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> 371da177e4SLinus Torvalds #include <asm/uaccess.h> 381da177e4SLinus Torvalds 39e81e3f4dSEric Paris #include "internal.h" 40c7105365SAl Viro #include "mount.h" 41e81e3f4dSEric Paris 421da177e4SLinus Torvalds /* [Feb-1997 T. Schoebel-Theuer] 431da177e4SLinus Torvalds * Fundamental changes in the pathname lookup mechanisms (namei) 441da177e4SLinus Torvalds * were necessary because of omirr. The reason is that omirr needs 451da177e4SLinus Torvalds * to know the _real_ pathname, not the user-supplied one, in case 461da177e4SLinus Torvalds * of symlinks (and also when transname replacements occur). 471da177e4SLinus Torvalds * 481da177e4SLinus Torvalds * The new code replaces the old recursive symlink resolution with 491da177e4SLinus Torvalds * an iterative one (in case of non-nested symlink chains). It does 501da177e4SLinus Torvalds * this with calls to <fs>_follow_link(). 511da177e4SLinus Torvalds * As a side effect, dir_namei(), _namei() and follow_link() are now 521da177e4SLinus Torvalds * replaced with a single function lookup_dentry() that can handle all 531da177e4SLinus Torvalds * the special cases of the former code. 541da177e4SLinus Torvalds * 551da177e4SLinus Torvalds * With the new dcache, the pathname is stored at each inode, at least as 561da177e4SLinus Torvalds * long as the refcount of the inode is positive. As a side effect, the 571da177e4SLinus Torvalds * size of the dcache depends on the inode cache and thus is dynamic. 581da177e4SLinus Torvalds * 591da177e4SLinus Torvalds * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink 601da177e4SLinus Torvalds * resolution to correspond with current state of the code. 611da177e4SLinus Torvalds * 621da177e4SLinus Torvalds * Note that the symlink resolution is not *completely* iterative. 631da177e4SLinus Torvalds * There is still a significant amount of tail- and mid- recursion in 641da177e4SLinus Torvalds * the algorithm. Also, note that <fs>_readlink() is not used in 651da177e4SLinus Torvalds * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink() 661da177e4SLinus Torvalds * may return different results than <fs>_follow_link(). Many virtual 671da177e4SLinus Torvalds * filesystems (including /proc) exhibit this behavior. 681da177e4SLinus Torvalds */ 691da177e4SLinus Torvalds 701da177e4SLinus Torvalds /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation: 711da177e4SLinus Torvalds * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL 721da177e4SLinus Torvalds * and the name already exists in form of a symlink, try to create the new 731da177e4SLinus Torvalds * name indicated by the symlink. The old code always complained that the 741da177e4SLinus Torvalds * name already exists, due to not following the symlink even if its target 751da177e4SLinus Torvalds * is nonexistent. The new semantics affects also mknod() and link() when 7625985edcSLucas De Marchi * the name is a symlink pointing to a non-existent name. 771da177e4SLinus Torvalds * 781da177e4SLinus Torvalds * I don't know which semantics is the right one, since I have no access 791da177e4SLinus Torvalds * to standards. But I found by trial that HP-UX 9.0 has the full "new" 801da177e4SLinus Torvalds * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the 811da177e4SLinus Torvalds * "old" one. Personally, I think the new semantics is much more logical. 821da177e4SLinus Torvalds * Note that "ln old new" where "new" is a symlink pointing to a non-existing 831da177e4SLinus Torvalds * file does succeed in both HP-UX and SunOs, but not in Solaris 841da177e4SLinus Torvalds * and in the old Linux semantics. 851da177e4SLinus Torvalds */ 861da177e4SLinus Torvalds 871da177e4SLinus Torvalds /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink 881da177e4SLinus Torvalds * semantics. See the comments in "open_namei" and "do_link" below. 891da177e4SLinus Torvalds * 901da177e4SLinus Torvalds * [10-Sep-98 Alan Modra] Another symlink change. 911da177e4SLinus Torvalds */ 921da177e4SLinus Torvalds 931da177e4SLinus Torvalds /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks: 941da177e4SLinus Torvalds * inside the path - always follow. 951da177e4SLinus Torvalds * in the last component in creation/removal/renaming - never follow. 961da177e4SLinus Torvalds * if LOOKUP_FOLLOW passed - follow. 971da177e4SLinus Torvalds * if the pathname has trailing slashes - follow. 981da177e4SLinus Torvalds * otherwise - don't follow. 991da177e4SLinus Torvalds * (applied in that order). 1001da177e4SLinus Torvalds * 1011da177e4SLinus Torvalds * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT 1021da177e4SLinus Torvalds * restored for 2.4. This is the last surviving part of old 4.2BSD bug. 1031da177e4SLinus Torvalds * During the 2.4 we need to fix the userland stuff depending on it - 1041da177e4SLinus Torvalds * hopefully we will be able to get rid of that wart in 2.5. So far only 1051da177e4SLinus Torvalds * XEmacs seems to be relying on it... 1061da177e4SLinus Torvalds */ 1071da177e4SLinus Torvalds /* 1081da177e4SLinus Torvalds * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland) 109a11f3a05SArjan van de Ven * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives 1101da177e4SLinus Torvalds * any extra contention... 1111da177e4SLinus Torvalds */ 1121da177e4SLinus Torvalds 1131da177e4SLinus Torvalds /* In order to reduce some races, while at the same time doing additional 1141da177e4SLinus Torvalds * checking and hopefully speeding things up, we copy filenames to the 1151da177e4SLinus Torvalds * kernel data space before using them.. 1161da177e4SLinus Torvalds * 1171da177e4SLinus Torvalds * POSIX.1 2.4: an empty pathname is invalid (ENOENT). 1181da177e4SLinus Torvalds * PATH_MAX includes the nul terminator --RR. 1191da177e4SLinus Torvalds */ 12091a27b2aSJeff Layton void final_putname(struct filename *name) 1211da177e4SLinus Torvalds { 1227950e385SJeff Layton if (name->separate) { 12391a27b2aSJeff Layton __putname(name->name); 12491a27b2aSJeff Layton kfree(name); 1257950e385SJeff Layton } else { 1267950e385SJeff Layton __putname(name); 12791a27b2aSJeff Layton } 1287950e385SJeff Layton } 1297950e385SJeff Layton 1307950e385SJeff Layton #define EMBEDDED_NAME_MAX (PATH_MAX - sizeof(struct filename)) 13191a27b2aSJeff Layton 13291a27b2aSJeff Layton static struct filename * 13391a27b2aSJeff Layton getname_flags(const char __user *filename, int flags, int *empty) 13491a27b2aSJeff Layton { 13591a27b2aSJeff Layton struct filename *result, *err; 1363f9f0aa6SLinus Torvalds int len; 1377950e385SJeff Layton long max; 1387950e385SJeff Layton char *kname; 1391da177e4SLinus Torvalds 1407ac86265SJeff Layton result = audit_reusename(filename); 1417ac86265SJeff Layton if (result) 1427ac86265SJeff Layton return result; 1437ac86265SJeff Layton 1447950e385SJeff Layton result = __getname(); 1453f9f0aa6SLinus Torvalds if (unlikely(!result)) 1464043cde8SEric Paris return ERR_PTR(-ENOMEM); 1471da177e4SLinus Torvalds 1487950e385SJeff Layton /* 1497950e385SJeff Layton * First, try to embed the struct filename inside the names_cache 1507950e385SJeff Layton * allocation 1517950e385SJeff Layton */ 1527950e385SJeff Layton kname = (char *)result + sizeof(*result); 15391a27b2aSJeff Layton result->name = kname; 1547950e385SJeff Layton result->separate = false; 1557950e385SJeff Layton max = EMBEDDED_NAME_MAX; 1567950e385SJeff Layton 1577950e385SJeff Layton recopy: 1587950e385SJeff Layton len = strncpy_from_user(kname, filename, max); 15991a27b2aSJeff Layton if (unlikely(len < 0)) { 1603f9f0aa6SLinus Torvalds err = ERR_PTR(len); 1613f9f0aa6SLinus Torvalds goto error; 16291a27b2aSJeff Layton } 1633f9f0aa6SLinus Torvalds 1647950e385SJeff Layton /* 1657950e385SJeff Layton * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a 1667950e385SJeff Layton * separate struct filename so we can dedicate the entire 1677950e385SJeff Layton * names_cache allocation for the pathname, and re-do the copy from 1687950e385SJeff Layton * userland. 1697950e385SJeff Layton */ 1707950e385SJeff Layton if (len == EMBEDDED_NAME_MAX && max == EMBEDDED_NAME_MAX) { 1717950e385SJeff Layton kname = (char *)result; 1727950e385SJeff Layton 1737950e385SJeff Layton result = kzalloc(sizeof(*result), GFP_KERNEL); 1747950e385SJeff Layton if (!result) { 1757950e385SJeff Layton err = ERR_PTR(-ENOMEM); 1767950e385SJeff Layton result = (struct filename *)kname; 1777950e385SJeff Layton goto error; 1787950e385SJeff Layton } 1797950e385SJeff Layton result->name = kname; 1807950e385SJeff Layton result->separate = true; 1817950e385SJeff Layton max = PATH_MAX; 1827950e385SJeff Layton goto recopy; 1837950e385SJeff Layton } 1847950e385SJeff Layton 1853f9f0aa6SLinus Torvalds /* The empty path is special. */ 1863f9f0aa6SLinus Torvalds if (unlikely(!len)) { 1873f9f0aa6SLinus Torvalds if (empty) 1881fa1e7f6SAndy Whitcroft *empty = 1; 1893f9f0aa6SLinus Torvalds err = ERR_PTR(-ENOENT); 1903f9f0aa6SLinus Torvalds if (!(flags & LOOKUP_EMPTY)) 1913f9f0aa6SLinus Torvalds goto error; 1921da177e4SLinus Torvalds } 1933f9f0aa6SLinus Torvalds 1943f9f0aa6SLinus Torvalds err = ERR_PTR(-ENAMETOOLONG); 1957950e385SJeff Layton if (unlikely(len >= PATH_MAX)) 1967950e385SJeff Layton goto error; 1977950e385SJeff Layton 1987950e385SJeff Layton result->uptr = filename; 1991da177e4SLinus Torvalds audit_getname(result); 2001da177e4SLinus Torvalds return result; 2011da177e4SLinus Torvalds 2023f9f0aa6SLinus Torvalds error: 2037950e385SJeff Layton final_putname(result); 2043f9f0aa6SLinus Torvalds return err; 2053f9f0aa6SLinus Torvalds } 2063f9f0aa6SLinus Torvalds 20791a27b2aSJeff Layton struct filename * 20891a27b2aSJeff Layton getname(const char __user * filename) 209f52e0c11SAl Viro { 210f7493e5dSLinus Torvalds return getname_flags(filename, 0, NULL); 211f52e0c11SAl Viro } 21291a27b2aSJeff Layton EXPORT_SYMBOL(getname); 213f52e0c11SAl Viro 2141da177e4SLinus Torvalds #ifdef CONFIG_AUDITSYSCALL 21591a27b2aSJeff Layton void putname(struct filename *name) 2161da177e4SLinus Torvalds { 2175ac3a9c2SAl Viro if (unlikely(!audit_dummy_context())) 21891a27b2aSJeff Layton return audit_putname(name); 21991a27b2aSJeff Layton final_putname(name); 2201da177e4SLinus Torvalds } 2211da177e4SLinus Torvalds #endif 2221da177e4SLinus Torvalds 223e77819e5SLinus Torvalds static int check_acl(struct inode *inode, int mask) 224e77819e5SLinus Torvalds { 22584635d68SLinus Torvalds #ifdef CONFIG_FS_POSIX_ACL 226e77819e5SLinus Torvalds struct posix_acl *acl; 227e77819e5SLinus Torvalds 228e77819e5SLinus Torvalds if (mask & MAY_NOT_BLOCK) { 2293567866bSAl Viro acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS); 2303567866bSAl Viro if (!acl) 231e77819e5SLinus Torvalds return -EAGAIN; 2323567866bSAl Viro /* no ->get_acl() calls in RCU mode... */ 2333567866bSAl Viro if (acl == ACL_NOT_CACHED) 234e77819e5SLinus Torvalds return -ECHILD; 235206b1d09SAri Savolainen return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK); 236e77819e5SLinus Torvalds } 237e77819e5SLinus Torvalds 238e77819e5SLinus Torvalds acl = get_cached_acl(inode, ACL_TYPE_ACCESS); 239e77819e5SLinus Torvalds 240e77819e5SLinus Torvalds /* 2414e34e719SChristoph Hellwig * A filesystem can force a ACL callback by just never filling the 2424e34e719SChristoph Hellwig * ACL cache. But normally you'd fill the cache either at inode 2434e34e719SChristoph Hellwig * instantiation time, or on the first ->get_acl call. 244e77819e5SLinus Torvalds * 2454e34e719SChristoph Hellwig * If the filesystem doesn't have a get_acl() function at all, we'll 2464e34e719SChristoph Hellwig * just create the negative cache entry. 247e77819e5SLinus Torvalds */ 248e77819e5SLinus Torvalds if (acl == ACL_NOT_CACHED) { 2494e34e719SChristoph Hellwig if (inode->i_op->get_acl) { 2504e34e719SChristoph Hellwig acl = inode->i_op->get_acl(inode, ACL_TYPE_ACCESS); 2514e34e719SChristoph Hellwig if (IS_ERR(acl)) 2524e34e719SChristoph Hellwig return PTR_ERR(acl); 2534e34e719SChristoph Hellwig } else { 254e77819e5SLinus Torvalds set_cached_acl(inode, ACL_TYPE_ACCESS, NULL); 255e77819e5SLinus Torvalds return -EAGAIN; 256e77819e5SLinus Torvalds } 2574e34e719SChristoph Hellwig } 258e77819e5SLinus Torvalds 259e77819e5SLinus Torvalds if (acl) { 260e77819e5SLinus Torvalds int error = posix_acl_permission(inode, acl, mask); 261e77819e5SLinus Torvalds posix_acl_release(acl); 262e77819e5SLinus Torvalds return error; 263e77819e5SLinus Torvalds } 26484635d68SLinus Torvalds #endif 265e77819e5SLinus Torvalds 266e77819e5SLinus Torvalds return -EAGAIN; 267e77819e5SLinus Torvalds } 268e77819e5SLinus Torvalds 2695909ccaaSLinus Torvalds /* 270948409c7SAndreas Gruenbacher * This does the basic permission checking 2715909ccaaSLinus Torvalds */ 2727e40145eSAl Viro static int acl_permission_check(struct inode *inode, int mask) 2735909ccaaSLinus Torvalds { 27426cf46beSLinus Torvalds unsigned int mode = inode->i_mode; 2755909ccaaSLinus Torvalds 2768e96e3b7SEric W. Biederman if (likely(uid_eq(current_fsuid(), inode->i_uid))) 2775909ccaaSLinus Torvalds mode >>= 6; 2785909ccaaSLinus Torvalds else { 279e77819e5SLinus Torvalds if (IS_POSIXACL(inode) && (mode & S_IRWXG)) { 2807e40145eSAl Viro int error = check_acl(inode, mask); 2815909ccaaSLinus Torvalds if (error != -EAGAIN) 2825909ccaaSLinus Torvalds return error; 2835909ccaaSLinus Torvalds } 2845909ccaaSLinus Torvalds 2855909ccaaSLinus Torvalds if (in_group_p(inode->i_gid)) 2865909ccaaSLinus Torvalds mode >>= 3; 2875909ccaaSLinus Torvalds } 2885909ccaaSLinus Torvalds 2895909ccaaSLinus Torvalds /* 2905909ccaaSLinus Torvalds * If the DACs are ok we don't need any capability check. 2915909ccaaSLinus Torvalds */ 2929c2c7039SAl Viro if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) 2935909ccaaSLinus Torvalds return 0; 2945909ccaaSLinus Torvalds return -EACCES; 2955909ccaaSLinus Torvalds } 2961da177e4SLinus Torvalds 2971da177e4SLinus Torvalds /** 2981da177e4SLinus Torvalds * generic_permission - check for access rights on a Posix-like filesystem 2991da177e4SLinus Torvalds * @inode: inode to check access rights for 3008fd90c8dSAndreas Gruenbacher * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...) 3011da177e4SLinus Torvalds * 3021da177e4SLinus Torvalds * Used to check for read/write/execute permissions on a file. 3031da177e4SLinus Torvalds * We use "fsuid" for this, letting us set arbitrary permissions 3041da177e4SLinus Torvalds * for filesystem access without changing the "normal" uids which 305b74c79e9SNick Piggin * are used for other things. 306b74c79e9SNick Piggin * 307b74c79e9SNick Piggin * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk 308b74c79e9SNick Piggin * request cannot be satisfied (eg. requires blocking or too much complexity). 309b74c79e9SNick Piggin * It would then be called again in ref-walk mode. 3101da177e4SLinus Torvalds */ 3112830ba7fSAl Viro int generic_permission(struct inode *inode, int mask) 3121da177e4SLinus Torvalds { 3135909ccaaSLinus Torvalds int ret; 3141da177e4SLinus Torvalds 3151da177e4SLinus Torvalds /* 316948409c7SAndreas Gruenbacher * Do the basic permission checks. 3171da177e4SLinus Torvalds */ 3187e40145eSAl Viro ret = acl_permission_check(inode, mask); 3195909ccaaSLinus Torvalds if (ret != -EACCES) 3205909ccaaSLinus Torvalds return ret; 3211da177e4SLinus Torvalds 322d594e7ecSAl Viro if (S_ISDIR(inode->i_mode)) { 323d594e7ecSAl Viro /* DACs are overridable for directories */ 3241a48e2acSEric W. Biederman if (inode_capable(inode, CAP_DAC_OVERRIDE)) 325d594e7ecSAl Viro return 0; 326d594e7ecSAl Viro if (!(mask & MAY_WRITE)) 3271a48e2acSEric W. Biederman if (inode_capable(inode, CAP_DAC_READ_SEARCH)) 328d594e7ecSAl Viro return 0; 329d594e7ecSAl Viro return -EACCES; 330d594e7ecSAl Viro } 3311da177e4SLinus Torvalds /* 3321da177e4SLinus Torvalds * Read/write DACs are always overridable. 333d594e7ecSAl Viro * Executable DACs are overridable when there is 334d594e7ecSAl Viro * at least one exec bit set. 3351da177e4SLinus Torvalds */ 336d594e7ecSAl Viro if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO)) 3371a48e2acSEric W. Biederman if (inode_capable(inode, CAP_DAC_OVERRIDE)) 3381da177e4SLinus Torvalds return 0; 3391da177e4SLinus Torvalds 3401da177e4SLinus Torvalds /* 3411da177e4SLinus Torvalds * Searching includes executable on directories, else just read. 3421da177e4SLinus Torvalds */ 3437ea66001SSerge E. Hallyn mask &= MAY_READ | MAY_WRITE | MAY_EXEC; 344d594e7ecSAl Viro if (mask == MAY_READ) 3451a48e2acSEric W. Biederman if (inode_capable(inode, CAP_DAC_READ_SEARCH)) 3461da177e4SLinus Torvalds return 0; 3471da177e4SLinus Torvalds 3481da177e4SLinus Torvalds return -EACCES; 3491da177e4SLinus Torvalds } 3501da177e4SLinus Torvalds 3513ddcd056SLinus Torvalds /* 3523ddcd056SLinus Torvalds * We _really_ want to just do "generic_permission()" without 3533ddcd056SLinus Torvalds * even looking at the inode->i_op values. So we keep a cache 3543ddcd056SLinus Torvalds * flag in inode->i_opflags, that says "this has not special 3553ddcd056SLinus Torvalds * permission function, use the fast case". 3563ddcd056SLinus Torvalds */ 3573ddcd056SLinus Torvalds static inline int do_inode_permission(struct inode *inode, int mask) 3583ddcd056SLinus Torvalds { 3593ddcd056SLinus Torvalds if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) { 3603ddcd056SLinus Torvalds if (likely(inode->i_op->permission)) 3613ddcd056SLinus Torvalds return inode->i_op->permission(inode, mask); 3623ddcd056SLinus Torvalds 3633ddcd056SLinus Torvalds /* This gets set once for the inode lifetime */ 3643ddcd056SLinus Torvalds spin_lock(&inode->i_lock); 3653ddcd056SLinus Torvalds inode->i_opflags |= IOP_FASTPERM; 3663ddcd056SLinus Torvalds spin_unlock(&inode->i_lock); 3673ddcd056SLinus Torvalds } 3683ddcd056SLinus Torvalds return generic_permission(inode, mask); 3693ddcd056SLinus Torvalds } 3703ddcd056SLinus Torvalds 371cb23beb5SChristoph Hellwig /** 3720bdaea90SDavid Howells * __inode_permission - Check for access rights to a given inode 3730bdaea90SDavid Howells * @inode: Inode to check permission on 3740bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 375cb23beb5SChristoph Hellwig * 3760bdaea90SDavid Howells * Check for read/write/execute permissions on an inode. 377948409c7SAndreas Gruenbacher * 378948409c7SAndreas Gruenbacher * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. 3790bdaea90SDavid Howells * 3800bdaea90SDavid Howells * This does not check for a read-only file system. You probably want 3810bdaea90SDavid Howells * inode_permission(). 382cb23beb5SChristoph Hellwig */ 3830bdaea90SDavid Howells int __inode_permission(struct inode *inode, int mask) 3841da177e4SLinus Torvalds { 385e6305c43SAl Viro int retval; 3861da177e4SLinus Torvalds 3873ddcd056SLinus Torvalds if (unlikely(mask & MAY_WRITE)) { 3881da177e4SLinus Torvalds /* 3891da177e4SLinus Torvalds * Nobody gets write access to an immutable file. 3901da177e4SLinus Torvalds */ 3911da177e4SLinus Torvalds if (IS_IMMUTABLE(inode)) 3921da177e4SLinus Torvalds return -EACCES; 3931da177e4SLinus Torvalds } 3941da177e4SLinus Torvalds 3953ddcd056SLinus Torvalds retval = do_inode_permission(inode, mask); 3961da177e4SLinus Torvalds if (retval) 3971da177e4SLinus Torvalds return retval; 3981da177e4SLinus Torvalds 39908ce5f16SSerge E. Hallyn retval = devcgroup_inode_permission(inode, mask); 40008ce5f16SSerge E. Hallyn if (retval) 40108ce5f16SSerge E. Hallyn return retval; 40208ce5f16SSerge E. Hallyn 403d09ca739SEric Paris return security_inode_permission(inode, mask); 4041da177e4SLinus Torvalds } 4051da177e4SLinus Torvalds 406f4d6ff89SAl Viro /** 4070bdaea90SDavid Howells * sb_permission - Check superblock-level permissions 4080bdaea90SDavid Howells * @sb: Superblock of inode to check permission on 40955852635SRandy Dunlap * @inode: Inode to check permission on 4100bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 4110bdaea90SDavid Howells * 4120bdaea90SDavid Howells * Separate out file-system wide checks from inode-specific permission checks. 4130bdaea90SDavid Howells */ 4140bdaea90SDavid Howells static int sb_permission(struct super_block *sb, struct inode *inode, int mask) 4150bdaea90SDavid Howells { 4160bdaea90SDavid Howells if (unlikely(mask & MAY_WRITE)) { 4170bdaea90SDavid Howells umode_t mode = inode->i_mode; 4180bdaea90SDavid Howells 4190bdaea90SDavid Howells /* Nobody gets write access to a read-only fs. */ 4200bdaea90SDavid Howells if ((sb->s_flags & MS_RDONLY) && 4210bdaea90SDavid Howells (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) 4220bdaea90SDavid Howells return -EROFS; 4230bdaea90SDavid Howells } 4240bdaea90SDavid Howells return 0; 4250bdaea90SDavid Howells } 4260bdaea90SDavid Howells 4270bdaea90SDavid Howells /** 4280bdaea90SDavid Howells * inode_permission - Check for access rights to a given inode 4290bdaea90SDavid Howells * @inode: Inode to check permission on 4300bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 4310bdaea90SDavid Howells * 4320bdaea90SDavid Howells * Check for read/write/execute permissions on an inode. We use fs[ug]id for 4330bdaea90SDavid Howells * this, letting us set arbitrary permissions for filesystem access without 4340bdaea90SDavid Howells * changing the "normal" UIDs which are used for other things. 4350bdaea90SDavid Howells * 4360bdaea90SDavid Howells * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. 4370bdaea90SDavid Howells */ 4380bdaea90SDavid Howells int inode_permission(struct inode *inode, int mask) 4390bdaea90SDavid Howells { 4400bdaea90SDavid Howells int retval; 4410bdaea90SDavid Howells 4420bdaea90SDavid Howells retval = sb_permission(inode->i_sb, inode, mask); 4430bdaea90SDavid Howells if (retval) 4440bdaea90SDavid Howells return retval; 4450bdaea90SDavid Howells return __inode_permission(inode, mask); 4460bdaea90SDavid Howells } 4470bdaea90SDavid Howells 4480bdaea90SDavid Howells /** 4495dd784d0SJan Blunck * path_get - get a reference to a path 4505dd784d0SJan Blunck * @path: path to get the reference to 4515dd784d0SJan Blunck * 4525dd784d0SJan Blunck * Given a path increment the reference count to the dentry and the vfsmount. 4535dd784d0SJan Blunck */ 454dcf787f3SAl Viro void path_get(const struct path *path) 4555dd784d0SJan Blunck { 4565dd784d0SJan Blunck mntget(path->mnt); 4575dd784d0SJan Blunck dget(path->dentry); 4585dd784d0SJan Blunck } 4595dd784d0SJan Blunck EXPORT_SYMBOL(path_get); 4605dd784d0SJan Blunck 4615dd784d0SJan Blunck /** 4621d957f9bSJan Blunck * path_put - put a reference to a path 4631d957f9bSJan Blunck * @path: path to put the reference to 4641d957f9bSJan Blunck * 4651d957f9bSJan Blunck * Given a path decrement the reference count to the dentry and the vfsmount. 4661d957f9bSJan Blunck */ 467dcf787f3SAl Viro void path_put(const struct path *path) 4681da177e4SLinus Torvalds { 4691d957f9bSJan Blunck dput(path->dentry); 4701d957f9bSJan Blunck mntput(path->mnt); 4711da177e4SLinus Torvalds } 4721d957f9bSJan Blunck EXPORT_SYMBOL(path_put); 4731da177e4SLinus Torvalds 47419660af7SAl Viro /* 47531e6b01fSNick Piggin * Path walking has 2 modes, rcu-walk and ref-walk (see 47619660af7SAl Viro * Documentation/filesystems/path-lookup.txt). In situations when we can't 47719660af7SAl Viro * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab 47819660af7SAl Viro * normal reference counts on dentries and vfsmounts to transition to rcu-walk 47919660af7SAl Viro * mode. Refcounts are grabbed at the last known good point before rcu-walk 48019660af7SAl Viro * got stuck, so ref-walk may continue from there. If this is not successful 48119660af7SAl Viro * (eg. a seqcount has changed), then failure is returned and it's up to caller 48219660af7SAl Viro * to restart the path walk from the beginning in ref-walk mode. 48331e6b01fSNick Piggin */ 48431e6b01fSNick Piggin 48532a7991bSAl Viro static inline void lock_rcu_walk(void) 48632a7991bSAl Viro { 48732a7991bSAl Viro br_read_lock(&vfsmount_lock); 48832a7991bSAl Viro rcu_read_lock(); 48932a7991bSAl Viro } 49032a7991bSAl Viro 49132a7991bSAl Viro static inline void unlock_rcu_walk(void) 49232a7991bSAl Viro { 49332a7991bSAl Viro rcu_read_unlock(); 49432a7991bSAl Viro br_read_unlock(&vfsmount_lock); 49532a7991bSAl Viro } 49632a7991bSAl Viro 49731e6b01fSNick Piggin /** 49819660af7SAl Viro * unlazy_walk - try to switch to ref-walk mode. 49919660af7SAl Viro * @nd: nameidata pathwalk data 50019660af7SAl Viro * @dentry: child of nd->path.dentry or NULL 50139191628SRandy Dunlap * Returns: 0 on success, -ECHILD on failure 50231e6b01fSNick Piggin * 50319660af7SAl Viro * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry 50419660af7SAl Viro * for ref-walk mode. @dentry must be a path found by a do_lookup call on 50519660af7SAl Viro * @nd or NULL. Must be called from rcu-walk context. 50631e6b01fSNick Piggin */ 50719660af7SAl Viro static int unlazy_walk(struct nameidata *nd, struct dentry *dentry) 50831e6b01fSNick Piggin { 50931e6b01fSNick Piggin struct fs_struct *fs = current->fs; 51031e6b01fSNick Piggin struct dentry *parent = nd->path.dentry; 51131e6b01fSNick Piggin 51231e6b01fSNick Piggin BUG_ON(!(nd->flags & LOOKUP_RCU)); 513e5c832d5SLinus Torvalds 514e5c832d5SLinus Torvalds /* 515e5c832d5SLinus Torvalds * Get a reference to the parent first: we're 516e5c832d5SLinus Torvalds * going to make "path_put(nd->path)" valid in 517e5c832d5SLinus Torvalds * non-RCU context for "terminate_walk()". 518e5c832d5SLinus Torvalds * 519e5c832d5SLinus Torvalds * If this doesn't work, return immediately with 520e5c832d5SLinus Torvalds * RCU walking still active (and then we will do 521e5c832d5SLinus Torvalds * the RCU walk cleanup in terminate_walk()). 522e5c832d5SLinus Torvalds */ 523e5c832d5SLinus Torvalds if (!lockref_get_not_dead(&parent->d_lockref)) 524e5c832d5SLinus Torvalds return -ECHILD; 525e5c832d5SLinus Torvalds 526e5c832d5SLinus Torvalds /* 527e5c832d5SLinus Torvalds * After the mntget(), we terminate_walk() will do 528e5c832d5SLinus Torvalds * the right thing for non-RCU mode, and all our 529e5c832d5SLinus Torvalds * subsequent exit cases should unlock_rcu_walk() 530e5c832d5SLinus Torvalds * before returning. 531e5c832d5SLinus Torvalds */ 532e5c832d5SLinus Torvalds mntget(nd->path.mnt); 533e5c832d5SLinus Torvalds nd->flags &= ~LOOKUP_RCU; 53415570086SLinus Torvalds 53515570086SLinus Torvalds /* 53615570086SLinus Torvalds * For a negative lookup, the lookup sequence point is the parents 53715570086SLinus Torvalds * sequence point, and it only needs to revalidate the parent dentry. 53815570086SLinus Torvalds * 53915570086SLinus Torvalds * For a positive lookup, we need to move both the parent and the 54015570086SLinus Torvalds * dentry from the RCU domain to be properly refcounted. And the 54115570086SLinus Torvalds * sequence number in the dentry validates *both* dentry counters, 54215570086SLinus Torvalds * since we checked the sequence number of the parent after we got 54315570086SLinus Torvalds * the child sequence number. So we know the parent must still 54415570086SLinus Torvalds * be valid if the child sequence number is still valid. 54515570086SLinus Torvalds */ 54619660af7SAl Viro if (!dentry) { 547e5c832d5SLinus Torvalds if (read_seqcount_retry(&parent->d_seq, nd->seq)) 548e5c832d5SLinus Torvalds goto out; 54919660af7SAl Viro BUG_ON(nd->inode != parent->d_inode); 55019660af7SAl Viro } else { 551e5c832d5SLinus Torvalds if (!lockref_get_not_dead(&dentry->d_lockref)) 552e5c832d5SLinus Torvalds goto out; 553e5c832d5SLinus Torvalds if (read_seqcount_retry(&dentry->d_seq, nd->seq)) 554e5c832d5SLinus Torvalds goto drop_dentry; 55519660af7SAl Viro } 556e5c832d5SLinus Torvalds 557e5c832d5SLinus Torvalds /* 558e5c832d5SLinus Torvalds * Sequence counts matched. Now make sure that the root is 559e5c832d5SLinus Torvalds * still valid and get it if required. 560e5c832d5SLinus Torvalds */ 561e5c832d5SLinus Torvalds if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 562e5c832d5SLinus Torvalds spin_lock(&fs->lock); 563e5c832d5SLinus Torvalds if (nd->root.mnt != fs->root.mnt || nd->root.dentry != fs->root.dentry) 564e5c832d5SLinus Torvalds goto unlock_and_drop_dentry; 56531e6b01fSNick Piggin path_get(&nd->root); 56631e6b01fSNick Piggin spin_unlock(&fs->lock); 56731e6b01fSNick Piggin } 56831e6b01fSNick Piggin 56932a7991bSAl Viro unlock_rcu_walk(); 57031e6b01fSNick Piggin return 0; 57119660af7SAl Viro 572e5c832d5SLinus Torvalds unlock_and_drop_dentry: 57331e6b01fSNick Piggin spin_unlock(&fs->lock); 574e5c832d5SLinus Torvalds drop_dentry: 575e5c832d5SLinus Torvalds unlock_rcu_walk(); 576e5c832d5SLinus Torvalds dput(dentry); 577d0d27277SLinus Torvalds goto drop_root_mnt; 578e5c832d5SLinus Torvalds out: 579e5c832d5SLinus Torvalds unlock_rcu_walk(); 580d0d27277SLinus Torvalds drop_root_mnt: 581d0d27277SLinus Torvalds if (!(nd->flags & LOOKUP_ROOT)) 582d0d27277SLinus Torvalds nd->root.mnt = NULL; 58331e6b01fSNick Piggin return -ECHILD; 58431e6b01fSNick Piggin } 58531e6b01fSNick Piggin 5864ce16ef3SAl Viro static inline int d_revalidate(struct dentry *dentry, unsigned int flags) 58734286d66SNick Piggin { 5884ce16ef3SAl Viro return dentry->d_op->d_revalidate(dentry, flags); 58934286d66SNick Piggin } 59034286d66SNick Piggin 5919f1fafeeSAl Viro /** 5929f1fafeeSAl Viro * complete_walk - successful completion of path walk 5939f1fafeeSAl Viro * @nd: pointer nameidata 59439159de2SJeff Layton * 5959f1fafeeSAl Viro * If we had been in RCU mode, drop out of it and legitimize nd->path. 5969f1fafeeSAl Viro * Revalidate the final result, unless we'd already done that during 5979f1fafeeSAl Viro * the path walk or the filesystem doesn't ask for it. Return 0 on 5989f1fafeeSAl Viro * success, -error on failure. In case of failure caller does not 5999f1fafeeSAl Viro * need to drop nd->path. 60039159de2SJeff Layton */ 6019f1fafeeSAl Viro static int complete_walk(struct nameidata *nd) 60239159de2SJeff Layton { 60316c2cd71SAl Viro struct dentry *dentry = nd->path.dentry; 60439159de2SJeff Layton int status; 60539159de2SJeff Layton 6069f1fafeeSAl Viro if (nd->flags & LOOKUP_RCU) { 6079f1fafeeSAl Viro nd->flags &= ~LOOKUP_RCU; 6089f1fafeeSAl Viro if (!(nd->flags & LOOKUP_ROOT)) 6099f1fafeeSAl Viro nd->root.mnt = NULL; 61015570086SLinus Torvalds 611e5c832d5SLinus Torvalds if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) { 61232a7991bSAl Viro unlock_rcu_walk(); 6139f1fafeeSAl Viro return -ECHILD; 6149f1fafeeSAl Viro } 615e5c832d5SLinus Torvalds if (read_seqcount_retry(&dentry->d_seq, nd->seq)) { 616e5c832d5SLinus Torvalds unlock_rcu_walk(); 617e5c832d5SLinus Torvalds dput(dentry); 618e5c832d5SLinus Torvalds return -ECHILD; 619e5c832d5SLinus Torvalds } 6209f1fafeeSAl Viro mntget(nd->path.mnt); 62132a7991bSAl Viro unlock_rcu_walk(); 6229f1fafeeSAl Viro } 6239f1fafeeSAl Viro 62416c2cd71SAl Viro if (likely(!(nd->flags & LOOKUP_JUMPED))) 62539159de2SJeff Layton return 0; 62639159de2SJeff Layton 627ecf3d1f1SJeff Layton if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) 62816c2cd71SAl Viro return 0; 62916c2cd71SAl Viro 630ecf3d1f1SJeff Layton status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); 63139159de2SJeff Layton if (status > 0) 63239159de2SJeff Layton return 0; 63339159de2SJeff Layton 63416c2cd71SAl Viro if (!status) 63539159de2SJeff Layton status = -ESTALE; 63616c2cd71SAl Viro 6379f1fafeeSAl Viro path_put(&nd->path); 63839159de2SJeff Layton return status; 63939159de2SJeff Layton } 64039159de2SJeff Layton 6412a737871SAl Viro static __always_inline void set_root(struct nameidata *nd) 6422a737871SAl Viro { 643f7ad3c6bSMiklos Szeredi if (!nd->root.mnt) 644f7ad3c6bSMiklos Szeredi get_fs_root(current->fs, &nd->root); 6452a737871SAl Viro } 6462a737871SAl Viro 6476de88d72SAl Viro static int link_path_walk(const char *, struct nameidata *); 6486de88d72SAl Viro 64931e6b01fSNick Piggin static __always_inline void set_root_rcu(struct nameidata *nd) 65031e6b01fSNick Piggin { 65131e6b01fSNick Piggin if (!nd->root.mnt) { 65231e6b01fSNick Piggin struct fs_struct *fs = current->fs; 653c28cc364SNick Piggin unsigned seq; 654c28cc364SNick Piggin 655c28cc364SNick Piggin do { 656c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 65731e6b01fSNick Piggin nd->root = fs->root; 658c1530019STim Chen nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq); 659c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 66031e6b01fSNick Piggin } 66131e6b01fSNick Piggin } 66231e6b01fSNick Piggin 663f1662356SArjan van de Ven static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link) 6641da177e4SLinus Torvalds { 66531e6b01fSNick Piggin int ret; 66631e6b01fSNick Piggin 6671da177e4SLinus Torvalds if (IS_ERR(link)) 6681da177e4SLinus Torvalds goto fail; 6691da177e4SLinus Torvalds 6701da177e4SLinus Torvalds if (*link == '/') { 6712a737871SAl Viro set_root(nd); 6721d957f9bSJan Blunck path_put(&nd->path); 6732a737871SAl Viro nd->path = nd->root; 6742a737871SAl Viro path_get(&nd->root); 67516c2cd71SAl Viro nd->flags |= LOOKUP_JUMPED; 6761da177e4SLinus Torvalds } 67731e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 678b4091d5fSChristoph Hellwig 67931e6b01fSNick Piggin ret = link_path_walk(link, nd); 68031e6b01fSNick Piggin return ret; 6811da177e4SLinus Torvalds fail: 6821d957f9bSJan Blunck path_put(&nd->path); 6831da177e4SLinus Torvalds return PTR_ERR(link); 6841da177e4SLinus Torvalds } 6851da177e4SLinus Torvalds 6861d957f9bSJan Blunck static void path_put_conditional(struct path *path, struct nameidata *nd) 687051d3812SIan Kent { 688051d3812SIan Kent dput(path->dentry); 6894ac91378SJan Blunck if (path->mnt != nd->path.mnt) 690051d3812SIan Kent mntput(path->mnt); 691051d3812SIan Kent } 692051d3812SIan Kent 6937b9337aaSNick Piggin static inline void path_to_nameidata(const struct path *path, 6947b9337aaSNick Piggin struct nameidata *nd) 695051d3812SIan Kent { 69631e6b01fSNick Piggin if (!(nd->flags & LOOKUP_RCU)) { 6974ac91378SJan Blunck dput(nd->path.dentry); 69831e6b01fSNick Piggin if (nd->path.mnt != path->mnt) 6994ac91378SJan Blunck mntput(nd->path.mnt); 7009a229683SHuang Shijie } 70131e6b01fSNick Piggin nd->path.mnt = path->mnt; 7024ac91378SJan Blunck nd->path.dentry = path->dentry; 703051d3812SIan Kent } 704051d3812SIan Kent 705b5fb63c1SChristoph Hellwig /* 706b5fb63c1SChristoph Hellwig * Helper to directly jump to a known parsed path from ->follow_link, 707b5fb63c1SChristoph Hellwig * caller must have taken a reference to path beforehand. 708b5fb63c1SChristoph Hellwig */ 709b5fb63c1SChristoph Hellwig void nd_jump_link(struct nameidata *nd, struct path *path) 710b5fb63c1SChristoph Hellwig { 711b5fb63c1SChristoph Hellwig path_put(&nd->path); 712b5fb63c1SChristoph Hellwig 713b5fb63c1SChristoph Hellwig nd->path = *path; 714b5fb63c1SChristoph Hellwig nd->inode = nd->path.dentry->d_inode; 715b5fb63c1SChristoph Hellwig nd->flags |= LOOKUP_JUMPED; 716b5fb63c1SChristoph Hellwig } 717b5fb63c1SChristoph Hellwig 718574197e0SAl Viro static inline void put_link(struct nameidata *nd, struct path *link, void *cookie) 719574197e0SAl Viro { 720574197e0SAl Viro struct inode *inode = link->dentry->d_inode; 7216d7b5aaeSAl Viro if (inode->i_op->put_link) 722574197e0SAl Viro inode->i_op->put_link(link->dentry, nd, cookie); 723574197e0SAl Viro path_put(link); 724574197e0SAl Viro } 725574197e0SAl Viro 726561ec64aSLinus Torvalds int sysctl_protected_symlinks __read_mostly = 0; 727561ec64aSLinus Torvalds int sysctl_protected_hardlinks __read_mostly = 0; 728800179c9SKees Cook 729800179c9SKees Cook /** 730800179c9SKees Cook * may_follow_link - Check symlink following for unsafe situations 731800179c9SKees Cook * @link: The path of the symlink 73255852635SRandy Dunlap * @nd: nameidata pathwalk data 733800179c9SKees Cook * 734800179c9SKees Cook * In the case of the sysctl_protected_symlinks sysctl being enabled, 735800179c9SKees Cook * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is 736800179c9SKees Cook * in a sticky world-writable directory. This is to protect privileged 737800179c9SKees Cook * processes from failing races against path names that may change out 738800179c9SKees Cook * from under them by way of other users creating malicious symlinks. 739800179c9SKees Cook * It will permit symlinks to be followed only when outside a sticky 740800179c9SKees Cook * world-writable directory, or when the uid of the symlink and follower 741800179c9SKees Cook * match, or when the directory owner matches the symlink's owner. 742800179c9SKees Cook * 743800179c9SKees Cook * Returns 0 if following the symlink is allowed, -ve on error. 744800179c9SKees Cook */ 745800179c9SKees Cook static inline int may_follow_link(struct path *link, struct nameidata *nd) 746800179c9SKees Cook { 747800179c9SKees Cook const struct inode *inode; 748800179c9SKees Cook const struct inode *parent; 749800179c9SKees Cook 750800179c9SKees Cook if (!sysctl_protected_symlinks) 751800179c9SKees Cook return 0; 752800179c9SKees Cook 753800179c9SKees Cook /* Allowed if owner and follower match. */ 754800179c9SKees Cook inode = link->dentry->d_inode; 75581abe27bSEric W. Biederman if (uid_eq(current_cred()->fsuid, inode->i_uid)) 756800179c9SKees Cook return 0; 757800179c9SKees Cook 758800179c9SKees Cook /* Allowed if parent directory not sticky and world-writable. */ 759800179c9SKees Cook parent = nd->path.dentry->d_inode; 760800179c9SKees Cook if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) 761800179c9SKees Cook return 0; 762800179c9SKees Cook 763800179c9SKees Cook /* Allowed if parent directory and link owner match. */ 76481abe27bSEric W. Biederman if (uid_eq(parent->i_uid, inode->i_uid)) 765800179c9SKees Cook return 0; 766800179c9SKees Cook 767ffd8d101SSasha Levin audit_log_link_denied("follow_link", link); 768800179c9SKees Cook path_put_conditional(link, nd); 769800179c9SKees Cook path_put(&nd->path); 770800179c9SKees Cook return -EACCES; 771800179c9SKees Cook } 772800179c9SKees Cook 773800179c9SKees Cook /** 774800179c9SKees Cook * safe_hardlink_source - Check for safe hardlink conditions 775800179c9SKees Cook * @inode: the source inode to hardlink from 776800179c9SKees Cook * 777800179c9SKees Cook * Return false if at least one of the following conditions: 778800179c9SKees Cook * - inode is not a regular file 779800179c9SKees Cook * - inode is setuid 780800179c9SKees Cook * - inode is setgid and group-exec 781800179c9SKees Cook * - access failure for read and write 782800179c9SKees Cook * 783800179c9SKees Cook * Otherwise returns true. 784800179c9SKees Cook */ 785800179c9SKees Cook static bool safe_hardlink_source(struct inode *inode) 786800179c9SKees Cook { 787800179c9SKees Cook umode_t mode = inode->i_mode; 788800179c9SKees Cook 789800179c9SKees Cook /* Special files should not get pinned to the filesystem. */ 790800179c9SKees Cook if (!S_ISREG(mode)) 791800179c9SKees Cook return false; 792800179c9SKees Cook 793800179c9SKees Cook /* Setuid files should not get pinned to the filesystem. */ 794800179c9SKees Cook if (mode & S_ISUID) 795800179c9SKees Cook return false; 796800179c9SKees Cook 797800179c9SKees Cook /* Executable setgid files should not get pinned to the filesystem. */ 798800179c9SKees Cook if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) 799800179c9SKees Cook return false; 800800179c9SKees Cook 801800179c9SKees Cook /* Hardlinking to unreadable or unwritable sources is dangerous. */ 802800179c9SKees Cook if (inode_permission(inode, MAY_READ | MAY_WRITE)) 803800179c9SKees Cook return false; 804800179c9SKees Cook 805800179c9SKees Cook return true; 806800179c9SKees Cook } 807800179c9SKees Cook 808800179c9SKees Cook /** 809800179c9SKees Cook * may_linkat - Check permissions for creating a hardlink 810800179c9SKees Cook * @link: the source to hardlink from 811800179c9SKees Cook * 812800179c9SKees Cook * Block hardlink when all of: 813800179c9SKees Cook * - sysctl_protected_hardlinks enabled 814800179c9SKees Cook * - fsuid does not match inode 815800179c9SKees Cook * - hardlink source is unsafe (see safe_hardlink_source() above) 816800179c9SKees Cook * - not CAP_FOWNER 817800179c9SKees Cook * 818800179c9SKees Cook * Returns 0 if successful, -ve on error. 819800179c9SKees Cook */ 820800179c9SKees Cook static int may_linkat(struct path *link) 821800179c9SKees Cook { 822800179c9SKees Cook const struct cred *cred; 823800179c9SKees Cook struct inode *inode; 824800179c9SKees Cook 825800179c9SKees Cook if (!sysctl_protected_hardlinks) 826800179c9SKees Cook return 0; 827800179c9SKees Cook 828800179c9SKees Cook cred = current_cred(); 829800179c9SKees Cook inode = link->dentry->d_inode; 830800179c9SKees Cook 831800179c9SKees Cook /* Source inode owner (or CAP_FOWNER) can hardlink all they like, 832800179c9SKees Cook * otherwise, it must be a safe source. 833800179c9SKees Cook */ 83481abe27bSEric W. Biederman if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) || 835800179c9SKees Cook capable(CAP_FOWNER)) 836800179c9SKees Cook return 0; 837800179c9SKees Cook 838a51d9eaaSKees Cook audit_log_link_denied("linkat", link); 839800179c9SKees Cook return -EPERM; 840800179c9SKees Cook } 841800179c9SKees Cook 842def4af30SAl Viro static __always_inline int 843574197e0SAl Viro follow_link(struct path *link, struct nameidata *nd, void **p) 8441da177e4SLinus Torvalds { 8457b9337aaSNick Piggin struct dentry *dentry = link->dentry; 8466d7b5aaeSAl Viro int error; 8476d7b5aaeSAl Viro char *s; 8481da177e4SLinus Torvalds 849844a3917SAl Viro BUG_ON(nd->flags & LOOKUP_RCU); 850844a3917SAl Viro 8510e794589SAl Viro if (link->mnt == nd->path.mnt) 8520e794589SAl Viro mntget(link->mnt); 8530e794589SAl Viro 8546d7b5aaeSAl Viro error = -ELOOP; 8556d7b5aaeSAl Viro if (unlikely(current->total_link_count >= 40)) 8566d7b5aaeSAl Viro goto out_put_nd_path; 8576d7b5aaeSAl Viro 858574197e0SAl Viro cond_resched(); 859574197e0SAl Viro current->total_link_count++; 860574197e0SAl Viro 86168ac1234SAl Viro touch_atime(link); 8621da177e4SLinus Torvalds nd_set_link(nd, NULL); 863cd4e91d3SAl Viro 86436f3b4f6SAl Viro error = security_inode_follow_link(link->dentry, nd); 8656d7b5aaeSAl Viro if (error) 8666d7b5aaeSAl Viro goto out_put_nd_path; 86736f3b4f6SAl Viro 86886acdca1SAl Viro nd->last_type = LAST_BIND; 869def4af30SAl Viro *p = dentry->d_inode->i_op->follow_link(dentry, nd); 870def4af30SAl Viro error = PTR_ERR(*p); 8716d7b5aaeSAl Viro if (IS_ERR(*p)) 872408ef013SChristoph Hellwig goto out_put_nd_path; 8736d7b5aaeSAl Viro 874cc314eefSLinus Torvalds error = 0; 8756d7b5aaeSAl Viro s = nd_get_link(nd); 8766d7b5aaeSAl Viro if (s) { 8771da177e4SLinus Torvalds error = __vfs_follow_link(nd, s); 8786d7b5aaeSAl Viro if (unlikely(error)) 8796d7b5aaeSAl Viro put_link(nd, link, *p); 880b5fb63c1SChristoph Hellwig } 8816d7b5aaeSAl Viro 8826d7b5aaeSAl Viro return error; 8836d7b5aaeSAl Viro 8846d7b5aaeSAl Viro out_put_nd_path: 88598f6ef64SArnd Bergmann *p = NULL; 8866d7b5aaeSAl Viro path_put(&nd->path); 8876d7b5aaeSAl Viro path_put(link); 8881da177e4SLinus Torvalds return error; 8891da177e4SLinus Torvalds } 8901da177e4SLinus Torvalds 89131e6b01fSNick Piggin static int follow_up_rcu(struct path *path) 89231e6b01fSNick Piggin { 8930714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 8940714a533SAl Viro struct mount *parent; 89531e6b01fSNick Piggin struct dentry *mountpoint; 89631e6b01fSNick Piggin 8970714a533SAl Viro parent = mnt->mnt_parent; 8980714a533SAl Viro if (&parent->mnt == path->mnt) 89931e6b01fSNick Piggin return 0; 900a73324daSAl Viro mountpoint = mnt->mnt_mountpoint; 90131e6b01fSNick Piggin path->dentry = mountpoint; 9020714a533SAl Viro path->mnt = &parent->mnt; 90331e6b01fSNick Piggin return 1; 90431e6b01fSNick Piggin } 90531e6b01fSNick Piggin 906f015f126SDavid Howells /* 907f015f126SDavid Howells * follow_up - Find the mountpoint of path's vfsmount 908f015f126SDavid Howells * 909f015f126SDavid Howells * Given a path, find the mountpoint of its source file system. 910f015f126SDavid Howells * Replace @path with the path of the mountpoint in the parent mount. 911f015f126SDavid Howells * Up is towards /. 912f015f126SDavid Howells * 913f015f126SDavid Howells * Return 1 if we went up a level and 0 if we were already at the 914f015f126SDavid Howells * root. 915f015f126SDavid Howells */ 916bab77ebfSAl Viro int follow_up(struct path *path) 9171da177e4SLinus Torvalds { 9180714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 9190714a533SAl Viro struct mount *parent; 9201da177e4SLinus Torvalds struct dentry *mountpoint; 92199b7db7bSNick Piggin 922962830dfSAndi Kleen br_read_lock(&vfsmount_lock); 9230714a533SAl Viro parent = mnt->mnt_parent; 9243c0a6163SAl Viro if (parent == mnt) { 925962830dfSAndi Kleen br_read_unlock(&vfsmount_lock); 9261da177e4SLinus Torvalds return 0; 9271da177e4SLinus Torvalds } 9280714a533SAl Viro mntget(&parent->mnt); 929a73324daSAl Viro mountpoint = dget(mnt->mnt_mountpoint); 930962830dfSAndi Kleen br_read_unlock(&vfsmount_lock); 931bab77ebfSAl Viro dput(path->dentry); 932bab77ebfSAl Viro path->dentry = mountpoint; 933bab77ebfSAl Viro mntput(path->mnt); 9340714a533SAl Viro path->mnt = &parent->mnt; 9351da177e4SLinus Torvalds return 1; 9361da177e4SLinus Torvalds } 9371da177e4SLinus Torvalds 938b5c84bf6SNick Piggin /* 9399875cf80SDavid Howells * Perform an automount 9409875cf80SDavid Howells * - return -EISDIR to tell follow_managed() to stop and return the path we 9419875cf80SDavid Howells * were called with. 9421da177e4SLinus Torvalds */ 9439875cf80SDavid Howells static int follow_automount(struct path *path, unsigned flags, 9449875cf80SDavid Howells bool *need_mntput) 94531e6b01fSNick Piggin { 9469875cf80SDavid Howells struct vfsmount *mnt; 947ea5b778aSDavid Howells int err; 9489875cf80SDavid Howells 9499875cf80SDavid Howells if (!path->dentry->d_op || !path->dentry->d_op->d_automount) 9509875cf80SDavid Howells return -EREMOTE; 9519875cf80SDavid Howells 9520ec26fd0SMiklos Szeredi /* We don't want to mount if someone's just doing a stat - 9530ec26fd0SMiklos Szeredi * unless they're stat'ing a directory and appended a '/' to 9540ec26fd0SMiklos Szeredi * the name. 9550ec26fd0SMiklos Szeredi * 9560ec26fd0SMiklos Szeredi * We do, however, want to mount if someone wants to open or 9575a30d8a2SDavid Howells * create a file of any type under the mountpoint, wants to 9580ec26fd0SMiklos Szeredi * traverse through the mountpoint or wants to open the 9590ec26fd0SMiklos Szeredi * mounted directory. Also, autofs may mark negative dentries 9600ec26fd0SMiklos Szeredi * as being automount points. These will need the attentions 9610ec26fd0SMiklos Szeredi * of the daemon to instantiate them before they can be used. 9625a30d8a2SDavid Howells */ 9635a30d8a2SDavid Howells if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | 964d94c177bSLinus Torvalds LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && 9655a30d8a2SDavid Howells path->dentry->d_inode) 9669875cf80SDavid Howells return -EISDIR; 9670ec26fd0SMiklos Szeredi 9689875cf80SDavid Howells current->total_link_count++; 9699875cf80SDavid Howells if (current->total_link_count >= 40) 9709875cf80SDavid Howells return -ELOOP; 9719875cf80SDavid Howells 9729875cf80SDavid Howells mnt = path->dentry->d_op->d_automount(path); 9739875cf80SDavid Howells if (IS_ERR(mnt)) { 9749875cf80SDavid Howells /* 9759875cf80SDavid Howells * The filesystem is allowed to return -EISDIR here to indicate 9769875cf80SDavid Howells * it doesn't want to automount. For instance, autofs would do 9779875cf80SDavid Howells * this so that its userspace daemon can mount on this dentry. 9789875cf80SDavid Howells * 9799875cf80SDavid Howells * However, we can only permit this if it's a terminal point in 9809875cf80SDavid Howells * the path being looked up; if it wasn't then the remainder of 9819875cf80SDavid Howells * the path is inaccessible and we should say so. 9829875cf80SDavid Howells */ 98349084c3bSAl Viro if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT)) 9849875cf80SDavid Howells return -EREMOTE; 9859875cf80SDavid Howells return PTR_ERR(mnt); 98631e6b01fSNick Piggin } 987ea5b778aSDavid Howells 9889875cf80SDavid Howells if (!mnt) /* mount collision */ 9899875cf80SDavid Howells return 0; 9909875cf80SDavid Howells 9918aef1884SAl Viro if (!*need_mntput) { 9928aef1884SAl Viro /* lock_mount() may release path->mnt on error */ 9938aef1884SAl Viro mntget(path->mnt); 9948aef1884SAl Viro *need_mntput = true; 9958aef1884SAl Viro } 99619a167afSAl Viro err = finish_automount(mnt, path); 997ea5b778aSDavid Howells 998ea5b778aSDavid Howells switch (err) { 999ea5b778aSDavid Howells case -EBUSY: 1000ea5b778aSDavid Howells /* Someone else made a mount here whilst we were busy */ 100119a167afSAl Viro return 0; 1002ea5b778aSDavid Howells case 0: 10038aef1884SAl Viro path_put(path); 10049875cf80SDavid Howells path->mnt = mnt; 10059875cf80SDavid Howells path->dentry = dget(mnt->mnt_root); 10069875cf80SDavid Howells return 0; 100719a167afSAl Viro default: 100819a167afSAl Viro return err; 10099875cf80SDavid Howells } 101019a167afSAl Viro 1011ea5b778aSDavid Howells } 10129875cf80SDavid Howells 10139875cf80SDavid Howells /* 10149875cf80SDavid Howells * Handle a dentry that is managed in some way. 1015cc53ce53SDavid Howells * - Flagged for transit management (autofs) 10169875cf80SDavid Howells * - Flagged as mountpoint 10179875cf80SDavid Howells * - Flagged as automount point 10189875cf80SDavid Howells * 10199875cf80SDavid Howells * This may only be called in refwalk mode. 10209875cf80SDavid Howells * 10219875cf80SDavid Howells * Serialization is taken care of in namespace.c 10229875cf80SDavid Howells */ 10239875cf80SDavid Howells static int follow_managed(struct path *path, unsigned flags) 10249875cf80SDavid Howells { 10258aef1884SAl Viro struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */ 10269875cf80SDavid Howells unsigned managed; 10279875cf80SDavid Howells bool need_mntput = false; 10288aef1884SAl Viro int ret = 0; 10299875cf80SDavid Howells 10309875cf80SDavid Howells /* Given that we're not holding a lock here, we retain the value in a 10319875cf80SDavid Howells * local variable for each dentry as we look at it so that we don't see 10329875cf80SDavid Howells * the components of that value change under us */ 10339875cf80SDavid Howells while (managed = ACCESS_ONCE(path->dentry->d_flags), 10349875cf80SDavid Howells managed &= DCACHE_MANAGED_DENTRY, 10359875cf80SDavid Howells unlikely(managed != 0)) { 1036cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1037cc53ce53SDavid Howells * being held. */ 1038cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 1039cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1040cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 10411aed3e42SAl Viro ret = path->dentry->d_op->d_manage(path->dentry, false); 1042cc53ce53SDavid Howells if (ret < 0) 10438aef1884SAl Viro break; 1044cc53ce53SDavid Howells } 1045cc53ce53SDavid Howells 10469875cf80SDavid Howells /* Transit to a mounted filesystem. */ 10479875cf80SDavid Howells if (managed & DCACHE_MOUNTED) { 10489875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 10499875cf80SDavid Howells if (mounted) { 10509875cf80SDavid Howells dput(path->dentry); 10519875cf80SDavid Howells if (need_mntput) 1052463ffb2eSAl Viro mntput(path->mnt); 1053463ffb2eSAl Viro path->mnt = mounted; 1054463ffb2eSAl Viro path->dentry = dget(mounted->mnt_root); 10559875cf80SDavid Howells need_mntput = true; 10569875cf80SDavid Howells continue; 1057463ffb2eSAl Viro } 1058463ffb2eSAl Viro 10599875cf80SDavid Howells /* Something is mounted on this dentry in another 10609875cf80SDavid Howells * namespace and/or whatever was mounted there in this 10619875cf80SDavid Howells * namespace got unmounted before we managed to get the 10629875cf80SDavid Howells * vfsmount_lock */ 10631da177e4SLinus Torvalds } 10649875cf80SDavid Howells 10659875cf80SDavid Howells /* Handle an automount point */ 10669875cf80SDavid Howells if (managed & DCACHE_NEED_AUTOMOUNT) { 10679875cf80SDavid Howells ret = follow_automount(path, flags, &need_mntput); 10689875cf80SDavid Howells if (ret < 0) 10698aef1884SAl Viro break; 10709875cf80SDavid Howells continue; 10719875cf80SDavid Howells } 10729875cf80SDavid Howells 10739875cf80SDavid Howells /* We didn't change the current path point */ 10749875cf80SDavid Howells break; 10759875cf80SDavid Howells } 10768aef1884SAl Viro 10778aef1884SAl Viro if (need_mntput && path->mnt == mnt) 10788aef1884SAl Viro mntput(path->mnt); 10798aef1884SAl Viro if (ret == -EISDIR) 10808aef1884SAl Viro ret = 0; 1081a3fbbde7SAl Viro return ret < 0 ? ret : need_mntput; 10821da177e4SLinus Torvalds } 10831da177e4SLinus Torvalds 1084cc53ce53SDavid Howells int follow_down_one(struct path *path) 10851da177e4SLinus Torvalds { 10861da177e4SLinus Torvalds struct vfsmount *mounted; 10871da177e4SLinus Torvalds 10881c755af4SAl Viro mounted = lookup_mnt(path); 10891da177e4SLinus Torvalds if (mounted) { 10909393bd07SAl Viro dput(path->dentry); 10919393bd07SAl Viro mntput(path->mnt); 10929393bd07SAl Viro path->mnt = mounted; 10939393bd07SAl Viro path->dentry = dget(mounted->mnt_root); 10941da177e4SLinus Torvalds return 1; 10951da177e4SLinus Torvalds } 10961da177e4SLinus Torvalds return 0; 10971da177e4SLinus Torvalds } 10981da177e4SLinus Torvalds 109962a7375eSIan Kent static inline bool managed_dentry_might_block(struct dentry *dentry) 110062a7375eSIan Kent { 110162a7375eSIan Kent return (dentry->d_flags & DCACHE_MANAGE_TRANSIT && 110262a7375eSIan Kent dentry->d_op->d_manage(dentry, true) < 0); 110362a7375eSIan Kent } 110462a7375eSIan Kent 11059875cf80SDavid Howells /* 1106287548e4SAl Viro * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if 1107287548e4SAl Viro * we meet a managed dentry that would need blocking. 11089875cf80SDavid Howells */ 11099875cf80SDavid Howells static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, 1110287548e4SAl Viro struct inode **inode) 11119875cf80SDavid Howells { 111262a7375eSIan Kent for (;;) { 1113c7105365SAl Viro struct mount *mounted; 111462a7375eSIan Kent /* 111562a7375eSIan Kent * Don't forget we might have a non-mountpoint managed dentry 111662a7375eSIan Kent * that wants to block transit. 111762a7375eSIan Kent */ 1118287548e4SAl Viro if (unlikely(managed_dentry_might_block(path->dentry))) 1119ab90911fSDavid Howells return false; 112062a7375eSIan Kent 112162a7375eSIan Kent if (!d_mountpoint(path->dentry)) 112262a7375eSIan Kent break; 112362a7375eSIan Kent 11249875cf80SDavid Howells mounted = __lookup_mnt(path->mnt, path->dentry, 1); 11259875cf80SDavid Howells if (!mounted) 11269875cf80SDavid Howells break; 1127c7105365SAl Viro path->mnt = &mounted->mnt; 1128c7105365SAl Viro path->dentry = mounted->mnt.mnt_root; 1129a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 11309875cf80SDavid Howells nd->seq = read_seqcount_begin(&path->dentry->d_seq); 113159430262SLinus Torvalds /* 113259430262SLinus Torvalds * Update the inode too. We don't need to re-check the 113359430262SLinus Torvalds * dentry sequence number here after this d_inode read, 113459430262SLinus Torvalds * because a mount-point is always pinned. 113559430262SLinus Torvalds */ 113659430262SLinus Torvalds *inode = path->dentry->d_inode; 11379875cf80SDavid Howells } 11389875cf80SDavid Howells return true; 11399875cf80SDavid Howells } 11409875cf80SDavid Howells 1141dea39376SAl Viro static void follow_mount_rcu(struct nameidata *nd) 1142287548e4SAl Viro { 1143dea39376SAl Viro while (d_mountpoint(nd->path.dentry)) { 1144c7105365SAl Viro struct mount *mounted; 1145dea39376SAl Viro mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1); 1146287548e4SAl Viro if (!mounted) 1147287548e4SAl Viro break; 1148c7105365SAl Viro nd->path.mnt = &mounted->mnt; 1149c7105365SAl Viro nd->path.dentry = mounted->mnt.mnt_root; 1150dea39376SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 1151287548e4SAl Viro } 1152287548e4SAl Viro } 1153287548e4SAl Viro 115431e6b01fSNick Piggin static int follow_dotdot_rcu(struct nameidata *nd) 115531e6b01fSNick Piggin { 115631e6b01fSNick Piggin set_root_rcu(nd); 115731e6b01fSNick Piggin 115831e6b01fSNick Piggin while (1) { 115931e6b01fSNick Piggin if (nd->path.dentry == nd->root.dentry && 116031e6b01fSNick Piggin nd->path.mnt == nd->root.mnt) { 116131e6b01fSNick Piggin break; 116231e6b01fSNick Piggin } 116331e6b01fSNick Piggin if (nd->path.dentry != nd->path.mnt->mnt_root) { 116431e6b01fSNick Piggin struct dentry *old = nd->path.dentry; 116531e6b01fSNick Piggin struct dentry *parent = old->d_parent; 116631e6b01fSNick Piggin unsigned seq; 116731e6b01fSNick Piggin 116831e6b01fSNick Piggin seq = read_seqcount_begin(&parent->d_seq); 116931e6b01fSNick Piggin if (read_seqcount_retry(&old->d_seq, nd->seq)) 1170ef7562d5SAl Viro goto failed; 117131e6b01fSNick Piggin nd->path.dentry = parent; 117231e6b01fSNick Piggin nd->seq = seq; 117331e6b01fSNick Piggin break; 117431e6b01fSNick Piggin } 117531e6b01fSNick Piggin if (!follow_up_rcu(&nd->path)) 117631e6b01fSNick Piggin break; 117731e6b01fSNick Piggin nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 117831e6b01fSNick Piggin } 1179dea39376SAl Viro follow_mount_rcu(nd); 1180dea39376SAl Viro nd->inode = nd->path.dentry->d_inode; 118131e6b01fSNick Piggin return 0; 1182ef7562d5SAl Viro 1183ef7562d5SAl Viro failed: 1184ef7562d5SAl Viro nd->flags &= ~LOOKUP_RCU; 11855b6ca027SAl Viro if (!(nd->flags & LOOKUP_ROOT)) 1186ef7562d5SAl Viro nd->root.mnt = NULL; 118732a7991bSAl Viro unlock_rcu_walk(); 1188ef7562d5SAl Viro return -ECHILD; 118931e6b01fSNick Piggin } 119031e6b01fSNick Piggin 11919875cf80SDavid Howells /* 1192cc53ce53SDavid Howells * Follow down to the covering mount currently visible to userspace. At each 1193cc53ce53SDavid Howells * point, the filesystem owning that dentry may be queried as to whether the 1194cc53ce53SDavid Howells * caller is permitted to proceed or not. 1195cc53ce53SDavid Howells */ 11967cc90cc3SAl Viro int follow_down(struct path *path) 1197cc53ce53SDavid Howells { 1198cc53ce53SDavid Howells unsigned managed; 1199cc53ce53SDavid Howells int ret; 1200cc53ce53SDavid Howells 1201cc53ce53SDavid Howells while (managed = ACCESS_ONCE(path->dentry->d_flags), 1202cc53ce53SDavid Howells unlikely(managed & DCACHE_MANAGED_DENTRY)) { 1203cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1204cc53ce53SDavid Howells * being held. 1205cc53ce53SDavid Howells * 1206cc53ce53SDavid Howells * We indicate to the filesystem if someone is trying to mount 1207cc53ce53SDavid Howells * something here. This gives autofs the chance to deny anyone 1208cc53ce53SDavid Howells * other than its daemon the right to mount on its 1209cc53ce53SDavid Howells * superstructure. 1210cc53ce53SDavid Howells * 1211cc53ce53SDavid Howells * The filesystem may sleep at this point. 1212cc53ce53SDavid Howells */ 1213cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 1214cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1215cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 1216ab90911fSDavid Howells ret = path->dentry->d_op->d_manage( 12171aed3e42SAl Viro path->dentry, false); 1218cc53ce53SDavid Howells if (ret < 0) 1219cc53ce53SDavid Howells return ret == -EISDIR ? 0 : ret; 1220cc53ce53SDavid Howells } 1221cc53ce53SDavid Howells 1222cc53ce53SDavid Howells /* Transit to a mounted filesystem. */ 1223cc53ce53SDavid Howells if (managed & DCACHE_MOUNTED) { 1224cc53ce53SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 1225cc53ce53SDavid Howells if (!mounted) 1226cc53ce53SDavid Howells break; 1227cc53ce53SDavid Howells dput(path->dentry); 1228cc53ce53SDavid Howells mntput(path->mnt); 1229cc53ce53SDavid Howells path->mnt = mounted; 1230cc53ce53SDavid Howells path->dentry = dget(mounted->mnt_root); 1231cc53ce53SDavid Howells continue; 1232cc53ce53SDavid Howells } 1233cc53ce53SDavid Howells 1234cc53ce53SDavid Howells /* Don't handle automount points here */ 1235cc53ce53SDavid Howells break; 1236cc53ce53SDavid Howells } 1237cc53ce53SDavid Howells return 0; 1238cc53ce53SDavid Howells } 1239cc53ce53SDavid Howells 1240cc53ce53SDavid Howells /* 12419875cf80SDavid Howells * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() 12429875cf80SDavid Howells */ 12439875cf80SDavid Howells static void follow_mount(struct path *path) 12449875cf80SDavid Howells { 12459875cf80SDavid Howells while (d_mountpoint(path->dentry)) { 12469875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 12479875cf80SDavid Howells if (!mounted) 12489875cf80SDavid Howells break; 12499875cf80SDavid Howells dput(path->dentry); 12509875cf80SDavid Howells mntput(path->mnt); 12519875cf80SDavid Howells path->mnt = mounted; 12529875cf80SDavid Howells path->dentry = dget(mounted->mnt_root); 12539875cf80SDavid Howells } 12549875cf80SDavid Howells } 12559875cf80SDavid Howells 125631e6b01fSNick Piggin static void follow_dotdot(struct nameidata *nd) 12571da177e4SLinus Torvalds { 12582a737871SAl Viro set_root(nd); 1259e518ddb7SAndreas Mohr 12601da177e4SLinus Torvalds while(1) { 12614ac91378SJan Blunck struct dentry *old = nd->path.dentry; 12621da177e4SLinus Torvalds 12632a737871SAl Viro if (nd->path.dentry == nd->root.dentry && 12642a737871SAl Viro nd->path.mnt == nd->root.mnt) { 12651da177e4SLinus Torvalds break; 12661da177e4SLinus Torvalds } 12674ac91378SJan Blunck if (nd->path.dentry != nd->path.mnt->mnt_root) { 12683088dd70SAl Viro /* rare case of legitimate dget_parent()... */ 12693088dd70SAl Viro nd->path.dentry = dget_parent(nd->path.dentry); 12701da177e4SLinus Torvalds dput(old); 12711da177e4SLinus Torvalds break; 12721da177e4SLinus Torvalds } 12733088dd70SAl Viro if (!follow_up(&nd->path)) 12741da177e4SLinus Torvalds break; 12751da177e4SLinus Torvalds } 127679ed0226SAl Viro follow_mount(&nd->path); 127731e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 12781da177e4SLinus Torvalds } 12791da177e4SLinus Torvalds 12801da177e4SLinus Torvalds /* 1281bad61189SMiklos Szeredi * This looks up the name in dcache, possibly revalidates the old dentry and 1282bad61189SMiklos Szeredi * allocates a new one if not found or not valid. In the need_lookup argument 1283bad61189SMiklos Szeredi * returns whether i_op->lookup is necessary. 1284bad61189SMiklos Szeredi * 1285bad61189SMiklos Szeredi * dir->d_inode->i_mutex must be held 1286baa03890SNick Piggin */ 1287bad61189SMiklos Szeredi static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir, 1288201f956eSAl Viro unsigned int flags, bool *need_lookup) 1289baa03890SNick Piggin { 1290baa03890SNick Piggin struct dentry *dentry; 1291bad61189SMiklos Szeredi int error; 1292baa03890SNick Piggin 1293bad61189SMiklos Szeredi *need_lookup = false; 1294bad61189SMiklos Szeredi dentry = d_lookup(dir, name); 1295bad61189SMiklos Szeredi if (dentry) { 129639e3c955SJeff Layton if (dentry->d_flags & DCACHE_OP_REVALIDATE) { 1297201f956eSAl Viro error = d_revalidate(dentry, flags); 1298bad61189SMiklos Szeredi if (unlikely(error <= 0)) { 1299bad61189SMiklos Szeredi if (error < 0) { 1300bad61189SMiklos Szeredi dput(dentry); 1301bad61189SMiklos Szeredi return ERR_PTR(error); 1302bad61189SMiklos Szeredi } else if (!d_invalidate(dentry)) { 1303bad61189SMiklos Szeredi dput(dentry); 1304bad61189SMiklos Szeredi dentry = NULL; 1305bad61189SMiklos Szeredi } 1306bad61189SMiklos Szeredi } 1307bad61189SMiklos Szeredi } 1308bad61189SMiklos Szeredi } 1309baa03890SNick Piggin 1310bad61189SMiklos Szeredi if (!dentry) { 1311bad61189SMiklos Szeredi dentry = d_alloc(dir, name); 1312baa03890SNick Piggin if (unlikely(!dentry)) 1313baa03890SNick Piggin return ERR_PTR(-ENOMEM); 1314baa03890SNick Piggin 1315bad61189SMiklos Szeredi *need_lookup = true; 1316baa03890SNick Piggin } 1317baa03890SNick Piggin return dentry; 1318baa03890SNick Piggin } 1319baa03890SNick Piggin 1320baa03890SNick Piggin /* 1321bad61189SMiklos Szeredi * Call i_op->lookup on the dentry. The dentry must be negative but may be 1322bad61189SMiklos Szeredi * hashed if it was pouplated with DCACHE_NEED_LOOKUP. 1323bad61189SMiklos Szeredi * 1324bad61189SMiklos Szeredi * dir->d_inode->i_mutex must be held 132544396f4bSJosef Bacik */ 1326bad61189SMiklos Szeredi static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry, 132772bd866aSAl Viro unsigned int flags) 132844396f4bSJosef Bacik { 132944396f4bSJosef Bacik struct dentry *old; 133044396f4bSJosef Bacik 133144396f4bSJosef Bacik /* Don't create child dentry for a dead directory. */ 1332bad61189SMiklos Szeredi if (unlikely(IS_DEADDIR(dir))) { 1333e188dc02SMiklos Szeredi dput(dentry); 133444396f4bSJosef Bacik return ERR_PTR(-ENOENT); 1335e188dc02SMiklos Szeredi } 133644396f4bSJosef Bacik 133772bd866aSAl Viro old = dir->i_op->lookup(dir, dentry, flags); 133844396f4bSJosef Bacik if (unlikely(old)) { 133944396f4bSJosef Bacik dput(dentry); 134044396f4bSJosef Bacik dentry = old; 134144396f4bSJosef Bacik } 134244396f4bSJosef Bacik return dentry; 134344396f4bSJosef Bacik } 134444396f4bSJosef Bacik 1345a3255546SAl Viro static struct dentry *__lookup_hash(struct qstr *name, 134672bd866aSAl Viro struct dentry *base, unsigned int flags) 1347a3255546SAl Viro { 1348bad61189SMiklos Szeredi bool need_lookup; 1349a3255546SAl Viro struct dentry *dentry; 1350a3255546SAl Viro 135172bd866aSAl Viro dentry = lookup_dcache(name, base, flags, &need_lookup); 1352bad61189SMiklos Szeredi if (!need_lookup) 1353a3255546SAl Viro return dentry; 1354bad61189SMiklos Szeredi 135572bd866aSAl Viro return lookup_real(base->d_inode, dentry, flags); 1356a3255546SAl Viro } 1357a3255546SAl Viro 135844396f4bSJosef Bacik /* 13591da177e4SLinus Torvalds * It's more convoluted than I'd like it to be, but... it's still fairly 13601da177e4SLinus Torvalds * small and for now I'd prefer to have fast path as straight as possible. 13611da177e4SLinus Torvalds * It _is_ time-critical. 13621da177e4SLinus Torvalds */ 1363e97cdc87SAl Viro static int lookup_fast(struct nameidata *nd, 136431e6b01fSNick Piggin struct path *path, struct inode **inode) 13651da177e4SLinus Torvalds { 13664ac91378SJan Blunck struct vfsmount *mnt = nd->path.mnt; 136731e6b01fSNick Piggin struct dentry *dentry, *parent = nd->path.dentry; 13685a18fff2SAl Viro int need_reval = 1; 13695a18fff2SAl Viro int status = 1; 13709875cf80SDavid Howells int err; 13719875cf80SDavid Howells 13723cac260aSAl Viro /* 1373b04f784eSNick Piggin * Rename seqlock is not required here because in the off chance 1374b04f784eSNick Piggin * of a false negative due to a concurrent rename, we're going to 1375b04f784eSNick Piggin * do the non-racy lookup, below. 1376b04f784eSNick Piggin */ 137731e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) { 137831e6b01fSNick Piggin unsigned seq; 1379da53be12SLinus Torvalds dentry = __d_lookup_rcu(parent, &nd->last, &seq); 13805a18fff2SAl Viro if (!dentry) 13815a18fff2SAl Viro goto unlazy; 13825a18fff2SAl Viro 138312f8ad4bSLinus Torvalds /* 138412f8ad4bSLinus Torvalds * This sequence count validates that the inode matches 138512f8ad4bSLinus Torvalds * the dentry name information from lookup. 138612f8ad4bSLinus Torvalds */ 138712f8ad4bSLinus Torvalds *inode = dentry->d_inode; 138812f8ad4bSLinus Torvalds if (read_seqcount_retry(&dentry->d_seq, seq)) 138912f8ad4bSLinus Torvalds return -ECHILD; 139012f8ad4bSLinus Torvalds 139112f8ad4bSLinus Torvalds /* 139212f8ad4bSLinus Torvalds * This sequence count validates that the parent had no 139312f8ad4bSLinus Torvalds * changes while we did the lookup of the dentry above. 139412f8ad4bSLinus Torvalds * 139512f8ad4bSLinus Torvalds * The memory barrier in read_seqcount_begin of child is 139612f8ad4bSLinus Torvalds * enough, we can use __read_seqcount_retry here. 139712f8ad4bSLinus Torvalds */ 139831e6b01fSNick Piggin if (__read_seqcount_retry(&parent->d_seq, nd->seq)) 139931e6b01fSNick Piggin return -ECHILD; 140031e6b01fSNick Piggin nd->seq = seq; 14015a18fff2SAl Viro 140224643087SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) { 14034ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 14045a18fff2SAl Viro if (unlikely(status <= 0)) { 14055a18fff2SAl Viro if (status != -ECHILD) 14065a18fff2SAl Viro need_reval = 0; 14075a18fff2SAl Viro goto unlazy; 14085a18fff2SAl Viro } 140924643087SAl Viro } 141031e6b01fSNick Piggin path->mnt = mnt; 141131e6b01fSNick Piggin path->dentry = dentry; 1412d6e9bd25SAl Viro if (unlikely(!__follow_mount_rcu(nd, path, inode))) 1413d6e9bd25SAl Viro goto unlazy; 1414d6e9bd25SAl Viro if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT)) 1415d6e9bd25SAl Viro goto unlazy; 14169875cf80SDavid Howells return 0; 14175a18fff2SAl Viro unlazy: 141819660af7SAl Viro if (unlazy_walk(nd, dentry)) 14195a18fff2SAl Viro return -ECHILD; 14205a18fff2SAl Viro } else { 1421e97cdc87SAl Viro dentry = __d_lookup(parent, &nd->last); 142224643087SAl Viro } 14235a18fff2SAl Viro 142481e6f520SAl Viro if (unlikely(!dentry)) 142581e6f520SAl Viro goto need_lookup; 14265a18fff2SAl Viro 14275a18fff2SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval) 14284ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 14295a18fff2SAl Viro if (unlikely(status <= 0)) { 14305a18fff2SAl Viro if (status < 0) { 14315a18fff2SAl Viro dput(dentry); 14325a18fff2SAl Viro return status; 14335a18fff2SAl Viro } 14345a18fff2SAl Viro if (!d_invalidate(dentry)) { 14355a18fff2SAl Viro dput(dentry); 143681e6f520SAl Viro goto need_lookup; 14375a18fff2SAl Viro } 14385a18fff2SAl Viro } 1439697f514dSMiklos Szeredi 14401da177e4SLinus Torvalds path->mnt = mnt; 14411da177e4SLinus Torvalds path->dentry = dentry; 14429875cf80SDavid Howells err = follow_managed(path, nd->flags); 144389312214SIan Kent if (unlikely(err < 0)) { 144489312214SIan Kent path_put_conditional(path, nd); 14459875cf80SDavid Howells return err; 144689312214SIan Kent } 1447a3fbbde7SAl Viro if (err) 1448a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 144931e6b01fSNick Piggin *inode = path->dentry->d_inode; 14501da177e4SLinus Torvalds return 0; 145181e6f520SAl Viro 145281e6f520SAl Viro need_lookup: 1453697f514dSMiklos Szeredi return 1; 1454697f514dSMiklos Szeredi } 1455697f514dSMiklos Szeredi 1456697f514dSMiklos Szeredi /* Fast lookup failed, do it the slow way */ 1457cc2a5271SAl Viro static int lookup_slow(struct nameidata *nd, struct path *path) 1458697f514dSMiklos Szeredi { 1459697f514dSMiklos Szeredi struct dentry *dentry, *parent; 1460697f514dSMiklos Szeredi int err; 1461697f514dSMiklos Szeredi 1462697f514dSMiklos Szeredi parent = nd->path.dentry; 146381e6f520SAl Viro BUG_ON(nd->inode != parent->d_inode); 146481e6f520SAl Viro 146581e6f520SAl Viro mutex_lock(&parent->d_inode->i_mutex); 1466cc2a5271SAl Viro dentry = __lookup_hash(&nd->last, parent, nd->flags); 146781e6f520SAl Viro mutex_unlock(&parent->d_inode->i_mutex); 146881e6f520SAl Viro if (IS_ERR(dentry)) 146981e6f520SAl Viro return PTR_ERR(dentry); 1470697f514dSMiklos Szeredi path->mnt = nd->path.mnt; 1471697f514dSMiklos Szeredi path->dentry = dentry; 1472697f514dSMiklos Szeredi err = follow_managed(path, nd->flags); 1473697f514dSMiklos Szeredi if (unlikely(err < 0)) { 1474697f514dSMiklos Szeredi path_put_conditional(path, nd); 1475697f514dSMiklos Szeredi return err; 1476697f514dSMiklos Szeredi } 1477697f514dSMiklos Szeredi if (err) 1478697f514dSMiklos Szeredi nd->flags |= LOOKUP_JUMPED; 1479697f514dSMiklos Szeredi return 0; 14801da177e4SLinus Torvalds } 14811da177e4SLinus Torvalds 148252094c8aSAl Viro static inline int may_lookup(struct nameidata *nd) 148352094c8aSAl Viro { 148452094c8aSAl Viro if (nd->flags & LOOKUP_RCU) { 14854ad5abb3SAl Viro int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); 148652094c8aSAl Viro if (err != -ECHILD) 148752094c8aSAl Viro return err; 148819660af7SAl Viro if (unlazy_walk(nd, NULL)) 148952094c8aSAl Viro return -ECHILD; 149052094c8aSAl Viro } 14914ad5abb3SAl Viro return inode_permission(nd->inode, MAY_EXEC); 149252094c8aSAl Viro } 149352094c8aSAl Viro 14949856fa1bSAl Viro static inline int handle_dots(struct nameidata *nd, int type) 14959856fa1bSAl Viro { 14969856fa1bSAl Viro if (type == LAST_DOTDOT) { 14979856fa1bSAl Viro if (nd->flags & LOOKUP_RCU) { 14989856fa1bSAl Viro if (follow_dotdot_rcu(nd)) 14999856fa1bSAl Viro return -ECHILD; 15009856fa1bSAl Viro } else 15019856fa1bSAl Viro follow_dotdot(nd); 15029856fa1bSAl Viro } 15039856fa1bSAl Viro return 0; 15049856fa1bSAl Viro } 15059856fa1bSAl Viro 1506951361f9SAl Viro static void terminate_walk(struct nameidata *nd) 1507951361f9SAl Viro { 1508951361f9SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1509951361f9SAl Viro path_put(&nd->path); 1510951361f9SAl Viro } else { 1511951361f9SAl Viro nd->flags &= ~LOOKUP_RCU; 15125b6ca027SAl Viro if (!(nd->flags & LOOKUP_ROOT)) 1513951361f9SAl Viro nd->root.mnt = NULL; 151432a7991bSAl Viro unlock_rcu_walk(); 1515951361f9SAl Viro } 1516951361f9SAl Viro } 1517951361f9SAl Viro 15183ddcd056SLinus Torvalds /* 15193ddcd056SLinus Torvalds * Do we need to follow links? We _really_ want to be able 15203ddcd056SLinus Torvalds * to do this check without having to look at inode->i_op, 15213ddcd056SLinus Torvalds * so we keep a cache of "no, this doesn't need follow_link" 15223ddcd056SLinus Torvalds * for the common case. 15233ddcd056SLinus Torvalds */ 15247813b94aSLinus Torvalds static inline int should_follow_link(struct inode *inode, int follow) 15253ddcd056SLinus Torvalds { 15263ddcd056SLinus Torvalds if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) { 15273ddcd056SLinus Torvalds if (likely(inode->i_op->follow_link)) 15283ddcd056SLinus Torvalds return follow; 15293ddcd056SLinus Torvalds 15303ddcd056SLinus Torvalds /* This gets set once for the inode lifetime */ 15313ddcd056SLinus Torvalds spin_lock(&inode->i_lock); 15323ddcd056SLinus Torvalds inode->i_opflags |= IOP_NOFOLLOW; 15333ddcd056SLinus Torvalds spin_unlock(&inode->i_lock); 15343ddcd056SLinus Torvalds } 15353ddcd056SLinus Torvalds return 0; 15363ddcd056SLinus Torvalds } 15373ddcd056SLinus Torvalds 1538ce57dfc1SAl Viro static inline int walk_component(struct nameidata *nd, struct path *path, 153921b9b073SAl Viro int follow) 1540ce57dfc1SAl Viro { 1541ce57dfc1SAl Viro struct inode *inode; 1542ce57dfc1SAl Viro int err; 1543ce57dfc1SAl Viro /* 1544ce57dfc1SAl Viro * "." and ".." are special - ".." especially so because it has 1545ce57dfc1SAl Viro * to be able to know about the current root directory and 1546ce57dfc1SAl Viro * parent relationships. 1547ce57dfc1SAl Viro */ 154821b9b073SAl Viro if (unlikely(nd->last_type != LAST_NORM)) 154921b9b073SAl Viro return handle_dots(nd, nd->last_type); 1550e97cdc87SAl Viro err = lookup_fast(nd, path, &inode); 1551ce57dfc1SAl Viro if (unlikely(err)) { 1552697f514dSMiklos Szeredi if (err < 0) 1553697f514dSMiklos Szeredi goto out_err; 1554697f514dSMiklos Szeredi 1555cc2a5271SAl Viro err = lookup_slow(nd, path); 1556697f514dSMiklos Szeredi if (err < 0) 1557697f514dSMiklos Szeredi goto out_err; 1558697f514dSMiklos Szeredi 1559697f514dSMiklos Szeredi inode = path->dentry->d_inode; 1560ce57dfc1SAl Viro } 1561697f514dSMiklos Szeredi err = -ENOENT; 1562697f514dSMiklos Szeredi if (!inode) 1563697f514dSMiklos Szeredi goto out_path_put; 1564697f514dSMiklos Szeredi 15657813b94aSLinus Torvalds if (should_follow_link(inode, follow)) { 156619660af7SAl Viro if (nd->flags & LOOKUP_RCU) { 156719660af7SAl Viro if (unlikely(unlazy_walk(nd, path->dentry))) { 1568697f514dSMiklos Szeredi err = -ECHILD; 1569697f514dSMiklos Szeredi goto out_err; 157019660af7SAl Viro } 157119660af7SAl Viro } 1572ce57dfc1SAl Viro BUG_ON(inode != path->dentry->d_inode); 1573ce57dfc1SAl Viro return 1; 1574ce57dfc1SAl Viro } 1575ce57dfc1SAl Viro path_to_nameidata(path, nd); 1576ce57dfc1SAl Viro nd->inode = inode; 1577ce57dfc1SAl Viro return 0; 1578697f514dSMiklos Szeredi 1579697f514dSMiklos Szeredi out_path_put: 1580697f514dSMiklos Szeredi path_to_nameidata(path, nd); 1581697f514dSMiklos Szeredi out_err: 1582697f514dSMiklos Szeredi terminate_walk(nd); 1583697f514dSMiklos Szeredi return err; 1584ce57dfc1SAl Viro } 1585ce57dfc1SAl Viro 15861da177e4SLinus Torvalds /* 1587b356379aSAl Viro * This limits recursive symlink follows to 8, while 1588b356379aSAl Viro * limiting consecutive symlinks to 40. 1589b356379aSAl Viro * 1590b356379aSAl Viro * Without that kind of total limit, nasty chains of consecutive 1591b356379aSAl Viro * symlinks can cause almost arbitrarily long lookups. 1592b356379aSAl Viro */ 1593b356379aSAl Viro static inline int nested_symlink(struct path *path, struct nameidata *nd) 1594b356379aSAl Viro { 1595b356379aSAl Viro int res; 1596b356379aSAl Viro 1597b356379aSAl Viro if (unlikely(current->link_count >= MAX_NESTED_LINKS)) { 1598b356379aSAl Viro path_put_conditional(path, nd); 1599b356379aSAl Viro path_put(&nd->path); 1600b356379aSAl Viro return -ELOOP; 1601b356379aSAl Viro } 16021a4022f8SErez Zadok BUG_ON(nd->depth >= MAX_NESTED_LINKS); 1603b356379aSAl Viro 1604b356379aSAl Viro nd->depth++; 1605b356379aSAl Viro current->link_count++; 1606b356379aSAl Viro 1607b356379aSAl Viro do { 1608b356379aSAl Viro struct path link = *path; 1609b356379aSAl Viro void *cookie; 1610574197e0SAl Viro 1611574197e0SAl Viro res = follow_link(&link, nd, &cookie); 16126d7b5aaeSAl Viro if (res) 16136d7b5aaeSAl Viro break; 161421b9b073SAl Viro res = walk_component(nd, path, LOOKUP_FOLLOW); 1615574197e0SAl Viro put_link(nd, &link, cookie); 1616b356379aSAl Viro } while (res > 0); 1617b356379aSAl Viro 1618b356379aSAl Viro current->link_count--; 1619b356379aSAl Viro nd->depth--; 1620b356379aSAl Viro return res; 1621b356379aSAl Viro } 1622b356379aSAl Viro 1623b356379aSAl Viro /* 16243ddcd056SLinus Torvalds * We really don't want to look at inode->i_op->lookup 16253ddcd056SLinus Torvalds * when we don't have to. So we keep a cache bit in 16263ddcd056SLinus Torvalds * the inode ->i_opflags field that says "yes, we can 16273ddcd056SLinus Torvalds * do lookup on this inode". 16283ddcd056SLinus Torvalds */ 16293ddcd056SLinus Torvalds static inline int can_lookup(struct inode *inode) 16303ddcd056SLinus Torvalds { 16313ddcd056SLinus Torvalds if (likely(inode->i_opflags & IOP_LOOKUP)) 16323ddcd056SLinus Torvalds return 1; 16333ddcd056SLinus Torvalds if (likely(!inode->i_op->lookup)) 16343ddcd056SLinus Torvalds return 0; 16353ddcd056SLinus Torvalds 16363ddcd056SLinus Torvalds /* We do this once for the lifetime of the inode */ 16373ddcd056SLinus Torvalds spin_lock(&inode->i_lock); 16383ddcd056SLinus Torvalds inode->i_opflags |= IOP_LOOKUP; 16393ddcd056SLinus Torvalds spin_unlock(&inode->i_lock); 16403ddcd056SLinus Torvalds return 1; 16413ddcd056SLinus Torvalds } 16423ddcd056SLinus Torvalds 1643bfcfaa77SLinus Torvalds /* 1644bfcfaa77SLinus Torvalds * We can do the critical dentry name comparison and hashing 1645bfcfaa77SLinus Torvalds * operations one word at a time, but we are limited to: 1646bfcfaa77SLinus Torvalds * 1647bfcfaa77SLinus Torvalds * - Architectures with fast unaligned word accesses. We could 1648bfcfaa77SLinus Torvalds * do a "get_unaligned()" if this helps and is sufficiently 1649bfcfaa77SLinus Torvalds * fast. 1650bfcfaa77SLinus Torvalds * 1651bfcfaa77SLinus Torvalds * - Little-endian machines (so that we can generate the mask 1652bfcfaa77SLinus Torvalds * of low bytes efficiently). Again, we *could* do a byte 1653bfcfaa77SLinus Torvalds * swapping load on big-endian architectures if that is not 1654bfcfaa77SLinus Torvalds * expensive enough to make the optimization worthless. 1655bfcfaa77SLinus Torvalds * 1656bfcfaa77SLinus Torvalds * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we 1657bfcfaa77SLinus Torvalds * do not trap on the (extremely unlikely) case of a page 1658bfcfaa77SLinus Torvalds * crossing operation. 1659bfcfaa77SLinus Torvalds * 1660bfcfaa77SLinus Torvalds * - Furthermore, we need an efficient 64-bit compile for the 1661bfcfaa77SLinus Torvalds * 64-bit case in order to generate the "number of bytes in 1662bfcfaa77SLinus Torvalds * the final mask". Again, that could be replaced with a 1663bfcfaa77SLinus Torvalds * efficient population count instruction or similar. 1664bfcfaa77SLinus Torvalds */ 1665bfcfaa77SLinus Torvalds #ifdef CONFIG_DCACHE_WORD_ACCESS 1666bfcfaa77SLinus Torvalds 1667f68e556eSLinus Torvalds #include <asm/word-at-a-time.h> 1668bfcfaa77SLinus Torvalds 1669f68e556eSLinus Torvalds #ifdef CONFIG_64BIT 1670bfcfaa77SLinus Torvalds 1671bfcfaa77SLinus Torvalds static inline unsigned int fold_hash(unsigned long hash) 1672bfcfaa77SLinus Torvalds { 1673bfcfaa77SLinus Torvalds hash += hash >> (8*sizeof(int)); 1674bfcfaa77SLinus Torvalds return hash; 1675bfcfaa77SLinus Torvalds } 1676bfcfaa77SLinus Torvalds 1677bfcfaa77SLinus Torvalds #else /* 32-bit case */ 1678bfcfaa77SLinus Torvalds 1679bfcfaa77SLinus Torvalds #define fold_hash(x) (x) 1680bfcfaa77SLinus Torvalds 1681bfcfaa77SLinus Torvalds #endif 1682bfcfaa77SLinus Torvalds 1683bfcfaa77SLinus Torvalds unsigned int full_name_hash(const unsigned char *name, unsigned int len) 1684bfcfaa77SLinus Torvalds { 1685bfcfaa77SLinus Torvalds unsigned long a, mask; 1686bfcfaa77SLinus Torvalds unsigned long hash = 0; 1687bfcfaa77SLinus Torvalds 1688bfcfaa77SLinus Torvalds for (;;) { 1689e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name); 1690bfcfaa77SLinus Torvalds if (len < sizeof(unsigned long)) 1691bfcfaa77SLinus Torvalds break; 1692bfcfaa77SLinus Torvalds hash += a; 1693f132c5beSAl Viro hash *= 9; 1694bfcfaa77SLinus Torvalds name += sizeof(unsigned long); 1695bfcfaa77SLinus Torvalds len -= sizeof(unsigned long); 1696bfcfaa77SLinus Torvalds if (!len) 1697bfcfaa77SLinus Torvalds goto done; 1698bfcfaa77SLinus Torvalds } 1699bfcfaa77SLinus Torvalds mask = ~(~0ul << len*8); 1700bfcfaa77SLinus Torvalds hash += mask & a; 1701bfcfaa77SLinus Torvalds done: 1702bfcfaa77SLinus Torvalds return fold_hash(hash); 1703bfcfaa77SLinus Torvalds } 1704bfcfaa77SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 1705bfcfaa77SLinus Torvalds 1706bfcfaa77SLinus Torvalds /* 1707bfcfaa77SLinus Torvalds * Calculate the length and hash of the path component, and 1708bfcfaa77SLinus Torvalds * return the length of the component; 1709bfcfaa77SLinus Torvalds */ 1710bfcfaa77SLinus Torvalds static inline unsigned long hash_name(const char *name, unsigned int *hashp) 1711bfcfaa77SLinus Torvalds { 171236126f8fSLinus Torvalds unsigned long a, b, adata, bdata, mask, hash, len; 171336126f8fSLinus Torvalds const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 1714bfcfaa77SLinus Torvalds 1715bfcfaa77SLinus Torvalds hash = a = 0; 1716bfcfaa77SLinus Torvalds len = -sizeof(unsigned long); 1717bfcfaa77SLinus Torvalds do { 1718bfcfaa77SLinus Torvalds hash = (hash + a) * 9; 1719bfcfaa77SLinus Torvalds len += sizeof(unsigned long); 1720e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name+len); 172136126f8fSLinus Torvalds b = a ^ REPEAT_BYTE('/'); 172236126f8fSLinus Torvalds } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); 1723bfcfaa77SLinus Torvalds 172436126f8fSLinus Torvalds adata = prep_zero_mask(a, adata, &constants); 172536126f8fSLinus Torvalds bdata = prep_zero_mask(b, bdata, &constants); 172636126f8fSLinus Torvalds 172736126f8fSLinus Torvalds mask = create_zero_mask(adata | bdata); 172836126f8fSLinus Torvalds 172936126f8fSLinus Torvalds hash += a & zero_bytemask(mask); 1730bfcfaa77SLinus Torvalds *hashp = fold_hash(hash); 1731bfcfaa77SLinus Torvalds 173236126f8fSLinus Torvalds return len + find_zero(mask); 1733bfcfaa77SLinus Torvalds } 1734bfcfaa77SLinus Torvalds 1735bfcfaa77SLinus Torvalds #else 1736bfcfaa77SLinus Torvalds 17370145acc2SLinus Torvalds unsigned int full_name_hash(const unsigned char *name, unsigned int len) 17380145acc2SLinus Torvalds { 17390145acc2SLinus Torvalds unsigned long hash = init_name_hash(); 17400145acc2SLinus Torvalds while (len--) 17410145acc2SLinus Torvalds hash = partial_name_hash(*name++, hash); 17420145acc2SLinus Torvalds return end_name_hash(hash); 17430145acc2SLinus Torvalds } 1744ae942ae7SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 17450145acc2SLinus Torvalds 17463ddcd056SLinus Torvalds /* 1747200e9ef7SLinus Torvalds * We know there's a real path component here of at least 1748200e9ef7SLinus Torvalds * one character. 1749200e9ef7SLinus Torvalds */ 1750200e9ef7SLinus Torvalds static inline unsigned long hash_name(const char *name, unsigned int *hashp) 1751200e9ef7SLinus Torvalds { 1752200e9ef7SLinus Torvalds unsigned long hash = init_name_hash(); 1753200e9ef7SLinus Torvalds unsigned long len = 0, c; 1754200e9ef7SLinus Torvalds 1755200e9ef7SLinus Torvalds c = (unsigned char)*name; 1756200e9ef7SLinus Torvalds do { 1757200e9ef7SLinus Torvalds len++; 1758200e9ef7SLinus Torvalds hash = partial_name_hash(c, hash); 1759200e9ef7SLinus Torvalds c = (unsigned char)name[len]; 1760200e9ef7SLinus Torvalds } while (c && c != '/'); 1761200e9ef7SLinus Torvalds *hashp = end_name_hash(hash); 1762200e9ef7SLinus Torvalds return len; 1763200e9ef7SLinus Torvalds } 1764200e9ef7SLinus Torvalds 1765bfcfaa77SLinus Torvalds #endif 1766bfcfaa77SLinus Torvalds 1767200e9ef7SLinus Torvalds /* 17681da177e4SLinus Torvalds * Name resolution. 1769ea3834d9SPrasanna Meda * This is the basic name resolution function, turning a pathname into 1770ea3834d9SPrasanna Meda * the final dentry. We expect 'base' to be positive and a directory. 17711da177e4SLinus Torvalds * 1772ea3834d9SPrasanna Meda * Returns 0 and nd will have valid dentry and mnt on success. 1773ea3834d9SPrasanna Meda * Returns error and drops reference to input namei data on failure. 17741da177e4SLinus Torvalds */ 17756de88d72SAl Viro static int link_path_walk(const char *name, struct nameidata *nd) 17761da177e4SLinus Torvalds { 17771da177e4SLinus Torvalds struct path next; 17781da177e4SLinus Torvalds int err; 17791da177e4SLinus Torvalds 17801da177e4SLinus Torvalds while (*name=='/') 17811da177e4SLinus Torvalds name++; 17821da177e4SLinus Torvalds if (!*name) 1783086e183aSAl Viro return 0; 17841da177e4SLinus Torvalds 17851da177e4SLinus Torvalds /* At this point we know we have a real path component. */ 17861da177e4SLinus Torvalds for(;;) { 17871da177e4SLinus Torvalds struct qstr this; 1788200e9ef7SLinus Torvalds long len; 1789fe479a58SAl Viro int type; 17901da177e4SLinus Torvalds 179152094c8aSAl Viro err = may_lookup(nd); 17921da177e4SLinus Torvalds if (err) 17931da177e4SLinus Torvalds break; 17941da177e4SLinus Torvalds 1795200e9ef7SLinus Torvalds len = hash_name(name, &this.hash); 17961da177e4SLinus Torvalds this.name = name; 1797200e9ef7SLinus Torvalds this.len = len; 17981da177e4SLinus Torvalds 1799fe479a58SAl Viro type = LAST_NORM; 1800200e9ef7SLinus Torvalds if (name[0] == '.') switch (len) { 1801fe479a58SAl Viro case 2: 1802200e9ef7SLinus Torvalds if (name[1] == '.') { 1803fe479a58SAl Viro type = LAST_DOTDOT; 180416c2cd71SAl Viro nd->flags |= LOOKUP_JUMPED; 180516c2cd71SAl Viro } 1806fe479a58SAl Viro break; 1807fe479a58SAl Viro case 1: 1808fe479a58SAl Viro type = LAST_DOT; 1809fe479a58SAl Viro } 18105a202bcdSAl Viro if (likely(type == LAST_NORM)) { 18115a202bcdSAl Viro struct dentry *parent = nd->path.dentry; 181216c2cd71SAl Viro nd->flags &= ~LOOKUP_JUMPED; 18135a202bcdSAl Viro if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { 1814da53be12SLinus Torvalds err = parent->d_op->d_hash(parent, &this); 18155a202bcdSAl Viro if (err < 0) 18165a202bcdSAl Viro break; 18175a202bcdSAl Viro } 18185a202bcdSAl Viro } 1819fe479a58SAl Viro 18205f4a6a69SAl Viro nd->last = this; 18215f4a6a69SAl Viro nd->last_type = type; 18225f4a6a69SAl Viro 1823200e9ef7SLinus Torvalds if (!name[len]) 18245f4a6a69SAl Viro return 0; 1825200e9ef7SLinus Torvalds /* 1826200e9ef7SLinus Torvalds * If it wasn't NUL, we know it was '/'. Skip that 1827200e9ef7SLinus Torvalds * slash, and continue until no more slashes. 1828200e9ef7SLinus Torvalds */ 1829200e9ef7SLinus Torvalds do { 1830200e9ef7SLinus Torvalds len++; 1831200e9ef7SLinus Torvalds } while (unlikely(name[len] == '/')); 1832200e9ef7SLinus Torvalds if (!name[len]) 18335f4a6a69SAl Viro return 0; 18345f4a6a69SAl Viro 1835200e9ef7SLinus Torvalds name += len; 18361da177e4SLinus Torvalds 183721b9b073SAl Viro err = walk_component(nd, &next, LOOKUP_FOLLOW); 1838ce57dfc1SAl Viro if (err < 0) 1839ce57dfc1SAl Viro return err; 1840fe479a58SAl Viro 1841ce57dfc1SAl Viro if (err) { 1842b356379aSAl Viro err = nested_symlink(&next, nd); 18431da177e4SLinus Torvalds if (err) 1844a7472babSAl Viro return err; 184531e6b01fSNick Piggin } 18465f4a6a69SAl Viro if (!can_lookup(nd->inode)) { 18473ddcd056SLinus Torvalds err = -ENOTDIR; 18483ddcd056SLinus Torvalds break; 18495f4a6a69SAl Viro } 1850ce57dfc1SAl Viro } 1851951361f9SAl Viro terminate_walk(nd); 18521da177e4SLinus Torvalds return err; 18531da177e4SLinus Torvalds } 18541da177e4SLinus Torvalds 185570e9b357SAl Viro static int path_init(int dfd, const char *name, unsigned int flags, 185670e9b357SAl Viro struct nameidata *nd, struct file **fp) 185731e6b01fSNick Piggin { 185831e6b01fSNick Piggin int retval = 0; 185931e6b01fSNick Piggin 186031e6b01fSNick Piggin nd->last_type = LAST_ROOT; /* if there are only slashes... */ 186116c2cd71SAl Viro nd->flags = flags | LOOKUP_JUMPED; 186231e6b01fSNick Piggin nd->depth = 0; 18635b6ca027SAl Viro if (flags & LOOKUP_ROOT) { 18645b6ca027SAl Viro struct inode *inode = nd->root.dentry->d_inode; 186573d049a4SAl Viro if (*name) { 1866741b7c3fSAl Viro if (!can_lookup(inode)) 18675b6ca027SAl Viro return -ENOTDIR; 18685b6ca027SAl Viro retval = inode_permission(inode, MAY_EXEC); 18695b6ca027SAl Viro if (retval) 18705b6ca027SAl Viro return retval; 187173d049a4SAl Viro } 18725b6ca027SAl Viro nd->path = nd->root; 18735b6ca027SAl Viro nd->inode = inode; 18745b6ca027SAl Viro if (flags & LOOKUP_RCU) { 187532a7991bSAl Viro lock_rcu_walk(); 18765b6ca027SAl Viro nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 18775b6ca027SAl Viro } else { 18785b6ca027SAl Viro path_get(&nd->path); 18795b6ca027SAl Viro } 18805b6ca027SAl Viro return 0; 18815b6ca027SAl Viro } 18825b6ca027SAl Viro 188331e6b01fSNick Piggin nd->root.mnt = NULL; 188431e6b01fSNick Piggin 188531e6b01fSNick Piggin if (*name=='/') { 1886e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 188732a7991bSAl Viro lock_rcu_walk(); 1888e41f7d4eSAl Viro set_root_rcu(nd); 1889e41f7d4eSAl Viro } else { 1890e41f7d4eSAl Viro set_root(nd); 1891e41f7d4eSAl Viro path_get(&nd->root); 1892e41f7d4eSAl Viro } 189331e6b01fSNick Piggin nd->path = nd->root; 189431e6b01fSNick Piggin } else if (dfd == AT_FDCWD) { 1895e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 189631e6b01fSNick Piggin struct fs_struct *fs = current->fs; 1897c28cc364SNick Piggin unsigned seq; 189831e6b01fSNick Piggin 189932a7991bSAl Viro lock_rcu_walk(); 190031e6b01fSNick Piggin 1901c28cc364SNick Piggin do { 1902c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 190331e6b01fSNick Piggin nd->path = fs->pwd; 1904c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 1905c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 1906e41f7d4eSAl Viro } else { 1907e41f7d4eSAl Viro get_fs_pwd(current->fs, &nd->path); 1908e41f7d4eSAl Viro } 190931e6b01fSNick Piggin } else { 1910582aa64aSJeff Layton /* Caller must check execute permissions on the starting path component */ 19112903ff01SAl Viro struct fd f = fdget_raw(dfd); 191231e6b01fSNick Piggin struct dentry *dentry; 191331e6b01fSNick Piggin 19142903ff01SAl Viro if (!f.file) 19152903ff01SAl Viro return -EBADF; 191631e6b01fSNick Piggin 19172903ff01SAl Viro dentry = f.file->f_path.dentry; 191831e6b01fSNick Piggin 1919f52e0c11SAl Viro if (*name) { 1920741b7c3fSAl Viro if (!can_lookup(dentry->d_inode)) { 19212903ff01SAl Viro fdput(f); 19222903ff01SAl Viro return -ENOTDIR; 1923f52e0c11SAl Viro } 19242903ff01SAl Viro } 19252903ff01SAl Viro 19262903ff01SAl Viro nd->path = f.file->f_path; 1927e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 19282903ff01SAl Viro if (f.need_put) 19292903ff01SAl Viro *fp = f.file; 1930c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 193132a7991bSAl Viro lock_rcu_walk(); 19325590ff0dSUlrich Drepper } else { 19332903ff01SAl Viro path_get(&nd->path); 19342903ff01SAl Viro fdput(f); 19351da177e4SLinus Torvalds } 1936e41f7d4eSAl Viro } 1937e41f7d4eSAl Viro 193831e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 19399b4a9b14SAl Viro return 0; 19409b4a9b14SAl Viro } 19419b4a9b14SAl Viro 1942bd92d7feSAl Viro static inline int lookup_last(struct nameidata *nd, struct path *path) 1943bd92d7feSAl Viro { 1944bd92d7feSAl Viro if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) 1945bd92d7feSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 1946bd92d7feSAl Viro 1947bd92d7feSAl Viro nd->flags &= ~LOOKUP_PARENT; 194821b9b073SAl Viro return walk_component(nd, path, nd->flags & LOOKUP_FOLLOW); 1949bd92d7feSAl Viro } 1950bd92d7feSAl Viro 19519b4a9b14SAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 1952ee0827cdSAl Viro static int path_lookupat(int dfd, const char *name, 19539b4a9b14SAl Viro unsigned int flags, struct nameidata *nd) 19549b4a9b14SAl Viro { 195570e9b357SAl Viro struct file *base = NULL; 1956bd92d7feSAl Viro struct path path; 1957bd92d7feSAl Viro int err; 195831e6b01fSNick Piggin 195931e6b01fSNick Piggin /* 196031e6b01fSNick Piggin * Path walking is largely split up into 2 different synchronisation 196131e6b01fSNick Piggin * schemes, rcu-walk and ref-walk (explained in 196231e6b01fSNick Piggin * Documentation/filesystems/path-lookup.txt). These share much of the 196331e6b01fSNick Piggin * path walk code, but some things particularly setup, cleanup, and 196431e6b01fSNick Piggin * following mounts are sufficiently divergent that functions are 196531e6b01fSNick Piggin * duplicated. Typically there is a function foo(), and its RCU 196631e6b01fSNick Piggin * analogue, foo_rcu(). 196731e6b01fSNick Piggin * 196831e6b01fSNick Piggin * -ECHILD is the error number of choice (just to avoid clashes) that 196931e6b01fSNick Piggin * is returned if some aspect of an rcu-walk fails. Such an error must 197031e6b01fSNick Piggin * be handled by restarting a traditional ref-walk (which will always 197131e6b01fSNick Piggin * be able to complete). 197231e6b01fSNick Piggin */ 1973bd92d7feSAl Viro err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base); 1974ee0827cdSAl Viro 1975bd92d7feSAl Viro if (unlikely(err)) 1976bd92d7feSAl Viro return err; 1977ee0827cdSAl Viro 1978ee0827cdSAl Viro current->total_link_count = 0; 1979bd92d7feSAl Viro err = link_path_walk(name, nd); 1980bd92d7feSAl Viro 1981bd92d7feSAl Viro if (!err && !(flags & LOOKUP_PARENT)) { 1982bd92d7feSAl Viro err = lookup_last(nd, &path); 1983bd92d7feSAl Viro while (err > 0) { 1984bd92d7feSAl Viro void *cookie; 1985bd92d7feSAl Viro struct path link = path; 1986800179c9SKees Cook err = may_follow_link(&link, nd); 1987800179c9SKees Cook if (unlikely(err)) 1988800179c9SKees Cook break; 1989bd92d7feSAl Viro nd->flags |= LOOKUP_PARENT; 1990574197e0SAl Viro err = follow_link(&link, nd, &cookie); 19916d7b5aaeSAl Viro if (err) 19926d7b5aaeSAl Viro break; 1993bd92d7feSAl Viro err = lookup_last(nd, &path); 1994574197e0SAl Viro put_link(nd, &link, cookie); 1995bd92d7feSAl Viro } 1996bd92d7feSAl Viro } 1997ee0827cdSAl Viro 19989f1fafeeSAl Viro if (!err) 19999f1fafeeSAl Viro err = complete_walk(nd); 2000bd92d7feSAl Viro 2001bd92d7feSAl Viro if (!err && nd->flags & LOOKUP_DIRECTORY) { 200205252901SAl Viro if (!can_lookup(nd->inode)) { 2003bd92d7feSAl Viro path_put(&nd->path); 2004bd23a539SAl Viro err = -ENOTDIR; 2005bd92d7feSAl Viro } 2006bd92d7feSAl Viro } 200716c2cd71SAl Viro 200870e9b357SAl Viro if (base) 200970e9b357SAl Viro fput(base); 2010ee0827cdSAl Viro 20115b6ca027SAl Viro if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 201231e6b01fSNick Piggin path_put(&nd->root); 201331e6b01fSNick Piggin nd->root.mnt = NULL; 201431e6b01fSNick Piggin } 2015bd92d7feSAl Viro return err; 201631e6b01fSNick Piggin } 201731e6b01fSNick Piggin 2018873f1eedSJeff Layton static int filename_lookup(int dfd, struct filename *name, 2019873f1eedSJeff Layton unsigned int flags, struct nameidata *nd) 2020873f1eedSJeff Layton { 2021873f1eedSJeff Layton int retval = path_lookupat(dfd, name->name, flags | LOOKUP_RCU, nd); 2022873f1eedSJeff Layton if (unlikely(retval == -ECHILD)) 2023873f1eedSJeff Layton retval = path_lookupat(dfd, name->name, flags, nd); 2024873f1eedSJeff Layton if (unlikely(retval == -ESTALE)) 2025873f1eedSJeff Layton retval = path_lookupat(dfd, name->name, 2026873f1eedSJeff Layton flags | LOOKUP_REVAL, nd); 2027873f1eedSJeff Layton 2028873f1eedSJeff Layton if (likely(!retval)) 2029adb5c247SJeff Layton audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT); 2030873f1eedSJeff Layton return retval; 2031873f1eedSJeff Layton } 2032873f1eedSJeff Layton 2033ee0827cdSAl Viro static int do_path_lookup(int dfd, const char *name, 2034ee0827cdSAl Viro unsigned int flags, struct nameidata *nd) 2035ee0827cdSAl Viro { 2036873f1eedSJeff Layton struct filename filename = { .name = name }; 2037ee0827cdSAl Viro 2038873f1eedSJeff Layton return filename_lookup(dfd, &filename, flags, nd); 20391da177e4SLinus Torvalds } 20401da177e4SLinus Torvalds 204179714f72SAl Viro /* does lookup, returns the object with parent locked */ 204279714f72SAl Viro struct dentry *kern_path_locked(const char *name, struct path *path) 20435590ff0dSUlrich Drepper { 204479714f72SAl Viro struct nameidata nd; 204579714f72SAl Viro struct dentry *d; 204679714f72SAl Viro int err = do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, &nd); 204779714f72SAl Viro if (err) 204879714f72SAl Viro return ERR_PTR(err); 204979714f72SAl Viro if (nd.last_type != LAST_NORM) { 205079714f72SAl Viro path_put(&nd.path); 205179714f72SAl Viro return ERR_PTR(-EINVAL); 205279714f72SAl Viro } 205379714f72SAl Viro mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 20541e0ea001SAl Viro d = __lookup_hash(&nd.last, nd.path.dentry, 0); 205579714f72SAl Viro if (IS_ERR(d)) { 205679714f72SAl Viro mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 205779714f72SAl Viro path_put(&nd.path); 205879714f72SAl Viro return d; 205979714f72SAl Viro } 206079714f72SAl Viro *path = nd.path; 206179714f72SAl Viro return d; 20625590ff0dSUlrich Drepper } 20635590ff0dSUlrich Drepper 2064d1811465SAl Viro int kern_path(const char *name, unsigned int flags, struct path *path) 2065d1811465SAl Viro { 2066d1811465SAl Viro struct nameidata nd; 2067d1811465SAl Viro int res = do_path_lookup(AT_FDCWD, name, flags, &nd); 2068d1811465SAl Viro if (!res) 2069d1811465SAl Viro *path = nd.path; 2070d1811465SAl Viro return res; 2071d1811465SAl Viro } 2072d1811465SAl Viro 207316f18200SJosef 'Jeff' Sipek /** 207416f18200SJosef 'Jeff' Sipek * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair 207516f18200SJosef 'Jeff' Sipek * @dentry: pointer to dentry of the base directory 207616f18200SJosef 'Jeff' Sipek * @mnt: pointer to vfs mount of the base directory 207716f18200SJosef 'Jeff' Sipek * @name: pointer to file name 207816f18200SJosef 'Jeff' Sipek * @flags: lookup flags 2079e0a01249SAl Viro * @path: pointer to struct path to fill 208016f18200SJosef 'Jeff' Sipek */ 208116f18200SJosef 'Jeff' Sipek int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, 208216f18200SJosef 'Jeff' Sipek const char *name, unsigned int flags, 2083e0a01249SAl Viro struct path *path) 208416f18200SJosef 'Jeff' Sipek { 2085e0a01249SAl Viro struct nameidata nd; 2086e0a01249SAl Viro int err; 2087e0a01249SAl Viro nd.root.dentry = dentry; 2088e0a01249SAl Viro nd.root.mnt = mnt; 2089e0a01249SAl Viro BUG_ON(flags & LOOKUP_PARENT); 20905b6ca027SAl Viro /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */ 2091e0a01249SAl Viro err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd); 2092e0a01249SAl Viro if (!err) 2093e0a01249SAl Viro *path = nd.path; 2094e0a01249SAl Viro return err; 209516f18200SJosef 'Jeff' Sipek } 209616f18200SJosef 'Jeff' Sipek 2097057f6c01SJames Morris /* 2098057f6c01SJames Morris * Restricted form of lookup. Doesn't follow links, single-component only, 2099057f6c01SJames Morris * needs parent already locked. Doesn't follow mounts. 2100057f6c01SJames Morris * SMP-safe. 2101057f6c01SJames Morris */ 2102a244e169SAdrian Bunk static struct dentry *lookup_hash(struct nameidata *nd) 21031da177e4SLinus Torvalds { 210472bd866aSAl Viro return __lookup_hash(&nd->last, nd->path.dentry, nd->flags); 21051da177e4SLinus Torvalds } 21061da177e4SLinus Torvalds 2107eead1911SChristoph Hellwig /** 2108a6b91919SRandy Dunlap * lookup_one_len - filesystem helper to lookup single pathname component 2109eead1911SChristoph Hellwig * @name: pathname component to lookup 2110eead1911SChristoph Hellwig * @base: base directory to lookup from 2111eead1911SChristoph Hellwig * @len: maximum length @len should be interpreted to 2112eead1911SChristoph Hellwig * 2113a6b91919SRandy Dunlap * Note that this routine is purely a helper for filesystem usage and should 2114a6b91919SRandy Dunlap * not be called by generic code. Also note that by using this function the 2115eead1911SChristoph Hellwig * nameidata argument is passed to the filesystem methods and a filesystem 2116eead1911SChristoph Hellwig * using this helper needs to be prepared for that. 2117eead1911SChristoph Hellwig */ 2118057f6c01SJames Morris struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 2119057f6c01SJames Morris { 2120057f6c01SJames Morris struct qstr this; 21216a96ba54SAl Viro unsigned int c; 2122cda309deSMiklos Szeredi int err; 2123057f6c01SJames Morris 21242f9092e1SDavid Woodhouse WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex)); 21252f9092e1SDavid Woodhouse 21266a96ba54SAl Viro this.name = name; 21276a96ba54SAl Viro this.len = len; 21280145acc2SLinus Torvalds this.hash = full_name_hash(name, len); 21296a96ba54SAl Viro if (!len) 21306a96ba54SAl Viro return ERR_PTR(-EACCES); 21316a96ba54SAl Viro 213221d8a15aSAl Viro if (unlikely(name[0] == '.')) { 213321d8a15aSAl Viro if (len < 2 || (len == 2 && name[1] == '.')) 213421d8a15aSAl Viro return ERR_PTR(-EACCES); 213521d8a15aSAl Viro } 213621d8a15aSAl Viro 21376a96ba54SAl Viro while (len--) { 21386a96ba54SAl Viro c = *(const unsigned char *)name++; 21396a96ba54SAl Viro if (c == '/' || c == '\0') 21406a96ba54SAl Viro return ERR_PTR(-EACCES); 21416a96ba54SAl Viro } 21425a202bcdSAl Viro /* 21435a202bcdSAl Viro * See if the low-level filesystem might want 21445a202bcdSAl Viro * to use its own hash.. 21455a202bcdSAl Viro */ 21465a202bcdSAl Viro if (base->d_flags & DCACHE_OP_HASH) { 2147da53be12SLinus Torvalds int err = base->d_op->d_hash(base, &this); 21485a202bcdSAl Viro if (err < 0) 21495a202bcdSAl Viro return ERR_PTR(err); 21505a202bcdSAl Viro } 2151eead1911SChristoph Hellwig 2152cda309deSMiklos Szeredi err = inode_permission(base->d_inode, MAY_EXEC); 2153cda309deSMiklos Szeredi if (err) 2154cda309deSMiklos Szeredi return ERR_PTR(err); 2155cda309deSMiklos Szeredi 215672bd866aSAl Viro return __lookup_hash(&this, base, 0); 2157057f6c01SJames Morris } 2158057f6c01SJames Morris 21591fa1e7f6SAndy Whitcroft int user_path_at_empty(int dfd, const char __user *name, unsigned flags, 21601fa1e7f6SAndy Whitcroft struct path *path, int *empty) 21611da177e4SLinus Torvalds { 21622d8f3038SAl Viro struct nameidata nd; 216391a27b2aSJeff Layton struct filename *tmp = getname_flags(name, flags, empty); 21641da177e4SLinus Torvalds int err = PTR_ERR(tmp); 21651da177e4SLinus Torvalds if (!IS_ERR(tmp)) { 21662d8f3038SAl Viro 21672d8f3038SAl Viro BUG_ON(flags & LOOKUP_PARENT); 21682d8f3038SAl Viro 2169873f1eedSJeff Layton err = filename_lookup(dfd, tmp, flags, &nd); 21701da177e4SLinus Torvalds putname(tmp); 21712d8f3038SAl Viro if (!err) 21722d8f3038SAl Viro *path = nd.path; 21731da177e4SLinus Torvalds } 21741da177e4SLinus Torvalds return err; 21751da177e4SLinus Torvalds } 21761da177e4SLinus Torvalds 21771fa1e7f6SAndy Whitcroft int user_path_at(int dfd, const char __user *name, unsigned flags, 21781fa1e7f6SAndy Whitcroft struct path *path) 21791fa1e7f6SAndy Whitcroft { 2180f7493e5dSLinus Torvalds return user_path_at_empty(dfd, name, flags, path, NULL); 21811fa1e7f6SAndy Whitcroft } 21821fa1e7f6SAndy Whitcroft 2183873f1eedSJeff Layton /* 2184873f1eedSJeff Layton * NB: most callers don't do anything directly with the reference to the 2185873f1eedSJeff Layton * to struct filename, but the nd->last pointer points into the name string 2186873f1eedSJeff Layton * allocated by getname. So we must hold the reference to it until all 2187873f1eedSJeff Layton * path-walking is complete. 2188873f1eedSJeff Layton */ 218991a27b2aSJeff Layton static struct filename * 21909e790bd6SJeff Layton user_path_parent(int dfd, const char __user *path, struct nameidata *nd, 21919e790bd6SJeff Layton unsigned int flags) 21922ad94ae6SAl Viro { 219391a27b2aSJeff Layton struct filename *s = getname(path); 21942ad94ae6SAl Viro int error; 21952ad94ae6SAl Viro 21969e790bd6SJeff Layton /* only LOOKUP_REVAL is allowed in extra flags */ 21979e790bd6SJeff Layton flags &= LOOKUP_REVAL; 21989e790bd6SJeff Layton 21992ad94ae6SAl Viro if (IS_ERR(s)) 220091a27b2aSJeff Layton return s; 22012ad94ae6SAl Viro 22029e790bd6SJeff Layton error = filename_lookup(dfd, s, flags | LOOKUP_PARENT, nd); 220391a27b2aSJeff Layton if (error) { 22042ad94ae6SAl Viro putname(s); 220591a27b2aSJeff Layton return ERR_PTR(error); 220691a27b2aSJeff Layton } 22072ad94ae6SAl Viro 220891a27b2aSJeff Layton return s; 22092ad94ae6SAl Viro } 22102ad94ae6SAl Viro 22118033426eSJeff Layton /** 22128033426eSJeff Layton * umount_lookup_last - look up last component for umount 22138033426eSJeff Layton * @nd: pathwalk nameidata - currently pointing at parent directory of "last" 22148033426eSJeff Layton * @path: pointer to container for result 22158033426eSJeff Layton * 22168033426eSJeff Layton * This is a special lookup_last function just for umount. In this case, we 22178033426eSJeff Layton * need to resolve the path without doing any revalidation. 22188033426eSJeff Layton * 22198033426eSJeff Layton * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since 22208033426eSJeff Layton * mountpoints are always pinned in the dcache, their ancestors are too. Thus, 22218033426eSJeff Layton * in almost all cases, this lookup will be served out of the dcache. The only 22228033426eSJeff Layton * cases where it won't are if nd->last refers to a symlink or the path is 22238033426eSJeff Layton * bogus and it doesn't exist. 22248033426eSJeff Layton * 22258033426eSJeff Layton * Returns: 22268033426eSJeff Layton * -error: if there was an error during lookup. This includes -ENOENT if the 22278033426eSJeff Layton * lookup found a negative dentry. The nd->path reference will also be 22288033426eSJeff Layton * put in this case. 22298033426eSJeff Layton * 22308033426eSJeff Layton * 0: if we successfully resolved nd->path and found it to not to be a 22318033426eSJeff Layton * symlink that needs to be followed. "path" will also be populated. 22328033426eSJeff Layton * The nd->path reference will also be put. 22338033426eSJeff Layton * 22348033426eSJeff Layton * 1: if we successfully resolved nd->last and found it to be a symlink 22358033426eSJeff Layton * that needs to be followed. "path" will be populated with the path 22368033426eSJeff Layton * to the link, and nd->path will *not* be put. 22378033426eSJeff Layton */ 22388033426eSJeff Layton static int 22398033426eSJeff Layton umount_lookup_last(struct nameidata *nd, struct path *path) 22408033426eSJeff Layton { 22418033426eSJeff Layton int error = 0; 22428033426eSJeff Layton struct dentry *dentry; 22438033426eSJeff Layton struct dentry *dir = nd->path.dentry; 22448033426eSJeff Layton 22458033426eSJeff Layton if (unlikely(nd->flags & LOOKUP_RCU)) { 22468033426eSJeff Layton WARN_ON_ONCE(1); 22478033426eSJeff Layton error = -ECHILD; 22488033426eSJeff Layton goto error_check; 22498033426eSJeff Layton } 22508033426eSJeff Layton 22518033426eSJeff Layton nd->flags &= ~LOOKUP_PARENT; 22528033426eSJeff Layton 22538033426eSJeff Layton if (unlikely(nd->last_type != LAST_NORM)) { 22548033426eSJeff Layton error = handle_dots(nd, nd->last_type); 22558033426eSJeff Layton if (!error) 22568033426eSJeff Layton dentry = dget(nd->path.dentry); 22578033426eSJeff Layton goto error_check; 22588033426eSJeff Layton } 22598033426eSJeff Layton 22608033426eSJeff Layton mutex_lock(&dir->d_inode->i_mutex); 22618033426eSJeff Layton dentry = d_lookup(dir, &nd->last); 22628033426eSJeff Layton if (!dentry) { 22638033426eSJeff Layton /* 22648033426eSJeff Layton * No cached dentry. Mounted dentries are pinned in the cache, 22658033426eSJeff Layton * so that means that this dentry is probably a symlink or the 22668033426eSJeff Layton * path doesn't actually point to a mounted dentry. 22678033426eSJeff Layton */ 22688033426eSJeff Layton dentry = d_alloc(dir, &nd->last); 22698033426eSJeff Layton if (!dentry) { 22708033426eSJeff Layton error = -ENOMEM; 22718033426eSJeff Layton } else { 22728033426eSJeff Layton dentry = lookup_real(dir->d_inode, dentry, nd->flags); 22738033426eSJeff Layton if (IS_ERR(dentry)) 22748033426eSJeff Layton error = PTR_ERR(dentry); 22758033426eSJeff Layton } 22768033426eSJeff Layton } 22778033426eSJeff Layton mutex_unlock(&dir->d_inode->i_mutex); 22788033426eSJeff Layton 22798033426eSJeff Layton error_check: 22808033426eSJeff Layton if (!error) { 22818033426eSJeff Layton if (!dentry->d_inode) { 22828033426eSJeff Layton error = -ENOENT; 22838033426eSJeff Layton dput(dentry); 22848033426eSJeff Layton } else { 22858033426eSJeff Layton path->dentry = dentry; 22868033426eSJeff Layton path->mnt = mntget(nd->path.mnt); 22878033426eSJeff Layton if (should_follow_link(dentry->d_inode, 22888033426eSJeff Layton nd->flags & LOOKUP_FOLLOW)) 22898033426eSJeff Layton return 1; 22908033426eSJeff Layton follow_mount(path); 22918033426eSJeff Layton } 22928033426eSJeff Layton } 22938033426eSJeff Layton terminate_walk(nd); 22948033426eSJeff Layton return error; 22958033426eSJeff Layton } 22968033426eSJeff Layton 22978033426eSJeff Layton /** 22988033426eSJeff Layton * path_umountat - look up a path to be umounted 22998033426eSJeff Layton * @dfd: directory file descriptor to start walk from 23008033426eSJeff Layton * @name: full pathname to walk 23018033426eSJeff Layton * @flags: lookup flags 23028033426eSJeff Layton * @nd: pathwalk nameidata 23038033426eSJeff Layton * 23048033426eSJeff Layton * Look up the given name, but don't attempt to revalidate the last component. 23058033426eSJeff Layton * Returns 0 and "path" will be valid on success; Retuns error otherwise. 23068033426eSJeff Layton */ 23078033426eSJeff Layton static int 23088033426eSJeff Layton path_umountat(int dfd, const char *name, struct path *path, unsigned int flags) 23098033426eSJeff Layton { 23108033426eSJeff Layton struct file *base = NULL; 23118033426eSJeff Layton struct nameidata nd; 23128033426eSJeff Layton int err; 23138033426eSJeff Layton 23148033426eSJeff Layton err = path_init(dfd, name, flags | LOOKUP_PARENT, &nd, &base); 23158033426eSJeff Layton if (unlikely(err)) 23168033426eSJeff Layton return err; 23178033426eSJeff Layton 23188033426eSJeff Layton current->total_link_count = 0; 23198033426eSJeff Layton err = link_path_walk(name, &nd); 23208033426eSJeff Layton if (err) 23218033426eSJeff Layton goto out; 23228033426eSJeff Layton 23238033426eSJeff Layton /* If we're in rcuwalk, drop out of it to handle last component */ 23248033426eSJeff Layton if (nd.flags & LOOKUP_RCU) { 23258033426eSJeff Layton err = unlazy_walk(&nd, NULL); 23268033426eSJeff Layton if (err) { 23278033426eSJeff Layton terminate_walk(&nd); 23288033426eSJeff Layton goto out; 23298033426eSJeff Layton } 23308033426eSJeff Layton } 23318033426eSJeff Layton 23328033426eSJeff Layton err = umount_lookup_last(&nd, path); 23338033426eSJeff Layton while (err > 0) { 23348033426eSJeff Layton void *cookie; 23358033426eSJeff Layton struct path link = *path; 23368033426eSJeff Layton err = may_follow_link(&link, &nd); 23378033426eSJeff Layton if (unlikely(err)) 23388033426eSJeff Layton break; 23398033426eSJeff Layton nd.flags |= LOOKUP_PARENT; 23408033426eSJeff Layton err = follow_link(&link, &nd, &cookie); 23418033426eSJeff Layton if (err) 23428033426eSJeff Layton break; 23438033426eSJeff Layton err = umount_lookup_last(&nd, path); 23448033426eSJeff Layton put_link(&nd, &link, cookie); 23458033426eSJeff Layton } 23468033426eSJeff Layton out: 23478033426eSJeff Layton if (base) 23488033426eSJeff Layton fput(base); 23498033426eSJeff Layton 23508033426eSJeff Layton if (nd.root.mnt && !(nd.flags & LOOKUP_ROOT)) 23518033426eSJeff Layton path_put(&nd.root); 23528033426eSJeff Layton 23538033426eSJeff Layton return err; 23548033426eSJeff Layton } 23558033426eSJeff Layton 23568033426eSJeff Layton /** 23578033426eSJeff Layton * user_path_umountat - lookup a path from userland in order to umount it 23588033426eSJeff Layton * @dfd: directory file descriptor 23598033426eSJeff Layton * @name: pathname from userland 23608033426eSJeff Layton * @flags: lookup flags 23618033426eSJeff Layton * @path: pointer to container to hold result 23628033426eSJeff Layton * 23638033426eSJeff Layton * A umount is a special case for path walking. We're not actually interested 23648033426eSJeff Layton * in the inode in this situation, and ESTALE errors can be a problem. We 23658033426eSJeff Layton * simply want track down the dentry and vfsmount attached at the mountpoint 23668033426eSJeff Layton * and avoid revalidating the last component. 23678033426eSJeff Layton * 23688033426eSJeff Layton * Returns 0 and populates "path" on success. 23698033426eSJeff Layton */ 23708033426eSJeff Layton int 23718033426eSJeff Layton user_path_umountat(int dfd, const char __user *name, unsigned int flags, 23728033426eSJeff Layton struct path *path) 23738033426eSJeff Layton { 23748033426eSJeff Layton struct filename *s = getname(name); 23758033426eSJeff Layton int error; 23768033426eSJeff Layton 23778033426eSJeff Layton if (IS_ERR(s)) 23788033426eSJeff Layton return PTR_ERR(s); 23798033426eSJeff Layton 23808033426eSJeff Layton error = path_umountat(dfd, s->name, path, flags | LOOKUP_RCU); 23818033426eSJeff Layton if (unlikely(error == -ECHILD)) 23828033426eSJeff Layton error = path_umountat(dfd, s->name, path, flags); 23838033426eSJeff Layton if (unlikely(error == -ESTALE)) 23848033426eSJeff Layton error = path_umountat(dfd, s->name, path, flags | LOOKUP_REVAL); 23858033426eSJeff Layton 23868033426eSJeff Layton if (likely(!error)) 23878033426eSJeff Layton audit_inode(s, path->dentry, 0); 23888033426eSJeff Layton 23898033426eSJeff Layton putname(s); 23908033426eSJeff Layton return error; 23918033426eSJeff Layton } 23928033426eSJeff Layton 23931da177e4SLinus Torvalds /* 23941da177e4SLinus Torvalds * It's inline, so penalty for filesystems that don't use sticky bit is 23951da177e4SLinus Torvalds * minimal. 23961da177e4SLinus Torvalds */ 23971da177e4SLinus Torvalds static inline int check_sticky(struct inode *dir, struct inode *inode) 23981da177e4SLinus Torvalds { 23998e96e3b7SEric W. Biederman kuid_t fsuid = current_fsuid(); 2400da9592edSDavid Howells 24011da177e4SLinus Torvalds if (!(dir->i_mode & S_ISVTX)) 24021da177e4SLinus Torvalds return 0; 24038e96e3b7SEric W. Biederman if (uid_eq(inode->i_uid, fsuid)) 24041da177e4SLinus Torvalds return 0; 24058e96e3b7SEric W. Biederman if (uid_eq(dir->i_uid, fsuid)) 24061da177e4SLinus Torvalds return 0; 24071a48e2acSEric W. Biederman return !inode_capable(inode, CAP_FOWNER); 24081da177e4SLinus Torvalds } 24091da177e4SLinus Torvalds 24101da177e4SLinus Torvalds /* 24111da177e4SLinus Torvalds * Check whether we can remove a link victim from directory dir, check 24121da177e4SLinus Torvalds * whether the type of victim is right. 24131da177e4SLinus Torvalds * 1. We can't do it if dir is read-only (done in permission()) 24141da177e4SLinus Torvalds * 2. We should have write and exec permissions on dir 24151da177e4SLinus Torvalds * 3. We can't remove anything from append-only dir 24161da177e4SLinus Torvalds * 4. We can't do anything with immutable dir (done in permission()) 24171da177e4SLinus Torvalds * 5. If the sticky bit on dir is set we should either 24181da177e4SLinus Torvalds * a. be owner of dir, or 24191da177e4SLinus Torvalds * b. be owner of victim, or 24201da177e4SLinus Torvalds * c. have CAP_FOWNER capability 24211da177e4SLinus Torvalds * 6. If the victim is append-only or immutable we can't do antyhing with 24221da177e4SLinus Torvalds * links pointing to it. 24231da177e4SLinus Torvalds * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. 24241da177e4SLinus Torvalds * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. 24251da177e4SLinus Torvalds * 9. We can't remove a root or mountpoint. 24261da177e4SLinus Torvalds * 10. We don't allow removal of NFS sillyrenamed files; it's handled by 24271da177e4SLinus Torvalds * nfs_async_unlink(). 24281da177e4SLinus Torvalds */ 2429858119e1SArjan van de Ven static int may_delete(struct inode *dir,struct dentry *victim,int isdir) 24301da177e4SLinus Torvalds { 24311da177e4SLinus Torvalds int error; 24321da177e4SLinus Torvalds 24331da177e4SLinus Torvalds if (!victim->d_inode) 24341da177e4SLinus Torvalds return -ENOENT; 24351da177e4SLinus Torvalds 24361da177e4SLinus Torvalds BUG_ON(victim->d_parent->d_inode != dir); 24374fa6b5ecSJeff Layton audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); 24381da177e4SLinus Torvalds 2439f419a2e3SAl Viro error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 24401da177e4SLinus Torvalds if (error) 24411da177e4SLinus Torvalds return error; 24421da177e4SLinus Torvalds if (IS_APPEND(dir)) 24431da177e4SLinus Torvalds return -EPERM; 24441da177e4SLinus Torvalds if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)|| 2445f9454548SHugh Dickins IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode)) 24461da177e4SLinus Torvalds return -EPERM; 24471da177e4SLinus Torvalds if (isdir) { 24481da177e4SLinus Torvalds if (!S_ISDIR(victim->d_inode->i_mode)) 24491da177e4SLinus Torvalds return -ENOTDIR; 24501da177e4SLinus Torvalds if (IS_ROOT(victim)) 24511da177e4SLinus Torvalds return -EBUSY; 24521da177e4SLinus Torvalds } else if (S_ISDIR(victim->d_inode->i_mode)) 24531da177e4SLinus Torvalds return -EISDIR; 24541da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 24551da177e4SLinus Torvalds return -ENOENT; 24561da177e4SLinus Torvalds if (victim->d_flags & DCACHE_NFSFS_RENAMED) 24571da177e4SLinus Torvalds return -EBUSY; 24581da177e4SLinus Torvalds return 0; 24591da177e4SLinus Torvalds } 24601da177e4SLinus Torvalds 24611da177e4SLinus Torvalds /* Check whether we can create an object with dentry child in directory 24621da177e4SLinus Torvalds * dir. 24631da177e4SLinus Torvalds * 1. We can't do it if child already exists (open has special treatment for 24641da177e4SLinus Torvalds * this case, but since we are inlined it's OK) 24651da177e4SLinus Torvalds * 2. We can't do it if dir is read-only (done in permission()) 24661da177e4SLinus Torvalds * 3. We should have write and exec permissions on dir 24671da177e4SLinus Torvalds * 4. We can't do it if dir is immutable (done in permission()) 24681da177e4SLinus Torvalds */ 2469a95164d9SMiklos Szeredi static inline int may_create(struct inode *dir, struct dentry *child) 24701da177e4SLinus Torvalds { 24711da177e4SLinus Torvalds if (child->d_inode) 24721da177e4SLinus Torvalds return -EEXIST; 24731da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 24741da177e4SLinus Torvalds return -ENOENT; 2475f419a2e3SAl Viro return inode_permission(dir, MAY_WRITE | MAY_EXEC); 24761da177e4SLinus Torvalds } 24771da177e4SLinus Torvalds 24781da177e4SLinus Torvalds /* 24791da177e4SLinus Torvalds * p1 and p2 should be directories on the same fs. 24801da177e4SLinus Torvalds */ 24811da177e4SLinus Torvalds struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 24821da177e4SLinus Torvalds { 24831da177e4SLinus Torvalds struct dentry *p; 24841da177e4SLinus Torvalds 24851da177e4SLinus Torvalds if (p1 == p2) { 2486f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 24871da177e4SLinus Torvalds return NULL; 24881da177e4SLinus Torvalds } 24891da177e4SLinus Torvalds 2490a11f3a05SArjan van de Ven mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 24911da177e4SLinus Torvalds 2492e2761a11SOGAWA Hirofumi p = d_ancestor(p2, p1); 2493e2761a11SOGAWA Hirofumi if (p) { 2494f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT); 2495f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD); 24961da177e4SLinus Torvalds return p; 24971da177e4SLinus Torvalds } 24981da177e4SLinus Torvalds 2499e2761a11SOGAWA Hirofumi p = d_ancestor(p1, p2); 2500e2761a11SOGAWA Hirofumi if (p) { 2501f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 2502f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); 25031da177e4SLinus Torvalds return p; 25041da177e4SLinus Torvalds } 25051da177e4SLinus Torvalds 2506f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 2507f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); 25081da177e4SLinus Torvalds return NULL; 25091da177e4SLinus Torvalds } 25101da177e4SLinus Torvalds 25111da177e4SLinus Torvalds void unlock_rename(struct dentry *p1, struct dentry *p2) 25121da177e4SLinus Torvalds { 25131b1dcc1bSJes Sorensen mutex_unlock(&p1->d_inode->i_mutex); 25141da177e4SLinus Torvalds if (p1 != p2) { 25151b1dcc1bSJes Sorensen mutex_unlock(&p2->d_inode->i_mutex); 2516a11f3a05SArjan van de Ven mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 25171da177e4SLinus Torvalds } 25181da177e4SLinus Torvalds } 25191da177e4SLinus Torvalds 25204acdaf27SAl Viro int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, 2521312b63fbSAl Viro bool want_excl) 25221da177e4SLinus Torvalds { 2523a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 25241da177e4SLinus Torvalds if (error) 25251da177e4SLinus Torvalds return error; 25261da177e4SLinus Torvalds 2527acfa4380SAl Viro if (!dir->i_op->create) 25281da177e4SLinus Torvalds return -EACCES; /* shouldn't it be ENOSYS? */ 25291da177e4SLinus Torvalds mode &= S_IALLUGO; 25301da177e4SLinus Torvalds mode |= S_IFREG; 25311da177e4SLinus Torvalds error = security_inode_create(dir, dentry, mode); 25321da177e4SLinus Torvalds if (error) 25331da177e4SLinus Torvalds return error; 2534312b63fbSAl Viro error = dir->i_op->create(dir, dentry, mode, want_excl); 2535a74574aaSStephen Smalley if (!error) 2536f38aa942SAmy Griffis fsnotify_create(dir, dentry); 25371da177e4SLinus Torvalds return error; 25381da177e4SLinus Torvalds } 25391da177e4SLinus Torvalds 254073d049a4SAl Viro static int may_open(struct path *path, int acc_mode, int flag) 25411da177e4SLinus Torvalds { 25423fb64190SChristoph Hellwig struct dentry *dentry = path->dentry; 25431da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 25441da177e4SLinus Torvalds int error; 25451da177e4SLinus Torvalds 2546bcda7652SAl Viro /* O_PATH? */ 2547bcda7652SAl Viro if (!acc_mode) 2548bcda7652SAl Viro return 0; 2549bcda7652SAl Viro 25501da177e4SLinus Torvalds if (!inode) 25511da177e4SLinus Torvalds return -ENOENT; 25521da177e4SLinus Torvalds 2553c8fe8f30SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 2554c8fe8f30SChristoph Hellwig case S_IFLNK: 25551da177e4SLinus Torvalds return -ELOOP; 2556c8fe8f30SChristoph Hellwig case S_IFDIR: 2557c8fe8f30SChristoph Hellwig if (acc_mode & MAY_WRITE) 25581da177e4SLinus Torvalds return -EISDIR; 2559c8fe8f30SChristoph Hellwig break; 2560c8fe8f30SChristoph Hellwig case S_IFBLK: 2561c8fe8f30SChristoph Hellwig case S_IFCHR: 25623fb64190SChristoph Hellwig if (path->mnt->mnt_flags & MNT_NODEV) 25631da177e4SLinus Torvalds return -EACCES; 2564c8fe8f30SChristoph Hellwig /*FALLTHRU*/ 2565c8fe8f30SChristoph Hellwig case S_IFIFO: 2566c8fe8f30SChristoph Hellwig case S_IFSOCK: 25671da177e4SLinus Torvalds flag &= ~O_TRUNC; 2568c8fe8f30SChristoph Hellwig break; 25694a3fd211SDave Hansen } 2570b41572e9SDave Hansen 25713fb64190SChristoph Hellwig error = inode_permission(inode, acc_mode); 2572b41572e9SDave Hansen if (error) 2573b41572e9SDave Hansen return error; 25746146f0d5SMimi Zohar 25751da177e4SLinus Torvalds /* 25761da177e4SLinus Torvalds * An append-only file must be opened in append mode for writing. 25771da177e4SLinus Torvalds */ 25781da177e4SLinus Torvalds if (IS_APPEND(inode)) { 25798737c930SAl Viro if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) 25807715b521SAl Viro return -EPERM; 25811da177e4SLinus Torvalds if (flag & O_TRUNC) 25827715b521SAl Viro return -EPERM; 25831da177e4SLinus Torvalds } 25841da177e4SLinus Torvalds 25851da177e4SLinus Torvalds /* O_NOATIME can only be set by the owner or superuser */ 25862e149670SSerge E. Hallyn if (flag & O_NOATIME && !inode_owner_or_capable(inode)) 25877715b521SAl Viro return -EPERM; 25881da177e4SLinus Torvalds 2589f3c7691eSJ. Bruce Fields return 0; 25907715b521SAl Viro } 25917715b521SAl Viro 2592e1181ee6SJeff Layton static int handle_truncate(struct file *filp) 25937715b521SAl Viro { 2594e1181ee6SJeff Layton struct path *path = &filp->f_path; 25957715b521SAl Viro struct inode *inode = path->dentry->d_inode; 25967715b521SAl Viro int error = get_write_access(inode); 25971da177e4SLinus Torvalds if (error) 25987715b521SAl Viro return error; 25991da177e4SLinus Torvalds /* 26001da177e4SLinus Torvalds * Refuse to truncate files with mandatory locks held on them. 26011da177e4SLinus Torvalds */ 26021da177e4SLinus Torvalds error = locks_verify_locked(inode); 2603be6d3e56SKentaro Takeda if (!error) 2604ea0d3ab2STetsuo Handa error = security_path_truncate(path); 26051da177e4SLinus Torvalds if (!error) { 26067715b521SAl Viro error = do_truncate(path->dentry, 0, 2607d139d7ffSMiklos Szeredi ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, 2608e1181ee6SJeff Layton filp); 26091da177e4SLinus Torvalds } 26101da177e4SLinus Torvalds put_write_access(inode); 2611acd0c935SMimi Zohar return error; 26121da177e4SLinus Torvalds } 26131da177e4SLinus Torvalds 2614d57999e1SDave Hansen static inline int open_to_namei_flags(int flag) 2615d57999e1SDave Hansen { 26168a5e929dSAl Viro if ((flag & O_ACCMODE) == 3) 26178a5e929dSAl Viro flag--; 2618d57999e1SDave Hansen return flag; 2619d57999e1SDave Hansen } 2620d57999e1SDave Hansen 2621d18e9008SMiklos Szeredi static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode) 2622d18e9008SMiklos Szeredi { 2623d18e9008SMiklos Szeredi int error = security_path_mknod(dir, dentry, mode, 0); 2624d18e9008SMiklos Szeredi if (error) 2625d18e9008SMiklos Szeredi return error; 2626d18e9008SMiklos Szeredi 2627d18e9008SMiklos Szeredi error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC); 2628d18e9008SMiklos Szeredi if (error) 2629d18e9008SMiklos Szeredi return error; 2630d18e9008SMiklos Szeredi 2631d18e9008SMiklos Szeredi return security_inode_create(dir->dentry->d_inode, dentry, mode); 2632d18e9008SMiklos Szeredi } 2633d18e9008SMiklos Szeredi 26341acf0af9SDavid Howells /* 26351acf0af9SDavid Howells * Attempt to atomically look up, create and open a file from a negative 26361acf0af9SDavid Howells * dentry. 26371acf0af9SDavid Howells * 26381acf0af9SDavid Howells * Returns 0 if successful. The file will have been created and attached to 26391acf0af9SDavid Howells * @file by the filesystem calling finish_open(). 26401acf0af9SDavid Howells * 26411acf0af9SDavid Howells * Returns 1 if the file was looked up only or didn't need creating. The 26421acf0af9SDavid Howells * caller will need to perform the open themselves. @path will have been 26431acf0af9SDavid Howells * updated to point to the new dentry. This may be negative. 26441acf0af9SDavid Howells * 26451acf0af9SDavid Howells * Returns an error code otherwise. 26461acf0af9SDavid Howells */ 26472675a4ebSAl Viro static int atomic_open(struct nameidata *nd, struct dentry *dentry, 264830d90494SAl Viro struct path *path, struct file *file, 2649015c3bbcSMiklos Szeredi const struct open_flags *op, 265064894cf8SAl Viro bool got_write, bool need_lookup, 265147237687SAl Viro int *opened) 2652d18e9008SMiklos Szeredi { 2653d18e9008SMiklos Szeredi struct inode *dir = nd->path.dentry->d_inode; 2654d18e9008SMiklos Szeredi unsigned open_flag = open_to_namei_flags(op->open_flag); 2655d18e9008SMiklos Szeredi umode_t mode; 2656d18e9008SMiklos Szeredi int error; 2657d18e9008SMiklos Szeredi int acc_mode; 2658d18e9008SMiklos Szeredi int create_error = 0; 2659d18e9008SMiklos Szeredi struct dentry *const DENTRY_NOT_SET = (void *) -1UL; 2660d18e9008SMiklos Szeredi 2661d18e9008SMiklos Szeredi BUG_ON(dentry->d_inode); 2662d18e9008SMiklos Szeredi 2663d18e9008SMiklos Szeredi /* Don't create child dentry for a dead directory. */ 2664d18e9008SMiklos Szeredi if (unlikely(IS_DEADDIR(dir))) { 26652675a4ebSAl Viro error = -ENOENT; 2666d18e9008SMiklos Szeredi goto out; 2667d18e9008SMiklos Szeredi } 2668d18e9008SMiklos Szeredi 266962b259d8SMiklos Szeredi mode = op->mode; 2670d18e9008SMiklos Szeredi if ((open_flag & O_CREAT) && !IS_POSIXACL(dir)) 2671d18e9008SMiklos Szeredi mode &= ~current_umask(); 2672d18e9008SMiklos Szeredi 2673f8310c59SAl Viro if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT)) { 2674d18e9008SMiklos Szeredi open_flag &= ~O_TRUNC; 267547237687SAl Viro *opened |= FILE_CREATED; 2676d18e9008SMiklos Szeredi } 2677d18e9008SMiklos Szeredi 2678d18e9008SMiklos Szeredi /* 2679d18e9008SMiklos Szeredi * Checking write permission is tricky, bacuse we don't know if we are 2680d18e9008SMiklos Szeredi * going to actually need it: O_CREAT opens should work as long as the 2681d18e9008SMiklos Szeredi * file exists. But checking existence breaks atomicity. The trick is 2682d18e9008SMiklos Szeredi * to check access and if not granted clear O_CREAT from the flags. 2683d18e9008SMiklos Szeredi * 2684d18e9008SMiklos Szeredi * Another problem is returing the "right" error value (e.g. for an 2685d18e9008SMiklos Szeredi * O_EXCL open we want to return EEXIST not EROFS). 2686d18e9008SMiklos Szeredi */ 268764894cf8SAl Viro if (((open_flag & (O_CREAT | O_TRUNC)) || 268864894cf8SAl Viro (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) { 268964894cf8SAl Viro if (!(open_flag & O_CREAT)) { 2690d18e9008SMiklos Szeredi /* 2691d18e9008SMiklos Szeredi * No O_CREATE -> atomicity not a requirement -> fall 2692d18e9008SMiklos Szeredi * back to lookup + open 2693d18e9008SMiklos Szeredi */ 2694d18e9008SMiklos Szeredi goto no_open; 2695d18e9008SMiklos Szeredi } else if (open_flag & (O_EXCL | O_TRUNC)) { 2696d18e9008SMiklos Szeredi /* Fall back and fail with the right error */ 269764894cf8SAl Viro create_error = -EROFS; 2698d18e9008SMiklos Szeredi goto no_open; 2699d18e9008SMiklos Szeredi } else { 2700d18e9008SMiklos Szeredi /* No side effects, safe to clear O_CREAT */ 270164894cf8SAl Viro create_error = -EROFS; 2702d18e9008SMiklos Szeredi open_flag &= ~O_CREAT; 2703d18e9008SMiklos Szeredi } 2704d18e9008SMiklos Szeredi } 2705d18e9008SMiklos Szeredi 2706d18e9008SMiklos Szeredi if (open_flag & O_CREAT) { 270738227f78SMiklos Szeredi error = may_o_create(&nd->path, dentry, mode); 2708d18e9008SMiklos Szeredi if (error) { 2709d18e9008SMiklos Szeredi create_error = error; 2710d18e9008SMiklos Szeredi if (open_flag & O_EXCL) 2711d18e9008SMiklos Szeredi goto no_open; 2712d18e9008SMiklos Szeredi open_flag &= ~O_CREAT; 2713d18e9008SMiklos Szeredi } 2714d18e9008SMiklos Szeredi } 2715d18e9008SMiklos Szeredi 2716d18e9008SMiklos Szeredi if (nd->flags & LOOKUP_DIRECTORY) 2717d18e9008SMiklos Szeredi open_flag |= O_DIRECTORY; 2718d18e9008SMiklos Szeredi 271930d90494SAl Viro file->f_path.dentry = DENTRY_NOT_SET; 272030d90494SAl Viro file->f_path.mnt = nd->path.mnt; 272130d90494SAl Viro error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode, 272247237687SAl Viro opened); 2723d9585277SAl Viro if (error < 0) { 2724d9585277SAl Viro if (create_error && error == -ENOENT) 2725d9585277SAl Viro error = create_error; 2726d18e9008SMiklos Szeredi goto out; 2727d18e9008SMiklos Szeredi } 2728d18e9008SMiklos Szeredi 2729d18e9008SMiklos Szeredi acc_mode = op->acc_mode; 273047237687SAl Viro if (*opened & FILE_CREATED) { 2731d18e9008SMiklos Szeredi fsnotify_create(dir, dentry); 2732d18e9008SMiklos Szeredi acc_mode = MAY_OPEN; 2733d18e9008SMiklos Szeredi } 2734d18e9008SMiklos Szeredi 2735d9585277SAl Viro if (error) { /* returned 1, that is */ 273630d90494SAl Viro if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { 27372675a4ebSAl Viro error = -EIO; 2738d18e9008SMiklos Szeredi goto out; 2739d18e9008SMiklos Szeredi } 274030d90494SAl Viro if (file->f_path.dentry) { 2741d18e9008SMiklos Szeredi dput(dentry); 274230d90494SAl Viro dentry = file->f_path.dentry; 2743d18e9008SMiklos Szeredi } 274462b2ce96SSage Weil if (create_error && dentry->d_inode == NULL) { 274562b2ce96SSage Weil error = create_error; 274662b2ce96SSage Weil goto out; 274762b2ce96SSage Weil } 2748d18e9008SMiklos Szeredi goto looked_up; 2749d18e9008SMiklos Szeredi } 2750d18e9008SMiklos Szeredi 2751d18e9008SMiklos Szeredi /* 2752d18e9008SMiklos Szeredi * We didn't have the inode before the open, so check open permission 2753d18e9008SMiklos Szeredi * here. 2754d18e9008SMiklos Szeredi */ 27552675a4ebSAl Viro error = may_open(&file->f_path, acc_mode, open_flag); 27562675a4ebSAl Viro if (error) 27572675a4ebSAl Viro fput(file); 2758d18e9008SMiklos Szeredi 2759d18e9008SMiklos Szeredi out: 2760d18e9008SMiklos Szeredi dput(dentry); 27612675a4ebSAl Viro return error; 2762d18e9008SMiklos Szeredi 2763d18e9008SMiklos Szeredi no_open: 2764d18e9008SMiklos Szeredi if (need_lookup) { 276572bd866aSAl Viro dentry = lookup_real(dir, dentry, nd->flags); 2766d18e9008SMiklos Szeredi if (IS_ERR(dentry)) 27672675a4ebSAl Viro return PTR_ERR(dentry); 2768d18e9008SMiklos Szeredi 2769d18e9008SMiklos Szeredi if (create_error) { 2770d18e9008SMiklos Szeredi int open_flag = op->open_flag; 2771d18e9008SMiklos Szeredi 27722675a4ebSAl Viro error = create_error; 2773d18e9008SMiklos Szeredi if ((open_flag & O_EXCL)) { 2774d18e9008SMiklos Szeredi if (!dentry->d_inode) 2775d18e9008SMiklos Szeredi goto out; 2776d18e9008SMiklos Szeredi } else if (!dentry->d_inode) { 2777d18e9008SMiklos Szeredi goto out; 2778d18e9008SMiklos Szeredi } else if ((open_flag & O_TRUNC) && 2779d18e9008SMiklos Szeredi S_ISREG(dentry->d_inode->i_mode)) { 2780d18e9008SMiklos Szeredi goto out; 2781d18e9008SMiklos Szeredi } 2782d18e9008SMiklos Szeredi /* will fail later, go on to get the right error */ 2783d18e9008SMiklos Szeredi } 2784d18e9008SMiklos Szeredi } 2785d18e9008SMiklos Szeredi looked_up: 2786d18e9008SMiklos Szeredi path->dentry = dentry; 2787d18e9008SMiklos Szeredi path->mnt = nd->path.mnt; 27882675a4ebSAl Viro return 1; 2789d18e9008SMiklos Szeredi } 2790d18e9008SMiklos Szeredi 279131e6b01fSNick Piggin /* 27921acf0af9SDavid Howells * Look up and maybe create and open the last component. 2793d58ffd35SMiklos Szeredi * 2794d58ffd35SMiklos Szeredi * Must be called with i_mutex held on parent. 2795d58ffd35SMiklos Szeredi * 27961acf0af9SDavid Howells * Returns 0 if the file was successfully atomically created (if necessary) and 27971acf0af9SDavid Howells * opened. In this case the file will be returned attached to @file. 27981acf0af9SDavid Howells * 27991acf0af9SDavid Howells * Returns 1 if the file was not completely opened at this time, though lookups 28001acf0af9SDavid Howells * and creations will have been performed and the dentry returned in @path will 28011acf0af9SDavid Howells * be positive upon return if O_CREAT was specified. If O_CREAT wasn't 28021acf0af9SDavid Howells * specified then a negative dentry may be returned. 28031acf0af9SDavid Howells * 28041acf0af9SDavid Howells * An error code is returned otherwise. 28051acf0af9SDavid Howells * 28061acf0af9SDavid Howells * FILE_CREATE will be set in @*opened if the dentry was created and will be 28071acf0af9SDavid Howells * cleared otherwise prior to returning. 2808d58ffd35SMiklos Szeredi */ 28092675a4ebSAl Viro static int lookup_open(struct nameidata *nd, struct path *path, 281030d90494SAl Viro struct file *file, 2811d58ffd35SMiklos Szeredi const struct open_flags *op, 281264894cf8SAl Viro bool got_write, int *opened) 2813d58ffd35SMiklos Szeredi { 2814d58ffd35SMiklos Szeredi struct dentry *dir = nd->path.dentry; 281554ef4872SMiklos Szeredi struct inode *dir_inode = dir->d_inode; 2816d58ffd35SMiklos Szeredi struct dentry *dentry; 2817d58ffd35SMiklos Szeredi int error; 281854ef4872SMiklos Szeredi bool need_lookup; 2819d58ffd35SMiklos Szeredi 282047237687SAl Viro *opened &= ~FILE_CREATED; 2821201f956eSAl Viro dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup); 2822d58ffd35SMiklos Szeredi if (IS_ERR(dentry)) 28232675a4ebSAl Viro return PTR_ERR(dentry); 2824d58ffd35SMiklos Szeredi 2825d18e9008SMiklos Szeredi /* Cached positive dentry: will open in f_op->open */ 2826d18e9008SMiklos Szeredi if (!need_lookup && dentry->d_inode) 2827d18e9008SMiklos Szeredi goto out_no_open; 2828d18e9008SMiklos Szeredi 2829d18e9008SMiklos Szeredi if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) { 283064894cf8SAl Viro return atomic_open(nd, dentry, path, file, op, got_write, 283147237687SAl Viro need_lookup, opened); 2832d18e9008SMiklos Szeredi } 2833d18e9008SMiklos Szeredi 283454ef4872SMiklos Szeredi if (need_lookup) { 283554ef4872SMiklos Szeredi BUG_ON(dentry->d_inode); 283654ef4872SMiklos Szeredi 283772bd866aSAl Viro dentry = lookup_real(dir_inode, dentry, nd->flags); 283854ef4872SMiklos Szeredi if (IS_ERR(dentry)) 28392675a4ebSAl Viro return PTR_ERR(dentry); 284054ef4872SMiklos Szeredi } 284154ef4872SMiklos Szeredi 2842d58ffd35SMiklos Szeredi /* Negative dentry, just create the file */ 2843d58ffd35SMiklos Szeredi if (!dentry->d_inode && (op->open_flag & O_CREAT)) { 2844d58ffd35SMiklos Szeredi umode_t mode = op->mode; 2845d58ffd35SMiklos Szeredi if (!IS_POSIXACL(dir->d_inode)) 2846d58ffd35SMiklos Szeredi mode &= ~current_umask(); 2847d58ffd35SMiklos Szeredi /* 2848d58ffd35SMiklos Szeredi * This write is needed to ensure that a 2849d58ffd35SMiklos Szeredi * rw->ro transition does not occur between 2850d58ffd35SMiklos Szeredi * the time when the file is created and when 2851d58ffd35SMiklos Szeredi * a permanent write count is taken through 2852015c3bbcSMiklos Szeredi * the 'struct file' in finish_open(). 2853d58ffd35SMiklos Szeredi */ 285464894cf8SAl Viro if (!got_write) { 285564894cf8SAl Viro error = -EROFS; 2856d58ffd35SMiklos Szeredi goto out_dput; 285764894cf8SAl Viro } 285847237687SAl Viro *opened |= FILE_CREATED; 2859d58ffd35SMiklos Szeredi error = security_path_mknod(&nd->path, dentry, mode, 0); 2860d58ffd35SMiklos Szeredi if (error) 2861d58ffd35SMiklos Szeredi goto out_dput; 2862312b63fbSAl Viro error = vfs_create(dir->d_inode, dentry, mode, 2863312b63fbSAl Viro nd->flags & LOOKUP_EXCL); 2864d58ffd35SMiklos Szeredi if (error) 2865d58ffd35SMiklos Szeredi goto out_dput; 2866d58ffd35SMiklos Szeredi } 2867d18e9008SMiklos Szeredi out_no_open: 2868d58ffd35SMiklos Szeredi path->dentry = dentry; 2869d58ffd35SMiklos Szeredi path->mnt = nd->path.mnt; 28702675a4ebSAl Viro return 1; 2871d58ffd35SMiklos Szeredi 2872d58ffd35SMiklos Szeredi out_dput: 2873d58ffd35SMiklos Szeredi dput(dentry); 28742675a4ebSAl Viro return error; 2875d58ffd35SMiklos Szeredi } 2876d58ffd35SMiklos Szeredi 2877d58ffd35SMiklos Szeredi /* 2878fe2d35ffSAl Viro * Handle the last step of open() 287931e6b01fSNick Piggin */ 28802675a4ebSAl Viro static int do_last(struct nameidata *nd, struct path *path, 288130d90494SAl Viro struct file *file, const struct open_flags *op, 2882669abf4eSJeff Layton int *opened, struct filename *name) 2883fb1cc555SAl Viro { 2884a1e28038SAl Viro struct dentry *dir = nd->path.dentry; 2885ca344a89SAl Viro int open_flag = op->open_flag; 288677d660a8SMiklos Szeredi bool will_truncate = (open_flag & O_TRUNC) != 0; 288764894cf8SAl Viro bool got_write = false; 2888bcda7652SAl Viro int acc_mode = op->acc_mode; 2889a1eb3315SMiklos Szeredi struct inode *inode; 289077d660a8SMiklos Szeredi bool symlink_ok = false; 289116b1c1cdSMiklos Szeredi struct path save_parent = { .dentry = NULL, .mnt = NULL }; 289216b1c1cdSMiklos Szeredi bool retried = false; 289316c2cd71SAl Viro int error; 2894fb1cc555SAl Viro 2895c3e380b0SAl Viro nd->flags &= ~LOOKUP_PARENT; 2896c3e380b0SAl Viro nd->flags |= op->intent; 2897c3e380b0SAl Viro 2898bc77daa7SAl Viro if (nd->last_type != LAST_NORM) { 2899fe2d35ffSAl Viro error = handle_dots(nd, nd->last_type); 2900fe2d35ffSAl Viro if (error) 29012675a4ebSAl Viro return error; 2902e83db167SMiklos Szeredi goto finish_open; 29031f36f774SAl Viro } 2904a2c36b45SAl Viro 2905ca344a89SAl Viro if (!(open_flag & O_CREAT)) { 2906fe2d35ffSAl Viro if (nd->last.name[nd->last.len]) 2907fe2d35ffSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 2908bcda7652SAl Viro if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW)) 290977d660a8SMiklos Szeredi symlink_ok = true; 2910fe2d35ffSAl Viro /* we _can_ be in RCU mode here */ 2911e97cdc87SAl Viro error = lookup_fast(nd, path, &inode); 291271574865SMiklos Szeredi if (likely(!error)) 291371574865SMiklos Szeredi goto finish_lookup; 291471574865SMiklos Szeredi 2915ce57dfc1SAl Viro if (error < 0) 29162675a4ebSAl Viro goto out; 2917a1eb3315SMiklos Szeredi 291837d7fffcSMiklos Szeredi BUG_ON(nd->inode != dir->d_inode); 2919b6183df7SMiklos Szeredi } else { 2920fe2d35ffSAl Viro /* create side of things */ 2921a3fbbde7SAl Viro /* 2922b6183df7SMiklos Szeredi * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED 2923b6183df7SMiklos Szeredi * has been cleared when we got to the last component we are 2924b6183df7SMiklos Szeredi * about to look up 2925a3fbbde7SAl Viro */ 29269f1fafeeSAl Viro error = complete_walk(nd); 29279f1fafeeSAl Viro if (error) 29282675a4ebSAl Viro return error; 2929fe2d35ffSAl Viro 293033e2208aSJeff Layton audit_inode(name, dir, LOOKUP_PARENT); 293116c2cd71SAl Viro error = -EISDIR; 29321f36f774SAl Viro /* trailing slashes? */ 293331e6b01fSNick Piggin if (nd->last.name[nd->last.len]) 29342675a4ebSAl Viro goto out; 2935b6183df7SMiklos Szeredi } 29361f36f774SAl Viro 293716b1c1cdSMiklos Szeredi retry_lookup: 293864894cf8SAl Viro if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { 293964894cf8SAl Viro error = mnt_want_write(nd->path.mnt); 294064894cf8SAl Viro if (!error) 294164894cf8SAl Viro got_write = true; 294264894cf8SAl Viro /* 294364894cf8SAl Viro * do _not_ fail yet - we might not need that or fail with 294464894cf8SAl Viro * a different error; let lookup_open() decide; we'll be 294564894cf8SAl Viro * dropping this one anyway. 294664894cf8SAl Viro */ 294764894cf8SAl Viro } 2948a1e28038SAl Viro mutex_lock(&dir->d_inode->i_mutex); 294964894cf8SAl Viro error = lookup_open(nd, path, file, op, got_write, opened); 2950fb1cc555SAl Viro mutex_unlock(&dir->d_inode->i_mutex); 2951fb1cc555SAl Viro 29522675a4ebSAl Viro if (error <= 0) { 29532675a4ebSAl Viro if (error) 2954d58ffd35SMiklos Szeredi goto out; 29556c0d46c4SAl Viro 295647237687SAl Viro if ((*opened & FILE_CREATED) || 2957496ad9aaSAl Viro !S_ISREG(file_inode(file)->i_mode)) 295877d660a8SMiklos Szeredi will_truncate = false; 2959d18e9008SMiklos Szeredi 2960adb5c247SJeff Layton audit_inode(name, file->f_path.dentry, 0); 2961d18e9008SMiklos Szeredi goto opened; 2962d18e9008SMiklos Szeredi } 2963d18e9008SMiklos Szeredi 296447237687SAl Viro if (*opened & FILE_CREATED) { 29659b44f1b3SAl Viro /* Don't check for write permission, don't truncate */ 2966ca344a89SAl Viro open_flag &= ~O_TRUNC; 296777d660a8SMiklos Szeredi will_truncate = false; 2968bcda7652SAl Viro acc_mode = MAY_OPEN; 2969d58ffd35SMiklos Szeredi path_to_nameidata(path, nd); 2970e83db167SMiklos Szeredi goto finish_open_created; 2971fb1cc555SAl Viro } 2972fb1cc555SAl Viro 2973fb1cc555SAl Viro /* 29743134f37eSJeff Layton * create/update audit record if it already exists. 2975fb1cc555SAl Viro */ 29763134f37eSJeff Layton if (path->dentry->d_inode) 2977adb5c247SJeff Layton audit_inode(name, path->dentry, 0); 2978fb1cc555SAl Viro 2979d18e9008SMiklos Szeredi /* 2980d18e9008SMiklos Szeredi * If atomic_open() acquired write access it is dropped now due to 2981d18e9008SMiklos Szeredi * possible mount and symlink following (this might be optimized away if 2982d18e9008SMiklos Szeredi * necessary...) 2983d18e9008SMiklos Szeredi */ 298464894cf8SAl Viro if (got_write) { 2985d18e9008SMiklos Szeredi mnt_drop_write(nd->path.mnt); 298664894cf8SAl Viro got_write = false; 2987d18e9008SMiklos Szeredi } 2988d18e9008SMiklos Szeredi 2989fb1cc555SAl Viro error = -EEXIST; 2990f8310c59SAl Viro if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT)) 2991fb1cc555SAl Viro goto exit_dput; 2992fb1cc555SAl Viro 29939875cf80SDavid Howells error = follow_managed(path, nd->flags); 29949875cf80SDavid Howells if (error < 0) 2995fb1cc555SAl Viro goto exit_dput; 2996fb1cc555SAl Viro 2997a3fbbde7SAl Viro if (error) 2998a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 2999a3fbbde7SAl Viro 3000decf3400SMiklos Szeredi BUG_ON(nd->flags & LOOKUP_RCU); 3001decf3400SMiklos Szeredi inode = path->dentry->d_inode; 30025f5daac1SMiklos Szeredi finish_lookup: 30035f5daac1SMiklos Szeredi /* we _can_ be in RCU mode here */ 3004fb1cc555SAl Viro error = -ENOENT; 300554c33e7fSMiklos Szeredi if (!inode) { 300654c33e7fSMiklos Szeredi path_to_nameidata(path, nd); 30072675a4ebSAl Viro goto out; 300854c33e7fSMiklos Szeredi } 30099e67f361SAl Viro 3010d45ea867SMiklos Szeredi if (should_follow_link(inode, !symlink_ok)) { 3011d45ea867SMiklos Szeredi if (nd->flags & LOOKUP_RCU) { 3012d45ea867SMiklos Szeredi if (unlikely(unlazy_walk(nd, path->dentry))) { 3013d45ea867SMiklos Szeredi error = -ECHILD; 30142675a4ebSAl Viro goto out; 3015d45ea867SMiklos Szeredi } 3016d45ea867SMiklos Szeredi } 3017d45ea867SMiklos Szeredi BUG_ON(inode != path->dentry->d_inode); 30182675a4ebSAl Viro return 1; 3019d45ea867SMiklos Szeredi } 3020fb1cc555SAl Viro 302116b1c1cdSMiklos Szeredi if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) { 3022fb1cc555SAl Viro path_to_nameidata(path, nd); 302316b1c1cdSMiklos Szeredi } else { 302416b1c1cdSMiklos Szeredi save_parent.dentry = nd->path.dentry; 302516b1c1cdSMiklos Szeredi save_parent.mnt = mntget(path->mnt); 302616b1c1cdSMiklos Szeredi nd->path.dentry = path->dentry; 302716b1c1cdSMiklos Szeredi 302816b1c1cdSMiklos Szeredi } 3029decf3400SMiklos Szeredi nd->inode = inode; 3030a3fbbde7SAl Viro /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ 3031bc77daa7SAl Viro finish_open: 3032a3fbbde7SAl Viro error = complete_walk(nd); 303316b1c1cdSMiklos Szeredi if (error) { 303416b1c1cdSMiklos Szeredi path_put(&save_parent); 30352675a4ebSAl Viro return error; 303616b1c1cdSMiklos Szeredi } 3037bc77daa7SAl Viro audit_inode(name, nd->path.dentry, 0); 3038fb1cc555SAl Viro error = -EISDIR; 3039050ac841SMiklos Szeredi if ((open_flag & O_CREAT) && S_ISDIR(nd->inode->i_mode)) 30402675a4ebSAl Viro goto out; 3041af2f5542SMiklos Szeredi error = -ENOTDIR; 304205252901SAl Viro if ((nd->flags & LOOKUP_DIRECTORY) && !can_lookup(nd->inode)) 30432675a4ebSAl Viro goto out; 30446c0d46c4SAl Viro if (!S_ISREG(nd->inode->i_mode)) 304577d660a8SMiklos Szeredi will_truncate = false; 30466c0d46c4SAl Viro 30470f9d1a10SAl Viro if (will_truncate) { 30480f9d1a10SAl Viro error = mnt_want_write(nd->path.mnt); 30490f9d1a10SAl Viro if (error) 30502675a4ebSAl Viro goto out; 305164894cf8SAl Viro got_write = true; 30520f9d1a10SAl Viro } 3053e83db167SMiklos Szeredi finish_open_created: 3054bcda7652SAl Viro error = may_open(&nd->path, acc_mode, open_flag); 3055ca344a89SAl Viro if (error) 30562675a4ebSAl Viro goto out; 305730d90494SAl Viro file->f_path.mnt = nd->path.mnt; 305830d90494SAl Viro error = finish_open(file, nd->path.dentry, NULL, opened); 305930d90494SAl Viro if (error) { 306030d90494SAl Viro if (error == -EOPENSTALE) 3061f60dc3dbSMiklos Szeredi goto stale_open; 3062015c3bbcSMiklos Szeredi goto out; 3063f60dc3dbSMiklos Szeredi } 3064a8277b9bSMiklos Szeredi opened: 30652675a4ebSAl Viro error = open_check_o_direct(file); 3066015c3bbcSMiklos Szeredi if (error) 3067015c3bbcSMiklos Szeredi goto exit_fput; 30682675a4ebSAl Viro error = ima_file_check(file, op->acc_mode); 3069aa4caadbSMiklos Szeredi if (error) 3070aa4caadbSMiklos Szeredi goto exit_fput; 3071aa4caadbSMiklos Szeredi 30720f9d1a10SAl Viro if (will_truncate) { 30732675a4ebSAl Viro error = handle_truncate(file); 3074aa4caadbSMiklos Szeredi if (error) 3075aa4caadbSMiklos Szeredi goto exit_fput; 30760f9d1a10SAl Viro } 3077ca344a89SAl Viro out: 307864894cf8SAl Viro if (got_write) 30790f9d1a10SAl Viro mnt_drop_write(nd->path.mnt); 308016b1c1cdSMiklos Szeredi path_put(&save_parent); 3081e276ae67SMiklos Szeredi terminate_walk(nd); 30822675a4ebSAl Viro return error; 3083fb1cc555SAl Viro 3084fb1cc555SAl Viro exit_dput: 3085fb1cc555SAl Viro path_put_conditional(path, nd); 3086ca344a89SAl Viro goto out; 3087015c3bbcSMiklos Szeredi exit_fput: 30882675a4ebSAl Viro fput(file); 30892675a4ebSAl Viro goto out; 3090015c3bbcSMiklos Szeredi 3091f60dc3dbSMiklos Szeredi stale_open: 3092f60dc3dbSMiklos Szeredi /* If no saved parent or already retried then can't retry */ 3093f60dc3dbSMiklos Szeredi if (!save_parent.dentry || retried) 3094f60dc3dbSMiklos Szeredi goto out; 3095f60dc3dbSMiklos Szeredi 3096f60dc3dbSMiklos Szeredi BUG_ON(save_parent.dentry != dir); 3097f60dc3dbSMiklos Szeredi path_put(&nd->path); 3098f60dc3dbSMiklos Szeredi nd->path = save_parent; 3099f60dc3dbSMiklos Szeredi nd->inode = dir->d_inode; 3100f60dc3dbSMiklos Szeredi save_parent.mnt = NULL; 3101f60dc3dbSMiklos Szeredi save_parent.dentry = NULL; 310264894cf8SAl Viro if (got_write) { 3103f60dc3dbSMiklos Szeredi mnt_drop_write(nd->path.mnt); 310464894cf8SAl Viro got_write = false; 3105f60dc3dbSMiklos Szeredi } 3106f60dc3dbSMiklos Szeredi retried = true; 3107f60dc3dbSMiklos Szeredi goto retry_lookup; 3108fb1cc555SAl Viro } 3109fb1cc555SAl Viro 311060545d0dSAl Viro static int do_tmpfile(int dfd, struct filename *pathname, 311160545d0dSAl Viro struct nameidata *nd, int flags, 311260545d0dSAl Viro const struct open_flags *op, 311360545d0dSAl Viro struct file *file, int *opened) 311460545d0dSAl Viro { 311560545d0dSAl Viro static const struct qstr name = QSTR_INIT("/", 1); 311660545d0dSAl Viro struct dentry *dentry, *child; 311760545d0dSAl Viro struct inode *dir; 311860545d0dSAl Viro int error = path_lookupat(dfd, pathname->name, 311960545d0dSAl Viro flags | LOOKUP_DIRECTORY, nd); 312060545d0dSAl Viro if (unlikely(error)) 312160545d0dSAl Viro return error; 312260545d0dSAl Viro error = mnt_want_write(nd->path.mnt); 312360545d0dSAl Viro if (unlikely(error)) 312460545d0dSAl Viro goto out; 312560545d0dSAl Viro /* we want directory to be writable */ 312660545d0dSAl Viro error = inode_permission(nd->inode, MAY_WRITE | MAY_EXEC); 312760545d0dSAl Viro if (error) 312860545d0dSAl Viro goto out2; 312960545d0dSAl Viro dentry = nd->path.dentry; 313060545d0dSAl Viro dir = dentry->d_inode; 313160545d0dSAl Viro if (!dir->i_op->tmpfile) { 313260545d0dSAl Viro error = -EOPNOTSUPP; 313360545d0dSAl Viro goto out2; 313460545d0dSAl Viro } 313560545d0dSAl Viro child = d_alloc(dentry, &name); 313660545d0dSAl Viro if (unlikely(!child)) { 313760545d0dSAl Viro error = -ENOMEM; 313860545d0dSAl Viro goto out2; 313960545d0dSAl Viro } 314060545d0dSAl Viro nd->flags &= ~LOOKUP_DIRECTORY; 314160545d0dSAl Viro nd->flags |= op->intent; 314260545d0dSAl Viro dput(nd->path.dentry); 314360545d0dSAl Viro nd->path.dentry = child; 314460545d0dSAl Viro error = dir->i_op->tmpfile(dir, nd->path.dentry, op->mode); 314560545d0dSAl Viro if (error) 314660545d0dSAl Viro goto out2; 314760545d0dSAl Viro audit_inode(pathname, nd->path.dentry, 0); 314860545d0dSAl Viro error = may_open(&nd->path, op->acc_mode, op->open_flag); 314960545d0dSAl Viro if (error) 315060545d0dSAl Viro goto out2; 315160545d0dSAl Viro file->f_path.mnt = nd->path.mnt; 315260545d0dSAl Viro error = finish_open(file, nd->path.dentry, NULL, opened); 315360545d0dSAl Viro if (error) 315460545d0dSAl Viro goto out2; 315560545d0dSAl Viro error = open_check_o_direct(file); 3156f4e0c30cSAl Viro if (error) { 315760545d0dSAl Viro fput(file); 3158f4e0c30cSAl Viro } else if (!(op->open_flag & O_EXCL)) { 3159f4e0c30cSAl Viro struct inode *inode = file_inode(file); 3160f4e0c30cSAl Viro spin_lock(&inode->i_lock); 3161f4e0c30cSAl Viro inode->i_state |= I_LINKABLE; 3162f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 3163f4e0c30cSAl Viro } 316460545d0dSAl Viro out2: 316560545d0dSAl Viro mnt_drop_write(nd->path.mnt); 316660545d0dSAl Viro out: 316760545d0dSAl Viro path_put(&nd->path); 316860545d0dSAl Viro return error; 316960545d0dSAl Viro } 317060545d0dSAl Viro 3171669abf4eSJeff Layton static struct file *path_openat(int dfd, struct filename *pathname, 317273d049a4SAl Viro struct nameidata *nd, const struct open_flags *op, int flags) 31731da177e4SLinus Torvalds { 3174fe2d35ffSAl Viro struct file *base = NULL; 317530d90494SAl Viro struct file *file; 31769850c056SAl Viro struct path path; 317747237687SAl Viro int opened = 0; 317813aab428SAl Viro int error; 317931e6b01fSNick Piggin 318030d90494SAl Viro file = get_empty_filp(); 31811afc99beSAl Viro if (IS_ERR(file)) 31821afc99beSAl Viro return file; 318331e6b01fSNick Piggin 318430d90494SAl Viro file->f_flags = op->open_flag; 318531e6b01fSNick Piggin 3186bb458c64SAl Viro if (unlikely(file->f_flags & __O_TMPFILE)) { 318760545d0dSAl Viro error = do_tmpfile(dfd, pathname, nd, flags, op, file, &opened); 318860545d0dSAl Viro goto out; 318960545d0dSAl Viro } 319060545d0dSAl Viro 3191669abf4eSJeff Layton error = path_init(dfd, pathname->name, flags | LOOKUP_PARENT, nd, &base); 319231e6b01fSNick Piggin if (unlikely(error)) 31932675a4ebSAl Viro goto out; 319431e6b01fSNick Piggin 3195fe2d35ffSAl Viro current->total_link_count = 0; 3196669abf4eSJeff Layton error = link_path_walk(pathname->name, nd); 319731e6b01fSNick Piggin if (unlikely(error)) 31982675a4ebSAl Viro goto out; 31991da177e4SLinus Torvalds 32002675a4ebSAl Viro error = do_last(nd, &path, file, op, &opened, pathname); 32012675a4ebSAl Viro while (unlikely(error > 0)) { /* trailing symlink */ 32027b9337aaSNick Piggin struct path link = path; 3203def4af30SAl Viro void *cookie; 3204574197e0SAl Viro if (!(nd->flags & LOOKUP_FOLLOW)) { 320573d049a4SAl Viro path_put_conditional(&path, nd); 320673d049a4SAl Viro path_put(&nd->path); 32072675a4ebSAl Viro error = -ELOOP; 320840b39136SAl Viro break; 320940b39136SAl Viro } 3210800179c9SKees Cook error = may_follow_link(&link, nd); 3211800179c9SKees Cook if (unlikely(error)) 3212800179c9SKees Cook break; 321373d049a4SAl Viro nd->flags |= LOOKUP_PARENT; 321473d049a4SAl Viro nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); 3215574197e0SAl Viro error = follow_link(&link, nd, &cookie); 3216c3e380b0SAl Viro if (unlikely(error)) 32172675a4ebSAl Viro break; 32182675a4ebSAl Viro error = do_last(nd, &path, file, op, &opened, pathname); 3219574197e0SAl Viro put_link(nd, &link, cookie); 3220806b681cSAl Viro } 322110fa8e62SAl Viro out: 322273d049a4SAl Viro if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) 322373d049a4SAl Viro path_put(&nd->root); 3224fe2d35ffSAl Viro if (base) 3225fe2d35ffSAl Viro fput(base); 32262675a4ebSAl Viro if (!(opened & FILE_OPENED)) { 32272675a4ebSAl Viro BUG_ON(!error); 322830d90494SAl Viro put_filp(file); 3229015c3bbcSMiklos Szeredi } 32302675a4ebSAl Viro if (unlikely(error)) { 32312675a4ebSAl Viro if (error == -EOPENSTALE) { 32322675a4ebSAl Viro if (flags & LOOKUP_RCU) 32332675a4ebSAl Viro error = -ECHILD; 32342675a4ebSAl Viro else 32352675a4ebSAl Viro error = -ESTALE; 32362675a4ebSAl Viro } 32372675a4ebSAl Viro file = ERR_PTR(error); 32382675a4ebSAl Viro } 32392675a4ebSAl Viro return file; 3240de459215SKirill Korotaev } 32411da177e4SLinus Torvalds 3242669abf4eSJeff Layton struct file *do_filp_open(int dfd, struct filename *pathname, 3243f9652e10SAl Viro const struct open_flags *op) 324413aab428SAl Viro { 324573d049a4SAl Viro struct nameidata nd; 3246f9652e10SAl Viro int flags = op->lookup_flags; 324713aab428SAl Viro struct file *filp; 324813aab428SAl Viro 324973d049a4SAl Viro filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU); 325013aab428SAl Viro if (unlikely(filp == ERR_PTR(-ECHILD))) 325173d049a4SAl Viro filp = path_openat(dfd, pathname, &nd, op, flags); 325213aab428SAl Viro if (unlikely(filp == ERR_PTR(-ESTALE))) 325373d049a4SAl Viro filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL); 325413aab428SAl Viro return filp; 325513aab428SAl Viro } 325613aab428SAl Viro 325773d049a4SAl Viro struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, 3258f9652e10SAl Viro const char *name, const struct open_flags *op) 325973d049a4SAl Viro { 326073d049a4SAl Viro struct nameidata nd; 326173d049a4SAl Viro struct file *file; 3262669abf4eSJeff Layton struct filename filename = { .name = name }; 3263f9652e10SAl Viro int flags = op->lookup_flags | LOOKUP_ROOT; 326473d049a4SAl Viro 326573d049a4SAl Viro nd.root.mnt = mnt; 326673d049a4SAl Viro nd.root.dentry = dentry; 326773d049a4SAl Viro 3268bcda7652SAl Viro if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN) 326973d049a4SAl Viro return ERR_PTR(-ELOOP); 327073d049a4SAl Viro 3271669abf4eSJeff Layton file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_RCU); 327273d049a4SAl Viro if (unlikely(file == ERR_PTR(-ECHILD))) 3273669abf4eSJeff Layton file = path_openat(-1, &filename, &nd, op, flags); 327473d049a4SAl Viro if (unlikely(file == ERR_PTR(-ESTALE))) 3275669abf4eSJeff Layton file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_REVAL); 327673d049a4SAl Viro return file; 327773d049a4SAl Viro } 327873d049a4SAl Viro 32791ac12b4bSJeff Layton struct dentry *kern_path_create(int dfd, const char *pathname, 32801ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 32811da177e4SLinus Torvalds { 3282c663e5d8SChristoph Hellwig struct dentry *dentry = ERR_PTR(-EEXIST); 3283ed75e95dSAl Viro struct nameidata nd; 3284c30dabfeSJan Kara int err2; 32851ac12b4bSJeff Layton int error; 32861ac12b4bSJeff Layton bool is_dir = (lookup_flags & LOOKUP_DIRECTORY); 32871ac12b4bSJeff Layton 32881ac12b4bSJeff Layton /* 32891ac12b4bSJeff Layton * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any 32901ac12b4bSJeff Layton * other flags passed in are ignored! 32911ac12b4bSJeff Layton */ 32921ac12b4bSJeff Layton lookup_flags &= LOOKUP_REVAL; 32931ac12b4bSJeff Layton 32941ac12b4bSJeff Layton error = do_path_lookup(dfd, pathname, LOOKUP_PARENT|lookup_flags, &nd); 3295ed75e95dSAl Viro if (error) 3296ed75e95dSAl Viro return ERR_PTR(error); 32971da177e4SLinus Torvalds 3298c663e5d8SChristoph Hellwig /* 3299c663e5d8SChristoph Hellwig * Yucky last component or no last component at all? 3300c663e5d8SChristoph Hellwig * (foo/., foo/.., /////) 3301c663e5d8SChristoph Hellwig */ 3302ed75e95dSAl Viro if (nd.last_type != LAST_NORM) 3303ed75e95dSAl Viro goto out; 3304ed75e95dSAl Viro nd.flags &= ~LOOKUP_PARENT; 3305ed75e95dSAl Viro nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL; 3306c663e5d8SChristoph Hellwig 3307c30dabfeSJan Kara /* don't fail immediately if it's r/o, at least try to report other errors */ 3308c30dabfeSJan Kara err2 = mnt_want_write(nd.path.mnt); 3309c663e5d8SChristoph Hellwig /* 3310c663e5d8SChristoph Hellwig * Do the final lookup. 3311c663e5d8SChristoph Hellwig */ 3312ed75e95dSAl Viro mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 3313ed75e95dSAl Viro dentry = lookup_hash(&nd); 33141da177e4SLinus Torvalds if (IS_ERR(dentry)) 3315a8104a9fSAl Viro goto unlock; 3316c663e5d8SChristoph Hellwig 3317a8104a9fSAl Viro error = -EEXIST; 3318e9baf6e5SAl Viro if (dentry->d_inode) 3319a8104a9fSAl Viro goto fail; 3320c663e5d8SChristoph Hellwig /* 3321c663e5d8SChristoph Hellwig * Special case - lookup gave negative, but... we had foo/bar/ 3322c663e5d8SChristoph Hellwig * From the vfs_mknod() POV we just have a negative dentry - 3323c663e5d8SChristoph Hellwig * all is fine. Let's be bastards - you had / on the end, you've 3324c663e5d8SChristoph Hellwig * been asking for (non-existent) directory. -ENOENT for you. 3325c663e5d8SChristoph Hellwig */ 3326ed75e95dSAl Viro if (unlikely(!is_dir && nd.last.name[nd.last.len])) { 3327a8104a9fSAl Viro error = -ENOENT; 3328ed75e95dSAl Viro goto fail; 3329e9baf6e5SAl Viro } 3330c30dabfeSJan Kara if (unlikely(err2)) { 3331c30dabfeSJan Kara error = err2; 3332a8104a9fSAl Viro goto fail; 3333c30dabfeSJan Kara } 3334ed75e95dSAl Viro *path = nd.path; 3335e9baf6e5SAl Viro return dentry; 33361da177e4SLinus Torvalds fail: 3337a8104a9fSAl Viro dput(dentry); 3338a8104a9fSAl Viro dentry = ERR_PTR(error); 3339a8104a9fSAl Viro unlock: 3340dae6ad8fSAl Viro mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 3341c30dabfeSJan Kara if (!err2) 3342c30dabfeSJan Kara mnt_drop_write(nd.path.mnt); 3343ed75e95dSAl Viro out: 3344dae6ad8fSAl Viro path_put(&nd.path); 3345ed75e95dSAl Viro return dentry; 3346dae6ad8fSAl Viro } 3347dae6ad8fSAl Viro EXPORT_SYMBOL(kern_path_create); 3348dae6ad8fSAl Viro 3349921a1650SAl Viro void done_path_create(struct path *path, struct dentry *dentry) 3350921a1650SAl Viro { 3351921a1650SAl Viro dput(dentry); 3352921a1650SAl Viro mutex_unlock(&path->dentry->d_inode->i_mutex); 3353a8104a9fSAl Viro mnt_drop_write(path->mnt); 3354921a1650SAl Viro path_put(path); 3355921a1650SAl Viro } 3356921a1650SAl Viro EXPORT_SYMBOL(done_path_create); 3357921a1650SAl Viro 33581ac12b4bSJeff Layton struct dentry *user_path_create(int dfd, const char __user *pathname, 33591ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 3360dae6ad8fSAl Viro { 336191a27b2aSJeff Layton struct filename *tmp = getname(pathname); 3362dae6ad8fSAl Viro struct dentry *res; 3363dae6ad8fSAl Viro if (IS_ERR(tmp)) 3364dae6ad8fSAl Viro return ERR_CAST(tmp); 33651ac12b4bSJeff Layton res = kern_path_create(dfd, tmp->name, path, lookup_flags); 3366dae6ad8fSAl Viro putname(tmp); 3367dae6ad8fSAl Viro return res; 3368dae6ad8fSAl Viro } 3369dae6ad8fSAl Viro EXPORT_SYMBOL(user_path_create); 3370dae6ad8fSAl Viro 33711a67aafbSAl Viro int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) 33721da177e4SLinus Torvalds { 3373a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 33741da177e4SLinus Torvalds 33751da177e4SLinus Torvalds if (error) 33761da177e4SLinus Torvalds return error; 33771da177e4SLinus Torvalds 3378975d6b39SEric W. Biederman if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) 33791da177e4SLinus Torvalds return -EPERM; 33801da177e4SLinus Torvalds 3381acfa4380SAl Viro if (!dir->i_op->mknod) 33821da177e4SLinus Torvalds return -EPERM; 33831da177e4SLinus Torvalds 338408ce5f16SSerge E. Hallyn error = devcgroup_inode_mknod(mode, dev); 338508ce5f16SSerge E. Hallyn if (error) 338608ce5f16SSerge E. Hallyn return error; 338708ce5f16SSerge E. Hallyn 33881da177e4SLinus Torvalds error = security_inode_mknod(dir, dentry, mode, dev); 33891da177e4SLinus Torvalds if (error) 33901da177e4SLinus Torvalds return error; 33911da177e4SLinus Torvalds 33921da177e4SLinus Torvalds error = dir->i_op->mknod(dir, dentry, mode, dev); 3393a74574aaSStephen Smalley if (!error) 3394f38aa942SAmy Griffis fsnotify_create(dir, dentry); 33951da177e4SLinus Torvalds return error; 33961da177e4SLinus Torvalds } 33971da177e4SLinus Torvalds 3398f69aac00SAl Viro static int may_mknod(umode_t mode) 3399463c3197SDave Hansen { 3400463c3197SDave Hansen switch (mode & S_IFMT) { 3401463c3197SDave Hansen case S_IFREG: 3402463c3197SDave Hansen case S_IFCHR: 3403463c3197SDave Hansen case S_IFBLK: 3404463c3197SDave Hansen case S_IFIFO: 3405463c3197SDave Hansen case S_IFSOCK: 3406463c3197SDave Hansen case 0: /* zero mode translates to S_IFREG */ 3407463c3197SDave Hansen return 0; 3408463c3197SDave Hansen case S_IFDIR: 3409463c3197SDave Hansen return -EPERM; 3410463c3197SDave Hansen default: 3411463c3197SDave Hansen return -EINVAL; 3412463c3197SDave Hansen } 3413463c3197SDave Hansen } 3414463c3197SDave Hansen 34158208a22bSAl Viro SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, 34162e4d0924SHeiko Carstens unsigned, dev) 34171da177e4SLinus Torvalds { 34181da177e4SLinus Torvalds struct dentry *dentry; 3419dae6ad8fSAl Viro struct path path; 3420dae6ad8fSAl Viro int error; 3421972567f1SJeff Layton unsigned int lookup_flags = 0; 34221da177e4SLinus Torvalds 34238e4bfca1SAl Viro error = may_mknod(mode); 34248e4bfca1SAl Viro if (error) 34258e4bfca1SAl Viro return error; 3426972567f1SJeff Layton retry: 3427972567f1SJeff Layton dentry = user_path_create(dfd, filename, &path, lookup_flags); 3428dae6ad8fSAl Viro if (IS_ERR(dentry)) 3429dae6ad8fSAl Viro return PTR_ERR(dentry); 34302ad94ae6SAl Viro 3431dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3432ce3b0f8dSAl Viro mode &= ~current_umask(); 3433dae6ad8fSAl Viro error = security_path_mknod(&path, dentry, mode, dev); 3434be6d3e56SKentaro Takeda if (error) 3435a8104a9fSAl Viro goto out; 34361da177e4SLinus Torvalds switch (mode & S_IFMT) { 34371da177e4SLinus Torvalds case 0: case S_IFREG: 3438312b63fbSAl Viro error = vfs_create(path.dentry->d_inode,dentry,mode,true); 34391da177e4SLinus Torvalds break; 34401da177e4SLinus Torvalds case S_IFCHR: case S_IFBLK: 3441dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode, 34421da177e4SLinus Torvalds new_decode_dev(dev)); 34431da177e4SLinus Torvalds break; 34441da177e4SLinus Torvalds case S_IFIFO: case S_IFSOCK: 3445dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); 34461da177e4SLinus Torvalds break; 34471da177e4SLinus Torvalds } 3448a8104a9fSAl Viro out: 3449921a1650SAl Viro done_path_create(&path, dentry); 3450972567f1SJeff Layton if (retry_estale(error, lookup_flags)) { 3451972567f1SJeff Layton lookup_flags |= LOOKUP_REVAL; 3452972567f1SJeff Layton goto retry; 3453972567f1SJeff Layton } 34541da177e4SLinus Torvalds return error; 34551da177e4SLinus Torvalds } 34561da177e4SLinus Torvalds 34578208a22bSAl Viro SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) 34585590ff0dSUlrich Drepper { 34595590ff0dSUlrich Drepper return sys_mknodat(AT_FDCWD, filename, mode, dev); 34605590ff0dSUlrich Drepper } 34615590ff0dSUlrich Drepper 346218bb1db3SAl Viro int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 34631da177e4SLinus Torvalds { 3464a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 34658de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 34661da177e4SLinus Torvalds 34671da177e4SLinus Torvalds if (error) 34681da177e4SLinus Torvalds return error; 34691da177e4SLinus Torvalds 3470acfa4380SAl Viro if (!dir->i_op->mkdir) 34711da177e4SLinus Torvalds return -EPERM; 34721da177e4SLinus Torvalds 34731da177e4SLinus Torvalds mode &= (S_IRWXUGO|S_ISVTX); 34741da177e4SLinus Torvalds error = security_inode_mkdir(dir, dentry, mode); 34751da177e4SLinus Torvalds if (error) 34761da177e4SLinus Torvalds return error; 34771da177e4SLinus Torvalds 34788de52778SAl Viro if (max_links && dir->i_nlink >= max_links) 34798de52778SAl Viro return -EMLINK; 34808de52778SAl Viro 34811da177e4SLinus Torvalds error = dir->i_op->mkdir(dir, dentry, mode); 3482a74574aaSStephen Smalley if (!error) 3483f38aa942SAmy Griffis fsnotify_mkdir(dir, dentry); 34841da177e4SLinus Torvalds return error; 34851da177e4SLinus Torvalds } 34861da177e4SLinus Torvalds 3487a218d0fdSAl Viro SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) 34881da177e4SLinus Torvalds { 34896902d925SDave Hansen struct dentry *dentry; 3490dae6ad8fSAl Viro struct path path; 3491dae6ad8fSAl Viro int error; 3492b76d8b82SJeff Layton unsigned int lookup_flags = LOOKUP_DIRECTORY; 34931da177e4SLinus Torvalds 3494b76d8b82SJeff Layton retry: 3495b76d8b82SJeff Layton dentry = user_path_create(dfd, pathname, &path, lookup_flags); 34966902d925SDave Hansen if (IS_ERR(dentry)) 3497dae6ad8fSAl Viro return PTR_ERR(dentry); 34986902d925SDave Hansen 3499dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3500ce3b0f8dSAl Viro mode &= ~current_umask(); 3501dae6ad8fSAl Viro error = security_path_mkdir(&path, dentry, mode); 3502a8104a9fSAl Viro if (!error) 3503dae6ad8fSAl Viro error = vfs_mkdir(path.dentry->d_inode, dentry, mode); 3504921a1650SAl Viro done_path_create(&path, dentry); 3505b76d8b82SJeff Layton if (retry_estale(error, lookup_flags)) { 3506b76d8b82SJeff Layton lookup_flags |= LOOKUP_REVAL; 3507b76d8b82SJeff Layton goto retry; 3508b76d8b82SJeff Layton } 35091da177e4SLinus Torvalds return error; 35101da177e4SLinus Torvalds } 35111da177e4SLinus Torvalds 3512a218d0fdSAl Viro SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) 35135590ff0dSUlrich Drepper { 35145590ff0dSUlrich Drepper return sys_mkdirat(AT_FDCWD, pathname, mode); 35155590ff0dSUlrich Drepper } 35165590ff0dSUlrich Drepper 35171da177e4SLinus Torvalds /* 3518a71905f0SSage Weil * The dentry_unhash() helper will try to drop the dentry early: we 3519c0d02594SJ. Bruce Fields * should have a usage count of 1 if we're the only user of this 3520a71905f0SSage Weil * dentry, and if that is true (possibly after pruning the dcache), 3521a71905f0SSage Weil * then we drop the dentry now. 35221da177e4SLinus Torvalds * 35231da177e4SLinus Torvalds * A low-level filesystem can, if it choses, legally 35241da177e4SLinus Torvalds * do a 35251da177e4SLinus Torvalds * 35261da177e4SLinus Torvalds * if (!d_unhashed(dentry)) 35271da177e4SLinus Torvalds * return -EBUSY; 35281da177e4SLinus Torvalds * 35291da177e4SLinus Torvalds * if it cannot handle the case of removing a directory 35301da177e4SLinus Torvalds * that is still in use by something else.. 35311da177e4SLinus Torvalds */ 35321da177e4SLinus Torvalds void dentry_unhash(struct dentry *dentry) 35331da177e4SLinus Torvalds { 35341da177e4SLinus Torvalds shrink_dcache_parent(dentry); 35351da177e4SLinus Torvalds spin_lock(&dentry->d_lock); 353698474236SWaiman Long if (dentry->d_lockref.count == 1) 35371da177e4SLinus Torvalds __d_drop(dentry); 35381da177e4SLinus Torvalds spin_unlock(&dentry->d_lock); 35391da177e4SLinus Torvalds } 35401da177e4SLinus Torvalds 35411da177e4SLinus Torvalds int vfs_rmdir(struct inode *dir, struct dentry *dentry) 35421da177e4SLinus Torvalds { 35431da177e4SLinus Torvalds int error = may_delete(dir, dentry, 1); 35441da177e4SLinus Torvalds 35451da177e4SLinus Torvalds if (error) 35461da177e4SLinus Torvalds return error; 35471da177e4SLinus Torvalds 3548acfa4380SAl Viro if (!dir->i_op->rmdir) 35491da177e4SLinus Torvalds return -EPERM; 35501da177e4SLinus Torvalds 35511d2ef590SAl Viro dget(dentry); 35521b1dcc1bSJes Sorensen mutex_lock(&dentry->d_inode->i_mutex); 3553912dbc15SSage Weil 35541da177e4SLinus Torvalds error = -EBUSY; 3555912dbc15SSage Weil if (d_mountpoint(dentry)) 3556912dbc15SSage Weil goto out; 3557912dbc15SSage Weil 35581da177e4SLinus Torvalds error = security_inode_rmdir(dir, dentry); 3559912dbc15SSage Weil if (error) 3560912dbc15SSage Weil goto out; 3561912dbc15SSage Weil 35623cebde24SSage Weil shrink_dcache_parent(dentry); 35631da177e4SLinus Torvalds error = dir->i_op->rmdir(dir, dentry); 3564912dbc15SSage Weil if (error) 3565912dbc15SSage Weil goto out; 3566912dbc15SSage Weil 35671da177e4SLinus Torvalds dentry->d_inode->i_flags |= S_DEAD; 3568d83c49f3SAl Viro dont_mount(dentry); 35691da177e4SLinus Torvalds 3570912dbc15SSage Weil out: 3571912dbc15SSage Weil mutex_unlock(&dentry->d_inode->i_mutex); 35721d2ef590SAl Viro dput(dentry); 3573912dbc15SSage Weil if (!error) 3574912dbc15SSage Weil d_delete(dentry); 35751da177e4SLinus Torvalds return error; 35761da177e4SLinus Torvalds } 35771da177e4SLinus Torvalds 35785590ff0dSUlrich Drepper static long do_rmdir(int dfd, const char __user *pathname) 35791da177e4SLinus Torvalds { 35801da177e4SLinus Torvalds int error = 0; 358191a27b2aSJeff Layton struct filename *name; 35821da177e4SLinus Torvalds struct dentry *dentry; 35831da177e4SLinus Torvalds struct nameidata nd; 3584c6ee9206SJeff Layton unsigned int lookup_flags = 0; 3585c6ee9206SJeff Layton retry: 3586c6ee9206SJeff Layton name = user_path_parent(dfd, pathname, &nd, lookup_flags); 358791a27b2aSJeff Layton if (IS_ERR(name)) 358891a27b2aSJeff Layton return PTR_ERR(name); 35891da177e4SLinus Torvalds 35901da177e4SLinus Torvalds switch(nd.last_type) { 35911da177e4SLinus Torvalds case LAST_DOTDOT: 35921da177e4SLinus Torvalds error = -ENOTEMPTY; 35931da177e4SLinus Torvalds goto exit1; 35941da177e4SLinus Torvalds case LAST_DOT: 35951da177e4SLinus Torvalds error = -EINVAL; 35961da177e4SLinus Torvalds goto exit1; 35971da177e4SLinus Torvalds case LAST_ROOT: 35981da177e4SLinus Torvalds error = -EBUSY; 35991da177e4SLinus Torvalds goto exit1; 36001da177e4SLinus Torvalds } 36010612d9fbSOGAWA Hirofumi 36020612d9fbSOGAWA Hirofumi nd.flags &= ~LOOKUP_PARENT; 3603c30dabfeSJan Kara error = mnt_want_write(nd.path.mnt); 3604c30dabfeSJan Kara if (error) 3605c30dabfeSJan Kara goto exit1; 36060612d9fbSOGAWA Hirofumi 36074ac91378SJan Blunck mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 360849705b77SChristoph Hellwig dentry = lookup_hash(&nd); 36091da177e4SLinus Torvalds error = PTR_ERR(dentry); 36106902d925SDave Hansen if (IS_ERR(dentry)) 36116902d925SDave Hansen goto exit2; 3612e6bc45d6STheodore Ts'o if (!dentry->d_inode) { 3613e6bc45d6STheodore Ts'o error = -ENOENT; 3614e6bc45d6STheodore Ts'o goto exit3; 3615e6bc45d6STheodore Ts'o } 3616be6d3e56SKentaro Takeda error = security_path_rmdir(&nd.path, dentry); 3617be6d3e56SKentaro Takeda if (error) 3618c30dabfeSJan Kara goto exit3; 36194ac91378SJan Blunck error = vfs_rmdir(nd.path.dentry->d_inode, dentry); 36200622753bSDave Hansen exit3: 36211da177e4SLinus Torvalds dput(dentry); 36226902d925SDave Hansen exit2: 36234ac91378SJan Blunck mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 3624c30dabfeSJan Kara mnt_drop_write(nd.path.mnt); 36251da177e4SLinus Torvalds exit1: 36261d957f9bSJan Blunck path_put(&nd.path); 36271da177e4SLinus Torvalds putname(name); 3628c6ee9206SJeff Layton if (retry_estale(error, lookup_flags)) { 3629c6ee9206SJeff Layton lookup_flags |= LOOKUP_REVAL; 3630c6ee9206SJeff Layton goto retry; 3631c6ee9206SJeff Layton } 36321da177e4SLinus Torvalds return error; 36331da177e4SLinus Torvalds } 36341da177e4SLinus Torvalds 36353cdad428SHeiko Carstens SYSCALL_DEFINE1(rmdir, const char __user *, pathname) 36365590ff0dSUlrich Drepper { 36375590ff0dSUlrich Drepper return do_rmdir(AT_FDCWD, pathname); 36385590ff0dSUlrich Drepper } 36395590ff0dSUlrich Drepper 36401da177e4SLinus Torvalds int vfs_unlink(struct inode *dir, struct dentry *dentry) 36411da177e4SLinus Torvalds { 36421da177e4SLinus Torvalds int error = may_delete(dir, dentry, 0); 36431da177e4SLinus Torvalds 36441da177e4SLinus Torvalds if (error) 36451da177e4SLinus Torvalds return error; 36461da177e4SLinus Torvalds 3647acfa4380SAl Viro if (!dir->i_op->unlink) 36481da177e4SLinus Torvalds return -EPERM; 36491da177e4SLinus Torvalds 36501b1dcc1bSJes Sorensen mutex_lock(&dentry->d_inode->i_mutex); 36511da177e4SLinus Torvalds if (d_mountpoint(dentry)) 36521da177e4SLinus Torvalds error = -EBUSY; 36531da177e4SLinus Torvalds else { 36541da177e4SLinus Torvalds error = security_inode_unlink(dir, dentry); 3655bec1052eSAl Viro if (!error) { 36561da177e4SLinus Torvalds error = dir->i_op->unlink(dir, dentry); 3657bec1052eSAl Viro if (!error) 3658d83c49f3SAl Viro dont_mount(dentry); 3659bec1052eSAl Viro } 36601da177e4SLinus Torvalds } 36611b1dcc1bSJes Sorensen mutex_unlock(&dentry->d_inode->i_mutex); 36621da177e4SLinus Torvalds 36631da177e4SLinus Torvalds /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 36641da177e4SLinus Torvalds if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { 3665ece95912SJan Kara fsnotify_link_count(dentry->d_inode); 36661da177e4SLinus Torvalds d_delete(dentry); 36671da177e4SLinus Torvalds } 36680eeca283SRobert Love 36691da177e4SLinus Torvalds return error; 36701da177e4SLinus Torvalds } 36711da177e4SLinus Torvalds 36721da177e4SLinus Torvalds /* 36731da177e4SLinus Torvalds * Make sure that the actual truncation of the file will occur outside its 36741b1dcc1bSJes Sorensen * directory's i_mutex. Truncate can take a long time if there is a lot of 36751da177e4SLinus Torvalds * writeout happening, and we don't want to prevent access to the directory 36761da177e4SLinus Torvalds * while waiting on the I/O. 36771da177e4SLinus Torvalds */ 36785590ff0dSUlrich Drepper static long do_unlinkat(int dfd, const char __user *pathname) 36791da177e4SLinus Torvalds { 36802ad94ae6SAl Viro int error; 368191a27b2aSJeff Layton struct filename *name; 36821da177e4SLinus Torvalds struct dentry *dentry; 36831da177e4SLinus Torvalds struct nameidata nd; 36841da177e4SLinus Torvalds struct inode *inode = NULL; 36855d18f813SJeff Layton unsigned int lookup_flags = 0; 36865d18f813SJeff Layton retry: 36875d18f813SJeff Layton name = user_path_parent(dfd, pathname, &nd, lookup_flags); 368891a27b2aSJeff Layton if (IS_ERR(name)) 368991a27b2aSJeff Layton return PTR_ERR(name); 36902ad94ae6SAl Viro 36911da177e4SLinus Torvalds error = -EISDIR; 36921da177e4SLinus Torvalds if (nd.last_type != LAST_NORM) 36931da177e4SLinus Torvalds goto exit1; 36940612d9fbSOGAWA Hirofumi 36950612d9fbSOGAWA Hirofumi nd.flags &= ~LOOKUP_PARENT; 3696c30dabfeSJan Kara error = mnt_want_write(nd.path.mnt); 3697c30dabfeSJan Kara if (error) 3698c30dabfeSJan Kara goto exit1; 36990612d9fbSOGAWA Hirofumi 37004ac91378SJan Blunck mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 370149705b77SChristoph Hellwig dentry = lookup_hash(&nd); 37021da177e4SLinus Torvalds error = PTR_ERR(dentry); 37031da177e4SLinus Torvalds if (!IS_ERR(dentry)) { 37041da177e4SLinus Torvalds /* Why not before? Because we want correct error value */ 370550338b88STörök Edwin if (nd.last.name[nd.last.len]) 370650338b88STörök Edwin goto slashes; 37071da177e4SLinus Torvalds inode = dentry->d_inode; 370850338b88STörök Edwin if (!inode) 3709e6bc45d6STheodore Ts'o goto slashes; 37107de9c6eeSAl Viro ihold(inode); 3711be6d3e56SKentaro Takeda error = security_path_unlink(&nd.path, dentry); 3712be6d3e56SKentaro Takeda if (error) 3713c30dabfeSJan Kara goto exit2; 37144ac91378SJan Blunck error = vfs_unlink(nd.path.dentry->d_inode, dentry); 37151da177e4SLinus Torvalds exit2: 37161da177e4SLinus Torvalds dput(dentry); 37171da177e4SLinus Torvalds } 37184ac91378SJan Blunck mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 37191da177e4SLinus Torvalds if (inode) 37201da177e4SLinus Torvalds iput(inode); /* truncate the inode here */ 3721c30dabfeSJan Kara mnt_drop_write(nd.path.mnt); 37221da177e4SLinus Torvalds exit1: 37231d957f9bSJan Blunck path_put(&nd.path); 37241da177e4SLinus Torvalds putname(name); 37255d18f813SJeff Layton if (retry_estale(error, lookup_flags)) { 37265d18f813SJeff Layton lookup_flags |= LOOKUP_REVAL; 37275d18f813SJeff Layton inode = NULL; 37285d18f813SJeff Layton goto retry; 37295d18f813SJeff Layton } 37301da177e4SLinus Torvalds return error; 37311da177e4SLinus Torvalds 37321da177e4SLinus Torvalds slashes: 37331da177e4SLinus Torvalds error = !dentry->d_inode ? -ENOENT : 37341da177e4SLinus Torvalds S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR; 37351da177e4SLinus Torvalds goto exit2; 37361da177e4SLinus Torvalds } 37371da177e4SLinus Torvalds 37382e4d0924SHeiko Carstens SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) 37395590ff0dSUlrich Drepper { 37405590ff0dSUlrich Drepper if ((flag & ~AT_REMOVEDIR) != 0) 37415590ff0dSUlrich Drepper return -EINVAL; 37425590ff0dSUlrich Drepper 37435590ff0dSUlrich Drepper if (flag & AT_REMOVEDIR) 37445590ff0dSUlrich Drepper return do_rmdir(dfd, pathname); 37455590ff0dSUlrich Drepper 37465590ff0dSUlrich Drepper return do_unlinkat(dfd, pathname); 37475590ff0dSUlrich Drepper } 37485590ff0dSUlrich Drepper 37493480b257SHeiko Carstens SYSCALL_DEFINE1(unlink, const char __user *, pathname) 37505590ff0dSUlrich Drepper { 37515590ff0dSUlrich Drepper return do_unlinkat(AT_FDCWD, pathname); 37525590ff0dSUlrich Drepper } 37535590ff0dSUlrich Drepper 3754db2e747bSMiklos Szeredi int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) 37551da177e4SLinus Torvalds { 3756a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 37571da177e4SLinus Torvalds 37581da177e4SLinus Torvalds if (error) 37591da177e4SLinus Torvalds return error; 37601da177e4SLinus Torvalds 3761acfa4380SAl Viro if (!dir->i_op->symlink) 37621da177e4SLinus Torvalds return -EPERM; 37631da177e4SLinus Torvalds 37641da177e4SLinus Torvalds error = security_inode_symlink(dir, dentry, oldname); 37651da177e4SLinus Torvalds if (error) 37661da177e4SLinus Torvalds return error; 37671da177e4SLinus Torvalds 37681da177e4SLinus Torvalds error = dir->i_op->symlink(dir, dentry, oldname); 3769a74574aaSStephen Smalley if (!error) 3770f38aa942SAmy Griffis fsnotify_create(dir, dentry); 37711da177e4SLinus Torvalds return error; 37721da177e4SLinus Torvalds } 37731da177e4SLinus Torvalds 37742e4d0924SHeiko Carstens SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, 37752e4d0924SHeiko Carstens int, newdfd, const char __user *, newname) 37761da177e4SLinus Torvalds { 37772ad94ae6SAl Viro int error; 377891a27b2aSJeff Layton struct filename *from; 37796902d925SDave Hansen struct dentry *dentry; 3780dae6ad8fSAl Viro struct path path; 3781f46d3567SJeff Layton unsigned int lookup_flags = 0; 37821da177e4SLinus Torvalds 37831da177e4SLinus Torvalds from = getname(oldname); 37841da177e4SLinus Torvalds if (IS_ERR(from)) 37851da177e4SLinus Torvalds return PTR_ERR(from); 3786f46d3567SJeff Layton retry: 3787f46d3567SJeff Layton dentry = user_path_create(newdfd, newname, &path, lookup_flags); 37881da177e4SLinus Torvalds error = PTR_ERR(dentry); 37896902d925SDave Hansen if (IS_ERR(dentry)) 3790dae6ad8fSAl Viro goto out_putname; 37916902d925SDave Hansen 379291a27b2aSJeff Layton error = security_path_symlink(&path, dentry, from->name); 3793a8104a9fSAl Viro if (!error) 379491a27b2aSJeff Layton error = vfs_symlink(path.dentry->d_inode, dentry, from->name); 3795921a1650SAl Viro done_path_create(&path, dentry); 3796f46d3567SJeff Layton if (retry_estale(error, lookup_flags)) { 3797f46d3567SJeff Layton lookup_flags |= LOOKUP_REVAL; 3798f46d3567SJeff Layton goto retry; 3799f46d3567SJeff Layton } 38006902d925SDave Hansen out_putname: 38011da177e4SLinus Torvalds putname(from); 38021da177e4SLinus Torvalds return error; 38031da177e4SLinus Torvalds } 38041da177e4SLinus Torvalds 38053480b257SHeiko Carstens SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) 38065590ff0dSUlrich Drepper { 38075590ff0dSUlrich Drepper return sys_symlinkat(oldname, AT_FDCWD, newname); 38085590ff0dSUlrich Drepper } 38095590ff0dSUlrich Drepper 38101da177e4SLinus Torvalds int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry) 38111da177e4SLinus Torvalds { 38121da177e4SLinus Torvalds struct inode *inode = old_dentry->d_inode; 38138de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 38141da177e4SLinus Torvalds int error; 38151da177e4SLinus Torvalds 38161da177e4SLinus Torvalds if (!inode) 38171da177e4SLinus Torvalds return -ENOENT; 38181da177e4SLinus Torvalds 3819a95164d9SMiklos Szeredi error = may_create(dir, new_dentry); 38201da177e4SLinus Torvalds if (error) 38211da177e4SLinus Torvalds return error; 38221da177e4SLinus Torvalds 38231da177e4SLinus Torvalds if (dir->i_sb != inode->i_sb) 38241da177e4SLinus Torvalds return -EXDEV; 38251da177e4SLinus Torvalds 38261da177e4SLinus Torvalds /* 38271da177e4SLinus Torvalds * A link to an append-only or immutable file cannot be created. 38281da177e4SLinus Torvalds */ 38291da177e4SLinus Torvalds if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 38301da177e4SLinus Torvalds return -EPERM; 3831acfa4380SAl Viro if (!dir->i_op->link) 38321da177e4SLinus Torvalds return -EPERM; 38337e79eedbSTetsuo Handa if (S_ISDIR(inode->i_mode)) 38341da177e4SLinus Torvalds return -EPERM; 38351da177e4SLinus Torvalds 38361da177e4SLinus Torvalds error = security_inode_link(old_dentry, dir, new_dentry); 38371da177e4SLinus Torvalds if (error) 38381da177e4SLinus Torvalds return error; 38391da177e4SLinus Torvalds 38407e79eedbSTetsuo Handa mutex_lock(&inode->i_mutex); 3841aae8a97dSAneesh Kumar K.V /* Make sure we don't allow creating hardlink to an unlinked file */ 3842f4e0c30cSAl Viro if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) 3843aae8a97dSAneesh Kumar K.V error = -ENOENT; 38448de52778SAl Viro else if (max_links && inode->i_nlink >= max_links) 38458de52778SAl Viro error = -EMLINK; 3846aae8a97dSAneesh Kumar K.V else 38471da177e4SLinus Torvalds error = dir->i_op->link(old_dentry, dir, new_dentry); 3848f4e0c30cSAl Viro 3849f4e0c30cSAl Viro if (!error && (inode->i_state & I_LINKABLE)) { 3850f4e0c30cSAl Viro spin_lock(&inode->i_lock); 3851f4e0c30cSAl Viro inode->i_state &= ~I_LINKABLE; 3852f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 3853f4e0c30cSAl Viro } 38547e79eedbSTetsuo Handa mutex_unlock(&inode->i_mutex); 3855e31e14ecSStephen Smalley if (!error) 38567e79eedbSTetsuo Handa fsnotify_link(dir, inode, new_dentry); 38571da177e4SLinus Torvalds return error; 38581da177e4SLinus Torvalds } 38591da177e4SLinus Torvalds 38601da177e4SLinus Torvalds /* 38611da177e4SLinus Torvalds * Hardlinks are often used in delicate situations. We avoid 38621da177e4SLinus Torvalds * security-related surprises by not following symlinks on the 38631da177e4SLinus Torvalds * newname. --KAB 38641da177e4SLinus Torvalds * 38651da177e4SLinus Torvalds * We don't follow them on the oldname either to be compatible 38661da177e4SLinus Torvalds * with linux 2.0, and to avoid hard-linking to directories 38671da177e4SLinus Torvalds * and other special files. --ADM 38681da177e4SLinus Torvalds */ 38692e4d0924SHeiko Carstens SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, 38702e4d0924SHeiko Carstens int, newdfd, const char __user *, newname, int, flags) 38711da177e4SLinus Torvalds { 38721da177e4SLinus Torvalds struct dentry *new_dentry; 3873dae6ad8fSAl Viro struct path old_path, new_path; 387411a7b371SAneesh Kumar K.V int how = 0; 38751da177e4SLinus Torvalds int error; 38761da177e4SLinus Torvalds 387711a7b371SAneesh Kumar K.V if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) 3878c04030e1SUlrich Drepper return -EINVAL; 387911a7b371SAneesh Kumar K.V /* 3880f0cc6ffbSLinus Torvalds * To use null names we require CAP_DAC_READ_SEARCH 3881f0cc6ffbSLinus Torvalds * This ensures that not everyone will be able to create 3882f0cc6ffbSLinus Torvalds * handlink using the passed filedescriptor. 388311a7b371SAneesh Kumar K.V */ 3884f0cc6ffbSLinus Torvalds if (flags & AT_EMPTY_PATH) { 3885f0cc6ffbSLinus Torvalds if (!capable(CAP_DAC_READ_SEARCH)) 3886f0cc6ffbSLinus Torvalds return -ENOENT; 388711a7b371SAneesh Kumar K.V how = LOOKUP_EMPTY; 3888f0cc6ffbSLinus Torvalds } 3889c04030e1SUlrich Drepper 389011a7b371SAneesh Kumar K.V if (flags & AT_SYMLINK_FOLLOW) 389111a7b371SAneesh Kumar K.V how |= LOOKUP_FOLLOW; 3892442e31caSJeff Layton retry: 389311a7b371SAneesh Kumar K.V error = user_path_at(olddfd, oldname, how, &old_path); 38941da177e4SLinus Torvalds if (error) 38952ad94ae6SAl Viro return error; 38962ad94ae6SAl Viro 3897442e31caSJeff Layton new_dentry = user_path_create(newdfd, newname, &new_path, 3898442e31caSJeff Layton (how & LOOKUP_REVAL)); 38991da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 39006902d925SDave Hansen if (IS_ERR(new_dentry)) 3901dae6ad8fSAl Viro goto out; 3902dae6ad8fSAl Viro 3903dae6ad8fSAl Viro error = -EXDEV; 3904dae6ad8fSAl Viro if (old_path.mnt != new_path.mnt) 3905dae6ad8fSAl Viro goto out_dput; 3906800179c9SKees Cook error = may_linkat(&old_path); 3907800179c9SKees Cook if (unlikely(error)) 3908800179c9SKees Cook goto out_dput; 3909dae6ad8fSAl Viro error = security_path_link(old_path.dentry, &new_path, new_dentry); 3910be6d3e56SKentaro Takeda if (error) 3911a8104a9fSAl Viro goto out_dput; 3912dae6ad8fSAl Viro error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry); 391375c3f29dSDave Hansen out_dput: 3914921a1650SAl Viro done_path_create(&new_path, new_dentry); 3915442e31caSJeff Layton if (retry_estale(error, how)) { 3916442e31caSJeff Layton how |= LOOKUP_REVAL; 3917442e31caSJeff Layton goto retry; 3918442e31caSJeff Layton } 39191da177e4SLinus Torvalds out: 39202d8f3038SAl Viro path_put(&old_path); 39211da177e4SLinus Torvalds 39221da177e4SLinus Torvalds return error; 39231da177e4SLinus Torvalds } 39241da177e4SLinus Torvalds 39253480b257SHeiko Carstens SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) 39265590ff0dSUlrich Drepper { 3927c04030e1SUlrich Drepper return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 39285590ff0dSUlrich Drepper } 39295590ff0dSUlrich Drepper 39301da177e4SLinus Torvalds /* 39311da177e4SLinus Torvalds * The worst of all namespace operations - renaming directory. "Perverted" 39321da177e4SLinus Torvalds * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 39331da177e4SLinus Torvalds * Problems: 39341da177e4SLinus Torvalds * a) we can get into loop creation. Check is done in is_subdir(). 39351da177e4SLinus Torvalds * b) race potential - two innocent renames can create a loop together. 39361da177e4SLinus Torvalds * That's where 4.4 screws up. Current fix: serialization on 3937a11f3a05SArjan van de Ven * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 39381da177e4SLinus Torvalds * story. 39391da177e4SLinus Torvalds * c) we have to lock _three_ objects - parents and victim (if it exists). 39401b1dcc1bSJes Sorensen * And that - after we got ->i_mutex on parents (until then we don't know 39411da177e4SLinus Torvalds * whether the target exists). Solution: try to be smart with locking 39421da177e4SLinus Torvalds * order for inodes. We rely on the fact that tree topology may change 3943a11f3a05SArjan van de Ven * only under ->s_vfs_rename_mutex _and_ that parent of the object we 39441da177e4SLinus Torvalds * move will be locked. Thus we can rank directories by the tree 39451da177e4SLinus Torvalds * (ancestors first) and rank all non-directories after them. 39461da177e4SLinus Torvalds * That works since everybody except rename does "lock parent, lookup, 3947a11f3a05SArjan van de Ven * lock child" and rename is under ->s_vfs_rename_mutex. 39481da177e4SLinus Torvalds * HOWEVER, it relies on the assumption that any object with ->lookup() 39491da177e4SLinus Torvalds * has no more than 1 dentry. If "hybrid" objects will ever appear, 39501da177e4SLinus Torvalds * we'd better make sure that there's no link(2) for them. 3951e4eaac06SSage Weil * d) conversion from fhandle to dentry may come in the wrong moment - when 39521b1dcc1bSJes Sorensen * we are removing the target. Solution: we will have to grab ->i_mutex 39531da177e4SLinus Torvalds * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 3954c41b20e7SAdam Buchbinder * ->i_mutex on parents, which works but leads to some truly excessive 39551da177e4SLinus Torvalds * locking]. 39561da177e4SLinus Torvalds */ 395775c96f85SAdrian Bunk static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry, 39581da177e4SLinus Torvalds struct inode *new_dir, struct dentry *new_dentry) 39591da177e4SLinus Torvalds { 39601da177e4SLinus Torvalds int error = 0; 39619055cba7SSage Weil struct inode *target = new_dentry->d_inode; 39628de52778SAl Viro unsigned max_links = new_dir->i_sb->s_max_links; 39631da177e4SLinus Torvalds 39641da177e4SLinus Torvalds /* 39651da177e4SLinus Torvalds * If we are going to change the parent - check write permissions, 39661da177e4SLinus Torvalds * we'll need to flip '..'. 39671da177e4SLinus Torvalds */ 39681da177e4SLinus Torvalds if (new_dir != old_dir) { 3969f419a2e3SAl Viro error = inode_permission(old_dentry->d_inode, MAY_WRITE); 39701da177e4SLinus Torvalds if (error) 39711da177e4SLinus Torvalds return error; 39721da177e4SLinus Torvalds } 39731da177e4SLinus Torvalds 39741da177e4SLinus Torvalds error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); 39751da177e4SLinus Torvalds if (error) 39761da177e4SLinus Torvalds return error; 39771da177e4SLinus Torvalds 39781d2ef590SAl Viro dget(new_dentry); 3979d83c49f3SAl Viro if (target) 39801b1dcc1bSJes Sorensen mutex_lock(&target->i_mutex); 39819055cba7SSage Weil 39821da177e4SLinus Torvalds error = -EBUSY; 39839055cba7SSage Weil if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry)) 39849055cba7SSage Weil goto out; 39859055cba7SSage Weil 39868de52778SAl Viro error = -EMLINK; 39878de52778SAl Viro if (max_links && !target && new_dir != old_dir && 39888de52778SAl Viro new_dir->i_nlink >= max_links) 39898de52778SAl Viro goto out; 39908de52778SAl Viro 39913cebde24SSage Weil if (target) 39923cebde24SSage Weil shrink_dcache_parent(new_dentry); 39931da177e4SLinus Torvalds error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); 39949055cba7SSage Weil if (error) 39959055cba7SSage Weil goto out; 39969055cba7SSage Weil 39971da177e4SLinus Torvalds if (target) { 39981da177e4SLinus Torvalds target->i_flags |= S_DEAD; 3999d83c49f3SAl Viro dont_mount(new_dentry); 4000d83c49f3SAl Viro } 40019055cba7SSage Weil out: 40029055cba7SSage Weil if (target) 40031b1dcc1bSJes Sorensen mutex_unlock(&target->i_mutex); 40041d2ef590SAl Viro dput(new_dentry); 4005e31e14ecSStephen Smalley if (!error) 4006349457ccSMark Fasheh if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) 40071da177e4SLinus Torvalds d_move(old_dentry,new_dentry); 40081da177e4SLinus Torvalds return error; 40091da177e4SLinus Torvalds } 40101da177e4SLinus Torvalds 401175c96f85SAdrian Bunk static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry, 40121da177e4SLinus Torvalds struct inode *new_dir, struct dentry *new_dentry) 40131da177e4SLinus Torvalds { 401451892bbbSSage Weil struct inode *target = new_dentry->d_inode; 40151da177e4SLinus Torvalds int error; 40161da177e4SLinus Torvalds 40171da177e4SLinus Torvalds error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); 40181da177e4SLinus Torvalds if (error) 40191da177e4SLinus Torvalds return error; 40201da177e4SLinus Torvalds 40211da177e4SLinus Torvalds dget(new_dentry); 40221da177e4SLinus Torvalds if (target) 40231b1dcc1bSJes Sorensen mutex_lock(&target->i_mutex); 402451892bbbSSage Weil 40251da177e4SLinus Torvalds error = -EBUSY; 402651892bbbSSage Weil if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry)) 402751892bbbSSage Weil goto out; 402851892bbbSSage Weil 40291da177e4SLinus Torvalds error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); 403051892bbbSSage Weil if (error) 403151892bbbSSage Weil goto out; 403251892bbbSSage Weil 4033bec1052eSAl Viro if (target) 4034d83c49f3SAl Viro dont_mount(new_dentry); 4035349457ccSMark Fasheh if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) 40361da177e4SLinus Torvalds d_move(old_dentry, new_dentry); 403751892bbbSSage Weil out: 40381da177e4SLinus Torvalds if (target) 40391b1dcc1bSJes Sorensen mutex_unlock(&target->i_mutex); 40401da177e4SLinus Torvalds dput(new_dentry); 40411da177e4SLinus Torvalds return error; 40421da177e4SLinus Torvalds } 40431da177e4SLinus Torvalds 40441da177e4SLinus Torvalds int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, 40451da177e4SLinus Torvalds struct inode *new_dir, struct dentry *new_dentry) 40461da177e4SLinus Torvalds { 40471da177e4SLinus Torvalds int error; 40481da177e4SLinus Torvalds int is_dir = S_ISDIR(old_dentry->d_inode->i_mode); 404959b0df21SEric Paris const unsigned char *old_name; 40501da177e4SLinus Torvalds 40511da177e4SLinus Torvalds if (old_dentry->d_inode == new_dentry->d_inode) 40521da177e4SLinus Torvalds return 0; 40531da177e4SLinus Torvalds 40541da177e4SLinus Torvalds error = may_delete(old_dir, old_dentry, is_dir); 40551da177e4SLinus Torvalds if (error) 40561da177e4SLinus Torvalds return error; 40571da177e4SLinus Torvalds 40581da177e4SLinus Torvalds if (!new_dentry->d_inode) 4059a95164d9SMiklos Szeredi error = may_create(new_dir, new_dentry); 40601da177e4SLinus Torvalds else 40611da177e4SLinus Torvalds error = may_delete(new_dir, new_dentry, is_dir); 40621da177e4SLinus Torvalds if (error) 40631da177e4SLinus Torvalds return error; 40641da177e4SLinus Torvalds 4065acfa4380SAl Viro if (!old_dir->i_op->rename) 40661da177e4SLinus Torvalds return -EPERM; 40671da177e4SLinus Torvalds 40680eeca283SRobert Love old_name = fsnotify_oldname_init(old_dentry->d_name.name); 40690eeca283SRobert Love 40701da177e4SLinus Torvalds if (is_dir) 40711da177e4SLinus Torvalds error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry); 40721da177e4SLinus Torvalds else 40731da177e4SLinus Torvalds error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry); 4074123df294SAl Viro if (!error) 4075123df294SAl Viro fsnotify_move(old_dir, new_dir, old_name, is_dir, 40765a190ae6SAl Viro new_dentry->d_inode, old_dentry); 40770eeca283SRobert Love fsnotify_oldname_free(old_name); 40780eeca283SRobert Love 40791da177e4SLinus Torvalds return error; 40801da177e4SLinus Torvalds } 40811da177e4SLinus Torvalds 40822e4d0924SHeiko Carstens SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, 40832e4d0924SHeiko Carstens int, newdfd, const char __user *, newname) 40841da177e4SLinus Torvalds { 40851da177e4SLinus Torvalds struct dentry *old_dir, *new_dir; 40861da177e4SLinus Torvalds struct dentry *old_dentry, *new_dentry; 40871da177e4SLinus Torvalds struct dentry *trap; 40881da177e4SLinus Torvalds struct nameidata oldnd, newnd; 408991a27b2aSJeff Layton struct filename *from; 409091a27b2aSJeff Layton struct filename *to; 4091c6a94284SJeff Layton unsigned int lookup_flags = 0; 4092c6a94284SJeff Layton bool should_retry = false; 40932ad94ae6SAl Viro int error; 4094c6a94284SJeff Layton retry: 4095c6a94284SJeff Layton from = user_path_parent(olddfd, oldname, &oldnd, lookup_flags); 409691a27b2aSJeff Layton if (IS_ERR(from)) { 409791a27b2aSJeff Layton error = PTR_ERR(from); 40981da177e4SLinus Torvalds goto exit; 409991a27b2aSJeff Layton } 41001da177e4SLinus Torvalds 4101c6a94284SJeff Layton to = user_path_parent(newdfd, newname, &newnd, lookup_flags); 410291a27b2aSJeff Layton if (IS_ERR(to)) { 410391a27b2aSJeff Layton error = PTR_ERR(to); 41041da177e4SLinus Torvalds goto exit1; 410591a27b2aSJeff Layton } 41061da177e4SLinus Torvalds 41071da177e4SLinus Torvalds error = -EXDEV; 41084ac91378SJan Blunck if (oldnd.path.mnt != newnd.path.mnt) 41091da177e4SLinus Torvalds goto exit2; 41101da177e4SLinus Torvalds 41114ac91378SJan Blunck old_dir = oldnd.path.dentry; 41121da177e4SLinus Torvalds error = -EBUSY; 41131da177e4SLinus Torvalds if (oldnd.last_type != LAST_NORM) 41141da177e4SLinus Torvalds goto exit2; 41151da177e4SLinus Torvalds 41164ac91378SJan Blunck new_dir = newnd.path.dentry; 41171da177e4SLinus Torvalds if (newnd.last_type != LAST_NORM) 41181da177e4SLinus Torvalds goto exit2; 41191da177e4SLinus Torvalds 4120c30dabfeSJan Kara error = mnt_want_write(oldnd.path.mnt); 4121c30dabfeSJan Kara if (error) 4122c30dabfeSJan Kara goto exit2; 4123c30dabfeSJan Kara 41240612d9fbSOGAWA Hirofumi oldnd.flags &= ~LOOKUP_PARENT; 41250612d9fbSOGAWA Hirofumi newnd.flags &= ~LOOKUP_PARENT; 41264e9ed2f8SOGAWA Hirofumi newnd.flags |= LOOKUP_RENAME_TARGET; 41270612d9fbSOGAWA Hirofumi 41281da177e4SLinus Torvalds trap = lock_rename(new_dir, old_dir); 41291da177e4SLinus Torvalds 413049705b77SChristoph Hellwig old_dentry = lookup_hash(&oldnd); 41311da177e4SLinus Torvalds error = PTR_ERR(old_dentry); 41321da177e4SLinus Torvalds if (IS_ERR(old_dentry)) 41331da177e4SLinus Torvalds goto exit3; 41341da177e4SLinus Torvalds /* source must exist */ 41351da177e4SLinus Torvalds error = -ENOENT; 41361da177e4SLinus Torvalds if (!old_dentry->d_inode) 41371da177e4SLinus Torvalds goto exit4; 41381da177e4SLinus Torvalds /* unless the source is a directory trailing slashes give -ENOTDIR */ 41391da177e4SLinus Torvalds if (!S_ISDIR(old_dentry->d_inode->i_mode)) { 41401da177e4SLinus Torvalds error = -ENOTDIR; 41411da177e4SLinus Torvalds if (oldnd.last.name[oldnd.last.len]) 41421da177e4SLinus Torvalds goto exit4; 41431da177e4SLinus Torvalds if (newnd.last.name[newnd.last.len]) 41441da177e4SLinus Torvalds goto exit4; 41451da177e4SLinus Torvalds } 41461da177e4SLinus Torvalds /* source should not be ancestor of target */ 41471da177e4SLinus Torvalds error = -EINVAL; 41481da177e4SLinus Torvalds if (old_dentry == trap) 41491da177e4SLinus Torvalds goto exit4; 415049705b77SChristoph Hellwig new_dentry = lookup_hash(&newnd); 41511da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 41521da177e4SLinus Torvalds if (IS_ERR(new_dentry)) 41531da177e4SLinus Torvalds goto exit4; 41541da177e4SLinus Torvalds /* target should not be an ancestor of source */ 41551da177e4SLinus Torvalds error = -ENOTEMPTY; 41561da177e4SLinus Torvalds if (new_dentry == trap) 41571da177e4SLinus Torvalds goto exit5; 41581da177e4SLinus Torvalds 4159be6d3e56SKentaro Takeda error = security_path_rename(&oldnd.path, old_dentry, 4160be6d3e56SKentaro Takeda &newnd.path, new_dentry); 4161be6d3e56SKentaro Takeda if (error) 4162c30dabfeSJan Kara goto exit5; 41631da177e4SLinus Torvalds error = vfs_rename(old_dir->d_inode, old_dentry, 41641da177e4SLinus Torvalds new_dir->d_inode, new_dentry); 41651da177e4SLinus Torvalds exit5: 41661da177e4SLinus Torvalds dput(new_dentry); 41671da177e4SLinus Torvalds exit4: 41681da177e4SLinus Torvalds dput(old_dentry); 41691da177e4SLinus Torvalds exit3: 41701da177e4SLinus Torvalds unlock_rename(new_dir, old_dir); 4171c30dabfeSJan Kara mnt_drop_write(oldnd.path.mnt); 41721da177e4SLinus Torvalds exit2: 4173c6a94284SJeff Layton if (retry_estale(error, lookup_flags)) 4174c6a94284SJeff Layton should_retry = true; 41751d957f9bSJan Blunck path_put(&newnd.path); 41762ad94ae6SAl Viro putname(to); 41771da177e4SLinus Torvalds exit1: 41781d957f9bSJan Blunck path_put(&oldnd.path); 41791da177e4SLinus Torvalds putname(from); 4180c6a94284SJeff Layton if (should_retry) { 4181c6a94284SJeff Layton should_retry = false; 4182c6a94284SJeff Layton lookup_flags |= LOOKUP_REVAL; 4183c6a94284SJeff Layton goto retry; 4184c6a94284SJeff Layton } 41852ad94ae6SAl Viro exit: 41861da177e4SLinus Torvalds return error; 41871da177e4SLinus Torvalds } 41881da177e4SLinus Torvalds 4189a26eab24SHeiko Carstens SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) 41905590ff0dSUlrich Drepper { 41915590ff0dSUlrich Drepper return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname); 41925590ff0dSUlrich Drepper } 41935590ff0dSUlrich Drepper 41941da177e4SLinus Torvalds int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link) 41951da177e4SLinus Torvalds { 41961da177e4SLinus Torvalds int len; 41971da177e4SLinus Torvalds 41981da177e4SLinus Torvalds len = PTR_ERR(link); 41991da177e4SLinus Torvalds if (IS_ERR(link)) 42001da177e4SLinus Torvalds goto out; 42011da177e4SLinus Torvalds 42021da177e4SLinus Torvalds len = strlen(link); 42031da177e4SLinus Torvalds if (len > (unsigned) buflen) 42041da177e4SLinus Torvalds len = buflen; 42051da177e4SLinus Torvalds if (copy_to_user(buffer, link, len)) 42061da177e4SLinus Torvalds len = -EFAULT; 42071da177e4SLinus Torvalds out: 42081da177e4SLinus Torvalds return len; 42091da177e4SLinus Torvalds } 42101da177e4SLinus Torvalds 42111da177e4SLinus Torvalds /* 42121da177e4SLinus Torvalds * A helper for ->readlink(). This should be used *ONLY* for symlinks that 42131da177e4SLinus Torvalds * have ->follow_link() touching nd only in nd_set_link(). Using (or not 42141da177e4SLinus Torvalds * using) it for any given inode is up to filesystem. 42151da177e4SLinus Torvalds */ 42161da177e4SLinus Torvalds int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen) 42171da177e4SLinus Torvalds { 42181da177e4SLinus Torvalds struct nameidata nd; 4219cc314eefSLinus Torvalds void *cookie; 4220694a1764SMarcin Slusarz int res; 4221cc314eefSLinus Torvalds 42221da177e4SLinus Torvalds nd.depth = 0; 4223cc314eefSLinus Torvalds cookie = dentry->d_inode->i_op->follow_link(dentry, &nd); 4224694a1764SMarcin Slusarz if (IS_ERR(cookie)) 4225694a1764SMarcin Slusarz return PTR_ERR(cookie); 4226694a1764SMarcin Slusarz 4227694a1764SMarcin Slusarz res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd)); 42281da177e4SLinus Torvalds if (dentry->d_inode->i_op->put_link) 4229cc314eefSLinus Torvalds dentry->d_inode->i_op->put_link(dentry, &nd, cookie); 4230694a1764SMarcin Slusarz return res; 42311da177e4SLinus Torvalds } 42321da177e4SLinus Torvalds 42331da177e4SLinus Torvalds int vfs_follow_link(struct nameidata *nd, const char *link) 42341da177e4SLinus Torvalds { 42351da177e4SLinus Torvalds return __vfs_follow_link(nd, link); 42361da177e4SLinus Torvalds } 42371da177e4SLinus Torvalds 42381da177e4SLinus Torvalds /* get the link contents into pagecache */ 42391da177e4SLinus Torvalds static char *page_getlink(struct dentry * dentry, struct page **ppage) 42401da177e4SLinus Torvalds { 4241ebd09abbSDuane Griffin char *kaddr; 42421da177e4SLinus Torvalds struct page *page; 42431da177e4SLinus Torvalds struct address_space *mapping = dentry->d_inode->i_mapping; 4244090d2b18SPekka Enberg page = read_mapping_page(mapping, 0, NULL); 42451da177e4SLinus Torvalds if (IS_ERR(page)) 42466fe6900eSNick Piggin return (char*)page; 42471da177e4SLinus Torvalds *ppage = page; 4248ebd09abbSDuane Griffin kaddr = kmap(page); 4249ebd09abbSDuane Griffin nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1); 4250ebd09abbSDuane Griffin return kaddr; 42511da177e4SLinus Torvalds } 42521da177e4SLinus Torvalds 42531da177e4SLinus Torvalds int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 42541da177e4SLinus Torvalds { 42551da177e4SLinus Torvalds struct page *page = NULL; 42561da177e4SLinus Torvalds char *s = page_getlink(dentry, &page); 42571da177e4SLinus Torvalds int res = vfs_readlink(dentry,buffer,buflen,s); 42581da177e4SLinus Torvalds if (page) { 42591da177e4SLinus Torvalds kunmap(page); 42601da177e4SLinus Torvalds page_cache_release(page); 42611da177e4SLinus Torvalds } 42621da177e4SLinus Torvalds return res; 42631da177e4SLinus Torvalds } 42641da177e4SLinus Torvalds 4265cc314eefSLinus Torvalds void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd) 42661da177e4SLinus Torvalds { 4267cc314eefSLinus Torvalds struct page *page = NULL; 42681da177e4SLinus Torvalds nd_set_link(nd, page_getlink(dentry, &page)); 4269cc314eefSLinus Torvalds return page; 42701da177e4SLinus Torvalds } 42711da177e4SLinus Torvalds 4272cc314eefSLinus Torvalds void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) 42731da177e4SLinus Torvalds { 4274cc314eefSLinus Torvalds struct page *page = cookie; 4275cc314eefSLinus Torvalds 4276cc314eefSLinus Torvalds if (page) { 42771da177e4SLinus Torvalds kunmap(page); 42781da177e4SLinus Torvalds page_cache_release(page); 42791da177e4SLinus Torvalds } 42801da177e4SLinus Torvalds } 42811da177e4SLinus Torvalds 428254566b2cSNick Piggin /* 428354566b2cSNick Piggin * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS 428454566b2cSNick Piggin */ 428554566b2cSNick Piggin int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) 42861da177e4SLinus Torvalds { 42871da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 42880adb25d2SKirill Korotaev struct page *page; 4289afddba49SNick Piggin void *fsdata; 4290beb497abSDmitriy Monakhov int err; 42911da177e4SLinus Torvalds char *kaddr; 429254566b2cSNick Piggin unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE; 429354566b2cSNick Piggin if (nofs) 429454566b2cSNick Piggin flags |= AOP_FLAG_NOFS; 42951da177e4SLinus Torvalds 42967e53cac4SNeilBrown retry: 4297afddba49SNick Piggin err = pagecache_write_begin(NULL, mapping, 0, len-1, 429854566b2cSNick Piggin flags, &page, &fsdata); 42991da177e4SLinus Torvalds if (err) 4300afddba49SNick Piggin goto fail; 4301afddba49SNick Piggin 4302e8e3c3d6SCong Wang kaddr = kmap_atomic(page); 43031da177e4SLinus Torvalds memcpy(kaddr, symname, len-1); 4304e8e3c3d6SCong Wang kunmap_atomic(kaddr); 4305afddba49SNick Piggin 4306afddba49SNick Piggin err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, 4307afddba49SNick Piggin page, fsdata); 43081da177e4SLinus Torvalds if (err < 0) 43091da177e4SLinus Torvalds goto fail; 4310afddba49SNick Piggin if (err < len-1) 4311afddba49SNick Piggin goto retry; 4312afddba49SNick Piggin 43131da177e4SLinus Torvalds mark_inode_dirty(inode); 43141da177e4SLinus Torvalds return 0; 43151da177e4SLinus Torvalds fail: 43161da177e4SLinus Torvalds return err; 43171da177e4SLinus Torvalds } 43181da177e4SLinus Torvalds 43190adb25d2SKirill Korotaev int page_symlink(struct inode *inode, const char *symname, int len) 43200adb25d2SKirill Korotaev { 43210adb25d2SKirill Korotaev return __page_symlink(inode, symname, len, 432254566b2cSNick Piggin !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS)); 43230adb25d2SKirill Korotaev } 43240adb25d2SKirill Korotaev 432592e1d5beSArjan van de Ven const struct inode_operations page_symlink_inode_operations = { 43261da177e4SLinus Torvalds .readlink = generic_readlink, 43271da177e4SLinus Torvalds .follow_link = page_follow_link_light, 43281da177e4SLinus Torvalds .put_link = page_put_link, 43291da177e4SLinus Torvalds }; 43301da177e4SLinus Torvalds 43312d8f3038SAl Viro EXPORT_SYMBOL(user_path_at); 4332cc53ce53SDavid Howells EXPORT_SYMBOL(follow_down_one); 43331da177e4SLinus Torvalds EXPORT_SYMBOL(follow_down); 43341da177e4SLinus Torvalds EXPORT_SYMBOL(follow_up); 4335f6d2ac5cSAl Viro EXPORT_SYMBOL(get_write_access); /* nfsd */ 43361da177e4SLinus Torvalds EXPORT_SYMBOL(lock_rename); 43371da177e4SLinus Torvalds EXPORT_SYMBOL(lookup_one_len); 43381da177e4SLinus Torvalds EXPORT_SYMBOL(page_follow_link_light); 43391da177e4SLinus Torvalds EXPORT_SYMBOL(page_put_link); 43401da177e4SLinus Torvalds EXPORT_SYMBOL(page_readlink); 43410adb25d2SKirill Korotaev EXPORT_SYMBOL(__page_symlink); 43421da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink); 43431da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink_inode_operations); 4344d1811465SAl Viro EXPORT_SYMBOL(kern_path); 434516f18200SJosef 'Jeff' Sipek EXPORT_SYMBOL(vfs_path_lookup); 4346f419a2e3SAl Viro EXPORT_SYMBOL(inode_permission); 43471da177e4SLinus Torvalds EXPORT_SYMBOL(unlock_rename); 43481da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_create); 43491da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_follow_link); 43501da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_link); 43511da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_mkdir); 43521da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_mknod); 43531da177e4SLinus Torvalds EXPORT_SYMBOL(generic_permission); 43541da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_readlink); 43551da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_rename); 43561da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_rmdir); 43571da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_symlink); 43581da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_unlink); 43591da177e4SLinus Torvalds EXPORT_SYMBOL(dentry_unhash); 43601da177e4SLinus Torvalds EXPORT_SYMBOL(generic_readlink); 4361