11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * linux/fs/namei.c 31da177e4SLinus Torvalds * 41da177e4SLinus Torvalds * Copyright (C) 1991, 1992 Linus Torvalds 51da177e4SLinus Torvalds */ 61da177e4SLinus Torvalds 71da177e4SLinus Torvalds /* 81da177e4SLinus Torvalds * Some corrections by tytso. 91da177e4SLinus Torvalds */ 101da177e4SLinus Torvalds 111da177e4SLinus Torvalds /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname 121da177e4SLinus Torvalds * lookup logic. 131da177e4SLinus Torvalds */ 141da177e4SLinus Torvalds /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture. 151da177e4SLinus Torvalds */ 161da177e4SLinus Torvalds 171da177e4SLinus Torvalds #include <linux/init.h> 18630d9c47SPaul Gortmaker #include <linux/export.h> 1944696908SDavid S. Miller #include <linux/kernel.h> 201da177e4SLinus Torvalds #include <linux/slab.h> 211da177e4SLinus Torvalds #include <linux/fs.h> 221da177e4SLinus Torvalds #include <linux/namei.h> 231da177e4SLinus Torvalds #include <linux/pagemap.h> 240eeca283SRobert Love #include <linux/fsnotify.h> 251da177e4SLinus Torvalds #include <linux/personality.h> 261da177e4SLinus Torvalds #include <linux/security.h> 276146f0d5SMimi Zohar #include <linux/ima.h> 281da177e4SLinus Torvalds #include <linux/syscalls.h> 291da177e4SLinus Torvalds #include <linux/mount.h> 301da177e4SLinus Torvalds #include <linux/audit.h> 3116f7e0feSRandy Dunlap #include <linux/capability.h> 32834f2a4aSTrond Myklebust #include <linux/file.h> 335590ff0dSUlrich Drepper #include <linux/fcntl.h> 3408ce5f16SSerge E. Hallyn #include <linux/device_cgroup.h> 355ad4e53bSAl Viro #include <linux/fs_struct.h> 36e77819e5SLinus Torvalds #include <linux/posix_acl.h> 3799d263d4SLinus Torvalds #include <linux/hash.h> 381da177e4SLinus Torvalds #include <asm/uaccess.h> 391da177e4SLinus Torvalds 40e81e3f4dSEric Paris #include "internal.h" 41c7105365SAl Viro #include "mount.h" 42e81e3f4dSEric Paris 431da177e4SLinus Torvalds /* [Feb-1997 T. Schoebel-Theuer] 441da177e4SLinus Torvalds * Fundamental changes in the pathname lookup mechanisms (namei) 451da177e4SLinus Torvalds * were necessary because of omirr. The reason is that omirr needs 461da177e4SLinus Torvalds * to know the _real_ pathname, not the user-supplied one, in case 471da177e4SLinus Torvalds * of symlinks (and also when transname replacements occur). 481da177e4SLinus Torvalds * 491da177e4SLinus Torvalds * The new code replaces the old recursive symlink resolution with 501da177e4SLinus Torvalds * an iterative one (in case of non-nested symlink chains). It does 511da177e4SLinus Torvalds * this with calls to <fs>_follow_link(). 521da177e4SLinus Torvalds * As a side effect, dir_namei(), _namei() and follow_link() are now 531da177e4SLinus Torvalds * replaced with a single function lookup_dentry() that can handle all 541da177e4SLinus Torvalds * the special cases of the former code. 551da177e4SLinus Torvalds * 561da177e4SLinus Torvalds * With the new dcache, the pathname is stored at each inode, at least as 571da177e4SLinus Torvalds * long as the refcount of the inode is positive. As a side effect, the 581da177e4SLinus Torvalds * size of the dcache depends on the inode cache and thus is dynamic. 591da177e4SLinus Torvalds * 601da177e4SLinus Torvalds * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink 611da177e4SLinus Torvalds * resolution to correspond with current state of the code. 621da177e4SLinus Torvalds * 631da177e4SLinus Torvalds * Note that the symlink resolution is not *completely* iterative. 641da177e4SLinus Torvalds * There is still a significant amount of tail- and mid- recursion in 651da177e4SLinus Torvalds * the algorithm. Also, note that <fs>_readlink() is not used in 661da177e4SLinus Torvalds * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink() 671da177e4SLinus Torvalds * may return different results than <fs>_follow_link(). Many virtual 681da177e4SLinus Torvalds * filesystems (including /proc) exhibit this behavior. 691da177e4SLinus Torvalds */ 701da177e4SLinus Torvalds 711da177e4SLinus Torvalds /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation: 721da177e4SLinus Torvalds * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL 731da177e4SLinus Torvalds * and the name already exists in form of a symlink, try to create the new 741da177e4SLinus Torvalds * name indicated by the symlink. The old code always complained that the 751da177e4SLinus Torvalds * name already exists, due to not following the symlink even if its target 761da177e4SLinus Torvalds * is nonexistent. The new semantics affects also mknod() and link() when 7725985edcSLucas De Marchi * the name is a symlink pointing to a non-existent name. 781da177e4SLinus Torvalds * 791da177e4SLinus Torvalds * I don't know which semantics is the right one, since I have no access 801da177e4SLinus Torvalds * to standards. But I found by trial that HP-UX 9.0 has the full "new" 811da177e4SLinus Torvalds * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the 821da177e4SLinus Torvalds * "old" one. Personally, I think the new semantics is much more logical. 831da177e4SLinus Torvalds * Note that "ln old new" where "new" is a symlink pointing to a non-existing 841da177e4SLinus Torvalds * file does succeed in both HP-UX and SunOs, but not in Solaris 851da177e4SLinus Torvalds * and in the old Linux semantics. 861da177e4SLinus Torvalds */ 871da177e4SLinus Torvalds 881da177e4SLinus Torvalds /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink 891da177e4SLinus Torvalds * semantics. See the comments in "open_namei" and "do_link" below. 901da177e4SLinus Torvalds * 911da177e4SLinus Torvalds * [10-Sep-98 Alan Modra] Another symlink change. 921da177e4SLinus Torvalds */ 931da177e4SLinus Torvalds 941da177e4SLinus Torvalds /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks: 951da177e4SLinus Torvalds * inside the path - always follow. 961da177e4SLinus Torvalds * in the last component in creation/removal/renaming - never follow. 971da177e4SLinus Torvalds * if LOOKUP_FOLLOW passed - follow. 981da177e4SLinus Torvalds * if the pathname has trailing slashes - follow. 991da177e4SLinus Torvalds * otherwise - don't follow. 1001da177e4SLinus Torvalds * (applied in that order). 1011da177e4SLinus Torvalds * 1021da177e4SLinus Torvalds * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT 1031da177e4SLinus Torvalds * restored for 2.4. This is the last surviving part of old 4.2BSD bug. 1041da177e4SLinus Torvalds * During the 2.4 we need to fix the userland stuff depending on it - 1051da177e4SLinus Torvalds * hopefully we will be able to get rid of that wart in 2.5. So far only 1061da177e4SLinus Torvalds * XEmacs seems to be relying on it... 1071da177e4SLinus Torvalds */ 1081da177e4SLinus Torvalds /* 1091da177e4SLinus Torvalds * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland) 110a11f3a05SArjan van de Ven * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives 1111da177e4SLinus Torvalds * any extra contention... 1121da177e4SLinus Torvalds */ 1131da177e4SLinus Torvalds 1141da177e4SLinus Torvalds /* In order to reduce some races, while at the same time doing additional 1151da177e4SLinus Torvalds * checking and hopefully speeding things up, we copy filenames to the 1161da177e4SLinus Torvalds * kernel data space before using them.. 1171da177e4SLinus Torvalds * 1181da177e4SLinus Torvalds * POSIX.1 2.4: an empty pathname is invalid (ENOENT). 1191da177e4SLinus Torvalds * PATH_MAX includes the nul terminator --RR. 1201da177e4SLinus Torvalds */ 1217950e385SJeff Layton 122fd2f7cb5SAl Viro #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname)) 12391a27b2aSJeff Layton 12451f39a1fSDavid Drysdale struct filename * 12591a27b2aSJeff Layton getname_flags(const char __user *filename, int flags, int *empty) 12691a27b2aSJeff Layton { 12794b5d262SAl Viro struct filename *result; 1287950e385SJeff Layton char *kname; 12994b5d262SAl Viro int len; 1301da177e4SLinus Torvalds 1317ac86265SJeff Layton result = audit_reusename(filename); 1327ac86265SJeff Layton if (result) 1337ac86265SJeff Layton return result; 1347ac86265SJeff Layton 1357950e385SJeff Layton result = __getname(); 1363f9f0aa6SLinus Torvalds if (unlikely(!result)) 1374043cde8SEric Paris return ERR_PTR(-ENOMEM); 1381da177e4SLinus Torvalds 1397950e385SJeff Layton /* 1407950e385SJeff Layton * First, try to embed the struct filename inside the names_cache 1417950e385SJeff Layton * allocation 1427950e385SJeff Layton */ 143fd2f7cb5SAl Viro kname = (char *)result->iname; 14491a27b2aSJeff Layton result->name = kname; 1457950e385SJeff Layton 14694b5d262SAl Viro len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX); 14791a27b2aSJeff Layton if (unlikely(len < 0)) { 14894b5d262SAl Viro __putname(result); 14994b5d262SAl Viro return ERR_PTR(len); 15091a27b2aSJeff Layton } 1513f9f0aa6SLinus Torvalds 1527950e385SJeff Layton /* 1537950e385SJeff Layton * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a 1547950e385SJeff Layton * separate struct filename so we can dedicate the entire 1557950e385SJeff Layton * names_cache allocation for the pathname, and re-do the copy from 1567950e385SJeff Layton * userland. 1577950e385SJeff Layton */ 15894b5d262SAl Viro if (unlikely(len == EMBEDDED_NAME_MAX)) { 159fd2f7cb5SAl Viro const size_t size = offsetof(struct filename, iname[1]); 1607950e385SJeff Layton kname = (char *)result; 1617950e385SJeff Layton 162fd2f7cb5SAl Viro /* 163fd2f7cb5SAl Viro * size is chosen that way we to guarantee that 164fd2f7cb5SAl Viro * result->iname[0] is within the same object and that 165fd2f7cb5SAl Viro * kname can't be equal to result->iname, no matter what. 166fd2f7cb5SAl Viro */ 167fd2f7cb5SAl Viro result = kzalloc(size, GFP_KERNEL); 16894b5d262SAl Viro if (unlikely(!result)) { 16994b5d262SAl Viro __putname(kname); 17094b5d262SAl Viro return ERR_PTR(-ENOMEM); 1717950e385SJeff Layton } 1727950e385SJeff Layton result->name = kname; 17394b5d262SAl Viro len = strncpy_from_user(kname, filename, PATH_MAX); 17494b5d262SAl Viro if (unlikely(len < 0)) { 17594b5d262SAl Viro __putname(kname); 17694b5d262SAl Viro kfree(result); 17794b5d262SAl Viro return ERR_PTR(len); 17894b5d262SAl Viro } 17994b5d262SAl Viro if (unlikely(len == PATH_MAX)) { 18094b5d262SAl Viro __putname(kname); 18194b5d262SAl Viro kfree(result); 18294b5d262SAl Viro return ERR_PTR(-ENAMETOOLONG); 18394b5d262SAl Viro } 1847950e385SJeff Layton } 1857950e385SJeff Layton 18694b5d262SAl Viro result->refcnt = 1; 1873f9f0aa6SLinus Torvalds /* The empty path is special. */ 1883f9f0aa6SLinus Torvalds if (unlikely(!len)) { 1893f9f0aa6SLinus Torvalds if (empty) 1901fa1e7f6SAndy Whitcroft *empty = 1; 19194b5d262SAl Viro if (!(flags & LOOKUP_EMPTY)) { 19294b5d262SAl Viro putname(result); 19394b5d262SAl Viro return ERR_PTR(-ENOENT); 1941da177e4SLinus Torvalds } 19594b5d262SAl Viro } 1967950e385SJeff Layton 1977950e385SJeff Layton result->uptr = filename; 198c4ad8f98SLinus Torvalds result->aname = NULL; 1991da177e4SLinus Torvalds audit_getname(result); 2001da177e4SLinus Torvalds return result; 2013f9f0aa6SLinus Torvalds } 2023f9f0aa6SLinus Torvalds 20391a27b2aSJeff Layton struct filename * 20491a27b2aSJeff Layton getname(const char __user * filename) 205f52e0c11SAl Viro { 206f7493e5dSLinus Torvalds return getname_flags(filename, 0, NULL); 207f52e0c11SAl Viro } 208f52e0c11SAl Viro 209c4ad8f98SLinus Torvalds struct filename * 210c4ad8f98SLinus Torvalds getname_kernel(const char * filename) 211c4ad8f98SLinus Torvalds { 212c4ad8f98SLinus Torvalds struct filename *result; 21308518549SPaul Moore int len = strlen(filename) + 1; 214c4ad8f98SLinus Torvalds 215c4ad8f98SLinus Torvalds result = __getname(); 216c4ad8f98SLinus Torvalds if (unlikely(!result)) 217c4ad8f98SLinus Torvalds return ERR_PTR(-ENOMEM); 218c4ad8f98SLinus Torvalds 21908518549SPaul Moore if (len <= EMBEDDED_NAME_MAX) { 220fd2f7cb5SAl Viro result->name = (char *)result->iname; 22108518549SPaul Moore } else if (len <= PATH_MAX) { 22208518549SPaul Moore struct filename *tmp; 22308518549SPaul Moore 22408518549SPaul Moore tmp = kmalloc(sizeof(*tmp), GFP_KERNEL); 22508518549SPaul Moore if (unlikely(!tmp)) { 22608518549SPaul Moore __putname(result); 22708518549SPaul Moore return ERR_PTR(-ENOMEM); 22808518549SPaul Moore } 22908518549SPaul Moore tmp->name = (char *)result; 23008518549SPaul Moore result = tmp; 23108518549SPaul Moore } else { 23208518549SPaul Moore __putname(result); 23308518549SPaul Moore return ERR_PTR(-ENAMETOOLONG); 23408518549SPaul Moore } 23508518549SPaul Moore memcpy((char *)result->name, filename, len); 236c4ad8f98SLinus Torvalds result->uptr = NULL; 237c4ad8f98SLinus Torvalds result->aname = NULL; 23855422d0bSPaul Moore result->refcnt = 1; 239fd3522fdSPaul Moore audit_getname(result); 240c4ad8f98SLinus Torvalds 241c4ad8f98SLinus Torvalds return result; 242c4ad8f98SLinus Torvalds } 243c4ad8f98SLinus Torvalds 24491a27b2aSJeff Layton void putname(struct filename *name) 2451da177e4SLinus Torvalds { 24655422d0bSPaul Moore BUG_ON(name->refcnt <= 0); 24755422d0bSPaul Moore 24855422d0bSPaul Moore if (--name->refcnt > 0) 24955422d0bSPaul Moore return; 25055422d0bSPaul Moore 251fd2f7cb5SAl Viro if (name->name != name->iname) { 25255422d0bSPaul Moore __putname(name->name); 25355422d0bSPaul Moore kfree(name); 25455422d0bSPaul Moore } else 25555422d0bSPaul Moore __putname(name); 2561da177e4SLinus Torvalds } 2571da177e4SLinus Torvalds 258e77819e5SLinus Torvalds static int check_acl(struct inode *inode, int mask) 259e77819e5SLinus Torvalds { 26084635d68SLinus Torvalds #ifdef CONFIG_FS_POSIX_ACL 261e77819e5SLinus Torvalds struct posix_acl *acl; 262e77819e5SLinus Torvalds 263e77819e5SLinus Torvalds if (mask & MAY_NOT_BLOCK) { 2643567866bSAl Viro acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS); 2653567866bSAl Viro if (!acl) 266e77819e5SLinus Torvalds return -EAGAIN; 2673567866bSAl Viro /* no ->get_acl() calls in RCU mode... */ 2683567866bSAl Viro if (acl == ACL_NOT_CACHED) 269e77819e5SLinus Torvalds return -ECHILD; 270206b1d09SAri Savolainen return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK); 271e77819e5SLinus Torvalds } 272e77819e5SLinus Torvalds 2732982baa2SChristoph Hellwig acl = get_acl(inode, ACL_TYPE_ACCESS); 2744e34e719SChristoph Hellwig if (IS_ERR(acl)) 2754e34e719SChristoph Hellwig return PTR_ERR(acl); 276e77819e5SLinus Torvalds if (acl) { 277e77819e5SLinus Torvalds int error = posix_acl_permission(inode, acl, mask); 278e77819e5SLinus Torvalds posix_acl_release(acl); 279e77819e5SLinus Torvalds return error; 280e77819e5SLinus Torvalds } 28184635d68SLinus Torvalds #endif 282e77819e5SLinus Torvalds 283e77819e5SLinus Torvalds return -EAGAIN; 284e77819e5SLinus Torvalds } 285e77819e5SLinus Torvalds 2865909ccaaSLinus Torvalds /* 287948409c7SAndreas Gruenbacher * This does the basic permission checking 2885909ccaaSLinus Torvalds */ 2897e40145eSAl Viro static int acl_permission_check(struct inode *inode, int mask) 2905909ccaaSLinus Torvalds { 29126cf46beSLinus Torvalds unsigned int mode = inode->i_mode; 2925909ccaaSLinus Torvalds 2938e96e3b7SEric W. Biederman if (likely(uid_eq(current_fsuid(), inode->i_uid))) 2945909ccaaSLinus Torvalds mode >>= 6; 2955909ccaaSLinus Torvalds else { 296e77819e5SLinus Torvalds if (IS_POSIXACL(inode) && (mode & S_IRWXG)) { 2977e40145eSAl Viro int error = check_acl(inode, mask); 2985909ccaaSLinus Torvalds if (error != -EAGAIN) 2995909ccaaSLinus Torvalds return error; 3005909ccaaSLinus Torvalds } 3015909ccaaSLinus Torvalds 3025909ccaaSLinus Torvalds if (in_group_p(inode->i_gid)) 3035909ccaaSLinus Torvalds mode >>= 3; 3045909ccaaSLinus Torvalds } 3055909ccaaSLinus Torvalds 3065909ccaaSLinus Torvalds /* 3075909ccaaSLinus Torvalds * If the DACs are ok we don't need any capability check. 3085909ccaaSLinus Torvalds */ 3099c2c7039SAl Viro if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) 3105909ccaaSLinus Torvalds return 0; 3115909ccaaSLinus Torvalds return -EACCES; 3125909ccaaSLinus Torvalds } 3131da177e4SLinus Torvalds 3141da177e4SLinus Torvalds /** 3151da177e4SLinus Torvalds * generic_permission - check for access rights on a Posix-like filesystem 3161da177e4SLinus Torvalds * @inode: inode to check access rights for 3178fd90c8dSAndreas Gruenbacher * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...) 3181da177e4SLinus Torvalds * 3191da177e4SLinus Torvalds * Used to check for read/write/execute permissions on a file. 3201da177e4SLinus Torvalds * We use "fsuid" for this, letting us set arbitrary permissions 3211da177e4SLinus Torvalds * for filesystem access without changing the "normal" uids which 322b74c79e9SNick Piggin * are used for other things. 323b74c79e9SNick Piggin * 324b74c79e9SNick Piggin * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk 325b74c79e9SNick Piggin * request cannot be satisfied (eg. requires blocking or too much complexity). 326b74c79e9SNick Piggin * It would then be called again in ref-walk mode. 3271da177e4SLinus Torvalds */ 3282830ba7fSAl Viro int generic_permission(struct inode *inode, int mask) 3291da177e4SLinus Torvalds { 3305909ccaaSLinus Torvalds int ret; 3311da177e4SLinus Torvalds 3321da177e4SLinus Torvalds /* 333948409c7SAndreas Gruenbacher * Do the basic permission checks. 3341da177e4SLinus Torvalds */ 3357e40145eSAl Viro ret = acl_permission_check(inode, mask); 3365909ccaaSLinus Torvalds if (ret != -EACCES) 3375909ccaaSLinus Torvalds return ret; 3381da177e4SLinus Torvalds 339d594e7ecSAl Viro if (S_ISDIR(inode->i_mode)) { 340d594e7ecSAl Viro /* DACs are overridable for directories */ 34123adbe12SAndy Lutomirski if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE)) 342d594e7ecSAl Viro return 0; 343d594e7ecSAl Viro if (!(mask & MAY_WRITE)) 34423adbe12SAndy Lutomirski if (capable_wrt_inode_uidgid(inode, 34523adbe12SAndy Lutomirski CAP_DAC_READ_SEARCH)) 346d594e7ecSAl Viro return 0; 347d594e7ecSAl Viro return -EACCES; 348d594e7ecSAl Viro } 3491da177e4SLinus Torvalds /* 3501da177e4SLinus Torvalds * Read/write DACs are always overridable. 351d594e7ecSAl Viro * Executable DACs are overridable when there is 352d594e7ecSAl Viro * at least one exec bit set. 3531da177e4SLinus Torvalds */ 354d594e7ecSAl Viro if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO)) 35523adbe12SAndy Lutomirski if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE)) 3561da177e4SLinus Torvalds return 0; 3571da177e4SLinus Torvalds 3581da177e4SLinus Torvalds /* 3591da177e4SLinus Torvalds * Searching includes executable on directories, else just read. 3601da177e4SLinus Torvalds */ 3617ea66001SSerge E. Hallyn mask &= MAY_READ | MAY_WRITE | MAY_EXEC; 362d594e7ecSAl Viro if (mask == MAY_READ) 36323adbe12SAndy Lutomirski if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH)) 3641da177e4SLinus Torvalds return 0; 3651da177e4SLinus Torvalds 3661da177e4SLinus Torvalds return -EACCES; 3671da177e4SLinus Torvalds } 3684d359507SAl Viro EXPORT_SYMBOL(generic_permission); 3691da177e4SLinus Torvalds 3703ddcd056SLinus Torvalds /* 3713ddcd056SLinus Torvalds * We _really_ want to just do "generic_permission()" without 3723ddcd056SLinus Torvalds * even looking at the inode->i_op values. So we keep a cache 3733ddcd056SLinus Torvalds * flag in inode->i_opflags, that says "this has not special 3743ddcd056SLinus Torvalds * permission function, use the fast case". 3753ddcd056SLinus Torvalds */ 3763ddcd056SLinus Torvalds static inline int do_inode_permission(struct inode *inode, int mask) 3773ddcd056SLinus Torvalds { 3783ddcd056SLinus Torvalds if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) { 3793ddcd056SLinus Torvalds if (likely(inode->i_op->permission)) 3803ddcd056SLinus Torvalds return inode->i_op->permission(inode, mask); 3813ddcd056SLinus Torvalds 3823ddcd056SLinus Torvalds /* This gets set once for the inode lifetime */ 3833ddcd056SLinus Torvalds spin_lock(&inode->i_lock); 3843ddcd056SLinus Torvalds inode->i_opflags |= IOP_FASTPERM; 3853ddcd056SLinus Torvalds spin_unlock(&inode->i_lock); 3863ddcd056SLinus Torvalds } 3873ddcd056SLinus Torvalds return generic_permission(inode, mask); 3883ddcd056SLinus Torvalds } 3893ddcd056SLinus Torvalds 390cb23beb5SChristoph Hellwig /** 3910bdaea90SDavid Howells * __inode_permission - Check for access rights to a given inode 3920bdaea90SDavid Howells * @inode: Inode to check permission on 3930bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 394cb23beb5SChristoph Hellwig * 3950bdaea90SDavid Howells * Check for read/write/execute permissions on an inode. 396948409c7SAndreas Gruenbacher * 397948409c7SAndreas Gruenbacher * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. 3980bdaea90SDavid Howells * 3990bdaea90SDavid Howells * This does not check for a read-only file system. You probably want 4000bdaea90SDavid Howells * inode_permission(). 401cb23beb5SChristoph Hellwig */ 4020bdaea90SDavid Howells int __inode_permission(struct inode *inode, int mask) 4031da177e4SLinus Torvalds { 404e6305c43SAl Viro int retval; 4051da177e4SLinus Torvalds 4063ddcd056SLinus Torvalds if (unlikely(mask & MAY_WRITE)) { 4071da177e4SLinus Torvalds /* 4081da177e4SLinus Torvalds * Nobody gets write access to an immutable file. 4091da177e4SLinus Torvalds */ 4101da177e4SLinus Torvalds if (IS_IMMUTABLE(inode)) 4111da177e4SLinus Torvalds return -EACCES; 4121da177e4SLinus Torvalds } 4131da177e4SLinus Torvalds 4143ddcd056SLinus Torvalds retval = do_inode_permission(inode, mask); 4151da177e4SLinus Torvalds if (retval) 4161da177e4SLinus Torvalds return retval; 4171da177e4SLinus Torvalds 41808ce5f16SSerge E. Hallyn retval = devcgroup_inode_permission(inode, mask); 41908ce5f16SSerge E. Hallyn if (retval) 42008ce5f16SSerge E. Hallyn return retval; 42108ce5f16SSerge E. Hallyn 422d09ca739SEric Paris return security_inode_permission(inode, mask); 4231da177e4SLinus Torvalds } 424bd5d0856SMiklos Szeredi EXPORT_SYMBOL(__inode_permission); 4251da177e4SLinus Torvalds 426f4d6ff89SAl Viro /** 4270bdaea90SDavid Howells * sb_permission - Check superblock-level permissions 4280bdaea90SDavid Howells * @sb: Superblock of inode to check permission on 42955852635SRandy Dunlap * @inode: Inode to check permission on 4300bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 4310bdaea90SDavid Howells * 4320bdaea90SDavid Howells * Separate out file-system wide checks from inode-specific permission checks. 4330bdaea90SDavid Howells */ 4340bdaea90SDavid Howells static int sb_permission(struct super_block *sb, struct inode *inode, int mask) 4350bdaea90SDavid Howells { 4360bdaea90SDavid Howells if (unlikely(mask & MAY_WRITE)) { 4370bdaea90SDavid Howells umode_t mode = inode->i_mode; 4380bdaea90SDavid Howells 4390bdaea90SDavid Howells /* Nobody gets write access to a read-only fs. */ 4400bdaea90SDavid Howells if ((sb->s_flags & MS_RDONLY) && 4410bdaea90SDavid Howells (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) 4420bdaea90SDavid Howells return -EROFS; 4430bdaea90SDavid Howells } 4440bdaea90SDavid Howells return 0; 4450bdaea90SDavid Howells } 4460bdaea90SDavid Howells 4470bdaea90SDavid Howells /** 4480bdaea90SDavid Howells * inode_permission - Check for access rights to a given inode 4490bdaea90SDavid Howells * @inode: Inode to check permission on 4500bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 4510bdaea90SDavid Howells * 4520bdaea90SDavid Howells * Check for read/write/execute permissions on an inode. We use fs[ug]id for 4530bdaea90SDavid Howells * this, letting us set arbitrary permissions for filesystem access without 4540bdaea90SDavid Howells * changing the "normal" UIDs which are used for other things. 4550bdaea90SDavid Howells * 4560bdaea90SDavid Howells * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. 4570bdaea90SDavid Howells */ 4580bdaea90SDavid Howells int inode_permission(struct inode *inode, int mask) 4590bdaea90SDavid Howells { 4600bdaea90SDavid Howells int retval; 4610bdaea90SDavid Howells 4620bdaea90SDavid Howells retval = sb_permission(inode->i_sb, inode, mask); 4630bdaea90SDavid Howells if (retval) 4640bdaea90SDavid Howells return retval; 4650bdaea90SDavid Howells return __inode_permission(inode, mask); 4660bdaea90SDavid Howells } 4674d359507SAl Viro EXPORT_SYMBOL(inode_permission); 4680bdaea90SDavid Howells 4690bdaea90SDavid Howells /** 4705dd784d0SJan Blunck * path_get - get a reference to a path 4715dd784d0SJan Blunck * @path: path to get the reference to 4725dd784d0SJan Blunck * 4735dd784d0SJan Blunck * Given a path increment the reference count to the dentry and the vfsmount. 4745dd784d0SJan Blunck */ 475dcf787f3SAl Viro void path_get(const struct path *path) 4765dd784d0SJan Blunck { 4775dd784d0SJan Blunck mntget(path->mnt); 4785dd784d0SJan Blunck dget(path->dentry); 4795dd784d0SJan Blunck } 4805dd784d0SJan Blunck EXPORT_SYMBOL(path_get); 4815dd784d0SJan Blunck 4825dd784d0SJan Blunck /** 4831d957f9bSJan Blunck * path_put - put a reference to a path 4841d957f9bSJan Blunck * @path: path to put the reference to 4851d957f9bSJan Blunck * 4861d957f9bSJan Blunck * Given a path decrement the reference count to the dentry and the vfsmount. 4871d957f9bSJan Blunck */ 488dcf787f3SAl Viro void path_put(const struct path *path) 4891da177e4SLinus Torvalds { 4901d957f9bSJan Blunck dput(path->dentry); 4911d957f9bSJan Blunck mntput(path->mnt); 4921da177e4SLinus Torvalds } 4931d957f9bSJan Blunck EXPORT_SYMBOL(path_put); 4941da177e4SLinus Torvalds 495894bc8c4SAl Viro #define EMBEDDED_LEVELS 2 4961f55a6ecSAl Viro struct nameidata { 4971f55a6ecSAl Viro struct path path; 4981f55a6ecSAl Viro struct qstr last; 4991f55a6ecSAl Viro struct path root; 5001f55a6ecSAl Viro struct inode *inode; /* path.dentry.d_inode */ 5011f55a6ecSAl Viro unsigned int flags; 5029883d185SAl Viro unsigned seq, m_seq; 5031f55a6ecSAl Viro int last_type; 5041f55a6ecSAl Viro unsigned depth; 505756daf26SNeilBrown int total_link_count; 506697fc6caSAl Viro struct saved { 507697fc6caSAl Viro struct path link; 508fceef393SAl Viro struct delayed_call done; 509697fc6caSAl Viro const char *name; 5100450b2d1SAl Viro unsigned seq; 511894bc8c4SAl Viro } *stack, internal[EMBEDDED_LEVELS]; 5129883d185SAl Viro struct filename *name; 5139883d185SAl Viro struct nameidata *saved; 514fceef393SAl Viro struct inode *link_inode; 5159883d185SAl Viro unsigned root_seq; 5169883d185SAl Viro int dfd; 5171f55a6ecSAl Viro }; 5181f55a6ecSAl Viro 5199883d185SAl Viro static void set_nameidata(struct nameidata *p, int dfd, struct filename *name) 520894bc8c4SAl Viro { 521756daf26SNeilBrown struct nameidata *old = current->nameidata; 522756daf26SNeilBrown p->stack = p->internal; 523c8a53ee5SAl Viro p->dfd = dfd; 524c8a53ee5SAl Viro p->name = name; 525756daf26SNeilBrown p->total_link_count = old ? old->total_link_count : 0; 5269883d185SAl Viro p->saved = old; 527756daf26SNeilBrown current->nameidata = p; 528894bc8c4SAl Viro } 529894bc8c4SAl Viro 5309883d185SAl Viro static void restore_nameidata(void) 531894bc8c4SAl Viro { 5329883d185SAl Viro struct nameidata *now = current->nameidata, *old = now->saved; 533756daf26SNeilBrown 534756daf26SNeilBrown current->nameidata = old; 535756daf26SNeilBrown if (old) 536756daf26SNeilBrown old->total_link_count = now->total_link_count; 537e1a63bbcSAl Viro if (now->stack != now->internal) 538756daf26SNeilBrown kfree(now->stack); 539894bc8c4SAl Viro } 540894bc8c4SAl Viro 541894bc8c4SAl Viro static int __nd_alloc_stack(struct nameidata *nd) 542894bc8c4SAl Viro { 543bc40aee0SAl Viro struct saved *p; 544bc40aee0SAl Viro 545bc40aee0SAl Viro if (nd->flags & LOOKUP_RCU) { 546bc40aee0SAl Viro p= kmalloc(MAXSYMLINKS * sizeof(struct saved), 547bc40aee0SAl Viro GFP_ATOMIC); 548bc40aee0SAl Viro if (unlikely(!p)) 549bc40aee0SAl Viro return -ECHILD; 550bc40aee0SAl Viro } else { 551bc40aee0SAl Viro p= kmalloc(MAXSYMLINKS * sizeof(struct saved), 552894bc8c4SAl Viro GFP_KERNEL); 553894bc8c4SAl Viro if (unlikely(!p)) 554894bc8c4SAl Viro return -ENOMEM; 555bc40aee0SAl Viro } 556894bc8c4SAl Viro memcpy(p, nd->internal, sizeof(nd->internal)); 557894bc8c4SAl Viro nd->stack = p; 558894bc8c4SAl Viro return 0; 559894bc8c4SAl Viro } 560894bc8c4SAl Viro 561397d425dSEric W. Biederman /** 562397d425dSEric W. Biederman * path_connected - Verify that a path->dentry is below path->mnt.mnt_root 563397d425dSEric W. Biederman * @path: nameidate to verify 564397d425dSEric W. Biederman * 565397d425dSEric W. Biederman * Rename can sometimes move a file or directory outside of a bind 566397d425dSEric W. Biederman * mount, path_connected allows those cases to be detected. 567397d425dSEric W. Biederman */ 568397d425dSEric W. Biederman static bool path_connected(const struct path *path) 569397d425dSEric W. Biederman { 570397d425dSEric W. Biederman struct vfsmount *mnt = path->mnt; 571397d425dSEric W. Biederman 572397d425dSEric W. Biederman /* Only bind mounts can have disconnected paths */ 573397d425dSEric W. Biederman if (mnt->mnt_root == mnt->mnt_sb->s_root) 574397d425dSEric W. Biederman return true; 575397d425dSEric W. Biederman 576397d425dSEric W. Biederman return is_subdir(path->dentry, mnt->mnt_root); 577397d425dSEric W. Biederman } 578397d425dSEric W. Biederman 579894bc8c4SAl Viro static inline int nd_alloc_stack(struct nameidata *nd) 580894bc8c4SAl Viro { 581da4e0be0SAl Viro if (likely(nd->depth != EMBEDDED_LEVELS)) 582894bc8c4SAl Viro return 0; 583894bc8c4SAl Viro if (likely(nd->stack != nd->internal)) 584894bc8c4SAl Viro return 0; 585894bc8c4SAl Viro return __nd_alloc_stack(nd); 586894bc8c4SAl Viro } 587894bc8c4SAl Viro 5887973387aSAl Viro static void drop_links(struct nameidata *nd) 5897973387aSAl Viro { 5907973387aSAl Viro int i = nd->depth; 5917973387aSAl Viro while (i--) { 5927973387aSAl Viro struct saved *last = nd->stack + i; 593fceef393SAl Viro do_delayed_call(&last->done); 594fceef393SAl Viro clear_delayed_call(&last->done); 5957973387aSAl Viro } 5967973387aSAl Viro } 5977973387aSAl Viro 5987973387aSAl Viro static void terminate_walk(struct nameidata *nd) 5997973387aSAl Viro { 6007973387aSAl Viro drop_links(nd); 6017973387aSAl Viro if (!(nd->flags & LOOKUP_RCU)) { 6027973387aSAl Viro int i; 6037973387aSAl Viro path_put(&nd->path); 6047973387aSAl Viro for (i = 0; i < nd->depth; i++) 6057973387aSAl Viro path_put(&nd->stack[i].link); 606102b8af2SAl Viro if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 607102b8af2SAl Viro path_put(&nd->root); 608102b8af2SAl Viro nd->root.mnt = NULL; 609102b8af2SAl Viro } 6107973387aSAl Viro } else { 6117973387aSAl Viro nd->flags &= ~LOOKUP_RCU; 6127973387aSAl Viro if (!(nd->flags & LOOKUP_ROOT)) 6137973387aSAl Viro nd->root.mnt = NULL; 6147973387aSAl Viro rcu_read_unlock(); 6157973387aSAl Viro } 6167973387aSAl Viro nd->depth = 0; 6177973387aSAl Viro } 6187973387aSAl Viro 6197973387aSAl Viro /* path_put is needed afterwards regardless of success or failure */ 6207973387aSAl Viro static bool legitimize_path(struct nameidata *nd, 6217973387aSAl Viro struct path *path, unsigned seq) 6227973387aSAl Viro { 6237973387aSAl Viro int res = __legitimize_mnt(path->mnt, nd->m_seq); 6247973387aSAl Viro if (unlikely(res)) { 6257973387aSAl Viro if (res > 0) 6267973387aSAl Viro path->mnt = NULL; 6277973387aSAl Viro path->dentry = NULL; 6287973387aSAl Viro return false; 6297973387aSAl Viro } 6307973387aSAl Viro if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) { 6317973387aSAl Viro path->dentry = NULL; 6327973387aSAl Viro return false; 6337973387aSAl Viro } 6347973387aSAl Viro return !read_seqcount_retry(&path->dentry->d_seq, seq); 6357973387aSAl Viro } 6367973387aSAl Viro 6377973387aSAl Viro static bool legitimize_links(struct nameidata *nd) 6387973387aSAl Viro { 6397973387aSAl Viro int i; 6407973387aSAl Viro for (i = 0; i < nd->depth; i++) { 6417973387aSAl Viro struct saved *last = nd->stack + i; 6427973387aSAl Viro if (unlikely(!legitimize_path(nd, &last->link, last->seq))) { 6437973387aSAl Viro drop_links(nd); 6447973387aSAl Viro nd->depth = i + 1; 6457973387aSAl Viro return false; 6467973387aSAl Viro } 6477973387aSAl Viro } 6487973387aSAl Viro return true; 6497973387aSAl Viro } 6507973387aSAl Viro 65119660af7SAl Viro /* 65231e6b01fSNick Piggin * Path walking has 2 modes, rcu-walk and ref-walk (see 65319660af7SAl Viro * Documentation/filesystems/path-lookup.txt). In situations when we can't 65419660af7SAl Viro * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab 65557e3715cSMike Marshall * normal reference counts on dentries and vfsmounts to transition to ref-walk 65619660af7SAl Viro * mode. Refcounts are grabbed at the last known good point before rcu-walk 65719660af7SAl Viro * got stuck, so ref-walk may continue from there. If this is not successful 65819660af7SAl Viro * (eg. a seqcount has changed), then failure is returned and it's up to caller 65919660af7SAl Viro * to restart the path walk from the beginning in ref-walk mode. 66031e6b01fSNick Piggin */ 66131e6b01fSNick Piggin 66231e6b01fSNick Piggin /** 66319660af7SAl Viro * unlazy_walk - try to switch to ref-walk mode. 66419660af7SAl Viro * @nd: nameidata pathwalk data 66519660af7SAl Viro * @dentry: child of nd->path.dentry or NULL 6666e9918b7SAl Viro * @seq: seq number to check dentry against 66739191628SRandy Dunlap * Returns: 0 on success, -ECHILD on failure 66831e6b01fSNick Piggin * 66919660af7SAl Viro * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry 67019660af7SAl Viro * for ref-walk mode. @dentry must be a path found by a do_lookup call on 67119660af7SAl Viro * @nd or NULL. Must be called from rcu-walk context. 6727973387aSAl Viro * Nothing should touch nameidata between unlazy_walk() failure and 6737973387aSAl Viro * terminate_walk(). 67431e6b01fSNick Piggin */ 6756e9918b7SAl Viro static int unlazy_walk(struct nameidata *nd, struct dentry *dentry, unsigned seq) 67631e6b01fSNick Piggin { 67731e6b01fSNick Piggin struct dentry *parent = nd->path.dentry; 67831e6b01fSNick Piggin 67931e6b01fSNick Piggin BUG_ON(!(nd->flags & LOOKUP_RCU)); 680e5c832d5SLinus Torvalds 681e5c832d5SLinus Torvalds nd->flags &= ~LOOKUP_RCU; 6827973387aSAl Viro if (unlikely(!legitimize_links(nd))) 6837973387aSAl Viro goto out2; 6847973387aSAl Viro if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq))) 6857973387aSAl Viro goto out2; 6867973387aSAl Viro if (unlikely(!lockref_get_not_dead(&parent->d_lockref))) 6877973387aSAl Viro goto out1; 68848a066e7SAl Viro 68915570086SLinus Torvalds /* 69015570086SLinus Torvalds * For a negative lookup, the lookup sequence point is the parents 69115570086SLinus Torvalds * sequence point, and it only needs to revalidate the parent dentry. 69215570086SLinus Torvalds * 69315570086SLinus Torvalds * For a positive lookup, we need to move both the parent and the 69415570086SLinus Torvalds * dentry from the RCU domain to be properly refcounted. And the 69515570086SLinus Torvalds * sequence number in the dentry validates *both* dentry counters, 69615570086SLinus Torvalds * since we checked the sequence number of the parent after we got 69715570086SLinus Torvalds * the child sequence number. So we know the parent must still 69815570086SLinus Torvalds * be valid if the child sequence number is still valid. 69915570086SLinus Torvalds */ 70019660af7SAl Viro if (!dentry) { 701e5c832d5SLinus Torvalds if (read_seqcount_retry(&parent->d_seq, nd->seq)) 702e5c832d5SLinus Torvalds goto out; 70319660af7SAl Viro BUG_ON(nd->inode != parent->d_inode); 70419660af7SAl Viro } else { 705e5c832d5SLinus Torvalds if (!lockref_get_not_dead(&dentry->d_lockref)) 706e5c832d5SLinus Torvalds goto out; 7076e9918b7SAl Viro if (read_seqcount_retry(&dentry->d_seq, seq)) 708e5c832d5SLinus Torvalds goto drop_dentry; 70919660af7SAl Viro } 710e5c832d5SLinus Torvalds 711e5c832d5SLinus Torvalds /* 712e5c832d5SLinus Torvalds * Sequence counts matched. Now make sure that the root is 713e5c832d5SLinus Torvalds * still valid and get it if required. 714e5c832d5SLinus Torvalds */ 715e5c832d5SLinus Torvalds if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 7165a8d87e8SAl Viro if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) { 7175a8d87e8SAl Viro rcu_read_unlock(); 7185a8d87e8SAl Viro dput(dentry); 7195a8d87e8SAl Viro return -ECHILD; 7207973387aSAl Viro } 72131e6b01fSNick Piggin } 72231e6b01fSNick Piggin 7238b61e74fSAl Viro rcu_read_unlock(); 72431e6b01fSNick Piggin return 0; 72519660af7SAl Viro 726e5c832d5SLinus Torvalds drop_dentry: 7278b61e74fSAl Viro rcu_read_unlock(); 728e5c832d5SLinus Torvalds dput(dentry); 729d0d27277SLinus Torvalds goto drop_root_mnt; 7307973387aSAl Viro out2: 7317973387aSAl Viro nd->path.mnt = NULL; 7327973387aSAl Viro out1: 7337973387aSAl Viro nd->path.dentry = NULL; 734e5c832d5SLinus Torvalds out: 7358b61e74fSAl Viro rcu_read_unlock(); 736d0d27277SLinus Torvalds drop_root_mnt: 737d0d27277SLinus Torvalds if (!(nd->flags & LOOKUP_ROOT)) 738d0d27277SLinus Torvalds nd->root.mnt = NULL; 73931e6b01fSNick Piggin return -ECHILD; 74031e6b01fSNick Piggin } 74131e6b01fSNick Piggin 7427973387aSAl Viro static int unlazy_link(struct nameidata *nd, struct path *link, unsigned seq) 7437973387aSAl Viro { 7447973387aSAl Viro if (unlikely(!legitimize_path(nd, link, seq))) { 7457973387aSAl Viro drop_links(nd); 7467973387aSAl Viro nd->depth = 0; 7477973387aSAl Viro nd->flags &= ~LOOKUP_RCU; 7487973387aSAl Viro nd->path.mnt = NULL; 7497973387aSAl Viro nd->path.dentry = NULL; 7507973387aSAl Viro if (!(nd->flags & LOOKUP_ROOT)) 7517973387aSAl Viro nd->root.mnt = NULL; 7527973387aSAl Viro rcu_read_unlock(); 7537973387aSAl Viro } else if (likely(unlazy_walk(nd, NULL, 0)) == 0) { 7547973387aSAl Viro return 0; 7557973387aSAl Viro } 7567973387aSAl Viro path_put(link); 7577973387aSAl Viro return -ECHILD; 7587973387aSAl Viro } 7597973387aSAl Viro 7604ce16ef3SAl Viro static inline int d_revalidate(struct dentry *dentry, unsigned int flags) 76134286d66SNick Piggin { 7624ce16ef3SAl Viro return dentry->d_op->d_revalidate(dentry, flags); 76334286d66SNick Piggin } 76434286d66SNick Piggin 7659f1fafeeSAl Viro /** 7669f1fafeeSAl Viro * complete_walk - successful completion of path walk 7679f1fafeeSAl Viro * @nd: pointer nameidata 76839159de2SJeff Layton * 7699f1fafeeSAl Viro * If we had been in RCU mode, drop out of it and legitimize nd->path. 7709f1fafeeSAl Viro * Revalidate the final result, unless we'd already done that during 7719f1fafeeSAl Viro * the path walk or the filesystem doesn't ask for it. Return 0 on 7729f1fafeeSAl Viro * success, -error on failure. In case of failure caller does not 7739f1fafeeSAl Viro * need to drop nd->path. 77439159de2SJeff Layton */ 7759f1fafeeSAl Viro static int complete_walk(struct nameidata *nd) 77639159de2SJeff Layton { 77716c2cd71SAl Viro struct dentry *dentry = nd->path.dentry; 77839159de2SJeff Layton int status; 77939159de2SJeff Layton 7809f1fafeeSAl Viro if (nd->flags & LOOKUP_RCU) { 7819f1fafeeSAl Viro if (!(nd->flags & LOOKUP_ROOT)) 7829f1fafeeSAl Viro nd->root.mnt = NULL; 7836e9918b7SAl Viro if (unlikely(unlazy_walk(nd, NULL, 0))) 78448a066e7SAl Viro return -ECHILD; 78548a066e7SAl Viro } 7869f1fafeeSAl Viro 78716c2cd71SAl Viro if (likely(!(nd->flags & LOOKUP_JUMPED))) 78839159de2SJeff Layton return 0; 78939159de2SJeff Layton 790ecf3d1f1SJeff Layton if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) 79116c2cd71SAl Viro return 0; 79216c2cd71SAl Viro 793ecf3d1f1SJeff Layton status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); 79439159de2SJeff Layton if (status > 0) 79539159de2SJeff Layton return 0; 79639159de2SJeff Layton 79716c2cd71SAl Viro if (!status) 79839159de2SJeff Layton status = -ESTALE; 79916c2cd71SAl Viro 80039159de2SJeff Layton return status; 80139159de2SJeff Layton } 80239159de2SJeff Layton 80318d8c860SAl Viro static void set_root(struct nameidata *nd) 8042a737871SAl Viro { 80531e6b01fSNick Piggin struct fs_struct *fs = current->fs; 8069e6697e2SAl Viro 8079e6697e2SAl Viro if (nd->flags & LOOKUP_RCU) { 8088f47a016SAl Viro unsigned seq; 809c28cc364SNick Piggin 810c28cc364SNick Piggin do { 811c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 81231e6b01fSNick Piggin nd->root = fs->root; 8138f47a016SAl Viro nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq); 814c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 8159e6697e2SAl Viro } else { 8169e6697e2SAl Viro get_fs_root(fs, &nd->root); 8179e6697e2SAl Viro } 81831e6b01fSNick Piggin } 81931e6b01fSNick Piggin 8201d957f9bSJan Blunck static void path_put_conditional(struct path *path, struct nameidata *nd) 821051d3812SIan Kent { 822051d3812SIan Kent dput(path->dentry); 8234ac91378SJan Blunck if (path->mnt != nd->path.mnt) 824051d3812SIan Kent mntput(path->mnt); 825051d3812SIan Kent } 826051d3812SIan Kent 8277b9337aaSNick Piggin static inline void path_to_nameidata(const struct path *path, 8287b9337aaSNick Piggin struct nameidata *nd) 829051d3812SIan Kent { 83031e6b01fSNick Piggin if (!(nd->flags & LOOKUP_RCU)) { 8314ac91378SJan Blunck dput(nd->path.dentry); 83231e6b01fSNick Piggin if (nd->path.mnt != path->mnt) 8334ac91378SJan Blunck mntput(nd->path.mnt); 8349a229683SHuang Shijie } 83531e6b01fSNick Piggin nd->path.mnt = path->mnt; 8364ac91378SJan Blunck nd->path.dentry = path->dentry; 837051d3812SIan Kent } 838051d3812SIan Kent 839248fb5b9SAl Viro static int nd_jump_root(struct nameidata *nd) 840248fb5b9SAl Viro { 841248fb5b9SAl Viro if (nd->flags & LOOKUP_RCU) { 842248fb5b9SAl Viro struct dentry *d; 843248fb5b9SAl Viro nd->path = nd->root; 844248fb5b9SAl Viro d = nd->path.dentry; 845248fb5b9SAl Viro nd->inode = d->d_inode; 846248fb5b9SAl Viro nd->seq = nd->root_seq; 847248fb5b9SAl Viro if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq))) 848248fb5b9SAl Viro return -ECHILD; 849248fb5b9SAl Viro } else { 850248fb5b9SAl Viro path_put(&nd->path); 851248fb5b9SAl Viro nd->path = nd->root; 852248fb5b9SAl Viro path_get(&nd->path); 853248fb5b9SAl Viro nd->inode = nd->path.dentry->d_inode; 854248fb5b9SAl Viro } 855248fb5b9SAl Viro nd->flags |= LOOKUP_JUMPED; 856248fb5b9SAl Viro return 0; 857248fb5b9SAl Viro } 858248fb5b9SAl Viro 859b5fb63c1SChristoph Hellwig /* 8606b255391SAl Viro * Helper to directly jump to a known parsed path from ->get_link, 861b5fb63c1SChristoph Hellwig * caller must have taken a reference to path beforehand. 862b5fb63c1SChristoph Hellwig */ 8636e77137bSAl Viro void nd_jump_link(struct path *path) 864b5fb63c1SChristoph Hellwig { 8656e77137bSAl Viro struct nameidata *nd = current->nameidata; 866b5fb63c1SChristoph Hellwig path_put(&nd->path); 867b5fb63c1SChristoph Hellwig 868b5fb63c1SChristoph Hellwig nd->path = *path; 869b5fb63c1SChristoph Hellwig nd->inode = nd->path.dentry->d_inode; 870b5fb63c1SChristoph Hellwig nd->flags |= LOOKUP_JUMPED; 871b5fb63c1SChristoph Hellwig } 872b5fb63c1SChristoph Hellwig 873b9ff4429SAl Viro static inline void put_link(struct nameidata *nd) 874574197e0SAl Viro { 87521c3003dSAl Viro struct saved *last = nd->stack + --nd->depth; 876fceef393SAl Viro do_delayed_call(&last->done); 8776548fae2SAl Viro if (!(nd->flags & LOOKUP_RCU)) 878b9ff4429SAl Viro path_put(&last->link); 879574197e0SAl Viro } 880574197e0SAl Viro 881561ec64aSLinus Torvalds int sysctl_protected_symlinks __read_mostly = 0; 882561ec64aSLinus Torvalds int sysctl_protected_hardlinks __read_mostly = 0; 883800179c9SKees Cook 884800179c9SKees Cook /** 885800179c9SKees Cook * may_follow_link - Check symlink following for unsafe situations 88655852635SRandy Dunlap * @nd: nameidata pathwalk data 887800179c9SKees Cook * 888800179c9SKees Cook * In the case of the sysctl_protected_symlinks sysctl being enabled, 889800179c9SKees Cook * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is 890800179c9SKees Cook * in a sticky world-writable directory. This is to protect privileged 891800179c9SKees Cook * processes from failing races against path names that may change out 892800179c9SKees Cook * from under them by way of other users creating malicious symlinks. 893800179c9SKees Cook * It will permit symlinks to be followed only when outside a sticky 894800179c9SKees Cook * world-writable directory, or when the uid of the symlink and follower 895800179c9SKees Cook * match, or when the directory owner matches the symlink's owner. 896800179c9SKees Cook * 897800179c9SKees Cook * Returns 0 if following the symlink is allowed, -ve on error. 898800179c9SKees Cook */ 899fec2fa24SAl Viro static inline int may_follow_link(struct nameidata *nd) 900800179c9SKees Cook { 901800179c9SKees Cook const struct inode *inode; 902800179c9SKees Cook const struct inode *parent; 903800179c9SKees Cook 904800179c9SKees Cook if (!sysctl_protected_symlinks) 905800179c9SKees Cook return 0; 906800179c9SKees Cook 907800179c9SKees Cook /* Allowed if owner and follower match. */ 908fceef393SAl Viro inode = nd->link_inode; 90981abe27bSEric W. Biederman if (uid_eq(current_cred()->fsuid, inode->i_uid)) 910800179c9SKees Cook return 0; 911800179c9SKees Cook 912800179c9SKees Cook /* Allowed if parent directory not sticky and world-writable. */ 913aa65fa35SAl Viro parent = nd->inode; 914800179c9SKees Cook if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) 915800179c9SKees Cook return 0; 916800179c9SKees Cook 917800179c9SKees Cook /* Allowed if parent directory and link owner match. */ 91881abe27bSEric W. Biederman if (uid_eq(parent->i_uid, inode->i_uid)) 919800179c9SKees Cook return 0; 920800179c9SKees Cook 92131956502SAl Viro if (nd->flags & LOOKUP_RCU) 92231956502SAl Viro return -ECHILD; 92331956502SAl Viro 9241cf2665bSAl Viro audit_log_link_denied("follow_link", &nd->stack[0].link); 925800179c9SKees Cook return -EACCES; 926800179c9SKees Cook } 927800179c9SKees Cook 928800179c9SKees Cook /** 929800179c9SKees Cook * safe_hardlink_source - Check for safe hardlink conditions 930800179c9SKees Cook * @inode: the source inode to hardlink from 931800179c9SKees Cook * 932800179c9SKees Cook * Return false if at least one of the following conditions: 933800179c9SKees Cook * - inode is not a regular file 934800179c9SKees Cook * - inode is setuid 935800179c9SKees Cook * - inode is setgid and group-exec 936800179c9SKees Cook * - access failure for read and write 937800179c9SKees Cook * 938800179c9SKees Cook * Otherwise returns true. 939800179c9SKees Cook */ 940800179c9SKees Cook static bool safe_hardlink_source(struct inode *inode) 941800179c9SKees Cook { 942800179c9SKees Cook umode_t mode = inode->i_mode; 943800179c9SKees Cook 944800179c9SKees Cook /* Special files should not get pinned to the filesystem. */ 945800179c9SKees Cook if (!S_ISREG(mode)) 946800179c9SKees Cook return false; 947800179c9SKees Cook 948800179c9SKees Cook /* Setuid files should not get pinned to the filesystem. */ 949800179c9SKees Cook if (mode & S_ISUID) 950800179c9SKees Cook return false; 951800179c9SKees Cook 952800179c9SKees Cook /* Executable setgid files should not get pinned to the filesystem. */ 953800179c9SKees Cook if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) 954800179c9SKees Cook return false; 955800179c9SKees Cook 956800179c9SKees Cook /* Hardlinking to unreadable or unwritable sources is dangerous. */ 957800179c9SKees Cook if (inode_permission(inode, MAY_READ | MAY_WRITE)) 958800179c9SKees Cook return false; 959800179c9SKees Cook 960800179c9SKees Cook return true; 961800179c9SKees Cook } 962800179c9SKees Cook 963800179c9SKees Cook /** 964800179c9SKees Cook * may_linkat - Check permissions for creating a hardlink 965800179c9SKees Cook * @link: the source to hardlink from 966800179c9SKees Cook * 967800179c9SKees Cook * Block hardlink when all of: 968800179c9SKees Cook * - sysctl_protected_hardlinks enabled 969800179c9SKees Cook * - fsuid does not match inode 970800179c9SKees Cook * - hardlink source is unsafe (see safe_hardlink_source() above) 971f2ca3796SDirk Steinmetz * - not CAP_FOWNER in a namespace with the inode owner uid mapped 972800179c9SKees Cook * 973800179c9SKees Cook * Returns 0 if successful, -ve on error. 974800179c9SKees Cook */ 975800179c9SKees Cook static int may_linkat(struct path *link) 976800179c9SKees Cook { 977800179c9SKees Cook struct inode *inode; 978800179c9SKees Cook 979800179c9SKees Cook if (!sysctl_protected_hardlinks) 980800179c9SKees Cook return 0; 981800179c9SKees Cook 982800179c9SKees Cook inode = link->dentry->d_inode; 983800179c9SKees Cook 984800179c9SKees Cook /* Source inode owner (or CAP_FOWNER) can hardlink all they like, 985800179c9SKees Cook * otherwise, it must be a safe source. 986800179c9SKees Cook */ 987f2ca3796SDirk Steinmetz if (inode_owner_or_capable(inode) || safe_hardlink_source(inode)) 988800179c9SKees Cook return 0; 989800179c9SKees Cook 990a51d9eaaSKees Cook audit_log_link_denied("linkat", link); 991800179c9SKees Cook return -EPERM; 992800179c9SKees Cook } 993800179c9SKees Cook 9943b2e7f75SAl Viro static __always_inline 9953b2e7f75SAl Viro const char *get_link(struct nameidata *nd) 9961da177e4SLinus Torvalds { 997ab104923SAl Viro struct saved *last = nd->stack + nd->depth - 1; 9981cf2665bSAl Viro struct dentry *dentry = last->link.dentry; 999fceef393SAl Viro struct inode *inode = nd->link_inode; 10006d7b5aaeSAl Viro int error; 10010a959df5SAl Viro const char *res; 10021da177e4SLinus Torvalds 10038fa9dd24SNeilBrown if (!(nd->flags & LOOKUP_RCU)) { 10048fa9dd24SNeilBrown touch_atime(&last->link); 10058fa9dd24SNeilBrown cond_resched(); 10068fa9dd24SNeilBrown } else if (atime_needs_update(&last->link, inode)) { 10078fa9dd24SNeilBrown if (unlikely(unlazy_walk(nd, NULL, 0))) 10088fa9dd24SNeilBrown return ERR_PTR(-ECHILD); 10098fa9dd24SNeilBrown touch_atime(&last->link); 10108fa9dd24SNeilBrown } 10118fa9dd24SNeilBrown 1012bda0be7aSNeilBrown error = security_inode_follow_link(dentry, inode, 1013bda0be7aSNeilBrown nd->flags & LOOKUP_RCU); 1014bda0be7aSNeilBrown if (unlikely(error)) 10156920a440SAl Viro return ERR_PTR(error); 101636f3b4f6SAl Viro 101786acdca1SAl Viro nd->last_type = LAST_BIND; 1018d4dee48bSAl Viro res = inode->i_link; 1019d4dee48bSAl Viro if (!res) { 1020fceef393SAl Viro const char * (*get)(struct dentry *, struct inode *, 1021fceef393SAl Viro struct delayed_call *); 1022fceef393SAl Viro get = inode->i_op->get_link; 10238c1b4566SAl Viro if (nd->flags & LOOKUP_RCU) { 1024fceef393SAl Viro res = get(NULL, inode, &last->done); 10256b255391SAl Viro if (res == ERR_PTR(-ECHILD)) { 10268c1b4566SAl Viro if (unlikely(unlazy_walk(nd, NULL, 0))) 10278c1b4566SAl Viro return ERR_PTR(-ECHILD); 1028fceef393SAl Viro res = get(dentry, inode, &last->done); 10298c1b4566SAl Viro } 10306b255391SAl Viro } else { 1031fceef393SAl Viro res = get(dentry, inode, &last->done); 10326b255391SAl Viro } 1033fceef393SAl Viro if (IS_ERR_OR_NULL(res)) 1034fab51e8aSAl Viro return res; 10350a959df5SAl Viro } 1036fab51e8aSAl Viro if (*res == '/') { 1037fab51e8aSAl Viro if (!nd->root.mnt) 1038fab51e8aSAl Viro set_root(nd); 1039248fb5b9SAl Viro if (unlikely(nd_jump_root(nd))) 1040fab51e8aSAl Viro return ERR_PTR(-ECHILD); 1041fab51e8aSAl Viro while (unlikely(*++res == '/')) 1042fab51e8aSAl Viro ; 1043fab51e8aSAl Viro } 1044fab51e8aSAl Viro if (!*res) 1045fab51e8aSAl Viro res = NULL; 10460a959df5SAl Viro return res; 10470a959df5SAl Viro } 10486d7b5aaeSAl Viro 1049f015f126SDavid Howells /* 1050f015f126SDavid Howells * follow_up - Find the mountpoint of path's vfsmount 1051f015f126SDavid Howells * 1052f015f126SDavid Howells * Given a path, find the mountpoint of its source file system. 1053f015f126SDavid Howells * Replace @path with the path of the mountpoint in the parent mount. 1054f015f126SDavid Howells * Up is towards /. 1055f015f126SDavid Howells * 1056f015f126SDavid Howells * Return 1 if we went up a level and 0 if we were already at the 1057f015f126SDavid Howells * root. 1058f015f126SDavid Howells */ 1059bab77ebfSAl Viro int follow_up(struct path *path) 10601da177e4SLinus Torvalds { 10610714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 10620714a533SAl Viro struct mount *parent; 10631da177e4SLinus Torvalds struct dentry *mountpoint; 106499b7db7bSNick Piggin 106548a066e7SAl Viro read_seqlock_excl(&mount_lock); 10660714a533SAl Viro parent = mnt->mnt_parent; 10673c0a6163SAl Viro if (parent == mnt) { 106848a066e7SAl Viro read_sequnlock_excl(&mount_lock); 10691da177e4SLinus Torvalds return 0; 10701da177e4SLinus Torvalds } 10710714a533SAl Viro mntget(&parent->mnt); 1072a73324daSAl Viro mountpoint = dget(mnt->mnt_mountpoint); 107348a066e7SAl Viro read_sequnlock_excl(&mount_lock); 1074bab77ebfSAl Viro dput(path->dentry); 1075bab77ebfSAl Viro path->dentry = mountpoint; 1076bab77ebfSAl Viro mntput(path->mnt); 10770714a533SAl Viro path->mnt = &parent->mnt; 10781da177e4SLinus Torvalds return 1; 10791da177e4SLinus Torvalds } 10804d359507SAl Viro EXPORT_SYMBOL(follow_up); 10811da177e4SLinus Torvalds 1082b5c84bf6SNick Piggin /* 10839875cf80SDavid Howells * Perform an automount 10849875cf80SDavid Howells * - return -EISDIR to tell follow_managed() to stop and return the path we 10859875cf80SDavid Howells * were called with. 10861da177e4SLinus Torvalds */ 1087756daf26SNeilBrown static int follow_automount(struct path *path, struct nameidata *nd, 10889875cf80SDavid Howells bool *need_mntput) 108931e6b01fSNick Piggin { 10909875cf80SDavid Howells struct vfsmount *mnt; 1091ea5b778aSDavid Howells int err; 10929875cf80SDavid Howells 10939875cf80SDavid Howells if (!path->dentry->d_op || !path->dentry->d_op->d_automount) 10949875cf80SDavid Howells return -EREMOTE; 10959875cf80SDavid Howells 10960ec26fd0SMiklos Szeredi /* We don't want to mount if someone's just doing a stat - 10970ec26fd0SMiklos Szeredi * unless they're stat'ing a directory and appended a '/' to 10980ec26fd0SMiklos Szeredi * the name. 10990ec26fd0SMiklos Szeredi * 11000ec26fd0SMiklos Szeredi * We do, however, want to mount if someone wants to open or 11015a30d8a2SDavid Howells * create a file of any type under the mountpoint, wants to 11020ec26fd0SMiklos Szeredi * traverse through the mountpoint or wants to open the 11030ec26fd0SMiklos Szeredi * mounted directory. Also, autofs may mark negative dentries 11040ec26fd0SMiklos Szeredi * as being automount points. These will need the attentions 11050ec26fd0SMiklos Szeredi * of the daemon to instantiate them before they can be used. 11065a30d8a2SDavid Howells */ 1107756daf26SNeilBrown if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | 1108d94c177bSLinus Torvalds LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && 11095a30d8a2SDavid Howells path->dentry->d_inode) 11109875cf80SDavid Howells return -EISDIR; 11110ec26fd0SMiklos Szeredi 1112756daf26SNeilBrown nd->total_link_count++; 1113756daf26SNeilBrown if (nd->total_link_count >= 40) 11149875cf80SDavid Howells return -ELOOP; 11159875cf80SDavid Howells 11169875cf80SDavid Howells mnt = path->dentry->d_op->d_automount(path); 11179875cf80SDavid Howells if (IS_ERR(mnt)) { 11189875cf80SDavid Howells /* 11199875cf80SDavid Howells * The filesystem is allowed to return -EISDIR here to indicate 11209875cf80SDavid Howells * it doesn't want to automount. For instance, autofs would do 11219875cf80SDavid Howells * this so that its userspace daemon can mount on this dentry. 11229875cf80SDavid Howells * 11239875cf80SDavid Howells * However, we can only permit this if it's a terminal point in 11249875cf80SDavid Howells * the path being looked up; if it wasn't then the remainder of 11259875cf80SDavid Howells * the path is inaccessible and we should say so. 11269875cf80SDavid Howells */ 1127756daf26SNeilBrown if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT)) 11289875cf80SDavid Howells return -EREMOTE; 11299875cf80SDavid Howells return PTR_ERR(mnt); 113031e6b01fSNick Piggin } 1131ea5b778aSDavid Howells 11329875cf80SDavid Howells if (!mnt) /* mount collision */ 11339875cf80SDavid Howells return 0; 11349875cf80SDavid Howells 11358aef1884SAl Viro if (!*need_mntput) { 11368aef1884SAl Viro /* lock_mount() may release path->mnt on error */ 11378aef1884SAl Viro mntget(path->mnt); 11388aef1884SAl Viro *need_mntput = true; 11398aef1884SAl Viro } 114019a167afSAl Viro err = finish_automount(mnt, path); 1141ea5b778aSDavid Howells 1142ea5b778aSDavid Howells switch (err) { 1143ea5b778aSDavid Howells case -EBUSY: 1144ea5b778aSDavid Howells /* Someone else made a mount here whilst we were busy */ 114519a167afSAl Viro return 0; 1146ea5b778aSDavid Howells case 0: 11478aef1884SAl Viro path_put(path); 11489875cf80SDavid Howells path->mnt = mnt; 11499875cf80SDavid Howells path->dentry = dget(mnt->mnt_root); 11509875cf80SDavid Howells return 0; 115119a167afSAl Viro default: 115219a167afSAl Viro return err; 11539875cf80SDavid Howells } 115419a167afSAl Viro 1155ea5b778aSDavid Howells } 11569875cf80SDavid Howells 11579875cf80SDavid Howells /* 11589875cf80SDavid Howells * Handle a dentry that is managed in some way. 1159cc53ce53SDavid Howells * - Flagged for transit management (autofs) 11609875cf80SDavid Howells * - Flagged as mountpoint 11619875cf80SDavid Howells * - Flagged as automount point 11629875cf80SDavid Howells * 11639875cf80SDavid Howells * This may only be called in refwalk mode. 11649875cf80SDavid Howells * 11659875cf80SDavid Howells * Serialization is taken care of in namespace.c 11669875cf80SDavid Howells */ 1167756daf26SNeilBrown static int follow_managed(struct path *path, struct nameidata *nd) 11689875cf80SDavid Howells { 11698aef1884SAl Viro struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */ 11709875cf80SDavid Howells unsigned managed; 11719875cf80SDavid Howells bool need_mntput = false; 11728aef1884SAl Viro int ret = 0; 11739875cf80SDavid Howells 11749875cf80SDavid Howells /* Given that we're not holding a lock here, we retain the value in a 11759875cf80SDavid Howells * local variable for each dentry as we look at it so that we don't see 11769875cf80SDavid Howells * the components of that value change under us */ 11779875cf80SDavid Howells while (managed = ACCESS_ONCE(path->dentry->d_flags), 11789875cf80SDavid Howells managed &= DCACHE_MANAGED_DENTRY, 11799875cf80SDavid Howells unlikely(managed != 0)) { 1180cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1181cc53ce53SDavid Howells * being held. */ 1182cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 1183cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1184cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 11851aed3e42SAl Viro ret = path->dentry->d_op->d_manage(path->dentry, false); 1186cc53ce53SDavid Howells if (ret < 0) 11878aef1884SAl Viro break; 1188cc53ce53SDavid Howells } 1189cc53ce53SDavid Howells 11909875cf80SDavid Howells /* Transit to a mounted filesystem. */ 11919875cf80SDavid Howells if (managed & DCACHE_MOUNTED) { 11929875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 11939875cf80SDavid Howells if (mounted) { 11949875cf80SDavid Howells dput(path->dentry); 11959875cf80SDavid Howells if (need_mntput) 1196463ffb2eSAl Viro mntput(path->mnt); 1197463ffb2eSAl Viro path->mnt = mounted; 1198463ffb2eSAl Viro path->dentry = dget(mounted->mnt_root); 11999875cf80SDavid Howells need_mntput = true; 12009875cf80SDavid Howells continue; 1201463ffb2eSAl Viro } 1202463ffb2eSAl Viro 12039875cf80SDavid Howells /* Something is mounted on this dentry in another 12049875cf80SDavid Howells * namespace and/or whatever was mounted there in this 120548a066e7SAl Viro * namespace got unmounted before lookup_mnt() could 120648a066e7SAl Viro * get it */ 12071da177e4SLinus Torvalds } 12089875cf80SDavid Howells 12099875cf80SDavid Howells /* Handle an automount point */ 12109875cf80SDavid Howells if (managed & DCACHE_NEED_AUTOMOUNT) { 1211756daf26SNeilBrown ret = follow_automount(path, nd, &need_mntput); 12129875cf80SDavid Howells if (ret < 0) 12138aef1884SAl Viro break; 12149875cf80SDavid Howells continue; 12159875cf80SDavid Howells } 12169875cf80SDavid Howells 12179875cf80SDavid Howells /* We didn't change the current path point */ 12189875cf80SDavid Howells break; 12199875cf80SDavid Howells } 12208aef1884SAl Viro 12218aef1884SAl Viro if (need_mntput && path->mnt == mnt) 12228aef1884SAl Viro mntput(path->mnt); 1223e9742b53SAl Viro if (ret == -EISDIR || !ret) 1224e9742b53SAl Viro ret = 1; 12258402752eSAl Viro if (need_mntput) 12268402752eSAl Viro nd->flags |= LOOKUP_JUMPED; 12278402752eSAl Viro if (unlikely(ret < 0)) 12288402752eSAl Viro path_put_conditional(path, nd); 12298402752eSAl Viro return ret; 12301da177e4SLinus Torvalds } 12311da177e4SLinus Torvalds 1232cc53ce53SDavid Howells int follow_down_one(struct path *path) 12331da177e4SLinus Torvalds { 12341da177e4SLinus Torvalds struct vfsmount *mounted; 12351da177e4SLinus Torvalds 12361c755af4SAl Viro mounted = lookup_mnt(path); 12371da177e4SLinus Torvalds if (mounted) { 12389393bd07SAl Viro dput(path->dentry); 12399393bd07SAl Viro mntput(path->mnt); 12409393bd07SAl Viro path->mnt = mounted; 12419393bd07SAl Viro path->dentry = dget(mounted->mnt_root); 12421da177e4SLinus Torvalds return 1; 12431da177e4SLinus Torvalds } 12441da177e4SLinus Torvalds return 0; 12451da177e4SLinus Torvalds } 12464d359507SAl Viro EXPORT_SYMBOL(follow_down_one); 12471da177e4SLinus Torvalds 1248b8faf035SNeilBrown static inline int managed_dentry_rcu(struct dentry *dentry) 124962a7375eSIan Kent { 1250b8faf035SNeilBrown return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ? 1251b8faf035SNeilBrown dentry->d_op->d_manage(dentry, true) : 0; 125262a7375eSIan Kent } 125362a7375eSIan Kent 12549875cf80SDavid Howells /* 1255287548e4SAl Viro * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if 1256287548e4SAl Viro * we meet a managed dentry that would need blocking. 12579875cf80SDavid Howells */ 12589875cf80SDavid Howells static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, 1259254cf582SAl Viro struct inode **inode, unsigned *seqp) 12609875cf80SDavid Howells { 126162a7375eSIan Kent for (;;) { 1262c7105365SAl Viro struct mount *mounted; 126362a7375eSIan Kent /* 126462a7375eSIan Kent * Don't forget we might have a non-mountpoint managed dentry 126562a7375eSIan Kent * that wants to block transit. 126662a7375eSIan Kent */ 1267b8faf035SNeilBrown switch (managed_dentry_rcu(path->dentry)) { 1268b8faf035SNeilBrown case -ECHILD: 1269b8faf035SNeilBrown default: 1270ab90911fSDavid Howells return false; 1271b8faf035SNeilBrown case -EISDIR: 1272b8faf035SNeilBrown return true; 1273b8faf035SNeilBrown case 0: 1274b8faf035SNeilBrown break; 1275b8faf035SNeilBrown } 127662a7375eSIan Kent 127762a7375eSIan Kent if (!d_mountpoint(path->dentry)) 1278b8faf035SNeilBrown return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); 127962a7375eSIan Kent 1280474279dcSAl Viro mounted = __lookup_mnt(path->mnt, path->dentry); 12819875cf80SDavid Howells if (!mounted) 12829875cf80SDavid Howells break; 1283c7105365SAl Viro path->mnt = &mounted->mnt; 1284c7105365SAl Viro path->dentry = mounted->mnt.mnt_root; 1285a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 1286254cf582SAl Viro *seqp = read_seqcount_begin(&path->dentry->d_seq); 128759430262SLinus Torvalds /* 128859430262SLinus Torvalds * Update the inode too. We don't need to re-check the 128959430262SLinus Torvalds * dentry sequence number here after this d_inode read, 129059430262SLinus Torvalds * because a mount-point is always pinned. 129159430262SLinus Torvalds */ 129259430262SLinus Torvalds *inode = path->dentry->d_inode; 12939875cf80SDavid Howells } 1294f5be3e29SAl Viro return !read_seqretry(&mount_lock, nd->m_seq) && 1295b8faf035SNeilBrown !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); 1296287548e4SAl Viro } 1297287548e4SAl Viro 129831e6b01fSNick Piggin static int follow_dotdot_rcu(struct nameidata *nd) 129931e6b01fSNick Piggin { 13004023bfc9SAl Viro struct inode *inode = nd->inode; 130131e6b01fSNick Piggin 130231e6b01fSNick Piggin while (1) { 1303aed434adSAl Viro if (path_equal(&nd->path, &nd->root)) 130431e6b01fSNick Piggin break; 130531e6b01fSNick Piggin if (nd->path.dentry != nd->path.mnt->mnt_root) { 130631e6b01fSNick Piggin struct dentry *old = nd->path.dentry; 130731e6b01fSNick Piggin struct dentry *parent = old->d_parent; 130831e6b01fSNick Piggin unsigned seq; 130931e6b01fSNick Piggin 13104023bfc9SAl Viro inode = parent->d_inode; 131131e6b01fSNick Piggin seq = read_seqcount_begin(&parent->d_seq); 1312aed434adSAl Viro if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq))) 1313aed434adSAl Viro return -ECHILD; 131431e6b01fSNick Piggin nd->path.dentry = parent; 131531e6b01fSNick Piggin nd->seq = seq; 1316397d425dSEric W. Biederman if (unlikely(!path_connected(&nd->path))) 1317397d425dSEric W. Biederman return -ENOENT; 131831e6b01fSNick Piggin break; 1319aed434adSAl Viro } else { 1320aed434adSAl Viro struct mount *mnt = real_mount(nd->path.mnt); 1321aed434adSAl Viro struct mount *mparent = mnt->mnt_parent; 1322aed434adSAl Viro struct dentry *mountpoint = mnt->mnt_mountpoint; 1323aed434adSAl Viro struct inode *inode2 = mountpoint->d_inode; 1324aed434adSAl Viro unsigned seq = read_seqcount_begin(&mountpoint->d_seq); 1325aed434adSAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1326aed434adSAl Viro return -ECHILD; 1327aed434adSAl Viro if (&mparent->mnt == nd->path.mnt) 132831e6b01fSNick Piggin break; 1329aed434adSAl Viro /* we know that mountpoint was pinned */ 1330aed434adSAl Viro nd->path.dentry = mountpoint; 1331aed434adSAl Viro nd->path.mnt = &mparent->mnt; 1332aed434adSAl Viro inode = inode2; 1333aed434adSAl Viro nd->seq = seq; 133431e6b01fSNick Piggin } 1335aed434adSAl Viro } 1336aed434adSAl Viro while (unlikely(d_mountpoint(nd->path.dentry))) { 1337b37199e6SAl Viro struct mount *mounted; 1338b37199e6SAl Viro mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry); 1339aed434adSAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1340aed434adSAl Viro return -ECHILD; 1341b37199e6SAl Viro if (!mounted) 1342b37199e6SAl Viro break; 1343b37199e6SAl Viro nd->path.mnt = &mounted->mnt; 1344b37199e6SAl Viro nd->path.dentry = mounted->mnt.mnt_root; 13454023bfc9SAl Viro inode = nd->path.dentry->d_inode; 1346b37199e6SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 1347b37199e6SAl Viro } 13484023bfc9SAl Viro nd->inode = inode; 134931e6b01fSNick Piggin return 0; 135031e6b01fSNick Piggin } 135131e6b01fSNick Piggin 13529875cf80SDavid Howells /* 1353cc53ce53SDavid Howells * Follow down to the covering mount currently visible to userspace. At each 1354cc53ce53SDavid Howells * point, the filesystem owning that dentry may be queried as to whether the 1355cc53ce53SDavid Howells * caller is permitted to proceed or not. 1356cc53ce53SDavid Howells */ 13577cc90cc3SAl Viro int follow_down(struct path *path) 1358cc53ce53SDavid Howells { 1359cc53ce53SDavid Howells unsigned managed; 1360cc53ce53SDavid Howells int ret; 1361cc53ce53SDavid Howells 1362cc53ce53SDavid Howells while (managed = ACCESS_ONCE(path->dentry->d_flags), 1363cc53ce53SDavid Howells unlikely(managed & DCACHE_MANAGED_DENTRY)) { 1364cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1365cc53ce53SDavid Howells * being held. 1366cc53ce53SDavid Howells * 1367cc53ce53SDavid Howells * We indicate to the filesystem if someone is trying to mount 1368cc53ce53SDavid Howells * something here. This gives autofs the chance to deny anyone 1369cc53ce53SDavid Howells * other than its daemon the right to mount on its 1370cc53ce53SDavid Howells * superstructure. 1371cc53ce53SDavid Howells * 1372cc53ce53SDavid Howells * The filesystem may sleep at this point. 1373cc53ce53SDavid Howells */ 1374cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 1375cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1376cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 1377ab90911fSDavid Howells ret = path->dentry->d_op->d_manage( 13781aed3e42SAl Viro path->dentry, false); 1379cc53ce53SDavid Howells if (ret < 0) 1380cc53ce53SDavid Howells return ret == -EISDIR ? 0 : ret; 1381cc53ce53SDavid Howells } 1382cc53ce53SDavid Howells 1383cc53ce53SDavid Howells /* Transit to a mounted filesystem. */ 1384cc53ce53SDavid Howells if (managed & DCACHE_MOUNTED) { 1385cc53ce53SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 1386cc53ce53SDavid Howells if (!mounted) 1387cc53ce53SDavid Howells break; 1388cc53ce53SDavid Howells dput(path->dentry); 1389cc53ce53SDavid Howells mntput(path->mnt); 1390cc53ce53SDavid Howells path->mnt = mounted; 1391cc53ce53SDavid Howells path->dentry = dget(mounted->mnt_root); 1392cc53ce53SDavid Howells continue; 1393cc53ce53SDavid Howells } 1394cc53ce53SDavid Howells 1395cc53ce53SDavid Howells /* Don't handle automount points here */ 1396cc53ce53SDavid Howells break; 1397cc53ce53SDavid Howells } 1398cc53ce53SDavid Howells return 0; 1399cc53ce53SDavid Howells } 14004d359507SAl Viro EXPORT_SYMBOL(follow_down); 1401cc53ce53SDavid Howells 1402cc53ce53SDavid Howells /* 14039875cf80SDavid Howells * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() 14049875cf80SDavid Howells */ 14059875cf80SDavid Howells static void follow_mount(struct path *path) 14069875cf80SDavid Howells { 14079875cf80SDavid Howells while (d_mountpoint(path->dentry)) { 14089875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 14099875cf80SDavid Howells if (!mounted) 14109875cf80SDavid Howells break; 14119875cf80SDavid Howells dput(path->dentry); 14129875cf80SDavid Howells mntput(path->mnt); 14139875cf80SDavid Howells path->mnt = mounted; 14149875cf80SDavid Howells path->dentry = dget(mounted->mnt_root); 14159875cf80SDavid Howells } 14169875cf80SDavid Howells } 14179875cf80SDavid Howells 1418397d425dSEric W. Biederman static int follow_dotdot(struct nameidata *nd) 14191da177e4SLinus Torvalds { 14201da177e4SLinus Torvalds while(1) { 14214ac91378SJan Blunck struct dentry *old = nd->path.dentry; 14221da177e4SLinus Torvalds 14232a737871SAl Viro if (nd->path.dentry == nd->root.dentry && 14242a737871SAl Viro nd->path.mnt == nd->root.mnt) { 14251da177e4SLinus Torvalds break; 14261da177e4SLinus Torvalds } 14274ac91378SJan Blunck if (nd->path.dentry != nd->path.mnt->mnt_root) { 14283088dd70SAl Viro /* rare case of legitimate dget_parent()... */ 14293088dd70SAl Viro nd->path.dentry = dget_parent(nd->path.dentry); 14301da177e4SLinus Torvalds dput(old); 1431397d425dSEric W. Biederman if (unlikely(!path_connected(&nd->path))) 1432397d425dSEric W. Biederman return -ENOENT; 14331da177e4SLinus Torvalds break; 14341da177e4SLinus Torvalds } 14353088dd70SAl Viro if (!follow_up(&nd->path)) 14361da177e4SLinus Torvalds break; 14371da177e4SLinus Torvalds } 143879ed0226SAl Viro follow_mount(&nd->path); 143931e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 1440397d425dSEric W. Biederman return 0; 14411da177e4SLinus Torvalds } 14421da177e4SLinus Torvalds 14431da177e4SLinus Torvalds /* 1444bad61189SMiklos Szeredi * This looks up the name in dcache, possibly revalidates the old dentry and 1445bad61189SMiklos Szeredi * allocates a new one if not found or not valid. In the need_lookup argument 1446bad61189SMiklos Szeredi * returns whether i_op->lookup is necessary. 1447bad61189SMiklos Szeredi * 1448bad61189SMiklos Szeredi * dir->d_inode->i_mutex must be held 1449baa03890SNick Piggin */ 1450bad61189SMiklos Szeredi static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir, 14516c51e513SAl Viro unsigned int flags) 1452baa03890SNick Piggin { 1453baa03890SNick Piggin struct dentry *dentry; 1454bad61189SMiklos Szeredi int error; 1455baa03890SNick Piggin 1456bad61189SMiklos Szeredi dentry = d_lookup(dir, name); 1457bad61189SMiklos Szeredi if (dentry) { 145839e3c955SJeff Layton if (dentry->d_flags & DCACHE_OP_REVALIDATE) { 1459201f956eSAl Viro error = d_revalidate(dentry, flags); 1460bad61189SMiklos Szeredi if (unlikely(error <= 0)) { 1461bad61189SMiklos Szeredi if (error < 0) { 1462bad61189SMiklos Szeredi dput(dentry); 1463bad61189SMiklos Szeredi return ERR_PTR(error); 14645542aa2fSEric W. Biederman } else { 14655542aa2fSEric W. Biederman d_invalidate(dentry); 1466bad61189SMiklos Szeredi dput(dentry); 1467bad61189SMiklos Szeredi dentry = NULL; 1468bad61189SMiklos Szeredi } 1469bad61189SMiklos Szeredi } 1470bad61189SMiklos Szeredi } 1471bad61189SMiklos Szeredi } 1472baa03890SNick Piggin return dentry; 1473baa03890SNick Piggin } 1474baa03890SNick Piggin 1475baa03890SNick Piggin /* 147613a2c3beSJ. Bruce Fields * Call i_op->lookup on the dentry. The dentry must be negative and 147713a2c3beSJ. Bruce Fields * unhashed. 1478bad61189SMiklos Szeredi * 1479bad61189SMiklos Szeredi * dir->d_inode->i_mutex must be held 148044396f4bSJosef Bacik */ 1481bad61189SMiklos Szeredi static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry, 148272bd866aSAl Viro unsigned int flags) 148344396f4bSJosef Bacik { 148444396f4bSJosef Bacik struct dentry *old; 148544396f4bSJosef Bacik 148644396f4bSJosef Bacik /* Don't create child dentry for a dead directory. */ 1487bad61189SMiklos Szeredi if (unlikely(IS_DEADDIR(dir))) { 1488e188dc02SMiklos Szeredi dput(dentry); 148944396f4bSJosef Bacik return ERR_PTR(-ENOENT); 1490e188dc02SMiklos Szeredi } 149144396f4bSJosef Bacik 149272bd866aSAl Viro old = dir->i_op->lookup(dir, dentry, flags); 149344396f4bSJosef Bacik if (unlikely(old)) { 149444396f4bSJosef Bacik dput(dentry); 149544396f4bSJosef Bacik dentry = old; 149644396f4bSJosef Bacik } 149744396f4bSJosef Bacik return dentry; 149844396f4bSJosef Bacik } 149944396f4bSJosef Bacik 1500a3255546SAl Viro static struct dentry *__lookup_hash(struct qstr *name, 150172bd866aSAl Viro struct dentry *base, unsigned int flags) 1502a3255546SAl Viro { 15036c51e513SAl Viro struct dentry *dentry = lookup_dcache(name, base, flags); 1504a3255546SAl Viro 15056c51e513SAl Viro if (dentry) 1506a3255546SAl Viro return dentry; 1507bad61189SMiklos Szeredi 15086c51e513SAl Viro dentry = d_alloc(base, name); 15096c51e513SAl Viro if (unlikely(!dentry)) 15106c51e513SAl Viro return ERR_PTR(-ENOMEM); 15116c51e513SAl Viro 151272bd866aSAl Viro return lookup_real(base->d_inode, dentry, flags); 1513a3255546SAl Viro } 1514a3255546SAl Viro 1515e97cdc87SAl Viro static int lookup_fast(struct nameidata *nd, 1516254cf582SAl Viro struct path *path, struct inode **inode, 1517254cf582SAl Viro unsigned *seqp) 15181da177e4SLinus Torvalds { 15194ac91378SJan Blunck struct vfsmount *mnt = nd->path.mnt; 152031e6b01fSNick Piggin struct dentry *dentry, *parent = nd->path.dentry; 15215a18fff2SAl Viro int status = 1; 15229875cf80SDavid Howells int err; 15239875cf80SDavid Howells 15243cac260aSAl Viro /* 1525b04f784eSNick Piggin * Rename seqlock is not required here because in the off chance 15265d0f49c1SAl Viro * of a false negative due to a concurrent rename, the caller is 15275d0f49c1SAl Viro * going to fall back to non-racy lookup. 1528b04f784eSNick Piggin */ 152931e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) { 153031e6b01fSNick Piggin unsigned seq; 1531766c4cbfSAl Viro bool negative; 1532da53be12SLinus Torvalds dentry = __d_lookup_rcu(parent, &nd->last, &seq); 15335d0f49c1SAl Viro if (unlikely(!dentry)) { 15345d0f49c1SAl Viro if (unlazy_walk(nd, NULL, 0)) 15355d0f49c1SAl Viro return -ECHILD; 1536e9742b53SAl Viro return 0; 15375d0f49c1SAl Viro } 15385a18fff2SAl Viro 153912f8ad4bSLinus Torvalds /* 154012f8ad4bSLinus Torvalds * This sequence count validates that the inode matches 154112f8ad4bSLinus Torvalds * the dentry name information from lookup. 154212f8ad4bSLinus Torvalds */ 154363afdfc7SDavid Howells *inode = d_backing_inode(dentry); 1544766c4cbfSAl Viro negative = d_is_negative(dentry); 15455d0f49c1SAl Viro if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) 154612f8ad4bSLinus Torvalds return -ECHILD; 154712f8ad4bSLinus Torvalds 154812f8ad4bSLinus Torvalds /* 154912f8ad4bSLinus Torvalds * This sequence count validates that the parent had no 155012f8ad4bSLinus Torvalds * changes while we did the lookup of the dentry above. 155112f8ad4bSLinus Torvalds * 155212f8ad4bSLinus Torvalds * The memory barrier in read_seqcount_begin of child is 155312f8ad4bSLinus Torvalds * enough, we can use __read_seqcount_retry here. 155412f8ad4bSLinus Torvalds */ 15555d0f49c1SAl Viro if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq))) 155631e6b01fSNick Piggin return -ECHILD; 15575a18fff2SAl Viro 1558254cf582SAl Viro *seqp = seq; 15595d0f49c1SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) 15604ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 15615a18fff2SAl Viro if (unlikely(status <= 0)) { 15625d0f49c1SAl Viro if (unlazy_walk(nd, dentry, seq)) 15635d0f49c1SAl Viro return -ECHILD; 15645d0f49c1SAl Viro if (status == -ECHILD) 15655d0f49c1SAl Viro status = d_revalidate(dentry, nd->flags); 15665d0f49c1SAl Viro } else { 1567daf3761cSTrond Myklebust /* 1568daf3761cSTrond Myklebust * Note: do negative dentry check after revalidation in 1569daf3761cSTrond Myklebust * case that drops it. 1570daf3761cSTrond Myklebust */ 15715d0f49c1SAl Viro if (unlikely(negative)) 1572daf3761cSTrond Myklebust return -ENOENT; 157331e6b01fSNick Piggin path->mnt = mnt; 157431e6b01fSNick Piggin path->dentry = dentry; 1575254cf582SAl Viro if (likely(__follow_mount_rcu(nd, path, inode, seqp))) 1576e9742b53SAl Viro return 1; 1577254cf582SAl Viro if (unlazy_walk(nd, dentry, seq)) 15785a18fff2SAl Viro return -ECHILD; 15795d0f49c1SAl Viro } 15805a18fff2SAl Viro } else { 1581e97cdc87SAl Viro dentry = __d_lookup(parent, &nd->last); 158281e6f520SAl Viro if (unlikely(!dentry)) 1583e9742b53SAl Viro return 0; 15845d0f49c1SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) 15854ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 15865d0f49c1SAl Viro } 15875a18fff2SAl Viro if (unlikely(status <= 0)) { 1588e9742b53SAl Viro if (!status) 15895d0f49c1SAl Viro d_invalidate(dentry); 15905a18fff2SAl Viro dput(dentry); 15915a18fff2SAl Viro return status; 15925a18fff2SAl Viro } 1593766c4cbfSAl Viro if (unlikely(d_is_negative(dentry))) { 1594766c4cbfSAl Viro dput(dentry); 1595766c4cbfSAl Viro return -ENOENT; 1596766c4cbfSAl Viro } 15975d0f49c1SAl Viro 15981da177e4SLinus Torvalds path->mnt = mnt; 15991da177e4SLinus Torvalds path->dentry = dentry; 1600756daf26SNeilBrown err = follow_managed(path, nd); 1601e9742b53SAl Viro if (likely(err > 0)) 160263afdfc7SDavid Howells *inode = d_backing_inode(path->dentry); 16038402752eSAl Viro return err; 1604697f514dSMiklos Szeredi } 1605697f514dSMiklos Szeredi 1606697f514dSMiklos Szeredi /* Fast lookup failed, do it the slow way */ 1607cc2a5271SAl Viro static int lookup_slow(struct nameidata *nd, struct path *path) 1608697f514dSMiklos Szeredi { 1609697f514dSMiklos Szeredi struct dentry *dentry, *parent; 1610697f514dSMiklos Szeredi 1611697f514dSMiklos Szeredi parent = nd->path.dentry; 161281e6f520SAl Viro BUG_ON(nd->inode != parent->d_inode); 161381e6f520SAl Viro 16145955102cSAl Viro inode_lock(parent->d_inode); 1615cc2a5271SAl Viro dentry = __lookup_hash(&nd->last, parent, nd->flags); 16165955102cSAl Viro inode_unlock(parent->d_inode); 161781e6f520SAl Viro if (IS_ERR(dentry)) 161881e6f520SAl Viro return PTR_ERR(dentry); 1619697f514dSMiklos Szeredi path->mnt = nd->path.mnt; 1620697f514dSMiklos Szeredi path->dentry = dentry; 16218402752eSAl Viro return follow_managed(path, nd); 16221da177e4SLinus Torvalds } 16231da177e4SLinus Torvalds 162452094c8aSAl Viro static inline int may_lookup(struct nameidata *nd) 162552094c8aSAl Viro { 162652094c8aSAl Viro if (nd->flags & LOOKUP_RCU) { 16274ad5abb3SAl Viro int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); 162852094c8aSAl Viro if (err != -ECHILD) 162952094c8aSAl Viro return err; 16306e9918b7SAl Viro if (unlazy_walk(nd, NULL, 0)) 163152094c8aSAl Viro return -ECHILD; 163252094c8aSAl Viro } 16334ad5abb3SAl Viro return inode_permission(nd->inode, MAY_EXEC); 163452094c8aSAl Viro } 163552094c8aSAl Viro 16369856fa1bSAl Viro static inline int handle_dots(struct nameidata *nd, int type) 16379856fa1bSAl Viro { 16389856fa1bSAl Viro if (type == LAST_DOTDOT) { 16399e6697e2SAl Viro if (!nd->root.mnt) 16409e6697e2SAl Viro set_root(nd); 16419856fa1bSAl Viro if (nd->flags & LOOKUP_RCU) { 164270291aecSAl Viro return follow_dotdot_rcu(nd); 16439856fa1bSAl Viro } else 1644397d425dSEric W. Biederman return follow_dotdot(nd); 16459856fa1bSAl Viro } 16469856fa1bSAl Viro return 0; 16479856fa1bSAl Viro } 16489856fa1bSAl Viro 1649181548c0SAl Viro static int pick_link(struct nameidata *nd, struct path *link, 1650181548c0SAl Viro struct inode *inode, unsigned seq) 1651d63ff28fSAl Viro { 1652626de996SAl Viro int error; 16531cf2665bSAl Viro struct saved *last; 1654756daf26SNeilBrown if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) { 1655626de996SAl Viro path_to_nameidata(link, nd); 1656626de996SAl Viro return -ELOOP; 1657626de996SAl Viro } 1658bc40aee0SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1659cd179f44SAl Viro if (link->mnt == nd->path.mnt) 1660cd179f44SAl Viro mntget(link->mnt); 16617973387aSAl Viro } 1662626de996SAl Viro error = nd_alloc_stack(nd); 1663626de996SAl Viro if (unlikely(error)) { 1664bc40aee0SAl Viro if (error == -ECHILD) { 1665bc40aee0SAl Viro if (unlikely(unlazy_link(nd, link, seq))) 1666bc40aee0SAl Viro return -ECHILD; 1667bc40aee0SAl Viro error = nd_alloc_stack(nd); 1668bc40aee0SAl Viro } 1669bc40aee0SAl Viro if (error) { 1670cd179f44SAl Viro path_put(link); 1671626de996SAl Viro return error; 1672626de996SAl Viro } 1673bc40aee0SAl Viro } 1674626de996SAl Viro 1675ab104923SAl Viro last = nd->stack + nd->depth++; 16761cf2665bSAl Viro last->link = *link; 1677fceef393SAl Viro clear_delayed_call(&last->done); 1678fceef393SAl Viro nd->link_inode = inode; 16790450b2d1SAl Viro last->seq = seq; 1680d63ff28fSAl Viro return 1; 1681d63ff28fSAl Viro } 1682d63ff28fSAl Viro 16833ddcd056SLinus Torvalds /* 16843ddcd056SLinus Torvalds * Do we need to follow links? We _really_ want to be able 16853ddcd056SLinus Torvalds * to do this check without having to look at inode->i_op, 16863ddcd056SLinus Torvalds * so we keep a cache of "no, this doesn't need follow_link" 16873ddcd056SLinus Torvalds * for the common case. 16883ddcd056SLinus Torvalds */ 1689254cf582SAl Viro static inline int should_follow_link(struct nameidata *nd, struct path *link, 1690181548c0SAl Viro int follow, 1691181548c0SAl Viro struct inode *inode, unsigned seq) 16923ddcd056SLinus Torvalds { 1693d63ff28fSAl Viro if (likely(!d_is_symlink(link->dentry))) 1694d63ff28fSAl Viro return 0; 1695d63ff28fSAl Viro if (!follow) 1696d63ff28fSAl Viro return 0; 1697a7f77542SAl Viro /* make sure that d_is_symlink above matches inode */ 1698a7f77542SAl Viro if (nd->flags & LOOKUP_RCU) { 1699a7f77542SAl Viro if (read_seqcount_retry(&link->dentry->d_seq, seq)) 1700a7f77542SAl Viro return -ECHILD; 1701a7f77542SAl Viro } 1702181548c0SAl Viro return pick_link(nd, link, inode, seq); 17033ddcd056SLinus Torvalds } 17043ddcd056SLinus Torvalds 17054693a547SAl Viro enum {WALK_GET = 1, WALK_PUT = 2}; 17064693a547SAl Viro 17074693a547SAl Viro static int walk_component(struct nameidata *nd, int flags) 1708ce57dfc1SAl Viro { 1709caa85634SAl Viro struct path path; 1710ce57dfc1SAl Viro struct inode *inode; 1711254cf582SAl Viro unsigned seq; 1712ce57dfc1SAl Viro int err; 1713ce57dfc1SAl Viro /* 1714ce57dfc1SAl Viro * "." and ".." are special - ".." especially so because it has 1715ce57dfc1SAl Viro * to be able to know about the current root directory and 1716ce57dfc1SAl Viro * parent relationships. 1717ce57dfc1SAl Viro */ 17184693a547SAl Viro if (unlikely(nd->last_type != LAST_NORM)) { 17194693a547SAl Viro err = handle_dots(nd, nd->last_type); 17204693a547SAl Viro if (flags & WALK_PUT) 17214693a547SAl Viro put_link(nd); 17224693a547SAl Viro return err; 17234693a547SAl Viro } 1724254cf582SAl Viro err = lookup_fast(nd, &path, &inode, &seq); 1725e9742b53SAl Viro if (unlikely(err <= 0)) { 1726697f514dSMiklos Szeredi if (err < 0) 1727f0a9ba70SAl Viro return err; 1728697f514dSMiklos Szeredi 1729caa85634SAl Viro err = lookup_slow(nd, &path); 1730697f514dSMiklos Szeredi if (err < 0) 1731f0a9ba70SAl Viro return err; 1732697f514dSMiklos Szeredi 1733254cf582SAl Viro seq = 0; /* we are already out of RCU mode */ 1734697f514dSMiklos Szeredi err = -ENOENT; 1735caa85634SAl Viro if (d_is_negative(path.dentry)) 1736697f514dSMiklos Szeredi goto out_path_put; 1737d4565649SAl Viro inode = d_backing_inode(path.dentry); 1738766c4cbfSAl Viro } 1739697f514dSMiklos Szeredi 17404693a547SAl Viro if (flags & WALK_PUT) 17414693a547SAl Viro put_link(nd); 1742181548c0SAl Viro err = should_follow_link(nd, &path, flags & WALK_GET, inode, seq); 1743d63ff28fSAl Viro if (unlikely(err)) 1744d63ff28fSAl Viro return err; 1745caa85634SAl Viro path_to_nameidata(&path, nd); 1746ce57dfc1SAl Viro nd->inode = inode; 1747254cf582SAl Viro nd->seq = seq; 1748ce57dfc1SAl Viro return 0; 1749697f514dSMiklos Szeredi 1750697f514dSMiklos Szeredi out_path_put: 1751caa85634SAl Viro path_to_nameidata(&path, nd); 1752697f514dSMiklos Szeredi return err; 1753ce57dfc1SAl Viro } 1754ce57dfc1SAl Viro 17551da177e4SLinus Torvalds /* 1756bfcfaa77SLinus Torvalds * We can do the critical dentry name comparison and hashing 1757bfcfaa77SLinus Torvalds * operations one word at a time, but we are limited to: 1758bfcfaa77SLinus Torvalds * 1759bfcfaa77SLinus Torvalds * - Architectures with fast unaligned word accesses. We could 1760bfcfaa77SLinus Torvalds * do a "get_unaligned()" if this helps and is sufficiently 1761bfcfaa77SLinus Torvalds * fast. 1762bfcfaa77SLinus Torvalds * 1763bfcfaa77SLinus Torvalds * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we 1764bfcfaa77SLinus Torvalds * do not trap on the (extremely unlikely) case of a page 1765bfcfaa77SLinus Torvalds * crossing operation. 1766bfcfaa77SLinus Torvalds * 1767bfcfaa77SLinus Torvalds * - Furthermore, we need an efficient 64-bit compile for the 1768bfcfaa77SLinus Torvalds * 64-bit case in order to generate the "number of bytes in 1769bfcfaa77SLinus Torvalds * the final mask". Again, that could be replaced with a 1770bfcfaa77SLinus Torvalds * efficient population count instruction or similar. 1771bfcfaa77SLinus Torvalds */ 1772bfcfaa77SLinus Torvalds #ifdef CONFIG_DCACHE_WORD_ACCESS 1773bfcfaa77SLinus Torvalds 1774f68e556eSLinus Torvalds #include <asm/word-at-a-time.h> 1775bfcfaa77SLinus Torvalds 1776f68e556eSLinus Torvalds #ifdef CONFIG_64BIT 1777bfcfaa77SLinus Torvalds 1778bfcfaa77SLinus Torvalds static inline unsigned int fold_hash(unsigned long hash) 1779bfcfaa77SLinus Torvalds { 178099d263d4SLinus Torvalds return hash_64(hash, 32); 1781bfcfaa77SLinus Torvalds } 1782bfcfaa77SLinus Torvalds 1783bfcfaa77SLinus Torvalds #else /* 32-bit case */ 1784bfcfaa77SLinus Torvalds 1785bfcfaa77SLinus Torvalds #define fold_hash(x) (x) 1786bfcfaa77SLinus Torvalds 1787bfcfaa77SLinus Torvalds #endif 1788bfcfaa77SLinus Torvalds 1789bfcfaa77SLinus Torvalds unsigned int full_name_hash(const unsigned char *name, unsigned int len) 1790bfcfaa77SLinus Torvalds { 1791bfcfaa77SLinus Torvalds unsigned long a, mask; 1792bfcfaa77SLinus Torvalds unsigned long hash = 0; 1793bfcfaa77SLinus Torvalds 1794bfcfaa77SLinus Torvalds for (;;) { 1795e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name); 1796bfcfaa77SLinus Torvalds if (len < sizeof(unsigned long)) 1797bfcfaa77SLinus Torvalds break; 1798bfcfaa77SLinus Torvalds hash += a; 1799f132c5beSAl Viro hash *= 9; 1800bfcfaa77SLinus Torvalds name += sizeof(unsigned long); 1801bfcfaa77SLinus Torvalds len -= sizeof(unsigned long); 1802bfcfaa77SLinus Torvalds if (!len) 1803bfcfaa77SLinus Torvalds goto done; 1804bfcfaa77SLinus Torvalds } 1805a5c21dceSWill Deacon mask = bytemask_from_count(len); 1806bfcfaa77SLinus Torvalds hash += mask & a; 1807bfcfaa77SLinus Torvalds done: 1808bfcfaa77SLinus Torvalds return fold_hash(hash); 1809bfcfaa77SLinus Torvalds } 1810bfcfaa77SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 1811bfcfaa77SLinus Torvalds 1812bfcfaa77SLinus Torvalds /* 1813bfcfaa77SLinus Torvalds * Calculate the length and hash of the path component, and 1814d6bb3e90SLinus Torvalds * return the "hash_len" as the result. 1815bfcfaa77SLinus Torvalds */ 1816d6bb3e90SLinus Torvalds static inline u64 hash_name(const char *name) 1817bfcfaa77SLinus Torvalds { 181836126f8fSLinus Torvalds unsigned long a, b, adata, bdata, mask, hash, len; 181936126f8fSLinus Torvalds const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 1820bfcfaa77SLinus Torvalds 1821bfcfaa77SLinus Torvalds hash = a = 0; 1822bfcfaa77SLinus Torvalds len = -sizeof(unsigned long); 1823bfcfaa77SLinus Torvalds do { 1824bfcfaa77SLinus Torvalds hash = (hash + a) * 9; 1825bfcfaa77SLinus Torvalds len += sizeof(unsigned long); 1826e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name+len); 182736126f8fSLinus Torvalds b = a ^ REPEAT_BYTE('/'); 182836126f8fSLinus Torvalds } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); 1829bfcfaa77SLinus Torvalds 183036126f8fSLinus Torvalds adata = prep_zero_mask(a, adata, &constants); 183136126f8fSLinus Torvalds bdata = prep_zero_mask(b, bdata, &constants); 183236126f8fSLinus Torvalds 183336126f8fSLinus Torvalds mask = create_zero_mask(adata | bdata); 183436126f8fSLinus Torvalds 183536126f8fSLinus Torvalds hash += a & zero_bytemask(mask); 18369226b5b4SLinus Torvalds len += find_zero(mask); 1837d6bb3e90SLinus Torvalds return hashlen_create(fold_hash(hash), len); 1838bfcfaa77SLinus Torvalds } 1839bfcfaa77SLinus Torvalds 1840bfcfaa77SLinus Torvalds #else 1841bfcfaa77SLinus Torvalds 18420145acc2SLinus Torvalds unsigned int full_name_hash(const unsigned char *name, unsigned int len) 18430145acc2SLinus Torvalds { 18440145acc2SLinus Torvalds unsigned long hash = init_name_hash(); 18450145acc2SLinus Torvalds while (len--) 18460145acc2SLinus Torvalds hash = partial_name_hash(*name++, hash); 18470145acc2SLinus Torvalds return end_name_hash(hash); 18480145acc2SLinus Torvalds } 1849ae942ae7SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 18500145acc2SLinus Torvalds 18513ddcd056SLinus Torvalds /* 1852200e9ef7SLinus Torvalds * We know there's a real path component here of at least 1853200e9ef7SLinus Torvalds * one character. 1854200e9ef7SLinus Torvalds */ 1855d6bb3e90SLinus Torvalds static inline u64 hash_name(const char *name) 1856200e9ef7SLinus Torvalds { 1857200e9ef7SLinus Torvalds unsigned long hash = init_name_hash(); 1858200e9ef7SLinus Torvalds unsigned long len = 0, c; 1859200e9ef7SLinus Torvalds 1860200e9ef7SLinus Torvalds c = (unsigned char)*name; 1861200e9ef7SLinus Torvalds do { 1862200e9ef7SLinus Torvalds len++; 1863200e9ef7SLinus Torvalds hash = partial_name_hash(c, hash); 1864200e9ef7SLinus Torvalds c = (unsigned char)name[len]; 1865200e9ef7SLinus Torvalds } while (c && c != '/'); 1866d6bb3e90SLinus Torvalds return hashlen_create(end_name_hash(hash), len); 1867200e9ef7SLinus Torvalds } 1868200e9ef7SLinus Torvalds 1869bfcfaa77SLinus Torvalds #endif 1870bfcfaa77SLinus Torvalds 1871200e9ef7SLinus Torvalds /* 18721da177e4SLinus Torvalds * Name resolution. 1873ea3834d9SPrasanna Meda * This is the basic name resolution function, turning a pathname into 1874ea3834d9SPrasanna Meda * the final dentry. We expect 'base' to be positive and a directory. 18751da177e4SLinus Torvalds * 1876ea3834d9SPrasanna Meda * Returns 0 and nd will have valid dentry and mnt on success. 1877ea3834d9SPrasanna Meda * Returns error and drops reference to input namei data on failure. 18781da177e4SLinus Torvalds */ 18796de88d72SAl Viro static int link_path_walk(const char *name, struct nameidata *nd) 18801da177e4SLinus Torvalds { 18811da177e4SLinus Torvalds int err; 18821da177e4SLinus Torvalds 18831da177e4SLinus Torvalds while (*name=='/') 18841da177e4SLinus Torvalds name++; 18851da177e4SLinus Torvalds if (!*name) 18869e18f10aSAl Viro return 0; 18871da177e4SLinus Torvalds 18881da177e4SLinus Torvalds /* At this point we know we have a real path component. */ 18891da177e4SLinus Torvalds for(;;) { 1890d6bb3e90SLinus Torvalds u64 hash_len; 1891fe479a58SAl Viro int type; 18921da177e4SLinus Torvalds 189352094c8aSAl Viro err = may_lookup(nd); 18941da177e4SLinus Torvalds if (err) 18953595e234SAl Viro return err; 18961da177e4SLinus Torvalds 1897d6bb3e90SLinus Torvalds hash_len = hash_name(name); 18981da177e4SLinus Torvalds 1899fe479a58SAl Viro type = LAST_NORM; 1900d6bb3e90SLinus Torvalds if (name[0] == '.') switch (hashlen_len(hash_len)) { 1901fe479a58SAl Viro case 2: 1902200e9ef7SLinus Torvalds if (name[1] == '.') { 1903fe479a58SAl Viro type = LAST_DOTDOT; 190416c2cd71SAl Viro nd->flags |= LOOKUP_JUMPED; 190516c2cd71SAl Viro } 1906fe479a58SAl Viro break; 1907fe479a58SAl Viro case 1: 1908fe479a58SAl Viro type = LAST_DOT; 1909fe479a58SAl Viro } 19105a202bcdSAl Viro if (likely(type == LAST_NORM)) { 19115a202bcdSAl Viro struct dentry *parent = nd->path.dentry; 191216c2cd71SAl Viro nd->flags &= ~LOOKUP_JUMPED; 19135a202bcdSAl Viro if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { 1914a060dc50SJames Hogan struct qstr this = { { .hash_len = hash_len }, .name = name }; 1915da53be12SLinus Torvalds err = parent->d_op->d_hash(parent, &this); 19165a202bcdSAl Viro if (err < 0) 19173595e234SAl Viro return err; 1918d6bb3e90SLinus Torvalds hash_len = this.hash_len; 1919d6bb3e90SLinus Torvalds name = this.name; 19205a202bcdSAl Viro } 19215a202bcdSAl Viro } 1922fe479a58SAl Viro 1923d6bb3e90SLinus Torvalds nd->last.hash_len = hash_len; 1924d6bb3e90SLinus Torvalds nd->last.name = name; 19255f4a6a69SAl Viro nd->last_type = type; 19265f4a6a69SAl Viro 1927d6bb3e90SLinus Torvalds name += hashlen_len(hash_len); 1928d6bb3e90SLinus Torvalds if (!*name) 1929bdf6cbf1SAl Viro goto OK; 1930200e9ef7SLinus Torvalds /* 1931200e9ef7SLinus Torvalds * If it wasn't NUL, we know it was '/'. Skip that 1932200e9ef7SLinus Torvalds * slash, and continue until no more slashes. 1933200e9ef7SLinus Torvalds */ 1934200e9ef7SLinus Torvalds do { 1935d6bb3e90SLinus Torvalds name++; 1936d6bb3e90SLinus Torvalds } while (unlikely(*name == '/')); 19378620c238SAl Viro if (unlikely(!*name)) { 19388620c238SAl Viro OK: 1939368ee9baSAl Viro /* pathname body, done */ 19408620c238SAl Viro if (!nd->depth) 19418620c238SAl Viro return 0; 19428620c238SAl Viro name = nd->stack[nd->depth - 1].name; 1943368ee9baSAl Viro /* trailing symlink, done */ 19448620c238SAl Viro if (!name) 19458620c238SAl Viro return 0; 19468620c238SAl Viro /* last component of nested symlink */ 19474693a547SAl Viro err = walk_component(nd, WALK_GET | WALK_PUT); 19488620c238SAl Viro } else { 19494693a547SAl Viro err = walk_component(nd, WALK_GET); 19508620c238SAl Viro } 1951ce57dfc1SAl Viro if (err < 0) 19523595e234SAl Viro return err; 1953fe479a58SAl Viro 1954ce57dfc1SAl Viro if (err) { 1955626de996SAl Viro const char *s = get_link(nd); 19565a460275SAl Viro 1957a1c83681SViresh Kumar if (IS_ERR(s)) 19583595e234SAl Viro return PTR_ERR(s); 1959172a39a0SAl Viro err = 0; 196012b09578SAl Viro if (unlikely(!s)) { 196112b09578SAl Viro /* jumped */ 1962b9ff4429SAl Viro put_link(nd); 196312b09578SAl Viro } else { 1964071bf501SAl Viro nd->stack[nd->depth - 1].name = name; 196532cd7468SAl Viro name = s; 19669e18f10aSAl Viro continue; 196748c8b0c5SAl Viro } 196831e6b01fSNick Piggin } 196997242f99SAl Viro if (unlikely(!d_can_lookup(nd->path.dentry))) { 197097242f99SAl Viro if (nd->flags & LOOKUP_RCU) { 197197242f99SAl Viro if (unlazy_walk(nd, NULL, 0)) 197297242f99SAl Viro return -ECHILD; 197397242f99SAl Viro } 19743595e234SAl Viro return -ENOTDIR; 19755f4a6a69SAl Viro } 1976ce57dfc1SAl Viro } 197797242f99SAl Viro } 19781da177e4SLinus Torvalds 1979c8a53ee5SAl Viro static const char *path_init(struct nameidata *nd, unsigned flags) 198031e6b01fSNick Piggin { 198131e6b01fSNick Piggin int retval = 0; 1982c8a53ee5SAl Viro const char *s = nd->name->name; 198331e6b01fSNick Piggin 198431e6b01fSNick Piggin nd->last_type = LAST_ROOT; /* if there are only slashes... */ 1985980f3ea2SAl Viro nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT; 198631e6b01fSNick Piggin nd->depth = 0; 19875b6ca027SAl Viro if (flags & LOOKUP_ROOT) { 1988b18825a7SDavid Howells struct dentry *root = nd->root.dentry; 1989b18825a7SDavid Howells struct inode *inode = root->d_inode; 1990fd2f7cb5SAl Viro if (*s) { 199144b1d530SMiklos Szeredi if (!d_can_lookup(root)) 1992368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 19935b6ca027SAl Viro retval = inode_permission(inode, MAY_EXEC); 19945b6ca027SAl Viro if (retval) 1995368ee9baSAl Viro return ERR_PTR(retval); 199673d049a4SAl Viro } 19975b6ca027SAl Viro nd->path = nd->root; 19985b6ca027SAl Viro nd->inode = inode; 19995b6ca027SAl Viro if (flags & LOOKUP_RCU) { 20008b61e74fSAl Viro rcu_read_lock(); 20015b6ca027SAl Viro nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 20028f47a016SAl Viro nd->root_seq = nd->seq; 200348a066e7SAl Viro nd->m_seq = read_seqbegin(&mount_lock); 20045b6ca027SAl Viro } else { 20055b6ca027SAl Viro path_get(&nd->path); 20065b6ca027SAl Viro } 2007368ee9baSAl Viro return s; 20085b6ca027SAl Viro } 20095b6ca027SAl Viro 201031e6b01fSNick Piggin nd->root.mnt = NULL; 2011248fb5b9SAl Viro nd->path.mnt = NULL; 2012248fb5b9SAl Viro nd->path.dentry = NULL; 201331e6b01fSNick Piggin 201448a066e7SAl Viro nd->m_seq = read_seqbegin(&mount_lock); 2015fd2f7cb5SAl Viro if (*s == '/') { 20169e6697e2SAl Viro if (flags & LOOKUP_RCU) 20178b61e74fSAl Viro rcu_read_lock(); 2018e41f7d4eSAl Viro set_root(nd); 2019248fb5b9SAl Viro if (likely(!nd_jump_root(nd))) 2020ef55d917SAl Viro return s; 2021ef55d917SAl Viro nd->root.mnt = NULL; 2022ef55d917SAl Viro rcu_read_unlock(); 2023ef55d917SAl Viro return ERR_PTR(-ECHILD); 2024c8a53ee5SAl Viro } else if (nd->dfd == AT_FDCWD) { 2025e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 202631e6b01fSNick Piggin struct fs_struct *fs = current->fs; 2027c28cc364SNick Piggin unsigned seq; 202831e6b01fSNick Piggin 20298b61e74fSAl Viro rcu_read_lock(); 203031e6b01fSNick Piggin 2031c28cc364SNick Piggin do { 2032c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 203331e6b01fSNick Piggin nd->path = fs->pwd; 2034ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2035c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 2036c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 2037e41f7d4eSAl Viro } else { 2038e41f7d4eSAl Viro get_fs_pwd(current->fs, &nd->path); 2039ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2040e41f7d4eSAl Viro } 2041ef55d917SAl Viro return s; 204231e6b01fSNick Piggin } else { 2043582aa64aSJeff Layton /* Caller must check execute permissions on the starting path component */ 2044c8a53ee5SAl Viro struct fd f = fdget_raw(nd->dfd); 204531e6b01fSNick Piggin struct dentry *dentry; 204631e6b01fSNick Piggin 20472903ff01SAl Viro if (!f.file) 2048368ee9baSAl Viro return ERR_PTR(-EBADF); 204931e6b01fSNick Piggin 20502903ff01SAl Viro dentry = f.file->f_path.dentry; 205131e6b01fSNick Piggin 2052fd2f7cb5SAl Viro if (*s) { 205344b1d530SMiklos Szeredi if (!d_can_lookup(dentry)) { 20542903ff01SAl Viro fdput(f); 2055368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 2056f52e0c11SAl Viro } 20572903ff01SAl Viro } 20582903ff01SAl Viro 20592903ff01SAl Viro nd->path = f.file->f_path; 2060e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 20618b61e74fSAl Viro rcu_read_lock(); 206234a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 206334a26b99SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 20645590ff0dSUlrich Drepper } else { 20652903ff01SAl Viro path_get(&nd->path); 206634a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 20671da177e4SLinus Torvalds } 206834a26b99SAl Viro fdput(f); 2069368ee9baSAl Viro return s; 2070e41f7d4eSAl Viro } 20719b4a9b14SAl Viro } 20729b4a9b14SAl Viro 20733bdba28bSAl Viro static const char *trailing_symlink(struct nameidata *nd) 207495fa25d9SAl Viro { 207595fa25d9SAl Viro const char *s; 2076fec2fa24SAl Viro int error = may_follow_link(nd); 2077deb106c6SAl Viro if (unlikely(error)) 20783bdba28bSAl Viro return ERR_PTR(error); 207995fa25d9SAl Viro nd->flags |= LOOKUP_PARENT; 2080fab51e8aSAl Viro nd->stack[0].name = NULL; 20813b2e7f75SAl Viro s = get_link(nd); 2082deb106c6SAl Viro return s ? s : ""; 208395fa25d9SAl Viro } 208495fa25d9SAl Viro 2085caa85634SAl Viro static inline int lookup_last(struct nameidata *nd) 2086bd92d7feSAl Viro { 2087bd92d7feSAl Viro if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) 2088bd92d7feSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 2089bd92d7feSAl Viro 2090bd92d7feSAl Viro nd->flags &= ~LOOKUP_PARENT; 2091deb106c6SAl Viro return walk_component(nd, 20924693a547SAl Viro nd->flags & LOOKUP_FOLLOW 20934693a547SAl Viro ? nd->depth 20944693a547SAl Viro ? WALK_PUT | WALK_GET 20954693a547SAl Viro : WALK_GET 20964693a547SAl Viro : 0); 2097bd92d7feSAl Viro } 2098bd92d7feSAl Viro 20999b4a9b14SAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2100c8a53ee5SAl Viro static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path) 21019b4a9b14SAl Viro { 2102c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2103bd92d7feSAl Viro int err; 210431e6b01fSNick Piggin 2105368ee9baSAl Viro if (IS_ERR(s)) 2106368ee9baSAl Viro return PTR_ERR(s); 21073bdba28bSAl Viro while (!(err = link_path_walk(s, nd)) 21083bdba28bSAl Viro && ((err = lookup_last(nd)) > 0)) { 21093bdba28bSAl Viro s = trailing_symlink(nd); 21103bdba28bSAl Viro if (IS_ERR(s)) { 21113bdba28bSAl Viro err = PTR_ERR(s); 21126d7b5aaeSAl Viro break; 2113bd92d7feSAl Viro } 2114bd92d7feSAl Viro } 21159f1fafeeSAl Viro if (!err) 21169f1fafeeSAl Viro err = complete_walk(nd); 2117bd92d7feSAl Viro 2118deb106c6SAl Viro if (!err && nd->flags & LOOKUP_DIRECTORY) 2119deb106c6SAl Viro if (!d_can_lookup(nd->path.dentry)) 2120bd23a539SAl Viro err = -ENOTDIR; 2121625b6d10SAl Viro if (!err) { 2122625b6d10SAl Viro *path = nd->path; 2123625b6d10SAl Viro nd->path.mnt = NULL; 2124625b6d10SAl Viro nd->path.dentry = NULL; 2125625b6d10SAl Viro } 2126deb106c6SAl Viro terminate_walk(nd); 2127bd92d7feSAl Viro return err; 212831e6b01fSNick Piggin } 212931e6b01fSNick Piggin 2130625b6d10SAl Viro static int filename_lookup(int dfd, struct filename *name, unsigned flags, 21319ad1aaa6SAl Viro struct path *path, struct path *root) 2132873f1eedSJeff Layton { 2133894bc8c4SAl Viro int retval; 21349883d185SAl Viro struct nameidata nd; 2135abc9f5beSAl Viro if (IS_ERR(name)) 2136abc9f5beSAl Viro return PTR_ERR(name); 21379ad1aaa6SAl Viro if (unlikely(root)) { 21389ad1aaa6SAl Viro nd.root = *root; 21399ad1aaa6SAl Viro flags |= LOOKUP_ROOT; 21409ad1aaa6SAl Viro } 21419883d185SAl Viro set_nameidata(&nd, dfd, name); 2142c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_RCU, path); 2143873f1eedSJeff Layton if (unlikely(retval == -ECHILD)) 2144c8a53ee5SAl Viro retval = path_lookupat(&nd, flags, path); 2145873f1eedSJeff Layton if (unlikely(retval == -ESTALE)) 2146c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path); 2147873f1eedSJeff Layton 2148873f1eedSJeff Layton if (likely(!retval)) 2149625b6d10SAl Viro audit_inode(name, path->dentry, flags & LOOKUP_PARENT); 21509883d185SAl Viro restore_nameidata(); 2151e4bd1c1aSAl Viro putname(name); 2152873f1eedSJeff Layton return retval; 2153873f1eedSJeff Layton } 2154873f1eedSJeff Layton 21558bcb77faSAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2156c8a53ee5SAl Viro static int path_parentat(struct nameidata *nd, unsigned flags, 2157391172c4SAl Viro struct path *parent) 21588bcb77faSAl Viro { 2159c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2160368ee9baSAl Viro int err; 2161368ee9baSAl Viro if (IS_ERR(s)) 2162368ee9baSAl Viro return PTR_ERR(s); 2163368ee9baSAl Viro err = link_path_walk(s, nd); 21648bcb77faSAl Viro if (!err) 21658bcb77faSAl Viro err = complete_walk(nd); 2166391172c4SAl Viro if (!err) { 2167391172c4SAl Viro *parent = nd->path; 2168391172c4SAl Viro nd->path.mnt = NULL; 2169391172c4SAl Viro nd->path.dentry = NULL; 2170391172c4SAl Viro } 2171deb106c6SAl Viro terminate_walk(nd); 21728bcb77faSAl Viro return err; 21738bcb77faSAl Viro } 21748bcb77faSAl Viro 21755c31b6ceSAl Viro static struct filename *filename_parentat(int dfd, struct filename *name, 2176391172c4SAl Viro unsigned int flags, struct path *parent, 2177391172c4SAl Viro struct qstr *last, int *type) 21788bcb77faSAl Viro { 21798bcb77faSAl Viro int retval; 21809883d185SAl Viro struct nameidata nd; 21818bcb77faSAl Viro 21825c31b6ceSAl Viro if (IS_ERR(name)) 21835c31b6ceSAl Viro return name; 21849883d185SAl Viro set_nameidata(&nd, dfd, name); 2185c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_RCU, parent); 21868bcb77faSAl Viro if (unlikely(retval == -ECHILD)) 2187c8a53ee5SAl Viro retval = path_parentat(&nd, flags, parent); 21888bcb77faSAl Viro if (unlikely(retval == -ESTALE)) 2189c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent); 2190391172c4SAl Viro if (likely(!retval)) { 2191391172c4SAl Viro *last = nd.last; 2192391172c4SAl Viro *type = nd.last_type; 2193391172c4SAl Viro audit_inode(name, parent->dentry, LOOKUP_PARENT); 21945c31b6ceSAl Viro } else { 21955c31b6ceSAl Viro putname(name); 21965c31b6ceSAl Viro name = ERR_PTR(retval); 2197391172c4SAl Viro } 21989883d185SAl Viro restore_nameidata(); 21995c31b6ceSAl Viro return name; 22008bcb77faSAl Viro } 22018bcb77faSAl Viro 220279714f72SAl Viro /* does lookup, returns the object with parent locked */ 220379714f72SAl Viro struct dentry *kern_path_locked(const char *name, struct path *path) 22045590ff0dSUlrich Drepper { 22055c31b6ceSAl Viro struct filename *filename; 22065c31b6ceSAl Viro struct dentry *d; 2207391172c4SAl Viro struct qstr last; 2208391172c4SAl Viro int type; 220951689104SPaul Moore 22105c31b6ceSAl Viro filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path, 22115c31b6ceSAl Viro &last, &type); 221251689104SPaul Moore if (IS_ERR(filename)) 221351689104SPaul Moore return ERR_CAST(filename); 22145c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) { 2215391172c4SAl Viro path_put(path); 22165c31b6ceSAl Viro putname(filename); 22175c31b6ceSAl Viro return ERR_PTR(-EINVAL); 221879714f72SAl Viro } 22195955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 2220391172c4SAl Viro d = __lookup_hash(&last, path->dentry, 0); 222179714f72SAl Viro if (IS_ERR(d)) { 22225955102cSAl Viro inode_unlock(path->dentry->d_inode); 2223391172c4SAl Viro path_put(path); 222479714f72SAl Viro } 222551689104SPaul Moore putname(filename); 222679714f72SAl Viro return d; 22275590ff0dSUlrich Drepper } 22285590ff0dSUlrich Drepper 2229d1811465SAl Viro int kern_path(const char *name, unsigned int flags, struct path *path) 2230d1811465SAl Viro { 2231abc9f5beSAl Viro return filename_lookup(AT_FDCWD, getname_kernel(name), 2232abc9f5beSAl Viro flags, path, NULL); 2233d1811465SAl Viro } 22344d359507SAl Viro EXPORT_SYMBOL(kern_path); 2235d1811465SAl Viro 223616f18200SJosef 'Jeff' Sipek /** 223716f18200SJosef 'Jeff' Sipek * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair 223816f18200SJosef 'Jeff' Sipek * @dentry: pointer to dentry of the base directory 223916f18200SJosef 'Jeff' Sipek * @mnt: pointer to vfs mount of the base directory 224016f18200SJosef 'Jeff' Sipek * @name: pointer to file name 224116f18200SJosef 'Jeff' Sipek * @flags: lookup flags 2242e0a01249SAl Viro * @path: pointer to struct path to fill 224316f18200SJosef 'Jeff' Sipek */ 224416f18200SJosef 'Jeff' Sipek int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, 224516f18200SJosef 'Jeff' Sipek const char *name, unsigned int flags, 2246e0a01249SAl Viro struct path *path) 224716f18200SJosef 'Jeff' Sipek { 22489ad1aaa6SAl Viro struct path root = {.mnt = mnt, .dentry = dentry}; 22499ad1aaa6SAl Viro /* the first argument of filename_lookup() is ignored with root */ 2250abc9f5beSAl Viro return filename_lookup(AT_FDCWD, getname_kernel(name), 2251abc9f5beSAl Viro flags , path, &root); 225216f18200SJosef 'Jeff' Sipek } 22534d359507SAl Viro EXPORT_SYMBOL(vfs_path_lookup); 225416f18200SJosef 'Jeff' Sipek 2255eead1911SChristoph Hellwig /** 2256a6b91919SRandy Dunlap * lookup_one_len - filesystem helper to lookup single pathname component 2257eead1911SChristoph Hellwig * @name: pathname component to lookup 2258eead1911SChristoph Hellwig * @base: base directory to lookup from 2259eead1911SChristoph Hellwig * @len: maximum length @len should be interpreted to 2260eead1911SChristoph Hellwig * 2261a6b91919SRandy Dunlap * Note that this routine is purely a helper for filesystem usage and should 22629e7543e9SAl Viro * not be called by generic code. 2263bbddca8eSNeilBrown * 2264bbddca8eSNeilBrown * The caller must hold base->i_mutex. 2265eead1911SChristoph Hellwig */ 2266057f6c01SJames Morris struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 2267057f6c01SJames Morris { 2268057f6c01SJames Morris struct qstr this; 22696a96ba54SAl Viro unsigned int c; 2270cda309deSMiklos Szeredi int err; 2271057f6c01SJames Morris 22725955102cSAl Viro WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 22732f9092e1SDavid Woodhouse 22746a96ba54SAl Viro this.name = name; 22756a96ba54SAl Viro this.len = len; 22760145acc2SLinus Torvalds this.hash = full_name_hash(name, len); 22776a96ba54SAl Viro if (!len) 22786a96ba54SAl Viro return ERR_PTR(-EACCES); 22796a96ba54SAl Viro 228021d8a15aSAl Viro if (unlikely(name[0] == '.')) { 228121d8a15aSAl Viro if (len < 2 || (len == 2 && name[1] == '.')) 228221d8a15aSAl Viro return ERR_PTR(-EACCES); 228321d8a15aSAl Viro } 228421d8a15aSAl Viro 22856a96ba54SAl Viro while (len--) { 22866a96ba54SAl Viro c = *(const unsigned char *)name++; 22876a96ba54SAl Viro if (c == '/' || c == '\0') 22886a96ba54SAl Viro return ERR_PTR(-EACCES); 22896a96ba54SAl Viro } 22905a202bcdSAl Viro /* 22915a202bcdSAl Viro * See if the low-level filesystem might want 22925a202bcdSAl Viro * to use its own hash.. 22935a202bcdSAl Viro */ 22945a202bcdSAl Viro if (base->d_flags & DCACHE_OP_HASH) { 2295da53be12SLinus Torvalds int err = base->d_op->d_hash(base, &this); 22965a202bcdSAl Viro if (err < 0) 22975a202bcdSAl Viro return ERR_PTR(err); 22985a202bcdSAl Viro } 2299eead1911SChristoph Hellwig 2300cda309deSMiklos Szeredi err = inode_permission(base->d_inode, MAY_EXEC); 2301cda309deSMiklos Szeredi if (err) 2302cda309deSMiklos Szeredi return ERR_PTR(err); 2303cda309deSMiklos Szeredi 230472bd866aSAl Viro return __lookup_hash(&this, base, 0); 2305057f6c01SJames Morris } 23064d359507SAl Viro EXPORT_SYMBOL(lookup_one_len); 2307057f6c01SJames Morris 2308bbddca8eSNeilBrown /** 2309bbddca8eSNeilBrown * lookup_one_len_unlocked - filesystem helper to lookup single pathname component 2310bbddca8eSNeilBrown * @name: pathname component to lookup 2311bbddca8eSNeilBrown * @base: base directory to lookup from 2312bbddca8eSNeilBrown * @len: maximum length @len should be interpreted to 2313bbddca8eSNeilBrown * 2314bbddca8eSNeilBrown * Note that this routine is purely a helper for filesystem usage and should 2315bbddca8eSNeilBrown * not be called by generic code. 2316bbddca8eSNeilBrown * 2317bbddca8eSNeilBrown * Unlike lookup_one_len, it should be called without the parent 2318bbddca8eSNeilBrown * i_mutex held, and will take the i_mutex itself if necessary. 2319bbddca8eSNeilBrown */ 2320bbddca8eSNeilBrown struct dentry *lookup_one_len_unlocked(const char *name, 2321bbddca8eSNeilBrown struct dentry *base, int len) 2322bbddca8eSNeilBrown { 2323bbddca8eSNeilBrown struct qstr this; 2324bbddca8eSNeilBrown unsigned int c; 2325bbddca8eSNeilBrown int err; 2326bbddca8eSNeilBrown struct dentry *ret; 2327bbddca8eSNeilBrown 2328bbddca8eSNeilBrown this.name = name; 2329bbddca8eSNeilBrown this.len = len; 2330bbddca8eSNeilBrown this.hash = full_name_hash(name, len); 2331bbddca8eSNeilBrown if (!len) 2332bbddca8eSNeilBrown return ERR_PTR(-EACCES); 2333bbddca8eSNeilBrown 2334bbddca8eSNeilBrown if (unlikely(name[0] == '.')) { 2335bbddca8eSNeilBrown if (len < 2 || (len == 2 && name[1] == '.')) 2336bbddca8eSNeilBrown return ERR_PTR(-EACCES); 2337bbddca8eSNeilBrown } 2338bbddca8eSNeilBrown 2339bbddca8eSNeilBrown while (len--) { 2340bbddca8eSNeilBrown c = *(const unsigned char *)name++; 2341bbddca8eSNeilBrown if (c == '/' || c == '\0') 2342bbddca8eSNeilBrown return ERR_PTR(-EACCES); 2343bbddca8eSNeilBrown } 2344bbddca8eSNeilBrown /* 2345bbddca8eSNeilBrown * See if the low-level filesystem might want 2346bbddca8eSNeilBrown * to use its own hash.. 2347bbddca8eSNeilBrown */ 2348bbddca8eSNeilBrown if (base->d_flags & DCACHE_OP_HASH) { 2349bbddca8eSNeilBrown int err = base->d_op->d_hash(base, &this); 2350bbddca8eSNeilBrown if (err < 0) 2351bbddca8eSNeilBrown return ERR_PTR(err); 2352bbddca8eSNeilBrown } 2353bbddca8eSNeilBrown 2354bbddca8eSNeilBrown err = inode_permission(base->d_inode, MAY_EXEC); 2355bbddca8eSNeilBrown if (err) 2356bbddca8eSNeilBrown return ERR_PTR(err); 2357bbddca8eSNeilBrown 2358bbddca8eSNeilBrown /* 2359bbddca8eSNeilBrown * __d_lookup() is used to try to get a quick answer and avoid the 2360bbddca8eSNeilBrown * mutex. A false-negative does no harm. 2361bbddca8eSNeilBrown */ 2362bbddca8eSNeilBrown ret = __d_lookup(base, &this); 2363bbddca8eSNeilBrown if (ret && unlikely(ret->d_flags & DCACHE_OP_REVALIDATE)) { 2364bbddca8eSNeilBrown dput(ret); 2365bbddca8eSNeilBrown ret = NULL; 2366bbddca8eSNeilBrown } 2367bbddca8eSNeilBrown if (ret) 2368bbddca8eSNeilBrown return ret; 2369bbddca8eSNeilBrown 23705955102cSAl Viro inode_lock(base->d_inode); 2371bbddca8eSNeilBrown ret = __lookup_hash(&this, base, 0); 23725955102cSAl Viro inode_unlock(base->d_inode); 2373bbddca8eSNeilBrown return ret; 2374bbddca8eSNeilBrown } 2375bbddca8eSNeilBrown EXPORT_SYMBOL(lookup_one_len_unlocked); 2376bbddca8eSNeilBrown 23771fa1e7f6SAndy Whitcroft int user_path_at_empty(int dfd, const char __user *name, unsigned flags, 23781fa1e7f6SAndy Whitcroft struct path *path, int *empty) 23791da177e4SLinus Torvalds { 2380abc9f5beSAl Viro return filename_lookup(dfd, getname_flags(name, flags, empty), 2381abc9f5beSAl Viro flags, path, NULL); 23821da177e4SLinus Torvalds } 2383b853a161SAl Viro EXPORT_SYMBOL(user_path_at_empty); 23841fa1e7f6SAndy Whitcroft 2385873f1eedSJeff Layton /* 2386873f1eedSJeff Layton * NB: most callers don't do anything directly with the reference to the 2387873f1eedSJeff Layton * to struct filename, but the nd->last pointer points into the name string 2388873f1eedSJeff Layton * allocated by getname. So we must hold the reference to it until all 2389873f1eedSJeff Layton * path-walking is complete. 2390873f1eedSJeff Layton */ 2391a2ec4a2dSAl Viro static inline struct filename * 2392f5beed75SAl Viro user_path_parent(int dfd, const char __user *path, 2393f5beed75SAl Viro struct path *parent, 2394f5beed75SAl Viro struct qstr *last, 2395f5beed75SAl Viro int *type, 23969e790bd6SJeff Layton unsigned int flags) 23972ad94ae6SAl Viro { 23989e790bd6SJeff Layton /* only LOOKUP_REVAL is allowed in extra flags */ 23995c31b6ceSAl Viro return filename_parentat(dfd, getname(path), flags & LOOKUP_REVAL, 24005c31b6ceSAl Viro parent, last, type); 24012ad94ae6SAl Viro } 24022ad94ae6SAl Viro 24038033426eSJeff Layton /** 2404197df04cSAl Viro * mountpoint_last - look up last component for umount 24058033426eSJeff Layton * @nd: pathwalk nameidata - currently pointing at parent directory of "last" 24068033426eSJeff Layton * @path: pointer to container for result 24078033426eSJeff Layton * 24088033426eSJeff Layton * This is a special lookup_last function just for umount. In this case, we 24098033426eSJeff Layton * need to resolve the path without doing any revalidation. 24108033426eSJeff Layton * 24118033426eSJeff Layton * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since 24128033426eSJeff Layton * mountpoints are always pinned in the dcache, their ancestors are too. Thus, 24138033426eSJeff Layton * in almost all cases, this lookup will be served out of the dcache. The only 24148033426eSJeff Layton * cases where it won't are if nd->last refers to a symlink or the path is 24158033426eSJeff Layton * bogus and it doesn't exist. 24168033426eSJeff Layton * 24178033426eSJeff Layton * Returns: 24188033426eSJeff Layton * -error: if there was an error during lookup. This includes -ENOENT if the 24198033426eSJeff Layton * lookup found a negative dentry. The nd->path reference will also be 24208033426eSJeff Layton * put in this case. 24218033426eSJeff Layton * 24228033426eSJeff Layton * 0: if we successfully resolved nd->path and found it to not to be a 24238033426eSJeff Layton * symlink that needs to be followed. "path" will also be populated. 24248033426eSJeff Layton * The nd->path reference will also be put. 24258033426eSJeff Layton * 24268033426eSJeff Layton * 1: if we successfully resolved nd->last and found it to be a symlink 24278033426eSJeff Layton * that needs to be followed. "path" will be populated with the path 24288033426eSJeff Layton * to the link, and nd->path will *not* be put. 24298033426eSJeff Layton */ 24308033426eSJeff Layton static int 2431197df04cSAl Viro mountpoint_last(struct nameidata *nd, struct path *path) 24328033426eSJeff Layton { 24338033426eSJeff Layton int error = 0; 24348033426eSJeff Layton struct dentry *dentry; 24358033426eSJeff Layton struct dentry *dir = nd->path.dentry; 24368033426eSJeff Layton 243735759521SAl Viro /* If we're in rcuwalk, drop out of it to handle last component */ 243835759521SAl Viro if (nd->flags & LOOKUP_RCU) { 24396e9918b7SAl Viro if (unlazy_walk(nd, NULL, 0)) 2440deb106c6SAl Viro return -ECHILD; 24418033426eSJeff Layton } 24428033426eSJeff Layton 24438033426eSJeff Layton nd->flags &= ~LOOKUP_PARENT; 24448033426eSJeff Layton 24458033426eSJeff Layton if (unlikely(nd->last_type != LAST_NORM)) { 24468033426eSJeff Layton error = handle_dots(nd, nd->last_type); 244735759521SAl Viro if (error) 2448deb106c6SAl Viro return error; 24498033426eSJeff Layton dentry = dget(nd->path.dentry); 245035759521SAl Viro goto done; 24518033426eSJeff Layton } 24528033426eSJeff Layton 24535955102cSAl Viro inode_lock(dir->d_inode); 24548033426eSJeff Layton dentry = d_lookup(dir, &nd->last); 24558033426eSJeff Layton if (!dentry) { 24568033426eSJeff Layton /* 24578033426eSJeff Layton * No cached dentry. Mounted dentries are pinned in the cache, 24588033426eSJeff Layton * so that means that this dentry is probably a symlink or the 24598033426eSJeff Layton * path doesn't actually point to a mounted dentry. 24608033426eSJeff Layton */ 24618033426eSJeff Layton dentry = d_alloc(dir, &nd->last); 24628033426eSJeff Layton if (!dentry) { 24635955102cSAl Viro inode_unlock(dir->d_inode); 2464deb106c6SAl Viro return -ENOMEM; 24658033426eSJeff Layton } 246635759521SAl Viro dentry = lookup_real(dir->d_inode, dentry, nd->flags); 2467bcceeebaSDave Jones if (IS_ERR(dentry)) { 24685955102cSAl Viro inode_unlock(dir->d_inode); 2469deb106c6SAl Viro return PTR_ERR(dentry); 24708033426eSJeff Layton } 2471bcceeebaSDave Jones } 24725955102cSAl Viro inode_unlock(dir->d_inode); 24738033426eSJeff Layton 247435759521SAl Viro done: 2475698934dfSDavid Howells if (d_is_negative(dentry)) { 24768033426eSJeff Layton dput(dentry); 2477deb106c6SAl Viro return -ENOENT; 247835759521SAl Viro } 2479191d7f73SAl Viro if (nd->depth) 2480191d7f73SAl Viro put_link(nd); 24818033426eSJeff Layton path->dentry = dentry; 2482295dc39dSVasily Averin path->mnt = nd->path.mnt; 2483181548c0SAl Viro error = should_follow_link(nd, path, nd->flags & LOOKUP_FOLLOW, 2484181548c0SAl Viro d_backing_inode(dentry), 0); 2485deb106c6SAl Viro if (unlikely(error)) 2486d63ff28fSAl Viro return error; 2487295dc39dSVasily Averin mntget(path->mnt); 24888033426eSJeff Layton follow_mount(path); 2489deb106c6SAl Viro return 0; 24908033426eSJeff Layton } 24918033426eSJeff Layton 24928033426eSJeff Layton /** 2493197df04cSAl Viro * path_mountpoint - look up a path to be umounted 24942a78b857SMasanari Iida * @nd: lookup context 24958033426eSJeff Layton * @flags: lookup flags 2496c8a53ee5SAl Viro * @path: pointer to container for result 24978033426eSJeff Layton * 24988033426eSJeff Layton * Look up the given name, but don't attempt to revalidate the last component. 2499606d6fe3SRandy Dunlap * Returns 0 and "path" will be valid on success; Returns error otherwise. 25008033426eSJeff Layton */ 25018033426eSJeff Layton static int 2502c8a53ee5SAl Viro path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path) 25038033426eSJeff Layton { 2504c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2505368ee9baSAl Viro int err; 2506368ee9baSAl Viro if (IS_ERR(s)) 2507368ee9baSAl Viro return PTR_ERR(s); 25083bdba28bSAl Viro while (!(err = link_path_walk(s, nd)) && 25093bdba28bSAl Viro (err = mountpoint_last(nd, path)) > 0) { 25103bdba28bSAl Viro s = trailing_symlink(nd); 25113bdba28bSAl Viro if (IS_ERR(s)) { 25123bdba28bSAl Viro err = PTR_ERR(s); 25138033426eSJeff Layton break; 25148033426eSJeff Layton } 25153bdba28bSAl Viro } 2516deb106c6SAl Viro terminate_walk(nd); 25178033426eSJeff Layton return err; 25188033426eSJeff Layton } 25198033426eSJeff Layton 25202d864651SAl Viro static int 2521668696dcSAl Viro filename_mountpoint(int dfd, struct filename *name, struct path *path, 25222d864651SAl Viro unsigned int flags) 25232d864651SAl Viro { 25249883d185SAl Viro struct nameidata nd; 2525cbaab2dbSAl Viro int error; 2526668696dcSAl Viro if (IS_ERR(name)) 2527668696dcSAl Viro return PTR_ERR(name); 25289883d185SAl Viro set_nameidata(&nd, dfd, name); 2529c8a53ee5SAl Viro error = path_mountpoint(&nd, flags | LOOKUP_RCU, path); 25302d864651SAl Viro if (unlikely(error == -ECHILD)) 2531c8a53ee5SAl Viro error = path_mountpoint(&nd, flags, path); 25322d864651SAl Viro if (unlikely(error == -ESTALE)) 2533c8a53ee5SAl Viro error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path); 25342d864651SAl Viro if (likely(!error)) 2535668696dcSAl Viro audit_inode(name, path->dentry, 0); 25369883d185SAl Viro restore_nameidata(); 2537668696dcSAl Viro putname(name); 25382d864651SAl Viro return error; 25392d864651SAl Viro } 25402d864651SAl Viro 25418033426eSJeff Layton /** 2542197df04cSAl Viro * user_path_mountpoint_at - lookup a path from userland in order to umount it 25438033426eSJeff Layton * @dfd: directory file descriptor 25448033426eSJeff Layton * @name: pathname from userland 25458033426eSJeff Layton * @flags: lookup flags 25468033426eSJeff Layton * @path: pointer to container to hold result 25478033426eSJeff Layton * 25488033426eSJeff Layton * A umount is a special case for path walking. We're not actually interested 25498033426eSJeff Layton * in the inode in this situation, and ESTALE errors can be a problem. We 25508033426eSJeff Layton * simply want track down the dentry and vfsmount attached at the mountpoint 25518033426eSJeff Layton * and avoid revalidating the last component. 25528033426eSJeff Layton * 25538033426eSJeff Layton * Returns 0 and populates "path" on success. 25548033426eSJeff Layton */ 25558033426eSJeff Layton int 2556197df04cSAl Viro user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags, 25578033426eSJeff Layton struct path *path) 25588033426eSJeff Layton { 2559cbaab2dbSAl Viro return filename_mountpoint(dfd, getname(name), path, flags); 25608033426eSJeff Layton } 25618033426eSJeff Layton 25622d864651SAl Viro int 25632d864651SAl Viro kern_path_mountpoint(int dfd, const char *name, struct path *path, 25642d864651SAl Viro unsigned int flags) 25652d864651SAl Viro { 2566cbaab2dbSAl Viro return filename_mountpoint(dfd, getname_kernel(name), path, flags); 25672d864651SAl Viro } 25682d864651SAl Viro EXPORT_SYMBOL(kern_path_mountpoint); 25692d864651SAl Viro 2570cbdf35bcSMiklos Szeredi int __check_sticky(struct inode *dir, struct inode *inode) 25711da177e4SLinus Torvalds { 25728e96e3b7SEric W. Biederman kuid_t fsuid = current_fsuid(); 2573da9592edSDavid Howells 25748e96e3b7SEric W. Biederman if (uid_eq(inode->i_uid, fsuid)) 25751da177e4SLinus Torvalds return 0; 25768e96e3b7SEric W. Biederman if (uid_eq(dir->i_uid, fsuid)) 25771da177e4SLinus Torvalds return 0; 257823adbe12SAndy Lutomirski return !capable_wrt_inode_uidgid(inode, CAP_FOWNER); 25791da177e4SLinus Torvalds } 2580cbdf35bcSMiklos Szeredi EXPORT_SYMBOL(__check_sticky); 25811da177e4SLinus Torvalds 25821da177e4SLinus Torvalds /* 25831da177e4SLinus Torvalds * Check whether we can remove a link victim from directory dir, check 25841da177e4SLinus Torvalds * whether the type of victim is right. 25851da177e4SLinus Torvalds * 1. We can't do it if dir is read-only (done in permission()) 25861da177e4SLinus Torvalds * 2. We should have write and exec permissions on dir 25871da177e4SLinus Torvalds * 3. We can't remove anything from append-only dir 25881da177e4SLinus Torvalds * 4. We can't do anything with immutable dir (done in permission()) 25891da177e4SLinus Torvalds * 5. If the sticky bit on dir is set we should either 25901da177e4SLinus Torvalds * a. be owner of dir, or 25911da177e4SLinus Torvalds * b. be owner of victim, or 25921da177e4SLinus Torvalds * c. have CAP_FOWNER capability 25931da177e4SLinus Torvalds * 6. If the victim is append-only or immutable we can't do antyhing with 25941da177e4SLinus Torvalds * links pointing to it. 25951da177e4SLinus Torvalds * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. 25961da177e4SLinus Torvalds * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. 25971da177e4SLinus Torvalds * 9. We can't remove a root or mountpoint. 25981da177e4SLinus Torvalds * 10. We don't allow removal of NFS sillyrenamed files; it's handled by 25991da177e4SLinus Torvalds * nfs_async_unlink(). 26001da177e4SLinus Torvalds */ 2601b18825a7SDavid Howells static int may_delete(struct inode *dir, struct dentry *victim, bool isdir) 26021da177e4SLinus Torvalds { 260363afdfc7SDavid Howells struct inode *inode = d_backing_inode(victim); 26041da177e4SLinus Torvalds int error; 26051da177e4SLinus Torvalds 2606b18825a7SDavid Howells if (d_is_negative(victim)) 26071da177e4SLinus Torvalds return -ENOENT; 2608b18825a7SDavid Howells BUG_ON(!inode); 26091da177e4SLinus Torvalds 26101da177e4SLinus Torvalds BUG_ON(victim->d_parent->d_inode != dir); 26114fa6b5ecSJeff Layton audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); 26121da177e4SLinus Torvalds 2613f419a2e3SAl Viro error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 26141da177e4SLinus Torvalds if (error) 26151da177e4SLinus Torvalds return error; 26161da177e4SLinus Torvalds if (IS_APPEND(dir)) 26171da177e4SLinus Torvalds return -EPERM; 2618b18825a7SDavid Howells 2619b18825a7SDavid Howells if (check_sticky(dir, inode) || IS_APPEND(inode) || 2620b18825a7SDavid Howells IS_IMMUTABLE(inode) || IS_SWAPFILE(inode)) 26211da177e4SLinus Torvalds return -EPERM; 26221da177e4SLinus Torvalds if (isdir) { 262344b1d530SMiklos Szeredi if (!d_is_dir(victim)) 26241da177e4SLinus Torvalds return -ENOTDIR; 26251da177e4SLinus Torvalds if (IS_ROOT(victim)) 26261da177e4SLinus Torvalds return -EBUSY; 262744b1d530SMiklos Szeredi } else if (d_is_dir(victim)) 26281da177e4SLinus Torvalds return -EISDIR; 26291da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 26301da177e4SLinus Torvalds return -ENOENT; 26311da177e4SLinus Torvalds if (victim->d_flags & DCACHE_NFSFS_RENAMED) 26321da177e4SLinus Torvalds return -EBUSY; 26331da177e4SLinus Torvalds return 0; 26341da177e4SLinus Torvalds } 26351da177e4SLinus Torvalds 26361da177e4SLinus Torvalds /* Check whether we can create an object with dentry child in directory 26371da177e4SLinus Torvalds * dir. 26381da177e4SLinus Torvalds * 1. We can't do it if child already exists (open has special treatment for 26391da177e4SLinus Torvalds * this case, but since we are inlined it's OK) 26401da177e4SLinus Torvalds * 2. We can't do it if dir is read-only (done in permission()) 26411da177e4SLinus Torvalds * 3. We should have write and exec permissions on dir 26421da177e4SLinus Torvalds * 4. We can't do it if dir is immutable (done in permission()) 26431da177e4SLinus Torvalds */ 2644a95164d9SMiklos Szeredi static inline int may_create(struct inode *dir, struct dentry *child) 26451da177e4SLinus Torvalds { 264614e972b4SJeff Layton audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); 26471da177e4SLinus Torvalds if (child->d_inode) 26481da177e4SLinus Torvalds return -EEXIST; 26491da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 26501da177e4SLinus Torvalds return -ENOENT; 2651f419a2e3SAl Viro return inode_permission(dir, MAY_WRITE | MAY_EXEC); 26521da177e4SLinus Torvalds } 26531da177e4SLinus Torvalds 26541da177e4SLinus Torvalds /* 26551da177e4SLinus Torvalds * p1 and p2 should be directories on the same fs. 26561da177e4SLinus Torvalds */ 26571da177e4SLinus Torvalds struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 26581da177e4SLinus Torvalds { 26591da177e4SLinus Torvalds struct dentry *p; 26601da177e4SLinus Torvalds 26611da177e4SLinus Torvalds if (p1 == p2) { 26625955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 26631da177e4SLinus Torvalds return NULL; 26641da177e4SLinus Torvalds } 26651da177e4SLinus Torvalds 2666a11f3a05SArjan van de Ven mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 26671da177e4SLinus Torvalds 2668e2761a11SOGAWA Hirofumi p = d_ancestor(p2, p1); 2669e2761a11SOGAWA Hirofumi if (p) { 26705955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT); 26715955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_CHILD); 26721da177e4SLinus Torvalds return p; 26731da177e4SLinus Torvalds } 26741da177e4SLinus Torvalds 2675e2761a11SOGAWA Hirofumi p = d_ancestor(p1, p2); 2676e2761a11SOGAWA Hirofumi if (p) { 26775955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 26785955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_CHILD); 26791da177e4SLinus Torvalds return p; 26801da177e4SLinus Torvalds } 26811da177e4SLinus Torvalds 26825955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 26835955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2); 26841da177e4SLinus Torvalds return NULL; 26851da177e4SLinus Torvalds } 26864d359507SAl Viro EXPORT_SYMBOL(lock_rename); 26871da177e4SLinus Torvalds 26881da177e4SLinus Torvalds void unlock_rename(struct dentry *p1, struct dentry *p2) 26891da177e4SLinus Torvalds { 26905955102cSAl Viro inode_unlock(p1->d_inode); 26911da177e4SLinus Torvalds if (p1 != p2) { 26925955102cSAl Viro inode_unlock(p2->d_inode); 2693a11f3a05SArjan van de Ven mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 26941da177e4SLinus Torvalds } 26951da177e4SLinus Torvalds } 26964d359507SAl Viro EXPORT_SYMBOL(unlock_rename); 26971da177e4SLinus Torvalds 26984acdaf27SAl Viro int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, 2699312b63fbSAl Viro bool want_excl) 27001da177e4SLinus Torvalds { 2701a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 27021da177e4SLinus Torvalds if (error) 27031da177e4SLinus Torvalds return error; 27041da177e4SLinus Torvalds 2705acfa4380SAl Viro if (!dir->i_op->create) 27061da177e4SLinus Torvalds return -EACCES; /* shouldn't it be ENOSYS? */ 27071da177e4SLinus Torvalds mode &= S_IALLUGO; 27081da177e4SLinus Torvalds mode |= S_IFREG; 27091da177e4SLinus Torvalds error = security_inode_create(dir, dentry, mode); 27101da177e4SLinus Torvalds if (error) 27111da177e4SLinus Torvalds return error; 2712312b63fbSAl Viro error = dir->i_op->create(dir, dentry, mode, want_excl); 2713a74574aaSStephen Smalley if (!error) 2714f38aa942SAmy Griffis fsnotify_create(dir, dentry); 27151da177e4SLinus Torvalds return error; 27161da177e4SLinus Torvalds } 27174d359507SAl Viro EXPORT_SYMBOL(vfs_create); 27181da177e4SLinus Torvalds 271973d049a4SAl Viro static int may_open(struct path *path, int acc_mode, int flag) 27201da177e4SLinus Torvalds { 27213fb64190SChristoph Hellwig struct dentry *dentry = path->dentry; 27221da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 27231da177e4SLinus Torvalds int error; 27241da177e4SLinus Torvalds 27251da177e4SLinus Torvalds if (!inode) 27261da177e4SLinus Torvalds return -ENOENT; 27271da177e4SLinus Torvalds 2728c8fe8f30SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 2729c8fe8f30SChristoph Hellwig case S_IFLNK: 27301da177e4SLinus Torvalds return -ELOOP; 2731c8fe8f30SChristoph Hellwig case S_IFDIR: 2732c8fe8f30SChristoph Hellwig if (acc_mode & MAY_WRITE) 27331da177e4SLinus Torvalds return -EISDIR; 2734c8fe8f30SChristoph Hellwig break; 2735c8fe8f30SChristoph Hellwig case S_IFBLK: 2736c8fe8f30SChristoph Hellwig case S_IFCHR: 27373fb64190SChristoph Hellwig if (path->mnt->mnt_flags & MNT_NODEV) 27381da177e4SLinus Torvalds return -EACCES; 2739c8fe8f30SChristoph Hellwig /*FALLTHRU*/ 2740c8fe8f30SChristoph Hellwig case S_IFIFO: 2741c8fe8f30SChristoph Hellwig case S_IFSOCK: 27421da177e4SLinus Torvalds flag &= ~O_TRUNC; 2743c8fe8f30SChristoph Hellwig break; 27444a3fd211SDave Hansen } 2745b41572e9SDave Hansen 274662fb4a15SAl Viro error = inode_permission(inode, MAY_OPEN | acc_mode); 2747b41572e9SDave Hansen if (error) 2748b41572e9SDave Hansen return error; 27496146f0d5SMimi Zohar 27501da177e4SLinus Torvalds /* 27511da177e4SLinus Torvalds * An append-only file must be opened in append mode for writing. 27521da177e4SLinus Torvalds */ 27531da177e4SLinus Torvalds if (IS_APPEND(inode)) { 27548737c930SAl Viro if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) 27557715b521SAl Viro return -EPERM; 27561da177e4SLinus Torvalds if (flag & O_TRUNC) 27577715b521SAl Viro return -EPERM; 27581da177e4SLinus Torvalds } 27591da177e4SLinus Torvalds 27601da177e4SLinus Torvalds /* O_NOATIME can only be set by the owner or superuser */ 27612e149670SSerge E. Hallyn if (flag & O_NOATIME && !inode_owner_or_capable(inode)) 27627715b521SAl Viro return -EPERM; 27631da177e4SLinus Torvalds 2764f3c7691eSJ. Bruce Fields return 0; 27657715b521SAl Viro } 27667715b521SAl Viro 2767e1181ee6SJeff Layton static int handle_truncate(struct file *filp) 27687715b521SAl Viro { 2769e1181ee6SJeff Layton struct path *path = &filp->f_path; 27707715b521SAl Viro struct inode *inode = path->dentry->d_inode; 27717715b521SAl Viro int error = get_write_access(inode); 27721da177e4SLinus Torvalds if (error) 27737715b521SAl Viro return error; 27741da177e4SLinus Torvalds /* 27751da177e4SLinus Torvalds * Refuse to truncate files with mandatory locks held on them. 27761da177e4SLinus Torvalds */ 2777d7a06983SJeff Layton error = locks_verify_locked(filp); 2778be6d3e56SKentaro Takeda if (!error) 2779ea0d3ab2STetsuo Handa error = security_path_truncate(path); 27801da177e4SLinus Torvalds if (!error) { 27817715b521SAl Viro error = do_truncate(path->dentry, 0, 2782d139d7ffSMiklos Szeredi ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, 2783e1181ee6SJeff Layton filp); 27841da177e4SLinus Torvalds } 27851da177e4SLinus Torvalds put_write_access(inode); 2786acd0c935SMimi Zohar return error; 27871da177e4SLinus Torvalds } 27881da177e4SLinus Torvalds 2789d57999e1SDave Hansen static inline int open_to_namei_flags(int flag) 2790d57999e1SDave Hansen { 27918a5e929dSAl Viro if ((flag & O_ACCMODE) == 3) 27928a5e929dSAl Viro flag--; 2793d57999e1SDave Hansen return flag; 2794d57999e1SDave Hansen } 2795d57999e1SDave Hansen 2796d18e9008SMiklos Szeredi static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode) 2797d18e9008SMiklos Szeredi { 2798d18e9008SMiklos Szeredi int error = security_path_mknod(dir, dentry, mode, 0); 2799d18e9008SMiklos Szeredi if (error) 2800d18e9008SMiklos Szeredi return error; 2801d18e9008SMiklos Szeredi 2802d18e9008SMiklos Szeredi error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC); 2803d18e9008SMiklos Szeredi if (error) 2804d18e9008SMiklos Szeredi return error; 2805d18e9008SMiklos Szeredi 2806d18e9008SMiklos Szeredi return security_inode_create(dir->dentry->d_inode, dentry, mode); 2807d18e9008SMiklos Szeredi } 2808d18e9008SMiklos Szeredi 28091acf0af9SDavid Howells /* 28101acf0af9SDavid Howells * Attempt to atomically look up, create and open a file from a negative 28111acf0af9SDavid Howells * dentry. 28121acf0af9SDavid Howells * 28131acf0af9SDavid Howells * Returns 0 if successful. The file will have been created and attached to 28141acf0af9SDavid Howells * @file by the filesystem calling finish_open(). 28151acf0af9SDavid Howells * 28161acf0af9SDavid Howells * Returns 1 if the file was looked up only or didn't need creating. The 28171acf0af9SDavid Howells * caller will need to perform the open themselves. @path will have been 28181acf0af9SDavid Howells * updated to point to the new dentry. This may be negative. 28191acf0af9SDavid Howells * 28201acf0af9SDavid Howells * Returns an error code otherwise. 28211acf0af9SDavid Howells */ 28222675a4ebSAl Viro static int atomic_open(struct nameidata *nd, struct dentry *dentry, 282330d90494SAl Viro struct path *path, struct file *file, 2824015c3bbcSMiklos Szeredi const struct open_flags *op, 282564894cf8SAl Viro bool got_write, bool need_lookup, 282647237687SAl Viro int *opened) 2827d18e9008SMiklos Szeredi { 2828d18e9008SMiklos Szeredi struct inode *dir = nd->path.dentry->d_inode; 2829d18e9008SMiklos Szeredi unsigned open_flag = open_to_namei_flags(op->open_flag); 2830d18e9008SMiklos Szeredi umode_t mode; 2831d18e9008SMiklos Szeredi int error; 2832d18e9008SMiklos Szeredi int acc_mode; 2833d18e9008SMiklos Szeredi int create_error = 0; 2834d18e9008SMiklos Szeredi struct dentry *const DENTRY_NOT_SET = (void *) -1UL; 2835116cc022SMiklos Szeredi bool excl; 2836d18e9008SMiklos Szeredi 2837d18e9008SMiklos Szeredi BUG_ON(dentry->d_inode); 2838d18e9008SMiklos Szeredi 2839d18e9008SMiklos Szeredi /* Don't create child dentry for a dead directory. */ 2840d18e9008SMiklos Szeredi if (unlikely(IS_DEADDIR(dir))) { 28412675a4ebSAl Viro error = -ENOENT; 2842d18e9008SMiklos Szeredi goto out; 2843d18e9008SMiklos Szeredi } 2844d18e9008SMiklos Szeredi 284562b259d8SMiklos Szeredi mode = op->mode; 2846d18e9008SMiklos Szeredi if ((open_flag & O_CREAT) && !IS_POSIXACL(dir)) 2847d18e9008SMiklos Szeredi mode &= ~current_umask(); 2848d18e9008SMiklos Szeredi 2849116cc022SMiklos Szeredi excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT); 2850116cc022SMiklos Szeredi if (excl) 2851d18e9008SMiklos Szeredi open_flag &= ~O_TRUNC; 2852d18e9008SMiklos Szeredi 2853d18e9008SMiklos Szeredi /* 2854d18e9008SMiklos Szeredi * Checking write permission is tricky, bacuse we don't know if we are 2855d18e9008SMiklos Szeredi * going to actually need it: O_CREAT opens should work as long as the 2856d18e9008SMiklos Szeredi * file exists. But checking existence breaks atomicity. The trick is 2857d18e9008SMiklos Szeredi * to check access and if not granted clear O_CREAT from the flags. 2858d18e9008SMiklos Szeredi * 2859d18e9008SMiklos Szeredi * Another problem is returing the "right" error value (e.g. for an 2860d18e9008SMiklos Szeredi * O_EXCL open we want to return EEXIST not EROFS). 2861d18e9008SMiklos Szeredi */ 286264894cf8SAl Viro if (((open_flag & (O_CREAT | O_TRUNC)) || 286364894cf8SAl Viro (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) { 286464894cf8SAl Viro if (!(open_flag & O_CREAT)) { 2865d18e9008SMiklos Szeredi /* 2866d18e9008SMiklos Szeredi * No O_CREATE -> atomicity not a requirement -> fall 2867d18e9008SMiklos Szeredi * back to lookup + open 2868d18e9008SMiklos Szeredi */ 2869d18e9008SMiklos Szeredi goto no_open; 2870d18e9008SMiklos Szeredi } else if (open_flag & (O_EXCL | O_TRUNC)) { 2871d18e9008SMiklos Szeredi /* Fall back and fail with the right error */ 287264894cf8SAl Viro create_error = -EROFS; 2873d18e9008SMiklos Szeredi goto no_open; 2874d18e9008SMiklos Szeredi } else { 2875d18e9008SMiklos Szeredi /* No side effects, safe to clear O_CREAT */ 287664894cf8SAl Viro create_error = -EROFS; 2877d18e9008SMiklos Szeredi open_flag &= ~O_CREAT; 2878d18e9008SMiklos Szeredi } 2879d18e9008SMiklos Szeredi } 2880d18e9008SMiklos Szeredi 2881d18e9008SMiklos Szeredi if (open_flag & O_CREAT) { 288238227f78SMiklos Szeredi error = may_o_create(&nd->path, dentry, mode); 2883d18e9008SMiklos Szeredi if (error) { 2884d18e9008SMiklos Szeredi create_error = error; 2885d18e9008SMiklos Szeredi if (open_flag & O_EXCL) 2886d18e9008SMiklos Szeredi goto no_open; 2887d18e9008SMiklos Szeredi open_flag &= ~O_CREAT; 2888d18e9008SMiklos Szeredi } 2889d18e9008SMiklos Szeredi } 2890d18e9008SMiklos Szeredi 2891d18e9008SMiklos Szeredi if (nd->flags & LOOKUP_DIRECTORY) 2892d18e9008SMiklos Szeredi open_flag |= O_DIRECTORY; 2893d18e9008SMiklos Szeredi 289430d90494SAl Viro file->f_path.dentry = DENTRY_NOT_SET; 289530d90494SAl Viro file->f_path.mnt = nd->path.mnt; 289630d90494SAl Viro error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode, 289747237687SAl Viro opened); 2898d9585277SAl Viro if (error < 0) { 2899d9585277SAl Viro if (create_error && error == -ENOENT) 2900d9585277SAl Viro error = create_error; 2901d18e9008SMiklos Szeredi goto out; 2902d18e9008SMiklos Szeredi } 2903d18e9008SMiklos Szeredi 2904d9585277SAl Viro if (error) { /* returned 1, that is */ 290530d90494SAl Viro if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { 29062675a4ebSAl Viro error = -EIO; 2907d18e9008SMiklos Szeredi goto out; 2908d18e9008SMiklos Szeredi } 290930d90494SAl Viro if (file->f_path.dentry) { 2910d18e9008SMiklos Szeredi dput(dentry); 291130d90494SAl Viro dentry = file->f_path.dentry; 2912d18e9008SMiklos Szeredi } 291303da633aSAl Viro if (*opened & FILE_CREATED) 291403da633aSAl Viro fsnotify_create(dir, dentry); 291503da633aSAl Viro if (!dentry->d_inode) { 291603da633aSAl Viro WARN_ON(*opened & FILE_CREATED); 291703da633aSAl Viro if (create_error) { 291862b2ce96SSage Weil error = create_error; 291962b2ce96SSage Weil goto out; 292062b2ce96SSage Weil } 292103da633aSAl Viro } else { 292203da633aSAl Viro if (excl && !(*opened & FILE_CREATED)) { 292303da633aSAl Viro error = -EEXIST; 292403da633aSAl Viro goto out; 292503da633aSAl Viro } 292603da633aSAl Viro } 2927d18e9008SMiklos Szeredi goto looked_up; 2928d18e9008SMiklos Szeredi } 2929d18e9008SMiklos Szeredi 2930d18e9008SMiklos Szeredi /* 2931d18e9008SMiklos Szeredi * We didn't have the inode before the open, so check open permission 2932d18e9008SMiklos Szeredi * here. 2933d18e9008SMiklos Szeredi */ 293403da633aSAl Viro acc_mode = op->acc_mode; 293503da633aSAl Viro if (*opened & FILE_CREATED) { 293603da633aSAl Viro WARN_ON(!(open_flag & O_CREAT)); 293703da633aSAl Viro fsnotify_create(dir, dentry); 293862fb4a15SAl Viro acc_mode = 0; 293903da633aSAl Viro } 29402675a4ebSAl Viro error = may_open(&file->f_path, acc_mode, open_flag); 29412675a4ebSAl Viro if (error) 29422675a4ebSAl Viro fput(file); 2943d18e9008SMiklos Szeredi 2944d18e9008SMiklos Szeredi out: 2945d18e9008SMiklos Szeredi dput(dentry); 29462675a4ebSAl Viro return error; 2947d18e9008SMiklos Szeredi 2948d18e9008SMiklos Szeredi no_open: 2949d18e9008SMiklos Szeredi if (need_lookup) { 295072bd866aSAl Viro dentry = lookup_real(dir, dentry, nd->flags); 2951d18e9008SMiklos Szeredi if (IS_ERR(dentry)) 29522675a4ebSAl Viro return PTR_ERR(dentry); 2953d18e9008SMiklos Szeredi 2954d18e9008SMiklos Szeredi if (create_error) { 2955d18e9008SMiklos Szeredi int open_flag = op->open_flag; 2956d18e9008SMiklos Szeredi 29572675a4ebSAl Viro error = create_error; 2958d18e9008SMiklos Szeredi if ((open_flag & O_EXCL)) { 2959d18e9008SMiklos Szeredi if (!dentry->d_inode) 2960d18e9008SMiklos Szeredi goto out; 2961d18e9008SMiklos Szeredi } else if (!dentry->d_inode) { 2962d18e9008SMiklos Szeredi goto out; 2963d18e9008SMiklos Szeredi } else if ((open_flag & O_TRUNC) && 2964e36cb0b8SDavid Howells d_is_reg(dentry)) { 2965d18e9008SMiklos Szeredi goto out; 2966d18e9008SMiklos Szeredi } 2967d18e9008SMiklos Szeredi /* will fail later, go on to get the right error */ 2968d18e9008SMiklos Szeredi } 2969d18e9008SMiklos Szeredi } 2970d18e9008SMiklos Szeredi looked_up: 2971d18e9008SMiklos Szeredi path->dentry = dentry; 2972d18e9008SMiklos Szeredi path->mnt = nd->path.mnt; 29732675a4ebSAl Viro return 1; 2974d18e9008SMiklos Szeredi } 2975d18e9008SMiklos Szeredi 297631e6b01fSNick Piggin /* 29771acf0af9SDavid Howells * Look up and maybe create and open the last component. 2978d58ffd35SMiklos Szeredi * 2979d58ffd35SMiklos Szeredi * Must be called with i_mutex held on parent. 2980d58ffd35SMiklos Szeredi * 29811acf0af9SDavid Howells * Returns 0 if the file was successfully atomically created (if necessary) and 29821acf0af9SDavid Howells * opened. In this case the file will be returned attached to @file. 29831acf0af9SDavid Howells * 29841acf0af9SDavid Howells * Returns 1 if the file was not completely opened at this time, though lookups 29851acf0af9SDavid Howells * and creations will have been performed and the dentry returned in @path will 29861acf0af9SDavid Howells * be positive upon return if O_CREAT was specified. If O_CREAT wasn't 29871acf0af9SDavid Howells * specified then a negative dentry may be returned. 29881acf0af9SDavid Howells * 29891acf0af9SDavid Howells * An error code is returned otherwise. 29901acf0af9SDavid Howells * 29911acf0af9SDavid Howells * FILE_CREATE will be set in @*opened if the dentry was created and will be 29921acf0af9SDavid Howells * cleared otherwise prior to returning. 2993d58ffd35SMiklos Szeredi */ 29942675a4ebSAl Viro static int lookup_open(struct nameidata *nd, struct path *path, 299530d90494SAl Viro struct file *file, 2996d58ffd35SMiklos Szeredi const struct open_flags *op, 299764894cf8SAl Viro bool got_write, int *opened) 2998d58ffd35SMiklos Szeredi { 2999d58ffd35SMiklos Szeredi struct dentry *dir = nd->path.dentry; 300054ef4872SMiklos Szeredi struct inode *dir_inode = dir->d_inode; 3001d58ffd35SMiklos Szeredi struct dentry *dentry; 3002d58ffd35SMiklos Szeredi int error; 30036c51e513SAl Viro bool need_lookup = false; 3004d58ffd35SMiklos Szeredi 300547237687SAl Viro *opened &= ~FILE_CREATED; 30066c51e513SAl Viro dentry = lookup_dcache(&nd->last, dir, nd->flags); 3007d58ffd35SMiklos Szeredi if (IS_ERR(dentry)) 30082675a4ebSAl Viro return PTR_ERR(dentry); 3009d58ffd35SMiklos Szeredi 30106c51e513SAl Viro if (!dentry) { 30116c51e513SAl Viro dentry = d_alloc(dir, &nd->last); 30126c51e513SAl Viro if (unlikely(!dentry)) 30136c51e513SAl Viro return -ENOMEM; 30146c51e513SAl Viro need_lookup = true; 30156c51e513SAl Viro } else if (dentry->d_inode) { 3016d18e9008SMiklos Szeredi /* Cached positive dentry: will open in f_op->open */ 3017d18e9008SMiklos Szeredi goto out_no_open; 30186c51e513SAl Viro } 3019d18e9008SMiklos Szeredi 3020d18e9008SMiklos Szeredi if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) { 302164894cf8SAl Viro return atomic_open(nd, dentry, path, file, op, got_write, 302247237687SAl Viro need_lookup, opened); 3023d18e9008SMiklos Szeredi } 3024d18e9008SMiklos Szeredi 302554ef4872SMiklos Szeredi if (need_lookup) { 302654ef4872SMiklos Szeredi BUG_ON(dentry->d_inode); 302754ef4872SMiklos Szeredi 302872bd866aSAl Viro dentry = lookup_real(dir_inode, dentry, nd->flags); 302954ef4872SMiklos Szeredi if (IS_ERR(dentry)) 30302675a4ebSAl Viro return PTR_ERR(dentry); 303154ef4872SMiklos Szeredi } 303254ef4872SMiklos Szeredi 3033d58ffd35SMiklos Szeredi /* Negative dentry, just create the file */ 3034d58ffd35SMiklos Szeredi if (!dentry->d_inode && (op->open_flag & O_CREAT)) { 3035d58ffd35SMiklos Szeredi umode_t mode = op->mode; 3036d58ffd35SMiklos Szeredi if (!IS_POSIXACL(dir->d_inode)) 3037d58ffd35SMiklos Szeredi mode &= ~current_umask(); 3038d58ffd35SMiklos Szeredi /* 3039d58ffd35SMiklos Szeredi * This write is needed to ensure that a 3040d58ffd35SMiklos Szeredi * rw->ro transition does not occur between 3041d58ffd35SMiklos Szeredi * the time when the file is created and when 3042d58ffd35SMiklos Szeredi * a permanent write count is taken through 3043015c3bbcSMiklos Szeredi * the 'struct file' in finish_open(). 3044d58ffd35SMiklos Szeredi */ 304564894cf8SAl Viro if (!got_write) { 304664894cf8SAl Viro error = -EROFS; 3047d58ffd35SMiklos Szeredi goto out_dput; 304864894cf8SAl Viro } 304947237687SAl Viro *opened |= FILE_CREATED; 3050d58ffd35SMiklos Szeredi error = security_path_mknod(&nd->path, dentry, mode, 0); 3051d58ffd35SMiklos Szeredi if (error) 3052d58ffd35SMiklos Szeredi goto out_dput; 3053312b63fbSAl Viro error = vfs_create(dir->d_inode, dentry, mode, 3054312b63fbSAl Viro nd->flags & LOOKUP_EXCL); 3055d58ffd35SMiklos Szeredi if (error) 3056d58ffd35SMiklos Szeredi goto out_dput; 3057d58ffd35SMiklos Szeredi } 3058d18e9008SMiklos Szeredi out_no_open: 3059d58ffd35SMiklos Szeredi path->dentry = dentry; 3060d58ffd35SMiklos Szeredi path->mnt = nd->path.mnt; 30612675a4ebSAl Viro return 1; 3062d58ffd35SMiklos Szeredi 3063d58ffd35SMiklos Szeredi out_dput: 3064d58ffd35SMiklos Szeredi dput(dentry); 30652675a4ebSAl Viro return error; 3066d58ffd35SMiklos Szeredi } 3067d58ffd35SMiklos Szeredi 3068d58ffd35SMiklos Szeredi /* 3069fe2d35ffSAl Viro * Handle the last step of open() 307031e6b01fSNick Piggin */ 3071896475d5SAl Viro static int do_last(struct nameidata *nd, 307230d90494SAl Viro struct file *file, const struct open_flags *op, 307376ae2a5aSAl Viro int *opened) 3074fb1cc555SAl Viro { 3075a1e28038SAl Viro struct dentry *dir = nd->path.dentry; 3076ca344a89SAl Viro int open_flag = op->open_flag; 307777d660a8SMiklos Szeredi bool will_truncate = (open_flag & O_TRUNC) != 0; 307864894cf8SAl Viro bool got_write = false; 3079bcda7652SAl Viro int acc_mode = op->acc_mode; 3080254cf582SAl Viro unsigned seq; 3081a1eb3315SMiklos Szeredi struct inode *inode; 308216b1c1cdSMiklos Szeredi struct path save_parent = { .dentry = NULL, .mnt = NULL }; 3083896475d5SAl Viro struct path path; 308416b1c1cdSMiklos Szeredi bool retried = false; 308516c2cd71SAl Viro int error; 3086fb1cc555SAl Viro 3087c3e380b0SAl Viro nd->flags &= ~LOOKUP_PARENT; 3088c3e380b0SAl Viro nd->flags |= op->intent; 3089c3e380b0SAl Viro 3090bc77daa7SAl Viro if (nd->last_type != LAST_NORM) { 3091fe2d35ffSAl Viro error = handle_dots(nd, nd->last_type); 3092deb106c6SAl Viro if (unlikely(error)) 30932675a4ebSAl Viro return error; 3094e83db167SMiklos Szeredi goto finish_open; 30951f36f774SAl Viro } 3096a2c36b45SAl Viro 3097ca344a89SAl Viro if (!(open_flag & O_CREAT)) { 3098fe2d35ffSAl Viro if (nd->last.name[nd->last.len]) 3099fe2d35ffSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 3100fe2d35ffSAl Viro /* we _can_ be in RCU mode here */ 3101254cf582SAl Viro error = lookup_fast(nd, &path, &inode, &seq); 3102e9742b53SAl Viro if (likely(error > 0)) 310371574865SMiklos Szeredi goto finish_lookup; 310471574865SMiklos Szeredi 3105ce57dfc1SAl Viro if (error < 0) 3106deb106c6SAl Viro return error; 3107a1eb3315SMiklos Szeredi 310837d7fffcSMiklos Szeredi BUG_ON(nd->inode != dir->d_inode); 31096583fe22SAl Viro BUG_ON(nd->flags & LOOKUP_RCU); 3110b6183df7SMiklos Szeredi } else { 3111fe2d35ffSAl Viro /* create side of things */ 3112a3fbbde7SAl Viro /* 3113b6183df7SMiklos Szeredi * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED 3114b6183df7SMiklos Szeredi * has been cleared when we got to the last component we are 3115b6183df7SMiklos Szeredi * about to look up 3116a3fbbde7SAl Viro */ 31179f1fafeeSAl Viro error = complete_walk(nd); 3118e8bb73dfSAl Viro if (error) 31192675a4ebSAl Viro return error; 3120fe2d35ffSAl Viro 312176ae2a5aSAl Viro audit_inode(nd->name, dir, LOOKUP_PARENT); 31221f36f774SAl Viro /* trailing slashes? */ 3123deb106c6SAl Viro if (unlikely(nd->last.name[nd->last.len])) 3124deb106c6SAl Viro return -EISDIR; 3125b6183df7SMiklos Szeredi } 31261f36f774SAl Viro 312716b1c1cdSMiklos Szeredi retry_lookup: 312864894cf8SAl Viro if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { 312964894cf8SAl Viro error = mnt_want_write(nd->path.mnt); 313064894cf8SAl Viro if (!error) 313164894cf8SAl Viro got_write = true; 313264894cf8SAl Viro /* 313364894cf8SAl Viro * do _not_ fail yet - we might not need that or fail with 313464894cf8SAl Viro * a different error; let lookup_open() decide; we'll be 313564894cf8SAl Viro * dropping this one anyway. 313664894cf8SAl Viro */ 313764894cf8SAl Viro } 31385955102cSAl Viro inode_lock(dir->d_inode); 3139896475d5SAl Viro error = lookup_open(nd, &path, file, op, got_write, opened); 31405955102cSAl Viro inode_unlock(dir->d_inode); 3141fb1cc555SAl Viro 31422675a4ebSAl Viro if (error <= 0) { 31432675a4ebSAl Viro if (error) 3144d58ffd35SMiklos Szeredi goto out; 31456c0d46c4SAl Viro 314647237687SAl Viro if ((*opened & FILE_CREATED) || 3147496ad9aaSAl Viro !S_ISREG(file_inode(file)->i_mode)) 314877d660a8SMiklos Szeredi will_truncate = false; 3149d18e9008SMiklos Szeredi 315076ae2a5aSAl Viro audit_inode(nd->name, file->f_path.dentry, 0); 3151d18e9008SMiklos Szeredi goto opened; 3152d18e9008SMiklos Szeredi } 3153d18e9008SMiklos Szeredi 315447237687SAl Viro if (*opened & FILE_CREATED) { 31559b44f1b3SAl Viro /* Don't check for write permission, don't truncate */ 3156ca344a89SAl Viro open_flag &= ~O_TRUNC; 315777d660a8SMiklos Szeredi will_truncate = false; 315862fb4a15SAl Viro acc_mode = 0; 3159896475d5SAl Viro path_to_nameidata(&path, nd); 3160e83db167SMiklos Szeredi goto finish_open_created; 3161fb1cc555SAl Viro } 3162fb1cc555SAl Viro 3163fb1cc555SAl Viro /* 3164d18e9008SMiklos Szeredi * If atomic_open() acquired write access it is dropped now due to 3165d18e9008SMiklos Szeredi * possible mount and symlink following (this might be optimized away if 3166d18e9008SMiklos Szeredi * necessary...) 3167d18e9008SMiklos Szeredi */ 316864894cf8SAl Viro if (got_write) { 3169d18e9008SMiklos Szeredi mnt_drop_write(nd->path.mnt); 317064894cf8SAl Viro got_write = false; 3171d18e9008SMiklos Szeredi } 3172d18e9008SMiklos Szeredi 31736583fe22SAl Viro if (unlikely(d_is_negative(path.dentry))) { 31746583fe22SAl Viro path_to_nameidata(&path, nd); 31756583fe22SAl Viro return -ENOENT; 31766583fe22SAl Viro } 31776583fe22SAl Viro 31786583fe22SAl Viro /* 31796583fe22SAl Viro * create/update audit record if it already exists. 31806583fe22SAl Viro */ 31816583fe22SAl Viro audit_inode(nd->name, path.dentry, 0); 31826583fe22SAl Viro 3183deb106c6SAl Viro if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) { 3184deb106c6SAl Viro path_to_nameidata(&path, nd); 3185deb106c6SAl Viro return -EEXIST; 3186deb106c6SAl Viro } 3187fb1cc555SAl Viro 3188756daf26SNeilBrown error = follow_managed(&path, nd); 3189deb106c6SAl Viro if (unlikely(error < 0)) 3190deb106c6SAl Viro return error; 3191a3fbbde7SAl Viro 3192254cf582SAl Viro seq = 0; /* out of RCU mode, so the value doesn't matter */ 3193d4565649SAl Viro inode = d_backing_inode(path.dentry); 3194766c4cbfSAl Viro finish_lookup: 3195191d7f73SAl Viro if (nd->depth) 3196191d7f73SAl Viro put_link(nd); 3197181548c0SAl Viro error = should_follow_link(nd, &path, nd->flags & LOOKUP_FOLLOW, 3198181548c0SAl Viro inode, seq); 3199deb106c6SAl Viro if (unlikely(error)) 3200d63ff28fSAl Viro return error; 3201fb1cc555SAl Viro 3202896475d5SAl Viro if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path.mnt) { 3203896475d5SAl Viro path_to_nameidata(&path, nd); 320416b1c1cdSMiklos Szeredi } else { 320516b1c1cdSMiklos Szeredi save_parent.dentry = nd->path.dentry; 3206896475d5SAl Viro save_parent.mnt = mntget(path.mnt); 3207896475d5SAl Viro nd->path.dentry = path.dentry; 320816b1c1cdSMiklos Szeredi 320916b1c1cdSMiklos Szeredi } 3210decf3400SMiklos Szeredi nd->inode = inode; 3211254cf582SAl Viro nd->seq = seq; 3212a3fbbde7SAl Viro /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ 3213bc77daa7SAl Viro finish_open: 3214a3fbbde7SAl Viro error = complete_walk(nd); 321516b1c1cdSMiklos Szeredi if (error) { 321616b1c1cdSMiklos Szeredi path_put(&save_parent); 32172675a4ebSAl Viro return error; 321816b1c1cdSMiklos Szeredi } 321976ae2a5aSAl Viro audit_inode(nd->name, nd->path.dentry, 0); 32205129fa48SAl Viro if (unlikely(d_is_symlink(nd->path.dentry)) && !(open_flag & O_PATH)) { 32215129fa48SAl Viro error = -ELOOP; 32225129fa48SAl Viro goto out; 32235129fa48SAl Viro } 3224fb1cc555SAl Viro error = -EISDIR; 322544b1d530SMiklos Szeredi if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry)) 32262675a4ebSAl Viro goto out; 3227af2f5542SMiklos Szeredi error = -ENOTDIR; 322844b1d530SMiklos Szeredi if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry)) 32292675a4ebSAl Viro goto out; 32304bbcbd3bSDavid Howells if (!d_is_reg(nd->path.dentry)) 323177d660a8SMiklos Szeredi will_truncate = false; 32326c0d46c4SAl Viro 32330f9d1a10SAl Viro if (will_truncate) { 32340f9d1a10SAl Viro error = mnt_want_write(nd->path.mnt); 32350f9d1a10SAl Viro if (error) 32362675a4ebSAl Viro goto out; 323764894cf8SAl Viro got_write = true; 32380f9d1a10SAl Viro } 3239e83db167SMiklos Szeredi finish_open_created: 324062fb4a15SAl Viro if (likely(!(open_flag & O_PATH))) { 3241bcda7652SAl Viro error = may_open(&nd->path, acc_mode, open_flag); 3242ca344a89SAl Viro if (error) 32432675a4ebSAl Viro goto out; 324462fb4a15SAl Viro } 32454aa7c634SMiklos Szeredi BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */ 32464aa7c634SMiklos Szeredi error = vfs_open(&nd->path, file, current_cred()); 32474aa7c634SMiklos Szeredi if (!error) { 32484aa7c634SMiklos Szeredi *opened |= FILE_OPENED; 32494aa7c634SMiklos Szeredi } else { 325030d90494SAl Viro if (error == -EOPENSTALE) 3251f60dc3dbSMiklos Szeredi goto stale_open; 3252015c3bbcSMiklos Szeredi goto out; 3253f60dc3dbSMiklos Szeredi } 3254a8277b9bSMiklos Szeredi opened: 32552675a4ebSAl Viro error = open_check_o_direct(file); 3256015c3bbcSMiklos Szeredi if (error) 3257015c3bbcSMiklos Szeredi goto exit_fput; 32583034a146SDmitry Kasatkin error = ima_file_check(file, op->acc_mode, *opened); 3259aa4caadbSMiklos Szeredi if (error) 3260aa4caadbSMiklos Szeredi goto exit_fput; 3261aa4caadbSMiklos Szeredi 32620f9d1a10SAl Viro if (will_truncate) { 32632675a4ebSAl Viro error = handle_truncate(file); 3264aa4caadbSMiklos Szeredi if (error) 3265aa4caadbSMiklos Szeredi goto exit_fput; 32660f9d1a10SAl Viro } 3267ca344a89SAl Viro out: 3268c80567c8SAl Viro if (unlikely(error > 0)) { 3269c80567c8SAl Viro WARN_ON(1); 3270c80567c8SAl Viro error = -EINVAL; 3271c80567c8SAl Viro } 327264894cf8SAl Viro if (got_write) 32730f9d1a10SAl Viro mnt_drop_write(nd->path.mnt); 327416b1c1cdSMiklos Szeredi path_put(&save_parent); 32752675a4ebSAl Viro return error; 3276fb1cc555SAl Viro 3277015c3bbcSMiklos Szeredi exit_fput: 32782675a4ebSAl Viro fput(file); 32792675a4ebSAl Viro goto out; 3280015c3bbcSMiklos Szeredi 3281f60dc3dbSMiklos Szeredi stale_open: 3282f60dc3dbSMiklos Szeredi /* If no saved parent or already retried then can't retry */ 3283f60dc3dbSMiklos Szeredi if (!save_parent.dentry || retried) 3284f60dc3dbSMiklos Szeredi goto out; 3285f60dc3dbSMiklos Szeredi 3286f60dc3dbSMiklos Szeredi BUG_ON(save_parent.dentry != dir); 3287f60dc3dbSMiklos Szeredi path_put(&nd->path); 3288f60dc3dbSMiklos Szeredi nd->path = save_parent; 3289f60dc3dbSMiklos Szeredi nd->inode = dir->d_inode; 3290f60dc3dbSMiklos Szeredi save_parent.mnt = NULL; 3291f60dc3dbSMiklos Szeredi save_parent.dentry = NULL; 329264894cf8SAl Viro if (got_write) { 3293f60dc3dbSMiklos Szeredi mnt_drop_write(nd->path.mnt); 329464894cf8SAl Viro got_write = false; 3295f60dc3dbSMiklos Szeredi } 3296f60dc3dbSMiklos Szeredi retried = true; 3297f60dc3dbSMiklos Szeredi goto retry_lookup; 3298fb1cc555SAl Viro } 3299fb1cc555SAl Viro 3300c8a53ee5SAl Viro static int do_tmpfile(struct nameidata *nd, unsigned flags, 330160545d0dSAl Viro const struct open_flags *op, 330260545d0dSAl Viro struct file *file, int *opened) 330360545d0dSAl Viro { 330460545d0dSAl Viro static const struct qstr name = QSTR_INIT("/", 1); 3305625b6d10SAl Viro struct dentry *child; 330660545d0dSAl Viro struct inode *dir; 3307625b6d10SAl Viro struct path path; 3308c8a53ee5SAl Viro int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path); 330960545d0dSAl Viro if (unlikely(error)) 331060545d0dSAl Viro return error; 3311625b6d10SAl Viro error = mnt_want_write(path.mnt); 331260545d0dSAl Viro if (unlikely(error)) 331360545d0dSAl Viro goto out; 3314625b6d10SAl Viro dir = path.dentry->d_inode; 331560545d0dSAl Viro /* we want directory to be writable */ 3316625b6d10SAl Viro error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 331760545d0dSAl Viro if (error) 331860545d0dSAl Viro goto out2; 331960545d0dSAl Viro if (!dir->i_op->tmpfile) { 332060545d0dSAl Viro error = -EOPNOTSUPP; 332160545d0dSAl Viro goto out2; 332260545d0dSAl Viro } 3323625b6d10SAl Viro child = d_alloc(path.dentry, &name); 332460545d0dSAl Viro if (unlikely(!child)) { 332560545d0dSAl Viro error = -ENOMEM; 332660545d0dSAl Viro goto out2; 332760545d0dSAl Viro } 3328625b6d10SAl Viro dput(path.dentry); 3329625b6d10SAl Viro path.dentry = child; 3330625b6d10SAl Viro error = dir->i_op->tmpfile(dir, child, op->mode); 333160545d0dSAl Viro if (error) 333260545d0dSAl Viro goto out2; 3333c8a53ee5SAl Viro audit_inode(nd->name, child, 0); 333469a91c23SEric Rannaud /* Don't check for other permissions, the inode was just created */ 333562fb4a15SAl Viro error = may_open(&path, 0, op->open_flag); 333660545d0dSAl Viro if (error) 333760545d0dSAl Viro goto out2; 3338625b6d10SAl Viro file->f_path.mnt = path.mnt; 3339625b6d10SAl Viro error = finish_open(file, child, NULL, opened); 334060545d0dSAl Viro if (error) 334160545d0dSAl Viro goto out2; 334260545d0dSAl Viro error = open_check_o_direct(file); 3343f4e0c30cSAl Viro if (error) { 334460545d0dSAl Viro fput(file); 3345f4e0c30cSAl Viro } else if (!(op->open_flag & O_EXCL)) { 3346f4e0c30cSAl Viro struct inode *inode = file_inode(file); 3347f4e0c30cSAl Viro spin_lock(&inode->i_lock); 3348f4e0c30cSAl Viro inode->i_state |= I_LINKABLE; 3349f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 3350f4e0c30cSAl Viro } 335160545d0dSAl Viro out2: 3352625b6d10SAl Viro mnt_drop_write(path.mnt); 335360545d0dSAl Viro out: 3354625b6d10SAl Viro path_put(&path); 335560545d0dSAl Viro return error; 335660545d0dSAl Viro } 335760545d0dSAl Viro 3358c8a53ee5SAl Viro static struct file *path_openat(struct nameidata *nd, 3359c8a53ee5SAl Viro const struct open_flags *op, unsigned flags) 33601da177e4SLinus Torvalds { 3361368ee9baSAl Viro const char *s; 336230d90494SAl Viro struct file *file; 336347237687SAl Viro int opened = 0; 336413aab428SAl Viro int error; 336531e6b01fSNick Piggin 336630d90494SAl Viro file = get_empty_filp(); 33671afc99beSAl Viro if (IS_ERR(file)) 33681afc99beSAl Viro return file; 336931e6b01fSNick Piggin 337030d90494SAl Viro file->f_flags = op->open_flag; 337131e6b01fSNick Piggin 3372bb458c64SAl Viro if (unlikely(file->f_flags & __O_TMPFILE)) { 3373c8a53ee5SAl Viro error = do_tmpfile(nd, flags, op, file, &opened); 3374f15133dfSAl Viro goto out2; 337560545d0dSAl Viro } 337660545d0dSAl Viro 3377c8a53ee5SAl Viro s = path_init(nd, flags); 3378368ee9baSAl Viro if (IS_ERR(s)) { 3379368ee9baSAl Viro put_filp(file); 3380368ee9baSAl Viro return ERR_CAST(s); 3381368ee9baSAl Viro } 33823bdba28bSAl Viro while (!(error = link_path_walk(s, nd)) && 338376ae2a5aSAl Viro (error = do_last(nd, file, op, &opened)) > 0) { 338473d049a4SAl Viro nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); 33853bdba28bSAl Viro s = trailing_symlink(nd); 33863bdba28bSAl Viro if (IS_ERR(s)) { 33873bdba28bSAl Viro error = PTR_ERR(s); 33882675a4ebSAl Viro break; 3389806b681cSAl Viro } 33903bdba28bSAl Viro } 3391deb106c6SAl Viro terminate_walk(nd); 3392f15133dfSAl Viro out2: 33932675a4ebSAl Viro if (!(opened & FILE_OPENED)) { 33942675a4ebSAl Viro BUG_ON(!error); 339530d90494SAl Viro put_filp(file); 3396015c3bbcSMiklos Szeredi } 33972675a4ebSAl Viro if (unlikely(error)) { 33982675a4ebSAl Viro if (error == -EOPENSTALE) { 33992675a4ebSAl Viro if (flags & LOOKUP_RCU) 34002675a4ebSAl Viro error = -ECHILD; 34012675a4ebSAl Viro else 34022675a4ebSAl Viro error = -ESTALE; 34032675a4ebSAl Viro } 34042675a4ebSAl Viro file = ERR_PTR(error); 34052675a4ebSAl Viro } 34062675a4ebSAl Viro return file; 3407de459215SKirill Korotaev } 34081da177e4SLinus Torvalds 3409669abf4eSJeff Layton struct file *do_filp_open(int dfd, struct filename *pathname, 3410f9652e10SAl Viro const struct open_flags *op) 341113aab428SAl Viro { 34129883d185SAl Viro struct nameidata nd; 3413f9652e10SAl Viro int flags = op->lookup_flags; 341413aab428SAl Viro struct file *filp; 341513aab428SAl Viro 34169883d185SAl Viro set_nameidata(&nd, dfd, pathname); 3417c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_RCU); 341813aab428SAl Viro if (unlikely(filp == ERR_PTR(-ECHILD))) 3419c8a53ee5SAl Viro filp = path_openat(&nd, op, flags); 342013aab428SAl Viro if (unlikely(filp == ERR_PTR(-ESTALE))) 3421c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_REVAL); 34229883d185SAl Viro restore_nameidata(); 342313aab428SAl Viro return filp; 342413aab428SAl Viro } 342513aab428SAl Viro 342673d049a4SAl Viro struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, 3427f9652e10SAl Viro const char *name, const struct open_flags *op) 342873d049a4SAl Viro { 34299883d185SAl Viro struct nameidata nd; 343073d049a4SAl Viro struct file *file; 343151689104SPaul Moore struct filename *filename; 3432f9652e10SAl Viro int flags = op->lookup_flags | LOOKUP_ROOT; 343373d049a4SAl Viro 343473d049a4SAl Viro nd.root.mnt = mnt; 343573d049a4SAl Viro nd.root.dentry = dentry; 343673d049a4SAl Viro 3437b18825a7SDavid Howells if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN) 343873d049a4SAl Viro return ERR_PTR(-ELOOP); 343973d049a4SAl Viro 344051689104SPaul Moore filename = getname_kernel(name); 3441a1c83681SViresh Kumar if (IS_ERR(filename)) 344251689104SPaul Moore return ERR_CAST(filename); 344351689104SPaul Moore 34449883d185SAl Viro set_nameidata(&nd, -1, filename); 3445c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_RCU); 344673d049a4SAl Viro if (unlikely(file == ERR_PTR(-ECHILD))) 3447c8a53ee5SAl Viro file = path_openat(&nd, op, flags); 344873d049a4SAl Viro if (unlikely(file == ERR_PTR(-ESTALE))) 3449c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_REVAL); 34509883d185SAl Viro restore_nameidata(); 345151689104SPaul Moore putname(filename); 345273d049a4SAl Viro return file; 345373d049a4SAl Viro } 345473d049a4SAl Viro 3455fa14a0b8SAl Viro static struct dentry *filename_create(int dfd, struct filename *name, 34561ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 34571da177e4SLinus Torvalds { 3458c663e5d8SChristoph Hellwig struct dentry *dentry = ERR_PTR(-EEXIST); 3459391172c4SAl Viro struct qstr last; 3460391172c4SAl Viro int type; 3461c30dabfeSJan Kara int err2; 34621ac12b4bSJeff Layton int error; 34631ac12b4bSJeff Layton bool is_dir = (lookup_flags & LOOKUP_DIRECTORY); 34641ac12b4bSJeff Layton 34651ac12b4bSJeff Layton /* 34661ac12b4bSJeff Layton * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any 34671ac12b4bSJeff Layton * other flags passed in are ignored! 34681ac12b4bSJeff Layton */ 34691ac12b4bSJeff Layton lookup_flags &= LOOKUP_REVAL; 34701ac12b4bSJeff Layton 34715c31b6ceSAl Viro name = filename_parentat(dfd, name, lookup_flags, path, &last, &type); 34725c31b6ceSAl Viro if (IS_ERR(name)) 34735c31b6ceSAl Viro return ERR_CAST(name); 34741da177e4SLinus Torvalds 3475c663e5d8SChristoph Hellwig /* 3476c663e5d8SChristoph Hellwig * Yucky last component or no last component at all? 3477c663e5d8SChristoph Hellwig * (foo/., foo/.., /////) 3478c663e5d8SChristoph Hellwig */ 34795c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) 3480ed75e95dSAl Viro goto out; 3481c663e5d8SChristoph Hellwig 3482c30dabfeSJan Kara /* don't fail immediately if it's r/o, at least try to report other errors */ 3483391172c4SAl Viro err2 = mnt_want_write(path->mnt); 3484c663e5d8SChristoph Hellwig /* 3485c663e5d8SChristoph Hellwig * Do the final lookup. 3486c663e5d8SChristoph Hellwig */ 3487391172c4SAl Viro lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL; 34885955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 3489391172c4SAl Viro dentry = __lookup_hash(&last, path->dentry, lookup_flags); 34901da177e4SLinus Torvalds if (IS_ERR(dentry)) 3491a8104a9fSAl Viro goto unlock; 3492c663e5d8SChristoph Hellwig 3493a8104a9fSAl Viro error = -EEXIST; 3494b18825a7SDavid Howells if (d_is_positive(dentry)) 3495a8104a9fSAl Viro goto fail; 3496b18825a7SDavid Howells 3497c663e5d8SChristoph Hellwig /* 3498c663e5d8SChristoph Hellwig * Special case - lookup gave negative, but... we had foo/bar/ 3499c663e5d8SChristoph Hellwig * From the vfs_mknod() POV we just have a negative dentry - 3500c663e5d8SChristoph Hellwig * all is fine. Let's be bastards - you had / on the end, you've 3501c663e5d8SChristoph Hellwig * been asking for (non-existent) directory. -ENOENT for you. 3502c663e5d8SChristoph Hellwig */ 3503391172c4SAl Viro if (unlikely(!is_dir && last.name[last.len])) { 3504a8104a9fSAl Viro error = -ENOENT; 3505ed75e95dSAl Viro goto fail; 3506e9baf6e5SAl Viro } 3507c30dabfeSJan Kara if (unlikely(err2)) { 3508c30dabfeSJan Kara error = err2; 3509a8104a9fSAl Viro goto fail; 3510c30dabfeSJan Kara } 3511181c37b6SAl Viro putname(name); 3512e9baf6e5SAl Viro return dentry; 35131da177e4SLinus Torvalds fail: 3514a8104a9fSAl Viro dput(dentry); 3515a8104a9fSAl Viro dentry = ERR_PTR(error); 3516a8104a9fSAl Viro unlock: 35175955102cSAl Viro inode_unlock(path->dentry->d_inode); 3518c30dabfeSJan Kara if (!err2) 3519391172c4SAl Viro mnt_drop_write(path->mnt); 3520ed75e95dSAl Viro out: 3521391172c4SAl Viro path_put(path); 3522181c37b6SAl Viro putname(name); 3523ed75e95dSAl Viro return dentry; 3524dae6ad8fSAl Viro } 3525fa14a0b8SAl Viro 3526fa14a0b8SAl Viro struct dentry *kern_path_create(int dfd, const char *pathname, 3527fa14a0b8SAl Viro struct path *path, unsigned int lookup_flags) 3528fa14a0b8SAl Viro { 3529181c37b6SAl Viro return filename_create(dfd, getname_kernel(pathname), 3530181c37b6SAl Viro path, lookup_flags); 3531fa14a0b8SAl Viro } 3532dae6ad8fSAl Viro EXPORT_SYMBOL(kern_path_create); 3533dae6ad8fSAl Viro 3534921a1650SAl Viro void done_path_create(struct path *path, struct dentry *dentry) 3535921a1650SAl Viro { 3536921a1650SAl Viro dput(dentry); 35375955102cSAl Viro inode_unlock(path->dentry->d_inode); 3538a8104a9fSAl Viro mnt_drop_write(path->mnt); 3539921a1650SAl Viro path_put(path); 3540921a1650SAl Viro } 3541921a1650SAl Viro EXPORT_SYMBOL(done_path_create); 3542921a1650SAl Viro 3543520ae687SAl Viro inline struct dentry *user_path_create(int dfd, const char __user *pathname, 35441ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 3545dae6ad8fSAl Viro { 3546181c37b6SAl Viro return filename_create(dfd, getname(pathname), path, lookup_flags); 3547dae6ad8fSAl Viro } 3548dae6ad8fSAl Viro EXPORT_SYMBOL(user_path_create); 3549dae6ad8fSAl Viro 35501a67aafbSAl Viro int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) 35511da177e4SLinus Torvalds { 3552a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 35531da177e4SLinus Torvalds 35541da177e4SLinus Torvalds if (error) 35551da177e4SLinus Torvalds return error; 35561da177e4SLinus Torvalds 3557975d6b39SEric W. Biederman if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) 35581da177e4SLinus Torvalds return -EPERM; 35591da177e4SLinus Torvalds 3560acfa4380SAl Viro if (!dir->i_op->mknod) 35611da177e4SLinus Torvalds return -EPERM; 35621da177e4SLinus Torvalds 356308ce5f16SSerge E. Hallyn error = devcgroup_inode_mknod(mode, dev); 356408ce5f16SSerge E. Hallyn if (error) 356508ce5f16SSerge E. Hallyn return error; 356608ce5f16SSerge E. Hallyn 35671da177e4SLinus Torvalds error = security_inode_mknod(dir, dentry, mode, dev); 35681da177e4SLinus Torvalds if (error) 35691da177e4SLinus Torvalds return error; 35701da177e4SLinus Torvalds 35711da177e4SLinus Torvalds error = dir->i_op->mknod(dir, dentry, mode, dev); 3572a74574aaSStephen Smalley if (!error) 3573f38aa942SAmy Griffis fsnotify_create(dir, dentry); 35741da177e4SLinus Torvalds return error; 35751da177e4SLinus Torvalds } 35764d359507SAl Viro EXPORT_SYMBOL(vfs_mknod); 35771da177e4SLinus Torvalds 3578f69aac00SAl Viro static int may_mknod(umode_t mode) 3579463c3197SDave Hansen { 3580463c3197SDave Hansen switch (mode & S_IFMT) { 3581463c3197SDave Hansen case S_IFREG: 3582463c3197SDave Hansen case S_IFCHR: 3583463c3197SDave Hansen case S_IFBLK: 3584463c3197SDave Hansen case S_IFIFO: 3585463c3197SDave Hansen case S_IFSOCK: 3586463c3197SDave Hansen case 0: /* zero mode translates to S_IFREG */ 3587463c3197SDave Hansen return 0; 3588463c3197SDave Hansen case S_IFDIR: 3589463c3197SDave Hansen return -EPERM; 3590463c3197SDave Hansen default: 3591463c3197SDave Hansen return -EINVAL; 3592463c3197SDave Hansen } 3593463c3197SDave Hansen } 3594463c3197SDave Hansen 35958208a22bSAl Viro SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, 35962e4d0924SHeiko Carstens unsigned, dev) 35971da177e4SLinus Torvalds { 35981da177e4SLinus Torvalds struct dentry *dentry; 3599dae6ad8fSAl Viro struct path path; 3600dae6ad8fSAl Viro int error; 3601972567f1SJeff Layton unsigned int lookup_flags = 0; 36021da177e4SLinus Torvalds 36038e4bfca1SAl Viro error = may_mknod(mode); 36048e4bfca1SAl Viro if (error) 36058e4bfca1SAl Viro return error; 3606972567f1SJeff Layton retry: 3607972567f1SJeff Layton dentry = user_path_create(dfd, filename, &path, lookup_flags); 3608dae6ad8fSAl Viro if (IS_ERR(dentry)) 3609dae6ad8fSAl Viro return PTR_ERR(dentry); 36102ad94ae6SAl Viro 3611dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3612ce3b0f8dSAl Viro mode &= ~current_umask(); 3613dae6ad8fSAl Viro error = security_path_mknod(&path, dentry, mode, dev); 3614be6d3e56SKentaro Takeda if (error) 3615a8104a9fSAl Viro goto out; 36161da177e4SLinus Torvalds switch (mode & S_IFMT) { 36171da177e4SLinus Torvalds case 0: case S_IFREG: 3618312b63fbSAl Viro error = vfs_create(path.dentry->d_inode,dentry,mode,true); 36191da177e4SLinus Torvalds break; 36201da177e4SLinus Torvalds case S_IFCHR: case S_IFBLK: 3621dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode, 36221da177e4SLinus Torvalds new_decode_dev(dev)); 36231da177e4SLinus Torvalds break; 36241da177e4SLinus Torvalds case S_IFIFO: case S_IFSOCK: 3625dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); 36261da177e4SLinus Torvalds break; 36271da177e4SLinus Torvalds } 3628a8104a9fSAl Viro out: 3629921a1650SAl Viro done_path_create(&path, dentry); 3630972567f1SJeff Layton if (retry_estale(error, lookup_flags)) { 3631972567f1SJeff Layton lookup_flags |= LOOKUP_REVAL; 3632972567f1SJeff Layton goto retry; 3633972567f1SJeff Layton } 36341da177e4SLinus Torvalds return error; 36351da177e4SLinus Torvalds } 36361da177e4SLinus Torvalds 36378208a22bSAl Viro SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) 36385590ff0dSUlrich Drepper { 36395590ff0dSUlrich Drepper return sys_mknodat(AT_FDCWD, filename, mode, dev); 36405590ff0dSUlrich Drepper } 36415590ff0dSUlrich Drepper 364218bb1db3SAl Viro int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 36431da177e4SLinus Torvalds { 3644a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 36458de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 36461da177e4SLinus Torvalds 36471da177e4SLinus Torvalds if (error) 36481da177e4SLinus Torvalds return error; 36491da177e4SLinus Torvalds 3650acfa4380SAl Viro if (!dir->i_op->mkdir) 36511da177e4SLinus Torvalds return -EPERM; 36521da177e4SLinus Torvalds 36531da177e4SLinus Torvalds mode &= (S_IRWXUGO|S_ISVTX); 36541da177e4SLinus Torvalds error = security_inode_mkdir(dir, dentry, mode); 36551da177e4SLinus Torvalds if (error) 36561da177e4SLinus Torvalds return error; 36571da177e4SLinus Torvalds 36588de52778SAl Viro if (max_links && dir->i_nlink >= max_links) 36598de52778SAl Viro return -EMLINK; 36608de52778SAl Viro 36611da177e4SLinus Torvalds error = dir->i_op->mkdir(dir, dentry, mode); 3662a74574aaSStephen Smalley if (!error) 3663f38aa942SAmy Griffis fsnotify_mkdir(dir, dentry); 36641da177e4SLinus Torvalds return error; 36651da177e4SLinus Torvalds } 36664d359507SAl Viro EXPORT_SYMBOL(vfs_mkdir); 36671da177e4SLinus Torvalds 3668a218d0fdSAl Viro SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) 36691da177e4SLinus Torvalds { 36706902d925SDave Hansen struct dentry *dentry; 3671dae6ad8fSAl Viro struct path path; 3672dae6ad8fSAl Viro int error; 3673b76d8b82SJeff Layton unsigned int lookup_flags = LOOKUP_DIRECTORY; 36741da177e4SLinus Torvalds 3675b76d8b82SJeff Layton retry: 3676b76d8b82SJeff Layton dentry = user_path_create(dfd, pathname, &path, lookup_flags); 36776902d925SDave Hansen if (IS_ERR(dentry)) 3678dae6ad8fSAl Viro return PTR_ERR(dentry); 36796902d925SDave Hansen 3680dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3681ce3b0f8dSAl Viro mode &= ~current_umask(); 3682dae6ad8fSAl Viro error = security_path_mkdir(&path, dentry, mode); 3683a8104a9fSAl Viro if (!error) 3684dae6ad8fSAl Viro error = vfs_mkdir(path.dentry->d_inode, dentry, mode); 3685921a1650SAl Viro done_path_create(&path, dentry); 3686b76d8b82SJeff Layton if (retry_estale(error, lookup_flags)) { 3687b76d8b82SJeff Layton lookup_flags |= LOOKUP_REVAL; 3688b76d8b82SJeff Layton goto retry; 3689b76d8b82SJeff Layton } 36901da177e4SLinus Torvalds return error; 36911da177e4SLinus Torvalds } 36921da177e4SLinus Torvalds 3693a218d0fdSAl Viro SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) 36945590ff0dSUlrich Drepper { 36955590ff0dSUlrich Drepper return sys_mkdirat(AT_FDCWD, pathname, mode); 36965590ff0dSUlrich Drepper } 36975590ff0dSUlrich Drepper 36981da177e4SLinus Torvalds /* 3699a71905f0SSage Weil * The dentry_unhash() helper will try to drop the dentry early: we 3700c0d02594SJ. Bruce Fields * should have a usage count of 1 if we're the only user of this 3701a71905f0SSage Weil * dentry, and if that is true (possibly after pruning the dcache), 3702a71905f0SSage Weil * then we drop the dentry now. 37031da177e4SLinus Torvalds * 37041da177e4SLinus Torvalds * A low-level filesystem can, if it choses, legally 37051da177e4SLinus Torvalds * do a 37061da177e4SLinus Torvalds * 37071da177e4SLinus Torvalds * if (!d_unhashed(dentry)) 37081da177e4SLinus Torvalds * return -EBUSY; 37091da177e4SLinus Torvalds * 37101da177e4SLinus Torvalds * if it cannot handle the case of removing a directory 37111da177e4SLinus Torvalds * that is still in use by something else.. 37121da177e4SLinus Torvalds */ 37131da177e4SLinus Torvalds void dentry_unhash(struct dentry *dentry) 37141da177e4SLinus Torvalds { 37151da177e4SLinus Torvalds shrink_dcache_parent(dentry); 37161da177e4SLinus Torvalds spin_lock(&dentry->d_lock); 371798474236SWaiman Long if (dentry->d_lockref.count == 1) 37181da177e4SLinus Torvalds __d_drop(dentry); 37191da177e4SLinus Torvalds spin_unlock(&dentry->d_lock); 37201da177e4SLinus Torvalds } 37214d359507SAl Viro EXPORT_SYMBOL(dentry_unhash); 37221da177e4SLinus Torvalds 37231da177e4SLinus Torvalds int vfs_rmdir(struct inode *dir, struct dentry *dentry) 37241da177e4SLinus Torvalds { 37251da177e4SLinus Torvalds int error = may_delete(dir, dentry, 1); 37261da177e4SLinus Torvalds 37271da177e4SLinus Torvalds if (error) 37281da177e4SLinus Torvalds return error; 37291da177e4SLinus Torvalds 3730acfa4380SAl Viro if (!dir->i_op->rmdir) 37311da177e4SLinus Torvalds return -EPERM; 37321da177e4SLinus Torvalds 37331d2ef590SAl Viro dget(dentry); 37345955102cSAl Viro inode_lock(dentry->d_inode); 3735912dbc15SSage Weil 37361da177e4SLinus Torvalds error = -EBUSY; 37377af1364fSEric W. Biederman if (is_local_mountpoint(dentry)) 3738912dbc15SSage Weil goto out; 3739912dbc15SSage Weil 37401da177e4SLinus Torvalds error = security_inode_rmdir(dir, dentry); 3741912dbc15SSage Weil if (error) 3742912dbc15SSage Weil goto out; 3743912dbc15SSage Weil 37443cebde24SSage Weil shrink_dcache_parent(dentry); 37451da177e4SLinus Torvalds error = dir->i_op->rmdir(dir, dentry); 3746912dbc15SSage Weil if (error) 3747912dbc15SSage Weil goto out; 3748912dbc15SSage Weil 37491da177e4SLinus Torvalds dentry->d_inode->i_flags |= S_DEAD; 3750d83c49f3SAl Viro dont_mount(dentry); 37518ed936b5SEric W. Biederman detach_mounts(dentry); 37521da177e4SLinus Torvalds 3753912dbc15SSage Weil out: 37545955102cSAl Viro inode_unlock(dentry->d_inode); 37551d2ef590SAl Viro dput(dentry); 3756912dbc15SSage Weil if (!error) 3757912dbc15SSage Weil d_delete(dentry); 37581da177e4SLinus Torvalds return error; 37591da177e4SLinus Torvalds } 37604d359507SAl Viro EXPORT_SYMBOL(vfs_rmdir); 37611da177e4SLinus Torvalds 37625590ff0dSUlrich Drepper static long do_rmdir(int dfd, const char __user *pathname) 37631da177e4SLinus Torvalds { 37641da177e4SLinus Torvalds int error = 0; 376591a27b2aSJeff Layton struct filename *name; 37661da177e4SLinus Torvalds struct dentry *dentry; 3767f5beed75SAl Viro struct path path; 3768f5beed75SAl Viro struct qstr last; 3769f5beed75SAl Viro int type; 3770c6ee9206SJeff Layton unsigned int lookup_flags = 0; 3771c6ee9206SJeff Layton retry: 3772f5beed75SAl Viro name = user_path_parent(dfd, pathname, 3773f5beed75SAl Viro &path, &last, &type, lookup_flags); 377491a27b2aSJeff Layton if (IS_ERR(name)) 377591a27b2aSJeff Layton return PTR_ERR(name); 37761da177e4SLinus Torvalds 3777f5beed75SAl Viro switch (type) { 37781da177e4SLinus Torvalds case LAST_DOTDOT: 37791da177e4SLinus Torvalds error = -ENOTEMPTY; 37801da177e4SLinus Torvalds goto exit1; 37811da177e4SLinus Torvalds case LAST_DOT: 37821da177e4SLinus Torvalds error = -EINVAL; 37831da177e4SLinus Torvalds goto exit1; 37841da177e4SLinus Torvalds case LAST_ROOT: 37851da177e4SLinus Torvalds error = -EBUSY; 37861da177e4SLinus Torvalds goto exit1; 37871da177e4SLinus Torvalds } 37880612d9fbSOGAWA Hirofumi 3789f5beed75SAl Viro error = mnt_want_write(path.mnt); 3790c30dabfeSJan Kara if (error) 3791c30dabfeSJan Kara goto exit1; 37920612d9fbSOGAWA Hirofumi 37935955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 3794f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 37951da177e4SLinus Torvalds error = PTR_ERR(dentry); 37966902d925SDave Hansen if (IS_ERR(dentry)) 37976902d925SDave Hansen goto exit2; 3798e6bc45d6STheodore Ts'o if (!dentry->d_inode) { 3799e6bc45d6STheodore Ts'o error = -ENOENT; 3800e6bc45d6STheodore Ts'o goto exit3; 3801e6bc45d6STheodore Ts'o } 3802f5beed75SAl Viro error = security_path_rmdir(&path, dentry); 3803be6d3e56SKentaro Takeda if (error) 3804c30dabfeSJan Kara goto exit3; 3805f5beed75SAl Viro error = vfs_rmdir(path.dentry->d_inode, dentry); 38060622753bSDave Hansen exit3: 38071da177e4SLinus Torvalds dput(dentry); 38086902d925SDave Hansen exit2: 38095955102cSAl Viro inode_unlock(path.dentry->d_inode); 3810f5beed75SAl Viro mnt_drop_write(path.mnt); 38111da177e4SLinus Torvalds exit1: 3812f5beed75SAl Viro path_put(&path); 38131da177e4SLinus Torvalds putname(name); 3814c6ee9206SJeff Layton if (retry_estale(error, lookup_flags)) { 3815c6ee9206SJeff Layton lookup_flags |= LOOKUP_REVAL; 3816c6ee9206SJeff Layton goto retry; 3817c6ee9206SJeff Layton } 38181da177e4SLinus Torvalds return error; 38191da177e4SLinus Torvalds } 38201da177e4SLinus Torvalds 38213cdad428SHeiko Carstens SYSCALL_DEFINE1(rmdir, const char __user *, pathname) 38225590ff0dSUlrich Drepper { 38235590ff0dSUlrich Drepper return do_rmdir(AT_FDCWD, pathname); 38245590ff0dSUlrich Drepper } 38255590ff0dSUlrich Drepper 3826b21996e3SJ. Bruce Fields /** 3827b21996e3SJ. Bruce Fields * vfs_unlink - unlink a filesystem object 3828b21996e3SJ. Bruce Fields * @dir: parent directory 3829b21996e3SJ. Bruce Fields * @dentry: victim 3830b21996e3SJ. Bruce Fields * @delegated_inode: returns victim inode, if the inode is delegated. 3831b21996e3SJ. Bruce Fields * 3832b21996e3SJ. Bruce Fields * The caller must hold dir->i_mutex. 3833b21996e3SJ. Bruce Fields * 3834b21996e3SJ. Bruce Fields * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and 3835b21996e3SJ. Bruce Fields * return a reference to the inode in delegated_inode. The caller 3836b21996e3SJ. Bruce Fields * should then break the delegation on that inode and retry. Because 3837b21996e3SJ. Bruce Fields * breaking a delegation may take a long time, the caller should drop 3838b21996e3SJ. Bruce Fields * dir->i_mutex before doing so. 3839b21996e3SJ. Bruce Fields * 3840b21996e3SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 3841b21996e3SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 3842b21996e3SJ. Bruce Fields * to be NFS exported. 3843b21996e3SJ. Bruce Fields */ 3844b21996e3SJ. Bruce Fields int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode) 38451da177e4SLinus Torvalds { 38469accbb97SJ. Bruce Fields struct inode *target = dentry->d_inode; 38471da177e4SLinus Torvalds int error = may_delete(dir, dentry, 0); 38481da177e4SLinus Torvalds 38491da177e4SLinus Torvalds if (error) 38501da177e4SLinus Torvalds return error; 38511da177e4SLinus Torvalds 3852acfa4380SAl Viro if (!dir->i_op->unlink) 38531da177e4SLinus Torvalds return -EPERM; 38541da177e4SLinus Torvalds 38555955102cSAl Viro inode_lock(target); 38568ed936b5SEric W. Biederman if (is_local_mountpoint(dentry)) 38571da177e4SLinus Torvalds error = -EBUSY; 38581da177e4SLinus Torvalds else { 38591da177e4SLinus Torvalds error = security_inode_unlink(dir, dentry); 3860bec1052eSAl Viro if (!error) { 38615a14696cSJ. Bruce Fields error = try_break_deleg(target, delegated_inode); 38625a14696cSJ. Bruce Fields if (error) 3863b21996e3SJ. Bruce Fields goto out; 38641da177e4SLinus Torvalds error = dir->i_op->unlink(dir, dentry); 38658ed936b5SEric W. Biederman if (!error) { 3866d83c49f3SAl Viro dont_mount(dentry); 38678ed936b5SEric W. Biederman detach_mounts(dentry); 38688ed936b5SEric W. Biederman } 3869bec1052eSAl Viro } 38701da177e4SLinus Torvalds } 3871b21996e3SJ. Bruce Fields out: 38725955102cSAl Viro inode_unlock(target); 38731da177e4SLinus Torvalds 38741da177e4SLinus Torvalds /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 38751da177e4SLinus Torvalds if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { 38769accbb97SJ. Bruce Fields fsnotify_link_count(target); 38771da177e4SLinus Torvalds d_delete(dentry); 38781da177e4SLinus Torvalds } 38790eeca283SRobert Love 38801da177e4SLinus Torvalds return error; 38811da177e4SLinus Torvalds } 38824d359507SAl Viro EXPORT_SYMBOL(vfs_unlink); 38831da177e4SLinus Torvalds 38841da177e4SLinus Torvalds /* 38851da177e4SLinus Torvalds * Make sure that the actual truncation of the file will occur outside its 38861b1dcc1bSJes Sorensen * directory's i_mutex. Truncate can take a long time if there is a lot of 38871da177e4SLinus Torvalds * writeout happening, and we don't want to prevent access to the directory 38881da177e4SLinus Torvalds * while waiting on the I/O. 38891da177e4SLinus Torvalds */ 38905590ff0dSUlrich Drepper static long do_unlinkat(int dfd, const char __user *pathname) 38911da177e4SLinus Torvalds { 38922ad94ae6SAl Viro int error; 389391a27b2aSJeff Layton struct filename *name; 38941da177e4SLinus Torvalds struct dentry *dentry; 3895f5beed75SAl Viro struct path path; 3896f5beed75SAl Viro struct qstr last; 3897f5beed75SAl Viro int type; 38981da177e4SLinus Torvalds struct inode *inode = NULL; 3899b21996e3SJ. Bruce Fields struct inode *delegated_inode = NULL; 39005d18f813SJeff Layton unsigned int lookup_flags = 0; 39015d18f813SJeff Layton retry: 3902f5beed75SAl Viro name = user_path_parent(dfd, pathname, 3903f5beed75SAl Viro &path, &last, &type, lookup_flags); 390491a27b2aSJeff Layton if (IS_ERR(name)) 390591a27b2aSJeff Layton return PTR_ERR(name); 39062ad94ae6SAl Viro 39071da177e4SLinus Torvalds error = -EISDIR; 3908f5beed75SAl Viro if (type != LAST_NORM) 39091da177e4SLinus Torvalds goto exit1; 39100612d9fbSOGAWA Hirofumi 3911f5beed75SAl Viro error = mnt_want_write(path.mnt); 3912c30dabfeSJan Kara if (error) 3913c30dabfeSJan Kara goto exit1; 3914b21996e3SJ. Bruce Fields retry_deleg: 39155955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 3916f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 39171da177e4SLinus Torvalds error = PTR_ERR(dentry); 39181da177e4SLinus Torvalds if (!IS_ERR(dentry)) { 39191da177e4SLinus Torvalds /* Why not before? Because we want correct error value */ 3920f5beed75SAl Viro if (last.name[last.len]) 392150338b88STörök Edwin goto slashes; 39221da177e4SLinus Torvalds inode = dentry->d_inode; 3923b18825a7SDavid Howells if (d_is_negative(dentry)) 3924e6bc45d6STheodore Ts'o goto slashes; 39257de9c6eeSAl Viro ihold(inode); 3926f5beed75SAl Viro error = security_path_unlink(&path, dentry); 3927be6d3e56SKentaro Takeda if (error) 3928c30dabfeSJan Kara goto exit2; 3929f5beed75SAl Viro error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode); 39301da177e4SLinus Torvalds exit2: 39311da177e4SLinus Torvalds dput(dentry); 39321da177e4SLinus Torvalds } 39335955102cSAl Viro inode_unlock(path.dentry->d_inode); 39341da177e4SLinus Torvalds if (inode) 39351da177e4SLinus Torvalds iput(inode); /* truncate the inode here */ 3936b21996e3SJ. Bruce Fields inode = NULL; 3937b21996e3SJ. Bruce Fields if (delegated_inode) { 39385a14696cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 3939b21996e3SJ. Bruce Fields if (!error) 3940b21996e3SJ. Bruce Fields goto retry_deleg; 3941b21996e3SJ. Bruce Fields } 3942f5beed75SAl Viro mnt_drop_write(path.mnt); 39431da177e4SLinus Torvalds exit1: 3944f5beed75SAl Viro path_put(&path); 39451da177e4SLinus Torvalds putname(name); 39465d18f813SJeff Layton if (retry_estale(error, lookup_flags)) { 39475d18f813SJeff Layton lookup_flags |= LOOKUP_REVAL; 39485d18f813SJeff Layton inode = NULL; 39495d18f813SJeff Layton goto retry; 39505d18f813SJeff Layton } 39511da177e4SLinus Torvalds return error; 39521da177e4SLinus Torvalds 39531da177e4SLinus Torvalds slashes: 3954b18825a7SDavid Howells if (d_is_negative(dentry)) 3955b18825a7SDavid Howells error = -ENOENT; 395644b1d530SMiklos Szeredi else if (d_is_dir(dentry)) 3957b18825a7SDavid Howells error = -EISDIR; 3958b18825a7SDavid Howells else 3959b18825a7SDavid Howells error = -ENOTDIR; 39601da177e4SLinus Torvalds goto exit2; 39611da177e4SLinus Torvalds } 39621da177e4SLinus Torvalds 39632e4d0924SHeiko Carstens SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) 39645590ff0dSUlrich Drepper { 39655590ff0dSUlrich Drepper if ((flag & ~AT_REMOVEDIR) != 0) 39665590ff0dSUlrich Drepper return -EINVAL; 39675590ff0dSUlrich Drepper 39685590ff0dSUlrich Drepper if (flag & AT_REMOVEDIR) 39695590ff0dSUlrich Drepper return do_rmdir(dfd, pathname); 39705590ff0dSUlrich Drepper 39715590ff0dSUlrich Drepper return do_unlinkat(dfd, pathname); 39725590ff0dSUlrich Drepper } 39735590ff0dSUlrich Drepper 39743480b257SHeiko Carstens SYSCALL_DEFINE1(unlink, const char __user *, pathname) 39755590ff0dSUlrich Drepper { 39765590ff0dSUlrich Drepper return do_unlinkat(AT_FDCWD, pathname); 39775590ff0dSUlrich Drepper } 39785590ff0dSUlrich Drepper 3979db2e747bSMiklos Szeredi int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) 39801da177e4SLinus Torvalds { 3981a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 39821da177e4SLinus Torvalds 39831da177e4SLinus Torvalds if (error) 39841da177e4SLinus Torvalds return error; 39851da177e4SLinus Torvalds 3986acfa4380SAl Viro if (!dir->i_op->symlink) 39871da177e4SLinus Torvalds return -EPERM; 39881da177e4SLinus Torvalds 39891da177e4SLinus Torvalds error = security_inode_symlink(dir, dentry, oldname); 39901da177e4SLinus Torvalds if (error) 39911da177e4SLinus Torvalds return error; 39921da177e4SLinus Torvalds 39931da177e4SLinus Torvalds error = dir->i_op->symlink(dir, dentry, oldname); 3994a74574aaSStephen Smalley if (!error) 3995f38aa942SAmy Griffis fsnotify_create(dir, dentry); 39961da177e4SLinus Torvalds return error; 39971da177e4SLinus Torvalds } 39984d359507SAl Viro EXPORT_SYMBOL(vfs_symlink); 39991da177e4SLinus Torvalds 40002e4d0924SHeiko Carstens SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, 40012e4d0924SHeiko Carstens int, newdfd, const char __user *, newname) 40021da177e4SLinus Torvalds { 40032ad94ae6SAl Viro int error; 400491a27b2aSJeff Layton struct filename *from; 40056902d925SDave Hansen struct dentry *dentry; 4006dae6ad8fSAl Viro struct path path; 4007f46d3567SJeff Layton unsigned int lookup_flags = 0; 40081da177e4SLinus Torvalds 40091da177e4SLinus Torvalds from = getname(oldname); 40101da177e4SLinus Torvalds if (IS_ERR(from)) 40111da177e4SLinus Torvalds return PTR_ERR(from); 4012f46d3567SJeff Layton retry: 4013f46d3567SJeff Layton dentry = user_path_create(newdfd, newname, &path, lookup_flags); 40141da177e4SLinus Torvalds error = PTR_ERR(dentry); 40156902d925SDave Hansen if (IS_ERR(dentry)) 4016dae6ad8fSAl Viro goto out_putname; 40176902d925SDave Hansen 401891a27b2aSJeff Layton error = security_path_symlink(&path, dentry, from->name); 4019a8104a9fSAl Viro if (!error) 402091a27b2aSJeff Layton error = vfs_symlink(path.dentry->d_inode, dentry, from->name); 4021921a1650SAl Viro done_path_create(&path, dentry); 4022f46d3567SJeff Layton if (retry_estale(error, lookup_flags)) { 4023f46d3567SJeff Layton lookup_flags |= LOOKUP_REVAL; 4024f46d3567SJeff Layton goto retry; 4025f46d3567SJeff Layton } 40266902d925SDave Hansen out_putname: 40271da177e4SLinus Torvalds putname(from); 40281da177e4SLinus Torvalds return error; 40291da177e4SLinus Torvalds } 40301da177e4SLinus Torvalds 40313480b257SHeiko Carstens SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) 40325590ff0dSUlrich Drepper { 40335590ff0dSUlrich Drepper return sys_symlinkat(oldname, AT_FDCWD, newname); 40345590ff0dSUlrich Drepper } 40355590ff0dSUlrich Drepper 4036146a8595SJ. Bruce Fields /** 4037146a8595SJ. Bruce Fields * vfs_link - create a new link 4038146a8595SJ. Bruce Fields * @old_dentry: object to be linked 4039146a8595SJ. Bruce Fields * @dir: new parent 4040146a8595SJ. Bruce Fields * @new_dentry: where to create the new link 4041146a8595SJ. Bruce Fields * @delegated_inode: returns inode needing a delegation break 4042146a8595SJ. Bruce Fields * 4043146a8595SJ. Bruce Fields * The caller must hold dir->i_mutex 4044146a8595SJ. Bruce Fields * 4045146a8595SJ. Bruce Fields * If vfs_link discovers a delegation on the to-be-linked file in need 4046146a8595SJ. Bruce Fields * of breaking, it will return -EWOULDBLOCK and return a reference to the 4047146a8595SJ. Bruce Fields * inode in delegated_inode. The caller should then break the delegation 4048146a8595SJ. Bruce Fields * and retry. Because breaking a delegation may take a long time, the 4049146a8595SJ. Bruce Fields * caller should drop the i_mutex before doing so. 4050146a8595SJ. Bruce Fields * 4051146a8595SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 4052146a8595SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 4053146a8595SJ. Bruce Fields * to be NFS exported. 4054146a8595SJ. Bruce Fields */ 4055146a8595SJ. Bruce Fields int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode) 40561da177e4SLinus Torvalds { 40571da177e4SLinus Torvalds struct inode *inode = old_dentry->d_inode; 40588de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 40591da177e4SLinus Torvalds int error; 40601da177e4SLinus Torvalds 40611da177e4SLinus Torvalds if (!inode) 40621da177e4SLinus Torvalds return -ENOENT; 40631da177e4SLinus Torvalds 4064a95164d9SMiklos Szeredi error = may_create(dir, new_dentry); 40651da177e4SLinus Torvalds if (error) 40661da177e4SLinus Torvalds return error; 40671da177e4SLinus Torvalds 40681da177e4SLinus Torvalds if (dir->i_sb != inode->i_sb) 40691da177e4SLinus Torvalds return -EXDEV; 40701da177e4SLinus Torvalds 40711da177e4SLinus Torvalds /* 40721da177e4SLinus Torvalds * A link to an append-only or immutable file cannot be created. 40731da177e4SLinus Torvalds */ 40741da177e4SLinus Torvalds if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 40751da177e4SLinus Torvalds return -EPERM; 4076acfa4380SAl Viro if (!dir->i_op->link) 40771da177e4SLinus Torvalds return -EPERM; 40787e79eedbSTetsuo Handa if (S_ISDIR(inode->i_mode)) 40791da177e4SLinus Torvalds return -EPERM; 40801da177e4SLinus Torvalds 40811da177e4SLinus Torvalds error = security_inode_link(old_dentry, dir, new_dentry); 40821da177e4SLinus Torvalds if (error) 40831da177e4SLinus Torvalds return error; 40841da177e4SLinus Torvalds 40855955102cSAl Viro inode_lock(inode); 4086aae8a97dSAneesh Kumar K.V /* Make sure we don't allow creating hardlink to an unlinked file */ 4087f4e0c30cSAl Viro if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) 4088aae8a97dSAneesh Kumar K.V error = -ENOENT; 40898de52778SAl Viro else if (max_links && inode->i_nlink >= max_links) 40908de52778SAl Viro error = -EMLINK; 4091146a8595SJ. Bruce Fields else { 4092146a8595SJ. Bruce Fields error = try_break_deleg(inode, delegated_inode); 4093146a8595SJ. Bruce Fields if (!error) 40941da177e4SLinus Torvalds error = dir->i_op->link(old_dentry, dir, new_dentry); 4095146a8595SJ. Bruce Fields } 4096f4e0c30cSAl Viro 4097f4e0c30cSAl Viro if (!error && (inode->i_state & I_LINKABLE)) { 4098f4e0c30cSAl Viro spin_lock(&inode->i_lock); 4099f4e0c30cSAl Viro inode->i_state &= ~I_LINKABLE; 4100f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 4101f4e0c30cSAl Viro } 41025955102cSAl Viro inode_unlock(inode); 4103e31e14ecSStephen Smalley if (!error) 41047e79eedbSTetsuo Handa fsnotify_link(dir, inode, new_dentry); 41051da177e4SLinus Torvalds return error; 41061da177e4SLinus Torvalds } 41074d359507SAl Viro EXPORT_SYMBOL(vfs_link); 41081da177e4SLinus Torvalds 41091da177e4SLinus Torvalds /* 41101da177e4SLinus Torvalds * Hardlinks are often used in delicate situations. We avoid 41111da177e4SLinus Torvalds * security-related surprises by not following symlinks on the 41121da177e4SLinus Torvalds * newname. --KAB 41131da177e4SLinus Torvalds * 41141da177e4SLinus Torvalds * We don't follow them on the oldname either to be compatible 41151da177e4SLinus Torvalds * with linux 2.0, and to avoid hard-linking to directories 41161da177e4SLinus Torvalds * and other special files. --ADM 41171da177e4SLinus Torvalds */ 41182e4d0924SHeiko Carstens SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, 41192e4d0924SHeiko Carstens int, newdfd, const char __user *, newname, int, flags) 41201da177e4SLinus Torvalds { 41211da177e4SLinus Torvalds struct dentry *new_dentry; 4122dae6ad8fSAl Viro struct path old_path, new_path; 4123146a8595SJ. Bruce Fields struct inode *delegated_inode = NULL; 412411a7b371SAneesh Kumar K.V int how = 0; 41251da177e4SLinus Torvalds int error; 41261da177e4SLinus Torvalds 412711a7b371SAneesh Kumar K.V if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) 4128c04030e1SUlrich Drepper return -EINVAL; 412911a7b371SAneesh Kumar K.V /* 4130f0cc6ffbSLinus Torvalds * To use null names we require CAP_DAC_READ_SEARCH 4131f0cc6ffbSLinus Torvalds * This ensures that not everyone will be able to create 4132f0cc6ffbSLinus Torvalds * handlink using the passed filedescriptor. 413311a7b371SAneesh Kumar K.V */ 4134f0cc6ffbSLinus Torvalds if (flags & AT_EMPTY_PATH) { 4135f0cc6ffbSLinus Torvalds if (!capable(CAP_DAC_READ_SEARCH)) 4136f0cc6ffbSLinus Torvalds return -ENOENT; 413711a7b371SAneesh Kumar K.V how = LOOKUP_EMPTY; 4138f0cc6ffbSLinus Torvalds } 4139c04030e1SUlrich Drepper 414011a7b371SAneesh Kumar K.V if (flags & AT_SYMLINK_FOLLOW) 414111a7b371SAneesh Kumar K.V how |= LOOKUP_FOLLOW; 4142442e31caSJeff Layton retry: 414311a7b371SAneesh Kumar K.V error = user_path_at(olddfd, oldname, how, &old_path); 41441da177e4SLinus Torvalds if (error) 41452ad94ae6SAl Viro return error; 41462ad94ae6SAl Viro 4147442e31caSJeff Layton new_dentry = user_path_create(newdfd, newname, &new_path, 4148442e31caSJeff Layton (how & LOOKUP_REVAL)); 41491da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 41506902d925SDave Hansen if (IS_ERR(new_dentry)) 4151dae6ad8fSAl Viro goto out; 4152dae6ad8fSAl Viro 4153dae6ad8fSAl Viro error = -EXDEV; 4154dae6ad8fSAl Viro if (old_path.mnt != new_path.mnt) 4155dae6ad8fSAl Viro goto out_dput; 4156800179c9SKees Cook error = may_linkat(&old_path); 4157800179c9SKees Cook if (unlikely(error)) 4158800179c9SKees Cook goto out_dput; 4159dae6ad8fSAl Viro error = security_path_link(old_path.dentry, &new_path, new_dentry); 4160be6d3e56SKentaro Takeda if (error) 4161a8104a9fSAl Viro goto out_dput; 4162146a8595SJ. Bruce Fields error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode); 416375c3f29dSDave Hansen out_dput: 4164921a1650SAl Viro done_path_create(&new_path, new_dentry); 4165146a8595SJ. Bruce Fields if (delegated_inode) { 4166146a8595SJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4167d22e6338SOleg Drokin if (!error) { 4168d22e6338SOleg Drokin path_put(&old_path); 4169146a8595SJ. Bruce Fields goto retry; 4170146a8595SJ. Bruce Fields } 4171d22e6338SOleg Drokin } 4172442e31caSJeff Layton if (retry_estale(error, how)) { 4173d22e6338SOleg Drokin path_put(&old_path); 4174442e31caSJeff Layton how |= LOOKUP_REVAL; 4175442e31caSJeff Layton goto retry; 4176442e31caSJeff Layton } 41771da177e4SLinus Torvalds out: 41782d8f3038SAl Viro path_put(&old_path); 41791da177e4SLinus Torvalds 41801da177e4SLinus Torvalds return error; 41811da177e4SLinus Torvalds } 41821da177e4SLinus Torvalds 41833480b257SHeiko Carstens SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) 41845590ff0dSUlrich Drepper { 4185c04030e1SUlrich Drepper return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 41865590ff0dSUlrich Drepper } 41875590ff0dSUlrich Drepper 4188bc27027aSMiklos Szeredi /** 4189bc27027aSMiklos Szeredi * vfs_rename - rename a filesystem object 4190bc27027aSMiklos Szeredi * @old_dir: parent of source 4191bc27027aSMiklos Szeredi * @old_dentry: source 4192bc27027aSMiklos Szeredi * @new_dir: parent of destination 4193bc27027aSMiklos Szeredi * @new_dentry: destination 4194bc27027aSMiklos Szeredi * @delegated_inode: returns an inode needing a delegation break 4195520c8b16SMiklos Szeredi * @flags: rename flags 4196bc27027aSMiklos Szeredi * 4197bc27027aSMiklos Szeredi * The caller must hold multiple mutexes--see lock_rename()). 4198bc27027aSMiklos Szeredi * 4199bc27027aSMiklos Szeredi * If vfs_rename discovers a delegation in need of breaking at either 4200bc27027aSMiklos Szeredi * the source or destination, it will return -EWOULDBLOCK and return a 4201bc27027aSMiklos Szeredi * reference to the inode in delegated_inode. The caller should then 4202bc27027aSMiklos Szeredi * break the delegation and retry. Because breaking a delegation may 4203bc27027aSMiklos Szeredi * take a long time, the caller should drop all locks before doing 4204bc27027aSMiklos Szeredi * so. 4205bc27027aSMiklos Szeredi * 4206bc27027aSMiklos Szeredi * Alternatively, a caller may pass NULL for delegated_inode. This may 4207bc27027aSMiklos Szeredi * be appropriate for callers that expect the underlying filesystem not 4208bc27027aSMiklos Szeredi * to be NFS exported. 4209bc27027aSMiklos Szeredi * 42101da177e4SLinus Torvalds * The worst of all namespace operations - renaming directory. "Perverted" 42111da177e4SLinus Torvalds * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 42121da177e4SLinus Torvalds * Problems: 4213d03b29a2SJ. Bruce Fields * a) we can get into loop creation. 42141da177e4SLinus Torvalds * b) race potential - two innocent renames can create a loop together. 42151da177e4SLinus Torvalds * That's where 4.4 screws up. Current fix: serialization on 4216a11f3a05SArjan van de Ven * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 42171da177e4SLinus Torvalds * story. 42186cedba89SJ. Bruce Fields * c) we have to lock _four_ objects - parents and victim (if it exists), 42196cedba89SJ. Bruce Fields * and source (if it is not a directory). 42201b1dcc1bSJes Sorensen * And that - after we got ->i_mutex on parents (until then we don't know 42211da177e4SLinus Torvalds * whether the target exists). Solution: try to be smart with locking 42221da177e4SLinus Torvalds * order for inodes. We rely on the fact that tree topology may change 4223a11f3a05SArjan van de Ven * only under ->s_vfs_rename_mutex _and_ that parent of the object we 42241da177e4SLinus Torvalds * move will be locked. Thus we can rank directories by the tree 42251da177e4SLinus Torvalds * (ancestors first) and rank all non-directories after them. 42261da177e4SLinus Torvalds * That works since everybody except rename does "lock parent, lookup, 4227a11f3a05SArjan van de Ven * lock child" and rename is under ->s_vfs_rename_mutex. 42281da177e4SLinus Torvalds * HOWEVER, it relies on the assumption that any object with ->lookup() 42291da177e4SLinus Torvalds * has no more than 1 dentry. If "hybrid" objects will ever appear, 42301da177e4SLinus Torvalds * we'd better make sure that there's no link(2) for them. 4231e4eaac06SSage Weil * d) conversion from fhandle to dentry may come in the wrong moment - when 42321b1dcc1bSJes Sorensen * we are removing the target. Solution: we will have to grab ->i_mutex 42331da177e4SLinus Torvalds * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 4234c41b20e7SAdam Buchbinder * ->i_mutex on parents, which works but leads to some truly excessive 42351da177e4SLinus Torvalds * locking]. 42361da177e4SLinus Torvalds */ 42371da177e4SLinus Torvalds int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, 42388e6d782cSJ. Bruce Fields struct inode *new_dir, struct dentry *new_dentry, 4239520c8b16SMiklos Szeredi struct inode **delegated_inode, unsigned int flags) 42401da177e4SLinus Torvalds { 42411da177e4SLinus Torvalds int error; 4242bc27027aSMiklos Szeredi bool is_dir = d_is_dir(old_dentry); 424359b0df21SEric Paris const unsigned char *old_name; 4244bc27027aSMiklos Szeredi struct inode *source = old_dentry->d_inode; 4245bc27027aSMiklos Szeredi struct inode *target = new_dentry->d_inode; 4246da1ce067SMiklos Szeredi bool new_is_dir = false; 4247da1ce067SMiklos Szeredi unsigned max_links = new_dir->i_sb->s_max_links; 42481da177e4SLinus Torvalds 4249bc27027aSMiklos Szeredi if (source == target) 42501da177e4SLinus Torvalds return 0; 42511da177e4SLinus Torvalds 42521da177e4SLinus Torvalds error = may_delete(old_dir, old_dentry, is_dir); 42531da177e4SLinus Torvalds if (error) 42541da177e4SLinus Torvalds return error; 42551da177e4SLinus Torvalds 4256da1ce067SMiklos Szeredi if (!target) { 4257a95164d9SMiklos Szeredi error = may_create(new_dir, new_dentry); 4258da1ce067SMiklos Szeredi } else { 4259da1ce067SMiklos Szeredi new_is_dir = d_is_dir(new_dentry); 4260da1ce067SMiklos Szeredi 4261da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 42621da177e4SLinus Torvalds error = may_delete(new_dir, new_dentry, is_dir); 4263da1ce067SMiklos Szeredi else 4264da1ce067SMiklos Szeredi error = may_delete(new_dir, new_dentry, new_is_dir); 4265da1ce067SMiklos Szeredi } 42661da177e4SLinus Torvalds if (error) 42671da177e4SLinus Torvalds return error; 42681da177e4SLinus Torvalds 42697177a9c4SMiklos Szeredi if (!old_dir->i_op->rename && !old_dir->i_op->rename2) 42701da177e4SLinus Torvalds return -EPERM; 42711da177e4SLinus Torvalds 4272520c8b16SMiklos Szeredi if (flags && !old_dir->i_op->rename2) 4273520c8b16SMiklos Szeredi return -EINVAL; 4274520c8b16SMiklos Szeredi 4275bc27027aSMiklos Szeredi /* 4276bc27027aSMiklos Szeredi * If we are going to change the parent - check write permissions, 4277bc27027aSMiklos Szeredi * we'll need to flip '..'. 4278bc27027aSMiklos Szeredi */ 4279da1ce067SMiklos Szeredi if (new_dir != old_dir) { 4280da1ce067SMiklos Szeredi if (is_dir) { 4281bc27027aSMiklos Szeredi error = inode_permission(source, MAY_WRITE); 4282bc27027aSMiklos Szeredi if (error) 4283bc27027aSMiklos Szeredi return error; 4284bc27027aSMiklos Szeredi } 4285da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && new_is_dir) { 4286da1ce067SMiklos Szeredi error = inode_permission(target, MAY_WRITE); 4287da1ce067SMiklos Szeredi if (error) 4288da1ce067SMiklos Szeredi return error; 4289da1ce067SMiklos Szeredi } 4290da1ce067SMiklos Szeredi } 42910eeca283SRobert Love 42920b3974ebSMiklos Szeredi error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry, 42930b3974ebSMiklos Szeredi flags); 4294bc27027aSMiklos Szeredi if (error) 4295bc27027aSMiklos Szeredi return error; 4296bc27027aSMiklos Szeredi 4297bc27027aSMiklos Szeredi old_name = fsnotify_oldname_init(old_dentry->d_name.name); 4298bc27027aSMiklos Szeredi dget(new_dentry); 4299da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4300bc27027aSMiklos Szeredi lock_two_nondirectories(source, target); 4301bc27027aSMiklos Szeredi else if (target) 43025955102cSAl Viro inode_lock(target); 4303bc27027aSMiklos Szeredi 4304bc27027aSMiklos Szeredi error = -EBUSY; 43057af1364fSEric W. Biederman if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry)) 4306bc27027aSMiklos Szeredi goto out; 4307bc27027aSMiklos Szeredi 4308da1ce067SMiklos Szeredi if (max_links && new_dir != old_dir) { 4309bc27027aSMiklos Szeredi error = -EMLINK; 4310da1ce067SMiklos Szeredi if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links) 4311bc27027aSMiklos Szeredi goto out; 4312da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir && 4313da1ce067SMiklos Szeredi old_dir->i_nlink >= max_links) 4314da1ce067SMiklos Szeredi goto out; 4315da1ce067SMiklos Szeredi } 4316da1ce067SMiklos Szeredi if (is_dir && !(flags & RENAME_EXCHANGE) && target) 4317bc27027aSMiklos Szeredi shrink_dcache_parent(new_dentry); 4318da1ce067SMiklos Szeredi if (!is_dir) { 4319bc27027aSMiklos Szeredi error = try_break_deleg(source, delegated_inode); 4320bc27027aSMiklos Szeredi if (error) 4321bc27027aSMiklos Szeredi goto out; 4322da1ce067SMiklos Szeredi } 4323da1ce067SMiklos Szeredi if (target && !new_is_dir) { 4324bc27027aSMiklos Szeredi error = try_break_deleg(target, delegated_inode); 4325bc27027aSMiklos Szeredi if (error) 4326bc27027aSMiklos Szeredi goto out; 4327bc27027aSMiklos Szeredi } 43287177a9c4SMiklos Szeredi if (!old_dir->i_op->rename2) { 4329520c8b16SMiklos Szeredi error = old_dir->i_op->rename(old_dir, old_dentry, 4330520c8b16SMiklos Szeredi new_dir, new_dentry); 4331520c8b16SMiklos Szeredi } else { 43327177a9c4SMiklos Szeredi WARN_ON(old_dir->i_op->rename != NULL); 4333520c8b16SMiklos Szeredi error = old_dir->i_op->rename2(old_dir, old_dentry, 4334520c8b16SMiklos Szeredi new_dir, new_dentry, flags); 4335520c8b16SMiklos Szeredi } 4336bc27027aSMiklos Szeredi if (error) 4337bc27027aSMiklos Szeredi goto out; 4338bc27027aSMiklos Szeredi 4339da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE) && target) { 43401da177e4SLinus Torvalds if (is_dir) 4341bc27027aSMiklos Szeredi target->i_flags |= S_DEAD; 4342bc27027aSMiklos Szeredi dont_mount(new_dentry); 43438ed936b5SEric W. Biederman detach_mounts(new_dentry); 4344bc27027aSMiklos Szeredi } 4345da1ce067SMiklos Szeredi if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) { 4346da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 4347bc27027aSMiklos Szeredi d_move(old_dentry, new_dentry); 4348da1ce067SMiklos Szeredi else 4349da1ce067SMiklos Szeredi d_exchange(old_dentry, new_dentry); 4350da1ce067SMiklos Szeredi } 4351bc27027aSMiklos Szeredi out: 4352da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4353bc27027aSMiklos Szeredi unlock_two_nondirectories(source, target); 4354bc27027aSMiklos Szeredi else if (target) 43555955102cSAl Viro inode_unlock(target); 4356bc27027aSMiklos Szeredi dput(new_dentry); 4357da1ce067SMiklos Szeredi if (!error) { 4358123df294SAl Viro fsnotify_move(old_dir, new_dir, old_name, is_dir, 4359da1ce067SMiklos Szeredi !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry); 4360da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4361da1ce067SMiklos Szeredi fsnotify_move(new_dir, old_dir, old_dentry->d_name.name, 4362da1ce067SMiklos Szeredi new_is_dir, NULL, new_dentry); 4363da1ce067SMiklos Szeredi } 4364da1ce067SMiklos Szeredi } 43650eeca283SRobert Love fsnotify_oldname_free(old_name); 43660eeca283SRobert Love 43671da177e4SLinus Torvalds return error; 43681da177e4SLinus Torvalds } 43694d359507SAl Viro EXPORT_SYMBOL(vfs_rename); 43701da177e4SLinus Torvalds 4371520c8b16SMiklos Szeredi SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname, 4372520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname, unsigned int, flags) 43731da177e4SLinus Torvalds { 43741da177e4SLinus Torvalds struct dentry *old_dentry, *new_dentry; 43751da177e4SLinus Torvalds struct dentry *trap; 4376f5beed75SAl Viro struct path old_path, new_path; 4377f5beed75SAl Viro struct qstr old_last, new_last; 4378f5beed75SAl Viro int old_type, new_type; 43798e6d782cSJ. Bruce Fields struct inode *delegated_inode = NULL; 438091a27b2aSJeff Layton struct filename *from; 438191a27b2aSJeff Layton struct filename *to; 4382f5beed75SAl Viro unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET; 4383c6a94284SJeff Layton bool should_retry = false; 43842ad94ae6SAl Viro int error; 4385520c8b16SMiklos Szeredi 43860d7a8555SMiklos Szeredi if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) 4387da1ce067SMiklos Szeredi return -EINVAL; 4388da1ce067SMiklos Szeredi 43890d7a8555SMiklos Szeredi if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) && 43900d7a8555SMiklos Szeredi (flags & RENAME_EXCHANGE)) 4391520c8b16SMiklos Szeredi return -EINVAL; 4392520c8b16SMiklos Szeredi 43930d7a8555SMiklos Szeredi if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD)) 43940d7a8555SMiklos Szeredi return -EPERM; 43950d7a8555SMiklos Szeredi 4396f5beed75SAl Viro if (flags & RENAME_EXCHANGE) 4397f5beed75SAl Viro target_flags = 0; 4398f5beed75SAl Viro 4399c6a94284SJeff Layton retry: 4400f5beed75SAl Viro from = user_path_parent(olddfd, oldname, 4401f5beed75SAl Viro &old_path, &old_last, &old_type, lookup_flags); 440291a27b2aSJeff Layton if (IS_ERR(from)) { 440391a27b2aSJeff Layton error = PTR_ERR(from); 44041da177e4SLinus Torvalds goto exit; 440591a27b2aSJeff Layton } 44061da177e4SLinus Torvalds 4407f5beed75SAl Viro to = user_path_parent(newdfd, newname, 4408f5beed75SAl Viro &new_path, &new_last, &new_type, lookup_flags); 440991a27b2aSJeff Layton if (IS_ERR(to)) { 441091a27b2aSJeff Layton error = PTR_ERR(to); 44111da177e4SLinus Torvalds goto exit1; 441291a27b2aSJeff Layton } 44131da177e4SLinus Torvalds 44141da177e4SLinus Torvalds error = -EXDEV; 4415f5beed75SAl Viro if (old_path.mnt != new_path.mnt) 44161da177e4SLinus Torvalds goto exit2; 44171da177e4SLinus Torvalds 44181da177e4SLinus Torvalds error = -EBUSY; 4419f5beed75SAl Viro if (old_type != LAST_NORM) 44201da177e4SLinus Torvalds goto exit2; 44211da177e4SLinus Torvalds 44220a7c3937SMiklos Szeredi if (flags & RENAME_NOREPLACE) 44230a7c3937SMiklos Szeredi error = -EEXIST; 4424f5beed75SAl Viro if (new_type != LAST_NORM) 44251da177e4SLinus Torvalds goto exit2; 44261da177e4SLinus Torvalds 4427f5beed75SAl Viro error = mnt_want_write(old_path.mnt); 4428c30dabfeSJan Kara if (error) 4429c30dabfeSJan Kara goto exit2; 4430c30dabfeSJan Kara 44318e6d782cSJ. Bruce Fields retry_deleg: 4432f5beed75SAl Viro trap = lock_rename(new_path.dentry, old_path.dentry); 44331da177e4SLinus Torvalds 4434f5beed75SAl Viro old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags); 44351da177e4SLinus Torvalds error = PTR_ERR(old_dentry); 44361da177e4SLinus Torvalds if (IS_ERR(old_dentry)) 44371da177e4SLinus Torvalds goto exit3; 44381da177e4SLinus Torvalds /* source must exist */ 44391da177e4SLinus Torvalds error = -ENOENT; 4440b18825a7SDavid Howells if (d_is_negative(old_dentry)) 44411da177e4SLinus Torvalds goto exit4; 4442f5beed75SAl Viro new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags); 44431da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 44441da177e4SLinus Torvalds if (IS_ERR(new_dentry)) 44451da177e4SLinus Torvalds goto exit4; 44460a7c3937SMiklos Szeredi error = -EEXIST; 44470a7c3937SMiklos Szeredi if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry)) 44480a7c3937SMiklos Szeredi goto exit5; 4449da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4450da1ce067SMiklos Szeredi error = -ENOENT; 4451da1ce067SMiklos Szeredi if (d_is_negative(new_dentry)) 4452da1ce067SMiklos Szeredi goto exit5; 4453da1ce067SMiklos Szeredi 4454da1ce067SMiklos Szeredi if (!d_is_dir(new_dentry)) { 4455da1ce067SMiklos Szeredi error = -ENOTDIR; 4456f5beed75SAl Viro if (new_last.name[new_last.len]) 4457da1ce067SMiklos Szeredi goto exit5; 4458da1ce067SMiklos Szeredi } 4459da1ce067SMiklos Szeredi } 44600a7c3937SMiklos Szeredi /* unless the source is a directory trailing slashes give -ENOTDIR */ 44610a7c3937SMiklos Szeredi if (!d_is_dir(old_dentry)) { 44620a7c3937SMiklos Szeredi error = -ENOTDIR; 4463f5beed75SAl Viro if (old_last.name[old_last.len]) 44640a7c3937SMiklos Szeredi goto exit5; 4465f5beed75SAl Viro if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len]) 44660a7c3937SMiklos Szeredi goto exit5; 44670a7c3937SMiklos Szeredi } 44680a7c3937SMiklos Szeredi /* source should not be ancestor of target */ 44690a7c3937SMiklos Szeredi error = -EINVAL; 44700a7c3937SMiklos Szeredi if (old_dentry == trap) 44710a7c3937SMiklos Szeredi goto exit5; 44721da177e4SLinus Torvalds /* target should not be an ancestor of source */ 4473da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 44741da177e4SLinus Torvalds error = -ENOTEMPTY; 44751da177e4SLinus Torvalds if (new_dentry == trap) 44761da177e4SLinus Torvalds goto exit5; 44771da177e4SLinus Torvalds 4478f5beed75SAl Viro error = security_path_rename(&old_path, old_dentry, 4479f5beed75SAl Viro &new_path, new_dentry, flags); 4480be6d3e56SKentaro Takeda if (error) 4481c30dabfeSJan Kara goto exit5; 4482f5beed75SAl Viro error = vfs_rename(old_path.dentry->d_inode, old_dentry, 4483f5beed75SAl Viro new_path.dentry->d_inode, new_dentry, 4484520c8b16SMiklos Szeredi &delegated_inode, flags); 44851da177e4SLinus Torvalds exit5: 44861da177e4SLinus Torvalds dput(new_dentry); 44871da177e4SLinus Torvalds exit4: 44881da177e4SLinus Torvalds dput(old_dentry); 44891da177e4SLinus Torvalds exit3: 4490f5beed75SAl Viro unlock_rename(new_path.dentry, old_path.dentry); 44918e6d782cSJ. Bruce Fields if (delegated_inode) { 44928e6d782cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 44938e6d782cSJ. Bruce Fields if (!error) 44948e6d782cSJ. Bruce Fields goto retry_deleg; 44958e6d782cSJ. Bruce Fields } 4496f5beed75SAl Viro mnt_drop_write(old_path.mnt); 44971da177e4SLinus Torvalds exit2: 4498c6a94284SJeff Layton if (retry_estale(error, lookup_flags)) 4499c6a94284SJeff Layton should_retry = true; 4500f5beed75SAl Viro path_put(&new_path); 45012ad94ae6SAl Viro putname(to); 45021da177e4SLinus Torvalds exit1: 4503f5beed75SAl Viro path_put(&old_path); 45041da177e4SLinus Torvalds putname(from); 4505c6a94284SJeff Layton if (should_retry) { 4506c6a94284SJeff Layton should_retry = false; 4507c6a94284SJeff Layton lookup_flags |= LOOKUP_REVAL; 4508c6a94284SJeff Layton goto retry; 4509c6a94284SJeff Layton } 45102ad94ae6SAl Viro exit: 45111da177e4SLinus Torvalds return error; 45121da177e4SLinus Torvalds } 45131da177e4SLinus Torvalds 4514520c8b16SMiklos Szeredi SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, 4515520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname) 4516520c8b16SMiklos Szeredi { 4517520c8b16SMiklos Szeredi return sys_renameat2(olddfd, oldname, newdfd, newname, 0); 4518520c8b16SMiklos Szeredi } 4519520c8b16SMiklos Szeredi 4520a26eab24SHeiko Carstens SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) 45215590ff0dSUlrich Drepper { 4522520c8b16SMiklos Szeredi return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 45235590ff0dSUlrich Drepper } 45245590ff0dSUlrich Drepper 4525787fb6bcSMiklos Szeredi int vfs_whiteout(struct inode *dir, struct dentry *dentry) 4526787fb6bcSMiklos Szeredi { 4527787fb6bcSMiklos Szeredi int error = may_create(dir, dentry); 4528787fb6bcSMiklos Szeredi if (error) 4529787fb6bcSMiklos Szeredi return error; 4530787fb6bcSMiklos Szeredi 4531787fb6bcSMiklos Szeredi if (!dir->i_op->mknod) 4532787fb6bcSMiklos Szeredi return -EPERM; 4533787fb6bcSMiklos Szeredi 4534787fb6bcSMiklos Szeredi return dir->i_op->mknod(dir, dentry, 4535787fb6bcSMiklos Szeredi S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV); 4536787fb6bcSMiklos Szeredi } 4537787fb6bcSMiklos Szeredi EXPORT_SYMBOL(vfs_whiteout); 4538787fb6bcSMiklos Szeredi 45395d826c84SAl Viro int readlink_copy(char __user *buffer, int buflen, const char *link) 45401da177e4SLinus Torvalds { 45415d826c84SAl Viro int len = PTR_ERR(link); 45421da177e4SLinus Torvalds if (IS_ERR(link)) 45431da177e4SLinus Torvalds goto out; 45441da177e4SLinus Torvalds 45451da177e4SLinus Torvalds len = strlen(link); 45461da177e4SLinus Torvalds if (len > (unsigned) buflen) 45471da177e4SLinus Torvalds len = buflen; 45481da177e4SLinus Torvalds if (copy_to_user(buffer, link, len)) 45491da177e4SLinus Torvalds len = -EFAULT; 45501da177e4SLinus Torvalds out: 45511da177e4SLinus Torvalds return len; 45521da177e4SLinus Torvalds } 45535d826c84SAl Viro EXPORT_SYMBOL(readlink_copy); 45541da177e4SLinus Torvalds 45551da177e4SLinus Torvalds /* 45561da177e4SLinus Torvalds * A helper for ->readlink(). This should be used *ONLY* for symlinks that 45576b255391SAl Viro * have ->get_link() not calling nd_jump_link(). Using (or not using) it 45586b255391SAl Viro * for any given inode is up to filesystem. 45591da177e4SLinus Torvalds */ 45601da177e4SLinus Torvalds int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen) 45611da177e4SLinus Torvalds { 4562fceef393SAl Viro DEFINE_DELAYED_CALL(done); 45635f2c4179SAl Viro struct inode *inode = d_inode(dentry); 45645f2c4179SAl Viro const char *link = inode->i_link; 4565694a1764SMarcin Slusarz int res; 4566cc314eefSLinus Torvalds 4567d4dee48bSAl Viro if (!link) { 4568fceef393SAl Viro link = inode->i_op->get_link(dentry, inode, &done); 4569680baacbSAl Viro if (IS_ERR(link)) 4570680baacbSAl Viro return PTR_ERR(link); 4571d4dee48bSAl Viro } 4572680baacbSAl Viro res = readlink_copy(buffer, buflen, link); 4573fceef393SAl Viro do_delayed_call(&done); 4574694a1764SMarcin Slusarz return res; 45751da177e4SLinus Torvalds } 45764d359507SAl Viro EXPORT_SYMBOL(generic_readlink); 45771da177e4SLinus Torvalds 45781da177e4SLinus Torvalds /* get the link contents into pagecache */ 45796b255391SAl Viro const char *page_get_link(struct dentry *dentry, struct inode *inode, 4580fceef393SAl Viro struct delayed_call *callback) 45811da177e4SLinus Torvalds { 4582ebd09abbSDuane Griffin char *kaddr; 45831da177e4SLinus Torvalds struct page *page; 45846b255391SAl Viro struct address_space *mapping = inode->i_mapping; 45856b255391SAl Viro 4586d3883d4fSAl Viro if (!dentry) { 4587d3883d4fSAl Viro page = find_get_page(mapping, 0); 4588d3883d4fSAl Viro if (!page) 45896b255391SAl Viro return ERR_PTR(-ECHILD); 4590d3883d4fSAl Viro if (!PageUptodate(page)) { 4591d3883d4fSAl Viro put_page(page); 4592d3883d4fSAl Viro return ERR_PTR(-ECHILD); 4593d3883d4fSAl Viro } 4594d3883d4fSAl Viro } else { 4595090d2b18SPekka Enberg page = read_mapping_page(mapping, 0, NULL); 45961da177e4SLinus Torvalds if (IS_ERR(page)) 45976fe6900eSNick Piggin return (char*)page; 4598d3883d4fSAl Viro } 4599fceef393SAl Viro set_delayed_call(callback, page_put_link, page); 460021fc61c7SAl Viro BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM); 460121fc61c7SAl Viro kaddr = page_address(page); 46026b255391SAl Viro nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1); 4603ebd09abbSDuane Griffin return kaddr; 46041da177e4SLinus Torvalds } 46051da177e4SLinus Torvalds 46066b255391SAl Viro EXPORT_SYMBOL(page_get_link); 46071da177e4SLinus Torvalds 4608fceef393SAl Viro void page_put_link(void *arg) 46091da177e4SLinus Torvalds { 4610fceef393SAl Viro put_page(arg); 46111da177e4SLinus Torvalds } 46124d359507SAl Viro EXPORT_SYMBOL(page_put_link); 46131da177e4SLinus Torvalds 4614aa80deabSAl Viro int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 4615aa80deabSAl Viro { 4616fceef393SAl Viro DEFINE_DELAYED_CALL(done); 46176b255391SAl Viro int res = readlink_copy(buffer, buflen, 46186b255391SAl Viro page_get_link(dentry, d_inode(dentry), 4619fceef393SAl Viro &done)); 4620fceef393SAl Viro do_delayed_call(&done); 4621aa80deabSAl Viro return res; 4622aa80deabSAl Viro } 4623aa80deabSAl Viro EXPORT_SYMBOL(page_readlink); 4624aa80deabSAl Viro 462554566b2cSNick Piggin /* 462654566b2cSNick Piggin * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS 462754566b2cSNick Piggin */ 462854566b2cSNick Piggin int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) 46291da177e4SLinus Torvalds { 46301da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 46310adb25d2SKirill Korotaev struct page *page; 4632afddba49SNick Piggin void *fsdata; 4633beb497abSDmitriy Monakhov int err; 463454566b2cSNick Piggin unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE; 463554566b2cSNick Piggin if (nofs) 463654566b2cSNick Piggin flags |= AOP_FLAG_NOFS; 46371da177e4SLinus Torvalds 46387e53cac4SNeilBrown retry: 4639afddba49SNick Piggin err = pagecache_write_begin(NULL, mapping, 0, len-1, 464054566b2cSNick Piggin flags, &page, &fsdata); 46411da177e4SLinus Torvalds if (err) 4642afddba49SNick Piggin goto fail; 4643afddba49SNick Piggin 464421fc61c7SAl Viro memcpy(page_address(page), symname, len-1); 4645afddba49SNick Piggin 4646afddba49SNick Piggin err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, 4647afddba49SNick Piggin page, fsdata); 46481da177e4SLinus Torvalds if (err < 0) 46491da177e4SLinus Torvalds goto fail; 4650afddba49SNick Piggin if (err < len-1) 4651afddba49SNick Piggin goto retry; 4652afddba49SNick Piggin 46531da177e4SLinus Torvalds mark_inode_dirty(inode); 46541da177e4SLinus Torvalds return 0; 46551da177e4SLinus Torvalds fail: 46561da177e4SLinus Torvalds return err; 46571da177e4SLinus Torvalds } 46584d359507SAl Viro EXPORT_SYMBOL(__page_symlink); 46591da177e4SLinus Torvalds 46600adb25d2SKirill Korotaev int page_symlink(struct inode *inode, const char *symname, int len) 46610adb25d2SKirill Korotaev { 46620adb25d2SKirill Korotaev return __page_symlink(inode, symname, len, 4663c62d2555SMichal Hocko !mapping_gfp_constraint(inode->i_mapping, __GFP_FS)); 46640adb25d2SKirill Korotaev } 46654d359507SAl Viro EXPORT_SYMBOL(page_symlink); 46660adb25d2SKirill Korotaev 466792e1d5beSArjan van de Ven const struct inode_operations page_symlink_inode_operations = { 46681da177e4SLinus Torvalds .readlink = generic_readlink, 46696b255391SAl Viro .get_link = page_get_link, 46701da177e4SLinus Torvalds }; 46711da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink_inode_operations); 4672