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> 382a18da7aSGeorge Spelvin #include <linux/bitops.h> 39aeaa4a79SEric W. Biederman #include <linux/init_task.h> 407c0f6ba6SLinus Torvalds #include <linux/uaccess.h> 411da177e4SLinus Torvalds 42e81e3f4dSEric Paris #include "internal.h" 43c7105365SAl Viro #include "mount.h" 44e81e3f4dSEric Paris 451da177e4SLinus Torvalds /* [Feb-1997 T. Schoebel-Theuer] 461da177e4SLinus Torvalds * Fundamental changes in the pathname lookup mechanisms (namei) 471da177e4SLinus Torvalds * were necessary because of omirr. The reason is that omirr needs 481da177e4SLinus Torvalds * to know the _real_ pathname, not the user-supplied one, in case 491da177e4SLinus Torvalds * of symlinks (and also when transname replacements occur). 501da177e4SLinus Torvalds * 511da177e4SLinus Torvalds * The new code replaces the old recursive symlink resolution with 521da177e4SLinus Torvalds * an iterative one (in case of non-nested symlink chains). It does 531da177e4SLinus Torvalds * this with calls to <fs>_follow_link(). 541da177e4SLinus Torvalds * As a side effect, dir_namei(), _namei() and follow_link() are now 551da177e4SLinus Torvalds * replaced with a single function lookup_dentry() that can handle all 561da177e4SLinus Torvalds * the special cases of the former code. 571da177e4SLinus Torvalds * 581da177e4SLinus Torvalds * With the new dcache, the pathname is stored at each inode, at least as 591da177e4SLinus Torvalds * long as the refcount of the inode is positive. As a side effect, the 601da177e4SLinus Torvalds * size of the dcache depends on the inode cache and thus is dynamic. 611da177e4SLinus Torvalds * 621da177e4SLinus Torvalds * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink 631da177e4SLinus Torvalds * resolution to correspond with current state of the code. 641da177e4SLinus Torvalds * 651da177e4SLinus Torvalds * Note that the symlink resolution is not *completely* iterative. 661da177e4SLinus Torvalds * There is still a significant amount of tail- and mid- recursion in 671da177e4SLinus Torvalds * the algorithm. Also, note that <fs>_readlink() is not used in 681da177e4SLinus Torvalds * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink() 691da177e4SLinus Torvalds * may return different results than <fs>_follow_link(). Many virtual 701da177e4SLinus Torvalds * filesystems (including /proc) exhibit this behavior. 711da177e4SLinus Torvalds */ 721da177e4SLinus Torvalds 731da177e4SLinus Torvalds /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation: 741da177e4SLinus Torvalds * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL 751da177e4SLinus Torvalds * and the name already exists in form of a symlink, try to create the new 761da177e4SLinus Torvalds * name indicated by the symlink. The old code always complained that the 771da177e4SLinus Torvalds * name already exists, due to not following the symlink even if its target 781da177e4SLinus Torvalds * is nonexistent. The new semantics affects also mknod() and link() when 7925985edcSLucas De Marchi * the name is a symlink pointing to a non-existent name. 801da177e4SLinus Torvalds * 811da177e4SLinus Torvalds * I don't know which semantics is the right one, since I have no access 821da177e4SLinus Torvalds * to standards. But I found by trial that HP-UX 9.0 has the full "new" 831da177e4SLinus Torvalds * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the 841da177e4SLinus Torvalds * "old" one. Personally, I think the new semantics is much more logical. 851da177e4SLinus Torvalds * Note that "ln old new" where "new" is a symlink pointing to a non-existing 861da177e4SLinus Torvalds * file does succeed in both HP-UX and SunOs, but not in Solaris 871da177e4SLinus Torvalds * and in the old Linux semantics. 881da177e4SLinus Torvalds */ 891da177e4SLinus Torvalds 901da177e4SLinus Torvalds /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink 911da177e4SLinus Torvalds * semantics. See the comments in "open_namei" and "do_link" below. 921da177e4SLinus Torvalds * 931da177e4SLinus Torvalds * [10-Sep-98 Alan Modra] Another symlink change. 941da177e4SLinus Torvalds */ 951da177e4SLinus Torvalds 961da177e4SLinus Torvalds /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks: 971da177e4SLinus Torvalds * inside the path - always follow. 981da177e4SLinus Torvalds * in the last component in creation/removal/renaming - never follow. 991da177e4SLinus Torvalds * if LOOKUP_FOLLOW passed - follow. 1001da177e4SLinus Torvalds * if the pathname has trailing slashes - follow. 1011da177e4SLinus Torvalds * otherwise - don't follow. 1021da177e4SLinus Torvalds * (applied in that order). 1031da177e4SLinus Torvalds * 1041da177e4SLinus Torvalds * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT 1051da177e4SLinus Torvalds * restored for 2.4. This is the last surviving part of old 4.2BSD bug. 1061da177e4SLinus Torvalds * During the 2.4 we need to fix the userland stuff depending on it - 1071da177e4SLinus Torvalds * hopefully we will be able to get rid of that wart in 2.5. So far only 1081da177e4SLinus Torvalds * XEmacs seems to be relying on it... 1091da177e4SLinus Torvalds */ 1101da177e4SLinus Torvalds /* 1111da177e4SLinus Torvalds * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland) 112a11f3a05SArjan van de Ven * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives 1131da177e4SLinus Torvalds * any extra contention... 1141da177e4SLinus Torvalds */ 1151da177e4SLinus Torvalds 1161da177e4SLinus Torvalds /* In order to reduce some races, while at the same time doing additional 1171da177e4SLinus Torvalds * checking and hopefully speeding things up, we copy filenames to the 1181da177e4SLinus Torvalds * kernel data space before using them.. 1191da177e4SLinus Torvalds * 1201da177e4SLinus Torvalds * POSIX.1 2.4: an empty pathname is invalid (ENOENT). 1211da177e4SLinus Torvalds * PATH_MAX includes the nul terminator --RR. 1221da177e4SLinus Torvalds */ 1237950e385SJeff Layton 124fd2f7cb5SAl Viro #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname)) 12591a27b2aSJeff Layton 12651f39a1fSDavid Drysdale struct filename * 12791a27b2aSJeff Layton getname_flags(const char __user *filename, int flags, int *empty) 12891a27b2aSJeff Layton { 12994b5d262SAl Viro struct filename *result; 1307950e385SJeff Layton char *kname; 13194b5d262SAl Viro int len; 1321da177e4SLinus Torvalds 1337ac86265SJeff Layton result = audit_reusename(filename); 1347ac86265SJeff Layton if (result) 1357ac86265SJeff Layton return result; 1367ac86265SJeff Layton 1377950e385SJeff Layton result = __getname(); 1383f9f0aa6SLinus Torvalds if (unlikely(!result)) 1394043cde8SEric Paris return ERR_PTR(-ENOMEM); 1401da177e4SLinus Torvalds 1417950e385SJeff Layton /* 1427950e385SJeff Layton * First, try to embed the struct filename inside the names_cache 1437950e385SJeff Layton * allocation 1447950e385SJeff Layton */ 145fd2f7cb5SAl Viro kname = (char *)result->iname; 14691a27b2aSJeff Layton result->name = kname; 1477950e385SJeff Layton 14894b5d262SAl Viro len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX); 14991a27b2aSJeff Layton if (unlikely(len < 0)) { 15094b5d262SAl Viro __putname(result); 15194b5d262SAl Viro return ERR_PTR(len); 15291a27b2aSJeff Layton } 1533f9f0aa6SLinus Torvalds 1547950e385SJeff Layton /* 1557950e385SJeff Layton * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a 1567950e385SJeff Layton * separate struct filename so we can dedicate the entire 1577950e385SJeff Layton * names_cache allocation for the pathname, and re-do the copy from 1587950e385SJeff Layton * userland. 1597950e385SJeff Layton */ 16094b5d262SAl Viro if (unlikely(len == EMBEDDED_NAME_MAX)) { 161fd2f7cb5SAl Viro const size_t size = offsetof(struct filename, iname[1]); 1627950e385SJeff Layton kname = (char *)result; 1637950e385SJeff Layton 164fd2f7cb5SAl Viro /* 165fd2f7cb5SAl Viro * size is chosen that way we to guarantee that 166fd2f7cb5SAl Viro * result->iname[0] is within the same object and that 167fd2f7cb5SAl Viro * kname can't be equal to result->iname, no matter what. 168fd2f7cb5SAl Viro */ 169fd2f7cb5SAl Viro result = kzalloc(size, GFP_KERNEL); 17094b5d262SAl Viro if (unlikely(!result)) { 17194b5d262SAl Viro __putname(kname); 17294b5d262SAl Viro return ERR_PTR(-ENOMEM); 1737950e385SJeff Layton } 1747950e385SJeff Layton result->name = kname; 17594b5d262SAl Viro len = strncpy_from_user(kname, filename, PATH_MAX); 17694b5d262SAl Viro if (unlikely(len < 0)) { 17794b5d262SAl Viro __putname(kname); 17894b5d262SAl Viro kfree(result); 17994b5d262SAl Viro return ERR_PTR(len); 18094b5d262SAl Viro } 18194b5d262SAl Viro if (unlikely(len == PATH_MAX)) { 18294b5d262SAl Viro __putname(kname); 18394b5d262SAl Viro kfree(result); 18494b5d262SAl Viro return ERR_PTR(-ENAMETOOLONG); 18594b5d262SAl Viro } 1867950e385SJeff Layton } 1877950e385SJeff Layton 18894b5d262SAl Viro result->refcnt = 1; 1893f9f0aa6SLinus Torvalds /* The empty path is special. */ 1903f9f0aa6SLinus Torvalds if (unlikely(!len)) { 1913f9f0aa6SLinus Torvalds if (empty) 1921fa1e7f6SAndy Whitcroft *empty = 1; 19394b5d262SAl Viro if (!(flags & LOOKUP_EMPTY)) { 19494b5d262SAl Viro putname(result); 19594b5d262SAl Viro return ERR_PTR(-ENOENT); 1961da177e4SLinus Torvalds } 19794b5d262SAl Viro } 1987950e385SJeff Layton 1997950e385SJeff Layton result->uptr = filename; 200c4ad8f98SLinus Torvalds result->aname = NULL; 2011da177e4SLinus Torvalds audit_getname(result); 2021da177e4SLinus Torvalds return result; 2033f9f0aa6SLinus Torvalds } 2043f9f0aa6SLinus Torvalds 20591a27b2aSJeff Layton struct filename * 20691a27b2aSJeff Layton getname(const char __user * filename) 207f52e0c11SAl Viro { 208f7493e5dSLinus Torvalds return getname_flags(filename, 0, NULL); 209f52e0c11SAl Viro } 210f52e0c11SAl Viro 211c4ad8f98SLinus Torvalds struct filename * 212c4ad8f98SLinus Torvalds getname_kernel(const char * filename) 213c4ad8f98SLinus Torvalds { 214c4ad8f98SLinus Torvalds struct filename *result; 21508518549SPaul Moore int len = strlen(filename) + 1; 216c4ad8f98SLinus Torvalds 217c4ad8f98SLinus Torvalds result = __getname(); 218c4ad8f98SLinus Torvalds if (unlikely(!result)) 219c4ad8f98SLinus Torvalds return ERR_PTR(-ENOMEM); 220c4ad8f98SLinus Torvalds 22108518549SPaul Moore if (len <= EMBEDDED_NAME_MAX) { 222fd2f7cb5SAl Viro result->name = (char *)result->iname; 22308518549SPaul Moore } else if (len <= PATH_MAX) { 22408518549SPaul Moore struct filename *tmp; 22508518549SPaul Moore 22608518549SPaul Moore tmp = kmalloc(sizeof(*tmp), GFP_KERNEL); 22708518549SPaul Moore if (unlikely(!tmp)) { 22808518549SPaul Moore __putname(result); 22908518549SPaul Moore return ERR_PTR(-ENOMEM); 23008518549SPaul Moore } 23108518549SPaul Moore tmp->name = (char *)result; 23208518549SPaul Moore result = tmp; 23308518549SPaul Moore } else { 23408518549SPaul Moore __putname(result); 23508518549SPaul Moore return ERR_PTR(-ENAMETOOLONG); 23608518549SPaul Moore } 23708518549SPaul Moore memcpy((char *)result->name, filename, len); 238c4ad8f98SLinus Torvalds result->uptr = NULL; 239c4ad8f98SLinus Torvalds result->aname = NULL; 24055422d0bSPaul Moore result->refcnt = 1; 241fd3522fdSPaul Moore audit_getname(result); 242c4ad8f98SLinus Torvalds 243c4ad8f98SLinus Torvalds return result; 244c4ad8f98SLinus Torvalds } 245c4ad8f98SLinus Torvalds 24691a27b2aSJeff Layton void putname(struct filename *name) 2471da177e4SLinus Torvalds { 24855422d0bSPaul Moore BUG_ON(name->refcnt <= 0); 24955422d0bSPaul Moore 25055422d0bSPaul Moore if (--name->refcnt > 0) 25155422d0bSPaul Moore return; 25255422d0bSPaul Moore 253fd2f7cb5SAl Viro if (name->name != name->iname) { 25455422d0bSPaul Moore __putname(name->name); 25555422d0bSPaul Moore kfree(name); 25655422d0bSPaul Moore } else 25755422d0bSPaul Moore __putname(name); 2581da177e4SLinus Torvalds } 2591da177e4SLinus Torvalds 260e77819e5SLinus Torvalds static int check_acl(struct inode *inode, int mask) 261e77819e5SLinus Torvalds { 26284635d68SLinus Torvalds #ifdef CONFIG_FS_POSIX_ACL 263e77819e5SLinus Torvalds struct posix_acl *acl; 264e77819e5SLinus Torvalds 265e77819e5SLinus Torvalds if (mask & MAY_NOT_BLOCK) { 2663567866bSAl Viro acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS); 2673567866bSAl Viro if (!acl) 268e77819e5SLinus Torvalds return -EAGAIN; 2693567866bSAl Viro /* no ->get_acl() calls in RCU mode... */ 270b8a7a3a6SAndreas Gruenbacher if (is_uncached_acl(acl)) 271e77819e5SLinus Torvalds return -ECHILD; 272206b1d09SAri Savolainen return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK); 273e77819e5SLinus Torvalds } 274e77819e5SLinus Torvalds 2752982baa2SChristoph Hellwig acl = get_acl(inode, ACL_TYPE_ACCESS); 2764e34e719SChristoph Hellwig if (IS_ERR(acl)) 2774e34e719SChristoph Hellwig return PTR_ERR(acl); 278e77819e5SLinus Torvalds if (acl) { 279e77819e5SLinus Torvalds int error = posix_acl_permission(inode, acl, mask); 280e77819e5SLinus Torvalds posix_acl_release(acl); 281e77819e5SLinus Torvalds return error; 282e77819e5SLinus Torvalds } 28384635d68SLinus Torvalds #endif 284e77819e5SLinus Torvalds 285e77819e5SLinus Torvalds return -EAGAIN; 286e77819e5SLinus Torvalds } 287e77819e5SLinus Torvalds 2885909ccaaSLinus Torvalds /* 289948409c7SAndreas Gruenbacher * This does the basic permission checking 2905909ccaaSLinus Torvalds */ 2917e40145eSAl Viro static int acl_permission_check(struct inode *inode, int mask) 2925909ccaaSLinus Torvalds { 29326cf46beSLinus Torvalds unsigned int mode = inode->i_mode; 2945909ccaaSLinus Torvalds 2958e96e3b7SEric W. Biederman if (likely(uid_eq(current_fsuid(), inode->i_uid))) 2965909ccaaSLinus Torvalds mode >>= 6; 2975909ccaaSLinus Torvalds else { 298e77819e5SLinus Torvalds if (IS_POSIXACL(inode) && (mode & S_IRWXG)) { 2997e40145eSAl Viro int error = check_acl(inode, mask); 3005909ccaaSLinus Torvalds if (error != -EAGAIN) 3015909ccaaSLinus Torvalds return error; 3025909ccaaSLinus Torvalds } 3035909ccaaSLinus Torvalds 3045909ccaaSLinus Torvalds if (in_group_p(inode->i_gid)) 3055909ccaaSLinus Torvalds mode >>= 3; 3065909ccaaSLinus Torvalds } 3075909ccaaSLinus Torvalds 3085909ccaaSLinus Torvalds /* 3095909ccaaSLinus Torvalds * If the DACs are ok we don't need any capability check. 3105909ccaaSLinus Torvalds */ 3119c2c7039SAl Viro if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) 3125909ccaaSLinus Torvalds return 0; 3135909ccaaSLinus Torvalds return -EACCES; 3145909ccaaSLinus Torvalds } 3151da177e4SLinus Torvalds 3161da177e4SLinus Torvalds /** 3171da177e4SLinus Torvalds * generic_permission - check for access rights on a Posix-like filesystem 3181da177e4SLinus Torvalds * @inode: inode to check access rights for 3198fd90c8dSAndreas Gruenbacher * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...) 3201da177e4SLinus Torvalds * 3211da177e4SLinus Torvalds * Used to check for read/write/execute permissions on a file. 3221da177e4SLinus Torvalds * We use "fsuid" for this, letting us set arbitrary permissions 3231da177e4SLinus Torvalds * for filesystem access without changing the "normal" uids which 324b74c79e9SNick Piggin * are used for other things. 325b74c79e9SNick Piggin * 326b74c79e9SNick Piggin * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk 327b74c79e9SNick Piggin * request cannot be satisfied (eg. requires blocking or too much complexity). 328b74c79e9SNick Piggin * It would then be called again in ref-walk mode. 3291da177e4SLinus Torvalds */ 3302830ba7fSAl Viro int generic_permission(struct inode *inode, int mask) 3311da177e4SLinus Torvalds { 3325909ccaaSLinus Torvalds int ret; 3331da177e4SLinus Torvalds 3341da177e4SLinus Torvalds /* 335948409c7SAndreas Gruenbacher * Do the basic permission checks. 3361da177e4SLinus Torvalds */ 3377e40145eSAl Viro ret = acl_permission_check(inode, mask); 3385909ccaaSLinus Torvalds if (ret != -EACCES) 3395909ccaaSLinus Torvalds return ret; 3401da177e4SLinus Torvalds 341d594e7ecSAl Viro if (S_ISDIR(inode->i_mode)) { 342d594e7ecSAl Viro /* DACs are overridable for directories */ 34323adbe12SAndy Lutomirski if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE)) 344d594e7ecSAl Viro return 0; 345d594e7ecSAl Viro if (!(mask & MAY_WRITE)) 34623adbe12SAndy Lutomirski if (capable_wrt_inode_uidgid(inode, 34723adbe12SAndy Lutomirski CAP_DAC_READ_SEARCH)) 348d594e7ecSAl Viro return 0; 349d594e7ecSAl Viro return -EACCES; 350d594e7ecSAl Viro } 3511da177e4SLinus Torvalds /* 3521da177e4SLinus Torvalds * Read/write DACs are always overridable. 353d594e7ecSAl Viro * Executable DACs are overridable when there is 354d594e7ecSAl Viro * at least one exec bit set. 3551da177e4SLinus Torvalds */ 356d594e7ecSAl Viro if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO)) 35723adbe12SAndy Lutomirski if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE)) 3581da177e4SLinus Torvalds return 0; 3591da177e4SLinus Torvalds 3601da177e4SLinus Torvalds /* 3611da177e4SLinus Torvalds * Searching includes executable on directories, else just read. 3621da177e4SLinus Torvalds */ 3637ea66001SSerge E. Hallyn mask &= MAY_READ | MAY_WRITE | MAY_EXEC; 364d594e7ecSAl Viro if (mask == MAY_READ) 36523adbe12SAndy Lutomirski if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH)) 3661da177e4SLinus Torvalds return 0; 3671da177e4SLinus Torvalds 3681da177e4SLinus Torvalds return -EACCES; 3691da177e4SLinus Torvalds } 3704d359507SAl Viro EXPORT_SYMBOL(generic_permission); 3711da177e4SLinus Torvalds 3723ddcd056SLinus Torvalds /* 3733ddcd056SLinus Torvalds * We _really_ want to just do "generic_permission()" without 3743ddcd056SLinus Torvalds * even looking at the inode->i_op values. So we keep a cache 3753ddcd056SLinus Torvalds * flag in inode->i_opflags, that says "this has not special 3763ddcd056SLinus Torvalds * permission function, use the fast case". 3773ddcd056SLinus Torvalds */ 3783ddcd056SLinus Torvalds static inline int do_inode_permission(struct inode *inode, int mask) 3793ddcd056SLinus Torvalds { 3803ddcd056SLinus Torvalds if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) { 3813ddcd056SLinus Torvalds if (likely(inode->i_op->permission)) 3823ddcd056SLinus Torvalds return inode->i_op->permission(inode, mask); 3833ddcd056SLinus Torvalds 3843ddcd056SLinus Torvalds /* This gets set once for the inode lifetime */ 3853ddcd056SLinus Torvalds spin_lock(&inode->i_lock); 3863ddcd056SLinus Torvalds inode->i_opflags |= IOP_FASTPERM; 3873ddcd056SLinus Torvalds spin_unlock(&inode->i_lock); 3883ddcd056SLinus Torvalds } 3893ddcd056SLinus Torvalds return generic_permission(inode, mask); 3903ddcd056SLinus Torvalds } 3913ddcd056SLinus Torvalds 392cb23beb5SChristoph Hellwig /** 3930bdaea90SDavid Howells * __inode_permission - Check for access rights to a given inode 3940bdaea90SDavid Howells * @inode: Inode to check permission on 3950bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 396cb23beb5SChristoph Hellwig * 3970bdaea90SDavid Howells * Check for read/write/execute permissions on an inode. 398948409c7SAndreas Gruenbacher * 399948409c7SAndreas Gruenbacher * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. 4000bdaea90SDavid Howells * 4010bdaea90SDavid Howells * This does not check for a read-only file system. You probably want 4020bdaea90SDavid Howells * inode_permission(). 403cb23beb5SChristoph Hellwig */ 4040bdaea90SDavid Howells int __inode_permission(struct inode *inode, int mask) 4051da177e4SLinus Torvalds { 406e6305c43SAl Viro int retval; 4071da177e4SLinus Torvalds 4083ddcd056SLinus Torvalds if (unlikely(mask & MAY_WRITE)) { 4091da177e4SLinus Torvalds /* 4101da177e4SLinus Torvalds * Nobody gets write access to an immutable file. 4111da177e4SLinus Torvalds */ 4121da177e4SLinus Torvalds if (IS_IMMUTABLE(inode)) 413337684a1SEryu Guan return -EPERM; 4140bd23d09SEric W. Biederman 4150bd23d09SEric W. Biederman /* 4160bd23d09SEric W. Biederman * Updating mtime will likely cause i_uid and i_gid to be 4170bd23d09SEric W. Biederman * written back improperly if their true value is unknown 4180bd23d09SEric W. Biederman * to the vfs. 4190bd23d09SEric W. Biederman */ 4200bd23d09SEric W. Biederman if (HAS_UNMAPPED_ID(inode)) 4210bd23d09SEric W. Biederman return -EACCES; 4221da177e4SLinus Torvalds } 4231da177e4SLinus Torvalds 4243ddcd056SLinus Torvalds retval = do_inode_permission(inode, mask); 4251da177e4SLinus Torvalds if (retval) 4261da177e4SLinus Torvalds return retval; 4271da177e4SLinus Torvalds 42808ce5f16SSerge E. Hallyn retval = devcgroup_inode_permission(inode, mask); 42908ce5f16SSerge E. Hallyn if (retval) 43008ce5f16SSerge E. Hallyn return retval; 43108ce5f16SSerge E. Hallyn 432d09ca739SEric Paris return security_inode_permission(inode, mask); 4331da177e4SLinus Torvalds } 434bd5d0856SMiklos Szeredi EXPORT_SYMBOL(__inode_permission); 4351da177e4SLinus Torvalds 436f4d6ff89SAl Viro /** 4370bdaea90SDavid Howells * sb_permission - Check superblock-level permissions 4380bdaea90SDavid Howells * @sb: Superblock of inode to check permission on 43955852635SRandy Dunlap * @inode: Inode to check permission on 4400bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 4410bdaea90SDavid Howells * 4420bdaea90SDavid Howells * Separate out file-system wide checks from inode-specific permission checks. 4430bdaea90SDavid Howells */ 4440bdaea90SDavid Howells static int sb_permission(struct super_block *sb, struct inode *inode, int mask) 4450bdaea90SDavid Howells { 4460bdaea90SDavid Howells if (unlikely(mask & MAY_WRITE)) { 4470bdaea90SDavid Howells umode_t mode = inode->i_mode; 4480bdaea90SDavid Howells 4490bdaea90SDavid Howells /* Nobody gets write access to a read-only fs. */ 4500bdaea90SDavid Howells if ((sb->s_flags & MS_RDONLY) && 4510bdaea90SDavid Howells (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) 4520bdaea90SDavid Howells return -EROFS; 4530bdaea90SDavid Howells } 4540bdaea90SDavid Howells return 0; 4550bdaea90SDavid Howells } 4560bdaea90SDavid Howells 4570bdaea90SDavid Howells /** 4580bdaea90SDavid Howells * inode_permission - Check for access rights to a given inode 4590bdaea90SDavid Howells * @inode: Inode to check permission on 4600bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 4610bdaea90SDavid Howells * 4620bdaea90SDavid Howells * Check for read/write/execute permissions on an inode. We use fs[ug]id for 4630bdaea90SDavid Howells * this, letting us set arbitrary permissions for filesystem access without 4640bdaea90SDavid Howells * changing the "normal" UIDs which are used for other things. 4650bdaea90SDavid Howells * 4660bdaea90SDavid Howells * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. 4670bdaea90SDavid Howells */ 4680bdaea90SDavid Howells int inode_permission(struct inode *inode, int mask) 4690bdaea90SDavid Howells { 4700bdaea90SDavid Howells int retval; 4710bdaea90SDavid Howells 4720bdaea90SDavid Howells retval = sb_permission(inode->i_sb, inode, mask); 4730bdaea90SDavid Howells if (retval) 4740bdaea90SDavid Howells return retval; 4750bdaea90SDavid Howells return __inode_permission(inode, mask); 4760bdaea90SDavid Howells } 4774d359507SAl Viro EXPORT_SYMBOL(inode_permission); 4780bdaea90SDavid Howells 4790bdaea90SDavid Howells /** 4805dd784d0SJan Blunck * path_get - get a reference to a path 4815dd784d0SJan Blunck * @path: path to get the reference to 4825dd784d0SJan Blunck * 4835dd784d0SJan Blunck * Given a path increment the reference count to the dentry and the vfsmount. 4845dd784d0SJan Blunck */ 485dcf787f3SAl Viro void path_get(const struct path *path) 4865dd784d0SJan Blunck { 4875dd784d0SJan Blunck mntget(path->mnt); 4885dd784d0SJan Blunck dget(path->dentry); 4895dd784d0SJan Blunck } 4905dd784d0SJan Blunck EXPORT_SYMBOL(path_get); 4915dd784d0SJan Blunck 4925dd784d0SJan Blunck /** 4931d957f9bSJan Blunck * path_put - put a reference to a path 4941d957f9bSJan Blunck * @path: path to put the reference to 4951d957f9bSJan Blunck * 4961d957f9bSJan Blunck * Given a path decrement the reference count to the dentry and the vfsmount. 4971d957f9bSJan Blunck */ 498dcf787f3SAl Viro void path_put(const struct path *path) 4991da177e4SLinus Torvalds { 5001d957f9bSJan Blunck dput(path->dentry); 5011d957f9bSJan Blunck mntput(path->mnt); 5021da177e4SLinus Torvalds } 5031d957f9bSJan Blunck EXPORT_SYMBOL(path_put); 5041da177e4SLinus Torvalds 505894bc8c4SAl Viro #define EMBEDDED_LEVELS 2 5061f55a6ecSAl Viro struct nameidata { 5071f55a6ecSAl Viro struct path path; 5081f55a6ecSAl Viro struct qstr last; 5091f55a6ecSAl Viro struct path root; 5101f55a6ecSAl Viro struct inode *inode; /* path.dentry.d_inode */ 5111f55a6ecSAl Viro unsigned int flags; 5129883d185SAl Viro unsigned seq, m_seq; 5131f55a6ecSAl Viro int last_type; 5141f55a6ecSAl Viro unsigned depth; 515756daf26SNeilBrown int total_link_count; 516697fc6caSAl Viro struct saved { 517697fc6caSAl Viro struct path link; 518fceef393SAl Viro struct delayed_call done; 519697fc6caSAl Viro const char *name; 5200450b2d1SAl Viro unsigned seq; 521894bc8c4SAl Viro } *stack, internal[EMBEDDED_LEVELS]; 5229883d185SAl Viro struct filename *name; 5239883d185SAl Viro struct nameidata *saved; 524fceef393SAl Viro struct inode *link_inode; 5259883d185SAl Viro unsigned root_seq; 5269883d185SAl Viro int dfd; 5271f55a6ecSAl Viro }; 5281f55a6ecSAl Viro 5299883d185SAl Viro static void set_nameidata(struct nameidata *p, int dfd, struct filename *name) 530894bc8c4SAl Viro { 531756daf26SNeilBrown struct nameidata *old = current->nameidata; 532756daf26SNeilBrown p->stack = p->internal; 533c8a53ee5SAl Viro p->dfd = dfd; 534c8a53ee5SAl Viro p->name = name; 535756daf26SNeilBrown p->total_link_count = old ? old->total_link_count : 0; 5369883d185SAl Viro p->saved = old; 537756daf26SNeilBrown current->nameidata = p; 538894bc8c4SAl Viro } 539894bc8c4SAl Viro 5409883d185SAl Viro static void restore_nameidata(void) 541894bc8c4SAl Viro { 5429883d185SAl Viro struct nameidata *now = current->nameidata, *old = now->saved; 543756daf26SNeilBrown 544756daf26SNeilBrown current->nameidata = old; 545756daf26SNeilBrown if (old) 546756daf26SNeilBrown old->total_link_count = now->total_link_count; 547e1a63bbcSAl Viro if (now->stack != now->internal) 548756daf26SNeilBrown kfree(now->stack); 549894bc8c4SAl Viro } 550894bc8c4SAl Viro 551894bc8c4SAl Viro static int __nd_alloc_stack(struct nameidata *nd) 552894bc8c4SAl Viro { 553bc40aee0SAl Viro struct saved *p; 554bc40aee0SAl Viro 555bc40aee0SAl Viro if (nd->flags & LOOKUP_RCU) { 556bc40aee0SAl Viro p= kmalloc(MAXSYMLINKS * sizeof(struct saved), 557bc40aee0SAl Viro GFP_ATOMIC); 558bc40aee0SAl Viro if (unlikely(!p)) 559bc40aee0SAl Viro return -ECHILD; 560bc40aee0SAl Viro } else { 561bc40aee0SAl Viro p= kmalloc(MAXSYMLINKS * sizeof(struct saved), 562894bc8c4SAl Viro GFP_KERNEL); 563894bc8c4SAl Viro if (unlikely(!p)) 564894bc8c4SAl Viro return -ENOMEM; 565bc40aee0SAl Viro } 566894bc8c4SAl Viro memcpy(p, nd->internal, sizeof(nd->internal)); 567894bc8c4SAl Viro nd->stack = p; 568894bc8c4SAl Viro return 0; 569894bc8c4SAl Viro } 570894bc8c4SAl Viro 571397d425dSEric W. Biederman /** 572397d425dSEric W. Biederman * path_connected - Verify that a path->dentry is below path->mnt.mnt_root 573397d425dSEric W. Biederman * @path: nameidate to verify 574397d425dSEric W. Biederman * 575397d425dSEric W. Biederman * Rename can sometimes move a file or directory outside of a bind 576397d425dSEric W. Biederman * mount, path_connected allows those cases to be detected. 577397d425dSEric W. Biederman */ 578397d425dSEric W. Biederman static bool path_connected(const struct path *path) 579397d425dSEric W. Biederman { 580397d425dSEric W. Biederman struct vfsmount *mnt = path->mnt; 581397d425dSEric W. Biederman 582397d425dSEric W. Biederman /* Only bind mounts can have disconnected paths */ 583397d425dSEric W. Biederman if (mnt->mnt_root == mnt->mnt_sb->s_root) 584397d425dSEric W. Biederman return true; 585397d425dSEric W. Biederman 586397d425dSEric W. Biederman return is_subdir(path->dentry, mnt->mnt_root); 587397d425dSEric W. Biederman } 588397d425dSEric W. Biederman 589894bc8c4SAl Viro static inline int nd_alloc_stack(struct nameidata *nd) 590894bc8c4SAl Viro { 591da4e0be0SAl Viro if (likely(nd->depth != EMBEDDED_LEVELS)) 592894bc8c4SAl Viro return 0; 593894bc8c4SAl Viro if (likely(nd->stack != nd->internal)) 594894bc8c4SAl Viro return 0; 595894bc8c4SAl Viro return __nd_alloc_stack(nd); 596894bc8c4SAl Viro } 597894bc8c4SAl Viro 5987973387aSAl Viro static void drop_links(struct nameidata *nd) 5997973387aSAl Viro { 6007973387aSAl Viro int i = nd->depth; 6017973387aSAl Viro while (i--) { 6027973387aSAl Viro struct saved *last = nd->stack + i; 603fceef393SAl Viro do_delayed_call(&last->done); 604fceef393SAl Viro clear_delayed_call(&last->done); 6057973387aSAl Viro } 6067973387aSAl Viro } 6077973387aSAl Viro 6087973387aSAl Viro static void terminate_walk(struct nameidata *nd) 6097973387aSAl Viro { 6107973387aSAl Viro drop_links(nd); 6117973387aSAl Viro if (!(nd->flags & LOOKUP_RCU)) { 6127973387aSAl Viro int i; 6137973387aSAl Viro path_put(&nd->path); 6147973387aSAl Viro for (i = 0; i < nd->depth; i++) 6157973387aSAl Viro path_put(&nd->stack[i].link); 616102b8af2SAl Viro if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 617102b8af2SAl Viro path_put(&nd->root); 618102b8af2SAl Viro nd->root.mnt = NULL; 619102b8af2SAl Viro } 6207973387aSAl Viro } else { 6217973387aSAl Viro nd->flags &= ~LOOKUP_RCU; 6227973387aSAl Viro if (!(nd->flags & LOOKUP_ROOT)) 6237973387aSAl Viro nd->root.mnt = NULL; 6247973387aSAl Viro rcu_read_unlock(); 6257973387aSAl Viro } 6267973387aSAl Viro nd->depth = 0; 6277973387aSAl Viro } 6287973387aSAl Viro 6297973387aSAl Viro /* path_put is needed afterwards regardless of success or failure */ 6307973387aSAl Viro static bool legitimize_path(struct nameidata *nd, 6317973387aSAl Viro struct path *path, unsigned seq) 6327973387aSAl Viro { 6337973387aSAl Viro int res = __legitimize_mnt(path->mnt, nd->m_seq); 6347973387aSAl Viro if (unlikely(res)) { 6357973387aSAl Viro if (res > 0) 6367973387aSAl Viro path->mnt = NULL; 6377973387aSAl Viro path->dentry = NULL; 6387973387aSAl Viro return false; 6397973387aSAl Viro } 6407973387aSAl Viro if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) { 6417973387aSAl Viro path->dentry = NULL; 6427973387aSAl Viro return false; 6437973387aSAl Viro } 6447973387aSAl Viro return !read_seqcount_retry(&path->dentry->d_seq, seq); 6457973387aSAl Viro } 6467973387aSAl Viro 6477973387aSAl Viro static bool legitimize_links(struct nameidata *nd) 6487973387aSAl Viro { 6497973387aSAl Viro int i; 6507973387aSAl Viro for (i = 0; i < nd->depth; i++) { 6517973387aSAl Viro struct saved *last = nd->stack + i; 6527973387aSAl Viro if (unlikely(!legitimize_path(nd, &last->link, last->seq))) { 6537973387aSAl Viro drop_links(nd); 6547973387aSAl Viro nd->depth = i + 1; 6557973387aSAl Viro return false; 6567973387aSAl Viro } 6577973387aSAl Viro } 6587973387aSAl Viro return true; 6597973387aSAl Viro } 6607973387aSAl Viro 66119660af7SAl Viro /* 66231e6b01fSNick Piggin * Path walking has 2 modes, rcu-walk and ref-walk (see 66319660af7SAl Viro * Documentation/filesystems/path-lookup.txt). In situations when we can't 66419660af7SAl Viro * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab 66557e3715cSMike Marshall * normal reference counts on dentries and vfsmounts to transition to ref-walk 66619660af7SAl Viro * mode. Refcounts are grabbed at the last known good point before rcu-walk 66719660af7SAl Viro * got stuck, so ref-walk may continue from there. If this is not successful 66819660af7SAl Viro * (eg. a seqcount has changed), then failure is returned and it's up to caller 66919660af7SAl Viro * to restart the path walk from the beginning in ref-walk mode. 67031e6b01fSNick Piggin */ 67131e6b01fSNick Piggin 67231e6b01fSNick Piggin /** 67319660af7SAl Viro * unlazy_walk - try to switch to ref-walk mode. 67419660af7SAl Viro * @nd: nameidata pathwalk data 67539191628SRandy Dunlap * Returns: 0 on success, -ECHILD on failure 67631e6b01fSNick Piggin * 6774675ac39SAl Viro * unlazy_walk attempts to legitimize the current nd->path and nd->root 6784675ac39SAl Viro * for ref-walk mode. 6794675ac39SAl Viro * Must be called from rcu-walk context. 6807973387aSAl Viro * Nothing should touch nameidata between unlazy_walk() failure and 6817973387aSAl Viro * terminate_walk(). 68231e6b01fSNick Piggin */ 6834675ac39SAl Viro static int unlazy_walk(struct nameidata *nd) 68431e6b01fSNick Piggin { 68531e6b01fSNick Piggin struct dentry *parent = nd->path.dentry; 68631e6b01fSNick Piggin 68731e6b01fSNick Piggin BUG_ON(!(nd->flags & LOOKUP_RCU)); 688e5c832d5SLinus Torvalds 689e5c832d5SLinus Torvalds nd->flags &= ~LOOKUP_RCU; 6907973387aSAl Viro if (unlikely(!legitimize_links(nd))) 6917973387aSAl Viro goto out2; 6924675ac39SAl Viro if (unlikely(!legitimize_path(nd, &nd->path, nd->seq))) 6934675ac39SAl Viro goto out1; 6944675ac39SAl Viro if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 6954675ac39SAl Viro if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) 6964675ac39SAl Viro goto out; 6974675ac39SAl Viro } 6984675ac39SAl Viro rcu_read_unlock(); 6994675ac39SAl Viro BUG_ON(nd->inode != parent->d_inode); 7004675ac39SAl Viro return 0; 7014675ac39SAl Viro 7024675ac39SAl Viro out2: 7034675ac39SAl Viro nd->path.mnt = NULL; 7044675ac39SAl Viro nd->path.dentry = NULL; 7054675ac39SAl Viro out1: 7064675ac39SAl Viro if (!(nd->flags & LOOKUP_ROOT)) 7074675ac39SAl Viro nd->root.mnt = NULL; 7084675ac39SAl Viro out: 7094675ac39SAl Viro rcu_read_unlock(); 7104675ac39SAl Viro return -ECHILD; 7114675ac39SAl Viro } 7124675ac39SAl Viro 7134675ac39SAl Viro /** 7144675ac39SAl Viro * unlazy_child - try to switch to ref-walk mode. 7154675ac39SAl Viro * @nd: nameidata pathwalk data 7164675ac39SAl Viro * @dentry: child of nd->path.dentry 7174675ac39SAl Viro * @seq: seq number to check dentry against 7184675ac39SAl Viro * Returns: 0 on success, -ECHILD on failure 7194675ac39SAl Viro * 7204675ac39SAl Viro * unlazy_child attempts to legitimize the current nd->path, nd->root and dentry 7214675ac39SAl Viro * for ref-walk mode. @dentry must be a path found by a do_lookup call on 7224675ac39SAl Viro * @nd. Must be called from rcu-walk context. 7234675ac39SAl Viro * Nothing should touch nameidata between unlazy_child() failure and 7244675ac39SAl Viro * terminate_walk(). 7254675ac39SAl Viro */ 7264675ac39SAl Viro static int unlazy_child(struct nameidata *nd, struct dentry *dentry, unsigned seq) 7274675ac39SAl Viro { 7284675ac39SAl Viro BUG_ON(!(nd->flags & LOOKUP_RCU)); 7294675ac39SAl Viro 7304675ac39SAl Viro nd->flags &= ~LOOKUP_RCU; 7314675ac39SAl Viro if (unlikely(!legitimize_links(nd))) 7324675ac39SAl Viro goto out2; 7337973387aSAl Viro if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq))) 7347973387aSAl Viro goto out2; 7354675ac39SAl Viro if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref))) 7367973387aSAl Viro goto out1; 73748a066e7SAl Viro 73815570086SLinus Torvalds /* 7394675ac39SAl Viro * We need to move both the parent and the dentry from the RCU domain 7404675ac39SAl Viro * to be properly refcounted. And the sequence number in the dentry 7414675ac39SAl Viro * validates *both* dentry counters, since we checked the sequence 7424675ac39SAl Viro * number of the parent after we got the child sequence number. So we 7434675ac39SAl Viro * know the parent must still be valid if the child sequence number is 74415570086SLinus Torvalds */ 7454675ac39SAl Viro if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) 746e5c832d5SLinus Torvalds goto out; 7474675ac39SAl Viro if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) { 7484675ac39SAl Viro rcu_read_unlock(); 7494675ac39SAl Viro dput(dentry); 7504675ac39SAl Viro goto drop_root_mnt; 75119660af7SAl Viro } 752e5c832d5SLinus Torvalds /* 753e5c832d5SLinus Torvalds * Sequence counts matched. Now make sure that the root is 754e5c832d5SLinus Torvalds * still valid and get it if required. 755e5c832d5SLinus Torvalds */ 756e5c832d5SLinus Torvalds if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 7575a8d87e8SAl Viro if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) { 7585a8d87e8SAl Viro rcu_read_unlock(); 7595a8d87e8SAl Viro dput(dentry); 7605a8d87e8SAl Viro return -ECHILD; 7617973387aSAl Viro } 76231e6b01fSNick Piggin } 76331e6b01fSNick Piggin 7648b61e74fSAl Viro rcu_read_unlock(); 76531e6b01fSNick Piggin return 0; 76619660af7SAl Viro 7677973387aSAl Viro out2: 7687973387aSAl Viro nd->path.mnt = NULL; 7697973387aSAl Viro out1: 7707973387aSAl Viro nd->path.dentry = NULL; 771e5c832d5SLinus Torvalds out: 7728b61e74fSAl Viro rcu_read_unlock(); 773d0d27277SLinus Torvalds drop_root_mnt: 774d0d27277SLinus Torvalds if (!(nd->flags & LOOKUP_ROOT)) 775d0d27277SLinus Torvalds nd->root.mnt = NULL; 77631e6b01fSNick Piggin return -ECHILD; 77731e6b01fSNick Piggin } 77831e6b01fSNick Piggin 7794ce16ef3SAl Viro static inline int d_revalidate(struct dentry *dentry, unsigned int flags) 78034286d66SNick Piggin { 781a89f8337SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) 7824ce16ef3SAl Viro return dentry->d_op->d_revalidate(dentry, flags); 783a89f8337SAl Viro else 784a89f8337SAl Viro return 1; 78534286d66SNick Piggin } 78634286d66SNick Piggin 7879f1fafeeSAl Viro /** 7889f1fafeeSAl Viro * complete_walk - successful completion of path walk 7899f1fafeeSAl Viro * @nd: pointer nameidata 79039159de2SJeff Layton * 7919f1fafeeSAl Viro * If we had been in RCU mode, drop out of it and legitimize nd->path. 7929f1fafeeSAl Viro * Revalidate the final result, unless we'd already done that during 7939f1fafeeSAl Viro * the path walk or the filesystem doesn't ask for it. Return 0 on 7949f1fafeeSAl Viro * success, -error on failure. In case of failure caller does not 7959f1fafeeSAl Viro * need to drop nd->path. 79639159de2SJeff Layton */ 7979f1fafeeSAl Viro static int complete_walk(struct nameidata *nd) 79839159de2SJeff Layton { 79916c2cd71SAl Viro struct dentry *dentry = nd->path.dentry; 80039159de2SJeff Layton int status; 80139159de2SJeff Layton 8029f1fafeeSAl Viro if (nd->flags & LOOKUP_RCU) { 8039f1fafeeSAl Viro if (!(nd->flags & LOOKUP_ROOT)) 8049f1fafeeSAl Viro nd->root.mnt = NULL; 8054675ac39SAl Viro if (unlikely(unlazy_walk(nd))) 80648a066e7SAl Viro return -ECHILD; 80748a066e7SAl Viro } 8089f1fafeeSAl Viro 80916c2cd71SAl Viro if (likely(!(nd->flags & LOOKUP_JUMPED))) 81039159de2SJeff Layton return 0; 81139159de2SJeff Layton 812ecf3d1f1SJeff Layton if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) 81316c2cd71SAl Viro return 0; 81416c2cd71SAl Viro 815ecf3d1f1SJeff Layton status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); 81639159de2SJeff Layton if (status > 0) 81739159de2SJeff Layton return 0; 81839159de2SJeff Layton 81916c2cd71SAl Viro if (!status) 82039159de2SJeff Layton status = -ESTALE; 82116c2cd71SAl Viro 82239159de2SJeff Layton return status; 82339159de2SJeff Layton } 82439159de2SJeff Layton 82518d8c860SAl Viro static void set_root(struct nameidata *nd) 8262a737871SAl Viro { 82731e6b01fSNick Piggin struct fs_struct *fs = current->fs; 8289e6697e2SAl Viro 8299e6697e2SAl Viro if (nd->flags & LOOKUP_RCU) { 8308f47a016SAl Viro unsigned seq; 831c28cc364SNick Piggin 832c28cc364SNick Piggin do { 833c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 83431e6b01fSNick Piggin nd->root = fs->root; 8358f47a016SAl Viro nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq); 836c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 8379e6697e2SAl Viro } else { 8389e6697e2SAl Viro get_fs_root(fs, &nd->root); 8399e6697e2SAl Viro } 84031e6b01fSNick Piggin } 84131e6b01fSNick Piggin 8421d957f9bSJan Blunck static void path_put_conditional(struct path *path, struct nameidata *nd) 843051d3812SIan Kent { 844051d3812SIan Kent dput(path->dentry); 8454ac91378SJan Blunck if (path->mnt != nd->path.mnt) 846051d3812SIan Kent mntput(path->mnt); 847051d3812SIan Kent } 848051d3812SIan Kent 8497b9337aaSNick Piggin static inline void path_to_nameidata(const struct path *path, 8507b9337aaSNick Piggin struct nameidata *nd) 851051d3812SIan Kent { 85231e6b01fSNick Piggin if (!(nd->flags & LOOKUP_RCU)) { 8534ac91378SJan Blunck dput(nd->path.dentry); 85431e6b01fSNick Piggin if (nd->path.mnt != path->mnt) 8554ac91378SJan Blunck mntput(nd->path.mnt); 8569a229683SHuang Shijie } 85731e6b01fSNick Piggin nd->path.mnt = path->mnt; 8584ac91378SJan Blunck nd->path.dentry = path->dentry; 859051d3812SIan Kent } 860051d3812SIan Kent 861248fb5b9SAl Viro static int nd_jump_root(struct nameidata *nd) 862248fb5b9SAl Viro { 863248fb5b9SAl Viro if (nd->flags & LOOKUP_RCU) { 864248fb5b9SAl Viro struct dentry *d; 865248fb5b9SAl Viro nd->path = nd->root; 866248fb5b9SAl Viro d = nd->path.dentry; 867248fb5b9SAl Viro nd->inode = d->d_inode; 868248fb5b9SAl Viro nd->seq = nd->root_seq; 869248fb5b9SAl Viro if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq))) 870248fb5b9SAl Viro return -ECHILD; 871248fb5b9SAl Viro } else { 872248fb5b9SAl Viro path_put(&nd->path); 873248fb5b9SAl Viro nd->path = nd->root; 874248fb5b9SAl Viro path_get(&nd->path); 875248fb5b9SAl Viro nd->inode = nd->path.dentry->d_inode; 876248fb5b9SAl Viro } 877248fb5b9SAl Viro nd->flags |= LOOKUP_JUMPED; 878248fb5b9SAl Viro return 0; 879248fb5b9SAl Viro } 880248fb5b9SAl Viro 881b5fb63c1SChristoph Hellwig /* 8826b255391SAl Viro * Helper to directly jump to a known parsed path from ->get_link, 883b5fb63c1SChristoph Hellwig * caller must have taken a reference to path beforehand. 884b5fb63c1SChristoph Hellwig */ 8856e77137bSAl Viro void nd_jump_link(struct path *path) 886b5fb63c1SChristoph Hellwig { 8876e77137bSAl Viro struct nameidata *nd = current->nameidata; 888b5fb63c1SChristoph Hellwig path_put(&nd->path); 889b5fb63c1SChristoph Hellwig 890b5fb63c1SChristoph Hellwig nd->path = *path; 891b5fb63c1SChristoph Hellwig nd->inode = nd->path.dentry->d_inode; 892b5fb63c1SChristoph Hellwig nd->flags |= LOOKUP_JUMPED; 893b5fb63c1SChristoph Hellwig } 894b5fb63c1SChristoph Hellwig 895b9ff4429SAl Viro static inline void put_link(struct nameidata *nd) 896574197e0SAl Viro { 89721c3003dSAl Viro struct saved *last = nd->stack + --nd->depth; 898fceef393SAl Viro do_delayed_call(&last->done); 8996548fae2SAl Viro if (!(nd->flags & LOOKUP_RCU)) 900b9ff4429SAl Viro path_put(&last->link); 901574197e0SAl Viro } 902574197e0SAl Viro 903561ec64aSLinus Torvalds int sysctl_protected_symlinks __read_mostly = 0; 904561ec64aSLinus Torvalds int sysctl_protected_hardlinks __read_mostly = 0; 905800179c9SKees Cook 906800179c9SKees Cook /** 907800179c9SKees Cook * may_follow_link - Check symlink following for unsafe situations 90855852635SRandy Dunlap * @nd: nameidata pathwalk data 909800179c9SKees Cook * 910800179c9SKees Cook * In the case of the sysctl_protected_symlinks sysctl being enabled, 911800179c9SKees Cook * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is 912800179c9SKees Cook * in a sticky world-writable directory. This is to protect privileged 913800179c9SKees Cook * processes from failing races against path names that may change out 914800179c9SKees Cook * from under them by way of other users creating malicious symlinks. 915800179c9SKees Cook * It will permit symlinks to be followed only when outside a sticky 916800179c9SKees Cook * world-writable directory, or when the uid of the symlink and follower 917800179c9SKees Cook * match, or when the directory owner matches the symlink's owner. 918800179c9SKees Cook * 919800179c9SKees Cook * Returns 0 if following the symlink is allowed, -ve on error. 920800179c9SKees Cook */ 921fec2fa24SAl Viro static inline int may_follow_link(struct nameidata *nd) 922800179c9SKees Cook { 923800179c9SKees Cook const struct inode *inode; 924800179c9SKees Cook const struct inode *parent; 9252d7f9e2aSSeth Forshee kuid_t puid; 926800179c9SKees Cook 927800179c9SKees Cook if (!sysctl_protected_symlinks) 928800179c9SKees Cook return 0; 929800179c9SKees Cook 930800179c9SKees Cook /* Allowed if owner and follower match. */ 931fceef393SAl Viro inode = nd->link_inode; 93281abe27bSEric W. Biederman if (uid_eq(current_cred()->fsuid, inode->i_uid)) 933800179c9SKees Cook return 0; 934800179c9SKees Cook 935800179c9SKees Cook /* Allowed if parent directory not sticky and world-writable. */ 936aa65fa35SAl Viro parent = nd->inode; 937800179c9SKees Cook if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) 938800179c9SKees Cook return 0; 939800179c9SKees Cook 940800179c9SKees Cook /* Allowed if parent directory and link owner match. */ 9412d7f9e2aSSeth Forshee puid = parent->i_uid; 9422d7f9e2aSSeth Forshee if (uid_valid(puid) && uid_eq(puid, inode->i_uid)) 943800179c9SKees Cook return 0; 944800179c9SKees Cook 94531956502SAl Viro if (nd->flags & LOOKUP_RCU) 94631956502SAl Viro return -ECHILD; 94731956502SAl Viro 9481cf2665bSAl Viro audit_log_link_denied("follow_link", &nd->stack[0].link); 949800179c9SKees Cook return -EACCES; 950800179c9SKees Cook } 951800179c9SKees Cook 952800179c9SKees Cook /** 953800179c9SKees Cook * safe_hardlink_source - Check for safe hardlink conditions 954800179c9SKees Cook * @inode: the source inode to hardlink from 955800179c9SKees Cook * 956800179c9SKees Cook * Return false if at least one of the following conditions: 957800179c9SKees Cook * - inode is not a regular file 958800179c9SKees Cook * - inode is setuid 959800179c9SKees Cook * - inode is setgid and group-exec 960800179c9SKees Cook * - access failure for read and write 961800179c9SKees Cook * 962800179c9SKees Cook * Otherwise returns true. 963800179c9SKees Cook */ 964800179c9SKees Cook static bool safe_hardlink_source(struct inode *inode) 965800179c9SKees Cook { 966800179c9SKees Cook umode_t mode = inode->i_mode; 967800179c9SKees Cook 968800179c9SKees Cook /* Special files should not get pinned to the filesystem. */ 969800179c9SKees Cook if (!S_ISREG(mode)) 970800179c9SKees Cook return false; 971800179c9SKees Cook 972800179c9SKees Cook /* Setuid files should not get pinned to the filesystem. */ 973800179c9SKees Cook if (mode & S_ISUID) 974800179c9SKees Cook return false; 975800179c9SKees Cook 976800179c9SKees Cook /* Executable setgid files should not get pinned to the filesystem. */ 977800179c9SKees Cook if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) 978800179c9SKees Cook return false; 979800179c9SKees Cook 980800179c9SKees Cook /* Hardlinking to unreadable or unwritable sources is dangerous. */ 981800179c9SKees Cook if (inode_permission(inode, MAY_READ | MAY_WRITE)) 982800179c9SKees Cook return false; 983800179c9SKees Cook 984800179c9SKees Cook return true; 985800179c9SKees Cook } 986800179c9SKees Cook 987800179c9SKees Cook /** 988800179c9SKees Cook * may_linkat - Check permissions for creating a hardlink 989800179c9SKees Cook * @link: the source to hardlink from 990800179c9SKees Cook * 991800179c9SKees Cook * Block hardlink when all of: 992800179c9SKees Cook * - sysctl_protected_hardlinks enabled 993800179c9SKees Cook * - fsuid does not match inode 994800179c9SKees Cook * - hardlink source is unsafe (see safe_hardlink_source() above) 995f2ca3796SDirk Steinmetz * - not CAP_FOWNER in a namespace with the inode owner uid mapped 996800179c9SKees Cook * 997800179c9SKees Cook * Returns 0 if successful, -ve on error. 998800179c9SKees Cook */ 999800179c9SKees Cook static int may_linkat(struct path *link) 1000800179c9SKees Cook { 1001800179c9SKees Cook struct inode *inode; 1002800179c9SKees Cook 1003800179c9SKees Cook if (!sysctl_protected_hardlinks) 1004800179c9SKees Cook return 0; 1005800179c9SKees Cook 1006800179c9SKees Cook inode = link->dentry->d_inode; 1007800179c9SKees Cook 1008800179c9SKees Cook /* Source inode owner (or CAP_FOWNER) can hardlink all they like, 1009800179c9SKees Cook * otherwise, it must be a safe source. 1010800179c9SKees Cook */ 1011f2ca3796SDirk Steinmetz if (inode_owner_or_capable(inode) || safe_hardlink_source(inode)) 1012800179c9SKees Cook return 0; 1013800179c9SKees Cook 1014a51d9eaaSKees Cook audit_log_link_denied("linkat", link); 1015800179c9SKees Cook return -EPERM; 1016800179c9SKees Cook } 1017800179c9SKees Cook 10183b2e7f75SAl Viro static __always_inline 10193b2e7f75SAl Viro const char *get_link(struct nameidata *nd) 10201da177e4SLinus Torvalds { 1021ab104923SAl Viro struct saved *last = nd->stack + nd->depth - 1; 10221cf2665bSAl Viro struct dentry *dentry = last->link.dentry; 1023fceef393SAl Viro struct inode *inode = nd->link_inode; 10246d7b5aaeSAl Viro int error; 10250a959df5SAl Viro const char *res; 10261da177e4SLinus Torvalds 10278fa9dd24SNeilBrown if (!(nd->flags & LOOKUP_RCU)) { 10288fa9dd24SNeilBrown touch_atime(&last->link); 10298fa9dd24SNeilBrown cond_resched(); 1030598e3c8fSMiklos Szeredi } else if (atime_needs_update_rcu(&last->link, inode)) { 10314675ac39SAl Viro if (unlikely(unlazy_walk(nd))) 10328fa9dd24SNeilBrown return ERR_PTR(-ECHILD); 10338fa9dd24SNeilBrown touch_atime(&last->link); 10348fa9dd24SNeilBrown } 10358fa9dd24SNeilBrown 1036bda0be7aSNeilBrown error = security_inode_follow_link(dentry, inode, 1037bda0be7aSNeilBrown nd->flags & LOOKUP_RCU); 1038bda0be7aSNeilBrown if (unlikely(error)) 10396920a440SAl Viro return ERR_PTR(error); 104036f3b4f6SAl Viro 104186acdca1SAl Viro nd->last_type = LAST_BIND; 1042d4dee48bSAl Viro res = inode->i_link; 1043d4dee48bSAl Viro if (!res) { 1044fceef393SAl Viro const char * (*get)(struct dentry *, struct inode *, 1045fceef393SAl Viro struct delayed_call *); 1046fceef393SAl Viro get = inode->i_op->get_link; 10478c1b4566SAl Viro if (nd->flags & LOOKUP_RCU) { 1048fceef393SAl Viro res = get(NULL, inode, &last->done); 10496b255391SAl Viro if (res == ERR_PTR(-ECHILD)) { 10504675ac39SAl Viro if (unlikely(unlazy_walk(nd))) 10518c1b4566SAl Viro return ERR_PTR(-ECHILD); 1052fceef393SAl Viro res = get(dentry, inode, &last->done); 10538c1b4566SAl Viro } 10546b255391SAl Viro } else { 1055fceef393SAl Viro res = get(dentry, inode, &last->done); 10566b255391SAl Viro } 1057fceef393SAl Viro if (IS_ERR_OR_NULL(res)) 1058fab51e8aSAl Viro return res; 10590a959df5SAl Viro } 1060fab51e8aSAl Viro if (*res == '/') { 1061fab51e8aSAl Viro if (!nd->root.mnt) 1062fab51e8aSAl Viro set_root(nd); 1063248fb5b9SAl Viro if (unlikely(nd_jump_root(nd))) 1064fab51e8aSAl Viro return ERR_PTR(-ECHILD); 1065fab51e8aSAl Viro while (unlikely(*++res == '/')) 1066fab51e8aSAl Viro ; 1067fab51e8aSAl Viro } 1068fab51e8aSAl Viro if (!*res) 1069fab51e8aSAl Viro res = NULL; 10700a959df5SAl Viro return res; 10710a959df5SAl Viro } 10726d7b5aaeSAl Viro 1073f015f126SDavid Howells /* 1074f015f126SDavid Howells * follow_up - Find the mountpoint of path's vfsmount 1075f015f126SDavid Howells * 1076f015f126SDavid Howells * Given a path, find the mountpoint of its source file system. 1077f015f126SDavid Howells * Replace @path with the path of the mountpoint in the parent mount. 1078f015f126SDavid Howells * Up is towards /. 1079f015f126SDavid Howells * 1080f015f126SDavid Howells * Return 1 if we went up a level and 0 if we were already at the 1081f015f126SDavid Howells * root. 1082f015f126SDavid Howells */ 1083bab77ebfSAl Viro int follow_up(struct path *path) 10841da177e4SLinus Torvalds { 10850714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 10860714a533SAl Viro struct mount *parent; 10871da177e4SLinus Torvalds struct dentry *mountpoint; 108899b7db7bSNick Piggin 108948a066e7SAl Viro read_seqlock_excl(&mount_lock); 10900714a533SAl Viro parent = mnt->mnt_parent; 10913c0a6163SAl Viro if (parent == mnt) { 109248a066e7SAl Viro read_sequnlock_excl(&mount_lock); 10931da177e4SLinus Torvalds return 0; 10941da177e4SLinus Torvalds } 10950714a533SAl Viro mntget(&parent->mnt); 1096a73324daSAl Viro mountpoint = dget(mnt->mnt_mountpoint); 109748a066e7SAl Viro read_sequnlock_excl(&mount_lock); 1098bab77ebfSAl Viro dput(path->dentry); 1099bab77ebfSAl Viro path->dentry = mountpoint; 1100bab77ebfSAl Viro mntput(path->mnt); 11010714a533SAl Viro path->mnt = &parent->mnt; 11021da177e4SLinus Torvalds return 1; 11031da177e4SLinus Torvalds } 11044d359507SAl Viro EXPORT_SYMBOL(follow_up); 11051da177e4SLinus Torvalds 1106b5c84bf6SNick Piggin /* 11079875cf80SDavid Howells * Perform an automount 11089875cf80SDavid Howells * - return -EISDIR to tell follow_managed() to stop and return the path we 11099875cf80SDavid Howells * were called with. 11101da177e4SLinus Torvalds */ 1111756daf26SNeilBrown static int follow_automount(struct path *path, struct nameidata *nd, 11129875cf80SDavid Howells bool *need_mntput) 111331e6b01fSNick Piggin { 11149875cf80SDavid Howells struct vfsmount *mnt; 1115ea5b778aSDavid Howells int err; 11169875cf80SDavid Howells 11179875cf80SDavid Howells if (!path->dentry->d_op || !path->dentry->d_op->d_automount) 11189875cf80SDavid Howells return -EREMOTE; 11199875cf80SDavid Howells 11200ec26fd0SMiklos Szeredi /* We don't want to mount if someone's just doing a stat - 11210ec26fd0SMiklos Szeredi * unless they're stat'ing a directory and appended a '/' to 11220ec26fd0SMiklos Szeredi * the name. 11230ec26fd0SMiklos Szeredi * 11240ec26fd0SMiklos Szeredi * We do, however, want to mount if someone wants to open or 11255a30d8a2SDavid Howells * create a file of any type under the mountpoint, wants to 11260ec26fd0SMiklos Szeredi * traverse through the mountpoint or wants to open the 11270ec26fd0SMiklos Szeredi * mounted directory. Also, autofs may mark negative dentries 11280ec26fd0SMiklos Szeredi * as being automount points. These will need the attentions 11290ec26fd0SMiklos Szeredi * of the daemon to instantiate them before they can be used. 11305a30d8a2SDavid Howells */ 1131756daf26SNeilBrown if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | 1132d94c177bSLinus Torvalds LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && 11335a30d8a2SDavid Howells path->dentry->d_inode) 11349875cf80SDavid Howells return -EISDIR; 11350ec26fd0SMiklos Szeredi 1136aeaa4a79SEric W. Biederman if (path->dentry->d_sb->s_user_ns != &init_user_ns) 1137aeaa4a79SEric W. Biederman return -EACCES; 1138aeaa4a79SEric W. Biederman 1139756daf26SNeilBrown nd->total_link_count++; 1140756daf26SNeilBrown if (nd->total_link_count >= 40) 11419875cf80SDavid Howells return -ELOOP; 11429875cf80SDavid Howells 11439875cf80SDavid Howells mnt = path->dentry->d_op->d_automount(path); 11449875cf80SDavid Howells if (IS_ERR(mnt)) { 11459875cf80SDavid Howells /* 11469875cf80SDavid Howells * The filesystem is allowed to return -EISDIR here to indicate 11479875cf80SDavid Howells * it doesn't want to automount. For instance, autofs would do 11489875cf80SDavid Howells * this so that its userspace daemon can mount on this dentry. 11499875cf80SDavid Howells * 11509875cf80SDavid Howells * However, we can only permit this if it's a terminal point in 11519875cf80SDavid Howells * the path being looked up; if it wasn't then the remainder of 11529875cf80SDavid Howells * the path is inaccessible and we should say so. 11539875cf80SDavid Howells */ 1154756daf26SNeilBrown if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT)) 11559875cf80SDavid Howells return -EREMOTE; 11569875cf80SDavid Howells return PTR_ERR(mnt); 115731e6b01fSNick Piggin } 1158ea5b778aSDavid Howells 11599875cf80SDavid Howells if (!mnt) /* mount collision */ 11609875cf80SDavid Howells return 0; 11619875cf80SDavid Howells 11628aef1884SAl Viro if (!*need_mntput) { 11638aef1884SAl Viro /* lock_mount() may release path->mnt on error */ 11648aef1884SAl Viro mntget(path->mnt); 11658aef1884SAl Viro *need_mntput = true; 11668aef1884SAl Viro } 116719a167afSAl Viro err = finish_automount(mnt, path); 1168ea5b778aSDavid Howells 1169ea5b778aSDavid Howells switch (err) { 1170ea5b778aSDavid Howells case -EBUSY: 1171ea5b778aSDavid Howells /* Someone else made a mount here whilst we were busy */ 117219a167afSAl Viro return 0; 1173ea5b778aSDavid Howells case 0: 11748aef1884SAl Viro path_put(path); 11759875cf80SDavid Howells path->mnt = mnt; 11769875cf80SDavid Howells path->dentry = dget(mnt->mnt_root); 11779875cf80SDavid Howells return 0; 117819a167afSAl Viro default: 117919a167afSAl Viro return err; 11809875cf80SDavid Howells } 118119a167afSAl Viro 1182ea5b778aSDavid Howells } 11839875cf80SDavid Howells 11849875cf80SDavid Howells /* 11859875cf80SDavid Howells * Handle a dentry that is managed in some way. 1186cc53ce53SDavid Howells * - Flagged for transit management (autofs) 11879875cf80SDavid Howells * - Flagged as mountpoint 11889875cf80SDavid Howells * - Flagged as automount point 11899875cf80SDavid Howells * 11909875cf80SDavid Howells * This may only be called in refwalk mode. 11919875cf80SDavid Howells * 11929875cf80SDavid Howells * Serialization is taken care of in namespace.c 11939875cf80SDavid Howells */ 1194756daf26SNeilBrown static int follow_managed(struct path *path, struct nameidata *nd) 11959875cf80SDavid Howells { 11968aef1884SAl Viro struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */ 11979875cf80SDavid Howells unsigned managed; 11989875cf80SDavid Howells bool need_mntput = false; 11998aef1884SAl Viro int ret = 0; 12009875cf80SDavid Howells 12019875cf80SDavid Howells /* Given that we're not holding a lock here, we retain the value in a 12029875cf80SDavid Howells * local variable for each dentry as we look at it so that we don't see 12039875cf80SDavid Howells * the components of that value change under us */ 12049875cf80SDavid Howells while (managed = ACCESS_ONCE(path->dentry->d_flags), 12059875cf80SDavid Howells managed &= DCACHE_MANAGED_DENTRY, 12069875cf80SDavid Howells unlikely(managed != 0)) { 1207cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1208cc53ce53SDavid Howells * being held. */ 1209cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 1210cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1211cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 1212fb5f51c7SIan Kent ret = path->dentry->d_op->d_manage(path, false); 1213cc53ce53SDavid Howells if (ret < 0) 12148aef1884SAl Viro break; 1215cc53ce53SDavid Howells } 1216cc53ce53SDavid Howells 12179875cf80SDavid Howells /* Transit to a mounted filesystem. */ 12189875cf80SDavid Howells if (managed & DCACHE_MOUNTED) { 12199875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 12209875cf80SDavid Howells if (mounted) { 12219875cf80SDavid Howells dput(path->dentry); 12229875cf80SDavid Howells if (need_mntput) 1223463ffb2eSAl Viro mntput(path->mnt); 1224463ffb2eSAl Viro path->mnt = mounted; 1225463ffb2eSAl Viro path->dentry = dget(mounted->mnt_root); 12269875cf80SDavid Howells need_mntput = true; 12279875cf80SDavid Howells continue; 1228463ffb2eSAl Viro } 1229463ffb2eSAl Viro 12309875cf80SDavid Howells /* Something is mounted on this dentry in another 12319875cf80SDavid Howells * namespace and/or whatever was mounted there in this 123248a066e7SAl Viro * namespace got unmounted before lookup_mnt() could 123348a066e7SAl Viro * get it */ 12341da177e4SLinus Torvalds } 12359875cf80SDavid Howells 12369875cf80SDavid Howells /* Handle an automount point */ 12379875cf80SDavid Howells if (managed & DCACHE_NEED_AUTOMOUNT) { 1238756daf26SNeilBrown ret = follow_automount(path, nd, &need_mntput); 12399875cf80SDavid Howells if (ret < 0) 12408aef1884SAl Viro break; 12419875cf80SDavid Howells continue; 12429875cf80SDavid Howells } 12439875cf80SDavid Howells 12449875cf80SDavid Howells /* We didn't change the current path point */ 12459875cf80SDavid Howells break; 12469875cf80SDavid Howells } 12478aef1884SAl Viro 12488aef1884SAl Viro if (need_mntput && path->mnt == mnt) 12498aef1884SAl Viro mntput(path->mnt); 1250e9742b53SAl Viro if (ret == -EISDIR || !ret) 1251e9742b53SAl Viro ret = 1; 12528402752eSAl Viro if (need_mntput) 12538402752eSAl Viro nd->flags |= LOOKUP_JUMPED; 12548402752eSAl Viro if (unlikely(ret < 0)) 12558402752eSAl Viro path_put_conditional(path, nd); 12568402752eSAl Viro return ret; 12571da177e4SLinus Torvalds } 12581da177e4SLinus Torvalds 1259cc53ce53SDavid Howells int follow_down_one(struct path *path) 12601da177e4SLinus Torvalds { 12611da177e4SLinus Torvalds struct vfsmount *mounted; 12621da177e4SLinus Torvalds 12631c755af4SAl Viro mounted = lookup_mnt(path); 12641da177e4SLinus Torvalds if (mounted) { 12659393bd07SAl Viro dput(path->dentry); 12669393bd07SAl Viro mntput(path->mnt); 12679393bd07SAl Viro path->mnt = mounted; 12689393bd07SAl Viro path->dentry = dget(mounted->mnt_root); 12691da177e4SLinus Torvalds return 1; 12701da177e4SLinus Torvalds } 12711da177e4SLinus Torvalds return 0; 12721da177e4SLinus Torvalds } 12734d359507SAl Viro EXPORT_SYMBOL(follow_down_one); 12741da177e4SLinus Torvalds 1275fb5f51c7SIan Kent static inline int managed_dentry_rcu(const struct path *path) 127662a7375eSIan Kent { 1277fb5f51c7SIan Kent return (path->dentry->d_flags & DCACHE_MANAGE_TRANSIT) ? 1278fb5f51c7SIan Kent path->dentry->d_op->d_manage(path, true) : 0; 127962a7375eSIan Kent } 128062a7375eSIan Kent 12819875cf80SDavid Howells /* 1282287548e4SAl Viro * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if 1283287548e4SAl Viro * we meet a managed dentry that would need blocking. 12849875cf80SDavid Howells */ 12859875cf80SDavid Howells static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, 1286254cf582SAl Viro struct inode **inode, unsigned *seqp) 12879875cf80SDavid Howells { 128862a7375eSIan Kent for (;;) { 1289c7105365SAl Viro struct mount *mounted; 129062a7375eSIan Kent /* 129162a7375eSIan Kent * Don't forget we might have a non-mountpoint managed dentry 129262a7375eSIan Kent * that wants to block transit. 129362a7375eSIan Kent */ 1294fb5f51c7SIan Kent switch (managed_dentry_rcu(path)) { 1295b8faf035SNeilBrown case -ECHILD: 1296b8faf035SNeilBrown default: 1297ab90911fSDavid Howells return false; 1298b8faf035SNeilBrown case -EISDIR: 1299b8faf035SNeilBrown return true; 1300b8faf035SNeilBrown case 0: 1301b8faf035SNeilBrown break; 1302b8faf035SNeilBrown } 130362a7375eSIan Kent 130462a7375eSIan Kent if (!d_mountpoint(path->dentry)) 1305b8faf035SNeilBrown return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); 130662a7375eSIan Kent 1307474279dcSAl Viro mounted = __lookup_mnt(path->mnt, path->dentry); 13089875cf80SDavid Howells if (!mounted) 13099875cf80SDavid Howells break; 1310c7105365SAl Viro path->mnt = &mounted->mnt; 1311c7105365SAl Viro path->dentry = mounted->mnt.mnt_root; 1312a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 1313254cf582SAl Viro *seqp = read_seqcount_begin(&path->dentry->d_seq); 131459430262SLinus Torvalds /* 131559430262SLinus Torvalds * Update the inode too. We don't need to re-check the 131659430262SLinus Torvalds * dentry sequence number here after this d_inode read, 131759430262SLinus Torvalds * because a mount-point is always pinned. 131859430262SLinus Torvalds */ 131959430262SLinus Torvalds *inode = path->dentry->d_inode; 13209875cf80SDavid Howells } 1321f5be3e29SAl Viro return !read_seqretry(&mount_lock, nd->m_seq) && 1322b8faf035SNeilBrown !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); 1323287548e4SAl Viro } 1324287548e4SAl Viro 132531e6b01fSNick Piggin static int follow_dotdot_rcu(struct nameidata *nd) 132631e6b01fSNick Piggin { 13274023bfc9SAl Viro struct inode *inode = nd->inode; 132831e6b01fSNick Piggin 132931e6b01fSNick Piggin while (1) { 1330aed434adSAl Viro if (path_equal(&nd->path, &nd->root)) 133131e6b01fSNick Piggin break; 133231e6b01fSNick Piggin if (nd->path.dentry != nd->path.mnt->mnt_root) { 133331e6b01fSNick Piggin struct dentry *old = nd->path.dentry; 133431e6b01fSNick Piggin struct dentry *parent = old->d_parent; 133531e6b01fSNick Piggin unsigned seq; 133631e6b01fSNick Piggin 13374023bfc9SAl Viro inode = parent->d_inode; 133831e6b01fSNick Piggin seq = read_seqcount_begin(&parent->d_seq); 1339aed434adSAl Viro if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq))) 1340aed434adSAl Viro return -ECHILD; 134131e6b01fSNick Piggin nd->path.dentry = parent; 134231e6b01fSNick Piggin nd->seq = seq; 1343397d425dSEric W. Biederman if (unlikely(!path_connected(&nd->path))) 1344397d425dSEric W. Biederman return -ENOENT; 134531e6b01fSNick Piggin break; 1346aed434adSAl Viro } else { 1347aed434adSAl Viro struct mount *mnt = real_mount(nd->path.mnt); 1348aed434adSAl Viro struct mount *mparent = mnt->mnt_parent; 1349aed434adSAl Viro struct dentry *mountpoint = mnt->mnt_mountpoint; 1350aed434adSAl Viro struct inode *inode2 = mountpoint->d_inode; 1351aed434adSAl Viro unsigned seq = read_seqcount_begin(&mountpoint->d_seq); 1352aed434adSAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1353aed434adSAl Viro return -ECHILD; 1354aed434adSAl Viro if (&mparent->mnt == nd->path.mnt) 135531e6b01fSNick Piggin break; 1356aed434adSAl Viro /* we know that mountpoint was pinned */ 1357aed434adSAl Viro nd->path.dentry = mountpoint; 1358aed434adSAl Viro nd->path.mnt = &mparent->mnt; 1359aed434adSAl Viro inode = inode2; 1360aed434adSAl Viro nd->seq = seq; 136131e6b01fSNick Piggin } 1362aed434adSAl Viro } 1363aed434adSAl Viro while (unlikely(d_mountpoint(nd->path.dentry))) { 1364b37199e6SAl Viro struct mount *mounted; 1365b37199e6SAl Viro mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry); 1366aed434adSAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1367aed434adSAl Viro return -ECHILD; 1368b37199e6SAl Viro if (!mounted) 1369b37199e6SAl Viro break; 1370b37199e6SAl Viro nd->path.mnt = &mounted->mnt; 1371b37199e6SAl Viro nd->path.dentry = mounted->mnt.mnt_root; 13724023bfc9SAl Viro inode = nd->path.dentry->d_inode; 1373b37199e6SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 1374b37199e6SAl Viro } 13754023bfc9SAl Viro nd->inode = inode; 137631e6b01fSNick Piggin return 0; 137731e6b01fSNick Piggin } 137831e6b01fSNick Piggin 13799875cf80SDavid Howells /* 1380cc53ce53SDavid Howells * Follow down to the covering mount currently visible to userspace. At each 1381cc53ce53SDavid Howells * point, the filesystem owning that dentry may be queried as to whether the 1382cc53ce53SDavid Howells * caller is permitted to proceed or not. 1383cc53ce53SDavid Howells */ 13847cc90cc3SAl Viro int follow_down(struct path *path) 1385cc53ce53SDavid Howells { 1386cc53ce53SDavid Howells unsigned managed; 1387cc53ce53SDavid Howells int ret; 1388cc53ce53SDavid Howells 1389cc53ce53SDavid Howells while (managed = ACCESS_ONCE(path->dentry->d_flags), 1390cc53ce53SDavid Howells unlikely(managed & DCACHE_MANAGED_DENTRY)) { 1391cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1392cc53ce53SDavid Howells * being held. 1393cc53ce53SDavid Howells * 1394cc53ce53SDavid Howells * We indicate to the filesystem if someone is trying to mount 1395cc53ce53SDavid Howells * something here. This gives autofs the chance to deny anyone 1396cc53ce53SDavid Howells * other than its daemon the right to mount on its 1397cc53ce53SDavid Howells * superstructure. 1398cc53ce53SDavid Howells * 1399cc53ce53SDavid Howells * The filesystem may sleep at this point. 1400cc53ce53SDavid Howells */ 1401cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 1402cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1403cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 1404fb5f51c7SIan Kent ret = path->dentry->d_op->d_manage(path, false); 1405cc53ce53SDavid Howells if (ret < 0) 1406cc53ce53SDavid Howells return ret == -EISDIR ? 0 : ret; 1407cc53ce53SDavid Howells } 1408cc53ce53SDavid Howells 1409cc53ce53SDavid Howells /* Transit to a mounted filesystem. */ 1410cc53ce53SDavid Howells if (managed & DCACHE_MOUNTED) { 1411cc53ce53SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 1412cc53ce53SDavid Howells if (!mounted) 1413cc53ce53SDavid Howells break; 1414cc53ce53SDavid Howells dput(path->dentry); 1415cc53ce53SDavid Howells mntput(path->mnt); 1416cc53ce53SDavid Howells path->mnt = mounted; 1417cc53ce53SDavid Howells path->dentry = dget(mounted->mnt_root); 1418cc53ce53SDavid Howells continue; 1419cc53ce53SDavid Howells } 1420cc53ce53SDavid Howells 1421cc53ce53SDavid Howells /* Don't handle automount points here */ 1422cc53ce53SDavid Howells break; 1423cc53ce53SDavid Howells } 1424cc53ce53SDavid Howells return 0; 1425cc53ce53SDavid Howells } 14264d359507SAl Viro EXPORT_SYMBOL(follow_down); 1427cc53ce53SDavid Howells 1428cc53ce53SDavid Howells /* 14299875cf80SDavid Howells * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() 14309875cf80SDavid Howells */ 14319875cf80SDavid Howells static void follow_mount(struct path *path) 14329875cf80SDavid Howells { 14339875cf80SDavid Howells while (d_mountpoint(path->dentry)) { 14349875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 14359875cf80SDavid Howells if (!mounted) 14369875cf80SDavid Howells break; 14379875cf80SDavid Howells dput(path->dentry); 14389875cf80SDavid Howells mntput(path->mnt); 14399875cf80SDavid Howells path->mnt = mounted; 14409875cf80SDavid Howells path->dentry = dget(mounted->mnt_root); 14419875cf80SDavid Howells } 14429875cf80SDavid Howells } 14439875cf80SDavid Howells 1444eedf265aSEric W. Biederman static int path_parent_directory(struct path *path) 1445eedf265aSEric W. Biederman { 1446eedf265aSEric W. Biederman struct dentry *old = path->dentry; 1447eedf265aSEric W. Biederman /* rare case of legitimate dget_parent()... */ 1448eedf265aSEric W. Biederman path->dentry = dget_parent(path->dentry); 1449eedf265aSEric W. Biederman dput(old); 1450eedf265aSEric W. Biederman if (unlikely(!path_connected(path))) 1451eedf265aSEric W. Biederman return -ENOENT; 1452eedf265aSEric W. Biederman return 0; 1453eedf265aSEric W. Biederman } 1454eedf265aSEric W. Biederman 1455397d425dSEric W. Biederman static int follow_dotdot(struct nameidata *nd) 14561da177e4SLinus Torvalds { 14571da177e4SLinus Torvalds while(1) { 14582a737871SAl Viro if (nd->path.dentry == nd->root.dentry && 14592a737871SAl Viro nd->path.mnt == nd->root.mnt) { 14601da177e4SLinus Torvalds break; 14611da177e4SLinus Torvalds } 14624ac91378SJan Blunck if (nd->path.dentry != nd->path.mnt->mnt_root) { 1463eedf265aSEric W. Biederman int ret = path_parent_directory(&nd->path); 1464eedf265aSEric W. Biederman if (ret) 1465eedf265aSEric W. Biederman return ret; 14661da177e4SLinus Torvalds break; 14671da177e4SLinus Torvalds } 14683088dd70SAl Viro if (!follow_up(&nd->path)) 14691da177e4SLinus Torvalds break; 14701da177e4SLinus Torvalds } 147179ed0226SAl Viro follow_mount(&nd->path); 147231e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 1473397d425dSEric W. Biederman return 0; 14741da177e4SLinus Torvalds } 14751da177e4SLinus Torvalds 14761da177e4SLinus Torvalds /* 1477f4fdace9SOleg Drokin * This looks up the name in dcache and possibly revalidates the found dentry. 1478f4fdace9SOleg Drokin * NULL is returned if the dentry does not exist in the cache. 1479baa03890SNick Piggin */ 1480e3c13928SAl Viro static struct dentry *lookup_dcache(const struct qstr *name, 1481e3c13928SAl Viro struct dentry *dir, 14826c51e513SAl Viro unsigned int flags) 1483baa03890SNick Piggin { 1484a89f8337SAl Viro struct dentry *dentry = d_lookup(dir, name); 1485bad61189SMiklos Szeredi if (dentry) { 1486a89f8337SAl Viro int error = d_revalidate(dentry, flags); 1487bad61189SMiklos Szeredi if (unlikely(error <= 0)) { 148874ff0ffcSAl Viro if (!error) 14895542aa2fSEric W. Biederman d_invalidate(dentry); 1490bad61189SMiklos Szeredi dput(dentry); 149174ff0ffcSAl Viro return ERR_PTR(error); 1492bad61189SMiklos Szeredi } 1493bad61189SMiklos Szeredi } 1494baa03890SNick Piggin return dentry; 1495baa03890SNick Piggin } 1496baa03890SNick Piggin 1497baa03890SNick Piggin /* 149813a2c3beSJ. Bruce Fields * Call i_op->lookup on the dentry. The dentry must be negative and 149913a2c3beSJ. Bruce Fields * unhashed. 1500bad61189SMiklos Szeredi * 1501bad61189SMiklos Szeredi * dir->d_inode->i_mutex must be held 150244396f4bSJosef Bacik */ 1503bad61189SMiklos Szeredi static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry, 150472bd866aSAl Viro unsigned int flags) 150544396f4bSJosef Bacik { 150644396f4bSJosef Bacik struct dentry *old; 150744396f4bSJosef Bacik 150844396f4bSJosef Bacik /* Don't create child dentry for a dead directory. */ 1509bad61189SMiklos Szeredi if (unlikely(IS_DEADDIR(dir))) { 1510e188dc02SMiklos Szeredi dput(dentry); 151144396f4bSJosef Bacik return ERR_PTR(-ENOENT); 1512e188dc02SMiklos Szeredi } 151344396f4bSJosef Bacik 151472bd866aSAl Viro old = dir->i_op->lookup(dir, dentry, flags); 151544396f4bSJosef Bacik if (unlikely(old)) { 151644396f4bSJosef Bacik dput(dentry); 151744396f4bSJosef Bacik dentry = old; 151844396f4bSJosef Bacik } 151944396f4bSJosef Bacik return dentry; 152044396f4bSJosef Bacik } 152144396f4bSJosef Bacik 1522e3c13928SAl Viro static struct dentry *__lookup_hash(const struct qstr *name, 152372bd866aSAl Viro struct dentry *base, unsigned int flags) 1524a3255546SAl Viro { 15256c51e513SAl Viro struct dentry *dentry = lookup_dcache(name, base, flags); 1526a3255546SAl Viro 15276c51e513SAl Viro if (dentry) 1528a3255546SAl Viro return dentry; 1529bad61189SMiklos Szeredi 15306c51e513SAl Viro dentry = d_alloc(base, name); 15316c51e513SAl Viro if (unlikely(!dentry)) 15326c51e513SAl Viro return ERR_PTR(-ENOMEM); 15336c51e513SAl Viro 153472bd866aSAl Viro return lookup_real(base->d_inode, dentry, flags); 1535a3255546SAl Viro } 1536a3255546SAl Viro 1537e97cdc87SAl Viro static int lookup_fast(struct nameidata *nd, 1538254cf582SAl Viro struct path *path, struct inode **inode, 1539254cf582SAl Viro unsigned *seqp) 15401da177e4SLinus Torvalds { 15414ac91378SJan Blunck struct vfsmount *mnt = nd->path.mnt; 154231e6b01fSNick Piggin struct dentry *dentry, *parent = nd->path.dentry; 15435a18fff2SAl Viro int status = 1; 15449875cf80SDavid Howells int err; 15459875cf80SDavid Howells 15463cac260aSAl Viro /* 1547b04f784eSNick Piggin * Rename seqlock is not required here because in the off chance 15485d0f49c1SAl Viro * of a false negative due to a concurrent rename, the caller is 15495d0f49c1SAl Viro * going to fall back to non-racy lookup. 1550b04f784eSNick Piggin */ 155131e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) { 155231e6b01fSNick Piggin unsigned seq; 1553766c4cbfSAl Viro bool negative; 1554da53be12SLinus Torvalds dentry = __d_lookup_rcu(parent, &nd->last, &seq); 15555d0f49c1SAl Viro if (unlikely(!dentry)) { 15564675ac39SAl Viro if (unlazy_walk(nd)) 15575d0f49c1SAl Viro return -ECHILD; 1558e9742b53SAl Viro return 0; 15595d0f49c1SAl Viro } 15605a18fff2SAl Viro 156112f8ad4bSLinus Torvalds /* 156212f8ad4bSLinus Torvalds * This sequence count validates that the inode matches 156312f8ad4bSLinus Torvalds * the dentry name information from lookup. 156412f8ad4bSLinus Torvalds */ 156563afdfc7SDavid Howells *inode = d_backing_inode(dentry); 1566766c4cbfSAl Viro negative = d_is_negative(dentry); 15675d0f49c1SAl Viro if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) 156812f8ad4bSLinus Torvalds return -ECHILD; 156912f8ad4bSLinus Torvalds 157012f8ad4bSLinus Torvalds /* 157112f8ad4bSLinus Torvalds * This sequence count validates that the parent had no 157212f8ad4bSLinus Torvalds * changes while we did the lookup of the dentry above. 157312f8ad4bSLinus Torvalds * 157412f8ad4bSLinus Torvalds * The memory barrier in read_seqcount_begin of child is 157512f8ad4bSLinus Torvalds * enough, we can use __read_seqcount_retry here. 157612f8ad4bSLinus Torvalds */ 15775d0f49c1SAl Viro if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq))) 157831e6b01fSNick Piggin return -ECHILD; 15795a18fff2SAl Viro 1580254cf582SAl Viro *seqp = seq; 15814ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 1582209a7fb2SAl Viro if (likely(status > 0)) { 1583daf3761cSTrond Myklebust /* 1584daf3761cSTrond Myklebust * Note: do negative dentry check after revalidation in 1585daf3761cSTrond Myklebust * case that drops it. 1586daf3761cSTrond Myklebust */ 15875d0f49c1SAl Viro if (unlikely(negative)) 1588daf3761cSTrond Myklebust return -ENOENT; 158931e6b01fSNick Piggin path->mnt = mnt; 159031e6b01fSNick Piggin path->dentry = dentry; 1591254cf582SAl Viro if (likely(__follow_mount_rcu(nd, path, inode, seqp))) 1592e9742b53SAl Viro return 1; 15935d0f49c1SAl Viro } 15944675ac39SAl Viro if (unlazy_child(nd, dentry, seq)) 1595254cf582SAl Viro return -ECHILD; 1596209a7fb2SAl Viro if (unlikely(status == -ECHILD)) 1597209a7fb2SAl Viro /* we'd been told to redo it in non-rcu mode */ 1598209a7fb2SAl Viro status = d_revalidate(dentry, nd->flags); 15995a18fff2SAl Viro } else { 1600e97cdc87SAl Viro dentry = __d_lookup(parent, &nd->last); 160181e6f520SAl Viro if (unlikely(!dentry)) 1602e9742b53SAl Viro return 0; 16034ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 16045d0f49c1SAl Viro } 16055a18fff2SAl Viro if (unlikely(status <= 0)) { 1606e9742b53SAl Viro if (!status) 16075d0f49c1SAl Viro d_invalidate(dentry); 16085a18fff2SAl Viro dput(dentry); 16095a18fff2SAl Viro return status; 16105a18fff2SAl Viro } 1611766c4cbfSAl Viro if (unlikely(d_is_negative(dentry))) { 1612766c4cbfSAl Viro dput(dentry); 1613766c4cbfSAl Viro return -ENOENT; 1614766c4cbfSAl Viro } 16155d0f49c1SAl Viro 16161da177e4SLinus Torvalds path->mnt = mnt; 16171da177e4SLinus Torvalds path->dentry = dentry; 1618756daf26SNeilBrown err = follow_managed(path, nd); 1619e9742b53SAl Viro if (likely(err > 0)) 162063afdfc7SDavid Howells *inode = d_backing_inode(path->dentry); 16218402752eSAl Viro return err; 1622697f514dSMiklos Szeredi } 1623697f514dSMiklos Szeredi 1624697f514dSMiklos Szeredi /* Fast lookup failed, do it the slow way */ 1625e3c13928SAl Viro static struct dentry *lookup_slow(const struct qstr *name, 1626e3c13928SAl Viro struct dentry *dir, 1627e3c13928SAl Viro unsigned int flags) 1628697f514dSMiklos Szeredi { 162994bdd655SAl Viro struct dentry *dentry = ERR_PTR(-ENOENT), *old; 16301936386eSAl Viro struct inode *inode = dir->d_inode; 1631d9171b93SAl Viro DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 16321936386eSAl Viro 16339902af79SAl Viro inode_lock_shared(inode); 16341936386eSAl Viro /* Don't go there if it's already dead */ 163594bdd655SAl Viro if (unlikely(IS_DEADDIR(inode))) 163694bdd655SAl Viro goto out; 163794bdd655SAl Viro again: 1638d9171b93SAl Viro dentry = d_alloc_parallel(dir, name, &wq); 163994bdd655SAl Viro if (IS_ERR(dentry)) 164094bdd655SAl Viro goto out; 164194bdd655SAl Viro if (unlikely(!d_in_lookup(dentry))) { 1642a89f8337SAl Viro if (!(flags & LOOKUP_NO_REVAL)) { 1643949a852eSAl Viro int error = d_revalidate(dentry, flags); 1644949a852eSAl Viro if (unlikely(error <= 0)) { 164594bdd655SAl Viro if (!error) { 1646949a852eSAl Viro d_invalidate(dentry); 1647949a852eSAl Viro dput(dentry); 164894bdd655SAl Viro goto again; 164994bdd655SAl Viro } 165094bdd655SAl Viro dput(dentry); 1651949a852eSAl Viro dentry = ERR_PTR(error); 1652949a852eSAl Viro } 1653949a852eSAl Viro } 165494bdd655SAl Viro } else { 16551936386eSAl Viro old = inode->i_op->lookup(inode, dentry, flags); 165685c7f810SAl Viro d_lookup_done(dentry); 16571936386eSAl Viro if (unlikely(old)) { 16581936386eSAl Viro dput(dentry); 16591936386eSAl Viro dentry = old; 1660949a852eSAl Viro } 1661949a852eSAl Viro } 166294bdd655SAl Viro out: 16639902af79SAl Viro inode_unlock_shared(inode); 1664e3c13928SAl Viro return dentry; 16651da177e4SLinus Torvalds } 16661da177e4SLinus Torvalds 166752094c8aSAl Viro static inline int may_lookup(struct nameidata *nd) 166852094c8aSAl Viro { 166952094c8aSAl Viro if (nd->flags & LOOKUP_RCU) { 16704ad5abb3SAl Viro int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); 167152094c8aSAl Viro if (err != -ECHILD) 167252094c8aSAl Viro return err; 16734675ac39SAl Viro if (unlazy_walk(nd)) 167452094c8aSAl Viro return -ECHILD; 167552094c8aSAl Viro } 16764ad5abb3SAl Viro return inode_permission(nd->inode, MAY_EXEC); 167752094c8aSAl Viro } 167852094c8aSAl Viro 16799856fa1bSAl Viro static inline int handle_dots(struct nameidata *nd, int type) 16809856fa1bSAl Viro { 16819856fa1bSAl Viro if (type == LAST_DOTDOT) { 16829e6697e2SAl Viro if (!nd->root.mnt) 16839e6697e2SAl Viro set_root(nd); 16849856fa1bSAl Viro if (nd->flags & LOOKUP_RCU) { 168570291aecSAl Viro return follow_dotdot_rcu(nd); 16869856fa1bSAl Viro } else 1687397d425dSEric W. Biederman return follow_dotdot(nd); 16889856fa1bSAl Viro } 16899856fa1bSAl Viro return 0; 16909856fa1bSAl Viro } 16919856fa1bSAl Viro 1692181548c0SAl Viro static int pick_link(struct nameidata *nd, struct path *link, 1693181548c0SAl Viro struct inode *inode, unsigned seq) 1694d63ff28fSAl Viro { 1695626de996SAl Viro int error; 16961cf2665bSAl Viro struct saved *last; 1697756daf26SNeilBrown if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) { 1698626de996SAl Viro path_to_nameidata(link, nd); 1699626de996SAl Viro return -ELOOP; 1700626de996SAl Viro } 1701bc40aee0SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1702cd179f44SAl Viro if (link->mnt == nd->path.mnt) 1703cd179f44SAl Viro mntget(link->mnt); 17047973387aSAl Viro } 1705626de996SAl Viro error = nd_alloc_stack(nd); 1706626de996SAl Viro if (unlikely(error)) { 1707bc40aee0SAl Viro if (error == -ECHILD) { 1708ad1633a1SAl Viro if (unlikely(!legitimize_path(nd, link, seq))) { 1709ad1633a1SAl Viro drop_links(nd); 1710ad1633a1SAl Viro nd->depth = 0; 1711ad1633a1SAl Viro nd->flags &= ~LOOKUP_RCU; 1712ad1633a1SAl Viro nd->path.mnt = NULL; 1713ad1633a1SAl Viro nd->path.dentry = NULL; 1714ad1633a1SAl Viro if (!(nd->flags & LOOKUP_ROOT)) 1715ad1633a1SAl Viro nd->root.mnt = NULL; 1716ad1633a1SAl Viro rcu_read_unlock(); 17174675ac39SAl Viro } else if (likely(unlazy_walk(nd)) == 0) 1718bc40aee0SAl Viro error = nd_alloc_stack(nd); 1719bc40aee0SAl Viro } 1720bc40aee0SAl Viro if (error) { 1721cd179f44SAl Viro path_put(link); 1722626de996SAl Viro return error; 1723626de996SAl Viro } 1724bc40aee0SAl Viro } 1725626de996SAl Viro 1726ab104923SAl Viro last = nd->stack + nd->depth++; 17271cf2665bSAl Viro last->link = *link; 1728fceef393SAl Viro clear_delayed_call(&last->done); 1729fceef393SAl Viro nd->link_inode = inode; 17300450b2d1SAl Viro last->seq = seq; 1731d63ff28fSAl Viro return 1; 1732d63ff28fSAl Viro } 1733d63ff28fSAl Viro 17348f64fb1cSAl Viro enum {WALK_FOLLOW = 1, WALK_MORE = 2}; 173531d66bcdSAl Viro 17363ddcd056SLinus Torvalds /* 17373ddcd056SLinus Torvalds * Do we need to follow links? We _really_ want to be able 17383ddcd056SLinus Torvalds * to do this check without having to look at inode->i_op, 17393ddcd056SLinus Torvalds * so we keep a cache of "no, this doesn't need follow_link" 17403ddcd056SLinus Torvalds * for the common case. 17413ddcd056SLinus Torvalds */ 17428f64fb1cSAl Viro static inline int step_into(struct nameidata *nd, struct path *path, 17438f64fb1cSAl Viro int flags, struct inode *inode, unsigned seq) 17443ddcd056SLinus Torvalds { 174531d66bcdSAl Viro if (!(flags & WALK_MORE) && nd->depth) 174631d66bcdSAl Viro put_link(nd); 17478f64fb1cSAl Viro if (likely(!d_is_symlink(path->dentry)) || 17488f64fb1cSAl Viro !(flags & WALK_FOLLOW || nd->flags & LOOKUP_FOLLOW)) { 17498f64fb1cSAl Viro /* not a symlink or should not follow */ 17508f64fb1cSAl Viro path_to_nameidata(path, nd); 17518f64fb1cSAl Viro nd->inode = inode; 17528f64fb1cSAl Viro nd->seq = seq; 1753d63ff28fSAl Viro return 0; 17548f64fb1cSAl Viro } 1755a7f77542SAl Viro /* make sure that d_is_symlink above matches inode */ 1756a7f77542SAl Viro if (nd->flags & LOOKUP_RCU) { 17578f64fb1cSAl Viro if (read_seqcount_retry(&path->dentry->d_seq, seq)) 1758a7f77542SAl Viro return -ECHILD; 1759a7f77542SAl Viro } 17608f64fb1cSAl Viro return pick_link(nd, path, inode, seq); 17613ddcd056SLinus Torvalds } 17623ddcd056SLinus Torvalds 17634693a547SAl Viro static int walk_component(struct nameidata *nd, int flags) 1764ce57dfc1SAl Viro { 1765caa85634SAl Viro struct path path; 1766ce57dfc1SAl Viro struct inode *inode; 1767254cf582SAl Viro unsigned seq; 1768ce57dfc1SAl Viro int err; 1769ce57dfc1SAl Viro /* 1770ce57dfc1SAl Viro * "." and ".." are special - ".." especially so because it has 1771ce57dfc1SAl Viro * to be able to know about the current root directory and 1772ce57dfc1SAl Viro * parent relationships. 1773ce57dfc1SAl Viro */ 17744693a547SAl Viro if (unlikely(nd->last_type != LAST_NORM)) { 17754693a547SAl Viro err = handle_dots(nd, nd->last_type); 17761c4ff1a8SAl Viro if (!(flags & WALK_MORE) && nd->depth) 17774693a547SAl Viro put_link(nd); 17784693a547SAl Viro return err; 17794693a547SAl Viro } 1780254cf582SAl Viro err = lookup_fast(nd, &path, &inode, &seq); 1781e9742b53SAl Viro if (unlikely(err <= 0)) { 1782697f514dSMiklos Szeredi if (err < 0) 1783f0a9ba70SAl Viro return err; 1784e3c13928SAl Viro path.dentry = lookup_slow(&nd->last, nd->path.dentry, 1785e3c13928SAl Viro nd->flags); 1786e3c13928SAl Viro if (IS_ERR(path.dentry)) 1787e3c13928SAl Viro return PTR_ERR(path.dentry); 17887500c38aSAl Viro 1789e3c13928SAl Viro path.mnt = nd->path.mnt; 1790e3c13928SAl Viro err = follow_managed(&path, nd); 1791e3c13928SAl Viro if (unlikely(err < 0)) 1792f0a9ba70SAl Viro return err; 1793697f514dSMiklos Szeredi 17947500c38aSAl Viro if (unlikely(d_is_negative(path.dentry))) { 17957500c38aSAl Viro path_to_nameidata(&path, nd); 17967500c38aSAl Viro return -ENOENT; 17977500c38aSAl Viro } 17987500c38aSAl Viro 1799254cf582SAl Viro seq = 0; /* we are already out of RCU mode */ 1800d4565649SAl Viro inode = d_backing_inode(path.dentry); 1801766c4cbfSAl Viro } 1802697f514dSMiklos Szeredi 18038f64fb1cSAl Viro return step_into(nd, &path, flags, inode, seq); 1804ce57dfc1SAl Viro } 1805ce57dfc1SAl Viro 18061da177e4SLinus Torvalds /* 1807bfcfaa77SLinus Torvalds * We can do the critical dentry name comparison and hashing 1808bfcfaa77SLinus Torvalds * operations one word at a time, but we are limited to: 1809bfcfaa77SLinus Torvalds * 1810bfcfaa77SLinus Torvalds * - Architectures with fast unaligned word accesses. We could 1811bfcfaa77SLinus Torvalds * do a "get_unaligned()" if this helps and is sufficiently 1812bfcfaa77SLinus Torvalds * fast. 1813bfcfaa77SLinus Torvalds * 1814bfcfaa77SLinus Torvalds * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we 1815bfcfaa77SLinus Torvalds * do not trap on the (extremely unlikely) case of a page 1816bfcfaa77SLinus Torvalds * crossing operation. 1817bfcfaa77SLinus Torvalds * 1818bfcfaa77SLinus Torvalds * - Furthermore, we need an efficient 64-bit compile for the 1819bfcfaa77SLinus Torvalds * 64-bit case in order to generate the "number of bytes in 1820bfcfaa77SLinus Torvalds * the final mask". Again, that could be replaced with a 1821bfcfaa77SLinus Torvalds * efficient population count instruction or similar. 1822bfcfaa77SLinus Torvalds */ 1823bfcfaa77SLinus Torvalds #ifdef CONFIG_DCACHE_WORD_ACCESS 1824bfcfaa77SLinus Torvalds 1825f68e556eSLinus Torvalds #include <asm/word-at-a-time.h> 1826bfcfaa77SLinus Torvalds 1827468a9428SGeorge Spelvin #ifdef HASH_MIX 1828bfcfaa77SLinus Torvalds 1829468a9428SGeorge Spelvin /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */ 1830468a9428SGeorge Spelvin 1831468a9428SGeorge Spelvin #elif defined(CONFIG_64BIT) 18322a18da7aSGeorge Spelvin /* 18332a18da7aSGeorge Spelvin * Register pressure in the mixing function is an issue, particularly 18342a18da7aSGeorge Spelvin * on 32-bit x86, but almost any function requires one state value and 18352a18da7aSGeorge Spelvin * one temporary. Instead, use a function designed for two state values 18362a18da7aSGeorge Spelvin * and no temporaries. 18372a18da7aSGeorge Spelvin * 18382a18da7aSGeorge Spelvin * This function cannot create a collision in only two iterations, so 18392a18da7aSGeorge Spelvin * we have two iterations to achieve avalanche. In those two iterations, 18402a18da7aSGeorge Spelvin * we have six layers of mixing, which is enough to spread one bit's 18412a18da7aSGeorge Spelvin * influence out to 2^6 = 64 state bits. 18422a18da7aSGeorge Spelvin * 18432a18da7aSGeorge Spelvin * Rotate constants are scored by considering either 64 one-bit input 18442a18da7aSGeorge Spelvin * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the 18452a18da7aSGeorge Spelvin * probability of that delta causing a change to each of the 128 output 18462a18da7aSGeorge Spelvin * bits, using a sample of random initial states. 18472a18da7aSGeorge Spelvin * 18482a18da7aSGeorge Spelvin * The Shannon entropy of the computed probabilities is then summed 18492a18da7aSGeorge Spelvin * to produce a score. Ideally, any input change has a 50% chance of 18502a18da7aSGeorge Spelvin * toggling any given output bit. 18512a18da7aSGeorge Spelvin * 18522a18da7aSGeorge Spelvin * Mixing scores (in bits) for (12,45): 18532a18da7aSGeorge Spelvin * Input delta: 1-bit 2-bit 18542a18da7aSGeorge Spelvin * 1 round: 713.3 42542.6 18552a18da7aSGeorge Spelvin * 2 rounds: 2753.7 140389.8 18562a18da7aSGeorge Spelvin * 3 rounds: 5954.1 233458.2 18572a18da7aSGeorge Spelvin * 4 rounds: 7862.6 256672.2 18582a18da7aSGeorge Spelvin * Perfect: 8192 258048 18592a18da7aSGeorge Spelvin * (64*128) (64*63/2 * 128) 18602a18da7aSGeorge Spelvin */ 18612a18da7aSGeorge Spelvin #define HASH_MIX(x, y, a) \ 18622a18da7aSGeorge Spelvin ( x ^= (a), \ 18632a18da7aSGeorge Spelvin y ^= x, x = rol64(x,12),\ 18642a18da7aSGeorge Spelvin x += y, y = rol64(y,45),\ 18652a18da7aSGeorge Spelvin y *= 9 ) 1866bfcfaa77SLinus Torvalds 18670fed3ac8SGeorge Spelvin /* 18682a18da7aSGeorge Spelvin * Fold two longs into one 32-bit hash value. This must be fast, but 18692a18da7aSGeorge Spelvin * latency isn't quite as critical, as there is a fair bit of additional 18702a18da7aSGeorge Spelvin * work done before the hash value is used. 18710fed3ac8SGeorge Spelvin */ 18722a18da7aSGeorge Spelvin static inline unsigned int fold_hash(unsigned long x, unsigned long y) 18730fed3ac8SGeorge Spelvin { 18742a18da7aSGeorge Spelvin y ^= x * GOLDEN_RATIO_64; 18752a18da7aSGeorge Spelvin y *= GOLDEN_RATIO_64; 18762a18da7aSGeorge Spelvin return y >> 32; 18770fed3ac8SGeorge Spelvin } 18780fed3ac8SGeorge Spelvin 1879bfcfaa77SLinus Torvalds #else /* 32-bit case */ 1880bfcfaa77SLinus Torvalds 18812a18da7aSGeorge Spelvin /* 18822a18da7aSGeorge Spelvin * Mixing scores (in bits) for (7,20): 18832a18da7aSGeorge Spelvin * Input delta: 1-bit 2-bit 18842a18da7aSGeorge Spelvin * 1 round: 330.3 9201.6 18852a18da7aSGeorge Spelvin * 2 rounds: 1246.4 25475.4 18862a18da7aSGeorge Spelvin * 3 rounds: 1907.1 31295.1 18872a18da7aSGeorge Spelvin * 4 rounds: 2042.3 31718.6 18882a18da7aSGeorge Spelvin * Perfect: 2048 31744 18892a18da7aSGeorge Spelvin * (32*64) (32*31/2 * 64) 18902a18da7aSGeorge Spelvin */ 18912a18da7aSGeorge Spelvin #define HASH_MIX(x, y, a) \ 18922a18da7aSGeorge Spelvin ( x ^= (a), \ 18932a18da7aSGeorge Spelvin y ^= x, x = rol32(x, 7),\ 18942a18da7aSGeorge Spelvin x += y, y = rol32(y,20),\ 18952a18da7aSGeorge Spelvin y *= 9 ) 1896bfcfaa77SLinus Torvalds 18972a18da7aSGeorge Spelvin static inline unsigned int fold_hash(unsigned long x, unsigned long y) 18980fed3ac8SGeorge Spelvin { 18992a18da7aSGeorge Spelvin /* Use arch-optimized multiply if one exists */ 19002a18da7aSGeorge Spelvin return __hash_32(y ^ __hash_32(x)); 19010fed3ac8SGeorge Spelvin } 19020fed3ac8SGeorge Spelvin 1903bfcfaa77SLinus Torvalds #endif 1904bfcfaa77SLinus Torvalds 19052a18da7aSGeorge Spelvin /* 19062a18da7aSGeorge Spelvin * Return the hash of a string of known length. This is carfully 19072a18da7aSGeorge Spelvin * designed to match hash_name(), which is the more critical function. 19082a18da7aSGeorge Spelvin * In particular, we must end by hashing a final word containing 0..7 19092a18da7aSGeorge Spelvin * payload bytes, to match the way that hash_name() iterates until it 19102a18da7aSGeorge Spelvin * finds the delimiter after the name. 19112a18da7aSGeorge Spelvin */ 19128387ff25SLinus Torvalds unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) 1913bfcfaa77SLinus Torvalds { 19148387ff25SLinus Torvalds unsigned long a, x = 0, y = (unsigned long)salt; 1915bfcfaa77SLinus Torvalds 1916bfcfaa77SLinus Torvalds for (;;) { 1917fcfd2fbfSGeorge Spelvin if (!len) 1918fcfd2fbfSGeorge Spelvin goto done; 1919e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name); 1920bfcfaa77SLinus Torvalds if (len < sizeof(unsigned long)) 1921bfcfaa77SLinus Torvalds break; 19222a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 1923bfcfaa77SLinus Torvalds name += sizeof(unsigned long); 1924bfcfaa77SLinus Torvalds len -= sizeof(unsigned long); 1925bfcfaa77SLinus Torvalds } 19262a18da7aSGeorge Spelvin x ^= a & bytemask_from_count(len); 1927bfcfaa77SLinus Torvalds done: 19282a18da7aSGeorge Spelvin return fold_hash(x, y); 1929bfcfaa77SLinus Torvalds } 1930bfcfaa77SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 1931bfcfaa77SLinus Torvalds 1932fcfd2fbfSGeorge Spelvin /* Return the "hash_len" (hash and length) of a null-terminated string */ 19338387ff25SLinus Torvalds u64 hashlen_string(const void *salt, const char *name) 1934fcfd2fbfSGeorge Spelvin { 19358387ff25SLinus Torvalds unsigned long a = 0, x = 0, y = (unsigned long)salt; 19368387ff25SLinus Torvalds unsigned long adata, mask, len; 1937fcfd2fbfSGeorge Spelvin const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 1938fcfd2fbfSGeorge Spelvin 19398387ff25SLinus Torvalds len = 0; 19408387ff25SLinus Torvalds goto inside; 19418387ff25SLinus Torvalds 1942fcfd2fbfSGeorge Spelvin do { 19432a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 1944fcfd2fbfSGeorge Spelvin len += sizeof(unsigned long); 19458387ff25SLinus Torvalds inside: 1946fcfd2fbfSGeorge Spelvin a = load_unaligned_zeropad(name+len); 1947fcfd2fbfSGeorge Spelvin } while (!has_zero(a, &adata, &constants)); 1948fcfd2fbfSGeorge Spelvin 1949fcfd2fbfSGeorge Spelvin adata = prep_zero_mask(a, adata, &constants); 1950fcfd2fbfSGeorge Spelvin mask = create_zero_mask(adata); 19512a18da7aSGeorge Spelvin x ^= a & zero_bytemask(mask); 1952fcfd2fbfSGeorge Spelvin 19532a18da7aSGeorge Spelvin return hashlen_create(fold_hash(x, y), len + find_zero(mask)); 1954fcfd2fbfSGeorge Spelvin } 1955fcfd2fbfSGeorge Spelvin EXPORT_SYMBOL(hashlen_string); 1956fcfd2fbfSGeorge Spelvin 1957bfcfaa77SLinus Torvalds /* 1958bfcfaa77SLinus Torvalds * Calculate the length and hash of the path component, and 1959d6bb3e90SLinus Torvalds * return the "hash_len" as the result. 1960bfcfaa77SLinus Torvalds */ 19618387ff25SLinus Torvalds static inline u64 hash_name(const void *salt, const char *name) 1962bfcfaa77SLinus Torvalds { 19638387ff25SLinus Torvalds unsigned long a = 0, b, x = 0, y = (unsigned long)salt; 19648387ff25SLinus Torvalds unsigned long adata, bdata, mask, len; 196536126f8fSLinus Torvalds const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 1966bfcfaa77SLinus Torvalds 19678387ff25SLinus Torvalds len = 0; 19688387ff25SLinus Torvalds goto inside; 19698387ff25SLinus Torvalds 1970bfcfaa77SLinus Torvalds do { 19712a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 1972bfcfaa77SLinus Torvalds len += sizeof(unsigned long); 19738387ff25SLinus Torvalds inside: 1974e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name+len); 197536126f8fSLinus Torvalds b = a ^ REPEAT_BYTE('/'); 197636126f8fSLinus Torvalds } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); 1977bfcfaa77SLinus Torvalds 197836126f8fSLinus Torvalds adata = prep_zero_mask(a, adata, &constants); 197936126f8fSLinus Torvalds bdata = prep_zero_mask(b, bdata, &constants); 198036126f8fSLinus Torvalds mask = create_zero_mask(adata | bdata); 19812a18da7aSGeorge Spelvin x ^= a & zero_bytemask(mask); 198236126f8fSLinus Torvalds 19832a18da7aSGeorge Spelvin return hashlen_create(fold_hash(x, y), len + find_zero(mask)); 1984bfcfaa77SLinus Torvalds } 1985bfcfaa77SLinus Torvalds 19862a18da7aSGeorge Spelvin #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */ 1987bfcfaa77SLinus Torvalds 1988fcfd2fbfSGeorge Spelvin /* Return the hash of a string of known length */ 19898387ff25SLinus Torvalds unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) 19900145acc2SLinus Torvalds { 19918387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 19920145acc2SLinus Torvalds while (len--) 1993fcfd2fbfSGeorge Spelvin hash = partial_name_hash((unsigned char)*name++, hash); 19940145acc2SLinus Torvalds return end_name_hash(hash); 19950145acc2SLinus Torvalds } 1996ae942ae7SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 19970145acc2SLinus Torvalds 1998fcfd2fbfSGeorge Spelvin /* Return the "hash_len" (hash and length) of a null-terminated string */ 19998387ff25SLinus Torvalds u64 hashlen_string(const void *salt, const char *name) 2000fcfd2fbfSGeorge Spelvin { 20018387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 2002fcfd2fbfSGeorge Spelvin unsigned long len = 0, c; 2003fcfd2fbfSGeorge Spelvin 2004fcfd2fbfSGeorge Spelvin c = (unsigned char)*name; 2005e0ab7af9SGeorge Spelvin while (c) { 2006fcfd2fbfSGeorge Spelvin len++; 2007fcfd2fbfSGeorge Spelvin hash = partial_name_hash(c, hash); 2008fcfd2fbfSGeorge Spelvin c = (unsigned char)name[len]; 2009e0ab7af9SGeorge Spelvin } 2010fcfd2fbfSGeorge Spelvin return hashlen_create(end_name_hash(hash), len); 2011fcfd2fbfSGeorge Spelvin } 2012f2a031b6SGeorge Spelvin EXPORT_SYMBOL(hashlen_string); 2013fcfd2fbfSGeorge Spelvin 20143ddcd056SLinus Torvalds /* 2015200e9ef7SLinus Torvalds * We know there's a real path component here of at least 2016200e9ef7SLinus Torvalds * one character. 2017200e9ef7SLinus Torvalds */ 20188387ff25SLinus Torvalds static inline u64 hash_name(const void *salt, const char *name) 2019200e9ef7SLinus Torvalds { 20208387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 2021200e9ef7SLinus Torvalds unsigned long len = 0, c; 2022200e9ef7SLinus Torvalds 2023200e9ef7SLinus Torvalds c = (unsigned char)*name; 2024200e9ef7SLinus Torvalds do { 2025200e9ef7SLinus Torvalds len++; 2026200e9ef7SLinus Torvalds hash = partial_name_hash(c, hash); 2027200e9ef7SLinus Torvalds c = (unsigned char)name[len]; 2028200e9ef7SLinus Torvalds } while (c && c != '/'); 2029d6bb3e90SLinus Torvalds return hashlen_create(end_name_hash(hash), len); 2030200e9ef7SLinus Torvalds } 2031200e9ef7SLinus Torvalds 2032bfcfaa77SLinus Torvalds #endif 2033bfcfaa77SLinus Torvalds 2034200e9ef7SLinus Torvalds /* 20351da177e4SLinus Torvalds * Name resolution. 2036ea3834d9SPrasanna Meda * This is the basic name resolution function, turning a pathname into 2037ea3834d9SPrasanna Meda * the final dentry. We expect 'base' to be positive and a directory. 20381da177e4SLinus Torvalds * 2039ea3834d9SPrasanna Meda * Returns 0 and nd will have valid dentry and mnt on success. 2040ea3834d9SPrasanna Meda * Returns error and drops reference to input namei data on failure. 20411da177e4SLinus Torvalds */ 20426de88d72SAl Viro static int link_path_walk(const char *name, struct nameidata *nd) 20431da177e4SLinus Torvalds { 20441da177e4SLinus Torvalds int err; 20451da177e4SLinus Torvalds 20461da177e4SLinus Torvalds while (*name=='/') 20471da177e4SLinus Torvalds name++; 20481da177e4SLinus Torvalds if (!*name) 20499e18f10aSAl Viro return 0; 20501da177e4SLinus Torvalds 20511da177e4SLinus Torvalds /* At this point we know we have a real path component. */ 20521da177e4SLinus Torvalds for(;;) { 2053d6bb3e90SLinus Torvalds u64 hash_len; 2054fe479a58SAl Viro int type; 20551da177e4SLinus Torvalds 205652094c8aSAl Viro err = may_lookup(nd); 20571da177e4SLinus Torvalds if (err) 20583595e234SAl Viro return err; 20591da177e4SLinus Torvalds 20608387ff25SLinus Torvalds hash_len = hash_name(nd->path.dentry, name); 20611da177e4SLinus Torvalds 2062fe479a58SAl Viro type = LAST_NORM; 2063d6bb3e90SLinus Torvalds if (name[0] == '.') switch (hashlen_len(hash_len)) { 2064fe479a58SAl Viro case 2: 2065200e9ef7SLinus Torvalds if (name[1] == '.') { 2066fe479a58SAl Viro type = LAST_DOTDOT; 206716c2cd71SAl Viro nd->flags |= LOOKUP_JUMPED; 206816c2cd71SAl Viro } 2069fe479a58SAl Viro break; 2070fe479a58SAl Viro case 1: 2071fe479a58SAl Viro type = LAST_DOT; 2072fe479a58SAl Viro } 20735a202bcdSAl Viro if (likely(type == LAST_NORM)) { 20745a202bcdSAl Viro struct dentry *parent = nd->path.dentry; 207516c2cd71SAl Viro nd->flags &= ~LOOKUP_JUMPED; 20765a202bcdSAl Viro if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { 2077a060dc50SJames Hogan struct qstr this = { { .hash_len = hash_len }, .name = name }; 2078da53be12SLinus Torvalds err = parent->d_op->d_hash(parent, &this); 20795a202bcdSAl Viro if (err < 0) 20803595e234SAl Viro return err; 2081d6bb3e90SLinus Torvalds hash_len = this.hash_len; 2082d6bb3e90SLinus Torvalds name = this.name; 20835a202bcdSAl Viro } 20845a202bcdSAl Viro } 2085fe479a58SAl Viro 2086d6bb3e90SLinus Torvalds nd->last.hash_len = hash_len; 2087d6bb3e90SLinus Torvalds nd->last.name = name; 20885f4a6a69SAl Viro nd->last_type = type; 20895f4a6a69SAl Viro 2090d6bb3e90SLinus Torvalds name += hashlen_len(hash_len); 2091d6bb3e90SLinus Torvalds if (!*name) 2092bdf6cbf1SAl Viro goto OK; 2093200e9ef7SLinus Torvalds /* 2094200e9ef7SLinus Torvalds * If it wasn't NUL, we know it was '/'. Skip that 2095200e9ef7SLinus Torvalds * slash, and continue until no more slashes. 2096200e9ef7SLinus Torvalds */ 2097200e9ef7SLinus Torvalds do { 2098d6bb3e90SLinus Torvalds name++; 2099d6bb3e90SLinus Torvalds } while (unlikely(*name == '/')); 21008620c238SAl Viro if (unlikely(!*name)) { 21018620c238SAl Viro OK: 2102368ee9baSAl Viro /* pathname body, done */ 21038620c238SAl Viro if (!nd->depth) 21048620c238SAl Viro return 0; 21058620c238SAl Viro name = nd->stack[nd->depth - 1].name; 2106368ee9baSAl Viro /* trailing symlink, done */ 21078620c238SAl Viro if (!name) 21088620c238SAl Viro return 0; 21098620c238SAl Viro /* last component of nested symlink */ 21108f64fb1cSAl Viro err = walk_component(nd, WALK_FOLLOW); 21111c4ff1a8SAl Viro } else { 21121c4ff1a8SAl Viro /* not the last component */ 21138f64fb1cSAl Viro err = walk_component(nd, WALK_FOLLOW | WALK_MORE); 21148620c238SAl Viro } 2115ce57dfc1SAl Viro if (err < 0) 21163595e234SAl Viro return err; 2117fe479a58SAl Viro 2118ce57dfc1SAl Viro if (err) { 2119626de996SAl Viro const char *s = get_link(nd); 21205a460275SAl Viro 2121a1c83681SViresh Kumar if (IS_ERR(s)) 21223595e234SAl Viro return PTR_ERR(s); 2123172a39a0SAl Viro err = 0; 212412b09578SAl Viro if (unlikely(!s)) { 212512b09578SAl Viro /* jumped */ 2126b9ff4429SAl Viro put_link(nd); 212712b09578SAl Viro } else { 2128071bf501SAl Viro nd->stack[nd->depth - 1].name = name; 212932cd7468SAl Viro name = s; 21309e18f10aSAl Viro continue; 213148c8b0c5SAl Viro } 213231e6b01fSNick Piggin } 213397242f99SAl Viro if (unlikely(!d_can_lookup(nd->path.dentry))) { 213497242f99SAl Viro if (nd->flags & LOOKUP_RCU) { 21354675ac39SAl Viro if (unlazy_walk(nd)) 213697242f99SAl Viro return -ECHILD; 213797242f99SAl Viro } 21383595e234SAl Viro return -ENOTDIR; 21395f4a6a69SAl Viro } 2140ce57dfc1SAl Viro } 214197242f99SAl Viro } 21421da177e4SLinus Torvalds 2143c8a53ee5SAl Viro static const char *path_init(struct nameidata *nd, unsigned flags) 214431e6b01fSNick Piggin { 2145c8a53ee5SAl Viro const char *s = nd->name->name; 214631e6b01fSNick Piggin 2147c0eb027eSLinus Torvalds if (!*s) 2148c0eb027eSLinus Torvalds flags &= ~LOOKUP_RCU; 2149c0eb027eSLinus Torvalds 215031e6b01fSNick Piggin nd->last_type = LAST_ROOT; /* if there are only slashes... */ 2151980f3ea2SAl Viro nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT; 215231e6b01fSNick Piggin nd->depth = 0; 21535b6ca027SAl Viro if (flags & LOOKUP_ROOT) { 2154b18825a7SDavid Howells struct dentry *root = nd->root.dentry; 2155b18825a7SDavid Howells struct inode *inode = root->d_inode; 215693893862SAl Viro if (*s && unlikely(!d_can_lookup(root))) 2157368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 21585b6ca027SAl Viro nd->path = nd->root; 21595b6ca027SAl Viro nd->inode = inode; 21605b6ca027SAl Viro if (flags & LOOKUP_RCU) { 21618b61e74fSAl Viro rcu_read_lock(); 21625b6ca027SAl Viro nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 21638f47a016SAl Viro nd->root_seq = nd->seq; 216448a066e7SAl Viro nd->m_seq = read_seqbegin(&mount_lock); 21655b6ca027SAl Viro } else { 21665b6ca027SAl Viro path_get(&nd->path); 21675b6ca027SAl Viro } 2168368ee9baSAl Viro return s; 21695b6ca027SAl Viro } 21705b6ca027SAl Viro 217131e6b01fSNick Piggin nd->root.mnt = NULL; 2172248fb5b9SAl Viro nd->path.mnt = NULL; 2173248fb5b9SAl Viro nd->path.dentry = NULL; 217431e6b01fSNick Piggin 217548a066e7SAl Viro nd->m_seq = read_seqbegin(&mount_lock); 2176fd2f7cb5SAl Viro if (*s == '/') { 21779e6697e2SAl Viro if (flags & LOOKUP_RCU) 21788b61e74fSAl Viro rcu_read_lock(); 2179e41f7d4eSAl Viro set_root(nd); 2180248fb5b9SAl Viro if (likely(!nd_jump_root(nd))) 2181ef55d917SAl Viro return s; 2182ef55d917SAl Viro nd->root.mnt = NULL; 2183ef55d917SAl Viro rcu_read_unlock(); 2184ef55d917SAl Viro return ERR_PTR(-ECHILD); 2185c8a53ee5SAl Viro } else if (nd->dfd == AT_FDCWD) { 2186e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 218731e6b01fSNick Piggin struct fs_struct *fs = current->fs; 2188c28cc364SNick Piggin unsigned seq; 218931e6b01fSNick Piggin 21908b61e74fSAl Viro rcu_read_lock(); 219131e6b01fSNick Piggin 2192c28cc364SNick Piggin do { 2193c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 219431e6b01fSNick Piggin nd->path = fs->pwd; 2195ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2196c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 2197c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 2198e41f7d4eSAl Viro } else { 2199e41f7d4eSAl Viro get_fs_pwd(current->fs, &nd->path); 2200ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2201e41f7d4eSAl Viro } 2202ef55d917SAl Viro return s; 220331e6b01fSNick Piggin } else { 2204582aa64aSJeff Layton /* Caller must check execute permissions on the starting path component */ 2205c8a53ee5SAl Viro struct fd f = fdget_raw(nd->dfd); 220631e6b01fSNick Piggin struct dentry *dentry; 220731e6b01fSNick Piggin 22082903ff01SAl Viro if (!f.file) 2209368ee9baSAl Viro return ERR_PTR(-EBADF); 221031e6b01fSNick Piggin 22112903ff01SAl Viro dentry = f.file->f_path.dentry; 221231e6b01fSNick Piggin 2213fd2f7cb5SAl Viro if (*s) { 221444b1d530SMiklos Szeredi if (!d_can_lookup(dentry)) { 22152903ff01SAl Viro fdput(f); 2216368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 2217f52e0c11SAl Viro } 22182903ff01SAl Viro } 22192903ff01SAl Viro 22202903ff01SAl Viro nd->path = f.file->f_path; 2221e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 22228b61e74fSAl Viro rcu_read_lock(); 222334a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 222434a26b99SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 22255590ff0dSUlrich Drepper } else { 22262903ff01SAl Viro path_get(&nd->path); 222734a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 22281da177e4SLinus Torvalds } 222934a26b99SAl Viro fdput(f); 2230368ee9baSAl Viro return s; 2231e41f7d4eSAl Viro } 22329b4a9b14SAl Viro } 22339b4a9b14SAl Viro 22343bdba28bSAl Viro static const char *trailing_symlink(struct nameidata *nd) 223595fa25d9SAl Viro { 223695fa25d9SAl Viro const char *s; 2237fec2fa24SAl Viro int error = may_follow_link(nd); 2238deb106c6SAl Viro if (unlikely(error)) 22393bdba28bSAl Viro return ERR_PTR(error); 224095fa25d9SAl Viro nd->flags |= LOOKUP_PARENT; 2241fab51e8aSAl Viro nd->stack[0].name = NULL; 22423b2e7f75SAl Viro s = get_link(nd); 2243deb106c6SAl Viro return s ? s : ""; 224495fa25d9SAl Viro } 224595fa25d9SAl Viro 2246caa85634SAl Viro static inline int lookup_last(struct nameidata *nd) 2247bd92d7feSAl Viro { 2248bd92d7feSAl Viro if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) 2249bd92d7feSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 2250bd92d7feSAl Viro 2251bd92d7feSAl Viro nd->flags &= ~LOOKUP_PARENT; 22521c4ff1a8SAl Viro return walk_component(nd, 0); 2253bd92d7feSAl Viro } 2254bd92d7feSAl Viro 22554f757f3cSAl Viro static int handle_lookup_down(struct nameidata *nd) 22564f757f3cSAl Viro { 22574f757f3cSAl Viro struct path path = nd->path; 22584f757f3cSAl Viro struct inode *inode = nd->inode; 22594f757f3cSAl Viro unsigned seq = nd->seq; 22604f757f3cSAl Viro int err; 22614f757f3cSAl Viro 22624f757f3cSAl Viro if (nd->flags & LOOKUP_RCU) { 22634f757f3cSAl Viro /* 22644f757f3cSAl Viro * don't bother with unlazy_walk on failure - we are 22654f757f3cSAl Viro * at the very beginning of walk, so we lose nothing 22664f757f3cSAl Viro * if we simply redo everything in non-RCU mode 22674f757f3cSAl Viro */ 22684f757f3cSAl Viro if (unlikely(!__follow_mount_rcu(nd, &path, &inode, &seq))) 22694f757f3cSAl Viro return -ECHILD; 22704f757f3cSAl Viro } else { 22714f757f3cSAl Viro dget(path.dentry); 22724f757f3cSAl Viro err = follow_managed(&path, nd); 22734f757f3cSAl Viro if (unlikely(err < 0)) 22744f757f3cSAl Viro return err; 22754f757f3cSAl Viro inode = d_backing_inode(path.dentry); 22764f757f3cSAl Viro seq = 0; 22774f757f3cSAl Viro } 22784f757f3cSAl Viro path_to_nameidata(&path, nd); 22794f757f3cSAl Viro nd->inode = inode; 22804f757f3cSAl Viro nd->seq = seq; 22814f757f3cSAl Viro return 0; 22824f757f3cSAl Viro } 22834f757f3cSAl Viro 22849b4a9b14SAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2285c8a53ee5SAl Viro static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path) 22869b4a9b14SAl Viro { 2287c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2288bd92d7feSAl Viro int err; 228931e6b01fSNick Piggin 2290368ee9baSAl Viro if (IS_ERR(s)) 2291368ee9baSAl Viro return PTR_ERR(s); 22924f757f3cSAl Viro 22934f757f3cSAl Viro if (unlikely(flags & LOOKUP_DOWN)) { 22944f757f3cSAl Viro err = handle_lookup_down(nd); 22954f757f3cSAl Viro if (unlikely(err < 0)) { 22964f757f3cSAl Viro terminate_walk(nd); 22974f757f3cSAl Viro return err; 22984f757f3cSAl Viro } 22994f757f3cSAl Viro } 23004f757f3cSAl Viro 23013bdba28bSAl Viro while (!(err = link_path_walk(s, nd)) 23023bdba28bSAl Viro && ((err = lookup_last(nd)) > 0)) { 23033bdba28bSAl Viro s = trailing_symlink(nd); 23043bdba28bSAl Viro if (IS_ERR(s)) { 23053bdba28bSAl Viro err = PTR_ERR(s); 23066d7b5aaeSAl Viro break; 2307bd92d7feSAl Viro } 2308bd92d7feSAl Viro } 23099f1fafeeSAl Viro if (!err) 23109f1fafeeSAl Viro err = complete_walk(nd); 2311bd92d7feSAl Viro 2312deb106c6SAl Viro if (!err && nd->flags & LOOKUP_DIRECTORY) 2313deb106c6SAl Viro if (!d_can_lookup(nd->path.dentry)) 2314bd23a539SAl Viro err = -ENOTDIR; 2315625b6d10SAl Viro if (!err) { 2316625b6d10SAl Viro *path = nd->path; 2317625b6d10SAl Viro nd->path.mnt = NULL; 2318625b6d10SAl Viro nd->path.dentry = NULL; 2319625b6d10SAl Viro } 2320deb106c6SAl Viro terminate_walk(nd); 2321bd92d7feSAl Viro return err; 232231e6b01fSNick Piggin } 232331e6b01fSNick Piggin 2324625b6d10SAl Viro static int filename_lookup(int dfd, struct filename *name, unsigned flags, 23259ad1aaa6SAl Viro struct path *path, struct path *root) 2326873f1eedSJeff Layton { 2327894bc8c4SAl Viro int retval; 23289883d185SAl Viro struct nameidata nd; 2329abc9f5beSAl Viro if (IS_ERR(name)) 2330abc9f5beSAl Viro return PTR_ERR(name); 23319ad1aaa6SAl Viro if (unlikely(root)) { 23329ad1aaa6SAl Viro nd.root = *root; 23339ad1aaa6SAl Viro flags |= LOOKUP_ROOT; 23349ad1aaa6SAl Viro } 23359883d185SAl Viro set_nameidata(&nd, dfd, name); 2336c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_RCU, path); 2337873f1eedSJeff Layton if (unlikely(retval == -ECHILD)) 2338c8a53ee5SAl Viro retval = path_lookupat(&nd, flags, path); 2339873f1eedSJeff Layton if (unlikely(retval == -ESTALE)) 2340c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path); 2341873f1eedSJeff Layton 2342873f1eedSJeff Layton if (likely(!retval)) 2343625b6d10SAl Viro audit_inode(name, path->dentry, flags & LOOKUP_PARENT); 23449883d185SAl Viro restore_nameidata(); 2345e4bd1c1aSAl Viro putname(name); 2346873f1eedSJeff Layton return retval; 2347873f1eedSJeff Layton } 2348873f1eedSJeff Layton 23498bcb77faSAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2350c8a53ee5SAl Viro static int path_parentat(struct nameidata *nd, unsigned flags, 2351391172c4SAl Viro struct path *parent) 23528bcb77faSAl Viro { 2353c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2354368ee9baSAl Viro int err; 2355368ee9baSAl Viro if (IS_ERR(s)) 2356368ee9baSAl Viro return PTR_ERR(s); 2357368ee9baSAl Viro err = link_path_walk(s, nd); 23588bcb77faSAl Viro if (!err) 23598bcb77faSAl Viro err = complete_walk(nd); 2360391172c4SAl Viro if (!err) { 2361391172c4SAl Viro *parent = nd->path; 2362391172c4SAl Viro nd->path.mnt = NULL; 2363391172c4SAl Viro nd->path.dentry = NULL; 2364391172c4SAl Viro } 2365deb106c6SAl Viro terminate_walk(nd); 23668bcb77faSAl Viro return err; 23678bcb77faSAl Viro } 23688bcb77faSAl Viro 23695c31b6ceSAl Viro static struct filename *filename_parentat(int dfd, struct filename *name, 2370391172c4SAl Viro unsigned int flags, struct path *parent, 2371391172c4SAl Viro struct qstr *last, int *type) 23728bcb77faSAl Viro { 23738bcb77faSAl Viro int retval; 23749883d185SAl Viro struct nameidata nd; 23758bcb77faSAl Viro 23765c31b6ceSAl Viro if (IS_ERR(name)) 23775c31b6ceSAl Viro return name; 23789883d185SAl Viro set_nameidata(&nd, dfd, name); 2379c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_RCU, parent); 23808bcb77faSAl Viro if (unlikely(retval == -ECHILD)) 2381c8a53ee5SAl Viro retval = path_parentat(&nd, flags, parent); 23828bcb77faSAl Viro if (unlikely(retval == -ESTALE)) 2383c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent); 2384391172c4SAl Viro if (likely(!retval)) { 2385391172c4SAl Viro *last = nd.last; 2386391172c4SAl Viro *type = nd.last_type; 2387391172c4SAl Viro audit_inode(name, parent->dentry, LOOKUP_PARENT); 23885c31b6ceSAl Viro } else { 23895c31b6ceSAl Viro putname(name); 23905c31b6ceSAl Viro name = ERR_PTR(retval); 2391391172c4SAl Viro } 23929883d185SAl Viro restore_nameidata(); 23935c31b6ceSAl Viro return name; 23948bcb77faSAl Viro } 23958bcb77faSAl Viro 239679714f72SAl Viro /* does lookup, returns the object with parent locked */ 239779714f72SAl Viro struct dentry *kern_path_locked(const char *name, struct path *path) 23985590ff0dSUlrich Drepper { 23995c31b6ceSAl Viro struct filename *filename; 24005c31b6ceSAl Viro struct dentry *d; 2401391172c4SAl Viro struct qstr last; 2402391172c4SAl Viro int type; 240351689104SPaul Moore 24045c31b6ceSAl Viro filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path, 24055c31b6ceSAl Viro &last, &type); 240651689104SPaul Moore if (IS_ERR(filename)) 240751689104SPaul Moore return ERR_CAST(filename); 24085c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) { 2409391172c4SAl Viro path_put(path); 24105c31b6ceSAl Viro putname(filename); 24115c31b6ceSAl Viro return ERR_PTR(-EINVAL); 241279714f72SAl Viro } 24135955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 2414391172c4SAl Viro d = __lookup_hash(&last, path->dentry, 0); 241579714f72SAl Viro if (IS_ERR(d)) { 24165955102cSAl Viro inode_unlock(path->dentry->d_inode); 2417391172c4SAl Viro path_put(path); 241879714f72SAl Viro } 241951689104SPaul Moore putname(filename); 242079714f72SAl Viro return d; 24215590ff0dSUlrich Drepper } 24225590ff0dSUlrich Drepper 2423d1811465SAl Viro int kern_path(const char *name, unsigned int flags, struct path *path) 2424d1811465SAl Viro { 2425abc9f5beSAl Viro return filename_lookup(AT_FDCWD, getname_kernel(name), 2426abc9f5beSAl Viro flags, path, NULL); 2427d1811465SAl Viro } 24284d359507SAl Viro EXPORT_SYMBOL(kern_path); 2429d1811465SAl Viro 243016f18200SJosef 'Jeff' Sipek /** 243116f18200SJosef 'Jeff' Sipek * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair 243216f18200SJosef 'Jeff' Sipek * @dentry: pointer to dentry of the base directory 243316f18200SJosef 'Jeff' Sipek * @mnt: pointer to vfs mount of the base directory 243416f18200SJosef 'Jeff' Sipek * @name: pointer to file name 243516f18200SJosef 'Jeff' Sipek * @flags: lookup flags 2436e0a01249SAl Viro * @path: pointer to struct path to fill 243716f18200SJosef 'Jeff' Sipek */ 243816f18200SJosef 'Jeff' Sipek int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, 243916f18200SJosef 'Jeff' Sipek const char *name, unsigned int flags, 2440e0a01249SAl Viro struct path *path) 244116f18200SJosef 'Jeff' Sipek { 24429ad1aaa6SAl Viro struct path root = {.mnt = mnt, .dentry = dentry}; 24439ad1aaa6SAl Viro /* the first argument of filename_lookup() is ignored with root */ 2444abc9f5beSAl Viro return filename_lookup(AT_FDCWD, getname_kernel(name), 2445abc9f5beSAl Viro flags , path, &root); 244616f18200SJosef 'Jeff' Sipek } 24474d359507SAl Viro EXPORT_SYMBOL(vfs_path_lookup); 244816f18200SJosef 'Jeff' Sipek 2449eead1911SChristoph Hellwig /** 2450a6b91919SRandy Dunlap * lookup_one_len - filesystem helper to lookup single pathname component 2451eead1911SChristoph Hellwig * @name: pathname component to lookup 2452eead1911SChristoph Hellwig * @base: base directory to lookup from 2453eead1911SChristoph Hellwig * @len: maximum length @len should be interpreted to 2454eead1911SChristoph Hellwig * 2455a6b91919SRandy Dunlap * Note that this routine is purely a helper for filesystem usage and should 24569e7543e9SAl Viro * not be called by generic code. 2457bbddca8eSNeilBrown * 2458bbddca8eSNeilBrown * The caller must hold base->i_mutex. 2459eead1911SChristoph Hellwig */ 2460057f6c01SJames Morris struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 2461057f6c01SJames Morris { 2462057f6c01SJames Morris struct qstr this; 24636a96ba54SAl Viro unsigned int c; 2464cda309deSMiklos Szeredi int err; 2465057f6c01SJames Morris 24665955102cSAl Viro WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 24672f9092e1SDavid Woodhouse 24686a96ba54SAl Viro this.name = name; 24696a96ba54SAl Viro this.len = len; 24708387ff25SLinus Torvalds this.hash = full_name_hash(base, name, len); 24716a96ba54SAl Viro if (!len) 24726a96ba54SAl Viro return ERR_PTR(-EACCES); 24736a96ba54SAl Viro 247421d8a15aSAl Viro if (unlikely(name[0] == '.')) { 247521d8a15aSAl Viro if (len < 2 || (len == 2 && name[1] == '.')) 247621d8a15aSAl Viro return ERR_PTR(-EACCES); 247721d8a15aSAl Viro } 247821d8a15aSAl Viro 24796a96ba54SAl Viro while (len--) { 24806a96ba54SAl Viro c = *(const unsigned char *)name++; 24816a96ba54SAl Viro if (c == '/' || c == '\0') 24826a96ba54SAl Viro return ERR_PTR(-EACCES); 24836a96ba54SAl Viro } 24845a202bcdSAl Viro /* 24855a202bcdSAl Viro * See if the low-level filesystem might want 24865a202bcdSAl Viro * to use its own hash.. 24875a202bcdSAl Viro */ 24885a202bcdSAl Viro if (base->d_flags & DCACHE_OP_HASH) { 2489da53be12SLinus Torvalds int err = base->d_op->d_hash(base, &this); 24905a202bcdSAl Viro if (err < 0) 24915a202bcdSAl Viro return ERR_PTR(err); 24925a202bcdSAl Viro } 2493eead1911SChristoph Hellwig 2494cda309deSMiklos Szeredi err = inode_permission(base->d_inode, MAY_EXEC); 2495cda309deSMiklos Szeredi if (err) 2496cda309deSMiklos Szeredi return ERR_PTR(err); 2497cda309deSMiklos Szeredi 249872bd866aSAl Viro return __lookup_hash(&this, base, 0); 2499057f6c01SJames Morris } 25004d359507SAl Viro EXPORT_SYMBOL(lookup_one_len); 2501057f6c01SJames Morris 2502bbddca8eSNeilBrown /** 2503bbddca8eSNeilBrown * lookup_one_len_unlocked - filesystem helper to lookup single pathname component 2504bbddca8eSNeilBrown * @name: pathname component to lookup 2505bbddca8eSNeilBrown * @base: base directory to lookup from 2506bbddca8eSNeilBrown * @len: maximum length @len should be interpreted to 2507bbddca8eSNeilBrown * 2508bbddca8eSNeilBrown * Note that this routine is purely a helper for filesystem usage and should 2509bbddca8eSNeilBrown * not be called by generic code. 2510bbddca8eSNeilBrown * 2511bbddca8eSNeilBrown * Unlike lookup_one_len, it should be called without the parent 2512bbddca8eSNeilBrown * i_mutex held, and will take the i_mutex itself if necessary. 2513bbddca8eSNeilBrown */ 2514bbddca8eSNeilBrown struct dentry *lookup_one_len_unlocked(const char *name, 2515bbddca8eSNeilBrown struct dentry *base, int len) 2516bbddca8eSNeilBrown { 2517bbddca8eSNeilBrown struct qstr this; 2518bbddca8eSNeilBrown unsigned int c; 2519bbddca8eSNeilBrown int err; 252020d00ee8SLinus Torvalds struct dentry *ret; 2521bbddca8eSNeilBrown 2522bbddca8eSNeilBrown this.name = name; 2523bbddca8eSNeilBrown this.len = len; 25248387ff25SLinus Torvalds this.hash = full_name_hash(base, name, len); 2525bbddca8eSNeilBrown if (!len) 2526bbddca8eSNeilBrown return ERR_PTR(-EACCES); 2527bbddca8eSNeilBrown 2528bbddca8eSNeilBrown if (unlikely(name[0] == '.')) { 2529bbddca8eSNeilBrown if (len < 2 || (len == 2 && name[1] == '.')) 2530bbddca8eSNeilBrown return ERR_PTR(-EACCES); 2531bbddca8eSNeilBrown } 2532bbddca8eSNeilBrown 2533bbddca8eSNeilBrown while (len--) { 2534bbddca8eSNeilBrown c = *(const unsigned char *)name++; 2535bbddca8eSNeilBrown if (c == '/' || c == '\0') 2536bbddca8eSNeilBrown return ERR_PTR(-EACCES); 2537bbddca8eSNeilBrown } 2538bbddca8eSNeilBrown /* 2539bbddca8eSNeilBrown * See if the low-level filesystem might want 2540bbddca8eSNeilBrown * to use its own hash.. 2541bbddca8eSNeilBrown */ 2542bbddca8eSNeilBrown if (base->d_flags & DCACHE_OP_HASH) { 2543bbddca8eSNeilBrown int err = base->d_op->d_hash(base, &this); 2544bbddca8eSNeilBrown if (err < 0) 2545bbddca8eSNeilBrown return ERR_PTR(err); 2546bbddca8eSNeilBrown } 2547bbddca8eSNeilBrown 2548bbddca8eSNeilBrown err = inode_permission(base->d_inode, MAY_EXEC); 2549bbddca8eSNeilBrown if (err) 2550bbddca8eSNeilBrown return ERR_PTR(err); 2551bbddca8eSNeilBrown 255220d00ee8SLinus Torvalds ret = lookup_dcache(&this, base, 0); 255320d00ee8SLinus Torvalds if (!ret) 255420d00ee8SLinus Torvalds ret = lookup_slow(&this, base, 0); 255520d00ee8SLinus Torvalds return ret; 2556bbddca8eSNeilBrown } 2557bbddca8eSNeilBrown EXPORT_SYMBOL(lookup_one_len_unlocked); 2558bbddca8eSNeilBrown 2559eedf265aSEric W. Biederman #ifdef CONFIG_UNIX98_PTYS 2560eedf265aSEric W. Biederman int path_pts(struct path *path) 2561eedf265aSEric W. Biederman { 2562eedf265aSEric W. Biederman /* Find something mounted on "pts" in the same directory as 2563eedf265aSEric W. Biederman * the input path. 2564eedf265aSEric W. Biederman */ 2565eedf265aSEric W. Biederman struct dentry *child, *parent; 2566eedf265aSEric W. Biederman struct qstr this; 2567eedf265aSEric W. Biederman int ret; 2568eedf265aSEric W. Biederman 2569eedf265aSEric W. Biederman ret = path_parent_directory(path); 2570eedf265aSEric W. Biederman if (ret) 2571eedf265aSEric W. Biederman return ret; 2572eedf265aSEric W. Biederman 2573eedf265aSEric W. Biederman parent = path->dentry; 2574eedf265aSEric W. Biederman this.name = "pts"; 2575eedf265aSEric W. Biederman this.len = 3; 2576eedf265aSEric W. Biederman child = d_hash_and_lookup(parent, &this); 2577eedf265aSEric W. Biederman if (!child) 2578eedf265aSEric W. Biederman return -ENOENT; 2579eedf265aSEric W. Biederman 2580eedf265aSEric W. Biederman path->dentry = child; 2581eedf265aSEric W. Biederman dput(parent); 2582eedf265aSEric W. Biederman follow_mount(path); 2583eedf265aSEric W. Biederman return 0; 2584eedf265aSEric W. Biederman } 2585eedf265aSEric W. Biederman #endif 2586eedf265aSEric W. Biederman 25871fa1e7f6SAndy Whitcroft int user_path_at_empty(int dfd, const char __user *name, unsigned flags, 25881fa1e7f6SAndy Whitcroft struct path *path, int *empty) 25891da177e4SLinus Torvalds { 2590abc9f5beSAl Viro return filename_lookup(dfd, getname_flags(name, flags, empty), 2591abc9f5beSAl Viro flags, path, NULL); 25921da177e4SLinus Torvalds } 2593b853a161SAl Viro EXPORT_SYMBOL(user_path_at_empty); 25941fa1e7f6SAndy Whitcroft 25958033426eSJeff Layton /** 2596197df04cSAl Viro * mountpoint_last - look up last component for umount 25978033426eSJeff Layton * @nd: pathwalk nameidata - currently pointing at parent directory of "last" 25988033426eSJeff Layton * 25998033426eSJeff Layton * This is a special lookup_last function just for umount. In this case, we 26008033426eSJeff Layton * need to resolve the path without doing any revalidation. 26018033426eSJeff Layton * 26028033426eSJeff Layton * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since 26038033426eSJeff Layton * mountpoints are always pinned in the dcache, their ancestors are too. Thus, 26048033426eSJeff Layton * in almost all cases, this lookup will be served out of the dcache. The only 26058033426eSJeff Layton * cases where it won't are if nd->last refers to a symlink or the path is 26068033426eSJeff Layton * bogus and it doesn't exist. 26078033426eSJeff Layton * 26088033426eSJeff Layton * Returns: 26098033426eSJeff Layton * -error: if there was an error during lookup. This includes -ENOENT if the 2610ba8f4613SAl Viro * lookup found a negative dentry. 26118033426eSJeff Layton * 2612ba8f4613SAl Viro * 0: if we successfully resolved nd->last and found it to not to be a 2613ba8f4613SAl Viro * symlink that needs to be followed. 26148033426eSJeff Layton * 26158033426eSJeff Layton * 1: if we successfully resolved nd->last and found it to be a symlink 2616ba8f4613SAl Viro * that needs to be followed. 26178033426eSJeff Layton */ 26188033426eSJeff Layton static int 2619ba8f4613SAl Viro mountpoint_last(struct nameidata *nd) 26208033426eSJeff Layton { 26218033426eSJeff Layton int error = 0; 26228033426eSJeff Layton struct dentry *dir = nd->path.dentry; 2623ba8f4613SAl Viro struct path path; 26248033426eSJeff Layton 262535759521SAl Viro /* If we're in rcuwalk, drop out of it to handle last component */ 262635759521SAl Viro if (nd->flags & LOOKUP_RCU) { 26274675ac39SAl Viro if (unlazy_walk(nd)) 2628deb106c6SAl Viro return -ECHILD; 26298033426eSJeff Layton } 26308033426eSJeff Layton 26318033426eSJeff Layton nd->flags &= ~LOOKUP_PARENT; 26328033426eSJeff Layton 26338033426eSJeff Layton if (unlikely(nd->last_type != LAST_NORM)) { 26348033426eSJeff Layton error = handle_dots(nd, nd->last_type); 263535759521SAl Viro if (error) 2636deb106c6SAl Viro return error; 2637ba8f4613SAl Viro path.dentry = dget(nd->path.dentry); 2638949a852eSAl Viro } else { 2639ba8f4613SAl Viro path.dentry = d_lookup(dir, &nd->last); 2640ba8f4613SAl Viro if (!path.dentry) { 26418033426eSJeff Layton /* 2642949a852eSAl Viro * No cached dentry. Mounted dentries are pinned in the 2643949a852eSAl Viro * cache, so that means that this dentry is probably 2644949a852eSAl Viro * a symlink or the path doesn't actually point 2645949a852eSAl Viro * to a mounted dentry. 26468033426eSJeff Layton */ 2647ba8f4613SAl Viro path.dentry = lookup_slow(&nd->last, dir, 2648949a852eSAl Viro nd->flags | LOOKUP_NO_REVAL); 2649ba8f4613SAl Viro if (IS_ERR(path.dentry)) 2650ba8f4613SAl Viro return PTR_ERR(path.dentry); 26518033426eSJeff Layton } 2652bcceeebaSDave Jones } 2653ba8f4613SAl Viro if (d_is_negative(path.dentry)) { 2654ba8f4613SAl Viro dput(path.dentry); 2655deb106c6SAl Viro return -ENOENT; 265635759521SAl Viro } 2657ba8f4613SAl Viro path.mnt = nd->path.mnt; 26588f64fb1cSAl Viro return step_into(nd, &path, 0, d_backing_inode(path.dentry), 0); 26598033426eSJeff Layton } 26608033426eSJeff Layton 26618033426eSJeff Layton /** 2662197df04cSAl Viro * path_mountpoint - look up a path to be umounted 26632a78b857SMasanari Iida * @nd: lookup context 26648033426eSJeff Layton * @flags: lookup flags 2665c8a53ee5SAl Viro * @path: pointer to container for result 26668033426eSJeff Layton * 26678033426eSJeff Layton * Look up the given name, but don't attempt to revalidate the last component. 2668606d6fe3SRandy Dunlap * Returns 0 and "path" will be valid on success; Returns error otherwise. 26698033426eSJeff Layton */ 26708033426eSJeff Layton static int 2671c8a53ee5SAl Viro path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path) 26728033426eSJeff Layton { 2673c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2674368ee9baSAl Viro int err; 2675368ee9baSAl Viro if (IS_ERR(s)) 2676368ee9baSAl Viro return PTR_ERR(s); 26773bdba28bSAl Viro while (!(err = link_path_walk(s, nd)) && 2678ba8f4613SAl Viro (err = mountpoint_last(nd)) > 0) { 26793bdba28bSAl Viro s = trailing_symlink(nd); 26803bdba28bSAl Viro if (IS_ERR(s)) { 26813bdba28bSAl Viro err = PTR_ERR(s); 26828033426eSJeff Layton break; 26838033426eSJeff Layton } 26843bdba28bSAl Viro } 2685ba8f4613SAl Viro if (!err) { 2686ba8f4613SAl Viro *path = nd->path; 2687ba8f4613SAl Viro nd->path.mnt = NULL; 2688ba8f4613SAl Viro nd->path.dentry = NULL; 2689ba8f4613SAl Viro follow_mount(path); 2690ba8f4613SAl Viro } 2691deb106c6SAl Viro terminate_walk(nd); 26928033426eSJeff Layton return err; 26938033426eSJeff Layton } 26948033426eSJeff Layton 26952d864651SAl Viro static int 2696668696dcSAl Viro filename_mountpoint(int dfd, struct filename *name, struct path *path, 26972d864651SAl Viro unsigned int flags) 26982d864651SAl Viro { 26999883d185SAl Viro struct nameidata nd; 2700cbaab2dbSAl Viro int error; 2701668696dcSAl Viro if (IS_ERR(name)) 2702668696dcSAl Viro return PTR_ERR(name); 27039883d185SAl Viro set_nameidata(&nd, dfd, name); 2704c8a53ee5SAl Viro error = path_mountpoint(&nd, flags | LOOKUP_RCU, path); 27052d864651SAl Viro if (unlikely(error == -ECHILD)) 2706c8a53ee5SAl Viro error = path_mountpoint(&nd, flags, path); 27072d864651SAl Viro if (unlikely(error == -ESTALE)) 2708c8a53ee5SAl Viro error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path); 27092d864651SAl Viro if (likely(!error)) 2710668696dcSAl Viro audit_inode(name, path->dentry, 0); 27119883d185SAl Viro restore_nameidata(); 2712668696dcSAl Viro putname(name); 27132d864651SAl Viro return error; 27142d864651SAl Viro } 27152d864651SAl Viro 27168033426eSJeff Layton /** 2717197df04cSAl Viro * user_path_mountpoint_at - lookup a path from userland in order to umount it 27188033426eSJeff Layton * @dfd: directory file descriptor 27198033426eSJeff Layton * @name: pathname from userland 27208033426eSJeff Layton * @flags: lookup flags 27218033426eSJeff Layton * @path: pointer to container to hold result 27228033426eSJeff Layton * 27238033426eSJeff Layton * A umount is a special case for path walking. We're not actually interested 27248033426eSJeff Layton * in the inode in this situation, and ESTALE errors can be a problem. We 27258033426eSJeff Layton * simply want track down the dentry and vfsmount attached at the mountpoint 27268033426eSJeff Layton * and avoid revalidating the last component. 27278033426eSJeff Layton * 27288033426eSJeff Layton * Returns 0 and populates "path" on success. 27298033426eSJeff Layton */ 27308033426eSJeff Layton int 2731197df04cSAl Viro user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags, 27328033426eSJeff Layton struct path *path) 27338033426eSJeff Layton { 2734cbaab2dbSAl Viro return filename_mountpoint(dfd, getname(name), path, flags); 27358033426eSJeff Layton } 27368033426eSJeff Layton 27372d864651SAl Viro int 27382d864651SAl Viro kern_path_mountpoint(int dfd, const char *name, struct path *path, 27392d864651SAl Viro unsigned int flags) 27402d864651SAl Viro { 2741cbaab2dbSAl Viro return filename_mountpoint(dfd, getname_kernel(name), path, flags); 27422d864651SAl Viro } 27432d864651SAl Viro EXPORT_SYMBOL(kern_path_mountpoint); 27442d864651SAl Viro 2745cbdf35bcSMiklos Szeredi int __check_sticky(struct inode *dir, struct inode *inode) 27461da177e4SLinus Torvalds { 27478e96e3b7SEric W. Biederman kuid_t fsuid = current_fsuid(); 2748da9592edSDavid Howells 27498e96e3b7SEric W. Biederman if (uid_eq(inode->i_uid, fsuid)) 27501da177e4SLinus Torvalds return 0; 27518e96e3b7SEric W. Biederman if (uid_eq(dir->i_uid, fsuid)) 27521da177e4SLinus Torvalds return 0; 275323adbe12SAndy Lutomirski return !capable_wrt_inode_uidgid(inode, CAP_FOWNER); 27541da177e4SLinus Torvalds } 2755cbdf35bcSMiklos Szeredi EXPORT_SYMBOL(__check_sticky); 27561da177e4SLinus Torvalds 27571da177e4SLinus Torvalds /* 27581da177e4SLinus Torvalds * Check whether we can remove a link victim from directory dir, check 27591da177e4SLinus Torvalds * whether the type of victim is right. 27601da177e4SLinus Torvalds * 1. We can't do it if dir is read-only (done in permission()) 27611da177e4SLinus Torvalds * 2. We should have write and exec permissions on dir 27621da177e4SLinus Torvalds * 3. We can't remove anything from append-only dir 27631da177e4SLinus Torvalds * 4. We can't do anything with immutable dir (done in permission()) 27641da177e4SLinus Torvalds * 5. If the sticky bit on dir is set we should either 27651da177e4SLinus Torvalds * a. be owner of dir, or 27661da177e4SLinus Torvalds * b. be owner of victim, or 27671da177e4SLinus Torvalds * c. have CAP_FOWNER capability 27681da177e4SLinus Torvalds * 6. If the victim is append-only or immutable we can't do antyhing with 27691da177e4SLinus Torvalds * links pointing to it. 27700bd23d09SEric W. Biederman * 7. If the victim has an unknown uid or gid we can't change the inode. 27710bd23d09SEric W. Biederman * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR. 27720bd23d09SEric W. Biederman * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR. 27730bd23d09SEric W. Biederman * 10. We can't remove a root or mountpoint. 27740bd23d09SEric W. Biederman * 11. We don't allow removal of NFS sillyrenamed files; it's handled by 27751da177e4SLinus Torvalds * nfs_async_unlink(). 27761da177e4SLinus Torvalds */ 2777b18825a7SDavid Howells static int may_delete(struct inode *dir, struct dentry *victim, bool isdir) 27781da177e4SLinus Torvalds { 277963afdfc7SDavid Howells struct inode *inode = d_backing_inode(victim); 27801da177e4SLinus Torvalds int error; 27811da177e4SLinus Torvalds 2782b18825a7SDavid Howells if (d_is_negative(victim)) 27831da177e4SLinus Torvalds return -ENOENT; 2784b18825a7SDavid Howells BUG_ON(!inode); 27851da177e4SLinus Torvalds 27861da177e4SLinus Torvalds BUG_ON(victim->d_parent->d_inode != dir); 27874fa6b5ecSJeff Layton audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); 27881da177e4SLinus Torvalds 2789f419a2e3SAl Viro error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 27901da177e4SLinus Torvalds if (error) 27911da177e4SLinus Torvalds return error; 27921da177e4SLinus Torvalds if (IS_APPEND(dir)) 27931da177e4SLinus Torvalds return -EPERM; 2794b18825a7SDavid Howells 2795b18825a7SDavid Howells if (check_sticky(dir, inode) || IS_APPEND(inode) || 27960bd23d09SEric W. Biederman IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || HAS_UNMAPPED_ID(inode)) 27971da177e4SLinus Torvalds return -EPERM; 27981da177e4SLinus Torvalds if (isdir) { 279944b1d530SMiklos Szeredi if (!d_is_dir(victim)) 28001da177e4SLinus Torvalds return -ENOTDIR; 28011da177e4SLinus Torvalds if (IS_ROOT(victim)) 28021da177e4SLinus Torvalds return -EBUSY; 280344b1d530SMiklos Szeredi } else if (d_is_dir(victim)) 28041da177e4SLinus Torvalds return -EISDIR; 28051da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 28061da177e4SLinus Torvalds return -ENOENT; 28071da177e4SLinus Torvalds if (victim->d_flags & DCACHE_NFSFS_RENAMED) 28081da177e4SLinus Torvalds return -EBUSY; 28091da177e4SLinus Torvalds return 0; 28101da177e4SLinus Torvalds } 28111da177e4SLinus Torvalds 28121da177e4SLinus Torvalds /* Check whether we can create an object with dentry child in directory 28131da177e4SLinus Torvalds * dir. 28141da177e4SLinus Torvalds * 1. We can't do it if child already exists (open has special treatment for 28151da177e4SLinus Torvalds * this case, but since we are inlined it's OK) 28161da177e4SLinus Torvalds * 2. We can't do it if dir is read-only (done in permission()) 2817036d5236SEric W. Biederman * 3. We can't do it if the fs can't represent the fsuid or fsgid. 2818036d5236SEric W. Biederman * 4. We should have write and exec permissions on dir 2819036d5236SEric W. Biederman * 5. We can't do it if dir is immutable (done in permission()) 28201da177e4SLinus Torvalds */ 2821a95164d9SMiklos Szeredi static inline int may_create(struct inode *dir, struct dentry *child) 28221da177e4SLinus Torvalds { 2823036d5236SEric W. Biederman struct user_namespace *s_user_ns; 282414e972b4SJeff Layton audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); 28251da177e4SLinus Torvalds if (child->d_inode) 28261da177e4SLinus Torvalds return -EEXIST; 28271da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 28281da177e4SLinus Torvalds return -ENOENT; 2829036d5236SEric W. Biederman s_user_ns = dir->i_sb->s_user_ns; 2830036d5236SEric W. Biederman if (!kuid_has_mapping(s_user_ns, current_fsuid()) || 2831036d5236SEric W. Biederman !kgid_has_mapping(s_user_ns, current_fsgid())) 2832036d5236SEric W. Biederman return -EOVERFLOW; 2833f419a2e3SAl Viro return inode_permission(dir, MAY_WRITE | MAY_EXEC); 28341da177e4SLinus Torvalds } 28351da177e4SLinus Torvalds 28361da177e4SLinus Torvalds /* 28371da177e4SLinus Torvalds * p1 and p2 should be directories on the same fs. 28381da177e4SLinus Torvalds */ 28391da177e4SLinus Torvalds struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 28401da177e4SLinus Torvalds { 28411da177e4SLinus Torvalds struct dentry *p; 28421da177e4SLinus Torvalds 28431da177e4SLinus Torvalds if (p1 == p2) { 28445955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 28451da177e4SLinus Torvalds return NULL; 28461da177e4SLinus Torvalds } 28471da177e4SLinus Torvalds 2848fc64005cSAl Viro mutex_lock(&p1->d_sb->s_vfs_rename_mutex); 28491da177e4SLinus Torvalds 2850e2761a11SOGAWA Hirofumi p = d_ancestor(p2, p1); 2851e2761a11SOGAWA Hirofumi if (p) { 28525955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT); 28535955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_CHILD); 28541da177e4SLinus Torvalds return p; 28551da177e4SLinus Torvalds } 28561da177e4SLinus Torvalds 2857e2761a11SOGAWA Hirofumi p = d_ancestor(p1, p2); 2858e2761a11SOGAWA Hirofumi if (p) { 28595955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 28605955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_CHILD); 28611da177e4SLinus Torvalds return p; 28621da177e4SLinus Torvalds } 28631da177e4SLinus Torvalds 28645955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 28655955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2); 28661da177e4SLinus Torvalds return NULL; 28671da177e4SLinus Torvalds } 28684d359507SAl Viro EXPORT_SYMBOL(lock_rename); 28691da177e4SLinus Torvalds 28701da177e4SLinus Torvalds void unlock_rename(struct dentry *p1, struct dentry *p2) 28711da177e4SLinus Torvalds { 28725955102cSAl Viro inode_unlock(p1->d_inode); 28731da177e4SLinus Torvalds if (p1 != p2) { 28745955102cSAl Viro inode_unlock(p2->d_inode); 2875fc64005cSAl Viro mutex_unlock(&p1->d_sb->s_vfs_rename_mutex); 28761da177e4SLinus Torvalds } 28771da177e4SLinus Torvalds } 28784d359507SAl Viro EXPORT_SYMBOL(unlock_rename); 28791da177e4SLinus Torvalds 28804acdaf27SAl Viro int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, 2881312b63fbSAl Viro bool want_excl) 28821da177e4SLinus Torvalds { 2883a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 28841da177e4SLinus Torvalds if (error) 28851da177e4SLinus Torvalds return error; 28861da177e4SLinus Torvalds 2887acfa4380SAl Viro if (!dir->i_op->create) 28881da177e4SLinus Torvalds return -EACCES; /* shouldn't it be ENOSYS? */ 28891da177e4SLinus Torvalds mode &= S_IALLUGO; 28901da177e4SLinus Torvalds mode |= S_IFREG; 28911da177e4SLinus Torvalds error = security_inode_create(dir, dentry, mode); 28921da177e4SLinus Torvalds if (error) 28931da177e4SLinus Torvalds return error; 2894312b63fbSAl Viro error = dir->i_op->create(dir, dentry, mode, want_excl); 2895a74574aaSStephen Smalley if (!error) 2896f38aa942SAmy Griffis fsnotify_create(dir, dentry); 28971da177e4SLinus Torvalds return error; 28981da177e4SLinus Torvalds } 28994d359507SAl Viro EXPORT_SYMBOL(vfs_create); 29001da177e4SLinus Torvalds 2901a2982cc9SEric W. Biederman bool may_open_dev(const struct path *path) 2902a2982cc9SEric W. Biederman { 2903a2982cc9SEric W. Biederman return !(path->mnt->mnt_flags & MNT_NODEV) && 2904a2982cc9SEric W. Biederman !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV); 2905a2982cc9SEric W. Biederman } 2906a2982cc9SEric W. Biederman 2907f0bb5aafSAl Viro static int may_open(const struct path *path, int acc_mode, int flag) 29081da177e4SLinus Torvalds { 29093fb64190SChristoph Hellwig struct dentry *dentry = path->dentry; 29101da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 29111da177e4SLinus Torvalds int error; 29121da177e4SLinus Torvalds 29131da177e4SLinus Torvalds if (!inode) 29141da177e4SLinus Torvalds return -ENOENT; 29151da177e4SLinus Torvalds 2916c8fe8f30SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 2917c8fe8f30SChristoph Hellwig case S_IFLNK: 29181da177e4SLinus Torvalds return -ELOOP; 2919c8fe8f30SChristoph Hellwig case S_IFDIR: 2920c8fe8f30SChristoph Hellwig if (acc_mode & MAY_WRITE) 29211da177e4SLinus Torvalds return -EISDIR; 2922c8fe8f30SChristoph Hellwig break; 2923c8fe8f30SChristoph Hellwig case S_IFBLK: 2924c8fe8f30SChristoph Hellwig case S_IFCHR: 2925a2982cc9SEric W. Biederman if (!may_open_dev(path)) 29261da177e4SLinus Torvalds return -EACCES; 2927c8fe8f30SChristoph Hellwig /*FALLTHRU*/ 2928c8fe8f30SChristoph Hellwig case S_IFIFO: 2929c8fe8f30SChristoph Hellwig case S_IFSOCK: 29301da177e4SLinus Torvalds flag &= ~O_TRUNC; 2931c8fe8f30SChristoph Hellwig break; 29324a3fd211SDave Hansen } 2933b41572e9SDave Hansen 293462fb4a15SAl Viro error = inode_permission(inode, MAY_OPEN | acc_mode); 2935b41572e9SDave Hansen if (error) 2936b41572e9SDave Hansen return error; 29376146f0d5SMimi Zohar 29381da177e4SLinus Torvalds /* 29391da177e4SLinus Torvalds * An append-only file must be opened in append mode for writing. 29401da177e4SLinus Torvalds */ 29411da177e4SLinus Torvalds if (IS_APPEND(inode)) { 29428737c930SAl Viro if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) 29437715b521SAl Viro return -EPERM; 29441da177e4SLinus Torvalds if (flag & O_TRUNC) 29457715b521SAl Viro return -EPERM; 29461da177e4SLinus Torvalds } 29471da177e4SLinus Torvalds 29481da177e4SLinus Torvalds /* O_NOATIME can only be set by the owner or superuser */ 29492e149670SSerge E. Hallyn if (flag & O_NOATIME && !inode_owner_or_capable(inode)) 29507715b521SAl Viro return -EPERM; 29511da177e4SLinus Torvalds 2952f3c7691eSJ. Bruce Fields return 0; 29537715b521SAl Viro } 29547715b521SAl Viro 2955e1181ee6SJeff Layton static int handle_truncate(struct file *filp) 29567715b521SAl Viro { 2957f0bb5aafSAl Viro const struct path *path = &filp->f_path; 29587715b521SAl Viro struct inode *inode = path->dentry->d_inode; 29597715b521SAl Viro int error = get_write_access(inode); 29601da177e4SLinus Torvalds if (error) 29617715b521SAl Viro return error; 29621da177e4SLinus Torvalds /* 29631da177e4SLinus Torvalds * Refuse to truncate files with mandatory locks held on them. 29641da177e4SLinus Torvalds */ 2965d7a06983SJeff Layton error = locks_verify_locked(filp); 2966be6d3e56SKentaro Takeda if (!error) 2967ea0d3ab2STetsuo Handa error = security_path_truncate(path); 29681da177e4SLinus Torvalds if (!error) { 29697715b521SAl Viro error = do_truncate(path->dentry, 0, 2970d139d7ffSMiklos Szeredi ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, 2971e1181ee6SJeff Layton filp); 29721da177e4SLinus Torvalds } 29731da177e4SLinus Torvalds put_write_access(inode); 2974acd0c935SMimi Zohar return error; 29751da177e4SLinus Torvalds } 29761da177e4SLinus Torvalds 2977d57999e1SDave Hansen static inline int open_to_namei_flags(int flag) 2978d57999e1SDave Hansen { 29798a5e929dSAl Viro if ((flag & O_ACCMODE) == 3) 29808a5e929dSAl Viro flag--; 2981d57999e1SDave Hansen return flag; 2982d57999e1SDave Hansen } 2983d57999e1SDave Hansen 2984d3607752SAl Viro static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode) 2985d18e9008SMiklos Szeredi { 29861328c727SSeth Forshee struct user_namespace *s_user_ns; 2987d18e9008SMiklos Szeredi int error = security_path_mknod(dir, dentry, mode, 0); 2988d18e9008SMiklos Szeredi if (error) 2989d18e9008SMiklos Szeredi return error; 2990d18e9008SMiklos Szeredi 29911328c727SSeth Forshee s_user_ns = dir->dentry->d_sb->s_user_ns; 29921328c727SSeth Forshee if (!kuid_has_mapping(s_user_ns, current_fsuid()) || 29931328c727SSeth Forshee !kgid_has_mapping(s_user_ns, current_fsgid())) 29941328c727SSeth Forshee return -EOVERFLOW; 29951328c727SSeth Forshee 2996d18e9008SMiklos Szeredi error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC); 2997d18e9008SMiklos Szeredi if (error) 2998d18e9008SMiklos Szeredi return error; 2999d18e9008SMiklos Szeredi 3000d18e9008SMiklos Szeredi return security_inode_create(dir->dentry->d_inode, dentry, mode); 3001d18e9008SMiklos Szeredi } 3002d18e9008SMiklos Szeredi 30031acf0af9SDavid Howells /* 30041acf0af9SDavid Howells * Attempt to atomically look up, create and open a file from a negative 30051acf0af9SDavid Howells * dentry. 30061acf0af9SDavid Howells * 30071acf0af9SDavid Howells * Returns 0 if successful. The file will have been created and attached to 30081acf0af9SDavid Howells * @file by the filesystem calling finish_open(). 30091acf0af9SDavid Howells * 30101acf0af9SDavid Howells * Returns 1 if the file was looked up only or didn't need creating. The 30111acf0af9SDavid Howells * caller will need to perform the open themselves. @path will have been 30121acf0af9SDavid Howells * updated to point to the new dentry. This may be negative. 30131acf0af9SDavid Howells * 30141acf0af9SDavid Howells * Returns an error code otherwise. 30151acf0af9SDavid Howells */ 30162675a4ebSAl Viro static int atomic_open(struct nameidata *nd, struct dentry *dentry, 301730d90494SAl Viro struct path *path, struct file *file, 3018015c3bbcSMiklos Szeredi const struct open_flags *op, 30191643b43fSAl Viro int open_flag, umode_t mode, 302047237687SAl Viro int *opened) 3021d18e9008SMiklos Szeredi { 3022d18e9008SMiklos Szeredi struct dentry *const DENTRY_NOT_SET = (void *) -1UL; 3023d18e9008SMiklos Szeredi struct inode *dir = nd->path.dentry->d_inode; 3024d18e9008SMiklos Szeredi int error; 3025d18e9008SMiklos Szeredi 3026384f26e2SAl Viro if (!(~open_flag & (O_EXCL | O_CREAT))) /* both O_EXCL and O_CREAT */ 3027d18e9008SMiklos Szeredi open_flag &= ~O_TRUNC; 3028d18e9008SMiklos Szeredi 3029d18e9008SMiklos Szeredi if (nd->flags & LOOKUP_DIRECTORY) 3030d18e9008SMiklos Szeredi open_flag |= O_DIRECTORY; 3031d18e9008SMiklos Szeredi 303230d90494SAl Viro file->f_path.dentry = DENTRY_NOT_SET; 303330d90494SAl Viro file->f_path.mnt = nd->path.mnt; 30340fb1ea09SAl Viro error = dir->i_op->atomic_open(dir, dentry, file, 30350fb1ea09SAl Viro open_to_namei_flags(open_flag), 30360fb1ea09SAl Viro mode, opened); 30376fbd0714SAl Viro d_lookup_done(dentry); 3038384f26e2SAl Viro if (!error) { 3039d18e9008SMiklos Szeredi /* 3040384f26e2SAl Viro * We didn't have the inode before the open, so check open 3041384f26e2SAl Viro * permission here. 3042d18e9008SMiklos Szeredi */ 3043384f26e2SAl Viro int acc_mode = op->acc_mode; 304403da633aSAl Viro if (*opened & FILE_CREATED) { 304503da633aSAl Viro WARN_ON(!(open_flag & O_CREAT)); 304603da633aSAl Viro fsnotify_create(dir, dentry); 304762fb4a15SAl Viro acc_mode = 0; 304803da633aSAl Viro } 30492675a4ebSAl Viro error = may_open(&file->f_path, acc_mode, open_flag); 3050384f26e2SAl Viro if (WARN_ON(error > 0)) 3051384f26e2SAl Viro error = -EINVAL; 3052384f26e2SAl Viro } else if (error > 0) { 3053d18e9008SMiklos Szeredi if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { 3054d18e9008SMiklos Szeredi error = -EIO; 3055384f26e2SAl Viro } else { 3056d18e9008SMiklos Szeredi if (file->f_path.dentry) { 3057d18e9008SMiklos Szeredi dput(dentry); 3058d18e9008SMiklos Szeredi dentry = file->f_path.dentry; 305910c64ceaSAl Viro } 3060d18e9008SMiklos Szeredi if (*opened & FILE_CREATED) 3061d18e9008SMiklos Szeredi fsnotify_create(dir, dentry); 3062a01e718fSAl Viro if (unlikely(d_is_negative(dentry))) { 3063a01e718fSAl Viro error = -ENOENT; 3064a01e718fSAl Viro } else { 3065d18e9008SMiklos Szeredi path->dentry = dentry; 3066d18e9008SMiklos Szeredi path->mnt = nd->path.mnt; 30672675a4ebSAl Viro return 1; 3068d18e9008SMiklos Szeredi } 3069d18e9008SMiklos Szeredi } 3070a01e718fSAl Viro } 3071d18e9008SMiklos Szeredi dput(dentry); 3072d18e9008SMiklos Szeredi return error; 3073d18e9008SMiklos Szeredi } 3074d18e9008SMiklos Szeredi 307531e6b01fSNick Piggin /* 30761acf0af9SDavid Howells * Look up and maybe create and open the last component. 3077d58ffd35SMiklos Szeredi * 3078d58ffd35SMiklos Szeredi * Must be called with i_mutex held on parent. 3079d58ffd35SMiklos Szeredi * 30801acf0af9SDavid Howells * Returns 0 if the file was successfully atomically created (if necessary) and 30811acf0af9SDavid Howells * opened. In this case the file will be returned attached to @file. 30821acf0af9SDavid Howells * 30831acf0af9SDavid Howells * Returns 1 if the file was not completely opened at this time, though lookups 30841acf0af9SDavid Howells * and creations will have been performed and the dentry returned in @path will 30851acf0af9SDavid Howells * be positive upon return if O_CREAT was specified. If O_CREAT wasn't 30861acf0af9SDavid Howells * specified then a negative dentry may be returned. 30871acf0af9SDavid Howells * 30881acf0af9SDavid Howells * An error code is returned otherwise. 30891acf0af9SDavid Howells * 30901acf0af9SDavid Howells * FILE_CREATE will be set in @*opened if the dentry was created and will be 30911acf0af9SDavid Howells * cleared otherwise prior to returning. 3092d58ffd35SMiklos Szeredi */ 30932675a4ebSAl Viro static int lookup_open(struct nameidata *nd, struct path *path, 309430d90494SAl Viro struct file *file, 3095d58ffd35SMiklos Szeredi const struct open_flags *op, 309664894cf8SAl Viro bool got_write, int *opened) 3097d58ffd35SMiklos Szeredi { 3098d58ffd35SMiklos Szeredi struct dentry *dir = nd->path.dentry; 309954ef4872SMiklos Szeredi struct inode *dir_inode = dir->d_inode; 31001643b43fSAl Viro int open_flag = op->open_flag; 3101d58ffd35SMiklos Szeredi struct dentry *dentry; 31021643b43fSAl Viro int error, create_error = 0; 31031643b43fSAl Viro umode_t mode = op->mode; 31046fbd0714SAl Viro DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 3105d58ffd35SMiklos Szeredi 3106ce8644fcSAl Viro if (unlikely(IS_DEADDIR(dir_inode))) 3107ce8644fcSAl Viro return -ENOENT; 3108d58ffd35SMiklos Szeredi 310947237687SAl Viro *opened &= ~FILE_CREATED; 31106fbd0714SAl Viro dentry = d_lookup(dir, &nd->last); 31116fbd0714SAl Viro for (;;) { 31126fbd0714SAl Viro if (!dentry) { 31136fbd0714SAl Viro dentry = d_alloc_parallel(dir, &nd->last, &wq); 3114d58ffd35SMiklos Szeredi if (IS_ERR(dentry)) 31152675a4ebSAl Viro return PTR_ERR(dentry); 31166fbd0714SAl Viro } 31176fbd0714SAl Viro if (d_in_lookup(dentry)) 31186fbd0714SAl Viro break; 3119d58ffd35SMiklos Szeredi 31206fbd0714SAl Viro error = d_revalidate(dentry, nd->flags); 31216fbd0714SAl Viro if (likely(error > 0)) 31226fbd0714SAl Viro break; 31236fbd0714SAl Viro if (error) 31246fbd0714SAl Viro goto out_dput; 31256fbd0714SAl Viro d_invalidate(dentry); 31266fbd0714SAl Viro dput(dentry); 31276fbd0714SAl Viro dentry = NULL; 31286fbd0714SAl Viro } 31296fbd0714SAl Viro if (dentry->d_inode) { 3130d18e9008SMiklos Szeredi /* Cached positive dentry: will open in f_op->open */ 3131d18e9008SMiklos Szeredi goto out_no_open; 31326c51e513SAl Viro } 3133d18e9008SMiklos Szeredi 31341643b43fSAl Viro /* 31351643b43fSAl Viro * Checking write permission is tricky, bacuse we don't know if we are 31361643b43fSAl Viro * going to actually need it: O_CREAT opens should work as long as the 31371643b43fSAl Viro * file exists. But checking existence breaks atomicity. The trick is 31381643b43fSAl Viro * to check access and if not granted clear O_CREAT from the flags. 31391643b43fSAl Viro * 31401643b43fSAl Viro * Another problem is returing the "right" error value (e.g. for an 31411643b43fSAl Viro * O_EXCL open we want to return EEXIST not EROFS). 31421643b43fSAl Viro */ 31431643b43fSAl Viro if (open_flag & O_CREAT) { 31441643b43fSAl Viro if (!IS_POSIXACL(dir->d_inode)) 31451643b43fSAl Viro mode &= ~current_umask(); 31461643b43fSAl Viro if (unlikely(!got_write)) { 31471643b43fSAl Viro create_error = -EROFS; 31481643b43fSAl Viro open_flag &= ~O_CREAT; 31491643b43fSAl Viro if (open_flag & (O_EXCL | O_TRUNC)) 31501643b43fSAl Viro goto no_open; 31511643b43fSAl Viro /* No side effects, safe to clear O_CREAT */ 31521643b43fSAl Viro } else { 31531643b43fSAl Viro create_error = may_o_create(&nd->path, dentry, mode); 31541643b43fSAl Viro if (create_error) { 31551643b43fSAl Viro open_flag &= ~O_CREAT; 31561643b43fSAl Viro if (open_flag & O_EXCL) 31571643b43fSAl Viro goto no_open; 31581643b43fSAl Viro } 31591643b43fSAl Viro } 31601643b43fSAl Viro } else if ((open_flag & (O_TRUNC|O_WRONLY|O_RDWR)) && 31611643b43fSAl Viro unlikely(!got_write)) { 31621643b43fSAl Viro /* 31631643b43fSAl Viro * No O_CREATE -> atomicity not a requirement -> fall 31641643b43fSAl Viro * back to lookup + open 31651643b43fSAl Viro */ 31661643b43fSAl Viro goto no_open; 3167d18e9008SMiklos Szeredi } 3168d18e9008SMiklos Szeredi 31691643b43fSAl Viro if (dir_inode->i_op->atomic_open) { 31701643b43fSAl Viro error = atomic_open(nd, dentry, path, file, op, open_flag, 31711643b43fSAl Viro mode, opened); 31721643b43fSAl Viro if (unlikely(error == -ENOENT) && create_error) 31731643b43fSAl Viro error = create_error; 31741643b43fSAl Viro return error; 31751643b43fSAl Viro } 317654ef4872SMiklos Szeredi 31771643b43fSAl Viro no_open: 31786fbd0714SAl Viro if (d_in_lookup(dentry)) { 317912fa5e24SAl Viro struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry, 318012fa5e24SAl Viro nd->flags); 31816fbd0714SAl Viro d_lookup_done(dentry); 318212fa5e24SAl Viro if (unlikely(res)) { 318312fa5e24SAl Viro if (IS_ERR(res)) { 318412fa5e24SAl Viro error = PTR_ERR(res); 318512fa5e24SAl Viro goto out_dput; 318612fa5e24SAl Viro } 318712fa5e24SAl Viro dput(dentry); 318812fa5e24SAl Viro dentry = res; 318912fa5e24SAl Viro } 319054ef4872SMiklos Szeredi } 319154ef4872SMiklos Szeredi 3192d58ffd35SMiklos Szeredi /* Negative dentry, just create the file */ 31931643b43fSAl Viro if (!dentry->d_inode && (open_flag & O_CREAT)) { 319447237687SAl Viro *opened |= FILE_CREATED; 3195ce8644fcSAl Viro audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE); 3196ce8644fcSAl Viro if (!dir_inode->i_op->create) { 3197ce8644fcSAl Viro error = -EACCES; 3198d58ffd35SMiklos Szeredi goto out_dput; 319977d660a8SMiklos Szeredi } 3200ce8644fcSAl Viro error = dir_inode->i_op->create(dir_inode, dentry, mode, 32011643b43fSAl Viro open_flag & O_EXCL); 3202d58ffd35SMiklos Szeredi if (error) 3203d58ffd35SMiklos Szeredi goto out_dput; 3204ce8644fcSAl Viro fsnotify_create(dir_inode, dentry); 3205d58ffd35SMiklos Szeredi } 32061643b43fSAl Viro if (unlikely(create_error) && !dentry->d_inode) { 32071643b43fSAl Viro error = create_error; 3208d58ffd35SMiklos Szeredi goto out_dput; 3209d58ffd35SMiklos Szeredi } 3210d18e9008SMiklos Szeredi out_no_open: 3211d58ffd35SMiklos Szeredi path->dentry = dentry; 3212d58ffd35SMiklos Szeredi path->mnt = nd->path.mnt; 32132675a4ebSAl Viro return 1; 3214d58ffd35SMiklos Szeredi 3215d58ffd35SMiklos Szeredi out_dput: 3216d58ffd35SMiklos Szeredi dput(dentry); 32172675a4ebSAl Viro return error; 3218d58ffd35SMiklos Szeredi } 3219d58ffd35SMiklos Szeredi 3220d58ffd35SMiklos Szeredi /* 3221fe2d35ffSAl Viro * Handle the last step of open() 322231e6b01fSNick Piggin */ 3223896475d5SAl Viro static int do_last(struct nameidata *nd, 322430d90494SAl Viro struct file *file, const struct open_flags *op, 322576ae2a5aSAl Viro int *opened) 3226fb1cc555SAl Viro { 3227a1e28038SAl Viro struct dentry *dir = nd->path.dentry; 3228ca344a89SAl Viro int open_flag = op->open_flag; 322977d660a8SMiklos Szeredi bool will_truncate = (open_flag & O_TRUNC) != 0; 323064894cf8SAl Viro bool got_write = false; 3231bcda7652SAl Viro int acc_mode = op->acc_mode; 3232254cf582SAl Viro unsigned seq; 3233a1eb3315SMiklos Szeredi struct inode *inode; 3234896475d5SAl Viro struct path path; 323516c2cd71SAl Viro int error; 3236fb1cc555SAl Viro 3237c3e380b0SAl Viro nd->flags &= ~LOOKUP_PARENT; 3238c3e380b0SAl Viro nd->flags |= op->intent; 3239c3e380b0SAl Viro 3240bc77daa7SAl Viro if (nd->last_type != LAST_NORM) { 3241fe2d35ffSAl Viro error = handle_dots(nd, nd->last_type); 3242deb106c6SAl Viro if (unlikely(error)) 32432675a4ebSAl Viro return error; 3244e83db167SMiklos Szeredi goto finish_open; 32451f36f774SAl Viro } 3246a2c36b45SAl Viro 3247ca344a89SAl Viro if (!(open_flag & O_CREAT)) { 3248fe2d35ffSAl Viro if (nd->last.name[nd->last.len]) 3249fe2d35ffSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 3250fe2d35ffSAl Viro /* we _can_ be in RCU mode here */ 3251254cf582SAl Viro error = lookup_fast(nd, &path, &inode, &seq); 3252e9742b53SAl Viro if (likely(error > 0)) 325371574865SMiklos Szeredi goto finish_lookup; 325471574865SMiklos Szeredi 3255ce57dfc1SAl Viro if (error < 0) 3256deb106c6SAl Viro return error; 3257a1eb3315SMiklos Szeredi 325837d7fffcSMiklos Szeredi BUG_ON(nd->inode != dir->d_inode); 32596583fe22SAl Viro BUG_ON(nd->flags & LOOKUP_RCU); 3260b6183df7SMiklos Szeredi } else { 3261fe2d35ffSAl Viro /* create side of things */ 3262a3fbbde7SAl Viro /* 3263b6183df7SMiklos Szeredi * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED 3264b6183df7SMiklos Szeredi * has been cleared when we got to the last component we are 3265b6183df7SMiklos Szeredi * about to look up 3266a3fbbde7SAl Viro */ 32679f1fafeeSAl Viro error = complete_walk(nd); 3268e8bb73dfSAl Viro if (error) 32692675a4ebSAl Viro return error; 3270fe2d35ffSAl Viro 327176ae2a5aSAl Viro audit_inode(nd->name, dir, LOOKUP_PARENT); 32721f36f774SAl Viro /* trailing slashes? */ 3273deb106c6SAl Viro if (unlikely(nd->last.name[nd->last.len])) 3274deb106c6SAl Viro return -EISDIR; 3275b6183df7SMiklos Szeredi } 32761f36f774SAl Viro 32779cf843e3SAl Viro if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { 327864894cf8SAl Viro error = mnt_want_write(nd->path.mnt); 327964894cf8SAl Viro if (!error) 328064894cf8SAl Viro got_write = true; 328164894cf8SAl Viro /* 328264894cf8SAl Viro * do _not_ fail yet - we might not need that or fail with 328364894cf8SAl Viro * a different error; let lookup_open() decide; we'll be 328464894cf8SAl Viro * dropping this one anyway. 328564894cf8SAl Viro */ 328664894cf8SAl Viro } 32879cf843e3SAl Viro if (open_flag & O_CREAT) 32885955102cSAl Viro inode_lock(dir->d_inode); 32899cf843e3SAl Viro else 32909cf843e3SAl Viro inode_lock_shared(dir->d_inode); 3291896475d5SAl Viro error = lookup_open(nd, &path, file, op, got_write, opened); 32929cf843e3SAl Viro if (open_flag & O_CREAT) 32935955102cSAl Viro inode_unlock(dir->d_inode); 32949cf843e3SAl Viro else 32959cf843e3SAl Viro inode_unlock_shared(dir->d_inode); 3296fb1cc555SAl Viro 32972675a4ebSAl Viro if (error <= 0) { 32982675a4ebSAl Viro if (error) 3299d58ffd35SMiklos Szeredi goto out; 33006c0d46c4SAl Viro 330147237687SAl Viro if ((*opened & FILE_CREATED) || 3302496ad9aaSAl Viro !S_ISREG(file_inode(file)->i_mode)) 330377d660a8SMiklos Szeredi will_truncate = false; 3304d18e9008SMiklos Szeredi 330576ae2a5aSAl Viro audit_inode(nd->name, file->f_path.dentry, 0); 3306d18e9008SMiklos Szeredi goto opened; 3307d18e9008SMiklos Szeredi } 3308d18e9008SMiklos Szeredi 330947237687SAl Viro if (*opened & FILE_CREATED) { 33109b44f1b3SAl Viro /* Don't check for write permission, don't truncate */ 3311ca344a89SAl Viro open_flag &= ~O_TRUNC; 331277d660a8SMiklos Szeredi will_truncate = false; 331362fb4a15SAl Viro acc_mode = 0; 3314896475d5SAl Viro path_to_nameidata(&path, nd); 3315e83db167SMiklos Szeredi goto finish_open_created; 3316fb1cc555SAl Viro } 3317fb1cc555SAl Viro 3318fb1cc555SAl Viro /* 3319d18e9008SMiklos Szeredi * If atomic_open() acquired write access it is dropped now due to 3320d18e9008SMiklos Szeredi * possible mount and symlink following (this might be optimized away if 3321d18e9008SMiklos Szeredi * necessary...) 3322d18e9008SMiklos Szeredi */ 332364894cf8SAl Viro if (got_write) { 3324d18e9008SMiklos Szeredi mnt_drop_write(nd->path.mnt); 332564894cf8SAl Viro got_write = false; 3326d18e9008SMiklos Szeredi } 3327d18e9008SMiklos Szeredi 3328e6ec03a2SAl Viro error = follow_managed(&path, nd); 3329e6ec03a2SAl Viro if (unlikely(error < 0)) 3330e6ec03a2SAl Viro return error; 3331e6ec03a2SAl Viro 33326583fe22SAl Viro if (unlikely(d_is_negative(path.dentry))) { 33336583fe22SAl Viro path_to_nameidata(&path, nd); 33346583fe22SAl Viro return -ENOENT; 33356583fe22SAl Viro } 33366583fe22SAl Viro 33376583fe22SAl Viro /* 33386583fe22SAl Viro * create/update audit record if it already exists. 33396583fe22SAl Viro */ 33406583fe22SAl Viro audit_inode(nd->name, path.dentry, 0); 33416583fe22SAl Viro 3342deb106c6SAl Viro if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) { 3343deb106c6SAl Viro path_to_nameidata(&path, nd); 3344deb106c6SAl Viro return -EEXIST; 3345deb106c6SAl Viro } 3346fb1cc555SAl Viro 3347254cf582SAl Viro seq = 0; /* out of RCU mode, so the value doesn't matter */ 3348d4565649SAl Viro inode = d_backing_inode(path.dentry); 3349766c4cbfSAl Viro finish_lookup: 33508f64fb1cSAl Viro error = step_into(nd, &path, 0, inode, seq); 3351deb106c6SAl Viro if (unlikely(error)) 3352d63ff28fSAl Viro return error; 3353bc77daa7SAl Viro finish_open: 33548f64fb1cSAl Viro /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ 3355a3fbbde7SAl Viro error = complete_walk(nd); 3356fac7d191SAl Viro if (error) 33572675a4ebSAl Viro return error; 335876ae2a5aSAl Viro audit_inode(nd->name, nd->path.dentry, 0); 3359fb1cc555SAl Viro error = -EISDIR; 336044b1d530SMiklos Szeredi if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry)) 33612675a4ebSAl Viro goto out; 3362af2f5542SMiklos Szeredi error = -ENOTDIR; 336344b1d530SMiklos Szeredi if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry)) 33642675a4ebSAl Viro goto out; 33654bbcbd3bSDavid Howells if (!d_is_reg(nd->path.dentry)) 336677d660a8SMiklos Szeredi will_truncate = false; 33676c0d46c4SAl Viro 33680f9d1a10SAl Viro if (will_truncate) { 33690f9d1a10SAl Viro error = mnt_want_write(nd->path.mnt); 33700f9d1a10SAl Viro if (error) 33712675a4ebSAl Viro goto out; 337264894cf8SAl Viro got_write = true; 33730f9d1a10SAl Viro } 3374e83db167SMiklos Szeredi finish_open_created: 3375bcda7652SAl Viro error = may_open(&nd->path, acc_mode, open_flag); 3376ca344a89SAl Viro if (error) 33772675a4ebSAl Viro goto out; 33784aa7c634SMiklos Szeredi BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */ 33794aa7c634SMiklos Szeredi error = vfs_open(&nd->path, file, current_cred()); 3380fac7d191SAl Viro if (error) 3381015c3bbcSMiklos Szeredi goto out; 3382fac7d191SAl Viro *opened |= FILE_OPENED; 3383a8277b9bSMiklos Szeredi opened: 33842675a4ebSAl Viro error = open_check_o_direct(file); 3385fe9ec829SAl Viro if (!error) 33863034a146SDmitry Kasatkin error = ima_file_check(file, op->acc_mode, *opened); 3387fe9ec829SAl Viro if (!error && will_truncate) 33882675a4ebSAl Viro error = handle_truncate(file); 3389ca344a89SAl Viro out: 3390fe9ec829SAl Viro if (unlikely(error) && (*opened & FILE_OPENED)) 3391fe9ec829SAl Viro fput(file); 3392c80567c8SAl Viro if (unlikely(error > 0)) { 3393c80567c8SAl Viro WARN_ON(1); 3394c80567c8SAl Viro error = -EINVAL; 3395c80567c8SAl Viro } 339664894cf8SAl Viro if (got_write) 33970f9d1a10SAl Viro mnt_drop_write(nd->path.mnt); 33982675a4ebSAl Viro return error; 3399fb1cc555SAl Viro } 3400fb1cc555SAl Viro 3401af7bd4dcSAmir Goldstein struct dentry *vfs_tmpfile(struct dentry *dentry, umode_t mode, int open_flag) 3402af7bd4dcSAmir Goldstein { 3403af7bd4dcSAmir Goldstein static const struct qstr name = QSTR_INIT("/", 1); 3404af7bd4dcSAmir Goldstein struct dentry *child = NULL; 3405af7bd4dcSAmir Goldstein struct inode *dir = dentry->d_inode; 3406af7bd4dcSAmir Goldstein struct inode *inode; 3407af7bd4dcSAmir Goldstein int error; 3408af7bd4dcSAmir Goldstein 3409af7bd4dcSAmir Goldstein /* we want directory to be writable */ 3410af7bd4dcSAmir Goldstein error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 3411af7bd4dcSAmir Goldstein if (error) 3412af7bd4dcSAmir Goldstein goto out_err; 3413af7bd4dcSAmir Goldstein error = -EOPNOTSUPP; 3414af7bd4dcSAmir Goldstein if (!dir->i_op->tmpfile) 3415af7bd4dcSAmir Goldstein goto out_err; 3416af7bd4dcSAmir Goldstein error = -ENOMEM; 3417af7bd4dcSAmir Goldstein child = d_alloc(dentry, &name); 3418af7bd4dcSAmir Goldstein if (unlikely(!child)) 3419af7bd4dcSAmir Goldstein goto out_err; 3420af7bd4dcSAmir Goldstein error = dir->i_op->tmpfile(dir, child, mode); 3421af7bd4dcSAmir Goldstein if (error) 3422af7bd4dcSAmir Goldstein goto out_err; 3423af7bd4dcSAmir Goldstein error = -ENOENT; 3424af7bd4dcSAmir Goldstein inode = child->d_inode; 3425af7bd4dcSAmir Goldstein if (unlikely(!inode)) 3426af7bd4dcSAmir Goldstein goto out_err; 3427af7bd4dcSAmir Goldstein if (!(open_flag & O_EXCL)) { 3428af7bd4dcSAmir Goldstein spin_lock(&inode->i_lock); 3429af7bd4dcSAmir Goldstein inode->i_state |= I_LINKABLE; 3430af7bd4dcSAmir Goldstein spin_unlock(&inode->i_lock); 3431af7bd4dcSAmir Goldstein } 3432af7bd4dcSAmir Goldstein return child; 3433af7bd4dcSAmir Goldstein 3434af7bd4dcSAmir Goldstein out_err: 3435af7bd4dcSAmir Goldstein dput(child); 3436af7bd4dcSAmir Goldstein return ERR_PTR(error); 3437af7bd4dcSAmir Goldstein } 3438af7bd4dcSAmir Goldstein EXPORT_SYMBOL(vfs_tmpfile); 3439af7bd4dcSAmir Goldstein 3440c8a53ee5SAl Viro static int do_tmpfile(struct nameidata *nd, unsigned flags, 344160545d0dSAl Viro const struct open_flags *op, 344260545d0dSAl Viro struct file *file, int *opened) 344360545d0dSAl Viro { 3444625b6d10SAl Viro struct dentry *child; 3445625b6d10SAl Viro struct path path; 3446c8a53ee5SAl Viro int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path); 344760545d0dSAl Viro if (unlikely(error)) 344860545d0dSAl Viro return error; 3449625b6d10SAl Viro error = mnt_want_write(path.mnt); 345060545d0dSAl Viro if (unlikely(error)) 345160545d0dSAl Viro goto out; 3452af7bd4dcSAmir Goldstein child = vfs_tmpfile(path.dentry, op->mode, op->open_flag); 3453af7bd4dcSAmir Goldstein error = PTR_ERR(child); 3454af7bd4dcSAmir Goldstein if (unlikely(IS_ERR(child))) 345560545d0dSAl Viro goto out2; 3456625b6d10SAl Viro dput(path.dentry); 3457625b6d10SAl Viro path.dentry = child; 3458c8a53ee5SAl Viro audit_inode(nd->name, child, 0); 345969a91c23SEric Rannaud /* Don't check for other permissions, the inode was just created */ 346062fb4a15SAl Viro error = may_open(&path, 0, op->open_flag); 346160545d0dSAl Viro if (error) 346260545d0dSAl Viro goto out2; 3463625b6d10SAl Viro file->f_path.mnt = path.mnt; 3464625b6d10SAl Viro error = finish_open(file, child, NULL, opened); 346560545d0dSAl Viro if (error) 346660545d0dSAl Viro goto out2; 346760545d0dSAl Viro error = open_check_o_direct(file); 3468af7bd4dcSAmir Goldstein if (error) 346960545d0dSAl Viro fput(file); 347060545d0dSAl Viro out2: 3471625b6d10SAl Viro mnt_drop_write(path.mnt); 347260545d0dSAl Viro out: 3473625b6d10SAl Viro path_put(&path); 347460545d0dSAl Viro return error; 347560545d0dSAl Viro } 347660545d0dSAl Viro 34776ac08709SAl Viro static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file) 34786ac08709SAl Viro { 34796ac08709SAl Viro struct path path; 34806ac08709SAl Viro int error = path_lookupat(nd, flags, &path); 34816ac08709SAl Viro if (!error) { 34826ac08709SAl Viro audit_inode(nd->name, path.dentry, 0); 34836ac08709SAl Viro error = vfs_open(&path, file, current_cred()); 34846ac08709SAl Viro path_put(&path); 34856ac08709SAl Viro } 34866ac08709SAl Viro return error; 34876ac08709SAl Viro } 34886ac08709SAl Viro 3489c8a53ee5SAl Viro static struct file *path_openat(struct nameidata *nd, 3490c8a53ee5SAl Viro const struct open_flags *op, unsigned flags) 34911da177e4SLinus Torvalds { 3492368ee9baSAl Viro const char *s; 349330d90494SAl Viro struct file *file; 349447237687SAl Viro int opened = 0; 349513aab428SAl Viro int error; 349631e6b01fSNick Piggin 349730d90494SAl Viro file = get_empty_filp(); 34981afc99beSAl Viro if (IS_ERR(file)) 34991afc99beSAl Viro return file; 350031e6b01fSNick Piggin 350130d90494SAl Viro file->f_flags = op->open_flag; 350231e6b01fSNick Piggin 3503bb458c64SAl Viro if (unlikely(file->f_flags & __O_TMPFILE)) { 3504c8a53ee5SAl Viro error = do_tmpfile(nd, flags, op, file, &opened); 3505f15133dfSAl Viro goto out2; 350660545d0dSAl Viro } 350760545d0dSAl Viro 35086ac08709SAl Viro if (unlikely(file->f_flags & O_PATH)) { 35096ac08709SAl Viro error = do_o_path(nd, flags, file); 35106ac08709SAl Viro if (!error) 35116ac08709SAl Viro opened |= FILE_OPENED; 35126ac08709SAl Viro goto out2; 35136ac08709SAl Viro } 35146ac08709SAl Viro 3515c8a53ee5SAl Viro s = path_init(nd, flags); 3516368ee9baSAl Viro if (IS_ERR(s)) { 3517368ee9baSAl Viro put_filp(file); 3518368ee9baSAl Viro return ERR_CAST(s); 3519368ee9baSAl Viro } 35203bdba28bSAl Viro while (!(error = link_path_walk(s, nd)) && 352176ae2a5aSAl Viro (error = do_last(nd, file, op, &opened)) > 0) { 352273d049a4SAl Viro nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); 35233bdba28bSAl Viro s = trailing_symlink(nd); 35243bdba28bSAl Viro if (IS_ERR(s)) { 35253bdba28bSAl Viro error = PTR_ERR(s); 35262675a4ebSAl Viro break; 3527806b681cSAl Viro } 35283bdba28bSAl Viro } 3529deb106c6SAl Viro terminate_walk(nd); 3530f15133dfSAl Viro out2: 35312675a4ebSAl Viro if (!(opened & FILE_OPENED)) { 35322675a4ebSAl Viro BUG_ON(!error); 353330d90494SAl Viro put_filp(file); 3534015c3bbcSMiklos Szeredi } 35352675a4ebSAl Viro if (unlikely(error)) { 35362675a4ebSAl Viro if (error == -EOPENSTALE) { 35372675a4ebSAl Viro if (flags & LOOKUP_RCU) 35382675a4ebSAl Viro error = -ECHILD; 35392675a4ebSAl Viro else 35402675a4ebSAl Viro error = -ESTALE; 35412675a4ebSAl Viro } 35422675a4ebSAl Viro file = ERR_PTR(error); 35432675a4ebSAl Viro } 35442675a4ebSAl Viro return file; 3545de459215SKirill Korotaev } 35461da177e4SLinus Torvalds 3547669abf4eSJeff Layton struct file *do_filp_open(int dfd, struct filename *pathname, 3548f9652e10SAl Viro const struct open_flags *op) 354913aab428SAl Viro { 35509883d185SAl Viro struct nameidata nd; 3551f9652e10SAl Viro int flags = op->lookup_flags; 355213aab428SAl Viro struct file *filp; 355313aab428SAl Viro 35549883d185SAl Viro set_nameidata(&nd, dfd, pathname); 3555c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_RCU); 355613aab428SAl Viro if (unlikely(filp == ERR_PTR(-ECHILD))) 3557c8a53ee5SAl Viro filp = path_openat(&nd, op, flags); 355813aab428SAl Viro if (unlikely(filp == ERR_PTR(-ESTALE))) 3559c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_REVAL); 35609883d185SAl Viro restore_nameidata(); 356113aab428SAl Viro return filp; 356213aab428SAl Viro } 356313aab428SAl Viro 356473d049a4SAl Viro struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, 3565f9652e10SAl Viro const char *name, const struct open_flags *op) 356673d049a4SAl Viro { 35679883d185SAl Viro struct nameidata nd; 356873d049a4SAl Viro struct file *file; 356951689104SPaul Moore struct filename *filename; 3570f9652e10SAl Viro int flags = op->lookup_flags | LOOKUP_ROOT; 357173d049a4SAl Viro 357273d049a4SAl Viro nd.root.mnt = mnt; 357373d049a4SAl Viro nd.root.dentry = dentry; 357473d049a4SAl Viro 3575b18825a7SDavid Howells if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN) 357673d049a4SAl Viro return ERR_PTR(-ELOOP); 357773d049a4SAl Viro 357851689104SPaul Moore filename = getname_kernel(name); 3579a1c83681SViresh Kumar if (IS_ERR(filename)) 358051689104SPaul Moore return ERR_CAST(filename); 358151689104SPaul Moore 35829883d185SAl Viro set_nameidata(&nd, -1, filename); 3583c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_RCU); 358473d049a4SAl Viro if (unlikely(file == ERR_PTR(-ECHILD))) 3585c8a53ee5SAl Viro file = path_openat(&nd, op, flags); 358673d049a4SAl Viro if (unlikely(file == ERR_PTR(-ESTALE))) 3587c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_REVAL); 35889883d185SAl Viro restore_nameidata(); 358951689104SPaul Moore putname(filename); 359073d049a4SAl Viro return file; 359173d049a4SAl Viro } 359273d049a4SAl Viro 3593fa14a0b8SAl Viro static struct dentry *filename_create(int dfd, struct filename *name, 35941ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 35951da177e4SLinus Torvalds { 3596c663e5d8SChristoph Hellwig struct dentry *dentry = ERR_PTR(-EEXIST); 3597391172c4SAl Viro struct qstr last; 3598391172c4SAl Viro int type; 3599c30dabfeSJan Kara int err2; 36001ac12b4bSJeff Layton int error; 36011ac12b4bSJeff Layton bool is_dir = (lookup_flags & LOOKUP_DIRECTORY); 36021ac12b4bSJeff Layton 36031ac12b4bSJeff Layton /* 36041ac12b4bSJeff Layton * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any 36051ac12b4bSJeff Layton * other flags passed in are ignored! 36061ac12b4bSJeff Layton */ 36071ac12b4bSJeff Layton lookup_flags &= LOOKUP_REVAL; 36081ac12b4bSJeff Layton 36095c31b6ceSAl Viro name = filename_parentat(dfd, name, lookup_flags, path, &last, &type); 36105c31b6ceSAl Viro if (IS_ERR(name)) 36115c31b6ceSAl Viro return ERR_CAST(name); 36121da177e4SLinus Torvalds 3613c663e5d8SChristoph Hellwig /* 3614c663e5d8SChristoph Hellwig * Yucky last component or no last component at all? 3615c663e5d8SChristoph Hellwig * (foo/., foo/.., /////) 3616c663e5d8SChristoph Hellwig */ 36175c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) 3618ed75e95dSAl Viro goto out; 3619c663e5d8SChristoph Hellwig 3620c30dabfeSJan Kara /* don't fail immediately if it's r/o, at least try to report other errors */ 3621391172c4SAl Viro err2 = mnt_want_write(path->mnt); 3622c663e5d8SChristoph Hellwig /* 3623c663e5d8SChristoph Hellwig * Do the final lookup. 3624c663e5d8SChristoph Hellwig */ 3625391172c4SAl Viro lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL; 36265955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 3627391172c4SAl Viro dentry = __lookup_hash(&last, path->dentry, lookup_flags); 36281da177e4SLinus Torvalds if (IS_ERR(dentry)) 3629a8104a9fSAl Viro goto unlock; 3630c663e5d8SChristoph Hellwig 3631a8104a9fSAl Viro error = -EEXIST; 3632b18825a7SDavid Howells if (d_is_positive(dentry)) 3633a8104a9fSAl Viro goto fail; 3634b18825a7SDavid Howells 3635c663e5d8SChristoph Hellwig /* 3636c663e5d8SChristoph Hellwig * Special case - lookup gave negative, but... we had foo/bar/ 3637c663e5d8SChristoph Hellwig * From the vfs_mknod() POV we just have a negative dentry - 3638c663e5d8SChristoph Hellwig * all is fine. Let's be bastards - you had / on the end, you've 3639c663e5d8SChristoph Hellwig * been asking for (non-existent) directory. -ENOENT for you. 3640c663e5d8SChristoph Hellwig */ 3641391172c4SAl Viro if (unlikely(!is_dir && last.name[last.len])) { 3642a8104a9fSAl Viro error = -ENOENT; 3643ed75e95dSAl Viro goto fail; 3644e9baf6e5SAl Viro } 3645c30dabfeSJan Kara if (unlikely(err2)) { 3646c30dabfeSJan Kara error = err2; 3647a8104a9fSAl Viro goto fail; 3648c30dabfeSJan Kara } 3649181c37b6SAl Viro putname(name); 3650e9baf6e5SAl Viro return dentry; 36511da177e4SLinus Torvalds fail: 3652a8104a9fSAl Viro dput(dentry); 3653a8104a9fSAl Viro dentry = ERR_PTR(error); 3654a8104a9fSAl Viro unlock: 36555955102cSAl Viro inode_unlock(path->dentry->d_inode); 3656c30dabfeSJan Kara if (!err2) 3657391172c4SAl Viro mnt_drop_write(path->mnt); 3658ed75e95dSAl Viro out: 3659391172c4SAl Viro path_put(path); 3660181c37b6SAl Viro putname(name); 3661ed75e95dSAl Viro return dentry; 3662dae6ad8fSAl Viro } 3663fa14a0b8SAl Viro 3664fa14a0b8SAl Viro struct dentry *kern_path_create(int dfd, const char *pathname, 3665fa14a0b8SAl Viro struct path *path, unsigned int lookup_flags) 3666fa14a0b8SAl Viro { 3667181c37b6SAl Viro return filename_create(dfd, getname_kernel(pathname), 3668181c37b6SAl Viro path, lookup_flags); 3669fa14a0b8SAl Viro } 3670dae6ad8fSAl Viro EXPORT_SYMBOL(kern_path_create); 3671dae6ad8fSAl Viro 3672921a1650SAl Viro void done_path_create(struct path *path, struct dentry *dentry) 3673921a1650SAl Viro { 3674921a1650SAl Viro dput(dentry); 36755955102cSAl Viro inode_unlock(path->dentry->d_inode); 3676a8104a9fSAl Viro mnt_drop_write(path->mnt); 3677921a1650SAl Viro path_put(path); 3678921a1650SAl Viro } 3679921a1650SAl Viro EXPORT_SYMBOL(done_path_create); 3680921a1650SAl Viro 3681520ae687SAl Viro inline struct dentry *user_path_create(int dfd, const char __user *pathname, 36821ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 3683dae6ad8fSAl Viro { 3684181c37b6SAl Viro return filename_create(dfd, getname(pathname), path, lookup_flags); 3685dae6ad8fSAl Viro } 3686dae6ad8fSAl Viro EXPORT_SYMBOL(user_path_create); 3687dae6ad8fSAl Viro 36881a67aafbSAl Viro int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) 36891da177e4SLinus Torvalds { 3690a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 36911da177e4SLinus Torvalds 36921da177e4SLinus Torvalds if (error) 36931da177e4SLinus Torvalds return error; 36941da177e4SLinus Torvalds 3695975d6b39SEric W. Biederman if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) 36961da177e4SLinus Torvalds return -EPERM; 36971da177e4SLinus Torvalds 3698acfa4380SAl Viro if (!dir->i_op->mknod) 36991da177e4SLinus Torvalds return -EPERM; 37001da177e4SLinus Torvalds 370108ce5f16SSerge E. Hallyn error = devcgroup_inode_mknod(mode, dev); 370208ce5f16SSerge E. Hallyn if (error) 370308ce5f16SSerge E. Hallyn return error; 370408ce5f16SSerge E. Hallyn 37051da177e4SLinus Torvalds error = security_inode_mknod(dir, dentry, mode, dev); 37061da177e4SLinus Torvalds if (error) 37071da177e4SLinus Torvalds return error; 37081da177e4SLinus Torvalds 37091da177e4SLinus Torvalds error = dir->i_op->mknod(dir, dentry, mode, dev); 3710a74574aaSStephen Smalley if (!error) 3711f38aa942SAmy Griffis fsnotify_create(dir, dentry); 37121da177e4SLinus Torvalds return error; 37131da177e4SLinus Torvalds } 37144d359507SAl Viro EXPORT_SYMBOL(vfs_mknod); 37151da177e4SLinus Torvalds 3716f69aac00SAl Viro static int may_mknod(umode_t mode) 3717463c3197SDave Hansen { 3718463c3197SDave Hansen switch (mode & S_IFMT) { 3719463c3197SDave Hansen case S_IFREG: 3720463c3197SDave Hansen case S_IFCHR: 3721463c3197SDave Hansen case S_IFBLK: 3722463c3197SDave Hansen case S_IFIFO: 3723463c3197SDave Hansen case S_IFSOCK: 3724463c3197SDave Hansen case 0: /* zero mode translates to S_IFREG */ 3725463c3197SDave Hansen return 0; 3726463c3197SDave Hansen case S_IFDIR: 3727463c3197SDave Hansen return -EPERM; 3728463c3197SDave Hansen default: 3729463c3197SDave Hansen return -EINVAL; 3730463c3197SDave Hansen } 3731463c3197SDave Hansen } 3732463c3197SDave Hansen 37338208a22bSAl Viro SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, 37342e4d0924SHeiko Carstens unsigned, dev) 37351da177e4SLinus Torvalds { 37361da177e4SLinus Torvalds struct dentry *dentry; 3737dae6ad8fSAl Viro struct path path; 3738dae6ad8fSAl Viro int error; 3739972567f1SJeff Layton unsigned int lookup_flags = 0; 37401da177e4SLinus Torvalds 37418e4bfca1SAl Viro error = may_mknod(mode); 37428e4bfca1SAl Viro if (error) 37438e4bfca1SAl Viro return error; 3744972567f1SJeff Layton retry: 3745972567f1SJeff Layton dentry = user_path_create(dfd, filename, &path, lookup_flags); 3746dae6ad8fSAl Viro if (IS_ERR(dentry)) 3747dae6ad8fSAl Viro return PTR_ERR(dentry); 37482ad94ae6SAl Viro 3749dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3750ce3b0f8dSAl Viro mode &= ~current_umask(); 3751dae6ad8fSAl Viro error = security_path_mknod(&path, dentry, mode, dev); 3752be6d3e56SKentaro Takeda if (error) 3753a8104a9fSAl Viro goto out; 37541da177e4SLinus Torvalds switch (mode & S_IFMT) { 37551da177e4SLinus Torvalds case 0: case S_IFREG: 3756312b63fbSAl Viro error = vfs_create(path.dentry->d_inode,dentry,mode,true); 375705d1a717SMimi Zohar if (!error) 375805d1a717SMimi Zohar ima_post_path_mknod(dentry); 37591da177e4SLinus Torvalds break; 37601da177e4SLinus Torvalds case S_IFCHR: case S_IFBLK: 3761dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode, 37621da177e4SLinus Torvalds new_decode_dev(dev)); 37631da177e4SLinus Torvalds break; 37641da177e4SLinus Torvalds case S_IFIFO: case S_IFSOCK: 3765dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); 37661da177e4SLinus Torvalds break; 37671da177e4SLinus Torvalds } 3768a8104a9fSAl Viro out: 3769921a1650SAl Viro done_path_create(&path, dentry); 3770972567f1SJeff Layton if (retry_estale(error, lookup_flags)) { 3771972567f1SJeff Layton lookup_flags |= LOOKUP_REVAL; 3772972567f1SJeff Layton goto retry; 3773972567f1SJeff Layton } 37741da177e4SLinus Torvalds return error; 37751da177e4SLinus Torvalds } 37761da177e4SLinus Torvalds 37778208a22bSAl Viro SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) 37785590ff0dSUlrich Drepper { 37795590ff0dSUlrich Drepper return sys_mknodat(AT_FDCWD, filename, mode, dev); 37805590ff0dSUlrich Drepper } 37815590ff0dSUlrich Drepper 378218bb1db3SAl Viro int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 37831da177e4SLinus Torvalds { 3784a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 37858de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 37861da177e4SLinus Torvalds 37871da177e4SLinus Torvalds if (error) 37881da177e4SLinus Torvalds return error; 37891da177e4SLinus Torvalds 3790acfa4380SAl Viro if (!dir->i_op->mkdir) 37911da177e4SLinus Torvalds return -EPERM; 37921da177e4SLinus Torvalds 37931da177e4SLinus Torvalds mode &= (S_IRWXUGO|S_ISVTX); 37941da177e4SLinus Torvalds error = security_inode_mkdir(dir, dentry, mode); 37951da177e4SLinus Torvalds if (error) 37961da177e4SLinus Torvalds return error; 37971da177e4SLinus Torvalds 37988de52778SAl Viro if (max_links && dir->i_nlink >= max_links) 37998de52778SAl Viro return -EMLINK; 38008de52778SAl Viro 38011da177e4SLinus Torvalds error = dir->i_op->mkdir(dir, dentry, mode); 3802a74574aaSStephen Smalley if (!error) 3803f38aa942SAmy Griffis fsnotify_mkdir(dir, dentry); 38041da177e4SLinus Torvalds return error; 38051da177e4SLinus Torvalds } 38064d359507SAl Viro EXPORT_SYMBOL(vfs_mkdir); 38071da177e4SLinus Torvalds 3808a218d0fdSAl Viro SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) 38091da177e4SLinus Torvalds { 38106902d925SDave Hansen struct dentry *dentry; 3811dae6ad8fSAl Viro struct path path; 3812dae6ad8fSAl Viro int error; 3813b76d8b82SJeff Layton unsigned int lookup_flags = LOOKUP_DIRECTORY; 38141da177e4SLinus Torvalds 3815b76d8b82SJeff Layton retry: 3816b76d8b82SJeff Layton dentry = user_path_create(dfd, pathname, &path, lookup_flags); 38176902d925SDave Hansen if (IS_ERR(dentry)) 3818dae6ad8fSAl Viro return PTR_ERR(dentry); 38196902d925SDave Hansen 3820dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3821ce3b0f8dSAl Viro mode &= ~current_umask(); 3822dae6ad8fSAl Viro error = security_path_mkdir(&path, dentry, mode); 3823a8104a9fSAl Viro if (!error) 3824dae6ad8fSAl Viro error = vfs_mkdir(path.dentry->d_inode, dentry, mode); 3825921a1650SAl Viro done_path_create(&path, dentry); 3826b76d8b82SJeff Layton if (retry_estale(error, lookup_flags)) { 3827b76d8b82SJeff Layton lookup_flags |= LOOKUP_REVAL; 3828b76d8b82SJeff Layton goto retry; 3829b76d8b82SJeff Layton } 38301da177e4SLinus Torvalds return error; 38311da177e4SLinus Torvalds } 38321da177e4SLinus Torvalds 3833a218d0fdSAl Viro SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) 38345590ff0dSUlrich Drepper { 38355590ff0dSUlrich Drepper return sys_mkdirat(AT_FDCWD, pathname, mode); 38365590ff0dSUlrich Drepper } 38375590ff0dSUlrich Drepper 38381da177e4SLinus Torvalds int vfs_rmdir(struct inode *dir, struct dentry *dentry) 38391da177e4SLinus Torvalds { 38401da177e4SLinus Torvalds int error = may_delete(dir, dentry, 1); 38411da177e4SLinus Torvalds 38421da177e4SLinus Torvalds if (error) 38431da177e4SLinus Torvalds return error; 38441da177e4SLinus Torvalds 3845acfa4380SAl Viro if (!dir->i_op->rmdir) 38461da177e4SLinus Torvalds return -EPERM; 38471da177e4SLinus Torvalds 38481d2ef590SAl Viro dget(dentry); 38495955102cSAl Viro inode_lock(dentry->d_inode); 3850912dbc15SSage Weil 38511da177e4SLinus Torvalds error = -EBUSY; 38527af1364fSEric W. Biederman if (is_local_mountpoint(dentry)) 3853912dbc15SSage Weil goto out; 3854912dbc15SSage Weil 38551da177e4SLinus Torvalds error = security_inode_rmdir(dir, dentry); 3856912dbc15SSage Weil if (error) 3857912dbc15SSage Weil goto out; 3858912dbc15SSage Weil 38593cebde24SSage Weil shrink_dcache_parent(dentry); 38601da177e4SLinus Torvalds error = dir->i_op->rmdir(dir, dentry); 3861912dbc15SSage Weil if (error) 3862912dbc15SSage Weil goto out; 3863912dbc15SSage Weil 38641da177e4SLinus Torvalds dentry->d_inode->i_flags |= S_DEAD; 3865d83c49f3SAl Viro dont_mount(dentry); 38668ed936b5SEric W. Biederman detach_mounts(dentry); 38671da177e4SLinus Torvalds 3868912dbc15SSage Weil out: 38695955102cSAl Viro inode_unlock(dentry->d_inode); 38701d2ef590SAl Viro dput(dentry); 3871912dbc15SSage Weil if (!error) 3872912dbc15SSage Weil d_delete(dentry); 38731da177e4SLinus Torvalds return error; 38741da177e4SLinus Torvalds } 38754d359507SAl Viro EXPORT_SYMBOL(vfs_rmdir); 38761da177e4SLinus Torvalds 38775590ff0dSUlrich Drepper static long do_rmdir(int dfd, const char __user *pathname) 38781da177e4SLinus Torvalds { 38791da177e4SLinus Torvalds int error = 0; 388091a27b2aSJeff Layton struct filename *name; 38811da177e4SLinus Torvalds struct dentry *dentry; 3882f5beed75SAl Viro struct path path; 3883f5beed75SAl Viro struct qstr last; 3884f5beed75SAl Viro int type; 3885c6ee9206SJeff Layton unsigned int lookup_flags = 0; 3886c6ee9206SJeff Layton retry: 3887c1d4dd27SAl Viro name = filename_parentat(dfd, getname(pathname), lookup_flags, 3888c1d4dd27SAl Viro &path, &last, &type); 388991a27b2aSJeff Layton if (IS_ERR(name)) 389091a27b2aSJeff Layton return PTR_ERR(name); 38911da177e4SLinus Torvalds 3892f5beed75SAl Viro switch (type) { 38931da177e4SLinus Torvalds case LAST_DOTDOT: 38941da177e4SLinus Torvalds error = -ENOTEMPTY; 38951da177e4SLinus Torvalds goto exit1; 38961da177e4SLinus Torvalds case LAST_DOT: 38971da177e4SLinus Torvalds error = -EINVAL; 38981da177e4SLinus Torvalds goto exit1; 38991da177e4SLinus Torvalds case LAST_ROOT: 39001da177e4SLinus Torvalds error = -EBUSY; 39011da177e4SLinus Torvalds goto exit1; 39021da177e4SLinus Torvalds } 39030612d9fbSOGAWA Hirofumi 3904f5beed75SAl Viro error = mnt_want_write(path.mnt); 3905c30dabfeSJan Kara if (error) 3906c30dabfeSJan Kara goto exit1; 39070612d9fbSOGAWA Hirofumi 39085955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 3909f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 39101da177e4SLinus Torvalds error = PTR_ERR(dentry); 39116902d925SDave Hansen if (IS_ERR(dentry)) 39126902d925SDave Hansen goto exit2; 3913e6bc45d6STheodore Ts'o if (!dentry->d_inode) { 3914e6bc45d6STheodore Ts'o error = -ENOENT; 3915e6bc45d6STheodore Ts'o goto exit3; 3916e6bc45d6STheodore Ts'o } 3917f5beed75SAl Viro error = security_path_rmdir(&path, dentry); 3918be6d3e56SKentaro Takeda if (error) 3919c30dabfeSJan Kara goto exit3; 3920f5beed75SAl Viro error = vfs_rmdir(path.dentry->d_inode, dentry); 39210622753bSDave Hansen exit3: 39221da177e4SLinus Torvalds dput(dentry); 39236902d925SDave Hansen exit2: 39245955102cSAl Viro inode_unlock(path.dentry->d_inode); 3925f5beed75SAl Viro mnt_drop_write(path.mnt); 39261da177e4SLinus Torvalds exit1: 3927f5beed75SAl Viro path_put(&path); 39281da177e4SLinus Torvalds putname(name); 3929c6ee9206SJeff Layton if (retry_estale(error, lookup_flags)) { 3930c6ee9206SJeff Layton lookup_flags |= LOOKUP_REVAL; 3931c6ee9206SJeff Layton goto retry; 3932c6ee9206SJeff Layton } 39331da177e4SLinus Torvalds return error; 39341da177e4SLinus Torvalds } 39351da177e4SLinus Torvalds 39363cdad428SHeiko Carstens SYSCALL_DEFINE1(rmdir, const char __user *, pathname) 39375590ff0dSUlrich Drepper { 39385590ff0dSUlrich Drepper return do_rmdir(AT_FDCWD, pathname); 39395590ff0dSUlrich Drepper } 39405590ff0dSUlrich Drepper 3941b21996e3SJ. Bruce Fields /** 3942b21996e3SJ. Bruce Fields * vfs_unlink - unlink a filesystem object 3943b21996e3SJ. Bruce Fields * @dir: parent directory 3944b21996e3SJ. Bruce Fields * @dentry: victim 3945b21996e3SJ. Bruce Fields * @delegated_inode: returns victim inode, if the inode is delegated. 3946b21996e3SJ. Bruce Fields * 3947b21996e3SJ. Bruce Fields * The caller must hold dir->i_mutex. 3948b21996e3SJ. Bruce Fields * 3949b21996e3SJ. Bruce Fields * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and 3950b21996e3SJ. Bruce Fields * return a reference to the inode in delegated_inode. The caller 3951b21996e3SJ. Bruce Fields * should then break the delegation on that inode and retry. Because 3952b21996e3SJ. Bruce Fields * breaking a delegation may take a long time, the caller should drop 3953b21996e3SJ. Bruce Fields * dir->i_mutex before doing so. 3954b21996e3SJ. Bruce Fields * 3955b21996e3SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 3956b21996e3SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 3957b21996e3SJ. Bruce Fields * to be NFS exported. 3958b21996e3SJ. Bruce Fields */ 3959b21996e3SJ. Bruce Fields int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode) 39601da177e4SLinus Torvalds { 39619accbb97SJ. Bruce Fields struct inode *target = dentry->d_inode; 39621da177e4SLinus Torvalds int error = may_delete(dir, dentry, 0); 39631da177e4SLinus Torvalds 39641da177e4SLinus Torvalds if (error) 39651da177e4SLinus Torvalds return error; 39661da177e4SLinus Torvalds 3967acfa4380SAl Viro if (!dir->i_op->unlink) 39681da177e4SLinus Torvalds return -EPERM; 39691da177e4SLinus Torvalds 39705955102cSAl Viro inode_lock(target); 39718ed936b5SEric W. Biederman if (is_local_mountpoint(dentry)) 39721da177e4SLinus Torvalds error = -EBUSY; 39731da177e4SLinus Torvalds else { 39741da177e4SLinus Torvalds error = security_inode_unlink(dir, dentry); 3975bec1052eSAl Viro if (!error) { 39765a14696cSJ. Bruce Fields error = try_break_deleg(target, delegated_inode); 39775a14696cSJ. Bruce Fields if (error) 3978b21996e3SJ. Bruce Fields goto out; 39791da177e4SLinus Torvalds error = dir->i_op->unlink(dir, dentry); 39808ed936b5SEric W. Biederman if (!error) { 3981d83c49f3SAl Viro dont_mount(dentry); 39828ed936b5SEric W. Biederman detach_mounts(dentry); 39838ed936b5SEric W. Biederman } 3984bec1052eSAl Viro } 39851da177e4SLinus Torvalds } 3986b21996e3SJ. Bruce Fields out: 39875955102cSAl Viro inode_unlock(target); 39881da177e4SLinus Torvalds 39891da177e4SLinus Torvalds /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 39901da177e4SLinus Torvalds if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { 39919accbb97SJ. Bruce Fields fsnotify_link_count(target); 39921da177e4SLinus Torvalds d_delete(dentry); 39931da177e4SLinus Torvalds } 39940eeca283SRobert Love 39951da177e4SLinus Torvalds return error; 39961da177e4SLinus Torvalds } 39974d359507SAl Viro EXPORT_SYMBOL(vfs_unlink); 39981da177e4SLinus Torvalds 39991da177e4SLinus Torvalds /* 40001da177e4SLinus Torvalds * Make sure that the actual truncation of the file will occur outside its 40011b1dcc1bSJes Sorensen * directory's i_mutex. Truncate can take a long time if there is a lot of 40021da177e4SLinus Torvalds * writeout happening, and we don't want to prevent access to the directory 40031da177e4SLinus Torvalds * while waiting on the I/O. 40041da177e4SLinus Torvalds */ 40055590ff0dSUlrich Drepper static long do_unlinkat(int dfd, const char __user *pathname) 40061da177e4SLinus Torvalds { 40072ad94ae6SAl Viro int error; 400891a27b2aSJeff Layton struct filename *name; 40091da177e4SLinus Torvalds struct dentry *dentry; 4010f5beed75SAl Viro struct path path; 4011f5beed75SAl Viro struct qstr last; 4012f5beed75SAl Viro int type; 40131da177e4SLinus Torvalds struct inode *inode = NULL; 4014b21996e3SJ. Bruce Fields struct inode *delegated_inode = NULL; 40155d18f813SJeff Layton unsigned int lookup_flags = 0; 40165d18f813SJeff Layton retry: 4017c1d4dd27SAl Viro name = filename_parentat(dfd, getname(pathname), lookup_flags, 4018c1d4dd27SAl Viro &path, &last, &type); 401991a27b2aSJeff Layton if (IS_ERR(name)) 402091a27b2aSJeff Layton return PTR_ERR(name); 40212ad94ae6SAl Viro 40221da177e4SLinus Torvalds error = -EISDIR; 4023f5beed75SAl Viro if (type != LAST_NORM) 40241da177e4SLinus Torvalds goto exit1; 40250612d9fbSOGAWA Hirofumi 4026f5beed75SAl Viro error = mnt_want_write(path.mnt); 4027c30dabfeSJan Kara if (error) 4028c30dabfeSJan Kara goto exit1; 4029b21996e3SJ. Bruce Fields retry_deleg: 40305955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 4031f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 40321da177e4SLinus Torvalds error = PTR_ERR(dentry); 40331da177e4SLinus Torvalds if (!IS_ERR(dentry)) { 40341da177e4SLinus Torvalds /* Why not before? Because we want correct error value */ 4035f5beed75SAl Viro if (last.name[last.len]) 403650338b88STörök Edwin goto slashes; 40371da177e4SLinus Torvalds inode = dentry->d_inode; 4038b18825a7SDavid Howells if (d_is_negative(dentry)) 4039e6bc45d6STheodore Ts'o goto slashes; 40407de9c6eeSAl Viro ihold(inode); 4041f5beed75SAl Viro error = security_path_unlink(&path, dentry); 4042be6d3e56SKentaro Takeda if (error) 4043c30dabfeSJan Kara goto exit2; 4044f5beed75SAl Viro error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode); 40451da177e4SLinus Torvalds exit2: 40461da177e4SLinus Torvalds dput(dentry); 40471da177e4SLinus Torvalds } 40485955102cSAl Viro inode_unlock(path.dentry->d_inode); 40491da177e4SLinus Torvalds if (inode) 40501da177e4SLinus Torvalds iput(inode); /* truncate the inode here */ 4051b21996e3SJ. Bruce Fields inode = NULL; 4052b21996e3SJ. Bruce Fields if (delegated_inode) { 40535a14696cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4054b21996e3SJ. Bruce Fields if (!error) 4055b21996e3SJ. Bruce Fields goto retry_deleg; 4056b21996e3SJ. Bruce Fields } 4057f5beed75SAl Viro mnt_drop_write(path.mnt); 40581da177e4SLinus Torvalds exit1: 4059f5beed75SAl Viro path_put(&path); 40601da177e4SLinus Torvalds putname(name); 40615d18f813SJeff Layton if (retry_estale(error, lookup_flags)) { 40625d18f813SJeff Layton lookup_flags |= LOOKUP_REVAL; 40635d18f813SJeff Layton inode = NULL; 40645d18f813SJeff Layton goto retry; 40655d18f813SJeff Layton } 40661da177e4SLinus Torvalds return error; 40671da177e4SLinus Torvalds 40681da177e4SLinus Torvalds slashes: 4069b18825a7SDavid Howells if (d_is_negative(dentry)) 4070b18825a7SDavid Howells error = -ENOENT; 407144b1d530SMiklos Szeredi else if (d_is_dir(dentry)) 4072b18825a7SDavid Howells error = -EISDIR; 4073b18825a7SDavid Howells else 4074b18825a7SDavid Howells error = -ENOTDIR; 40751da177e4SLinus Torvalds goto exit2; 40761da177e4SLinus Torvalds } 40771da177e4SLinus Torvalds 40782e4d0924SHeiko Carstens SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) 40795590ff0dSUlrich Drepper { 40805590ff0dSUlrich Drepper if ((flag & ~AT_REMOVEDIR) != 0) 40815590ff0dSUlrich Drepper return -EINVAL; 40825590ff0dSUlrich Drepper 40835590ff0dSUlrich Drepper if (flag & AT_REMOVEDIR) 40845590ff0dSUlrich Drepper return do_rmdir(dfd, pathname); 40855590ff0dSUlrich Drepper 40865590ff0dSUlrich Drepper return do_unlinkat(dfd, pathname); 40875590ff0dSUlrich Drepper } 40885590ff0dSUlrich Drepper 40893480b257SHeiko Carstens SYSCALL_DEFINE1(unlink, const char __user *, pathname) 40905590ff0dSUlrich Drepper { 40915590ff0dSUlrich Drepper return do_unlinkat(AT_FDCWD, pathname); 40925590ff0dSUlrich Drepper } 40935590ff0dSUlrich Drepper 4094db2e747bSMiklos Szeredi int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) 40951da177e4SLinus Torvalds { 4096a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 40971da177e4SLinus Torvalds 40981da177e4SLinus Torvalds if (error) 40991da177e4SLinus Torvalds return error; 41001da177e4SLinus Torvalds 4101acfa4380SAl Viro if (!dir->i_op->symlink) 41021da177e4SLinus Torvalds return -EPERM; 41031da177e4SLinus Torvalds 41041da177e4SLinus Torvalds error = security_inode_symlink(dir, dentry, oldname); 41051da177e4SLinus Torvalds if (error) 41061da177e4SLinus Torvalds return error; 41071da177e4SLinus Torvalds 41081da177e4SLinus Torvalds error = dir->i_op->symlink(dir, dentry, oldname); 4109a74574aaSStephen Smalley if (!error) 4110f38aa942SAmy Griffis fsnotify_create(dir, dentry); 41111da177e4SLinus Torvalds return error; 41121da177e4SLinus Torvalds } 41134d359507SAl Viro EXPORT_SYMBOL(vfs_symlink); 41141da177e4SLinus Torvalds 41152e4d0924SHeiko Carstens SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, 41162e4d0924SHeiko Carstens int, newdfd, const char __user *, newname) 41171da177e4SLinus Torvalds { 41182ad94ae6SAl Viro int error; 411991a27b2aSJeff Layton struct filename *from; 41206902d925SDave Hansen struct dentry *dentry; 4121dae6ad8fSAl Viro struct path path; 4122f46d3567SJeff Layton unsigned int lookup_flags = 0; 41231da177e4SLinus Torvalds 41241da177e4SLinus Torvalds from = getname(oldname); 41251da177e4SLinus Torvalds if (IS_ERR(from)) 41261da177e4SLinus Torvalds return PTR_ERR(from); 4127f46d3567SJeff Layton retry: 4128f46d3567SJeff Layton dentry = user_path_create(newdfd, newname, &path, lookup_flags); 41291da177e4SLinus Torvalds error = PTR_ERR(dentry); 41306902d925SDave Hansen if (IS_ERR(dentry)) 4131dae6ad8fSAl Viro goto out_putname; 41326902d925SDave Hansen 413391a27b2aSJeff Layton error = security_path_symlink(&path, dentry, from->name); 4134a8104a9fSAl Viro if (!error) 413591a27b2aSJeff Layton error = vfs_symlink(path.dentry->d_inode, dentry, from->name); 4136921a1650SAl Viro done_path_create(&path, dentry); 4137f46d3567SJeff Layton if (retry_estale(error, lookup_flags)) { 4138f46d3567SJeff Layton lookup_flags |= LOOKUP_REVAL; 4139f46d3567SJeff Layton goto retry; 4140f46d3567SJeff Layton } 41416902d925SDave Hansen out_putname: 41421da177e4SLinus Torvalds putname(from); 41431da177e4SLinus Torvalds return error; 41441da177e4SLinus Torvalds } 41451da177e4SLinus Torvalds 41463480b257SHeiko Carstens SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) 41475590ff0dSUlrich Drepper { 41485590ff0dSUlrich Drepper return sys_symlinkat(oldname, AT_FDCWD, newname); 41495590ff0dSUlrich Drepper } 41505590ff0dSUlrich Drepper 4151146a8595SJ. Bruce Fields /** 4152146a8595SJ. Bruce Fields * vfs_link - create a new link 4153146a8595SJ. Bruce Fields * @old_dentry: object to be linked 4154146a8595SJ. Bruce Fields * @dir: new parent 4155146a8595SJ. Bruce Fields * @new_dentry: where to create the new link 4156146a8595SJ. Bruce Fields * @delegated_inode: returns inode needing a delegation break 4157146a8595SJ. Bruce Fields * 4158146a8595SJ. Bruce Fields * The caller must hold dir->i_mutex 4159146a8595SJ. Bruce Fields * 4160146a8595SJ. Bruce Fields * If vfs_link discovers a delegation on the to-be-linked file in need 4161146a8595SJ. Bruce Fields * of breaking, it will return -EWOULDBLOCK and return a reference to the 4162146a8595SJ. Bruce Fields * inode in delegated_inode. The caller should then break the delegation 4163146a8595SJ. Bruce Fields * and retry. Because breaking a delegation may take a long time, the 4164146a8595SJ. Bruce Fields * caller should drop the i_mutex before doing so. 4165146a8595SJ. Bruce Fields * 4166146a8595SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 4167146a8595SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 4168146a8595SJ. Bruce Fields * to be NFS exported. 4169146a8595SJ. Bruce Fields */ 4170146a8595SJ. Bruce Fields int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode) 41711da177e4SLinus Torvalds { 41721da177e4SLinus Torvalds struct inode *inode = old_dentry->d_inode; 41738de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 41741da177e4SLinus Torvalds int error; 41751da177e4SLinus Torvalds 41761da177e4SLinus Torvalds if (!inode) 41771da177e4SLinus Torvalds return -ENOENT; 41781da177e4SLinus Torvalds 4179a95164d9SMiklos Szeredi error = may_create(dir, new_dentry); 41801da177e4SLinus Torvalds if (error) 41811da177e4SLinus Torvalds return error; 41821da177e4SLinus Torvalds 41831da177e4SLinus Torvalds if (dir->i_sb != inode->i_sb) 41841da177e4SLinus Torvalds return -EXDEV; 41851da177e4SLinus Torvalds 41861da177e4SLinus Torvalds /* 41871da177e4SLinus Torvalds * A link to an append-only or immutable file cannot be created. 41881da177e4SLinus Torvalds */ 41891da177e4SLinus Torvalds if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 41901da177e4SLinus Torvalds return -EPERM; 41910bd23d09SEric W. Biederman /* 41920bd23d09SEric W. Biederman * Updating the link count will likely cause i_uid and i_gid to 41930bd23d09SEric W. Biederman * be writen back improperly if their true value is unknown to 41940bd23d09SEric W. Biederman * the vfs. 41950bd23d09SEric W. Biederman */ 41960bd23d09SEric W. Biederman if (HAS_UNMAPPED_ID(inode)) 41970bd23d09SEric W. Biederman return -EPERM; 4198acfa4380SAl Viro if (!dir->i_op->link) 41991da177e4SLinus Torvalds return -EPERM; 42007e79eedbSTetsuo Handa if (S_ISDIR(inode->i_mode)) 42011da177e4SLinus Torvalds return -EPERM; 42021da177e4SLinus Torvalds 42031da177e4SLinus Torvalds error = security_inode_link(old_dentry, dir, new_dentry); 42041da177e4SLinus Torvalds if (error) 42051da177e4SLinus Torvalds return error; 42061da177e4SLinus Torvalds 42075955102cSAl Viro inode_lock(inode); 4208aae8a97dSAneesh Kumar K.V /* Make sure we don't allow creating hardlink to an unlinked file */ 4209f4e0c30cSAl Viro if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) 4210aae8a97dSAneesh Kumar K.V error = -ENOENT; 42118de52778SAl Viro else if (max_links && inode->i_nlink >= max_links) 42128de52778SAl Viro error = -EMLINK; 4213146a8595SJ. Bruce Fields else { 4214146a8595SJ. Bruce Fields error = try_break_deleg(inode, delegated_inode); 4215146a8595SJ. Bruce Fields if (!error) 42161da177e4SLinus Torvalds error = dir->i_op->link(old_dentry, dir, new_dentry); 4217146a8595SJ. Bruce Fields } 4218f4e0c30cSAl Viro 4219f4e0c30cSAl Viro if (!error && (inode->i_state & I_LINKABLE)) { 4220f4e0c30cSAl Viro spin_lock(&inode->i_lock); 4221f4e0c30cSAl Viro inode->i_state &= ~I_LINKABLE; 4222f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 4223f4e0c30cSAl Viro } 42245955102cSAl Viro inode_unlock(inode); 4225e31e14ecSStephen Smalley if (!error) 42267e79eedbSTetsuo Handa fsnotify_link(dir, inode, new_dentry); 42271da177e4SLinus Torvalds return error; 42281da177e4SLinus Torvalds } 42294d359507SAl Viro EXPORT_SYMBOL(vfs_link); 42301da177e4SLinus Torvalds 42311da177e4SLinus Torvalds /* 42321da177e4SLinus Torvalds * Hardlinks are often used in delicate situations. We avoid 42331da177e4SLinus Torvalds * security-related surprises by not following symlinks on the 42341da177e4SLinus Torvalds * newname. --KAB 42351da177e4SLinus Torvalds * 42361da177e4SLinus Torvalds * We don't follow them on the oldname either to be compatible 42371da177e4SLinus Torvalds * with linux 2.0, and to avoid hard-linking to directories 42381da177e4SLinus Torvalds * and other special files. --ADM 42391da177e4SLinus Torvalds */ 42402e4d0924SHeiko Carstens SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, 42412e4d0924SHeiko Carstens int, newdfd, const char __user *, newname, int, flags) 42421da177e4SLinus Torvalds { 42431da177e4SLinus Torvalds struct dentry *new_dentry; 4244dae6ad8fSAl Viro struct path old_path, new_path; 4245146a8595SJ. Bruce Fields struct inode *delegated_inode = NULL; 424611a7b371SAneesh Kumar K.V int how = 0; 42471da177e4SLinus Torvalds int error; 42481da177e4SLinus Torvalds 424911a7b371SAneesh Kumar K.V if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) 4250c04030e1SUlrich Drepper return -EINVAL; 425111a7b371SAneesh Kumar K.V /* 4252f0cc6ffbSLinus Torvalds * To use null names we require CAP_DAC_READ_SEARCH 4253f0cc6ffbSLinus Torvalds * This ensures that not everyone will be able to create 4254f0cc6ffbSLinus Torvalds * handlink using the passed filedescriptor. 425511a7b371SAneesh Kumar K.V */ 4256f0cc6ffbSLinus Torvalds if (flags & AT_EMPTY_PATH) { 4257f0cc6ffbSLinus Torvalds if (!capable(CAP_DAC_READ_SEARCH)) 4258f0cc6ffbSLinus Torvalds return -ENOENT; 425911a7b371SAneesh Kumar K.V how = LOOKUP_EMPTY; 4260f0cc6ffbSLinus Torvalds } 4261c04030e1SUlrich Drepper 426211a7b371SAneesh Kumar K.V if (flags & AT_SYMLINK_FOLLOW) 426311a7b371SAneesh Kumar K.V how |= LOOKUP_FOLLOW; 4264442e31caSJeff Layton retry: 426511a7b371SAneesh Kumar K.V error = user_path_at(olddfd, oldname, how, &old_path); 42661da177e4SLinus Torvalds if (error) 42672ad94ae6SAl Viro return error; 42682ad94ae6SAl Viro 4269442e31caSJeff Layton new_dentry = user_path_create(newdfd, newname, &new_path, 4270442e31caSJeff Layton (how & LOOKUP_REVAL)); 42711da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 42726902d925SDave Hansen if (IS_ERR(new_dentry)) 4273dae6ad8fSAl Viro goto out; 4274dae6ad8fSAl Viro 4275dae6ad8fSAl Viro error = -EXDEV; 4276dae6ad8fSAl Viro if (old_path.mnt != new_path.mnt) 4277dae6ad8fSAl Viro goto out_dput; 4278800179c9SKees Cook error = may_linkat(&old_path); 4279800179c9SKees Cook if (unlikely(error)) 4280800179c9SKees Cook goto out_dput; 4281dae6ad8fSAl Viro error = security_path_link(old_path.dentry, &new_path, new_dentry); 4282be6d3e56SKentaro Takeda if (error) 4283a8104a9fSAl Viro goto out_dput; 4284146a8595SJ. Bruce Fields error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode); 428575c3f29dSDave Hansen out_dput: 4286921a1650SAl Viro done_path_create(&new_path, new_dentry); 4287146a8595SJ. Bruce Fields if (delegated_inode) { 4288146a8595SJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4289d22e6338SOleg Drokin if (!error) { 4290d22e6338SOleg Drokin path_put(&old_path); 4291146a8595SJ. Bruce Fields goto retry; 4292146a8595SJ. Bruce Fields } 4293d22e6338SOleg Drokin } 4294442e31caSJeff Layton if (retry_estale(error, how)) { 4295d22e6338SOleg Drokin path_put(&old_path); 4296442e31caSJeff Layton how |= LOOKUP_REVAL; 4297442e31caSJeff Layton goto retry; 4298442e31caSJeff Layton } 42991da177e4SLinus Torvalds out: 43002d8f3038SAl Viro path_put(&old_path); 43011da177e4SLinus Torvalds 43021da177e4SLinus Torvalds return error; 43031da177e4SLinus Torvalds } 43041da177e4SLinus Torvalds 43053480b257SHeiko Carstens SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) 43065590ff0dSUlrich Drepper { 4307c04030e1SUlrich Drepper return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 43085590ff0dSUlrich Drepper } 43095590ff0dSUlrich Drepper 4310bc27027aSMiklos Szeredi /** 4311bc27027aSMiklos Szeredi * vfs_rename - rename a filesystem object 4312bc27027aSMiklos Szeredi * @old_dir: parent of source 4313bc27027aSMiklos Szeredi * @old_dentry: source 4314bc27027aSMiklos Szeredi * @new_dir: parent of destination 4315bc27027aSMiklos Szeredi * @new_dentry: destination 4316bc27027aSMiklos Szeredi * @delegated_inode: returns an inode needing a delegation break 4317520c8b16SMiklos Szeredi * @flags: rename flags 4318bc27027aSMiklos Szeredi * 4319bc27027aSMiklos Szeredi * The caller must hold multiple mutexes--see lock_rename()). 4320bc27027aSMiklos Szeredi * 4321bc27027aSMiklos Szeredi * If vfs_rename discovers a delegation in need of breaking at either 4322bc27027aSMiklos Szeredi * the source or destination, it will return -EWOULDBLOCK and return a 4323bc27027aSMiklos Szeredi * reference to the inode in delegated_inode. The caller should then 4324bc27027aSMiklos Szeredi * break the delegation and retry. Because breaking a delegation may 4325bc27027aSMiklos Szeredi * take a long time, the caller should drop all locks before doing 4326bc27027aSMiklos Szeredi * so. 4327bc27027aSMiklos Szeredi * 4328bc27027aSMiklos Szeredi * Alternatively, a caller may pass NULL for delegated_inode. This may 4329bc27027aSMiklos Szeredi * be appropriate for callers that expect the underlying filesystem not 4330bc27027aSMiklos Szeredi * to be NFS exported. 4331bc27027aSMiklos Szeredi * 43321da177e4SLinus Torvalds * The worst of all namespace operations - renaming directory. "Perverted" 43331da177e4SLinus Torvalds * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 43341da177e4SLinus Torvalds * Problems: 4335d03b29a2SJ. Bruce Fields * a) we can get into loop creation. 43361da177e4SLinus Torvalds * b) race potential - two innocent renames can create a loop together. 43371da177e4SLinus Torvalds * That's where 4.4 screws up. Current fix: serialization on 4338a11f3a05SArjan van de Ven * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 43391da177e4SLinus Torvalds * story. 43406cedba89SJ. Bruce Fields * c) we have to lock _four_ objects - parents and victim (if it exists), 43416cedba89SJ. Bruce Fields * and source (if it is not a directory). 43421b1dcc1bSJes Sorensen * And that - after we got ->i_mutex on parents (until then we don't know 43431da177e4SLinus Torvalds * whether the target exists). Solution: try to be smart with locking 43441da177e4SLinus Torvalds * order for inodes. We rely on the fact that tree topology may change 4345a11f3a05SArjan van de Ven * only under ->s_vfs_rename_mutex _and_ that parent of the object we 43461da177e4SLinus Torvalds * move will be locked. Thus we can rank directories by the tree 43471da177e4SLinus Torvalds * (ancestors first) and rank all non-directories after them. 43481da177e4SLinus Torvalds * That works since everybody except rename does "lock parent, lookup, 4349a11f3a05SArjan van de Ven * lock child" and rename is under ->s_vfs_rename_mutex. 43501da177e4SLinus Torvalds * HOWEVER, it relies on the assumption that any object with ->lookup() 43511da177e4SLinus Torvalds * has no more than 1 dentry. If "hybrid" objects will ever appear, 43521da177e4SLinus Torvalds * we'd better make sure that there's no link(2) for them. 4353e4eaac06SSage Weil * d) conversion from fhandle to dentry may come in the wrong moment - when 43541b1dcc1bSJes Sorensen * we are removing the target. Solution: we will have to grab ->i_mutex 43551da177e4SLinus Torvalds * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 4356c41b20e7SAdam Buchbinder * ->i_mutex on parents, which works but leads to some truly excessive 43571da177e4SLinus Torvalds * locking]. 43581da177e4SLinus Torvalds */ 43591da177e4SLinus Torvalds int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, 43608e6d782cSJ. Bruce Fields struct inode *new_dir, struct dentry *new_dentry, 4361520c8b16SMiklos Szeredi struct inode **delegated_inode, unsigned int flags) 43621da177e4SLinus Torvalds { 43631da177e4SLinus Torvalds int error; 4364bc27027aSMiklos Szeredi bool is_dir = d_is_dir(old_dentry); 436559b0df21SEric Paris const unsigned char *old_name; 4366bc27027aSMiklos Szeredi struct inode *source = old_dentry->d_inode; 4367bc27027aSMiklos Szeredi struct inode *target = new_dentry->d_inode; 4368da1ce067SMiklos Szeredi bool new_is_dir = false; 4369da1ce067SMiklos Szeredi unsigned max_links = new_dir->i_sb->s_max_links; 43701da177e4SLinus Torvalds 43718d3e2936SMiklos Szeredi if (source == target) 43721da177e4SLinus Torvalds return 0; 43731da177e4SLinus Torvalds 43741da177e4SLinus Torvalds error = may_delete(old_dir, old_dentry, is_dir); 43751da177e4SLinus Torvalds if (error) 43761da177e4SLinus Torvalds return error; 43771da177e4SLinus Torvalds 4378da1ce067SMiklos Szeredi if (!target) { 4379a95164d9SMiklos Szeredi error = may_create(new_dir, new_dentry); 4380da1ce067SMiklos Szeredi } else { 4381da1ce067SMiklos Szeredi new_is_dir = d_is_dir(new_dentry); 4382da1ce067SMiklos Szeredi 4383da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 43841da177e4SLinus Torvalds error = may_delete(new_dir, new_dentry, is_dir); 4385da1ce067SMiklos Szeredi else 4386da1ce067SMiklos Szeredi error = may_delete(new_dir, new_dentry, new_is_dir); 4387da1ce067SMiklos Szeredi } 43881da177e4SLinus Torvalds if (error) 43891da177e4SLinus Torvalds return error; 43901da177e4SLinus Torvalds 43912773bf00SMiklos Szeredi if (!old_dir->i_op->rename) 43921da177e4SLinus Torvalds return -EPERM; 43931da177e4SLinus Torvalds 4394bc27027aSMiklos Szeredi /* 4395bc27027aSMiklos Szeredi * If we are going to change the parent - check write permissions, 4396bc27027aSMiklos Szeredi * we'll need to flip '..'. 4397bc27027aSMiklos Szeredi */ 4398da1ce067SMiklos Szeredi if (new_dir != old_dir) { 4399da1ce067SMiklos Szeredi if (is_dir) { 4400bc27027aSMiklos Szeredi error = inode_permission(source, MAY_WRITE); 4401bc27027aSMiklos Szeredi if (error) 4402bc27027aSMiklos Szeredi return error; 4403bc27027aSMiklos Szeredi } 4404da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && new_is_dir) { 4405da1ce067SMiklos Szeredi error = inode_permission(target, MAY_WRITE); 4406da1ce067SMiklos Szeredi if (error) 4407da1ce067SMiklos Szeredi return error; 4408da1ce067SMiklos Szeredi } 4409da1ce067SMiklos Szeredi } 44100eeca283SRobert Love 44110b3974ebSMiklos Szeredi error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry, 44120b3974ebSMiklos Szeredi flags); 4413bc27027aSMiklos Szeredi if (error) 4414bc27027aSMiklos Szeredi return error; 4415bc27027aSMiklos Szeredi 4416bc27027aSMiklos Szeredi old_name = fsnotify_oldname_init(old_dentry->d_name.name); 4417bc27027aSMiklos Szeredi dget(new_dentry); 4418da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4419bc27027aSMiklos Szeredi lock_two_nondirectories(source, target); 4420bc27027aSMiklos Szeredi else if (target) 44215955102cSAl Viro inode_lock(target); 4422bc27027aSMiklos Szeredi 4423bc27027aSMiklos Szeredi error = -EBUSY; 44247af1364fSEric W. Biederman if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry)) 4425bc27027aSMiklos Szeredi goto out; 4426bc27027aSMiklos Szeredi 4427da1ce067SMiklos Szeredi if (max_links && new_dir != old_dir) { 4428bc27027aSMiklos Szeredi error = -EMLINK; 4429da1ce067SMiklos Szeredi if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links) 4430bc27027aSMiklos Szeredi goto out; 4431da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir && 4432da1ce067SMiklos Szeredi old_dir->i_nlink >= max_links) 4433da1ce067SMiklos Szeredi goto out; 4434da1ce067SMiklos Szeredi } 4435da1ce067SMiklos Szeredi if (is_dir && !(flags & RENAME_EXCHANGE) && target) 4436bc27027aSMiklos Szeredi shrink_dcache_parent(new_dentry); 4437da1ce067SMiklos Szeredi if (!is_dir) { 4438bc27027aSMiklos Szeredi error = try_break_deleg(source, delegated_inode); 4439bc27027aSMiklos Szeredi if (error) 4440bc27027aSMiklos Szeredi goto out; 4441da1ce067SMiklos Szeredi } 4442da1ce067SMiklos Szeredi if (target && !new_is_dir) { 4443bc27027aSMiklos Szeredi error = try_break_deleg(target, delegated_inode); 4444bc27027aSMiklos Szeredi if (error) 4445bc27027aSMiklos Szeredi goto out; 4446bc27027aSMiklos Szeredi } 4447520c8b16SMiklos Szeredi error = old_dir->i_op->rename(old_dir, old_dentry, 4448520c8b16SMiklos Szeredi new_dir, new_dentry, flags); 4449bc27027aSMiklos Szeredi if (error) 4450bc27027aSMiklos Szeredi goto out; 4451bc27027aSMiklos Szeredi 4452da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE) && target) { 44531da177e4SLinus Torvalds if (is_dir) 4454bc27027aSMiklos Szeredi target->i_flags |= S_DEAD; 4455bc27027aSMiklos Szeredi dont_mount(new_dentry); 44568ed936b5SEric W. Biederman detach_mounts(new_dentry); 4457bc27027aSMiklos Szeredi } 4458da1ce067SMiklos Szeredi if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) { 4459da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 4460bc27027aSMiklos Szeredi d_move(old_dentry, new_dentry); 4461da1ce067SMiklos Szeredi else 4462da1ce067SMiklos Szeredi d_exchange(old_dentry, new_dentry); 4463da1ce067SMiklos Szeredi } 4464bc27027aSMiklos Szeredi out: 4465da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4466bc27027aSMiklos Szeredi unlock_two_nondirectories(source, target); 4467bc27027aSMiklos Szeredi else if (target) 44685955102cSAl Viro inode_unlock(target); 4469bc27027aSMiklos Szeredi dput(new_dentry); 4470da1ce067SMiklos Szeredi if (!error) { 4471123df294SAl Viro fsnotify_move(old_dir, new_dir, old_name, is_dir, 4472da1ce067SMiklos Szeredi !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry); 4473da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4474da1ce067SMiklos Szeredi fsnotify_move(new_dir, old_dir, old_dentry->d_name.name, 4475da1ce067SMiklos Szeredi new_is_dir, NULL, new_dentry); 4476da1ce067SMiklos Szeredi } 4477da1ce067SMiklos Szeredi } 44780eeca283SRobert Love fsnotify_oldname_free(old_name); 44790eeca283SRobert Love 44801da177e4SLinus Torvalds return error; 44811da177e4SLinus Torvalds } 44824d359507SAl Viro EXPORT_SYMBOL(vfs_rename); 44831da177e4SLinus Torvalds 4484520c8b16SMiklos Szeredi SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname, 4485520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname, unsigned int, flags) 44861da177e4SLinus Torvalds { 44871da177e4SLinus Torvalds struct dentry *old_dentry, *new_dentry; 44881da177e4SLinus Torvalds struct dentry *trap; 4489f5beed75SAl Viro struct path old_path, new_path; 4490f5beed75SAl Viro struct qstr old_last, new_last; 4491f5beed75SAl Viro int old_type, new_type; 44928e6d782cSJ. Bruce Fields struct inode *delegated_inode = NULL; 449391a27b2aSJeff Layton struct filename *from; 449491a27b2aSJeff Layton struct filename *to; 4495f5beed75SAl Viro unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET; 4496c6a94284SJeff Layton bool should_retry = false; 44972ad94ae6SAl Viro int error; 4498520c8b16SMiklos Szeredi 44990d7a8555SMiklos Szeredi if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) 4500da1ce067SMiklos Szeredi return -EINVAL; 4501da1ce067SMiklos Szeredi 45020d7a8555SMiklos Szeredi if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) && 45030d7a8555SMiklos Szeredi (flags & RENAME_EXCHANGE)) 4504520c8b16SMiklos Szeredi return -EINVAL; 4505520c8b16SMiklos Szeredi 45060d7a8555SMiklos Szeredi if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD)) 45070d7a8555SMiklos Szeredi return -EPERM; 45080d7a8555SMiklos Szeredi 4509f5beed75SAl Viro if (flags & RENAME_EXCHANGE) 4510f5beed75SAl Viro target_flags = 0; 4511f5beed75SAl Viro 4512c6a94284SJeff Layton retry: 4513c1d4dd27SAl Viro from = filename_parentat(olddfd, getname(oldname), lookup_flags, 4514c1d4dd27SAl Viro &old_path, &old_last, &old_type); 451591a27b2aSJeff Layton if (IS_ERR(from)) { 451691a27b2aSJeff Layton error = PTR_ERR(from); 45171da177e4SLinus Torvalds goto exit; 451891a27b2aSJeff Layton } 45191da177e4SLinus Torvalds 4520c1d4dd27SAl Viro to = filename_parentat(newdfd, getname(newname), lookup_flags, 4521c1d4dd27SAl Viro &new_path, &new_last, &new_type); 452291a27b2aSJeff Layton if (IS_ERR(to)) { 452391a27b2aSJeff Layton error = PTR_ERR(to); 45241da177e4SLinus Torvalds goto exit1; 452591a27b2aSJeff Layton } 45261da177e4SLinus Torvalds 45271da177e4SLinus Torvalds error = -EXDEV; 4528f5beed75SAl Viro if (old_path.mnt != new_path.mnt) 45291da177e4SLinus Torvalds goto exit2; 45301da177e4SLinus Torvalds 45311da177e4SLinus Torvalds error = -EBUSY; 4532f5beed75SAl Viro if (old_type != LAST_NORM) 45331da177e4SLinus Torvalds goto exit2; 45341da177e4SLinus Torvalds 45350a7c3937SMiklos Szeredi if (flags & RENAME_NOREPLACE) 45360a7c3937SMiklos Szeredi error = -EEXIST; 4537f5beed75SAl Viro if (new_type != LAST_NORM) 45381da177e4SLinus Torvalds goto exit2; 45391da177e4SLinus Torvalds 4540f5beed75SAl Viro error = mnt_want_write(old_path.mnt); 4541c30dabfeSJan Kara if (error) 4542c30dabfeSJan Kara goto exit2; 4543c30dabfeSJan Kara 45448e6d782cSJ. Bruce Fields retry_deleg: 4545f5beed75SAl Viro trap = lock_rename(new_path.dentry, old_path.dentry); 45461da177e4SLinus Torvalds 4547f5beed75SAl Viro old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags); 45481da177e4SLinus Torvalds error = PTR_ERR(old_dentry); 45491da177e4SLinus Torvalds if (IS_ERR(old_dentry)) 45501da177e4SLinus Torvalds goto exit3; 45511da177e4SLinus Torvalds /* source must exist */ 45521da177e4SLinus Torvalds error = -ENOENT; 4553b18825a7SDavid Howells if (d_is_negative(old_dentry)) 45541da177e4SLinus Torvalds goto exit4; 4555f5beed75SAl Viro new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags); 45561da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 45571da177e4SLinus Torvalds if (IS_ERR(new_dentry)) 45581da177e4SLinus Torvalds goto exit4; 45590a7c3937SMiklos Szeredi error = -EEXIST; 45600a7c3937SMiklos Szeredi if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry)) 45610a7c3937SMiklos Szeredi goto exit5; 4562da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4563da1ce067SMiklos Szeredi error = -ENOENT; 4564da1ce067SMiklos Szeredi if (d_is_negative(new_dentry)) 4565da1ce067SMiklos Szeredi goto exit5; 4566da1ce067SMiklos Szeredi 4567da1ce067SMiklos Szeredi if (!d_is_dir(new_dentry)) { 4568da1ce067SMiklos Szeredi error = -ENOTDIR; 4569f5beed75SAl Viro if (new_last.name[new_last.len]) 4570da1ce067SMiklos Szeredi goto exit5; 4571da1ce067SMiklos Szeredi } 4572da1ce067SMiklos Szeredi } 45730a7c3937SMiklos Szeredi /* unless the source is a directory trailing slashes give -ENOTDIR */ 45740a7c3937SMiklos Szeredi if (!d_is_dir(old_dentry)) { 45750a7c3937SMiklos Szeredi error = -ENOTDIR; 4576f5beed75SAl Viro if (old_last.name[old_last.len]) 45770a7c3937SMiklos Szeredi goto exit5; 4578f5beed75SAl Viro if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len]) 45790a7c3937SMiklos Szeredi goto exit5; 45800a7c3937SMiklos Szeredi } 45810a7c3937SMiklos Szeredi /* source should not be ancestor of target */ 45820a7c3937SMiklos Szeredi error = -EINVAL; 45830a7c3937SMiklos Szeredi if (old_dentry == trap) 45840a7c3937SMiklos Szeredi goto exit5; 45851da177e4SLinus Torvalds /* target should not be an ancestor of source */ 4586da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 45871da177e4SLinus Torvalds error = -ENOTEMPTY; 45881da177e4SLinus Torvalds if (new_dentry == trap) 45891da177e4SLinus Torvalds goto exit5; 45901da177e4SLinus Torvalds 4591f5beed75SAl Viro error = security_path_rename(&old_path, old_dentry, 4592f5beed75SAl Viro &new_path, new_dentry, flags); 4593be6d3e56SKentaro Takeda if (error) 4594c30dabfeSJan Kara goto exit5; 4595f5beed75SAl Viro error = vfs_rename(old_path.dentry->d_inode, old_dentry, 4596f5beed75SAl Viro new_path.dentry->d_inode, new_dentry, 4597520c8b16SMiklos Szeredi &delegated_inode, flags); 45981da177e4SLinus Torvalds exit5: 45991da177e4SLinus Torvalds dput(new_dentry); 46001da177e4SLinus Torvalds exit4: 46011da177e4SLinus Torvalds dput(old_dentry); 46021da177e4SLinus Torvalds exit3: 4603f5beed75SAl Viro unlock_rename(new_path.dentry, old_path.dentry); 46048e6d782cSJ. Bruce Fields if (delegated_inode) { 46058e6d782cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 46068e6d782cSJ. Bruce Fields if (!error) 46078e6d782cSJ. Bruce Fields goto retry_deleg; 46088e6d782cSJ. Bruce Fields } 4609f5beed75SAl Viro mnt_drop_write(old_path.mnt); 46101da177e4SLinus Torvalds exit2: 4611c6a94284SJeff Layton if (retry_estale(error, lookup_flags)) 4612c6a94284SJeff Layton should_retry = true; 4613f5beed75SAl Viro path_put(&new_path); 46142ad94ae6SAl Viro putname(to); 46151da177e4SLinus Torvalds exit1: 4616f5beed75SAl Viro path_put(&old_path); 46171da177e4SLinus Torvalds putname(from); 4618c6a94284SJeff Layton if (should_retry) { 4619c6a94284SJeff Layton should_retry = false; 4620c6a94284SJeff Layton lookup_flags |= LOOKUP_REVAL; 4621c6a94284SJeff Layton goto retry; 4622c6a94284SJeff Layton } 46232ad94ae6SAl Viro exit: 46241da177e4SLinus Torvalds return error; 46251da177e4SLinus Torvalds } 46261da177e4SLinus Torvalds 4627520c8b16SMiklos Szeredi SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, 4628520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname) 4629520c8b16SMiklos Szeredi { 4630520c8b16SMiklos Szeredi return sys_renameat2(olddfd, oldname, newdfd, newname, 0); 4631520c8b16SMiklos Szeredi } 4632520c8b16SMiklos Szeredi 4633a26eab24SHeiko Carstens SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) 46345590ff0dSUlrich Drepper { 4635520c8b16SMiklos Szeredi return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 46365590ff0dSUlrich Drepper } 46375590ff0dSUlrich Drepper 4638787fb6bcSMiklos Szeredi int vfs_whiteout(struct inode *dir, struct dentry *dentry) 4639787fb6bcSMiklos Szeredi { 4640787fb6bcSMiklos Szeredi int error = may_create(dir, dentry); 4641787fb6bcSMiklos Szeredi if (error) 4642787fb6bcSMiklos Szeredi return error; 4643787fb6bcSMiklos Szeredi 4644787fb6bcSMiklos Szeredi if (!dir->i_op->mknod) 4645787fb6bcSMiklos Szeredi return -EPERM; 4646787fb6bcSMiklos Szeredi 4647787fb6bcSMiklos Szeredi return dir->i_op->mknod(dir, dentry, 4648787fb6bcSMiklos Szeredi S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV); 4649787fb6bcSMiklos Szeredi } 4650787fb6bcSMiklos Szeredi EXPORT_SYMBOL(vfs_whiteout); 4651787fb6bcSMiklos Szeredi 46525d826c84SAl Viro int readlink_copy(char __user *buffer, int buflen, const char *link) 46531da177e4SLinus Torvalds { 46545d826c84SAl Viro int len = PTR_ERR(link); 46551da177e4SLinus Torvalds if (IS_ERR(link)) 46561da177e4SLinus Torvalds goto out; 46571da177e4SLinus Torvalds 46581da177e4SLinus Torvalds len = strlen(link); 46591da177e4SLinus Torvalds if (len > (unsigned) buflen) 46601da177e4SLinus Torvalds len = buflen; 46611da177e4SLinus Torvalds if (copy_to_user(buffer, link, len)) 46621da177e4SLinus Torvalds len = -EFAULT; 46631da177e4SLinus Torvalds out: 46641da177e4SLinus Torvalds return len; 46651da177e4SLinus Torvalds } 46661da177e4SLinus Torvalds 46671da177e4SLinus Torvalds /* 46681da177e4SLinus Torvalds * A helper for ->readlink(). This should be used *ONLY* for symlinks that 46696b255391SAl Viro * have ->get_link() not calling nd_jump_link(). Using (or not using) it 46706b255391SAl Viro * for any given inode is up to filesystem. 46711da177e4SLinus Torvalds */ 4672d16744ecSMiklos Szeredi static int generic_readlink(struct dentry *dentry, char __user *buffer, 4673d16744ecSMiklos Szeredi int buflen) 46741da177e4SLinus Torvalds { 4675fceef393SAl Viro DEFINE_DELAYED_CALL(done); 46765f2c4179SAl Viro struct inode *inode = d_inode(dentry); 46775f2c4179SAl Viro const char *link = inode->i_link; 4678694a1764SMarcin Slusarz int res; 4679cc314eefSLinus Torvalds 4680d4dee48bSAl Viro if (!link) { 4681fceef393SAl Viro link = inode->i_op->get_link(dentry, inode, &done); 4682680baacbSAl Viro if (IS_ERR(link)) 4683680baacbSAl Viro return PTR_ERR(link); 4684d4dee48bSAl Viro } 4685680baacbSAl Viro res = readlink_copy(buffer, buflen, link); 4686fceef393SAl Viro do_delayed_call(&done); 4687694a1764SMarcin Slusarz return res; 46881da177e4SLinus Torvalds } 46891da177e4SLinus Torvalds 4690d60874cdSMiklos Szeredi /** 4691fd4a0edfSMiklos Szeredi * vfs_readlink - copy symlink body into userspace buffer 4692fd4a0edfSMiklos Szeredi * @dentry: dentry on which to get symbolic link 4693fd4a0edfSMiklos Szeredi * @buffer: user memory pointer 4694fd4a0edfSMiklos Szeredi * @buflen: size of buffer 4695fd4a0edfSMiklos Szeredi * 4696fd4a0edfSMiklos Szeredi * Does not touch atime. That's up to the caller if necessary 4697fd4a0edfSMiklos Szeredi * 4698fd4a0edfSMiklos Szeredi * Does not call security hook. 4699fd4a0edfSMiklos Szeredi */ 4700fd4a0edfSMiklos Szeredi int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen) 4701fd4a0edfSMiklos Szeredi { 4702fd4a0edfSMiklos Szeredi struct inode *inode = d_inode(dentry); 4703fd4a0edfSMiklos Szeredi 470476fca90eSMiklos Szeredi if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) { 470576fca90eSMiklos Szeredi if (unlikely(inode->i_op->readlink)) 470676fca90eSMiklos Szeredi return inode->i_op->readlink(dentry, buffer, buflen); 470776fca90eSMiklos Szeredi 470876fca90eSMiklos Szeredi if (!d_is_symlink(dentry)) 4709fd4a0edfSMiklos Szeredi return -EINVAL; 4710fd4a0edfSMiklos Szeredi 471176fca90eSMiklos Szeredi spin_lock(&inode->i_lock); 471276fca90eSMiklos Szeredi inode->i_opflags |= IOP_DEFAULT_READLINK; 471376fca90eSMiklos Szeredi spin_unlock(&inode->i_lock); 471476fca90eSMiklos Szeredi } 471576fca90eSMiklos Szeredi 471676fca90eSMiklos Szeredi return generic_readlink(dentry, buffer, buflen); 4717fd4a0edfSMiklos Szeredi } 4718fd4a0edfSMiklos Szeredi EXPORT_SYMBOL(vfs_readlink); 47191da177e4SLinus Torvalds 4720d60874cdSMiklos Szeredi /** 4721d60874cdSMiklos Szeredi * vfs_get_link - get symlink body 4722d60874cdSMiklos Szeredi * @dentry: dentry on which to get symbolic link 4723d60874cdSMiklos Szeredi * @done: caller needs to free returned data with this 4724d60874cdSMiklos Szeredi * 4725d60874cdSMiklos Szeredi * Calls security hook and i_op->get_link() on the supplied inode. 4726d60874cdSMiklos Szeredi * 4727d60874cdSMiklos Szeredi * It does not touch atime. That's up to the caller if necessary. 4728d60874cdSMiklos Szeredi * 4729d60874cdSMiklos Szeredi * Does not work on "special" symlinks like /proc/$$/fd/N 4730d60874cdSMiklos Szeredi */ 4731d60874cdSMiklos Szeredi const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done) 4732d60874cdSMiklos Szeredi { 4733d60874cdSMiklos Szeredi const char *res = ERR_PTR(-EINVAL); 4734d60874cdSMiklos Szeredi struct inode *inode = d_inode(dentry); 4735d60874cdSMiklos Szeredi 4736d60874cdSMiklos Szeredi if (d_is_symlink(dentry)) { 4737d60874cdSMiklos Szeredi res = ERR_PTR(security_inode_readlink(dentry)); 4738d60874cdSMiklos Szeredi if (!res) 4739d60874cdSMiklos Szeredi res = inode->i_op->get_link(dentry, inode, done); 4740d60874cdSMiklos Szeredi } 4741d60874cdSMiklos Szeredi return res; 4742d60874cdSMiklos Szeredi } 4743d60874cdSMiklos Szeredi EXPORT_SYMBOL(vfs_get_link); 4744d60874cdSMiklos Szeredi 47451da177e4SLinus Torvalds /* get the link contents into pagecache */ 47466b255391SAl Viro const char *page_get_link(struct dentry *dentry, struct inode *inode, 4747fceef393SAl Viro struct delayed_call *callback) 47481da177e4SLinus Torvalds { 4749ebd09abbSDuane Griffin char *kaddr; 47501da177e4SLinus Torvalds struct page *page; 47516b255391SAl Viro struct address_space *mapping = inode->i_mapping; 47526b255391SAl Viro 4753d3883d4fSAl Viro if (!dentry) { 4754d3883d4fSAl Viro page = find_get_page(mapping, 0); 4755d3883d4fSAl Viro if (!page) 47566b255391SAl Viro return ERR_PTR(-ECHILD); 4757d3883d4fSAl Viro if (!PageUptodate(page)) { 4758d3883d4fSAl Viro put_page(page); 4759d3883d4fSAl Viro return ERR_PTR(-ECHILD); 4760d3883d4fSAl Viro } 4761d3883d4fSAl Viro } else { 4762090d2b18SPekka Enberg page = read_mapping_page(mapping, 0, NULL); 47631da177e4SLinus Torvalds if (IS_ERR(page)) 47646fe6900eSNick Piggin return (char*)page; 4765d3883d4fSAl Viro } 4766fceef393SAl Viro set_delayed_call(callback, page_put_link, page); 476721fc61c7SAl Viro BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM); 476821fc61c7SAl Viro kaddr = page_address(page); 47696b255391SAl Viro nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1); 4770ebd09abbSDuane Griffin return kaddr; 47711da177e4SLinus Torvalds } 47721da177e4SLinus Torvalds 47736b255391SAl Viro EXPORT_SYMBOL(page_get_link); 47741da177e4SLinus Torvalds 4775fceef393SAl Viro void page_put_link(void *arg) 47761da177e4SLinus Torvalds { 4777fceef393SAl Viro put_page(arg); 47781da177e4SLinus Torvalds } 47794d359507SAl Viro EXPORT_SYMBOL(page_put_link); 47801da177e4SLinus Torvalds 4781aa80deabSAl Viro int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 4782aa80deabSAl Viro { 4783fceef393SAl Viro DEFINE_DELAYED_CALL(done); 47846b255391SAl Viro int res = readlink_copy(buffer, buflen, 47856b255391SAl Viro page_get_link(dentry, d_inode(dentry), 4786fceef393SAl Viro &done)); 4787fceef393SAl Viro do_delayed_call(&done); 4788aa80deabSAl Viro return res; 4789aa80deabSAl Viro } 4790aa80deabSAl Viro EXPORT_SYMBOL(page_readlink); 4791aa80deabSAl Viro 479254566b2cSNick Piggin /* 479354566b2cSNick Piggin * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS 479454566b2cSNick Piggin */ 479554566b2cSNick Piggin int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) 47961da177e4SLinus Torvalds { 47971da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 47980adb25d2SKirill Korotaev struct page *page; 4799afddba49SNick Piggin void *fsdata; 4800beb497abSDmitriy Monakhov int err; 480154566b2cSNick Piggin unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE; 480254566b2cSNick Piggin if (nofs) 480354566b2cSNick Piggin flags |= AOP_FLAG_NOFS; 48041da177e4SLinus Torvalds 48057e53cac4SNeilBrown retry: 4806afddba49SNick Piggin err = pagecache_write_begin(NULL, mapping, 0, len-1, 480754566b2cSNick Piggin flags, &page, &fsdata); 48081da177e4SLinus Torvalds if (err) 4809afddba49SNick Piggin goto fail; 4810afddba49SNick Piggin 481121fc61c7SAl Viro memcpy(page_address(page), symname, len-1); 4812afddba49SNick Piggin 4813afddba49SNick Piggin err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, 4814afddba49SNick Piggin page, fsdata); 48151da177e4SLinus Torvalds if (err < 0) 48161da177e4SLinus Torvalds goto fail; 4817afddba49SNick Piggin if (err < len-1) 4818afddba49SNick Piggin goto retry; 4819afddba49SNick Piggin 48201da177e4SLinus Torvalds mark_inode_dirty(inode); 48211da177e4SLinus Torvalds return 0; 48221da177e4SLinus Torvalds fail: 48231da177e4SLinus Torvalds return err; 48241da177e4SLinus Torvalds } 48254d359507SAl Viro EXPORT_SYMBOL(__page_symlink); 48261da177e4SLinus Torvalds 48270adb25d2SKirill Korotaev int page_symlink(struct inode *inode, const char *symname, int len) 48280adb25d2SKirill Korotaev { 48290adb25d2SKirill Korotaev return __page_symlink(inode, symname, len, 4830c62d2555SMichal Hocko !mapping_gfp_constraint(inode->i_mapping, __GFP_FS)); 48310adb25d2SKirill Korotaev } 48324d359507SAl Viro EXPORT_SYMBOL(page_symlink); 48330adb25d2SKirill Korotaev 483492e1d5beSArjan van de Ven const struct inode_operations page_symlink_inode_operations = { 48356b255391SAl Viro .get_link = page_get_link, 48361da177e4SLinus Torvalds }; 48371da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink_inode_operations); 4838