11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * linux/fs/namei.c 31da177e4SLinus Torvalds * 41da177e4SLinus Torvalds * Copyright (C) 1991, 1992 Linus Torvalds 51da177e4SLinus Torvalds */ 61da177e4SLinus Torvalds 71da177e4SLinus Torvalds /* 81da177e4SLinus Torvalds * Some corrections by tytso. 91da177e4SLinus Torvalds */ 101da177e4SLinus Torvalds 111da177e4SLinus Torvalds /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname 121da177e4SLinus Torvalds * lookup logic. 131da177e4SLinus Torvalds */ 141da177e4SLinus Torvalds /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture. 151da177e4SLinus Torvalds */ 161da177e4SLinus Torvalds 171da177e4SLinus Torvalds #include <linux/init.h> 18630d9c47SPaul Gortmaker #include <linux/export.h> 1944696908SDavid S. Miller #include <linux/kernel.h> 201da177e4SLinus Torvalds #include <linux/slab.h> 211da177e4SLinus Torvalds #include <linux/fs.h> 221da177e4SLinus Torvalds #include <linux/namei.h> 231da177e4SLinus Torvalds #include <linux/pagemap.h> 240eeca283SRobert Love #include <linux/fsnotify.h> 251da177e4SLinus Torvalds #include <linux/personality.h> 261da177e4SLinus Torvalds #include <linux/security.h> 276146f0d5SMimi Zohar #include <linux/ima.h> 281da177e4SLinus Torvalds #include <linux/syscalls.h> 291da177e4SLinus Torvalds #include <linux/mount.h> 301da177e4SLinus Torvalds #include <linux/audit.h> 3116f7e0feSRandy Dunlap #include <linux/capability.h> 32834f2a4aSTrond Myklebust #include <linux/file.h> 335590ff0dSUlrich Drepper #include <linux/fcntl.h> 3408ce5f16SSerge E. Hallyn #include <linux/device_cgroup.h> 355ad4e53bSAl Viro #include <linux/fs_struct.h> 36e77819e5SLinus Torvalds #include <linux/posix_acl.h> 3799d263d4SLinus Torvalds #include <linux/hash.h> 381da177e4SLinus Torvalds #include <asm/uaccess.h> 391da177e4SLinus Torvalds 40e81e3f4dSEric Paris #include "internal.h" 41c7105365SAl Viro #include "mount.h" 42e81e3f4dSEric Paris 431da177e4SLinus Torvalds /* [Feb-1997 T. Schoebel-Theuer] 441da177e4SLinus Torvalds * Fundamental changes in the pathname lookup mechanisms (namei) 451da177e4SLinus Torvalds * were necessary because of omirr. The reason is that omirr needs 461da177e4SLinus Torvalds * to know the _real_ pathname, not the user-supplied one, in case 471da177e4SLinus Torvalds * of symlinks (and also when transname replacements occur). 481da177e4SLinus Torvalds * 491da177e4SLinus Torvalds * The new code replaces the old recursive symlink resolution with 501da177e4SLinus Torvalds * an iterative one (in case of non-nested symlink chains). It does 511da177e4SLinus Torvalds * this with calls to <fs>_follow_link(). 521da177e4SLinus Torvalds * As a side effect, dir_namei(), _namei() and follow_link() are now 531da177e4SLinus Torvalds * replaced with a single function lookup_dentry() that can handle all 541da177e4SLinus Torvalds * the special cases of the former code. 551da177e4SLinus Torvalds * 561da177e4SLinus Torvalds * With the new dcache, the pathname is stored at each inode, at least as 571da177e4SLinus Torvalds * long as the refcount of the inode is positive. As a side effect, the 581da177e4SLinus Torvalds * size of the dcache depends on the inode cache and thus is dynamic. 591da177e4SLinus Torvalds * 601da177e4SLinus Torvalds * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink 611da177e4SLinus Torvalds * resolution to correspond with current state of the code. 621da177e4SLinus Torvalds * 631da177e4SLinus Torvalds * Note that the symlink resolution is not *completely* iterative. 641da177e4SLinus Torvalds * There is still a significant amount of tail- and mid- recursion in 651da177e4SLinus Torvalds * the algorithm. Also, note that <fs>_readlink() is not used in 661da177e4SLinus Torvalds * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink() 671da177e4SLinus Torvalds * may return different results than <fs>_follow_link(). Many virtual 681da177e4SLinus Torvalds * filesystems (including /proc) exhibit this behavior. 691da177e4SLinus Torvalds */ 701da177e4SLinus Torvalds 711da177e4SLinus Torvalds /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation: 721da177e4SLinus Torvalds * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL 731da177e4SLinus Torvalds * and the name already exists in form of a symlink, try to create the new 741da177e4SLinus Torvalds * name indicated by the symlink. The old code always complained that the 751da177e4SLinus Torvalds * name already exists, due to not following the symlink even if its target 761da177e4SLinus Torvalds * is nonexistent. The new semantics affects also mknod() and link() when 7725985edcSLucas De Marchi * the name is a symlink pointing to a non-existent name. 781da177e4SLinus Torvalds * 791da177e4SLinus Torvalds * I don't know which semantics is the right one, since I have no access 801da177e4SLinus Torvalds * to standards. But I found by trial that HP-UX 9.0 has the full "new" 811da177e4SLinus Torvalds * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the 821da177e4SLinus Torvalds * "old" one. Personally, I think the new semantics is much more logical. 831da177e4SLinus Torvalds * Note that "ln old new" where "new" is a symlink pointing to a non-existing 841da177e4SLinus Torvalds * file does succeed in both HP-UX and SunOs, but not in Solaris 851da177e4SLinus Torvalds * and in the old Linux semantics. 861da177e4SLinus Torvalds */ 871da177e4SLinus Torvalds 881da177e4SLinus Torvalds /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink 891da177e4SLinus Torvalds * semantics. See the comments in "open_namei" and "do_link" below. 901da177e4SLinus Torvalds * 911da177e4SLinus Torvalds * [10-Sep-98 Alan Modra] Another symlink change. 921da177e4SLinus Torvalds */ 931da177e4SLinus Torvalds 941da177e4SLinus Torvalds /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks: 951da177e4SLinus Torvalds * inside the path - always follow. 961da177e4SLinus Torvalds * in the last component in creation/removal/renaming - never follow. 971da177e4SLinus Torvalds * if LOOKUP_FOLLOW passed - follow. 981da177e4SLinus Torvalds * if the pathname has trailing slashes - follow. 991da177e4SLinus Torvalds * otherwise - don't follow. 1001da177e4SLinus Torvalds * (applied in that order). 1011da177e4SLinus Torvalds * 1021da177e4SLinus Torvalds * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT 1031da177e4SLinus Torvalds * restored for 2.4. This is the last surviving part of old 4.2BSD bug. 1041da177e4SLinus Torvalds * During the 2.4 we need to fix the userland stuff depending on it - 1051da177e4SLinus Torvalds * hopefully we will be able to get rid of that wart in 2.5. So far only 1061da177e4SLinus Torvalds * XEmacs seems to be relying on it... 1071da177e4SLinus Torvalds */ 1081da177e4SLinus Torvalds /* 1091da177e4SLinus Torvalds * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland) 110a11f3a05SArjan van de Ven * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives 1111da177e4SLinus Torvalds * any extra contention... 1121da177e4SLinus Torvalds */ 1131da177e4SLinus Torvalds 1141da177e4SLinus Torvalds /* In order to reduce some races, while at the same time doing additional 1151da177e4SLinus Torvalds * checking and hopefully speeding things up, we copy filenames to the 1161da177e4SLinus Torvalds * kernel data space before using them.. 1171da177e4SLinus Torvalds * 1181da177e4SLinus Torvalds * POSIX.1 2.4: an empty pathname is invalid (ENOENT). 1191da177e4SLinus Torvalds * PATH_MAX includes the nul terminator --RR. 1201da177e4SLinus Torvalds */ 1217950e385SJeff Layton 122fd2f7cb5SAl Viro #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname)) 12391a27b2aSJeff Layton 12451f39a1fSDavid Drysdale struct filename * 12591a27b2aSJeff Layton getname_flags(const char __user *filename, int flags, int *empty) 12691a27b2aSJeff Layton { 12794b5d262SAl Viro struct filename *result; 1287950e385SJeff Layton char *kname; 12994b5d262SAl Viro int len; 1301da177e4SLinus Torvalds 1317ac86265SJeff Layton result = audit_reusename(filename); 1327ac86265SJeff Layton if (result) 1337ac86265SJeff Layton return result; 1347ac86265SJeff Layton 1357950e385SJeff Layton result = __getname(); 1363f9f0aa6SLinus Torvalds if (unlikely(!result)) 1374043cde8SEric Paris return ERR_PTR(-ENOMEM); 1381da177e4SLinus Torvalds 1397950e385SJeff Layton /* 1407950e385SJeff Layton * First, try to embed the struct filename inside the names_cache 1417950e385SJeff Layton * allocation 1427950e385SJeff Layton */ 143fd2f7cb5SAl Viro kname = (char *)result->iname; 14491a27b2aSJeff Layton result->name = kname; 1457950e385SJeff Layton 14694b5d262SAl Viro len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX); 14791a27b2aSJeff Layton if (unlikely(len < 0)) { 14894b5d262SAl Viro __putname(result); 14994b5d262SAl Viro return ERR_PTR(len); 15091a27b2aSJeff Layton } 1513f9f0aa6SLinus Torvalds 1527950e385SJeff Layton /* 1537950e385SJeff Layton * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a 1547950e385SJeff Layton * separate struct filename so we can dedicate the entire 1557950e385SJeff Layton * names_cache allocation for the pathname, and re-do the copy from 1567950e385SJeff Layton * userland. 1577950e385SJeff Layton */ 15894b5d262SAl Viro if (unlikely(len == EMBEDDED_NAME_MAX)) { 159fd2f7cb5SAl Viro const size_t size = offsetof(struct filename, iname[1]); 1607950e385SJeff Layton kname = (char *)result; 1617950e385SJeff Layton 162fd2f7cb5SAl Viro /* 163fd2f7cb5SAl Viro * size is chosen that way we to guarantee that 164fd2f7cb5SAl Viro * result->iname[0] is within the same object and that 165fd2f7cb5SAl Viro * kname can't be equal to result->iname, no matter what. 166fd2f7cb5SAl Viro */ 167fd2f7cb5SAl Viro result = kzalloc(size, GFP_KERNEL); 16894b5d262SAl Viro if (unlikely(!result)) { 16994b5d262SAl Viro __putname(kname); 17094b5d262SAl Viro return ERR_PTR(-ENOMEM); 1717950e385SJeff Layton } 1727950e385SJeff Layton result->name = kname; 17394b5d262SAl Viro len = strncpy_from_user(kname, filename, PATH_MAX); 17494b5d262SAl Viro if (unlikely(len < 0)) { 17594b5d262SAl Viro __putname(kname); 17694b5d262SAl Viro kfree(result); 17794b5d262SAl Viro return ERR_PTR(len); 17894b5d262SAl Viro } 17994b5d262SAl Viro if (unlikely(len == PATH_MAX)) { 18094b5d262SAl Viro __putname(kname); 18194b5d262SAl Viro kfree(result); 18294b5d262SAl Viro return ERR_PTR(-ENAMETOOLONG); 18394b5d262SAl Viro } 1847950e385SJeff Layton } 1857950e385SJeff Layton 18694b5d262SAl Viro result->refcnt = 1; 1873f9f0aa6SLinus Torvalds /* The empty path is special. */ 1883f9f0aa6SLinus Torvalds if (unlikely(!len)) { 1893f9f0aa6SLinus Torvalds if (empty) 1901fa1e7f6SAndy Whitcroft *empty = 1; 19194b5d262SAl Viro if (!(flags & LOOKUP_EMPTY)) { 19294b5d262SAl Viro putname(result); 19394b5d262SAl Viro return ERR_PTR(-ENOENT); 1941da177e4SLinus Torvalds } 19594b5d262SAl Viro } 1967950e385SJeff Layton 1977950e385SJeff Layton result->uptr = filename; 198c4ad8f98SLinus Torvalds result->aname = NULL; 1991da177e4SLinus Torvalds audit_getname(result); 2001da177e4SLinus Torvalds return result; 2013f9f0aa6SLinus Torvalds } 2023f9f0aa6SLinus Torvalds 20391a27b2aSJeff Layton struct filename * 20491a27b2aSJeff Layton getname(const char __user * filename) 205f52e0c11SAl Viro { 206f7493e5dSLinus Torvalds return getname_flags(filename, 0, NULL); 207f52e0c11SAl Viro } 208f52e0c11SAl Viro 209c4ad8f98SLinus Torvalds struct filename * 210c4ad8f98SLinus Torvalds getname_kernel(const char * filename) 211c4ad8f98SLinus Torvalds { 212c4ad8f98SLinus Torvalds struct filename *result; 21308518549SPaul Moore int len = strlen(filename) + 1; 214c4ad8f98SLinus Torvalds 215c4ad8f98SLinus Torvalds result = __getname(); 216c4ad8f98SLinus Torvalds if (unlikely(!result)) 217c4ad8f98SLinus Torvalds return ERR_PTR(-ENOMEM); 218c4ad8f98SLinus Torvalds 21908518549SPaul Moore if (len <= EMBEDDED_NAME_MAX) { 220fd2f7cb5SAl Viro result->name = (char *)result->iname; 22108518549SPaul Moore } else if (len <= PATH_MAX) { 22208518549SPaul Moore struct filename *tmp; 22308518549SPaul Moore 22408518549SPaul Moore tmp = kmalloc(sizeof(*tmp), GFP_KERNEL); 22508518549SPaul Moore if (unlikely(!tmp)) { 22608518549SPaul Moore __putname(result); 22708518549SPaul Moore return ERR_PTR(-ENOMEM); 22808518549SPaul Moore } 22908518549SPaul Moore tmp->name = (char *)result; 23008518549SPaul Moore result = tmp; 23108518549SPaul Moore } else { 23208518549SPaul Moore __putname(result); 23308518549SPaul Moore return ERR_PTR(-ENAMETOOLONG); 23408518549SPaul Moore } 23508518549SPaul Moore memcpy((char *)result->name, filename, len); 236c4ad8f98SLinus Torvalds result->uptr = NULL; 237c4ad8f98SLinus Torvalds result->aname = NULL; 23855422d0bSPaul Moore result->refcnt = 1; 239fd3522fdSPaul Moore audit_getname(result); 240c4ad8f98SLinus Torvalds 241c4ad8f98SLinus Torvalds return result; 242c4ad8f98SLinus Torvalds } 243c4ad8f98SLinus Torvalds 24491a27b2aSJeff Layton void putname(struct filename *name) 2451da177e4SLinus Torvalds { 24655422d0bSPaul Moore BUG_ON(name->refcnt <= 0); 24755422d0bSPaul Moore 24855422d0bSPaul Moore if (--name->refcnt > 0) 24955422d0bSPaul Moore return; 25055422d0bSPaul Moore 251fd2f7cb5SAl Viro if (name->name != name->iname) { 25255422d0bSPaul Moore __putname(name->name); 25355422d0bSPaul Moore kfree(name); 25455422d0bSPaul Moore } else 25555422d0bSPaul Moore __putname(name); 2561da177e4SLinus Torvalds } 2571da177e4SLinus Torvalds 258e77819e5SLinus Torvalds static int check_acl(struct inode *inode, int mask) 259e77819e5SLinus Torvalds { 26084635d68SLinus Torvalds #ifdef CONFIG_FS_POSIX_ACL 261e77819e5SLinus Torvalds struct posix_acl *acl; 262e77819e5SLinus Torvalds 263e77819e5SLinus Torvalds if (mask & MAY_NOT_BLOCK) { 2643567866bSAl Viro acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS); 2653567866bSAl Viro if (!acl) 266e77819e5SLinus Torvalds return -EAGAIN; 2673567866bSAl Viro /* no ->get_acl() calls in RCU mode... */ 2683567866bSAl Viro if (acl == ACL_NOT_CACHED) 269e77819e5SLinus Torvalds return -ECHILD; 270206b1d09SAri Savolainen return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK); 271e77819e5SLinus Torvalds } 272e77819e5SLinus Torvalds 2732982baa2SChristoph Hellwig acl = get_acl(inode, ACL_TYPE_ACCESS); 2744e34e719SChristoph Hellwig if (IS_ERR(acl)) 2754e34e719SChristoph Hellwig return PTR_ERR(acl); 276e77819e5SLinus Torvalds if (acl) { 277e77819e5SLinus Torvalds int error = posix_acl_permission(inode, acl, mask); 278e77819e5SLinus Torvalds posix_acl_release(acl); 279e77819e5SLinus Torvalds return error; 280e77819e5SLinus Torvalds } 28184635d68SLinus Torvalds #endif 282e77819e5SLinus Torvalds 283e77819e5SLinus Torvalds return -EAGAIN; 284e77819e5SLinus Torvalds } 285e77819e5SLinus Torvalds 2865909ccaaSLinus Torvalds /* 287948409c7SAndreas Gruenbacher * This does the basic permission checking 2885909ccaaSLinus Torvalds */ 2897e40145eSAl Viro static int acl_permission_check(struct inode *inode, int mask) 2905909ccaaSLinus Torvalds { 29126cf46beSLinus Torvalds unsigned int mode = inode->i_mode; 2925909ccaaSLinus Torvalds 2938e96e3b7SEric W. Biederman if (likely(uid_eq(current_fsuid(), inode->i_uid))) 2945909ccaaSLinus Torvalds mode >>= 6; 2955909ccaaSLinus Torvalds else { 296e77819e5SLinus Torvalds if (IS_POSIXACL(inode) && (mode & S_IRWXG)) { 2977e40145eSAl Viro int error = check_acl(inode, mask); 2985909ccaaSLinus Torvalds if (error != -EAGAIN) 2995909ccaaSLinus Torvalds return error; 3005909ccaaSLinus Torvalds } 3015909ccaaSLinus Torvalds 3025909ccaaSLinus Torvalds if (in_group_p(inode->i_gid)) 3035909ccaaSLinus Torvalds mode >>= 3; 3045909ccaaSLinus Torvalds } 3055909ccaaSLinus Torvalds 3065909ccaaSLinus Torvalds /* 3075909ccaaSLinus Torvalds * If the DACs are ok we don't need any capability check. 3085909ccaaSLinus Torvalds */ 3099c2c7039SAl Viro if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) 3105909ccaaSLinus Torvalds return 0; 3115909ccaaSLinus Torvalds return -EACCES; 3125909ccaaSLinus Torvalds } 3131da177e4SLinus Torvalds 3141da177e4SLinus Torvalds /** 3151da177e4SLinus Torvalds * generic_permission - check for access rights on a Posix-like filesystem 3161da177e4SLinus Torvalds * @inode: inode to check access rights for 3178fd90c8dSAndreas Gruenbacher * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...) 3181da177e4SLinus Torvalds * 3191da177e4SLinus Torvalds * Used to check for read/write/execute permissions on a file. 3201da177e4SLinus Torvalds * We use "fsuid" for this, letting us set arbitrary permissions 3211da177e4SLinus Torvalds * for filesystem access without changing the "normal" uids which 322b74c79e9SNick Piggin * are used for other things. 323b74c79e9SNick Piggin * 324b74c79e9SNick Piggin * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk 325b74c79e9SNick Piggin * request cannot be satisfied (eg. requires blocking or too much complexity). 326b74c79e9SNick Piggin * It would then be called again in ref-walk mode. 3271da177e4SLinus Torvalds */ 3282830ba7fSAl Viro int generic_permission(struct inode *inode, int mask) 3291da177e4SLinus Torvalds { 3305909ccaaSLinus Torvalds int ret; 3311da177e4SLinus Torvalds 3321da177e4SLinus Torvalds /* 333948409c7SAndreas Gruenbacher * Do the basic permission checks. 3341da177e4SLinus Torvalds */ 3357e40145eSAl Viro ret = acl_permission_check(inode, mask); 3365909ccaaSLinus Torvalds if (ret != -EACCES) 3375909ccaaSLinus Torvalds return ret; 3381da177e4SLinus Torvalds 339d594e7ecSAl Viro if (S_ISDIR(inode->i_mode)) { 340d594e7ecSAl Viro /* DACs are overridable for directories */ 34123adbe12SAndy Lutomirski if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE)) 342d594e7ecSAl Viro return 0; 343d594e7ecSAl Viro if (!(mask & MAY_WRITE)) 34423adbe12SAndy Lutomirski if (capable_wrt_inode_uidgid(inode, 34523adbe12SAndy Lutomirski CAP_DAC_READ_SEARCH)) 346d594e7ecSAl Viro return 0; 347d594e7ecSAl Viro return -EACCES; 348d594e7ecSAl Viro } 3491da177e4SLinus Torvalds /* 3501da177e4SLinus Torvalds * Read/write DACs are always overridable. 351d594e7ecSAl Viro * Executable DACs are overridable when there is 352d594e7ecSAl Viro * at least one exec bit set. 3531da177e4SLinus Torvalds */ 354d594e7ecSAl Viro if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO)) 35523adbe12SAndy Lutomirski if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE)) 3561da177e4SLinus Torvalds return 0; 3571da177e4SLinus Torvalds 3581da177e4SLinus Torvalds /* 3591da177e4SLinus Torvalds * Searching includes executable on directories, else just read. 3601da177e4SLinus Torvalds */ 3617ea66001SSerge E. Hallyn mask &= MAY_READ | MAY_WRITE | MAY_EXEC; 362d594e7ecSAl Viro if (mask == MAY_READ) 36323adbe12SAndy Lutomirski if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH)) 3641da177e4SLinus Torvalds return 0; 3651da177e4SLinus Torvalds 3661da177e4SLinus Torvalds return -EACCES; 3671da177e4SLinus Torvalds } 3684d359507SAl Viro EXPORT_SYMBOL(generic_permission); 3691da177e4SLinus Torvalds 3703ddcd056SLinus Torvalds /* 3713ddcd056SLinus Torvalds * We _really_ want to just do "generic_permission()" without 3723ddcd056SLinus Torvalds * even looking at the inode->i_op values. So we keep a cache 3733ddcd056SLinus Torvalds * flag in inode->i_opflags, that says "this has not special 3743ddcd056SLinus Torvalds * permission function, use the fast case". 3753ddcd056SLinus Torvalds */ 3763ddcd056SLinus Torvalds static inline int do_inode_permission(struct inode *inode, int mask) 3773ddcd056SLinus Torvalds { 3783ddcd056SLinus Torvalds if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) { 3793ddcd056SLinus Torvalds if (likely(inode->i_op->permission)) 3803ddcd056SLinus Torvalds return inode->i_op->permission(inode, mask); 3813ddcd056SLinus Torvalds 3823ddcd056SLinus Torvalds /* This gets set once for the inode lifetime */ 3833ddcd056SLinus Torvalds spin_lock(&inode->i_lock); 3843ddcd056SLinus Torvalds inode->i_opflags |= IOP_FASTPERM; 3853ddcd056SLinus Torvalds spin_unlock(&inode->i_lock); 3863ddcd056SLinus Torvalds } 3873ddcd056SLinus Torvalds return generic_permission(inode, mask); 3883ddcd056SLinus Torvalds } 3893ddcd056SLinus Torvalds 390cb23beb5SChristoph Hellwig /** 3910bdaea90SDavid Howells * __inode_permission - Check for access rights to a given inode 3920bdaea90SDavid Howells * @inode: Inode to check permission on 3930bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 394cb23beb5SChristoph Hellwig * 3950bdaea90SDavid Howells * Check for read/write/execute permissions on an inode. 396948409c7SAndreas Gruenbacher * 397948409c7SAndreas Gruenbacher * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. 3980bdaea90SDavid Howells * 3990bdaea90SDavid Howells * This does not check for a read-only file system. You probably want 4000bdaea90SDavid Howells * inode_permission(). 401cb23beb5SChristoph Hellwig */ 4020bdaea90SDavid Howells int __inode_permission(struct inode *inode, int mask) 4031da177e4SLinus Torvalds { 404e6305c43SAl Viro int retval; 4051da177e4SLinus Torvalds 4063ddcd056SLinus Torvalds if (unlikely(mask & MAY_WRITE)) { 4071da177e4SLinus Torvalds /* 4081da177e4SLinus Torvalds * Nobody gets write access to an immutable file. 4091da177e4SLinus Torvalds */ 4101da177e4SLinus Torvalds if (IS_IMMUTABLE(inode)) 4111da177e4SLinus Torvalds return -EACCES; 4121da177e4SLinus Torvalds } 4131da177e4SLinus Torvalds 4143ddcd056SLinus Torvalds retval = do_inode_permission(inode, mask); 4151da177e4SLinus Torvalds if (retval) 4161da177e4SLinus Torvalds return retval; 4171da177e4SLinus Torvalds 41808ce5f16SSerge E. Hallyn retval = devcgroup_inode_permission(inode, mask); 41908ce5f16SSerge E. Hallyn if (retval) 42008ce5f16SSerge E. Hallyn return retval; 42108ce5f16SSerge E. Hallyn 422d09ca739SEric Paris return security_inode_permission(inode, mask); 4231da177e4SLinus Torvalds } 424bd5d0856SMiklos Szeredi EXPORT_SYMBOL(__inode_permission); 4251da177e4SLinus Torvalds 426f4d6ff89SAl Viro /** 4270bdaea90SDavid Howells * sb_permission - Check superblock-level permissions 4280bdaea90SDavid Howells * @sb: Superblock of inode to check permission on 42955852635SRandy Dunlap * @inode: Inode to check permission on 4300bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 4310bdaea90SDavid Howells * 4320bdaea90SDavid Howells * Separate out file-system wide checks from inode-specific permission checks. 4330bdaea90SDavid Howells */ 4340bdaea90SDavid Howells static int sb_permission(struct super_block *sb, struct inode *inode, int mask) 4350bdaea90SDavid Howells { 4360bdaea90SDavid Howells if (unlikely(mask & MAY_WRITE)) { 4370bdaea90SDavid Howells umode_t mode = inode->i_mode; 4380bdaea90SDavid Howells 4390bdaea90SDavid Howells /* Nobody gets write access to a read-only fs. */ 4400bdaea90SDavid Howells if ((sb->s_flags & MS_RDONLY) && 4410bdaea90SDavid Howells (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) 4420bdaea90SDavid Howells return -EROFS; 4430bdaea90SDavid Howells } 4440bdaea90SDavid Howells return 0; 4450bdaea90SDavid Howells } 4460bdaea90SDavid Howells 4470bdaea90SDavid Howells /** 4480bdaea90SDavid Howells * inode_permission - Check for access rights to a given inode 4490bdaea90SDavid Howells * @inode: Inode to check permission on 4500bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 4510bdaea90SDavid Howells * 4520bdaea90SDavid Howells * Check for read/write/execute permissions on an inode. We use fs[ug]id for 4530bdaea90SDavid Howells * this, letting us set arbitrary permissions for filesystem access without 4540bdaea90SDavid Howells * changing the "normal" UIDs which are used for other things. 4550bdaea90SDavid Howells * 4560bdaea90SDavid Howells * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. 4570bdaea90SDavid Howells */ 4580bdaea90SDavid Howells int inode_permission(struct inode *inode, int mask) 4590bdaea90SDavid Howells { 4600bdaea90SDavid Howells int retval; 4610bdaea90SDavid Howells 4620bdaea90SDavid Howells retval = sb_permission(inode->i_sb, inode, mask); 4630bdaea90SDavid Howells if (retval) 4640bdaea90SDavid Howells return retval; 4650bdaea90SDavid Howells return __inode_permission(inode, mask); 4660bdaea90SDavid Howells } 4674d359507SAl Viro EXPORT_SYMBOL(inode_permission); 4680bdaea90SDavid Howells 4690bdaea90SDavid Howells /** 4705dd784d0SJan Blunck * path_get - get a reference to a path 4715dd784d0SJan Blunck * @path: path to get the reference to 4725dd784d0SJan Blunck * 4735dd784d0SJan Blunck * Given a path increment the reference count to the dentry and the vfsmount. 4745dd784d0SJan Blunck */ 475dcf787f3SAl Viro void path_get(const struct path *path) 4765dd784d0SJan Blunck { 4775dd784d0SJan Blunck mntget(path->mnt); 4785dd784d0SJan Blunck dget(path->dentry); 4795dd784d0SJan Blunck } 4805dd784d0SJan Blunck EXPORT_SYMBOL(path_get); 4815dd784d0SJan Blunck 4825dd784d0SJan Blunck /** 4831d957f9bSJan Blunck * path_put - put a reference to a path 4841d957f9bSJan Blunck * @path: path to put the reference to 4851d957f9bSJan Blunck * 4861d957f9bSJan Blunck * Given a path decrement the reference count to the dentry and the vfsmount. 4871d957f9bSJan Blunck */ 488dcf787f3SAl Viro void path_put(const struct path *path) 4891da177e4SLinus Torvalds { 4901d957f9bSJan Blunck dput(path->dentry); 4911d957f9bSJan Blunck mntput(path->mnt); 4921da177e4SLinus Torvalds } 4931d957f9bSJan Blunck EXPORT_SYMBOL(path_put); 4941da177e4SLinus Torvalds 495894bc8c4SAl Viro #define EMBEDDED_LEVELS 2 4961f55a6ecSAl Viro struct nameidata { 4971f55a6ecSAl Viro struct path path; 4981f55a6ecSAl Viro struct qstr last; 4991f55a6ecSAl Viro struct path root; 5001f55a6ecSAl Viro struct inode *inode; /* path.dentry.d_inode */ 5011f55a6ecSAl Viro unsigned int flags; 5029883d185SAl Viro unsigned seq, m_seq; 5031f55a6ecSAl Viro int last_type; 5041f55a6ecSAl Viro unsigned depth; 505756daf26SNeilBrown int total_link_count; 506697fc6caSAl Viro struct saved { 507697fc6caSAl Viro struct path link; 508697fc6caSAl Viro void *cookie; 509697fc6caSAl Viro const char *name; 510237d8b32SAl Viro struct inode *inode; 5110450b2d1SAl Viro unsigned seq; 512894bc8c4SAl Viro } *stack, internal[EMBEDDED_LEVELS]; 5139883d185SAl Viro struct filename *name; 5149883d185SAl Viro struct nameidata *saved; 5159883d185SAl Viro unsigned root_seq; 5169883d185SAl Viro int dfd; 5171f55a6ecSAl Viro }; 5181f55a6ecSAl Viro 5199883d185SAl Viro static void set_nameidata(struct nameidata *p, int dfd, struct filename *name) 520894bc8c4SAl Viro { 521756daf26SNeilBrown struct nameidata *old = current->nameidata; 522756daf26SNeilBrown p->stack = p->internal; 523c8a53ee5SAl Viro p->dfd = dfd; 524c8a53ee5SAl Viro p->name = name; 525756daf26SNeilBrown p->total_link_count = old ? old->total_link_count : 0; 5269883d185SAl Viro p->saved = old; 527756daf26SNeilBrown current->nameidata = p; 528894bc8c4SAl Viro } 529894bc8c4SAl Viro 5309883d185SAl Viro static void restore_nameidata(void) 531894bc8c4SAl Viro { 5329883d185SAl Viro struct nameidata *now = current->nameidata, *old = now->saved; 533756daf26SNeilBrown 534756daf26SNeilBrown current->nameidata = old; 535756daf26SNeilBrown if (old) 536756daf26SNeilBrown old->total_link_count = now->total_link_count; 537756daf26SNeilBrown if (now->stack != now->internal) { 538756daf26SNeilBrown kfree(now->stack); 539756daf26SNeilBrown now->stack = now->internal; 540894bc8c4SAl Viro } 541894bc8c4SAl Viro } 542894bc8c4SAl Viro 543894bc8c4SAl Viro static int __nd_alloc_stack(struct nameidata *nd) 544894bc8c4SAl Viro { 545bc40aee0SAl Viro struct saved *p; 546bc40aee0SAl Viro 547bc40aee0SAl Viro if (nd->flags & LOOKUP_RCU) { 548bc40aee0SAl Viro p= kmalloc(MAXSYMLINKS * sizeof(struct saved), 549bc40aee0SAl Viro GFP_ATOMIC); 550bc40aee0SAl Viro if (unlikely(!p)) 551bc40aee0SAl Viro return -ECHILD; 552bc40aee0SAl Viro } else { 553bc40aee0SAl Viro p= kmalloc(MAXSYMLINKS * sizeof(struct saved), 554894bc8c4SAl Viro GFP_KERNEL); 555894bc8c4SAl Viro if (unlikely(!p)) 556894bc8c4SAl Viro return -ENOMEM; 557bc40aee0SAl Viro } 558894bc8c4SAl Viro memcpy(p, nd->internal, sizeof(nd->internal)); 559894bc8c4SAl Viro nd->stack = p; 560894bc8c4SAl Viro return 0; 561894bc8c4SAl Viro } 562894bc8c4SAl Viro 563397d425dSEric W. Biederman /** 564397d425dSEric W. Biederman * path_connected - Verify that a path->dentry is below path->mnt.mnt_root 565397d425dSEric W. Biederman * @path: nameidate to verify 566397d425dSEric W. Biederman * 567397d425dSEric W. Biederman * Rename can sometimes move a file or directory outside of a bind 568397d425dSEric W. Biederman * mount, path_connected allows those cases to be detected. 569397d425dSEric W. Biederman */ 570397d425dSEric W. Biederman static bool path_connected(const struct path *path) 571397d425dSEric W. Biederman { 572397d425dSEric W. Biederman struct vfsmount *mnt = path->mnt; 573397d425dSEric W. Biederman 574397d425dSEric W. Biederman /* Only bind mounts can have disconnected paths */ 575397d425dSEric W. Biederman if (mnt->mnt_root == mnt->mnt_sb->s_root) 576397d425dSEric W. Biederman return true; 577397d425dSEric W. Biederman 578397d425dSEric W. Biederman return is_subdir(path->dentry, mnt->mnt_root); 579397d425dSEric W. Biederman } 580397d425dSEric W. Biederman 581894bc8c4SAl Viro static inline int nd_alloc_stack(struct nameidata *nd) 582894bc8c4SAl Viro { 583da4e0be0SAl Viro if (likely(nd->depth != EMBEDDED_LEVELS)) 584894bc8c4SAl Viro return 0; 585894bc8c4SAl Viro if (likely(nd->stack != nd->internal)) 586894bc8c4SAl Viro return 0; 587894bc8c4SAl Viro return __nd_alloc_stack(nd); 588894bc8c4SAl Viro } 589894bc8c4SAl Viro 5907973387aSAl Viro static void drop_links(struct nameidata *nd) 5917973387aSAl Viro { 5927973387aSAl Viro int i = nd->depth; 5937973387aSAl Viro while (i--) { 5947973387aSAl Viro struct saved *last = nd->stack + i; 5957973387aSAl Viro struct inode *inode = last->inode; 5967973387aSAl Viro if (last->cookie && inode->i_op->put_link) { 5977973387aSAl Viro inode->i_op->put_link(inode, last->cookie); 5987973387aSAl Viro last->cookie = NULL; 5997973387aSAl Viro } 6007973387aSAl Viro } 6017973387aSAl Viro } 6027973387aSAl Viro 6037973387aSAl Viro static void terminate_walk(struct nameidata *nd) 6047973387aSAl Viro { 6057973387aSAl Viro drop_links(nd); 6067973387aSAl Viro if (!(nd->flags & LOOKUP_RCU)) { 6077973387aSAl Viro int i; 6087973387aSAl Viro path_put(&nd->path); 6097973387aSAl Viro for (i = 0; i < nd->depth; i++) 6107973387aSAl Viro path_put(&nd->stack[i].link); 611102b8af2SAl Viro if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 612102b8af2SAl Viro path_put(&nd->root); 613102b8af2SAl Viro nd->root.mnt = NULL; 614102b8af2SAl Viro } 6157973387aSAl Viro } else { 6167973387aSAl Viro nd->flags &= ~LOOKUP_RCU; 6177973387aSAl Viro if (!(nd->flags & LOOKUP_ROOT)) 6187973387aSAl Viro nd->root.mnt = NULL; 6197973387aSAl Viro rcu_read_unlock(); 6207973387aSAl Viro } 6217973387aSAl Viro nd->depth = 0; 6227973387aSAl Viro } 6237973387aSAl Viro 6247973387aSAl Viro /* path_put is needed afterwards regardless of success or failure */ 6257973387aSAl Viro static bool legitimize_path(struct nameidata *nd, 6267973387aSAl Viro struct path *path, unsigned seq) 6277973387aSAl Viro { 6287973387aSAl Viro int res = __legitimize_mnt(path->mnt, nd->m_seq); 6297973387aSAl Viro if (unlikely(res)) { 6307973387aSAl Viro if (res > 0) 6317973387aSAl Viro path->mnt = NULL; 6327973387aSAl Viro path->dentry = NULL; 6337973387aSAl Viro return false; 6347973387aSAl Viro } 6357973387aSAl Viro if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) { 6367973387aSAl Viro path->dentry = NULL; 6377973387aSAl Viro return false; 6387973387aSAl Viro } 6397973387aSAl Viro return !read_seqcount_retry(&path->dentry->d_seq, seq); 6407973387aSAl Viro } 6417973387aSAl Viro 6427973387aSAl Viro static bool legitimize_links(struct nameidata *nd) 6437973387aSAl Viro { 6447973387aSAl Viro int i; 6457973387aSAl Viro for (i = 0; i < nd->depth; i++) { 6467973387aSAl Viro struct saved *last = nd->stack + i; 6477973387aSAl Viro if (unlikely(!legitimize_path(nd, &last->link, last->seq))) { 6487973387aSAl Viro drop_links(nd); 6497973387aSAl Viro nd->depth = i + 1; 6507973387aSAl Viro return false; 6517973387aSAl Viro } 6527973387aSAl Viro } 6537973387aSAl Viro return true; 6547973387aSAl Viro } 6557973387aSAl Viro 65619660af7SAl Viro /* 65731e6b01fSNick Piggin * Path walking has 2 modes, rcu-walk and ref-walk (see 65819660af7SAl Viro * Documentation/filesystems/path-lookup.txt). In situations when we can't 65919660af7SAl Viro * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab 66057e3715cSMike Marshall * normal reference counts on dentries and vfsmounts to transition to ref-walk 66119660af7SAl Viro * mode. Refcounts are grabbed at the last known good point before rcu-walk 66219660af7SAl Viro * got stuck, so ref-walk may continue from there. If this is not successful 66319660af7SAl Viro * (eg. a seqcount has changed), then failure is returned and it's up to caller 66419660af7SAl Viro * to restart the path walk from the beginning in ref-walk mode. 66531e6b01fSNick Piggin */ 66631e6b01fSNick Piggin 66731e6b01fSNick Piggin /** 66819660af7SAl Viro * unlazy_walk - try to switch to ref-walk mode. 66919660af7SAl Viro * @nd: nameidata pathwalk data 67019660af7SAl Viro * @dentry: child of nd->path.dentry or NULL 6716e9918b7SAl Viro * @seq: seq number to check dentry against 67239191628SRandy Dunlap * Returns: 0 on success, -ECHILD on failure 67331e6b01fSNick Piggin * 67419660af7SAl Viro * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry 67519660af7SAl Viro * for ref-walk mode. @dentry must be a path found by a do_lookup call on 67619660af7SAl Viro * @nd or NULL. Must be called from rcu-walk context. 6777973387aSAl Viro * Nothing should touch nameidata between unlazy_walk() failure and 6787973387aSAl Viro * terminate_walk(). 67931e6b01fSNick Piggin */ 6806e9918b7SAl Viro static int unlazy_walk(struct nameidata *nd, struct dentry *dentry, unsigned seq) 68131e6b01fSNick Piggin { 68231e6b01fSNick Piggin struct dentry *parent = nd->path.dentry; 68331e6b01fSNick Piggin 68431e6b01fSNick Piggin BUG_ON(!(nd->flags & LOOKUP_RCU)); 685e5c832d5SLinus Torvalds 686e5c832d5SLinus Torvalds nd->flags &= ~LOOKUP_RCU; 6877973387aSAl Viro if (unlikely(!legitimize_links(nd))) 6887973387aSAl Viro goto out2; 6897973387aSAl Viro if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq))) 6907973387aSAl Viro goto out2; 6917973387aSAl Viro if (unlikely(!lockref_get_not_dead(&parent->d_lockref))) 6927973387aSAl Viro goto out1; 69348a066e7SAl Viro 69415570086SLinus Torvalds /* 69515570086SLinus Torvalds * For a negative lookup, the lookup sequence point is the parents 69615570086SLinus Torvalds * sequence point, and it only needs to revalidate the parent dentry. 69715570086SLinus Torvalds * 69815570086SLinus Torvalds * For a positive lookup, we need to move both the parent and the 69915570086SLinus Torvalds * dentry from the RCU domain to be properly refcounted. And the 70015570086SLinus Torvalds * sequence number in the dentry validates *both* dentry counters, 70115570086SLinus Torvalds * since we checked the sequence number of the parent after we got 70215570086SLinus Torvalds * the child sequence number. So we know the parent must still 70315570086SLinus Torvalds * be valid if the child sequence number is still valid. 70415570086SLinus Torvalds */ 70519660af7SAl Viro if (!dentry) { 706e5c832d5SLinus Torvalds if (read_seqcount_retry(&parent->d_seq, nd->seq)) 707e5c832d5SLinus Torvalds goto out; 70819660af7SAl Viro BUG_ON(nd->inode != parent->d_inode); 70919660af7SAl Viro } else { 710e5c832d5SLinus Torvalds if (!lockref_get_not_dead(&dentry->d_lockref)) 711e5c832d5SLinus Torvalds goto out; 7126e9918b7SAl Viro if (read_seqcount_retry(&dentry->d_seq, seq)) 713e5c832d5SLinus Torvalds goto drop_dentry; 71419660af7SAl Viro } 715e5c832d5SLinus Torvalds 716e5c832d5SLinus Torvalds /* 717e5c832d5SLinus Torvalds * Sequence counts matched. Now make sure that the root is 718e5c832d5SLinus Torvalds * still valid and get it if required. 719e5c832d5SLinus Torvalds */ 720e5c832d5SLinus Torvalds if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 7215a8d87e8SAl Viro if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) { 7225a8d87e8SAl Viro rcu_read_unlock(); 7235a8d87e8SAl Viro dput(dentry); 7245a8d87e8SAl Viro return -ECHILD; 7257973387aSAl Viro } 72631e6b01fSNick Piggin } 72731e6b01fSNick Piggin 7288b61e74fSAl Viro rcu_read_unlock(); 72931e6b01fSNick Piggin return 0; 73019660af7SAl Viro 731e5c832d5SLinus Torvalds drop_dentry: 7328b61e74fSAl Viro rcu_read_unlock(); 733e5c832d5SLinus Torvalds dput(dentry); 734d0d27277SLinus Torvalds goto drop_root_mnt; 7357973387aSAl Viro out2: 7367973387aSAl Viro nd->path.mnt = NULL; 7377973387aSAl Viro out1: 7387973387aSAl Viro nd->path.dentry = NULL; 739e5c832d5SLinus Torvalds out: 7408b61e74fSAl Viro rcu_read_unlock(); 741d0d27277SLinus Torvalds drop_root_mnt: 742d0d27277SLinus Torvalds if (!(nd->flags & LOOKUP_ROOT)) 743d0d27277SLinus Torvalds nd->root.mnt = NULL; 74431e6b01fSNick Piggin return -ECHILD; 74531e6b01fSNick Piggin } 74631e6b01fSNick Piggin 7477973387aSAl Viro static int unlazy_link(struct nameidata *nd, struct path *link, unsigned seq) 7487973387aSAl Viro { 7497973387aSAl Viro if (unlikely(!legitimize_path(nd, link, seq))) { 7507973387aSAl Viro drop_links(nd); 7517973387aSAl Viro nd->depth = 0; 7527973387aSAl Viro nd->flags &= ~LOOKUP_RCU; 7537973387aSAl Viro nd->path.mnt = NULL; 7547973387aSAl Viro nd->path.dentry = NULL; 7557973387aSAl Viro if (!(nd->flags & LOOKUP_ROOT)) 7567973387aSAl Viro nd->root.mnt = NULL; 7577973387aSAl Viro rcu_read_unlock(); 7587973387aSAl Viro } else if (likely(unlazy_walk(nd, NULL, 0)) == 0) { 7597973387aSAl Viro return 0; 7607973387aSAl Viro } 7617973387aSAl Viro path_put(link); 7627973387aSAl Viro return -ECHILD; 7637973387aSAl Viro } 7647973387aSAl Viro 7654ce16ef3SAl Viro static inline int d_revalidate(struct dentry *dentry, unsigned int flags) 76634286d66SNick Piggin { 7674ce16ef3SAl Viro return dentry->d_op->d_revalidate(dentry, flags); 76834286d66SNick Piggin } 76934286d66SNick Piggin 7709f1fafeeSAl Viro /** 7719f1fafeeSAl Viro * complete_walk - successful completion of path walk 7729f1fafeeSAl Viro * @nd: pointer nameidata 77339159de2SJeff Layton * 7749f1fafeeSAl Viro * If we had been in RCU mode, drop out of it and legitimize nd->path. 7759f1fafeeSAl Viro * Revalidate the final result, unless we'd already done that during 7769f1fafeeSAl Viro * the path walk or the filesystem doesn't ask for it. Return 0 on 7779f1fafeeSAl Viro * success, -error on failure. In case of failure caller does not 7789f1fafeeSAl Viro * need to drop nd->path. 77939159de2SJeff Layton */ 7809f1fafeeSAl Viro static int complete_walk(struct nameidata *nd) 78139159de2SJeff Layton { 78216c2cd71SAl Viro struct dentry *dentry = nd->path.dentry; 78339159de2SJeff Layton int status; 78439159de2SJeff Layton 7859f1fafeeSAl Viro if (nd->flags & LOOKUP_RCU) { 7869f1fafeeSAl Viro if (!(nd->flags & LOOKUP_ROOT)) 7879f1fafeeSAl Viro nd->root.mnt = NULL; 7886e9918b7SAl Viro if (unlikely(unlazy_walk(nd, NULL, 0))) 78948a066e7SAl Viro return -ECHILD; 79048a066e7SAl Viro } 7919f1fafeeSAl Viro 79216c2cd71SAl Viro if (likely(!(nd->flags & LOOKUP_JUMPED))) 79339159de2SJeff Layton return 0; 79439159de2SJeff Layton 795ecf3d1f1SJeff Layton if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) 79616c2cd71SAl Viro return 0; 79716c2cd71SAl Viro 798ecf3d1f1SJeff Layton status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); 79939159de2SJeff Layton if (status > 0) 80039159de2SJeff Layton return 0; 80139159de2SJeff Layton 80216c2cd71SAl Viro if (!status) 80339159de2SJeff Layton status = -ESTALE; 80416c2cd71SAl Viro 80539159de2SJeff Layton return status; 80639159de2SJeff Layton } 80739159de2SJeff Layton 80818d8c860SAl Viro static void set_root(struct nameidata *nd) 8092a737871SAl Viro { 810f7ad3c6bSMiklos Szeredi get_fs_root(current->fs, &nd->root); 8112a737871SAl Viro } 8122a737871SAl Viro 81306d7137eSAl Viro static void set_root_rcu(struct nameidata *nd) 81431e6b01fSNick Piggin { 81531e6b01fSNick Piggin struct fs_struct *fs = current->fs; 8168f47a016SAl Viro unsigned seq; 817c28cc364SNick Piggin 818c28cc364SNick Piggin do { 819c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 82031e6b01fSNick Piggin nd->root = fs->root; 8218f47a016SAl Viro nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq); 822c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 82331e6b01fSNick Piggin } 82431e6b01fSNick Piggin 8251d957f9bSJan Blunck static void path_put_conditional(struct path *path, struct nameidata *nd) 826051d3812SIan Kent { 827051d3812SIan Kent dput(path->dentry); 8284ac91378SJan Blunck if (path->mnt != nd->path.mnt) 829051d3812SIan Kent mntput(path->mnt); 830051d3812SIan Kent } 831051d3812SIan Kent 8327b9337aaSNick Piggin static inline void path_to_nameidata(const struct path *path, 8337b9337aaSNick Piggin struct nameidata *nd) 834051d3812SIan Kent { 83531e6b01fSNick Piggin if (!(nd->flags & LOOKUP_RCU)) { 8364ac91378SJan Blunck dput(nd->path.dentry); 83731e6b01fSNick Piggin if (nd->path.mnt != path->mnt) 8384ac91378SJan Blunck mntput(nd->path.mnt); 8399a229683SHuang Shijie } 84031e6b01fSNick Piggin nd->path.mnt = path->mnt; 8414ac91378SJan Blunck nd->path.dentry = path->dentry; 842051d3812SIan Kent } 843051d3812SIan Kent 844b5fb63c1SChristoph Hellwig /* 845b5fb63c1SChristoph Hellwig * Helper to directly jump to a known parsed path from ->follow_link, 846b5fb63c1SChristoph Hellwig * caller must have taken a reference to path beforehand. 847b5fb63c1SChristoph Hellwig */ 8486e77137bSAl Viro void nd_jump_link(struct path *path) 849b5fb63c1SChristoph Hellwig { 8506e77137bSAl Viro struct nameidata *nd = current->nameidata; 851b5fb63c1SChristoph Hellwig path_put(&nd->path); 852b5fb63c1SChristoph Hellwig 853b5fb63c1SChristoph Hellwig nd->path = *path; 854b5fb63c1SChristoph Hellwig nd->inode = nd->path.dentry->d_inode; 855b5fb63c1SChristoph Hellwig nd->flags |= LOOKUP_JUMPED; 856b5fb63c1SChristoph Hellwig } 857b5fb63c1SChristoph Hellwig 858b9ff4429SAl Viro static inline void put_link(struct nameidata *nd) 859574197e0SAl Viro { 86021c3003dSAl Viro struct saved *last = nd->stack + --nd->depth; 861237d8b32SAl Viro struct inode *inode = last->inode; 862b9ff4429SAl Viro if (last->cookie && inode->i_op->put_link) 8635f2c4179SAl Viro inode->i_op->put_link(inode, last->cookie); 8646548fae2SAl Viro if (!(nd->flags & LOOKUP_RCU)) 865b9ff4429SAl Viro path_put(&last->link); 866574197e0SAl Viro } 867574197e0SAl Viro 868561ec64aSLinus Torvalds int sysctl_protected_symlinks __read_mostly = 0; 869561ec64aSLinus Torvalds int sysctl_protected_hardlinks __read_mostly = 0; 870800179c9SKees Cook 871800179c9SKees Cook /** 872800179c9SKees Cook * may_follow_link - Check symlink following for unsafe situations 87355852635SRandy Dunlap * @nd: nameidata pathwalk data 874800179c9SKees Cook * 875800179c9SKees Cook * In the case of the sysctl_protected_symlinks sysctl being enabled, 876800179c9SKees Cook * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is 877800179c9SKees Cook * in a sticky world-writable directory. This is to protect privileged 878800179c9SKees Cook * processes from failing races against path names that may change out 879800179c9SKees Cook * from under them by way of other users creating malicious symlinks. 880800179c9SKees Cook * It will permit symlinks to be followed only when outside a sticky 881800179c9SKees Cook * world-writable directory, or when the uid of the symlink and follower 882800179c9SKees Cook * match, or when the directory owner matches the symlink's owner. 883800179c9SKees Cook * 884800179c9SKees Cook * Returns 0 if following the symlink is allowed, -ve on error. 885800179c9SKees Cook */ 886fec2fa24SAl Viro static inline int may_follow_link(struct nameidata *nd) 887800179c9SKees Cook { 888800179c9SKees Cook const struct inode *inode; 889800179c9SKees Cook const struct inode *parent; 890800179c9SKees Cook 891800179c9SKees Cook if (!sysctl_protected_symlinks) 892800179c9SKees Cook return 0; 893800179c9SKees Cook 894800179c9SKees Cook /* Allowed if owner and follower match. */ 895237d8b32SAl Viro inode = nd->stack[0].inode; 89681abe27bSEric W. Biederman if (uid_eq(current_cred()->fsuid, inode->i_uid)) 897800179c9SKees Cook return 0; 898800179c9SKees Cook 899800179c9SKees Cook /* Allowed if parent directory not sticky and world-writable. */ 900aa65fa35SAl Viro parent = nd->inode; 901800179c9SKees Cook if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) 902800179c9SKees Cook return 0; 903800179c9SKees Cook 904800179c9SKees Cook /* Allowed if parent directory and link owner match. */ 90581abe27bSEric W. Biederman if (uid_eq(parent->i_uid, inode->i_uid)) 906800179c9SKees Cook return 0; 907800179c9SKees Cook 90831956502SAl Viro if (nd->flags & LOOKUP_RCU) 90931956502SAl Viro return -ECHILD; 91031956502SAl Viro 9111cf2665bSAl Viro audit_log_link_denied("follow_link", &nd->stack[0].link); 912800179c9SKees Cook return -EACCES; 913800179c9SKees Cook } 914800179c9SKees Cook 915800179c9SKees Cook /** 916800179c9SKees Cook * safe_hardlink_source - Check for safe hardlink conditions 917800179c9SKees Cook * @inode: the source inode to hardlink from 918800179c9SKees Cook * 919800179c9SKees Cook * Return false if at least one of the following conditions: 920800179c9SKees Cook * - inode is not a regular file 921800179c9SKees Cook * - inode is setuid 922800179c9SKees Cook * - inode is setgid and group-exec 923800179c9SKees Cook * - access failure for read and write 924800179c9SKees Cook * 925800179c9SKees Cook * Otherwise returns true. 926800179c9SKees Cook */ 927800179c9SKees Cook static bool safe_hardlink_source(struct inode *inode) 928800179c9SKees Cook { 929800179c9SKees Cook umode_t mode = inode->i_mode; 930800179c9SKees Cook 931800179c9SKees Cook /* Special files should not get pinned to the filesystem. */ 932800179c9SKees Cook if (!S_ISREG(mode)) 933800179c9SKees Cook return false; 934800179c9SKees Cook 935800179c9SKees Cook /* Setuid files should not get pinned to the filesystem. */ 936800179c9SKees Cook if (mode & S_ISUID) 937800179c9SKees Cook return false; 938800179c9SKees Cook 939800179c9SKees Cook /* Executable setgid files should not get pinned to the filesystem. */ 940800179c9SKees Cook if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) 941800179c9SKees Cook return false; 942800179c9SKees Cook 943800179c9SKees Cook /* Hardlinking to unreadable or unwritable sources is dangerous. */ 944800179c9SKees Cook if (inode_permission(inode, MAY_READ | MAY_WRITE)) 945800179c9SKees Cook return false; 946800179c9SKees Cook 947800179c9SKees Cook return true; 948800179c9SKees Cook } 949800179c9SKees Cook 950800179c9SKees Cook /** 951800179c9SKees Cook * may_linkat - Check permissions for creating a hardlink 952800179c9SKees Cook * @link: the source to hardlink from 953800179c9SKees Cook * 954800179c9SKees Cook * Block hardlink when all of: 955800179c9SKees Cook * - sysctl_protected_hardlinks enabled 956800179c9SKees Cook * - fsuid does not match inode 957800179c9SKees Cook * - hardlink source is unsafe (see safe_hardlink_source() above) 958f2ca3796SDirk Steinmetz * - not CAP_FOWNER in a namespace with the inode owner uid mapped 959800179c9SKees Cook * 960800179c9SKees Cook * Returns 0 if successful, -ve on error. 961800179c9SKees Cook */ 962800179c9SKees Cook static int may_linkat(struct path *link) 963800179c9SKees Cook { 964800179c9SKees Cook struct inode *inode; 965800179c9SKees Cook 966800179c9SKees Cook if (!sysctl_protected_hardlinks) 967800179c9SKees Cook return 0; 968800179c9SKees Cook 969800179c9SKees Cook inode = link->dentry->d_inode; 970800179c9SKees Cook 971800179c9SKees Cook /* Source inode owner (or CAP_FOWNER) can hardlink all they like, 972800179c9SKees Cook * otherwise, it must be a safe source. 973800179c9SKees Cook */ 974f2ca3796SDirk Steinmetz if (inode_owner_or_capable(inode) || safe_hardlink_source(inode)) 975800179c9SKees Cook return 0; 976800179c9SKees Cook 977a51d9eaaSKees Cook audit_log_link_denied("linkat", link); 978800179c9SKees Cook return -EPERM; 979800179c9SKees Cook } 980800179c9SKees Cook 9813b2e7f75SAl Viro static __always_inline 9823b2e7f75SAl Viro const char *get_link(struct nameidata *nd) 9831da177e4SLinus Torvalds { 984ab104923SAl Viro struct saved *last = nd->stack + nd->depth - 1; 9851cf2665bSAl Viro struct dentry *dentry = last->link.dentry; 986237d8b32SAl Viro struct inode *inode = last->inode; 9876d7b5aaeSAl Viro int error; 9880a959df5SAl Viro const char *res; 9891da177e4SLinus Torvalds 9908fa9dd24SNeilBrown if (!(nd->flags & LOOKUP_RCU)) { 9918fa9dd24SNeilBrown touch_atime(&last->link); 9928fa9dd24SNeilBrown cond_resched(); 9938fa9dd24SNeilBrown } else if (atime_needs_update(&last->link, inode)) { 9948fa9dd24SNeilBrown if (unlikely(unlazy_walk(nd, NULL, 0))) 9958fa9dd24SNeilBrown return ERR_PTR(-ECHILD); 9968fa9dd24SNeilBrown touch_atime(&last->link); 9978fa9dd24SNeilBrown } 9988fa9dd24SNeilBrown 999bda0be7aSNeilBrown error = security_inode_follow_link(dentry, inode, 1000bda0be7aSNeilBrown nd->flags & LOOKUP_RCU); 1001bda0be7aSNeilBrown if (unlikely(error)) 10026920a440SAl Viro return ERR_PTR(error); 100336f3b4f6SAl Viro 100486acdca1SAl Viro nd->last_type = LAST_BIND; 1005d4dee48bSAl Viro res = inode->i_link; 1006d4dee48bSAl Viro if (!res) { 10078c1b4566SAl Viro if (nd->flags & LOOKUP_RCU) { 10088c1b4566SAl Viro if (unlikely(unlazy_walk(nd, NULL, 0))) 10098c1b4566SAl Viro return ERR_PTR(-ECHILD); 10108c1b4566SAl Viro } 10116e77137bSAl Viro res = inode->i_op->follow_link(dentry, &last->cookie); 1012fab51e8aSAl Viro if (IS_ERR_OR_NULL(res)) { 10136920a440SAl Viro last->cookie = NULL; 1014fab51e8aSAl Viro return res; 10150a959df5SAl Viro } 1016fab51e8aSAl Viro } 1017fab51e8aSAl Viro if (*res == '/') { 10188c1b4566SAl Viro if (nd->flags & LOOKUP_RCU) { 10198f47a016SAl Viro struct dentry *d; 10208f47a016SAl Viro if (!nd->root.mnt) 10218f47a016SAl Viro set_root_rcu(nd); 10228f47a016SAl Viro nd->path = nd->root; 10238f47a016SAl Viro d = nd->path.dentry; 10248f47a016SAl Viro nd->inode = d->d_inode; 10258f47a016SAl Viro nd->seq = nd->root_seq; 10268f47a016SAl Viro if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq))) 10278c1b4566SAl Viro return ERR_PTR(-ECHILD); 10288f47a016SAl Viro } else { 1029fab51e8aSAl Viro if (!nd->root.mnt) 1030fab51e8aSAl Viro set_root(nd); 1031fab51e8aSAl Viro path_put(&nd->path); 1032fab51e8aSAl Viro nd->path = nd->root; 1033fab51e8aSAl Viro path_get(&nd->root); 1034fab51e8aSAl Viro nd->inode = nd->path.dentry->d_inode; 10358f47a016SAl Viro } 1036fab51e8aSAl Viro nd->flags |= LOOKUP_JUMPED; 1037fab51e8aSAl Viro while (unlikely(*++res == '/')) 1038fab51e8aSAl Viro ; 1039fab51e8aSAl Viro } 1040fab51e8aSAl Viro if (!*res) 1041fab51e8aSAl Viro res = NULL; 10420a959df5SAl Viro return res; 10430a959df5SAl Viro } 10446d7b5aaeSAl Viro 1045f015f126SDavid Howells /* 1046f015f126SDavid Howells * follow_up - Find the mountpoint of path's vfsmount 1047f015f126SDavid Howells * 1048f015f126SDavid Howells * Given a path, find the mountpoint of its source file system. 1049f015f126SDavid Howells * Replace @path with the path of the mountpoint in the parent mount. 1050f015f126SDavid Howells * Up is towards /. 1051f015f126SDavid Howells * 1052f015f126SDavid Howells * Return 1 if we went up a level and 0 if we were already at the 1053f015f126SDavid Howells * root. 1054f015f126SDavid Howells */ 1055bab77ebfSAl Viro int follow_up(struct path *path) 10561da177e4SLinus Torvalds { 10570714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 10580714a533SAl Viro struct mount *parent; 10591da177e4SLinus Torvalds struct dentry *mountpoint; 106099b7db7bSNick Piggin 106148a066e7SAl Viro read_seqlock_excl(&mount_lock); 10620714a533SAl Viro parent = mnt->mnt_parent; 10633c0a6163SAl Viro if (parent == mnt) { 106448a066e7SAl Viro read_sequnlock_excl(&mount_lock); 10651da177e4SLinus Torvalds return 0; 10661da177e4SLinus Torvalds } 10670714a533SAl Viro mntget(&parent->mnt); 1068a73324daSAl Viro mountpoint = dget(mnt->mnt_mountpoint); 106948a066e7SAl Viro read_sequnlock_excl(&mount_lock); 1070bab77ebfSAl Viro dput(path->dentry); 1071bab77ebfSAl Viro path->dentry = mountpoint; 1072bab77ebfSAl Viro mntput(path->mnt); 10730714a533SAl Viro path->mnt = &parent->mnt; 10741da177e4SLinus Torvalds return 1; 10751da177e4SLinus Torvalds } 10764d359507SAl Viro EXPORT_SYMBOL(follow_up); 10771da177e4SLinus Torvalds 1078b5c84bf6SNick Piggin /* 10799875cf80SDavid Howells * Perform an automount 10809875cf80SDavid Howells * - return -EISDIR to tell follow_managed() to stop and return the path we 10819875cf80SDavid Howells * were called with. 10821da177e4SLinus Torvalds */ 1083756daf26SNeilBrown static int follow_automount(struct path *path, struct nameidata *nd, 10849875cf80SDavid Howells bool *need_mntput) 108531e6b01fSNick Piggin { 10869875cf80SDavid Howells struct vfsmount *mnt; 1087ea5b778aSDavid Howells int err; 10889875cf80SDavid Howells 10899875cf80SDavid Howells if (!path->dentry->d_op || !path->dentry->d_op->d_automount) 10909875cf80SDavid Howells return -EREMOTE; 10919875cf80SDavid Howells 10920ec26fd0SMiklos Szeredi /* We don't want to mount if someone's just doing a stat - 10930ec26fd0SMiklos Szeredi * unless they're stat'ing a directory and appended a '/' to 10940ec26fd0SMiklos Szeredi * the name. 10950ec26fd0SMiklos Szeredi * 10960ec26fd0SMiklos Szeredi * We do, however, want to mount if someone wants to open or 10975a30d8a2SDavid Howells * create a file of any type under the mountpoint, wants to 10980ec26fd0SMiklos Szeredi * traverse through the mountpoint or wants to open the 10990ec26fd0SMiklos Szeredi * mounted directory. Also, autofs may mark negative dentries 11000ec26fd0SMiklos Szeredi * as being automount points. These will need the attentions 11010ec26fd0SMiklos Szeredi * of the daemon to instantiate them before they can be used. 11025a30d8a2SDavid Howells */ 1103756daf26SNeilBrown if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | 1104d94c177bSLinus Torvalds LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && 11055a30d8a2SDavid Howells path->dentry->d_inode) 11069875cf80SDavid Howells return -EISDIR; 11070ec26fd0SMiklos Szeredi 1108756daf26SNeilBrown nd->total_link_count++; 1109756daf26SNeilBrown if (nd->total_link_count >= 40) 11109875cf80SDavid Howells return -ELOOP; 11119875cf80SDavid Howells 11129875cf80SDavid Howells mnt = path->dentry->d_op->d_automount(path); 11139875cf80SDavid Howells if (IS_ERR(mnt)) { 11149875cf80SDavid Howells /* 11159875cf80SDavid Howells * The filesystem is allowed to return -EISDIR here to indicate 11169875cf80SDavid Howells * it doesn't want to automount. For instance, autofs would do 11179875cf80SDavid Howells * this so that its userspace daemon can mount on this dentry. 11189875cf80SDavid Howells * 11199875cf80SDavid Howells * However, we can only permit this if it's a terminal point in 11209875cf80SDavid Howells * the path being looked up; if it wasn't then the remainder of 11219875cf80SDavid Howells * the path is inaccessible and we should say so. 11229875cf80SDavid Howells */ 1123756daf26SNeilBrown if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT)) 11249875cf80SDavid Howells return -EREMOTE; 11259875cf80SDavid Howells return PTR_ERR(mnt); 112631e6b01fSNick Piggin } 1127ea5b778aSDavid Howells 11289875cf80SDavid Howells if (!mnt) /* mount collision */ 11299875cf80SDavid Howells return 0; 11309875cf80SDavid Howells 11318aef1884SAl Viro if (!*need_mntput) { 11328aef1884SAl Viro /* lock_mount() may release path->mnt on error */ 11338aef1884SAl Viro mntget(path->mnt); 11348aef1884SAl Viro *need_mntput = true; 11358aef1884SAl Viro } 113619a167afSAl Viro err = finish_automount(mnt, path); 1137ea5b778aSDavid Howells 1138ea5b778aSDavid Howells switch (err) { 1139ea5b778aSDavid Howells case -EBUSY: 1140ea5b778aSDavid Howells /* Someone else made a mount here whilst we were busy */ 114119a167afSAl Viro return 0; 1142ea5b778aSDavid Howells case 0: 11438aef1884SAl Viro path_put(path); 11449875cf80SDavid Howells path->mnt = mnt; 11459875cf80SDavid Howells path->dentry = dget(mnt->mnt_root); 11469875cf80SDavid Howells return 0; 114719a167afSAl Viro default: 114819a167afSAl Viro return err; 11499875cf80SDavid Howells } 115019a167afSAl Viro 1151ea5b778aSDavid Howells } 11529875cf80SDavid Howells 11539875cf80SDavid Howells /* 11549875cf80SDavid Howells * Handle a dentry that is managed in some way. 1155cc53ce53SDavid Howells * - Flagged for transit management (autofs) 11569875cf80SDavid Howells * - Flagged as mountpoint 11579875cf80SDavid Howells * - Flagged as automount point 11589875cf80SDavid Howells * 11599875cf80SDavid Howells * This may only be called in refwalk mode. 11609875cf80SDavid Howells * 11619875cf80SDavid Howells * Serialization is taken care of in namespace.c 11629875cf80SDavid Howells */ 1163756daf26SNeilBrown static int follow_managed(struct path *path, struct nameidata *nd) 11649875cf80SDavid Howells { 11658aef1884SAl Viro struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */ 11669875cf80SDavid Howells unsigned managed; 11679875cf80SDavid Howells bool need_mntput = false; 11688aef1884SAl Viro int ret = 0; 11699875cf80SDavid Howells 11709875cf80SDavid Howells /* Given that we're not holding a lock here, we retain the value in a 11719875cf80SDavid Howells * local variable for each dentry as we look at it so that we don't see 11729875cf80SDavid Howells * the components of that value change under us */ 11739875cf80SDavid Howells while (managed = ACCESS_ONCE(path->dentry->d_flags), 11749875cf80SDavid Howells managed &= DCACHE_MANAGED_DENTRY, 11759875cf80SDavid Howells unlikely(managed != 0)) { 1176cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1177cc53ce53SDavid Howells * being held. */ 1178cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 1179cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1180cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 11811aed3e42SAl Viro ret = path->dentry->d_op->d_manage(path->dentry, false); 1182cc53ce53SDavid Howells if (ret < 0) 11838aef1884SAl Viro break; 1184cc53ce53SDavid Howells } 1185cc53ce53SDavid Howells 11869875cf80SDavid Howells /* Transit to a mounted filesystem. */ 11879875cf80SDavid Howells if (managed & DCACHE_MOUNTED) { 11889875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 11899875cf80SDavid Howells if (mounted) { 11909875cf80SDavid Howells dput(path->dentry); 11919875cf80SDavid Howells if (need_mntput) 1192463ffb2eSAl Viro mntput(path->mnt); 1193463ffb2eSAl Viro path->mnt = mounted; 1194463ffb2eSAl Viro path->dentry = dget(mounted->mnt_root); 11959875cf80SDavid Howells need_mntput = true; 11969875cf80SDavid Howells continue; 1197463ffb2eSAl Viro } 1198463ffb2eSAl Viro 11999875cf80SDavid Howells /* Something is mounted on this dentry in another 12009875cf80SDavid Howells * namespace and/or whatever was mounted there in this 120148a066e7SAl Viro * namespace got unmounted before lookup_mnt() could 120248a066e7SAl Viro * get it */ 12031da177e4SLinus Torvalds } 12049875cf80SDavid Howells 12059875cf80SDavid Howells /* Handle an automount point */ 12069875cf80SDavid Howells if (managed & DCACHE_NEED_AUTOMOUNT) { 1207756daf26SNeilBrown ret = follow_automount(path, nd, &need_mntput); 12089875cf80SDavid Howells if (ret < 0) 12098aef1884SAl Viro break; 12109875cf80SDavid Howells continue; 12119875cf80SDavid Howells } 12129875cf80SDavid Howells 12139875cf80SDavid Howells /* We didn't change the current path point */ 12149875cf80SDavid Howells break; 12159875cf80SDavid Howells } 12168aef1884SAl Viro 12178aef1884SAl Viro if (need_mntput && path->mnt == mnt) 12188aef1884SAl Viro mntput(path->mnt); 12198aef1884SAl Viro if (ret == -EISDIR) 12208aef1884SAl Viro ret = 0; 12218402752eSAl Viro if (need_mntput) 12228402752eSAl Viro nd->flags |= LOOKUP_JUMPED; 12238402752eSAl Viro if (unlikely(ret < 0)) 12248402752eSAl Viro path_put_conditional(path, nd); 12258402752eSAl Viro return ret; 12261da177e4SLinus Torvalds } 12271da177e4SLinus Torvalds 1228cc53ce53SDavid Howells int follow_down_one(struct path *path) 12291da177e4SLinus Torvalds { 12301da177e4SLinus Torvalds struct vfsmount *mounted; 12311da177e4SLinus Torvalds 12321c755af4SAl Viro mounted = lookup_mnt(path); 12331da177e4SLinus Torvalds if (mounted) { 12349393bd07SAl Viro dput(path->dentry); 12359393bd07SAl Viro mntput(path->mnt); 12369393bd07SAl Viro path->mnt = mounted; 12379393bd07SAl Viro path->dentry = dget(mounted->mnt_root); 12381da177e4SLinus Torvalds return 1; 12391da177e4SLinus Torvalds } 12401da177e4SLinus Torvalds return 0; 12411da177e4SLinus Torvalds } 12424d359507SAl Viro EXPORT_SYMBOL(follow_down_one); 12431da177e4SLinus Torvalds 1244b8faf035SNeilBrown static inline int managed_dentry_rcu(struct dentry *dentry) 124562a7375eSIan Kent { 1246b8faf035SNeilBrown return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ? 1247b8faf035SNeilBrown dentry->d_op->d_manage(dentry, true) : 0; 124862a7375eSIan Kent } 124962a7375eSIan Kent 12509875cf80SDavid Howells /* 1251287548e4SAl Viro * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if 1252287548e4SAl Viro * we meet a managed dentry that would need blocking. 12539875cf80SDavid Howells */ 12549875cf80SDavid Howells static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, 1255254cf582SAl Viro struct inode **inode, unsigned *seqp) 12569875cf80SDavid Howells { 125762a7375eSIan Kent for (;;) { 1258c7105365SAl Viro struct mount *mounted; 125962a7375eSIan Kent /* 126062a7375eSIan Kent * Don't forget we might have a non-mountpoint managed dentry 126162a7375eSIan Kent * that wants to block transit. 126262a7375eSIan Kent */ 1263b8faf035SNeilBrown switch (managed_dentry_rcu(path->dentry)) { 1264b8faf035SNeilBrown case -ECHILD: 1265b8faf035SNeilBrown default: 1266ab90911fSDavid Howells return false; 1267b8faf035SNeilBrown case -EISDIR: 1268b8faf035SNeilBrown return true; 1269b8faf035SNeilBrown case 0: 1270b8faf035SNeilBrown break; 1271b8faf035SNeilBrown } 127262a7375eSIan Kent 127362a7375eSIan Kent if (!d_mountpoint(path->dentry)) 1274b8faf035SNeilBrown return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); 127562a7375eSIan Kent 1276474279dcSAl Viro mounted = __lookup_mnt(path->mnt, path->dentry); 12779875cf80SDavid Howells if (!mounted) 12789875cf80SDavid Howells break; 1279c7105365SAl Viro path->mnt = &mounted->mnt; 1280c7105365SAl Viro path->dentry = mounted->mnt.mnt_root; 1281a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 1282254cf582SAl Viro *seqp = read_seqcount_begin(&path->dentry->d_seq); 128359430262SLinus Torvalds /* 128459430262SLinus Torvalds * Update the inode too. We don't need to re-check the 128559430262SLinus Torvalds * dentry sequence number here after this d_inode read, 128659430262SLinus Torvalds * because a mount-point is always pinned. 128759430262SLinus Torvalds */ 128859430262SLinus Torvalds *inode = path->dentry->d_inode; 12899875cf80SDavid Howells } 1290f5be3e29SAl Viro return !read_seqretry(&mount_lock, nd->m_seq) && 1291b8faf035SNeilBrown !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); 1292287548e4SAl Viro } 1293287548e4SAl Viro 129431e6b01fSNick Piggin static int follow_dotdot_rcu(struct nameidata *nd) 129531e6b01fSNick Piggin { 12964023bfc9SAl Viro struct inode *inode = nd->inode; 12977bd88377SAl Viro if (!nd->root.mnt) 129831e6b01fSNick Piggin set_root_rcu(nd); 129931e6b01fSNick Piggin 130031e6b01fSNick Piggin while (1) { 1301aed434adSAl Viro if (path_equal(&nd->path, &nd->root)) 130231e6b01fSNick Piggin break; 130331e6b01fSNick Piggin if (nd->path.dentry != nd->path.mnt->mnt_root) { 130431e6b01fSNick Piggin struct dentry *old = nd->path.dentry; 130531e6b01fSNick Piggin struct dentry *parent = old->d_parent; 130631e6b01fSNick Piggin unsigned seq; 130731e6b01fSNick Piggin 13084023bfc9SAl Viro inode = parent->d_inode; 130931e6b01fSNick Piggin seq = read_seqcount_begin(&parent->d_seq); 1310aed434adSAl Viro if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq))) 1311aed434adSAl Viro return -ECHILD; 131231e6b01fSNick Piggin nd->path.dentry = parent; 131331e6b01fSNick Piggin nd->seq = seq; 1314397d425dSEric W. Biederman if (unlikely(!path_connected(&nd->path))) 1315397d425dSEric W. Biederman return -ENOENT; 131631e6b01fSNick Piggin break; 1317aed434adSAl Viro } else { 1318aed434adSAl Viro struct mount *mnt = real_mount(nd->path.mnt); 1319aed434adSAl Viro struct mount *mparent = mnt->mnt_parent; 1320aed434adSAl Viro struct dentry *mountpoint = mnt->mnt_mountpoint; 1321aed434adSAl Viro struct inode *inode2 = mountpoint->d_inode; 1322aed434adSAl Viro unsigned seq = read_seqcount_begin(&mountpoint->d_seq); 1323aed434adSAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1324aed434adSAl Viro return -ECHILD; 1325aed434adSAl Viro if (&mparent->mnt == nd->path.mnt) 132631e6b01fSNick Piggin break; 1327aed434adSAl Viro /* we know that mountpoint was pinned */ 1328aed434adSAl Viro nd->path.dentry = mountpoint; 1329aed434adSAl Viro nd->path.mnt = &mparent->mnt; 1330aed434adSAl Viro inode = inode2; 1331aed434adSAl Viro nd->seq = seq; 133231e6b01fSNick Piggin } 1333aed434adSAl Viro } 1334aed434adSAl Viro while (unlikely(d_mountpoint(nd->path.dentry))) { 1335b37199e6SAl Viro struct mount *mounted; 1336b37199e6SAl Viro mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry); 1337aed434adSAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1338aed434adSAl Viro return -ECHILD; 1339b37199e6SAl Viro if (!mounted) 1340b37199e6SAl Viro break; 1341b37199e6SAl Viro nd->path.mnt = &mounted->mnt; 1342b37199e6SAl Viro nd->path.dentry = mounted->mnt.mnt_root; 13434023bfc9SAl Viro inode = nd->path.dentry->d_inode; 1344b37199e6SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 1345b37199e6SAl Viro } 13464023bfc9SAl Viro nd->inode = inode; 134731e6b01fSNick Piggin return 0; 134831e6b01fSNick Piggin } 134931e6b01fSNick Piggin 13509875cf80SDavid Howells /* 1351cc53ce53SDavid Howells * Follow down to the covering mount currently visible to userspace. At each 1352cc53ce53SDavid Howells * point, the filesystem owning that dentry may be queried as to whether the 1353cc53ce53SDavid Howells * caller is permitted to proceed or not. 1354cc53ce53SDavid Howells */ 13557cc90cc3SAl Viro int follow_down(struct path *path) 1356cc53ce53SDavid Howells { 1357cc53ce53SDavid Howells unsigned managed; 1358cc53ce53SDavid Howells int ret; 1359cc53ce53SDavid Howells 1360cc53ce53SDavid Howells while (managed = ACCESS_ONCE(path->dentry->d_flags), 1361cc53ce53SDavid Howells unlikely(managed & DCACHE_MANAGED_DENTRY)) { 1362cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1363cc53ce53SDavid Howells * being held. 1364cc53ce53SDavid Howells * 1365cc53ce53SDavid Howells * We indicate to the filesystem if someone is trying to mount 1366cc53ce53SDavid Howells * something here. This gives autofs the chance to deny anyone 1367cc53ce53SDavid Howells * other than its daemon the right to mount on its 1368cc53ce53SDavid Howells * superstructure. 1369cc53ce53SDavid Howells * 1370cc53ce53SDavid Howells * The filesystem may sleep at this point. 1371cc53ce53SDavid Howells */ 1372cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 1373cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1374cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 1375ab90911fSDavid Howells ret = path->dentry->d_op->d_manage( 13761aed3e42SAl Viro path->dentry, false); 1377cc53ce53SDavid Howells if (ret < 0) 1378cc53ce53SDavid Howells return ret == -EISDIR ? 0 : ret; 1379cc53ce53SDavid Howells } 1380cc53ce53SDavid Howells 1381cc53ce53SDavid Howells /* Transit to a mounted filesystem. */ 1382cc53ce53SDavid Howells if (managed & DCACHE_MOUNTED) { 1383cc53ce53SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 1384cc53ce53SDavid Howells if (!mounted) 1385cc53ce53SDavid Howells break; 1386cc53ce53SDavid Howells dput(path->dentry); 1387cc53ce53SDavid Howells mntput(path->mnt); 1388cc53ce53SDavid Howells path->mnt = mounted; 1389cc53ce53SDavid Howells path->dentry = dget(mounted->mnt_root); 1390cc53ce53SDavid Howells continue; 1391cc53ce53SDavid Howells } 1392cc53ce53SDavid Howells 1393cc53ce53SDavid Howells /* Don't handle automount points here */ 1394cc53ce53SDavid Howells break; 1395cc53ce53SDavid Howells } 1396cc53ce53SDavid Howells return 0; 1397cc53ce53SDavid Howells } 13984d359507SAl Viro EXPORT_SYMBOL(follow_down); 1399cc53ce53SDavid Howells 1400cc53ce53SDavid Howells /* 14019875cf80SDavid Howells * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() 14029875cf80SDavid Howells */ 14039875cf80SDavid Howells static void follow_mount(struct path *path) 14049875cf80SDavid Howells { 14059875cf80SDavid Howells while (d_mountpoint(path->dentry)) { 14069875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 14079875cf80SDavid Howells if (!mounted) 14089875cf80SDavid Howells break; 14099875cf80SDavid Howells dput(path->dentry); 14109875cf80SDavid Howells mntput(path->mnt); 14119875cf80SDavid Howells path->mnt = mounted; 14129875cf80SDavid Howells path->dentry = dget(mounted->mnt_root); 14139875cf80SDavid Howells } 14149875cf80SDavid Howells } 14159875cf80SDavid Howells 1416397d425dSEric W. Biederman static int follow_dotdot(struct nameidata *nd) 14171da177e4SLinus Torvalds { 14187bd88377SAl Viro if (!nd->root.mnt) 14192a737871SAl Viro set_root(nd); 1420e518ddb7SAndreas Mohr 14211da177e4SLinus Torvalds while(1) { 14224ac91378SJan Blunck struct dentry *old = nd->path.dentry; 14231da177e4SLinus Torvalds 14242a737871SAl Viro if (nd->path.dentry == nd->root.dentry && 14252a737871SAl Viro nd->path.mnt == nd->root.mnt) { 14261da177e4SLinus Torvalds break; 14271da177e4SLinus Torvalds } 14284ac91378SJan Blunck if (nd->path.dentry != nd->path.mnt->mnt_root) { 14293088dd70SAl Viro /* rare case of legitimate dget_parent()... */ 14303088dd70SAl Viro nd->path.dentry = dget_parent(nd->path.dentry); 14311da177e4SLinus Torvalds dput(old); 1432397d425dSEric W. Biederman if (unlikely(!path_connected(&nd->path))) 1433397d425dSEric W. Biederman return -ENOENT; 14341da177e4SLinus Torvalds break; 14351da177e4SLinus Torvalds } 14363088dd70SAl Viro if (!follow_up(&nd->path)) 14371da177e4SLinus Torvalds break; 14381da177e4SLinus Torvalds } 143979ed0226SAl Viro follow_mount(&nd->path); 144031e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 1441397d425dSEric W. Biederman return 0; 14421da177e4SLinus Torvalds } 14431da177e4SLinus Torvalds 14441da177e4SLinus Torvalds /* 1445bad61189SMiklos Szeredi * This looks up the name in dcache, possibly revalidates the old dentry and 1446bad61189SMiklos Szeredi * allocates a new one if not found or not valid. In the need_lookup argument 1447bad61189SMiklos Szeredi * returns whether i_op->lookup is necessary. 1448bad61189SMiklos Szeredi * 1449bad61189SMiklos Szeredi * dir->d_inode->i_mutex must be held 1450baa03890SNick Piggin */ 1451bad61189SMiklos Szeredi static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir, 1452201f956eSAl Viro unsigned int flags, bool *need_lookup) 1453baa03890SNick Piggin { 1454baa03890SNick Piggin struct dentry *dentry; 1455bad61189SMiklos Szeredi int error; 1456baa03890SNick Piggin 1457bad61189SMiklos Szeredi *need_lookup = false; 1458bad61189SMiklos Szeredi dentry = d_lookup(dir, name); 1459bad61189SMiklos Szeredi if (dentry) { 146039e3c955SJeff Layton if (dentry->d_flags & DCACHE_OP_REVALIDATE) { 1461201f956eSAl Viro error = d_revalidate(dentry, flags); 1462bad61189SMiklos Szeredi if (unlikely(error <= 0)) { 1463bad61189SMiklos Szeredi if (error < 0) { 1464bad61189SMiklos Szeredi dput(dentry); 1465bad61189SMiklos Szeredi return ERR_PTR(error); 14665542aa2fSEric W. Biederman } else { 14675542aa2fSEric W. Biederman d_invalidate(dentry); 1468bad61189SMiklos Szeredi dput(dentry); 1469bad61189SMiklos Szeredi dentry = NULL; 1470bad61189SMiklos Szeredi } 1471bad61189SMiklos Szeredi } 1472bad61189SMiklos Szeredi } 1473bad61189SMiklos Szeredi } 1474baa03890SNick Piggin 1475bad61189SMiklos Szeredi if (!dentry) { 1476bad61189SMiklos Szeredi dentry = d_alloc(dir, name); 1477baa03890SNick Piggin if (unlikely(!dentry)) 1478baa03890SNick Piggin return ERR_PTR(-ENOMEM); 1479baa03890SNick Piggin 1480bad61189SMiklos Szeredi *need_lookup = true; 1481baa03890SNick Piggin } 1482baa03890SNick Piggin return dentry; 1483baa03890SNick Piggin } 1484baa03890SNick Piggin 1485baa03890SNick Piggin /* 148613a2c3beSJ. Bruce Fields * Call i_op->lookup on the dentry. The dentry must be negative and 148713a2c3beSJ. Bruce Fields * unhashed. 1488bad61189SMiklos Szeredi * 1489bad61189SMiklos Szeredi * dir->d_inode->i_mutex must be held 149044396f4bSJosef Bacik */ 1491bad61189SMiklos Szeredi static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry, 149272bd866aSAl Viro unsigned int flags) 149344396f4bSJosef Bacik { 149444396f4bSJosef Bacik struct dentry *old; 149544396f4bSJosef Bacik 149644396f4bSJosef Bacik /* Don't create child dentry for a dead directory. */ 1497bad61189SMiklos Szeredi if (unlikely(IS_DEADDIR(dir))) { 1498e188dc02SMiklos Szeredi dput(dentry); 149944396f4bSJosef Bacik return ERR_PTR(-ENOENT); 1500e188dc02SMiklos Szeredi } 150144396f4bSJosef Bacik 150272bd866aSAl Viro old = dir->i_op->lookup(dir, dentry, flags); 150344396f4bSJosef Bacik if (unlikely(old)) { 150444396f4bSJosef Bacik dput(dentry); 150544396f4bSJosef Bacik dentry = old; 150644396f4bSJosef Bacik } 150744396f4bSJosef Bacik return dentry; 150844396f4bSJosef Bacik } 150944396f4bSJosef Bacik 1510a3255546SAl Viro static struct dentry *__lookup_hash(struct qstr *name, 151172bd866aSAl Viro struct dentry *base, unsigned int flags) 1512a3255546SAl Viro { 1513bad61189SMiklos Szeredi bool need_lookup; 1514a3255546SAl Viro struct dentry *dentry; 1515a3255546SAl Viro 151672bd866aSAl Viro dentry = lookup_dcache(name, base, flags, &need_lookup); 1517bad61189SMiklos Szeredi if (!need_lookup) 1518a3255546SAl Viro return dentry; 1519bad61189SMiklos Szeredi 152072bd866aSAl Viro return lookup_real(base->d_inode, dentry, flags); 1521a3255546SAl Viro } 1522a3255546SAl Viro 152344396f4bSJosef Bacik /* 15241da177e4SLinus Torvalds * It's more convoluted than I'd like it to be, but... it's still fairly 15251da177e4SLinus Torvalds * small and for now I'd prefer to have fast path as straight as possible. 15261da177e4SLinus Torvalds * It _is_ time-critical. 15271da177e4SLinus Torvalds */ 1528e97cdc87SAl Viro static int lookup_fast(struct nameidata *nd, 1529254cf582SAl Viro struct path *path, struct inode **inode, 1530254cf582SAl Viro unsigned *seqp) 15311da177e4SLinus Torvalds { 15324ac91378SJan Blunck struct vfsmount *mnt = nd->path.mnt; 153331e6b01fSNick Piggin struct dentry *dentry, *parent = nd->path.dentry; 15345a18fff2SAl Viro int need_reval = 1; 15355a18fff2SAl Viro int status = 1; 15369875cf80SDavid Howells int err; 15379875cf80SDavid Howells 15383cac260aSAl Viro /* 1539b04f784eSNick Piggin * Rename seqlock is not required here because in the off chance 1540b04f784eSNick Piggin * of a false negative due to a concurrent rename, we're going to 1541b04f784eSNick Piggin * do the non-racy lookup, below. 1542b04f784eSNick Piggin */ 154331e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) { 154431e6b01fSNick Piggin unsigned seq; 1545766c4cbfSAl Viro bool negative; 1546da53be12SLinus Torvalds dentry = __d_lookup_rcu(parent, &nd->last, &seq); 15475a18fff2SAl Viro if (!dentry) 15485a18fff2SAl Viro goto unlazy; 15495a18fff2SAl Viro 155012f8ad4bSLinus Torvalds /* 155112f8ad4bSLinus Torvalds * This sequence count validates that the inode matches 155212f8ad4bSLinus Torvalds * the dentry name information from lookup. 155312f8ad4bSLinus Torvalds */ 155463afdfc7SDavid Howells *inode = d_backing_inode(dentry); 1555766c4cbfSAl Viro negative = d_is_negative(dentry); 155612f8ad4bSLinus Torvalds if (read_seqcount_retry(&dentry->d_seq, seq)) 155712f8ad4bSLinus Torvalds return -ECHILD; 155812f8ad4bSLinus Torvalds 155912f8ad4bSLinus Torvalds /* 156012f8ad4bSLinus Torvalds * This sequence count validates that the parent had no 156112f8ad4bSLinus Torvalds * changes while we did the lookup of the dentry above. 156212f8ad4bSLinus Torvalds * 156312f8ad4bSLinus Torvalds * The memory barrier in read_seqcount_begin of child is 156412f8ad4bSLinus Torvalds * enough, we can use __read_seqcount_retry here. 156512f8ad4bSLinus Torvalds */ 156631e6b01fSNick Piggin if (__read_seqcount_retry(&parent->d_seq, nd->seq)) 156731e6b01fSNick Piggin return -ECHILD; 15685a18fff2SAl Viro 1569254cf582SAl Viro *seqp = seq; 157024643087SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) { 15714ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 15725a18fff2SAl Viro if (unlikely(status <= 0)) { 15735a18fff2SAl Viro if (status != -ECHILD) 15745a18fff2SAl Viro need_reval = 0; 15755a18fff2SAl Viro goto unlazy; 15765a18fff2SAl Viro } 157724643087SAl Viro } 1578daf3761cSTrond Myklebust /* 1579daf3761cSTrond Myklebust * Note: do negative dentry check after revalidation in 1580daf3761cSTrond Myklebust * case that drops it. 1581daf3761cSTrond Myklebust */ 1582daf3761cSTrond Myklebust if (negative) 1583daf3761cSTrond Myklebust return -ENOENT; 158431e6b01fSNick Piggin path->mnt = mnt; 158531e6b01fSNick Piggin path->dentry = dentry; 1586254cf582SAl Viro if (likely(__follow_mount_rcu(nd, path, inode, seqp))) 15879875cf80SDavid Howells return 0; 15885a18fff2SAl Viro unlazy: 1589254cf582SAl Viro if (unlazy_walk(nd, dentry, seq)) 15905a18fff2SAl Viro return -ECHILD; 15915a18fff2SAl Viro } else { 1592e97cdc87SAl Viro dentry = __d_lookup(parent, &nd->last); 159324643087SAl Viro } 15945a18fff2SAl Viro 159581e6f520SAl Viro if (unlikely(!dentry)) 159681e6f520SAl Viro goto need_lookup; 15975a18fff2SAl Viro 15985a18fff2SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval) 15994ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 16005a18fff2SAl Viro if (unlikely(status <= 0)) { 16015a18fff2SAl Viro if (status < 0) { 16025a18fff2SAl Viro dput(dentry); 16035a18fff2SAl Viro return status; 16045a18fff2SAl Viro } 16055542aa2fSEric W. Biederman d_invalidate(dentry); 16065a18fff2SAl Viro dput(dentry); 160781e6f520SAl Viro goto need_lookup; 16085a18fff2SAl Viro } 1609697f514dSMiklos Szeredi 1610766c4cbfSAl Viro if (unlikely(d_is_negative(dentry))) { 1611766c4cbfSAl Viro dput(dentry); 1612766c4cbfSAl Viro return -ENOENT; 1613766c4cbfSAl Viro } 16141da177e4SLinus Torvalds path->mnt = mnt; 16151da177e4SLinus Torvalds path->dentry = dentry; 1616756daf26SNeilBrown err = follow_managed(path, nd); 16178402752eSAl Viro if (likely(!err)) 161863afdfc7SDavid Howells *inode = d_backing_inode(path->dentry); 16198402752eSAl Viro return err; 162081e6f520SAl Viro 162181e6f520SAl Viro need_lookup: 1622697f514dSMiklos Szeredi return 1; 1623697f514dSMiklos Szeredi } 1624697f514dSMiklos Szeredi 1625697f514dSMiklos Szeredi /* Fast lookup failed, do it the slow way */ 1626cc2a5271SAl Viro static int lookup_slow(struct nameidata *nd, struct path *path) 1627697f514dSMiklos Szeredi { 1628697f514dSMiklos Szeredi struct dentry *dentry, *parent; 1629697f514dSMiklos Szeredi 1630697f514dSMiklos Szeredi parent = nd->path.dentry; 163181e6f520SAl Viro BUG_ON(nd->inode != parent->d_inode); 163281e6f520SAl Viro 163381e6f520SAl Viro mutex_lock(&parent->d_inode->i_mutex); 1634cc2a5271SAl Viro dentry = __lookup_hash(&nd->last, parent, nd->flags); 163581e6f520SAl Viro mutex_unlock(&parent->d_inode->i_mutex); 163681e6f520SAl Viro if (IS_ERR(dentry)) 163781e6f520SAl Viro return PTR_ERR(dentry); 1638697f514dSMiklos Szeredi path->mnt = nd->path.mnt; 1639697f514dSMiklos Szeredi path->dentry = dentry; 16408402752eSAl Viro return follow_managed(path, nd); 16411da177e4SLinus Torvalds } 16421da177e4SLinus Torvalds 164352094c8aSAl Viro static inline int may_lookup(struct nameidata *nd) 164452094c8aSAl Viro { 164552094c8aSAl Viro if (nd->flags & LOOKUP_RCU) { 16464ad5abb3SAl Viro int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); 164752094c8aSAl Viro if (err != -ECHILD) 164852094c8aSAl Viro return err; 16496e9918b7SAl Viro if (unlazy_walk(nd, NULL, 0)) 165052094c8aSAl Viro return -ECHILD; 165152094c8aSAl Viro } 16524ad5abb3SAl Viro return inode_permission(nd->inode, MAY_EXEC); 165352094c8aSAl Viro } 165452094c8aSAl Viro 16559856fa1bSAl Viro static inline int handle_dots(struct nameidata *nd, int type) 16569856fa1bSAl Viro { 16579856fa1bSAl Viro if (type == LAST_DOTDOT) { 16589856fa1bSAl Viro if (nd->flags & LOOKUP_RCU) { 165970291aecSAl Viro return follow_dotdot_rcu(nd); 16609856fa1bSAl Viro } else 1661397d425dSEric W. Biederman return follow_dotdot(nd); 16629856fa1bSAl Viro } 16639856fa1bSAl Viro return 0; 16649856fa1bSAl Viro } 16659856fa1bSAl Viro 1666181548c0SAl Viro static int pick_link(struct nameidata *nd, struct path *link, 1667181548c0SAl Viro struct inode *inode, unsigned seq) 1668d63ff28fSAl Viro { 1669626de996SAl Viro int error; 16701cf2665bSAl Viro struct saved *last; 1671756daf26SNeilBrown if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) { 1672626de996SAl Viro path_to_nameidata(link, nd); 1673626de996SAl Viro return -ELOOP; 1674626de996SAl Viro } 1675bc40aee0SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1676cd179f44SAl Viro if (link->mnt == nd->path.mnt) 1677cd179f44SAl Viro mntget(link->mnt); 16787973387aSAl Viro } 1679626de996SAl Viro error = nd_alloc_stack(nd); 1680626de996SAl Viro if (unlikely(error)) { 1681bc40aee0SAl Viro if (error == -ECHILD) { 1682bc40aee0SAl Viro if (unlikely(unlazy_link(nd, link, seq))) 1683bc40aee0SAl Viro return -ECHILD; 1684bc40aee0SAl Viro error = nd_alloc_stack(nd); 1685bc40aee0SAl Viro } 1686bc40aee0SAl Viro if (error) { 1687cd179f44SAl Viro path_put(link); 1688626de996SAl Viro return error; 1689626de996SAl Viro } 1690bc40aee0SAl Viro } 1691626de996SAl Viro 1692ab104923SAl Viro last = nd->stack + nd->depth++; 16931cf2665bSAl Viro last->link = *link; 1694ab104923SAl Viro last->cookie = NULL; 1695181548c0SAl Viro last->inode = inode; 16960450b2d1SAl Viro last->seq = seq; 1697d63ff28fSAl Viro return 1; 1698d63ff28fSAl Viro } 1699d63ff28fSAl Viro 17003ddcd056SLinus Torvalds /* 17013ddcd056SLinus Torvalds * Do we need to follow links? We _really_ want to be able 17023ddcd056SLinus Torvalds * to do this check without having to look at inode->i_op, 17033ddcd056SLinus Torvalds * so we keep a cache of "no, this doesn't need follow_link" 17043ddcd056SLinus Torvalds * for the common case. 17053ddcd056SLinus Torvalds */ 1706254cf582SAl Viro static inline int should_follow_link(struct nameidata *nd, struct path *link, 1707181548c0SAl Viro int follow, 1708181548c0SAl Viro struct inode *inode, unsigned seq) 17093ddcd056SLinus Torvalds { 1710d63ff28fSAl Viro if (likely(!d_is_symlink(link->dentry))) 1711d63ff28fSAl Viro return 0; 1712d63ff28fSAl Viro if (!follow) 1713d63ff28fSAl Viro return 0; 1714181548c0SAl Viro return pick_link(nd, link, inode, seq); 17153ddcd056SLinus Torvalds } 17163ddcd056SLinus Torvalds 17174693a547SAl Viro enum {WALK_GET = 1, WALK_PUT = 2}; 17184693a547SAl Viro 17194693a547SAl Viro static int walk_component(struct nameidata *nd, int flags) 1720ce57dfc1SAl Viro { 1721caa85634SAl Viro struct path path; 1722ce57dfc1SAl Viro struct inode *inode; 1723254cf582SAl Viro unsigned seq; 1724ce57dfc1SAl Viro int err; 1725ce57dfc1SAl Viro /* 1726ce57dfc1SAl Viro * "." and ".." are special - ".." especially so because it has 1727ce57dfc1SAl Viro * to be able to know about the current root directory and 1728ce57dfc1SAl Viro * parent relationships. 1729ce57dfc1SAl Viro */ 17304693a547SAl Viro if (unlikely(nd->last_type != LAST_NORM)) { 17314693a547SAl Viro err = handle_dots(nd, nd->last_type); 17324693a547SAl Viro if (flags & WALK_PUT) 17334693a547SAl Viro put_link(nd); 17344693a547SAl Viro return err; 17354693a547SAl Viro } 1736254cf582SAl Viro err = lookup_fast(nd, &path, &inode, &seq); 1737ce57dfc1SAl Viro if (unlikely(err)) { 1738697f514dSMiklos Szeredi if (err < 0) 1739f0a9ba70SAl Viro return err; 1740697f514dSMiklos Szeredi 1741caa85634SAl Viro err = lookup_slow(nd, &path); 1742697f514dSMiklos Szeredi if (err < 0) 1743f0a9ba70SAl Viro return err; 1744697f514dSMiklos Szeredi 174563afdfc7SDavid Howells inode = d_backing_inode(path.dentry); 1746254cf582SAl Viro seq = 0; /* we are already out of RCU mode */ 1747697f514dSMiklos Szeredi err = -ENOENT; 1748caa85634SAl Viro if (d_is_negative(path.dentry)) 1749697f514dSMiklos Szeredi goto out_path_put; 1750766c4cbfSAl Viro } 1751697f514dSMiklos Szeredi 17524693a547SAl Viro if (flags & WALK_PUT) 17534693a547SAl Viro put_link(nd); 1754181548c0SAl Viro err = should_follow_link(nd, &path, flags & WALK_GET, inode, seq); 1755d63ff28fSAl Viro if (unlikely(err)) 1756d63ff28fSAl Viro return err; 1757caa85634SAl Viro path_to_nameidata(&path, nd); 1758ce57dfc1SAl Viro nd->inode = inode; 1759254cf582SAl Viro nd->seq = seq; 1760ce57dfc1SAl Viro return 0; 1761697f514dSMiklos Szeredi 1762697f514dSMiklos Szeredi out_path_put: 1763caa85634SAl Viro path_to_nameidata(&path, nd); 1764697f514dSMiklos Szeredi return err; 1765ce57dfc1SAl Viro } 1766ce57dfc1SAl Viro 17671da177e4SLinus Torvalds /* 1768bfcfaa77SLinus Torvalds * We can do the critical dentry name comparison and hashing 1769bfcfaa77SLinus Torvalds * operations one word at a time, but we are limited to: 1770bfcfaa77SLinus Torvalds * 1771bfcfaa77SLinus Torvalds * - Architectures with fast unaligned word accesses. We could 1772bfcfaa77SLinus Torvalds * do a "get_unaligned()" if this helps and is sufficiently 1773bfcfaa77SLinus Torvalds * fast. 1774bfcfaa77SLinus Torvalds * 1775bfcfaa77SLinus Torvalds * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we 1776bfcfaa77SLinus Torvalds * do not trap on the (extremely unlikely) case of a page 1777bfcfaa77SLinus Torvalds * crossing operation. 1778bfcfaa77SLinus Torvalds * 1779bfcfaa77SLinus Torvalds * - Furthermore, we need an efficient 64-bit compile for the 1780bfcfaa77SLinus Torvalds * 64-bit case in order to generate the "number of bytes in 1781bfcfaa77SLinus Torvalds * the final mask". Again, that could be replaced with a 1782bfcfaa77SLinus Torvalds * efficient population count instruction or similar. 1783bfcfaa77SLinus Torvalds */ 1784bfcfaa77SLinus Torvalds #ifdef CONFIG_DCACHE_WORD_ACCESS 1785bfcfaa77SLinus Torvalds 1786f68e556eSLinus Torvalds #include <asm/word-at-a-time.h> 1787bfcfaa77SLinus Torvalds 1788f68e556eSLinus Torvalds #ifdef CONFIG_64BIT 1789bfcfaa77SLinus Torvalds 1790bfcfaa77SLinus Torvalds static inline unsigned int fold_hash(unsigned long hash) 1791bfcfaa77SLinus Torvalds { 179299d263d4SLinus Torvalds return hash_64(hash, 32); 1793bfcfaa77SLinus Torvalds } 1794bfcfaa77SLinus Torvalds 1795bfcfaa77SLinus Torvalds #else /* 32-bit case */ 1796bfcfaa77SLinus Torvalds 1797bfcfaa77SLinus Torvalds #define fold_hash(x) (x) 1798bfcfaa77SLinus Torvalds 1799bfcfaa77SLinus Torvalds #endif 1800bfcfaa77SLinus Torvalds 1801bfcfaa77SLinus Torvalds unsigned int full_name_hash(const unsigned char *name, unsigned int len) 1802bfcfaa77SLinus Torvalds { 1803bfcfaa77SLinus Torvalds unsigned long a, mask; 1804bfcfaa77SLinus Torvalds unsigned long hash = 0; 1805bfcfaa77SLinus Torvalds 1806bfcfaa77SLinus Torvalds for (;;) { 1807e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name); 1808bfcfaa77SLinus Torvalds if (len < sizeof(unsigned long)) 1809bfcfaa77SLinus Torvalds break; 1810bfcfaa77SLinus Torvalds hash += a; 1811f132c5beSAl Viro hash *= 9; 1812bfcfaa77SLinus Torvalds name += sizeof(unsigned long); 1813bfcfaa77SLinus Torvalds len -= sizeof(unsigned long); 1814bfcfaa77SLinus Torvalds if (!len) 1815bfcfaa77SLinus Torvalds goto done; 1816bfcfaa77SLinus Torvalds } 1817a5c21dceSWill Deacon mask = bytemask_from_count(len); 1818bfcfaa77SLinus Torvalds hash += mask & a; 1819bfcfaa77SLinus Torvalds done: 1820bfcfaa77SLinus Torvalds return fold_hash(hash); 1821bfcfaa77SLinus Torvalds } 1822bfcfaa77SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 1823bfcfaa77SLinus Torvalds 1824bfcfaa77SLinus Torvalds /* 1825bfcfaa77SLinus Torvalds * Calculate the length and hash of the path component, and 1826d6bb3e90SLinus Torvalds * return the "hash_len" as the result. 1827bfcfaa77SLinus Torvalds */ 1828d6bb3e90SLinus Torvalds static inline u64 hash_name(const char *name) 1829bfcfaa77SLinus Torvalds { 183036126f8fSLinus Torvalds unsigned long a, b, adata, bdata, mask, hash, len; 183136126f8fSLinus Torvalds const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 1832bfcfaa77SLinus Torvalds 1833bfcfaa77SLinus Torvalds hash = a = 0; 1834bfcfaa77SLinus Torvalds len = -sizeof(unsigned long); 1835bfcfaa77SLinus Torvalds do { 1836bfcfaa77SLinus Torvalds hash = (hash + a) * 9; 1837bfcfaa77SLinus Torvalds len += sizeof(unsigned long); 1838e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name+len); 183936126f8fSLinus Torvalds b = a ^ REPEAT_BYTE('/'); 184036126f8fSLinus Torvalds } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); 1841bfcfaa77SLinus Torvalds 184236126f8fSLinus Torvalds adata = prep_zero_mask(a, adata, &constants); 184336126f8fSLinus Torvalds bdata = prep_zero_mask(b, bdata, &constants); 184436126f8fSLinus Torvalds 184536126f8fSLinus Torvalds mask = create_zero_mask(adata | bdata); 184636126f8fSLinus Torvalds 184736126f8fSLinus Torvalds hash += a & zero_bytemask(mask); 18489226b5b4SLinus Torvalds len += find_zero(mask); 1849d6bb3e90SLinus Torvalds return hashlen_create(fold_hash(hash), len); 1850bfcfaa77SLinus Torvalds } 1851bfcfaa77SLinus Torvalds 1852bfcfaa77SLinus Torvalds #else 1853bfcfaa77SLinus Torvalds 18540145acc2SLinus Torvalds unsigned int full_name_hash(const unsigned char *name, unsigned int len) 18550145acc2SLinus Torvalds { 18560145acc2SLinus Torvalds unsigned long hash = init_name_hash(); 18570145acc2SLinus Torvalds while (len--) 18580145acc2SLinus Torvalds hash = partial_name_hash(*name++, hash); 18590145acc2SLinus Torvalds return end_name_hash(hash); 18600145acc2SLinus Torvalds } 1861ae942ae7SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 18620145acc2SLinus Torvalds 18633ddcd056SLinus Torvalds /* 1864200e9ef7SLinus Torvalds * We know there's a real path component here of at least 1865200e9ef7SLinus Torvalds * one character. 1866200e9ef7SLinus Torvalds */ 1867d6bb3e90SLinus Torvalds static inline u64 hash_name(const char *name) 1868200e9ef7SLinus Torvalds { 1869200e9ef7SLinus Torvalds unsigned long hash = init_name_hash(); 1870200e9ef7SLinus Torvalds unsigned long len = 0, c; 1871200e9ef7SLinus Torvalds 1872200e9ef7SLinus Torvalds c = (unsigned char)*name; 1873200e9ef7SLinus Torvalds do { 1874200e9ef7SLinus Torvalds len++; 1875200e9ef7SLinus Torvalds hash = partial_name_hash(c, hash); 1876200e9ef7SLinus Torvalds c = (unsigned char)name[len]; 1877200e9ef7SLinus Torvalds } while (c && c != '/'); 1878d6bb3e90SLinus Torvalds return hashlen_create(end_name_hash(hash), len); 1879200e9ef7SLinus Torvalds } 1880200e9ef7SLinus Torvalds 1881bfcfaa77SLinus Torvalds #endif 1882bfcfaa77SLinus Torvalds 1883200e9ef7SLinus Torvalds /* 18841da177e4SLinus Torvalds * Name resolution. 1885ea3834d9SPrasanna Meda * This is the basic name resolution function, turning a pathname into 1886ea3834d9SPrasanna Meda * the final dentry. We expect 'base' to be positive and a directory. 18871da177e4SLinus Torvalds * 1888ea3834d9SPrasanna Meda * Returns 0 and nd will have valid dentry and mnt on success. 1889ea3834d9SPrasanna Meda * Returns error and drops reference to input namei data on failure. 18901da177e4SLinus Torvalds */ 18916de88d72SAl Viro static int link_path_walk(const char *name, struct nameidata *nd) 18921da177e4SLinus Torvalds { 18931da177e4SLinus Torvalds int err; 18941da177e4SLinus Torvalds 18951da177e4SLinus Torvalds while (*name=='/') 18961da177e4SLinus Torvalds name++; 18971da177e4SLinus Torvalds if (!*name) 18989e18f10aSAl Viro return 0; 18991da177e4SLinus Torvalds 19001da177e4SLinus Torvalds /* At this point we know we have a real path component. */ 19011da177e4SLinus Torvalds for(;;) { 1902d6bb3e90SLinus Torvalds u64 hash_len; 1903fe479a58SAl Viro int type; 19041da177e4SLinus Torvalds 190552094c8aSAl Viro err = may_lookup(nd); 19061da177e4SLinus Torvalds if (err) 19073595e234SAl Viro return err; 19081da177e4SLinus Torvalds 1909d6bb3e90SLinus Torvalds hash_len = hash_name(name); 19101da177e4SLinus Torvalds 1911fe479a58SAl Viro type = LAST_NORM; 1912d6bb3e90SLinus Torvalds if (name[0] == '.') switch (hashlen_len(hash_len)) { 1913fe479a58SAl Viro case 2: 1914200e9ef7SLinus Torvalds if (name[1] == '.') { 1915fe479a58SAl Viro type = LAST_DOTDOT; 191616c2cd71SAl Viro nd->flags |= LOOKUP_JUMPED; 191716c2cd71SAl Viro } 1918fe479a58SAl Viro break; 1919fe479a58SAl Viro case 1: 1920fe479a58SAl Viro type = LAST_DOT; 1921fe479a58SAl Viro } 19225a202bcdSAl Viro if (likely(type == LAST_NORM)) { 19235a202bcdSAl Viro struct dentry *parent = nd->path.dentry; 192416c2cd71SAl Viro nd->flags &= ~LOOKUP_JUMPED; 19255a202bcdSAl Viro if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { 1926a060dc50SJames Hogan struct qstr this = { { .hash_len = hash_len }, .name = name }; 1927da53be12SLinus Torvalds err = parent->d_op->d_hash(parent, &this); 19285a202bcdSAl Viro if (err < 0) 19293595e234SAl Viro return err; 1930d6bb3e90SLinus Torvalds hash_len = this.hash_len; 1931d6bb3e90SLinus Torvalds name = this.name; 19325a202bcdSAl Viro } 19335a202bcdSAl Viro } 1934fe479a58SAl Viro 1935d6bb3e90SLinus Torvalds nd->last.hash_len = hash_len; 1936d6bb3e90SLinus Torvalds nd->last.name = name; 19375f4a6a69SAl Viro nd->last_type = type; 19385f4a6a69SAl Viro 1939d6bb3e90SLinus Torvalds name += hashlen_len(hash_len); 1940d6bb3e90SLinus Torvalds if (!*name) 1941bdf6cbf1SAl Viro goto OK; 1942200e9ef7SLinus Torvalds /* 1943200e9ef7SLinus Torvalds * If it wasn't NUL, we know it was '/'. Skip that 1944200e9ef7SLinus Torvalds * slash, and continue until no more slashes. 1945200e9ef7SLinus Torvalds */ 1946200e9ef7SLinus Torvalds do { 1947d6bb3e90SLinus Torvalds name++; 1948d6bb3e90SLinus Torvalds } while (unlikely(*name == '/')); 19498620c238SAl Viro if (unlikely(!*name)) { 19508620c238SAl Viro OK: 1951368ee9baSAl Viro /* pathname body, done */ 19528620c238SAl Viro if (!nd->depth) 19538620c238SAl Viro return 0; 19548620c238SAl Viro name = nd->stack[nd->depth - 1].name; 1955368ee9baSAl Viro /* trailing symlink, done */ 19568620c238SAl Viro if (!name) 19578620c238SAl Viro return 0; 19588620c238SAl Viro /* last component of nested symlink */ 19594693a547SAl Viro err = walk_component(nd, WALK_GET | WALK_PUT); 19608620c238SAl Viro } else { 19614693a547SAl Viro err = walk_component(nd, WALK_GET); 19628620c238SAl Viro } 1963ce57dfc1SAl Viro if (err < 0) 19643595e234SAl Viro return err; 1965fe479a58SAl Viro 1966ce57dfc1SAl Viro if (err) { 1967626de996SAl Viro const char *s = get_link(nd); 19685a460275SAl Viro 1969a1c83681SViresh Kumar if (IS_ERR(s)) 19703595e234SAl Viro return PTR_ERR(s); 1971172a39a0SAl Viro err = 0; 197212b09578SAl Viro if (unlikely(!s)) { 197312b09578SAl Viro /* jumped */ 1974b9ff4429SAl Viro put_link(nd); 197512b09578SAl Viro } else { 1976071bf501SAl Viro nd->stack[nd->depth - 1].name = name; 197732cd7468SAl Viro name = s; 19789e18f10aSAl Viro continue; 197948c8b0c5SAl Viro } 198031e6b01fSNick Piggin } 198197242f99SAl Viro if (unlikely(!d_can_lookup(nd->path.dentry))) { 198297242f99SAl Viro if (nd->flags & LOOKUP_RCU) { 198397242f99SAl Viro if (unlazy_walk(nd, NULL, 0)) 198497242f99SAl Viro return -ECHILD; 198597242f99SAl Viro } 19863595e234SAl Viro return -ENOTDIR; 19875f4a6a69SAl Viro } 1988ce57dfc1SAl Viro } 198997242f99SAl Viro } 19901da177e4SLinus Torvalds 1991c8a53ee5SAl Viro static const char *path_init(struct nameidata *nd, unsigned flags) 199231e6b01fSNick Piggin { 199331e6b01fSNick Piggin int retval = 0; 1994c8a53ee5SAl Viro const char *s = nd->name->name; 199531e6b01fSNick Piggin 199631e6b01fSNick Piggin nd->last_type = LAST_ROOT; /* if there are only slashes... */ 1997980f3ea2SAl Viro nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT; 199831e6b01fSNick Piggin nd->depth = 0; 19995b6ca027SAl Viro if (flags & LOOKUP_ROOT) { 2000b18825a7SDavid Howells struct dentry *root = nd->root.dentry; 2001b18825a7SDavid Howells struct inode *inode = root->d_inode; 2002fd2f7cb5SAl Viro if (*s) { 200344b1d530SMiklos Szeredi if (!d_can_lookup(root)) 2004368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 20055b6ca027SAl Viro retval = inode_permission(inode, MAY_EXEC); 20065b6ca027SAl Viro if (retval) 2007368ee9baSAl Viro return ERR_PTR(retval); 200873d049a4SAl Viro } 20095b6ca027SAl Viro nd->path = nd->root; 20105b6ca027SAl Viro nd->inode = inode; 20115b6ca027SAl Viro if (flags & LOOKUP_RCU) { 20128b61e74fSAl Viro rcu_read_lock(); 20135b6ca027SAl Viro nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 20148f47a016SAl Viro nd->root_seq = nd->seq; 201548a066e7SAl Viro nd->m_seq = read_seqbegin(&mount_lock); 20165b6ca027SAl Viro } else { 20175b6ca027SAl Viro path_get(&nd->path); 20185b6ca027SAl Viro } 2019368ee9baSAl Viro return s; 20205b6ca027SAl Viro } 20215b6ca027SAl Viro 202231e6b01fSNick Piggin nd->root.mnt = NULL; 202331e6b01fSNick Piggin 202448a066e7SAl Viro nd->m_seq = read_seqbegin(&mount_lock); 2025fd2f7cb5SAl Viro if (*s == '/') { 2026e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 20278b61e74fSAl Viro rcu_read_lock(); 202806d7137eSAl Viro set_root_rcu(nd); 202906d7137eSAl Viro nd->seq = nd->root_seq; 2030e41f7d4eSAl Viro } else { 2031e41f7d4eSAl Viro set_root(nd); 2032e41f7d4eSAl Viro path_get(&nd->root); 2033e41f7d4eSAl Viro } 203431e6b01fSNick Piggin nd->path = nd->root; 2035c8a53ee5SAl Viro } else if (nd->dfd == AT_FDCWD) { 2036e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 203731e6b01fSNick Piggin struct fs_struct *fs = current->fs; 2038c28cc364SNick Piggin unsigned seq; 203931e6b01fSNick Piggin 20408b61e74fSAl Viro rcu_read_lock(); 204131e6b01fSNick Piggin 2042c28cc364SNick Piggin do { 2043c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 204431e6b01fSNick Piggin nd->path = fs->pwd; 2045c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 2046c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 2047e41f7d4eSAl Viro } else { 2048e41f7d4eSAl Viro get_fs_pwd(current->fs, &nd->path); 2049e41f7d4eSAl Viro } 205031e6b01fSNick Piggin } else { 2051582aa64aSJeff Layton /* Caller must check execute permissions on the starting path component */ 2052c8a53ee5SAl Viro struct fd f = fdget_raw(nd->dfd); 205331e6b01fSNick Piggin struct dentry *dentry; 205431e6b01fSNick Piggin 20552903ff01SAl Viro if (!f.file) 2056368ee9baSAl Viro return ERR_PTR(-EBADF); 205731e6b01fSNick Piggin 20582903ff01SAl Viro dentry = f.file->f_path.dentry; 205931e6b01fSNick Piggin 2060fd2f7cb5SAl Viro if (*s) { 206144b1d530SMiklos Szeredi if (!d_can_lookup(dentry)) { 20622903ff01SAl Viro fdput(f); 2063368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 2064f52e0c11SAl Viro } 20652903ff01SAl Viro } 20662903ff01SAl Viro 20672903ff01SAl Viro nd->path = f.file->f_path; 2068e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 20698b61e74fSAl Viro rcu_read_lock(); 207034a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 207134a26b99SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 20725590ff0dSUlrich Drepper } else { 20732903ff01SAl Viro path_get(&nd->path); 207434a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 20751da177e4SLinus Torvalds } 207634a26b99SAl Viro fdput(f); 2077368ee9baSAl Viro return s; 2078e41f7d4eSAl Viro } 2079e41f7d4eSAl Viro 208031e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 20814023bfc9SAl Viro if (!(flags & LOOKUP_RCU)) 2082368ee9baSAl Viro return s; 20834023bfc9SAl Viro if (likely(!read_seqcount_retry(&nd->path.dentry->d_seq, nd->seq))) 2084368ee9baSAl Viro return s; 20854023bfc9SAl Viro if (!(nd->flags & LOOKUP_ROOT)) 20864023bfc9SAl Viro nd->root.mnt = NULL; 20874023bfc9SAl Viro rcu_read_unlock(); 2088368ee9baSAl Viro return ERR_PTR(-ECHILD); 20899b4a9b14SAl Viro } 20909b4a9b14SAl Viro 20913bdba28bSAl Viro static const char *trailing_symlink(struct nameidata *nd) 209295fa25d9SAl Viro { 209395fa25d9SAl Viro const char *s; 2094fec2fa24SAl Viro int error = may_follow_link(nd); 2095deb106c6SAl Viro if (unlikely(error)) 20963bdba28bSAl Viro return ERR_PTR(error); 209795fa25d9SAl Viro nd->flags |= LOOKUP_PARENT; 2098fab51e8aSAl Viro nd->stack[0].name = NULL; 20993b2e7f75SAl Viro s = get_link(nd); 2100deb106c6SAl Viro return s ? s : ""; 210195fa25d9SAl Viro } 210295fa25d9SAl Viro 2103caa85634SAl Viro static inline int lookup_last(struct nameidata *nd) 2104bd92d7feSAl Viro { 2105bd92d7feSAl Viro if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) 2106bd92d7feSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 2107bd92d7feSAl Viro 2108bd92d7feSAl Viro nd->flags &= ~LOOKUP_PARENT; 2109deb106c6SAl Viro return walk_component(nd, 21104693a547SAl Viro nd->flags & LOOKUP_FOLLOW 21114693a547SAl Viro ? nd->depth 21124693a547SAl Viro ? WALK_PUT | WALK_GET 21134693a547SAl Viro : WALK_GET 21144693a547SAl Viro : 0); 2115bd92d7feSAl Viro } 2116bd92d7feSAl Viro 21179b4a9b14SAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2118c8a53ee5SAl Viro static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path) 21199b4a9b14SAl Viro { 2120c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2121bd92d7feSAl Viro int err; 212231e6b01fSNick Piggin 2123368ee9baSAl Viro if (IS_ERR(s)) 2124368ee9baSAl Viro return PTR_ERR(s); 21253bdba28bSAl Viro while (!(err = link_path_walk(s, nd)) 21263bdba28bSAl Viro && ((err = lookup_last(nd)) > 0)) { 21273bdba28bSAl Viro s = trailing_symlink(nd); 21283bdba28bSAl Viro if (IS_ERR(s)) { 21293bdba28bSAl Viro err = PTR_ERR(s); 21306d7b5aaeSAl Viro break; 2131bd92d7feSAl Viro } 2132bd92d7feSAl Viro } 21339f1fafeeSAl Viro if (!err) 21349f1fafeeSAl Viro err = complete_walk(nd); 2135bd92d7feSAl Viro 2136deb106c6SAl Viro if (!err && nd->flags & LOOKUP_DIRECTORY) 2137deb106c6SAl Viro if (!d_can_lookup(nd->path.dentry)) 2138bd23a539SAl Viro err = -ENOTDIR; 2139625b6d10SAl Viro if (!err) { 2140625b6d10SAl Viro *path = nd->path; 2141625b6d10SAl Viro nd->path.mnt = NULL; 2142625b6d10SAl Viro nd->path.dentry = NULL; 2143625b6d10SAl Viro } 2144deb106c6SAl Viro terminate_walk(nd); 2145bd92d7feSAl Viro return err; 214631e6b01fSNick Piggin } 214731e6b01fSNick Piggin 2148625b6d10SAl Viro static int filename_lookup(int dfd, struct filename *name, unsigned flags, 21499ad1aaa6SAl Viro struct path *path, struct path *root) 2150873f1eedSJeff Layton { 2151894bc8c4SAl Viro int retval; 21529883d185SAl Viro struct nameidata nd; 2153abc9f5beSAl Viro if (IS_ERR(name)) 2154abc9f5beSAl Viro return PTR_ERR(name); 21559ad1aaa6SAl Viro if (unlikely(root)) { 21569ad1aaa6SAl Viro nd.root = *root; 21579ad1aaa6SAl Viro flags |= LOOKUP_ROOT; 21589ad1aaa6SAl Viro } 21599883d185SAl Viro set_nameidata(&nd, dfd, name); 2160c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_RCU, path); 2161873f1eedSJeff Layton if (unlikely(retval == -ECHILD)) 2162c8a53ee5SAl Viro retval = path_lookupat(&nd, flags, path); 2163873f1eedSJeff Layton if (unlikely(retval == -ESTALE)) 2164c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path); 2165873f1eedSJeff Layton 2166873f1eedSJeff Layton if (likely(!retval)) 2167625b6d10SAl Viro audit_inode(name, path->dentry, flags & LOOKUP_PARENT); 21689883d185SAl Viro restore_nameidata(); 2169e4bd1c1aSAl Viro putname(name); 2170873f1eedSJeff Layton return retval; 2171873f1eedSJeff Layton } 2172873f1eedSJeff Layton 21738bcb77faSAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2174c8a53ee5SAl Viro static int path_parentat(struct nameidata *nd, unsigned flags, 2175391172c4SAl Viro struct path *parent) 21768bcb77faSAl Viro { 2177c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2178368ee9baSAl Viro int err; 2179368ee9baSAl Viro if (IS_ERR(s)) 2180368ee9baSAl Viro return PTR_ERR(s); 2181368ee9baSAl Viro err = link_path_walk(s, nd); 21828bcb77faSAl Viro if (!err) 21838bcb77faSAl Viro err = complete_walk(nd); 2184391172c4SAl Viro if (!err) { 2185391172c4SAl Viro *parent = nd->path; 2186391172c4SAl Viro nd->path.mnt = NULL; 2187391172c4SAl Viro nd->path.dentry = NULL; 2188391172c4SAl Viro } 2189deb106c6SAl Viro terminate_walk(nd); 21908bcb77faSAl Viro return err; 21918bcb77faSAl Viro } 21928bcb77faSAl Viro 21935c31b6ceSAl Viro static struct filename *filename_parentat(int dfd, struct filename *name, 2194391172c4SAl Viro unsigned int flags, struct path *parent, 2195391172c4SAl Viro struct qstr *last, int *type) 21968bcb77faSAl Viro { 21978bcb77faSAl Viro int retval; 21989883d185SAl Viro struct nameidata nd; 21998bcb77faSAl Viro 22005c31b6ceSAl Viro if (IS_ERR(name)) 22015c31b6ceSAl Viro return name; 22029883d185SAl Viro set_nameidata(&nd, dfd, name); 2203c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_RCU, parent); 22048bcb77faSAl Viro if (unlikely(retval == -ECHILD)) 2205c8a53ee5SAl Viro retval = path_parentat(&nd, flags, parent); 22068bcb77faSAl Viro if (unlikely(retval == -ESTALE)) 2207c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent); 2208391172c4SAl Viro if (likely(!retval)) { 2209391172c4SAl Viro *last = nd.last; 2210391172c4SAl Viro *type = nd.last_type; 2211391172c4SAl Viro audit_inode(name, parent->dentry, LOOKUP_PARENT); 22125c31b6ceSAl Viro } else { 22135c31b6ceSAl Viro putname(name); 22145c31b6ceSAl Viro name = ERR_PTR(retval); 2215391172c4SAl Viro } 22169883d185SAl Viro restore_nameidata(); 22175c31b6ceSAl Viro return name; 22188bcb77faSAl Viro } 22198bcb77faSAl Viro 222079714f72SAl Viro /* does lookup, returns the object with parent locked */ 222179714f72SAl Viro struct dentry *kern_path_locked(const char *name, struct path *path) 22225590ff0dSUlrich Drepper { 22235c31b6ceSAl Viro struct filename *filename; 22245c31b6ceSAl Viro struct dentry *d; 2225391172c4SAl Viro struct qstr last; 2226391172c4SAl Viro int type; 222751689104SPaul Moore 22285c31b6ceSAl Viro filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path, 22295c31b6ceSAl Viro &last, &type); 223051689104SPaul Moore if (IS_ERR(filename)) 223151689104SPaul Moore return ERR_CAST(filename); 22325c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) { 2233391172c4SAl Viro path_put(path); 22345c31b6ceSAl Viro putname(filename); 22355c31b6ceSAl Viro return ERR_PTR(-EINVAL); 223679714f72SAl Viro } 2237391172c4SAl Viro mutex_lock_nested(&path->dentry->d_inode->i_mutex, I_MUTEX_PARENT); 2238391172c4SAl Viro d = __lookup_hash(&last, path->dentry, 0); 223979714f72SAl Viro if (IS_ERR(d)) { 2240391172c4SAl Viro mutex_unlock(&path->dentry->d_inode->i_mutex); 2241391172c4SAl Viro path_put(path); 224279714f72SAl Viro } 224351689104SPaul Moore putname(filename); 224479714f72SAl Viro return d; 22455590ff0dSUlrich Drepper } 22465590ff0dSUlrich Drepper 2247d1811465SAl Viro int kern_path(const char *name, unsigned int flags, struct path *path) 2248d1811465SAl Viro { 2249abc9f5beSAl Viro return filename_lookup(AT_FDCWD, getname_kernel(name), 2250abc9f5beSAl Viro flags, path, NULL); 2251d1811465SAl Viro } 22524d359507SAl Viro EXPORT_SYMBOL(kern_path); 2253d1811465SAl Viro 225416f18200SJosef 'Jeff' Sipek /** 225516f18200SJosef 'Jeff' Sipek * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair 225616f18200SJosef 'Jeff' Sipek * @dentry: pointer to dentry of the base directory 225716f18200SJosef 'Jeff' Sipek * @mnt: pointer to vfs mount of the base directory 225816f18200SJosef 'Jeff' Sipek * @name: pointer to file name 225916f18200SJosef 'Jeff' Sipek * @flags: lookup flags 2260e0a01249SAl Viro * @path: pointer to struct path to fill 226116f18200SJosef 'Jeff' Sipek */ 226216f18200SJosef 'Jeff' Sipek int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, 226316f18200SJosef 'Jeff' Sipek const char *name, unsigned int flags, 2264e0a01249SAl Viro struct path *path) 226516f18200SJosef 'Jeff' Sipek { 22669ad1aaa6SAl Viro struct path root = {.mnt = mnt, .dentry = dentry}; 22679ad1aaa6SAl Viro /* the first argument of filename_lookup() is ignored with root */ 2268abc9f5beSAl Viro return filename_lookup(AT_FDCWD, getname_kernel(name), 2269abc9f5beSAl Viro flags , path, &root); 227016f18200SJosef 'Jeff' Sipek } 22714d359507SAl Viro EXPORT_SYMBOL(vfs_path_lookup); 227216f18200SJosef 'Jeff' Sipek 2273eead1911SChristoph Hellwig /** 2274a6b91919SRandy Dunlap * lookup_one_len - filesystem helper to lookup single pathname component 2275eead1911SChristoph Hellwig * @name: pathname component to lookup 2276eead1911SChristoph Hellwig * @base: base directory to lookup from 2277eead1911SChristoph Hellwig * @len: maximum length @len should be interpreted to 2278eead1911SChristoph Hellwig * 2279a6b91919SRandy Dunlap * Note that this routine is purely a helper for filesystem usage and should 22809e7543e9SAl Viro * not be called by generic code. 2281eead1911SChristoph Hellwig */ 2282057f6c01SJames Morris struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 2283057f6c01SJames Morris { 2284057f6c01SJames Morris struct qstr this; 22856a96ba54SAl Viro unsigned int c; 2286cda309deSMiklos Szeredi int err; 2287057f6c01SJames Morris 22882f9092e1SDavid Woodhouse WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex)); 22892f9092e1SDavid Woodhouse 22906a96ba54SAl Viro this.name = name; 22916a96ba54SAl Viro this.len = len; 22920145acc2SLinus Torvalds this.hash = full_name_hash(name, len); 22936a96ba54SAl Viro if (!len) 22946a96ba54SAl Viro return ERR_PTR(-EACCES); 22956a96ba54SAl Viro 229621d8a15aSAl Viro if (unlikely(name[0] == '.')) { 229721d8a15aSAl Viro if (len < 2 || (len == 2 && name[1] == '.')) 229821d8a15aSAl Viro return ERR_PTR(-EACCES); 229921d8a15aSAl Viro } 230021d8a15aSAl Viro 23016a96ba54SAl Viro while (len--) { 23026a96ba54SAl Viro c = *(const unsigned char *)name++; 23036a96ba54SAl Viro if (c == '/' || c == '\0') 23046a96ba54SAl Viro return ERR_PTR(-EACCES); 23056a96ba54SAl Viro } 23065a202bcdSAl Viro /* 23075a202bcdSAl Viro * See if the low-level filesystem might want 23085a202bcdSAl Viro * to use its own hash.. 23095a202bcdSAl Viro */ 23105a202bcdSAl Viro if (base->d_flags & DCACHE_OP_HASH) { 2311da53be12SLinus Torvalds int err = base->d_op->d_hash(base, &this); 23125a202bcdSAl Viro if (err < 0) 23135a202bcdSAl Viro return ERR_PTR(err); 23145a202bcdSAl Viro } 2315eead1911SChristoph Hellwig 2316cda309deSMiklos Szeredi err = inode_permission(base->d_inode, MAY_EXEC); 2317cda309deSMiklos Szeredi if (err) 2318cda309deSMiklos Szeredi return ERR_PTR(err); 2319cda309deSMiklos Szeredi 232072bd866aSAl Viro return __lookup_hash(&this, base, 0); 2321057f6c01SJames Morris } 23224d359507SAl Viro EXPORT_SYMBOL(lookup_one_len); 2323057f6c01SJames Morris 23241fa1e7f6SAndy Whitcroft int user_path_at_empty(int dfd, const char __user *name, unsigned flags, 23251fa1e7f6SAndy Whitcroft struct path *path, int *empty) 23261da177e4SLinus Torvalds { 2327abc9f5beSAl Viro return filename_lookup(dfd, getname_flags(name, flags, empty), 2328abc9f5beSAl Viro flags, path, NULL); 23291da177e4SLinus Torvalds } 2330b853a161SAl Viro EXPORT_SYMBOL(user_path_at_empty); 23311fa1e7f6SAndy Whitcroft 2332873f1eedSJeff Layton /* 2333873f1eedSJeff Layton * NB: most callers don't do anything directly with the reference to the 2334873f1eedSJeff Layton * to struct filename, but the nd->last pointer points into the name string 2335873f1eedSJeff Layton * allocated by getname. So we must hold the reference to it until all 2336873f1eedSJeff Layton * path-walking is complete. 2337873f1eedSJeff Layton */ 2338a2ec4a2dSAl Viro static inline struct filename * 2339f5beed75SAl Viro user_path_parent(int dfd, const char __user *path, 2340f5beed75SAl Viro struct path *parent, 2341f5beed75SAl Viro struct qstr *last, 2342f5beed75SAl Viro int *type, 23439e790bd6SJeff Layton unsigned int flags) 23442ad94ae6SAl Viro { 23459e790bd6SJeff Layton /* only LOOKUP_REVAL is allowed in extra flags */ 23465c31b6ceSAl Viro return filename_parentat(dfd, getname(path), flags & LOOKUP_REVAL, 23475c31b6ceSAl Viro parent, last, type); 23482ad94ae6SAl Viro } 23492ad94ae6SAl Viro 23508033426eSJeff Layton /** 2351197df04cSAl Viro * mountpoint_last - look up last component for umount 23528033426eSJeff Layton * @nd: pathwalk nameidata - currently pointing at parent directory of "last" 23538033426eSJeff Layton * @path: pointer to container for result 23548033426eSJeff Layton * 23558033426eSJeff Layton * This is a special lookup_last function just for umount. In this case, we 23568033426eSJeff Layton * need to resolve the path without doing any revalidation. 23578033426eSJeff Layton * 23588033426eSJeff Layton * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since 23598033426eSJeff Layton * mountpoints are always pinned in the dcache, their ancestors are too. Thus, 23608033426eSJeff Layton * in almost all cases, this lookup will be served out of the dcache. The only 23618033426eSJeff Layton * cases where it won't are if nd->last refers to a symlink or the path is 23628033426eSJeff Layton * bogus and it doesn't exist. 23638033426eSJeff Layton * 23648033426eSJeff Layton * Returns: 23658033426eSJeff Layton * -error: if there was an error during lookup. This includes -ENOENT if the 23668033426eSJeff Layton * lookup found a negative dentry. The nd->path reference will also be 23678033426eSJeff Layton * put in this case. 23688033426eSJeff Layton * 23698033426eSJeff Layton * 0: if we successfully resolved nd->path and found it to not to be a 23708033426eSJeff Layton * symlink that needs to be followed. "path" will also be populated. 23718033426eSJeff Layton * The nd->path reference will also be put. 23728033426eSJeff Layton * 23738033426eSJeff Layton * 1: if we successfully resolved nd->last and found it to be a symlink 23748033426eSJeff Layton * that needs to be followed. "path" will be populated with the path 23758033426eSJeff Layton * to the link, and nd->path will *not* be put. 23768033426eSJeff Layton */ 23778033426eSJeff Layton static int 2378197df04cSAl Viro mountpoint_last(struct nameidata *nd, struct path *path) 23798033426eSJeff Layton { 23808033426eSJeff Layton int error = 0; 23818033426eSJeff Layton struct dentry *dentry; 23828033426eSJeff Layton struct dentry *dir = nd->path.dentry; 23838033426eSJeff Layton 238435759521SAl Viro /* If we're in rcuwalk, drop out of it to handle last component */ 238535759521SAl Viro if (nd->flags & LOOKUP_RCU) { 23866e9918b7SAl Viro if (unlazy_walk(nd, NULL, 0)) 2387deb106c6SAl Viro return -ECHILD; 23888033426eSJeff Layton } 23898033426eSJeff Layton 23908033426eSJeff Layton nd->flags &= ~LOOKUP_PARENT; 23918033426eSJeff Layton 23928033426eSJeff Layton if (unlikely(nd->last_type != LAST_NORM)) { 23938033426eSJeff Layton error = handle_dots(nd, nd->last_type); 239435759521SAl Viro if (error) 2395deb106c6SAl Viro return error; 23968033426eSJeff Layton dentry = dget(nd->path.dentry); 239735759521SAl Viro goto done; 23988033426eSJeff Layton } 23998033426eSJeff Layton 24008033426eSJeff Layton mutex_lock(&dir->d_inode->i_mutex); 24018033426eSJeff Layton dentry = d_lookup(dir, &nd->last); 24028033426eSJeff Layton if (!dentry) { 24038033426eSJeff Layton /* 24048033426eSJeff Layton * No cached dentry. Mounted dentries are pinned in the cache, 24058033426eSJeff Layton * so that means that this dentry is probably a symlink or the 24068033426eSJeff Layton * path doesn't actually point to a mounted dentry. 24078033426eSJeff Layton */ 24088033426eSJeff Layton dentry = d_alloc(dir, &nd->last); 24098033426eSJeff Layton if (!dentry) { 2410bcceeebaSDave Jones mutex_unlock(&dir->d_inode->i_mutex); 2411deb106c6SAl Viro return -ENOMEM; 24128033426eSJeff Layton } 241335759521SAl Viro dentry = lookup_real(dir->d_inode, dentry, nd->flags); 2414bcceeebaSDave Jones if (IS_ERR(dentry)) { 2415bcceeebaSDave Jones mutex_unlock(&dir->d_inode->i_mutex); 2416deb106c6SAl Viro return PTR_ERR(dentry); 24178033426eSJeff Layton } 2418bcceeebaSDave Jones } 24198033426eSJeff Layton mutex_unlock(&dir->d_inode->i_mutex); 24208033426eSJeff Layton 242135759521SAl Viro done: 2422698934dfSDavid Howells if (d_is_negative(dentry)) { 24238033426eSJeff Layton dput(dentry); 2424deb106c6SAl Viro return -ENOENT; 242535759521SAl Viro } 2426191d7f73SAl Viro if (nd->depth) 2427191d7f73SAl Viro put_link(nd); 24288033426eSJeff Layton path->dentry = dentry; 2429295dc39dSVasily Averin path->mnt = nd->path.mnt; 2430181548c0SAl Viro error = should_follow_link(nd, path, nd->flags & LOOKUP_FOLLOW, 2431181548c0SAl Viro d_backing_inode(dentry), 0); 2432deb106c6SAl Viro if (unlikely(error)) 2433d63ff28fSAl Viro return error; 2434295dc39dSVasily Averin mntget(path->mnt); 24358033426eSJeff Layton follow_mount(path); 2436deb106c6SAl Viro return 0; 24378033426eSJeff Layton } 24388033426eSJeff Layton 24398033426eSJeff Layton /** 2440197df04cSAl Viro * path_mountpoint - look up a path to be umounted 24412a78b857SMasanari Iida * @nd: lookup context 24428033426eSJeff Layton * @flags: lookup flags 2443c8a53ee5SAl Viro * @path: pointer to container for result 24448033426eSJeff Layton * 24458033426eSJeff Layton * Look up the given name, but don't attempt to revalidate the last component. 2446606d6fe3SRandy Dunlap * Returns 0 and "path" will be valid on success; Returns error otherwise. 24478033426eSJeff Layton */ 24488033426eSJeff Layton static int 2449c8a53ee5SAl Viro path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path) 24508033426eSJeff Layton { 2451c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2452368ee9baSAl Viro int err; 2453368ee9baSAl Viro if (IS_ERR(s)) 2454368ee9baSAl Viro return PTR_ERR(s); 24553bdba28bSAl Viro while (!(err = link_path_walk(s, nd)) && 24563bdba28bSAl Viro (err = mountpoint_last(nd, path)) > 0) { 24573bdba28bSAl Viro s = trailing_symlink(nd); 24583bdba28bSAl Viro if (IS_ERR(s)) { 24593bdba28bSAl Viro err = PTR_ERR(s); 24608033426eSJeff Layton break; 24618033426eSJeff Layton } 24623bdba28bSAl Viro } 2463deb106c6SAl Viro terminate_walk(nd); 24648033426eSJeff Layton return err; 24658033426eSJeff Layton } 24668033426eSJeff Layton 24672d864651SAl Viro static int 2468668696dcSAl Viro filename_mountpoint(int dfd, struct filename *name, struct path *path, 24692d864651SAl Viro unsigned int flags) 24702d864651SAl Viro { 24719883d185SAl Viro struct nameidata nd; 2472cbaab2dbSAl Viro int error; 2473668696dcSAl Viro if (IS_ERR(name)) 2474668696dcSAl Viro return PTR_ERR(name); 24759883d185SAl Viro set_nameidata(&nd, dfd, name); 2476c8a53ee5SAl Viro error = path_mountpoint(&nd, flags | LOOKUP_RCU, path); 24772d864651SAl Viro if (unlikely(error == -ECHILD)) 2478c8a53ee5SAl Viro error = path_mountpoint(&nd, flags, path); 24792d864651SAl Viro if (unlikely(error == -ESTALE)) 2480c8a53ee5SAl Viro error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path); 24812d864651SAl Viro if (likely(!error)) 2482668696dcSAl Viro audit_inode(name, path->dentry, 0); 24839883d185SAl Viro restore_nameidata(); 2484668696dcSAl Viro putname(name); 24852d864651SAl Viro return error; 24862d864651SAl Viro } 24872d864651SAl Viro 24888033426eSJeff Layton /** 2489197df04cSAl Viro * user_path_mountpoint_at - lookup a path from userland in order to umount it 24908033426eSJeff Layton * @dfd: directory file descriptor 24918033426eSJeff Layton * @name: pathname from userland 24928033426eSJeff Layton * @flags: lookup flags 24938033426eSJeff Layton * @path: pointer to container to hold result 24948033426eSJeff Layton * 24958033426eSJeff Layton * A umount is a special case for path walking. We're not actually interested 24968033426eSJeff Layton * in the inode in this situation, and ESTALE errors can be a problem. We 24978033426eSJeff Layton * simply want track down the dentry and vfsmount attached at the mountpoint 24988033426eSJeff Layton * and avoid revalidating the last component. 24998033426eSJeff Layton * 25008033426eSJeff Layton * Returns 0 and populates "path" on success. 25018033426eSJeff Layton */ 25028033426eSJeff Layton int 2503197df04cSAl Viro user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags, 25048033426eSJeff Layton struct path *path) 25058033426eSJeff Layton { 2506cbaab2dbSAl Viro return filename_mountpoint(dfd, getname(name), path, flags); 25078033426eSJeff Layton } 25088033426eSJeff Layton 25092d864651SAl Viro int 25102d864651SAl Viro kern_path_mountpoint(int dfd, const char *name, struct path *path, 25112d864651SAl Viro unsigned int flags) 25122d864651SAl Viro { 2513cbaab2dbSAl Viro return filename_mountpoint(dfd, getname_kernel(name), path, flags); 25142d864651SAl Viro } 25152d864651SAl Viro EXPORT_SYMBOL(kern_path_mountpoint); 25162d864651SAl Viro 2517cbdf35bcSMiklos Szeredi int __check_sticky(struct inode *dir, struct inode *inode) 25181da177e4SLinus Torvalds { 25198e96e3b7SEric W. Biederman kuid_t fsuid = current_fsuid(); 2520da9592edSDavid Howells 25218e96e3b7SEric W. Biederman if (uid_eq(inode->i_uid, fsuid)) 25221da177e4SLinus Torvalds return 0; 25238e96e3b7SEric W. Biederman if (uid_eq(dir->i_uid, fsuid)) 25241da177e4SLinus Torvalds return 0; 252523adbe12SAndy Lutomirski return !capable_wrt_inode_uidgid(inode, CAP_FOWNER); 25261da177e4SLinus Torvalds } 2527cbdf35bcSMiklos Szeredi EXPORT_SYMBOL(__check_sticky); 25281da177e4SLinus Torvalds 25291da177e4SLinus Torvalds /* 25301da177e4SLinus Torvalds * Check whether we can remove a link victim from directory dir, check 25311da177e4SLinus Torvalds * whether the type of victim is right. 25321da177e4SLinus Torvalds * 1. We can't do it if dir is read-only (done in permission()) 25331da177e4SLinus Torvalds * 2. We should have write and exec permissions on dir 25341da177e4SLinus Torvalds * 3. We can't remove anything from append-only dir 25351da177e4SLinus Torvalds * 4. We can't do anything with immutable dir (done in permission()) 25361da177e4SLinus Torvalds * 5. If the sticky bit on dir is set we should either 25371da177e4SLinus Torvalds * a. be owner of dir, or 25381da177e4SLinus Torvalds * b. be owner of victim, or 25391da177e4SLinus Torvalds * c. have CAP_FOWNER capability 25401da177e4SLinus Torvalds * 6. If the victim is append-only or immutable we can't do antyhing with 25411da177e4SLinus Torvalds * links pointing to it. 25421da177e4SLinus Torvalds * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. 25431da177e4SLinus Torvalds * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. 25441da177e4SLinus Torvalds * 9. We can't remove a root or mountpoint. 25451da177e4SLinus Torvalds * 10. We don't allow removal of NFS sillyrenamed files; it's handled by 25461da177e4SLinus Torvalds * nfs_async_unlink(). 25471da177e4SLinus Torvalds */ 2548b18825a7SDavid Howells static int may_delete(struct inode *dir, struct dentry *victim, bool isdir) 25491da177e4SLinus Torvalds { 255063afdfc7SDavid Howells struct inode *inode = d_backing_inode(victim); 25511da177e4SLinus Torvalds int error; 25521da177e4SLinus Torvalds 2553b18825a7SDavid Howells if (d_is_negative(victim)) 25541da177e4SLinus Torvalds return -ENOENT; 2555b18825a7SDavid Howells BUG_ON(!inode); 25561da177e4SLinus Torvalds 25571da177e4SLinus Torvalds BUG_ON(victim->d_parent->d_inode != dir); 25584fa6b5ecSJeff Layton audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); 25591da177e4SLinus Torvalds 2560f419a2e3SAl Viro error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 25611da177e4SLinus Torvalds if (error) 25621da177e4SLinus Torvalds return error; 25631da177e4SLinus Torvalds if (IS_APPEND(dir)) 25641da177e4SLinus Torvalds return -EPERM; 2565b18825a7SDavid Howells 2566b18825a7SDavid Howells if (check_sticky(dir, inode) || IS_APPEND(inode) || 2567b18825a7SDavid Howells IS_IMMUTABLE(inode) || IS_SWAPFILE(inode)) 25681da177e4SLinus Torvalds return -EPERM; 25691da177e4SLinus Torvalds if (isdir) { 257044b1d530SMiklos Szeredi if (!d_is_dir(victim)) 25711da177e4SLinus Torvalds return -ENOTDIR; 25721da177e4SLinus Torvalds if (IS_ROOT(victim)) 25731da177e4SLinus Torvalds return -EBUSY; 257444b1d530SMiklos Szeredi } else if (d_is_dir(victim)) 25751da177e4SLinus Torvalds return -EISDIR; 25761da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 25771da177e4SLinus Torvalds return -ENOENT; 25781da177e4SLinus Torvalds if (victim->d_flags & DCACHE_NFSFS_RENAMED) 25791da177e4SLinus Torvalds return -EBUSY; 25801da177e4SLinus Torvalds return 0; 25811da177e4SLinus Torvalds } 25821da177e4SLinus Torvalds 25831da177e4SLinus Torvalds /* Check whether we can create an object with dentry child in directory 25841da177e4SLinus Torvalds * dir. 25851da177e4SLinus Torvalds * 1. We can't do it if child already exists (open has special treatment for 25861da177e4SLinus Torvalds * this case, but since we are inlined it's OK) 25871da177e4SLinus Torvalds * 2. We can't do it if dir is read-only (done in permission()) 25881da177e4SLinus Torvalds * 3. We should have write and exec permissions on dir 25891da177e4SLinus Torvalds * 4. We can't do it if dir is immutable (done in permission()) 25901da177e4SLinus Torvalds */ 2591a95164d9SMiklos Szeredi static inline int may_create(struct inode *dir, struct dentry *child) 25921da177e4SLinus Torvalds { 259314e972b4SJeff Layton audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); 25941da177e4SLinus Torvalds if (child->d_inode) 25951da177e4SLinus Torvalds return -EEXIST; 25961da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 25971da177e4SLinus Torvalds return -ENOENT; 2598f419a2e3SAl Viro return inode_permission(dir, MAY_WRITE | MAY_EXEC); 25991da177e4SLinus Torvalds } 26001da177e4SLinus Torvalds 26011da177e4SLinus Torvalds /* 26021da177e4SLinus Torvalds * p1 and p2 should be directories on the same fs. 26031da177e4SLinus Torvalds */ 26041da177e4SLinus Torvalds struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 26051da177e4SLinus Torvalds { 26061da177e4SLinus Torvalds struct dentry *p; 26071da177e4SLinus Torvalds 26081da177e4SLinus Torvalds if (p1 == p2) { 2609f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 26101da177e4SLinus Torvalds return NULL; 26111da177e4SLinus Torvalds } 26121da177e4SLinus Torvalds 2613a11f3a05SArjan van de Ven mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 26141da177e4SLinus Torvalds 2615e2761a11SOGAWA Hirofumi p = d_ancestor(p2, p1); 2616e2761a11SOGAWA Hirofumi if (p) { 2617f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT); 2618f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD); 26191da177e4SLinus Torvalds return p; 26201da177e4SLinus Torvalds } 26211da177e4SLinus Torvalds 2622e2761a11SOGAWA Hirofumi p = d_ancestor(p1, p2); 2623e2761a11SOGAWA Hirofumi if (p) { 2624f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 2625f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); 26261da177e4SLinus Torvalds return p; 26271da177e4SLinus Torvalds } 26281da177e4SLinus Torvalds 2629f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 2630d1b72cc6SMiklos Szeredi mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT2); 26311da177e4SLinus Torvalds return NULL; 26321da177e4SLinus Torvalds } 26334d359507SAl Viro EXPORT_SYMBOL(lock_rename); 26341da177e4SLinus Torvalds 26351da177e4SLinus Torvalds void unlock_rename(struct dentry *p1, struct dentry *p2) 26361da177e4SLinus Torvalds { 26371b1dcc1bSJes Sorensen mutex_unlock(&p1->d_inode->i_mutex); 26381da177e4SLinus Torvalds if (p1 != p2) { 26391b1dcc1bSJes Sorensen mutex_unlock(&p2->d_inode->i_mutex); 2640a11f3a05SArjan van de Ven mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 26411da177e4SLinus Torvalds } 26421da177e4SLinus Torvalds } 26434d359507SAl Viro EXPORT_SYMBOL(unlock_rename); 26441da177e4SLinus Torvalds 26454acdaf27SAl Viro int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, 2646312b63fbSAl Viro bool want_excl) 26471da177e4SLinus Torvalds { 2648a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 26491da177e4SLinus Torvalds if (error) 26501da177e4SLinus Torvalds return error; 26511da177e4SLinus Torvalds 2652acfa4380SAl Viro if (!dir->i_op->create) 26531da177e4SLinus Torvalds return -EACCES; /* shouldn't it be ENOSYS? */ 26541da177e4SLinus Torvalds mode &= S_IALLUGO; 26551da177e4SLinus Torvalds mode |= S_IFREG; 26561da177e4SLinus Torvalds error = security_inode_create(dir, dentry, mode); 26571da177e4SLinus Torvalds if (error) 26581da177e4SLinus Torvalds return error; 2659312b63fbSAl Viro error = dir->i_op->create(dir, dentry, mode, want_excl); 2660a74574aaSStephen Smalley if (!error) 2661f38aa942SAmy Griffis fsnotify_create(dir, dentry); 26621da177e4SLinus Torvalds return error; 26631da177e4SLinus Torvalds } 26644d359507SAl Viro EXPORT_SYMBOL(vfs_create); 26651da177e4SLinus Torvalds 266673d049a4SAl Viro static int may_open(struct path *path, int acc_mode, int flag) 26671da177e4SLinus Torvalds { 26683fb64190SChristoph Hellwig struct dentry *dentry = path->dentry; 26691da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 26701da177e4SLinus Torvalds int error; 26711da177e4SLinus Torvalds 2672bcda7652SAl Viro /* O_PATH? */ 2673bcda7652SAl Viro if (!acc_mode) 2674bcda7652SAl Viro return 0; 2675bcda7652SAl Viro 26761da177e4SLinus Torvalds if (!inode) 26771da177e4SLinus Torvalds return -ENOENT; 26781da177e4SLinus Torvalds 2679c8fe8f30SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 2680c8fe8f30SChristoph Hellwig case S_IFLNK: 26811da177e4SLinus Torvalds return -ELOOP; 2682c8fe8f30SChristoph Hellwig case S_IFDIR: 2683c8fe8f30SChristoph Hellwig if (acc_mode & MAY_WRITE) 26841da177e4SLinus Torvalds return -EISDIR; 2685c8fe8f30SChristoph Hellwig break; 2686c8fe8f30SChristoph Hellwig case S_IFBLK: 2687c8fe8f30SChristoph Hellwig case S_IFCHR: 26883fb64190SChristoph Hellwig if (path->mnt->mnt_flags & MNT_NODEV) 26891da177e4SLinus Torvalds return -EACCES; 2690c8fe8f30SChristoph Hellwig /*FALLTHRU*/ 2691c8fe8f30SChristoph Hellwig case S_IFIFO: 2692c8fe8f30SChristoph Hellwig case S_IFSOCK: 26931da177e4SLinus Torvalds flag &= ~O_TRUNC; 2694c8fe8f30SChristoph Hellwig break; 26954a3fd211SDave Hansen } 2696b41572e9SDave Hansen 26973fb64190SChristoph Hellwig error = inode_permission(inode, acc_mode); 2698b41572e9SDave Hansen if (error) 2699b41572e9SDave Hansen return error; 27006146f0d5SMimi Zohar 27011da177e4SLinus Torvalds /* 27021da177e4SLinus Torvalds * An append-only file must be opened in append mode for writing. 27031da177e4SLinus Torvalds */ 27041da177e4SLinus Torvalds if (IS_APPEND(inode)) { 27058737c930SAl Viro if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) 27067715b521SAl Viro return -EPERM; 27071da177e4SLinus Torvalds if (flag & O_TRUNC) 27087715b521SAl Viro return -EPERM; 27091da177e4SLinus Torvalds } 27101da177e4SLinus Torvalds 27111da177e4SLinus Torvalds /* O_NOATIME can only be set by the owner or superuser */ 27122e149670SSerge E. Hallyn if (flag & O_NOATIME && !inode_owner_or_capable(inode)) 27137715b521SAl Viro return -EPERM; 27141da177e4SLinus Torvalds 2715f3c7691eSJ. Bruce Fields return 0; 27167715b521SAl Viro } 27177715b521SAl Viro 2718e1181ee6SJeff Layton static int handle_truncate(struct file *filp) 27197715b521SAl Viro { 2720e1181ee6SJeff Layton struct path *path = &filp->f_path; 27217715b521SAl Viro struct inode *inode = path->dentry->d_inode; 27227715b521SAl Viro int error = get_write_access(inode); 27231da177e4SLinus Torvalds if (error) 27247715b521SAl Viro return error; 27251da177e4SLinus Torvalds /* 27261da177e4SLinus Torvalds * Refuse to truncate files with mandatory locks held on them. 27271da177e4SLinus Torvalds */ 2728d7a06983SJeff Layton error = locks_verify_locked(filp); 2729be6d3e56SKentaro Takeda if (!error) 2730ea0d3ab2STetsuo Handa error = security_path_truncate(path); 27311da177e4SLinus Torvalds if (!error) { 27327715b521SAl Viro error = do_truncate(path->dentry, 0, 2733d139d7ffSMiklos Szeredi ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, 2734e1181ee6SJeff Layton filp); 27351da177e4SLinus Torvalds } 27361da177e4SLinus Torvalds put_write_access(inode); 2737acd0c935SMimi Zohar return error; 27381da177e4SLinus Torvalds } 27391da177e4SLinus Torvalds 2740d57999e1SDave Hansen static inline int open_to_namei_flags(int flag) 2741d57999e1SDave Hansen { 27428a5e929dSAl Viro if ((flag & O_ACCMODE) == 3) 27438a5e929dSAl Viro flag--; 2744d57999e1SDave Hansen return flag; 2745d57999e1SDave Hansen } 2746d57999e1SDave Hansen 2747d18e9008SMiklos Szeredi static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode) 2748d18e9008SMiklos Szeredi { 2749d18e9008SMiklos Szeredi int error = security_path_mknod(dir, dentry, mode, 0); 2750d18e9008SMiklos Szeredi if (error) 2751d18e9008SMiklos Szeredi return error; 2752d18e9008SMiklos Szeredi 2753d18e9008SMiklos Szeredi error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC); 2754d18e9008SMiklos Szeredi if (error) 2755d18e9008SMiklos Szeredi return error; 2756d18e9008SMiklos Szeredi 2757d18e9008SMiklos Szeredi return security_inode_create(dir->dentry->d_inode, dentry, mode); 2758d18e9008SMiklos Szeredi } 2759d18e9008SMiklos Szeredi 27601acf0af9SDavid Howells /* 27611acf0af9SDavid Howells * Attempt to atomically look up, create and open a file from a negative 27621acf0af9SDavid Howells * dentry. 27631acf0af9SDavid Howells * 27641acf0af9SDavid Howells * Returns 0 if successful. The file will have been created and attached to 27651acf0af9SDavid Howells * @file by the filesystem calling finish_open(). 27661acf0af9SDavid Howells * 27671acf0af9SDavid Howells * Returns 1 if the file was looked up only or didn't need creating. The 27681acf0af9SDavid Howells * caller will need to perform the open themselves. @path will have been 27691acf0af9SDavid Howells * updated to point to the new dentry. This may be negative. 27701acf0af9SDavid Howells * 27711acf0af9SDavid Howells * Returns an error code otherwise. 27721acf0af9SDavid Howells */ 27732675a4ebSAl Viro static int atomic_open(struct nameidata *nd, struct dentry *dentry, 277430d90494SAl Viro struct path *path, struct file *file, 2775015c3bbcSMiklos Szeredi const struct open_flags *op, 277664894cf8SAl Viro bool got_write, bool need_lookup, 277747237687SAl Viro int *opened) 2778d18e9008SMiklos Szeredi { 2779d18e9008SMiklos Szeredi struct inode *dir = nd->path.dentry->d_inode; 2780d18e9008SMiklos Szeredi unsigned open_flag = open_to_namei_flags(op->open_flag); 2781d18e9008SMiklos Szeredi umode_t mode; 2782d18e9008SMiklos Szeredi int error; 2783d18e9008SMiklos Szeredi int acc_mode; 2784d18e9008SMiklos Szeredi int create_error = 0; 2785d18e9008SMiklos Szeredi struct dentry *const DENTRY_NOT_SET = (void *) -1UL; 2786116cc022SMiklos Szeredi bool excl; 2787d18e9008SMiklos Szeredi 2788d18e9008SMiklos Szeredi BUG_ON(dentry->d_inode); 2789d18e9008SMiklos Szeredi 2790d18e9008SMiklos Szeredi /* Don't create child dentry for a dead directory. */ 2791d18e9008SMiklos Szeredi if (unlikely(IS_DEADDIR(dir))) { 27922675a4ebSAl Viro error = -ENOENT; 2793d18e9008SMiklos Szeredi goto out; 2794d18e9008SMiklos Szeredi } 2795d18e9008SMiklos Szeredi 279662b259d8SMiklos Szeredi mode = op->mode; 2797d18e9008SMiklos Szeredi if ((open_flag & O_CREAT) && !IS_POSIXACL(dir)) 2798d18e9008SMiklos Szeredi mode &= ~current_umask(); 2799d18e9008SMiklos Szeredi 2800116cc022SMiklos Szeredi excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT); 2801116cc022SMiklos Szeredi if (excl) 2802d18e9008SMiklos Szeredi open_flag &= ~O_TRUNC; 2803d18e9008SMiklos Szeredi 2804d18e9008SMiklos Szeredi /* 2805d18e9008SMiklos Szeredi * Checking write permission is tricky, bacuse we don't know if we are 2806d18e9008SMiklos Szeredi * going to actually need it: O_CREAT opens should work as long as the 2807d18e9008SMiklos Szeredi * file exists. But checking existence breaks atomicity. The trick is 2808d18e9008SMiklos Szeredi * to check access and if not granted clear O_CREAT from the flags. 2809d18e9008SMiklos Szeredi * 2810d18e9008SMiklos Szeredi * Another problem is returing the "right" error value (e.g. for an 2811d18e9008SMiklos Szeredi * O_EXCL open we want to return EEXIST not EROFS). 2812d18e9008SMiklos Szeredi */ 281364894cf8SAl Viro if (((open_flag & (O_CREAT | O_TRUNC)) || 281464894cf8SAl Viro (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) { 281564894cf8SAl Viro if (!(open_flag & O_CREAT)) { 2816d18e9008SMiklos Szeredi /* 2817d18e9008SMiklos Szeredi * No O_CREATE -> atomicity not a requirement -> fall 2818d18e9008SMiklos Szeredi * back to lookup + open 2819d18e9008SMiklos Szeredi */ 2820d18e9008SMiklos Szeredi goto no_open; 2821d18e9008SMiklos Szeredi } else if (open_flag & (O_EXCL | O_TRUNC)) { 2822d18e9008SMiklos Szeredi /* Fall back and fail with the right error */ 282364894cf8SAl Viro create_error = -EROFS; 2824d18e9008SMiklos Szeredi goto no_open; 2825d18e9008SMiklos Szeredi } else { 2826d18e9008SMiklos Szeredi /* No side effects, safe to clear O_CREAT */ 282764894cf8SAl Viro create_error = -EROFS; 2828d18e9008SMiklos Szeredi open_flag &= ~O_CREAT; 2829d18e9008SMiklos Szeredi } 2830d18e9008SMiklos Szeredi } 2831d18e9008SMiklos Szeredi 2832d18e9008SMiklos Szeredi if (open_flag & O_CREAT) { 283338227f78SMiklos Szeredi error = may_o_create(&nd->path, dentry, mode); 2834d18e9008SMiklos Szeredi if (error) { 2835d18e9008SMiklos Szeredi create_error = error; 2836d18e9008SMiklos Szeredi if (open_flag & O_EXCL) 2837d18e9008SMiklos Szeredi goto no_open; 2838d18e9008SMiklos Szeredi open_flag &= ~O_CREAT; 2839d18e9008SMiklos Szeredi } 2840d18e9008SMiklos Szeredi } 2841d18e9008SMiklos Szeredi 2842d18e9008SMiklos Szeredi if (nd->flags & LOOKUP_DIRECTORY) 2843d18e9008SMiklos Szeredi open_flag |= O_DIRECTORY; 2844d18e9008SMiklos Szeredi 284530d90494SAl Viro file->f_path.dentry = DENTRY_NOT_SET; 284630d90494SAl Viro file->f_path.mnt = nd->path.mnt; 284730d90494SAl Viro error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode, 284847237687SAl Viro opened); 2849d9585277SAl Viro if (error < 0) { 2850d9585277SAl Viro if (create_error && error == -ENOENT) 2851d9585277SAl Viro error = create_error; 2852d18e9008SMiklos Szeredi goto out; 2853d18e9008SMiklos Szeredi } 2854d18e9008SMiklos Szeredi 2855d9585277SAl Viro if (error) { /* returned 1, that is */ 285630d90494SAl Viro if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { 28572675a4ebSAl Viro error = -EIO; 2858d18e9008SMiklos Szeredi goto out; 2859d18e9008SMiklos Szeredi } 286030d90494SAl Viro if (file->f_path.dentry) { 2861d18e9008SMiklos Szeredi dput(dentry); 286230d90494SAl Viro dentry = file->f_path.dentry; 2863d18e9008SMiklos Szeredi } 286403da633aSAl Viro if (*opened & FILE_CREATED) 286503da633aSAl Viro fsnotify_create(dir, dentry); 286603da633aSAl Viro if (!dentry->d_inode) { 286703da633aSAl Viro WARN_ON(*opened & FILE_CREATED); 286803da633aSAl Viro if (create_error) { 286962b2ce96SSage Weil error = create_error; 287062b2ce96SSage Weil goto out; 287162b2ce96SSage Weil } 287203da633aSAl Viro } else { 287303da633aSAl Viro if (excl && !(*opened & FILE_CREATED)) { 287403da633aSAl Viro error = -EEXIST; 287503da633aSAl Viro goto out; 287603da633aSAl Viro } 287703da633aSAl Viro } 2878d18e9008SMiklos Szeredi goto looked_up; 2879d18e9008SMiklos Szeredi } 2880d18e9008SMiklos Szeredi 2881d18e9008SMiklos Szeredi /* 2882d18e9008SMiklos Szeredi * We didn't have the inode before the open, so check open permission 2883d18e9008SMiklos Szeredi * here. 2884d18e9008SMiklos Szeredi */ 288503da633aSAl Viro acc_mode = op->acc_mode; 288603da633aSAl Viro if (*opened & FILE_CREATED) { 288703da633aSAl Viro WARN_ON(!(open_flag & O_CREAT)); 288803da633aSAl Viro fsnotify_create(dir, dentry); 288903da633aSAl Viro acc_mode = MAY_OPEN; 289003da633aSAl Viro } 28912675a4ebSAl Viro error = may_open(&file->f_path, acc_mode, open_flag); 28922675a4ebSAl Viro if (error) 28932675a4ebSAl Viro fput(file); 2894d18e9008SMiklos Szeredi 2895d18e9008SMiklos Szeredi out: 2896d18e9008SMiklos Szeredi dput(dentry); 28972675a4ebSAl Viro return error; 2898d18e9008SMiklos Szeredi 2899d18e9008SMiklos Szeredi no_open: 2900d18e9008SMiklos Szeredi if (need_lookup) { 290172bd866aSAl Viro dentry = lookup_real(dir, dentry, nd->flags); 2902d18e9008SMiklos Szeredi if (IS_ERR(dentry)) 29032675a4ebSAl Viro return PTR_ERR(dentry); 2904d18e9008SMiklos Szeredi 2905d18e9008SMiklos Szeredi if (create_error) { 2906d18e9008SMiklos Szeredi int open_flag = op->open_flag; 2907d18e9008SMiklos Szeredi 29082675a4ebSAl Viro error = create_error; 2909d18e9008SMiklos Szeredi if ((open_flag & O_EXCL)) { 2910d18e9008SMiklos Szeredi if (!dentry->d_inode) 2911d18e9008SMiklos Szeredi goto out; 2912d18e9008SMiklos Szeredi } else if (!dentry->d_inode) { 2913d18e9008SMiklos Szeredi goto out; 2914d18e9008SMiklos Szeredi } else if ((open_flag & O_TRUNC) && 2915e36cb0b8SDavid Howells d_is_reg(dentry)) { 2916d18e9008SMiklos Szeredi goto out; 2917d18e9008SMiklos Szeredi } 2918d18e9008SMiklos Szeredi /* will fail later, go on to get the right error */ 2919d18e9008SMiklos Szeredi } 2920d18e9008SMiklos Szeredi } 2921d18e9008SMiklos Szeredi looked_up: 2922d18e9008SMiklos Szeredi path->dentry = dentry; 2923d18e9008SMiklos Szeredi path->mnt = nd->path.mnt; 29242675a4ebSAl Viro return 1; 2925d18e9008SMiklos Szeredi } 2926d18e9008SMiklos Szeredi 292731e6b01fSNick Piggin /* 29281acf0af9SDavid Howells * Look up and maybe create and open the last component. 2929d58ffd35SMiklos Szeredi * 2930d58ffd35SMiklos Szeredi * Must be called with i_mutex held on parent. 2931d58ffd35SMiklos Szeredi * 29321acf0af9SDavid Howells * Returns 0 if the file was successfully atomically created (if necessary) and 29331acf0af9SDavid Howells * opened. In this case the file will be returned attached to @file. 29341acf0af9SDavid Howells * 29351acf0af9SDavid Howells * Returns 1 if the file was not completely opened at this time, though lookups 29361acf0af9SDavid Howells * and creations will have been performed and the dentry returned in @path will 29371acf0af9SDavid Howells * be positive upon return if O_CREAT was specified. If O_CREAT wasn't 29381acf0af9SDavid Howells * specified then a negative dentry may be returned. 29391acf0af9SDavid Howells * 29401acf0af9SDavid Howells * An error code is returned otherwise. 29411acf0af9SDavid Howells * 29421acf0af9SDavid Howells * FILE_CREATE will be set in @*opened if the dentry was created and will be 29431acf0af9SDavid Howells * cleared otherwise prior to returning. 2944d58ffd35SMiklos Szeredi */ 29452675a4ebSAl Viro static int lookup_open(struct nameidata *nd, struct path *path, 294630d90494SAl Viro struct file *file, 2947d58ffd35SMiklos Szeredi const struct open_flags *op, 294864894cf8SAl Viro bool got_write, int *opened) 2949d58ffd35SMiklos Szeredi { 2950d58ffd35SMiklos Szeredi struct dentry *dir = nd->path.dentry; 295154ef4872SMiklos Szeredi struct inode *dir_inode = dir->d_inode; 2952d58ffd35SMiklos Szeredi struct dentry *dentry; 2953d58ffd35SMiklos Szeredi int error; 295454ef4872SMiklos Szeredi bool need_lookup; 2955d58ffd35SMiklos Szeredi 295647237687SAl Viro *opened &= ~FILE_CREATED; 2957201f956eSAl Viro dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup); 2958d58ffd35SMiklos Szeredi if (IS_ERR(dentry)) 29592675a4ebSAl Viro return PTR_ERR(dentry); 2960d58ffd35SMiklos Szeredi 2961d18e9008SMiklos Szeredi /* Cached positive dentry: will open in f_op->open */ 2962d18e9008SMiklos Szeredi if (!need_lookup && dentry->d_inode) 2963d18e9008SMiklos Szeredi goto out_no_open; 2964d18e9008SMiklos Szeredi 2965d18e9008SMiklos Szeredi if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) { 296664894cf8SAl Viro return atomic_open(nd, dentry, path, file, op, got_write, 296747237687SAl Viro need_lookup, opened); 2968d18e9008SMiklos Szeredi } 2969d18e9008SMiklos Szeredi 297054ef4872SMiklos Szeredi if (need_lookup) { 297154ef4872SMiklos Szeredi BUG_ON(dentry->d_inode); 297254ef4872SMiklos Szeredi 297372bd866aSAl Viro dentry = lookup_real(dir_inode, dentry, nd->flags); 297454ef4872SMiklos Szeredi if (IS_ERR(dentry)) 29752675a4ebSAl Viro return PTR_ERR(dentry); 297654ef4872SMiklos Szeredi } 297754ef4872SMiklos Szeredi 2978d58ffd35SMiklos Szeredi /* Negative dentry, just create the file */ 2979d58ffd35SMiklos Szeredi if (!dentry->d_inode && (op->open_flag & O_CREAT)) { 2980d58ffd35SMiklos Szeredi umode_t mode = op->mode; 2981d58ffd35SMiklos Szeredi if (!IS_POSIXACL(dir->d_inode)) 2982d58ffd35SMiklos Szeredi mode &= ~current_umask(); 2983d58ffd35SMiklos Szeredi /* 2984d58ffd35SMiklos Szeredi * This write is needed to ensure that a 2985d58ffd35SMiklos Szeredi * rw->ro transition does not occur between 2986d58ffd35SMiklos Szeredi * the time when the file is created and when 2987d58ffd35SMiklos Szeredi * a permanent write count is taken through 2988015c3bbcSMiklos Szeredi * the 'struct file' in finish_open(). 2989d58ffd35SMiklos Szeredi */ 299064894cf8SAl Viro if (!got_write) { 299164894cf8SAl Viro error = -EROFS; 2992d58ffd35SMiklos Szeredi goto out_dput; 299364894cf8SAl Viro } 299447237687SAl Viro *opened |= FILE_CREATED; 2995d58ffd35SMiklos Szeredi error = security_path_mknod(&nd->path, dentry, mode, 0); 2996d58ffd35SMiklos Szeredi if (error) 2997d58ffd35SMiklos Szeredi goto out_dput; 2998312b63fbSAl Viro error = vfs_create(dir->d_inode, dentry, mode, 2999312b63fbSAl Viro nd->flags & LOOKUP_EXCL); 3000d58ffd35SMiklos Szeredi if (error) 3001d58ffd35SMiklos Szeredi goto out_dput; 3002d58ffd35SMiklos Szeredi } 3003d18e9008SMiklos Szeredi out_no_open: 3004d58ffd35SMiklos Szeredi path->dentry = dentry; 3005d58ffd35SMiklos Szeredi path->mnt = nd->path.mnt; 30062675a4ebSAl Viro return 1; 3007d58ffd35SMiklos Szeredi 3008d58ffd35SMiklos Szeredi out_dput: 3009d58ffd35SMiklos Szeredi dput(dentry); 30102675a4ebSAl Viro return error; 3011d58ffd35SMiklos Szeredi } 3012d58ffd35SMiklos Szeredi 3013d58ffd35SMiklos Szeredi /* 3014fe2d35ffSAl Viro * Handle the last step of open() 301531e6b01fSNick Piggin */ 3016896475d5SAl Viro static int do_last(struct nameidata *nd, 301730d90494SAl Viro struct file *file, const struct open_flags *op, 301876ae2a5aSAl Viro int *opened) 3019fb1cc555SAl Viro { 3020a1e28038SAl Viro struct dentry *dir = nd->path.dentry; 3021ca344a89SAl Viro int open_flag = op->open_flag; 302277d660a8SMiklos Szeredi bool will_truncate = (open_flag & O_TRUNC) != 0; 302364894cf8SAl Viro bool got_write = false; 3024bcda7652SAl Viro int acc_mode = op->acc_mode; 3025254cf582SAl Viro unsigned seq; 3026a1eb3315SMiklos Szeredi struct inode *inode; 302716b1c1cdSMiklos Szeredi struct path save_parent = { .dentry = NULL, .mnt = NULL }; 3028896475d5SAl Viro struct path path; 302916b1c1cdSMiklos Szeredi bool retried = false; 303016c2cd71SAl Viro int error; 3031fb1cc555SAl Viro 3032c3e380b0SAl Viro nd->flags &= ~LOOKUP_PARENT; 3033c3e380b0SAl Viro nd->flags |= op->intent; 3034c3e380b0SAl Viro 3035bc77daa7SAl Viro if (nd->last_type != LAST_NORM) { 3036fe2d35ffSAl Viro error = handle_dots(nd, nd->last_type); 3037deb106c6SAl Viro if (unlikely(error)) 30382675a4ebSAl Viro return error; 3039e83db167SMiklos Szeredi goto finish_open; 30401f36f774SAl Viro } 3041a2c36b45SAl Viro 3042ca344a89SAl Viro if (!(open_flag & O_CREAT)) { 3043fe2d35ffSAl Viro if (nd->last.name[nd->last.len]) 3044fe2d35ffSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 3045fe2d35ffSAl Viro /* we _can_ be in RCU mode here */ 3046254cf582SAl Viro error = lookup_fast(nd, &path, &inode, &seq); 304771574865SMiklos Szeredi if (likely(!error)) 304871574865SMiklos Szeredi goto finish_lookup; 304971574865SMiklos Szeredi 3050ce57dfc1SAl Viro if (error < 0) 3051deb106c6SAl Viro return error; 3052a1eb3315SMiklos Szeredi 305337d7fffcSMiklos Szeredi BUG_ON(nd->inode != dir->d_inode); 3054b6183df7SMiklos Szeredi } else { 3055fe2d35ffSAl Viro /* create side of things */ 3056a3fbbde7SAl Viro /* 3057b6183df7SMiklos Szeredi * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED 3058b6183df7SMiklos Szeredi * has been cleared when we got to the last component we are 3059b6183df7SMiklos Szeredi * about to look up 3060a3fbbde7SAl Viro */ 30619f1fafeeSAl Viro error = complete_walk(nd); 3062e8bb73dfSAl Viro if (error) 30632675a4ebSAl Viro return error; 3064fe2d35ffSAl Viro 306576ae2a5aSAl Viro audit_inode(nd->name, dir, LOOKUP_PARENT); 30661f36f774SAl Viro /* trailing slashes? */ 3067deb106c6SAl Viro if (unlikely(nd->last.name[nd->last.len])) 3068deb106c6SAl Viro return -EISDIR; 3069b6183df7SMiklos Szeredi } 30701f36f774SAl Viro 307116b1c1cdSMiklos Szeredi retry_lookup: 307264894cf8SAl Viro if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { 307364894cf8SAl Viro error = mnt_want_write(nd->path.mnt); 307464894cf8SAl Viro if (!error) 307564894cf8SAl Viro got_write = true; 307664894cf8SAl Viro /* 307764894cf8SAl Viro * do _not_ fail yet - we might not need that or fail with 307864894cf8SAl Viro * a different error; let lookup_open() decide; we'll be 307964894cf8SAl Viro * dropping this one anyway. 308064894cf8SAl Viro */ 308164894cf8SAl Viro } 3082a1e28038SAl Viro mutex_lock(&dir->d_inode->i_mutex); 3083896475d5SAl Viro error = lookup_open(nd, &path, file, op, got_write, opened); 3084fb1cc555SAl Viro mutex_unlock(&dir->d_inode->i_mutex); 3085fb1cc555SAl Viro 30862675a4ebSAl Viro if (error <= 0) { 30872675a4ebSAl Viro if (error) 3088d58ffd35SMiklos Szeredi goto out; 30896c0d46c4SAl Viro 309047237687SAl Viro if ((*opened & FILE_CREATED) || 3091496ad9aaSAl Viro !S_ISREG(file_inode(file)->i_mode)) 309277d660a8SMiklos Szeredi will_truncate = false; 3093d18e9008SMiklos Szeredi 309476ae2a5aSAl Viro audit_inode(nd->name, file->f_path.dentry, 0); 3095d18e9008SMiklos Szeredi goto opened; 3096d18e9008SMiklos Szeredi } 3097d18e9008SMiklos Szeredi 309847237687SAl Viro if (*opened & FILE_CREATED) { 30999b44f1b3SAl Viro /* Don't check for write permission, don't truncate */ 3100ca344a89SAl Viro open_flag &= ~O_TRUNC; 310177d660a8SMiklos Szeredi will_truncate = false; 3102bcda7652SAl Viro acc_mode = MAY_OPEN; 3103896475d5SAl Viro path_to_nameidata(&path, nd); 3104e83db167SMiklos Szeredi goto finish_open_created; 3105fb1cc555SAl Viro } 3106fb1cc555SAl Viro 3107fb1cc555SAl Viro /* 31083134f37eSJeff Layton * create/update audit record if it already exists. 3109fb1cc555SAl Viro */ 3110896475d5SAl Viro if (d_is_positive(path.dentry)) 311176ae2a5aSAl Viro audit_inode(nd->name, path.dentry, 0); 3112fb1cc555SAl Viro 3113d18e9008SMiklos Szeredi /* 3114d18e9008SMiklos Szeredi * If atomic_open() acquired write access it is dropped now due to 3115d18e9008SMiklos Szeredi * possible mount and symlink following (this might be optimized away if 3116d18e9008SMiklos Szeredi * necessary...) 3117d18e9008SMiklos Szeredi */ 311864894cf8SAl Viro if (got_write) { 3119d18e9008SMiklos Szeredi mnt_drop_write(nd->path.mnt); 312064894cf8SAl Viro got_write = false; 3121d18e9008SMiklos Szeredi } 3122d18e9008SMiklos Szeredi 3123deb106c6SAl Viro if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) { 3124deb106c6SAl Viro path_to_nameidata(&path, nd); 3125deb106c6SAl Viro return -EEXIST; 3126deb106c6SAl Viro } 3127fb1cc555SAl Viro 3128756daf26SNeilBrown error = follow_managed(&path, nd); 3129deb106c6SAl Viro if (unlikely(error < 0)) 3130deb106c6SAl Viro return error; 3131a3fbbde7SAl Viro 3132decf3400SMiklos Szeredi BUG_ON(nd->flags & LOOKUP_RCU); 313363afdfc7SDavid Howells inode = d_backing_inode(path.dentry); 3134254cf582SAl Viro seq = 0; /* out of RCU mode, so the value doesn't matter */ 3135deb106c6SAl Viro if (unlikely(d_is_negative(path.dentry))) { 3136896475d5SAl Viro path_to_nameidata(&path, nd); 3137deb106c6SAl Viro return -ENOENT; 313854c33e7fSMiklos Szeredi } 3139766c4cbfSAl Viro finish_lookup: 3140191d7f73SAl Viro if (nd->depth) 3141191d7f73SAl Viro put_link(nd); 3142181548c0SAl Viro error = should_follow_link(nd, &path, nd->flags & LOOKUP_FOLLOW, 3143181548c0SAl Viro inode, seq); 3144deb106c6SAl Viro if (unlikely(error)) 3145d63ff28fSAl Viro return error; 3146fb1cc555SAl Viro 3147896475d5SAl Viro if (unlikely(d_is_symlink(path.dentry)) && !(open_flag & O_PATH)) { 3148896475d5SAl Viro path_to_nameidata(&path, nd); 3149deb106c6SAl Viro return -ELOOP; 3150a5cfe2d5SAl Viro } 3151a5cfe2d5SAl Viro 3152896475d5SAl Viro if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path.mnt) { 3153896475d5SAl Viro path_to_nameidata(&path, nd); 315416b1c1cdSMiklos Szeredi } else { 315516b1c1cdSMiklos Szeredi save_parent.dentry = nd->path.dentry; 3156896475d5SAl Viro save_parent.mnt = mntget(path.mnt); 3157896475d5SAl Viro nd->path.dentry = path.dentry; 315816b1c1cdSMiklos Szeredi 315916b1c1cdSMiklos Szeredi } 3160decf3400SMiklos Szeredi nd->inode = inode; 3161254cf582SAl Viro nd->seq = seq; 3162a3fbbde7SAl Viro /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ 3163bc77daa7SAl Viro finish_open: 3164a3fbbde7SAl Viro error = complete_walk(nd); 316516b1c1cdSMiklos Szeredi if (error) { 316616b1c1cdSMiklos Szeredi path_put(&save_parent); 31672675a4ebSAl Viro return error; 316816b1c1cdSMiklos Szeredi } 316976ae2a5aSAl Viro audit_inode(nd->name, nd->path.dentry, 0); 3170fb1cc555SAl Viro error = -EISDIR; 317144b1d530SMiklos Szeredi if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry)) 31722675a4ebSAl Viro goto out; 3173af2f5542SMiklos Szeredi error = -ENOTDIR; 317444b1d530SMiklos Szeredi if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry)) 31752675a4ebSAl Viro goto out; 31764bbcbd3bSDavid Howells if (!d_is_reg(nd->path.dentry)) 317777d660a8SMiklos Szeredi will_truncate = false; 31786c0d46c4SAl Viro 31790f9d1a10SAl Viro if (will_truncate) { 31800f9d1a10SAl Viro error = mnt_want_write(nd->path.mnt); 31810f9d1a10SAl Viro if (error) 31822675a4ebSAl Viro goto out; 318364894cf8SAl Viro got_write = true; 31840f9d1a10SAl Viro } 3185e83db167SMiklos Szeredi finish_open_created: 3186bcda7652SAl Viro error = may_open(&nd->path, acc_mode, open_flag); 3187ca344a89SAl Viro if (error) 31882675a4ebSAl Viro goto out; 31894aa7c634SMiklos Szeredi 31904aa7c634SMiklos Szeredi BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */ 31914aa7c634SMiklos Szeredi error = vfs_open(&nd->path, file, current_cred()); 31924aa7c634SMiklos Szeredi if (!error) { 31934aa7c634SMiklos Szeredi *opened |= FILE_OPENED; 31944aa7c634SMiklos Szeredi } else { 319530d90494SAl Viro if (error == -EOPENSTALE) 3196f60dc3dbSMiklos Szeredi goto stale_open; 3197015c3bbcSMiklos Szeredi goto out; 3198f60dc3dbSMiklos Szeredi } 3199a8277b9bSMiklos Szeredi opened: 32002675a4ebSAl Viro error = open_check_o_direct(file); 3201015c3bbcSMiklos Szeredi if (error) 3202015c3bbcSMiklos Szeredi goto exit_fput; 32033034a146SDmitry Kasatkin error = ima_file_check(file, op->acc_mode, *opened); 3204aa4caadbSMiklos Szeredi if (error) 3205aa4caadbSMiklos Szeredi goto exit_fput; 3206aa4caadbSMiklos Szeredi 32070f9d1a10SAl Viro if (will_truncate) { 32082675a4ebSAl Viro error = handle_truncate(file); 3209aa4caadbSMiklos Szeredi if (error) 3210aa4caadbSMiklos Szeredi goto exit_fput; 32110f9d1a10SAl Viro } 3212ca344a89SAl Viro out: 321364894cf8SAl Viro if (got_write) 32140f9d1a10SAl Viro mnt_drop_write(nd->path.mnt); 321516b1c1cdSMiklos Szeredi path_put(&save_parent); 32162675a4ebSAl Viro return error; 3217fb1cc555SAl Viro 3218015c3bbcSMiklos Szeredi exit_fput: 32192675a4ebSAl Viro fput(file); 32202675a4ebSAl Viro goto out; 3221015c3bbcSMiklos Szeredi 3222f60dc3dbSMiklos Szeredi stale_open: 3223f60dc3dbSMiklos Szeredi /* If no saved parent or already retried then can't retry */ 3224f60dc3dbSMiklos Szeredi if (!save_parent.dentry || retried) 3225f60dc3dbSMiklos Szeredi goto out; 3226f60dc3dbSMiklos Szeredi 3227f60dc3dbSMiklos Szeredi BUG_ON(save_parent.dentry != dir); 3228f60dc3dbSMiklos Szeredi path_put(&nd->path); 3229f60dc3dbSMiklos Szeredi nd->path = save_parent; 3230f60dc3dbSMiklos Szeredi nd->inode = dir->d_inode; 3231f60dc3dbSMiklos Szeredi save_parent.mnt = NULL; 3232f60dc3dbSMiklos Szeredi save_parent.dentry = NULL; 323364894cf8SAl Viro if (got_write) { 3234f60dc3dbSMiklos Szeredi mnt_drop_write(nd->path.mnt); 323564894cf8SAl Viro got_write = false; 3236f60dc3dbSMiklos Szeredi } 3237f60dc3dbSMiklos Szeredi retried = true; 3238f60dc3dbSMiklos Szeredi goto retry_lookup; 3239fb1cc555SAl Viro } 3240fb1cc555SAl Viro 3241c8a53ee5SAl Viro static int do_tmpfile(struct nameidata *nd, unsigned flags, 324260545d0dSAl Viro const struct open_flags *op, 324360545d0dSAl Viro struct file *file, int *opened) 324460545d0dSAl Viro { 324560545d0dSAl Viro static const struct qstr name = QSTR_INIT("/", 1); 3246625b6d10SAl Viro struct dentry *child; 324760545d0dSAl Viro struct inode *dir; 3248625b6d10SAl Viro struct path path; 3249c8a53ee5SAl Viro int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path); 325060545d0dSAl Viro if (unlikely(error)) 325160545d0dSAl Viro return error; 3252625b6d10SAl Viro error = mnt_want_write(path.mnt); 325360545d0dSAl Viro if (unlikely(error)) 325460545d0dSAl Viro goto out; 3255625b6d10SAl Viro dir = path.dentry->d_inode; 325660545d0dSAl Viro /* we want directory to be writable */ 3257625b6d10SAl Viro error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 325860545d0dSAl Viro if (error) 325960545d0dSAl Viro goto out2; 326060545d0dSAl Viro if (!dir->i_op->tmpfile) { 326160545d0dSAl Viro error = -EOPNOTSUPP; 326260545d0dSAl Viro goto out2; 326360545d0dSAl Viro } 3264625b6d10SAl Viro child = d_alloc(path.dentry, &name); 326560545d0dSAl Viro if (unlikely(!child)) { 326660545d0dSAl Viro error = -ENOMEM; 326760545d0dSAl Viro goto out2; 326860545d0dSAl Viro } 3269625b6d10SAl Viro dput(path.dentry); 3270625b6d10SAl Viro path.dentry = child; 3271625b6d10SAl Viro error = dir->i_op->tmpfile(dir, child, op->mode); 327260545d0dSAl Viro if (error) 327360545d0dSAl Viro goto out2; 3274c8a53ee5SAl Viro audit_inode(nd->name, child, 0); 327569a91c23SEric Rannaud /* Don't check for other permissions, the inode was just created */ 3276625b6d10SAl Viro error = may_open(&path, MAY_OPEN, op->open_flag); 327760545d0dSAl Viro if (error) 327860545d0dSAl Viro goto out2; 3279625b6d10SAl Viro file->f_path.mnt = path.mnt; 3280625b6d10SAl Viro error = finish_open(file, child, NULL, opened); 328160545d0dSAl Viro if (error) 328260545d0dSAl Viro goto out2; 328360545d0dSAl Viro error = open_check_o_direct(file); 3284f4e0c30cSAl Viro if (error) { 328560545d0dSAl Viro fput(file); 3286f4e0c30cSAl Viro } else if (!(op->open_flag & O_EXCL)) { 3287f4e0c30cSAl Viro struct inode *inode = file_inode(file); 3288f4e0c30cSAl Viro spin_lock(&inode->i_lock); 3289f4e0c30cSAl Viro inode->i_state |= I_LINKABLE; 3290f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 3291f4e0c30cSAl Viro } 329260545d0dSAl Viro out2: 3293625b6d10SAl Viro mnt_drop_write(path.mnt); 329460545d0dSAl Viro out: 3295625b6d10SAl Viro path_put(&path); 329660545d0dSAl Viro return error; 329760545d0dSAl Viro } 329860545d0dSAl Viro 3299c8a53ee5SAl Viro static struct file *path_openat(struct nameidata *nd, 3300c8a53ee5SAl Viro const struct open_flags *op, unsigned flags) 33011da177e4SLinus Torvalds { 3302368ee9baSAl Viro const char *s; 330330d90494SAl Viro struct file *file; 330447237687SAl Viro int opened = 0; 330513aab428SAl Viro int error; 330631e6b01fSNick Piggin 330730d90494SAl Viro file = get_empty_filp(); 33081afc99beSAl Viro if (IS_ERR(file)) 33091afc99beSAl Viro return file; 331031e6b01fSNick Piggin 331130d90494SAl Viro file->f_flags = op->open_flag; 331231e6b01fSNick Piggin 3313bb458c64SAl Viro if (unlikely(file->f_flags & __O_TMPFILE)) { 3314c8a53ee5SAl Viro error = do_tmpfile(nd, flags, op, file, &opened); 3315f15133dfSAl Viro goto out2; 331660545d0dSAl Viro } 331760545d0dSAl Viro 3318c8a53ee5SAl Viro s = path_init(nd, flags); 3319368ee9baSAl Viro if (IS_ERR(s)) { 3320368ee9baSAl Viro put_filp(file); 3321368ee9baSAl Viro return ERR_CAST(s); 3322368ee9baSAl Viro } 33233bdba28bSAl Viro while (!(error = link_path_walk(s, nd)) && 332476ae2a5aSAl Viro (error = do_last(nd, file, op, &opened)) > 0) { 332573d049a4SAl Viro nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); 33263bdba28bSAl Viro s = trailing_symlink(nd); 33273bdba28bSAl Viro if (IS_ERR(s)) { 33283bdba28bSAl Viro error = PTR_ERR(s); 33292675a4ebSAl Viro break; 3330806b681cSAl Viro } 33313bdba28bSAl Viro } 3332deb106c6SAl Viro terminate_walk(nd); 3333f15133dfSAl Viro out2: 33342675a4ebSAl Viro if (!(opened & FILE_OPENED)) { 33352675a4ebSAl Viro BUG_ON(!error); 333630d90494SAl Viro put_filp(file); 3337015c3bbcSMiklos Szeredi } 33382675a4ebSAl Viro if (unlikely(error)) { 33392675a4ebSAl Viro if (error == -EOPENSTALE) { 33402675a4ebSAl Viro if (flags & LOOKUP_RCU) 33412675a4ebSAl Viro error = -ECHILD; 33422675a4ebSAl Viro else 33432675a4ebSAl Viro error = -ESTALE; 33442675a4ebSAl Viro } 33452675a4ebSAl Viro file = ERR_PTR(error); 33462675a4ebSAl Viro } 33472675a4ebSAl Viro return file; 3348de459215SKirill Korotaev } 33491da177e4SLinus Torvalds 3350669abf4eSJeff Layton struct file *do_filp_open(int dfd, struct filename *pathname, 3351f9652e10SAl Viro const struct open_flags *op) 335213aab428SAl Viro { 33539883d185SAl Viro struct nameidata nd; 3354f9652e10SAl Viro int flags = op->lookup_flags; 335513aab428SAl Viro struct file *filp; 335613aab428SAl Viro 33579883d185SAl Viro set_nameidata(&nd, dfd, pathname); 3358c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_RCU); 335913aab428SAl Viro if (unlikely(filp == ERR_PTR(-ECHILD))) 3360c8a53ee5SAl Viro filp = path_openat(&nd, op, flags); 336113aab428SAl Viro if (unlikely(filp == ERR_PTR(-ESTALE))) 3362c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_REVAL); 33639883d185SAl Viro restore_nameidata(); 336413aab428SAl Viro return filp; 336513aab428SAl Viro } 336613aab428SAl Viro 336773d049a4SAl Viro struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, 3368f9652e10SAl Viro const char *name, const struct open_flags *op) 336973d049a4SAl Viro { 33709883d185SAl Viro struct nameidata nd; 337173d049a4SAl Viro struct file *file; 337251689104SPaul Moore struct filename *filename; 3373f9652e10SAl Viro int flags = op->lookup_flags | LOOKUP_ROOT; 337473d049a4SAl Viro 337573d049a4SAl Viro nd.root.mnt = mnt; 337673d049a4SAl Viro nd.root.dentry = dentry; 337773d049a4SAl Viro 3378b18825a7SDavid Howells if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN) 337973d049a4SAl Viro return ERR_PTR(-ELOOP); 338073d049a4SAl Viro 338151689104SPaul Moore filename = getname_kernel(name); 3382a1c83681SViresh Kumar if (IS_ERR(filename)) 338351689104SPaul Moore return ERR_CAST(filename); 338451689104SPaul Moore 33859883d185SAl Viro set_nameidata(&nd, -1, filename); 3386c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_RCU); 338773d049a4SAl Viro if (unlikely(file == ERR_PTR(-ECHILD))) 3388c8a53ee5SAl Viro file = path_openat(&nd, op, flags); 338973d049a4SAl Viro if (unlikely(file == ERR_PTR(-ESTALE))) 3390c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_REVAL); 33919883d185SAl Viro restore_nameidata(); 339251689104SPaul Moore putname(filename); 339373d049a4SAl Viro return file; 339473d049a4SAl Viro } 339573d049a4SAl Viro 3396fa14a0b8SAl Viro static struct dentry *filename_create(int dfd, struct filename *name, 33971ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 33981da177e4SLinus Torvalds { 3399c663e5d8SChristoph Hellwig struct dentry *dentry = ERR_PTR(-EEXIST); 3400391172c4SAl Viro struct qstr last; 3401391172c4SAl Viro int type; 3402c30dabfeSJan Kara int err2; 34031ac12b4bSJeff Layton int error; 34041ac12b4bSJeff Layton bool is_dir = (lookup_flags & LOOKUP_DIRECTORY); 34051ac12b4bSJeff Layton 34061ac12b4bSJeff Layton /* 34071ac12b4bSJeff Layton * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any 34081ac12b4bSJeff Layton * other flags passed in are ignored! 34091ac12b4bSJeff Layton */ 34101ac12b4bSJeff Layton lookup_flags &= LOOKUP_REVAL; 34111ac12b4bSJeff Layton 34125c31b6ceSAl Viro name = filename_parentat(dfd, name, lookup_flags, path, &last, &type); 34135c31b6ceSAl Viro if (IS_ERR(name)) 34145c31b6ceSAl Viro return ERR_CAST(name); 34151da177e4SLinus Torvalds 3416c663e5d8SChristoph Hellwig /* 3417c663e5d8SChristoph Hellwig * Yucky last component or no last component at all? 3418c663e5d8SChristoph Hellwig * (foo/., foo/.., /////) 3419c663e5d8SChristoph Hellwig */ 34205c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) 3421ed75e95dSAl Viro goto out; 3422c663e5d8SChristoph Hellwig 3423c30dabfeSJan Kara /* don't fail immediately if it's r/o, at least try to report other errors */ 3424391172c4SAl Viro err2 = mnt_want_write(path->mnt); 3425c663e5d8SChristoph Hellwig /* 3426c663e5d8SChristoph Hellwig * Do the final lookup. 3427c663e5d8SChristoph Hellwig */ 3428391172c4SAl Viro lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL; 3429391172c4SAl Viro mutex_lock_nested(&path->dentry->d_inode->i_mutex, I_MUTEX_PARENT); 3430391172c4SAl Viro dentry = __lookup_hash(&last, path->dentry, lookup_flags); 34311da177e4SLinus Torvalds if (IS_ERR(dentry)) 3432a8104a9fSAl Viro goto unlock; 3433c663e5d8SChristoph Hellwig 3434a8104a9fSAl Viro error = -EEXIST; 3435b18825a7SDavid Howells if (d_is_positive(dentry)) 3436a8104a9fSAl Viro goto fail; 3437b18825a7SDavid Howells 3438c663e5d8SChristoph Hellwig /* 3439c663e5d8SChristoph Hellwig * Special case - lookup gave negative, but... we had foo/bar/ 3440c663e5d8SChristoph Hellwig * From the vfs_mknod() POV we just have a negative dentry - 3441c663e5d8SChristoph Hellwig * all is fine. Let's be bastards - you had / on the end, you've 3442c663e5d8SChristoph Hellwig * been asking for (non-existent) directory. -ENOENT for you. 3443c663e5d8SChristoph Hellwig */ 3444391172c4SAl Viro if (unlikely(!is_dir && last.name[last.len])) { 3445a8104a9fSAl Viro error = -ENOENT; 3446ed75e95dSAl Viro goto fail; 3447e9baf6e5SAl Viro } 3448c30dabfeSJan Kara if (unlikely(err2)) { 3449c30dabfeSJan Kara error = err2; 3450a8104a9fSAl Viro goto fail; 3451c30dabfeSJan Kara } 3452181c37b6SAl Viro putname(name); 3453e9baf6e5SAl Viro return dentry; 34541da177e4SLinus Torvalds fail: 3455a8104a9fSAl Viro dput(dentry); 3456a8104a9fSAl Viro dentry = ERR_PTR(error); 3457a8104a9fSAl Viro unlock: 3458391172c4SAl Viro mutex_unlock(&path->dentry->d_inode->i_mutex); 3459c30dabfeSJan Kara if (!err2) 3460391172c4SAl Viro mnt_drop_write(path->mnt); 3461ed75e95dSAl Viro out: 3462391172c4SAl Viro path_put(path); 3463181c37b6SAl Viro putname(name); 3464ed75e95dSAl Viro return dentry; 3465dae6ad8fSAl Viro } 3466fa14a0b8SAl Viro 3467fa14a0b8SAl Viro struct dentry *kern_path_create(int dfd, const char *pathname, 3468fa14a0b8SAl Viro struct path *path, unsigned int lookup_flags) 3469fa14a0b8SAl Viro { 3470181c37b6SAl Viro return filename_create(dfd, getname_kernel(pathname), 3471181c37b6SAl Viro path, lookup_flags); 3472fa14a0b8SAl Viro } 3473dae6ad8fSAl Viro EXPORT_SYMBOL(kern_path_create); 3474dae6ad8fSAl Viro 3475921a1650SAl Viro void done_path_create(struct path *path, struct dentry *dentry) 3476921a1650SAl Viro { 3477921a1650SAl Viro dput(dentry); 3478921a1650SAl Viro mutex_unlock(&path->dentry->d_inode->i_mutex); 3479a8104a9fSAl Viro mnt_drop_write(path->mnt); 3480921a1650SAl Viro path_put(path); 3481921a1650SAl Viro } 3482921a1650SAl Viro EXPORT_SYMBOL(done_path_create); 3483921a1650SAl Viro 3484520ae687SAl Viro inline struct dentry *user_path_create(int dfd, const char __user *pathname, 34851ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 3486dae6ad8fSAl Viro { 3487181c37b6SAl Viro return filename_create(dfd, getname(pathname), path, lookup_flags); 3488dae6ad8fSAl Viro } 3489dae6ad8fSAl Viro EXPORT_SYMBOL(user_path_create); 3490dae6ad8fSAl Viro 34911a67aafbSAl Viro int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) 34921da177e4SLinus Torvalds { 3493a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 34941da177e4SLinus Torvalds 34951da177e4SLinus Torvalds if (error) 34961da177e4SLinus Torvalds return error; 34971da177e4SLinus Torvalds 3498975d6b39SEric W. Biederman if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) 34991da177e4SLinus Torvalds return -EPERM; 35001da177e4SLinus Torvalds 3501acfa4380SAl Viro if (!dir->i_op->mknod) 35021da177e4SLinus Torvalds return -EPERM; 35031da177e4SLinus Torvalds 350408ce5f16SSerge E. Hallyn error = devcgroup_inode_mknod(mode, dev); 350508ce5f16SSerge E. Hallyn if (error) 350608ce5f16SSerge E. Hallyn return error; 350708ce5f16SSerge E. Hallyn 35081da177e4SLinus Torvalds error = security_inode_mknod(dir, dentry, mode, dev); 35091da177e4SLinus Torvalds if (error) 35101da177e4SLinus Torvalds return error; 35111da177e4SLinus Torvalds 35121da177e4SLinus Torvalds error = dir->i_op->mknod(dir, dentry, mode, dev); 3513a74574aaSStephen Smalley if (!error) 3514f38aa942SAmy Griffis fsnotify_create(dir, dentry); 35151da177e4SLinus Torvalds return error; 35161da177e4SLinus Torvalds } 35174d359507SAl Viro EXPORT_SYMBOL(vfs_mknod); 35181da177e4SLinus Torvalds 3519f69aac00SAl Viro static int may_mknod(umode_t mode) 3520463c3197SDave Hansen { 3521463c3197SDave Hansen switch (mode & S_IFMT) { 3522463c3197SDave Hansen case S_IFREG: 3523463c3197SDave Hansen case S_IFCHR: 3524463c3197SDave Hansen case S_IFBLK: 3525463c3197SDave Hansen case S_IFIFO: 3526463c3197SDave Hansen case S_IFSOCK: 3527463c3197SDave Hansen case 0: /* zero mode translates to S_IFREG */ 3528463c3197SDave Hansen return 0; 3529463c3197SDave Hansen case S_IFDIR: 3530463c3197SDave Hansen return -EPERM; 3531463c3197SDave Hansen default: 3532463c3197SDave Hansen return -EINVAL; 3533463c3197SDave Hansen } 3534463c3197SDave Hansen } 3535463c3197SDave Hansen 35368208a22bSAl Viro SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, 35372e4d0924SHeiko Carstens unsigned, dev) 35381da177e4SLinus Torvalds { 35391da177e4SLinus Torvalds struct dentry *dentry; 3540dae6ad8fSAl Viro struct path path; 3541dae6ad8fSAl Viro int error; 3542972567f1SJeff Layton unsigned int lookup_flags = 0; 35431da177e4SLinus Torvalds 35448e4bfca1SAl Viro error = may_mknod(mode); 35458e4bfca1SAl Viro if (error) 35468e4bfca1SAl Viro return error; 3547972567f1SJeff Layton retry: 3548972567f1SJeff Layton dentry = user_path_create(dfd, filename, &path, lookup_flags); 3549dae6ad8fSAl Viro if (IS_ERR(dentry)) 3550dae6ad8fSAl Viro return PTR_ERR(dentry); 35512ad94ae6SAl Viro 3552dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3553ce3b0f8dSAl Viro mode &= ~current_umask(); 3554dae6ad8fSAl Viro error = security_path_mknod(&path, dentry, mode, dev); 3555be6d3e56SKentaro Takeda if (error) 3556a8104a9fSAl Viro goto out; 35571da177e4SLinus Torvalds switch (mode & S_IFMT) { 35581da177e4SLinus Torvalds case 0: case S_IFREG: 3559312b63fbSAl Viro error = vfs_create(path.dentry->d_inode,dentry,mode,true); 35601da177e4SLinus Torvalds break; 35611da177e4SLinus Torvalds case S_IFCHR: case S_IFBLK: 3562dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode, 35631da177e4SLinus Torvalds new_decode_dev(dev)); 35641da177e4SLinus Torvalds break; 35651da177e4SLinus Torvalds case S_IFIFO: case S_IFSOCK: 3566dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); 35671da177e4SLinus Torvalds break; 35681da177e4SLinus Torvalds } 3569a8104a9fSAl Viro out: 3570921a1650SAl Viro done_path_create(&path, dentry); 3571972567f1SJeff Layton if (retry_estale(error, lookup_flags)) { 3572972567f1SJeff Layton lookup_flags |= LOOKUP_REVAL; 3573972567f1SJeff Layton goto retry; 3574972567f1SJeff Layton } 35751da177e4SLinus Torvalds return error; 35761da177e4SLinus Torvalds } 35771da177e4SLinus Torvalds 35788208a22bSAl Viro SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) 35795590ff0dSUlrich Drepper { 35805590ff0dSUlrich Drepper return sys_mknodat(AT_FDCWD, filename, mode, dev); 35815590ff0dSUlrich Drepper } 35825590ff0dSUlrich Drepper 358318bb1db3SAl Viro int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 35841da177e4SLinus Torvalds { 3585a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 35868de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 35871da177e4SLinus Torvalds 35881da177e4SLinus Torvalds if (error) 35891da177e4SLinus Torvalds return error; 35901da177e4SLinus Torvalds 3591acfa4380SAl Viro if (!dir->i_op->mkdir) 35921da177e4SLinus Torvalds return -EPERM; 35931da177e4SLinus Torvalds 35941da177e4SLinus Torvalds mode &= (S_IRWXUGO|S_ISVTX); 35951da177e4SLinus Torvalds error = security_inode_mkdir(dir, dentry, mode); 35961da177e4SLinus Torvalds if (error) 35971da177e4SLinus Torvalds return error; 35981da177e4SLinus Torvalds 35998de52778SAl Viro if (max_links && dir->i_nlink >= max_links) 36008de52778SAl Viro return -EMLINK; 36018de52778SAl Viro 36021da177e4SLinus Torvalds error = dir->i_op->mkdir(dir, dentry, mode); 3603a74574aaSStephen Smalley if (!error) 3604f38aa942SAmy Griffis fsnotify_mkdir(dir, dentry); 36051da177e4SLinus Torvalds return error; 36061da177e4SLinus Torvalds } 36074d359507SAl Viro EXPORT_SYMBOL(vfs_mkdir); 36081da177e4SLinus Torvalds 3609a218d0fdSAl Viro SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) 36101da177e4SLinus Torvalds { 36116902d925SDave Hansen struct dentry *dentry; 3612dae6ad8fSAl Viro struct path path; 3613dae6ad8fSAl Viro int error; 3614b76d8b82SJeff Layton unsigned int lookup_flags = LOOKUP_DIRECTORY; 36151da177e4SLinus Torvalds 3616b76d8b82SJeff Layton retry: 3617b76d8b82SJeff Layton dentry = user_path_create(dfd, pathname, &path, lookup_flags); 36186902d925SDave Hansen if (IS_ERR(dentry)) 3619dae6ad8fSAl Viro return PTR_ERR(dentry); 36206902d925SDave Hansen 3621dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3622ce3b0f8dSAl Viro mode &= ~current_umask(); 3623dae6ad8fSAl Viro error = security_path_mkdir(&path, dentry, mode); 3624a8104a9fSAl Viro if (!error) 3625dae6ad8fSAl Viro error = vfs_mkdir(path.dentry->d_inode, dentry, mode); 3626921a1650SAl Viro done_path_create(&path, dentry); 3627b76d8b82SJeff Layton if (retry_estale(error, lookup_flags)) { 3628b76d8b82SJeff Layton lookup_flags |= LOOKUP_REVAL; 3629b76d8b82SJeff Layton goto retry; 3630b76d8b82SJeff Layton } 36311da177e4SLinus Torvalds return error; 36321da177e4SLinus Torvalds } 36331da177e4SLinus Torvalds 3634a218d0fdSAl Viro SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) 36355590ff0dSUlrich Drepper { 36365590ff0dSUlrich Drepper return sys_mkdirat(AT_FDCWD, pathname, mode); 36375590ff0dSUlrich Drepper } 36385590ff0dSUlrich Drepper 36391da177e4SLinus Torvalds /* 3640a71905f0SSage Weil * The dentry_unhash() helper will try to drop the dentry early: we 3641c0d02594SJ. Bruce Fields * should have a usage count of 1 if we're the only user of this 3642a71905f0SSage Weil * dentry, and if that is true (possibly after pruning the dcache), 3643a71905f0SSage Weil * then we drop the dentry now. 36441da177e4SLinus Torvalds * 36451da177e4SLinus Torvalds * A low-level filesystem can, if it choses, legally 36461da177e4SLinus Torvalds * do a 36471da177e4SLinus Torvalds * 36481da177e4SLinus Torvalds * if (!d_unhashed(dentry)) 36491da177e4SLinus Torvalds * return -EBUSY; 36501da177e4SLinus Torvalds * 36511da177e4SLinus Torvalds * if it cannot handle the case of removing a directory 36521da177e4SLinus Torvalds * that is still in use by something else.. 36531da177e4SLinus Torvalds */ 36541da177e4SLinus Torvalds void dentry_unhash(struct dentry *dentry) 36551da177e4SLinus Torvalds { 36561da177e4SLinus Torvalds shrink_dcache_parent(dentry); 36571da177e4SLinus Torvalds spin_lock(&dentry->d_lock); 365898474236SWaiman Long if (dentry->d_lockref.count == 1) 36591da177e4SLinus Torvalds __d_drop(dentry); 36601da177e4SLinus Torvalds spin_unlock(&dentry->d_lock); 36611da177e4SLinus Torvalds } 36624d359507SAl Viro EXPORT_SYMBOL(dentry_unhash); 36631da177e4SLinus Torvalds 36641da177e4SLinus Torvalds int vfs_rmdir(struct inode *dir, struct dentry *dentry) 36651da177e4SLinus Torvalds { 36661da177e4SLinus Torvalds int error = may_delete(dir, dentry, 1); 36671da177e4SLinus Torvalds 36681da177e4SLinus Torvalds if (error) 36691da177e4SLinus Torvalds return error; 36701da177e4SLinus Torvalds 3671acfa4380SAl Viro if (!dir->i_op->rmdir) 36721da177e4SLinus Torvalds return -EPERM; 36731da177e4SLinus Torvalds 36741d2ef590SAl Viro dget(dentry); 36751b1dcc1bSJes Sorensen mutex_lock(&dentry->d_inode->i_mutex); 3676912dbc15SSage Weil 36771da177e4SLinus Torvalds error = -EBUSY; 36787af1364fSEric W. Biederman if (is_local_mountpoint(dentry)) 3679912dbc15SSage Weil goto out; 3680912dbc15SSage Weil 36811da177e4SLinus Torvalds error = security_inode_rmdir(dir, dentry); 3682912dbc15SSage Weil if (error) 3683912dbc15SSage Weil goto out; 3684912dbc15SSage Weil 36853cebde24SSage Weil shrink_dcache_parent(dentry); 36861da177e4SLinus Torvalds error = dir->i_op->rmdir(dir, dentry); 3687912dbc15SSage Weil if (error) 3688912dbc15SSage Weil goto out; 3689912dbc15SSage Weil 36901da177e4SLinus Torvalds dentry->d_inode->i_flags |= S_DEAD; 3691d83c49f3SAl Viro dont_mount(dentry); 36928ed936b5SEric W. Biederman detach_mounts(dentry); 36931da177e4SLinus Torvalds 3694912dbc15SSage Weil out: 3695912dbc15SSage Weil mutex_unlock(&dentry->d_inode->i_mutex); 36961d2ef590SAl Viro dput(dentry); 3697912dbc15SSage Weil if (!error) 3698912dbc15SSage Weil d_delete(dentry); 36991da177e4SLinus Torvalds return error; 37001da177e4SLinus Torvalds } 37014d359507SAl Viro EXPORT_SYMBOL(vfs_rmdir); 37021da177e4SLinus Torvalds 37035590ff0dSUlrich Drepper static long do_rmdir(int dfd, const char __user *pathname) 37041da177e4SLinus Torvalds { 37051da177e4SLinus Torvalds int error = 0; 370691a27b2aSJeff Layton struct filename *name; 37071da177e4SLinus Torvalds struct dentry *dentry; 3708f5beed75SAl Viro struct path path; 3709f5beed75SAl Viro struct qstr last; 3710f5beed75SAl Viro int type; 3711c6ee9206SJeff Layton unsigned int lookup_flags = 0; 3712c6ee9206SJeff Layton retry: 3713f5beed75SAl Viro name = user_path_parent(dfd, pathname, 3714f5beed75SAl Viro &path, &last, &type, lookup_flags); 371591a27b2aSJeff Layton if (IS_ERR(name)) 371691a27b2aSJeff Layton return PTR_ERR(name); 37171da177e4SLinus Torvalds 3718f5beed75SAl Viro switch (type) { 37191da177e4SLinus Torvalds case LAST_DOTDOT: 37201da177e4SLinus Torvalds error = -ENOTEMPTY; 37211da177e4SLinus Torvalds goto exit1; 37221da177e4SLinus Torvalds case LAST_DOT: 37231da177e4SLinus Torvalds error = -EINVAL; 37241da177e4SLinus Torvalds goto exit1; 37251da177e4SLinus Torvalds case LAST_ROOT: 37261da177e4SLinus Torvalds error = -EBUSY; 37271da177e4SLinus Torvalds goto exit1; 37281da177e4SLinus Torvalds } 37290612d9fbSOGAWA Hirofumi 3730f5beed75SAl Viro error = mnt_want_write(path.mnt); 3731c30dabfeSJan Kara if (error) 3732c30dabfeSJan Kara goto exit1; 37330612d9fbSOGAWA Hirofumi 3734f5beed75SAl Viro mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 3735f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 37361da177e4SLinus Torvalds error = PTR_ERR(dentry); 37376902d925SDave Hansen if (IS_ERR(dentry)) 37386902d925SDave Hansen goto exit2; 3739e6bc45d6STheodore Ts'o if (!dentry->d_inode) { 3740e6bc45d6STheodore Ts'o error = -ENOENT; 3741e6bc45d6STheodore Ts'o goto exit3; 3742e6bc45d6STheodore Ts'o } 3743f5beed75SAl Viro error = security_path_rmdir(&path, dentry); 3744be6d3e56SKentaro Takeda if (error) 3745c30dabfeSJan Kara goto exit3; 3746f5beed75SAl Viro error = vfs_rmdir(path.dentry->d_inode, dentry); 37470622753bSDave Hansen exit3: 37481da177e4SLinus Torvalds dput(dentry); 37496902d925SDave Hansen exit2: 3750f5beed75SAl Viro mutex_unlock(&path.dentry->d_inode->i_mutex); 3751f5beed75SAl Viro mnt_drop_write(path.mnt); 37521da177e4SLinus Torvalds exit1: 3753f5beed75SAl Viro path_put(&path); 37541da177e4SLinus Torvalds putname(name); 3755c6ee9206SJeff Layton if (retry_estale(error, lookup_flags)) { 3756c6ee9206SJeff Layton lookup_flags |= LOOKUP_REVAL; 3757c6ee9206SJeff Layton goto retry; 3758c6ee9206SJeff Layton } 37591da177e4SLinus Torvalds return error; 37601da177e4SLinus Torvalds } 37611da177e4SLinus Torvalds 37623cdad428SHeiko Carstens SYSCALL_DEFINE1(rmdir, const char __user *, pathname) 37635590ff0dSUlrich Drepper { 37645590ff0dSUlrich Drepper return do_rmdir(AT_FDCWD, pathname); 37655590ff0dSUlrich Drepper } 37665590ff0dSUlrich Drepper 3767b21996e3SJ. Bruce Fields /** 3768b21996e3SJ. Bruce Fields * vfs_unlink - unlink a filesystem object 3769b21996e3SJ. Bruce Fields * @dir: parent directory 3770b21996e3SJ. Bruce Fields * @dentry: victim 3771b21996e3SJ. Bruce Fields * @delegated_inode: returns victim inode, if the inode is delegated. 3772b21996e3SJ. Bruce Fields * 3773b21996e3SJ. Bruce Fields * The caller must hold dir->i_mutex. 3774b21996e3SJ. Bruce Fields * 3775b21996e3SJ. Bruce Fields * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and 3776b21996e3SJ. Bruce Fields * return a reference to the inode in delegated_inode. The caller 3777b21996e3SJ. Bruce Fields * should then break the delegation on that inode and retry. Because 3778b21996e3SJ. Bruce Fields * breaking a delegation may take a long time, the caller should drop 3779b21996e3SJ. Bruce Fields * dir->i_mutex before doing so. 3780b21996e3SJ. Bruce Fields * 3781b21996e3SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 3782b21996e3SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 3783b21996e3SJ. Bruce Fields * to be NFS exported. 3784b21996e3SJ. Bruce Fields */ 3785b21996e3SJ. Bruce Fields int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode) 37861da177e4SLinus Torvalds { 37879accbb97SJ. Bruce Fields struct inode *target = dentry->d_inode; 37881da177e4SLinus Torvalds int error = may_delete(dir, dentry, 0); 37891da177e4SLinus Torvalds 37901da177e4SLinus Torvalds if (error) 37911da177e4SLinus Torvalds return error; 37921da177e4SLinus Torvalds 3793acfa4380SAl Viro if (!dir->i_op->unlink) 37941da177e4SLinus Torvalds return -EPERM; 37951da177e4SLinus Torvalds 37969accbb97SJ. Bruce Fields mutex_lock(&target->i_mutex); 37978ed936b5SEric W. Biederman if (is_local_mountpoint(dentry)) 37981da177e4SLinus Torvalds error = -EBUSY; 37991da177e4SLinus Torvalds else { 38001da177e4SLinus Torvalds error = security_inode_unlink(dir, dentry); 3801bec1052eSAl Viro if (!error) { 38025a14696cSJ. Bruce Fields error = try_break_deleg(target, delegated_inode); 38035a14696cSJ. Bruce Fields if (error) 3804b21996e3SJ. Bruce Fields goto out; 38051da177e4SLinus Torvalds error = dir->i_op->unlink(dir, dentry); 38068ed936b5SEric W. Biederman if (!error) { 3807d83c49f3SAl Viro dont_mount(dentry); 38088ed936b5SEric W. Biederman detach_mounts(dentry); 38098ed936b5SEric W. Biederman } 3810bec1052eSAl Viro } 38111da177e4SLinus Torvalds } 3812b21996e3SJ. Bruce Fields out: 38139accbb97SJ. Bruce Fields mutex_unlock(&target->i_mutex); 38141da177e4SLinus Torvalds 38151da177e4SLinus Torvalds /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 38161da177e4SLinus Torvalds if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { 38179accbb97SJ. Bruce Fields fsnotify_link_count(target); 38181da177e4SLinus Torvalds d_delete(dentry); 38191da177e4SLinus Torvalds } 38200eeca283SRobert Love 38211da177e4SLinus Torvalds return error; 38221da177e4SLinus Torvalds } 38234d359507SAl Viro EXPORT_SYMBOL(vfs_unlink); 38241da177e4SLinus Torvalds 38251da177e4SLinus Torvalds /* 38261da177e4SLinus Torvalds * Make sure that the actual truncation of the file will occur outside its 38271b1dcc1bSJes Sorensen * directory's i_mutex. Truncate can take a long time if there is a lot of 38281da177e4SLinus Torvalds * writeout happening, and we don't want to prevent access to the directory 38291da177e4SLinus Torvalds * while waiting on the I/O. 38301da177e4SLinus Torvalds */ 38315590ff0dSUlrich Drepper static long do_unlinkat(int dfd, const char __user *pathname) 38321da177e4SLinus Torvalds { 38332ad94ae6SAl Viro int error; 383491a27b2aSJeff Layton struct filename *name; 38351da177e4SLinus Torvalds struct dentry *dentry; 3836f5beed75SAl Viro struct path path; 3837f5beed75SAl Viro struct qstr last; 3838f5beed75SAl Viro int type; 38391da177e4SLinus Torvalds struct inode *inode = NULL; 3840b21996e3SJ. Bruce Fields struct inode *delegated_inode = NULL; 38415d18f813SJeff Layton unsigned int lookup_flags = 0; 38425d18f813SJeff Layton retry: 3843f5beed75SAl Viro name = user_path_parent(dfd, pathname, 3844f5beed75SAl Viro &path, &last, &type, lookup_flags); 384591a27b2aSJeff Layton if (IS_ERR(name)) 384691a27b2aSJeff Layton return PTR_ERR(name); 38472ad94ae6SAl Viro 38481da177e4SLinus Torvalds error = -EISDIR; 3849f5beed75SAl Viro if (type != LAST_NORM) 38501da177e4SLinus Torvalds goto exit1; 38510612d9fbSOGAWA Hirofumi 3852f5beed75SAl Viro error = mnt_want_write(path.mnt); 3853c30dabfeSJan Kara if (error) 3854c30dabfeSJan Kara goto exit1; 3855b21996e3SJ. Bruce Fields retry_deleg: 3856f5beed75SAl Viro mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 3857f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 38581da177e4SLinus Torvalds error = PTR_ERR(dentry); 38591da177e4SLinus Torvalds if (!IS_ERR(dentry)) { 38601da177e4SLinus Torvalds /* Why not before? Because we want correct error value */ 3861f5beed75SAl Viro if (last.name[last.len]) 386250338b88STörök Edwin goto slashes; 38631da177e4SLinus Torvalds inode = dentry->d_inode; 3864b18825a7SDavid Howells if (d_is_negative(dentry)) 3865e6bc45d6STheodore Ts'o goto slashes; 38667de9c6eeSAl Viro ihold(inode); 3867f5beed75SAl Viro error = security_path_unlink(&path, dentry); 3868be6d3e56SKentaro Takeda if (error) 3869c30dabfeSJan Kara goto exit2; 3870f5beed75SAl Viro error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode); 38711da177e4SLinus Torvalds exit2: 38721da177e4SLinus Torvalds dput(dentry); 38731da177e4SLinus Torvalds } 3874f5beed75SAl Viro mutex_unlock(&path.dentry->d_inode->i_mutex); 38751da177e4SLinus Torvalds if (inode) 38761da177e4SLinus Torvalds iput(inode); /* truncate the inode here */ 3877b21996e3SJ. Bruce Fields inode = NULL; 3878b21996e3SJ. Bruce Fields if (delegated_inode) { 38795a14696cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 3880b21996e3SJ. Bruce Fields if (!error) 3881b21996e3SJ. Bruce Fields goto retry_deleg; 3882b21996e3SJ. Bruce Fields } 3883f5beed75SAl Viro mnt_drop_write(path.mnt); 38841da177e4SLinus Torvalds exit1: 3885f5beed75SAl Viro path_put(&path); 38861da177e4SLinus Torvalds putname(name); 38875d18f813SJeff Layton if (retry_estale(error, lookup_flags)) { 38885d18f813SJeff Layton lookup_flags |= LOOKUP_REVAL; 38895d18f813SJeff Layton inode = NULL; 38905d18f813SJeff Layton goto retry; 38915d18f813SJeff Layton } 38921da177e4SLinus Torvalds return error; 38931da177e4SLinus Torvalds 38941da177e4SLinus Torvalds slashes: 3895b18825a7SDavid Howells if (d_is_negative(dentry)) 3896b18825a7SDavid Howells error = -ENOENT; 389744b1d530SMiklos Szeredi else if (d_is_dir(dentry)) 3898b18825a7SDavid Howells error = -EISDIR; 3899b18825a7SDavid Howells else 3900b18825a7SDavid Howells error = -ENOTDIR; 39011da177e4SLinus Torvalds goto exit2; 39021da177e4SLinus Torvalds } 39031da177e4SLinus Torvalds 39042e4d0924SHeiko Carstens SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) 39055590ff0dSUlrich Drepper { 39065590ff0dSUlrich Drepper if ((flag & ~AT_REMOVEDIR) != 0) 39075590ff0dSUlrich Drepper return -EINVAL; 39085590ff0dSUlrich Drepper 39095590ff0dSUlrich Drepper if (flag & AT_REMOVEDIR) 39105590ff0dSUlrich Drepper return do_rmdir(dfd, pathname); 39115590ff0dSUlrich Drepper 39125590ff0dSUlrich Drepper return do_unlinkat(dfd, pathname); 39135590ff0dSUlrich Drepper } 39145590ff0dSUlrich Drepper 39153480b257SHeiko Carstens SYSCALL_DEFINE1(unlink, const char __user *, pathname) 39165590ff0dSUlrich Drepper { 39175590ff0dSUlrich Drepper return do_unlinkat(AT_FDCWD, pathname); 39185590ff0dSUlrich Drepper } 39195590ff0dSUlrich Drepper 3920db2e747bSMiklos Szeredi int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) 39211da177e4SLinus Torvalds { 3922a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 39231da177e4SLinus Torvalds 39241da177e4SLinus Torvalds if (error) 39251da177e4SLinus Torvalds return error; 39261da177e4SLinus Torvalds 3927acfa4380SAl Viro if (!dir->i_op->symlink) 39281da177e4SLinus Torvalds return -EPERM; 39291da177e4SLinus Torvalds 39301da177e4SLinus Torvalds error = security_inode_symlink(dir, dentry, oldname); 39311da177e4SLinus Torvalds if (error) 39321da177e4SLinus Torvalds return error; 39331da177e4SLinus Torvalds 39341da177e4SLinus Torvalds error = dir->i_op->symlink(dir, dentry, oldname); 3935a74574aaSStephen Smalley if (!error) 3936f38aa942SAmy Griffis fsnotify_create(dir, dentry); 39371da177e4SLinus Torvalds return error; 39381da177e4SLinus Torvalds } 39394d359507SAl Viro EXPORT_SYMBOL(vfs_symlink); 39401da177e4SLinus Torvalds 39412e4d0924SHeiko Carstens SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, 39422e4d0924SHeiko Carstens int, newdfd, const char __user *, newname) 39431da177e4SLinus Torvalds { 39442ad94ae6SAl Viro int error; 394591a27b2aSJeff Layton struct filename *from; 39466902d925SDave Hansen struct dentry *dentry; 3947dae6ad8fSAl Viro struct path path; 3948f46d3567SJeff Layton unsigned int lookup_flags = 0; 39491da177e4SLinus Torvalds 39501da177e4SLinus Torvalds from = getname(oldname); 39511da177e4SLinus Torvalds if (IS_ERR(from)) 39521da177e4SLinus Torvalds return PTR_ERR(from); 3953f46d3567SJeff Layton retry: 3954f46d3567SJeff Layton dentry = user_path_create(newdfd, newname, &path, lookup_flags); 39551da177e4SLinus Torvalds error = PTR_ERR(dentry); 39566902d925SDave Hansen if (IS_ERR(dentry)) 3957dae6ad8fSAl Viro goto out_putname; 39586902d925SDave Hansen 395991a27b2aSJeff Layton error = security_path_symlink(&path, dentry, from->name); 3960a8104a9fSAl Viro if (!error) 396191a27b2aSJeff Layton error = vfs_symlink(path.dentry->d_inode, dentry, from->name); 3962921a1650SAl Viro done_path_create(&path, dentry); 3963f46d3567SJeff Layton if (retry_estale(error, lookup_flags)) { 3964f46d3567SJeff Layton lookup_flags |= LOOKUP_REVAL; 3965f46d3567SJeff Layton goto retry; 3966f46d3567SJeff Layton } 39676902d925SDave Hansen out_putname: 39681da177e4SLinus Torvalds putname(from); 39691da177e4SLinus Torvalds return error; 39701da177e4SLinus Torvalds } 39711da177e4SLinus Torvalds 39723480b257SHeiko Carstens SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) 39735590ff0dSUlrich Drepper { 39745590ff0dSUlrich Drepper return sys_symlinkat(oldname, AT_FDCWD, newname); 39755590ff0dSUlrich Drepper } 39765590ff0dSUlrich Drepper 3977146a8595SJ. Bruce Fields /** 3978146a8595SJ. Bruce Fields * vfs_link - create a new link 3979146a8595SJ. Bruce Fields * @old_dentry: object to be linked 3980146a8595SJ. Bruce Fields * @dir: new parent 3981146a8595SJ. Bruce Fields * @new_dentry: where to create the new link 3982146a8595SJ. Bruce Fields * @delegated_inode: returns inode needing a delegation break 3983146a8595SJ. Bruce Fields * 3984146a8595SJ. Bruce Fields * The caller must hold dir->i_mutex 3985146a8595SJ. Bruce Fields * 3986146a8595SJ. Bruce Fields * If vfs_link discovers a delegation on the to-be-linked file in need 3987146a8595SJ. Bruce Fields * of breaking, it will return -EWOULDBLOCK and return a reference to the 3988146a8595SJ. Bruce Fields * inode in delegated_inode. The caller should then break the delegation 3989146a8595SJ. Bruce Fields * and retry. Because breaking a delegation may take a long time, the 3990146a8595SJ. Bruce Fields * caller should drop the i_mutex before doing so. 3991146a8595SJ. Bruce Fields * 3992146a8595SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 3993146a8595SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 3994146a8595SJ. Bruce Fields * to be NFS exported. 3995146a8595SJ. Bruce Fields */ 3996146a8595SJ. Bruce Fields int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode) 39971da177e4SLinus Torvalds { 39981da177e4SLinus Torvalds struct inode *inode = old_dentry->d_inode; 39998de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 40001da177e4SLinus Torvalds int error; 40011da177e4SLinus Torvalds 40021da177e4SLinus Torvalds if (!inode) 40031da177e4SLinus Torvalds return -ENOENT; 40041da177e4SLinus Torvalds 4005a95164d9SMiklos Szeredi error = may_create(dir, new_dentry); 40061da177e4SLinus Torvalds if (error) 40071da177e4SLinus Torvalds return error; 40081da177e4SLinus Torvalds 40091da177e4SLinus Torvalds if (dir->i_sb != inode->i_sb) 40101da177e4SLinus Torvalds return -EXDEV; 40111da177e4SLinus Torvalds 40121da177e4SLinus Torvalds /* 40131da177e4SLinus Torvalds * A link to an append-only or immutable file cannot be created. 40141da177e4SLinus Torvalds */ 40151da177e4SLinus Torvalds if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 40161da177e4SLinus Torvalds return -EPERM; 4017acfa4380SAl Viro if (!dir->i_op->link) 40181da177e4SLinus Torvalds return -EPERM; 40197e79eedbSTetsuo Handa if (S_ISDIR(inode->i_mode)) 40201da177e4SLinus Torvalds return -EPERM; 40211da177e4SLinus Torvalds 40221da177e4SLinus Torvalds error = security_inode_link(old_dentry, dir, new_dentry); 40231da177e4SLinus Torvalds if (error) 40241da177e4SLinus Torvalds return error; 40251da177e4SLinus Torvalds 40267e79eedbSTetsuo Handa mutex_lock(&inode->i_mutex); 4027aae8a97dSAneesh Kumar K.V /* Make sure we don't allow creating hardlink to an unlinked file */ 4028f4e0c30cSAl Viro if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) 4029aae8a97dSAneesh Kumar K.V error = -ENOENT; 40308de52778SAl Viro else if (max_links && inode->i_nlink >= max_links) 40318de52778SAl Viro error = -EMLINK; 4032146a8595SJ. Bruce Fields else { 4033146a8595SJ. Bruce Fields error = try_break_deleg(inode, delegated_inode); 4034146a8595SJ. Bruce Fields if (!error) 40351da177e4SLinus Torvalds error = dir->i_op->link(old_dentry, dir, new_dentry); 4036146a8595SJ. Bruce Fields } 4037f4e0c30cSAl Viro 4038f4e0c30cSAl Viro if (!error && (inode->i_state & I_LINKABLE)) { 4039f4e0c30cSAl Viro spin_lock(&inode->i_lock); 4040f4e0c30cSAl Viro inode->i_state &= ~I_LINKABLE; 4041f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 4042f4e0c30cSAl Viro } 40437e79eedbSTetsuo Handa mutex_unlock(&inode->i_mutex); 4044e31e14ecSStephen Smalley if (!error) 40457e79eedbSTetsuo Handa fsnotify_link(dir, inode, new_dentry); 40461da177e4SLinus Torvalds return error; 40471da177e4SLinus Torvalds } 40484d359507SAl Viro EXPORT_SYMBOL(vfs_link); 40491da177e4SLinus Torvalds 40501da177e4SLinus Torvalds /* 40511da177e4SLinus Torvalds * Hardlinks are often used in delicate situations. We avoid 40521da177e4SLinus Torvalds * security-related surprises by not following symlinks on the 40531da177e4SLinus Torvalds * newname. --KAB 40541da177e4SLinus Torvalds * 40551da177e4SLinus Torvalds * We don't follow them on the oldname either to be compatible 40561da177e4SLinus Torvalds * with linux 2.0, and to avoid hard-linking to directories 40571da177e4SLinus Torvalds * and other special files. --ADM 40581da177e4SLinus Torvalds */ 40592e4d0924SHeiko Carstens SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, 40602e4d0924SHeiko Carstens int, newdfd, const char __user *, newname, int, flags) 40611da177e4SLinus Torvalds { 40621da177e4SLinus Torvalds struct dentry *new_dentry; 4063dae6ad8fSAl Viro struct path old_path, new_path; 4064146a8595SJ. Bruce Fields struct inode *delegated_inode = NULL; 406511a7b371SAneesh Kumar K.V int how = 0; 40661da177e4SLinus Torvalds int error; 40671da177e4SLinus Torvalds 406811a7b371SAneesh Kumar K.V if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) 4069c04030e1SUlrich Drepper return -EINVAL; 407011a7b371SAneesh Kumar K.V /* 4071f0cc6ffbSLinus Torvalds * To use null names we require CAP_DAC_READ_SEARCH 4072f0cc6ffbSLinus Torvalds * This ensures that not everyone will be able to create 4073f0cc6ffbSLinus Torvalds * handlink using the passed filedescriptor. 407411a7b371SAneesh Kumar K.V */ 4075f0cc6ffbSLinus Torvalds if (flags & AT_EMPTY_PATH) { 4076f0cc6ffbSLinus Torvalds if (!capable(CAP_DAC_READ_SEARCH)) 4077f0cc6ffbSLinus Torvalds return -ENOENT; 407811a7b371SAneesh Kumar K.V how = LOOKUP_EMPTY; 4079f0cc6ffbSLinus Torvalds } 4080c04030e1SUlrich Drepper 408111a7b371SAneesh Kumar K.V if (flags & AT_SYMLINK_FOLLOW) 408211a7b371SAneesh Kumar K.V how |= LOOKUP_FOLLOW; 4083442e31caSJeff Layton retry: 408411a7b371SAneesh Kumar K.V error = user_path_at(olddfd, oldname, how, &old_path); 40851da177e4SLinus Torvalds if (error) 40862ad94ae6SAl Viro return error; 40872ad94ae6SAl Viro 4088442e31caSJeff Layton new_dentry = user_path_create(newdfd, newname, &new_path, 4089442e31caSJeff Layton (how & LOOKUP_REVAL)); 40901da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 40916902d925SDave Hansen if (IS_ERR(new_dentry)) 4092dae6ad8fSAl Viro goto out; 4093dae6ad8fSAl Viro 4094dae6ad8fSAl Viro error = -EXDEV; 4095dae6ad8fSAl Viro if (old_path.mnt != new_path.mnt) 4096dae6ad8fSAl Viro goto out_dput; 4097800179c9SKees Cook error = may_linkat(&old_path); 4098800179c9SKees Cook if (unlikely(error)) 4099800179c9SKees Cook goto out_dput; 4100dae6ad8fSAl Viro error = security_path_link(old_path.dentry, &new_path, new_dentry); 4101be6d3e56SKentaro Takeda if (error) 4102a8104a9fSAl Viro goto out_dput; 4103146a8595SJ. Bruce Fields error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode); 410475c3f29dSDave Hansen out_dput: 4105921a1650SAl Viro done_path_create(&new_path, new_dentry); 4106146a8595SJ. Bruce Fields if (delegated_inode) { 4107146a8595SJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4108d22e6338SOleg Drokin if (!error) { 4109d22e6338SOleg Drokin path_put(&old_path); 4110146a8595SJ. Bruce Fields goto retry; 4111146a8595SJ. Bruce Fields } 4112d22e6338SOleg Drokin } 4113442e31caSJeff Layton if (retry_estale(error, how)) { 4114d22e6338SOleg Drokin path_put(&old_path); 4115442e31caSJeff Layton how |= LOOKUP_REVAL; 4116442e31caSJeff Layton goto retry; 4117442e31caSJeff Layton } 41181da177e4SLinus Torvalds out: 41192d8f3038SAl Viro path_put(&old_path); 41201da177e4SLinus Torvalds 41211da177e4SLinus Torvalds return error; 41221da177e4SLinus Torvalds } 41231da177e4SLinus Torvalds 41243480b257SHeiko Carstens SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) 41255590ff0dSUlrich Drepper { 4126c04030e1SUlrich Drepper return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 41275590ff0dSUlrich Drepper } 41285590ff0dSUlrich Drepper 4129bc27027aSMiklos Szeredi /** 4130bc27027aSMiklos Szeredi * vfs_rename - rename a filesystem object 4131bc27027aSMiklos Szeredi * @old_dir: parent of source 4132bc27027aSMiklos Szeredi * @old_dentry: source 4133bc27027aSMiklos Szeredi * @new_dir: parent of destination 4134bc27027aSMiklos Szeredi * @new_dentry: destination 4135bc27027aSMiklos Szeredi * @delegated_inode: returns an inode needing a delegation break 4136520c8b16SMiklos Szeredi * @flags: rename flags 4137bc27027aSMiklos Szeredi * 4138bc27027aSMiklos Szeredi * The caller must hold multiple mutexes--see lock_rename()). 4139bc27027aSMiklos Szeredi * 4140bc27027aSMiklos Szeredi * If vfs_rename discovers a delegation in need of breaking at either 4141bc27027aSMiklos Szeredi * the source or destination, it will return -EWOULDBLOCK and return a 4142bc27027aSMiklos Szeredi * reference to the inode in delegated_inode. The caller should then 4143bc27027aSMiklos Szeredi * break the delegation and retry. Because breaking a delegation may 4144bc27027aSMiklos Szeredi * take a long time, the caller should drop all locks before doing 4145bc27027aSMiklos Szeredi * so. 4146bc27027aSMiklos Szeredi * 4147bc27027aSMiklos Szeredi * Alternatively, a caller may pass NULL for delegated_inode. This may 4148bc27027aSMiklos Szeredi * be appropriate for callers that expect the underlying filesystem not 4149bc27027aSMiklos Szeredi * to be NFS exported. 4150bc27027aSMiklos Szeredi * 41511da177e4SLinus Torvalds * The worst of all namespace operations - renaming directory. "Perverted" 41521da177e4SLinus Torvalds * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 41531da177e4SLinus Torvalds * Problems: 4154d03b29a2SJ. Bruce Fields * a) we can get into loop creation. 41551da177e4SLinus Torvalds * b) race potential - two innocent renames can create a loop together. 41561da177e4SLinus Torvalds * That's where 4.4 screws up. Current fix: serialization on 4157a11f3a05SArjan van de Ven * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 41581da177e4SLinus Torvalds * story. 41596cedba89SJ. Bruce Fields * c) we have to lock _four_ objects - parents and victim (if it exists), 41606cedba89SJ. Bruce Fields * and source (if it is not a directory). 41611b1dcc1bSJes Sorensen * And that - after we got ->i_mutex on parents (until then we don't know 41621da177e4SLinus Torvalds * whether the target exists). Solution: try to be smart with locking 41631da177e4SLinus Torvalds * order for inodes. We rely on the fact that tree topology may change 4164a11f3a05SArjan van de Ven * only under ->s_vfs_rename_mutex _and_ that parent of the object we 41651da177e4SLinus Torvalds * move will be locked. Thus we can rank directories by the tree 41661da177e4SLinus Torvalds * (ancestors first) and rank all non-directories after them. 41671da177e4SLinus Torvalds * That works since everybody except rename does "lock parent, lookup, 4168a11f3a05SArjan van de Ven * lock child" and rename is under ->s_vfs_rename_mutex. 41691da177e4SLinus Torvalds * HOWEVER, it relies on the assumption that any object with ->lookup() 41701da177e4SLinus Torvalds * has no more than 1 dentry. If "hybrid" objects will ever appear, 41711da177e4SLinus Torvalds * we'd better make sure that there's no link(2) for them. 4172e4eaac06SSage Weil * d) conversion from fhandle to dentry may come in the wrong moment - when 41731b1dcc1bSJes Sorensen * we are removing the target. Solution: we will have to grab ->i_mutex 41741da177e4SLinus Torvalds * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 4175c41b20e7SAdam Buchbinder * ->i_mutex on parents, which works but leads to some truly excessive 41761da177e4SLinus Torvalds * locking]. 41771da177e4SLinus Torvalds */ 41781da177e4SLinus Torvalds int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, 41798e6d782cSJ. Bruce Fields struct inode *new_dir, struct dentry *new_dentry, 4180520c8b16SMiklos Szeredi struct inode **delegated_inode, unsigned int flags) 41811da177e4SLinus Torvalds { 41821da177e4SLinus Torvalds int error; 4183bc27027aSMiklos Szeredi bool is_dir = d_is_dir(old_dentry); 418459b0df21SEric Paris const unsigned char *old_name; 4185bc27027aSMiklos Szeredi struct inode *source = old_dentry->d_inode; 4186bc27027aSMiklos Szeredi struct inode *target = new_dentry->d_inode; 4187da1ce067SMiklos Szeredi bool new_is_dir = false; 4188da1ce067SMiklos Szeredi unsigned max_links = new_dir->i_sb->s_max_links; 41891da177e4SLinus Torvalds 4190bc27027aSMiklos Szeredi if (source == target) 41911da177e4SLinus Torvalds return 0; 41921da177e4SLinus Torvalds 41931da177e4SLinus Torvalds error = may_delete(old_dir, old_dentry, is_dir); 41941da177e4SLinus Torvalds if (error) 41951da177e4SLinus Torvalds return error; 41961da177e4SLinus Torvalds 4197da1ce067SMiklos Szeredi if (!target) { 4198a95164d9SMiklos Szeredi error = may_create(new_dir, new_dentry); 4199da1ce067SMiklos Szeredi } else { 4200da1ce067SMiklos Szeredi new_is_dir = d_is_dir(new_dentry); 4201da1ce067SMiklos Szeredi 4202da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 42031da177e4SLinus Torvalds error = may_delete(new_dir, new_dentry, is_dir); 4204da1ce067SMiklos Szeredi else 4205da1ce067SMiklos Szeredi error = may_delete(new_dir, new_dentry, new_is_dir); 4206da1ce067SMiklos Szeredi } 42071da177e4SLinus Torvalds if (error) 42081da177e4SLinus Torvalds return error; 42091da177e4SLinus Torvalds 42107177a9c4SMiklos Szeredi if (!old_dir->i_op->rename && !old_dir->i_op->rename2) 42111da177e4SLinus Torvalds return -EPERM; 42121da177e4SLinus Torvalds 4213520c8b16SMiklos Szeredi if (flags && !old_dir->i_op->rename2) 4214520c8b16SMiklos Szeredi return -EINVAL; 4215520c8b16SMiklos Szeredi 4216bc27027aSMiklos Szeredi /* 4217bc27027aSMiklos Szeredi * If we are going to change the parent - check write permissions, 4218bc27027aSMiklos Szeredi * we'll need to flip '..'. 4219bc27027aSMiklos Szeredi */ 4220da1ce067SMiklos Szeredi if (new_dir != old_dir) { 4221da1ce067SMiklos Szeredi if (is_dir) { 4222bc27027aSMiklos Szeredi error = inode_permission(source, MAY_WRITE); 4223bc27027aSMiklos Szeredi if (error) 4224bc27027aSMiklos Szeredi return error; 4225bc27027aSMiklos Szeredi } 4226da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && new_is_dir) { 4227da1ce067SMiklos Szeredi error = inode_permission(target, MAY_WRITE); 4228da1ce067SMiklos Szeredi if (error) 4229da1ce067SMiklos Szeredi return error; 4230da1ce067SMiklos Szeredi } 4231da1ce067SMiklos Szeredi } 42320eeca283SRobert Love 42330b3974ebSMiklos Szeredi error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry, 42340b3974ebSMiklos Szeredi flags); 4235bc27027aSMiklos Szeredi if (error) 4236bc27027aSMiklos Szeredi return error; 4237bc27027aSMiklos Szeredi 4238bc27027aSMiklos Szeredi old_name = fsnotify_oldname_init(old_dentry->d_name.name); 4239bc27027aSMiklos Szeredi dget(new_dentry); 4240da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4241bc27027aSMiklos Szeredi lock_two_nondirectories(source, target); 4242bc27027aSMiklos Szeredi else if (target) 4243bc27027aSMiklos Szeredi mutex_lock(&target->i_mutex); 4244bc27027aSMiklos Szeredi 4245bc27027aSMiklos Szeredi error = -EBUSY; 42467af1364fSEric W. Biederman if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry)) 4247bc27027aSMiklos Szeredi goto out; 4248bc27027aSMiklos Szeredi 4249da1ce067SMiklos Szeredi if (max_links && new_dir != old_dir) { 4250bc27027aSMiklos Szeredi error = -EMLINK; 4251da1ce067SMiklos Szeredi if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links) 4252bc27027aSMiklos Szeredi goto out; 4253da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir && 4254da1ce067SMiklos Szeredi old_dir->i_nlink >= max_links) 4255da1ce067SMiklos Szeredi goto out; 4256da1ce067SMiklos Szeredi } 4257da1ce067SMiklos Szeredi if (is_dir && !(flags & RENAME_EXCHANGE) && target) 4258bc27027aSMiklos Szeredi shrink_dcache_parent(new_dentry); 4259da1ce067SMiklos Szeredi if (!is_dir) { 4260bc27027aSMiklos Szeredi error = try_break_deleg(source, delegated_inode); 4261bc27027aSMiklos Szeredi if (error) 4262bc27027aSMiklos Szeredi goto out; 4263da1ce067SMiklos Szeredi } 4264da1ce067SMiklos Szeredi if (target && !new_is_dir) { 4265bc27027aSMiklos Szeredi error = try_break_deleg(target, delegated_inode); 4266bc27027aSMiklos Szeredi if (error) 4267bc27027aSMiklos Szeredi goto out; 4268bc27027aSMiklos Szeredi } 42697177a9c4SMiklos Szeredi if (!old_dir->i_op->rename2) { 4270520c8b16SMiklos Szeredi error = old_dir->i_op->rename(old_dir, old_dentry, 4271520c8b16SMiklos Szeredi new_dir, new_dentry); 4272520c8b16SMiklos Szeredi } else { 42737177a9c4SMiklos Szeredi WARN_ON(old_dir->i_op->rename != NULL); 4274520c8b16SMiklos Szeredi error = old_dir->i_op->rename2(old_dir, old_dentry, 4275520c8b16SMiklos Szeredi new_dir, new_dentry, flags); 4276520c8b16SMiklos Szeredi } 4277bc27027aSMiklos Szeredi if (error) 4278bc27027aSMiklos Szeredi goto out; 4279bc27027aSMiklos Szeredi 4280da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE) && target) { 42811da177e4SLinus Torvalds if (is_dir) 4282bc27027aSMiklos Szeredi target->i_flags |= S_DEAD; 4283bc27027aSMiklos Szeredi dont_mount(new_dentry); 42848ed936b5SEric W. Biederman detach_mounts(new_dentry); 4285bc27027aSMiklos Szeredi } 4286da1ce067SMiklos Szeredi if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) { 4287da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 4288bc27027aSMiklos Szeredi d_move(old_dentry, new_dentry); 4289da1ce067SMiklos Szeredi else 4290da1ce067SMiklos Szeredi d_exchange(old_dentry, new_dentry); 4291da1ce067SMiklos Szeredi } 4292bc27027aSMiklos Szeredi out: 4293da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4294bc27027aSMiklos Szeredi unlock_two_nondirectories(source, target); 4295bc27027aSMiklos Szeredi else if (target) 4296bc27027aSMiklos Szeredi mutex_unlock(&target->i_mutex); 4297bc27027aSMiklos Szeredi dput(new_dentry); 4298da1ce067SMiklos Szeredi if (!error) { 4299123df294SAl Viro fsnotify_move(old_dir, new_dir, old_name, is_dir, 4300da1ce067SMiklos Szeredi !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry); 4301da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4302da1ce067SMiklos Szeredi fsnotify_move(new_dir, old_dir, old_dentry->d_name.name, 4303da1ce067SMiklos Szeredi new_is_dir, NULL, new_dentry); 4304da1ce067SMiklos Szeredi } 4305da1ce067SMiklos Szeredi } 43060eeca283SRobert Love fsnotify_oldname_free(old_name); 43070eeca283SRobert Love 43081da177e4SLinus Torvalds return error; 43091da177e4SLinus Torvalds } 43104d359507SAl Viro EXPORT_SYMBOL(vfs_rename); 43111da177e4SLinus Torvalds 4312520c8b16SMiklos Szeredi SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname, 4313520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname, unsigned int, flags) 43141da177e4SLinus Torvalds { 43151da177e4SLinus Torvalds struct dentry *old_dentry, *new_dentry; 43161da177e4SLinus Torvalds struct dentry *trap; 4317f5beed75SAl Viro struct path old_path, new_path; 4318f5beed75SAl Viro struct qstr old_last, new_last; 4319f5beed75SAl Viro int old_type, new_type; 43208e6d782cSJ. Bruce Fields struct inode *delegated_inode = NULL; 432191a27b2aSJeff Layton struct filename *from; 432291a27b2aSJeff Layton struct filename *to; 4323f5beed75SAl Viro unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET; 4324c6a94284SJeff Layton bool should_retry = false; 43252ad94ae6SAl Viro int error; 4326520c8b16SMiklos Szeredi 43270d7a8555SMiklos Szeredi if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) 4328da1ce067SMiklos Szeredi return -EINVAL; 4329da1ce067SMiklos Szeredi 43300d7a8555SMiklos Szeredi if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) && 43310d7a8555SMiklos Szeredi (flags & RENAME_EXCHANGE)) 4332520c8b16SMiklos Szeredi return -EINVAL; 4333520c8b16SMiklos Szeredi 43340d7a8555SMiklos Szeredi if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD)) 43350d7a8555SMiklos Szeredi return -EPERM; 43360d7a8555SMiklos Szeredi 4337f5beed75SAl Viro if (flags & RENAME_EXCHANGE) 4338f5beed75SAl Viro target_flags = 0; 4339f5beed75SAl Viro 4340c6a94284SJeff Layton retry: 4341f5beed75SAl Viro from = user_path_parent(olddfd, oldname, 4342f5beed75SAl Viro &old_path, &old_last, &old_type, lookup_flags); 434391a27b2aSJeff Layton if (IS_ERR(from)) { 434491a27b2aSJeff Layton error = PTR_ERR(from); 43451da177e4SLinus Torvalds goto exit; 434691a27b2aSJeff Layton } 43471da177e4SLinus Torvalds 4348f5beed75SAl Viro to = user_path_parent(newdfd, newname, 4349f5beed75SAl Viro &new_path, &new_last, &new_type, lookup_flags); 435091a27b2aSJeff Layton if (IS_ERR(to)) { 435191a27b2aSJeff Layton error = PTR_ERR(to); 43521da177e4SLinus Torvalds goto exit1; 435391a27b2aSJeff Layton } 43541da177e4SLinus Torvalds 43551da177e4SLinus Torvalds error = -EXDEV; 4356f5beed75SAl Viro if (old_path.mnt != new_path.mnt) 43571da177e4SLinus Torvalds goto exit2; 43581da177e4SLinus Torvalds 43591da177e4SLinus Torvalds error = -EBUSY; 4360f5beed75SAl Viro if (old_type != LAST_NORM) 43611da177e4SLinus Torvalds goto exit2; 43621da177e4SLinus Torvalds 43630a7c3937SMiklos Szeredi if (flags & RENAME_NOREPLACE) 43640a7c3937SMiklos Szeredi error = -EEXIST; 4365f5beed75SAl Viro if (new_type != LAST_NORM) 43661da177e4SLinus Torvalds goto exit2; 43671da177e4SLinus Torvalds 4368f5beed75SAl Viro error = mnt_want_write(old_path.mnt); 4369c30dabfeSJan Kara if (error) 4370c30dabfeSJan Kara goto exit2; 4371c30dabfeSJan Kara 43728e6d782cSJ. Bruce Fields retry_deleg: 4373f5beed75SAl Viro trap = lock_rename(new_path.dentry, old_path.dentry); 43741da177e4SLinus Torvalds 4375f5beed75SAl Viro old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags); 43761da177e4SLinus Torvalds error = PTR_ERR(old_dentry); 43771da177e4SLinus Torvalds if (IS_ERR(old_dentry)) 43781da177e4SLinus Torvalds goto exit3; 43791da177e4SLinus Torvalds /* source must exist */ 43801da177e4SLinus Torvalds error = -ENOENT; 4381b18825a7SDavid Howells if (d_is_negative(old_dentry)) 43821da177e4SLinus Torvalds goto exit4; 4383f5beed75SAl Viro new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags); 43841da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 43851da177e4SLinus Torvalds if (IS_ERR(new_dentry)) 43861da177e4SLinus Torvalds goto exit4; 43870a7c3937SMiklos Szeredi error = -EEXIST; 43880a7c3937SMiklos Szeredi if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry)) 43890a7c3937SMiklos Szeredi goto exit5; 4390da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4391da1ce067SMiklos Szeredi error = -ENOENT; 4392da1ce067SMiklos Szeredi if (d_is_negative(new_dentry)) 4393da1ce067SMiklos Szeredi goto exit5; 4394da1ce067SMiklos Szeredi 4395da1ce067SMiklos Szeredi if (!d_is_dir(new_dentry)) { 4396da1ce067SMiklos Szeredi error = -ENOTDIR; 4397f5beed75SAl Viro if (new_last.name[new_last.len]) 4398da1ce067SMiklos Szeredi goto exit5; 4399da1ce067SMiklos Szeredi } 4400da1ce067SMiklos Szeredi } 44010a7c3937SMiklos Szeredi /* unless the source is a directory trailing slashes give -ENOTDIR */ 44020a7c3937SMiklos Szeredi if (!d_is_dir(old_dentry)) { 44030a7c3937SMiklos Szeredi error = -ENOTDIR; 4404f5beed75SAl Viro if (old_last.name[old_last.len]) 44050a7c3937SMiklos Szeredi goto exit5; 4406f5beed75SAl Viro if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len]) 44070a7c3937SMiklos Szeredi goto exit5; 44080a7c3937SMiklos Szeredi } 44090a7c3937SMiklos Szeredi /* source should not be ancestor of target */ 44100a7c3937SMiklos Szeredi error = -EINVAL; 44110a7c3937SMiklos Szeredi if (old_dentry == trap) 44120a7c3937SMiklos Szeredi goto exit5; 44131da177e4SLinus Torvalds /* target should not be an ancestor of source */ 4414da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 44151da177e4SLinus Torvalds error = -ENOTEMPTY; 44161da177e4SLinus Torvalds if (new_dentry == trap) 44171da177e4SLinus Torvalds goto exit5; 44181da177e4SLinus Torvalds 4419f5beed75SAl Viro error = security_path_rename(&old_path, old_dentry, 4420f5beed75SAl Viro &new_path, new_dentry, flags); 4421be6d3e56SKentaro Takeda if (error) 4422c30dabfeSJan Kara goto exit5; 4423f5beed75SAl Viro error = vfs_rename(old_path.dentry->d_inode, old_dentry, 4424f5beed75SAl Viro new_path.dentry->d_inode, new_dentry, 4425520c8b16SMiklos Szeredi &delegated_inode, flags); 44261da177e4SLinus Torvalds exit5: 44271da177e4SLinus Torvalds dput(new_dentry); 44281da177e4SLinus Torvalds exit4: 44291da177e4SLinus Torvalds dput(old_dentry); 44301da177e4SLinus Torvalds exit3: 4431f5beed75SAl Viro unlock_rename(new_path.dentry, old_path.dentry); 44328e6d782cSJ. Bruce Fields if (delegated_inode) { 44338e6d782cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 44348e6d782cSJ. Bruce Fields if (!error) 44358e6d782cSJ. Bruce Fields goto retry_deleg; 44368e6d782cSJ. Bruce Fields } 4437f5beed75SAl Viro mnt_drop_write(old_path.mnt); 44381da177e4SLinus Torvalds exit2: 4439c6a94284SJeff Layton if (retry_estale(error, lookup_flags)) 4440c6a94284SJeff Layton should_retry = true; 4441f5beed75SAl Viro path_put(&new_path); 44422ad94ae6SAl Viro putname(to); 44431da177e4SLinus Torvalds exit1: 4444f5beed75SAl Viro path_put(&old_path); 44451da177e4SLinus Torvalds putname(from); 4446c6a94284SJeff Layton if (should_retry) { 4447c6a94284SJeff Layton should_retry = false; 4448c6a94284SJeff Layton lookup_flags |= LOOKUP_REVAL; 4449c6a94284SJeff Layton goto retry; 4450c6a94284SJeff Layton } 44512ad94ae6SAl Viro exit: 44521da177e4SLinus Torvalds return error; 44531da177e4SLinus Torvalds } 44541da177e4SLinus Torvalds 4455520c8b16SMiklos Szeredi SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, 4456520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname) 4457520c8b16SMiklos Szeredi { 4458520c8b16SMiklos Szeredi return sys_renameat2(olddfd, oldname, newdfd, newname, 0); 4459520c8b16SMiklos Szeredi } 4460520c8b16SMiklos Szeredi 4461a26eab24SHeiko Carstens SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) 44625590ff0dSUlrich Drepper { 4463520c8b16SMiklos Szeredi return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 44645590ff0dSUlrich Drepper } 44655590ff0dSUlrich Drepper 4466787fb6bcSMiklos Szeredi int vfs_whiteout(struct inode *dir, struct dentry *dentry) 4467787fb6bcSMiklos Szeredi { 4468787fb6bcSMiklos Szeredi int error = may_create(dir, dentry); 4469787fb6bcSMiklos Szeredi if (error) 4470787fb6bcSMiklos Szeredi return error; 4471787fb6bcSMiklos Szeredi 4472787fb6bcSMiklos Szeredi if (!dir->i_op->mknod) 4473787fb6bcSMiklos Szeredi return -EPERM; 4474787fb6bcSMiklos Szeredi 4475787fb6bcSMiklos Szeredi return dir->i_op->mknod(dir, dentry, 4476787fb6bcSMiklos Szeredi S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV); 4477787fb6bcSMiklos Szeredi } 4478787fb6bcSMiklos Szeredi EXPORT_SYMBOL(vfs_whiteout); 4479787fb6bcSMiklos Szeredi 44805d826c84SAl Viro int readlink_copy(char __user *buffer, int buflen, const char *link) 44811da177e4SLinus Torvalds { 44825d826c84SAl Viro int len = PTR_ERR(link); 44831da177e4SLinus Torvalds if (IS_ERR(link)) 44841da177e4SLinus Torvalds goto out; 44851da177e4SLinus Torvalds 44861da177e4SLinus Torvalds len = strlen(link); 44871da177e4SLinus Torvalds if (len > (unsigned) buflen) 44881da177e4SLinus Torvalds len = buflen; 44891da177e4SLinus Torvalds if (copy_to_user(buffer, link, len)) 44901da177e4SLinus Torvalds len = -EFAULT; 44911da177e4SLinus Torvalds out: 44921da177e4SLinus Torvalds return len; 44931da177e4SLinus Torvalds } 44945d826c84SAl Viro EXPORT_SYMBOL(readlink_copy); 44951da177e4SLinus Torvalds 44961da177e4SLinus Torvalds /* 44971da177e4SLinus Torvalds * A helper for ->readlink(). This should be used *ONLY* for symlinks that 44981da177e4SLinus Torvalds * have ->follow_link() touching nd only in nd_set_link(). Using (or not 44991da177e4SLinus Torvalds * using) it for any given inode is up to filesystem. 45001da177e4SLinus Torvalds */ 45011da177e4SLinus Torvalds int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen) 45021da177e4SLinus Torvalds { 4503cc314eefSLinus Torvalds void *cookie; 45045f2c4179SAl Viro struct inode *inode = d_inode(dentry); 45055f2c4179SAl Viro const char *link = inode->i_link; 4506694a1764SMarcin Slusarz int res; 4507cc314eefSLinus Torvalds 4508d4dee48bSAl Viro if (!link) { 45095f2c4179SAl Viro link = inode->i_op->follow_link(dentry, &cookie); 4510680baacbSAl Viro if (IS_ERR(link)) 4511680baacbSAl Viro return PTR_ERR(link); 4512d4dee48bSAl Viro } 4513680baacbSAl Viro res = readlink_copy(buffer, buflen, link); 45145f2c4179SAl Viro if (inode->i_op->put_link) 45155f2c4179SAl Viro inode->i_op->put_link(inode, cookie); 4516694a1764SMarcin Slusarz return res; 45171da177e4SLinus Torvalds } 45184d359507SAl Viro EXPORT_SYMBOL(generic_readlink); 45191da177e4SLinus Torvalds 45201da177e4SLinus Torvalds /* get the link contents into pagecache */ 45211da177e4SLinus Torvalds static char *page_getlink(struct dentry * dentry, struct page **ppage) 45221da177e4SLinus Torvalds { 4523ebd09abbSDuane Griffin char *kaddr; 45241da177e4SLinus Torvalds struct page *page; 45251da177e4SLinus Torvalds struct address_space *mapping = dentry->d_inode->i_mapping; 4526090d2b18SPekka Enberg page = read_mapping_page(mapping, 0, NULL); 45271da177e4SLinus Torvalds if (IS_ERR(page)) 45286fe6900eSNick Piggin return (char*)page; 45291da177e4SLinus Torvalds *ppage = page; 4530ebd09abbSDuane Griffin kaddr = kmap(page); 4531ebd09abbSDuane Griffin nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1); 4532ebd09abbSDuane Griffin return kaddr; 45331da177e4SLinus Torvalds } 45341da177e4SLinus Torvalds 45351da177e4SLinus Torvalds int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 45361da177e4SLinus Torvalds { 45371da177e4SLinus Torvalds struct page *page = NULL; 45385d826c84SAl Viro int res = readlink_copy(buffer, buflen, page_getlink(dentry, &page)); 45391da177e4SLinus Torvalds if (page) { 45401da177e4SLinus Torvalds kunmap(page); 45411da177e4SLinus Torvalds page_cache_release(page); 45421da177e4SLinus Torvalds } 45431da177e4SLinus Torvalds return res; 45441da177e4SLinus Torvalds } 45454d359507SAl Viro EXPORT_SYMBOL(page_readlink); 45461da177e4SLinus Torvalds 45476e77137bSAl Viro const char *page_follow_link_light(struct dentry *dentry, void **cookie) 45481da177e4SLinus Torvalds { 4549cc314eefSLinus Torvalds struct page *page = NULL; 4550680baacbSAl Viro char *res = page_getlink(dentry, &page); 4551680baacbSAl Viro if (!IS_ERR(res)) 4552680baacbSAl Viro *cookie = page; 4553680baacbSAl Viro return res; 45541da177e4SLinus Torvalds } 45554d359507SAl Viro EXPORT_SYMBOL(page_follow_link_light); 45561da177e4SLinus Torvalds 45575f2c4179SAl Viro void page_put_link(struct inode *unused, void *cookie) 45581da177e4SLinus Torvalds { 4559cc314eefSLinus Torvalds struct page *page = cookie; 45601da177e4SLinus Torvalds kunmap(page); 45611da177e4SLinus Torvalds page_cache_release(page); 45621da177e4SLinus Torvalds } 45634d359507SAl Viro EXPORT_SYMBOL(page_put_link); 45641da177e4SLinus Torvalds 456554566b2cSNick Piggin /* 456654566b2cSNick Piggin * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS 456754566b2cSNick Piggin */ 456854566b2cSNick Piggin int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) 45691da177e4SLinus Torvalds { 45701da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 45710adb25d2SKirill Korotaev struct page *page; 4572afddba49SNick Piggin void *fsdata; 4573beb497abSDmitriy Monakhov int err; 45741da177e4SLinus Torvalds char *kaddr; 457554566b2cSNick Piggin unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE; 457654566b2cSNick Piggin if (nofs) 457754566b2cSNick Piggin flags |= AOP_FLAG_NOFS; 45781da177e4SLinus Torvalds 45797e53cac4SNeilBrown retry: 4580afddba49SNick Piggin err = pagecache_write_begin(NULL, mapping, 0, len-1, 458154566b2cSNick Piggin flags, &page, &fsdata); 45821da177e4SLinus Torvalds if (err) 4583afddba49SNick Piggin goto fail; 4584afddba49SNick Piggin 4585e8e3c3d6SCong Wang kaddr = kmap_atomic(page); 45861da177e4SLinus Torvalds memcpy(kaddr, symname, len-1); 4587e8e3c3d6SCong Wang kunmap_atomic(kaddr); 4588afddba49SNick Piggin 4589afddba49SNick Piggin err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, 4590afddba49SNick Piggin page, fsdata); 45911da177e4SLinus Torvalds if (err < 0) 45921da177e4SLinus Torvalds goto fail; 4593afddba49SNick Piggin if (err < len-1) 4594afddba49SNick Piggin goto retry; 4595afddba49SNick Piggin 45961da177e4SLinus Torvalds mark_inode_dirty(inode); 45971da177e4SLinus Torvalds return 0; 45981da177e4SLinus Torvalds fail: 45991da177e4SLinus Torvalds return err; 46001da177e4SLinus Torvalds } 46014d359507SAl Viro EXPORT_SYMBOL(__page_symlink); 46021da177e4SLinus Torvalds 46030adb25d2SKirill Korotaev int page_symlink(struct inode *inode, const char *symname, int len) 46040adb25d2SKirill Korotaev { 46050adb25d2SKirill Korotaev return __page_symlink(inode, symname, len, 4606c62d2555SMichal Hocko !mapping_gfp_constraint(inode->i_mapping, __GFP_FS)); 46070adb25d2SKirill Korotaev } 46084d359507SAl Viro EXPORT_SYMBOL(page_symlink); 46090adb25d2SKirill Korotaev 461092e1d5beSArjan van de Ven const struct inode_operations page_symlink_inode_operations = { 46111da177e4SLinus Torvalds .readlink = generic_readlink, 46121da177e4SLinus Torvalds .follow_link = page_follow_link_light, 46131da177e4SLinus Torvalds .put_link = page_put_link, 46141da177e4SLinus Torvalds }; 46151da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink_inode_operations); 4616