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; 508fceef393SAl Viro struct delayed_call done; 509697fc6caSAl Viro const char *name; 5100450b2d1SAl Viro unsigned seq; 511894bc8c4SAl Viro } *stack, internal[EMBEDDED_LEVELS]; 5129883d185SAl Viro struct filename *name; 5139883d185SAl Viro struct nameidata *saved; 514fceef393SAl Viro struct inode *link_inode; 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; 537e1a63bbcSAl Viro if (now->stack != now->internal) 538756daf26SNeilBrown kfree(now->stack); 539894bc8c4SAl Viro } 540894bc8c4SAl Viro 541894bc8c4SAl Viro static int __nd_alloc_stack(struct nameidata *nd) 542894bc8c4SAl Viro { 543bc40aee0SAl Viro struct saved *p; 544bc40aee0SAl Viro 545bc40aee0SAl Viro if (nd->flags & LOOKUP_RCU) { 546bc40aee0SAl Viro p= kmalloc(MAXSYMLINKS * sizeof(struct saved), 547bc40aee0SAl Viro GFP_ATOMIC); 548bc40aee0SAl Viro if (unlikely(!p)) 549bc40aee0SAl Viro return -ECHILD; 550bc40aee0SAl Viro } else { 551bc40aee0SAl Viro p= kmalloc(MAXSYMLINKS * sizeof(struct saved), 552894bc8c4SAl Viro GFP_KERNEL); 553894bc8c4SAl Viro if (unlikely(!p)) 554894bc8c4SAl Viro return -ENOMEM; 555bc40aee0SAl Viro } 556894bc8c4SAl Viro memcpy(p, nd->internal, sizeof(nd->internal)); 557894bc8c4SAl Viro nd->stack = p; 558894bc8c4SAl Viro return 0; 559894bc8c4SAl Viro } 560894bc8c4SAl Viro 561397d425dSEric W. Biederman /** 562397d425dSEric W. Biederman * path_connected - Verify that a path->dentry is below path->mnt.mnt_root 563397d425dSEric W. Biederman * @path: nameidate to verify 564397d425dSEric W. Biederman * 565397d425dSEric W. Biederman * Rename can sometimes move a file or directory outside of a bind 566397d425dSEric W. Biederman * mount, path_connected allows those cases to be detected. 567397d425dSEric W. Biederman */ 568397d425dSEric W. Biederman static bool path_connected(const struct path *path) 569397d425dSEric W. Biederman { 570397d425dSEric W. Biederman struct vfsmount *mnt = path->mnt; 571397d425dSEric W. Biederman 572397d425dSEric W. Biederman /* Only bind mounts can have disconnected paths */ 573397d425dSEric W. Biederman if (mnt->mnt_root == mnt->mnt_sb->s_root) 574397d425dSEric W. Biederman return true; 575397d425dSEric W. Biederman 576397d425dSEric W. Biederman return is_subdir(path->dentry, mnt->mnt_root); 577397d425dSEric W. Biederman } 578397d425dSEric W. Biederman 579894bc8c4SAl Viro static inline int nd_alloc_stack(struct nameidata *nd) 580894bc8c4SAl Viro { 581da4e0be0SAl Viro if (likely(nd->depth != EMBEDDED_LEVELS)) 582894bc8c4SAl Viro return 0; 583894bc8c4SAl Viro if (likely(nd->stack != nd->internal)) 584894bc8c4SAl Viro return 0; 585894bc8c4SAl Viro return __nd_alloc_stack(nd); 586894bc8c4SAl Viro } 587894bc8c4SAl Viro 5887973387aSAl Viro static void drop_links(struct nameidata *nd) 5897973387aSAl Viro { 5907973387aSAl Viro int i = nd->depth; 5917973387aSAl Viro while (i--) { 5927973387aSAl Viro struct saved *last = nd->stack + i; 593fceef393SAl Viro do_delayed_call(&last->done); 594fceef393SAl Viro clear_delayed_call(&last->done); 5957973387aSAl Viro } 5967973387aSAl Viro } 5977973387aSAl Viro 5987973387aSAl Viro static void terminate_walk(struct nameidata *nd) 5997973387aSAl Viro { 6007973387aSAl Viro drop_links(nd); 6017973387aSAl Viro if (!(nd->flags & LOOKUP_RCU)) { 6027973387aSAl Viro int i; 6037973387aSAl Viro path_put(&nd->path); 6047973387aSAl Viro for (i = 0; i < nd->depth; i++) 6057973387aSAl Viro path_put(&nd->stack[i].link); 606102b8af2SAl Viro if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 607102b8af2SAl Viro path_put(&nd->root); 608102b8af2SAl Viro nd->root.mnt = NULL; 609102b8af2SAl Viro } 6107973387aSAl Viro } else { 6117973387aSAl Viro nd->flags &= ~LOOKUP_RCU; 6127973387aSAl Viro if (!(nd->flags & LOOKUP_ROOT)) 6137973387aSAl Viro nd->root.mnt = NULL; 6147973387aSAl Viro rcu_read_unlock(); 6157973387aSAl Viro } 6167973387aSAl Viro nd->depth = 0; 6177973387aSAl Viro } 6187973387aSAl Viro 6197973387aSAl Viro /* path_put is needed afterwards regardless of success or failure */ 6207973387aSAl Viro static bool legitimize_path(struct nameidata *nd, 6217973387aSAl Viro struct path *path, unsigned seq) 6227973387aSAl Viro { 6237973387aSAl Viro int res = __legitimize_mnt(path->mnt, nd->m_seq); 6247973387aSAl Viro if (unlikely(res)) { 6257973387aSAl Viro if (res > 0) 6267973387aSAl Viro path->mnt = NULL; 6277973387aSAl Viro path->dentry = NULL; 6287973387aSAl Viro return false; 6297973387aSAl Viro } 6307973387aSAl Viro if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) { 6317973387aSAl Viro path->dentry = NULL; 6327973387aSAl Viro return false; 6337973387aSAl Viro } 6347973387aSAl Viro return !read_seqcount_retry(&path->dentry->d_seq, seq); 6357973387aSAl Viro } 6367973387aSAl Viro 6377973387aSAl Viro static bool legitimize_links(struct nameidata *nd) 6387973387aSAl Viro { 6397973387aSAl Viro int i; 6407973387aSAl Viro for (i = 0; i < nd->depth; i++) { 6417973387aSAl Viro struct saved *last = nd->stack + i; 6427973387aSAl Viro if (unlikely(!legitimize_path(nd, &last->link, last->seq))) { 6437973387aSAl Viro drop_links(nd); 6447973387aSAl Viro nd->depth = i + 1; 6457973387aSAl Viro return false; 6467973387aSAl Viro } 6477973387aSAl Viro } 6487973387aSAl Viro return true; 6497973387aSAl Viro } 6507973387aSAl Viro 65119660af7SAl Viro /* 65231e6b01fSNick Piggin * Path walking has 2 modes, rcu-walk and ref-walk (see 65319660af7SAl Viro * Documentation/filesystems/path-lookup.txt). In situations when we can't 65419660af7SAl Viro * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab 65557e3715cSMike Marshall * normal reference counts on dentries and vfsmounts to transition to ref-walk 65619660af7SAl Viro * mode. Refcounts are grabbed at the last known good point before rcu-walk 65719660af7SAl Viro * got stuck, so ref-walk may continue from there. If this is not successful 65819660af7SAl Viro * (eg. a seqcount has changed), then failure is returned and it's up to caller 65919660af7SAl Viro * to restart the path walk from the beginning in ref-walk mode. 66031e6b01fSNick Piggin */ 66131e6b01fSNick Piggin 66231e6b01fSNick Piggin /** 66319660af7SAl Viro * unlazy_walk - try to switch to ref-walk mode. 66419660af7SAl Viro * @nd: nameidata pathwalk data 66519660af7SAl Viro * @dentry: child of nd->path.dentry or NULL 6666e9918b7SAl Viro * @seq: seq number to check dentry against 66739191628SRandy Dunlap * Returns: 0 on success, -ECHILD on failure 66831e6b01fSNick Piggin * 66919660af7SAl Viro * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry 67019660af7SAl Viro * for ref-walk mode. @dentry must be a path found by a do_lookup call on 67119660af7SAl Viro * @nd or NULL. Must be called from rcu-walk context. 6727973387aSAl Viro * Nothing should touch nameidata between unlazy_walk() failure and 6737973387aSAl Viro * terminate_walk(). 67431e6b01fSNick Piggin */ 6756e9918b7SAl Viro static int unlazy_walk(struct nameidata *nd, struct dentry *dentry, unsigned seq) 67631e6b01fSNick Piggin { 67731e6b01fSNick Piggin struct dentry *parent = nd->path.dentry; 67831e6b01fSNick Piggin 67931e6b01fSNick Piggin BUG_ON(!(nd->flags & LOOKUP_RCU)); 680e5c832d5SLinus Torvalds 681e5c832d5SLinus Torvalds nd->flags &= ~LOOKUP_RCU; 6827973387aSAl Viro if (unlikely(!legitimize_links(nd))) 6837973387aSAl Viro goto out2; 6847973387aSAl Viro if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq))) 6857973387aSAl Viro goto out2; 6867973387aSAl Viro if (unlikely(!lockref_get_not_dead(&parent->d_lockref))) 6877973387aSAl Viro goto out1; 68848a066e7SAl Viro 68915570086SLinus Torvalds /* 69015570086SLinus Torvalds * For a negative lookup, the lookup sequence point is the parents 69115570086SLinus Torvalds * sequence point, and it only needs to revalidate the parent dentry. 69215570086SLinus Torvalds * 69315570086SLinus Torvalds * For a positive lookup, we need to move both the parent and the 69415570086SLinus Torvalds * dentry from the RCU domain to be properly refcounted. And the 69515570086SLinus Torvalds * sequence number in the dentry validates *both* dentry counters, 69615570086SLinus Torvalds * since we checked the sequence number of the parent after we got 69715570086SLinus Torvalds * the child sequence number. So we know the parent must still 69815570086SLinus Torvalds * be valid if the child sequence number is still valid. 69915570086SLinus Torvalds */ 70019660af7SAl Viro if (!dentry) { 701e5c832d5SLinus Torvalds if (read_seqcount_retry(&parent->d_seq, nd->seq)) 702e5c832d5SLinus Torvalds goto out; 70319660af7SAl Viro BUG_ON(nd->inode != parent->d_inode); 70419660af7SAl Viro } else { 705e5c832d5SLinus Torvalds if (!lockref_get_not_dead(&dentry->d_lockref)) 706e5c832d5SLinus Torvalds goto out; 7076e9918b7SAl Viro if (read_seqcount_retry(&dentry->d_seq, seq)) 708e5c832d5SLinus Torvalds goto drop_dentry; 70919660af7SAl Viro } 710e5c832d5SLinus Torvalds 711e5c832d5SLinus Torvalds /* 712e5c832d5SLinus Torvalds * Sequence counts matched. Now make sure that the root is 713e5c832d5SLinus Torvalds * still valid and get it if required. 714e5c832d5SLinus Torvalds */ 715e5c832d5SLinus Torvalds if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 7165a8d87e8SAl Viro if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) { 7175a8d87e8SAl Viro rcu_read_unlock(); 7185a8d87e8SAl Viro dput(dentry); 7195a8d87e8SAl Viro return -ECHILD; 7207973387aSAl Viro } 72131e6b01fSNick Piggin } 72231e6b01fSNick Piggin 7238b61e74fSAl Viro rcu_read_unlock(); 72431e6b01fSNick Piggin return 0; 72519660af7SAl Viro 726e5c832d5SLinus Torvalds drop_dentry: 7278b61e74fSAl Viro rcu_read_unlock(); 728e5c832d5SLinus Torvalds dput(dentry); 729d0d27277SLinus Torvalds goto drop_root_mnt; 7307973387aSAl Viro out2: 7317973387aSAl Viro nd->path.mnt = NULL; 7327973387aSAl Viro out1: 7337973387aSAl Viro nd->path.dentry = NULL; 734e5c832d5SLinus Torvalds out: 7358b61e74fSAl Viro rcu_read_unlock(); 736d0d27277SLinus Torvalds drop_root_mnt: 737d0d27277SLinus Torvalds if (!(nd->flags & LOOKUP_ROOT)) 738d0d27277SLinus Torvalds nd->root.mnt = NULL; 73931e6b01fSNick Piggin return -ECHILD; 74031e6b01fSNick Piggin } 74131e6b01fSNick Piggin 7427973387aSAl Viro static int unlazy_link(struct nameidata *nd, struct path *link, unsigned seq) 7437973387aSAl Viro { 7447973387aSAl Viro if (unlikely(!legitimize_path(nd, link, seq))) { 7457973387aSAl Viro drop_links(nd); 7467973387aSAl Viro nd->depth = 0; 7477973387aSAl Viro nd->flags &= ~LOOKUP_RCU; 7487973387aSAl Viro nd->path.mnt = NULL; 7497973387aSAl Viro nd->path.dentry = NULL; 7507973387aSAl Viro if (!(nd->flags & LOOKUP_ROOT)) 7517973387aSAl Viro nd->root.mnt = NULL; 7527973387aSAl Viro rcu_read_unlock(); 7537973387aSAl Viro } else if (likely(unlazy_walk(nd, NULL, 0)) == 0) { 7547973387aSAl Viro return 0; 7557973387aSAl Viro } 7567973387aSAl Viro path_put(link); 7577973387aSAl Viro return -ECHILD; 7587973387aSAl Viro } 7597973387aSAl Viro 7604ce16ef3SAl Viro static inline int d_revalidate(struct dentry *dentry, unsigned int flags) 76134286d66SNick Piggin { 7624ce16ef3SAl Viro return dentry->d_op->d_revalidate(dentry, flags); 76334286d66SNick Piggin } 76434286d66SNick Piggin 7659f1fafeeSAl Viro /** 7669f1fafeeSAl Viro * complete_walk - successful completion of path walk 7679f1fafeeSAl Viro * @nd: pointer nameidata 76839159de2SJeff Layton * 7699f1fafeeSAl Viro * If we had been in RCU mode, drop out of it and legitimize nd->path. 7709f1fafeeSAl Viro * Revalidate the final result, unless we'd already done that during 7719f1fafeeSAl Viro * the path walk or the filesystem doesn't ask for it. Return 0 on 7729f1fafeeSAl Viro * success, -error on failure. In case of failure caller does not 7739f1fafeeSAl Viro * need to drop nd->path. 77439159de2SJeff Layton */ 7759f1fafeeSAl Viro static int complete_walk(struct nameidata *nd) 77639159de2SJeff Layton { 77716c2cd71SAl Viro struct dentry *dentry = nd->path.dentry; 77839159de2SJeff Layton int status; 77939159de2SJeff Layton 7809f1fafeeSAl Viro if (nd->flags & LOOKUP_RCU) { 7819f1fafeeSAl Viro if (!(nd->flags & LOOKUP_ROOT)) 7829f1fafeeSAl Viro nd->root.mnt = NULL; 7836e9918b7SAl Viro if (unlikely(unlazy_walk(nd, NULL, 0))) 78448a066e7SAl Viro return -ECHILD; 78548a066e7SAl Viro } 7869f1fafeeSAl Viro 78716c2cd71SAl Viro if (likely(!(nd->flags & LOOKUP_JUMPED))) 78839159de2SJeff Layton return 0; 78939159de2SJeff Layton 790ecf3d1f1SJeff Layton if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) 79116c2cd71SAl Viro return 0; 79216c2cd71SAl Viro 793ecf3d1f1SJeff Layton status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); 79439159de2SJeff Layton if (status > 0) 79539159de2SJeff Layton return 0; 79639159de2SJeff Layton 79716c2cd71SAl Viro if (!status) 79839159de2SJeff Layton status = -ESTALE; 79916c2cd71SAl Viro 80039159de2SJeff Layton return status; 80139159de2SJeff Layton } 80239159de2SJeff Layton 80318d8c860SAl Viro static void set_root(struct nameidata *nd) 8042a737871SAl Viro { 80531e6b01fSNick Piggin struct fs_struct *fs = current->fs; 8069e6697e2SAl Viro 8079e6697e2SAl Viro if (nd->flags & LOOKUP_RCU) { 8088f47a016SAl Viro unsigned seq; 809c28cc364SNick Piggin 810c28cc364SNick Piggin do { 811c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 81231e6b01fSNick Piggin nd->root = fs->root; 8138f47a016SAl Viro nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq); 814c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 8159e6697e2SAl Viro } else { 8169e6697e2SAl Viro get_fs_root(fs, &nd->root); 8179e6697e2SAl Viro } 81831e6b01fSNick Piggin } 81931e6b01fSNick Piggin 8201d957f9bSJan Blunck static void path_put_conditional(struct path *path, struct nameidata *nd) 821051d3812SIan Kent { 822051d3812SIan Kent dput(path->dentry); 8234ac91378SJan Blunck if (path->mnt != nd->path.mnt) 824051d3812SIan Kent mntput(path->mnt); 825051d3812SIan Kent } 826051d3812SIan Kent 8277b9337aaSNick Piggin static inline void path_to_nameidata(const struct path *path, 8287b9337aaSNick Piggin struct nameidata *nd) 829051d3812SIan Kent { 83031e6b01fSNick Piggin if (!(nd->flags & LOOKUP_RCU)) { 8314ac91378SJan Blunck dput(nd->path.dentry); 83231e6b01fSNick Piggin if (nd->path.mnt != path->mnt) 8334ac91378SJan Blunck mntput(nd->path.mnt); 8349a229683SHuang Shijie } 83531e6b01fSNick Piggin nd->path.mnt = path->mnt; 8364ac91378SJan Blunck nd->path.dentry = path->dentry; 837051d3812SIan Kent } 838051d3812SIan Kent 839248fb5b9SAl Viro static int nd_jump_root(struct nameidata *nd) 840248fb5b9SAl Viro { 841248fb5b9SAl Viro if (nd->flags & LOOKUP_RCU) { 842248fb5b9SAl Viro struct dentry *d; 843248fb5b9SAl Viro nd->path = nd->root; 844248fb5b9SAl Viro d = nd->path.dentry; 845248fb5b9SAl Viro nd->inode = d->d_inode; 846248fb5b9SAl Viro nd->seq = nd->root_seq; 847248fb5b9SAl Viro if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq))) 848248fb5b9SAl Viro return -ECHILD; 849248fb5b9SAl Viro } else { 850248fb5b9SAl Viro path_put(&nd->path); 851248fb5b9SAl Viro nd->path = nd->root; 852248fb5b9SAl Viro path_get(&nd->path); 853248fb5b9SAl Viro nd->inode = nd->path.dentry->d_inode; 854248fb5b9SAl Viro } 855248fb5b9SAl Viro nd->flags |= LOOKUP_JUMPED; 856248fb5b9SAl Viro return 0; 857248fb5b9SAl Viro } 858248fb5b9SAl Viro 859b5fb63c1SChristoph Hellwig /* 8606b255391SAl Viro * Helper to directly jump to a known parsed path from ->get_link, 861b5fb63c1SChristoph Hellwig * caller must have taken a reference to path beforehand. 862b5fb63c1SChristoph Hellwig */ 8636e77137bSAl Viro void nd_jump_link(struct path *path) 864b5fb63c1SChristoph Hellwig { 8656e77137bSAl Viro struct nameidata *nd = current->nameidata; 866b5fb63c1SChristoph Hellwig path_put(&nd->path); 867b5fb63c1SChristoph Hellwig 868b5fb63c1SChristoph Hellwig nd->path = *path; 869b5fb63c1SChristoph Hellwig nd->inode = nd->path.dentry->d_inode; 870b5fb63c1SChristoph Hellwig nd->flags |= LOOKUP_JUMPED; 871b5fb63c1SChristoph Hellwig } 872b5fb63c1SChristoph Hellwig 873b9ff4429SAl Viro static inline void put_link(struct nameidata *nd) 874574197e0SAl Viro { 87521c3003dSAl Viro struct saved *last = nd->stack + --nd->depth; 876fceef393SAl Viro do_delayed_call(&last->done); 8776548fae2SAl Viro if (!(nd->flags & LOOKUP_RCU)) 878b9ff4429SAl Viro path_put(&last->link); 879574197e0SAl Viro } 880574197e0SAl Viro 881561ec64aSLinus Torvalds int sysctl_protected_symlinks __read_mostly = 0; 882561ec64aSLinus Torvalds int sysctl_protected_hardlinks __read_mostly = 0; 883800179c9SKees Cook 884800179c9SKees Cook /** 885800179c9SKees Cook * may_follow_link - Check symlink following for unsafe situations 88655852635SRandy Dunlap * @nd: nameidata pathwalk data 887800179c9SKees Cook * 888800179c9SKees Cook * In the case of the sysctl_protected_symlinks sysctl being enabled, 889800179c9SKees Cook * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is 890800179c9SKees Cook * in a sticky world-writable directory. This is to protect privileged 891800179c9SKees Cook * processes from failing races against path names that may change out 892800179c9SKees Cook * from under them by way of other users creating malicious symlinks. 893800179c9SKees Cook * It will permit symlinks to be followed only when outside a sticky 894800179c9SKees Cook * world-writable directory, or when the uid of the symlink and follower 895800179c9SKees Cook * match, or when the directory owner matches the symlink's owner. 896800179c9SKees Cook * 897800179c9SKees Cook * Returns 0 if following the symlink is allowed, -ve on error. 898800179c9SKees Cook */ 899fec2fa24SAl Viro static inline int may_follow_link(struct nameidata *nd) 900800179c9SKees Cook { 901800179c9SKees Cook const struct inode *inode; 902800179c9SKees Cook const struct inode *parent; 903800179c9SKees Cook 904800179c9SKees Cook if (!sysctl_protected_symlinks) 905800179c9SKees Cook return 0; 906800179c9SKees Cook 907800179c9SKees Cook /* Allowed if owner and follower match. */ 908fceef393SAl Viro inode = nd->link_inode; 90981abe27bSEric W. Biederman if (uid_eq(current_cred()->fsuid, inode->i_uid)) 910800179c9SKees Cook return 0; 911800179c9SKees Cook 912800179c9SKees Cook /* Allowed if parent directory not sticky and world-writable. */ 913aa65fa35SAl Viro parent = nd->inode; 914800179c9SKees Cook if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) 915800179c9SKees Cook return 0; 916800179c9SKees Cook 917800179c9SKees Cook /* Allowed if parent directory and link owner match. */ 91881abe27bSEric W. Biederman if (uid_eq(parent->i_uid, inode->i_uid)) 919800179c9SKees Cook return 0; 920800179c9SKees Cook 92131956502SAl Viro if (nd->flags & LOOKUP_RCU) 92231956502SAl Viro return -ECHILD; 92331956502SAl Viro 9241cf2665bSAl Viro audit_log_link_denied("follow_link", &nd->stack[0].link); 925800179c9SKees Cook return -EACCES; 926800179c9SKees Cook } 927800179c9SKees Cook 928800179c9SKees Cook /** 929800179c9SKees Cook * safe_hardlink_source - Check for safe hardlink conditions 930800179c9SKees Cook * @inode: the source inode to hardlink from 931800179c9SKees Cook * 932800179c9SKees Cook * Return false if at least one of the following conditions: 933800179c9SKees Cook * - inode is not a regular file 934800179c9SKees Cook * - inode is setuid 935800179c9SKees Cook * - inode is setgid and group-exec 936800179c9SKees Cook * - access failure for read and write 937800179c9SKees Cook * 938800179c9SKees Cook * Otherwise returns true. 939800179c9SKees Cook */ 940800179c9SKees Cook static bool safe_hardlink_source(struct inode *inode) 941800179c9SKees Cook { 942800179c9SKees Cook umode_t mode = inode->i_mode; 943800179c9SKees Cook 944800179c9SKees Cook /* Special files should not get pinned to the filesystem. */ 945800179c9SKees Cook if (!S_ISREG(mode)) 946800179c9SKees Cook return false; 947800179c9SKees Cook 948800179c9SKees Cook /* Setuid files should not get pinned to the filesystem. */ 949800179c9SKees Cook if (mode & S_ISUID) 950800179c9SKees Cook return false; 951800179c9SKees Cook 952800179c9SKees Cook /* Executable setgid files should not get pinned to the filesystem. */ 953800179c9SKees Cook if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) 954800179c9SKees Cook return false; 955800179c9SKees Cook 956800179c9SKees Cook /* Hardlinking to unreadable or unwritable sources is dangerous. */ 957800179c9SKees Cook if (inode_permission(inode, MAY_READ | MAY_WRITE)) 958800179c9SKees Cook return false; 959800179c9SKees Cook 960800179c9SKees Cook return true; 961800179c9SKees Cook } 962800179c9SKees Cook 963800179c9SKees Cook /** 964800179c9SKees Cook * may_linkat - Check permissions for creating a hardlink 965800179c9SKees Cook * @link: the source to hardlink from 966800179c9SKees Cook * 967800179c9SKees Cook * Block hardlink when all of: 968800179c9SKees Cook * - sysctl_protected_hardlinks enabled 969800179c9SKees Cook * - fsuid does not match inode 970800179c9SKees Cook * - hardlink source is unsafe (see safe_hardlink_source() above) 971f2ca3796SDirk Steinmetz * - not CAP_FOWNER in a namespace with the inode owner uid mapped 972800179c9SKees Cook * 973800179c9SKees Cook * Returns 0 if successful, -ve on error. 974800179c9SKees Cook */ 975800179c9SKees Cook static int may_linkat(struct path *link) 976800179c9SKees Cook { 977800179c9SKees Cook struct inode *inode; 978800179c9SKees Cook 979800179c9SKees Cook if (!sysctl_protected_hardlinks) 980800179c9SKees Cook return 0; 981800179c9SKees Cook 982800179c9SKees Cook inode = link->dentry->d_inode; 983800179c9SKees Cook 984800179c9SKees Cook /* Source inode owner (or CAP_FOWNER) can hardlink all they like, 985800179c9SKees Cook * otherwise, it must be a safe source. 986800179c9SKees Cook */ 987f2ca3796SDirk Steinmetz if (inode_owner_or_capable(inode) || safe_hardlink_source(inode)) 988800179c9SKees Cook return 0; 989800179c9SKees Cook 990a51d9eaaSKees Cook audit_log_link_denied("linkat", link); 991800179c9SKees Cook return -EPERM; 992800179c9SKees Cook } 993800179c9SKees Cook 9943b2e7f75SAl Viro static __always_inline 9953b2e7f75SAl Viro const char *get_link(struct nameidata *nd) 9961da177e4SLinus Torvalds { 997ab104923SAl Viro struct saved *last = nd->stack + nd->depth - 1; 9981cf2665bSAl Viro struct dentry *dentry = last->link.dentry; 999fceef393SAl Viro struct inode *inode = nd->link_inode; 10006d7b5aaeSAl Viro int error; 10010a959df5SAl Viro const char *res; 10021da177e4SLinus Torvalds 10038fa9dd24SNeilBrown if (!(nd->flags & LOOKUP_RCU)) { 10048fa9dd24SNeilBrown touch_atime(&last->link); 10058fa9dd24SNeilBrown cond_resched(); 10068fa9dd24SNeilBrown } else if (atime_needs_update(&last->link, inode)) { 10078fa9dd24SNeilBrown if (unlikely(unlazy_walk(nd, NULL, 0))) 10088fa9dd24SNeilBrown return ERR_PTR(-ECHILD); 10098fa9dd24SNeilBrown touch_atime(&last->link); 10108fa9dd24SNeilBrown } 10118fa9dd24SNeilBrown 1012bda0be7aSNeilBrown error = security_inode_follow_link(dentry, inode, 1013bda0be7aSNeilBrown nd->flags & LOOKUP_RCU); 1014bda0be7aSNeilBrown if (unlikely(error)) 10156920a440SAl Viro return ERR_PTR(error); 101636f3b4f6SAl Viro 101786acdca1SAl Viro nd->last_type = LAST_BIND; 1018d4dee48bSAl Viro res = inode->i_link; 1019d4dee48bSAl Viro if (!res) { 1020fceef393SAl Viro const char * (*get)(struct dentry *, struct inode *, 1021fceef393SAl Viro struct delayed_call *); 1022fceef393SAl Viro get = inode->i_op->get_link; 10238c1b4566SAl Viro if (nd->flags & LOOKUP_RCU) { 1024fceef393SAl Viro res = get(NULL, inode, &last->done); 10256b255391SAl Viro if (res == ERR_PTR(-ECHILD)) { 10268c1b4566SAl Viro if (unlikely(unlazy_walk(nd, NULL, 0))) 10278c1b4566SAl Viro return ERR_PTR(-ECHILD); 1028fceef393SAl Viro res = get(dentry, inode, &last->done); 10298c1b4566SAl Viro } 10306b255391SAl Viro } else { 1031fceef393SAl Viro res = get(dentry, inode, &last->done); 10326b255391SAl Viro } 1033fceef393SAl Viro if (IS_ERR_OR_NULL(res)) 1034fab51e8aSAl Viro return res; 10350a959df5SAl Viro } 1036fab51e8aSAl Viro if (*res == '/') { 1037fab51e8aSAl Viro if (!nd->root.mnt) 1038fab51e8aSAl Viro set_root(nd); 1039248fb5b9SAl Viro if (unlikely(nd_jump_root(nd))) 1040fab51e8aSAl Viro return ERR_PTR(-ECHILD); 1041fab51e8aSAl Viro while (unlikely(*++res == '/')) 1042fab51e8aSAl Viro ; 1043fab51e8aSAl Viro } 1044fab51e8aSAl Viro if (!*res) 1045fab51e8aSAl Viro res = NULL; 10460a959df5SAl Viro return res; 10470a959df5SAl Viro } 10486d7b5aaeSAl Viro 1049f015f126SDavid Howells /* 1050f015f126SDavid Howells * follow_up - Find the mountpoint of path's vfsmount 1051f015f126SDavid Howells * 1052f015f126SDavid Howells * Given a path, find the mountpoint of its source file system. 1053f015f126SDavid Howells * Replace @path with the path of the mountpoint in the parent mount. 1054f015f126SDavid Howells * Up is towards /. 1055f015f126SDavid Howells * 1056f015f126SDavid Howells * Return 1 if we went up a level and 0 if we were already at the 1057f015f126SDavid Howells * root. 1058f015f126SDavid Howells */ 1059bab77ebfSAl Viro int follow_up(struct path *path) 10601da177e4SLinus Torvalds { 10610714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 10620714a533SAl Viro struct mount *parent; 10631da177e4SLinus Torvalds struct dentry *mountpoint; 106499b7db7bSNick Piggin 106548a066e7SAl Viro read_seqlock_excl(&mount_lock); 10660714a533SAl Viro parent = mnt->mnt_parent; 10673c0a6163SAl Viro if (parent == mnt) { 106848a066e7SAl Viro read_sequnlock_excl(&mount_lock); 10691da177e4SLinus Torvalds return 0; 10701da177e4SLinus Torvalds } 10710714a533SAl Viro mntget(&parent->mnt); 1072a73324daSAl Viro mountpoint = dget(mnt->mnt_mountpoint); 107348a066e7SAl Viro read_sequnlock_excl(&mount_lock); 1074bab77ebfSAl Viro dput(path->dentry); 1075bab77ebfSAl Viro path->dentry = mountpoint; 1076bab77ebfSAl Viro mntput(path->mnt); 10770714a533SAl Viro path->mnt = &parent->mnt; 10781da177e4SLinus Torvalds return 1; 10791da177e4SLinus Torvalds } 10804d359507SAl Viro EXPORT_SYMBOL(follow_up); 10811da177e4SLinus Torvalds 1082b5c84bf6SNick Piggin /* 10839875cf80SDavid Howells * Perform an automount 10849875cf80SDavid Howells * - return -EISDIR to tell follow_managed() to stop and return the path we 10859875cf80SDavid Howells * were called with. 10861da177e4SLinus Torvalds */ 1087756daf26SNeilBrown static int follow_automount(struct path *path, struct nameidata *nd, 10889875cf80SDavid Howells bool *need_mntput) 108931e6b01fSNick Piggin { 10909875cf80SDavid Howells struct vfsmount *mnt; 1091ea5b778aSDavid Howells int err; 10929875cf80SDavid Howells 10939875cf80SDavid Howells if (!path->dentry->d_op || !path->dentry->d_op->d_automount) 10949875cf80SDavid Howells return -EREMOTE; 10959875cf80SDavid Howells 10960ec26fd0SMiklos Szeredi /* We don't want to mount if someone's just doing a stat - 10970ec26fd0SMiklos Szeredi * unless they're stat'ing a directory and appended a '/' to 10980ec26fd0SMiklos Szeredi * the name. 10990ec26fd0SMiklos Szeredi * 11000ec26fd0SMiklos Szeredi * We do, however, want to mount if someone wants to open or 11015a30d8a2SDavid Howells * create a file of any type under the mountpoint, wants to 11020ec26fd0SMiklos Szeredi * traverse through the mountpoint or wants to open the 11030ec26fd0SMiklos Szeredi * mounted directory. Also, autofs may mark negative dentries 11040ec26fd0SMiklos Szeredi * as being automount points. These will need the attentions 11050ec26fd0SMiklos Szeredi * of the daemon to instantiate them before they can be used. 11065a30d8a2SDavid Howells */ 1107756daf26SNeilBrown if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | 1108d94c177bSLinus Torvalds LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && 11095a30d8a2SDavid Howells path->dentry->d_inode) 11109875cf80SDavid Howells return -EISDIR; 11110ec26fd0SMiklos Szeredi 1112756daf26SNeilBrown nd->total_link_count++; 1113756daf26SNeilBrown if (nd->total_link_count >= 40) 11149875cf80SDavid Howells return -ELOOP; 11159875cf80SDavid Howells 11169875cf80SDavid Howells mnt = path->dentry->d_op->d_automount(path); 11179875cf80SDavid Howells if (IS_ERR(mnt)) { 11189875cf80SDavid Howells /* 11199875cf80SDavid Howells * The filesystem is allowed to return -EISDIR here to indicate 11209875cf80SDavid Howells * it doesn't want to automount. For instance, autofs would do 11219875cf80SDavid Howells * this so that its userspace daemon can mount on this dentry. 11229875cf80SDavid Howells * 11239875cf80SDavid Howells * However, we can only permit this if it's a terminal point in 11249875cf80SDavid Howells * the path being looked up; if it wasn't then the remainder of 11259875cf80SDavid Howells * the path is inaccessible and we should say so. 11269875cf80SDavid Howells */ 1127756daf26SNeilBrown if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT)) 11289875cf80SDavid Howells return -EREMOTE; 11299875cf80SDavid Howells return PTR_ERR(mnt); 113031e6b01fSNick Piggin } 1131ea5b778aSDavid Howells 11329875cf80SDavid Howells if (!mnt) /* mount collision */ 11339875cf80SDavid Howells return 0; 11349875cf80SDavid Howells 11358aef1884SAl Viro if (!*need_mntput) { 11368aef1884SAl Viro /* lock_mount() may release path->mnt on error */ 11378aef1884SAl Viro mntget(path->mnt); 11388aef1884SAl Viro *need_mntput = true; 11398aef1884SAl Viro } 114019a167afSAl Viro err = finish_automount(mnt, path); 1141ea5b778aSDavid Howells 1142ea5b778aSDavid Howells switch (err) { 1143ea5b778aSDavid Howells case -EBUSY: 1144ea5b778aSDavid Howells /* Someone else made a mount here whilst we were busy */ 114519a167afSAl Viro return 0; 1146ea5b778aSDavid Howells case 0: 11478aef1884SAl Viro path_put(path); 11489875cf80SDavid Howells path->mnt = mnt; 11499875cf80SDavid Howells path->dentry = dget(mnt->mnt_root); 11509875cf80SDavid Howells return 0; 115119a167afSAl Viro default: 115219a167afSAl Viro return err; 11539875cf80SDavid Howells } 115419a167afSAl Viro 1155ea5b778aSDavid Howells } 11569875cf80SDavid Howells 11579875cf80SDavid Howells /* 11589875cf80SDavid Howells * Handle a dentry that is managed in some way. 1159cc53ce53SDavid Howells * - Flagged for transit management (autofs) 11609875cf80SDavid Howells * - Flagged as mountpoint 11619875cf80SDavid Howells * - Flagged as automount point 11629875cf80SDavid Howells * 11639875cf80SDavid Howells * This may only be called in refwalk mode. 11649875cf80SDavid Howells * 11659875cf80SDavid Howells * Serialization is taken care of in namespace.c 11669875cf80SDavid Howells */ 1167756daf26SNeilBrown static int follow_managed(struct path *path, struct nameidata *nd) 11689875cf80SDavid Howells { 11698aef1884SAl Viro struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */ 11709875cf80SDavid Howells unsigned managed; 11719875cf80SDavid Howells bool need_mntput = false; 11728aef1884SAl Viro int ret = 0; 11739875cf80SDavid Howells 11749875cf80SDavid Howells /* Given that we're not holding a lock here, we retain the value in a 11759875cf80SDavid Howells * local variable for each dentry as we look at it so that we don't see 11769875cf80SDavid Howells * the components of that value change under us */ 11779875cf80SDavid Howells while (managed = ACCESS_ONCE(path->dentry->d_flags), 11789875cf80SDavid Howells managed &= DCACHE_MANAGED_DENTRY, 11799875cf80SDavid Howells unlikely(managed != 0)) { 1180cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1181cc53ce53SDavid Howells * being held. */ 1182cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 1183cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1184cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 11851aed3e42SAl Viro ret = path->dentry->d_op->d_manage(path->dentry, false); 1186cc53ce53SDavid Howells if (ret < 0) 11878aef1884SAl Viro break; 1188cc53ce53SDavid Howells } 1189cc53ce53SDavid Howells 11909875cf80SDavid Howells /* Transit to a mounted filesystem. */ 11919875cf80SDavid Howells if (managed & DCACHE_MOUNTED) { 11929875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 11939875cf80SDavid Howells if (mounted) { 11949875cf80SDavid Howells dput(path->dentry); 11959875cf80SDavid Howells if (need_mntput) 1196463ffb2eSAl Viro mntput(path->mnt); 1197463ffb2eSAl Viro path->mnt = mounted; 1198463ffb2eSAl Viro path->dentry = dget(mounted->mnt_root); 11999875cf80SDavid Howells need_mntput = true; 12009875cf80SDavid Howells continue; 1201463ffb2eSAl Viro } 1202463ffb2eSAl Viro 12039875cf80SDavid Howells /* Something is mounted on this dentry in another 12049875cf80SDavid Howells * namespace and/or whatever was mounted there in this 120548a066e7SAl Viro * namespace got unmounted before lookup_mnt() could 120648a066e7SAl Viro * get it */ 12071da177e4SLinus Torvalds } 12089875cf80SDavid Howells 12099875cf80SDavid Howells /* Handle an automount point */ 12109875cf80SDavid Howells if (managed & DCACHE_NEED_AUTOMOUNT) { 1211756daf26SNeilBrown ret = follow_automount(path, nd, &need_mntput); 12129875cf80SDavid Howells if (ret < 0) 12138aef1884SAl Viro break; 12149875cf80SDavid Howells continue; 12159875cf80SDavid Howells } 12169875cf80SDavid Howells 12179875cf80SDavid Howells /* We didn't change the current path point */ 12189875cf80SDavid Howells break; 12199875cf80SDavid Howells } 12208aef1884SAl Viro 12218aef1884SAl Viro if (need_mntput && path->mnt == mnt) 12228aef1884SAl Viro mntput(path->mnt); 12238aef1884SAl Viro if (ret == -EISDIR) 12248aef1884SAl Viro ret = 0; 12258402752eSAl Viro if (need_mntput) 12268402752eSAl Viro nd->flags |= LOOKUP_JUMPED; 12278402752eSAl Viro if (unlikely(ret < 0)) 12288402752eSAl Viro path_put_conditional(path, nd); 12298402752eSAl Viro return ret; 12301da177e4SLinus Torvalds } 12311da177e4SLinus Torvalds 1232cc53ce53SDavid Howells int follow_down_one(struct path *path) 12331da177e4SLinus Torvalds { 12341da177e4SLinus Torvalds struct vfsmount *mounted; 12351da177e4SLinus Torvalds 12361c755af4SAl Viro mounted = lookup_mnt(path); 12371da177e4SLinus Torvalds if (mounted) { 12389393bd07SAl Viro dput(path->dentry); 12399393bd07SAl Viro mntput(path->mnt); 12409393bd07SAl Viro path->mnt = mounted; 12419393bd07SAl Viro path->dentry = dget(mounted->mnt_root); 12421da177e4SLinus Torvalds return 1; 12431da177e4SLinus Torvalds } 12441da177e4SLinus Torvalds return 0; 12451da177e4SLinus Torvalds } 12464d359507SAl Viro EXPORT_SYMBOL(follow_down_one); 12471da177e4SLinus Torvalds 1248b8faf035SNeilBrown static inline int managed_dentry_rcu(struct dentry *dentry) 124962a7375eSIan Kent { 1250b8faf035SNeilBrown return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ? 1251b8faf035SNeilBrown dentry->d_op->d_manage(dentry, true) : 0; 125262a7375eSIan Kent } 125362a7375eSIan Kent 12549875cf80SDavid Howells /* 1255287548e4SAl Viro * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if 1256287548e4SAl Viro * we meet a managed dentry that would need blocking. 12579875cf80SDavid Howells */ 12589875cf80SDavid Howells static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, 1259254cf582SAl Viro struct inode **inode, unsigned *seqp) 12609875cf80SDavid Howells { 126162a7375eSIan Kent for (;;) { 1262c7105365SAl Viro struct mount *mounted; 126362a7375eSIan Kent /* 126462a7375eSIan Kent * Don't forget we might have a non-mountpoint managed dentry 126562a7375eSIan Kent * that wants to block transit. 126662a7375eSIan Kent */ 1267b8faf035SNeilBrown switch (managed_dentry_rcu(path->dentry)) { 1268b8faf035SNeilBrown case -ECHILD: 1269b8faf035SNeilBrown default: 1270ab90911fSDavid Howells return false; 1271b8faf035SNeilBrown case -EISDIR: 1272b8faf035SNeilBrown return true; 1273b8faf035SNeilBrown case 0: 1274b8faf035SNeilBrown break; 1275b8faf035SNeilBrown } 127662a7375eSIan Kent 127762a7375eSIan Kent if (!d_mountpoint(path->dentry)) 1278b8faf035SNeilBrown return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); 127962a7375eSIan Kent 1280474279dcSAl Viro mounted = __lookup_mnt(path->mnt, path->dentry); 12819875cf80SDavid Howells if (!mounted) 12829875cf80SDavid Howells break; 1283c7105365SAl Viro path->mnt = &mounted->mnt; 1284c7105365SAl Viro path->dentry = mounted->mnt.mnt_root; 1285a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 1286254cf582SAl Viro *seqp = read_seqcount_begin(&path->dentry->d_seq); 128759430262SLinus Torvalds /* 128859430262SLinus Torvalds * Update the inode too. We don't need to re-check the 128959430262SLinus Torvalds * dentry sequence number here after this d_inode read, 129059430262SLinus Torvalds * because a mount-point is always pinned. 129159430262SLinus Torvalds */ 129259430262SLinus Torvalds *inode = path->dentry->d_inode; 12939875cf80SDavid Howells } 1294f5be3e29SAl Viro return !read_seqretry(&mount_lock, nd->m_seq) && 1295b8faf035SNeilBrown !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); 1296287548e4SAl Viro } 1297287548e4SAl Viro 129831e6b01fSNick Piggin static int follow_dotdot_rcu(struct nameidata *nd) 129931e6b01fSNick Piggin { 13004023bfc9SAl Viro struct inode *inode = nd->inode; 130131e6b01fSNick Piggin 130231e6b01fSNick Piggin while (1) { 1303aed434adSAl Viro if (path_equal(&nd->path, &nd->root)) 130431e6b01fSNick Piggin break; 130531e6b01fSNick Piggin if (nd->path.dentry != nd->path.mnt->mnt_root) { 130631e6b01fSNick Piggin struct dentry *old = nd->path.dentry; 130731e6b01fSNick Piggin struct dentry *parent = old->d_parent; 130831e6b01fSNick Piggin unsigned seq; 130931e6b01fSNick Piggin 13104023bfc9SAl Viro inode = parent->d_inode; 131131e6b01fSNick Piggin seq = read_seqcount_begin(&parent->d_seq); 1312aed434adSAl Viro if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq))) 1313aed434adSAl Viro return -ECHILD; 131431e6b01fSNick Piggin nd->path.dentry = parent; 131531e6b01fSNick Piggin nd->seq = seq; 1316397d425dSEric W. Biederman if (unlikely(!path_connected(&nd->path))) 1317397d425dSEric W. Biederman return -ENOENT; 131831e6b01fSNick Piggin break; 1319aed434adSAl Viro } else { 1320aed434adSAl Viro struct mount *mnt = real_mount(nd->path.mnt); 1321aed434adSAl Viro struct mount *mparent = mnt->mnt_parent; 1322aed434adSAl Viro struct dentry *mountpoint = mnt->mnt_mountpoint; 1323aed434adSAl Viro struct inode *inode2 = mountpoint->d_inode; 1324aed434adSAl Viro unsigned seq = read_seqcount_begin(&mountpoint->d_seq); 1325aed434adSAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1326aed434adSAl Viro return -ECHILD; 1327aed434adSAl Viro if (&mparent->mnt == nd->path.mnt) 132831e6b01fSNick Piggin break; 1329aed434adSAl Viro /* we know that mountpoint was pinned */ 1330aed434adSAl Viro nd->path.dentry = mountpoint; 1331aed434adSAl Viro nd->path.mnt = &mparent->mnt; 1332aed434adSAl Viro inode = inode2; 1333aed434adSAl Viro nd->seq = seq; 133431e6b01fSNick Piggin } 1335aed434adSAl Viro } 1336aed434adSAl Viro while (unlikely(d_mountpoint(nd->path.dentry))) { 1337b37199e6SAl Viro struct mount *mounted; 1338b37199e6SAl Viro mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry); 1339aed434adSAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1340aed434adSAl Viro return -ECHILD; 1341b37199e6SAl Viro if (!mounted) 1342b37199e6SAl Viro break; 1343b37199e6SAl Viro nd->path.mnt = &mounted->mnt; 1344b37199e6SAl Viro nd->path.dentry = mounted->mnt.mnt_root; 13454023bfc9SAl Viro inode = nd->path.dentry->d_inode; 1346b37199e6SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 1347b37199e6SAl Viro } 13484023bfc9SAl Viro nd->inode = inode; 134931e6b01fSNick Piggin return 0; 135031e6b01fSNick Piggin } 135131e6b01fSNick Piggin 13529875cf80SDavid Howells /* 1353cc53ce53SDavid Howells * Follow down to the covering mount currently visible to userspace. At each 1354cc53ce53SDavid Howells * point, the filesystem owning that dentry may be queried as to whether the 1355cc53ce53SDavid Howells * caller is permitted to proceed or not. 1356cc53ce53SDavid Howells */ 13577cc90cc3SAl Viro int follow_down(struct path *path) 1358cc53ce53SDavid Howells { 1359cc53ce53SDavid Howells unsigned managed; 1360cc53ce53SDavid Howells int ret; 1361cc53ce53SDavid Howells 1362cc53ce53SDavid Howells while (managed = ACCESS_ONCE(path->dentry->d_flags), 1363cc53ce53SDavid Howells unlikely(managed & DCACHE_MANAGED_DENTRY)) { 1364cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1365cc53ce53SDavid Howells * being held. 1366cc53ce53SDavid Howells * 1367cc53ce53SDavid Howells * We indicate to the filesystem if someone is trying to mount 1368cc53ce53SDavid Howells * something here. This gives autofs the chance to deny anyone 1369cc53ce53SDavid Howells * other than its daemon the right to mount on its 1370cc53ce53SDavid Howells * superstructure. 1371cc53ce53SDavid Howells * 1372cc53ce53SDavid Howells * The filesystem may sleep at this point. 1373cc53ce53SDavid Howells */ 1374cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 1375cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1376cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 1377ab90911fSDavid Howells ret = path->dentry->d_op->d_manage( 13781aed3e42SAl Viro path->dentry, false); 1379cc53ce53SDavid Howells if (ret < 0) 1380cc53ce53SDavid Howells return ret == -EISDIR ? 0 : ret; 1381cc53ce53SDavid Howells } 1382cc53ce53SDavid Howells 1383cc53ce53SDavid Howells /* Transit to a mounted filesystem. */ 1384cc53ce53SDavid Howells if (managed & DCACHE_MOUNTED) { 1385cc53ce53SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 1386cc53ce53SDavid Howells if (!mounted) 1387cc53ce53SDavid Howells break; 1388cc53ce53SDavid Howells dput(path->dentry); 1389cc53ce53SDavid Howells mntput(path->mnt); 1390cc53ce53SDavid Howells path->mnt = mounted; 1391cc53ce53SDavid Howells path->dentry = dget(mounted->mnt_root); 1392cc53ce53SDavid Howells continue; 1393cc53ce53SDavid Howells } 1394cc53ce53SDavid Howells 1395cc53ce53SDavid Howells /* Don't handle automount points here */ 1396cc53ce53SDavid Howells break; 1397cc53ce53SDavid Howells } 1398cc53ce53SDavid Howells return 0; 1399cc53ce53SDavid Howells } 14004d359507SAl Viro EXPORT_SYMBOL(follow_down); 1401cc53ce53SDavid Howells 1402cc53ce53SDavid Howells /* 14039875cf80SDavid Howells * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() 14049875cf80SDavid Howells */ 14059875cf80SDavid Howells static void follow_mount(struct path *path) 14069875cf80SDavid Howells { 14079875cf80SDavid Howells while (d_mountpoint(path->dentry)) { 14089875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 14099875cf80SDavid Howells if (!mounted) 14109875cf80SDavid Howells break; 14119875cf80SDavid Howells dput(path->dentry); 14129875cf80SDavid Howells mntput(path->mnt); 14139875cf80SDavid Howells path->mnt = mounted; 14149875cf80SDavid Howells path->dentry = dget(mounted->mnt_root); 14159875cf80SDavid Howells } 14169875cf80SDavid Howells } 14179875cf80SDavid Howells 1418397d425dSEric W. Biederman static int follow_dotdot(struct nameidata *nd) 14191da177e4SLinus Torvalds { 14201da177e4SLinus Torvalds while(1) { 14214ac91378SJan Blunck struct dentry *old = nd->path.dentry; 14221da177e4SLinus Torvalds 14232a737871SAl Viro if (nd->path.dentry == nd->root.dentry && 14242a737871SAl Viro nd->path.mnt == nd->root.mnt) { 14251da177e4SLinus Torvalds break; 14261da177e4SLinus Torvalds } 14274ac91378SJan Blunck if (nd->path.dentry != nd->path.mnt->mnt_root) { 14283088dd70SAl Viro /* rare case of legitimate dget_parent()... */ 14293088dd70SAl Viro nd->path.dentry = dget_parent(nd->path.dentry); 14301da177e4SLinus Torvalds dput(old); 1431397d425dSEric W. Biederman if (unlikely(!path_connected(&nd->path))) 1432397d425dSEric W. Biederman return -ENOENT; 14331da177e4SLinus Torvalds break; 14341da177e4SLinus Torvalds } 14353088dd70SAl Viro if (!follow_up(&nd->path)) 14361da177e4SLinus Torvalds break; 14371da177e4SLinus Torvalds } 143879ed0226SAl Viro follow_mount(&nd->path); 143931e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 1440397d425dSEric W. Biederman return 0; 14411da177e4SLinus Torvalds } 14421da177e4SLinus Torvalds 14431da177e4SLinus Torvalds /* 1444bad61189SMiklos Szeredi * This looks up the name in dcache, possibly revalidates the old dentry and 1445bad61189SMiklos Szeredi * allocates a new one if not found or not valid. In the need_lookup argument 1446bad61189SMiklos Szeredi * returns whether i_op->lookup is necessary. 1447bad61189SMiklos Szeredi * 1448bad61189SMiklos Szeredi * dir->d_inode->i_mutex must be held 1449baa03890SNick Piggin */ 1450bad61189SMiklos Szeredi static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir, 1451201f956eSAl Viro unsigned int flags, bool *need_lookup) 1452baa03890SNick Piggin { 1453baa03890SNick Piggin struct dentry *dentry; 1454bad61189SMiklos Szeredi int error; 1455baa03890SNick Piggin 1456bad61189SMiklos Szeredi *need_lookup = false; 1457bad61189SMiklos Szeredi dentry = d_lookup(dir, name); 1458bad61189SMiklos Szeredi if (dentry) { 145939e3c955SJeff Layton if (dentry->d_flags & DCACHE_OP_REVALIDATE) { 1460201f956eSAl Viro error = d_revalidate(dentry, flags); 1461bad61189SMiklos Szeredi if (unlikely(error <= 0)) { 1462bad61189SMiklos Szeredi if (error < 0) { 1463bad61189SMiklos Szeredi dput(dentry); 1464bad61189SMiklos Szeredi return ERR_PTR(error); 14655542aa2fSEric W. Biederman } else { 14665542aa2fSEric W. Biederman d_invalidate(dentry); 1467bad61189SMiklos Szeredi dput(dentry); 1468bad61189SMiklos Szeredi dentry = NULL; 1469bad61189SMiklos Szeredi } 1470bad61189SMiklos Szeredi } 1471bad61189SMiklos Szeredi } 1472bad61189SMiklos Szeredi } 1473baa03890SNick Piggin 1474bad61189SMiklos Szeredi if (!dentry) { 1475bad61189SMiklos Szeredi dentry = d_alloc(dir, name); 1476baa03890SNick Piggin if (unlikely(!dentry)) 1477baa03890SNick Piggin return ERR_PTR(-ENOMEM); 1478baa03890SNick Piggin 1479bad61189SMiklos Szeredi *need_lookup = true; 1480baa03890SNick Piggin } 1481baa03890SNick Piggin return dentry; 1482baa03890SNick Piggin } 1483baa03890SNick Piggin 1484baa03890SNick Piggin /* 148513a2c3beSJ. Bruce Fields * Call i_op->lookup on the dentry. The dentry must be negative and 148613a2c3beSJ. Bruce Fields * unhashed. 1487bad61189SMiklos Szeredi * 1488bad61189SMiklos Szeredi * dir->d_inode->i_mutex must be held 148944396f4bSJosef Bacik */ 1490bad61189SMiklos Szeredi static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry, 149172bd866aSAl Viro unsigned int flags) 149244396f4bSJosef Bacik { 149344396f4bSJosef Bacik struct dentry *old; 149444396f4bSJosef Bacik 149544396f4bSJosef Bacik /* Don't create child dentry for a dead directory. */ 1496bad61189SMiklos Szeredi if (unlikely(IS_DEADDIR(dir))) { 1497e188dc02SMiklos Szeredi dput(dentry); 149844396f4bSJosef Bacik return ERR_PTR(-ENOENT); 1499e188dc02SMiklos Szeredi } 150044396f4bSJosef Bacik 150172bd866aSAl Viro old = dir->i_op->lookup(dir, dentry, flags); 150244396f4bSJosef Bacik if (unlikely(old)) { 150344396f4bSJosef Bacik dput(dentry); 150444396f4bSJosef Bacik dentry = old; 150544396f4bSJosef Bacik } 150644396f4bSJosef Bacik return dentry; 150744396f4bSJosef Bacik } 150844396f4bSJosef Bacik 1509a3255546SAl Viro static struct dentry *__lookup_hash(struct qstr *name, 151072bd866aSAl Viro struct dentry *base, unsigned int flags) 1511a3255546SAl Viro { 1512bad61189SMiklos Szeredi bool need_lookup; 1513a3255546SAl Viro struct dentry *dentry; 1514a3255546SAl Viro 151572bd866aSAl Viro dentry = lookup_dcache(name, base, flags, &need_lookup); 1516bad61189SMiklos Szeredi if (!need_lookup) 1517a3255546SAl Viro return dentry; 1518bad61189SMiklos Szeredi 151972bd866aSAl Viro return lookup_real(base->d_inode, dentry, flags); 1520a3255546SAl Viro } 1521a3255546SAl Viro 152244396f4bSJosef Bacik /* 15231da177e4SLinus Torvalds * It's more convoluted than I'd like it to be, but... it's still fairly 15241da177e4SLinus Torvalds * small and for now I'd prefer to have fast path as straight as possible. 15251da177e4SLinus Torvalds * It _is_ time-critical. 15261da177e4SLinus Torvalds */ 1527e97cdc87SAl Viro static int lookup_fast(struct nameidata *nd, 1528254cf582SAl Viro struct path *path, struct inode **inode, 1529254cf582SAl Viro unsigned *seqp) 15301da177e4SLinus Torvalds { 15314ac91378SJan Blunck struct vfsmount *mnt = nd->path.mnt; 153231e6b01fSNick Piggin struct dentry *dentry, *parent = nd->path.dentry; 15335a18fff2SAl Viro int need_reval = 1; 15345a18fff2SAl Viro int status = 1; 15359875cf80SDavid Howells int err; 15369875cf80SDavid Howells 15373cac260aSAl Viro /* 1538b04f784eSNick Piggin * Rename seqlock is not required here because in the off chance 1539b04f784eSNick Piggin * of a false negative due to a concurrent rename, we're going to 1540b04f784eSNick Piggin * do the non-racy lookup, below. 1541b04f784eSNick Piggin */ 154231e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) { 154331e6b01fSNick Piggin unsigned seq; 1544766c4cbfSAl Viro bool negative; 1545da53be12SLinus Torvalds dentry = __d_lookup_rcu(parent, &nd->last, &seq); 15465a18fff2SAl Viro if (!dentry) 15475a18fff2SAl Viro goto unlazy; 15485a18fff2SAl Viro 154912f8ad4bSLinus Torvalds /* 155012f8ad4bSLinus Torvalds * This sequence count validates that the inode matches 155112f8ad4bSLinus Torvalds * the dentry name information from lookup. 155212f8ad4bSLinus Torvalds */ 155363afdfc7SDavid Howells *inode = d_backing_inode(dentry); 1554766c4cbfSAl Viro negative = d_is_negative(dentry); 155512f8ad4bSLinus Torvalds if (read_seqcount_retry(&dentry->d_seq, seq)) 155612f8ad4bSLinus Torvalds return -ECHILD; 155712f8ad4bSLinus Torvalds 155812f8ad4bSLinus Torvalds /* 155912f8ad4bSLinus Torvalds * This sequence count validates that the parent had no 156012f8ad4bSLinus Torvalds * changes while we did the lookup of the dentry above. 156112f8ad4bSLinus Torvalds * 156212f8ad4bSLinus Torvalds * The memory barrier in read_seqcount_begin of child is 156312f8ad4bSLinus Torvalds * enough, we can use __read_seqcount_retry here. 156412f8ad4bSLinus Torvalds */ 156531e6b01fSNick Piggin if (__read_seqcount_retry(&parent->d_seq, nd->seq)) 156631e6b01fSNick Piggin return -ECHILD; 15675a18fff2SAl Viro 1568254cf582SAl Viro *seqp = seq; 156924643087SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) { 15704ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 15715a18fff2SAl Viro if (unlikely(status <= 0)) { 15725a18fff2SAl Viro if (status != -ECHILD) 15735a18fff2SAl Viro need_reval = 0; 15745a18fff2SAl Viro goto unlazy; 15755a18fff2SAl Viro } 157624643087SAl Viro } 1577daf3761cSTrond Myklebust /* 1578daf3761cSTrond Myklebust * Note: do negative dentry check after revalidation in 1579daf3761cSTrond Myklebust * case that drops it. 1580daf3761cSTrond Myklebust */ 1581daf3761cSTrond Myklebust if (negative) 1582daf3761cSTrond Myklebust return -ENOENT; 158331e6b01fSNick Piggin path->mnt = mnt; 158431e6b01fSNick Piggin path->dentry = dentry; 1585254cf582SAl Viro if (likely(__follow_mount_rcu(nd, path, inode, seqp))) 15869875cf80SDavid Howells return 0; 15875a18fff2SAl Viro unlazy: 1588254cf582SAl Viro if (unlazy_walk(nd, dentry, seq)) 15895a18fff2SAl Viro return -ECHILD; 15905a18fff2SAl Viro } else { 1591e97cdc87SAl Viro dentry = __d_lookup(parent, &nd->last); 159224643087SAl Viro } 15935a18fff2SAl Viro 159481e6f520SAl Viro if (unlikely(!dentry)) 159581e6f520SAl Viro goto need_lookup; 15965a18fff2SAl Viro 15975a18fff2SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval) 15984ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 15995a18fff2SAl Viro if (unlikely(status <= 0)) { 16005a18fff2SAl Viro if (status < 0) { 16015a18fff2SAl Viro dput(dentry); 16025a18fff2SAl Viro return status; 16035a18fff2SAl Viro } 16045542aa2fSEric W. Biederman d_invalidate(dentry); 16055a18fff2SAl Viro dput(dentry); 160681e6f520SAl Viro goto need_lookup; 16075a18fff2SAl Viro } 1608697f514dSMiklos Szeredi 1609766c4cbfSAl Viro if (unlikely(d_is_negative(dentry))) { 1610766c4cbfSAl Viro dput(dentry); 1611766c4cbfSAl Viro return -ENOENT; 1612766c4cbfSAl Viro } 16131da177e4SLinus Torvalds path->mnt = mnt; 16141da177e4SLinus Torvalds path->dentry = dentry; 1615756daf26SNeilBrown err = follow_managed(path, nd); 16168402752eSAl Viro if (likely(!err)) 161763afdfc7SDavid Howells *inode = d_backing_inode(path->dentry); 16188402752eSAl Viro return err; 161981e6f520SAl Viro 162081e6f520SAl Viro need_lookup: 1621697f514dSMiklos Szeredi return 1; 1622697f514dSMiklos Szeredi } 1623697f514dSMiklos Szeredi 1624697f514dSMiklos Szeredi /* Fast lookup failed, do it the slow way */ 1625cc2a5271SAl Viro static int lookup_slow(struct nameidata *nd, struct path *path) 1626697f514dSMiklos Szeredi { 1627697f514dSMiklos Szeredi struct dentry *dentry, *parent; 1628697f514dSMiklos Szeredi 1629697f514dSMiklos Szeredi parent = nd->path.dentry; 163081e6f520SAl Viro BUG_ON(nd->inode != parent->d_inode); 163181e6f520SAl Viro 16325955102cSAl Viro inode_lock(parent->d_inode); 1633cc2a5271SAl Viro dentry = __lookup_hash(&nd->last, parent, nd->flags); 16345955102cSAl Viro inode_unlock(parent->d_inode); 163581e6f520SAl Viro if (IS_ERR(dentry)) 163681e6f520SAl Viro return PTR_ERR(dentry); 1637697f514dSMiklos Szeredi path->mnt = nd->path.mnt; 1638697f514dSMiklos Szeredi path->dentry = dentry; 16398402752eSAl Viro return follow_managed(path, nd); 16401da177e4SLinus Torvalds } 16411da177e4SLinus Torvalds 164252094c8aSAl Viro static inline int may_lookup(struct nameidata *nd) 164352094c8aSAl Viro { 164452094c8aSAl Viro if (nd->flags & LOOKUP_RCU) { 16454ad5abb3SAl Viro int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); 164652094c8aSAl Viro if (err != -ECHILD) 164752094c8aSAl Viro return err; 16486e9918b7SAl Viro if (unlazy_walk(nd, NULL, 0)) 164952094c8aSAl Viro return -ECHILD; 165052094c8aSAl Viro } 16514ad5abb3SAl Viro return inode_permission(nd->inode, MAY_EXEC); 165252094c8aSAl Viro } 165352094c8aSAl Viro 16549856fa1bSAl Viro static inline int handle_dots(struct nameidata *nd, int type) 16559856fa1bSAl Viro { 16569856fa1bSAl Viro if (type == LAST_DOTDOT) { 16579e6697e2SAl Viro if (!nd->root.mnt) 16589e6697e2SAl Viro set_root(nd); 16599856fa1bSAl Viro if (nd->flags & LOOKUP_RCU) { 166070291aecSAl Viro return follow_dotdot_rcu(nd); 16619856fa1bSAl Viro } else 1662397d425dSEric W. Biederman return follow_dotdot(nd); 16639856fa1bSAl Viro } 16649856fa1bSAl Viro return 0; 16659856fa1bSAl Viro } 16669856fa1bSAl Viro 1667181548c0SAl Viro static int pick_link(struct nameidata *nd, struct path *link, 1668181548c0SAl Viro struct inode *inode, unsigned seq) 1669d63ff28fSAl Viro { 1670626de996SAl Viro int error; 16711cf2665bSAl Viro struct saved *last; 1672756daf26SNeilBrown if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) { 1673626de996SAl Viro path_to_nameidata(link, nd); 1674626de996SAl Viro return -ELOOP; 1675626de996SAl Viro } 1676bc40aee0SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1677cd179f44SAl Viro if (link->mnt == nd->path.mnt) 1678cd179f44SAl Viro mntget(link->mnt); 16797973387aSAl Viro } 1680626de996SAl Viro error = nd_alloc_stack(nd); 1681626de996SAl Viro if (unlikely(error)) { 1682bc40aee0SAl Viro if (error == -ECHILD) { 1683bc40aee0SAl Viro if (unlikely(unlazy_link(nd, link, seq))) 1684bc40aee0SAl Viro return -ECHILD; 1685bc40aee0SAl Viro error = nd_alloc_stack(nd); 1686bc40aee0SAl Viro } 1687bc40aee0SAl Viro if (error) { 1688cd179f44SAl Viro path_put(link); 1689626de996SAl Viro return error; 1690626de996SAl Viro } 1691bc40aee0SAl Viro } 1692626de996SAl Viro 1693ab104923SAl Viro last = nd->stack + nd->depth++; 16941cf2665bSAl Viro last->link = *link; 1695fceef393SAl Viro clear_delayed_call(&last->done); 1696fceef393SAl Viro nd->link_inode = inode; 16970450b2d1SAl Viro last->seq = seq; 1698d63ff28fSAl Viro return 1; 1699d63ff28fSAl Viro } 1700d63ff28fSAl Viro 17013ddcd056SLinus Torvalds /* 17023ddcd056SLinus Torvalds * Do we need to follow links? We _really_ want to be able 17033ddcd056SLinus Torvalds * to do this check without having to look at inode->i_op, 17043ddcd056SLinus Torvalds * so we keep a cache of "no, this doesn't need follow_link" 17053ddcd056SLinus Torvalds * for the common case. 17063ddcd056SLinus Torvalds */ 1707254cf582SAl Viro static inline int should_follow_link(struct nameidata *nd, struct path *link, 1708181548c0SAl Viro int follow, 1709181548c0SAl Viro struct inode *inode, unsigned seq) 17103ddcd056SLinus Torvalds { 1711d63ff28fSAl Viro if (likely(!d_is_symlink(link->dentry))) 1712d63ff28fSAl Viro return 0; 1713d63ff28fSAl Viro if (!follow) 1714d63ff28fSAl Viro return 0; 1715181548c0SAl Viro return pick_link(nd, link, inode, seq); 17163ddcd056SLinus Torvalds } 17173ddcd056SLinus Torvalds 17184693a547SAl Viro enum {WALK_GET = 1, WALK_PUT = 2}; 17194693a547SAl Viro 17204693a547SAl Viro static int walk_component(struct nameidata *nd, int flags) 1721ce57dfc1SAl Viro { 1722caa85634SAl Viro struct path path; 1723ce57dfc1SAl Viro struct inode *inode; 1724254cf582SAl Viro unsigned seq; 1725ce57dfc1SAl Viro int err; 1726ce57dfc1SAl Viro /* 1727ce57dfc1SAl Viro * "." and ".." are special - ".." especially so because it has 1728ce57dfc1SAl Viro * to be able to know about the current root directory and 1729ce57dfc1SAl Viro * parent relationships. 1730ce57dfc1SAl Viro */ 17314693a547SAl Viro if (unlikely(nd->last_type != LAST_NORM)) { 17324693a547SAl Viro err = handle_dots(nd, nd->last_type); 17334693a547SAl Viro if (flags & WALK_PUT) 17344693a547SAl Viro put_link(nd); 17354693a547SAl Viro return err; 17364693a547SAl Viro } 1737254cf582SAl Viro err = lookup_fast(nd, &path, &inode, &seq); 1738ce57dfc1SAl Viro if (unlikely(err)) { 1739697f514dSMiklos Szeredi if (err < 0) 1740f0a9ba70SAl Viro return err; 1741697f514dSMiklos Szeredi 1742caa85634SAl Viro err = lookup_slow(nd, &path); 1743697f514dSMiklos Szeredi if (err < 0) 1744f0a9ba70SAl Viro return err; 1745697f514dSMiklos Szeredi 174663afdfc7SDavid Howells inode = d_backing_inode(path.dentry); 1747254cf582SAl Viro seq = 0; /* we are already out of RCU mode */ 1748697f514dSMiklos Szeredi err = -ENOENT; 1749caa85634SAl Viro if (d_is_negative(path.dentry)) 1750697f514dSMiklos Szeredi goto out_path_put; 1751766c4cbfSAl Viro } 1752697f514dSMiklos Szeredi 17534693a547SAl Viro if (flags & WALK_PUT) 17544693a547SAl Viro put_link(nd); 1755181548c0SAl Viro err = should_follow_link(nd, &path, flags & WALK_GET, inode, seq); 1756d63ff28fSAl Viro if (unlikely(err)) 1757d63ff28fSAl Viro return err; 1758caa85634SAl Viro path_to_nameidata(&path, nd); 1759ce57dfc1SAl Viro nd->inode = inode; 1760254cf582SAl Viro nd->seq = seq; 1761ce57dfc1SAl Viro return 0; 1762697f514dSMiklos Szeredi 1763697f514dSMiklos Szeredi out_path_put: 1764caa85634SAl Viro path_to_nameidata(&path, nd); 1765697f514dSMiklos Szeredi return err; 1766ce57dfc1SAl Viro } 1767ce57dfc1SAl Viro 17681da177e4SLinus Torvalds /* 1769bfcfaa77SLinus Torvalds * We can do the critical dentry name comparison and hashing 1770bfcfaa77SLinus Torvalds * operations one word at a time, but we are limited to: 1771bfcfaa77SLinus Torvalds * 1772bfcfaa77SLinus Torvalds * - Architectures with fast unaligned word accesses. We could 1773bfcfaa77SLinus Torvalds * do a "get_unaligned()" if this helps and is sufficiently 1774bfcfaa77SLinus Torvalds * fast. 1775bfcfaa77SLinus Torvalds * 1776bfcfaa77SLinus Torvalds * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we 1777bfcfaa77SLinus Torvalds * do not trap on the (extremely unlikely) case of a page 1778bfcfaa77SLinus Torvalds * crossing operation. 1779bfcfaa77SLinus Torvalds * 1780bfcfaa77SLinus Torvalds * - Furthermore, we need an efficient 64-bit compile for the 1781bfcfaa77SLinus Torvalds * 64-bit case in order to generate the "number of bytes in 1782bfcfaa77SLinus Torvalds * the final mask". Again, that could be replaced with a 1783bfcfaa77SLinus Torvalds * efficient population count instruction or similar. 1784bfcfaa77SLinus Torvalds */ 1785bfcfaa77SLinus Torvalds #ifdef CONFIG_DCACHE_WORD_ACCESS 1786bfcfaa77SLinus Torvalds 1787f68e556eSLinus Torvalds #include <asm/word-at-a-time.h> 1788bfcfaa77SLinus Torvalds 1789f68e556eSLinus Torvalds #ifdef CONFIG_64BIT 1790bfcfaa77SLinus Torvalds 1791bfcfaa77SLinus Torvalds static inline unsigned int fold_hash(unsigned long hash) 1792bfcfaa77SLinus Torvalds { 179399d263d4SLinus Torvalds return hash_64(hash, 32); 1794bfcfaa77SLinus Torvalds } 1795bfcfaa77SLinus Torvalds 1796bfcfaa77SLinus Torvalds #else /* 32-bit case */ 1797bfcfaa77SLinus Torvalds 1798bfcfaa77SLinus Torvalds #define fold_hash(x) (x) 1799bfcfaa77SLinus Torvalds 1800bfcfaa77SLinus Torvalds #endif 1801bfcfaa77SLinus Torvalds 1802bfcfaa77SLinus Torvalds unsigned int full_name_hash(const unsigned char *name, unsigned int len) 1803bfcfaa77SLinus Torvalds { 1804bfcfaa77SLinus Torvalds unsigned long a, mask; 1805bfcfaa77SLinus Torvalds unsigned long hash = 0; 1806bfcfaa77SLinus Torvalds 1807bfcfaa77SLinus Torvalds for (;;) { 1808e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name); 1809bfcfaa77SLinus Torvalds if (len < sizeof(unsigned long)) 1810bfcfaa77SLinus Torvalds break; 1811bfcfaa77SLinus Torvalds hash += a; 1812f132c5beSAl Viro hash *= 9; 1813bfcfaa77SLinus Torvalds name += sizeof(unsigned long); 1814bfcfaa77SLinus Torvalds len -= sizeof(unsigned long); 1815bfcfaa77SLinus Torvalds if (!len) 1816bfcfaa77SLinus Torvalds goto done; 1817bfcfaa77SLinus Torvalds } 1818a5c21dceSWill Deacon mask = bytemask_from_count(len); 1819bfcfaa77SLinus Torvalds hash += mask & a; 1820bfcfaa77SLinus Torvalds done: 1821bfcfaa77SLinus Torvalds return fold_hash(hash); 1822bfcfaa77SLinus Torvalds } 1823bfcfaa77SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 1824bfcfaa77SLinus Torvalds 1825bfcfaa77SLinus Torvalds /* 1826bfcfaa77SLinus Torvalds * Calculate the length and hash of the path component, and 1827d6bb3e90SLinus Torvalds * return the "hash_len" as the result. 1828bfcfaa77SLinus Torvalds */ 1829d6bb3e90SLinus Torvalds static inline u64 hash_name(const char *name) 1830bfcfaa77SLinus Torvalds { 183136126f8fSLinus Torvalds unsigned long a, b, adata, bdata, mask, hash, len; 183236126f8fSLinus Torvalds const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 1833bfcfaa77SLinus Torvalds 1834bfcfaa77SLinus Torvalds hash = a = 0; 1835bfcfaa77SLinus Torvalds len = -sizeof(unsigned long); 1836bfcfaa77SLinus Torvalds do { 1837bfcfaa77SLinus Torvalds hash = (hash + a) * 9; 1838bfcfaa77SLinus Torvalds len += sizeof(unsigned long); 1839e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name+len); 184036126f8fSLinus Torvalds b = a ^ REPEAT_BYTE('/'); 184136126f8fSLinus Torvalds } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); 1842bfcfaa77SLinus Torvalds 184336126f8fSLinus Torvalds adata = prep_zero_mask(a, adata, &constants); 184436126f8fSLinus Torvalds bdata = prep_zero_mask(b, bdata, &constants); 184536126f8fSLinus Torvalds 184636126f8fSLinus Torvalds mask = create_zero_mask(adata | bdata); 184736126f8fSLinus Torvalds 184836126f8fSLinus Torvalds hash += a & zero_bytemask(mask); 18499226b5b4SLinus Torvalds len += find_zero(mask); 1850d6bb3e90SLinus Torvalds return hashlen_create(fold_hash(hash), len); 1851bfcfaa77SLinus Torvalds } 1852bfcfaa77SLinus Torvalds 1853bfcfaa77SLinus Torvalds #else 1854bfcfaa77SLinus Torvalds 18550145acc2SLinus Torvalds unsigned int full_name_hash(const unsigned char *name, unsigned int len) 18560145acc2SLinus Torvalds { 18570145acc2SLinus Torvalds unsigned long hash = init_name_hash(); 18580145acc2SLinus Torvalds while (len--) 18590145acc2SLinus Torvalds hash = partial_name_hash(*name++, hash); 18600145acc2SLinus Torvalds return end_name_hash(hash); 18610145acc2SLinus Torvalds } 1862ae942ae7SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 18630145acc2SLinus Torvalds 18643ddcd056SLinus Torvalds /* 1865200e9ef7SLinus Torvalds * We know there's a real path component here of at least 1866200e9ef7SLinus Torvalds * one character. 1867200e9ef7SLinus Torvalds */ 1868d6bb3e90SLinus Torvalds static inline u64 hash_name(const char *name) 1869200e9ef7SLinus Torvalds { 1870200e9ef7SLinus Torvalds unsigned long hash = init_name_hash(); 1871200e9ef7SLinus Torvalds unsigned long len = 0, c; 1872200e9ef7SLinus Torvalds 1873200e9ef7SLinus Torvalds c = (unsigned char)*name; 1874200e9ef7SLinus Torvalds do { 1875200e9ef7SLinus Torvalds len++; 1876200e9ef7SLinus Torvalds hash = partial_name_hash(c, hash); 1877200e9ef7SLinus Torvalds c = (unsigned char)name[len]; 1878200e9ef7SLinus Torvalds } while (c && c != '/'); 1879d6bb3e90SLinus Torvalds return hashlen_create(end_name_hash(hash), len); 1880200e9ef7SLinus Torvalds } 1881200e9ef7SLinus Torvalds 1882bfcfaa77SLinus Torvalds #endif 1883bfcfaa77SLinus Torvalds 1884200e9ef7SLinus Torvalds /* 18851da177e4SLinus Torvalds * Name resolution. 1886ea3834d9SPrasanna Meda * This is the basic name resolution function, turning a pathname into 1887ea3834d9SPrasanna Meda * the final dentry. We expect 'base' to be positive and a directory. 18881da177e4SLinus Torvalds * 1889ea3834d9SPrasanna Meda * Returns 0 and nd will have valid dentry and mnt on success. 1890ea3834d9SPrasanna Meda * Returns error and drops reference to input namei data on failure. 18911da177e4SLinus Torvalds */ 18926de88d72SAl Viro static int link_path_walk(const char *name, struct nameidata *nd) 18931da177e4SLinus Torvalds { 18941da177e4SLinus Torvalds int err; 18951da177e4SLinus Torvalds 18961da177e4SLinus Torvalds while (*name=='/') 18971da177e4SLinus Torvalds name++; 18981da177e4SLinus Torvalds if (!*name) 18999e18f10aSAl Viro return 0; 19001da177e4SLinus Torvalds 19011da177e4SLinus Torvalds /* At this point we know we have a real path component. */ 19021da177e4SLinus Torvalds for(;;) { 1903d6bb3e90SLinus Torvalds u64 hash_len; 1904fe479a58SAl Viro int type; 19051da177e4SLinus Torvalds 190652094c8aSAl Viro err = may_lookup(nd); 19071da177e4SLinus Torvalds if (err) 19083595e234SAl Viro return err; 19091da177e4SLinus Torvalds 1910d6bb3e90SLinus Torvalds hash_len = hash_name(name); 19111da177e4SLinus Torvalds 1912fe479a58SAl Viro type = LAST_NORM; 1913d6bb3e90SLinus Torvalds if (name[0] == '.') switch (hashlen_len(hash_len)) { 1914fe479a58SAl Viro case 2: 1915200e9ef7SLinus Torvalds if (name[1] == '.') { 1916fe479a58SAl Viro type = LAST_DOTDOT; 191716c2cd71SAl Viro nd->flags |= LOOKUP_JUMPED; 191816c2cd71SAl Viro } 1919fe479a58SAl Viro break; 1920fe479a58SAl Viro case 1: 1921fe479a58SAl Viro type = LAST_DOT; 1922fe479a58SAl Viro } 19235a202bcdSAl Viro if (likely(type == LAST_NORM)) { 19245a202bcdSAl Viro struct dentry *parent = nd->path.dentry; 192516c2cd71SAl Viro nd->flags &= ~LOOKUP_JUMPED; 19265a202bcdSAl Viro if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { 1927a060dc50SJames Hogan struct qstr this = { { .hash_len = hash_len }, .name = name }; 1928da53be12SLinus Torvalds err = parent->d_op->d_hash(parent, &this); 19295a202bcdSAl Viro if (err < 0) 19303595e234SAl Viro return err; 1931d6bb3e90SLinus Torvalds hash_len = this.hash_len; 1932d6bb3e90SLinus Torvalds name = this.name; 19335a202bcdSAl Viro } 19345a202bcdSAl Viro } 1935fe479a58SAl Viro 1936d6bb3e90SLinus Torvalds nd->last.hash_len = hash_len; 1937d6bb3e90SLinus Torvalds nd->last.name = name; 19385f4a6a69SAl Viro nd->last_type = type; 19395f4a6a69SAl Viro 1940d6bb3e90SLinus Torvalds name += hashlen_len(hash_len); 1941d6bb3e90SLinus Torvalds if (!*name) 1942bdf6cbf1SAl Viro goto OK; 1943200e9ef7SLinus Torvalds /* 1944200e9ef7SLinus Torvalds * If it wasn't NUL, we know it was '/'. Skip that 1945200e9ef7SLinus Torvalds * slash, and continue until no more slashes. 1946200e9ef7SLinus Torvalds */ 1947200e9ef7SLinus Torvalds do { 1948d6bb3e90SLinus Torvalds name++; 1949d6bb3e90SLinus Torvalds } while (unlikely(*name == '/')); 19508620c238SAl Viro if (unlikely(!*name)) { 19518620c238SAl Viro OK: 1952368ee9baSAl Viro /* pathname body, done */ 19538620c238SAl Viro if (!nd->depth) 19548620c238SAl Viro return 0; 19558620c238SAl Viro name = nd->stack[nd->depth - 1].name; 1956368ee9baSAl Viro /* trailing symlink, done */ 19578620c238SAl Viro if (!name) 19588620c238SAl Viro return 0; 19598620c238SAl Viro /* last component of nested symlink */ 19604693a547SAl Viro err = walk_component(nd, WALK_GET | WALK_PUT); 19618620c238SAl Viro } else { 19624693a547SAl Viro err = walk_component(nd, WALK_GET); 19638620c238SAl Viro } 1964ce57dfc1SAl Viro if (err < 0) 19653595e234SAl Viro return err; 1966fe479a58SAl Viro 1967ce57dfc1SAl Viro if (err) { 1968626de996SAl Viro const char *s = get_link(nd); 19695a460275SAl Viro 1970a1c83681SViresh Kumar if (IS_ERR(s)) 19713595e234SAl Viro return PTR_ERR(s); 1972172a39a0SAl Viro err = 0; 197312b09578SAl Viro if (unlikely(!s)) { 197412b09578SAl Viro /* jumped */ 1975b9ff4429SAl Viro put_link(nd); 197612b09578SAl Viro } else { 1977071bf501SAl Viro nd->stack[nd->depth - 1].name = name; 197832cd7468SAl Viro name = s; 19799e18f10aSAl Viro continue; 198048c8b0c5SAl Viro } 198131e6b01fSNick Piggin } 198297242f99SAl Viro if (unlikely(!d_can_lookup(nd->path.dentry))) { 198397242f99SAl Viro if (nd->flags & LOOKUP_RCU) { 198497242f99SAl Viro if (unlazy_walk(nd, NULL, 0)) 198597242f99SAl Viro return -ECHILD; 198697242f99SAl Viro } 19873595e234SAl Viro return -ENOTDIR; 19885f4a6a69SAl Viro } 1989ce57dfc1SAl Viro } 199097242f99SAl Viro } 19911da177e4SLinus Torvalds 1992c8a53ee5SAl Viro static const char *path_init(struct nameidata *nd, unsigned flags) 199331e6b01fSNick Piggin { 199431e6b01fSNick Piggin int retval = 0; 1995c8a53ee5SAl Viro const char *s = nd->name->name; 199631e6b01fSNick Piggin 199731e6b01fSNick Piggin nd->last_type = LAST_ROOT; /* if there are only slashes... */ 1998980f3ea2SAl Viro nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT; 199931e6b01fSNick Piggin nd->depth = 0; 20005b6ca027SAl Viro if (flags & LOOKUP_ROOT) { 2001b18825a7SDavid Howells struct dentry *root = nd->root.dentry; 2002b18825a7SDavid Howells struct inode *inode = root->d_inode; 2003fd2f7cb5SAl Viro if (*s) { 200444b1d530SMiklos Szeredi if (!d_can_lookup(root)) 2005368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 20065b6ca027SAl Viro retval = inode_permission(inode, MAY_EXEC); 20075b6ca027SAl Viro if (retval) 2008368ee9baSAl Viro return ERR_PTR(retval); 200973d049a4SAl Viro } 20105b6ca027SAl Viro nd->path = nd->root; 20115b6ca027SAl Viro nd->inode = inode; 20125b6ca027SAl Viro if (flags & LOOKUP_RCU) { 20138b61e74fSAl Viro rcu_read_lock(); 20145b6ca027SAl Viro nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 20158f47a016SAl Viro nd->root_seq = nd->seq; 201648a066e7SAl Viro nd->m_seq = read_seqbegin(&mount_lock); 20175b6ca027SAl Viro } else { 20185b6ca027SAl Viro path_get(&nd->path); 20195b6ca027SAl Viro } 2020368ee9baSAl Viro return s; 20215b6ca027SAl Viro } 20225b6ca027SAl Viro 202331e6b01fSNick Piggin nd->root.mnt = NULL; 2024248fb5b9SAl Viro nd->path.mnt = NULL; 2025248fb5b9SAl Viro nd->path.dentry = NULL; 202631e6b01fSNick Piggin 202748a066e7SAl Viro nd->m_seq = read_seqbegin(&mount_lock); 2028fd2f7cb5SAl Viro if (*s == '/') { 20299e6697e2SAl Viro if (flags & LOOKUP_RCU) 20308b61e74fSAl Viro rcu_read_lock(); 2031e41f7d4eSAl Viro set_root(nd); 2032248fb5b9SAl Viro if (likely(!nd_jump_root(nd))) 2033ef55d917SAl Viro return s; 2034ef55d917SAl Viro nd->root.mnt = NULL; 2035ef55d917SAl Viro rcu_read_unlock(); 2036ef55d917SAl Viro return ERR_PTR(-ECHILD); 2037c8a53ee5SAl Viro } else if (nd->dfd == AT_FDCWD) { 2038e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 203931e6b01fSNick Piggin struct fs_struct *fs = current->fs; 2040c28cc364SNick Piggin unsigned seq; 204131e6b01fSNick Piggin 20428b61e74fSAl Viro rcu_read_lock(); 204331e6b01fSNick Piggin 2044c28cc364SNick Piggin do { 2045c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 204631e6b01fSNick Piggin nd->path = fs->pwd; 2047ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2048c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 2049c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 2050e41f7d4eSAl Viro } else { 2051e41f7d4eSAl Viro get_fs_pwd(current->fs, &nd->path); 2052ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2053e41f7d4eSAl Viro } 2054ef55d917SAl Viro return s; 205531e6b01fSNick Piggin } else { 2056582aa64aSJeff Layton /* Caller must check execute permissions on the starting path component */ 2057c8a53ee5SAl Viro struct fd f = fdget_raw(nd->dfd); 205831e6b01fSNick Piggin struct dentry *dentry; 205931e6b01fSNick Piggin 20602903ff01SAl Viro if (!f.file) 2061368ee9baSAl Viro return ERR_PTR(-EBADF); 206231e6b01fSNick Piggin 20632903ff01SAl Viro dentry = f.file->f_path.dentry; 206431e6b01fSNick Piggin 2065fd2f7cb5SAl Viro if (*s) { 206644b1d530SMiklos Szeredi if (!d_can_lookup(dentry)) { 20672903ff01SAl Viro fdput(f); 2068368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 2069f52e0c11SAl Viro } 20702903ff01SAl Viro } 20712903ff01SAl Viro 20722903ff01SAl Viro nd->path = f.file->f_path; 2073e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 20748b61e74fSAl Viro rcu_read_lock(); 207534a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 207634a26b99SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 20775590ff0dSUlrich Drepper } else { 20782903ff01SAl Viro path_get(&nd->path); 207934a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 20801da177e4SLinus Torvalds } 208134a26b99SAl Viro fdput(f); 2082368ee9baSAl Viro return s; 2083e41f7d4eSAl Viro } 20849b4a9b14SAl Viro } 20859b4a9b14SAl Viro 20863bdba28bSAl Viro static const char *trailing_symlink(struct nameidata *nd) 208795fa25d9SAl Viro { 208895fa25d9SAl Viro const char *s; 2089fec2fa24SAl Viro int error = may_follow_link(nd); 2090deb106c6SAl Viro if (unlikely(error)) 20913bdba28bSAl Viro return ERR_PTR(error); 209295fa25d9SAl Viro nd->flags |= LOOKUP_PARENT; 2093fab51e8aSAl Viro nd->stack[0].name = NULL; 20943b2e7f75SAl Viro s = get_link(nd); 2095deb106c6SAl Viro return s ? s : ""; 209695fa25d9SAl Viro } 209795fa25d9SAl Viro 2098caa85634SAl Viro static inline int lookup_last(struct nameidata *nd) 2099bd92d7feSAl Viro { 2100bd92d7feSAl Viro if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) 2101bd92d7feSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 2102bd92d7feSAl Viro 2103bd92d7feSAl Viro nd->flags &= ~LOOKUP_PARENT; 2104deb106c6SAl Viro return walk_component(nd, 21054693a547SAl Viro nd->flags & LOOKUP_FOLLOW 21064693a547SAl Viro ? nd->depth 21074693a547SAl Viro ? WALK_PUT | WALK_GET 21084693a547SAl Viro : WALK_GET 21094693a547SAl Viro : 0); 2110bd92d7feSAl Viro } 2111bd92d7feSAl Viro 21129b4a9b14SAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2113c8a53ee5SAl Viro static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path) 21149b4a9b14SAl Viro { 2115c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2116bd92d7feSAl Viro int err; 211731e6b01fSNick Piggin 2118368ee9baSAl Viro if (IS_ERR(s)) 2119368ee9baSAl Viro return PTR_ERR(s); 21203bdba28bSAl Viro while (!(err = link_path_walk(s, nd)) 21213bdba28bSAl Viro && ((err = lookup_last(nd)) > 0)) { 21223bdba28bSAl Viro s = trailing_symlink(nd); 21233bdba28bSAl Viro if (IS_ERR(s)) { 21243bdba28bSAl Viro err = PTR_ERR(s); 21256d7b5aaeSAl Viro break; 2126bd92d7feSAl Viro } 2127bd92d7feSAl Viro } 21289f1fafeeSAl Viro if (!err) 21299f1fafeeSAl Viro err = complete_walk(nd); 2130bd92d7feSAl Viro 2131deb106c6SAl Viro if (!err && nd->flags & LOOKUP_DIRECTORY) 2132deb106c6SAl Viro if (!d_can_lookup(nd->path.dentry)) 2133bd23a539SAl Viro err = -ENOTDIR; 2134625b6d10SAl Viro if (!err) { 2135625b6d10SAl Viro *path = nd->path; 2136625b6d10SAl Viro nd->path.mnt = NULL; 2137625b6d10SAl Viro nd->path.dentry = NULL; 2138625b6d10SAl Viro } 2139deb106c6SAl Viro terminate_walk(nd); 2140bd92d7feSAl Viro return err; 214131e6b01fSNick Piggin } 214231e6b01fSNick Piggin 2143625b6d10SAl Viro static int filename_lookup(int dfd, struct filename *name, unsigned flags, 21449ad1aaa6SAl Viro struct path *path, struct path *root) 2145873f1eedSJeff Layton { 2146894bc8c4SAl Viro int retval; 21479883d185SAl Viro struct nameidata nd; 2148abc9f5beSAl Viro if (IS_ERR(name)) 2149abc9f5beSAl Viro return PTR_ERR(name); 21509ad1aaa6SAl Viro if (unlikely(root)) { 21519ad1aaa6SAl Viro nd.root = *root; 21529ad1aaa6SAl Viro flags |= LOOKUP_ROOT; 21539ad1aaa6SAl Viro } 21549883d185SAl Viro set_nameidata(&nd, dfd, name); 2155c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_RCU, path); 2156873f1eedSJeff Layton if (unlikely(retval == -ECHILD)) 2157c8a53ee5SAl Viro retval = path_lookupat(&nd, flags, path); 2158873f1eedSJeff Layton if (unlikely(retval == -ESTALE)) 2159c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path); 2160873f1eedSJeff Layton 2161873f1eedSJeff Layton if (likely(!retval)) 2162625b6d10SAl Viro audit_inode(name, path->dentry, flags & LOOKUP_PARENT); 21639883d185SAl Viro restore_nameidata(); 2164e4bd1c1aSAl Viro putname(name); 2165873f1eedSJeff Layton return retval; 2166873f1eedSJeff Layton } 2167873f1eedSJeff Layton 21688bcb77faSAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2169c8a53ee5SAl Viro static int path_parentat(struct nameidata *nd, unsigned flags, 2170391172c4SAl Viro struct path *parent) 21718bcb77faSAl Viro { 2172c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2173368ee9baSAl Viro int err; 2174368ee9baSAl Viro if (IS_ERR(s)) 2175368ee9baSAl Viro return PTR_ERR(s); 2176368ee9baSAl Viro err = link_path_walk(s, nd); 21778bcb77faSAl Viro if (!err) 21788bcb77faSAl Viro err = complete_walk(nd); 2179391172c4SAl Viro if (!err) { 2180391172c4SAl Viro *parent = nd->path; 2181391172c4SAl Viro nd->path.mnt = NULL; 2182391172c4SAl Viro nd->path.dentry = NULL; 2183391172c4SAl Viro } 2184deb106c6SAl Viro terminate_walk(nd); 21858bcb77faSAl Viro return err; 21868bcb77faSAl Viro } 21878bcb77faSAl Viro 21885c31b6ceSAl Viro static struct filename *filename_parentat(int dfd, struct filename *name, 2189391172c4SAl Viro unsigned int flags, struct path *parent, 2190391172c4SAl Viro struct qstr *last, int *type) 21918bcb77faSAl Viro { 21928bcb77faSAl Viro int retval; 21939883d185SAl Viro struct nameidata nd; 21948bcb77faSAl Viro 21955c31b6ceSAl Viro if (IS_ERR(name)) 21965c31b6ceSAl Viro return name; 21979883d185SAl Viro set_nameidata(&nd, dfd, name); 2198c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_RCU, parent); 21998bcb77faSAl Viro if (unlikely(retval == -ECHILD)) 2200c8a53ee5SAl Viro retval = path_parentat(&nd, flags, parent); 22018bcb77faSAl Viro if (unlikely(retval == -ESTALE)) 2202c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent); 2203391172c4SAl Viro if (likely(!retval)) { 2204391172c4SAl Viro *last = nd.last; 2205391172c4SAl Viro *type = nd.last_type; 2206391172c4SAl Viro audit_inode(name, parent->dentry, LOOKUP_PARENT); 22075c31b6ceSAl Viro } else { 22085c31b6ceSAl Viro putname(name); 22095c31b6ceSAl Viro name = ERR_PTR(retval); 2210391172c4SAl Viro } 22119883d185SAl Viro restore_nameidata(); 22125c31b6ceSAl Viro return name; 22138bcb77faSAl Viro } 22148bcb77faSAl Viro 221579714f72SAl Viro /* does lookup, returns the object with parent locked */ 221679714f72SAl Viro struct dentry *kern_path_locked(const char *name, struct path *path) 22175590ff0dSUlrich Drepper { 22185c31b6ceSAl Viro struct filename *filename; 22195c31b6ceSAl Viro struct dentry *d; 2220391172c4SAl Viro struct qstr last; 2221391172c4SAl Viro int type; 222251689104SPaul Moore 22235c31b6ceSAl Viro filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path, 22245c31b6ceSAl Viro &last, &type); 222551689104SPaul Moore if (IS_ERR(filename)) 222651689104SPaul Moore return ERR_CAST(filename); 22275c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) { 2228391172c4SAl Viro path_put(path); 22295c31b6ceSAl Viro putname(filename); 22305c31b6ceSAl Viro return ERR_PTR(-EINVAL); 223179714f72SAl Viro } 22325955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 2233391172c4SAl Viro d = __lookup_hash(&last, path->dentry, 0); 223479714f72SAl Viro if (IS_ERR(d)) { 22355955102cSAl Viro inode_unlock(path->dentry->d_inode); 2236391172c4SAl Viro path_put(path); 223779714f72SAl Viro } 223851689104SPaul Moore putname(filename); 223979714f72SAl Viro return d; 22405590ff0dSUlrich Drepper } 22415590ff0dSUlrich Drepper 2242d1811465SAl Viro int kern_path(const char *name, unsigned int flags, struct path *path) 2243d1811465SAl Viro { 2244abc9f5beSAl Viro return filename_lookup(AT_FDCWD, getname_kernel(name), 2245abc9f5beSAl Viro flags, path, NULL); 2246d1811465SAl Viro } 22474d359507SAl Viro EXPORT_SYMBOL(kern_path); 2248d1811465SAl Viro 224916f18200SJosef 'Jeff' Sipek /** 225016f18200SJosef 'Jeff' Sipek * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair 225116f18200SJosef 'Jeff' Sipek * @dentry: pointer to dentry of the base directory 225216f18200SJosef 'Jeff' Sipek * @mnt: pointer to vfs mount of the base directory 225316f18200SJosef 'Jeff' Sipek * @name: pointer to file name 225416f18200SJosef 'Jeff' Sipek * @flags: lookup flags 2255e0a01249SAl Viro * @path: pointer to struct path to fill 225616f18200SJosef 'Jeff' Sipek */ 225716f18200SJosef 'Jeff' Sipek int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, 225816f18200SJosef 'Jeff' Sipek const char *name, unsigned int flags, 2259e0a01249SAl Viro struct path *path) 226016f18200SJosef 'Jeff' Sipek { 22619ad1aaa6SAl Viro struct path root = {.mnt = mnt, .dentry = dentry}; 22629ad1aaa6SAl Viro /* the first argument of filename_lookup() is ignored with root */ 2263abc9f5beSAl Viro return filename_lookup(AT_FDCWD, getname_kernel(name), 2264abc9f5beSAl Viro flags , path, &root); 226516f18200SJosef 'Jeff' Sipek } 22664d359507SAl Viro EXPORT_SYMBOL(vfs_path_lookup); 226716f18200SJosef 'Jeff' Sipek 2268eead1911SChristoph Hellwig /** 2269a6b91919SRandy Dunlap * lookup_one_len - filesystem helper to lookup single pathname component 2270eead1911SChristoph Hellwig * @name: pathname component to lookup 2271eead1911SChristoph Hellwig * @base: base directory to lookup from 2272eead1911SChristoph Hellwig * @len: maximum length @len should be interpreted to 2273eead1911SChristoph Hellwig * 2274a6b91919SRandy Dunlap * Note that this routine is purely a helper for filesystem usage and should 22759e7543e9SAl Viro * not be called by generic code. 2276bbddca8eSNeilBrown * 2277bbddca8eSNeilBrown * The caller must hold base->i_mutex. 2278eead1911SChristoph Hellwig */ 2279057f6c01SJames Morris struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 2280057f6c01SJames Morris { 2281057f6c01SJames Morris struct qstr this; 22826a96ba54SAl Viro unsigned int c; 2283cda309deSMiklos Szeredi int err; 2284057f6c01SJames Morris 22855955102cSAl Viro WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 22862f9092e1SDavid Woodhouse 22876a96ba54SAl Viro this.name = name; 22886a96ba54SAl Viro this.len = len; 22890145acc2SLinus Torvalds this.hash = full_name_hash(name, len); 22906a96ba54SAl Viro if (!len) 22916a96ba54SAl Viro return ERR_PTR(-EACCES); 22926a96ba54SAl Viro 229321d8a15aSAl Viro if (unlikely(name[0] == '.')) { 229421d8a15aSAl Viro if (len < 2 || (len == 2 && name[1] == '.')) 229521d8a15aSAl Viro return ERR_PTR(-EACCES); 229621d8a15aSAl Viro } 229721d8a15aSAl Viro 22986a96ba54SAl Viro while (len--) { 22996a96ba54SAl Viro c = *(const unsigned char *)name++; 23006a96ba54SAl Viro if (c == '/' || c == '\0') 23016a96ba54SAl Viro return ERR_PTR(-EACCES); 23026a96ba54SAl Viro } 23035a202bcdSAl Viro /* 23045a202bcdSAl Viro * See if the low-level filesystem might want 23055a202bcdSAl Viro * to use its own hash.. 23065a202bcdSAl Viro */ 23075a202bcdSAl Viro if (base->d_flags & DCACHE_OP_HASH) { 2308da53be12SLinus Torvalds int err = base->d_op->d_hash(base, &this); 23095a202bcdSAl Viro if (err < 0) 23105a202bcdSAl Viro return ERR_PTR(err); 23115a202bcdSAl Viro } 2312eead1911SChristoph Hellwig 2313cda309deSMiklos Szeredi err = inode_permission(base->d_inode, MAY_EXEC); 2314cda309deSMiklos Szeredi if (err) 2315cda309deSMiklos Szeredi return ERR_PTR(err); 2316cda309deSMiklos Szeredi 231772bd866aSAl Viro return __lookup_hash(&this, base, 0); 2318057f6c01SJames Morris } 23194d359507SAl Viro EXPORT_SYMBOL(lookup_one_len); 2320057f6c01SJames Morris 2321bbddca8eSNeilBrown /** 2322bbddca8eSNeilBrown * lookup_one_len_unlocked - filesystem helper to lookup single pathname component 2323bbddca8eSNeilBrown * @name: pathname component to lookup 2324bbddca8eSNeilBrown * @base: base directory to lookup from 2325bbddca8eSNeilBrown * @len: maximum length @len should be interpreted to 2326bbddca8eSNeilBrown * 2327bbddca8eSNeilBrown * Note that this routine is purely a helper for filesystem usage and should 2328bbddca8eSNeilBrown * not be called by generic code. 2329bbddca8eSNeilBrown * 2330bbddca8eSNeilBrown * Unlike lookup_one_len, it should be called without the parent 2331bbddca8eSNeilBrown * i_mutex held, and will take the i_mutex itself if necessary. 2332bbddca8eSNeilBrown */ 2333bbddca8eSNeilBrown struct dentry *lookup_one_len_unlocked(const char *name, 2334bbddca8eSNeilBrown struct dentry *base, int len) 2335bbddca8eSNeilBrown { 2336bbddca8eSNeilBrown struct qstr this; 2337bbddca8eSNeilBrown unsigned int c; 2338bbddca8eSNeilBrown int err; 2339bbddca8eSNeilBrown struct dentry *ret; 2340bbddca8eSNeilBrown 2341bbddca8eSNeilBrown this.name = name; 2342bbddca8eSNeilBrown this.len = len; 2343bbddca8eSNeilBrown this.hash = full_name_hash(name, len); 2344bbddca8eSNeilBrown if (!len) 2345bbddca8eSNeilBrown return ERR_PTR(-EACCES); 2346bbddca8eSNeilBrown 2347bbddca8eSNeilBrown if (unlikely(name[0] == '.')) { 2348bbddca8eSNeilBrown if (len < 2 || (len == 2 && name[1] == '.')) 2349bbddca8eSNeilBrown return ERR_PTR(-EACCES); 2350bbddca8eSNeilBrown } 2351bbddca8eSNeilBrown 2352bbddca8eSNeilBrown while (len--) { 2353bbddca8eSNeilBrown c = *(const unsigned char *)name++; 2354bbddca8eSNeilBrown if (c == '/' || c == '\0') 2355bbddca8eSNeilBrown return ERR_PTR(-EACCES); 2356bbddca8eSNeilBrown } 2357bbddca8eSNeilBrown /* 2358bbddca8eSNeilBrown * See if the low-level filesystem might want 2359bbddca8eSNeilBrown * to use its own hash.. 2360bbddca8eSNeilBrown */ 2361bbddca8eSNeilBrown if (base->d_flags & DCACHE_OP_HASH) { 2362bbddca8eSNeilBrown int err = base->d_op->d_hash(base, &this); 2363bbddca8eSNeilBrown if (err < 0) 2364bbddca8eSNeilBrown return ERR_PTR(err); 2365bbddca8eSNeilBrown } 2366bbddca8eSNeilBrown 2367bbddca8eSNeilBrown err = inode_permission(base->d_inode, MAY_EXEC); 2368bbddca8eSNeilBrown if (err) 2369bbddca8eSNeilBrown return ERR_PTR(err); 2370bbddca8eSNeilBrown 2371bbddca8eSNeilBrown /* 2372bbddca8eSNeilBrown * __d_lookup() is used to try to get a quick answer and avoid the 2373bbddca8eSNeilBrown * mutex. A false-negative does no harm. 2374bbddca8eSNeilBrown */ 2375bbddca8eSNeilBrown ret = __d_lookup(base, &this); 2376bbddca8eSNeilBrown if (ret && unlikely(ret->d_flags & DCACHE_OP_REVALIDATE)) { 2377bbddca8eSNeilBrown dput(ret); 2378bbddca8eSNeilBrown ret = NULL; 2379bbddca8eSNeilBrown } 2380bbddca8eSNeilBrown if (ret) 2381bbddca8eSNeilBrown return ret; 2382bbddca8eSNeilBrown 23835955102cSAl Viro inode_lock(base->d_inode); 2384bbddca8eSNeilBrown ret = __lookup_hash(&this, base, 0); 23855955102cSAl Viro inode_unlock(base->d_inode); 2386bbddca8eSNeilBrown return ret; 2387bbddca8eSNeilBrown } 2388bbddca8eSNeilBrown EXPORT_SYMBOL(lookup_one_len_unlocked); 2389bbddca8eSNeilBrown 23901fa1e7f6SAndy Whitcroft int user_path_at_empty(int dfd, const char __user *name, unsigned flags, 23911fa1e7f6SAndy Whitcroft struct path *path, int *empty) 23921da177e4SLinus Torvalds { 2393abc9f5beSAl Viro return filename_lookup(dfd, getname_flags(name, flags, empty), 2394abc9f5beSAl Viro flags, path, NULL); 23951da177e4SLinus Torvalds } 2396b853a161SAl Viro EXPORT_SYMBOL(user_path_at_empty); 23971fa1e7f6SAndy Whitcroft 2398873f1eedSJeff Layton /* 2399873f1eedSJeff Layton * NB: most callers don't do anything directly with the reference to the 2400873f1eedSJeff Layton * to struct filename, but the nd->last pointer points into the name string 2401873f1eedSJeff Layton * allocated by getname. So we must hold the reference to it until all 2402873f1eedSJeff Layton * path-walking is complete. 2403873f1eedSJeff Layton */ 2404a2ec4a2dSAl Viro static inline struct filename * 2405f5beed75SAl Viro user_path_parent(int dfd, const char __user *path, 2406f5beed75SAl Viro struct path *parent, 2407f5beed75SAl Viro struct qstr *last, 2408f5beed75SAl Viro int *type, 24099e790bd6SJeff Layton unsigned int flags) 24102ad94ae6SAl Viro { 24119e790bd6SJeff Layton /* only LOOKUP_REVAL is allowed in extra flags */ 24125c31b6ceSAl Viro return filename_parentat(dfd, getname(path), flags & LOOKUP_REVAL, 24135c31b6ceSAl Viro parent, last, type); 24142ad94ae6SAl Viro } 24152ad94ae6SAl Viro 24168033426eSJeff Layton /** 2417197df04cSAl Viro * mountpoint_last - look up last component for umount 24188033426eSJeff Layton * @nd: pathwalk nameidata - currently pointing at parent directory of "last" 24198033426eSJeff Layton * @path: pointer to container for result 24208033426eSJeff Layton * 24218033426eSJeff Layton * This is a special lookup_last function just for umount. In this case, we 24228033426eSJeff Layton * need to resolve the path without doing any revalidation. 24238033426eSJeff Layton * 24248033426eSJeff Layton * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since 24258033426eSJeff Layton * mountpoints are always pinned in the dcache, their ancestors are too. Thus, 24268033426eSJeff Layton * in almost all cases, this lookup will be served out of the dcache. The only 24278033426eSJeff Layton * cases where it won't are if nd->last refers to a symlink or the path is 24288033426eSJeff Layton * bogus and it doesn't exist. 24298033426eSJeff Layton * 24308033426eSJeff Layton * Returns: 24318033426eSJeff Layton * -error: if there was an error during lookup. This includes -ENOENT if the 24328033426eSJeff Layton * lookup found a negative dentry. The nd->path reference will also be 24338033426eSJeff Layton * put in this case. 24348033426eSJeff Layton * 24358033426eSJeff Layton * 0: if we successfully resolved nd->path and found it to not to be a 24368033426eSJeff Layton * symlink that needs to be followed. "path" will also be populated. 24378033426eSJeff Layton * The nd->path reference will also be put. 24388033426eSJeff Layton * 24398033426eSJeff Layton * 1: if we successfully resolved nd->last and found it to be a symlink 24408033426eSJeff Layton * that needs to be followed. "path" will be populated with the path 24418033426eSJeff Layton * to the link, and nd->path will *not* be put. 24428033426eSJeff Layton */ 24438033426eSJeff Layton static int 2444197df04cSAl Viro mountpoint_last(struct nameidata *nd, struct path *path) 24458033426eSJeff Layton { 24468033426eSJeff Layton int error = 0; 24478033426eSJeff Layton struct dentry *dentry; 24488033426eSJeff Layton struct dentry *dir = nd->path.dentry; 24498033426eSJeff Layton 245035759521SAl Viro /* If we're in rcuwalk, drop out of it to handle last component */ 245135759521SAl Viro if (nd->flags & LOOKUP_RCU) { 24526e9918b7SAl Viro if (unlazy_walk(nd, NULL, 0)) 2453deb106c6SAl Viro return -ECHILD; 24548033426eSJeff Layton } 24558033426eSJeff Layton 24568033426eSJeff Layton nd->flags &= ~LOOKUP_PARENT; 24578033426eSJeff Layton 24588033426eSJeff Layton if (unlikely(nd->last_type != LAST_NORM)) { 24598033426eSJeff Layton error = handle_dots(nd, nd->last_type); 246035759521SAl Viro if (error) 2461deb106c6SAl Viro return error; 24628033426eSJeff Layton dentry = dget(nd->path.dentry); 246335759521SAl Viro goto done; 24648033426eSJeff Layton } 24658033426eSJeff Layton 24665955102cSAl Viro inode_lock(dir->d_inode); 24678033426eSJeff Layton dentry = d_lookup(dir, &nd->last); 24688033426eSJeff Layton if (!dentry) { 24698033426eSJeff Layton /* 24708033426eSJeff Layton * No cached dentry. Mounted dentries are pinned in the cache, 24718033426eSJeff Layton * so that means that this dentry is probably a symlink or the 24728033426eSJeff Layton * path doesn't actually point to a mounted dentry. 24738033426eSJeff Layton */ 24748033426eSJeff Layton dentry = d_alloc(dir, &nd->last); 24758033426eSJeff Layton if (!dentry) { 24765955102cSAl Viro inode_unlock(dir->d_inode); 2477deb106c6SAl Viro return -ENOMEM; 24788033426eSJeff Layton } 247935759521SAl Viro dentry = lookup_real(dir->d_inode, dentry, nd->flags); 2480bcceeebaSDave Jones if (IS_ERR(dentry)) { 24815955102cSAl Viro inode_unlock(dir->d_inode); 2482deb106c6SAl Viro return PTR_ERR(dentry); 24838033426eSJeff Layton } 2484bcceeebaSDave Jones } 24855955102cSAl Viro inode_unlock(dir->d_inode); 24868033426eSJeff Layton 248735759521SAl Viro done: 2488698934dfSDavid Howells if (d_is_negative(dentry)) { 24898033426eSJeff Layton dput(dentry); 2490deb106c6SAl Viro return -ENOENT; 249135759521SAl Viro } 2492191d7f73SAl Viro if (nd->depth) 2493191d7f73SAl Viro put_link(nd); 24948033426eSJeff Layton path->dentry = dentry; 2495295dc39dSVasily Averin path->mnt = nd->path.mnt; 2496181548c0SAl Viro error = should_follow_link(nd, path, nd->flags & LOOKUP_FOLLOW, 2497181548c0SAl Viro d_backing_inode(dentry), 0); 2498deb106c6SAl Viro if (unlikely(error)) 2499d63ff28fSAl Viro return error; 2500295dc39dSVasily Averin mntget(path->mnt); 25018033426eSJeff Layton follow_mount(path); 2502deb106c6SAl Viro return 0; 25038033426eSJeff Layton } 25048033426eSJeff Layton 25058033426eSJeff Layton /** 2506197df04cSAl Viro * path_mountpoint - look up a path to be umounted 25072a78b857SMasanari Iida * @nd: lookup context 25088033426eSJeff Layton * @flags: lookup flags 2509c8a53ee5SAl Viro * @path: pointer to container for result 25108033426eSJeff Layton * 25118033426eSJeff Layton * Look up the given name, but don't attempt to revalidate the last component. 2512606d6fe3SRandy Dunlap * Returns 0 and "path" will be valid on success; Returns error otherwise. 25138033426eSJeff Layton */ 25148033426eSJeff Layton static int 2515c8a53ee5SAl Viro path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path) 25168033426eSJeff Layton { 2517c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2518368ee9baSAl Viro int err; 2519368ee9baSAl Viro if (IS_ERR(s)) 2520368ee9baSAl Viro return PTR_ERR(s); 25213bdba28bSAl Viro while (!(err = link_path_walk(s, nd)) && 25223bdba28bSAl Viro (err = mountpoint_last(nd, path)) > 0) { 25233bdba28bSAl Viro s = trailing_symlink(nd); 25243bdba28bSAl Viro if (IS_ERR(s)) { 25253bdba28bSAl Viro err = PTR_ERR(s); 25268033426eSJeff Layton break; 25278033426eSJeff Layton } 25283bdba28bSAl Viro } 2529deb106c6SAl Viro terminate_walk(nd); 25308033426eSJeff Layton return err; 25318033426eSJeff Layton } 25328033426eSJeff Layton 25332d864651SAl Viro static int 2534668696dcSAl Viro filename_mountpoint(int dfd, struct filename *name, struct path *path, 25352d864651SAl Viro unsigned int flags) 25362d864651SAl Viro { 25379883d185SAl Viro struct nameidata nd; 2538cbaab2dbSAl Viro int error; 2539668696dcSAl Viro if (IS_ERR(name)) 2540668696dcSAl Viro return PTR_ERR(name); 25419883d185SAl Viro set_nameidata(&nd, dfd, name); 2542c8a53ee5SAl Viro error = path_mountpoint(&nd, flags | LOOKUP_RCU, path); 25432d864651SAl Viro if (unlikely(error == -ECHILD)) 2544c8a53ee5SAl Viro error = path_mountpoint(&nd, flags, path); 25452d864651SAl Viro if (unlikely(error == -ESTALE)) 2546c8a53ee5SAl Viro error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path); 25472d864651SAl Viro if (likely(!error)) 2548668696dcSAl Viro audit_inode(name, path->dentry, 0); 25499883d185SAl Viro restore_nameidata(); 2550668696dcSAl Viro putname(name); 25512d864651SAl Viro return error; 25522d864651SAl Viro } 25532d864651SAl Viro 25548033426eSJeff Layton /** 2555197df04cSAl Viro * user_path_mountpoint_at - lookup a path from userland in order to umount it 25568033426eSJeff Layton * @dfd: directory file descriptor 25578033426eSJeff Layton * @name: pathname from userland 25588033426eSJeff Layton * @flags: lookup flags 25598033426eSJeff Layton * @path: pointer to container to hold result 25608033426eSJeff Layton * 25618033426eSJeff Layton * A umount is a special case for path walking. We're not actually interested 25628033426eSJeff Layton * in the inode in this situation, and ESTALE errors can be a problem. We 25638033426eSJeff Layton * simply want track down the dentry and vfsmount attached at the mountpoint 25648033426eSJeff Layton * and avoid revalidating the last component. 25658033426eSJeff Layton * 25668033426eSJeff Layton * Returns 0 and populates "path" on success. 25678033426eSJeff Layton */ 25688033426eSJeff Layton int 2569197df04cSAl Viro user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags, 25708033426eSJeff Layton struct path *path) 25718033426eSJeff Layton { 2572cbaab2dbSAl Viro return filename_mountpoint(dfd, getname(name), path, flags); 25738033426eSJeff Layton } 25748033426eSJeff Layton 25752d864651SAl Viro int 25762d864651SAl Viro kern_path_mountpoint(int dfd, const char *name, struct path *path, 25772d864651SAl Viro unsigned int flags) 25782d864651SAl Viro { 2579cbaab2dbSAl Viro return filename_mountpoint(dfd, getname_kernel(name), path, flags); 25802d864651SAl Viro } 25812d864651SAl Viro EXPORT_SYMBOL(kern_path_mountpoint); 25822d864651SAl Viro 2583cbdf35bcSMiklos Szeredi int __check_sticky(struct inode *dir, struct inode *inode) 25841da177e4SLinus Torvalds { 25858e96e3b7SEric W. Biederman kuid_t fsuid = current_fsuid(); 2586da9592edSDavid Howells 25878e96e3b7SEric W. Biederman if (uid_eq(inode->i_uid, fsuid)) 25881da177e4SLinus Torvalds return 0; 25898e96e3b7SEric W. Biederman if (uid_eq(dir->i_uid, fsuid)) 25901da177e4SLinus Torvalds return 0; 259123adbe12SAndy Lutomirski return !capable_wrt_inode_uidgid(inode, CAP_FOWNER); 25921da177e4SLinus Torvalds } 2593cbdf35bcSMiklos Szeredi EXPORT_SYMBOL(__check_sticky); 25941da177e4SLinus Torvalds 25951da177e4SLinus Torvalds /* 25961da177e4SLinus Torvalds * Check whether we can remove a link victim from directory dir, check 25971da177e4SLinus Torvalds * whether the type of victim is right. 25981da177e4SLinus Torvalds * 1. We can't do it if dir is read-only (done in permission()) 25991da177e4SLinus Torvalds * 2. We should have write and exec permissions on dir 26001da177e4SLinus Torvalds * 3. We can't remove anything from append-only dir 26011da177e4SLinus Torvalds * 4. We can't do anything with immutable dir (done in permission()) 26021da177e4SLinus Torvalds * 5. If the sticky bit on dir is set we should either 26031da177e4SLinus Torvalds * a. be owner of dir, or 26041da177e4SLinus Torvalds * b. be owner of victim, or 26051da177e4SLinus Torvalds * c. have CAP_FOWNER capability 26061da177e4SLinus Torvalds * 6. If the victim is append-only or immutable we can't do antyhing with 26071da177e4SLinus Torvalds * links pointing to it. 26081da177e4SLinus Torvalds * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. 26091da177e4SLinus Torvalds * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. 26101da177e4SLinus Torvalds * 9. We can't remove a root or mountpoint. 26111da177e4SLinus Torvalds * 10. We don't allow removal of NFS sillyrenamed files; it's handled by 26121da177e4SLinus Torvalds * nfs_async_unlink(). 26131da177e4SLinus Torvalds */ 2614b18825a7SDavid Howells static int may_delete(struct inode *dir, struct dentry *victim, bool isdir) 26151da177e4SLinus Torvalds { 261663afdfc7SDavid Howells struct inode *inode = d_backing_inode(victim); 26171da177e4SLinus Torvalds int error; 26181da177e4SLinus Torvalds 2619b18825a7SDavid Howells if (d_is_negative(victim)) 26201da177e4SLinus Torvalds return -ENOENT; 2621b18825a7SDavid Howells BUG_ON(!inode); 26221da177e4SLinus Torvalds 26231da177e4SLinus Torvalds BUG_ON(victim->d_parent->d_inode != dir); 26244fa6b5ecSJeff Layton audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); 26251da177e4SLinus Torvalds 2626f419a2e3SAl Viro error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 26271da177e4SLinus Torvalds if (error) 26281da177e4SLinus Torvalds return error; 26291da177e4SLinus Torvalds if (IS_APPEND(dir)) 26301da177e4SLinus Torvalds return -EPERM; 2631b18825a7SDavid Howells 2632b18825a7SDavid Howells if (check_sticky(dir, inode) || IS_APPEND(inode) || 2633b18825a7SDavid Howells IS_IMMUTABLE(inode) || IS_SWAPFILE(inode)) 26341da177e4SLinus Torvalds return -EPERM; 26351da177e4SLinus Torvalds if (isdir) { 263644b1d530SMiklos Szeredi if (!d_is_dir(victim)) 26371da177e4SLinus Torvalds return -ENOTDIR; 26381da177e4SLinus Torvalds if (IS_ROOT(victim)) 26391da177e4SLinus Torvalds return -EBUSY; 264044b1d530SMiklos Szeredi } else if (d_is_dir(victim)) 26411da177e4SLinus Torvalds return -EISDIR; 26421da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 26431da177e4SLinus Torvalds return -ENOENT; 26441da177e4SLinus Torvalds if (victim->d_flags & DCACHE_NFSFS_RENAMED) 26451da177e4SLinus Torvalds return -EBUSY; 26461da177e4SLinus Torvalds return 0; 26471da177e4SLinus Torvalds } 26481da177e4SLinus Torvalds 26491da177e4SLinus Torvalds /* Check whether we can create an object with dentry child in directory 26501da177e4SLinus Torvalds * dir. 26511da177e4SLinus Torvalds * 1. We can't do it if child already exists (open has special treatment for 26521da177e4SLinus Torvalds * this case, but since we are inlined it's OK) 26531da177e4SLinus Torvalds * 2. We can't do it if dir is read-only (done in permission()) 26541da177e4SLinus Torvalds * 3. We should have write and exec permissions on dir 26551da177e4SLinus Torvalds * 4. We can't do it if dir is immutable (done in permission()) 26561da177e4SLinus Torvalds */ 2657a95164d9SMiklos Szeredi static inline int may_create(struct inode *dir, struct dentry *child) 26581da177e4SLinus Torvalds { 265914e972b4SJeff Layton audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); 26601da177e4SLinus Torvalds if (child->d_inode) 26611da177e4SLinus Torvalds return -EEXIST; 26621da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 26631da177e4SLinus Torvalds return -ENOENT; 2664f419a2e3SAl Viro return inode_permission(dir, MAY_WRITE | MAY_EXEC); 26651da177e4SLinus Torvalds } 26661da177e4SLinus Torvalds 26671da177e4SLinus Torvalds /* 26681da177e4SLinus Torvalds * p1 and p2 should be directories on the same fs. 26691da177e4SLinus Torvalds */ 26701da177e4SLinus Torvalds struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 26711da177e4SLinus Torvalds { 26721da177e4SLinus Torvalds struct dentry *p; 26731da177e4SLinus Torvalds 26741da177e4SLinus Torvalds if (p1 == p2) { 26755955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 26761da177e4SLinus Torvalds return NULL; 26771da177e4SLinus Torvalds } 26781da177e4SLinus Torvalds 2679a11f3a05SArjan van de Ven mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 26801da177e4SLinus Torvalds 2681e2761a11SOGAWA Hirofumi p = d_ancestor(p2, p1); 2682e2761a11SOGAWA Hirofumi if (p) { 26835955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT); 26845955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_CHILD); 26851da177e4SLinus Torvalds return p; 26861da177e4SLinus Torvalds } 26871da177e4SLinus Torvalds 2688e2761a11SOGAWA Hirofumi p = d_ancestor(p1, p2); 2689e2761a11SOGAWA Hirofumi if (p) { 26905955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 26915955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_CHILD); 26921da177e4SLinus Torvalds return p; 26931da177e4SLinus Torvalds } 26941da177e4SLinus Torvalds 26955955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 26965955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2); 26971da177e4SLinus Torvalds return NULL; 26981da177e4SLinus Torvalds } 26994d359507SAl Viro EXPORT_SYMBOL(lock_rename); 27001da177e4SLinus Torvalds 27011da177e4SLinus Torvalds void unlock_rename(struct dentry *p1, struct dentry *p2) 27021da177e4SLinus Torvalds { 27035955102cSAl Viro inode_unlock(p1->d_inode); 27041da177e4SLinus Torvalds if (p1 != p2) { 27055955102cSAl Viro inode_unlock(p2->d_inode); 2706a11f3a05SArjan van de Ven mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 27071da177e4SLinus Torvalds } 27081da177e4SLinus Torvalds } 27094d359507SAl Viro EXPORT_SYMBOL(unlock_rename); 27101da177e4SLinus Torvalds 27114acdaf27SAl Viro int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, 2712312b63fbSAl Viro bool want_excl) 27131da177e4SLinus Torvalds { 2714a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 27151da177e4SLinus Torvalds if (error) 27161da177e4SLinus Torvalds return error; 27171da177e4SLinus Torvalds 2718acfa4380SAl Viro if (!dir->i_op->create) 27191da177e4SLinus Torvalds return -EACCES; /* shouldn't it be ENOSYS? */ 27201da177e4SLinus Torvalds mode &= S_IALLUGO; 27211da177e4SLinus Torvalds mode |= S_IFREG; 27221da177e4SLinus Torvalds error = security_inode_create(dir, dentry, mode); 27231da177e4SLinus Torvalds if (error) 27241da177e4SLinus Torvalds return error; 2725312b63fbSAl Viro error = dir->i_op->create(dir, dentry, mode, want_excl); 2726a74574aaSStephen Smalley if (!error) 2727f38aa942SAmy Griffis fsnotify_create(dir, dentry); 27281da177e4SLinus Torvalds return error; 27291da177e4SLinus Torvalds } 27304d359507SAl Viro EXPORT_SYMBOL(vfs_create); 27311da177e4SLinus Torvalds 273273d049a4SAl Viro static int may_open(struct path *path, int acc_mode, int flag) 27331da177e4SLinus Torvalds { 27343fb64190SChristoph Hellwig struct dentry *dentry = path->dentry; 27351da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 27361da177e4SLinus Torvalds int error; 27371da177e4SLinus Torvalds 27381da177e4SLinus Torvalds if (!inode) 27391da177e4SLinus Torvalds return -ENOENT; 27401da177e4SLinus Torvalds 2741c8fe8f30SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 2742c8fe8f30SChristoph Hellwig case S_IFLNK: 27431da177e4SLinus Torvalds return -ELOOP; 2744c8fe8f30SChristoph Hellwig case S_IFDIR: 2745c8fe8f30SChristoph Hellwig if (acc_mode & MAY_WRITE) 27461da177e4SLinus Torvalds return -EISDIR; 2747c8fe8f30SChristoph Hellwig break; 2748c8fe8f30SChristoph Hellwig case S_IFBLK: 2749c8fe8f30SChristoph Hellwig case S_IFCHR: 27503fb64190SChristoph Hellwig if (path->mnt->mnt_flags & MNT_NODEV) 27511da177e4SLinus Torvalds return -EACCES; 2752c8fe8f30SChristoph Hellwig /*FALLTHRU*/ 2753c8fe8f30SChristoph Hellwig case S_IFIFO: 2754c8fe8f30SChristoph Hellwig case S_IFSOCK: 27551da177e4SLinus Torvalds flag &= ~O_TRUNC; 2756c8fe8f30SChristoph Hellwig break; 27574a3fd211SDave Hansen } 2758b41572e9SDave Hansen 275962fb4a15SAl Viro error = inode_permission(inode, MAY_OPEN | acc_mode); 2760b41572e9SDave Hansen if (error) 2761b41572e9SDave Hansen return error; 27626146f0d5SMimi Zohar 27631da177e4SLinus Torvalds /* 27641da177e4SLinus Torvalds * An append-only file must be opened in append mode for writing. 27651da177e4SLinus Torvalds */ 27661da177e4SLinus Torvalds if (IS_APPEND(inode)) { 27678737c930SAl Viro if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) 27687715b521SAl Viro return -EPERM; 27691da177e4SLinus Torvalds if (flag & O_TRUNC) 27707715b521SAl Viro return -EPERM; 27711da177e4SLinus Torvalds } 27721da177e4SLinus Torvalds 27731da177e4SLinus Torvalds /* O_NOATIME can only be set by the owner or superuser */ 27742e149670SSerge E. Hallyn if (flag & O_NOATIME && !inode_owner_or_capable(inode)) 27757715b521SAl Viro return -EPERM; 27761da177e4SLinus Torvalds 2777f3c7691eSJ. Bruce Fields return 0; 27787715b521SAl Viro } 27797715b521SAl Viro 2780e1181ee6SJeff Layton static int handle_truncate(struct file *filp) 27817715b521SAl Viro { 2782e1181ee6SJeff Layton struct path *path = &filp->f_path; 27837715b521SAl Viro struct inode *inode = path->dentry->d_inode; 27847715b521SAl Viro int error = get_write_access(inode); 27851da177e4SLinus Torvalds if (error) 27867715b521SAl Viro return error; 27871da177e4SLinus Torvalds /* 27881da177e4SLinus Torvalds * Refuse to truncate files with mandatory locks held on them. 27891da177e4SLinus Torvalds */ 2790d7a06983SJeff Layton error = locks_verify_locked(filp); 2791be6d3e56SKentaro Takeda if (!error) 2792ea0d3ab2STetsuo Handa error = security_path_truncate(path); 27931da177e4SLinus Torvalds if (!error) { 27947715b521SAl Viro error = do_truncate(path->dentry, 0, 2795d139d7ffSMiklos Szeredi ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, 2796e1181ee6SJeff Layton filp); 27971da177e4SLinus Torvalds } 27981da177e4SLinus Torvalds put_write_access(inode); 2799acd0c935SMimi Zohar return error; 28001da177e4SLinus Torvalds } 28011da177e4SLinus Torvalds 2802d57999e1SDave Hansen static inline int open_to_namei_flags(int flag) 2803d57999e1SDave Hansen { 28048a5e929dSAl Viro if ((flag & O_ACCMODE) == 3) 28058a5e929dSAl Viro flag--; 2806d57999e1SDave Hansen return flag; 2807d57999e1SDave Hansen } 2808d57999e1SDave Hansen 2809d18e9008SMiklos Szeredi static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode) 2810d18e9008SMiklos Szeredi { 2811d18e9008SMiklos Szeredi int error = security_path_mknod(dir, dentry, mode, 0); 2812d18e9008SMiklos Szeredi if (error) 2813d18e9008SMiklos Szeredi return error; 2814d18e9008SMiklos Szeredi 2815d18e9008SMiklos Szeredi error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC); 2816d18e9008SMiklos Szeredi if (error) 2817d18e9008SMiklos Szeredi return error; 2818d18e9008SMiklos Szeredi 2819d18e9008SMiklos Szeredi return security_inode_create(dir->dentry->d_inode, dentry, mode); 2820d18e9008SMiklos Szeredi } 2821d18e9008SMiklos Szeredi 28221acf0af9SDavid Howells /* 28231acf0af9SDavid Howells * Attempt to atomically look up, create and open a file from a negative 28241acf0af9SDavid Howells * dentry. 28251acf0af9SDavid Howells * 28261acf0af9SDavid Howells * Returns 0 if successful. The file will have been created and attached to 28271acf0af9SDavid Howells * @file by the filesystem calling finish_open(). 28281acf0af9SDavid Howells * 28291acf0af9SDavid Howells * Returns 1 if the file was looked up only or didn't need creating. The 28301acf0af9SDavid Howells * caller will need to perform the open themselves. @path will have been 28311acf0af9SDavid Howells * updated to point to the new dentry. This may be negative. 28321acf0af9SDavid Howells * 28331acf0af9SDavid Howells * Returns an error code otherwise. 28341acf0af9SDavid Howells */ 28352675a4ebSAl Viro static int atomic_open(struct nameidata *nd, struct dentry *dentry, 283630d90494SAl Viro struct path *path, struct file *file, 2837015c3bbcSMiklos Szeredi const struct open_flags *op, 283864894cf8SAl Viro bool got_write, bool need_lookup, 283947237687SAl Viro int *opened) 2840d18e9008SMiklos Szeredi { 2841d18e9008SMiklos Szeredi struct inode *dir = nd->path.dentry->d_inode; 2842d18e9008SMiklos Szeredi unsigned open_flag = open_to_namei_flags(op->open_flag); 2843d18e9008SMiklos Szeredi umode_t mode; 2844d18e9008SMiklos Szeredi int error; 2845d18e9008SMiklos Szeredi int acc_mode; 2846d18e9008SMiklos Szeredi int create_error = 0; 2847d18e9008SMiklos Szeredi struct dentry *const DENTRY_NOT_SET = (void *) -1UL; 2848116cc022SMiklos Szeredi bool excl; 2849d18e9008SMiklos Szeredi 2850d18e9008SMiklos Szeredi BUG_ON(dentry->d_inode); 2851d18e9008SMiklos Szeredi 2852d18e9008SMiklos Szeredi /* Don't create child dentry for a dead directory. */ 2853d18e9008SMiklos Szeredi if (unlikely(IS_DEADDIR(dir))) { 28542675a4ebSAl Viro error = -ENOENT; 2855d18e9008SMiklos Szeredi goto out; 2856d18e9008SMiklos Szeredi } 2857d18e9008SMiklos Szeredi 285862b259d8SMiklos Szeredi mode = op->mode; 2859d18e9008SMiklos Szeredi if ((open_flag & O_CREAT) && !IS_POSIXACL(dir)) 2860d18e9008SMiklos Szeredi mode &= ~current_umask(); 2861d18e9008SMiklos Szeredi 2862116cc022SMiklos Szeredi excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT); 2863116cc022SMiklos Szeredi if (excl) 2864d18e9008SMiklos Szeredi open_flag &= ~O_TRUNC; 2865d18e9008SMiklos Szeredi 2866d18e9008SMiklos Szeredi /* 2867d18e9008SMiklos Szeredi * Checking write permission is tricky, bacuse we don't know if we are 2868d18e9008SMiklos Szeredi * going to actually need it: O_CREAT opens should work as long as the 2869d18e9008SMiklos Szeredi * file exists. But checking existence breaks atomicity. The trick is 2870d18e9008SMiklos Szeredi * to check access and if not granted clear O_CREAT from the flags. 2871d18e9008SMiklos Szeredi * 2872d18e9008SMiklos Szeredi * Another problem is returing the "right" error value (e.g. for an 2873d18e9008SMiklos Szeredi * O_EXCL open we want to return EEXIST not EROFS). 2874d18e9008SMiklos Szeredi */ 287564894cf8SAl Viro if (((open_flag & (O_CREAT | O_TRUNC)) || 287664894cf8SAl Viro (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) { 287764894cf8SAl Viro if (!(open_flag & O_CREAT)) { 2878d18e9008SMiklos Szeredi /* 2879d18e9008SMiklos Szeredi * No O_CREATE -> atomicity not a requirement -> fall 2880d18e9008SMiklos Szeredi * back to lookup + open 2881d18e9008SMiklos Szeredi */ 2882d18e9008SMiklos Szeredi goto no_open; 2883d18e9008SMiklos Szeredi } else if (open_flag & (O_EXCL | O_TRUNC)) { 2884d18e9008SMiklos Szeredi /* Fall back and fail with the right error */ 288564894cf8SAl Viro create_error = -EROFS; 2886d18e9008SMiklos Szeredi goto no_open; 2887d18e9008SMiklos Szeredi } else { 2888d18e9008SMiklos Szeredi /* No side effects, safe to clear O_CREAT */ 288964894cf8SAl Viro create_error = -EROFS; 2890d18e9008SMiklos Szeredi open_flag &= ~O_CREAT; 2891d18e9008SMiklos Szeredi } 2892d18e9008SMiklos Szeredi } 2893d18e9008SMiklos Szeredi 2894d18e9008SMiklos Szeredi if (open_flag & O_CREAT) { 289538227f78SMiklos Szeredi error = may_o_create(&nd->path, dentry, mode); 2896d18e9008SMiklos Szeredi if (error) { 2897d18e9008SMiklos Szeredi create_error = error; 2898d18e9008SMiklos Szeredi if (open_flag & O_EXCL) 2899d18e9008SMiklos Szeredi goto no_open; 2900d18e9008SMiklos Szeredi open_flag &= ~O_CREAT; 2901d18e9008SMiklos Szeredi } 2902d18e9008SMiklos Szeredi } 2903d18e9008SMiklos Szeredi 2904d18e9008SMiklos Szeredi if (nd->flags & LOOKUP_DIRECTORY) 2905d18e9008SMiklos Szeredi open_flag |= O_DIRECTORY; 2906d18e9008SMiklos Szeredi 290730d90494SAl Viro file->f_path.dentry = DENTRY_NOT_SET; 290830d90494SAl Viro file->f_path.mnt = nd->path.mnt; 290930d90494SAl Viro error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode, 291047237687SAl Viro opened); 2911d9585277SAl Viro if (error < 0) { 2912d9585277SAl Viro if (create_error && error == -ENOENT) 2913d9585277SAl Viro error = create_error; 2914d18e9008SMiklos Szeredi goto out; 2915d18e9008SMiklos Szeredi } 2916d18e9008SMiklos Szeredi 2917d9585277SAl Viro if (error) { /* returned 1, that is */ 291830d90494SAl Viro if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { 29192675a4ebSAl Viro error = -EIO; 2920d18e9008SMiklos Szeredi goto out; 2921d18e9008SMiklos Szeredi } 292230d90494SAl Viro if (file->f_path.dentry) { 2923d18e9008SMiklos Szeredi dput(dentry); 292430d90494SAl Viro dentry = file->f_path.dentry; 2925d18e9008SMiklos Szeredi } 292603da633aSAl Viro if (*opened & FILE_CREATED) 292703da633aSAl Viro fsnotify_create(dir, dentry); 292803da633aSAl Viro if (!dentry->d_inode) { 292903da633aSAl Viro WARN_ON(*opened & FILE_CREATED); 293003da633aSAl Viro if (create_error) { 293162b2ce96SSage Weil error = create_error; 293262b2ce96SSage Weil goto out; 293362b2ce96SSage Weil } 293403da633aSAl Viro } else { 293503da633aSAl Viro if (excl && !(*opened & FILE_CREATED)) { 293603da633aSAl Viro error = -EEXIST; 293703da633aSAl Viro goto out; 293803da633aSAl Viro } 293903da633aSAl Viro } 2940d18e9008SMiklos Szeredi goto looked_up; 2941d18e9008SMiklos Szeredi } 2942d18e9008SMiklos Szeredi 2943d18e9008SMiklos Szeredi /* 2944d18e9008SMiklos Szeredi * We didn't have the inode before the open, so check open permission 2945d18e9008SMiklos Szeredi * here. 2946d18e9008SMiklos Szeredi */ 294703da633aSAl Viro acc_mode = op->acc_mode; 294803da633aSAl Viro if (*opened & FILE_CREATED) { 294903da633aSAl Viro WARN_ON(!(open_flag & O_CREAT)); 295003da633aSAl Viro fsnotify_create(dir, dentry); 295162fb4a15SAl Viro acc_mode = 0; 295203da633aSAl Viro } 29532675a4ebSAl Viro error = may_open(&file->f_path, acc_mode, open_flag); 29542675a4ebSAl Viro if (error) 29552675a4ebSAl Viro fput(file); 2956d18e9008SMiklos Szeredi 2957d18e9008SMiklos Szeredi out: 2958d18e9008SMiklos Szeredi dput(dentry); 29592675a4ebSAl Viro return error; 2960d18e9008SMiklos Szeredi 2961d18e9008SMiklos Szeredi no_open: 2962d18e9008SMiklos Szeredi if (need_lookup) { 296372bd866aSAl Viro dentry = lookup_real(dir, dentry, nd->flags); 2964d18e9008SMiklos Szeredi if (IS_ERR(dentry)) 29652675a4ebSAl Viro return PTR_ERR(dentry); 2966d18e9008SMiklos Szeredi 2967d18e9008SMiklos Szeredi if (create_error) { 2968d18e9008SMiklos Szeredi int open_flag = op->open_flag; 2969d18e9008SMiklos Szeredi 29702675a4ebSAl Viro error = create_error; 2971d18e9008SMiklos Szeredi if ((open_flag & O_EXCL)) { 2972d18e9008SMiklos Szeredi if (!dentry->d_inode) 2973d18e9008SMiklos Szeredi goto out; 2974d18e9008SMiklos Szeredi } else if (!dentry->d_inode) { 2975d18e9008SMiklos Szeredi goto out; 2976d18e9008SMiklos Szeredi } else if ((open_flag & O_TRUNC) && 2977e36cb0b8SDavid Howells d_is_reg(dentry)) { 2978d18e9008SMiklos Szeredi goto out; 2979d18e9008SMiklos Szeredi } 2980d18e9008SMiklos Szeredi /* will fail later, go on to get the right error */ 2981d18e9008SMiklos Szeredi } 2982d18e9008SMiklos Szeredi } 2983d18e9008SMiklos Szeredi looked_up: 2984d18e9008SMiklos Szeredi path->dentry = dentry; 2985d18e9008SMiklos Szeredi path->mnt = nd->path.mnt; 29862675a4ebSAl Viro return 1; 2987d18e9008SMiklos Szeredi } 2988d18e9008SMiklos Szeredi 298931e6b01fSNick Piggin /* 29901acf0af9SDavid Howells * Look up and maybe create and open the last component. 2991d58ffd35SMiklos Szeredi * 2992d58ffd35SMiklos Szeredi * Must be called with i_mutex held on parent. 2993d58ffd35SMiklos Szeredi * 29941acf0af9SDavid Howells * Returns 0 if the file was successfully atomically created (if necessary) and 29951acf0af9SDavid Howells * opened. In this case the file will be returned attached to @file. 29961acf0af9SDavid Howells * 29971acf0af9SDavid Howells * Returns 1 if the file was not completely opened at this time, though lookups 29981acf0af9SDavid Howells * and creations will have been performed and the dentry returned in @path will 29991acf0af9SDavid Howells * be positive upon return if O_CREAT was specified. If O_CREAT wasn't 30001acf0af9SDavid Howells * specified then a negative dentry may be returned. 30011acf0af9SDavid Howells * 30021acf0af9SDavid Howells * An error code is returned otherwise. 30031acf0af9SDavid Howells * 30041acf0af9SDavid Howells * FILE_CREATE will be set in @*opened if the dentry was created and will be 30051acf0af9SDavid Howells * cleared otherwise prior to returning. 3006d58ffd35SMiklos Szeredi */ 30072675a4ebSAl Viro static int lookup_open(struct nameidata *nd, struct path *path, 300830d90494SAl Viro struct file *file, 3009d58ffd35SMiklos Szeredi const struct open_flags *op, 301064894cf8SAl Viro bool got_write, int *opened) 3011d58ffd35SMiklos Szeredi { 3012d58ffd35SMiklos Szeredi struct dentry *dir = nd->path.dentry; 301354ef4872SMiklos Szeredi struct inode *dir_inode = dir->d_inode; 3014d58ffd35SMiklos Szeredi struct dentry *dentry; 3015d58ffd35SMiklos Szeredi int error; 301654ef4872SMiklos Szeredi bool need_lookup; 3017d58ffd35SMiklos Szeredi 301847237687SAl Viro *opened &= ~FILE_CREATED; 3019201f956eSAl Viro dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup); 3020d58ffd35SMiklos Szeredi if (IS_ERR(dentry)) 30212675a4ebSAl Viro return PTR_ERR(dentry); 3022d58ffd35SMiklos Szeredi 3023d18e9008SMiklos Szeredi /* Cached positive dentry: will open in f_op->open */ 3024d18e9008SMiklos Szeredi if (!need_lookup && dentry->d_inode) 3025d18e9008SMiklos Szeredi goto out_no_open; 3026d18e9008SMiklos Szeredi 3027d18e9008SMiklos Szeredi if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) { 302864894cf8SAl Viro return atomic_open(nd, dentry, path, file, op, got_write, 302947237687SAl Viro need_lookup, opened); 3030d18e9008SMiklos Szeredi } 3031d18e9008SMiklos Szeredi 303254ef4872SMiklos Szeredi if (need_lookup) { 303354ef4872SMiklos Szeredi BUG_ON(dentry->d_inode); 303454ef4872SMiklos Szeredi 303572bd866aSAl Viro dentry = lookup_real(dir_inode, dentry, nd->flags); 303654ef4872SMiklos Szeredi if (IS_ERR(dentry)) 30372675a4ebSAl Viro return PTR_ERR(dentry); 303854ef4872SMiklos Szeredi } 303954ef4872SMiklos Szeredi 3040d58ffd35SMiklos Szeredi /* Negative dentry, just create the file */ 3041d58ffd35SMiklos Szeredi if (!dentry->d_inode && (op->open_flag & O_CREAT)) { 3042d58ffd35SMiklos Szeredi umode_t mode = op->mode; 3043d58ffd35SMiklos Szeredi if (!IS_POSIXACL(dir->d_inode)) 3044d58ffd35SMiklos Szeredi mode &= ~current_umask(); 3045d58ffd35SMiklos Szeredi /* 3046d58ffd35SMiklos Szeredi * This write is needed to ensure that a 3047d58ffd35SMiklos Szeredi * rw->ro transition does not occur between 3048d58ffd35SMiklos Szeredi * the time when the file is created and when 3049d58ffd35SMiklos Szeredi * a permanent write count is taken through 3050015c3bbcSMiklos Szeredi * the 'struct file' in finish_open(). 3051d58ffd35SMiklos Szeredi */ 305264894cf8SAl Viro if (!got_write) { 305364894cf8SAl Viro error = -EROFS; 3054d58ffd35SMiklos Szeredi goto out_dput; 305564894cf8SAl Viro } 305647237687SAl Viro *opened |= FILE_CREATED; 3057d58ffd35SMiklos Szeredi error = security_path_mknod(&nd->path, dentry, mode, 0); 3058d58ffd35SMiklos Szeredi if (error) 3059d58ffd35SMiklos Szeredi goto out_dput; 3060312b63fbSAl Viro error = vfs_create(dir->d_inode, dentry, mode, 3061312b63fbSAl Viro nd->flags & LOOKUP_EXCL); 3062d58ffd35SMiklos Szeredi if (error) 3063d58ffd35SMiklos Szeredi goto out_dput; 3064d58ffd35SMiklos Szeredi } 3065d18e9008SMiklos Szeredi out_no_open: 3066d58ffd35SMiklos Szeredi path->dentry = dentry; 3067d58ffd35SMiklos Szeredi path->mnt = nd->path.mnt; 30682675a4ebSAl Viro return 1; 3069d58ffd35SMiklos Szeredi 3070d58ffd35SMiklos Szeredi out_dput: 3071d58ffd35SMiklos Szeredi dput(dentry); 30722675a4ebSAl Viro return error; 3073d58ffd35SMiklos Szeredi } 3074d58ffd35SMiklos Szeredi 3075d58ffd35SMiklos Szeredi /* 3076fe2d35ffSAl Viro * Handle the last step of open() 307731e6b01fSNick Piggin */ 3078896475d5SAl Viro static int do_last(struct nameidata *nd, 307930d90494SAl Viro struct file *file, const struct open_flags *op, 308076ae2a5aSAl Viro int *opened) 3081fb1cc555SAl Viro { 3082a1e28038SAl Viro struct dentry *dir = nd->path.dentry; 3083ca344a89SAl Viro int open_flag = op->open_flag; 308477d660a8SMiklos Szeredi bool will_truncate = (open_flag & O_TRUNC) != 0; 308564894cf8SAl Viro bool got_write = false; 3086bcda7652SAl Viro int acc_mode = op->acc_mode; 3087254cf582SAl Viro unsigned seq; 3088a1eb3315SMiklos Szeredi struct inode *inode; 308916b1c1cdSMiklos Szeredi struct path save_parent = { .dentry = NULL, .mnt = NULL }; 3090896475d5SAl Viro struct path path; 309116b1c1cdSMiklos Szeredi bool retried = false; 309216c2cd71SAl Viro int error; 3093fb1cc555SAl Viro 3094c3e380b0SAl Viro nd->flags &= ~LOOKUP_PARENT; 3095c3e380b0SAl Viro nd->flags |= op->intent; 3096c3e380b0SAl Viro 3097bc77daa7SAl Viro if (nd->last_type != LAST_NORM) { 3098fe2d35ffSAl Viro error = handle_dots(nd, nd->last_type); 3099deb106c6SAl Viro if (unlikely(error)) 31002675a4ebSAl Viro return error; 3101e83db167SMiklos Szeredi goto finish_open; 31021f36f774SAl Viro } 3103a2c36b45SAl Viro 3104ca344a89SAl Viro if (!(open_flag & O_CREAT)) { 3105fe2d35ffSAl Viro if (nd->last.name[nd->last.len]) 3106fe2d35ffSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 3107fe2d35ffSAl Viro /* we _can_ be in RCU mode here */ 3108254cf582SAl Viro error = lookup_fast(nd, &path, &inode, &seq); 310971574865SMiklos Szeredi if (likely(!error)) 311071574865SMiklos Szeredi goto finish_lookup; 311171574865SMiklos Szeredi 3112ce57dfc1SAl Viro if (error < 0) 3113deb106c6SAl Viro return error; 3114a1eb3315SMiklos Szeredi 311537d7fffcSMiklos Szeredi BUG_ON(nd->inode != dir->d_inode); 3116b6183df7SMiklos Szeredi } else { 3117fe2d35ffSAl Viro /* create side of things */ 3118a3fbbde7SAl Viro /* 3119b6183df7SMiklos Szeredi * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED 3120b6183df7SMiklos Szeredi * has been cleared when we got to the last component we are 3121b6183df7SMiklos Szeredi * about to look up 3122a3fbbde7SAl Viro */ 31239f1fafeeSAl Viro error = complete_walk(nd); 3124e8bb73dfSAl Viro if (error) 31252675a4ebSAl Viro return error; 3126fe2d35ffSAl Viro 312776ae2a5aSAl Viro audit_inode(nd->name, dir, LOOKUP_PARENT); 31281f36f774SAl Viro /* trailing slashes? */ 3129deb106c6SAl Viro if (unlikely(nd->last.name[nd->last.len])) 3130deb106c6SAl Viro return -EISDIR; 3131b6183df7SMiklos Szeredi } 31321f36f774SAl Viro 313316b1c1cdSMiklos Szeredi retry_lookup: 313464894cf8SAl Viro if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { 313564894cf8SAl Viro error = mnt_want_write(nd->path.mnt); 313664894cf8SAl Viro if (!error) 313764894cf8SAl Viro got_write = true; 313864894cf8SAl Viro /* 313964894cf8SAl Viro * do _not_ fail yet - we might not need that or fail with 314064894cf8SAl Viro * a different error; let lookup_open() decide; we'll be 314164894cf8SAl Viro * dropping this one anyway. 314264894cf8SAl Viro */ 314364894cf8SAl Viro } 31445955102cSAl Viro inode_lock(dir->d_inode); 3145896475d5SAl Viro error = lookup_open(nd, &path, file, op, got_write, opened); 31465955102cSAl Viro inode_unlock(dir->d_inode); 3147fb1cc555SAl Viro 31482675a4ebSAl Viro if (error <= 0) { 31492675a4ebSAl Viro if (error) 3150d58ffd35SMiklos Szeredi goto out; 31516c0d46c4SAl Viro 315247237687SAl Viro if ((*opened & FILE_CREATED) || 3153496ad9aaSAl Viro !S_ISREG(file_inode(file)->i_mode)) 315477d660a8SMiklos Szeredi will_truncate = false; 3155d18e9008SMiklos Szeredi 315676ae2a5aSAl Viro audit_inode(nd->name, file->f_path.dentry, 0); 3157d18e9008SMiklos Szeredi goto opened; 3158d18e9008SMiklos Szeredi } 3159d18e9008SMiklos Szeredi 316047237687SAl Viro if (*opened & FILE_CREATED) { 31619b44f1b3SAl Viro /* Don't check for write permission, don't truncate */ 3162ca344a89SAl Viro open_flag &= ~O_TRUNC; 316377d660a8SMiklos Szeredi will_truncate = false; 316462fb4a15SAl Viro acc_mode = 0; 3165896475d5SAl Viro path_to_nameidata(&path, nd); 3166e83db167SMiklos Szeredi goto finish_open_created; 3167fb1cc555SAl Viro } 3168fb1cc555SAl Viro 3169fb1cc555SAl Viro /* 31703134f37eSJeff Layton * create/update audit record if it already exists. 3171fb1cc555SAl Viro */ 3172896475d5SAl Viro if (d_is_positive(path.dentry)) 317376ae2a5aSAl Viro audit_inode(nd->name, path.dentry, 0); 3174fb1cc555SAl Viro 3175d18e9008SMiklos Szeredi /* 3176d18e9008SMiklos Szeredi * If atomic_open() acquired write access it is dropped now due to 3177d18e9008SMiklos Szeredi * possible mount and symlink following (this might be optimized away if 3178d18e9008SMiklos Szeredi * necessary...) 3179d18e9008SMiklos Szeredi */ 318064894cf8SAl Viro if (got_write) { 3181d18e9008SMiklos Szeredi mnt_drop_write(nd->path.mnt); 318264894cf8SAl Viro got_write = false; 3183d18e9008SMiklos Szeredi } 3184d18e9008SMiklos Szeredi 3185deb106c6SAl Viro if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) { 3186deb106c6SAl Viro path_to_nameidata(&path, nd); 3187deb106c6SAl Viro return -EEXIST; 3188deb106c6SAl Viro } 3189fb1cc555SAl Viro 3190756daf26SNeilBrown error = follow_managed(&path, nd); 3191deb106c6SAl Viro if (unlikely(error < 0)) 3192deb106c6SAl Viro return error; 3193a3fbbde7SAl Viro 3194decf3400SMiklos Szeredi BUG_ON(nd->flags & LOOKUP_RCU); 319563afdfc7SDavid Howells inode = d_backing_inode(path.dentry); 3196254cf582SAl Viro seq = 0; /* out of RCU mode, so the value doesn't matter */ 3197deb106c6SAl Viro if (unlikely(d_is_negative(path.dentry))) { 3198896475d5SAl Viro path_to_nameidata(&path, nd); 3199deb106c6SAl Viro return -ENOENT; 320054c33e7fSMiklos Szeredi } 3201766c4cbfSAl Viro finish_lookup: 3202191d7f73SAl Viro if (nd->depth) 3203191d7f73SAl Viro put_link(nd); 3204181548c0SAl Viro error = should_follow_link(nd, &path, nd->flags & LOOKUP_FOLLOW, 3205181548c0SAl Viro inode, seq); 3206deb106c6SAl Viro if (unlikely(error)) 3207d63ff28fSAl Viro return error; 3208fb1cc555SAl Viro 3209896475d5SAl Viro if (unlikely(d_is_symlink(path.dentry)) && !(open_flag & O_PATH)) { 3210896475d5SAl Viro path_to_nameidata(&path, nd); 3211deb106c6SAl Viro return -ELOOP; 3212a5cfe2d5SAl Viro } 3213a5cfe2d5SAl Viro 3214896475d5SAl Viro if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path.mnt) { 3215896475d5SAl Viro path_to_nameidata(&path, nd); 321616b1c1cdSMiklos Szeredi } else { 321716b1c1cdSMiklos Szeredi save_parent.dentry = nd->path.dentry; 3218896475d5SAl Viro save_parent.mnt = mntget(path.mnt); 3219896475d5SAl Viro nd->path.dentry = path.dentry; 322016b1c1cdSMiklos Szeredi 322116b1c1cdSMiklos Szeredi } 3222decf3400SMiklos Szeredi nd->inode = inode; 3223254cf582SAl Viro nd->seq = seq; 3224a3fbbde7SAl Viro /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ 3225bc77daa7SAl Viro finish_open: 3226a3fbbde7SAl Viro error = complete_walk(nd); 322716b1c1cdSMiklos Szeredi if (error) { 322816b1c1cdSMiklos Szeredi path_put(&save_parent); 32292675a4ebSAl Viro return error; 323016b1c1cdSMiklos Szeredi } 323176ae2a5aSAl Viro audit_inode(nd->name, nd->path.dentry, 0); 3232fb1cc555SAl Viro error = -EISDIR; 323344b1d530SMiklos Szeredi if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry)) 32342675a4ebSAl Viro goto out; 3235af2f5542SMiklos Szeredi error = -ENOTDIR; 323644b1d530SMiklos Szeredi if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry)) 32372675a4ebSAl Viro goto out; 32384bbcbd3bSDavid Howells if (!d_is_reg(nd->path.dentry)) 323977d660a8SMiklos Szeredi will_truncate = false; 32406c0d46c4SAl Viro 32410f9d1a10SAl Viro if (will_truncate) { 32420f9d1a10SAl Viro error = mnt_want_write(nd->path.mnt); 32430f9d1a10SAl Viro if (error) 32442675a4ebSAl Viro goto out; 324564894cf8SAl Viro got_write = true; 32460f9d1a10SAl Viro } 3247e83db167SMiklos Szeredi finish_open_created: 324862fb4a15SAl Viro if (likely(!(open_flag & O_PATH))) { 3249bcda7652SAl Viro error = may_open(&nd->path, acc_mode, open_flag); 3250ca344a89SAl Viro if (error) 32512675a4ebSAl Viro goto out; 325262fb4a15SAl Viro } 32534aa7c634SMiklos Szeredi BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */ 32544aa7c634SMiklos Szeredi error = vfs_open(&nd->path, file, current_cred()); 32554aa7c634SMiklos Szeredi if (!error) { 32564aa7c634SMiklos Szeredi *opened |= FILE_OPENED; 32574aa7c634SMiklos Szeredi } else { 325830d90494SAl Viro if (error == -EOPENSTALE) 3259f60dc3dbSMiklos Szeredi goto stale_open; 3260015c3bbcSMiklos Szeredi goto out; 3261f60dc3dbSMiklos Szeredi } 3262a8277b9bSMiklos Szeredi opened: 32632675a4ebSAl Viro error = open_check_o_direct(file); 3264015c3bbcSMiklos Szeredi if (error) 3265015c3bbcSMiklos Szeredi goto exit_fput; 32663034a146SDmitry Kasatkin error = ima_file_check(file, op->acc_mode, *opened); 3267aa4caadbSMiklos Szeredi if (error) 3268aa4caadbSMiklos Szeredi goto exit_fput; 3269aa4caadbSMiklos Szeredi 32700f9d1a10SAl Viro if (will_truncate) { 32712675a4ebSAl Viro error = handle_truncate(file); 3272aa4caadbSMiklos Szeredi if (error) 3273aa4caadbSMiklos Szeredi goto exit_fput; 32740f9d1a10SAl Viro } 3275ca344a89SAl Viro out: 3276c80567c8SAl Viro if (unlikely(error > 0)) { 3277c80567c8SAl Viro WARN_ON(1); 3278c80567c8SAl Viro error = -EINVAL; 3279c80567c8SAl Viro } 328064894cf8SAl Viro if (got_write) 32810f9d1a10SAl Viro mnt_drop_write(nd->path.mnt); 328216b1c1cdSMiklos Szeredi path_put(&save_parent); 32832675a4ebSAl Viro return error; 3284fb1cc555SAl Viro 3285015c3bbcSMiklos Szeredi exit_fput: 32862675a4ebSAl Viro fput(file); 32872675a4ebSAl Viro goto out; 3288015c3bbcSMiklos Szeredi 3289f60dc3dbSMiklos Szeredi stale_open: 3290f60dc3dbSMiklos Szeredi /* If no saved parent or already retried then can't retry */ 3291f60dc3dbSMiklos Szeredi if (!save_parent.dentry || retried) 3292f60dc3dbSMiklos Szeredi goto out; 3293f60dc3dbSMiklos Szeredi 3294f60dc3dbSMiklos Szeredi BUG_ON(save_parent.dentry != dir); 3295f60dc3dbSMiklos Szeredi path_put(&nd->path); 3296f60dc3dbSMiklos Szeredi nd->path = save_parent; 3297f60dc3dbSMiklos Szeredi nd->inode = dir->d_inode; 3298f60dc3dbSMiklos Szeredi save_parent.mnt = NULL; 3299f60dc3dbSMiklos Szeredi save_parent.dentry = NULL; 330064894cf8SAl Viro if (got_write) { 3301f60dc3dbSMiklos Szeredi mnt_drop_write(nd->path.mnt); 330264894cf8SAl Viro got_write = false; 3303f60dc3dbSMiklos Szeredi } 3304f60dc3dbSMiklos Szeredi retried = true; 3305f60dc3dbSMiklos Szeredi goto retry_lookup; 3306fb1cc555SAl Viro } 3307fb1cc555SAl Viro 3308c8a53ee5SAl Viro static int do_tmpfile(struct nameidata *nd, unsigned flags, 330960545d0dSAl Viro const struct open_flags *op, 331060545d0dSAl Viro struct file *file, int *opened) 331160545d0dSAl Viro { 331260545d0dSAl Viro static const struct qstr name = QSTR_INIT("/", 1); 3313625b6d10SAl Viro struct dentry *child; 331460545d0dSAl Viro struct inode *dir; 3315625b6d10SAl Viro struct path path; 3316c8a53ee5SAl Viro int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path); 331760545d0dSAl Viro if (unlikely(error)) 331860545d0dSAl Viro return error; 3319625b6d10SAl Viro error = mnt_want_write(path.mnt); 332060545d0dSAl Viro if (unlikely(error)) 332160545d0dSAl Viro goto out; 3322625b6d10SAl Viro dir = path.dentry->d_inode; 332360545d0dSAl Viro /* we want directory to be writable */ 3324625b6d10SAl Viro error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 332560545d0dSAl Viro if (error) 332660545d0dSAl Viro goto out2; 332760545d0dSAl Viro if (!dir->i_op->tmpfile) { 332860545d0dSAl Viro error = -EOPNOTSUPP; 332960545d0dSAl Viro goto out2; 333060545d0dSAl Viro } 3331625b6d10SAl Viro child = d_alloc(path.dentry, &name); 333260545d0dSAl Viro if (unlikely(!child)) { 333360545d0dSAl Viro error = -ENOMEM; 333460545d0dSAl Viro goto out2; 333560545d0dSAl Viro } 3336625b6d10SAl Viro dput(path.dentry); 3337625b6d10SAl Viro path.dentry = child; 3338625b6d10SAl Viro error = dir->i_op->tmpfile(dir, child, op->mode); 333960545d0dSAl Viro if (error) 334060545d0dSAl Viro goto out2; 3341c8a53ee5SAl Viro audit_inode(nd->name, child, 0); 334269a91c23SEric Rannaud /* Don't check for other permissions, the inode was just created */ 334362fb4a15SAl Viro error = may_open(&path, 0, op->open_flag); 334460545d0dSAl Viro if (error) 334560545d0dSAl Viro goto out2; 3346625b6d10SAl Viro file->f_path.mnt = path.mnt; 3347625b6d10SAl Viro error = finish_open(file, child, NULL, opened); 334860545d0dSAl Viro if (error) 334960545d0dSAl Viro goto out2; 335060545d0dSAl Viro error = open_check_o_direct(file); 3351f4e0c30cSAl Viro if (error) { 335260545d0dSAl Viro fput(file); 3353f4e0c30cSAl Viro } else if (!(op->open_flag & O_EXCL)) { 3354f4e0c30cSAl Viro struct inode *inode = file_inode(file); 3355f4e0c30cSAl Viro spin_lock(&inode->i_lock); 3356f4e0c30cSAl Viro inode->i_state |= I_LINKABLE; 3357f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 3358f4e0c30cSAl Viro } 335960545d0dSAl Viro out2: 3360625b6d10SAl Viro mnt_drop_write(path.mnt); 336160545d0dSAl Viro out: 3362625b6d10SAl Viro path_put(&path); 336360545d0dSAl Viro return error; 336460545d0dSAl Viro } 336560545d0dSAl Viro 3366c8a53ee5SAl Viro static struct file *path_openat(struct nameidata *nd, 3367c8a53ee5SAl Viro const struct open_flags *op, unsigned flags) 33681da177e4SLinus Torvalds { 3369368ee9baSAl Viro const char *s; 337030d90494SAl Viro struct file *file; 337147237687SAl Viro int opened = 0; 337213aab428SAl Viro int error; 337331e6b01fSNick Piggin 337430d90494SAl Viro file = get_empty_filp(); 33751afc99beSAl Viro if (IS_ERR(file)) 33761afc99beSAl Viro return file; 337731e6b01fSNick Piggin 337830d90494SAl Viro file->f_flags = op->open_flag; 337931e6b01fSNick Piggin 3380bb458c64SAl Viro if (unlikely(file->f_flags & __O_TMPFILE)) { 3381c8a53ee5SAl Viro error = do_tmpfile(nd, flags, op, file, &opened); 3382f15133dfSAl Viro goto out2; 338360545d0dSAl Viro } 338460545d0dSAl Viro 3385c8a53ee5SAl Viro s = path_init(nd, flags); 3386368ee9baSAl Viro if (IS_ERR(s)) { 3387368ee9baSAl Viro put_filp(file); 3388368ee9baSAl Viro return ERR_CAST(s); 3389368ee9baSAl Viro } 33903bdba28bSAl Viro while (!(error = link_path_walk(s, nd)) && 339176ae2a5aSAl Viro (error = do_last(nd, file, op, &opened)) > 0) { 339273d049a4SAl Viro nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); 33933bdba28bSAl Viro s = trailing_symlink(nd); 33943bdba28bSAl Viro if (IS_ERR(s)) { 33953bdba28bSAl Viro error = PTR_ERR(s); 33962675a4ebSAl Viro break; 3397806b681cSAl Viro } 33983bdba28bSAl Viro } 3399deb106c6SAl Viro terminate_walk(nd); 3400f15133dfSAl Viro out2: 34012675a4ebSAl Viro if (!(opened & FILE_OPENED)) { 34022675a4ebSAl Viro BUG_ON(!error); 340330d90494SAl Viro put_filp(file); 3404015c3bbcSMiklos Szeredi } 34052675a4ebSAl Viro if (unlikely(error)) { 34062675a4ebSAl Viro if (error == -EOPENSTALE) { 34072675a4ebSAl Viro if (flags & LOOKUP_RCU) 34082675a4ebSAl Viro error = -ECHILD; 34092675a4ebSAl Viro else 34102675a4ebSAl Viro error = -ESTALE; 34112675a4ebSAl Viro } 34122675a4ebSAl Viro file = ERR_PTR(error); 34132675a4ebSAl Viro } 34142675a4ebSAl Viro return file; 3415de459215SKirill Korotaev } 34161da177e4SLinus Torvalds 3417669abf4eSJeff Layton struct file *do_filp_open(int dfd, struct filename *pathname, 3418f9652e10SAl Viro const struct open_flags *op) 341913aab428SAl Viro { 34209883d185SAl Viro struct nameidata nd; 3421f9652e10SAl Viro int flags = op->lookup_flags; 342213aab428SAl Viro struct file *filp; 342313aab428SAl Viro 34249883d185SAl Viro set_nameidata(&nd, dfd, pathname); 3425c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_RCU); 342613aab428SAl Viro if (unlikely(filp == ERR_PTR(-ECHILD))) 3427c8a53ee5SAl Viro filp = path_openat(&nd, op, flags); 342813aab428SAl Viro if (unlikely(filp == ERR_PTR(-ESTALE))) 3429c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_REVAL); 34309883d185SAl Viro restore_nameidata(); 343113aab428SAl Viro return filp; 343213aab428SAl Viro } 343313aab428SAl Viro 343473d049a4SAl Viro struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, 3435f9652e10SAl Viro const char *name, const struct open_flags *op) 343673d049a4SAl Viro { 34379883d185SAl Viro struct nameidata nd; 343873d049a4SAl Viro struct file *file; 343951689104SPaul Moore struct filename *filename; 3440f9652e10SAl Viro int flags = op->lookup_flags | LOOKUP_ROOT; 344173d049a4SAl Viro 344273d049a4SAl Viro nd.root.mnt = mnt; 344373d049a4SAl Viro nd.root.dentry = dentry; 344473d049a4SAl Viro 3445b18825a7SDavid Howells if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN) 344673d049a4SAl Viro return ERR_PTR(-ELOOP); 344773d049a4SAl Viro 344851689104SPaul Moore filename = getname_kernel(name); 3449a1c83681SViresh Kumar if (IS_ERR(filename)) 345051689104SPaul Moore return ERR_CAST(filename); 345151689104SPaul Moore 34529883d185SAl Viro set_nameidata(&nd, -1, filename); 3453c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_RCU); 345473d049a4SAl Viro if (unlikely(file == ERR_PTR(-ECHILD))) 3455c8a53ee5SAl Viro file = path_openat(&nd, op, flags); 345673d049a4SAl Viro if (unlikely(file == ERR_PTR(-ESTALE))) 3457c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_REVAL); 34589883d185SAl Viro restore_nameidata(); 345951689104SPaul Moore putname(filename); 346073d049a4SAl Viro return file; 346173d049a4SAl Viro } 346273d049a4SAl Viro 3463fa14a0b8SAl Viro static struct dentry *filename_create(int dfd, struct filename *name, 34641ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 34651da177e4SLinus Torvalds { 3466c663e5d8SChristoph Hellwig struct dentry *dentry = ERR_PTR(-EEXIST); 3467391172c4SAl Viro struct qstr last; 3468391172c4SAl Viro int type; 3469c30dabfeSJan Kara int err2; 34701ac12b4bSJeff Layton int error; 34711ac12b4bSJeff Layton bool is_dir = (lookup_flags & LOOKUP_DIRECTORY); 34721ac12b4bSJeff Layton 34731ac12b4bSJeff Layton /* 34741ac12b4bSJeff Layton * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any 34751ac12b4bSJeff Layton * other flags passed in are ignored! 34761ac12b4bSJeff Layton */ 34771ac12b4bSJeff Layton lookup_flags &= LOOKUP_REVAL; 34781ac12b4bSJeff Layton 34795c31b6ceSAl Viro name = filename_parentat(dfd, name, lookup_flags, path, &last, &type); 34805c31b6ceSAl Viro if (IS_ERR(name)) 34815c31b6ceSAl Viro return ERR_CAST(name); 34821da177e4SLinus Torvalds 3483c663e5d8SChristoph Hellwig /* 3484c663e5d8SChristoph Hellwig * Yucky last component or no last component at all? 3485c663e5d8SChristoph Hellwig * (foo/., foo/.., /////) 3486c663e5d8SChristoph Hellwig */ 34875c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) 3488ed75e95dSAl Viro goto out; 3489c663e5d8SChristoph Hellwig 3490c30dabfeSJan Kara /* don't fail immediately if it's r/o, at least try to report other errors */ 3491391172c4SAl Viro err2 = mnt_want_write(path->mnt); 3492c663e5d8SChristoph Hellwig /* 3493c663e5d8SChristoph Hellwig * Do the final lookup. 3494c663e5d8SChristoph Hellwig */ 3495391172c4SAl Viro lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL; 34965955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 3497391172c4SAl Viro dentry = __lookup_hash(&last, path->dentry, lookup_flags); 34981da177e4SLinus Torvalds if (IS_ERR(dentry)) 3499a8104a9fSAl Viro goto unlock; 3500c663e5d8SChristoph Hellwig 3501a8104a9fSAl Viro error = -EEXIST; 3502b18825a7SDavid Howells if (d_is_positive(dentry)) 3503a8104a9fSAl Viro goto fail; 3504b18825a7SDavid Howells 3505c663e5d8SChristoph Hellwig /* 3506c663e5d8SChristoph Hellwig * Special case - lookup gave negative, but... we had foo/bar/ 3507c663e5d8SChristoph Hellwig * From the vfs_mknod() POV we just have a negative dentry - 3508c663e5d8SChristoph Hellwig * all is fine. Let's be bastards - you had / on the end, you've 3509c663e5d8SChristoph Hellwig * been asking for (non-existent) directory. -ENOENT for you. 3510c663e5d8SChristoph Hellwig */ 3511391172c4SAl Viro if (unlikely(!is_dir && last.name[last.len])) { 3512a8104a9fSAl Viro error = -ENOENT; 3513ed75e95dSAl Viro goto fail; 3514e9baf6e5SAl Viro } 3515c30dabfeSJan Kara if (unlikely(err2)) { 3516c30dabfeSJan Kara error = err2; 3517a8104a9fSAl Viro goto fail; 3518c30dabfeSJan Kara } 3519181c37b6SAl Viro putname(name); 3520e9baf6e5SAl Viro return dentry; 35211da177e4SLinus Torvalds fail: 3522a8104a9fSAl Viro dput(dentry); 3523a8104a9fSAl Viro dentry = ERR_PTR(error); 3524a8104a9fSAl Viro unlock: 35255955102cSAl Viro inode_unlock(path->dentry->d_inode); 3526c30dabfeSJan Kara if (!err2) 3527391172c4SAl Viro mnt_drop_write(path->mnt); 3528ed75e95dSAl Viro out: 3529391172c4SAl Viro path_put(path); 3530181c37b6SAl Viro putname(name); 3531ed75e95dSAl Viro return dentry; 3532dae6ad8fSAl Viro } 3533fa14a0b8SAl Viro 3534fa14a0b8SAl Viro struct dentry *kern_path_create(int dfd, const char *pathname, 3535fa14a0b8SAl Viro struct path *path, unsigned int lookup_flags) 3536fa14a0b8SAl Viro { 3537181c37b6SAl Viro return filename_create(dfd, getname_kernel(pathname), 3538181c37b6SAl Viro path, lookup_flags); 3539fa14a0b8SAl Viro } 3540dae6ad8fSAl Viro EXPORT_SYMBOL(kern_path_create); 3541dae6ad8fSAl Viro 3542921a1650SAl Viro void done_path_create(struct path *path, struct dentry *dentry) 3543921a1650SAl Viro { 3544921a1650SAl Viro dput(dentry); 35455955102cSAl Viro inode_unlock(path->dentry->d_inode); 3546a8104a9fSAl Viro mnt_drop_write(path->mnt); 3547921a1650SAl Viro path_put(path); 3548921a1650SAl Viro } 3549921a1650SAl Viro EXPORT_SYMBOL(done_path_create); 3550921a1650SAl Viro 3551520ae687SAl Viro inline struct dentry *user_path_create(int dfd, const char __user *pathname, 35521ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 3553dae6ad8fSAl Viro { 3554181c37b6SAl Viro return filename_create(dfd, getname(pathname), path, lookup_flags); 3555dae6ad8fSAl Viro } 3556dae6ad8fSAl Viro EXPORT_SYMBOL(user_path_create); 3557dae6ad8fSAl Viro 35581a67aafbSAl Viro int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) 35591da177e4SLinus Torvalds { 3560a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 35611da177e4SLinus Torvalds 35621da177e4SLinus Torvalds if (error) 35631da177e4SLinus Torvalds return error; 35641da177e4SLinus Torvalds 3565975d6b39SEric W. Biederman if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) 35661da177e4SLinus Torvalds return -EPERM; 35671da177e4SLinus Torvalds 3568acfa4380SAl Viro if (!dir->i_op->mknod) 35691da177e4SLinus Torvalds return -EPERM; 35701da177e4SLinus Torvalds 357108ce5f16SSerge E. Hallyn error = devcgroup_inode_mknod(mode, dev); 357208ce5f16SSerge E. Hallyn if (error) 357308ce5f16SSerge E. Hallyn return error; 357408ce5f16SSerge E. Hallyn 35751da177e4SLinus Torvalds error = security_inode_mknod(dir, dentry, mode, dev); 35761da177e4SLinus Torvalds if (error) 35771da177e4SLinus Torvalds return error; 35781da177e4SLinus Torvalds 35791da177e4SLinus Torvalds error = dir->i_op->mknod(dir, dentry, mode, dev); 3580a74574aaSStephen Smalley if (!error) 3581f38aa942SAmy Griffis fsnotify_create(dir, dentry); 35821da177e4SLinus Torvalds return error; 35831da177e4SLinus Torvalds } 35844d359507SAl Viro EXPORT_SYMBOL(vfs_mknod); 35851da177e4SLinus Torvalds 3586f69aac00SAl Viro static int may_mknod(umode_t mode) 3587463c3197SDave Hansen { 3588463c3197SDave Hansen switch (mode & S_IFMT) { 3589463c3197SDave Hansen case S_IFREG: 3590463c3197SDave Hansen case S_IFCHR: 3591463c3197SDave Hansen case S_IFBLK: 3592463c3197SDave Hansen case S_IFIFO: 3593463c3197SDave Hansen case S_IFSOCK: 3594463c3197SDave Hansen case 0: /* zero mode translates to S_IFREG */ 3595463c3197SDave Hansen return 0; 3596463c3197SDave Hansen case S_IFDIR: 3597463c3197SDave Hansen return -EPERM; 3598463c3197SDave Hansen default: 3599463c3197SDave Hansen return -EINVAL; 3600463c3197SDave Hansen } 3601463c3197SDave Hansen } 3602463c3197SDave Hansen 36038208a22bSAl Viro SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, 36042e4d0924SHeiko Carstens unsigned, dev) 36051da177e4SLinus Torvalds { 36061da177e4SLinus Torvalds struct dentry *dentry; 3607dae6ad8fSAl Viro struct path path; 3608dae6ad8fSAl Viro int error; 3609972567f1SJeff Layton unsigned int lookup_flags = 0; 36101da177e4SLinus Torvalds 36118e4bfca1SAl Viro error = may_mknod(mode); 36128e4bfca1SAl Viro if (error) 36138e4bfca1SAl Viro return error; 3614972567f1SJeff Layton retry: 3615972567f1SJeff Layton dentry = user_path_create(dfd, filename, &path, lookup_flags); 3616dae6ad8fSAl Viro if (IS_ERR(dentry)) 3617dae6ad8fSAl Viro return PTR_ERR(dentry); 36182ad94ae6SAl Viro 3619dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3620ce3b0f8dSAl Viro mode &= ~current_umask(); 3621dae6ad8fSAl Viro error = security_path_mknod(&path, dentry, mode, dev); 3622be6d3e56SKentaro Takeda if (error) 3623a8104a9fSAl Viro goto out; 36241da177e4SLinus Torvalds switch (mode & S_IFMT) { 36251da177e4SLinus Torvalds case 0: case S_IFREG: 3626312b63fbSAl Viro error = vfs_create(path.dentry->d_inode,dentry,mode,true); 36271da177e4SLinus Torvalds break; 36281da177e4SLinus Torvalds case S_IFCHR: case S_IFBLK: 3629dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode, 36301da177e4SLinus Torvalds new_decode_dev(dev)); 36311da177e4SLinus Torvalds break; 36321da177e4SLinus Torvalds case S_IFIFO: case S_IFSOCK: 3633dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); 36341da177e4SLinus Torvalds break; 36351da177e4SLinus Torvalds } 3636a8104a9fSAl Viro out: 3637921a1650SAl Viro done_path_create(&path, dentry); 3638972567f1SJeff Layton if (retry_estale(error, lookup_flags)) { 3639972567f1SJeff Layton lookup_flags |= LOOKUP_REVAL; 3640972567f1SJeff Layton goto retry; 3641972567f1SJeff Layton } 36421da177e4SLinus Torvalds return error; 36431da177e4SLinus Torvalds } 36441da177e4SLinus Torvalds 36458208a22bSAl Viro SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) 36465590ff0dSUlrich Drepper { 36475590ff0dSUlrich Drepper return sys_mknodat(AT_FDCWD, filename, mode, dev); 36485590ff0dSUlrich Drepper } 36495590ff0dSUlrich Drepper 365018bb1db3SAl Viro int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 36511da177e4SLinus Torvalds { 3652a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 36538de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 36541da177e4SLinus Torvalds 36551da177e4SLinus Torvalds if (error) 36561da177e4SLinus Torvalds return error; 36571da177e4SLinus Torvalds 3658acfa4380SAl Viro if (!dir->i_op->mkdir) 36591da177e4SLinus Torvalds return -EPERM; 36601da177e4SLinus Torvalds 36611da177e4SLinus Torvalds mode &= (S_IRWXUGO|S_ISVTX); 36621da177e4SLinus Torvalds error = security_inode_mkdir(dir, dentry, mode); 36631da177e4SLinus Torvalds if (error) 36641da177e4SLinus Torvalds return error; 36651da177e4SLinus Torvalds 36668de52778SAl Viro if (max_links && dir->i_nlink >= max_links) 36678de52778SAl Viro return -EMLINK; 36688de52778SAl Viro 36691da177e4SLinus Torvalds error = dir->i_op->mkdir(dir, dentry, mode); 3670a74574aaSStephen Smalley if (!error) 3671f38aa942SAmy Griffis fsnotify_mkdir(dir, dentry); 36721da177e4SLinus Torvalds return error; 36731da177e4SLinus Torvalds } 36744d359507SAl Viro EXPORT_SYMBOL(vfs_mkdir); 36751da177e4SLinus Torvalds 3676a218d0fdSAl Viro SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) 36771da177e4SLinus Torvalds { 36786902d925SDave Hansen struct dentry *dentry; 3679dae6ad8fSAl Viro struct path path; 3680dae6ad8fSAl Viro int error; 3681b76d8b82SJeff Layton unsigned int lookup_flags = LOOKUP_DIRECTORY; 36821da177e4SLinus Torvalds 3683b76d8b82SJeff Layton retry: 3684b76d8b82SJeff Layton dentry = user_path_create(dfd, pathname, &path, lookup_flags); 36856902d925SDave Hansen if (IS_ERR(dentry)) 3686dae6ad8fSAl Viro return PTR_ERR(dentry); 36876902d925SDave Hansen 3688dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3689ce3b0f8dSAl Viro mode &= ~current_umask(); 3690dae6ad8fSAl Viro error = security_path_mkdir(&path, dentry, mode); 3691a8104a9fSAl Viro if (!error) 3692dae6ad8fSAl Viro error = vfs_mkdir(path.dentry->d_inode, dentry, mode); 3693921a1650SAl Viro done_path_create(&path, dentry); 3694b76d8b82SJeff Layton if (retry_estale(error, lookup_flags)) { 3695b76d8b82SJeff Layton lookup_flags |= LOOKUP_REVAL; 3696b76d8b82SJeff Layton goto retry; 3697b76d8b82SJeff Layton } 36981da177e4SLinus Torvalds return error; 36991da177e4SLinus Torvalds } 37001da177e4SLinus Torvalds 3701a218d0fdSAl Viro SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) 37025590ff0dSUlrich Drepper { 37035590ff0dSUlrich Drepper return sys_mkdirat(AT_FDCWD, pathname, mode); 37045590ff0dSUlrich Drepper } 37055590ff0dSUlrich Drepper 37061da177e4SLinus Torvalds /* 3707a71905f0SSage Weil * The dentry_unhash() helper will try to drop the dentry early: we 3708c0d02594SJ. Bruce Fields * should have a usage count of 1 if we're the only user of this 3709a71905f0SSage Weil * dentry, and if that is true (possibly after pruning the dcache), 3710a71905f0SSage Weil * then we drop the dentry now. 37111da177e4SLinus Torvalds * 37121da177e4SLinus Torvalds * A low-level filesystem can, if it choses, legally 37131da177e4SLinus Torvalds * do a 37141da177e4SLinus Torvalds * 37151da177e4SLinus Torvalds * if (!d_unhashed(dentry)) 37161da177e4SLinus Torvalds * return -EBUSY; 37171da177e4SLinus Torvalds * 37181da177e4SLinus Torvalds * if it cannot handle the case of removing a directory 37191da177e4SLinus Torvalds * that is still in use by something else.. 37201da177e4SLinus Torvalds */ 37211da177e4SLinus Torvalds void dentry_unhash(struct dentry *dentry) 37221da177e4SLinus Torvalds { 37231da177e4SLinus Torvalds shrink_dcache_parent(dentry); 37241da177e4SLinus Torvalds spin_lock(&dentry->d_lock); 372598474236SWaiman Long if (dentry->d_lockref.count == 1) 37261da177e4SLinus Torvalds __d_drop(dentry); 37271da177e4SLinus Torvalds spin_unlock(&dentry->d_lock); 37281da177e4SLinus Torvalds } 37294d359507SAl Viro EXPORT_SYMBOL(dentry_unhash); 37301da177e4SLinus Torvalds 37311da177e4SLinus Torvalds int vfs_rmdir(struct inode *dir, struct dentry *dentry) 37321da177e4SLinus Torvalds { 37331da177e4SLinus Torvalds int error = may_delete(dir, dentry, 1); 37341da177e4SLinus Torvalds 37351da177e4SLinus Torvalds if (error) 37361da177e4SLinus Torvalds return error; 37371da177e4SLinus Torvalds 3738acfa4380SAl Viro if (!dir->i_op->rmdir) 37391da177e4SLinus Torvalds return -EPERM; 37401da177e4SLinus Torvalds 37411d2ef590SAl Viro dget(dentry); 37425955102cSAl Viro inode_lock(dentry->d_inode); 3743912dbc15SSage Weil 37441da177e4SLinus Torvalds error = -EBUSY; 37457af1364fSEric W. Biederman if (is_local_mountpoint(dentry)) 3746912dbc15SSage Weil goto out; 3747912dbc15SSage Weil 37481da177e4SLinus Torvalds error = security_inode_rmdir(dir, dentry); 3749912dbc15SSage Weil if (error) 3750912dbc15SSage Weil goto out; 3751912dbc15SSage Weil 37523cebde24SSage Weil shrink_dcache_parent(dentry); 37531da177e4SLinus Torvalds error = dir->i_op->rmdir(dir, dentry); 3754912dbc15SSage Weil if (error) 3755912dbc15SSage Weil goto out; 3756912dbc15SSage Weil 37571da177e4SLinus Torvalds dentry->d_inode->i_flags |= S_DEAD; 3758d83c49f3SAl Viro dont_mount(dentry); 37598ed936b5SEric W. Biederman detach_mounts(dentry); 37601da177e4SLinus Torvalds 3761912dbc15SSage Weil out: 37625955102cSAl Viro inode_unlock(dentry->d_inode); 37631d2ef590SAl Viro dput(dentry); 3764912dbc15SSage Weil if (!error) 3765912dbc15SSage Weil d_delete(dentry); 37661da177e4SLinus Torvalds return error; 37671da177e4SLinus Torvalds } 37684d359507SAl Viro EXPORT_SYMBOL(vfs_rmdir); 37691da177e4SLinus Torvalds 37705590ff0dSUlrich Drepper static long do_rmdir(int dfd, const char __user *pathname) 37711da177e4SLinus Torvalds { 37721da177e4SLinus Torvalds int error = 0; 377391a27b2aSJeff Layton struct filename *name; 37741da177e4SLinus Torvalds struct dentry *dentry; 3775f5beed75SAl Viro struct path path; 3776f5beed75SAl Viro struct qstr last; 3777f5beed75SAl Viro int type; 3778c6ee9206SJeff Layton unsigned int lookup_flags = 0; 3779c6ee9206SJeff Layton retry: 3780f5beed75SAl Viro name = user_path_parent(dfd, pathname, 3781f5beed75SAl Viro &path, &last, &type, lookup_flags); 378291a27b2aSJeff Layton if (IS_ERR(name)) 378391a27b2aSJeff Layton return PTR_ERR(name); 37841da177e4SLinus Torvalds 3785f5beed75SAl Viro switch (type) { 37861da177e4SLinus Torvalds case LAST_DOTDOT: 37871da177e4SLinus Torvalds error = -ENOTEMPTY; 37881da177e4SLinus Torvalds goto exit1; 37891da177e4SLinus Torvalds case LAST_DOT: 37901da177e4SLinus Torvalds error = -EINVAL; 37911da177e4SLinus Torvalds goto exit1; 37921da177e4SLinus Torvalds case LAST_ROOT: 37931da177e4SLinus Torvalds error = -EBUSY; 37941da177e4SLinus Torvalds goto exit1; 37951da177e4SLinus Torvalds } 37960612d9fbSOGAWA Hirofumi 3797f5beed75SAl Viro error = mnt_want_write(path.mnt); 3798c30dabfeSJan Kara if (error) 3799c30dabfeSJan Kara goto exit1; 38000612d9fbSOGAWA Hirofumi 38015955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 3802f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 38031da177e4SLinus Torvalds error = PTR_ERR(dentry); 38046902d925SDave Hansen if (IS_ERR(dentry)) 38056902d925SDave Hansen goto exit2; 3806e6bc45d6STheodore Ts'o if (!dentry->d_inode) { 3807e6bc45d6STheodore Ts'o error = -ENOENT; 3808e6bc45d6STheodore Ts'o goto exit3; 3809e6bc45d6STheodore Ts'o } 3810f5beed75SAl Viro error = security_path_rmdir(&path, dentry); 3811be6d3e56SKentaro Takeda if (error) 3812c30dabfeSJan Kara goto exit3; 3813f5beed75SAl Viro error = vfs_rmdir(path.dentry->d_inode, dentry); 38140622753bSDave Hansen exit3: 38151da177e4SLinus Torvalds dput(dentry); 38166902d925SDave Hansen exit2: 38175955102cSAl Viro inode_unlock(path.dentry->d_inode); 3818f5beed75SAl Viro mnt_drop_write(path.mnt); 38191da177e4SLinus Torvalds exit1: 3820f5beed75SAl Viro path_put(&path); 38211da177e4SLinus Torvalds putname(name); 3822c6ee9206SJeff Layton if (retry_estale(error, lookup_flags)) { 3823c6ee9206SJeff Layton lookup_flags |= LOOKUP_REVAL; 3824c6ee9206SJeff Layton goto retry; 3825c6ee9206SJeff Layton } 38261da177e4SLinus Torvalds return error; 38271da177e4SLinus Torvalds } 38281da177e4SLinus Torvalds 38293cdad428SHeiko Carstens SYSCALL_DEFINE1(rmdir, const char __user *, pathname) 38305590ff0dSUlrich Drepper { 38315590ff0dSUlrich Drepper return do_rmdir(AT_FDCWD, pathname); 38325590ff0dSUlrich Drepper } 38335590ff0dSUlrich Drepper 3834b21996e3SJ. Bruce Fields /** 3835b21996e3SJ. Bruce Fields * vfs_unlink - unlink a filesystem object 3836b21996e3SJ. Bruce Fields * @dir: parent directory 3837b21996e3SJ. Bruce Fields * @dentry: victim 3838b21996e3SJ. Bruce Fields * @delegated_inode: returns victim inode, if the inode is delegated. 3839b21996e3SJ. Bruce Fields * 3840b21996e3SJ. Bruce Fields * The caller must hold dir->i_mutex. 3841b21996e3SJ. Bruce Fields * 3842b21996e3SJ. Bruce Fields * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and 3843b21996e3SJ. Bruce Fields * return a reference to the inode in delegated_inode. The caller 3844b21996e3SJ. Bruce Fields * should then break the delegation on that inode and retry. Because 3845b21996e3SJ. Bruce Fields * breaking a delegation may take a long time, the caller should drop 3846b21996e3SJ. Bruce Fields * dir->i_mutex before doing so. 3847b21996e3SJ. Bruce Fields * 3848b21996e3SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 3849b21996e3SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 3850b21996e3SJ. Bruce Fields * to be NFS exported. 3851b21996e3SJ. Bruce Fields */ 3852b21996e3SJ. Bruce Fields int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode) 38531da177e4SLinus Torvalds { 38549accbb97SJ. Bruce Fields struct inode *target = dentry->d_inode; 38551da177e4SLinus Torvalds int error = may_delete(dir, dentry, 0); 38561da177e4SLinus Torvalds 38571da177e4SLinus Torvalds if (error) 38581da177e4SLinus Torvalds return error; 38591da177e4SLinus Torvalds 3860acfa4380SAl Viro if (!dir->i_op->unlink) 38611da177e4SLinus Torvalds return -EPERM; 38621da177e4SLinus Torvalds 38635955102cSAl Viro inode_lock(target); 38648ed936b5SEric W. Biederman if (is_local_mountpoint(dentry)) 38651da177e4SLinus Torvalds error = -EBUSY; 38661da177e4SLinus Torvalds else { 38671da177e4SLinus Torvalds error = security_inode_unlink(dir, dentry); 3868bec1052eSAl Viro if (!error) { 38695a14696cSJ. Bruce Fields error = try_break_deleg(target, delegated_inode); 38705a14696cSJ. Bruce Fields if (error) 3871b21996e3SJ. Bruce Fields goto out; 38721da177e4SLinus Torvalds error = dir->i_op->unlink(dir, dentry); 38738ed936b5SEric W. Biederman if (!error) { 3874d83c49f3SAl Viro dont_mount(dentry); 38758ed936b5SEric W. Biederman detach_mounts(dentry); 38768ed936b5SEric W. Biederman } 3877bec1052eSAl Viro } 38781da177e4SLinus Torvalds } 3879b21996e3SJ. Bruce Fields out: 38805955102cSAl Viro inode_unlock(target); 38811da177e4SLinus Torvalds 38821da177e4SLinus Torvalds /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 38831da177e4SLinus Torvalds if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { 38849accbb97SJ. Bruce Fields fsnotify_link_count(target); 38851da177e4SLinus Torvalds d_delete(dentry); 38861da177e4SLinus Torvalds } 38870eeca283SRobert Love 38881da177e4SLinus Torvalds return error; 38891da177e4SLinus Torvalds } 38904d359507SAl Viro EXPORT_SYMBOL(vfs_unlink); 38911da177e4SLinus Torvalds 38921da177e4SLinus Torvalds /* 38931da177e4SLinus Torvalds * Make sure that the actual truncation of the file will occur outside its 38941b1dcc1bSJes Sorensen * directory's i_mutex. Truncate can take a long time if there is a lot of 38951da177e4SLinus Torvalds * writeout happening, and we don't want to prevent access to the directory 38961da177e4SLinus Torvalds * while waiting on the I/O. 38971da177e4SLinus Torvalds */ 38985590ff0dSUlrich Drepper static long do_unlinkat(int dfd, const char __user *pathname) 38991da177e4SLinus Torvalds { 39002ad94ae6SAl Viro int error; 390191a27b2aSJeff Layton struct filename *name; 39021da177e4SLinus Torvalds struct dentry *dentry; 3903f5beed75SAl Viro struct path path; 3904f5beed75SAl Viro struct qstr last; 3905f5beed75SAl Viro int type; 39061da177e4SLinus Torvalds struct inode *inode = NULL; 3907b21996e3SJ. Bruce Fields struct inode *delegated_inode = NULL; 39085d18f813SJeff Layton unsigned int lookup_flags = 0; 39095d18f813SJeff Layton retry: 3910f5beed75SAl Viro name = user_path_parent(dfd, pathname, 3911f5beed75SAl Viro &path, &last, &type, lookup_flags); 391291a27b2aSJeff Layton if (IS_ERR(name)) 391391a27b2aSJeff Layton return PTR_ERR(name); 39142ad94ae6SAl Viro 39151da177e4SLinus Torvalds error = -EISDIR; 3916f5beed75SAl Viro if (type != LAST_NORM) 39171da177e4SLinus Torvalds goto exit1; 39180612d9fbSOGAWA Hirofumi 3919f5beed75SAl Viro error = mnt_want_write(path.mnt); 3920c30dabfeSJan Kara if (error) 3921c30dabfeSJan Kara goto exit1; 3922b21996e3SJ. Bruce Fields retry_deleg: 39235955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 3924f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 39251da177e4SLinus Torvalds error = PTR_ERR(dentry); 39261da177e4SLinus Torvalds if (!IS_ERR(dentry)) { 39271da177e4SLinus Torvalds /* Why not before? Because we want correct error value */ 3928f5beed75SAl Viro if (last.name[last.len]) 392950338b88STörök Edwin goto slashes; 39301da177e4SLinus Torvalds inode = dentry->d_inode; 3931b18825a7SDavid Howells if (d_is_negative(dentry)) 3932e6bc45d6STheodore Ts'o goto slashes; 39337de9c6eeSAl Viro ihold(inode); 3934f5beed75SAl Viro error = security_path_unlink(&path, dentry); 3935be6d3e56SKentaro Takeda if (error) 3936c30dabfeSJan Kara goto exit2; 3937f5beed75SAl Viro error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode); 39381da177e4SLinus Torvalds exit2: 39391da177e4SLinus Torvalds dput(dentry); 39401da177e4SLinus Torvalds } 39415955102cSAl Viro inode_unlock(path.dentry->d_inode); 39421da177e4SLinus Torvalds if (inode) 39431da177e4SLinus Torvalds iput(inode); /* truncate the inode here */ 3944b21996e3SJ. Bruce Fields inode = NULL; 3945b21996e3SJ. Bruce Fields if (delegated_inode) { 39465a14696cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 3947b21996e3SJ. Bruce Fields if (!error) 3948b21996e3SJ. Bruce Fields goto retry_deleg; 3949b21996e3SJ. Bruce Fields } 3950f5beed75SAl Viro mnt_drop_write(path.mnt); 39511da177e4SLinus Torvalds exit1: 3952f5beed75SAl Viro path_put(&path); 39531da177e4SLinus Torvalds putname(name); 39545d18f813SJeff Layton if (retry_estale(error, lookup_flags)) { 39555d18f813SJeff Layton lookup_flags |= LOOKUP_REVAL; 39565d18f813SJeff Layton inode = NULL; 39575d18f813SJeff Layton goto retry; 39585d18f813SJeff Layton } 39591da177e4SLinus Torvalds return error; 39601da177e4SLinus Torvalds 39611da177e4SLinus Torvalds slashes: 3962b18825a7SDavid Howells if (d_is_negative(dentry)) 3963b18825a7SDavid Howells error = -ENOENT; 396444b1d530SMiklos Szeredi else if (d_is_dir(dentry)) 3965b18825a7SDavid Howells error = -EISDIR; 3966b18825a7SDavid Howells else 3967b18825a7SDavid Howells error = -ENOTDIR; 39681da177e4SLinus Torvalds goto exit2; 39691da177e4SLinus Torvalds } 39701da177e4SLinus Torvalds 39712e4d0924SHeiko Carstens SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) 39725590ff0dSUlrich Drepper { 39735590ff0dSUlrich Drepper if ((flag & ~AT_REMOVEDIR) != 0) 39745590ff0dSUlrich Drepper return -EINVAL; 39755590ff0dSUlrich Drepper 39765590ff0dSUlrich Drepper if (flag & AT_REMOVEDIR) 39775590ff0dSUlrich Drepper return do_rmdir(dfd, pathname); 39785590ff0dSUlrich Drepper 39795590ff0dSUlrich Drepper return do_unlinkat(dfd, pathname); 39805590ff0dSUlrich Drepper } 39815590ff0dSUlrich Drepper 39823480b257SHeiko Carstens SYSCALL_DEFINE1(unlink, const char __user *, pathname) 39835590ff0dSUlrich Drepper { 39845590ff0dSUlrich Drepper return do_unlinkat(AT_FDCWD, pathname); 39855590ff0dSUlrich Drepper } 39865590ff0dSUlrich Drepper 3987db2e747bSMiklos Szeredi int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) 39881da177e4SLinus Torvalds { 3989a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 39901da177e4SLinus Torvalds 39911da177e4SLinus Torvalds if (error) 39921da177e4SLinus Torvalds return error; 39931da177e4SLinus Torvalds 3994acfa4380SAl Viro if (!dir->i_op->symlink) 39951da177e4SLinus Torvalds return -EPERM; 39961da177e4SLinus Torvalds 39971da177e4SLinus Torvalds error = security_inode_symlink(dir, dentry, oldname); 39981da177e4SLinus Torvalds if (error) 39991da177e4SLinus Torvalds return error; 40001da177e4SLinus Torvalds 40011da177e4SLinus Torvalds error = dir->i_op->symlink(dir, dentry, oldname); 4002a74574aaSStephen Smalley if (!error) 4003f38aa942SAmy Griffis fsnotify_create(dir, dentry); 40041da177e4SLinus Torvalds return error; 40051da177e4SLinus Torvalds } 40064d359507SAl Viro EXPORT_SYMBOL(vfs_symlink); 40071da177e4SLinus Torvalds 40082e4d0924SHeiko Carstens SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, 40092e4d0924SHeiko Carstens int, newdfd, const char __user *, newname) 40101da177e4SLinus Torvalds { 40112ad94ae6SAl Viro int error; 401291a27b2aSJeff Layton struct filename *from; 40136902d925SDave Hansen struct dentry *dentry; 4014dae6ad8fSAl Viro struct path path; 4015f46d3567SJeff Layton unsigned int lookup_flags = 0; 40161da177e4SLinus Torvalds 40171da177e4SLinus Torvalds from = getname(oldname); 40181da177e4SLinus Torvalds if (IS_ERR(from)) 40191da177e4SLinus Torvalds return PTR_ERR(from); 4020f46d3567SJeff Layton retry: 4021f46d3567SJeff Layton dentry = user_path_create(newdfd, newname, &path, lookup_flags); 40221da177e4SLinus Torvalds error = PTR_ERR(dentry); 40236902d925SDave Hansen if (IS_ERR(dentry)) 4024dae6ad8fSAl Viro goto out_putname; 40256902d925SDave Hansen 402691a27b2aSJeff Layton error = security_path_symlink(&path, dentry, from->name); 4027a8104a9fSAl Viro if (!error) 402891a27b2aSJeff Layton error = vfs_symlink(path.dentry->d_inode, dentry, from->name); 4029921a1650SAl Viro done_path_create(&path, dentry); 4030f46d3567SJeff Layton if (retry_estale(error, lookup_flags)) { 4031f46d3567SJeff Layton lookup_flags |= LOOKUP_REVAL; 4032f46d3567SJeff Layton goto retry; 4033f46d3567SJeff Layton } 40346902d925SDave Hansen out_putname: 40351da177e4SLinus Torvalds putname(from); 40361da177e4SLinus Torvalds return error; 40371da177e4SLinus Torvalds } 40381da177e4SLinus Torvalds 40393480b257SHeiko Carstens SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) 40405590ff0dSUlrich Drepper { 40415590ff0dSUlrich Drepper return sys_symlinkat(oldname, AT_FDCWD, newname); 40425590ff0dSUlrich Drepper } 40435590ff0dSUlrich Drepper 4044146a8595SJ. Bruce Fields /** 4045146a8595SJ. Bruce Fields * vfs_link - create a new link 4046146a8595SJ. Bruce Fields * @old_dentry: object to be linked 4047146a8595SJ. Bruce Fields * @dir: new parent 4048146a8595SJ. Bruce Fields * @new_dentry: where to create the new link 4049146a8595SJ. Bruce Fields * @delegated_inode: returns inode needing a delegation break 4050146a8595SJ. Bruce Fields * 4051146a8595SJ. Bruce Fields * The caller must hold dir->i_mutex 4052146a8595SJ. Bruce Fields * 4053146a8595SJ. Bruce Fields * If vfs_link discovers a delegation on the to-be-linked file in need 4054146a8595SJ. Bruce Fields * of breaking, it will return -EWOULDBLOCK and return a reference to the 4055146a8595SJ. Bruce Fields * inode in delegated_inode. The caller should then break the delegation 4056146a8595SJ. Bruce Fields * and retry. Because breaking a delegation may take a long time, the 4057146a8595SJ. Bruce Fields * caller should drop the i_mutex before doing so. 4058146a8595SJ. Bruce Fields * 4059146a8595SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 4060146a8595SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 4061146a8595SJ. Bruce Fields * to be NFS exported. 4062146a8595SJ. Bruce Fields */ 4063146a8595SJ. Bruce Fields int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode) 40641da177e4SLinus Torvalds { 40651da177e4SLinus Torvalds struct inode *inode = old_dentry->d_inode; 40668de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 40671da177e4SLinus Torvalds int error; 40681da177e4SLinus Torvalds 40691da177e4SLinus Torvalds if (!inode) 40701da177e4SLinus Torvalds return -ENOENT; 40711da177e4SLinus Torvalds 4072a95164d9SMiklos Szeredi error = may_create(dir, new_dentry); 40731da177e4SLinus Torvalds if (error) 40741da177e4SLinus Torvalds return error; 40751da177e4SLinus Torvalds 40761da177e4SLinus Torvalds if (dir->i_sb != inode->i_sb) 40771da177e4SLinus Torvalds return -EXDEV; 40781da177e4SLinus Torvalds 40791da177e4SLinus Torvalds /* 40801da177e4SLinus Torvalds * A link to an append-only or immutable file cannot be created. 40811da177e4SLinus Torvalds */ 40821da177e4SLinus Torvalds if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 40831da177e4SLinus Torvalds return -EPERM; 4084acfa4380SAl Viro if (!dir->i_op->link) 40851da177e4SLinus Torvalds return -EPERM; 40867e79eedbSTetsuo Handa if (S_ISDIR(inode->i_mode)) 40871da177e4SLinus Torvalds return -EPERM; 40881da177e4SLinus Torvalds 40891da177e4SLinus Torvalds error = security_inode_link(old_dentry, dir, new_dentry); 40901da177e4SLinus Torvalds if (error) 40911da177e4SLinus Torvalds return error; 40921da177e4SLinus Torvalds 40935955102cSAl Viro inode_lock(inode); 4094aae8a97dSAneesh Kumar K.V /* Make sure we don't allow creating hardlink to an unlinked file */ 4095f4e0c30cSAl Viro if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) 4096aae8a97dSAneesh Kumar K.V error = -ENOENT; 40978de52778SAl Viro else if (max_links && inode->i_nlink >= max_links) 40988de52778SAl Viro error = -EMLINK; 4099146a8595SJ. Bruce Fields else { 4100146a8595SJ. Bruce Fields error = try_break_deleg(inode, delegated_inode); 4101146a8595SJ. Bruce Fields if (!error) 41021da177e4SLinus Torvalds error = dir->i_op->link(old_dentry, dir, new_dentry); 4103146a8595SJ. Bruce Fields } 4104f4e0c30cSAl Viro 4105f4e0c30cSAl Viro if (!error && (inode->i_state & I_LINKABLE)) { 4106f4e0c30cSAl Viro spin_lock(&inode->i_lock); 4107f4e0c30cSAl Viro inode->i_state &= ~I_LINKABLE; 4108f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 4109f4e0c30cSAl Viro } 41105955102cSAl Viro inode_unlock(inode); 4111e31e14ecSStephen Smalley if (!error) 41127e79eedbSTetsuo Handa fsnotify_link(dir, inode, new_dentry); 41131da177e4SLinus Torvalds return error; 41141da177e4SLinus Torvalds } 41154d359507SAl Viro EXPORT_SYMBOL(vfs_link); 41161da177e4SLinus Torvalds 41171da177e4SLinus Torvalds /* 41181da177e4SLinus Torvalds * Hardlinks are often used in delicate situations. We avoid 41191da177e4SLinus Torvalds * security-related surprises by not following symlinks on the 41201da177e4SLinus Torvalds * newname. --KAB 41211da177e4SLinus Torvalds * 41221da177e4SLinus Torvalds * We don't follow them on the oldname either to be compatible 41231da177e4SLinus Torvalds * with linux 2.0, and to avoid hard-linking to directories 41241da177e4SLinus Torvalds * and other special files. --ADM 41251da177e4SLinus Torvalds */ 41262e4d0924SHeiko Carstens SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, 41272e4d0924SHeiko Carstens int, newdfd, const char __user *, newname, int, flags) 41281da177e4SLinus Torvalds { 41291da177e4SLinus Torvalds struct dentry *new_dentry; 4130dae6ad8fSAl Viro struct path old_path, new_path; 4131146a8595SJ. Bruce Fields struct inode *delegated_inode = NULL; 413211a7b371SAneesh Kumar K.V int how = 0; 41331da177e4SLinus Torvalds int error; 41341da177e4SLinus Torvalds 413511a7b371SAneesh Kumar K.V if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) 4136c04030e1SUlrich Drepper return -EINVAL; 413711a7b371SAneesh Kumar K.V /* 4138f0cc6ffbSLinus Torvalds * To use null names we require CAP_DAC_READ_SEARCH 4139f0cc6ffbSLinus Torvalds * This ensures that not everyone will be able to create 4140f0cc6ffbSLinus Torvalds * handlink using the passed filedescriptor. 414111a7b371SAneesh Kumar K.V */ 4142f0cc6ffbSLinus Torvalds if (flags & AT_EMPTY_PATH) { 4143f0cc6ffbSLinus Torvalds if (!capable(CAP_DAC_READ_SEARCH)) 4144f0cc6ffbSLinus Torvalds return -ENOENT; 414511a7b371SAneesh Kumar K.V how = LOOKUP_EMPTY; 4146f0cc6ffbSLinus Torvalds } 4147c04030e1SUlrich Drepper 414811a7b371SAneesh Kumar K.V if (flags & AT_SYMLINK_FOLLOW) 414911a7b371SAneesh Kumar K.V how |= LOOKUP_FOLLOW; 4150442e31caSJeff Layton retry: 415111a7b371SAneesh Kumar K.V error = user_path_at(olddfd, oldname, how, &old_path); 41521da177e4SLinus Torvalds if (error) 41532ad94ae6SAl Viro return error; 41542ad94ae6SAl Viro 4155442e31caSJeff Layton new_dentry = user_path_create(newdfd, newname, &new_path, 4156442e31caSJeff Layton (how & LOOKUP_REVAL)); 41571da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 41586902d925SDave Hansen if (IS_ERR(new_dentry)) 4159dae6ad8fSAl Viro goto out; 4160dae6ad8fSAl Viro 4161dae6ad8fSAl Viro error = -EXDEV; 4162dae6ad8fSAl Viro if (old_path.mnt != new_path.mnt) 4163dae6ad8fSAl Viro goto out_dput; 4164800179c9SKees Cook error = may_linkat(&old_path); 4165800179c9SKees Cook if (unlikely(error)) 4166800179c9SKees Cook goto out_dput; 4167dae6ad8fSAl Viro error = security_path_link(old_path.dentry, &new_path, new_dentry); 4168be6d3e56SKentaro Takeda if (error) 4169a8104a9fSAl Viro goto out_dput; 4170146a8595SJ. Bruce Fields error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode); 417175c3f29dSDave Hansen out_dput: 4172921a1650SAl Viro done_path_create(&new_path, new_dentry); 4173146a8595SJ. Bruce Fields if (delegated_inode) { 4174146a8595SJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4175d22e6338SOleg Drokin if (!error) { 4176d22e6338SOleg Drokin path_put(&old_path); 4177146a8595SJ. Bruce Fields goto retry; 4178146a8595SJ. Bruce Fields } 4179d22e6338SOleg Drokin } 4180442e31caSJeff Layton if (retry_estale(error, how)) { 4181d22e6338SOleg Drokin path_put(&old_path); 4182442e31caSJeff Layton how |= LOOKUP_REVAL; 4183442e31caSJeff Layton goto retry; 4184442e31caSJeff Layton } 41851da177e4SLinus Torvalds out: 41862d8f3038SAl Viro path_put(&old_path); 41871da177e4SLinus Torvalds 41881da177e4SLinus Torvalds return error; 41891da177e4SLinus Torvalds } 41901da177e4SLinus Torvalds 41913480b257SHeiko Carstens SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) 41925590ff0dSUlrich Drepper { 4193c04030e1SUlrich Drepper return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 41945590ff0dSUlrich Drepper } 41955590ff0dSUlrich Drepper 4196bc27027aSMiklos Szeredi /** 4197bc27027aSMiklos Szeredi * vfs_rename - rename a filesystem object 4198bc27027aSMiklos Szeredi * @old_dir: parent of source 4199bc27027aSMiklos Szeredi * @old_dentry: source 4200bc27027aSMiklos Szeredi * @new_dir: parent of destination 4201bc27027aSMiklos Szeredi * @new_dentry: destination 4202bc27027aSMiklos Szeredi * @delegated_inode: returns an inode needing a delegation break 4203520c8b16SMiklos Szeredi * @flags: rename flags 4204bc27027aSMiklos Szeredi * 4205bc27027aSMiklos Szeredi * The caller must hold multiple mutexes--see lock_rename()). 4206bc27027aSMiklos Szeredi * 4207bc27027aSMiklos Szeredi * If vfs_rename discovers a delegation in need of breaking at either 4208bc27027aSMiklos Szeredi * the source or destination, it will return -EWOULDBLOCK and return a 4209bc27027aSMiklos Szeredi * reference to the inode in delegated_inode. The caller should then 4210bc27027aSMiklos Szeredi * break the delegation and retry. Because breaking a delegation may 4211bc27027aSMiklos Szeredi * take a long time, the caller should drop all locks before doing 4212bc27027aSMiklos Szeredi * so. 4213bc27027aSMiklos Szeredi * 4214bc27027aSMiklos Szeredi * Alternatively, a caller may pass NULL for delegated_inode. This may 4215bc27027aSMiklos Szeredi * be appropriate for callers that expect the underlying filesystem not 4216bc27027aSMiklos Szeredi * to be NFS exported. 4217bc27027aSMiklos Szeredi * 42181da177e4SLinus Torvalds * The worst of all namespace operations - renaming directory. "Perverted" 42191da177e4SLinus Torvalds * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 42201da177e4SLinus Torvalds * Problems: 4221d03b29a2SJ. Bruce Fields * a) we can get into loop creation. 42221da177e4SLinus Torvalds * b) race potential - two innocent renames can create a loop together. 42231da177e4SLinus Torvalds * That's where 4.4 screws up. Current fix: serialization on 4224a11f3a05SArjan van de Ven * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 42251da177e4SLinus Torvalds * story. 42266cedba89SJ. Bruce Fields * c) we have to lock _four_ objects - parents and victim (if it exists), 42276cedba89SJ. Bruce Fields * and source (if it is not a directory). 42281b1dcc1bSJes Sorensen * And that - after we got ->i_mutex on parents (until then we don't know 42291da177e4SLinus Torvalds * whether the target exists). Solution: try to be smart with locking 42301da177e4SLinus Torvalds * order for inodes. We rely on the fact that tree topology may change 4231a11f3a05SArjan van de Ven * only under ->s_vfs_rename_mutex _and_ that parent of the object we 42321da177e4SLinus Torvalds * move will be locked. Thus we can rank directories by the tree 42331da177e4SLinus Torvalds * (ancestors first) and rank all non-directories after them. 42341da177e4SLinus Torvalds * That works since everybody except rename does "lock parent, lookup, 4235a11f3a05SArjan van de Ven * lock child" and rename is under ->s_vfs_rename_mutex. 42361da177e4SLinus Torvalds * HOWEVER, it relies on the assumption that any object with ->lookup() 42371da177e4SLinus Torvalds * has no more than 1 dentry. If "hybrid" objects will ever appear, 42381da177e4SLinus Torvalds * we'd better make sure that there's no link(2) for them. 4239e4eaac06SSage Weil * d) conversion from fhandle to dentry may come in the wrong moment - when 42401b1dcc1bSJes Sorensen * we are removing the target. Solution: we will have to grab ->i_mutex 42411da177e4SLinus Torvalds * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 4242c41b20e7SAdam Buchbinder * ->i_mutex on parents, which works but leads to some truly excessive 42431da177e4SLinus Torvalds * locking]. 42441da177e4SLinus Torvalds */ 42451da177e4SLinus Torvalds int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, 42468e6d782cSJ. Bruce Fields struct inode *new_dir, struct dentry *new_dentry, 4247520c8b16SMiklos Szeredi struct inode **delegated_inode, unsigned int flags) 42481da177e4SLinus Torvalds { 42491da177e4SLinus Torvalds int error; 4250bc27027aSMiklos Szeredi bool is_dir = d_is_dir(old_dentry); 425159b0df21SEric Paris const unsigned char *old_name; 4252bc27027aSMiklos Szeredi struct inode *source = old_dentry->d_inode; 4253bc27027aSMiklos Szeredi struct inode *target = new_dentry->d_inode; 4254da1ce067SMiklos Szeredi bool new_is_dir = false; 4255da1ce067SMiklos Szeredi unsigned max_links = new_dir->i_sb->s_max_links; 42561da177e4SLinus Torvalds 4257bc27027aSMiklos Szeredi if (source == target) 42581da177e4SLinus Torvalds return 0; 42591da177e4SLinus Torvalds 42601da177e4SLinus Torvalds error = may_delete(old_dir, old_dentry, is_dir); 42611da177e4SLinus Torvalds if (error) 42621da177e4SLinus Torvalds return error; 42631da177e4SLinus Torvalds 4264da1ce067SMiklos Szeredi if (!target) { 4265a95164d9SMiklos Szeredi error = may_create(new_dir, new_dentry); 4266da1ce067SMiklos Szeredi } else { 4267da1ce067SMiklos Szeredi new_is_dir = d_is_dir(new_dentry); 4268da1ce067SMiklos Szeredi 4269da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 42701da177e4SLinus Torvalds error = may_delete(new_dir, new_dentry, is_dir); 4271da1ce067SMiklos Szeredi else 4272da1ce067SMiklos Szeredi error = may_delete(new_dir, new_dentry, new_is_dir); 4273da1ce067SMiklos Szeredi } 42741da177e4SLinus Torvalds if (error) 42751da177e4SLinus Torvalds return error; 42761da177e4SLinus Torvalds 42777177a9c4SMiklos Szeredi if (!old_dir->i_op->rename && !old_dir->i_op->rename2) 42781da177e4SLinus Torvalds return -EPERM; 42791da177e4SLinus Torvalds 4280520c8b16SMiklos Szeredi if (flags && !old_dir->i_op->rename2) 4281520c8b16SMiklos Szeredi return -EINVAL; 4282520c8b16SMiklos Szeredi 4283bc27027aSMiklos Szeredi /* 4284bc27027aSMiklos Szeredi * If we are going to change the parent - check write permissions, 4285bc27027aSMiklos Szeredi * we'll need to flip '..'. 4286bc27027aSMiklos Szeredi */ 4287da1ce067SMiklos Szeredi if (new_dir != old_dir) { 4288da1ce067SMiklos Szeredi if (is_dir) { 4289bc27027aSMiklos Szeredi error = inode_permission(source, MAY_WRITE); 4290bc27027aSMiklos Szeredi if (error) 4291bc27027aSMiklos Szeredi return error; 4292bc27027aSMiklos Szeredi } 4293da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && new_is_dir) { 4294da1ce067SMiklos Szeredi error = inode_permission(target, MAY_WRITE); 4295da1ce067SMiklos Szeredi if (error) 4296da1ce067SMiklos Szeredi return error; 4297da1ce067SMiklos Szeredi } 4298da1ce067SMiklos Szeredi } 42990eeca283SRobert Love 43000b3974ebSMiklos Szeredi error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry, 43010b3974ebSMiklos Szeredi flags); 4302bc27027aSMiklos Szeredi if (error) 4303bc27027aSMiklos Szeredi return error; 4304bc27027aSMiklos Szeredi 4305bc27027aSMiklos Szeredi old_name = fsnotify_oldname_init(old_dentry->d_name.name); 4306bc27027aSMiklos Szeredi dget(new_dentry); 4307da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4308bc27027aSMiklos Szeredi lock_two_nondirectories(source, target); 4309bc27027aSMiklos Szeredi else if (target) 43105955102cSAl Viro inode_lock(target); 4311bc27027aSMiklos Szeredi 4312bc27027aSMiklos Szeredi error = -EBUSY; 43137af1364fSEric W. Biederman if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry)) 4314bc27027aSMiklos Szeredi goto out; 4315bc27027aSMiklos Szeredi 4316da1ce067SMiklos Szeredi if (max_links && new_dir != old_dir) { 4317bc27027aSMiklos Szeredi error = -EMLINK; 4318da1ce067SMiklos Szeredi if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links) 4319bc27027aSMiklos Szeredi goto out; 4320da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir && 4321da1ce067SMiklos Szeredi old_dir->i_nlink >= max_links) 4322da1ce067SMiklos Szeredi goto out; 4323da1ce067SMiklos Szeredi } 4324da1ce067SMiklos Szeredi if (is_dir && !(flags & RENAME_EXCHANGE) && target) 4325bc27027aSMiklos Szeredi shrink_dcache_parent(new_dentry); 4326da1ce067SMiklos Szeredi if (!is_dir) { 4327bc27027aSMiklos Szeredi error = try_break_deleg(source, delegated_inode); 4328bc27027aSMiklos Szeredi if (error) 4329bc27027aSMiklos Szeredi goto out; 4330da1ce067SMiklos Szeredi } 4331da1ce067SMiklos Szeredi if (target && !new_is_dir) { 4332bc27027aSMiklos Szeredi error = try_break_deleg(target, delegated_inode); 4333bc27027aSMiklos Szeredi if (error) 4334bc27027aSMiklos Szeredi goto out; 4335bc27027aSMiklos Szeredi } 43367177a9c4SMiklos Szeredi if (!old_dir->i_op->rename2) { 4337520c8b16SMiklos Szeredi error = old_dir->i_op->rename(old_dir, old_dentry, 4338520c8b16SMiklos Szeredi new_dir, new_dentry); 4339520c8b16SMiklos Szeredi } else { 43407177a9c4SMiklos Szeredi WARN_ON(old_dir->i_op->rename != NULL); 4341520c8b16SMiklos Szeredi error = old_dir->i_op->rename2(old_dir, old_dentry, 4342520c8b16SMiklos Szeredi new_dir, new_dentry, flags); 4343520c8b16SMiklos Szeredi } 4344bc27027aSMiklos Szeredi if (error) 4345bc27027aSMiklos Szeredi goto out; 4346bc27027aSMiklos Szeredi 4347da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE) && target) { 43481da177e4SLinus Torvalds if (is_dir) 4349bc27027aSMiklos Szeredi target->i_flags |= S_DEAD; 4350bc27027aSMiklos Szeredi dont_mount(new_dentry); 43518ed936b5SEric W. Biederman detach_mounts(new_dentry); 4352bc27027aSMiklos Szeredi } 4353da1ce067SMiklos Szeredi if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) { 4354da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 4355bc27027aSMiklos Szeredi d_move(old_dentry, new_dentry); 4356da1ce067SMiklos Szeredi else 4357da1ce067SMiklos Szeredi d_exchange(old_dentry, new_dentry); 4358da1ce067SMiklos Szeredi } 4359bc27027aSMiklos Szeredi out: 4360da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4361bc27027aSMiklos Szeredi unlock_two_nondirectories(source, target); 4362bc27027aSMiklos Szeredi else if (target) 43635955102cSAl Viro inode_unlock(target); 4364bc27027aSMiklos Szeredi dput(new_dentry); 4365da1ce067SMiklos Szeredi if (!error) { 4366123df294SAl Viro fsnotify_move(old_dir, new_dir, old_name, is_dir, 4367da1ce067SMiklos Szeredi !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry); 4368da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4369da1ce067SMiklos Szeredi fsnotify_move(new_dir, old_dir, old_dentry->d_name.name, 4370da1ce067SMiklos Szeredi new_is_dir, NULL, new_dentry); 4371da1ce067SMiklos Szeredi } 4372da1ce067SMiklos Szeredi } 43730eeca283SRobert Love fsnotify_oldname_free(old_name); 43740eeca283SRobert Love 43751da177e4SLinus Torvalds return error; 43761da177e4SLinus Torvalds } 43774d359507SAl Viro EXPORT_SYMBOL(vfs_rename); 43781da177e4SLinus Torvalds 4379520c8b16SMiklos Szeredi SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname, 4380520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname, unsigned int, flags) 43811da177e4SLinus Torvalds { 43821da177e4SLinus Torvalds struct dentry *old_dentry, *new_dentry; 43831da177e4SLinus Torvalds struct dentry *trap; 4384f5beed75SAl Viro struct path old_path, new_path; 4385f5beed75SAl Viro struct qstr old_last, new_last; 4386f5beed75SAl Viro int old_type, new_type; 43878e6d782cSJ. Bruce Fields struct inode *delegated_inode = NULL; 438891a27b2aSJeff Layton struct filename *from; 438991a27b2aSJeff Layton struct filename *to; 4390f5beed75SAl Viro unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET; 4391c6a94284SJeff Layton bool should_retry = false; 43922ad94ae6SAl Viro int error; 4393520c8b16SMiklos Szeredi 43940d7a8555SMiklos Szeredi if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) 4395da1ce067SMiklos Szeredi return -EINVAL; 4396da1ce067SMiklos Szeredi 43970d7a8555SMiklos Szeredi if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) && 43980d7a8555SMiklos Szeredi (flags & RENAME_EXCHANGE)) 4399520c8b16SMiklos Szeredi return -EINVAL; 4400520c8b16SMiklos Szeredi 44010d7a8555SMiklos Szeredi if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD)) 44020d7a8555SMiklos Szeredi return -EPERM; 44030d7a8555SMiklos Szeredi 4404f5beed75SAl Viro if (flags & RENAME_EXCHANGE) 4405f5beed75SAl Viro target_flags = 0; 4406f5beed75SAl Viro 4407c6a94284SJeff Layton retry: 4408f5beed75SAl Viro from = user_path_parent(olddfd, oldname, 4409f5beed75SAl Viro &old_path, &old_last, &old_type, lookup_flags); 441091a27b2aSJeff Layton if (IS_ERR(from)) { 441191a27b2aSJeff Layton error = PTR_ERR(from); 44121da177e4SLinus Torvalds goto exit; 441391a27b2aSJeff Layton } 44141da177e4SLinus Torvalds 4415f5beed75SAl Viro to = user_path_parent(newdfd, newname, 4416f5beed75SAl Viro &new_path, &new_last, &new_type, lookup_flags); 441791a27b2aSJeff Layton if (IS_ERR(to)) { 441891a27b2aSJeff Layton error = PTR_ERR(to); 44191da177e4SLinus Torvalds goto exit1; 442091a27b2aSJeff Layton } 44211da177e4SLinus Torvalds 44221da177e4SLinus Torvalds error = -EXDEV; 4423f5beed75SAl Viro if (old_path.mnt != new_path.mnt) 44241da177e4SLinus Torvalds goto exit2; 44251da177e4SLinus Torvalds 44261da177e4SLinus Torvalds error = -EBUSY; 4427f5beed75SAl Viro if (old_type != LAST_NORM) 44281da177e4SLinus Torvalds goto exit2; 44291da177e4SLinus Torvalds 44300a7c3937SMiklos Szeredi if (flags & RENAME_NOREPLACE) 44310a7c3937SMiklos Szeredi error = -EEXIST; 4432f5beed75SAl Viro if (new_type != LAST_NORM) 44331da177e4SLinus Torvalds goto exit2; 44341da177e4SLinus Torvalds 4435f5beed75SAl Viro error = mnt_want_write(old_path.mnt); 4436c30dabfeSJan Kara if (error) 4437c30dabfeSJan Kara goto exit2; 4438c30dabfeSJan Kara 44398e6d782cSJ. Bruce Fields retry_deleg: 4440f5beed75SAl Viro trap = lock_rename(new_path.dentry, old_path.dentry); 44411da177e4SLinus Torvalds 4442f5beed75SAl Viro old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags); 44431da177e4SLinus Torvalds error = PTR_ERR(old_dentry); 44441da177e4SLinus Torvalds if (IS_ERR(old_dentry)) 44451da177e4SLinus Torvalds goto exit3; 44461da177e4SLinus Torvalds /* source must exist */ 44471da177e4SLinus Torvalds error = -ENOENT; 4448b18825a7SDavid Howells if (d_is_negative(old_dentry)) 44491da177e4SLinus Torvalds goto exit4; 4450f5beed75SAl Viro new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags); 44511da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 44521da177e4SLinus Torvalds if (IS_ERR(new_dentry)) 44531da177e4SLinus Torvalds goto exit4; 44540a7c3937SMiklos Szeredi error = -EEXIST; 44550a7c3937SMiklos Szeredi if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry)) 44560a7c3937SMiklos Szeredi goto exit5; 4457da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4458da1ce067SMiklos Szeredi error = -ENOENT; 4459da1ce067SMiklos Szeredi if (d_is_negative(new_dentry)) 4460da1ce067SMiklos Szeredi goto exit5; 4461da1ce067SMiklos Szeredi 4462da1ce067SMiklos Szeredi if (!d_is_dir(new_dentry)) { 4463da1ce067SMiklos Szeredi error = -ENOTDIR; 4464f5beed75SAl Viro if (new_last.name[new_last.len]) 4465da1ce067SMiklos Szeredi goto exit5; 4466da1ce067SMiklos Szeredi } 4467da1ce067SMiklos Szeredi } 44680a7c3937SMiklos Szeredi /* unless the source is a directory trailing slashes give -ENOTDIR */ 44690a7c3937SMiklos Szeredi if (!d_is_dir(old_dentry)) { 44700a7c3937SMiklos Szeredi error = -ENOTDIR; 4471f5beed75SAl Viro if (old_last.name[old_last.len]) 44720a7c3937SMiklos Szeredi goto exit5; 4473f5beed75SAl Viro if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len]) 44740a7c3937SMiklos Szeredi goto exit5; 44750a7c3937SMiklos Szeredi } 44760a7c3937SMiklos Szeredi /* source should not be ancestor of target */ 44770a7c3937SMiklos Szeredi error = -EINVAL; 44780a7c3937SMiklos Szeredi if (old_dentry == trap) 44790a7c3937SMiklos Szeredi goto exit5; 44801da177e4SLinus Torvalds /* target should not be an ancestor of source */ 4481da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 44821da177e4SLinus Torvalds error = -ENOTEMPTY; 44831da177e4SLinus Torvalds if (new_dentry == trap) 44841da177e4SLinus Torvalds goto exit5; 44851da177e4SLinus Torvalds 4486f5beed75SAl Viro error = security_path_rename(&old_path, old_dentry, 4487f5beed75SAl Viro &new_path, new_dentry, flags); 4488be6d3e56SKentaro Takeda if (error) 4489c30dabfeSJan Kara goto exit5; 4490f5beed75SAl Viro error = vfs_rename(old_path.dentry->d_inode, old_dentry, 4491f5beed75SAl Viro new_path.dentry->d_inode, new_dentry, 4492520c8b16SMiklos Szeredi &delegated_inode, flags); 44931da177e4SLinus Torvalds exit5: 44941da177e4SLinus Torvalds dput(new_dentry); 44951da177e4SLinus Torvalds exit4: 44961da177e4SLinus Torvalds dput(old_dentry); 44971da177e4SLinus Torvalds exit3: 4498f5beed75SAl Viro unlock_rename(new_path.dentry, old_path.dentry); 44998e6d782cSJ. Bruce Fields if (delegated_inode) { 45008e6d782cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 45018e6d782cSJ. Bruce Fields if (!error) 45028e6d782cSJ. Bruce Fields goto retry_deleg; 45038e6d782cSJ. Bruce Fields } 4504f5beed75SAl Viro mnt_drop_write(old_path.mnt); 45051da177e4SLinus Torvalds exit2: 4506c6a94284SJeff Layton if (retry_estale(error, lookup_flags)) 4507c6a94284SJeff Layton should_retry = true; 4508f5beed75SAl Viro path_put(&new_path); 45092ad94ae6SAl Viro putname(to); 45101da177e4SLinus Torvalds exit1: 4511f5beed75SAl Viro path_put(&old_path); 45121da177e4SLinus Torvalds putname(from); 4513c6a94284SJeff Layton if (should_retry) { 4514c6a94284SJeff Layton should_retry = false; 4515c6a94284SJeff Layton lookup_flags |= LOOKUP_REVAL; 4516c6a94284SJeff Layton goto retry; 4517c6a94284SJeff Layton } 45182ad94ae6SAl Viro exit: 45191da177e4SLinus Torvalds return error; 45201da177e4SLinus Torvalds } 45211da177e4SLinus Torvalds 4522520c8b16SMiklos Szeredi SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, 4523520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname) 4524520c8b16SMiklos Szeredi { 4525520c8b16SMiklos Szeredi return sys_renameat2(olddfd, oldname, newdfd, newname, 0); 4526520c8b16SMiklos Szeredi } 4527520c8b16SMiklos Szeredi 4528a26eab24SHeiko Carstens SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) 45295590ff0dSUlrich Drepper { 4530520c8b16SMiklos Szeredi return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 45315590ff0dSUlrich Drepper } 45325590ff0dSUlrich Drepper 4533787fb6bcSMiklos Szeredi int vfs_whiteout(struct inode *dir, struct dentry *dentry) 4534787fb6bcSMiklos Szeredi { 4535787fb6bcSMiklos Szeredi int error = may_create(dir, dentry); 4536787fb6bcSMiklos Szeredi if (error) 4537787fb6bcSMiklos Szeredi return error; 4538787fb6bcSMiklos Szeredi 4539787fb6bcSMiklos Szeredi if (!dir->i_op->mknod) 4540787fb6bcSMiklos Szeredi return -EPERM; 4541787fb6bcSMiklos Szeredi 4542787fb6bcSMiklos Szeredi return dir->i_op->mknod(dir, dentry, 4543787fb6bcSMiklos Szeredi S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV); 4544787fb6bcSMiklos Szeredi } 4545787fb6bcSMiklos Szeredi EXPORT_SYMBOL(vfs_whiteout); 4546787fb6bcSMiklos Szeredi 45475d826c84SAl Viro int readlink_copy(char __user *buffer, int buflen, const char *link) 45481da177e4SLinus Torvalds { 45495d826c84SAl Viro int len = PTR_ERR(link); 45501da177e4SLinus Torvalds if (IS_ERR(link)) 45511da177e4SLinus Torvalds goto out; 45521da177e4SLinus Torvalds 45531da177e4SLinus Torvalds len = strlen(link); 45541da177e4SLinus Torvalds if (len > (unsigned) buflen) 45551da177e4SLinus Torvalds len = buflen; 45561da177e4SLinus Torvalds if (copy_to_user(buffer, link, len)) 45571da177e4SLinus Torvalds len = -EFAULT; 45581da177e4SLinus Torvalds out: 45591da177e4SLinus Torvalds return len; 45601da177e4SLinus Torvalds } 45615d826c84SAl Viro EXPORT_SYMBOL(readlink_copy); 45621da177e4SLinus Torvalds 45631da177e4SLinus Torvalds /* 45641da177e4SLinus Torvalds * A helper for ->readlink(). This should be used *ONLY* for symlinks that 45656b255391SAl Viro * have ->get_link() not calling nd_jump_link(). Using (or not using) it 45666b255391SAl Viro * for any given inode is up to filesystem. 45671da177e4SLinus Torvalds */ 45681da177e4SLinus Torvalds int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen) 45691da177e4SLinus Torvalds { 4570fceef393SAl Viro DEFINE_DELAYED_CALL(done); 45715f2c4179SAl Viro struct inode *inode = d_inode(dentry); 45725f2c4179SAl Viro const char *link = inode->i_link; 4573694a1764SMarcin Slusarz int res; 4574cc314eefSLinus Torvalds 4575d4dee48bSAl Viro if (!link) { 4576fceef393SAl Viro link = inode->i_op->get_link(dentry, inode, &done); 4577680baacbSAl Viro if (IS_ERR(link)) 4578680baacbSAl Viro return PTR_ERR(link); 4579d4dee48bSAl Viro } 4580680baacbSAl Viro res = readlink_copy(buffer, buflen, link); 4581fceef393SAl Viro do_delayed_call(&done); 4582694a1764SMarcin Slusarz return res; 45831da177e4SLinus Torvalds } 45844d359507SAl Viro EXPORT_SYMBOL(generic_readlink); 45851da177e4SLinus Torvalds 45861da177e4SLinus Torvalds /* get the link contents into pagecache */ 45876b255391SAl Viro const char *page_get_link(struct dentry *dentry, struct inode *inode, 4588fceef393SAl Viro struct delayed_call *callback) 45891da177e4SLinus Torvalds { 4590ebd09abbSDuane Griffin char *kaddr; 45911da177e4SLinus Torvalds struct page *page; 45926b255391SAl Viro struct address_space *mapping = inode->i_mapping; 45936b255391SAl Viro 4594d3883d4fSAl Viro if (!dentry) { 4595d3883d4fSAl Viro page = find_get_page(mapping, 0); 4596d3883d4fSAl Viro if (!page) 45976b255391SAl Viro return ERR_PTR(-ECHILD); 4598d3883d4fSAl Viro if (!PageUptodate(page)) { 4599d3883d4fSAl Viro put_page(page); 4600d3883d4fSAl Viro return ERR_PTR(-ECHILD); 4601d3883d4fSAl Viro } 4602d3883d4fSAl Viro } else { 4603090d2b18SPekka Enberg page = read_mapping_page(mapping, 0, NULL); 46041da177e4SLinus Torvalds if (IS_ERR(page)) 46056fe6900eSNick Piggin return (char*)page; 4606d3883d4fSAl Viro } 4607fceef393SAl Viro set_delayed_call(callback, page_put_link, page); 460821fc61c7SAl Viro BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM); 460921fc61c7SAl Viro kaddr = page_address(page); 46106b255391SAl Viro nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1); 4611ebd09abbSDuane Griffin return kaddr; 46121da177e4SLinus Torvalds } 46131da177e4SLinus Torvalds 46146b255391SAl Viro EXPORT_SYMBOL(page_get_link); 46151da177e4SLinus Torvalds 4616fceef393SAl Viro void page_put_link(void *arg) 46171da177e4SLinus Torvalds { 4618fceef393SAl Viro put_page(arg); 46191da177e4SLinus Torvalds } 46204d359507SAl Viro EXPORT_SYMBOL(page_put_link); 46211da177e4SLinus Torvalds 4622aa80deabSAl Viro int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 4623aa80deabSAl Viro { 4624fceef393SAl Viro DEFINE_DELAYED_CALL(done); 46256b255391SAl Viro int res = readlink_copy(buffer, buflen, 46266b255391SAl Viro page_get_link(dentry, d_inode(dentry), 4627fceef393SAl Viro &done)); 4628fceef393SAl Viro do_delayed_call(&done); 4629aa80deabSAl Viro return res; 4630aa80deabSAl Viro } 4631aa80deabSAl Viro EXPORT_SYMBOL(page_readlink); 4632aa80deabSAl Viro 463354566b2cSNick Piggin /* 463454566b2cSNick Piggin * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS 463554566b2cSNick Piggin */ 463654566b2cSNick Piggin int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) 46371da177e4SLinus Torvalds { 46381da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 46390adb25d2SKirill Korotaev struct page *page; 4640afddba49SNick Piggin void *fsdata; 4641beb497abSDmitriy Monakhov int err; 464254566b2cSNick Piggin unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE; 464354566b2cSNick Piggin if (nofs) 464454566b2cSNick Piggin flags |= AOP_FLAG_NOFS; 46451da177e4SLinus Torvalds 46467e53cac4SNeilBrown retry: 4647afddba49SNick Piggin err = pagecache_write_begin(NULL, mapping, 0, len-1, 464854566b2cSNick Piggin flags, &page, &fsdata); 46491da177e4SLinus Torvalds if (err) 4650afddba49SNick Piggin goto fail; 4651afddba49SNick Piggin 465221fc61c7SAl Viro memcpy(page_address(page), symname, len-1); 4653afddba49SNick Piggin 4654afddba49SNick Piggin err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, 4655afddba49SNick Piggin page, fsdata); 46561da177e4SLinus Torvalds if (err < 0) 46571da177e4SLinus Torvalds goto fail; 4658afddba49SNick Piggin if (err < len-1) 4659afddba49SNick Piggin goto retry; 4660afddba49SNick Piggin 46611da177e4SLinus Torvalds mark_inode_dirty(inode); 46621da177e4SLinus Torvalds return 0; 46631da177e4SLinus Torvalds fail: 46641da177e4SLinus Torvalds return err; 46651da177e4SLinus Torvalds } 46664d359507SAl Viro EXPORT_SYMBOL(__page_symlink); 46671da177e4SLinus Torvalds 46680adb25d2SKirill Korotaev int page_symlink(struct inode *inode, const char *symname, int len) 46690adb25d2SKirill Korotaev { 46700adb25d2SKirill Korotaev return __page_symlink(inode, symname, len, 4671c62d2555SMichal Hocko !mapping_gfp_constraint(inode->i_mapping, __GFP_FS)); 46720adb25d2SKirill Korotaev } 46734d359507SAl Viro EXPORT_SYMBOL(page_symlink); 46740adb25d2SKirill Korotaev 467592e1d5beSArjan van de Ven const struct inode_operations page_symlink_inode_operations = { 46761da177e4SLinus Torvalds .readlink = generic_readlink, 46776b255391SAl Viro .get_link = page_get_link, 46781da177e4SLinus Torvalds }; 46791da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink_inode_operations); 4680