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 67519660af7SAl Viro * @dentry: child of nd->path.dentry or NULL 6766e9918b7SAl Viro * @seq: seq number to check dentry against 67739191628SRandy Dunlap * Returns: 0 on success, -ECHILD on failure 67831e6b01fSNick Piggin * 67919660af7SAl Viro * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry 68019660af7SAl Viro * for ref-walk mode. @dentry must be a path found by a do_lookup call on 68119660af7SAl Viro * @nd or NULL. Must be called from rcu-walk context. 6827973387aSAl Viro * Nothing should touch nameidata between unlazy_walk() failure and 6837973387aSAl Viro * terminate_walk(). 68431e6b01fSNick Piggin */ 6856e9918b7SAl Viro static int unlazy_walk(struct nameidata *nd, struct dentry *dentry, unsigned seq) 68631e6b01fSNick Piggin { 68731e6b01fSNick Piggin struct dentry *parent = nd->path.dentry; 68831e6b01fSNick Piggin 68931e6b01fSNick Piggin BUG_ON(!(nd->flags & LOOKUP_RCU)); 690e5c832d5SLinus Torvalds 691e5c832d5SLinus Torvalds nd->flags &= ~LOOKUP_RCU; 6927973387aSAl Viro if (unlikely(!legitimize_links(nd))) 6937973387aSAl Viro goto out2; 6947973387aSAl Viro if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq))) 6957973387aSAl Viro goto out2; 6967973387aSAl Viro if (unlikely(!lockref_get_not_dead(&parent->d_lockref))) 6977973387aSAl Viro goto out1; 69848a066e7SAl Viro 69915570086SLinus Torvalds /* 70015570086SLinus Torvalds * For a negative lookup, the lookup sequence point is the parents 70115570086SLinus Torvalds * sequence point, and it only needs to revalidate the parent dentry. 70215570086SLinus Torvalds * 70315570086SLinus Torvalds * For a positive lookup, we need to move both the parent and the 70415570086SLinus Torvalds * dentry from the RCU domain to be properly refcounted. And the 70515570086SLinus Torvalds * sequence number in the dentry validates *both* dentry counters, 70615570086SLinus Torvalds * since we checked the sequence number of the parent after we got 70715570086SLinus Torvalds * the child sequence number. So we know the parent must still 70815570086SLinus Torvalds * be valid if the child sequence number is still valid. 70915570086SLinus Torvalds */ 71019660af7SAl Viro if (!dentry) { 711e5c832d5SLinus Torvalds if (read_seqcount_retry(&parent->d_seq, nd->seq)) 712e5c832d5SLinus Torvalds goto out; 71319660af7SAl Viro BUG_ON(nd->inode != parent->d_inode); 71419660af7SAl Viro } else { 715e5c832d5SLinus Torvalds if (!lockref_get_not_dead(&dentry->d_lockref)) 716e5c832d5SLinus Torvalds goto out; 7176e9918b7SAl Viro if (read_seqcount_retry(&dentry->d_seq, seq)) 718e5c832d5SLinus Torvalds goto drop_dentry; 71919660af7SAl Viro } 720e5c832d5SLinus Torvalds 721e5c832d5SLinus Torvalds /* 722e5c832d5SLinus Torvalds * Sequence counts matched. Now make sure that the root is 723e5c832d5SLinus Torvalds * still valid and get it if required. 724e5c832d5SLinus Torvalds */ 725e5c832d5SLinus Torvalds if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 7265a8d87e8SAl Viro if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) { 7275a8d87e8SAl Viro rcu_read_unlock(); 7285a8d87e8SAl Viro dput(dentry); 7295a8d87e8SAl Viro return -ECHILD; 7307973387aSAl Viro } 73131e6b01fSNick Piggin } 73231e6b01fSNick Piggin 7338b61e74fSAl Viro rcu_read_unlock(); 73431e6b01fSNick Piggin return 0; 73519660af7SAl Viro 736e5c832d5SLinus Torvalds drop_dentry: 7378b61e74fSAl Viro rcu_read_unlock(); 738e5c832d5SLinus Torvalds dput(dentry); 739d0d27277SLinus Torvalds goto drop_root_mnt; 7407973387aSAl Viro out2: 7417973387aSAl Viro nd->path.mnt = NULL; 7427973387aSAl Viro out1: 7437973387aSAl Viro nd->path.dentry = NULL; 744e5c832d5SLinus Torvalds out: 7458b61e74fSAl Viro rcu_read_unlock(); 746d0d27277SLinus Torvalds drop_root_mnt: 747d0d27277SLinus Torvalds if (!(nd->flags & LOOKUP_ROOT)) 748d0d27277SLinus Torvalds nd->root.mnt = NULL; 74931e6b01fSNick Piggin return -ECHILD; 75031e6b01fSNick Piggin } 75131e6b01fSNick Piggin 7527973387aSAl Viro static int unlazy_link(struct nameidata *nd, struct path *link, unsigned seq) 7537973387aSAl Viro { 7547973387aSAl Viro if (unlikely(!legitimize_path(nd, link, seq))) { 7557973387aSAl Viro drop_links(nd); 7567973387aSAl Viro nd->depth = 0; 7577973387aSAl Viro nd->flags &= ~LOOKUP_RCU; 7587973387aSAl Viro nd->path.mnt = NULL; 7597973387aSAl Viro nd->path.dentry = NULL; 7607973387aSAl Viro if (!(nd->flags & LOOKUP_ROOT)) 7617973387aSAl Viro nd->root.mnt = NULL; 7627973387aSAl Viro rcu_read_unlock(); 7637973387aSAl Viro } else if (likely(unlazy_walk(nd, NULL, 0)) == 0) { 7647973387aSAl Viro return 0; 7657973387aSAl Viro } 7667973387aSAl Viro path_put(link); 7677973387aSAl Viro return -ECHILD; 7687973387aSAl Viro } 7697973387aSAl Viro 7704ce16ef3SAl Viro static inline int d_revalidate(struct dentry *dentry, unsigned int flags) 77134286d66SNick Piggin { 7724ce16ef3SAl Viro return dentry->d_op->d_revalidate(dentry, flags); 77334286d66SNick Piggin } 77434286d66SNick Piggin 7759f1fafeeSAl Viro /** 7769f1fafeeSAl Viro * complete_walk - successful completion of path walk 7779f1fafeeSAl Viro * @nd: pointer nameidata 77839159de2SJeff Layton * 7799f1fafeeSAl Viro * If we had been in RCU mode, drop out of it and legitimize nd->path. 7809f1fafeeSAl Viro * Revalidate the final result, unless we'd already done that during 7819f1fafeeSAl Viro * the path walk or the filesystem doesn't ask for it. Return 0 on 7829f1fafeeSAl Viro * success, -error on failure. In case of failure caller does not 7839f1fafeeSAl Viro * need to drop nd->path. 78439159de2SJeff Layton */ 7859f1fafeeSAl Viro static int complete_walk(struct nameidata *nd) 78639159de2SJeff Layton { 78716c2cd71SAl Viro struct dentry *dentry = nd->path.dentry; 78839159de2SJeff Layton int status; 78939159de2SJeff Layton 7909f1fafeeSAl Viro if (nd->flags & LOOKUP_RCU) { 7919f1fafeeSAl Viro if (!(nd->flags & LOOKUP_ROOT)) 7929f1fafeeSAl Viro nd->root.mnt = NULL; 7936e9918b7SAl Viro if (unlikely(unlazy_walk(nd, NULL, 0))) 79448a066e7SAl Viro return -ECHILD; 79548a066e7SAl Viro } 7969f1fafeeSAl Viro 79716c2cd71SAl Viro if (likely(!(nd->flags & LOOKUP_JUMPED))) 79839159de2SJeff Layton return 0; 79939159de2SJeff Layton 800ecf3d1f1SJeff Layton if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) 80116c2cd71SAl Viro return 0; 80216c2cd71SAl Viro 803ecf3d1f1SJeff Layton status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); 80439159de2SJeff Layton if (status > 0) 80539159de2SJeff Layton return 0; 80639159de2SJeff Layton 80716c2cd71SAl Viro if (!status) 80839159de2SJeff Layton status = -ESTALE; 80916c2cd71SAl Viro 81039159de2SJeff Layton return status; 81139159de2SJeff Layton } 81239159de2SJeff Layton 81318d8c860SAl Viro static void set_root(struct nameidata *nd) 8142a737871SAl Viro { 81531e6b01fSNick Piggin struct fs_struct *fs = current->fs; 8169e6697e2SAl Viro 8179e6697e2SAl Viro if (nd->flags & LOOKUP_RCU) { 8188f47a016SAl Viro unsigned seq; 819c28cc364SNick Piggin 820c28cc364SNick Piggin do { 821c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 82231e6b01fSNick Piggin nd->root = fs->root; 8238f47a016SAl Viro nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq); 824c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 8259e6697e2SAl Viro } else { 8269e6697e2SAl Viro get_fs_root(fs, &nd->root); 8279e6697e2SAl Viro } 82831e6b01fSNick Piggin } 82931e6b01fSNick Piggin 8301d957f9bSJan Blunck static void path_put_conditional(struct path *path, struct nameidata *nd) 831051d3812SIan Kent { 832051d3812SIan Kent dput(path->dentry); 8334ac91378SJan Blunck if (path->mnt != nd->path.mnt) 834051d3812SIan Kent mntput(path->mnt); 835051d3812SIan Kent } 836051d3812SIan Kent 8377b9337aaSNick Piggin static inline void path_to_nameidata(const struct path *path, 8387b9337aaSNick Piggin struct nameidata *nd) 839051d3812SIan Kent { 84031e6b01fSNick Piggin if (!(nd->flags & LOOKUP_RCU)) { 8414ac91378SJan Blunck dput(nd->path.dentry); 84231e6b01fSNick Piggin if (nd->path.mnt != path->mnt) 8434ac91378SJan Blunck mntput(nd->path.mnt); 8449a229683SHuang Shijie } 84531e6b01fSNick Piggin nd->path.mnt = path->mnt; 8464ac91378SJan Blunck nd->path.dentry = path->dentry; 847051d3812SIan Kent } 848051d3812SIan Kent 849248fb5b9SAl Viro static int nd_jump_root(struct nameidata *nd) 850248fb5b9SAl Viro { 851248fb5b9SAl Viro if (nd->flags & LOOKUP_RCU) { 852248fb5b9SAl Viro struct dentry *d; 853248fb5b9SAl Viro nd->path = nd->root; 854248fb5b9SAl Viro d = nd->path.dentry; 855248fb5b9SAl Viro nd->inode = d->d_inode; 856248fb5b9SAl Viro nd->seq = nd->root_seq; 857248fb5b9SAl Viro if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq))) 858248fb5b9SAl Viro return -ECHILD; 859248fb5b9SAl Viro } else { 860248fb5b9SAl Viro path_put(&nd->path); 861248fb5b9SAl Viro nd->path = nd->root; 862248fb5b9SAl Viro path_get(&nd->path); 863248fb5b9SAl Viro nd->inode = nd->path.dentry->d_inode; 864248fb5b9SAl Viro } 865248fb5b9SAl Viro nd->flags |= LOOKUP_JUMPED; 866248fb5b9SAl Viro return 0; 867248fb5b9SAl Viro } 868248fb5b9SAl Viro 869b5fb63c1SChristoph Hellwig /* 8706b255391SAl Viro * Helper to directly jump to a known parsed path from ->get_link, 871b5fb63c1SChristoph Hellwig * caller must have taken a reference to path beforehand. 872b5fb63c1SChristoph Hellwig */ 8736e77137bSAl Viro void nd_jump_link(struct path *path) 874b5fb63c1SChristoph Hellwig { 8756e77137bSAl Viro struct nameidata *nd = current->nameidata; 876b5fb63c1SChristoph Hellwig path_put(&nd->path); 877b5fb63c1SChristoph Hellwig 878b5fb63c1SChristoph Hellwig nd->path = *path; 879b5fb63c1SChristoph Hellwig nd->inode = nd->path.dentry->d_inode; 880b5fb63c1SChristoph Hellwig nd->flags |= LOOKUP_JUMPED; 881b5fb63c1SChristoph Hellwig } 882b5fb63c1SChristoph Hellwig 883b9ff4429SAl Viro static inline void put_link(struct nameidata *nd) 884574197e0SAl Viro { 88521c3003dSAl Viro struct saved *last = nd->stack + --nd->depth; 886fceef393SAl Viro do_delayed_call(&last->done); 8876548fae2SAl Viro if (!(nd->flags & LOOKUP_RCU)) 888b9ff4429SAl Viro path_put(&last->link); 889574197e0SAl Viro } 890574197e0SAl Viro 891561ec64aSLinus Torvalds int sysctl_protected_symlinks __read_mostly = 0; 892561ec64aSLinus Torvalds int sysctl_protected_hardlinks __read_mostly = 0; 893800179c9SKees Cook 894800179c9SKees Cook /** 895800179c9SKees Cook * may_follow_link - Check symlink following for unsafe situations 89655852635SRandy Dunlap * @nd: nameidata pathwalk data 897800179c9SKees Cook * 898800179c9SKees Cook * In the case of the sysctl_protected_symlinks sysctl being enabled, 899800179c9SKees Cook * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is 900800179c9SKees Cook * in a sticky world-writable directory. This is to protect privileged 901800179c9SKees Cook * processes from failing races against path names that may change out 902800179c9SKees Cook * from under them by way of other users creating malicious symlinks. 903800179c9SKees Cook * It will permit symlinks to be followed only when outside a sticky 904800179c9SKees Cook * world-writable directory, or when the uid of the symlink and follower 905800179c9SKees Cook * match, or when the directory owner matches the symlink's owner. 906800179c9SKees Cook * 907800179c9SKees Cook * Returns 0 if following the symlink is allowed, -ve on error. 908800179c9SKees Cook */ 909fec2fa24SAl Viro static inline int may_follow_link(struct nameidata *nd) 910800179c9SKees Cook { 911800179c9SKees Cook const struct inode *inode; 912800179c9SKees Cook const struct inode *parent; 9132d7f9e2aSSeth Forshee kuid_t puid; 914800179c9SKees Cook 915800179c9SKees Cook if (!sysctl_protected_symlinks) 916800179c9SKees Cook return 0; 917800179c9SKees Cook 918800179c9SKees Cook /* Allowed if owner and follower match. */ 919fceef393SAl Viro inode = nd->link_inode; 92081abe27bSEric W. Biederman if (uid_eq(current_cred()->fsuid, inode->i_uid)) 921800179c9SKees Cook return 0; 922800179c9SKees Cook 923800179c9SKees Cook /* Allowed if parent directory not sticky and world-writable. */ 924aa65fa35SAl Viro parent = nd->inode; 925800179c9SKees Cook if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) 926800179c9SKees Cook return 0; 927800179c9SKees Cook 928800179c9SKees Cook /* Allowed if parent directory and link owner match. */ 9292d7f9e2aSSeth Forshee puid = parent->i_uid; 9302d7f9e2aSSeth Forshee if (uid_valid(puid) && uid_eq(puid, inode->i_uid)) 931800179c9SKees Cook return 0; 932800179c9SKees Cook 93331956502SAl Viro if (nd->flags & LOOKUP_RCU) 93431956502SAl Viro return -ECHILD; 93531956502SAl Viro 9361cf2665bSAl Viro audit_log_link_denied("follow_link", &nd->stack[0].link); 937800179c9SKees Cook return -EACCES; 938800179c9SKees Cook } 939800179c9SKees Cook 940800179c9SKees Cook /** 941800179c9SKees Cook * safe_hardlink_source - Check for safe hardlink conditions 942800179c9SKees Cook * @inode: the source inode to hardlink from 943800179c9SKees Cook * 944800179c9SKees Cook * Return false if at least one of the following conditions: 945800179c9SKees Cook * - inode is not a regular file 946800179c9SKees Cook * - inode is setuid 947800179c9SKees Cook * - inode is setgid and group-exec 948800179c9SKees Cook * - access failure for read and write 949800179c9SKees Cook * 950800179c9SKees Cook * Otherwise returns true. 951800179c9SKees Cook */ 952800179c9SKees Cook static bool safe_hardlink_source(struct inode *inode) 953800179c9SKees Cook { 954800179c9SKees Cook umode_t mode = inode->i_mode; 955800179c9SKees Cook 956800179c9SKees Cook /* Special files should not get pinned to the filesystem. */ 957800179c9SKees Cook if (!S_ISREG(mode)) 958800179c9SKees Cook return false; 959800179c9SKees Cook 960800179c9SKees Cook /* Setuid files should not get pinned to the filesystem. */ 961800179c9SKees Cook if (mode & S_ISUID) 962800179c9SKees Cook return false; 963800179c9SKees Cook 964800179c9SKees Cook /* Executable setgid files should not get pinned to the filesystem. */ 965800179c9SKees Cook if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) 966800179c9SKees Cook return false; 967800179c9SKees Cook 968800179c9SKees Cook /* Hardlinking to unreadable or unwritable sources is dangerous. */ 969800179c9SKees Cook if (inode_permission(inode, MAY_READ | MAY_WRITE)) 970800179c9SKees Cook return false; 971800179c9SKees Cook 972800179c9SKees Cook return true; 973800179c9SKees Cook } 974800179c9SKees Cook 975800179c9SKees Cook /** 976800179c9SKees Cook * may_linkat - Check permissions for creating a hardlink 977800179c9SKees Cook * @link: the source to hardlink from 978800179c9SKees Cook * 979800179c9SKees Cook * Block hardlink when all of: 980800179c9SKees Cook * - sysctl_protected_hardlinks enabled 981800179c9SKees Cook * - fsuid does not match inode 982800179c9SKees Cook * - hardlink source is unsafe (see safe_hardlink_source() above) 983f2ca3796SDirk Steinmetz * - not CAP_FOWNER in a namespace with the inode owner uid mapped 984800179c9SKees Cook * 985800179c9SKees Cook * Returns 0 if successful, -ve on error. 986800179c9SKees Cook */ 987800179c9SKees Cook static int may_linkat(struct path *link) 988800179c9SKees Cook { 989800179c9SKees Cook struct inode *inode; 990800179c9SKees Cook 991800179c9SKees Cook if (!sysctl_protected_hardlinks) 992800179c9SKees Cook return 0; 993800179c9SKees Cook 994800179c9SKees Cook inode = link->dentry->d_inode; 995800179c9SKees Cook 996800179c9SKees Cook /* Source inode owner (or CAP_FOWNER) can hardlink all they like, 997800179c9SKees Cook * otherwise, it must be a safe source. 998800179c9SKees Cook */ 999f2ca3796SDirk Steinmetz if (inode_owner_or_capable(inode) || safe_hardlink_source(inode)) 1000800179c9SKees Cook return 0; 1001800179c9SKees Cook 1002a51d9eaaSKees Cook audit_log_link_denied("linkat", link); 1003800179c9SKees Cook return -EPERM; 1004800179c9SKees Cook } 1005800179c9SKees Cook 10063b2e7f75SAl Viro static __always_inline 10073b2e7f75SAl Viro const char *get_link(struct nameidata *nd) 10081da177e4SLinus Torvalds { 1009ab104923SAl Viro struct saved *last = nd->stack + nd->depth - 1; 10101cf2665bSAl Viro struct dentry *dentry = last->link.dentry; 1011fceef393SAl Viro struct inode *inode = nd->link_inode; 10126d7b5aaeSAl Viro int error; 10130a959df5SAl Viro const char *res; 10141da177e4SLinus Torvalds 10158fa9dd24SNeilBrown if (!(nd->flags & LOOKUP_RCU)) { 10168fa9dd24SNeilBrown touch_atime(&last->link); 10178fa9dd24SNeilBrown cond_resched(); 1018598e3c8fSMiklos Szeredi } else if (atime_needs_update_rcu(&last->link, inode)) { 10198fa9dd24SNeilBrown if (unlikely(unlazy_walk(nd, NULL, 0))) 10208fa9dd24SNeilBrown return ERR_PTR(-ECHILD); 10218fa9dd24SNeilBrown touch_atime(&last->link); 10228fa9dd24SNeilBrown } 10238fa9dd24SNeilBrown 1024bda0be7aSNeilBrown error = security_inode_follow_link(dentry, inode, 1025bda0be7aSNeilBrown nd->flags & LOOKUP_RCU); 1026bda0be7aSNeilBrown if (unlikely(error)) 10276920a440SAl Viro return ERR_PTR(error); 102836f3b4f6SAl Viro 102986acdca1SAl Viro nd->last_type = LAST_BIND; 1030d4dee48bSAl Viro res = inode->i_link; 1031d4dee48bSAl Viro if (!res) { 1032fceef393SAl Viro const char * (*get)(struct dentry *, struct inode *, 1033fceef393SAl Viro struct delayed_call *); 1034fceef393SAl Viro get = inode->i_op->get_link; 10358c1b4566SAl Viro if (nd->flags & LOOKUP_RCU) { 1036fceef393SAl Viro res = get(NULL, inode, &last->done); 10376b255391SAl Viro if (res == ERR_PTR(-ECHILD)) { 10388c1b4566SAl Viro if (unlikely(unlazy_walk(nd, NULL, 0))) 10398c1b4566SAl Viro return ERR_PTR(-ECHILD); 1040fceef393SAl Viro res = get(dentry, inode, &last->done); 10418c1b4566SAl Viro } 10426b255391SAl Viro } else { 1043fceef393SAl Viro res = get(dentry, inode, &last->done); 10446b255391SAl Viro } 1045fceef393SAl Viro if (IS_ERR_OR_NULL(res)) 1046fab51e8aSAl Viro return res; 10470a959df5SAl Viro } 1048fab51e8aSAl Viro if (*res == '/') { 1049fab51e8aSAl Viro if (!nd->root.mnt) 1050fab51e8aSAl Viro set_root(nd); 1051248fb5b9SAl Viro if (unlikely(nd_jump_root(nd))) 1052fab51e8aSAl Viro return ERR_PTR(-ECHILD); 1053fab51e8aSAl Viro while (unlikely(*++res == '/')) 1054fab51e8aSAl Viro ; 1055fab51e8aSAl Viro } 1056fab51e8aSAl Viro if (!*res) 1057fab51e8aSAl Viro res = NULL; 10580a959df5SAl Viro return res; 10590a959df5SAl Viro } 10606d7b5aaeSAl Viro 1061f015f126SDavid Howells /* 1062f015f126SDavid Howells * follow_up - Find the mountpoint of path's vfsmount 1063f015f126SDavid Howells * 1064f015f126SDavid Howells * Given a path, find the mountpoint of its source file system. 1065f015f126SDavid Howells * Replace @path with the path of the mountpoint in the parent mount. 1066f015f126SDavid Howells * Up is towards /. 1067f015f126SDavid Howells * 1068f015f126SDavid Howells * Return 1 if we went up a level and 0 if we were already at the 1069f015f126SDavid Howells * root. 1070f015f126SDavid Howells */ 1071bab77ebfSAl Viro int follow_up(struct path *path) 10721da177e4SLinus Torvalds { 10730714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 10740714a533SAl Viro struct mount *parent; 10751da177e4SLinus Torvalds struct dentry *mountpoint; 107699b7db7bSNick Piggin 107748a066e7SAl Viro read_seqlock_excl(&mount_lock); 10780714a533SAl Viro parent = mnt->mnt_parent; 10793c0a6163SAl Viro if (parent == mnt) { 108048a066e7SAl Viro read_sequnlock_excl(&mount_lock); 10811da177e4SLinus Torvalds return 0; 10821da177e4SLinus Torvalds } 10830714a533SAl Viro mntget(&parent->mnt); 1084a73324daSAl Viro mountpoint = dget(mnt->mnt_mountpoint); 108548a066e7SAl Viro read_sequnlock_excl(&mount_lock); 1086bab77ebfSAl Viro dput(path->dentry); 1087bab77ebfSAl Viro path->dentry = mountpoint; 1088bab77ebfSAl Viro mntput(path->mnt); 10890714a533SAl Viro path->mnt = &parent->mnt; 10901da177e4SLinus Torvalds return 1; 10911da177e4SLinus Torvalds } 10924d359507SAl Viro EXPORT_SYMBOL(follow_up); 10931da177e4SLinus Torvalds 1094b5c84bf6SNick Piggin /* 10959875cf80SDavid Howells * Perform an automount 10969875cf80SDavid Howells * - return -EISDIR to tell follow_managed() to stop and return the path we 10979875cf80SDavid Howells * were called with. 10981da177e4SLinus Torvalds */ 1099756daf26SNeilBrown static int follow_automount(struct path *path, struct nameidata *nd, 11009875cf80SDavid Howells bool *need_mntput) 110131e6b01fSNick Piggin { 11029875cf80SDavid Howells struct vfsmount *mnt; 1103aeaa4a79SEric W. Biederman const struct cred *old_cred; 1104ea5b778aSDavid Howells int err; 11059875cf80SDavid Howells 11069875cf80SDavid Howells if (!path->dentry->d_op || !path->dentry->d_op->d_automount) 11079875cf80SDavid Howells return -EREMOTE; 11089875cf80SDavid Howells 11090ec26fd0SMiklos Szeredi /* We don't want to mount if someone's just doing a stat - 11100ec26fd0SMiklos Szeredi * unless they're stat'ing a directory and appended a '/' to 11110ec26fd0SMiklos Szeredi * the name. 11120ec26fd0SMiklos Szeredi * 11130ec26fd0SMiklos Szeredi * We do, however, want to mount if someone wants to open or 11145a30d8a2SDavid Howells * create a file of any type under the mountpoint, wants to 11150ec26fd0SMiklos Szeredi * traverse through the mountpoint or wants to open the 11160ec26fd0SMiklos Szeredi * mounted directory. Also, autofs may mark negative dentries 11170ec26fd0SMiklos Szeredi * as being automount points. These will need the attentions 11180ec26fd0SMiklos Szeredi * of the daemon to instantiate them before they can be used. 11195a30d8a2SDavid Howells */ 1120756daf26SNeilBrown if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | 1121d94c177bSLinus Torvalds LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && 11225a30d8a2SDavid Howells path->dentry->d_inode) 11239875cf80SDavid Howells return -EISDIR; 11240ec26fd0SMiklos Szeredi 1125aeaa4a79SEric W. Biederman if (path->dentry->d_sb->s_user_ns != &init_user_ns) 1126aeaa4a79SEric W. Biederman return -EACCES; 1127aeaa4a79SEric W. Biederman 1128756daf26SNeilBrown nd->total_link_count++; 1129756daf26SNeilBrown if (nd->total_link_count >= 40) 11309875cf80SDavid Howells return -ELOOP; 11319875cf80SDavid Howells 1132aeaa4a79SEric W. Biederman old_cred = override_creds(&init_cred); 11339875cf80SDavid Howells mnt = path->dentry->d_op->d_automount(path); 1134aeaa4a79SEric W. Biederman revert_creds(old_cred); 11359875cf80SDavid Howells if (IS_ERR(mnt)) { 11369875cf80SDavid Howells /* 11379875cf80SDavid Howells * The filesystem is allowed to return -EISDIR here to indicate 11389875cf80SDavid Howells * it doesn't want to automount. For instance, autofs would do 11399875cf80SDavid Howells * this so that its userspace daemon can mount on this dentry. 11409875cf80SDavid Howells * 11419875cf80SDavid Howells * However, we can only permit this if it's a terminal point in 11429875cf80SDavid Howells * the path being looked up; if it wasn't then the remainder of 11439875cf80SDavid Howells * the path is inaccessible and we should say so. 11449875cf80SDavid Howells */ 1145756daf26SNeilBrown if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT)) 11469875cf80SDavid Howells return -EREMOTE; 11479875cf80SDavid Howells return PTR_ERR(mnt); 114831e6b01fSNick Piggin } 1149ea5b778aSDavid Howells 11509875cf80SDavid Howells if (!mnt) /* mount collision */ 11519875cf80SDavid Howells return 0; 11529875cf80SDavid Howells 11538aef1884SAl Viro if (!*need_mntput) { 11548aef1884SAl Viro /* lock_mount() may release path->mnt on error */ 11558aef1884SAl Viro mntget(path->mnt); 11568aef1884SAl Viro *need_mntput = true; 11578aef1884SAl Viro } 115819a167afSAl Viro err = finish_automount(mnt, path); 1159ea5b778aSDavid Howells 1160ea5b778aSDavid Howells switch (err) { 1161ea5b778aSDavid Howells case -EBUSY: 1162ea5b778aSDavid Howells /* Someone else made a mount here whilst we were busy */ 116319a167afSAl Viro return 0; 1164ea5b778aSDavid Howells case 0: 11658aef1884SAl Viro path_put(path); 11669875cf80SDavid Howells path->mnt = mnt; 11679875cf80SDavid Howells path->dentry = dget(mnt->mnt_root); 11689875cf80SDavid Howells return 0; 116919a167afSAl Viro default: 117019a167afSAl Viro return err; 11719875cf80SDavid Howells } 117219a167afSAl Viro 1173ea5b778aSDavid Howells } 11749875cf80SDavid Howells 11759875cf80SDavid Howells /* 11769875cf80SDavid Howells * Handle a dentry that is managed in some way. 1177cc53ce53SDavid Howells * - Flagged for transit management (autofs) 11789875cf80SDavid Howells * - Flagged as mountpoint 11799875cf80SDavid Howells * - Flagged as automount point 11809875cf80SDavid Howells * 11819875cf80SDavid Howells * This may only be called in refwalk mode. 11829875cf80SDavid Howells * 11839875cf80SDavid Howells * Serialization is taken care of in namespace.c 11849875cf80SDavid Howells */ 1185756daf26SNeilBrown static int follow_managed(struct path *path, struct nameidata *nd) 11869875cf80SDavid Howells { 11878aef1884SAl Viro struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */ 11889875cf80SDavid Howells unsigned managed; 11899875cf80SDavid Howells bool need_mntput = false; 11908aef1884SAl Viro int ret = 0; 11919875cf80SDavid Howells 11929875cf80SDavid Howells /* Given that we're not holding a lock here, we retain the value in a 11939875cf80SDavid Howells * local variable for each dentry as we look at it so that we don't see 11949875cf80SDavid Howells * the components of that value change under us */ 11959875cf80SDavid Howells while (managed = ACCESS_ONCE(path->dentry->d_flags), 11969875cf80SDavid Howells managed &= DCACHE_MANAGED_DENTRY, 11979875cf80SDavid Howells unlikely(managed != 0)) { 1198cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1199cc53ce53SDavid Howells * being held. */ 1200cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 1201cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1202cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 1203fb5f51c7SIan Kent ret = path->dentry->d_op->d_manage(path, false); 1204cc53ce53SDavid Howells if (ret < 0) 12058aef1884SAl Viro break; 1206cc53ce53SDavid Howells } 1207cc53ce53SDavid Howells 12089875cf80SDavid Howells /* Transit to a mounted filesystem. */ 12099875cf80SDavid Howells if (managed & DCACHE_MOUNTED) { 12109875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 12119875cf80SDavid Howells if (mounted) { 12129875cf80SDavid Howells dput(path->dentry); 12139875cf80SDavid Howells if (need_mntput) 1214463ffb2eSAl Viro mntput(path->mnt); 1215463ffb2eSAl Viro path->mnt = mounted; 1216463ffb2eSAl Viro path->dentry = dget(mounted->mnt_root); 12179875cf80SDavid Howells need_mntput = true; 12189875cf80SDavid Howells continue; 1219463ffb2eSAl Viro } 1220463ffb2eSAl Viro 12219875cf80SDavid Howells /* Something is mounted on this dentry in another 12229875cf80SDavid Howells * namespace and/or whatever was mounted there in this 122348a066e7SAl Viro * namespace got unmounted before lookup_mnt() could 122448a066e7SAl Viro * get it */ 12251da177e4SLinus Torvalds } 12269875cf80SDavid Howells 12279875cf80SDavid Howells /* Handle an automount point */ 12289875cf80SDavid Howells if (managed & DCACHE_NEED_AUTOMOUNT) { 1229756daf26SNeilBrown ret = follow_automount(path, nd, &need_mntput); 12309875cf80SDavid Howells if (ret < 0) 12318aef1884SAl Viro break; 12329875cf80SDavid Howells continue; 12339875cf80SDavid Howells } 12349875cf80SDavid Howells 12359875cf80SDavid Howells /* We didn't change the current path point */ 12369875cf80SDavid Howells break; 12379875cf80SDavid Howells } 12388aef1884SAl Viro 12398aef1884SAl Viro if (need_mntput && path->mnt == mnt) 12408aef1884SAl Viro mntput(path->mnt); 1241e9742b53SAl Viro if (ret == -EISDIR || !ret) 1242e9742b53SAl Viro ret = 1; 12438402752eSAl Viro if (need_mntput) 12448402752eSAl Viro nd->flags |= LOOKUP_JUMPED; 12458402752eSAl Viro if (unlikely(ret < 0)) 12468402752eSAl Viro path_put_conditional(path, nd); 12478402752eSAl Viro return ret; 12481da177e4SLinus Torvalds } 12491da177e4SLinus Torvalds 1250cc53ce53SDavid Howells int follow_down_one(struct path *path) 12511da177e4SLinus Torvalds { 12521da177e4SLinus Torvalds struct vfsmount *mounted; 12531da177e4SLinus Torvalds 12541c755af4SAl Viro mounted = lookup_mnt(path); 12551da177e4SLinus Torvalds if (mounted) { 12569393bd07SAl Viro dput(path->dentry); 12579393bd07SAl Viro mntput(path->mnt); 12589393bd07SAl Viro path->mnt = mounted; 12599393bd07SAl Viro path->dentry = dget(mounted->mnt_root); 12601da177e4SLinus Torvalds return 1; 12611da177e4SLinus Torvalds } 12621da177e4SLinus Torvalds return 0; 12631da177e4SLinus Torvalds } 12644d359507SAl Viro EXPORT_SYMBOL(follow_down_one); 12651da177e4SLinus Torvalds 1266fb5f51c7SIan Kent static inline int managed_dentry_rcu(const struct path *path) 126762a7375eSIan Kent { 1268fb5f51c7SIan Kent return (path->dentry->d_flags & DCACHE_MANAGE_TRANSIT) ? 1269fb5f51c7SIan Kent path->dentry->d_op->d_manage(path, true) : 0; 127062a7375eSIan Kent } 127162a7375eSIan Kent 12729875cf80SDavid Howells /* 1273287548e4SAl Viro * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if 1274287548e4SAl Viro * we meet a managed dentry that would need blocking. 12759875cf80SDavid Howells */ 12769875cf80SDavid Howells static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, 1277254cf582SAl Viro struct inode **inode, unsigned *seqp) 12789875cf80SDavid Howells { 127962a7375eSIan Kent for (;;) { 1280c7105365SAl Viro struct mount *mounted; 128162a7375eSIan Kent /* 128262a7375eSIan Kent * Don't forget we might have a non-mountpoint managed dentry 128362a7375eSIan Kent * that wants to block transit. 128462a7375eSIan Kent */ 1285fb5f51c7SIan Kent switch (managed_dentry_rcu(path)) { 1286b8faf035SNeilBrown case -ECHILD: 1287b8faf035SNeilBrown default: 1288ab90911fSDavid Howells return false; 1289b8faf035SNeilBrown case -EISDIR: 1290b8faf035SNeilBrown return true; 1291b8faf035SNeilBrown case 0: 1292b8faf035SNeilBrown break; 1293b8faf035SNeilBrown } 129462a7375eSIan Kent 129562a7375eSIan Kent if (!d_mountpoint(path->dentry)) 1296b8faf035SNeilBrown return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); 129762a7375eSIan Kent 1298474279dcSAl Viro mounted = __lookup_mnt(path->mnt, path->dentry); 12999875cf80SDavid Howells if (!mounted) 13009875cf80SDavid Howells break; 1301c7105365SAl Viro path->mnt = &mounted->mnt; 1302c7105365SAl Viro path->dentry = mounted->mnt.mnt_root; 1303a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 1304254cf582SAl Viro *seqp = read_seqcount_begin(&path->dentry->d_seq); 130559430262SLinus Torvalds /* 130659430262SLinus Torvalds * Update the inode too. We don't need to re-check the 130759430262SLinus Torvalds * dentry sequence number here after this d_inode read, 130859430262SLinus Torvalds * because a mount-point is always pinned. 130959430262SLinus Torvalds */ 131059430262SLinus Torvalds *inode = path->dentry->d_inode; 13119875cf80SDavid Howells } 1312f5be3e29SAl Viro return !read_seqretry(&mount_lock, nd->m_seq) && 1313b8faf035SNeilBrown !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); 1314287548e4SAl Viro } 1315287548e4SAl Viro 131631e6b01fSNick Piggin static int follow_dotdot_rcu(struct nameidata *nd) 131731e6b01fSNick Piggin { 13184023bfc9SAl Viro struct inode *inode = nd->inode; 131931e6b01fSNick Piggin 132031e6b01fSNick Piggin while (1) { 1321aed434adSAl Viro if (path_equal(&nd->path, &nd->root)) 132231e6b01fSNick Piggin break; 132331e6b01fSNick Piggin if (nd->path.dentry != nd->path.mnt->mnt_root) { 132431e6b01fSNick Piggin struct dentry *old = nd->path.dentry; 132531e6b01fSNick Piggin struct dentry *parent = old->d_parent; 132631e6b01fSNick Piggin unsigned seq; 132731e6b01fSNick Piggin 13284023bfc9SAl Viro inode = parent->d_inode; 132931e6b01fSNick Piggin seq = read_seqcount_begin(&parent->d_seq); 1330aed434adSAl Viro if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq))) 1331aed434adSAl Viro return -ECHILD; 133231e6b01fSNick Piggin nd->path.dentry = parent; 133331e6b01fSNick Piggin nd->seq = seq; 1334397d425dSEric W. Biederman if (unlikely(!path_connected(&nd->path))) 1335397d425dSEric W. Biederman return -ENOENT; 133631e6b01fSNick Piggin break; 1337aed434adSAl Viro } else { 1338aed434adSAl Viro struct mount *mnt = real_mount(nd->path.mnt); 1339aed434adSAl Viro struct mount *mparent = mnt->mnt_parent; 1340aed434adSAl Viro struct dentry *mountpoint = mnt->mnt_mountpoint; 1341aed434adSAl Viro struct inode *inode2 = mountpoint->d_inode; 1342aed434adSAl Viro unsigned seq = read_seqcount_begin(&mountpoint->d_seq); 1343aed434adSAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1344aed434adSAl Viro return -ECHILD; 1345aed434adSAl Viro if (&mparent->mnt == nd->path.mnt) 134631e6b01fSNick Piggin break; 1347aed434adSAl Viro /* we know that mountpoint was pinned */ 1348aed434adSAl Viro nd->path.dentry = mountpoint; 1349aed434adSAl Viro nd->path.mnt = &mparent->mnt; 1350aed434adSAl Viro inode = inode2; 1351aed434adSAl Viro nd->seq = seq; 135231e6b01fSNick Piggin } 1353aed434adSAl Viro } 1354aed434adSAl Viro while (unlikely(d_mountpoint(nd->path.dentry))) { 1355b37199e6SAl Viro struct mount *mounted; 1356b37199e6SAl Viro mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry); 1357aed434adSAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1358aed434adSAl Viro return -ECHILD; 1359b37199e6SAl Viro if (!mounted) 1360b37199e6SAl Viro break; 1361b37199e6SAl Viro nd->path.mnt = &mounted->mnt; 1362b37199e6SAl Viro nd->path.dentry = mounted->mnt.mnt_root; 13634023bfc9SAl Viro inode = nd->path.dentry->d_inode; 1364b37199e6SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 1365b37199e6SAl Viro } 13664023bfc9SAl Viro nd->inode = inode; 136731e6b01fSNick Piggin return 0; 136831e6b01fSNick Piggin } 136931e6b01fSNick Piggin 13709875cf80SDavid Howells /* 1371cc53ce53SDavid Howells * Follow down to the covering mount currently visible to userspace. At each 1372cc53ce53SDavid Howells * point, the filesystem owning that dentry may be queried as to whether the 1373cc53ce53SDavid Howells * caller is permitted to proceed or not. 1374cc53ce53SDavid Howells */ 13757cc90cc3SAl Viro int follow_down(struct path *path) 1376cc53ce53SDavid Howells { 1377cc53ce53SDavid Howells unsigned managed; 1378cc53ce53SDavid Howells int ret; 1379cc53ce53SDavid Howells 1380cc53ce53SDavid Howells while (managed = ACCESS_ONCE(path->dentry->d_flags), 1381cc53ce53SDavid Howells unlikely(managed & DCACHE_MANAGED_DENTRY)) { 1382cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1383cc53ce53SDavid Howells * being held. 1384cc53ce53SDavid Howells * 1385cc53ce53SDavid Howells * We indicate to the filesystem if someone is trying to mount 1386cc53ce53SDavid Howells * something here. This gives autofs the chance to deny anyone 1387cc53ce53SDavid Howells * other than its daemon the right to mount on its 1388cc53ce53SDavid Howells * superstructure. 1389cc53ce53SDavid Howells * 1390cc53ce53SDavid Howells * The filesystem may sleep at this point. 1391cc53ce53SDavid Howells */ 1392cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 1393cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1394cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 1395fb5f51c7SIan Kent ret = path->dentry->d_op->d_manage(path, false); 1396cc53ce53SDavid Howells if (ret < 0) 1397cc53ce53SDavid Howells return ret == -EISDIR ? 0 : ret; 1398cc53ce53SDavid Howells } 1399cc53ce53SDavid Howells 1400cc53ce53SDavid Howells /* Transit to a mounted filesystem. */ 1401cc53ce53SDavid Howells if (managed & DCACHE_MOUNTED) { 1402cc53ce53SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 1403cc53ce53SDavid Howells if (!mounted) 1404cc53ce53SDavid Howells break; 1405cc53ce53SDavid Howells dput(path->dentry); 1406cc53ce53SDavid Howells mntput(path->mnt); 1407cc53ce53SDavid Howells path->mnt = mounted; 1408cc53ce53SDavid Howells path->dentry = dget(mounted->mnt_root); 1409cc53ce53SDavid Howells continue; 1410cc53ce53SDavid Howells } 1411cc53ce53SDavid Howells 1412cc53ce53SDavid Howells /* Don't handle automount points here */ 1413cc53ce53SDavid Howells break; 1414cc53ce53SDavid Howells } 1415cc53ce53SDavid Howells return 0; 1416cc53ce53SDavid Howells } 14174d359507SAl Viro EXPORT_SYMBOL(follow_down); 1418cc53ce53SDavid Howells 1419cc53ce53SDavid Howells /* 14209875cf80SDavid Howells * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() 14219875cf80SDavid Howells */ 14229875cf80SDavid Howells static void follow_mount(struct path *path) 14239875cf80SDavid Howells { 14249875cf80SDavid Howells while (d_mountpoint(path->dentry)) { 14259875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 14269875cf80SDavid Howells if (!mounted) 14279875cf80SDavid Howells break; 14289875cf80SDavid Howells dput(path->dentry); 14299875cf80SDavid Howells mntput(path->mnt); 14309875cf80SDavid Howells path->mnt = mounted; 14319875cf80SDavid Howells path->dentry = dget(mounted->mnt_root); 14329875cf80SDavid Howells } 14339875cf80SDavid Howells } 14349875cf80SDavid Howells 1435eedf265aSEric W. Biederman static int path_parent_directory(struct path *path) 1436eedf265aSEric W. Biederman { 1437eedf265aSEric W. Biederman struct dentry *old = path->dentry; 1438eedf265aSEric W. Biederman /* rare case of legitimate dget_parent()... */ 1439eedf265aSEric W. Biederman path->dentry = dget_parent(path->dentry); 1440eedf265aSEric W. Biederman dput(old); 1441eedf265aSEric W. Biederman if (unlikely(!path_connected(path))) 1442eedf265aSEric W. Biederman return -ENOENT; 1443eedf265aSEric W. Biederman return 0; 1444eedf265aSEric W. Biederman } 1445eedf265aSEric W. Biederman 1446397d425dSEric W. Biederman static int follow_dotdot(struct nameidata *nd) 14471da177e4SLinus Torvalds { 14481da177e4SLinus Torvalds while(1) { 14492a737871SAl Viro if (nd->path.dentry == nd->root.dentry && 14502a737871SAl Viro nd->path.mnt == nd->root.mnt) { 14511da177e4SLinus Torvalds break; 14521da177e4SLinus Torvalds } 14534ac91378SJan Blunck if (nd->path.dentry != nd->path.mnt->mnt_root) { 1454eedf265aSEric W. Biederman int ret = path_parent_directory(&nd->path); 1455eedf265aSEric W. Biederman if (ret) 1456eedf265aSEric W. Biederman return ret; 14571da177e4SLinus Torvalds break; 14581da177e4SLinus Torvalds } 14593088dd70SAl Viro if (!follow_up(&nd->path)) 14601da177e4SLinus Torvalds break; 14611da177e4SLinus Torvalds } 146279ed0226SAl Viro follow_mount(&nd->path); 146331e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 1464397d425dSEric W. Biederman return 0; 14651da177e4SLinus Torvalds } 14661da177e4SLinus Torvalds 14671da177e4SLinus Torvalds /* 1468f4fdace9SOleg Drokin * This looks up the name in dcache and possibly revalidates the found dentry. 1469f4fdace9SOleg Drokin * NULL is returned if the dentry does not exist in the cache. 1470baa03890SNick Piggin */ 1471e3c13928SAl Viro static struct dentry *lookup_dcache(const struct qstr *name, 1472e3c13928SAl Viro struct dentry *dir, 14736c51e513SAl Viro unsigned int flags) 1474baa03890SNick Piggin { 1475baa03890SNick Piggin struct dentry *dentry; 1476bad61189SMiklos Szeredi int error; 1477baa03890SNick Piggin 1478bad61189SMiklos Szeredi dentry = d_lookup(dir, name); 1479bad61189SMiklos Szeredi if (dentry) { 148039e3c955SJeff Layton if (dentry->d_flags & DCACHE_OP_REVALIDATE) { 1481201f956eSAl Viro error = d_revalidate(dentry, flags); 1482bad61189SMiklos Szeredi if (unlikely(error <= 0)) { 148374ff0ffcSAl Viro if (!error) 14845542aa2fSEric W. Biederman d_invalidate(dentry); 1485bad61189SMiklos Szeredi dput(dentry); 148674ff0ffcSAl Viro return ERR_PTR(error); 1487bad61189SMiklos Szeredi } 1488bad61189SMiklos Szeredi } 1489bad61189SMiklos Szeredi } 1490baa03890SNick Piggin return dentry; 1491baa03890SNick Piggin } 1492baa03890SNick Piggin 1493baa03890SNick Piggin /* 149413a2c3beSJ. Bruce Fields * Call i_op->lookup on the dentry. The dentry must be negative and 149513a2c3beSJ. Bruce Fields * unhashed. 1496bad61189SMiklos Szeredi * 1497bad61189SMiklos Szeredi * dir->d_inode->i_mutex must be held 149844396f4bSJosef Bacik */ 1499bad61189SMiklos Szeredi static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry, 150072bd866aSAl Viro unsigned int flags) 150144396f4bSJosef Bacik { 150244396f4bSJosef Bacik struct dentry *old; 150344396f4bSJosef Bacik 150444396f4bSJosef Bacik /* Don't create child dentry for a dead directory. */ 1505bad61189SMiklos Szeredi if (unlikely(IS_DEADDIR(dir))) { 1506e188dc02SMiklos Szeredi dput(dentry); 150744396f4bSJosef Bacik return ERR_PTR(-ENOENT); 1508e188dc02SMiklos Szeredi } 150944396f4bSJosef Bacik 151072bd866aSAl Viro old = dir->i_op->lookup(dir, dentry, flags); 151144396f4bSJosef Bacik if (unlikely(old)) { 151244396f4bSJosef Bacik dput(dentry); 151344396f4bSJosef Bacik dentry = old; 151444396f4bSJosef Bacik } 151544396f4bSJosef Bacik return dentry; 151644396f4bSJosef Bacik } 151744396f4bSJosef Bacik 1518e3c13928SAl Viro static struct dentry *__lookup_hash(const struct qstr *name, 151972bd866aSAl Viro struct dentry *base, unsigned int flags) 1520a3255546SAl Viro { 15216c51e513SAl Viro struct dentry *dentry = lookup_dcache(name, base, flags); 1522a3255546SAl Viro 15236c51e513SAl Viro if (dentry) 1524a3255546SAl Viro return dentry; 1525bad61189SMiklos Szeredi 15266c51e513SAl Viro dentry = d_alloc(base, name); 15276c51e513SAl Viro if (unlikely(!dentry)) 15286c51e513SAl Viro return ERR_PTR(-ENOMEM); 15296c51e513SAl Viro 153072bd866aSAl Viro return lookup_real(base->d_inode, dentry, flags); 1531a3255546SAl Viro } 1532a3255546SAl Viro 1533e97cdc87SAl Viro static int lookup_fast(struct nameidata *nd, 1534254cf582SAl Viro struct path *path, struct inode **inode, 1535254cf582SAl Viro unsigned *seqp) 15361da177e4SLinus Torvalds { 15374ac91378SJan Blunck struct vfsmount *mnt = nd->path.mnt; 153831e6b01fSNick Piggin struct dentry *dentry, *parent = nd->path.dentry; 15395a18fff2SAl Viro int status = 1; 15409875cf80SDavid Howells int err; 15419875cf80SDavid Howells 15423cac260aSAl Viro /* 1543b04f784eSNick Piggin * Rename seqlock is not required here because in the off chance 15445d0f49c1SAl Viro * of a false negative due to a concurrent rename, the caller is 15455d0f49c1SAl Viro * going to fall back to non-racy lookup. 1546b04f784eSNick Piggin */ 154731e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) { 154831e6b01fSNick Piggin unsigned seq; 1549766c4cbfSAl Viro bool negative; 1550da53be12SLinus Torvalds dentry = __d_lookup_rcu(parent, &nd->last, &seq); 15515d0f49c1SAl Viro if (unlikely(!dentry)) { 15525d0f49c1SAl Viro if (unlazy_walk(nd, NULL, 0)) 15535d0f49c1SAl Viro return -ECHILD; 1554e9742b53SAl Viro return 0; 15555d0f49c1SAl Viro } 15565a18fff2SAl Viro 155712f8ad4bSLinus Torvalds /* 155812f8ad4bSLinus Torvalds * This sequence count validates that the inode matches 155912f8ad4bSLinus Torvalds * the dentry name information from lookup. 156012f8ad4bSLinus Torvalds */ 156163afdfc7SDavid Howells *inode = d_backing_inode(dentry); 1562766c4cbfSAl Viro negative = d_is_negative(dentry); 15635d0f49c1SAl Viro if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) 156412f8ad4bSLinus Torvalds return -ECHILD; 156512f8ad4bSLinus Torvalds 156612f8ad4bSLinus Torvalds /* 156712f8ad4bSLinus Torvalds * This sequence count validates that the parent had no 156812f8ad4bSLinus Torvalds * changes while we did the lookup of the dentry above. 156912f8ad4bSLinus Torvalds * 157012f8ad4bSLinus Torvalds * The memory barrier in read_seqcount_begin of child is 157112f8ad4bSLinus Torvalds * enough, we can use __read_seqcount_retry here. 157212f8ad4bSLinus Torvalds */ 15735d0f49c1SAl Viro if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq))) 157431e6b01fSNick Piggin return -ECHILD; 15755a18fff2SAl Viro 1576254cf582SAl Viro *seqp = seq; 15775d0f49c1SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) 15784ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 15795a18fff2SAl Viro if (unlikely(status <= 0)) { 15805d0f49c1SAl Viro if (unlazy_walk(nd, dentry, seq)) 15815d0f49c1SAl Viro return -ECHILD; 15825d0f49c1SAl Viro if (status == -ECHILD) 15835d0f49c1SAl Viro status = d_revalidate(dentry, nd->flags); 15845d0f49c1SAl Viro } else { 1585daf3761cSTrond Myklebust /* 1586daf3761cSTrond Myklebust * Note: do negative dentry check after revalidation in 1587daf3761cSTrond Myklebust * case that drops it. 1588daf3761cSTrond Myklebust */ 15895d0f49c1SAl Viro if (unlikely(negative)) 1590daf3761cSTrond Myklebust return -ENOENT; 159131e6b01fSNick Piggin path->mnt = mnt; 159231e6b01fSNick Piggin path->dentry = dentry; 1593254cf582SAl Viro if (likely(__follow_mount_rcu(nd, path, inode, seqp))) 1594e9742b53SAl Viro return 1; 1595254cf582SAl Viro if (unlazy_walk(nd, dentry, seq)) 15965a18fff2SAl Viro return -ECHILD; 15975d0f49c1SAl Viro } 15985a18fff2SAl Viro } else { 1599e97cdc87SAl Viro dentry = __d_lookup(parent, &nd->last); 160081e6f520SAl Viro if (unlikely(!dentry)) 1601e9742b53SAl Viro return 0; 16025d0f49c1SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) 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))) { 1642949a852eSAl Viro if ((dentry->d_flags & DCACHE_OP_REVALIDATE) && 1643949a852eSAl Viro !(flags & LOOKUP_NO_REVAL)) { 1644949a852eSAl Viro int error = d_revalidate(dentry, flags); 1645949a852eSAl Viro if (unlikely(error <= 0)) { 164694bdd655SAl Viro if (!error) { 1647949a852eSAl Viro d_invalidate(dentry); 1648949a852eSAl Viro dput(dentry); 164994bdd655SAl Viro goto again; 165094bdd655SAl Viro } 165194bdd655SAl Viro dput(dentry); 1652949a852eSAl Viro dentry = ERR_PTR(error); 1653949a852eSAl Viro } 1654949a852eSAl Viro } 165594bdd655SAl Viro } else { 16561936386eSAl Viro old = inode->i_op->lookup(inode, dentry, flags); 165785c7f810SAl Viro d_lookup_done(dentry); 16581936386eSAl Viro if (unlikely(old)) { 16591936386eSAl Viro dput(dentry); 16601936386eSAl Viro dentry = old; 1661949a852eSAl Viro } 1662949a852eSAl Viro } 166394bdd655SAl Viro out: 16649902af79SAl Viro inode_unlock_shared(inode); 1665e3c13928SAl Viro return dentry; 16661da177e4SLinus Torvalds } 16671da177e4SLinus Torvalds 166852094c8aSAl Viro static inline int may_lookup(struct nameidata *nd) 166952094c8aSAl Viro { 167052094c8aSAl Viro if (nd->flags & LOOKUP_RCU) { 16714ad5abb3SAl Viro int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); 167252094c8aSAl Viro if (err != -ECHILD) 167352094c8aSAl Viro return err; 16746e9918b7SAl Viro if (unlazy_walk(nd, NULL, 0)) 167552094c8aSAl Viro return -ECHILD; 167652094c8aSAl Viro } 16774ad5abb3SAl Viro return inode_permission(nd->inode, MAY_EXEC); 167852094c8aSAl Viro } 167952094c8aSAl Viro 16809856fa1bSAl Viro static inline int handle_dots(struct nameidata *nd, int type) 16819856fa1bSAl Viro { 16829856fa1bSAl Viro if (type == LAST_DOTDOT) { 16839e6697e2SAl Viro if (!nd->root.mnt) 16849e6697e2SAl Viro set_root(nd); 16859856fa1bSAl Viro if (nd->flags & LOOKUP_RCU) { 168670291aecSAl Viro return follow_dotdot_rcu(nd); 16879856fa1bSAl Viro } else 1688397d425dSEric W. Biederman return follow_dotdot(nd); 16899856fa1bSAl Viro } 16909856fa1bSAl Viro return 0; 16919856fa1bSAl Viro } 16929856fa1bSAl Viro 1693181548c0SAl Viro static int pick_link(struct nameidata *nd, struct path *link, 1694181548c0SAl Viro struct inode *inode, unsigned seq) 1695d63ff28fSAl Viro { 1696626de996SAl Viro int error; 16971cf2665bSAl Viro struct saved *last; 1698756daf26SNeilBrown if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) { 1699626de996SAl Viro path_to_nameidata(link, nd); 1700626de996SAl Viro return -ELOOP; 1701626de996SAl Viro } 1702bc40aee0SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1703cd179f44SAl Viro if (link->mnt == nd->path.mnt) 1704cd179f44SAl Viro mntget(link->mnt); 17057973387aSAl Viro } 1706626de996SAl Viro error = nd_alloc_stack(nd); 1707626de996SAl Viro if (unlikely(error)) { 1708bc40aee0SAl Viro if (error == -ECHILD) { 1709bc40aee0SAl Viro if (unlikely(unlazy_link(nd, link, seq))) 1710bc40aee0SAl Viro return -ECHILD; 1711bc40aee0SAl Viro error = nd_alloc_stack(nd); 1712bc40aee0SAl Viro } 1713bc40aee0SAl Viro if (error) { 1714cd179f44SAl Viro path_put(link); 1715626de996SAl Viro return error; 1716626de996SAl Viro } 1717bc40aee0SAl Viro } 1718626de996SAl Viro 1719ab104923SAl Viro last = nd->stack + nd->depth++; 17201cf2665bSAl Viro last->link = *link; 1721fceef393SAl Viro clear_delayed_call(&last->done); 1722fceef393SAl Viro nd->link_inode = inode; 17230450b2d1SAl Viro last->seq = seq; 1724d63ff28fSAl Viro return 1; 1725d63ff28fSAl Viro } 1726d63ff28fSAl Viro 17278f64fb1cSAl Viro enum {WALK_FOLLOW = 1, WALK_MORE = 2}; 172831d66bcdSAl Viro 17293ddcd056SLinus Torvalds /* 17303ddcd056SLinus Torvalds * Do we need to follow links? We _really_ want to be able 17313ddcd056SLinus Torvalds * to do this check without having to look at inode->i_op, 17323ddcd056SLinus Torvalds * so we keep a cache of "no, this doesn't need follow_link" 17333ddcd056SLinus Torvalds * for the common case. 17343ddcd056SLinus Torvalds */ 17358f64fb1cSAl Viro static inline int step_into(struct nameidata *nd, struct path *path, 17368f64fb1cSAl Viro int flags, struct inode *inode, unsigned seq) 17373ddcd056SLinus Torvalds { 173831d66bcdSAl Viro if (!(flags & WALK_MORE) && nd->depth) 173931d66bcdSAl Viro put_link(nd); 17408f64fb1cSAl Viro if (likely(!d_is_symlink(path->dentry)) || 17418f64fb1cSAl Viro !(flags & WALK_FOLLOW || nd->flags & LOOKUP_FOLLOW)) { 17428f64fb1cSAl Viro /* not a symlink or should not follow */ 17438f64fb1cSAl Viro path_to_nameidata(path, nd); 17448f64fb1cSAl Viro nd->inode = inode; 17458f64fb1cSAl Viro nd->seq = seq; 1746d63ff28fSAl Viro return 0; 17478f64fb1cSAl Viro } 1748a7f77542SAl Viro /* make sure that d_is_symlink above matches inode */ 1749a7f77542SAl Viro if (nd->flags & LOOKUP_RCU) { 17508f64fb1cSAl Viro if (read_seqcount_retry(&path->dentry->d_seq, seq)) 1751a7f77542SAl Viro return -ECHILD; 1752a7f77542SAl Viro } 17538f64fb1cSAl Viro return pick_link(nd, path, inode, seq); 17543ddcd056SLinus Torvalds } 17553ddcd056SLinus Torvalds 17564693a547SAl Viro static int walk_component(struct nameidata *nd, int flags) 1757ce57dfc1SAl Viro { 1758caa85634SAl Viro struct path path; 1759ce57dfc1SAl Viro struct inode *inode; 1760254cf582SAl Viro unsigned seq; 1761ce57dfc1SAl Viro int err; 1762ce57dfc1SAl Viro /* 1763ce57dfc1SAl Viro * "." and ".." are special - ".." especially so because it has 1764ce57dfc1SAl Viro * to be able to know about the current root directory and 1765ce57dfc1SAl Viro * parent relationships. 1766ce57dfc1SAl Viro */ 17674693a547SAl Viro if (unlikely(nd->last_type != LAST_NORM)) { 17684693a547SAl Viro err = handle_dots(nd, nd->last_type); 17691c4ff1a8SAl Viro if (!(flags & WALK_MORE) && nd->depth) 17704693a547SAl Viro put_link(nd); 17714693a547SAl Viro return err; 17724693a547SAl Viro } 1773254cf582SAl Viro err = lookup_fast(nd, &path, &inode, &seq); 1774e9742b53SAl Viro if (unlikely(err <= 0)) { 1775697f514dSMiklos Szeredi if (err < 0) 1776f0a9ba70SAl Viro return err; 1777e3c13928SAl Viro path.dentry = lookup_slow(&nd->last, nd->path.dentry, 1778e3c13928SAl Viro nd->flags); 1779e3c13928SAl Viro if (IS_ERR(path.dentry)) 1780e3c13928SAl Viro return PTR_ERR(path.dentry); 17817500c38aSAl Viro 1782e3c13928SAl Viro path.mnt = nd->path.mnt; 1783e3c13928SAl Viro err = follow_managed(&path, nd); 1784e3c13928SAl Viro if (unlikely(err < 0)) 1785f0a9ba70SAl Viro return err; 1786697f514dSMiklos Szeredi 17877500c38aSAl Viro if (unlikely(d_is_negative(path.dentry))) { 17887500c38aSAl Viro path_to_nameidata(&path, nd); 17897500c38aSAl Viro return -ENOENT; 17907500c38aSAl Viro } 17917500c38aSAl Viro 1792254cf582SAl Viro seq = 0; /* we are already out of RCU mode */ 1793d4565649SAl Viro inode = d_backing_inode(path.dentry); 1794766c4cbfSAl Viro } 1795697f514dSMiklos Szeredi 17968f64fb1cSAl Viro return step_into(nd, &path, flags, inode, seq); 1797ce57dfc1SAl Viro } 1798ce57dfc1SAl Viro 17991da177e4SLinus Torvalds /* 1800bfcfaa77SLinus Torvalds * We can do the critical dentry name comparison and hashing 1801bfcfaa77SLinus Torvalds * operations one word at a time, but we are limited to: 1802bfcfaa77SLinus Torvalds * 1803bfcfaa77SLinus Torvalds * - Architectures with fast unaligned word accesses. We could 1804bfcfaa77SLinus Torvalds * do a "get_unaligned()" if this helps and is sufficiently 1805bfcfaa77SLinus Torvalds * fast. 1806bfcfaa77SLinus Torvalds * 1807bfcfaa77SLinus Torvalds * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we 1808bfcfaa77SLinus Torvalds * do not trap on the (extremely unlikely) case of a page 1809bfcfaa77SLinus Torvalds * crossing operation. 1810bfcfaa77SLinus Torvalds * 1811bfcfaa77SLinus Torvalds * - Furthermore, we need an efficient 64-bit compile for the 1812bfcfaa77SLinus Torvalds * 64-bit case in order to generate the "number of bytes in 1813bfcfaa77SLinus Torvalds * the final mask". Again, that could be replaced with a 1814bfcfaa77SLinus Torvalds * efficient population count instruction or similar. 1815bfcfaa77SLinus Torvalds */ 1816bfcfaa77SLinus Torvalds #ifdef CONFIG_DCACHE_WORD_ACCESS 1817bfcfaa77SLinus Torvalds 1818f68e556eSLinus Torvalds #include <asm/word-at-a-time.h> 1819bfcfaa77SLinus Torvalds 1820468a9428SGeorge Spelvin #ifdef HASH_MIX 1821bfcfaa77SLinus Torvalds 1822468a9428SGeorge Spelvin /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */ 1823468a9428SGeorge Spelvin 1824468a9428SGeorge Spelvin #elif defined(CONFIG_64BIT) 18252a18da7aSGeorge Spelvin /* 18262a18da7aSGeorge Spelvin * Register pressure in the mixing function is an issue, particularly 18272a18da7aSGeorge Spelvin * on 32-bit x86, but almost any function requires one state value and 18282a18da7aSGeorge Spelvin * one temporary. Instead, use a function designed for two state values 18292a18da7aSGeorge Spelvin * and no temporaries. 18302a18da7aSGeorge Spelvin * 18312a18da7aSGeorge Spelvin * This function cannot create a collision in only two iterations, so 18322a18da7aSGeorge Spelvin * we have two iterations to achieve avalanche. In those two iterations, 18332a18da7aSGeorge Spelvin * we have six layers of mixing, which is enough to spread one bit's 18342a18da7aSGeorge Spelvin * influence out to 2^6 = 64 state bits. 18352a18da7aSGeorge Spelvin * 18362a18da7aSGeorge Spelvin * Rotate constants are scored by considering either 64 one-bit input 18372a18da7aSGeorge Spelvin * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the 18382a18da7aSGeorge Spelvin * probability of that delta causing a change to each of the 128 output 18392a18da7aSGeorge Spelvin * bits, using a sample of random initial states. 18402a18da7aSGeorge Spelvin * 18412a18da7aSGeorge Spelvin * The Shannon entropy of the computed probabilities is then summed 18422a18da7aSGeorge Spelvin * to produce a score. Ideally, any input change has a 50% chance of 18432a18da7aSGeorge Spelvin * toggling any given output bit. 18442a18da7aSGeorge Spelvin * 18452a18da7aSGeorge Spelvin * Mixing scores (in bits) for (12,45): 18462a18da7aSGeorge Spelvin * Input delta: 1-bit 2-bit 18472a18da7aSGeorge Spelvin * 1 round: 713.3 42542.6 18482a18da7aSGeorge Spelvin * 2 rounds: 2753.7 140389.8 18492a18da7aSGeorge Spelvin * 3 rounds: 5954.1 233458.2 18502a18da7aSGeorge Spelvin * 4 rounds: 7862.6 256672.2 18512a18da7aSGeorge Spelvin * Perfect: 8192 258048 18522a18da7aSGeorge Spelvin * (64*128) (64*63/2 * 128) 18532a18da7aSGeorge Spelvin */ 18542a18da7aSGeorge Spelvin #define HASH_MIX(x, y, a) \ 18552a18da7aSGeorge Spelvin ( x ^= (a), \ 18562a18da7aSGeorge Spelvin y ^= x, x = rol64(x,12),\ 18572a18da7aSGeorge Spelvin x += y, y = rol64(y,45),\ 18582a18da7aSGeorge Spelvin y *= 9 ) 1859bfcfaa77SLinus Torvalds 18600fed3ac8SGeorge Spelvin /* 18612a18da7aSGeorge Spelvin * Fold two longs into one 32-bit hash value. This must be fast, but 18622a18da7aSGeorge Spelvin * latency isn't quite as critical, as there is a fair bit of additional 18632a18da7aSGeorge Spelvin * work done before the hash value is used. 18640fed3ac8SGeorge Spelvin */ 18652a18da7aSGeorge Spelvin static inline unsigned int fold_hash(unsigned long x, unsigned long y) 18660fed3ac8SGeorge Spelvin { 18672a18da7aSGeorge Spelvin y ^= x * GOLDEN_RATIO_64; 18682a18da7aSGeorge Spelvin y *= GOLDEN_RATIO_64; 18692a18da7aSGeorge Spelvin return y >> 32; 18700fed3ac8SGeorge Spelvin } 18710fed3ac8SGeorge Spelvin 1872bfcfaa77SLinus Torvalds #else /* 32-bit case */ 1873bfcfaa77SLinus Torvalds 18742a18da7aSGeorge Spelvin /* 18752a18da7aSGeorge Spelvin * Mixing scores (in bits) for (7,20): 18762a18da7aSGeorge Spelvin * Input delta: 1-bit 2-bit 18772a18da7aSGeorge Spelvin * 1 round: 330.3 9201.6 18782a18da7aSGeorge Spelvin * 2 rounds: 1246.4 25475.4 18792a18da7aSGeorge Spelvin * 3 rounds: 1907.1 31295.1 18802a18da7aSGeorge Spelvin * 4 rounds: 2042.3 31718.6 18812a18da7aSGeorge Spelvin * Perfect: 2048 31744 18822a18da7aSGeorge Spelvin * (32*64) (32*31/2 * 64) 18832a18da7aSGeorge Spelvin */ 18842a18da7aSGeorge Spelvin #define HASH_MIX(x, y, a) \ 18852a18da7aSGeorge Spelvin ( x ^= (a), \ 18862a18da7aSGeorge Spelvin y ^= x, x = rol32(x, 7),\ 18872a18da7aSGeorge Spelvin x += y, y = rol32(y,20),\ 18882a18da7aSGeorge Spelvin y *= 9 ) 1889bfcfaa77SLinus Torvalds 18902a18da7aSGeorge Spelvin static inline unsigned int fold_hash(unsigned long x, unsigned long y) 18910fed3ac8SGeorge Spelvin { 18922a18da7aSGeorge Spelvin /* Use arch-optimized multiply if one exists */ 18932a18da7aSGeorge Spelvin return __hash_32(y ^ __hash_32(x)); 18940fed3ac8SGeorge Spelvin } 18950fed3ac8SGeorge Spelvin 1896bfcfaa77SLinus Torvalds #endif 1897bfcfaa77SLinus Torvalds 18982a18da7aSGeorge Spelvin /* 18992a18da7aSGeorge Spelvin * Return the hash of a string of known length. This is carfully 19002a18da7aSGeorge Spelvin * designed to match hash_name(), which is the more critical function. 19012a18da7aSGeorge Spelvin * In particular, we must end by hashing a final word containing 0..7 19022a18da7aSGeorge Spelvin * payload bytes, to match the way that hash_name() iterates until it 19032a18da7aSGeorge Spelvin * finds the delimiter after the name. 19042a18da7aSGeorge Spelvin */ 19058387ff25SLinus Torvalds unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) 1906bfcfaa77SLinus Torvalds { 19078387ff25SLinus Torvalds unsigned long a, x = 0, y = (unsigned long)salt; 1908bfcfaa77SLinus Torvalds 1909bfcfaa77SLinus Torvalds for (;;) { 1910fcfd2fbfSGeorge Spelvin if (!len) 1911fcfd2fbfSGeorge Spelvin goto done; 1912e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name); 1913bfcfaa77SLinus Torvalds if (len < sizeof(unsigned long)) 1914bfcfaa77SLinus Torvalds break; 19152a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 1916bfcfaa77SLinus Torvalds name += sizeof(unsigned long); 1917bfcfaa77SLinus Torvalds len -= sizeof(unsigned long); 1918bfcfaa77SLinus Torvalds } 19192a18da7aSGeorge Spelvin x ^= a & bytemask_from_count(len); 1920bfcfaa77SLinus Torvalds done: 19212a18da7aSGeorge Spelvin return fold_hash(x, y); 1922bfcfaa77SLinus Torvalds } 1923bfcfaa77SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 1924bfcfaa77SLinus Torvalds 1925fcfd2fbfSGeorge Spelvin /* Return the "hash_len" (hash and length) of a null-terminated string */ 19268387ff25SLinus Torvalds u64 hashlen_string(const void *salt, const char *name) 1927fcfd2fbfSGeorge Spelvin { 19288387ff25SLinus Torvalds unsigned long a = 0, x = 0, y = (unsigned long)salt; 19298387ff25SLinus Torvalds unsigned long adata, mask, len; 1930fcfd2fbfSGeorge Spelvin const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 1931fcfd2fbfSGeorge Spelvin 19328387ff25SLinus Torvalds len = 0; 19338387ff25SLinus Torvalds goto inside; 19348387ff25SLinus Torvalds 1935fcfd2fbfSGeorge Spelvin do { 19362a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 1937fcfd2fbfSGeorge Spelvin len += sizeof(unsigned long); 19388387ff25SLinus Torvalds inside: 1939fcfd2fbfSGeorge Spelvin a = load_unaligned_zeropad(name+len); 1940fcfd2fbfSGeorge Spelvin } while (!has_zero(a, &adata, &constants)); 1941fcfd2fbfSGeorge Spelvin 1942fcfd2fbfSGeorge Spelvin adata = prep_zero_mask(a, adata, &constants); 1943fcfd2fbfSGeorge Spelvin mask = create_zero_mask(adata); 19442a18da7aSGeorge Spelvin x ^= a & zero_bytemask(mask); 1945fcfd2fbfSGeorge Spelvin 19462a18da7aSGeorge Spelvin return hashlen_create(fold_hash(x, y), len + find_zero(mask)); 1947fcfd2fbfSGeorge Spelvin } 1948fcfd2fbfSGeorge Spelvin EXPORT_SYMBOL(hashlen_string); 1949fcfd2fbfSGeorge Spelvin 1950bfcfaa77SLinus Torvalds /* 1951bfcfaa77SLinus Torvalds * Calculate the length and hash of the path component, and 1952d6bb3e90SLinus Torvalds * return the "hash_len" as the result. 1953bfcfaa77SLinus Torvalds */ 19548387ff25SLinus Torvalds static inline u64 hash_name(const void *salt, const char *name) 1955bfcfaa77SLinus Torvalds { 19568387ff25SLinus Torvalds unsigned long a = 0, b, x = 0, y = (unsigned long)salt; 19578387ff25SLinus Torvalds unsigned long adata, bdata, mask, len; 195836126f8fSLinus Torvalds const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 1959bfcfaa77SLinus Torvalds 19608387ff25SLinus Torvalds len = 0; 19618387ff25SLinus Torvalds goto inside; 19628387ff25SLinus Torvalds 1963bfcfaa77SLinus Torvalds do { 19642a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 1965bfcfaa77SLinus Torvalds len += sizeof(unsigned long); 19668387ff25SLinus Torvalds inside: 1967e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name+len); 196836126f8fSLinus Torvalds b = a ^ REPEAT_BYTE('/'); 196936126f8fSLinus Torvalds } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); 1970bfcfaa77SLinus Torvalds 197136126f8fSLinus Torvalds adata = prep_zero_mask(a, adata, &constants); 197236126f8fSLinus Torvalds bdata = prep_zero_mask(b, bdata, &constants); 197336126f8fSLinus Torvalds mask = create_zero_mask(adata | bdata); 19742a18da7aSGeorge Spelvin x ^= a & zero_bytemask(mask); 197536126f8fSLinus Torvalds 19762a18da7aSGeorge Spelvin return hashlen_create(fold_hash(x, y), len + find_zero(mask)); 1977bfcfaa77SLinus Torvalds } 1978bfcfaa77SLinus Torvalds 19792a18da7aSGeorge Spelvin #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */ 1980bfcfaa77SLinus Torvalds 1981fcfd2fbfSGeorge Spelvin /* Return the hash of a string of known length */ 19828387ff25SLinus Torvalds unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) 19830145acc2SLinus Torvalds { 19848387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 19850145acc2SLinus Torvalds while (len--) 1986fcfd2fbfSGeorge Spelvin hash = partial_name_hash((unsigned char)*name++, hash); 19870145acc2SLinus Torvalds return end_name_hash(hash); 19880145acc2SLinus Torvalds } 1989ae942ae7SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 19900145acc2SLinus Torvalds 1991fcfd2fbfSGeorge Spelvin /* Return the "hash_len" (hash and length) of a null-terminated string */ 19928387ff25SLinus Torvalds u64 hashlen_string(const void *salt, const char *name) 1993fcfd2fbfSGeorge Spelvin { 19948387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 1995fcfd2fbfSGeorge Spelvin unsigned long len = 0, c; 1996fcfd2fbfSGeorge Spelvin 1997fcfd2fbfSGeorge Spelvin c = (unsigned char)*name; 1998e0ab7af9SGeorge Spelvin while (c) { 1999fcfd2fbfSGeorge Spelvin len++; 2000fcfd2fbfSGeorge Spelvin hash = partial_name_hash(c, hash); 2001fcfd2fbfSGeorge Spelvin c = (unsigned char)name[len]; 2002e0ab7af9SGeorge Spelvin } 2003fcfd2fbfSGeorge Spelvin return hashlen_create(end_name_hash(hash), len); 2004fcfd2fbfSGeorge Spelvin } 2005f2a031b6SGeorge Spelvin EXPORT_SYMBOL(hashlen_string); 2006fcfd2fbfSGeorge Spelvin 20073ddcd056SLinus Torvalds /* 2008200e9ef7SLinus Torvalds * We know there's a real path component here of at least 2009200e9ef7SLinus Torvalds * one character. 2010200e9ef7SLinus Torvalds */ 20118387ff25SLinus Torvalds static inline u64 hash_name(const void *salt, const char *name) 2012200e9ef7SLinus Torvalds { 20138387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 2014200e9ef7SLinus Torvalds unsigned long len = 0, c; 2015200e9ef7SLinus Torvalds 2016200e9ef7SLinus Torvalds c = (unsigned char)*name; 2017200e9ef7SLinus Torvalds do { 2018200e9ef7SLinus Torvalds len++; 2019200e9ef7SLinus Torvalds hash = partial_name_hash(c, hash); 2020200e9ef7SLinus Torvalds c = (unsigned char)name[len]; 2021200e9ef7SLinus Torvalds } while (c && c != '/'); 2022d6bb3e90SLinus Torvalds return hashlen_create(end_name_hash(hash), len); 2023200e9ef7SLinus Torvalds } 2024200e9ef7SLinus Torvalds 2025bfcfaa77SLinus Torvalds #endif 2026bfcfaa77SLinus Torvalds 2027200e9ef7SLinus Torvalds /* 20281da177e4SLinus Torvalds * Name resolution. 2029ea3834d9SPrasanna Meda * This is the basic name resolution function, turning a pathname into 2030ea3834d9SPrasanna Meda * the final dentry. We expect 'base' to be positive and a directory. 20311da177e4SLinus Torvalds * 2032ea3834d9SPrasanna Meda * Returns 0 and nd will have valid dentry and mnt on success. 2033ea3834d9SPrasanna Meda * Returns error and drops reference to input namei data on failure. 20341da177e4SLinus Torvalds */ 20356de88d72SAl Viro static int link_path_walk(const char *name, struct nameidata *nd) 20361da177e4SLinus Torvalds { 20371da177e4SLinus Torvalds int err; 20381da177e4SLinus Torvalds 20391da177e4SLinus Torvalds while (*name=='/') 20401da177e4SLinus Torvalds name++; 20411da177e4SLinus Torvalds if (!*name) 20429e18f10aSAl Viro return 0; 20431da177e4SLinus Torvalds 20441da177e4SLinus Torvalds /* At this point we know we have a real path component. */ 20451da177e4SLinus Torvalds for(;;) { 2046d6bb3e90SLinus Torvalds u64 hash_len; 2047fe479a58SAl Viro int type; 20481da177e4SLinus Torvalds 204952094c8aSAl Viro err = may_lookup(nd); 20501da177e4SLinus Torvalds if (err) 20513595e234SAl Viro return err; 20521da177e4SLinus Torvalds 20538387ff25SLinus Torvalds hash_len = hash_name(nd->path.dentry, name); 20541da177e4SLinus Torvalds 2055fe479a58SAl Viro type = LAST_NORM; 2056d6bb3e90SLinus Torvalds if (name[0] == '.') switch (hashlen_len(hash_len)) { 2057fe479a58SAl Viro case 2: 2058200e9ef7SLinus Torvalds if (name[1] == '.') { 2059fe479a58SAl Viro type = LAST_DOTDOT; 206016c2cd71SAl Viro nd->flags |= LOOKUP_JUMPED; 206116c2cd71SAl Viro } 2062fe479a58SAl Viro break; 2063fe479a58SAl Viro case 1: 2064fe479a58SAl Viro type = LAST_DOT; 2065fe479a58SAl Viro } 20665a202bcdSAl Viro if (likely(type == LAST_NORM)) { 20675a202bcdSAl Viro struct dentry *parent = nd->path.dentry; 206816c2cd71SAl Viro nd->flags &= ~LOOKUP_JUMPED; 20695a202bcdSAl Viro if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { 2070a060dc50SJames Hogan struct qstr this = { { .hash_len = hash_len }, .name = name }; 2071da53be12SLinus Torvalds err = parent->d_op->d_hash(parent, &this); 20725a202bcdSAl Viro if (err < 0) 20733595e234SAl Viro return err; 2074d6bb3e90SLinus Torvalds hash_len = this.hash_len; 2075d6bb3e90SLinus Torvalds name = this.name; 20765a202bcdSAl Viro } 20775a202bcdSAl Viro } 2078fe479a58SAl Viro 2079d6bb3e90SLinus Torvalds nd->last.hash_len = hash_len; 2080d6bb3e90SLinus Torvalds nd->last.name = name; 20815f4a6a69SAl Viro nd->last_type = type; 20825f4a6a69SAl Viro 2083d6bb3e90SLinus Torvalds name += hashlen_len(hash_len); 2084d6bb3e90SLinus Torvalds if (!*name) 2085bdf6cbf1SAl Viro goto OK; 2086200e9ef7SLinus Torvalds /* 2087200e9ef7SLinus Torvalds * If it wasn't NUL, we know it was '/'. Skip that 2088200e9ef7SLinus Torvalds * slash, and continue until no more slashes. 2089200e9ef7SLinus Torvalds */ 2090200e9ef7SLinus Torvalds do { 2091d6bb3e90SLinus Torvalds name++; 2092d6bb3e90SLinus Torvalds } while (unlikely(*name == '/')); 20938620c238SAl Viro if (unlikely(!*name)) { 20948620c238SAl Viro OK: 2095368ee9baSAl Viro /* pathname body, done */ 20968620c238SAl Viro if (!nd->depth) 20978620c238SAl Viro return 0; 20988620c238SAl Viro name = nd->stack[nd->depth - 1].name; 2099368ee9baSAl Viro /* trailing symlink, done */ 21008620c238SAl Viro if (!name) 21018620c238SAl Viro return 0; 21028620c238SAl Viro /* last component of nested symlink */ 21038f64fb1cSAl Viro err = walk_component(nd, WALK_FOLLOW); 21041c4ff1a8SAl Viro } else { 21051c4ff1a8SAl Viro /* not the last component */ 21068f64fb1cSAl Viro err = walk_component(nd, WALK_FOLLOW | WALK_MORE); 21078620c238SAl Viro } 2108ce57dfc1SAl Viro if (err < 0) 21093595e234SAl Viro return err; 2110fe479a58SAl Viro 2111ce57dfc1SAl Viro if (err) { 2112626de996SAl Viro const char *s = get_link(nd); 21135a460275SAl Viro 2114a1c83681SViresh Kumar if (IS_ERR(s)) 21153595e234SAl Viro return PTR_ERR(s); 2116172a39a0SAl Viro err = 0; 211712b09578SAl Viro if (unlikely(!s)) { 211812b09578SAl Viro /* jumped */ 2119b9ff4429SAl Viro put_link(nd); 212012b09578SAl Viro } else { 2121071bf501SAl Viro nd->stack[nd->depth - 1].name = name; 212232cd7468SAl Viro name = s; 21239e18f10aSAl Viro continue; 212448c8b0c5SAl Viro } 212531e6b01fSNick Piggin } 212697242f99SAl Viro if (unlikely(!d_can_lookup(nd->path.dentry))) { 212797242f99SAl Viro if (nd->flags & LOOKUP_RCU) { 212897242f99SAl Viro if (unlazy_walk(nd, NULL, 0)) 212997242f99SAl Viro return -ECHILD; 213097242f99SAl Viro } 21313595e234SAl Viro return -ENOTDIR; 21325f4a6a69SAl Viro } 2133ce57dfc1SAl Viro } 213497242f99SAl Viro } 21351da177e4SLinus Torvalds 2136c8a53ee5SAl Viro static const char *path_init(struct nameidata *nd, unsigned flags) 213731e6b01fSNick Piggin { 213831e6b01fSNick Piggin int retval = 0; 2139c8a53ee5SAl Viro const char *s = nd->name->name; 214031e6b01fSNick Piggin 214131e6b01fSNick Piggin nd->last_type = LAST_ROOT; /* if there are only slashes... */ 2142980f3ea2SAl Viro nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT; 214331e6b01fSNick Piggin nd->depth = 0; 21445b6ca027SAl Viro if (flags & LOOKUP_ROOT) { 2145b18825a7SDavid Howells struct dentry *root = nd->root.dentry; 2146b18825a7SDavid Howells struct inode *inode = root->d_inode; 2147fd2f7cb5SAl Viro if (*s) { 214844b1d530SMiklos Szeredi if (!d_can_lookup(root)) 2149368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 21505b6ca027SAl Viro retval = inode_permission(inode, MAY_EXEC); 21515b6ca027SAl Viro if (retval) 2152368ee9baSAl Viro return ERR_PTR(retval); 215373d049a4SAl Viro } 21545b6ca027SAl Viro nd->path = nd->root; 21555b6ca027SAl Viro nd->inode = inode; 21565b6ca027SAl Viro if (flags & LOOKUP_RCU) { 21578b61e74fSAl Viro rcu_read_lock(); 21585b6ca027SAl Viro nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 21598f47a016SAl Viro nd->root_seq = nd->seq; 216048a066e7SAl Viro nd->m_seq = read_seqbegin(&mount_lock); 21615b6ca027SAl Viro } else { 21625b6ca027SAl Viro path_get(&nd->path); 21635b6ca027SAl Viro } 2164368ee9baSAl Viro return s; 21655b6ca027SAl Viro } 21665b6ca027SAl Viro 216731e6b01fSNick Piggin nd->root.mnt = NULL; 2168248fb5b9SAl Viro nd->path.mnt = NULL; 2169248fb5b9SAl Viro nd->path.dentry = NULL; 217031e6b01fSNick Piggin 217148a066e7SAl Viro nd->m_seq = read_seqbegin(&mount_lock); 2172fd2f7cb5SAl Viro if (*s == '/') { 21739e6697e2SAl Viro if (flags & LOOKUP_RCU) 21748b61e74fSAl Viro rcu_read_lock(); 2175e41f7d4eSAl Viro set_root(nd); 2176248fb5b9SAl Viro if (likely(!nd_jump_root(nd))) 2177ef55d917SAl Viro return s; 2178ef55d917SAl Viro nd->root.mnt = NULL; 2179ef55d917SAl Viro rcu_read_unlock(); 2180ef55d917SAl Viro return ERR_PTR(-ECHILD); 2181c8a53ee5SAl Viro } else if (nd->dfd == AT_FDCWD) { 2182e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 218331e6b01fSNick Piggin struct fs_struct *fs = current->fs; 2184c28cc364SNick Piggin unsigned seq; 218531e6b01fSNick Piggin 21868b61e74fSAl Viro rcu_read_lock(); 218731e6b01fSNick Piggin 2188c28cc364SNick Piggin do { 2189c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 219031e6b01fSNick Piggin nd->path = fs->pwd; 2191ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2192c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 2193c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 2194e41f7d4eSAl Viro } else { 2195e41f7d4eSAl Viro get_fs_pwd(current->fs, &nd->path); 2196ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2197e41f7d4eSAl Viro } 2198ef55d917SAl Viro return s; 219931e6b01fSNick Piggin } else { 2200582aa64aSJeff Layton /* Caller must check execute permissions on the starting path component */ 2201c8a53ee5SAl Viro struct fd f = fdget_raw(nd->dfd); 220231e6b01fSNick Piggin struct dentry *dentry; 220331e6b01fSNick Piggin 22042903ff01SAl Viro if (!f.file) 2205368ee9baSAl Viro return ERR_PTR(-EBADF); 220631e6b01fSNick Piggin 22072903ff01SAl Viro dentry = f.file->f_path.dentry; 220831e6b01fSNick Piggin 2209fd2f7cb5SAl Viro if (*s) { 221044b1d530SMiklos Szeredi if (!d_can_lookup(dentry)) { 22112903ff01SAl Viro fdput(f); 2212368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 2213f52e0c11SAl Viro } 22142903ff01SAl Viro } 22152903ff01SAl Viro 22162903ff01SAl Viro nd->path = f.file->f_path; 2217e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 22188b61e74fSAl Viro rcu_read_lock(); 221934a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 222034a26b99SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 22215590ff0dSUlrich Drepper } else { 22222903ff01SAl Viro path_get(&nd->path); 222334a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 22241da177e4SLinus Torvalds } 222534a26b99SAl Viro fdput(f); 2226368ee9baSAl Viro return s; 2227e41f7d4eSAl Viro } 22289b4a9b14SAl Viro } 22299b4a9b14SAl Viro 22303bdba28bSAl Viro static const char *trailing_symlink(struct nameidata *nd) 223195fa25d9SAl Viro { 223295fa25d9SAl Viro const char *s; 2233fec2fa24SAl Viro int error = may_follow_link(nd); 2234deb106c6SAl Viro if (unlikely(error)) 22353bdba28bSAl Viro return ERR_PTR(error); 223695fa25d9SAl Viro nd->flags |= LOOKUP_PARENT; 2237fab51e8aSAl Viro nd->stack[0].name = NULL; 22383b2e7f75SAl Viro s = get_link(nd); 2239deb106c6SAl Viro return s ? s : ""; 224095fa25d9SAl Viro } 224195fa25d9SAl Viro 2242caa85634SAl Viro static inline int lookup_last(struct nameidata *nd) 2243bd92d7feSAl Viro { 2244bd92d7feSAl Viro if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) 2245bd92d7feSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 2246bd92d7feSAl Viro 2247bd92d7feSAl Viro nd->flags &= ~LOOKUP_PARENT; 22481c4ff1a8SAl Viro return walk_component(nd, 0); 2249bd92d7feSAl Viro } 2250bd92d7feSAl Viro 22519b4a9b14SAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2252c8a53ee5SAl Viro static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path) 22539b4a9b14SAl Viro { 2254c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2255bd92d7feSAl Viro int err; 225631e6b01fSNick Piggin 2257368ee9baSAl Viro if (IS_ERR(s)) 2258368ee9baSAl Viro return PTR_ERR(s); 22593bdba28bSAl Viro while (!(err = link_path_walk(s, nd)) 22603bdba28bSAl Viro && ((err = lookup_last(nd)) > 0)) { 22613bdba28bSAl Viro s = trailing_symlink(nd); 22623bdba28bSAl Viro if (IS_ERR(s)) { 22633bdba28bSAl Viro err = PTR_ERR(s); 22646d7b5aaeSAl Viro break; 2265bd92d7feSAl Viro } 2266bd92d7feSAl Viro } 22679f1fafeeSAl Viro if (!err) 22689f1fafeeSAl Viro err = complete_walk(nd); 2269bd92d7feSAl Viro 2270deb106c6SAl Viro if (!err && nd->flags & LOOKUP_DIRECTORY) 2271deb106c6SAl Viro if (!d_can_lookup(nd->path.dentry)) 2272bd23a539SAl Viro err = -ENOTDIR; 2273625b6d10SAl Viro if (!err) { 2274625b6d10SAl Viro *path = nd->path; 2275625b6d10SAl Viro nd->path.mnt = NULL; 2276625b6d10SAl Viro nd->path.dentry = NULL; 2277625b6d10SAl Viro } 2278deb106c6SAl Viro terminate_walk(nd); 2279bd92d7feSAl Viro return err; 228031e6b01fSNick Piggin } 228131e6b01fSNick Piggin 2282625b6d10SAl Viro static int filename_lookup(int dfd, struct filename *name, unsigned flags, 22839ad1aaa6SAl Viro struct path *path, struct path *root) 2284873f1eedSJeff Layton { 2285894bc8c4SAl Viro int retval; 22869883d185SAl Viro struct nameidata nd; 2287abc9f5beSAl Viro if (IS_ERR(name)) 2288abc9f5beSAl Viro return PTR_ERR(name); 22899ad1aaa6SAl Viro if (unlikely(root)) { 22909ad1aaa6SAl Viro nd.root = *root; 22919ad1aaa6SAl Viro flags |= LOOKUP_ROOT; 22929ad1aaa6SAl Viro } 22939883d185SAl Viro set_nameidata(&nd, dfd, name); 2294c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_RCU, path); 2295873f1eedSJeff Layton if (unlikely(retval == -ECHILD)) 2296c8a53ee5SAl Viro retval = path_lookupat(&nd, flags, path); 2297873f1eedSJeff Layton if (unlikely(retval == -ESTALE)) 2298c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path); 2299873f1eedSJeff Layton 2300873f1eedSJeff Layton if (likely(!retval)) 2301625b6d10SAl Viro audit_inode(name, path->dentry, flags & LOOKUP_PARENT); 23029883d185SAl Viro restore_nameidata(); 2303e4bd1c1aSAl Viro putname(name); 2304873f1eedSJeff Layton return retval; 2305873f1eedSJeff Layton } 2306873f1eedSJeff Layton 23078bcb77faSAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2308c8a53ee5SAl Viro static int path_parentat(struct nameidata *nd, unsigned flags, 2309391172c4SAl Viro struct path *parent) 23108bcb77faSAl Viro { 2311c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2312368ee9baSAl Viro int err; 2313368ee9baSAl Viro if (IS_ERR(s)) 2314368ee9baSAl Viro return PTR_ERR(s); 2315368ee9baSAl Viro err = link_path_walk(s, nd); 23168bcb77faSAl Viro if (!err) 23178bcb77faSAl Viro err = complete_walk(nd); 2318391172c4SAl Viro if (!err) { 2319391172c4SAl Viro *parent = nd->path; 2320391172c4SAl Viro nd->path.mnt = NULL; 2321391172c4SAl Viro nd->path.dentry = NULL; 2322391172c4SAl Viro } 2323deb106c6SAl Viro terminate_walk(nd); 23248bcb77faSAl Viro return err; 23258bcb77faSAl Viro } 23268bcb77faSAl Viro 23275c31b6ceSAl Viro static struct filename *filename_parentat(int dfd, struct filename *name, 2328391172c4SAl Viro unsigned int flags, struct path *parent, 2329391172c4SAl Viro struct qstr *last, int *type) 23308bcb77faSAl Viro { 23318bcb77faSAl Viro int retval; 23329883d185SAl Viro struct nameidata nd; 23338bcb77faSAl Viro 23345c31b6ceSAl Viro if (IS_ERR(name)) 23355c31b6ceSAl Viro return name; 23369883d185SAl Viro set_nameidata(&nd, dfd, name); 2337c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_RCU, parent); 23388bcb77faSAl Viro if (unlikely(retval == -ECHILD)) 2339c8a53ee5SAl Viro retval = path_parentat(&nd, flags, parent); 23408bcb77faSAl Viro if (unlikely(retval == -ESTALE)) 2341c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent); 2342391172c4SAl Viro if (likely(!retval)) { 2343391172c4SAl Viro *last = nd.last; 2344391172c4SAl Viro *type = nd.last_type; 2345391172c4SAl Viro audit_inode(name, parent->dentry, LOOKUP_PARENT); 23465c31b6ceSAl Viro } else { 23475c31b6ceSAl Viro putname(name); 23485c31b6ceSAl Viro name = ERR_PTR(retval); 2349391172c4SAl Viro } 23509883d185SAl Viro restore_nameidata(); 23515c31b6ceSAl Viro return name; 23528bcb77faSAl Viro } 23538bcb77faSAl Viro 235479714f72SAl Viro /* does lookup, returns the object with parent locked */ 235579714f72SAl Viro struct dentry *kern_path_locked(const char *name, struct path *path) 23565590ff0dSUlrich Drepper { 23575c31b6ceSAl Viro struct filename *filename; 23585c31b6ceSAl Viro struct dentry *d; 2359391172c4SAl Viro struct qstr last; 2360391172c4SAl Viro int type; 236151689104SPaul Moore 23625c31b6ceSAl Viro filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path, 23635c31b6ceSAl Viro &last, &type); 236451689104SPaul Moore if (IS_ERR(filename)) 236551689104SPaul Moore return ERR_CAST(filename); 23665c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) { 2367391172c4SAl Viro path_put(path); 23685c31b6ceSAl Viro putname(filename); 23695c31b6ceSAl Viro return ERR_PTR(-EINVAL); 237079714f72SAl Viro } 23715955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 2372391172c4SAl Viro d = __lookup_hash(&last, path->dentry, 0); 237379714f72SAl Viro if (IS_ERR(d)) { 23745955102cSAl Viro inode_unlock(path->dentry->d_inode); 2375391172c4SAl Viro path_put(path); 237679714f72SAl Viro } 237751689104SPaul Moore putname(filename); 237879714f72SAl Viro return d; 23795590ff0dSUlrich Drepper } 23805590ff0dSUlrich Drepper 2381d1811465SAl Viro int kern_path(const char *name, unsigned int flags, struct path *path) 2382d1811465SAl Viro { 2383abc9f5beSAl Viro return filename_lookup(AT_FDCWD, getname_kernel(name), 2384abc9f5beSAl Viro flags, path, NULL); 2385d1811465SAl Viro } 23864d359507SAl Viro EXPORT_SYMBOL(kern_path); 2387d1811465SAl Viro 238816f18200SJosef 'Jeff' Sipek /** 238916f18200SJosef 'Jeff' Sipek * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair 239016f18200SJosef 'Jeff' Sipek * @dentry: pointer to dentry of the base directory 239116f18200SJosef 'Jeff' Sipek * @mnt: pointer to vfs mount of the base directory 239216f18200SJosef 'Jeff' Sipek * @name: pointer to file name 239316f18200SJosef 'Jeff' Sipek * @flags: lookup flags 2394e0a01249SAl Viro * @path: pointer to struct path to fill 239516f18200SJosef 'Jeff' Sipek */ 239616f18200SJosef 'Jeff' Sipek int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, 239716f18200SJosef 'Jeff' Sipek const char *name, unsigned int flags, 2398e0a01249SAl Viro struct path *path) 239916f18200SJosef 'Jeff' Sipek { 24009ad1aaa6SAl Viro struct path root = {.mnt = mnt, .dentry = dentry}; 24019ad1aaa6SAl Viro /* the first argument of filename_lookup() is ignored with root */ 2402abc9f5beSAl Viro return filename_lookup(AT_FDCWD, getname_kernel(name), 2403abc9f5beSAl Viro flags , path, &root); 240416f18200SJosef 'Jeff' Sipek } 24054d359507SAl Viro EXPORT_SYMBOL(vfs_path_lookup); 240616f18200SJosef 'Jeff' Sipek 2407eead1911SChristoph Hellwig /** 2408a6b91919SRandy Dunlap * lookup_one_len - filesystem helper to lookup single pathname component 2409eead1911SChristoph Hellwig * @name: pathname component to lookup 2410eead1911SChristoph Hellwig * @base: base directory to lookup from 2411eead1911SChristoph Hellwig * @len: maximum length @len should be interpreted to 2412eead1911SChristoph Hellwig * 2413a6b91919SRandy Dunlap * Note that this routine is purely a helper for filesystem usage and should 24149e7543e9SAl Viro * not be called by generic code. 2415bbddca8eSNeilBrown * 2416bbddca8eSNeilBrown * The caller must hold base->i_mutex. 2417eead1911SChristoph Hellwig */ 2418057f6c01SJames Morris struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 2419057f6c01SJames Morris { 2420057f6c01SJames Morris struct qstr this; 24216a96ba54SAl Viro unsigned int c; 2422cda309deSMiklos Szeredi int err; 2423057f6c01SJames Morris 24245955102cSAl Viro WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 24252f9092e1SDavid Woodhouse 24266a96ba54SAl Viro this.name = name; 24276a96ba54SAl Viro this.len = len; 24288387ff25SLinus Torvalds this.hash = full_name_hash(base, name, len); 24296a96ba54SAl Viro if (!len) 24306a96ba54SAl Viro return ERR_PTR(-EACCES); 24316a96ba54SAl Viro 243221d8a15aSAl Viro if (unlikely(name[0] == '.')) { 243321d8a15aSAl Viro if (len < 2 || (len == 2 && name[1] == '.')) 243421d8a15aSAl Viro return ERR_PTR(-EACCES); 243521d8a15aSAl Viro } 243621d8a15aSAl Viro 24376a96ba54SAl Viro while (len--) { 24386a96ba54SAl Viro c = *(const unsigned char *)name++; 24396a96ba54SAl Viro if (c == '/' || c == '\0') 24406a96ba54SAl Viro return ERR_PTR(-EACCES); 24416a96ba54SAl Viro } 24425a202bcdSAl Viro /* 24435a202bcdSAl Viro * See if the low-level filesystem might want 24445a202bcdSAl Viro * to use its own hash.. 24455a202bcdSAl Viro */ 24465a202bcdSAl Viro if (base->d_flags & DCACHE_OP_HASH) { 2447da53be12SLinus Torvalds int err = base->d_op->d_hash(base, &this); 24485a202bcdSAl Viro if (err < 0) 24495a202bcdSAl Viro return ERR_PTR(err); 24505a202bcdSAl Viro } 2451eead1911SChristoph Hellwig 2452cda309deSMiklos Szeredi err = inode_permission(base->d_inode, MAY_EXEC); 2453cda309deSMiklos Szeredi if (err) 2454cda309deSMiklos Szeredi return ERR_PTR(err); 2455cda309deSMiklos Szeredi 245672bd866aSAl Viro return __lookup_hash(&this, base, 0); 2457057f6c01SJames Morris } 24584d359507SAl Viro EXPORT_SYMBOL(lookup_one_len); 2459057f6c01SJames Morris 2460bbddca8eSNeilBrown /** 2461bbddca8eSNeilBrown * lookup_one_len_unlocked - filesystem helper to lookup single pathname component 2462bbddca8eSNeilBrown * @name: pathname component to lookup 2463bbddca8eSNeilBrown * @base: base directory to lookup from 2464bbddca8eSNeilBrown * @len: maximum length @len should be interpreted to 2465bbddca8eSNeilBrown * 2466bbddca8eSNeilBrown * Note that this routine is purely a helper for filesystem usage and should 2467bbddca8eSNeilBrown * not be called by generic code. 2468bbddca8eSNeilBrown * 2469bbddca8eSNeilBrown * Unlike lookup_one_len, it should be called without the parent 2470bbddca8eSNeilBrown * i_mutex held, and will take the i_mutex itself if necessary. 2471bbddca8eSNeilBrown */ 2472bbddca8eSNeilBrown struct dentry *lookup_one_len_unlocked(const char *name, 2473bbddca8eSNeilBrown struct dentry *base, int len) 2474bbddca8eSNeilBrown { 2475bbddca8eSNeilBrown struct qstr this; 2476bbddca8eSNeilBrown unsigned int c; 2477bbddca8eSNeilBrown int err; 247820d00ee8SLinus Torvalds struct dentry *ret; 2479bbddca8eSNeilBrown 2480bbddca8eSNeilBrown this.name = name; 2481bbddca8eSNeilBrown this.len = len; 24828387ff25SLinus Torvalds this.hash = full_name_hash(base, name, len); 2483bbddca8eSNeilBrown if (!len) 2484bbddca8eSNeilBrown return ERR_PTR(-EACCES); 2485bbddca8eSNeilBrown 2486bbddca8eSNeilBrown if (unlikely(name[0] == '.')) { 2487bbddca8eSNeilBrown if (len < 2 || (len == 2 && name[1] == '.')) 2488bbddca8eSNeilBrown return ERR_PTR(-EACCES); 2489bbddca8eSNeilBrown } 2490bbddca8eSNeilBrown 2491bbddca8eSNeilBrown while (len--) { 2492bbddca8eSNeilBrown c = *(const unsigned char *)name++; 2493bbddca8eSNeilBrown if (c == '/' || c == '\0') 2494bbddca8eSNeilBrown return ERR_PTR(-EACCES); 2495bbddca8eSNeilBrown } 2496bbddca8eSNeilBrown /* 2497bbddca8eSNeilBrown * See if the low-level filesystem might want 2498bbddca8eSNeilBrown * to use its own hash.. 2499bbddca8eSNeilBrown */ 2500bbddca8eSNeilBrown if (base->d_flags & DCACHE_OP_HASH) { 2501bbddca8eSNeilBrown int err = base->d_op->d_hash(base, &this); 2502bbddca8eSNeilBrown if (err < 0) 2503bbddca8eSNeilBrown return ERR_PTR(err); 2504bbddca8eSNeilBrown } 2505bbddca8eSNeilBrown 2506bbddca8eSNeilBrown err = inode_permission(base->d_inode, MAY_EXEC); 2507bbddca8eSNeilBrown if (err) 2508bbddca8eSNeilBrown return ERR_PTR(err); 2509bbddca8eSNeilBrown 251020d00ee8SLinus Torvalds ret = lookup_dcache(&this, base, 0); 251120d00ee8SLinus Torvalds if (!ret) 251220d00ee8SLinus Torvalds ret = lookup_slow(&this, base, 0); 251320d00ee8SLinus Torvalds return ret; 2514bbddca8eSNeilBrown } 2515bbddca8eSNeilBrown EXPORT_SYMBOL(lookup_one_len_unlocked); 2516bbddca8eSNeilBrown 2517eedf265aSEric W. Biederman #ifdef CONFIG_UNIX98_PTYS 2518eedf265aSEric W. Biederman int path_pts(struct path *path) 2519eedf265aSEric W. Biederman { 2520eedf265aSEric W. Biederman /* Find something mounted on "pts" in the same directory as 2521eedf265aSEric W. Biederman * the input path. 2522eedf265aSEric W. Biederman */ 2523eedf265aSEric W. Biederman struct dentry *child, *parent; 2524eedf265aSEric W. Biederman struct qstr this; 2525eedf265aSEric W. Biederman int ret; 2526eedf265aSEric W. Biederman 2527eedf265aSEric W. Biederman ret = path_parent_directory(path); 2528eedf265aSEric W. Biederman if (ret) 2529eedf265aSEric W. Biederman return ret; 2530eedf265aSEric W. Biederman 2531eedf265aSEric W. Biederman parent = path->dentry; 2532eedf265aSEric W. Biederman this.name = "pts"; 2533eedf265aSEric W. Biederman this.len = 3; 2534eedf265aSEric W. Biederman child = d_hash_and_lookup(parent, &this); 2535eedf265aSEric W. Biederman if (!child) 2536eedf265aSEric W. Biederman return -ENOENT; 2537eedf265aSEric W. Biederman 2538eedf265aSEric W. Biederman path->dentry = child; 2539eedf265aSEric W. Biederman dput(parent); 2540eedf265aSEric W. Biederman follow_mount(path); 2541eedf265aSEric W. Biederman return 0; 2542eedf265aSEric W. Biederman } 2543eedf265aSEric W. Biederman #endif 2544eedf265aSEric W. Biederman 25451fa1e7f6SAndy Whitcroft int user_path_at_empty(int dfd, const char __user *name, unsigned flags, 25461fa1e7f6SAndy Whitcroft struct path *path, int *empty) 25471da177e4SLinus Torvalds { 2548abc9f5beSAl Viro return filename_lookup(dfd, getname_flags(name, flags, empty), 2549abc9f5beSAl Viro flags, path, NULL); 25501da177e4SLinus Torvalds } 2551b853a161SAl Viro EXPORT_SYMBOL(user_path_at_empty); 25521fa1e7f6SAndy Whitcroft 25538033426eSJeff Layton /** 2554197df04cSAl Viro * mountpoint_last - look up last component for umount 25558033426eSJeff Layton * @nd: pathwalk nameidata - currently pointing at parent directory of "last" 25568033426eSJeff Layton * 25578033426eSJeff Layton * This is a special lookup_last function just for umount. In this case, we 25588033426eSJeff Layton * need to resolve the path without doing any revalidation. 25598033426eSJeff Layton * 25608033426eSJeff Layton * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since 25618033426eSJeff Layton * mountpoints are always pinned in the dcache, their ancestors are too. Thus, 25628033426eSJeff Layton * in almost all cases, this lookup will be served out of the dcache. The only 25638033426eSJeff Layton * cases where it won't are if nd->last refers to a symlink or the path is 25648033426eSJeff Layton * bogus and it doesn't exist. 25658033426eSJeff Layton * 25668033426eSJeff Layton * Returns: 25678033426eSJeff Layton * -error: if there was an error during lookup. This includes -ENOENT if the 2568ba8f4613SAl Viro * lookup found a negative dentry. 25698033426eSJeff Layton * 2570ba8f4613SAl Viro * 0: if we successfully resolved nd->last and found it to not to be a 2571ba8f4613SAl Viro * symlink that needs to be followed. 25728033426eSJeff Layton * 25738033426eSJeff Layton * 1: if we successfully resolved nd->last and found it to be a symlink 2574ba8f4613SAl Viro * that needs to be followed. 25758033426eSJeff Layton */ 25768033426eSJeff Layton static int 2577ba8f4613SAl Viro mountpoint_last(struct nameidata *nd) 25788033426eSJeff Layton { 25798033426eSJeff Layton int error = 0; 25808033426eSJeff Layton struct dentry *dir = nd->path.dentry; 2581ba8f4613SAl Viro struct path path; 25828033426eSJeff Layton 258335759521SAl Viro /* If we're in rcuwalk, drop out of it to handle last component */ 258435759521SAl Viro if (nd->flags & LOOKUP_RCU) { 25856e9918b7SAl Viro if (unlazy_walk(nd, NULL, 0)) 2586deb106c6SAl Viro return -ECHILD; 25878033426eSJeff Layton } 25888033426eSJeff Layton 25898033426eSJeff Layton nd->flags &= ~LOOKUP_PARENT; 25908033426eSJeff Layton 25918033426eSJeff Layton if (unlikely(nd->last_type != LAST_NORM)) { 25928033426eSJeff Layton error = handle_dots(nd, nd->last_type); 259335759521SAl Viro if (error) 2594deb106c6SAl Viro return error; 2595ba8f4613SAl Viro path.dentry = dget(nd->path.dentry); 2596949a852eSAl Viro } else { 2597ba8f4613SAl Viro path.dentry = d_lookup(dir, &nd->last); 2598ba8f4613SAl Viro if (!path.dentry) { 25998033426eSJeff Layton /* 2600949a852eSAl Viro * No cached dentry. Mounted dentries are pinned in the 2601949a852eSAl Viro * cache, so that means that this dentry is probably 2602949a852eSAl Viro * a symlink or the path doesn't actually point 2603949a852eSAl Viro * to a mounted dentry. 26048033426eSJeff Layton */ 2605ba8f4613SAl Viro path.dentry = lookup_slow(&nd->last, dir, 2606949a852eSAl Viro nd->flags | LOOKUP_NO_REVAL); 2607ba8f4613SAl Viro if (IS_ERR(path.dentry)) 2608ba8f4613SAl Viro return PTR_ERR(path.dentry); 26098033426eSJeff Layton } 2610bcceeebaSDave Jones } 2611ba8f4613SAl Viro if (d_is_negative(path.dentry)) { 2612ba8f4613SAl Viro dput(path.dentry); 2613deb106c6SAl Viro return -ENOENT; 261435759521SAl Viro } 2615ba8f4613SAl Viro path.mnt = nd->path.mnt; 26168f64fb1cSAl Viro return step_into(nd, &path, 0, d_backing_inode(path.dentry), 0); 26178033426eSJeff Layton } 26188033426eSJeff Layton 26198033426eSJeff Layton /** 2620197df04cSAl Viro * path_mountpoint - look up a path to be umounted 26212a78b857SMasanari Iida * @nd: lookup context 26228033426eSJeff Layton * @flags: lookup flags 2623c8a53ee5SAl Viro * @path: pointer to container for result 26248033426eSJeff Layton * 26258033426eSJeff Layton * Look up the given name, but don't attempt to revalidate the last component. 2626606d6fe3SRandy Dunlap * Returns 0 and "path" will be valid on success; Returns error otherwise. 26278033426eSJeff Layton */ 26288033426eSJeff Layton static int 2629c8a53ee5SAl Viro path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path) 26308033426eSJeff Layton { 2631c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2632368ee9baSAl Viro int err; 2633368ee9baSAl Viro if (IS_ERR(s)) 2634368ee9baSAl Viro return PTR_ERR(s); 26353bdba28bSAl Viro while (!(err = link_path_walk(s, nd)) && 2636ba8f4613SAl Viro (err = mountpoint_last(nd)) > 0) { 26373bdba28bSAl Viro s = trailing_symlink(nd); 26383bdba28bSAl Viro if (IS_ERR(s)) { 26393bdba28bSAl Viro err = PTR_ERR(s); 26408033426eSJeff Layton break; 26418033426eSJeff Layton } 26423bdba28bSAl Viro } 2643ba8f4613SAl Viro if (!err) { 2644ba8f4613SAl Viro *path = nd->path; 2645ba8f4613SAl Viro nd->path.mnt = NULL; 2646ba8f4613SAl Viro nd->path.dentry = NULL; 2647ba8f4613SAl Viro follow_mount(path); 2648ba8f4613SAl Viro } 2649deb106c6SAl Viro terminate_walk(nd); 26508033426eSJeff Layton return err; 26518033426eSJeff Layton } 26528033426eSJeff Layton 26532d864651SAl Viro static int 2654668696dcSAl Viro filename_mountpoint(int dfd, struct filename *name, struct path *path, 26552d864651SAl Viro unsigned int flags) 26562d864651SAl Viro { 26579883d185SAl Viro struct nameidata nd; 2658cbaab2dbSAl Viro int error; 2659668696dcSAl Viro if (IS_ERR(name)) 2660668696dcSAl Viro return PTR_ERR(name); 26619883d185SAl Viro set_nameidata(&nd, dfd, name); 2662c8a53ee5SAl Viro error = path_mountpoint(&nd, flags | LOOKUP_RCU, path); 26632d864651SAl Viro if (unlikely(error == -ECHILD)) 2664c8a53ee5SAl Viro error = path_mountpoint(&nd, flags, path); 26652d864651SAl Viro if (unlikely(error == -ESTALE)) 2666c8a53ee5SAl Viro error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path); 26672d864651SAl Viro if (likely(!error)) 2668668696dcSAl Viro audit_inode(name, path->dentry, 0); 26699883d185SAl Viro restore_nameidata(); 2670668696dcSAl Viro putname(name); 26712d864651SAl Viro return error; 26722d864651SAl Viro } 26732d864651SAl Viro 26748033426eSJeff Layton /** 2675197df04cSAl Viro * user_path_mountpoint_at - lookup a path from userland in order to umount it 26768033426eSJeff Layton * @dfd: directory file descriptor 26778033426eSJeff Layton * @name: pathname from userland 26788033426eSJeff Layton * @flags: lookup flags 26798033426eSJeff Layton * @path: pointer to container to hold result 26808033426eSJeff Layton * 26818033426eSJeff Layton * A umount is a special case for path walking. We're not actually interested 26828033426eSJeff Layton * in the inode in this situation, and ESTALE errors can be a problem. We 26838033426eSJeff Layton * simply want track down the dentry and vfsmount attached at the mountpoint 26848033426eSJeff Layton * and avoid revalidating the last component. 26858033426eSJeff Layton * 26868033426eSJeff Layton * Returns 0 and populates "path" on success. 26878033426eSJeff Layton */ 26888033426eSJeff Layton int 2689197df04cSAl Viro user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags, 26908033426eSJeff Layton struct path *path) 26918033426eSJeff Layton { 2692cbaab2dbSAl Viro return filename_mountpoint(dfd, getname(name), path, flags); 26938033426eSJeff Layton } 26948033426eSJeff Layton 26952d864651SAl Viro int 26962d864651SAl Viro kern_path_mountpoint(int dfd, const char *name, struct path *path, 26972d864651SAl Viro unsigned int flags) 26982d864651SAl Viro { 2699cbaab2dbSAl Viro return filename_mountpoint(dfd, getname_kernel(name), path, flags); 27002d864651SAl Viro } 27012d864651SAl Viro EXPORT_SYMBOL(kern_path_mountpoint); 27022d864651SAl Viro 2703cbdf35bcSMiklos Szeredi int __check_sticky(struct inode *dir, struct inode *inode) 27041da177e4SLinus Torvalds { 27058e96e3b7SEric W. Biederman kuid_t fsuid = current_fsuid(); 2706da9592edSDavid Howells 27078e96e3b7SEric W. Biederman if (uid_eq(inode->i_uid, fsuid)) 27081da177e4SLinus Torvalds return 0; 27098e96e3b7SEric W. Biederman if (uid_eq(dir->i_uid, fsuid)) 27101da177e4SLinus Torvalds return 0; 271123adbe12SAndy Lutomirski return !capable_wrt_inode_uidgid(inode, CAP_FOWNER); 27121da177e4SLinus Torvalds } 2713cbdf35bcSMiklos Szeredi EXPORT_SYMBOL(__check_sticky); 27141da177e4SLinus Torvalds 27151da177e4SLinus Torvalds /* 27161da177e4SLinus Torvalds * Check whether we can remove a link victim from directory dir, check 27171da177e4SLinus Torvalds * whether the type of victim is right. 27181da177e4SLinus Torvalds * 1. We can't do it if dir is read-only (done in permission()) 27191da177e4SLinus Torvalds * 2. We should have write and exec permissions on dir 27201da177e4SLinus Torvalds * 3. We can't remove anything from append-only dir 27211da177e4SLinus Torvalds * 4. We can't do anything with immutable dir (done in permission()) 27221da177e4SLinus Torvalds * 5. If the sticky bit on dir is set we should either 27231da177e4SLinus Torvalds * a. be owner of dir, or 27241da177e4SLinus Torvalds * b. be owner of victim, or 27251da177e4SLinus Torvalds * c. have CAP_FOWNER capability 27261da177e4SLinus Torvalds * 6. If the victim is append-only or immutable we can't do antyhing with 27271da177e4SLinus Torvalds * links pointing to it. 27280bd23d09SEric W. Biederman * 7. If the victim has an unknown uid or gid we can't change the inode. 27290bd23d09SEric W. Biederman * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR. 27300bd23d09SEric W. Biederman * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR. 27310bd23d09SEric W. Biederman * 10. We can't remove a root or mountpoint. 27320bd23d09SEric W. Biederman * 11. We don't allow removal of NFS sillyrenamed files; it's handled by 27331da177e4SLinus Torvalds * nfs_async_unlink(). 27341da177e4SLinus Torvalds */ 2735b18825a7SDavid Howells static int may_delete(struct inode *dir, struct dentry *victim, bool isdir) 27361da177e4SLinus Torvalds { 273763afdfc7SDavid Howells struct inode *inode = d_backing_inode(victim); 27381da177e4SLinus Torvalds int error; 27391da177e4SLinus Torvalds 2740b18825a7SDavid Howells if (d_is_negative(victim)) 27411da177e4SLinus Torvalds return -ENOENT; 2742b18825a7SDavid Howells BUG_ON(!inode); 27431da177e4SLinus Torvalds 27441da177e4SLinus Torvalds BUG_ON(victim->d_parent->d_inode != dir); 27454fa6b5ecSJeff Layton audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); 27461da177e4SLinus Torvalds 2747f419a2e3SAl Viro error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 27481da177e4SLinus Torvalds if (error) 27491da177e4SLinus Torvalds return error; 27501da177e4SLinus Torvalds if (IS_APPEND(dir)) 27511da177e4SLinus Torvalds return -EPERM; 2752b18825a7SDavid Howells 2753b18825a7SDavid Howells if (check_sticky(dir, inode) || IS_APPEND(inode) || 27540bd23d09SEric W. Biederman IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || HAS_UNMAPPED_ID(inode)) 27551da177e4SLinus Torvalds return -EPERM; 27561da177e4SLinus Torvalds if (isdir) { 275744b1d530SMiklos Szeredi if (!d_is_dir(victim)) 27581da177e4SLinus Torvalds return -ENOTDIR; 27591da177e4SLinus Torvalds if (IS_ROOT(victim)) 27601da177e4SLinus Torvalds return -EBUSY; 276144b1d530SMiklos Szeredi } else if (d_is_dir(victim)) 27621da177e4SLinus Torvalds return -EISDIR; 27631da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 27641da177e4SLinus Torvalds return -ENOENT; 27651da177e4SLinus Torvalds if (victim->d_flags & DCACHE_NFSFS_RENAMED) 27661da177e4SLinus Torvalds return -EBUSY; 27671da177e4SLinus Torvalds return 0; 27681da177e4SLinus Torvalds } 27691da177e4SLinus Torvalds 27701da177e4SLinus Torvalds /* Check whether we can create an object with dentry child in directory 27711da177e4SLinus Torvalds * dir. 27721da177e4SLinus Torvalds * 1. We can't do it if child already exists (open has special treatment for 27731da177e4SLinus Torvalds * this case, but since we are inlined it's OK) 27741da177e4SLinus Torvalds * 2. We can't do it if dir is read-only (done in permission()) 2775036d5236SEric W. Biederman * 3. We can't do it if the fs can't represent the fsuid or fsgid. 2776036d5236SEric W. Biederman * 4. We should have write and exec permissions on dir 2777036d5236SEric W. Biederman * 5. We can't do it if dir is immutable (done in permission()) 27781da177e4SLinus Torvalds */ 2779a95164d9SMiklos Szeredi static inline int may_create(struct inode *dir, struct dentry *child) 27801da177e4SLinus Torvalds { 2781036d5236SEric W. Biederman struct user_namespace *s_user_ns; 278214e972b4SJeff Layton audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); 27831da177e4SLinus Torvalds if (child->d_inode) 27841da177e4SLinus Torvalds return -EEXIST; 27851da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 27861da177e4SLinus Torvalds return -ENOENT; 2787036d5236SEric W. Biederman s_user_ns = dir->i_sb->s_user_ns; 2788036d5236SEric W. Biederman if (!kuid_has_mapping(s_user_ns, current_fsuid()) || 2789036d5236SEric W. Biederman !kgid_has_mapping(s_user_ns, current_fsgid())) 2790036d5236SEric W. Biederman return -EOVERFLOW; 2791f419a2e3SAl Viro return inode_permission(dir, MAY_WRITE | MAY_EXEC); 27921da177e4SLinus Torvalds } 27931da177e4SLinus Torvalds 27941da177e4SLinus Torvalds /* 27951da177e4SLinus Torvalds * p1 and p2 should be directories on the same fs. 27961da177e4SLinus Torvalds */ 27971da177e4SLinus Torvalds struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 27981da177e4SLinus Torvalds { 27991da177e4SLinus Torvalds struct dentry *p; 28001da177e4SLinus Torvalds 28011da177e4SLinus Torvalds if (p1 == p2) { 28025955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 28031da177e4SLinus Torvalds return NULL; 28041da177e4SLinus Torvalds } 28051da177e4SLinus Torvalds 2806fc64005cSAl Viro mutex_lock(&p1->d_sb->s_vfs_rename_mutex); 28071da177e4SLinus Torvalds 2808e2761a11SOGAWA Hirofumi p = d_ancestor(p2, p1); 2809e2761a11SOGAWA Hirofumi if (p) { 28105955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT); 28115955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_CHILD); 28121da177e4SLinus Torvalds return p; 28131da177e4SLinus Torvalds } 28141da177e4SLinus Torvalds 2815e2761a11SOGAWA Hirofumi p = d_ancestor(p1, p2); 2816e2761a11SOGAWA Hirofumi if (p) { 28175955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 28185955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_CHILD); 28191da177e4SLinus Torvalds return p; 28201da177e4SLinus Torvalds } 28211da177e4SLinus Torvalds 28225955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 28235955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2); 28241da177e4SLinus Torvalds return NULL; 28251da177e4SLinus Torvalds } 28264d359507SAl Viro EXPORT_SYMBOL(lock_rename); 28271da177e4SLinus Torvalds 28281da177e4SLinus Torvalds void unlock_rename(struct dentry *p1, struct dentry *p2) 28291da177e4SLinus Torvalds { 28305955102cSAl Viro inode_unlock(p1->d_inode); 28311da177e4SLinus Torvalds if (p1 != p2) { 28325955102cSAl Viro inode_unlock(p2->d_inode); 2833fc64005cSAl Viro mutex_unlock(&p1->d_sb->s_vfs_rename_mutex); 28341da177e4SLinus Torvalds } 28351da177e4SLinus Torvalds } 28364d359507SAl Viro EXPORT_SYMBOL(unlock_rename); 28371da177e4SLinus Torvalds 28384acdaf27SAl Viro int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, 2839312b63fbSAl Viro bool want_excl) 28401da177e4SLinus Torvalds { 2841a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 28421da177e4SLinus Torvalds if (error) 28431da177e4SLinus Torvalds return error; 28441da177e4SLinus Torvalds 2845acfa4380SAl Viro if (!dir->i_op->create) 28461da177e4SLinus Torvalds return -EACCES; /* shouldn't it be ENOSYS? */ 28471da177e4SLinus Torvalds mode &= S_IALLUGO; 28481da177e4SLinus Torvalds mode |= S_IFREG; 28491da177e4SLinus Torvalds error = security_inode_create(dir, dentry, mode); 28501da177e4SLinus Torvalds if (error) 28511da177e4SLinus Torvalds return error; 2852312b63fbSAl Viro error = dir->i_op->create(dir, dentry, mode, want_excl); 2853a74574aaSStephen Smalley if (!error) 2854f38aa942SAmy Griffis fsnotify_create(dir, dentry); 28551da177e4SLinus Torvalds return error; 28561da177e4SLinus Torvalds } 28574d359507SAl Viro EXPORT_SYMBOL(vfs_create); 28581da177e4SLinus Torvalds 2859a2982cc9SEric W. Biederman bool may_open_dev(const struct path *path) 2860a2982cc9SEric W. Biederman { 2861a2982cc9SEric W. Biederman return !(path->mnt->mnt_flags & MNT_NODEV) && 2862a2982cc9SEric W. Biederman !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV); 2863a2982cc9SEric W. Biederman } 2864a2982cc9SEric W. Biederman 2865f0bb5aafSAl Viro static int may_open(const struct path *path, int acc_mode, int flag) 28661da177e4SLinus Torvalds { 28673fb64190SChristoph Hellwig struct dentry *dentry = path->dentry; 28681da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 28691da177e4SLinus Torvalds int error; 28701da177e4SLinus Torvalds 28711da177e4SLinus Torvalds if (!inode) 28721da177e4SLinus Torvalds return -ENOENT; 28731da177e4SLinus Torvalds 2874c8fe8f30SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 2875c8fe8f30SChristoph Hellwig case S_IFLNK: 28761da177e4SLinus Torvalds return -ELOOP; 2877c8fe8f30SChristoph Hellwig case S_IFDIR: 2878c8fe8f30SChristoph Hellwig if (acc_mode & MAY_WRITE) 28791da177e4SLinus Torvalds return -EISDIR; 2880c8fe8f30SChristoph Hellwig break; 2881c8fe8f30SChristoph Hellwig case S_IFBLK: 2882c8fe8f30SChristoph Hellwig case S_IFCHR: 2883a2982cc9SEric W. Biederman if (!may_open_dev(path)) 28841da177e4SLinus Torvalds return -EACCES; 2885c8fe8f30SChristoph Hellwig /*FALLTHRU*/ 2886c8fe8f30SChristoph Hellwig case S_IFIFO: 2887c8fe8f30SChristoph Hellwig case S_IFSOCK: 28881da177e4SLinus Torvalds flag &= ~O_TRUNC; 2889c8fe8f30SChristoph Hellwig break; 28904a3fd211SDave Hansen } 2891b41572e9SDave Hansen 289262fb4a15SAl Viro error = inode_permission(inode, MAY_OPEN | acc_mode); 2893b41572e9SDave Hansen if (error) 2894b41572e9SDave Hansen return error; 28956146f0d5SMimi Zohar 28961da177e4SLinus Torvalds /* 28971da177e4SLinus Torvalds * An append-only file must be opened in append mode for writing. 28981da177e4SLinus Torvalds */ 28991da177e4SLinus Torvalds if (IS_APPEND(inode)) { 29008737c930SAl Viro if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) 29017715b521SAl Viro return -EPERM; 29021da177e4SLinus Torvalds if (flag & O_TRUNC) 29037715b521SAl Viro return -EPERM; 29041da177e4SLinus Torvalds } 29051da177e4SLinus Torvalds 29061da177e4SLinus Torvalds /* O_NOATIME can only be set by the owner or superuser */ 29072e149670SSerge E. Hallyn if (flag & O_NOATIME && !inode_owner_or_capable(inode)) 29087715b521SAl Viro return -EPERM; 29091da177e4SLinus Torvalds 2910f3c7691eSJ. Bruce Fields return 0; 29117715b521SAl Viro } 29127715b521SAl Viro 2913e1181ee6SJeff Layton static int handle_truncate(struct file *filp) 29147715b521SAl Viro { 2915f0bb5aafSAl Viro const struct path *path = &filp->f_path; 29167715b521SAl Viro struct inode *inode = path->dentry->d_inode; 29177715b521SAl Viro int error = get_write_access(inode); 29181da177e4SLinus Torvalds if (error) 29197715b521SAl Viro return error; 29201da177e4SLinus Torvalds /* 29211da177e4SLinus Torvalds * Refuse to truncate files with mandatory locks held on them. 29221da177e4SLinus Torvalds */ 2923d7a06983SJeff Layton error = locks_verify_locked(filp); 2924be6d3e56SKentaro Takeda if (!error) 2925ea0d3ab2STetsuo Handa error = security_path_truncate(path); 29261da177e4SLinus Torvalds if (!error) { 29277715b521SAl Viro error = do_truncate(path->dentry, 0, 2928d139d7ffSMiklos Szeredi ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, 2929e1181ee6SJeff Layton filp); 29301da177e4SLinus Torvalds } 29311da177e4SLinus Torvalds put_write_access(inode); 2932acd0c935SMimi Zohar return error; 29331da177e4SLinus Torvalds } 29341da177e4SLinus Torvalds 2935d57999e1SDave Hansen static inline int open_to_namei_flags(int flag) 2936d57999e1SDave Hansen { 29378a5e929dSAl Viro if ((flag & O_ACCMODE) == 3) 29388a5e929dSAl Viro flag--; 2939d57999e1SDave Hansen return flag; 2940d57999e1SDave Hansen } 2941d57999e1SDave Hansen 2942d3607752SAl Viro static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode) 2943d18e9008SMiklos Szeredi { 2944d18e9008SMiklos Szeredi int error = security_path_mknod(dir, dentry, mode, 0); 2945d18e9008SMiklos Szeredi if (error) 2946d18e9008SMiklos Szeredi return error; 2947d18e9008SMiklos Szeredi 2948d18e9008SMiklos Szeredi error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC); 2949d18e9008SMiklos Szeredi if (error) 2950d18e9008SMiklos Szeredi return error; 2951d18e9008SMiklos Szeredi 2952d18e9008SMiklos Szeredi return security_inode_create(dir->dentry->d_inode, dentry, mode); 2953d18e9008SMiklos Szeredi } 2954d18e9008SMiklos Szeredi 29551acf0af9SDavid Howells /* 29561acf0af9SDavid Howells * Attempt to atomically look up, create and open a file from a negative 29571acf0af9SDavid Howells * dentry. 29581acf0af9SDavid Howells * 29591acf0af9SDavid Howells * Returns 0 if successful. The file will have been created and attached to 29601acf0af9SDavid Howells * @file by the filesystem calling finish_open(). 29611acf0af9SDavid Howells * 29621acf0af9SDavid Howells * Returns 1 if the file was looked up only or didn't need creating. The 29631acf0af9SDavid Howells * caller will need to perform the open themselves. @path will have been 29641acf0af9SDavid Howells * updated to point to the new dentry. This may be negative. 29651acf0af9SDavid Howells * 29661acf0af9SDavid Howells * Returns an error code otherwise. 29671acf0af9SDavid Howells */ 29682675a4ebSAl Viro static int atomic_open(struct nameidata *nd, struct dentry *dentry, 296930d90494SAl Viro struct path *path, struct file *file, 2970015c3bbcSMiklos Szeredi const struct open_flags *op, 29711643b43fSAl Viro int open_flag, umode_t mode, 297247237687SAl Viro int *opened) 2973d18e9008SMiklos Szeredi { 2974d18e9008SMiklos Szeredi struct dentry *const DENTRY_NOT_SET = (void *) -1UL; 2975d18e9008SMiklos Szeredi struct inode *dir = nd->path.dentry->d_inode; 2976d18e9008SMiklos Szeredi int error; 2977d18e9008SMiklos Szeredi 2978384f26e2SAl Viro if (!(~open_flag & (O_EXCL | O_CREAT))) /* both O_EXCL and O_CREAT */ 2979d18e9008SMiklos Szeredi open_flag &= ~O_TRUNC; 2980d18e9008SMiklos Szeredi 2981d18e9008SMiklos Szeredi if (nd->flags & LOOKUP_DIRECTORY) 2982d18e9008SMiklos Szeredi open_flag |= O_DIRECTORY; 2983d18e9008SMiklos Szeredi 298430d90494SAl Viro file->f_path.dentry = DENTRY_NOT_SET; 298530d90494SAl Viro file->f_path.mnt = nd->path.mnt; 29860fb1ea09SAl Viro error = dir->i_op->atomic_open(dir, dentry, file, 29870fb1ea09SAl Viro open_to_namei_flags(open_flag), 29880fb1ea09SAl Viro mode, opened); 29896fbd0714SAl Viro d_lookup_done(dentry); 2990384f26e2SAl Viro if (!error) { 2991d18e9008SMiklos Szeredi /* 2992384f26e2SAl Viro * We didn't have the inode before the open, so check open 2993384f26e2SAl Viro * permission here. 2994d18e9008SMiklos Szeredi */ 2995384f26e2SAl Viro int acc_mode = op->acc_mode; 299603da633aSAl Viro if (*opened & FILE_CREATED) { 299703da633aSAl Viro WARN_ON(!(open_flag & O_CREAT)); 299803da633aSAl Viro fsnotify_create(dir, dentry); 299962fb4a15SAl Viro acc_mode = 0; 300003da633aSAl Viro } 30012675a4ebSAl Viro error = may_open(&file->f_path, acc_mode, open_flag); 3002384f26e2SAl Viro if (WARN_ON(error > 0)) 3003384f26e2SAl Viro error = -EINVAL; 3004384f26e2SAl Viro } else if (error > 0) { 3005d18e9008SMiklos Szeredi if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { 3006d18e9008SMiklos Szeredi error = -EIO; 3007384f26e2SAl Viro } else { 3008d18e9008SMiklos Szeredi if (file->f_path.dentry) { 3009d18e9008SMiklos Szeredi dput(dentry); 3010d18e9008SMiklos Szeredi dentry = file->f_path.dentry; 301110c64ceaSAl Viro } 3012d18e9008SMiklos Szeredi if (*opened & FILE_CREATED) 3013d18e9008SMiklos Szeredi fsnotify_create(dir, dentry); 3014a01e718fSAl Viro if (unlikely(d_is_negative(dentry))) { 3015a01e718fSAl Viro error = -ENOENT; 3016a01e718fSAl Viro } else { 3017d18e9008SMiklos Szeredi path->dentry = dentry; 3018d18e9008SMiklos Szeredi path->mnt = nd->path.mnt; 30192675a4ebSAl Viro return 1; 3020d18e9008SMiklos Szeredi } 3021d18e9008SMiklos Szeredi } 3022a01e718fSAl Viro } 3023d18e9008SMiklos Szeredi dput(dentry); 3024d18e9008SMiklos Szeredi return error; 3025d18e9008SMiklos Szeredi } 3026d18e9008SMiklos Szeredi 302731e6b01fSNick Piggin /* 30281acf0af9SDavid Howells * Look up and maybe create and open the last component. 3029d58ffd35SMiklos Szeredi * 3030d58ffd35SMiklos Szeredi * Must be called with i_mutex held on parent. 3031d58ffd35SMiklos Szeredi * 30321acf0af9SDavid Howells * Returns 0 if the file was successfully atomically created (if necessary) and 30331acf0af9SDavid Howells * opened. In this case the file will be returned attached to @file. 30341acf0af9SDavid Howells * 30351acf0af9SDavid Howells * Returns 1 if the file was not completely opened at this time, though lookups 30361acf0af9SDavid Howells * and creations will have been performed and the dentry returned in @path will 30371acf0af9SDavid Howells * be positive upon return if O_CREAT was specified. If O_CREAT wasn't 30381acf0af9SDavid Howells * specified then a negative dentry may be returned. 30391acf0af9SDavid Howells * 30401acf0af9SDavid Howells * An error code is returned otherwise. 30411acf0af9SDavid Howells * 30421acf0af9SDavid Howells * FILE_CREATE will be set in @*opened if the dentry was created and will be 30431acf0af9SDavid Howells * cleared otherwise prior to returning. 3044d58ffd35SMiklos Szeredi */ 30452675a4ebSAl Viro static int lookup_open(struct nameidata *nd, struct path *path, 304630d90494SAl Viro struct file *file, 3047d58ffd35SMiklos Szeredi const struct open_flags *op, 304864894cf8SAl Viro bool got_write, int *opened) 3049d58ffd35SMiklos Szeredi { 3050d58ffd35SMiklos Szeredi struct dentry *dir = nd->path.dentry; 305154ef4872SMiklos Szeredi struct inode *dir_inode = dir->d_inode; 30521643b43fSAl Viro int open_flag = op->open_flag; 3053d58ffd35SMiklos Szeredi struct dentry *dentry; 30541643b43fSAl Viro int error, create_error = 0; 30551643b43fSAl Viro umode_t mode = op->mode; 30566fbd0714SAl Viro DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 3057d58ffd35SMiklos Szeredi 3058ce8644fcSAl Viro if (unlikely(IS_DEADDIR(dir_inode))) 3059ce8644fcSAl Viro return -ENOENT; 3060d58ffd35SMiklos Szeredi 306147237687SAl Viro *opened &= ~FILE_CREATED; 30626fbd0714SAl Viro dentry = d_lookup(dir, &nd->last); 30636fbd0714SAl Viro for (;;) { 30646fbd0714SAl Viro if (!dentry) { 30656fbd0714SAl Viro dentry = d_alloc_parallel(dir, &nd->last, &wq); 3066d58ffd35SMiklos Szeredi if (IS_ERR(dentry)) 30672675a4ebSAl Viro return PTR_ERR(dentry); 30686fbd0714SAl Viro } 30696fbd0714SAl Viro if (d_in_lookup(dentry)) 30706fbd0714SAl Viro break; 3071d58ffd35SMiklos Szeredi 30726fbd0714SAl Viro if (!(dentry->d_flags & DCACHE_OP_REVALIDATE)) 30736fbd0714SAl Viro break; 30746fbd0714SAl Viro 30756fbd0714SAl Viro error = d_revalidate(dentry, nd->flags); 30766fbd0714SAl Viro if (likely(error > 0)) 30776fbd0714SAl Viro break; 30786fbd0714SAl Viro if (error) 30796fbd0714SAl Viro goto out_dput; 30806fbd0714SAl Viro d_invalidate(dentry); 30816fbd0714SAl Viro dput(dentry); 30826fbd0714SAl Viro dentry = NULL; 30836fbd0714SAl Viro } 30846fbd0714SAl Viro if (dentry->d_inode) { 3085d18e9008SMiklos Szeredi /* Cached positive dentry: will open in f_op->open */ 3086d18e9008SMiklos Szeredi goto out_no_open; 30876c51e513SAl Viro } 3088d18e9008SMiklos Szeredi 30891643b43fSAl Viro /* 30901643b43fSAl Viro * Checking write permission is tricky, bacuse we don't know if we are 30911643b43fSAl Viro * going to actually need it: O_CREAT opens should work as long as the 30921643b43fSAl Viro * file exists. But checking existence breaks atomicity. The trick is 30931643b43fSAl Viro * to check access and if not granted clear O_CREAT from the flags. 30941643b43fSAl Viro * 30951643b43fSAl Viro * Another problem is returing the "right" error value (e.g. for an 30961643b43fSAl Viro * O_EXCL open we want to return EEXIST not EROFS). 30971643b43fSAl Viro */ 30981643b43fSAl Viro if (open_flag & O_CREAT) { 30991643b43fSAl Viro if (!IS_POSIXACL(dir->d_inode)) 31001643b43fSAl Viro mode &= ~current_umask(); 31011643b43fSAl Viro if (unlikely(!got_write)) { 31021643b43fSAl Viro create_error = -EROFS; 31031643b43fSAl Viro open_flag &= ~O_CREAT; 31041643b43fSAl Viro if (open_flag & (O_EXCL | O_TRUNC)) 31051643b43fSAl Viro goto no_open; 31061643b43fSAl Viro /* No side effects, safe to clear O_CREAT */ 31071643b43fSAl Viro } else { 31081643b43fSAl Viro create_error = may_o_create(&nd->path, dentry, mode); 31091643b43fSAl Viro if (create_error) { 31101643b43fSAl Viro open_flag &= ~O_CREAT; 31111643b43fSAl Viro if (open_flag & O_EXCL) 31121643b43fSAl Viro goto no_open; 31131643b43fSAl Viro } 31141643b43fSAl Viro } 31151643b43fSAl Viro } else if ((open_flag & (O_TRUNC|O_WRONLY|O_RDWR)) && 31161643b43fSAl Viro unlikely(!got_write)) { 31171643b43fSAl Viro /* 31181643b43fSAl Viro * No O_CREATE -> atomicity not a requirement -> fall 31191643b43fSAl Viro * back to lookup + open 31201643b43fSAl Viro */ 31211643b43fSAl Viro goto no_open; 3122d18e9008SMiklos Szeredi } 3123d18e9008SMiklos Szeredi 31241643b43fSAl Viro if (dir_inode->i_op->atomic_open) { 31251643b43fSAl Viro error = atomic_open(nd, dentry, path, file, op, open_flag, 31261643b43fSAl Viro mode, opened); 31271643b43fSAl Viro if (unlikely(error == -ENOENT) && create_error) 31281643b43fSAl Viro error = create_error; 31291643b43fSAl Viro return error; 31301643b43fSAl Viro } 313154ef4872SMiklos Szeredi 31321643b43fSAl Viro no_open: 31336fbd0714SAl Viro if (d_in_lookup(dentry)) { 313412fa5e24SAl Viro struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry, 313512fa5e24SAl Viro nd->flags); 31366fbd0714SAl Viro d_lookup_done(dentry); 313712fa5e24SAl Viro if (unlikely(res)) { 313812fa5e24SAl Viro if (IS_ERR(res)) { 313912fa5e24SAl Viro error = PTR_ERR(res); 314012fa5e24SAl Viro goto out_dput; 314112fa5e24SAl Viro } 314212fa5e24SAl Viro dput(dentry); 314312fa5e24SAl Viro dentry = res; 314412fa5e24SAl Viro } 314554ef4872SMiklos Szeredi } 314654ef4872SMiklos Szeredi 3147d58ffd35SMiklos Szeredi /* Negative dentry, just create the file */ 31481643b43fSAl Viro if (!dentry->d_inode && (open_flag & O_CREAT)) { 314947237687SAl Viro *opened |= FILE_CREATED; 3150ce8644fcSAl Viro audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE); 3151ce8644fcSAl Viro if (!dir_inode->i_op->create) { 3152ce8644fcSAl Viro error = -EACCES; 3153d58ffd35SMiklos Szeredi goto out_dput; 315477d660a8SMiklos Szeredi } 3155ce8644fcSAl Viro error = dir_inode->i_op->create(dir_inode, dentry, mode, 31561643b43fSAl Viro open_flag & O_EXCL); 3157d58ffd35SMiklos Szeredi if (error) 3158d58ffd35SMiklos Szeredi goto out_dput; 3159ce8644fcSAl Viro fsnotify_create(dir_inode, dentry); 3160d58ffd35SMiklos Szeredi } 31611643b43fSAl Viro if (unlikely(create_error) && !dentry->d_inode) { 31621643b43fSAl Viro error = create_error; 3163d58ffd35SMiklos Szeredi goto out_dput; 3164d58ffd35SMiklos Szeredi } 3165d18e9008SMiklos Szeredi out_no_open: 3166d58ffd35SMiklos Szeredi path->dentry = dentry; 3167d58ffd35SMiklos Szeredi path->mnt = nd->path.mnt; 31682675a4ebSAl Viro return 1; 3169d58ffd35SMiklos Szeredi 3170d58ffd35SMiklos Szeredi out_dput: 3171d58ffd35SMiklos Szeredi dput(dentry); 31722675a4ebSAl Viro return error; 3173d58ffd35SMiklos Szeredi } 3174d58ffd35SMiklos Szeredi 3175d58ffd35SMiklos Szeredi /* 3176fe2d35ffSAl Viro * Handle the last step of open() 317731e6b01fSNick Piggin */ 3178896475d5SAl Viro static int do_last(struct nameidata *nd, 317930d90494SAl Viro struct file *file, const struct open_flags *op, 318076ae2a5aSAl Viro int *opened) 3181fb1cc555SAl Viro { 3182a1e28038SAl Viro struct dentry *dir = nd->path.dentry; 3183ca344a89SAl Viro int open_flag = op->open_flag; 318477d660a8SMiklos Szeredi bool will_truncate = (open_flag & O_TRUNC) != 0; 318564894cf8SAl Viro bool got_write = false; 3186bcda7652SAl Viro int acc_mode = op->acc_mode; 3187254cf582SAl Viro unsigned seq; 3188a1eb3315SMiklos Szeredi struct inode *inode; 3189896475d5SAl Viro struct path path; 319016c2cd71SAl Viro int error; 3191fb1cc555SAl Viro 3192c3e380b0SAl Viro nd->flags &= ~LOOKUP_PARENT; 3193c3e380b0SAl Viro nd->flags |= op->intent; 3194c3e380b0SAl Viro 3195bc77daa7SAl Viro if (nd->last_type != LAST_NORM) { 3196fe2d35ffSAl Viro error = handle_dots(nd, nd->last_type); 3197deb106c6SAl Viro if (unlikely(error)) 31982675a4ebSAl Viro return error; 3199e83db167SMiklos Szeredi goto finish_open; 32001f36f774SAl Viro } 3201a2c36b45SAl Viro 3202ca344a89SAl Viro if (!(open_flag & O_CREAT)) { 3203fe2d35ffSAl Viro if (nd->last.name[nd->last.len]) 3204fe2d35ffSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 3205fe2d35ffSAl Viro /* we _can_ be in RCU mode here */ 3206254cf582SAl Viro error = lookup_fast(nd, &path, &inode, &seq); 3207e9742b53SAl Viro if (likely(error > 0)) 320871574865SMiklos Szeredi goto finish_lookup; 320971574865SMiklos Szeredi 3210ce57dfc1SAl Viro if (error < 0) 3211deb106c6SAl Viro return error; 3212a1eb3315SMiklos Szeredi 321337d7fffcSMiklos Szeredi BUG_ON(nd->inode != dir->d_inode); 32146583fe22SAl Viro BUG_ON(nd->flags & LOOKUP_RCU); 3215b6183df7SMiklos Szeredi } else { 3216fe2d35ffSAl Viro /* create side of things */ 3217a3fbbde7SAl Viro /* 3218b6183df7SMiklos Szeredi * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED 3219b6183df7SMiklos Szeredi * has been cleared when we got to the last component we are 3220b6183df7SMiklos Szeredi * about to look up 3221a3fbbde7SAl Viro */ 32229f1fafeeSAl Viro error = complete_walk(nd); 3223e8bb73dfSAl Viro if (error) 32242675a4ebSAl Viro return error; 3225fe2d35ffSAl Viro 322676ae2a5aSAl Viro audit_inode(nd->name, dir, LOOKUP_PARENT); 32271f36f774SAl Viro /* trailing slashes? */ 3228deb106c6SAl Viro if (unlikely(nd->last.name[nd->last.len])) 3229deb106c6SAl Viro return -EISDIR; 3230b6183df7SMiklos Szeredi } 32311f36f774SAl Viro 32329cf843e3SAl Viro if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { 323364894cf8SAl Viro error = mnt_want_write(nd->path.mnt); 323464894cf8SAl Viro if (!error) 323564894cf8SAl Viro got_write = true; 323664894cf8SAl Viro /* 323764894cf8SAl Viro * do _not_ fail yet - we might not need that or fail with 323864894cf8SAl Viro * a different error; let lookup_open() decide; we'll be 323964894cf8SAl Viro * dropping this one anyway. 324064894cf8SAl Viro */ 324164894cf8SAl Viro } 32429cf843e3SAl Viro if (open_flag & O_CREAT) 32435955102cSAl Viro inode_lock(dir->d_inode); 32449cf843e3SAl Viro else 32459cf843e3SAl Viro inode_lock_shared(dir->d_inode); 3246896475d5SAl Viro error = lookup_open(nd, &path, file, op, got_write, opened); 32479cf843e3SAl Viro if (open_flag & O_CREAT) 32485955102cSAl Viro inode_unlock(dir->d_inode); 32499cf843e3SAl Viro else 32509cf843e3SAl Viro inode_unlock_shared(dir->d_inode); 3251fb1cc555SAl Viro 32522675a4ebSAl Viro if (error <= 0) { 32532675a4ebSAl Viro if (error) 3254d58ffd35SMiklos Szeredi goto out; 32556c0d46c4SAl Viro 325647237687SAl Viro if ((*opened & FILE_CREATED) || 3257496ad9aaSAl Viro !S_ISREG(file_inode(file)->i_mode)) 325877d660a8SMiklos Szeredi will_truncate = false; 3259d18e9008SMiklos Szeredi 326076ae2a5aSAl Viro audit_inode(nd->name, file->f_path.dentry, 0); 3261d18e9008SMiklos Szeredi goto opened; 3262d18e9008SMiklos Szeredi } 3263d18e9008SMiklos Szeredi 326447237687SAl Viro if (*opened & FILE_CREATED) { 32659b44f1b3SAl Viro /* Don't check for write permission, don't truncate */ 3266ca344a89SAl Viro open_flag &= ~O_TRUNC; 326777d660a8SMiklos Szeredi will_truncate = false; 326862fb4a15SAl Viro acc_mode = 0; 3269896475d5SAl Viro path_to_nameidata(&path, nd); 3270e83db167SMiklos Szeredi goto finish_open_created; 3271fb1cc555SAl Viro } 3272fb1cc555SAl Viro 3273fb1cc555SAl Viro /* 3274d18e9008SMiklos Szeredi * If atomic_open() acquired write access it is dropped now due to 3275d18e9008SMiklos Szeredi * possible mount and symlink following (this might be optimized away if 3276d18e9008SMiklos Szeredi * necessary...) 3277d18e9008SMiklos Szeredi */ 327864894cf8SAl Viro if (got_write) { 3279d18e9008SMiklos Szeredi mnt_drop_write(nd->path.mnt); 328064894cf8SAl Viro got_write = false; 3281d18e9008SMiklos Szeredi } 3282d18e9008SMiklos Szeredi 3283e6ec03a2SAl Viro error = follow_managed(&path, nd); 3284e6ec03a2SAl Viro if (unlikely(error < 0)) 3285e6ec03a2SAl Viro return error; 3286e6ec03a2SAl Viro 32876583fe22SAl Viro if (unlikely(d_is_negative(path.dentry))) { 32886583fe22SAl Viro path_to_nameidata(&path, nd); 32896583fe22SAl Viro return -ENOENT; 32906583fe22SAl Viro } 32916583fe22SAl Viro 32926583fe22SAl Viro /* 32936583fe22SAl Viro * create/update audit record if it already exists. 32946583fe22SAl Viro */ 32956583fe22SAl Viro audit_inode(nd->name, path.dentry, 0); 32966583fe22SAl Viro 3297deb106c6SAl Viro if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) { 3298deb106c6SAl Viro path_to_nameidata(&path, nd); 3299deb106c6SAl Viro return -EEXIST; 3300deb106c6SAl Viro } 3301fb1cc555SAl Viro 3302254cf582SAl Viro seq = 0; /* out of RCU mode, so the value doesn't matter */ 3303d4565649SAl Viro inode = d_backing_inode(path.dentry); 3304766c4cbfSAl Viro finish_lookup: 33058f64fb1cSAl Viro error = step_into(nd, &path, 0, inode, seq); 3306deb106c6SAl Viro if (unlikely(error)) 3307d63ff28fSAl Viro return error; 3308bc77daa7SAl Viro finish_open: 33098f64fb1cSAl Viro /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ 3310a3fbbde7SAl Viro error = complete_walk(nd); 3311fac7d191SAl Viro if (error) 33122675a4ebSAl Viro return error; 331376ae2a5aSAl Viro audit_inode(nd->name, nd->path.dentry, 0); 3314fb1cc555SAl Viro error = -EISDIR; 331544b1d530SMiklos Szeredi if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry)) 33162675a4ebSAl Viro goto out; 3317af2f5542SMiklos Szeredi error = -ENOTDIR; 331844b1d530SMiklos Szeredi if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry)) 33192675a4ebSAl Viro goto out; 33204bbcbd3bSDavid Howells if (!d_is_reg(nd->path.dentry)) 332177d660a8SMiklos Szeredi will_truncate = false; 33226c0d46c4SAl Viro 33230f9d1a10SAl Viro if (will_truncate) { 33240f9d1a10SAl Viro error = mnt_want_write(nd->path.mnt); 33250f9d1a10SAl Viro if (error) 33262675a4ebSAl Viro goto out; 332764894cf8SAl Viro got_write = true; 33280f9d1a10SAl Viro } 3329e83db167SMiklos Szeredi finish_open_created: 3330bcda7652SAl Viro error = may_open(&nd->path, acc_mode, open_flag); 3331ca344a89SAl Viro if (error) 33322675a4ebSAl Viro goto out; 33334aa7c634SMiklos Szeredi BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */ 33344aa7c634SMiklos Szeredi error = vfs_open(&nd->path, file, current_cred()); 3335fac7d191SAl Viro if (error) 3336015c3bbcSMiklos Szeredi goto out; 3337fac7d191SAl Viro *opened |= FILE_OPENED; 3338a8277b9bSMiklos Szeredi opened: 33392675a4ebSAl Viro error = open_check_o_direct(file); 3340fe9ec829SAl Viro if (!error) 33413034a146SDmitry Kasatkin error = ima_file_check(file, op->acc_mode, *opened); 3342fe9ec829SAl Viro if (!error && will_truncate) 33432675a4ebSAl Viro error = handle_truncate(file); 3344ca344a89SAl Viro out: 3345fe9ec829SAl Viro if (unlikely(error) && (*opened & FILE_OPENED)) 3346fe9ec829SAl Viro fput(file); 3347c80567c8SAl Viro if (unlikely(error > 0)) { 3348c80567c8SAl Viro WARN_ON(1); 3349c80567c8SAl Viro error = -EINVAL; 3350c80567c8SAl Viro } 335164894cf8SAl Viro if (got_write) 33520f9d1a10SAl Viro mnt_drop_write(nd->path.mnt); 33532675a4ebSAl Viro return error; 3354fb1cc555SAl Viro } 3355fb1cc555SAl Viro 3356c8a53ee5SAl Viro static int do_tmpfile(struct nameidata *nd, unsigned flags, 335760545d0dSAl Viro const struct open_flags *op, 335860545d0dSAl Viro struct file *file, int *opened) 335960545d0dSAl Viro { 336060545d0dSAl Viro static const struct qstr name = QSTR_INIT("/", 1); 3361625b6d10SAl Viro struct dentry *child; 336260545d0dSAl Viro struct inode *dir; 3363625b6d10SAl Viro struct path path; 3364c8a53ee5SAl Viro int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path); 336560545d0dSAl Viro if (unlikely(error)) 336660545d0dSAl Viro return error; 3367625b6d10SAl Viro error = mnt_want_write(path.mnt); 336860545d0dSAl Viro if (unlikely(error)) 336960545d0dSAl Viro goto out; 3370625b6d10SAl Viro dir = path.dentry->d_inode; 337160545d0dSAl Viro /* we want directory to be writable */ 3372625b6d10SAl Viro error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 337360545d0dSAl Viro if (error) 337460545d0dSAl Viro goto out2; 337560545d0dSAl Viro if (!dir->i_op->tmpfile) { 337660545d0dSAl Viro error = -EOPNOTSUPP; 337760545d0dSAl Viro goto out2; 337860545d0dSAl Viro } 3379625b6d10SAl Viro child = d_alloc(path.dentry, &name); 338060545d0dSAl Viro if (unlikely(!child)) { 338160545d0dSAl Viro error = -ENOMEM; 338260545d0dSAl Viro goto out2; 338360545d0dSAl Viro } 3384625b6d10SAl Viro dput(path.dentry); 3385625b6d10SAl Viro path.dentry = child; 3386625b6d10SAl Viro error = dir->i_op->tmpfile(dir, child, op->mode); 338760545d0dSAl Viro if (error) 338860545d0dSAl Viro goto out2; 3389c8a53ee5SAl Viro audit_inode(nd->name, child, 0); 339069a91c23SEric Rannaud /* Don't check for other permissions, the inode was just created */ 339162fb4a15SAl Viro error = may_open(&path, 0, op->open_flag); 339260545d0dSAl Viro if (error) 339360545d0dSAl Viro goto out2; 3394625b6d10SAl Viro file->f_path.mnt = path.mnt; 3395625b6d10SAl Viro error = finish_open(file, child, NULL, opened); 339660545d0dSAl Viro if (error) 339760545d0dSAl Viro goto out2; 339860545d0dSAl Viro error = open_check_o_direct(file); 3399f4e0c30cSAl Viro if (error) { 340060545d0dSAl Viro fput(file); 3401f4e0c30cSAl Viro } else if (!(op->open_flag & O_EXCL)) { 3402f4e0c30cSAl Viro struct inode *inode = file_inode(file); 3403f4e0c30cSAl Viro spin_lock(&inode->i_lock); 3404f4e0c30cSAl Viro inode->i_state |= I_LINKABLE; 3405f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 3406f4e0c30cSAl Viro } 340760545d0dSAl Viro out2: 3408625b6d10SAl Viro mnt_drop_write(path.mnt); 340960545d0dSAl Viro out: 3410625b6d10SAl Viro path_put(&path); 341160545d0dSAl Viro return error; 341260545d0dSAl Viro } 341360545d0dSAl Viro 34146ac08709SAl Viro static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file) 34156ac08709SAl Viro { 34166ac08709SAl Viro struct path path; 34176ac08709SAl Viro int error = path_lookupat(nd, flags, &path); 34186ac08709SAl Viro if (!error) { 34196ac08709SAl Viro audit_inode(nd->name, path.dentry, 0); 34206ac08709SAl Viro error = vfs_open(&path, file, current_cred()); 34216ac08709SAl Viro path_put(&path); 34226ac08709SAl Viro } 34236ac08709SAl Viro return error; 34246ac08709SAl Viro } 34256ac08709SAl Viro 3426c8a53ee5SAl Viro static struct file *path_openat(struct nameidata *nd, 3427c8a53ee5SAl Viro const struct open_flags *op, unsigned flags) 34281da177e4SLinus Torvalds { 3429368ee9baSAl Viro const char *s; 343030d90494SAl Viro struct file *file; 343147237687SAl Viro int opened = 0; 343213aab428SAl Viro int error; 343331e6b01fSNick Piggin 343430d90494SAl Viro file = get_empty_filp(); 34351afc99beSAl Viro if (IS_ERR(file)) 34361afc99beSAl Viro return file; 343731e6b01fSNick Piggin 343830d90494SAl Viro file->f_flags = op->open_flag; 343931e6b01fSNick Piggin 3440bb458c64SAl Viro if (unlikely(file->f_flags & __O_TMPFILE)) { 3441c8a53ee5SAl Viro error = do_tmpfile(nd, flags, op, file, &opened); 3442f15133dfSAl Viro goto out2; 344360545d0dSAl Viro } 344460545d0dSAl Viro 34456ac08709SAl Viro if (unlikely(file->f_flags & O_PATH)) { 34466ac08709SAl Viro error = do_o_path(nd, flags, file); 34476ac08709SAl Viro if (!error) 34486ac08709SAl Viro opened |= FILE_OPENED; 34496ac08709SAl Viro goto out2; 34506ac08709SAl Viro } 34516ac08709SAl Viro 3452c8a53ee5SAl Viro s = path_init(nd, flags); 3453368ee9baSAl Viro if (IS_ERR(s)) { 3454368ee9baSAl Viro put_filp(file); 3455368ee9baSAl Viro return ERR_CAST(s); 3456368ee9baSAl Viro } 34573bdba28bSAl Viro while (!(error = link_path_walk(s, nd)) && 345876ae2a5aSAl Viro (error = do_last(nd, file, op, &opened)) > 0) { 345973d049a4SAl Viro nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); 34603bdba28bSAl Viro s = trailing_symlink(nd); 34613bdba28bSAl Viro if (IS_ERR(s)) { 34623bdba28bSAl Viro error = PTR_ERR(s); 34632675a4ebSAl Viro break; 3464806b681cSAl Viro } 34653bdba28bSAl Viro } 3466deb106c6SAl Viro terminate_walk(nd); 3467f15133dfSAl Viro out2: 34682675a4ebSAl Viro if (!(opened & FILE_OPENED)) { 34692675a4ebSAl Viro BUG_ON(!error); 347030d90494SAl Viro put_filp(file); 3471015c3bbcSMiklos Szeredi } 34722675a4ebSAl Viro if (unlikely(error)) { 34732675a4ebSAl Viro if (error == -EOPENSTALE) { 34742675a4ebSAl Viro if (flags & LOOKUP_RCU) 34752675a4ebSAl Viro error = -ECHILD; 34762675a4ebSAl Viro else 34772675a4ebSAl Viro error = -ESTALE; 34782675a4ebSAl Viro } 34792675a4ebSAl Viro file = ERR_PTR(error); 34802675a4ebSAl Viro } 34812675a4ebSAl Viro return file; 3482de459215SKirill Korotaev } 34831da177e4SLinus Torvalds 3484669abf4eSJeff Layton struct file *do_filp_open(int dfd, struct filename *pathname, 3485f9652e10SAl Viro const struct open_flags *op) 348613aab428SAl Viro { 34879883d185SAl Viro struct nameidata nd; 3488f9652e10SAl Viro int flags = op->lookup_flags; 348913aab428SAl Viro struct file *filp; 349013aab428SAl Viro 34919883d185SAl Viro set_nameidata(&nd, dfd, pathname); 3492c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_RCU); 349313aab428SAl Viro if (unlikely(filp == ERR_PTR(-ECHILD))) 3494c8a53ee5SAl Viro filp = path_openat(&nd, op, flags); 349513aab428SAl Viro if (unlikely(filp == ERR_PTR(-ESTALE))) 3496c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_REVAL); 34979883d185SAl Viro restore_nameidata(); 349813aab428SAl Viro return filp; 349913aab428SAl Viro } 350013aab428SAl Viro 350173d049a4SAl Viro struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, 3502f9652e10SAl Viro const char *name, const struct open_flags *op) 350373d049a4SAl Viro { 35049883d185SAl Viro struct nameidata nd; 350573d049a4SAl Viro struct file *file; 350651689104SPaul Moore struct filename *filename; 3507f9652e10SAl Viro int flags = op->lookup_flags | LOOKUP_ROOT; 350873d049a4SAl Viro 350973d049a4SAl Viro nd.root.mnt = mnt; 351073d049a4SAl Viro nd.root.dentry = dentry; 351173d049a4SAl Viro 3512b18825a7SDavid Howells if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN) 351373d049a4SAl Viro return ERR_PTR(-ELOOP); 351473d049a4SAl Viro 351551689104SPaul Moore filename = getname_kernel(name); 3516a1c83681SViresh Kumar if (IS_ERR(filename)) 351751689104SPaul Moore return ERR_CAST(filename); 351851689104SPaul Moore 35199883d185SAl Viro set_nameidata(&nd, -1, filename); 3520c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_RCU); 352173d049a4SAl Viro if (unlikely(file == ERR_PTR(-ECHILD))) 3522c8a53ee5SAl Viro file = path_openat(&nd, op, flags); 352373d049a4SAl Viro if (unlikely(file == ERR_PTR(-ESTALE))) 3524c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_REVAL); 35259883d185SAl Viro restore_nameidata(); 352651689104SPaul Moore putname(filename); 352773d049a4SAl Viro return file; 352873d049a4SAl Viro } 352973d049a4SAl Viro 3530fa14a0b8SAl Viro static struct dentry *filename_create(int dfd, struct filename *name, 35311ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 35321da177e4SLinus Torvalds { 3533c663e5d8SChristoph Hellwig struct dentry *dentry = ERR_PTR(-EEXIST); 3534391172c4SAl Viro struct qstr last; 3535391172c4SAl Viro int type; 3536c30dabfeSJan Kara int err2; 35371ac12b4bSJeff Layton int error; 35381ac12b4bSJeff Layton bool is_dir = (lookup_flags & LOOKUP_DIRECTORY); 35391ac12b4bSJeff Layton 35401ac12b4bSJeff Layton /* 35411ac12b4bSJeff Layton * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any 35421ac12b4bSJeff Layton * other flags passed in are ignored! 35431ac12b4bSJeff Layton */ 35441ac12b4bSJeff Layton lookup_flags &= LOOKUP_REVAL; 35451ac12b4bSJeff Layton 35465c31b6ceSAl Viro name = filename_parentat(dfd, name, lookup_flags, path, &last, &type); 35475c31b6ceSAl Viro if (IS_ERR(name)) 35485c31b6ceSAl Viro return ERR_CAST(name); 35491da177e4SLinus Torvalds 3550c663e5d8SChristoph Hellwig /* 3551c663e5d8SChristoph Hellwig * Yucky last component or no last component at all? 3552c663e5d8SChristoph Hellwig * (foo/., foo/.., /////) 3553c663e5d8SChristoph Hellwig */ 35545c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) 3555ed75e95dSAl Viro goto out; 3556c663e5d8SChristoph Hellwig 3557c30dabfeSJan Kara /* don't fail immediately if it's r/o, at least try to report other errors */ 3558391172c4SAl Viro err2 = mnt_want_write(path->mnt); 3559c663e5d8SChristoph Hellwig /* 3560c663e5d8SChristoph Hellwig * Do the final lookup. 3561c663e5d8SChristoph Hellwig */ 3562391172c4SAl Viro lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL; 35635955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 3564391172c4SAl Viro dentry = __lookup_hash(&last, path->dentry, lookup_flags); 35651da177e4SLinus Torvalds if (IS_ERR(dentry)) 3566a8104a9fSAl Viro goto unlock; 3567c663e5d8SChristoph Hellwig 3568a8104a9fSAl Viro error = -EEXIST; 3569b18825a7SDavid Howells if (d_is_positive(dentry)) 3570a8104a9fSAl Viro goto fail; 3571b18825a7SDavid Howells 3572c663e5d8SChristoph Hellwig /* 3573c663e5d8SChristoph Hellwig * Special case - lookup gave negative, but... we had foo/bar/ 3574c663e5d8SChristoph Hellwig * From the vfs_mknod() POV we just have a negative dentry - 3575c663e5d8SChristoph Hellwig * all is fine. Let's be bastards - you had / on the end, you've 3576c663e5d8SChristoph Hellwig * been asking for (non-existent) directory. -ENOENT for you. 3577c663e5d8SChristoph Hellwig */ 3578391172c4SAl Viro if (unlikely(!is_dir && last.name[last.len])) { 3579a8104a9fSAl Viro error = -ENOENT; 3580ed75e95dSAl Viro goto fail; 3581e9baf6e5SAl Viro } 3582c30dabfeSJan Kara if (unlikely(err2)) { 3583c30dabfeSJan Kara error = err2; 3584a8104a9fSAl Viro goto fail; 3585c30dabfeSJan Kara } 3586181c37b6SAl Viro putname(name); 3587e9baf6e5SAl Viro return dentry; 35881da177e4SLinus Torvalds fail: 3589a8104a9fSAl Viro dput(dentry); 3590a8104a9fSAl Viro dentry = ERR_PTR(error); 3591a8104a9fSAl Viro unlock: 35925955102cSAl Viro inode_unlock(path->dentry->d_inode); 3593c30dabfeSJan Kara if (!err2) 3594391172c4SAl Viro mnt_drop_write(path->mnt); 3595ed75e95dSAl Viro out: 3596391172c4SAl Viro path_put(path); 3597181c37b6SAl Viro putname(name); 3598ed75e95dSAl Viro return dentry; 3599dae6ad8fSAl Viro } 3600fa14a0b8SAl Viro 3601fa14a0b8SAl Viro struct dentry *kern_path_create(int dfd, const char *pathname, 3602fa14a0b8SAl Viro struct path *path, unsigned int lookup_flags) 3603fa14a0b8SAl Viro { 3604181c37b6SAl Viro return filename_create(dfd, getname_kernel(pathname), 3605181c37b6SAl Viro path, lookup_flags); 3606fa14a0b8SAl Viro } 3607dae6ad8fSAl Viro EXPORT_SYMBOL(kern_path_create); 3608dae6ad8fSAl Viro 3609921a1650SAl Viro void done_path_create(struct path *path, struct dentry *dentry) 3610921a1650SAl Viro { 3611921a1650SAl Viro dput(dentry); 36125955102cSAl Viro inode_unlock(path->dentry->d_inode); 3613a8104a9fSAl Viro mnt_drop_write(path->mnt); 3614921a1650SAl Viro path_put(path); 3615921a1650SAl Viro } 3616921a1650SAl Viro EXPORT_SYMBOL(done_path_create); 3617921a1650SAl Viro 3618520ae687SAl Viro inline struct dentry *user_path_create(int dfd, const char __user *pathname, 36191ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 3620dae6ad8fSAl Viro { 3621181c37b6SAl Viro return filename_create(dfd, getname(pathname), path, lookup_flags); 3622dae6ad8fSAl Viro } 3623dae6ad8fSAl Viro EXPORT_SYMBOL(user_path_create); 3624dae6ad8fSAl Viro 36251a67aafbSAl Viro int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) 36261da177e4SLinus Torvalds { 3627a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 36281da177e4SLinus Torvalds 36291da177e4SLinus Torvalds if (error) 36301da177e4SLinus Torvalds return error; 36311da177e4SLinus Torvalds 3632975d6b39SEric W. Biederman if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) 36331da177e4SLinus Torvalds return -EPERM; 36341da177e4SLinus Torvalds 3635acfa4380SAl Viro if (!dir->i_op->mknod) 36361da177e4SLinus Torvalds return -EPERM; 36371da177e4SLinus Torvalds 363808ce5f16SSerge E. Hallyn error = devcgroup_inode_mknod(mode, dev); 363908ce5f16SSerge E. Hallyn if (error) 364008ce5f16SSerge E. Hallyn return error; 364108ce5f16SSerge E. Hallyn 36421da177e4SLinus Torvalds error = security_inode_mknod(dir, dentry, mode, dev); 36431da177e4SLinus Torvalds if (error) 36441da177e4SLinus Torvalds return error; 36451da177e4SLinus Torvalds 36461da177e4SLinus Torvalds error = dir->i_op->mknod(dir, dentry, mode, dev); 3647a74574aaSStephen Smalley if (!error) 3648f38aa942SAmy Griffis fsnotify_create(dir, dentry); 36491da177e4SLinus Torvalds return error; 36501da177e4SLinus Torvalds } 36514d359507SAl Viro EXPORT_SYMBOL(vfs_mknod); 36521da177e4SLinus Torvalds 3653f69aac00SAl Viro static int may_mknod(umode_t mode) 3654463c3197SDave Hansen { 3655463c3197SDave Hansen switch (mode & S_IFMT) { 3656463c3197SDave Hansen case S_IFREG: 3657463c3197SDave Hansen case S_IFCHR: 3658463c3197SDave Hansen case S_IFBLK: 3659463c3197SDave Hansen case S_IFIFO: 3660463c3197SDave Hansen case S_IFSOCK: 3661463c3197SDave Hansen case 0: /* zero mode translates to S_IFREG */ 3662463c3197SDave Hansen return 0; 3663463c3197SDave Hansen case S_IFDIR: 3664463c3197SDave Hansen return -EPERM; 3665463c3197SDave Hansen default: 3666463c3197SDave Hansen return -EINVAL; 3667463c3197SDave Hansen } 3668463c3197SDave Hansen } 3669463c3197SDave Hansen 36708208a22bSAl Viro SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, 36712e4d0924SHeiko Carstens unsigned, dev) 36721da177e4SLinus Torvalds { 36731da177e4SLinus Torvalds struct dentry *dentry; 3674dae6ad8fSAl Viro struct path path; 3675dae6ad8fSAl Viro int error; 3676972567f1SJeff Layton unsigned int lookup_flags = 0; 36771da177e4SLinus Torvalds 36788e4bfca1SAl Viro error = may_mknod(mode); 36798e4bfca1SAl Viro if (error) 36808e4bfca1SAl Viro return error; 3681972567f1SJeff Layton retry: 3682972567f1SJeff Layton dentry = user_path_create(dfd, filename, &path, lookup_flags); 3683dae6ad8fSAl Viro if (IS_ERR(dentry)) 3684dae6ad8fSAl Viro return PTR_ERR(dentry); 36852ad94ae6SAl Viro 3686dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3687ce3b0f8dSAl Viro mode &= ~current_umask(); 3688dae6ad8fSAl Viro error = security_path_mknod(&path, dentry, mode, dev); 3689be6d3e56SKentaro Takeda if (error) 3690a8104a9fSAl Viro goto out; 36911da177e4SLinus Torvalds switch (mode & S_IFMT) { 36921da177e4SLinus Torvalds case 0: case S_IFREG: 3693312b63fbSAl Viro error = vfs_create(path.dentry->d_inode,dentry,mode,true); 369405d1a717SMimi Zohar if (!error) 369505d1a717SMimi Zohar ima_post_path_mknod(dentry); 36961da177e4SLinus Torvalds break; 36971da177e4SLinus Torvalds case S_IFCHR: case S_IFBLK: 3698dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode, 36991da177e4SLinus Torvalds new_decode_dev(dev)); 37001da177e4SLinus Torvalds break; 37011da177e4SLinus Torvalds case S_IFIFO: case S_IFSOCK: 3702dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); 37031da177e4SLinus Torvalds break; 37041da177e4SLinus Torvalds } 3705a8104a9fSAl Viro out: 3706921a1650SAl Viro done_path_create(&path, dentry); 3707972567f1SJeff Layton if (retry_estale(error, lookup_flags)) { 3708972567f1SJeff Layton lookup_flags |= LOOKUP_REVAL; 3709972567f1SJeff Layton goto retry; 3710972567f1SJeff Layton } 37111da177e4SLinus Torvalds return error; 37121da177e4SLinus Torvalds } 37131da177e4SLinus Torvalds 37148208a22bSAl Viro SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) 37155590ff0dSUlrich Drepper { 37165590ff0dSUlrich Drepper return sys_mknodat(AT_FDCWD, filename, mode, dev); 37175590ff0dSUlrich Drepper } 37185590ff0dSUlrich Drepper 371918bb1db3SAl Viro int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 37201da177e4SLinus Torvalds { 3721a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 37228de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 37231da177e4SLinus Torvalds 37241da177e4SLinus Torvalds if (error) 37251da177e4SLinus Torvalds return error; 37261da177e4SLinus Torvalds 3727acfa4380SAl Viro if (!dir->i_op->mkdir) 37281da177e4SLinus Torvalds return -EPERM; 37291da177e4SLinus Torvalds 37301da177e4SLinus Torvalds mode &= (S_IRWXUGO|S_ISVTX); 37311da177e4SLinus Torvalds error = security_inode_mkdir(dir, dentry, mode); 37321da177e4SLinus Torvalds if (error) 37331da177e4SLinus Torvalds return error; 37341da177e4SLinus Torvalds 37358de52778SAl Viro if (max_links && dir->i_nlink >= max_links) 37368de52778SAl Viro return -EMLINK; 37378de52778SAl Viro 37381da177e4SLinus Torvalds error = dir->i_op->mkdir(dir, dentry, mode); 3739a74574aaSStephen Smalley if (!error) 3740f38aa942SAmy Griffis fsnotify_mkdir(dir, dentry); 37411da177e4SLinus Torvalds return error; 37421da177e4SLinus Torvalds } 37434d359507SAl Viro EXPORT_SYMBOL(vfs_mkdir); 37441da177e4SLinus Torvalds 3745a218d0fdSAl Viro SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) 37461da177e4SLinus Torvalds { 37476902d925SDave Hansen struct dentry *dentry; 3748dae6ad8fSAl Viro struct path path; 3749dae6ad8fSAl Viro int error; 3750b76d8b82SJeff Layton unsigned int lookup_flags = LOOKUP_DIRECTORY; 37511da177e4SLinus Torvalds 3752b76d8b82SJeff Layton retry: 3753b76d8b82SJeff Layton dentry = user_path_create(dfd, pathname, &path, lookup_flags); 37546902d925SDave Hansen if (IS_ERR(dentry)) 3755dae6ad8fSAl Viro return PTR_ERR(dentry); 37566902d925SDave Hansen 3757dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3758ce3b0f8dSAl Viro mode &= ~current_umask(); 3759dae6ad8fSAl Viro error = security_path_mkdir(&path, dentry, mode); 3760a8104a9fSAl Viro if (!error) 3761dae6ad8fSAl Viro error = vfs_mkdir(path.dentry->d_inode, dentry, mode); 3762921a1650SAl Viro done_path_create(&path, dentry); 3763b76d8b82SJeff Layton if (retry_estale(error, lookup_flags)) { 3764b76d8b82SJeff Layton lookup_flags |= LOOKUP_REVAL; 3765b76d8b82SJeff Layton goto retry; 3766b76d8b82SJeff Layton } 37671da177e4SLinus Torvalds return error; 37681da177e4SLinus Torvalds } 37691da177e4SLinus Torvalds 3770a218d0fdSAl Viro SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) 37715590ff0dSUlrich Drepper { 37725590ff0dSUlrich Drepper return sys_mkdirat(AT_FDCWD, pathname, mode); 37735590ff0dSUlrich Drepper } 37745590ff0dSUlrich Drepper 37751da177e4SLinus Torvalds int vfs_rmdir(struct inode *dir, struct dentry *dentry) 37761da177e4SLinus Torvalds { 37771da177e4SLinus Torvalds int error = may_delete(dir, dentry, 1); 37781da177e4SLinus Torvalds 37791da177e4SLinus Torvalds if (error) 37801da177e4SLinus Torvalds return error; 37811da177e4SLinus Torvalds 3782acfa4380SAl Viro if (!dir->i_op->rmdir) 37831da177e4SLinus Torvalds return -EPERM; 37841da177e4SLinus Torvalds 37851d2ef590SAl Viro dget(dentry); 37865955102cSAl Viro inode_lock(dentry->d_inode); 3787912dbc15SSage Weil 37881da177e4SLinus Torvalds error = -EBUSY; 37897af1364fSEric W. Biederman if (is_local_mountpoint(dentry)) 3790912dbc15SSage Weil goto out; 3791912dbc15SSage Weil 37921da177e4SLinus Torvalds error = security_inode_rmdir(dir, dentry); 3793912dbc15SSage Weil if (error) 3794912dbc15SSage Weil goto out; 3795912dbc15SSage Weil 37963cebde24SSage Weil shrink_dcache_parent(dentry); 37971da177e4SLinus Torvalds error = dir->i_op->rmdir(dir, dentry); 3798912dbc15SSage Weil if (error) 3799912dbc15SSage Weil goto out; 3800912dbc15SSage Weil 38011da177e4SLinus Torvalds dentry->d_inode->i_flags |= S_DEAD; 3802d83c49f3SAl Viro dont_mount(dentry); 38038ed936b5SEric W. Biederman detach_mounts(dentry); 38041da177e4SLinus Torvalds 3805912dbc15SSage Weil out: 38065955102cSAl Viro inode_unlock(dentry->d_inode); 38071d2ef590SAl Viro dput(dentry); 3808912dbc15SSage Weil if (!error) 3809912dbc15SSage Weil d_delete(dentry); 38101da177e4SLinus Torvalds return error; 38111da177e4SLinus Torvalds } 38124d359507SAl Viro EXPORT_SYMBOL(vfs_rmdir); 38131da177e4SLinus Torvalds 38145590ff0dSUlrich Drepper static long do_rmdir(int dfd, const char __user *pathname) 38151da177e4SLinus Torvalds { 38161da177e4SLinus Torvalds int error = 0; 381791a27b2aSJeff Layton struct filename *name; 38181da177e4SLinus Torvalds struct dentry *dentry; 3819f5beed75SAl Viro struct path path; 3820f5beed75SAl Viro struct qstr last; 3821f5beed75SAl Viro int type; 3822c6ee9206SJeff Layton unsigned int lookup_flags = 0; 3823c6ee9206SJeff Layton retry: 3824c1d4dd27SAl Viro name = filename_parentat(dfd, getname(pathname), lookup_flags, 3825c1d4dd27SAl Viro &path, &last, &type); 382691a27b2aSJeff Layton if (IS_ERR(name)) 382791a27b2aSJeff Layton return PTR_ERR(name); 38281da177e4SLinus Torvalds 3829f5beed75SAl Viro switch (type) { 38301da177e4SLinus Torvalds case LAST_DOTDOT: 38311da177e4SLinus Torvalds error = -ENOTEMPTY; 38321da177e4SLinus Torvalds goto exit1; 38331da177e4SLinus Torvalds case LAST_DOT: 38341da177e4SLinus Torvalds error = -EINVAL; 38351da177e4SLinus Torvalds goto exit1; 38361da177e4SLinus Torvalds case LAST_ROOT: 38371da177e4SLinus Torvalds error = -EBUSY; 38381da177e4SLinus Torvalds goto exit1; 38391da177e4SLinus Torvalds } 38400612d9fbSOGAWA Hirofumi 3841f5beed75SAl Viro error = mnt_want_write(path.mnt); 3842c30dabfeSJan Kara if (error) 3843c30dabfeSJan Kara goto exit1; 38440612d9fbSOGAWA Hirofumi 38455955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 3846f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 38471da177e4SLinus Torvalds error = PTR_ERR(dentry); 38486902d925SDave Hansen if (IS_ERR(dentry)) 38496902d925SDave Hansen goto exit2; 3850e6bc45d6STheodore Ts'o if (!dentry->d_inode) { 3851e6bc45d6STheodore Ts'o error = -ENOENT; 3852e6bc45d6STheodore Ts'o goto exit3; 3853e6bc45d6STheodore Ts'o } 3854f5beed75SAl Viro error = security_path_rmdir(&path, dentry); 3855be6d3e56SKentaro Takeda if (error) 3856c30dabfeSJan Kara goto exit3; 3857f5beed75SAl Viro error = vfs_rmdir(path.dentry->d_inode, dentry); 38580622753bSDave Hansen exit3: 38591da177e4SLinus Torvalds dput(dentry); 38606902d925SDave Hansen exit2: 38615955102cSAl Viro inode_unlock(path.dentry->d_inode); 3862f5beed75SAl Viro mnt_drop_write(path.mnt); 38631da177e4SLinus Torvalds exit1: 3864f5beed75SAl Viro path_put(&path); 38651da177e4SLinus Torvalds putname(name); 3866c6ee9206SJeff Layton if (retry_estale(error, lookup_flags)) { 3867c6ee9206SJeff Layton lookup_flags |= LOOKUP_REVAL; 3868c6ee9206SJeff Layton goto retry; 3869c6ee9206SJeff Layton } 38701da177e4SLinus Torvalds return error; 38711da177e4SLinus Torvalds } 38721da177e4SLinus Torvalds 38733cdad428SHeiko Carstens SYSCALL_DEFINE1(rmdir, const char __user *, pathname) 38745590ff0dSUlrich Drepper { 38755590ff0dSUlrich Drepper return do_rmdir(AT_FDCWD, pathname); 38765590ff0dSUlrich Drepper } 38775590ff0dSUlrich Drepper 3878b21996e3SJ. Bruce Fields /** 3879b21996e3SJ. Bruce Fields * vfs_unlink - unlink a filesystem object 3880b21996e3SJ. Bruce Fields * @dir: parent directory 3881b21996e3SJ. Bruce Fields * @dentry: victim 3882b21996e3SJ. Bruce Fields * @delegated_inode: returns victim inode, if the inode is delegated. 3883b21996e3SJ. Bruce Fields * 3884b21996e3SJ. Bruce Fields * The caller must hold dir->i_mutex. 3885b21996e3SJ. Bruce Fields * 3886b21996e3SJ. Bruce Fields * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and 3887b21996e3SJ. Bruce Fields * return a reference to the inode in delegated_inode. The caller 3888b21996e3SJ. Bruce Fields * should then break the delegation on that inode and retry. Because 3889b21996e3SJ. Bruce Fields * breaking a delegation may take a long time, the caller should drop 3890b21996e3SJ. Bruce Fields * dir->i_mutex before doing so. 3891b21996e3SJ. Bruce Fields * 3892b21996e3SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 3893b21996e3SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 3894b21996e3SJ. Bruce Fields * to be NFS exported. 3895b21996e3SJ. Bruce Fields */ 3896b21996e3SJ. Bruce Fields int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode) 38971da177e4SLinus Torvalds { 38989accbb97SJ. Bruce Fields struct inode *target = dentry->d_inode; 38991da177e4SLinus Torvalds int error = may_delete(dir, dentry, 0); 39001da177e4SLinus Torvalds 39011da177e4SLinus Torvalds if (error) 39021da177e4SLinus Torvalds return error; 39031da177e4SLinus Torvalds 3904acfa4380SAl Viro if (!dir->i_op->unlink) 39051da177e4SLinus Torvalds return -EPERM; 39061da177e4SLinus Torvalds 39075955102cSAl Viro inode_lock(target); 39088ed936b5SEric W. Biederman if (is_local_mountpoint(dentry)) 39091da177e4SLinus Torvalds error = -EBUSY; 39101da177e4SLinus Torvalds else { 39111da177e4SLinus Torvalds error = security_inode_unlink(dir, dentry); 3912bec1052eSAl Viro if (!error) { 39135a14696cSJ. Bruce Fields error = try_break_deleg(target, delegated_inode); 39145a14696cSJ. Bruce Fields if (error) 3915b21996e3SJ. Bruce Fields goto out; 39161da177e4SLinus Torvalds error = dir->i_op->unlink(dir, dentry); 39178ed936b5SEric W. Biederman if (!error) { 3918d83c49f3SAl Viro dont_mount(dentry); 39198ed936b5SEric W. Biederman detach_mounts(dentry); 39208ed936b5SEric W. Biederman } 3921bec1052eSAl Viro } 39221da177e4SLinus Torvalds } 3923b21996e3SJ. Bruce Fields out: 39245955102cSAl Viro inode_unlock(target); 39251da177e4SLinus Torvalds 39261da177e4SLinus Torvalds /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 39271da177e4SLinus Torvalds if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { 39289accbb97SJ. Bruce Fields fsnotify_link_count(target); 39291da177e4SLinus Torvalds d_delete(dentry); 39301da177e4SLinus Torvalds } 39310eeca283SRobert Love 39321da177e4SLinus Torvalds return error; 39331da177e4SLinus Torvalds } 39344d359507SAl Viro EXPORT_SYMBOL(vfs_unlink); 39351da177e4SLinus Torvalds 39361da177e4SLinus Torvalds /* 39371da177e4SLinus Torvalds * Make sure that the actual truncation of the file will occur outside its 39381b1dcc1bSJes Sorensen * directory's i_mutex. Truncate can take a long time if there is a lot of 39391da177e4SLinus Torvalds * writeout happening, and we don't want to prevent access to the directory 39401da177e4SLinus Torvalds * while waiting on the I/O. 39411da177e4SLinus Torvalds */ 39425590ff0dSUlrich Drepper static long do_unlinkat(int dfd, const char __user *pathname) 39431da177e4SLinus Torvalds { 39442ad94ae6SAl Viro int error; 394591a27b2aSJeff Layton struct filename *name; 39461da177e4SLinus Torvalds struct dentry *dentry; 3947f5beed75SAl Viro struct path path; 3948f5beed75SAl Viro struct qstr last; 3949f5beed75SAl Viro int type; 39501da177e4SLinus Torvalds struct inode *inode = NULL; 3951b21996e3SJ. Bruce Fields struct inode *delegated_inode = NULL; 39525d18f813SJeff Layton unsigned int lookup_flags = 0; 39535d18f813SJeff Layton retry: 3954c1d4dd27SAl Viro name = filename_parentat(dfd, getname(pathname), lookup_flags, 3955c1d4dd27SAl Viro &path, &last, &type); 395691a27b2aSJeff Layton if (IS_ERR(name)) 395791a27b2aSJeff Layton return PTR_ERR(name); 39582ad94ae6SAl Viro 39591da177e4SLinus Torvalds error = -EISDIR; 3960f5beed75SAl Viro if (type != LAST_NORM) 39611da177e4SLinus Torvalds goto exit1; 39620612d9fbSOGAWA Hirofumi 3963f5beed75SAl Viro error = mnt_want_write(path.mnt); 3964c30dabfeSJan Kara if (error) 3965c30dabfeSJan Kara goto exit1; 3966b21996e3SJ. Bruce Fields retry_deleg: 39675955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 3968f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 39691da177e4SLinus Torvalds error = PTR_ERR(dentry); 39701da177e4SLinus Torvalds if (!IS_ERR(dentry)) { 39711da177e4SLinus Torvalds /* Why not before? Because we want correct error value */ 3972f5beed75SAl Viro if (last.name[last.len]) 397350338b88STörök Edwin goto slashes; 39741da177e4SLinus Torvalds inode = dentry->d_inode; 3975b18825a7SDavid Howells if (d_is_negative(dentry)) 3976e6bc45d6STheodore Ts'o goto slashes; 39777de9c6eeSAl Viro ihold(inode); 3978f5beed75SAl Viro error = security_path_unlink(&path, dentry); 3979be6d3e56SKentaro Takeda if (error) 3980c30dabfeSJan Kara goto exit2; 3981f5beed75SAl Viro error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode); 39821da177e4SLinus Torvalds exit2: 39831da177e4SLinus Torvalds dput(dentry); 39841da177e4SLinus Torvalds } 39855955102cSAl Viro inode_unlock(path.dentry->d_inode); 39861da177e4SLinus Torvalds if (inode) 39871da177e4SLinus Torvalds iput(inode); /* truncate the inode here */ 3988b21996e3SJ. Bruce Fields inode = NULL; 3989b21996e3SJ. Bruce Fields if (delegated_inode) { 39905a14696cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 3991b21996e3SJ. Bruce Fields if (!error) 3992b21996e3SJ. Bruce Fields goto retry_deleg; 3993b21996e3SJ. Bruce Fields } 3994f5beed75SAl Viro mnt_drop_write(path.mnt); 39951da177e4SLinus Torvalds exit1: 3996f5beed75SAl Viro path_put(&path); 39971da177e4SLinus Torvalds putname(name); 39985d18f813SJeff Layton if (retry_estale(error, lookup_flags)) { 39995d18f813SJeff Layton lookup_flags |= LOOKUP_REVAL; 40005d18f813SJeff Layton inode = NULL; 40015d18f813SJeff Layton goto retry; 40025d18f813SJeff Layton } 40031da177e4SLinus Torvalds return error; 40041da177e4SLinus Torvalds 40051da177e4SLinus Torvalds slashes: 4006b18825a7SDavid Howells if (d_is_negative(dentry)) 4007b18825a7SDavid Howells error = -ENOENT; 400844b1d530SMiklos Szeredi else if (d_is_dir(dentry)) 4009b18825a7SDavid Howells error = -EISDIR; 4010b18825a7SDavid Howells else 4011b18825a7SDavid Howells error = -ENOTDIR; 40121da177e4SLinus Torvalds goto exit2; 40131da177e4SLinus Torvalds } 40141da177e4SLinus Torvalds 40152e4d0924SHeiko Carstens SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) 40165590ff0dSUlrich Drepper { 40175590ff0dSUlrich Drepper if ((flag & ~AT_REMOVEDIR) != 0) 40185590ff0dSUlrich Drepper return -EINVAL; 40195590ff0dSUlrich Drepper 40205590ff0dSUlrich Drepper if (flag & AT_REMOVEDIR) 40215590ff0dSUlrich Drepper return do_rmdir(dfd, pathname); 40225590ff0dSUlrich Drepper 40235590ff0dSUlrich Drepper return do_unlinkat(dfd, pathname); 40245590ff0dSUlrich Drepper } 40255590ff0dSUlrich Drepper 40263480b257SHeiko Carstens SYSCALL_DEFINE1(unlink, const char __user *, pathname) 40275590ff0dSUlrich Drepper { 40285590ff0dSUlrich Drepper return do_unlinkat(AT_FDCWD, pathname); 40295590ff0dSUlrich Drepper } 40305590ff0dSUlrich Drepper 4031db2e747bSMiklos Szeredi int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) 40321da177e4SLinus Torvalds { 4033a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 40341da177e4SLinus Torvalds 40351da177e4SLinus Torvalds if (error) 40361da177e4SLinus Torvalds return error; 40371da177e4SLinus Torvalds 4038acfa4380SAl Viro if (!dir->i_op->symlink) 40391da177e4SLinus Torvalds return -EPERM; 40401da177e4SLinus Torvalds 40411da177e4SLinus Torvalds error = security_inode_symlink(dir, dentry, oldname); 40421da177e4SLinus Torvalds if (error) 40431da177e4SLinus Torvalds return error; 40441da177e4SLinus Torvalds 40451da177e4SLinus Torvalds error = dir->i_op->symlink(dir, dentry, oldname); 4046a74574aaSStephen Smalley if (!error) 4047f38aa942SAmy Griffis fsnotify_create(dir, dentry); 40481da177e4SLinus Torvalds return error; 40491da177e4SLinus Torvalds } 40504d359507SAl Viro EXPORT_SYMBOL(vfs_symlink); 40511da177e4SLinus Torvalds 40522e4d0924SHeiko Carstens SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, 40532e4d0924SHeiko Carstens int, newdfd, const char __user *, newname) 40541da177e4SLinus Torvalds { 40552ad94ae6SAl Viro int error; 405691a27b2aSJeff Layton struct filename *from; 40576902d925SDave Hansen struct dentry *dentry; 4058dae6ad8fSAl Viro struct path path; 4059f46d3567SJeff Layton unsigned int lookup_flags = 0; 40601da177e4SLinus Torvalds 40611da177e4SLinus Torvalds from = getname(oldname); 40621da177e4SLinus Torvalds if (IS_ERR(from)) 40631da177e4SLinus Torvalds return PTR_ERR(from); 4064f46d3567SJeff Layton retry: 4065f46d3567SJeff Layton dentry = user_path_create(newdfd, newname, &path, lookup_flags); 40661da177e4SLinus Torvalds error = PTR_ERR(dentry); 40676902d925SDave Hansen if (IS_ERR(dentry)) 4068dae6ad8fSAl Viro goto out_putname; 40696902d925SDave Hansen 407091a27b2aSJeff Layton error = security_path_symlink(&path, dentry, from->name); 4071a8104a9fSAl Viro if (!error) 407291a27b2aSJeff Layton error = vfs_symlink(path.dentry->d_inode, dentry, from->name); 4073921a1650SAl Viro done_path_create(&path, dentry); 4074f46d3567SJeff Layton if (retry_estale(error, lookup_flags)) { 4075f46d3567SJeff Layton lookup_flags |= LOOKUP_REVAL; 4076f46d3567SJeff Layton goto retry; 4077f46d3567SJeff Layton } 40786902d925SDave Hansen out_putname: 40791da177e4SLinus Torvalds putname(from); 40801da177e4SLinus Torvalds return error; 40811da177e4SLinus Torvalds } 40821da177e4SLinus Torvalds 40833480b257SHeiko Carstens SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) 40845590ff0dSUlrich Drepper { 40855590ff0dSUlrich Drepper return sys_symlinkat(oldname, AT_FDCWD, newname); 40865590ff0dSUlrich Drepper } 40875590ff0dSUlrich Drepper 4088146a8595SJ. Bruce Fields /** 4089146a8595SJ. Bruce Fields * vfs_link - create a new link 4090146a8595SJ. Bruce Fields * @old_dentry: object to be linked 4091146a8595SJ. Bruce Fields * @dir: new parent 4092146a8595SJ. Bruce Fields * @new_dentry: where to create the new link 4093146a8595SJ. Bruce Fields * @delegated_inode: returns inode needing a delegation break 4094146a8595SJ. Bruce Fields * 4095146a8595SJ. Bruce Fields * The caller must hold dir->i_mutex 4096146a8595SJ. Bruce Fields * 4097146a8595SJ. Bruce Fields * If vfs_link discovers a delegation on the to-be-linked file in need 4098146a8595SJ. Bruce Fields * of breaking, it will return -EWOULDBLOCK and return a reference to the 4099146a8595SJ. Bruce Fields * inode in delegated_inode. The caller should then break the delegation 4100146a8595SJ. Bruce Fields * and retry. Because breaking a delegation may take a long time, the 4101146a8595SJ. Bruce Fields * caller should drop the i_mutex before doing so. 4102146a8595SJ. Bruce Fields * 4103146a8595SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 4104146a8595SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 4105146a8595SJ. Bruce Fields * to be NFS exported. 4106146a8595SJ. Bruce Fields */ 4107146a8595SJ. Bruce Fields int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode) 41081da177e4SLinus Torvalds { 41091da177e4SLinus Torvalds struct inode *inode = old_dentry->d_inode; 41108de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 41111da177e4SLinus Torvalds int error; 41121da177e4SLinus Torvalds 41131da177e4SLinus Torvalds if (!inode) 41141da177e4SLinus Torvalds return -ENOENT; 41151da177e4SLinus Torvalds 4116a95164d9SMiklos Szeredi error = may_create(dir, new_dentry); 41171da177e4SLinus Torvalds if (error) 41181da177e4SLinus Torvalds return error; 41191da177e4SLinus Torvalds 41201da177e4SLinus Torvalds if (dir->i_sb != inode->i_sb) 41211da177e4SLinus Torvalds return -EXDEV; 41221da177e4SLinus Torvalds 41231da177e4SLinus Torvalds /* 41241da177e4SLinus Torvalds * A link to an append-only or immutable file cannot be created. 41251da177e4SLinus Torvalds */ 41261da177e4SLinus Torvalds if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 41271da177e4SLinus Torvalds return -EPERM; 41280bd23d09SEric W. Biederman /* 41290bd23d09SEric W. Biederman * Updating the link count will likely cause i_uid and i_gid to 41300bd23d09SEric W. Biederman * be writen back improperly if their true value is unknown to 41310bd23d09SEric W. Biederman * the vfs. 41320bd23d09SEric W. Biederman */ 41330bd23d09SEric W. Biederman if (HAS_UNMAPPED_ID(inode)) 41340bd23d09SEric W. Biederman return -EPERM; 4135acfa4380SAl Viro if (!dir->i_op->link) 41361da177e4SLinus Torvalds return -EPERM; 41377e79eedbSTetsuo Handa if (S_ISDIR(inode->i_mode)) 41381da177e4SLinus Torvalds return -EPERM; 41391da177e4SLinus Torvalds 41401da177e4SLinus Torvalds error = security_inode_link(old_dentry, dir, new_dentry); 41411da177e4SLinus Torvalds if (error) 41421da177e4SLinus Torvalds return error; 41431da177e4SLinus Torvalds 41445955102cSAl Viro inode_lock(inode); 4145aae8a97dSAneesh Kumar K.V /* Make sure we don't allow creating hardlink to an unlinked file */ 4146f4e0c30cSAl Viro if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) 4147aae8a97dSAneesh Kumar K.V error = -ENOENT; 41488de52778SAl Viro else if (max_links && inode->i_nlink >= max_links) 41498de52778SAl Viro error = -EMLINK; 4150146a8595SJ. Bruce Fields else { 4151146a8595SJ. Bruce Fields error = try_break_deleg(inode, delegated_inode); 4152146a8595SJ. Bruce Fields if (!error) 41531da177e4SLinus Torvalds error = dir->i_op->link(old_dentry, dir, new_dentry); 4154146a8595SJ. Bruce Fields } 4155f4e0c30cSAl Viro 4156f4e0c30cSAl Viro if (!error && (inode->i_state & I_LINKABLE)) { 4157f4e0c30cSAl Viro spin_lock(&inode->i_lock); 4158f4e0c30cSAl Viro inode->i_state &= ~I_LINKABLE; 4159f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 4160f4e0c30cSAl Viro } 41615955102cSAl Viro inode_unlock(inode); 4162e31e14ecSStephen Smalley if (!error) 41637e79eedbSTetsuo Handa fsnotify_link(dir, inode, new_dentry); 41641da177e4SLinus Torvalds return error; 41651da177e4SLinus Torvalds } 41664d359507SAl Viro EXPORT_SYMBOL(vfs_link); 41671da177e4SLinus Torvalds 41681da177e4SLinus Torvalds /* 41691da177e4SLinus Torvalds * Hardlinks are often used in delicate situations. We avoid 41701da177e4SLinus Torvalds * security-related surprises by not following symlinks on the 41711da177e4SLinus Torvalds * newname. --KAB 41721da177e4SLinus Torvalds * 41731da177e4SLinus Torvalds * We don't follow them on the oldname either to be compatible 41741da177e4SLinus Torvalds * with linux 2.0, and to avoid hard-linking to directories 41751da177e4SLinus Torvalds * and other special files. --ADM 41761da177e4SLinus Torvalds */ 41772e4d0924SHeiko Carstens SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, 41782e4d0924SHeiko Carstens int, newdfd, const char __user *, newname, int, flags) 41791da177e4SLinus Torvalds { 41801da177e4SLinus Torvalds struct dentry *new_dentry; 4181dae6ad8fSAl Viro struct path old_path, new_path; 4182146a8595SJ. Bruce Fields struct inode *delegated_inode = NULL; 418311a7b371SAneesh Kumar K.V int how = 0; 41841da177e4SLinus Torvalds int error; 41851da177e4SLinus Torvalds 418611a7b371SAneesh Kumar K.V if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) 4187c04030e1SUlrich Drepper return -EINVAL; 418811a7b371SAneesh Kumar K.V /* 4189f0cc6ffbSLinus Torvalds * To use null names we require CAP_DAC_READ_SEARCH 4190f0cc6ffbSLinus Torvalds * This ensures that not everyone will be able to create 4191f0cc6ffbSLinus Torvalds * handlink using the passed filedescriptor. 419211a7b371SAneesh Kumar K.V */ 4193f0cc6ffbSLinus Torvalds if (flags & AT_EMPTY_PATH) { 4194f0cc6ffbSLinus Torvalds if (!capable(CAP_DAC_READ_SEARCH)) 4195f0cc6ffbSLinus Torvalds return -ENOENT; 419611a7b371SAneesh Kumar K.V how = LOOKUP_EMPTY; 4197f0cc6ffbSLinus Torvalds } 4198c04030e1SUlrich Drepper 419911a7b371SAneesh Kumar K.V if (flags & AT_SYMLINK_FOLLOW) 420011a7b371SAneesh Kumar K.V how |= LOOKUP_FOLLOW; 4201442e31caSJeff Layton retry: 420211a7b371SAneesh Kumar K.V error = user_path_at(olddfd, oldname, how, &old_path); 42031da177e4SLinus Torvalds if (error) 42042ad94ae6SAl Viro return error; 42052ad94ae6SAl Viro 4206442e31caSJeff Layton new_dentry = user_path_create(newdfd, newname, &new_path, 4207442e31caSJeff Layton (how & LOOKUP_REVAL)); 42081da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 42096902d925SDave Hansen if (IS_ERR(new_dentry)) 4210dae6ad8fSAl Viro goto out; 4211dae6ad8fSAl Viro 4212dae6ad8fSAl Viro error = -EXDEV; 4213dae6ad8fSAl Viro if (old_path.mnt != new_path.mnt) 4214dae6ad8fSAl Viro goto out_dput; 4215800179c9SKees Cook error = may_linkat(&old_path); 4216800179c9SKees Cook if (unlikely(error)) 4217800179c9SKees Cook goto out_dput; 4218dae6ad8fSAl Viro error = security_path_link(old_path.dentry, &new_path, new_dentry); 4219be6d3e56SKentaro Takeda if (error) 4220a8104a9fSAl Viro goto out_dput; 4221146a8595SJ. Bruce Fields error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode); 422275c3f29dSDave Hansen out_dput: 4223921a1650SAl Viro done_path_create(&new_path, new_dentry); 4224146a8595SJ. Bruce Fields if (delegated_inode) { 4225146a8595SJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4226d22e6338SOleg Drokin if (!error) { 4227d22e6338SOleg Drokin path_put(&old_path); 4228146a8595SJ. Bruce Fields goto retry; 4229146a8595SJ. Bruce Fields } 4230d22e6338SOleg Drokin } 4231442e31caSJeff Layton if (retry_estale(error, how)) { 4232d22e6338SOleg Drokin path_put(&old_path); 4233442e31caSJeff Layton how |= LOOKUP_REVAL; 4234442e31caSJeff Layton goto retry; 4235442e31caSJeff Layton } 42361da177e4SLinus Torvalds out: 42372d8f3038SAl Viro path_put(&old_path); 42381da177e4SLinus Torvalds 42391da177e4SLinus Torvalds return error; 42401da177e4SLinus Torvalds } 42411da177e4SLinus Torvalds 42423480b257SHeiko Carstens SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) 42435590ff0dSUlrich Drepper { 4244c04030e1SUlrich Drepper return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 42455590ff0dSUlrich Drepper } 42465590ff0dSUlrich Drepper 4247bc27027aSMiklos Szeredi /** 4248bc27027aSMiklos Szeredi * vfs_rename - rename a filesystem object 4249bc27027aSMiklos Szeredi * @old_dir: parent of source 4250bc27027aSMiklos Szeredi * @old_dentry: source 4251bc27027aSMiklos Szeredi * @new_dir: parent of destination 4252bc27027aSMiklos Szeredi * @new_dentry: destination 4253bc27027aSMiklos Szeredi * @delegated_inode: returns an inode needing a delegation break 4254520c8b16SMiklos Szeredi * @flags: rename flags 4255bc27027aSMiklos Szeredi * 4256bc27027aSMiklos Szeredi * The caller must hold multiple mutexes--see lock_rename()). 4257bc27027aSMiklos Szeredi * 4258bc27027aSMiklos Szeredi * If vfs_rename discovers a delegation in need of breaking at either 4259bc27027aSMiklos Szeredi * the source or destination, it will return -EWOULDBLOCK and return a 4260bc27027aSMiklos Szeredi * reference to the inode in delegated_inode. The caller should then 4261bc27027aSMiklos Szeredi * break the delegation and retry. Because breaking a delegation may 4262bc27027aSMiklos Szeredi * take a long time, the caller should drop all locks before doing 4263bc27027aSMiklos Szeredi * so. 4264bc27027aSMiklos Szeredi * 4265bc27027aSMiklos Szeredi * Alternatively, a caller may pass NULL for delegated_inode. This may 4266bc27027aSMiklos Szeredi * be appropriate for callers that expect the underlying filesystem not 4267bc27027aSMiklos Szeredi * to be NFS exported. 4268bc27027aSMiklos Szeredi * 42691da177e4SLinus Torvalds * The worst of all namespace operations - renaming directory. "Perverted" 42701da177e4SLinus Torvalds * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 42711da177e4SLinus Torvalds * Problems: 4272d03b29a2SJ. Bruce Fields * a) we can get into loop creation. 42731da177e4SLinus Torvalds * b) race potential - two innocent renames can create a loop together. 42741da177e4SLinus Torvalds * That's where 4.4 screws up. Current fix: serialization on 4275a11f3a05SArjan van de Ven * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 42761da177e4SLinus Torvalds * story. 42776cedba89SJ. Bruce Fields * c) we have to lock _four_ objects - parents and victim (if it exists), 42786cedba89SJ. Bruce Fields * and source (if it is not a directory). 42791b1dcc1bSJes Sorensen * And that - after we got ->i_mutex on parents (until then we don't know 42801da177e4SLinus Torvalds * whether the target exists). Solution: try to be smart with locking 42811da177e4SLinus Torvalds * order for inodes. We rely on the fact that tree topology may change 4282a11f3a05SArjan van de Ven * only under ->s_vfs_rename_mutex _and_ that parent of the object we 42831da177e4SLinus Torvalds * move will be locked. Thus we can rank directories by the tree 42841da177e4SLinus Torvalds * (ancestors first) and rank all non-directories after them. 42851da177e4SLinus Torvalds * That works since everybody except rename does "lock parent, lookup, 4286a11f3a05SArjan van de Ven * lock child" and rename is under ->s_vfs_rename_mutex. 42871da177e4SLinus Torvalds * HOWEVER, it relies on the assumption that any object with ->lookup() 42881da177e4SLinus Torvalds * has no more than 1 dentry. If "hybrid" objects will ever appear, 42891da177e4SLinus Torvalds * we'd better make sure that there's no link(2) for them. 4290e4eaac06SSage Weil * d) conversion from fhandle to dentry may come in the wrong moment - when 42911b1dcc1bSJes Sorensen * we are removing the target. Solution: we will have to grab ->i_mutex 42921da177e4SLinus Torvalds * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 4293c41b20e7SAdam Buchbinder * ->i_mutex on parents, which works but leads to some truly excessive 42941da177e4SLinus Torvalds * locking]. 42951da177e4SLinus Torvalds */ 42961da177e4SLinus Torvalds int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, 42978e6d782cSJ. Bruce Fields struct inode *new_dir, struct dentry *new_dentry, 4298520c8b16SMiklos Szeredi struct inode **delegated_inode, unsigned int flags) 42991da177e4SLinus Torvalds { 43001da177e4SLinus Torvalds int error; 4301bc27027aSMiklos Szeredi bool is_dir = d_is_dir(old_dentry); 430259b0df21SEric Paris const unsigned char *old_name; 4303bc27027aSMiklos Szeredi struct inode *source = old_dentry->d_inode; 4304bc27027aSMiklos Szeredi struct inode *target = new_dentry->d_inode; 4305da1ce067SMiklos Szeredi bool new_is_dir = false; 4306da1ce067SMiklos Szeredi unsigned max_links = new_dir->i_sb->s_max_links; 43071da177e4SLinus Torvalds 43088d3e2936SMiklos Szeredi if (source == target) 43091da177e4SLinus Torvalds return 0; 43101da177e4SLinus Torvalds 43111da177e4SLinus Torvalds error = may_delete(old_dir, old_dentry, is_dir); 43121da177e4SLinus Torvalds if (error) 43131da177e4SLinus Torvalds return error; 43141da177e4SLinus Torvalds 4315da1ce067SMiklos Szeredi if (!target) { 4316a95164d9SMiklos Szeredi error = may_create(new_dir, new_dentry); 4317da1ce067SMiklos Szeredi } else { 4318da1ce067SMiklos Szeredi new_is_dir = d_is_dir(new_dentry); 4319da1ce067SMiklos Szeredi 4320da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 43211da177e4SLinus Torvalds error = may_delete(new_dir, new_dentry, is_dir); 4322da1ce067SMiklos Szeredi else 4323da1ce067SMiklos Szeredi error = may_delete(new_dir, new_dentry, new_is_dir); 4324da1ce067SMiklos Szeredi } 43251da177e4SLinus Torvalds if (error) 43261da177e4SLinus Torvalds return error; 43271da177e4SLinus Torvalds 43282773bf00SMiklos Szeredi if (!old_dir->i_op->rename) 43291da177e4SLinus Torvalds return -EPERM; 43301da177e4SLinus Torvalds 4331bc27027aSMiklos Szeredi /* 4332bc27027aSMiklos Szeredi * If we are going to change the parent - check write permissions, 4333bc27027aSMiklos Szeredi * we'll need to flip '..'. 4334bc27027aSMiklos Szeredi */ 4335da1ce067SMiklos Szeredi if (new_dir != old_dir) { 4336da1ce067SMiklos Szeredi if (is_dir) { 4337bc27027aSMiklos Szeredi error = inode_permission(source, MAY_WRITE); 4338bc27027aSMiklos Szeredi if (error) 4339bc27027aSMiklos Szeredi return error; 4340bc27027aSMiklos Szeredi } 4341da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && new_is_dir) { 4342da1ce067SMiklos Szeredi error = inode_permission(target, MAY_WRITE); 4343da1ce067SMiklos Szeredi if (error) 4344da1ce067SMiklos Szeredi return error; 4345da1ce067SMiklos Szeredi } 4346da1ce067SMiklos Szeredi } 43470eeca283SRobert Love 43480b3974ebSMiklos Szeredi error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry, 43490b3974ebSMiklos Szeredi flags); 4350bc27027aSMiklos Szeredi if (error) 4351bc27027aSMiklos Szeredi return error; 4352bc27027aSMiklos Szeredi 4353bc27027aSMiklos Szeredi old_name = fsnotify_oldname_init(old_dentry->d_name.name); 4354bc27027aSMiklos Szeredi dget(new_dentry); 4355da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4356bc27027aSMiklos Szeredi lock_two_nondirectories(source, target); 4357bc27027aSMiklos Szeredi else if (target) 43585955102cSAl Viro inode_lock(target); 4359bc27027aSMiklos Szeredi 4360bc27027aSMiklos Szeredi error = -EBUSY; 43617af1364fSEric W. Biederman if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry)) 4362bc27027aSMiklos Szeredi goto out; 4363bc27027aSMiklos Szeredi 4364da1ce067SMiklos Szeredi if (max_links && new_dir != old_dir) { 4365bc27027aSMiklos Szeredi error = -EMLINK; 4366da1ce067SMiklos Szeredi if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links) 4367bc27027aSMiklos Szeredi goto out; 4368da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir && 4369da1ce067SMiklos Szeredi old_dir->i_nlink >= max_links) 4370da1ce067SMiklos Szeredi goto out; 4371da1ce067SMiklos Szeredi } 4372da1ce067SMiklos Szeredi if (is_dir && !(flags & RENAME_EXCHANGE) && target) 4373bc27027aSMiklos Szeredi shrink_dcache_parent(new_dentry); 4374da1ce067SMiklos Szeredi if (!is_dir) { 4375bc27027aSMiklos Szeredi error = try_break_deleg(source, delegated_inode); 4376bc27027aSMiklos Szeredi if (error) 4377bc27027aSMiklos Szeredi goto out; 4378da1ce067SMiklos Szeredi } 4379da1ce067SMiklos Szeredi if (target && !new_is_dir) { 4380bc27027aSMiklos Szeredi error = try_break_deleg(target, delegated_inode); 4381bc27027aSMiklos Szeredi if (error) 4382bc27027aSMiklos Szeredi goto out; 4383bc27027aSMiklos Szeredi } 4384520c8b16SMiklos Szeredi error = old_dir->i_op->rename(old_dir, old_dentry, 4385520c8b16SMiklos Szeredi new_dir, new_dentry, flags); 4386bc27027aSMiklos Szeredi if (error) 4387bc27027aSMiklos Szeredi goto out; 4388bc27027aSMiklos Szeredi 4389da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE) && target) { 43901da177e4SLinus Torvalds if (is_dir) 4391bc27027aSMiklos Szeredi target->i_flags |= S_DEAD; 4392bc27027aSMiklos Szeredi dont_mount(new_dentry); 43938ed936b5SEric W. Biederman detach_mounts(new_dentry); 4394bc27027aSMiklos Szeredi } 4395da1ce067SMiklos Szeredi if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) { 4396da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 4397bc27027aSMiklos Szeredi d_move(old_dentry, new_dentry); 4398da1ce067SMiklos Szeredi else 4399da1ce067SMiklos Szeredi d_exchange(old_dentry, new_dentry); 4400da1ce067SMiklos Szeredi } 4401bc27027aSMiklos Szeredi out: 4402da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4403bc27027aSMiklos Szeredi unlock_two_nondirectories(source, target); 4404bc27027aSMiklos Szeredi else if (target) 44055955102cSAl Viro inode_unlock(target); 4406bc27027aSMiklos Szeredi dput(new_dentry); 4407da1ce067SMiklos Szeredi if (!error) { 4408123df294SAl Viro fsnotify_move(old_dir, new_dir, old_name, is_dir, 4409da1ce067SMiklos Szeredi !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry); 4410da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4411da1ce067SMiklos Szeredi fsnotify_move(new_dir, old_dir, old_dentry->d_name.name, 4412da1ce067SMiklos Szeredi new_is_dir, NULL, new_dentry); 4413da1ce067SMiklos Szeredi } 4414da1ce067SMiklos Szeredi } 44150eeca283SRobert Love fsnotify_oldname_free(old_name); 44160eeca283SRobert Love 44171da177e4SLinus Torvalds return error; 44181da177e4SLinus Torvalds } 44194d359507SAl Viro EXPORT_SYMBOL(vfs_rename); 44201da177e4SLinus Torvalds 4421520c8b16SMiklos Szeredi SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname, 4422520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname, unsigned int, flags) 44231da177e4SLinus Torvalds { 44241da177e4SLinus Torvalds struct dentry *old_dentry, *new_dentry; 44251da177e4SLinus Torvalds struct dentry *trap; 4426f5beed75SAl Viro struct path old_path, new_path; 4427f5beed75SAl Viro struct qstr old_last, new_last; 4428f5beed75SAl Viro int old_type, new_type; 44298e6d782cSJ. Bruce Fields struct inode *delegated_inode = NULL; 443091a27b2aSJeff Layton struct filename *from; 443191a27b2aSJeff Layton struct filename *to; 4432f5beed75SAl Viro unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET; 4433c6a94284SJeff Layton bool should_retry = false; 44342ad94ae6SAl Viro int error; 4435520c8b16SMiklos Szeredi 44360d7a8555SMiklos Szeredi if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) 4437da1ce067SMiklos Szeredi return -EINVAL; 4438da1ce067SMiklos Szeredi 44390d7a8555SMiklos Szeredi if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) && 44400d7a8555SMiklos Szeredi (flags & RENAME_EXCHANGE)) 4441520c8b16SMiklos Szeredi return -EINVAL; 4442520c8b16SMiklos Szeredi 44430d7a8555SMiklos Szeredi if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD)) 44440d7a8555SMiklos Szeredi return -EPERM; 44450d7a8555SMiklos Szeredi 4446f5beed75SAl Viro if (flags & RENAME_EXCHANGE) 4447f5beed75SAl Viro target_flags = 0; 4448f5beed75SAl Viro 4449c6a94284SJeff Layton retry: 4450c1d4dd27SAl Viro from = filename_parentat(olddfd, getname(oldname), lookup_flags, 4451c1d4dd27SAl Viro &old_path, &old_last, &old_type); 445291a27b2aSJeff Layton if (IS_ERR(from)) { 445391a27b2aSJeff Layton error = PTR_ERR(from); 44541da177e4SLinus Torvalds goto exit; 445591a27b2aSJeff Layton } 44561da177e4SLinus Torvalds 4457c1d4dd27SAl Viro to = filename_parentat(newdfd, getname(newname), lookup_flags, 4458c1d4dd27SAl Viro &new_path, &new_last, &new_type); 445991a27b2aSJeff Layton if (IS_ERR(to)) { 446091a27b2aSJeff Layton error = PTR_ERR(to); 44611da177e4SLinus Torvalds goto exit1; 446291a27b2aSJeff Layton } 44631da177e4SLinus Torvalds 44641da177e4SLinus Torvalds error = -EXDEV; 4465f5beed75SAl Viro if (old_path.mnt != new_path.mnt) 44661da177e4SLinus Torvalds goto exit2; 44671da177e4SLinus Torvalds 44681da177e4SLinus Torvalds error = -EBUSY; 4469f5beed75SAl Viro if (old_type != LAST_NORM) 44701da177e4SLinus Torvalds goto exit2; 44711da177e4SLinus Torvalds 44720a7c3937SMiklos Szeredi if (flags & RENAME_NOREPLACE) 44730a7c3937SMiklos Szeredi error = -EEXIST; 4474f5beed75SAl Viro if (new_type != LAST_NORM) 44751da177e4SLinus Torvalds goto exit2; 44761da177e4SLinus Torvalds 4477f5beed75SAl Viro error = mnt_want_write(old_path.mnt); 4478c30dabfeSJan Kara if (error) 4479c30dabfeSJan Kara goto exit2; 4480c30dabfeSJan Kara 44818e6d782cSJ. Bruce Fields retry_deleg: 4482f5beed75SAl Viro trap = lock_rename(new_path.dentry, old_path.dentry); 44831da177e4SLinus Torvalds 4484f5beed75SAl Viro old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags); 44851da177e4SLinus Torvalds error = PTR_ERR(old_dentry); 44861da177e4SLinus Torvalds if (IS_ERR(old_dentry)) 44871da177e4SLinus Torvalds goto exit3; 44881da177e4SLinus Torvalds /* source must exist */ 44891da177e4SLinus Torvalds error = -ENOENT; 4490b18825a7SDavid Howells if (d_is_negative(old_dentry)) 44911da177e4SLinus Torvalds goto exit4; 4492f5beed75SAl Viro new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags); 44931da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 44941da177e4SLinus Torvalds if (IS_ERR(new_dentry)) 44951da177e4SLinus Torvalds goto exit4; 44960a7c3937SMiklos Szeredi error = -EEXIST; 44970a7c3937SMiklos Szeredi if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry)) 44980a7c3937SMiklos Szeredi goto exit5; 4499da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4500da1ce067SMiklos Szeredi error = -ENOENT; 4501da1ce067SMiklos Szeredi if (d_is_negative(new_dentry)) 4502da1ce067SMiklos Szeredi goto exit5; 4503da1ce067SMiklos Szeredi 4504da1ce067SMiklos Szeredi if (!d_is_dir(new_dentry)) { 4505da1ce067SMiklos Szeredi error = -ENOTDIR; 4506f5beed75SAl Viro if (new_last.name[new_last.len]) 4507da1ce067SMiklos Szeredi goto exit5; 4508da1ce067SMiklos Szeredi } 4509da1ce067SMiklos Szeredi } 45100a7c3937SMiklos Szeredi /* unless the source is a directory trailing slashes give -ENOTDIR */ 45110a7c3937SMiklos Szeredi if (!d_is_dir(old_dentry)) { 45120a7c3937SMiklos Szeredi error = -ENOTDIR; 4513f5beed75SAl Viro if (old_last.name[old_last.len]) 45140a7c3937SMiklos Szeredi goto exit5; 4515f5beed75SAl Viro if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len]) 45160a7c3937SMiklos Szeredi goto exit5; 45170a7c3937SMiklos Szeredi } 45180a7c3937SMiklos Szeredi /* source should not be ancestor of target */ 45190a7c3937SMiklos Szeredi error = -EINVAL; 45200a7c3937SMiklos Szeredi if (old_dentry == trap) 45210a7c3937SMiklos Szeredi goto exit5; 45221da177e4SLinus Torvalds /* target should not be an ancestor of source */ 4523da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 45241da177e4SLinus Torvalds error = -ENOTEMPTY; 45251da177e4SLinus Torvalds if (new_dentry == trap) 45261da177e4SLinus Torvalds goto exit5; 45271da177e4SLinus Torvalds 4528f5beed75SAl Viro error = security_path_rename(&old_path, old_dentry, 4529f5beed75SAl Viro &new_path, new_dentry, flags); 4530be6d3e56SKentaro Takeda if (error) 4531c30dabfeSJan Kara goto exit5; 4532f5beed75SAl Viro error = vfs_rename(old_path.dentry->d_inode, old_dentry, 4533f5beed75SAl Viro new_path.dentry->d_inode, new_dentry, 4534520c8b16SMiklos Szeredi &delegated_inode, flags); 45351da177e4SLinus Torvalds exit5: 45361da177e4SLinus Torvalds dput(new_dentry); 45371da177e4SLinus Torvalds exit4: 45381da177e4SLinus Torvalds dput(old_dentry); 45391da177e4SLinus Torvalds exit3: 4540f5beed75SAl Viro unlock_rename(new_path.dentry, old_path.dentry); 45418e6d782cSJ. Bruce Fields if (delegated_inode) { 45428e6d782cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 45438e6d782cSJ. Bruce Fields if (!error) 45448e6d782cSJ. Bruce Fields goto retry_deleg; 45458e6d782cSJ. Bruce Fields } 4546f5beed75SAl Viro mnt_drop_write(old_path.mnt); 45471da177e4SLinus Torvalds exit2: 4548c6a94284SJeff Layton if (retry_estale(error, lookup_flags)) 4549c6a94284SJeff Layton should_retry = true; 4550f5beed75SAl Viro path_put(&new_path); 45512ad94ae6SAl Viro putname(to); 45521da177e4SLinus Torvalds exit1: 4553f5beed75SAl Viro path_put(&old_path); 45541da177e4SLinus Torvalds putname(from); 4555c6a94284SJeff Layton if (should_retry) { 4556c6a94284SJeff Layton should_retry = false; 4557c6a94284SJeff Layton lookup_flags |= LOOKUP_REVAL; 4558c6a94284SJeff Layton goto retry; 4559c6a94284SJeff Layton } 45602ad94ae6SAl Viro exit: 45611da177e4SLinus Torvalds return error; 45621da177e4SLinus Torvalds } 45631da177e4SLinus Torvalds 4564520c8b16SMiklos Szeredi SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, 4565520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname) 4566520c8b16SMiklos Szeredi { 4567520c8b16SMiklos Szeredi return sys_renameat2(olddfd, oldname, newdfd, newname, 0); 4568520c8b16SMiklos Szeredi } 4569520c8b16SMiklos Szeredi 4570a26eab24SHeiko Carstens SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) 45715590ff0dSUlrich Drepper { 4572520c8b16SMiklos Szeredi return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 45735590ff0dSUlrich Drepper } 45745590ff0dSUlrich Drepper 4575787fb6bcSMiklos Szeredi int vfs_whiteout(struct inode *dir, struct dentry *dentry) 4576787fb6bcSMiklos Szeredi { 4577787fb6bcSMiklos Szeredi int error = may_create(dir, dentry); 4578787fb6bcSMiklos Szeredi if (error) 4579787fb6bcSMiklos Szeredi return error; 4580787fb6bcSMiklos Szeredi 4581787fb6bcSMiklos Szeredi if (!dir->i_op->mknod) 4582787fb6bcSMiklos Szeredi return -EPERM; 4583787fb6bcSMiklos Szeredi 4584787fb6bcSMiklos Szeredi return dir->i_op->mknod(dir, dentry, 4585787fb6bcSMiklos Szeredi S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV); 4586787fb6bcSMiklos Szeredi } 4587787fb6bcSMiklos Szeredi EXPORT_SYMBOL(vfs_whiteout); 4588787fb6bcSMiklos Szeredi 45895d826c84SAl Viro int readlink_copy(char __user *buffer, int buflen, const char *link) 45901da177e4SLinus Torvalds { 45915d826c84SAl Viro int len = PTR_ERR(link); 45921da177e4SLinus Torvalds if (IS_ERR(link)) 45931da177e4SLinus Torvalds goto out; 45941da177e4SLinus Torvalds 45951da177e4SLinus Torvalds len = strlen(link); 45961da177e4SLinus Torvalds if (len > (unsigned) buflen) 45971da177e4SLinus Torvalds len = buflen; 45981da177e4SLinus Torvalds if (copy_to_user(buffer, link, len)) 45991da177e4SLinus Torvalds len = -EFAULT; 46001da177e4SLinus Torvalds out: 46011da177e4SLinus Torvalds return len; 46021da177e4SLinus Torvalds } 46031da177e4SLinus Torvalds 46041da177e4SLinus Torvalds /* 46051da177e4SLinus Torvalds * A helper for ->readlink(). This should be used *ONLY* for symlinks that 46066b255391SAl Viro * have ->get_link() not calling nd_jump_link(). Using (or not using) it 46076b255391SAl Viro * for any given inode is up to filesystem. 46081da177e4SLinus Torvalds */ 4609d16744ecSMiklos Szeredi static int generic_readlink(struct dentry *dentry, char __user *buffer, 4610d16744ecSMiklos Szeredi int buflen) 46111da177e4SLinus Torvalds { 4612fceef393SAl Viro DEFINE_DELAYED_CALL(done); 46135f2c4179SAl Viro struct inode *inode = d_inode(dentry); 46145f2c4179SAl Viro const char *link = inode->i_link; 4615694a1764SMarcin Slusarz int res; 4616cc314eefSLinus Torvalds 4617d4dee48bSAl Viro if (!link) { 4618fceef393SAl Viro link = inode->i_op->get_link(dentry, inode, &done); 4619680baacbSAl Viro if (IS_ERR(link)) 4620680baacbSAl Viro return PTR_ERR(link); 4621d4dee48bSAl Viro } 4622680baacbSAl Viro res = readlink_copy(buffer, buflen, link); 4623fceef393SAl Viro do_delayed_call(&done); 4624694a1764SMarcin Slusarz return res; 46251da177e4SLinus Torvalds } 46261da177e4SLinus Torvalds 4627d60874cdSMiklos Szeredi /** 4628fd4a0edfSMiklos Szeredi * vfs_readlink - copy symlink body into userspace buffer 4629fd4a0edfSMiklos Szeredi * @dentry: dentry on which to get symbolic link 4630fd4a0edfSMiklos Szeredi * @buffer: user memory pointer 4631fd4a0edfSMiklos Szeredi * @buflen: size of buffer 4632fd4a0edfSMiklos Szeredi * 4633fd4a0edfSMiklos Szeredi * Does not touch atime. That's up to the caller if necessary 4634fd4a0edfSMiklos Szeredi * 4635fd4a0edfSMiklos Szeredi * Does not call security hook. 4636fd4a0edfSMiklos Szeredi */ 4637fd4a0edfSMiklos Szeredi int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen) 4638fd4a0edfSMiklos Szeredi { 4639fd4a0edfSMiklos Szeredi struct inode *inode = d_inode(dentry); 4640fd4a0edfSMiklos Szeredi 464176fca90eSMiklos Szeredi if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) { 464276fca90eSMiklos Szeredi if (unlikely(inode->i_op->readlink)) 464376fca90eSMiklos Szeredi return inode->i_op->readlink(dentry, buffer, buflen); 464476fca90eSMiklos Szeredi 464576fca90eSMiklos Szeredi if (!d_is_symlink(dentry)) 4646fd4a0edfSMiklos Szeredi return -EINVAL; 4647fd4a0edfSMiklos Szeredi 464876fca90eSMiklos Szeredi spin_lock(&inode->i_lock); 464976fca90eSMiklos Szeredi inode->i_opflags |= IOP_DEFAULT_READLINK; 465076fca90eSMiklos Szeredi spin_unlock(&inode->i_lock); 465176fca90eSMiklos Szeredi } 465276fca90eSMiklos Szeredi 465376fca90eSMiklos Szeredi return generic_readlink(dentry, buffer, buflen); 4654fd4a0edfSMiklos Szeredi } 4655fd4a0edfSMiklos Szeredi EXPORT_SYMBOL(vfs_readlink); 46561da177e4SLinus Torvalds 4657d60874cdSMiklos Szeredi /** 4658d60874cdSMiklos Szeredi * vfs_get_link - get symlink body 4659d60874cdSMiklos Szeredi * @dentry: dentry on which to get symbolic link 4660d60874cdSMiklos Szeredi * @done: caller needs to free returned data with this 4661d60874cdSMiklos Szeredi * 4662d60874cdSMiklos Szeredi * Calls security hook and i_op->get_link() on the supplied inode. 4663d60874cdSMiklos Szeredi * 4664d60874cdSMiklos Szeredi * It does not touch atime. That's up to the caller if necessary. 4665d60874cdSMiklos Szeredi * 4666d60874cdSMiklos Szeredi * Does not work on "special" symlinks like /proc/$$/fd/N 4667d60874cdSMiklos Szeredi */ 4668d60874cdSMiklos Szeredi const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done) 4669d60874cdSMiklos Szeredi { 4670d60874cdSMiklos Szeredi const char *res = ERR_PTR(-EINVAL); 4671d60874cdSMiklos Szeredi struct inode *inode = d_inode(dentry); 4672d60874cdSMiklos Szeredi 4673d60874cdSMiklos Szeredi if (d_is_symlink(dentry)) { 4674d60874cdSMiklos Szeredi res = ERR_PTR(security_inode_readlink(dentry)); 4675d60874cdSMiklos Szeredi if (!res) 4676d60874cdSMiklos Szeredi res = inode->i_op->get_link(dentry, inode, done); 4677d60874cdSMiklos Szeredi } 4678d60874cdSMiklos Szeredi return res; 4679d60874cdSMiklos Szeredi } 4680d60874cdSMiklos Szeredi EXPORT_SYMBOL(vfs_get_link); 4681d60874cdSMiklos Szeredi 46821da177e4SLinus Torvalds /* get the link contents into pagecache */ 46836b255391SAl Viro const char *page_get_link(struct dentry *dentry, struct inode *inode, 4684fceef393SAl Viro struct delayed_call *callback) 46851da177e4SLinus Torvalds { 4686ebd09abbSDuane Griffin char *kaddr; 46871da177e4SLinus Torvalds struct page *page; 46886b255391SAl Viro struct address_space *mapping = inode->i_mapping; 46896b255391SAl Viro 4690d3883d4fSAl Viro if (!dentry) { 4691d3883d4fSAl Viro page = find_get_page(mapping, 0); 4692d3883d4fSAl Viro if (!page) 46936b255391SAl Viro return ERR_PTR(-ECHILD); 4694d3883d4fSAl Viro if (!PageUptodate(page)) { 4695d3883d4fSAl Viro put_page(page); 4696d3883d4fSAl Viro return ERR_PTR(-ECHILD); 4697d3883d4fSAl Viro } 4698d3883d4fSAl Viro } else { 4699090d2b18SPekka Enberg page = read_mapping_page(mapping, 0, NULL); 47001da177e4SLinus Torvalds if (IS_ERR(page)) 47016fe6900eSNick Piggin return (char*)page; 4702d3883d4fSAl Viro } 4703fceef393SAl Viro set_delayed_call(callback, page_put_link, page); 470421fc61c7SAl Viro BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM); 470521fc61c7SAl Viro kaddr = page_address(page); 47066b255391SAl Viro nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1); 4707ebd09abbSDuane Griffin return kaddr; 47081da177e4SLinus Torvalds } 47091da177e4SLinus Torvalds 47106b255391SAl Viro EXPORT_SYMBOL(page_get_link); 47111da177e4SLinus Torvalds 4712fceef393SAl Viro void page_put_link(void *arg) 47131da177e4SLinus Torvalds { 4714fceef393SAl Viro put_page(arg); 47151da177e4SLinus Torvalds } 47164d359507SAl Viro EXPORT_SYMBOL(page_put_link); 47171da177e4SLinus Torvalds 4718aa80deabSAl Viro int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 4719aa80deabSAl Viro { 4720fceef393SAl Viro DEFINE_DELAYED_CALL(done); 47216b255391SAl Viro int res = readlink_copy(buffer, buflen, 47226b255391SAl Viro page_get_link(dentry, d_inode(dentry), 4723fceef393SAl Viro &done)); 4724fceef393SAl Viro do_delayed_call(&done); 4725aa80deabSAl Viro return res; 4726aa80deabSAl Viro } 4727aa80deabSAl Viro EXPORT_SYMBOL(page_readlink); 4728aa80deabSAl Viro 472954566b2cSNick Piggin /* 473054566b2cSNick Piggin * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS 473154566b2cSNick Piggin */ 473254566b2cSNick Piggin int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) 47331da177e4SLinus Torvalds { 47341da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 47350adb25d2SKirill Korotaev struct page *page; 4736afddba49SNick Piggin void *fsdata; 4737beb497abSDmitriy Monakhov int err; 473854566b2cSNick Piggin unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE; 473954566b2cSNick Piggin if (nofs) 474054566b2cSNick Piggin flags |= AOP_FLAG_NOFS; 47411da177e4SLinus Torvalds 47427e53cac4SNeilBrown retry: 4743afddba49SNick Piggin err = pagecache_write_begin(NULL, mapping, 0, len-1, 474454566b2cSNick Piggin flags, &page, &fsdata); 47451da177e4SLinus Torvalds if (err) 4746afddba49SNick Piggin goto fail; 4747afddba49SNick Piggin 474821fc61c7SAl Viro memcpy(page_address(page), symname, len-1); 4749afddba49SNick Piggin 4750afddba49SNick Piggin err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, 4751afddba49SNick Piggin page, fsdata); 47521da177e4SLinus Torvalds if (err < 0) 47531da177e4SLinus Torvalds goto fail; 4754afddba49SNick Piggin if (err < len-1) 4755afddba49SNick Piggin goto retry; 4756afddba49SNick Piggin 47571da177e4SLinus Torvalds mark_inode_dirty(inode); 47581da177e4SLinus Torvalds return 0; 47591da177e4SLinus Torvalds fail: 47601da177e4SLinus Torvalds return err; 47611da177e4SLinus Torvalds } 47624d359507SAl Viro EXPORT_SYMBOL(__page_symlink); 47631da177e4SLinus Torvalds 47640adb25d2SKirill Korotaev int page_symlink(struct inode *inode, const char *symname, int len) 47650adb25d2SKirill Korotaev { 47660adb25d2SKirill Korotaev return __page_symlink(inode, symname, len, 4767c62d2555SMichal Hocko !mapping_gfp_constraint(inode->i_mapping, __GFP_FS)); 47680adb25d2SKirill Korotaev } 47694d359507SAl Viro EXPORT_SYMBOL(page_symlink); 47700adb25d2SKirill Korotaev 477192e1d5beSArjan van de Ven const struct inode_operations page_symlink_inode_operations = { 47726b255391SAl Viro .get_link = page_get_link, 47731da177e4SLinus Torvalds }; 47741da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink_inode_operations); 4775