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; 5028f47a016SAl Viro unsigned seq, m_seq, root_seq; 5031f55a6ecSAl Viro int last_type; 5041f55a6ecSAl Viro unsigned depth; 505756daf26SNeilBrown int total_link_count; 506697fc6caSAl Viro struct saved { 507697fc6caSAl Viro struct path link; 508697fc6caSAl Viro void *cookie; 509697fc6caSAl Viro const char *name; 510237d8b32SAl Viro struct inode *inode; 5110450b2d1SAl Viro unsigned seq; 512894bc8c4SAl Viro } *stack, internal[EMBEDDED_LEVELS]; 5131f55a6ecSAl Viro }; 5141f55a6ecSAl Viro 515756daf26SNeilBrown static struct nameidata *set_nameidata(struct nameidata *p) 516894bc8c4SAl Viro { 517756daf26SNeilBrown struct nameidata *old = current->nameidata; 518756daf26SNeilBrown p->stack = p->internal; 519756daf26SNeilBrown p->total_link_count = old ? old->total_link_count : 0; 520756daf26SNeilBrown current->nameidata = p; 521756daf26SNeilBrown return old; 522894bc8c4SAl Viro } 523894bc8c4SAl Viro 524756daf26SNeilBrown static void restore_nameidata(struct nameidata *old) 525894bc8c4SAl Viro { 526756daf26SNeilBrown struct nameidata *now = current->nameidata; 527756daf26SNeilBrown 528756daf26SNeilBrown current->nameidata = old; 529756daf26SNeilBrown if (old) 530756daf26SNeilBrown old->total_link_count = now->total_link_count; 531756daf26SNeilBrown if (now->stack != now->internal) { 532756daf26SNeilBrown kfree(now->stack); 533756daf26SNeilBrown now->stack = now->internal; 534894bc8c4SAl Viro } 535894bc8c4SAl Viro } 536894bc8c4SAl Viro 537894bc8c4SAl Viro static int __nd_alloc_stack(struct nameidata *nd) 538894bc8c4SAl Viro { 539bc40aee0SAl Viro struct saved *p; 540bc40aee0SAl Viro 541bc40aee0SAl Viro if (nd->flags & LOOKUP_RCU) { 542bc40aee0SAl Viro p= kmalloc(MAXSYMLINKS * sizeof(struct saved), 543bc40aee0SAl Viro GFP_ATOMIC); 544bc40aee0SAl Viro if (unlikely(!p)) 545bc40aee0SAl Viro return -ECHILD; 546bc40aee0SAl Viro } else { 547bc40aee0SAl Viro p= kmalloc(MAXSYMLINKS * sizeof(struct saved), 548894bc8c4SAl Viro GFP_KERNEL); 549894bc8c4SAl Viro if (unlikely(!p)) 550894bc8c4SAl Viro return -ENOMEM; 551bc40aee0SAl Viro } 552894bc8c4SAl Viro memcpy(p, nd->internal, sizeof(nd->internal)); 553894bc8c4SAl Viro nd->stack = p; 554894bc8c4SAl Viro return 0; 555894bc8c4SAl Viro } 556894bc8c4SAl Viro 557894bc8c4SAl Viro static inline int nd_alloc_stack(struct nameidata *nd) 558894bc8c4SAl Viro { 559da4e0be0SAl Viro if (likely(nd->depth != EMBEDDED_LEVELS)) 560894bc8c4SAl Viro return 0; 561894bc8c4SAl Viro if (likely(nd->stack != nd->internal)) 562894bc8c4SAl Viro return 0; 563894bc8c4SAl Viro return __nd_alloc_stack(nd); 564894bc8c4SAl Viro } 565894bc8c4SAl Viro 5667973387aSAl Viro static void drop_links(struct nameidata *nd) 5677973387aSAl Viro { 5687973387aSAl Viro int i = nd->depth; 5697973387aSAl Viro while (i--) { 5707973387aSAl Viro struct saved *last = nd->stack + i; 5717973387aSAl Viro struct inode *inode = last->inode; 5727973387aSAl Viro if (last->cookie && inode->i_op->put_link) { 5737973387aSAl Viro inode->i_op->put_link(inode, last->cookie); 5747973387aSAl Viro last->cookie = NULL; 5757973387aSAl Viro } 5767973387aSAl Viro } 5777973387aSAl Viro } 5787973387aSAl Viro 5797973387aSAl Viro static void terminate_walk(struct nameidata *nd) 5807973387aSAl Viro { 5817973387aSAl Viro drop_links(nd); 5827973387aSAl Viro if (!(nd->flags & LOOKUP_RCU)) { 5837973387aSAl Viro int i; 5847973387aSAl Viro path_put(&nd->path); 5857973387aSAl Viro for (i = 0; i < nd->depth; i++) 5867973387aSAl Viro path_put(&nd->stack[i].link); 587102b8af2SAl Viro if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 588102b8af2SAl Viro path_put(&nd->root); 589102b8af2SAl Viro nd->root.mnt = NULL; 590102b8af2SAl Viro } 5917973387aSAl Viro } else { 5927973387aSAl Viro nd->flags &= ~LOOKUP_RCU; 5937973387aSAl Viro if (!(nd->flags & LOOKUP_ROOT)) 5947973387aSAl Viro nd->root.mnt = NULL; 5957973387aSAl Viro rcu_read_unlock(); 5967973387aSAl Viro } 5977973387aSAl Viro nd->depth = 0; 5987973387aSAl Viro } 5997973387aSAl Viro 6007973387aSAl Viro /* path_put is needed afterwards regardless of success or failure */ 6017973387aSAl Viro static bool legitimize_path(struct nameidata *nd, 6027973387aSAl Viro struct path *path, unsigned seq) 6037973387aSAl Viro { 6047973387aSAl Viro int res = __legitimize_mnt(path->mnt, nd->m_seq); 6057973387aSAl Viro if (unlikely(res)) { 6067973387aSAl Viro if (res > 0) 6077973387aSAl Viro path->mnt = NULL; 6087973387aSAl Viro path->dentry = NULL; 6097973387aSAl Viro return false; 6107973387aSAl Viro } 6117973387aSAl Viro if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) { 6127973387aSAl Viro path->dentry = NULL; 6137973387aSAl Viro return false; 6147973387aSAl Viro } 6157973387aSAl Viro return !read_seqcount_retry(&path->dentry->d_seq, seq); 6167973387aSAl Viro } 6177973387aSAl Viro 6187973387aSAl Viro static bool legitimize_links(struct nameidata *nd) 6197973387aSAl Viro { 6207973387aSAl Viro int i; 6217973387aSAl Viro for (i = 0; i < nd->depth; i++) { 6227973387aSAl Viro struct saved *last = nd->stack + i; 6237973387aSAl Viro if (unlikely(!legitimize_path(nd, &last->link, last->seq))) { 6247973387aSAl Viro drop_links(nd); 6257973387aSAl Viro nd->depth = i + 1; 6267973387aSAl Viro return false; 6277973387aSAl Viro } 6287973387aSAl Viro } 6297973387aSAl Viro return true; 6307973387aSAl Viro } 6317973387aSAl Viro 63219660af7SAl Viro /* 63331e6b01fSNick Piggin * Path walking has 2 modes, rcu-walk and ref-walk (see 63419660af7SAl Viro * Documentation/filesystems/path-lookup.txt). In situations when we can't 63519660af7SAl Viro * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab 63619660af7SAl Viro * normal reference counts on dentries and vfsmounts to transition to rcu-walk 63719660af7SAl Viro * mode. Refcounts are grabbed at the last known good point before rcu-walk 63819660af7SAl Viro * got stuck, so ref-walk may continue from there. If this is not successful 63919660af7SAl Viro * (eg. a seqcount has changed), then failure is returned and it's up to caller 64019660af7SAl Viro * to restart the path walk from the beginning in ref-walk mode. 64131e6b01fSNick Piggin */ 64231e6b01fSNick Piggin 64331e6b01fSNick Piggin /** 64419660af7SAl Viro * unlazy_walk - try to switch to ref-walk mode. 64519660af7SAl Viro * @nd: nameidata pathwalk data 64619660af7SAl Viro * @dentry: child of nd->path.dentry or NULL 6476e9918b7SAl Viro * @seq: seq number to check dentry against 64839191628SRandy Dunlap * Returns: 0 on success, -ECHILD on failure 64931e6b01fSNick Piggin * 65019660af7SAl Viro * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry 65119660af7SAl Viro * for ref-walk mode. @dentry must be a path found by a do_lookup call on 65219660af7SAl Viro * @nd or NULL. Must be called from rcu-walk context. 6537973387aSAl Viro * Nothing should touch nameidata between unlazy_walk() failure and 6547973387aSAl Viro * terminate_walk(). 65531e6b01fSNick Piggin */ 6566e9918b7SAl Viro static int unlazy_walk(struct nameidata *nd, struct dentry *dentry, unsigned seq) 65731e6b01fSNick Piggin { 65831e6b01fSNick Piggin struct dentry *parent = nd->path.dentry; 65931e6b01fSNick Piggin 66031e6b01fSNick Piggin BUG_ON(!(nd->flags & LOOKUP_RCU)); 661e5c832d5SLinus Torvalds 662e5c832d5SLinus Torvalds nd->flags &= ~LOOKUP_RCU; 6637973387aSAl Viro if (unlikely(!legitimize_links(nd))) 6647973387aSAl Viro goto out2; 6657973387aSAl Viro if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq))) 6667973387aSAl Viro goto out2; 6677973387aSAl Viro if (unlikely(!lockref_get_not_dead(&parent->d_lockref))) 6687973387aSAl Viro goto out1; 66948a066e7SAl Viro 67015570086SLinus Torvalds /* 67115570086SLinus Torvalds * For a negative lookup, the lookup sequence point is the parents 67215570086SLinus Torvalds * sequence point, and it only needs to revalidate the parent dentry. 67315570086SLinus Torvalds * 67415570086SLinus Torvalds * For a positive lookup, we need to move both the parent and the 67515570086SLinus Torvalds * dentry from the RCU domain to be properly refcounted. And the 67615570086SLinus Torvalds * sequence number in the dentry validates *both* dentry counters, 67715570086SLinus Torvalds * since we checked the sequence number of the parent after we got 67815570086SLinus Torvalds * the child sequence number. So we know the parent must still 67915570086SLinus Torvalds * be valid if the child sequence number is still valid. 68015570086SLinus Torvalds */ 68119660af7SAl Viro if (!dentry) { 682e5c832d5SLinus Torvalds if (read_seqcount_retry(&parent->d_seq, nd->seq)) 683e5c832d5SLinus Torvalds goto out; 68419660af7SAl Viro BUG_ON(nd->inode != parent->d_inode); 68519660af7SAl Viro } else { 686e5c832d5SLinus Torvalds if (!lockref_get_not_dead(&dentry->d_lockref)) 687e5c832d5SLinus Torvalds goto out; 6886e9918b7SAl Viro if (read_seqcount_retry(&dentry->d_seq, seq)) 689e5c832d5SLinus Torvalds goto drop_dentry; 69019660af7SAl Viro } 691e5c832d5SLinus Torvalds 692e5c832d5SLinus Torvalds /* 693e5c832d5SLinus Torvalds * Sequence counts matched. Now make sure that the root is 694e5c832d5SLinus Torvalds * still valid and get it if required. 695e5c832d5SLinus Torvalds */ 696e5c832d5SLinus Torvalds if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 6975a8d87e8SAl Viro if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) { 6985a8d87e8SAl Viro rcu_read_unlock(); 6995a8d87e8SAl Viro dput(dentry); 7005a8d87e8SAl Viro return -ECHILD; 7017973387aSAl Viro } 70231e6b01fSNick Piggin } 70331e6b01fSNick Piggin 7048b61e74fSAl Viro rcu_read_unlock(); 70531e6b01fSNick Piggin return 0; 70619660af7SAl Viro 707e5c832d5SLinus Torvalds drop_dentry: 7088b61e74fSAl Viro rcu_read_unlock(); 709e5c832d5SLinus Torvalds dput(dentry); 710d0d27277SLinus Torvalds goto drop_root_mnt; 7117973387aSAl Viro out2: 7127973387aSAl Viro nd->path.mnt = NULL; 7137973387aSAl Viro out1: 7147973387aSAl Viro nd->path.dentry = NULL; 715e5c832d5SLinus Torvalds out: 7168b61e74fSAl Viro rcu_read_unlock(); 717d0d27277SLinus Torvalds drop_root_mnt: 718d0d27277SLinus Torvalds if (!(nd->flags & LOOKUP_ROOT)) 719d0d27277SLinus Torvalds nd->root.mnt = NULL; 72031e6b01fSNick Piggin return -ECHILD; 72131e6b01fSNick Piggin } 72231e6b01fSNick Piggin 7237973387aSAl Viro static int unlazy_link(struct nameidata *nd, struct path *link, unsigned seq) 7247973387aSAl Viro { 7257973387aSAl Viro if (unlikely(!legitimize_path(nd, link, seq))) { 7267973387aSAl Viro drop_links(nd); 7277973387aSAl Viro nd->depth = 0; 7287973387aSAl Viro nd->flags &= ~LOOKUP_RCU; 7297973387aSAl Viro nd->path.mnt = NULL; 7307973387aSAl Viro nd->path.dentry = NULL; 7317973387aSAl Viro if (!(nd->flags & LOOKUP_ROOT)) 7327973387aSAl Viro nd->root.mnt = NULL; 7337973387aSAl Viro rcu_read_unlock(); 7347973387aSAl Viro } else if (likely(unlazy_walk(nd, NULL, 0)) == 0) { 7357973387aSAl Viro return 0; 7367973387aSAl Viro } 7377973387aSAl Viro path_put(link); 7387973387aSAl Viro return -ECHILD; 7397973387aSAl Viro } 7407973387aSAl Viro 7414ce16ef3SAl Viro static inline int d_revalidate(struct dentry *dentry, unsigned int flags) 74234286d66SNick Piggin { 7434ce16ef3SAl Viro return dentry->d_op->d_revalidate(dentry, flags); 74434286d66SNick Piggin } 74534286d66SNick Piggin 7469f1fafeeSAl Viro /** 7479f1fafeeSAl Viro * complete_walk - successful completion of path walk 7489f1fafeeSAl Viro * @nd: pointer nameidata 74939159de2SJeff Layton * 7509f1fafeeSAl Viro * If we had been in RCU mode, drop out of it and legitimize nd->path. 7519f1fafeeSAl Viro * Revalidate the final result, unless we'd already done that during 7529f1fafeeSAl Viro * the path walk or the filesystem doesn't ask for it. Return 0 on 7539f1fafeeSAl Viro * success, -error on failure. In case of failure caller does not 7549f1fafeeSAl Viro * need to drop nd->path. 75539159de2SJeff Layton */ 7569f1fafeeSAl Viro static int complete_walk(struct nameidata *nd) 75739159de2SJeff Layton { 75816c2cd71SAl Viro struct dentry *dentry = nd->path.dentry; 75939159de2SJeff Layton int status; 76039159de2SJeff Layton 7619f1fafeeSAl Viro if (nd->flags & LOOKUP_RCU) { 7629f1fafeeSAl Viro if (!(nd->flags & LOOKUP_ROOT)) 7639f1fafeeSAl Viro nd->root.mnt = NULL; 7646e9918b7SAl Viro if (unlikely(unlazy_walk(nd, NULL, 0))) 76548a066e7SAl Viro return -ECHILD; 76648a066e7SAl Viro } 7679f1fafeeSAl Viro 76816c2cd71SAl Viro if (likely(!(nd->flags & LOOKUP_JUMPED))) 76939159de2SJeff Layton return 0; 77039159de2SJeff Layton 771ecf3d1f1SJeff Layton if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) 77216c2cd71SAl Viro return 0; 77316c2cd71SAl Viro 774ecf3d1f1SJeff Layton status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); 77539159de2SJeff Layton if (status > 0) 77639159de2SJeff Layton return 0; 77739159de2SJeff Layton 77816c2cd71SAl Viro if (!status) 77939159de2SJeff Layton status = -ESTALE; 78016c2cd71SAl Viro 78139159de2SJeff Layton return status; 78239159de2SJeff Layton } 78339159de2SJeff Layton 78418d8c860SAl Viro static void set_root(struct nameidata *nd) 7852a737871SAl Viro { 786f7ad3c6bSMiklos Szeredi get_fs_root(current->fs, &nd->root); 7872a737871SAl Viro } 7882a737871SAl Viro 78918d8c860SAl Viro static unsigned set_root_rcu(struct nameidata *nd) 79031e6b01fSNick Piggin { 79131e6b01fSNick Piggin struct fs_struct *fs = current->fs; 7928f47a016SAl Viro unsigned seq; 793c28cc364SNick Piggin 794c28cc364SNick Piggin do { 795c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 79631e6b01fSNick Piggin nd->root = fs->root; 7978f47a016SAl Viro nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq); 798c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 7998f47a016SAl Viro return nd->root_seq; 80031e6b01fSNick Piggin } 80131e6b01fSNick Piggin 8021d957f9bSJan Blunck static void path_put_conditional(struct path *path, struct nameidata *nd) 803051d3812SIan Kent { 804051d3812SIan Kent dput(path->dentry); 8054ac91378SJan Blunck if (path->mnt != nd->path.mnt) 806051d3812SIan Kent mntput(path->mnt); 807051d3812SIan Kent } 808051d3812SIan Kent 8097b9337aaSNick Piggin static inline void path_to_nameidata(const struct path *path, 8107b9337aaSNick Piggin struct nameidata *nd) 811051d3812SIan Kent { 81231e6b01fSNick Piggin if (!(nd->flags & LOOKUP_RCU)) { 8134ac91378SJan Blunck dput(nd->path.dentry); 81431e6b01fSNick Piggin if (nd->path.mnt != path->mnt) 8154ac91378SJan Blunck mntput(nd->path.mnt); 8169a229683SHuang Shijie } 81731e6b01fSNick Piggin nd->path.mnt = path->mnt; 8184ac91378SJan Blunck nd->path.dentry = path->dentry; 819051d3812SIan Kent } 820051d3812SIan Kent 821b5fb63c1SChristoph Hellwig /* 822b5fb63c1SChristoph Hellwig * Helper to directly jump to a known parsed path from ->follow_link, 823b5fb63c1SChristoph Hellwig * caller must have taken a reference to path beforehand. 824b5fb63c1SChristoph Hellwig */ 8256e77137bSAl Viro void nd_jump_link(struct path *path) 826b5fb63c1SChristoph Hellwig { 8276e77137bSAl Viro struct nameidata *nd = current->nameidata; 828b5fb63c1SChristoph Hellwig path_put(&nd->path); 829b5fb63c1SChristoph Hellwig 830b5fb63c1SChristoph Hellwig nd->path = *path; 831b5fb63c1SChristoph Hellwig nd->inode = nd->path.dentry->d_inode; 832b5fb63c1SChristoph Hellwig nd->flags |= LOOKUP_JUMPED; 833b5fb63c1SChristoph Hellwig } 834b5fb63c1SChristoph Hellwig 835b9ff4429SAl Viro static inline void put_link(struct nameidata *nd) 836574197e0SAl Viro { 83721c3003dSAl Viro struct saved *last = nd->stack + --nd->depth; 838237d8b32SAl Viro struct inode *inode = last->inode; 839b9ff4429SAl Viro if (last->cookie && inode->i_op->put_link) 8405f2c4179SAl Viro inode->i_op->put_link(inode, last->cookie); 8416548fae2SAl Viro if (!(nd->flags & LOOKUP_RCU)) 842b9ff4429SAl Viro path_put(&last->link); 843574197e0SAl Viro } 844574197e0SAl Viro 845561ec64aSLinus Torvalds int sysctl_protected_symlinks __read_mostly = 0; 846561ec64aSLinus Torvalds int sysctl_protected_hardlinks __read_mostly = 0; 847800179c9SKees Cook 848800179c9SKees Cook /** 849800179c9SKees Cook * may_follow_link - Check symlink following for unsafe situations 85055852635SRandy Dunlap * @nd: nameidata pathwalk data 851800179c9SKees Cook * 852800179c9SKees Cook * In the case of the sysctl_protected_symlinks sysctl being enabled, 853800179c9SKees Cook * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is 854800179c9SKees Cook * in a sticky world-writable directory. This is to protect privileged 855800179c9SKees Cook * processes from failing races against path names that may change out 856800179c9SKees Cook * from under them by way of other users creating malicious symlinks. 857800179c9SKees Cook * It will permit symlinks to be followed only when outside a sticky 858800179c9SKees Cook * world-writable directory, or when the uid of the symlink and follower 859800179c9SKees Cook * match, or when the directory owner matches the symlink's owner. 860800179c9SKees Cook * 861800179c9SKees Cook * Returns 0 if following the symlink is allowed, -ve on error. 862800179c9SKees Cook */ 863fec2fa24SAl Viro static inline int may_follow_link(struct nameidata *nd) 864800179c9SKees Cook { 865800179c9SKees Cook const struct inode *inode; 866800179c9SKees Cook const struct inode *parent; 867800179c9SKees Cook 868800179c9SKees Cook if (!sysctl_protected_symlinks) 869800179c9SKees Cook return 0; 870800179c9SKees Cook 871800179c9SKees Cook /* Allowed if owner and follower match. */ 872237d8b32SAl Viro inode = nd->stack[0].inode; 87381abe27bSEric W. Biederman if (uid_eq(current_cred()->fsuid, inode->i_uid)) 874800179c9SKees Cook return 0; 875800179c9SKees Cook 876800179c9SKees Cook /* Allowed if parent directory not sticky and world-writable. */ 877800179c9SKees Cook parent = nd->path.dentry->d_inode; 878800179c9SKees Cook if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) 879800179c9SKees Cook return 0; 880800179c9SKees Cook 881800179c9SKees Cook /* Allowed if parent directory and link owner match. */ 88281abe27bSEric W. Biederman if (uid_eq(parent->i_uid, inode->i_uid)) 883800179c9SKees Cook return 0; 884800179c9SKees Cook 88531956502SAl Viro if (nd->flags & LOOKUP_RCU) 88631956502SAl Viro return -ECHILD; 88731956502SAl Viro 8881cf2665bSAl Viro audit_log_link_denied("follow_link", &nd->stack[0].link); 889800179c9SKees Cook return -EACCES; 890800179c9SKees Cook } 891800179c9SKees Cook 892800179c9SKees Cook /** 893800179c9SKees Cook * safe_hardlink_source - Check for safe hardlink conditions 894800179c9SKees Cook * @inode: the source inode to hardlink from 895800179c9SKees Cook * 896800179c9SKees Cook * Return false if at least one of the following conditions: 897800179c9SKees Cook * - inode is not a regular file 898800179c9SKees Cook * - inode is setuid 899800179c9SKees Cook * - inode is setgid and group-exec 900800179c9SKees Cook * - access failure for read and write 901800179c9SKees Cook * 902800179c9SKees Cook * Otherwise returns true. 903800179c9SKees Cook */ 904800179c9SKees Cook static bool safe_hardlink_source(struct inode *inode) 905800179c9SKees Cook { 906800179c9SKees Cook umode_t mode = inode->i_mode; 907800179c9SKees Cook 908800179c9SKees Cook /* Special files should not get pinned to the filesystem. */ 909800179c9SKees Cook if (!S_ISREG(mode)) 910800179c9SKees Cook return false; 911800179c9SKees Cook 912800179c9SKees Cook /* Setuid files should not get pinned to the filesystem. */ 913800179c9SKees Cook if (mode & S_ISUID) 914800179c9SKees Cook return false; 915800179c9SKees Cook 916800179c9SKees Cook /* Executable setgid files should not get pinned to the filesystem. */ 917800179c9SKees Cook if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) 918800179c9SKees Cook return false; 919800179c9SKees Cook 920800179c9SKees Cook /* Hardlinking to unreadable or unwritable sources is dangerous. */ 921800179c9SKees Cook if (inode_permission(inode, MAY_READ | MAY_WRITE)) 922800179c9SKees Cook return false; 923800179c9SKees Cook 924800179c9SKees Cook return true; 925800179c9SKees Cook } 926800179c9SKees Cook 927800179c9SKees Cook /** 928800179c9SKees Cook * may_linkat - Check permissions for creating a hardlink 929800179c9SKees Cook * @link: the source to hardlink from 930800179c9SKees Cook * 931800179c9SKees Cook * Block hardlink when all of: 932800179c9SKees Cook * - sysctl_protected_hardlinks enabled 933800179c9SKees Cook * - fsuid does not match inode 934800179c9SKees Cook * - hardlink source is unsafe (see safe_hardlink_source() above) 935800179c9SKees Cook * - not CAP_FOWNER 936800179c9SKees Cook * 937800179c9SKees Cook * Returns 0 if successful, -ve on error. 938800179c9SKees Cook */ 939800179c9SKees Cook static int may_linkat(struct path *link) 940800179c9SKees Cook { 941800179c9SKees Cook const struct cred *cred; 942800179c9SKees Cook struct inode *inode; 943800179c9SKees Cook 944800179c9SKees Cook if (!sysctl_protected_hardlinks) 945800179c9SKees Cook return 0; 946800179c9SKees Cook 947800179c9SKees Cook cred = current_cred(); 948800179c9SKees Cook inode = link->dentry->d_inode; 949800179c9SKees Cook 950800179c9SKees Cook /* Source inode owner (or CAP_FOWNER) can hardlink all they like, 951800179c9SKees Cook * otherwise, it must be a safe source. 952800179c9SKees Cook */ 95381abe27bSEric W. Biederman if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) || 954800179c9SKees Cook capable(CAP_FOWNER)) 955800179c9SKees Cook return 0; 956800179c9SKees Cook 957a51d9eaaSKees Cook audit_log_link_denied("linkat", link); 958800179c9SKees Cook return -EPERM; 959800179c9SKees Cook } 960800179c9SKees Cook 9613b2e7f75SAl Viro static __always_inline 9623b2e7f75SAl Viro const char *get_link(struct nameidata *nd) 9631da177e4SLinus Torvalds { 964ab104923SAl Viro struct saved *last = nd->stack + nd->depth - 1; 9651cf2665bSAl Viro struct dentry *dentry = last->link.dentry; 966237d8b32SAl Viro struct inode *inode = last->inode; 9676d7b5aaeSAl Viro int error; 9680a959df5SAl Viro const char *res; 9691da177e4SLinus Torvalds 9708fa9dd24SNeilBrown if (!(nd->flags & LOOKUP_RCU)) { 9718fa9dd24SNeilBrown touch_atime(&last->link); 9728fa9dd24SNeilBrown cond_resched(); 9738fa9dd24SNeilBrown } else if (atime_needs_update(&last->link, inode)) { 9748fa9dd24SNeilBrown if (unlikely(unlazy_walk(nd, NULL, 0))) 9758fa9dd24SNeilBrown return ERR_PTR(-ECHILD); 9768fa9dd24SNeilBrown touch_atime(&last->link); 9778fa9dd24SNeilBrown } 9788fa9dd24SNeilBrown 979bda0be7aSNeilBrown error = security_inode_follow_link(dentry, inode, 980bda0be7aSNeilBrown nd->flags & LOOKUP_RCU); 981bda0be7aSNeilBrown if (unlikely(error)) 9826920a440SAl Viro return ERR_PTR(error); 98336f3b4f6SAl Viro 98486acdca1SAl Viro nd->last_type = LAST_BIND; 985d4dee48bSAl Viro res = inode->i_link; 986d4dee48bSAl Viro if (!res) { 9878c1b4566SAl Viro if (nd->flags & LOOKUP_RCU) { 9888c1b4566SAl Viro if (unlikely(unlazy_walk(nd, NULL, 0))) 9898c1b4566SAl Viro return ERR_PTR(-ECHILD); 9908c1b4566SAl Viro } 9916e77137bSAl Viro res = inode->i_op->follow_link(dentry, &last->cookie); 992fab51e8aSAl Viro if (IS_ERR_OR_NULL(res)) { 9936920a440SAl Viro last->cookie = NULL; 994fab51e8aSAl Viro return res; 9950a959df5SAl Viro } 996fab51e8aSAl Viro } 997fab51e8aSAl Viro if (*res == '/') { 9988c1b4566SAl Viro if (nd->flags & LOOKUP_RCU) { 9998f47a016SAl Viro struct dentry *d; 10008f47a016SAl Viro if (!nd->root.mnt) 10018f47a016SAl Viro set_root_rcu(nd); 10028f47a016SAl Viro nd->path = nd->root; 10038f47a016SAl Viro d = nd->path.dentry; 10048f47a016SAl Viro nd->inode = d->d_inode; 10058f47a016SAl Viro nd->seq = nd->root_seq; 10068f47a016SAl Viro if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq))) 10078c1b4566SAl Viro return ERR_PTR(-ECHILD); 10088f47a016SAl Viro } else { 1009fab51e8aSAl Viro if (!nd->root.mnt) 1010fab51e8aSAl Viro set_root(nd); 1011fab51e8aSAl Viro path_put(&nd->path); 1012fab51e8aSAl Viro nd->path = nd->root; 1013fab51e8aSAl Viro path_get(&nd->root); 1014fab51e8aSAl Viro nd->inode = nd->path.dentry->d_inode; 10158f47a016SAl Viro } 1016fab51e8aSAl Viro nd->flags |= LOOKUP_JUMPED; 1017fab51e8aSAl Viro while (unlikely(*++res == '/')) 1018fab51e8aSAl Viro ; 1019fab51e8aSAl Viro } 1020fab51e8aSAl Viro if (!*res) 1021fab51e8aSAl Viro res = NULL; 10220a959df5SAl Viro return res; 10230a959df5SAl Viro } 10246d7b5aaeSAl Viro 1025f015f126SDavid Howells /* 1026f015f126SDavid Howells * follow_up - Find the mountpoint of path's vfsmount 1027f015f126SDavid Howells * 1028f015f126SDavid Howells * Given a path, find the mountpoint of its source file system. 1029f015f126SDavid Howells * Replace @path with the path of the mountpoint in the parent mount. 1030f015f126SDavid Howells * Up is towards /. 1031f015f126SDavid Howells * 1032f015f126SDavid Howells * Return 1 if we went up a level and 0 if we were already at the 1033f015f126SDavid Howells * root. 1034f015f126SDavid Howells */ 1035bab77ebfSAl Viro int follow_up(struct path *path) 10361da177e4SLinus Torvalds { 10370714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 10380714a533SAl Viro struct mount *parent; 10391da177e4SLinus Torvalds struct dentry *mountpoint; 104099b7db7bSNick Piggin 104148a066e7SAl Viro read_seqlock_excl(&mount_lock); 10420714a533SAl Viro parent = mnt->mnt_parent; 10433c0a6163SAl Viro if (parent == mnt) { 104448a066e7SAl Viro read_sequnlock_excl(&mount_lock); 10451da177e4SLinus Torvalds return 0; 10461da177e4SLinus Torvalds } 10470714a533SAl Viro mntget(&parent->mnt); 1048a73324daSAl Viro mountpoint = dget(mnt->mnt_mountpoint); 104948a066e7SAl Viro read_sequnlock_excl(&mount_lock); 1050bab77ebfSAl Viro dput(path->dentry); 1051bab77ebfSAl Viro path->dentry = mountpoint; 1052bab77ebfSAl Viro mntput(path->mnt); 10530714a533SAl Viro path->mnt = &parent->mnt; 10541da177e4SLinus Torvalds return 1; 10551da177e4SLinus Torvalds } 10564d359507SAl Viro EXPORT_SYMBOL(follow_up); 10571da177e4SLinus Torvalds 1058b5c84bf6SNick Piggin /* 10599875cf80SDavid Howells * Perform an automount 10609875cf80SDavid Howells * - return -EISDIR to tell follow_managed() to stop and return the path we 10619875cf80SDavid Howells * were called with. 10621da177e4SLinus Torvalds */ 1063756daf26SNeilBrown static int follow_automount(struct path *path, struct nameidata *nd, 10649875cf80SDavid Howells bool *need_mntput) 106531e6b01fSNick Piggin { 10669875cf80SDavid Howells struct vfsmount *mnt; 1067ea5b778aSDavid Howells int err; 10689875cf80SDavid Howells 10699875cf80SDavid Howells if (!path->dentry->d_op || !path->dentry->d_op->d_automount) 10709875cf80SDavid Howells return -EREMOTE; 10719875cf80SDavid Howells 10720ec26fd0SMiklos Szeredi /* We don't want to mount if someone's just doing a stat - 10730ec26fd0SMiklos Szeredi * unless they're stat'ing a directory and appended a '/' to 10740ec26fd0SMiklos Szeredi * the name. 10750ec26fd0SMiklos Szeredi * 10760ec26fd0SMiklos Szeredi * We do, however, want to mount if someone wants to open or 10775a30d8a2SDavid Howells * create a file of any type under the mountpoint, wants to 10780ec26fd0SMiklos Szeredi * traverse through the mountpoint or wants to open the 10790ec26fd0SMiklos Szeredi * mounted directory. Also, autofs may mark negative dentries 10800ec26fd0SMiklos Szeredi * as being automount points. These will need the attentions 10810ec26fd0SMiklos Szeredi * of the daemon to instantiate them before they can be used. 10825a30d8a2SDavid Howells */ 1083756daf26SNeilBrown if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | 1084d94c177bSLinus Torvalds LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && 10855a30d8a2SDavid Howells path->dentry->d_inode) 10869875cf80SDavid Howells return -EISDIR; 10870ec26fd0SMiklos Szeredi 1088756daf26SNeilBrown nd->total_link_count++; 1089756daf26SNeilBrown if (nd->total_link_count >= 40) 10909875cf80SDavid Howells return -ELOOP; 10919875cf80SDavid Howells 10929875cf80SDavid Howells mnt = path->dentry->d_op->d_automount(path); 10939875cf80SDavid Howells if (IS_ERR(mnt)) { 10949875cf80SDavid Howells /* 10959875cf80SDavid Howells * The filesystem is allowed to return -EISDIR here to indicate 10969875cf80SDavid Howells * it doesn't want to automount. For instance, autofs would do 10979875cf80SDavid Howells * this so that its userspace daemon can mount on this dentry. 10989875cf80SDavid Howells * 10999875cf80SDavid Howells * However, we can only permit this if it's a terminal point in 11009875cf80SDavid Howells * the path being looked up; if it wasn't then the remainder of 11019875cf80SDavid Howells * the path is inaccessible and we should say so. 11029875cf80SDavid Howells */ 1103756daf26SNeilBrown if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT)) 11049875cf80SDavid Howells return -EREMOTE; 11059875cf80SDavid Howells return PTR_ERR(mnt); 110631e6b01fSNick Piggin } 1107ea5b778aSDavid Howells 11089875cf80SDavid Howells if (!mnt) /* mount collision */ 11099875cf80SDavid Howells return 0; 11109875cf80SDavid Howells 11118aef1884SAl Viro if (!*need_mntput) { 11128aef1884SAl Viro /* lock_mount() may release path->mnt on error */ 11138aef1884SAl Viro mntget(path->mnt); 11148aef1884SAl Viro *need_mntput = true; 11158aef1884SAl Viro } 111619a167afSAl Viro err = finish_automount(mnt, path); 1117ea5b778aSDavid Howells 1118ea5b778aSDavid Howells switch (err) { 1119ea5b778aSDavid Howells case -EBUSY: 1120ea5b778aSDavid Howells /* Someone else made a mount here whilst we were busy */ 112119a167afSAl Viro return 0; 1122ea5b778aSDavid Howells case 0: 11238aef1884SAl Viro path_put(path); 11249875cf80SDavid Howells path->mnt = mnt; 11259875cf80SDavid Howells path->dentry = dget(mnt->mnt_root); 11269875cf80SDavid Howells return 0; 112719a167afSAl Viro default: 112819a167afSAl Viro return err; 11299875cf80SDavid Howells } 113019a167afSAl Viro 1131ea5b778aSDavid Howells } 11329875cf80SDavid Howells 11339875cf80SDavid Howells /* 11349875cf80SDavid Howells * Handle a dentry that is managed in some way. 1135cc53ce53SDavid Howells * - Flagged for transit management (autofs) 11369875cf80SDavid Howells * - Flagged as mountpoint 11379875cf80SDavid Howells * - Flagged as automount point 11389875cf80SDavid Howells * 11399875cf80SDavid Howells * This may only be called in refwalk mode. 11409875cf80SDavid Howells * 11419875cf80SDavid Howells * Serialization is taken care of in namespace.c 11429875cf80SDavid Howells */ 1143756daf26SNeilBrown static int follow_managed(struct path *path, struct nameidata *nd) 11449875cf80SDavid Howells { 11458aef1884SAl Viro struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */ 11469875cf80SDavid Howells unsigned managed; 11479875cf80SDavid Howells bool need_mntput = false; 11488aef1884SAl Viro int ret = 0; 11499875cf80SDavid Howells 11509875cf80SDavid Howells /* Given that we're not holding a lock here, we retain the value in a 11519875cf80SDavid Howells * local variable for each dentry as we look at it so that we don't see 11529875cf80SDavid Howells * the components of that value change under us */ 11539875cf80SDavid Howells while (managed = ACCESS_ONCE(path->dentry->d_flags), 11549875cf80SDavid Howells managed &= DCACHE_MANAGED_DENTRY, 11559875cf80SDavid Howells unlikely(managed != 0)) { 1156cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1157cc53ce53SDavid Howells * being held. */ 1158cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 1159cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1160cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 11611aed3e42SAl Viro ret = path->dentry->d_op->d_manage(path->dentry, false); 1162cc53ce53SDavid Howells if (ret < 0) 11638aef1884SAl Viro break; 1164cc53ce53SDavid Howells } 1165cc53ce53SDavid Howells 11669875cf80SDavid Howells /* Transit to a mounted filesystem. */ 11679875cf80SDavid Howells if (managed & DCACHE_MOUNTED) { 11689875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 11699875cf80SDavid Howells if (mounted) { 11709875cf80SDavid Howells dput(path->dentry); 11719875cf80SDavid Howells if (need_mntput) 1172463ffb2eSAl Viro mntput(path->mnt); 1173463ffb2eSAl Viro path->mnt = mounted; 1174463ffb2eSAl Viro path->dentry = dget(mounted->mnt_root); 11759875cf80SDavid Howells need_mntput = true; 11769875cf80SDavid Howells continue; 1177463ffb2eSAl Viro } 1178463ffb2eSAl Viro 11799875cf80SDavid Howells /* Something is mounted on this dentry in another 11809875cf80SDavid Howells * namespace and/or whatever was mounted there in this 118148a066e7SAl Viro * namespace got unmounted before lookup_mnt() could 118248a066e7SAl Viro * get it */ 11831da177e4SLinus Torvalds } 11849875cf80SDavid Howells 11859875cf80SDavid Howells /* Handle an automount point */ 11869875cf80SDavid Howells if (managed & DCACHE_NEED_AUTOMOUNT) { 1187756daf26SNeilBrown ret = follow_automount(path, nd, &need_mntput); 11889875cf80SDavid Howells if (ret < 0) 11898aef1884SAl Viro break; 11909875cf80SDavid Howells continue; 11919875cf80SDavid Howells } 11929875cf80SDavid Howells 11939875cf80SDavid Howells /* We didn't change the current path point */ 11949875cf80SDavid Howells break; 11959875cf80SDavid Howells } 11968aef1884SAl Viro 11978aef1884SAl Viro if (need_mntput && path->mnt == mnt) 11988aef1884SAl Viro mntput(path->mnt); 11998aef1884SAl Viro if (ret == -EISDIR) 12008aef1884SAl Viro ret = 0; 12018402752eSAl Viro if (need_mntput) 12028402752eSAl Viro nd->flags |= LOOKUP_JUMPED; 12038402752eSAl Viro if (unlikely(ret < 0)) 12048402752eSAl Viro path_put_conditional(path, nd); 12058402752eSAl Viro return ret; 12061da177e4SLinus Torvalds } 12071da177e4SLinus Torvalds 1208cc53ce53SDavid Howells int follow_down_one(struct path *path) 12091da177e4SLinus Torvalds { 12101da177e4SLinus Torvalds struct vfsmount *mounted; 12111da177e4SLinus Torvalds 12121c755af4SAl Viro mounted = lookup_mnt(path); 12131da177e4SLinus Torvalds if (mounted) { 12149393bd07SAl Viro dput(path->dentry); 12159393bd07SAl Viro mntput(path->mnt); 12169393bd07SAl Viro path->mnt = mounted; 12179393bd07SAl Viro path->dentry = dget(mounted->mnt_root); 12181da177e4SLinus Torvalds return 1; 12191da177e4SLinus Torvalds } 12201da177e4SLinus Torvalds return 0; 12211da177e4SLinus Torvalds } 12224d359507SAl Viro EXPORT_SYMBOL(follow_down_one); 12231da177e4SLinus Torvalds 1224b8faf035SNeilBrown static inline int managed_dentry_rcu(struct dentry *dentry) 122562a7375eSIan Kent { 1226b8faf035SNeilBrown return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ? 1227b8faf035SNeilBrown dentry->d_op->d_manage(dentry, true) : 0; 122862a7375eSIan Kent } 122962a7375eSIan Kent 12309875cf80SDavid Howells /* 1231287548e4SAl Viro * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if 1232287548e4SAl Viro * we meet a managed dentry that would need blocking. 12339875cf80SDavid Howells */ 12349875cf80SDavid Howells static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, 1235254cf582SAl Viro struct inode **inode, unsigned *seqp) 12369875cf80SDavid Howells { 123762a7375eSIan Kent for (;;) { 1238c7105365SAl Viro struct mount *mounted; 123962a7375eSIan Kent /* 124062a7375eSIan Kent * Don't forget we might have a non-mountpoint managed dentry 124162a7375eSIan Kent * that wants to block transit. 124262a7375eSIan Kent */ 1243b8faf035SNeilBrown switch (managed_dentry_rcu(path->dentry)) { 1244b8faf035SNeilBrown case -ECHILD: 1245b8faf035SNeilBrown default: 1246ab90911fSDavid Howells return false; 1247b8faf035SNeilBrown case -EISDIR: 1248b8faf035SNeilBrown return true; 1249b8faf035SNeilBrown case 0: 1250b8faf035SNeilBrown break; 1251b8faf035SNeilBrown } 125262a7375eSIan Kent 125362a7375eSIan Kent if (!d_mountpoint(path->dentry)) 1254b8faf035SNeilBrown return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); 125562a7375eSIan Kent 1256474279dcSAl Viro mounted = __lookup_mnt(path->mnt, path->dentry); 12579875cf80SDavid Howells if (!mounted) 12589875cf80SDavid Howells break; 1259c7105365SAl Viro path->mnt = &mounted->mnt; 1260c7105365SAl Viro path->dentry = mounted->mnt.mnt_root; 1261a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 1262254cf582SAl Viro *seqp = read_seqcount_begin(&path->dentry->d_seq); 126359430262SLinus Torvalds /* 126459430262SLinus Torvalds * Update the inode too. We don't need to re-check the 126559430262SLinus Torvalds * dentry sequence number here after this d_inode read, 126659430262SLinus Torvalds * because a mount-point is always pinned. 126759430262SLinus Torvalds */ 126859430262SLinus Torvalds *inode = path->dentry->d_inode; 12699875cf80SDavid Howells } 1270f5be3e29SAl Viro return !read_seqretry(&mount_lock, nd->m_seq) && 1271b8faf035SNeilBrown !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); 1272287548e4SAl Viro } 1273287548e4SAl Viro 127431e6b01fSNick Piggin static int follow_dotdot_rcu(struct nameidata *nd) 127531e6b01fSNick Piggin { 12764023bfc9SAl Viro struct inode *inode = nd->inode; 12777bd88377SAl Viro if (!nd->root.mnt) 127831e6b01fSNick Piggin set_root_rcu(nd); 127931e6b01fSNick Piggin 128031e6b01fSNick Piggin while (1) { 1281aed434adSAl Viro if (path_equal(&nd->path, &nd->root)) 128231e6b01fSNick Piggin break; 128331e6b01fSNick Piggin if (nd->path.dentry != nd->path.mnt->mnt_root) { 128431e6b01fSNick Piggin struct dentry *old = nd->path.dentry; 128531e6b01fSNick Piggin struct dentry *parent = old->d_parent; 128631e6b01fSNick Piggin unsigned seq; 128731e6b01fSNick Piggin 12884023bfc9SAl Viro inode = parent->d_inode; 128931e6b01fSNick Piggin seq = read_seqcount_begin(&parent->d_seq); 1290aed434adSAl Viro if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq))) 1291aed434adSAl Viro return -ECHILD; 129231e6b01fSNick Piggin nd->path.dentry = parent; 129331e6b01fSNick Piggin nd->seq = seq; 129431e6b01fSNick Piggin break; 1295aed434adSAl Viro } else { 1296aed434adSAl Viro struct mount *mnt = real_mount(nd->path.mnt); 1297aed434adSAl Viro struct mount *mparent = mnt->mnt_parent; 1298aed434adSAl Viro struct dentry *mountpoint = mnt->mnt_mountpoint; 1299aed434adSAl Viro struct inode *inode2 = mountpoint->d_inode; 1300aed434adSAl Viro unsigned seq = read_seqcount_begin(&mountpoint->d_seq); 1301aed434adSAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1302aed434adSAl Viro return -ECHILD; 1303aed434adSAl Viro if (&mparent->mnt == nd->path.mnt) 130431e6b01fSNick Piggin break; 1305aed434adSAl Viro /* we know that mountpoint was pinned */ 1306aed434adSAl Viro nd->path.dentry = mountpoint; 1307aed434adSAl Viro nd->path.mnt = &mparent->mnt; 1308aed434adSAl Viro inode = inode2; 1309aed434adSAl Viro nd->seq = seq; 131031e6b01fSNick Piggin } 1311aed434adSAl Viro } 1312aed434adSAl Viro while (unlikely(d_mountpoint(nd->path.dentry))) { 1313b37199e6SAl Viro struct mount *mounted; 1314b37199e6SAl Viro mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry); 1315aed434adSAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1316aed434adSAl Viro return -ECHILD; 1317b37199e6SAl Viro if (!mounted) 1318b37199e6SAl Viro break; 1319b37199e6SAl Viro nd->path.mnt = &mounted->mnt; 1320b37199e6SAl Viro nd->path.dentry = mounted->mnt.mnt_root; 13214023bfc9SAl Viro inode = nd->path.dentry->d_inode; 1322b37199e6SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 1323b37199e6SAl Viro } 13244023bfc9SAl Viro nd->inode = inode; 132531e6b01fSNick Piggin return 0; 132631e6b01fSNick Piggin } 132731e6b01fSNick Piggin 13289875cf80SDavid Howells /* 1329cc53ce53SDavid Howells * Follow down to the covering mount currently visible to userspace. At each 1330cc53ce53SDavid Howells * point, the filesystem owning that dentry may be queried as to whether the 1331cc53ce53SDavid Howells * caller is permitted to proceed or not. 1332cc53ce53SDavid Howells */ 13337cc90cc3SAl Viro int follow_down(struct path *path) 1334cc53ce53SDavid Howells { 1335cc53ce53SDavid Howells unsigned managed; 1336cc53ce53SDavid Howells int ret; 1337cc53ce53SDavid Howells 1338cc53ce53SDavid Howells while (managed = ACCESS_ONCE(path->dentry->d_flags), 1339cc53ce53SDavid Howells unlikely(managed & DCACHE_MANAGED_DENTRY)) { 1340cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1341cc53ce53SDavid Howells * being held. 1342cc53ce53SDavid Howells * 1343cc53ce53SDavid Howells * We indicate to the filesystem if someone is trying to mount 1344cc53ce53SDavid Howells * something here. This gives autofs the chance to deny anyone 1345cc53ce53SDavid Howells * other than its daemon the right to mount on its 1346cc53ce53SDavid Howells * superstructure. 1347cc53ce53SDavid Howells * 1348cc53ce53SDavid Howells * The filesystem may sleep at this point. 1349cc53ce53SDavid Howells */ 1350cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 1351cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1352cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 1353ab90911fSDavid Howells ret = path->dentry->d_op->d_manage( 13541aed3e42SAl Viro path->dentry, false); 1355cc53ce53SDavid Howells if (ret < 0) 1356cc53ce53SDavid Howells return ret == -EISDIR ? 0 : ret; 1357cc53ce53SDavid Howells } 1358cc53ce53SDavid Howells 1359cc53ce53SDavid Howells /* Transit to a mounted filesystem. */ 1360cc53ce53SDavid Howells if (managed & DCACHE_MOUNTED) { 1361cc53ce53SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 1362cc53ce53SDavid Howells if (!mounted) 1363cc53ce53SDavid Howells break; 1364cc53ce53SDavid Howells dput(path->dentry); 1365cc53ce53SDavid Howells mntput(path->mnt); 1366cc53ce53SDavid Howells path->mnt = mounted; 1367cc53ce53SDavid Howells path->dentry = dget(mounted->mnt_root); 1368cc53ce53SDavid Howells continue; 1369cc53ce53SDavid Howells } 1370cc53ce53SDavid Howells 1371cc53ce53SDavid Howells /* Don't handle automount points here */ 1372cc53ce53SDavid Howells break; 1373cc53ce53SDavid Howells } 1374cc53ce53SDavid Howells return 0; 1375cc53ce53SDavid Howells } 13764d359507SAl Viro EXPORT_SYMBOL(follow_down); 1377cc53ce53SDavid Howells 1378cc53ce53SDavid Howells /* 13799875cf80SDavid Howells * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() 13809875cf80SDavid Howells */ 13819875cf80SDavid Howells static void follow_mount(struct path *path) 13829875cf80SDavid Howells { 13839875cf80SDavid Howells while (d_mountpoint(path->dentry)) { 13849875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 13859875cf80SDavid Howells if (!mounted) 13869875cf80SDavid Howells break; 13879875cf80SDavid Howells dput(path->dentry); 13889875cf80SDavid Howells mntput(path->mnt); 13899875cf80SDavid Howells path->mnt = mounted; 13909875cf80SDavid Howells path->dentry = dget(mounted->mnt_root); 13919875cf80SDavid Howells } 13929875cf80SDavid Howells } 13939875cf80SDavid Howells 139431e6b01fSNick Piggin static void follow_dotdot(struct nameidata *nd) 13951da177e4SLinus Torvalds { 13967bd88377SAl Viro if (!nd->root.mnt) 13972a737871SAl Viro set_root(nd); 1398e518ddb7SAndreas Mohr 13991da177e4SLinus Torvalds while(1) { 14004ac91378SJan Blunck struct dentry *old = nd->path.dentry; 14011da177e4SLinus Torvalds 14022a737871SAl Viro if (nd->path.dentry == nd->root.dentry && 14032a737871SAl Viro nd->path.mnt == nd->root.mnt) { 14041da177e4SLinus Torvalds break; 14051da177e4SLinus Torvalds } 14064ac91378SJan Blunck if (nd->path.dentry != nd->path.mnt->mnt_root) { 14073088dd70SAl Viro /* rare case of legitimate dget_parent()... */ 14083088dd70SAl Viro nd->path.dentry = dget_parent(nd->path.dentry); 14091da177e4SLinus Torvalds dput(old); 14101da177e4SLinus Torvalds break; 14111da177e4SLinus Torvalds } 14123088dd70SAl Viro if (!follow_up(&nd->path)) 14131da177e4SLinus Torvalds break; 14141da177e4SLinus Torvalds } 141579ed0226SAl Viro follow_mount(&nd->path); 141631e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 14171da177e4SLinus Torvalds } 14181da177e4SLinus Torvalds 14191da177e4SLinus Torvalds /* 1420bad61189SMiklos Szeredi * This looks up the name in dcache, possibly revalidates the old dentry and 1421bad61189SMiklos Szeredi * allocates a new one if not found or not valid. In the need_lookup argument 1422bad61189SMiklos Szeredi * returns whether i_op->lookup is necessary. 1423bad61189SMiklos Szeredi * 1424bad61189SMiklos Szeredi * dir->d_inode->i_mutex must be held 1425baa03890SNick Piggin */ 1426bad61189SMiklos Szeredi static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir, 1427201f956eSAl Viro unsigned int flags, bool *need_lookup) 1428baa03890SNick Piggin { 1429baa03890SNick Piggin struct dentry *dentry; 1430bad61189SMiklos Szeredi int error; 1431baa03890SNick Piggin 1432bad61189SMiklos Szeredi *need_lookup = false; 1433bad61189SMiklos Szeredi dentry = d_lookup(dir, name); 1434bad61189SMiklos Szeredi if (dentry) { 143539e3c955SJeff Layton if (dentry->d_flags & DCACHE_OP_REVALIDATE) { 1436201f956eSAl Viro error = d_revalidate(dentry, flags); 1437bad61189SMiklos Szeredi if (unlikely(error <= 0)) { 1438bad61189SMiklos Szeredi if (error < 0) { 1439bad61189SMiklos Szeredi dput(dentry); 1440bad61189SMiklos Szeredi return ERR_PTR(error); 14415542aa2fSEric W. Biederman } else { 14425542aa2fSEric W. Biederman d_invalidate(dentry); 1443bad61189SMiklos Szeredi dput(dentry); 1444bad61189SMiklos Szeredi dentry = NULL; 1445bad61189SMiklos Szeredi } 1446bad61189SMiklos Szeredi } 1447bad61189SMiklos Szeredi } 1448bad61189SMiklos Szeredi } 1449baa03890SNick Piggin 1450bad61189SMiklos Szeredi if (!dentry) { 1451bad61189SMiklos Szeredi dentry = d_alloc(dir, name); 1452baa03890SNick Piggin if (unlikely(!dentry)) 1453baa03890SNick Piggin return ERR_PTR(-ENOMEM); 1454baa03890SNick Piggin 1455bad61189SMiklos Szeredi *need_lookup = true; 1456baa03890SNick Piggin } 1457baa03890SNick Piggin return dentry; 1458baa03890SNick Piggin } 1459baa03890SNick Piggin 1460baa03890SNick Piggin /* 146113a2c3beSJ. Bruce Fields * Call i_op->lookup on the dentry. The dentry must be negative and 146213a2c3beSJ. Bruce Fields * unhashed. 1463bad61189SMiklos Szeredi * 1464bad61189SMiklos Szeredi * dir->d_inode->i_mutex must be held 146544396f4bSJosef Bacik */ 1466bad61189SMiklos Szeredi static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry, 146772bd866aSAl Viro unsigned int flags) 146844396f4bSJosef Bacik { 146944396f4bSJosef Bacik struct dentry *old; 147044396f4bSJosef Bacik 147144396f4bSJosef Bacik /* Don't create child dentry for a dead directory. */ 1472bad61189SMiklos Szeredi if (unlikely(IS_DEADDIR(dir))) { 1473e188dc02SMiklos Szeredi dput(dentry); 147444396f4bSJosef Bacik return ERR_PTR(-ENOENT); 1475e188dc02SMiklos Szeredi } 147644396f4bSJosef Bacik 147772bd866aSAl Viro old = dir->i_op->lookup(dir, dentry, flags); 147844396f4bSJosef Bacik if (unlikely(old)) { 147944396f4bSJosef Bacik dput(dentry); 148044396f4bSJosef Bacik dentry = old; 148144396f4bSJosef Bacik } 148244396f4bSJosef Bacik return dentry; 148344396f4bSJosef Bacik } 148444396f4bSJosef Bacik 1485a3255546SAl Viro static struct dentry *__lookup_hash(struct qstr *name, 148672bd866aSAl Viro struct dentry *base, unsigned int flags) 1487a3255546SAl Viro { 1488bad61189SMiklos Szeredi bool need_lookup; 1489a3255546SAl Viro struct dentry *dentry; 1490a3255546SAl Viro 149172bd866aSAl Viro dentry = lookup_dcache(name, base, flags, &need_lookup); 1492bad61189SMiklos Szeredi if (!need_lookup) 1493a3255546SAl Viro return dentry; 1494bad61189SMiklos Szeredi 149572bd866aSAl Viro return lookup_real(base->d_inode, dentry, flags); 1496a3255546SAl Viro } 1497a3255546SAl Viro 149844396f4bSJosef Bacik /* 14991da177e4SLinus Torvalds * It's more convoluted than I'd like it to be, but... it's still fairly 15001da177e4SLinus Torvalds * small and for now I'd prefer to have fast path as straight as possible. 15011da177e4SLinus Torvalds * It _is_ time-critical. 15021da177e4SLinus Torvalds */ 1503e97cdc87SAl Viro static int lookup_fast(struct nameidata *nd, 1504254cf582SAl Viro struct path *path, struct inode **inode, 1505254cf582SAl Viro unsigned *seqp) 15061da177e4SLinus Torvalds { 15074ac91378SJan Blunck struct vfsmount *mnt = nd->path.mnt; 150831e6b01fSNick Piggin struct dentry *dentry, *parent = nd->path.dentry; 15095a18fff2SAl Viro int need_reval = 1; 15105a18fff2SAl Viro int status = 1; 15119875cf80SDavid Howells int err; 15129875cf80SDavid Howells 15133cac260aSAl Viro /* 1514b04f784eSNick Piggin * Rename seqlock is not required here because in the off chance 1515b04f784eSNick Piggin * of a false negative due to a concurrent rename, we're going to 1516b04f784eSNick Piggin * do the non-racy lookup, below. 1517b04f784eSNick Piggin */ 151831e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) { 151931e6b01fSNick Piggin unsigned seq; 1520766c4cbfSAl Viro bool negative; 1521da53be12SLinus Torvalds dentry = __d_lookup_rcu(parent, &nd->last, &seq); 15225a18fff2SAl Viro if (!dentry) 15235a18fff2SAl Viro goto unlazy; 15245a18fff2SAl Viro 152512f8ad4bSLinus Torvalds /* 152612f8ad4bSLinus Torvalds * This sequence count validates that the inode matches 152712f8ad4bSLinus Torvalds * the dentry name information from lookup. 152812f8ad4bSLinus Torvalds */ 152963afdfc7SDavid Howells *inode = d_backing_inode(dentry); 1530766c4cbfSAl Viro negative = d_is_negative(dentry); 153112f8ad4bSLinus Torvalds if (read_seqcount_retry(&dentry->d_seq, seq)) 153212f8ad4bSLinus Torvalds return -ECHILD; 1533766c4cbfSAl Viro if (negative) 1534766c4cbfSAl Viro return -ENOENT; 153512f8ad4bSLinus Torvalds 153612f8ad4bSLinus Torvalds /* 153712f8ad4bSLinus Torvalds * This sequence count validates that the parent had no 153812f8ad4bSLinus Torvalds * changes while we did the lookup of the dentry above. 153912f8ad4bSLinus Torvalds * 154012f8ad4bSLinus Torvalds * The memory barrier in read_seqcount_begin of child is 154112f8ad4bSLinus Torvalds * enough, we can use __read_seqcount_retry here. 154212f8ad4bSLinus Torvalds */ 154331e6b01fSNick Piggin if (__read_seqcount_retry(&parent->d_seq, nd->seq)) 154431e6b01fSNick Piggin return -ECHILD; 15455a18fff2SAl Viro 1546254cf582SAl Viro *seqp = seq; 154724643087SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) { 15484ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 15495a18fff2SAl Viro if (unlikely(status <= 0)) { 15505a18fff2SAl Viro if (status != -ECHILD) 15515a18fff2SAl Viro need_reval = 0; 15525a18fff2SAl Viro goto unlazy; 15535a18fff2SAl Viro } 155424643087SAl Viro } 155531e6b01fSNick Piggin path->mnt = mnt; 155631e6b01fSNick Piggin path->dentry = dentry; 1557254cf582SAl Viro if (likely(__follow_mount_rcu(nd, path, inode, seqp))) 15589875cf80SDavid Howells return 0; 15595a18fff2SAl Viro unlazy: 1560254cf582SAl Viro if (unlazy_walk(nd, dentry, seq)) 15615a18fff2SAl Viro return -ECHILD; 15625a18fff2SAl Viro } else { 1563e97cdc87SAl Viro dentry = __d_lookup(parent, &nd->last); 156424643087SAl Viro } 15655a18fff2SAl Viro 156681e6f520SAl Viro if (unlikely(!dentry)) 156781e6f520SAl Viro goto need_lookup; 15685a18fff2SAl Viro 15695a18fff2SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval) 15704ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 15715a18fff2SAl Viro if (unlikely(status <= 0)) { 15725a18fff2SAl Viro if (status < 0) { 15735a18fff2SAl Viro dput(dentry); 15745a18fff2SAl Viro return status; 15755a18fff2SAl Viro } 15765542aa2fSEric W. Biederman d_invalidate(dentry); 15775a18fff2SAl Viro dput(dentry); 157881e6f520SAl Viro goto need_lookup; 15795a18fff2SAl Viro } 1580697f514dSMiklos Szeredi 1581766c4cbfSAl Viro if (unlikely(d_is_negative(dentry))) { 1582766c4cbfSAl Viro dput(dentry); 1583766c4cbfSAl Viro return -ENOENT; 1584766c4cbfSAl Viro } 15851da177e4SLinus Torvalds path->mnt = mnt; 15861da177e4SLinus Torvalds path->dentry = dentry; 1587756daf26SNeilBrown err = follow_managed(path, nd); 15888402752eSAl Viro if (likely(!err)) 158963afdfc7SDavid Howells *inode = d_backing_inode(path->dentry); 15908402752eSAl Viro return err; 159181e6f520SAl Viro 159281e6f520SAl Viro need_lookup: 1593697f514dSMiklos Szeredi return 1; 1594697f514dSMiklos Szeredi } 1595697f514dSMiklos Szeredi 1596697f514dSMiklos Szeredi /* Fast lookup failed, do it the slow way */ 1597cc2a5271SAl Viro static int lookup_slow(struct nameidata *nd, struct path *path) 1598697f514dSMiklos Szeredi { 1599697f514dSMiklos Szeredi struct dentry *dentry, *parent; 1600697f514dSMiklos Szeredi 1601697f514dSMiklos Szeredi parent = nd->path.dentry; 160281e6f520SAl Viro BUG_ON(nd->inode != parent->d_inode); 160381e6f520SAl Viro 160481e6f520SAl Viro mutex_lock(&parent->d_inode->i_mutex); 1605cc2a5271SAl Viro dentry = __lookup_hash(&nd->last, parent, nd->flags); 160681e6f520SAl Viro mutex_unlock(&parent->d_inode->i_mutex); 160781e6f520SAl Viro if (IS_ERR(dentry)) 160881e6f520SAl Viro return PTR_ERR(dentry); 1609697f514dSMiklos Szeredi path->mnt = nd->path.mnt; 1610697f514dSMiklos Szeredi path->dentry = dentry; 16118402752eSAl Viro return follow_managed(path, nd); 16121da177e4SLinus Torvalds } 16131da177e4SLinus Torvalds 161452094c8aSAl Viro static inline int may_lookup(struct nameidata *nd) 161552094c8aSAl Viro { 161652094c8aSAl Viro if (nd->flags & LOOKUP_RCU) { 16174ad5abb3SAl Viro int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); 161852094c8aSAl Viro if (err != -ECHILD) 161952094c8aSAl Viro return err; 16206e9918b7SAl Viro if (unlazy_walk(nd, NULL, 0)) 162152094c8aSAl Viro return -ECHILD; 162252094c8aSAl Viro } 16234ad5abb3SAl Viro return inode_permission(nd->inode, MAY_EXEC); 162452094c8aSAl Viro } 162552094c8aSAl Viro 16269856fa1bSAl Viro static inline int handle_dots(struct nameidata *nd, int type) 16279856fa1bSAl Viro { 16289856fa1bSAl Viro if (type == LAST_DOTDOT) { 16299856fa1bSAl Viro if (nd->flags & LOOKUP_RCU) { 163070291aecSAl Viro return follow_dotdot_rcu(nd); 16319856fa1bSAl Viro } else 16329856fa1bSAl Viro follow_dotdot(nd); 16339856fa1bSAl Viro } 16349856fa1bSAl Viro return 0; 16359856fa1bSAl Viro } 16369856fa1bSAl Viro 1637181548c0SAl Viro static int pick_link(struct nameidata *nd, struct path *link, 1638181548c0SAl Viro struct inode *inode, unsigned seq) 1639d63ff28fSAl Viro { 1640626de996SAl Viro int error; 16411cf2665bSAl Viro struct saved *last; 1642756daf26SNeilBrown if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) { 1643626de996SAl Viro path_to_nameidata(link, nd); 1644626de996SAl Viro return -ELOOP; 1645626de996SAl Viro } 1646bc40aee0SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1647cd179f44SAl Viro if (link->mnt == nd->path.mnt) 1648cd179f44SAl Viro mntget(link->mnt); 16497973387aSAl Viro } 1650626de996SAl Viro error = nd_alloc_stack(nd); 1651626de996SAl Viro if (unlikely(error)) { 1652bc40aee0SAl Viro if (error == -ECHILD) { 1653bc40aee0SAl Viro if (unlikely(unlazy_link(nd, link, seq))) 1654bc40aee0SAl Viro return -ECHILD; 1655bc40aee0SAl Viro error = nd_alloc_stack(nd); 1656bc40aee0SAl Viro } 1657bc40aee0SAl Viro if (error) { 1658cd179f44SAl Viro path_put(link); 1659626de996SAl Viro return error; 1660626de996SAl Viro } 1661bc40aee0SAl Viro } 1662626de996SAl Viro 1663ab104923SAl Viro last = nd->stack + nd->depth++; 16641cf2665bSAl Viro last->link = *link; 1665ab104923SAl Viro last->cookie = NULL; 1666181548c0SAl Viro last->inode = inode; 16670450b2d1SAl Viro last->seq = seq; 1668d63ff28fSAl Viro return 1; 1669d63ff28fSAl Viro } 1670d63ff28fSAl Viro 16713ddcd056SLinus Torvalds /* 16723ddcd056SLinus Torvalds * Do we need to follow links? We _really_ want to be able 16733ddcd056SLinus Torvalds * to do this check without having to look at inode->i_op, 16743ddcd056SLinus Torvalds * so we keep a cache of "no, this doesn't need follow_link" 16753ddcd056SLinus Torvalds * for the common case. 16763ddcd056SLinus Torvalds */ 1677254cf582SAl Viro static inline int should_follow_link(struct nameidata *nd, struct path *link, 1678181548c0SAl Viro int follow, 1679181548c0SAl Viro struct inode *inode, unsigned seq) 16803ddcd056SLinus Torvalds { 1681d63ff28fSAl Viro if (likely(!d_is_symlink(link->dentry))) 1682d63ff28fSAl Viro return 0; 1683d63ff28fSAl Viro if (!follow) 1684d63ff28fSAl Viro return 0; 1685181548c0SAl Viro return pick_link(nd, link, inode, seq); 16863ddcd056SLinus Torvalds } 16873ddcd056SLinus Torvalds 16884693a547SAl Viro enum {WALK_GET = 1, WALK_PUT = 2}; 16894693a547SAl Viro 16904693a547SAl Viro static int walk_component(struct nameidata *nd, int flags) 1691ce57dfc1SAl Viro { 1692caa85634SAl Viro struct path path; 1693ce57dfc1SAl Viro struct inode *inode; 1694254cf582SAl Viro unsigned seq; 1695ce57dfc1SAl Viro int err; 1696ce57dfc1SAl Viro /* 1697ce57dfc1SAl Viro * "." and ".." are special - ".." especially so because it has 1698ce57dfc1SAl Viro * to be able to know about the current root directory and 1699ce57dfc1SAl Viro * parent relationships. 1700ce57dfc1SAl Viro */ 17014693a547SAl Viro if (unlikely(nd->last_type != LAST_NORM)) { 17024693a547SAl Viro err = handle_dots(nd, nd->last_type); 17034693a547SAl Viro if (flags & WALK_PUT) 17044693a547SAl Viro put_link(nd); 17054693a547SAl Viro return err; 17064693a547SAl Viro } 1707254cf582SAl Viro err = lookup_fast(nd, &path, &inode, &seq); 1708ce57dfc1SAl Viro if (unlikely(err)) { 1709697f514dSMiklos Szeredi if (err < 0) 1710f0a9ba70SAl Viro return err; 1711697f514dSMiklos Szeredi 1712caa85634SAl Viro err = lookup_slow(nd, &path); 1713697f514dSMiklos Szeredi if (err < 0) 1714f0a9ba70SAl Viro return err; 1715697f514dSMiklos Szeredi 171663afdfc7SDavid Howells inode = d_backing_inode(path.dentry); 1717254cf582SAl Viro seq = 0; /* we are already out of RCU mode */ 1718697f514dSMiklos Szeredi err = -ENOENT; 1719caa85634SAl Viro if (d_is_negative(path.dentry)) 1720697f514dSMiklos Szeredi goto out_path_put; 1721766c4cbfSAl Viro } 1722697f514dSMiklos Szeredi 17234693a547SAl Viro if (flags & WALK_PUT) 17244693a547SAl Viro put_link(nd); 1725181548c0SAl Viro err = should_follow_link(nd, &path, flags & WALK_GET, inode, seq); 1726d63ff28fSAl Viro if (unlikely(err)) 1727d63ff28fSAl Viro return err; 1728caa85634SAl Viro path_to_nameidata(&path, nd); 1729ce57dfc1SAl Viro nd->inode = inode; 1730254cf582SAl Viro nd->seq = seq; 1731ce57dfc1SAl Viro return 0; 1732697f514dSMiklos Szeredi 1733697f514dSMiklos Szeredi out_path_put: 1734caa85634SAl Viro path_to_nameidata(&path, nd); 1735697f514dSMiklos Szeredi return err; 1736ce57dfc1SAl Viro } 1737ce57dfc1SAl Viro 17381da177e4SLinus Torvalds /* 1739bfcfaa77SLinus Torvalds * We can do the critical dentry name comparison and hashing 1740bfcfaa77SLinus Torvalds * operations one word at a time, but we are limited to: 1741bfcfaa77SLinus Torvalds * 1742bfcfaa77SLinus Torvalds * - Architectures with fast unaligned word accesses. We could 1743bfcfaa77SLinus Torvalds * do a "get_unaligned()" if this helps and is sufficiently 1744bfcfaa77SLinus Torvalds * fast. 1745bfcfaa77SLinus Torvalds * 1746bfcfaa77SLinus Torvalds * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we 1747bfcfaa77SLinus Torvalds * do not trap on the (extremely unlikely) case of a page 1748bfcfaa77SLinus Torvalds * crossing operation. 1749bfcfaa77SLinus Torvalds * 1750bfcfaa77SLinus Torvalds * - Furthermore, we need an efficient 64-bit compile for the 1751bfcfaa77SLinus Torvalds * 64-bit case in order to generate the "number of bytes in 1752bfcfaa77SLinus Torvalds * the final mask". Again, that could be replaced with a 1753bfcfaa77SLinus Torvalds * efficient population count instruction or similar. 1754bfcfaa77SLinus Torvalds */ 1755bfcfaa77SLinus Torvalds #ifdef CONFIG_DCACHE_WORD_ACCESS 1756bfcfaa77SLinus Torvalds 1757f68e556eSLinus Torvalds #include <asm/word-at-a-time.h> 1758bfcfaa77SLinus Torvalds 1759f68e556eSLinus Torvalds #ifdef CONFIG_64BIT 1760bfcfaa77SLinus Torvalds 1761bfcfaa77SLinus Torvalds static inline unsigned int fold_hash(unsigned long hash) 1762bfcfaa77SLinus Torvalds { 176399d263d4SLinus Torvalds return hash_64(hash, 32); 1764bfcfaa77SLinus Torvalds } 1765bfcfaa77SLinus Torvalds 1766bfcfaa77SLinus Torvalds #else /* 32-bit case */ 1767bfcfaa77SLinus Torvalds 1768bfcfaa77SLinus Torvalds #define fold_hash(x) (x) 1769bfcfaa77SLinus Torvalds 1770bfcfaa77SLinus Torvalds #endif 1771bfcfaa77SLinus Torvalds 1772bfcfaa77SLinus Torvalds unsigned int full_name_hash(const unsigned char *name, unsigned int len) 1773bfcfaa77SLinus Torvalds { 1774bfcfaa77SLinus Torvalds unsigned long a, mask; 1775bfcfaa77SLinus Torvalds unsigned long hash = 0; 1776bfcfaa77SLinus Torvalds 1777bfcfaa77SLinus Torvalds for (;;) { 1778e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name); 1779bfcfaa77SLinus Torvalds if (len < sizeof(unsigned long)) 1780bfcfaa77SLinus Torvalds break; 1781bfcfaa77SLinus Torvalds hash += a; 1782f132c5beSAl Viro hash *= 9; 1783bfcfaa77SLinus Torvalds name += sizeof(unsigned long); 1784bfcfaa77SLinus Torvalds len -= sizeof(unsigned long); 1785bfcfaa77SLinus Torvalds if (!len) 1786bfcfaa77SLinus Torvalds goto done; 1787bfcfaa77SLinus Torvalds } 1788a5c21dceSWill Deacon mask = bytemask_from_count(len); 1789bfcfaa77SLinus Torvalds hash += mask & a; 1790bfcfaa77SLinus Torvalds done: 1791bfcfaa77SLinus Torvalds return fold_hash(hash); 1792bfcfaa77SLinus Torvalds } 1793bfcfaa77SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 1794bfcfaa77SLinus Torvalds 1795bfcfaa77SLinus Torvalds /* 1796bfcfaa77SLinus Torvalds * Calculate the length and hash of the path component, and 1797d6bb3e90SLinus Torvalds * return the "hash_len" as the result. 1798bfcfaa77SLinus Torvalds */ 1799d6bb3e90SLinus Torvalds static inline u64 hash_name(const char *name) 1800bfcfaa77SLinus Torvalds { 180136126f8fSLinus Torvalds unsigned long a, b, adata, bdata, mask, hash, len; 180236126f8fSLinus Torvalds const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 1803bfcfaa77SLinus Torvalds 1804bfcfaa77SLinus Torvalds hash = a = 0; 1805bfcfaa77SLinus Torvalds len = -sizeof(unsigned long); 1806bfcfaa77SLinus Torvalds do { 1807bfcfaa77SLinus Torvalds hash = (hash + a) * 9; 1808bfcfaa77SLinus Torvalds len += sizeof(unsigned long); 1809e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name+len); 181036126f8fSLinus Torvalds b = a ^ REPEAT_BYTE('/'); 181136126f8fSLinus Torvalds } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); 1812bfcfaa77SLinus Torvalds 181336126f8fSLinus Torvalds adata = prep_zero_mask(a, adata, &constants); 181436126f8fSLinus Torvalds bdata = prep_zero_mask(b, bdata, &constants); 181536126f8fSLinus Torvalds 181636126f8fSLinus Torvalds mask = create_zero_mask(adata | bdata); 181736126f8fSLinus Torvalds 181836126f8fSLinus Torvalds hash += a & zero_bytemask(mask); 18199226b5b4SLinus Torvalds len += find_zero(mask); 1820d6bb3e90SLinus Torvalds return hashlen_create(fold_hash(hash), len); 1821bfcfaa77SLinus Torvalds } 1822bfcfaa77SLinus Torvalds 1823bfcfaa77SLinus Torvalds #else 1824bfcfaa77SLinus Torvalds 18250145acc2SLinus Torvalds unsigned int full_name_hash(const unsigned char *name, unsigned int len) 18260145acc2SLinus Torvalds { 18270145acc2SLinus Torvalds unsigned long hash = init_name_hash(); 18280145acc2SLinus Torvalds while (len--) 18290145acc2SLinus Torvalds hash = partial_name_hash(*name++, hash); 18300145acc2SLinus Torvalds return end_name_hash(hash); 18310145acc2SLinus Torvalds } 1832ae942ae7SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 18330145acc2SLinus Torvalds 18343ddcd056SLinus Torvalds /* 1835200e9ef7SLinus Torvalds * We know there's a real path component here of at least 1836200e9ef7SLinus Torvalds * one character. 1837200e9ef7SLinus Torvalds */ 1838d6bb3e90SLinus Torvalds static inline u64 hash_name(const char *name) 1839200e9ef7SLinus Torvalds { 1840200e9ef7SLinus Torvalds unsigned long hash = init_name_hash(); 1841200e9ef7SLinus Torvalds unsigned long len = 0, c; 1842200e9ef7SLinus Torvalds 1843200e9ef7SLinus Torvalds c = (unsigned char)*name; 1844200e9ef7SLinus Torvalds do { 1845200e9ef7SLinus Torvalds len++; 1846200e9ef7SLinus Torvalds hash = partial_name_hash(c, hash); 1847200e9ef7SLinus Torvalds c = (unsigned char)name[len]; 1848200e9ef7SLinus Torvalds } while (c && c != '/'); 1849d6bb3e90SLinus Torvalds return hashlen_create(end_name_hash(hash), len); 1850200e9ef7SLinus Torvalds } 1851200e9ef7SLinus Torvalds 1852bfcfaa77SLinus Torvalds #endif 1853bfcfaa77SLinus Torvalds 1854200e9ef7SLinus Torvalds /* 18551da177e4SLinus Torvalds * Name resolution. 1856ea3834d9SPrasanna Meda * This is the basic name resolution function, turning a pathname into 1857ea3834d9SPrasanna Meda * the final dentry. We expect 'base' to be positive and a directory. 18581da177e4SLinus Torvalds * 1859ea3834d9SPrasanna Meda * Returns 0 and nd will have valid dentry and mnt on success. 1860ea3834d9SPrasanna Meda * Returns error and drops reference to input namei data on failure. 18611da177e4SLinus Torvalds */ 18626de88d72SAl Viro static int link_path_walk(const char *name, struct nameidata *nd) 18631da177e4SLinus Torvalds { 18641da177e4SLinus Torvalds int err; 18651da177e4SLinus Torvalds 18661da177e4SLinus Torvalds while (*name=='/') 18671da177e4SLinus Torvalds name++; 18681da177e4SLinus Torvalds if (!*name) 18699e18f10aSAl Viro return 0; 18701da177e4SLinus Torvalds 18711da177e4SLinus Torvalds /* At this point we know we have a real path component. */ 18721da177e4SLinus Torvalds for(;;) { 1873d6bb3e90SLinus Torvalds u64 hash_len; 1874fe479a58SAl Viro int type; 18751da177e4SLinus Torvalds 187652094c8aSAl Viro err = may_lookup(nd); 18771da177e4SLinus Torvalds if (err) 18783595e234SAl Viro return err; 18791da177e4SLinus Torvalds 1880d6bb3e90SLinus Torvalds hash_len = hash_name(name); 18811da177e4SLinus Torvalds 1882fe479a58SAl Viro type = LAST_NORM; 1883d6bb3e90SLinus Torvalds if (name[0] == '.') switch (hashlen_len(hash_len)) { 1884fe479a58SAl Viro case 2: 1885200e9ef7SLinus Torvalds if (name[1] == '.') { 1886fe479a58SAl Viro type = LAST_DOTDOT; 188716c2cd71SAl Viro nd->flags |= LOOKUP_JUMPED; 188816c2cd71SAl Viro } 1889fe479a58SAl Viro break; 1890fe479a58SAl Viro case 1: 1891fe479a58SAl Viro type = LAST_DOT; 1892fe479a58SAl Viro } 18935a202bcdSAl Viro if (likely(type == LAST_NORM)) { 18945a202bcdSAl Viro struct dentry *parent = nd->path.dentry; 189516c2cd71SAl Viro nd->flags &= ~LOOKUP_JUMPED; 18965a202bcdSAl Viro if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { 1897a060dc50SJames Hogan struct qstr this = { { .hash_len = hash_len }, .name = name }; 1898da53be12SLinus Torvalds err = parent->d_op->d_hash(parent, &this); 18995a202bcdSAl Viro if (err < 0) 19003595e234SAl Viro return err; 1901d6bb3e90SLinus Torvalds hash_len = this.hash_len; 1902d6bb3e90SLinus Torvalds name = this.name; 19035a202bcdSAl Viro } 19045a202bcdSAl Viro } 1905fe479a58SAl Viro 1906d6bb3e90SLinus Torvalds nd->last.hash_len = hash_len; 1907d6bb3e90SLinus Torvalds nd->last.name = name; 19085f4a6a69SAl Viro nd->last_type = type; 19095f4a6a69SAl Viro 1910d6bb3e90SLinus Torvalds name += hashlen_len(hash_len); 1911d6bb3e90SLinus Torvalds if (!*name) 1912bdf6cbf1SAl Viro goto OK; 1913200e9ef7SLinus Torvalds /* 1914200e9ef7SLinus Torvalds * If it wasn't NUL, we know it was '/'. Skip that 1915200e9ef7SLinus Torvalds * slash, and continue until no more slashes. 1916200e9ef7SLinus Torvalds */ 1917200e9ef7SLinus Torvalds do { 1918d6bb3e90SLinus Torvalds name++; 1919d6bb3e90SLinus Torvalds } while (unlikely(*name == '/')); 19208620c238SAl Viro if (unlikely(!*name)) { 19218620c238SAl Viro OK: 1922368ee9baSAl Viro /* pathname body, done */ 19238620c238SAl Viro if (!nd->depth) 19248620c238SAl Viro return 0; 19258620c238SAl Viro name = nd->stack[nd->depth - 1].name; 1926368ee9baSAl Viro /* trailing symlink, done */ 19278620c238SAl Viro if (!name) 19288620c238SAl Viro return 0; 19298620c238SAl Viro /* last component of nested symlink */ 19304693a547SAl Viro err = walk_component(nd, WALK_GET | WALK_PUT); 19318620c238SAl Viro } else { 19324693a547SAl Viro err = walk_component(nd, WALK_GET); 19338620c238SAl Viro } 1934ce57dfc1SAl Viro if (err < 0) 19353595e234SAl Viro return err; 1936fe479a58SAl Viro 1937ce57dfc1SAl Viro if (err) { 1938626de996SAl Viro const char *s = get_link(nd); 19395a460275SAl Viro 19403595e234SAl Viro if (unlikely(IS_ERR(s))) 19413595e234SAl Viro return PTR_ERR(s); 1942172a39a0SAl Viro err = 0; 194312b09578SAl Viro if (unlikely(!s)) { 194412b09578SAl Viro /* jumped */ 1945b9ff4429SAl Viro put_link(nd); 194612b09578SAl Viro } else { 1947071bf501SAl Viro nd->stack[nd->depth - 1].name = name; 194832cd7468SAl Viro name = s; 19499e18f10aSAl Viro continue; 195048c8b0c5SAl Viro } 195131e6b01fSNick Piggin } 19523595e234SAl Viro if (unlikely(!d_can_lookup(nd->path.dentry))) 19533595e234SAl Viro return -ENOTDIR; 19545f4a6a69SAl Viro } 1955ce57dfc1SAl Viro } 19561da177e4SLinus Torvalds 1957368ee9baSAl Viro static const char *path_init(int dfd, const struct filename *name, 1958368ee9baSAl Viro unsigned int flags, struct nameidata *nd) 195931e6b01fSNick Piggin { 196031e6b01fSNick Piggin int retval = 0; 1961fd2f7cb5SAl Viro const char *s = name->name; 196231e6b01fSNick Piggin 196331e6b01fSNick Piggin nd->last_type = LAST_ROOT; /* if there are only slashes... */ 1964980f3ea2SAl Viro nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT; 196531e6b01fSNick Piggin nd->depth = 0; 1966368ee9baSAl Viro nd->total_link_count = 0; 19675b6ca027SAl Viro if (flags & LOOKUP_ROOT) { 1968b18825a7SDavid Howells struct dentry *root = nd->root.dentry; 1969b18825a7SDavid Howells struct inode *inode = root->d_inode; 1970fd2f7cb5SAl Viro if (*s) { 197144b1d530SMiklos Szeredi if (!d_can_lookup(root)) 1972368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 19735b6ca027SAl Viro retval = inode_permission(inode, MAY_EXEC); 19745b6ca027SAl Viro if (retval) 1975368ee9baSAl Viro return ERR_PTR(retval); 197673d049a4SAl Viro } 19775b6ca027SAl Viro nd->path = nd->root; 19785b6ca027SAl Viro nd->inode = inode; 19795b6ca027SAl Viro if (flags & LOOKUP_RCU) { 19808b61e74fSAl Viro rcu_read_lock(); 19815b6ca027SAl Viro nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 19828f47a016SAl Viro nd->root_seq = nd->seq; 198348a066e7SAl Viro nd->m_seq = read_seqbegin(&mount_lock); 19845b6ca027SAl Viro } else { 19855b6ca027SAl Viro path_get(&nd->path); 19865b6ca027SAl Viro } 1987368ee9baSAl Viro return s; 19885b6ca027SAl Viro } 19895b6ca027SAl Viro 199031e6b01fSNick Piggin nd->root.mnt = NULL; 199131e6b01fSNick Piggin 199248a066e7SAl Viro nd->m_seq = read_seqbegin(&mount_lock); 1993fd2f7cb5SAl Viro if (*s == '/') { 1994e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 19958b61e74fSAl Viro rcu_read_lock(); 19967bd88377SAl Viro nd->seq = set_root_rcu(nd); 1997e41f7d4eSAl Viro } else { 1998e41f7d4eSAl Viro set_root(nd); 1999e41f7d4eSAl Viro path_get(&nd->root); 2000e41f7d4eSAl Viro } 200131e6b01fSNick Piggin nd->path = nd->root; 200231e6b01fSNick Piggin } else if (dfd == AT_FDCWD) { 2003e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 200431e6b01fSNick Piggin struct fs_struct *fs = current->fs; 2005c28cc364SNick Piggin unsigned seq; 200631e6b01fSNick Piggin 20078b61e74fSAl Viro rcu_read_lock(); 200831e6b01fSNick Piggin 2009c28cc364SNick Piggin do { 2010c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 201131e6b01fSNick Piggin nd->path = fs->pwd; 2012c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 2013c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 2014e41f7d4eSAl Viro } else { 2015e41f7d4eSAl Viro get_fs_pwd(current->fs, &nd->path); 2016e41f7d4eSAl Viro } 201731e6b01fSNick Piggin } else { 2018582aa64aSJeff Layton /* Caller must check execute permissions on the starting path component */ 20192903ff01SAl Viro struct fd f = fdget_raw(dfd); 202031e6b01fSNick Piggin struct dentry *dentry; 202131e6b01fSNick Piggin 20222903ff01SAl Viro if (!f.file) 2023368ee9baSAl Viro return ERR_PTR(-EBADF); 202431e6b01fSNick Piggin 20252903ff01SAl Viro dentry = f.file->f_path.dentry; 202631e6b01fSNick Piggin 2027fd2f7cb5SAl Viro if (*s) { 202844b1d530SMiklos Szeredi if (!d_can_lookup(dentry)) { 20292903ff01SAl Viro fdput(f); 2030368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 2031f52e0c11SAl Viro } 20322903ff01SAl Viro } 20332903ff01SAl Viro 20342903ff01SAl Viro nd->path = f.file->f_path; 2035e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 20368b61e74fSAl Viro rcu_read_lock(); 203734a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 203834a26b99SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 20395590ff0dSUlrich Drepper } else { 20402903ff01SAl Viro path_get(&nd->path); 204134a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 20421da177e4SLinus Torvalds } 204334a26b99SAl Viro fdput(f); 2044368ee9baSAl Viro return s; 2045e41f7d4eSAl Viro } 2046e41f7d4eSAl Viro 204731e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 20484023bfc9SAl Viro if (!(flags & LOOKUP_RCU)) 2049368ee9baSAl Viro return s; 20504023bfc9SAl Viro if (likely(!read_seqcount_retry(&nd->path.dentry->d_seq, nd->seq))) 2051368ee9baSAl Viro return s; 20524023bfc9SAl Viro if (!(nd->flags & LOOKUP_ROOT)) 20534023bfc9SAl Viro nd->root.mnt = NULL; 20544023bfc9SAl Viro rcu_read_unlock(); 2055368ee9baSAl Viro return ERR_PTR(-ECHILD); 20569b4a9b14SAl Viro } 20579b4a9b14SAl Viro 20583bdba28bSAl Viro static const char *trailing_symlink(struct nameidata *nd) 205995fa25d9SAl Viro { 206095fa25d9SAl Viro const char *s; 2061fec2fa24SAl Viro int error = may_follow_link(nd); 2062deb106c6SAl Viro if (unlikely(error)) 20633bdba28bSAl Viro return ERR_PTR(error); 206495fa25d9SAl Viro nd->flags |= LOOKUP_PARENT; 2065fab51e8aSAl Viro nd->stack[0].name = NULL; 20663b2e7f75SAl Viro s = get_link(nd); 2067deb106c6SAl Viro return s ? s : ""; 206895fa25d9SAl Viro } 206995fa25d9SAl Viro 2070caa85634SAl Viro static inline int lookup_last(struct nameidata *nd) 2071bd92d7feSAl Viro { 2072bd92d7feSAl Viro if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) 2073bd92d7feSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 2074bd92d7feSAl Viro 2075bd92d7feSAl Viro nd->flags &= ~LOOKUP_PARENT; 2076deb106c6SAl Viro return walk_component(nd, 20774693a547SAl Viro nd->flags & LOOKUP_FOLLOW 20784693a547SAl Viro ? nd->depth 20794693a547SAl Viro ? WALK_PUT | WALK_GET 20804693a547SAl Viro : WALK_GET 20814693a547SAl Viro : 0); 2082bd92d7feSAl Viro } 2083bd92d7feSAl Viro 20849b4a9b14SAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2085625b6d10SAl Viro static int path_lookupat(int dfd, const struct filename *name, unsigned flags, 2086625b6d10SAl Viro struct nameidata *nd, struct path *path) 20879b4a9b14SAl Viro { 2088368ee9baSAl Viro const char *s = path_init(dfd, name, flags, nd); 2089bd92d7feSAl Viro int err; 209031e6b01fSNick Piggin 2091368ee9baSAl Viro if (IS_ERR(s)) 2092368ee9baSAl Viro return PTR_ERR(s); 20933bdba28bSAl Viro while (!(err = link_path_walk(s, nd)) 20943bdba28bSAl Viro && ((err = lookup_last(nd)) > 0)) { 20953bdba28bSAl Viro s = trailing_symlink(nd); 20963bdba28bSAl Viro if (IS_ERR(s)) { 20973bdba28bSAl Viro err = PTR_ERR(s); 20986d7b5aaeSAl Viro break; 2099bd92d7feSAl Viro } 2100bd92d7feSAl Viro } 21019f1fafeeSAl Viro if (!err) 21029f1fafeeSAl Viro err = complete_walk(nd); 2103bd92d7feSAl Viro 2104deb106c6SAl Viro if (!err && nd->flags & LOOKUP_DIRECTORY) 2105deb106c6SAl Viro if (!d_can_lookup(nd->path.dentry)) 2106bd23a539SAl Viro err = -ENOTDIR; 2107625b6d10SAl Viro if (!err) { 2108625b6d10SAl Viro *path = nd->path; 2109625b6d10SAl Viro nd->path.mnt = NULL; 2110625b6d10SAl Viro nd->path.dentry = NULL; 2111625b6d10SAl Viro } 2112deb106c6SAl Viro terminate_walk(nd); 2113bd92d7feSAl Viro return err; 211431e6b01fSNick Piggin } 211531e6b01fSNick Piggin 2116625b6d10SAl Viro static int filename_lookup(int dfd, struct filename *name, unsigned flags, 21179ad1aaa6SAl Viro struct path *path, struct path *root) 2118873f1eedSJeff Layton { 2119894bc8c4SAl Viro int retval; 2120abc9f5beSAl Viro struct nameidata nd, *saved_nd; 2121abc9f5beSAl Viro if (IS_ERR(name)) 2122abc9f5beSAl Viro return PTR_ERR(name); 2123abc9f5beSAl Viro saved_nd = set_nameidata(&nd); 21249ad1aaa6SAl Viro if (unlikely(root)) { 21259ad1aaa6SAl Viro nd.root = *root; 21269ad1aaa6SAl Viro flags |= LOOKUP_ROOT; 21279ad1aaa6SAl Viro } 21289ad1aaa6SAl Viro retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, &nd, path); 2129873f1eedSJeff Layton if (unlikely(retval == -ECHILD)) 21309ad1aaa6SAl Viro retval = path_lookupat(dfd, name, flags, &nd, path); 2131873f1eedSJeff Layton if (unlikely(retval == -ESTALE)) 2132625b6d10SAl Viro retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, 21339ad1aaa6SAl Viro &nd, path); 2134873f1eedSJeff Layton 2135873f1eedSJeff Layton if (likely(!retval)) 2136625b6d10SAl Viro audit_inode(name, path->dentry, flags & LOOKUP_PARENT); 2137756daf26SNeilBrown restore_nameidata(saved_nd); 2138e4bd1c1aSAl Viro putname(name); 2139873f1eedSJeff Layton return retval; 2140873f1eedSJeff Layton } 2141873f1eedSJeff Layton 21428bcb77faSAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 21438bcb77faSAl Viro static int path_parentat(int dfd, const struct filename *name, 2144391172c4SAl Viro unsigned int flags, struct nameidata *nd, 2145391172c4SAl Viro struct path *parent) 21468bcb77faSAl Viro { 2147368ee9baSAl Viro const char *s = path_init(dfd, name, flags, nd); 2148368ee9baSAl Viro int err; 2149368ee9baSAl Viro if (IS_ERR(s)) 2150368ee9baSAl Viro return PTR_ERR(s); 2151368ee9baSAl Viro err = link_path_walk(s, nd); 21528bcb77faSAl Viro if (!err) 21538bcb77faSAl Viro err = complete_walk(nd); 2154391172c4SAl Viro if (!err) { 2155391172c4SAl Viro *parent = nd->path; 2156391172c4SAl Viro nd->path.mnt = NULL; 2157391172c4SAl Viro nd->path.dentry = NULL; 2158391172c4SAl Viro } 2159deb106c6SAl Viro terminate_walk(nd); 21608bcb77faSAl Viro return err; 21618bcb77faSAl Viro } 21628bcb77faSAl Viro 21635c31b6ceSAl Viro static struct filename *filename_parentat(int dfd, struct filename *name, 2164391172c4SAl Viro unsigned int flags, struct path *parent, 2165391172c4SAl Viro struct qstr *last, int *type) 21668bcb77faSAl Viro { 21678bcb77faSAl Viro int retval; 21685c31b6ceSAl Viro struct nameidata nd, *saved_nd; 21698bcb77faSAl Viro 21705c31b6ceSAl Viro if (IS_ERR(name)) 21715c31b6ceSAl Viro return name; 21725c31b6ceSAl Viro saved_nd = set_nameidata(&nd); 2173391172c4SAl Viro retval = path_parentat(dfd, name, flags | LOOKUP_RCU, &nd, parent); 21748bcb77faSAl Viro if (unlikely(retval == -ECHILD)) 2175391172c4SAl Viro retval = path_parentat(dfd, name, flags, &nd, parent); 21768bcb77faSAl Viro if (unlikely(retval == -ESTALE)) 2177391172c4SAl Viro retval = path_parentat(dfd, name, flags | LOOKUP_REVAL, 2178391172c4SAl Viro &nd, parent); 2179391172c4SAl Viro if (likely(!retval)) { 2180391172c4SAl Viro *last = nd.last; 2181391172c4SAl Viro *type = nd.last_type; 2182391172c4SAl Viro audit_inode(name, parent->dentry, LOOKUP_PARENT); 21835c31b6ceSAl Viro } else { 21845c31b6ceSAl Viro putname(name); 21855c31b6ceSAl Viro name = ERR_PTR(retval); 2186391172c4SAl Viro } 21878bcb77faSAl Viro restore_nameidata(saved_nd); 21885c31b6ceSAl Viro return name; 21898bcb77faSAl Viro } 21908bcb77faSAl Viro 219179714f72SAl Viro /* does lookup, returns the object with parent locked */ 219279714f72SAl Viro struct dentry *kern_path_locked(const char *name, struct path *path) 21935590ff0dSUlrich Drepper { 21945c31b6ceSAl Viro struct filename *filename; 21955c31b6ceSAl Viro struct dentry *d; 2196391172c4SAl Viro struct qstr last; 2197391172c4SAl Viro int type; 219851689104SPaul Moore 21995c31b6ceSAl Viro filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path, 22005c31b6ceSAl Viro &last, &type); 220151689104SPaul Moore if (IS_ERR(filename)) 220251689104SPaul Moore return ERR_CAST(filename); 22035c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) { 2204391172c4SAl Viro path_put(path); 22055c31b6ceSAl Viro putname(filename); 22065c31b6ceSAl Viro return ERR_PTR(-EINVAL); 220779714f72SAl Viro } 2208391172c4SAl Viro mutex_lock_nested(&path->dentry->d_inode->i_mutex, I_MUTEX_PARENT); 2209391172c4SAl Viro d = __lookup_hash(&last, path->dentry, 0); 221079714f72SAl Viro if (IS_ERR(d)) { 2211391172c4SAl Viro mutex_unlock(&path->dentry->d_inode->i_mutex); 2212391172c4SAl Viro path_put(path); 221379714f72SAl Viro } 221451689104SPaul Moore putname(filename); 221579714f72SAl Viro return d; 22165590ff0dSUlrich Drepper } 22175590ff0dSUlrich Drepper 2218d1811465SAl Viro int kern_path(const char *name, unsigned int flags, struct path *path) 2219d1811465SAl Viro { 2220abc9f5beSAl Viro return filename_lookup(AT_FDCWD, getname_kernel(name), 2221abc9f5beSAl Viro flags, path, NULL); 2222d1811465SAl Viro } 22234d359507SAl Viro EXPORT_SYMBOL(kern_path); 2224d1811465SAl Viro 222516f18200SJosef 'Jeff' Sipek /** 222616f18200SJosef 'Jeff' Sipek * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair 222716f18200SJosef 'Jeff' Sipek * @dentry: pointer to dentry of the base directory 222816f18200SJosef 'Jeff' Sipek * @mnt: pointer to vfs mount of the base directory 222916f18200SJosef 'Jeff' Sipek * @name: pointer to file name 223016f18200SJosef 'Jeff' Sipek * @flags: lookup flags 2231e0a01249SAl Viro * @path: pointer to struct path to fill 223216f18200SJosef 'Jeff' Sipek */ 223316f18200SJosef 'Jeff' Sipek int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, 223416f18200SJosef 'Jeff' Sipek const char *name, unsigned int flags, 2235e0a01249SAl Viro struct path *path) 223616f18200SJosef 'Jeff' Sipek { 22379ad1aaa6SAl Viro struct path root = {.mnt = mnt, .dentry = dentry}; 22389ad1aaa6SAl Viro /* the first argument of filename_lookup() is ignored with root */ 2239abc9f5beSAl Viro return filename_lookup(AT_FDCWD, getname_kernel(name), 2240abc9f5beSAl Viro flags , path, &root); 224116f18200SJosef 'Jeff' Sipek } 22424d359507SAl Viro EXPORT_SYMBOL(vfs_path_lookup); 224316f18200SJosef 'Jeff' Sipek 2244eead1911SChristoph Hellwig /** 2245a6b91919SRandy Dunlap * lookup_one_len - filesystem helper to lookup single pathname component 2246eead1911SChristoph Hellwig * @name: pathname component to lookup 2247eead1911SChristoph Hellwig * @base: base directory to lookup from 2248eead1911SChristoph Hellwig * @len: maximum length @len should be interpreted to 2249eead1911SChristoph Hellwig * 2250a6b91919SRandy Dunlap * Note that this routine is purely a helper for filesystem usage and should 22519e7543e9SAl Viro * not be called by generic code. 2252eead1911SChristoph Hellwig */ 2253057f6c01SJames Morris struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 2254057f6c01SJames Morris { 2255057f6c01SJames Morris struct qstr this; 22566a96ba54SAl Viro unsigned int c; 2257cda309deSMiklos Szeredi int err; 2258057f6c01SJames Morris 22592f9092e1SDavid Woodhouse WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex)); 22602f9092e1SDavid Woodhouse 22616a96ba54SAl Viro this.name = name; 22626a96ba54SAl Viro this.len = len; 22630145acc2SLinus Torvalds this.hash = full_name_hash(name, len); 22646a96ba54SAl Viro if (!len) 22656a96ba54SAl Viro return ERR_PTR(-EACCES); 22666a96ba54SAl Viro 226721d8a15aSAl Viro if (unlikely(name[0] == '.')) { 226821d8a15aSAl Viro if (len < 2 || (len == 2 && name[1] == '.')) 226921d8a15aSAl Viro return ERR_PTR(-EACCES); 227021d8a15aSAl Viro } 227121d8a15aSAl Viro 22726a96ba54SAl Viro while (len--) { 22736a96ba54SAl Viro c = *(const unsigned char *)name++; 22746a96ba54SAl Viro if (c == '/' || c == '\0') 22756a96ba54SAl Viro return ERR_PTR(-EACCES); 22766a96ba54SAl Viro } 22775a202bcdSAl Viro /* 22785a202bcdSAl Viro * See if the low-level filesystem might want 22795a202bcdSAl Viro * to use its own hash.. 22805a202bcdSAl Viro */ 22815a202bcdSAl Viro if (base->d_flags & DCACHE_OP_HASH) { 2282da53be12SLinus Torvalds int err = base->d_op->d_hash(base, &this); 22835a202bcdSAl Viro if (err < 0) 22845a202bcdSAl Viro return ERR_PTR(err); 22855a202bcdSAl Viro } 2286eead1911SChristoph Hellwig 2287cda309deSMiklos Szeredi err = inode_permission(base->d_inode, MAY_EXEC); 2288cda309deSMiklos Szeredi if (err) 2289cda309deSMiklos Szeredi return ERR_PTR(err); 2290cda309deSMiklos Szeredi 229172bd866aSAl Viro return __lookup_hash(&this, base, 0); 2292057f6c01SJames Morris } 22934d359507SAl Viro EXPORT_SYMBOL(lookup_one_len); 2294057f6c01SJames Morris 22951fa1e7f6SAndy Whitcroft int user_path_at_empty(int dfd, const char __user *name, unsigned flags, 22961fa1e7f6SAndy Whitcroft struct path *path, int *empty) 22971da177e4SLinus Torvalds { 2298abc9f5beSAl Viro return filename_lookup(dfd, getname_flags(name, flags, empty), 2299abc9f5beSAl Viro flags, path, NULL); 23001da177e4SLinus Torvalds } 23011da177e4SLinus Torvalds 23021fa1e7f6SAndy Whitcroft int user_path_at(int dfd, const char __user *name, unsigned flags, 23031fa1e7f6SAndy Whitcroft struct path *path) 23041fa1e7f6SAndy Whitcroft { 2305f7493e5dSLinus Torvalds return user_path_at_empty(dfd, name, flags, path, NULL); 23061fa1e7f6SAndy Whitcroft } 23074d359507SAl Viro EXPORT_SYMBOL(user_path_at); 23081fa1e7f6SAndy Whitcroft 2309873f1eedSJeff Layton /* 2310873f1eedSJeff Layton * NB: most callers don't do anything directly with the reference to the 2311873f1eedSJeff Layton * to struct filename, but the nd->last pointer points into the name string 2312873f1eedSJeff Layton * allocated by getname. So we must hold the reference to it until all 2313873f1eedSJeff Layton * path-walking is complete. 2314873f1eedSJeff Layton */ 231591a27b2aSJeff Layton static struct filename * 2316f5beed75SAl Viro user_path_parent(int dfd, const char __user *path, 2317f5beed75SAl Viro struct path *parent, 2318f5beed75SAl Viro struct qstr *last, 2319f5beed75SAl Viro int *type, 23209e790bd6SJeff Layton unsigned int flags) 23212ad94ae6SAl Viro { 23229e790bd6SJeff Layton /* only LOOKUP_REVAL is allowed in extra flags */ 23235c31b6ceSAl Viro return filename_parentat(dfd, getname(path), flags & LOOKUP_REVAL, 23245c31b6ceSAl Viro parent, last, type); 23252ad94ae6SAl Viro } 23262ad94ae6SAl Viro 23278033426eSJeff Layton /** 2328197df04cSAl Viro * mountpoint_last - look up last component for umount 23298033426eSJeff Layton * @nd: pathwalk nameidata - currently pointing at parent directory of "last" 23308033426eSJeff Layton * @path: pointer to container for result 23318033426eSJeff Layton * 23328033426eSJeff Layton * This is a special lookup_last function just for umount. In this case, we 23338033426eSJeff Layton * need to resolve the path without doing any revalidation. 23348033426eSJeff Layton * 23358033426eSJeff Layton * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since 23368033426eSJeff Layton * mountpoints are always pinned in the dcache, their ancestors are too. Thus, 23378033426eSJeff Layton * in almost all cases, this lookup will be served out of the dcache. The only 23388033426eSJeff Layton * cases where it won't are if nd->last refers to a symlink or the path is 23398033426eSJeff Layton * bogus and it doesn't exist. 23408033426eSJeff Layton * 23418033426eSJeff Layton * Returns: 23428033426eSJeff Layton * -error: if there was an error during lookup. This includes -ENOENT if the 23438033426eSJeff Layton * lookup found a negative dentry. The nd->path reference will also be 23448033426eSJeff Layton * put in this case. 23458033426eSJeff Layton * 23468033426eSJeff Layton * 0: if we successfully resolved nd->path and found it to not to be a 23478033426eSJeff Layton * symlink that needs to be followed. "path" will also be populated. 23488033426eSJeff Layton * The nd->path reference will also be put. 23498033426eSJeff Layton * 23508033426eSJeff Layton * 1: if we successfully resolved nd->last and found it to be a symlink 23518033426eSJeff Layton * that needs to be followed. "path" will be populated with the path 23528033426eSJeff Layton * to the link, and nd->path will *not* be put. 23538033426eSJeff Layton */ 23548033426eSJeff Layton static int 2355197df04cSAl Viro mountpoint_last(struct nameidata *nd, struct path *path) 23568033426eSJeff Layton { 23578033426eSJeff Layton int error = 0; 23588033426eSJeff Layton struct dentry *dentry; 23598033426eSJeff Layton struct dentry *dir = nd->path.dentry; 23608033426eSJeff Layton 236135759521SAl Viro /* If we're in rcuwalk, drop out of it to handle last component */ 236235759521SAl Viro if (nd->flags & LOOKUP_RCU) { 23636e9918b7SAl Viro if (unlazy_walk(nd, NULL, 0)) 2364deb106c6SAl Viro return -ECHILD; 23658033426eSJeff Layton } 23668033426eSJeff Layton 23678033426eSJeff Layton nd->flags &= ~LOOKUP_PARENT; 23688033426eSJeff Layton 23698033426eSJeff Layton if (unlikely(nd->last_type != LAST_NORM)) { 23708033426eSJeff Layton error = handle_dots(nd, nd->last_type); 237135759521SAl Viro if (error) 2372deb106c6SAl Viro return error; 23738033426eSJeff Layton dentry = dget(nd->path.dentry); 237435759521SAl Viro goto done; 23758033426eSJeff Layton } 23768033426eSJeff Layton 23778033426eSJeff Layton mutex_lock(&dir->d_inode->i_mutex); 23788033426eSJeff Layton dentry = d_lookup(dir, &nd->last); 23798033426eSJeff Layton if (!dentry) { 23808033426eSJeff Layton /* 23818033426eSJeff Layton * No cached dentry. Mounted dentries are pinned in the cache, 23828033426eSJeff Layton * so that means that this dentry is probably a symlink or the 23838033426eSJeff Layton * path doesn't actually point to a mounted dentry. 23848033426eSJeff Layton */ 23858033426eSJeff Layton dentry = d_alloc(dir, &nd->last); 23868033426eSJeff Layton if (!dentry) { 2387bcceeebaSDave Jones mutex_unlock(&dir->d_inode->i_mutex); 2388deb106c6SAl Viro return -ENOMEM; 23898033426eSJeff Layton } 239035759521SAl Viro dentry = lookup_real(dir->d_inode, dentry, nd->flags); 2391bcceeebaSDave Jones if (IS_ERR(dentry)) { 2392bcceeebaSDave Jones mutex_unlock(&dir->d_inode->i_mutex); 2393deb106c6SAl Viro return PTR_ERR(dentry); 23948033426eSJeff Layton } 2395bcceeebaSDave Jones } 23968033426eSJeff Layton mutex_unlock(&dir->d_inode->i_mutex); 23978033426eSJeff Layton 239835759521SAl Viro done: 2399698934dfSDavid Howells if (d_is_negative(dentry)) { 24008033426eSJeff Layton dput(dentry); 2401deb106c6SAl Viro return -ENOENT; 240235759521SAl Viro } 2403191d7f73SAl Viro if (nd->depth) 2404191d7f73SAl Viro put_link(nd); 24058033426eSJeff Layton path->dentry = dentry; 2406295dc39dSVasily Averin path->mnt = nd->path.mnt; 2407181548c0SAl Viro error = should_follow_link(nd, path, nd->flags & LOOKUP_FOLLOW, 2408181548c0SAl Viro d_backing_inode(dentry), 0); 2409deb106c6SAl Viro if (unlikely(error)) 2410d63ff28fSAl Viro return error; 2411295dc39dSVasily Averin mntget(path->mnt); 24128033426eSJeff Layton follow_mount(path); 2413deb106c6SAl Viro return 0; 24148033426eSJeff Layton } 24158033426eSJeff Layton 24168033426eSJeff Layton /** 2417197df04cSAl Viro * path_mountpoint - look up a path to be umounted 24188033426eSJeff Layton * @dfd: directory file descriptor to start walk from 24198033426eSJeff Layton * @name: full pathname to walk 2420606d6fe3SRandy Dunlap * @path: pointer to container for result 24218033426eSJeff Layton * @flags: lookup flags 24228033426eSJeff Layton * 24238033426eSJeff Layton * Look up the given name, but don't attempt to revalidate the last component. 2424606d6fe3SRandy Dunlap * Returns 0 and "path" will be valid on success; Returns error otherwise. 24258033426eSJeff Layton */ 24268033426eSJeff Layton static int 2427668696dcSAl Viro path_mountpoint(int dfd, const struct filename *name, struct path *path, 242846afd6f6SAl Viro struct nameidata *nd, unsigned int flags) 24298033426eSJeff Layton { 2430368ee9baSAl Viro const char *s = path_init(dfd, name, flags, nd); 2431368ee9baSAl Viro int err; 2432368ee9baSAl Viro if (IS_ERR(s)) 2433368ee9baSAl Viro return PTR_ERR(s); 24343bdba28bSAl Viro while (!(err = link_path_walk(s, nd)) && 24353bdba28bSAl Viro (err = mountpoint_last(nd, path)) > 0) { 24363bdba28bSAl Viro s = trailing_symlink(nd); 24373bdba28bSAl Viro if (IS_ERR(s)) { 24383bdba28bSAl Viro err = PTR_ERR(s); 24398033426eSJeff Layton break; 24408033426eSJeff Layton } 24413bdba28bSAl Viro } 2442deb106c6SAl Viro terminate_walk(nd); 24438033426eSJeff Layton return err; 24448033426eSJeff Layton } 24458033426eSJeff Layton 24462d864651SAl Viro static int 2447668696dcSAl Viro filename_mountpoint(int dfd, struct filename *name, struct path *path, 24482d864651SAl Viro unsigned int flags) 24492d864651SAl Viro { 2450756daf26SNeilBrown struct nameidata nd, *saved; 2451cbaab2dbSAl Viro int error; 2452668696dcSAl Viro if (IS_ERR(name)) 2453668696dcSAl Viro return PTR_ERR(name); 2454756daf26SNeilBrown saved = set_nameidata(&nd); 245546afd6f6SAl Viro error = path_mountpoint(dfd, name, path, &nd, flags | LOOKUP_RCU); 24562d864651SAl Viro if (unlikely(error == -ECHILD)) 245746afd6f6SAl Viro error = path_mountpoint(dfd, name, path, &nd, flags); 24582d864651SAl Viro if (unlikely(error == -ESTALE)) 245946afd6f6SAl Viro error = path_mountpoint(dfd, name, path, &nd, flags | LOOKUP_REVAL); 24602d864651SAl Viro if (likely(!error)) 2461668696dcSAl Viro audit_inode(name, path->dentry, 0); 2462756daf26SNeilBrown restore_nameidata(saved); 2463668696dcSAl Viro putname(name); 24642d864651SAl Viro return error; 24652d864651SAl Viro } 24662d864651SAl Viro 24678033426eSJeff Layton /** 2468197df04cSAl Viro * user_path_mountpoint_at - lookup a path from userland in order to umount it 24698033426eSJeff Layton * @dfd: directory file descriptor 24708033426eSJeff Layton * @name: pathname from userland 24718033426eSJeff Layton * @flags: lookup flags 24728033426eSJeff Layton * @path: pointer to container to hold result 24738033426eSJeff Layton * 24748033426eSJeff Layton * A umount is a special case for path walking. We're not actually interested 24758033426eSJeff Layton * in the inode in this situation, and ESTALE errors can be a problem. We 24768033426eSJeff Layton * simply want track down the dentry and vfsmount attached at the mountpoint 24778033426eSJeff Layton * and avoid revalidating the last component. 24788033426eSJeff Layton * 24798033426eSJeff Layton * Returns 0 and populates "path" on success. 24808033426eSJeff Layton */ 24818033426eSJeff Layton int 2482197df04cSAl Viro user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags, 24838033426eSJeff Layton struct path *path) 24848033426eSJeff Layton { 2485cbaab2dbSAl Viro return filename_mountpoint(dfd, getname(name), path, flags); 24868033426eSJeff Layton } 24878033426eSJeff Layton 24882d864651SAl Viro int 24892d864651SAl Viro kern_path_mountpoint(int dfd, const char *name, struct path *path, 24902d864651SAl Viro unsigned int flags) 24912d864651SAl Viro { 2492cbaab2dbSAl Viro return filename_mountpoint(dfd, getname_kernel(name), path, flags); 24932d864651SAl Viro } 24942d864651SAl Viro EXPORT_SYMBOL(kern_path_mountpoint); 24952d864651SAl Viro 2496cbdf35bcSMiklos Szeredi int __check_sticky(struct inode *dir, struct inode *inode) 24971da177e4SLinus Torvalds { 24988e96e3b7SEric W. Biederman kuid_t fsuid = current_fsuid(); 2499da9592edSDavid Howells 25008e96e3b7SEric W. Biederman if (uid_eq(inode->i_uid, fsuid)) 25011da177e4SLinus Torvalds return 0; 25028e96e3b7SEric W. Biederman if (uid_eq(dir->i_uid, fsuid)) 25031da177e4SLinus Torvalds return 0; 250423adbe12SAndy Lutomirski return !capable_wrt_inode_uidgid(inode, CAP_FOWNER); 25051da177e4SLinus Torvalds } 2506cbdf35bcSMiklos Szeredi EXPORT_SYMBOL(__check_sticky); 25071da177e4SLinus Torvalds 25081da177e4SLinus Torvalds /* 25091da177e4SLinus Torvalds * Check whether we can remove a link victim from directory dir, check 25101da177e4SLinus Torvalds * whether the type of victim is right. 25111da177e4SLinus Torvalds * 1. We can't do it if dir is read-only (done in permission()) 25121da177e4SLinus Torvalds * 2. We should have write and exec permissions on dir 25131da177e4SLinus Torvalds * 3. We can't remove anything from append-only dir 25141da177e4SLinus Torvalds * 4. We can't do anything with immutable dir (done in permission()) 25151da177e4SLinus Torvalds * 5. If the sticky bit on dir is set we should either 25161da177e4SLinus Torvalds * a. be owner of dir, or 25171da177e4SLinus Torvalds * b. be owner of victim, or 25181da177e4SLinus Torvalds * c. have CAP_FOWNER capability 25191da177e4SLinus Torvalds * 6. If the victim is append-only or immutable we can't do antyhing with 25201da177e4SLinus Torvalds * links pointing to it. 25211da177e4SLinus Torvalds * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. 25221da177e4SLinus Torvalds * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. 25231da177e4SLinus Torvalds * 9. We can't remove a root or mountpoint. 25241da177e4SLinus Torvalds * 10. We don't allow removal of NFS sillyrenamed files; it's handled by 25251da177e4SLinus Torvalds * nfs_async_unlink(). 25261da177e4SLinus Torvalds */ 2527b18825a7SDavid Howells static int may_delete(struct inode *dir, struct dentry *victim, bool isdir) 25281da177e4SLinus Torvalds { 252963afdfc7SDavid Howells struct inode *inode = d_backing_inode(victim); 25301da177e4SLinus Torvalds int error; 25311da177e4SLinus Torvalds 2532b18825a7SDavid Howells if (d_is_negative(victim)) 25331da177e4SLinus Torvalds return -ENOENT; 2534b18825a7SDavid Howells BUG_ON(!inode); 25351da177e4SLinus Torvalds 25361da177e4SLinus Torvalds BUG_ON(victim->d_parent->d_inode != dir); 25374fa6b5ecSJeff Layton audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); 25381da177e4SLinus Torvalds 2539f419a2e3SAl Viro error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 25401da177e4SLinus Torvalds if (error) 25411da177e4SLinus Torvalds return error; 25421da177e4SLinus Torvalds if (IS_APPEND(dir)) 25431da177e4SLinus Torvalds return -EPERM; 2544b18825a7SDavid Howells 2545b18825a7SDavid Howells if (check_sticky(dir, inode) || IS_APPEND(inode) || 2546b18825a7SDavid Howells IS_IMMUTABLE(inode) || IS_SWAPFILE(inode)) 25471da177e4SLinus Torvalds return -EPERM; 25481da177e4SLinus Torvalds if (isdir) { 254944b1d530SMiklos Szeredi if (!d_is_dir(victim)) 25501da177e4SLinus Torvalds return -ENOTDIR; 25511da177e4SLinus Torvalds if (IS_ROOT(victim)) 25521da177e4SLinus Torvalds return -EBUSY; 255344b1d530SMiklos Szeredi } else if (d_is_dir(victim)) 25541da177e4SLinus Torvalds return -EISDIR; 25551da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 25561da177e4SLinus Torvalds return -ENOENT; 25571da177e4SLinus Torvalds if (victim->d_flags & DCACHE_NFSFS_RENAMED) 25581da177e4SLinus Torvalds return -EBUSY; 25591da177e4SLinus Torvalds return 0; 25601da177e4SLinus Torvalds } 25611da177e4SLinus Torvalds 25621da177e4SLinus Torvalds /* Check whether we can create an object with dentry child in directory 25631da177e4SLinus Torvalds * dir. 25641da177e4SLinus Torvalds * 1. We can't do it if child already exists (open has special treatment for 25651da177e4SLinus Torvalds * this case, but since we are inlined it's OK) 25661da177e4SLinus Torvalds * 2. We can't do it if dir is read-only (done in permission()) 25671da177e4SLinus Torvalds * 3. We should have write and exec permissions on dir 25681da177e4SLinus Torvalds * 4. We can't do it if dir is immutable (done in permission()) 25691da177e4SLinus Torvalds */ 2570a95164d9SMiklos Szeredi static inline int may_create(struct inode *dir, struct dentry *child) 25711da177e4SLinus Torvalds { 257214e972b4SJeff Layton audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); 25731da177e4SLinus Torvalds if (child->d_inode) 25741da177e4SLinus Torvalds return -EEXIST; 25751da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 25761da177e4SLinus Torvalds return -ENOENT; 2577f419a2e3SAl Viro return inode_permission(dir, MAY_WRITE | MAY_EXEC); 25781da177e4SLinus Torvalds } 25791da177e4SLinus Torvalds 25801da177e4SLinus Torvalds /* 25811da177e4SLinus Torvalds * p1 and p2 should be directories on the same fs. 25821da177e4SLinus Torvalds */ 25831da177e4SLinus Torvalds struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 25841da177e4SLinus Torvalds { 25851da177e4SLinus Torvalds struct dentry *p; 25861da177e4SLinus Torvalds 25871da177e4SLinus Torvalds if (p1 == p2) { 2588f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 25891da177e4SLinus Torvalds return NULL; 25901da177e4SLinus Torvalds } 25911da177e4SLinus Torvalds 2592a11f3a05SArjan van de Ven mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 25931da177e4SLinus Torvalds 2594e2761a11SOGAWA Hirofumi p = d_ancestor(p2, p1); 2595e2761a11SOGAWA Hirofumi if (p) { 2596f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT); 2597f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD); 25981da177e4SLinus Torvalds return p; 25991da177e4SLinus Torvalds } 26001da177e4SLinus Torvalds 2601e2761a11SOGAWA Hirofumi p = d_ancestor(p1, p2); 2602e2761a11SOGAWA Hirofumi if (p) { 2603f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 2604f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); 26051da177e4SLinus Torvalds return p; 26061da177e4SLinus Torvalds } 26071da177e4SLinus Torvalds 2608f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 2609d1b72cc6SMiklos Szeredi mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT2); 26101da177e4SLinus Torvalds return NULL; 26111da177e4SLinus Torvalds } 26124d359507SAl Viro EXPORT_SYMBOL(lock_rename); 26131da177e4SLinus Torvalds 26141da177e4SLinus Torvalds void unlock_rename(struct dentry *p1, struct dentry *p2) 26151da177e4SLinus Torvalds { 26161b1dcc1bSJes Sorensen mutex_unlock(&p1->d_inode->i_mutex); 26171da177e4SLinus Torvalds if (p1 != p2) { 26181b1dcc1bSJes Sorensen mutex_unlock(&p2->d_inode->i_mutex); 2619a11f3a05SArjan van de Ven mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 26201da177e4SLinus Torvalds } 26211da177e4SLinus Torvalds } 26224d359507SAl Viro EXPORT_SYMBOL(unlock_rename); 26231da177e4SLinus Torvalds 26244acdaf27SAl Viro int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, 2625312b63fbSAl Viro bool want_excl) 26261da177e4SLinus Torvalds { 2627a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 26281da177e4SLinus Torvalds if (error) 26291da177e4SLinus Torvalds return error; 26301da177e4SLinus Torvalds 2631acfa4380SAl Viro if (!dir->i_op->create) 26321da177e4SLinus Torvalds return -EACCES; /* shouldn't it be ENOSYS? */ 26331da177e4SLinus Torvalds mode &= S_IALLUGO; 26341da177e4SLinus Torvalds mode |= S_IFREG; 26351da177e4SLinus Torvalds error = security_inode_create(dir, dentry, mode); 26361da177e4SLinus Torvalds if (error) 26371da177e4SLinus Torvalds return error; 2638312b63fbSAl Viro error = dir->i_op->create(dir, dentry, mode, want_excl); 2639a74574aaSStephen Smalley if (!error) 2640f38aa942SAmy Griffis fsnotify_create(dir, dentry); 26411da177e4SLinus Torvalds return error; 26421da177e4SLinus Torvalds } 26434d359507SAl Viro EXPORT_SYMBOL(vfs_create); 26441da177e4SLinus Torvalds 264573d049a4SAl Viro static int may_open(struct path *path, int acc_mode, int flag) 26461da177e4SLinus Torvalds { 26473fb64190SChristoph Hellwig struct dentry *dentry = path->dentry; 26481da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 26491da177e4SLinus Torvalds int error; 26501da177e4SLinus Torvalds 2651bcda7652SAl Viro /* O_PATH? */ 2652bcda7652SAl Viro if (!acc_mode) 2653bcda7652SAl Viro return 0; 2654bcda7652SAl Viro 26551da177e4SLinus Torvalds if (!inode) 26561da177e4SLinus Torvalds return -ENOENT; 26571da177e4SLinus Torvalds 2658c8fe8f30SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 2659c8fe8f30SChristoph Hellwig case S_IFLNK: 26601da177e4SLinus Torvalds return -ELOOP; 2661c8fe8f30SChristoph Hellwig case S_IFDIR: 2662c8fe8f30SChristoph Hellwig if (acc_mode & MAY_WRITE) 26631da177e4SLinus Torvalds return -EISDIR; 2664c8fe8f30SChristoph Hellwig break; 2665c8fe8f30SChristoph Hellwig case S_IFBLK: 2666c8fe8f30SChristoph Hellwig case S_IFCHR: 26673fb64190SChristoph Hellwig if (path->mnt->mnt_flags & MNT_NODEV) 26681da177e4SLinus Torvalds return -EACCES; 2669c8fe8f30SChristoph Hellwig /*FALLTHRU*/ 2670c8fe8f30SChristoph Hellwig case S_IFIFO: 2671c8fe8f30SChristoph Hellwig case S_IFSOCK: 26721da177e4SLinus Torvalds flag &= ~O_TRUNC; 2673c8fe8f30SChristoph Hellwig break; 26744a3fd211SDave Hansen } 2675b41572e9SDave Hansen 26763fb64190SChristoph Hellwig error = inode_permission(inode, acc_mode); 2677b41572e9SDave Hansen if (error) 2678b41572e9SDave Hansen return error; 26796146f0d5SMimi Zohar 26801da177e4SLinus Torvalds /* 26811da177e4SLinus Torvalds * An append-only file must be opened in append mode for writing. 26821da177e4SLinus Torvalds */ 26831da177e4SLinus Torvalds if (IS_APPEND(inode)) { 26848737c930SAl Viro if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) 26857715b521SAl Viro return -EPERM; 26861da177e4SLinus Torvalds if (flag & O_TRUNC) 26877715b521SAl Viro return -EPERM; 26881da177e4SLinus Torvalds } 26891da177e4SLinus Torvalds 26901da177e4SLinus Torvalds /* O_NOATIME can only be set by the owner or superuser */ 26912e149670SSerge E. Hallyn if (flag & O_NOATIME && !inode_owner_or_capable(inode)) 26927715b521SAl Viro return -EPERM; 26931da177e4SLinus Torvalds 2694f3c7691eSJ. Bruce Fields return 0; 26957715b521SAl Viro } 26967715b521SAl Viro 2697e1181ee6SJeff Layton static int handle_truncate(struct file *filp) 26987715b521SAl Viro { 2699e1181ee6SJeff Layton struct path *path = &filp->f_path; 27007715b521SAl Viro struct inode *inode = path->dentry->d_inode; 27017715b521SAl Viro int error = get_write_access(inode); 27021da177e4SLinus Torvalds if (error) 27037715b521SAl Viro return error; 27041da177e4SLinus Torvalds /* 27051da177e4SLinus Torvalds * Refuse to truncate files with mandatory locks held on them. 27061da177e4SLinus Torvalds */ 2707d7a06983SJeff Layton error = locks_verify_locked(filp); 2708be6d3e56SKentaro Takeda if (!error) 2709ea0d3ab2STetsuo Handa error = security_path_truncate(path); 27101da177e4SLinus Torvalds if (!error) { 27117715b521SAl Viro error = do_truncate(path->dentry, 0, 2712d139d7ffSMiklos Szeredi ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, 2713e1181ee6SJeff Layton filp); 27141da177e4SLinus Torvalds } 27151da177e4SLinus Torvalds put_write_access(inode); 2716acd0c935SMimi Zohar return error; 27171da177e4SLinus Torvalds } 27181da177e4SLinus Torvalds 2719d57999e1SDave Hansen static inline int open_to_namei_flags(int flag) 2720d57999e1SDave Hansen { 27218a5e929dSAl Viro if ((flag & O_ACCMODE) == 3) 27228a5e929dSAl Viro flag--; 2723d57999e1SDave Hansen return flag; 2724d57999e1SDave Hansen } 2725d57999e1SDave Hansen 2726d18e9008SMiklos Szeredi static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode) 2727d18e9008SMiklos Szeredi { 2728d18e9008SMiklos Szeredi int error = security_path_mknod(dir, dentry, mode, 0); 2729d18e9008SMiklos Szeredi if (error) 2730d18e9008SMiklos Szeredi return error; 2731d18e9008SMiklos Szeredi 2732d18e9008SMiklos Szeredi error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC); 2733d18e9008SMiklos Szeredi if (error) 2734d18e9008SMiklos Szeredi return error; 2735d18e9008SMiklos Szeredi 2736d18e9008SMiklos Szeredi return security_inode_create(dir->dentry->d_inode, dentry, mode); 2737d18e9008SMiklos Szeredi } 2738d18e9008SMiklos Szeredi 27391acf0af9SDavid Howells /* 27401acf0af9SDavid Howells * Attempt to atomically look up, create and open a file from a negative 27411acf0af9SDavid Howells * dentry. 27421acf0af9SDavid Howells * 27431acf0af9SDavid Howells * Returns 0 if successful. The file will have been created and attached to 27441acf0af9SDavid Howells * @file by the filesystem calling finish_open(). 27451acf0af9SDavid Howells * 27461acf0af9SDavid Howells * Returns 1 if the file was looked up only or didn't need creating. The 27471acf0af9SDavid Howells * caller will need to perform the open themselves. @path will have been 27481acf0af9SDavid Howells * updated to point to the new dentry. This may be negative. 27491acf0af9SDavid Howells * 27501acf0af9SDavid Howells * Returns an error code otherwise. 27511acf0af9SDavid Howells */ 27522675a4ebSAl Viro static int atomic_open(struct nameidata *nd, struct dentry *dentry, 275330d90494SAl Viro struct path *path, struct file *file, 2754015c3bbcSMiklos Szeredi const struct open_flags *op, 275564894cf8SAl Viro bool got_write, bool need_lookup, 275647237687SAl Viro int *opened) 2757d18e9008SMiklos Szeredi { 2758d18e9008SMiklos Szeredi struct inode *dir = nd->path.dentry->d_inode; 2759d18e9008SMiklos Szeredi unsigned open_flag = open_to_namei_flags(op->open_flag); 2760d18e9008SMiklos Szeredi umode_t mode; 2761d18e9008SMiklos Szeredi int error; 2762d18e9008SMiklos Szeredi int acc_mode; 2763d18e9008SMiklos Szeredi int create_error = 0; 2764d18e9008SMiklos Szeredi struct dentry *const DENTRY_NOT_SET = (void *) -1UL; 2765116cc022SMiklos Szeredi bool excl; 2766d18e9008SMiklos Szeredi 2767d18e9008SMiklos Szeredi BUG_ON(dentry->d_inode); 2768d18e9008SMiklos Szeredi 2769d18e9008SMiklos Szeredi /* Don't create child dentry for a dead directory. */ 2770d18e9008SMiklos Szeredi if (unlikely(IS_DEADDIR(dir))) { 27712675a4ebSAl Viro error = -ENOENT; 2772d18e9008SMiklos Szeredi goto out; 2773d18e9008SMiklos Szeredi } 2774d18e9008SMiklos Szeredi 277562b259d8SMiklos Szeredi mode = op->mode; 2776d18e9008SMiklos Szeredi if ((open_flag & O_CREAT) && !IS_POSIXACL(dir)) 2777d18e9008SMiklos Szeredi mode &= ~current_umask(); 2778d18e9008SMiklos Szeredi 2779116cc022SMiklos Szeredi excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT); 2780116cc022SMiklos Szeredi if (excl) 2781d18e9008SMiklos Szeredi open_flag &= ~O_TRUNC; 2782d18e9008SMiklos Szeredi 2783d18e9008SMiklos Szeredi /* 2784d18e9008SMiklos Szeredi * Checking write permission is tricky, bacuse we don't know if we are 2785d18e9008SMiklos Szeredi * going to actually need it: O_CREAT opens should work as long as the 2786d18e9008SMiklos Szeredi * file exists. But checking existence breaks atomicity. The trick is 2787d18e9008SMiklos Szeredi * to check access and if not granted clear O_CREAT from the flags. 2788d18e9008SMiklos Szeredi * 2789d18e9008SMiklos Szeredi * Another problem is returing the "right" error value (e.g. for an 2790d18e9008SMiklos Szeredi * O_EXCL open we want to return EEXIST not EROFS). 2791d18e9008SMiklos Szeredi */ 279264894cf8SAl Viro if (((open_flag & (O_CREAT | O_TRUNC)) || 279364894cf8SAl Viro (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) { 279464894cf8SAl Viro if (!(open_flag & O_CREAT)) { 2795d18e9008SMiklos Szeredi /* 2796d18e9008SMiklos Szeredi * No O_CREATE -> atomicity not a requirement -> fall 2797d18e9008SMiklos Szeredi * back to lookup + open 2798d18e9008SMiklos Szeredi */ 2799d18e9008SMiklos Szeredi goto no_open; 2800d18e9008SMiklos Szeredi } else if (open_flag & (O_EXCL | O_TRUNC)) { 2801d18e9008SMiklos Szeredi /* Fall back and fail with the right error */ 280264894cf8SAl Viro create_error = -EROFS; 2803d18e9008SMiklos Szeredi goto no_open; 2804d18e9008SMiklos Szeredi } else { 2805d18e9008SMiklos Szeredi /* No side effects, safe to clear O_CREAT */ 280664894cf8SAl Viro create_error = -EROFS; 2807d18e9008SMiklos Szeredi open_flag &= ~O_CREAT; 2808d18e9008SMiklos Szeredi } 2809d18e9008SMiklos Szeredi } 2810d18e9008SMiklos Szeredi 2811d18e9008SMiklos Szeredi if (open_flag & O_CREAT) { 281238227f78SMiklos Szeredi error = may_o_create(&nd->path, dentry, mode); 2813d18e9008SMiklos Szeredi if (error) { 2814d18e9008SMiklos Szeredi create_error = error; 2815d18e9008SMiklos Szeredi if (open_flag & O_EXCL) 2816d18e9008SMiklos Szeredi goto no_open; 2817d18e9008SMiklos Szeredi open_flag &= ~O_CREAT; 2818d18e9008SMiklos Szeredi } 2819d18e9008SMiklos Szeredi } 2820d18e9008SMiklos Szeredi 2821d18e9008SMiklos Szeredi if (nd->flags & LOOKUP_DIRECTORY) 2822d18e9008SMiklos Szeredi open_flag |= O_DIRECTORY; 2823d18e9008SMiklos Szeredi 282430d90494SAl Viro file->f_path.dentry = DENTRY_NOT_SET; 282530d90494SAl Viro file->f_path.mnt = nd->path.mnt; 282630d90494SAl Viro error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode, 282747237687SAl Viro opened); 2828d9585277SAl Viro if (error < 0) { 2829d9585277SAl Viro if (create_error && error == -ENOENT) 2830d9585277SAl Viro error = create_error; 2831d18e9008SMiklos Szeredi goto out; 2832d18e9008SMiklos Szeredi } 2833d18e9008SMiklos Szeredi 2834d9585277SAl Viro if (error) { /* returned 1, that is */ 283530d90494SAl Viro if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { 28362675a4ebSAl Viro error = -EIO; 2837d18e9008SMiklos Szeredi goto out; 2838d18e9008SMiklos Szeredi } 283930d90494SAl Viro if (file->f_path.dentry) { 2840d18e9008SMiklos Szeredi dput(dentry); 284130d90494SAl Viro dentry = file->f_path.dentry; 2842d18e9008SMiklos Szeredi } 284303da633aSAl Viro if (*opened & FILE_CREATED) 284403da633aSAl Viro fsnotify_create(dir, dentry); 284503da633aSAl Viro if (!dentry->d_inode) { 284603da633aSAl Viro WARN_ON(*opened & FILE_CREATED); 284703da633aSAl Viro if (create_error) { 284862b2ce96SSage Weil error = create_error; 284962b2ce96SSage Weil goto out; 285062b2ce96SSage Weil } 285103da633aSAl Viro } else { 285203da633aSAl Viro if (excl && !(*opened & FILE_CREATED)) { 285303da633aSAl Viro error = -EEXIST; 285403da633aSAl Viro goto out; 285503da633aSAl Viro } 285603da633aSAl Viro } 2857d18e9008SMiklos Szeredi goto looked_up; 2858d18e9008SMiklos Szeredi } 2859d18e9008SMiklos Szeredi 2860d18e9008SMiklos Szeredi /* 2861d18e9008SMiklos Szeredi * We didn't have the inode before the open, so check open permission 2862d18e9008SMiklos Szeredi * here. 2863d18e9008SMiklos Szeredi */ 286403da633aSAl Viro acc_mode = op->acc_mode; 286503da633aSAl Viro if (*opened & FILE_CREATED) { 286603da633aSAl Viro WARN_ON(!(open_flag & O_CREAT)); 286703da633aSAl Viro fsnotify_create(dir, dentry); 286803da633aSAl Viro acc_mode = MAY_OPEN; 286903da633aSAl Viro } 28702675a4ebSAl Viro error = may_open(&file->f_path, acc_mode, open_flag); 28712675a4ebSAl Viro if (error) 28722675a4ebSAl Viro fput(file); 2873d18e9008SMiklos Szeredi 2874d18e9008SMiklos Szeredi out: 2875d18e9008SMiklos Szeredi dput(dentry); 28762675a4ebSAl Viro return error; 2877d18e9008SMiklos Szeredi 2878d18e9008SMiklos Szeredi no_open: 2879d18e9008SMiklos Szeredi if (need_lookup) { 288072bd866aSAl Viro dentry = lookup_real(dir, dentry, nd->flags); 2881d18e9008SMiklos Szeredi if (IS_ERR(dentry)) 28822675a4ebSAl Viro return PTR_ERR(dentry); 2883d18e9008SMiklos Szeredi 2884d18e9008SMiklos Szeredi if (create_error) { 2885d18e9008SMiklos Szeredi int open_flag = op->open_flag; 2886d18e9008SMiklos Szeredi 28872675a4ebSAl Viro error = create_error; 2888d18e9008SMiklos Szeredi if ((open_flag & O_EXCL)) { 2889d18e9008SMiklos Szeredi if (!dentry->d_inode) 2890d18e9008SMiklos Szeredi goto out; 2891d18e9008SMiklos Szeredi } else if (!dentry->d_inode) { 2892d18e9008SMiklos Szeredi goto out; 2893d18e9008SMiklos Szeredi } else if ((open_flag & O_TRUNC) && 2894e36cb0b8SDavid Howells d_is_reg(dentry)) { 2895d18e9008SMiklos Szeredi goto out; 2896d18e9008SMiklos Szeredi } 2897d18e9008SMiklos Szeredi /* will fail later, go on to get the right error */ 2898d18e9008SMiklos Szeredi } 2899d18e9008SMiklos Szeredi } 2900d18e9008SMiklos Szeredi looked_up: 2901d18e9008SMiklos Szeredi path->dentry = dentry; 2902d18e9008SMiklos Szeredi path->mnt = nd->path.mnt; 29032675a4ebSAl Viro return 1; 2904d18e9008SMiklos Szeredi } 2905d18e9008SMiklos Szeredi 290631e6b01fSNick Piggin /* 29071acf0af9SDavid Howells * Look up and maybe create and open the last component. 2908d58ffd35SMiklos Szeredi * 2909d58ffd35SMiklos Szeredi * Must be called with i_mutex held on parent. 2910d58ffd35SMiklos Szeredi * 29111acf0af9SDavid Howells * Returns 0 if the file was successfully atomically created (if necessary) and 29121acf0af9SDavid Howells * opened. In this case the file will be returned attached to @file. 29131acf0af9SDavid Howells * 29141acf0af9SDavid Howells * Returns 1 if the file was not completely opened at this time, though lookups 29151acf0af9SDavid Howells * and creations will have been performed and the dentry returned in @path will 29161acf0af9SDavid Howells * be positive upon return if O_CREAT was specified. If O_CREAT wasn't 29171acf0af9SDavid Howells * specified then a negative dentry may be returned. 29181acf0af9SDavid Howells * 29191acf0af9SDavid Howells * An error code is returned otherwise. 29201acf0af9SDavid Howells * 29211acf0af9SDavid Howells * FILE_CREATE will be set in @*opened if the dentry was created and will be 29221acf0af9SDavid Howells * cleared otherwise prior to returning. 2923d58ffd35SMiklos Szeredi */ 29242675a4ebSAl Viro static int lookup_open(struct nameidata *nd, struct path *path, 292530d90494SAl Viro struct file *file, 2926d58ffd35SMiklos Szeredi const struct open_flags *op, 292764894cf8SAl Viro bool got_write, int *opened) 2928d58ffd35SMiklos Szeredi { 2929d58ffd35SMiklos Szeredi struct dentry *dir = nd->path.dentry; 293054ef4872SMiklos Szeredi struct inode *dir_inode = dir->d_inode; 2931d58ffd35SMiklos Szeredi struct dentry *dentry; 2932d58ffd35SMiklos Szeredi int error; 293354ef4872SMiklos Szeredi bool need_lookup; 2934d58ffd35SMiklos Szeredi 293547237687SAl Viro *opened &= ~FILE_CREATED; 2936201f956eSAl Viro dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup); 2937d58ffd35SMiklos Szeredi if (IS_ERR(dentry)) 29382675a4ebSAl Viro return PTR_ERR(dentry); 2939d58ffd35SMiklos Szeredi 2940d18e9008SMiklos Szeredi /* Cached positive dentry: will open in f_op->open */ 2941d18e9008SMiklos Szeredi if (!need_lookup && dentry->d_inode) 2942d18e9008SMiklos Szeredi goto out_no_open; 2943d18e9008SMiklos Szeredi 2944d18e9008SMiklos Szeredi if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) { 294564894cf8SAl Viro return atomic_open(nd, dentry, path, file, op, got_write, 294647237687SAl Viro need_lookup, opened); 2947d18e9008SMiklos Szeredi } 2948d18e9008SMiklos Szeredi 294954ef4872SMiklos Szeredi if (need_lookup) { 295054ef4872SMiklos Szeredi BUG_ON(dentry->d_inode); 295154ef4872SMiklos Szeredi 295272bd866aSAl Viro dentry = lookup_real(dir_inode, dentry, nd->flags); 295354ef4872SMiklos Szeredi if (IS_ERR(dentry)) 29542675a4ebSAl Viro return PTR_ERR(dentry); 295554ef4872SMiklos Szeredi } 295654ef4872SMiklos Szeredi 2957d58ffd35SMiklos Szeredi /* Negative dentry, just create the file */ 2958d58ffd35SMiklos Szeredi if (!dentry->d_inode && (op->open_flag & O_CREAT)) { 2959d58ffd35SMiklos Szeredi umode_t mode = op->mode; 2960d58ffd35SMiklos Szeredi if (!IS_POSIXACL(dir->d_inode)) 2961d58ffd35SMiklos Szeredi mode &= ~current_umask(); 2962d58ffd35SMiklos Szeredi /* 2963d58ffd35SMiklos Szeredi * This write is needed to ensure that a 2964d58ffd35SMiklos Szeredi * rw->ro transition does not occur between 2965d58ffd35SMiklos Szeredi * the time when the file is created and when 2966d58ffd35SMiklos Szeredi * a permanent write count is taken through 2967015c3bbcSMiklos Szeredi * the 'struct file' in finish_open(). 2968d58ffd35SMiklos Szeredi */ 296964894cf8SAl Viro if (!got_write) { 297064894cf8SAl Viro error = -EROFS; 2971d58ffd35SMiklos Szeredi goto out_dput; 297264894cf8SAl Viro } 297347237687SAl Viro *opened |= FILE_CREATED; 2974d58ffd35SMiklos Szeredi error = security_path_mknod(&nd->path, dentry, mode, 0); 2975d58ffd35SMiklos Szeredi if (error) 2976d58ffd35SMiklos Szeredi goto out_dput; 2977312b63fbSAl Viro error = vfs_create(dir->d_inode, dentry, mode, 2978312b63fbSAl Viro nd->flags & LOOKUP_EXCL); 2979d58ffd35SMiklos Szeredi if (error) 2980d58ffd35SMiklos Szeredi goto out_dput; 2981d58ffd35SMiklos Szeredi } 2982d18e9008SMiklos Szeredi out_no_open: 2983d58ffd35SMiklos Szeredi path->dentry = dentry; 2984d58ffd35SMiklos Szeredi path->mnt = nd->path.mnt; 29852675a4ebSAl Viro return 1; 2986d58ffd35SMiklos Szeredi 2987d58ffd35SMiklos Szeredi out_dput: 2988d58ffd35SMiklos Szeredi dput(dentry); 29892675a4ebSAl Viro return error; 2990d58ffd35SMiklos Szeredi } 2991d58ffd35SMiklos Szeredi 2992d58ffd35SMiklos Szeredi /* 2993fe2d35ffSAl Viro * Handle the last step of open() 299431e6b01fSNick Piggin */ 2995896475d5SAl Viro static int do_last(struct nameidata *nd, 299630d90494SAl Viro struct file *file, const struct open_flags *op, 2997669abf4eSJeff Layton int *opened, struct filename *name) 2998fb1cc555SAl Viro { 2999a1e28038SAl Viro struct dentry *dir = nd->path.dentry; 3000ca344a89SAl Viro int open_flag = op->open_flag; 300177d660a8SMiklos Szeredi bool will_truncate = (open_flag & O_TRUNC) != 0; 300264894cf8SAl Viro bool got_write = false; 3003bcda7652SAl Viro int acc_mode = op->acc_mode; 3004254cf582SAl Viro unsigned seq; 3005a1eb3315SMiklos Szeredi struct inode *inode; 300616b1c1cdSMiklos Szeredi struct path save_parent = { .dentry = NULL, .mnt = NULL }; 3007896475d5SAl Viro struct path path; 300816b1c1cdSMiklos Szeredi bool retried = false; 300916c2cd71SAl Viro int error; 3010fb1cc555SAl Viro 3011c3e380b0SAl Viro nd->flags &= ~LOOKUP_PARENT; 3012c3e380b0SAl Viro nd->flags |= op->intent; 3013c3e380b0SAl Viro 3014bc77daa7SAl Viro if (nd->last_type != LAST_NORM) { 3015fe2d35ffSAl Viro error = handle_dots(nd, nd->last_type); 3016deb106c6SAl Viro if (unlikely(error)) 30172675a4ebSAl Viro return error; 3018e83db167SMiklos Szeredi goto finish_open; 30191f36f774SAl Viro } 3020a2c36b45SAl Viro 3021ca344a89SAl Viro if (!(open_flag & O_CREAT)) { 3022fe2d35ffSAl Viro if (nd->last.name[nd->last.len]) 3023fe2d35ffSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 3024fe2d35ffSAl Viro /* we _can_ be in RCU mode here */ 3025254cf582SAl Viro error = lookup_fast(nd, &path, &inode, &seq); 302671574865SMiklos Szeredi if (likely(!error)) 302771574865SMiklos Szeredi goto finish_lookup; 302871574865SMiklos Szeredi 3029ce57dfc1SAl Viro if (error < 0) 3030deb106c6SAl Viro return error; 3031a1eb3315SMiklos Szeredi 303237d7fffcSMiklos Szeredi BUG_ON(nd->inode != dir->d_inode); 3033b6183df7SMiklos Szeredi } else { 3034fe2d35ffSAl Viro /* create side of things */ 3035a3fbbde7SAl Viro /* 3036b6183df7SMiklos Szeredi * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED 3037b6183df7SMiklos Szeredi * has been cleared when we got to the last component we are 3038b6183df7SMiklos Szeredi * about to look up 3039a3fbbde7SAl Viro */ 30409f1fafeeSAl Viro error = complete_walk(nd); 3041e8bb73dfSAl Viro if (error) 30422675a4ebSAl Viro return error; 3043fe2d35ffSAl Viro 304433e2208aSJeff Layton audit_inode(name, dir, LOOKUP_PARENT); 30451f36f774SAl Viro /* trailing slashes? */ 3046deb106c6SAl Viro if (unlikely(nd->last.name[nd->last.len])) 3047deb106c6SAl Viro return -EISDIR; 3048b6183df7SMiklos Szeredi } 30491f36f774SAl Viro 305016b1c1cdSMiklos Szeredi retry_lookup: 305164894cf8SAl Viro if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { 305264894cf8SAl Viro error = mnt_want_write(nd->path.mnt); 305364894cf8SAl Viro if (!error) 305464894cf8SAl Viro got_write = true; 305564894cf8SAl Viro /* 305664894cf8SAl Viro * do _not_ fail yet - we might not need that or fail with 305764894cf8SAl Viro * a different error; let lookup_open() decide; we'll be 305864894cf8SAl Viro * dropping this one anyway. 305964894cf8SAl Viro */ 306064894cf8SAl Viro } 3061a1e28038SAl Viro mutex_lock(&dir->d_inode->i_mutex); 3062896475d5SAl Viro error = lookup_open(nd, &path, file, op, got_write, opened); 3063fb1cc555SAl Viro mutex_unlock(&dir->d_inode->i_mutex); 3064fb1cc555SAl Viro 30652675a4ebSAl Viro if (error <= 0) { 30662675a4ebSAl Viro if (error) 3067d58ffd35SMiklos Szeredi goto out; 30686c0d46c4SAl Viro 306947237687SAl Viro if ((*opened & FILE_CREATED) || 3070496ad9aaSAl Viro !S_ISREG(file_inode(file)->i_mode)) 307177d660a8SMiklos Szeredi will_truncate = false; 3072d18e9008SMiklos Szeredi 3073adb5c247SJeff Layton audit_inode(name, file->f_path.dentry, 0); 3074d18e9008SMiklos Szeredi goto opened; 3075d18e9008SMiklos Szeredi } 3076d18e9008SMiklos Szeredi 307747237687SAl Viro if (*opened & FILE_CREATED) { 30789b44f1b3SAl Viro /* Don't check for write permission, don't truncate */ 3079ca344a89SAl Viro open_flag &= ~O_TRUNC; 308077d660a8SMiklos Szeredi will_truncate = false; 3081bcda7652SAl Viro acc_mode = MAY_OPEN; 3082896475d5SAl Viro path_to_nameidata(&path, nd); 3083e83db167SMiklos Szeredi goto finish_open_created; 3084fb1cc555SAl Viro } 3085fb1cc555SAl Viro 3086fb1cc555SAl Viro /* 30873134f37eSJeff Layton * create/update audit record if it already exists. 3088fb1cc555SAl Viro */ 3089896475d5SAl Viro if (d_is_positive(path.dentry)) 3090896475d5SAl Viro audit_inode(name, path.dentry, 0); 3091fb1cc555SAl Viro 3092d18e9008SMiklos Szeredi /* 3093d18e9008SMiklos Szeredi * If atomic_open() acquired write access it is dropped now due to 3094d18e9008SMiklos Szeredi * possible mount and symlink following (this might be optimized away if 3095d18e9008SMiklos Szeredi * necessary...) 3096d18e9008SMiklos Szeredi */ 309764894cf8SAl Viro if (got_write) { 3098d18e9008SMiklos Szeredi mnt_drop_write(nd->path.mnt); 309964894cf8SAl Viro got_write = false; 3100d18e9008SMiklos Szeredi } 3101d18e9008SMiklos Szeredi 3102deb106c6SAl Viro if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) { 3103deb106c6SAl Viro path_to_nameidata(&path, nd); 3104deb106c6SAl Viro return -EEXIST; 3105deb106c6SAl Viro } 3106fb1cc555SAl Viro 3107756daf26SNeilBrown error = follow_managed(&path, nd); 3108deb106c6SAl Viro if (unlikely(error < 0)) 3109deb106c6SAl Viro return error; 3110a3fbbde7SAl Viro 3111decf3400SMiklos Szeredi BUG_ON(nd->flags & LOOKUP_RCU); 311263afdfc7SDavid Howells inode = d_backing_inode(path.dentry); 3113254cf582SAl Viro seq = 0; /* out of RCU mode, so the value doesn't matter */ 3114deb106c6SAl Viro if (unlikely(d_is_negative(path.dentry))) { 3115896475d5SAl Viro path_to_nameidata(&path, nd); 3116deb106c6SAl Viro return -ENOENT; 311754c33e7fSMiklos Szeredi } 3118766c4cbfSAl Viro finish_lookup: 3119191d7f73SAl Viro if (nd->depth) 3120191d7f73SAl Viro put_link(nd); 3121181548c0SAl Viro error = should_follow_link(nd, &path, nd->flags & LOOKUP_FOLLOW, 3122181548c0SAl Viro inode, seq); 3123deb106c6SAl Viro if (unlikely(error)) 3124d63ff28fSAl Viro return error; 3125fb1cc555SAl Viro 3126896475d5SAl Viro if (unlikely(d_is_symlink(path.dentry)) && !(open_flag & O_PATH)) { 3127896475d5SAl Viro path_to_nameidata(&path, nd); 3128deb106c6SAl Viro return -ELOOP; 3129a5cfe2d5SAl Viro } 3130a5cfe2d5SAl Viro 3131896475d5SAl Viro if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path.mnt) { 3132896475d5SAl Viro path_to_nameidata(&path, nd); 313316b1c1cdSMiklos Szeredi } else { 313416b1c1cdSMiklos Szeredi save_parent.dentry = nd->path.dentry; 3135896475d5SAl Viro save_parent.mnt = mntget(path.mnt); 3136896475d5SAl Viro nd->path.dentry = path.dentry; 313716b1c1cdSMiklos Szeredi 313816b1c1cdSMiklos Szeredi } 3139decf3400SMiklos Szeredi nd->inode = inode; 3140254cf582SAl Viro nd->seq = seq; 3141a3fbbde7SAl Viro /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ 3142bc77daa7SAl Viro finish_open: 3143a3fbbde7SAl Viro error = complete_walk(nd); 314416b1c1cdSMiklos Szeredi if (error) { 314516b1c1cdSMiklos Szeredi path_put(&save_parent); 31462675a4ebSAl Viro return error; 314716b1c1cdSMiklos Szeredi } 3148bc77daa7SAl Viro audit_inode(name, nd->path.dentry, 0); 3149fb1cc555SAl Viro error = -EISDIR; 315044b1d530SMiklos Szeredi if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry)) 31512675a4ebSAl Viro goto out; 3152af2f5542SMiklos Szeredi error = -ENOTDIR; 315344b1d530SMiklos Szeredi if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry)) 31542675a4ebSAl Viro goto out; 31554bbcbd3bSDavid Howells if (!d_is_reg(nd->path.dentry)) 315677d660a8SMiklos Szeredi will_truncate = false; 31576c0d46c4SAl Viro 31580f9d1a10SAl Viro if (will_truncate) { 31590f9d1a10SAl Viro error = mnt_want_write(nd->path.mnt); 31600f9d1a10SAl Viro if (error) 31612675a4ebSAl Viro goto out; 316264894cf8SAl Viro got_write = true; 31630f9d1a10SAl Viro } 3164e83db167SMiklos Szeredi finish_open_created: 3165bcda7652SAl Viro error = may_open(&nd->path, acc_mode, open_flag); 3166ca344a89SAl Viro if (error) 31672675a4ebSAl Viro goto out; 31684aa7c634SMiklos Szeredi 31694aa7c634SMiklos Szeredi BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */ 31704aa7c634SMiklos Szeredi error = vfs_open(&nd->path, file, current_cred()); 31714aa7c634SMiklos Szeredi if (!error) { 31724aa7c634SMiklos Szeredi *opened |= FILE_OPENED; 31734aa7c634SMiklos Szeredi } else { 317430d90494SAl Viro if (error == -EOPENSTALE) 3175f60dc3dbSMiklos Szeredi goto stale_open; 3176015c3bbcSMiklos Szeredi goto out; 3177f60dc3dbSMiklos Szeredi } 3178a8277b9bSMiklos Szeredi opened: 31792675a4ebSAl Viro error = open_check_o_direct(file); 3180015c3bbcSMiklos Szeredi if (error) 3181015c3bbcSMiklos Szeredi goto exit_fput; 31823034a146SDmitry Kasatkin error = ima_file_check(file, op->acc_mode, *opened); 3183aa4caadbSMiklos Szeredi if (error) 3184aa4caadbSMiklos Szeredi goto exit_fput; 3185aa4caadbSMiklos Szeredi 31860f9d1a10SAl Viro if (will_truncate) { 31872675a4ebSAl Viro error = handle_truncate(file); 3188aa4caadbSMiklos Szeredi if (error) 3189aa4caadbSMiklos Szeredi goto exit_fput; 31900f9d1a10SAl Viro } 3191ca344a89SAl Viro out: 319264894cf8SAl Viro if (got_write) 31930f9d1a10SAl Viro mnt_drop_write(nd->path.mnt); 319416b1c1cdSMiklos Szeredi path_put(&save_parent); 31952675a4ebSAl Viro return error; 3196fb1cc555SAl Viro 3197015c3bbcSMiklos Szeredi exit_fput: 31982675a4ebSAl Viro fput(file); 31992675a4ebSAl Viro goto out; 3200015c3bbcSMiklos Szeredi 3201f60dc3dbSMiklos Szeredi stale_open: 3202f60dc3dbSMiklos Szeredi /* If no saved parent or already retried then can't retry */ 3203f60dc3dbSMiklos Szeredi if (!save_parent.dentry || retried) 3204f60dc3dbSMiklos Szeredi goto out; 3205f60dc3dbSMiklos Szeredi 3206f60dc3dbSMiklos Szeredi BUG_ON(save_parent.dentry != dir); 3207f60dc3dbSMiklos Szeredi path_put(&nd->path); 3208f60dc3dbSMiklos Szeredi nd->path = save_parent; 3209f60dc3dbSMiklos Szeredi nd->inode = dir->d_inode; 3210f60dc3dbSMiklos Szeredi save_parent.mnt = NULL; 3211f60dc3dbSMiklos Szeredi save_parent.dentry = NULL; 321264894cf8SAl Viro if (got_write) { 3213f60dc3dbSMiklos Szeredi mnt_drop_write(nd->path.mnt); 321464894cf8SAl Viro got_write = false; 3215f60dc3dbSMiklos Szeredi } 3216f60dc3dbSMiklos Szeredi retried = true; 3217f60dc3dbSMiklos Szeredi goto retry_lookup; 3218fb1cc555SAl Viro } 3219fb1cc555SAl Viro 322060545d0dSAl Viro static int do_tmpfile(int dfd, struct filename *pathname, 322160545d0dSAl Viro struct nameidata *nd, int flags, 322260545d0dSAl Viro const struct open_flags *op, 322360545d0dSAl Viro struct file *file, int *opened) 322460545d0dSAl Viro { 322560545d0dSAl Viro static const struct qstr name = QSTR_INIT("/", 1); 3226625b6d10SAl Viro struct dentry *child; 322760545d0dSAl Viro struct inode *dir; 3228625b6d10SAl Viro struct path path; 32295eb6b495SAl Viro int error = path_lookupat(dfd, pathname, 3230625b6d10SAl Viro flags | LOOKUP_DIRECTORY, nd, &path); 323160545d0dSAl Viro if (unlikely(error)) 323260545d0dSAl Viro return error; 3233625b6d10SAl Viro error = mnt_want_write(path.mnt); 323460545d0dSAl Viro if (unlikely(error)) 323560545d0dSAl Viro goto out; 3236625b6d10SAl Viro dir = path.dentry->d_inode; 323760545d0dSAl Viro /* we want directory to be writable */ 3238625b6d10SAl Viro error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 323960545d0dSAl Viro if (error) 324060545d0dSAl Viro goto out2; 324160545d0dSAl Viro if (!dir->i_op->tmpfile) { 324260545d0dSAl Viro error = -EOPNOTSUPP; 324360545d0dSAl Viro goto out2; 324460545d0dSAl Viro } 3245625b6d10SAl Viro child = d_alloc(path.dentry, &name); 324660545d0dSAl Viro if (unlikely(!child)) { 324760545d0dSAl Viro error = -ENOMEM; 324860545d0dSAl Viro goto out2; 324960545d0dSAl Viro } 3250625b6d10SAl Viro dput(path.dentry); 3251625b6d10SAl Viro path.dentry = child; 3252625b6d10SAl Viro error = dir->i_op->tmpfile(dir, child, op->mode); 325360545d0dSAl Viro if (error) 325460545d0dSAl Viro goto out2; 3255625b6d10SAl Viro audit_inode(pathname, child, 0); 325669a91c23SEric Rannaud /* Don't check for other permissions, the inode was just created */ 3257625b6d10SAl Viro error = may_open(&path, MAY_OPEN, op->open_flag); 325860545d0dSAl Viro if (error) 325960545d0dSAl Viro goto out2; 3260625b6d10SAl Viro file->f_path.mnt = path.mnt; 3261625b6d10SAl Viro error = finish_open(file, child, NULL, opened); 326260545d0dSAl Viro if (error) 326360545d0dSAl Viro goto out2; 326460545d0dSAl Viro error = open_check_o_direct(file); 3265f4e0c30cSAl Viro if (error) { 326660545d0dSAl Viro fput(file); 3267f4e0c30cSAl Viro } else if (!(op->open_flag & O_EXCL)) { 3268f4e0c30cSAl Viro struct inode *inode = file_inode(file); 3269f4e0c30cSAl Viro spin_lock(&inode->i_lock); 3270f4e0c30cSAl Viro inode->i_state |= I_LINKABLE; 3271f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 3272f4e0c30cSAl Viro } 327360545d0dSAl Viro out2: 3274625b6d10SAl Viro mnt_drop_write(path.mnt); 327560545d0dSAl Viro out: 3276625b6d10SAl Viro path_put(&path); 327760545d0dSAl Viro return error; 327860545d0dSAl Viro } 327960545d0dSAl Viro 3280669abf4eSJeff Layton static struct file *path_openat(int dfd, struct filename *pathname, 328173d049a4SAl Viro struct nameidata *nd, const struct open_flags *op, int flags) 32821da177e4SLinus Torvalds { 3283368ee9baSAl Viro const char *s; 328430d90494SAl Viro struct file *file; 328547237687SAl Viro int opened = 0; 328613aab428SAl Viro int error; 328731e6b01fSNick Piggin 328830d90494SAl Viro file = get_empty_filp(); 32891afc99beSAl Viro if (IS_ERR(file)) 32901afc99beSAl Viro return file; 329131e6b01fSNick Piggin 329230d90494SAl Viro file->f_flags = op->open_flag; 329331e6b01fSNick Piggin 3294bb458c64SAl Viro if (unlikely(file->f_flags & __O_TMPFILE)) { 329560545d0dSAl Viro error = do_tmpfile(dfd, pathname, nd, flags, op, file, &opened); 3296f15133dfSAl Viro goto out2; 329760545d0dSAl Viro } 329860545d0dSAl Viro 3299368ee9baSAl Viro s = path_init(dfd, pathname, flags, nd); 3300368ee9baSAl Viro if (IS_ERR(s)) { 3301368ee9baSAl Viro put_filp(file); 3302368ee9baSAl Viro return ERR_CAST(s); 3303368ee9baSAl Viro } 33043bdba28bSAl Viro while (!(error = link_path_walk(s, nd)) && 33053bdba28bSAl Viro (error = do_last(nd, file, op, &opened, pathname)) > 0) { 330673d049a4SAl Viro nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); 33073bdba28bSAl Viro s = trailing_symlink(nd); 33083bdba28bSAl Viro if (IS_ERR(s)) { 33093bdba28bSAl Viro error = PTR_ERR(s); 33102675a4ebSAl Viro break; 3311806b681cSAl Viro } 33123bdba28bSAl Viro } 3313deb106c6SAl Viro terminate_walk(nd); 3314f15133dfSAl Viro out2: 33152675a4ebSAl Viro if (!(opened & FILE_OPENED)) { 33162675a4ebSAl Viro BUG_ON(!error); 331730d90494SAl Viro put_filp(file); 3318015c3bbcSMiklos Szeredi } 33192675a4ebSAl Viro if (unlikely(error)) { 33202675a4ebSAl Viro if (error == -EOPENSTALE) { 33212675a4ebSAl Viro if (flags & LOOKUP_RCU) 33222675a4ebSAl Viro error = -ECHILD; 33232675a4ebSAl Viro else 33242675a4ebSAl Viro error = -ESTALE; 33252675a4ebSAl Viro } 33262675a4ebSAl Viro file = ERR_PTR(error); 33272675a4ebSAl Viro } 33282675a4ebSAl Viro return file; 3329de459215SKirill Korotaev } 33301da177e4SLinus Torvalds 3331669abf4eSJeff Layton struct file *do_filp_open(int dfd, struct filename *pathname, 3332f9652e10SAl Viro const struct open_flags *op) 333313aab428SAl Viro { 3334756daf26SNeilBrown struct nameidata nd, *saved_nd = set_nameidata(&nd); 3335f9652e10SAl Viro int flags = op->lookup_flags; 333613aab428SAl Viro struct file *filp; 333713aab428SAl Viro 333873d049a4SAl Viro filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU); 333913aab428SAl Viro if (unlikely(filp == ERR_PTR(-ECHILD))) 334073d049a4SAl Viro filp = path_openat(dfd, pathname, &nd, op, flags); 334113aab428SAl Viro if (unlikely(filp == ERR_PTR(-ESTALE))) 334273d049a4SAl Viro filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL); 3343756daf26SNeilBrown restore_nameidata(saved_nd); 334413aab428SAl Viro return filp; 334513aab428SAl Viro } 334613aab428SAl Viro 334773d049a4SAl Viro struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, 3348f9652e10SAl Viro const char *name, const struct open_flags *op) 334973d049a4SAl Viro { 3350756daf26SNeilBrown struct nameidata nd, *saved_nd; 335173d049a4SAl Viro struct file *file; 335251689104SPaul Moore struct filename *filename; 3353f9652e10SAl Viro int flags = op->lookup_flags | LOOKUP_ROOT; 335473d049a4SAl Viro 335573d049a4SAl Viro nd.root.mnt = mnt; 335673d049a4SAl Viro nd.root.dentry = dentry; 335773d049a4SAl Viro 3358b18825a7SDavid Howells if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN) 335973d049a4SAl Viro return ERR_PTR(-ELOOP); 336073d049a4SAl Viro 336151689104SPaul Moore filename = getname_kernel(name); 336251689104SPaul Moore if (unlikely(IS_ERR(filename))) 336351689104SPaul Moore return ERR_CAST(filename); 336451689104SPaul Moore 3365756daf26SNeilBrown saved_nd = set_nameidata(&nd); 336651689104SPaul Moore file = path_openat(-1, filename, &nd, op, flags | LOOKUP_RCU); 336773d049a4SAl Viro if (unlikely(file == ERR_PTR(-ECHILD))) 336851689104SPaul Moore file = path_openat(-1, filename, &nd, op, flags); 336973d049a4SAl Viro if (unlikely(file == ERR_PTR(-ESTALE))) 337051689104SPaul Moore file = path_openat(-1, filename, &nd, op, flags | LOOKUP_REVAL); 3371756daf26SNeilBrown restore_nameidata(saved_nd); 337251689104SPaul Moore putname(filename); 337373d049a4SAl Viro return file; 337473d049a4SAl Viro } 337573d049a4SAl Viro 3376fa14a0b8SAl Viro static struct dentry *filename_create(int dfd, struct filename *name, 33771ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 33781da177e4SLinus Torvalds { 3379c663e5d8SChristoph Hellwig struct dentry *dentry = ERR_PTR(-EEXIST); 3380391172c4SAl Viro struct qstr last; 3381391172c4SAl Viro int type; 3382c30dabfeSJan Kara int err2; 33831ac12b4bSJeff Layton int error; 33841ac12b4bSJeff Layton bool is_dir = (lookup_flags & LOOKUP_DIRECTORY); 33851ac12b4bSJeff Layton 33861ac12b4bSJeff Layton /* 33871ac12b4bSJeff Layton * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any 33881ac12b4bSJeff Layton * other flags passed in are ignored! 33891ac12b4bSJeff Layton */ 33901ac12b4bSJeff Layton lookup_flags &= LOOKUP_REVAL; 33911ac12b4bSJeff Layton 33925c31b6ceSAl Viro name = filename_parentat(dfd, name, lookup_flags, path, &last, &type); 33935c31b6ceSAl Viro if (IS_ERR(name)) 33945c31b6ceSAl Viro return ERR_CAST(name); 33951da177e4SLinus Torvalds 3396c663e5d8SChristoph Hellwig /* 3397c663e5d8SChristoph Hellwig * Yucky last component or no last component at all? 3398c663e5d8SChristoph Hellwig * (foo/., foo/.., /////) 3399c663e5d8SChristoph Hellwig */ 34005c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) 3401ed75e95dSAl Viro goto out; 3402c663e5d8SChristoph Hellwig 3403c30dabfeSJan Kara /* don't fail immediately if it's r/o, at least try to report other errors */ 3404391172c4SAl Viro err2 = mnt_want_write(path->mnt); 3405c663e5d8SChristoph Hellwig /* 3406c663e5d8SChristoph Hellwig * Do the final lookup. 3407c663e5d8SChristoph Hellwig */ 3408391172c4SAl Viro lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL; 3409391172c4SAl Viro mutex_lock_nested(&path->dentry->d_inode->i_mutex, I_MUTEX_PARENT); 3410391172c4SAl Viro dentry = __lookup_hash(&last, path->dentry, lookup_flags); 34111da177e4SLinus Torvalds if (IS_ERR(dentry)) 3412a8104a9fSAl Viro goto unlock; 3413c663e5d8SChristoph Hellwig 3414a8104a9fSAl Viro error = -EEXIST; 3415b18825a7SDavid Howells if (d_is_positive(dentry)) 3416a8104a9fSAl Viro goto fail; 3417b18825a7SDavid Howells 3418c663e5d8SChristoph Hellwig /* 3419c663e5d8SChristoph Hellwig * Special case - lookup gave negative, but... we had foo/bar/ 3420c663e5d8SChristoph Hellwig * From the vfs_mknod() POV we just have a negative dentry - 3421c663e5d8SChristoph Hellwig * all is fine. Let's be bastards - you had / on the end, you've 3422c663e5d8SChristoph Hellwig * been asking for (non-existent) directory. -ENOENT for you. 3423c663e5d8SChristoph Hellwig */ 3424391172c4SAl Viro if (unlikely(!is_dir && last.name[last.len])) { 3425a8104a9fSAl Viro error = -ENOENT; 3426ed75e95dSAl Viro goto fail; 3427e9baf6e5SAl Viro } 3428c30dabfeSJan Kara if (unlikely(err2)) { 3429c30dabfeSJan Kara error = err2; 3430a8104a9fSAl Viro goto fail; 3431c30dabfeSJan Kara } 3432181c37b6SAl Viro putname(name); 3433e9baf6e5SAl Viro return dentry; 34341da177e4SLinus Torvalds fail: 3435a8104a9fSAl Viro dput(dentry); 3436a8104a9fSAl Viro dentry = ERR_PTR(error); 3437a8104a9fSAl Viro unlock: 3438391172c4SAl Viro mutex_unlock(&path->dentry->d_inode->i_mutex); 3439c30dabfeSJan Kara if (!err2) 3440391172c4SAl Viro mnt_drop_write(path->mnt); 3441ed75e95dSAl Viro out: 3442391172c4SAl Viro path_put(path); 3443181c37b6SAl Viro putname(name); 3444ed75e95dSAl Viro return dentry; 3445dae6ad8fSAl Viro } 3446fa14a0b8SAl Viro 3447fa14a0b8SAl Viro struct dentry *kern_path_create(int dfd, const char *pathname, 3448fa14a0b8SAl Viro struct path *path, unsigned int lookup_flags) 3449fa14a0b8SAl Viro { 3450181c37b6SAl Viro return filename_create(dfd, getname_kernel(pathname), 3451181c37b6SAl Viro path, lookup_flags); 3452fa14a0b8SAl Viro } 3453dae6ad8fSAl Viro EXPORT_SYMBOL(kern_path_create); 3454dae6ad8fSAl Viro 3455921a1650SAl Viro void done_path_create(struct path *path, struct dentry *dentry) 3456921a1650SAl Viro { 3457921a1650SAl Viro dput(dentry); 3458921a1650SAl Viro mutex_unlock(&path->dentry->d_inode->i_mutex); 3459a8104a9fSAl Viro mnt_drop_write(path->mnt); 3460921a1650SAl Viro path_put(path); 3461921a1650SAl Viro } 3462921a1650SAl Viro EXPORT_SYMBOL(done_path_create); 3463921a1650SAl Viro 34641ac12b4bSJeff Layton struct dentry *user_path_create(int dfd, const char __user *pathname, 34651ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 3466dae6ad8fSAl Viro { 3467181c37b6SAl Viro return filename_create(dfd, getname(pathname), path, lookup_flags); 3468dae6ad8fSAl Viro } 3469dae6ad8fSAl Viro EXPORT_SYMBOL(user_path_create); 3470dae6ad8fSAl Viro 34711a67aafbSAl Viro int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) 34721da177e4SLinus Torvalds { 3473a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 34741da177e4SLinus Torvalds 34751da177e4SLinus Torvalds if (error) 34761da177e4SLinus Torvalds return error; 34771da177e4SLinus Torvalds 3478975d6b39SEric W. Biederman if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) 34791da177e4SLinus Torvalds return -EPERM; 34801da177e4SLinus Torvalds 3481acfa4380SAl Viro if (!dir->i_op->mknod) 34821da177e4SLinus Torvalds return -EPERM; 34831da177e4SLinus Torvalds 348408ce5f16SSerge E. Hallyn error = devcgroup_inode_mknod(mode, dev); 348508ce5f16SSerge E. Hallyn if (error) 348608ce5f16SSerge E. Hallyn return error; 348708ce5f16SSerge E. Hallyn 34881da177e4SLinus Torvalds error = security_inode_mknod(dir, dentry, mode, dev); 34891da177e4SLinus Torvalds if (error) 34901da177e4SLinus Torvalds return error; 34911da177e4SLinus Torvalds 34921da177e4SLinus Torvalds error = dir->i_op->mknod(dir, dentry, mode, dev); 3493a74574aaSStephen Smalley if (!error) 3494f38aa942SAmy Griffis fsnotify_create(dir, dentry); 34951da177e4SLinus Torvalds return error; 34961da177e4SLinus Torvalds } 34974d359507SAl Viro EXPORT_SYMBOL(vfs_mknod); 34981da177e4SLinus Torvalds 3499f69aac00SAl Viro static int may_mknod(umode_t mode) 3500463c3197SDave Hansen { 3501463c3197SDave Hansen switch (mode & S_IFMT) { 3502463c3197SDave Hansen case S_IFREG: 3503463c3197SDave Hansen case S_IFCHR: 3504463c3197SDave Hansen case S_IFBLK: 3505463c3197SDave Hansen case S_IFIFO: 3506463c3197SDave Hansen case S_IFSOCK: 3507463c3197SDave Hansen case 0: /* zero mode translates to S_IFREG */ 3508463c3197SDave Hansen return 0; 3509463c3197SDave Hansen case S_IFDIR: 3510463c3197SDave Hansen return -EPERM; 3511463c3197SDave Hansen default: 3512463c3197SDave Hansen return -EINVAL; 3513463c3197SDave Hansen } 3514463c3197SDave Hansen } 3515463c3197SDave Hansen 35168208a22bSAl Viro SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, 35172e4d0924SHeiko Carstens unsigned, dev) 35181da177e4SLinus Torvalds { 35191da177e4SLinus Torvalds struct dentry *dentry; 3520dae6ad8fSAl Viro struct path path; 3521dae6ad8fSAl Viro int error; 3522972567f1SJeff Layton unsigned int lookup_flags = 0; 35231da177e4SLinus Torvalds 35248e4bfca1SAl Viro error = may_mknod(mode); 35258e4bfca1SAl Viro if (error) 35268e4bfca1SAl Viro return error; 3527972567f1SJeff Layton retry: 3528972567f1SJeff Layton dentry = user_path_create(dfd, filename, &path, lookup_flags); 3529dae6ad8fSAl Viro if (IS_ERR(dentry)) 3530dae6ad8fSAl Viro return PTR_ERR(dentry); 35312ad94ae6SAl Viro 3532dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3533ce3b0f8dSAl Viro mode &= ~current_umask(); 3534dae6ad8fSAl Viro error = security_path_mknod(&path, dentry, mode, dev); 3535be6d3e56SKentaro Takeda if (error) 3536a8104a9fSAl Viro goto out; 35371da177e4SLinus Torvalds switch (mode & S_IFMT) { 35381da177e4SLinus Torvalds case 0: case S_IFREG: 3539312b63fbSAl Viro error = vfs_create(path.dentry->d_inode,dentry,mode,true); 35401da177e4SLinus Torvalds break; 35411da177e4SLinus Torvalds case S_IFCHR: case S_IFBLK: 3542dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode, 35431da177e4SLinus Torvalds new_decode_dev(dev)); 35441da177e4SLinus Torvalds break; 35451da177e4SLinus Torvalds case S_IFIFO: case S_IFSOCK: 3546dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); 35471da177e4SLinus Torvalds break; 35481da177e4SLinus Torvalds } 3549a8104a9fSAl Viro out: 3550921a1650SAl Viro done_path_create(&path, dentry); 3551972567f1SJeff Layton if (retry_estale(error, lookup_flags)) { 3552972567f1SJeff Layton lookup_flags |= LOOKUP_REVAL; 3553972567f1SJeff Layton goto retry; 3554972567f1SJeff Layton } 35551da177e4SLinus Torvalds return error; 35561da177e4SLinus Torvalds } 35571da177e4SLinus Torvalds 35588208a22bSAl Viro SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) 35595590ff0dSUlrich Drepper { 35605590ff0dSUlrich Drepper return sys_mknodat(AT_FDCWD, filename, mode, dev); 35615590ff0dSUlrich Drepper } 35625590ff0dSUlrich Drepper 356318bb1db3SAl Viro int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 35641da177e4SLinus Torvalds { 3565a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 35668de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 35671da177e4SLinus Torvalds 35681da177e4SLinus Torvalds if (error) 35691da177e4SLinus Torvalds return error; 35701da177e4SLinus Torvalds 3571acfa4380SAl Viro if (!dir->i_op->mkdir) 35721da177e4SLinus Torvalds return -EPERM; 35731da177e4SLinus Torvalds 35741da177e4SLinus Torvalds mode &= (S_IRWXUGO|S_ISVTX); 35751da177e4SLinus Torvalds error = security_inode_mkdir(dir, dentry, mode); 35761da177e4SLinus Torvalds if (error) 35771da177e4SLinus Torvalds return error; 35781da177e4SLinus Torvalds 35798de52778SAl Viro if (max_links && dir->i_nlink >= max_links) 35808de52778SAl Viro return -EMLINK; 35818de52778SAl Viro 35821da177e4SLinus Torvalds error = dir->i_op->mkdir(dir, dentry, mode); 3583a74574aaSStephen Smalley if (!error) 3584f38aa942SAmy Griffis fsnotify_mkdir(dir, dentry); 35851da177e4SLinus Torvalds return error; 35861da177e4SLinus Torvalds } 35874d359507SAl Viro EXPORT_SYMBOL(vfs_mkdir); 35881da177e4SLinus Torvalds 3589a218d0fdSAl Viro SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) 35901da177e4SLinus Torvalds { 35916902d925SDave Hansen struct dentry *dentry; 3592dae6ad8fSAl Viro struct path path; 3593dae6ad8fSAl Viro int error; 3594b76d8b82SJeff Layton unsigned int lookup_flags = LOOKUP_DIRECTORY; 35951da177e4SLinus Torvalds 3596b76d8b82SJeff Layton retry: 3597b76d8b82SJeff Layton dentry = user_path_create(dfd, pathname, &path, lookup_flags); 35986902d925SDave Hansen if (IS_ERR(dentry)) 3599dae6ad8fSAl Viro return PTR_ERR(dentry); 36006902d925SDave Hansen 3601dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3602ce3b0f8dSAl Viro mode &= ~current_umask(); 3603dae6ad8fSAl Viro error = security_path_mkdir(&path, dentry, mode); 3604a8104a9fSAl Viro if (!error) 3605dae6ad8fSAl Viro error = vfs_mkdir(path.dentry->d_inode, dentry, mode); 3606921a1650SAl Viro done_path_create(&path, dentry); 3607b76d8b82SJeff Layton if (retry_estale(error, lookup_flags)) { 3608b76d8b82SJeff Layton lookup_flags |= LOOKUP_REVAL; 3609b76d8b82SJeff Layton goto retry; 3610b76d8b82SJeff Layton } 36111da177e4SLinus Torvalds return error; 36121da177e4SLinus Torvalds } 36131da177e4SLinus Torvalds 3614a218d0fdSAl Viro SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) 36155590ff0dSUlrich Drepper { 36165590ff0dSUlrich Drepper return sys_mkdirat(AT_FDCWD, pathname, mode); 36175590ff0dSUlrich Drepper } 36185590ff0dSUlrich Drepper 36191da177e4SLinus Torvalds /* 3620a71905f0SSage Weil * The dentry_unhash() helper will try to drop the dentry early: we 3621c0d02594SJ. Bruce Fields * should have a usage count of 1 if we're the only user of this 3622a71905f0SSage Weil * dentry, and if that is true (possibly after pruning the dcache), 3623a71905f0SSage Weil * then we drop the dentry now. 36241da177e4SLinus Torvalds * 36251da177e4SLinus Torvalds * A low-level filesystem can, if it choses, legally 36261da177e4SLinus Torvalds * do a 36271da177e4SLinus Torvalds * 36281da177e4SLinus Torvalds * if (!d_unhashed(dentry)) 36291da177e4SLinus Torvalds * return -EBUSY; 36301da177e4SLinus Torvalds * 36311da177e4SLinus Torvalds * if it cannot handle the case of removing a directory 36321da177e4SLinus Torvalds * that is still in use by something else.. 36331da177e4SLinus Torvalds */ 36341da177e4SLinus Torvalds void dentry_unhash(struct dentry *dentry) 36351da177e4SLinus Torvalds { 36361da177e4SLinus Torvalds shrink_dcache_parent(dentry); 36371da177e4SLinus Torvalds spin_lock(&dentry->d_lock); 363898474236SWaiman Long if (dentry->d_lockref.count == 1) 36391da177e4SLinus Torvalds __d_drop(dentry); 36401da177e4SLinus Torvalds spin_unlock(&dentry->d_lock); 36411da177e4SLinus Torvalds } 36424d359507SAl Viro EXPORT_SYMBOL(dentry_unhash); 36431da177e4SLinus Torvalds 36441da177e4SLinus Torvalds int vfs_rmdir(struct inode *dir, struct dentry *dentry) 36451da177e4SLinus Torvalds { 36461da177e4SLinus Torvalds int error = may_delete(dir, dentry, 1); 36471da177e4SLinus Torvalds 36481da177e4SLinus Torvalds if (error) 36491da177e4SLinus Torvalds return error; 36501da177e4SLinus Torvalds 3651acfa4380SAl Viro if (!dir->i_op->rmdir) 36521da177e4SLinus Torvalds return -EPERM; 36531da177e4SLinus Torvalds 36541d2ef590SAl Viro dget(dentry); 36551b1dcc1bSJes Sorensen mutex_lock(&dentry->d_inode->i_mutex); 3656912dbc15SSage Weil 36571da177e4SLinus Torvalds error = -EBUSY; 36587af1364fSEric W. Biederman if (is_local_mountpoint(dentry)) 3659912dbc15SSage Weil goto out; 3660912dbc15SSage Weil 36611da177e4SLinus Torvalds error = security_inode_rmdir(dir, dentry); 3662912dbc15SSage Weil if (error) 3663912dbc15SSage Weil goto out; 3664912dbc15SSage Weil 36653cebde24SSage Weil shrink_dcache_parent(dentry); 36661da177e4SLinus Torvalds error = dir->i_op->rmdir(dir, dentry); 3667912dbc15SSage Weil if (error) 3668912dbc15SSage Weil goto out; 3669912dbc15SSage Weil 36701da177e4SLinus Torvalds dentry->d_inode->i_flags |= S_DEAD; 3671d83c49f3SAl Viro dont_mount(dentry); 36728ed936b5SEric W. Biederman detach_mounts(dentry); 36731da177e4SLinus Torvalds 3674912dbc15SSage Weil out: 3675912dbc15SSage Weil mutex_unlock(&dentry->d_inode->i_mutex); 36761d2ef590SAl Viro dput(dentry); 3677912dbc15SSage Weil if (!error) 3678912dbc15SSage Weil d_delete(dentry); 36791da177e4SLinus Torvalds return error; 36801da177e4SLinus Torvalds } 36814d359507SAl Viro EXPORT_SYMBOL(vfs_rmdir); 36821da177e4SLinus Torvalds 36835590ff0dSUlrich Drepper static long do_rmdir(int dfd, const char __user *pathname) 36841da177e4SLinus Torvalds { 36851da177e4SLinus Torvalds int error = 0; 368691a27b2aSJeff Layton struct filename *name; 36871da177e4SLinus Torvalds struct dentry *dentry; 3688f5beed75SAl Viro struct path path; 3689f5beed75SAl Viro struct qstr last; 3690f5beed75SAl Viro int type; 3691c6ee9206SJeff Layton unsigned int lookup_flags = 0; 3692c6ee9206SJeff Layton retry: 3693f5beed75SAl Viro name = user_path_parent(dfd, pathname, 3694f5beed75SAl Viro &path, &last, &type, lookup_flags); 369591a27b2aSJeff Layton if (IS_ERR(name)) 369691a27b2aSJeff Layton return PTR_ERR(name); 36971da177e4SLinus Torvalds 3698f5beed75SAl Viro switch (type) { 36991da177e4SLinus Torvalds case LAST_DOTDOT: 37001da177e4SLinus Torvalds error = -ENOTEMPTY; 37011da177e4SLinus Torvalds goto exit1; 37021da177e4SLinus Torvalds case LAST_DOT: 37031da177e4SLinus Torvalds error = -EINVAL; 37041da177e4SLinus Torvalds goto exit1; 37051da177e4SLinus Torvalds case LAST_ROOT: 37061da177e4SLinus Torvalds error = -EBUSY; 37071da177e4SLinus Torvalds goto exit1; 37081da177e4SLinus Torvalds } 37090612d9fbSOGAWA Hirofumi 3710f5beed75SAl Viro error = mnt_want_write(path.mnt); 3711c30dabfeSJan Kara if (error) 3712c30dabfeSJan Kara goto exit1; 37130612d9fbSOGAWA Hirofumi 3714f5beed75SAl Viro mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 3715f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 37161da177e4SLinus Torvalds error = PTR_ERR(dentry); 37176902d925SDave Hansen if (IS_ERR(dentry)) 37186902d925SDave Hansen goto exit2; 3719e6bc45d6STheodore Ts'o if (!dentry->d_inode) { 3720e6bc45d6STheodore Ts'o error = -ENOENT; 3721e6bc45d6STheodore Ts'o goto exit3; 3722e6bc45d6STheodore Ts'o } 3723f5beed75SAl Viro error = security_path_rmdir(&path, dentry); 3724be6d3e56SKentaro Takeda if (error) 3725c30dabfeSJan Kara goto exit3; 3726f5beed75SAl Viro error = vfs_rmdir(path.dentry->d_inode, dentry); 37270622753bSDave Hansen exit3: 37281da177e4SLinus Torvalds dput(dentry); 37296902d925SDave Hansen exit2: 3730f5beed75SAl Viro mutex_unlock(&path.dentry->d_inode->i_mutex); 3731f5beed75SAl Viro mnt_drop_write(path.mnt); 37321da177e4SLinus Torvalds exit1: 3733f5beed75SAl Viro path_put(&path); 37341da177e4SLinus Torvalds putname(name); 3735c6ee9206SJeff Layton if (retry_estale(error, lookup_flags)) { 3736c6ee9206SJeff Layton lookup_flags |= LOOKUP_REVAL; 3737c6ee9206SJeff Layton goto retry; 3738c6ee9206SJeff Layton } 37391da177e4SLinus Torvalds return error; 37401da177e4SLinus Torvalds } 37411da177e4SLinus Torvalds 37423cdad428SHeiko Carstens SYSCALL_DEFINE1(rmdir, const char __user *, pathname) 37435590ff0dSUlrich Drepper { 37445590ff0dSUlrich Drepper return do_rmdir(AT_FDCWD, pathname); 37455590ff0dSUlrich Drepper } 37465590ff0dSUlrich Drepper 3747b21996e3SJ. Bruce Fields /** 3748b21996e3SJ. Bruce Fields * vfs_unlink - unlink a filesystem object 3749b21996e3SJ. Bruce Fields * @dir: parent directory 3750b21996e3SJ. Bruce Fields * @dentry: victim 3751b21996e3SJ. Bruce Fields * @delegated_inode: returns victim inode, if the inode is delegated. 3752b21996e3SJ. Bruce Fields * 3753b21996e3SJ. Bruce Fields * The caller must hold dir->i_mutex. 3754b21996e3SJ. Bruce Fields * 3755b21996e3SJ. Bruce Fields * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and 3756b21996e3SJ. Bruce Fields * return a reference to the inode in delegated_inode. The caller 3757b21996e3SJ. Bruce Fields * should then break the delegation on that inode and retry. Because 3758b21996e3SJ. Bruce Fields * breaking a delegation may take a long time, the caller should drop 3759b21996e3SJ. Bruce Fields * dir->i_mutex before doing so. 3760b21996e3SJ. Bruce Fields * 3761b21996e3SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 3762b21996e3SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 3763b21996e3SJ. Bruce Fields * to be NFS exported. 3764b21996e3SJ. Bruce Fields */ 3765b21996e3SJ. Bruce Fields int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode) 37661da177e4SLinus Torvalds { 37679accbb97SJ. Bruce Fields struct inode *target = dentry->d_inode; 37681da177e4SLinus Torvalds int error = may_delete(dir, dentry, 0); 37691da177e4SLinus Torvalds 37701da177e4SLinus Torvalds if (error) 37711da177e4SLinus Torvalds return error; 37721da177e4SLinus Torvalds 3773acfa4380SAl Viro if (!dir->i_op->unlink) 37741da177e4SLinus Torvalds return -EPERM; 37751da177e4SLinus Torvalds 37769accbb97SJ. Bruce Fields mutex_lock(&target->i_mutex); 37778ed936b5SEric W. Biederman if (is_local_mountpoint(dentry)) 37781da177e4SLinus Torvalds error = -EBUSY; 37791da177e4SLinus Torvalds else { 37801da177e4SLinus Torvalds error = security_inode_unlink(dir, dentry); 3781bec1052eSAl Viro if (!error) { 37825a14696cSJ. Bruce Fields error = try_break_deleg(target, delegated_inode); 37835a14696cSJ. Bruce Fields if (error) 3784b21996e3SJ. Bruce Fields goto out; 37851da177e4SLinus Torvalds error = dir->i_op->unlink(dir, dentry); 37868ed936b5SEric W. Biederman if (!error) { 3787d83c49f3SAl Viro dont_mount(dentry); 37888ed936b5SEric W. Biederman detach_mounts(dentry); 37898ed936b5SEric W. Biederman } 3790bec1052eSAl Viro } 37911da177e4SLinus Torvalds } 3792b21996e3SJ. Bruce Fields out: 37939accbb97SJ. Bruce Fields mutex_unlock(&target->i_mutex); 37941da177e4SLinus Torvalds 37951da177e4SLinus Torvalds /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 37961da177e4SLinus Torvalds if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { 37979accbb97SJ. Bruce Fields fsnotify_link_count(target); 37981da177e4SLinus Torvalds d_delete(dentry); 37991da177e4SLinus Torvalds } 38000eeca283SRobert Love 38011da177e4SLinus Torvalds return error; 38021da177e4SLinus Torvalds } 38034d359507SAl Viro EXPORT_SYMBOL(vfs_unlink); 38041da177e4SLinus Torvalds 38051da177e4SLinus Torvalds /* 38061da177e4SLinus Torvalds * Make sure that the actual truncation of the file will occur outside its 38071b1dcc1bSJes Sorensen * directory's i_mutex. Truncate can take a long time if there is a lot of 38081da177e4SLinus Torvalds * writeout happening, and we don't want to prevent access to the directory 38091da177e4SLinus Torvalds * while waiting on the I/O. 38101da177e4SLinus Torvalds */ 38115590ff0dSUlrich Drepper static long do_unlinkat(int dfd, const char __user *pathname) 38121da177e4SLinus Torvalds { 38132ad94ae6SAl Viro int error; 381491a27b2aSJeff Layton struct filename *name; 38151da177e4SLinus Torvalds struct dentry *dentry; 3816f5beed75SAl Viro struct path path; 3817f5beed75SAl Viro struct qstr last; 3818f5beed75SAl Viro int type; 38191da177e4SLinus Torvalds struct inode *inode = NULL; 3820b21996e3SJ. Bruce Fields struct inode *delegated_inode = NULL; 38215d18f813SJeff Layton unsigned int lookup_flags = 0; 38225d18f813SJeff Layton retry: 3823f5beed75SAl Viro name = user_path_parent(dfd, pathname, 3824f5beed75SAl Viro &path, &last, &type, lookup_flags); 382591a27b2aSJeff Layton if (IS_ERR(name)) 382691a27b2aSJeff Layton return PTR_ERR(name); 38272ad94ae6SAl Viro 38281da177e4SLinus Torvalds error = -EISDIR; 3829f5beed75SAl Viro if (type != LAST_NORM) 38301da177e4SLinus Torvalds goto exit1; 38310612d9fbSOGAWA Hirofumi 3832f5beed75SAl Viro error = mnt_want_write(path.mnt); 3833c30dabfeSJan Kara if (error) 3834c30dabfeSJan Kara goto exit1; 3835b21996e3SJ. Bruce Fields retry_deleg: 3836f5beed75SAl Viro mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 3837f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 38381da177e4SLinus Torvalds error = PTR_ERR(dentry); 38391da177e4SLinus Torvalds if (!IS_ERR(dentry)) { 38401da177e4SLinus Torvalds /* Why not before? Because we want correct error value */ 3841f5beed75SAl Viro if (last.name[last.len]) 384250338b88STörök Edwin goto slashes; 38431da177e4SLinus Torvalds inode = dentry->d_inode; 3844b18825a7SDavid Howells if (d_is_negative(dentry)) 3845e6bc45d6STheodore Ts'o goto slashes; 38467de9c6eeSAl Viro ihold(inode); 3847f5beed75SAl Viro error = security_path_unlink(&path, dentry); 3848be6d3e56SKentaro Takeda if (error) 3849c30dabfeSJan Kara goto exit2; 3850f5beed75SAl Viro error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode); 38511da177e4SLinus Torvalds exit2: 38521da177e4SLinus Torvalds dput(dentry); 38531da177e4SLinus Torvalds } 3854f5beed75SAl Viro mutex_unlock(&path.dentry->d_inode->i_mutex); 38551da177e4SLinus Torvalds if (inode) 38561da177e4SLinus Torvalds iput(inode); /* truncate the inode here */ 3857b21996e3SJ. Bruce Fields inode = NULL; 3858b21996e3SJ. Bruce Fields if (delegated_inode) { 38595a14696cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 3860b21996e3SJ. Bruce Fields if (!error) 3861b21996e3SJ. Bruce Fields goto retry_deleg; 3862b21996e3SJ. Bruce Fields } 3863f5beed75SAl Viro mnt_drop_write(path.mnt); 38641da177e4SLinus Torvalds exit1: 3865f5beed75SAl Viro path_put(&path); 38661da177e4SLinus Torvalds putname(name); 38675d18f813SJeff Layton if (retry_estale(error, lookup_flags)) { 38685d18f813SJeff Layton lookup_flags |= LOOKUP_REVAL; 38695d18f813SJeff Layton inode = NULL; 38705d18f813SJeff Layton goto retry; 38715d18f813SJeff Layton } 38721da177e4SLinus Torvalds return error; 38731da177e4SLinus Torvalds 38741da177e4SLinus Torvalds slashes: 3875b18825a7SDavid Howells if (d_is_negative(dentry)) 3876b18825a7SDavid Howells error = -ENOENT; 387744b1d530SMiklos Szeredi else if (d_is_dir(dentry)) 3878b18825a7SDavid Howells error = -EISDIR; 3879b18825a7SDavid Howells else 3880b18825a7SDavid Howells error = -ENOTDIR; 38811da177e4SLinus Torvalds goto exit2; 38821da177e4SLinus Torvalds } 38831da177e4SLinus Torvalds 38842e4d0924SHeiko Carstens SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) 38855590ff0dSUlrich Drepper { 38865590ff0dSUlrich Drepper if ((flag & ~AT_REMOVEDIR) != 0) 38875590ff0dSUlrich Drepper return -EINVAL; 38885590ff0dSUlrich Drepper 38895590ff0dSUlrich Drepper if (flag & AT_REMOVEDIR) 38905590ff0dSUlrich Drepper return do_rmdir(dfd, pathname); 38915590ff0dSUlrich Drepper 38925590ff0dSUlrich Drepper return do_unlinkat(dfd, pathname); 38935590ff0dSUlrich Drepper } 38945590ff0dSUlrich Drepper 38953480b257SHeiko Carstens SYSCALL_DEFINE1(unlink, const char __user *, pathname) 38965590ff0dSUlrich Drepper { 38975590ff0dSUlrich Drepper return do_unlinkat(AT_FDCWD, pathname); 38985590ff0dSUlrich Drepper } 38995590ff0dSUlrich Drepper 3900db2e747bSMiklos Szeredi int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) 39011da177e4SLinus Torvalds { 3902a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 39031da177e4SLinus Torvalds 39041da177e4SLinus Torvalds if (error) 39051da177e4SLinus Torvalds return error; 39061da177e4SLinus Torvalds 3907acfa4380SAl Viro if (!dir->i_op->symlink) 39081da177e4SLinus Torvalds return -EPERM; 39091da177e4SLinus Torvalds 39101da177e4SLinus Torvalds error = security_inode_symlink(dir, dentry, oldname); 39111da177e4SLinus Torvalds if (error) 39121da177e4SLinus Torvalds return error; 39131da177e4SLinus Torvalds 39141da177e4SLinus Torvalds error = dir->i_op->symlink(dir, dentry, oldname); 3915a74574aaSStephen Smalley if (!error) 3916f38aa942SAmy Griffis fsnotify_create(dir, dentry); 39171da177e4SLinus Torvalds return error; 39181da177e4SLinus Torvalds } 39194d359507SAl Viro EXPORT_SYMBOL(vfs_symlink); 39201da177e4SLinus Torvalds 39212e4d0924SHeiko Carstens SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, 39222e4d0924SHeiko Carstens int, newdfd, const char __user *, newname) 39231da177e4SLinus Torvalds { 39242ad94ae6SAl Viro int error; 392591a27b2aSJeff Layton struct filename *from; 39266902d925SDave Hansen struct dentry *dentry; 3927dae6ad8fSAl Viro struct path path; 3928f46d3567SJeff Layton unsigned int lookup_flags = 0; 39291da177e4SLinus Torvalds 39301da177e4SLinus Torvalds from = getname(oldname); 39311da177e4SLinus Torvalds if (IS_ERR(from)) 39321da177e4SLinus Torvalds return PTR_ERR(from); 3933f46d3567SJeff Layton retry: 3934f46d3567SJeff Layton dentry = user_path_create(newdfd, newname, &path, lookup_flags); 39351da177e4SLinus Torvalds error = PTR_ERR(dentry); 39366902d925SDave Hansen if (IS_ERR(dentry)) 3937dae6ad8fSAl Viro goto out_putname; 39386902d925SDave Hansen 393991a27b2aSJeff Layton error = security_path_symlink(&path, dentry, from->name); 3940a8104a9fSAl Viro if (!error) 394191a27b2aSJeff Layton error = vfs_symlink(path.dentry->d_inode, dentry, from->name); 3942921a1650SAl Viro done_path_create(&path, dentry); 3943f46d3567SJeff Layton if (retry_estale(error, lookup_flags)) { 3944f46d3567SJeff Layton lookup_flags |= LOOKUP_REVAL; 3945f46d3567SJeff Layton goto retry; 3946f46d3567SJeff Layton } 39476902d925SDave Hansen out_putname: 39481da177e4SLinus Torvalds putname(from); 39491da177e4SLinus Torvalds return error; 39501da177e4SLinus Torvalds } 39511da177e4SLinus Torvalds 39523480b257SHeiko Carstens SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) 39535590ff0dSUlrich Drepper { 39545590ff0dSUlrich Drepper return sys_symlinkat(oldname, AT_FDCWD, newname); 39555590ff0dSUlrich Drepper } 39565590ff0dSUlrich Drepper 3957146a8595SJ. Bruce Fields /** 3958146a8595SJ. Bruce Fields * vfs_link - create a new link 3959146a8595SJ. Bruce Fields * @old_dentry: object to be linked 3960146a8595SJ. Bruce Fields * @dir: new parent 3961146a8595SJ. Bruce Fields * @new_dentry: where to create the new link 3962146a8595SJ. Bruce Fields * @delegated_inode: returns inode needing a delegation break 3963146a8595SJ. Bruce Fields * 3964146a8595SJ. Bruce Fields * The caller must hold dir->i_mutex 3965146a8595SJ. Bruce Fields * 3966146a8595SJ. Bruce Fields * If vfs_link discovers a delegation on the to-be-linked file in need 3967146a8595SJ. Bruce Fields * of breaking, it will return -EWOULDBLOCK and return a reference to the 3968146a8595SJ. Bruce Fields * inode in delegated_inode. The caller should then break the delegation 3969146a8595SJ. Bruce Fields * and retry. Because breaking a delegation may take a long time, the 3970146a8595SJ. Bruce Fields * caller should drop the i_mutex before doing so. 3971146a8595SJ. Bruce Fields * 3972146a8595SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 3973146a8595SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 3974146a8595SJ. Bruce Fields * to be NFS exported. 3975146a8595SJ. Bruce Fields */ 3976146a8595SJ. Bruce Fields int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode) 39771da177e4SLinus Torvalds { 39781da177e4SLinus Torvalds struct inode *inode = old_dentry->d_inode; 39798de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 39801da177e4SLinus Torvalds int error; 39811da177e4SLinus Torvalds 39821da177e4SLinus Torvalds if (!inode) 39831da177e4SLinus Torvalds return -ENOENT; 39841da177e4SLinus Torvalds 3985a95164d9SMiklos Szeredi error = may_create(dir, new_dentry); 39861da177e4SLinus Torvalds if (error) 39871da177e4SLinus Torvalds return error; 39881da177e4SLinus Torvalds 39891da177e4SLinus Torvalds if (dir->i_sb != inode->i_sb) 39901da177e4SLinus Torvalds return -EXDEV; 39911da177e4SLinus Torvalds 39921da177e4SLinus Torvalds /* 39931da177e4SLinus Torvalds * A link to an append-only or immutable file cannot be created. 39941da177e4SLinus Torvalds */ 39951da177e4SLinus Torvalds if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 39961da177e4SLinus Torvalds return -EPERM; 3997acfa4380SAl Viro if (!dir->i_op->link) 39981da177e4SLinus Torvalds return -EPERM; 39997e79eedbSTetsuo Handa if (S_ISDIR(inode->i_mode)) 40001da177e4SLinus Torvalds return -EPERM; 40011da177e4SLinus Torvalds 40021da177e4SLinus Torvalds error = security_inode_link(old_dentry, dir, new_dentry); 40031da177e4SLinus Torvalds if (error) 40041da177e4SLinus Torvalds return error; 40051da177e4SLinus Torvalds 40067e79eedbSTetsuo Handa mutex_lock(&inode->i_mutex); 4007aae8a97dSAneesh Kumar K.V /* Make sure we don't allow creating hardlink to an unlinked file */ 4008f4e0c30cSAl Viro if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) 4009aae8a97dSAneesh Kumar K.V error = -ENOENT; 40108de52778SAl Viro else if (max_links && inode->i_nlink >= max_links) 40118de52778SAl Viro error = -EMLINK; 4012146a8595SJ. Bruce Fields else { 4013146a8595SJ. Bruce Fields error = try_break_deleg(inode, delegated_inode); 4014146a8595SJ. Bruce Fields if (!error) 40151da177e4SLinus Torvalds error = dir->i_op->link(old_dentry, dir, new_dentry); 4016146a8595SJ. Bruce Fields } 4017f4e0c30cSAl Viro 4018f4e0c30cSAl Viro if (!error && (inode->i_state & I_LINKABLE)) { 4019f4e0c30cSAl Viro spin_lock(&inode->i_lock); 4020f4e0c30cSAl Viro inode->i_state &= ~I_LINKABLE; 4021f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 4022f4e0c30cSAl Viro } 40237e79eedbSTetsuo Handa mutex_unlock(&inode->i_mutex); 4024e31e14ecSStephen Smalley if (!error) 40257e79eedbSTetsuo Handa fsnotify_link(dir, inode, new_dentry); 40261da177e4SLinus Torvalds return error; 40271da177e4SLinus Torvalds } 40284d359507SAl Viro EXPORT_SYMBOL(vfs_link); 40291da177e4SLinus Torvalds 40301da177e4SLinus Torvalds /* 40311da177e4SLinus Torvalds * Hardlinks are often used in delicate situations. We avoid 40321da177e4SLinus Torvalds * security-related surprises by not following symlinks on the 40331da177e4SLinus Torvalds * newname. --KAB 40341da177e4SLinus Torvalds * 40351da177e4SLinus Torvalds * We don't follow them on the oldname either to be compatible 40361da177e4SLinus Torvalds * with linux 2.0, and to avoid hard-linking to directories 40371da177e4SLinus Torvalds * and other special files. --ADM 40381da177e4SLinus Torvalds */ 40392e4d0924SHeiko Carstens SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, 40402e4d0924SHeiko Carstens int, newdfd, const char __user *, newname, int, flags) 40411da177e4SLinus Torvalds { 40421da177e4SLinus Torvalds struct dentry *new_dentry; 4043dae6ad8fSAl Viro struct path old_path, new_path; 4044146a8595SJ. Bruce Fields struct inode *delegated_inode = NULL; 404511a7b371SAneesh Kumar K.V int how = 0; 40461da177e4SLinus Torvalds int error; 40471da177e4SLinus Torvalds 404811a7b371SAneesh Kumar K.V if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) 4049c04030e1SUlrich Drepper return -EINVAL; 405011a7b371SAneesh Kumar K.V /* 4051f0cc6ffbSLinus Torvalds * To use null names we require CAP_DAC_READ_SEARCH 4052f0cc6ffbSLinus Torvalds * This ensures that not everyone will be able to create 4053f0cc6ffbSLinus Torvalds * handlink using the passed filedescriptor. 405411a7b371SAneesh Kumar K.V */ 4055f0cc6ffbSLinus Torvalds if (flags & AT_EMPTY_PATH) { 4056f0cc6ffbSLinus Torvalds if (!capable(CAP_DAC_READ_SEARCH)) 4057f0cc6ffbSLinus Torvalds return -ENOENT; 405811a7b371SAneesh Kumar K.V how = LOOKUP_EMPTY; 4059f0cc6ffbSLinus Torvalds } 4060c04030e1SUlrich Drepper 406111a7b371SAneesh Kumar K.V if (flags & AT_SYMLINK_FOLLOW) 406211a7b371SAneesh Kumar K.V how |= LOOKUP_FOLLOW; 4063442e31caSJeff Layton retry: 406411a7b371SAneesh Kumar K.V error = user_path_at(olddfd, oldname, how, &old_path); 40651da177e4SLinus Torvalds if (error) 40662ad94ae6SAl Viro return error; 40672ad94ae6SAl Viro 4068442e31caSJeff Layton new_dentry = user_path_create(newdfd, newname, &new_path, 4069442e31caSJeff Layton (how & LOOKUP_REVAL)); 40701da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 40716902d925SDave Hansen if (IS_ERR(new_dentry)) 4072dae6ad8fSAl Viro goto out; 4073dae6ad8fSAl Viro 4074dae6ad8fSAl Viro error = -EXDEV; 4075dae6ad8fSAl Viro if (old_path.mnt != new_path.mnt) 4076dae6ad8fSAl Viro goto out_dput; 4077800179c9SKees Cook error = may_linkat(&old_path); 4078800179c9SKees Cook if (unlikely(error)) 4079800179c9SKees Cook goto out_dput; 4080dae6ad8fSAl Viro error = security_path_link(old_path.dentry, &new_path, new_dentry); 4081be6d3e56SKentaro Takeda if (error) 4082a8104a9fSAl Viro goto out_dput; 4083146a8595SJ. Bruce Fields error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode); 408475c3f29dSDave Hansen out_dput: 4085921a1650SAl Viro done_path_create(&new_path, new_dentry); 4086146a8595SJ. Bruce Fields if (delegated_inode) { 4087146a8595SJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4088d22e6338SOleg Drokin if (!error) { 4089d22e6338SOleg Drokin path_put(&old_path); 4090146a8595SJ. Bruce Fields goto retry; 4091146a8595SJ. Bruce Fields } 4092d22e6338SOleg Drokin } 4093442e31caSJeff Layton if (retry_estale(error, how)) { 4094d22e6338SOleg Drokin path_put(&old_path); 4095442e31caSJeff Layton how |= LOOKUP_REVAL; 4096442e31caSJeff Layton goto retry; 4097442e31caSJeff Layton } 40981da177e4SLinus Torvalds out: 40992d8f3038SAl Viro path_put(&old_path); 41001da177e4SLinus Torvalds 41011da177e4SLinus Torvalds return error; 41021da177e4SLinus Torvalds } 41031da177e4SLinus Torvalds 41043480b257SHeiko Carstens SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) 41055590ff0dSUlrich Drepper { 4106c04030e1SUlrich Drepper return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 41075590ff0dSUlrich Drepper } 41085590ff0dSUlrich Drepper 4109bc27027aSMiklos Szeredi /** 4110bc27027aSMiklos Szeredi * vfs_rename - rename a filesystem object 4111bc27027aSMiklos Szeredi * @old_dir: parent of source 4112bc27027aSMiklos Szeredi * @old_dentry: source 4113bc27027aSMiklos Szeredi * @new_dir: parent of destination 4114bc27027aSMiklos Szeredi * @new_dentry: destination 4115bc27027aSMiklos Szeredi * @delegated_inode: returns an inode needing a delegation break 4116520c8b16SMiklos Szeredi * @flags: rename flags 4117bc27027aSMiklos Szeredi * 4118bc27027aSMiklos Szeredi * The caller must hold multiple mutexes--see lock_rename()). 4119bc27027aSMiklos Szeredi * 4120bc27027aSMiklos Szeredi * If vfs_rename discovers a delegation in need of breaking at either 4121bc27027aSMiklos Szeredi * the source or destination, it will return -EWOULDBLOCK and return a 4122bc27027aSMiklos Szeredi * reference to the inode in delegated_inode. The caller should then 4123bc27027aSMiklos Szeredi * break the delegation and retry. Because breaking a delegation may 4124bc27027aSMiklos Szeredi * take a long time, the caller should drop all locks before doing 4125bc27027aSMiklos Szeredi * so. 4126bc27027aSMiklos Szeredi * 4127bc27027aSMiklos Szeredi * Alternatively, a caller may pass NULL for delegated_inode. This may 4128bc27027aSMiklos Szeredi * be appropriate for callers that expect the underlying filesystem not 4129bc27027aSMiklos Szeredi * to be NFS exported. 4130bc27027aSMiklos Szeredi * 41311da177e4SLinus Torvalds * The worst of all namespace operations - renaming directory. "Perverted" 41321da177e4SLinus Torvalds * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 41331da177e4SLinus Torvalds * Problems: 4134d03b29a2SJ. Bruce Fields * a) we can get into loop creation. 41351da177e4SLinus Torvalds * b) race potential - two innocent renames can create a loop together. 41361da177e4SLinus Torvalds * That's where 4.4 screws up. Current fix: serialization on 4137a11f3a05SArjan van de Ven * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 41381da177e4SLinus Torvalds * story. 41396cedba89SJ. Bruce Fields * c) we have to lock _four_ objects - parents and victim (if it exists), 41406cedba89SJ. Bruce Fields * and source (if it is not a directory). 41411b1dcc1bSJes Sorensen * And that - after we got ->i_mutex on parents (until then we don't know 41421da177e4SLinus Torvalds * whether the target exists). Solution: try to be smart with locking 41431da177e4SLinus Torvalds * order for inodes. We rely on the fact that tree topology may change 4144a11f3a05SArjan van de Ven * only under ->s_vfs_rename_mutex _and_ that parent of the object we 41451da177e4SLinus Torvalds * move will be locked. Thus we can rank directories by the tree 41461da177e4SLinus Torvalds * (ancestors first) and rank all non-directories after them. 41471da177e4SLinus Torvalds * That works since everybody except rename does "lock parent, lookup, 4148a11f3a05SArjan van de Ven * lock child" and rename is under ->s_vfs_rename_mutex. 41491da177e4SLinus Torvalds * HOWEVER, it relies on the assumption that any object with ->lookup() 41501da177e4SLinus Torvalds * has no more than 1 dentry. If "hybrid" objects will ever appear, 41511da177e4SLinus Torvalds * we'd better make sure that there's no link(2) for them. 4152e4eaac06SSage Weil * d) conversion from fhandle to dentry may come in the wrong moment - when 41531b1dcc1bSJes Sorensen * we are removing the target. Solution: we will have to grab ->i_mutex 41541da177e4SLinus Torvalds * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 4155c41b20e7SAdam Buchbinder * ->i_mutex on parents, which works but leads to some truly excessive 41561da177e4SLinus Torvalds * locking]. 41571da177e4SLinus Torvalds */ 41581da177e4SLinus Torvalds int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, 41598e6d782cSJ. Bruce Fields struct inode *new_dir, struct dentry *new_dentry, 4160520c8b16SMiklos Szeredi struct inode **delegated_inode, unsigned int flags) 41611da177e4SLinus Torvalds { 41621da177e4SLinus Torvalds int error; 4163bc27027aSMiklos Szeredi bool is_dir = d_is_dir(old_dentry); 416459b0df21SEric Paris const unsigned char *old_name; 4165bc27027aSMiklos Szeredi struct inode *source = old_dentry->d_inode; 4166bc27027aSMiklos Szeredi struct inode *target = new_dentry->d_inode; 4167da1ce067SMiklos Szeredi bool new_is_dir = false; 4168da1ce067SMiklos Szeredi unsigned max_links = new_dir->i_sb->s_max_links; 41691da177e4SLinus Torvalds 4170bc27027aSMiklos Szeredi if (source == target) 41711da177e4SLinus Torvalds return 0; 41721da177e4SLinus Torvalds 41731da177e4SLinus Torvalds error = may_delete(old_dir, old_dentry, is_dir); 41741da177e4SLinus Torvalds if (error) 41751da177e4SLinus Torvalds return error; 41761da177e4SLinus Torvalds 4177da1ce067SMiklos Szeredi if (!target) { 4178a95164d9SMiklos Szeredi error = may_create(new_dir, new_dentry); 4179da1ce067SMiklos Szeredi } else { 4180da1ce067SMiklos Szeredi new_is_dir = d_is_dir(new_dentry); 4181da1ce067SMiklos Szeredi 4182da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 41831da177e4SLinus Torvalds error = may_delete(new_dir, new_dentry, is_dir); 4184da1ce067SMiklos Szeredi else 4185da1ce067SMiklos Szeredi error = may_delete(new_dir, new_dentry, new_is_dir); 4186da1ce067SMiklos Szeredi } 41871da177e4SLinus Torvalds if (error) 41881da177e4SLinus Torvalds return error; 41891da177e4SLinus Torvalds 41907177a9c4SMiklos Szeredi if (!old_dir->i_op->rename && !old_dir->i_op->rename2) 41911da177e4SLinus Torvalds return -EPERM; 41921da177e4SLinus Torvalds 4193520c8b16SMiklos Szeredi if (flags && !old_dir->i_op->rename2) 4194520c8b16SMiklos Szeredi return -EINVAL; 4195520c8b16SMiklos Szeredi 4196bc27027aSMiklos Szeredi /* 4197bc27027aSMiklos Szeredi * If we are going to change the parent - check write permissions, 4198bc27027aSMiklos Szeredi * we'll need to flip '..'. 4199bc27027aSMiklos Szeredi */ 4200da1ce067SMiklos Szeredi if (new_dir != old_dir) { 4201da1ce067SMiklos Szeredi if (is_dir) { 4202bc27027aSMiklos Szeredi error = inode_permission(source, MAY_WRITE); 4203bc27027aSMiklos Szeredi if (error) 4204bc27027aSMiklos Szeredi return error; 4205bc27027aSMiklos Szeredi } 4206da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && new_is_dir) { 4207da1ce067SMiklos Szeredi error = inode_permission(target, MAY_WRITE); 4208da1ce067SMiklos Szeredi if (error) 4209da1ce067SMiklos Szeredi return error; 4210da1ce067SMiklos Szeredi } 4211da1ce067SMiklos Szeredi } 42120eeca283SRobert Love 42130b3974ebSMiklos Szeredi error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry, 42140b3974ebSMiklos Szeredi flags); 4215bc27027aSMiklos Szeredi if (error) 4216bc27027aSMiklos Szeredi return error; 4217bc27027aSMiklos Szeredi 4218bc27027aSMiklos Szeredi old_name = fsnotify_oldname_init(old_dentry->d_name.name); 4219bc27027aSMiklos Szeredi dget(new_dentry); 4220da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4221bc27027aSMiklos Szeredi lock_two_nondirectories(source, target); 4222bc27027aSMiklos Szeredi else if (target) 4223bc27027aSMiklos Szeredi mutex_lock(&target->i_mutex); 4224bc27027aSMiklos Szeredi 4225bc27027aSMiklos Szeredi error = -EBUSY; 42267af1364fSEric W. Biederman if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry)) 4227bc27027aSMiklos Szeredi goto out; 4228bc27027aSMiklos Szeredi 4229da1ce067SMiklos Szeredi if (max_links && new_dir != old_dir) { 4230bc27027aSMiklos Szeredi error = -EMLINK; 4231da1ce067SMiklos Szeredi if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links) 4232bc27027aSMiklos Szeredi goto out; 4233da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir && 4234da1ce067SMiklos Szeredi old_dir->i_nlink >= max_links) 4235da1ce067SMiklos Szeredi goto out; 4236da1ce067SMiklos Szeredi } 4237da1ce067SMiklos Szeredi if (is_dir && !(flags & RENAME_EXCHANGE) && target) 4238bc27027aSMiklos Szeredi shrink_dcache_parent(new_dentry); 4239da1ce067SMiklos Szeredi if (!is_dir) { 4240bc27027aSMiklos Szeredi error = try_break_deleg(source, delegated_inode); 4241bc27027aSMiklos Szeredi if (error) 4242bc27027aSMiklos Szeredi goto out; 4243da1ce067SMiklos Szeredi } 4244da1ce067SMiklos Szeredi if (target && !new_is_dir) { 4245bc27027aSMiklos Szeredi error = try_break_deleg(target, delegated_inode); 4246bc27027aSMiklos Szeredi if (error) 4247bc27027aSMiklos Szeredi goto out; 4248bc27027aSMiklos Szeredi } 42497177a9c4SMiklos Szeredi if (!old_dir->i_op->rename2) { 4250520c8b16SMiklos Szeredi error = old_dir->i_op->rename(old_dir, old_dentry, 4251520c8b16SMiklos Szeredi new_dir, new_dentry); 4252520c8b16SMiklos Szeredi } else { 42537177a9c4SMiklos Szeredi WARN_ON(old_dir->i_op->rename != NULL); 4254520c8b16SMiklos Szeredi error = old_dir->i_op->rename2(old_dir, old_dentry, 4255520c8b16SMiklos Szeredi new_dir, new_dentry, flags); 4256520c8b16SMiklos Szeredi } 4257bc27027aSMiklos Szeredi if (error) 4258bc27027aSMiklos Szeredi goto out; 4259bc27027aSMiklos Szeredi 4260da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE) && target) { 42611da177e4SLinus Torvalds if (is_dir) 4262bc27027aSMiklos Szeredi target->i_flags |= S_DEAD; 4263bc27027aSMiklos Szeredi dont_mount(new_dentry); 42648ed936b5SEric W. Biederman detach_mounts(new_dentry); 4265bc27027aSMiklos Szeredi } 4266da1ce067SMiklos Szeredi if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) { 4267da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 4268bc27027aSMiklos Szeredi d_move(old_dentry, new_dentry); 4269da1ce067SMiklos Szeredi else 4270da1ce067SMiklos Szeredi d_exchange(old_dentry, new_dentry); 4271da1ce067SMiklos Szeredi } 4272bc27027aSMiklos Szeredi out: 4273da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4274bc27027aSMiklos Szeredi unlock_two_nondirectories(source, target); 4275bc27027aSMiklos Szeredi else if (target) 4276bc27027aSMiklos Szeredi mutex_unlock(&target->i_mutex); 4277bc27027aSMiklos Szeredi dput(new_dentry); 4278da1ce067SMiklos Szeredi if (!error) { 4279123df294SAl Viro fsnotify_move(old_dir, new_dir, old_name, is_dir, 4280da1ce067SMiklos Szeredi !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry); 4281da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4282da1ce067SMiklos Szeredi fsnotify_move(new_dir, old_dir, old_dentry->d_name.name, 4283da1ce067SMiklos Szeredi new_is_dir, NULL, new_dentry); 4284da1ce067SMiklos Szeredi } 4285da1ce067SMiklos Szeredi } 42860eeca283SRobert Love fsnotify_oldname_free(old_name); 42870eeca283SRobert Love 42881da177e4SLinus Torvalds return error; 42891da177e4SLinus Torvalds } 42904d359507SAl Viro EXPORT_SYMBOL(vfs_rename); 42911da177e4SLinus Torvalds 4292520c8b16SMiklos Szeredi SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname, 4293520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname, unsigned int, flags) 42941da177e4SLinus Torvalds { 42951da177e4SLinus Torvalds struct dentry *old_dentry, *new_dentry; 42961da177e4SLinus Torvalds struct dentry *trap; 4297f5beed75SAl Viro struct path old_path, new_path; 4298f5beed75SAl Viro struct qstr old_last, new_last; 4299f5beed75SAl Viro int old_type, new_type; 43008e6d782cSJ. Bruce Fields struct inode *delegated_inode = NULL; 430191a27b2aSJeff Layton struct filename *from; 430291a27b2aSJeff Layton struct filename *to; 4303f5beed75SAl Viro unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET; 4304c6a94284SJeff Layton bool should_retry = false; 43052ad94ae6SAl Viro int error; 4306520c8b16SMiklos Szeredi 43070d7a8555SMiklos Szeredi if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) 4308da1ce067SMiklos Szeredi return -EINVAL; 4309da1ce067SMiklos Szeredi 43100d7a8555SMiklos Szeredi if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) && 43110d7a8555SMiklos Szeredi (flags & RENAME_EXCHANGE)) 4312520c8b16SMiklos Szeredi return -EINVAL; 4313520c8b16SMiklos Szeredi 43140d7a8555SMiklos Szeredi if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD)) 43150d7a8555SMiklos Szeredi return -EPERM; 43160d7a8555SMiklos Szeredi 4317f5beed75SAl Viro if (flags & RENAME_EXCHANGE) 4318f5beed75SAl Viro target_flags = 0; 4319f5beed75SAl Viro 4320c6a94284SJeff Layton retry: 4321f5beed75SAl Viro from = user_path_parent(olddfd, oldname, 4322f5beed75SAl Viro &old_path, &old_last, &old_type, lookup_flags); 432391a27b2aSJeff Layton if (IS_ERR(from)) { 432491a27b2aSJeff Layton error = PTR_ERR(from); 43251da177e4SLinus Torvalds goto exit; 432691a27b2aSJeff Layton } 43271da177e4SLinus Torvalds 4328f5beed75SAl Viro to = user_path_parent(newdfd, newname, 4329f5beed75SAl Viro &new_path, &new_last, &new_type, lookup_flags); 433091a27b2aSJeff Layton if (IS_ERR(to)) { 433191a27b2aSJeff Layton error = PTR_ERR(to); 43321da177e4SLinus Torvalds goto exit1; 433391a27b2aSJeff Layton } 43341da177e4SLinus Torvalds 43351da177e4SLinus Torvalds error = -EXDEV; 4336f5beed75SAl Viro if (old_path.mnt != new_path.mnt) 43371da177e4SLinus Torvalds goto exit2; 43381da177e4SLinus Torvalds 43391da177e4SLinus Torvalds error = -EBUSY; 4340f5beed75SAl Viro if (old_type != LAST_NORM) 43411da177e4SLinus Torvalds goto exit2; 43421da177e4SLinus Torvalds 43430a7c3937SMiklos Szeredi if (flags & RENAME_NOREPLACE) 43440a7c3937SMiklos Szeredi error = -EEXIST; 4345f5beed75SAl Viro if (new_type != LAST_NORM) 43461da177e4SLinus Torvalds goto exit2; 43471da177e4SLinus Torvalds 4348f5beed75SAl Viro error = mnt_want_write(old_path.mnt); 4349c30dabfeSJan Kara if (error) 4350c30dabfeSJan Kara goto exit2; 4351c30dabfeSJan Kara 43528e6d782cSJ. Bruce Fields retry_deleg: 4353f5beed75SAl Viro trap = lock_rename(new_path.dentry, old_path.dentry); 43541da177e4SLinus Torvalds 4355f5beed75SAl Viro old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags); 43561da177e4SLinus Torvalds error = PTR_ERR(old_dentry); 43571da177e4SLinus Torvalds if (IS_ERR(old_dentry)) 43581da177e4SLinus Torvalds goto exit3; 43591da177e4SLinus Torvalds /* source must exist */ 43601da177e4SLinus Torvalds error = -ENOENT; 4361b18825a7SDavid Howells if (d_is_negative(old_dentry)) 43621da177e4SLinus Torvalds goto exit4; 4363f5beed75SAl Viro new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags); 43641da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 43651da177e4SLinus Torvalds if (IS_ERR(new_dentry)) 43661da177e4SLinus Torvalds goto exit4; 43670a7c3937SMiklos Szeredi error = -EEXIST; 43680a7c3937SMiklos Szeredi if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry)) 43690a7c3937SMiklos Szeredi goto exit5; 4370da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4371da1ce067SMiklos Szeredi error = -ENOENT; 4372da1ce067SMiklos Szeredi if (d_is_negative(new_dentry)) 4373da1ce067SMiklos Szeredi goto exit5; 4374da1ce067SMiklos Szeredi 4375da1ce067SMiklos Szeredi if (!d_is_dir(new_dentry)) { 4376da1ce067SMiklos Szeredi error = -ENOTDIR; 4377f5beed75SAl Viro if (new_last.name[new_last.len]) 4378da1ce067SMiklos Szeredi goto exit5; 4379da1ce067SMiklos Szeredi } 4380da1ce067SMiklos Szeredi } 43810a7c3937SMiklos Szeredi /* unless the source is a directory trailing slashes give -ENOTDIR */ 43820a7c3937SMiklos Szeredi if (!d_is_dir(old_dentry)) { 43830a7c3937SMiklos Szeredi error = -ENOTDIR; 4384f5beed75SAl Viro if (old_last.name[old_last.len]) 43850a7c3937SMiklos Szeredi goto exit5; 4386f5beed75SAl Viro if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len]) 43870a7c3937SMiklos Szeredi goto exit5; 43880a7c3937SMiklos Szeredi } 43890a7c3937SMiklos Szeredi /* source should not be ancestor of target */ 43900a7c3937SMiklos Szeredi error = -EINVAL; 43910a7c3937SMiklos Szeredi if (old_dentry == trap) 43920a7c3937SMiklos Szeredi goto exit5; 43931da177e4SLinus Torvalds /* target should not be an ancestor of source */ 4394da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 43951da177e4SLinus Torvalds error = -ENOTEMPTY; 43961da177e4SLinus Torvalds if (new_dentry == trap) 43971da177e4SLinus Torvalds goto exit5; 43981da177e4SLinus Torvalds 4399f5beed75SAl Viro error = security_path_rename(&old_path, old_dentry, 4400f5beed75SAl Viro &new_path, new_dentry, flags); 4401be6d3e56SKentaro Takeda if (error) 4402c30dabfeSJan Kara goto exit5; 4403f5beed75SAl Viro error = vfs_rename(old_path.dentry->d_inode, old_dentry, 4404f5beed75SAl Viro new_path.dentry->d_inode, new_dentry, 4405520c8b16SMiklos Szeredi &delegated_inode, flags); 44061da177e4SLinus Torvalds exit5: 44071da177e4SLinus Torvalds dput(new_dentry); 44081da177e4SLinus Torvalds exit4: 44091da177e4SLinus Torvalds dput(old_dentry); 44101da177e4SLinus Torvalds exit3: 4411f5beed75SAl Viro unlock_rename(new_path.dentry, old_path.dentry); 44128e6d782cSJ. Bruce Fields if (delegated_inode) { 44138e6d782cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 44148e6d782cSJ. Bruce Fields if (!error) 44158e6d782cSJ. Bruce Fields goto retry_deleg; 44168e6d782cSJ. Bruce Fields } 4417f5beed75SAl Viro mnt_drop_write(old_path.mnt); 44181da177e4SLinus Torvalds exit2: 4419c6a94284SJeff Layton if (retry_estale(error, lookup_flags)) 4420c6a94284SJeff Layton should_retry = true; 4421f5beed75SAl Viro path_put(&new_path); 44222ad94ae6SAl Viro putname(to); 44231da177e4SLinus Torvalds exit1: 4424f5beed75SAl Viro path_put(&old_path); 44251da177e4SLinus Torvalds putname(from); 4426c6a94284SJeff Layton if (should_retry) { 4427c6a94284SJeff Layton should_retry = false; 4428c6a94284SJeff Layton lookup_flags |= LOOKUP_REVAL; 4429c6a94284SJeff Layton goto retry; 4430c6a94284SJeff Layton } 44312ad94ae6SAl Viro exit: 44321da177e4SLinus Torvalds return error; 44331da177e4SLinus Torvalds } 44341da177e4SLinus Torvalds 4435520c8b16SMiklos Szeredi SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, 4436520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname) 4437520c8b16SMiklos Szeredi { 4438520c8b16SMiklos Szeredi return sys_renameat2(olddfd, oldname, newdfd, newname, 0); 4439520c8b16SMiklos Szeredi } 4440520c8b16SMiklos Szeredi 4441a26eab24SHeiko Carstens SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) 44425590ff0dSUlrich Drepper { 4443520c8b16SMiklos Szeredi return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 44445590ff0dSUlrich Drepper } 44455590ff0dSUlrich Drepper 4446787fb6bcSMiklos Szeredi int vfs_whiteout(struct inode *dir, struct dentry *dentry) 4447787fb6bcSMiklos Szeredi { 4448787fb6bcSMiklos Szeredi int error = may_create(dir, dentry); 4449787fb6bcSMiklos Szeredi if (error) 4450787fb6bcSMiklos Szeredi return error; 4451787fb6bcSMiklos Szeredi 4452787fb6bcSMiklos Szeredi if (!dir->i_op->mknod) 4453787fb6bcSMiklos Szeredi return -EPERM; 4454787fb6bcSMiklos Szeredi 4455787fb6bcSMiklos Szeredi return dir->i_op->mknod(dir, dentry, 4456787fb6bcSMiklos Szeredi S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV); 4457787fb6bcSMiklos Szeredi } 4458787fb6bcSMiklos Szeredi EXPORT_SYMBOL(vfs_whiteout); 4459787fb6bcSMiklos Szeredi 44605d826c84SAl Viro int readlink_copy(char __user *buffer, int buflen, const char *link) 44611da177e4SLinus Torvalds { 44625d826c84SAl Viro int len = PTR_ERR(link); 44631da177e4SLinus Torvalds if (IS_ERR(link)) 44641da177e4SLinus Torvalds goto out; 44651da177e4SLinus Torvalds 44661da177e4SLinus Torvalds len = strlen(link); 44671da177e4SLinus Torvalds if (len > (unsigned) buflen) 44681da177e4SLinus Torvalds len = buflen; 44691da177e4SLinus Torvalds if (copy_to_user(buffer, link, len)) 44701da177e4SLinus Torvalds len = -EFAULT; 44711da177e4SLinus Torvalds out: 44721da177e4SLinus Torvalds return len; 44731da177e4SLinus Torvalds } 44745d826c84SAl Viro EXPORT_SYMBOL(readlink_copy); 44751da177e4SLinus Torvalds 44761da177e4SLinus Torvalds /* 44771da177e4SLinus Torvalds * A helper for ->readlink(). This should be used *ONLY* for symlinks that 44781da177e4SLinus Torvalds * have ->follow_link() touching nd only in nd_set_link(). Using (or not 44791da177e4SLinus Torvalds * using) it for any given inode is up to filesystem. 44801da177e4SLinus Torvalds */ 44811da177e4SLinus Torvalds int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen) 44821da177e4SLinus Torvalds { 4483cc314eefSLinus Torvalds void *cookie; 44845f2c4179SAl Viro struct inode *inode = d_inode(dentry); 44855f2c4179SAl Viro const char *link = inode->i_link; 4486694a1764SMarcin Slusarz int res; 4487cc314eefSLinus Torvalds 4488d4dee48bSAl Viro if (!link) { 44895f2c4179SAl Viro link = inode->i_op->follow_link(dentry, &cookie); 4490680baacbSAl Viro if (IS_ERR(link)) 4491680baacbSAl Viro return PTR_ERR(link); 4492d4dee48bSAl Viro } 4493680baacbSAl Viro res = readlink_copy(buffer, buflen, link); 44945f2c4179SAl Viro if (inode->i_op->put_link) 44955f2c4179SAl Viro inode->i_op->put_link(inode, cookie); 4496694a1764SMarcin Slusarz return res; 44971da177e4SLinus Torvalds } 44984d359507SAl Viro EXPORT_SYMBOL(generic_readlink); 44991da177e4SLinus Torvalds 45001da177e4SLinus Torvalds /* get the link contents into pagecache */ 45011da177e4SLinus Torvalds static char *page_getlink(struct dentry * dentry, struct page **ppage) 45021da177e4SLinus Torvalds { 4503ebd09abbSDuane Griffin char *kaddr; 45041da177e4SLinus Torvalds struct page *page; 45051da177e4SLinus Torvalds struct address_space *mapping = dentry->d_inode->i_mapping; 4506090d2b18SPekka Enberg page = read_mapping_page(mapping, 0, NULL); 45071da177e4SLinus Torvalds if (IS_ERR(page)) 45086fe6900eSNick Piggin return (char*)page; 45091da177e4SLinus Torvalds *ppage = page; 4510ebd09abbSDuane Griffin kaddr = kmap(page); 4511ebd09abbSDuane Griffin nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1); 4512ebd09abbSDuane Griffin return kaddr; 45131da177e4SLinus Torvalds } 45141da177e4SLinus Torvalds 45151da177e4SLinus Torvalds int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 45161da177e4SLinus Torvalds { 45171da177e4SLinus Torvalds struct page *page = NULL; 45185d826c84SAl Viro int res = readlink_copy(buffer, buflen, page_getlink(dentry, &page)); 45191da177e4SLinus Torvalds if (page) { 45201da177e4SLinus Torvalds kunmap(page); 45211da177e4SLinus Torvalds page_cache_release(page); 45221da177e4SLinus Torvalds } 45231da177e4SLinus Torvalds return res; 45241da177e4SLinus Torvalds } 45254d359507SAl Viro EXPORT_SYMBOL(page_readlink); 45261da177e4SLinus Torvalds 45276e77137bSAl Viro const char *page_follow_link_light(struct dentry *dentry, void **cookie) 45281da177e4SLinus Torvalds { 4529cc314eefSLinus Torvalds struct page *page = NULL; 4530680baacbSAl Viro char *res = page_getlink(dentry, &page); 4531680baacbSAl Viro if (!IS_ERR(res)) 4532680baacbSAl Viro *cookie = page; 4533680baacbSAl Viro return res; 45341da177e4SLinus Torvalds } 45354d359507SAl Viro EXPORT_SYMBOL(page_follow_link_light); 45361da177e4SLinus Torvalds 45375f2c4179SAl Viro void page_put_link(struct inode *unused, void *cookie) 45381da177e4SLinus Torvalds { 4539cc314eefSLinus Torvalds struct page *page = cookie; 45401da177e4SLinus Torvalds kunmap(page); 45411da177e4SLinus Torvalds page_cache_release(page); 45421da177e4SLinus Torvalds } 45434d359507SAl Viro EXPORT_SYMBOL(page_put_link); 45441da177e4SLinus Torvalds 454554566b2cSNick Piggin /* 454654566b2cSNick Piggin * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS 454754566b2cSNick Piggin */ 454854566b2cSNick Piggin int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) 45491da177e4SLinus Torvalds { 45501da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 45510adb25d2SKirill Korotaev struct page *page; 4552afddba49SNick Piggin void *fsdata; 4553beb497abSDmitriy Monakhov int err; 45541da177e4SLinus Torvalds char *kaddr; 455554566b2cSNick Piggin unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE; 455654566b2cSNick Piggin if (nofs) 455754566b2cSNick Piggin flags |= AOP_FLAG_NOFS; 45581da177e4SLinus Torvalds 45597e53cac4SNeilBrown retry: 4560afddba49SNick Piggin err = pagecache_write_begin(NULL, mapping, 0, len-1, 456154566b2cSNick Piggin flags, &page, &fsdata); 45621da177e4SLinus Torvalds if (err) 4563afddba49SNick Piggin goto fail; 4564afddba49SNick Piggin 4565e8e3c3d6SCong Wang kaddr = kmap_atomic(page); 45661da177e4SLinus Torvalds memcpy(kaddr, symname, len-1); 4567e8e3c3d6SCong Wang kunmap_atomic(kaddr); 4568afddba49SNick Piggin 4569afddba49SNick Piggin err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, 4570afddba49SNick Piggin page, fsdata); 45711da177e4SLinus Torvalds if (err < 0) 45721da177e4SLinus Torvalds goto fail; 4573afddba49SNick Piggin if (err < len-1) 4574afddba49SNick Piggin goto retry; 4575afddba49SNick Piggin 45761da177e4SLinus Torvalds mark_inode_dirty(inode); 45771da177e4SLinus Torvalds return 0; 45781da177e4SLinus Torvalds fail: 45791da177e4SLinus Torvalds return err; 45801da177e4SLinus Torvalds } 45814d359507SAl Viro EXPORT_SYMBOL(__page_symlink); 45821da177e4SLinus Torvalds 45830adb25d2SKirill Korotaev int page_symlink(struct inode *inode, const char *symname, int len) 45840adb25d2SKirill Korotaev { 45850adb25d2SKirill Korotaev return __page_symlink(inode, symname, len, 458654566b2cSNick Piggin !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS)); 45870adb25d2SKirill Korotaev } 45884d359507SAl Viro EXPORT_SYMBOL(page_symlink); 45890adb25d2SKirill Korotaev 459092e1d5beSArjan van de Ven const struct inode_operations page_symlink_inode_operations = { 45911da177e4SLinus Torvalds .readlink = generic_readlink, 45921da177e4SLinus Torvalds .follow_link = page_follow_link_light, 45931da177e4SLinus Torvalds .put_link = page_put_link, 45941da177e4SLinus Torvalds }; 45951da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink_inode_operations); 4596