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 48531e6b01fSNick Piggin /** 48619660af7SAl Viro * unlazy_walk - try to switch to ref-walk mode. 48719660af7SAl Viro * @nd: nameidata pathwalk data 48819660af7SAl Viro * @dentry: child of nd->path.dentry or NULL 48939191628SRandy Dunlap * Returns: 0 on success, -ECHILD on failure 49031e6b01fSNick Piggin * 49119660af7SAl Viro * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry 49219660af7SAl Viro * for ref-walk mode. @dentry must be a path found by a do_lookup call on 49319660af7SAl Viro * @nd or NULL. Must be called from rcu-walk context. 49431e6b01fSNick Piggin */ 49519660af7SAl Viro static int unlazy_walk(struct nameidata *nd, struct dentry *dentry) 49631e6b01fSNick Piggin { 49731e6b01fSNick Piggin struct fs_struct *fs = current->fs; 49831e6b01fSNick Piggin struct dentry *parent = nd->path.dentry; 49931e6b01fSNick Piggin 50031e6b01fSNick Piggin BUG_ON(!(nd->flags & LOOKUP_RCU)); 501e5c832d5SLinus Torvalds 502e5c832d5SLinus Torvalds /* 50348a066e7SAl Viro * After legitimizing the bastards, terminate_walk() 50448a066e7SAl Viro * will do the right thing for non-RCU mode, and all our 50548a066e7SAl Viro * subsequent exit cases should rcu_read_unlock() 50648a066e7SAl Viro * before returning. Do vfsmount first; if dentry 50748a066e7SAl Viro * can't be legitimized, just set nd->path.dentry to NULL 50848a066e7SAl Viro * and rely on dput(NULL) being a no-op. 509e5c832d5SLinus Torvalds */ 51048a066e7SAl Viro if (!legitimize_mnt(nd->path.mnt, nd->m_seq)) 511e5c832d5SLinus Torvalds return -ECHILD; 512e5c832d5SLinus Torvalds nd->flags &= ~LOOKUP_RCU; 51315570086SLinus Torvalds 51448a066e7SAl Viro if (!lockref_get_not_dead(&parent->d_lockref)) { 51548a066e7SAl Viro nd->path.dentry = NULL; 5168b61e74fSAl Viro rcu_read_unlock(); 51748a066e7SAl Viro return -ECHILD; 51848a066e7SAl Viro } 51948a066e7SAl Viro 52015570086SLinus Torvalds /* 52115570086SLinus Torvalds * For a negative lookup, the lookup sequence point is the parents 52215570086SLinus Torvalds * sequence point, and it only needs to revalidate the parent dentry. 52315570086SLinus Torvalds * 52415570086SLinus Torvalds * For a positive lookup, we need to move both the parent and the 52515570086SLinus Torvalds * dentry from the RCU domain to be properly refcounted. And the 52615570086SLinus Torvalds * sequence number in the dentry validates *both* dentry counters, 52715570086SLinus Torvalds * since we checked the sequence number of the parent after we got 52815570086SLinus Torvalds * the child sequence number. So we know the parent must still 52915570086SLinus Torvalds * be valid if the child sequence number is still valid. 53015570086SLinus Torvalds */ 53119660af7SAl Viro if (!dentry) { 532e5c832d5SLinus Torvalds if (read_seqcount_retry(&parent->d_seq, nd->seq)) 533e5c832d5SLinus Torvalds goto out; 53419660af7SAl Viro BUG_ON(nd->inode != parent->d_inode); 53519660af7SAl Viro } else { 536e5c832d5SLinus Torvalds if (!lockref_get_not_dead(&dentry->d_lockref)) 537e5c832d5SLinus Torvalds goto out; 538e5c832d5SLinus Torvalds if (read_seqcount_retry(&dentry->d_seq, nd->seq)) 539e5c832d5SLinus Torvalds goto drop_dentry; 54019660af7SAl Viro } 541e5c832d5SLinus Torvalds 542e5c832d5SLinus Torvalds /* 543e5c832d5SLinus Torvalds * Sequence counts matched. Now make sure that the root is 544e5c832d5SLinus Torvalds * still valid and get it if required. 545e5c832d5SLinus Torvalds */ 546e5c832d5SLinus Torvalds if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 547e5c832d5SLinus Torvalds spin_lock(&fs->lock); 548e5c832d5SLinus Torvalds if (nd->root.mnt != fs->root.mnt || nd->root.dentry != fs->root.dentry) 549e5c832d5SLinus Torvalds goto unlock_and_drop_dentry; 55031e6b01fSNick Piggin path_get(&nd->root); 55131e6b01fSNick Piggin spin_unlock(&fs->lock); 55231e6b01fSNick Piggin } 55331e6b01fSNick Piggin 5548b61e74fSAl Viro rcu_read_unlock(); 55531e6b01fSNick Piggin return 0; 55619660af7SAl Viro 557e5c832d5SLinus Torvalds unlock_and_drop_dentry: 55831e6b01fSNick Piggin spin_unlock(&fs->lock); 559e5c832d5SLinus Torvalds drop_dentry: 5608b61e74fSAl Viro rcu_read_unlock(); 561e5c832d5SLinus Torvalds dput(dentry); 562d0d27277SLinus Torvalds goto drop_root_mnt; 563e5c832d5SLinus Torvalds out: 5648b61e74fSAl Viro rcu_read_unlock(); 565d0d27277SLinus Torvalds drop_root_mnt: 566d0d27277SLinus Torvalds if (!(nd->flags & LOOKUP_ROOT)) 567d0d27277SLinus Torvalds nd->root.mnt = NULL; 56831e6b01fSNick Piggin return -ECHILD; 56931e6b01fSNick Piggin } 57031e6b01fSNick Piggin 5714ce16ef3SAl Viro static inline int d_revalidate(struct dentry *dentry, unsigned int flags) 57234286d66SNick Piggin { 5734ce16ef3SAl Viro return dentry->d_op->d_revalidate(dentry, flags); 57434286d66SNick Piggin } 57534286d66SNick Piggin 5769f1fafeeSAl Viro /** 5779f1fafeeSAl Viro * complete_walk - successful completion of path walk 5789f1fafeeSAl Viro * @nd: pointer nameidata 57939159de2SJeff Layton * 5809f1fafeeSAl Viro * If we had been in RCU mode, drop out of it and legitimize nd->path. 5819f1fafeeSAl Viro * Revalidate the final result, unless we'd already done that during 5829f1fafeeSAl Viro * the path walk or the filesystem doesn't ask for it. Return 0 on 5839f1fafeeSAl Viro * success, -error on failure. In case of failure caller does not 5849f1fafeeSAl Viro * need to drop nd->path. 58539159de2SJeff Layton */ 5869f1fafeeSAl Viro static int complete_walk(struct nameidata *nd) 58739159de2SJeff Layton { 58816c2cd71SAl Viro struct dentry *dentry = nd->path.dentry; 58939159de2SJeff Layton int status; 59039159de2SJeff Layton 5919f1fafeeSAl Viro if (nd->flags & LOOKUP_RCU) { 5929f1fafeeSAl Viro nd->flags &= ~LOOKUP_RCU; 5939f1fafeeSAl Viro if (!(nd->flags & LOOKUP_ROOT)) 5949f1fafeeSAl Viro nd->root.mnt = NULL; 59515570086SLinus Torvalds 59648a066e7SAl Viro if (!legitimize_mnt(nd->path.mnt, nd->m_seq)) { 5978b61e74fSAl Viro rcu_read_unlock(); 59848a066e7SAl Viro return -ECHILD; 59948a066e7SAl Viro } 600e5c832d5SLinus Torvalds if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) { 6018b61e74fSAl Viro rcu_read_unlock(); 60248a066e7SAl Viro mntput(nd->path.mnt); 6039f1fafeeSAl Viro return -ECHILD; 6049f1fafeeSAl Viro } 605e5c832d5SLinus Torvalds if (read_seqcount_retry(&dentry->d_seq, nd->seq)) { 6068b61e74fSAl Viro rcu_read_unlock(); 607e5c832d5SLinus Torvalds dput(dentry); 60848a066e7SAl Viro mntput(nd->path.mnt); 609e5c832d5SLinus Torvalds return -ECHILD; 610e5c832d5SLinus Torvalds } 6118b61e74fSAl Viro rcu_read_unlock(); 6129f1fafeeSAl Viro } 6139f1fafeeSAl Viro 61416c2cd71SAl Viro if (likely(!(nd->flags & LOOKUP_JUMPED))) 61539159de2SJeff Layton return 0; 61639159de2SJeff Layton 617ecf3d1f1SJeff Layton if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) 61816c2cd71SAl Viro return 0; 61916c2cd71SAl Viro 620ecf3d1f1SJeff Layton status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); 62139159de2SJeff Layton if (status > 0) 62239159de2SJeff Layton return 0; 62339159de2SJeff Layton 62416c2cd71SAl Viro if (!status) 62539159de2SJeff Layton status = -ESTALE; 62616c2cd71SAl Viro 6279f1fafeeSAl Viro path_put(&nd->path); 62839159de2SJeff Layton return status; 62939159de2SJeff Layton } 63039159de2SJeff Layton 6312a737871SAl Viro static __always_inline void set_root(struct nameidata *nd) 6322a737871SAl Viro { 633f7ad3c6bSMiklos Szeredi if (!nd->root.mnt) 634f7ad3c6bSMiklos Szeredi get_fs_root(current->fs, &nd->root); 6352a737871SAl Viro } 6362a737871SAl Viro 6376de88d72SAl Viro static int link_path_walk(const char *, struct nameidata *); 6386de88d72SAl Viro 63931e6b01fSNick Piggin static __always_inline void set_root_rcu(struct nameidata *nd) 64031e6b01fSNick Piggin { 64131e6b01fSNick Piggin if (!nd->root.mnt) { 64231e6b01fSNick Piggin struct fs_struct *fs = current->fs; 643c28cc364SNick Piggin unsigned seq; 644c28cc364SNick Piggin 645c28cc364SNick Piggin do { 646c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 64731e6b01fSNick Piggin nd->root = fs->root; 648c1530019STim Chen nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq); 649c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 65031e6b01fSNick Piggin } 65131e6b01fSNick Piggin } 65231e6b01fSNick Piggin 6531d957f9bSJan Blunck static void path_put_conditional(struct path *path, struct nameidata *nd) 654051d3812SIan Kent { 655051d3812SIan Kent dput(path->dentry); 6564ac91378SJan Blunck if (path->mnt != nd->path.mnt) 657051d3812SIan Kent mntput(path->mnt); 658051d3812SIan Kent } 659051d3812SIan Kent 6607b9337aaSNick Piggin static inline void path_to_nameidata(const struct path *path, 6617b9337aaSNick Piggin struct nameidata *nd) 662051d3812SIan Kent { 66331e6b01fSNick Piggin if (!(nd->flags & LOOKUP_RCU)) { 6644ac91378SJan Blunck dput(nd->path.dentry); 66531e6b01fSNick Piggin if (nd->path.mnt != path->mnt) 6664ac91378SJan Blunck mntput(nd->path.mnt); 6679a229683SHuang Shijie } 66831e6b01fSNick Piggin nd->path.mnt = path->mnt; 6694ac91378SJan Blunck nd->path.dentry = path->dentry; 670051d3812SIan Kent } 671051d3812SIan Kent 672b5fb63c1SChristoph Hellwig /* 673b5fb63c1SChristoph Hellwig * Helper to directly jump to a known parsed path from ->follow_link, 674b5fb63c1SChristoph Hellwig * caller must have taken a reference to path beforehand. 675b5fb63c1SChristoph Hellwig */ 676b5fb63c1SChristoph Hellwig void nd_jump_link(struct nameidata *nd, struct path *path) 677b5fb63c1SChristoph Hellwig { 678b5fb63c1SChristoph Hellwig path_put(&nd->path); 679b5fb63c1SChristoph Hellwig 680b5fb63c1SChristoph Hellwig nd->path = *path; 681b5fb63c1SChristoph Hellwig nd->inode = nd->path.dentry->d_inode; 682b5fb63c1SChristoph Hellwig nd->flags |= LOOKUP_JUMPED; 683b5fb63c1SChristoph Hellwig } 684b5fb63c1SChristoph Hellwig 685574197e0SAl Viro static inline void put_link(struct nameidata *nd, struct path *link, void *cookie) 686574197e0SAl Viro { 687574197e0SAl Viro struct inode *inode = link->dentry->d_inode; 6886d7b5aaeSAl Viro if (inode->i_op->put_link) 689574197e0SAl Viro inode->i_op->put_link(link->dentry, nd, cookie); 690574197e0SAl Viro path_put(link); 691574197e0SAl Viro } 692574197e0SAl Viro 693561ec64aSLinus Torvalds int sysctl_protected_symlinks __read_mostly = 0; 694561ec64aSLinus Torvalds int sysctl_protected_hardlinks __read_mostly = 0; 695800179c9SKees Cook 696800179c9SKees Cook /** 697800179c9SKees Cook * may_follow_link - Check symlink following for unsafe situations 698800179c9SKees Cook * @link: The path of the symlink 69955852635SRandy Dunlap * @nd: nameidata pathwalk data 700800179c9SKees Cook * 701800179c9SKees Cook * In the case of the sysctl_protected_symlinks sysctl being enabled, 702800179c9SKees Cook * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is 703800179c9SKees Cook * in a sticky world-writable directory. This is to protect privileged 704800179c9SKees Cook * processes from failing races against path names that may change out 705800179c9SKees Cook * from under them by way of other users creating malicious symlinks. 706800179c9SKees Cook * It will permit symlinks to be followed only when outside a sticky 707800179c9SKees Cook * world-writable directory, or when the uid of the symlink and follower 708800179c9SKees Cook * match, or when the directory owner matches the symlink's owner. 709800179c9SKees Cook * 710800179c9SKees Cook * Returns 0 if following the symlink is allowed, -ve on error. 711800179c9SKees Cook */ 712800179c9SKees Cook static inline int may_follow_link(struct path *link, struct nameidata *nd) 713800179c9SKees Cook { 714800179c9SKees Cook const struct inode *inode; 715800179c9SKees Cook const struct inode *parent; 716800179c9SKees Cook 717800179c9SKees Cook if (!sysctl_protected_symlinks) 718800179c9SKees Cook return 0; 719800179c9SKees Cook 720800179c9SKees Cook /* Allowed if owner and follower match. */ 721800179c9SKees Cook inode = link->dentry->d_inode; 72281abe27bSEric W. Biederman if (uid_eq(current_cred()->fsuid, inode->i_uid)) 723800179c9SKees Cook return 0; 724800179c9SKees Cook 725800179c9SKees Cook /* Allowed if parent directory not sticky and world-writable. */ 726800179c9SKees Cook parent = nd->path.dentry->d_inode; 727800179c9SKees Cook if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) 728800179c9SKees Cook return 0; 729800179c9SKees Cook 730800179c9SKees Cook /* Allowed if parent directory and link owner match. */ 73181abe27bSEric W. Biederman if (uid_eq(parent->i_uid, inode->i_uid)) 732800179c9SKees Cook return 0; 733800179c9SKees Cook 734ffd8d101SSasha Levin audit_log_link_denied("follow_link", link); 735800179c9SKees Cook path_put_conditional(link, nd); 736800179c9SKees Cook path_put(&nd->path); 737800179c9SKees Cook return -EACCES; 738800179c9SKees Cook } 739800179c9SKees Cook 740800179c9SKees Cook /** 741800179c9SKees Cook * safe_hardlink_source - Check for safe hardlink conditions 742800179c9SKees Cook * @inode: the source inode to hardlink from 743800179c9SKees Cook * 744800179c9SKees Cook * Return false if at least one of the following conditions: 745800179c9SKees Cook * - inode is not a regular file 746800179c9SKees Cook * - inode is setuid 747800179c9SKees Cook * - inode is setgid and group-exec 748800179c9SKees Cook * - access failure for read and write 749800179c9SKees Cook * 750800179c9SKees Cook * Otherwise returns true. 751800179c9SKees Cook */ 752800179c9SKees Cook static bool safe_hardlink_source(struct inode *inode) 753800179c9SKees Cook { 754800179c9SKees Cook umode_t mode = inode->i_mode; 755800179c9SKees Cook 756800179c9SKees Cook /* Special files should not get pinned to the filesystem. */ 757800179c9SKees Cook if (!S_ISREG(mode)) 758800179c9SKees Cook return false; 759800179c9SKees Cook 760800179c9SKees Cook /* Setuid files should not get pinned to the filesystem. */ 761800179c9SKees Cook if (mode & S_ISUID) 762800179c9SKees Cook return false; 763800179c9SKees Cook 764800179c9SKees Cook /* Executable setgid files should not get pinned to the filesystem. */ 765800179c9SKees Cook if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) 766800179c9SKees Cook return false; 767800179c9SKees Cook 768800179c9SKees Cook /* Hardlinking to unreadable or unwritable sources is dangerous. */ 769800179c9SKees Cook if (inode_permission(inode, MAY_READ | MAY_WRITE)) 770800179c9SKees Cook return false; 771800179c9SKees Cook 772800179c9SKees Cook return true; 773800179c9SKees Cook } 774800179c9SKees Cook 775800179c9SKees Cook /** 776800179c9SKees Cook * may_linkat - Check permissions for creating a hardlink 777800179c9SKees Cook * @link: the source to hardlink from 778800179c9SKees Cook * 779800179c9SKees Cook * Block hardlink when all of: 780800179c9SKees Cook * - sysctl_protected_hardlinks enabled 781800179c9SKees Cook * - fsuid does not match inode 782800179c9SKees Cook * - hardlink source is unsafe (see safe_hardlink_source() above) 783800179c9SKees Cook * - not CAP_FOWNER 784800179c9SKees Cook * 785800179c9SKees Cook * Returns 0 if successful, -ve on error. 786800179c9SKees Cook */ 787800179c9SKees Cook static int may_linkat(struct path *link) 788800179c9SKees Cook { 789800179c9SKees Cook const struct cred *cred; 790800179c9SKees Cook struct inode *inode; 791800179c9SKees Cook 792800179c9SKees Cook if (!sysctl_protected_hardlinks) 793800179c9SKees Cook return 0; 794800179c9SKees Cook 795800179c9SKees Cook cred = current_cred(); 796800179c9SKees Cook inode = link->dentry->d_inode; 797800179c9SKees Cook 798800179c9SKees Cook /* Source inode owner (or CAP_FOWNER) can hardlink all they like, 799800179c9SKees Cook * otherwise, it must be a safe source. 800800179c9SKees Cook */ 80181abe27bSEric W. Biederman if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) || 802800179c9SKees Cook capable(CAP_FOWNER)) 803800179c9SKees Cook return 0; 804800179c9SKees Cook 805a51d9eaaSKees Cook audit_log_link_denied("linkat", link); 806800179c9SKees Cook return -EPERM; 807800179c9SKees Cook } 808800179c9SKees Cook 809def4af30SAl Viro static __always_inline int 810574197e0SAl Viro follow_link(struct path *link, struct nameidata *nd, void **p) 8111da177e4SLinus Torvalds { 8127b9337aaSNick Piggin struct dentry *dentry = link->dentry; 8136d7b5aaeSAl Viro int error; 8146d7b5aaeSAl Viro char *s; 8151da177e4SLinus Torvalds 816844a3917SAl Viro BUG_ON(nd->flags & LOOKUP_RCU); 817844a3917SAl Viro 8180e794589SAl Viro if (link->mnt == nd->path.mnt) 8190e794589SAl Viro mntget(link->mnt); 8200e794589SAl Viro 8216d7b5aaeSAl Viro error = -ELOOP; 8226d7b5aaeSAl Viro if (unlikely(current->total_link_count >= 40)) 8236d7b5aaeSAl Viro goto out_put_nd_path; 8246d7b5aaeSAl Viro 825574197e0SAl Viro cond_resched(); 826574197e0SAl Viro current->total_link_count++; 827574197e0SAl Viro 82868ac1234SAl Viro touch_atime(link); 8291da177e4SLinus Torvalds nd_set_link(nd, NULL); 830cd4e91d3SAl Viro 83136f3b4f6SAl Viro error = security_inode_follow_link(link->dentry, nd); 8326d7b5aaeSAl Viro if (error) 8336d7b5aaeSAl Viro goto out_put_nd_path; 83436f3b4f6SAl Viro 83586acdca1SAl Viro nd->last_type = LAST_BIND; 836def4af30SAl Viro *p = dentry->d_inode->i_op->follow_link(dentry, nd); 837def4af30SAl Viro error = PTR_ERR(*p); 8386d7b5aaeSAl Viro if (IS_ERR(*p)) 839408ef013SChristoph Hellwig goto out_put_nd_path; 8406d7b5aaeSAl Viro 841cc314eefSLinus Torvalds error = 0; 8426d7b5aaeSAl Viro s = nd_get_link(nd); 8436d7b5aaeSAl Viro if (s) { 844443ed254SAl Viro if (unlikely(IS_ERR(s))) { 845443ed254SAl Viro path_put(&nd->path); 846443ed254SAl Viro put_link(nd, link, *p); 847443ed254SAl Viro return PTR_ERR(s); 848443ed254SAl Viro } 849443ed254SAl Viro if (*s == '/') { 850443ed254SAl Viro set_root(nd); 851443ed254SAl Viro path_put(&nd->path); 852443ed254SAl Viro nd->path = nd->root; 853443ed254SAl Viro path_get(&nd->root); 854443ed254SAl Viro nd->flags |= LOOKUP_JUMPED; 855443ed254SAl Viro } 856443ed254SAl Viro nd->inode = nd->path.dentry->d_inode; 857443ed254SAl Viro error = link_path_walk(s, nd); 8586d7b5aaeSAl Viro if (unlikely(error)) 8596d7b5aaeSAl Viro put_link(nd, link, *p); 860b5fb63c1SChristoph Hellwig } 8616d7b5aaeSAl Viro 8626d7b5aaeSAl Viro return error; 8636d7b5aaeSAl Viro 8646d7b5aaeSAl Viro out_put_nd_path: 86598f6ef64SArnd Bergmann *p = NULL; 8666d7b5aaeSAl Viro path_put(&nd->path); 8676d7b5aaeSAl Viro path_put(link); 8681da177e4SLinus Torvalds return error; 8691da177e4SLinus Torvalds } 8701da177e4SLinus Torvalds 87131e6b01fSNick Piggin static int follow_up_rcu(struct path *path) 87231e6b01fSNick Piggin { 8730714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 8740714a533SAl Viro struct mount *parent; 87531e6b01fSNick Piggin struct dentry *mountpoint; 87631e6b01fSNick Piggin 8770714a533SAl Viro parent = mnt->mnt_parent; 8780714a533SAl Viro if (&parent->mnt == path->mnt) 87931e6b01fSNick Piggin return 0; 880a73324daSAl Viro mountpoint = mnt->mnt_mountpoint; 88131e6b01fSNick Piggin path->dentry = mountpoint; 8820714a533SAl Viro path->mnt = &parent->mnt; 88331e6b01fSNick Piggin return 1; 88431e6b01fSNick Piggin } 88531e6b01fSNick Piggin 886f015f126SDavid Howells /* 887f015f126SDavid Howells * follow_up - Find the mountpoint of path's vfsmount 888f015f126SDavid Howells * 889f015f126SDavid Howells * Given a path, find the mountpoint of its source file system. 890f015f126SDavid Howells * Replace @path with the path of the mountpoint in the parent mount. 891f015f126SDavid Howells * Up is towards /. 892f015f126SDavid Howells * 893f015f126SDavid Howells * Return 1 if we went up a level and 0 if we were already at the 894f015f126SDavid Howells * root. 895f015f126SDavid Howells */ 896bab77ebfSAl Viro int follow_up(struct path *path) 8971da177e4SLinus Torvalds { 8980714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 8990714a533SAl Viro struct mount *parent; 9001da177e4SLinus Torvalds struct dentry *mountpoint; 90199b7db7bSNick Piggin 90248a066e7SAl Viro read_seqlock_excl(&mount_lock); 9030714a533SAl Viro parent = mnt->mnt_parent; 9043c0a6163SAl Viro if (parent == mnt) { 90548a066e7SAl Viro read_sequnlock_excl(&mount_lock); 9061da177e4SLinus Torvalds return 0; 9071da177e4SLinus Torvalds } 9080714a533SAl Viro mntget(&parent->mnt); 909a73324daSAl Viro mountpoint = dget(mnt->mnt_mountpoint); 91048a066e7SAl Viro read_sequnlock_excl(&mount_lock); 911bab77ebfSAl Viro dput(path->dentry); 912bab77ebfSAl Viro path->dentry = mountpoint; 913bab77ebfSAl Viro mntput(path->mnt); 9140714a533SAl Viro path->mnt = &parent->mnt; 9151da177e4SLinus Torvalds return 1; 9161da177e4SLinus Torvalds } 9171da177e4SLinus Torvalds 918b5c84bf6SNick Piggin /* 9199875cf80SDavid Howells * Perform an automount 9209875cf80SDavid Howells * - return -EISDIR to tell follow_managed() to stop and return the path we 9219875cf80SDavid Howells * were called with. 9221da177e4SLinus Torvalds */ 9239875cf80SDavid Howells static int follow_automount(struct path *path, unsigned flags, 9249875cf80SDavid Howells bool *need_mntput) 92531e6b01fSNick Piggin { 9269875cf80SDavid Howells struct vfsmount *mnt; 927ea5b778aSDavid Howells int err; 9289875cf80SDavid Howells 9299875cf80SDavid Howells if (!path->dentry->d_op || !path->dentry->d_op->d_automount) 9309875cf80SDavid Howells return -EREMOTE; 9319875cf80SDavid Howells 9320ec26fd0SMiklos Szeredi /* We don't want to mount if someone's just doing a stat - 9330ec26fd0SMiklos Szeredi * unless they're stat'ing a directory and appended a '/' to 9340ec26fd0SMiklos Szeredi * the name. 9350ec26fd0SMiklos Szeredi * 9360ec26fd0SMiklos Szeredi * We do, however, want to mount if someone wants to open or 9375a30d8a2SDavid Howells * create a file of any type under the mountpoint, wants to 9380ec26fd0SMiklos Szeredi * traverse through the mountpoint or wants to open the 9390ec26fd0SMiklos Szeredi * mounted directory. Also, autofs may mark negative dentries 9400ec26fd0SMiklos Szeredi * as being automount points. These will need the attentions 9410ec26fd0SMiklos Szeredi * of the daemon to instantiate them before they can be used. 9425a30d8a2SDavid Howells */ 9435a30d8a2SDavid Howells if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | 944d94c177bSLinus Torvalds LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && 9455a30d8a2SDavid Howells path->dentry->d_inode) 9469875cf80SDavid Howells return -EISDIR; 9470ec26fd0SMiklos Szeredi 9489875cf80SDavid Howells current->total_link_count++; 9499875cf80SDavid Howells if (current->total_link_count >= 40) 9509875cf80SDavid Howells return -ELOOP; 9519875cf80SDavid Howells 9529875cf80SDavid Howells mnt = path->dentry->d_op->d_automount(path); 9539875cf80SDavid Howells if (IS_ERR(mnt)) { 9549875cf80SDavid Howells /* 9559875cf80SDavid Howells * The filesystem is allowed to return -EISDIR here to indicate 9569875cf80SDavid Howells * it doesn't want to automount. For instance, autofs would do 9579875cf80SDavid Howells * this so that its userspace daemon can mount on this dentry. 9589875cf80SDavid Howells * 9599875cf80SDavid Howells * However, we can only permit this if it's a terminal point in 9609875cf80SDavid Howells * the path being looked up; if it wasn't then the remainder of 9619875cf80SDavid Howells * the path is inaccessible and we should say so. 9629875cf80SDavid Howells */ 96349084c3bSAl Viro if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT)) 9649875cf80SDavid Howells return -EREMOTE; 9659875cf80SDavid Howells return PTR_ERR(mnt); 96631e6b01fSNick Piggin } 967ea5b778aSDavid Howells 9689875cf80SDavid Howells if (!mnt) /* mount collision */ 9699875cf80SDavid Howells return 0; 9709875cf80SDavid Howells 9718aef1884SAl Viro if (!*need_mntput) { 9728aef1884SAl Viro /* lock_mount() may release path->mnt on error */ 9738aef1884SAl Viro mntget(path->mnt); 9748aef1884SAl Viro *need_mntput = true; 9758aef1884SAl Viro } 97619a167afSAl Viro err = finish_automount(mnt, path); 977ea5b778aSDavid Howells 978ea5b778aSDavid Howells switch (err) { 979ea5b778aSDavid Howells case -EBUSY: 980ea5b778aSDavid Howells /* Someone else made a mount here whilst we were busy */ 98119a167afSAl Viro return 0; 982ea5b778aSDavid Howells case 0: 9838aef1884SAl Viro path_put(path); 9849875cf80SDavid Howells path->mnt = mnt; 9859875cf80SDavid Howells path->dentry = dget(mnt->mnt_root); 9869875cf80SDavid Howells return 0; 98719a167afSAl Viro default: 98819a167afSAl Viro return err; 9899875cf80SDavid Howells } 99019a167afSAl Viro 991ea5b778aSDavid Howells } 9929875cf80SDavid Howells 9939875cf80SDavid Howells /* 9949875cf80SDavid Howells * Handle a dentry that is managed in some way. 995cc53ce53SDavid Howells * - Flagged for transit management (autofs) 9969875cf80SDavid Howells * - Flagged as mountpoint 9979875cf80SDavid Howells * - Flagged as automount point 9989875cf80SDavid Howells * 9999875cf80SDavid Howells * This may only be called in refwalk mode. 10009875cf80SDavid Howells * 10019875cf80SDavid Howells * Serialization is taken care of in namespace.c 10029875cf80SDavid Howells */ 10039875cf80SDavid Howells static int follow_managed(struct path *path, unsigned flags) 10049875cf80SDavid Howells { 10058aef1884SAl Viro struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */ 10069875cf80SDavid Howells unsigned managed; 10079875cf80SDavid Howells bool need_mntput = false; 10088aef1884SAl Viro int ret = 0; 10099875cf80SDavid Howells 10109875cf80SDavid Howells /* Given that we're not holding a lock here, we retain the value in a 10119875cf80SDavid Howells * local variable for each dentry as we look at it so that we don't see 10129875cf80SDavid Howells * the components of that value change under us */ 10139875cf80SDavid Howells while (managed = ACCESS_ONCE(path->dentry->d_flags), 10149875cf80SDavid Howells managed &= DCACHE_MANAGED_DENTRY, 10159875cf80SDavid Howells unlikely(managed != 0)) { 1016cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1017cc53ce53SDavid Howells * being held. */ 1018cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 1019cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1020cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 10211aed3e42SAl Viro ret = path->dentry->d_op->d_manage(path->dentry, false); 1022cc53ce53SDavid Howells if (ret < 0) 10238aef1884SAl Viro break; 1024cc53ce53SDavid Howells } 1025cc53ce53SDavid Howells 10269875cf80SDavid Howells /* Transit to a mounted filesystem. */ 10279875cf80SDavid Howells if (managed & DCACHE_MOUNTED) { 10289875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 10299875cf80SDavid Howells if (mounted) { 10309875cf80SDavid Howells dput(path->dentry); 10319875cf80SDavid Howells if (need_mntput) 1032463ffb2eSAl Viro mntput(path->mnt); 1033463ffb2eSAl Viro path->mnt = mounted; 1034463ffb2eSAl Viro path->dentry = dget(mounted->mnt_root); 10359875cf80SDavid Howells need_mntput = true; 10369875cf80SDavid Howells continue; 1037463ffb2eSAl Viro } 1038463ffb2eSAl Viro 10399875cf80SDavid Howells /* Something is mounted on this dentry in another 10409875cf80SDavid Howells * namespace and/or whatever was mounted there in this 104148a066e7SAl Viro * namespace got unmounted before lookup_mnt() could 104248a066e7SAl Viro * get it */ 10431da177e4SLinus Torvalds } 10449875cf80SDavid Howells 10459875cf80SDavid Howells /* Handle an automount point */ 10469875cf80SDavid Howells if (managed & DCACHE_NEED_AUTOMOUNT) { 10479875cf80SDavid Howells ret = follow_automount(path, flags, &need_mntput); 10489875cf80SDavid Howells if (ret < 0) 10498aef1884SAl Viro break; 10509875cf80SDavid Howells continue; 10519875cf80SDavid Howells } 10529875cf80SDavid Howells 10539875cf80SDavid Howells /* We didn't change the current path point */ 10549875cf80SDavid Howells break; 10559875cf80SDavid Howells } 10568aef1884SAl Viro 10578aef1884SAl Viro if (need_mntput && path->mnt == mnt) 10588aef1884SAl Viro mntput(path->mnt); 10598aef1884SAl Viro if (ret == -EISDIR) 10608aef1884SAl Viro ret = 0; 1061a3fbbde7SAl Viro return ret < 0 ? ret : need_mntput; 10621da177e4SLinus Torvalds } 10631da177e4SLinus Torvalds 1064cc53ce53SDavid Howells int follow_down_one(struct path *path) 10651da177e4SLinus Torvalds { 10661da177e4SLinus Torvalds struct vfsmount *mounted; 10671da177e4SLinus Torvalds 10681c755af4SAl Viro mounted = lookup_mnt(path); 10691da177e4SLinus Torvalds if (mounted) { 10709393bd07SAl Viro dput(path->dentry); 10719393bd07SAl Viro mntput(path->mnt); 10729393bd07SAl Viro path->mnt = mounted; 10739393bd07SAl Viro path->dentry = dget(mounted->mnt_root); 10741da177e4SLinus Torvalds return 1; 10751da177e4SLinus Torvalds } 10761da177e4SLinus Torvalds return 0; 10771da177e4SLinus Torvalds } 10781da177e4SLinus Torvalds 107962a7375eSIan Kent static inline bool managed_dentry_might_block(struct dentry *dentry) 108062a7375eSIan Kent { 108162a7375eSIan Kent return (dentry->d_flags & DCACHE_MANAGE_TRANSIT && 108262a7375eSIan Kent dentry->d_op->d_manage(dentry, true) < 0); 108362a7375eSIan Kent } 108462a7375eSIan Kent 10859875cf80SDavid Howells /* 1086287548e4SAl Viro * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if 1087287548e4SAl Viro * we meet a managed dentry that would need blocking. 10889875cf80SDavid Howells */ 10899875cf80SDavid Howells static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, 1090287548e4SAl Viro struct inode **inode) 10919875cf80SDavid Howells { 109262a7375eSIan Kent for (;;) { 1093c7105365SAl Viro struct mount *mounted; 109462a7375eSIan Kent /* 109562a7375eSIan Kent * Don't forget we might have a non-mountpoint managed dentry 109662a7375eSIan Kent * that wants to block transit. 109762a7375eSIan Kent */ 1098287548e4SAl Viro if (unlikely(managed_dentry_might_block(path->dentry))) 1099ab90911fSDavid Howells return false; 110062a7375eSIan Kent 110162a7375eSIan Kent if (!d_mountpoint(path->dentry)) 110262a7375eSIan Kent break; 110362a7375eSIan Kent 1104474279dcSAl Viro mounted = __lookup_mnt(path->mnt, path->dentry); 11059875cf80SDavid Howells if (!mounted) 11069875cf80SDavid Howells break; 1107c7105365SAl Viro path->mnt = &mounted->mnt; 1108c7105365SAl Viro path->dentry = mounted->mnt.mnt_root; 1109a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 11109875cf80SDavid Howells nd->seq = read_seqcount_begin(&path->dentry->d_seq); 111159430262SLinus Torvalds /* 111259430262SLinus Torvalds * Update the inode too. We don't need to re-check the 111359430262SLinus Torvalds * dentry sequence number here after this d_inode read, 111459430262SLinus Torvalds * because a mount-point is always pinned. 111559430262SLinus Torvalds */ 111659430262SLinus Torvalds *inode = path->dentry->d_inode; 11179875cf80SDavid Howells } 11189875cf80SDavid Howells return true; 11199875cf80SDavid Howells } 11209875cf80SDavid Howells 1121dea39376SAl Viro static void follow_mount_rcu(struct nameidata *nd) 1122287548e4SAl Viro { 1123dea39376SAl Viro while (d_mountpoint(nd->path.dentry)) { 1124c7105365SAl Viro struct mount *mounted; 1125474279dcSAl Viro mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry); 1126287548e4SAl Viro if (!mounted) 1127287548e4SAl Viro break; 1128c7105365SAl Viro nd->path.mnt = &mounted->mnt; 1129c7105365SAl Viro nd->path.dentry = mounted->mnt.mnt_root; 1130dea39376SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 1131287548e4SAl Viro } 1132287548e4SAl Viro } 1133287548e4SAl Viro 113431e6b01fSNick Piggin static int follow_dotdot_rcu(struct nameidata *nd) 113531e6b01fSNick Piggin { 113631e6b01fSNick Piggin set_root_rcu(nd); 113731e6b01fSNick Piggin 113831e6b01fSNick Piggin while (1) { 113931e6b01fSNick Piggin if (nd->path.dentry == nd->root.dentry && 114031e6b01fSNick Piggin nd->path.mnt == nd->root.mnt) { 114131e6b01fSNick Piggin break; 114231e6b01fSNick Piggin } 114331e6b01fSNick Piggin if (nd->path.dentry != nd->path.mnt->mnt_root) { 114431e6b01fSNick Piggin struct dentry *old = nd->path.dentry; 114531e6b01fSNick Piggin struct dentry *parent = old->d_parent; 114631e6b01fSNick Piggin unsigned seq; 114731e6b01fSNick Piggin 114831e6b01fSNick Piggin seq = read_seqcount_begin(&parent->d_seq); 114931e6b01fSNick Piggin if (read_seqcount_retry(&old->d_seq, nd->seq)) 1150ef7562d5SAl Viro goto failed; 115131e6b01fSNick Piggin nd->path.dentry = parent; 115231e6b01fSNick Piggin nd->seq = seq; 115331e6b01fSNick Piggin break; 115431e6b01fSNick Piggin } 115531e6b01fSNick Piggin if (!follow_up_rcu(&nd->path)) 115631e6b01fSNick Piggin break; 115731e6b01fSNick Piggin nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 115831e6b01fSNick Piggin } 1159dea39376SAl Viro follow_mount_rcu(nd); 1160dea39376SAl Viro nd->inode = nd->path.dentry->d_inode; 116131e6b01fSNick Piggin return 0; 1162ef7562d5SAl Viro 1163ef7562d5SAl Viro failed: 1164ef7562d5SAl Viro nd->flags &= ~LOOKUP_RCU; 11655b6ca027SAl Viro if (!(nd->flags & LOOKUP_ROOT)) 1166ef7562d5SAl Viro nd->root.mnt = NULL; 11678b61e74fSAl Viro rcu_read_unlock(); 1168ef7562d5SAl Viro return -ECHILD; 116931e6b01fSNick Piggin } 117031e6b01fSNick Piggin 11719875cf80SDavid Howells /* 1172cc53ce53SDavid Howells * Follow down to the covering mount currently visible to userspace. At each 1173cc53ce53SDavid Howells * point, the filesystem owning that dentry may be queried as to whether the 1174cc53ce53SDavid Howells * caller is permitted to proceed or not. 1175cc53ce53SDavid Howells */ 11767cc90cc3SAl Viro int follow_down(struct path *path) 1177cc53ce53SDavid Howells { 1178cc53ce53SDavid Howells unsigned managed; 1179cc53ce53SDavid Howells int ret; 1180cc53ce53SDavid Howells 1181cc53ce53SDavid Howells while (managed = ACCESS_ONCE(path->dentry->d_flags), 1182cc53ce53SDavid Howells unlikely(managed & DCACHE_MANAGED_DENTRY)) { 1183cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1184cc53ce53SDavid Howells * being held. 1185cc53ce53SDavid Howells * 1186cc53ce53SDavid Howells * We indicate to the filesystem if someone is trying to mount 1187cc53ce53SDavid Howells * something here. This gives autofs the chance to deny anyone 1188cc53ce53SDavid Howells * other than its daemon the right to mount on its 1189cc53ce53SDavid Howells * superstructure. 1190cc53ce53SDavid Howells * 1191cc53ce53SDavid Howells * The filesystem may sleep at this point. 1192cc53ce53SDavid Howells */ 1193cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 1194cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1195cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 1196ab90911fSDavid Howells ret = path->dentry->d_op->d_manage( 11971aed3e42SAl Viro path->dentry, false); 1198cc53ce53SDavid Howells if (ret < 0) 1199cc53ce53SDavid Howells return ret == -EISDIR ? 0 : ret; 1200cc53ce53SDavid Howells } 1201cc53ce53SDavid Howells 1202cc53ce53SDavid Howells /* Transit to a mounted filesystem. */ 1203cc53ce53SDavid Howells if (managed & DCACHE_MOUNTED) { 1204cc53ce53SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 1205cc53ce53SDavid Howells if (!mounted) 1206cc53ce53SDavid Howells break; 1207cc53ce53SDavid Howells dput(path->dentry); 1208cc53ce53SDavid Howells mntput(path->mnt); 1209cc53ce53SDavid Howells path->mnt = mounted; 1210cc53ce53SDavid Howells path->dentry = dget(mounted->mnt_root); 1211cc53ce53SDavid Howells continue; 1212cc53ce53SDavid Howells } 1213cc53ce53SDavid Howells 1214cc53ce53SDavid Howells /* Don't handle automount points here */ 1215cc53ce53SDavid Howells break; 1216cc53ce53SDavid Howells } 1217cc53ce53SDavid Howells return 0; 1218cc53ce53SDavid Howells } 1219cc53ce53SDavid Howells 1220cc53ce53SDavid Howells /* 12219875cf80SDavid Howells * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() 12229875cf80SDavid Howells */ 12239875cf80SDavid Howells static void follow_mount(struct path *path) 12249875cf80SDavid Howells { 12259875cf80SDavid Howells while (d_mountpoint(path->dentry)) { 12269875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 12279875cf80SDavid Howells if (!mounted) 12289875cf80SDavid Howells break; 12299875cf80SDavid Howells dput(path->dentry); 12309875cf80SDavid Howells mntput(path->mnt); 12319875cf80SDavid Howells path->mnt = mounted; 12329875cf80SDavid Howells path->dentry = dget(mounted->mnt_root); 12339875cf80SDavid Howells } 12349875cf80SDavid Howells } 12359875cf80SDavid Howells 123631e6b01fSNick Piggin static void follow_dotdot(struct nameidata *nd) 12371da177e4SLinus Torvalds { 12382a737871SAl Viro set_root(nd); 1239e518ddb7SAndreas Mohr 12401da177e4SLinus Torvalds while(1) { 12414ac91378SJan Blunck struct dentry *old = nd->path.dentry; 12421da177e4SLinus Torvalds 12432a737871SAl Viro if (nd->path.dentry == nd->root.dentry && 12442a737871SAl Viro nd->path.mnt == nd->root.mnt) { 12451da177e4SLinus Torvalds break; 12461da177e4SLinus Torvalds } 12474ac91378SJan Blunck if (nd->path.dentry != nd->path.mnt->mnt_root) { 12483088dd70SAl Viro /* rare case of legitimate dget_parent()... */ 12493088dd70SAl Viro nd->path.dentry = dget_parent(nd->path.dentry); 12501da177e4SLinus Torvalds dput(old); 12511da177e4SLinus Torvalds break; 12521da177e4SLinus Torvalds } 12533088dd70SAl Viro if (!follow_up(&nd->path)) 12541da177e4SLinus Torvalds break; 12551da177e4SLinus Torvalds } 125679ed0226SAl Viro follow_mount(&nd->path); 125731e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 12581da177e4SLinus Torvalds } 12591da177e4SLinus Torvalds 12601da177e4SLinus Torvalds /* 1261bad61189SMiklos Szeredi * This looks up the name in dcache, possibly revalidates the old dentry and 1262bad61189SMiklos Szeredi * allocates a new one if not found or not valid. In the need_lookup argument 1263bad61189SMiklos Szeredi * returns whether i_op->lookup is necessary. 1264bad61189SMiklos Szeredi * 1265bad61189SMiklos Szeredi * dir->d_inode->i_mutex must be held 1266baa03890SNick Piggin */ 1267bad61189SMiklos Szeredi static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir, 1268201f956eSAl Viro unsigned int flags, bool *need_lookup) 1269baa03890SNick Piggin { 1270baa03890SNick Piggin struct dentry *dentry; 1271bad61189SMiklos Szeredi int error; 1272baa03890SNick Piggin 1273bad61189SMiklos Szeredi *need_lookup = false; 1274bad61189SMiklos Szeredi dentry = d_lookup(dir, name); 1275bad61189SMiklos Szeredi if (dentry) { 127639e3c955SJeff Layton if (dentry->d_flags & DCACHE_OP_REVALIDATE) { 1277201f956eSAl Viro error = d_revalidate(dentry, flags); 1278bad61189SMiklos Szeredi if (unlikely(error <= 0)) { 1279bad61189SMiklos Szeredi if (error < 0) { 1280bad61189SMiklos Szeredi dput(dentry); 1281bad61189SMiklos Szeredi return ERR_PTR(error); 1282bad61189SMiklos Szeredi } else if (!d_invalidate(dentry)) { 1283bad61189SMiklos Szeredi dput(dentry); 1284bad61189SMiklos Szeredi dentry = NULL; 1285bad61189SMiklos Szeredi } 1286bad61189SMiklos Szeredi } 1287bad61189SMiklos Szeredi } 1288bad61189SMiklos Szeredi } 1289baa03890SNick Piggin 1290bad61189SMiklos Szeredi if (!dentry) { 1291bad61189SMiklos Szeredi dentry = d_alloc(dir, name); 1292baa03890SNick Piggin if (unlikely(!dentry)) 1293baa03890SNick Piggin return ERR_PTR(-ENOMEM); 1294baa03890SNick Piggin 1295bad61189SMiklos Szeredi *need_lookup = true; 1296baa03890SNick Piggin } 1297baa03890SNick Piggin return dentry; 1298baa03890SNick Piggin } 1299baa03890SNick Piggin 1300baa03890SNick Piggin /* 1301bad61189SMiklos Szeredi * Call i_op->lookup on the dentry. The dentry must be negative but may be 1302bad61189SMiklos Szeredi * hashed if it was pouplated with DCACHE_NEED_LOOKUP. 1303bad61189SMiklos Szeredi * 1304bad61189SMiklos Szeredi * dir->d_inode->i_mutex must be held 130544396f4bSJosef Bacik */ 1306bad61189SMiklos Szeredi static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry, 130772bd866aSAl Viro unsigned int flags) 130844396f4bSJosef Bacik { 130944396f4bSJosef Bacik struct dentry *old; 131044396f4bSJosef Bacik 131144396f4bSJosef Bacik /* Don't create child dentry for a dead directory. */ 1312bad61189SMiklos Szeredi if (unlikely(IS_DEADDIR(dir))) { 1313e188dc02SMiklos Szeredi dput(dentry); 131444396f4bSJosef Bacik return ERR_PTR(-ENOENT); 1315e188dc02SMiklos Szeredi } 131644396f4bSJosef Bacik 131772bd866aSAl Viro old = dir->i_op->lookup(dir, dentry, flags); 131844396f4bSJosef Bacik if (unlikely(old)) { 131944396f4bSJosef Bacik dput(dentry); 132044396f4bSJosef Bacik dentry = old; 132144396f4bSJosef Bacik } 132244396f4bSJosef Bacik return dentry; 132344396f4bSJosef Bacik } 132444396f4bSJosef Bacik 1325a3255546SAl Viro static struct dentry *__lookup_hash(struct qstr *name, 132672bd866aSAl Viro struct dentry *base, unsigned int flags) 1327a3255546SAl Viro { 1328bad61189SMiklos Szeredi bool need_lookup; 1329a3255546SAl Viro struct dentry *dentry; 1330a3255546SAl Viro 133172bd866aSAl Viro dentry = lookup_dcache(name, base, flags, &need_lookup); 1332bad61189SMiklos Szeredi if (!need_lookup) 1333a3255546SAl Viro return dentry; 1334bad61189SMiklos Szeredi 133572bd866aSAl Viro return lookup_real(base->d_inode, dentry, flags); 1336a3255546SAl Viro } 1337a3255546SAl Viro 133844396f4bSJosef Bacik /* 13391da177e4SLinus Torvalds * It's more convoluted than I'd like it to be, but... it's still fairly 13401da177e4SLinus Torvalds * small and for now I'd prefer to have fast path as straight as possible. 13411da177e4SLinus Torvalds * It _is_ time-critical. 13421da177e4SLinus Torvalds */ 1343e97cdc87SAl Viro static int lookup_fast(struct nameidata *nd, 134431e6b01fSNick Piggin struct path *path, struct inode **inode) 13451da177e4SLinus Torvalds { 13464ac91378SJan Blunck struct vfsmount *mnt = nd->path.mnt; 134731e6b01fSNick Piggin struct dentry *dentry, *parent = nd->path.dentry; 13485a18fff2SAl Viro int need_reval = 1; 13495a18fff2SAl Viro int status = 1; 13509875cf80SDavid Howells int err; 13519875cf80SDavid Howells 13523cac260aSAl Viro /* 1353b04f784eSNick Piggin * Rename seqlock is not required here because in the off chance 1354b04f784eSNick Piggin * of a false negative due to a concurrent rename, we're going to 1355b04f784eSNick Piggin * do the non-racy lookup, below. 1356b04f784eSNick Piggin */ 135731e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) { 135831e6b01fSNick Piggin unsigned seq; 1359da53be12SLinus Torvalds dentry = __d_lookup_rcu(parent, &nd->last, &seq); 13605a18fff2SAl Viro if (!dentry) 13615a18fff2SAl Viro goto unlazy; 13625a18fff2SAl Viro 136312f8ad4bSLinus Torvalds /* 136412f8ad4bSLinus Torvalds * This sequence count validates that the inode matches 136512f8ad4bSLinus Torvalds * the dentry name information from lookup. 136612f8ad4bSLinus Torvalds */ 136712f8ad4bSLinus Torvalds *inode = dentry->d_inode; 136812f8ad4bSLinus Torvalds if (read_seqcount_retry(&dentry->d_seq, seq)) 136912f8ad4bSLinus Torvalds return -ECHILD; 137012f8ad4bSLinus Torvalds 137112f8ad4bSLinus Torvalds /* 137212f8ad4bSLinus Torvalds * This sequence count validates that the parent had no 137312f8ad4bSLinus Torvalds * changes while we did the lookup of the dentry above. 137412f8ad4bSLinus Torvalds * 137512f8ad4bSLinus Torvalds * The memory barrier in read_seqcount_begin of child is 137612f8ad4bSLinus Torvalds * enough, we can use __read_seqcount_retry here. 137712f8ad4bSLinus Torvalds */ 137831e6b01fSNick Piggin if (__read_seqcount_retry(&parent->d_seq, nd->seq)) 137931e6b01fSNick Piggin return -ECHILD; 138031e6b01fSNick Piggin nd->seq = seq; 13815a18fff2SAl Viro 138224643087SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) { 13834ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 13845a18fff2SAl Viro if (unlikely(status <= 0)) { 13855a18fff2SAl Viro if (status != -ECHILD) 13865a18fff2SAl Viro need_reval = 0; 13875a18fff2SAl Viro goto unlazy; 13885a18fff2SAl Viro } 138924643087SAl Viro } 139031e6b01fSNick Piggin path->mnt = mnt; 139131e6b01fSNick Piggin path->dentry = dentry; 1392d6e9bd25SAl Viro if (unlikely(!__follow_mount_rcu(nd, path, inode))) 1393d6e9bd25SAl Viro goto unlazy; 1394d6e9bd25SAl Viro if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT)) 1395d6e9bd25SAl Viro goto unlazy; 13969875cf80SDavid Howells return 0; 13975a18fff2SAl Viro unlazy: 139819660af7SAl Viro if (unlazy_walk(nd, dentry)) 13995a18fff2SAl Viro return -ECHILD; 14005a18fff2SAl Viro } else { 1401e97cdc87SAl Viro dentry = __d_lookup(parent, &nd->last); 140224643087SAl Viro } 14035a18fff2SAl Viro 140481e6f520SAl Viro if (unlikely(!dentry)) 140581e6f520SAl Viro goto need_lookup; 14065a18fff2SAl Viro 14075a18fff2SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval) 14084ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 14095a18fff2SAl Viro if (unlikely(status <= 0)) { 14105a18fff2SAl Viro if (status < 0) { 14115a18fff2SAl Viro dput(dentry); 14125a18fff2SAl Viro return status; 14135a18fff2SAl Viro } 14145a18fff2SAl Viro if (!d_invalidate(dentry)) { 14155a18fff2SAl Viro dput(dentry); 141681e6f520SAl Viro goto need_lookup; 14175a18fff2SAl Viro } 14185a18fff2SAl Viro } 1419697f514dSMiklos Szeredi 14201da177e4SLinus Torvalds path->mnt = mnt; 14211da177e4SLinus Torvalds path->dentry = dentry; 14229875cf80SDavid Howells err = follow_managed(path, nd->flags); 142389312214SIan Kent if (unlikely(err < 0)) { 142489312214SIan Kent path_put_conditional(path, nd); 14259875cf80SDavid Howells return err; 142689312214SIan Kent } 1427a3fbbde7SAl Viro if (err) 1428a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 142931e6b01fSNick Piggin *inode = path->dentry->d_inode; 14301da177e4SLinus Torvalds return 0; 143181e6f520SAl Viro 143281e6f520SAl Viro need_lookup: 1433697f514dSMiklos Szeredi return 1; 1434697f514dSMiklos Szeredi } 1435697f514dSMiklos Szeredi 1436697f514dSMiklos Szeredi /* Fast lookup failed, do it the slow way */ 1437cc2a5271SAl Viro static int lookup_slow(struct nameidata *nd, struct path *path) 1438697f514dSMiklos Szeredi { 1439697f514dSMiklos Szeredi struct dentry *dentry, *parent; 1440697f514dSMiklos Szeredi int err; 1441697f514dSMiklos Szeredi 1442697f514dSMiklos Szeredi parent = nd->path.dentry; 144381e6f520SAl Viro BUG_ON(nd->inode != parent->d_inode); 144481e6f520SAl Viro 144581e6f520SAl Viro mutex_lock(&parent->d_inode->i_mutex); 1446cc2a5271SAl Viro dentry = __lookup_hash(&nd->last, parent, nd->flags); 144781e6f520SAl Viro mutex_unlock(&parent->d_inode->i_mutex); 144881e6f520SAl Viro if (IS_ERR(dentry)) 144981e6f520SAl Viro return PTR_ERR(dentry); 1450697f514dSMiklos Szeredi path->mnt = nd->path.mnt; 1451697f514dSMiklos Szeredi path->dentry = dentry; 1452697f514dSMiklos Szeredi err = follow_managed(path, nd->flags); 1453697f514dSMiklos Szeredi if (unlikely(err < 0)) { 1454697f514dSMiklos Szeredi path_put_conditional(path, nd); 1455697f514dSMiklos Szeredi return err; 1456697f514dSMiklos Szeredi } 1457697f514dSMiklos Szeredi if (err) 1458697f514dSMiklos Szeredi nd->flags |= LOOKUP_JUMPED; 1459697f514dSMiklos Szeredi return 0; 14601da177e4SLinus Torvalds } 14611da177e4SLinus Torvalds 146252094c8aSAl Viro static inline int may_lookup(struct nameidata *nd) 146352094c8aSAl Viro { 146452094c8aSAl Viro if (nd->flags & LOOKUP_RCU) { 14654ad5abb3SAl Viro int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); 146652094c8aSAl Viro if (err != -ECHILD) 146752094c8aSAl Viro return err; 146819660af7SAl Viro if (unlazy_walk(nd, NULL)) 146952094c8aSAl Viro return -ECHILD; 147052094c8aSAl Viro } 14714ad5abb3SAl Viro return inode_permission(nd->inode, MAY_EXEC); 147252094c8aSAl Viro } 147352094c8aSAl Viro 14749856fa1bSAl Viro static inline int handle_dots(struct nameidata *nd, int type) 14759856fa1bSAl Viro { 14769856fa1bSAl Viro if (type == LAST_DOTDOT) { 14779856fa1bSAl Viro if (nd->flags & LOOKUP_RCU) { 14789856fa1bSAl Viro if (follow_dotdot_rcu(nd)) 14799856fa1bSAl Viro return -ECHILD; 14809856fa1bSAl Viro } else 14819856fa1bSAl Viro follow_dotdot(nd); 14829856fa1bSAl Viro } 14839856fa1bSAl Viro return 0; 14849856fa1bSAl Viro } 14859856fa1bSAl Viro 1486951361f9SAl Viro static void terminate_walk(struct nameidata *nd) 1487951361f9SAl Viro { 1488951361f9SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1489951361f9SAl Viro path_put(&nd->path); 1490951361f9SAl Viro } else { 1491951361f9SAl Viro nd->flags &= ~LOOKUP_RCU; 14925b6ca027SAl Viro if (!(nd->flags & LOOKUP_ROOT)) 1493951361f9SAl Viro nd->root.mnt = NULL; 14948b61e74fSAl Viro rcu_read_unlock(); 1495951361f9SAl Viro } 1496951361f9SAl Viro } 1497951361f9SAl Viro 14983ddcd056SLinus Torvalds /* 14993ddcd056SLinus Torvalds * Do we need to follow links? We _really_ want to be able 15003ddcd056SLinus Torvalds * to do this check without having to look at inode->i_op, 15013ddcd056SLinus Torvalds * so we keep a cache of "no, this doesn't need follow_link" 15023ddcd056SLinus Torvalds * for the common case. 15033ddcd056SLinus Torvalds */ 1504b18825a7SDavid Howells static inline int should_follow_link(struct dentry *dentry, int follow) 15053ddcd056SLinus Torvalds { 1506b18825a7SDavid Howells return unlikely(d_is_symlink(dentry)) ? follow : 0; 15073ddcd056SLinus Torvalds } 15083ddcd056SLinus Torvalds 1509ce57dfc1SAl Viro static inline int walk_component(struct nameidata *nd, struct path *path, 151021b9b073SAl Viro int follow) 1511ce57dfc1SAl Viro { 1512ce57dfc1SAl Viro struct inode *inode; 1513ce57dfc1SAl Viro int err; 1514ce57dfc1SAl Viro /* 1515ce57dfc1SAl Viro * "." and ".." are special - ".." especially so because it has 1516ce57dfc1SAl Viro * to be able to know about the current root directory and 1517ce57dfc1SAl Viro * parent relationships. 1518ce57dfc1SAl Viro */ 151921b9b073SAl Viro if (unlikely(nd->last_type != LAST_NORM)) 152021b9b073SAl Viro return handle_dots(nd, nd->last_type); 1521e97cdc87SAl Viro err = lookup_fast(nd, path, &inode); 1522ce57dfc1SAl Viro if (unlikely(err)) { 1523697f514dSMiklos Szeredi if (err < 0) 1524697f514dSMiklos Szeredi goto out_err; 1525697f514dSMiklos Szeredi 1526cc2a5271SAl Viro err = lookup_slow(nd, path); 1527697f514dSMiklos Szeredi if (err < 0) 1528697f514dSMiklos Szeredi goto out_err; 1529697f514dSMiklos Szeredi 1530697f514dSMiklos Szeredi inode = path->dentry->d_inode; 1531ce57dfc1SAl Viro } 1532697f514dSMiklos Szeredi err = -ENOENT; 1533697f514dSMiklos Szeredi if (!inode) 1534697f514dSMiklos Szeredi goto out_path_put; 1535697f514dSMiklos Szeredi 1536b18825a7SDavid Howells if (should_follow_link(path->dentry, follow)) { 153719660af7SAl Viro if (nd->flags & LOOKUP_RCU) { 153819660af7SAl Viro if (unlikely(unlazy_walk(nd, path->dentry))) { 1539697f514dSMiklos Szeredi err = -ECHILD; 1540697f514dSMiklos Szeredi goto out_err; 154119660af7SAl Viro } 154219660af7SAl Viro } 1543ce57dfc1SAl Viro BUG_ON(inode != path->dentry->d_inode); 1544ce57dfc1SAl Viro return 1; 1545ce57dfc1SAl Viro } 1546ce57dfc1SAl Viro path_to_nameidata(path, nd); 1547ce57dfc1SAl Viro nd->inode = inode; 1548ce57dfc1SAl Viro return 0; 1549697f514dSMiklos Szeredi 1550697f514dSMiklos Szeredi out_path_put: 1551697f514dSMiklos Szeredi path_to_nameidata(path, nd); 1552697f514dSMiklos Szeredi out_err: 1553697f514dSMiklos Szeredi terminate_walk(nd); 1554697f514dSMiklos Szeredi return err; 1555ce57dfc1SAl Viro } 1556ce57dfc1SAl Viro 15571da177e4SLinus Torvalds /* 1558b356379aSAl Viro * This limits recursive symlink follows to 8, while 1559b356379aSAl Viro * limiting consecutive symlinks to 40. 1560b356379aSAl Viro * 1561b356379aSAl Viro * Without that kind of total limit, nasty chains of consecutive 1562b356379aSAl Viro * symlinks can cause almost arbitrarily long lookups. 1563b356379aSAl Viro */ 1564b356379aSAl Viro static inline int nested_symlink(struct path *path, struct nameidata *nd) 1565b356379aSAl Viro { 1566b356379aSAl Viro int res; 1567b356379aSAl Viro 1568b356379aSAl Viro if (unlikely(current->link_count >= MAX_NESTED_LINKS)) { 1569b356379aSAl Viro path_put_conditional(path, nd); 1570b356379aSAl Viro path_put(&nd->path); 1571b356379aSAl Viro return -ELOOP; 1572b356379aSAl Viro } 15731a4022f8SErez Zadok BUG_ON(nd->depth >= MAX_NESTED_LINKS); 1574b356379aSAl Viro 1575b356379aSAl Viro nd->depth++; 1576b356379aSAl Viro current->link_count++; 1577b356379aSAl Viro 1578b356379aSAl Viro do { 1579b356379aSAl Viro struct path link = *path; 1580b356379aSAl Viro void *cookie; 1581574197e0SAl Viro 1582574197e0SAl Viro res = follow_link(&link, nd, &cookie); 15836d7b5aaeSAl Viro if (res) 15846d7b5aaeSAl Viro break; 158521b9b073SAl Viro res = walk_component(nd, path, LOOKUP_FOLLOW); 1586574197e0SAl Viro put_link(nd, &link, cookie); 1587b356379aSAl Viro } while (res > 0); 1588b356379aSAl Viro 1589b356379aSAl Viro current->link_count--; 1590b356379aSAl Viro nd->depth--; 1591b356379aSAl Viro return res; 1592b356379aSAl Viro } 1593b356379aSAl Viro 1594b356379aSAl Viro /* 1595bfcfaa77SLinus Torvalds * We can do the critical dentry name comparison and hashing 1596bfcfaa77SLinus Torvalds * operations one word at a time, but we are limited to: 1597bfcfaa77SLinus Torvalds * 1598bfcfaa77SLinus Torvalds * - Architectures with fast unaligned word accesses. We could 1599bfcfaa77SLinus Torvalds * do a "get_unaligned()" if this helps and is sufficiently 1600bfcfaa77SLinus Torvalds * fast. 1601bfcfaa77SLinus Torvalds * 1602bfcfaa77SLinus Torvalds * - Little-endian machines (so that we can generate the mask 1603bfcfaa77SLinus Torvalds * of low bytes efficiently). Again, we *could* do a byte 1604bfcfaa77SLinus Torvalds * swapping load on big-endian architectures if that is not 1605bfcfaa77SLinus Torvalds * expensive enough to make the optimization worthless. 1606bfcfaa77SLinus Torvalds * 1607bfcfaa77SLinus Torvalds * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we 1608bfcfaa77SLinus Torvalds * do not trap on the (extremely unlikely) case of a page 1609bfcfaa77SLinus Torvalds * crossing operation. 1610bfcfaa77SLinus Torvalds * 1611bfcfaa77SLinus Torvalds * - Furthermore, we need an efficient 64-bit compile for the 1612bfcfaa77SLinus Torvalds * 64-bit case in order to generate the "number of bytes in 1613bfcfaa77SLinus Torvalds * the final mask". Again, that could be replaced with a 1614bfcfaa77SLinus Torvalds * efficient population count instruction or similar. 1615bfcfaa77SLinus Torvalds */ 1616bfcfaa77SLinus Torvalds #ifdef CONFIG_DCACHE_WORD_ACCESS 1617bfcfaa77SLinus Torvalds 1618f68e556eSLinus Torvalds #include <asm/word-at-a-time.h> 1619bfcfaa77SLinus Torvalds 1620f68e556eSLinus Torvalds #ifdef CONFIG_64BIT 1621bfcfaa77SLinus Torvalds 1622bfcfaa77SLinus Torvalds static inline unsigned int fold_hash(unsigned long hash) 1623bfcfaa77SLinus Torvalds { 1624bfcfaa77SLinus Torvalds hash += hash >> (8*sizeof(int)); 1625bfcfaa77SLinus Torvalds return hash; 1626bfcfaa77SLinus Torvalds } 1627bfcfaa77SLinus Torvalds 1628bfcfaa77SLinus Torvalds #else /* 32-bit case */ 1629bfcfaa77SLinus Torvalds 1630bfcfaa77SLinus Torvalds #define fold_hash(x) (x) 1631bfcfaa77SLinus Torvalds 1632bfcfaa77SLinus Torvalds #endif 1633bfcfaa77SLinus Torvalds 1634bfcfaa77SLinus Torvalds unsigned int full_name_hash(const unsigned char *name, unsigned int len) 1635bfcfaa77SLinus Torvalds { 1636bfcfaa77SLinus Torvalds unsigned long a, mask; 1637bfcfaa77SLinus Torvalds unsigned long hash = 0; 1638bfcfaa77SLinus Torvalds 1639bfcfaa77SLinus Torvalds for (;;) { 1640e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name); 1641bfcfaa77SLinus Torvalds if (len < sizeof(unsigned long)) 1642bfcfaa77SLinus Torvalds break; 1643bfcfaa77SLinus Torvalds hash += a; 1644f132c5beSAl Viro hash *= 9; 1645bfcfaa77SLinus Torvalds name += sizeof(unsigned long); 1646bfcfaa77SLinus Torvalds len -= sizeof(unsigned long); 1647bfcfaa77SLinus Torvalds if (!len) 1648bfcfaa77SLinus Torvalds goto done; 1649bfcfaa77SLinus Torvalds } 1650bfcfaa77SLinus Torvalds mask = ~(~0ul << len*8); 1651bfcfaa77SLinus Torvalds hash += mask & a; 1652bfcfaa77SLinus Torvalds done: 1653bfcfaa77SLinus Torvalds return fold_hash(hash); 1654bfcfaa77SLinus Torvalds } 1655bfcfaa77SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 1656bfcfaa77SLinus Torvalds 1657bfcfaa77SLinus Torvalds /* 1658bfcfaa77SLinus Torvalds * Calculate the length and hash of the path component, and 1659bfcfaa77SLinus Torvalds * return the length of the component; 1660bfcfaa77SLinus Torvalds */ 1661bfcfaa77SLinus Torvalds static inline unsigned long hash_name(const char *name, unsigned int *hashp) 1662bfcfaa77SLinus Torvalds { 166336126f8fSLinus Torvalds unsigned long a, b, adata, bdata, mask, hash, len; 166436126f8fSLinus Torvalds const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 1665bfcfaa77SLinus Torvalds 1666bfcfaa77SLinus Torvalds hash = a = 0; 1667bfcfaa77SLinus Torvalds len = -sizeof(unsigned long); 1668bfcfaa77SLinus Torvalds do { 1669bfcfaa77SLinus Torvalds hash = (hash + a) * 9; 1670bfcfaa77SLinus Torvalds len += sizeof(unsigned long); 1671e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name+len); 167236126f8fSLinus Torvalds b = a ^ REPEAT_BYTE('/'); 167336126f8fSLinus Torvalds } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); 1674bfcfaa77SLinus Torvalds 167536126f8fSLinus Torvalds adata = prep_zero_mask(a, adata, &constants); 167636126f8fSLinus Torvalds bdata = prep_zero_mask(b, bdata, &constants); 167736126f8fSLinus Torvalds 167836126f8fSLinus Torvalds mask = create_zero_mask(adata | bdata); 167936126f8fSLinus Torvalds 168036126f8fSLinus Torvalds hash += a & zero_bytemask(mask); 1681bfcfaa77SLinus Torvalds *hashp = fold_hash(hash); 1682bfcfaa77SLinus Torvalds 168336126f8fSLinus Torvalds return len + find_zero(mask); 1684bfcfaa77SLinus Torvalds } 1685bfcfaa77SLinus Torvalds 1686bfcfaa77SLinus Torvalds #else 1687bfcfaa77SLinus Torvalds 16880145acc2SLinus Torvalds unsigned int full_name_hash(const unsigned char *name, unsigned int len) 16890145acc2SLinus Torvalds { 16900145acc2SLinus Torvalds unsigned long hash = init_name_hash(); 16910145acc2SLinus Torvalds while (len--) 16920145acc2SLinus Torvalds hash = partial_name_hash(*name++, hash); 16930145acc2SLinus Torvalds return end_name_hash(hash); 16940145acc2SLinus Torvalds } 1695ae942ae7SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 16960145acc2SLinus Torvalds 16973ddcd056SLinus Torvalds /* 1698200e9ef7SLinus Torvalds * We know there's a real path component here of at least 1699200e9ef7SLinus Torvalds * one character. 1700200e9ef7SLinus Torvalds */ 1701200e9ef7SLinus Torvalds static inline unsigned long hash_name(const char *name, unsigned int *hashp) 1702200e9ef7SLinus Torvalds { 1703200e9ef7SLinus Torvalds unsigned long hash = init_name_hash(); 1704200e9ef7SLinus Torvalds unsigned long len = 0, c; 1705200e9ef7SLinus Torvalds 1706200e9ef7SLinus Torvalds c = (unsigned char)*name; 1707200e9ef7SLinus Torvalds do { 1708200e9ef7SLinus Torvalds len++; 1709200e9ef7SLinus Torvalds hash = partial_name_hash(c, hash); 1710200e9ef7SLinus Torvalds c = (unsigned char)name[len]; 1711200e9ef7SLinus Torvalds } while (c && c != '/'); 1712200e9ef7SLinus Torvalds *hashp = end_name_hash(hash); 1713200e9ef7SLinus Torvalds return len; 1714200e9ef7SLinus Torvalds } 1715200e9ef7SLinus Torvalds 1716bfcfaa77SLinus Torvalds #endif 1717bfcfaa77SLinus Torvalds 1718200e9ef7SLinus Torvalds /* 17191da177e4SLinus Torvalds * Name resolution. 1720ea3834d9SPrasanna Meda * This is the basic name resolution function, turning a pathname into 1721ea3834d9SPrasanna Meda * the final dentry. We expect 'base' to be positive and a directory. 17221da177e4SLinus Torvalds * 1723ea3834d9SPrasanna Meda * Returns 0 and nd will have valid dentry and mnt on success. 1724ea3834d9SPrasanna Meda * Returns error and drops reference to input namei data on failure. 17251da177e4SLinus Torvalds */ 17266de88d72SAl Viro static int link_path_walk(const char *name, struct nameidata *nd) 17271da177e4SLinus Torvalds { 17281da177e4SLinus Torvalds struct path next; 17291da177e4SLinus Torvalds int err; 17301da177e4SLinus Torvalds 17311da177e4SLinus Torvalds while (*name=='/') 17321da177e4SLinus Torvalds name++; 17331da177e4SLinus Torvalds if (!*name) 1734086e183aSAl Viro return 0; 17351da177e4SLinus Torvalds 17361da177e4SLinus Torvalds /* At this point we know we have a real path component. */ 17371da177e4SLinus Torvalds for(;;) { 17381da177e4SLinus Torvalds struct qstr this; 1739200e9ef7SLinus Torvalds long len; 1740fe479a58SAl Viro int type; 17411da177e4SLinus Torvalds 174252094c8aSAl Viro err = may_lookup(nd); 17431da177e4SLinus Torvalds if (err) 17441da177e4SLinus Torvalds break; 17451da177e4SLinus Torvalds 1746200e9ef7SLinus Torvalds len = hash_name(name, &this.hash); 17471da177e4SLinus Torvalds this.name = name; 1748200e9ef7SLinus Torvalds this.len = len; 17491da177e4SLinus Torvalds 1750fe479a58SAl Viro type = LAST_NORM; 1751200e9ef7SLinus Torvalds if (name[0] == '.') switch (len) { 1752fe479a58SAl Viro case 2: 1753200e9ef7SLinus Torvalds if (name[1] == '.') { 1754fe479a58SAl Viro type = LAST_DOTDOT; 175516c2cd71SAl Viro nd->flags |= LOOKUP_JUMPED; 175616c2cd71SAl Viro } 1757fe479a58SAl Viro break; 1758fe479a58SAl Viro case 1: 1759fe479a58SAl Viro type = LAST_DOT; 1760fe479a58SAl Viro } 17615a202bcdSAl Viro if (likely(type == LAST_NORM)) { 17625a202bcdSAl Viro struct dentry *parent = nd->path.dentry; 176316c2cd71SAl Viro nd->flags &= ~LOOKUP_JUMPED; 17645a202bcdSAl Viro if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { 1765da53be12SLinus Torvalds err = parent->d_op->d_hash(parent, &this); 17665a202bcdSAl Viro if (err < 0) 17675a202bcdSAl Viro break; 17685a202bcdSAl Viro } 17695a202bcdSAl Viro } 1770fe479a58SAl Viro 17715f4a6a69SAl Viro nd->last = this; 17725f4a6a69SAl Viro nd->last_type = type; 17735f4a6a69SAl Viro 1774200e9ef7SLinus Torvalds if (!name[len]) 17755f4a6a69SAl Viro return 0; 1776200e9ef7SLinus Torvalds /* 1777200e9ef7SLinus Torvalds * If it wasn't NUL, we know it was '/'. Skip that 1778200e9ef7SLinus Torvalds * slash, and continue until no more slashes. 1779200e9ef7SLinus Torvalds */ 1780200e9ef7SLinus Torvalds do { 1781200e9ef7SLinus Torvalds len++; 1782200e9ef7SLinus Torvalds } while (unlikely(name[len] == '/')); 1783200e9ef7SLinus Torvalds if (!name[len]) 17845f4a6a69SAl Viro return 0; 17855f4a6a69SAl Viro 1786200e9ef7SLinus Torvalds name += len; 17871da177e4SLinus Torvalds 178821b9b073SAl Viro err = walk_component(nd, &next, LOOKUP_FOLLOW); 1789ce57dfc1SAl Viro if (err < 0) 1790ce57dfc1SAl Viro return err; 1791fe479a58SAl Viro 1792ce57dfc1SAl Viro if (err) { 1793b356379aSAl Viro err = nested_symlink(&next, nd); 17941da177e4SLinus Torvalds if (err) 1795a7472babSAl Viro return err; 179631e6b01fSNick Piggin } 1797b18825a7SDavid Howells if (!d_is_directory(nd->path.dentry)) { 17983ddcd056SLinus Torvalds err = -ENOTDIR; 17993ddcd056SLinus Torvalds break; 18005f4a6a69SAl Viro } 1801ce57dfc1SAl Viro } 1802951361f9SAl Viro terminate_walk(nd); 18031da177e4SLinus Torvalds return err; 18041da177e4SLinus Torvalds } 18051da177e4SLinus Torvalds 180670e9b357SAl Viro static int path_init(int dfd, const char *name, unsigned int flags, 180770e9b357SAl Viro struct nameidata *nd, struct file **fp) 180831e6b01fSNick Piggin { 180931e6b01fSNick Piggin int retval = 0; 181031e6b01fSNick Piggin 181131e6b01fSNick Piggin nd->last_type = LAST_ROOT; /* if there are only slashes... */ 181216c2cd71SAl Viro nd->flags = flags | LOOKUP_JUMPED; 181331e6b01fSNick Piggin nd->depth = 0; 18145b6ca027SAl Viro if (flags & LOOKUP_ROOT) { 1815b18825a7SDavid Howells struct dentry *root = nd->root.dentry; 1816b18825a7SDavid Howells struct inode *inode = root->d_inode; 181773d049a4SAl Viro if (*name) { 1818b18825a7SDavid Howells if (!d_is_directory(root)) 18195b6ca027SAl Viro return -ENOTDIR; 18205b6ca027SAl Viro retval = inode_permission(inode, MAY_EXEC); 18215b6ca027SAl Viro if (retval) 18225b6ca027SAl Viro return retval; 182373d049a4SAl Viro } 18245b6ca027SAl Viro nd->path = nd->root; 18255b6ca027SAl Viro nd->inode = inode; 18265b6ca027SAl Viro if (flags & LOOKUP_RCU) { 18278b61e74fSAl Viro rcu_read_lock(); 18285b6ca027SAl Viro nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 182948a066e7SAl Viro nd->m_seq = read_seqbegin(&mount_lock); 18305b6ca027SAl Viro } else { 18315b6ca027SAl Viro path_get(&nd->path); 18325b6ca027SAl Viro } 18335b6ca027SAl Viro return 0; 18345b6ca027SAl Viro } 18355b6ca027SAl Viro 183631e6b01fSNick Piggin nd->root.mnt = NULL; 183731e6b01fSNick Piggin 183848a066e7SAl Viro nd->m_seq = read_seqbegin(&mount_lock); 183931e6b01fSNick Piggin if (*name=='/') { 1840e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 18418b61e74fSAl Viro rcu_read_lock(); 1842e41f7d4eSAl Viro set_root_rcu(nd); 1843e41f7d4eSAl Viro } else { 1844e41f7d4eSAl Viro set_root(nd); 1845e41f7d4eSAl Viro path_get(&nd->root); 1846e41f7d4eSAl Viro } 184731e6b01fSNick Piggin nd->path = nd->root; 184831e6b01fSNick Piggin } else if (dfd == AT_FDCWD) { 1849e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 185031e6b01fSNick Piggin struct fs_struct *fs = current->fs; 1851c28cc364SNick Piggin unsigned seq; 185231e6b01fSNick Piggin 18538b61e74fSAl Viro rcu_read_lock(); 185431e6b01fSNick Piggin 1855c28cc364SNick Piggin do { 1856c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 185731e6b01fSNick Piggin nd->path = fs->pwd; 1858c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 1859c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 1860e41f7d4eSAl Viro } else { 1861e41f7d4eSAl Viro get_fs_pwd(current->fs, &nd->path); 1862e41f7d4eSAl Viro } 186331e6b01fSNick Piggin } else { 1864582aa64aSJeff Layton /* Caller must check execute permissions on the starting path component */ 18652903ff01SAl Viro struct fd f = fdget_raw(dfd); 186631e6b01fSNick Piggin struct dentry *dentry; 186731e6b01fSNick Piggin 18682903ff01SAl Viro if (!f.file) 18692903ff01SAl Viro return -EBADF; 187031e6b01fSNick Piggin 18712903ff01SAl Viro dentry = f.file->f_path.dentry; 187231e6b01fSNick Piggin 1873f52e0c11SAl Viro if (*name) { 1874b18825a7SDavid Howells if (!d_is_directory(dentry)) { 18752903ff01SAl Viro fdput(f); 18762903ff01SAl Viro return -ENOTDIR; 1877f52e0c11SAl Viro } 18782903ff01SAl Viro } 18792903ff01SAl Viro 18802903ff01SAl Viro nd->path = f.file->f_path; 1881e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 18822903ff01SAl Viro if (f.need_put) 18832903ff01SAl Viro *fp = f.file; 1884c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 18858b61e74fSAl Viro rcu_read_lock(); 18865590ff0dSUlrich Drepper } else { 18872903ff01SAl Viro path_get(&nd->path); 18882903ff01SAl Viro fdput(f); 18891da177e4SLinus Torvalds } 1890e41f7d4eSAl Viro } 1891e41f7d4eSAl Viro 189231e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 18939b4a9b14SAl Viro return 0; 18949b4a9b14SAl Viro } 18959b4a9b14SAl Viro 1896bd92d7feSAl Viro static inline int lookup_last(struct nameidata *nd, struct path *path) 1897bd92d7feSAl Viro { 1898bd92d7feSAl Viro if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) 1899bd92d7feSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 1900bd92d7feSAl Viro 1901bd92d7feSAl Viro nd->flags &= ~LOOKUP_PARENT; 190221b9b073SAl Viro return walk_component(nd, path, nd->flags & LOOKUP_FOLLOW); 1903bd92d7feSAl Viro } 1904bd92d7feSAl Viro 19059b4a9b14SAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 1906ee0827cdSAl Viro static int path_lookupat(int dfd, const char *name, 19079b4a9b14SAl Viro unsigned int flags, struct nameidata *nd) 19089b4a9b14SAl Viro { 190970e9b357SAl Viro struct file *base = NULL; 1910bd92d7feSAl Viro struct path path; 1911bd92d7feSAl Viro int err; 191231e6b01fSNick Piggin 191331e6b01fSNick Piggin /* 191431e6b01fSNick Piggin * Path walking is largely split up into 2 different synchronisation 191531e6b01fSNick Piggin * schemes, rcu-walk and ref-walk (explained in 191631e6b01fSNick Piggin * Documentation/filesystems/path-lookup.txt). These share much of the 191731e6b01fSNick Piggin * path walk code, but some things particularly setup, cleanup, and 191831e6b01fSNick Piggin * following mounts are sufficiently divergent that functions are 191931e6b01fSNick Piggin * duplicated. Typically there is a function foo(), and its RCU 192031e6b01fSNick Piggin * analogue, foo_rcu(). 192131e6b01fSNick Piggin * 192231e6b01fSNick Piggin * -ECHILD is the error number of choice (just to avoid clashes) that 192331e6b01fSNick Piggin * is returned if some aspect of an rcu-walk fails. Such an error must 192431e6b01fSNick Piggin * be handled by restarting a traditional ref-walk (which will always 192531e6b01fSNick Piggin * be able to complete). 192631e6b01fSNick Piggin */ 1927bd92d7feSAl Viro err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base); 1928ee0827cdSAl Viro 1929bd92d7feSAl Viro if (unlikely(err)) 1930bd92d7feSAl Viro return err; 1931ee0827cdSAl Viro 1932ee0827cdSAl Viro current->total_link_count = 0; 1933bd92d7feSAl Viro err = link_path_walk(name, nd); 1934bd92d7feSAl Viro 1935bd92d7feSAl Viro if (!err && !(flags & LOOKUP_PARENT)) { 1936bd92d7feSAl Viro err = lookup_last(nd, &path); 1937bd92d7feSAl Viro while (err > 0) { 1938bd92d7feSAl Viro void *cookie; 1939bd92d7feSAl Viro struct path link = path; 1940800179c9SKees Cook err = may_follow_link(&link, nd); 1941800179c9SKees Cook if (unlikely(err)) 1942800179c9SKees Cook break; 1943bd92d7feSAl Viro nd->flags |= LOOKUP_PARENT; 1944574197e0SAl Viro err = follow_link(&link, nd, &cookie); 19456d7b5aaeSAl Viro if (err) 19466d7b5aaeSAl Viro break; 1947bd92d7feSAl Viro err = lookup_last(nd, &path); 1948574197e0SAl Viro put_link(nd, &link, cookie); 1949bd92d7feSAl Viro } 1950bd92d7feSAl Viro } 1951ee0827cdSAl Viro 19529f1fafeeSAl Viro if (!err) 19539f1fafeeSAl Viro err = complete_walk(nd); 1954bd92d7feSAl Viro 1955bd92d7feSAl Viro if (!err && nd->flags & LOOKUP_DIRECTORY) { 1956b18825a7SDavid Howells if (!d_is_directory(nd->path.dentry)) { 1957bd92d7feSAl Viro path_put(&nd->path); 1958bd23a539SAl Viro err = -ENOTDIR; 1959bd92d7feSAl Viro } 1960bd92d7feSAl Viro } 196116c2cd71SAl Viro 196270e9b357SAl Viro if (base) 196370e9b357SAl Viro fput(base); 1964ee0827cdSAl Viro 19655b6ca027SAl Viro if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 196631e6b01fSNick Piggin path_put(&nd->root); 196731e6b01fSNick Piggin nd->root.mnt = NULL; 196831e6b01fSNick Piggin } 1969bd92d7feSAl Viro return err; 197031e6b01fSNick Piggin } 197131e6b01fSNick Piggin 1972873f1eedSJeff Layton static int filename_lookup(int dfd, struct filename *name, 1973873f1eedSJeff Layton unsigned int flags, struct nameidata *nd) 1974873f1eedSJeff Layton { 1975873f1eedSJeff Layton int retval = path_lookupat(dfd, name->name, flags | LOOKUP_RCU, nd); 1976873f1eedSJeff Layton if (unlikely(retval == -ECHILD)) 1977873f1eedSJeff Layton retval = path_lookupat(dfd, name->name, flags, nd); 1978873f1eedSJeff Layton if (unlikely(retval == -ESTALE)) 1979873f1eedSJeff Layton retval = path_lookupat(dfd, name->name, 1980873f1eedSJeff Layton flags | LOOKUP_REVAL, nd); 1981873f1eedSJeff Layton 1982873f1eedSJeff Layton if (likely(!retval)) 1983adb5c247SJeff Layton audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT); 1984873f1eedSJeff Layton return retval; 1985873f1eedSJeff Layton } 1986873f1eedSJeff Layton 1987ee0827cdSAl Viro static int do_path_lookup(int dfd, const char *name, 1988ee0827cdSAl Viro unsigned int flags, struct nameidata *nd) 1989ee0827cdSAl Viro { 1990873f1eedSJeff Layton struct filename filename = { .name = name }; 1991ee0827cdSAl Viro 1992873f1eedSJeff Layton return filename_lookup(dfd, &filename, flags, nd); 19931da177e4SLinus Torvalds } 19941da177e4SLinus Torvalds 199579714f72SAl Viro /* does lookup, returns the object with parent locked */ 199679714f72SAl Viro struct dentry *kern_path_locked(const char *name, struct path *path) 19975590ff0dSUlrich Drepper { 199879714f72SAl Viro struct nameidata nd; 199979714f72SAl Viro struct dentry *d; 200079714f72SAl Viro int err = do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, &nd); 200179714f72SAl Viro if (err) 200279714f72SAl Viro return ERR_PTR(err); 200379714f72SAl Viro if (nd.last_type != LAST_NORM) { 200479714f72SAl Viro path_put(&nd.path); 200579714f72SAl Viro return ERR_PTR(-EINVAL); 200679714f72SAl Viro } 200779714f72SAl Viro mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 20081e0ea001SAl Viro d = __lookup_hash(&nd.last, nd.path.dentry, 0); 200979714f72SAl Viro if (IS_ERR(d)) { 201079714f72SAl Viro mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 201179714f72SAl Viro path_put(&nd.path); 201279714f72SAl Viro return d; 201379714f72SAl Viro } 201479714f72SAl Viro *path = nd.path; 201579714f72SAl Viro return d; 20165590ff0dSUlrich Drepper } 20175590ff0dSUlrich Drepper 2018d1811465SAl Viro int kern_path(const char *name, unsigned int flags, struct path *path) 2019d1811465SAl Viro { 2020d1811465SAl Viro struct nameidata nd; 2021d1811465SAl Viro int res = do_path_lookup(AT_FDCWD, name, flags, &nd); 2022d1811465SAl Viro if (!res) 2023d1811465SAl Viro *path = nd.path; 2024d1811465SAl Viro return res; 2025d1811465SAl Viro } 2026d1811465SAl Viro 202716f18200SJosef 'Jeff' Sipek /** 202816f18200SJosef 'Jeff' Sipek * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair 202916f18200SJosef 'Jeff' Sipek * @dentry: pointer to dentry of the base directory 203016f18200SJosef 'Jeff' Sipek * @mnt: pointer to vfs mount of the base directory 203116f18200SJosef 'Jeff' Sipek * @name: pointer to file name 203216f18200SJosef 'Jeff' Sipek * @flags: lookup flags 2033e0a01249SAl Viro * @path: pointer to struct path to fill 203416f18200SJosef 'Jeff' Sipek */ 203516f18200SJosef 'Jeff' Sipek int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, 203616f18200SJosef 'Jeff' Sipek const char *name, unsigned int flags, 2037e0a01249SAl Viro struct path *path) 203816f18200SJosef 'Jeff' Sipek { 2039e0a01249SAl Viro struct nameidata nd; 2040e0a01249SAl Viro int err; 2041e0a01249SAl Viro nd.root.dentry = dentry; 2042e0a01249SAl Viro nd.root.mnt = mnt; 2043e0a01249SAl Viro BUG_ON(flags & LOOKUP_PARENT); 20445b6ca027SAl Viro /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */ 2045e0a01249SAl Viro err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd); 2046e0a01249SAl Viro if (!err) 2047e0a01249SAl Viro *path = nd.path; 2048e0a01249SAl Viro return err; 204916f18200SJosef 'Jeff' Sipek } 205016f18200SJosef 'Jeff' Sipek 2051057f6c01SJames Morris /* 2052057f6c01SJames Morris * Restricted form of lookup. Doesn't follow links, single-component only, 2053057f6c01SJames Morris * needs parent already locked. Doesn't follow mounts. 2054057f6c01SJames Morris * SMP-safe. 2055057f6c01SJames Morris */ 2056a244e169SAdrian Bunk static struct dentry *lookup_hash(struct nameidata *nd) 20571da177e4SLinus Torvalds { 205872bd866aSAl Viro return __lookup_hash(&nd->last, nd->path.dentry, nd->flags); 20591da177e4SLinus Torvalds } 20601da177e4SLinus Torvalds 2061eead1911SChristoph Hellwig /** 2062a6b91919SRandy Dunlap * lookup_one_len - filesystem helper to lookup single pathname component 2063eead1911SChristoph Hellwig * @name: pathname component to lookup 2064eead1911SChristoph Hellwig * @base: base directory to lookup from 2065eead1911SChristoph Hellwig * @len: maximum length @len should be interpreted to 2066eead1911SChristoph Hellwig * 2067a6b91919SRandy Dunlap * Note that this routine is purely a helper for filesystem usage and should 2068a6b91919SRandy Dunlap * not be called by generic code. Also note that by using this function the 2069eead1911SChristoph Hellwig * nameidata argument is passed to the filesystem methods and a filesystem 2070eead1911SChristoph Hellwig * using this helper needs to be prepared for that. 2071eead1911SChristoph Hellwig */ 2072057f6c01SJames Morris struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 2073057f6c01SJames Morris { 2074057f6c01SJames Morris struct qstr this; 20756a96ba54SAl Viro unsigned int c; 2076cda309deSMiklos Szeredi int err; 2077057f6c01SJames Morris 20782f9092e1SDavid Woodhouse WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex)); 20792f9092e1SDavid Woodhouse 20806a96ba54SAl Viro this.name = name; 20816a96ba54SAl Viro this.len = len; 20820145acc2SLinus Torvalds this.hash = full_name_hash(name, len); 20836a96ba54SAl Viro if (!len) 20846a96ba54SAl Viro return ERR_PTR(-EACCES); 20856a96ba54SAl Viro 208621d8a15aSAl Viro if (unlikely(name[0] == '.')) { 208721d8a15aSAl Viro if (len < 2 || (len == 2 && name[1] == '.')) 208821d8a15aSAl Viro return ERR_PTR(-EACCES); 208921d8a15aSAl Viro } 209021d8a15aSAl Viro 20916a96ba54SAl Viro while (len--) { 20926a96ba54SAl Viro c = *(const unsigned char *)name++; 20936a96ba54SAl Viro if (c == '/' || c == '\0') 20946a96ba54SAl Viro return ERR_PTR(-EACCES); 20956a96ba54SAl Viro } 20965a202bcdSAl Viro /* 20975a202bcdSAl Viro * See if the low-level filesystem might want 20985a202bcdSAl Viro * to use its own hash.. 20995a202bcdSAl Viro */ 21005a202bcdSAl Viro if (base->d_flags & DCACHE_OP_HASH) { 2101da53be12SLinus Torvalds int err = base->d_op->d_hash(base, &this); 21025a202bcdSAl Viro if (err < 0) 21035a202bcdSAl Viro return ERR_PTR(err); 21045a202bcdSAl Viro } 2105eead1911SChristoph Hellwig 2106cda309deSMiklos Szeredi err = inode_permission(base->d_inode, MAY_EXEC); 2107cda309deSMiklos Szeredi if (err) 2108cda309deSMiklos Szeredi return ERR_PTR(err); 2109cda309deSMiklos Szeredi 211072bd866aSAl Viro return __lookup_hash(&this, base, 0); 2111057f6c01SJames Morris } 2112057f6c01SJames Morris 21131fa1e7f6SAndy Whitcroft int user_path_at_empty(int dfd, const char __user *name, unsigned flags, 21141fa1e7f6SAndy Whitcroft struct path *path, int *empty) 21151da177e4SLinus Torvalds { 21162d8f3038SAl Viro struct nameidata nd; 211791a27b2aSJeff Layton struct filename *tmp = getname_flags(name, flags, empty); 21181da177e4SLinus Torvalds int err = PTR_ERR(tmp); 21191da177e4SLinus Torvalds if (!IS_ERR(tmp)) { 21202d8f3038SAl Viro 21212d8f3038SAl Viro BUG_ON(flags & LOOKUP_PARENT); 21222d8f3038SAl Viro 2123873f1eedSJeff Layton err = filename_lookup(dfd, tmp, flags, &nd); 21241da177e4SLinus Torvalds putname(tmp); 21252d8f3038SAl Viro if (!err) 21262d8f3038SAl Viro *path = nd.path; 21271da177e4SLinus Torvalds } 21281da177e4SLinus Torvalds return err; 21291da177e4SLinus Torvalds } 21301da177e4SLinus Torvalds 21311fa1e7f6SAndy Whitcroft int user_path_at(int dfd, const char __user *name, unsigned flags, 21321fa1e7f6SAndy Whitcroft struct path *path) 21331fa1e7f6SAndy Whitcroft { 2134f7493e5dSLinus Torvalds return user_path_at_empty(dfd, name, flags, path, NULL); 21351fa1e7f6SAndy Whitcroft } 21361fa1e7f6SAndy Whitcroft 2137873f1eedSJeff Layton /* 2138873f1eedSJeff Layton * NB: most callers don't do anything directly with the reference to the 2139873f1eedSJeff Layton * to struct filename, but the nd->last pointer points into the name string 2140873f1eedSJeff Layton * allocated by getname. So we must hold the reference to it until all 2141873f1eedSJeff Layton * path-walking is complete. 2142873f1eedSJeff Layton */ 214391a27b2aSJeff Layton static struct filename * 21449e790bd6SJeff Layton user_path_parent(int dfd, const char __user *path, struct nameidata *nd, 21459e790bd6SJeff Layton unsigned int flags) 21462ad94ae6SAl Viro { 214791a27b2aSJeff Layton struct filename *s = getname(path); 21482ad94ae6SAl Viro int error; 21492ad94ae6SAl Viro 21509e790bd6SJeff Layton /* only LOOKUP_REVAL is allowed in extra flags */ 21519e790bd6SJeff Layton flags &= LOOKUP_REVAL; 21529e790bd6SJeff Layton 21532ad94ae6SAl Viro if (IS_ERR(s)) 215491a27b2aSJeff Layton return s; 21552ad94ae6SAl Viro 21569e790bd6SJeff Layton error = filename_lookup(dfd, s, flags | LOOKUP_PARENT, nd); 215791a27b2aSJeff Layton if (error) { 21582ad94ae6SAl Viro putname(s); 215991a27b2aSJeff Layton return ERR_PTR(error); 216091a27b2aSJeff Layton } 21612ad94ae6SAl Viro 216291a27b2aSJeff Layton return s; 21632ad94ae6SAl Viro } 21642ad94ae6SAl Viro 21658033426eSJeff Layton /** 2166197df04cSAl Viro * mountpoint_last - look up last component for umount 21678033426eSJeff Layton * @nd: pathwalk nameidata - currently pointing at parent directory of "last" 21688033426eSJeff Layton * @path: pointer to container for result 21698033426eSJeff Layton * 21708033426eSJeff Layton * This is a special lookup_last function just for umount. In this case, we 21718033426eSJeff Layton * need to resolve the path without doing any revalidation. 21728033426eSJeff Layton * 21738033426eSJeff Layton * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since 21748033426eSJeff Layton * mountpoints are always pinned in the dcache, their ancestors are too. Thus, 21758033426eSJeff Layton * in almost all cases, this lookup will be served out of the dcache. The only 21768033426eSJeff Layton * cases where it won't are if nd->last refers to a symlink or the path is 21778033426eSJeff Layton * bogus and it doesn't exist. 21788033426eSJeff Layton * 21798033426eSJeff Layton * Returns: 21808033426eSJeff Layton * -error: if there was an error during lookup. This includes -ENOENT if the 21818033426eSJeff Layton * lookup found a negative dentry. The nd->path reference will also be 21828033426eSJeff Layton * put in this case. 21838033426eSJeff Layton * 21848033426eSJeff Layton * 0: if we successfully resolved nd->path and found it to not to be a 21858033426eSJeff Layton * symlink that needs to be followed. "path" will also be populated. 21868033426eSJeff Layton * The nd->path reference will also be put. 21878033426eSJeff Layton * 21888033426eSJeff Layton * 1: if we successfully resolved nd->last and found it to be a symlink 21898033426eSJeff Layton * that needs to be followed. "path" will be populated with the path 21908033426eSJeff Layton * to the link, and nd->path will *not* be put. 21918033426eSJeff Layton */ 21928033426eSJeff Layton static int 2193197df04cSAl Viro mountpoint_last(struct nameidata *nd, struct path *path) 21948033426eSJeff Layton { 21958033426eSJeff Layton int error = 0; 21968033426eSJeff Layton struct dentry *dentry; 21978033426eSJeff Layton struct dentry *dir = nd->path.dentry; 21988033426eSJeff Layton 219935759521SAl Viro /* If we're in rcuwalk, drop out of it to handle last component */ 220035759521SAl Viro if (nd->flags & LOOKUP_RCU) { 220135759521SAl Viro if (unlazy_walk(nd, NULL)) { 22028033426eSJeff Layton error = -ECHILD; 220335759521SAl Viro goto out; 220435759521SAl Viro } 22058033426eSJeff Layton } 22068033426eSJeff Layton 22078033426eSJeff Layton nd->flags &= ~LOOKUP_PARENT; 22088033426eSJeff Layton 22098033426eSJeff Layton if (unlikely(nd->last_type != LAST_NORM)) { 22108033426eSJeff Layton error = handle_dots(nd, nd->last_type); 221135759521SAl Viro if (error) 221235759521SAl Viro goto out; 22138033426eSJeff Layton dentry = dget(nd->path.dentry); 221435759521SAl Viro goto done; 22158033426eSJeff Layton } 22168033426eSJeff Layton 22178033426eSJeff Layton mutex_lock(&dir->d_inode->i_mutex); 22188033426eSJeff Layton dentry = d_lookup(dir, &nd->last); 22198033426eSJeff Layton if (!dentry) { 22208033426eSJeff Layton /* 22218033426eSJeff Layton * No cached dentry. Mounted dentries are pinned in the cache, 22228033426eSJeff Layton * so that means that this dentry is probably a symlink or the 22238033426eSJeff Layton * path doesn't actually point to a mounted dentry. 22248033426eSJeff Layton */ 22258033426eSJeff Layton dentry = d_alloc(dir, &nd->last); 22268033426eSJeff Layton if (!dentry) { 22278033426eSJeff Layton error = -ENOMEM; 2228bcceeebaSDave Jones mutex_unlock(&dir->d_inode->i_mutex); 222935759521SAl Viro goto out; 22308033426eSJeff Layton } 223135759521SAl Viro dentry = lookup_real(dir->d_inode, dentry, nd->flags); 223235759521SAl Viro error = PTR_ERR(dentry); 2233bcceeebaSDave Jones if (IS_ERR(dentry)) { 2234bcceeebaSDave Jones mutex_unlock(&dir->d_inode->i_mutex); 223535759521SAl Viro goto out; 22368033426eSJeff Layton } 2237bcceeebaSDave Jones } 22388033426eSJeff Layton mutex_unlock(&dir->d_inode->i_mutex); 22398033426eSJeff Layton 224035759521SAl Viro done: 22418033426eSJeff Layton if (!dentry->d_inode) { 22428033426eSJeff Layton error = -ENOENT; 22438033426eSJeff Layton dput(dentry); 224435759521SAl Viro goto out; 224535759521SAl Viro } 22468033426eSJeff Layton path->dentry = dentry; 22478033426eSJeff Layton path->mnt = mntget(nd->path.mnt); 2248b18825a7SDavid Howells if (should_follow_link(dentry, nd->flags & LOOKUP_FOLLOW)) 22498033426eSJeff Layton return 1; 22508033426eSJeff Layton follow_mount(path); 225135759521SAl Viro error = 0; 225235759521SAl Viro out: 22538033426eSJeff Layton terminate_walk(nd); 22548033426eSJeff Layton return error; 22558033426eSJeff Layton } 22568033426eSJeff Layton 22578033426eSJeff Layton /** 2258197df04cSAl Viro * path_mountpoint - look up a path to be umounted 22598033426eSJeff Layton * @dfd: directory file descriptor to start walk from 22608033426eSJeff Layton * @name: full pathname to walk 22618033426eSJeff Layton * @flags: lookup flags 22628033426eSJeff Layton * 22638033426eSJeff Layton * Look up the given name, but don't attempt to revalidate the last component. 22648033426eSJeff Layton * Returns 0 and "path" will be valid on success; Retuns error otherwise. 22658033426eSJeff Layton */ 22668033426eSJeff Layton static int 2267197df04cSAl Viro path_mountpoint(int dfd, const char *name, struct path *path, unsigned int flags) 22688033426eSJeff Layton { 22698033426eSJeff Layton struct file *base = NULL; 22708033426eSJeff Layton struct nameidata nd; 22718033426eSJeff Layton int err; 22728033426eSJeff Layton 22738033426eSJeff Layton err = path_init(dfd, name, flags | LOOKUP_PARENT, &nd, &base); 22748033426eSJeff Layton if (unlikely(err)) 22758033426eSJeff Layton return err; 22768033426eSJeff Layton 22778033426eSJeff Layton current->total_link_count = 0; 22788033426eSJeff Layton err = link_path_walk(name, &nd); 22798033426eSJeff Layton if (err) 22808033426eSJeff Layton goto out; 22818033426eSJeff Layton 2282197df04cSAl Viro err = mountpoint_last(&nd, path); 22838033426eSJeff Layton while (err > 0) { 22848033426eSJeff Layton void *cookie; 22858033426eSJeff Layton struct path link = *path; 22868033426eSJeff Layton err = may_follow_link(&link, &nd); 22878033426eSJeff Layton if (unlikely(err)) 22888033426eSJeff Layton break; 22898033426eSJeff Layton nd.flags |= LOOKUP_PARENT; 22908033426eSJeff Layton err = follow_link(&link, &nd, &cookie); 22918033426eSJeff Layton if (err) 22928033426eSJeff Layton break; 2293197df04cSAl Viro err = mountpoint_last(&nd, path); 22948033426eSJeff Layton put_link(&nd, &link, cookie); 22958033426eSJeff Layton } 22968033426eSJeff Layton out: 22978033426eSJeff Layton if (base) 22988033426eSJeff Layton fput(base); 22998033426eSJeff Layton 23008033426eSJeff Layton if (nd.root.mnt && !(nd.flags & LOOKUP_ROOT)) 23018033426eSJeff Layton path_put(&nd.root); 23028033426eSJeff Layton 23038033426eSJeff Layton return err; 23048033426eSJeff Layton } 23058033426eSJeff Layton 23062d864651SAl Viro static int 23072d864651SAl Viro filename_mountpoint(int dfd, struct filename *s, struct path *path, 23082d864651SAl Viro unsigned int flags) 23092d864651SAl Viro { 23102d864651SAl Viro int error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_RCU); 23112d864651SAl Viro if (unlikely(error == -ECHILD)) 23122d864651SAl Viro error = path_mountpoint(dfd, s->name, path, flags); 23132d864651SAl Viro if (unlikely(error == -ESTALE)) 23142d864651SAl Viro error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_REVAL); 23152d864651SAl Viro if (likely(!error)) 23162d864651SAl Viro audit_inode(s, path->dentry, 0); 23172d864651SAl Viro return error; 23182d864651SAl Viro } 23192d864651SAl Viro 23208033426eSJeff Layton /** 2321197df04cSAl Viro * user_path_mountpoint_at - lookup a path from userland in order to umount it 23228033426eSJeff Layton * @dfd: directory file descriptor 23238033426eSJeff Layton * @name: pathname from userland 23248033426eSJeff Layton * @flags: lookup flags 23258033426eSJeff Layton * @path: pointer to container to hold result 23268033426eSJeff Layton * 23278033426eSJeff Layton * A umount is a special case for path walking. We're not actually interested 23288033426eSJeff Layton * in the inode in this situation, and ESTALE errors can be a problem. We 23298033426eSJeff Layton * simply want track down the dentry and vfsmount attached at the mountpoint 23308033426eSJeff Layton * and avoid revalidating the last component. 23318033426eSJeff Layton * 23328033426eSJeff Layton * Returns 0 and populates "path" on success. 23338033426eSJeff Layton */ 23348033426eSJeff Layton int 2335197df04cSAl Viro user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags, 23368033426eSJeff Layton struct path *path) 23378033426eSJeff Layton { 23388033426eSJeff Layton struct filename *s = getname(name); 23398033426eSJeff Layton int error; 23408033426eSJeff Layton if (IS_ERR(s)) 23418033426eSJeff Layton return PTR_ERR(s); 23422d864651SAl Viro error = filename_mountpoint(dfd, s, path, flags); 23438033426eSJeff Layton putname(s); 23448033426eSJeff Layton return error; 23458033426eSJeff Layton } 23468033426eSJeff Layton 23472d864651SAl Viro int 23482d864651SAl Viro kern_path_mountpoint(int dfd, const char *name, struct path *path, 23492d864651SAl Viro unsigned int flags) 23502d864651SAl Viro { 23512d864651SAl Viro struct filename s = {.name = name}; 23522d864651SAl Viro return filename_mountpoint(dfd, &s, path, flags); 23532d864651SAl Viro } 23542d864651SAl Viro EXPORT_SYMBOL(kern_path_mountpoint); 23552d864651SAl Viro 23561da177e4SLinus Torvalds /* 23571da177e4SLinus Torvalds * It's inline, so penalty for filesystems that don't use sticky bit is 23581da177e4SLinus Torvalds * minimal. 23591da177e4SLinus Torvalds */ 23601da177e4SLinus Torvalds static inline int check_sticky(struct inode *dir, struct inode *inode) 23611da177e4SLinus Torvalds { 23628e96e3b7SEric W. Biederman kuid_t fsuid = current_fsuid(); 2363da9592edSDavid Howells 23641da177e4SLinus Torvalds if (!(dir->i_mode & S_ISVTX)) 23651da177e4SLinus Torvalds return 0; 23668e96e3b7SEric W. Biederman if (uid_eq(inode->i_uid, fsuid)) 23671da177e4SLinus Torvalds return 0; 23688e96e3b7SEric W. Biederman if (uid_eq(dir->i_uid, fsuid)) 23691da177e4SLinus Torvalds return 0; 23701a48e2acSEric W. Biederman return !inode_capable(inode, CAP_FOWNER); 23711da177e4SLinus Torvalds } 23721da177e4SLinus Torvalds 23731da177e4SLinus Torvalds /* 23741da177e4SLinus Torvalds * Check whether we can remove a link victim from directory dir, check 23751da177e4SLinus Torvalds * whether the type of victim is right. 23761da177e4SLinus Torvalds * 1. We can't do it if dir is read-only (done in permission()) 23771da177e4SLinus Torvalds * 2. We should have write and exec permissions on dir 23781da177e4SLinus Torvalds * 3. We can't remove anything from append-only dir 23791da177e4SLinus Torvalds * 4. We can't do anything with immutable dir (done in permission()) 23801da177e4SLinus Torvalds * 5. If the sticky bit on dir is set we should either 23811da177e4SLinus Torvalds * a. be owner of dir, or 23821da177e4SLinus Torvalds * b. be owner of victim, or 23831da177e4SLinus Torvalds * c. have CAP_FOWNER capability 23841da177e4SLinus Torvalds * 6. If the victim is append-only or immutable we can't do antyhing with 23851da177e4SLinus Torvalds * links pointing to it. 23861da177e4SLinus Torvalds * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. 23871da177e4SLinus Torvalds * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. 23881da177e4SLinus Torvalds * 9. We can't remove a root or mountpoint. 23891da177e4SLinus Torvalds * 10. We don't allow removal of NFS sillyrenamed files; it's handled by 23901da177e4SLinus Torvalds * nfs_async_unlink(). 23911da177e4SLinus Torvalds */ 2392b18825a7SDavid Howells static int may_delete(struct inode *dir, struct dentry *victim, bool isdir) 23931da177e4SLinus Torvalds { 2394b18825a7SDavid Howells struct inode *inode = victim->d_inode; 23951da177e4SLinus Torvalds int error; 23961da177e4SLinus Torvalds 2397b18825a7SDavid Howells if (d_is_negative(victim)) 23981da177e4SLinus Torvalds return -ENOENT; 2399b18825a7SDavid Howells BUG_ON(!inode); 24001da177e4SLinus Torvalds 24011da177e4SLinus Torvalds BUG_ON(victim->d_parent->d_inode != dir); 24024fa6b5ecSJeff Layton audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); 24031da177e4SLinus Torvalds 2404f419a2e3SAl Viro error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 24051da177e4SLinus Torvalds if (error) 24061da177e4SLinus Torvalds return error; 24071da177e4SLinus Torvalds if (IS_APPEND(dir)) 24081da177e4SLinus Torvalds return -EPERM; 2409b18825a7SDavid Howells 2410b18825a7SDavid Howells if (check_sticky(dir, inode) || IS_APPEND(inode) || 2411b18825a7SDavid Howells IS_IMMUTABLE(inode) || IS_SWAPFILE(inode)) 24121da177e4SLinus Torvalds return -EPERM; 24131da177e4SLinus Torvalds if (isdir) { 2414b18825a7SDavid Howells if (!d_is_directory(victim) && !d_is_autodir(victim)) 24151da177e4SLinus Torvalds return -ENOTDIR; 24161da177e4SLinus Torvalds if (IS_ROOT(victim)) 24171da177e4SLinus Torvalds return -EBUSY; 2418b18825a7SDavid Howells } else if (d_is_directory(victim) || d_is_autodir(victim)) 24191da177e4SLinus Torvalds return -EISDIR; 24201da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 24211da177e4SLinus Torvalds return -ENOENT; 24221da177e4SLinus Torvalds if (victim->d_flags & DCACHE_NFSFS_RENAMED) 24231da177e4SLinus Torvalds return -EBUSY; 24241da177e4SLinus Torvalds return 0; 24251da177e4SLinus Torvalds } 24261da177e4SLinus Torvalds 24271da177e4SLinus Torvalds /* Check whether we can create an object with dentry child in directory 24281da177e4SLinus Torvalds * dir. 24291da177e4SLinus Torvalds * 1. We can't do it if child already exists (open has special treatment for 24301da177e4SLinus Torvalds * this case, but since we are inlined it's OK) 24311da177e4SLinus Torvalds * 2. We can't do it if dir is read-only (done in permission()) 24321da177e4SLinus Torvalds * 3. We should have write and exec permissions on dir 24331da177e4SLinus Torvalds * 4. We can't do it if dir is immutable (done in permission()) 24341da177e4SLinus Torvalds */ 2435a95164d9SMiklos Szeredi static inline int may_create(struct inode *dir, struct dentry *child) 24361da177e4SLinus Torvalds { 24371da177e4SLinus Torvalds if (child->d_inode) 24381da177e4SLinus Torvalds return -EEXIST; 24391da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 24401da177e4SLinus Torvalds return -ENOENT; 2441f419a2e3SAl Viro return inode_permission(dir, MAY_WRITE | MAY_EXEC); 24421da177e4SLinus Torvalds } 24431da177e4SLinus Torvalds 24441da177e4SLinus Torvalds /* 24451da177e4SLinus Torvalds * p1 and p2 should be directories on the same fs. 24461da177e4SLinus Torvalds */ 24471da177e4SLinus Torvalds struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 24481da177e4SLinus Torvalds { 24491da177e4SLinus Torvalds struct dentry *p; 24501da177e4SLinus Torvalds 24511da177e4SLinus Torvalds if (p1 == p2) { 2452f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 24531da177e4SLinus Torvalds return NULL; 24541da177e4SLinus Torvalds } 24551da177e4SLinus Torvalds 2456a11f3a05SArjan van de Ven mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 24571da177e4SLinus Torvalds 2458e2761a11SOGAWA Hirofumi p = d_ancestor(p2, p1); 2459e2761a11SOGAWA Hirofumi if (p) { 2460f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT); 2461f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD); 24621da177e4SLinus Torvalds return p; 24631da177e4SLinus Torvalds } 24641da177e4SLinus Torvalds 2465e2761a11SOGAWA Hirofumi p = d_ancestor(p1, p2); 2466e2761a11SOGAWA Hirofumi if (p) { 2467f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 2468f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); 24691da177e4SLinus Torvalds return p; 24701da177e4SLinus Torvalds } 24711da177e4SLinus Torvalds 2472f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 2473f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); 24741da177e4SLinus Torvalds return NULL; 24751da177e4SLinus Torvalds } 24761da177e4SLinus Torvalds 24771da177e4SLinus Torvalds void unlock_rename(struct dentry *p1, struct dentry *p2) 24781da177e4SLinus Torvalds { 24791b1dcc1bSJes Sorensen mutex_unlock(&p1->d_inode->i_mutex); 24801da177e4SLinus Torvalds if (p1 != p2) { 24811b1dcc1bSJes Sorensen mutex_unlock(&p2->d_inode->i_mutex); 2482a11f3a05SArjan van de Ven mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 24831da177e4SLinus Torvalds } 24841da177e4SLinus Torvalds } 24851da177e4SLinus Torvalds 24864acdaf27SAl Viro int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, 2487312b63fbSAl Viro bool want_excl) 24881da177e4SLinus Torvalds { 2489a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 24901da177e4SLinus Torvalds if (error) 24911da177e4SLinus Torvalds return error; 24921da177e4SLinus Torvalds 2493acfa4380SAl Viro if (!dir->i_op->create) 24941da177e4SLinus Torvalds return -EACCES; /* shouldn't it be ENOSYS? */ 24951da177e4SLinus Torvalds mode &= S_IALLUGO; 24961da177e4SLinus Torvalds mode |= S_IFREG; 24971da177e4SLinus Torvalds error = security_inode_create(dir, dentry, mode); 24981da177e4SLinus Torvalds if (error) 24991da177e4SLinus Torvalds return error; 2500312b63fbSAl Viro error = dir->i_op->create(dir, dentry, mode, want_excl); 2501a74574aaSStephen Smalley if (!error) 2502f38aa942SAmy Griffis fsnotify_create(dir, dentry); 25031da177e4SLinus Torvalds return error; 25041da177e4SLinus Torvalds } 25051da177e4SLinus Torvalds 250673d049a4SAl Viro static int may_open(struct path *path, int acc_mode, int flag) 25071da177e4SLinus Torvalds { 25083fb64190SChristoph Hellwig struct dentry *dentry = path->dentry; 25091da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 25101da177e4SLinus Torvalds int error; 25111da177e4SLinus Torvalds 2512bcda7652SAl Viro /* O_PATH? */ 2513bcda7652SAl Viro if (!acc_mode) 2514bcda7652SAl Viro return 0; 2515bcda7652SAl Viro 25161da177e4SLinus Torvalds if (!inode) 25171da177e4SLinus Torvalds return -ENOENT; 25181da177e4SLinus Torvalds 2519c8fe8f30SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 2520c8fe8f30SChristoph Hellwig case S_IFLNK: 25211da177e4SLinus Torvalds return -ELOOP; 2522c8fe8f30SChristoph Hellwig case S_IFDIR: 2523c8fe8f30SChristoph Hellwig if (acc_mode & MAY_WRITE) 25241da177e4SLinus Torvalds return -EISDIR; 2525c8fe8f30SChristoph Hellwig break; 2526c8fe8f30SChristoph Hellwig case S_IFBLK: 2527c8fe8f30SChristoph Hellwig case S_IFCHR: 25283fb64190SChristoph Hellwig if (path->mnt->mnt_flags & MNT_NODEV) 25291da177e4SLinus Torvalds return -EACCES; 2530c8fe8f30SChristoph Hellwig /*FALLTHRU*/ 2531c8fe8f30SChristoph Hellwig case S_IFIFO: 2532c8fe8f30SChristoph Hellwig case S_IFSOCK: 25331da177e4SLinus Torvalds flag &= ~O_TRUNC; 2534c8fe8f30SChristoph Hellwig break; 25354a3fd211SDave Hansen } 2536b41572e9SDave Hansen 25373fb64190SChristoph Hellwig error = inode_permission(inode, acc_mode); 2538b41572e9SDave Hansen if (error) 2539b41572e9SDave Hansen return error; 25406146f0d5SMimi Zohar 25411da177e4SLinus Torvalds /* 25421da177e4SLinus Torvalds * An append-only file must be opened in append mode for writing. 25431da177e4SLinus Torvalds */ 25441da177e4SLinus Torvalds if (IS_APPEND(inode)) { 25458737c930SAl Viro if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) 25467715b521SAl Viro return -EPERM; 25471da177e4SLinus Torvalds if (flag & O_TRUNC) 25487715b521SAl Viro return -EPERM; 25491da177e4SLinus Torvalds } 25501da177e4SLinus Torvalds 25511da177e4SLinus Torvalds /* O_NOATIME can only be set by the owner or superuser */ 25522e149670SSerge E. Hallyn if (flag & O_NOATIME && !inode_owner_or_capable(inode)) 25537715b521SAl Viro return -EPERM; 25541da177e4SLinus Torvalds 2555f3c7691eSJ. Bruce Fields return 0; 25567715b521SAl Viro } 25577715b521SAl Viro 2558e1181ee6SJeff Layton static int handle_truncate(struct file *filp) 25597715b521SAl Viro { 2560e1181ee6SJeff Layton struct path *path = &filp->f_path; 25617715b521SAl Viro struct inode *inode = path->dentry->d_inode; 25627715b521SAl Viro int error = get_write_access(inode); 25631da177e4SLinus Torvalds if (error) 25647715b521SAl Viro return error; 25651da177e4SLinus Torvalds /* 25661da177e4SLinus Torvalds * Refuse to truncate files with mandatory locks held on them. 25671da177e4SLinus Torvalds */ 25681da177e4SLinus Torvalds error = locks_verify_locked(inode); 2569be6d3e56SKentaro Takeda if (!error) 2570ea0d3ab2STetsuo Handa error = security_path_truncate(path); 25711da177e4SLinus Torvalds if (!error) { 25727715b521SAl Viro error = do_truncate(path->dentry, 0, 2573d139d7ffSMiklos Szeredi ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, 2574e1181ee6SJeff Layton filp); 25751da177e4SLinus Torvalds } 25761da177e4SLinus Torvalds put_write_access(inode); 2577acd0c935SMimi Zohar return error; 25781da177e4SLinus Torvalds } 25791da177e4SLinus Torvalds 2580d57999e1SDave Hansen static inline int open_to_namei_flags(int flag) 2581d57999e1SDave Hansen { 25828a5e929dSAl Viro if ((flag & O_ACCMODE) == 3) 25838a5e929dSAl Viro flag--; 2584d57999e1SDave Hansen return flag; 2585d57999e1SDave Hansen } 2586d57999e1SDave Hansen 2587d18e9008SMiklos Szeredi static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode) 2588d18e9008SMiklos Szeredi { 2589d18e9008SMiklos Szeredi int error = security_path_mknod(dir, dentry, mode, 0); 2590d18e9008SMiklos Szeredi if (error) 2591d18e9008SMiklos Szeredi return error; 2592d18e9008SMiklos Szeredi 2593d18e9008SMiklos Szeredi error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC); 2594d18e9008SMiklos Szeredi if (error) 2595d18e9008SMiklos Szeredi return error; 2596d18e9008SMiklos Szeredi 2597d18e9008SMiklos Szeredi return security_inode_create(dir->dentry->d_inode, dentry, mode); 2598d18e9008SMiklos Szeredi } 2599d18e9008SMiklos Szeredi 26001acf0af9SDavid Howells /* 26011acf0af9SDavid Howells * Attempt to atomically look up, create and open a file from a negative 26021acf0af9SDavid Howells * dentry. 26031acf0af9SDavid Howells * 26041acf0af9SDavid Howells * Returns 0 if successful. The file will have been created and attached to 26051acf0af9SDavid Howells * @file by the filesystem calling finish_open(). 26061acf0af9SDavid Howells * 26071acf0af9SDavid Howells * Returns 1 if the file was looked up only or didn't need creating. The 26081acf0af9SDavid Howells * caller will need to perform the open themselves. @path will have been 26091acf0af9SDavid Howells * updated to point to the new dentry. This may be negative. 26101acf0af9SDavid Howells * 26111acf0af9SDavid Howells * Returns an error code otherwise. 26121acf0af9SDavid Howells */ 26132675a4ebSAl Viro static int atomic_open(struct nameidata *nd, struct dentry *dentry, 261430d90494SAl Viro struct path *path, struct file *file, 2615015c3bbcSMiklos Szeredi const struct open_flags *op, 261664894cf8SAl Viro bool got_write, bool need_lookup, 261747237687SAl Viro int *opened) 2618d18e9008SMiklos Szeredi { 2619d18e9008SMiklos Szeredi struct inode *dir = nd->path.dentry->d_inode; 2620d18e9008SMiklos Szeredi unsigned open_flag = open_to_namei_flags(op->open_flag); 2621d18e9008SMiklos Szeredi umode_t mode; 2622d18e9008SMiklos Szeredi int error; 2623d18e9008SMiklos Szeredi int acc_mode; 2624d18e9008SMiklos Szeredi int create_error = 0; 2625d18e9008SMiklos Szeredi struct dentry *const DENTRY_NOT_SET = (void *) -1UL; 2626116cc022SMiklos Szeredi bool excl; 2627d18e9008SMiklos Szeredi 2628d18e9008SMiklos Szeredi BUG_ON(dentry->d_inode); 2629d18e9008SMiklos Szeredi 2630d18e9008SMiklos Szeredi /* Don't create child dentry for a dead directory. */ 2631d18e9008SMiklos Szeredi if (unlikely(IS_DEADDIR(dir))) { 26322675a4ebSAl Viro error = -ENOENT; 2633d18e9008SMiklos Szeredi goto out; 2634d18e9008SMiklos Szeredi } 2635d18e9008SMiklos Szeredi 263662b259d8SMiklos Szeredi mode = op->mode; 2637d18e9008SMiklos Szeredi if ((open_flag & O_CREAT) && !IS_POSIXACL(dir)) 2638d18e9008SMiklos Szeredi mode &= ~current_umask(); 2639d18e9008SMiklos Szeredi 2640116cc022SMiklos Szeredi excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT); 2641116cc022SMiklos Szeredi if (excl) 2642d18e9008SMiklos Szeredi open_flag &= ~O_TRUNC; 2643d18e9008SMiklos Szeredi 2644d18e9008SMiklos Szeredi /* 2645d18e9008SMiklos Szeredi * Checking write permission is tricky, bacuse we don't know if we are 2646d18e9008SMiklos Szeredi * going to actually need it: O_CREAT opens should work as long as the 2647d18e9008SMiklos Szeredi * file exists. But checking existence breaks atomicity. The trick is 2648d18e9008SMiklos Szeredi * to check access and if not granted clear O_CREAT from the flags. 2649d18e9008SMiklos Szeredi * 2650d18e9008SMiklos Szeredi * Another problem is returing the "right" error value (e.g. for an 2651d18e9008SMiklos Szeredi * O_EXCL open we want to return EEXIST not EROFS). 2652d18e9008SMiklos Szeredi */ 265364894cf8SAl Viro if (((open_flag & (O_CREAT | O_TRUNC)) || 265464894cf8SAl Viro (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) { 265564894cf8SAl Viro if (!(open_flag & O_CREAT)) { 2656d18e9008SMiklos Szeredi /* 2657d18e9008SMiklos Szeredi * No O_CREATE -> atomicity not a requirement -> fall 2658d18e9008SMiklos Szeredi * back to lookup + open 2659d18e9008SMiklos Szeredi */ 2660d18e9008SMiklos Szeredi goto no_open; 2661d18e9008SMiklos Szeredi } else if (open_flag & (O_EXCL | O_TRUNC)) { 2662d18e9008SMiklos Szeredi /* Fall back and fail with the right error */ 266364894cf8SAl Viro create_error = -EROFS; 2664d18e9008SMiklos Szeredi goto no_open; 2665d18e9008SMiklos Szeredi } else { 2666d18e9008SMiklos Szeredi /* No side effects, safe to clear O_CREAT */ 266764894cf8SAl Viro create_error = -EROFS; 2668d18e9008SMiklos Szeredi open_flag &= ~O_CREAT; 2669d18e9008SMiklos Szeredi } 2670d18e9008SMiklos Szeredi } 2671d18e9008SMiklos Szeredi 2672d18e9008SMiklos Szeredi if (open_flag & O_CREAT) { 267338227f78SMiklos Szeredi error = may_o_create(&nd->path, dentry, mode); 2674d18e9008SMiklos Szeredi if (error) { 2675d18e9008SMiklos Szeredi create_error = error; 2676d18e9008SMiklos Szeredi if (open_flag & O_EXCL) 2677d18e9008SMiklos Szeredi goto no_open; 2678d18e9008SMiklos Szeredi open_flag &= ~O_CREAT; 2679d18e9008SMiklos Szeredi } 2680d18e9008SMiklos Szeredi } 2681d18e9008SMiklos Szeredi 2682d18e9008SMiklos Szeredi if (nd->flags & LOOKUP_DIRECTORY) 2683d18e9008SMiklos Szeredi open_flag |= O_DIRECTORY; 2684d18e9008SMiklos Szeredi 268530d90494SAl Viro file->f_path.dentry = DENTRY_NOT_SET; 268630d90494SAl Viro file->f_path.mnt = nd->path.mnt; 268730d90494SAl Viro error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode, 268847237687SAl Viro opened); 2689d9585277SAl Viro if (error < 0) { 2690d9585277SAl Viro if (create_error && error == -ENOENT) 2691d9585277SAl Viro error = create_error; 2692d18e9008SMiklos Szeredi goto out; 2693d18e9008SMiklos Szeredi } 2694d18e9008SMiklos Szeredi 2695d9585277SAl Viro if (error) { /* returned 1, that is */ 269630d90494SAl Viro if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { 26972675a4ebSAl Viro error = -EIO; 2698d18e9008SMiklos Szeredi goto out; 2699d18e9008SMiklos Szeredi } 270030d90494SAl Viro if (file->f_path.dentry) { 2701d18e9008SMiklos Szeredi dput(dentry); 270230d90494SAl Viro dentry = file->f_path.dentry; 2703d18e9008SMiklos Szeredi } 270403da633aSAl Viro if (*opened & FILE_CREATED) 270503da633aSAl Viro fsnotify_create(dir, dentry); 270603da633aSAl Viro if (!dentry->d_inode) { 270703da633aSAl Viro WARN_ON(*opened & FILE_CREATED); 270803da633aSAl Viro if (create_error) { 270962b2ce96SSage Weil error = create_error; 271062b2ce96SSage Weil goto out; 271162b2ce96SSage Weil } 271203da633aSAl Viro } else { 271303da633aSAl Viro if (excl && !(*opened & FILE_CREATED)) { 271403da633aSAl Viro error = -EEXIST; 271503da633aSAl Viro goto out; 271603da633aSAl Viro } 271703da633aSAl Viro } 2718d18e9008SMiklos Szeredi goto looked_up; 2719d18e9008SMiklos Szeredi } 2720d18e9008SMiklos Szeredi 2721d18e9008SMiklos Szeredi /* 2722d18e9008SMiklos Szeredi * We didn't have the inode before the open, so check open permission 2723d18e9008SMiklos Szeredi * here. 2724d18e9008SMiklos Szeredi */ 272503da633aSAl Viro acc_mode = op->acc_mode; 272603da633aSAl Viro if (*opened & FILE_CREATED) { 272703da633aSAl Viro WARN_ON(!(open_flag & O_CREAT)); 272803da633aSAl Viro fsnotify_create(dir, dentry); 272903da633aSAl Viro acc_mode = MAY_OPEN; 273003da633aSAl Viro } 27312675a4ebSAl Viro error = may_open(&file->f_path, acc_mode, open_flag); 27322675a4ebSAl Viro if (error) 27332675a4ebSAl Viro fput(file); 2734d18e9008SMiklos Szeredi 2735d18e9008SMiklos Szeredi out: 2736d18e9008SMiklos Szeredi dput(dentry); 27372675a4ebSAl Viro return error; 2738d18e9008SMiklos Szeredi 2739d18e9008SMiklos Szeredi no_open: 2740d18e9008SMiklos Szeredi if (need_lookup) { 274172bd866aSAl Viro dentry = lookup_real(dir, dentry, nd->flags); 2742d18e9008SMiklos Szeredi if (IS_ERR(dentry)) 27432675a4ebSAl Viro return PTR_ERR(dentry); 2744d18e9008SMiklos Szeredi 2745d18e9008SMiklos Szeredi if (create_error) { 2746d18e9008SMiklos Szeredi int open_flag = op->open_flag; 2747d18e9008SMiklos Szeredi 27482675a4ebSAl Viro error = create_error; 2749d18e9008SMiklos Szeredi if ((open_flag & O_EXCL)) { 2750d18e9008SMiklos Szeredi if (!dentry->d_inode) 2751d18e9008SMiklos Szeredi goto out; 2752d18e9008SMiklos Szeredi } else if (!dentry->d_inode) { 2753d18e9008SMiklos Szeredi goto out; 2754d18e9008SMiklos Szeredi } else if ((open_flag & O_TRUNC) && 2755d18e9008SMiklos Szeredi S_ISREG(dentry->d_inode->i_mode)) { 2756d18e9008SMiklos Szeredi goto out; 2757d18e9008SMiklos Szeredi } 2758d18e9008SMiklos Szeredi /* will fail later, go on to get the right error */ 2759d18e9008SMiklos Szeredi } 2760d18e9008SMiklos Szeredi } 2761d18e9008SMiklos Szeredi looked_up: 2762d18e9008SMiklos Szeredi path->dentry = dentry; 2763d18e9008SMiklos Szeredi path->mnt = nd->path.mnt; 27642675a4ebSAl Viro return 1; 2765d18e9008SMiklos Szeredi } 2766d18e9008SMiklos Szeredi 276731e6b01fSNick Piggin /* 27681acf0af9SDavid Howells * Look up and maybe create and open the last component. 2769d58ffd35SMiklos Szeredi * 2770d58ffd35SMiklos Szeredi * Must be called with i_mutex held on parent. 2771d58ffd35SMiklos Szeredi * 27721acf0af9SDavid Howells * Returns 0 if the file was successfully atomically created (if necessary) and 27731acf0af9SDavid Howells * opened. In this case the file will be returned attached to @file. 27741acf0af9SDavid Howells * 27751acf0af9SDavid Howells * Returns 1 if the file was not completely opened at this time, though lookups 27761acf0af9SDavid Howells * and creations will have been performed and the dentry returned in @path will 27771acf0af9SDavid Howells * be positive upon return if O_CREAT was specified. If O_CREAT wasn't 27781acf0af9SDavid Howells * specified then a negative dentry may be returned. 27791acf0af9SDavid Howells * 27801acf0af9SDavid Howells * An error code is returned otherwise. 27811acf0af9SDavid Howells * 27821acf0af9SDavid Howells * FILE_CREATE will be set in @*opened if the dentry was created and will be 27831acf0af9SDavid Howells * cleared otherwise prior to returning. 2784d58ffd35SMiklos Szeredi */ 27852675a4ebSAl Viro static int lookup_open(struct nameidata *nd, struct path *path, 278630d90494SAl Viro struct file *file, 2787d58ffd35SMiklos Szeredi const struct open_flags *op, 278864894cf8SAl Viro bool got_write, int *opened) 2789d58ffd35SMiklos Szeredi { 2790d58ffd35SMiklos Szeredi struct dentry *dir = nd->path.dentry; 279154ef4872SMiklos Szeredi struct inode *dir_inode = dir->d_inode; 2792d58ffd35SMiklos Szeredi struct dentry *dentry; 2793d58ffd35SMiklos Szeredi int error; 279454ef4872SMiklos Szeredi bool need_lookup; 2795d58ffd35SMiklos Szeredi 279647237687SAl Viro *opened &= ~FILE_CREATED; 2797201f956eSAl Viro dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup); 2798d58ffd35SMiklos Szeredi if (IS_ERR(dentry)) 27992675a4ebSAl Viro return PTR_ERR(dentry); 2800d58ffd35SMiklos Szeredi 2801d18e9008SMiklos Szeredi /* Cached positive dentry: will open in f_op->open */ 2802d18e9008SMiklos Szeredi if (!need_lookup && dentry->d_inode) 2803d18e9008SMiklos Szeredi goto out_no_open; 2804d18e9008SMiklos Szeredi 2805d18e9008SMiklos Szeredi if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) { 280664894cf8SAl Viro return atomic_open(nd, dentry, path, file, op, got_write, 280747237687SAl Viro need_lookup, opened); 2808d18e9008SMiklos Szeredi } 2809d18e9008SMiklos Szeredi 281054ef4872SMiklos Szeredi if (need_lookup) { 281154ef4872SMiklos Szeredi BUG_ON(dentry->d_inode); 281254ef4872SMiklos Szeredi 281372bd866aSAl Viro dentry = lookup_real(dir_inode, dentry, nd->flags); 281454ef4872SMiklos Szeredi if (IS_ERR(dentry)) 28152675a4ebSAl Viro return PTR_ERR(dentry); 281654ef4872SMiklos Szeredi } 281754ef4872SMiklos Szeredi 2818d58ffd35SMiklos Szeredi /* Negative dentry, just create the file */ 2819d58ffd35SMiklos Szeredi if (!dentry->d_inode && (op->open_flag & O_CREAT)) { 2820d58ffd35SMiklos Szeredi umode_t mode = op->mode; 2821d58ffd35SMiklos Szeredi if (!IS_POSIXACL(dir->d_inode)) 2822d58ffd35SMiklos Szeredi mode &= ~current_umask(); 2823d58ffd35SMiklos Szeredi /* 2824d58ffd35SMiklos Szeredi * This write is needed to ensure that a 2825d58ffd35SMiklos Szeredi * rw->ro transition does not occur between 2826d58ffd35SMiklos Szeredi * the time when the file is created and when 2827d58ffd35SMiklos Szeredi * a permanent write count is taken through 2828015c3bbcSMiklos Szeredi * the 'struct file' in finish_open(). 2829d58ffd35SMiklos Szeredi */ 283064894cf8SAl Viro if (!got_write) { 283164894cf8SAl Viro error = -EROFS; 2832d58ffd35SMiklos Szeredi goto out_dput; 283364894cf8SAl Viro } 283447237687SAl Viro *opened |= FILE_CREATED; 2835d58ffd35SMiklos Szeredi error = security_path_mknod(&nd->path, dentry, mode, 0); 2836d58ffd35SMiklos Szeredi if (error) 2837d58ffd35SMiklos Szeredi goto out_dput; 2838312b63fbSAl Viro error = vfs_create(dir->d_inode, dentry, mode, 2839312b63fbSAl Viro nd->flags & LOOKUP_EXCL); 2840d58ffd35SMiklos Szeredi if (error) 2841d58ffd35SMiklos Szeredi goto out_dput; 2842d58ffd35SMiklos Szeredi } 2843d18e9008SMiklos Szeredi out_no_open: 2844d58ffd35SMiklos Szeredi path->dentry = dentry; 2845d58ffd35SMiklos Szeredi path->mnt = nd->path.mnt; 28462675a4ebSAl Viro return 1; 2847d58ffd35SMiklos Szeredi 2848d58ffd35SMiklos Szeredi out_dput: 2849d58ffd35SMiklos Szeredi dput(dentry); 28502675a4ebSAl Viro return error; 2851d58ffd35SMiklos Szeredi } 2852d58ffd35SMiklos Szeredi 2853d58ffd35SMiklos Szeredi /* 2854fe2d35ffSAl Viro * Handle the last step of open() 285531e6b01fSNick Piggin */ 28562675a4ebSAl Viro static int do_last(struct nameidata *nd, struct path *path, 285730d90494SAl Viro struct file *file, const struct open_flags *op, 2858669abf4eSJeff Layton int *opened, struct filename *name) 2859fb1cc555SAl Viro { 2860a1e28038SAl Viro struct dentry *dir = nd->path.dentry; 2861ca344a89SAl Viro int open_flag = op->open_flag; 286277d660a8SMiklos Szeredi bool will_truncate = (open_flag & O_TRUNC) != 0; 286364894cf8SAl Viro bool got_write = false; 2864bcda7652SAl Viro int acc_mode = op->acc_mode; 2865a1eb3315SMiklos Szeredi struct inode *inode; 286677d660a8SMiklos Szeredi bool symlink_ok = false; 286716b1c1cdSMiklos Szeredi struct path save_parent = { .dentry = NULL, .mnt = NULL }; 286816b1c1cdSMiklos Szeredi bool retried = false; 286916c2cd71SAl Viro int error; 2870fb1cc555SAl Viro 2871c3e380b0SAl Viro nd->flags &= ~LOOKUP_PARENT; 2872c3e380b0SAl Viro nd->flags |= op->intent; 2873c3e380b0SAl Viro 2874bc77daa7SAl Viro if (nd->last_type != LAST_NORM) { 2875fe2d35ffSAl Viro error = handle_dots(nd, nd->last_type); 2876fe2d35ffSAl Viro if (error) 28772675a4ebSAl Viro return error; 2878e83db167SMiklos Szeredi goto finish_open; 28791f36f774SAl Viro } 2880a2c36b45SAl Viro 2881ca344a89SAl Viro if (!(open_flag & O_CREAT)) { 2882fe2d35ffSAl Viro if (nd->last.name[nd->last.len]) 2883fe2d35ffSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 2884bcda7652SAl Viro if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW)) 288577d660a8SMiklos Szeredi symlink_ok = true; 2886fe2d35ffSAl Viro /* we _can_ be in RCU mode here */ 2887e97cdc87SAl Viro error = lookup_fast(nd, path, &inode); 288871574865SMiklos Szeredi if (likely(!error)) 288971574865SMiklos Szeredi goto finish_lookup; 289071574865SMiklos Szeredi 2891ce57dfc1SAl Viro if (error < 0) 28922675a4ebSAl Viro goto out; 2893a1eb3315SMiklos Szeredi 289437d7fffcSMiklos Szeredi BUG_ON(nd->inode != dir->d_inode); 2895b6183df7SMiklos Szeredi } else { 2896fe2d35ffSAl Viro /* create side of things */ 2897a3fbbde7SAl Viro /* 2898b6183df7SMiklos Szeredi * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED 2899b6183df7SMiklos Szeredi * has been cleared when we got to the last component we are 2900b6183df7SMiklos Szeredi * about to look up 2901a3fbbde7SAl Viro */ 29029f1fafeeSAl Viro error = complete_walk(nd); 29039f1fafeeSAl Viro if (error) 29042675a4ebSAl Viro return error; 2905fe2d35ffSAl Viro 290633e2208aSJeff Layton audit_inode(name, dir, LOOKUP_PARENT); 290716c2cd71SAl Viro error = -EISDIR; 29081f36f774SAl Viro /* trailing slashes? */ 290931e6b01fSNick Piggin if (nd->last.name[nd->last.len]) 29102675a4ebSAl Viro goto out; 2911b6183df7SMiklos Szeredi } 29121f36f774SAl Viro 291316b1c1cdSMiklos Szeredi retry_lookup: 291464894cf8SAl Viro if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { 291564894cf8SAl Viro error = mnt_want_write(nd->path.mnt); 291664894cf8SAl Viro if (!error) 291764894cf8SAl Viro got_write = true; 291864894cf8SAl Viro /* 291964894cf8SAl Viro * do _not_ fail yet - we might not need that or fail with 292064894cf8SAl Viro * a different error; let lookup_open() decide; we'll be 292164894cf8SAl Viro * dropping this one anyway. 292264894cf8SAl Viro */ 292364894cf8SAl Viro } 2924a1e28038SAl Viro mutex_lock(&dir->d_inode->i_mutex); 292564894cf8SAl Viro error = lookup_open(nd, path, file, op, got_write, opened); 2926fb1cc555SAl Viro mutex_unlock(&dir->d_inode->i_mutex); 2927fb1cc555SAl Viro 29282675a4ebSAl Viro if (error <= 0) { 29292675a4ebSAl Viro if (error) 2930d58ffd35SMiklos Szeredi goto out; 29316c0d46c4SAl Viro 293247237687SAl Viro if ((*opened & FILE_CREATED) || 2933496ad9aaSAl Viro !S_ISREG(file_inode(file)->i_mode)) 293477d660a8SMiklos Szeredi will_truncate = false; 2935d18e9008SMiklos Szeredi 2936adb5c247SJeff Layton audit_inode(name, file->f_path.dentry, 0); 2937d18e9008SMiklos Szeredi goto opened; 2938d18e9008SMiklos Szeredi } 2939d18e9008SMiklos Szeredi 294047237687SAl Viro if (*opened & FILE_CREATED) { 29419b44f1b3SAl Viro /* Don't check for write permission, don't truncate */ 2942ca344a89SAl Viro open_flag &= ~O_TRUNC; 294377d660a8SMiklos Szeredi will_truncate = false; 2944bcda7652SAl Viro acc_mode = MAY_OPEN; 2945d58ffd35SMiklos Szeredi path_to_nameidata(path, nd); 2946e83db167SMiklos Szeredi goto finish_open_created; 2947fb1cc555SAl Viro } 2948fb1cc555SAl Viro 2949fb1cc555SAl Viro /* 29503134f37eSJeff Layton * create/update audit record if it already exists. 2951fb1cc555SAl Viro */ 2952b18825a7SDavid Howells if (d_is_positive(path->dentry)) 2953adb5c247SJeff Layton audit_inode(name, path->dentry, 0); 2954fb1cc555SAl Viro 2955d18e9008SMiklos Szeredi /* 2956d18e9008SMiklos Szeredi * If atomic_open() acquired write access it is dropped now due to 2957d18e9008SMiklos Szeredi * possible mount and symlink following (this might be optimized away if 2958d18e9008SMiklos Szeredi * necessary...) 2959d18e9008SMiklos Szeredi */ 296064894cf8SAl Viro if (got_write) { 2961d18e9008SMiklos Szeredi mnt_drop_write(nd->path.mnt); 296264894cf8SAl Viro got_write = false; 2963d18e9008SMiklos Szeredi } 2964d18e9008SMiklos Szeredi 2965fb1cc555SAl Viro error = -EEXIST; 2966f8310c59SAl Viro if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT)) 2967fb1cc555SAl Viro goto exit_dput; 2968fb1cc555SAl Viro 29699875cf80SDavid Howells error = follow_managed(path, nd->flags); 29709875cf80SDavid Howells if (error < 0) 2971fb1cc555SAl Viro goto exit_dput; 2972fb1cc555SAl Viro 2973a3fbbde7SAl Viro if (error) 2974a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 2975a3fbbde7SAl Viro 2976decf3400SMiklos Szeredi BUG_ON(nd->flags & LOOKUP_RCU); 2977decf3400SMiklos Szeredi inode = path->dentry->d_inode; 29785f5daac1SMiklos Szeredi finish_lookup: 29795f5daac1SMiklos Szeredi /* we _can_ be in RCU mode here */ 2980fb1cc555SAl Viro error = -ENOENT; 2981b18825a7SDavid Howells if (d_is_negative(path->dentry)) { 298254c33e7fSMiklos Szeredi path_to_nameidata(path, nd); 29832675a4ebSAl Viro goto out; 298454c33e7fSMiklos Szeredi } 29859e67f361SAl Viro 2986b18825a7SDavid Howells if (should_follow_link(path->dentry, !symlink_ok)) { 2987d45ea867SMiklos Szeredi if (nd->flags & LOOKUP_RCU) { 2988d45ea867SMiklos Szeredi if (unlikely(unlazy_walk(nd, path->dentry))) { 2989d45ea867SMiklos Szeredi error = -ECHILD; 29902675a4ebSAl Viro goto out; 2991d45ea867SMiklos Szeredi } 2992d45ea867SMiklos Szeredi } 2993d45ea867SMiklos Szeredi BUG_ON(inode != path->dentry->d_inode); 29942675a4ebSAl Viro return 1; 2995d45ea867SMiklos Szeredi } 2996fb1cc555SAl Viro 299716b1c1cdSMiklos Szeredi if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) { 2998fb1cc555SAl Viro path_to_nameidata(path, nd); 299916b1c1cdSMiklos Szeredi } else { 300016b1c1cdSMiklos Szeredi save_parent.dentry = nd->path.dentry; 300116b1c1cdSMiklos Szeredi save_parent.mnt = mntget(path->mnt); 300216b1c1cdSMiklos Szeredi nd->path.dentry = path->dentry; 300316b1c1cdSMiklos Szeredi 300416b1c1cdSMiklos Szeredi } 3005decf3400SMiklos Szeredi nd->inode = inode; 3006a3fbbde7SAl Viro /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ 3007bc77daa7SAl Viro finish_open: 3008a3fbbde7SAl Viro error = complete_walk(nd); 300916b1c1cdSMiklos Szeredi if (error) { 301016b1c1cdSMiklos Szeredi path_put(&save_parent); 30112675a4ebSAl Viro return error; 301216b1c1cdSMiklos Szeredi } 3013bc77daa7SAl Viro audit_inode(name, nd->path.dentry, 0); 3014fb1cc555SAl Viro error = -EISDIR; 3015b18825a7SDavid Howells if ((open_flag & O_CREAT) && 3016b18825a7SDavid Howells (d_is_directory(nd->path.dentry) || d_is_autodir(nd->path.dentry))) 30172675a4ebSAl Viro goto out; 3018af2f5542SMiklos Szeredi error = -ENOTDIR; 3019b18825a7SDavid Howells if ((nd->flags & LOOKUP_DIRECTORY) && !d_is_directory(nd->path.dentry)) 30202675a4ebSAl Viro goto out; 30216c0d46c4SAl Viro if (!S_ISREG(nd->inode->i_mode)) 302277d660a8SMiklos Szeredi will_truncate = false; 30236c0d46c4SAl Viro 30240f9d1a10SAl Viro if (will_truncate) { 30250f9d1a10SAl Viro error = mnt_want_write(nd->path.mnt); 30260f9d1a10SAl Viro if (error) 30272675a4ebSAl Viro goto out; 302864894cf8SAl Viro got_write = true; 30290f9d1a10SAl Viro } 3030e83db167SMiklos Szeredi finish_open_created: 3031bcda7652SAl Viro error = may_open(&nd->path, acc_mode, open_flag); 3032ca344a89SAl Viro if (error) 30332675a4ebSAl Viro goto out; 303430d90494SAl Viro file->f_path.mnt = nd->path.mnt; 303530d90494SAl Viro error = finish_open(file, nd->path.dentry, NULL, opened); 303630d90494SAl Viro if (error) { 303730d90494SAl Viro if (error == -EOPENSTALE) 3038f60dc3dbSMiklos Szeredi goto stale_open; 3039015c3bbcSMiklos Szeredi goto out; 3040f60dc3dbSMiklos Szeredi } 3041a8277b9bSMiklos Szeredi opened: 30422675a4ebSAl Viro error = open_check_o_direct(file); 3043015c3bbcSMiklos Szeredi if (error) 3044015c3bbcSMiklos Szeredi goto exit_fput; 30452675a4ebSAl Viro error = ima_file_check(file, op->acc_mode); 3046aa4caadbSMiklos Szeredi if (error) 3047aa4caadbSMiklos Szeredi goto exit_fput; 3048aa4caadbSMiklos Szeredi 30490f9d1a10SAl Viro if (will_truncate) { 30502675a4ebSAl Viro error = handle_truncate(file); 3051aa4caadbSMiklos Szeredi if (error) 3052aa4caadbSMiklos Szeredi goto exit_fput; 30530f9d1a10SAl Viro } 3054ca344a89SAl Viro out: 305564894cf8SAl Viro if (got_write) 30560f9d1a10SAl Viro mnt_drop_write(nd->path.mnt); 305716b1c1cdSMiklos Szeredi path_put(&save_parent); 3058e276ae67SMiklos Szeredi terminate_walk(nd); 30592675a4ebSAl Viro return error; 3060fb1cc555SAl Viro 3061fb1cc555SAl Viro exit_dput: 3062fb1cc555SAl Viro path_put_conditional(path, nd); 3063ca344a89SAl Viro goto out; 3064015c3bbcSMiklos Szeredi exit_fput: 30652675a4ebSAl Viro fput(file); 30662675a4ebSAl Viro goto out; 3067015c3bbcSMiklos Szeredi 3068f60dc3dbSMiklos Szeredi stale_open: 3069f60dc3dbSMiklos Szeredi /* If no saved parent or already retried then can't retry */ 3070f60dc3dbSMiklos Szeredi if (!save_parent.dentry || retried) 3071f60dc3dbSMiklos Szeredi goto out; 3072f60dc3dbSMiklos Szeredi 3073f60dc3dbSMiklos Szeredi BUG_ON(save_parent.dentry != dir); 3074f60dc3dbSMiklos Szeredi path_put(&nd->path); 3075f60dc3dbSMiklos Szeredi nd->path = save_parent; 3076f60dc3dbSMiklos Szeredi nd->inode = dir->d_inode; 3077f60dc3dbSMiklos Szeredi save_parent.mnt = NULL; 3078f60dc3dbSMiklos Szeredi save_parent.dentry = NULL; 307964894cf8SAl Viro if (got_write) { 3080f60dc3dbSMiklos Szeredi mnt_drop_write(nd->path.mnt); 308164894cf8SAl Viro got_write = false; 3082f60dc3dbSMiklos Szeredi } 3083f60dc3dbSMiklos Szeredi retried = true; 3084f60dc3dbSMiklos Szeredi goto retry_lookup; 3085fb1cc555SAl Viro } 3086fb1cc555SAl Viro 308760545d0dSAl Viro static int do_tmpfile(int dfd, struct filename *pathname, 308860545d0dSAl Viro struct nameidata *nd, int flags, 308960545d0dSAl Viro const struct open_flags *op, 309060545d0dSAl Viro struct file *file, int *opened) 309160545d0dSAl Viro { 309260545d0dSAl Viro static const struct qstr name = QSTR_INIT("/", 1); 309360545d0dSAl Viro struct dentry *dentry, *child; 309460545d0dSAl Viro struct inode *dir; 309560545d0dSAl Viro int error = path_lookupat(dfd, pathname->name, 309660545d0dSAl Viro flags | LOOKUP_DIRECTORY, nd); 309760545d0dSAl Viro if (unlikely(error)) 309860545d0dSAl Viro return error; 309960545d0dSAl Viro error = mnt_want_write(nd->path.mnt); 310060545d0dSAl Viro if (unlikely(error)) 310160545d0dSAl Viro goto out; 310260545d0dSAl Viro /* we want directory to be writable */ 310360545d0dSAl Viro error = inode_permission(nd->inode, MAY_WRITE | MAY_EXEC); 310460545d0dSAl Viro if (error) 310560545d0dSAl Viro goto out2; 310660545d0dSAl Viro dentry = nd->path.dentry; 310760545d0dSAl Viro dir = dentry->d_inode; 310860545d0dSAl Viro if (!dir->i_op->tmpfile) { 310960545d0dSAl Viro error = -EOPNOTSUPP; 311060545d0dSAl Viro goto out2; 311160545d0dSAl Viro } 311260545d0dSAl Viro child = d_alloc(dentry, &name); 311360545d0dSAl Viro if (unlikely(!child)) { 311460545d0dSAl Viro error = -ENOMEM; 311560545d0dSAl Viro goto out2; 311660545d0dSAl Viro } 311760545d0dSAl Viro nd->flags &= ~LOOKUP_DIRECTORY; 311860545d0dSAl Viro nd->flags |= op->intent; 311960545d0dSAl Viro dput(nd->path.dentry); 312060545d0dSAl Viro nd->path.dentry = child; 312160545d0dSAl Viro error = dir->i_op->tmpfile(dir, nd->path.dentry, op->mode); 312260545d0dSAl Viro if (error) 312360545d0dSAl Viro goto out2; 312460545d0dSAl Viro audit_inode(pathname, nd->path.dentry, 0); 312560545d0dSAl Viro error = may_open(&nd->path, op->acc_mode, op->open_flag); 312660545d0dSAl Viro if (error) 312760545d0dSAl Viro goto out2; 312860545d0dSAl Viro file->f_path.mnt = nd->path.mnt; 312960545d0dSAl Viro error = finish_open(file, nd->path.dentry, NULL, opened); 313060545d0dSAl Viro if (error) 313160545d0dSAl Viro goto out2; 313260545d0dSAl Viro error = open_check_o_direct(file); 3133f4e0c30cSAl Viro if (error) { 313460545d0dSAl Viro fput(file); 3135f4e0c30cSAl Viro } else if (!(op->open_flag & O_EXCL)) { 3136f4e0c30cSAl Viro struct inode *inode = file_inode(file); 3137f4e0c30cSAl Viro spin_lock(&inode->i_lock); 3138f4e0c30cSAl Viro inode->i_state |= I_LINKABLE; 3139f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 3140f4e0c30cSAl Viro } 314160545d0dSAl Viro out2: 314260545d0dSAl Viro mnt_drop_write(nd->path.mnt); 314360545d0dSAl Viro out: 314460545d0dSAl Viro path_put(&nd->path); 314560545d0dSAl Viro return error; 314660545d0dSAl Viro } 314760545d0dSAl Viro 3148669abf4eSJeff Layton static struct file *path_openat(int dfd, struct filename *pathname, 314973d049a4SAl Viro struct nameidata *nd, const struct open_flags *op, int flags) 31501da177e4SLinus Torvalds { 3151fe2d35ffSAl Viro struct file *base = NULL; 315230d90494SAl Viro struct file *file; 31539850c056SAl Viro struct path path; 315447237687SAl Viro int opened = 0; 315513aab428SAl Viro int error; 315631e6b01fSNick Piggin 315730d90494SAl Viro file = get_empty_filp(); 31581afc99beSAl Viro if (IS_ERR(file)) 31591afc99beSAl Viro return file; 316031e6b01fSNick Piggin 316130d90494SAl Viro file->f_flags = op->open_flag; 316231e6b01fSNick Piggin 3163bb458c64SAl Viro if (unlikely(file->f_flags & __O_TMPFILE)) { 316460545d0dSAl Viro error = do_tmpfile(dfd, pathname, nd, flags, op, file, &opened); 316560545d0dSAl Viro goto out; 316660545d0dSAl Viro } 316760545d0dSAl Viro 3168669abf4eSJeff Layton error = path_init(dfd, pathname->name, flags | LOOKUP_PARENT, nd, &base); 316931e6b01fSNick Piggin if (unlikely(error)) 31702675a4ebSAl Viro goto out; 317131e6b01fSNick Piggin 3172fe2d35ffSAl Viro current->total_link_count = 0; 3173669abf4eSJeff Layton error = link_path_walk(pathname->name, nd); 317431e6b01fSNick Piggin if (unlikely(error)) 31752675a4ebSAl Viro goto out; 31761da177e4SLinus Torvalds 31772675a4ebSAl Viro error = do_last(nd, &path, file, op, &opened, pathname); 31782675a4ebSAl Viro while (unlikely(error > 0)) { /* trailing symlink */ 31797b9337aaSNick Piggin struct path link = path; 3180def4af30SAl Viro void *cookie; 3181574197e0SAl Viro if (!(nd->flags & LOOKUP_FOLLOW)) { 318273d049a4SAl Viro path_put_conditional(&path, nd); 318373d049a4SAl Viro path_put(&nd->path); 31842675a4ebSAl Viro error = -ELOOP; 318540b39136SAl Viro break; 318640b39136SAl Viro } 3187800179c9SKees Cook error = may_follow_link(&link, nd); 3188800179c9SKees Cook if (unlikely(error)) 3189800179c9SKees Cook break; 319073d049a4SAl Viro nd->flags |= LOOKUP_PARENT; 319173d049a4SAl Viro nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); 3192574197e0SAl Viro error = follow_link(&link, nd, &cookie); 3193c3e380b0SAl Viro if (unlikely(error)) 31942675a4ebSAl Viro break; 31952675a4ebSAl Viro error = do_last(nd, &path, file, op, &opened, pathname); 3196574197e0SAl Viro put_link(nd, &link, cookie); 3197806b681cSAl Viro } 319810fa8e62SAl Viro out: 319973d049a4SAl Viro if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) 320073d049a4SAl Viro path_put(&nd->root); 3201fe2d35ffSAl Viro if (base) 3202fe2d35ffSAl Viro fput(base); 32032675a4ebSAl Viro if (!(opened & FILE_OPENED)) { 32042675a4ebSAl Viro BUG_ON(!error); 320530d90494SAl Viro put_filp(file); 3206015c3bbcSMiklos Szeredi } 32072675a4ebSAl Viro if (unlikely(error)) { 32082675a4ebSAl Viro if (error == -EOPENSTALE) { 32092675a4ebSAl Viro if (flags & LOOKUP_RCU) 32102675a4ebSAl Viro error = -ECHILD; 32112675a4ebSAl Viro else 32122675a4ebSAl Viro error = -ESTALE; 32132675a4ebSAl Viro } 32142675a4ebSAl Viro file = ERR_PTR(error); 32152675a4ebSAl Viro } 32162675a4ebSAl Viro return file; 3217de459215SKirill Korotaev } 32181da177e4SLinus Torvalds 3219669abf4eSJeff Layton struct file *do_filp_open(int dfd, struct filename *pathname, 3220f9652e10SAl Viro const struct open_flags *op) 322113aab428SAl Viro { 322273d049a4SAl Viro struct nameidata nd; 3223f9652e10SAl Viro int flags = op->lookup_flags; 322413aab428SAl Viro struct file *filp; 322513aab428SAl Viro 322673d049a4SAl Viro filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU); 322713aab428SAl Viro if (unlikely(filp == ERR_PTR(-ECHILD))) 322873d049a4SAl Viro filp = path_openat(dfd, pathname, &nd, op, flags); 322913aab428SAl Viro if (unlikely(filp == ERR_PTR(-ESTALE))) 323073d049a4SAl Viro filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL); 323113aab428SAl Viro return filp; 323213aab428SAl Viro } 323313aab428SAl Viro 323473d049a4SAl Viro struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, 3235f9652e10SAl Viro const char *name, const struct open_flags *op) 323673d049a4SAl Viro { 323773d049a4SAl Viro struct nameidata nd; 323873d049a4SAl Viro struct file *file; 3239669abf4eSJeff Layton struct filename filename = { .name = name }; 3240f9652e10SAl Viro int flags = op->lookup_flags | LOOKUP_ROOT; 324173d049a4SAl Viro 324273d049a4SAl Viro nd.root.mnt = mnt; 324373d049a4SAl Viro nd.root.dentry = dentry; 324473d049a4SAl Viro 3245b18825a7SDavid Howells if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN) 324673d049a4SAl Viro return ERR_PTR(-ELOOP); 324773d049a4SAl Viro 3248669abf4eSJeff Layton file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_RCU); 324973d049a4SAl Viro if (unlikely(file == ERR_PTR(-ECHILD))) 3250669abf4eSJeff Layton file = path_openat(-1, &filename, &nd, op, flags); 325173d049a4SAl Viro if (unlikely(file == ERR_PTR(-ESTALE))) 3252669abf4eSJeff Layton file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_REVAL); 325373d049a4SAl Viro return file; 325473d049a4SAl Viro } 325573d049a4SAl Viro 32561ac12b4bSJeff Layton struct dentry *kern_path_create(int dfd, const char *pathname, 32571ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 32581da177e4SLinus Torvalds { 3259c663e5d8SChristoph Hellwig struct dentry *dentry = ERR_PTR(-EEXIST); 3260ed75e95dSAl Viro struct nameidata nd; 3261c30dabfeSJan Kara int err2; 32621ac12b4bSJeff Layton int error; 32631ac12b4bSJeff Layton bool is_dir = (lookup_flags & LOOKUP_DIRECTORY); 32641ac12b4bSJeff Layton 32651ac12b4bSJeff Layton /* 32661ac12b4bSJeff Layton * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any 32671ac12b4bSJeff Layton * other flags passed in are ignored! 32681ac12b4bSJeff Layton */ 32691ac12b4bSJeff Layton lookup_flags &= LOOKUP_REVAL; 32701ac12b4bSJeff Layton 32711ac12b4bSJeff Layton error = do_path_lookup(dfd, pathname, LOOKUP_PARENT|lookup_flags, &nd); 3272ed75e95dSAl Viro if (error) 3273ed75e95dSAl Viro return ERR_PTR(error); 32741da177e4SLinus Torvalds 3275c663e5d8SChristoph Hellwig /* 3276c663e5d8SChristoph Hellwig * Yucky last component or no last component at all? 3277c663e5d8SChristoph Hellwig * (foo/., foo/.., /////) 3278c663e5d8SChristoph Hellwig */ 3279ed75e95dSAl Viro if (nd.last_type != LAST_NORM) 3280ed75e95dSAl Viro goto out; 3281ed75e95dSAl Viro nd.flags &= ~LOOKUP_PARENT; 3282ed75e95dSAl Viro nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL; 3283c663e5d8SChristoph Hellwig 3284c30dabfeSJan Kara /* don't fail immediately if it's r/o, at least try to report other errors */ 3285c30dabfeSJan Kara err2 = mnt_want_write(nd.path.mnt); 3286c663e5d8SChristoph Hellwig /* 3287c663e5d8SChristoph Hellwig * Do the final lookup. 3288c663e5d8SChristoph Hellwig */ 3289ed75e95dSAl Viro mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 3290ed75e95dSAl Viro dentry = lookup_hash(&nd); 32911da177e4SLinus Torvalds if (IS_ERR(dentry)) 3292a8104a9fSAl Viro goto unlock; 3293c663e5d8SChristoph Hellwig 3294a8104a9fSAl Viro error = -EEXIST; 3295b18825a7SDavid Howells if (d_is_positive(dentry)) 3296a8104a9fSAl Viro goto fail; 3297b18825a7SDavid Howells 3298c663e5d8SChristoph Hellwig /* 3299c663e5d8SChristoph Hellwig * Special case - lookup gave negative, but... we had foo/bar/ 3300c663e5d8SChristoph Hellwig * From the vfs_mknod() POV we just have a negative dentry - 3301c663e5d8SChristoph Hellwig * all is fine. Let's be bastards - you had / on the end, you've 3302c663e5d8SChristoph Hellwig * been asking for (non-existent) directory. -ENOENT for you. 3303c663e5d8SChristoph Hellwig */ 3304ed75e95dSAl Viro if (unlikely(!is_dir && nd.last.name[nd.last.len])) { 3305a8104a9fSAl Viro error = -ENOENT; 3306ed75e95dSAl Viro goto fail; 3307e9baf6e5SAl Viro } 3308c30dabfeSJan Kara if (unlikely(err2)) { 3309c30dabfeSJan Kara error = err2; 3310a8104a9fSAl Viro goto fail; 3311c30dabfeSJan Kara } 3312ed75e95dSAl Viro *path = nd.path; 3313e9baf6e5SAl Viro return dentry; 33141da177e4SLinus Torvalds fail: 3315a8104a9fSAl Viro dput(dentry); 3316a8104a9fSAl Viro dentry = ERR_PTR(error); 3317a8104a9fSAl Viro unlock: 3318dae6ad8fSAl Viro mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 3319c30dabfeSJan Kara if (!err2) 3320c30dabfeSJan Kara mnt_drop_write(nd.path.mnt); 3321ed75e95dSAl Viro out: 3322dae6ad8fSAl Viro path_put(&nd.path); 3323ed75e95dSAl Viro return dentry; 3324dae6ad8fSAl Viro } 3325dae6ad8fSAl Viro EXPORT_SYMBOL(kern_path_create); 3326dae6ad8fSAl Viro 3327921a1650SAl Viro void done_path_create(struct path *path, struct dentry *dentry) 3328921a1650SAl Viro { 3329921a1650SAl Viro dput(dentry); 3330921a1650SAl Viro mutex_unlock(&path->dentry->d_inode->i_mutex); 3331a8104a9fSAl Viro mnt_drop_write(path->mnt); 3332921a1650SAl Viro path_put(path); 3333921a1650SAl Viro } 3334921a1650SAl Viro EXPORT_SYMBOL(done_path_create); 3335921a1650SAl Viro 33361ac12b4bSJeff Layton struct dentry *user_path_create(int dfd, const char __user *pathname, 33371ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 3338dae6ad8fSAl Viro { 333991a27b2aSJeff Layton struct filename *tmp = getname(pathname); 3340dae6ad8fSAl Viro struct dentry *res; 3341dae6ad8fSAl Viro if (IS_ERR(tmp)) 3342dae6ad8fSAl Viro return ERR_CAST(tmp); 33431ac12b4bSJeff Layton res = kern_path_create(dfd, tmp->name, path, lookup_flags); 3344dae6ad8fSAl Viro putname(tmp); 3345dae6ad8fSAl Viro return res; 3346dae6ad8fSAl Viro } 3347dae6ad8fSAl Viro EXPORT_SYMBOL(user_path_create); 3348dae6ad8fSAl Viro 33491a67aafbSAl Viro int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) 33501da177e4SLinus Torvalds { 3351a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 33521da177e4SLinus Torvalds 33531da177e4SLinus Torvalds if (error) 33541da177e4SLinus Torvalds return error; 33551da177e4SLinus Torvalds 3356975d6b39SEric W. Biederman if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) 33571da177e4SLinus Torvalds return -EPERM; 33581da177e4SLinus Torvalds 3359acfa4380SAl Viro if (!dir->i_op->mknod) 33601da177e4SLinus Torvalds return -EPERM; 33611da177e4SLinus Torvalds 336208ce5f16SSerge E. Hallyn error = devcgroup_inode_mknod(mode, dev); 336308ce5f16SSerge E. Hallyn if (error) 336408ce5f16SSerge E. Hallyn return error; 336508ce5f16SSerge E. Hallyn 33661da177e4SLinus Torvalds error = security_inode_mknod(dir, dentry, mode, dev); 33671da177e4SLinus Torvalds if (error) 33681da177e4SLinus Torvalds return error; 33691da177e4SLinus Torvalds 33701da177e4SLinus Torvalds error = dir->i_op->mknod(dir, dentry, mode, dev); 3371a74574aaSStephen Smalley if (!error) 3372f38aa942SAmy Griffis fsnotify_create(dir, dentry); 33731da177e4SLinus Torvalds return error; 33741da177e4SLinus Torvalds } 33751da177e4SLinus Torvalds 3376f69aac00SAl Viro static int may_mknod(umode_t mode) 3377463c3197SDave Hansen { 3378463c3197SDave Hansen switch (mode & S_IFMT) { 3379463c3197SDave Hansen case S_IFREG: 3380463c3197SDave Hansen case S_IFCHR: 3381463c3197SDave Hansen case S_IFBLK: 3382463c3197SDave Hansen case S_IFIFO: 3383463c3197SDave Hansen case S_IFSOCK: 3384463c3197SDave Hansen case 0: /* zero mode translates to S_IFREG */ 3385463c3197SDave Hansen return 0; 3386463c3197SDave Hansen case S_IFDIR: 3387463c3197SDave Hansen return -EPERM; 3388463c3197SDave Hansen default: 3389463c3197SDave Hansen return -EINVAL; 3390463c3197SDave Hansen } 3391463c3197SDave Hansen } 3392463c3197SDave Hansen 33938208a22bSAl Viro SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, 33942e4d0924SHeiko Carstens unsigned, dev) 33951da177e4SLinus Torvalds { 33961da177e4SLinus Torvalds struct dentry *dentry; 3397dae6ad8fSAl Viro struct path path; 3398dae6ad8fSAl Viro int error; 3399972567f1SJeff Layton unsigned int lookup_flags = 0; 34001da177e4SLinus Torvalds 34018e4bfca1SAl Viro error = may_mknod(mode); 34028e4bfca1SAl Viro if (error) 34038e4bfca1SAl Viro return error; 3404972567f1SJeff Layton retry: 3405972567f1SJeff Layton dentry = user_path_create(dfd, filename, &path, lookup_flags); 3406dae6ad8fSAl Viro if (IS_ERR(dentry)) 3407dae6ad8fSAl Viro return PTR_ERR(dentry); 34082ad94ae6SAl Viro 3409dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3410ce3b0f8dSAl Viro mode &= ~current_umask(); 3411dae6ad8fSAl Viro error = security_path_mknod(&path, dentry, mode, dev); 3412be6d3e56SKentaro Takeda if (error) 3413a8104a9fSAl Viro goto out; 34141da177e4SLinus Torvalds switch (mode & S_IFMT) { 34151da177e4SLinus Torvalds case 0: case S_IFREG: 3416312b63fbSAl Viro error = vfs_create(path.dentry->d_inode,dentry,mode,true); 34171da177e4SLinus Torvalds break; 34181da177e4SLinus Torvalds case S_IFCHR: case S_IFBLK: 3419dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode, 34201da177e4SLinus Torvalds new_decode_dev(dev)); 34211da177e4SLinus Torvalds break; 34221da177e4SLinus Torvalds case S_IFIFO: case S_IFSOCK: 3423dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); 34241da177e4SLinus Torvalds break; 34251da177e4SLinus Torvalds } 3426a8104a9fSAl Viro out: 3427921a1650SAl Viro done_path_create(&path, dentry); 3428972567f1SJeff Layton if (retry_estale(error, lookup_flags)) { 3429972567f1SJeff Layton lookup_flags |= LOOKUP_REVAL; 3430972567f1SJeff Layton goto retry; 3431972567f1SJeff Layton } 34321da177e4SLinus Torvalds return error; 34331da177e4SLinus Torvalds } 34341da177e4SLinus Torvalds 34358208a22bSAl Viro SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) 34365590ff0dSUlrich Drepper { 34375590ff0dSUlrich Drepper return sys_mknodat(AT_FDCWD, filename, mode, dev); 34385590ff0dSUlrich Drepper } 34395590ff0dSUlrich Drepper 344018bb1db3SAl Viro int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 34411da177e4SLinus Torvalds { 3442a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 34438de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 34441da177e4SLinus Torvalds 34451da177e4SLinus Torvalds if (error) 34461da177e4SLinus Torvalds return error; 34471da177e4SLinus Torvalds 3448acfa4380SAl Viro if (!dir->i_op->mkdir) 34491da177e4SLinus Torvalds return -EPERM; 34501da177e4SLinus Torvalds 34511da177e4SLinus Torvalds mode &= (S_IRWXUGO|S_ISVTX); 34521da177e4SLinus Torvalds error = security_inode_mkdir(dir, dentry, mode); 34531da177e4SLinus Torvalds if (error) 34541da177e4SLinus Torvalds return error; 34551da177e4SLinus Torvalds 34568de52778SAl Viro if (max_links && dir->i_nlink >= max_links) 34578de52778SAl Viro return -EMLINK; 34588de52778SAl Viro 34591da177e4SLinus Torvalds error = dir->i_op->mkdir(dir, dentry, mode); 3460a74574aaSStephen Smalley if (!error) 3461f38aa942SAmy Griffis fsnotify_mkdir(dir, dentry); 34621da177e4SLinus Torvalds return error; 34631da177e4SLinus Torvalds } 34641da177e4SLinus Torvalds 3465a218d0fdSAl Viro SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) 34661da177e4SLinus Torvalds { 34676902d925SDave Hansen struct dentry *dentry; 3468dae6ad8fSAl Viro struct path path; 3469dae6ad8fSAl Viro int error; 3470b76d8b82SJeff Layton unsigned int lookup_flags = LOOKUP_DIRECTORY; 34711da177e4SLinus Torvalds 3472b76d8b82SJeff Layton retry: 3473b76d8b82SJeff Layton dentry = user_path_create(dfd, pathname, &path, lookup_flags); 34746902d925SDave Hansen if (IS_ERR(dentry)) 3475dae6ad8fSAl Viro return PTR_ERR(dentry); 34766902d925SDave Hansen 3477dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3478ce3b0f8dSAl Viro mode &= ~current_umask(); 3479dae6ad8fSAl Viro error = security_path_mkdir(&path, dentry, mode); 3480a8104a9fSAl Viro if (!error) 3481dae6ad8fSAl Viro error = vfs_mkdir(path.dentry->d_inode, dentry, mode); 3482921a1650SAl Viro done_path_create(&path, dentry); 3483b76d8b82SJeff Layton if (retry_estale(error, lookup_flags)) { 3484b76d8b82SJeff Layton lookup_flags |= LOOKUP_REVAL; 3485b76d8b82SJeff Layton goto retry; 3486b76d8b82SJeff Layton } 34871da177e4SLinus Torvalds return error; 34881da177e4SLinus Torvalds } 34891da177e4SLinus Torvalds 3490a218d0fdSAl Viro SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) 34915590ff0dSUlrich Drepper { 34925590ff0dSUlrich Drepper return sys_mkdirat(AT_FDCWD, pathname, mode); 34935590ff0dSUlrich Drepper } 34945590ff0dSUlrich Drepper 34951da177e4SLinus Torvalds /* 3496a71905f0SSage Weil * The dentry_unhash() helper will try to drop the dentry early: we 3497c0d02594SJ. Bruce Fields * should have a usage count of 1 if we're the only user of this 3498a71905f0SSage Weil * dentry, and if that is true (possibly after pruning the dcache), 3499a71905f0SSage Weil * then we drop the dentry now. 35001da177e4SLinus Torvalds * 35011da177e4SLinus Torvalds * A low-level filesystem can, if it choses, legally 35021da177e4SLinus Torvalds * do a 35031da177e4SLinus Torvalds * 35041da177e4SLinus Torvalds * if (!d_unhashed(dentry)) 35051da177e4SLinus Torvalds * return -EBUSY; 35061da177e4SLinus Torvalds * 35071da177e4SLinus Torvalds * if it cannot handle the case of removing a directory 35081da177e4SLinus Torvalds * that is still in use by something else.. 35091da177e4SLinus Torvalds */ 35101da177e4SLinus Torvalds void dentry_unhash(struct dentry *dentry) 35111da177e4SLinus Torvalds { 35121da177e4SLinus Torvalds shrink_dcache_parent(dentry); 35131da177e4SLinus Torvalds spin_lock(&dentry->d_lock); 351498474236SWaiman Long if (dentry->d_lockref.count == 1) 35151da177e4SLinus Torvalds __d_drop(dentry); 35161da177e4SLinus Torvalds spin_unlock(&dentry->d_lock); 35171da177e4SLinus Torvalds } 35181da177e4SLinus Torvalds 35191da177e4SLinus Torvalds int vfs_rmdir(struct inode *dir, struct dentry *dentry) 35201da177e4SLinus Torvalds { 35211da177e4SLinus Torvalds int error = may_delete(dir, dentry, 1); 35221da177e4SLinus Torvalds 35231da177e4SLinus Torvalds if (error) 35241da177e4SLinus Torvalds return error; 35251da177e4SLinus Torvalds 3526acfa4380SAl Viro if (!dir->i_op->rmdir) 35271da177e4SLinus Torvalds return -EPERM; 35281da177e4SLinus Torvalds 35291d2ef590SAl Viro dget(dentry); 35301b1dcc1bSJes Sorensen mutex_lock(&dentry->d_inode->i_mutex); 3531912dbc15SSage Weil 35321da177e4SLinus Torvalds error = -EBUSY; 3533912dbc15SSage Weil if (d_mountpoint(dentry)) 3534912dbc15SSage Weil goto out; 3535912dbc15SSage Weil 35361da177e4SLinus Torvalds error = security_inode_rmdir(dir, dentry); 3537912dbc15SSage Weil if (error) 3538912dbc15SSage Weil goto out; 3539912dbc15SSage Weil 35403cebde24SSage Weil shrink_dcache_parent(dentry); 35411da177e4SLinus Torvalds error = dir->i_op->rmdir(dir, dentry); 3542912dbc15SSage Weil if (error) 3543912dbc15SSage Weil goto out; 3544912dbc15SSage Weil 35451da177e4SLinus Torvalds dentry->d_inode->i_flags |= S_DEAD; 3546d83c49f3SAl Viro dont_mount(dentry); 35471da177e4SLinus Torvalds 3548912dbc15SSage Weil out: 3549912dbc15SSage Weil mutex_unlock(&dentry->d_inode->i_mutex); 35501d2ef590SAl Viro dput(dentry); 3551912dbc15SSage Weil if (!error) 3552912dbc15SSage Weil d_delete(dentry); 35531da177e4SLinus Torvalds return error; 35541da177e4SLinus Torvalds } 35551da177e4SLinus Torvalds 35565590ff0dSUlrich Drepper static long do_rmdir(int dfd, const char __user *pathname) 35571da177e4SLinus Torvalds { 35581da177e4SLinus Torvalds int error = 0; 355991a27b2aSJeff Layton struct filename *name; 35601da177e4SLinus Torvalds struct dentry *dentry; 35611da177e4SLinus Torvalds struct nameidata nd; 3562c6ee9206SJeff Layton unsigned int lookup_flags = 0; 3563c6ee9206SJeff Layton retry: 3564c6ee9206SJeff Layton name = user_path_parent(dfd, pathname, &nd, lookup_flags); 356591a27b2aSJeff Layton if (IS_ERR(name)) 356691a27b2aSJeff Layton return PTR_ERR(name); 35671da177e4SLinus Torvalds 35681da177e4SLinus Torvalds switch(nd.last_type) { 35691da177e4SLinus Torvalds case LAST_DOTDOT: 35701da177e4SLinus Torvalds error = -ENOTEMPTY; 35711da177e4SLinus Torvalds goto exit1; 35721da177e4SLinus Torvalds case LAST_DOT: 35731da177e4SLinus Torvalds error = -EINVAL; 35741da177e4SLinus Torvalds goto exit1; 35751da177e4SLinus Torvalds case LAST_ROOT: 35761da177e4SLinus Torvalds error = -EBUSY; 35771da177e4SLinus Torvalds goto exit1; 35781da177e4SLinus Torvalds } 35790612d9fbSOGAWA Hirofumi 35800612d9fbSOGAWA Hirofumi nd.flags &= ~LOOKUP_PARENT; 3581c30dabfeSJan Kara error = mnt_want_write(nd.path.mnt); 3582c30dabfeSJan Kara if (error) 3583c30dabfeSJan Kara goto exit1; 35840612d9fbSOGAWA Hirofumi 35854ac91378SJan Blunck mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 358649705b77SChristoph Hellwig dentry = lookup_hash(&nd); 35871da177e4SLinus Torvalds error = PTR_ERR(dentry); 35886902d925SDave Hansen if (IS_ERR(dentry)) 35896902d925SDave Hansen goto exit2; 3590e6bc45d6STheodore Ts'o if (!dentry->d_inode) { 3591e6bc45d6STheodore Ts'o error = -ENOENT; 3592e6bc45d6STheodore Ts'o goto exit3; 3593e6bc45d6STheodore Ts'o } 3594be6d3e56SKentaro Takeda error = security_path_rmdir(&nd.path, dentry); 3595be6d3e56SKentaro Takeda if (error) 3596c30dabfeSJan Kara goto exit3; 35974ac91378SJan Blunck error = vfs_rmdir(nd.path.dentry->d_inode, dentry); 35980622753bSDave Hansen exit3: 35991da177e4SLinus Torvalds dput(dentry); 36006902d925SDave Hansen exit2: 36014ac91378SJan Blunck mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 3602c30dabfeSJan Kara mnt_drop_write(nd.path.mnt); 36031da177e4SLinus Torvalds exit1: 36041d957f9bSJan Blunck path_put(&nd.path); 36051da177e4SLinus Torvalds putname(name); 3606c6ee9206SJeff Layton if (retry_estale(error, lookup_flags)) { 3607c6ee9206SJeff Layton lookup_flags |= LOOKUP_REVAL; 3608c6ee9206SJeff Layton goto retry; 3609c6ee9206SJeff Layton } 36101da177e4SLinus Torvalds return error; 36111da177e4SLinus Torvalds } 36121da177e4SLinus Torvalds 36133cdad428SHeiko Carstens SYSCALL_DEFINE1(rmdir, const char __user *, pathname) 36145590ff0dSUlrich Drepper { 36155590ff0dSUlrich Drepper return do_rmdir(AT_FDCWD, pathname); 36165590ff0dSUlrich Drepper } 36175590ff0dSUlrich Drepper 36181da177e4SLinus Torvalds int vfs_unlink(struct inode *dir, struct dentry *dentry) 36191da177e4SLinus Torvalds { 36201da177e4SLinus Torvalds int error = may_delete(dir, dentry, 0); 36211da177e4SLinus Torvalds 36221da177e4SLinus Torvalds if (error) 36231da177e4SLinus Torvalds return error; 36241da177e4SLinus Torvalds 3625acfa4380SAl Viro if (!dir->i_op->unlink) 36261da177e4SLinus Torvalds return -EPERM; 36271da177e4SLinus Torvalds 36281b1dcc1bSJes Sorensen mutex_lock(&dentry->d_inode->i_mutex); 36291da177e4SLinus Torvalds if (d_mountpoint(dentry)) 36301da177e4SLinus Torvalds error = -EBUSY; 36311da177e4SLinus Torvalds else { 36321da177e4SLinus Torvalds error = security_inode_unlink(dir, dentry); 3633bec1052eSAl Viro if (!error) { 36341da177e4SLinus Torvalds error = dir->i_op->unlink(dir, dentry); 3635bec1052eSAl Viro if (!error) 3636d83c49f3SAl Viro dont_mount(dentry); 3637bec1052eSAl Viro } 36381da177e4SLinus Torvalds } 36391b1dcc1bSJes Sorensen mutex_unlock(&dentry->d_inode->i_mutex); 36401da177e4SLinus Torvalds 36411da177e4SLinus Torvalds /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 36421da177e4SLinus Torvalds if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { 3643ece95912SJan Kara fsnotify_link_count(dentry->d_inode); 36441da177e4SLinus Torvalds d_delete(dentry); 36451da177e4SLinus Torvalds } 36460eeca283SRobert Love 36471da177e4SLinus Torvalds return error; 36481da177e4SLinus Torvalds } 36491da177e4SLinus Torvalds 36501da177e4SLinus Torvalds /* 36511da177e4SLinus Torvalds * Make sure that the actual truncation of the file will occur outside its 36521b1dcc1bSJes Sorensen * directory's i_mutex. Truncate can take a long time if there is a lot of 36531da177e4SLinus Torvalds * writeout happening, and we don't want to prevent access to the directory 36541da177e4SLinus Torvalds * while waiting on the I/O. 36551da177e4SLinus Torvalds */ 36565590ff0dSUlrich Drepper static long do_unlinkat(int dfd, const char __user *pathname) 36571da177e4SLinus Torvalds { 36582ad94ae6SAl Viro int error; 365991a27b2aSJeff Layton struct filename *name; 36601da177e4SLinus Torvalds struct dentry *dentry; 36611da177e4SLinus Torvalds struct nameidata nd; 36621da177e4SLinus Torvalds struct inode *inode = NULL; 36635d18f813SJeff Layton unsigned int lookup_flags = 0; 36645d18f813SJeff Layton retry: 36655d18f813SJeff Layton name = user_path_parent(dfd, pathname, &nd, lookup_flags); 366691a27b2aSJeff Layton if (IS_ERR(name)) 366791a27b2aSJeff Layton return PTR_ERR(name); 36682ad94ae6SAl Viro 36691da177e4SLinus Torvalds error = -EISDIR; 36701da177e4SLinus Torvalds if (nd.last_type != LAST_NORM) 36711da177e4SLinus Torvalds goto exit1; 36720612d9fbSOGAWA Hirofumi 36730612d9fbSOGAWA Hirofumi nd.flags &= ~LOOKUP_PARENT; 3674c30dabfeSJan Kara error = mnt_want_write(nd.path.mnt); 3675c30dabfeSJan Kara if (error) 3676c30dabfeSJan Kara goto exit1; 36770612d9fbSOGAWA Hirofumi 36784ac91378SJan Blunck mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 367949705b77SChristoph Hellwig dentry = lookup_hash(&nd); 36801da177e4SLinus Torvalds error = PTR_ERR(dentry); 36811da177e4SLinus Torvalds if (!IS_ERR(dentry)) { 36821da177e4SLinus Torvalds /* Why not before? Because we want correct error value */ 368350338b88STörök Edwin if (nd.last.name[nd.last.len]) 368450338b88STörök Edwin goto slashes; 36851da177e4SLinus Torvalds inode = dentry->d_inode; 3686b18825a7SDavid Howells if (d_is_negative(dentry)) 3687e6bc45d6STheodore Ts'o goto slashes; 36887de9c6eeSAl Viro ihold(inode); 3689be6d3e56SKentaro Takeda error = security_path_unlink(&nd.path, dentry); 3690be6d3e56SKentaro Takeda if (error) 3691c30dabfeSJan Kara goto exit2; 36924ac91378SJan Blunck error = vfs_unlink(nd.path.dentry->d_inode, dentry); 36931da177e4SLinus Torvalds exit2: 36941da177e4SLinus Torvalds dput(dentry); 36951da177e4SLinus Torvalds } 36964ac91378SJan Blunck mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 36971da177e4SLinus Torvalds if (inode) 36981da177e4SLinus Torvalds iput(inode); /* truncate the inode here */ 3699c30dabfeSJan Kara mnt_drop_write(nd.path.mnt); 37001da177e4SLinus Torvalds exit1: 37011d957f9bSJan Blunck path_put(&nd.path); 37021da177e4SLinus Torvalds putname(name); 37035d18f813SJeff Layton if (retry_estale(error, lookup_flags)) { 37045d18f813SJeff Layton lookup_flags |= LOOKUP_REVAL; 37055d18f813SJeff Layton inode = NULL; 37065d18f813SJeff Layton goto retry; 37075d18f813SJeff Layton } 37081da177e4SLinus Torvalds return error; 37091da177e4SLinus Torvalds 37101da177e4SLinus Torvalds slashes: 3711b18825a7SDavid Howells if (d_is_negative(dentry)) 3712b18825a7SDavid Howells error = -ENOENT; 3713b18825a7SDavid Howells else if (d_is_directory(dentry) || d_is_autodir(dentry)) 3714b18825a7SDavid Howells error = -EISDIR; 3715b18825a7SDavid Howells else 3716b18825a7SDavid Howells error = -ENOTDIR; 37171da177e4SLinus Torvalds goto exit2; 37181da177e4SLinus Torvalds } 37191da177e4SLinus Torvalds 37202e4d0924SHeiko Carstens SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) 37215590ff0dSUlrich Drepper { 37225590ff0dSUlrich Drepper if ((flag & ~AT_REMOVEDIR) != 0) 37235590ff0dSUlrich Drepper return -EINVAL; 37245590ff0dSUlrich Drepper 37255590ff0dSUlrich Drepper if (flag & AT_REMOVEDIR) 37265590ff0dSUlrich Drepper return do_rmdir(dfd, pathname); 37275590ff0dSUlrich Drepper 37285590ff0dSUlrich Drepper return do_unlinkat(dfd, pathname); 37295590ff0dSUlrich Drepper } 37305590ff0dSUlrich Drepper 37313480b257SHeiko Carstens SYSCALL_DEFINE1(unlink, const char __user *, pathname) 37325590ff0dSUlrich Drepper { 37335590ff0dSUlrich Drepper return do_unlinkat(AT_FDCWD, pathname); 37345590ff0dSUlrich Drepper } 37355590ff0dSUlrich Drepper 3736db2e747bSMiklos Szeredi int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) 37371da177e4SLinus Torvalds { 3738a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 37391da177e4SLinus Torvalds 37401da177e4SLinus Torvalds if (error) 37411da177e4SLinus Torvalds return error; 37421da177e4SLinus Torvalds 3743acfa4380SAl Viro if (!dir->i_op->symlink) 37441da177e4SLinus Torvalds return -EPERM; 37451da177e4SLinus Torvalds 37461da177e4SLinus Torvalds error = security_inode_symlink(dir, dentry, oldname); 37471da177e4SLinus Torvalds if (error) 37481da177e4SLinus Torvalds return error; 37491da177e4SLinus Torvalds 37501da177e4SLinus Torvalds error = dir->i_op->symlink(dir, dentry, oldname); 3751a74574aaSStephen Smalley if (!error) 3752f38aa942SAmy Griffis fsnotify_create(dir, dentry); 37531da177e4SLinus Torvalds return error; 37541da177e4SLinus Torvalds } 37551da177e4SLinus Torvalds 37562e4d0924SHeiko Carstens SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, 37572e4d0924SHeiko Carstens int, newdfd, const char __user *, newname) 37581da177e4SLinus Torvalds { 37592ad94ae6SAl Viro int error; 376091a27b2aSJeff Layton struct filename *from; 37616902d925SDave Hansen struct dentry *dentry; 3762dae6ad8fSAl Viro struct path path; 3763f46d3567SJeff Layton unsigned int lookup_flags = 0; 37641da177e4SLinus Torvalds 37651da177e4SLinus Torvalds from = getname(oldname); 37661da177e4SLinus Torvalds if (IS_ERR(from)) 37671da177e4SLinus Torvalds return PTR_ERR(from); 3768f46d3567SJeff Layton retry: 3769f46d3567SJeff Layton dentry = user_path_create(newdfd, newname, &path, lookup_flags); 37701da177e4SLinus Torvalds error = PTR_ERR(dentry); 37716902d925SDave Hansen if (IS_ERR(dentry)) 3772dae6ad8fSAl Viro goto out_putname; 37736902d925SDave Hansen 377491a27b2aSJeff Layton error = security_path_symlink(&path, dentry, from->name); 3775a8104a9fSAl Viro if (!error) 377691a27b2aSJeff Layton error = vfs_symlink(path.dentry->d_inode, dentry, from->name); 3777921a1650SAl Viro done_path_create(&path, dentry); 3778f46d3567SJeff Layton if (retry_estale(error, lookup_flags)) { 3779f46d3567SJeff Layton lookup_flags |= LOOKUP_REVAL; 3780f46d3567SJeff Layton goto retry; 3781f46d3567SJeff Layton } 37826902d925SDave Hansen out_putname: 37831da177e4SLinus Torvalds putname(from); 37841da177e4SLinus Torvalds return error; 37851da177e4SLinus Torvalds } 37861da177e4SLinus Torvalds 37873480b257SHeiko Carstens SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) 37885590ff0dSUlrich Drepper { 37895590ff0dSUlrich Drepper return sys_symlinkat(oldname, AT_FDCWD, newname); 37905590ff0dSUlrich Drepper } 37915590ff0dSUlrich Drepper 37921da177e4SLinus Torvalds int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry) 37931da177e4SLinus Torvalds { 37941da177e4SLinus Torvalds struct inode *inode = old_dentry->d_inode; 37958de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 37961da177e4SLinus Torvalds int error; 37971da177e4SLinus Torvalds 37981da177e4SLinus Torvalds if (!inode) 37991da177e4SLinus Torvalds return -ENOENT; 38001da177e4SLinus Torvalds 3801a95164d9SMiklos Szeredi error = may_create(dir, new_dentry); 38021da177e4SLinus Torvalds if (error) 38031da177e4SLinus Torvalds return error; 38041da177e4SLinus Torvalds 38051da177e4SLinus Torvalds if (dir->i_sb != inode->i_sb) 38061da177e4SLinus Torvalds return -EXDEV; 38071da177e4SLinus Torvalds 38081da177e4SLinus Torvalds /* 38091da177e4SLinus Torvalds * A link to an append-only or immutable file cannot be created. 38101da177e4SLinus Torvalds */ 38111da177e4SLinus Torvalds if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 38121da177e4SLinus Torvalds return -EPERM; 3813acfa4380SAl Viro if (!dir->i_op->link) 38141da177e4SLinus Torvalds return -EPERM; 38157e79eedbSTetsuo Handa if (S_ISDIR(inode->i_mode)) 38161da177e4SLinus Torvalds return -EPERM; 38171da177e4SLinus Torvalds 38181da177e4SLinus Torvalds error = security_inode_link(old_dentry, dir, new_dentry); 38191da177e4SLinus Torvalds if (error) 38201da177e4SLinus Torvalds return error; 38211da177e4SLinus Torvalds 38227e79eedbSTetsuo Handa mutex_lock(&inode->i_mutex); 3823aae8a97dSAneesh Kumar K.V /* Make sure we don't allow creating hardlink to an unlinked file */ 3824f4e0c30cSAl Viro if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) 3825aae8a97dSAneesh Kumar K.V error = -ENOENT; 38268de52778SAl Viro else if (max_links && inode->i_nlink >= max_links) 38278de52778SAl Viro error = -EMLINK; 3828aae8a97dSAneesh Kumar K.V else 38291da177e4SLinus Torvalds error = dir->i_op->link(old_dentry, dir, new_dentry); 3830f4e0c30cSAl Viro 3831f4e0c30cSAl Viro if (!error && (inode->i_state & I_LINKABLE)) { 3832f4e0c30cSAl Viro spin_lock(&inode->i_lock); 3833f4e0c30cSAl Viro inode->i_state &= ~I_LINKABLE; 3834f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 3835f4e0c30cSAl Viro } 38367e79eedbSTetsuo Handa mutex_unlock(&inode->i_mutex); 3837e31e14ecSStephen Smalley if (!error) 38387e79eedbSTetsuo Handa fsnotify_link(dir, inode, new_dentry); 38391da177e4SLinus Torvalds return error; 38401da177e4SLinus Torvalds } 38411da177e4SLinus Torvalds 38421da177e4SLinus Torvalds /* 38431da177e4SLinus Torvalds * Hardlinks are often used in delicate situations. We avoid 38441da177e4SLinus Torvalds * security-related surprises by not following symlinks on the 38451da177e4SLinus Torvalds * newname. --KAB 38461da177e4SLinus Torvalds * 38471da177e4SLinus Torvalds * We don't follow them on the oldname either to be compatible 38481da177e4SLinus Torvalds * with linux 2.0, and to avoid hard-linking to directories 38491da177e4SLinus Torvalds * and other special files. --ADM 38501da177e4SLinus Torvalds */ 38512e4d0924SHeiko Carstens SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, 38522e4d0924SHeiko Carstens int, newdfd, const char __user *, newname, int, flags) 38531da177e4SLinus Torvalds { 38541da177e4SLinus Torvalds struct dentry *new_dentry; 3855dae6ad8fSAl Viro struct path old_path, new_path; 385611a7b371SAneesh Kumar K.V int how = 0; 38571da177e4SLinus Torvalds int error; 38581da177e4SLinus Torvalds 385911a7b371SAneesh Kumar K.V if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) 3860c04030e1SUlrich Drepper return -EINVAL; 386111a7b371SAneesh Kumar K.V /* 3862f0cc6ffbSLinus Torvalds * To use null names we require CAP_DAC_READ_SEARCH 3863f0cc6ffbSLinus Torvalds * This ensures that not everyone will be able to create 3864f0cc6ffbSLinus Torvalds * handlink using the passed filedescriptor. 386511a7b371SAneesh Kumar K.V */ 3866f0cc6ffbSLinus Torvalds if (flags & AT_EMPTY_PATH) { 3867f0cc6ffbSLinus Torvalds if (!capable(CAP_DAC_READ_SEARCH)) 3868f0cc6ffbSLinus Torvalds return -ENOENT; 386911a7b371SAneesh Kumar K.V how = LOOKUP_EMPTY; 3870f0cc6ffbSLinus Torvalds } 3871c04030e1SUlrich Drepper 387211a7b371SAneesh Kumar K.V if (flags & AT_SYMLINK_FOLLOW) 387311a7b371SAneesh Kumar K.V how |= LOOKUP_FOLLOW; 3874442e31caSJeff Layton retry: 387511a7b371SAneesh Kumar K.V error = user_path_at(olddfd, oldname, how, &old_path); 38761da177e4SLinus Torvalds if (error) 38772ad94ae6SAl Viro return error; 38782ad94ae6SAl Viro 3879442e31caSJeff Layton new_dentry = user_path_create(newdfd, newname, &new_path, 3880442e31caSJeff Layton (how & LOOKUP_REVAL)); 38811da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 38826902d925SDave Hansen if (IS_ERR(new_dentry)) 3883dae6ad8fSAl Viro goto out; 3884dae6ad8fSAl Viro 3885dae6ad8fSAl Viro error = -EXDEV; 3886dae6ad8fSAl Viro if (old_path.mnt != new_path.mnt) 3887dae6ad8fSAl Viro goto out_dput; 3888800179c9SKees Cook error = may_linkat(&old_path); 3889800179c9SKees Cook if (unlikely(error)) 3890800179c9SKees Cook goto out_dput; 3891dae6ad8fSAl Viro error = security_path_link(old_path.dentry, &new_path, new_dentry); 3892be6d3e56SKentaro Takeda if (error) 3893a8104a9fSAl Viro goto out_dput; 3894dae6ad8fSAl Viro error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry); 389575c3f29dSDave Hansen out_dput: 3896921a1650SAl Viro done_path_create(&new_path, new_dentry); 3897442e31caSJeff Layton if (retry_estale(error, how)) { 3898442e31caSJeff Layton how |= LOOKUP_REVAL; 3899442e31caSJeff Layton goto retry; 3900442e31caSJeff Layton } 39011da177e4SLinus Torvalds out: 39022d8f3038SAl Viro path_put(&old_path); 39031da177e4SLinus Torvalds 39041da177e4SLinus Torvalds return error; 39051da177e4SLinus Torvalds } 39061da177e4SLinus Torvalds 39073480b257SHeiko Carstens SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) 39085590ff0dSUlrich Drepper { 3909c04030e1SUlrich Drepper return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 39105590ff0dSUlrich Drepper } 39115590ff0dSUlrich Drepper 39121da177e4SLinus Torvalds /* 39131da177e4SLinus Torvalds * The worst of all namespace operations - renaming directory. "Perverted" 39141da177e4SLinus Torvalds * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 39151da177e4SLinus Torvalds * Problems: 39161da177e4SLinus Torvalds * a) we can get into loop creation. Check is done in is_subdir(). 39171da177e4SLinus Torvalds * b) race potential - two innocent renames can create a loop together. 39181da177e4SLinus Torvalds * That's where 4.4 screws up. Current fix: serialization on 3919a11f3a05SArjan van de Ven * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 39201da177e4SLinus Torvalds * story. 39211da177e4SLinus Torvalds * c) we have to lock _three_ objects - parents and victim (if it exists). 39221b1dcc1bSJes Sorensen * And that - after we got ->i_mutex on parents (until then we don't know 39231da177e4SLinus Torvalds * whether the target exists). Solution: try to be smart with locking 39241da177e4SLinus Torvalds * order for inodes. We rely on the fact that tree topology may change 3925a11f3a05SArjan van de Ven * only under ->s_vfs_rename_mutex _and_ that parent of the object we 39261da177e4SLinus Torvalds * move will be locked. Thus we can rank directories by the tree 39271da177e4SLinus Torvalds * (ancestors first) and rank all non-directories after them. 39281da177e4SLinus Torvalds * That works since everybody except rename does "lock parent, lookup, 3929a11f3a05SArjan van de Ven * lock child" and rename is under ->s_vfs_rename_mutex. 39301da177e4SLinus Torvalds * HOWEVER, it relies on the assumption that any object with ->lookup() 39311da177e4SLinus Torvalds * has no more than 1 dentry. If "hybrid" objects will ever appear, 39321da177e4SLinus Torvalds * we'd better make sure that there's no link(2) for them. 3933e4eaac06SSage Weil * d) conversion from fhandle to dentry may come in the wrong moment - when 39341b1dcc1bSJes Sorensen * we are removing the target. Solution: we will have to grab ->i_mutex 39351da177e4SLinus Torvalds * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 3936c41b20e7SAdam Buchbinder * ->i_mutex on parents, which works but leads to some truly excessive 39371da177e4SLinus Torvalds * locking]. 39381da177e4SLinus Torvalds */ 393975c96f85SAdrian Bunk static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry, 39401da177e4SLinus Torvalds struct inode *new_dir, struct dentry *new_dentry) 39411da177e4SLinus Torvalds { 39421da177e4SLinus Torvalds int error = 0; 39439055cba7SSage Weil struct inode *target = new_dentry->d_inode; 39448de52778SAl Viro unsigned max_links = new_dir->i_sb->s_max_links; 39451da177e4SLinus Torvalds 39461da177e4SLinus Torvalds /* 39471da177e4SLinus Torvalds * If we are going to change the parent - check write permissions, 39481da177e4SLinus Torvalds * we'll need to flip '..'. 39491da177e4SLinus Torvalds */ 39501da177e4SLinus Torvalds if (new_dir != old_dir) { 3951f419a2e3SAl Viro error = inode_permission(old_dentry->d_inode, MAY_WRITE); 39521da177e4SLinus Torvalds if (error) 39531da177e4SLinus Torvalds return error; 39541da177e4SLinus Torvalds } 39551da177e4SLinus Torvalds 39561da177e4SLinus Torvalds error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); 39571da177e4SLinus Torvalds if (error) 39581da177e4SLinus Torvalds return error; 39591da177e4SLinus Torvalds 39601d2ef590SAl Viro dget(new_dentry); 3961d83c49f3SAl Viro if (target) 39621b1dcc1bSJes Sorensen mutex_lock(&target->i_mutex); 39639055cba7SSage Weil 39641da177e4SLinus Torvalds error = -EBUSY; 39659055cba7SSage Weil if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry)) 39669055cba7SSage Weil goto out; 39679055cba7SSage Weil 39688de52778SAl Viro error = -EMLINK; 39698de52778SAl Viro if (max_links && !target && new_dir != old_dir && 39708de52778SAl Viro new_dir->i_nlink >= max_links) 39718de52778SAl Viro goto out; 39728de52778SAl Viro 39733cebde24SSage Weil if (target) 39743cebde24SSage Weil shrink_dcache_parent(new_dentry); 39751da177e4SLinus Torvalds error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); 39769055cba7SSage Weil if (error) 39779055cba7SSage Weil goto out; 39789055cba7SSage Weil 39791da177e4SLinus Torvalds if (target) { 39801da177e4SLinus Torvalds target->i_flags |= S_DEAD; 3981d83c49f3SAl Viro dont_mount(new_dentry); 3982d83c49f3SAl Viro } 39839055cba7SSage Weil out: 39849055cba7SSage Weil if (target) 39851b1dcc1bSJes Sorensen mutex_unlock(&target->i_mutex); 39861d2ef590SAl Viro dput(new_dentry); 3987e31e14ecSStephen Smalley if (!error) 3988349457ccSMark Fasheh if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) 39891da177e4SLinus Torvalds d_move(old_dentry,new_dentry); 39901da177e4SLinus Torvalds return error; 39911da177e4SLinus Torvalds } 39921da177e4SLinus Torvalds 399375c96f85SAdrian Bunk static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry, 39941da177e4SLinus Torvalds struct inode *new_dir, struct dentry *new_dentry) 39951da177e4SLinus Torvalds { 399651892bbbSSage Weil struct inode *target = new_dentry->d_inode; 39971da177e4SLinus Torvalds int error; 39981da177e4SLinus Torvalds 39991da177e4SLinus Torvalds error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); 40001da177e4SLinus Torvalds if (error) 40011da177e4SLinus Torvalds return error; 40021da177e4SLinus Torvalds 40031da177e4SLinus Torvalds dget(new_dentry); 40041da177e4SLinus Torvalds if (target) 40051b1dcc1bSJes Sorensen mutex_lock(&target->i_mutex); 400651892bbbSSage Weil 40071da177e4SLinus Torvalds error = -EBUSY; 400851892bbbSSage Weil if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry)) 400951892bbbSSage Weil goto out; 401051892bbbSSage Weil 40111da177e4SLinus Torvalds error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); 401251892bbbSSage Weil if (error) 401351892bbbSSage Weil goto out; 401451892bbbSSage Weil 4015bec1052eSAl Viro if (target) 4016d83c49f3SAl Viro dont_mount(new_dentry); 4017349457ccSMark Fasheh if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) 40181da177e4SLinus Torvalds d_move(old_dentry, new_dentry); 401951892bbbSSage Weil out: 40201da177e4SLinus Torvalds if (target) 40211b1dcc1bSJes Sorensen mutex_unlock(&target->i_mutex); 40221da177e4SLinus Torvalds dput(new_dentry); 40231da177e4SLinus Torvalds return error; 40241da177e4SLinus Torvalds } 40251da177e4SLinus Torvalds 40261da177e4SLinus Torvalds int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, 40271da177e4SLinus Torvalds struct inode *new_dir, struct dentry *new_dentry) 40281da177e4SLinus Torvalds { 40291da177e4SLinus Torvalds int error; 4030b18825a7SDavid Howells int is_dir = d_is_directory(old_dentry) || d_is_autodir(old_dentry); 403159b0df21SEric Paris const unsigned char *old_name; 40321da177e4SLinus Torvalds 40331da177e4SLinus Torvalds if (old_dentry->d_inode == new_dentry->d_inode) 40341da177e4SLinus Torvalds return 0; 40351da177e4SLinus Torvalds 40361da177e4SLinus Torvalds error = may_delete(old_dir, old_dentry, is_dir); 40371da177e4SLinus Torvalds if (error) 40381da177e4SLinus Torvalds return error; 40391da177e4SLinus Torvalds 40401da177e4SLinus Torvalds if (!new_dentry->d_inode) 4041a95164d9SMiklos Szeredi error = may_create(new_dir, new_dentry); 40421da177e4SLinus Torvalds else 40431da177e4SLinus Torvalds error = may_delete(new_dir, new_dentry, is_dir); 40441da177e4SLinus Torvalds if (error) 40451da177e4SLinus Torvalds return error; 40461da177e4SLinus Torvalds 4047acfa4380SAl Viro if (!old_dir->i_op->rename) 40481da177e4SLinus Torvalds return -EPERM; 40491da177e4SLinus Torvalds 40500eeca283SRobert Love old_name = fsnotify_oldname_init(old_dentry->d_name.name); 40510eeca283SRobert Love 40521da177e4SLinus Torvalds if (is_dir) 40531da177e4SLinus Torvalds error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry); 40541da177e4SLinus Torvalds else 40551da177e4SLinus Torvalds error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry); 4056123df294SAl Viro if (!error) 4057123df294SAl Viro fsnotify_move(old_dir, new_dir, old_name, is_dir, 40585a190ae6SAl Viro new_dentry->d_inode, old_dentry); 40590eeca283SRobert Love fsnotify_oldname_free(old_name); 40600eeca283SRobert Love 40611da177e4SLinus Torvalds return error; 40621da177e4SLinus Torvalds } 40631da177e4SLinus Torvalds 40642e4d0924SHeiko Carstens SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, 40652e4d0924SHeiko Carstens int, newdfd, const char __user *, newname) 40661da177e4SLinus Torvalds { 40671da177e4SLinus Torvalds struct dentry *old_dir, *new_dir; 40681da177e4SLinus Torvalds struct dentry *old_dentry, *new_dentry; 40691da177e4SLinus Torvalds struct dentry *trap; 40701da177e4SLinus Torvalds struct nameidata oldnd, newnd; 407191a27b2aSJeff Layton struct filename *from; 407291a27b2aSJeff Layton struct filename *to; 4073c6a94284SJeff Layton unsigned int lookup_flags = 0; 4074c6a94284SJeff Layton bool should_retry = false; 40752ad94ae6SAl Viro int error; 4076c6a94284SJeff Layton retry: 4077c6a94284SJeff Layton from = user_path_parent(olddfd, oldname, &oldnd, lookup_flags); 407891a27b2aSJeff Layton if (IS_ERR(from)) { 407991a27b2aSJeff Layton error = PTR_ERR(from); 40801da177e4SLinus Torvalds goto exit; 408191a27b2aSJeff Layton } 40821da177e4SLinus Torvalds 4083c6a94284SJeff Layton to = user_path_parent(newdfd, newname, &newnd, lookup_flags); 408491a27b2aSJeff Layton if (IS_ERR(to)) { 408591a27b2aSJeff Layton error = PTR_ERR(to); 40861da177e4SLinus Torvalds goto exit1; 408791a27b2aSJeff Layton } 40881da177e4SLinus Torvalds 40891da177e4SLinus Torvalds error = -EXDEV; 40904ac91378SJan Blunck if (oldnd.path.mnt != newnd.path.mnt) 40911da177e4SLinus Torvalds goto exit2; 40921da177e4SLinus Torvalds 40934ac91378SJan Blunck old_dir = oldnd.path.dentry; 40941da177e4SLinus Torvalds error = -EBUSY; 40951da177e4SLinus Torvalds if (oldnd.last_type != LAST_NORM) 40961da177e4SLinus Torvalds goto exit2; 40971da177e4SLinus Torvalds 40984ac91378SJan Blunck new_dir = newnd.path.dentry; 40991da177e4SLinus Torvalds if (newnd.last_type != LAST_NORM) 41001da177e4SLinus Torvalds goto exit2; 41011da177e4SLinus Torvalds 4102c30dabfeSJan Kara error = mnt_want_write(oldnd.path.mnt); 4103c30dabfeSJan Kara if (error) 4104c30dabfeSJan Kara goto exit2; 4105c30dabfeSJan Kara 41060612d9fbSOGAWA Hirofumi oldnd.flags &= ~LOOKUP_PARENT; 41070612d9fbSOGAWA Hirofumi newnd.flags &= ~LOOKUP_PARENT; 41084e9ed2f8SOGAWA Hirofumi newnd.flags |= LOOKUP_RENAME_TARGET; 41090612d9fbSOGAWA Hirofumi 41101da177e4SLinus Torvalds trap = lock_rename(new_dir, old_dir); 41111da177e4SLinus Torvalds 411249705b77SChristoph Hellwig old_dentry = lookup_hash(&oldnd); 41131da177e4SLinus Torvalds error = PTR_ERR(old_dentry); 41141da177e4SLinus Torvalds if (IS_ERR(old_dentry)) 41151da177e4SLinus Torvalds goto exit3; 41161da177e4SLinus Torvalds /* source must exist */ 41171da177e4SLinus Torvalds error = -ENOENT; 4118b18825a7SDavid Howells if (d_is_negative(old_dentry)) 41191da177e4SLinus Torvalds goto exit4; 41201da177e4SLinus Torvalds /* unless the source is a directory trailing slashes give -ENOTDIR */ 4121b18825a7SDavid Howells if (!d_is_directory(old_dentry) && !d_is_autodir(old_dentry)) { 41221da177e4SLinus Torvalds error = -ENOTDIR; 41231da177e4SLinus Torvalds if (oldnd.last.name[oldnd.last.len]) 41241da177e4SLinus Torvalds goto exit4; 41251da177e4SLinus Torvalds if (newnd.last.name[newnd.last.len]) 41261da177e4SLinus Torvalds goto exit4; 41271da177e4SLinus Torvalds } 41281da177e4SLinus Torvalds /* source should not be ancestor of target */ 41291da177e4SLinus Torvalds error = -EINVAL; 41301da177e4SLinus Torvalds if (old_dentry == trap) 41311da177e4SLinus Torvalds goto exit4; 413249705b77SChristoph Hellwig new_dentry = lookup_hash(&newnd); 41331da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 41341da177e4SLinus Torvalds if (IS_ERR(new_dentry)) 41351da177e4SLinus Torvalds goto exit4; 41361da177e4SLinus Torvalds /* target should not be an ancestor of source */ 41371da177e4SLinus Torvalds error = -ENOTEMPTY; 41381da177e4SLinus Torvalds if (new_dentry == trap) 41391da177e4SLinus Torvalds goto exit5; 41401da177e4SLinus Torvalds 4141be6d3e56SKentaro Takeda error = security_path_rename(&oldnd.path, old_dentry, 4142be6d3e56SKentaro Takeda &newnd.path, new_dentry); 4143be6d3e56SKentaro Takeda if (error) 4144c30dabfeSJan Kara goto exit5; 41451da177e4SLinus Torvalds error = vfs_rename(old_dir->d_inode, old_dentry, 41461da177e4SLinus Torvalds new_dir->d_inode, new_dentry); 41471da177e4SLinus Torvalds exit5: 41481da177e4SLinus Torvalds dput(new_dentry); 41491da177e4SLinus Torvalds exit4: 41501da177e4SLinus Torvalds dput(old_dentry); 41511da177e4SLinus Torvalds exit3: 41521da177e4SLinus Torvalds unlock_rename(new_dir, old_dir); 4153c30dabfeSJan Kara mnt_drop_write(oldnd.path.mnt); 41541da177e4SLinus Torvalds exit2: 4155c6a94284SJeff Layton if (retry_estale(error, lookup_flags)) 4156c6a94284SJeff Layton should_retry = true; 41571d957f9bSJan Blunck path_put(&newnd.path); 41582ad94ae6SAl Viro putname(to); 41591da177e4SLinus Torvalds exit1: 41601d957f9bSJan Blunck path_put(&oldnd.path); 41611da177e4SLinus Torvalds putname(from); 4162c6a94284SJeff Layton if (should_retry) { 4163c6a94284SJeff Layton should_retry = false; 4164c6a94284SJeff Layton lookup_flags |= LOOKUP_REVAL; 4165c6a94284SJeff Layton goto retry; 4166c6a94284SJeff Layton } 41672ad94ae6SAl Viro exit: 41681da177e4SLinus Torvalds return error; 41691da177e4SLinus Torvalds } 41701da177e4SLinus Torvalds 4171a26eab24SHeiko Carstens SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) 41725590ff0dSUlrich Drepper { 41735590ff0dSUlrich Drepper return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname); 41745590ff0dSUlrich Drepper } 41755590ff0dSUlrich Drepper 41761da177e4SLinus Torvalds int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link) 41771da177e4SLinus Torvalds { 41781da177e4SLinus Torvalds int len; 41791da177e4SLinus Torvalds 41801da177e4SLinus Torvalds len = PTR_ERR(link); 41811da177e4SLinus Torvalds if (IS_ERR(link)) 41821da177e4SLinus Torvalds goto out; 41831da177e4SLinus Torvalds 41841da177e4SLinus Torvalds len = strlen(link); 41851da177e4SLinus Torvalds if (len > (unsigned) buflen) 41861da177e4SLinus Torvalds len = buflen; 41871da177e4SLinus Torvalds if (copy_to_user(buffer, link, len)) 41881da177e4SLinus Torvalds len = -EFAULT; 41891da177e4SLinus Torvalds out: 41901da177e4SLinus Torvalds return len; 41911da177e4SLinus Torvalds } 41921da177e4SLinus Torvalds 41931da177e4SLinus Torvalds /* 41941da177e4SLinus Torvalds * A helper for ->readlink(). This should be used *ONLY* for symlinks that 41951da177e4SLinus Torvalds * have ->follow_link() touching nd only in nd_set_link(). Using (or not 41961da177e4SLinus Torvalds * using) it for any given inode is up to filesystem. 41971da177e4SLinus Torvalds */ 41981da177e4SLinus Torvalds int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen) 41991da177e4SLinus Torvalds { 42001da177e4SLinus Torvalds struct nameidata nd; 4201cc314eefSLinus Torvalds void *cookie; 4202694a1764SMarcin Slusarz int res; 4203cc314eefSLinus Torvalds 42041da177e4SLinus Torvalds nd.depth = 0; 4205cc314eefSLinus Torvalds cookie = dentry->d_inode->i_op->follow_link(dentry, &nd); 4206694a1764SMarcin Slusarz if (IS_ERR(cookie)) 4207694a1764SMarcin Slusarz return PTR_ERR(cookie); 4208694a1764SMarcin Slusarz 4209694a1764SMarcin Slusarz res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd)); 42101da177e4SLinus Torvalds if (dentry->d_inode->i_op->put_link) 4211cc314eefSLinus Torvalds dentry->d_inode->i_op->put_link(dentry, &nd, cookie); 4212694a1764SMarcin Slusarz return res; 42131da177e4SLinus Torvalds } 42141da177e4SLinus Torvalds 42151da177e4SLinus Torvalds /* get the link contents into pagecache */ 42161da177e4SLinus Torvalds static char *page_getlink(struct dentry * dentry, struct page **ppage) 42171da177e4SLinus Torvalds { 4218ebd09abbSDuane Griffin char *kaddr; 42191da177e4SLinus Torvalds struct page *page; 42201da177e4SLinus Torvalds struct address_space *mapping = dentry->d_inode->i_mapping; 4221090d2b18SPekka Enberg page = read_mapping_page(mapping, 0, NULL); 42221da177e4SLinus Torvalds if (IS_ERR(page)) 42236fe6900eSNick Piggin return (char*)page; 42241da177e4SLinus Torvalds *ppage = page; 4225ebd09abbSDuane Griffin kaddr = kmap(page); 4226ebd09abbSDuane Griffin nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1); 4227ebd09abbSDuane Griffin return kaddr; 42281da177e4SLinus Torvalds } 42291da177e4SLinus Torvalds 42301da177e4SLinus Torvalds int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 42311da177e4SLinus Torvalds { 42321da177e4SLinus Torvalds struct page *page = NULL; 42331da177e4SLinus Torvalds char *s = page_getlink(dentry, &page); 42341da177e4SLinus Torvalds int res = vfs_readlink(dentry,buffer,buflen,s); 42351da177e4SLinus Torvalds if (page) { 42361da177e4SLinus Torvalds kunmap(page); 42371da177e4SLinus Torvalds page_cache_release(page); 42381da177e4SLinus Torvalds } 42391da177e4SLinus Torvalds return res; 42401da177e4SLinus Torvalds } 42411da177e4SLinus Torvalds 4242cc314eefSLinus Torvalds void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd) 42431da177e4SLinus Torvalds { 4244cc314eefSLinus Torvalds struct page *page = NULL; 42451da177e4SLinus Torvalds nd_set_link(nd, page_getlink(dentry, &page)); 4246cc314eefSLinus Torvalds return page; 42471da177e4SLinus Torvalds } 42481da177e4SLinus Torvalds 4249cc314eefSLinus Torvalds void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) 42501da177e4SLinus Torvalds { 4251cc314eefSLinus Torvalds struct page *page = cookie; 4252cc314eefSLinus Torvalds 4253cc314eefSLinus Torvalds if (page) { 42541da177e4SLinus Torvalds kunmap(page); 42551da177e4SLinus Torvalds page_cache_release(page); 42561da177e4SLinus Torvalds } 42571da177e4SLinus Torvalds } 42581da177e4SLinus Torvalds 425954566b2cSNick Piggin /* 426054566b2cSNick Piggin * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS 426154566b2cSNick Piggin */ 426254566b2cSNick Piggin int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) 42631da177e4SLinus Torvalds { 42641da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 42650adb25d2SKirill Korotaev struct page *page; 4266afddba49SNick Piggin void *fsdata; 4267beb497abSDmitriy Monakhov int err; 42681da177e4SLinus Torvalds char *kaddr; 426954566b2cSNick Piggin unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE; 427054566b2cSNick Piggin if (nofs) 427154566b2cSNick Piggin flags |= AOP_FLAG_NOFS; 42721da177e4SLinus Torvalds 42737e53cac4SNeilBrown retry: 4274afddba49SNick Piggin err = pagecache_write_begin(NULL, mapping, 0, len-1, 427554566b2cSNick Piggin flags, &page, &fsdata); 42761da177e4SLinus Torvalds if (err) 4277afddba49SNick Piggin goto fail; 4278afddba49SNick Piggin 4279e8e3c3d6SCong Wang kaddr = kmap_atomic(page); 42801da177e4SLinus Torvalds memcpy(kaddr, symname, len-1); 4281e8e3c3d6SCong Wang kunmap_atomic(kaddr); 4282afddba49SNick Piggin 4283afddba49SNick Piggin err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, 4284afddba49SNick Piggin page, fsdata); 42851da177e4SLinus Torvalds if (err < 0) 42861da177e4SLinus Torvalds goto fail; 4287afddba49SNick Piggin if (err < len-1) 4288afddba49SNick Piggin goto retry; 4289afddba49SNick Piggin 42901da177e4SLinus Torvalds mark_inode_dirty(inode); 42911da177e4SLinus Torvalds return 0; 42921da177e4SLinus Torvalds fail: 42931da177e4SLinus Torvalds return err; 42941da177e4SLinus Torvalds } 42951da177e4SLinus Torvalds 42960adb25d2SKirill Korotaev int page_symlink(struct inode *inode, const char *symname, int len) 42970adb25d2SKirill Korotaev { 42980adb25d2SKirill Korotaev return __page_symlink(inode, symname, len, 429954566b2cSNick Piggin !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS)); 43000adb25d2SKirill Korotaev } 43010adb25d2SKirill Korotaev 430292e1d5beSArjan van de Ven const struct inode_operations page_symlink_inode_operations = { 43031da177e4SLinus Torvalds .readlink = generic_readlink, 43041da177e4SLinus Torvalds .follow_link = page_follow_link_light, 43051da177e4SLinus Torvalds .put_link = page_put_link, 43061da177e4SLinus Torvalds }; 43071da177e4SLinus Torvalds 43082d8f3038SAl Viro EXPORT_SYMBOL(user_path_at); 4309cc53ce53SDavid Howells EXPORT_SYMBOL(follow_down_one); 43101da177e4SLinus Torvalds EXPORT_SYMBOL(follow_down); 43111da177e4SLinus Torvalds EXPORT_SYMBOL(follow_up); 4312f6d2ac5cSAl Viro EXPORT_SYMBOL(get_write_access); /* nfsd */ 43131da177e4SLinus Torvalds EXPORT_SYMBOL(lock_rename); 43141da177e4SLinus Torvalds EXPORT_SYMBOL(lookup_one_len); 43151da177e4SLinus Torvalds EXPORT_SYMBOL(page_follow_link_light); 43161da177e4SLinus Torvalds EXPORT_SYMBOL(page_put_link); 43171da177e4SLinus Torvalds EXPORT_SYMBOL(page_readlink); 43180adb25d2SKirill Korotaev EXPORT_SYMBOL(__page_symlink); 43191da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink); 43201da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink_inode_operations); 4321d1811465SAl Viro EXPORT_SYMBOL(kern_path); 432216f18200SJosef 'Jeff' Sipek EXPORT_SYMBOL(vfs_path_lookup); 4323f419a2e3SAl Viro EXPORT_SYMBOL(inode_permission); 43241da177e4SLinus Torvalds EXPORT_SYMBOL(unlock_rename); 43251da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_create); 43261da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_link); 43271da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_mkdir); 43281da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_mknod); 43291da177e4SLinus Torvalds EXPORT_SYMBOL(generic_permission); 43301da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_readlink); 43311da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_rename); 43321da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_rmdir); 43331da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_symlink); 43341da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_unlink); 43351da177e4SLinus Torvalds EXPORT_SYMBOL(dentry_unhash); 43361da177e4SLinus Torvalds EXPORT_SYMBOL(generic_readlink); 4337