b61e74fSAl Viro rcu_read_unlock(); 733e5c832d5SLinus Torvalds dput(dentry); 734d0d27277SLinus Torvalds goto drop_root_mnt; 7357973387aSAl Viro out2: 7367973387aSAl Viro nd->path.mnt = NULL; 7377973387aSAl Viro out1: 7387973387aSAl Viro nd->path.dentry = NULL; 739e5c832d5SLinus Torvalds out: 7408b61e74fSAl Viro rcu_read_unlock(); 741d0d27277SLinus Torvalds drop_root_mnt: 742d0d27277SLinus Torvalds if (!(nd->flags & LOOKUP_ROOT)) 743d0d27277SLinus Torvalds nd->root.mnt = NULL; 74431e6b01fSNick Piggin return -ECHILD; 74531e6b01fSNick Piggin } 74631e6b01fSNick Piggin 7477973387aSAl Viro static int unlazy_link(struct nameidata *nd, struct path *link, unsigned seq) 7487973387aSAl Viro { 7497973387aSAl Viro if (unlikely(!legitimize_path(nd, link, seq))) { 7507973387aSAl Viro drop_links(nd); 7517973387aSAl Viro nd->depth = 0; 7527973387aSAl Viro nd->flags &= ~LOOKUP_RCU; 7537973387aSAl Viro nd->path.mnt = NULL; 7547973387aSAl Viro nd->path.dentry = NULL; 7557973387aSAl Viro if (!(nd->flags & LOOKUP_ROOT)) 7567973387aSAl Viro nd->root.mnt = NULL; 7577973387aSAl Viro rcu_read_unlock(); 7587973387aSAl Viro } else if (likely(unlazy_walk(nd, NULL, 0)) == 0) { 7597973387aSAl Viro return 0; 7607973387aSAl Viro } 7617973387aSAl Viro path_put(link); 7627973387aSAl Viro return -ECHILD; 7637973387aSAl Viro } 7647973387aSAl Viro 7654ce16ef3SAl Viro static inline int d_revalidate(struct dentry *dentry, unsigned int flags) 76634286d66SNick Piggin { 7674ce16ef3SAl Viro return dentry->d_op->d_revalidate(dentry, flags); 76834286d66SNick Piggin } 76934286d66SNick Piggin 7709f1fafeeSAl Viro /** 7719f1fafeeSAl Viro * complete_walk - successful completion of path walk 7729f1fafeeSAl Viro * @nd: pointer nameidata 77339159de2SJeff Layton * 7749f1fafeeSAl Viro * If we had been in RCU mode, drop out of it and legitimize nd->path. 7759f1fafeeSAl Viro * Revalidate the final result, unless we'd already done that during 7769f1fafeeSAl Viro * the path walk or the filesystem doesn't ask for it. Return 0 on 7779f1fafeeSAl Viro * success, -error on failure. In case of failure caller does not 7789f1fafeeSAl Viro * need to drop nd->path. 77939159de2SJeff Layton */ 7809f1fafeeSAl Viro static int complete_walk(struct nameidata *nd) 78139159de2SJeff Layton { 78216c2cd71SAl Viro struct dentry *dentry = nd->path.dentry; 78339159de2SJeff Layton int status; 78439159de2SJeff Layton 7859f1fafeeSAl Viro if (nd->flags & LOOKUP_RCU) { 7869f1fafeeSAl Viro if (!(nd->flags & LOOKUP_ROOT)) 7879f1fafeeSAl Viro nd->root.mnt = NULL; 7886e9918b7SAl Viro if (unlikely(unlazy_walk(nd, NULL, 0))) 78948a066e7SAl Viro return -ECHILD; 79048a066e7SAl Viro } 7919f1fafeeSAl Viro 79216c2cd71SAl Viro if (likely(!(nd->flags & LOOKUP_JUMPED))) 79339159de2SJeff Layton return 0; 79439159de2SJeff Layton 795ecf3d1f1SJeff Layton if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) 79616c2cd71SAl Viro return 0; 79716c2cd71SAl Viro 798ecf3d1f1SJeff Layton status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); 79939159de2SJeff Layton if (status > 0) 80039159de2SJeff Layton return 0; 80139159de2SJeff Layton 80216c2cd71SAl Viro if (!status) 80339159de2SJeff Layton status = -ESTALE; 80416c2cd71SAl Viro 80539159de2SJeff Layton return status; 80639159de2SJeff Layton } 80739159de2SJeff Layton 80818d8c860SAl Viro static void set_root(struct nameidata *nd) 8092a737871SAl Viro { 81031e6b01fSNick Piggin struct fs_struct *fs = current->fs; 8119e6697e2SAl Viro 8129e6697e2SAl Viro if (nd->flags & LOOKUP_RCU) { 8138f47a016SAl Viro unsigned seq; 814c28cc364SNick Piggin 815c28cc364SNick Piggin do { 816c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 81731e6b01fSNick Piggin nd->root = fs->root; 8188f47a016SAl Viro nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq); 819c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 8209e6697e2SAl Viro } else { 8219e6697e2SAl Viro get_fs_root(fs, &nd->root); 8229e6697e2SAl Viro } 82331e6b01fSNick Piggin } 82431e6b01fSNick Piggin 8251d957f9bSJan Blunck static void path_put_conditional(struct path *path, struct nameidata *nd) 826051d3812SIan Kent { 827051d3812SIan Kent dput(path->dentry); 8284ac91378SJan Blunck if (path->mnt != nd->path.mnt) 829051d3812SIan Kent mntput(path->mnt); 830051d3812SIan Kent } 831051d3812SIan Kent 8327b9337aaSNick Piggin static inline void path_to_nameidata(const struct path *path, 8337b9337aaSNick Piggin struct nameidata *nd) 834051d3812SIan Kent { 83531e6b01fSNick Piggin if (!(nd->flags & LOOKUP_RCU)) { 8364ac91378SJan Blunck dput(nd->path.dentry); 83731e6b01fSNick Piggin if (nd->path.mnt != path->mnt) 8384ac91378SJan Blunck mntput(nd->path.mnt); 8399a229683SHuang Shijie } 84031e6b01fSNick Piggin nd->path.mnt = path->mnt; 8414ac91378SJan Blunck nd->path.dentry = path->dentry; 842051d3812SIan Kent } 843051d3812SIan Kent 844b5fb63c1SChristoph Hellwig /* 845b5fb63c1SChristoph Hellwig * Helper to directly jump to a known parsed path from ->follow_link, 846b5fb63c1SChristoph Hellwig * caller must have taken a reference to path beforehand. 847b5fb63c1SChristoph Hellwig */ 8486e77137bSAl Viro void nd_jump_link(struct path *path) 849b5fb63c1SChristoph Hellwig { 8506e77137bSAl Viro struct nameidata *nd = current->nameidata; 851b5fb63c1SChristoph Hellwig path_put(&nd->path); 852b5fb63c1SChristoph Hellwig 853b5fb63c1SChristoph Hellwig nd->path = *path; 854b5fb63c1SChristoph Hellwig nd->inode = nd->path.dentry->d_inode; 855b5fb63c1SChristoph Hellwig nd->flags |= LOOKUP_JUMPED; 856b5fb63c1SChristoph Hellwig } 857b5fb63c1SChristoph Hellwig 858b9ff4429SAl Viro static inline void put_link(struct nameidata *nd) 859574197e0SAl Viro { 86021c3003dSAl Viro struct saved *last = nd->stack + --nd->depth; 861237d8b32SAl Viro struct inode *inode = last->inode; 862b9ff4429SAl Viro if (last->cookie && inode->i_op->put_link) 8635f2c4179SAl Viro inode->i_op->put_link(inode, last->cookie); 8646548fae2SAl Viro if (!(nd->flags & LOOKUP_RCU)) 865b9ff4429SAl Viro path_put(&last->link); 866574197e0SAl Viro } 867574197e0SAl Viro 868561ec64aSLinus Torvalds int sysctl_protected_symlinks __read_mostly = 0; 869561ec64aSLinus Torvalds int sysctl_protected_hardlinks __read_mostly = 0; 870800179c9SKees Cook 871800179c9SKees Cook /** 872800179c9SKees Cook * may_follow_link - Check symlink following for unsafe situations 87355852635SRandy Dunlap * @nd: nameidata pathwalk data 874800179c9SKees Cook * 875800179c9SKees Cook * In the case of the sysctl_protected_symlinks sysctl being enabled, 876800179c9SKees Cook * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is 877800179c9SKees Cook * in a sticky world-writable directory. This is to protect privileged 878800179c9SKees Cook * processes from failing races against path names that may change out 879800179c9SKees Cook * from under them by way of other users creating malicious symlinks. 880800179c9SKees Cook * It will permit symlinks to be followed only when outside a sticky 881800179c9SKees Cook * world-writable directory, or when the uid of the symlink and follower 882800179c9SKees Cook * match, or when the directory owner matches the symlink's owner. 883800179c9SKees Cook * 884800179c9SKees Cook * Returns 0 if following the symlink is allowed, -ve on error. 885800179c9SKees Cook */ 886fec2fa24SAl Viro static inline int may_follow_link(struct nameidata *nd) 887800179c9SKees Cook { 888800179c9SKees Cook const struct inode *inode; 889800179c9SKees Cook const struct inode *parent; 890800179c9SKees Cook 891800179c9SKees Cook if (!sysctl_protected_symlinks) 892800179c9SKees Cook return 0; 893800179c9SKees Cook 894800179c9SKees Cook /* Allowed if owner and follower match. */ 895237d8b32SAl Viro inode = nd->stack[0].inode; 89681abe27bSEric W. Biederman if (uid_eq(current_cred()->fsuid, inode->i_uid)) 897800179c9SKees Cook return 0; 898800179c9SKees Cook 899800179c9SKees Cook /* Allowed if parent directory not sticky and world-writable. */ 900aa65fa35SAl Viro parent = nd->inode; 901800179c9SKees Cook if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) 902800179c9SKees Cook return 0; 903800179c9SKees Cook 904800179c9SKees Cook /* Allowed if parent directory and link owner match. */ 90581abe27bSEric W. Biederman if (uid_eq(parent->i_uid, inode->i_uid)) 906800179c9SKees Cook return 0; 907800179c9SKees Cook 90831956502SAl Viro if (nd->flags & LOOKUP_RCU) 90931956502SAl Viro return -ECHILD; 91031956502SAl Viro 9111cf2665bSAl Viro audit_log_link_denied("follow_link", &nd->stack[0].link); 912800179c9SKees Cook return -EACCES; 913800179c9SKees Cook } 914800179c9SKees Cook 915800179c9SKees Cook /** 916800179c9SKees Cook * safe_hardlink_source - Check for safe hardlink conditions 917800179c9SKees Cook * @inode: the source inode to hardlink from 918800179c9SKees Cook * 919800179c9SKees Cook * Return false if at least one of the following conditions: 920800179c9SKees Cook * - inode is not a regular file 921800179c9SKees Cook * - inode is setuid 922800179c9SKees Cook * - inode is setgid and group-exec 923800179c9SKees Cook * - access failure for read and write 924800179c9SKees Cook * 925800179c9SKees Cook * Otherwise returns true. 926800179c9SKees Cook */ 927800179c9SKees Cook static bool safe_hardlink_source(struct inode *inode) 928800179c9SKees Cook { 929800179c9SKees Cook umode_t mode = inode->i_mode; 930800179c9SKees Cook 931800179c9SKees Cook /* Special files should not get pinned to the filesystem. */ 932800179c9SKees Cook if (!S_ISREG(mode)) 933800179c9SKees Cook return false; 934800179c9SKees Cook 935800179c9SKees Cook /* Setuid files should not get pinned to the filesystem. */ 936800179c9SKees Cook if (mode & S_ISUID) 937800179c9SKees Cook return false; 938800179c9SKees Cook 939800179c9SKees Cook /* Executable setgid files should not get pinned to the filesystem. */ 940800179c9SKees Cook if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) 941800179c9SKees Cook return false; 942800179c9SKees Cook 943800179c9SKees Cook /* Hardlinking to unreadable or unwritable sources is dangerous. */ 944800179c9SKees Cook if (inode_permission(inode, MAY_READ | MAY_WRITE)) 945800179c9SKees Cook return false; 946800179c9SKees Cook 947800179c9SKees Cook return true; 948800179c9SKees Cook } 949800179c9SKees Cook 950800179c9SKees Cook /** 951800179c9SKees Cook * may_linkat - Check permissions for creating a hardlink 952800179c9SKees Cook * @link: the source to hardlink from 953800179c9SKees Cook * 954800179c9SKees Cook * Block hardlink when all of: 955800179c9SKees Cook * - sysctl_protected_hardlinks enabled 956800179c9SKees Cook * - fsuid does not match inode 957800179c9SKees Cook * - hardlink source is unsafe (see safe_hardlink_source() above) 958f2ca3796SDirk Steinmetz * - not CAP_FOWNER in a namespace with the inode owner uid mapped 959800179c9SKees Cook * 960800179c9SKees Cook * Returns 0 if successful, -ve on error. 961800179c9SKees Cook */ 962800179c9SKees Cook static int may_linkat(struct path *link) 963800179c9SKees Cook { 964800179c9SKees Cook struct inode *inode; 965800179c9SKees Cook 966800179c9SKees Cook if (!sysctl_protected_hardlinks) 967800179c9SKees Cook return 0; 968800179c9SKees Cook 969800179c9SKees Cook inode = link->dentry->d_inode; 970800179c9SKees Cook 971800179c9SKees Cook /* Source inode owner (or CAP_FOWNER) can hardlink all they like, 972800179c9SKees Cook * otherwise, it must be a safe source. 973800179c9SKees Cook */ 974f2ca3796SDirk Steinmetz if (inode_owner_or_capable(inode) || safe_hardlink_source(inode)) 975800179c9SKees Cook return 0; 976800179c9SKees Cook 977a51d9eaaSKees Cook audit_log_link_denied("linkat", link); 978800179c9SKees Cook return -EPERM; 979800179c9SKees Cook } 980800179c9SKees Cook 9813b2e7f75SAl Viro static __always_inline 9823b2e7f75SAl Viro const char *get_link(struct nameidata *nd) 9831da177e4SLinus Torvalds { 984ab104923SAl Viro struct saved *last = nd->stack + nd->depth - 1; 9851cf2665bSAl Viro struct dentry *dentry = last->link.dentry; 986237d8b32SAl Viro struct inode *inode = last->inode; 9876d7b5aaeSAl Viro int error; 9880a959df5SAl Viro const char *res; 9891da177e4SLinus Torvalds 9908fa9dd24SNeilBrown if (!(nd->flags & LOOKUP_RCU)) { 9918fa9dd24SNeilBrown touch_atime(&last->link); 9928fa9dd24SNeilBrown cond_resched(); 9938fa9dd24SNeilBrown } else if (atime_needs_update(&last->link, inode)) { 9948fa9dd24SNeilBrown if (unlikely(unlazy_walk(nd, NULL, 0))) 9958fa9dd24SNeilBrown return ERR_PTR(-ECHILD); 9968fa9dd24SNeilBrown touch_atime(&last->link); 9978fa9dd24SNeilBrown } 9988fa9dd24SNeilBrown 999bda0be7aSNeilBrown error = security_inode_follow_link(dentry, inode, 1000bda0be7aSNeilBrown nd->flags & LOOKUP_RCU); 1001bda0be7aSNeilBrown if (unlikely(error)) 10026920a440SAl Viro return ERR_PTR(error); 100336f3b4f6SAl Viro 100486acdca1SAl Viro nd->last_type = LAST_BIND; 1005d4dee48bSAl Viro res = inode->i_link; 1006d4dee48bSAl Viro if (!res) { 10078c1b4566SAl Viro if (nd->flags & LOOKUP_RCU) { 10088c1b4566SAl Viro if (unlikely(unlazy_walk(nd, NULL, 0))) 10098c1b4566SAl Viro return ERR_PTR(-ECHILD); 10108c1b4566SAl Viro } 10116e77137bSAl Viro res = inode->i_op->follow_link(dentry, &last->cookie); 1012fab51e8aSAl Viro if (IS_ERR_OR_NULL(res)) { 10136920a440SAl Viro last->cookie = NULL; 1014fab51e8aSAl Viro return res; 10150a959df5SAl Viro } 1016fab51e8aSAl Viro } 1017fab51e8aSAl Viro if (*res == '/') { 10189e6697e2SAl Viro if (!nd->root.mnt) 10199e6697e2SAl Viro set_root(nd); 10208c1b4566SAl Viro if (nd->flags & LOOKUP_RCU) { 10218f47a016SAl Viro struct dentry *d; 10228f47a016SAl Viro nd->path = nd->root; 10238f47a016SAl Viro d = nd->path.dentry; 10248f47a016SAl Viro nd->inode = d->d_inode; 10258f47a016SAl Viro nd->seq = nd->root_seq; 10268f47a016SAl Viro if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq))) 10278c1b4566SAl Viro return ERR_PTR(-ECHILD); 10288f47a016SAl Viro } else { 1029fab51e8aSAl Viro path_put(&nd->path); 1030fab51e8aSAl Viro nd->path = nd->root; 1031fab51e8aSAl Viro path_get(&nd->root); 1032fab51e8aSAl Viro nd->inode = nd->path.dentry->d_inode; 10338f47a016SAl Viro } 1034fab51e8aSAl Viro nd->flags |= LOOKUP_JUMPED; 1035fab51e8aSAl Viro while (unlikely(*++res == '/')) 1036fab51e8aSAl Viro ; 1037fab51e8aSAl Viro } 1038fab51e8aSAl Viro if (!*res) 1039fab51e8aSAl Viro res = NULL; 10400a959df5SAl Viro return res; 10410a959df5SAl Viro } 10426d7b5aaeSAl Viro 1043f015f126SDavid Howells /* 1044f015f126SDavid Howells * follow_up - Find the mountpoint of path's vfsmount 1045f015f126SDavid Howells * 1046f015f126SDavid Howells * Given a path, find the mountpoint of its source file system. 1047f015f126SDavid Howells * Replace @path with the path of the mountpoint in the parent mount. 1048f015f126SDavid Howells * Up is towards /. 1049f015f126SDavid Howells * 1050f015f126SDavid Howells * Return 1 if we went up a level and 0 if we were already at the 1051f015f126SDavid Howells * root. 1052f015f126SDavid Howells */ 1053bab77ebfSAl Viro int follow_up(struct path *path) 10541da177e4SLinus Torvalds { 10550714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 10560714a533SAl Viro struct mount *parent; 10571da177e4SLinus Torvalds struct dentry *mountpoint; 105899b7db7bSNick Piggin 105948a066e7SAl Viro read_seqlock_excl(&mount_lock); 10600714a533SAl Viro parent = mnt->mnt_parent; 10613c0a6163SAl Viro if (parent == mnt) { 106248a066e7SAl Viro read_sequnlock_excl(&mount_lock); 10631da177e4SLinus Torvalds return 0; 10641da177e4SLinus Torvalds } 10650714a533SAl Viro mntget(&parent->mnt); 1066a73324daSAl Viro mountpoint = dget(mnt->mnt_mountpoint); 106748a066e7SAl Viro read_sequnlock_excl(&mount_lock); 1068bab77ebfSAl Viro dput(path->dentry); 1069bab77ebfSAl Viro path->dentry = mountpoint; 1070bab77ebfSAl Viro mntput(path->mnt); 10710714a533SAl Viro path->mnt = &parent->mnt; 10721da177e4SLinus Torvalds return 1; 10731da177e4SLinus Torvalds } 10744d359507SAl Viro EXPORT_SYMBOL(follow_up); 10751da177e4SLinus Torvalds 1076b5c84bf6SNick Piggin /* 10779875cf80SDavid Howells * Perform an automount 10789875cf80SDavid Howells * - return -EISDIR to tell follow_managed() to stop and return the path we 10799875cf80SDavid Howells * were called with. 10801da177e4SLinus Torvalds */ 1081756daf26SNeilBrown static int follow_automount(struct path *path, struct nameidata *nd, 10829875cf80SDavid Howells bool *need_mntput) 108331e6b01fSNick Piggin { 10849875cf80SDavid Howells struct vfsmount *mnt; 1085ea5b778aSDavid Howells int err; 10869875cf80SDavid Howells 10879875cf80SDavid Howells if (!path->dentry->d_op || !path->dentry->d_op->d_automount) 10889875cf80SDavid Howells return -EREMOTE; 10899875cf80SDavid Howells 10900ec26fd0SMiklos Szeredi /* We don't want to mount if someone's just doing a stat - 10910ec26fd0SMiklos Szeredi * unless they're stat'ing a directory and appended a '/' to 10920ec26fd0SMiklos Szeredi * the name. 10930ec26fd0SMiklos Szeredi * 10940ec26fd0SMiklos Szeredi * We do, however, want to mount if someone wants to open or 10955a30d8a2SDavid Howells * create a file of any type under the mountpoint, wants to 10960ec26fd0SMiklos Szeredi * traverse through the mountpoint or wants to open the 10970ec26fd0SMiklos Szeredi * mounted directory. Also, autofs may mark negative dentries 10980ec26fd0SMiklos Szeredi * as being automount points. These will need the attentions 10990ec26fd0SMiklos Szeredi * of the daemon to instantiate them before they can be used. 11005a30d8a2SDavid Howells */ 1101756daf26SNeilBrown if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | 1102d94c177bSLinus Torvalds LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && 11035a30d8a2SDavid Howells path->dentry->d_inode) 11049875cf80SDavid Howells return -EISDIR; 11050ec26fd0SMiklos Szeredi 1106756daf26SNeilBrown nd->total_link_count++; 1107756daf26SNeilBrown if (nd->total_link_count >= 40) 11089875cf80SDavid Howells return -ELOOP; 11099875cf80SDavid Howells 11109875cf80SDavid Howells mnt = path->dentry->d_op->d_automount(path); 11119875cf80SDavid Howells if (IS_ERR(mnt)) { 11129875cf80SDavid Howells /* 11139875cf80SDavid Howells * The filesystem is allowed to return -EISDIR here to indicate 11149875cf80SDavid Howells * it doesn't want to automount. For instance, autofs would do 11159875cf80SDavid Howells * this so that its userspace daemon can mount on this dentry. 11169875cf80SDavid Howells * 11179875cf80SDavid Howells * However, we can only permit this if it's a terminal point in 11189875cf80SDavid Howells * the path being looked up; if it wasn't then the remainder of 11199875cf80SDavid Howells * the path is inaccessible and we should say so. 11209875cf80SDavid Howells */ 1121756daf26SNeilBrown if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT)) 11229875cf80SDavid Howells return -EREMOTE; 11239875cf80SDavid Howells return PTR_ERR(mnt); 112431e6b01fSNick Piggin } 1125ea5b778aSDavid Howells 11269875cf80SDavid Howells if (!mnt) /* mount collision */ 11279875cf80SDavid Howells return 0; 11289875cf80SDavid Howells 11298aef1884SAl Viro if (!*need_mntput) { 11308aef1884SAl Viro /* lock_mount() may release path->mnt on error */ 11318aef1884SAl Viro mntget(path->mnt); 11328aef1884SAl Viro *need_mntput = true; 11338aef1884SAl Viro } 113419a167afSAl Viro err = finish_automount(mnt, path); 1135ea5b778aSDavid Howells 1136ea5b778aSDavid Howells switch (err) { 1137ea5b778aSDavid Howells case -EBUSY: 1138ea5b778aSDavid Howells /* Someone else made a mount here whilst we were busy */ 113919a167afSAl Viro return 0; 1140ea5b778aSDavid Howells case 0: 11418aef1884SAl Viro path_put(path); 11429875cf80SDavid Howells path->mnt = mnt; 11439875cf80SDavid Howells path->dentry = dget(mnt->mnt_root); 11449875cf80SDavid Howells return 0; 114519a167afSAl Viro default: 114619a167afSAl Viro return err; 11479875cf80SDavid Howells } 114819a167afSAl Viro 1149ea5b778aSDavid Howells } 11509875cf80SDavid Howells 11519875cf80SDavid Howells /* 11529875cf80SDavid Howells * Handle a dentry that is managed in some way. 1153cc53ce53SDavid Howells * - Flagged for transit management (autofs) 11549875cf80SDavid Howells * - Flagged as mountpoint 11559875cf80SDavid Howells * - Flagged as automount point 11569875cf80SDavid Howells * 11579875cf80SDavid Howells * This may only be called in refwalk mode. 11589875cf80SDavid Howells * 11599875cf80SDavid Howells * Serialization is taken care of in namespace.c 11609875cf80SDavid Howells */ 1161756daf26SNeilBrown static int follow_managed(struct path *path, struct nameidata *nd) 11629875cf80SDavid Howells { 11638aef1884SAl Viro struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */ 11649875cf80SDavid Howells unsigned managed; 11659875cf80SDavid Howells bool need_mntput = false; 11668aef1884SAl Viro int ret = 0; 11679875cf80SDavid Howells 11689875cf80SDavid Howells /* Given that we're not holding a lock here, we retain the value in a 11699875cf80SDavid Howells * local variable for each dentry as we look at it so that we don't see 11709875cf80SDavid Howells * the components of that value change under us */ 11719875cf80SDavid Howells while (managed = ACCESS_ONCE(path->dentry->d_flags), 11729875cf80SDavid Howells managed &= DCACHE_MANAGED_DENTRY, 11739875cf80SDavid Howells unlikely(managed != 0)) { 1174cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1175cc53ce53SDavid Howells * being held. */ 1176cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 1177cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1178cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 11791aed3e42SAl Viro ret = path->dentry->d_op->d_manage(path->dentry, false); 1180cc53ce53SDavid Howells if (ret < 0) 11818aef1884SAl Viro break; 1182cc53ce53SDavid Howells } 1183cc53ce53SDavid Howells 11849875cf80SDavid Howells /* Transit to a mounted filesystem. */ 11859875cf80SDavid Howells if (managed & DCACHE_MOUNTED) { 11869875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 11879875cf80SDavid Howells if (mounted) { 11889875cf80SDavid Howells dput(path->dentry); 11899875cf80SDavid Howells if (need_mntput) 1190463ffb2eSAl Viro mntput(path->mnt); 1191463ffb2eSAl Viro path->mnt = mounted; 1192463ffb2eSAl Viro path->dentry = dget(mounted->mnt_root); 11939875cf80SDavid Howells need_mntput = true; 11949875cf80SDavid Howells continue; 1195463ffb2eSAl Viro } 1196463ffb2eSAl Viro 11979875cf80SDavid Howells /* Something is mounted on this dentry in another 11989875cf80SDavid Howells * namespace and/or whatever was mounted there in this 119948a066e7SAl Viro * namespace got unmounted before lookup_mnt() could 120048a066e7SAl Viro * get it */ 12011da177e4SLinus Torvalds } 12029875cf80SDavid Howells 12039875cf80SDavid Howells /* Handle an automount point */ 12049875cf80SDavid Howells if (managed & DCACHE_NEED_AUTOMOUNT) { 1205756daf26SNeilBrown ret = follow_automount(path, nd, &need_mntput); 12069875cf80SDavid Howells if (ret < 0) 12078aef1884SAl Viro break; 12089875cf80SDavid Howells continue; 12099875cf80SDavid Howells } 12109875cf80SDavid Howells 12119875cf80SDavid Howells /* We didn't change the current path point */ 12129875cf80SDavid Howells break; 12139875cf80SDavid Howells } 12148aef1884SAl Viro 12158aef1884SAl Viro if (need_mntput && path->mnt == mnt) 12168aef1884SAl Viro mntput(path->mnt); 12178aef1884SAl Viro if (ret == -EISDIR) 12188aef1884SAl Viro ret = 0; 12198402752eSAl Viro if (need_mntput) 12208402752eSAl Viro nd->flags |= LOOKUP_JUMPED; 12218402752eSAl Viro if (unlikely(ret < 0)) 12228402752eSAl Viro path_put_conditional(path, nd); 12238402752eSAl Viro return ret; 12241da177e4SLinus Torvalds } 12251da177e4SLinus Torvalds 1226cc53ce53SDavid Howells int follow_down_one(struct path *path) 12271da177e4SLinus Torvalds { 12281da177e4SLinus Torvalds struct vfsmount *mounted; 12291da177e4SLinus Torvalds 12301c755af4SAl Viro mounted = lookup_mnt(path); 12311da177e4SLinus Torvalds if (mounted) { 12329393bd07SAl Viro dput(path->dentry); 12339393bd07SAl Viro mntput(path->mnt); 12349393bd07SAl Viro path->mnt = mounted; 12359393bd07SAl Viro path->dentry = dget(mounted->mnt_root); 12361da177e4SLinus Torvalds return 1; 12371da177e4SLinus Torvalds } 12381da177e4SLinus Torvalds return 0; 12391da177e4SLinus Torvalds } 12404d359507SAl Viro EXPORT_SYMBOL(follow_down_one); 12411da177e4SLinus Torvalds 1242b8faf035SNeilBrown static inline int managed_dentry_rcu(struct dentry *dentry) 124362a7375eSIan Kent { 1244b8faf035SNeilBrown return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ? 1245b8faf035SNeilBrown dentry->d_op->d_manage(dentry, true) : 0; 124662a7375eSIan Kent } 124762a7375eSIan Kent 12489875cf80SDavid Howells /* 1249287548e4SAl Viro * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if 1250287548e4SAl Viro * we meet a managed dentry that would need blocking. 12519875cf80SDavid Howells */ 12529875cf80SDavid Howells static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, 1253254cf582SAl Viro struct inode **inode, unsigned *seqp) 12549875cf80SDavid Howells { 125562a7375eSIan Kent for (;;) { 1256c7105365SAl Viro struct mount *mounted; 125762a7375eSIan Kent /* 125862a7375eSIan Kent * Don't forget we might have a non-mountpoint managed dentry 125962a7375eSIan Kent * that wants to block transit. 126062a7375eSIan Kent */ 1261b8faf035SNeilBrown switch (managed_dentry_rcu(path->dentry)) { 1262b8faf035SNeilBrown case -ECHILD: 1263b8faf035SNeilBrown default: 1264ab90911fSDavid Howells return false; 1265b8faf035SNeilBrown case -EISDIR: 1266b8faf035SNeilBrown return true; 1267b8faf035SNeilBrown case 0: 1268b8faf035SNeilBrown break; 1269b8faf035SNeilBrown } 127062a7375eSIan Kent 127162a7375eSIan Kent if (!d_mountpoint(path->dentry)) 1272b8faf035SNeilBrown return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); 127362a7375eSIan Kent 1274474279dcSAl Viro mounted = __lookup_mnt(path->mnt, path->dentry); 12759875cf80SDavid Howells if (!mounted) 12769875cf80SDavid Howells break; 1277c7105365SAl Viro path->mnt = &mounted->mnt; 1278c7105365SAl Viro path->dentry = mounted->mnt.mnt_root; 1279a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 1280254cf582SAl Viro *seqp = read_seqcount_begin(&path->dentry->d_seq); 128159430262SLinus Torvalds /* 128259430262SLinus Torvalds * Update the inode too. We don't need to re-check the 128359430262SLinus Torvalds * dentry sequence number here after this d_inode read, 128459430262SLinus Torvalds * because a mount-point is always pinned. 128559430262SLinus Torvalds */ 128659430262SLinus Torvalds *inode = path->dentry->d_inode; 12879875cf80SDavid Howells } 1288f5be3e29SAl Viro return !read_seqretry(&mount_lock, nd->m_seq) && 1289b8faf035SNeilBrown !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); 1290287548e4SAl Viro } 1291287548e4SAl Viro 129231e6b01fSNick Piggin static int follow_dotdot_rcu(struct nameidata *nd) 129331e6b01fSNick Piggin { 12944023bfc9SAl Viro struct inode *inode = nd->inode; 129531e6b01fSNick Piggin 129631e6b01fSNick Piggin while (1) { 1297aed434adSAl Viro if (path_equal(&nd->path, &nd->root)) 129831e6b01fSNick Piggin break; 129931e6b01fSNick Piggin if (nd->path.dentry != nd->path.mnt->mnt_root) { 130031e6b01fSNick Piggin struct dentry *old = nd->path.dentry; 130131e6b01fSNick Piggin struct dentry *parent = old->d_parent; 130231e6b01fSNick Piggin unsigned seq; 130331e6b01fSNick Piggin 13044023bfc9SAl Viro inode = parent->d_inode; 130531e6b01fSNick Piggin seq = read_seqcount_begin(&parent->d_seq); 1306aed434adSAl Viro if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq))) 1307aed434adSAl Viro return -ECHILD; 130831e6b01fSNick Piggin nd->path.dentry = parent; 130931e6b01fSNick Piggin nd->seq = seq; 1310397d425dSEric W. Biederman if (unlikely(!path_connected(&nd->path))) 1311397d425dSEric W. Biederman return -ENOENT; 131231e6b01fSNick Piggin break; 1313aed434adSAl Viro } else { 1314aed434adSAl Viro struct mount *mnt = real_mount(nd->path.mnt); 1315aed434adSAl Viro struct mount *mparent = mnt->mnt_parent; 1316aed434adSAl Viro struct dentry *mountpoint = mnt->mnt_mountpoint; 1317aed434adSAl Viro struct inode *inode2 = mountpoint->d_inode; 1318aed434adSAl Viro unsigned seq = read_seqcount_begin(&mountpoint->d_seq); 1319aed434adSAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1320aed434adSAl Viro return -ECHILD; 1321aed434adSAl Viro if (&mparent->mnt == nd->path.mnt) 132231e6b01fSNick Piggin break; 1323aed434adSAl Viro /* we know that mountpoint was pinned */ 1324aed434adSAl Viro nd->path.dentry = mountpoint; 1325aed434adSAl Viro nd->path.mnt = &mparent->mnt; 1326aed434adSAl Viro inode = inode2; 1327aed434adSAl Viro nd->seq = seq; 132831e6b01fSNick Piggin } 1329aed434adSAl Viro } 1330aed434adSAl Viro while (unlikely(d_mountpoint(nd->path.dentry))) { 1331b37199e6SAl Viro struct mount *mounted; 1332b37199e6SAl Viro mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry); 1333aed434adSAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1334aed434adSAl Viro return -ECHILD; 1335b37199e6SAl Viro if (!mounted) 1336b37199e6SAl Viro break; 1337b37199e6SAl Viro nd->path.mnt = &mounted->mnt; 1338b37199e6SAl Viro nd->path.dentry = mounted->mnt.mnt_root; 13394023bfc9SAl Viro inode = nd->path.dentry->d_inode; 1340b37199e6SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 1341b37199e6SAl Viro } 13424023bfc9SAl Viro nd->inode = inode; 134331e6b01fSNick Piggin return 0; 134431e6b01fSNick Piggin } 134531e6b01fSNick Piggin 13469875cf80SDavid Howells /* 1347cc53ce53SDavid Howells * Follow down to the covering mount currently visible to userspace. At each 1348cc53ce53SDavid Howells * point, the filesystem owning that dentry may be queried as to whether the 1349cc53ce53SDavid Howells * caller is permitted to proceed or not. 1350cc53ce53SDavid Howells */ 13517cc90cc3SAl Viro int follow_down(struct path *path) 1352cc53ce53SDavid Howells { 1353cc53ce53SDavid Howells unsigned managed; 1354cc53ce53SDavid Howells int ret; 1355cc53ce53SDavid Howells 1356cc53ce53SDavid Howells while (managed = ACCESS_ONCE(path->dentry->d_flags), 1357cc53ce53SDavid Howells unlikely(managed & DCACHE_MANAGED_DENTRY)) { 1358cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1359cc53ce53SDavid Howells * being held. 1360cc53ce53SDavid Howells * 1361cc53ce53SDavid Howells * We indicate to the filesystem if someone is trying to mount 1362cc53ce53SDavid Howells * something here. This gives autofs the chance to deny anyone 1363cc53ce53SDavid Howells * other than its daemon the right to mount on its 1364cc53ce53SDavid Howells * superstructure. 1365cc53ce53SDavid Howells * 1366cc53ce53SDavid Howells * The filesystem may sleep at this point. 1367cc53ce53SDavid Howells */ 1368cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 1369cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1370cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 1371ab90911fSDavid Howells ret = path->dentry->d_op->d_manage( 13721aed3e42SAl Viro path->dentry, false); 1373cc53ce53SDavid Howells if (ret < 0) 1374cc53ce53SDavid Howells return ret == -EISDIR ? 0 : ret; 1375cc53ce53SDavid Howells } 1376cc53ce53SDavid Howells 1377cc53ce53SDavid Howells /* Transit to a mounted filesystem. */ 1378cc53ce53SDavid Howells if (managed & DCACHE_MOUNTED) { 1379cc53ce53SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 1380cc53ce53SDavid Howells if (!mounted) 1381cc53ce53SDavid Howells break; 1382cc53ce53SDavid Howells dput(path->dentry); 1383cc53ce53SDavid Howells mntput(path->mnt); 1384cc53ce53SDavid Howells path->mnt = mounted; 1385cc53ce53SDavid Howells path->dentry = dget(mounted->mnt_root); 1386cc53ce53SDavid Howells continue; 1387cc53ce53SDavid Howells } 1388cc53ce53SDavid Howells 1389cc53ce53SDavid Howells /* Don't handle automount points here */ 1390cc53ce53SDavid Howells break; 1391cc53ce53SDavid Howells } 1392cc53ce53SDavid Howells return 0; 1393cc53ce53SDavid Howells } 13944d359507SAl Viro EXPORT_SYMBOL(follow_down); 1395cc53ce53SDavid Howells 1396cc53ce53SDavid Howells /* 13979875cf80SDavid Howells * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() 13989875cf80SDavid Howells */ 13999875cf80SDavid Howells static void follow_mount(struct path *path) 14009875cf80SDavid Howells { 14019875cf80SDavid Howells while (d_mountpoint(path->dentry)) { 14029875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 14039875cf80SDavid Howells if (!mounted) 14049875cf80SDavid Howells break; 14059875cf80SDavid Howells dput(path->dentry); 14069875cf80SDavid Howells mntput(path->mnt); 14079875cf80SDavid Howells path->mnt = mounted; 14089875cf80SDavid Howells path->dentry = dget(mounted->mnt_root); 14099875cf80SDavid Howells } 14109875cf80SDavid Howells } 14119875cf80SDavid Howells 1412397d425dSEric W. Biederman static int follow_dotdot(struct nameidata *nd) 14131da177e4SLinus Torvalds { 14141da177e4SLinus Torvalds while(1) { 14154ac91378SJan Blunck struct dentry *old = nd->path.dentry; 14161da177e4SLinus Torvalds 14172a737871SAl Viro if (nd->path.dentry == nd->root.dentry && 14182a737871SAl Viro nd->path.mnt == nd->root.mnt) { 14191da177e4SLinus Torvalds break; 14201da177e4SLinus Torvalds } 14214ac91378SJan Blunck if (nd->path.dentry != nd->path.mnt->mnt_root) { 14223088dd70SAl Viro /* rare case of legitimate dget_parent()... */ 14233088dd70SAl Viro nd->path.dentry = dget_parent(nd->path.dentry); 14241da177e4SLinus Torvalds dput(old); 1425397d425dSEric W. Biederman if (unlikely(!path_connected(&nd->path))) 1426397d425dSEric W. Biederman return -ENOENT; 14271da177e4SLinus Torvalds break; 14281da177e4SLinus Torvalds } 14293088dd70SAl Viro if (!follow_up(&nd->path)) 14301da177e4SLinus Torvalds break; 14311da177e4SLinus Torvalds } 143279ed0226SAl Viro follow_mount(&nd->path); 143331e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 1434397d425dSEric W. Biederman return 0; 14351da177e4SLinus Torvalds } 14361da177e4SLinus Torvalds 14371da177e4SLinus Torvalds /* 1438bad61189SMiklos Szeredi * This looks up the name in dcache, possibly revalidates the old dentry and 1439bad61189SMiklos Szeredi * allocates a new one if not found or not valid. In the need_lookup argument 1440bad61189SMiklos Szeredi * returns whether i_op->lookup is necessary. 1441bad61189SMiklos Szeredi * 1442bad61189SMiklos Szeredi * dir->d_inode->i_mutex must be held 1443baa03890SNick Piggin */ 1444bad61189SMiklos Szeredi static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir, 1445201f956eSAl Viro unsigned int flags, bool *need_lookup) 1446baa03890SNick Piggin { 1447baa03890SNick Piggin struct dentry *dentry; 1448bad61189SMiklos Szeredi int error; 1449baa03890SNick Piggin 1450bad61189SMiklos Szeredi *need_lookup = false; 1451bad61189SMiklos Szeredi dentry = d_lookup(dir, name); 1452bad61189SMiklos Szeredi if (dentry) { 145339e3c955SJeff Layton if (dentry->d_flags & DCACHE_OP_REVALIDATE) { 1454201f956eSAl Viro error = d_revalidate(dentry, flags); 1455bad61189SMiklos Szeredi if (unlikely(error <= 0)) { 1456bad61189SMiklos Szeredi if (error < 0) { 1457bad61189SMiklos Szeredi dput(dentry); 1458bad61189SMiklos Szeredi return ERR_PTR(error); 14595542aa2fSEric W. Biederman } else { 14605542aa2fSEric W. Biederman d_invalidate(dentry); 1461bad61189SMiklos Szeredi dput(dentry); 1462bad61189SMiklos Szeredi dentry = NULL; 1463bad61189SMiklos Szeredi } 1464bad61189SMiklos Szeredi } 1465bad61189SMiklos Szeredi } 1466bad61189SMiklos Szeredi } 1467baa03890SNick Piggin 1468bad61189SMiklos Szeredi if (!dentry) { 1469bad61189SMiklos Szeredi dentry = d_alloc(dir, name); 1470baa03890SNick Piggin if (unlikely(!dentry)) 1471baa03890SNick Piggin return ERR_PTR(-ENOMEM); 1472baa03890SNick Piggin 1473bad61189SMiklos Szeredi *need_lookup = true; 1474baa03890SNick Piggin } 1475baa03890SNick Piggin return dentry; 1476baa03890SNick Piggin } 1477baa03890SNick Piggin 1478baa03890SNick Piggin /* 147913a2c3beSJ. Bruce Fields * Call i_op->lookup on the dentry. The dentry must be negative and 148013a2c3beSJ. Bruce Fields * unhashed. 1481bad61189SMiklos Szeredi * 1482bad61189SMiklos Szeredi * dir->d_inode->i_mutex must be held 148344396f4bSJosef Bacik */ 1484bad61189SMiklos Szeredi static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry, 148572bd866aSAl Viro unsigned int flags) 148644396f4bSJosef Bacik { 148744396f4bSJosef Bacik struct dentry *old; 148844396f4bSJosef Bacik 148944396f4bSJosef Bacik /* Don't create child dentry for a dead directory. */ 1490bad61189SMiklos Szeredi if (unlikely(IS_DEADDIR(dir))) { 1491e188dc02SMiklos Szeredi dput(dentry); 149244396f4bSJosef Bacik return ERR_PTR(-ENOENT); 1493e188dc02SMiklos Szeredi } 149444396f4bSJosef Bacik 149572bd866aSAl Viro old = dir->i_op->lookup(dir, dentry, flags); 149644396f4bSJosef Bacik if (unlikely(old)) { 149744396f4bSJosef Bacik dput(dentry); 149844396f4bSJosef Bacik dentry = old; 149944396f4bSJosef Bacik } 150044396f4bSJosef Bacik return dentry; 150144396f4bSJosef Bacik } 150244396f4bSJosef Bacik 1503a3255546SAl Viro static struct dentry *__lookup_hash(struct qstr *name, 150472bd866aSAl Viro struct dentry *base, unsigned int flags) 1505a3255546SAl Viro { 1506bad61189SMiklos Szeredi bool need_lookup; 1507a3255546SAl Viro struct dentry *dentry; 1508a3255546SAl Viro 150972bd866aSAl Viro dentry = lookup_dcache(name, base, flags, &need_lookup); 1510bad61189SMiklos Szeredi if (!need_lookup) 1511a3255546SAl Viro return dentry; 1512bad61189SMiklos Szeredi 151372bd866aSAl Viro return lookup_real(base->d_inode, dentry, flags); 1514a3255546SAl Viro } 1515a3255546SAl Viro 151644396f4bSJosef Bacik /* 15171da177e4SLinus Torvalds * It's more convoluted than I'd like it to be, but... it's still fairly 15181da177e4SLinus Torvalds * small and for now I'd prefer to have fast path as straight as possible. 15191da177e4SLinus Torvalds * It _is_ time-critical. 15201da177e4SLinus Torvalds */ 1521e97cdc87SAl Viro static int lookup_fast(struct nameidata *nd, 1522254cf582SAl Viro struct path *path, struct inode **inode, 1523254cf582SAl Viro unsigned *seqp) 15241da177e4SLinus Torvalds { 15254ac91378SJan Blunck struct vfsmount *mnt = nd->path.mnt; 152631e6b01fSNick Piggin struct dentry *dentry, *parent = nd->path.dentry; 15275a18fff2SAl Viro int need_reval = 1; 15285a18fff2SAl Viro int status = 1; 15299875cf80SDavid Howells int err; 15309875cf80SDavid Howells 15313cac260aSAl Viro /* 1532b04f784eSNick Piggin * Rename seqlock is not required here because in the off chance 1533b04f784eSNick Piggin * of a false negative due to a concurrent rename, we're going to 1534b04f784eSNick Piggin * do the non-racy lookup, below. 1535b04f784eSNick Piggin */ 153631e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) { 153731e6b01fSNick Piggin unsigned seq; 1538766c4cbfSAl Viro bool negative; 1539da53be12SLinus Torvalds dentry = __d_lookup_rcu(parent, &nd->last, &seq); 15405a18fff2SAl Viro if (!dentry) 15415a18fff2SAl Viro goto unlazy; 15425a18fff2SAl Viro 154312f8ad4bSLinus Torvalds /* 154412f8ad4bSLinus Torvalds * This sequence count validates that the inode matches 154512f8ad4bSLinus Torvalds * the dentry name information from lookup. 154612f8ad4bSLinus Torvalds */ 154763afdfc7SDavid Howells *inode = d_backing_inode(dentry); 1548766c4cbfSAl Viro negative = d_is_negative(dentry); 154912f8ad4bSLinus Torvalds if (read_seqcount_retry(&dentry->d_seq, seq)) 155012f8ad4bSLinus Torvalds return -ECHILD; 155112f8ad4bSLinus Torvalds 155212f8ad4bSLinus Torvalds /* 155312f8ad4bSLinus Torvalds * This sequence count validates that the parent had no 155412f8ad4bSLinus Torvalds * changes while we did the lookup of the dentry above. 155512f8ad4bSLinus Torvalds * 155612f8ad4bSLinus Torvalds * The memory barrier in read_seqcount_begin of child is 155712f8ad4bSLinus Torvalds * enough, we can use __read_seqcount_retry here. 155812f8ad4bSLinus Torvalds */ 155931e6b01fSNick Piggin if (__read_seqcount_retry(&parent->d_seq, nd->seq)) 156031e6b01fSNick Piggin return -ECHILD; 15615a18fff2SAl Viro 1562254cf582SAl Viro *seqp = seq; 156324643087SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) { 15644ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 15655a18fff2SAl Viro if (unlikely(status <= 0)) { 15665a18fff2SAl Viro if (status != -ECHILD) 15675a18fff2SAl Viro need_reval = 0; 15685a18fff2SAl Viro goto unlazy; 15695a18fff2SAl Viro } 157024643087SAl Viro } 1571daf3761cSTrond Myklebust /* 1572daf3761cSTrond Myklebust * Note: do negative dentry check after revalidation in 1573daf3761cSTrond Myklebust * case that drops it. 1574daf3761cSTrond Myklebust */ 1575daf3761cSTrond Myklebust if (negative) 1576daf3761cSTrond Myklebust return -ENOENT; 157731e6b01fSNick Piggin path->mnt = mnt; 157831e6b01fSNick Piggin path->dentry = dentry; 1579254cf582SAl Viro if (likely(__follow_mount_rcu(nd, path, inode, seqp))) 15809875cf80SDavid Howells return 0; 15815a18fff2SAl Viro unlazy: 1582254cf582SAl Viro if (unlazy_walk(nd, dentry, seq)) 15835a18fff2SAl Viro return -ECHILD; 15845a18fff2SAl Viro } else { 1585e97cdc87SAl Viro dentry = __d_lookup(parent, &nd->last); 158624643087SAl Viro } 15875a18fff2SAl Viro 158881e6f520SAl Viro if (unlikely(!dentry)) 158981e6f520SAl Viro goto need_lookup; 15905a18fff2SAl Viro 15915a18fff2SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval) 15924ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 15935a18fff2SAl Viro if (unlikely(status <= 0)) { 15945a18fff2SAl Viro if (status < 0) { 15955a18fff2SAl Viro dput(dentry); 15965a18fff2SAl Viro return status; 15975a18fff2SAl Viro } 15985542aa2fSEric W. Biederman d_invalidate(dentry); 15995a18fff2SAl Viro dput(dentry); 160081e6f520SAl Viro goto need_lookup; 16015a18fff2SAl Viro } 1602697f514dSMiklos Szeredi 1603766c4cbfSAl Viro if (unlikely(d_is_negative(dentry))) { 1604766c4cbfSAl Viro dput(dentry); 1605766c4cbfSAl Viro return -ENOENT; 1606766c4cbfSAl Viro } 16071da177e4SLinus Torvalds path->mnt = mnt; 16081da177e4SLinus Torvalds path->dentry = dentry; 1609756daf26SNeilBrown err = follow_managed(path, nd); 16108402752eSAl Viro if (likely(!err)) 161163afdfc7SDavid Howells *inode = d_backing_inode(path->dentry); 16128402752eSAl Viro return err; 161381e6f520SAl Viro 161481e6f520SAl Viro need_lookup: 1615697f514dSMiklos Szeredi return 1; 1616697f514dSMiklos Szeredi } 1617697f514dSMiklos Szeredi 1618697f514dSMiklos Szeredi /* Fast lookup failed, do it the slow way */ 1619cc2a5271SAl Viro static int lookup_slow(struct nameidata *nd, struct path *path) 1620697f514dSMiklos Szeredi { 1621697f514dSMiklos Szeredi struct dentry *dentry, *parent; 1622697f514dSMiklos Szeredi 1623697f514dSMiklos Szeredi parent = nd->path.dentry; 162481e6f520SAl Viro BUG_ON(nd->inode != parent->d_inode); 162581e6f520SAl Viro 162681e6f520SAl Viro mutex_lock(&parent->d_inode->i_mutex); 1627cc2a5271SAl Viro dentry = __lookup_hash(&nd->last, parent, nd->flags); 162881e6f520SAl Viro mutex_unlock(&parent->d_inode->i_mutex); 162981e6f520SAl Viro if (IS_ERR(dentry)) 163081e6f520SAl Viro return PTR_ERR(dentry); 1631697f514dSMiklos Szeredi path->mnt = nd->path.mnt; 1632697f514dSMiklos Szeredi path->dentry = dentry; 16338402752eSAl Viro return follow_managed(path, nd); 16341da177e4SLinus Torvalds } 16351da177e4SLinus Torvalds 163652094c8aSAl Viro static inline int may_lookup(struct nameidata *nd) 163752094c8aSAl Viro { 163852094c8aSAl Viro if (nd->flags & LOOKUP_RCU) { 16394ad5abb3SAl Viro int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); 164052094c8aSAl Viro if (err != -ECHILD) 164152094c8aSAl Viro return err; 16426e9918b7SAl Viro if (unlazy_walk(nd, NULL, 0)) 164352094c8aSAl Viro return -ECHILD; 164452094c8aSAl Viro } 16454ad5abb3SAl Viro return inode_permission(nd->inode, MAY_EXEC); 164652094c8aSAl Viro } 164752094c8aSAl Viro 16489856fa1bSAl Viro static inline int handle_dots(struct nameidata *nd, int type) 16499856fa1bSAl Viro { 16509856fa1bSAl Viro if (type == LAST_DOTDOT) { 16519e6697e2SAl Viro if (!nd->root.mnt) 16529e6697e2SAl Viro set_root(nd); 16539856fa1bSAl Viro if (nd->flags & LOOKUP_RCU) { 165470291aecSAl Viro return follow_dotdot_rcu(nd); 16559856fa1bSAl Viro } else 1656397d425dSEric W. Biederman return follow_dotdot(nd); 16579856fa1bSAl Viro } 16589856fa1bSAl Viro return 0; 16599856fa1bSAl Viro } 16609856fa1bSAl Viro 1661181548c0SAl Viro static int pick_link(struct nameidata *nd, struct path *link, 1662181548c0SAl Viro struct inode *inode, unsigned seq) 1663d63ff28fSAl Viro { 1664626de996SAl Viro int error; 16651cf2665bSAl Viro struct saved *last; 1666756daf26SNeilBrown if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) { 1667626de996SAl Viro path_to_nameidata(link, nd); 1668626de996SAl Viro return -ELOOP; 1669626de996SAl Viro } 1670bc40aee0SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1671cd179f44SAl Viro if (link->mnt == nd->path.mnt) 1672cd179f44SAl Viro mntget(link->mnt); 16737973387aSAl Viro } 1674626de996SAl Viro error = nd_alloc_stack(nd); 1675626de996SAl Viro if (unlikely(error)) { 1676bc40aee0SAl Viro if (error == -ECHILD) { 1677bc40aee0SAl Viro if (unlikely(unlazy_link(nd, link, seq))) 1678bc40aee0SAl Viro return -ECHILD; 1679bc40aee0SAl Viro error = nd_alloc_stack(nd); 1680bc40aee0SAl Viro } 1681bc40aee0SAl Viro if (error) { 1682cd179f44SAl Viro path_put(link); 1683626de996SAl Viro return error; 1684626de996SAl Viro } 1685bc40aee0SAl Viro } 1686626de996SAl Viro 1687ab104923SAl Viro last = nd->stack + nd->depth++; 16881cf2665bSAl Viro last->link = *link; 1689ab104923SAl Viro last->cookie = NULL; 1690181548c0SAl Viro last->inode = inode; 16910450b2d1SAl Viro last->seq = seq; 1692d63ff28fSAl Viro return 1; 1693d63ff28fSAl Viro } 1694d63ff28fSAl Viro 16953ddcd056SLinus Torvalds /* 16963ddcd056SLinus Torvalds * Do we need to follow links? We _really_ want to be able 16973ddcd056SLinus Torvalds * to do this check without having to look at inode->i_op, 16983ddcd056SLinus Torvalds * so we keep a cache of "no, this doesn't need follow_link" 16993ddcd056SLinus Torvalds * for the common case. 17003ddcd056SLinus Torvalds */ 1701254cf582SAl Viro static inline int should_follow_link(struct nameidata *nd, struct path *link, 1702181548c0SAl Viro int follow, 1703181548c0SAl Viro struct inode *inode, unsigned seq) 17043ddcd056SLinus Torvalds { 1705d63ff28fSAl Viro if (likely(!d_is_symlink(link->dentry))) 1706d63ff28fSAl Viro return 0; 1707d63ff28fSAl Viro if (!follow) 1708d63ff28fSAl Viro return 0; 1709181548c0SAl Viro return pick_link(nd, link, inode, seq); 17103ddcd056SLinus Torvalds } 17113ddcd056SLinus Torvalds 17124693a547SAl Viro enum {WALK_GET = 1, WALK_PUT = 2}; 17134693a547SAl Viro 17144693a547SAl Viro static int walk_component(struct nameidata *nd, int flags) 1715ce57dfc1SAl Viro { 1716caa85634SAl Viro struct path path; 1717ce57dfc1SAl Viro struct inode *inode; 1718254cf582SAl Viro unsigned seq; 1719ce57dfc1SAl Viro int err; 1720ce57dfc1SAl Viro /* 1721ce57dfc1SAl Viro * "." and ".." are special - ".." especially so because it has 1722ce57dfc1SAl Viro * to be able to know about the current root directory and 1723ce57dfc1SAl Viro * parent relationships. 1724ce57dfc1SAl Viro */ 17254693a547SAl Viro if (unlikely(nd->last_type != LAST_NORM)) { 17264693a547SAl Viro err = handle_dots(nd, nd->last_type); 17274693a547SAl Viro if (flags & WALK_PUT) 17284693a547SAl Viro put_link(nd); 17294693a547SAl Viro return err; 17304693a547SAl Viro } 1731254cf582SAl Viro err = lookup_fast(nd, &path, &inode, &seq); 1732ce57dfc1SAl Viro if (unlikely(err)) { 1733697f514dSMiklos Szeredi if (err < 0) 1734f0a9ba70SAl Viro return err; 1735697f514dSMiklos Szeredi 1736caa85634SAl Viro err = lookup_slow(nd, &path); 1737697f514dSMiklos Szeredi if (err < 0) 1738f0a9ba70SAl Viro return err; 1739697f514dSMiklos Szeredi 174063afdfc7SDavid Howells inode = d_backing_inode(path.dentry); 1741254cf582SAl Viro seq = 0; /* we are already out of RCU mode */ 1742697f514dSMiklos Szeredi err = -ENOENT; 1743caa85634SAl Viro if (d_is_negative(path.dentry)) 1744697f514dSMiklos Szeredi goto out_path_put; 1745766c4cbfSAl Viro } 1746697f514dSMiklos Szeredi 17474693a547SAl Viro if (flags & WALK_PUT) 17484693a547SAl Viro put_link(nd); 1749181548c0SAl Viro err = should_follow_link(nd, &path, flags & WALK_GET, inode, seq); 1750d63ff28fSAl Viro if (unlikely(err)) 1751d63ff28fSAl Viro return err; 1752caa85634SAl Viro path_to_nameidata(&path, nd); 1753ce57dfc1SAl Viro nd->inode = inode; 1754254cf582SAl Viro nd->seq = seq; 1755ce57dfc1SAl Viro return 0; 1756697f514dSMiklos Szeredi 1757697f514dSMiklos Szeredi out_path_put: 1758caa85634SAl Viro path_to_nameidata(&path, nd); 1759697f514dSMiklos Szeredi return err; 1760ce57dfc1SAl Viro } 1761ce57dfc1SAl Viro 17621da177e4SLinus Torvalds /* 1763bfcfaa77SLinus Torvalds * We can do the critical dentry name comparison and hashing 1764bfcfaa77SLinus Torvalds * operations one word at a time, but we are limited to: 1765bfcfaa77SLinus Torvalds * 1766bfcfaa77SLinus Torvalds * - Architectures with fast unaligned word accesses. We could 1767bfcfaa77SLinus Torvalds * do a "get_unaligned()" if this helps and is sufficiently 1768bfcfaa77SLinus Torvalds * fast. 1769bfcfaa77SLinus Torvalds * 1770bfcfaa77SLinus Torvalds * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we 1771bfcfaa77SLinus Torvalds * do not trap on the (extremely unlikely) case of a page 1772bfcfaa77SLinus Torvalds * crossing operation. 1773bfcfaa77SLinus Torvalds * 1774bfcfaa77SLinus Torvalds * - Furthermore, we need an efficient 64-bit compile for the 1775bfcfaa77SLinus Torvalds * 64-bit case in order to generate the "number of bytes in 1776bfcfaa77SLinus Torvalds * the final mask". Again, that could be replaced with a 1777bfcfaa77SLinus Torvalds * efficient population count instruction or similar. 1778bfcfaa77SLinus Torvalds */ 1779bfcfaa77SLinus Torvalds #ifdef CONFIG_DCACHE_WORD_ACCESS 1780bfcfaa77SLinus Torvalds 1781f68e556eSLinus Torvalds #include <asm/word-at-a-time.h> 1782bfcfaa77SLinus Torvalds 1783f68e556eSLinus Torvalds #ifdef CONFIG_64BIT 1784bfcfaa77SLinus Torvalds 1785bfcfaa77SLinus Torvalds static inline unsigned int fold_hash(unsigned long hash) 1786bfcfaa77SLinus Torvalds { 178799d263d4SLinus Torvalds return hash_64(hash, 32); 1788bfcfaa77SLinus Torvalds } 1789bfcfaa77SLinus Torvalds 1790bfcfaa77SLinus Torvalds #else /* 32-bit case */ 1791bfcfaa77SLinus Torvalds 1792bfcfaa77SLinus Torvalds #define fold_hash(x) (x) 1793bfcfaa77SLinus Torvalds 1794bfcfaa77SLinus Torvalds #endif 1795bfcfaa77SLinus Torvalds 1796bfcfaa77SLinus Torvalds unsigned int full_name_hash(const unsigned char *name, unsigned int len) 1797bfcfaa77SLinus Torvalds { 1798bfcfaa77SLinus Torvalds unsigned long a, mask; 1799bfcfaa77SLinus Torvalds unsigned long hash = 0; 1800bfcfaa77SLinus Torvalds 1801bfcfaa77SLinus Torvalds for (;;) { 1802e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name); 1803bfcfaa77SLinus Torvalds if (len < sizeof(unsigned long)) 1804bfcfaa77SLinus Torvalds break; 1805bfcfaa77SLinus Torvalds hash += a; 1806f132c5beSAl Viro hash *= 9; 1807bfcfaa77SLinus Torvalds name += sizeof(unsigned long); 1808bfcfaa77SLinus Torvalds len -= sizeof(unsigned long); 1809bfcfaa77SLinus Torvalds if (!len) 1810bfcfaa77SLinus Torvalds goto done; 1811bfcfaa77SLinus Torvalds } 1812a5c21dceSWill Deacon mask = bytemask_from_count(len); 1813bfcfaa77SLinus Torvalds hash += mask & a; 1814bfcfaa77SLinus Torvalds done: 1815bfcfaa77SLinus Torvalds return fold_hash(hash); 1816bfcfaa77SLinus Torvalds } 1817bfcfaa77SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 1818bfcfaa77SLinus Torvalds 1819bfcfaa77SLinus Torvalds /* 1820bfcfaa77SLinus Torvalds * Calculate the length and hash of the path component, and 1821d6bb3e90SLinus Torvalds * return the "hash_len" as the result. 1822bfcfaa77SLinus Torvalds */ 1823d6bb3e90SLinus Torvalds static inline u64 hash_name(const char *name) 1824bfcfaa77SLinus Torvalds { 182536126f8fSLinus Torvalds unsigned long a, b, adata, bdata, mask, hash, len; 182636126f8fSLinus Torvalds const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 1827bfcfaa77SLinus Torvalds 1828bfcfaa77SLinus Torvalds hash = a = 0; 1829bfcfaa77SLinus Torvalds len = -sizeof(unsigned long); 1830bfcfaa77SLinus Torvalds do { 1831bfcfaa77SLinus Torvalds hash = (hash + a) * 9; 1832bfcfaa77SLinus Torvalds len += sizeof(unsigned long); 1833e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name+len); 183436126f8fSLinus Torvalds b = a ^ REPEAT_BYTE('/'); 183536126f8fSLinus Torvalds } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); 1836bfcfaa77SLinus Torvalds 183736126f8fSLinus Torvalds adata = prep_zero_mask(a, adata, &constants); 183836126f8fSLinus Torvalds bdata = prep_zero_mask(b, bdata, &constants); 183936126f8fSLinus Torvalds 184036126f8fSLinus Torvalds mask = create_zero_mask(adata | bdata); 184136126f8fSLinus Torvalds 184236126f8fSLinus Torvalds hash += a & zero_bytemask(mask); 18439226b5b4SLinus Torvalds len += find_zero(mask); 1844d6bb3e90SLinus Torvalds return hashlen_create(fold_hash(hash), len); 1845bfcfaa77SLinus Torvalds } 1846bfcfaa77SLinus Torvalds 1847bfcfaa77SLinus Torvalds #else 1848bfcfaa77SLinus Torvalds 18490145acc2SLinus Torvalds unsigned int full_name_hash(const unsigned char *name, unsigned int len) 18500145acc2SLinus Torvalds { 18510145acc2SLinus Torvalds unsigned long hash = init_name_hash(); 18520145acc2SLinus Torvalds while (len--) 18530145acc2SLinus Torvalds hash = partial_name_hash(*name++, hash); 18540145acc2SLinus Torvalds return end_name_hash(hash); 18550145acc2SLinus Torvalds } 1856ae942ae7SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 18570145acc2SLinus Torvalds 18583ddcd056SLinus Torvalds /* 1859200e9ef7SLinus Torvalds * We know there's a real path component here of at least 1860200e9ef7SLinus Torvalds * one character. 1861200e9ef7SLinus Torvalds */ 1862d6bb3e90SLinus Torvalds static inline u64 hash_name(const char *name) 1863200e9ef7SLinus Torvalds { 1864200e9ef7SLinus Torvalds unsigned long hash = init_name_hash(); 1865200e9ef7SLinus Torvalds unsigned long len = 0, c; 1866200e9ef7SLinus Torvalds 1867200e9ef7SLinus Torvalds c = (unsigned char)*name; 1868200e9ef7SLinus Torvalds do { 1869200e9ef7SLinus Torvalds len++; 1870200e9ef7SLinus Torvalds hash = partial_name_hash(c, hash); 1871200e9ef7SLinus Torvalds c = (unsigned char)name[len]; 1872200e9ef7SLinus Torvalds } while (c && c != '/'); 1873d6bb3e90SLinus Torvalds return hashlen_create(end_name_hash(hash), len); 1874200e9ef7SLinus Torvalds } 1875200e9ef7SLinus Torvalds 1876bfcfaa77SLinus Torvalds #endif 1877bfcfaa77SLinus Torvalds 1878200e9ef7SLinus Torvalds /* 18791da177e4SLinus Torvalds * Name resolution. 1880ea3834d9SPrasanna Meda * This is the basic name resolution function, turning a pathname into 1881ea3834d9SPrasanna Meda * the final dentry. We expect 'base' to be positive and a directory. 18821da177e4SLinus Torvalds * 1883ea3834d9SPrasanna Meda * Returns 0 and nd will have valid dentry and mnt on success. 1884ea3834d9SPrasanna Meda * Returns error and drops reference to input namei data on failure. 18851da177e4SLinus Torvalds */ 18866de88d72SAl Viro static int link_path_walk(const char *name, struct nameidata *nd) 18871da177e4SLinus Torvalds { 18881da177e4SLinus Torvalds int err; 18891da177e4SLinus Torvalds 18901da177e4SLinus Torvalds while (*name=='/') 18911da177e4SLinus Torvalds name++; 18921da177e4SLinus Torvalds if (!*name) 18939e18f10aSAl Viro return 0; 18941da177e4SLinus Torvalds 18951da177e4SLinus Torvalds /* At this point we know we have a real path component. */ 18961da177e4SLinus Torvalds for(;;) { 1897d6bb3e90SLinus Torvalds u64 hash_len; 1898fe479a58SAl Viro int type; 18991da177e4SLinus Torvalds 190052094c8aSAl Viro err = may_lookup(nd); 19011da177e4SLinus Torvalds if (err) 19023595e234SAl Viro return err; 19031da177e4SLinus Torvalds 1904d6bb3e90SLinus Torvalds hash_len = hash_name(name); 19051da177e4SLinus Torvalds 1906fe479a58SAl Viro type = LAST_NORM; 1907d6bb3e90SLinus Torvalds if (name[0] == '.') switch (hashlen_len(hash_len)) { 1908fe479a58SAl Viro case 2: 1909200e9ef7SLinus Torvalds if (name[1] == '.') { 1910fe479a58SAl Viro type = LAST_DOTDOT; 191116c2cd71SAl Viro nd->flags |= LOOKUP_JUMPED; 191216c2cd71SAl Viro } 1913fe479a58SAl Viro break; 1914fe479a58SAl Viro case 1: 1915fe479a58SAl Viro type = LAST_DOT; 1916fe479a58SAl Viro } 19175a202bcdSAl Viro if (likely(type == LAST_NORM)) { 19185a202bcdSAl Viro struct dentry *parent = nd->path.dentry; 191916c2cd71SAl Viro nd->flags &= ~LOOKUP_JUMPED; 19205a202bcdSAl Viro if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { 1921a060dc50SJames Hogan struct qstr this = { { .hash_len = hash_len }, .name = name }; 1922da53be12SLinus Torvalds err = parent->d_op->d_hash(parent, &this); 19235a202bcdSAl Viro if (err < 0) 19243595e234SAl Viro return err; 1925d6bb3e90SLinus Torvalds hash_len = this.hash_len; 1926d6bb3e90SLinus Torvalds name = this.name; 19275a202bcdSAl Viro } 19285a202bcdSAl Viro } 1929fe479a58SAl Viro 1930d6bb3e90SLinus Torvalds nd->last.hash_len = hash_len; 1931d6bb3e90SLinus Torvalds nd->last.name = name; 19325f4a6a69SAl Viro nd->last_type = type; 19335f4a6a69SAl Viro 1934d6bb3e90SLinus Torvalds name += hashlen_len(hash_len); 1935d6bb3e90SLinus Torvalds if (!*name) 1936bdf6cbf1SAl Viro goto OK; 1937200e9ef7SLinus Torvalds /* 1938200e9ef7SLinus Torvalds * If it wasn't NUL, we know it was '/'. Skip that 1939200e9ef7SLinus Torvalds * slash, and continue until no more slashes. 1940200e9ef7SLinus Torvalds */ 1941200e9ef7SLinus Torvalds do { 1942d6bb3e90SLinus Torvalds name++; 1943d6bb3e90SLinus Torvalds } while (unlikely(*name == '/')); 19448620c238SAl Viro if (unlikely(!*name)) { 19458620c238SAl Viro OK: 1946368ee9baSAl Viro /* pathname body, done */ 19478620c238SAl Viro if (!nd->depth) 19488620c238SAl Viro return 0; 19498620c238SAl Viro name = nd->stack[nd->depth - 1].name; 1950368ee9baSAl Viro /* trailing symlink, done */ 19518620c238SAl Viro if (!name) 19528620c238SAl Viro return 0; 19538620c238SAl Viro /* last component of nested symlink */ 19544693a547SAl Viro err = walk_component(nd, WALK_GET | WALK_PUT); 19558620c238SAl Viro } else { 19564693a547SAl Viro err = walk_component(nd, WALK_GET); 19578620c238SAl Viro } 1958ce57dfc1SAl Viro if (err < 0) 19593595e234SAl Viro return err; 1960fe479a58SAl Viro 1961ce57dfc1SAl Viro if (err) { 1962626de996SAl Viro const char *s = get_link(nd); 19635a460275SAl Viro 1964a1c83681SViresh Kumar if (IS_ERR(s)) 19653595e234SAl Viro return PTR_ERR(s); 1966172a39a0SAl Viro err = 0; 196712b09578SAl Viro if (unlikely(!s)) { 196812b09578SAl Viro /* jumped */ 1969b9ff4429SAl Viro put_link(nd); 197012b09578SAl Viro } else { 1971071bf501SAl Viro nd->stack[nd->depth - 1].name = name; 197232cd7468SAl Viro name = s; 19739e18f10aSAl Viro continue; 197448c8b0c5SAl Viro } 197531e6b01fSNick Piggin } 197697242f99SAl Viro if (unlikely(!d_can_lookup(nd->path.dentry))) { 197797242f99SAl Viro if (nd->flags & LOOKUP_RCU) { 197897242f99SAl Viro if (unlazy_walk(nd, NULL, 0)) 197997242f99SAl Viro return -ECHILD; 198097242f99SAl Viro } 19813595e234SAl Viro return -ENOTDIR; 19825f4a6a69SAl Viro } 1983ce57dfc1SAl Viro } 198497242f99SAl Viro } 19851da177e4SLinus Torvalds 1986c8a53ee5SAl Viro static const char *path_init(struct nameidata *nd, unsigned flags) 198731e6b01fSNick Piggin { 198831e6b01fSNick Piggin int retval = 0; 1989c8a53ee5SAl Viro const char *s = nd->name->name; 199031e6b01fSNick Piggin 199131e6b01fSNick Piggin nd->last_type = LAST_ROOT; /* if there are only slashes... */ 1992980f3ea2SAl Viro nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT; 199331e6b01fSNick Piggin nd->depth = 0; 19945b6ca027SAl Viro if (flags & LOOKUP_ROOT) { 1995b18825a7SDavid Howells struct dentry *root = nd->root.dentry; 1996b18825a7SDavid Howells struct inode *inode = root->d_inode; 1997fd2f7cb5SAl Viro if (*s) { 199844b1d530SMiklos Szeredi if (!d_can_lookup(root)) 1999368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 20005b6ca027SAl Viro retval = inode_permission(inode, MAY_EXEC); 20015b6ca027SAl Viro if (retval) 2002368ee9baSAl Viro return ERR_PTR(retval); 200373d049a4SAl Viro } 20045b6ca027SAl Viro nd->path = nd->root; 20055b6ca027SAl Viro nd->inode = inode; 20065b6ca027SAl Viro if (flags & LOOKUP_RCU) { 20078b61e74fSAl Viro rcu_read_lock(); 20085b6ca027SAl Viro nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 20098f47a016SAl Viro nd->root_seq = nd->seq; 201048a066e7SAl Viro nd->m_seq = read_seqbegin(&mount_lock); 20115b6ca027SAl Viro } else { 20125b6ca027SAl Viro path_get(&nd->path); 20135b6ca027SAl Viro } 2014368ee9baSAl Viro return s; 20155b6ca027SAl Viro } 20165b6ca027SAl Viro 201731e6b01fSNick Piggin nd->root.mnt = NULL; 201831e6b01fSNick Piggin 201948a066e7SAl Viro nd->m_seq = read_seqbegin(&mount_lock); 2020fd2f7cb5SAl Viro if (*s == '/') { 20219e6697e2SAl Viro if (flags & LOOKUP_RCU) 20228b61e74fSAl Viro rcu_read_lock(); 2023e41f7d4eSAl Viro set_root(nd); 20249e6697e2SAl Viro if (flags & LOOKUP_RCU) { 20259e6697e2SAl Viro nd->seq = nd->root_seq; 202631e6b01fSNick Piggin nd->path = nd->root; 20279e6697e2SAl Viro } else { 20289e6697e2SAl Viro path_get(&nd->root); 20299e6697e2SAl Viro nd->path = nd->root; 20309e6697e2SAl Viro } 2031ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2032ef55d917SAl Viro if (!(flags & LOOKUP_RCU)) 2033ef55d917SAl Viro return s; 2034ef55d917SAl Viro if (likely(!read_seqcount_retry(&nd->path.dentry->d_seq, nd->seq))) 2035ef55d917SAl Viro return s; 2036ef55d917SAl Viro if (!(nd->flags & LOOKUP_ROOT)) 2037ef55d917SAl Viro nd->root.mnt = NULL; 2038ef55d917SAl Viro rcu_read_unlock(); 2039ef55d917SAl Viro return ERR_PTR(-ECHILD); 2040c8a53ee5SAl Viro } else if (nd->dfd == AT_FDCWD) { 2041e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 204231e6b01fSNick Piggin struct fs_struct *fs = current->fs; 2043c28cc364SNick Piggin unsigned seq; 204431e6b01fSNick Piggin 20458b61e74fSAl Viro rcu_read_lock(); 204631e6b01fSNick Piggin 2047c28cc364SNick Piggin do { 2048c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 204931e6b01fSNick Piggin nd->path = fs->pwd; 2050ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2051c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 2052c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 2053e41f7d4eSAl Viro } else { 2054e41f7d4eSAl Viro get_fs_pwd(current->fs, &nd->path); 2055ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2056e41f7d4eSAl Viro } 2057ef55d917SAl Viro return s; 205831e6b01fSNick Piggin } else { 2059582aa64aSJeff Layton /* Caller must check execute permissions on the starting path component */ 2060c8a53ee5SAl Viro struct fd f = fdget_raw(nd->dfd); 206131e6b01fSNick Piggin struct dentry *dentry; 206231e6b01fSNick Piggin 20632903ff01SAl Viro if (!f.file) 2064368ee9baSAl Viro return ERR_PTR(-EBADF); 206531e6b01fSNick Piggin 20662903ff01SAl Viro dentry = f.file->f_path.dentry; 206731e6b01fSNick Piggin 2068fd2f7cb5SAl Viro if (*s) { 206944b1d530SMiklos Szeredi if (!d_can_lookup(dentry)) { 20702903ff01SAl Viro fdput(f); 2071368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 2072f52e0c11SAl Viro } 20732903ff01SAl Viro } 20742903ff01SAl Viro 20752903ff01SAl Viro nd->path = f.file->f_path; 2076e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 20778b61e74fSAl Viro rcu_read_lock(); 207834a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 207934a26b99SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 20805590ff0dSUlrich Drepper } else { 20812903ff01SAl Viro path_get(&nd->path); 208234a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 20831da177e4SLinus Torvalds } 208434a26b99SAl Viro fdput(f); 2085368ee9baSAl Viro return s; 2086e41f7d4eSAl Viro } 20879b4a9b14SAl Viro } 20889b4a9b14SAl Viro 20893bdba28bSAl Viro static const char *trailing_symlink(struct nameidata *nd) 209095fa25d9SAl Viro { 209195fa25d9SAl Viro const char *s; 2092fec2fa24SAl Viro int error = may_follow_link(nd); 2093deb106c6SAl Viro if (unlikely(error)) 20943bdba28bSAl Viro return ERR_PTR(error); 209595fa25d9SAl Viro nd->flags |= LOOKUP_PARENT; 2096fab51e8aSAl Viro nd->stack[0].name = NULL; 20973b2e7f75SAl Viro s = get_link(nd); 2098deb106c6SAl Viro return s ? s : ""; 209995fa25d9SAl Viro } 210095fa25d9SAl Viro 2101caa85634SAl Viro static inline int lookup_last(struct nameidata *nd) 2102bd92d7feSAl Viro { 2103bd92d7feSAl Viro if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) 2104bd92d7feSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 2105bd92d7feSAl Viro 2106bd92d7feSAl Viro nd->flags &= ~LOOKUP_PARENT; 2107deb106c6SAl Viro return walk_component(nd, 21084693a547SAl Viro nd->flags & LOOKUP_FOLLOW 21094693a547SAl Viro ? nd->depth 21104693a547SAl Viro ? WALK_PUT | WALK_GET 21114693a547SAl Viro : WALK_GET 21124693a547SAl Viro : 0); 2113bd92d7feSAl Viro } 2114bd92d7feSAl Viro 21159b4a9b14SAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2116c8a53ee5SAl Viro static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path) 21179b4a9b14SAl Viro { 2118c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2119bd92d7feSAl Viro int err; 212031e6b01fSNick Piggin 2121368ee9baSAl Viro if (IS_ERR(s)) 2122368ee9baSAl Viro return PTR_ERR(s); 21233bdba28bSAl Viro while (!(err = link_path_walk(s, nd)) 21243bdba28bSAl Viro && ((err = lookup_last(nd)) > 0)) { 21253bdba28bSAl Viro s = trailing_symlink(nd); 21263bdba28bSAl Viro if (IS_ERR(s)) { 21273bdba28bSAl Viro err = PTR_ERR(s); 21286d7b5aaeSAl Viro break; 2129bd92d7feSAl Viro } 2130bd92d7feSAl Viro } 21319f1fafeeSAl Viro if (!err) 21329f1fafeeSAl Viro err = complete_walk(nd); 2133bd92d7feSAl Viro 2134deb106c6SAl Viro if (!err && nd->flags & LOOKUP_DIRECTORY) 2135deb106c6SAl Viro if (!d_can_lookup(nd->path.dentry)) 2136bd23a539SAl Viro err = -ENOTDIR; 2137625b6d10SAl Viro if (!err) { 2138625b6d10SAl Viro *path = nd->path; 2139625b6d10SAl Viro nd->path.mnt = NULL; 2140625b6d10SAl Viro nd->path.dentry = NULL; 2141625b6d10SAl Viro } 2142deb106c6SAl Viro terminate_walk(nd); 2143bd92d7feSAl Viro return err; 214431e6b01fSNick Piggin } 214531e6b01fSNick Piggin 2146625b6d10SAl Viro static int filename_lookup(int dfd, struct filename *name, unsigned flags, 21479ad1aaa6SAl Viro struct path *path, struct path *root) 2148873f1eedSJeff Layton { 2149894bc8c4SAl Viro int retval; 21509883d185SAl Viro struct nameidata nd; 2151abc9f5beSAl Viro if (IS_ERR(name)) 2152abc9f5beSAl Viro return PTR_ERR(name); 21539ad1aaa6SAl Viro if (unlikely(root)) { 21549ad1aaa6SAl Viro nd.root = *root; 21559ad1aaa6SAl Viro flags |= LOOKUP_ROOT; 21569ad1aaa6SAl Viro } 21579883d185SAl Viro set_nameidata(&nd, dfd, name); 2158c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_RCU, path); 2159873f1eedSJeff Layton if (unlikely(retval == -ECHILD)) 2160c8a53ee5SAl Viro retval = path_lookupat(&nd, flags, path); 2161873f1eedSJeff Layton if (unlikely(retval == -ESTALE)) 2162c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path); 2163873f1eedSJeff Layton 2164873f1eedSJeff Layton if (likely(!retval)) 2165625b6d10SAl Viro audit_inode(name, path->dentry, flags & LOOKUP_PARENT); 21669883d185SAl Viro restore_nameidata(); 2167e4bd1c1aSAl Viro putname(name); 2168873f1eedSJeff Layton return retval; 2169873f1eedSJeff Layton } 2170873f1eedSJeff Layton 21718bcb77faSAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2172c8a53ee5SAl Viro static int path_parentat(struct nameidata *nd, unsigned flags, 2173391172c4SAl Viro struct path *parent) 21748bcb77faSAl Viro { 2175c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2176368ee9baSAl Viro int err; 2177368ee9baSAl Viro if (IS_ERR(s)) 2178368ee9baSAl Viro return PTR_ERR(s); 2179368ee9baSAl Viro err = link_path_walk(s, nd); 21808bcb77faSAl Viro if (!err) 21818bcb77faSAl Viro err = complete_walk(nd); 2182391172c4SAl Viro if (!err) { 2183391172c4SAl Viro *parent = nd->path; 2184391172c4SAl Viro nd->path.mnt = NULL; 2185391172c4SAl Viro nd->path.dentry = NULL; 2186391172c4SAl Viro } 2187deb106c6SAl Viro terminate_walk(nd); 21888bcb77faSAl Viro return err; 21898bcb77faSAl Viro } 21908bcb77faSAl Viro 21915c31b6ceSAl Viro static struct filename *filename_parentat(int dfd, struct filename *name, 2192391172c4SAl Viro unsigned int flags, struct path *parent, 2193391172c4SAl Viro struct qstr *last, int *type) 21948bcb77faSAl Viro { 21958bcb77faSAl Viro int retval; 21969883d185SAl Viro struct nameidata nd; 21978bcb77faSAl Viro 21985c31b6ceSAl Viro if (IS_ERR(name)) 21995c31b6ceSAl Viro return name; 22009883d185SAl Viro set_nameidata(&nd, dfd, name); 2201c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_RCU, parent); 22028bcb77faSAl Viro if (unlikely(retval == -ECHILD)) 2203c8a53ee5SAl Viro retval = path_parentat(&nd, flags, parent); 22048bcb77faSAl Viro if (unlikely(retval == -ESTALE)) 2205c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent); 2206391172c4SAl Viro if (likely(!retval)) { 2207391172c4SAl Viro *last = nd.last; 2208391172c4SAl Viro *type = nd.last_type; 2209391172c4SAl Viro audit_inode(name, parent->dentry, LOOKUP_PARENT); 22105c31b6ceSAl Viro } else { 22115c31b6ceSAl Viro putname(name); 22125c31b6ceSAl Viro name = ERR_PTR(retval); 2213391172c4SAl Viro } 22149883d185SAl Viro restore_nameidata(); 22155c31b6ceSAl Viro return name; 22168bcb77faSAl Viro } 22178bcb77faSAl Viro 221879714f72SAl Viro /* does lookup, returns the object with parent locked */ 221979714f72SAl Viro struct dentry *kern_path_locked(const char *name, struct path *path) 22205590ff0dSUlrich Drepper { 22215c31b6ceSAl Viro struct filename *filename; 22225c31b6ceSAl Viro struct dentry *d; 2223391172c4SAl Viro struct qstr last; 2224391172c4SAl Viro int type; 222551689104SPaul Moore 22265c31b6ceSAl Viro filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path, 22275c31b6ceSAl Viro &last, &type); 222851689104SPaul Moore if (IS_ERR(filename)) 222951689104SPaul Moore return ERR_CAST(filename); 22305c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) { 2231391172c4SAl Viro path_put(path); 22325c31b6ceSAl Viro putname(filename); 22335c31b6ceSAl Viro return ERR_PTR(-EINVAL); 223479714f72SAl Viro } 2235391172c4SAl Viro mutex_lock_nested(&path->dentry->d_inode->i_mutex, I_MUTEX_PARENT); 2236391172c4SAl Viro d = __lookup_hash(&last, path->dentry, 0); 223779714f72SAl Viro if (IS_ERR(d)) { 2238391172c4SAl Viro mutex_unlock(&path->dentry->d_inode->i_mutex); 2239391172c4SAl Viro path_put(path); 224079714f72SAl Viro } 224151689104SPaul Moore putname(filename); 224279714f72SAl Viro return d; 22435590ff0dSUlrich Drepper } 22445590ff0dSUlrich Drepper 2245d1811465SAl Viro int kern_path(const char *name, unsigned int flags, struct path *path) 2246d1811465SAl Viro { 2247abc9f5beSAl Viro return filename_lookup(AT_FDCWD, getname_kernel(name), 2248abc9f5beSAl Viro flags, path, NULL); 2249d1811465SAl Viro } 22504d359507SAl Viro EXPORT_SYMBOL(kern_path); 2251d1811465SAl Viro 225216f18200SJosef 'Jeff' Sipek /** 225316f18200SJosef 'Jeff' Sipek * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair 225416f18200SJosef 'Jeff' Sipek * @dentry: pointer to dentry of the base directory 225516f18200SJosef 'Jeff' Sipek * @mnt: pointer to vfs mount of the base directory 225616f18200SJosef 'Jeff' Sipek * @name: pointer to file name 225716f18200SJosef 'Jeff' Sipek * @flags: lookup flags 2258e0a01249SAl Viro * @path: pointer to struct path to fill 225916f18200SJosef 'Jeff' Sipek */ 226016f18200SJosef 'Jeff' Sipek int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, 226116f18200SJosef 'Jeff' Sipek const char *name, unsigned int flags, 2262e0a01249SAl Viro struct path *path) 226316f18200SJosef 'Jeff' Sipek { 22649ad1aaa6SAl Viro struct path root = {.mnt = mnt, .dentry = dentry}; 22659ad1aaa6SAl Viro /* the first argument of filename_lookup() is ignored with root */ 2266abc9f5beSAl Viro return filename_lookup(AT_FDCWD, getname_kernel(name), 2267abc9f5beSAl Viro flags , path, &root); 226816f18200SJosef 'Jeff' Sipek } 22694d359507SAl Viro EXPORT_SYMBOL(vfs_path_lookup); 227016f18200SJosef 'Jeff' Sipek 2271eead1911SChristoph Hellwig /** 2272a6b91919SRandy Dunlap * lookup_one_len - filesystem helper to lookup single pathname component 2273eead1911SChristoph Hellwig * @name: pathname component to lookup 2274eead1911SChristoph Hellwig * @base: base directory to lookup from 2275eead1911SChristoph Hellwig * @len: maximum length @len should be interpreted to 2276eead1911SChristoph Hellwig * 2277a6b91919SRandy Dunlap * Note that this routine is purely a helper for filesystem usage and should 22789e7543e9SAl Viro * not be called by generic code. 2279eead1911SChristoph Hellwig */ 2280057f6c01SJames Morris struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 2281057f6c01SJames Morris { 2282057f6c01SJames Morris struct qstr this; 22836a96ba54SAl Viro unsigned int c; 2284cda309deSMiklos Szeredi int err; 2285057f6c01SJames Morris 22862f9092e1SDavid Woodhouse WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex)); 22872f9092e1SDavid Woodhouse 22886a96ba54SAl Viro this.name = name; 22896a96ba54SAl Viro this.len = len; 22900145acc2SLinus Torvalds this.hash = full_name_hash(name, len); 22916a96ba54SAl Viro if (!len) 22926a96ba54SAl Viro return ERR_PTR(-EACCES); 22936a96ba54SAl Viro 229421d8a15aSAl Viro if (unlikely(name[0] == '.')) { 229521d8a15aSAl Viro if (len < 2 || (len == 2 && name[1] == '.')) 229621d8a15aSAl Viro return ERR_PTR(-EACCES); 229721d8a15aSAl Viro } 229821d8a15aSAl Viro 22996a96ba54SAl Viro while (len--) { 23006a96ba54SAl Viro c = *(const unsigned char *)name++; 23016a96ba54SAl Viro if (c == '/' || c == '\0') 23026a96ba54SAl Viro return ERR_PTR(-EACCES); 23036a96ba54SAl Viro } 23045a202bcdSAl Viro /* 23055a202bcdSAl Viro * See if the low-level filesystem might want 23065a202bcdSAl Viro * to use its own hash.. 23075a202bcdSAl Viro */ 23085a202bcdSAl Viro if (base->d_flags & DCACHE_OP_HASH) { 2309da53be12SLinus Torvalds int err = base->d_op->d_hash(base, &this); 23105a202bcdSAl Viro if (err < 0) 23115a202bcdSAl Viro return ERR_PTR(err); 23125a202bcdSAl Viro } 2313eead1911SChristoph Hellwig 2314cda309deSMiklos Szeredi err = inode_permission(base->d_inode, MAY_EXEC); 2315cda309deSMiklos Szeredi if (err) 2316cda309deSMiklos Szeredi return ERR_PTR(err); 2317cda309deSMiklos Szeredi 231872bd866aSAl Viro return __lookup_hash(&this, base, 0); 2319057f6c01SJames Morris } 23204d359507SAl Viro EXPORT_SYMBOL(lookup_one_len); 2321057f6c01SJames Morris 23221fa1e7f6SAndy Whitcroft int user_path_at_empty(int dfd, const char __user *name, unsigned flags, 23231fa1e7f6SAndy Whitcroft struct path *path, int *empty) 23241da177e4SLinus Torvalds { 2325abc9f5beSAl Viro return filename_lookup(dfd, getname_flags(name, flags, empty), 2326abc9f5beSAl Viro flags, path, NULL); 23271da177e4SLinus Torvalds } 2328b853a161SAl Viro EXPORT_SYMBOL(user_path_at_empty); 23291fa1e7f6SAndy Whitcroft 2330873f1eedSJeff Layton /* 2331873f1eedSJeff Layton * NB: most callers don't do anything directly with the reference to the 2332873f1eedSJeff Layton * to struct filename, but the nd->last pointer points into the name string 2333873f1eedSJeff Layton * allocated by getname. So we must hold the reference to it until all 2334873f1eedSJeff Layton * path-walking is complete. 2335873f1eedSJeff Layton */ 2336a2ec4a2dSAl Viro static inline struct filename * 2337f5beed75SAl Viro user_path_parent(int dfd, const char __user *path, 2338f5beed75SAl Viro struct path *parent, 2339f5beed75SAl Viro struct qstr *last, 2340f5beed75SAl Viro int *type, 23419e790bd6SJeff Layton unsigned int flags) 23422ad94ae6SAl Viro { 23439e790bd6SJeff Layton /* only LOOKUP_REVAL is allowed in extra flags */ 23445c31b6ceSAl Viro return filename_parentat(dfd, getname(path), flags & LOOKUP_REVAL, 23455c31b6ceSAl Viro parent, last, type); 23462ad94ae6SAl Viro } 23472ad94ae6SAl Viro 23488033426eSJeff Layton /** 2349197df04cSAl Viro * mountpoint_last - look up last component for umount 23508033426eSJeff Layton * @nd: pathwalk nameidata - currently pointing at parent directory of "last" 23518033426eSJeff Layton * @path: pointer to container for result 23528033426eSJeff Layton * 23538033426eSJeff Layton * This is a special lookup_last function just for umount. In this case, we 23548033426eSJeff Layton * need to resolve the path without doing any revalidation. 23558033426eSJeff Layton * 23568033426eSJeff Layton * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since 23578033426eSJeff Layton * mountpoints are always pinned in the dcache, their ancestors are too. Thus, 23588033426eSJeff Layton * in almost all cases, this lookup will be served out of the dcache. The only 23598033426eSJeff Layton * cases where it won't are if nd->last refers to a symlink or the path is 23608033426eSJeff Layton * bogus and it doesn't exist. 23618033426eSJeff Layton * 23628033426eSJeff Layton * Returns: 23638033426eSJeff Layton * -error: if there was an error during lookup. This includes -ENOENT if the 23648033426eSJeff Layton * lookup found a negative dentry. The nd->path reference will also be 23658033426eSJeff Layton * put in this case. 23668033426eSJeff Layton * 23678033426eSJeff Layton * 0: if we successfully resolved nd->path and found it to not to be a 23688033426eSJeff Layton * symlink that needs to be followed. "path" will also be populated. 23698033426eSJeff Layton * The nd->path reference will also be put. 23708033426eSJeff Layton * 23718033426eSJeff Layton * 1: if we successfully resolved nd->last and found it to be a symlink 23728033426eSJeff Layton * that needs to be followed. "path" will be populated with the path 23738033426eSJeff Layton * to the link, and nd->path will *not* be put. 23748033426eSJeff Layton */ 23758033426eSJeff Layton static int 2376197df04cSAl Viro mountpoint_last(struct nameidata *nd, struct path *path) 23778033426eSJeff Layton { 23788033426eSJeff Layton int error = 0; 23798033426eSJeff Layton struct dentry *dentry; 23808033426eSJeff Layton struct dentry *dir = nd->path.dentry; 23818033426eSJeff Layton 238235759521SAl Viro /* If we're in rcuwalk, drop out of it to handle last component */ 238335759521SAl Viro if (nd->flags & LOOKUP_RCU) { 23846e9918b7SAl Viro if (unlazy_walk(nd, NULL, 0)) 2385deb106c6SAl Viro return -ECHILD; 23868033426eSJeff Layton } 23878033426eSJeff Layton 23888033426eSJeff Layton nd->flags &= ~LOOKUP_PARENT; 23898033426eSJeff Layton 23908033426eSJeff Layton if (unlikely(nd->last_type != LAST_NORM)) { 23918033426eSJeff Layton error = handle_dots(nd, nd->last_type); 239235759521SAl Viro if (error) 2393deb106c6SAl Viro return error; 23948033426eSJeff Layton dentry = dget(nd->path.dentry); 239535759521SAl Viro goto done; 23968033426eSJeff Layton } 23978033426eSJeff Layton 23988033426eSJeff Layton mutex_lock(&dir->d_inode->i_mutex); 23998033426eSJeff Layton dentry = d_lookup(dir, &nd->last); 24008033426eSJeff Layton if (!dentry) { 24018033426eSJeff Layton /* 24028033426eSJeff Layton * No cached dentry. Mounted dentries are pinned in the cache, 24038033426eSJeff Layton * so that means that this dentry is probably a symlink or the 24048033426eSJeff Layton * path doesn't actually point to a mounted dentry. 24058033426eSJeff Layton */ 24068033426eSJeff Layton dentry = d_alloc(dir, &nd->last); 24078033426eSJeff Layton if (!dentry) { 2408bcceeebaSDave Jones mutex_unlock(&dir->d_inode->i_mutex); 2409deb106c6SAl Viro return -ENOMEM; 24108033426eSJeff Layton } 241135759521SAl Viro dentry = lookup_real(dir->d_inode, dentry, nd->flags); 2412bcceeebaSDave Jones if (IS_ERR(dentry)) { 2413bcceeebaSDave Jones mutex_unlock(&dir->d_inode->i_mutex); 2414deb106c6SAl Viro return PTR_ERR(dentry); 24158033426eSJeff Layton } 2416bcceeebaSDave Jones } 24178033426eSJeff Layton mutex_unlock(&dir->d_inode->i_mutex); 24188033426eSJeff Layton 241935759521SAl Viro done: 2420698934dfSDavid Howells if (d_is_negative(dentry)) { 24218033426eSJeff Layton dput(dentry); 2422deb106c6SAl Viro return -ENOENT; 242335759521SAl Viro } 2424191d7f73SAl Viro if (nd->depth) 2425191d7f73SAl Viro put_link(nd); 24268033426eSJeff Layton path->dentry = dentry; 2427295dc39dSVasily Averin path->mnt = nd->path.mnt; 2428181548c0SAl Viro error = should_follow_link(nd, path, nd->flags & LOOKUP_FOLLOW, 2429181548c0SAl Viro d_backing_inode(dentry), 0); 2430deb106c6SAl Viro if (unlikely(error)) 2431d63ff28fSAl Viro return error; 2432295dc39dSVasily Averin mntget(path->mnt); 24338033426eSJeff Layton follow_mount(path); 2434deb106c6SAl Viro return 0; 24358033426eSJeff Layton } 24368033426eSJeff Layton 24378033426eSJeff Layton /** 2438197df04cSAl Viro * path_mountpoint - look up a path to be umounted 24392a78b857SMasanari Iida * @nd: lookup context 24408033426eSJeff Layton * @flags: lookup flags 2441c8a53ee5SAl Viro * @path: pointer to container for result 24428033426eSJeff Layton * 24438033426eSJeff Layton * Look up the given name, but don't attempt to revalidate the last component. 2444606d6fe3SRandy Dunlap * Returns 0 and "path" will be valid on success; Returns error otherwise. 24458033426eSJeff Layton */ 24468033426eSJeff Layton static int 2447c8a53ee5SAl Viro path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path) 24488033426eSJeff Layton { 2449c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2450368ee9baSAl Viro int err; 2451368ee9baSAl Viro if (IS_ERR(s)) 2452368ee9baSAl Viro return PTR_ERR(s); 24533bdba28bSAl Viro while (!(err = link_path_walk(s, nd)) && 24543bdba28bSAl Viro (err = mountpoint_last(nd, path)) > 0) { 24553bdba28bSAl Viro s = trailing_symlink(nd); 24563bdba28bSAl Viro if (IS_ERR(s)) { 24573bdba28bSAl Viro err = PTR_ERR(s); 24588033426eSJeff Layton break; 24598033426eSJeff Layton } 24603bdba28bSAl Viro } 2461deb106c6SAl Viro terminate_walk(nd); 24628033426eSJeff Layton return err; 24638033426eSJeff Layton } 24648033426eSJeff Layton 24652d864651SAl Viro static int 2466668696dcSAl Viro filename_mountpoint(int dfd, struct filename *name, struct path *path, 24672d864651SAl Viro unsigned int flags) 24682d864651SAl Viro { 24699883d185SAl Viro struct nameidata nd; 2470cbaab2dbSAl Viro int error; 2471668696dcSAl Viro if (IS_ERR(name)) 2472668696dcSAl Viro return PTR_ERR(name); 24739883d185SAl Viro set_nameidata(&nd, dfd, name); 2474c8a53ee5SAl Viro error = path_mountpoint(&nd, flags | LOOKUP_RCU, path); 24752d864651SAl Viro if (unlikely(error == -ECHILD)) 2476c8a53ee5SAl Viro error = path_mountpoint(&nd, flags, path); 24772d864651SAl Viro if (unlikely(error == -ESTALE)) 2478c8a53ee5SAl Viro error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path); 24792d864651SAl Viro if (likely(!error)) 2480668696dcSAl Viro audit_inode(name, path->dentry, 0); 24819883d185SAl Viro restore_nameidata(); 2482668696dcSAl Viro putname(name); 24832d864651SAl Viro return error; 24842d864651SAl Viro } 24852d864651SAl Viro 24868033426eSJeff Layton /** 2487197df04cSAl Viro * user_path_mountpoint_at - lookup a path from userland in order to umount it 24888033426eSJeff Layton * @dfd: directory file descriptor 24898033426eSJeff Layton * @name: pathname from userland 24908033426eSJeff Layton * @flags: lookup flags 24918033426eSJeff Layton * @path: pointer to container to hold result 24928033426eSJeff Layton * 24938033426eSJeff Layton * A umount is a special case for path walking. We're not actually interested 24948033426eSJeff Layton * in the inode in this situation, and ESTALE errors can be a problem. We 24958033426eSJeff Layton * simply want track down the dentry and vfsmount attached at the mountpoint 24968033426eSJeff Layton * and avoid revalidating the last component. 24978033426eSJeff Layton * 24988033426eSJeff Layton * Returns 0 and populates "path" on success. 24998033426eSJeff Layton */ 25008033426eSJeff Layton int 2501197df04cSAl Viro user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags, 25028033426eSJeff Layton struct path *path) 25038033426eSJeff Layton { 2504cbaab2dbSAl Viro return filename_mountpoint(dfd, getname(name), path, flags); 25058033426eSJeff Layton } 25068033426eSJeff Layton 25072d864651SAl Viro int 25082d864651SAl Viro kern_path_mountpoint(int dfd, const char *name, struct path *path, 25092d864651SAl Viro unsigned int flags) 25102d864651SAl Viro { 2511cbaab2dbSAl Viro return filename_mountpoint(dfd, getname_kernel(name), path, flags); 25122d864651SAl Viro } 25132d864651SAl Viro EXPORT_SYMBOL(kern_path_mountpoint); 25142d864651SAl Viro 2515cbdf35bcSMiklos Szeredi int __check_sticky(struct inode *dir, struct inode *inode) 25161da177e4SLinus Torvalds { 25178e96e3b7SEric W. Biederman kuid_t fsuid = current_fsuid(); 2518da9592edSDavid Howells 25198e96e3b7SEric W. Biederman if (uid_eq(inode->i_uid, fsuid)) 25201da177e4SLinus Torvalds return 0; 25218e96e3b7SEric W. Biederman if (uid_eq(dir->i_uid, fsuid)) 25221da177e4SLinus Torvalds return 0; 252323adbe12SAndy Lutomirski return !capable_wrt_inode_uidgid(inode, CAP_FOWNER); 25241da177e4SLinus Torvalds } 2525cbdf35bcSMiklos Szeredi EXPORT_SYMBOL(__check_sticky); 25261da177e4SLinus Torvalds 25271da177e4SLinus Torvalds /* 25281da177e4SLinus Torvalds * Check whether we can remove a link victim from directory dir, check 25291da177e4SLinus Torvalds * whether the type of victim is right. 25301da177e4SLinus Torvalds * 1. We can't do it if dir is read-only (done in permission()) 25311da177e4SLinus Torvalds * 2. We should have write and exec permissions on dir 25321da177e4SLinus Torvalds * 3. We can't remove anything from append-only dir 25331da177e4SLinus Torvalds * 4. We can't do anything with immutable dir (done in permission()) 25341da177e4SLinus Torvalds * 5. If the sticky bit on dir is set we should either 25351da177e4SLinus Torvalds * a. be owner of dir, or 25361da177e4SLinus Torvalds * b. be owner of victim, or 25371da177e4SLinus Torvalds * c. have CAP_FOWNER capability 25381da177e4SLinus Torvalds * 6. If the victim is append-only or immutable we can't do antyhing with 25391da177e4SLinus Torvalds * links pointing to it. 25401da177e4SLinus Torvalds * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. 25411da177e4SLinus Torvalds * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. 25421da177e4SLinus Torvalds * 9. We can't remove a root or mountpoint. 25431da177e4SLinus Torvalds * 10. We don't allow removal of NFS sillyrenamed files; it's handled by 25441da177e4SLinus Torvalds * nfs_async_unlink(). 25451da177e4SLinus Torvalds */ 2546b18825a7SDavid Howells static int may_delete(struct inode *dir, struct dentry *victim, bool isdir) 25471da177e4SLinus Torvalds { 254863afdfc7SDavid Howells struct inode *inode = d_backing_inode(victim); 25491da177e4SLinus Torvalds int error; 25501da177e4SLinus Torvalds 2551b18825a7SDavid Howells if (d_is_negative(victim)) 25521da177e4SLinus Torvalds return -ENOENT; 2553b18825a7SDavid Howells BUG_ON(!inode); 25541da177e4SLinus Torvalds 25551da177e4SLinus Torvalds BUG_ON(victim->d_parent->d_inode != dir); 25564fa6b5ecSJeff Layton audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); 25571da177e4SLinus Torvalds 2558f419a2e3SAl Viro error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 25591da177e4SLinus Torvalds if (error) 25601da177e4SLinus Torvalds return error; 25611da177e4SLinus Torvalds if (IS_APPEND(dir)) 25621da177e4SLinus Torvalds return -EPERM; 2563b18825a7SDavid Howells 2564b18825a7SDavid Howells if (check_sticky(dir, inode) || IS_APPEND(inode) || 2565b18825a7SDavid Howells IS_IMMUTABLE(inode) || IS_SWAPFILE(inode)) 25661da177e4SLinus Torvalds return -EPERM; 25671da177e4SLinus Torvalds if (isdir) { 256844b1d530SMiklos Szeredi if (!d_is_dir(victim)) 25691da177e4SLinus Torvalds return -ENOTDIR; 25701da177e4SLinus Torvalds if (IS_ROOT(victim)) 25711da177e4SLinus Torvalds return -EBUSY; 257244b1d530SMiklos Szeredi } else if (d_is_dir(victim)) 25731da177e4SLinus Torvalds return -EISDIR; 25741da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 25751da177e4SLinus Torvalds return -ENOENT; 25761da177e4SLinus Torvalds if (victim->d_flags & DCACHE_NFSFS_RENAMED) 25771da177e4SLinus Torvalds return -EBUSY; 25781da177e4SLinus Torvalds return 0; 25791da177e4SLinus Torvalds } 25801da177e4SLinus Torvalds 25811da177e4SLinus Torvalds /* Check whether we can create an object with dentry child in directory 25821da177e4SLinus Torvalds * dir. 25831da177e4SLinus Torvalds * 1. We can't do it if child already exists (open has special treatment for 25841da177e4SLinus Torvalds * this case, but since we are inlined it's OK) 25851da177e4SLinus Torvalds * 2. We can't do it if dir is read-only (done in permission()) 25861da177e4SLinus Torvalds * 3. We should have write and exec permissions on dir 25871da177e4SLinus Torvalds * 4. We can't do it if dir is immutable (done in permission()) 25881da177e4SLinus Torvalds */ 2589a95164d9SMiklos Szeredi static inline int may_create(struct inode *dir, struct dentry *child) 25901da177e4SLinus Torvalds { 259114e972b4SJeff Layton audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); 25921da177e4SLinus Torvalds if (child->d_inode) 25931da177e4SLinus Torvalds return -EEXIST; 25941da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 25951da177e4SLinus Torvalds return -ENOENT; 2596f419a2e3SAl Viro return inode_permission(dir, MAY_WRITE | MAY_EXEC); 25971da177e4SLinus Torvalds } 25981da177e4SLinus Torvalds 25991da177e4SLinus Torvalds /* 26001da177e4SLinus Torvalds * p1 and p2 should be directories on the same fs. 26011da177e4SLinus Torvalds */ 26021da177e4SLinus Torvalds struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 26031da177e4SLinus Torvalds { 26041da177e4SLinus Torvalds struct dentry *p; 26051da177e4SLinus Torvalds 26061da177e4SLinus Torvalds if (p1 == p2) { 2607f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 26081da177e4SLinus Torvalds return NULL; 26091da177e4SLinus Torvalds } 26101da177e4SLinus Torvalds 2611a11f3a05SArjan van de Ven mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 26121da177e4SLinus Torvalds 2613e2761a11SOGAWA Hirofumi p = d_ancestor(p2, p1); 2614e2761a11SOGAWA Hirofumi if (p) { 2615f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT); 2616f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD); 26171da177e4SLinus Torvalds return p; 26181da177e4SLinus Torvalds } 26191da177e4SLinus Torvalds 2620e2761a11SOGAWA Hirofumi p = d_ancestor(p1, p2); 2621e2761a11SOGAWA Hirofumi if (p) { 2622f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 2623f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); 26241da177e4SLinus Torvalds return p; 26251da177e4SLinus Torvalds } 26261da177e4SLinus Torvalds 2627f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 2628d1b72cc6SMiklos Szeredi mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT2); 26291da177e4SLinus Torvalds return NULL; 26301da177e4SLinus Torvalds } 26314d359507SAl Viro EXPORT_SYMBOL(lock_rename); 26321da177e4SLinus Torvalds 26331da177e4SLinus Torvalds void unlock_rename(struct dentry *p1, struct dentry *p2) 26341da177e4SLinus Torvalds { 26351b1dcc1bSJes Sorensen mutex_unlock(&p1->d_inode->i_mutex); 26361da177e4SLinus Torvalds if (p1 != p2) { 26371b1dcc1bSJes Sorensen mutex_unlock(&p2->d_inode->i_mutex); 2638a11f3a05SArjan van de Ven mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 26391da177e4SLinus Torvalds } 26401da177e4SLinus Torvalds } 26414d359507SAl Viro EXPORT_SYMBOL(unlock_rename); 26421da177e4SLinus Torvalds 26434acdaf27SAl Viro int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, 2644312b63fbSAl Viro bool want_excl) 26451da177e4SLinus Torvalds { 2646a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 26471da177e4SLinus Torvalds if (error) 26481da177e4SLinus Torvalds return error; 26491da177e4SLinus Torvalds 2650acfa4380SAl Viro if (!dir->i_op->create) 26511da177e4SLinus Torvalds return -EACCES; /* shouldn't it be ENOSYS? */ 26521da177e4SLinus Torvalds mode &= S_IALLUGO; 26531da177e4SLinus Torvalds mode |= S_IFREG; 26541da177e4SLinus Torvalds error = security_inode_create(dir, dentry, mode); 26551da177e4SLinus Torvalds if (error) 26561da177e4SLinus Torvalds return error; 2657312b63fbSAl Viro error = dir->i_op->create(dir, dentry, mode, want_excl); 2658a74574aaSStephen Smalley if (!error) 2659f38aa942SAmy Griffis fsnotify_create(dir, dentry); 26601da177e4SLinus Torvalds return error; 26611da177e4SLinus Torvalds } 26624d359507SAl Viro EXPORT_SYMBOL(vfs_create); 26631da177e4SLinus Torvalds 266473d049a4SAl Viro static int may_open(struct path *path, int acc_mode, int flag) 26651da177e4SLinus Torvalds { 26663fb64190SChristoph Hellwig struct dentry *dentry = path->dentry; 26671da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 26681da177e4SLinus Torvalds int error; 26691da177e4SLinus Torvalds 2670bcda7652SAl Viro /* O_PATH? */ 2671bcda7652SAl Viro if (!acc_mode) 2672bcda7652SAl Viro return 0; 2673bcda7652SAl Viro 26741da177e4SLinus Torvalds if (!inode) 26751da177e4SLinus Torvalds return -ENOENT; 26761da177e4SLinus Torvalds 2677c8fe8f30SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 2678c8fe8f30SChristoph Hellwig case S_IFLNK: 26791da177e4SLinus Torvalds return -ELOOP; 2680c8fe8f30SChristoph Hellwig case S_IFDIR: 2681c8fe8f30SChristoph Hellwig if (acc_mode & MAY_WRITE) 26821da177e4SLinus Torvalds return -EISDIR; 2683c8fe8f30SChristoph Hellwig break; 2684c8fe8f30SChristoph Hellwig case S_IFBLK: 2685c8fe8f30SChristoph Hellwig case S_IFCHR: 26863fb64190SChristoph Hellwig if (path->mnt->mnt_flags & MNT_NODEV) 26871da177e4SLinus Torvalds return -EACCES; 2688c8fe8f30SChristoph Hellwig /*FALLTHRU*/ 2689c8fe8f30SChristoph Hellwig case S_IFIFO: 2690c8fe8f30SChristoph Hellwig case S_IFSOCK: 26911da177e4SLinus Torvalds flag &= ~O_TRUNC; 2692c8fe8f30SChristoph Hellwig break; 26934a3fd211SDave Hansen } 2694b41572e9SDave Hansen 26953fb64190SChristoph Hellwig error = inode_permission(inode, acc_mode); 2696b41572e9SDave Hansen if (error) 2697b41572e9SDave Hansen return error; 26986146f0d5SMimi Zohar 26991da177e4SLinus Torvalds /* 27001da177e4SLinus Torvalds * An append-only file must be opened in append mode for writing. 27011da177e4SLinus Torvalds */ 27021da177e4SLinus Torvalds if (IS_APPEND(inode)) { 27038737c930SAl Viro if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) 27047715b521SAl Viro return -EPERM; 27051da177e4SLinus Torvalds if (flag & O_TRUNC) 27067715b521SAl Viro return -EPERM; 27071da177e4SLinus Torvalds } 27081da177e4SLinus Torvalds 27091da177e4SLinus Torvalds /* O_NOATIME can only be set by the owner or superuser */ 27102e149670SSerge E. Hallyn if (flag & O_NOATIME && !inode_owner_or_capable(inode)) 27117715b521SAl Viro return -EPERM; 27121da177e4SLinus Torvalds 2713f3c7691eSJ. Bruce Fields return 0; 27147715b521SAl Viro } 27157715b521SAl Viro 2716e1181ee6SJeff Layton static int handle_truncate(struct file *filp) 27177715b521SAl Viro { 2718e1181ee6SJeff Layton struct path *path = &filp->f_path; 27197715b521SAl Viro struct inode *inode = path->dentry->d_inode; 27207715b521SAl Viro int error = get_write_access(inode); 27211da177e4SLinus Torvalds if (error) 27227715b521SAl Viro return error; 27231da177e4SLinus Torvalds /* 27241da177e4SLinus Torvalds * Refuse to truncate files with mandatory locks held on them. 27251da177e4SLinus Torvalds */ 2726d7a06983SJeff Layton error = locks_verify_locked(filp); 2727be6d3e56SKentaro Takeda if (!error) 2728ea0d3ab2STetsuo Handa error = security_path_truncate(path); 27291da177e4SLinus Torvalds if (!error) { 27307715b521SAl Viro error = do_truncate(path->dentry, 0, 2731d139d7ffSMiklos Szeredi ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, 2732e1181ee6SJeff Layton filp); 27331da177e4SLinus Torvalds } 27341da177e4SLinus Torvalds put_write_access(inode); 2735acd0c935SMimi Zohar return error; 27361da177e4SLinus Torvalds } 27371da177e4SLinus Torvalds 2738d57999e1SDave Hansen static inline int open_to_namei_flags(int flag) 2739d57999e1SDave Hansen { 27408a5e929dSAl Viro if ((flag & O_ACCMODE) == 3) 27418a5e929dSAl Viro flag--; 2742d57999e1SDave Hansen return flag; 2743d57999e1SDave Hansen } 2744d57999e1SDave Hansen 2745d18e9008SMiklos Szeredi static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode) 2746d18e9008SMiklos Szeredi { 2747d18e9008SMiklos Szeredi int error = security_path_mknod(dir, dentry, mode, 0); 2748d18e9008SMiklos Szeredi if (error) 2749d18e9008SMiklos Szeredi return error; 2750d18e9008SMiklos Szeredi 2751d18e9008SMiklos Szeredi error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC); 2752d18e9008SMiklos Szeredi if (error) 2753d18e9008SMiklos Szeredi return error; 2754d18e9008SMiklos Szeredi 2755d18e9008SMiklos Szeredi return security_inode_create(dir->dentry->d_inode, dentry, mode); 2756d18e9008SMiklos Szeredi } 2757d18e9008SMiklos Szeredi 27581acf0af9SDavid Howells /* 27591acf0af9SDavid Howells * Attempt to atomically look up, create and open a file from a negative 27601acf0af9SDavid Howells * dentry. 27611acf0af9SDavid Howells * 27621acf0af9SDavid Howells * Returns 0 if successful. The file will have been created and attached to 27631acf0af9SDavid Howells * @file by the filesystem calling finish_open(). 27641acf0af9SDavid Howells * 27651acf0af9SDavid Howells * Returns 1 if the file was looked up only or didn't need creating. The 27661acf0af9SDavid Howells * caller will need to perform the open themselves. @path will have been 27671acf0af9SDavid Howells * updated to point to the new dentry. This may be negative. 27681acf0af9SDavid Howells * 27691acf0af9SDavid Howells * Returns an error code otherwise. 27701acf0af9SDavid Howells */ 27712675a4ebSAl Viro static int atomic_open(struct nameidata *nd, struct dentry *dentry, 277230d90494SAl Viro struct path *path, struct file *file, 2773015c3bbcSMiklos Szeredi const struct open_flags *op, 277464894cf8SAl Viro bool got_write, bool need_lookup, 277547237687SAl Viro int *opened) 2776d18e9008SMiklos Szeredi { 2777d18e9008SMiklos Szeredi struct inode *dir = nd->path.dentry->d_inode; 2778d18e9008SMiklos Szeredi unsigned open_flag = open_to_namei_flags(op->open_flag); 2779d18e9008SMiklos Szeredi umode_t mode; 2780d18e9008SMiklos Szeredi int error; 2781d18e9008SMiklos Szeredi int acc_mode; 2782d18e9008SMiklos Szeredi int create_error = 0; 2783d18e9008SMiklos Szeredi struct dentry *const DENTRY_NOT_SET = (void *) -1UL; 2784116cc022SMiklos Szeredi bool excl; 2785d18e9008SMiklos Szeredi 2786d18e9008SMiklos Szeredi BUG_ON(dentry->d_inode); 2787d18e9008SMiklos Szeredi 2788d18e9008SMiklos Szeredi /* Don't create child dentry for a dead directory. */ 2789d18e9008SMiklos Szeredi if (unlikely(IS_DEADDIR(dir))) { 27902675a4ebSAl Viro error = -ENOENT; 2791d18e9008SMiklos Szeredi goto out; 2792d18e9008SMiklos Szeredi } 2793d18e9008SMiklos Szeredi 279462b259d8SMiklos Szeredi mode = op->mode; 2795d18e9008SMiklos Szeredi if ((open_flag & O_CREAT) && !IS_POSIXACL(dir)) 2796d18e9008SMiklos Szeredi mode &= ~current_umask(); 2797d18e9008SMiklos Szeredi 2798116cc022SMiklos Szeredi excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT); 2799116cc022SMiklos Szeredi if (excl) 2800d18e9008SMiklos Szeredi open_flag &= ~O_TRUNC; 2801d18e9008SMiklos Szeredi 2802d18e9008SMiklos Szeredi /* 2803d18e9008SMiklos Szeredi * Checking write permission is tricky, bacuse we don't know if we are 2804d18e9008SMiklos Szeredi * going to actually need it: O_CREAT opens should work as long as the 2805d18e9008SMiklos Szeredi * file exists. But checking existence breaks atomicity. The trick is 2806d18e9008SMiklos Szeredi * to check access and if not granted clear O_CREAT from the flags. 2807d18e9008SMiklos Szeredi * 2808d18e9008SMiklos Szeredi * Another problem is returing the "right" error value (e.g. for an 2809d18e9008SMiklos Szeredi * O_EXCL open we want to return EEXIST not EROFS). 2810d18e9008SMiklos Szeredi */ 281164894cf8SAl Viro if (((open_flag & (O_CREAT | O_TRUNC)) || 281264894cf8SAl Viro (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) { 281364894cf8SAl Viro if (!(open_flag & O_CREAT)) { 2814d18e9008SMiklos Szeredi /* 2815d18e9008SMiklos Szeredi * No O_CREATE -> atomicity not a requirement -> fall 2816d18e9008SMiklos Szeredi * back to lookup + open 2817d18e9008SMiklos Szeredi */ 2818d18e9008SMiklos Szeredi goto no_open; 2819d18e9008SMiklos Szeredi } else if (open_flag & (O_EXCL | O_TRUNC)) { 2820d18e9008SMiklos Szeredi /* Fall back and fail with the right error */ 282164894cf8SAl Viro create_error = -EROFS; 2822d18e9008SMiklos Szeredi goto no_open; 2823d18e9008SMiklos Szeredi } else { 2824d18e9008SMiklos Szeredi /* No side effects, safe to clear O_CREAT */ 282564894cf8SAl Viro create_error = -EROFS; 2826d18e9008SMiklos Szeredi open_flag &= ~O_CREAT; 2827d18e9008SMiklos Szeredi } 2828d18e9008SMiklos Szeredi } 2829d18e9008SMiklos Szeredi 2830d18e9008SMiklos Szeredi if (open_flag & O_CREAT) { 283138227f78SMiklos Szeredi error = may_o_create(&nd->path, dentry, mode); 2832d18e9008SMiklos Szeredi if (error) { 2833d18e9008SMiklos Szeredi create_error = error; 2834d18e9008SMiklos Szeredi if (open_flag & O_EXCL) 2835d18e9008SMiklos Szeredi goto no_open; 2836d18e9008SMiklos Szeredi open_flag &= ~O_CREAT; 2837d18e9008SMiklos Szeredi } 2838d18e9008SMiklos Szeredi } 2839d18e9008SMiklos Szeredi 2840d18e9008SMiklos Szeredi if (nd->flags & LOOKUP_DIRECTORY) 2841d18e9008SMiklos Szeredi open_flag |= O_DIRECTORY; 2842d18e9008SMiklos Szeredi 284330d90494SAl Viro file->f_path.dentry = DENTRY_NOT_SET; 284430d90494SAl Viro file->f_path.mnt = nd->path.mnt; 284530d90494SAl Viro error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode, 284647237687SAl Viro opened); 2847d9585277SAl Viro if (error < 0) { 2848d9585277SAl Viro if (create_error && error == -ENOENT) 2849d9585277SAl Viro error = create_error; 2850d18e9008SMiklos Szeredi goto out; 2851d18e9008SMiklos Szeredi } 2852d18e9008SMiklos Szeredi 2853d9585277SAl Viro if (error) { /* returned 1, that is */ 285430d90494SAl Viro if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { 28552675a4ebSAl Viro error = -EIO; 2856d18e9008SMiklos Szeredi goto out; 2857d18e9008SMiklos Szeredi } 285830d90494SAl Viro if (file->f_path.dentry) { 2859d18e9008SMiklos Szeredi dput(dentry); 286030d90494SAl Viro dentry = file->f_path.dentry; 2861d18e9008SMiklos Szeredi } 286203da633aSAl Viro if (*opened & FILE_CREATED) 286303da633aSAl Viro fsnotify_create(dir, dentry); 286403da633aSAl Viro if (!dentry->d_inode) { 286503da633aSAl Viro WARN_ON(*opened & FILE_CREATED); 286603da633aSAl Viro if (create_error) { 286762b2ce96SSage Weil error = create_error; 286862b2ce96SSage Weil goto out; 286962b2ce96SSage Weil } 287003da633aSAl Viro } else { 287103da633aSAl Viro if (excl && !(*opened & FILE_CREATED)) { 287203da633aSAl Viro error = -EEXIST; 287303da633aSAl Viro goto out; 287403da633aSAl Viro } 287503da633aSAl Viro } 2876d18e9008SMiklos Szeredi goto looked_up; 2877d18e9008SMiklos Szeredi } 2878d18e9008SMiklos Szeredi 2879d18e9008SMiklos Szeredi /* 2880d18e9008SMiklos Szeredi * We didn't have the inode before the open, so check open permission 2881d18e9008SMiklos Szeredi * here. 2882d18e9008SMiklos Szeredi */ 288303da633aSAl Viro acc_mode = op->acc_mode; 288403da633aSAl Viro if (*opened & FILE_CREATED) { 288503da633aSAl Viro WARN_ON(!(open_flag & O_CREAT)); 288603da633aSAl Viro fsnotify_create(dir, dentry); 288703da633aSAl Viro acc_mode = MAY_OPEN; 288803da633aSAl Viro } 28892675a4ebSAl Viro error = may_open(&file->f_path, acc_mode, open_flag); 28902675a4ebSAl Viro if (error) 28912675a4ebSAl Viro fput(file); 2892d18e9008SMiklos Szeredi 2893d18e9008SMiklos Szeredi out: 2894d18e9008SMiklos Szeredi dput(dentry); 28952675a4ebSAl Viro return error; 2896d18e9008SMiklos Szeredi 2897d18e9008SMiklos Szeredi no_open: 2898d18e9008SMiklos Szeredi if (need_lookup) { 289972bd866aSAl Viro dentry = lookup_real(dir, dentry, nd->flags); 2900d18e9008SMiklos Szeredi if (IS_ERR(dentry)) 29012675a4ebSAl Viro return PTR_ERR(dentry); 2902d18e9008SMiklos Szeredi 2903d18e9008SMiklos Szeredi if (create_error) { 2904d18e9008SMiklos Szeredi int open_flag = op->open_flag; 2905d18e9008SMiklos Szeredi 29062675a4ebSAl Viro error = create_error; 2907d18e9008SMiklos Szeredi if ((open_flag & O_EXCL)) { 2908d18e9008SMiklos Szeredi if (!dentry->d_inode) 2909d18e9008SMiklos Szeredi goto out; 2910d18e9008SMiklos Szeredi } else if (!dentry->d_inode) { 2911d18e9008SMiklos Szeredi goto out; 2912d18e9008SMiklos Szeredi } else if ((open_flag & O_TRUNC) && 2913e36cb0b8SDavid Howells d_is_reg(dentry)) { 2914d18e9008SMiklos Szeredi goto out; 2915d18e9008SMiklos Szeredi } 2916d18e9008SMiklos Szeredi /* will fail later, go on to get the right error */ 2917d18e9008SMiklos Szeredi } 2918d18e9008SMiklos Szeredi } 2919d18e9008SMiklos Szeredi looked_up: 2920d18e9008SMiklos Szeredi path->dentry = dentry; 2921d18e9008SMiklos Szeredi path->mnt = nd->path.mnt; 29222675a4ebSAl Viro return 1; 2923d18e9008SMiklos Szeredi } 2924d18e9008SMiklos Szeredi 292531e6b01fSNick Piggin /* 29261acf0af9SDavid Howells * Look up and maybe create and open the last component. 2927d58ffd35SMiklos Szeredi * 2928d58ffd35SMiklos Szeredi * Must be called with i_mutex held on parent. 2929d58ffd35SMiklos Szeredi * 29301acf0af9SDavid Howells * Returns 0 if the file was successfully atomically created (if necessary) and 29311acf0af9SDavid Howells * opened. In this case the file will be returned attached to @file. 29321acf0af9SDavid Howells * 29331acf0af9SDavid Howells * Returns 1 if the file was not completely opened at this time, though lookups 29341acf0af9SDavid Howells * and creations will have been performed and the dentry returned in @path will 29351acf0af9SDavid Howells * be positive upon return if O_CREAT was specified. If O_CREAT wasn't 29361acf0af9SDavid Howells * specified then a negative dentry may be returned. 29371acf0af9SDavid Howells * 29381acf0af9SDavid Howells * An error code is returned otherwise. 29391acf0af9SDavid Howells * 29401acf0af9SDavid Howells * FILE_CREATE will be set in @*opened if the dentry was created and will be 29411acf0af9SDavid Howells * cleared otherwise prior to returning. 2942d58ffd35SMiklos Szeredi */ 29432675a4ebSAl Viro static int lookup_open(struct nameidata *nd, struct path *path, 294430d90494SAl Viro struct file *file, 2945d58ffd35SMiklos Szeredi const struct open_flags *op, 294664894cf8SAl Viro bool got_write, int *opened) 2947d58ffd35SMiklos Szeredi { 2948d58ffd35SMiklos Szeredi struct dentry *dir = nd->path.dentry; 294954ef4872SMiklos Szeredi struct inode *dir_inode = dir->d_inode; 2950d58ffd35SMiklos Szeredi struct dentry *dentry; 2951d58ffd35SMiklos Szeredi int error; 295254ef4872SMiklos Szeredi bool need_lookup; 2953d58ffd35SMiklos Szeredi 295447237687SAl Viro *opened &= ~FILE_CREATED; 2955201f956eSAl Viro dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup); 2956d58ffd35SMiklos Szeredi if (IS_ERR(dentry)) 29572675a4ebSAl Viro return PTR_ERR(dentry); 2958d58ffd35SMiklos Szeredi 2959d18e9008SMiklos Szeredi /* Cached positive dentry: will open in f_op->open */ 2960d18e9008SMiklos Szeredi if (!need_lookup && dentry->d_inode) 2961d18e9008SMiklos Szeredi goto out_no_open; 2962d18e9008SMiklos Szeredi 2963d18e9008SMiklos Szeredi if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) { 296464894cf8SAl Viro return atomic_open(nd, dentry, path, file, op, got_write, 296547237687SAl Viro need_lookup, opened); 2966d18e9008SMiklos Szeredi } 2967d18e9008SMiklos Szeredi 296854ef4872SMiklos Szeredi if (need_lookup) { 296954ef4872SMiklos Szeredi BUG_ON(dentry->d_inode); 297054ef4872SMiklos Szeredi 297172bd866aSAl Viro dentry = lookup_real(dir_inode, dentry, nd->flags); 297254ef4872SMiklos Szeredi if (IS_ERR(dentry)) 29732675a4ebSAl Viro return PTR_ERR(dentry); 297454ef4872SMiklos Szeredi } 297554ef4872SMiklos Szeredi 2976d58ffd35SMiklos Szeredi /* Negative dentry, just create the file */ 2977d58ffd35SMiklos Szeredi if (!dentry->d_inode && (op->open_flag & O_CREAT)) { 2978d58ffd35SMiklos Szeredi umode_t mode = op->mode; 2979d58ffd35SMiklos Szeredi if (!IS_POSIXACL(dir->d_inode)) 2980d58ffd35SMiklos Szeredi mode &= ~current_umask(); 2981d58ffd35SMiklos Szeredi /* 2982d58ffd35SMiklos Szeredi * This write is needed to ensure that a 2983d58ffd35SMiklos Szeredi * rw->ro transition does not occur between 2984d58ffd35SMiklos Szeredi * the time when the file is created and when 2985d58ffd35SMiklos Szeredi * a permanent write count is taken through 2986015c3bbcSMiklos Szeredi * the 'struct file' in finish_open(). 2987d58ffd35SMiklos Szeredi */ 298864894cf8SAl Viro if (!got_write) { 298964894cf8SAl Viro error = -EROFS; 2990d58ffd35SMiklos Szeredi goto out_dput; 299164894cf8SAl Viro } 299247237687SAl Viro *opened |= FILE_CREATED; 2993d58ffd35SMiklos Szeredi error = security_path_mknod(&nd->path, dentry, mode, 0); 2994d58ffd35SMiklos Szeredi if (error) 2995d58ffd35SMiklos Szeredi goto out_dput; 2996312b63fbSAl Viro error = vfs_create(dir->d_inode, dentry, mode, 2997312b63fbSAl Viro nd->flags & LOOKUP_EXCL); 2998d58ffd35SMiklos Szeredi if (error) 2999d58ffd35SMiklos Szeredi goto out_dput; 3000d58ffd35SMiklos Szeredi } 3001d18e9008SMiklos Szeredi out_no_open: 3002d58ffd35SMiklos Szeredi path->dentry = dentry; 3003d58ffd35SMiklos Szeredi path->mnt = nd->path.mnt; 30042675a4ebSAl Viro return 1; 3005d58ffd35SMiklos Szeredi 3006d58ffd35SMiklos Szeredi out_dput: 3007d58ffd35SMiklos Szeredi dput(dentry); 30082675a4ebSAl Viro return error; 3009d58ffd35SMiklos Szeredi } 3010d58ffd35SMiklos Szeredi 3011d58ffd35SMiklos Szeredi /* 3012fe2d35ffSAl Viro * Handle the last step of open() 301331e6b01fSNick Piggin */ 3014896475d5SAl Viro static int do_last(struct nameidata *nd, 301530d90494SAl Viro struct file *file, const struct open_flags *op, 301676ae2a5aSAl Viro int *opened) 3017fb1cc555SAl Viro { 3018a1e28038SAl Viro struct dentry *dir = nd->path.dentry; 3019ca344a89SAl Viro int open_flag = op->open_flag; 302077d660a8SMiklos Szeredi bool will_truncate = (open_flag & O_TRUNC) != 0; 302164894cf8SAl Viro bool got_write = false; 3022bcda7652SAl Viro int acc_mode = op->acc_mode; 3023254cf582SAl Viro unsigned seq; 3024a1eb3315SMiklos Szeredi struct inode *inode; 302516b1c1cdSMiklos Szeredi struct path save_parent = { .dentry = NULL, .mnt = NULL }; 3026896475d5SAl Viro struct path path; 302716b1c1cdSMiklos Szeredi bool retried = false; 302816c2cd71SAl Viro int error; 3029fb1cc555SAl Viro 3030c3e380b0SAl Viro nd->flags &= ~LOOKUP_PARENT; 3031c3e380b0SAl Viro nd->flags |= op->intent; 3032c3e380b0SAl Viro 3033bc77daa7SAl Viro if (nd->last_type != LAST_NORM) { 3034fe2d35ffSAl Viro error = handle_dots(nd, nd->last_type); 3035deb106c6SAl Viro if (unlikely(error)) 30362675a4ebSAl Viro return error; 3037e83db167SMiklos Szeredi goto finish_open; 30381f36f774SAl Viro } 3039a2c36b45SAl Viro 3040ca344a89SAl Viro if (!(open_flag & O_CREAT)) { 3041fe2d35ffSAl Viro if (nd->last.name[nd->last.len]) 3042fe2d35ffSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 3043fe2d35ffSAl Viro /* we _can_ be in RCU mode here */ 3044254cf582SAl Viro error = lookup_fast(nd, &path, &inode, &seq); 304571574865SMiklos Szeredi if (likely(!error)) 304671574865SMiklos Szeredi goto finish_lookup; 304771574865SMiklos Szeredi 3048ce57dfc1SAl Viro if (error < 0) 3049deb106c6SAl Viro return error; 3050a1eb3315SMiklos Szeredi 305137d7fffcSMiklos Szeredi BUG_ON(nd->inode != dir->d_inode); 3052b6183df7SMiklos Szeredi } else { 3053fe2d35ffSAl Viro /* create side of things */ 3054a3fbbde7SAl Viro /* 3055b6183df7SMiklos Szeredi * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED 3056b6183df7SMiklos Szeredi * has been cleared when we got to the last component we are 3057b6183df7SMiklos Szeredi * about to look up 3058a3fbbde7SAl Viro */ 30599f1fafeeSAl Viro error = complete_walk(nd); 3060e8bb73dfSAl Viro if (error) 30612675a4ebSAl Viro return error; 3062fe2d35ffSAl Viro 306376ae2a5aSAl Viro audit_inode(nd->name, dir, LOOKUP_PARENT); 30641f36f774SAl Viro /* trailing slashes? */ 3065deb106c6SAl Viro if (unlikely(nd->last.name[nd->last.len])) 3066deb106c6SAl Viro return -EISDIR; 3067b6183df7SMiklos Szeredi } 30681f36f774SAl Viro 306916b1c1cdSMiklos Szeredi retry_lookup: 307064894cf8SAl Viro if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { 307164894cf8SAl Viro error = mnt_want_write(nd->path.mnt); 307264894cf8SAl Viro if (!error) 307364894cf8SAl Viro got_write = true; 307464894cf8SAl Viro /* 307564894cf8SAl Viro * do _not_ fail yet - we might not need that or fail with 307664894cf8SAl Viro * a different error; let lookup_open() decide; we'll be 307764894cf8SAl Viro * dropping this one anyway. 307864894cf8SAl Viro */ 307964894cf8SAl Viro } 3080a1e28038SAl Viro mutex_lock(&dir->d_inode->i_mutex); 3081896475d5SAl Viro error = lookup_open(nd, &path, file, op, got_write, opened); 3082fb1cc555SAl Viro mutex_unlock(&dir->d_inode->i_mutex); 3083fb1cc555SAl Viro 30842675a4ebSAl Viro if (error <= 0) { 30852675a4ebSAl Viro if (error) 3086d58ffd35SMiklos Szeredi goto out; 30876c0d46c4SAl Viro 308847237687SAl Viro if ((*opened & FILE_CREATED) || 3089496ad9aaSAl Viro !S_ISREG(file_inode(file)->i_mode)) 309077d660a8SMiklos Szeredi will_truncate = false; 3091d18e9008SMiklos Szeredi 309276ae2a5aSAl Viro audit_inode(nd->name, file->f_path.dentry, 0); 3093d18e9008SMiklos Szeredi goto opened; 3094d18e9008SMiklos Szeredi } 3095d18e9008SMiklos Szeredi 309647237687SAl Viro if (*opened & FILE_CREATED) { 30979b44f1b3SAl Viro /* Don't check for write permission, don't truncate */ 3098ca344a89SAl Viro open_flag &= ~O_TRUNC; 309977d660a8SMiklos Szeredi will_truncate = false; 3100bcda7652SAl Viro acc_mode = MAY_OPEN; 3101896475d5SAl Viro path_to_nameidata(&path, nd); 3102e83db167SMiklos Szeredi goto finish_open_created; 3103fb1cc555SAl Viro } 3104fb1cc555SAl Viro 3105fb1cc555SAl Viro /* 31063134f37eSJeff Layton * create/update audit record if it already exists. 3107fb1cc555SAl Viro */ 3108896475d5SAl Viro if (d_is_positive(path.dentry)) 310976ae2a5aSAl Viro audit_inode(nd->name, path.dentry, 0); 3110fb1cc555SAl Viro 3111d18e9008SMiklos Szeredi /* 3112d18e9008SMiklos Szeredi * If atomic_open() acquired write access it is dropped now due to 3113d18e9008SMiklos Szeredi * possible mount and symlink following (this might be optimized away if 3114d18e9008SMiklos Szeredi * necessary...) 3115d18e9008SMiklos Szeredi */ 311664894cf8SAl Viro if (got_write) { 3117d18e9008SMiklos Szeredi mnt_drop_write(nd->path.mnt); 311864894cf8SAl Viro got_write = false; 3119d18e9008SMiklos Szeredi } 3120d18e9008SMiklos Szeredi 3121deb106c6SAl Viro if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) { 3122deb106c6SAl Viro path_to_nameidata(&path, nd); 3123deb106c6SAl Viro return -EEXIST; 3124deb106c6SAl Viro } 3125fb1cc555SAl Viro 3126756daf26SNeilBrown error = follow_managed(&path, nd); 3127deb106c6SAl Viro if (unlikely(error < 0)) 3128deb106c6SAl Viro return error; 3129a3fbbde7SAl Viro 3130decf3400SMiklos Szeredi BUG_ON(nd->flags & LOOKUP_RCU); 313163afdfc7SDavid Howells inode = d_backing_inode(path.dentry); 3132254cf582SAl Viro seq = 0; /* out of RCU mode, so the value doesn't matter */ 3133deb106c6SAl Viro if (unlikely(d_is_negative(path.dentry))) { 3134896475d5SAl Viro path_to_nameidata(&path, nd); 3135deb106c6SAl Viro return -ENOENT; 313654c33e7fSMiklos Szeredi } 3137766c4cbfSAl Viro finish_lookup: 3138191d7f73SAl Viro if (nd->depth) 3139191d7f73SAl Viro put_link(nd); 3140181548c0SAl Viro error = should_follow_link(nd, &path, nd->flags & LOOKUP_FOLLOW, 3141181548c0SAl Viro inode, seq); 3142deb106c6SAl Viro if (unlikely(error)) 3143d63ff28fSAl Viro return error; 3144fb1cc555SAl Viro 3145896475d5SAl Viro if (unlikely(d_is_symlink(path.dentry)) && !(open_flag & O_PATH)) { 3146896475d5SAl Viro path_to_nameidata(&path, nd); 3147deb106c6SAl Viro return -ELOOP; 3148a5cfe2d5SAl Viro } 3149a5cfe2d5SAl Viro 3150896475d5SAl Viro if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path.mnt) { 3151896475d5SAl Viro path_to_nameidata(&path, nd); 315216b1c1cdSMiklos Szeredi } else { 315316b1c1cdSMiklos Szeredi save_parent.dentry = nd->path.dentry; 3154896475d5SAl Viro save_parent.mnt = mntget(path.mnt); 3155896475d5SAl Viro nd->path.dentry = path.dentry; 315616b1c1cdSMiklos Szeredi 315716b1c1cdSMiklos Szeredi } 3158decf3400SMiklos Szeredi nd->inode = inode; 3159254cf582SAl Viro nd->seq = seq; 3160a3fbbde7SAl Viro /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ 3161bc77daa7SAl Viro finish_open: 3162a3fbbde7SAl Viro error = complete_walk(nd); 316316b1c1cdSMiklos Szeredi if (error) { 316416b1c1cdSMiklos Szeredi path_put(&save_parent); 31652675a4ebSAl Viro return error; 316616b1c1cdSMiklos Szeredi } 316776ae2a5aSAl Viro audit_inode(nd->name, nd->path.dentry, 0); 3168fb1cc555SAl Viro error = -EISDIR; 316944b1d530SMiklos Szeredi if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry)) 31702675a4ebSAl Viro goto out; 3171af2f5542SMiklos Szeredi error = -ENOTDIR; 317244b1d530SMiklos Szeredi if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry)) 31732675a4ebSAl Viro goto out; 31744bbcbd3bSDavid Howells if (!d_is_reg(nd->path.dentry)) 317577d660a8SMiklos Szeredi will_truncate = false; 31766c0d46c4SAl Viro 31770f9d1a10SAl Viro if (will_truncate) { 31780f9d1a10SAl Viro error = mnt_want_write(nd->path.mnt); 31790f9d1a10SAl Viro if (error) 31802675a4ebSAl Viro goto out; 318164894cf8SAl Viro got_write = true; 31820f9d1a10SAl Viro } 3183e83db167SMiklos Szeredi finish_open_created: 3184bcda7652SAl Viro error = may_open(&nd->path, acc_mode, open_flag); 3185ca344a89SAl Viro if (error) 31862675a4ebSAl Viro goto out; 31874aa7c634SMiklos Szeredi 31884aa7c634SMiklos Szeredi BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */ 31894aa7c634SMiklos Szeredi error = vfs_open(&nd->path, file, current_cred()); 31904aa7c634SMiklos Szeredi if (!error) { 31914aa7c634SMiklos Szeredi *opened |= FILE_OPENED; 31924aa7c634SMiklos Szeredi } else { 319330d90494SAl Viro if (error == -EOPENSTALE) 3194f60dc3dbSMiklos Szeredi goto stale_open; 3195015c3bbcSMiklos Szeredi goto out; 3196f60dc3dbSMiklos Szeredi } 3197a8277b9bSMiklos Szeredi opened: 31982675a4ebSAl Viro error = open_check_o_direct(file); 3199015c3bbcSMiklos Szeredi if (error) 3200015c3bbcSMiklos Szeredi goto exit_fput; 32013034a146SDmitry Kasatkin error = ima_file_check(file, op->acc_mode, *opened); 3202aa4caadbSMiklos Szeredi if (error) 3203aa4caadbSMiklos Szeredi goto exit_fput; 3204aa4caadbSMiklos Szeredi 32050f9d1a10SAl Viro if (will_truncate) { 32062675a4ebSAl Viro error = handle_truncate(file); 3207aa4caadbSMiklos Szeredi if (error) 3208aa4caadbSMiklos Szeredi goto exit_fput; 32090f9d1a10SAl Viro } 3210ca344a89SAl Viro out: 321164894cf8SAl Viro if (got_write) 32120f9d1a10SAl Viro mnt_drop_write(nd->path.mnt); 321316b1c1cdSMiklos Szeredi path_put(&save_parent); 32142675a4ebSAl Viro return error; 3215fb1cc555SAl Viro 3216015c3bbcSMiklos Szeredi exit_fput: 32172675a4ebSAl Viro fput(file); 32182675a4ebSAl Viro goto out; 3219015c3bbcSMiklos Szeredi 3220f60dc3dbSMiklos Szeredi stale_open: 3221f60dc3dbSMiklos Szeredi /* If no saved parent or already retried then can't retry */ 3222f60dc3dbSMiklos Szeredi if (!save_parent.dentry || retried) 3223f60dc3dbSMiklos Szeredi goto out; 3224f60dc3dbSMiklos Szeredi 3225f60dc3dbSMiklos Szeredi BUG_ON(save_parent.dentry != dir); 3226f60dc3dbSMiklos Szeredi path_put(&nd->path); 3227f60dc3dbSMiklos Szeredi nd->path = save_parent; 3228f60dc3dbSMiklos Szeredi nd->inode = dir->d_inode; 3229f60dc3dbSMiklos Szeredi save_parent.mnt = NULL; 3230f60dc3dbSMiklos Szeredi save_parent.dentry = NULL; 323164894cf8SAl Viro if (got_write) { 3232f60dc3dbSMiklos Szeredi mnt_drop_write(nd->path.mnt); 323364894cf8SAl Viro got_write = false; 3234f60dc3dbSMiklos Szeredi } 3235f60dc3dbSMiklos Szeredi retried = true; 3236f60dc3dbSMiklos Szeredi goto retry_lookup; 3237fb1cc555SAl Viro } 3238fb1cc555SAl Viro 3239c8a53ee5SAl Viro static int do_tmpfile(struct nameidata *nd, unsigned flags, 324060545d0dSAl Viro const struct open_flags *op, 324160545d0dSAl Viro struct file *file, int *opened) 324260545d0dSAl Viro { 324360545d0dSAl Viro static const struct qstr name = QSTR_INIT("/", 1); 3244625b6d10SAl Viro struct dentry *child; 324560545d0dSAl Viro struct inode *dir; 3246625b6d10SAl Viro struct path path; 3247c8a53ee5SAl Viro int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path); 324860545d0dSAl Viro if (unlikely(error)) 324960545d0dSAl Viro return error; 3250625b6d10SAl Viro error = mnt_want_write(path.mnt); 325160545d0dSAl Viro if (unlikely(error)) 325260545d0dSAl Viro goto out; 3253625b6d10SAl Viro dir = path.dentry->d_inode; 325460545d0dSAl Viro /* we want directory to be writable */ 3255625b6d10SAl Viro error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 325660545d0dSAl Viro if (error) 325760545d0dSAl Viro goto out2; 325860545d0dSAl Viro if (!dir->i_op->tmpfile) { 325960545d0dSAl Viro error = -EOPNOTSUPP; 326060545d0dSAl Viro goto out2; 326160545d0dSAl Viro } 3262625b6d10SAl Viro child = d_alloc(path.dentry, &name); 326360545d0dSAl Viro if (unlikely(!child)) { 326460545d0dSAl Viro error = -ENOMEM; 326560545d0dSAl Viro goto out2; 326660545d0dSAl Viro } 3267625b6d10SAl Viro dput(path.dentry); 3268625b6d10SAl Viro path.dentry = child; 3269625b6d10SAl Viro error = dir->i_op->tmpfile(dir, child, op->mode); 327060545d0dSAl Viro if (error) 327160545d0dSAl Viro goto out2; 3272c8a53ee5SAl Viro audit_inode(nd->name, child, 0); 327369a91c23SEric Rannaud /* Don't check for other permissions, the inode was just created */ 3274625b6d10SAl Viro error = may_open(&path, MAY_OPEN, op->open_flag); 327560545d0dSAl Viro if (error) 327660545d0dSAl Viro goto out2; 3277625b6d10SAl Viro file->f_path.mnt = path.mnt; 3278625b6d10SAl Viro error = finish_open(file, child, NULL, opened); 327960545d0dSAl Viro if (error) 328060545d0dSAl Viro goto out2; 328160545d0dSAl Viro error = open_check_o_direct(file); 3282f4e0c30cSAl Viro if (error) { 328360545d0dSAl Viro fput(file); 3284f4e0c30cSAl Viro } else if (!(op->open_flag & O_EXCL)) { 3285f4e0c30cSAl Viro struct inode *inode = file_inode(file); 3286f4e0c30cSAl Viro spin_lock(&inode->i_lock); 3287f4e0c30cSAl Viro inode->i_state |= I_LINKABLE; 3288f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 3289f4e0c30cSAl Viro } 329060545d0dSAl Viro out2: 3291625b6d10SAl Viro mnt_drop_write(path.mnt); 329260545d0dSAl Viro out: 3293625b6d10SAl Viro path_put(&path); 329460545d0dSAl Viro return error; 329560545d0dSAl Viro } 329660545d0dSAl Viro 3297c8a53ee5SAl Viro static struct file *path_openat(struct nameidata *nd, 3298c8a53ee5SAl Viro const struct open_flags *op, unsigned flags) 32991da177e4SLinus Torvalds { 3300368ee9baSAl Viro const char *s; 330130d90494SAl Viro struct file *file; 330247237687SAl Viro int opened = 0; 330313aab428SAl Viro int error; 330431e6b01fSNick Piggin 330530d90494SAl Viro file = get_empty_filp(); 33061afc99beSAl Viro if (IS_ERR(file)) 33071afc99beSAl Viro return file; 330831e6b01fSNick Piggin 330930d90494SAl Viro file->f_flags = op->open_flag; 331031e6b01fSNick Piggin 3311bb458c64SAl Viro if (unlikely(file->f_flags & __O_TMPFILE)) { 3312c8a53ee5SAl Viro error = do_tmpfile(nd, flags, op, file, &opened); 3313f15133dfSAl Viro goto out2; 331460545d0dSAl Viro } 331560545d0dSAl Viro 3316c8a53ee5SAl Viro s = path_init(nd, flags); 3317368ee9baSAl Viro if (IS_ERR(s)) { 3318368ee9baSAl Viro put_filp(file); 3319368ee9baSAl Viro return ERR_CAST(s); 3320368ee9baSAl Viro } 33213bdba28bSAl Viro while (!(error = link_path_walk(s, nd)) && 332276ae2a5aSAl Viro (error = do_last(nd, file, op, &opened)) > 0) { 332373d049a4SAl Viro nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); 33243bdba28bSAl Viro s = trailing_symlink(nd); 33253bdba28bSAl Viro if (IS_ERR(s)) { 33263bdba28bSAl Viro error = PTR_ERR(s); 33272675a4ebSAl Viro break; 3328806b681cSAl Viro } 33293bdba28bSAl Viro } 3330deb106c6SAl Viro terminate_walk(nd); 3331f15133dfSAl Viro out2: 33322675a4ebSAl Viro if (!(opened & FILE_OPENED)) { 33332675a4ebSAl Viro BUG_ON(!error); 333430d90494SAl Viro put_filp(file); 3335015c3bbcSMiklos Szeredi } 33362675a4ebSAl Viro if (unlikely(error)) { 33372675a4ebSAl Viro if (error == -EOPENSTALE) { 33382675a4ebSAl Viro if (flags & LOOKUP_RCU) 33392675a4ebSAl Viro error = -ECHILD; 33402675a4ebSAl Viro else 33412675a4ebSAl Viro error = -ESTALE; 33422675a4ebSAl Viro } 33432675a4ebSAl Viro file = ERR_PTR(error); 33442675a4ebSAl Viro } 33452675a4ebSAl Viro return file; 3346de459215SKirill Korotaev } 33471da177e4SLinus Torvalds 3348669abf4eSJeff Layton struct file *do_filp_open(int dfd, struct filename *pathname, 3349f9652e10SAl Viro const struct open_flags *op) 335013aab428SAl Viro { 33519883d185SAl Viro struct nameidata nd; 3352f9652e10SAl Viro int flags = op->lookup_flags; 335313aab428SAl Viro struct file *filp; 335413aab428SAl Viro 33559883d185SAl Viro set_nameidata(&nd, dfd, pathname); 3356c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_RCU); 335713aab428SAl Viro if (unlikely(filp == ERR_PTR(-ECHILD))) 3358c8a53ee5SAl Viro filp = path_openat(&nd, op, flags); 335913aab428SAl Viro if (unlikely(filp == ERR_PTR(-ESTALE))) 3360c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_REVAL); 33619883d185SAl Viro restore_nameidata(); 336213aab428SAl Viro return filp; 336313aab428SAl Viro } 336413aab428SAl Viro 336573d049a4SAl Viro struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, 3366f9652e10SAl Viro const char *name, const struct open_flags *op) 336773d049a4SAl Viro { 33689883d185SAl Viro struct nameidata nd; 336973d049a4SAl Viro struct file *file; 337051689104SPaul Moore struct filename *filename; 3371f9652e10SAl Viro int flags = op->lookup_flags | LOOKUP_ROOT; 337273d049a4SAl Viro 337373d049a4SAl Viro nd.root.mnt = mnt; 337473d049a4SAl Viro nd.root.dentry = dentry; 337573d049a4SAl Viro 3376b18825a7SDavid Howells if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN) 337773d049a4SAl Viro return ERR_PTR(-ELOOP); 337873d049a4SAl Viro 337951689104SPaul Moore filename = getname_kernel(name); 3380a1c83681SViresh Kumar if (IS_ERR(filename)) 338151689104SPaul Moore return ERR_CAST(filename); 338251689104SPaul Moore 33839883d185SAl Viro set_nameidata(&nd, -1, filename); 3384c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_RCU); 338573d049a4SAl Viro if (unlikely(file == ERR_PTR(-ECHILD))) 3386c8a53ee5SAl Viro file = path_openat(&nd, op, flags); 338773d049a4SAl Viro if (unlikely(file == ERR_PTR(-ESTALE))) 3388c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_REVAL); 33899883d185SAl Viro restore_nameidata(); 339051689104SPaul Moore putname(filename); 339173d049a4SAl Viro return file; 339273d049a4SAl Viro } 339373d049a4SAl Viro 3394fa14a0b8SAl Viro static struct dentry *filename_create(int dfd, struct filename *name, 33951ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 33961da177e4SLinus Torvalds { 3397c663e5d8SChristoph Hellwig struct dentry *dentry = ERR_PTR(-EEXIST); 3398391172c4SAl Viro struct qstr last; 3399391172c4SAl Viro int type; 3400c30dabfeSJan Kara int err2; 34011ac12b4bSJeff Layton int error; 34021ac12b4bSJeff Layton bool is_dir = (lookup_flags & LOOKUP_DIRECTORY); 34031ac12b4bSJeff Layton 34041ac12b4bSJeff Layton /* 34051ac12b4bSJeff Layton * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any 34061ac12b4bSJeff Layton * other flags passed in are ignored! 34071ac12b4bSJeff Layton */ 34081ac12b4bSJeff Layton lookup_flags &= LOOKUP_REVAL; 34091ac12b4bSJeff Layton 34105c31b6ceSAl Viro name = filename_parentat(dfd, name, lookup_flags, path, &last, &type); 34115c31b6ceSAl Viro if (IS_ERR(name)) 34125c31b6ceSAl Viro return ERR_CAST(name); 34131da177e4SLinus Torvalds 3414c663e5d8SChristoph Hellwig /* 3415c663e5d8SChristoph Hellwig * Yucky last component or no last component at all? 3416c663e5d8SChristoph Hellwig * (foo/., foo/.., /////) 3417c663e5d8SChristoph Hellwig */ 34185c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) 3419ed75e95dSAl Viro goto out; 3420c663e5d8SChristoph Hellwig 3421c30dabfeSJan Kara /* don't fail immediately if it's r/o, at least try to report other errors */ 3422391172c4SAl Viro err2 = mnt_want_write(path->mnt); 3423c663e5d8SChristoph Hellwig /* 3424c663e5d8SChristoph Hellwig * Do the final lookup. 3425c663e5d8SChristoph Hellwig */ 3426391172c4SAl Viro lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL; 3427391172c4SAl Viro mutex_lock_nested(&path->dentry->d_inode->i_mutex, I_MUTEX_PARENT); 3428391172c4SAl Viro dentry = __lookup_hash(&last, path->dentry, lookup_flags); 34291da177e4SLinus Torvalds if (IS_ERR(dentry)) 3430a8104a9fSAl Viro goto unlock; 3431c663e5d8SChristoph Hellwig 3432a8104a9fSAl Viro error = -EEXIST; 3433b18825a7SDavid Howells if (d_is_positive(dentry)) 3434a8104a9fSAl Viro goto fail; 3435b18825a7SDavid Howells 3436c663e5d8SChristoph Hellwig /* 3437c663e5d8SChristoph Hellwig * Special case - lookup gave negative, but... we had foo/bar/ 3438c663e5d8SChristoph Hellwig * From the vfs_mknod() POV we just have a negative dentry - 3439c663e5d8SChristoph Hellwig * all is fine. Let's be bastards - you had / on the end, you've 3440c663e5d8SChristoph Hellwig * been asking for (non-existent) directory. -ENOENT for you. 3441c663e5d8SChristoph Hellwig */ 3442391172c4SAl Viro if (unlikely(!is_dir && last.name[last.len])) { 3443a8104a9fSAl Viro error = -ENOENT; 3444ed75e95dSAl Viro goto fail; 3445e9baf6e5SAl Viro } 3446c30dabfeSJan Kara if (unlikely(err2)) { 3447c30dabfeSJan Kara error = err2; 3448a8104a9fSAl Viro goto fail; 3449c30dabfeSJan Kara } 3450181c37b6SAl Viro putname(name); 3451e9baf6e5SAl Viro return dentry; 34521da177e4SLinus Torvalds fail: 3453a8104a9fSAl Viro dput(dentry); 3454a8104a9fSAl Viro dentry = ERR_PTR(error); 3455a8104a9fSAl Viro unlock: 3456391172c4SAl Viro mutex_unlock(&path->dentry->d_inode->i_mutex); 3457c30dabfeSJan Kara if (!err2) 3458391172c4SAl Viro mnt_drop_write(path->mnt); 3459ed75e95dSAl Viro out: 3460391172c4SAl Viro path_put(path); 3461181c37b6SAl Viro putname(name); 3462ed75e95dSAl Viro return dentry; 3463dae6ad8fSAl Viro } 3464fa14a0b8SAl Viro 3465fa14a0b8SAl Viro struct dentry *kern_path_create(int dfd, const char *pathname, 3466fa14a0b8SAl Viro struct path *path, unsigned int lookup_flags) 3467fa14a0b8SAl Viro { 3468181c37b6SAl Viro return filename_create(dfd, getname_kernel(pathname), 3469181c37b6SAl Viro path, lookup_flags); 3470fa14a0b8SAl Viro } 3471dae6ad8fSAl Viro EXPORT_SYMBOL(kern_path_create); 3472dae6ad8fSAl Viro 3473921a1650SAl Viro void done_path_create(struct path *path, struct dentry *dentry) 3474921a1650SAl Viro { 3475921a1650SAl Viro dput(dentry); 3476921a1650SAl Viro mutex_unlock(&path->dentry->d_inode->i_mutex); 3477a8104a9fSAl Viro mnt_drop_write(path->mnt); 3478921a1650SAl Viro path_put(path); 3479921a1650SAl Viro } 3480921a1650SAl Viro EXPORT_SYMBOL(done_path_create); 3481921a1650SAl Viro 3482520ae687SAl Viro inline struct dentry *user_path_create(int dfd, const char __user *pathname, 34831ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 3484dae6ad8fSAl Viro { 3485181c37b6SAl Viro return filename_create(dfd, getname(pathname), path, lookup_flags); 3486dae6ad8fSAl Viro } 3487dae6ad8fSAl Viro EXPORT_SYMBOL(user_path_create); 3488dae6ad8fSAl Viro 34891a67aafbSAl Viro int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) 34901da177e4SLinus Torvalds { 3491a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 34921da177e4SLinus Torvalds 34931da177e4SLinus Torvalds if (error) 34941da177e4SLinus Torvalds return error; 34951da177e4SLinus Torvalds 3496975d6b39SEric W. Biederman if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) 34971da177e4SLinus Torvalds return -EPERM; 34981da177e4SLinus Torvalds 3499acfa4380SAl Viro if (!dir->i_op->mknod) 35001da177e4SLinus Torvalds return -EPERM; 35011da177e4SLinus Torvalds 350208ce5f16SSerge E. Hallyn error = devcgroup_inode_mknod(mode, dev); 350308ce5f16SSerge E. Hallyn if (error) 350408ce5f16SSerge E. Hallyn return error; 350508ce5f16SSerge E. Hallyn 35061da177e4SLinus Torvalds error = security_inode_mknod(dir, dentry, mode, dev); 35071da177e4SLinus Torvalds if (error) 35081da177e4SLinus Torvalds return error; 35091da177e4SLinus Torvalds 35101da177e4SLinus Torvalds error = dir->i_op->mknod(dir, dentry, mode, dev); 3511a74574aaSStephen Smalley if (!error) 3512f38aa942SAmy Griffis fsnotify_create(dir, dentry); 35131da177e4SLinus Torvalds return error; 35141da177e4SLinus Torvalds } 35154d359507SAl Viro EXPORT_SYMBOL(vfs_mknod); 35161da177e4SLinus Torvalds 3517f69aac00SAl Viro static int may_mknod(umode_t mode) 3518463c3197SDave Hansen { 3519463c3197SDave Hansen switch (mode & S_IFMT) { 3520463c3197SDave Hansen case S_IFREG: 3521463c3197SDave Hansen case S_IFCHR: 3522463c3197SDave Hansen case S_IFBLK: 3523463c3197SDave Hansen case S_IFIFO: 3524463c3197SDave Hansen case S_IFSOCK: 3525463c3197SDave Hansen case 0: /* zero mode translates to S_IFREG */ 3526463c3197SDave Hansen return 0; 3527463c3197SDave Hansen case S_IFDIR: 3528463c3197SDave Hansen return -EPERM; 3529463c3197SDave Hansen default: 3530463c3197SDave Hansen return -EINVAL; 3531463c3197SDave Hansen } 3532463c3197SDave Hansen } 3533463c3197SDave Hansen 35348208a22bSAl Viro SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, 35352e4d0924SHeiko Carstens unsigned, dev) 35361da177e4SLinus Torvalds { 35371da177e4SLinus Torvalds struct dentry *dentry; 3538dae6ad8fSAl Viro struct path path; 3539dae6ad8fSAl Viro int error; 3540972567f1SJeff Layton unsigned int lookup_flags = 0; 35411da177e4SLinus Torvalds 35428e4bfca1SAl Viro error = may_mknod(mode); 35438e4bfca1SAl Viro if (error) 35448e4bfca1SAl Viro return error; 3545972567f1SJeff Layton retry: 3546972567f1SJeff Layton dentry = user_path_create(dfd, filename, &path, lookup_flags); 3547dae6ad8fSAl Viro if (IS_ERR(dentry)) 3548dae6ad8fSAl Viro return PTR_ERR(dentry); 35492ad94ae6SAl Viro 3550dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3551ce3b0f8dSAl Viro mode &= ~current_umask(); 3552dae6ad8fSAl Viro error = security_path_mknod(&path, dentry, mode, dev); 3553be6d3e56SKentaro Takeda if (error) 3554a8104a9fSAl Viro goto out; 35551da177e4SLinus Torvalds switch (mode & S_IFMT) { 35561da177e4SLinus Torvalds case 0: case S_IFREG: 3557312b63fbSAl Viro error = vfs_create(path.dentry->d_inode,dentry,mode,true); 35581da177e4SLinus Torvalds break; 35591da177e4SLinus Torvalds case S_IFCHR: case S_IFBLK: 3560dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode, 35611da177e4SLinus Torvalds new_decode_dev(dev)); 35621da177e4SLinus Torvalds break; 35631da177e4SLinus Torvalds case S_IFIFO: case S_IFSOCK: 3564dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); 35651da177e4SLinus Torvalds break; 35661da177e4SLinus Torvalds } 3567a8104a9fSAl Viro out: 3568921a1650SAl Viro done_path_create(&path, dentry); 3569972567f1SJeff Layton if (retry_estale(error, lookup_flags)) { 3570972567f1SJeff Layton lookup_flags |= LOOKUP_REVAL; 3571972567f1SJeff Layton goto retry; 3572972567f1SJeff Layton } 35731da177e4SLinus Torvalds return error; 35741da177e4SLinus Torvalds } 35751da177e4SLinus Torvalds 35768208a22bSAl Viro SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) 35775590ff0dSUlrich Drepper { 35785590ff0dSUlrich Drepper return sys_mknodat(AT_FDCWD, filename, mode, dev); 35795590ff0dSUlrich Drepper } 35805590ff0dSUlrich Drepper 358118bb1db3SAl Viro int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 35821da177e4SLinus Torvalds { 3583a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 35848de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 35851da177e4SLinus Torvalds 35861da177e4SLinus Torvalds if (error) 35871da177e4SLinus Torvalds return error; 35881da177e4SLinus Torvalds 3589acfa4380SAl Viro if (!dir->i_op->mkdir) 35901da177e4SLinus Torvalds return -EPERM; 35911da177e4SLinus Torvalds 35921da177e4SLinus Torvalds mode &= (S_IRWXUGO|S_ISVTX); 35931da177e4SLinus Torvalds error = security_inode_mkdir(dir, dentry, mode); 35941da177e4SLinus Torvalds if (error) 35951da177e4SLinus Torvalds return error; 35961da177e4SLinus Torvalds 35978de52778SAl Viro if (max_links && dir->i_nlink >= max_links) 35988de52778SAl Viro return -EMLINK; 35998de52778SAl Viro 36001da177e4SLinus Torvalds error = dir->i_op->mkdir(dir, dentry, mode); 3601a74574aaSStephen Smalley if (!error) 3602f38aa942SAmy Griffis fsnotify_mkdir(dir, dentry); 36031da177e4SLinus Torvalds return error; 36041da177e4SLinus Torvalds } 36054d359507SAl Viro EXPORT_SYMBOL(vfs_mkdir); 36061da177e4SLinus Torvalds 3607a218d0fdSAl Viro SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) 36081da177e4SLinus Torvalds { 36096902d925SDave Hansen struct dentry *dentry; 3610dae6ad8fSAl Viro struct path path; 3611dae6ad8fSAl Viro int error; 3612b76d8b82SJeff Layton unsigned int lookup_flags = LOOKUP_DIRECTORY; 36131da177e4SLinus Torvalds 3614b76d8b82SJeff Layton retry: 3615b76d8b82SJeff Layton dentry = user_path_create(dfd, pathname, &path, lookup_flags); 36166902d925SDave Hansen if (IS_ERR(dentry)) 3617dae6ad8fSAl Viro return PTR_ERR(dentry); 36186902d925SDave Hansen 3619dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3620ce3b0f8dSAl Viro mode &= ~current_umask(); 3621dae6ad8fSAl Viro error = security_path_mkdir(&path, dentry, mode); 3622a8104a9fSAl Viro if (!error) 3623dae6ad8fSAl Viro error = vfs_mkdir(path.dentry->d_inode, dentry, mode); 3624921a1650SAl Viro done_path_create(&path, dentry); 3625b76d8b82SJeff Layton if (retry_estale(error, lookup_flags)) { 3626b76d8b82SJeff Layton lookup_flags |= LOOKUP_REVAL; 3627b76d8b82SJeff Layton goto retry; 3628b76d8b82SJeff Layton } 36291da177e4SLinus Torvalds return error; 36301da177e4SLinus Torvalds } 36311da177e4SLinus Torvalds 3632a218d0fdSAl Viro SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) 36335590ff0dSUlrich Drepper { 36345590ff0dSUlrich Drepper return sys_mkdirat(AT_FDCWD, pathname, mode); 36355590ff0dSUlrich Drepper } 36365590ff0dSUlrich Drepper 36371da177e4SLinus Torvalds /* 3638a71905f0SSage Weil * The dentry_unhash() helper will try to drop the dentry early: we 3639c0d02594SJ. Bruce Fields * should have a usage count of 1 if we're the only user of this 3640a71905f0SSage Weil * dentry, and if that is true (possibly after pruning the dcache), 3641a71905f0SSage Weil * then we drop the dentry now. 36421da177e4SLinus Torvalds * 36431da177e4SLinus Torvalds * A low-level filesystem can, if it choses, legally 36441da177e4SLinus Torvalds * do a 36451da177e4SLinus Torvalds * 36461da177e4SLinus Torvalds * if (!d_unhashed(dentry)) 36471da177e4SLinus Torvalds * return -EBUSY; 36481da177e4SLinus Torvalds * 36491da177e4SLinus Torvalds * if it cannot handle the case of removing a directory 36501da177e4SLinus Torvalds * that is still in use by something else.. 36511da177e4SLinus Torvalds */ 36521da177e4SLinus Torvalds void dentry_unhash(struct dentry *dentry) 36531da177e4SLinus Torvalds { 36541da177e4SLinus Torvalds shrink_dcache_parent(dentry); 36551da177e4SLinus Torvalds spin_lock(&dentry->d_lock); 365698474236SWaiman Long if (dentry->d_lockref.count == 1) 36571da177e4SLinus Torvalds __d_drop(dentry); 36581da177e4SLinus Torvalds spin_unlock(&dentry->d_lock); 36591da177e4SLinus Torvalds } 36604d359507SAl Viro EXPORT_SYMBOL(dentry_unhash); 36611da177e4SLinus Torvalds 36621da177e4SLinus Torvalds int vfs_rmdir(struct inode *dir, struct dentry *dentry) 36631da177e4SLinus Torvalds { 36641da177e4SLinus Torvalds int error = may_delete(dir, dentry, 1); 36651da177e4SLinus Torvalds 36661da177e4SLinus Torvalds if (error) 36671da177e4SLinus Torvalds return error; 36681da177e4SLinus Torvalds 3669acfa4380SAl Viro if (!dir->i_op->rmdir) 36701da177e4SLinus Torvalds return -EPERM; 36711da177e4SLinus Torvalds 36721d2ef590SAl Viro dget(dentry); 36731b1dcc1bSJes Sorensen mutex_lock(&dentry->d_inode->i_mutex); 3674912dbc15SSage Weil 36751da177e4SLinus Torvalds error = -EBUSY; 36767af1364fSEric W. Biederman if (is_local_mountpoint(dentry)) 3677912dbc15SSage Weil goto out; 3678912dbc15SSage Weil 36791da177e4SLinus Torvalds error = security_inode_rmdir(dir, dentry); 3680912dbc15SSage Weil if (error) 3681912dbc15SSage Weil goto out; 3682912dbc15SSage Weil 36833cebde24SSage Weil shrink_dcache_parent(dentry); 36841da177e4SLinus Torvalds error = dir->i_op->rmdir(dir, dentry); 3685912dbc15SSage Weil if (error) 3686912dbc15SSage Weil goto out; 3687912dbc15SSage Weil 36881da177e4SLinus Torvalds dentry->d_inode->i_flags |= S_DEAD; 3689d83c49f3SAl Viro dont_mount(dentry); 36908ed936b5SEric W. Biederman detach_mounts(dentry); 36911da177e4SLinus Torvalds 3692912dbc15SSage Weil out: 3693912dbc15SSage Weil mutex_unlock(&dentry->d_inode->i_mutex); 36941d2ef590SAl Viro dput(dentry); 3695912dbc15SSage Weil if (!error) 3696912dbc15SSage Weil d_delete(dentry); 36971da177e4SLinus Torvalds return error; 36981da177e4SLinus Torvalds } 36994d359507SAl Viro EXPORT_SYMBOL(vfs_rmdir); 37001da177e4SLinus Torvalds 37015590ff0dSUlrich Drepper static long do_rmdir(int dfd, const char __user *pathname) 37021da177e4SLinus Torvalds { 37031da177e4SLinus Torvalds int error = 0; 370491a27b2aSJeff Layton struct filename *name; 37051da177e4SLinus Torvalds struct dentry *dentry; 3706f5beed75SAl Viro struct path path; 3707f5beed75SAl Viro struct qstr last; 3708f5beed75SAl Viro int type; 3709c6ee9206SJeff Layton unsigned int lookup_flags = 0; 3710c6ee9206SJeff Layton retry: 3711f5beed75SAl Viro name = user_path_parent(dfd, pathname, 3712f5beed75SAl Viro &path, &last, &type, lookup_flags); 371391a27b2aSJeff Layton if (IS_ERR(name)) 371491a27b2aSJeff Layton return PTR_ERR(name); 37151da177e4SLinus Torvalds 3716f5beed75SAl Viro switch (type) { 37171da177e4SLinus Torvalds case LAST_DOTDOT: 37181da177e4SLinus Torvalds error = -ENOTEMPTY; 37191da177e4SLinus Torvalds goto exit1; 37201da177e4SLinus Torvalds case LAST_DOT: 37211da177e4SLinus Torvalds error = -EINVAL; 37221da177e4SLinus Torvalds goto exit1; 37231da177e4SLinus Torvalds case LAST_ROOT: 37241da177e4SLinus Torvalds error = -EBUSY; 37251da177e4SLinus Torvalds goto exit1; 37261da177e4SLinus Torvalds } 37270612d9fbSOGAWA Hirofumi 3728f5beed75SAl Viro error = mnt_want_write(path.mnt); 3729c30dabfeSJan Kara if (error) 3730c30dabfeSJan Kara goto exit1; 37310612d9fbSOGAWA Hirofumi 3732f5beed75SAl Viro mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 3733f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 37341da177e4SLinus Torvalds error = PTR_ERR(dentry); 37356902d925SDave Hansen if (IS_ERR(dentry)) 37366902d925SDave Hansen goto exit2; 3737e6bc45d6STheodore Ts'o if (!dentry->d_inode) { 3738e6bc45d6STheodore Ts'o error = -ENOENT; 3739e6bc45d6STheodore Ts'o goto exit3; 3740e6bc45d6STheodore Ts'o } 3741f5beed75SAl Viro error = security_path_rmdir(&path, dentry); 3742be6d3e56SKentaro Takeda if (error) 3743c30dabfeSJan Kara goto exit3; 3744f5beed75SAl Viro error = vfs_rmdir(path.dentry->d_inode, dentry); 37450622753bSDave Hansen exit3: 37461da177e4SLinus Torvalds dput(dentry); 37476902d925SDave Hansen exit2: 3748f5beed75SAl Viro mutex_unlock(&path.dentry->d_inode->i_mutex); 3749f5beed75SAl Viro mnt_drop_write(path.mnt); 37501da177e4SLinus Torvalds exit1: 3751f5beed75SAl Viro path_put(&path); 37521da177e4SLinus Torvalds putname(name); 3753c6ee9206SJeff Layton if (retry_estale(error, lookup_flags)) { 3754c6ee9206SJeff Layton lookup_flags |= LOOKUP_REVAL; 3755c6ee9206SJeff Layton goto retry; 3756c6ee9206SJeff Layton } 37571da177e4SLinus Torvalds return error; 37581da177e4SLinus Torvalds } 37591da177e4SLinus Torvalds 37603cdad428SHeiko Carstens SYSCALL_DEFINE1(rmdir, const char __user *, pathname) 37615590ff0dSUlrich Drepper { 37625590ff0dSUlrich Drepper return do_rmdir(AT_FDCWD, pathname); 37635590ff0dSUlrich Drepper } 37645590ff0dSUlrich Drepper 3765b21996e3SJ. Bruce Fields /** 3766b21996e3SJ. Bruce Fields * vfs_unlink - unlink a filesystem object 3767b21996e3SJ. Bruce Fields * @dir: parent directory 3768b21996e3SJ. Bruce Fields * @dentry: victim 3769b21996e3SJ. Bruce Fields * @delegated_inode: returns victim inode, if the inode is delegated. 3770b21996e3SJ. Bruce Fields * 3771b21996e3SJ. Bruce Fields * The caller must hold dir->i_mutex. 3772b21996e3SJ. Bruce Fields * 3773b21996e3SJ. Bruce Fields * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and 3774b21996e3SJ. Bruce Fields * return a reference to the inode in delegated_inode. The caller 3775b21996e3SJ. Bruce Fields * should then break the delegation on that inode and retry. Because 3776b21996e3SJ. Bruce Fields * breaking a delegation may take a long time, the caller should drop 3777b21996e3SJ. Bruce Fields * dir->i_mutex before doing so. 3778b21996e3SJ. Bruce Fields * 3779b21996e3SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 3780b21996e3SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 3781b21996e3SJ. Bruce Fields * to be NFS exported. 3782b21996e3SJ. Bruce Fields */ 3783b21996e3SJ. Bruce Fields int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode) 37841da177e4SLinus Torvalds { 37859accbb97SJ. Bruce Fields struct inode *target = dentry->d_inode; 37861da177e4SLinus Torvalds int error = may_delete(dir, dentry, 0); 37871da177e4SLinus Torvalds 37881da177e4SLinus Torvalds if (error) 37891da177e4SLinus Torvalds return error; 37901da177e4SLinus Torvalds 3791acfa4380SAl Viro if (!dir->i_op->unlink) 37921da177e4SLinus Torvalds return -EPERM; 37931da177e4SLinus Torvalds 37949accbb97SJ. Bruce Fields mutex_lock(&target->i_mutex); 37958ed936b5SEric W. Biederman if (is_local_mountpoint(dentry)) 37961da177e4SLinus Torvalds error = -EBUSY; 37971da177e4SLinus Torvalds else { 37981da177e4SLinus Torvalds error = security_inode_unlink(dir, dentry); 3799bec1052eSAl Viro if (!error) { 38005a14696cSJ. Bruce Fields error = try_break_deleg(target, delegated_inode); 38015a14696cSJ. Bruce Fields if (error) 3802b21996e3SJ. Bruce Fields goto out; 38031da177e4SLinus Torvalds error = dir->i_op->unlink(dir, dentry); 38048ed936b5SEric W. Biederman if (!error) { 3805d83c49f3SAl Viro dont_mount(dentry); 38068ed936b5SEric W. Biederman detach_mounts(dentry); 38078ed936b5SEric W. Biederman } 3808bec1052eSAl Viro } 38091da177e4SLinus Torvalds } 3810b21996e3SJ. Bruce Fields out: 38119accbb97SJ. Bruce Fields mutex_unlock(&target->i_mutex); 38121da177e4SLinus Torvalds 38131da177e4SLinus Torvalds /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 38141da177e4SLinus Torvalds if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { 38159accbb97SJ. Bruce Fields fsnotify_link_count(target); 38161da177e4SLinus Torvalds d_delete(dentry); 38171da177e4SLinus Torvalds } 38180eeca283SRobert Love 38191da177e4SLinus Torvalds return error; 38201da177e4SLinus Torvalds } 38214d359507SAl Viro EXPORT_SYMBOL(vfs_unlink); 38221da177e4SLinus Torvalds 38231da177e4SLinus Torvalds /* 38241da177e4SLinus Torvalds * Make sure that the actual truncation of the file will occur outside its 38251b1dcc1bSJes Sorensen * directory's i_mutex. Truncate can take a long time if there is a lot of 38261da177e4SLinus Torvalds * writeout happening, and we don't want to prevent access to the directory 38271da177e4SLinus Torvalds * while waiting on the I/O. 38281da177e4SLinus Torvalds */ 38295590ff0dSUlrich Drepper static long do_unlinkat(int dfd, const char __user *pathname) 38301da177e4SLinus Torvalds { 38312ad94ae6SAl Viro int error; 383291a27b2aSJeff Layton struct filename *name; 38331da177e4SLinus Torvalds struct dentry *dentry; 3834f5beed75SAl Viro struct path path; 3835f5beed75SAl Viro struct qstr last; 3836f5beed75SAl Viro int type; 38371da177e4SLinus Torvalds struct inode *inode = NULL; 3838b21996e3SJ. Bruce Fields struct inode *delegated_inode = NULL; 38395d18f813SJeff Layton unsigned int lookup_flags = 0; 38405d18f813SJeff Layton retry: 3841f5beed75SAl Viro name = user_path_parent(dfd, pathname, 3842f5beed75SAl Viro &path, &last, &type, lookup_flags); 384391a27b2aSJeff Layton if (IS_ERR(name)) 384491a27b2aSJeff Layton return PTR_ERR(name); 38452ad94ae6SAl Viro 38461da177e4SLinus Torvalds error = -EISDIR; 3847f5beed75SAl Viro if (type != LAST_NORM) 38481da177e4SLinus Torvalds goto exit1; 38490612d9fbSOGAWA Hirofumi 3850f5beed75SAl Viro error = mnt_want_write(path.mnt); 3851c30dabfeSJan Kara if (error) 3852c30dabfeSJan Kara goto exit1; 3853b21996e3SJ. Bruce Fields retry_deleg: 3854f5beed75SAl Viro mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 3855f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 38561da177e4SLinus Torvalds error = PTR_ERR(dentry); 38571da177e4SLinus Torvalds if (!IS_ERR(dentry)) { 38581da177e4SLinus Torvalds /* Why not before? Because we want correct error value */ 3859f5beed75SAl Viro if (last.name[last.len]) 386050338b88STörök Edwin goto slashes; 38611da177e4SLinus Torvalds inode = dentry->d_inode; 3862b18825a7SDavid Howells if (d_is_negative(dentry)) 3863e6bc45d6STheodore Ts'o goto slashes; 38647de9c6eeSAl Viro ihold(inode); 3865f5beed75SAl Viro error = security_path_unlink(&path, dentry); 3866be6d3e56SKentaro Takeda if (error) 3867c30dabfeSJan Kara goto exit2; 3868f5beed75SAl Viro error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode); 38691da177e4SLinus Torvalds exit2: 38701da177e4SLinus Torvalds dput(dentry); 38711da177e4SLinus Torvalds } 3872f5beed75SAl Viro mutex_unlock(&path.dentry->d_inode->i_mutex); 38731da177e4SLinus Torvalds if (inode) 38741da177e4SLinus Torvalds iput(inode); /* truncate the inode here */ 3875b21996e3SJ. Bruce Fields inode = NULL; 3876b21996e3SJ. Bruce Fields if (delegated_inode) { 38775a14696cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 3878b21996e3SJ. Bruce Fields if (!error) 3879b21996e3SJ. Bruce Fields goto retry_deleg; 3880b21996e3SJ. Bruce Fields } 3881f5beed75SAl Viro mnt_drop_write(path.mnt); 38821da177e4SLinus Torvalds exit1: 3883f5beed75SAl Viro path_put(&path); 38841da177e4SLinus Torvalds putname(name); 38855d18f813SJeff Layton if (retry_estale(error, lookup_flags)) { 38865d18f813SJeff Layton lookup_flags |= LOOKUP_REVAL; 38875d18f813SJeff Layton inode = NULL; 38885d18f813SJeff Layton goto retry; 38895d18f813SJeff Layton } 38901da177e4SLinus Torvalds return error; 38911da177e4SLinus Torvalds 38921da177e4SLinus Torvalds slashes: 3893b18825a7SDavid Howells if (d_is_negative(dentry)) 3894b18825a7SDavid Howells error = -ENOENT; 389544b1d530SMiklos Szeredi else if (d_is_dir(dentry)) 3896b18825a7SDavid Howells error = -EISDIR; 3897b18825a7SDavid Howells else 3898b18825a7SDavid Howells error = -ENOTDIR; 38991da177e4SLinus Torvalds goto exit2; 39001da177e4SLinus Torvalds } 39011da177e4SLinus Torvalds 39022e4d0924SHeiko Carstens SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) 39035590ff0dSUlrich Drepper { 39045590ff0dSUlrich Drepper if ((flag & ~AT_REMOVEDIR) != 0) 39055590ff0dSUlrich Drepper return -EINVAL; 39065590ff0dSUlrich Drepper 39075590ff0dSUlrich Drepper if (flag & AT_REMOVEDIR) 39085590ff0dSUlrich Drepper return do_rmdir(dfd, pathname); 39095590ff0dSUlrich Drepper 39105590ff0dSUlrich Drepper return do_unlinkat(dfd, pathname); 39115590ff0dSUlrich Drepper } 39125590ff0dSUlrich Drepper 39133480b257SHeiko Carstens SYSCALL_DEFINE1(unlink, const char __user *, pathname) 39145590ff0dSUlrich Drepper { 39155590ff0dSUlrich Drepper return do_unlinkat(AT_FDCWD, pathname); 39165590ff0dSUlrich Drepper } 39175590ff0dSUlrich Drepper 3918db2e747bSMiklos Szeredi int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) 39191da177e4SLinus Torvalds { 3920a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 39211da177e4SLinus Torvalds 39221da177e4SLinus Torvalds if (error) 39231da177e4SLinus Torvalds return error; 39241da177e4SLinus Torvalds 3925acfa4380SAl Viro if (!dir->i_op->symlink) 39261da177e4SLinus Torvalds return -EPERM; 39271da177e4SLinus Torvalds 39281da177e4SLinus Torvalds error = security_inode_symlink(dir, dentry, oldname); 39291da177e4SLinus Torvalds if (error) 39301da177e4SLinus Torvalds return error; 39311da177e4SLinus Torvalds 39321da177e4SLinus Torvalds error = dir->i_op->symlink(dir, dentry, oldname); 3933a74574aaSStephen Smalley if (!error) 3934f38aa942SAmy Griffis fsnotify_create(dir, dentry); 39351da177e4SLinus Torvalds return error; 39361da177e4SLinus Torvalds } 39374d359507SAl Viro EXPORT_SYMBOL(vfs_symlink); 39381da177e4SLinus Torvalds 39392e4d0924SHeiko Carstens SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, 39402e4d0924SHeiko Carstens int, newdfd, const char __user *, newname) 39411da177e4SLinus Torvalds { 39422ad94ae6SAl Viro int error; 394391a27b2aSJeff Layton struct filename *from; 39446902d925SDave Hansen struct dentry *dentry; 3945dae6ad8fSAl Viro struct path path; 3946f46d3567SJeff Layton unsigned int lookup_flags = 0; 39471da177e4SLinus Torvalds 39481da177e4SLinus Torvalds from = getname(oldname); 39491da177e4SLinus Torvalds if (IS_ERR(from)) 39501da177e4SLinus Torvalds return PTR_ERR(from); 3951f46d3567SJeff Layton retry: 3952f46d3567SJeff Layton dentry = user_path_create(newdfd, newname, &path, lookup_flags); 39531da177e4SLinus Torvalds error = PTR_ERR(dentry); 39546902d925SDave Hansen if (IS_ERR(dentry)) 3955dae6ad8fSAl Viro goto out_putname; 39566902d925SDave Hansen 395791a27b2aSJeff Layton error = security_path_symlink(&path, dentry, from->name); 3958a8104a9fSAl Viro if (!error) 395991a27b2aSJeff Layton error = vfs_symlink(path.dentry->d_inode, dentry, from->name); 3960921a1650SAl Viro done_path_create(&path, dentry); 3961f46d3567SJeff Layton if (retry_estale(error, lookup_flags)) { 3962f46d3567SJeff Layton lookup_flags |= LOOKUP_REVAL; 3963f46d3567SJeff Layton goto retry; 3964f46d3567SJeff Layton } 39656902d925SDave Hansen out_putname: 39661da177e4SLinus Torvalds putname(from); 39671da177e4SLinus Torvalds return error; 39681da177e4SLinus Torvalds } 39691da177e4SLinus Torvalds 39703480b257SHeiko Carstens SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) 39715590ff0dSUlrich Drepper { 39725590ff0dSUlrich Drepper return sys_symlinkat(oldname, AT_FDCWD, newname); 39735590ff0dSUlrich Drepper } 39745590ff0dSUlrich Drepper 3975146a8595SJ. Bruce Fields /** 3976146a8595SJ. Bruce Fields * vfs_link - create a new link 3977146a8595SJ. Bruce Fields * @old_dentry: object to be linked 3978146a8595SJ. Bruce Fields * @dir: new parent 3979146a8595SJ. Bruce Fields * @new_dentry: where to create the new link 3980146a8595SJ. Bruce Fields * @delegated_inode: returns inode needing a delegation break 3981146a8595SJ. Bruce Fields * 3982146a8595SJ. Bruce Fields * The caller must hold dir->i_mutex 3983146a8595SJ. Bruce Fields * 3984146a8595SJ. Bruce Fields * If vfs_link discovers a delegation on the to-be-linked file in need 3985146a8595SJ. Bruce Fields * of breaking, it will return -EWOULDBLOCK and return a reference to the 3986146a8595SJ. Bruce Fields * inode in delegated_inode. The caller should then break the delegation 3987146a8595SJ. Bruce Fields * and retry. Because breaking a delegation may take a long time, the 3988146a8595SJ. Bruce Fields * caller should drop the i_mutex before doing so. 3989146a8595SJ. Bruce Fields * 3990146a8595SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 3991146a8595SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 3992146a8595SJ. Bruce Fields * to be NFS exported. 3993146a8595SJ. Bruce Fields */ 3994146a8595SJ. Bruce Fields int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode) 39951da177e4SLinus Torvalds { 39961da177e4SLinus Torvalds struct inode *inode = old_dentry->d_inode; 39978de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 39981da177e4SLinus Torvalds int error; 39991da177e4SLinus Torvalds 40001da177e4SLinus Torvalds if (!inode) 40011da177e4SLinus Torvalds return -ENOENT; 40021da177e4SLinus Torvalds 4003a95164d9SMiklos Szeredi error = may_create(dir, new_dentry); 40041da177e4SLinus Torvalds if (error) 40051da177e4SLinus Torvalds return error; 40061da177e4SLinus Torvalds 40071da177e4SLinus Torvalds if (dir->i_sb != inode->i_sb) 40081da177e4SLinus Torvalds return -EXDEV; 40091da177e4SLinus Torvalds 40101da177e4SLinus Torvalds /* 40111da177e4SLinus Torvalds * A link to an append-only or immutable file cannot be created. 40121da177e4SLinus Torvalds */ 40131da177e4SLinus Torvalds if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 40141da177e4SLinus Torvalds return -EPERM; 4015acfa4380SAl Viro if (!dir->i_op->link) 40161da177e4SLinus Torvalds return -EPERM; 40177e79eedbSTetsuo Handa if (S_ISDIR(inode->i_mode)) 40181da177e4SLinus Torvalds return -EPERM; 40191da177e4SLinus Torvalds 40201da177e4SLinus Torvalds error = security_inode_link(old_dentry, dir, new_dentry); 40211da177e4SLinus Torvalds if (error) 40221da177e4SLinus Torvalds return error; 40231da177e4SLinus Torvalds 40247e79eedbSTetsuo Handa mutex_lock(&inode->i_mutex); 4025aae8a97dSAneesh Kumar K.V /* Make sure we don't allow creating hardlink to an unlinked file */ 4026f4e0c30cSAl Viro if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) 4027aae8a97dSAneesh Kumar K.V error = -ENOENT; 40288de52778SAl Viro else if (max_links && inode->i_nlink >= max_links) 40298de52778SAl Viro error = -EMLINK; 4030146a8595SJ. Bruce Fields else { 4031146a8595SJ. Bruce Fields error = try_break_deleg(inode, delegated_inode); 4032146a8595SJ. Bruce Fields if (!error) 40331da177e4SLinus Torvalds error = dir->i_op->link(old_dentry, dir, new_dentry); 4034146a8595SJ. Bruce Fields } 4035f4e0c30cSAl Viro 4036f4e0c30cSAl Viro if (!error && (inode->i_state & I_LINKABLE)) { 4037f4e0c30cSAl Viro spin_lock(&inode->i_lock); 4038f4e0c30cSAl Viro inode->i_state &= ~I_LINKABLE; 4039f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 4040f4e0c30cSAl Viro } 40417e79eedbSTetsuo Handa mutex_unlock(&inode->i_mutex); 4042e31e14ecSStephen Smalley if (!error) 40437e79eedbSTetsuo Handa fsnotify_link(dir, inode, new_dentry); 40441da177e4SLinus Torvalds return error; 40451da177e4SLinus Torvalds } 40464d359507SAl Viro EXPORT_SYMBOL(vfs_link); 40471da177e4SLinus Torvalds 40481da177e4SLinus Torvalds /* 40491da177e4SLinus Torvalds * Hardlinks are often used in delicate situations. We avoid 40501da177e4SLinus Torvalds * security-related surprises by not following symlinks on the 40511da177e4SLinus Torvalds * newname. --KAB 40521da177e4SLinus Torvalds * 40531da177e4SLinus Torvalds * We don't follow them on the oldname either to be compatible 40541da177e4SLinus Torvalds * with linux 2.0, and to avoid hard-linking to directories 40551da177e4SLinus Torvalds * and other special files. --ADM 40561da177e4SLinus Torvalds */ 40572e4d0924SHeiko Carstens SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, 40582e4d0924SHeiko Carstens int, newdfd, const char __user *, newname, int, flags) 40591da177e4SLinus Torvalds { 40601da177e4SLinus Torvalds struct dentry *new_dentry; 4061dae6ad8fSAl Viro struct path old_path, new_path; 4062146a8595SJ. Bruce Fields struct inode *delegated_inode = NULL; 406311a7b371SAneesh Kumar K.V int how = 0; 40641da177e4SLinus Torvalds int error; 40651da177e4SLinus Torvalds 406611a7b371SAneesh Kumar K.V if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) 4067c04030e1SUlrich Drepper return -EINVAL; 406811a7b371SAneesh Kumar K.V /* 4069f0cc6ffbSLinus Torvalds * To use null names we require CAP_DAC_READ_SEARCH 4070f0cc6ffbSLinus Torvalds * This ensures that not everyone will be able to create 4071f0cc6ffbSLinus Torvalds * handlink using the passed filedescriptor. 407211a7b371SAneesh Kumar K.V */ 4073f0cc6ffbSLinus Torvalds if (flags & AT_EMPTY_PATH) { 4074f0cc6ffbSLinus Torvalds if (!capable(CAP_DAC_READ_SEARCH)) 4075f0cc6ffbSLinus Torvalds return -ENOENT; 407611a7b371SAneesh Kumar K.V how = LOOKUP_EMPTY; 4077f0cc6ffbSLinus Torvalds } 4078c04030e1SUlrich Drepper 407911a7b371SAneesh Kumar K.V if (flags & AT_SYMLINK_FOLLOW) 408011a7b371SAneesh Kumar K.V how |= LOOKUP_FOLLOW; 4081442e31caSJeff Layton retry: 408211a7b371SAneesh Kumar K.V error = user_path_at(olddfd, oldname, how, &old_path); 40831da177e4SLinus Torvalds if (error) 40842ad94ae6SAl Viro return error; 40852ad94ae6SAl Viro 4086442e31caSJeff Layton new_dentry = user_path_create(newdfd, newname, &new_path, 4087442e31caSJeff Layton (how & LOOKUP_REVAL)); 40881da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 40896902d925SDave Hansen if (IS_ERR(new_dentry)) 4090dae6ad8fSAl Viro goto out; 4091dae6ad8fSAl Viro 4092dae6ad8fSAl Viro error = -EXDEV; 4093dae6ad8fSAl Viro if (old_path.mnt != new_path.mnt) 4094dae6ad8fSAl Viro goto out_dput; 4095800179c9SKees Cook error = may_linkat(&old_path); 4096800179c9SKees Cook if (unlikely(error)) 4097800179c9SKees Cook goto out_dput; 4098dae6ad8fSAl Viro error = security_path_link(old_path.dentry, &new_path, new_dentry); 4099be6d3e56SKentaro Takeda if (error) 4100a8104a9fSAl Viro goto out_dput; 4101146a8595SJ. Bruce Fields error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode); 410275c3f29dSDave Hansen out_dput: 4103921a1650SAl Viro done_path_create(&new_path, new_dentry); 4104146a8595SJ. Bruce Fields if (delegated_inode) { 4105146a8595SJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4106d22e6338SOleg Drokin if (!error) { 4107d22e6338SOleg Drokin path_put(&old_path); 4108146a8595SJ. Bruce Fields goto retry; 4109146a8595SJ. Bruce Fields } 4110d22e6338SOleg Drokin } 4111442e31caSJeff Layton if (retry_estale(error, how)) { 4112d22e6338SOleg Drokin path_put(&old_path); 4113442e31caSJeff Layton how |= LOOKUP_REVAL; 4114442e31caSJeff Layton goto retry; 4115442e31caSJeff Layton } 41161da177e4SLinus Torvalds out: 41172d8f3038SAl Viro path_put(&old_path); 41181da177e4SLinus Torvalds 41191da177e4SLinus Torvalds return error; 41201da177e4SLinus Torvalds } 41211da177e4SLinus Torvalds 41223480b257SHeiko Carstens SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) 41235590ff0dSUlrich Drepper { 4124c04030e1SUlrich Drepper return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 41255590ff0dSUlrich Drepper } 41265590ff0dSUlrich Drepper 4127bc27027aSMiklos Szeredi /** 4128bc27027aSMiklos Szeredi * vfs_rename - rename a filesystem object 4129bc27027aSMiklos Szeredi * @old_dir: parent of source 4130bc27027aSMiklos Szeredi * @old_dentry: source 4131bc27027aSMiklos Szeredi * @new_dir: parent of destination 4132bc27027aSMiklos Szeredi * @new_dentry: destination 4133bc27027aSMiklos Szeredi * @delegated_inode: returns an inode needing a delegation break 4134520c8b16SMiklos Szeredi * @flags: rename flags 4135bc27027aSMiklos Szeredi * 4136bc27027aSMiklos Szeredi * The caller must hold multiple mutexes--see lock_rename()). 4137bc27027aSMiklos Szeredi * 4138bc27027aSMiklos Szeredi * If vfs_rename discovers a delegation in need of breaking at either 4139bc27027aSMiklos Szeredi * the source or destination, it will return -EWOULDBLOCK and return a 4140bc27027aSMiklos Szeredi * reference to the inode in delegated_inode. The caller should then 4141bc27027aSMiklos Szeredi * break the delegation and retry. Because breaking a delegation may 4142bc27027aSMiklos Szeredi * take a long time, the caller should drop all locks before doing 4143bc27027aSMiklos Szeredi * so. 4144bc27027aSMiklos Szeredi * 4145bc27027aSMiklos Szeredi * Alternatively, a caller may pass NULL for delegated_inode. This may 4146bc27027aSMiklos Szeredi * be appropriate for callers that expect the underlying filesystem not 4147bc27027aSMiklos Szeredi * to be NFS exported. 4148bc27027aSMiklos Szeredi * 41491da177e4SLinus Torvalds * The worst of all namespace operations - renaming directory. "Perverted" 41501da177e4SLinus Torvalds * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 41511da177e4SLinus Torvalds * Problems: 4152d03b29a2SJ. Bruce Fields * a) we can get into loop creation. 41531da177e4SLinus Torvalds * b) race potential - two innocent renames can create a loop together. 41541da177e4SLinus Torvalds * That's where 4.4 screws up. Current fix: serialization on 4155a11f3a05SArjan van de Ven * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 41561da177e4SLinus Torvalds * story. 41576cedba89SJ. Bruce Fields * c) we have to lock _four_ objects - parents and victim (if it exists), 41586cedba89SJ. Bruce Fields * and source (if it is not a directory). 41591b1dcc1bSJes Sorensen * And that - after we got ->i_mutex on parents (until then we don't know 41601da177e4SLinus Torvalds * whether the target exists). Solution: try to be smart with locking 41611da177e4SLinus Torvalds * order for inodes. We rely on the fact that tree topology may change 4162a11f3a05SArjan van de Ven * only under ->s_vfs_rename_mutex _and_ that parent of the object we 41631da177e4SLinus Torvalds * move will be locked. Thus we can rank directories by the tree 41641da177e4SLinus Torvalds * (ancestors first) and rank all non-directories after them. 41651da177e4SLinus Torvalds * That works since everybody except rename does "lock parent, lookup, 4166a11f3a05SArjan van de Ven * lock child" and rename is under ->s_vfs_rename_mutex. 41671da177e4SLinus Torvalds * HOWEVER, it relies on the assumption that any object with ->lookup() 41681da177e4SLinus Torvalds * has no more than 1 dentry. If "hybrid" objects will ever appear, 41691da177e4SLinus Torvalds * we'd better make sure that there's no link(2) for them. 4170e4eaac06SSage Weil * d) conversion from fhandle to dentry may come in the wrong moment - when 41711b1dcc1bSJes Sorensen * we are removing the target. Solution: we will have to grab ->i_mutex 41721da177e4SLinus Torvalds * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 4173c41b20e7SAdam Buchbinder * ->i_mutex on parents, which works but leads to some truly excessive 41741da177e4SLinus Torvalds * locking]. 41751da177e4SLinus Torvalds */ 41761da177e4SLinus Torvalds int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, 41778e6d782cSJ. Bruce Fields struct inode *new_dir, struct dentry *new_dentry, 4178520c8b16SMiklos Szeredi struct inode **delegated_inode, unsigned int flags) 41791da177e4SLinus Torvalds { 41801da177e4SLinus Torvalds int error; 4181bc27027aSMiklos Szeredi bool is_dir = d_is_dir(old_dentry); 418259b0df21SEric Paris const unsigned char *old_name; 4183bc27027aSMiklos Szeredi struct inode *source = old_dentry->d_inode; 4184bc27027aSMiklos Szeredi struct inode *target = new_dentry->d_inode; 4185da1ce067SMiklos Szeredi bool new_is_dir = false; 4186da1ce067SMiklos Szeredi unsigned max_links = new_dir->i_sb->s_max_links; 41871da177e4SLinus Torvalds 4188bc27027aSMiklos Szeredi if (source == target) 41891da177e4SLinus Torvalds return 0; 41901da177e4SLinus Torvalds 41911da177e4SLinus Torvalds error = may_delete(old_dir, old_dentry, is_dir); 41921da177e4SLinus Torvalds if (error) 41931da177e4SLinus Torvalds return error; 41941da177e4SLinus Torvalds 4195da1ce067SMiklos Szeredi if (!target) { 4196a95164d9SMiklos Szeredi error = may_create(new_dir, new_dentry); 4197da1ce067SMiklos Szeredi } else { 4198da1ce067SMiklos Szeredi new_is_dir = d_is_dir(new_dentry); 4199da1ce067SMiklos Szeredi 4200da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 42011da177e4SLinus Torvalds error = may_delete(new_dir, new_dentry, is_dir); 4202da1ce067SMiklos Szeredi else 4203da1ce067SMiklos Szeredi error = may_delete(new_dir, new_dentry, new_is_dir); 4204da1ce067SMiklos Szeredi } 42051da177e4SLinus Torvalds if (error) 42061da177e4SLinus Torvalds return error; 42071da177e4SLinus Torvalds 42087177a9c4SMiklos Szeredi if (!old_dir->i_op->rename && !old_dir->i_op->rename2) 42091da177e4SLinus Torvalds return -EPERM; 42101da177e4SLinus Torvalds 4211520c8b16SMiklos Szeredi if (flags && !old_dir->i_op->rename2) 4212520c8b16SMiklos Szeredi return -EINVAL; 4213520c8b16SMiklos Szeredi 4214bc27027aSMiklos Szeredi /* 4215bc27027aSMiklos Szeredi * If we are going to change the parent - check write permissions, 4216bc27027aSMiklos Szeredi * we'll need to flip '..'. 4217bc27027aSMiklos Szeredi */ 4218da1ce067SMiklos Szeredi if (new_dir != old_dir) { 4219da1ce067SMiklos Szeredi if (is_dir) { 4220bc27027aSMiklos Szeredi error = inode_permission(source, MAY_WRITE); 4221bc27027aSMiklos Szeredi if (error) 4222bc27027aSMiklos Szeredi return error; 4223bc27027aSMiklos Szeredi } 4224da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && new_is_dir) { 4225da1ce067SMiklos Szeredi error = inode_permission(target, MAY_WRITE); 4226da1ce067SMiklos Szeredi if (error) 4227da1ce067SMiklos Szeredi return error; 4228da1ce067SMiklos Szeredi } 4229da1ce067SMiklos Szeredi } 42300eeca283SRobert Love 42310b3974ebSMiklos Szeredi error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry, 42320b3974ebSMiklos Szeredi flags); 4233bc27027aSMiklos Szeredi if (error) 4234bc27027aSMiklos Szeredi return error; 4235bc27027aSMiklos Szeredi 4236bc27027aSMiklos Szeredi old_name = fsnotify_oldname_init(old_dentry->d_name.name); 4237bc27027aSMiklos Szeredi dget(new_dentry); 4238da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4239bc27027aSMiklos Szeredi lock_two_nondirectories(source, target); 4240bc27027aSMiklos Szeredi else if (target) 4241bc27027aSMiklos Szeredi mutex_lock(&target->i_mutex); 4242bc27027aSMiklos Szeredi 4243bc27027aSMiklos Szeredi error = -EBUSY; 42447af1364fSEric W. Biederman if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry)) 4245bc27027aSMiklos Szeredi goto out; 4246bc27027aSMiklos Szeredi 4247da1ce067SMiklos Szeredi if (max_links && new_dir != old_dir) { 4248bc27027aSMiklos Szeredi error = -EMLINK; 4249da1ce067SMiklos Szeredi if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links) 4250bc27027aSMiklos Szeredi goto out; 4251da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir && 4252da1ce067SMiklos Szeredi old_dir->i_nlink >= max_links) 4253da1ce067SMiklos Szeredi goto out; 4254da1ce067SMiklos Szeredi } 4255da1ce067SMiklos Szeredi if (is_dir && !(flags & RENAME_EXCHANGE) && target) 4256bc27027aSMiklos Szeredi shrink_dcache_parent(new_dentry); 4257da1ce067SMiklos Szeredi if (!is_dir) { 4258bc27027aSMiklos Szeredi error = try_break_deleg(source, delegated_inode); 4259bc27027aSMiklos Szeredi if (error) 4260bc27027aSMiklos Szeredi goto out; 4261da1ce067SMiklos Szeredi } 4262da1ce067SMiklos Szeredi if (target && !new_is_dir) { 4263bc27027aSMiklos Szeredi error = try_break_deleg(target, delegated_inode); 4264bc27027aSMiklos Szeredi if (error) 4265bc27027aSMiklos Szeredi goto out; 4266bc27027aSMiklos Szeredi } 42677177a9c4SMiklos Szeredi if (!old_dir->i_op->rename2) { 4268520c8b16SMiklos Szeredi error = old_dir->i_op->rename(old_dir, old_dentry, 4269520c8b16SMiklos Szeredi new_dir, new_dentry); 4270520c8b16SMiklos Szeredi } else { 42717177a9c4SMiklos Szeredi WARN_ON(old_dir->i_op->rename != NULL); 4272520c8b16SMiklos Szeredi error = old_dir->i_op->rename2(old_dir, old_dentry, 4273520c8b16SMiklos Szeredi new_dir, new_dentry, flags); 4274520c8b16SMiklos Szeredi } 4275bc27027aSMiklos Szeredi if (error) 4276bc27027aSMiklos Szeredi goto out; 4277bc27027aSMiklos Szeredi 4278da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE) && target) { 42791da177e4SLinus Torvalds if (is_dir) 4280bc27027aSMiklos Szeredi target->i_flags |= S_DEAD; 4281bc27027aSMiklos Szeredi dont_mount(new_dentry); 42828ed936b5SEric W. Biederman detach_mounts(new_dentry); 4283bc27027aSMiklos Szeredi } 4284da1ce067SMiklos Szeredi if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) { 4285da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 4286bc27027aSMiklos Szeredi d_move(old_dentry, new_dentry); 4287da1ce067SMiklos Szeredi else 4288da1ce067SMiklos Szeredi d_exchange(old_dentry, new_dentry); 4289da1ce067SMiklos Szeredi } 4290bc27027aSMiklos Szeredi out: 4291da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4292bc27027aSMiklos Szeredi unlock_two_nondirectories(source, target); 4293bc27027aSMiklos Szeredi else if (target) 4294bc27027aSMiklos Szeredi mutex_unlock(&target->i_mutex); 4295bc27027aSMiklos Szeredi dput(new_dentry); 4296da1ce067SMiklos Szeredi if (!error) { 4297123df294SAl Viro fsnotify_move(old_dir, new_dir, old_name, is_dir, 4298da1ce067SMiklos Szeredi !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry); 4299da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4300da1ce067SMiklos Szeredi fsnotify_move(new_dir, old_dir, old_dentry->d_name.name, 4301da1ce067SMiklos Szeredi new_is_dir, NULL, new_dentry); 4302da1ce067SMiklos Szeredi } 4303da1ce067SMiklos Szeredi } 43040eeca283SRobert Love fsnotify_oldname_free(old_name); 43050eeca283SRobert Love 43061da177e4SLinus Torvalds return error; 43071da177e4SLinus Torvalds } 43084d359507SAl Viro EXPORT_SYMBOL(vfs_rename); 43091da177e4SLinus Torvalds 4310520c8b16SMiklos Szeredi SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname, 4311520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname, unsigned int, flags) 43121da177e4SLinus Torvalds { 43131da177e4SLinus Torvalds struct dentry *old_dentry, *new_dentry; 43141da177e4SLinus Torvalds struct dentry *trap; 4315f5beed75SAl Viro struct path old_path, new_path; 4316f5beed75SAl Viro struct qstr old_last, new_last; 4317f5beed75SAl Viro int old_type, new_type; 43188e6d782cSJ. Bruce Fields struct inode *delegated_inode = NULL; 431991a27b2aSJeff Layton struct filename *from; 432091a27b2aSJeff Layton struct filename *to; 4321f5beed75SAl Viro unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET; 4322c6a94284SJeff Layton bool should_retry = false; 43232ad94ae6SAl Viro int error; 4324520c8b16SMiklos Szeredi 43250d7a8555SMiklos Szeredi if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) 4326da1ce067SMiklos Szeredi return -EINVAL; 4327da1ce067SMiklos Szeredi 43280d7a8555SMiklos Szeredi if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) && 43290d7a8555SMiklos Szeredi (flags & RENAME_EXCHANGE)) 4330520c8b16SMiklos Szeredi return -EINVAL; 4331520c8b16SMiklos Szeredi 43320d7a8555SMiklos Szeredi if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD)) 43330d7a8555SMiklos Szeredi return -EPERM; 43340d7a8555SMiklos Szeredi 4335f5beed75SAl Viro if (flags & RENAME_EXCHANGE) 4336f5beed75SAl Viro target_flags = 0; 4337f5beed75SAl Viro 4338c6a94284SJeff Layton retry: 4339f5beed75SAl Viro from = user_path_parent(olddfd, oldname, 4340f5beed75SAl Viro &old_path, &old_last, &old_type, lookup_flags); 434191a27b2aSJeff Layton if (IS_ERR(from)) { 434291a27b2aSJeff Layton error = PTR_ERR(from); 43431da177e4SLinus Torvalds goto exit; 434491a27b2aSJeff Layton } 43451da177e4SLinus Torvalds 4346f5beed75SAl Viro to = user_path_parent(newdfd, newname, 4347f5beed75SAl Viro &new_path, &new_last, &new_type, lookup_flags); 434891a27b2aSJeff Layton if (IS_ERR(to)) { 434991a27b2aSJeff Layton error = PTR_ERR(to); 43501da177e4SLinus Torvalds goto exit1; 435191a27b2aSJeff Layton } 43521da177e4SLinus Torvalds 43531da177e4SLinus Torvalds error = -EXDEV; 4354f5beed75SAl Viro if (old_path.mnt != new_path.mnt) 43551da177e4SLinus Torvalds goto exit2; 43561da177e4SLinus Torvalds 43571da177e4SLinus Torvalds error = -EBUSY; 4358f5beed75SAl Viro if (old_type != LAST_NORM) 43591da177e4SLinus Torvalds goto exit2; 43601da177e4SLinus Torvalds 43610a7c3937SMiklos Szeredi if (flags & RENAME_NOREPLACE) 43620a7c3937SMiklos Szeredi error = -EEXIST; 4363f5beed75SAl Viro if (new_type != LAST_NORM) 43641da177e4SLinus Torvalds goto exit2; 43651da177e4SLinus Torvalds 4366f5beed75SAl Viro error = mnt_want_write(old_path.mnt); 4367c30dabfeSJan Kara if (error) 4368c30dabfeSJan Kara goto exit2; 4369c30dabfeSJan Kara 43708e6d782cSJ. Bruce Fields retry_deleg: 4371f5beed75SAl Viro trap = lock_rename(new_path.dentry, old_path.dentry); 43721da177e4SLinus Torvalds 4373f5beed75SAl Viro old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags); 43741da177e4SLinus Torvalds error = PTR_ERR(old_dentry); 43751da177e4SLinus Torvalds if (IS_ERR(old_dentry)) 43761da177e4SLinus Torvalds goto exit3; 43771da177e4SLinus Torvalds /* source must exist */ 43781da177e4SLinus Torvalds error = -ENOENT; 4379b18825a7SDavid Howells if (d_is_negative(old_dentry)) 43801da177e4SLinus Torvalds goto exit4; 4381f5beed75SAl Viro new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags); 43821da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 43831da177e4SLinus Torvalds if (IS_ERR(new_dentry)) 43841da177e4SLinus Torvalds goto exit4; 43850a7c3937SMiklos Szeredi error = -EEXIST; 43860a7c3937SMiklos Szeredi if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry)) 43870a7c3937SMiklos Szeredi goto exit5; 4388da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4389da1ce067SMiklos Szeredi error = -ENOENT; 4390da1ce067SMiklos Szeredi if (d_is_negative(new_dentry)) 4391da1ce067SMiklos Szeredi goto exit5; 4392da1ce067SMiklos Szeredi 4393da1ce067SMiklos Szeredi if (!d_is_dir(new_dentry)) { 4394da1ce067SMiklos Szeredi error = -ENOTDIR; 4395f5beed75SAl Viro if (new_last.name[new_last.len]) 4396da1ce067SMiklos Szeredi goto exit5; 4397da1ce067SMiklos Szeredi } 4398da1ce067SMiklos Szeredi } 43990a7c3937SMiklos Szeredi /* unless the source is a directory trailing slashes give -ENOTDIR */ 44000a7c3937SMiklos Szeredi if (!d_is_dir(old_dentry)) { 44010a7c3937SMiklos Szeredi error = -ENOTDIR; 4402f5beed75SAl Viro if (old_last.name[old_last.len]) 44030a7c3937SMiklos Szeredi goto exit5; 4404f5beed75SAl Viro if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len]) 44050a7c3937SMiklos Szeredi goto exit5; 44060a7c3937SMiklos Szeredi } 44070a7c3937SMiklos Szeredi /* source should not be ancestor of target */ 44080a7c3937SMiklos Szeredi error = -EINVAL; 44090a7c3937SMiklos Szeredi if (old_dentry == trap) 44100a7c3937SMiklos Szeredi goto exit5; 44111da177e4SLinus Torvalds /* target should not be an ancestor of source */ 4412da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 44131da177e4SLinus Torvalds error = -ENOTEMPTY; 44141da177e4SLinus Torvalds if (new_dentry == trap) 44151da177e4SLinus Torvalds goto exit5; 44161da177e4SLinus Torvalds 4417f5beed75SAl Viro error = security_path_rename(&old_path, old_dentry, 4418f5beed75SAl Viro &new_path, new_dentry, flags); 4419be6d3e56SKentaro Takeda if (error) 4420c30dabfeSJan Kara goto exit5; 4421f5beed75SAl Viro error = vfs_rename(old_path.dentry->d_inode, old_dentry, 4422f5beed75SAl Viro new_path.dentry->d_inode, new_dentry, 4423520c8b16SMiklos Szeredi &delegated_inode, flags); 44241da177e4SLinus Torvalds exit5: 44251da177e4SLinus Torvalds dput(new_dentry); 44261da177e4SLinus Torvalds exit4: 44271da177e4SLinus Torvalds dput(old_dentry); 44281da177e4SLinus Torvalds exit3: 4429f5beed75SAl Viro unlock_rename(new_path.dentry, old_path.dentry); 44308e6d782cSJ. Bruce Fields if (delegated_inode) { 44318e6d782cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 44328e6d782cSJ. Bruce Fields if (!error) 44338e6d782cSJ. Bruce Fields goto retry_deleg; 44348e6d782cSJ. Bruce Fields } 4435f5beed75SAl Viro mnt_drop_write(old_path.mnt); 44361da177e4SLinus Torvalds exit2: 4437c6a94284SJeff Layton if (retry_estale(error, lookup_flags)) 4438c6a94284SJeff Layton should_retry = true; 4439f5beed75SAl Viro path_put(&new_path); 44402ad94ae6SAl Viro putname(to); 44411da177e4SLinus Torvalds exit1: 4442f5beed75SAl Viro path_put(&old_path); 44431da177e4SLinus Torvalds putname(from); 4444c6a94284SJeff Layton if (should_retry) { 4445c6a94284SJeff Layton should_retry = false; 4446c6a94284SJeff Layton lookup_flags |= LOOKUP_REVAL; 4447c6a94284SJeff Layton goto retry; 4448c6a94284SJeff Layton } 44492ad94ae6SAl Viro exit: 44501da177e4SLinus Torvalds return error; 44511da177e4SLinus Torvalds } 44521da177e4SLinus Torvalds 4453520c8b16SMiklos Szeredi SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, 4454520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname) 4455520c8b16SMiklos Szeredi { 4456520c8b16SMiklos Szeredi return sys_renameat2(olddfd, oldname, newdfd, newname, 0); 4457520c8b16SMiklos Szeredi } 4458520c8b16SMiklos Szeredi 4459a26eab24SHeiko Carstens SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) 44605590ff0dSUlrich Drepper { 4461520c8b16SMiklos Szeredi return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 44625590ff0dSUlrich Drepper } 44635590ff0dSUlrich Drepper 4464787fb6bcSMiklos Szeredi int vfs_whiteout(struct inode *dir, struct dentry *dentry) 4465787fb6bcSMiklos Szeredi { 4466787fb6bcSMiklos Szeredi int error = may_create(dir, dentry); 4467787fb6bcSMiklos Szeredi if (error) 4468787fb6bcSMiklos Szeredi return error; 4469787fb6bcSMiklos Szeredi 4470787fb6bcSMiklos Szeredi if (!dir->i_op->mknod) 4471787fb6bcSMiklos Szeredi return -EPERM; 4472787fb6bcSMiklos Szeredi 4473787fb6bcSMiklos Szeredi return dir->i_op->mknod(dir, dentry, 4474787fb6bcSMiklos Szeredi S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV); 4475787fb6bcSMiklos Szeredi } 4476787fb6bcSMiklos Szeredi EXPORT_SYMBOL(vfs_whiteout); 4477787fb6bcSMiklos Szeredi 44785d826c84SAl Viro int readlink_copy(char __user *buffer, int buflen, const char *link) 44791da177e4SLinus Torvalds { 44805d826c84SAl Viro int len = PTR_ERR(link); 44811da177e4SLinus Torvalds if (IS_ERR(link)) 44821da177e4SLinus Torvalds goto out; 44831da177e4SLinus Torvalds 44841da177e4SLinus Torvalds len = strlen(link); 44851da177e4SLinus Torvalds if (len > (unsigned) buflen) 44861da177e4SLinus Torvalds len = buflen; 44871da177e4SLinus Torvalds if (copy_to_user(buffer, link, len)) 44881da177e4SLinus Torvalds len = -EFAULT; 44891da177e4SLinus Torvalds out: 44901da177e4SLinus Torvalds return len; 44911da177e4SLinus Torvalds } 44925d826c84SAl Viro EXPORT_SYMBOL(readlink_copy); 44931da177e4SLinus Torvalds 44941da177e4SLinus Torvalds /* 44951da177e4SLinus Torvalds * A helper for ->readlink(). This should be used *ONLY* for symlinks that 44961da177e4SLinus Torvalds * have ->follow_link() touching nd only in nd_set_link(). Using (or not 44971da177e4SLinus Torvalds * using) it for any given inode is up to filesystem. 44981da177e4SLinus Torvalds */ 44991da177e4SLinus Torvalds int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen) 45001da177e4SLinus Torvalds { 4501cc314eefSLinus Torvalds void *cookie; 45025f2c4179SAl Viro struct inode *inode = d_inode(dentry); 45035f2c4179SAl Viro const char *link = inode->i_link; 4504694a1764SMarcin Slusarz int res; 4505cc314eefSLinus Torvalds 4506d4dee48bSAl Viro if (!link) { 45075f2c4179SAl Viro link = inode->i_op->follow_link(dentry, &cookie); 4508680baacbSAl Viro if (IS_ERR(link)) 4509680baacbSAl Viro return PTR_ERR(link); 4510d4dee48bSAl Viro } 4511680baacbSAl Viro res = readlink_copy(buffer, buflen, link); 45125f2c4179SAl Viro if (inode->i_op->put_link) 45135f2c4179SAl Viro inode->i_op->put_link(inode, cookie); 4514694a1764SMarcin Slusarz return res; 45151da177e4SLinus Torvalds } 45164d359507SAl Viro EXPORT_SYMBOL(generic_readlink); 45171da177e4SLinus Torvalds 45181da177e4SLinus Torvalds /* get the link contents into pagecache */ 45191da177e4SLinus Torvalds static char *page_getlink(struct dentry * dentry, struct page **ppage) 45201da177e4SLinus Torvalds { 4521ebd09abbSDuane Griffin char *kaddr; 45221da177e4SLinus Torvalds struct page *page; 45231da177e4SLinus Torvalds struct address_space *mapping = dentry->d_inode->i_mapping; 4524090d2b18SPekka Enberg page = read_mapping_page(mapping, 0, NULL); 45251da177e4SLinus Torvalds if (IS_ERR(page)) 45266fe6900eSNick Piggin return (char*)page; 45271da177e4SLinus Torvalds *ppage = page; 4528ebd09abbSDuane Griffin kaddr = kmap(page); 4529ebd09abbSDuane Griffin nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1); 4530ebd09abbSDuane Griffin return kaddr; 45311da177e4SLinus Torvalds } 45321da177e4SLinus Torvalds 45331da177e4SLinus Torvalds int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 45341da177e4SLinus Torvalds { 45351da177e4SLinus Torvalds struct page *page = NULL; 45365d826c84SAl Viro int res = readlink_copy(buffer, buflen, page_getlink(dentry, &page)); 45371da177e4SLinus Torvalds if (page) { 45381da177e4SLinus Torvalds kunmap(page); 45391da177e4SLinus Torvalds page_cache_release(page); 45401da177e4SLinus Torvalds } 45411da177e4SLinus Torvalds return res; 45421da177e4SLinus Torvalds } 45434d359507SAl Viro EXPORT_SYMBOL(page_readlink); 45441da177e4SLinus Torvalds 45456e77137bSAl Viro const char *page_follow_link_light(struct dentry *dentry, void **cookie) 45461da177e4SLinus Torvalds { 4547cc314eefSLinus Torvalds struct page *page = NULL; 4548680baacbSAl Viro char *res = page_getlink(dentry, &page); 4549680baacbSAl Viro if (!IS_ERR(res)) 4550680baacbSAl Viro *cookie = page; 4551680baacbSAl Viro return res; 45521da177e4SLinus Torvalds } 45534d359507SAl Viro EXPORT_SYMBOL(page_follow_link_light); 45541da177e4SLinus Torvalds 45555f2c4179SAl Viro void page_put_link(struct inode *unused, void *cookie) 45561da177e4SLinus Torvalds { 4557cc314eefSLinus Torvalds struct page *page = cookie; 45581da177e4SLinus Torvalds kunmap(page); 45591da177e4SLinus Torvalds page_cache_release(page); 45601da177e4SLinus Torvalds } 45614d359507SAl Viro EXPORT_SYMBOL(page_put_link); 45621da177e4SLinus Torvalds 456354566b2cSNick Piggin /* 456454566b2cSNick Piggin * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS 456554566b2cSNick Piggin */ 456654566b2cSNick Piggin int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) 45671da177e4SLinus Torvalds { 45681da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 45690adb25d2SKirill Korotaev struct page *page; 4570afddba49SNick Piggin void *fsdata; 4571beb497abSDmitriy Monakhov int err; 45721da177e4SLinus Torvalds char *kaddr; 457354566b2cSNick Piggin unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE; 457454566b2cSNick Piggin if (nofs) 457554566b2cSNick Piggin flags |= AOP_FLAG_NOFS; 45761da177e4SLinus Torvalds 45777e53cac4SNeilBrown retry: 4578afddba49SNick Piggin err = pagecache_write_begin(NULL, mapping, 0, len-1, 457954566b2cSNick Piggin flags, &page, &fsdata); 45801da177e4SLinus Torvalds if (err) 4581afddba49SNick Piggin goto fail; 4582afddba49SNick Piggin 4583e8e3c3d6SCong Wang kaddr = kmap_atomic(page); 45841da177e4SLinus Torvalds memcpy(kaddr, symname, len-1); 4585e8e3c3d6SCong Wang kunmap_atomic(kaddr); 4586afddba49SNick Piggin 4587afddba49SNick Piggin err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, 4588afddba49SNick Piggin page, fsdata); 45891da177e4SLinus Torvalds if (err < 0) 45901da177e4SLinus Torvalds goto fail; 4591afddba49SNick Piggin if (err < len-1) 4592afddba49SNick Piggin goto retry; 4593afddba49SNick Piggin 45941da177e4SLinus Torvalds mark_inode_dirty(inode); 45951da177e4SLinus Torvalds return 0; 45961da177e4SLinus Torvalds fail: 45971da177e4SLinus Torvalds return err; 45981da177e4SLinus Torvalds } 45994d359507SAl Viro EXPORT_SYMBOL(__page_symlink); 46001da177e4SLinus Torvalds 46010adb25d2SKirill Korotaev int page_symlink(struct inode *inode, const char *symname, int len) 46020adb25d2SKirill Korotaev { 46030adb25d2SKirill Korotaev return __page_symlink(inode, symname, len, 4604c62d2555SMichal Hocko !mapping_gfp_constraint(inode->i_mapping, __GFP_FS)); 46050adb25d2SKirill Korotaev } 46064d359507SAl Viro EXPORT_SYMBOL(page_symlink); 46070adb25d2SKirill Korotaev 460892e1d5beSArjan van de Ven const struct inode_operations page_symlink_inode_operations = { 46091da177e4SLinus Torvalds .readlink = generic_readlink, 46101da177e4SLinus Torvalds .follow_link = page_follow_link_light, 46111da177e4SLinus Torvalds .put_link = page_put_link, 46121da177e4SLinus Torvalds }; 46131da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink_inode_operations); 4614