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 66019660af7SAl Viro * normal reference counts on dentries and vfsmounts to transition to rcu-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 { 810f7ad3c6bSMiklos Szeredi get_fs_root(current->fs, &nd->root); 8112a737871SAl Viro } 8122a737871SAl Viro 81306d7137eSAl Viro static void set_root_rcu(struct nameidata *nd) 81431e6b01fSNick Piggin { 81531e6b01fSNick Piggin struct fs_struct *fs = current->fs; 8168f47a016SAl Viro unsigned seq; 817c28cc364SNick Piggin 818c28cc364SNick Piggin do { 819c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 82031e6b01fSNick Piggin nd->root = fs->root; 8218f47a016SAl Viro nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq); 822c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 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 /* 8456b255391SAl Viro * Helper to directly jump to a known parsed path from ->get_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) { 10086b255391SAl Viro res = inode->i_op->get_link(NULL, inode, 10096b255391SAl Viro &last->cookie); 10106b255391SAl Viro if (res == ERR_PTR(-ECHILD)) { 10118c1b4566SAl Viro if (unlikely(unlazy_walk(nd, NULL, 0))) 10128c1b4566SAl Viro return ERR_PTR(-ECHILD); 10136b255391SAl Viro res = inode->i_op->get_link(dentry, inode, 10146b255391SAl Viro &last->cookie); 10158c1b4566SAl Viro } 10166b255391SAl Viro } else { 10176b255391SAl Viro res = inode->i_op->get_link(dentry, inode, 10186b255391SAl Viro &last->cookie); 10196b255391SAl Viro } 1020fab51e8aSAl Viro if (IS_ERR_OR_NULL(res)) { 10216920a440SAl Viro last->cookie = NULL; 1022fab51e8aSAl Viro return res; 10230a959df5SAl Viro } 1024fab51e8aSAl Viro } 1025fab51e8aSAl Viro if (*res == '/') { 10268c1b4566SAl Viro if (nd->flags & LOOKUP_RCU) { 10278f47a016SAl Viro struct dentry *d; 10288f47a016SAl Viro if (!nd->root.mnt) 10298f47a016SAl Viro set_root_rcu(nd); 10308f47a016SAl Viro nd->path = nd->root; 10318f47a016SAl Viro d = nd->path.dentry; 10328f47a016SAl Viro nd->inode = d->d_inode; 10338f47a016SAl Viro nd->seq = nd->root_seq; 10348f47a016SAl Viro if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq))) 10358c1b4566SAl Viro return ERR_PTR(-ECHILD); 10368f47a016SAl Viro } else { 1037fab51e8aSAl Viro if (!nd->root.mnt) 1038fab51e8aSAl Viro set_root(nd); 1039fab51e8aSAl Viro path_put(&nd->path); 1040fab51e8aSAl Viro nd->path = nd->root; 1041fab51e8aSAl Viro path_get(&nd->root); 1042fab51e8aSAl Viro nd->inode = nd->path.dentry->d_inode; 10438f47a016SAl Viro } 1044fab51e8aSAl Viro nd->flags |= LOOKUP_JUMPED; 1045fab51e8aSAl Viro while (unlikely(*++res == '/')) 1046fab51e8aSAl Viro ; 1047fab51e8aSAl Viro } 1048fab51e8aSAl Viro if (!*res) 1049fab51e8aSAl Viro res = NULL; 10500a959df5SAl Viro return res; 10510a959df5SAl Viro } 10526d7b5aaeSAl Viro 1053f015f126SDavid Howells /* 1054f015f126SDavid Howells * follow_up - Find the mountpoint of path's vfsmount 1055f015f126SDavid Howells * 1056f015f126SDavid Howells * Given a path, find the mountpoint of its source file system. 1057f015f126SDavid Howells * Replace @path with the path of the mountpoint in the parent mount. 1058f015f126SDavid Howells * Up is towards /. 1059f015f126SDavid Howells * 1060f015f126SDavid Howells * Return 1 if we went up a level and 0 if we were already at the 1061f015f126SDavid Howells * root. 1062f015f126SDavid Howells */ 1063bab77ebfSAl Viro int follow_up(struct path *path) 10641da177e4SLinus Torvalds { 10650714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 10660714a533SAl Viro struct mount *parent; 10671da177e4SLinus Torvalds struct dentry *mountpoint; 106899b7db7bSNick Piggin 106948a066e7SAl Viro read_seqlock_excl(&mount_lock); 10700714a533SAl Viro parent = mnt->mnt_parent; 10713c0a6163SAl Viro if (parent == mnt) { 107248a066e7SAl Viro read_sequnlock_excl(&mount_lock); 10731da177e4SLinus Torvalds return 0; 10741da177e4SLinus Torvalds } 10750714a533SAl Viro mntget(&parent->mnt); 1076a73324daSAl Viro mountpoint = dget(mnt->mnt_mountpoint); 107748a066e7SAl Viro read_sequnlock_excl(&mount_lock); 1078bab77ebfSAl Viro dput(path->dentry); 1079bab77ebfSAl Viro path->dentry = mountpoint; 1080bab77ebfSAl Viro mntput(path->mnt); 10810714a533SAl Viro path->mnt = &parent->mnt; 10821da177e4SLinus Torvalds return 1; 10831da177e4SLinus Torvalds } 10844d359507SAl Viro EXPORT_SYMBOL(follow_up); 10851da177e4SLinus Torvalds 1086b5c84bf6SNick Piggin /* 10879875cf80SDavid Howells * Perform an automount 10889875cf80SDavid Howells * - return -EISDIR to tell follow_managed() to stop and return the path we 10899875cf80SDavid Howells * were called with. 10901da177e4SLinus Torvalds */ 1091756daf26SNeilBrown static int follow_automount(struct path *path, struct nameidata *nd, 10929875cf80SDavid Howells bool *need_mntput) 109331e6b01fSNick Piggin { 10949875cf80SDavid Howells struct vfsmount *mnt; 1095ea5b778aSDavid Howells int err; 10969875cf80SDavid Howells 10979875cf80SDavid Howells if (!path->dentry->d_op || !path->dentry->d_op->d_automount) 10989875cf80SDavid Howells return -EREMOTE; 10999875cf80SDavid Howells 11000ec26fd0SMiklos Szeredi /* We don't want to mount if someone's just doing a stat - 11010ec26fd0SMiklos Szeredi * unless they're stat'ing a directory and appended a '/' to 11020ec26fd0SMiklos Szeredi * the name. 11030ec26fd0SMiklos Szeredi * 11040ec26fd0SMiklos Szeredi * We do, however, want to mount if someone wants to open or 11055a30d8a2SDavid Howells * create a file of any type under the mountpoint, wants to 11060ec26fd0SMiklos Szeredi * traverse through the mountpoint or wants to open the 11070ec26fd0SMiklos Szeredi * mounted directory. Also, autofs may mark negative dentries 11080ec26fd0SMiklos Szeredi * as being automount points. These will need the attentions 11090ec26fd0SMiklos Szeredi * of the daemon to instantiate them before they can be used. 11105a30d8a2SDavid Howells */ 1111756daf26SNeilBrown if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | 1112d94c177bSLinus Torvalds LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && 11135a30d8a2SDavid Howells path->dentry->d_inode) 11149875cf80SDavid Howells return -EISDIR; 11150ec26fd0SMiklos Szeredi 1116756daf26SNeilBrown nd->total_link_count++; 1117756daf26SNeilBrown if (nd->total_link_count >= 40) 11189875cf80SDavid Howells return -ELOOP; 11199875cf80SDavid Howells 11209875cf80SDavid Howells mnt = path->dentry->d_op->d_automount(path); 11219875cf80SDavid Howells if (IS_ERR(mnt)) { 11229875cf80SDavid Howells /* 11239875cf80SDavid Howells * The filesystem is allowed to return -EISDIR here to indicate 11249875cf80SDavid Howells * it doesn't want to automount. For instance, autofs would do 11259875cf80SDavid Howells * this so that its userspace daemon can mount on this dentry. 11269875cf80SDavid Howells * 11279875cf80SDavid Howells * However, we can only permit this if it's a terminal point in 11289875cf80SDavid Howells * the path being looked up; if it wasn't then the remainder of 11299875cf80SDavid Howells * the path is inaccessible and we should say so. 11309875cf80SDavid Howells */ 1131756daf26SNeilBrown if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT)) 11329875cf80SDavid Howells return -EREMOTE; 11339875cf80SDavid Howells return PTR_ERR(mnt); 113431e6b01fSNick Piggin } 1135ea5b778aSDavid Howells 11369875cf80SDavid Howells if (!mnt) /* mount collision */ 11379875cf80SDavid Howells return 0; 11389875cf80SDavid Howells 11398aef1884SAl Viro if (!*need_mntput) { 11408aef1884SAl Viro /* lock_mount() may release path->mnt on error */ 11418aef1884SAl Viro mntget(path->mnt); 11428aef1884SAl Viro *need_mntput = true; 11438aef1884SAl Viro } 114419a167afSAl Viro err = finish_automount(mnt, path); 1145ea5b778aSDavid Howells 1146ea5b778aSDavid Howells switch (err) { 1147ea5b778aSDavid Howells case -EBUSY: 1148ea5b778aSDavid Howells /* Someone else made a mount here whilst we were busy */ 114919a167afSAl Viro return 0; 1150ea5b778aSDavid Howells case 0: 11518aef1884SAl Viro path_put(path); 11529875cf80SDavid Howells path->mnt = mnt; 11539875cf80SDavid Howells path->dentry = dget(mnt->mnt_root); 11549875cf80SDavid Howells return 0; 115519a167afSAl Viro default: 115619a167afSAl Viro return err; 11579875cf80SDavid Howells } 115819a167afSAl Viro 1159ea5b778aSDavid Howells } 11609875cf80SDavid Howells 11619875cf80SDavid Howells /* 11629875cf80SDavid Howells * Handle a dentry that is managed in some way. 1163cc53ce53SDavid Howells * - Flagged for transit management (autofs) 11649875cf80SDavid Howells * - Flagged as mountpoint 11659875cf80SDavid Howells * - Flagged as automount point 11669875cf80SDavid Howells * 11679875cf80SDavid Howells * This may only be called in refwalk mode. 11689875cf80SDavid Howells * 11699875cf80SDavid Howells * Serialization is taken care of in namespace.c 11709875cf80SDavid Howells */ 1171756daf26SNeilBrown static int follow_managed(struct path *path, struct nameidata *nd) 11729875cf80SDavid Howells { 11738aef1884SAl Viro struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */ 11749875cf80SDavid Howells unsigned managed; 11759875cf80SDavid Howells bool need_mntput = false; 11768aef1884SAl Viro int ret = 0; 11779875cf80SDavid Howells 11789875cf80SDavid Howells /* Given that we're not holding a lock here, we retain the value in a 11799875cf80SDavid Howells * local variable for each dentry as we look at it so that we don't see 11809875cf80SDavid Howells * the components of that value change under us */ 11819875cf80SDavid Howells while (managed = ACCESS_ONCE(path->dentry->d_flags), 11829875cf80SDavid Howells managed &= DCACHE_MANAGED_DENTRY, 11839875cf80SDavid Howells unlikely(managed != 0)) { 1184cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1185cc53ce53SDavid Howells * being held. */ 1186cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 1187cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1188cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 11891aed3e42SAl Viro ret = path->dentry->d_op->d_manage(path->dentry, false); 1190cc53ce53SDavid Howells if (ret < 0) 11918aef1884SAl Viro break; 1192cc53ce53SDavid Howells } 1193cc53ce53SDavid Howells 11949875cf80SDavid Howells /* Transit to a mounted filesystem. */ 11959875cf80SDavid Howells if (managed & DCACHE_MOUNTED) { 11969875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 11979875cf80SDavid Howells if (mounted) { 11989875cf80SDavid Howells dput(path->dentry); 11999875cf80SDavid Howells if (need_mntput) 1200463ffb2eSAl Viro mntput(path->mnt); 1201463ffb2eSAl Viro path->mnt = mounted; 1202463ffb2eSAl Viro path->dentry = dget(mounted->mnt_root); 12039875cf80SDavid Howells need_mntput = true; 12049875cf80SDavid Howells continue; 1205463ffb2eSAl Viro } 1206463ffb2eSAl Viro 12079875cf80SDavid Howells /* Something is mounted on this dentry in another 12089875cf80SDavid Howells * namespace and/or whatever was mounted there in this 120948a066e7SAl Viro * namespace got unmounted before lookup_mnt() could 121048a066e7SAl Viro * get it */ 12111da177e4SLinus Torvalds } 12129875cf80SDavid Howells 12139875cf80SDavid Howells /* Handle an automount point */ 12149875cf80SDavid Howells if (managed & DCACHE_NEED_AUTOMOUNT) { 1215756daf26SNeilBrown ret = follow_automount(path, nd, &need_mntput); 12169875cf80SDavid Howells if (ret < 0) 12178aef1884SAl Viro break; 12189875cf80SDavid Howells continue; 12199875cf80SDavid Howells } 12209875cf80SDavid Howells 12219875cf80SDavid Howells /* We didn't change the current path point */ 12229875cf80SDavid Howells break; 12239875cf80SDavid Howells } 12248aef1884SAl Viro 12258aef1884SAl Viro if (need_mntput && path->mnt == mnt) 12268aef1884SAl Viro mntput(path->mnt); 12278aef1884SAl Viro if (ret == -EISDIR) 12288aef1884SAl Viro ret = 0; 12298402752eSAl Viro if (need_mntput) 12308402752eSAl Viro nd->flags |= LOOKUP_JUMPED; 12318402752eSAl Viro if (unlikely(ret < 0)) 12328402752eSAl Viro path_put_conditional(path, nd); 12338402752eSAl Viro return ret; 12341da177e4SLinus Torvalds } 12351da177e4SLinus Torvalds 1236cc53ce53SDavid Howells int follow_down_one(struct path *path) 12371da177e4SLinus Torvalds { 12381da177e4SLinus Torvalds struct vfsmount *mounted; 12391da177e4SLinus Torvalds 12401c755af4SAl Viro mounted = lookup_mnt(path); 12411da177e4SLinus Torvalds if (mounted) { 12429393bd07SAl Viro dput(path->dentry); 12439393bd07SAl Viro mntput(path->mnt); 12449393bd07SAl Viro path->mnt = mounted; 12459393bd07SAl Viro path->dentry = dget(mounted->mnt_root); 12461da177e4SLinus Torvalds return 1; 12471da177e4SLinus Torvalds } 12481da177e4SLinus Torvalds return 0; 12491da177e4SLinus Torvalds } 12504d359507SAl Viro EXPORT_SYMBOL(follow_down_one); 12511da177e4SLinus Torvalds 1252b8faf035SNeilBrown static inline int managed_dentry_rcu(struct dentry *dentry) 125362a7375eSIan Kent { 1254b8faf035SNeilBrown return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ? 1255b8faf035SNeilBrown dentry->d_op->d_manage(dentry, true) : 0; 125662a7375eSIan Kent } 125762a7375eSIan Kent 12589875cf80SDavid Howells /* 1259287548e4SAl Viro * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if 1260287548e4SAl Viro * we meet a managed dentry that would need blocking. 12619875cf80SDavid Howells */ 12629875cf80SDavid Howells static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, 1263254cf582SAl Viro struct inode **inode, unsigned *seqp) 12649875cf80SDavid Howells { 126562a7375eSIan Kent for (;;) { 1266c7105365SAl Viro struct mount *mounted; 126762a7375eSIan Kent /* 126862a7375eSIan Kent * Don't forget we might have a non-mountpoint managed dentry 126962a7375eSIan Kent * that wants to block transit. 127062a7375eSIan Kent */ 1271b8faf035SNeilBrown switch (managed_dentry_rcu(path->dentry)) { 1272b8faf035SNeilBrown case -ECHILD: 1273b8faf035SNeilBrown default: 1274ab90911fSDavid Howells return false; 1275b8faf035SNeilBrown case -EISDIR: 1276b8faf035SNeilBrown return true; 1277b8faf035SNeilBrown case 0: 1278b8faf035SNeilBrown break; 1279b8faf035SNeilBrown } 128062a7375eSIan Kent 128162a7375eSIan Kent if (!d_mountpoint(path->dentry)) 1282b8faf035SNeilBrown return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); 128362a7375eSIan Kent 1284474279dcSAl Viro mounted = __lookup_mnt(path->mnt, path->dentry); 12859875cf80SDavid Howells if (!mounted) 12869875cf80SDavid Howells break; 1287c7105365SAl Viro path->mnt = &mounted->mnt; 1288c7105365SAl Viro path->dentry = mounted->mnt.mnt_root; 1289a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 1290254cf582SAl Viro *seqp = read_seqcount_begin(&path->dentry->d_seq); 129159430262SLinus Torvalds /* 129259430262SLinus Torvalds * Update the inode too. We don't need to re-check the 129359430262SLinus Torvalds * dentry sequence number here after this d_inode read, 129459430262SLinus Torvalds * because a mount-point is always pinned. 129559430262SLinus Torvalds */ 129659430262SLinus Torvalds *inode = path->dentry->d_inode; 12979875cf80SDavid Howells } 1298f5be3e29SAl Viro return !read_seqretry(&mount_lock, nd->m_seq) && 1299b8faf035SNeilBrown !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); 1300287548e4SAl Viro } 1301287548e4SAl Viro 130231e6b01fSNick Piggin static int follow_dotdot_rcu(struct nameidata *nd) 130331e6b01fSNick Piggin { 13044023bfc9SAl Viro struct inode *inode = nd->inode; 13057bd88377SAl Viro if (!nd->root.mnt) 130631e6b01fSNick Piggin set_root_rcu(nd); 130731e6b01fSNick Piggin 130831e6b01fSNick Piggin while (1) { 1309aed434adSAl Viro if (path_equal(&nd->path, &nd->root)) 131031e6b01fSNick Piggin break; 131131e6b01fSNick Piggin if (nd->path.dentry != nd->path.mnt->mnt_root) { 131231e6b01fSNick Piggin struct dentry *old = nd->path.dentry; 131331e6b01fSNick Piggin struct dentry *parent = old->d_parent; 131431e6b01fSNick Piggin unsigned seq; 131531e6b01fSNick Piggin 13164023bfc9SAl Viro inode = parent->d_inode; 131731e6b01fSNick Piggin seq = read_seqcount_begin(&parent->d_seq); 1318aed434adSAl Viro if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq))) 1319aed434adSAl Viro return -ECHILD; 132031e6b01fSNick Piggin nd->path.dentry = parent; 132131e6b01fSNick Piggin nd->seq = seq; 1322397d425dSEric W. Biederman if (unlikely(!path_connected(&nd->path))) 1323397d425dSEric W. Biederman return -ENOENT; 132431e6b01fSNick Piggin break; 1325aed434adSAl Viro } else { 1326aed434adSAl Viro struct mount *mnt = real_mount(nd->path.mnt); 1327aed434adSAl Viro struct mount *mparent = mnt->mnt_parent; 1328aed434adSAl Viro struct dentry *mountpoint = mnt->mnt_mountpoint; 1329aed434adSAl Viro struct inode *inode2 = mountpoint->d_inode; 1330aed434adSAl Viro unsigned seq = read_seqcount_begin(&mountpoint->d_seq); 1331aed434adSAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1332aed434adSAl Viro return -ECHILD; 1333aed434adSAl Viro if (&mparent->mnt == nd->path.mnt) 133431e6b01fSNick Piggin break; 1335aed434adSAl Viro /* we know that mountpoint was pinned */ 1336aed434adSAl Viro nd->path.dentry = mountpoint; 1337aed434adSAl Viro nd->path.mnt = &mparent->mnt; 1338aed434adSAl Viro inode = inode2; 1339aed434adSAl Viro nd->seq = seq; 134031e6b01fSNick Piggin } 1341aed434adSAl Viro } 1342aed434adSAl Viro while (unlikely(d_mountpoint(nd->path.dentry))) { 1343b37199e6SAl Viro struct mount *mounted; 1344b37199e6SAl Viro mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry); 1345aed434adSAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1346aed434adSAl Viro return -ECHILD; 1347b37199e6SAl Viro if (!mounted) 1348b37199e6SAl Viro break; 1349b37199e6SAl Viro nd->path.mnt = &mounted->mnt; 1350b37199e6SAl Viro nd->path.dentry = mounted->mnt.mnt_root; 13514023bfc9SAl Viro inode = nd->path.dentry->d_inode; 1352b37199e6SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 1353b37199e6SAl Viro } 13544023bfc9SAl Viro nd->inode = inode; 135531e6b01fSNick Piggin return 0; 135631e6b01fSNick Piggin } 135731e6b01fSNick Piggin 13589875cf80SDavid Howells /* 1359cc53ce53SDavid Howells * Follow down to the covering mount currently visible to userspace. At each 1360cc53ce53SDavid Howells * point, the filesystem owning that dentry may be queried as to whether the 1361cc53ce53SDavid Howells * caller is permitted to proceed or not. 1362cc53ce53SDavid Howells */ 13637cc90cc3SAl Viro int follow_down(struct path *path) 1364cc53ce53SDavid Howells { 1365cc53ce53SDavid Howells unsigned managed; 1366cc53ce53SDavid Howells int ret; 1367cc53ce53SDavid Howells 1368cc53ce53SDavid Howells while (managed = ACCESS_ONCE(path->dentry->d_flags), 1369cc53ce53SDavid Howells unlikely(managed & DCACHE_MANAGED_DENTRY)) { 1370cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1371cc53ce53SDavid Howells * being held. 1372cc53ce53SDavid Howells * 1373cc53ce53SDavid Howells * We indicate to the filesystem if someone is trying to mount 1374cc53ce53SDavid Howells * something here. This gives autofs the chance to deny anyone 1375cc53ce53SDavid Howells * other than its daemon the right to mount on its 1376cc53ce53SDavid Howells * superstructure. 1377cc53ce53SDavid Howells * 1378cc53ce53SDavid Howells * The filesystem may sleep at this point. 1379cc53ce53SDavid Howells */ 1380cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 1381cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1382cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 1383ab90911fSDavid Howells ret = path->dentry->d_op->d_manage( 13841aed3e42SAl Viro path->dentry, false); 1385cc53ce53SDavid Howells if (ret < 0) 1386cc53ce53SDavid Howells return ret == -EISDIR ? 0 : ret; 1387cc53ce53SDavid Howells } 1388cc53ce53SDavid Howells 1389cc53ce53SDavid Howells /* Transit to a mounted filesystem. */ 1390cc53ce53SDavid Howells if (managed & DCACHE_MOUNTED) { 1391cc53ce53SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 1392cc53ce53SDavid Howells if (!mounted) 1393cc53ce53SDavid Howells break; 1394cc53ce53SDavid Howells dput(path->dentry); 1395cc53ce53SDavid Howells mntput(path->mnt); 1396cc53ce53SDavid Howells path->mnt = mounted; 1397cc53ce53SDavid Howells path->dentry = dget(mounted->mnt_root); 1398cc53ce53SDavid Howells continue; 1399cc53ce53SDavid Howells } 1400cc53ce53SDavid Howells 1401cc53ce53SDavid Howells /* Don't handle automount points here */ 1402cc53ce53SDavid Howells break; 1403cc53ce53SDavid Howells } 1404cc53ce53SDavid Howells return 0; 1405cc53ce53SDavid Howells } 14064d359507SAl Viro EXPORT_SYMBOL(follow_down); 1407cc53ce53SDavid Howells 1408cc53ce53SDavid Howells /* 14099875cf80SDavid Howells * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() 14109875cf80SDavid Howells */ 14119875cf80SDavid Howells static void follow_mount(struct path *path) 14129875cf80SDavid Howells { 14139875cf80SDavid Howells while (d_mountpoint(path->dentry)) { 14149875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 14159875cf80SDavid Howells if (!mounted) 14169875cf80SDavid Howells break; 14179875cf80SDavid Howells dput(path->dentry); 14189875cf80SDavid Howells mntput(path->mnt); 14199875cf80SDavid Howells path->mnt = mounted; 14209875cf80SDavid Howells path->dentry = dget(mounted->mnt_root); 14219875cf80SDavid Howells } 14229875cf80SDavid Howells } 14239875cf80SDavid Howells 1424397d425dSEric W. Biederman static int follow_dotdot(struct nameidata *nd) 14251da177e4SLinus Torvalds { 14267bd88377SAl Viro if (!nd->root.mnt) 14272a737871SAl Viro set_root(nd); 1428e518ddb7SAndreas Mohr 14291da177e4SLinus Torvalds while(1) { 14304ac91378SJan Blunck struct dentry *old = nd->path.dentry; 14311da177e4SLinus Torvalds 14322a737871SAl Viro if (nd->path.dentry == nd->root.dentry && 14332a737871SAl Viro nd->path.mnt == nd->root.mnt) { 14341da177e4SLinus Torvalds break; 14351da177e4SLinus Torvalds } 14364ac91378SJan Blunck if (nd->path.dentry != nd->path.mnt->mnt_root) { 14373088dd70SAl Viro /* rare case of legitimate dget_parent()... */ 14383088dd70SAl Viro nd->path.dentry = dget_parent(nd->path.dentry); 14391da177e4SLinus Torvalds dput(old); 1440397d425dSEric W. Biederman if (unlikely(!path_connected(&nd->path))) 1441397d425dSEric W. Biederman return -ENOENT; 14421da177e4SLinus Torvalds break; 14431da177e4SLinus Torvalds } 14443088dd70SAl Viro if (!follow_up(&nd->path)) 14451da177e4SLinus Torvalds break; 14461da177e4SLinus Torvalds } 144779ed0226SAl Viro follow_mount(&nd->path); 144831e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 1449397d425dSEric W. Biederman return 0; 14501da177e4SLinus Torvalds } 14511da177e4SLinus Torvalds 14521da177e4SLinus Torvalds /* 1453bad61189SMiklos Szeredi * This looks up the name in dcache, possibly revalidates the old dentry and 1454bad61189SMiklos Szeredi * allocates a new one if not found or not valid. In the need_lookup argument 1455bad61189SMiklos Szeredi * returns whether i_op->lookup is necessary. 1456bad61189SMiklos Szeredi * 1457bad61189SMiklos Szeredi * dir->d_inode->i_mutex must be held 1458baa03890SNick Piggin */ 1459bad61189SMiklos Szeredi static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir, 1460201f956eSAl Viro unsigned int flags, bool *need_lookup) 1461baa03890SNick Piggin { 1462baa03890SNick Piggin struct dentry *dentry; 1463bad61189SMiklos Szeredi int error; 1464baa03890SNick Piggin 1465bad61189SMiklos Szeredi *need_lookup = false; 1466bad61189SMiklos Szeredi dentry = d_lookup(dir, name); 1467bad61189SMiklos Szeredi if (dentry) { 146839e3c955SJeff Layton if (dentry->d_flags & DCACHE_OP_REVALIDATE) { 1469201f956eSAl Viro error = d_revalidate(dentry, flags); 1470bad61189SMiklos Szeredi if (unlikely(error <= 0)) { 1471bad61189SMiklos Szeredi if (error < 0) { 1472bad61189SMiklos Szeredi dput(dentry); 1473bad61189SMiklos Szeredi return ERR_PTR(error); 14745542aa2fSEric W. Biederman } else { 14755542aa2fSEric W. Biederman d_invalidate(dentry); 1476bad61189SMiklos Szeredi dput(dentry); 1477bad61189SMiklos Szeredi dentry = NULL; 1478bad61189SMiklos Szeredi } 1479bad61189SMiklos Szeredi } 1480bad61189SMiklos Szeredi } 1481bad61189SMiklos Szeredi } 1482baa03890SNick Piggin 1483bad61189SMiklos Szeredi if (!dentry) { 1484bad61189SMiklos Szeredi dentry = d_alloc(dir, name); 1485baa03890SNick Piggin if (unlikely(!dentry)) 1486baa03890SNick Piggin return ERR_PTR(-ENOMEM); 1487baa03890SNick Piggin 1488bad61189SMiklos Szeredi *need_lookup = true; 1489baa03890SNick Piggin } 1490baa03890SNick Piggin return dentry; 1491baa03890SNick Piggin } 1492baa03890SNick Piggin 1493baa03890SNick Piggin /* 149413a2c3beSJ. Bruce Fields * Call i_op->lookup on the dentry. The dentry must be negative and 149513a2c3beSJ. Bruce Fields * unhashed. 1496bad61189SMiklos Szeredi * 1497bad61189SMiklos Szeredi * dir->d_inode->i_mutex must be held 149844396f4bSJosef Bacik */ 1499bad61189SMiklos Szeredi static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry, 150072bd866aSAl Viro unsigned int flags) 150144396f4bSJosef Bacik { 150244396f4bSJosef Bacik struct dentry *old; 150344396f4bSJosef Bacik 150444396f4bSJosef Bacik /* Don't create child dentry for a dead directory. */ 1505bad61189SMiklos Szeredi if (unlikely(IS_DEADDIR(dir))) { 1506e188dc02SMiklos Szeredi dput(dentry); 150744396f4bSJosef Bacik return ERR_PTR(-ENOENT); 1508e188dc02SMiklos Szeredi } 150944396f4bSJosef Bacik 151072bd866aSAl Viro old = dir->i_op->lookup(dir, dentry, flags); 151144396f4bSJosef Bacik if (unlikely(old)) { 151244396f4bSJosef Bacik dput(dentry); 151344396f4bSJosef Bacik dentry = old; 151444396f4bSJosef Bacik } 151544396f4bSJosef Bacik return dentry; 151644396f4bSJosef Bacik } 151744396f4bSJosef Bacik 1518a3255546SAl Viro static struct dentry *__lookup_hash(struct qstr *name, 151972bd866aSAl Viro struct dentry *base, unsigned int flags) 1520a3255546SAl Viro { 1521bad61189SMiklos Szeredi bool need_lookup; 1522a3255546SAl Viro struct dentry *dentry; 1523a3255546SAl Viro 152472bd866aSAl Viro dentry = lookup_dcache(name, base, flags, &need_lookup); 1525bad61189SMiklos Szeredi if (!need_lookup) 1526a3255546SAl Viro return dentry; 1527bad61189SMiklos Szeredi 152872bd866aSAl Viro return lookup_real(base->d_inode, dentry, flags); 1529a3255546SAl Viro } 1530a3255546SAl Viro 153144396f4bSJosef Bacik /* 15321da177e4SLinus Torvalds * It's more convoluted than I'd like it to be, but... it's still fairly 15331da177e4SLinus Torvalds * small and for now I'd prefer to have fast path as straight as possible. 15341da177e4SLinus Torvalds * It _is_ time-critical. 15351da177e4SLinus Torvalds */ 1536e97cdc87SAl Viro static int lookup_fast(struct nameidata *nd, 1537254cf582SAl Viro struct path *path, struct inode **inode, 1538254cf582SAl Viro unsigned *seqp) 15391da177e4SLinus Torvalds { 15404ac91378SJan Blunck struct vfsmount *mnt = nd->path.mnt; 154131e6b01fSNick Piggin struct dentry *dentry, *parent = nd->path.dentry; 15425a18fff2SAl Viro int need_reval = 1; 15435a18fff2SAl Viro int status = 1; 15449875cf80SDavid Howells int err; 15459875cf80SDavid Howells 15463cac260aSAl Viro /* 1547b04f784eSNick Piggin * Rename seqlock is not required here because in the off chance 1548b04f784eSNick Piggin * of a false negative due to a concurrent rename, we're going to 1549b04f784eSNick Piggin * do the non-racy lookup, below. 1550b04f784eSNick Piggin */ 155131e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) { 155231e6b01fSNick Piggin unsigned seq; 1553766c4cbfSAl Viro bool negative; 1554da53be12SLinus Torvalds dentry = __d_lookup_rcu(parent, &nd->last, &seq); 15555a18fff2SAl Viro if (!dentry) 15565a18fff2SAl Viro goto unlazy; 15575a18fff2SAl Viro 155812f8ad4bSLinus Torvalds /* 155912f8ad4bSLinus Torvalds * This sequence count validates that the inode matches 156012f8ad4bSLinus Torvalds * the dentry name information from lookup. 156112f8ad4bSLinus Torvalds */ 156263afdfc7SDavid Howells *inode = d_backing_inode(dentry); 1563766c4cbfSAl Viro negative = d_is_negative(dentry); 156412f8ad4bSLinus Torvalds if (read_seqcount_retry(&dentry->d_seq, seq)) 156512f8ad4bSLinus Torvalds return -ECHILD; 156612f8ad4bSLinus Torvalds 156712f8ad4bSLinus Torvalds /* 156812f8ad4bSLinus Torvalds * This sequence count validates that the parent had no 156912f8ad4bSLinus Torvalds * changes while we did the lookup of the dentry above. 157012f8ad4bSLinus Torvalds * 157112f8ad4bSLinus Torvalds * The memory barrier in read_seqcount_begin of child is 157212f8ad4bSLinus Torvalds * enough, we can use __read_seqcount_retry here. 157312f8ad4bSLinus Torvalds */ 157431e6b01fSNick Piggin if (__read_seqcount_retry(&parent->d_seq, nd->seq)) 157531e6b01fSNick Piggin return -ECHILD; 15765a18fff2SAl Viro 1577254cf582SAl Viro *seqp = seq; 157824643087SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) { 15794ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 15805a18fff2SAl Viro if (unlikely(status <= 0)) { 15815a18fff2SAl Viro if (status != -ECHILD) 15825a18fff2SAl Viro need_reval = 0; 15835a18fff2SAl Viro goto unlazy; 15845a18fff2SAl Viro } 158524643087SAl Viro } 1586daf3761cSTrond Myklebust /* 1587daf3761cSTrond Myklebust * Note: do negative dentry check after revalidation in 1588daf3761cSTrond Myklebust * case that drops it. 1589daf3761cSTrond Myklebust */ 1590daf3761cSTrond Myklebust if (negative) 1591daf3761cSTrond Myklebust return -ENOENT; 159231e6b01fSNick Piggin path->mnt = mnt; 159331e6b01fSNick Piggin path->dentry = dentry; 1594254cf582SAl Viro if (likely(__follow_mount_rcu(nd, path, inode, seqp))) 15959875cf80SDavid Howells return 0; 15965a18fff2SAl Viro unlazy: 1597254cf582SAl Viro if (unlazy_walk(nd, dentry, seq)) 15985a18fff2SAl Viro return -ECHILD; 15995a18fff2SAl Viro } else { 1600e97cdc87SAl Viro dentry = __d_lookup(parent, &nd->last); 160124643087SAl Viro } 16025a18fff2SAl Viro 160381e6f520SAl Viro if (unlikely(!dentry)) 160481e6f520SAl Viro goto need_lookup; 16055a18fff2SAl Viro 16065a18fff2SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval) 16074ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 16085a18fff2SAl Viro if (unlikely(status <= 0)) { 16095a18fff2SAl Viro if (status < 0) { 16105a18fff2SAl Viro dput(dentry); 16115a18fff2SAl Viro return status; 16125a18fff2SAl Viro } 16135542aa2fSEric W. Biederman d_invalidate(dentry); 16145a18fff2SAl Viro dput(dentry); 161581e6f520SAl Viro goto need_lookup; 16165a18fff2SAl Viro } 1617697f514dSMiklos Szeredi 1618766c4cbfSAl Viro if (unlikely(d_is_negative(dentry))) { 1619766c4cbfSAl Viro dput(dentry); 1620766c4cbfSAl Viro return -ENOENT; 1621766c4cbfSAl Viro } 16221da177e4SLinus Torvalds path->mnt = mnt; 16231da177e4SLinus Torvalds path->dentry = dentry; 1624756daf26SNeilBrown err = follow_managed(path, nd); 16258402752eSAl Viro if (likely(!err)) 162663afdfc7SDavid Howells *inode = d_backing_inode(path->dentry); 16278402752eSAl Viro return err; 162881e6f520SAl Viro 162981e6f520SAl Viro need_lookup: 1630697f514dSMiklos Szeredi return 1; 1631697f514dSMiklos Szeredi } 1632697f514dSMiklos Szeredi 1633697f514dSMiklos Szeredi /* Fast lookup failed, do it the slow way */ 1634cc2a5271SAl Viro static int lookup_slow(struct nameidata *nd, struct path *path) 1635697f514dSMiklos Szeredi { 1636697f514dSMiklos Szeredi struct dentry *dentry, *parent; 1637697f514dSMiklos Szeredi 1638697f514dSMiklos Szeredi parent = nd->path.dentry; 163981e6f520SAl Viro BUG_ON(nd->inode != parent->d_inode); 164081e6f520SAl Viro 164181e6f520SAl Viro mutex_lock(&parent->d_inode->i_mutex); 1642cc2a5271SAl Viro dentry = __lookup_hash(&nd->last, parent, nd->flags); 164381e6f520SAl Viro mutex_unlock(&parent->d_inode->i_mutex); 164481e6f520SAl Viro if (IS_ERR(dentry)) 164581e6f520SAl Viro return PTR_ERR(dentry); 1646697f514dSMiklos Szeredi path->mnt = nd->path.mnt; 1647697f514dSMiklos Szeredi path->dentry = dentry; 16488402752eSAl Viro return follow_managed(path, nd); 16491da177e4SLinus Torvalds } 16501da177e4SLinus Torvalds 165152094c8aSAl Viro static inline int may_lookup(struct nameidata *nd) 165252094c8aSAl Viro { 165352094c8aSAl Viro if (nd->flags & LOOKUP_RCU) { 16544ad5abb3SAl Viro int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); 165552094c8aSAl Viro if (err != -ECHILD) 165652094c8aSAl Viro return err; 16576e9918b7SAl Viro if (unlazy_walk(nd, NULL, 0)) 165852094c8aSAl Viro return -ECHILD; 165952094c8aSAl Viro } 16604ad5abb3SAl Viro return inode_permission(nd->inode, MAY_EXEC); 166152094c8aSAl Viro } 166252094c8aSAl Viro 16639856fa1bSAl Viro static inline int handle_dots(struct nameidata *nd, int type) 16649856fa1bSAl Viro { 16659856fa1bSAl Viro if (type == LAST_DOTDOT) { 16669856fa1bSAl Viro if (nd->flags & LOOKUP_RCU) { 166770291aecSAl Viro return follow_dotdot_rcu(nd); 16689856fa1bSAl Viro } else 1669397d425dSEric W. Biederman return follow_dotdot(nd); 16709856fa1bSAl Viro } 16719856fa1bSAl Viro return 0; 16729856fa1bSAl Viro } 16739856fa1bSAl Viro 1674181548c0SAl Viro static int pick_link(struct nameidata *nd, struct path *link, 1675181548c0SAl Viro struct inode *inode, unsigned seq) 1676d63ff28fSAl Viro { 1677626de996SAl Viro int error; 16781cf2665bSAl Viro struct saved *last; 1679756daf26SNeilBrown if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) { 1680626de996SAl Viro path_to_nameidata(link, nd); 1681626de996SAl Viro return -ELOOP; 1682626de996SAl Viro } 1683bc40aee0SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1684cd179f44SAl Viro if (link->mnt == nd->path.mnt) 1685cd179f44SAl Viro mntget(link->mnt); 16867973387aSAl Viro } 1687626de996SAl Viro error = nd_alloc_stack(nd); 1688626de996SAl Viro if (unlikely(error)) { 1689bc40aee0SAl Viro if (error == -ECHILD) { 1690bc40aee0SAl Viro if (unlikely(unlazy_link(nd, link, seq))) 1691bc40aee0SAl Viro return -ECHILD; 1692bc40aee0SAl Viro error = nd_alloc_stack(nd); 1693bc40aee0SAl Viro } 1694bc40aee0SAl Viro if (error) { 1695cd179f44SAl Viro path_put(link); 1696626de996SAl Viro return error; 1697626de996SAl Viro } 1698bc40aee0SAl Viro } 1699626de996SAl Viro 1700ab104923SAl Viro last = nd->stack + nd->depth++; 17011cf2665bSAl Viro last->link = *link; 1702ab104923SAl Viro last->cookie = NULL; 1703181548c0SAl Viro last->inode = inode; 17040450b2d1SAl Viro last->seq = seq; 1705d63ff28fSAl Viro return 1; 1706d63ff28fSAl Viro } 1707d63ff28fSAl Viro 17083ddcd056SLinus Torvalds /* 17093ddcd056SLinus Torvalds * Do we need to follow links? We _really_ want to be able 17103ddcd056SLinus Torvalds * to do this check without having to look at inode->i_op, 17113ddcd056SLinus Torvalds * so we keep a cache of "no, this doesn't need follow_link" 17123ddcd056SLinus Torvalds * for the common case. 17133ddcd056SLinus Torvalds */ 1714254cf582SAl Viro static inline int should_follow_link(struct nameidata *nd, struct path *link, 1715181548c0SAl Viro int follow, 1716181548c0SAl Viro struct inode *inode, unsigned seq) 17173ddcd056SLinus Torvalds { 1718d63ff28fSAl Viro if (likely(!d_is_symlink(link->dentry))) 1719d63ff28fSAl Viro return 0; 1720d63ff28fSAl Viro if (!follow) 1721d63ff28fSAl Viro return 0; 1722181548c0SAl Viro return pick_link(nd, link, inode, seq); 17233ddcd056SLinus Torvalds } 17243ddcd056SLinus Torvalds 17254693a547SAl Viro enum {WALK_GET = 1, WALK_PUT = 2}; 17264693a547SAl Viro 17274693a547SAl Viro static int walk_component(struct nameidata *nd, int flags) 1728ce57dfc1SAl Viro { 1729caa85634SAl Viro struct path path; 1730ce57dfc1SAl Viro struct inode *inode; 1731254cf582SAl Viro unsigned seq; 1732ce57dfc1SAl Viro int err; 1733ce57dfc1SAl Viro /* 1734ce57dfc1SAl Viro * "." and ".." are special - ".." especially so because it has 1735ce57dfc1SAl Viro * to be able to know about the current root directory and 1736ce57dfc1SAl Viro * parent relationships. 1737ce57dfc1SAl Viro */ 17384693a547SAl Viro if (unlikely(nd->last_type != LAST_NORM)) { 17394693a547SAl Viro err = handle_dots(nd, nd->last_type); 17404693a547SAl Viro if (flags & WALK_PUT) 17414693a547SAl Viro put_link(nd); 17424693a547SAl Viro return err; 17434693a547SAl Viro } 1744254cf582SAl Viro err = lookup_fast(nd, &path, &inode, &seq); 1745ce57dfc1SAl Viro if (unlikely(err)) { 1746697f514dSMiklos Szeredi if (err < 0) 1747f0a9ba70SAl Viro return err; 1748697f514dSMiklos Szeredi 1749caa85634SAl Viro err = lookup_slow(nd, &path); 1750697f514dSMiklos Szeredi if (err < 0) 1751f0a9ba70SAl Viro return err; 1752697f514dSMiklos Szeredi 175363afdfc7SDavid Howells inode = d_backing_inode(path.dentry); 1754254cf582SAl Viro seq = 0; /* we are already out of RCU mode */ 1755697f514dSMiklos Szeredi err = -ENOENT; 1756caa85634SAl Viro if (d_is_negative(path.dentry)) 1757697f514dSMiklos Szeredi goto out_path_put; 1758766c4cbfSAl Viro } 1759697f514dSMiklos Szeredi 17604693a547SAl Viro if (flags & WALK_PUT) 17614693a547SAl Viro put_link(nd); 1762181548c0SAl Viro err = should_follow_link(nd, &path, flags & WALK_GET, inode, seq); 1763d63ff28fSAl Viro if (unlikely(err)) 1764d63ff28fSAl Viro return err; 1765caa85634SAl Viro path_to_nameidata(&path, nd); 1766ce57dfc1SAl Viro nd->inode = inode; 1767254cf582SAl Viro nd->seq = seq; 1768ce57dfc1SAl Viro return 0; 1769697f514dSMiklos Szeredi 1770697f514dSMiklos Szeredi out_path_put: 1771caa85634SAl Viro path_to_nameidata(&path, nd); 1772697f514dSMiklos Szeredi return err; 1773ce57dfc1SAl Viro } 1774ce57dfc1SAl Viro 17751da177e4SLinus Torvalds /* 1776bfcfaa77SLinus Torvalds * We can do the critical dentry name comparison and hashing 1777bfcfaa77SLinus Torvalds * operations one word at a time, but we are limited to: 1778bfcfaa77SLinus Torvalds * 1779bfcfaa77SLinus Torvalds * - Architectures with fast unaligned word accesses. We could 1780bfcfaa77SLinus Torvalds * do a "get_unaligned()" if this helps and is sufficiently 1781bfcfaa77SLinus Torvalds * fast. 1782bfcfaa77SLinus Torvalds * 1783bfcfaa77SLinus Torvalds * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we 1784bfcfaa77SLinus Torvalds * do not trap on the (extremely unlikely) case of a page 1785bfcfaa77SLinus Torvalds * crossing operation. 1786bfcfaa77SLinus Torvalds * 1787bfcfaa77SLinus Torvalds * - Furthermore, we need an efficient 64-bit compile for the 1788bfcfaa77SLinus Torvalds * 64-bit case in order to generate the "number of bytes in 1789bfcfaa77SLinus Torvalds * the final mask". Again, that could be replaced with a 1790bfcfaa77SLinus Torvalds * efficient population count instruction or similar. 1791bfcfaa77SLinus Torvalds */ 1792bfcfaa77SLinus Torvalds #ifdef CONFIG_DCACHE_WORD_ACCESS 1793bfcfaa77SLinus Torvalds 1794f68e556eSLinus Torvalds #include <asm/word-at-a-time.h> 1795bfcfaa77SLinus Torvalds 1796f68e556eSLinus Torvalds #ifdef CONFIG_64BIT 1797bfcfaa77SLinus Torvalds 1798bfcfaa77SLinus Torvalds static inline unsigned int fold_hash(unsigned long hash) 1799bfcfaa77SLinus Torvalds { 180099d263d4SLinus Torvalds return hash_64(hash, 32); 1801bfcfaa77SLinus Torvalds } 1802bfcfaa77SLinus Torvalds 1803bfcfaa77SLinus Torvalds #else /* 32-bit case */ 1804bfcfaa77SLinus Torvalds 1805bfcfaa77SLinus Torvalds #define fold_hash(x) (x) 1806bfcfaa77SLinus Torvalds 1807bfcfaa77SLinus Torvalds #endif 1808bfcfaa77SLinus Torvalds 1809bfcfaa77SLinus Torvalds unsigned int full_name_hash(const unsigned char *name, unsigned int len) 1810bfcfaa77SLinus Torvalds { 1811bfcfaa77SLinus Torvalds unsigned long a, mask; 1812bfcfaa77SLinus Torvalds unsigned long hash = 0; 1813bfcfaa77SLinus Torvalds 1814bfcfaa77SLinus Torvalds for (;;) { 1815e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name); 1816bfcfaa77SLinus Torvalds if (len < sizeof(unsigned long)) 1817bfcfaa77SLinus Torvalds break; 1818bfcfaa77SLinus Torvalds hash += a; 1819f132c5beSAl Viro hash *= 9; 1820bfcfaa77SLinus Torvalds name += sizeof(unsigned long); 1821bfcfaa77SLinus Torvalds len -= sizeof(unsigned long); 1822bfcfaa77SLinus Torvalds if (!len) 1823bfcfaa77SLinus Torvalds goto done; 1824bfcfaa77SLinus Torvalds } 1825a5c21dceSWill Deacon mask = bytemask_from_count(len); 1826bfcfaa77SLinus Torvalds hash += mask & a; 1827bfcfaa77SLinus Torvalds done: 1828bfcfaa77SLinus Torvalds return fold_hash(hash); 1829bfcfaa77SLinus Torvalds } 1830bfcfaa77SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 1831bfcfaa77SLinus Torvalds 1832bfcfaa77SLinus Torvalds /* 1833bfcfaa77SLinus Torvalds * Calculate the length and hash of the path component, and 1834d6bb3e90SLinus Torvalds * return the "hash_len" as the result. 1835bfcfaa77SLinus Torvalds */ 1836d6bb3e90SLinus Torvalds static inline u64 hash_name(const char *name) 1837bfcfaa77SLinus Torvalds { 183836126f8fSLinus Torvalds unsigned long a, b, adata, bdata, mask, hash, len; 183936126f8fSLinus Torvalds const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 1840bfcfaa77SLinus Torvalds 1841bfcfaa77SLinus Torvalds hash = a = 0; 1842bfcfaa77SLinus Torvalds len = -sizeof(unsigned long); 1843bfcfaa77SLinus Torvalds do { 1844bfcfaa77SLinus Torvalds hash = (hash + a) * 9; 1845bfcfaa77SLinus Torvalds len += sizeof(unsigned long); 1846e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name+len); 184736126f8fSLinus Torvalds b = a ^ REPEAT_BYTE('/'); 184836126f8fSLinus Torvalds } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); 1849bfcfaa77SLinus Torvalds 185036126f8fSLinus Torvalds adata = prep_zero_mask(a, adata, &constants); 185136126f8fSLinus Torvalds bdata = prep_zero_mask(b, bdata, &constants); 185236126f8fSLinus Torvalds 185336126f8fSLinus Torvalds mask = create_zero_mask(adata | bdata); 185436126f8fSLinus Torvalds 185536126f8fSLinus Torvalds hash += a & zero_bytemask(mask); 18569226b5b4SLinus Torvalds len += find_zero(mask); 1857d6bb3e90SLinus Torvalds return hashlen_create(fold_hash(hash), len); 1858bfcfaa77SLinus Torvalds } 1859bfcfaa77SLinus Torvalds 1860bfcfaa77SLinus Torvalds #else 1861bfcfaa77SLinus Torvalds 18620145acc2SLinus Torvalds unsigned int full_name_hash(const unsigned char *name, unsigned int len) 18630145acc2SLinus Torvalds { 18640145acc2SLinus Torvalds unsigned long hash = init_name_hash(); 18650145acc2SLinus Torvalds while (len--) 18660145acc2SLinus Torvalds hash = partial_name_hash(*name++, hash); 18670145acc2SLinus Torvalds return end_name_hash(hash); 18680145acc2SLinus Torvalds } 1869ae942ae7SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 18700145acc2SLinus Torvalds 18713ddcd056SLinus Torvalds /* 1872200e9ef7SLinus Torvalds * We know there's a real path component here of at least 1873200e9ef7SLinus Torvalds * one character. 1874200e9ef7SLinus Torvalds */ 1875d6bb3e90SLinus Torvalds static inline u64 hash_name(const char *name) 1876200e9ef7SLinus Torvalds { 1877200e9ef7SLinus Torvalds unsigned long hash = init_name_hash(); 1878200e9ef7SLinus Torvalds unsigned long len = 0, c; 1879200e9ef7SLinus Torvalds 1880200e9ef7SLinus Torvalds c = (unsigned char)*name; 1881200e9ef7SLinus Torvalds do { 1882200e9ef7SLinus Torvalds len++; 1883200e9ef7SLinus Torvalds hash = partial_name_hash(c, hash); 1884200e9ef7SLinus Torvalds c = (unsigned char)name[len]; 1885200e9ef7SLinus Torvalds } while (c && c != '/'); 1886d6bb3e90SLinus Torvalds return hashlen_create(end_name_hash(hash), len); 1887200e9ef7SLinus Torvalds } 1888200e9ef7SLinus Torvalds 1889bfcfaa77SLinus Torvalds #endif 1890bfcfaa77SLinus Torvalds 1891200e9ef7SLinus Torvalds /* 18921da177e4SLinus Torvalds * Name resolution. 1893ea3834d9SPrasanna Meda * This is the basic name resolution function, turning a pathname into 1894ea3834d9SPrasanna Meda * the final dentry. We expect 'base' to be positive and a directory. 18951da177e4SLinus Torvalds * 1896ea3834d9SPrasanna Meda * Returns 0 and nd will have valid dentry and mnt on success. 1897ea3834d9SPrasanna Meda * Returns error and drops reference to input namei data on failure. 18981da177e4SLinus Torvalds */ 18996de88d72SAl Viro static int link_path_walk(const char *name, struct nameidata *nd) 19001da177e4SLinus Torvalds { 19011da177e4SLinus Torvalds int err; 19021da177e4SLinus Torvalds 19031da177e4SLinus Torvalds while (*name=='/') 19041da177e4SLinus Torvalds name++; 19051da177e4SLinus Torvalds if (!*name) 19069e18f10aSAl Viro return 0; 19071da177e4SLinus Torvalds 19081da177e4SLinus Torvalds /* At this point we know we have a real path component. */ 19091da177e4SLinus Torvalds for(;;) { 1910d6bb3e90SLinus Torvalds u64 hash_len; 1911fe479a58SAl Viro int type; 19121da177e4SLinus Torvalds 191352094c8aSAl Viro err = may_lookup(nd); 19141da177e4SLinus Torvalds if (err) 19153595e234SAl Viro return err; 19161da177e4SLinus Torvalds 1917d6bb3e90SLinus Torvalds hash_len = hash_name(name); 19181da177e4SLinus Torvalds 1919fe479a58SAl Viro type = LAST_NORM; 1920d6bb3e90SLinus Torvalds if (name[0] == '.') switch (hashlen_len(hash_len)) { 1921fe479a58SAl Viro case 2: 1922200e9ef7SLinus Torvalds if (name[1] == '.') { 1923fe479a58SAl Viro type = LAST_DOTDOT; 192416c2cd71SAl Viro nd->flags |= LOOKUP_JUMPED; 192516c2cd71SAl Viro } 1926fe479a58SAl Viro break; 1927fe479a58SAl Viro case 1: 1928fe479a58SAl Viro type = LAST_DOT; 1929fe479a58SAl Viro } 19305a202bcdSAl Viro if (likely(type == LAST_NORM)) { 19315a202bcdSAl Viro struct dentry *parent = nd->path.dentry; 193216c2cd71SAl Viro nd->flags &= ~LOOKUP_JUMPED; 19335a202bcdSAl Viro if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { 1934a060dc50SJames Hogan struct qstr this = { { .hash_len = hash_len }, .name = name }; 1935da53be12SLinus Torvalds err = parent->d_op->d_hash(parent, &this); 19365a202bcdSAl Viro if (err < 0) 19373595e234SAl Viro return err; 1938d6bb3e90SLinus Torvalds hash_len = this.hash_len; 1939d6bb3e90SLinus Torvalds name = this.name; 19405a202bcdSAl Viro } 19415a202bcdSAl Viro } 1942fe479a58SAl Viro 1943d6bb3e90SLinus Torvalds nd->last.hash_len = hash_len; 1944d6bb3e90SLinus Torvalds nd->last.name = name; 19455f4a6a69SAl Viro nd->last_type = type; 19465f4a6a69SAl Viro 1947d6bb3e90SLinus Torvalds name += hashlen_len(hash_len); 1948d6bb3e90SLinus Torvalds if (!*name) 1949bdf6cbf1SAl Viro goto OK; 1950200e9ef7SLinus Torvalds /* 1951200e9ef7SLinus Torvalds * If it wasn't NUL, we know it was '/'. Skip that 1952200e9ef7SLinus Torvalds * slash, and continue until no more slashes. 1953200e9ef7SLinus Torvalds */ 1954200e9ef7SLinus Torvalds do { 1955d6bb3e90SLinus Torvalds name++; 1956d6bb3e90SLinus Torvalds } while (unlikely(*name == '/')); 19578620c238SAl Viro if (unlikely(!*name)) { 19588620c238SAl Viro OK: 1959368ee9baSAl Viro /* pathname body, done */ 19608620c238SAl Viro if (!nd->depth) 19618620c238SAl Viro return 0; 19628620c238SAl Viro name = nd->stack[nd->depth - 1].name; 1963368ee9baSAl Viro /* trailing symlink, done */ 19648620c238SAl Viro if (!name) 19658620c238SAl Viro return 0; 19668620c238SAl Viro /* last component of nested symlink */ 19674693a547SAl Viro err = walk_component(nd, WALK_GET | WALK_PUT); 19688620c238SAl Viro } else { 19694693a547SAl Viro err = walk_component(nd, WALK_GET); 19708620c238SAl Viro } 1971ce57dfc1SAl Viro if (err < 0) 19723595e234SAl Viro return err; 1973fe479a58SAl Viro 1974ce57dfc1SAl Viro if (err) { 1975626de996SAl Viro const char *s = get_link(nd); 19765a460275SAl Viro 1977a1c83681SViresh Kumar if (IS_ERR(s)) 19783595e234SAl Viro return PTR_ERR(s); 1979172a39a0SAl Viro err = 0; 198012b09578SAl Viro if (unlikely(!s)) { 198112b09578SAl Viro /* jumped */ 1982b9ff4429SAl Viro put_link(nd); 198312b09578SAl Viro } else { 1984071bf501SAl Viro nd->stack[nd->depth - 1].name = name; 198532cd7468SAl Viro name = s; 19869e18f10aSAl Viro continue; 198748c8b0c5SAl Viro } 198831e6b01fSNick Piggin } 198997242f99SAl Viro if (unlikely(!d_can_lookup(nd->path.dentry))) { 199097242f99SAl Viro if (nd->flags & LOOKUP_RCU) { 199197242f99SAl Viro if (unlazy_walk(nd, NULL, 0)) 199297242f99SAl Viro return -ECHILD; 199397242f99SAl Viro } 19943595e234SAl Viro return -ENOTDIR; 19955f4a6a69SAl Viro } 1996ce57dfc1SAl Viro } 199797242f99SAl Viro } 19981da177e4SLinus Torvalds 1999c8a53ee5SAl Viro static const char *path_init(struct nameidata *nd, unsigned flags) 200031e6b01fSNick Piggin { 200131e6b01fSNick Piggin int retval = 0; 2002c8a53ee5SAl Viro const char *s = nd->name->name; 200331e6b01fSNick Piggin 200431e6b01fSNick Piggin nd->last_type = LAST_ROOT; /* if there are only slashes... */ 2005980f3ea2SAl Viro nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT; 200631e6b01fSNick Piggin nd->depth = 0; 20075b6ca027SAl Viro if (flags & LOOKUP_ROOT) { 2008b18825a7SDavid Howells struct dentry *root = nd->root.dentry; 2009b18825a7SDavid Howells struct inode *inode = root->d_inode; 2010fd2f7cb5SAl Viro if (*s) { 201144b1d530SMiklos Szeredi if (!d_can_lookup(root)) 2012368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 20135b6ca027SAl Viro retval = inode_permission(inode, MAY_EXEC); 20145b6ca027SAl Viro if (retval) 2015368ee9baSAl Viro return ERR_PTR(retval); 201673d049a4SAl Viro } 20175b6ca027SAl Viro nd->path = nd->root; 20185b6ca027SAl Viro nd->inode = inode; 20195b6ca027SAl Viro if (flags & LOOKUP_RCU) { 20208b61e74fSAl Viro rcu_read_lock(); 20215b6ca027SAl Viro nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 20228f47a016SAl Viro nd->root_seq = nd->seq; 202348a066e7SAl Viro nd->m_seq = read_seqbegin(&mount_lock); 20245b6ca027SAl Viro } else { 20255b6ca027SAl Viro path_get(&nd->path); 20265b6ca027SAl Viro } 2027368ee9baSAl Viro return s; 20285b6ca027SAl Viro } 20295b6ca027SAl Viro 203031e6b01fSNick Piggin nd->root.mnt = NULL; 203131e6b01fSNick Piggin 203248a066e7SAl Viro nd->m_seq = read_seqbegin(&mount_lock); 2033fd2f7cb5SAl Viro if (*s == '/') { 2034e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 20358b61e74fSAl Viro rcu_read_lock(); 203606d7137eSAl Viro set_root_rcu(nd); 203706d7137eSAl Viro nd->seq = nd->root_seq; 2038e41f7d4eSAl Viro } else { 2039e41f7d4eSAl Viro set_root(nd); 2040e41f7d4eSAl Viro path_get(&nd->root); 2041e41f7d4eSAl Viro } 204231e6b01fSNick Piggin nd->path = nd->root; 2043c8a53ee5SAl Viro } else if (nd->dfd == AT_FDCWD) { 2044e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 204531e6b01fSNick Piggin struct fs_struct *fs = current->fs; 2046c28cc364SNick Piggin unsigned seq; 204731e6b01fSNick Piggin 20488b61e74fSAl Viro rcu_read_lock(); 204931e6b01fSNick Piggin 2050c28cc364SNick Piggin do { 2051c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 205231e6b01fSNick Piggin nd->path = fs->pwd; 2053c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 2054c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 2055e41f7d4eSAl Viro } else { 2056e41f7d4eSAl Viro get_fs_pwd(current->fs, &nd->path); 2057e41f7d4eSAl Viro } 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 } 2087e41f7d4eSAl Viro 208831e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 20894023bfc9SAl Viro if (!(flags & LOOKUP_RCU)) 2090368ee9baSAl Viro return s; 20914023bfc9SAl Viro if (likely(!read_seqcount_retry(&nd->path.dentry->d_seq, nd->seq))) 2092368ee9baSAl Viro return s; 20934023bfc9SAl Viro if (!(nd->flags & LOOKUP_ROOT)) 20944023bfc9SAl Viro nd->root.mnt = NULL; 20954023bfc9SAl Viro rcu_read_unlock(); 2096368ee9baSAl Viro return ERR_PTR(-ECHILD); 20979b4a9b14SAl Viro } 20989b4a9b14SAl Viro 20993bdba28bSAl Viro static const char *trailing_symlink(struct nameidata *nd) 210095fa25d9SAl Viro { 210195fa25d9SAl Viro const char *s; 2102fec2fa24SAl Viro int error = may_follow_link(nd); 2103deb106c6SAl Viro if (unlikely(error)) 21043bdba28bSAl Viro return ERR_PTR(error); 210595fa25d9SAl Viro nd->flags |= LOOKUP_PARENT; 2106fab51e8aSAl Viro nd->stack[0].name = NULL; 21073b2e7f75SAl Viro s = get_link(nd); 2108deb106c6SAl Viro return s ? s : ""; 210995fa25d9SAl Viro } 211095fa25d9SAl Viro 2111caa85634SAl Viro static inline int lookup_last(struct nameidata *nd) 2112bd92d7feSAl Viro { 2113bd92d7feSAl Viro if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) 2114bd92d7feSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 2115bd92d7feSAl Viro 2116bd92d7feSAl Viro nd->flags &= ~LOOKUP_PARENT; 2117deb106c6SAl Viro return walk_component(nd, 21184693a547SAl Viro nd->flags & LOOKUP_FOLLOW 21194693a547SAl Viro ? nd->depth 21204693a547SAl Viro ? WALK_PUT | WALK_GET 21214693a547SAl Viro : WALK_GET 21224693a547SAl Viro : 0); 2123bd92d7feSAl Viro } 2124bd92d7feSAl Viro 21259b4a9b14SAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2126c8a53ee5SAl Viro static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path) 21279b4a9b14SAl Viro { 2128c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2129bd92d7feSAl Viro int err; 213031e6b01fSNick Piggin 2131368ee9baSAl Viro if (IS_ERR(s)) 2132368ee9baSAl Viro return PTR_ERR(s); 21333bdba28bSAl Viro while (!(err = link_path_walk(s, nd)) 21343bdba28bSAl Viro && ((err = lookup_last(nd)) > 0)) { 21353bdba28bSAl Viro s = trailing_symlink(nd); 21363bdba28bSAl Viro if (IS_ERR(s)) { 21373bdba28bSAl Viro err = PTR_ERR(s); 21386d7b5aaeSAl Viro break; 2139bd92d7feSAl Viro } 2140bd92d7feSAl Viro } 21419f1fafeeSAl Viro if (!err) 21429f1fafeeSAl Viro err = complete_walk(nd); 2143bd92d7feSAl Viro 2144deb106c6SAl Viro if (!err && nd->flags & LOOKUP_DIRECTORY) 2145deb106c6SAl Viro if (!d_can_lookup(nd->path.dentry)) 2146bd23a539SAl Viro err = -ENOTDIR; 2147625b6d10SAl Viro if (!err) { 2148625b6d10SAl Viro *path = nd->path; 2149625b6d10SAl Viro nd->path.mnt = NULL; 2150625b6d10SAl Viro nd->path.dentry = NULL; 2151625b6d10SAl Viro } 2152deb106c6SAl Viro terminate_walk(nd); 2153bd92d7feSAl Viro return err; 215431e6b01fSNick Piggin } 215531e6b01fSNick Piggin 2156625b6d10SAl Viro static int filename_lookup(int dfd, struct filename *name, unsigned flags, 21579ad1aaa6SAl Viro struct path *path, struct path *root) 2158873f1eedSJeff Layton { 2159894bc8c4SAl Viro int retval; 21609883d185SAl Viro struct nameidata nd; 2161abc9f5beSAl Viro if (IS_ERR(name)) 2162abc9f5beSAl Viro return PTR_ERR(name); 21639ad1aaa6SAl Viro if (unlikely(root)) { 21649ad1aaa6SAl Viro nd.root = *root; 21659ad1aaa6SAl Viro flags |= LOOKUP_ROOT; 21669ad1aaa6SAl Viro } 21679883d185SAl Viro set_nameidata(&nd, dfd, name); 2168c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_RCU, path); 2169873f1eedSJeff Layton if (unlikely(retval == -ECHILD)) 2170c8a53ee5SAl Viro retval = path_lookupat(&nd, flags, path); 2171873f1eedSJeff Layton if (unlikely(retval == -ESTALE)) 2172c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path); 2173873f1eedSJeff Layton 2174873f1eedSJeff Layton if (likely(!retval)) 2175625b6d10SAl Viro audit_inode(name, path->dentry, flags & LOOKUP_PARENT); 21769883d185SAl Viro restore_nameidata(); 2177e4bd1c1aSAl Viro putname(name); 2178873f1eedSJeff Layton return retval; 2179873f1eedSJeff Layton } 2180873f1eedSJeff Layton 21818bcb77faSAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2182c8a53ee5SAl Viro static int path_parentat(struct nameidata *nd, unsigned flags, 2183391172c4SAl Viro struct path *parent) 21848bcb77faSAl Viro { 2185c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2186368ee9baSAl Viro int err; 2187368ee9baSAl Viro if (IS_ERR(s)) 2188368ee9baSAl Viro return PTR_ERR(s); 2189368ee9baSAl Viro err = link_path_walk(s, nd); 21908bcb77faSAl Viro if (!err) 21918bcb77faSAl Viro err = complete_walk(nd); 2192391172c4SAl Viro if (!err) { 2193391172c4SAl Viro *parent = nd->path; 2194391172c4SAl Viro nd->path.mnt = NULL; 2195391172c4SAl Viro nd->path.dentry = NULL; 2196391172c4SAl Viro } 2197deb106c6SAl Viro terminate_walk(nd); 21988bcb77faSAl Viro return err; 21998bcb77faSAl Viro } 22008bcb77faSAl Viro 22015c31b6ceSAl Viro static struct filename *filename_parentat(int dfd, struct filename *name, 2202391172c4SAl Viro unsigned int flags, struct path *parent, 2203391172c4SAl Viro struct qstr *last, int *type) 22048bcb77faSAl Viro { 22058bcb77faSAl Viro int retval; 22069883d185SAl Viro struct nameidata nd; 22078bcb77faSAl Viro 22085c31b6ceSAl Viro if (IS_ERR(name)) 22095c31b6ceSAl Viro return name; 22109883d185SAl Viro set_nameidata(&nd, dfd, name); 2211c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_RCU, parent); 22128bcb77faSAl Viro if (unlikely(retval == -ECHILD)) 2213c8a53ee5SAl Viro retval = path_parentat(&nd, flags, parent); 22148bcb77faSAl Viro if (unlikely(retval == -ESTALE)) 2215c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent); 2216391172c4SAl Viro if (likely(!retval)) { 2217391172c4SAl Viro *last = nd.last; 2218391172c4SAl Viro *type = nd.last_type; 2219391172c4SAl Viro audit_inode(name, parent->dentry, LOOKUP_PARENT); 22205c31b6ceSAl Viro } else { 22215c31b6ceSAl Viro putname(name); 22225c31b6ceSAl Viro name = ERR_PTR(retval); 2223391172c4SAl Viro } 22249883d185SAl Viro restore_nameidata(); 22255c31b6ceSAl Viro return name; 22268bcb77faSAl Viro } 22278bcb77faSAl Viro 222879714f72SAl Viro /* does lookup, returns the object with parent locked */ 222979714f72SAl Viro struct dentry *kern_path_locked(const char *name, struct path *path) 22305590ff0dSUlrich Drepper { 22315c31b6ceSAl Viro struct filename *filename; 22325c31b6ceSAl Viro struct dentry *d; 2233391172c4SAl Viro struct qstr last; 2234391172c4SAl Viro int type; 223551689104SPaul Moore 22365c31b6ceSAl Viro filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path, 22375c31b6ceSAl Viro &last, &type); 223851689104SPaul Moore if (IS_ERR(filename)) 223951689104SPaul Moore return ERR_CAST(filename); 22405c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) { 2241391172c4SAl Viro path_put(path); 22425c31b6ceSAl Viro putname(filename); 22435c31b6ceSAl Viro return ERR_PTR(-EINVAL); 224479714f72SAl Viro } 2245391172c4SAl Viro mutex_lock_nested(&path->dentry->d_inode->i_mutex, I_MUTEX_PARENT); 2246391172c4SAl Viro d = __lookup_hash(&last, path->dentry, 0); 224779714f72SAl Viro if (IS_ERR(d)) { 2248391172c4SAl Viro mutex_unlock(&path->dentry->d_inode->i_mutex); 2249391172c4SAl Viro path_put(path); 225079714f72SAl Viro } 225151689104SPaul Moore putname(filename); 225279714f72SAl Viro return d; 22535590ff0dSUlrich Drepper } 22545590ff0dSUlrich Drepper 2255d1811465SAl Viro int kern_path(const char *name, unsigned int flags, struct path *path) 2256d1811465SAl Viro { 2257abc9f5beSAl Viro return filename_lookup(AT_FDCWD, getname_kernel(name), 2258abc9f5beSAl Viro flags, path, NULL); 2259d1811465SAl Viro } 22604d359507SAl Viro EXPORT_SYMBOL(kern_path); 2261d1811465SAl Viro 226216f18200SJosef 'Jeff' Sipek /** 226316f18200SJosef 'Jeff' Sipek * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair 226416f18200SJosef 'Jeff' Sipek * @dentry: pointer to dentry of the base directory 226516f18200SJosef 'Jeff' Sipek * @mnt: pointer to vfs mount of the base directory 226616f18200SJosef 'Jeff' Sipek * @name: pointer to file name 226716f18200SJosef 'Jeff' Sipek * @flags: lookup flags 2268e0a01249SAl Viro * @path: pointer to struct path to fill 226916f18200SJosef 'Jeff' Sipek */ 227016f18200SJosef 'Jeff' Sipek int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, 227116f18200SJosef 'Jeff' Sipek const char *name, unsigned int flags, 2272e0a01249SAl Viro struct path *path) 227316f18200SJosef 'Jeff' Sipek { 22749ad1aaa6SAl Viro struct path root = {.mnt = mnt, .dentry = dentry}; 22759ad1aaa6SAl Viro /* the first argument of filename_lookup() is ignored with root */ 2276abc9f5beSAl Viro return filename_lookup(AT_FDCWD, getname_kernel(name), 2277abc9f5beSAl Viro flags , path, &root); 227816f18200SJosef 'Jeff' Sipek } 22794d359507SAl Viro EXPORT_SYMBOL(vfs_path_lookup); 228016f18200SJosef 'Jeff' Sipek 2281eead1911SChristoph Hellwig /** 2282a6b91919SRandy Dunlap * lookup_one_len - filesystem helper to lookup single pathname component 2283eead1911SChristoph Hellwig * @name: pathname component to lookup 2284eead1911SChristoph Hellwig * @base: base directory to lookup from 2285eead1911SChristoph Hellwig * @len: maximum length @len should be interpreted to 2286eead1911SChristoph Hellwig * 2287a6b91919SRandy Dunlap * Note that this routine is purely a helper for filesystem usage and should 22889e7543e9SAl Viro * not be called by generic code. 2289eead1911SChristoph Hellwig */ 2290057f6c01SJames Morris struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 2291057f6c01SJames Morris { 2292057f6c01SJames Morris struct qstr this; 22936a96ba54SAl Viro unsigned int c; 2294cda309deSMiklos Szeredi int err; 2295057f6c01SJames Morris 22962f9092e1SDavid Woodhouse WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex)); 22972f9092e1SDavid Woodhouse 22986a96ba54SAl Viro this.name = name; 22996a96ba54SAl Viro this.len = len; 23000145acc2SLinus Torvalds this.hash = full_name_hash(name, len); 23016a96ba54SAl Viro if (!len) 23026a96ba54SAl Viro return ERR_PTR(-EACCES); 23036a96ba54SAl Viro 230421d8a15aSAl Viro if (unlikely(name[0] == '.')) { 230521d8a15aSAl Viro if (len < 2 || (len == 2 && name[1] == '.')) 230621d8a15aSAl Viro return ERR_PTR(-EACCES); 230721d8a15aSAl Viro } 230821d8a15aSAl Viro 23096a96ba54SAl Viro while (len--) { 23106a96ba54SAl Viro c = *(const unsigned char *)name++; 23116a96ba54SAl Viro if (c == '/' || c == '\0') 23126a96ba54SAl Viro return ERR_PTR(-EACCES); 23136a96ba54SAl Viro } 23145a202bcdSAl Viro /* 23155a202bcdSAl Viro * See if the low-level filesystem might want 23165a202bcdSAl Viro * to use its own hash.. 23175a202bcdSAl Viro */ 23185a202bcdSAl Viro if (base->d_flags & DCACHE_OP_HASH) { 2319da53be12SLinus Torvalds int err = base->d_op->d_hash(base, &this); 23205a202bcdSAl Viro if (err < 0) 23215a202bcdSAl Viro return ERR_PTR(err); 23225a202bcdSAl Viro } 2323eead1911SChristoph Hellwig 2324cda309deSMiklos Szeredi err = inode_permission(base->d_inode, MAY_EXEC); 2325cda309deSMiklos Szeredi if (err) 2326cda309deSMiklos Szeredi return ERR_PTR(err); 2327cda309deSMiklos Szeredi 232872bd866aSAl Viro return __lookup_hash(&this, base, 0); 2329057f6c01SJames Morris } 23304d359507SAl Viro EXPORT_SYMBOL(lookup_one_len); 2331057f6c01SJames Morris 23321fa1e7f6SAndy Whitcroft int user_path_at_empty(int dfd, const char __user *name, unsigned flags, 23331fa1e7f6SAndy Whitcroft struct path *path, int *empty) 23341da177e4SLinus Torvalds { 2335abc9f5beSAl Viro return filename_lookup(dfd, getname_flags(name, flags, empty), 2336abc9f5beSAl Viro flags, path, NULL); 23371da177e4SLinus Torvalds } 2338b853a161SAl Viro EXPORT_SYMBOL(user_path_at_empty); 23391fa1e7f6SAndy Whitcroft 2340873f1eedSJeff Layton /* 2341873f1eedSJeff Layton * NB: most callers don't do anything directly with the reference to the 2342873f1eedSJeff Layton * to struct filename, but the nd->last pointer points into the name string 2343873f1eedSJeff Layton * allocated by getname. So we must hold the reference to it until all 2344873f1eedSJeff Layton * path-walking is complete. 2345873f1eedSJeff Layton */ 2346a2ec4a2dSAl Viro static inline struct filename * 2347f5beed75SAl Viro user_path_parent(int dfd, const char __user *path, 2348f5beed75SAl Viro struct path *parent, 2349f5beed75SAl Viro struct qstr *last, 2350f5beed75SAl Viro int *type, 23519e790bd6SJeff Layton unsigned int flags) 23522ad94ae6SAl Viro { 23539e790bd6SJeff Layton /* only LOOKUP_REVAL is allowed in extra flags */ 23545c31b6ceSAl Viro return filename_parentat(dfd, getname(path), flags & LOOKUP_REVAL, 23555c31b6ceSAl Viro parent, last, type); 23562ad94ae6SAl Viro } 23572ad94ae6SAl Viro 23588033426eSJeff Layton /** 2359197df04cSAl Viro * mountpoint_last - look up last component for umount 23608033426eSJeff Layton * @nd: pathwalk nameidata - currently pointing at parent directory of "last" 23618033426eSJeff Layton * @path: pointer to container for result 23628033426eSJeff Layton * 23638033426eSJeff Layton * This is a special lookup_last function just for umount. In this case, we 23648033426eSJeff Layton * need to resolve the path without doing any revalidation. 23658033426eSJeff Layton * 23668033426eSJeff Layton * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since 23678033426eSJeff Layton * mountpoints are always pinned in the dcache, their ancestors are too. Thus, 23688033426eSJeff Layton * in almost all cases, this lookup will be served out of the dcache. The only 23698033426eSJeff Layton * cases where it won't are if nd->last refers to a symlink or the path is 23708033426eSJeff Layton * bogus and it doesn't exist. 23718033426eSJeff Layton * 23728033426eSJeff Layton * Returns: 23738033426eSJeff Layton * -error: if there was an error during lookup. This includes -ENOENT if the 23748033426eSJeff Layton * lookup found a negative dentry. The nd->path reference will also be 23758033426eSJeff Layton * put in this case. 23768033426eSJeff Layton * 23778033426eSJeff Layton * 0: if we successfully resolved nd->path and found it to not to be a 23788033426eSJeff Layton * symlink that needs to be followed. "path" will also be populated. 23798033426eSJeff Layton * The nd->path reference will also be put. 23808033426eSJeff Layton * 23818033426eSJeff Layton * 1: if we successfully resolved nd->last and found it to be a symlink 23828033426eSJeff Layton * that needs to be followed. "path" will be populated with the path 23838033426eSJeff Layton * to the link, and nd->path will *not* be put. 23848033426eSJeff Layton */ 23858033426eSJeff Layton static int 2386197df04cSAl Viro mountpoint_last(struct nameidata *nd, struct path *path) 23878033426eSJeff Layton { 23888033426eSJeff Layton int error = 0; 23898033426eSJeff Layton struct dentry *dentry; 23908033426eSJeff Layton struct dentry *dir = nd->path.dentry; 23918033426eSJeff Layton 239235759521SAl Viro /* If we're in rcuwalk, drop out of it to handle last component */ 239335759521SAl Viro if (nd->flags & LOOKUP_RCU) { 23946e9918b7SAl Viro if (unlazy_walk(nd, NULL, 0)) 2395deb106c6SAl Viro return -ECHILD; 23968033426eSJeff Layton } 23978033426eSJeff Layton 23988033426eSJeff Layton nd->flags &= ~LOOKUP_PARENT; 23998033426eSJeff Layton 24008033426eSJeff Layton if (unlikely(nd->last_type != LAST_NORM)) { 24018033426eSJeff Layton error = handle_dots(nd, nd->last_type); 240235759521SAl Viro if (error) 2403deb106c6SAl Viro return error; 24048033426eSJeff Layton dentry = dget(nd->path.dentry); 240535759521SAl Viro goto done; 24068033426eSJeff Layton } 24078033426eSJeff Layton 24088033426eSJeff Layton mutex_lock(&dir->d_inode->i_mutex); 24098033426eSJeff Layton dentry = d_lookup(dir, &nd->last); 24108033426eSJeff Layton if (!dentry) { 24118033426eSJeff Layton /* 24128033426eSJeff Layton * No cached dentry. Mounted dentries are pinned in the cache, 24138033426eSJeff Layton * so that means that this dentry is probably a symlink or the 24148033426eSJeff Layton * path doesn't actually point to a mounted dentry. 24158033426eSJeff Layton */ 24168033426eSJeff Layton dentry = d_alloc(dir, &nd->last); 24178033426eSJeff Layton if (!dentry) { 2418bcceeebaSDave Jones mutex_unlock(&dir->d_inode->i_mutex); 2419deb106c6SAl Viro return -ENOMEM; 24208033426eSJeff Layton } 242135759521SAl Viro dentry = lookup_real(dir->d_inode, dentry, nd->flags); 2422bcceeebaSDave Jones if (IS_ERR(dentry)) { 2423bcceeebaSDave Jones mutex_unlock(&dir->d_inode->i_mutex); 2424deb106c6SAl Viro return PTR_ERR(dentry); 24258033426eSJeff Layton } 2426bcceeebaSDave Jones } 24278033426eSJeff Layton mutex_unlock(&dir->d_inode->i_mutex); 24288033426eSJeff Layton 242935759521SAl Viro done: 2430698934dfSDavid Howells if (d_is_negative(dentry)) { 24318033426eSJeff Layton dput(dentry); 2432deb106c6SAl Viro return -ENOENT; 243335759521SAl Viro } 2434191d7f73SAl Viro if (nd->depth) 2435191d7f73SAl Viro put_link(nd); 24368033426eSJeff Layton path->dentry = dentry; 2437295dc39dSVasily Averin path->mnt = nd->path.mnt; 2438181548c0SAl Viro error = should_follow_link(nd, path, nd->flags & LOOKUP_FOLLOW, 2439181548c0SAl Viro d_backing_inode(dentry), 0); 2440deb106c6SAl Viro if (unlikely(error)) 2441d63ff28fSAl Viro return error; 2442295dc39dSVasily Averin mntget(path->mnt); 24438033426eSJeff Layton follow_mount(path); 2444deb106c6SAl Viro return 0; 24458033426eSJeff Layton } 24468033426eSJeff Layton 24478033426eSJeff Layton /** 2448197df04cSAl Viro * path_mountpoint - look up a path to be umounted 24492a78b857SMasanari Iida * @nd: lookup context 24508033426eSJeff Layton * @flags: lookup flags 2451c8a53ee5SAl Viro * @path: pointer to container for result 24528033426eSJeff Layton * 24538033426eSJeff Layton * Look up the given name, but don't attempt to revalidate the last component. 2454606d6fe3SRandy Dunlap * Returns 0 and "path" will be valid on success; Returns error otherwise. 24558033426eSJeff Layton */ 24568033426eSJeff Layton static int 2457c8a53ee5SAl Viro path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path) 24588033426eSJeff Layton { 2459c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2460368ee9baSAl Viro int err; 2461368ee9baSAl Viro if (IS_ERR(s)) 2462368ee9baSAl Viro return PTR_ERR(s); 24633bdba28bSAl Viro while (!(err = link_path_walk(s, nd)) && 24643bdba28bSAl Viro (err = mountpoint_last(nd, path)) > 0) { 24653bdba28bSAl Viro s = trailing_symlink(nd); 24663bdba28bSAl Viro if (IS_ERR(s)) { 24673bdba28bSAl Viro err = PTR_ERR(s); 24688033426eSJeff Layton break; 24698033426eSJeff Layton } 24703bdba28bSAl Viro } 2471deb106c6SAl Viro terminate_walk(nd); 24728033426eSJeff Layton return err; 24738033426eSJeff Layton } 24748033426eSJeff Layton 24752d864651SAl Viro static int 2476668696dcSAl Viro filename_mountpoint(int dfd, struct filename *name, struct path *path, 24772d864651SAl Viro unsigned int flags) 24782d864651SAl Viro { 24799883d185SAl Viro struct nameidata nd; 2480cbaab2dbSAl Viro int error; 2481668696dcSAl Viro if (IS_ERR(name)) 2482668696dcSAl Viro return PTR_ERR(name); 24839883d185SAl Viro set_nameidata(&nd, dfd, name); 2484c8a53ee5SAl Viro error = path_mountpoint(&nd, flags | LOOKUP_RCU, path); 24852d864651SAl Viro if (unlikely(error == -ECHILD)) 2486c8a53ee5SAl Viro error = path_mountpoint(&nd, flags, path); 24872d864651SAl Viro if (unlikely(error == -ESTALE)) 2488c8a53ee5SAl Viro error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path); 24892d864651SAl Viro if (likely(!error)) 2490668696dcSAl Viro audit_inode(name, path->dentry, 0); 24919883d185SAl Viro restore_nameidata(); 2492668696dcSAl Viro putname(name); 24932d864651SAl Viro return error; 24942d864651SAl Viro } 24952d864651SAl Viro 24968033426eSJeff Layton /** 2497197df04cSAl Viro * user_path_mountpoint_at - lookup a path from userland in order to umount it 24988033426eSJeff Layton * @dfd: directory file descriptor 24998033426eSJeff Layton * @name: pathname from userland 25008033426eSJeff Layton * @flags: lookup flags 25018033426eSJeff Layton * @path: pointer to container to hold result 25028033426eSJeff Layton * 25038033426eSJeff Layton * A umount is a special case for path walking. We're not actually interested 25048033426eSJeff Layton * in the inode in this situation, and ESTALE errors can be a problem. We 25058033426eSJeff Layton * simply want track down the dentry and vfsmount attached at the mountpoint 25068033426eSJeff Layton * and avoid revalidating the last component. 25078033426eSJeff Layton * 25088033426eSJeff Layton * Returns 0 and populates "path" on success. 25098033426eSJeff Layton */ 25108033426eSJeff Layton int 2511197df04cSAl Viro user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags, 25128033426eSJeff Layton struct path *path) 25138033426eSJeff Layton { 2514cbaab2dbSAl Viro return filename_mountpoint(dfd, getname(name), path, flags); 25158033426eSJeff Layton } 25168033426eSJeff Layton 25172d864651SAl Viro int 25182d864651SAl Viro kern_path_mountpoint(int dfd, const char *name, struct path *path, 25192d864651SAl Viro unsigned int flags) 25202d864651SAl Viro { 2521cbaab2dbSAl Viro return filename_mountpoint(dfd, getname_kernel(name), path, flags); 25222d864651SAl Viro } 25232d864651SAl Viro EXPORT_SYMBOL(kern_path_mountpoint); 25242d864651SAl Viro 2525cbdf35bcSMiklos Szeredi int __check_sticky(struct inode *dir, struct inode *inode) 25261da177e4SLinus Torvalds { 25278e96e3b7SEric W. Biederman kuid_t fsuid = current_fsuid(); 2528da9592edSDavid Howells 25298e96e3b7SEric W. Biederman if (uid_eq(inode->i_uid, fsuid)) 25301da177e4SLinus Torvalds return 0; 25318e96e3b7SEric W. Biederman if (uid_eq(dir->i_uid, fsuid)) 25321da177e4SLinus Torvalds return 0; 253323adbe12SAndy Lutomirski return !capable_wrt_inode_uidgid(inode, CAP_FOWNER); 25341da177e4SLinus Torvalds } 2535cbdf35bcSMiklos Szeredi EXPORT_SYMBOL(__check_sticky); 25361da177e4SLinus Torvalds 25371da177e4SLinus Torvalds /* 25381da177e4SLinus Torvalds * Check whether we can remove a link victim from directory dir, check 25391da177e4SLinus Torvalds * whether the type of victim is right. 25401da177e4SLinus Torvalds * 1. We can't do it if dir is read-only (done in permission()) 25411da177e4SLinus Torvalds * 2. We should have write and exec permissions on dir 25421da177e4SLinus Torvalds * 3. We can't remove anything from append-only dir 25431da177e4SLinus Torvalds * 4. We can't do anything with immutable dir (done in permission()) 25441da177e4SLinus Torvalds * 5. If the sticky bit on dir is set we should either 25451da177e4SLinus Torvalds * a. be owner of dir, or 25461da177e4SLinus Torvalds * b. be owner of victim, or 25471da177e4SLinus Torvalds * c. have CAP_FOWNER capability 25481da177e4SLinus Torvalds * 6. If the victim is append-only or immutable we can't do antyhing with 25491da177e4SLinus Torvalds * links pointing to it. 25501da177e4SLinus Torvalds * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. 25511da177e4SLinus Torvalds * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. 25521da177e4SLinus Torvalds * 9. We can't remove a root or mountpoint. 25531da177e4SLinus Torvalds * 10. We don't allow removal of NFS sillyrenamed files; it's handled by 25541da177e4SLinus Torvalds * nfs_async_unlink(). 25551da177e4SLinus Torvalds */ 2556b18825a7SDavid Howells static int may_delete(struct inode *dir, struct dentry *victim, bool isdir) 25571da177e4SLinus Torvalds { 255863afdfc7SDavid Howells struct inode *inode = d_backing_inode(victim); 25591da177e4SLinus Torvalds int error; 25601da177e4SLinus Torvalds 2561b18825a7SDavid Howells if (d_is_negative(victim)) 25621da177e4SLinus Torvalds return -ENOENT; 2563b18825a7SDavid Howells BUG_ON(!inode); 25641da177e4SLinus Torvalds 25651da177e4SLinus Torvalds BUG_ON(victim->d_parent->d_inode != dir); 25664fa6b5ecSJeff Layton audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); 25671da177e4SLinus Torvalds 2568f419a2e3SAl Viro error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 25691da177e4SLinus Torvalds if (error) 25701da177e4SLinus Torvalds return error; 25711da177e4SLinus Torvalds if (IS_APPEND(dir)) 25721da177e4SLinus Torvalds return -EPERM; 2573b18825a7SDavid Howells 2574b18825a7SDavid Howells if (check_sticky(dir, inode) || IS_APPEND(inode) || 2575b18825a7SDavid Howells IS_IMMUTABLE(inode) || IS_SWAPFILE(inode)) 25761da177e4SLinus Torvalds return -EPERM; 25771da177e4SLinus Torvalds if (isdir) { 257844b1d530SMiklos Szeredi if (!d_is_dir(victim)) 25791da177e4SLinus Torvalds return -ENOTDIR; 25801da177e4SLinus Torvalds if (IS_ROOT(victim)) 25811da177e4SLinus Torvalds return -EBUSY; 258244b1d530SMiklos Szeredi } else if (d_is_dir(victim)) 25831da177e4SLinus Torvalds return -EISDIR; 25841da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 25851da177e4SLinus Torvalds return -ENOENT; 25861da177e4SLinus Torvalds if (victim->d_flags & DCACHE_NFSFS_RENAMED) 25871da177e4SLinus Torvalds return -EBUSY; 25881da177e4SLinus Torvalds return 0; 25891da177e4SLinus Torvalds } 25901da177e4SLinus Torvalds 25911da177e4SLinus Torvalds /* Check whether we can create an object with dentry child in directory 25921da177e4SLinus Torvalds * dir. 25931da177e4SLinus Torvalds * 1. We can't do it if child already exists (open has special treatment for 25941da177e4SLinus Torvalds * this case, but since we are inlined it's OK) 25951da177e4SLinus Torvalds * 2. We can't do it if dir is read-only (done in permission()) 25961da177e4SLinus Torvalds * 3. We should have write and exec permissions on dir 25971da177e4SLinus Torvalds * 4. We can't do it if dir is immutable (done in permission()) 25981da177e4SLinus Torvalds */ 2599a95164d9SMiklos Szeredi static inline int may_create(struct inode *dir, struct dentry *child) 26001da177e4SLinus Torvalds { 260114e972b4SJeff Layton audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); 26021da177e4SLinus Torvalds if (child->d_inode) 26031da177e4SLinus Torvalds return -EEXIST; 26041da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 26051da177e4SLinus Torvalds return -ENOENT; 2606f419a2e3SAl Viro return inode_permission(dir, MAY_WRITE | MAY_EXEC); 26071da177e4SLinus Torvalds } 26081da177e4SLinus Torvalds 26091da177e4SLinus Torvalds /* 26101da177e4SLinus Torvalds * p1 and p2 should be directories on the same fs. 26111da177e4SLinus Torvalds */ 26121da177e4SLinus Torvalds struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 26131da177e4SLinus Torvalds { 26141da177e4SLinus Torvalds struct dentry *p; 26151da177e4SLinus Torvalds 26161da177e4SLinus Torvalds if (p1 == p2) { 2617f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 26181da177e4SLinus Torvalds return NULL; 26191da177e4SLinus Torvalds } 26201da177e4SLinus Torvalds 2621a11f3a05SArjan van de Ven mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 26221da177e4SLinus Torvalds 2623e2761a11SOGAWA Hirofumi p = d_ancestor(p2, p1); 2624e2761a11SOGAWA Hirofumi if (p) { 2625f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT); 2626f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD); 26271da177e4SLinus Torvalds return p; 26281da177e4SLinus Torvalds } 26291da177e4SLinus Torvalds 2630e2761a11SOGAWA Hirofumi p = d_ancestor(p1, p2); 2631e2761a11SOGAWA Hirofumi if (p) { 2632f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 2633f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); 26341da177e4SLinus Torvalds return p; 26351da177e4SLinus Torvalds } 26361da177e4SLinus Torvalds 2637f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 2638d1b72cc6SMiklos Szeredi mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT2); 26391da177e4SLinus Torvalds return NULL; 26401da177e4SLinus Torvalds } 26414d359507SAl Viro EXPORT_SYMBOL(lock_rename); 26421da177e4SLinus Torvalds 26431da177e4SLinus Torvalds void unlock_rename(struct dentry *p1, struct dentry *p2) 26441da177e4SLinus Torvalds { 26451b1dcc1bSJes Sorensen mutex_unlock(&p1->d_inode->i_mutex); 26461da177e4SLinus Torvalds if (p1 != p2) { 26471b1dcc1bSJes Sorensen mutex_unlock(&p2->d_inode->i_mutex); 2648a11f3a05SArjan van de Ven mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 26491da177e4SLinus Torvalds } 26501da177e4SLinus Torvalds } 26514d359507SAl Viro EXPORT_SYMBOL(unlock_rename); 26521da177e4SLinus Torvalds 26534acdaf27SAl Viro int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, 2654312b63fbSAl Viro bool want_excl) 26551da177e4SLinus Torvalds { 2656a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 26571da177e4SLinus Torvalds if (error) 26581da177e4SLinus Torvalds return error; 26591da177e4SLinus Torvalds 2660acfa4380SAl Viro if (!dir->i_op->create) 26611da177e4SLinus Torvalds return -EACCES; /* shouldn't it be ENOSYS? */ 26621da177e4SLinus Torvalds mode &= S_IALLUGO; 26631da177e4SLinus Torvalds mode |= S_IFREG; 26641da177e4SLinus Torvalds error = security_inode_create(dir, dentry, mode); 26651da177e4SLinus Torvalds if (error) 26661da177e4SLinus Torvalds return error; 2667312b63fbSAl Viro error = dir->i_op->create(dir, dentry, mode, want_excl); 2668a74574aaSStephen Smalley if (!error) 2669f38aa942SAmy Griffis fsnotify_create(dir, dentry); 26701da177e4SLinus Torvalds return error; 26711da177e4SLinus Torvalds } 26724d359507SAl Viro EXPORT_SYMBOL(vfs_create); 26731da177e4SLinus Torvalds 267473d049a4SAl Viro static int may_open(struct path *path, int acc_mode, int flag) 26751da177e4SLinus Torvalds { 26763fb64190SChristoph Hellwig struct dentry *dentry = path->dentry; 26771da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 26781da177e4SLinus Torvalds int error; 26791da177e4SLinus Torvalds 2680bcda7652SAl Viro /* O_PATH? */ 2681bcda7652SAl Viro if (!acc_mode) 2682bcda7652SAl Viro return 0; 2683bcda7652SAl Viro 26841da177e4SLinus Torvalds if (!inode) 26851da177e4SLinus Torvalds return -ENOENT; 26861da177e4SLinus Torvalds 2687c8fe8f30SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 2688c8fe8f30SChristoph Hellwig case S_IFLNK: 26891da177e4SLinus Torvalds return -ELOOP; 2690c8fe8f30SChristoph Hellwig case S_IFDIR: 2691c8fe8f30SChristoph Hellwig if (acc_mode & MAY_WRITE) 26921da177e4SLinus Torvalds return -EISDIR; 2693c8fe8f30SChristoph Hellwig break; 2694c8fe8f30SChristoph Hellwig case S_IFBLK: 2695c8fe8f30SChristoph Hellwig case S_IFCHR: 26963fb64190SChristoph Hellwig if (path->mnt->mnt_flags & MNT_NODEV) 26971da177e4SLinus Torvalds return -EACCES; 2698c8fe8f30SChristoph Hellwig /*FALLTHRU*/ 2699c8fe8f30SChristoph Hellwig case S_IFIFO: 2700c8fe8f30SChristoph Hellwig case S_IFSOCK: 27011da177e4SLinus Torvalds flag &= ~O_TRUNC; 2702c8fe8f30SChristoph Hellwig break; 27034a3fd211SDave Hansen } 2704b41572e9SDave Hansen 27053fb64190SChristoph Hellwig error = inode_permission(inode, acc_mode); 2706b41572e9SDave Hansen if (error) 2707b41572e9SDave Hansen return error; 27086146f0d5SMimi Zohar 27091da177e4SLinus Torvalds /* 27101da177e4SLinus Torvalds * An append-only file must be opened in append mode for writing. 27111da177e4SLinus Torvalds */ 27121da177e4SLinus Torvalds if (IS_APPEND(inode)) { 27138737c930SAl Viro if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) 27147715b521SAl Viro return -EPERM; 27151da177e4SLinus Torvalds if (flag & O_TRUNC) 27167715b521SAl Viro return -EPERM; 27171da177e4SLinus Torvalds } 27181da177e4SLinus Torvalds 27191da177e4SLinus Torvalds /* O_NOATIME can only be set by the owner or superuser */ 27202e149670SSerge E. Hallyn if (flag & O_NOATIME && !inode_owner_or_capable(inode)) 27217715b521SAl Viro return -EPERM; 27221da177e4SLinus Torvalds 2723f3c7691eSJ. Bruce Fields return 0; 27247715b521SAl Viro } 27257715b521SAl Viro 2726e1181ee6SJeff Layton static int handle_truncate(struct file *filp) 27277715b521SAl Viro { 2728e1181ee6SJeff Layton struct path *path = &filp->f_path; 27297715b521SAl Viro struct inode *inode = path->dentry->d_inode; 27307715b521SAl Viro int error = get_write_access(inode); 27311da177e4SLinus Torvalds if (error) 27327715b521SAl Viro return error; 27331da177e4SLinus Torvalds /* 27341da177e4SLinus Torvalds * Refuse to truncate files with mandatory locks held on them. 27351da177e4SLinus Torvalds */ 2736d7a06983SJeff Layton error = locks_verify_locked(filp); 2737be6d3e56SKentaro Takeda if (!error) 2738ea0d3ab2STetsuo Handa error = security_path_truncate(path); 27391da177e4SLinus Torvalds if (!error) { 27407715b521SAl Viro error = do_truncate(path->dentry, 0, 2741d139d7ffSMiklos Szeredi ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, 2742e1181ee6SJeff Layton filp); 27431da177e4SLinus Torvalds } 27441da177e4SLinus Torvalds put_write_access(inode); 2745acd0c935SMimi Zohar return error; 27461da177e4SLinus Torvalds } 27471da177e4SLinus Torvalds 2748d57999e1SDave Hansen static inline int open_to_namei_flags(int flag) 2749d57999e1SDave Hansen { 27508a5e929dSAl Viro if ((flag & O_ACCMODE) == 3) 27518a5e929dSAl Viro flag--; 2752d57999e1SDave Hansen return flag; 2753d57999e1SDave Hansen } 2754d57999e1SDave Hansen 2755d18e9008SMiklos Szeredi static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode) 2756d18e9008SMiklos Szeredi { 2757d18e9008SMiklos Szeredi int error = security_path_mknod(dir, dentry, mode, 0); 2758d18e9008SMiklos Szeredi if (error) 2759d18e9008SMiklos Szeredi return error; 2760d18e9008SMiklos Szeredi 2761d18e9008SMiklos Szeredi error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC); 2762d18e9008SMiklos Szeredi if (error) 2763d18e9008SMiklos Szeredi return error; 2764d18e9008SMiklos Szeredi 2765d18e9008SMiklos Szeredi return security_inode_create(dir->dentry->d_inode, dentry, mode); 2766d18e9008SMiklos Szeredi } 2767d18e9008SMiklos Szeredi 27681acf0af9SDavid Howells /* 27691acf0af9SDavid Howells * Attempt to atomically look up, create and open a file from a negative 27701acf0af9SDavid Howells * dentry. 27711acf0af9SDavid Howells * 27721acf0af9SDavid Howells * Returns 0 if successful. The file will have been created and attached to 27731acf0af9SDavid Howells * @file by the filesystem calling finish_open(). 27741acf0af9SDavid Howells * 27751acf0af9SDavid Howells * Returns 1 if the file was looked up only or didn't need creating. The 27761acf0af9SDavid Howells * caller will need to perform the open themselves. @path will have been 27771acf0af9SDavid Howells * updated to point to the new dentry. This may be negative. 27781acf0af9SDavid Howells * 27791acf0af9SDavid Howells * Returns an error code otherwise. 27801acf0af9SDavid Howells */ 27812675a4ebSAl Viro static int atomic_open(struct nameidata *nd, struct dentry *dentry, 278230d90494SAl Viro struct path *path, struct file *file, 2783015c3bbcSMiklos Szeredi const struct open_flags *op, 278464894cf8SAl Viro bool got_write, bool need_lookup, 278547237687SAl Viro int *opened) 2786d18e9008SMiklos Szeredi { 2787d18e9008SMiklos Szeredi struct inode *dir = nd->path.dentry->d_inode; 2788d18e9008SMiklos Szeredi unsigned open_flag = open_to_namei_flags(op->open_flag); 2789d18e9008SMiklos Szeredi umode_t mode; 2790d18e9008SMiklos Szeredi int error; 2791d18e9008SMiklos Szeredi int acc_mode; 2792d18e9008SMiklos Szeredi int create_error = 0; 2793d18e9008SMiklos Szeredi struct dentry *const DENTRY_NOT_SET = (void *) -1UL; 2794116cc022SMiklos Szeredi bool excl; 2795d18e9008SMiklos Szeredi 2796d18e9008SMiklos Szeredi BUG_ON(dentry->d_inode); 2797d18e9008SMiklos Szeredi 2798d18e9008SMiklos Szeredi /* Don't create child dentry for a dead directory. */ 2799d18e9008SMiklos Szeredi if (unlikely(IS_DEADDIR(dir))) { 28002675a4ebSAl Viro error = -ENOENT; 2801d18e9008SMiklos Szeredi goto out; 2802d18e9008SMiklos Szeredi } 2803d18e9008SMiklos Szeredi 280462b259d8SMiklos Szeredi mode = op->mode; 2805d18e9008SMiklos Szeredi if ((open_flag & O_CREAT) && !IS_POSIXACL(dir)) 2806d18e9008SMiklos Szeredi mode &= ~current_umask(); 2807d18e9008SMiklos Szeredi 2808116cc022SMiklos Szeredi excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT); 2809116cc022SMiklos Szeredi if (excl) 2810d18e9008SMiklos Szeredi open_flag &= ~O_TRUNC; 2811d18e9008SMiklos Szeredi 2812d18e9008SMiklos Szeredi /* 2813d18e9008SMiklos Szeredi * Checking write permission is tricky, bacuse we don't know if we are 2814d18e9008SMiklos Szeredi * going to actually need it: O_CREAT opens should work as long as the 2815d18e9008SMiklos Szeredi * file exists. But checking existence breaks atomicity. The trick is 2816d18e9008SMiklos Szeredi * to check access and if not granted clear O_CREAT from the flags. 2817d18e9008SMiklos Szeredi * 2818d18e9008SMiklos Szeredi * Another problem is returing the "right" error value (e.g. for an 2819d18e9008SMiklos Szeredi * O_EXCL open we want to return EEXIST not EROFS). 2820d18e9008SMiklos Szeredi */ 282164894cf8SAl Viro if (((open_flag & (O_CREAT | O_TRUNC)) || 282264894cf8SAl Viro (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) { 282364894cf8SAl Viro if (!(open_flag & O_CREAT)) { 2824d18e9008SMiklos Szeredi /* 2825d18e9008SMiklos Szeredi * No O_CREATE -> atomicity not a requirement -> fall 2826d18e9008SMiklos Szeredi * back to lookup + open 2827d18e9008SMiklos Szeredi */ 2828d18e9008SMiklos Szeredi goto no_open; 2829d18e9008SMiklos Szeredi } else if (open_flag & (O_EXCL | O_TRUNC)) { 2830d18e9008SMiklos Szeredi /* Fall back and fail with the right error */ 283164894cf8SAl Viro create_error = -EROFS; 2832d18e9008SMiklos Szeredi goto no_open; 2833d18e9008SMiklos Szeredi } else { 2834d18e9008SMiklos Szeredi /* No side effects, safe to clear O_CREAT */ 283564894cf8SAl Viro create_error = -EROFS; 2836d18e9008SMiklos Szeredi open_flag &= ~O_CREAT; 2837d18e9008SMiklos Szeredi } 2838d18e9008SMiklos Szeredi } 2839d18e9008SMiklos Szeredi 2840d18e9008SMiklos Szeredi if (open_flag & O_CREAT) { 284138227f78SMiklos Szeredi error = may_o_create(&nd->path, dentry, mode); 2842d18e9008SMiklos Szeredi if (error) { 2843d18e9008SMiklos Szeredi create_error = error; 2844d18e9008SMiklos Szeredi if (open_flag & O_EXCL) 2845d18e9008SMiklos Szeredi goto no_open; 2846d18e9008SMiklos Szeredi open_flag &= ~O_CREAT; 2847d18e9008SMiklos Szeredi } 2848d18e9008SMiklos Szeredi } 2849d18e9008SMiklos Szeredi 2850d18e9008SMiklos Szeredi if (nd->flags & LOOKUP_DIRECTORY) 2851d18e9008SMiklos Szeredi open_flag |= O_DIRECTORY; 2852d18e9008SMiklos Szeredi 285330d90494SAl Viro file->f_path.dentry = DENTRY_NOT_SET; 285430d90494SAl Viro file->f_path.mnt = nd->path.mnt; 285530d90494SAl Viro error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode, 285647237687SAl Viro opened); 2857d9585277SAl Viro if (error < 0) { 2858d9585277SAl Viro if (create_error && error == -ENOENT) 2859d9585277SAl Viro error = create_error; 2860d18e9008SMiklos Szeredi goto out; 2861d18e9008SMiklos Szeredi } 2862d18e9008SMiklos Szeredi 2863d9585277SAl Viro if (error) { /* returned 1, that is */ 286430d90494SAl Viro if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { 28652675a4ebSAl Viro error = -EIO; 2866d18e9008SMiklos Szeredi goto out; 2867d18e9008SMiklos Szeredi } 286830d90494SAl Viro if (file->f_path.dentry) { 2869d18e9008SMiklos Szeredi dput(dentry); 287030d90494SAl Viro dentry = file->f_path.dentry; 2871d18e9008SMiklos Szeredi } 287203da633aSAl Viro if (*opened & FILE_CREATED) 287303da633aSAl Viro fsnotify_create(dir, dentry); 287403da633aSAl Viro if (!dentry->d_inode) { 287503da633aSAl Viro WARN_ON(*opened & FILE_CREATED); 287603da633aSAl Viro if (create_error) { 287762b2ce96SSage Weil error = create_error; 287862b2ce96SSage Weil goto out; 287962b2ce96SSage Weil } 288003da633aSAl Viro } else { 288103da633aSAl Viro if (excl && !(*opened & FILE_CREATED)) { 288203da633aSAl Viro error = -EEXIST; 288303da633aSAl Viro goto out; 288403da633aSAl Viro } 288503da633aSAl Viro } 2886d18e9008SMiklos Szeredi goto looked_up; 2887d18e9008SMiklos Szeredi } 2888d18e9008SMiklos Szeredi 2889d18e9008SMiklos Szeredi /* 2890d18e9008SMiklos Szeredi * We didn't have the inode before the open, so check open permission 2891d18e9008SMiklos Szeredi * here. 2892d18e9008SMiklos Szeredi */ 289303da633aSAl Viro acc_mode = op->acc_mode; 289403da633aSAl Viro if (*opened & FILE_CREATED) { 289503da633aSAl Viro WARN_ON(!(open_flag & O_CREAT)); 289603da633aSAl Viro fsnotify_create(dir, dentry); 289703da633aSAl Viro acc_mode = MAY_OPEN; 289803da633aSAl Viro } 28992675a4ebSAl Viro error = may_open(&file->f_path, acc_mode, open_flag); 29002675a4ebSAl Viro if (error) 29012675a4ebSAl Viro fput(file); 2902d18e9008SMiklos Szeredi 2903d18e9008SMiklos Szeredi out: 2904d18e9008SMiklos Szeredi dput(dentry); 29052675a4ebSAl Viro return error; 2906d18e9008SMiklos Szeredi 2907d18e9008SMiklos Szeredi no_open: 2908d18e9008SMiklos Szeredi if (need_lookup) { 290972bd866aSAl Viro dentry = lookup_real(dir, dentry, nd->flags); 2910d18e9008SMiklos Szeredi if (IS_ERR(dentry)) 29112675a4ebSAl Viro return PTR_ERR(dentry); 2912d18e9008SMiklos Szeredi 2913d18e9008SMiklos Szeredi if (create_error) { 2914d18e9008SMiklos Szeredi int open_flag = op->open_flag; 2915d18e9008SMiklos Szeredi 29162675a4ebSAl Viro error = create_error; 2917d18e9008SMiklos Szeredi if ((open_flag & O_EXCL)) { 2918d18e9008SMiklos Szeredi if (!dentry->d_inode) 2919d18e9008SMiklos Szeredi goto out; 2920d18e9008SMiklos Szeredi } else if (!dentry->d_inode) { 2921d18e9008SMiklos Szeredi goto out; 2922d18e9008SMiklos Szeredi } else if ((open_flag & O_TRUNC) && 2923e36cb0b8SDavid Howells d_is_reg(dentry)) { 2924d18e9008SMiklos Szeredi goto out; 2925d18e9008SMiklos Szeredi } 2926d18e9008SMiklos Szeredi /* will fail later, go on to get the right error */ 2927d18e9008SMiklos Szeredi } 2928d18e9008SMiklos Szeredi } 2929d18e9008SMiklos Szeredi looked_up: 2930d18e9008SMiklos Szeredi path->dentry = dentry; 2931d18e9008SMiklos Szeredi path->mnt = nd->path.mnt; 29322675a4ebSAl Viro return 1; 2933d18e9008SMiklos Szeredi } 2934d18e9008SMiklos Szeredi 293531e6b01fSNick Piggin /* 29361acf0af9SDavid Howells * Look up and maybe create and open the last component. 2937d58ffd35SMiklos Szeredi * 2938d58ffd35SMiklos Szeredi * Must be called with i_mutex held on parent. 2939d58ffd35SMiklos Szeredi * 29401acf0af9SDavid Howells * Returns 0 if the file was successfully atomically created (if necessary) and 29411acf0af9SDavid Howells * opened. In this case the file will be returned attached to @file. 29421acf0af9SDavid Howells * 29431acf0af9SDavid Howells * Returns 1 if the file was not completely opened at this time, though lookups 29441acf0af9SDavid Howells * and creations will have been performed and the dentry returned in @path will 29451acf0af9SDavid Howells * be positive upon return if O_CREAT was specified. If O_CREAT wasn't 29461acf0af9SDavid Howells * specified then a negative dentry may be returned. 29471acf0af9SDavid Howells * 29481acf0af9SDavid Howells * An error code is returned otherwise. 29491acf0af9SDavid Howells * 29501acf0af9SDavid Howells * FILE_CREATE will be set in @*opened if the dentry was created and will be 29511acf0af9SDavid Howells * cleared otherwise prior to returning. 2952d58ffd35SMiklos Szeredi */ 29532675a4ebSAl Viro static int lookup_open(struct nameidata *nd, struct path *path, 295430d90494SAl Viro struct file *file, 2955d58ffd35SMiklos Szeredi const struct open_flags *op, 295664894cf8SAl Viro bool got_write, int *opened) 2957d58ffd35SMiklos Szeredi { 2958d58ffd35SMiklos Szeredi struct dentry *dir = nd->path.dentry; 295954ef4872SMiklos Szeredi struct inode *dir_inode = dir->d_inode; 2960d58ffd35SMiklos Szeredi struct dentry *dentry; 2961d58ffd35SMiklos Szeredi int error; 296254ef4872SMiklos Szeredi bool need_lookup; 2963d58ffd35SMiklos Szeredi 296447237687SAl Viro *opened &= ~FILE_CREATED; 2965201f956eSAl Viro dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup); 2966d58ffd35SMiklos Szeredi if (IS_ERR(dentry)) 29672675a4ebSAl Viro return PTR_ERR(dentry); 2968d58ffd35SMiklos Szeredi 2969d18e9008SMiklos Szeredi /* Cached positive dentry: will open in f_op->open */ 2970d18e9008SMiklos Szeredi if (!need_lookup && dentry->d_inode) 2971d18e9008SMiklos Szeredi goto out_no_open; 2972d18e9008SMiklos Szeredi 2973d18e9008SMiklos Szeredi if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) { 297464894cf8SAl Viro return atomic_open(nd, dentry, path, file, op, got_write, 297547237687SAl Viro need_lookup, opened); 2976d18e9008SMiklos Szeredi } 2977d18e9008SMiklos Szeredi 297854ef4872SMiklos Szeredi if (need_lookup) { 297954ef4872SMiklos Szeredi BUG_ON(dentry->d_inode); 298054ef4872SMiklos Szeredi 298172bd866aSAl Viro dentry = lookup_real(dir_inode, dentry, nd->flags); 298254ef4872SMiklos Szeredi if (IS_ERR(dentry)) 29832675a4ebSAl Viro return PTR_ERR(dentry); 298454ef4872SMiklos Szeredi } 298554ef4872SMiklos Szeredi 2986d58ffd35SMiklos Szeredi /* Negative dentry, just create the file */ 2987d58ffd35SMiklos Szeredi if (!dentry->d_inode && (op->open_flag & O_CREAT)) { 2988d58ffd35SMiklos Szeredi umode_t mode = op->mode; 2989d58ffd35SMiklos Szeredi if (!IS_POSIXACL(dir->d_inode)) 2990d58ffd35SMiklos Szeredi mode &= ~current_umask(); 2991d58ffd35SMiklos Szeredi /* 2992d58ffd35SMiklos Szeredi * This write is needed to ensure that a 2993d58ffd35SMiklos Szeredi * rw->ro transition does not occur between 2994d58ffd35SMiklos Szeredi * the time when the file is created and when 2995d58ffd35SMiklos Szeredi * a permanent write count is taken through 2996015c3bbcSMiklos Szeredi * the 'struct file' in finish_open(). 2997d58ffd35SMiklos Szeredi */ 299864894cf8SAl Viro if (!got_write) { 299964894cf8SAl Viro error = -EROFS; 3000d58ffd35SMiklos Szeredi goto out_dput; 300164894cf8SAl Viro } 300247237687SAl Viro *opened |= FILE_CREATED; 3003d58ffd35SMiklos Szeredi error = security_path_mknod(&nd->path, dentry, mode, 0); 3004d58ffd35SMiklos Szeredi if (error) 3005d58ffd35SMiklos Szeredi goto out_dput; 3006312b63fbSAl Viro error = vfs_create(dir->d_inode, dentry, mode, 3007312b63fbSAl Viro nd->flags & LOOKUP_EXCL); 3008d58ffd35SMiklos Szeredi if (error) 3009d58ffd35SMiklos Szeredi goto out_dput; 3010d58ffd35SMiklos Szeredi } 3011d18e9008SMiklos Szeredi out_no_open: 3012d58ffd35SMiklos Szeredi path->dentry = dentry; 3013d58ffd35SMiklos Szeredi path->mnt = nd->path.mnt; 30142675a4ebSAl Viro return 1; 3015d58ffd35SMiklos Szeredi 3016d58ffd35SMiklos Szeredi out_dput: 3017d58ffd35SMiklos Szeredi dput(dentry); 30182675a4ebSAl Viro return error; 3019d58ffd35SMiklos Szeredi } 3020d58ffd35SMiklos Szeredi 3021d58ffd35SMiklos Szeredi /* 3022fe2d35ffSAl Viro * Handle the last step of open() 302331e6b01fSNick Piggin */ 3024896475d5SAl Viro static int do_last(struct nameidata *nd, 302530d90494SAl Viro struct file *file, const struct open_flags *op, 302676ae2a5aSAl Viro int *opened) 3027fb1cc555SAl Viro { 3028a1e28038SAl Viro struct dentry *dir = nd->path.dentry; 3029ca344a89SAl Viro int open_flag = op->open_flag; 303077d660a8SMiklos Szeredi bool will_truncate = (open_flag & O_TRUNC) != 0; 303164894cf8SAl Viro bool got_write = false; 3032bcda7652SAl Viro int acc_mode = op->acc_mode; 3033254cf582SAl Viro unsigned seq; 3034a1eb3315SMiklos Szeredi struct inode *inode; 303516b1c1cdSMiklos Szeredi struct path save_parent = { .dentry = NULL, .mnt = NULL }; 3036896475d5SAl Viro struct path path; 303716b1c1cdSMiklos Szeredi bool retried = false; 303816c2cd71SAl Viro int error; 3039fb1cc555SAl Viro 3040c3e380b0SAl Viro nd->flags &= ~LOOKUP_PARENT; 3041c3e380b0SAl Viro nd->flags |= op->intent; 3042c3e380b0SAl Viro 3043bc77daa7SAl Viro if (nd->last_type != LAST_NORM) { 3044fe2d35ffSAl Viro error = handle_dots(nd, nd->last_type); 3045deb106c6SAl Viro if (unlikely(error)) 30462675a4ebSAl Viro return error; 3047e83db167SMiklos Szeredi goto finish_open; 30481f36f774SAl Viro } 3049a2c36b45SAl Viro 3050ca344a89SAl Viro if (!(open_flag & O_CREAT)) { 3051fe2d35ffSAl Viro if (nd->last.name[nd->last.len]) 3052fe2d35ffSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 3053fe2d35ffSAl Viro /* we _can_ be in RCU mode here */ 3054254cf582SAl Viro error = lookup_fast(nd, &path, &inode, &seq); 305571574865SMiklos Szeredi if (likely(!error)) 305671574865SMiklos Szeredi goto finish_lookup; 305771574865SMiklos Szeredi 3058ce57dfc1SAl Viro if (error < 0) 3059deb106c6SAl Viro return error; 3060a1eb3315SMiklos Szeredi 306137d7fffcSMiklos Szeredi BUG_ON(nd->inode != dir->d_inode); 3062b6183df7SMiklos Szeredi } else { 3063fe2d35ffSAl Viro /* create side of things */ 3064a3fbbde7SAl Viro /* 3065b6183df7SMiklos Szeredi * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED 3066b6183df7SMiklos Szeredi * has been cleared when we got to the last component we are 3067b6183df7SMiklos Szeredi * about to look up 3068a3fbbde7SAl Viro */ 30699f1fafeeSAl Viro error = complete_walk(nd); 3070e8bb73dfSAl Viro if (error) 30712675a4ebSAl Viro return error; 3072fe2d35ffSAl Viro 307376ae2a5aSAl Viro audit_inode(nd->name, dir, LOOKUP_PARENT); 30741f36f774SAl Viro /* trailing slashes? */ 3075deb106c6SAl Viro if (unlikely(nd->last.name[nd->last.len])) 3076deb106c6SAl Viro return -EISDIR; 3077b6183df7SMiklos Szeredi } 30781f36f774SAl Viro 307916b1c1cdSMiklos Szeredi retry_lookup: 308064894cf8SAl Viro if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { 308164894cf8SAl Viro error = mnt_want_write(nd->path.mnt); 308264894cf8SAl Viro if (!error) 308364894cf8SAl Viro got_write = true; 308464894cf8SAl Viro /* 308564894cf8SAl Viro * do _not_ fail yet - we might not need that or fail with 308664894cf8SAl Viro * a different error; let lookup_open() decide; we'll be 308764894cf8SAl Viro * dropping this one anyway. 308864894cf8SAl Viro */ 308964894cf8SAl Viro } 3090a1e28038SAl Viro mutex_lock(&dir->d_inode->i_mutex); 3091896475d5SAl Viro error = lookup_open(nd, &path, file, op, got_write, opened); 3092fb1cc555SAl Viro mutex_unlock(&dir->d_inode->i_mutex); 3093fb1cc555SAl Viro 30942675a4ebSAl Viro if (error <= 0) { 30952675a4ebSAl Viro if (error) 3096d58ffd35SMiklos Szeredi goto out; 30976c0d46c4SAl Viro 309847237687SAl Viro if ((*opened & FILE_CREATED) || 3099496ad9aaSAl Viro !S_ISREG(file_inode(file)->i_mode)) 310077d660a8SMiklos Szeredi will_truncate = false; 3101d18e9008SMiklos Szeredi 310276ae2a5aSAl Viro audit_inode(nd->name, file->f_path.dentry, 0); 3103d18e9008SMiklos Szeredi goto opened; 3104d18e9008SMiklos Szeredi } 3105d18e9008SMiklos Szeredi 310647237687SAl Viro if (*opened & FILE_CREATED) { 31079b44f1b3SAl Viro /* Don't check for write permission, don't truncate */ 3108ca344a89SAl Viro open_flag &= ~O_TRUNC; 310977d660a8SMiklos Szeredi will_truncate = false; 3110bcda7652SAl Viro acc_mode = MAY_OPEN; 3111896475d5SAl Viro path_to_nameidata(&path, nd); 3112e83db167SMiklos Szeredi goto finish_open_created; 3113fb1cc555SAl Viro } 3114fb1cc555SAl Viro 3115fb1cc555SAl Viro /* 31163134f37eSJeff Layton * create/update audit record if it already exists. 3117fb1cc555SAl Viro */ 3118896475d5SAl Viro if (d_is_positive(path.dentry)) 311976ae2a5aSAl Viro audit_inode(nd->name, path.dentry, 0); 3120fb1cc555SAl Viro 3121d18e9008SMiklos Szeredi /* 3122d18e9008SMiklos Szeredi * If atomic_open() acquired write access it is dropped now due to 3123d18e9008SMiklos Szeredi * possible mount and symlink following (this might be optimized away if 3124d18e9008SMiklos Szeredi * necessary...) 3125d18e9008SMiklos Szeredi */ 312664894cf8SAl Viro if (got_write) { 3127d18e9008SMiklos Szeredi mnt_drop_write(nd->path.mnt); 312864894cf8SAl Viro got_write = false; 3129d18e9008SMiklos Szeredi } 3130d18e9008SMiklos Szeredi 3131deb106c6SAl Viro if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) { 3132deb106c6SAl Viro path_to_nameidata(&path, nd); 3133deb106c6SAl Viro return -EEXIST; 3134deb106c6SAl Viro } 3135fb1cc555SAl Viro 3136756daf26SNeilBrown error = follow_managed(&path, nd); 3137deb106c6SAl Viro if (unlikely(error < 0)) 3138deb106c6SAl Viro return error; 3139a3fbbde7SAl Viro 3140decf3400SMiklos Szeredi BUG_ON(nd->flags & LOOKUP_RCU); 314163afdfc7SDavid Howells inode = d_backing_inode(path.dentry); 3142254cf582SAl Viro seq = 0; /* out of RCU mode, so the value doesn't matter */ 3143deb106c6SAl Viro if (unlikely(d_is_negative(path.dentry))) { 3144896475d5SAl Viro path_to_nameidata(&path, nd); 3145deb106c6SAl Viro return -ENOENT; 314654c33e7fSMiklos Szeredi } 3147766c4cbfSAl Viro finish_lookup: 3148191d7f73SAl Viro if (nd->depth) 3149191d7f73SAl Viro put_link(nd); 3150181548c0SAl Viro error = should_follow_link(nd, &path, nd->flags & LOOKUP_FOLLOW, 3151181548c0SAl Viro inode, seq); 3152deb106c6SAl Viro if (unlikely(error)) 3153d63ff28fSAl Viro return error; 3154fb1cc555SAl Viro 3155896475d5SAl Viro if (unlikely(d_is_symlink(path.dentry)) && !(open_flag & O_PATH)) { 3156896475d5SAl Viro path_to_nameidata(&path, nd); 3157deb106c6SAl Viro return -ELOOP; 3158a5cfe2d5SAl Viro } 3159a5cfe2d5SAl Viro 3160896475d5SAl Viro if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path.mnt) { 3161896475d5SAl Viro path_to_nameidata(&path, nd); 316216b1c1cdSMiklos Szeredi } else { 316316b1c1cdSMiklos Szeredi save_parent.dentry = nd->path.dentry; 3164896475d5SAl Viro save_parent.mnt = mntget(path.mnt); 3165896475d5SAl Viro nd->path.dentry = path.dentry; 316616b1c1cdSMiklos Szeredi 316716b1c1cdSMiklos Szeredi } 3168decf3400SMiklos Szeredi nd->inode = inode; 3169254cf582SAl Viro nd->seq = seq; 3170a3fbbde7SAl Viro /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ 3171bc77daa7SAl Viro finish_open: 3172a3fbbde7SAl Viro error = complete_walk(nd); 317316b1c1cdSMiklos Szeredi if (error) { 317416b1c1cdSMiklos Szeredi path_put(&save_parent); 31752675a4ebSAl Viro return error; 317616b1c1cdSMiklos Szeredi } 317776ae2a5aSAl Viro audit_inode(nd->name, nd->path.dentry, 0); 3178fb1cc555SAl Viro error = -EISDIR; 317944b1d530SMiklos Szeredi if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry)) 31802675a4ebSAl Viro goto out; 3181af2f5542SMiklos Szeredi error = -ENOTDIR; 318244b1d530SMiklos Szeredi if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry)) 31832675a4ebSAl Viro goto out; 31844bbcbd3bSDavid Howells if (!d_is_reg(nd->path.dentry)) 318577d660a8SMiklos Szeredi will_truncate = false; 31866c0d46c4SAl Viro 31870f9d1a10SAl Viro if (will_truncate) { 31880f9d1a10SAl Viro error = mnt_want_write(nd->path.mnt); 31890f9d1a10SAl Viro if (error) 31902675a4ebSAl Viro goto out; 319164894cf8SAl Viro got_write = true; 31920f9d1a10SAl Viro } 3193e83db167SMiklos Szeredi finish_open_created: 3194bcda7652SAl Viro error = may_open(&nd->path, acc_mode, open_flag); 3195ca344a89SAl Viro if (error) 31962675a4ebSAl Viro goto out; 31974aa7c634SMiklos Szeredi 31984aa7c634SMiklos Szeredi BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */ 31994aa7c634SMiklos Szeredi error = vfs_open(&nd->path, file, current_cred()); 32004aa7c634SMiklos Szeredi if (!error) { 32014aa7c634SMiklos Szeredi *opened |= FILE_OPENED; 32024aa7c634SMiklos Szeredi } else { 320330d90494SAl Viro if (error == -EOPENSTALE) 3204f60dc3dbSMiklos Szeredi goto stale_open; 3205015c3bbcSMiklos Szeredi goto out; 3206f60dc3dbSMiklos Szeredi } 3207a8277b9bSMiklos Szeredi opened: 32082675a4ebSAl Viro error = open_check_o_direct(file); 3209015c3bbcSMiklos Szeredi if (error) 3210015c3bbcSMiklos Szeredi goto exit_fput; 32113034a146SDmitry Kasatkin error = ima_file_check(file, op->acc_mode, *opened); 3212aa4caadbSMiklos Szeredi if (error) 3213aa4caadbSMiklos Szeredi goto exit_fput; 3214aa4caadbSMiklos Szeredi 32150f9d1a10SAl Viro if (will_truncate) { 32162675a4ebSAl Viro error = handle_truncate(file); 3217aa4caadbSMiklos Szeredi if (error) 3218aa4caadbSMiklos Szeredi goto exit_fput; 32190f9d1a10SAl Viro } 3220ca344a89SAl Viro out: 322164894cf8SAl Viro if (got_write) 32220f9d1a10SAl Viro mnt_drop_write(nd->path.mnt); 322316b1c1cdSMiklos Szeredi path_put(&save_parent); 32242675a4ebSAl Viro return error; 3225fb1cc555SAl Viro 3226015c3bbcSMiklos Szeredi exit_fput: 32272675a4ebSAl Viro fput(file); 32282675a4ebSAl Viro goto out; 3229015c3bbcSMiklos Szeredi 3230f60dc3dbSMiklos Szeredi stale_open: 3231f60dc3dbSMiklos Szeredi /* If no saved parent or already retried then can't retry */ 3232f60dc3dbSMiklos Szeredi if (!save_parent.dentry || retried) 3233f60dc3dbSMiklos Szeredi goto out; 3234f60dc3dbSMiklos Szeredi 3235f60dc3dbSMiklos Szeredi BUG_ON(save_parent.dentry != dir); 3236f60dc3dbSMiklos Szeredi path_put(&nd->path); 3237f60dc3dbSMiklos Szeredi nd->path = save_parent; 3238f60dc3dbSMiklos Szeredi nd->inode = dir->d_inode; 3239f60dc3dbSMiklos Szeredi save_parent.mnt = NULL; 3240f60dc3dbSMiklos Szeredi save_parent.dentry = NULL; 324164894cf8SAl Viro if (got_write) { 3242f60dc3dbSMiklos Szeredi mnt_drop_write(nd->path.mnt); 324364894cf8SAl Viro got_write = false; 3244f60dc3dbSMiklos Szeredi } 3245f60dc3dbSMiklos Szeredi retried = true; 3246f60dc3dbSMiklos Szeredi goto retry_lookup; 3247fb1cc555SAl Viro } 3248fb1cc555SAl Viro 3249c8a53ee5SAl Viro static int do_tmpfile(struct nameidata *nd, unsigned flags, 325060545d0dSAl Viro const struct open_flags *op, 325160545d0dSAl Viro struct file *file, int *opened) 325260545d0dSAl Viro { 325360545d0dSAl Viro static const struct qstr name = QSTR_INIT("/", 1); 3254625b6d10SAl Viro struct dentry *child; 325560545d0dSAl Viro struct inode *dir; 3256625b6d10SAl Viro struct path path; 3257c8a53ee5SAl Viro int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path); 325860545d0dSAl Viro if (unlikely(error)) 325960545d0dSAl Viro return error; 3260625b6d10SAl Viro error = mnt_want_write(path.mnt); 326160545d0dSAl Viro if (unlikely(error)) 326260545d0dSAl Viro goto out; 3263625b6d10SAl Viro dir = path.dentry->d_inode; 326460545d0dSAl Viro /* we want directory to be writable */ 3265625b6d10SAl Viro error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 326660545d0dSAl Viro if (error) 326760545d0dSAl Viro goto out2; 326860545d0dSAl Viro if (!dir->i_op->tmpfile) { 326960545d0dSAl Viro error = -EOPNOTSUPP; 327060545d0dSAl Viro goto out2; 327160545d0dSAl Viro } 3272625b6d10SAl Viro child = d_alloc(path.dentry, &name); 327360545d0dSAl Viro if (unlikely(!child)) { 327460545d0dSAl Viro error = -ENOMEM; 327560545d0dSAl Viro goto out2; 327660545d0dSAl Viro } 3277625b6d10SAl Viro dput(path.dentry); 3278625b6d10SAl Viro path.dentry = child; 3279625b6d10SAl Viro error = dir->i_op->tmpfile(dir, child, op->mode); 328060545d0dSAl Viro if (error) 328160545d0dSAl Viro goto out2; 3282c8a53ee5SAl Viro audit_inode(nd->name, child, 0); 328369a91c23SEric Rannaud /* Don't check for other permissions, the inode was just created */ 3284625b6d10SAl Viro error = may_open(&path, MAY_OPEN, op->open_flag); 328560545d0dSAl Viro if (error) 328660545d0dSAl Viro goto out2; 3287625b6d10SAl Viro file->f_path.mnt = path.mnt; 3288625b6d10SAl Viro error = finish_open(file, child, NULL, opened); 328960545d0dSAl Viro if (error) 329060545d0dSAl Viro goto out2; 329160545d0dSAl Viro error = open_check_o_direct(file); 3292f4e0c30cSAl Viro if (error) { 329360545d0dSAl Viro fput(file); 3294f4e0c30cSAl Viro } else if (!(op->open_flag & O_EXCL)) { 3295f4e0c30cSAl Viro struct inode *inode = file_inode(file); 3296f4e0c30cSAl Viro spin_lock(&inode->i_lock); 3297f4e0c30cSAl Viro inode->i_state |= I_LINKABLE; 3298f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 3299f4e0c30cSAl Viro } 330060545d0dSAl Viro out2: 3301625b6d10SAl Viro mnt_drop_write(path.mnt); 330260545d0dSAl Viro out: 3303625b6d10SAl Viro path_put(&path); 330460545d0dSAl Viro return error; 330560545d0dSAl Viro } 330660545d0dSAl Viro 3307c8a53ee5SAl Viro static struct file *path_openat(struct nameidata *nd, 3308c8a53ee5SAl Viro const struct open_flags *op, unsigned flags) 33091da177e4SLinus Torvalds { 3310368ee9baSAl Viro const char *s; 331130d90494SAl Viro struct file *file; 331247237687SAl Viro int opened = 0; 331313aab428SAl Viro int error; 331431e6b01fSNick Piggin 331530d90494SAl Viro file = get_empty_filp(); 33161afc99beSAl Viro if (IS_ERR(file)) 33171afc99beSAl Viro return file; 331831e6b01fSNick Piggin 331930d90494SAl Viro file->f_flags = op->open_flag; 332031e6b01fSNick Piggin 3321bb458c64SAl Viro if (unlikely(file->f_flags & __O_TMPFILE)) { 3322c8a53ee5SAl Viro error = do_tmpfile(nd, flags, op, file, &opened); 3323f15133dfSAl Viro goto out2; 332460545d0dSAl Viro } 332560545d0dSAl Viro 3326c8a53ee5SAl Viro s = path_init(nd, flags); 3327368ee9baSAl Viro if (IS_ERR(s)) { 3328368ee9baSAl Viro put_filp(file); 3329368ee9baSAl Viro return ERR_CAST(s); 3330368ee9baSAl Viro } 33313bdba28bSAl Viro while (!(error = link_path_walk(s, nd)) && 333276ae2a5aSAl Viro (error = do_last(nd, file, op, &opened)) > 0) { 333373d049a4SAl Viro nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); 33343bdba28bSAl Viro s = trailing_symlink(nd); 33353bdba28bSAl Viro if (IS_ERR(s)) { 33363bdba28bSAl Viro error = PTR_ERR(s); 33372675a4ebSAl Viro break; 3338806b681cSAl Viro } 33393bdba28bSAl Viro } 3340deb106c6SAl Viro terminate_walk(nd); 3341f15133dfSAl Viro out2: 33422675a4ebSAl Viro if (!(opened & FILE_OPENED)) { 33432675a4ebSAl Viro BUG_ON(!error); 334430d90494SAl Viro put_filp(file); 3345015c3bbcSMiklos Szeredi } 33462675a4ebSAl Viro if (unlikely(error)) { 33472675a4ebSAl Viro if (error == -EOPENSTALE) { 33482675a4ebSAl Viro if (flags & LOOKUP_RCU) 33492675a4ebSAl Viro error = -ECHILD; 33502675a4ebSAl Viro else 33512675a4ebSAl Viro error = -ESTALE; 33522675a4ebSAl Viro } 33532675a4ebSAl Viro file = ERR_PTR(error); 33542675a4ebSAl Viro } 33552675a4ebSAl Viro return file; 3356de459215SKirill Korotaev } 33571da177e4SLinus Torvalds 3358669abf4eSJeff Layton struct file *do_filp_open(int dfd, struct filename *pathname, 3359f9652e10SAl Viro const struct open_flags *op) 336013aab428SAl Viro { 33619883d185SAl Viro struct nameidata nd; 3362f9652e10SAl Viro int flags = op->lookup_flags; 336313aab428SAl Viro struct file *filp; 336413aab428SAl Viro 33659883d185SAl Viro set_nameidata(&nd, dfd, pathname); 3366c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_RCU); 336713aab428SAl Viro if (unlikely(filp == ERR_PTR(-ECHILD))) 3368c8a53ee5SAl Viro filp = path_openat(&nd, op, flags); 336913aab428SAl Viro if (unlikely(filp == ERR_PTR(-ESTALE))) 3370c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_REVAL); 33719883d185SAl Viro restore_nameidata(); 337213aab428SAl Viro return filp; 337313aab428SAl Viro } 337413aab428SAl Viro 337573d049a4SAl Viro struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, 3376f9652e10SAl Viro const char *name, const struct open_flags *op) 337773d049a4SAl Viro { 33789883d185SAl Viro struct nameidata nd; 337973d049a4SAl Viro struct file *file; 338051689104SPaul Moore struct filename *filename; 3381f9652e10SAl Viro int flags = op->lookup_flags | LOOKUP_ROOT; 338273d049a4SAl Viro 338373d049a4SAl Viro nd.root.mnt = mnt; 338473d049a4SAl Viro nd.root.dentry = dentry; 338573d049a4SAl Viro 3386b18825a7SDavid Howells if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN) 338773d049a4SAl Viro return ERR_PTR(-ELOOP); 338873d049a4SAl Viro 338951689104SPaul Moore filename = getname_kernel(name); 3390a1c83681SViresh Kumar if (IS_ERR(filename)) 339151689104SPaul Moore return ERR_CAST(filename); 339251689104SPaul Moore 33939883d185SAl Viro set_nameidata(&nd, -1, filename); 3394c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_RCU); 339573d049a4SAl Viro if (unlikely(file == ERR_PTR(-ECHILD))) 3396c8a53ee5SAl Viro file = path_openat(&nd, op, flags); 339773d049a4SAl Viro if (unlikely(file == ERR_PTR(-ESTALE))) 3398c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_REVAL); 33999883d185SAl Viro restore_nameidata(); 340051689104SPaul Moore putname(filename); 340173d049a4SAl Viro return file; 340273d049a4SAl Viro } 340373d049a4SAl Viro 3404fa14a0b8SAl Viro static struct dentry *filename_create(int dfd, struct filename *name, 34051ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 34061da177e4SLinus Torvalds { 3407c663e5d8SChristoph Hellwig struct dentry *dentry = ERR_PTR(-EEXIST); 3408391172c4SAl Viro struct qstr last; 3409391172c4SAl Viro int type; 3410c30dabfeSJan Kara int err2; 34111ac12b4bSJeff Layton int error; 34121ac12b4bSJeff Layton bool is_dir = (lookup_flags & LOOKUP_DIRECTORY); 34131ac12b4bSJeff Layton 34141ac12b4bSJeff Layton /* 34151ac12b4bSJeff Layton * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any 34161ac12b4bSJeff Layton * other flags passed in are ignored! 34171ac12b4bSJeff Layton */ 34181ac12b4bSJeff Layton lookup_flags &= LOOKUP_REVAL; 34191ac12b4bSJeff Layton 34205c31b6ceSAl Viro name = filename_parentat(dfd, name, lookup_flags, path, &last, &type); 34215c31b6ceSAl Viro if (IS_ERR(name)) 34225c31b6ceSAl Viro return ERR_CAST(name); 34231da177e4SLinus Torvalds 3424c663e5d8SChristoph Hellwig /* 3425c663e5d8SChristoph Hellwig * Yucky last component or no last component at all? 3426c663e5d8SChristoph Hellwig * (foo/., foo/.., /////) 3427c663e5d8SChristoph Hellwig */ 34285c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) 3429ed75e95dSAl Viro goto out; 3430c663e5d8SChristoph Hellwig 3431c30dabfeSJan Kara /* don't fail immediately if it's r/o, at least try to report other errors */ 3432391172c4SAl Viro err2 = mnt_want_write(path->mnt); 3433c663e5d8SChristoph Hellwig /* 3434c663e5d8SChristoph Hellwig * Do the final lookup. 3435c663e5d8SChristoph Hellwig */ 3436391172c4SAl Viro lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL; 3437391172c4SAl Viro mutex_lock_nested(&path->dentry->d_inode->i_mutex, I_MUTEX_PARENT); 3438391172c4SAl Viro dentry = __lookup_hash(&last, path->dentry, lookup_flags); 34391da177e4SLinus Torvalds if (IS_ERR(dentry)) 3440a8104a9fSAl Viro goto unlock; 3441c663e5d8SChristoph Hellwig 3442a8104a9fSAl Viro error = -EEXIST; 3443b18825a7SDavid Howells if (d_is_positive(dentry)) 3444a8104a9fSAl Viro goto fail; 3445b18825a7SDavid Howells 3446c663e5d8SChristoph Hellwig /* 3447c663e5d8SChristoph Hellwig * Special case - lookup gave negative, but... we had foo/bar/ 3448c663e5d8SChristoph Hellwig * From the vfs_mknod() POV we just have a negative dentry - 3449c663e5d8SChristoph Hellwig * all is fine. Let's be bastards - you had / on the end, you've 3450c663e5d8SChristoph Hellwig * been asking for (non-existent) directory. -ENOENT for you. 3451c663e5d8SChristoph Hellwig */ 3452391172c4SAl Viro if (unlikely(!is_dir && last.name[last.len])) { 3453a8104a9fSAl Viro error = -ENOENT; 3454ed75e95dSAl Viro goto fail; 3455e9baf6e5SAl Viro } 3456c30dabfeSJan Kara if (unlikely(err2)) { 3457c30dabfeSJan Kara error = err2; 3458a8104a9fSAl Viro goto fail; 3459c30dabfeSJan Kara } 3460181c37b6SAl Viro putname(name); 3461e9baf6e5SAl Viro return dentry; 34621da177e4SLinus Torvalds fail: 3463a8104a9fSAl Viro dput(dentry); 3464a8104a9fSAl Viro dentry = ERR_PTR(error); 3465a8104a9fSAl Viro unlock: 3466391172c4SAl Viro mutex_unlock(&path->dentry->d_inode->i_mutex); 3467c30dabfeSJan Kara if (!err2) 3468391172c4SAl Viro mnt_drop_write(path->mnt); 3469ed75e95dSAl Viro out: 3470391172c4SAl Viro path_put(path); 3471181c37b6SAl Viro putname(name); 3472ed75e95dSAl Viro return dentry; 3473dae6ad8fSAl Viro } 3474fa14a0b8SAl Viro 3475fa14a0b8SAl Viro struct dentry *kern_path_create(int dfd, const char *pathname, 3476fa14a0b8SAl Viro struct path *path, unsigned int lookup_flags) 3477fa14a0b8SAl Viro { 3478181c37b6SAl Viro return filename_create(dfd, getname_kernel(pathname), 3479181c37b6SAl Viro path, lookup_flags); 3480fa14a0b8SAl Viro } 3481dae6ad8fSAl Viro EXPORT_SYMBOL(kern_path_create); 3482dae6ad8fSAl Viro 3483921a1650SAl Viro void done_path_create(struct path *path, struct dentry *dentry) 3484921a1650SAl Viro { 3485921a1650SAl Viro dput(dentry); 3486921a1650SAl Viro mutex_unlock(&path->dentry->d_inode->i_mutex); 3487a8104a9fSAl Viro mnt_drop_write(path->mnt); 3488921a1650SAl Viro path_put(path); 3489921a1650SAl Viro } 3490921a1650SAl Viro EXPORT_SYMBOL(done_path_create); 3491921a1650SAl Viro 3492520ae687SAl Viro inline struct dentry *user_path_create(int dfd, const char __user *pathname, 34931ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 3494dae6ad8fSAl Viro { 3495181c37b6SAl Viro return filename_create(dfd, getname(pathname), path, lookup_flags); 3496dae6ad8fSAl Viro } 3497dae6ad8fSAl Viro EXPORT_SYMBOL(user_path_create); 3498dae6ad8fSAl Viro 34991a67aafbSAl Viro int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) 35001da177e4SLinus Torvalds { 3501a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 35021da177e4SLinus Torvalds 35031da177e4SLinus Torvalds if (error) 35041da177e4SLinus Torvalds return error; 35051da177e4SLinus Torvalds 3506975d6b39SEric W. Biederman if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) 35071da177e4SLinus Torvalds return -EPERM; 35081da177e4SLinus Torvalds 3509acfa4380SAl Viro if (!dir->i_op->mknod) 35101da177e4SLinus Torvalds return -EPERM; 35111da177e4SLinus Torvalds 351208ce5f16SSerge E. Hallyn error = devcgroup_inode_mknod(mode, dev); 351308ce5f16SSerge E. Hallyn if (error) 351408ce5f16SSerge E. Hallyn return error; 351508ce5f16SSerge E. Hallyn 35161da177e4SLinus Torvalds error = security_inode_mknod(dir, dentry, mode, dev); 35171da177e4SLinus Torvalds if (error) 35181da177e4SLinus Torvalds return error; 35191da177e4SLinus Torvalds 35201da177e4SLinus Torvalds error = dir->i_op->mknod(dir, dentry, mode, dev); 3521a74574aaSStephen Smalley if (!error) 3522f38aa942SAmy Griffis fsnotify_create(dir, dentry); 35231da177e4SLinus Torvalds return error; 35241da177e4SLinus Torvalds } 35254d359507SAl Viro EXPORT_SYMBOL(vfs_mknod); 35261da177e4SLinus Torvalds 3527f69aac00SAl Viro static int may_mknod(umode_t mode) 3528463c3197SDave Hansen { 3529463c3197SDave Hansen switch (mode & S_IFMT) { 3530463c3197SDave Hansen case S_IFREG: 3531463c3197SDave Hansen case S_IFCHR: 3532463c3197SDave Hansen case S_IFBLK: 3533463c3197SDave Hansen case S_IFIFO: 3534463c3197SDave Hansen case S_IFSOCK: 3535463c3197SDave Hansen case 0: /* zero mode translates to S_IFREG */ 3536463c3197SDave Hansen return 0; 3537463c3197SDave Hansen case S_IFDIR: 3538463c3197SDave Hansen return -EPERM; 3539463c3197SDave Hansen default: 3540463c3197SDave Hansen return -EINVAL; 3541463c3197SDave Hansen } 3542463c3197SDave Hansen } 3543463c3197SDave Hansen 35448208a22bSAl Viro SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, 35452e4d0924SHeiko Carstens unsigned, dev) 35461da177e4SLinus Torvalds { 35471da177e4SLinus Torvalds struct dentry *dentry; 3548dae6ad8fSAl Viro struct path path; 3549dae6ad8fSAl Viro int error; 3550972567f1SJeff Layton unsigned int lookup_flags = 0; 35511da177e4SLinus Torvalds 35528e4bfca1SAl Viro error = may_mknod(mode); 35538e4bfca1SAl Viro if (error) 35548e4bfca1SAl Viro return error; 3555972567f1SJeff Layton retry: 3556972567f1SJeff Layton dentry = user_path_create(dfd, filename, &path, lookup_flags); 3557dae6ad8fSAl Viro if (IS_ERR(dentry)) 3558dae6ad8fSAl Viro return PTR_ERR(dentry); 35592ad94ae6SAl Viro 3560dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3561ce3b0f8dSAl Viro mode &= ~current_umask(); 3562dae6ad8fSAl Viro error = security_path_mknod(&path, dentry, mode, dev); 3563be6d3e56SKentaro Takeda if (error) 3564a8104a9fSAl Viro goto out; 35651da177e4SLinus Torvalds switch (mode & S_IFMT) { 35661da177e4SLinus Torvalds case 0: case S_IFREG: 3567312b63fbSAl Viro error = vfs_create(path.dentry->d_inode,dentry,mode,true); 35681da177e4SLinus Torvalds break; 35691da177e4SLinus Torvalds case S_IFCHR: case S_IFBLK: 3570dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode, 35711da177e4SLinus Torvalds new_decode_dev(dev)); 35721da177e4SLinus Torvalds break; 35731da177e4SLinus Torvalds case S_IFIFO: case S_IFSOCK: 3574dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); 35751da177e4SLinus Torvalds break; 35761da177e4SLinus Torvalds } 3577a8104a9fSAl Viro out: 3578921a1650SAl Viro done_path_create(&path, dentry); 3579972567f1SJeff Layton if (retry_estale(error, lookup_flags)) { 3580972567f1SJeff Layton lookup_flags |= LOOKUP_REVAL; 3581972567f1SJeff Layton goto retry; 3582972567f1SJeff Layton } 35831da177e4SLinus Torvalds return error; 35841da177e4SLinus Torvalds } 35851da177e4SLinus Torvalds 35868208a22bSAl Viro SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) 35875590ff0dSUlrich Drepper { 35885590ff0dSUlrich Drepper return sys_mknodat(AT_FDCWD, filename, mode, dev); 35895590ff0dSUlrich Drepper } 35905590ff0dSUlrich Drepper 359118bb1db3SAl Viro int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 35921da177e4SLinus Torvalds { 3593a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 35948de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 35951da177e4SLinus Torvalds 35961da177e4SLinus Torvalds if (error) 35971da177e4SLinus Torvalds return error; 35981da177e4SLinus Torvalds 3599acfa4380SAl Viro if (!dir->i_op->mkdir) 36001da177e4SLinus Torvalds return -EPERM; 36011da177e4SLinus Torvalds 36021da177e4SLinus Torvalds mode &= (S_IRWXUGO|S_ISVTX); 36031da177e4SLinus Torvalds error = security_inode_mkdir(dir, dentry, mode); 36041da177e4SLinus Torvalds if (error) 36051da177e4SLinus Torvalds return error; 36061da177e4SLinus Torvalds 36078de52778SAl Viro if (max_links && dir->i_nlink >= max_links) 36088de52778SAl Viro return -EMLINK; 36098de52778SAl Viro 36101da177e4SLinus Torvalds error = dir->i_op->mkdir(dir, dentry, mode); 3611a74574aaSStephen Smalley if (!error) 3612f38aa942SAmy Griffis fsnotify_mkdir(dir, dentry); 36131da177e4SLinus Torvalds return error; 36141da177e4SLinus Torvalds } 36154d359507SAl Viro EXPORT_SYMBOL(vfs_mkdir); 36161da177e4SLinus Torvalds 3617a218d0fdSAl Viro SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) 36181da177e4SLinus Torvalds { 36196902d925SDave Hansen struct dentry *dentry; 3620dae6ad8fSAl Viro struct path path; 3621dae6ad8fSAl Viro int error; 3622b76d8b82SJeff Layton unsigned int lookup_flags = LOOKUP_DIRECTORY; 36231da177e4SLinus Torvalds 3624b76d8b82SJeff Layton retry: 3625b76d8b82SJeff Layton dentry = user_path_create(dfd, pathname, &path, lookup_flags); 36266902d925SDave Hansen if (IS_ERR(dentry)) 3627dae6ad8fSAl Viro return PTR_ERR(dentry); 36286902d925SDave Hansen 3629dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3630ce3b0f8dSAl Viro mode &= ~current_umask(); 3631dae6ad8fSAl Viro error = security_path_mkdir(&path, dentry, mode); 3632a8104a9fSAl Viro if (!error) 3633dae6ad8fSAl Viro error = vfs_mkdir(path.dentry->d_inode, dentry, mode); 3634921a1650SAl Viro done_path_create(&path, dentry); 3635b76d8b82SJeff Layton if (retry_estale(error, lookup_flags)) { 3636b76d8b82SJeff Layton lookup_flags |= LOOKUP_REVAL; 3637b76d8b82SJeff Layton goto retry; 3638b76d8b82SJeff Layton } 36391da177e4SLinus Torvalds return error; 36401da177e4SLinus Torvalds } 36411da177e4SLinus Torvalds 3642a218d0fdSAl Viro SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) 36435590ff0dSUlrich Drepper { 36445590ff0dSUlrich Drepper return sys_mkdirat(AT_FDCWD, pathname, mode); 36455590ff0dSUlrich Drepper } 36465590ff0dSUlrich Drepper 36471da177e4SLinus Torvalds /* 3648a71905f0SSage Weil * The dentry_unhash() helper will try to drop the dentry early: we 3649c0d02594SJ. Bruce Fields * should have a usage count of 1 if we're the only user of this 3650a71905f0SSage Weil * dentry, and if that is true (possibly after pruning the dcache), 3651a71905f0SSage Weil * then we drop the dentry now. 36521da177e4SLinus Torvalds * 36531da177e4SLinus Torvalds * A low-level filesystem can, if it choses, legally 36541da177e4SLinus Torvalds * do a 36551da177e4SLinus Torvalds * 36561da177e4SLinus Torvalds * if (!d_unhashed(dentry)) 36571da177e4SLinus Torvalds * return -EBUSY; 36581da177e4SLinus Torvalds * 36591da177e4SLinus Torvalds * if it cannot handle the case of removing a directory 36601da177e4SLinus Torvalds * that is still in use by something else.. 36611da177e4SLinus Torvalds */ 36621da177e4SLinus Torvalds void dentry_unhash(struct dentry *dentry) 36631da177e4SLinus Torvalds { 36641da177e4SLinus Torvalds shrink_dcache_parent(dentry); 36651da177e4SLinus Torvalds spin_lock(&dentry->d_lock); 366698474236SWaiman Long if (dentry->d_lockref.count == 1) 36671da177e4SLinus Torvalds __d_drop(dentry); 36681da177e4SLinus Torvalds spin_unlock(&dentry->d_lock); 36691da177e4SLinus Torvalds } 36704d359507SAl Viro EXPORT_SYMBOL(dentry_unhash); 36711da177e4SLinus Torvalds 36721da177e4SLinus Torvalds int vfs_rmdir(struct inode *dir, struct dentry *dentry) 36731da177e4SLinus Torvalds { 36741da177e4SLinus Torvalds int error = may_delete(dir, dentry, 1); 36751da177e4SLinus Torvalds 36761da177e4SLinus Torvalds if (error) 36771da177e4SLinus Torvalds return error; 36781da177e4SLinus Torvalds 3679acfa4380SAl Viro if (!dir->i_op->rmdir) 36801da177e4SLinus Torvalds return -EPERM; 36811da177e4SLinus Torvalds 36821d2ef590SAl Viro dget(dentry); 36831b1dcc1bSJes Sorensen mutex_lock(&dentry->d_inode->i_mutex); 3684912dbc15SSage Weil 36851da177e4SLinus Torvalds error = -EBUSY; 36867af1364fSEric W. Biederman if (is_local_mountpoint(dentry)) 3687912dbc15SSage Weil goto out; 3688912dbc15SSage Weil 36891da177e4SLinus Torvalds error = security_inode_rmdir(dir, dentry); 3690912dbc15SSage Weil if (error) 3691912dbc15SSage Weil goto out; 3692912dbc15SSage Weil 36933cebde24SSage Weil shrink_dcache_parent(dentry); 36941da177e4SLinus Torvalds error = dir->i_op->rmdir(dir, dentry); 3695912dbc15SSage Weil if (error) 3696912dbc15SSage Weil goto out; 3697912dbc15SSage Weil 36981da177e4SLinus Torvalds dentry->d_inode->i_flags |= S_DEAD; 3699d83c49f3SAl Viro dont_mount(dentry); 37008ed936b5SEric W. Biederman detach_mounts(dentry); 37011da177e4SLinus Torvalds 3702912dbc15SSage Weil out: 3703912dbc15SSage Weil mutex_unlock(&dentry->d_inode->i_mutex); 37041d2ef590SAl Viro dput(dentry); 3705912dbc15SSage Weil if (!error) 3706912dbc15SSage Weil d_delete(dentry); 37071da177e4SLinus Torvalds return error; 37081da177e4SLinus Torvalds } 37094d359507SAl Viro EXPORT_SYMBOL(vfs_rmdir); 37101da177e4SLinus Torvalds 37115590ff0dSUlrich Drepper static long do_rmdir(int dfd, const char __user *pathname) 37121da177e4SLinus Torvalds { 37131da177e4SLinus Torvalds int error = 0; 371491a27b2aSJeff Layton struct filename *name; 37151da177e4SLinus Torvalds struct dentry *dentry; 3716f5beed75SAl Viro struct path path; 3717f5beed75SAl Viro struct qstr last; 3718f5beed75SAl Viro int type; 3719c6ee9206SJeff Layton unsigned int lookup_flags = 0; 3720c6ee9206SJeff Layton retry: 3721f5beed75SAl Viro name = user_path_parent(dfd, pathname, 3722f5beed75SAl Viro &path, &last, &type, lookup_flags); 372391a27b2aSJeff Layton if (IS_ERR(name)) 372491a27b2aSJeff Layton return PTR_ERR(name); 37251da177e4SLinus Torvalds 3726f5beed75SAl Viro switch (type) { 37271da177e4SLinus Torvalds case LAST_DOTDOT: 37281da177e4SLinus Torvalds error = -ENOTEMPTY; 37291da177e4SLinus Torvalds goto exit1; 37301da177e4SLinus Torvalds case LAST_DOT: 37311da177e4SLinus Torvalds error = -EINVAL; 37321da177e4SLinus Torvalds goto exit1; 37331da177e4SLinus Torvalds case LAST_ROOT: 37341da177e4SLinus Torvalds error = -EBUSY; 37351da177e4SLinus Torvalds goto exit1; 37361da177e4SLinus Torvalds } 37370612d9fbSOGAWA Hirofumi 3738f5beed75SAl Viro error = mnt_want_write(path.mnt); 3739c30dabfeSJan Kara if (error) 3740c30dabfeSJan Kara goto exit1; 37410612d9fbSOGAWA Hirofumi 3742f5beed75SAl Viro mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 3743f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 37441da177e4SLinus Torvalds error = PTR_ERR(dentry); 37456902d925SDave Hansen if (IS_ERR(dentry)) 37466902d925SDave Hansen goto exit2; 3747e6bc45d6STheodore Ts'o if (!dentry->d_inode) { 3748e6bc45d6STheodore Ts'o error = -ENOENT; 3749e6bc45d6STheodore Ts'o goto exit3; 3750e6bc45d6STheodore Ts'o } 3751f5beed75SAl Viro error = security_path_rmdir(&path, dentry); 3752be6d3e56SKentaro Takeda if (error) 3753c30dabfeSJan Kara goto exit3; 3754f5beed75SAl Viro error = vfs_rmdir(path.dentry->d_inode, dentry); 37550622753bSDave Hansen exit3: 37561da177e4SLinus Torvalds dput(dentry); 37576902d925SDave Hansen exit2: 3758f5beed75SAl Viro mutex_unlock(&path.dentry->d_inode->i_mutex); 3759f5beed75SAl Viro mnt_drop_write(path.mnt); 37601da177e4SLinus Torvalds exit1: 3761f5beed75SAl Viro path_put(&path); 37621da177e4SLinus Torvalds putname(name); 3763c6ee9206SJeff Layton if (retry_estale(error, lookup_flags)) { 3764c6ee9206SJeff Layton lookup_flags |= LOOKUP_REVAL; 3765c6ee9206SJeff Layton goto retry; 3766c6ee9206SJeff Layton } 37671da177e4SLinus Torvalds return error; 37681da177e4SLinus Torvalds } 37691da177e4SLinus Torvalds 37703cdad428SHeiko Carstens SYSCALL_DEFINE1(rmdir, const char __user *, pathname) 37715590ff0dSUlrich Drepper { 37725590ff0dSUlrich Drepper return do_rmdir(AT_FDCWD, pathname); 37735590ff0dSUlrich Drepper } 37745590ff0dSUlrich Drepper 3775b21996e3SJ. Bruce Fields /** 3776b21996e3SJ. Bruce Fields * vfs_unlink - unlink a filesystem object 3777b21996e3SJ. Bruce Fields * @dir: parent directory 3778b21996e3SJ. Bruce Fields * @dentry: victim 3779b21996e3SJ. Bruce Fields * @delegated_inode: returns victim inode, if the inode is delegated. 3780b21996e3SJ. Bruce Fields * 3781b21996e3SJ. Bruce Fields * The caller must hold dir->i_mutex. 3782b21996e3SJ. Bruce Fields * 3783b21996e3SJ. Bruce Fields * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and 3784b21996e3SJ. Bruce Fields * return a reference to the inode in delegated_inode. The caller 3785b21996e3SJ. Bruce Fields * should then break the delegation on that inode and retry. Because 3786b21996e3SJ. Bruce Fields * breaking a delegation may take a long time, the caller should drop 3787b21996e3SJ. Bruce Fields * dir->i_mutex before doing so. 3788b21996e3SJ. Bruce Fields * 3789b21996e3SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 3790b21996e3SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 3791b21996e3SJ. Bruce Fields * to be NFS exported. 3792b21996e3SJ. Bruce Fields */ 3793b21996e3SJ. Bruce Fields int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode) 37941da177e4SLinus Torvalds { 37959accbb97SJ. Bruce Fields struct inode *target = dentry->d_inode; 37961da177e4SLinus Torvalds int error = may_delete(dir, dentry, 0); 37971da177e4SLinus Torvalds 37981da177e4SLinus Torvalds if (error) 37991da177e4SLinus Torvalds return error; 38001da177e4SLinus Torvalds 3801acfa4380SAl Viro if (!dir->i_op->unlink) 38021da177e4SLinus Torvalds return -EPERM; 38031da177e4SLinus Torvalds 38049accbb97SJ. Bruce Fields mutex_lock(&target->i_mutex); 38058ed936b5SEric W. Biederman if (is_local_mountpoint(dentry)) 38061da177e4SLinus Torvalds error = -EBUSY; 38071da177e4SLinus Torvalds else { 38081da177e4SLinus Torvalds error = security_inode_unlink(dir, dentry); 3809bec1052eSAl Viro if (!error) { 38105a14696cSJ. Bruce Fields error = try_break_deleg(target, delegated_inode); 38115a14696cSJ. Bruce Fields if (error) 3812b21996e3SJ. Bruce Fields goto out; 38131da177e4SLinus Torvalds error = dir->i_op->unlink(dir, dentry); 38148ed936b5SEric W. Biederman if (!error) { 3815d83c49f3SAl Viro dont_mount(dentry); 38168ed936b5SEric W. Biederman detach_mounts(dentry); 38178ed936b5SEric W. Biederman } 3818bec1052eSAl Viro } 38191da177e4SLinus Torvalds } 3820b21996e3SJ. Bruce Fields out: 38219accbb97SJ. Bruce Fields mutex_unlock(&target->i_mutex); 38221da177e4SLinus Torvalds 38231da177e4SLinus Torvalds /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 38241da177e4SLinus Torvalds if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { 38259accbb97SJ. Bruce Fields fsnotify_link_count(target); 38261da177e4SLinus Torvalds d_delete(dentry); 38271da177e4SLinus Torvalds } 38280eeca283SRobert Love 38291da177e4SLinus Torvalds return error; 38301da177e4SLinus Torvalds } 38314d359507SAl Viro EXPORT_SYMBOL(vfs_unlink); 38321da177e4SLinus Torvalds 38331da177e4SLinus Torvalds /* 38341da177e4SLinus Torvalds * Make sure that the actual truncation of the file will occur outside its 38351b1dcc1bSJes Sorensen * directory's i_mutex. Truncate can take a long time if there is a lot of 38361da177e4SLinus Torvalds * writeout happening, and we don't want to prevent access to the directory 38371da177e4SLinus Torvalds * while waiting on the I/O. 38381da177e4SLinus Torvalds */ 38395590ff0dSUlrich Drepper static long do_unlinkat(int dfd, const char __user *pathname) 38401da177e4SLinus Torvalds { 38412ad94ae6SAl Viro int error; 384291a27b2aSJeff Layton struct filename *name; 38431da177e4SLinus Torvalds struct dentry *dentry; 3844f5beed75SAl Viro struct path path; 3845f5beed75SAl Viro struct qstr last; 3846f5beed75SAl Viro int type; 38471da177e4SLinus Torvalds struct inode *inode = NULL; 3848b21996e3SJ. Bruce Fields struct inode *delegated_inode = NULL; 38495d18f813SJeff Layton unsigned int lookup_flags = 0; 38505d18f813SJeff Layton retry: 3851f5beed75SAl Viro name = user_path_parent(dfd, pathname, 3852f5beed75SAl Viro &path, &last, &type, lookup_flags); 385391a27b2aSJeff Layton if (IS_ERR(name)) 385491a27b2aSJeff Layton return PTR_ERR(name); 38552ad94ae6SAl Viro 38561da177e4SLinus Torvalds error = -EISDIR; 3857f5beed75SAl Viro if (type != LAST_NORM) 38581da177e4SLinus Torvalds goto exit1; 38590612d9fbSOGAWA Hirofumi 3860f5beed75SAl Viro error = mnt_want_write(path.mnt); 3861c30dabfeSJan Kara if (error) 3862c30dabfeSJan Kara goto exit1; 3863b21996e3SJ. Bruce Fields retry_deleg: 3864f5beed75SAl Viro mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 3865f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 38661da177e4SLinus Torvalds error = PTR_ERR(dentry); 38671da177e4SLinus Torvalds if (!IS_ERR(dentry)) { 38681da177e4SLinus Torvalds /* Why not before? Because we want correct error value */ 3869f5beed75SAl Viro if (last.name[last.len]) 387050338b88STörök Edwin goto slashes; 38711da177e4SLinus Torvalds inode = dentry->d_inode; 3872b18825a7SDavid Howells if (d_is_negative(dentry)) 3873e6bc45d6STheodore Ts'o goto slashes; 38747de9c6eeSAl Viro ihold(inode); 3875f5beed75SAl Viro error = security_path_unlink(&path, dentry); 3876be6d3e56SKentaro Takeda if (error) 3877c30dabfeSJan Kara goto exit2; 3878f5beed75SAl Viro error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode); 38791da177e4SLinus Torvalds exit2: 38801da177e4SLinus Torvalds dput(dentry); 38811da177e4SLinus Torvalds } 3882f5beed75SAl Viro mutex_unlock(&path.dentry->d_inode->i_mutex); 38831da177e4SLinus Torvalds if (inode) 38841da177e4SLinus Torvalds iput(inode); /* truncate the inode here */ 3885b21996e3SJ. Bruce Fields inode = NULL; 3886b21996e3SJ. Bruce Fields if (delegated_inode) { 38875a14696cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 3888b21996e3SJ. Bruce Fields if (!error) 3889b21996e3SJ. Bruce Fields goto retry_deleg; 3890b21996e3SJ. Bruce Fields } 3891f5beed75SAl Viro mnt_drop_write(path.mnt); 38921da177e4SLinus Torvalds exit1: 3893f5beed75SAl Viro path_put(&path); 38941da177e4SLinus Torvalds putname(name); 38955d18f813SJeff Layton if (retry_estale(error, lookup_flags)) { 38965d18f813SJeff Layton lookup_flags |= LOOKUP_REVAL; 38975d18f813SJeff Layton inode = NULL; 38985d18f813SJeff Layton goto retry; 38995d18f813SJeff Layton } 39001da177e4SLinus Torvalds return error; 39011da177e4SLinus Torvalds 39021da177e4SLinus Torvalds slashes: 3903b18825a7SDavid Howells if (d_is_negative(dentry)) 3904b18825a7SDavid Howells error = -ENOENT; 390544b1d530SMiklos Szeredi else if (d_is_dir(dentry)) 3906b18825a7SDavid Howells error = -EISDIR; 3907b18825a7SDavid Howells else 3908b18825a7SDavid Howells error = -ENOTDIR; 39091da177e4SLinus Torvalds goto exit2; 39101da177e4SLinus Torvalds } 39111da177e4SLinus Torvalds 39122e4d0924SHeiko Carstens SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) 39135590ff0dSUlrich Drepper { 39145590ff0dSUlrich Drepper if ((flag & ~AT_REMOVEDIR) != 0) 39155590ff0dSUlrich Drepper return -EINVAL; 39165590ff0dSUlrich Drepper 39175590ff0dSUlrich Drepper if (flag & AT_REMOVEDIR) 39185590ff0dSUlrich Drepper return do_rmdir(dfd, pathname); 39195590ff0dSUlrich Drepper 39205590ff0dSUlrich Drepper return do_unlinkat(dfd, pathname); 39215590ff0dSUlrich Drepper } 39225590ff0dSUlrich Drepper 39233480b257SHeiko Carstens SYSCALL_DEFINE1(unlink, const char __user *, pathname) 39245590ff0dSUlrich Drepper { 39255590ff0dSUlrich Drepper return do_unlinkat(AT_FDCWD, pathname); 39265590ff0dSUlrich Drepper } 39275590ff0dSUlrich Drepper 3928db2e747bSMiklos Szeredi int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) 39291da177e4SLinus Torvalds { 3930a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 39311da177e4SLinus Torvalds 39321da177e4SLinus Torvalds if (error) 39331da177e4SLinus Torvalds return error; 39341da177e4SLinus Torvalds 3935acfa4380SAl Viro if (!dir->i_op->symlink) 39361da177e4SLinus Torvalds return -EPERM; 39371da177e4SLinus Torvalds 39381da177e4SLinus Torvalds error = security_inode_symlink(dir, dentry, oldname); 39391da177e4SLinus Torvalds if (error) 39401da177e4SLinus Torvalds return error; 39411da177e4SLinus Torvalds 39421da177e4SLinus Torvalds error = dir->i_op->symlink(dir, dentry, oldname); 3943a74574aaSStephen Smalley if (!error) 3944f38aa942SAmy Griffis fsnotify_create(dir, dentry); 39451da177e4SLinus Torvalds return error; 39461da177e4SLinus Torvalds } 39474d359507SAl Viro EXPORT_SYMBOL(vfs_symlink); 39481da177e4SLinus Torvalds 39492e4d0924SHeiko Carstens SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, 39502e4d0924SHeiko Carstens int, newdfd, const char __user *, newname) 39511da177e4SLinus Torvalds { 39522ad94ae6SAl Viro int error; 395391a27b2aSJeff Layton struct filename *from; 39546902d925SDave Hansen struct dentry *dentry; 3955dae6ad8fSAl Viro struct path path; 3956f46d3567SJeff Layton unsigned int lookup_flags = 0; 39571da177e4SLinus Torvalds 39581da177e4SLinus Torvalds from = getname(oldname); 39591da177e4SLinus Torvalds if (IS_ERR(from)) 39601da177e4SLinus Torvalds return PTR_ERR(from); 3961f46d3567SJeff Layton retry: 3962f46d3567SJeff Layton dentry = user_path_create(newdfd, newname, &path, lookup_flags); 39631da177e4SLinus Torvalds error = PTR_ERR(dentry); 39646902d925SDave Hansen if (IS_ERR(dentry)) 3965dae6ad8fSAl Viro goto out_putname; 39666902d925SDave Hansen 396791a27b2aSJeff Layton error = security_path_symlink(&path, dentry, from->name); 3968a8104a9fSAl Viro if (!error) 396991a27b2aSJeff Layton error = vfs_symlink(path.dentry->d_inode, dentry, from->name); 3970921a1650SAl Viro done_path_create(&path, dentry); 3971f46d3567SJeff Layton if (retry_estale(error, lookup_flags)) { 3972f46d3567SJeff Layton lookup_flags |= LOOKUP_REVAL; 3973f46d3567SJeff Layton goto retry; 3974f46d3567SJeff Layton } 39756902d925SDave Hansen out_putname: 39761da177e4SLinus Torvalds putname(from); 39771da177e4SLinus Torvalds return error; 39781da177e4SLinus Torvalds } 39791da177e4SLinus Torvalds 39803480b257SHeiko Carstens SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) 39815590ff0dSUlrich Drepper { 39825590ff0dSUlrich Drepper return sys_symlinkat(oldname, AT_FDCWD, newname); 39835590ff0dSUlrich Drepper } 39845590ff0dSUlrich Drepper 3985146a8595SJ. Bruce Fields /** 3986146a8595SJ. Bruce Fields * vfs_link - create a new link 3987146a8595SJ. Bruce Fields * @old_dentry: object to be linked 3988146a8595SJ. Bruce Fields * @dir: new parent 3989146a8595SJ. Bruce Fields * @new_dentry: where to create the new link 3990146a8595SJ. Bruce Fields * @delegated_inode: returns inode needing a delegation break 3991146a8595SJ. Bruce Fields * 3992146a8595SJ. Bruce Fields * The caller must hold dir->i_mutex 3993146a8595SJ. Bruce Fields * 3994146a8595SJ. Bruce Fields * If vfs_link discovers a delegation on the to-be-linked file in need 3995146a8595SJ. Bruce Fields * of breaking, it will return -EWOULDBLOCK and return a reference to the 3996146a8595SJ. Bruce Fields * inode in delegated_inode. The caller should then break the delegation 3997146a8595SJ. Bruce Fields * and retry. Because breaking a delegation may take a long time, the 3998146a8595SJ. Bruce Fields * caller should drop the i_mutex before doing so. 3999146a8595SJ. Bruce Fields * 4000146a8595SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 4001146a8595SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 4002146a8595SJ. Bruce Fields * to be NFS exported. 4003146a8595SJ. Bruce Fields */ 4004146a8595SJ. Bruce Fields int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode) 40051da177e4SLinus Torvalds { 40061da177e4SLinus Torvalds struct inode *inode = old_dentry->d_inode; 40078de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 40081da177e4SLinus Torvalds int error; 40091da177e4SLinus Torvalds 40101da177e4SLinus Torvalds if (!inode) 40111da177e4SLinus Torvalds return -ENOENT; 40121da177e4SLinus Torvalds 4013a95164d9SMiklos Szeredi error = may_create(dir, new_dentry); 40141da177e4SLinus Torvalds if (error) 40151da177e4SLinus Torvalds return error; 40161da177e4SLinus Torvalds 40171da177e4SLinus Torvalds if (dir->i_sb != inode->i_sb) 40181da177e4SLinus Torvalds return -EXDEV; 40191da177e4SLinus Torvalds 40201da177e4SLinus Torvalds /* 40211da177e4SLinus Torvalds * A link to an append-only or immutable file cannot be created. 40221da177e4SLinus Torvalds */ 40231da177e4SLinus Torvalds if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 40241da177e4SLinus Torvalds return -EPERM; 4025acfa4380SAl Viro if (!dir->i_op->link) 40261da177e4SLinus Torvalds return -EPERM; 40277e79eedbSTetsuo Handa if (S_ISDIR(inode->i_mode)) 40281da177e4SLinus Torvalds return -EPERM; 40291da177e4SLinus Torvalds 40301da177e4SLinus Torvalds error = security_inode_link(old_dentry, dir, new_dentry); 40311da177e4SLinus Torvalds if (error) 40321da177e4SLinus Torvalds return error; 40331da177e4SLinus Torvalds 40347e79eedbSTetsuo Handa mutex_lock(&inode->i_mutex); 4035aae8a97dSAneesh Kumar K.V /* Make sure we don't allow creating hardlink to an unlinked file */ 4036f4e0c30cSAl Viro if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) 4037aae8a97dSAneesh Kumar K.V error = -ENOENT; 40388de52778SAl Viro else if (max_links && inode->i_nlink >= max_links) 40398de52778SAl Viro error = -EMLINK; 4040146a8595SJ. Bruce Fields else { 4041146a8595SJ. Bruce Fields error = try_break_deleg(inode, delegated_inode); 4042146a8595SJ. Bruce Fields if (!error) 40431da177e4SLinus Torvalds error = dir->i_op->link(old_dentry, dir, new_dentry); 4044146a8595SJ. Bruce Fields } 4045f4e0c30cSAl Viro 4046f4e0c30cSAl Viro if (!error && (inode->i_state & I_LINKABLE)) { 4047f4e0c30cSAl Viro spin_lock(&inode->i_lock); 4048f4e0c30cSAl Viro inode->i_state &= ~I_LINKABLE; 4049f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 4050f4e0c30cSAl Viro } 40517e79eedbSTetsuo Handa mutex_unlock(&inode->i_mutex); 4052e31e14ecSStephen Smalley if (!error) 40537e79eedbSTetsuo Handa fsnotify_link(dir, inode, new_dentry); 40541da177e4SLinus Torvalds return error; 40551da177e4SLinus Torvalds } 40564d359507SAl Viro EXPORT_SYMBOL(vfs_link); 40571da177e4SLinus Torvalds 40581da177e4SLinus Torvalds /* 40591da177e4SLinus Torvalds * Hardlinks are often used in delicate situations. We avoid 40601da177e4SLinus Torvalds * security-related surprises by not following symlinks on the 40611da177e4SLinus Torvalds * newname. --KAB 40621da177e4SLinus Torvalds * 40631da177e4SLinus Torvalds * We don't follow them on the oldname either to be compatible 40641da177e4SLinus Torvalds * with linux 2.0, and to avoid hard-linking to directories 40651da177e4SLinus Torvalds * and other special files. --ADM 40661da177e4SLinus Torvalds */ 40672e4d0924SHeiko Carstens SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, 40682e4d0924SHeiko Carstens int, newdfd, const char __user *, newname, int, flags) 40691da177e4SLinus Torvalds { 40701da177e4SLinus Torvalds struct dentry *new_dentry; 4071dae6ad8fSAl Viro struct path old_path, new_path; 4072146a8595SJ. Bruce Fields struct inode *delegated_inode = NULL; 407311a7b371SAneesh Kumar K.V int how = 0; 40741da177e4SLinus Torvalds int error; 40751da177e4SLinus Torvalds 407611a7b371SAneesh Kumar K.V if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) 4077c04030e1SUlrich Drepper return -EINVAL; 407811a7b371SAneesh Kumar K.V /* 4079f0cc6ffbSLinus Torvalds * To use null names we require CAP_DAC_READ_SEARCH 4080f0cc6ffbSLinus Torvalds * This ensures that not everyone will be able to create 4081f0cc6ffbSLinus Torvalds * handlink using the passed filedescriptor. 408211a7b371SAneesh Kumar K.V */ 4083f0cc6ffbSLinus Torvalds if (flags & AT_EMPTY_PATH) { 4084f0cc6ffbSLinus Torvalds if (!capable(CAP_DAC_READ_SEARCH)) 4085f0cc6ffbSLinus Torvalds return -ENOENT; 408611a7b371SAneesh Kumar K.V how = LOOKUP_EMPTY; 4087f0cc6ffbSLinus Torvalds } 4088c04030e1SUlrich Drepper 408911a7b371SAneesh Kumar K.V if (flags & AT_SYMLINK_FOLLOW) 409011a7b371SAneesh Kumar K.V how |= LOOKUP_FOLLOW; 4091442e31caSJeff Layton retry: 409211a7b371SAneesh Kumar K.V error = user_path_at(olddfd, oldname, how, &old_path); 40931da177e4SLinus Torvalds if (error) 40942ad94ae6SAl Viro return error; 40952ad94ae6SAl Viro 4096442e31caSJeff Layton new_dentry = user_path_create(newdfd, newname, &new_path, 4097442e31caSJeff Layton (how & LOOKUP_REVAL)); 40981da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 40996902d925SDave Hansen if (IS_ERR(new_dentry)) 4100dae6ad8fSAl Viro goto out; 4101dae6ad8fSAl Viro 4102dae6ad8fSAl Viro error = -EXDEV; 4103dae6ad8fSAl Viro if (old_path.mnt != new_path.mnt) 4104dae6ad8fSAl Viro goto out_dput; 4105800179c9SKees Cook error = may_linkat(&old_path); 4106800179c9SKees Cook if (unlikely(error)) 4107800179c9SKees Cook goto out_dput; 4108dae6ad8fSAl Viro error = security_path_link(old_path.dentry, &new_path, new_dentry); 4109be6d3e56SKentaro Takeda if (error) 4110a8104a9fSAl Viro goto out_dput; 4111146a8595SJ. Bruce Fields error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode); 411275c3f29dSDave Hansen out_dput: 4113921a1650SAl Viro done_path_create(&new_path, new_dentry); 4114146a8595SJ. Bruce Fields if (delegated_inode) { 4115146a8595SJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4116d22e6338SOleg Drokin if (!error) { 4117d22e6338SOleg Drokin path_put(&old_path); 4118146a8595SJ. Bruce Fields goto retry; 4119146a8595SJ. Bruce Fields } 4120d22e6338SOleg Drokin } 4121442e31caSJeff Layton if (retry_estale(error, how)) { 4122d22e6338SOleg Drokin path_put(&old_path); 4123442e31caSJeff Layton how |= LOOKUP_REVAL; 4124442e31caSJeff Layton goto retry; 4125442e31caSJeff Layton } 41261da177e4SLinus Torvalds out: 41272d8f3038SAl Viro path_put(&old_path); 41281da177e4SLinus Torvalds 41291da177e4SLinus Torvalds return error; 41301da177e4SLinus Torvalds } 41311da177e4SLinus Torvalds 41323480b257SHeiko Carstens SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) 41335590ff0dSUlrich Drepper { 4134c04030e1SUlrich Drepper return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 41355590ff0dSUlrich Drepper } 41365590ff0dSUlrich Drepper 4137bc27027aSMiklos Szeredi /** 4138bc27027aSMiklos Szeredi * vfs_rename - rename a filesystem object 4139bc27027aSMiklos Szeredi * @old_dir: parent of source 4140bc27027aSMiklos Szeredi * @old_dentry: source 4141bc27027aSMiklos Szeredi * @new_dir: parent of destination 4142bc27027aSMiklos Szeredi * @new_dentry: destination 4143bc27027aSMiklos Szeredi * @delegated_inode: returns an inode needing a delegation break 4144520c8b16SMiklos Szeredi * @flags: rename flags 4145bc27027aSMiklos Szeredi * 4146bc27027aSMiklos Szeredi * The caller must hold multiple mutexes--see lock_rename()). 4147bc27027aSMiklos Szeredi * 4148bc27027aSMiklos Szeredi * If vfs_rename discovers a delegation in need of breaking at either 4149bc27027aSMiklos Szeredi * the source or destination, it will return -EWOULDBLOCK and return a 4150bc27027aSMiklos Szeredi * reference to the inode in delegated_inode. The caller should then 4151bc27027aSMiklos Szeredi * break the delegation and retry. Because breaking a delegation may 4152bc27027aSMiklos Szeredi * take a long time, the caller should drop all locks before doing 4153bc27027aSMiklos Szeredi * so. 4154bc27027aSMiklos Szeredi * 4155bc27027aSMiklos Szeredi * Alternatively, a caller may pass NULL for delegated_inode. This may 4156bc27027aSMiklos Szeredi * be appropriate for callers that expect the underlying filesystem not 4157bc27027aSMiklos Szeredi * to be NFS exported. 4158bc27027aSMiklos Szeredi * 41591da177e4SLinus Torvalds * The worst of all namespace operations - renaming directory. "Perverted" 41601da177e4SLinus Torvalds * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 41611da177e4SLinus Torvalds * Problems: 4162d03b29a2SJ. Bruce Fields * a) we can get into loop creation. 41631da177e4SLinus Torvalds * b) race potential - two innocent renames can create a loop together. 41641da177e4SLinus Torvalds * That's where 4.4 screws up. Current fix: serialization on 4165a11f3a05SArjan van de Ven * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 41661da177e4SLinus Torvalds * story. 41676cedba89SJ. Bruce Fields * c) we have to lock _four_ objects - parents and victim (if it exists), 41686cedba89SJ. Bruce Fields * and source (if it is not a directory). 41691b1dcc1bSJes Sorensen * And that - after we got ->i_mutex on parents (until then we don't know 41701da177e4SLinus Torvalds * whether the target exists). Solution: try to be smart with locking 41711da177e4SLinus Torvalds * order for inodes. We rely on the fact that tree topology may change 4172a11f3a05SArjan van de Ven * only under ->s_vfs_rename_mutex _and_ that parent of the object we 41731da177e4SLinus Torvalds * move will be locked. Thus we can rank directories by the tree 41741da177e4SLinus Torvalds * (ancestors first) and rank all non-directories after them. 41751da177e4SLinus Torvalds * That works since everybody except rename does "lock parent, lookup, 4176a11f3a05SArjan van de Ven * lock child" and rename is under ->s_vfs_rename_mutex. 41771da177e4SLinus Torvalds * HOWEVER, it relies on the assumption that any object with ->lookup() 41781da177e4SLinus Torvalds * has no more than 1 dentry. If "hybrid" objects will ever appear, 41791da177e4SLinus Torvalds * we'd better make sure that there's no link(2) for them. 4180e4eaac06SSage Weil * d) conversion from fhandle to dentry may come in the wrong moment - when 41811b1dcc1bSJes Sorensen * we are removing the target. Solution: we will have to grab ->i_mutex 41821da177e4SLinus Torvalds * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 4183c41b20e7SAdam Buchbinder * ->i_mutex on parents, which works but leads to some truly excessive 41841da177e4SLinus Torvalds * locking]. 41851da177e4SLinus Torvalds */ 41861da177e4SLinus Torvalds int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, 41878e6d782cSJ. Bruce Fields struct inode *new_dir, struct dentry *new_dentry, 4188520c8b16SMiklos Szeredi struct inode **delegated_inode, unsigned int flags) 41891da177e4SLinus Torvalds { 41901da177e4SLinus Torvalds int error; 4191bc27027aSMiklos Szeredi bool is_dir = d_is_dir(old_dentry); 419259b0df21SEric Paris const unsigned char *old_name; 4193bc27027aSMiklos Szeredi struct inode *source = old_dentry->d_inode; 4194bc27027aSMiklos Szeredi struct inode *target = new_dentry->d_inode; 4195da1ce067SMiklos Szeredi bool new_is_dir = false; 4196da1ce067SMiklos Szeredi unsigned max_links = new_dir->i_sb->s_max_links; 41971da177e4SLinus Torvalds 4198bc27027aSMiklos Szeredi if (source == target) 41991da177e4SLinus Torvalds return 0; 42001da177e4SLinus Torvalds 42011da177e4SLinus Torvalds error = may_delete(old_dir, old_dentry, is_dir); 42021da177e4SLinus Torvalds if (error) 42031da177e4SLinus Torvalds return error; 42041da177e4SLinus Torvalds 4205da1ce067SMiklos Szeredi if (!target) { 4206a95164d9SMiklos Szeredi error = may_create(new_dir, new_dentry); 4207da1ce067SMiklos Szeredi } else { 4208da1ce067SMiklos Szeredi new_is_dir = d_is_dir(new_dentry); 4209da1ce067SMiklos Szeredi 4210da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 42111da177e4SLinus Torvalds error = may_delete(new_dir, new_dentry, is_dir); 4212da1ce067SMiklos Szeredi else 4213da1ce067SMiklos Szeredi error = may_delete(new_dir, new_dentry, new_is_dir); 4214da1ce067SMiklos Szeredi } 42151da177e4SLinus Torvalds if (error) 42161da177e4SLinus Torvalds return error; 42171da177e4SLinus Torvalds 42187177a9c4SMiklos Szeredi if (!old_dir->i_op->rename && !old_dir->i_op->rename2) 42191da177e4SLinus Torvalds return -EPERM; 42201da177e4SLinus Torvalds 4221520c8b16SMiklos Szeredi if (flags && !old_dir->i_op->rename2) 4222520c8b16SMiklos Szeredi return -EINVAL; 4223520c8b16SMiklos Szeredi 4224bc27027aSMiklos Szeredi /* 4225bc27027aSMiklos Szeredi * If we are going to change the parent - check write permissions, 4226bc27027aSMiklos Szeredi * we'll need to flip '..'. 4227bc27027aSMiklos Szeredi */ 4228da1ce067SMiklos Szeredi if (new_dir != old_dir) { 4229da1ce067SMiklos Szeredi if (is_dir) { 4230bc27027aSMiklos Szeredi error = inode_permission(source, MAY_WRITE); 4231bc27027aSMiklos Szeredi if (error) 4232bc27027aSMiklos Szeredi return error; 4233bc27027aSMiklos Szeredi } 4234da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && new_is_dir) { 4235da1ce067SMiklos Szeredi error = inode_permission(target, MAY_WRITE); 4236da1ce067SMiklos Szeredi if (error) 4237da1ce067SMiklos Szeredi return error; 4238da1ce067SMiklos Szeredi } 4239da1ce067SMiklos Szeredi } 42400eeca283SRobert Love 42410b3974ebSMiklos Szeredi error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry, 42420b3974ebSMiklos Szeredi flags); 4243bc27027aSMiklos Szeredi if (error) 4244bc27027aSMiklos Szeredi return error; 4245bc27027aSMiklos Szeredi 4246bc27027aSMiklos Szeredi old_name = fsnotify_oldname_init(old_dentry->d_name.name); 4247bc27027aSMiklos Szeredi dget(new_dentry); 4248da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4249bc27027aSMiklos Szeredi lock_two_nondirectories(source, target); 4250bc27027aSMiklos Szeredi else if (target) 4251bc27027aSMiklos Szeredi mutex_lock(&target->i_mutex); 4252bc27027aSMiklos Szeredi 4253bc27027aSMiklos Szeredi error = -EBUSY; 42547af1364fSEric W. Biederman if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry)) 4255bc27027aSMiklos Szeredi goto out; 4256bc27027aSMiklos Szeredi 4257da1ce067SMiklos Szeredi if (max_links && new_dir != old_dir) { 4258bc27027aSMiklos Szeredi error = -EMLINK; 4259da1ce067SMiklos Szeredi if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links) 4260bc27027aSMiklos Szeredi goto out; 4261da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir && 4262da1ce067SMiklos Szeredi old_dir->i_nlink >= max_links) 4263da1ce067SMiklos Szeredi goto out; 4264da1ce067SMiklos Szeredi } 4265da1ce067SMiklos Szeredi if (is_dir && !(flags & RENAME_EXCHANGE) && target) 4266bc27027aSMiklos Szeredi shrink_dcache_parent(new_dentry); 4267da1ce067SMiklos Szeredi if (!is_dir) { 4268bc27027aSMiklos Szeredi error = try_break_deleg(source, delegated_inode); 4269bc27027aSMiklos Szeredi if (error) 4270bc27027aSMiklos Szeredi goto out; 4271da1ce067SMiklos Szeredi } 4272da1ce067SMiklos Szeredi if (target && !new_is_dir) { 4273bc27027aSMiklos Szeredi error = try_break_deleg(target, delegated_inode); 4274bc27027aSMiklos Szeredi if (error) 4275bc27027aSMiklos Szeredi goto out; 4276bc27027aSMiklos Szeredi } 42777177a9c4SMiklos Szeredi if (!old_dir->i_op->rename2) { 4278520c8b16SMiklos Szeredi error = old_dir->i_op->rename(old_dir, old_dentry, 4279520c8b16SMiklos Szeredi new_dir, new_dentry); 4280520c8b16SMiklos Szeredi } else { 42817177a9c4SMiklos Szeredi WARN_ON(old_dir->i_op->rename != NULL); 4282520c8b16SMiklos Szeredi error = old_dir->i_op->rename2(old_dir, old_dentry, 4283520c8b16SMiklos Szeredi new_dir, new_dentry, flags); 4284520c8b16SMiklos Szeredi } 4285bc27027aSMiklos Szeredi if (error) 4286bc27027aSMiklos Szeredi goto out; 4287bc27027aSMiklos Szeredi 4288da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE) && target) { 42891da177e4SLinus Torvalds if (is_dir) 4290bc27027aSMiklos Szeredi target->i_flags |= S_DEAD; 4291bc27027aSMiklos Szeredi dont_mount(new_dentry); 42928ed936b5SEric W. Biederman detach_mounts(new_dentry); 4293bc27027aSMiklos Szeredi } 4294da1ce067SMiklos Szeredi if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) { 4295da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 4296bc27027aSMiklos Szeredi d_move(old_dentry, new_dentry); 4297da1ce067SMiklos Szeredi else 4298da1ce067SMiklos Szeredi d_exchange(old_dentry, new_dentry); 4299da1ce067SMiklos Szeredi } 4300bc27027aSMiklos Szeredi out: 4301da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4302bc27027aSMiklos Szeredi unlock_two_nondirectories(source, target); 4303bc27027aSMiklos Szeredi else if (target) 4304bc27027aSMiklos Szeredi mutex_unlock(&target->i_mutex); 4305bc27027aSMiklos Szeredi dput(new_dentry); 4306da1ce067SMiklos Szeredi if (!error) { 4307123df294SAl Viro fsnotify_move(old_dir, new_dir, old_name, is_dir, 4308da1ce067SMiklos Szeredi !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry); 4309da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4310da1ce067SMiklos Szeredi fsnotify_move(new_dir, old_dir, old_dentry->d_name.name, 4311da1ce067SMiklos Szeredi new_is_dir, NULL, new_dentry); 4312da1ce067SMiklos Szeredi } 4313da1ce067SMiklos Szeredi } 43140eeca283SRobert Love fsnotify_oldname_free(old_name); 43150eeca283SRobert Love 43161da177e4SLinus Torvalds return error; 43171da177e4SLinus Torvalds } 43184d359507SAl Viro EXPORT_SYMBOL(vfs_rename); 43191da177e4SLinus Torvalds 4320520c8b16SMiklos Szeredi SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname, 4321520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname, unsigned int, flags) 43221da177e4SLinus Torvalds { 43231da177e4SLinus Torvalds struct dentry *old_dentry, *new_dentry; 43241da177e4SLinus Torvalds struct dentry *trap; 4325f5beed75SAl Viro struct path old_path, new_path; 4326f5beed75SAl Viro struct qstr old_last, new_last; 4327f5beed75SAl Viro int old_type, new_type; 43288e6d782cSJ. Bruce Fields struct inode *delegated_inode = NULL; 432991a27b2aSJeff Layton struct filename *from; 433091a27b2aSJeff Layton struct filename *to; 4331f5beed75SAl Viro unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET; 4332c6a94284SJeff Layton bool should_retry = false; 43332ad94ae6SAl Viro int error; 4334520c8b16SMiklos Szeredi 43350d7a8555SMiklos Szeredi if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) 4336da1ce067SMiklos Szeredi return -EINVAL; 4337da1ce067SMiklos Szeredi 43380d7a8555SMiklos Szeredi if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) && 43390d7a8555SMiklos Szeredi (flags & RENAME_EXCHANGE)) 4340520c8b16SMiklos Szeredi return -EINVAL; 4341520c8b16SMiklos Szeredi 43420d7a8555SMiklos Szeredi if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD)) 43430d7a8555SMiklos Szeredi return -EPERM; 43440d7a8555SMiklos Szeredi 4345f5beed75SAl Viro if (flags & RENAME_EXCHANGE) 4346f5beed75SAl Viro target_flags = 0; 4347f5beed75SAl Viro 4348c6a94284SJeff Layton retry: 4349f5beed75SAl Viro from = user_path_parent(olddfd, oldname, 4350f5beed75SAl Viro &old_path, &old_last, &old_type, lookup_flags); 435191a27b2aSJeff Layton if (IS_ERR(from)) { 435291a27b2aSJeff Layton error = PTR_ERR(from); 43531da177e4SLinus Torvalds goto exit; 435491a27b2aSJeff Layton } 43551da177e4SLinus Torvalds 4356f5beed75SAl Viro to = user_path_parent(newdfd, newname, 4357f5beed75SAl Viro &new_path, &new_last, &new_type, lookup_flags); 435891a27b2aSJeff Layton if (IS_ERR(to)) { 435991a27b2aSJeff Layton error = PTR_ERR(to); 43601da177e4SLinus Torvalds goto exit1; 436191a27b2aSJeff Layton } 43621da177e4SLinus Torvalds 43631da177e4SLinus Torvalds error = -EXDEV; 4364f5beed75SAl Viro if (old_path.mnt != new_path.mnt) 43651da177e4SLinus Torvalds goto exit2; 43661da177e4SLinus Torvalds 43671da177e4SLinus Torvalds error = -EBUSY; 4368f5beed75SAl Viro if (old_type != LAST_NORM) 43691da177e4SLinus Torvalds goto exit2; 43701da177e4SLinus Torvalds 43710a7c3937SMiklos Szeredi if (flags & RENAME_NOREPLACE) 43720a7c3937SMiklos Szeredi error = -EEXIST; 4373f5beed75SAl Viro if (new_type != LAST_NORM) 43741da177e4SLinus Torvalds goto exit2; 43751da177e4SLinus Torvalds 4376f5beed75SAl Viro error = mnt_want_write(old_path.mnt); 4377c30dabfeSJan Kara if (error) 4378c30dabfeSJan Kara goto exit2; 4379c30dabfeSJan Kara 43808e6d782cSJ. Bruce Fields retry_deleg: 4381f5beed75SAl Viro trap = lock_rename(new_path.dentry, old_path.dentry); 43821da177e4SLinus Torvalds 4383f5beed75SAl Viro old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags); 43841da177e4SLinus Torvalds error = PTR_ERR(old_dentry); 43851da177e4SLinus Torvalds if (IS_ERR(old_dentry)) 43861da177e4SLinus Torvalds goto exit3; 43871da177e4SLinus Torvalds /* source must exist */ 43881da177e4SLinus Torvalds error = -ENOENT; 4389b18825a7SDavid Howells if (d_is_negative(old_dentry)) 43901da177e4SLinus Torvalds goto exit4; 4391f5beed75SAl Viro new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags); 43921da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 43931da177e4SLinus Torvalds if (IS_ERR(new_dentry)) 43941da177e4SLinus Torvalds goto exit4; 43950a7c3937SMiklos Szeredi error = -EEXIST; 43960a7c3937SMiklos Szeredi if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry)) 43970a7c3937SMiklos Szeredi goto exit5; 4398da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4399da1ce067SMiklos Szeredi error = -ENOENT; 4400da1ce067SMiklos Szeredi if (d_is_negative(new_dentry)) 4401da1ce067SMiklos Szeredi goto exit5; 4402da1ce067SMiklos Szeredi 4403da1ce067SMiklos Szeredi if (!d_is_dir(new_dentry)) { 4404da1ce067SMiklos Szeredi error = -ENOTDIR; 4405f5beed75SAl Viro if (new_last.name[new_last.len]) 4406da1ce067SMiklos Szeredi goto exit5; 4407da1ce067SMiklos Szeredi } 4408da1ce067SMiklos Szeredi } 44090a7c3937SMiklos Szeredi /* unless the source is a directory trailing slashes give -ENOTDIR */ 44100a7c3937SMiklos Szeredi if (!d_is_dir(old_dentry)) { 44110a7c3937SMiklos Szeredi error = -ENOTDIR; 4412f5beed75SAl Viro if (old_last.name[old_last.len]) 44130a7c3937SMiklos Szeredi goto exit5; 4414f5beed75SAl Viro if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len]) 44150a7c3937SMiklos Szeredi goto exit5; 44160a7c3937SMiklos Szeredi } 44170a7c3937SMiklos Szeredi /* source should not be ancestor of target */ 44180a7c3937SMiklos Szeredi error = -EINVAL; 44190a7c3937SMiklos Szeredi if (old_dentry == trap) 44200a7c3937SMiklos Szeredi goto exit5; 44211da177e4SLinus Torvalds /* target should not be an ancestor of source */ 4422da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 44231da177e4SLinus Torvalds error = -ENOTEMPTY; 44241da177e4SLinus Torvalds if (new_dentry == trap) 44251da177e4SLinus Torvalds goto exit5; 44261da177e4SLinus Torvalds 4427f5beed75SAl Viro error = security_path_rename(&old_path, old_dentry, 4428f5beed75SAl Viro &new_path, new_dentry, flags); 4429be6d3e56SKentaro Takeda if (error) 4430c30dabfeSJan Kara goto exit5; 4431f5beed75SAl Viro error = vfs_rename(old_path.dentry->d_inode, old_dentry, 4432f5beed75SAl Viro new_path.dentry->d_inode, new_dentry, 4433520c8b16SMiklos Szeredi &delegated_inode, flags); 44341da177e4SLinus Torvalds exit5: 44351da177e4SLinus Torvalds dput(new_dentry); 44361da177e4SLinus Torvalds exit4: 44371da177e4SLinus Torvalds dput(old_dentry); 44381da177e4SLinus Torvalds exit3: 4439f5beed75SAl Viro unlock_rename(new_path.dentry, old_path.dentry); 44408e6d782cSJ. Bruce Fields if (delegated_inode) { 44418e6d782cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 44428e6d782cSJ. Bruce Fields if (!error) 44438e6d782cSJ. Bruce Fields goto retry_deleg; 44448e6d782cSJ. Bruce Fields } 4445f5beed75SAl Viro mnt_drop_write(old_path.mnt); 44461da177e4SLinus Torvalds exit2: 4447c6a94284SJeff Layton if (retry_estale(error, lookup_flags)) 4448c6a94284SJeff Layton should_retry = true; 4449f5beed75SAl Viro path_put(&new_path); 44502ad94ae6SAl Viro putname(to); 44511da177e4SLinus Torvalds exit1: 4452f5beed75SAl Viro path_put(&old_path); 44531da177e4SLinus Torvalds putname(from); 4454c6a94284SJeff Layton if (should_retry) { 4455c6a94284SJeff Layton should_retry = false; 4456c6a94284SJeff Layton lookup_flags |= LOOKUP_REVAL; 4457c6a94284SJeff Layton goto retry; 4458c6a94284SJeff Layton } 44592ad94ae6SAl Viro exit: 44601da177e4SLinus Torvalds return error; 44611da177e4SLinus Torvalds } 44621da177e4SLinus Torvalds 4463520c8b16SMiklos Szeredi SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, 4464520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname) 4465520c8b16SMiklos Szeredi { 4466520c8b16SMiklos Szeredi return sys_renameat2(olddfd, oldname, newdfd, newname, 0); 4467520c8b16SMiklos Szeredi } 4468520c8b16SMiklos Szeredi 4469a26eab24SHeiko Carstens SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) 44705590ff0dSUlrich Drepper { 4471520c8b16SMiklos Szeredi return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 44725590ff0dSUlrich Drepper } 44735590ff0dSUlrich Drepper 4474787fb6bcSMiklos Szeredi int vfs_whiteout(struct inode *dir, struct dentry *dentry) 4475787fb6bcSMiklos Szeredi { 4476787fb6bcSMiklos Szeredi int error = may_create(dir, dentry); 4477787fb6bcSMiklos Szeredi if (error) 4478787fb6bcSMiklos Szeredi return error; 4479787fb6bcSMiklos Szeredi 4480787fb6bcSMiklos Szeredi if (!dir->i_op->mknod) 4481787fb6bcSMiklos Szeredi return -EPERM; 4482787fb6bcSMiklos Szeredi 4483787fb6bcSMiklos Szeredi return dir->i_op->mknod(dir, dentry, 4484787fb6bcSMiklos Szeredi S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV); 4485787fb6bcSMiklos Szeredi } 4486787fb6bcSMiklos Szeredi EXPORT_SYMBOL(vfs_whiteout); 4487787fb6bcSMiklos Szeredi 44885d826c84SAl Viro int readlink_copy(char __user *buffer, int buflen, const char *link) 44891da177e4SLinus Torvalds { 44905d826c84SAl Viro int len = PTR_ERR(link); 44911da177e4SLinus Torvalds if (IS_ERR(link)) 44921da177e4SLinus Torvalds goto out; 44931da177e4SLinus Torvalds 44941da177e4SLinus Torvalds len = strlen(link); 44951da177e4SLinus Torvalds if (len > (unsigned) buflen) 44961da177e4SLinus Torvalds len = buflen; 44971da177e4SLinus Torvalds if (copy_to_user(buffer, link, len)) 44981da177e4SLinus Torvalds len = -EFAULT; 44991da177e4SLinus Torvalds out: 45001da177e4SLinus Torvalds return len; 45011da177e4SLinus Torvalds } 45025d826c84SAl Viro EXPORT_SYMBOL(readlink_copy); 45031da177e4SLinus Torvalds 45041da177e4SLinus Torvalds /* 45051da177e4SLinus Torvalds * A helper for ->readlink(). This should be used *ONLY* for symlinks that 45066b255391SAl Viro * have ->get_link() not calling nd_jump_link(). Using (or not using) it 45076b255391SAl Viro * for any given inode is up to filesystem. 45081da177e4SLinus Torvalds */ 45091da177e4SLinus Torvalds int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen) 45101da177e4SLinus Torvalds { 4511cc314eefSLinus Torvalds void *cookie; 45125f2c4179SAl Viro struct inode *inode = d_inode(dentry); 45135f2c4179SAl Viro const char *link = inode->i_link; 4514694a1764SMarcin Slusarz int res; 4515cc314eefSLinus Torvalds 4516d4dee48bSAl Viro if (!link) { 45176b255391SAl Viro link = inode->i_op->get_link(dentry, inode, &cookie); 4518680baacbSAl Viro if (IS_ERR(link)) 4519680baacbSAl Viro return PTR_ERR(link); 4520d4dee48bSAl Viro } 4521680baacbSAl Viro res = readlink_copy(buffer, buflen, link); 45225f2c4179SAl Viro if (inode->i_op->put_link) 45235f2c4179SAl Viro inode->i_op->put_link(inode, cookie); 4524694a1764SMarcin Slusarz return res; 45251da177e4SLinus Torvalds } 45264d359507SAl Viro EXPORT_SYMBOL(generic_readlink); 45271da177e4SLinus Torvalds 45281da177e4SLinus Torvalds /* get the link contents into pagecache */ 45296b255391SAl Viro const char *page_get_link(struct dentry *dentry, struct inode *inode, 45306b255391SAl Viro void **cookie) 45311da177e4SLinus Torvalds { 4532ebd09abbSDuane Griffin char *kaddr; 45331da177e4SLinus Torvalds struct page *page; 45346b255391SAl Viro struct address_space *mapping = inode->i_mapping; 45356b255391SAl Viro 45366b255391SAl Viro if (!dentry) 45376b255391SAl Viro return ERR_PTR(-ECHILD); 45386b255391SAl Viro 4539090d2b18SPekka Enberg page = read_mapping_page(mapping, 0, NULL); 45401da177e4SLinus Torvalds if (IS_ERR(page)) 45416fe6900eSNick Piggin return (char*)page; 4542aa80deabSAl Viro *cookie = page; 454321fc61c7SAl Viro BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM); 454421fc61c7SAl Viro kaddr = page_address(page); 45456b255391SAl Viro nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1); 4546ebd09abbSDuane Griffin return kaddr; 45471da177e4SLinus Torvalds } 45481da177e4SLinus Torvalds 45496b255391SAl Viro EXPORT_SYMBOL(page_get_link); 45501da177e4SLinus Torvalds 45515f2c4179SAl Viro void page_put_link(struct inode *unused, void *cookie) 45521da177e4SLinus Torvalds { 4553cc314eefSLinus Torvalds struct page *page = cookie; 45541da177e4SLinus Torvalds page_cache_release(page); 45551da177e4SLinus Torvalds } 45564d359507SAl Viro EXPORT_SYMBOL(page_put_link); 45571da177e4SLinus Torvalds 4558aa80deabSAl Viro int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 4559aa80deabSAl Viro { 4560aa80deabSAl Viro void *cookie = NULL; 45616b255391SAl Viro int res = readlink_copy(buffer, buflen, 45626b255391SAl Viro page_get_link(dentry, d_inode(dentry), 45636b255391SAl Viro &cookie)); 4564aa80deabSAl Viro if (cookie) 4565aa80deabSAl Viro page_put_link(NULL, cookie); 4566aa80deabSAl Viro return res; 4567aa80deabSAl Viro } 4568aa80deabSAl Viro EXPORT_SYMBOL(page_readlink); 4569aa80deabSAl Viro 457054566b2cSNick Piggin /* 457154566b2cSNick Piggin * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS 457254566b2cSNick Piggin */ 457354566b2cSNick Piggin int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) 45741da177e4SLinus Torvalds { 45751da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 45760adb25d2SKirill Korotaev struct page *page; 4577afddba49SNick Piggin void *fsdata; 4578beb497abSDmitriy Monakhov int err; 457954566b2cSNick Piggin unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE; 458054566b2cSNick Piggin if (nofs) 458154566b2cSNick Piggin flags |= AOP_FLAG_NOFS; 45821da177e4SLinus Torvalds 45837e53cac4SNeilBrown retry: 4584afddba49SNick Piggin err = pagecache_write_begin(NULL, mapping, 0, len-1, 458554566b2cSNick Piggin flags, &page, &fsdata); 45861da177e4SLinus Torvalds if (err) 4587afddba49SNick Piggin goto fail; 4588afddba49SNick Piggin 458921fc61c7SAl Viro memcpy(page_address(page), symname, len-1); 4590afddba49SNick Piggin 4591afddba49SNick Piggin err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, 4592afddba49SNick Piggin page, fsdata); 45931da177e4SLinus Torvalds if (err < 0) 45941da177e4SLinus Torvalds goto fail; 4595afddba49SNick Piggin if (err < len-1) 4596afddba49SNick Piggin goto retry; 4597afddba49SNick Piggin 45981da177e4SLinus Torvalds mark_inode_dirty(inode); 45991da177e4SLinus Torvalds return 0; 46001da177e4SLinus Torvalds fail: 46011da177e4SLinus Torvalds return err; 46021da177e4SLinus Torvalds } 46034d359507SAl Viro EXPORT_SYMBOL(__page_symlink); 46041da177e4SLinus Torvalds 46050adb25d2SKirill Korotaev int page_symlink(struct inode *inode, const char *symname, int len) 46060adb25d2SKirill Korotaev { 46070adb25d2SKirill Korotaev return __page_symlink(inode, symname, len, 4608c62d2555SMichal Hocko !mapping_gfp_constraint(inode->i_mapping, __GFP_FS)); 46090adb25d2SKirill Korotaev } 46104d359507SAl Viro EXPORT_SYMBOL(page_symlink); 46110adb25d2SKirill Korotaev 461292e1d5beSArjan van de Ven const struct inode_operations page_symlink_inode_operations = { 46131da177e4SLinus Torvalds .readlink = generic_readlink, 46146b255391SAl Viro .get_link = page_get_link, 46151da177e4SLinus Torvalds .put_link = page_put_link, 46161da177e4SLinus Torvalds }; 46171da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink_inode_operations); 4618