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 */ 501c8a53ee5SAl Viro struct filename *name; 502c8a53ee5SAl Viro int dfd; 5031f55a6ecSAl Viro unsigned int flags; 5048f47a016SAl Viro unsigned seq, m_seq, root_seq; 5051f55a6ecSAl Viro int last_type; 5061f55a6ecSAl Viro unsigned depth; 507756daf26SNeilBrown int total_link_count; 508697fc6caSAl Viro struct saved { 509697fc6caSAl Viro struct path link; 510697fc6caSAl Viro void *cookie; 511697fc6caSAl Viro const char *name; 512237d8b32SAl Viro struct inode *inode; 5130450b2d1SAl Viro unsigned seq; 514894bc8c4SAl Viro } *stack, internal[EMBEDDED_LEVELS]; 5151f55a6ecSAl Viro }; 5161f55a6ecSAl Viro 517c8a53ee5SAl Viro static struct nameidata *set_nameidata(struct nameidata *p, int dfd, 518c8a53ee5SAl Viro struct filename *name) 519894bc8c4SAl Viro { 520756daf26SNeilBrown struct nameidata *old = current->nameidata; 521756daf26SNeilBrown p->stack = p->internal; 522c8a53ee5SAl Viro p->dfd = dfd; 523c8a53ee5SAl Viro p->name = name; 524756daf26SNeilBrown p->total_link_count = old ? old->total_link_count : 0; 525756daf26SNeilBrown current->nameidata = p; 526756daf26SNeilBrown return old; 527894bc8c4SAl Viro } 528894bc8c4SAl Viro 529756daf26SNeilBrown static void restore_nameidata(struct nameidata *old) 530894bc8c4SAl Viro { 531756daf26SNeilBrown struct nameidata *now = current->nameidata; 532756daf26SNeilBrown 533756daf26SNeilBrown current->nameidata = old; 534756daf26SNeilBrown if (old) 535756daf26SNeilBrown old->total_link_count = now->total_link_count; 536756daf26SNeilBrown if (now->stack != now->internal) { 537756daf26SNeilBrown kfree(now->stack); 538756daf26SNeilBrown now->stack = now->internal; 539894bc8c4SAl Viro } 540894bc8c4SAl Viro } 541894bc8c4SAl Viro 542894bc8c4SAl Viro static int __nd_alloc_stack(struct nameidata *nd) 543894bc8c4SAl Viro { 544bc40aee0SAl Viro struct saved *p; 545bc40aee0SAl Viro 546bc40aee0SAl Viro if (nd->flags & LOOKUP_RCU) { 547bc40aee0SAl Viro p= kmalloc(MAXSYMLINKS * sizeof(struct saved), 548bc40aee0SAl Viro GFP_ATOMIC); 549bc40aee0SAl Viro if (unlikely(!p)) 550bc40aee0SAl Viro return -ECHILD; 551bc40aee0SAl Viro } else { 552bc40aee0SAl Viro p= kmalloc(MAXSYMLINKS * sizeof(struct saved), 553894bc8c4SAl Viro GFP_KERNEL); 554894bc8c4SAl Viro if (unlikely(!p)) 555894bc8c4SAl Viro return -ENOMEM; 556bc40aee0SAl Viro } 557894bc8c4SAl Viro memcpy(p, nd->internal, sizeof(nd->internal)); 558894bc8c4SAl Viro nd->stack = p; 559894bc8c4SAl Viro return 0; 560894bc8c4SAl Viro } 561894bc8c4SAl Viro 562894bc8c4SAl Viro static inline int nd_alloc_stack(struct nameidata *nd) 563894bc8c4SAl Viro { 564da4e0be0SAl Viro if (likely(nd->depth != EMBEDDED_LEVELS)) 565894bc8c4SAl Viro return 0; 566894bc8c4SAl Viro if (likely(nd->stack != nd->internal)) 567894bc8c4SAl Viro return 0; 568894bc8c4SAl Viro return __nd_alloc_stack(nd); 569894bc8c4SAl Viro } 570894bc8c4SAl Viro 5717973387aSAl Viro static void drop_links(struct nameidata *nd) 5727973387aSAl Viro { 5737973387aSAl Viro int i = nd->depth; 5747973387aSAl Viro while (i--) { 5757973387aSAl Viro struct saved *last = nd->stack + i; 5767973387aSAl Viro struct inode *inode = last->inode; 5777973387aSAl Viro if (last->cookie && inode->i_op->put_link) { 5787973387aSAl Viro inode->i_op->put_link(inode, last->cookie); 5797973387aSAl Viro last->cookie = NULL; 5807973387aSAl Viro } 5817973387aSAl Viro } 5827973387aSAl Viro } 5837973387aSAl Viro 5847973387aSAl Viro static void terminate_walk(struct nameidata *nd) 5857973387aSAl Viro { 5867973387aSAl Viro drop_links(nd); 5877973387aSAl Viro if (!(nd->flags & LOOKUP_RCU)) { 5887973387aSAl Viro int i; 5897973387aSAl Viro path_put(&nd->path); 5907973387aSAl Viro for (i = 0; i < nd->depth; i++) 5917973387aSAl Viro path_put(&nd->stack[i].link); 592102b8af2SAl Viro if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 593102b8af2SAl Viro path_put(&nd->root); 594102b8af2SAl Viro nd->root.mnt = NULL; 595102b8af2SAl Viro } 5967973387aSAl Viro } else { 5977973387aSAl Viro nd->flags &= ~LOOKUP_RCU; 5987973387aSAl Viro if (!(nd->flags & LOOKUP_ROOT)) 5997973387aSAl Viro nd->root.mnt = NULL; 6007973387aSAl Viro rcu_read_unlock(); 6017973387aSAl Viro } 6027973387aSAl Viro nd->depth = 0; 6037973387aSAl Viro } 6047973387aSAl Viro 6057973387aSAl Viro /* path_put is needed afterwards regardless of success or failure */ 6067973387aSAl Viro static bool legitimize_path(struct nameidata *nd, 6077973387aSAl Viro struct path *path, unsigned seq) 6087973387aSAl Viro { 6097973387aSAl Viro int res = __legitimize_mnt(path->mnt, nd->m_seq); 6107973387aSAl Viro if (unlikely(res)) { 6117973387aSAl Viro if (res > 0) 6127973387aSAl Viro path->mnt = NULL; 6137973387aSAl Viro path->dentry = NULL; 6147973387aSAl Viro return false; 6157973387aSAl Viro } 6167973387aSAl Viro if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) { 6177973387aSAl Viro path->dentry = NULL; 6187973387aSAl Viro return false; 6197973387aSAl Viro } 6207973387aSAl Viro return !read_seqcount_retry(&path->dentry->d_seq, seq); 6217973387aSAl Viro } 6227973387aSAl Viro 6237973387aSAl Viro static bool legitimize_links(struct nameidata *nd) 6247973387aSAl Viro { 6257973387aSAl Viro int i; 6267973387aSAl Viro for (i = 0; i < nd->depth; i++) { 6277973387aSAl Viro struct saved *last = nd->stack + i; 6287973387aSAl Viro if (unlikely(!legitimize_path(nd, &last->link, last->seq))) { 6297973387aSAl Viro drop_links(nd); 6307973387aSAl Viro nd->depth = i + 1; 6317973387aSAl Viro return false; 6327973387aSAl Viro } 6337973387aSAl Viro } 6347973387aSAl Viro return true; 6357973387aSAl Viro } 6367973387aSAl Viro 63719660af7SAl Viro /* 63831e6b01fSNick Piggin * Path walking has 2 modes, rcu-walk and ref-walk (see 63919660af7SAl Viro * Documentation/filesystems/path-lookup.txt). In situations when we can't 64019660af7SAl Viro * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab 64119660af7SAl Viro * normal reference counts on dentries and vfsmounts to transition to rcu-walk 64219660af7SAl Viro * mode. Refcounts are grabbed at the last known good point before rcu-walk 64319660af7SAl Viro * got stuck, so ref-walk may continue from there. If this is not successful 64419660af7SAl Viro * (eg. a seqcount has changed), then failure is returned and it's up to caller 64519660af7SAl Viro * to restart the path walk from the beginning in ref-walk mode. 64631e6b01fSNick Piggin */ 64731e6b01fSNick Piggin 64831e6b01fSNick Piggin /** 64919660af7SAl Viro * unlazy_walk - try to switch to ref-walk mode. 65019660af7SAl Viro * @nd: nameidata pathwalk data 65119660af7SAl Viro * @dentry: child of nd->path.dentry or NULL 6526e9918b7SAl Viro * @seq: seq number to check dentry against 65339191628SRandy Dunlap * Returns: 0 on success, -ECHILD on failure 65431e6b01fSNick Piggin * 65519660af7SAl Viro * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry 65619660af7SAl Viro * for ref-walk mode. @dentry must be a path found by a do_lookup call on 65719660af7SAl Viro * @nd or NULL. Must be called from rcu-walk context. 6587973387aSAl Viro * Nothing should touch nameidata between unlazy_walk() failure and 6597973387aSAl Viro * terminate_walk(). 66031e6b01fSNick Piggin */ 6616e9918b7SAl Viro static int unlazy_walk(struct nameidata *nd, struct dentry *dentry, unsigned seq) 66231e6b01fSNick Piggin { 66331e6b01fSNick Piggin struct dentry *parent = nd->path.dentry; 66431e6b01fSNick Piggin 66531e6b01fSNick Piggin BUG_ON(!(nd->flags & LOOKUP_RCU)); 666e5c832d5SLinus Torvalds 667e5c832d5SLinus Torvalds nd->flags &= ~LOOKUP_RCU; 6687973387aSAl Viro if (unlikely(!legitimize_links(nd))) 6697973387aSAl Viro goto out2; 6707973387aSAl Viro if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq))) 6717973387aSAl Viro goto out2; 6727973387aSAl Viro if (unlikely(!lockref_get_not_dead(&parent->d_lockref))) 6737973387aSAl Viro goto out1; 67448a066e7SAl Viro 67515570086SLinus Torvalds /* 67615570086SLinus Torvalds * For a negative lookup, the lookup sequence point is the parents 67715570086SLinus Torvalds * sequence point, and it only needs to revalidate the parent dentry. 67815570086SLinus Torvalds * 67915570086SLinus Torvalds * For a positive lookup, we need to move both the parent and the 68015570086SLinus Torvalds * dentry from the RCU domain to be properly refcounted. And the 68115570086SLinus Torvalds * sequence number in the dentry validates *both* dentry counters, 68215570086SLinus Torvalds * since we checked the sequence number of the parent after we got 68315570086SLinus Torvalds * the child sequence number. So we know the parent must still 68415570086SLinus Torvalds * be valid if the child sequence number is still valid. 68515570086SLinus Torvalds */ 68619660af7SAl Viro if (!dentry) { 687e5c832d5SLinus Torvalds if (read_seqcount_retry(&parent->d_seq, nd->seq)) 688e5c832d5SLinus Torvalds goto out; 68919660af7SAl Viro BUG_ON(nd->inode != parent->d_inode); 69019660af7SAl Viro } else { 691e5c832d5SLinus Torvalds if (!lockref_get_not_dead(&dentry->d_lockref)) 692e5c832d5SLinus Torvalds goto out; 6936e9918b7SAl Viro if (read_seqcount_retry(&dentry->d_seq, seq)) 694e5c832d5SLinus Torvalds goto drop_dentry; 69519660af7SAl Viro } 696e5c832d5SLinus Torvalds 697e5c832d5SLinus Torvalds /* 698e5c832d5SLinus Torvalds * Sequence counts matched. Now make sure that the root is 699e5c832d5SLinus Torvalds * still valid and get it if required. 700e5c832d5SLinus Torvalds */ 701e5c832d5SLinus Torvalds if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 7025a8d87e8SAl Viro if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) { 7035a8d87e8SAl Viro rcu_read_unlock(); 7045a8d87e8SAl Viro dput(dentry); 7055a8d87e8SAl Viro return -ECHILD; 7067973387aSAl Viro } 70731e6b01fSNick Piggin } 70831e6b01fSNick Piggin 7098b61e74fSAl Viro rcu_read_unlock(); 71031e6b01fSNick Piggin return 0; 71119660af7SAl Viro 712e5c832d5SLinus Torvalds drop_dentry: 7138b61e74fSAl Viro rcu_read_unlock(); 714e5c832d5SLinus Torvalds dput(dentry); 715d0d27277SLinus Torvalds goto drop_root_mnt; 7167973387aSAl Viro out2: 7177973387aSAl Viro nd->path.mnt = NULL; 7187973387aSAl Viro out1: 7197973387aSAl Viro nd->path.dentry = NULL; 720e5c832d5SLinus Torvalds out: 7218b61e74fSAl Viro rcu_read_unlock(); 722d0d27277SLinus Torvalds drop_root_mnt: 723d0d27277SLinus Torvalds if (!(nd->flags & LOOKUP_ROOT)) 724d0d27277SLinus Torvalds nd->root.mnt = NULL; 72531e6b01fSNick Piggin return -ECHILD; 72631e6b01fSNick Piggin } 72731e6b01fSNick Piggin 7287973387aSAl Viro static int unlazy_link(struct nameidata *nd, struct path *link, unsigned seq) 7297973387aSAl Viro { 7307973387aSAl Viro if (unlikely(!legitimize_path(nd, link, seq))) { 7317973387aSAl Viro drop_links(nd); 7327973387aSAl Viro nd->depth = 0; 7337973387aSAl Viro nd->flags &= ~LOOKUP_RCU; 7347973387aSAl Viro nd->path.mnt = NULL; 7357973387aSAl Viro nd->path.dentry = NULL; 7367973387aSAl Viro if (!(nd->flags & LOOKUP_ROOT)) 7377973387aSAl Viro nd->root.mnt = NULL; 7387973387aSAl Viro rcu_read_unlock(); 7397973387aSAl Viro } else if (likely(unlazy_walk(nd, NULL, 0)) == 0) { 7407973387aSAl Viro return 0; 7417973387aSAl Viro } 7427973387aSAl Viro path_put(link); 7437973387aSAl Viro return -ECHILD; 7447973387aSAl Viro } 7457973387aSAl Viro 7464ce16ef3SAl Viro static inline int d_revalidate(struct dentry *dentry, unsigned int flags) 74734286d66SNick Piggin { 7484ce16ef3SAl Viro return dentry->d_op->d_revalidate(dentry, flags); 74934286d66SNick Piggin } 75034286d66SNick Piggin 7519f1fafeeSAl Viro /** 7529f1fafeeSAl Viro * complete_walk - successful completion of path walk 7539f1fafeeSAl Viro * @nd: pointer nameidata 75439159de2SJeff Layton * 7559f1fafeeSAl Viro * If we had been in RCU mode, drop out of it and legitimize nd->path. 7569f1fafeeSAl Viro * Revalidate the final result, unless we'd already done that during 7579f1fafeeSAl Viro * the path walk or the filesystem doesn't ask for it. Return 0 on 7589f1fafeeSAl Viro * success, -error on failure. In case of failure caller does not 7599f1fafeeSAl Viro * need to drop nd->path. 76039159de2SJeff Layton */ 7619f1fafeeSAl Viro static int complete_walk(struct nameidata *nd) 76239159de2SJeff Layton { 76316c2cd71SAl Viro struct dentry *dentry = nd->path.dentry; 76439159de2SJeff Layton int status; 76539159de2SJeff Layton 7669f1fafeeSAl Viro if (nd->flags & LOOKUP_RCU) { 7679f1fafeeSAl Viro if (!(nd->flags & LOOKUP_ROOT)) 7689f1fafeeSAl Viro nd->root.mnt = NULL; 7696e9918b7SAl Viro if (unlikely(unlazy_walk(nd, NULL, 0))) 77048a066e7SAl Viro return -ECHILD; 77148a066e7SAl Viro } 7729f1fafeeSAl Viro 77316c2cd71SAl Viro if (likely(!(nd->flags & LOOKUP_JUMPED))) 77439159de2SJeff Layton return 0; 77539159de2SJeff Layton 776ecf3d1f1SJeff Layton if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) 77716c2cd71SAl Viro return 0; 77816c2cd71SAl Viro 779ecf3d1f1SJeff Layton status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); 78039159de2SJeff Layton if (status > 0) 78139159de2SJeff Layton return 0; 78239159de2SJeff Layton 78316c2cd71SAl Viro if (!status) 78439159de2SJeff Layton status = -ESTALE; 78516c2cd71SAl Viro 78639159de2SJeff Layton return status; 78739159de2SJeff Layton } 78839159de2SJeff Layton 78918d8c860SAl Viro static void set_root(struct nameidata *nd) 7902a737871SAl Viro { 791f7ad3c6bSMiklos Szeredi get_fs_root(current->fs, &nd->root); 7922a737871SAl Viro } 7932a737871SAl Viro 79418d8c860SAl Viro static unsigned set_root_rcu(struct nameidata *nd) 79531e6b01fSNick Piggin { 79631e6b01fSNick Piggin struct fs_struct *fs = current->fs; 7978f47a016SAl Viro unsigned seq; 798c28cc364SNick Piggin 799c28cc364SNick Piggin do { 800c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 80131e6b01fSNick Piggin nd->root = fs->root; 8028f47a016SAl Viro nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq); 803c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 8048f47a016SAl Viro return nd->root_seq; 80531e6b01fSNick Piggin } 80631e6b01fSNick Piggin 8071d957f9bSJan Blunck static void path_put_conditional(struct path *path, struct nameidata *nd) 808051d3812SIan Kent { 809051d3812SIan Kent dput(path->dentry); 8104ac91378SJan Blunck if (path->mnt != nd->path.mnt) 811051d3812SIan Kent mntput(path->mnt); 812051d3812SIan Kent } 813051d3812SIan Kent 8147b9337aaSNick Piggin static inline void path_to_nameidata(const struct path *path, 8157b9337aaSNick Piggin struct nameidata *nd) 816051d3812SIan Kent { 81731e6b01fSNick Piggin if (!(nd->flags & LOOKUP_RCU)) { 8184ac91378SJan Blunck dput(nd->path.dentry); 81931e6b01fSNick Piggin if (nd->path.mnt != path->mnt) 8204ac91378SJan Blunck mntput(nd->path.mnt); 8219a229683SHuang Shijie } 82231e6b01fSNick Piggin nd->path.mnt = path->mnt; 8234ac91378SJan Blunck nd->path.dentry = path->dentry; 824051d3812SIan Kent } 825051d3812SIan Kent 826b5fb63c1SChristoph Hellwig /* 827b5fb63c1SChristoph Hellwig * Helper to directly jump to a known parsed path from ->follow_link, 828b5fb63c1SChristoph Hellwig * caller must have taken a reference to path beforehand. 829b5fb63c1SChristoph Hellwig */ 8306e77137bSAl Viro void nd_jump_link(struct path *path) 831b5fb63c1SChristoph Hellwig { 8326e77137bSAl Viro struct nameidata *nd = current->nameidata; 833b5fb63c1SChristoph Hellwig path_put(&nd->path); 834b5fb63c1SChristoph Hellwig 835b5fb63c1SChristoph Hellwig nd->path = *path; 836b5fb63c1SChristoph Hellwig nd->inode = nd->path.dentry->d_inode; 837b5fb63c1SChristoph Hellwig nd->flags |= LOOKUP_JUMPED; 838b5fb63c1SChristoph Hellwig } 839b5fb63c1SChristoph Hellwig 840b9ff4429SAl Viro static inline void put_link(struct nameidata *nd) 841574197e0SAl Viro { 84221c3003dSAl Viro struct saved *last = nd->stack + --nd->depth; 843237d8b32SAl Viro struct inode *inode = last->inode; 844b9ff4429SAl Viro if (last->cookie && inode->i_op->put_link) 8455f2c4179SAl Viro inode->i_op->put_link(inode, last->cookie); 8466548fae2SAl Viro if (!(nd->flags & LOOKUP_RCU)) 847b9ff4429SAl Viro path_put(&last->link); 848574197e0SAl Viro } 849574197e0SAl Viro 850561ec64aSLinus Torvalds int sysctl_protected_symlinks __read_mostly = 0; 851561ec64aSLinus Torvalds int sysctl_protected_hardlinks __read_mostly = 0; 852800179c9SKees Cook 853800179c9SKees Cook /** 854800179c9SKees Cook * may_follow_link - Check symlink following for unsafe situations 85555852635SRandy Dunlap * @nd: nameidata pathwalk data 856800179c9SKees Cook * 857800179c9SKees Cook * In the case of the sysctl_protected_symlinks sysctl being enabled, 858800179c9SKees Cook * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is 859800179c9SKees Cook * in a sticky world-writable directory. This is to protect privileged 860800179c9SKees Cook * processes from failing races against path names that may change out 861800179c9SKees Cook * from under them by way of other users creating malicious symlinks. 862800179c9SKees Cook * It will permit symlinks to be followed only when outside a sticky 863800179c9SKees Cook * world-writable directory, or when the uid of the symlink and follower 864800179c9SKees Cook * match, or when the directory owner matches the symlink's owner. 865800179c9SKees Cook * 866800179c9SKees Cook * Returns 0 if following the symlink is allowed, -ve on error. 867800179c9SKees Cook */ 868fec2fa24SAl Viro static inline int may_follow_link(struct nameidata *nd) 869800179c9SKees Cook { 870800179c9SKees Cook const struct inode *inode; 871800179c9SKees Cook const struct inode *parent; 872800179c9SKees Cook 873800179c9SKees Cook if (!sysctl_protected_symlinks) 874800179c9SKees Cook return 0; 875800179c9SKees Cook 876800179c9SKees Cook /* Allowed if owner and follower match. */ 877237d8b32SAl Viro inode = nd->stack[0].inode; 87881abe27bSEric W. Biederman if (uid_eq(current_cred()->fsuid, inode->i_uid)) 879800179c9SKees Cook return 0; 880800179c9SKees Cook 881800179c9SKees Cook /* Allowed if parent directory not sticky and world-writable. */ 882800179c9SKees Cook parent = nd->path.dentry->d_inode; 883800179c9SKees Cook if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) 884800179c9SKees Cook return 0; 885800179c9SKees Cook 886800179c9SKees Cook /* Allowed if parent directory and link owner match. */ 88781abe27bSEric W. Biederman if (uid_eq(parent->i_uid, inode->i_uid)) 888800179c9SKees Cook return 0; 889800179c9SKees Cook 89031956502SAl Viro if (nd->flags & LOOKUP_RCU) 89131956502SAl Viro return -ECHILD; 89231956502SAl Viro 8931cf2665bSAl Viro audit_log_link_denied("follow_link", &nd->stack[0].link); 894800179c9SKees Cook return -EACCES; 895800179c9SKees Cook } 896800179c9SKees Cook 897800179c9SKees Cook /** 898800179c9SKees Cook * safe_hardlink_source - Check for safe hardlink conditions 899800179c9SKees Cook * @inode: the source inode to hardlink from 900800179c9SKees Cook * 901800179c9SKees Cook * Return false if at least one of the following conditions: 902800179c9SKees Cook * - inode is not a regular file 903800179c9SKees Cook * - inode is setuid 904800179c9SKees Cook * - inode is setgid and group-exec 905800179c9SKees Cook * - access failure for read and write 906800179c9SKees Cook * 907800179c9SKees Cook * Otherwise returns true. 908800179c9SKees Cook */ 909800179c9SKees Cook static bool safe_hardlink_source(struct inode *inode) 910800179c9SKees Cook { 911800179c9SKees Cook umode_t mode = inode->i_mode; 912800179c9SKees Cook 913800179c9SKees Cook /* Special files should not get pinned to the filesystem. */ 914800179c9SKees Cook if (!S_ISREG(mode)) 915800179c9SKees Cook return false; 916800179c9SKees Cook 917800179c9SKees Cook /* Setuid files should not get pinned to the filesystem. */ 918800179c9SKees Cook if (mode & S_ISUID) 919800179c9SKees Cook return false; 920800179c9SKees Cook 921800179c9SKees Cook /* Executable setgid files should not get pinned to the filesystem. */ 922800179c9SKees Cook if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) 923800179c9SKees Cook return false; 924800179c9SKees Cook 925800179c9SKees Cook /* Hardlinking to unreadable or unwritable sources is dangerous. */ 926800179c9SKees Cook if (inode_permission(inode, MAY_READ | MAY_WRITE)) 927800179c9SKees Cook return false; 928800179c9SKees Cook 929800179c9SKees Cook return true; 930800179c9SKees Cook } 931800179c9SKees Cook 932800179c9SKees Cook /** 933800179c9SKees Cook * may_linkat - Check permissions for creating a hardlink 934800179c9SKees Cook * @link: the source to hardlink from 935800179c9SKees Cook * 936800179c9SKees Cook * Block hardlink when all of: 937800179c9SKees Cook * - sysctl_protected_hardlinks enabled 938800179c9SKees Cook * - fsuid does not match inode 939800179c9SKees Cook * - hardlink source is unsafe (see safe_hardlink_source() above) 940800179c9SKees Cook * - not CAP_FOWNER 941800179c9SKees Cook * 942800179c9SKees Cook * Returns 0 if successful, -ve on error. 943800179c9SKees Cook */ 944800179c9SKees Cook static int may_linkat(struct path *link) 945800179c9SKees Cook { 946800179c9SKees Cook const struct cred *cred; 947800179c9SKees Cook struct inode *inode; 948800179c9SKees Cook 949800179c9SKees Cook if (!sysctl_protected_hardlinks) 950800179c9SKees Cook return 0; 951800179c9SKees Cook 952800179c9SKees Cook cred = current_cred(); 953800179c9SKees Cook inode = link->dentry->d_inode; 954800179c9SKees Cook 955800179c9SKees Cook /* Source inode owner (or CAP_FOWNER) can hardlink all they like, 956800179c9SKees Cook * otherwise, it must be a safe source. 957800179c9SKees Cook */ 95881abe27bSEric W. Biederman if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) || 959800179c9SKees Cook capable(CAP_FOWNER)) 960800179c9SKees Cook return 0; 961800179c9SKees Cook 962a51d9eaaSKees Cook audit_log_link_denied("linkat", link); 963800179c9SKees Cook return -EPERM; 964800179c9SKees Cook } 965800179c9SKees Cook 9663b2e7f75SAl Viro static __always_inline 9673b2e7f75SAl Viro const char *get_link(struct nameidata *nd) 9681da177e4SLinus Torvalds { 969ab104923SAl Viro struct saved *last = nd->stack + nd->depth - 1; 9701cf2665bSAl Viro struct dentry *dentry = last->link.dentry; 971237d8b32SAl Viro struct inode *inode = last->inode; 9726d7b5aaeSAl Viro int error; 9730a959df5SAl Viro const char *res; 9741da177e4SLinus Torvalds 9758fa9dd24SNeilBrown if (!(nd->flags & LOOKUP_RCU)) { 9768fa9dd24SNeilBrown touch_atime(&last->link); 9778fa9dd24SNeilBrown cond_resched(); 9788fa9dd24SNeilBrown } else if (atime_needs_update(&last->link, inode)) { 9798fa9dd24SNeilBrown if (unlikely(unlazy_walk(nd, NULL, 0))) 9808fa9dd24SNeilBrown return ERR_PTR(-ECHILD); 9818fa9dd24SNeilBrown touch_atime(&last->link); 9828fa9dd24SNeilBrown } 9838fa9dd24SNeilBrown 984bda0be7aSNeilBrown error = security_inode_follow_link(dentry, inode, 985bda0be7aSNeilBrown nd->flags & LOOKUP_RCU); 986bda0be7aSNeilBrown if (unlikely(error)) 9876920a440SAl Viro return ERR_PTR(error); 98836f3b4f6SAl Viro 98986acdca1SAl Viro nd->last_type = LAST_BIND; 990d4dee48bSAl Viro res = inode->i_link; 991d4dee48bSAl Viro if (!res) { 9928c1b4566SAl Viro if (nd->flags & LOOKUP_RCU) { 9938c1b4566SAl Viro if (unlikely(unlazy_walk(nd, NULL, 0))) 9948c1b4566SAl Viro return ERR_PTR(-ECHILD); 9958c1b4566SAl Viro } 9966e77137bSAl Viro res = inode->i_op->follow_link(dentry, &last->cookie); 997fab51e8aSAl Viro if (IS_ERR_OR_NULL(res)) { 9986920a440SAl Viro last->cookie = NULL; 999fab51e8aSAl Viro return res; 10000a959df5SAl Viro } 1001fab51e8aSAl Viro } 1002fab51e8aSAl Viro if (*res == '/') { 10038c1b4566SAl Viro if (nd->flags & LOOKUP_RCU) { 10048f47a016SAl Viro struct dentry *d; 10058f47a016SAl Viro if (!nd->root.mnt) 10068f47a016SAl Viro set_root_rcu(nd); 10078f47a016SAl Viro nd->path = nd->root; 10088f47a016SAl Viro d = nd->path.dentry; 10098f47a016SAl Viro nd->inode = d->d_inode; 10108f47a016SAl Viro nd->seq = nd->root_seq; 10118f47a016SAl Viro if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq))) 10128c1b4566SAl Viro return ERR_PTR(-ECHILD); 10138f47a016SAl Viro } else { 1014fab51e8aSAl Viro if (!nd->root.mnt) 1015fab51e8aSAl Viro set_root(nd); 1016fab51e8aSAl Viro path_put(&nd->path); 1017fab51e8aSAl Viro nd->path = nd->root; 1018fab51e8aSAl Viro path_get(&nd->root); 1019fab51e8aSAl Viro nd->inode = nd->path.dentry->d_inode; 10208f47a016SAl Viro } 1021fab51e8aSAl Viro nd->flags |= LOOKUP_JUMPED; 1022fab51e8aSAl Viro while (unlikely(*++res == '/')) 1023fab51e8aSAl Viro ; 1024fab51e8aSAl Viro } 1025fab51e8aSAl Viro if (!*res) 1026fab51e8aSAl Viro res = NULL; 10270a959df5SAl Viro return res; 10280a959df5SAl Viro } 10296d7b5aaeSAl Viro 1030f015f126SDavid Howells /* 1031f015f126SDavid Howells * follow_up - Find the mountpoint of path's vfsmount 1032f015f126SDavid Howells * 1033f015f126SDavid Howells * Given a path, find the mountpoint of its source file system. 1034f015f126SDavid Howells * Replace @path with the path of the mountpoint in the parent mount. 1035f015f126SDavid Howells * Up is towards /. 1036f015f126SDavid Howells * 1037f015f126SDavid Howells * Return 1 if we went up a level and 0 if we were already at the 1038f015f126SDavid Howells * root. 1039f015f126SDavid Howells */ 1040bab77ebfSAl Viro int follow_up(struct path *path) 10411da177e4SLinus Torvalds { 10420714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 10430714a533SAl Viro struct mount *parent; 10441da177e4SLinus Torvalds struct dentry *mountpoint; 104599b7db7bSNick Piggin 104648a066e7SAl Viro read_seqlock_excl(&mount_lock); 10470714a533SAl Viro parent = mnt->mnt_parent; 10483c0a6163SAl Viro if (parent == mnt) { 104948a066e7SAl Viro read_sequnlock_excl(&mount_lock); 10501da177e4SLinus Torvalds return 0; 10511da177e4SLinus Torvalds } 10520714a533SAl Viro mntget(&parent->mnt); 1053a73324daSAl Viro mountpoint = dget(mnt->mnt_mountpoint); 105448a066e7SAl Viro read_sequnlock_excl(&mount_lock); 1055bab77ebfSAl Viro dput(path->dentry); 1056bab77ebfSAl Viro path->dentry = mountpoint; 1057bab77ebfSAl Viro mntput(path->mnt); 10580714a533SAl Viro path->mnt = &parent->mnt; 10591da177e4SLinus Torvalds return 1; 10601da177e4SLinus Torvalds } 10614d359507SAl Viro EXPORT_SYMBOL(follow_up); 10621da177e4SLinus Torvalds 1063b5c84bf6SNick Piggin /* 10649875cf80SDavid Howells * Perform an automount 10659875cf80SDavid Howells * - return -EISDIR to tell follow_managed() to stop and return the path we 10669875cf80SDavid Howells * were called with. 10671da177e4SLinus Torvalds */ 1068756daf26SNeilBrown static int follow_automount(struct path *path, struct nameidata *nd, 10699875cf80SDavid Howells bool *need_mntput) 107031e6b01fSNick Piggin { 10719875cf80SDavid Howells struct vfsmount *mnt; 1072ea5b778aSDavid Howells int err; 10739875cf80SDavid Howells 10749875cf80SDavid Howells if (!path->dentry->d_op || !path->dentry->d_op->d_automount) 10759875cf80SDavid Howells return -EREMOTE; 10769875cf80SDavid Howells 10770ec26fd0SMiklos Szeredi /* We don't want to mount if someone's just doing a stat - 10780ec26fd0SMiklos Szeredi * unless they're stat'ing a directory and appended a '/' to 10790ec26fd0SMiklos Szeredi * the name. 10800ec26fd0SMiklos Szeredi * 10810ec26fd0SMiklos Szeredi * We do, however, want to mount if someone wants to open or 10825a30d8a2SDavid Howells * create a file of any type under the mountpoint, wants to 10830ec26fd0SMiklos Szeredi * traverse through the mountpoint or wants to open the 10840ec26fd0SMiklos Szeredi * mounted directory. Also, autofs may mark negative dentries 10850ec26fd0SMiklos Szeredi * as being automount points. These will need the attentions 10860ec26fd0SMiklos Szeredi * of the daemon to instantiate them before they can be used. 10875a30d8a2SDavid Howells */ 1088756daf26SNeilBrown if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | 1089d94c177bSLinus Torvalds LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && 10905a30d8a2SDavid Howells path->dentry->d_inode) 10919875cf80SDavid Howells return -EISDIR; 10920ec26fd0SMiklos Szeredi 1093756daf26SNeilBrown nd->total_link_count++; 1094756daf26SNeilBrown if (nd->total_link_count >= 40) 10959875cf80SDavid Howells return -ELOOP; 10969875cf80SDavid Howells 10979875cf80SDavid Howells mnt = path->dentry->d_op->d_automount(path); 10989875cf80SDavid Howells if (IS_ERR(mnt)) { 10999875cf80SDavid Howells /* 11009875cf80SDavid Howells * The filesystem is allowed to return -EISDIR here to indicate 11019875cf80SDavid Howells * it doesn't want to automount. For instance, autofs would do 11029875cf80SDavid Howells * this so that its userspace daemon can mount on this dentry. 11039875cf80SDavid Howells * 11049875cf80SDavid Howells * However, we can only permit this if it's a terminal point in 11059875cf80SDavid Howells * the path being looked up; if it wasn't then the remainder of 11069875cf80SDavid Howells * the path is inaccessible and we should say so. 11079875cf80SDavid Howells */ 1108756daf26SNeilBrown if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT)) 11099875cf80SDavid Howells return -EREMOTE; 11109875cf80SDavid Howells return PTR_ERR(mnt); 111131e6b01fSNick Piggin } 1112ea5b778aSDavid Howells 11139875cf80SDavid Howells if (!mnt) /* mount collision */ 11149875cf80SDavid Howells return 0; 11159875cf80SDavid Howells 11168aef1884SAl Viro if (!*need_mntput) { 11178aef1884SAl Viro /* lock_mount() may release path->mnt on error */ 11188aef1884SAl Viro mntget(path->mnt); 11198aef1884SAl Viro *need_mntput = true; 11208aef1884SAl Viro } 112119a167afSAl Viro err = finish_automount(mnt, path); 1122ea5b778aSDavid Howells 1123ea5b778aSDavid Howells switch (err) { 1124ea5b778aSDavid Howells case -EBUSY: 1125ea5b778aSDavid Howells /* Someone else made a mount here whilst we were busy */ 112619a167afSAl Viro return 0; 1127ea5b778aSDavid Howells case 0: 11288aef1884SAl Viro path_put(path); 11299875cf80SDavid Howells path->mnt = mnt; 11309875cf80SDavid Howells path->dentry = dget(mnt->mnt_root); 11319875cf80SDavid Howells return 0; 113219a167afSAl Viro default: 113319a167afSAl Viro return err; 11349875cf80SDavid Howells } 113519a167afSAl Viro 1136ea5b778aSDavid Howells } 11379875cf80SDavid Howells 11389875cf80SDavid Howells /* 11399875cf80SDavid Howells * Handle a dentry that is managed in some way. 1140cc53ce53SDavid Howells * - Flagged for transit management (autofs) 11419875cf80SDavid Howells * - Flagged as mountpoint 11429875cf80SDavid Howells * - Flagged as automount point 11439875cf80SDavid Howells * 11449875cf80SDavid Howells * This may only be called in refwalk mode. 11459875cf80SDavid Howells * 11469875cf80SDavid Howells * Serialization is taken care of in namespace.c 11479875cf80SDavid Howells */ 1148756daf26SNeilBrown static int follow_managed(struct path *path, struct nameidata *nd) 11499875cf80SDavid Howells { 11508aef1884SAl Viro struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */ 11519875cf80SDavid Howells unsigned managed; 11529875cf80SDavid Howells bool need_mntput = false; 11538aef1884SAl Viro int ret = 0; 11549875cf80SDavid Howells 11559875cf80SDavid Howells /* Given that we're not holding a lock here, we retain the value in a 11569875cf80SDavid Howells * local variable for each dentry as we look at it so that we don't see 11579875cf80SDavid Howells * the components of that value change under us */ 11589875cf80SDavid Howells while (managed = ACCESS_ONCE(path->dentry->d_flags), 11599875cf80SDavid Howells managed &= DCACHE_MANAGED_DENTRY, 11609875cf80SDavid Howells unlikely(managed != 0)) { 1161cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1162cc53ce53SDavid Howells * being held. */ 1163cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 1164cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1165cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 11661aed3e42SAl Viro ret = path->dentry->d_op->d_manage(path->dentry, false); 1167cc53ce53SDavid Howells if (ret < 0) 11688aef1884SAl Viro break; 1169cc53ce53SDavid Howells } 1170cc53ce53SDavid Howells 11719875cf80SDavid Howells /* Transit to a mounted filesystem. */ 11729875cf80SDavid Howells if (managed & DCACHE_MOUNTED) { 11739875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 11749875cf80SDavid Howells if (mounted) { 11759875cf80SDavid Howells dput(path->dentry); 11769875cf80SDavid Howells if (need_mntput) 1177463ffb2eSAl Viro mntput(path->mnt); 1178463ffb2eSAl Viro path->mnt = mounted; 1179463ffb2eSAl Viro path->dentry = dget(mounted->mnt_root); 11809875cf80SDavid Howells need_mntput = true; 11819875cf80SDavid Howells continue; 1182463ffb2eSAl Viro } 1183463ffb2eSAl Viro 11849875cf80SDavid Howells /* Something is mounted on this dentry in another 11859875cf80SDavid Howells * namespace and/or whatever was mounted there in this 118648a066e7SAl Viro * namespace got unmounted before lookup_mnt() could 118748a066e7SAl Viro * get it */ 11881da177e4SLinus Torvalds } 11899875cf80SDavid Howells 11909875cf80SDavid Howells /* Handle an automount point */ 11919875cf80SDavid Howells if (managed & DCACHE_NEED_AUTOMOUNT) { 1192756daf26SNeilBrown ret = follow_automount(path, nd, &need_mntput); 11939875cf80SDavid Howells if (ret < 0) 11948aef1884SAl Viro break; 11959875cf80SDavid Howells continue; 11969875cf80SDavid Howells } 11979875cf80SDavid Howells 11989875cf80SDavid Howells /* We didn't change the current path point */ 11999875cf80SDavid Howells break; 12009875cf80SDavid Howells } 12018aef1884SAl Viro 12028aef1884SAl Viro if (need_mntput && path->mnt == mnt) 12038aef1884SAl Viro mntput(path->mnt); 12048aef1884SAl Viro if (ret == -EISDIR) 12058aef1884SAl Viro ret = 0; 12068402752eSAl Viro if (need_mntput) 12078402752eSAl Viro nd->flags |= LOOKUP_JUMPED; 12088402752eSAl Viro if (unlikely(ret < 0)) 12098402752eSAl Viro path_put_conditional(path, nd); 12108402752eSAl Viro return ret; 12111da177e4SLinus Torvalds } 12121da177e4SLinus Torvalds 1213cc53ce53SDavid Howells int follow_down_one(struct path *path) 12141da177e4SLinus Torvalds { 12151da177e4SLinus Torvalds struct vfsmount *mounted; 12161da177e4SLinus Torvalds 12171c755af4SAl Viro mounted = lookup_mnt(path); 12181da177e4SLinus Torvalds if (mounted) { 12199393bd07SAl Viro dput(path->dentry); 12209393bd07SAl Viro mntput(path->mnt); 12219393bd07SAl Viro path->mnt = mounted; 12229393bd07SAl Viro path->dentry = dget(mounted->mnt_root); 12231da177e4SLinus Torvalds return 1; 12241da177e4SLinus Torvalds } 12251da177e4SLinus Torvalds return 0; 12261da177e4SLinus Torvalds } 12274d359507SAl Viro EXPORT_SYMBOL(follow_down_one); 12281da177e4SLinus Torvalds 1229b8faf035SNeilBrown static inline int managed_dentry_rcu(struct dentry *dentry) 123062a7375eSIan Kent { 1231b8faf035SNeilBrown return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ? 1232b8faf035SNeilBrown dentry->d_op->d_manage(dentry, true) : 0; 123362a7375eSIan Kent } 123462a7375eSIan Kent 12359875cf80SDavid Howells /* 1236287548e4SAl Viro * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if 1237287548e4SAl Viro * we meet a managed dentry that would need blocking. 12389875cf80SDavid Howells */ 12399875cf80SDavid Howells static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, 1240254cf582SAl Viro struct inode **inode, unsigned *seqp) 12419875cf80SDavid Howells { 124262a7375eSIan Kent for (;;) { 1243c7105365SAl Viro struct mount *mounted; 124462a7375eSIan Kent /* 124562a7375eSIan Kent * Don't forget we might have a non-mountpoint managed dentry 124662a7375eSIan Kent * that wants to block transit. 124762a7375eSIan Kent */ 1248b8faf035SNeilBrown switch (managed_dentry_rcu(path->dentry)) { 1249b8faf035SNeilBrown case -ECHILD: 1250b8faf035SNeilBrown default: 1251ab90911fSDavid Howells return false; 1252b8faf035SNeilBrown case -EISDIR: 1253b8faf035SNeilBrown return true; 1254b8faf035SNeilBrown case 0: 1255b8faf035SNeilBrown break; 1256b8faf035SNeilBrown } 125762a7375eSIan Kent 125862a7375eSIan Kent if (!d_mountpoint(path->dentry)) 1259b8faf035SNeilBrown return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); 126062a7375eSIan Kent 1261474279dcSAl Viro mounted = __lookup_mnt(path->mnt, path->dentry); 12629875cf80SDavid Howells if (!mounted) 12639875cf80SDavid Howells break; 1264c7105365SAl Viro path->mnt = &mounted->mnt; 1265c7105365SAl Viro path->dentry = mounted->mnt.mnt_root; 1266a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 1267254cf582SAl Viro *seqp = read_seqcount_begin(&path->dentry->d_seq); 126859430262SLinus Torvalds /* 126959430262SLinus Torvalds * Update the inode too. We don't need to re-check the 127059430262SLinus Torvalds * dentry sequence number here after this d_inode read, 127159430262SLinus Torvalds * because a mount-point is always pinned. 127259430262SLinus Torvalds */ 127359430262SLinus Torvalds *inode = path->dentry->d_inode; 12749875cf80SDavid Howells } 1275f5be3e29SAl Viro return !read_seqretry(&mount_lock, nd->m_seq) && 1276b8faf035SNeilBrown !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); 1277287548e4SAl Viro } 1278287548e4SAl Viro 127931e6b01fSNick Piggin static int follow_dotdot_rcu(struct nameidata *nd) 128031e6b01fSNick Piggin { 12814023bfc9SAl Viro struct inode *inode = nd->inode; 12827bd88377SAl Viro if (!nd->root.mnt) 128331e6b01fSNick Piggin set_root_rcu(nd); 128431e6b01fSNick Piggin 128531e6b01fSNick Piggin while (1) { 1286aed434adSAl Viro if (path_equal(&nd->path, &nd->root)) 128731e6b01fSNick Piggin break; 128831e6b01fSNick Piggin if (nd->path.dentry != nd->path.mnt->mnt_root) { 128931e6b01fSNick Piggin struct dentry *old = nd->path.dentry; 129031e6b01fSNick Piggin struct dentry *parent = old->d_parent; 129131e6b01fSNick Piggin unsigned seq; 129231e6b01fSNick Piggin 12934023bfc9SAl Viro inode = parent->d_inode; 129431e6b01fSNick Piggin seq = read_seqcount_begin(&parent->d_seq); 1295aed434adSAl Viro if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq))) 1296aed434adSAl Viro return -ECHILD; 129731e6b01fSNick Piggin nd->path.dentry = parent; 129831e6b01fSNick Piggin nd->seq = seq; 129931e6b01fSNick Piggin break; 1300aed434adSAl Viro } else { 1301aed434adSAl Viro struct mount *mnt = real_mount(nd->path.mnt); 1302aed434adSAl Viro struct mount *mparent = mnt->mnt_parent; 1303aed434adSAl Viro struct dentry *mountpoint = mnt->mnt_mountpoint; 1304aed434adSAl Viro struct inode *inode2 = mountpoint->d_inode; 1305aed434adSAl Viro unsigned seq = read_seqcount_begin(&mountpoint->d_seq); 1306aed434adSAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1307aed434adSAl Viro return -ECHILD; 1308aed434adSAl Viro if (&mparent->mnt == nd->path.mnt) 130931e6b01fSNick Piggin break; 1310aed434adSAl Viro /* we know that mountpoint was pinned */ 1311aed434adSAl Viro nd->path.dentry = mountpoint; 1312aed434adSAl Viro nd->path.mnt = &mparent->mnt; 1313aed434adSAl Viro inode = inode2; 1314aed434adSAl Viro nd->seq = seq; 131531e6b01fSNick Piggin } 1316aed434adSAl Viro } 1317aed434adSAl Viro while (unlikely(d_mountpoint(nd->path.dentry))) { 1318b37199e6SAl Viro struct mount *mounted; 1319b37199e6SAl Viro mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry); 1320aed434adSAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1321aed434adSAl Viro return -ECHILD; 1322b37199e6SAl Viro if (!mounted) 1323b37199e6SAl Viro break; 1324b37199e6SAl Viro nd->path.mnt = &mounted->mnt; 1325b37199e6SAl Viro nd->path.dentry = mounted->mnt.mnt_root; 13264023bfc9SAl Viro inode = nd->path.dentry->d_inode; 1327b37199e6SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 1328b37199e6SAl Viro } 13294023bfc9SAl Viro nd->inode = inode; 133031e6b01fSNick Piggin return 0; 133131e6b01fSNick Piggin } 133231e6b01fSNick Piggin 13339875cf80SDavid Howells /* 1334cc53ce53SDavid Howells * Follow down to the covering mount currently visible to userspace. At each 1335cc53ce53SDavid Howells * point, the filesystem owning that dentry may be queried as to whether the 1336cc53ce53SDavid Howells * caller is permitted to proceed or not. 1337cc53ce53SDavid Howells */ 13387cc90cc3SAl Viro int follow_down(struct path *path) 1339cc53ce53SDavid Howells { 1340cc53ce53SDavid Howells unsigned managed; 1341cc53ce53SDavid Howells int ret; 1342cc53ce53SDavid Howells 1343cc53ce53SDavid Howells while (managed = ACCESS_ONCE(path->dentry->d_flags), 1344cc53ce53SDavid Howells unlikely(managed & DCACHE_MANAGED_DENTRY)) { 1345cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1346cc53ce53SDavid Howells * being held. 1347cc53ce53SDavid Howells * 1348cc53ce53SDavid Howells * We indicate to the filesystem if someone is trying to mount 1349cc53ce53SDavid Howells * something here. This gives autofs the chance to deny anyone 1350cc53ce53SDavid Howells * other than its daemon the right to mount on its 1351cc53ce53SDavid Howells * superstructure. 1352cc53ce53SDavid Howells * 1353cc53ce53SDavid Howells * The filesystem may sleep at this point. 1354cc53ce53SDavid Howells */ 1355cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 1356cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1357cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 1358ab90911fSDavid Howells ret = path->dentry->d_op->d_manage( 13591aed3e42SAl Viro path->dentry, false); 1360cc53ce53SDavid Howells if (ret < 0) 1361cc53ce53SDavid Howells return ret == -EISDIR ? 0 : ret; 1362cc53ce53SDavid Howells } 1363cc53ce53SDavid Howells 1364cc53ce53SDavid Howells /* Transit to a mounted filesystem. */ 1365cc53ce53SDavid Howells if (managed & DCACHE_MOUNTED) { 1366cc53ce53SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 1367cc53ce53SDavid Howells if (!mounted) 1368cc53ce53SDavid Howells break; 1369cc53ce53SDavid Howells dput(path->dentry); 1370cc53ce53SDavid Howells mntput(path->mnt); 1371cc53ce53SDavid Howells path->mnt = mounted; 1372cc53ce53SDavid Howells path->dentry = dget(mounted->mnt_root); 1373cc53ce53SDavid Howells continue; 1374cc53ce53SDavid Howells } 1375cc53ce53SDavid Howells 1376cc53ce53SDavid Howells /* Don't handle automount points here */ 1377cc53ce53SDavid Howells break; 1378cc53ce53SDavid Howells } 1379cc53ce53SDavid Howells return 0; 1380cc53ce53SDavid Howells } 13814d359507SAl Viro EXPORT_SYMBOL(follow_down); 1382cc53ce53SDavid Howells 1383cc53ce53SDavid Howells /* 13849875cf80SDavid Howells * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() 13859875cf80SDavid Howells */ 13869875cf80SDavid Howells static void follow_mount(struct path *path) 13879875cf80SDavid Howells { 13889875cf80SDavid Howells while (d_mountpoint(path->dentry)) { 13899875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 13909875cf80SDavid Howells if (!mounted) 13919875cf80SDavid Howells break; 13929875cf80SDavid Howells dput(path->dentry); 13939875cf80SDavid Howells mntput(path->mnt); 13949875cf80SDavid Howells path->mnt = mounted; 13959875cf80SDavid Howells path->dentry = dget(mounted->mnt_root); 13969875cf80SDavid Howells } 13979875cf80SDavid Howells } 13989875cf80SDavid Howells 139931e6b01fSNick Piggin static void follow_dotdot(struct nameidata *nd) 14001da177e4SLinus Torvalds { 14017bd88377SAl Viro if (!nd->root.mnt) 14022a737871SAl Viro set_root(nd); 1403e518ddb7SAndreas Mohr 14041da177e4SLinus Torvalds while(1) { 14054ac91378SJan Blunck struct dentry *old = nd->path.dentry; 14061da177e4SLinus Torvalds 14072a737871SAl Viro if (nd->path.dentry == nd->root.dentry && 14082a737871SAl Viro nd->path.mnt == nd->root.mnt) { 14091da177e4SLinus Torvalds break; 14101da177e4SLinus Torvalds } 14114ac91378SJan Blunck if (nd->path.dentry != nd->path.mnt->mnt_root) { 14123088dd70SAl Viro /* rare case of legitimate dget_parent()... */ 14133088dd70SAl Viro nd->path.dentry = dget_parent(nd->path.dentry); 14141da177e4SLinus Torvalds dput(old); 14151da177e4SLinus Torvalds break; 14161da177e4SLinus Torvalds } 14173088dd70SAl Viro if (!follow_up(&nd->path)) 14181da177e4SLinus Torvalds break; 14191da177e4SLinus Torvalds } 142079ed0226SAl Viro follow_mount(&nd->path); 142131e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 14221da177e4SLinus Torvalds } 14231da177e4SLinus Torvalds 14241da177e4SLinus Torvalds /* 1425bad61189SMiklos Szeredi * This looks up the name in dcache, possibly revalidates the old dentry and 1426bad61189SMiklos Szeredi * allocates a new one if not found or not valid. In the need_lookup argument 1427bad61189SMiklos Szeredi * returns whether i_op->lookup is necessary. 1428bad61189SMiklos Szeredi * 1429bad61189SMiklos Szeredi * dir->d_inode->i_mutex must be held 1430baa03890SNick Piggin */ 1431bad61189SMiklos Szeredi static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir, 1432201f956eSAl Viro unsigned int flags, bool *need_lookup) 1433baa03890SNick Piggin { 1434baa03890SNick Piggin struct dentry *dentry; 1435bad61189SMiklos Szeredi int error; 1436baa03890SNick Piggin 1437bad61189SMiklos Szeredi *need_lookup = false; 1438bad61189SMiklos Szeredi dentry = d_lookup(dir, name); 1439bad61189SMiklos Szeredi if (dentry) { 144039e3c955SJeff Layton if (dentry->d_flags & DCACHE_OP_REVALIDATE) { 1441201f956eSAl Viro error = d_revalidate(dentry, flags); 1442bad61189SMiklos Szeredi if (unlikely(error <= 0)) { 1443bad61189SMiklos Szeredi if (error < 0) { 1444bad61189SMiklos Szeredi dput(dentry); 1445bad61189SMiklos Szeredi return ERR_PTR(error); 14465542aa2fSEric W. Biederman } else { 14475542aa2fSEric W. Biederman d_invalidate(dentry); 1448bad61189SMiklos Szeredi dput(dentry); 1449bad61189SMiklos Szeredi dentry = NULL; 1450bad61189SMiklos Szeredi } 1451bad61189SMiklos Szeredi } 1452bad61189SMiklos Szeredi } 1453bad61189SMiklos Szeredi } 1454baa03890SNick Piggin 1455bad61189SMiklos Szeredi if (!dentry) { 1456bad61189SMiklos Szeredi dentry = d_alloc(dir, name); 1457baa03890SNick Piggin if (unlikely(!dentry)) 1458baa03890SNick Piggin return ERR_PTR(-ENOMEM); 1459baa03890SNick Piggin 1460bad61189SMiklos Szeredi *need_lookup = true; 1461baa03890SNick Piggin } 1462baa03890SNick Piggin return dentry; 1463baa03890SNick Piggin } 1464baa03890SNick Piggin 1465baa03890SNick Piggin /* 146613a2c3beSJ. Bruce Fields * Call i_op->lookup on the dentry. The dentry must be negative and 146713a2c3beSJ. Bruce Fields * unhashed. 1468bad61189SMiklos Szeredi * 1469bad61189SMiklos Szeredi * dir->d_inode->i_mutex must be held 147044396f4bSJosef Bacik */ 1471bad61189SMiklos Szeredi static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry, 147272bd866aSAl Viro unsigned int flags) 147344396f4bSJosef Bacik { 147444396f4bSJosef Bacik struct dentry *old; 147544396f4bSJosef Bacik 147644396f4bSJosef Bacik /* Don't create child dentry for a dead directory. */ 1477bad61189SMiklos Szeredi if (unlikely(IS_DEADDIR(dir))) { 1478e188dc02SMiklos Szeredi dput(dentry); 147944396f4bSJosef Bacik return ERR_PTR(-ENOENT); 1480e188dc02SMiklos Szeredi } 148144396f4bSJosef Bacik 148272bd866aSAl Viro old = dir->i_op->lookup(dir, dentry, flags); 148344396f4bSJosef Bacik if (unlikely(old)) { 148444396f4bSJosef Bacik dput(dentry); 148544396f4bSJosef Bacik dentry = old; 148644396f4bSJosef Bacik } 148744396f4bSJosef Bacik return dentry; 148844396f4bSJosef Bacik } 148944396f4bSJosef Bacik 1490a3255546SAl Viro static struct dentry *__lookup_hash(struct qstr *name, 149172bd866aSAl Viro struct dentry *base, unsigned int flags) 1492a3255546SAl Viro { 1493bad61189SMiklos Szeredi bool need_lookup; 1494a3255546SAl Viro struct dentry *dentry; 1495a3255546SAl Viro 149672bd866aSAl Viro dentry = lookup_dcache(name, base, flags, &need_lookup); 1497bad61189SMiklos Szeredi if (!need_lookup) 1498a3255546SAl Viro return dentry; 1499bad61189SMiklos Szeredi 150072bd866aSAl Viro return lookup_real(base->d_inode, dentry, flags); 1501a3255546SAl Viro } 1502a3255546SAl Viro 150344396f4bSJosef Bacik /* 15041da177e4SLinus Torvalds * It's more convoluted than I'd like it to be, but... it's still fairly 15051da177e4SLinus Torvalds * small and for now I'd prefer to have fast path as straight as possible. 15061da177e4SLinus Torvalds * It _is_ time-critical. 15071da177e4SLinus Torvalds */ 1508e97cdc87SAl Viro static int lookup_fast(struct nameidata *nd, 1509254cf582SAl Viro struct path *path, struct inode **inode, 1510254cf582SAl Viro unsigned *seqp) 15111da177e4SLinus Torvalds { 15124ac91378SJan Blunck struct vfsmount *mnt = nd->path.mnt; 151331e6b01fSNick Piggin struct dentry *dentry, *parent = nd->path.dentry; 15145a18fff2SAl Viro int need_reval = 1; 15155a18fff2SAl Viro int status = 1; 15169875cf80SDavid Howells int err; 15179875cf80SDavid Howells 15183cac260aSAl Viro /* 1519b04f784eSNick Piggin * Rename seqlock is not required here because in the off chance 1520b04f784eSNick Piggin * of a false negative due to a concurrent rename, we're going to 1521b04f784eSNick Piggin * do the non-racy lookup, below. 1522b04f784eSNick Piggin */ 152331e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) { 152431e6b01fSNick Piggin unsigned seq; 1525766c4cbfSAl Viro bool negative; 1526da53be12SLinus Torvalds dentry = __d_lookup_rcu(parent, &nd->last, &seq); 15275a18fff2SAl Viro if (!dentry) 15285a18fff2SAl Viro goto unlazy; 15295a18fff2SAl Viro 153012f8ad4bSLinus Torvalds /* 153112f8ad4bSLinus Torvalds * This sequence count validates that the inode matches 153212f8ad4bSLinus Torvalds * the dentry name information from lookup. 153312f8ad4bSLinus Torvalds */ 153463afdfc7SDavid Howells *inode = d_backing_inode(dentry); 1535766c4cbfSAl Viro negative = d_is_negative(dentry); 153612f8ad4bSLinus Torvalds if (read_seqcount_retry(&dentry->d_seq, seq)) 153712f8ad4bSLinus Torvalds return -ECHILD; 1538766c4cbfSAl Viro if (negative) 1539766c4cbfSAl Viro return -ENOENT; 154012f8ad4bSLinus Torvalds 154112f8ad4bSLinus Torvalds /* 154212f8ad4bSLinus Torvalds * This sequence count validates that the parent had no 154312f8ad4bSLinus Torvalds * changes while we did the lookup of the dentry above. 154412f8ad4bSLinus Torvalds * 154512f8ad4bSLinus Torvalds * The memory barrier in read_seqcount_begin of child is 154612f8ad4bSLinus Torvalds * enough, we can use __read_seqcount_retry here. 154712f8ad4bSLinus Torvalds */ 154831e6b01fSNick Piggin if (__read_seqcount_retry(&parent->d_seq, nd->seq)) 154931e6b01fSNick Piggin return -ECHILD; 15505a18fff2SAl Viro 1551254cf582SAl Viro *seqp = seq; 155224643087SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) { 15534ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 15545a18fff2SAl Viro if (unlikely(status <= 0)) { 15555a18fff2SAl Viro if (status != -ECHILD) 15565a18fff2SAl Viro need_reval = 0; 15575a18fff2SAl Viro goto unlazy; 15585a18fff2SAl Viro } 155924643087SAl Viro } 156031e6b01fSNick Piggin path->mnt = mnt; 156131e6b01fSNick Piggin path->dentry = dentry; 1562254cf582SAl Viro if (likely(__follow_mount_rcu(nd, path, inode, seqp))) 15639875cf80SDavid Howells return 0; 15645a18fff2SAl Viro unlazy: 1565254cf582SAl Viro if (unlazy_walk(nd, dentry, seq)) 15665a18fff2SAl Viro return -ECHILD; 15675a18fff2SAl Viro } else { 1568e97cdc87SAl Viro dentry = __d_lookup(parent, &nd->last); 156924643087SAl Viro } 15705a18fff2SAl Viro 157181e6f520SAl Viro if (unlikely(!dentry)) 157281e6f520SAl Viro goto need_lookup; 15735a18fff2SAl Viro 15745a18fff2SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval) 15754ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 15765a18fff2SAl Viro if (unlikely(status <= 0)) { 15775a18fff2SAl Viro if (status < 0) { 15785a18fff2SAl Viro dput(dentry); 15795a18fff2SAl Viro return status; 15805a18fff2SAl Viro } 15815542aa2fSEric W. Biederman d_invalidate(dentry); 15825a18fff2SAl Viro dput(dentry); 158381e6f520SAl Viro goto need_lookup; 15845a18fff2SAl Viro } 1585697f514dSMiklos Szeredi 1586766c4cbfSAl Viro if (unlikely(d_is_negative(dentry))) { 1587766c4cbfSAl Viro dput(dentry); 1588766c4cbfSAl Viro return -ENOENT; 1589766c4cbfSAl Viro } 15901da177e4SLinus Torvalds path->mnt = mnt; 15911da177e4SLinus Torvalds path->dentry = dentry; 1592756daf26SNeilBrown err = follow_managed(path, nd); 15938402752eSAl Viro if (likely(!err)) 159463afdfc7SDavid Howells *inode = d_backing_inode(path->dentry); 15958402752eSAl Viro return err; 159681e6f520SAl Viro 159781e6f520SAl Viro need_lookup: 1598697f514dSMiklos Szeredi return 1; 1599697f514dSMiklos Szeredi } 1600697f514dSMiklos Szeredi 1601697f514dSMiklos Szeredi /* Fast lookup failed, do it the slow way */ 1602cc2a5271SAl Viro static int lookup_slow(struct nameidata *nd, struct path *path) 1603697f514dSMiklos Szeredi { 1604697f514dSMiklos Szeredi struct dentry *dentry, *parent; 1605697f514dSMiklos Szeredi 1606697f514dSMiklos Szeredi parent = nd->path.dentry; 160781e6f520SAl Viro BUG_ON(nd->inode != parent->d_inode); 160881e6f520SAl Viro 160981e6f520SAl Viro mutex_lock(&parent->d_inode->i_mutex); 1610cc2a5271SAl Viro dentry = __lookup_hash(&nd->last, parent, nd->flags); 161181e6f520SAl Viro mutex_unlock(&parent->d_inode->i_mutex); 161281e6f520SAl Viro if (IS_ERR(dentry)) 161381e6f520SAl Viro return PTR_ERR(dentry); 1614697f514dSMiklos Szeredi path->mnt = nd->path.mnt; 1615697f514dSMiklos Szeredi path->dentry = dentry; 16168402752eSAl Viro return follow_managed(path, nd); 16171da177e4SLinus Torvalds } 16181da177e4SLinus Torvalds 161952094c8aSAl Viro static inline int may_lookup(struct nameidata *nd) 162052094c8aSAl Viro { 162152094c8aSAl Viro if (nd->flags & LOOKUP_RCU) { 16224ad5abb3SAl Viro int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); 162352094c8aSAl Viro if (err != -ECHILD) 162452094c8aSAl Viro return err; 16256e9918b7SAl Viro if (unlazy_walk(nd, NULL, 0)) 162652094c8aSAl Viro return -ECHILD; 162752094c8aSAl Viro } 16284ad5abb3SAl Viro return inode_permission(nd->inode, MAY_EXEC); 162952094c8aSAl Viro } 163052094c8aSAl Viro 16319856fa1bSAl Viro static inline int handle_dots(struct nameidata *nd, int type) 16329856fa1bSAl Viro { 16339856fa1bSAl Viro if (type == LAST_DOTDOT) { 16349856fa1bSAl Viro if (nd->flags & LOOKUP_RCU) { 163570291aecSAl Viro return follow_dotdot_rcu(nd); 16369856fa1bSAl Viro } else 16379856fa1bSAl Viro follow_dotdot(nd); 16389856fa1bSAl Viro } 16399856fa1bSAl Viro return 0; 16409856fa1bSAl Viro } 16419856fa1bSAl Viro 1642181548c0SAl Viro static int pick_link(struct nameidata *nd, struct path *link, 1643181548c0SAl Viro struct inode *inode, unsigned seq) 1644d63ff28fSAl Viro { 1645626de996SAl Viro int error; 16461cf2665bSAl Viro struct saved *last; 1647756daf26SNeilBrown if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) { 1648626de996SAl Viro path_to_nameidata(link, nd); 1649626de996SAl Viro return -ELOOP; 1650626de996SAl Viro } 1651bc40aee0SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1652cd179f44SAl Viro if (link->mnt == nd->path.mnt) 1653cd179f44SAl Viro mntget(link->mnt); 16547973387aSAl Viro } 1655626de996SAl Viro error = nd_alloc_stack(nd); 1656626de996SAl Viro if (unlikely(error)) { 1657bc40aee0SAl Viro if (error == -ECHILD) { 1658bc40aee0SAl Viro if (unlikely(unlazy_link(nd, link, seq))) 1659bc40aee0SAl Viro return -ECHILD; 1660bc40aee0SAl Viro error = nd_alloc_stack(nd); 1661bc40aee0SAl Viro } 1662bc40aee0SAl Viro if (error) { 1663cd179f44SAl Viro path_put(link); 1664626de996SAl Viro return error; 1665626de996SAl Viro } 1666bc40aee0SAl Viro } 1667626de996SAl Viro 1668ab104923SAl Viro last = nd->stack + nd->depth++; 16691cf2665bSAl Viro last->link = *link; 1670ab104923SAl Viro last->cookie = NULL; 1671181548c0SAl Viro last->inode = inode; 16720450b2d1SAl Viro last->seq = seq; 1673d63ff28fSAl Viro return 1; 1674d63ff28fSAl Viro } 1675d63ff28fSAl Viro 16763ddcd056SLinus Torvalds /* 16773ddcd056SLinus Torvalds * Do we need to follow links? We _really_ want to be able 16783ddcd056SLinus Torvalds * to do this check without having to look at inode->i_op, 16793ddcd056SLinus Torvalds * so we keep a cache of "no, this doesn't need follow_link" 16803ddcd056SLinus Torvalds * for the common case. 16813ddcd056SLinus Torvalds */ 1682254cf582SAl Viro static inline int should_follow_link(struct nameidata *nd, struct path *link, 1683181548c0SAl Viro int follow, 1684181548c0SAl Viro struct inode *inode, unsigned seq) 16853ddcd056SLinus Torvalds { 1686d63ff28fSAl Viro if (likely(!d_is_symlink(link->dentry))) 1687d63ff28fSAl Viro return 0; 1688d63ff28fSAl Viro if (!follow) 1689d63ff28fSAl Viro return 0; 1690181548c0SAl Viro return pick_link(nd, link, inode, seq); 16913ddcd056SLinus Torvalds } 16923ddcd056SLinus Torvalds 16934693a547SAl Viro enum {WALK_GET = 1, WALK_PUT = 2}; 16944693a547SAl Viro 16954693a547SAl Viro static int walk_component(struct nameidata *nd, int flags) 1696ce57dfc1SAl Viro { 1697caa85634SAl Viro struct path path; 1698ce57dfc1SAl Viro struct inode *inode; 1699254cf582SAl Viro unsigned seq; 1700ce57dfc1SAl Viro int err; 1701ce57dfc1SAl Viro /* 1702ce57dfc1SAl Viro * "." and ".." are special - ".." especially so because it has 1703ce57dfc1SAl Viro * to be able to know about the current root directory and 1704ce57dfc1SAl Viro * parent relationships. 1705ce57dfc1SAl Viro */ 17064693a547SAl Viro if (unlikely(nd->last_type != LAST_NORM)) { 17074693a547SAl Viro err = handle_dots(nd, nd->last_type); 17084693a547SAl Viro if (flags & WALK_PUT) 17094693a547SAl Viro put_link(nd); 17104693a547SAl Viro return err; 17114693a547SAl Viro } 1712254cf582SAl Viro err = lookup_fast(nd, &path, &inode, &seq); 1713ce57dfc1SAl Viro if (unlikely(err)) { 1714697f514dSMiklos Szeredi if (err < 0) 1715f0a9ba70SAl Viro return err; 1716697f514dSMiklos Szeredi 1717caa85634SAl Viro err = lookup_slow(nd, &path); 1718697f514dSMiklos Szeredi if (err < 0) 1719f0a9ba70SAl Viro return err; 1720697f514dSMiklos Szeredi 172163afdfc7SDavid Howells inode = d_backing_inode(path.dentry); 1722254cf582SAl Viro seq = 0; /* we are already out of RCU mode */ 1723697f514dSMiklos Szeredi err = -ENOENT; 1724caa85634SAl Viro if (d_is_negative(path.dentry)) 1725697f514dSMiklos Szeredi goto out_path_put; 1726766c4cbfSAl Viro } 1727697f514dSMiklos Szeredi 17284693a547SAl Viro if (flags & WALK_PUT) 17294693a547SAl Viro put_link(nd); 1730181548c0SAl Viro err = should_follow_link(nd, &path, flags & WALK_GET, inode, seq); 1731d63ff28fSAl Viro if (unlikely(err)) 1732d63ff28fSAl Viro return err; 1733caa85634SAl Viro path_to_nameidata(&path, nd); 1734ce57dfc1SAl Viro nd->inode = inode; 1735254cf582SAl Viro nd->seq = seq; 1736ce57dfc1SAl Viro return 0; 1737697f514dSMiklos Szeredi 1738697f514dSMiklos Szeredi out_path_put: 1739caa85634SAl Viro path_to_nameidata(&path, nd); 1740697f514dSMiklos Szeredi return err; 1741ce57dfc1SAl Viro } 1742ce57dfc1SAl Viro 17431da177e4SLinus Torvalds /* 1744bfcfaa77SLinus Torvalds * We can do the critical dentry name comparison and hashing 1745bfcfaa77SLinus Torvalds * operations one word at a time, but we are limited to: 1746bfcfaa77SLinus Torvalds * 1747bfcfaa77SLinus Torvalds * - Architectures with fast unaligned word accesses. We could 1748bfcfaa77SLinus Torvalds * do a "get_unaligned()" if this helps and is sufficiently 1749bfcfaa77SLinus Torvalds * fast. 1750bfcfaa77SLinus Torvalds * 1751bfcfaa77SLinus Torvalds * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we 1752bfcfaa77SLinus Torvalds * do not trap on the (extremely unlikely) case of a page 1753bfcfaa77SLinus Torvalds * crossing operation. 1754bfcfaa77SLinus Torvalds * 1755bfcfaa77SLinus Torvalds * - Furthermore, we need an efficient 64-bit compile for the 1756bfcfaa77SLinus Torvalds * 64-bit case in order to generate the "number of bytes in 1757bfcfaa77SLinus Torvalds * the final mask". Again, that could be replaced with a 1758bfcfaa77SLinus Torvalds * efficient population count instruction or similar. 1759bfcfaa77SLinus Torvalds */ 1760bfcfaa77SLinus Torvalds #ifdef CONFIG_DCACHE_WORD_ACCESS 1761bfcfaa77SLinus Torvalds 1762f68e556eSLinus Torvalds #include <asm/word-at-a-time.h> 1763bfcfaa77SLinus Torvalds 1764f68e556eSLinus Torvalds #ifdef CONFIG_64BIT 1765bfcfaa77SLinus Torvalds 1766bfcfaa77SLinus Torvalds static inline unsigned int fold_hash(unsigned long hash) 1767bfcfaa77SLinus Torvalds { 176899d263d4SLinus Torvalds return hash_64(hash, 32); 1769bfcfaa77SLinus Torvalds } 1770bfcfaa77SLinus Torvalds 1771bfcfaa77SLinus Torvalds #else /* 32-bit case */ 1772bfcfaa77SLinus Torvalds 1773bfcfaa77SLinus Torvalds #define fold_hash(x) (x) 1774bfcfaa77SLinus Torvalds 1775bfcfaa77SLinus Torvalds #endif 1776bfcfaa77SLinus Torvalds 1777bfcfaa77SLinus Torvalds unsigned int full_name_hash(const unsigned char *name, unsigned int len) 1778bfcfaa77SLinus Torvalds { 1779bfcfaa77SLinus Torvalds unsigned long a, mask; 1780bfcfaa77SLinus Torvalds unsigned long hash = 0; 1781bfcfaa77SLinus Torvalds 1782bfcfaa77SLinus Torvalds for (;;) { 1783e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name); 1784bfcfaa77SLinus Torvalds if (len < sizeof(unsigned long)) 1785bfcfaa77SLinus Torvalds break; 1786bfcfaa77SLinus Torvalds hash += a; 1787f132c5beSAl Viro hash *= 9; 1788bfcfaa77SLinus Torvalds name += sizeof(unsigned long); 1789bfcfaa77SLinus Torvalds len -= sizeof(unsigned long); 1790bfcfaa77SLinus Torvalds if (!len) 1791bfcfaa77SLinus Torvalds goto done; 1792bfcfaa77SLinus Torvalds } 1793a5c21dceSWill Deacon mask = bytemask_from_count(len); 1794bfcfaa77SLinus Torvalds hash += mask & a; 1795bfcfaa77SLinus Torvalds done: 1796bfcfaa77SLinus Torvalds return fold_hash(hash); 1797bfcfaa77SLinus Torvalds } 1798bfcfaa77SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 1799bfcfaa77SLinus Torvalds 1800bfcfaa77SLinus Torvalds /* 1801bfcfaa77SLinus Torvalds * Calculate the length and hash of the path component, and 1802d6bb3e90SLinus Torvalds * return the "hash_len" as the result. 1803bfcfaa77SLinus Torvalds */ 1804d6bb3e90SLinus Torvalds static inline u64 hash_name(const char *name) 1805bfcfaa77SLinus Torvalds { 180636126f8fSLinus Torvalds unsigned long a, b, adata, bdata, mask, hash, len; 180736126f8fSLinus Torvalds const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 1808bfcfaa77SLinus Torvalds 1809bfcfaa77SLinus Torvalds hash = a = 0; 1810bfcfaa77SLinus Torvalds len = -sizeof(unsigned long); 1811bfcfaa77SLinus Torvalds do { 1812bfcfaa77SLinus Torvalds hash = (hash + a) * 9; 1813bfcfaa77SLinus Torvalds len += sizeof(unsigned long); 1814e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name+len); 181536126f8fSLinus Torvalds b = a ^ REPEAT_BYTE('/'); 181636126f8fSLinus Torvalds } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); 1817bfcfaa77SLinus Torvalds 181836126f8fSLinus Torvalds adata = prep_zero_mask(a, adata, &constants); 181936126f8fSLinus Torvalds bdata = prep_zero_mask(b, bdata, &constants); 182036126f8fSLinus Torvalds 182136126f8fSLinus Torvalds mask = create_zero_mask(adata | bdata); 182236126f8fSLinus Torvalds 182336126f8fSLinus Torvalds hash += a & zero_bytemask(mask); 18249226b5b4SLinus Torvalds len += find_zero(mask); 1825d6bb3e90SLinus Torvalds return hashlen_create(fold_hash(hash), len); 1826bfcfaa77SLinus Torvalds } 1827bfcfaa77SLinus Torvalds 1828bfcfaa77SLinus Torvalds #else 1829bfcfaa77SLinus Torvalds 18300145acc2SLinus Torvalds unsigned int full_name_hash(const unsigned char *name, unsigned int len) 18310145acc2SLinus Torvalds { 18320145acc2SLinus Torvalds unsigned long hash = init_name_hash(); 18330145acc2SLinus Torvalds while (len--) 18340145acc2SLinus Torvalds hash = partial_name_hash(*name++, hash); 18350145acc2SLinus Torvalds return end_name_hash(hash); 18360145acc2SLinus Torvalds } 1837ae942ae7SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 18380145acc2SLinus Torvalds 18393ddcd056SLinus Torvalds /* 1840200e9ef7SLinus Torvalds * We know there's a real path component here of at least 1841200e9ef7SLinus Torvalds * one character. 1842200e9ef7SLinus Torvalds */ 1843d6bb3e90SLinus Torvalds static inline u64 hash_name(const char *name) 1844200e9ef7SLinus Torvalds { 1845200e9ef7SLinus Torvalds unsigned long hash = init_name_hash(); 1846200e9ef7SLinus Torvalds unsigned long len = 0, c; 1847200e9ef7SLinus Torvalds 1848200e9ef7SLinus Torvalds c = (unsigned char)*name; 1849200e9ef7SLinus Torvalds do { 1850200e9ef7SLinus Torvalds len++; 1851200e9ef7SLinus Torvalds hash = partial_name_hash(c, hash); 1852200e9ef7SLinus Torvalds c = (unsigned char)name[len]; 1853200e9ef7SLinus Torvalds } while (c && c != '/'); 1854d6bb3e90SLinus Torvalds return hashlen_create(end_name_hash(hash), len); 1855200e9ef7SLinus Torvalds } 1856200e9ef7SLinus Torvalds 1857bfcfaa77SLinus Torvalds #endif 1858bfcfaa77SLinus Torvalds 1859200e9ef7SLinus Torvalds /* 18601da177e4SLinus Torvalds * Name resolution. 1861ea3834d9SPrasanna Meda * This is the basic name resolution function, turning a pathname into 1862ea3834d9SPrasanna Meda * the final dentry. We expect 'base' to be positive and a directory. 18631da177e4SLinus Torvalds * 1864ea3834d9SPrasanna Meda * Returns 0 and nd will have valid dentry and mnt on success. 1865ea3834d9SPrasanna Meda * Returns error and drops reference to input namei data on failure. 18661da177e4SLinus Torvalds */ 18676de88d72SAl Viro static int link_path_walk(const char *name, struct nameidata *nd) 18681da177e4SLinus Torvalds { 18691da177e4SLinus Torvalds int err; 18701da177e4SLinus Torvalds 18711da177e4SLinus Torvalds while (*name=='/') 18721da177e4SLinus Torvalds name++; 18731da177e4SLinus Torvalds if (!*name) 18749e18f10aSAl Viro return 0; 18751da177e4SLinus Torvalds 18761da177e4SLinus Torvalds /* At this point we know we have a real path component. */ 18771da177e4SLinus Torvalds for(;;) { 1878d6bb3e90SLinus Torvalds u64 hash_len; 1879fe479a58SAl Viro int type; 18801da177e4SLinus Torvalds 188152094c8aSAl Viro err = may_lookup(nd); 18821da177e4SLinus Torvalds if (err) 18833595e234SAl Viro return err; 18841da177e4SLinus Torvalds 1885d6bb3e90SLinus Torvalds hash_len = hash_name(name); 18861da177e4SLinus Torvalds 1887fe479a58SAl Viro type = LAST_NORM; 1888d6bb3e90SLinus Torvalds if (name[0] == '.') switch (hashlen_len(hash_len)) { 1889fe479a58SAl Viro case 2: 1890200e9ef7SLinus Torvalds if (name[1] == '.') { 1891fe479a58SAl Viro type = LAST_DOTDOT; 189216c2cd71SAl Viro nd->flags |= LOOKUP_JUMPED; 189316c2cd71SAl Viro } 1894fe479a58SAl Viro break; 1895fe479a58SAl Viro case 1: 1896fe479a58SAl Viro type = LAST_DOT; 1897fe479a58SAl Viro } 18985a202bcdSAl Viro if (likely(type == LAST_NORM)) { 18995a202bcdSAl Viro struct dentry *parent = nd->path.dentry; 190016c2cd71SAl Viro nd->flags &= ~LOOKUP_JUMPED; 19015a202bcdSAl Viro if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { 1902a060dc50SJames Hogan struct qstr this = { { .hash_len = hash_len }, .name = name }; 1903da53be12SLinus Torvalds err = parent->d_op->d_hash(parent, &this); 19045a202bcdSAl Viro if (err < 0) 19053595e234SAl Viro return err; 1906d6bb3e90SLinus Torvalds hash_len = this.hash_len; 1907d6bb3e90SLinus Torvalds name = this.name; 19085a202bcdSAl Viro } 19095a202bcdSAl Viro } 1910fe479a58SAl Viro 1911d6bb3e90SLinus Torvalds nd->last.hash_len = hash_len; 1912d6bb3e90SLinus Torvalds nd->last.name = name; 19135f4a6a69SAl Viro nd->last_type = type; 19145f4a6a69SAl Viro 1915d6bb3e90SLinus Torvalds name += hashlen_len(hash_len); 1916d6bb3e90SLinus Torvalds if (!*name) 1917bdf6cbf1SAl Viro goto OK; 1918200e9ef7SLinus Torvalds /* 1919200e9ef7SLinus Torvalds * If it wasn't NUL, we know it was '/'. Skip that 1920200e9ef7SLinus Torvalds * slash, and continue until no more slashes. 1921200e9ef7SLinus Torvalds */ 1922200e9ef7SLinus Torvalds do { 1923d6bb3e90SLinus Torvalds name++; 1924d6bb3e90SLinus Torvalds } while (unlikely(*name == '/')); 19258620c238SAl Viro if (unlikely(!*name)) { 19268620c238SAl Viro OK: 1927368ee9baSAl Viro /* pathname body, done */ 19288620c238SAl Viro if (!nd->depth) 19298620c238SAl Viro return 0; 19308620c238SAl Viro name = nd->stack[nd->depth - 1].name; 1931368ee9baSAl Viro /* trailing symlink, done */ 19328620c238SAl Viro if (!name) 19338620c238SAl Viro return 0; 19348620c238SAl Viro /* last component of nested symlink */ 19354693a547SAl Viro err = walk_component(nd, WALK_GET | WALK_PUT); 19368620c238SAl Viro } else { 19374693a547SAl Viro err = walk_component(nd, WALK_GET); 19388620c238SAl Viro } 1939ce57dfc1SAl Viro if (err < 0) 19403595e234SAl Viro return err; 1941fe479a58SAl Viro 1942ce57dfc1SAl Viro if (err) { 1943626de996SAl Viro const char *s = get_link(nd); 19445a460275SAl Viro 19453595e234SAl Viro if (unlikely(IS_ERR(s))) 19463595e234SAl Viro return PTR_ERR(s); 1947172a39a0SAl Viro err = 0; 194812b09578SAl Viro if (unlikely(!s)) { 194912b09578SAl Viro /* jumped */ 1950b9ff4429SAl Viro put_link(nd); 195112b09578SAl Viro } else { 1952071bf501SAl Viro nd->stack[nd->depth - 1].name = name; 195332cd7468SAl Viro name = s; 19549e18f10aSAl Viro continue; 195548c8b0c5SAl Viro } 195631e6b01fSNick Piggin } 19573595e234SAl Viro if (unlikely(!d_can_lookup(nd->path.dentry))) 19583595e234SAl Viro return -ENOTDIR; 19595f4a6a69SAl Viro } 1960ce57dfc1SAl Viro } 19611da177e4SLinus Torvalds 1962c8a53ee5SAl Viro static const char *path_init(struct nameidata *nd, unsigned flags) 196331e6b01fSNick Piggin { 196431e6b01fSNick Piggin int retval = 0; 1965c8a53ee5SAl Viro const char *s = nd->name->name; 196631e6b01fSNick Piggin 196731e6b01fSNick Piggin nd->last_type = LAST_ROOT; /* if there are only slashes... */ 1968980f3ea2SAl Viro nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT; 196931e6b01fSNick Piggin nd->depth = 0; 1970368ee9baSAl Viro nd->total_link_count = 0; 19715b6ca027SAl Viro if (flags & LOOKUP_ROOT) { 1972b18825a7SDavid Howells struct dentry *root = nd->root.dentry; 1973b18825a7SDavid Howells struct inode *inode = root->d_inode; 1974fd2f7cb5SAl Viro if (*s) { 197544b1d530SMiklos Szeredi if (!d_can_lookup(root)) 1976368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 19775b6ca027SAl Viro retval = inode_permission(inode, MAY_EXEC); 19785b6ca027SAl Viro if (retval) 1979368ee9baSAl Viro return ERR_PTR(retval); 198073d049a4SAl Viro } 19815b6ca027SAl Viro nd->path = nd->root; 19825b6ca027SAl Viro nd->inode = inode; 19835b6ca027SAl Viro if (flags & LOOKUP_RCU) { 19848b61e74fSAl Viro rcu_read_lock(); 19855b6ca027SAl Viro nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 19868f47a016SAl Viro nd->root_seq = nd->seq; 198748a066e7SAl Viro nd->m_seq = read_seqbegin(&mount_lock); 19885b6ca027SAl Viro } else { 19895b6ca027SAl Viro path_get(&nd->path); 19905b6ca027SAl Viro } 1991368ee9baSAl Viro return s; 19925b6ca027SAl Viro } 19935b6ca027SAl Viro 199431e6b01fSNick Piggin nd->root.mnt = NULL; 199531e6b01fSNick Piggin 199648a066e7SAl Viro nd->m_seq = read_seqbegin(&mount_lock); 1997fd2f7cb5SAl Viro if (*s == '/') { 1998e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 19998b61e74fSAl Viro rcu_read_lock(); 20007bd88377SAl Viro nd->seq = set_root_rcu(nd); 2001e41f7d4eSAl Viro } else { 2002e41f7d4eSAl Viro set_root(nd); 2003e41f7d4eSAl Viro path_get(&nd->root); 2004e41f7d4eSAl Viro } 200531e6b01fSNick Piggin nd->path = nd->root; 2006c8a53ee5SAl Viro } else if (nd->dfd == AT_FDCWD) { 2007e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 200831e6b01fSNick Piggin struct fs_struct *fs = current->fs; 2009c28cc364SNick Piggin unsigned seq; 201031e6b01fSNick Piggin 20118b61e74fSAl Viro rcu_read_lock(); 201231e6b01fSNick Piggin 2013c28cc364SNick Piggin do { 2014c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 201531e6b01fSNick Piggin nd->path = fs->pwd; 2016c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 2017c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 2018e41f7d4eSAl Viro } else { 2019e41f7d4eSAl Viro get_fs_pwd(current->fs, &nd->path); 2020e41f7d4eSAl Viro } 202131e6b01fSNick Piggin } else { 2022582aa64aSJeff Layton /* Caller must check execute permissions on the starting path component */ 2023c8a53ee5SAl Viro struct fd f = fdget_raw(nd->dfd); 202431e6b01fSNick Piggin struct dentry *dentry; 202531e6b01fSNick Piggin 20262903ff01SAl Viro if (!f.file) 2027368ee9baSAl Viro return ERR_PTR(-EBADF); 202831e6b01fSNick Piggin 20292903ff01SAl Viro dentry = f.file->f_path.dentry; 203031e6b01fSNick Piggin 2031fd2f7cb5SAl Viro if (*s) { 203244b1d530SMiklos Szeredi if (!d_can_lookup(dentry)) { 20332903ff01SAl Viro fdput(f); 2034368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 2035f52e0c11SAl Viro } 20362903ff01SAl Viro } 20372903ff01SAl Viro 20382903ff01SAl Viro nd->path = f.file->f_path; 2039e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 20408b61e74fSAl Viro rcu_read_lock(); 204134a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 204234a26b99SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 20435590ff0dSUlrich Drepper } else { 20442903ff01SAl Viro path_get(&nd->path); 204534a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 20461da177e4SLinus Torvalds } 204734a26b99SAl Viro fdput(f); 2048368ee9baSAl Viro return s; 2049e41f7d4eSAl Viro } 2050e41f7d4eSAl Viro 205131e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 20524023bfc9SAl Viro if (!(flags & LOOKUP_RCU)) 2053368ee9baSAl Viro return s; 20544023bfc9SAl Viro if (likely(!read_seqcount_retry(&nd->path.dentry->d_seq, nd->seq))) 2055368ee9baSAl Viro return s; 20564023bfc9SAl Viro if (!(nd->flags & LOOKUP_ROOT)) 20574023bfc9SAl Viro nd->root.mnt = NULL; 20584023bfc9SAl Viro rcu_read_unlock(); 2059368ee9baSAl Viro return ERR_PTR(-ECHILD); 20609b4a9b14SAl Viro } 20619b4a9b14SAl Viro 20623bdba28bSAl Viro static const char *trailing_symlink(struct nameidata *nd) 206395fa25d9SAl Viro { 206495fa25d9SAl Viro const char *s; 2065fec2fa24SAl Viro int error = may_follow_link(nd); 2066deb106c6SAl Viro if (unlikely(error)) 20673bdba28bSAl Viro return ERR_PTR(error); 206895fa25d9SAl Viro nd->flags |= LOOKUP_PARENT; 2069fab51e8aSAl Viro nd->stack[0].name = NULL; 20703b2e7f75SAl Viro s = get_link(nd); 2071deb106c6SAl Viro return s ? s : ""; 207295fa25d9SAl Viro } 207395fa25d9SAl Viro 2074caa85634SAl Viro static inline int lookup_last(struct nameidata *nd) 2075bd92d7feSAl Viro { 2076bd92d7feSAl Viro if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) 2077bd92d7feSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 2078bd92d7feSAl Viro 2079bd92d7feSAl Viro nd->flags &= ~LOOKUP_PARENT; 2080deb106c6SAl Viro return walk_component(nd, 20814693a547SAl Viro nd->flags & LOOKUP_FOLLOW 20824693a547SAl Viro ? nd->depth 20834693a547SAl Viro ? WALK_PUT | WALK_GET 20844693a547SAl Viro : WALK_GET 20854693a547SAl Viro : 0); 2086bd92d7feSAl Viro } 2087bd92d7feSAl Viro 20889b4a9b14SAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2089c8a53ee5SAl Viro static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path) 20909b4a9b14SAl Viro { 2091c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2092bd92d7feSAl Viro int err; 209331e6b01fSNick Piggin 2094368ee9baSAl Viro if (IS_ERR(s)) 2095368ee9baSAl Viro return PTR_ERR(s); 20963bdba28bSAl Viro while (!(err = link_path_walk(s, nd)) 20973bdba28bSAl Viro && ((err = lookup_last(nd)) > 0)) { 20983bdba28bSAl Viro s = trailing_symlink(nd); 20993bdba28bSAl Viro if (IS_ERR(s)) { 21003bdba28bSAl Viro err = PTR_ERR(s); 21016d7b5aaeSAl Viro break; 2102bd92d7feSAl Viro } 2103bd92d7feSAl Viro } 21049f1fafeeSAl Viro if (!err) 21059f1fafeeSAl Viro err = complete_walk(nd); 2106bd92d7feSAl Viro 2107deb106c6SAl Viro if (!err && nd->flags & LOOKUP_DIRECTORY) 2108deb106c6SAl Viro if (!d_can_lookup(nd->path.dentry)) 2109bd23a539SAl Viro err = -ENOTDIR; 2110625b6d10SAl Viro if (!err) { 2111625b6d10SAl Viro *path = nd->path; 2112625b6d10SAl Viro nd->path.mnt = NULL; 2113625b6d10SAl Viro nd->path.dentry = NULL; 2114625b6d10SAl Viro } 2115deb106c6SAl Viro terminate_walk(nd); 2116bd92d7feSAl Viro return err; 211731e6b01fSNick Piggin } 211831e6b01fSNick Piggin 2119625b6d10SAl Viro static int filename_lookup(int dfd, struct filename *name, unsigned flags, 21209ad1aaa6SAl Viro struct path *path, struct path *root) 2121873f1eedSJeff Layton { 2122894bc8c4SAl Viro int retval; 2123abc9f5beSAl Viro struct nameidata nd, *saved_nd; 2124abc9f5beSAl Viro if (IS_ERR(name)) 2125abc9f5beSAl Viro return PTR_ERR(name); 2126c8a53ee5SAl Viro saved_nd = set_nameidata(&nd, dfd, name); 21279ad1aaa6SAl Viro if (unlikely(root)) { 21289ad1aaa6SAl Viro nd.root = *root; 21299ad1aaa6SAl Viro flags |= LOOKUP_ROOT; 21309ad1aaa6SAl Viro } 2131c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_RCU, path); 2132873f1eedSJeff Layton if (unlikely(retval == -ECHILD)) 2133c8a53ee5SAl Viro retval = path_lookupat(&nd, flags, path); 2134873f1eedSJeff Layton if (unlikely(retval == -ESTALE)) 2135c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path); 2136873f1eedSJeff Layton 2137873f1eedSJeff Layton if (likely(!retval)) 2138625b6d10SAl Viro audit_inode(name, path->dentry, flags & LOOKUP_PARENT); 2139756daf26SNeilBrown restore_nameidata(saved_nd); 2140e4bd1c1aSAl Viro putname(name); 2141873f1eedSJeff Layton return retval; 2142873f1eedSJeff Layton } 2143873f1eedSJeff Layton 21448bcb77faSAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2145c8a53ee5SAl Viro static int path_parentat(struct nameidata *nd, unsigned flags, 2146391172c4SAl Viro struct path *parent) 21478bcb77faSAl Viro { 2148c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2149368ee9baSAl Viro int err; 2150368ee9baSAl Viro if (IS_ERR(s)) 2151368ee9baSAl Viro return PTR_ERR(s); 2152368ee9baSAl Viro err = link_path_walk(s, nd); 21538bcb77faSAl Viro if (!err) 21548bcb77faSAl Viro err = complete_walk(nd); 2155391172c4SAl Viro if (!err) { 2156391172c4SAl Viro *parent = nd->path; 2157391172c4SAl Viro nd->path.mnt = NULL; 2158391172c4SAl Viro nd->path.dentry = NULL; 2159391172c4SAl Viro } 2160deb106c6SAl Viro terminate_walk(nd); 21618bcb77faSAl Viro return err; 21628bcb77faSAl Viro } 21638bcb77faSAl Viro 21645c31b6ceSAl Viro static struct filename *filename_parentat(int dfd, struct filename *name, 2165391172c4SAl Viro unsigned int flags, struct path *parent, 2166391172c4SAl Viro struct qstr *last, int *type) 21678bcb77faSAl Viro { 21688bcb77faSAl Viro int retval; 21695c31b6ceSAl Viro struct nameidata nd, *saved_nd; 21708bcb77faSAl Viro 21715c31b6ceSAl Viro if (IS_ERR(name)) 21725c31b6ceSAl Viro return name; 2173c8a53ee5SAl Viro saved_nd = set_nameidata(&nd, dfd, name); 2174c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_RCU, parent); 21758bcb77faSAl Viro if (unlikely(retval == -ECHILD)) 2176c8a53ee5SAl Viro retval = path_parentat(&nd, flags, parent); 21778bcb77faSAl Viro if (unlikely(retval == -ESTALE)) 2178c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent); 2179391172c4SAl Viro if (likely(!retval)) { 2180391172c4SAl Viro *last = nd.last; 2181391172c4SAl Viro *type = nd.last_type; 2182391172c4SAl Viro audit_inode(name, parent->dentry, LOOKUP_PARENT); 21835c31b6ceSAl Viro } else { 21845c31b6ceSAl Viro putname(name); 21855c31b6ceSAl Viro name = ERR_PTR(retval); 2186391172c4SAl Viro } 21878bcb77faSAl Viro restore_nameidata(saved_nd); 21885c31b6ceSAl Viro return name; 21898bcb77faSAl Viro } 21908bcb77faSAl Viro 219179714f72SAl Viro /* does lookup, returns the object with parent locked */ 219279714f72SAl Viro struct dentry *kern_path_locked(const char *name, struct path *path) 21935590ff0dSUlrich Drepper { 21945c31b6ceSAl Viro struct filename *filename; 21955c31b6ceSAl Viro struct dentry *d; 2196391172c4SAl Viro struct qstr last; 2197391172c4SAl Viro int type; 219851689104SPaul Moore 21995c31b6ceSAl Viro filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path, 22005c31b6ceSAl Viro &last, &type); 220151689104SPaul Moore if (IS_ERR(filename)) 220251689104SPaul Moore return ERR_CAST(filename); 22035c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) { 2204391172c4SAl Viro path_put(path); 22055c31b6ceSAl Viro putname(filename); 22065c31b6ceSAl Viro return ERR_PTR(-EINVAL); 220779714f72SAl Viro } 2208391172c4SAl Viro mutex_lock_nested(&path->dentry->d_inode->i_mutex, I_MUTEX_PARENT); 2209391172c4SAl Viro d = __lookup_hash(&last, path->dentry, 0); 221079714f72SAl Viro if (IS_ERR(d)) { 2211391172c4SAl Viro mutex_unlock(&path->dentry->d_inode->i_mutex); 2212391172c4SAl Viro path_put(path); 221379714f72SAl Viro } 221451689104SPaul Moore putname(filename); 221579714f72SAl Viro return d; 22165590ff0dSUlrich Drepper } 22175590ff0dSUlrich Drepper 2218d1811465SAl Viro int kern_path(const char *name, unsigned int flags, struct path *path) 2219d1811465SAl Viro { 2220abc9f5beSAl Viro return filename_lookup(AT_FDCWD, getname_kernel(name), 2221abc9f5beSAl Viro flags, path, NULL); 2222d1811465SAl Viro } 22234d359507SAl Viro EXPORT_SYMBOL(kern_path); 2224d1811465SAl Viro 222516f18200SJosef 'Jeff' Sipek /** 222616f18200SJosef 'Jeff' Sipek * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair 222716f18200SJosef 'Jeff' Sipek * @dentry: pointer to dentry of the base directory 222816f18200SJosef 'Jeff' Sipek * @mnt: pointer to vfs mount of the base directory 222916f18200SJosef 'Jeff' Sipek * @name: pointer to file name 223016f18200SJosef 'Jeff' Sipek * @flags: lookup flags 2231e0a01249SAl Viro * @path: pointer to struct path to fill 223216f18200SJosef 'Jeff' Sipek */ 223316f18200SJosef 'Jeff' Sipek int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, 223416f18200SJosef 'Jeff' Sipek const char *name, unsigned int flags, 2235e0a01249SAl Viro struct path *path) 223616f18200SJosef 'Jeff' Sipek { 22379ad1aaa6SAl Viro struct path root = {.mnt = mnt, .dentry = dentry}; 22389ad1aaa6SAl Viro /* the first argument of filename_lookup() is ignored with root */ 2239abc9f5beSAl Viro return filename_lookup(AT_FDCWD, getname_kernel(name), 2240abc9f5beSAl Viro flags , path, &root); 224116f18200SJosef 'Jeff' Sipek } 22424d359507SAl Viro EXPORT_SYMBOL(vfs_path_lookup); 224316f18200SJosef 'Jeff' Sipek 2244eead1911SChristoph Hellwig /** 2245a6b91919SRandy Dunlap * lookup_one_len - filesystem helper to lookup single pathname component 2246eead1911SChristoph Hellwig * @name: pathname component to lookup 2247eead1911SChristoph Hellwig * @base: base directory to lookup from 2248eead1911SChristoph Hellwig * @len: maximum length @len should be interpreted to 2249eead1911SChristoph Hellwig * 2250a6b91919SRandy Dunlap * Note that this routine is purely a helper for filesystem usage and should 22519e7543e9SAl Viro * not be called by generic code. 2252eead1911SChristoph Hellwig */ 2253057f6c01SJames Morris struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 2254057f6c01SJames Morris { 2255057f6c01SJames Morris struct qstr this; 22566a96ba54SAl Viro unsigned int c; 2257cda309deSMiklos Szeredi int err; 2258057f6c01SJames Morris 22592f9092e1SDavid Woodhouse WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex)); 22602f9092e1SDavid Woodhouse 22616a96ba54SAl Viro this.name = name; 22626a96ba54SAl Viro this.len = len; 22630145acc2SLinus Torvalds this.hash = full_name_hash(name, len); 22646a96ba54SAl Viro if (!len) 22656a96ba54SAl Viro return ERR_PTR(-EACCES); 22666a96ba54SAl Viro 226721d8a15aSAl Viro if (unlikely(name[0] == '.')) { 226821d8a15aSAl Viro if (len < 2 || (len == 2 && name[1] == '.')) 226921d8a15aSAl Viro return ERR_PTR(-EACCES); 227021d8a15aSAl Viro } 227121d8a15aSAl Viro 22726a96ba54SAl Viro while (len--) { 22736a96ba54SAl Viro c = *(const unsigned char *)name++; 22746a96ba54SAl Viro if (c == '/' || c == '\0') 22756a96ba54SAl Viro return ERR_PTR(-EACCES); 22766a96ba54SAl Viro } 22775a202bcdSAl Viro /* 22785a202bcdSAl Viro * See if the low-level filesystem might want 22795a202bcdSAl Viro * to use its own hash.. 22805a202bcdSAl Viro */ 22815a202bcdSAl Viro if (base->d_flags & DCACHE_OP_HASH) { 2282da53be12SLinus Torvalds int err = base->d_op->d_hash(base, &this); 22835a202bcdSAl Viro if (err < 0) 22845a202bcdSAl Viro return ERR_PTR(err); 22855a202bcdSAl Viro } 2286eead1911SChristoph Hellwig 2287cda309deSMiklos Szeredi err = inode_permission(base->d_inode, MAY_EXEC); 2288cda309deSMiklos Szeredi if (err) 2289cda309deSMiklos Szeredi return ERR_PTR(err); 2290cda309deSMiklos Szeredi 229172bd866aSAl Viro return __lookup_hash(&this, base, 0); 2292057f6c01SJames Morris } 22934d359507SAl Viro EXPORT_SYMBOL(lookup_one_len); 2294057f6c01SJames Morris 22951fa1e7f6SAndy Whitcroft int user_path_at_empty(int dfd, const char __user *name, unsigned flags, 22961fa1e7f6SAndy Whitcroft struct path *path, int *empty) 22971da177e4SLinus Torvalds { 2298abc9f5beSAl Viro return filename_lookup(dfd, getname_flags(name, flags, empty), 2299abc9f5beSAl Viro flags, path, NULL); 23001da177e4SLinus Torvalds } 23011da177e4SLinus Torvalds 23021fa1e7f6SAndy Whitcroft int user_path_at(int dfd, const char __user *name, unsigned flags, 23031fa1e7f6SAndy Whitcroft struct path *path) 23041fa1e7f6SAndy Whitcroft { 2305f7493e5dSLinus Torvalds return user_path_at_empty(dfd, name, flags, path, NULL); 23061fa1e7f6SAndy Whitcroft } 23074d359507SAl Viro EXPORT_SYMBOL(user_path_at); 23081fa1e7f6SAndy Whitcroft 2309873f1eedSJeff Layton /* 2310873f1eedSJeff Layton * NB: most callers don't do anything directly with the reference to the 2311873f1eedSJeff Layton * to struct filename, but the nd->last pointer points into the name string 2312873f1eedSJeff Layton * allocated by getname. So we must hold the reference to it until all 2313873f1eedSJeff Layton * path-walking is complete. 2314873f1eedSJeff Layton */ 2315a2ec4a2dSAl Viro static inline struct filename * 2316f5beed75SAl Viro user_path_parent(int dfd, const char __user *path, 2317f5beed75SAl Viro struct path *parent, 2318f5beed75SAl Viro struct qstr *last, 2319f5beed75SAl Viro int *type, 23209e790bd6SJeff Layton unsigned int flags) 23212ad94ae6SAl Viro { 23229e790bd6SJeff Layton /* only LOOKUP_REVAL is allowed in extra flags */ 23235c31b6ceSAl Viro return filename_parentat(dfd, getname(path), flags & LOOKUP_REVAL, 23245c31b6ceSAl Viro parent, last, type); 23252ad94ae6SAl Viro } 23262ad94ae6SAl Viro 23278033426eSJeff Layton /** 2328197df04cSAl Viro * mountpoint_last - look up last component for umount 23298033426eSJeff Layton * @nd: pathwalk nameidata - currently pointing at parent directory of "last" 23308033426eSJeff Layton * @path: pointer to container for result 23318033426eSJeff Layton * 23328033426eSJeff Layton * This is a special lookup_last function just for umount. In this case, we 23338033426eSJeff Layton * need to resolve the path without doing any revalidation. 23348033426eSJeff Layton * 23358033426eSJeff Layton * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since 23368033426eSJeff Layton * mountpoints are always pinned in the dcache, their ancestors are too. Thus, 23378033426eSJeff Layton * in almost all cases, this lookup will be served out of the dcache. The only 23388033426eSJeff Layton * cases where it won't are if nd->last refers to a symlink or the path is 23398033426eSJeff Layton * bogus and it doesn't exist. 23408033426eSJeff Layton * 23418033426eSJeff Layton * Returns: 23428033426eSJeff Layton * -error: if there was an error during lookup. This includes -ENOENT if the 23438033426eSJeff Layton * lookup found a negative dentry. The nd->path reference will also be 23448033426eSJeff Layton * put in this case. 23458033426eSJeff Layton * 23468033426eSJeff Layton * 0: if we successfully resolved nd->path and found it to not to be a 23478033426eSJeff Layton * symlink that needs to be followed. "path" will also be populated. 23488033426eSJeff Layton * The nd->path reference will also be put. 23498033426eSJeff Layton * 23508033426eSJeff Layton * 1: if we successfully resolved nd->last and found it to be a symlink 23518033426eSJeff Layton * that needs to be followed. "path" will be populated with the path 23528033426eSJeff Layton * to the link, and nd->path will *not* be put. 23538033426eSJeff Layton */ 23548033426eSJeff Layton static int 2355197df04cSAl Viro mountpoint_last(struct nameidata *nd, struct path *path) 23568033426eSJeff Layton { 23578033426eSJeff Layton int error = 0; 23588033426eSJeff Layton struct dentry *dentry; 23598033426eSJeff Layton struct dentry *dir = nd->path.dentry; 23608033426eSJeff Layton 236135759521SAl Viro /* If we're in rcuwalk, drop out of it to handle last component */ 236235759521SAl Viro if (nd->flags & LOOKUP_RCU) { 23636e9918b7SAl Viro if (unlazy_walk(nd, NULL, 0)) 2364deb106c6SAl Viro return -ECHILD; 23658033426eSJeff Layton } 23668033426eSJeff Layton 23678033426eSJeff Layton nd->flags &= ~LOOKUP_PARENT; 23688033426eSJeff Layton 23698033426eSJeff Layton if (unlikely(nd->last_type != LAST_NORM)) { 23708033426eSJeff Layton error = handle_dots(nd, nd->last_type); 237135759521SAl Viro if (error) 2372deb106c6SAl Viro return error; 23738033426eSJeff Layton dentry = dget(nd->path.dentry); 237435759521SAl Viro goto done; 23758033426eSJeff Layton } 23768033426eSJeff Layton 23778033426eSJeff Layton mutex_lock(&dir->d_inode->i_mutex); 23788033426eSJeff Layton dentry = d_lookup(dir, &nd->last); 23798033426eSJeff Layton if (!dentry) { 23808033426eSJeff Layton /* 23818033426eSJeff Layton * No cached dentry. Mounted dentries are pinned in the cache, 23828033426eSJeff Layton * so that means that this dentry is probably a symlink or the 23838033426eSJeff Layton * path doesn't actually point to a mounted dentry. 23848033426eSJeff Layton */ 23858033426eSJeff Layton dentry = d_alloc(dir, &nd->last); 23868033426eSJeff Layton if (!dentry) { 2387bcceeebaSDave Jones mutex_unlock(&dir->d_inode->i_mutex); 2388deb106c6SAl Viro return -ENOMEM; 23898033426eSJeff Layton } 239035759521SAl Viro dentry = lookup_real(dir->d_inode, dentry, nd->flags); 2391bcceeebaSDave Jones if (IS_ERR(dentry)) { 2392bcceeebaSDave Jones mutex_unlock(&dir->d_inode->i_mutex); 2393deb106c6SAl Viro return PTR_ERR(dentry); 23948033426eSJeff Layton } 2395bcceeebaSDave Jones } 23968033426eSJeff Layton mutex_unlock(&dir->d_inode->i_mutex); 23978033426eSJeff Layton 239835759521SAl Viro done: 2399698934dfSDavid Howells if (d_is_negative(dentry)) { 24008033426eSJeff Layton dput(dentry); 2401deb106c6SAl Viro return -ENOENT; 240235759521SAl Viro } 2403191d7f73SAl Viro if (nd->depth) 2404191d7f73SAl Viro put_link(nd); 24058033426eSJeff Layton path->dentry = dentry; 2406295dc39dSVasily Averin path->mnt = nd->path.mnt; 2407181548c0SAl Viro error = should_follow_link(nd, path, nd->flags & LOOKUP_FOLLOW, 2408181548c0SAl Viro d_backing_inode(dentry), 0); 2409deb106c6SAl Viro if (unlikely(error)) 2410d63ff28fSAl Viro return error; 2411295dc39dSVasily Averin mntget(path->mnt); 24128033426eSJeff Layton follow_mount(path); 2413deb106c6SAl Viro return 0; 24148033426eSJeff Layton } 24158033426eSJeff Layton 24168033426eSJeff Layton /** 2417197df04cSAl Viro * path_mountpoint - look up a path to be umounted 2418c8a53ee5SAl Viro * @nameidata: lookup context 24198033426eSJeff Layton * @flags: lookup flags 2420c8a53ee5SAl Viro * @path: pointer to container for result 24218033426eSJeff Layton * 24228033426eSJeff Layton * Look up the given name, but don't attempt to revalidate the last component. 2423606d6fe3SRandy Dunlap * Returns 0 and "path" will be valid on success; Returns error otherwise. 24248033426eSJeff Layton */ 24258033426eSJeff Layton static int 2426c8a53ee5SAl Viro path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path) 24278033426eSJeff Layton { 2428c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2429368ee9baSAl Viro int err; 2430368ee9baSAl Viro if (IS_ERR(s)) 2431368ee9baSAl Viro return PTR_ERR(s); 24323bdba28bSAl Viro while (!(err = link_path_walk(s, nd)) && 24333bdba28bSAl Viro (err = mountpoint_last(nd, path)) > 0) { 24343bdba28bSAl Viro s = trailing_symlink(nd); 24353bdba28bSAl Viro if (IS_ERR(s)) { 24363bdba28bSAl Viro err = PTR_ERR(s); 24378033426eSJeff Layton break; 24388033426eSJeff Layton } 24393bdba28bSAl Viro } 2440deb106c6SAl Viro terminate_walk(nd); 24418033426eSJeff Layton return err; 24428033426eSJeff Layton } 24438033426eSJeff Layton 24442d864651SAl Viro static int 2445668696dcSAl Viro filename_mountpoint(int dfd, struct filename *name, struct path *path, 24462d864651SAl Viro unsigned int flags) 24472d864651SAl Viro { 2448756daf26SNeilBrown struct nameidata nd, *saved; 2449cbaab2dbSAl Viro int error; 2450668696dcSAl Viro if (IS_ERR(name)) 2451668696dcSAl Viro return PTR_ERR(name); 2452c8a53ee5SAl Viro saved = set_nameidata(&nd, dfd, name); 2453c8a53ee5SAl Viro error = path_mountpoint(&nd, flags | LOOKUP_RCU, path); 24542d864651SAl Viro if (unlikely(error == -ECHILD)) 2455c8a53ee5SAl Viro error = path_mountpoint(&nd, flags, path); 24562d864651SAl Viro if (unlikely(error == -ESTALE)) 2457c8a53ee5SAl Viro error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path); 24582d864651SAl Viro if (likely(!error)) 2459668696dcSAl Viro audit_inode(name, path->dentry, 0); 2460756daf26SNeilBrown restore_nameidata(saved); 2461668696dcSAl Viro putname(name); 24622d864651SAl Viro return error; 24632d864651SAl Viro } 24642d864651SAl Viro 24658033426eSJeff Layton /** 2466197df04cSAl Viro * user_path_mountpoint_at - lookup a path from userland in order to umount it 24678033426eSJeff Layton * @dfd: directory file descriptor 24688033426eSJeff Layton * @name: pathname from userland 24698033426eSJeff Layton * @flags: lookup flags 24708033426eSJeff Layton * @path: pointer to container to hold result 24718033426eSJeff Layton * 24728033426eSJeff Layton * A umount is a special case for path walking. We're not actually interested 24738033426eSJeff Layton * in the inode in this situation, and ESTALE errors can be a problem. We 24748033426eSJeff Layton * simply want track down the dentry and vfsmount attached at the mountpoint 24758033426eSJeff Layton * and avoid revalidating the last component. 24768033426eSJeff Layton * 24778033426eSJeff Layton * Returns 0 and populates "path" on success. 24788033426eSJeff Layton */ 24798033426eSJeff Layton int 2480197df04cSAl Viro user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags, 24818033426eSJeff Layton struct path *path) 24828033426eSJeff Layton { 2483cbaab2dbSAl Viro return filename_mountpoint(dfd, getname(name), path, flags); 24848033426eSJeff Layton } 24858033426eSJeff Layton 24862d864651SAl Viro int 24872d864651SAl Viro kern_path_mountpoint(int dfd, const char *name, struct path *path, 24882d864651SAl Viro unsigned int flags) 24892d864651SAl Viro { 2490cbaab2dbSAl Viro return filename_mountpoint(dfd, getname_kernel(name), path, flags); 24912d864651SAl Viro } 24922d864651SAl Viro EXPORT_SYMBOL(kern_path_mountpoint); 24932d864651SAl Viro 2494cbdf35bcSMiklos Szeredi int __check_sticky(struct inode *dir, struct inode *inode) 24951da177e4SLinus Torvalds { 24968e96e3b7SEric W. Biederman kuid_t fsuid = current_fsuid(); 2497da9592edSDavid Howells 24988e96e3b7SEric W. Biederman if (uid_eq(inode->i_uid, fsuid)) 24991da177e4SLinus Torvalds return 0; 25008e96e3b7SEric W. Biederman if (uid_eq(dir->i_uid, fsuid)) 25011da177e4SLinus Torvalds return 0; 250223adbe12SAndy Lutomirski return !capable_wrt_inode_uidgid(inode, CAP_FOWNER); 25031da177e4SLinus Torvalds } 2504cbdf35bcSMiklos Szeredi EXPORT_SYMBOL(__check_sticky); 25051da177e4SLinus Torvalds 25061da177e4SLinus Torvalds /* 25071da177e4SLinus Torvalds * Check whether we can remove a link victim from directory dir, check 25081da177e4SLinus Torvalds * whether the type of victim is right. 25091da177e4SLinus Torvalds * 1. We can't do it if dir is read-only (done in permission()) 25101da177e4SLinus Torvalds * 2. We should have write and exec permissions on dir 25111da177e4SLinus Torvalds * 3. We can't remove anything from append-only dir 25121da177e4SLinus Torvalds * 4. We can't do anything with immutable dir (done in permission()) 25131da177e4SLinus Torvalds * 5. If the sticky bit on dir is set we should either 25141da177e4SLinus Torvalds * a. be owner of dir, or 25151da177e4SLinus Torvalds * b. be owner of victim, or 25161da177e4SLinus Torvalds * c. have CAP_FOWNER capability 25171da177e4SLinus Torvalds * 6. If the victim is append-only or immutable we can't do antyhing with 25181da177e4SLinus Torvalds * links pointing to it. 25191da177e4SLinus Torvalds * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. 25201da177e4SLinus Torvalds * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. 25211da177e4SLinus Torvalds * 9. We can't remove a root or mountpoint. 25221da177e4SLinus Torvalds * 10. We don't allow removal of NFS sillyrenamed files; it's handled by 25231da177e4SLinus Torvalds * nfs_async_unlink(). 25241da177e4SLinus Torvalds */ 2525b18825a7SDavid Howells static int may_delete(struct inode *dir, struct dentry *victim, bool isdir) 25261da177e4SLinus Torvalds { 252763afdfc7SDavid Howells struct inode *inode = d_backing_inode(victim); 25281da177e4SLinus Torvalds int error; 25291da177e4SLinus Torvalds 2530b18825a7SDavid Howells if (d_is_negative(victim)) 25311da177e4SLinus Torvalds return -ENOENT; 2532b18825a7SDavid Howells BUG_ON(!inode); 25331da177e4SLinus Torvalds 25341da177e4SLinus Torvalds BUG_ON(victim->d_parent->d_inode != dir); 25354fa6b5ecSJeff Layton audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); 25361da177e4SLinus Torvalds 2537f419a2e3SAl Viro error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 25381da177e4SLinus Torvalds if (error) 25391da177e4SLinus Torvalds return error; 25401da177e4SLinus Torvalds if (IS_APPEND(dir)) 25411da177e4SLinus Torvalds return -EPERM; 2542b18825a7SDavid Howells 2543b18825a7SDavid Howells if (check_sticky(dir, inode) || IS_APPEND(inode) || 2544b18825a7SDavid Howells IS_IMMUTABLE(inode) || IS_SWAPFILE(inode)) 25451da177e4SLinus Torvalds return -EPERM; 25461da177e4SLinus Torvalds if (isdir) { 254744b1d530SMiklos Szeredi if (!d_is_dir(victim)) 25481da177e4SLinus Torvalds return -ENOTDIR; 25491da177e4SLinus Torvalds if (IS_ROOT(victim)) 25501da177e4SLinus Torvalds return -EBUSY; 255144b1d530SMiklos Szeredi } else if (d_is_dir(victim)) 25521da177e4SLinus Torvalds return -EISDIR; 25531da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 25541da177e4SLinus Torvalds return -ENOENT; 25551da177e4SLinus Torvalds if (victim->d_flags & DCACHE_NFSFS_RENAMED) 25561da177e4SLinus Torvalds return -EBUSY; 25571da177e4SLinus Torvalds return 0; 25581da177e4SLinus Torvalds } 25591da177e4SLinus Torvalds 25601da177e4SLinus Torvalds /* Check whether we can create an object with dentry child in directory 25611da177e4SLinus Torvalds * dir. 25621da177e4SLinus Torvalds * 1. We can't do it if child already exists (open has special treatment for 25631da177e4SLinus Torvalds * this case, but since we are inlined it's OK) 25641da177e4SLinus Torvalds * 2. We can't do it if dir is read-only (done in permission()) 25651da177e4SLinus Torvalds * 3. We should have write and exec permissions on dir 25661da177e4SLinus Torvalds * 4. We can't do it if dir is immutable (done in permission()) 25671da177e4SLinus Torvalds */ 2568a95164d9SMiklos Szeredi static inline int may_create(struct inode *dir, struct dentry *child) 25691da177e4SLinus Torvalds { 257014e972b4SJeff Layton audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); 25711da177e4SLinus Torvalds if (child->d_inode) 25721da177e4SLinus Torvalds return -EEXIST; 25731da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 25741da177e4SLinus Torvalds return -ENOENT; 2575f419a2e3SAl Viro return inode_permission(dir, MAY_WRITE | MAY_EXEC); 25761da177e4SLinus Torvalds } 25771da177e4SLinus Torvalds 25781da177e4SLinus Torvalds /* 25791da177e4SLinus Torvalds * p1 and p2 should be directories on the same fs. 25801da177e4SLinus Torvalds */ 25811da177e4SLinus Torvalds struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 25821da177e4SLinus Torvalds { 25831da177e4SLinus Torvalds struct dentry *p; 25841da177e4SLinus Torvalds 25851da177e4SLinus Torvalds if (p1 == p2) { 2586f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 25871da177e4SLinus Torvalds return NULL; 25881da177e4SLinus Torvalds } 25891da177e4SLinus Torvalds 2590a11f3a05SArjan van de Ven mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 25911da177e4SLinus Torvalds 2592e2761a11SOGAWA Hirofumi p = d_ancestor(p2, p1); 2593e2761a11SOGAWA Hirofumi if (p) { 2594f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT); 2595f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD); 25961da177e4SLinus Torvalds return p; 25971da177e4SLinus Torvalds } 25981da177e4SLinus Torvalds 2599e2761a11SOGAWA Hirofumi p = d_ancestor(p1, p2); 2600e2761a11SOGAWA Hirofumi if (p) { 2601f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 2602f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); 26031da177e4SLinus Torvalds return p; 26041da177e4SLinus Torvalds } 26051da177e4SLinus Torvalds 2606f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 2607d1b72cc6SMiklos Szeredi mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT2); 26081da177e4SLinus Torvalds return NULL; 26091da177e4SLinus Torvalds } 26104d359507SAl Viro EXPORT_SYMBOL(lock_rename); 26111da177e4SLinus Torvalds 26121da177e4SLinus Torvalds void unlock_rename(struct dentry *p1, struct dentry *p2) 26131da177e4SLinus Torvalds { 26141b1dcc1bSJes Sorensen mutex_unlock(&p1->d_inode->i_mutex); 26151da177e4SLinus Torvalds if (p1 != p2) { 26161b1dcc1bSJes Sorensen mutex_unlock(&p2->d_inode->i_mutex); 2617a11f3a05SArjan van de Ven mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 26181da177e4SLinus Torvalds } 26191da177e4SLinus Torvalds } 26204d359507SAl Viro EXPORT_SYMBOL(unlock_rename); 26211da177e4SLinus Torvalds 26224acdaf27SAl Viro int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, 2623312b63fbSAl Viro bool want_excl) 26241da177e4SLinus Torvalds { 2625a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 26261da177e4SLinus Torvalds if (error) 26271da177e4SLinus Torvalds return error; 26281da177e4SLinus Torvalds 2629acfa4380SAl Viro if (!dir->i_op->create) 26301da177e4SLinus Torvalds return -EACCES; /* shouldn't it be ENOSYS? */ 26311da177e4SLinus Torvalds mode &= S_IALLUGO; 26321da177e4SLinus Torvalds mode |= S_IFREG; 26331da177e4SLinus Torvalds error = security_inode_create(dir, dentry, mode); 26341da177e4SLinus Torvalds if (error) 26351da177e4SLinus Torvalds return error; 2636312b63fbSAl Viro error = dir->i_op->create(dir, dentry, mode, want_excl); 2637a74574aaSStephen Smalley if (!error) 2638f38aa942SAmy Griffis fsnotify_create(dir, dentry); 26391da177e4SLinus Torvalds return error; 26401da177e4SLinus Torvalds } 26414d359507SAl Viro EXPORT_SYMBOL(vfs_create); 26421da177e4SLinus Torvalds 264373d049a4SAl Viro static int may_open(struct path *path, int acc_mode, int flag) 26441da177e4SLinus Torvalds { 26453fb64190SChristoph Hellwig struct dentry *dentry = path->dentry; 26461da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 26471da177e4SLinus Torvalds int error; 26481da177e4SLinus Torvalds 2649bcda7652SAl Viro /* O_PATH? */ 2650bcda7652SAl Viro if (!acc_mode) 2651bcda7652SAl Viro return 0; 2652bcda7652SAl Viro 26531da177e4SLinus Torvalds if (!inode) 26541da177e4SLinus Torvalds return -ENOENT; 26551da177e4SLinus Torvalds 2656c8fe8f30SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 2657c8fe8f30SChristoph Hellwig case S_IFLNK: 26581da177e4SLinus Torvalds return -ELOOP; 2659c8fe8f30SChristoph Hellwig case S_IFDIR: 2660c8fe8f30SChristoph Hellwig if (acc_mode & MAY_WRITE) 26611da177e4SLinus Torvalds return -EISDIR; 2662c8fe8f30SChristoph Hellwig break; 2663c8fe8f30SChristoph Hellwig case S_IFBLK: 2664c8fe8f30SChristoph Hellwig case S_IFCHR: 26653fb64190SChristoph Hellwig if (path->mnt->mnt_flags & MNT_NODEV) 26661da177e4SLinus Torvalds return -EACCES; 2667c8fe8f30SChristoph Hellwig /*FALLTHRU*/ 2668c8fe8f30SChristoph Hellwig case S_IFIFO: 2669c8fe8f30SChristoph Hellwig case S_IFSOCK: 26701da177e4SLinus Torvalds flag &= ~O_TRUNC; 2671c8fe8f30SChristoph Hellwig break; 26724a3fd211SDave Hansen } 2673b41572e9SDave Hansen 26743fb64190SChristoph Hellwig error = inode_permission(inode, acc_mode); 2675b41572e9SDave Hansen if (error) 2676b41572e9SDave Hansen return error; 26776146f0d5SMimi Zohar 26781da177e4SLinus Torvalds /* 26791da177e4SLinus Torvalds * An append-only file must be opened in append mode for writing. 26801da177e4SLinus Torvalds */ 26811da177e4SLinus Torvalds if (IS_APPEND(inode)) { 26828737c930SAl Viro if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) 26837715b521SAl Viro return -EPERM; 26841da177e4SLinus Torvalds if (flag & O_TRUNC) 26857715b521SAl Viro return -EPERM; 26861da177e4SLinus Torvalds } 26871da177e4SLinus Torvalds 26881da177e4SLinus Torvalds /* O_NOATIME can only be set by the owner or superuser */ 26892e149670SSerge E. Hallyn if (flag & O_NOATIME && !inode_owner_or_capable(inode)) 26907715b521SAl Viro return -EPERM; 26911da177e4SLinus Torvalds 2692f3c7691eSJ. Bruce Fields return 0; 26937715b521SAl Viro } 26947715b521SAl Viro 2695e1181ee6SJeff Layton static int handle_truncate(struct file *filp) 26967715b521SAl Viro { 2697e1181ee6SJeff Layton struct path *path = &filp->f_path; 26987715b521SAl Viro struct inode *inode = path->dentry->d_inode; 26997715b521SAl Viro int error = get_write_access(inode); 27001da177e4SLinus Torvalds if (error) 27017715b521SAl Viro return error; 27021da177e4SLinus Torvalds /* 27031da177e4SLinus Torvalds * Refuse to truncate files with mandatory locks held on them. 27041da177e4SLinus Torvalds */ 2705d7a06983SJeff Layton error = locks_verify_locked(filp); 2706be6d3e56SKentaro Takeda if (!error) 2707ea0d3ab2STetsuo Handa error = security_path_truncate(path); 27081da177e4SLinus Torvalds if (!error) { 27097715b521SAl Viro error = do_truncate(path->dentry, 0, 2710d139d7ffSMiklos Szeredi ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, 2711e1181ee6SJeff Layton filp); 27121da177e4SLinus Torvalds } 27131da177e4SLinus Torvalds put_write_access(inode); 2714acd0c935SMimi Zohar return error; 27151da177e4SLinus Torvalds } 27161da177e4SLinus Torvalds 2717d57999e1SDave Hansen static inline int open_to_namei_flags(int flag) 2718d57999e1SDave Hansen { 27198a5e929dSAl Viro if ((flag & O_ACCMODE) == 3) 27208a5e929dSAl Viro flag--; 2721d57999e1SDave Hansen return flag; 2722d57999e1SDave Hansen } 2723d57999e1SDave Hansen 2724d18e9008SMiklos Szeredi static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode) 2725d18e9008SMiklos Szeredi { 2726d18e9008SMiklos Szeredi int error = security_path_mknod(dir, dentry, mode, 0); 2727d18e9008SMiklos Szeredi if (error) 2728d18e9008SMiklos Szeredi return error; 2729d18e9008SMiklos Szeredi 2730d18e9008SMiklos Szeredi error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC); 2731d18e9008SMiklos Szeredi if (error) 2732d18e9008SMiklos Szeredi return error; 2733d18e9008SMiklos Szeredi 2734d18e9008SMiklos Szeredi return security_inode_create(dir->dentry->d_inode, dentry, mode); 2735d18e9008SMiklos Szeredi } 2736d18e9008SMiklos Szeredi 27371acf0af9SDavid Howells /* 27381acf0af9SDavid Howells * Attempt to atomically look up, create and open a file from a negative 27391acf0af9SDavid Howells * dentry. 27401acf0af9SDavid Howells * 27411acf0af9SDavid Howells * Returns 0 if successful. The file will have been created and attached to 27421acf0af9SDavid Howells * @file by the filesystem calling finish_open(). 27431acf0af9SDavid Howells * 27441acf0af9SDavid Howells * Returns 1 if the file was looked up only or didn't need creating. The 27451acf0af9SDavid Howells * caller will need to perform the open themselves. @path will have been 27461acf0af9SDavid Howells * updated to point to the new dentry. This may be negative. 27471acf0af9SDavid Howells * 27481acf0af9SDavid Howells * Returns an error code otherwise. 27491acf0af9SDavid Howells */ 27502675a4ebSAl Viro static int atomic_open(struct nameidata *nd, struct dentry *dentry, 275130d90494SAl Viro struct path *path, struct file *file, 2752015c3bbcSMiklos Szeredi const struct open_flags *op, 275364894cf8SAl Viro bool got_write, bool need_lookup, 275447237687SAl Viro int *opened) 2755d18e9008SMiklos Szeredi { 2756d18e9008SMiklos Szeredi struct inode *dir = nd->path.dentry->d_inode; 2757d18e9008SMiklos Szeredi unsigned open_flag = open_to_namei_flags(op->open_flag); 2758d18e9008SMiklos Szeredi umode_t mode; 2759d18e9008SMiklos Szeredi int error; 2760d18e9008SMiklos Szeredi int acc_mode; 2761d18e9008SMiklos Szeredi int create_error = 0; 2762d18e9008SMiklos Szeredi struct dentry *const DENTRY_NOT_SET = (void *) -1UL; 2763116cc022SMiklos Szeredi bool excl; 2764d18e9008SMiklos Szeredi 2765d18e9008SMiklos Szeredi BUG_ON(dentry->d_inode); 2766d18e9008SMiklos Szeredi 2767d18e9008SMiklos Szeredi /* Don't create child dentry for a dead directory. */ 2768d18e9008SMiklos Szeredi if (unlikely(IS_DEADDIR(dir))) { 27692675a4ebSAl Viro error = -ENOENT; 2770d18e9008SMiklos Szeredi goto out; 2771d18e9008SMiklos Szeredi } 2772d18e9008SMiklos Szeredi 277362b259d8SMiklos Szeredi mode = op->mode; 2774d18e9008SMiklos Szeredi if ((open_flag & O_CREAT) && !IS_POSIXACL(dir)) 2775d18e9008SMiklos Szeredi mode &= ~current_umask(); 2776d18e9008SMiklos Szeredi 2777116cc022SMiklos Szeredi excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT); 2778116cc022SMiklos Szeredi if (excl) 2779d18e9008SMiklos Szeredi open_flag &= ~O_TRUNC; 2780d18e9008SMiklos Szeredi 2781d18e9008SMiklos Szeredi /* 2782d18e9008SMiklos Szeredi * Checking write permission is tricky, bacuse we don't know if we are 2783d18e9008SMiklos Szeredi * going to actually need it: O_CREAT opens should work as long as the 2784d18e9008SMiklos Szeredi * file exists. But checking existence breaks atomicity. The trick is 2785d18e9008SMiklos Szeredi * to check access and if not granted clear O_CREAT from the flags. 2786d18e9008SMiklos Szeredi * 2787d18e9008SMiklos Szeredi * Another problem is returing the "right" error value (e.g. for an 2788d18e9008SMiklos Szeredi * O_EXCL open we want to return EEXIST not EROFS). 2789d18e9008SMiklos Szeredi */ 279064894cf8SAl Viro if (((open_flag & (O_CREAT | O_TRUNC)) || 279164894cf8SAl Viro (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) { 279264894cf8SAl Viro if (!(open_flag & O_CREAT)) { 2793d18e9008SMiklos Szeredi /* 2794d18e9008SMiklos Szeredi * No O_CREATE -> atomicity not a requirement -> fall 2795d18e9008SMiklos Szeredi * back to lookup + open 2796d18e9008SMiklos Szeredi */ 2797d18e9008SMiklos Szeredi goto no_open; 2798d18e9008SMiklos Szeredi } else if (open_flag & (O_EXCL | O_TRUNC)) { 2799d18e9008SMiklos Szeredi /* Fall back and fail with the right error */ 280064894cf8SAl Viro create_error = -EROFS; 2801d18e9008SMiklos Szeredi goto no_open; 2802d18e9008SMiklos Szeredi } else { 2803d18e9008SMiklos Szeredi /* No side effects, safe to clear O_CREAT */ 280464894cf8SAl Viro create_error = -EROFS; 2805d18e9008SMiklos Szeredi open_flag &= ~O_CREAT; 2806d18e9008SMiklos Szeredi } 2807d18e9008SMiklos Szeredi } 2808d18e9008SMiklos Szeredi 2809d18e9008SMiklos Szeredi if (open_flag & O_CREAT) { 281038227f78SMiklos Szeredi error = may_o_create(&nd->path, dentry, mode); 2811d18e9008SMiklos Szeredi if (error) { 2812d18e9008SMiklos Szeredi create_error = error; 2813d18e9008SMiklos Szeredi if (open_flag & O_EXCL) 2814d18e9008SMiklos Szeredi goto no_open; 2815d18e9008SMiklos Szeredi open_flag &= ~O_CREAT; 2816d18e9008SMiklos Szeredi } 2817d18e9008SMiklos Szeredi } 2818d18e9008SMiklos Szeredi 2819d18e9008SMiklos Szeredi if (nd->flags & LOOKUP_DIRECTORY) 2820d18e9008SMiklos Szeredi open_flag |= O_DIRECTORY; 2821d18e9008SMiklos Szeredi 282230d90494SAl Viro file->f_path.dentry = DENTRY_NOT_SET; 282330d90494SAl Viro file->f_path.mnt = nd->path.mnt; 282430d90494SAl Viro error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode, 282547237687SAl Viro opened); 2826d9585277SAl Viro if (error < 0) { 2827d9585277SAl Viro if (create_error && error == -ENOENT) 2828d9585277SAl Viro error = create_error; 2829d18e9008SMiklos Szeredi goto out; 2830d18e9008SMiklos Szeredi } 2831d18e9008SMiklos Szeredi 2832d9585277SAl Viro if (error) { /* returned 1, that is */ 283330d90494SAl Viro if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { 28342675a4ebSAl Viro error = -EIO; 2835d18e9008SMiklos Szeredi goto out; 2836d18e9008SMiklos Szeredi } 283730d90494SAl Viro if (file->f_path.dentry) { 2838d18e9008SMiklos Szeredi dput(dentry); 283930d90494SAl Viro dentry = file->f_path.dentry; 2840d18e9008SMiklos Szeredi } 284103da633aSAl Viro if (*opened & FILE_CREATED) 284203da633aSAl Viro fsnotify_create(dir, dentry); 284303da633aSAl Viro if (!dentry->d_inode) { 284403da633aSAl Viro WARN_ON(*opened & FILE_CREATED); 284503da633aSAl Viro if (create_error) { 284662b2ce96SSage Weil error = create_error; 284762b2ce96SSage Weil goto out; 284862b2ce96SSage Weil } 284903da633aSAl Viro } else { 285003da633aSAl Viro if (excl && !(*opened & FILE_CREATED)) { 285103da633aSAl Viro error = -EEXIST; 285203da633aSAl Viro goto out; 285303da633aSAl Viro } 285403da633aSAl Viro } 2855d18e9008SMiklos Szeredi goto looked_up; 2856d18e9008SMiklos Szeredi } 2857d18e9008SMiklos Szeredi 2858d18e9008SMiklos Szeredi /* 2859d18e9008SMiklos Szeredi * We didn't have the inode before the open, so check open permission 2860d18e9008SMiklos Szeredi * here. 2861d18e9008SMiklos Szeredi */ 286203da633aSAl Viro acc_mode = op->acc_mode; 286303da633aSAl Viro if (*opened & FILE_CREATED) { 286403da633aSAl Viro WARN_ON(!(open_flag & O_CREAT)); 286503da633aSAl Viro fsnotify_create(dir, dentry); 286603da633aSAl Viro acc_mode = MAY_OPEN; 286703da633aSAl Viro } 28682675a4ebSAl Viro error = may_open(&file->f_path, acc_mode, open_flag); 28692675a4ebSAl Viro if (error) 28702675a4ebSAl Viro fput(file); 2871d18e9008SMiklos Szeredi 2872d18e9008SMiklos Szeredi out: 2873d18e9008SMiklos Szeredi dput(dentry); 28742675a4ebSAl Viro return error; 2875d18e9008SMiklos Szeredi 2876d18e9008SMiklos Szeredi no_open: 2877d18e9008SMiklos Szeredi if (need_lookup) { 287872bd866aSAl Viro dentry = lookup_real(dir, dentry, nd->flags); 2879d18e9008SMiklos Szeredi if (IS_ERR(dentry)) 28802675a4ebSAl Viro return PTR_ERR(dentry); 2881d18e9008SMiklos Szeredi 2882d18e9008SMiklos Szeredi if (create_error) { 2883d18e9008SMiklos Szeredi int open_flag = op->open_flag; 2884d18e9008SMiklos Szeredi 28852675a4ebSAl Viro error = create_error; 2886d18e9008SMiklos Szeredi if ((open_flag & O_EXCL)) { 2887d18e9008SMiklos Szeredi if (!dentry->d_inode) 2888d18e9008SMiklos Szeredi goto out; 2889d18e9008SMiklos Szeredi } else if (!dentry->d_inode) { 2890d18e9008SMiklos Szeredi goto out; 2891d18e9008SMiklos Szeredi } else if ((open_flag & O_TRUNC) && 2892e36cb0b8SDavid Howells d_is_reg(dentry)) { 2893d18e9008SMiklos Szeredi goto out; 2894d18e9008SMiklos Szeredi } 2895d18e9008SMiklos Szeredi /* will fail later, go on to get the right error */ 2896d18e9008SMiklos Szeredi } 2897d18e9008SMiklos Szeredi } 2898d18e9008SMiklos Szeredi looked_up: 2899d18e9008SMiklos Szeredi path->dentry = dentry; 2900d18e9008SMiklos Szeredi path->mnt = nd->path.mnt; 29012675a4ebSAl Viro return 1; 2902d18e9008SMiklos Szeredi } 2903d18e9008SMiklos Szeredi 290431e6b01fSNick Piggin /* 29051acf0af9SDavid Howells * Look up and maybe create and open the last component. 2906d58ffd35SMiklos Szeredi * 2907d58ffd35SMiklos Szeredi * Must be called with i_mutex held on parent. 2908d58ffd35SMiklos Szeredi * 29091acf0af9SDavid Howells * Returns 0 if the file was successfully atomically created (if necessary) and 29101acf0af9SDavid Howells * opened. In this case the file will be returned attached to @file. 29111acf0af9SDavid Howells * 29121acf0af9SDavid Howells * Returns 1 if the file was not completely opened at this time, though lookups 29131acf0af9SDavid Howells * and creations will have been performed and the dentry returned in @path will 29141acf0af9SDavid Howells * be positive upon return if O_CREAT was specified. If O_CREAT wasn't 29151acf0af9SDavid Howells * specified then a negative dentry may be returned. 29161acf0af9SDavid Howells * 29171acf0af9SDavid Howells * An error code is returned otherwise. 29181acf0af9SDavid Howells * 29191acf0af9SDavid Howells * FILE_CREATE will be set in @*opened if the dentry was created and will be 29201acf0af9SDavid Howells * cleared otherwise prior to returning. 2921d58ffd35SMiklos Szeredi */ 29222675a4ebSAl Viro static int lookup_open(struct nameidata *nd, struct path *path, 292330d90494SAl Viro struct file *file, 2924d58ffd35SMiklos Szeredi const struct open_flags *op, 292564894cf8SAl Viro bool got_write, int *opened) 2926d58ffd35SMiklos Szeredi { 2927d58ffd35SMiklos Szeredi struct dentry *dir = nd->path.dentry; 292854ef4872SMiklos Szeredi struct inode *dir_inode = dir->d_inode; 2929d58ffd35SMiklos Szeredi struct dentry *dentry; 2930d58ffd35SMiklos Szeredi int error; 293154ef4872SMiklos Szeredi bool need_lookup; 2932d58ffd35SMiklos Szeredi 293347237687SAl Viro *opened &= ~FILE_CREATED; 2934201f956eSAl Viro dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup); 2935d58ffd35SMiklos Szeredi if (IS_ERR(dentry)) 29362675a4ebSAl Viro return PTR_ERR(dentry); 2937d58ffd35SMiklos Szeredi 2938d18e9008SMiklos Szeredi /* Cached positive dentry: will open in f_op->open */ 2939d18e9008SMiklos Szeredi if (!need_lookup && dentry->d_inode) 2940d18e9008SMiklos Szeredi goto out_no_open; 2941d18e9008SMiklos Szeredi 2942d18e9008SMiklos Szeredi if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) { 294364894cf8SAl Viro return atomic_open(nd, dentry, path, file, op, got_write, 294447237687SAl Viro need_lookup, opened); 2945d18e9008SMiklos Szeredi } 2946d18e9008SMiklos Szeredi 294754ef4872SMiklos Szeredi if (need_lookup) { 294854ef4872SMiklos Szeredi BUG_ON(dentry->d_inode); 294954ef4872SMiklos Szeredi 295072bd866aSAl Viro dentry = lookup_real(dir_inode, dentry, nd->flags); 295154ef4872SMiklos Szeredi if (IS_ERR(dentry)) 29522675a4ebSAl Viro return PTR_ERR(dentry); 295354ef4872SMiklos Szeredi } 295454ef4872SMiklos Szeredi 2955d58ffd35SMiklos Szeredi /* Negative dentry, just create the file */ 2956d58ffd35SMiklos Szeredi if (!dentry->d_inode && (op->open_flag & O_CREAT)) { 2957d58ffd35SMiklos Szeredi umode_t mode = op->mode; 2958d58ffd35SMiklos Szeredi if (!IS_POSIXACL(dir->d_inode)) 2959d58ffd35SMiklos Szeredi mode &= ~current_umask(); 2960d58ffd35SMiklos Szeredi /* 2961d58ffd35SMiklos Szeredi * This write is needed to ensure that a 2962d58ffd35SMiklos Szeredi * rw->ro transition does not occur between 2963d58ffd35SMiklos Szeredi * the time when the file is created and when 2964d58ffd35SMiklos Szeredi * a permanent write count is taken through 2965015c3bbcSMiklos Szeredi * the 'struct file' in finish_open(). 2966d58ffd35SMiklos Szeredi */ 296764894cf8SAl Viro if (!got_write) { 296864894cf8SAl Viro error = -EROFS; 2969d58ffd35SMiklos Szeredi goto out_dput; 297064894cf8SAl Viro } 297147237687SAl Viro *opened |= FILE_CREATED; 2972d58ffd35SMiklos Szeredi error = security_path_mknod(&nd->path, dentry, mode, 0); 2973d58ffd35SMiklos Szeredi if (error) 2974d58ffd35SMiklos Szeredi goto out_dput; 2975312b63fbSAl Viro error = vfs_create(dir->d_inode, dentry, mode, 2976312b63fbSAl Viro nd->flags & LOOKUP_EXCL); 2977d58ffd35SMiklos Szeredi if (error) 2978d58ffd35SMiklos Szeredi goto out_dput; 2979d58ffd35SMiklos Szeredi } 2980d18e9008SMiklos Szeredi out_no_open: 2981d58ffd35SMiklos Szeredi path->dentry = dentry; 2982d58ffd35SMiklos Szeredi path->mnt = nd->path.mnt; 29832675a4ebSAl Viro return 1; 2984d58ffd35SMiklos Szeredi 2985d58ffd35SMiklos Szeredi out_dput: 2986d58ffd35SMiklos Szeredi dput(dentry); 29872675a4ebSAl Viro return error; 2988d58ffd35SMiklos Szeredi } 2989d58ffd35SMiklos Szeredi 2990d58ffd35SMiklos Szeredi /* 2991fe2d35ffSAl Viro * Handle the last step of open() 299231e6b01fSNick Piggin */ 2993896475d5SAl Viro static int do_last(struct nameidata *nd, 299430d90494SAl Viro struct file *file, const struct open_flags *op, 299576ae2a5aSAl Viro int *opened) 2996fb1cc555SAl Viro { 2997a1e28038SAl Viro struct dentry *dir = nd->path.dentry; 2998ca344a89SAl Viro int open_flag = op->open_flag; 299977d660a8SMiklos Szeredi bool will_truncate = (open_flag & O_TRUNC) != 0; 300064894cf8SAl Viro bool got_write = false; 3001bcda7652SAl Viro int acc_mode = op->acc_mode; 3002254cf582SAl Viro unsigned seq; 3003a1eb3315SMiklos Szeredi struct inode *inode; 300416b1c1cdSMiklos Szeredi struct path save_parent = { .dentry = NULL, .mnt = NULL }; 3005896475d5SAl Viro struct path path; 300616b1c1cdSMiklos Szeredi bool retried = false; 300716c2cd71SAl Viro int error; 3008fb1cc555SAl Viro 3009c3e380b0SAl Viro nd->flags &= ~LOOKUP_PARENT; 3010c3e380b0SAl Viro nd->flags |= op->intent; 3011c3e380b0SAl Viro 3012bc77daa7SAl Viro if (nd->last_type != LAST_NORM) { 3013fe2d35ffSAl Viro error = handle_dots(nd, nd->last_type); 3014deb106c6SAl Viro if (unlikely(error)) 30152675a4ebSAl Viro return error; 3016e83db167SMiklos Szeredi goto finish_open; 30171f36f774SAl Viro } 3018a2c36b45SAl Viro 3019ca344a89SAl Viro if (!(open_flag & O_CREAT)) { 3020fe2d35ffSAl Viro if (nd->last.name[nd->last.len]) 3021fe2d35ffSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 3022fe2d35ffSAl Viro /* we _can_ be in RCU mode here */ 3023254cf582SAl Viro error = lookup_fast(nd, &path, &inode, &seq); 302471574865SMiklos Szeredi if (likely(!error)) 302571574865SMiklos Szeredi goto finish_lookup; 302671574865SMiklos Szeredi 3027ce57dfc1SAl Viro if (error < 0) 3028deb106c6SAl Viro return error; 3029a1eb3315SMiklos Szeredi 303037d7fffcSMiklos Szeredi BUG_ON(nd->inode != dir->d_inode); 3031b6183df7SMiklos Szeredi } else { 3032fe2d35ffSAl Viro /* create side of things */ 3033a3fbbde7SAl Viro /* 3034b6183df7SMiklos Szeredi * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED 3035b6183df7SMiklos Szeredi * has been cleared when we got to the last component we are 3036b6183df7SMiklos Szeredi * about to look up 3037a3fbbde7SAl Viro */ 30389f1fafeeSAl Viro error = complete_walk(nd); 3039e8bb73dfSAl Viro if (error) 30402675a4ebSAl Viro return error; 3041fe2d35ffSAl Viro 304276ae2a5aSAl Viro audit_inode(nd->name, dir, LOOKUP_PARENT); 30431f36f774SAl Viro /* trailing slashes? */ 3044deb106c6SAl Viro if (unlikely(nd->last.name[nd->last.len])) 3045deb106c6SAl Viro return -EISDIR; 3046b6183df7SMiklos Szeredi } 30471f36f774SAl Viro 304816b1c1cdSMiklos Szeredi retry_lookup: 304964894cf8SAl Viro if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { 305064894cf8SAl Viro error = mnt_want_write(nd->path.mnt); 305164894cf8SAl Viro if (!error) 305264894cf8SAl Viro got_write = true; 305364894cf8SAl Viro /* 305464894cf8SAl Viro * do _not_ fail yet - we might not need that or fail with 305564894cf8SAl Viro * a different error; let lookup_open() decide; we'll be 305664894cf8SAl Viro * dropping this one anyway. 305764894cf8SAl Viro */ 305864894cf8SAl Viro } 3059a1e28038SAl Viro mutex_lock(&dir->d_inode->i_mutex); 3060896475d5SAl Viro error = lookup_open(nd, &path, file, op, got_write, opened); 3061fb1cc555SAl Viro mutex_unlock(&dir->d_inode->i_mutex); 3062fb1cc555SAl Viro 30632675a4ebSAl Viro if (error <= 0) { 30642675a4ebSAl Viro if (error) 3065d58ffd35SMiklos Szeredi goto out; 30666c0d46c4SAl Viro 306747237687SAl Viro if ((*opened & FILE_CREATED) || 3068496ad9aaSAl Viro !S_ISREG(file_inode(file)->i_mode)) 306977d660a8SMiklos Szeredi will_truncate = false; 3070d18e9008SMiklos Szeredi 307176ae2a5aSAl Viro audit_inode(nd->name, file->f_path.dentry, 0); 3072d18e9008SMiklos Szeredi goto opened; 3073d18e9008SMiklos Szeredi } 3074d18e9008SMiklos Szeredi 307547237687SAl Viro if (*opened & FILE_CREATED) { 30769b44f1b3SAl Viro /* Don't check for write permission, don't truncate */ 3077ca344a89SAl Viro open_flag &= ~O_TRUNC; 307877d660a8SMiklos Szeredi will_truncate = false; 3079bcda7652SAl Viro acc_mode = MAY_OPEN; 3080896475d5SAl Viro path_to_nameidata(&path, nd); 3081e83db167SMiklos Szeredi goto finish_open_created; 3082fb1cc555SAl Viro } 3083fb1cc555SAl Viro 3084fb1cc555SAl Viro /* 30853134f37eSJeff Layton * create/update audit record if it already exists. 3086fb1cc555SAl Viro */ 3087896475d5SAl Viro if (d_is_positive(path.dentry)) 308876ae2a5aSAl Viro audit_inode(nd->name, path.dentry, 0); 3089fb1cc555SAl Viro 3090d18e9008SMiklos Szeredi /* 3091d18e9008SMiklos Szeredi * If atomic_open() acquired write access it is dropped now due to 3092d18e9008SMiklos Szeredi * possible mount and symlink following (this might be optimized away if 3093d18e9008SMiklos Szeredi * necessary...) 3094d18e9008SMiklos Szeredi */ 309564894cf8SAl Viro if (got_write) { 3096d18e9008SMiklos Szeredi mnt_drop_write(nd->path.mnt); 309764894cf8SAl Viro got_write = false; 3098d18e9008SMiklos Szeredi } 3099d18e9008SMiklos Szeredi 3100deb106c6SAl Viro if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) { 3101deb106c6SAl Viro path_to_nameidata(&path, nd); 3102deb106c6SAl Viro return -EEXIST; 3103deb106c6SAl Viro } 3104fb1cc555SAl Viro 3105756daf26SNeilBrown error = follow_managed(&path, nd); 3106deb106c6SAl Viro if (unlikely(error < 0)) 3107deb106c6SAl Viro return error; 3108a3fbbde7SAl Viro 3109decf3400SMiklos Szeredi BUG_ON(nd->flags & LOOKUP_RCU); 311063afdfc7SDavid Howells inode = d_backing_inode(path.dentry); 3111254cf582SAl Viro seq = 0; /* out of RCU mode, so the value doesn't matter */ 3112deb106c6SAl Viro if (unlikely(d_is_negative(path.dentry))) { 3113896475d5SAl Viro path_to_nameidata(&path, nd); 3114deb106c6SAl Viro return -ENOENT; 311554c33e7fSMiklos Szeredi } 3116766c4cbfSAl Viro finish_lookup: 3117191d7f73SAl Viro if (nd->depth) 3118191d7f73SAl Viro put_link(nd); 3119181548c0SAl Viro error = should_follow_link(nd, &path, nd->flags & LOOKUP_FOLLOW, 3120181548c0SAl Viro inode, seq); 3121deb106c6SAl Viro if (unlikely(error)) 3122d63ff28fSAl Viro return error; 3123fb1cc555SAl Viro 3124896475d5SAl Viro if (unlikely(d_is_symlink(path.dentry)) && !(open_flag & O_PATH)) { 3125896475d5SAl Viro path_to_nameidata(&path, nd); 3126deb106c6SAl Viro return -ELOOP; 3127a5cfe2d5SAl Viro } 3128a5cfe2d5SAl Viro 3129896475d5SAl Viro if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path.mnt) { 3130896475d5SAl Viro path_to_nameidata(&path, nd); 313116b1c1cdSMiklos Szeredi } else { 313216b1c1cdSMiklos Szeredi save_parent.dentry = nd->path.dentry; 3133896475d5SAl Viro save_parent.mnt = mntget(path.mnt); 3134896475d5SAl Viro nd->path.dentry = path.dentry; 313516b1c1cdSMiklos Szeredi 313616b1c1cdSMiklos Szeredi } 3137decf3400SMiklos Szeredi nd->inode = inode; 3138254cf582SAl Viro nd->seq = seq; 3139a3fbbde7SAl Viro /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ 3140bc77daa7SAl Viro finish_open: 3141a3fbbde7SAl Viro error = complete_walk(nd); 314216b1c1cdSMiklos Szeredi if (error) { 314316b1c1cdSMiklos Szeredi path_put(&save_parent); 31442675a4ebSAl Viro return error; 314516b1c1cdSMiklos Szeredi } 314676ae2a5aSAl Viro audit_inode(nd->name, nd->path.dentry, 0); 3147fb1cc555SAl Viro error = -EISDIR; 314844b1d530SMiklos Szeredi if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry)) 31492675a4ebSAl Viro goto out; 3150af2f5542SMiklos Szeredi error = -ENOTDIR; 315144b1d530SMiklos Szeredi if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry)) 31522675a4ebSAl Viro goto out; 31534bbcbd3bSDavid Howells if (!d_is_reg(nd->path.dentry)) 315477d660a8SMiklos Szeredi will_truncate = false; 31556c0d46c4SAl Viro 31560f9d1a10SAl Viro if (will_truncate) { 31570f9d1a10SAl Viro error = mnt_want_write(nd->path.mnt); 31580f9d1a10SAl Viro if (error) 31592675a4ebSAl Viro goto out; 316064894cf8SAl Viro got_write = true; 31610f9d1a10SAl Viro } 3162e83db167SMiklos Szeredi finish_open_created: 3163bcda7652SAl Viro error = may_open(&nd->path, acc_mode, open_flag); 3164ca344a89SAl Viro if (error) 31652675a4ebSAl Viro goto out; 31664aa7c634SMiklos Szeredi 31674aa7c634SMiklos Szeredi BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */ 31684aa7c634SMiklos Szeredi error = vfs_open(&nd->path, file, current_cred()); 31694aa7c634SMiklos Szeredi if (!error) { 31704aa7c634SMiklos Szeredi *opened |= FILE_OPENED; 31714aa7c634SMiklos Szeredi } else { 317230d90494SAl Viro if (error == -EOPENSTALE) 3173f60dc3dbSMiklos Szeredi goto stale_open; 3174015c3bbcSMiklos Szeredi goto out; 3175f60dc3dbSMiklos Szeredi } 3176a8277b9bSMiklos Szeredi opened: 31772675a4ebSAl Viro error = open_check_o_direct(file); 3178015c3bbcSMiklos Szeredi if (error) 3179015c3bbcSMiklos Szeredi goto exit_fput; 31803034a146SDmitry Kasatkin error = ima_file_check(file, op->acc_mode, *opened); 3181aa4caadbSMiklos Szeredi if (error) 3182aa4caadbSMiklos Szeredi goto exit_fput; 3183aa4caadbSMiklos Szeredi 31840f9d1a10SAl Viro if (will_truncate) { 31852675a4ebSAl Viro error = handle_truncate(file); 3186aa4caadbSMiklos Szeredi if (error) 3187aa4caadbSMiklos Szeredi goto exit_fput; 31880f9d1a10SAl Viro } 3189ca344a89SAl Viro out: 319064894cf8SAl Viro if (got_write) 31910f9d1a10SAl Viro mnt_drop_write(nd->path.mnt); 319216b1c1cdSMiklos Szeredi path_put(&save_parent); 31932675a4ebSAl Viro return error; 3194fb1cc555SAl Viro 3195015c3bbcSMiklos Szeredi exit_fput: 31962675a4ebSAl Viro fput(file); 31972675a4ebSAl Viro goto out; 3198015c3bbcSMiklos Szeredi 3199f60dc3dbSMiklos Szeredi stale_open: 3200f60dc3dbSMiklos Szeredi /* If no saved parent or already retried then can't retry */ 3201f60dc3dbSMiklos Szeredi if (!save_parent.dentry || retried) 3202f60dc3dbSMiklos Szeredi goto out; 3203f60dc3dbSMiklos Szeredi 3204f60dc3dbSMiklos Szeredi BUG_ON(save_parent.dentry != dir); 3205f60dc3dbSMiklos Szeredi path_put(&nd->path); 3206f60dc3dbSMiklos Szeredi nd->path = save_parent; 3207f60dc3dbSMiklos Szeredi nd->inode = dir->d_inode; 3208f60dc3dbSMiklos Szeredi save_parent.mnt = NULL; 3209f60dc3dbSMiklos Szeredi save_parent.dentry = NULL; 321064894cf8SAl Viro if (got_write) { 3211f60dc3dbSMiklos Szeredi mnt_drop_write(nd->path.mnt); 321264894cf8SAl Viro got_write = false; 3213f60dc3dbSMiklos Szeredi } 3214f60dc3dbSMiklos Szeredi retried = true; 3215f60dc3dbSMiklos Szeredi goto retry_lookup; 3216fb1cc555SAl Viro } 3217fb1cc555SAl Viro 3218c8a53ee5SAl Viro static int do_tmpfile(struct nameidata *nd, unsigned flags, 321960545d0dSAl Viro const struct open_flags *op, 322060545d0dSAl Viro struct file *file, int *opened) 322160545d0dSAl Viro { 322260545d0dSAl Viro static const struct qstr name = QSTR_INIT("/", 1); 3223625b6d10SAl Viro struct dentry *child; 322460545d0dSAl Viro struct inode *dir; 3225625b6d10SAl Viro struct path path; 3226c8a53ee5SAl Viro int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path); 322760545d0dSAl Viro if (unlikely(error)) 322860545d0dSAl Viro return error; 3229625b6d10SAl Viro error = mnt_want_write(path.mnt); 323060545d0dSAl Viro if (unlikely(error)) 323160545d0dSAl Viro goto out; 3232625b6d10SAl Viro dir = path.dentry->d_inode; 323360545d0dSAl Viro /* we want directory to be writable */ 3234625b6d10SAl Viro error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 323560545d0dSAl Viro if (error) 323660545d0dSAl Viro goto out2; 323760545d0dSAl Viro if (!dir->i_op->tmpfile) { 323860545d0dSAl Viro error = -EOPNOTSUPP; 323960545d0dSAl Viro goto out2; 324060545d0dSAl Viro } 3241625b6d10SAl Viro child = d_alloc(path.dentry, &name); 324260545d0dSAl Viro if (unlikely(!child)) { 324360545d0dSAl Viro error = -ENOMEM; 324460545d0dSAl Viro goto out2; 324560545d0dSAl Viro } 3246625b6d10SAl Viro dput(path.dentry); 3247625b6d10SAl Viro path.dentry = child; 3248625b6d10SAl Viro error = dir->i_op->tmpfile(dir, child, op->mode); 324960545d0dSAl Viro if (error) 325060545d0dSAl Viro goto out2; 3251c8a53ee5SAl Viro audit_inode(nd->name, child, 0); 325269a91c23SEric Rannaud /* Don't check for other permissions, the inode was just created */ 3253625b6d10SAl Viro error = may_open(&path, MAY_OPEN, op->open_flag); 325460545d0dSAl Viro if (error) 325560545d0dSAl Viro goto out2; 3256625b6d10SAl Viro file->f_path.mnt = path.mnt; 3257625b6d10SAl Viro error = finish_open(file, child, NULL, opened); 325860545d0dSAl Viro if (error) 325960545d0dSAl Viro goto out2; 326060545d0dSAl Viro error = open_check_o_direct(file); 3261f4e0c30cSAl Viro if (error) { 326260545d0dSAl Viro fput(file); 3263f4e0c30cSAl Viro } else if (!(op->open_flag & O_EXCL)) { 3264f4e0c30cSAl Viro struct inode *inode = file_inode(file); 3265f4e0c30cSAl Viro spin_lock(&inode->i_lock); 3266f4e0c30cSAl Viro inode->i_state |= I_LINKABLE; 3267f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 3268f4e0c30cSAl Viro } 326960545d0dSAl Viro out2: 3270625b6d10SAl Viro mnt_drop_write(path.mnt); 327160545d0dSAl Viro out: 3272625b6d10SAl Viro path_put(&path); 327360545d0dSAl Viro return error; 327460545d0dSAl Viro } 327560545d0dSAl Viro 3276c8a53ee5SAl Viro static struct file *path_openat(struct nameidata *nd, 3277c8a53ee5SAl Viro const struct open_flags *op, unsigned flags) 32781da177e4SLinus Torvalds { 3279368ee9baSAl Viro const char *s; 328030d90494SAl Viro struct file *file; 328147237687SAl Viro int opened = 0; 328213aab428SAl Viro int error; 328331e6b01fSNick Piggin 328430d90494SAl Viro file = get_empty_filp(); 32851afc99beSAl Viro if (IS_ERR(file)) 32861afc99beSAl Viro return file; 328731e6b01fSNick Piggin 328830d90494SAl Viro file->f_flags = op->open_flag; 328931e6b01fSNick Piggin 3290bb458c64SAl Viro if (unlikely(file->f_flags & __O_TMPFILE)) { 3291c8a53ee5SAl Viro error = do_tmpfile(nd, flags, op, file, &opened); 3292f15133dfSAl Viro goto out2; 329360545d0dSAl Viro } 329460545d0dSAl Viro 3295c8a53ee5SAl Viro s = path_init(nd, flags); 3296368ee9baSAl Viro if (IS_ERR(s)) { 3297368ee9baSAl Viro put_filp(file); 3298368ee9baSAl Viro return ERR_CAST(s); 3299368ee9baSAl Viro } 33003bdba28bSAl Viro while (!(error = link_path_walk(s, nd)) && 330176ae2a5aSAl Viro (error = do_last(nd, file, op, &opened)) > 0) { 330273d049a4SAl Viro nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); 33033bdba28bSAl Viro s = trailing_symlink(nd); 33043bdba28bSAl Viro if (IS_ERR(s)) { 33053bdba28bSAl Viro error = PTR_ERR(s); 33062675a4ebSAl Viro break; 3307806b681cSAl Viro } 33083bdba28bSAl Viro } 3309deb106c6SAl Viro terminate_walk(nd); 3310f15133dfSAl Viro out2: 33112675a4ebSAl Viro if (!(opened & FILE_OPENED)) { 33122675a4ebSAl Viro BUG_ON(!error); 331330d90494SAl Viro put_filp(file); 3314015c3bbcSMiklos Szeredi } 33152675a4ebSAl Viro if (unlikely(error)) { 33162675a4ebSAl Viro if (error == -EOPENSTALE) { 33172675a4ebSAl Viro if (flags & LOOKUP_RCU) 33182675a4ebSAl Viro error = -ECHILD; 33192675a4ebSAl Viro else 33202675a4ebSAl Viro error = -ESTALE; 33212675a4ebSAl Viro } 33222675a4ebSAl Viro file = ERR_PTR(error); 33232675a4ebSAl Viro } 33242675a4ebSAl Viro return file; 3325de459215SKirill Korotaev } 33261da177e4SLinus Torvalds 3327669abf4eSJeff Layton struct file *do_filp_open(int dfd, struct filename *pathname, 3328f9652e10SAl Viro const struct open_flags *op) 332913aab428SAl Viro { 3330c8a53ee5SAl Viro struct nameidata nd, *saved_nd = set_nameidata(&nd, dfd, pathname); 3331f9652e10SAl Viro int flags = op->lookup_flags; 333213aab428SAl Viro struct file *filp; 333313aab428SAl Viro 3334c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_RCU); 333513aab428SAl Viro if (unlikely(filp == ERR_PTR(-ECHILD))) 3336c8a53ee5SAl Viro filp = path_openat(&nd, op, flags); 333713aab428SAl Viro if (unlikely(filp == ERR_PTR(-ESTALE))) 3338c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_REVAL); 3339756daf26SNeilBrown restore_nameidata(saved_nd); 334013aab428SAl Viro return filp; 334113aab428SAl Viro } 334213aab428SAl Viro 334373d049a4SAl Viro struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, 3344f9652e10SAl Viro const char *name, const struct open_flags *op) 334573d049a4SAl Viro { 3346756daf26SNeilBrown struct nameidata nd, *saved_nd; 334773d049a4SAl Viro struct file *file; 334851689104SPaul Moore struct filename *filename; 3349f9652e10SAl Viro int flags = op->lookup_flags | LOOKUP_ROOT; 335073d049a4SAl Viro 335173d049a4SAl Viro nd.root.mnt = mnt; 335273d049a4SAl Viro nd.root.dentry = dentry; 335373d049a4SAl Viro 3354b18825a7SDavid Howells if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN) 335573d049a4SAl Viro return ERR_PTR(-ELOOP); 335673d049a4SAl Viro 335751689104SPaul Moore filename = getname_kernel(name); 335851689104SPaul Moore if (unlikely(IS_ERR(filename))) 335951689104SPaul Moore return ERR_CAST(filename); 336051689104SPaul Moore 3361c8a53ee5SAl Viro saved_nd = set_nameidata(&nd, -1, filename); 3362c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_RCU); 336373d049a4SAl Viro if (unlikely(file == ERR_PTR(-ECHILD))) 3364c8a53ee5SAl Viro file = path_openat(&nd, op, flags); 336573d049a4SAl Viro if (unlikely(file == ERR_PTR(-ESTALE))) 3366c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_REVAL); 3367756daf26SNeilBrown restore_nameidata(saved_nd); 336851689104SPaul Moore putname(filename); 336973d049a4SAl Viro return file; 337073d049a4SAl Viro } 337173d049a4SAl Viro 3372fa14a0b8SAl Viro static struct dentry *filename_create(int dfd, struct filename *name, 33731ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 33741da177e4SLinus Torvalds { 3375c663e5d8SChristoph Hellwig struct dentry *dentry = ERR_PTR(-EEXIST); 3376391172c4SAl Viro struct qstr last; 3377391172c4SAl Viro int type; 3378c30dabfeSJan Kara int err2; 33791ac12b4bSJeff Layton int error; 33801ac12b4bSJeff Layton bool is_dir = (lookup_flags & LOOKUP_DIRECTORY); 33811ac12b4bSJeff Layton 33821ac12b4bSJeff Layton /* 33831ac12b4bSJeff Layton * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any 33841ac12b4bSJeff Layton * other flags passed in are ignored! 33851ac12b4bSJeff Layton */ 33861ac12b4bSJeff Layton lookup_flags &= LOOKUP_REVAL; 33871ac12b4bSJeff Layton 33885c31b6ceSAl Viro name = filename_parentat(dfd, name, lookup_flags, path, &last, &type); 33895c31b6ceSAl Viro if (IS_ERR(name)) 33905c31b6ceSAl Viro return ERR_CAST(name); 33911da177e4SLinus Torvalds 3392c663e5d8SChristoph Hellwig /* 3393c663e5d8SChristoph Hellwig * Yucky last component or no last component at all? 3394c663e5d8SChristoph Hellwig * (foo/., foo/.., /////) 3395c663e5d8SChristoph Hellwig */ 33965c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) 3397ed75e95dSAl Viro goto out; 3398c663e5d8SChristoph Hellwig 3399c30dabfeSJan Kara /* don't fail immediately if it's r/o, at least try to report other errors */ 3400391172c4SAl Viro err2 = mnt_want_write(path->mnt); 3401c663e5d8SChristoph Hellwig /* 3402c663e5d8SChristoph Hellwig * Do the final lookup. 3403c663e5d8SChristoph Hellwig */ 3404391172c4SAl Viro lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL; 3405391172c4SAl Viro mutex_lock_nested(&path->dentry->d_inode->i_mutex, I_MUTEX_PARENT); 3406391172c4SAl Viro dentry = __lookup_hash(&last, path->dentry, lookup_flags); 34071da177e4SLinus Torvalds if (IS_ERR(dentry)) 3408a8104a9fSAl Viro goto unlock; 3409c663e5d8SChristoph Hellwig 3410a8104a9fSAl Viro error = -EEXIST; 3411b18825a7SDavid Howells if (d_is_positive(dentry)) 3412a8104a9fSAl Viro goto fail; 3413b18825a7SDavid Howells 3414c663e5d8SChristoph Hellwig /* 3415c663e5d8SChristoph Hellwig * Special case - lookup gave negative, but... we had foo/bar/ 3416c663e5d8SChristoph Hellwig * From the vfs_mknod() POV we just have a negative dentry - 3417c663e5d8SChristoph Hellwig * all is fine. Let's be bastards - you had / on the end, you've 3418c663e5d8SChristoph Hellwig * been asking for (non-existent) directory. -ENOENT for you. 3419c663e5d8SChristoph Hellwig */ 3420391172c4SAl Viro if (unlikely(!is_dir && last.name[last.len])) { 3421a8104a9fSAl Viro error = -ENOENT; 3422ed75e95dSAl Viro goto fail; 3423e9baf6e5SAl Viro } 3424c30dabfeSJan Kara if (unlikely(err2)) { 3425c30dabfeSJan Kara error = err2; 3426a8104a9fSAl Viro goto fail; 3427c30dabfeSJan Kara } 3428181c37b6SAl Viro putname(name); 3429e9baf6e5SAl Viro return dentry; 34301da177e4SLinus Torvalds fail: 3431a8104a9fSAl Viro dput(dentry); 3432a8104a9fSAl Viro dentry = ERR_PTR(error); 3433a8104a9fSAl Viro unlock: 3434391172c4SAl Viro mutex_unlock(&path->dentry->d_inode->i_mutex); 3435c30dabfeSJan Kara if (!err2) 3436391172c4SAl Viro mnt_drop_write(path->mnt); 3437ed75e95dSAl Viro out: 3438391172c4SAl Viro path_put(path); 3439181c37b6SAl Viro putname(name); 3440ed75e95dSAl Viro return dentry; 3441dae6ad8fSAl Viro } 3442fa14a0b8SAl Viro 3443fa14a0b8SAl Viro struct dentry *kern_path_create(int dfd, const char *pathname, 3444fa14a0b8SAl Viro struct path *path, unsigned int lookup_flags) 3445fa14a0b8SAl Viro { 3446181c37b6SAl Viro return filename_create(dfd, getname_kernel(pathname), 3447181c37b6SAl Viro path, lookup_flags); 3448fa14a0b8SAl Viro } 3449dae6ad8fSAl Viro EXPORT_SYMBOL(kern_path_create); 3450dae6ad8fSAl Viro 3451921a1650SAl Viro void done_path_create(struct path *path, struct dentry *dentry) 3452921a1650SAl Viro { 3453921a1650SAl Viro dput(dentry); 3454921a1650SAl Viro mutex_unlock(&path->dentry->d_inode->i_mutex); 3455a8104a9fSAl Viro mnt_drop_write(path->mnt); 3456921a1650SAl Viro path_put(path); 3457921a1650SAl Viro } 3458921a1650SAl Viro EXPORT_SYMBOL(done_path_create); 3459921a1650SAl Viro 3460520ae687SAl Viro inline struct dentry *user_path_create(int dfd, const char __user *pathname, 34611ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 3462dae6ad8fSAl Viro { 3463181c37b6SAl Viro return filename_create(dfd, getname(pathname), path, lookup_flags); 3464dae6ad8fSAl Viro } 3465dae6ad8fSAl Viro EXPORT_SYMBOL(user_path_create); 3466dae6ad8fSAl Viro 34671a67aafbSAl Viro int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) 34681da177e4SLinus Torvalds { 3469a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 34701da177e4SLinus Torvalds 34711da177e4SLinus Torvalds if (error) 34721da177e4SLinus Torvalds return error; 34731da177e4SLinus Torvalds 3474975d6b39SEric W. Biederman if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) 34751da177e4SLinus Torvalds return -EPERM; 34761da177e4SLinus Torvalds 3477acfa4380SAl Viro if (!dir->i_op->mknod) 34781da177e4SLinus Torvalds return -EPERM; 34791da177e4SLinus Torvalds 348008ce5f16SSerge E. Hallyn error = devcgroup_inode_mknod(mode, dev); 348108ce5f16SSerge E. Hallyn if (error) 348208ce5f16SSerge E. Hallyn return error; 348308ce5f16SSerge E. Hallyn 34841da177e4SLinus Torvalds error = security_inode_mknod(dir, dentry, mode, dev); 34851da177e4SLinus Torvalds if (error) 34861da177e4SLinus Torvalds return error; 34871da177e4SLinus Torvalds 34881da177e4SLinus Torvalds error = dir->i_op->mknod(dir, dentry, mode, dev); 3489a74574aaSStephen Smalley if (!error) 3490f38aa942SAmy Griffis fsnotify_create(dir, dentry); 34911da177e4SLinus Torvalds return error; 34921da177e4SLinus Torvalds } 34934d359507SAl Viro EXPORT_SYMBOL(vfs_mknod); 34941da177e4SLinus Torvalds 3495f69aac00SAl Viro static int may_mknod(umode_t mode) 3496463c3197SDave Hansen { 3497463c3197SDave Hansen switch (mode & S_IFMT) { 3498463c3197SDave Hansen case S_IFREG: 3499463c3197SDave Hansen case S_IFCHR: 3500463c3197SDave Hansen case S_IFBLK: 3501463c3197SDave Hansen case S_IFIFO: 3502463c3197SDave Hansen case S_IFSOCK: 3503463c3197SDave Hansen case 0: /* zero mode translates to S_IFREG */ 3504463c3197SDave Hansen return 0; 3505463c3197SDave Hansen case S_IFDIR: 3506463c3197SDave Hansen return -EPERM; 3507463c3197SDave Hansen default: 3508463c3197SDave Hansen return -EINVAL; 3509463c3197SDave Hansen } 3510463c3197SDave Hansen } 3511463c3197SDave Hansen 35128208a22bSAl Viro SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, 35132e4d0924SHeiko Carstens unsigned, dev) 35141da177e4SLinus Torvalds { 35151da177e4SLinus Torvalds struct dentry *dentry; 3516dae6ad8fSAl Viro struct path path; 3517dae6ad8fSAl Viro int error; 3518972567f1SJeff Layton unsigned int lookup_flags = 0; 35191da177e4SLinus Torvalds 35208e4bfca1SAl Viro error = may_mknod(mode); 35218e4bfca1SAl Viro if (error) 35228e4bfca1SAl Viro return error; 3523972567f1SJeff Layton retry: 3524972567f1SJeff Layton dentry = user_path_create(dfd, filename, &path, lookup_flags); 3525dae6ad8fSAl Viro if (IS_ERR(dentry)) 3526dae6ad8fSAl Viro return PTR_ERR(dentry); 35272ad94ae6SAl Viro 3528dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3529ce3b0f8dSAl Viro mode &= ~current_umask(); 3530dae6ad8fSAl Viro error = security_path_mknod(&path, dentry, mode, dev); 3531be6d3e56SKentaro Takeda if (error) 3532a8104a9fSAl Viro goto out; 35331da177e4SLinus Torvalds switch (mode & S_IFMT) { 35341da177e4SLinus Torvalds case 0: case S_IFREG: 3535312b63fbSAl Viro error = vfs_create(path.dentry->d_inode,dentry,mode,true); 35361da177e4SLinus Torvalds break; 35371da177e4SLinus Torvalds case S_IFCHR: case S_IFBLK: 3538dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode, 35391da177e4SLinus Torvalds new_decode_dev(dev)); 35401da177e4SLinus Torvalds break; 35411da177e4SLinus Torvalds case S_IFIFO: case S_IFSOCK: 3542dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); 35431da177e4SLinus Torvalds break; 35441da177e4SLinus Torvalds } 3545a8104a9fSAl Viro out: 3546921a1650SAl Viro done_path_create(&path, dentry); 3547972567f1SJeff Layton if (retry_estale(error, lookup_flags)) { 3548972567f1SJeff Layton lookup_flags |= LOOKUP_REVAL; 3549972567f1SJeff Layton goto retry; 3550972567f1SJeff Layton } 35511da177e4SLinus Torvalds return error; 35521da177e4SLinus Torvalds } 35531da177e4SLinus Torvalds 35548208a22bSAl Viro SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) 35555590ff0dSUlrich Drepper { 35565590ff0dSUlrich Drepper return sys_mknodat(AT_FDCWD, filename, mode, dev); 35575590ff0dSUlrich Drepper } 35585590ff0dSUlrich Drepper 355918bb1db3SAl Viro int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 35601da177e4SLinus Torvalds { 3561a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 35628de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 35631da177e4SLinus Torvalds 35641da177e4SLinus Torvalds if (error) 35651da177e4SLinus Torvalds return error; 35661da177e4SLinus Torvalds 3567acfa4380SAl Viro if (!dir->i_op->mkdir) 35681da177e4SLinus Torvalds return -EPERM; 35691da177e4SLinus Torvalds 35701da177e4SLinus Torvalds mode &= (S_IRWXUGO|S_ISVTX); 35711da177e4SLinus Torvalds error = security_inode_mkdir(dir, dentry, mode); 35721da177e4SLinus Torvalds if (error) 35731da177e4SLinus Torvalds return error; 35741da177e4SLinus Torvalds 35758de52778SAl Viro if (max_links && dir->i_nlink >= max_links) 35768de52778SAl Viro return -EMLINK; 35778de52778SAl Viro 35781da177e4SLinus Torvalds error = dir->i_op->mkdir(dir, dentry, mode); 3579a74574aaSStephen Smalley if (!error) 3580f38aa942SAmy Griffis fsnotify_mkdir(dir, dentry); 35811da177e4SLinus Torvalds return error; 35821da177e4SLinus Torvalds } 35834d359507SAl Viro EXPORT_SYMBOL(vfs_mkdir); 35841da177e4SLinus Torvalds 3585a218d0fdSAl Viro SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) 35861da177e4SLinus Torvalds { 35876902d925SDave Hansen struct dentry *dentry; 3588dae6ad8fSAl Viro struct path path; 3589dae6ad8fSAl Viro int error; 3590b76d8b82SJeff Layton unsigned int lookup_flags = LOOKUP_DIRECTORY; 35911da177e4SLinus Torvalds 3592b76d8b82SJeff Layton retry: 3593b76d8b82SJeff Layton dentry = user_path_create(dfd, pathname, &path, lookup_flags); 35946902d925SDave Hansen if (IS_ERR(dentry)) 3595dae6ad8fSAl Viro return PTR_ERR(dentry); 35966902d925SDave Hansen 3597dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3598ce3b0f8dSAl Viro mode &= ~current_umask(); 3599dae6ad8fSAl Viro error = security_path_mkdir(&path, dentry, mode); 3600a8104a9fSAl Viro if (!error) 3601dae6ad8fSAl Viro error = vfs_mkdir(path.dentry->d_inode, dentry, mode); 3602921a1650SAl Viro done_path_create(&path, dentry); 3603b76d8b82SJeff Layton if (retry_estale(error, lookup_flags)) { 3604b76d8b82SJeff Layton lookup_flags |= LOOKUP_REVAL; 3605b76d8b82SJeff Layton goto retry; 3606b76d8b82SJeff Layton } 36071da177e4SLinus Torvalds return error; 36081da177e4SLinus Torvalds } 36091da177e4SLinus Torvalds 3610a218d0fdSAl Viro SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) 36115590ff0dSUlrich Drepper { 36125590ff0dSUlrich Drepper return sys_mkdirat(AT_FDCWD, pathname, mode); 36135590ff0dSUlrich Drepper } 36145590ff0dSUlrich Drepper 36151da177e4SLinus Torvalds /* 3616a71905f0SSage Weil * The dentry_unhash() helper will try to drop the dentry early: we 3617c0d02594SJ. Bruce Fields * should have a usage count of 1 if we're the only user of this 3618a71905f0SSage Weil * dentry, and if that is true (possibly after pruning the dcache), 3619a71905f0SSage Weil * then we drop the dentry now. 36201da177e4SLinus Torvalds * 36211da177e4SLinus Torvalds * A low-level filesystem can, if it choses, legally 36221da177e4SLinus Torvalds * do a 36231da177e4SLinus Torvalds * 36241da177e4SLinus Torvalds * if (!d_unhashed(dentry)) 36251da177e4SLinus Torvalds * return -EBUSY; 36261da177e4SLinus Torvalds * 36271da177e4SLinus Torvalds * if it cannot handle the case of removing a directory 36281da177e4SLinus Torvalds * that is still in use by something else.. 36291da177e4SLinus Torvalds */ 36301da177e4SLinus Torvalds void dentry_unhash(struct dentry *dentry) 36311da177e4SLinus Torvalds { 36321da177e4SLinus Torvalds shrink_dcache_parent(dentry); 36331da177e4SLinus Torvalds spin_lock(&dentry->d_lock); 363498474236SWaiman Long if (dentry->d_lockref.count == 1) 36351da177e4SLinus Torvalds __d_drop(dentry); 36361da177e4SLinus Torvalds spin_unlock(&dentry->d_lock); 36371da177e4SLinus Torvalds } 36384d359507SAl Viro EXPORT_SYMBOL(dentry_unhash); 36391da177e4SLinus Torvalds 36401da177e4SLinus Torvalds int vfs_rmdir(struct inode *dir, struct dentry *dentry) 36411da177e4SLinus Torvalds { 36421da177e4SLinus Torvalds int error = may_delete(dir, dentry, 1); 36431da177e4SLinus Torvalds 36441da177e4SLinus Torvalds if (error) 36451da177e4SLinus Torvalds return error; 36461da177e4SLinus Torvalds 3647acfa4380SAl Viro if (!dir->i_op->rmdir) 36481da177e4SLinus Torvalds return -EPERM; 36491da177e4SLinus Torvalds 36501d2ef590SAl Viro dget(dentry); 36511b1dcc1bSJes Sorensen mutex_lock(&dentry->d_inode->i_mutex); 3652912dbc15SSage Weil 36531da177e4SLinus Torvalds error = -EBUSY; 36547af1364fSEric W. Biederman if (is_local_mountpoint(dentry)) 3655912dbc15SSage Weil goto out; 3656912dbc15SSage Weil 36571da177e4SLinus Torvalds error = security_inode_rmdir(dir, dentry); 3658912dbc15SSage Weil if (error) 3659912dbc15SSage Weil goto out; 3660912dbc15SSage Weil 36613cebde24SSage Weil shrink_dcache_parent(dentry); 36621da177e4SLinus Torvalds error = dir->i_op->rmdir(dir, dentry); 3663912dbc15SSage Weil if (error) 3664912dbc15SSage Weil goto out; 3665912dbc15SSage Weil 36661da177e4SLinus Torvalds dentry->d_inode->i_flags |= S_DEAD; 3667d83c49f3SAl Viro dont_mount(dentry); 36688ed936b5SEric W. Biederman detach_mounts(dentry); 36691da177e4SLinus Torvalds 3670912dbc15SSage Weil out: 3671912dbc15SSage Weil mutex_unlock(&dentry->d_inode->i_mutex); 36721d2ef590SAl Viro dput(dentry); 3673912dbc15SSage Weil if (!error) 3674912dbc15SSage Weil d_delete(dentry); 36751da177e4SLinus Torvalds return error; 36761da177e4SLinus Torvalds } 36774d359507SAl Viro EXPORT_SYMBOL(vfs_rmdir); 36781da177e4SLinus Torvalds 36795590ff0dSUlrich Drepper static long do_rmdir(int dfd, const char __user *pathname) 36801da177e4SLinus Torvalds { 36811da177e4SLinus Torvalds int error = 0; 368291a27b2aSJeff Layton struct filename *name; 36831da177e4SLinus Torvalds struct dentry *dentry; 3684f5beed75SAl Viro struct path path; 3685f5beed75SAl Viro struct qstr last; 3686f5beed75SAl Viro int type; 3687c6ee9206SJeff Layton unsigned int lookup_flags = 0; 3688c6ee9206SJeff Layton retry: 3689f5beed75SAl Viro name = user_path_parent(dfd, pathname, 3690f5beed75SAl Viro &path, &last, &type, lookup_flags); 369191a27b2aSJeff Layton if (IS_ERR(name)) 369291a27b2aSJeff Layton return PTR_ERR(name); 36931da177e4SLinus Torvalds 3694f5beed75SAl Viro switch (type) { 36951da177e4SLinus Torvalds case LAST_DOTDOT: 36961da177e4SLinus Torvalds error = -ENOTEMPTY; 36971da177e4SLinus Torvalds goto exit1; 36981da177e4SLinus Torvalds case LAST_DOT: 36991da177e4SLinus Torvalds error = -EINVAL; 37001da177e4SLinus Torvalds goto exit1; 37011da177e4SLinus Torvalds case LAST_ROOT: 37021da177e4SLinus Torvalds error = -EBUSY; 37031da177e4SLinus Torvalds goto exit1; 37041da177e4SLinus Torvalds } 37050612d9fbSOGAWA Hirofumi 3706f5beed75SAl Viro error = mnt_want_write(path.mnt); 3707c30dabfeSJan Kara if (error) 3708c30dabfeSJan Kara goto exit1; 37090612d9fbSOGAWA Hirofumi 3710f5beed75SAl Viro mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 3711f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 37121da177e4SLinus Torvalds error = PTR_ERR(dentry); 37136902d925SDave Hansen if (IS_ERR(dentry)) 37146902d925SDave Hansen goto exit2; 3715e6bc45d6STheodore Ts'o if (!dentry->d_inode) { 3716e6bc45d6STheodore Ts'o error = -ENOENT; 3717e6bc45d6STheodore Ts'o goto exit3; 3718e6bc45d6STheodore Ts'o } 3719f5beed75SAl Viro error = security_path_rmdir(&path, dentry); 3720be6d3e56SKentaro Takeda if (error) 3721c30dabfeSJan Kara goto exit3; 3722f5beed75SAl Viro error = vfs_rmdir(path.dentry->d_inode, dentry); 37230622753bSDave Hansen exit3: 37241da177e4SLinus Torvalds dput(dentry); 37256902d925SDave Hansen exit2: 3726f5beed75SAl Viro mutex_unlock(&path.dentry->d_inode->i_mutex); 3727f5beed75SAl Viro mnt_drop_write(path.mnt); 37281da177e4SLinus Torvalds exit1: 3729f5beed75SAl Viro path_put(&path); 37301da177e4SLinus Torvalds putname(name); 3731c6ee9206SJeff Layton if (retry_estale(error, lookup_flags)) { 3732c6ee9206SJeff Layton lookup_flags |= LOOKUP_REVAL; 3733c6ee9206SJeff Layton goto retry; 3734c6ee9206SJeff Layton } 37351da177e4SLinus Torvalds return error; 37361da177e4SLinus Torvalds } 37371da177e4SLinus Torvalds 37383cdad428SHeiko Carstens SYSCALL_DEFINE1(rmdir, const char __user *, pathname) 37395590ff0dSUlrich Drepper { 37405590ff0dSUlrich Drepper return do_rmdir(AT_FDCWD, pathname); 37415590ff0dSUlrich Drepper } 37425590ff0dSUlrich Drepper 3743b21996e3SJ. Bruce Fields /** 3744b21996e3SJ. Bruce Fields * vfs_unlink - unlink a filesystem object 3745b21996e3SJ. Bruce Fields * @dir: parent directory 3746b21996e3SJ. Bruce Fields * @dentry: victim 3747b21996e3SJ. Bruce Fields * @delegated_inode: returns victim inode, if the inode is delegated. 3748b21996e3SJ. Bruce Fields * 3749b21996e3SJ. Bruce Fields * The caller must hold dir->i_mutex. 3750b21996e3SJ. Bruce Fields * 3751b21996e3SJ. Bruce Fields * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and 3752b21996e3SJ. Bruce Fields * return a reference to the inode in delegated_inode. The caller 3753b21996e3SJ. Bruce Fields * should then break the delegation on that inode and retry. Because 3754b21996e3SJ. Bruce Fields * breaking a delegation may take a long time, the caller should drop 3755b21996e3SJ. Bruce Fields * dir->i_mutex before doing so. 3756b21996e3SJ. Bruce Fields * 3757b21996e3SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 3758b21996e3SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 3759b21996e3SJ. Bruce Fields * to be NFS exported. 3760b21996e3SJ. Bruce Fields */ 3761b21996e3SJ. Bruce Fields int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode) 37621da177e4SLinus Torvalds { 37639accbb97SJ. Bruce Fields struct inode *target = dentry->d_inode; 37641da177e4SLinus Torvalds int error = may_delete(dir, dentry, 0); 37651da177e4SLinus Torvalds 37661da177e4SLinus Torvalds if (error) 37671da177e4SLinus Torvalds return error; 37681da177e4SLinus Torvalds 3769acfa4380SAl Viro if (!dir->i_op->unlink) 37701da177e4SLinus Torvalds return -EPERM; 37711da177e4SLinus Torvalds 37729accbb97SJ. Bruce Fields mutex_lock(&target->i_mutex); 37738ed936b5SEric W. Biederman if (is_local_mountpoint(dentry)) 37741da177e4SLinus Torvalds error = -EBUSY; 37751da177e4SLinus Torvalds else { 37761da177e4SLinus Torvalds error = security_inode_unlink(dir, dentry); 3777bec1052eSAl Viro if (!error) { 37785a14696cSJ. Bruce Fields error = try_break_deleg(target, delegated_inode); 37795a14696cSJ. Bruce Fields if (error) 3780b21996e3SJ. Bruce Fields goto out; 37811da177e4SLinus Torvalds error = dir->i_op->unlink(dir, dentry); 37828ed936b5SEric W. Biederman if (!error) { 3783d83c49f3SAl Viro dont_mount(dentry); 37848ed936b5SEric W. Biederman detach_mounts(dentry); 37858ed936b5SEric W. Biederman } 3786bec1052eSAl Viro } 37871da177e4SLinus Torvalds } 3788b21996e3SJ. Bruce Fields out: 37899accbb97SJ. Bruce Fields mutex_unlock(&target->i_mutex); 37901da177e4SLinus Torvalds 37911da177e4SLinus Torvalds /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 37921da177e4SLinus Torvalds if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { 37939accbb97SJ. Bruce Fields fsnotify_link_count(target); 37941da177e4SLinus Torvalds d_delete(dentry); 37951da177e4SLinus Torvalds } 37960eeca283SRobert Love 37971da177e4SLinus Torvalds return error; 37981da177e4SLinus Torvalds } 37994d359507SAl Viro EXPORT_SYMBOL(vfs_unlink); 38001da177e4SLinus Torvalds 38011da177e4SLinus Torvalds /* 38021da177e4SLinus Torvalds * Make sure that the actual truncation of the file will occur outside its 38031b1dcc1bSJes Sorensen * directory's i_mutex. Truncate can take a long time if there is a lot of 38041da177e4SLinus Torvalds * writeout happening, and we don't want to prevent access to the directory 38051da177e4SLinus Torvalds * while waiting on the I/O. 38061da177e4SLinus Torvalds */ 38075590ff0dSUlrich Drepper static long do_unlinkat(int dfd, const char __user *pathname) 38081da177e4SLinus Torvalds { 38092ad94ae6SAl Viro int error; 381091a27b2aSJeff Layton struct filename *name; 38111da177e4SLinus Torvalds struct dentry *dentry; 3812f5beed75SAl Viro struct path path; 3813f5beed75SAl Viro struct qstr last; 3814f5beed75SAl Viro int type; 38151da177e4SLinus Torvalds struct inode *inode = NULL; 3816b21996e3SJ. Bruce Fields struct inode *delegated_inode = NULL; 38175d18f813SJeff Layton unsigned int lookup_flags = 0; 38185d18f813SJeff Layton retry: 3819f5beed75SAl Viro name = user_path_parent(dfd, pathname, 3820f5beed75SAl Viro &path, &last, &type, lookup_flags); 382191a27b2aSJeff Layton if (IS_ERR(name)) 382291a27b2aSJeff Layton return PTR_ERR(name); 38232ad94ae6SAl Viro 38241da177e4SLinus Torvalds error = -EISDIR; 3825f5beed75SAl Viro if (type != LAST_NORM) 38261da177e4SLinus Torvalds goto exit1; 38270612d9fbSOGAWA Hirofumi 3828f5beed75SAl Viro error = mnt_want_write(path.mnt); 3829c30dabfeSJan Kara if (error) 3830c30dabfeSJan Kara goto exit1; 3831b21996e3SJ. Bruce Fields retry_deleg: 3832f5beed75SAl Viro mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 3833f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 38341da177e4SLinus Torvalds error = PTR_ERR(dentry); 38351da177e4SLinus Torvalds if (!IS_ERR(dentry)) { 38361da177e4SLinus Torvalds /* Why not before? Because we want correct error value */ 3837f5beed75SAl Viro if (last.name[last.len]) 383850338b88STörök Edwin goto slashes; 38391da177e4SLinus Torvalds inode = dentry->d_inode; 3840b18825a7SDavid Howells if (d_is_negative(dentry)) 3841e6bc45d6STheodore Ts'o goto slashes; 38427de9c6eeSAl Viro ihold(inode); 3843f5beed75SAl Viro error = security_path_unlink(&path, dentry); 3844be6d3e56SKentaro Takeda if (error) 3845c30dabfeSJan Kara goto exit2; 3846f5beed75SAl Viro error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode); 38471da177e4SLinus Torvalds exit2: 38481da177e4SLinus Torvalds dput(dentry); 38491da177e4SLinus Torvalds } 3850f5beed75SAl Viro mutex_unlock(&path.dentry->d_inode->i_mutex); 38511da177e4SLinus Torvalds if (inode) 38521da177e4SLinus Torvalds iput(inode); /* truncate the inode here */ 3853b21996e3SJ. Bruce Fields inode = NULL; 3854b21996e3SJ. Bruce Fields if (delegated_inode) { 38555a14696cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 3856b21996e3SJ. Bruce Fields if (!error) 3857b21996e3SJ. Bruce Fields goto retry_deleg; 3858b21996e3SJ. Bruce Fields } 3859f5beed75SAl Viro mnt_drop_write(path.mnt); 38601da177e4SLinus Torvalds exit1: 3861f5beed75SAl Viro path_put(&path); 38621da177e4SLinus Torvalds putname(name); 38635d18f813SJeff Layton if (retry_estale(error, lookup_flags)) { 38645d18f813SJeff Layton lookup_flags |= LOOKUP_REVAL; 38655d18f813SJeff Layton inode = NULL; 38665d18f813SJeff Layton goto retry; 38675d18f813SJeff Layton } 38681da177e4SLinus Torvalds return error; 38691da177e4SLinus Torvalds 38701da177e4SLinus Torvalds slashes: 3871b18825a7SDavid Howells if (d_is_negative(dentry)) 3872b18825a7SDavid Howells error = -ENOENT; 387344b1d530SMiklos Szeredi else if (d_is_dir(dentry)) 3874b18825a7SDavid Howells error = -EISDIR; 3875b18825a7SDavid Howells else 3876b18825a7SDavid Howells error = -ENOTDIR; 38771da177e4SLinus Torvalds goto exit2; 38781da177e4SLinus Torvalds } 38791da177e4SLinus Torvalds 38802e4d0924SHeiko Carstens SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) 38815590ff0dSUlrich Drepper { 38825590ff0dSUlrich Drepper if ((flag & ~AT_REMOVEDIR) != 0) 38835590ff0dSUlrich Drepper return -EINVAL; 38845590ff0dSUlrich Drepper 38855590ff0dSUlrich Drepper if (flag & AT_REMOVEDIR) 38865590ff0dSUlrich Drepper return do_rmdir(dfd, pathname); 38875590ff0dSUlrich Drepper 38885590ff0dSUlrich Drepper return do_unlinkat(dfd, pathname); 38895590ff0dSUlrich Drepper } 38905590ff0dSUlrich Drepper 38913480b257SHeiko Carstens SYSCALL_DEFINE1(unlink, const char __user *, pathname) 38925590ff0dSUlrich Drepper { 38935590ff0dSUlrich Drepper return do_unlinkat(AT_FDCWD, pathname); 38945590ff0dSUlrich Drepper } 38955590ff0dSUlrich Drepper 3896db2e747bSMiklos Szeredi int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) 38971da177e4SLinus Torvalds { 3898a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 38991da177e4SLinus Torvalds 39001da177e4SLinus Torvalds if (error) 39011da177e4SLinus Torvalds return error; 39021da177e4SLinus Torvalds 3903acfa4380SAl Viro if (!dir->i_op->symlink) 39041da177e4SLinus Torvalds return -EPERM; 39051da177e4SLinus Torvalds 39061da177e4SLinus Torvalds error = security_inode_symlink(dir, dentry, oldname); 39071da177e4SLinus Torvalds if (error) 39081da177e4SLinus Torvalds return error; 39091da177e4SLinus Torvalds 39101da177e4SLinus Torvalds error = dir->i_op->symlink(dir, dentry, oldname); 3911a74574aaSStephen Smalley if (!error) 3912f38aa942SAmy Griffis fsnotify_create(dir, dentry); 39131da177e4SLinus Torvalds return error; 39141da177e4SLinus Torvalds } 39154d359507SAl Viro EXPORT_SYMBOL(vfs_symlink); 39161da177e4SLinus Torvalds 39172e4d0924SHeiko Carstens SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, 39182e4d0924SHeiko Carstens int, newdfd, const char __user *, newname) 39191da177e4SLinus Torvalds { 39202ad94ae6SAl Viro int error; 392191a27b2aSJeff Layton struct filename *from; 39226902d925SDave Hansen struct dentry *dentry; 3923dae6ad8fSAl Viro struct path path; 3924f46d3567SJeff Layton unsigned int lookup_flags = 0; 39251da177e4SLinus Torvalds 39261da177e4SLinus Torvalds from = getname(oldname); 39271da177e4SLinus Torvalds if (IS_ERR(from)) 39281da177e4SLinus Torvalds return PTR_ERR(from); 3929f46d3567SJeff Layton retry: 3930f46d3567SJeff Layton dentry = user_path_create(newdfd, newname, &path, lookup_flags); 39311da177e4SLinus Torvalds error = PTR_ERR(dentry); 39326902d925SDave Hansen if (IS_ERR(dentry)) 3933dae6ad8fSAl Viro goto out_putname; 39346902d925SDave Hansen 393591a27b2aSJeff Layton error = security_path_symlink(&path, dentry, from->name); 3936a8104a9fSAl Viro if (!error) 393791a27b2aSJeff Layton error = vfs_symlink(path.dentry->d_inode, dentry, from->name); 3938921a1650SAl Viro done_path_create(&path, dentry); 3939f46d3567SJeff Layton if (retry_estale(error, lookup_flags)) { 3940f46d3567SJeff Layton lookup_flags |= LOOKUP_REVAL; 3941f46d3567SJeff Layton goto retry; 3942f46d3567SJeff Layton } 39436902d925SDave Hansen out_putname: 39441da177e4SLinus Torvalds putname(from); 39451da177e4SLinus Torvalds return error; 39461da177e4SLinus Torvalds } 39471da177e4SLinus Torvalds 39483480b257SHeiko Carstens SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) 39495590ff0dSUlrich Drepper { 39505590ff0dSUlrich Drepper return sys_symlinkat(oldname, AT_FDCWD, newname); 39515590ff0dSUlrich Drepper } 39525590ff0dSUlrich Drepper 3953146a8595SJ. Bruce Fields /** 3954146a8595SJ. Bruce Fields * vfs_link - create a new link 3955146a8595SJ. Bruce Fields * @old_dentry: object to be linked 3956146a8595SJ. Bruce Fields * @dir: new parent 3957146a8595SJ. Bruce Fields * @new_dentry: where to create the new link 3958146a8595SJ. Bruce Fields * @delegated_inode: returns inode needing a delegation break 3959146a8595SJ. Bruce Fields * 3960146a8595SJ. Bruce Fields * The caller must hold dir->i_mutex 3961146a8595SJ. Bruce Fields * 3962146a8595SJ. Bruce Fields * If vfs_link discovers a delegation on the to-be-linked file in need 3963146a8595SJ. Bruce Fields * of breaking, it will return -EWOULDBLOCK and return a reference to the 3964146a8595SJ. Bruce Fields * inode in delegated_inode. The caller should then break the delegation 3965146a8595SJ. Bruce Fields * and retry. Because breaking a delegation may take a long time, the 3966146a8595SJ. Bruce Fields * caller should drop the i_mutex before doing so. 3967146a8595SJ. Bruce Fields * 3968146a8595SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 3969146a8595SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 3970146a8595SJ. Bruce Fields * to be NFS exported. 3971146a8595SJ. Bruce Fields */ 3972146a8595SJ. Bruce Fields int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode) 39731da177e4SLinus Torvalds { 39741da177e4SLinus Torvalds struct inode *inode = old_dentry->d_inode; 39758de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 39761da177e4SLinus Torvalds int error; 39771da177e4SLinus Torvalds 39781da177e4SLinus Torvalds if (!inode) 39791da177e4SLinus Torvalds return -ENOENT; 39801da177e4SLinus Torvalds 3981a95164d9SMiklos Szeredi error = may_create(dir, new_dentry); 39821da177e4SLinus Torvalds if (error) 39831da177e4SLinus Torvalds return error; 39841da177e4SLinus Torvalds 39851da177e4SLinus Torvalds if (dir->i_sb != inode->i_sb) 39861da177e4SLinus Torvalds return -EXDEV; 39871da177e4SLinus Torvalds 39881da177e4SLinus Torvalds /* 39891da177e4SLinus Torvalds * A link to an append-only or immutable file cannot be created. 39901da177e4SLinus Torvalds */ 39911da177e4SLinus Torvalds if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 39921da177e4SLinus Torvalds return -EPERM; 3993acfa4380SAl Viro if (!dir->i_op->link) 39941da177e4SLinus Torvalds return -EPERM; 39957e79eedbSTetsuo Handa if (S_ISDIR(inode->i_mode)) 39961da177e4SLinus Torvalds return -EPERM; 39971da177e4SLinus Torvalds 39981da177e4SLinus Torvalds error = security_inode_link(old_dentry, dir, new_dentry); 39991da177e4SLinus Torvalds if (error) 40001da177e4SLinus Torvalds return error; 40011da177e4SLinus Torvalds 40027e79eedbSTetsuo Handa mutex_lock(&inode->i_mutex); 4003aae8a97dSAneesh Kumar K.V /* Make sure we don't allow creating hardlink to an unlinked file */ 4004f4e0c30cSAl Viro if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) 4005aae8a97dSAneesh Kumar K.V error = -ENOENT; 40068de52778SAl Viro else if (max_links && inode->i_nlink >= max_links) 40078de52778SAl Viro error = -EMLINK; 4008146a8595SJ. Bruce Fields else { 4009146a8595SJ. Bruce Fields error = try_break_deleg(inode, delegated_inode); 4010146a8595SJ. Bruce Fields if (!error) 40111da177e4SLinus Torvalds error = dir->i_op->link(old_dentry, dir, new_dentry); 4012146a8595SJ. Bruce Fields } 4013f4e0c30cSAl Viro 4014f4e0c30cSAl Viro if (!error && (inode->i_state & I_LINKABLE)) { 4015f4e0c30cSAl Viro spin_lock(&inode->i_lock); 4016f4e0c30cSAl Viro inode->i_state &= ~I_LINKABLE; 4017f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 4018f4e0c30cSAl Viro } 40197e79eedbSTetsuo Handa mutex_unlock(&inode->i_mutex); 4020e31e14ecSStephen Smalley if (!error) 40217e79eedbSTetsuo Handa fsnotify_link(dir, inode, new_dentry); 40221da177e4SLinus Torvalds return error; 40231da177e4SLinus Torvalds } 40244d359507SAl Viro EXPORT_SYMBOL(vfs_link); 40251da177e4SLinus Torvalds 40261da177e4SLinus Torvalds /* 40271da177e4SLinus Torvalds * Hardlinks are often used in delicate situations. We avoid 40281da177e4SLinus Torvalds * security-related surprises by not following symlinks on the 40291da177e4SLinus Torvalds * newname. --KAB 40301da177e4SLinus Torvalds * 40311da177e4SLinus Torvalds * We don't follow them on the oldname either to be compatible 40321da177e4SLinus Torvalds * with linux 2.0, and to avoid hard-linking to directories 40331da177e4SLinus Torvalds * and other special files. --ADM 40341da177e4SLinus Torvalds */ 40352e4d0924SHeiko Carstens SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, 40362e4d0924SHeiko Carstens int, newdfd, const char __user *, newname, int, flags) 40371da177e4SLinus Torvalds { 40381da177e4SLinus Torvalds struct dentry *new_dentry; 4039dae6ad8fSAl Viro struct path old_path, new_path; 4040146a8595SJ. Bruce Fields struct inode *delegated_inode = NULL; 404111a7b371SAneesh Kumar K.V int how = 0; 40421da177e4SLinus Torvalds int error; 40431da177e4SLinus Torvalds 404411a7b371SAneesh Kumar K.V if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) 4045c04030e1SUlrich Drepper return -EINVAL; 404611a7b371SAneesh Kumar K.V /* 4047f0cc6ffbSLinus Torvalds * To use null names we require CAP_DAC_READ_SEARCH 4048f0cc6ffbSLinus Torvalds * This ensures that not everyone will be able to create 4049f0cc6ffbSLinus Torvalds * handlink using the passed filedescriptor. 405011a7b371SAneesh Kumar K.V */ 4051f0cc6ffbSLinus Torvalds if (flags & AT_EMPTY_PATH) { 4052f0cc6ffbSLinus Torvalds if (!capable(CAP_DAC_READ_SEARCH)) 4053f0cc6ffbSLinus Torvalds return -ENOENT; 405411a7b371SAneesh Kumar K.V how = LOOKUP_EMPTY; 4055f0cc6ffbSLinus Torvalds } 4056c04030e1SUlrich Drepper 405711a7b371SAneesh Kumar K.V if (flags & AT_SYMLINK_FOLLOW) 405811a7b371SAneesh Kumar K.V how |= LOOKUP_FOLLOW; 4059442e31caSJeff Layton retry: 406011a7b371SAneesh Kumar K.V error = user_path_at(olddfd, oldname, how, &old_path); 40611da177e4SLinus Torvalds if (error) 40622ad94ae6SAl Viro return error; 40632ad94ae6SAl Viro 4064442e31caSJeff Layton new_dentry = user_path_create(newdfd, newname, &new_path, 4065442e31caSJeff Layton (how & LOOKUP_REVAL)); 40661da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 40676902d925SDave Hansen if (IS_ERR(new_dentry)) 4068dae6ad8fSAl Viro goto out; 4069dae6ad8fSAl Viro 4070dae6ad8fSAl Viro error = -EXDEV; 4071dae6ad8fSAl Viro if (old_path.mnt != new_path.mnt) 4072dae6ad8fSAl Viro goto out_dput; 4073800179c9SKees Cook error = may_linkat(&old_path); 4074800179c9SKees Cook if (unlikely(error)) 4075800179c9SKees Cook goto out_dput; 4076dae6ad8fSAl Viro error = security_path_link(old_path.dentry, &new_path, new_dentry); 4077be6d3e56SKentaro Takeda if (error) 4078a8104a9fSAl Viro goto out_dput; 4079146a8595SJ. Bruce Fields error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode); 408075c3f29dSDave Hansen out_dput: 4081921a1650SAl Viro done_path_create(&new_path, new_dentry); 4082146a8595SJ. Bruce Fields if (delegated_inode) { 4083146a8595SJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4084d22e6338SOleg Drokin if (!error) { 4085d22e6338SOleg Drokin path_put(&old_path); 4086146a8595SJ. Bruce Fields goto retry; 4087146a8595SJ. Bruce Fields } 4088d22e6338SOleg Drokin } 4089442e31caSJeff Layton if (retry_estale(error, how)) { 4090d22e6338SOleg Drokin path_put(&old_path); 4091442e31caSJeff Layton how |= LOOKUP_REVAL; 4092442e31caSJeff Layton goto retry; 4093442e31caSJeff Layton } 40941da177e4SLinus Torvalds out: 40952d8f3038SAl Viro path_put(&old_path); 40961da177e4SLinus Torvalds 40971da177e4SLinus Torvalds return error; 40981da177e4SLinus Torvalds } 40991da177e4SLinus Torvalds 41003480b257SHeiko Carstens SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) 41015590ff0dSUlrich Drepper { 4102c04030e1SUlrich Drepper return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 41035590ff0dSUlrich Drepper } 41045590ff0dSUlrich Drepper 4105bc27027aSMiklos Szeredi /** 4106bc27027aSMiklos Szeredi * vfs_rename - rename a filesystem object 4107bc27027aSMiklos Szeredi * @old_dir: parent of source 4108bc27027aSMiklos Szeredi * @old_dentry: source 4109bc27027aSMiklos Szeredi * @new_dir: parent of destination 4110bc27027aSMiklos Szeredi * @new_dentry: destination 4111bc27027aSMiklos Szeredi * @delegated_inode: returns an inode needing a delegation break 4112520c8b16SMiklos Szeredi * @flags: rename flags 4113bc27027aSMiklos Szeredi * 4114bc27027aSMiklos Szeredi * The caller must hold multiple mutexes--see lock_rename()). 4115bc27027aSMiklos Szeredi * 4116bc27027aSMiklos Szeredi * If vfs_rename discovers a delegation in need of breaking at either 4117bc27027aSMiklos Szeredi * the source or destination, it will return -EWOULDBLOCK and return a 4118bc27027aSMiklos Szeredi * reference to the inode in delegated_inode. The caller should then 4119bc27027aSMiklos Szeredi * break the delegation and retry. Because breaking a delegation may 4120bc27027aSMiklos Szeredi * take a long time, the caller should drop all locks before doing 4121bc27027aSMiklos Szeredi * so. 4122bc27027aSMiklos Szeredi * 4123bc27027aSMiklos Szeredi * Alternatively, a caller may pass NULL for delegated_inode. This may 4124bc27027aSMiklos Szeredi * be appropriate for callers that expect the underlying filesystem not 4125bc27027aSMiklos Szeredi * to be NFS exported. 4126bc27027aSMiklos Szeredi * 41271da177e4SLinus Torvalds * The worst of all namespace operations - renaming directory. "Perverted" 41281da177e4SLinus Torvalds * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 41291da177e4SLinus Torvalds * Problems: 4130d03b29a2SJ. Bruce Fields * a) we can get into loop creation. 41311da177e4SLinus Torvalds * b) race potential - two innocent renames can create a loop together. 41321da177e4SLinus Torvalds * That's where 4.4 screws up. Current fix: serialization on 4133a11f3a05SArjan van de Ven * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 41341da177e4SLinus Torvalds * story. 41356cedba89SJ. Bruce Fields * c) we have to lock _four_ objects - parents and victim (if it exists), 41366cedba89SJ. Bruce Fields * and source (if it is not a directory). 41371b1dcc1bSJes Sorensen * And that - after we got ->i_mutex on parents (until then we don't know 41381da177e4SLinus Torvalds * whether the target exists). Solution: try to be smart with locking 41391da177e4SLinus Torvalds * order for inodes. We rely on the fact that tree topology may change 4140a11f3a05SArjan van de Ven * only under ->s_vfs_rename_mutex _and_ that parent of the object we 41411da177e4SLinus Torvalds * move will be locked. Thus we can rank directories by the tree 41421da177e4SLinus Torvalds * (ancestors first) and rank all non-directories after them. 41431da177e4SLinus Torvalds * That works since everybody except rename does "lock parent, lookup, 4144a11f3a05SArjan van de Ven * lock child" and rename is under ->s_vfs_rename_mutex. 41451da177e4SLinus Torvalds * HOWEVER, it relies on the assumption that any object with ->lookup() 41461da177e4SLinus Torvalds * has no more than 1 dentry. If "hybrid" objects will ever appear, 41471da177e4SLinus Torvalds * we'd better make sure that there's no link(2) for them. 4148e4eaac06SSage Weil * d) conversion from fhandle to dentry may come in the wrong moment - when 41491b1dcc1bSJes Sorensen * we are removing the target. Solution: we will have to grab ->i_mutex 41501da177e4SLinus Torvalds * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 4151c41b20e7SAdam Buchbinder * ->i_mutex on parents, which works but leads to some truly excessive 41521da177e4SLinus Torvalds * locking]. 41531da177e4SLinus Torvalds */ 41541da177e4SLinus Torvalds int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, 41558e6d782cSJ. Bruce Fields struct inode *new_dir, struct dentry *new_dentry, 4156520c8b16SMiklos Szeredi struct inode **delegated_inode, unsigned int flags) 41571da177e4SLinus Torvalds { 41581da177e4SLinus Torvalds int error; 4159bc27027aSMiklos Szeredi bool is_dir = d_is_dir(old_dentry); 416059b0df21SEric Paris const unsigned char *old_name; 4161bc27027aSMiklos Szeredi struct inode *source = old_dentry->d_inode; 4162bc27027aSMiklos Szeredi struct inode *target = new_dentry->d_inode; 4163da1ce067SMiklos Szeredi bool new_is_dir = false; 4164da1ce067SMiklos Szeredi unsigned max_links = new_dir->i_sb->s_max_links; 41651da177e4SLinus Torvalds 4166bc27027aSMiklos Szeredi if (source == target) 41671da177e4SLinus Torvalds return 0; 41681da177e4SLinus Torvalds 41691da177e4SLinus Torvalds error = may_delete(old_dir, old_dentry, is_dir); 41701da177e4SLinus Torvalds if (error) 41711da177e4SLinus Torvalds return error; 41721da177e4SLinus Torvalds 4173da1ce067SMiklos Szeredi if (!target) { 4174a95164d9SMiklos Szeredi error = may_create(new_dir, new_dentry); 4175da1ce067SMiklos Szeredi } else { 4176da1ce067SMiklos Szeredi new_is_dir = d_is_dir(new_dentry); 4177da1ce067SMiklos Szeredi 4178da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 41791da177e4SLinus Torvalds error = may_delete(new_dir, new_dentry, is_dir); 4180da1ce067SMiklos Szeredi else 4181da1ce067SMiklos Szeredi error = may_delete(new_dir, new_dentry, new_is_dir); 4182da1ce067SMiklos Szeredi } 41831da177e4SLinus Torvalds if (error) 41841da177e4SLinus Torvalds return error; 41851da177e4SLinus Torvalds 41867177a9c4SMiklos Szeredi if (!old_dir->i_op->rename && !old_dir->i_op->rename2) 41871da177e4SLinus Torvalds return -EPERM; 41881da177e4SLinus Torvalds 4189520c8b16SMiklos Szeredi if (flags && !old_dir->i_op->rename2) 4190520c8b16SMiklos Szeredi return -EINVAL; 4191520c8b16SMiklos Szeredi 4192bc27027aSMiklos Szeredi /* 4193bc27027aSMiklos Szeredi * If we are going to change the parent - check write permissions, 4194bc27027aSMiklos Szeredi * we'll need to flip '..'. 4195bc27027aSMiklos Szeredi */ 4196da1ce067SMiklos Szeredi if (new_dir != old_dir) { 4197da1ce067SMiklos Szeredi if (is_dir) { 4198bc27027aSMiklos Szeredi error = inode_permission(source, MAY_WRITE); 4199bc27027aSMiklos Szeredi if (error) 4200bc27027aSMiklos Szeredi return error; 4201bc27027aSMiklos Szeredi } 4202da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && new_is_dir) { 4203da1ce067SMiklos Szeredi error = inode_permission(target, MAY_WRITE); 4204da1ce067SMiklos Szeredi if (error) 4205da1ce067SMiklos Szeredi return error; 4206da1ce067SMiklos Szeredi } 4207da1ce067SMiklos Szeredi } 42080eeca283SRobert Love 42090b3974ebSMiklos Szeredi error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry, 42100b3974ebSMiklos Szeredi flags); 4211bc27027aSMiklos Szeredi if (error) 4212bc27027aSMiklos Szeredi return error; 4213bc27027aSMiklos Szeredi 4214bc27027aSMiklos Szeredi old_name = fsnotify_oldname_init(old_dentry->d_name.name); 4215bc27027aSMiklos Szeredi dget(new_dentry); 4216da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4217bc27027aSMiklos Szeredi lock_two_nondirectories(source, target); 4218bc27027aSMiklos Szeredi else if (target) 4219bc27027aSMiklos Szeredi mutex_lock(&target->i_mutex); 4220bc27027aSMiklos Szeredi 4221bc27027aSMiklos Szeredi error = -EBUSY; 42227af1364fSEric W. Biederman if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry)) 4223bc27027aSMiklos Szeredi goto out; 4224bc27027aSMiklos Szeredi 4225da1ce067SMiklos Szeredi if (max_links && new_dir != old_dir) { 4226bc27027aSMiklos Szeredi error = -EMLINK; 4227da1ce067SMiklos Szeredi if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links) 4228bc27027aSMiklos Szeredi goto out; 4229da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir && 4230da1ce067SMiklos Szeredi old_dir->i_nlink >= max_links) 4231da1ce067SMiklos Szeredi goto out; 4232da1ce067SMiklos Szeredi } 4233da1ce067SMiklos Szeredi if (is_dir && !(flags & RENAME_EXCHANGE) && target) 4234bc27027aSMiklos Szeredi shrink_dcache_parent(new_dentry); 4235da1ce067SMiklos Szeredi if (!is_dir) { 4236bc27027aSMiklos Szeredi error = try_break_deleg(source, delegated_inode); 4237bc27027aSMiklos Szeredi if (error) 4238bc27027aSMiklos Szeredi goto out; 4239da1ce067SMiklos Szeredi } 4240da1ce067SMiklos Szeredi if (target && !new_is_dir) { 4241bc27027aSMiklos Szeredi error = try_break_deleg(target, delegated_inode); 4242bc27027aSMiklos Szeredi if (error) 4243bc27027aSMiklos Szeredi goto out; 4244bc27027aSMiklos Szeredi } 42457177a9c4SMiklos Szeredi if (!old_dir->i_op->rename2) { 4246520c8b16SMiklos Szeredi error = old_dir->i_op->rename(old_dir, old_dentry, 4247520c8b16SMiklos Szeredi new_dir, new_dentry); 4248520c8b16SMiklos Szeredi } else { 42497177a9c4SMiklos Szeredi WARN_ON(old_dir->i_op->rename != NULL); 4250520c8b16SMiklos Szeredi error = old_dir->i_op->rename2(old_dir, old_dentry, 4251520c8b16SMiklos Szeredi new_dir, new_dentry, flags); 4252520c8b16SMiklos Szeredi } 4253bc27027aSMiklos Szeredi if (error) 4254bc27027aSMiklos Szeredi goto out; 4255bc27027aSMiklos Szeredi 4256da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE) && target) { 42571da177e4SLinus Torvalds if (is_dir) 4258bc27027aSMiklos Szeredi target->i_flags |= S_DEAD; 4259bc27027aSMiklos Szeredi dont_mount(new_dentry); 42608ed936b5SEric W. Biederman detach_mounts(new_dentry); 4261bc27027aSMiklos Szeredi } 4262da1ce067SMiklos Szeredi if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) { 4263da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 4264bc27027aSMiklos Szeredi d_move(old_dentry, new_dentry); 4265da1ce067SMiklos Szeredi else 4266da1ce067SMiklos Szeredi d_exchange(old_dentry, new_dentry); 4267da1ce067SMiklos Szeredi } 4268bc27027aSMiklos Szeredi out: 4269da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4270bc27027aSMiklos Szeredi unlock_two_nondirectories(source, target); 4271bc27027aSMiklos Szeredi else if (target) 4272bc27027aSMiklos Szeredi mutex_unlock(&target->i_mutex); 4273bc27027aSMiklos Szeredi dput(new_dentry); 4274da1ce067SMiklos Szeredi if (!error) { 4275123df294SAl Viro fsnotify_move(old_dir, new_dir, old_name, is_dir, 4276da1ce067SMiklos Szeredi !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry); 4277da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4278da1ce067SMiklos Szeredi fsnotify_move(new_dir, old_dir, old_dentry->d_name.name, 4279da1ce067SMiklos Szeredi new_is_dir, NULL, new_dentry); 4280da1ce067SMiklos Szeredi } 4281da1ce067SMiklos Szeredi } 42820eeca283SRobert Love fsnotify_oldname_free(old_name); 42830eeca283SRobert Love 42841da177e4SLinus Torvalds return error; 42851da177e4SLinus Torvalds } 42864d359507SAl Viro EXPORT_SYMBOL(vfs_rename); 42871da177e4SLinus Torvalds 4288520c8b16SMiklos Szeredi SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname, 4289520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname, unsigned int, flags) 42901da177e4SLinus Torvalds { 42911da177e4SLinus Torvalds struct dentry *old_dentry, *new_dentry; 42921da177e4SLinus Torvalds struct dentry *trap; 4293f5beed75SAl Viro struct path old_path, new_path; 4294f5beed75SAl Viro struct qstr old_last, new_last; 4295f5beed75SAl Viro int old_type, new_type; 42968e6d782cSJ. Bruce Fields struct inode *delegated_inode = NULL; 429791a27b2aSJeff Layton struct filename *from; 429891a27b2aSJeff Layton struct filename *to; 4299f5beed75SAl Viro unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET; 4300c6a94284SJeff Layton bool should_retry = false; 43012ad94ae6SAl Viro int error; 4302520c8b16SMiklos Szeredi 43030d7a8555SMiklos Szeredi if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) 4304da1ce067SMiklos Szeredi return -EINVAL; 4305da1ce067SMiklos Szeredi 43060d7a8555SMiklos Szeredi if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) && 43070d7a8555SMiklos Szeredi (flags & RENAME_EXCHANGE)) 4308520c8b16SMiklos Szeredi return -EINVAL; 4309520c8b16SMiklos Szeredi 43100d7a8555SMiklos Szeredi if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD)) 43110d7a8555SMiklos Szeredi return -EPERM; 43120d7a8555SMiklos Szeredi 4313f5beed75SAl Viro if (flags & RENAME_EXCHANGE) 4314f5beed75SAl Viro target_flags = 0; 4315f5beed75SAl Viro 4316c6a94284SJeff Layton retry: 4317f5beed75SAl Viro from = user_path_parent(olddfd, oldname, 4318f5beed75SAl Viro &old_path, &old_last, &old_type, lookup_flags); 431991a27b2aSJeff Layton if (IS_ERR(from)) { 432091a27b2aSJeff Layton error = PTR_ERR(from); 43211da177e4SLinus Torvalds goto exit; 432291a27b2aSJeff Layton } 43231da177e4SLinus Torvalds 4324f5beed75SAl Viro to = user_path_parent(newdfd, newname, 4325f5beed75SAl Viro &new_path, &new_last, &new_type, lookup_flags); 432691a27b2aSJeff Layton if (IS_ERR(to)) { 432791a27b2aSJeff Layton error = PTR_ERR(to); 43281da177e4SLinus Torvalds goto exit1; 432991a27b2aSJeff Layton } 43301da177e4SLinus Torvalds 43311da177e4SLinus Torvalds error = -EXDEV; 4332f5beed75SAl Viro if (old_path.mnt != new_path.mnt) 43331da177e4SLinus Torvalds goto exit2; 43341da177e4SLinus Torvalds 43351da177e4SLinus Torvalds error = -EBUSY; 4336f5beed75SAl Viro if (old_type != LAST_NORM) 43371da177e4SLinus Torvalds goto exit2; 43381da177e4SLinus Torvalds 43390a7c3937SMiklos Szeredi if (flags & RENAME_NOREPLACE) 43400a7c3937SMiklos Szeredi error = -EEXIST; 4341f5beed75SAl Viro if (new_type != LAST_NORM) 43421da177e4SLinus Torvalds goto exit2; 43431da177e4SLinus Torvalds 4344f5beed75SAl Viro error = mnt_want_write(old_path.mnt); 4345c30dabfeSJan Kara if (error) 4346c30dabfeSJan Kara goto exit2; 4347c30dabfeSJan Kara 43488e6d782cSJ. Bruce Fields retry_deleg: 4349f5beed75SAl Viro trap = lock_rename(new_path.dentry, old_path.dentry); 43501da177e4SLinus Torvalds 4351f5beed75SAl Viro old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags); 43521da177e4SLinus Torvalds error = PTR_ERR(old_dentry); 43531da177e4SLinus Torvalds if (IS_ERR(old_dentry)) 43541da177e4SLinus Torvalds goto exit3; 43551da177e4SLinus Torvalds /* source must exist */ 43561da177e4SLinus Torvalds error = -ENOENT; 4357b18825a7SDavid Howells if (d_is_negative(old_dentry)) 43581da177e4SLinus Torvalds goto exit4; 4359f5beed75SAl Viro new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags); 43601da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 43611da177e4SLinus Torvalds if (IS_ERR(new_dentry)) 43621da177e4SLinus Torvalds goto exit4; 43630a7c3937SMiklos Szeredi error = -EEXIST; 43640a7c3937SMiklos Szeredi if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry)) 43650a7c3937SMiklos Szeredi goto exit5; 4366da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4367da1ce067SMiklos Szeredi error = -ENOENT; 4368da1ce067SMiklos Szeredi if (d_is_negative(new_dentry)) 4369da1ce067SMiklos Szeredi goto exit5; 4370da1ce067SMiklos Szeredi 4371da1ce067SMiklos Szeredi if (!d_is_dir(new_dentry)) { 4372da1ce067SMiklos Szeredi error = -ENOTDIR; 4373f5beed75SAl Viro if (new_last.name[new_last.len]) 4374da1ce067SMiklos Szeredi goto exit5; 4375da1ce067SMiklos Szeredi } 4376da1ce067SMiklos Szeredi } 43770a7c3937SMiklos Szeredi /* unless the source is a directory trailing slashes give -ENOTDIR */ 43780a7c3937SMiklos Szeredi if (!d_is_dir(old_dentry)) { 43790a7c3937SMiklos Szeredi error = -ENOTDIR; 4380f5beed75SAl Viro if (old_last.name[old_last.len]) 43810a7c3937SMiklos Szeredi goto exit5; 4382f5beed75SAl Viro if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len]) 43830a7c3937SMiklos Szeredi goto exit5; 43840a7c3937SMiklos Szeredi } 43850a7c3937SMiklos Szeredi /* source should not be ancestor of target */ 43860a7c3937SMiklos Szeredi error = -EINVAL; 43870a7c3937SMiklos Szeredi if (old_dentry == trap) 43880a7c3937SMiklos Szeredi goto exit5; 43891da177e4SLinus Torvalds /* target should not be an ancestor of source */ 4390da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 43911da177e4SLinus Torvalds error = -ENOTEMPTY; 43921da177e4SLinus Torvalds if (new_dentry == trap) 43931da177e4SLinus Torvalds goto exit5; 43941da177e4SLinus Torvalds 4395f5beed75SAl Viro error = security_path_rename(&old_path, old_dentry, 4396f5beed75SAl Viro &new_path, new_dentry, flags); 4397be6d3e56SKentaro Takeda if (error) 4398c30dabfeSJan Kara goto exit5; 4399f5beed75SAl Viro error = vfs_rename(old_path.dentry->d_inode, old_dentry, 4400f5beed75SAl Viro new_path.dentry->d_inode, new_dentry, 4401520c8b16SMiklos Szeredi &delegated_inode, flags); 44021da177e4SLinus Torvalds exit5: 44031da177e4SLinus Torvalds dput(new_dentry); 44041da177e4SLinus Torvalds exit4: 44051da177e4SLinus Torvalds dput(old_dentry); 44061da177e4SLinus Torvalds exit3: 4407f5beed75SAl Viro unlock_rename(new_path.dentry, old_path.dentry); 44088e6d782cSJ. Bruce Fields if (delegated_inode) { 44098e6d782cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 44108e6d782cSJ. Bruce Fields if (!error) 44118e6d782cSJ. Bruce Fields goto retry_deleg; 44128e6d782cSJ. Bruce Fields } 4413f5beed75SAl Viro mnt_drop_write(old_path.mnt); 44141da177e4SLinus Torvalds exit2: 4415c6a94284SJeff Layton if (retry_estale(error, lookup_flags)) 4416c6a94284SJeff Layton should_retry = true; 4417f5beed75SAl Viro path_put(&new_path); 44182ad94ae6SAl Viro putname(to); 44191da177e4SLinus Torvalds exit1: 4420f5beed75SAl Viro path_put(&old_path); 44211da177e4SLinus Torvalds putname(from); 4422c6a94284SJeff Layton if (should_retry) { 4423c6a94284SJeff Layton should_retry = false; 4424c6a94284SJeff Layton lookup_flags |= LOOKUP_REVAL; 4425c6a94284SJeff Layton goto retry; 4426c6a94284SJeff Layton } 44272ad94ae6SAl Viro exit: 44281da177e4SLinus Torvalds return error; 44291da177e4SLinus Torvalds } 44301da177e4SLinus Torvalds 4431520c8b16SMiklos Szeredi SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, 4432520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname) 4433520c8b16SMiklos Szeredi { 4434520c8b16SMiklos Szeredi return sys_renameat2(olddfd, oldname, newdfd, newname, 0); 4435520c8b16SMiklos Szeredi } 4436520c8b16SMiklos Szeredi 4437a26eab24SHeiko Carstens SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) 44385590ff0dSUlrich Drepper { 4439520c8b16SMiklos Szeredi return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 44405590ff0dSUlrich Drepper } 44415590ff0dSUlrich Drepper 4442787fb6bcSMiklos Szeredi int vfs_whiteout(struct inode *dir, struct dentry *dentry) 4443787fb6bcSMiklos Szeredi { 4444787fb6bcSMiklos Szeredi int error = may_create(dir, dentry); 4445787fb6bcSMiklos Szeredi if (error) 4446787fb6bcSMiklos Szeredi return error; 4447787fb6bcSMiklos Szeredi 4448787fb6bcSMiklos Szeredi if (!dir->i_op->mknod) 4449787fb6bcSMiklos Szeredi return -EPERM; 4450787fb6bcSMiklos Szeredi 4451787fb6bcSMiklos Szeredi return dir->i_op->mknod(dir, dentry, 4452787fb6bcSMiklos Szeredi S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV); 4453787fb6bcSMiklos Szeredi } 4454787fb6bcSMiklos Szeredi EXPORT_SYMBOL(vfs_whiteout); 4455787fb6bcSMiklos Szeredi 44565d826c84SAl Viro int readlink_copy(char __user *buffer, int buflen, const char *link) 44571da177e4SLinus Torvalds { 44585d826c84SAl Viro int len = PTR_ERR(link); 44591da177e4SLinus Torvalds if (IS_ERR(link)) 44601da177e4SLinus Torvalds goto out; 44611da177e4SLinus Torvalds 44621da177e4SLinus Torvalds len = strlen(link); 44631da177e4SLinus Torvalds if (len > (unsigned) buflen) 44641da177e4SLinus Torvalds len = buflen; 44651da177e4SLinus Torvalds if (copy_to_user(buffer, link, len)) 44661da177e4SLinus Torvalds len = -EFAULT; 44671da177e4SLinus Torvalds out: 44681da177e4SLinus Torvalds return len; 44691da177e4SLinus Torvalds } 44705d826c84SAl Viro EXPORT_SYMBOL(readlink_copy); 44711da177e4SLinus Torvalds 44721da177e4SLinus Torvalds /* 44731da177e4SLinus Torvalds * A helper for ->readlink(). This should be used *ONLY* for symlinks that 44741da177e4SLinus Torvalds * have ->follow_link() touching nd only in nd_set_link(). Using (or not 44751da177e4SLinus Torvalds * using) it for any given inode is up to filesystem. 44761da177e4SLinus Torvalds */ 44771da177e4SLinus Torvalds int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen) 44781da177e4SLinus Torvalds { 4479cc314eefSLinus Torvalds void *cookie; 44805f2c4179SAl Viro struct inode *inode = d_inode(dentry); 44815f2c4179SAl Viro const char *link = inode->i_link; 4482694a1764SMarcin Slusarz int res; 4483cc314eefSLinus Torvalds 4484d4dee48bSAl Viro if (!link) { 44855f2c4179SAl Viro link = inode->i_op->follow_link(dentry, &cookie); 4486680baacbSAl Viro if (IS_ERR(link)) 4487680baacbSAl Viro return PTR_ERR(link); 4488d4dee48bSAl Viro } 4489680baacbSAl Viro res = readlink_copy(buffer, buflen, link); 44905f2c4179SAl Viro if (inode->i_op->put_link) 44915f2c4179SAl Viro inode->i_op->put_link(inode, cookie); 4492694a1764SMarcin Slusarz return res; 44931da177e4SLinus Torvalds } 44944d359507SAl Viro EXPORT_SYMBOL(generic_readlink); 44951da177e4SLinus Torvalds 44961da177e4SLinus Torvalds /* get the link contents into pagecache */ 44971da177e4SLinus Torvalds static char *page_getlink(struct dentry * dentry, struct page **ppage) 44981da177e4SLinus Torvalds { 4499ebd09abbSDuane Griffin char *kaddr; 45001da177e4SLinus Torvalds struct page *page; 45011da177e4SLinus Torvalds struct address_space *mapping = dentry->d_inode->i_mapping; 4502090d2b18SPekka Enberg page = read_mapping_page(mapping, 0, NULL); 45031da177e4SLinus Torvalds if (IS_ERR(page)) 45046fe6900eSNick Piggin return (char*)page; 45051da177e4SLinus Torvalds *ppage = page; 4506ebd09abbSDuane Griffin kaddr = kmap(page); 4507ebd09abbSDuane Griffin nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1); 4508ebd09abbSDuane Griffin return kaddr; 45091da177e4SLinus Torvalds } 45101da177e4SLinus Torvalds 45111da177e4SLinus Torvalds int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 45121da177e4SLinus Torvalds { 45131da177e4SLinus Torvalds struct page *page = NULL; 45145d826c84SAl Viro int res = readlink_copy(buffer, buflen, page_getlink(dentry, &page)); 45151da177e4SLinus Torvalds if (page) { 45161da177e4SLinus Torvalds kunmap(page); 45171da177e4SLinus Torvalds page_cache_release(page); 45181da177e4SLinus Torvalds } 45191da177e4SLinus Torvalds return res; 45201da177e4SLinus Torvalds } 45214d359507SAl Viro EXPORT_SYMBOL(page_readlink); 45221da177e4SLinus Torvalds 45236e77137bSAl Viro const char *page_follow_link_light(struct dentry *dentry, void **cookie) 45241da177e4SLinus Torvalds { 4525cc314eefSLinus Torvalds struct page *page = NULL; 4526680baacbSAl Viro char *res = page_getlink(dentry, &page); 4527680baacbSAl Viro if (!IS_ERR(res)) 4528680baacbSAl Viro *cookie = page; 4529680baacbSAl Viro return res; 45301da177e4SLinus Torvalds } 45314d359507SAl Viro EXPORT_SYMBOL(page_follow_link_light); 45321da177e4SLinus Torvalds 45335f2c4179SAl Viro void page_put_link(struct inode *unused, void *cookie) 45341da177e4SLinus Torvalds { 4535cc314eefSLinus Torvalds struct page *page = cookie; 45361da177e4SLinus Torvalds kunmap(page); 45371da177e4SLinus Torvalds page_cache_release(page); 45381da177e4SLinus Torvalds } 45394d359507SAl Viro EXPORT_SYMBOL(page_put_link); 45401da177e4SLinus Torvalds 454154566b2cSNick Piggin /* 454254566b2cSNick Piggin * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS 454354566b2cSNick Piggin */ 454454566b2cSNick Piggin int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) 45451da177e4SLinus Torvalds { 45461da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 45470adb25d2SKirill Korotaev struct page *page; 4548afddba49SNick Piggin void *fsdata; 4549beb497abSDmitriy Monakhov int err; 45501da177e4SLinus Torvalds char *kaddr; 455154566b2cSNick Piggin unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE; 455254566b2cSNick Piggin if (nofs) 455354566b2cSNick Piggin flags |= AOP_FLAG_NOFS; 45541da177e4SLinus Torvalds 45557e53cac4SNeilBrown retry: 4556afddba49SNick Piggin err = pagecache_write_begin(NULL, mapping, 0, len-1, 455754566b2cSNick Piggin flags, &page, &fsdata); 45581da177e4SLinus Torvalds if (err) 4559afddba49SNick Piggin goto fail; 4560afddba49SNick Piggin 4561e8e3c3d6SCong Wang kaddr = kmap_atomic(page); 45621da177e4SLinus Torvalds memcpy(kaddr, symname, len-1); 4563e8e3c3d6SCong Wang kunmap_atomic(kaddr); 4564afddba49SNick Piggin 4565afddba49SNick Piggin err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, 4566afddba49SNick Piggin page, fsdata); 45671da177e4SLinus Torvalds if (err < 0) 45681da177e4SLinus Torvalds goto fail; 4569afddba49SNick Piggin if (err < len-1) 4570afddba49SNick Piggin goto retry; 4571afddba49SNick Piggin 45721da177e4SLinus Torvalds mark_inode_dirty(inode); 45731da177e4SLinus Torvalds return 0; 45741da177e4SLinus Torvalds fail: 45751da177e4SLinus Torvalds return err; 45761da177e4SLinus Torvalds } 45774d359507SAl Viro EXPORT_SYMBOL(__page_symlink); 45781da177e4SLinus Torvalds 45790adb25d2SKirill Korotaev int page_symlink(struct inode *inode, const char *symname, int len) 45800adb25d2SKirill Korotaev { 45810adb25d2SKirill Korotaev return __page_symlink(inode, symname, len, 458254566b2cSNick Piggin !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS)); 45830adb25d2SKirill Korotaev } 45844d359507SAl Viro EXPORT_SYMBOL(page_symlink); 45850adb25d2SKirill Korotaev 458692e1d5beSArjan van de Ven const struct inode_operations page_symlink_inode_operations = { 45871da177e4SLinus Torvalds .readlink = generic_readlink, 45881da177e4SLinus Torvalds .follow_link = page_follow_link_light, 45891da177e4SLinus Torvalds .put_link = page_put_link, 45901da177e4SLinus Torvalds }; 45911da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink_inode_operations); 4592