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 22559b4a9b14SAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2256c8a53ee5SAl Viro static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path) 22579b4a9b14SAl Viro { 2258c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2259bd92d7feSAl Viro int err; 226031e6b01fSNick Piggin 2261368ee9baSAl Viro if (IS_ERR(s)) 2262368ee9baSAl Viro return PTR_ERR(s); 22633bdba28bSAl Viro while (!(err = link_path_walk(s, nd)) 22643bdba28bSAl Viro && ((err = lookup_last(nd)) > 0)) { 22653bdba28bSAl Viro s = trailing_symlink(nd); 22663bdba28bSAl Viro if (IS_ERR(s)) { 22673bdba28bSAl Viro err = PTR_ERR(s); 22686d7b5aaeSAl Viro break; 2269bd92d7feSAl Viro } 2270bd92d7feSAl Viro } 22719f1fafeeSAl Viro if (!err) 22729f1fafeeSAl Viro err = complete_walk(nd); 2273bd92d7feSAl Viro 2274deb106c6SAl Viro if (!err && nd->flags & LOOKUP_DIRECTORY) 2275deb106c6SAl Viro if (!d_can_lookup(nd->path.dentry)) 2276bd23a539SAl Viro err = -ENOTDIR; 2277625b6d10SAl Viro if (!err) { 2278625b6d10SAl Viro *path = nd->path; 2279625b6d10SAl Viro nd->path.mnt = NULL; 2280625b6d10SAl Viro nd->path.dentry = NULL; 2281625b6d10SAl Viro } 2282deb106c6SAl Viro terminate_walk(nd); 2283bd92d7feSAl Viro return err; 228431e6b01fSNick Piggin } 228531e6b01fSNick Piggin 2286625b6d10SAl Viro static int filename_lookup(int dfd, struct filename *name, unsigned flags, 22879ad1aaa6SAl Viro struct path *path, struct path *root) 2288873f1eedSJeff Layton { 2289894bc8c4SAl Viro int retval; 22909883d185SAl Viro struct nameidata nd; 2291abc9f5beSAl Viro if (IS_ERR(name)) 2292abc9f5beSAl Viro return PTR_ERR(name); 22939ad1aaa6SAl Viro if (unlikely(root)) { 22949ad1aaa6SAl Viro nd.root = *root; 22959ad1aaa6SAl Viro flags |= LOOKUP_ROOT; 22969ad1aaa6SAl Viro } 22979883d185SAl Viro set_nameidata(&nd, dfd, name); 2298c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_RCU, path); 2299873f1eedSJeff Layton if (unlikely(retval == -ECHILD)) 2300c8a53ee5SAl Viro retval = path_lookupat(&nd, flags, path); 2301873f1eedSJeff Layton if (unlikely(retval == -ESTALE)) 2302c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path); 2303873f1eedSJeff Layton 2304873f1eedSJeff Layton if (likely(!retval)) 2305625b6d10SAl Viro audit_inode(name, path->dentry, flags & LOOKUP_PARENT); 23069883d185SAl Viro restore_nameidata(); 2307e4bd1c1aSAl Viro putname(name); 2308873f1eedSJeff Layton return retval; 2309873f1eedSJeff Layton } 2310873f1eedSJeff Layton 23118bcb77faSAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2312c8a53ee5SAl Viro static int path_parentat(struct nameidata *nd, unsigned flags, 2313391172c4SAl Viro struct path *parent) 23148bcb77faSAl Viro { 2315c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2316368ee9baSAl Viro int err; 2317368ee9baSAl Viro if (IS_ERR(s)) 2318368ee9baSAl Viro return PTR_ERR(s); 2319368ee9baSAl Viro err = link_path_walk(s, nd); 23208bcb77faSAl Viro if (!err) 23218bcb77faSAl Viro err = complete_walk(nd); 2322391172c4SAl Viro if (!err) { 2323391172c4SAl Viro *parent = nd->path; 2324391172c4SAl Viro nd->path.mnt = NULL; 2325391172c4SAl Viro nd->path.dentry = NULL; 2326391172c4SAl Viro } 2327deb106c6SAl Viro terminate_walk(nd); 23288bcb77faSAl Viro return err; 23298bcb77faSAl Viro } 23308bcb77faSAl Viro 23315c31b6ceSAl Viro static struct filename *filename_parentat(int dfd, struct filename *name, 2332391172c4SAl Viro unsigned int flags, struct path *parent, 2333391172c4SAl Viro struct qstr *last, int *type) 23348bcb77faSAl Viro { 23358bcb77faSAl Viro int retval; 23369883d185SAl Viro struct nameidata nd; 23378bcb77faSAl Viro 23385c31b6ceSAl Viro if (IS_ERR(name)) 23395c31b6ceSAl Viro return name; 23409883d185SAl Viro set_nameidata(&nd, dfd, name); 2341c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_RCU, parent); 23428bcb77faSAl Viro if (unlikely(retval == -ECHILD)) 2343c8a53ee5SAl Viro retval = path_parentat(&nd, flags, parent); 23448bcb77faSAl Viro if (unlikely(retval == -ESTALE)) 2345c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent); 2346391172c4SAl Viro if (likely(!retval)) { 2347391172c4SAl Viro *last = nd.last; 2348391172c4SAl Viro *type = nd.last_type; 2349391172c4SAl Viro audit_inode(name, parent->dentry, LOOKUP_PARENT); 23505c31b6ceSAl Viro } else { 23515c31b6ceSAl Viro putname(name); 23525c31b6ceSAl Viro name = ERR_PTR(retval); 2353391172c4SAl Viro } 23549883d185SAl Viro restore_nameidata(); 23555c31b6ceSAl Viro return name; 23568bcb77faSAl Viro } 23578bcb77faSAl Viro 235879714f72SAl Viro /* does lookup, returns the object with parent locked */ 235979714f72SAl Viro struct dentry *kern_path_locked(const char *name, struct path *path) 23605590ff0dSUlrich Drepper { 23615c31b6ceSAl Viro struct filename *filename; 23625c31b6ceSAl Viro struct dentry *d; 2363391172c4SAl Viro struct qstr last; 2364391172c4SAl Viro int type; 236551689104SPaul Moore 23665c31b6ceSAl Viro filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path, 23675c31b6ceSAl Viro &last, &type); 236851689104SPaul Moore if (IS_ERR(filename)) 236951689104SPaul Moore return ERR_CAST(filename); 23705c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) { 2371391172c4SAl Viro path_put(path); 23725c31b6ceSAl Viro putname(filename); 23735c31b6ceSAl Viro return ERR_PTR(-EINVAL); 237479714f72SAl Viro } 23755955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 2376391172c4SAl Viro d = __lookup_hash(&last, path->dentry, 0); 237779714f72SAl Viro if (IS_ERR(d)) { 23785955102cSAl Viro inode_unlock(path->dentry->d_inode); 2379391172c4SAl Viro path_put(path); 238079714f72SAl Viro } 238151689104SPaul Moore putname(filename); 238279714f72SAl Viro return d; 23835590ff0dSUlrich Drepper } 23845590ff0dSUlrich Drepper 2385d1811465SAl Viro int kern_path(const char *name, unsigned int flags, struct path *path) 2386d1811465SAl Viro { 2387abc9f5beSAl Viro return filename_lookup(AT_FDCWD, getname_kernel(name), 2388abc9f5beSAl Viro flags, path, NULL); 2389d1811465SAl Viro } 23904d359507SAl Viro EXPORT_SYMBOL(kern_path); 2391d1811465SAl Viro 239216f18200SJosef 'Jeff' Sipek /** 239316f18200SJosef 'Jeff' Sipek * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair 239416f18200SJosef 'Jeff' Sipek * @dentry: pointer to dentry of the base directory 239516f18200SJosef 'Jeff' Sipek * @mnt: pointer to vfs mount of the base directory 239616f18200SJosef 'Jeff' Sipek * @name: pointer to file name 239716f18200SJosef 'Jeff' Sipek * @flags: lookup flags 2398e0a01249SAl Viro * @path: pointer to struct path to fill 239916f18200SJosef 'Jeff' Sipek */ 240016f18200SJosef 'Jeff' Sipek int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, 240116f18200SJosef 'Jeff' Sipek const char *name, unsigned int flags, 2402e0a01249SAl Viro struct path *path) 240316f18200SJosef 'Jeff' Sipek { 24049ad1aaa6SAl Viro struct path root = {.mnt = mnt, .dentry = dentry}; 24059ad1aaa6SAl Viro /* the first argument of filename_lookup() is ignored with root */ 2406abc9f5beSAl Viro return filename_lookup(AT_FDCWD, getname_kernel(name), 2407abc9f5beSAl Viro flags , path, &root); 240816f18200SJosef 'Jeff' Sipek } 24094d359507SAl Viro EXPORT_SYMBOL(vfs_path_lookup); 241016f18200SJosef 'Jeff' Sipek 2411eead1911SChristoph Hellwig /** 2412a6b91919SRandy Dunlap * lookup_one_len - filesystem helper to lookup single pathname component 2413eead1911SChristoph Hellwig * @name: pathname component to lookup 2414eead1911SChristoph Hellwig * @base: base directory to lookup from 2415eead1911SChristoph Hellwig * @len: maximum length @len should be interpreted to 2416eead1911SChristoph Hellwig * 2417a6b91919SRandy Dunlap * Note that this routine is purely a helper for filesystem usage and should 24189e7543e9SAl Viro * not be called by generic code. 2419bbddca8eSNeilBrown * 2420bbddca8eSNeilBrown * The caller must hold base->i_mutex. 2421eead1911SChristoph Hellwig */ 2422057f6c01SJames Morris struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 2423057f6c01SJames Morris { 2424057f6c01SJames Morris struct qstr this; 24256a96ba54SAl Viro unsigned int c; 2426cda309deSMiklos Szeredi int err; 2427057f6c01SJames Morris 24285955102cSAl Viro WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 24292f9092e1SDavid Woodhouse 24306a96ba54SAl Viro this.name = name; 24316a96ba54SAl Viro this.len = len; 24328387ff25SLinus Torvalds this.hash = full_name_hash(base, name, len); 24336a96ba54SAl Viro if (!len) 24346a96ba54SAl Viro return ERR_PTR(-EACCES); 24356a96ba54SAl Viro 243621d8a15aSAl Viro if (unlikely(name[0] == '.')) { 243721d8a15aSAl Viro if (len < 2 || (len == 2 && name[1] == '.')) 243821d8a15aSAl Viro return ERR_PTR(-EACCES); 243921d8a15aSAl Viro } 244021d8a15aSAl Viro 24416a96ba54SAl Viro while (len--) { 24426a96ba54SAl Viro c = *(const unsigned char *)name++; 24436a96ba54SAl Viro if (c == '/' || c == '\0') 24446a96ba54SAl Viro return ERR_PTR(-EACCES); 24456a96ba54SAl Viro } 24465a202bcdSAl Viro /* 24475a202bcdSAl Viro * See if the low-level filesystem might want 24485a202bcdSAl Viro * to use its own hash.. 24495a202bcdSAl Viro */ 24505a202bcdSAl Viro if (base->d_flags & DCACHE_OP_HASH) { 2451da53be12SLinus Torvalds int err = base->d_op->d_hash(base, &this); 24525a202bcdSAl Viro if (err < 0) 24535a202bcdSAl Viro return ERR_PTR(err); 24545a202bcdSAl Viro } 2455eead1911SChristoph Hellwig 2456cda309deSMiklos Szeredi err = inode_permission(base->d_inode, MAY_EXEC); 2457cda309deSMiklos Szeredi if (err) 2458cda309deSMiklos Szeredi return ERR_PTR(err); 2459cda309deSMiklos Szeredi 246072bd866aSAl Viro return __lookup_hash(&this, base, 0); 2461057f6c01SJames Morris } 24624d359507SAl Viro EXPORT_SYMBOL(lookup_one_len); 2463057f6c01SJames Morris 2464bbddca8eSNeilBrown /** 2465bbddca8eSNeilBrown * lookup_one_len_unlocked - filesystem helper to lookup single pathname component 2466bbddca8eSNeilBrown * @name: pathname component to lookup 2467bbddca8eSNeilBrown * @base: base directory to lookup from 2468bbddca8eSNeilBrown * @len: maximum length @len should be interpreted to 2469bbddca8eSNeilBrown * 2470bbddca8eSNeilBrown * Note that this routine is purely a helper for filesystem usage and should 2471bbddca8eSNeilBrown * not be called by generic code. 2472bbddca8eSNeilBrown * 2473bbddca8eSNeilBrown * Unlike lookup_one_len, it should be called without the parent 2474bbddca8eSNeilBrown * i_mutex held, and will take the i_mutex itself if necessary. 2475bbddca8eSNeilBrown */ 2476bbddca8eSNeilBrown struct dentry *lookup_one_len_unlocked(const char *name, 2477bbddca8eSNeilBrown struct dentry *base, int len) 2478bbddca8eSNeilBrown { 2479bbddca8eSNeilBrown struct qstr this; 2480bbddca8eSNeilBrown unsigned int c; 2481bbddca8eSNeilBrown int err; 248220d00ee8SLinus Torvalds struct dentry *ret; 2483bbddca8eSNeilBrown 2484bbddca8eSNeilBrown this.name = name; 2485bbddca8eSNeilBrown this.len = len; 24868387ff25SLinus Torvalds this.hash = full_name_hash(base, name, len); 2487bbddca8eSNeilBrown if (!len) 2488bbddca8eSNeilBrown return ERR_PTR(-EACCES); 2489bbddca8eSNeilBrown 2490bbddca8eSNeilBrown if (unlikely(name[0] == '.')) { 2491bbddca8eSNeilBrown if (len < 2 || (len == 2 && name[1] == '.')) 2492bbddca8eSNeilBrown return ERR_PTR(-EACCES); 2493bbddca8eSNeilBrown } 2494bbddca8eSNeilBrown 2495bbddca8eSNeilBrown while (len--) { 2496bbddca8eSNeilBrown c = *(const unsigned char *)name++; 2497bbddca8eSNeilBrown if (c == '/' || c == '\0') 2498bbddca8eSNeilBrown return ERR_PTR(-EACCES); 2499bbddca8eSNeilBrown } 2500bbddca8eSNeilBrown /* 2501bbddca8eSNeilBrown * See if the low-level filesystem might want 2502bbddca8eSNeilBrown * to use its own hash.. 2503bbddca8eSNeilBrown */ 2504bbddca8eSNeilBrown if (base->d_flags & DCACHE_OP_HASH) { 2505bbddca8eSNeilBrown int err = base->d_op->d_hash(base, &this); 2506bbddca8eSNeilBrown if (err < 0) 2507bbddca8eSNeilBrown return ERR_PTR(err); 2508bbddca8eSNeilBrown } 2509bbddca8eSNeilBrown 2510bbddca8eSNeilBrown err = inode_permission(base->d_inode, MAY_EXEC); 2511bbddca8eSNeilBrown if (err) 2512bbddca8eSNeilBrown return ERR_PTR(err); 2513bbddca8eSNeilBrown 251420d00ee8SLinus Torvalds ret = lookup_dcache(&this, base, 0); 251520d00ee8SLinus Torvalds if (!ret) 251620d00ee8SLinus Torvalds ret = lookup_slow(&this, base, 0); 251720d00ee8SLinus Torvalds return ret; 2518bbddca8eSNeilBrown } 2519bbddca8eSNeilBrown EXPORT_SYMBOL(lookup_one_len_unlocked); 2520bbddca8eSNeilBrown 2521eedf265aSEric W. Biederman #ifdef CONFIG_UNIX98_PTYS 2522eedf265aSEric W. Biederman int path_pts(struct path *path) 2523eedf265aSEric W. Biederman { 2524eedf265aSEric W. Biederman /* Find something mounted on "pts" in the same directory as 2525eedf265aSEric W. Biederman * the input path. 2526eedf265aSEric W. Biederman */ 2527eedf265aSEric W. Biederman struct dentry *child, *parent; 2528eedf265aSEric W. Biederman struct qstr this; 2529eedf265aSEric W. Biederman int ret; 2530eedf265aSEric W. Biederman 2531eedf265aSEric W. Biederman ret = path_parent_directory(path); 2532eedf265aSEric W. Biederman if (ret) 2533eedf265aSEric W. Biederman return ret; 2534eedf265aSEric W. Biederman 2535eedf265aSEric W. Biederman parent = path->dentry; 2536eedf265aSEric W. Biederman this.name = "pts"; 2537eedf265aSEric W. Biederman this.len = 3; 2538eedf265aSEric W. Biederman child = d_hash_and_lookup(parent, &this); 2539eedf265aSEric W. Biederman if (!child) 2540eedf265aSEric W. Biederman return -ENOENT; 2541eedf265aSEric W. Biederman 2542eedf265aSEric W. Biederman path->dentry = child; 2543eedf265aSEric W. Biederman dput(parent); 2544eedf265aSEric W. Biederman follow_mount(path); 2545eedf265aSEric W. Biederman return 0; 2546eedf265aSEric W. Biederman } 2547eedf265aSEric W. Biederman #endif 2548eedf265aSEric W. Biederman 25491fa1e7f6SAndy Whitcroft int user_path_at_empty(int dfd, const char __user *name, unsigned flags, 25501fa1e7f6SAndy Whitcroft struct path *path, int *empty) 25511da177e4SLinus Torvalds { 2552abc9f5beSAl Viro return filename_lookup(dfd, getname_flags(name, flags, empty), 2553abc9f5beSAl Viro flags, path, NULL); 25541da177e4SLinus Torvalds } 2555b853a161SAl Viro EXPORT_SYMBOL(user_path_at_empty); 25561fa1e7f6SAndy Whitcroft 25578033426eSJeff Layton /** 2558197df04cSAl Viro * mountpoint_last - look up last component for umount 25598033426eSJeff Layton * @nd: pathwalk nameidata - currently pointing at parent directory of "last" 25608033426eSJeff Layton * 25618033426eSJeff Layton * This is a special lookup_last function just for umount. In this case, we 25628033426eSJeff Layton * need to resolve the path without doing any revalidation. 25638033426eSJeff Layton * 25648033426eSJeff Layton * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since 25658033426eSJeff Layton * mountpoints are always pinned in the dcache, their ancestors are too. Thus, 25668033426eSJeff Layton * in almost all cases, this lookup will be served out of the dcache. The only 25678033426eSJeff Layton * cases where it won't are if nd->last refers to a symlink or the path is 25688033426eSJeff Layton * bogus and it doesn't exist. 25698033426eSJeff Layton * 25708033426eSJeff Layton * Returns: 25718033426eSJeff Layton * -error: if there was an error during lookup. This includes -ENOENT if the 2572ba8f4613SAl Viro * lookup found a negative dentry. 25738033426eSJeff Layton * 2574ba8f4613SAl Viro * 0: if we successfully resolved nd->last and found it to not to be a 2575ba8f4613SAl Viro * symlink that needs to be followed. 25768033426eSJeff Layton * 25778033426eSJeff Layton * 1: if we successfully resolved nd->last and found it to be a symlink 2578ba8f4613SAl Viro * that needs to be followed. 25798033426eSJeff Layton */ 25808033426eSJeff Layton static int 2581ba8f4613SAl Viro mountpoint_last(struct nameidata *nd) 25828033426eSJeff Layton { 25838033426eSJeff Layton int error = 0; 25848033426eSJeff Layton struct dentry *dir = nd->path.dentry; 2585ba8f4613SAl Viro struct path path; 25868033426eSJeff Layton 258735759521SAl Viro /* If we're in rcuwalk, drop out of it to handle last component */ 258835759521SAl Viro if (nd->flags & LOOKUP_RCU) { 25894675ac39SAl Viro if (unlazy_walk(nd)) 2590deb106c6SAl Viro return -ECHILD; 25918033426eSJeff Layton } 25928033426eSJeff Layton 25938033426eSJeff Layton nd->flags &= ~LOOKUP_PARENT; 25948033426eSJeff Layton 25958033426eSJeff Layton if (unlikely(nd->last_type != LAST_NORM)) { 25968033426eSJeff Layton error = handle_dots(nd, nd->last_type); 259735759521SAl Viro if (error) 2598deb106c6SAl Viro return error; 2599ba8f4613SAl Viro path.dentry = dget(nd->path.dentry); 2600949a852eSAl Viro } else { 2601ba8f4613SAl Viro path.dentry = d_lookup(dir, &nd->last); 2602ba8f4613SAl Viro if (!path.dentry) { 26038033426eSJeff Layton /* 2604949a852eSAl Viro * No cached dentry. Mounted dentries are pinned in the 2605949a852eSAl Viro * cache, so that means that this dentry is probably 2606949a852eSAl Viro * a symlink or the path doesn't actually point 2607949a852eSAl Viro * to a mounted dentry. 26088033426eSJeff Layton */ 2609ba8f4613SAl Viro path.dentry = lookup_slow(&nd->last, dir, 2610949a852eSAl Viro nd->flags | LOOKUP_NO_REVAL); 2611ba8f4613SAl Viro if (IS_ERR(path.dentry)) 2612ba8f4613SAl Viro return PTR_ERR(path.dentry); 26138033426eSJeff Layton } 2614bcceeebaSDave Jones } 2615ba8f4613SAl Viro if (d_is_negative(path.dentry)) { 2616ba8f4613SAl Viro dput(path.dentry); 2617deb106c6SAl Viro return -ENOENT; 261835759521SAl Viro } 2619ba8f4613SAl Viro path.mnt = nd->path.mnt; 26208f64fb1cSAl Viro return step_into(nd, &path, 0, d_backing_inode(path.dentry), 0); 26218033426eSJeff Layton } 26228033426eSJeff Layton 26238033426eSJeff Layton /** 2624197df04cSAl Viro * path_mountpoint - look up a path to be umounted 26252a78b857SMasanari Iida * @nd: lookup context 26268033426eSJeff Layton * @flags: lookup flags 2627c8a53ee5SAl Viro * @path: pointer to container for result 26288033426eSJeff Layton * 26298033426eSJeff Layton * Look up the given name, but don't attempt to revalidate the last component. 2630606d6fe3SRandy Dunlap * Returns 0 and "path" will be valid on success; Returns error otherwise. 26318033426eSJeff Layton */ 26328033426eSJeff Layton static int 2633c8a53ee5SAl Viro path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path) 26348033426eSJeff Layton { 2635c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2636368ee9baSAl Viro int err; 2637368ee9baSAl Viro if (IS_ERR(s)) 2638368ee9baSAl Viro return PTR_ERR(s); 26393bdba28bSAl Viro while (!(err = link_path_walk(s, nd)) && 2640ba8f4613SAl Viro (err = mountpoint_last(nd)) > 0) { 26413bdba28bSAl Viro s = trailing_symlink(nd); 26423bdba28bSAl Viro if (IS_ERR(s)) { 26433bdba28bSAl Viro err = PTR_ERR(s); 26448033426eSJeff Layton break; 26458033426eSJeff Layton } 26463bdba28bSAl Viro } 2647ba8f4613SAl Viro if (!err) { 2648ba8f4613SAl Viro *path = nd->path; 2649ba8f4613SAl Viro nd->path.mnt = NULL; 2650ba8f4613SAl Viro nd->path.dentry = NULL; 2651ba8f4613SAl Viro follow_mount(path); 2652ba8f4613SAl Viro } 2653deb106c6SAl Viro terminate_walk(nd); 26548033426eSJeff Layton return err; 26558033426eSJeff Layton } 26568033426eSJeff Layton 26572d864651SAl Viro static int 2658668696dcSAl Viro filename_mountpoint(int dfd, struct filename *name, struct path *path, 26592d864651SAl Viro unsigned int flags) 26602d864651SAl Viro { 26619883d185SAl Viro struct nameidata nd; 2662cbaab2dbSAl Viro int error; 2663668696dcSAl Viro if (IS_ERR(name)) 2664668696dcSAl Viro return PTR_ERR(name); 26659883d185SAl Viro set_nameidata(&nd, dfd, name); 2666c8a53ee5SAl Viro error = path_mountpoint(&nd, flags | LOOKUP_RCU, path); 26672d864651SAl Viro if (unlikely(error == -ECHILD)) 2668c8a53ee5SAl Viro error = path_mountpoint(&nd, flags, path); 26692d864651SAl Viro if (unlikely(error == -ESTALE)) 2670c8a53ee5SAl Viro error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path); 26712d864651SAl Viro if (likely(!error)) 2672668696dcSAl Viro audit_inode(name, path->dentry, 0); 26739883d185SAl Viro restore_nameidata(); 2674668696dcSAl Viro putname(name); 26752d864651SAl Viro return error; 26762d864651SAl Viro } 26772d864651SAl Viro 26788033426eSJeff Layton /** 2679197df04cSAl Viro * user_path_mountpoint_at - lookup a path from userland in order to umount it 26808033426eSJeff Layton * @dfd: directory file descriptor 26818033426eSJeff Layton * @name: pathname from userland 26828033426eSJeff Layton * @flags: lookup flags 26838033426eSJeff Layton * @path: pointer to container to hold result 26848033426eSJeff Layton * 26858033426eSJeff Layton * A umount is a special case for path walking. We're not actually interested 26868033426eSJeff Layton * in the inode in this situation, and ESTALE errors can be a problem. We 26878033426eSJeff Layton * simply want track down the dentry and vfsmount attached at the mountpoint 26888033426eSJeff Layton * and avoid revalidating the last component. 26898033426eSJeff Layton * 26908033426eSJeff Layton * Returns 0 and populates "path" on success. 26918033426eSJeff Layton */ 26928033426eSJeff Layton int 2693197df04cSAl Viro user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags, 26948033426eSJeff Layton struct path *path) 26958033426eSJeff Layton { 2696cbaab2dbSAl Viro return filename_mountpoint(dfd, getname(name), path, flags); 26978033426eSJeff Layton } 26988033426eSJeff Layton 26992d864651SAl Viro int 27002d864651SAl Viro kern_path_mountpoint(int dfd, const char *name, struct path *path, 27012d864651SAl Viro unsigned int flags) 27022d864651SAl Viro { 2703cbaab2dbSAl Viro return filename_mountpoint(dfd, getname_kernel(name), path, flags); 27042d864651SAl Viro } 27052d864651SAl Viro EXPORT_SYMBOL(kern_path_mountpoint); 27062d864651SAl Viro 2707cbdf35bcSMiklos Szeredi int __check_sticky(struct inode *dir, struct inode *inode) 27081da177e4SLinus Torvalds { 27098e96e3b7SEric W. Biederman kuid_t fsuid = current_fsuid(); 2710da9592edSDavid Howells 27118e96e3b7SEric W. Biederman if (uid_eq(inode->i_uid, fsuid)) 27121da177e4SLinus Torvalds return 0; 27138e96e3b7SEric W. Biederman if (uid_eq(dir->i_uid, fsuid)) 27141da177e4SLinus Torvalds return 0; 271523adbe12SAndy Lutomirski return !capable_wrt_inode_uidgid(inode, CAP_FOWNER); 27161da177e4SLinus Torvalds } 2717cbdf35bcSMiklos Szeredi EXPORT_SYMBOL(__check_sticky); 27181da177e4SLinus Torvalds 27191da177e4SLinus Torvalds /* 27201da177e4SLinus Torvalds * Check whether we can remove a link victim from directory dir, check 27211da177e4SLinus Torvalds * whether the type of victim is right. 27221da177e4SLinus Torvalds * 1. We can't do it if dir is read-only (done in permission()) 27231da177e4SLinus Torvalds * 2. We should have write and exec permissions on dir 27241da177e4SLinus Torvalds * 3. We can't remove anything from append-only dir 27251da177e4SLinus Torvalds * 4. We can't do anything with immutable dir (done in permission()) 27261da177e4SLinus Torvalds * 5. If the sticky bit on dir is set we should either 27271da177e4SLinus Torvalds * a. be owner of dir, or 27281da177e4SLinus Torvalds * b. be owner of victim, or 27291da177e4SLinus Torvalds * c. have CAP_FOWNER capability 27301da177e4SLinus Torvalds * 6. If the victim is append-only or immutable we can't do antyhing with 27311da177e4SLinus Torvalds * links pointing to it. 27320bd23d09SEric W. Biederman * 7. If the victim has an unknown uid or gid we can't change the inode. 27330bd23d09SEric W. Biederman * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR. 27340bd23d09SEric W. Biederman * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR. 27350bd23d09SEric W. Biederman * 10. We can't remove a root or mountpoint. 27360bd23d09SEric W. Biederman * 11. We don't allow removal of NFS sillyrenamed files; it's handled by 27371da177e4SLinus Torvalds * nfs_async_unlink(). 27381da177e4SLinus Torvalds */ 2739b18825a7SDavid Howells static int may_delete(struct inode *dir, struct dentry *victim, bool isdir) 27401da177e4SLinus Torvalds { 274163afdfc7SDavid Howells struct inode *inode = d_backing_inode(victim); 27421da177e4SLinus Torvalds int error; 27431da177e4SLinus Torvalds 2744b18825a7SDavid Howells if (d_is_negative(victim)) 27451da177e4SLinus Torvalds return -ENOENT; 2746b18825a7SDavid Howells BUG_ON(!inode); 27471da177e4SLinus Torvalds 27481da177e4SLinus Torvalds BUG_ON(victim->d_parent->d_inode != dir); 27494fa6b5ecSJeff Layton audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); 27501da177e4SLinus Torvalds 2751f419a2e3SAl Viro error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 27521da177e4SLinus Torvalds if (error) 27531da177e4SLinus Torvalds return error; 27541da177e4SLinus Torvalds if (IS_APPEND(dir)) 27551da177e4SLinus Torvalds return -EPERM; 2756b18825a7SDavid Howells 2757b18825a7SDavid Howells if (check_sticky(dir, inode) || IS_APPEND(inode) || 27580bd23d09SEric W. Biederman IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || HAS_UNMAPPED_ID(inode)) 27591da177e4SLinus Torvalds return -EPERM; 27601da177e4SLinus Torvalds if (isdir) { 276144b1d530SMiklos Szeredi if (!d_is_dir(victim)) 27621da177e4SLinus Torvalds return -ENOTDIR; 27631da177e4SLinus Torvalds if (IS_ROOT(victim)) 27641da177e4SLinus Torvalds return -EBUSY; 276544b1d530SMiklos Szeredi } else if (d_is_dir(victim)) 27661da177e4SLinus Torvalds return -EISDIR; 27671da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 27681da177e4SLinus Torvalds return -ENOENT; 27691da177e4SLinus Torvalds if (victim->d_flags & DCACHE_NFSFS_RENAMED) 27701da177e4SLinus Torvalds return -EBUSY; 27711da177e4SLinus Torvalds return 0; 27721da177e4SLinus Torvalds } 27731da177e4SLinus Torvalds 27741da177e4SLinus Torvalds /* Check whether we can create an object with dentry child in directory 27751da177e4SLinus Torvalds * dir. 27761da177e4SLinus Torvalds * 1. We can't do it if child already exists (open has special treatment for 27771da177e4SLinus Torvalds * this case, but since we are inlined it's OK) 27781da177e4SLinus Torvalds * 2. We can't do it if dir is read-only (done in permission()) 2779036d5236SEric W. Biederman * 3. We can't do it if the fs can't represent the fsuid or fsgid. 2780036d5236SEric W. Biederman * 4. We should have write and exec permissions on dir 2781036d5236SEric W. Biederman * 5. We can't do it if dir is immutable (done in permission()) 27821da177e4SLinus Torvalds */ 2783a95164d9SMiklos Szeredi static inline int may_create(struct inode *dir, struct dentry *child) 27841da177e4SLinus Torvalds { 2785036d5236SEric W. Biederman struct user_namespace *s_user_ns; 278614e972b4SJeff Layton audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); 27871da177e4SLinus Torvalds if (child->d_inode) 27881da177e4SLinus Torvalds return -EEXIST; 27891da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 27901da177e4SLinus Torvalds return -ENOENT; 2791036d5236SEric W. Biederman s_user_ns = dir->i_sb->s_user_ns; 2792036d5236SEric W. Biederman if (!kuid_has_mapping(s_user_ns, current_fsuid()) || 2793036d5236SEric W. Biederman !kgid_has_mapping(s_user_ns, current_fsgid())) 2794036d5236SEric W. Biederman return -EOVERFLOW; 2795f419a2e3SAl Viro return inode_permission(dir, MAY_WRITE | MAY_EXEC); 27961da177e4SLinus Torvalds } 27971da177e4SLinus Torvalds 27981da177e4SLinus Torvalds /* 27991da177e4SLinus Torvalds * p1 and p2 should be directories on the same fs. 28001da177e4SLinus Torvalds */ 28011da177e4SLinus Torvalds struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 28021da177e4SLinus Torvalds { 28031da177e4SLinus Torvalds struct dentry *p; 28041da177e4SLinus Torvalds 28051da177e4SLinus Torvalds if (p1 == p2) { 28065955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 28071da177e4SLinus Torvalds return NULL; 28081da177e4SLinus Torvalds } 28091da177e4SLinus Torvalds 2810fc64005cSAl Viro mutex_lock(&p1->d_sb->s_vfs_rename_mutex); 28111da177e4SLinus Torvalds 2812e2761a11SOGAWA Hirofumi p = d_ancestor(p2, p1); 2813e2761a11SOGAWA Hirofumi if (p) { 28145955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT); 28155955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_CHILD); 28161da177e4SLinus Torvalds return p; 28171da177e4SLinus Torvalds } 28181da177e4SLinus Torvalds 2819e2761a11SOGAWA Hirofumi p = d_ancestor(p1, p2); 2820e2761a11SOGAWA Hirofumi if (p) { 28215955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 28225955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_CHILD); 28231da177e4SLinus Torvalds return p; 28241da177e4SLinus Torvalds } 28251da177e4SLinus Torvalds 28265955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 28275955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2); 28281da177e4SLinus Torvalds return NULL; 28291da177e4SLinus Torvalds } 28304d359507SAl Viro EXPORT_SYMBOL(lock_rename); 28311da177e4SLinus Torvalds 28321da177e4SLinus Torvalds void unlock_rename(struct dentry *p1, struct dentry *p2) 28331da177e4SLinus Torvalds { 28345955102cSAl Viro inode_unlock(p1->d_inode); 28351da177e4SLinus Torvalds if (p1 != p2) { 28365955102cSAl Viro inode_unlock(p2->d_inode); 2837fc64005cSAl Viro mutex_unlock(&p1->d_sb->s_vfs_rename_mutex); 28381da177e4SLinus Torvalds } 28391da177e4SLinus Torvalds } 28404d359507SAl Viro EXPORT_SYMBOL(unlock_rename); 28411da177e4SLinus Torvalds 28424acdaf27SAl Viro int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, 2843312b63fbSAl Viro bool want_excl) 28441da177e4SLinus Torvalds { 2845a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 28461da177e4SLinus Torvalds if (error) 28471da177e4SLinus Torvalds return error; 28481da177e4SLinus Torvalds 2849acfa4380SAl Viro if (!dir->i_op->create) 28501da177e4SLinus Torvalds return -EACCES; /* shouldn't it be ENOSYS? */ 28511da177e4SLinus Torvalds mode &= S_IALLUGO; 28521da177e4SLinus Torvalds mode |= S_IFREG; 28531da177e4SLinus Torvalds error = security_inode_create(dir, dentry, mode); 28541da177e4SLinus Torvalds if (error) 28551da177e4SLinus Torvalds return error; 2856312b63fbSAl Viro error = dir->i_op->create(dir, dentry, mode, want_excl); 2857a74574aaSStephen Smalley if (!error) 2858f38aa942SAmy Griffis fsnotify_create(dir, dentry); 28591da177e4SLinus Torvalds return error; 28601da177e4SLinus Torvalds } 28614d359507SAl Viro EXPORT_SYMBOL(vfs_create); 28621da177e4SLinus Torvalds 2863a2982cc9SEric W. Biederman bool may_open_dev(const struct path *path) 2864a2982cc9SEric W. Biederman { 2865a2982cc9SEric W. Biederman return !(path->mnt->mnt_flags & MNT_NODEV) && 2866a2982cc9SEric W. Biederman !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV); 2867a2982cc9SEric W. Biederman } 2868a2982cc9SEric W. Biederman 2869f0bb5aafSAl Viro static int may_open(const struct path *path, int acc_mode, int flag) 28701da177e4SLinus Torvalds { 28713fb64190SChristoph Hellwig struct dentry *dentry = path->dentry; 28721da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 28731da177e4SLinus Torvalds int error; 28741da177e4SLinus Torvalds 28751da177e4SLinus Torvalds if (!inode) 28761da177e4SLinus Torvalds return -ENOENT; 28771da177e4SLinus Torvalds 2878c8fe8f30SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 2879c8fe8f30SChristoph Hellwig case S_IFLNK: 28801da177e4SLinus Torvalds return -ELOOP; 2881c8fe8f30SChristoph Hellwig case S_IFDIR: 2882c8fe8f30SChristoph Hellwig if (acc_mode & MAY_WRITE) 28831da177e4SLinus Torvalds return -EISDIR; 2884c8fe8f30SChristoph Hellwig break; 2885c8fe8f30SChristoph Hellwig case S_IFBLK: 2886c8fe8f30SChristoph Hellwig case S_IFCHR: 2887a2982cc9SEric W. Biederman if (!may_open_dev(path)) 28881da177e4SLinus Torvalds return -EACCES; 2889c8fe8f30SChristoph Hellwig /*FALLTHRU*/ 2890c8fe8f30SChristoph Hellwig case S_IFIFO: 2891c8fe8f30SChristoph Hellwig case S_IFSOCK: 28921da177e4SLinus Torvalds flag &= ~O_TRUNC; 2893c8fe8f30SChristoph Hellwig break; 28944a3fd211SDave Hansen } 2895b41572e9SDave Hansen 289662fb4a15SAl Viro error = inode_permission(inode, MAY_OPEN | acc_mode); 2897b41572e9SDave Hansen if (error) 2898b41572e9SDave Hansen return error; 28996146f0d5SMimi Zohar 29001da177e4SLinus Torvalds /* 29011da177e4SLinus Torvalds * An append-only file must be opened in append mode for writing. 29021da177e4SLinus Torvalds */ 29031da177e4SLinus Torvalds if (IS_APPEND(inode)) { 29048737c930SAl Viro if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) 29057715b521SAl Viro return -EPERM; 29061da177e4SLinus Torvalds if (flag & O_TRUNC) 29077715b521SAl Viro return -EPERM; 29081da177e4SLinus Torvalds } 29091da177e4SLinus Torvalds 29101da177e4SLinus Torvalds /* O_NOATIME can only be set by the owner or superuser */ 29112e149670SSerge E. Hallyn if (flag & O_NOATIME && !inode_owner_or_capable(inode)) 29127715b521SAl Viro return -EPERM; 29131da177e4SLinus Torvalds 2914f3c7691eSJ. Bruce Fields return 0; 29157715b521SAl Viro } 29167715b521SAl Viro 2917e1181ee6SJeff Layton static int handle_truncate(struct file *filp) 29187715b521SAl Viro { 2919f0bb5aafSAl Viro const struct path *path = &filp->f_path; 29207715b521SAl Viro struct inode *inode = path->dentry->d_inode; 29217715b521SAl Viro int error = get_write_access(inode); 29221da177e4SLinus Torvalds if (error) 29237715b521SAl Viro return error; 29241da177e4SLinus Torvalds /* 29251da177e4SLinus Torvalds * Refuse to truncate files with mandatory locks held on them. 29261da177e4SLinus Torvalds */ 2927d7a06983SJeff Layton error = locks_verify_locked(filp); 2928be6d3e56SKentaro Takeda if (!error) 2929ea0d3ab2STetsuo Handa error = security_path_truncate(path); 29301da177e4SLinus Torvalds if (!error) { 29317715b521SAl Viro error = do_truncate(path->dentry, 0, 2932d139d7ffSMiklos Szeredi ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, 2933e1181ee6SJeff Layton filp); 29341da177e4SLinus Torvalds } 29351da177e4SLinus Torvalds put_write_access(inode); 2936acd0c935SMimi Zohar return error; 29371da177e4SLinus Torvalds } 29381da177e4SLinus Torvalds 2939d57999e1SDave Hansen static inline int open_to_namei_flags(int flag) 2940d57999e1SDave Hansen { 29418a5e929dSAl Viro if ((flag & O_ACCMODE) == 3) 29428a5e929dSAl Viro flag--; 2943d57999e1SDave Hansen return flag; 2944d57999e1SDave Hansen } 2945d57999e1SDave Hansen 2946d3607752SAl Viro static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode) 2947d18e9008SMiklos Szeredi { 29481328c727SSeth Forshee struct user_namespace *s_user_ns; 2949d18e9008SMiklos Szeredi int error = security_path_mknod(dir, dentry, mode, 0); 2950d18e9008SMiklos Szeredi if (error) 2951d18e9008SMiklos Szeredi return error; 2952d18e9008SMiklos Szeredi 29531328c727SSeth Forshee s_user_ns = dir->dentry->d_sb->s_user_ns; 29541328c727SSeth Forshee if (!kuid_has_mapping(s_user_ns, current_fsuid()) || 29551328c727SSeth Forshee !kgid_has_mapping(s_user_ns, current_fsgid())) 29561328c727SSeth Forshee return -EOVERFLOW; 29571328c727SSeth Forshee 2958d18e9008SMiklos Szeredi error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC); 2959d18e9008SMiklos Szeredi if (error) 2960d18e9008SMiklos Szeredi return error; 2961d18e9008SMiklos Szeredi 2962d18e9008SMiklos Szeredi return security_inode_create(dir->dentry->d_inode, dentry, mode); 2963d18e9008SMiklos Szeredi } 2964d18e9008SMiklos Szeredi 29651acf0af9SDavid Howells /* 29661acf0af9SDavid Howells * Attempt to atomically look up, create and open a file from a negative 29671acf0af9SDavid Howells * dentry. 29681acf0af9SDavid Howells * 29691acf0af9SDavid Howells * Returns 0 if successful. The file will have been created and attached to 29701acf0af9SDavid Howells * @file by the filesystem calling finish_open(). 29711acf0af9SDavid Howells * 29721acf0af9SDavid Howells * Returns 1 if the file was looked up only or didn't need creating. The 29731acf0af9SDavid Howells * caller will need to perform the open themselves. @path will have been 29741acf0af9SDavid Howells * updated to point to the new dentry. This may be negative. 29751acf0af9SDavid Howells * 29761acf0af9SDavid Howells * Returns an error code otherwise. 29771acf0af9SDavid Howells */ 29782675a4ebSAl Viro static int atomic_open(struct nameidata *nd, struct dentry *dentry, 297930d90494SAl Viro struct path *path, struct file *file, 2980015c3bbcSMiklos Szeredi const struct open_flags *op, 29811643b43fSAl Viro int open_flag, umode_t mode, 298247237687SAl Viro int *opened) 2983d18e9008SMiklos Szeredi { 2984d18e9008SMiklos Szeredi struct dentry *const DENTRY_NOT_SET = (void *) -1UL; 2985d18e9008SMiklos Szeredi struct inode *dir = nd->path.dentry->d_inode; 2986d18e9008SMiklos Szeredi int error; 2987d18e9008SMiklos Szeredi 2988384f26e2SAl Viro if (!(~open_flag & (O_EXCL | O_CREAT))) /* both O_EXCL and O_CREAT */ 2989d18e9008SMiklos Szeredi open_flag &= ~O_TRUNC; 2990d18e9008SMiklos Szeredi 2991d18e9008SMiklos Szeredi if (nd->flags & LOOKUP_DIRECTORY) 2992d18e9008SMiklos Szeredi open_flag |= O_DIRECTORY; 2993d18e9008SMiklos Szeredi 299430d90494SAl Viro file->f_path.dentry = DENTRY_NOT_SET; 299530d90494SAl Viro file->f_path.mnt = nd->path.mnt; 29960fb1ea09SAl Viro error = dir->i_op->atomic_open(dir, dentry, file, 29970fb1ea09SAl Viro open_to_namei_flags(open_flag), 29980fb1ea09SAl Viro mode, opened); 29996fbd0714SAl Viro d_lookup_done(dentry); 3000384f26e2SAl Viro if (!error) { 3001d18e9008SMiklos Szeredi /* 3002384f26e2SAl Viro * We didn't have the inode before the open, so check open 3003384f26e2SAl Viro * permission here. 3004d18e9008SMiklos Szeredi */ 3005384f26e2SAl Viro int acc_mode = op->acc_mode; 300603da633aSAl Viro if (*opened & FILE_CREATED) { 300703da633aSAl Viro WARN_ON(!(open_flag & O_CREAT)); 300803da633aSAl Viro fsnotify_create(dir, dentry); 300962fb4a15SAl Viro acc_mode = 0; 301003da633aSAl Viro } 30112675a4ebSAl Viro error = may_open(&file->f_path, acc_mode, open_flag); 3012384f26e2SAl Viro if (WARN_ON(error > 0)) 3013384f26e2SAl Viro error = -EINVAL; 3014384f26e2SAl Viro } else if (error > 0) { 3015d18e9008SMiklos Szeredi if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { 3016d18e9008SMiklos Szeredi error = -EIO; 3017384f26e2SAl Viro } else { 3018d18e9008SMiklos Szeredi if (file->f_path.dentry) { 3019d18e9008SMiklos Szeredi dput(dentry); 3020d18e9008SMiklos Szeredi dentry = file->f_path.dentry; 302110c64ceaSAl Viro } 3022d18e9008SMiklos Szeredi if (*opened & FILE_CREATED) 3023d18e9008SMiklos Szeredi fsnotify_create(dir, dentry); 3024a01e718fSAl Viro if (unlikely(d_is_negative(dentry))) { 3025a01e718fSAl Viro error = -ENOENT; 3026a01e718fSAl Viro } else { 3027d18e9008SMiklos Szeredi path->dentry = dentry; 3028d18e9008SMiklos Szeredi path->mnt = nd->path.mnt; 30292675a4ebSAl Viro return 1; 3030d18e9008SMiklos Szeredi } 3031d18e9008SMiklos Szeredi } 3032a01e718fSAl Viro } 3033d18e9008SMiklos Szeredi dput(dentry); 3034d18e9008SMiklos Szeredi return error; 3035d18e9008SMiklos Szeredi } 3036d18e9008SMiklos Szeredi 303731e6b01fSNick Piggin /* 30381acf0af9SDavid Howells * Look up and maybe create and open the last component. 3039d58ffd35SMiklos Szeredi * 3040d58ffd35SMiklos Szeredi * Must be called with i_mutex held on parent. 3041d58ffd35SMiklos Szeredi * 30421acf0af9SDavid Howells * Returns 0 if the file was successfully atomically created (if necessary) and 30431acf0af9SDavid Howells * opened. In this case the file will be returned attached to @file. 30441acf0af9SDavid Howells * 30451acf0af9SDavid Howells * Returns 1 if the file was not completely opened at this time, though lookups 30461acf0af9SDavid Howells * and creations will have been performed and the dentry returned in @path will 30471acf0af9SDavid Howells * be positive upon return if O_CREAT was specified. If O_CREAT wasn't 30481acf0af9SDavid Howells * specified then a negative dentry may be returned. 30491acf0af9SDavid Howells * 30501acf0af9SDavid Howells * An error code is returned otherwise. 30511acf0af9SDavid Howells * 30521acf0af9SDavid Howells * FILE_CREATE will be set in @*opened if the dentry was created and will be 30531acf0af9SDavid Howells * cleared otherwise prior to returning. 3054d58ffd35SMiklos Szeredi */ 30552675a4ebSAl Viro static int lookup_open(struct nameidata *nd, struct path *path, 305630d90494SAl Viro struct file *file, 3057d58ffd35SMiklos Szeredi const struct open_flags *op, 305864894cf8SAl Viro bool got_write, int *opened) 3059d58ffd35SMiklos Szeredi { 3060d58ffd35SMiklos Szeredi struct dentry *dir = nd->path.dentry; 306154ef4872SMiklos Szeredi struct inode *dir_inode = dir->d_inode; 30621643b43fSAl Viro int open_flag = op->open_flag; 3063d58ffd35SMiklos Szeredi struct dentry *dentry; 30641643b43fSAl Viro int error, create_error = 0; 30651643b43fSAl Viro umode_t mode = op->mode; 30666fbd0714SAl Viro DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 3067d58ffd35SMiklos Szeredi 3068ce8644fcSAl Viro if (unlikely(IS_DEADDIR(dir_inode))) 3069ce8644fcSAl Viro return -ENOENT; 3070d58ffd35SMiklos Szeredi 307147237687SAl Viro *opened &= ~FILE_CREATED; 30726fbd0714SAl Viro dentry = d_lookup(dir, &nd->last); 30736fbd0714SAl Viro for (;;) { 30746fbd0714SAl Viro if (!dentry) { 30756fbd0714SAl Viro dentry = d_alloc_parallel(dir, &nd->last, &wq); 3076d58ffd35SMiklos Szeredi if (IS_ERR(dentry)) 30772675a4ebSAl Viro return PTR_ERR(dentry); 30786fbd0714SAl Viro } 30796fbd0714SAl Viro if (d_in_lookup(dentry)) 30806fbd0714SAl Viro break; 3081d58ffd35SMiklos Szeredi 30826fbd0714SAl Viro error = d_revalidate(dentry, nd->flags); 30836fbd0714SAl Viro if (likely(error > 0)) 30846fbd0714SAl Viro break; 30856fbd0714SAl Viro if (error) 30866fbd0714SAl Viro goto out_dput; 30876fbd0714SAl Viro d_invalidate(dentry); 30886fbd0714SAl Viro dput(dentry); 30896fbd0714SAl Viro dentry = NULL; 30906fbd0714SAl Viro } 30916fbd0714SAl Viro if (dentry->d_inode) { 3092d18e9008SMiklos Szeredi /* Cached positive dentry: will open in f_op->open */ 3093d18e9008SMiklos Szeredi goto out_no_open; 30946c51e513SAl Viro } 3095d18e9008SMiklos Szeredi 30961643b43fSAl Viro /* 30971643b43fSAl Viro * Checking write permission is tricky, bacuse we don't know if we are 30981643b43fSAl Viro * going to actually need it: O_CREAT opens should work as long as the 30991643b43fSAl Viro * file exists. But checking existence breaks atomicity. The trick is 31001643b43fSAl Viro * to check access and if not granted clear O_CREAT from the flags. 31011643b43fSAl Viro * 31021643b43fSAl Viro * Another problem is returing the "right" error value (e.g. for an 31031643b43fSAl Viro * O_EXCL open we want to return EEXIST not EROFS). 31041643b43fSAl Viro */ 31051643b43fSAl Viro if (open_flag & O_CREAT) { 31061643b43fSAl Viro if (!IS_POSIXACL(dir->d_inode)) 31071643b43fSAl Viro mode &= ~current_umask(); 31081643b43fSAl Viro if (unlikely(!got_write)) { 31091643b43fSAl Viro create_error = -EROFS; 31101643b43fSAl Viro open_flag &= ~O_CREAT; 31111643b43fSAl Viro if (open_flag & (O_EXCL | O_TRUNC)) 31121643b43fSAl Viro goto no_open; 31131643b43fSAl Viro /* No side effects, safe to clear O_CREAT */ 31141643b43fSAl Viro } else { 31151643b43fSAl Viro create_error = may_o_create(&nd->path, dentry, mode); 31161643b43fSAl Viro if (create_error) { 31171643b43fSAl Viro open_flag &= ~O_CREAT; 31181643b43fSAl Viro if (open_flag & O_EXCL) 31191643b43fSAl Viro goto no_open; 31201643b43fSAl Viro } 31211643b43fSAl Viro } 31221643b43fSAl Viro } else if ((open_flag & (O_TRUNC|O_WRONLY|O_RDWR)) && 31231643b43fSAl Viro unlikely(!got_write)) { 31241643b43fSAl Viro /* 31251643b43fSAl Viro * No O_CREATE -> atomicity not a requirement -> fall 31261643b43fSAl Viro * back to lookup + open 31271643b43fSAl Viro */ 31281643b43fSAl Viro goto no_open; 3129d18e9008SMiklos Szeredi } 3130d18e9008SMiklos Szeredi 31311643b43fSAl Viro if (dir_inode->i_op->atomic_open) { 31321643b43fSAl Viro error = atomic_open(nd, dentry, path, file, op, open_flag, 31331643b43fSAl Viro mode, opened); 31341643b43fSAl Viro if (unlikely(error == -ENOENT) && create_error) 31351643b43fSAl Viro error = create_error; 31361643b43fSAl Viro return error; 31371643b43fSAl Viro } 313854ef4872SMiklos Szeredi 31391643b43fSAl Viro no_open: 31406fbd0714SAl Viro if (d_in_lookup(dentry)) { 314112fa5e24SAl Viro struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry, 314212fa5e24SAl Viro nd->flags); 31436fbd0714SAl Viro d_lookup_done(dentry); 314412fa5e24SAl Viro if (unlikely(res)) { 314512fa5e24SAl Viro if (IS_ERR(res)) { 314612fa5e24SAl Viro error = PTR_ERR(res); 314712fa5e24SAl Viro goto out_dput; 314812fa5e24SAl Viro } 314912fa5e24SAl Viro dput(dentry); 315012fa5e24SAl Viro dentry = res; 315112fa5e24SAl Viro } 315254ef4872SMiklos Szeredi } 315354ef4872SMiklos Szeredi 3154d58ffd35SMiklos Szeredi /* Negative dentry, just create the file */ 31551643b43fSAl Viro if (!dentry->d_inode && (open_flag & O_CREAT)) { 315647237687SAl Viro *opened |= FILE_CREATED; 3157ce8644fcSAl Viro audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE); 3158ce8644fcSAl Viro if (!dir_inode->i_op->create) { 3159ce8644fcSAl Viro error = -EACCES; 3160d58ffd35SMiklos Szeredi goto out_dput; 316177d660a8SMiklos Szeredi } 3162ce8644fcSAl Viro error = dir_inode->i_op->create(dir_inode, dentry, mode, 31631643b43fSAl Viro open_flag & O_EXCL); 3164d58ffd35SMiklos Szeredi if (error) 3165d58ffd35SMiklos Szeredi goto out_dput; 3166ce8644fcSAl Viro fsnotify_create(dir_inode, dentry); 3167d58ffd35SMiklos Szeredi } 31681643b43fSAl Viro if (unlikely(create_error) && !dentry->d_inode) { 31691643b43fSAl Viro error = create_error; 3170d58ffd35SMiklos Szeredi goto out_dput; 3171d58ffd35SMiklos Szeredi } 3172d18e9008SMiklos Szeredi out_no_open: 3173d58ffd35SMiklos Szeredi path->dentry = dentry; 3174d58ffd35SMiklos Szeredi path->mnt = nd->path.mnt; 31752675a4ebSAl Viro return 1; 3176d58ffd35SMiklos Szeredi 3177d58ffd35SMiklos Szeredi out_dput: 3178d58ffd35SMiklos Szeredi dput(dentry); 31792675a4ebSAl Viro return error; 3180d58ffd35SMiklos Szeredi } 3181d58ffd35SMiklos Szeredi 3182d58ffd35SMiklos Szeredi /* 3183fe2d35ffSAl Viro * Handle the last step of open() 318431e6b01fSNick Piggin */ 3185896475d5SAl Viro static int do_last(struct nameidata *nd, 318630d90494SAl Viro struct file *file, const struct open_flags *op, 318776ae2a5aSAl Viro int *opened) 3188fb1cc555SAl Viro { 3189a1e28038SAl Viro struct dentry *dir = nd->path.dentry; 3190ca344a89SAl Viro int open_flag = op->open_flag; 319177d660a8SMiklos Szeredi bool will_truncate = (open_flag & O_TRUNC) != 0; 319264894cf8SAl Viro bool got_write = false; 3193bcda7652SAl Viro int acc_mode = op->acc_mode; 3194254cf582SAl Viro unsigned seq; 3195a1eb3315SMiklos Szeredi struct inode *inode; 3196896475d5SAl Viro struct path path; 319716c2cd71SAl Viro int error; 3198fb1cc555SAl Viro 3199c3e380b0SAl Viro nd->flags &= ~LOOKUP_PARENT; 3200c3e380b0SAl Viro nd->flags |= op->intent; 3201c3e380b0SAl Viro 3202bc77daa7SAl Viro if (nd->last_type != LAST_NORM) { 3203fe2d35ffSAl Viro error = handle_dots(nd, nd->last_type); 3204deb106c6SAl Viro if (unlikely(error)) 32052675a4ebSAl Viro return error; 3206e83db167SMiklos Szeredi goto finish_open; 32071f36f774SAl Viro } 3208a2c36b45SAl Viro 3209ca344a89SAl Viro if (!(open_flag & O_CREAT)) { 3210fe2d35ffSAl Viro if (nd->last.name[nd->last.len]) 3211fe2d35ffSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 3212fe2d35ffSAl Viro /* we _can_ be in RCU mode here */ 3213254cf582SAl Viro error = lookup_fast(nd, &path, &inode, &seq); 3214e9742b53SAl Viro if (likely(error > 0)) 321571574865SMiklos Szeredi goto finish_lookup; 321671574865SMiklos Szeredi 3217ce57dfc1SAl Viro if (error < 0) 3218deb106c6SAl Viro return error; 3219a1eb3315SMiklos Szeredi 322037d7fffcSMiklos Szeredi BUG_ON(nd->inode != dir->d_inode); 32216583fe22SAl Viro BUG_ON(nd->flags & LOOKUP_RCU); 3222b6183df7SMiklos Szeredi } else { 3223fe2d35ffSAl Viro /* create side of things */ 3224a3fbbde7SAl Viro /* 3225b6183df7SMiklos Szeredi * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED 3226b6183df7SMiklos Szeredi * has been cleared when we got to the last component we are 3227b6183df7SMiklos Szeredi * about to look up 3228a3fbbde7SAl Viro */ 32299f1fafeeSAl Viro error = complete_walk(nd); 3230e8bb73dfSAl Viro if (error) 32312675a4ebSAl Viro return error; 3232fe2d35ffSAl Viro 323376ae2a5aSAl Viro audit_inode(nd->name, dir, LOOKUP_PARENT); 32341f36f774SAl Viro /* trailing slashes? */ 3235deb106c6SAl Viro if (unlikely(nd->last.name[nd->last.len])) 3236deb106c6SAl Viro return -EISDIR; 3237b6183df7SMiklos Szeredi } 32381f36f774SAl Viro 32399cf843e3SAl Viro if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { 324064894cf8SAl Viro error = mnt_want_write(nd->path.mnt); 324164894cf8SAl Viro if (!error) 324264894cf8SAl Viro got_write = true; 324364894cf8SAl Viro /* 324464894cf8SAl Viro * do _not_ fail yet - we might not need that or fail with 324564894cf8SAl Viro * a different error; let lookup_open() decide; we'll be 324664894cf8SAl Viro * dropping this one anyway. 324764894cf8SAl Viro */ 324864894cf8SAl Viro } 32499cf843e3SAl Viro if (open_flag & O_CREAT) 32505955102cSAl Viro inode_lock(dir->d_inode); 32519cf843e3SAl Viro else 32529cf843e3SAl Viro inode_lock_shared(dir->d_inode); 3253896475d5SAl Viro error = lookup_open(nd, &path, file, op, got_write, opened); 32549cf843e3SAl Viro if (open_flag & O_CREAT) 32555955102cSAl Viro inode_unlock(dir->d_inode); 32569cf843e3SAl Viro else 32579cf843e3SAl Viro inode_unlock_shared(dir->d_inode); 3258fb1cc555SAl Viro 32592675a4ebSAl Viro if (error <= 0) { 32602675a4ebSAl Viro if (error) 3261d58ffd35SMiklos Szeredi goto out; 32626c0d46c4SAl Viro 326347237687SAl Viro if ((*opened & FILE_CREATED) || 3264496ad9aaSAl Viro !S_ISREG(file_inode(file)->i_mode)) 326577d660a8SMiklos Szeredi will_truncate = false; 3266d18e9008SMiklos Szeredi 326776ae2a5aSAl Viro audit_inode(nd->name, file->f_path.dentry, 0); 3268d18e9008SMiklos Szeredi goto opened; 3269d18e9008SMiklos Szeredi } 3270d18e9008SMiklos Szeredi 327147237687SAl Viro if (*opened & FILE_CREATED) { 32729b44f1b3SAl Viro /* Don't check for write permission, don't truncate */ 3273ca344a89SAl Viro open_flag &= ~O_TRUNC; 327477d660a8SMiklos Szeredi will_truncate = false; 327562fb4a15SAl Viro acc_mode = 0; 3276896475d5SAl Viro path_to_nameidata(&path, nd); 3277e83db167SMiklos Szeredi goto finish_open_created; 3278fb1cc555SAl Viro } 3279fb1cc555SAl Viro 3280fb1cc555SAl Viro /* 3281d18e9008SMiklos Szeredi * If atomic_open() acquired write access it is dropped now due to 3282d18e9008SMiklos Szeredi * possible mount and symlink following (this might be optimized away if 3283d18e9008SMiklos Szeredi * necessary...) 3284d18e9008SMiklos Szeredi */ 328564894cf8SAl Viro if (got_write) { 3286d18e9008SMiklos Szeredi mnt_drop_write(nd->path.mnt); 328764894cf8SAl Viro got_write = false; 3288d18e9008SMiklos Szeredi } 3289d18e9008SMiklos Szeredi 3290e6ec03a2SAl Viro error = follow_managed(&path, nd); 3291e6ec03a2SAl Viro if (unlikely(error < 0)) 3292e6ec03a2SAl Viro return error; 3293e6ec03a2SAl Viro 32946583fe22SAl Viro if (unlikely(d_is_negative(path.dentry))) { 32956583fe22SAl Viro path_to_nameidata(&path, nd); 32966583fe22SAl Viro return -ENOENT; 32976583fe22SAl Viro } 32986583fe22SAl Viro 32996583fe22SAl Viro /* 33006583fe22SAl Viro * create/update audit record if it already exists. 33016583fe22SAl Viro */ 33026583fe22SAl Viro audit_inode(nd->name, path.dentry, 0); 33036583fe22SAl Viro 3304deb106c6SAl Viro if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) { 3305deb106c6SAl Viro path_to_nameidata(&path, nd); 3306deb106c6SAl Viro return -EEXIST; 3307deb106c6SAl Viro } 3308fb1cc555SAl Viro 3309254cf582SAl Viro seq = 0; /* out of RCU mode, so the value doesn't matter */ 3310d4565649SAl Viro inode = d_backing_inode(path.dentry); 3311766c4cbfSAl Viro finish_lookup: 33128f64fb1cSAl Viro error = step_into(nd, &path, 0, inode, seq); 3313deb106c6SAl Viro if (unlikely(error)) 3314d63ff28fSAl Viro return error; 3315bc77daa7SAl Viro finish_open: 33168f64fb1cSAl Viro /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ 3317a3fbbde7SAl Viro error = complete_walk(nd); 3318fac7d191SAl Viro if (error) 33192675a4ebSAl Viro return error; 332076ae2a5aSAl Viro audit_inode(nd->name, nd->path.dentry, 0); 3321fb1cc555SAl Viro error = -EISDIR; 332244b1d530SMiklos Szeredi if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry)) 33232675a4ebSAl Viro goto out; 3324af2f5542SMiklos Szeredi error = -ENOTDIR; 332544b1d530SMiklos Szeredi if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry)) 33262675a4ebSAl Viro goto out; 33274bbcbd3bSDavid Howells if (!d_is_reg(nd->path.dentry)) 332877d660a8SMiklos Szeredi will_truncate = false; 33296c0d46c4SAl Viro 33300f9d1a10SAl Viro if (will_truncate) { 33310f9d1a10SAl Viro error = mnt_want_write(nd->path.mnt); 33320f9d1a10SAl Viro if (error) 33332675a4ebSAl Viro goto out; 333464894cf8SAl Viro got_write = true; 33350f9d1a10SAl Viro } 3336e83db167SMiklos Szeredi finish_open_created: 3337bcda7652SAl Viro error = may_open(&nd->path, acc_mode, open_flag); 3338ca344a89SAl Viro if (error) 33392675a4ebSAl Viro goto out; 33404aa7c634SMiklos Szeredi BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */ 33414aa7c634SMiklos Szeredi error = vfs_open(&nd->path, file, current_cred()); 3342fac7d191SAl Viro if (error) 3343015c3bbcSMiklos Szeredi goto out; 3344fac7d191SAl Viro *opened |= FILE_OPENED; 3345a8277b9bSMiklos Szeredi opened: 33462675a4ebSAl Viro error = open_check_o_direct(file); 3347fe9ec829SAl Viro if (!error) 33483034a146SDmitry Kasatkin error = ima_file_check(file, op->acc_mode, *opened); 3349fe9ec829SAl Viro if (!error && will_truncate) 33502675a4ebSAl Viro error = handle_truncate(file); 3351ca344a89SAl Viro out: 3352fe9ec829SAl Viro if (unlikely(error) && (*opened & FILE_OPENED)) 3353fe9ec829SAl Viro fput(file); 3354c80567c8SAl Viro if (unlikely(error > 0)) { 3355c80567c8SAl Viro WARN_ON(1); 3356c80567c8SAl Viro error = -EINVAL; 3357c80567c8SAl Viro } 335864894cf8SAl Viro if (got_write) 33590f9d1a10SAl Viro mnt_drop_write(nd->path.mnt); 33602675a4ebSAl Viro return error; 3361fb1cc555SAl Viro } 3362fb1cc555SAl Viro 3363af7bd4dcSAmir Goldstein struct dentry *vfs_tmpfile(struct dentry *dentry, umode_t mode, int open_flag) 3364af7bd4dcSAmir Goldstein { 3365af7bd4dcSAmir Goldstein static const struct qstr name = QSTR_INIT("/", 1); 3366af7bd4dcSAmir Goldstein struct dentry *child = NULL; 3367af7bd4dcSAmir Goldstein struct inode *dir = dentry->d_inode; 3368af7bd4dcSAmir Goldstein struct inode *inode; 3369af7bd4dcSAmir Goldstein int error; 3370af7bd4dcSAmir Goldstein 3371af7bd4dcSAmir Goldstein /* we want directory to be writable */ 3372af7bd4dcSAmir Goldstein error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 3373af7bd4dcSAmir Goldstein if (error) 3374af7bd4dcSAmir Goldstein goto out_err; 3375af7bd4dcSAmir Goldstein error = -EOPNOTSUPP; 3376af7bd4dcSAmir Goldstein if (!dir->i_op->tmpfile) 3377af7bd4dcSAmir Goldstein goto out_err; 3378af7bd4dcSAmir Goldstein error = -ENOMEM; 3379af7bd4dcSAmir Goldstein child = d_alloc(dentry, &name); 3380af7bd4dcSAmir Goldstein if (unlikely(!child)) 3381af7bd4dcSAmir Goldstein goto out_err; 3382af7bd4dcSAmir Goldstein error = dir->i_op->tmpfile(dir, child, mode); 3383af7bd4dcSAmir Goldstein if (error) 3384af7bd4dcSAmir Goldstein goto out_err; 3385af7bd4dcSAmir Goldstein error = -ENOENT; 3386af7bd4dcSAmir Goldstein inode = child->d_inode; 3387af7bd4dcSAmir Goldstein if (unlikely(!inode)) 3388af7bd4dcSAmir Goldstein goto out_err; 3389af7bd4dcSAmir Goldstein if (!(open_flag & O_EXCL)) { 3390af7bd4dcSAmir Goldstein spin_lock(&inode->i_lock); 3391af7bd4dcSAmir Goldstein inode->i_state |= I_LINKABLE; 3392af7bd4dcSAmir Goldstein spin_unlock(&inode->i_lock); 3393af7bd4dcSAmir Goldstein } 3394af7bd4dcSAmir Goldstein return child; 3395af7bd4dcSAmir Goldstein 3396af7bd4dcSAmir Goldstein out_err: 3397af7bd4dcSAmir Goldstein dput(child); 3398af7bd4dcSAmir Goldstein return ERR_PTR(error); 3399af7bd4dcSAmir Goldstein } 3400af7bd4dcSAmir Goldstein EXPORT_SYMBOL(vfs_tmpfile); 3401af7bd4dcSAmir Goldstein 3402c8a53ee5SAl Viro static int do_tmpfile(struct nameidata *nd, unsigned flags, 340360545d0dSAl Viro const struct open_flags *op, 340460545d0dSAl Viro struct file *file, int *opened) 340560545d0dSAl Viro { 3406625b6d10SAl Viro struct dentry *child; 3407625b6d10SAl Viro struct path path; 3408c8a53ee5SAl Viro int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path); 340960545d0dSAl Viro if (unlikely(error)) 341060545d0dSAl Viro return error; 3411625b6d10SAl Viro error = mnt_want_write(path.mnt); 341260545d0dSAl Viro if (unlikely(error)) 341360545d0dSAl Viro goto out; 3414af7bd4dcSAmir Goldstein child = vfs_tmpfile(path.dentry, op->mode, op->open_flag); 3415af7bd4dcSAmir Goldstein error = PTR_ERR(child); 3416af7bd4dcSAmir Goldstein if (unlikely(IS_ERR(child))) 341760545d0dSAl Viro goto out2; 3418625b6d10SAl Viro dput(path.dentry); 3419625b6d10SAl Viro path.dentry = child; 3420c8a53ee5SAl Viro audit_inode(nd->name, child, 0); 342169a91c23SEric Rannaud /* Don't check for other permissions, the inode was just created */ 342262fb4a15SAl Viro error = may_open(&path, 0, op->open_flag); 342360545d0dSAl Viro if (error) 342460545d0dSAl Viro goto out2; 3425625b6d10SAl Viro file->f_path.mnt = path.mnt; 3426625b6d10SAl Viro error = finish_open(file, child, NULL, opened); 342760545d0dSAl Viro if (error) 342860545d0dSAl Viro goto out2; 342960545d0dSAl Viro error = open_check_o_direct(file); 3430af7bd4dcSAmir Goldstein if (error) 343160545d0dSAl Viro fput(file); 343260545d0dSAl Viro out2: 3433625b6d10SAl Viro mnt_drop_write(path.mnt); 343460545d0dSAl Viro out: 3435625b6d10SAl Viro path_put(&path); 343660545d0dSAl Viro return error; 343760545d0dSAl Viro } 343860545d0dSAl Viro 34396ac08709SAl Viro static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file) 34406ac08709SAl Viro { 34416ac08709SAl Viro struct path path; 34426ac08709SAl Viro int error = path_lookupat(nd, flags, &path); 34436ac08709SAl Viro if (!error) { 34446ac08709SAl Viro audit_inode(nd->name, path.dentry, 0); 34456ac08709SAl Viro error = vfs_open(&path, file, current_cred()); 34466ac08709SAl Viro path_put(&path); 34476ac08709SAl Viro } 34486ac08709SAl Viro return error; 34496ac08709SAl Viro } 34506ac08709SAl Viro 3451c8a53ee5SAl Viro static struct file *path_openat(struct nameidata *nd, 3452c8a53ee5SAl Viro const struct open_flags *op, unsigned flags) 34531da177e4SLinus Torvalds { 3454368ee9baSAl Viro const char *s; 345530d90494SAl Viro struct file *file; 345647237687SAl Viro int opened = 0; 345713aab428SAl Viro int error; 345831e6b01fSNick Piggin 345930d90494SAl Viro file = get_empty_filp(); 34601afc99beSAl Viro if (IS_ERR(file)) 34611afc99beSAl Viro return file; 346231e6b01fSNick Piggin 346330d90494SAl Viro file->f_flags = op->open_flag; 346431e6b01fSNick Piggin 3465bb458c64SAl Viro if (unlikely(file->f_flags & __O_TMPFILE)) { 3466c8a53ee5SAl Viro error = do_tmpfile(nd, flags, op, file, &opened); 3467f15133dfSAl Viro goto out2; 346860545d0dSAl Viro } 346960545d0dSAl Viro 34706ac08709SAl Viro if (unlikely(file->f_flags & O_PATH)) { 34716ac08709SAl Viro error = do_o_path(nd, flags, file); 34726ac08709SAl Viro if (!error) 34736ac08709SAl Viro opened |= FILE_OPENED; 34746ac08709SAl Viro goto out2; 34756ac08709SAl Viro } 34766ac08709SAl Viro 3477c8a53ee5SAl Viro s = path_init(nd, flags); 3478368ee9baSAl Viro if (IS_ERR(s)) { 3479368ee9baSAl Viro put_filp(file); 3480368ee9baSAl Viro return ERR_CAST(s); 3481368ee9baSAl Viro } 34823bdba28bSAl Viro while (!(error = link_path_walk(s, nd)) && 348376ae2a5aSAl Viro (error = do_last(nd, file, op, &opened)) > 0) { 348473d049a4SAl Viro nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); 34853bdba28bSAl Viro s = trailing_symlink(nd); 34863bdba28bSAl Viro if (IS_ERR(s)) { 34873bdba28bSAl Viro error = PTR_ERR(s); 34882675a4ebSAl Viro break; 3489806b681cSAl Viro } 34903bdba28bSAl Viro } 3491deb106c6SAl Viro terminate_walk(nd); 3492f15133dfSAl Viro out2: 34932675a4ebSAl Viro if (!(opened & FILE_OPENED)) { 34942675a4ebSAl Viro BUG_ON(!error); 349530d90494SAl Viro put_filp(file); 3496015c3bbcSMiklos Szeredi } 34972675a4ebSAl Viro if (unlikely(error)) { 34982675a4ebSAl Viro if (error == -EOPENSTALE) { 34992675a4ebSAl Viro if (flags & LOOKUP_RCU) 35002675a4ebSAl Viro error = -ECHILD; 35012675a4ebSAl Viro else 35022675a4ebSAl Viro error = -ESTALE; 35032675a4ebSAl Viro } 35042675a4ebSAl Viro file = ERR_PTR(error); 35052675a4ebSAl Viro } 35062675a4ebSAl Viro return file; 3507de459215SKirill Korotaev } 35081da177e4SLinus Torvalds 3509669abf4eSJeff Layton struct file *do_filp_open(int dfd, struct filename *pathname, 3510f9652e10SAl Viro const struct open_flags *op) 351113aab428SAl Viro { 35129883d185SAl Viro struct nameidata nd; 3513f9652e10SAl Viro int flags = op->lookup_flags; 351413aab428SAl Viro struct file *filp; 351513aab428SAl Viro 35169883d185SAl Viro set_nameidata(&nd, dfd, pathname); 3517c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_RCU); 351813aab428SAl Viro if (unlikely(filp == ERR_PTR(-ECHILD))) 3519c8a53ee5SAl Viro filp = path_openat(&nd, op, flags); 352013aab428SAl Viro if (unlikely(filp == ERR_PTR(-ESTALE))) 3521c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_REVAL); 35229883d185SAl Viro restore_nameidata(); 352313aab428SAl Viro return filp; 352413aab428SAl Viro } 352513aab428SAl Viro 352673d049a4SAl Viro struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, 3527f9652e10SAl Viro const char *name, const struct open_flags *op) 352873d049a4SAl Viro { 35299883d185SAl Viro struct nameidata nd; 353073d049a4SAl Viro struct file *file; 353151689104SPaul Moore struct filename *filename; 3532f9652e10SAl Viro int flags = op->lookup_flags | LOOKUP_ROOT; 353373d049a4SAl Viro 353473d049a4SAl Viro nd.root.mnt = mnt; 353573d049a4SAl Viro nd.root.dentry = dentry; 353673d049a4SAl Viro 3537b18825a7SDavid Howells if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN) 353873d049a4SAl Viro return ERR_PTR(-ELOOP); 353973d049a4SAl Viro 354051689104SPaul Moore filename = getname_kernel(name); 3541a1c83681SViresh Kumar if (IS_ERR(filename)) 354251689104SPaul Moore return ERR_CAST(filename); 354351689104SPaul Moore 35449883d185SAl Viro set_nameidata(&nd, -1, filename); 3545c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_RCU); 354673d049a4SAl Viro if (unlikely(file == ERR_PTR(-ECHILD))) 3547c8a53ee5SAl Viro file = path_openat(&nd, op, flags); 354873d049a4SAl Viro if (unlikely(file == ERR_PTR(-ESTALE))) 3549c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_REVAL); 35509883d185SAl Viro restore_nameidata(); 355151689104SPaul Moore putname(filename); 355273d049a4SAl Viro return file; 355373d049a4SAl Viro } 355473d049a4SAl Viro 3555fa14a0b8SAl Viro static struct dentry *filename_create(int dfd, struct filename *name, 35561ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 35571da177e4SLinus Torvalds { 3558c663e5d8SChristoph Hellwig struct dentry *dentry = ERR_PTR(-EEXIST); 3559391172c4SAl Viro struct qstr last; 3560391172c4SAl Viro int type; 3561c30dabfeSJan Kara int err2; 35621ac12b4bSJeff Layton int error; 35631ac12b4bSJeff Layton bool is_dir = (lookup_flags & LOOKUP_DIRECTORY); 35641ac12b4bSJeff Layton 35651ac12b4bSJeff Layton /* 35661ac12b4bSJeff Layton * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any 35671ac12b4bSJeff Layton * other flags passed in are ignored! 35681ac12b4bSJeff Layton */ 35691ac12b4bSJeff Layton lookup_flags &= LOOKUP_REVAL; 35701ac12b4bSJeff Layton 35715c31b6ceSAl Viro name = filename_parentat(dfd, name, lookup_flags, path, &last, &type); 35725c31b6ceSAl Viro if (IS_ERR(name)) 35735c31b6ceSAl Viro return ERR_CAST(name); 35741da177e4SLinus Torvalds 3575c663e5d8SChristoph Hellwig /* 3576c663e5d8SChristoph Hellwig * Yucky last component or no last component at all? 3577c663e5d8SChristoph Hellwig * (foo/., foo/.., /////) 3578c663e5d8SChristoph Hellwig */ 35795c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) 3580ed75e95dSAl Viro goto out; 3581c663e5d8SChristoph Hellwig 3582c30dabfeSJan Kara /* don't fail immediately if it's r/o, at least try to report other errors */ 3583391172c4SAl Viro err2 = mnt_want_write(path->mnt); 3584c663e5d8SChristoph Hellwig /* 3585c663e5d8SChristoph Hellwig * Do the final lookup. 3586c663e5d8SChristoph Hellwig */ 3587391172c4SAl Viro lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL; 35885955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 3589391172c4SAl Viro dentry = __lookup_hash(&last, path->dentry, lookup_flags); 35901da177e4SLinus Torvalds if (IS_ERR(dentry)) 3591a8104a9fSAl Viro goto unlock; 3592c663e5d8SChristoph Hellwig 3593a8104a9fSAl Viro error = -EEXIST; 3594b18825a7SDavid Howells if (d_is_positive(dentry)) 3595a8104a9fSAl Viro goto fail; 3596b18825a7SDavid Howells 3597c663e5d8SChristoph Hellwig /* 3598c663e5d8SChristoph Hellwig * Special case - lookup gave negative, but... we had foo/bar/ 3599c663e5d8SChristoph Hellwig * From the vfs_mknod() POV we just have a negative dentry - 3600c663e5d8SChristoph Hellwig * all is fine. Let's be bastards - you had / on the end, you've 3601c663e5d8SChristoph Hellwig * been asking for (non-existent) directory. -ENOENT for you. 3602c663e5d8SChristoph Hellwig */ 3603391172c4SAl Viro if (unlikely(!is_dir && last.name[last.len])) { 3604a8104a9fSAl Viro error = -ENOENT; 3605ed75e95dSAl Viro goto fail; 3606e9baf6e5SAl Viro } 3607c30dabfeSJan Kara if (unlikely(err2)) { 3608c30dabfeSJan Kara error = err2; 3609a8104a9fSAl Viro goto fail; 3610c30dabfeSJan Kara } 3611181c37b6SAl Viro putname(name); 3612e9baf6e5SAl Viro return dentry; 36131da177e4SLinus Torvalds fail: 3614a8104a9fSAl Viro dput(dentry); 3615a8104a9fSAl Viro dentry = ERR_PTR(error); 3616a8104a9fSAl Viro unlock: 36175955102cSAl Viro inode_unlock(path->dentry->d_inode); 3618c30dabfeSJan Kara if (!err2) 3619391172c4SAl Viro mnt_drop_write(path->mnt); 3620ed75e95dSAl Viro out: 3621391172c4SAl Viro path_put(path); 3622181c37b6SAl Viro putname(name); 3623ed75e95dSAl Viro return dentry; 3624dae6ad8fSAl Viro } 3625fa14a0b8SAl Viro 3626fa14a0b8SAl Viro struct dentry *kern_path_create(int dfd, const char *pathname, 3627fa14a0b8SAl Viro struct path *path, unsigned int lookup_flags) 3628fa14a0b8SAl Viro { 3629181c37b6SAl Viro return filename_create(dfd, getname_kernel(pathname), 3630181c37b6SAl Viro path, lookup_flags); 3631fa14a0b8SAl Viro } 3632dae6ad8fSAl Viro EXPORT_SYMBOL(kern_path_create); 3633dae6ad8fSAl Viro 3634921a1650SAl Viro void done_path_create(struct path *path, struct dentry *dentry) 3635921a1650SAl Viro { 3636921a1650SAl Viro dput(dentry); 36375955102cSAl Viro inode_unlock(path->dentry->d_inode); 3638a8104a9fSAl Viro mnt_drop_write(path->mnt); 3639921a1650SAl Viro path_put(path); 3640921a1650SAl Viro } 3641921a1650SAl Viro EXPORT_SYMBOL(done_path_create); 3642921a1650SAl Viro 3643520ae687SAl Viro inline struct dentry *user_path_create(int dfd, const char __user *pathname, 36441ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 3645dae6ad8fSAl Viro { 3646181c37b6SAl Viro return filename_create(dfd, getname(pathname), path, lookup_flags); 3647dae6ad8fSAl Viro } 3648dae6ad8fSAl Viro EXPORT_SYMBOL(user_path_create); 3649dae6ad8fSAl Viro 36501a67aafbSAl Viro int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) 36511da177e4SLinus Torvalds { 3652a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 36531da177e4SLinus Torvalds 36541da177e4SLinus Torvalds if (error) 36551da177e4SLinus Torvalds return error; 36561da177e4SLinus Torvalds 3657975d6b39SEric W. Biederman if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) 36581da177e4SLinus Torvalds return -EPERM; 36591da177e4SLinus Torvalds 3660acfa4380SAl Viro if (!dir->i_op->mknod) 36611da177e4SLinus Torvalds return -EPERM; 36621da177e4SLinus Torvalds 366308ce5f16SSerge E. Hallyn error = devcgroup_inode_mknod(mode, dev); 366408ce5f16SSerge E. Hallyn if (error) 366508ce5f16SSerge E. Hallyn return error; 366608ce5f16SSerge E. Hallyn 36671da177e4SLinus Torvalds error = security_inode_mknod(dir, dentry, mode, dev); 36681da177e4SLinus Torvalds if (error) 36691da177e4SLinus Torvalds return error; 36701da177e4SLinus Torvalds 36711da177e4SLinus Torvalds error = dir->i_op->mknod(dir, dentry, mode, dev); 3672a74574aaSStephen Smalley if (!error) 3673f38aa942SAmy Griffis fsnotify_create(dir, dentry); 36741da177e4SLinus Torvalds return error; 36751da177e4SLinus Torvalds } 36764d359507SAl Viro EXPORT_SYMBOL(vfs_mknod); 36771da177e4SLinus Torvalds 3678f69aac00SAl Viro static int may_mknod(umode_t mode) 3679463c3197SDave Hansen { 3680463c3197SDave Hansen switch (mode & S_IFMT) { 3681463c3197SDave Hansen case S_IFREG: 3682463c3197SDave Hansen case S_IFCHR: 3683463c3197SDave Hansen case S_IFBLK: 3684463c3197SDave Hansen case S_IFIFO: 3685463c3197SDave Hansen case S_IFSOCK: 3686463c3197SDave Hansen case 0: /* zero mode translates to S_IFREG */ 3687463c3197SDave Hansen return 0; 3688463c3197SDave Hansen case S_IFDIR: 3689463c3197SDave Hansen return -EPERM; 3690463c3197SDave Hansen default: 3691463c3197SDave Hansen return -EINVAL; 3692463c3197SDave Hansen } 3693463c3197SDave Hansen } 3694463c3197SDave Hansen 36958208a22bSAl Viro SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, 36962e4d0924SHeiko Carstens unsigned, dev) 36971da177e4SLinus Torvalds { 36981da177e4SLinus Torvalds struct dentry *dentry; 3699dae6ad8fSAl Viro struct path path; 3700dae6ad8fSAl Viro int error; 3701972567f1SJeff Layton unsigned int lookup_flags = 0; 37021da177e4SLinus Torvalds 37038e4bfca1SAl Viro error = may_mknod(mode); 37048e4bfca1SAl Viro if (error) 37058e4bfca1SAl Viro return error; 3706972567f1SJeff Layton retry: 3707972567f1SJeff Layton dentry = user_path_create(dfd, filename, &path, lookup_flags); 3708dae6ad8fSAl Viro if (IS_ERR(dentry)) 3709dae6ad8fSAl Viro return PTR_ERR(dentry); 37102ad94ae6SAl Viro 3711dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3712ce3b0f8dSAl Viro mode &= ~current_umask(); 3713dae6ad8fSAl Viro error = security_path_mknod(&path, dentry, mode, dev); 3714be6d3e56SKentaro Takeda if (error) 3715a8104a9fSAl Viro goto out; 37161da177e4SLinus Torvalds switch (mode & S_IFMT) { 37171da177e4SLinus Torvalds case 0: case S_IFREG: 3718312b63fbSAl Viro error = vfs_create(path.dentry->d_inode,dentry,mode,true); 371905d1a717SMimi Zohar if (!error) 372005d1a717SMimi Zohar ima_post_path_mknod(dentry); 37211da177e4SLinus Torvalds break; 37221da177e4SLinus Torvalds case S_IFCHR: case S_IFBLK: 3723dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode, 37241da177e4SLinus Torvalds new_decode_dev(dev)); 37251da177e4SLinus Torvalds break; 37261da177e4SLinus Torvalds case S_IFIFO: case S_IFSOCK: 3727dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); 37281da177e4SLinus Torvalds break; 37291da177e4SLinus Torvalds } 3730a8104a9fSAl Viro out: 3731921a1650SAl Viro done_path_create(&path, dentry); 3732972567f1SJeff Layton if (retry_estale(error, lookup_flags)) { 3733972567f1SJeff Layton lookup_flags |= LOOKUP_REVAL; 3734972567f1SJeff Layton goto retry; 3735972567f1SJeff Layton } 37361da177e4SLinus Torvalds return error; 37371da177e4SLinus Torvalds } 37381da177e4SLinus Torvalds 37398208a22bSAl Viro SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) 37405590ff0dSUlrich Drepper { 37415590ff0dSUlrich Drepper return sys_mknodat(AT_FDCWD, filename, mode, dev); 37425590ff0dSUlrich Drepper } 37435590ff0dSUlrich Drepper 374418bb1db3SAl Viro int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 37451da177e4SLinus Torvalds { 3746a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 37478de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 37481da177e4SLinus Torvalds 37491da177e4SLinus Torvalds if (error) 37501da177e4SLinus Torvalds return error; 37511da177e4SLinus Torvalds 3752acfa4380SAl Viro if (!dir->i_op->mkdir) 37531da177e4SLinus Torvalds return -EPERM; 37541da177e4SLinus Torvalds 37551da177e4SLinus Torvalds mode &= (S_IRWXUGO|S_ISVTX); 37561da177e4SLinus Torvalds error = security_inode_mkdir(dir, dentry, mode); 37571da177e4SLinus Torvalds if (error) 37581da177e4SLinus Torvalds return error; 37591da177e4SLinus Torvalds 37608de52778SAl Viro if (max_links && dir->i_nlink >= max_links) 37618de52778SAl Viro return -EMLINK; 37628de52778SAl Viro 37631da177e4SLinus Torvalds error = dir->i_op->mkdir(dir, dentry, mode); 3764a74574aaSStephen Smalley if (!error) 3765f38aa942SAmy Griffis fsnotify_mkdir(dir, dentry); 37661da177e4SLinus Torvalds return error; 37671da177e4SLinus Torvalds } 37684d359507SAl Viro EXPORT_SYMBOL(vfs_mkdir); 37691da177e4SLinus Torvalds 3770a218d0fdSAl Viro SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) 37711da177e4SLinus Torvalds { 37726902d925SDave Hansen struct dentry *dentry; 3773dae6ad8fSAl Viro struct path path; 3774dae6ad8fSAl Viro int error; 3775b76d8b82SJeff Layton unsigned int lookup_flags = LOOKUP_DIRECTORY; 37761da177e4SLinus Torvalds 3777b76d8b82SJeff Layton retry: 3778b76d8b82SJeff Layton dentry = user_path_create(dfd, pathname, &path, lookup_flags); 37796902d925SDave Hansen if (IS_ERR(dentry)) 3780dae6ad8fSAl Viro return PTR_ERR(dentry); 37816902d925SDave Hansen 3782dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3783ce3b0f8dSAl Viro mode &= ~current_umask(); 3784dae6ad8fSAl Viro error = security_path_mkdir(&path, dentry, mode); 3785a8104a9fSAl Viro if (!error) 3786dae6ad8fSAl Viro error = vfs_mkdir(path.dentry->d_inode, dentry, mode); 3787921a1650SAl Viro done_path_create(&path, dentry); 3788b76d8b82SJeff Layton if (retry_estale(error, lookup_flags)) { 3789b76d8b82SJeff Layton lookup_flags |= LOOKUP_REVAL; 3790b76d8b82SJeff Layton goto retry; 3791b76d8b82SJeff Layton } 37921da177e4SLinus Torvalds return error; 37931da177e4SLinus Torvalds } 37941da177e4SLinus Torvalds 3795a218d0fdSAl Viro SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) 37965590ff0dSUlrich Drepper { 37975590ff0dSUlrich Drepper return sys_mkdirat(AT_FDCWD, pathname, mode); 37985590ff0dSUlrich Drepper } 37995590ff0dSUlrich Drepper 38001da177e4SLinus Torvalds int vfs_rmdir(struct inode *dir, struct dentry *dentry) 38011da177e4SLinus Torvalds { 38021da177e4SLinus Torvalds int error = may_delete(dir, dentry, 1); 38031da177e4SLinus Torvalds 38041da177e4SLinus Torvalds if (error) 38051da177e4SLinus Torvalds return error; 38061da177e4SLinus Torvalds 3807acfa4380SAl Viro if (!dir->i_op->rmdir) 38081da177e4SLinus Torvalds return -EPERM; 38091da177e4SLinus Torvalds 38101d2ef590SAl Viro dget(dentry); 38115955102cSAl Viro inode_lock(dentry->d_inode); 3812912dbc15SSage Weil 38131da177e4SLinus Torvalds error = -EBUSY; 38147af1364fSEric W. Biederman if (is_local_mountpoint(dentry)) 3815912dbc15SSage Weil goto out; 3816912dbc15SSage Weil 38171da177e4SLinus Torvalds error = security_inode_rmdir(dir, dentry); 3818912dbc15SSage Weil if (error) 3819912dbc15SSage Weil goto out; 3820912dbc15SSage Weil 38213cebde24SSage Weil shrink_dcache_parent(dentry); 38221da177e4SLinus Torvalds error = dir->i_op->rmdir(dir, dentry); 3823912dbc15SSage Weil if (error) 3824912dbc15SSage Weil goto out; 3825912dbc15SSage Weil 38261da177e4SLinus Torvalds dentry->d_inode->i_flags |= S_DEAD; 3827d83c49f3SAl Viro dont_mount(dentry); 38288ed936b5SEric W. Biederman detach_mounts(dentry); 38291da177e4SLinus Torvalds 3830912dbc15SSage Weil out: 38315955102cSAl Viro inode_unlock(dentry->d_inode); 38321d2ef590SAl Viro dput(dentry); 3833912dbc15SSage Weil if (!error) 3834912dbc15SSage Weil d_delete(dentry); 38351da177e4SLinus Torvalds return error; 38361da177e4SLinus Torvalds } 38374d359507SAl Viro EXPORT_SYMBOL(vfs_rmdir); 38381da177e4SLinus Torvalds 38395590ff0dSUlrich Drepper static long do_rmdir(int dfd, const char __user *pathname) 38401da177e4SLinus Torvalds { 38411da177e4SLinus Torvalds int error = 0; 384291a27b2aSJeff Layton struct filename *name; 38431da177e4SLinus Torvalds struct dentry *dentry; 3844f5beed75SAl Viro struct path path; 3845f5beed75SAl Viro struct qstr last; 3846f5beed75SAl Viro int type; 3847c6ee9206SJeff Layton unsigned int lookup_flags = 0; 3848c6ee9206SJeff Layton retry: 3849c1d4dd27SAl Viro name = filename_parentat(dfd, getname(pathname), lookup_flags, 3850c1d4dd27SAl Viro &path, &last, &type); 385191a27b2aSJeff Layton if (IS_ERR(name)) 385291a27b2aSJeff Layton return PTR_ERR(name); 38531da177e4SLinus Torvalds 3854f5beed75SAl Viro switch (type) { 38551da177e4SLinus Torvalds case LAST_DOTDOT: 38561da177e4SLinus Torvalds error = -ENOTEMPTY; 38571da177e4SLinus Torvalds goto exit1; 38581da177e4SLinus Torvalds case LAST_DOT: 38591da177e4SLinus Torvalds error = -EINVAL; 38601da177e4SLinus Torvalds goto exit1; 38611da177e4SLinus Torvalds case LAST_ROOT: 38621da177e4SLinus Torvalds error = -EBUSY; 38631da177e4SLinus Torvalds goto exit1; 38641da177e4SLinus Torvalds } 38650612d9fbSOGAWA Hirofumi 3866f5beed75SAl Viro error = mnt_want_write(path.mnt); 3867c30dabfeSJan Kara if (error) 3868c30dabfeSJan Kara goto exit1; 38690612d9fbSOGAWA Hirofumi 38705955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 3871f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 38721da177e4SLinus Torvalds error = PTR_ERR(dentry); 38736902d925SDave Hansen if (IS_ERR(dentry)) 38746902d925SDave Hansen goto exit2; 3875e6bc45d6STheodore Ts'o if (!dentry->d_inode) { 3876e6bc45d6STheodore Ts'o error = -ENOENT; 3877e6bc45d6STheodore Ts'o goto exit3; 3878e6bc45d6STheodore Ts'o } 3879f5beed75SAl Viro error = security_path_rmdir(&path, dentry); 3880be6d3e56SKentaro Takeda if (error) 3881c30dabfeSJan Kara goto exit3; 3882f5beed75SAl Viro error = vfs_rmdir(path.dentry->d_inode, dentry); 38830622753bSDave Hansen exit3: 38841da177e4SLinus Torvalds dput(dentry); 38856902d925SDave Hansen exit2: 38865955102cSAl Viro inode_unlock(path.dentry->d_inode); 3887f5beed75SAl Viro mnt_drop_write(path.mnt); 38881da177e4SLinus Torvalds exit1: 3889f5beed75SAl Viro path_put(&path); 38901da177e4SLinus Torvalds putname(name); 3891c6ee9206SJeff Layton if (retry_estale(error, lookup_flags)) { 3892c6ee9206SJeff Layton lookup_flags |= LOOKUP_REVAL; 3893c6ee9206SJeff Layton goto retry; 3894c6ee9206SJeff Layton } 38951da177e4SLinus Torvalds return error; 38961da177e4SLinus Torvalds } 38971da177e4SLinus Torvalds 38983cdad428SHeiko Carstens SYSCALL_DEFINE1(rmdir, const char __user *, pathname) 38995590ff0dSUlrich Drepper { 39005590ff0dSUlrich Drepper return do_rmdir(AT_FDCWD, pathname); 39015590ff0dSUlrich Drepper } 39025590ff0dSUlrich Drepper 3903b21996e3SJ. Bruce Fields /** 3904b21996e3SJ. Bruce Fields * vfs_unlink - unlink a filesystem object 3905b21996e3SJ. Bruce Fields * @dir: parent directory 3906b21996e3SJ. Bruce Fields * @dentry: victim 3907b21996e3SJ. Bruce Fields * @delegated_inode: returns victim inode, if the inode is delegated. 3908b21996e3SJ. Bruce Fields * 3909b21996e3SJ. Bruce Fields * The caller must hold dir->i_mutex. 3910b21996e3SJ. Bruce Fields * 3911b21996e3SJ. Bruce Fields * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and 3912b21996e3SJ. Bruce Fields * return a reference to the inode in delegated_inode. The caller 3913b21996e3SJ. Bruce Fields * should then break the delegation on that inode and retry. Because 3914b21996e3SJ. Bruce Fields * breaking a delegation may take a long time, the caller should drop 3915b21996e3SJ. Bruce Fields * dir->i_mutex before doing so. 3916b21996e3SJ. Bruce Fields * 3917b21996e3SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 3918b21996e3SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 3919b21996e3SJ. Bruce Fields * to be NFS exported. 3920b21996e3SJ. Bruce Fields */ 3921b21996e3SJ. Bruce Fields int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode) 39221da177e4SLinus Torvalds { 39239accbb97SJ. Bruce Fields struct inode *target = dentry->d_inode; 39241da177e4SLinus Torvalds int error = may_delete(dir, dentry, 0); 39251da177e4SLinus Torvalds 39261da177e4SLinus Torvalds if (error) 39271da177e4SLinus Torvalds return error; 39281da177e4SLinus Torvalds 3929acfa4380SAl Viro if (!dir->i_op->unlink) 39301da177e4SLinus Torvalds return -EPERM; 39311da177e4SLinus Torvalds 39325955102cSAl Viro inode_lock(target); 39338ed936b5SEric W. Biederman if (is_local_mountpoint(dentry)) 39341da177e4SLinus Torvalds error = -EBUSY; 39351da177e4SLinus Torvalds else { 39361da177e4SLinus Torvalds error = security_inode_unlink(dir, dentry); 3937bec1052eSAl Viro if (!error) { 39385a14696cSJ. Bruce Fields error = try_break_deleg(target, delegated_inode); 39395a14696cSJ. Bruce Fields if (error) 3940b21996e3SJ. Bruce Fields goto out; 39411da177e4SLinus Torvalds error = dir->i_op->unlink(dir, dentry); 39428ed936b5SEric W. Biederman if (!error) { 3943d83c49f3SAl Viro dont_mount(dentry); 39448ed936b5SEric W. Biederman detach_mounts(dentry); 39458ed936b5SEric W. Biederman } 3946bec1052eSAl Viro } 39471da177e4SLinus Torvalds } 3948b21996e3SJ. Bruce Fields out: 39495955102cSAl Viro inode_unlock(target); 39501da177e4SLinus Torvalds 39511da177e4SLinus Torvalds /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 39521da177e4SLinus Torvalds if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { 39539accbb97SJ. Bruce Fields fsnotify_link_count(target); 39541da177e4SLinus Torvalds d_delete(dentry); 39551da177e4SLinus Torvalds } 39560eeca283SRobert Love 39571da177e4SLinus Torvalds return error; 39581da177e4SLinus Torvalds } 39594d359507SAl Viro EXPORT_SYMBOL(vfs_unlink); 39601da177e4SLinus Torvalds 39611da177e4SLinus Torvalds /* 39621da177e4SLinus Torvalds * Make sure that the actual truncation of the file will occur outside its 39631b1dcc1bSJes Sorensen * directory's i_mutex. Truncate can take a long time if there is a lot of 39641da177e4SLinus Torvalds * writeout happening, and we don't want to prevent access to the directory 39651da177e4SLinus Torvalds * while waiting on the I/O. 39661da177e4SLinus Torvalds */ 39675590ff0dSUlrich Drepper static long do_unlinkat(int dfd, const char __user *pathname) 39681da177e4SLinus Torvalds { 39692ad94ae6SAl Viro int error; 397091a27b2aSJeff Layton struct filename *name; 39711da177e4SLinus Torvalds struct dentry *dentry; 3972f5beed75SAl Viro struct path path; 3973f5beed75SAl Viro struct qstr last; 3974f5beed75SAl Viro int type; 39751da177e4SLinus Torvalds struct inode *inode = NULL; 3976b21996e3SJ. Bruce Fields struct inode *delegated_inode = NULL; 39775d18f813SJeff Layton unsigned int lookup_flags = 0; 39785d18f813SJeff Layton retry: 3979c1d4dd27SAl Viro name = filename_parentat(dfd, getname(pathname), lookup_flags, 3980c1d4dd27SAl Viro &path, &last, &type); 398191a27b2aSJeff Layton if (IS_ERR(name)) 398291a27b2aSJeff Layton return PTR_ERR(name); 39832ad94ae6SAl Viro 39841da177e4SLinus Torvalds error = -EISDIR; 3985f5beed75SAl Viro if (type != LAST_NORM) 39861da177e4SLinus Torvalds goto exit1; 39870612d9fbSOGAWA Hirofumi 3988f5beed75SAl Viro error = mnt_want_write(path.mnt); 3989c30dabfeSJan Kara if (error) 3990c30dabfeSJan Kara goto exit1; 3991b21996e3SJ. Bruce Fields retry_deleg: 39925955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 3993f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 39941da177e4SLinus Torvalds error = PTR_ERR(dentry); 39951da177e4SLinus Torvalds if (!IS_ERR(dentry)) { 39961da177e4SLinus Torvalds /* Why not before? Because we want correct error value */ 3997f5beed75SAl Viro if (last.name[last.len]) 399850338b88STörök Edwin goto slashes; 39991da177e4SLinus Torvalds inode = dentry->d_inode; 4000b18825a7SDavid Howells if (d_is_negative(dentry)) 4001e6bc45d6STheodore Ts'o goto slashes; 40027de9c6eeSAl Viro ihold(inode); 4003f5beed75SAl Viro error = security_path_unlink(&path, dentry); 4004be6d3e56SKentaro Takeda if (error) 4005c30dabfeSJan Kara goto exit2; 4006f5beed75SAl Viro error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode); 40071da177e4SLinus Torvalds exit2: 40081da177e4SLinus Torvalds dput(dentry); 40091da177e4SLinus Torvalds } 40105955102cSAl Viro inode_unlock(path.dentry->d_inode); 40111da177e4SLinus Torvalds if (inode) 40121da177e4SLinus Torvalds iput(inode); /* truncate the inode here */ 4013b21996e3SJ. Bruce Fields inode = NULL; 4014b21996e3SJ. Bruce Fields if (delegated_inode) { 40155a14696cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4016b21996e3SJ. Bruce Fields if (!error) 4017b21996e3SJ. Bruce Fields goto retry_deleg; 4018b21996e3SJ. Bruce Fields } 4019f5beed75SAl Viro mnt_drop_write(path.mnt); 40201da177e4SLinus Torvalds exit1: 4021f5beed75SAl Viro path_put(&path); 40221da177e4SLinus Torvalds putname(name); 40235d18f813SJeff Layton if (retry_estale(error, lookup_flags)) { 40245d18f813SJeff Layton lookup_flags |= LOOKUP_REVAL; 40255d18f813SJeff Layton inode = NULL; 40265d18f813SJeff Layton goto retry; 40275d18f813SJeff Layton } 40281da177e4SLinus Torvalds return error; 40291da177e4SLinus Torvalds 40301da177e4SLinus Torvalds slashes: 4031b18825a7SDavid Howells if (d_is_negative(dentry)) 4032b18825a7SDavid Howells error = -ENOENT; 403344b1d530SMiklos Szeredi else if (d_is_dir(dentry)) 4034b18825a7SDavid Howells error = -EISDIR; 4035b18825a7SDavid Howells else 4036b18825a7SDavid Howells error = -ENOTDIR; 40371da177e4SLinus Torvalds goto exit2; 40381da177e4SLinus Torvalds } 40391da177e4SLinus Torvalds 40402e4d0924SHeiko Carstens SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) 40415590ff0dSUlrich Drepper { 40425590ff0dSUlrich Drepper if ((flag & ~AT_REMOVEDIR) != 0) 40435590ff0dSUlrich Drepper return -EINVAL; 40445590ff0dSUlrich Drepper 40455590ff0dSUlrich Drepper if (flag & AT_REMOVEDIR) 40465590ff0dSUlrich Drepper return do_rmdir(dfd, pathname); 40475590ff0dSUlrich Drepper 40485590ff0dSUlrich Drepper return do_unlinkat(dfd, pathname); 40495590ff0dSUlrich Drepper } 40505590ff0dSUlrich Drepper 40513480b257SHeiko Carstens SYSCALL_DEFINE1(unlink, const char __user *, pathname) 40525590ff0dSUlrich Drepper { 40535590ff0dSUlrich Drepper return do_unlinkat(AT_FDCWD, pathname); 40545590ff0dSUlrich Drepper } 40555590ff0dSUlrich Drepper 4056db2e747bSMiklos Szeredi int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) 40571da177e4SLinus Torvalds { 4058a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 40591da177e4SLinus Torvalds 40601da177e4SLinus Torvalds if (error) 40611da177e4SLinus Torvalds return error; 40621da177e4SLinus Torvalds 4063acfa4380SAl Viro if (!dir->i_op->symlink) 40641da177e4SLinus Torvalds return -EPERM; 40651da177e4SLinus Torvalds 40661da177e4SLinus Torvalds error = security_inode_symlink(dir, dentry, oldname); 40671da177e4SLinus Torvalds if (error) 40681da177e4SLinus Torvalds return error; 40691da177e4SLinus Torvalds 40701da177e4SLinus Torvalds error = dir->i_op->symlink(dir, dentry, oldname); 4071a74574aaSStephen Smalley if (!error) 4072f38aa942SAmy Griffis fsnotify_create(dir, dentry); 40731da177e4SLinus Torvalds return error; 40741da177e4SLinus Torvalds } 40754d359507SAl Viro EXPORT_SYMBOL(vfs_symlink); 40761da177e4SLinus Torvalds 40772e4d0924SHeiko Carstens SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, 40782e4d0924SHeiko Carstens int, newdfd, const char __user *, newname) 40791da177e4SLinus Torvalds { 40802ad94ae6SAl Viro int error; 408191a27b2aSJeff Layton struct filename *from; 40826902d925SDave Hansen struct dentry *dentry; 4083dae6ad8fSAl Viro struct path path; 4084f46d3567SJeff Layton unsigned int lookup_flags = 0; 40851da177e4SLinus Torvalds 40861da177e4SLinus Torvalds from = getname(oldname); 40871da177e4SLinus Torvalds if (IS_ERR(from)) 40881da177e4SLinus Torvalds return PTR_ERR(from); 4089f46d3567SJeff Layton retry: 4090f46d3567SJeff Layton dentry = user_path_create(newdfd, newname, &path, lookup_flags); 40911da177e4SLinus Torvalds error = PTR_ERR(dentry); 40926902d925SDave Hansen if (IS_ERR(dentry)) 4093dae6ad8fSAl Viro goto out_putname; 40946902d925SDave Hansen 409591a27b2aSJeff Layton error = security_path_symlink(&path, dentry, from->name); 4096a8104a9fSAl Viro if (!error) 409791a27b2aSJeff Layton error = vfs_symlink(path.dentry->d_inode, dentry, from->name); 4098921a1650SAl Viro done_path_create(&path, dentry); 4099f46d3567SJeff Layton if (retry_estale(error, lookup_flags)) { 4100f46d3567SJeff Layton lookup_flags |= LOOKUP_REVAL; 4101f46d3567SJeff Layton goto retry; 4102f46d3567SJeff Layton } 41036902d925SDave Hansen out_putname: 41041da177e4SLinus Torvalds putname(from); 41051da177e4SLinus Torvalds return error; 41061da177e4SLinus Torvalds } 41071da177e4SLinus Torvalds 41083480b257SHeiko Carstens SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) 41095590ff0dSUlrich Drepper { 41105590ff0dSUlrich Drepper return sys_symlinkat(oldname, AT_FDCWD, newname); 41115590ff0dSUlrich Drepper } 41125590ff0dSUlrich Drepper 4113146a8595SJ. Bruce Fields /** 4114146a8595SJ. Bruce Fields * vfs_link - create a new link 4115146a8595SJ. Bruce Fields * @old_dentry: object to be linked 4116146a8595SJ. Bruce Fields * @dir: new parent 4117146a8595SJ. Bruce Fields * @new_dentry: where to create the new link 4118146a8595SJ. Bruce Fields * @delegated_inode: returns inode needing a delegation break 4119146a8595SJ. Bruce Fields * 4120146a8595SJ. Bruce Fields * The caller must hold dir->i_mutex 4121146a8595SJ. Bruce Fields * 4122146a8595SJ. Bruce Fields * If vfs_link discovers a delegation on the to-be-linked file in need 4123146a8595SJ. Bruce Fields * of breaking, it will return -EWOULDBLOCK and return a reference to the 4124146a8595SJ. Bruce Fields * inode in delegated_inode. The caller should then break the delegation 4125146a8595SJ. Bruce Fields * and retry. Because breaking a delegation may take a long time, the 4126146a8595SJ. Bruce Fields * caller should drop the i_mutex before doing so. 4127146a8595SJ. Bruce Fields * 4128146a8595SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 4129146a8595SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 4130146a8595SJ. Bruce Fields * to be NFS exported. 4131146a8595SJ. Bruce Fields */ 4132146a8595SJ. Bruce Fields int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode) 41331da177e4SLinus Torvalds { 41341da177e4SLinus Torvalds struct inode *inode = old_dentry->d_inode; 41358de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 41361da177e4SLinus Torvalds int error; 41371da177e4SLinus Torvalds 41381da177e4SLinus Torvalds if (!inode) 41391da177e4SLinus Torvalds return -ENOENT; 41401da177e4SLinus Torvalds 4141a95164d9SMiklos Szeredi error = may_create(dir, new_dentry); 41421da177e4SLinus Torvalds if (error) 41431da177e4SLinus Torvalds return error; 41441da177e4SLinus Torvalds 41451da177e4SLinus Torvalds if (dir->i_sb != inode->i_sb) 41461da177e4SLinus Torvalds return -EXDEV; 41471da177e4SLinus Torvalds 41481da177e4SLinus Torvalds /* 41491da177e4SLinus Torvalds * A link to an append-only or immutable file cannot be created. 41501da177e4SLinus Torvalds */ 41511da177e4SLinus Torvalds if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 41521da177e4SLinus Torvalds return -EPERM; 41530bd23d09SEric W. Biederman /* 41540bd23d09SEric W. Biederman * Updating the link count will likely cause i_uid and i_gid to 41550bd23d09SEric W. Biederman * be writen back improperly if their true value is unknown to 41560bd23d09SEric W. Biederman * the vfs. 41570bd23d09SEric W. Biederman */ 41580bd23d09SEric W. Biederman if (HAS_UNMAPPED_ID(inode)) 41590bd23d09SEric W. Biederman return -EPERM; 4160acfa4380SAl Viro if (!dir->i_op->link) 41611da177e4SLinus Torvalds return -EPERM; 41627e79eedbSTetsuo Handa if (S_ISDIR(inode->i_mode)) 41631da177e4SLinus Torvalds return -EPERM; 41641da177e4SLinus Torvalds 41651da177e4SLinus Torvalds error = security_inode_link(old_dentry, dir, new_dentry); 41661da177e4SLinus Torvalds if (error) 41671da177e4SLinus Torvalds return error; 41681da177e4SLinus Torvalds 41695955102cSAl Viro inode_lock(inode); 4170aae8a97dSAneesh Kumar K.V /* Make sure we don't allow creating hardlink to an unlinked file */ 4171f4e0c30cSAl Viro if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) 4172aae8a97dSAneesh Kumar K.V error = -ENOENT; 41738de52778SAl Viro else if (max_links && inode->i_nlink >= max_links) 41748de52778SAl Viro error = -EMLINK; 4175146a8595SJ. Bruce Fields else { 4176146a8595SJ. Bruce Fields error = try_break_deleg(inode, delegated_inode); 4177146a8595SJ. Bruce Fields if (!error) 41781da177e4SLinus Torvalds error = dir->i_op->link(old_dentry, dir, new_dentry); 4179146a8595SJ. Bruce Fields } 4180f4e0c30cSAl Viro 4181f4e0c30cSAl Viro if (!error && (inode->i_state & I_LINKABLE)) { 4182f4e0c30cSAl Viro spin_lock(&inode->i_lock); 4183f4e0c30cSAl Viro inode->i_state &= ~I_LINKABLE; 4184f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 4185f4e0c30cSAl Viro } 41865955102cSAl Viro inode_unlock(inode); 4187e31e14ecSStephen Smalley if (!error) 41887e79eedbSTetsuo Handa fsnotify_link(dir, inode, new_dentry); 41891da177e4SLinus Torvalds return error; 41901da177e4SLinus Torvalds } 41914d359507SAl Viro EXPORT_SYMBOL(vfs_link); 41921da177e4SLinus Torvalds 41931da177e4SLinus Torvalds /* 41941da177e4SLinus Torvalds * Hardlinks are often used in delicate situations. We avoid 41951da177e4SLinus Torvalds * security-related surprises by not following symlinks on the 41961da177e4SLinus Torvalds * newname. --KAB 41971da177e4SLinus Torvalds * 41981da177e4SLinus Torvalds * We don't follow them on the oldname either to be compatible 41991da177e4SLinus Torvalds * with linux 2.0, and to avoid hard-linking to directories 42001da177e4SLinus Torvalds * and other special files. --ADM 42011da177e4SLinus Torvalds */ 42022e4d0924SHeiko Carstens SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, 42032e4d0924SHeiko Carstens int, newdfd, const char __user *, newname, int, flags) 42041da177e4SLinus Torvalds { 42051da177e4SLinus Torvalds struct dentry *new_dentry; 4206dae6ad8fSAl Viro struct path old_path, new_path; 4207146a8595SJ. Bruce Fields struct inode *delegated_inode = NULL; 420811a7b371SAneesh Kumar K.V int how = 0; 42091da177e4SLinus Torvalds int error; 42101da177e4SLinus Torvalds 421111a7b371SAneesh Kumar K.V if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) 4212c04030e1SUlrich Drepper return -EINVAL; 421311a7b371SAneesh Kumar K.V /* 4214f0cc6ffbSLinus Torvalds * To use null names we require CAP_DAC_READ_SEARCH 4215f0cc6ffbSLinus Torvalds * This ensures that not everyone will be able to create 4216f0cc6ffbSLinus Torvalds * handlink using the passed filedescriptor. 421711a7b371SAneesh Kumar K.V */ 4218f0cc6ffbSLinus Torvalds if (flags & AT_EMPTY_PATH) { 4219f0cc6ffbSLinus Torvalds if (!capable(CAP_DAC_READ_SEARCH)) 4220f0cc6ffbSLinus Torvalds return -ENOENT; 422111a7b371SAneesh Kumar K.V how = LOOKUP_EMPTY; 4222f0cc6ffbSLinus Torvalds } 4223c04030e1SUlrich Drepper 422411a7b371SAneesh Kumar K.V if (flags & AT_SYMLINK_FOLLOW) 422511a7b371SAneesh Kumar K.V how |= LOOKUP_FOLLOW; 4226442e31caSJeff Layton retry: 422711a7b371SAneesh Kumar K.V error = user_path_at(olddfd, oldname, how, &old_path); 42281da177e4SLinus Torvalds if (error) 42292ad94ae6SAl Viro return error; 42302ad94ae6SAl Viro 4231442e31caSJeff Layton new_dentry = user_path_create(newdfd, newname, &new_path, 4232442e31caSJeff Layton (how & LOOKUP_REVAL)); 42331da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 42346902d925SDave Hansen if (IS_ERR(new_dentry)) 4235dae6ad8fSAl Viro goto out; 4236dae6ad8fSAl Viro 4237dae6ad8fSAl Viro error = -EXDEV; 4238dae6ad8fSAl Viro if (old_path.mnt != new_path.mnt) 4239dae6ad8fSAl Viro goto out_dput; 4240800179c9SKees Cook error = may_linkat(&old_path); 4241800179c9SKees Cook if (unlikely(error)) 4242800179c9SKees Cook goto out_dput; 4243dae6ad8fSAl Viro error = security_path_link(old_path.dentry, &new_path, new_dentry); 4244be6d3e56SKentaro Takeda if (error) 4245a8104a9fSAl Viro goto out_dput; 4246146a8595SJ. Bruce Fields error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode); 424775c3f29dSDave Hansen out_dput: 4248921a1650SAl Viro done_path_create(&new_path, new_dentry); 4249146a8595SJ. Bruce Fields if (delegated_inode) { 4250146a8595SJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4251d22e6338SOleg Drokin if (!error) { 4252d22e6338SOleg Drokin path_put(&old_path); 4253146a8595SJ. Bruce Fields goto retry; 4254146a8595SJ. Bruce Fields } 4255d22e6338SOleg Drokin } 4256442e31caSJeff Layton if (retry_estale(error, how)) { 4257d22e6338SOleg Drokin path_put(&old_path); 4258442e31caSJeff Layton how |= LOOKUP_REVAL; 4259442e31caSJeff Layton goto retry; 4260442e31caSJeff Layton } 42611da177e4SLinus Torvalds out: 42622d8f3038SAl Viro path_put(&old_path); 42631da177e4SLinus Torvalds 42641da177e4SLinus Torvalds return error; 42651da177e4SLinus Torvalds } 42661da177e4SLinus Torvalds 42673480b257SHeiko Carstens SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) 42685590ff0dSUlrich Drepper { 4269c04030e1SUlrich Drepper return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 42705590ff0dSUlrich Drepper } 42715590ff0dSUlrich Drepper 4272bc27027aSMiklos Szeredi /** 4273bc27027aSMiklos Szeredi * vfs_rename - rename a filesystem object 4274bc27027aSMiklos Szeredi * @old_dir: parent of source 4275bc27027aSMiklos Szeredi * @old_dentry: source 4276bc27027aSMiklos Szeredi * @new_dir: parent of destination 4277bc27027aSMiklos Szeredi * @new_dentry: destination 4278bc27027aSMiklos Szeredi * @delegated_inode: returns an inode needing a delegation break 4279520c8b16SMiklos Szeredi * @flags: rename flags 4280bc27027aSMiklos Szeredi * 4281bc27027aSMiklos Szeredi * The caller must hold multiple mutexes--see lock_rename()). 4282bc27027aSMiklos Szeredi * 4283bc27027aSMiklos Szeredi * If vfs_rename discovers a delegation in need of breaking at either 4284bc27027aSMiklos Szeredi * the source or destination, it will return -EWOULDBLOCK and return a 4285bc27027aSMiklos Szeredi * reference to the inode in delegated_inode. The caller should then 4286bc27027aSMiklos Szeredi * break the delegation and retry. Because breaking a delegation may 4287bc27027aSMiklos Szeredi * take a long time, the caller should drop all locks before doing 4288bc27027aSMiklos Szeredi * so. 4289bc27027aSMiklos Szeredi * 4290bc27027aSMiklos Szeredi * Alternatively, a caller may pass NULL for delegated_inode. This may 4291bc27027aSMiklos Szeredi * be appropriate for callers that expect the underlying filesystem not 4292bc27027aSMiklos Szeredi * to be NFS exported. 4293bc27027aSMiklos Szeredi * 42941da177e4SLinus Torvalds * The worst of all namespace operations - renaming directory. "Perverted" 42951da177e4SLinus Torvalds * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 42961da177e4SLinus Torvalds * Problems: 4297d03b29a2SJ. Bruce Fields * a) we can get into loop creation. 42981da177e4SLinus Torvalds * b) race potential - two innocent renames can create a loop together. 42991da177e4SLinus Torvalds * That's where 4.4 screws up. Current fix: serialization on 4300a11f3a05SArjan van de Ven * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 43011da177e4SLinus Torvalds * story. 43026cedba89SJ. Bruce Fields * c) we have to lock _four_ objects - parents and victim (if it exists), 43036cedba89SJ. Bruce Fields * and source (if it is not a directory). 43041b1dcc1bSJes Sorensen * And that - after we got ->i_mutex on parents (until then we don't know 43051da177e4SLinus Torvalds * whether the target exists). Solution: try to be smart with locking 43061da177e4SLinus Torvalds * order for inodes. We rely on the fact that tree topology may change 4307a11f3a05SArjan van de Ven * only under ->s_vfs_rename_mutex _and_ that parent of the object we 43081da177e4SLinus Torvalds * move will be locked. Thus we can rank directories by the tree 43091da177e4SLinus Torvalds * (ancestors first) and rank all non-directories after them. 43101da177e4SLinus Torvalds * That works since everybody except rename does "lock parent, lookup, 4311a11f3a05SArjan van de Ven * lock child" and rename is under ->s_vfs_rename_mutex. 43121da177e4SLinus Torvalds * HOWEVER, it relies on the assumption that any object with ->lookup() 43131da177e4SLinus Torvalds * has no more than 1 dentry. If "hybrid" objects will ever appear, 43141da177e4SLinus Torvalds * we'd better make sure that there's no link(2) for them. 4315e4eaac06SSage Weil * d) conversion from fhandle to dentry may come in the wrong moment - when 43161b1dcc1bSJes Sorensen * we are removing the target. Solution: we will have to grab ->i_mutex 43171da177e4SLinus Torvalds * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 4318c41b20e7SAdam Buchbinder * ->i_mutex on parents, which works but leads to some truly excessive 43191da177e4SLinus Torvalds * locking]. 43201da177e4SLinus Torvalds */ 43211da177e4SLinus Torvalds int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, 43228e6d782cSJ. Bruce Fields struct inode *new_dir, struct dentry *new_dentry, 4323520c8b16SMiklos Szeredi struct inode **delegated_inode, unsigned int flags) 43241da177e4SLinus Torvalds { 43251da177e4SLinus Torvalds int error; 4326bc27027aSMiklos Szeredi bool is_dir = d_is_dir(old_dentry); 432759b0df21SEric Paris const unsigned char *old_name; 4328bc27027aSMiklos Szeredi struct inode *source = old_dentry->d_inode; 4329bc27027aSMiklos Szeredi struct inode *target = new_dentry->d_inode; 4330da1ce067SMiklos Szeredi bool new_is_dir = false; 4331da1ce067SMiklos Szeredi unsigned max_links = new_dir->i_sb->s_max_links; 43321da177e4SLinus Torvalds 43338d3e2936SMiklos Szeredi if (source == target) 43341da177e4SLinus Torvalds return 0; 43351da177e4SLinus Torvalds 43361da177e4SLinus Torvalds error = may_delete(old_dir, old_dentry, is_dir); 43371da177e4SLinus Torvalds if (error) 43381da177e4SLinus Torvalds return error; 43391da177e4SLinus Torvalds 4340da1ce067SMiklos Szeredi if (!target) { 4341a95164d9SMiklos Szeredi error = may_create(new_dir, new_dentry); 4342da1ce067SMiklos Szeredi } else { 4343da1ce067SMiklos Szeredi new_is_dir = d_is_dir(new_dentry); 4344da1ce067SMiklos Szeredi 4345da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 43461da177e4SLinus Torvalds error = may_delete(new_dir, new_dentry, is_dir); 4347da1ce067SMiklos Szeredi else 4348da1ce067SMiklos Szeredi error = may_delete(new_dir, new_dentry, new_is_dir); 4349da1ce067SMiklos Szeredi } 43501da177e4SLinus Torvalds if (error) 43511da177e4SLinus Torvalds return error; 43521da177e4SLinus Torvalds 43532773bf00SMiklos Szeredi if (!old_dir->i_op->rename) 43541da177e4SLinus Torvalds return -EPERM; 43551da177e4SLinus Torvalds 4356bc27027aSMiklos Szeredi /* 4357bc27027aSMiklos Szeredi * If we are going to change the parent - check write permissions, 4358bc27027aSMiklos Szeredi * we'll need to flip '..'. 4359bc27027aSMiklos Szeredi */ 4360da1ce067SMiklos Szeredi if (new_dir != old_dir) { 4361da1ce067SMiklos Szeredi if (is_dir) { 4362bc27027aSMiklos Szeredi error = inode_permission(source, MAY_WRITE); 4363bc27027aSMiklos Szeredi if (error) 4364bc27027aSMiklos Szeredi return error; 4365bc27027aSMiklos Szeredi } 4366da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && new_is_dir) { 4367da1ce067SMiklos Szeredi error = inode_permission(target, MAY_WRITE); 4368da1ce067SMiklos Szeredi if (error) 4369da1ce067SMiklos Szeredi return error; 4370da1ce067SMiklos Szeredi } 4371da1ce067SMiklos Szeredi } 43720eeca283SRobert Love 43730b3974ebSMiklos Szeredi error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry, 43740b3974ebSMiklos Szeredi flags); 4375bc27027aSMiklos Szeredi if (error) 4376bc27027aSMiklos Szeredi return error; 4377bc27027aSMiklos Szeredi 4378bc27027aSMiklos Szeredi old_name = fsnotify_oldname_init(old_dentry->d_name.name); 4379bc27027aSMiklos Szeredi dget(new_dentry); 4380da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4381bc27027aSMiklos Szeredi lock_two_nondirectories(source, target); 4382bc27027aSMiklos Szeredi else if (target) 43835955102cSAl Viro inode_lock(target); 4384bc27027aSMiklos Szeredi 4385bc27027aSMiklos Szeredi error = -EBUSY; 43867af1364fSEric W. Biederman if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry)) 4387bc27027aSMiklos Szeredi goto out; 4388bc27027aSMiklos Szeredi 4389da1ce067SMiklos Szeredi if (max_links && new_dir != old_dir) { 4390bc27027aSMiklos Szeredi error = -EMLINK; 4391da1ce067SMiklos Szeredi if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links) 4392bc27027aSMiklos Szeredi goto out; 4393da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir && 4394da1ce067SMiklos Szeredi old_dir->i_nlink >= max_links) 4395da1ce067SMiklos Szeredi goto out; 4396da1ce067SMiklos Szeredi } 4397da1ce067SMiklos Szeredi if (is_dir && !(flags & RENAME_EXCHANGE) && target) 4398bc27027aSMiklos Szeredi shrink_dcache_parent(new_dentry); 4399da1ce067SMiklos Szeredi if (!is_dir) { 4400bc27027aSMiklos Szeredi error = try_break_deleg(source, delegated_inode); 4401bc27027aSMiklos Szeredi if (error) 4402bc27027aSMiklos Szeredi goto out; 4403da1ce067SMiklos Szeredi } 4404da1ce067SMiklos Szeredi if (target && !new_is_dir) { 4405bc27027aSMiklos Szeredi error = try_break_deleg(target, delegated_inode); 4406bc27027aSMiklos Szeredi if (error) 4407bc27027aSMiklos Szeredi goto out; 4408bc27027aSMiklos Szeredi } 4409520c8b16SMiklos Szeredi error = old_dir->i_op->rename(old_dir, old_dentry, 4410520c8b16SMiklos Szeredi new_dir, new_dentry, flags); 4411bc27027aSMiklos Szeredi if (error) 4412bc27027aSMiklos Szeredi goto out; 4413bc27027aSMiklos Szeredi 4414da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE) && target) { 44151da177e4SLinus Torvalds if (is_dir) 4416bc27027aSMiklos Szeredi target->i_flags |= S_DEAD; 4417bc27027aSMiklos Szeredi dont_mount(new_dentry); 44188ed936b5SEric W. Biederman detach_mounts(new_dentry); 4419bc27027aSMiklos Szeredi } 4420da1ce067SMiklos Szeredi if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) { 4421da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 4422bc27027aSMiklos Szeredi d_move(old_dentry, new_dentry); 4423da1ce067SMiklos Szeredi else 4424da1ce067SMiklos Szeredi d_exchange(old_dentry, new_dentry); 4425da1ce067SMiklos Szeredi } 4426bc27027aSMiklos Szeredi out: 4427da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4428bc27027aSMiklos Szeredi unlock_two_nondirectories(source, target); 4429bc27027aSMiklos Szeredi else if (target) 44305955102cSAl Viro inode_unlock(target); 4431bc27027aSMiklos Szeredi dput(new_dentry); 4432da1ce067SMiklos Szeredi if (!error) { 4433123df294SAl Viro fsnotify_move(old_dir, new_dir, old_name, is_dir, 4434da1ce067SMiklos Szeredi !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry); 4435da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4436da1ce067SMiklos Szeredi fsnotify_move(new_dir, old_dir, old_dentry->d_name.name, 4437da1ce067SMiklos Szeredi new_is_dir, NULL, new_dentry); 4438da1ce067SMiklos Szeredi } 4439da1ce067SMiklos Szeredi } 44400eeca283SRobert Love fsnotify_oldname_free(old_name); 44410eeca283SRobert Love 44421da177e4SLinus Torvalds return error; 44431da177e4SLinus Torvalds } 44444d359507SAl Viro EXPORT_SYMBOL(vfs_rename); 44451da177e4SLinus Torvalds 4446520c8b16SMiklos Szeredi SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname, 4447520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname, unsigned int, flags) 44481da177e4SLinus Torvalds { 44491da177e4SLinus Torvalds struct dentry *old_dentry, *new_dentry; 44501da177e4SLinus Torvalds struct dentry *trap; 4451f5beed75SAl Viro struct path old_path, new_path; 4452f5beed75SAl Viro struct qstr old_last, new_last; 4453f5beed75SAl Viro int old_type, new_type; 44548e6d782cSJ. Bruce Fields struct inode *delegated_inode = NULL; 445591a27b2aSJeff Layton struct filename *from; 445691a27b2aSJeff Layton struct filename *to; 4457f5beed75SAl Viro unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET; 4458c6a94284SJeff Layton bool should_retry = false; 44592ad94ae6SAl Viro int error; 4460520c8b16SMiklos Szeredi 44610d7a8555SMiklos Szeredi if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) 4462da1ce067SMiklos Szeredi return -EINVAL; 4463da1ce067SMiklos Szeredi 44640d7a8555SMiklos Szeredi if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) && 44650d7a8555SMiklos Szeredi (flags & RENAME_EXCHANGE)) 4466520c8b16SMiklos Szeredi return -EINVAL; 4467520c8b16SMiklos Szeredi 44680d7a8555SMiklos Szeredi if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD)) 44690d7a8555SMiklos Szeredi return -EPERM; 44700d7a8555SMiklos Szeredi 4471f5beed75SAl Viro if (flags & RENAME_EXCHANGE) 4472f5beed75SAl Viro target_flags = 0; 4473f5beed75SAl Viro 4474c6a94284SJeff Layton retry: 4475c1d4dd27SAl Viro from = filename_parentat(olddfd, getname(oldname), lookup_flags, 4476c1d4dd27SAl Viro &old_path, &old_last, &old_type); 447791a27b2aSJeff Layton if (IS_ERR(from)) { 447891a27b2aSJeff Layton error = PTR_ERR(from); 44791da177e4SLinus Torvalds goto exit; 448091a27b2aSJeff Layton } 44811da177e4SLinus Torvalds 4482c1d4dd27SAl Viro to = filename_parentat(newdfd, getname(newname), lookup_flags, 4483c1d4dd27SAl Viro &new_path, &new_last, &new_type); 448491a27b2aSJeff Layton if (IS_ERR(to)) { 448591a27b2aSJeff Layton error = PTR_ERR(to); 44861da177e4SLinus Torvalds goto exit1; 448791a27b2aSJeff Layton } 44881da177e4SLinus Torvalds 44891da177e4SLinus Torvalds error = -EXDEV; 4490f5beed75SAl Viro if (old_path.mnt != new_path.mnt) 44911da177e4SLinus Torvalds goto exit2; 44921da177e4SLinus Torvalds 44931da177e4SLinus Torvalds error = -EBUSY; 4494f5beed75SAl Viro if (old_type != LAST_NORM) 44951da177e4SLinus Torvalds goto exit2; 44961da177e4SLinus Torvalds 44970a7c3937SMiklos Szeredi if (flags & RENAME_NOREPLACE) 44980a7c3937SMiklos Szeredi error = -EEXIST; 4499f5beed75SAl Viro if (new_type != LAST_NORM) 45001da177e4SLinus Torvalds goto exit2; 45011da177e4SLinus Torvalds 4502f5beed75SAl Viro error = mnt_want_write(old_path.mnt); 4503c30dabfeSJan Kara if (error) 4504c30dabfeSJan Kara goto exit2; 4505c30dabfeSJan Kara 45068e6d782cSJ. Bruce Fields retry_deleg: 4507f5beed75SAl Viro trap = lock_rename(new_path.dentry, old_path.dentry); 45081da177e4SLinus Torvalds 4509f5beed75SAl Viro old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags); 45101da177e4SLinus Torvalds error = PTR_ERR(old_dentry); 45111da177e4SLinus Torvalds if (IS_ERR(old_dentry)) 45121da177e4SLinus Torvalds goto exit3; 45131da177e4SLinus Torvalds /* source must exist */ 45141da177e4SLinus Torvalds error = -ENOENT; 4515b18825a7SDavid Howells if (d_is_negative(old_dentry)) 45161da177e4SLinus Torvalds goto exit4; 4517f5beed75SAl Viro new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags); 45181da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 45191da177e4SLinus Torvalds if (IS_ERR(new_dentry)) 45201da177e4SLinus Torvalds goto exit4; 45210a7c3937SMiklos Szeredi error = -EEXIST; 45220a7c3937SMiklos Szeredi if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry)) 45230a7c3937SMiklos Szeredi goto exit5; 4524da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4525da1ce067SMiklos Szeredi error = -ENOENT; 4526da1ce067SMiklos Szeredi if (d_is_negative(new_dentry)) 4527da1ce067SMiklos Szeredi goto exit5; 4528da1ce067SMiklos Szeredi 4529da1ce067SMiklos Szeredi if (!d_is_dir(new_dentry)) { 4530da1ce067SMiklos Szeredi error = -ENOTDIR; 4531f5beed75SAl Viro if (new_last.name[new_last.len]) 4532da1ce067SMiklos Szeredi goto exit5; 4533da1ce067SMiklos Szeredi } 4534da1ce067SMiklos Szeredi } 45350a7c3937SMiklos Szeredi /* unless the source is a directory trailing slashes give -ENOTDIR */ 45360a7c3937SMiklos Szeredi if (!d_is_dir(old_dentry)) { 45370a7c3937SMiklos Szeredi error = -ENOTDIR; 4538f5beed75SAl Viro if (old_last.name[old_last.len]) 45390a7c3937SMiklos Szeredi goto exit5; 4540f5beed75SAl Viro if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len]) 45410a7c3937SMiklos Szeredi goto exit5; 45420a7c3937SMiklos Szeredi } 45430a7c3937SMiklos Szeredi /* source should not be ancestor of target */ 45440a7c3937SMiklos Szeredi error = -EINVAL; 45450a7c3937SMiklos Szeredi if (old_dentry == trap) 45460a7c3937SMiklos Szeredi goto exit5; 45471da177e4SLinus Torvalds /* target should not be an ancestor of source */ 4548da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 45491da177e4SLinus Torvalds error = -ENOTEMPTY; 45501da177e4SLinus Torvalds if (new_dentry == trap) 45511da177e4SLinus Torvalds goto exit5; 45521da177e4SLinus Torvalds 4553f5beed75SAl Viro error = security_path_rename(&old_path, old_dentry, 4554f5beed75SAl Viro &new_path, new_dentry, flags); 4555be6d3e56SKentaro Takeda if (error) 4556c30dabfeSJan Kara goto exit5; 4557f5beed75SAl Viro error = vfs_rename(old_path.dentry->d_inode, old_dentry, 4558f5beed75SAl Viro new_path.dentry->d_inode, new_dentry, 4559520c8b16SMiklos Szeredi &delegated_inode, flags); 45601da177e4SLinus Torvalds exit5: 45611da177e4SLinus Torvalds dput(new_dentry); 45621da177e4SLinus Torvalds exit4: 45631da177e4SLinus Torvalds dput(old_dentry); 45641da177e4SLinus Torvalds exit3: 4565f5beed75SAl Viro unlock_rename(new_path.dentry, old_path.dentry); 45668e6d782cSJ. Bruce Fields if (delegated_inode) { 45678e6d782cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 45688e6d782cSJ. Bruce Fields if (!error) 45698e6d782cSJ. Bruce Fields goto retry_deleg; 45708e6d782cSJ. Bruce Fields } 4571f5beed75SAl Viro mnt_drop_write(old_path.mnt); 45721da177e4SLinus Torvalds exit2: 4573c6a94284SJeff Layton if (retry_estale(error, lookup_flags)) 4574c6a94284SJeff Layton should_retry = true; 4575f5beed75SAl Viro path_put(&new_path); 45762ad94ae6SAl Viro putname(to); 45771da177e4SLinus Torvalds exit1: 4578f5beed75SAl Viro path_put(&old_path); 45791da177e4SLinus Torvalds putname(from); 4580c6a94284SJeff Layton if (should_retry) { 4581c6a94284SJeff Layton should_retry = false; 4582c6a94284SJeff Layton lookup_flags |= LOOKUP_REVAL; 4583c6a94284SJeff Layton goto retry; 4584c6a94284SJeff Layton } 45852ad94ae6SAl Viro exit: 45861da177e4SLinus Torvalds return error; 45871da177e4SLinus Torvalds } 45881da177e4SLinus Torvalds 4589520c8b16SMiklos Szeredi SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, 4590520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname) 4591520c8b16SMiklos Szeredi { 4592520c8b16SMiklos Szeredi return sys_renameat2(olddfd, oldname, newdfd, newname, 0); 4593520c8b16SMiklos Szeredi } 4594520c8b16SMiklos Szeredi 4595a26eab24SHeiko Carstens SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) 45965590ff0dSUlrich Drepper { 4597520c8b16SMiklos Szeredi return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 45985590ff0dSUlrich Drepper } 45995590ff0dSUlrich Drepper 4600787fb6bcSMiklos Szeredi int vfs_whiteout(struct inode *dir, struct dentry *dentry) 4601787fb6bcSMiklos Szeredi { 4602787fb6bcSMiklos Szeredi int error = may_create(dir, dentry); 4603787fb6bcSMiklos Szeredi if (error) 4604787fb6bcSMiklos Szeredi return error; 4605787fb6bcSMiklos Szeredi 4606787fb6bcSMiklos Szeredi if (!dir->i_op->mknod) 4607787fb6bcSMiklos Szeredi return -EPERM; 4608787fb6bcSMiklos Szeredi 4609787fb6bcSMiklos Szeredi return dir->i_op->mknod(dir, dentry, 4610787fb6bcSMiklos Szeredi S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV); 4611787fb6bcSMiklos Szeredi } 4612787fb6bcSMiklos Szeredi EXPORT_SYMBOL(vfs_whiteout); 4613787fb6bcSMiklos Szeredi 46145d826c84SAl Viro int readlink_copy(char __user *buffer, int buflen, const char *link) 46151da177e4SLinus Torvalds { 46165d826c84SAl Viro int len = PTR_ERR(link); 46171da177e4SLinus Torvalds if (IS_ERR(link)) 46181da177e4SLinus Torvalds goto out; 46191da177e4SLinus Torvalds 46201da177e4SLinus Torvalds len = strlen(link); 46211da177e4SLinus Torvalds if (len > (unsigned) buflen) 46221da177e4SLinus Torvalds len = buflen; 46231da177e4SLinus Torvalds if (copy_to_user(buffer, link, len)) 46241da177e4SLinus Torvalds len = -EFAULT; 46251da177e4SLinus Torvalds out: 46261da177e4SLinus Torvalds return len; 46271da177e4SLinus Torvalds } 46281da177e4SLinus Torvalds 46291da177e4SLinus Torvalds /* 46301da177e4SLinus Torvalds * A helper for ->readlink(). This should be used *ONLY* for symlinks that 46316b255391SAl Viro * have ->get_link() not calling nd_jump_link(). Using (or not using) it 46326b255391SAl Viro * for any given inode is up to filesystem. 46331da177e4SLinus Torvalds */ 4634d16744ecSMiklos Szeredi static int generic_readlink(struct dentry *dentry, char __user *buffer, 4635d16744ecSMiklos Szeredi int buflen) 46361da177e4SLinus Torvalds { 4637fceef393SAl Viro DEFINE_DELAYED_CALL(done); 46385f2c4179SAl Viro struct inode *inode = d_inode(dentry); 46395f2c4179SAl Viro const char *link = inode->i_link; 4640694a1764SMarcin Slusarz int res; 4641cc314eefSLinus Torvalds 4642d4dee48bSAl Viro if (!link) { 4643fceef393SAl Viro link = inode->i_op->get_link(dentry, inode, &done); 4644680baacbSAl Viro if (IS_ERR(link)) 4645680baacbSAl Viro return PTR_ERR(link); 4646d4dee48bSAl Viro } 4647680baacbSAl Viro res = readlink_copy(buffer, buflen, link); 4648fceef393SAl Viro do_delayed_call(&done); 4649694a1764SMarcin Slusarz return res; 46501da177e4SLinus Torvalds } 46511da177e4SLinus Torvalds 4652d60874cdSMiklos Szeredi /** 4653fd4a0edfSMiklos Szeredi * vfs_readlink - copy symlink body into userspace buffer 4654fd4a0edfSMiklos Szeredi * @dentry: dentry on which to get symbolic link 4655fd4a0edfSMiklos Szeredi * @buffer: user memory pointer 4656fd4a0edfSMiklos Szeredi * @buflen: size of buffer 4657fd4a0edfSMiklos Szeredi * 4658fd4a0edfSMiklos Szeredi * Does not touch atime. That's up to the caller if necessary 4659fd4a0edfSMiklos Szeredi * 4660fd4a0edfSMiklos Szeredi * Does not call security hook. 4661fd4a0edfSMiklos Szeredi */ 4662fd4a0edfSMiklos Szeredi int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen) 4663fd4a0edfSMiklos Szeredi { 4664fd4a0edfSMiklos Szeredi struct inode *inode = d_inode(dentry); 4665fd4a0edfSMiklos Szeredi 466676fca90eSMiklos Szeredi if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) { 466776fca90eSMiklos Szeredi if (unlikely(inode->i_op->readlink)) 466876fca90eSMiklos Szeredi return inode->i_op->readlink(dentry, buffer, buflen); 466976fca90eSMiklos Szeredi 467076fca90eSMiklos Szeredi if (!d_is_symlink(dentry)) 4671fd4a0edfSMiklos Szeredi return -EINVAL; 4672fd4a0edfSMiklos Szeredi 467376fca90eSMiklos Szeredi spin_lock(&inode->i_lock); 467476fca90eSMiklos Szeredi inode->i_opflags |= IOP_DEFAULT_READLINK; 467576fca90eSMiklos Szeredi spin_unlock(&inode->i_lock); 467676fca90eSMiklos Szeredi } 467776fca90eSMiklos Szeredi 467876fca90eSMiklos Szeredi return generic_readlink(dentry, buffer, buflen); 4679fd4a0edfSMiklos Szeredi } 4680fd4a0edfSMiklos Szeredi EXPORT_SYMBOL(vfs_readlink); 46811da177e4SLinus Torvalds 4682d60874cdSMiklos Szeredi /** 4683d60874cdSMiklos Szeredi * vfs_get_link - get symlink body 4684d60874cdSMiklos Szeredi * @dentry: dentry on which to get symbolic link 4685d60874cdSMiklos Szeredi * @done: caller needs to free returned data with this 4686d60874cdSMiklos Szeredi * 4687d60874cdSMiklos Szeredi * Calls security hook and i_op->get_link() on the supplied inode. 4688d60874cdSMiklos Szeredi * 4689d60874cdSMiklos Szeredi * It does not touch atime. That's up to the caller if necessary. 4690d60874cdSMiklos Szeredi * 4691d60874cdSMiklos Szeredi * Does not work on "special" symlinks like /proc/$$/fd/N 4692d60874cdSMiklos Szeredi */ 4693d60874cdSMiklos Szeredi const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done) 4694d60874cdSMiklos Szeredi { 4695d60874cdSMiklos Szeredi const char *res = ERR_PTR(-EINVAL); 4696d60874cdSMiklos Szeredi struct inode *inode = d_inode(dentry); 4697d60874cdSMiklos Szeredi 4698d60874cdSMiklos Szeredi if (d_is_symlink(dentry)) { 4699d60874cdSMiklos Szeredi res = ERR_PTR(security_inode_readlink(dentry)); 4700d60874cdSMiklos Szeredi if (!res) 4701d60874cdSMiklos Szeredi res = inode->i_op->get_link(dentry, inode, done); 4702d60874cdSMiklos Szeredi } 4703d60874cdSMiklos Szeredi return res; 4704d60874cdSMiklos Szeredi } 4705d60874cdSMiklos Szeredi EXPORT_SYMBOL(vfs_get_link); 4706d60874cdSMiklos Szeredi 47071da177e4SLinus Torvalds /* get the link contents into pagecache */ 47086b255391SAl Viro const char *page_get_link(struct dentry *dentry, struct inode *inode, 4709fceef393SAl Viro struct delayed_call *callback) 47101da177e4SLinus Torvalds { 4711ebd09abbSDuane Griffin char *kaddr; 47121da177e4SLinus Torvalds struct page *page; 47136b255391SAl Viro struct address_space *mapping = inode->i_mapping; 47146b255391SAl Viro 4715d3883d4fSAl Viro if (!dentry) { 4716d3883d4fSAl Viro page = find_get_page(mapping, 0); 4717d3883d4fSAl Viro if (!page) 47186b255391SAl Viro return ERR_PTR(-ECHILD); 4719d3883d4fSAl Viro if (!PageUptodate(page)) { 4720d3883d4fSAl Viro put_page(page); 4721d3883d4fSAl Viro return ERR_PTR(-ECHILD); 4722d3883d4fSAl Viro } 4723d3883d4fSAl Viro } else { 4724090d2b18SPekka Enberg page = read_mapping_page(mapping, 0, NULL); 47251da177e4SLinus Torvalds if (IS_ERR(page)) 47266fe6900eSNick Piggin return (char*)page; 4727d3883d4fSAl Viro } 4728fceef393SAl Viro set_delayed_call(callback, page_put_link, page); 472921fc61c7SAl Viro BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM); 473021fc61c7SAl Viro kaddr = page_address(page); 47316b255391SAl Viro nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1); 4732ebd09abbSDuane Griffin return kaddr; 47331da177e4SLinus Torvalds } 47341da177e4SLinus Torvalds 47356b255391SAl Viro EXPORT_SYMBOL(page_get_link); 47361da177e4SLinus Torvalds 4737fceef393SAl Viro void page_put_link(void *arg) 47381da177e4SLinus Torvalds { 4739fceef393SAl Viro put_page(arg); 47401da177e4SLinus Torvalds } 47414d359507SAl Viro EXPORT_SYMBOL(page_put_link); 47421da177e4SLinus Torvalds 4743aa80deabSAl Viro int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 4744aa80deabSAl Viro { 4745fceef393SAl Viro DEFINE_DELAYED_CALL(done); 47466b255391SAl Viro int res = readlink_copy(buffer, buflen, 47476b255391SAl Viro page_get_link(dentry, d_inode(dentry), 4748fceef393SAl Viro &done)); 4749fceef393SAl Viro do_delayed_call(&done); 4750aa80deabSAl Viro return res; 4751aa80deabSAl Viro } 4752aa80deabSAl Viro EXPORT_SYMBOL(page_readlink); 4753aa80deabSAl Viro 475454566b2cSNick Piggin /* 475554566b2cSNick Piggin * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS 475654566b2cSNick Piggin */ 475754566b2cSNick Piggin int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) 47581da177e4SLinus Torvalds { 47591da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 47600adb25d2SKirill Korotaev struct page *page; 4761afddba49SNick Piggin void *fsdata; 4762beb497abSDmitriy Monakhov int err; 476354566b2cSNick Piggin unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE; 476454566b2cSNick Piggin if (nofs) 476554566b2cSNick Piggin flags |= AOP_FLAG_NOFS; 47661da177e4SLinus Torvalds 47677e53cac4SNeilBrown retry: 4768afddba49SNick Piggin err = pagecache_write_begin(NULL, mapping, 0, len-1, 476954566b2cSNick Piggin flags, &page, &fsdata); 47701da177e4SLinus Torvalds if (err) 4771afddba49SNick Piggin goto fail; 4772afddba49SNick Piggin 477321fc61c7SAl Viro memcpy(page_address(page), symname, len-1); 4774afddba49SNick Piggin 4775afddba49SNick Piggin err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, 4776afddba49SNick Piggin page, fsdata); 47771da177e4SLinus Torvalds if (err < 0) 47781da177e4SLinus Torvalds goto fail; 4779afddba49SNick Piggin if (err < len-1) 4780afddba49SNick Piggin goto retry; 4781afddba49SNick Piggin 47821da177e4SLinus Torvalds mark_inode_dirty(inode); 47831da177e4SLinus Torvalds return 0; 47841da177e4SLinus Torvalds fail: 47851da177e4SLinus Torvalds return err; 47861da177e4SLinus Torvalds } 47874d359507SAl Viro EXPORT_SYMBOL(__page_symlink); 47881da177e4SLinus Torvalds 47890adb25d2SKirill Korotaev int page_symlink(struct inode *inode, const char *symname, int len) 47900adb25d2SKirill Korotaev { 47910adb25d2SKirill Korotaev return __page_symlink(inode, symname, len, 4792c62d2555SMichal Hocko !mapping_gfp_constraint(inode->i_mapping, __GFP_FS)); 47930adb25d2SKirill Korotaev } 47944d359507SAl Viro EXPORT_SYMBOL(page_symlink); 47950adb25d2SKirill Korotaev 479692e1d5beSArjan van de Ven const struct inode_operations page_symlink_inode_operations = { 47976b255391SAl Viro .get_link = page_get_link, 47981da177e4SLinus Torvalds }; 47991da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink_inode_operations); 4800