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 */ 1201fa1e7f6SAndy Whitcroft static char *getname_flags(const char __user *filename, int flags, int *empty) 1211da177e4SLinus Torvalds { 1223f9f0aa6SLinus Torvalds char *result = __getname(), *err; 1233f9f0aa6SLinus Torvalds int len; 1241da177e4SLinus Torvalds 1253f9f0aa6SLinus Torvalds if (unlikely(!result)) 1264043cde8SEric Paris return ERR_PTR(-ENOMEM); 1271da177e4SLinus Torvalds 1283f9f0aa6SLinus Torvalds len = strncpy_from_user(result, filename, PATH_MAX); 1293f9f0aa6SLinus Torvalds err = ERR_PTR(len); 1303f9f0aa6SLinus Torvalds if (unlikely(len < 0)) 1313f9f0aa6SLinus Torvalds goto error; 1323f9f0aa6SLinus Torvalds 1333f9f0aa6SLinus Torvalds /* The empty path is special. */ 1343f9f0aa6SLinus Torvalds if (unlikely(!len)) { 1353f9f0aa6SLinus Torvalds if (empty) 1361fa1e7f6SAndy Whitcroft *empty = 1; 1373f9f0aa6SLinus Torvalds err = ERR_PTR(-ENOENT); 1383f9f0aa6SLinus Torvalds if (!(flags & LOOKUP_EMPTY)) 1393f9f0aa6SLinus Torvalds goto error; 1401da177e4SLinus Torvalds } 1413f9f0aa6SLinus Torvalds 1423f9f0aa6SLinus Torvalds err = ERR_PTR(-ENAMETOOLONG); 1433f9f0aa6SLinus Torvalds if (likely(len < PATH_MAX)) { 1441da177e4SLinus Torvalds audit_getname(result); 1451da177e4SLinus Torvalds return result; 1461da177e4SLinus Torvalds } 1471da177e4SLinus Torvalds 1483f9f0aa6SLinus Torvalds error: 1493f9f0aa6SLinus Torvalds __putname(result); 1503f9f0aa6SLinus Torvalds return err; 1513f9f0aa6SLinus Torvalds } 1523f9f0aa6SLinus Torvalds 153f52e0c11SAl Viro char *getname(const char __user * filename) 154f52e0c11SAl Viro { 155f7493e5dSLinus Torvalds return getname_flags(filename, 0, NULL); 156f52e0c11SAl Viro } 157f52e0c11SAl Viro 1581da177e4SLinus Torvalds #ifdef CONFIG_AUDITSYSCALL 1591da177e4SLinus Torvalds void putname(const char *name) 1601da177e4SLinus Torvalds { 1615ac3a9c2SAl Viro if (unlikely(!audit_dummy_context())) 1621da177e4SLinus Torvalds audit_putname(name); 1631da177e4SLinus Torvalds else 1641da177e4SLinus Torvalds __putname(name); 1651da177e4SLinus Torvalds } 1661da177e4SLinus Torvalds EXPORT_SYMBOL(putname); 1671da177e4SLinus Torvalds #endif 1681da177e4SLinus Torvalds 169e77819e5SLinus Torvalds static int check_acl(struct inode *inode, int mask) 170e77819e5SLinus Torvalds { 17184635d68SLinus Torvalds #ifdef CONFIG_FS_POSIX_ACL 172e77819e5SLinus Torvalds struct posix_acl *acl; 173e77819e5SLinus Torvalds 174e77819e5SLinus Torvalds if (mask & MAY_NOT_BLOCK) { 1753567866bSAl Viro acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS); 1763567866bSAl Viro if (!acl) 177e77819e5SLinus Torvalds return -EAGAIN; 1783567866bSAl Viro /* no ->get_acl() calls in RCU mode... */ 1793567866bSAl Viro if (acl == ACL_NOT_CACHED) 180e77819e5SLinus Torvalds return -ECHILD; 181206b1d09SAri Savolainen return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK); 182e77819e5SLinus Torvalds } 183e77819e5SLinus Torvalds 184e77819e5SLinus Torvalds acl = get_cached_acl(inode, ACL_TYPE_ACCESS); 185e77819e5SLinus Torvalds 186e77819e5SLinus Torvalds /* 1874e34e719SChristoph Hellwig * A filesystem can force a ACL callback by just never filling the 1884e34e719SChristoph Hellwig * ACL cache. But normally you'd fill the cache either at inode 1894e34e719SChristoph Hellwig * instantiation time, or on the first ->get_acl call. 190e77819e5SLinus Torvalds * 1914e34e719SChristoph Hellwig * If the filesystem doesn't have a get_acl() function at all, we'll 1924e34e719SChristoph Hellwig * just create the negative cache entry. 193e77819e5SLinus Torvalds */ 194e77819e5SLinus Torvalds if (acl == ACL_NOT_CACHED) { 1954e34e719SChristoph Hellwig if (inode->i_op->get_acl) { 1964e34e719SChristoph Hellwig acl = inode->i_op->get_acl(inode, ACL_TYPE_ACCESS); 1974e34e719SChristoph Hellwig if (IS_ERR(acl)) 1984e34e719SChristoph Hellwig return PTR_ERR(acl); 1994e34e719SChristoph Hellwig } else { 200e77819e5SLinus Torvalds set_cached_acl(inode, ACL_TYPE_ACCESS, NULL); 201e77819e5SLinus Torvalds return -EAGAIN; 202e77819e5SLinus Torvalds } 2034e34e719SChristoph Hellwig } 204e77819e5SLinus Torvalds 205e77819e5SLinus Torvalds if (acl) { 206e77819e5SLinus Torvalds int error = posix_acl_permission(inode, acl, mask); 207e77819e5SLinus Torvalds posix_acl_release(acl); 208e77819e5SLinus Torvalds return error; 209e77819e5SLinus Torvalds } 21084635d68SLinus Torvalds #endif 211e77819e5SLinus Torvalds 212e77819e5SLinus Torvalds return -EAGAIN; 213e77819e5SLinus Torvalds } 214e77819e5SLinus Torvalds 2155909ccaaSLinus Torvalds /* 216948409c7SAndreas Gruenbacher * This does the basic permission checking 2175909ccaaSLinus Torvalds */ 2187e40145eSAl Viro static int acl_permission_check(struct inode *inode, int mask) 2195909ccaaSLinus Torvalds { 22026cf46beSLinus Torvalds unsigned int mode = inode->i_mode; 2215909ccaaSLinus Torvalds 2228e96e3b7SEric W. Biederman if (likely(uid_eq(current_fsuid(), inode->i_uid))) 2235909ccaaSLinus Torvalds mode >>= 6; 2245909ccaaSLinus Torvalds else { 225e77819e5SLinus Torvalds if (IS_POSIXACL(inode) && (mode & S_IRWXG)) { 2267e40145eSAl Viro int error = check_acl(inode, mask); 2275909ccaaSLinus Torvalds if (error != -EAGAIN) 2285909ccaaSLinus Torvalds return error; 2295909ccaaSLinus Torvalds } 2305909ccaaSLinus Torvalds 2315909ccaaSLinus Torvalds if (in_group_p(inode->i_gid)) 2325909ccaaSLinus Torvalds mode >>= 3; 2335909ccaaSLinus Torvalds } 2345909ccaaSLinus Torvalds 2355909ccaaSLinus Torvalds /* 2365909ccaaSLinus Torvalds * If the DACs are ok we don't need any capability check. 2375909ccaaSLinus Torvalds */ 2389c2c7039SAl Viro if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) 2395909ccaaSLinus Torvalds return 0; 2405909ccaaSLinus Torvalds return -EACCES; 2415909ccaaSLinus Torvalds } 2421da177e4SLinus Torvalds 2431da177e4SLinus Torvalds /** 2441da177e4SLinus Torvalds * generic_permission - check for access rights on a Posix-like filesystem 2451da177e4SLinus Torvalds * @inode: inode to check access rights for 2468fd90c8dSAndreas Gruenbacher * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...) 2471da177e4SLinus Torvalds * 2481da177e4SLinus Torvalds * Used to check for read/write/execute permissions on a file. 2491da177e4SLinus Torvalds * We use "fsuid" for this, letting us set arbitrary permissions 2501da177e4SLinus Torvalds * for filesystem access without changing the "normal" uids which 251b74c79e9SNick Piggin * are used for other things. 252b74c79e9SNick Piggin * 253b74c79e9SNick Piggin * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk 254b74c79e9SNick Piggin * request cannot be satisfied (eg. requires blocking or too much complexity). 255b74c79e9SNick Piggin * It would then be called again in ref-walk mode. 2561da177e4SLinus Torvalds */ 2572830ba7fSAl Viro int generic_permission(struct inode *inode, int mask) 2581da177e4SLinus Torvalds { 2595909ccaaSLinus Torvalds int ret; 2601da177e4SLinus Torvalds 2611da177e4SLinus Torvalds /* 262948409c7SAndreas Gruenbacher * Do the basic permission checks. 2631da177e4SLinus Torvalds */ 2647e40145eSAl Viro ret = acl_permission_check(inode, mask); 2655909ccaaSLinus Torvalds if (ret != -EACCES) 2665909ccaaSLinus Torvalds return ret; 2671da177e4SLinus Torvalds 268d594e7ecSAl Viro if (S_ISDIR(inode->i_mode)) { 269d594e7ecSAl Viro /* DACs are overridable for directories */ 2701a48e2acSEric W. Biederman if (inode_capable(inode, CAP_DAC_OVERRIDE)) 271d594e7ecSAl Viro return 0; 272d594e7ecSAl Viro if (!(mask & MAY_WRITE)) 2731a48e2acSEric W. Biederman if (inode_capable(inode, CAP_DAC_READ_SEARCH)) 274d594e7ecSAl Viro return 0; 275d594e7ecSAl Viro return -EACCES; 276d594e7ecSAl Viro } 2771da177e4SLinus Torvalds /* 2781da177e4SLinus Torvalds * Read/write DACs are always overridable. 279d594e7ecSAl Viro * Executable DACs are overridable when there is 280d594e7ecSAl Viro * at least one exec bit set. 2811da177e4SLinus Torvalds */ 282d594e7ecSAl Viro if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO)) 2831a48e2acSEric W. Biederman if (inode_capable(inode, CAP_DAC_OVERRIDE)) 2841da177e4SLinus Torvalds return 0; 2851da177e4SLinus Torvalds 2861da177e4SLinus Torvalds /* 2871da177e4SLinus Torvalds * Searching includes executable on directories, else just read. 2881da177e4SLinus Torvalds */ 2897ea66001SSerge E. Hallyn mask &= MAY_READ | MAY_WRITE | MAY_EXEC; 290d594e7ecSAl Viro if (mask == MAY_READ) 2911a48e2acSEric W. Biederman if (inode_capable(inode, CAP_DAC_READ_SEARCH)) 2921da177e4SLinus Torvalds return 0; 2931da177e4SLinus Torvalds 2941da177e4SLinus Torvalds return -EACCES; 2951da177e4SLinus Torvalds } 2961da177e4SLinus Torvalds 2973ddcd056SLinus Torvalds /* 2983ddcd056SLinus Torvalds * We _really_ want to just do "generic_permission()" without 2993ddcd056SLinus Torvalds * even looking at the inode->i_op values. So we keep a cache 3003ddcd056SLinus Torvalds * flag in inode->i_opflags, that says "this has not special 3013ddcd056SLinus Torvalds * permission function, use the fast case". 3023ddcd056SLinus Torvalds */ 3033ddcd056SLinus Torvalds static inline int do_inode_permission(struct inode *inode, int mask) 3043ddcd056SLinus Torvalds { 3053ddcd056SLinus Torvalds if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) { 3063ddcd056SLinus Torvalds if (likely(inode->i_op->permission)) 3073ddcd056SLinus Torvalds return inode->i_op->permission(inode, mask); 3083ddcd056SLinus Torvalds 3093ddcd056SLinus Torvalds /* This gets set once for the inode lifetime */ 3103ddcd056SLinus Torvalds spin_lock(&inode->i_lock); 3113ddcd056SLinus Torvalds inode->i_opflags |= IOP_FASTPERM; 3123ddcd056SLinus Torvalds spin_unlock(&inode->i_lock); 3133ddcd056SLinus Torvalds } 3143ddcd056SLinus Torvalds return generic_permission(inode, mask); 3153ddcd056SLinus Torvalds } 3163ddcd056SLinus Torvalds 317cb23beb5SChristoph Hellwig /** 3180bdaea90SDavid Howells * __inode_permission - Check for access rights to a given inode 3190bdaea90SDavid Howells * @inode: Inode to check permission on 3200bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 321cb23beb5SChristoph Hellwig * 3220bdaea90SDavid Howells * Check for read/write/execute permissions on an inode. 323948409c7SAndreas Gruenbacher * 324948409c7SAndreas Gruenbacher * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. 3250bdaea90SDavid Howells * 3260bdaea90SDavid Howells * This does not check for a read-only file system. You probably want 3270bdaea90SDavid Howells * inode_permission(). 328cb23beb5SChristoph Hellwig */ 3290bdaea90SDavid Howells int __inode_permission(struct inode *inode, int mask) 3301da177e4SLinus Torvalds { 331e6305c43SAl Viro int retval; 3321da177e4SLinus Torvalds 3333ddcd056SLinus Torvalds if (unlikely(mask & MAY_WRITE)) { 3341da177e4SLinus Torvalds /* 3351da177e4SLinus Torvalds * Nobody gets write access to an immutable file. 3361da177e4SLinus Torvalds */ 3371da177e4SLinus Torvalds if (IS_IMMUTABLE(inode)) 3381da177e4SLinus Torvalds return -EACCES; 3391da177e4SLinus Torvalds } 3401da177e4SLinus Torvalds 3413ddcd056SLinus Torvalds retval = do_inode_permission(inode, mask); 3421da177e4SLinus Torvalds if (retval) 3431da177e4SLinus Torvalds return retval; 3441da177e4SLinus Torvalds 34508ce5f16SSerge E. Hallyn retval = devcgroup_inode_permission(inode, mask); 34608ce5f16SSerge E. Hallyn if (retval) 34708ce5f16SSerge E. Hallyn return retval; 34808ce5f16SSerge E. Hallyn 349d09ca739SEric Paris return security_inode_permission(inode, mask); 3501da177e4SLinus Torvalds } 3511da177e4SLinus Torvalds 352f4d6ff89SAl Viro /** 3530bdaea90SDavid Howells * sb_permission - Check superblock-level permissions 3540bdaea90SDavid Howells * @sb: Superblock of inode to check permission on 3550bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 3560bdaea90SDavid Howells * 3570bdaea90SDavid Howells * Separate out file-system wide checks from inode-specific permission checks. 3580bdaea90SDavid Howells */ 3590bdaea90SDavid Howells static int sb_permission(struct super_block *sb, struct inode *inode, int mask) 3600bdaea90SDavid Howells { 3610bdaea90SDavid Howells if (unlikely(mask & MAY_WRITE)) { 3620bdaea90SDavid Howells umode_t mode = inode->i_mode; 3630bdaea90SDavid Howells 3640bdaea90SDavid Howells /* Nobody gets write access to a read-only fs. */ 3650bdaea90SDavid Howells if ((sb->s_flags & MS_RDONLY) && 3660bdaea90SDavid Howells (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) 3670bdaea90SDavid Howells return -EROFS; 3680bdaea90SDavid Howells } 3690bdaea90SDavid Howells return 0; 3700bdaea90SDavid Howells } 3710bdaea90SDavid Howells 3720bdaea90SDavid Howells /** 3730bdaea90SDavid Howells * inode_permission - Check for access rights to a given inode 3740bdaea90SDavid Howells * @inode: Inode to check permission on 3750bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 3760bdaea90SDavid Howells * 3770bdaea90SDavid Howells * Check for read/write/execute permissions on an inode. We use fs[ug]id for 3780bdaea90SDavid Howells * this, letting us set arbitrary permissions for filesystem access without 3790bdaea90SDavid Howells * changing the "normal" UIDs which are used for other things. 3800bdaea90SDavid Howells * 3810bdaea90SDavid Howells * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. 3820bdaea90SDavid Howells */ 3830bdaea90SDavid Howells int inode_permission(struct inode *inode, int mask) 3840bdaea90SDavid Howells { 3850bdaea90SDavid Howells int retval; 3860bdaea90SDavid Howells 3870bdaea90SDavid Howells retval = sb_permission(inode->i_sb, inode, mask); 3880bdaea90SDavid Howells if (retval) 3890bdaea90SDavid Howells return retval; 3900bdaea90SDavid Howells return __inode_permission(inode, mask); 3910bdaea90SDavid Howells } 3920bdaea90SDavid Howells 3930bdaea90SDavid Howells /** 3945dd784d0SJan Blunck * path_get - get a reference to a path 3955dd784d0SJan Blunck * @path: path to get the reference to 3965dd784d0SJan Blunck * 3975dd784d0SJan Blunck * Given a path increment the reference count to the dentry and the vfsmount. 3985dd784d0SJan Blunck */ 3995dd784d0SJan Blunck void path_get(struct path *path) 4005dd784d0SJan Blunck { 4015dd784d0SJan Blunck mntget(path->mnt); 4025dd784d0SJan Blunck dget(path->dentry); 4035dd784d0SJan Blunck } 4045dd784d0SJan Blunck EXPORT_SYMBOL(path_get); 4055dd784d0SJan Blunck 4065dd784d0SJan Blunck /** 4071d957f9bSJan Blunck * path_put - put a reference to a path 4081d957f9bSJan Blunck * @path: path to put the reference to 4091d957f9bSJan Blunck * 4101d957f9bSJan Blunck * Given a path decrement the reference count to the dentry and the vfsmount. 4111d957f9bSJan Blunck */ 4121d957f9bSJan Blunck void path_put(struct path *path) 4131da177e4SLinus Torvalds { 4141d957f9bSJan Blunck dput(path->dentry); 4151d957f9bSJan Blunck mntput(path->mnt); 4161da177e4SLinus Torvalds } 4171d957f9bSJan Blunck EXPORT_SYMBOL(path_put); 4181da177e4SLinus Torvalds 41919660af7SAl Viro /* 42031e6b01fSNick Piggin * Path walking has 2 modes, rcu-walk and ref-walk (see 42119660af7SAl Viro * Documentation/filesystems/path-lookup.txt). In situations when we can't 42219660af7SAl Viro * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab 42319660af7SAl Viro * normal reference counts on dentries and vfsmounts to transition to rcu-walk 42419660af7SAl Viro * mode. Refcounts are grabbed at the last known good point before rcu-walk 42519660af7SAl Viro * got stuck, so ref-walk may continue from there. If this is not successful 42619660af7SAl Viro * (eg. a seqcount has changed), then failure is returned and it's up to caller 42719660af7SAl Viro * to restart the path walk from the beginning in ref-walk mode. 42831e6b01fSNick Piggin */ 42931e6b01fSNick Piggin 43031e6b01fSNick Piggin /** 43119660af7SAl Viro * unlazy_walk - try to switch to ref-walk mode. 43219660af7SAl Viro * @nd: nameidata pathwalk data 43319660af7SAl Viro * @dentry: child of nd->path.dentry or NULL 43439191628SRandy Dunlap * Returns: 0 on success, -ECHILD on failure 43531e6b01fSNick Piggin * 43619660af7SAl Viro * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry 43719660af7SAl Viro * for ref-walk mode. @dentry must be a path found by a do_lookup call on 43819660af7SAl Viro * @nd or NULL. Must be called from rcu-walk context. 43931e6b01fSNick Piggin */ 44019660af7SAl Viro static int unlazy_walk(struct nameidata *nd, struct dentry *dentry) 44131e6b01fSNick Piggin { 44231e6b01fSNick Piggin struct fs_struct *fs = current->fs; 44331e6b01fSNick Piggin struct dentry *parent = nd->path.dentry; 4445b6ca027SAl Viro int want_root = 0; 44531e6b01fSNick Piggin 44631e6b01fSNick Piggin BUG_ON(!(nd->flags & LOOKUP_RCU)); 4475b6ca027SAl Viro if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 4485b6ca027SAl Viro want_root = 1; 44931e6b01fSNick Piggin spin_lock(&fs->lock); 45031e6b01fSNick Piggin if (nd->root.mnt != fs->root.mnt || 45131e6b01fSNick Piggin nd->root.dentry != fs->root.dentry) 45231e6b01fSNick Piggin goto err_root; 45331e6b01fSNick Piggin } 45431e6b01fSNick Piggin spin_lock(&parent->d_lock); 45519660af7SAl Viro if (!dentry) { 45619660af7SAl Viro if (!__d_rcu_to_refcount(parent, nd->seq)) 45719660af7SAl Viro goto err_parent; 45819660af7SAl Viro BUG_ON(nd->inode != parent->d_inode); 45919660af7SAl Viro } else { 46094c0d4ecSAl Viro if (dentry->d_parent != parent) 46194c0d4ecSAl Viro goto err_parent; 46231e6b01fSNick Piggin spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); 46331e6b01fSNick Piggin if (!__d_rcu_to_refcount(dentry, nd->seq)) 46419660af7SAl Viro goto err_child; 46531e6b01fSNick Piggin /* 46619660af7SAl Viro * If the sequence check on the child dentry passed, then 46719660af7SAl Viro * the child has not been removed from its parent. This 46819660af7SAl Viro * means the parent dentry must be valid and able to take 46919660af7SAl Viro * a reference at this point. 47031e6b01fSNick Piggin */ 47131e6b01fSNick Piggin BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent); 47231e6b01fSNick Piggin BUG_ON(!parent->d_count); 47331e6b01fSNick Piggin parent->d_count++; 47431e6b01fSNick Piggin spin_unlock(&dentry->d_lock); 47519660af7SAl Viro } 47631e6b01fSNick Piggin spin_unlock(&parent->d_lock); 4775b6ca027SAl Viro if (want_root) { 47831e6b01fSNick Piggin path_get(&nd->root); 47931e6b01fSNick Piggin spin_unlock(&fs->lock); 48031e6b01fSNick Piggin } 48131e6b01fSNick Piggin mntget(nd->path.mnt); 48231e6b01fSNick Piggin 48331e6b01fSNick Piggin rcu_read_unlock(); 484962830dfSAndi Kleen br_read_unlock(&vfsmount_lock); 48531e6b01fSNick Piggin nd->flags &= ~LOOKUP_RCU; 48631e6b01fSNick Piggin return 0; 48719660af7SAl Viro 48819660af7SAl Viro err_child: 48931e6b01fSNick Piggin spin_unlock(&dentry->d_lock); 49019660af7SAl Viro err_parent: 49131e6b01fSNick Piggin spin_unlock(&parent->d_lock); 49231e6b01fSNick Piggin err_root: 4935b6ca027SAl Viro if (want_root) 49431e6b01fSNick Piggin spin_unlock(&fs->lock); 49531e6b01fSNick Piggin return -ECHILD; 49631e6b01fSNick Piggin } 49731e6b01fSNick Piggin 4984ce16ef3SAl Viro static inline int d_revalidate(struct dentry *dentry, unsigned int flags) 49934286d66SNick Piggin { 5004ce16ef3SAl Viro return dentry->d_op->d_revalidate(dentry, flags); 50134286d66SNick Piggin } 50234286d66SNick Piggin 5039f1fafeeSAl Viro /** 5049f1fafeeSAl Viro * complete_walk - successful completion of path walk 5059f1fafeeSAl Viro * @nd: pointer nameidata 50639159de2SJeff Layton * 5079f1fafeeSAl Viro * If we had been in RCU mode, drop out of it and legitimize nd->path. 5089f1fafeeSAl Viro * Revalidate the final result, unless we'd already done that during 5099f1fafeeSAl Viro * the path walk or the filesystem doesn't ask for it. Return 0 on 5109f1fafeeSAl Viro * success, -error on failure. In case of failure caller does not 5119f1fafeeSAl Viro * need to drop nd->path. 51239159de2SJeff Layton */ 5139f1fafeeSAl Viro static int complete_walk(struct nameidata *nd) 51439159de2SJeff Layton { 51516c2cd71SAl Viro struct dentry *dentry = nd->path.dentry; 51639159de2SJeff Layton int status; 51739159de2SJeff Layton 5189f1fafeeSAl Viro if (nd->flags & LOOKUP_RCU) { 5199f1fafeeSAl Viro nd->flags &= ~LOOKUP_RCU; 5209f1fafeeSAl Viro if (!(nd->flags & LOOKUP_ROOT)) 5219f1fafeeSAl Viro nd->root.mnt = NULL; 5229f1fafeeSAl Viro spin_lock(&dentry->d_lock); 5239f1fafeeSAl Viro if (unlikely(!__d_rcu_to_refcount(dentry, nd->seq))) { 5249f1fafeeSAl Viro spin_unlock(&dentry->d_lock); 5259f1fafeeSAl Viro rcu_read_unlock(); 526962830dfSAndi Kleen br_read_unlock(&vfsmount_lock); 5279f1fafeeSAl Viro return -ECHILD; 5289f1fafeeSAl Viro } 5299f1fafeeSAl Viro BUG_ON(nd->inode != dentry->d_inode); 5309f1fafeeSAl Viro spin_unlock(&dentry->d_lock); 5319f1fafeeSAl Viro mntget(nd->path.mnt); 5329f1fafeeSAl Viro rcu_read_unlock(); 533962830dfSAndi Kleen br_read_unlock(&vfsmount_lock); 5349f1fafeeSAl Viro } 5359f1fafeeSAl Viro 53616c2cd71SAl Viro if (likely(!(nd->flags & LOOKUP_JUMPED))) 53739159de2SJeff Layton return 0; 53839159de2SJeff Layton 53916c2cd71SAl Viro if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE))) 54016c2cd71SAl Viro return 0; 54116c2cd71SAl Viro 54216c2cd71SAl Viro if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT))) 54316c2cd71SAl Viro return 0; 54416c2cd71SAl Viro 54516c2cd71SAl Viro /* Note: we do not d_invalidate() */ 5464ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 54739159de2SJeff Layton if (status > 0) 54839159de2SJeff Layton return 0; 54939159de2SJeff Layton 55016c2cd71SAl Viro if (!status) 55139159de2SJeff Layton status = -ESTALE; 55216c2cd71SAl Viro 5539f1fafeeSAl Viro path_put(&nd->path); 55439159de2SJeff Layton return status; 55539159de2SJeff Layton } 55639159de2SJeff Layton 5572a737871SAl Viro static __always_inline void set_root(struct nameidata *nd) 5582a737871SAl Viro { 559f7ad3c6bSMiklos Szeredi if (!nd->root.mnt) 560f7ad3c6bSMiklos Szeredi get_fs_root(current->fs, &nd->root); 5612a737871SAl Viro } 5622a737871SAl Viro 5636de88d72SAl Viro static int link_path_walk(const char *, struct nameidata *); 5646de88d72SAl Viro 56531e6b01fSNick Piggin static __always_inline void set_root_rcu(struct nameidata *nd) 56631e6b01fSNick Piggin { 56731e6b01fSNick Piggin if (!nd->root.mnt) { 56831e6b01fSNick Piggin struct fs_struct *fs = current->fs; 569c28cc364SNick Piggin unsigned seq; 570c28cc364SNick Piggin 571c28cc364SNick Piggin do { 572c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 57331e6b01fSNick Piggin nd->root = fs->root; 574c1530019STim Chen nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq); 575c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 57631e6b01fSNick Piggin } 57731e6b01fSNick Piggin } 57831e6b01fSNick Piggin 579f1662356SArjan van de Ven static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link) 5801da177e4SLinus Torvalds { 58131e6b01fSNick Piggin int ret; 58231e6b01fSNick Piggin 5831da177e4SLinus Torvalds if (IS_ERR(link)) 5841da177e4SLinus Torvalds goto fail; 5851da177e4SLinus Torvalds 5861da177e4SLinus Torvalds if (*link == '/') { 5872a737871SAl Viro set_root(nd); 5881d957f9bSJan Blunck path_put(&nd->path); 5892a737871SAl Viro nd->path = nd->root; 5902a737871SAl Viro path_get(&nd->root); 59116c2cd71SAl Viro nd->flags |= LOOKUP_JUMPED; 5921da177e4SLinus Torvalds } 59331e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 594b4091d5fSChristoph Hellwig 59531e6b01fSNick Piggin ret = link_path_walk(link, nd); 59631e6b01fSNick Piggin return ret; 5971da177e4SLinus Torvalds fail: 5981d957f9bSJan Blunck path_put(&nd->path); 5991da177e4SLinus Torvalds return PTR_ERR(link); 6001da177e4SLinus Torvalds } 6011da177e4SLinus Torvalds 6021d957f9bSJan Blunck static void path_put_conditional(struct path *path, struct nameidata *nd) 603051d3812SIan Kent { 604051d3812SIan Kent dput(path->dentry); 6054ac91378SJan Blunck if (path->mnt != nd->path.mnt) 606051d3812SIan Kent mntput(path->mnt); 607051d3812SIan Kent } 608051d3812SIan Kent 6097b9337aaSNick Piggin static inline void path_to_nameidata(const struct path *path, 6107b9337aaSNick Piggin struct nameidata *nd) 611051d3812SIan Kent { 61231e6b01fSNick Piggin if (!(nd->flags & LOOKUP_RCU)) { 6134ac91378SJan Blunck dput(nd->path.dentry); 61431e6b01fSNick Piggin if (nd->path.mnt != path->mnt) 6154ac91378SJan Blunck mntput(nd->path.mnt); 6169a229683SHuang Shijie } 61731e6b01fSNick Piggin nd->path.mnt = path->mnt; 6184ac91378SJan Blunck nd->path.dentry = path->dentry; 619051d3812SIan Kent } 620051d3812SIan Kent 621b5fb63c1SChristoph Hellwig /* 622b5fb63c1SChristoph Hellwig * Helper to directly jump to a known parsed path from ->follow_link, 623b5fb63c1SChristoph Hellwig * caller must have taken a reference to path beforehand. 624b5fb63c1SChristoph Hellwig */ 625b5fb63c1SChristoph Hellwig void nd_jump_link(struct nameidata *nd, struct path *path) 626b5fb63c1SChristoph Hellwig { 627b5fb63c1SChristoph Hellwig path_put(&nd->path); 628b5fb63c1SChristoph Hellwig 629b5fb63c1SChristoph Hellwig nd->path = *path; 630b5fb63c1SChristoph Hellwig nd->inode = nd->path.dentry->d_inode; 631b5fb63c1SChristoph Hellwig nd->flags |= LOOKUP_JUMPED; 632b5fb63c1SChristoph Hellwig 633b5fb63c1SChristoph Hellwig BUG_ON(nd->inode->i_op->follow_link); 634b5fb63c1SChristoph Hellwig } 635b5fb63c1SChristoph Hellwig 636574197e0SAl Viro static inline void put_link(struct nameidata *nd, struct path *link, void *cookie) 637574197e0SAl Viro { 638574197e0SAl Viro struct inode *inode = link->dentry->d_inode; 6396d7b5aaeSAl Viro if (inode->i_op->put_link) 640574197e0SAl Viro inode->i_op->put_link(link->dentry, nd, cookie); 641574197e0SAl Viro path_put(link); 642574197e0SAl Viro } 643574197e0SAl Viro 644def4af30SAl Viro static __always_inline int 645574197e0SAl Viro follow_link(struct path *link, struct nameidata *nd, void **p) 6461da177e4SLinus Torvalds { 6477b9337aaSNick Piggin struct dentry *dentry = link->dentry; 6486d7b5aaeSAl Viro int error; 6496d7b5aaeSAl Viro char *s; 6501da177e4SLinus Torvalds 651844a3917SAl Viro BUG_ON(nd->flags & LOOKUP_RCU); 652844a3917SAl Viro 6530e794589SAl Viro if (link->mnt == nd->path.mnt) 6540e794589SAl Viro mntget(link->mnt); 6550e794589SAl Viro 6566d7b5aaeSAl Viro error = -ELOOP; 6576d7b5aaeSAl Viro if (unlikely(current->total_link_count >= 40)) 6586d7b5aaeSAl Viro goto out_put_nd_path; 6596d7b5aaeSAl Viro 660574197e0SAl Viro cond_resched(); 661574197e0SAl Viro current->total_link_count++; 662574197e0SAl Viro 66368ac1234SAl Viro touch_atime(link); 6641da177e4SLinus Torvalds nd_set_link(nd, NULL); 665cd4e91d3SAl Viro 66636f3b4f6SAl Viro error = security_inode_follow_link(link->dentry, nd); 6676d7b5aaeSAl Viro if (error) 6686d7b5aaeSAl Viro goto out_put_nd_path; 66936f3b4f6SAl Viro 67086acdca1SAl Viro nd->last_type = LAST_BIND; 671def4af30SAl Viro *p = dentry->d_inode->i_op->follow_link(dentry, nd); 672def4af30SAl Viro error = PTR_ERR(*p); 6736d7b5aaeSAl Viro if (IS_ERR(*p)) 674408ef013SChristoph Hellwig goto out_put_nd_path; 6756d7b5aaeSAl Viro 676cc314eefSLinus Torvalds error = 0; 6776d7b5aaeSAl Viro s = nd_get_link(nd); 6786d7b5aaeSAl Viro if (s) { 6791da177e4SLinus Torvalds error = __vfs_follow_link(nd, s); 6806d7b5aaeSAl Viro if (unlikely(error)) 6816d7b5aaeSAl Viro put_link(nd, link, *p); 682b5fb63c1SChristoph Hellwig } 6836d7b5aaeSAl Viro 6846d7b5aaeSAl Viro return error; 6856d7b5aaeSAl Viro 6866d7b5aaeSAl Viro out_put_nd_path: 6876d7b5aaeSAl Viro path_put(&nd->path); 6886d7b5aaeSAl Viro path_put(link); 6891da177e4SLinus Torvalds return error; 6901da177e4SLinus Torvalds } 6911da177e4SLinus Torvalds 69231e6b01fSNick Piggin static int follow_up_rcu(struct path *path) 69331e6b01fSNick Piggin { 6940714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 6950714a533SAl Viro struct mount *parent; 69631e6b01fSNick Piggin struct dentry *mountpoint; 69731e6b01fSNick Piggin 6980714a533SAl Viro parent = mnt->mnt_parent; 6990714a533SAl Viro if (&parent->mnt == path->mnt) 70031e6b01fSNick Piggin return 0; 701a73324daSAl Viro mountpoint = mnt->mnt_mountpoint; 70231e6b01fSNick Piggin path->dentry = mountpoint; 7030714a533SAl Viro path->mnt = &parent->mnt; 70431e6b01fSNick Piggin return 1; 70531e6b01fSNick Piggin } 70631e6b01fSNick Piggin 707f015f126SDavid Howells /* 708f015f126SDavid Howells * follow_up - Find the mountpoint of path's vfsmount 709f015f126SDavid Howells * 710f015f126SDavid Howells * Given a path, find the mountpoint of its source file system. 711f015f126SDavid Howells * Replace @path with the path of the mountpoint in the parent mount. 712f015f126SDavid Howells * Up is towards /. 713f015f126SDavid Howells * 714f015f126SDavid Howells * Return 1 if we went up a level and 0 if we were already at the 715f015f126SDavid Howells * root. 716f015f126SDavid Howells */ 717bab77ebfSAl Viro int follow_up(struct path *path) 7181da177e4SLinus Torvalds { 7190714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 7200714a533SAl Viro struct mount *parent; 7211da177e4SLinus Torvalds struct dentry *mountpoint; 72299b7db7bSNick Piggin 723962830dfSAndi Kleen br_read_lock(&vfsmount_lock); 7240714a533SAl Viro parent = mnt->mnt_parent; 7250714a533SAl Viro if (&parent->mnt == path->mnt) { 726962830dfSAndi Kleen br_read_unlock(&vfsmount_lock); 7271da177e4SLinus Torvalds return 0; 7281da177e4SLinus Torvalds } 7290714a533SAl Viro mntget(&parent->mnt); 730a73324daSAl Viro mountpoint = dget(mnt->mnt_mountpoint); 731962830dfSAndi Kleen br_read_unlock(&vfsmount_lock); 732bab77ebfSAl Viro dput(path->dentry); 733bab77ebfSAl Viro path->dentry = mountpoint; 734bab77ebfSAl Viro mntput(path->mnt); 7350714a533SAl Viro path->mnt = &parent->mnt; 7361da177e4SLinus Torvalds return 1; 7371da177e4SLinus Torvalds } 7381da177e4SLinus Torvalds 739b5c84bf6SNick Piggin /* 7409875cf80SDavid Howells * Perform an automount 7419875cf80SDavid Howells * - return -EISDIR to tell follow_managed() to stop and return the path we 7429875cf80SDavid Howells * were called with. 7431da177e4SLinus Torvalds */ 7449875cf80SDavid Howells static int follow_automount(struct path *path, unsigned flags, 7459875cf80SDavid Howells bool *need_mntput) 74631e6b01fSNick Piggin { 7479875cf80SDavid Howells struct vfsmount *mnt; 748ea5b778aSDavid Howells int err; 7499875cf80SDavid Howells 7509875cf80SDavid Howells if (!path->dentry->d_op || !path->dentry->d_op->d_automount) 7519875cf80SDavid Howells return -EREMOTE; 7529875cf80SDavid Howells 7530ec26fd0SMiklos Szeredi /* We don't want to mount if someone's just doing a stat - 7540ec26fd0SMiklos Szeredi * unless they're stat'ing a directory and appended a '/' to 7550ec26fd0SMiklos Szeredi * the name. 7560ec26fd0SMiklos Szeredi * 7570ec26fd0SMiklos Szeredi * We do, however, want to mount if someone wants to open or 7585a30d8a2SDavid Howells * create a file of any type under the mountpoint, wants to 7590ec26fd0SMiklos Szeredi * traverse through the mountpoint or wants to open the 7600ec26fd0SMiklos Szeredi * mounted directory. Also, autofs may mark negative dentries 7610ec26fd0SMiklos Szeredi * as being automount points. These will need the attentions 7620ec26fd0SMiklos Szeredi * of the daemon to instantiate them before they can be used. 7635a30d8a2SDavid Howells */ 7645a30d8a2SDavid Howells if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | 765d94c177bSLinus Torvalds LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && 7665a30d8a2SDavid Howells path->dentry->d_inode) 7679875cf80SDavid Howells return -EISDIR; 7680ec26fd0SMiklos Szeredi 7699875cf80SDavid Howells current->total_link_count++; 7709875cf80SDavid Howells if (current->total_link_count >= 40) 7719875cf80SDavid Howells return -ELOOP; 7729875cf80SDavid Howells 7739875cf80SDavid Howells mnt = path->dentry->d_op->d_automount(path); 7749875cf80SDavid Howells if (IS_ERR(mnt)) { 7759875cf80SDavid Howells /* 7769875cf80SDavid Howells * The filesystem is allowed to return -EISDIR here to indicate 7779875cf80SDavid Howells * it doesn't want to automount. For instance, autofs would do 7789875cf80SDavid Howells * this so that its userspace daemon can mount on this dentry. 7799875cf80SDavid Howells * 7809875cf80SDavid Howells * However, we can only permit this if it's a terminal point in 7819875cf80SDavid Howells * the path being looked up; if it wasn't then the remainder of 7829875cf80SDavid Howells * the path is inaccessible and we should say so. 7839875cf80SDavid Howells */ 78449084c3bSAl Viro if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT)) 7859875cf80SDavid Howells return -EREMOTE; 7869875cf80SDavid Howells return PTR_ERR(mnt); 78731e6b01fSNick Piggin } 788ea5b778aSDavid Howells 7899875cf80SDavid Howells if (!mnt) /* mount collision */ 7909875cf80SDavid Howells return 0; 7919875cf80SDavid Howells 7928aef1884SAl Viro if (!*need_mntput) { 7938aef1884SAl Viro /* lock_mount() may release path->mnt on error */ 7948aef1884SAl Viro mntget(path->mnt); 7958aef1884SAl Viro *need_mntput = true; 7968aef1884SAl Viro } 79719a167afSAl Viro err = finish_automount(mnt, path); 798ea5b778aSDavid Howells 799ea5b778aSDavid Howells switch (err) { 800ea5b778aSDavid Howells case -EBUSY: 801ea5b778aSDavid Howells /* Someone else made a mount here whilst we were busy */ 80219a167afSAl Viro return 0; 803ea5b778aSDavid Howells case 0: 8048aef1884SAl Viro path_put(path); 8059875cf80SDavid Howells path->mnt = mnt; 8069875cf80SDavid Howells path->dentry = dget(mnt->mnt_root); 8079875cf80SDavid Howells return 0; 80819a167afSAl Viro default: 80919a167afSAl Viro return err; 8109875cf80SDavid Howells } 81119a167afSAl Viro 812ea5b778aSDavid Howells } 8139875cf80SDavid Howells 8149875cf80SDavid Howells /* 8159875cf80SDavid Howells * Handle a dentry that is managed in some way. 816cc53ce53SDavid Howells * - Flagged for transit management (autofs) 8179875cf80SDavid Howells * - Flagged as mountpoint 8189875cf80SDavid Howells * - Flagged as automount point 8199875cf80SDavid Howells * 8209875cf80SDavid Howells * This may only be called in refwalk mode. 8219875cf80SDavid Howells * 8229875cf80SDavid Howells * Serialization is taken care of in namespace.c 8239875cf80SDavid Howells */ 8249875cf80SDavid Howells static int follow_managed(struct path *path, unsigned flags) 8259875cf80SDavid Howells { 8268aef1884SAl Viro struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */ 8279875cf80SDavid Howells unsigned managed; 8289875cf80SDavid Howells bool need_mntput = false; 8298aef1884SAl Viro int ret = 0; 8309875cf80SDavid Howells 8319875cf80SDavid Howells /* Given that we're not holding a lock here, we retain the value in a 8329875cf80SDavid Howells * local variable for each dentry as we look at it so that we don't see 8339875cf80SDavid Howells * the components of that value change under us */ 8349875cf80SDavid Howells while (managed = ACCESS_ONCE(path->dentry->d_flags), 8359875cf80SDavid Howells managed &= DCACHE_MANAGED_DENTRY, 8369875cf80SDavid Howells unlikely(managed != 0)) { 837cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 838cc53ce53SDavid Howells * being held. */ 839cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 840cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 841cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 8421aed3e42SAl Viro ret = path->dentry->d_op->d_manage(path->dentry, false); 843cc53ce53SDavid Howells if (ret < 0) 8448aef1884SAl Viro break; 845cc53ce53SDavid Howells } 846cc53ce53SDavid Howells 8479875cf80SDavid Howells /* Transit to a mounted filesystem. */ 8489875cf80SDavid Howells if (managed & DCACHE_MOUNTED) { 8499875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 8509875cf80SDavid Howells if (mounted) { 8519875cf80SDavid Howells dput(path->dentry); 8529875cf80SDavid Howells if (need_mntput) 853463ffb2eSAl Viro mntput(path->mnt); 854463ffb2eSAl Viro path->mnt = mounted; 855463ffb2eSAl Viro path->dentry = dget(mounted->mnt_root); 8569875cf80SDavid Howells need_mntput = true; 8579875cf80SDavid Howells continue; 858463ffb2eSAl Viro } 859463ffb2eSAl Viro 8609875cf80SDavid Howells /* Something is mounted on this dentry in another 8619875cf80SDavid Howells * namespace and/or whatever was mounted there in this 8629875cf80SDavid Howells * namespace got unmounted before we managed to get the 8639875cf80SDavid Howells * vfsmount_lock */ 8641da177e4SLinus Torvalds } 8659875cf80SDavid Howells 8669875cf80SDavid Howells /* Handle an automount point */ 8679875cf80SDavid Howells if (managed & DCACHE_NEED_AUTOMOUNT) { 8689875cf80SDavid Howells ret = follow_automount(path, flags, &need_mntput); 8699875cf80SDavid Howells if (ret < 0) 8708aef1884SAl Viro break; 8719875cf80SDavid Howells continue; 8729875cf80SDavid Howells } 8739875cf80SDavid Howells 8749875cf80SDavid Howells /* We didn't change the current path point */ 8759875cf80SDavid Howells break; 8769875cf80SDavid Howells } 8778aef1884SAl Viro 8788aef1884SAl Viro if (need_mntput && path->mnt == mnt) 8798aef1884SAl Viro mntput(path->mnt); 8808aef1884SAl Viro if (ret == -EISDIR) 8818aef1884SAl Viro ret = 0; 882a3fbbde7SAl Viro return ret < 0 ? ret : need_mntput; 8831da177e4SLinus Torvalds } 8841da177e4SLinus Torvalds 885cc53ce53SDavid Howells int follow_down_one(struct path *path) 8861da177e4SLinus Torvalds { 8871da177e4SLinus Torvalds struct vfsmount *mounted; 8881da177e4SLinus Torvalds 8891c755af4SAl Viro mounted = lookup_mnt(path); 8901da177e4SLinus Torvalds if (mounted) { 8919393bd07SAl Viro dput(path->dentry); 8929393bd07SAl Viro mntput(path->mnt); 8939393bd07SAl Viro path->mnt = mounted; 8949393bd07SAl Viro path->dentry = dget(mounted->mnt_root); 8951da177e4SLinus Torvalds return 1; 8961da177e4SLinus Torvalds } 8971da177e4SLinus Torvalds return 0; 8981da177e4SLinus Torvalds } 8991da177e4SLinus Torvalds 90062a7375eSIan Kent static inline bool managed_dentry_might_block(struct dentry *dentry) 90162a7375eSIan Kent { 90262a7375eSIan Kent return (dentry->d_flags & DCACHE_MANAGE_TRANSIT && 90362a7375eSIan Kent dentry->d_op->d_manage(dentry, true) < 0); 90462a7375eSIan Kent } 90562a7375eSIan Kent 9069875cf80SDavid Howells /* 907287548e4SAl Viro * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if 908287548e4SAl Viro * we meet a managed dentry that would need blocking. 9099875cf80SDavid Howells */ 9109875cf80SDavid Howells static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, 911287548e4SAl Viro struct inode **inode) 9129875cf80SDavid Howells { 91362a7375eSIan Kent for (;;) { 914c7105365SAl Viro struct mount *mounted; 91562a7375eSIan Kent /* 91662a7375eSIan Kent * Don't forget we might have a non-mountpoint managed dentry 91762a7375eSIan Kent * that wants to block transit. 91862a7375eSIan Kent */ 919287548e4SAl Viro if (unlikely(managed_dentry_might_block(path->dentry))) 920ab90911fSDavid Howells return false; 92162a7375eSIan Kent 92262a7375eSIan Kent if (!d_mountpoint(path->dentry)) 92362a7375eSIan Kent break; 92462a7375eSIan Kent 9259875cf80SDavid Howells mounted = __lookup_mnt(path->mnt, path->dentry, 1); 9269875cf80SDavid Howells if (!mounted) 9279875cf80SDavid Howells break; 928c7105365SAl Viro path->mnt = &mounted->mnt; 929c7105365SAl Viro path->dentry = mounted->mnt.mnt_root; 930a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 9319875cf80SDavid Howells nd->seq = read_seqcount_begin(&path->dentry->d_seq); 93259430262SLinus Torvalds /* 93359430262SLinus Torvalds * Update the inode too. We don't need to re-check the 93459430262SLinus Torvalds * dentry sequence number here after this d_inode read, 93559430262SLinus Torvalds * because a mount-point is always pinned. 93659430262SLinus Torvalds */ 93759430262SLinus Torvalds *inode = path->dentry->d_inode; 9389875cf80SDavid Howells } 9399875cf80SDavid Howells return true; 9409875cf80SDavid Howells } 9419875cf80SDavid Howells 942dea39376SAl Viro static void follow_mount_rcu(struct nameidata *nd) 943287548e4SAl Viro { 944dea39376SAl Viro while (d_mountpoint(nd->path.dentry)) { 945c7105365SAl Viro struct mount *mounted; 946dea39376SAl Viro mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1); 947287548e4SAl Viro if (!mounted) 948287548e4SAl Viro break; 949c7105365SAl Viro nd->path.mnt = &mounted->mnt; 950c7105365SAl Viro nd->path.dentry = mounted->mnt.mnt_root; 951dea39376SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 952287548e4SAl Viro } 953287548e4SAl Viro } 954287548e4SAl Viro 95531e6b01fSNick Piggin static int follow_dotdot_rcu(struct nameidata *nd) 95631e6b01fSNick Piggin { 95731e6b01fSNick Piggin set_root_rcu(nd); 95831e6b01fSNick Piggin 95931e6b01fSNick Piggin while (1) { 96031e6b01fSNick Piggin if (nd->path.dentry == nd->root.dentry && 96131e6b01fSNick Piggin nd->path.mnt == nd->root.mnt) { 96231e6b01fSNick Piggin break; 96331e6b01fSNick Piggin } 96431e6b01fSNick Piggin if (nd->path.dentry != nd->path.mnt->mnt_root) { 96531e6b01fSNick Piggin struct dentry *old = nd->path.dentry; 96631e6b01fSNick Piggin struct dentry *parent = old->d_parent; 96731e6b01fSNick Piggin unsigned seq; 96831e6b01fSNick Piggin 96931e6b01fSNick Piggin seq = read_seqcount_begin(&parent->d_seq); 97031e6b01fSNick Piggin if (read_seqcount_retry(&old->d_seq, nd->seq)) 971ef7562d5SAl Viro goto failed; 97231e6b01fSNick Piggin nd->path.dentry = parent; 97331e6b01fSNick Piggin nd->seq = seq; 97431e6b01fSNick Piggin break; 97531e6b01fSNick Piggin } 97631e6b01fSNick Piggin if (!follow_up_rcu(&nd->path)) 97731e6b01fSNick Piggin break; 97831e6b01fSNick Piggin nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 97931e6b01fSNick Piggin } 980dea39376SAl Viro follow_mount_rcu(nd); 981dea39376SAl Viro nd->inode = nd->path.dentry->d_inode; 98231e6b01fSNick Piggin return 0; 983ef7562d5SAl Viro 984ef7562d5SAl Viro failed: 985ef7562d5SAl Viro nd->flags &= ~LOOKUP_RCU; 9865b6ca027SAl Viro if (!(nd->flags & LOOKUP_ROOT)) 987ef7562d5SAl Viro nd->root.mnt = NULL; 988ef7562d5SAl Viro rcu_read_unlock(); 989962830dfSAndi Kleen br_read_unlock(&vfsmount_lock); 990ef7562d5SAl Viro return -ECHILD; 99131e6b01fSNick Piggin } 99231e6b01fSNick Piggin 9939875cf80SDavid Howells /* 994cc53ce53SDavid Howells * Follow down to the covering mount currently visible to userspace. At each 995cc53ce53SDavid Howells * point, the filesystem owning that dentry may be queried as to whether the 996cc53ce53SDavid Howells * caller is permitted to proceed or not. 997cc53ce53SDavid Howells */ 9987cc90cc3SAl Viro int follow_down(struct path *path) 999cc53ce53SDavid Howells { 1000cc53ce53SDavid Howells unsigned managed; 1001cc53ce53SDavid Howells int ret; 1002cc53ce53SDavid Howells 1003cc53ce53SDavid Howells while (managed = ACCESS_ONCE(path->dentry->d_flags), 1004cc53ce53SDavid Howells unlikely(managed & DCACHE_MANAGED_DENTRY)) { 1005cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1006cc53ce53SDavid Howells * being held. 1007cc53ce53SDavid Howells * 1008cc53ce53SDavid Howells * We indicate to the filesystem if someone is trying to mount 1009cc53ce53SDavid Howells * something here. This gives autofs the chance to deny anyone 1010cc53ce53SDavid Howells * other than its daemon the right to mount on its 1011cc53ce53SDavid Howells * superstructure. 1012cc53ce53SDavid Howells * 1013cc53ce53SDavid Howells * The filesystem may sleep at this point. 1014cc53ce53SDavid Howells */ 1015cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 1016cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1017cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 1018ab90911fSDavid Howells ret = path->dentry->d_op->d_manage( 10191aed3e42SAl Viro path->dentry, false); 1020cc53ce53SDavid Howells if (ret < 0) 1021cc53ce53SDavid Howells return ret == -EISDIR ? 0 : ret; 1022cc53ce53SDavid Howells } 1023cc53ce53SDavid Howells 1024cc53ce53SDavid Howells /* Transit to a mounted filesystem. */ 1025cc53ce53SDavid Howells if (managed & DCACHE_MOUNTED) { 1026cc53ce53SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 1027cc53ce53SDavid Howells if (!mounted) 1028cc53ce53SDavid Howells break; 1029cc53ce53SDavid Howells dput(path->dentry); 1030cc53ce53SDavid Howells mntput(path->mnt); 1031cc53ce53SDavid Howells path->mnt = mounted; 1032cc53ce53SDavid Howells path->dentry = dget(mounted->mnt_root); 1033cc53ce53SDavid Howells continue; 1034cc53ce53SDavid Howells } 1035cc53ce53SDavid Howells 1036cc53ce53SDavid Howells /* Don't handle automount points here */ 1037cc53ce53SDavid Howells break; 1038cc53ce53SDavid Howells } 1039cc53ce53SDavid Howells return 0; 1040cc53ce53SDavid Howells } 1041cc53ce53SDavid Howells 1042cc53ce53SDavid Howells /* 10439875cf80SDavid Howells * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() 10449875cf80SDavid Howells */ 10459875cf80SDavid Howells static void follow_mount(struct path *path) 10469875cf80SDavid Howells { 10479875cf80SDavid Howells while (d_mountpoint(path->dentry)) { 10489875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 10499875cf80SDavid Howells if (!mounted) 10509875cf80SDavid Howells break; 10519875cf80SDavid Howells dput(path->dentry); 10529875cf80SDavid Howells mntput(path->mnt); 10539875cf80SDavid Howells path->mnt = mounted; 10549875cf80SDavid Howells path->dentry = dget(mounted->mnt_root); 10559875cf80SDavid Howells } 10569875cf80SDavid Howells } 10579875cf80SDavid Howells 105831e6b01fSNick Piggin static void follow_dotdot(struct nameidata *nd) 10591da177e4SLinus Torvalds { 10602a737871SAl Viro set_root(nd); 1061e518ddb7SAndreas Mohr 10621da177e4SLinus Torvalds while(1) { 10634ac91378SJan Blunck struct dentry *old = nd->path.dentry; 10641da177e4SLinus Torvalds 10652a737871SAl Viro if (nd->path.dentry == nd->root.dentry && 10662a737871SAl Viro nd->path.mnt == nd->root.mnt) { 10671da177e4SLinus Torvalds break; 10681da177e4SLinus Torvalds } 10694ac91378SJan Blunck if (nd->path.dentry != nd->path.mnt->mnt_root) { 10703088dd70SAl Viro /* rare case of legitimate dget_parent()... */ 10713088dd70SAl Viro nd->path.dentry = dget_parent(nd->path.dentry); 10721da177e4SLinus Torvalds dput(old); 10731da177e4SLinus Torvalds break; 10741da177e4SLinus Torvalds } 10753088dd70SAl Viro if (!follow_up(&nd->path)) 10761da177e4SLinus Torvalds break; 10771da177e4SLinus Torvalds } 107879ed0226SAl Viro follow_mount(&nd->path); 107931e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 10801da177e4SLinus Torvalds } 10811da177e4SLinus Torvalds 10821da177e4SLinus Torvalds /* 1083bad61189SMiklos Szeredi * This looks up the name in dcache, possibly revalidates the old dentry and 1084bad61189SMiklos Szeredi * allocates a new one if not found or not valid. In the need_lookup argument 1085bad61189SMiklos Szeredi * returns whether i_op->lookup is necessary. 1086bad61189SMiklos Szeredi * 1087bad61189SMiklos Szeredi * dir->d_inode->i_mutex must be held 1088baa03890SNick Piggin */ 1089bad61189SMiklos Szeredi static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir, 1090201f956eSAl Viro unsigned int flags, bool *need_lookup) 1091baa03890SNick Piggin { 1092baa03890SNick Piggin struct dentry *dentry; 1093bad61189SMiklos Szeredi int error; 1094baa03890SNick Piggin 1095bad61189SMiklos Szeredi *need_lookup = false; 1096bad61189SMiklos Szeredi dentry = d_lookup(dir, name); 1097bad61189SMiklos Szeredi if (dentry) { 1098bad61189SMiklos Szeredi if (d_need_lookup(dentry)) { 1099bad61189SMiklos Szeredi *need_lookup = true; 1100bad61189SMiklos Szeredi } else if (dentry->d_flags & DCACHE_OP_REVALIDATE) { 1101201f956eSAl Viro error = d_revalidate(dentry, flags); 1102bad61189SMiklos Szeredi if (unlikely(error <= 0)) { 1103bad61189SMiklos Szeredi if (error < 0) { 1104bad61189SMiklos Szeredi dput(dentry); 1105bad61189SMiklos Szeredi return ERR_PTR(error); 1106bad61189SMiklos Szeredi } else if (!d_invalidate(dentry)) { 1107bad61189SMiklos Szeredi dput(dentry); 1108bad61189SMiklos Szeredi dentry = NULL; 1109bad61189SMiklos Szeredi } 1110bad61189SMiklos Szeredi } 1111bad61189SMiklos Szeredi } 1112bad61189SMiklos Szeredi } 1113baa03890SNick Piggin 1114bad61189SMiklos Szeredi if (!dentry) { 1115bad61189SMiklos Szeredi dentry = d_alloc(dir, name); 1116baa03890SNick Piggin if (unlikely(!dentry)) 1117baa03890SNick Piggin return ERR_PTR(-ENOMEM); 1118baa03890SNick Piggin 1119bad61189SMiklos Szeredi *need_lookup = true; 1120baa03890SNick Piggin } 1121baa03890SNick Piggin return dentry; 1122baa03890SNick Piggin } 1123baa03890SNick Piggin 1124baa03890SNick Piggin /* 1125bad61189SMiklos Szeredi * Call i_op->lookup on the dentry. The dentry must be negative but may be 1126bad61189SMiklos Szeredi * hashed if it was pouplated with DCACHE_NEED_LOOKUP. 1127bad61189SMiklos Szeredi * 1128bad61189SMiklos Szeredi * dir->d_inode->i_mutex must be held 112944396f4bSJosef Bacik */ 1130bad61189SMiklos Szeredi static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry, 113172bd866aSAl Viro unsigned int flags) 113244396f4bSJosef Bacik { 113344396f4bSJosef Bacik struct dentry *old; 113444396f4bSJosef Bacik 113544396f4bSJosef Bacik /* Don't create child dentry for a dead directory. */ 1136bad61189SMiklos Szeredi if (unlikely(IS_DEADDIR(dir))) { 1137e188dc02SMiklos Szeredi dput(dentry); 113844396f4bSJosef Bacik return ERR_PTR(-ENOENT); 1139e188dc02SMiklos Szeredi } 114044396f4bSJosef Bacik 114172bd866aSAl Viro old = dir->i_op->lookup(dir, dentry, flags); 114244396f4bSJosef Bacik if (unlikely(old)) { 114344396f4bSJosef Bacik dput(dentry); 114444396f4bSJosef Bacik dentry = old; 114544396f4bSJosef Bacik } 114644396f4bSJosef Bacik return dentry; 114744396f4bSJosef Bacik } 114844396f4bSJosef Bacik 1149a3255546SAl Viro static struct dentry *__lookup_hash(struct qstr *name, 115072bd866aSAl Viro struct dentry *base, unsigned int flags) 1151a3255546SAl Viro { 1152bad61189SMiklos Szeredi bool need_lookup; 1153a3255546SAl Viro struct dentry *dentry; 1154a3255546SAl Viro 115572bd866aSAl Viro dentry = lookup_dcache(name, base, flags, &need_lookup); 1156bad61189SMiklos Szeredi if (!need_lookup) 1157a3255546SAl Viro return dentry; 1158bad61189SMiklos Szeredi 115972bd866aSAl Viro return lookup_real(base->d_inode, dentry, flags); 1160a3255546SAl Viro } 1161a3255546SAl Viro 116244396f4bSJosef Bacik /* 11631da177e4SLinus Torvalds * It's more convoluted than I'd like it to be, but... it's still fairly 11641da177e4SLinus Torvalds * small and for now I'd prefer to have fast path as straight as possible. 11651da177e4SLinus Torvalds * It _is_ time-critical. 11661da177e4SLinus Torvalds */ 1167697f514dSMiklos Szeredi static int lookup_fast(struct nameidata *nd, struct qstr *name, 116831e6b01fSNick Piggin struct path *path, struct inode **inode) 11691da177e4SLinus Torvalds { 11704ac91378SJan Blunck struct vfsmount *mnt = nd->path.mnt; 117131e6b01fSNick Piggin struct dentry *dentry, *parent = nd->path.dentry; 11725a18fff2SAl Viro int need_reval = 1; 11735a18fff2SAl Viro int status = 1; 11749875cf80SDavid Howells int err; 11759875cf80SDavid Howells 11763cac260aSAl Viro /* 1177b04f784eSNick Piggin * Rename seqlock is not required here because in the off chance 1178b04f784eSNick Piggin * of a false negative due to a concurrent rename, we're going to 1179b04f784eSNick Piggin * do the non-racy lookup, below. 1180b04f784eSNick Piggin */ 118131e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) { 118231e6b01fSNick Piggin unsigned seq; 118312f8ad4bSLinus Torvalds dentry = __d_lookup_rcu(parent, name, &seq, nd->inode); 11845a18fff2SAl Viro if (!dentry) 11855a18fff2SAl Viro goto unlazy; 11865a18fff2SAl Viro 118712f8ad4bSLinus Torvalds /* 118812f8ad4bSLinus Torvalds * This sequence count validates that the inode matches 118912f8ad4bSLinus Torvalds * the dentry name information from lookup. 119012f8ad4bSLinus Torvalds */ 119112f8ad4bSLinus Torvalds *inode = dentry->d_inode; 119212f8ad4bSLinus Torvalds if (read_seqcount_retry(&dentry->d_seq, seq)) 119312f8ad4bSLinus Torvalds return -ECHILD; 119412f8ad4bSLinus Torvalds 119512f8ad4bSLinus Torvalds /* 119612f8ad4bSLinus Torvalds * This sequence count validates that the parent had no 119712f8ad4bSLinus Torvalds * changes while we did the lookup of the dentry above. 119812f8ad4bSLinus Torvalds * 119912f8ad4bSLinus Torvalds * The memory barrier in read_seqcount_begin of child is 120012f8ad4bSLinus Torvalds * enough, we can use __read_seqcount_retry here. 120112f8ad4bSLinus Torvalds */ 120231e6b01fSNick Piggin if (__read_seqcount_retry(&parent->d_seq, nd->seq)) 120331e6b01fSNick Piggin return -ECHILD; 120431e6b01fSNick Piggin nd->seq = seq; 12055a18fff2SAl Viro 1206fa4ee159SMiklos Szeredi if (unlikely(d_need_lookup(dentry))) 1207fa4ee159SMiklos Szeredi goto unlazy; 120824643087SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) { 12094ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 12105a18fff2SAl Viro if (unlikely(status <= 0)) { 12115a18fff2SAl Viro if (status != -ECHILD) 12125a18fff2SAl Viro need_reval = 0; 12135a18fff2SAl Viro goto unlazy; 12145a18fff2SAl Viro } 121524643087SAl Viro } 121631e6b01fSNick Piggin path->mnt = mnt; 121731e6b01fSNick Piggin path->dentry = dentry; 1218d6e9bd25SAl Viro if (unlikely(!__follow_mount_rcu(nd, path, inode))) 1219d6e9bd25SAl Viro goto unlazy; 1220d6e9bd25SAl Viro if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT)) 1221d6e9bd25SAl Viro goto unlazy; 12229875cf80SDavid Howells return 0; 12235a18fff2SAl Viro unlazy: 122419660af7SAl Viro if (unlazy_walk(nd, dentry)) 12255a18fff2SAl Viro return -ECHILD; 12265a18fff2SAl Viro } else { 122731e6b01fSNick Piggin dentry = __d_lookup(parent, name); 122824643087SAl Viro } 12295a18fff2SAl Viro 123081e6f520SAl Viro if (unlikely(!dentry)) 123181e6f520SAl Viro goto need_lookup; 12325a18fff2SAl Viro 123381e6f520SAl Viro if (unlikely(d_need_lookup(dentry))) { 123481e6f520SAl Viro dput(dentry); 123581e6f520SAl Viro goto need_lookup; 12365a18fff2SAl Viro } 123781e6f520SAl Viro 12385a18fff2SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval) 12394ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 12405a18fff2SAl Viro if (unlikely(status <= 0)) { 12415a18fff2SAl Viro if (status < 0) { 12425a18fff2SAl Viro dput(dentry); 12435a18fff2SAl Viro return status; 12445a18fff2SAl Viro } 12455a18fff2SAl Viro if (!d_invalidate(dentry)) { 12465a18fff2SAl Viro dput(dentry); 124781e6f520SAl Viro goto need_lookup; 12485a18fff2SAl Viro } 12495a18fff2SAl Viro } 1250697f514dSMiklos Szeredi 12511da177e4SLinus Torvalds path->mnt = mnt; 12521da177e4SLinus Torvalds path->dentry = dentry; 12539875cf80SDavid Howells err = follow_managed(path, nd->flags); 125489312214SIan Kent if (unlikely(err < 0)) { 125589312214SIan Kent path_put_conditional(path, nd); 12569875cf80SDavid Howells return err; 125789312214SIan Kent } 1258a3fbbde7SAl Viro if (err) 1259a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 126031e6b01fSNick Piggin *inode = path->dentry->d_inode; 12611da177e4SLinus Torvalds return 0; 126281e6f520SAl Viro 126381e6f520SAl Viro need_lookup: 1264697f514dSMiklos Szeredi return 1; 1265697f514dSMiklos Szeredi } 1266697f514dSMiklos Szeredi 1267697f514dSMiklos Szeredi /* Fast lookup failed, do it the slow way */ 1268697f514dSMiklos Szeredi static int lookup_slow(struct nameidata *nd, struct qstr *name, 1269697f514dSMiklos Szeredi struct path *path) 1270697f514dSMiklos Szeredi { 1271697f514dSMiklos Szeredi struct dentry *dentry, *parent; 1272697f514dSMiklos Szeredi int err; 1273697f514dSMiklos Szeredi 1274697f514dSMiklos Szeredi parent = nd->path.dentry; 127581e6f520SAl Viro BUG_ON(nd->inode != parent->d_inode); 127681e6f520SAl Viro 127781e6f520SAl Viro mutex_lock(&parent->d_inode->i_mutex); 127872bd866aSAl Viro dentry = __lookup_hash(name, parent, nd->flags); 127981e6f520SAl Viro mutex_unlock(&parent->d_inode->i_mutex); 128081e6f520SAl Viro if (IS_ERR(dentry)) 128181e6f520SAl Viro return PTR_ERR(dentry); 1282697f514dSMiklos Szeredi path->mnt = nd->path.mnt; 1283697f514dSMiklos Szeredi path->dentry = dentry; 1284697f514dSMiklos Szeredi err = follow_managed(path, nd->flags); 1285697f514dSMiklos Szeredi if (unlikely(err < 0)) { 1286697f514dSMiklos Szeredi path_put_conditional(path, nd); 1287697f514dSMiklos Szeredi return err; 1288697f514dSMiklos Szeredi } 1289697f514dSMiklos Szeredi if (err) 1290697f514dSMiklos Szeredi nd->flags |= LOOKUP_JUMPED; 1291697f514dSMiklos Szeredi return 0; 12921da177e4SLinus Torvalds } 12931da177e4SLinus Torvalds 129452094c8aSAl Viro static inline int may_lookup(struct nameidata *nd) 129552094c8aSAl Viro { 129652094c8aSAl Viro if (nd->flags & LOOKUP_RCU) { 12974ad5abb3SAl Viro int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); 129852094c8aSAl Viro if (err != -ECHILD) 129952094c8aSAl Viro return err; 130019660af7SAl Viro if (unlazy_walk(nd, NULL)) 130152094c8aSAl Viro return -ECHILD; 130252094c8aSAl Viro } 13034ad5abb3SAl Viro return inode_permission(nd->inode, MAY_EXEC); 130452094c8aSAl Viro } 130552094c8aSAl Viro 13069856fa1bSAl Viro static inline int handle_dots(struct nameidata *nd, int type) 13079856fa1bSAl Viro { 13089856fa1bSAl Viro if (type == LAST_DOTDOT) { 13099856fa1bSAl Viro if (nd->flags & LOOKUP_RCU) { 13109856fa1bSAl Viro if (follow_dotdot_rcu(nd)) 13119856fa1bSAl Viro return -ECHILD; 13129856fa1bSAl Viro } else 13139856fa1bSAl Viro follow_dotdot(nd); 13149856fa1bSAl Viro } 13159856fa1bSAl Viro return 0; 13169856fa1bSAl Viro } 13179856fa1bSAl Viro 1318951361f9SAl Viro static void terminate_walk(struct nameidata *nd) 1319951361f9SAl Viro { 1320951361f9SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1321951361f9SAl Viro path_put(&nd->path); 1322951361f9SAl Viro } else { 1323951361f9SAl Viro nd->flags &= ~LOOKUP_RCU; 13245b6ca027SAl Viro if (!(nd->flags & LOOKUP_ROOT)) 1325951361f9SAl Viro nd->root.mnt = NULL; 1326951361f9SAl Viro rcu_read_unlock(); 1327962830dfSAndi Kleen br_read_unlock(&vfsmount_lock); 1328951361f9SAl Viro } 1329951361f9SAl Viro } 1330951361f9SAl Viro 13313ddcd056SLinus Torvalds /* 13323ddcd056SLinus Torvalds * Do we need to follow links? We _really_ want to be able 13333ddcd056SLinus Torvalds * to do this check without having to look at inode->i_op, 13343ddcd056SLinus Torvalds * so we keep a cache of "no, this doesn't need follow_link" 13353ddcd056SLinus Torvalds * for the common case. 13363ddcd056SLinus Torvalds */ 13377813b94aSLinus Torvalds static inline int should_follow_link(struct inode *inode, int follow) 13383ddcd056SLinus Torvalds { 13393ddcd056SLinus Torvalds if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) { 13403ddcd056SLinus Torvalds if (likely(inode->i_op->follow_link)) 13413ddcd056SLinus Torvalds return follow; 13423ddcd056SLinus Torvalds 13433ddcd056SLinus Torvalds /* This gets set once for the inode lifetime */ 13443ddcd056SLinus Torvalds spin_lock(&inode->i_lock); 13453ddcd056SLinus Torvalds inode->i_opflags |= IOP_NOFOLLOW; 13463ddcd056SLinus Torvalds spin_unlock(&inode->i_lock); 13473ddcd056SLinus Torvalds } 13483ddcd056SLinus Torvalds return 0; 13493ddcd056SLinus Torvalds } 13503ddcd056SLinus Torvalds 1351ce57dfc1SAl Viro static inline int walk_component(struct nameidata *nd, struct path *path, 1352ce57dfc1SAl Viro struct qstr *name, int type, int follow) 1353ce57dfc1SAl Viro { 1354ce57dfc1SAl Viro struct inode *inode; 1355ce57dfc1SAl Viro int err; 1356ce57dfc1SAl Viro /* 1357ce57dfc1SAl Viro * "." and ".." are special - ".." especially so because it has 1358ce57dfc1SAl Viro * to be able to know about the current root directory and 1359ce57dfc1SAl Viro * parent relationships. 1360ce57dfc1SAl Viro */ 1361ce57dfc1SAl Viro if (unlikely(type != LAST_NORM)) 1362ce57dfc1SAl Viro return handle_dots(nd, type); 1363697f514dSMiklos Szeredi err = lookup_fast(nd, name, path, &inode); 1364ce57dfc1SAl Viro if (unlikely(err)) { 1365697f514dSMiklos Szeredi if (err < 0) 1366697f514dSMiklos Szeredi goto out_err; 1367697f514dSMiklos Szeredi 1368697f514dSMiklos Szeredi err = lookup_slow(nd, name, path); 1369697f514dSMiklos Szeredi if (err < 0) 1370697f514dSMiklos Szeredi goto out_err; 1371697f514dSMiklos Szeredi 1372697f514dSMiklos Szeredi inode = path->dentry->d_inode; 1373ce57dfc1SAl Viro } 1374697f514dSMiklos Szeredi err = -ENOENT; 1375697f514dSMiklos Szeredi if (!inode) 1376697f514dSMiklos Szeredi goto out_path_put; 1377697f514dSMiklos Szeredi 13787813b94aSLinus Torvalds if (should_follow_link(inode, follow)) { 137919660af7SAl Viro if (nd->flags & LOOKUP_RCU) { 138019660af7SAl Viro if (unlikely(unlazy_walk(nd, path->dentry))) { 1381697f514dSMiklos Szeredi err = -ECHILD; 1382697f514dSMiklos Szeredi goto out_err; 138319660af7SAl Viro } 138419660af7SAl Viro } 1385ce57dfc1SAl Viro BUG_ON(inode != path->dentry->d_inode); 1386ce57dfc1SAl Viro return 1; 1387ce57dfc1SAl Viro } 1388ce57dfc1SAl Viro path_to_nameidata(path, nd); 1389ce57dfc1SAl Viro nd->inode = inode; 1390ce57dfc1SAl Viro return 0; 1391697f514dSMiklos Szeredi 1392697f514dSMiklos Szeredi out_path_put: 1393697f514dSMiklos Szeredi path_to_nameidata(path, nd); 1394697f514dSMiklos Szeredi out_err: 1395697f514dSMiklos Szeredi terminate_walk(nd); 1396697f514dSMiklos Szeredi return err; 1397ce57dfc1SAl Viro } 1398ce57dfc1SAl Viro 13991da177e4SLinus Torvalds /* 1400b356379aSAl Viro * This limits recursive symlink follows to 8, while 1401b356379aSAl Viro * limiting consecutive symlinks to 40. 1402b356379aSAl Viro * 1403b356379aSAl Viro * Without that kind of total limit, nasty chains of consecutive 1404b356379aSAl Viro * symlinks can cause almost arbitrarily long lookups. 1405b356379aSAl Viro */ 1406b356379aSAl Viro static inline int nested_symlink(struct path *path, struct nameidata *nd) 1407b356379aSAl Viro { 1408b356379aSAl Viro int res; 1409b356379aSAl Viro 1410b356379aSAl Viro if (unlikely(current->link_count >= MAX_NESTED_LINKS)) { 1411b356379aSAl Viro path_put_conditional(path, nd); 1412b356379aSAl Viro path_put(&nd->path); 1413b356379aSAl Viro return -ELOOP; 1414b356379aSAl Viro } 14151a4022f8SErez Zadok BUG_ON(nd->depth >= MAX_NESTED_LINKS); 1416b356379aSAl Viro 1417b356379aSAl Viro nd->depth++; 1418b356379aSAl Viro current->link_count++; 1419b356379aSAl Viro 1420b356379aSAl Viro do { 1421b356379aSAl Viro struct path link = *path; 1422b356379aSAl Viro void *cookie; 1423574197e0SAl Viro 1424574197e0SAl Viro res = follow_link(&link, nd, &cookie); 14256d7b5aaeSAl Viro if (res) 14266d7b5aaeSAl Viro break; 1427b356379aSAl Viro res = walk_component(nd, path, &nd->last, 1428b356379aSAl Viro nd->last_type, LOOKUP_FOLLOW); 1429574197e0SAl Viro put_link(nd, &link, cookie); 1430b356379aSAl Viro } while (res > 0); 1431b356379aSAl Viro 1432b356379aSAl Viro current->link_count--; 1433b356379aSAl Viro nd->depth--; 1434b356379aSAl Viro return res; 1435b356379aSAl Viro } 1436b356379aSAl Viro 1437b356379aSAl Viro /* 14383ddcd056SLinus Torvalds * We really don't want to look at inode->i_op->lookup 14393ddcd056SLinus Torvalds * when we don't have to. So we keep a cache bit in 14403ddcd056SLinus Torvalds * the inode ->i_opflags field that says "yes, we can 14413ddcd056SLinus Torvalds * do lookup on this inode". 14423ddcd056SLinus Torvalds */ 14433ddcd056SLinus Torvalds static inline int can_lookup(struct inode *inode) 14443ddcd056SLinus Torvalds { 14453ddcd056SLinus Torvalds if (likely(inode->i_opflags & IOP_LOOKUP)) 14463ddcd056SLinus Torvalds return 1; 14473ddcd056SLinus Torvalds if (likely(!inode->i_op->lookup)) 14483ddcd056SLinus Torvalds return 0; 14493ddcd056SLinus Torvalds 14503ddcd056SLinus Torvalds /* We do this once for the lifetime of the inode */ 14513ddcd056SLinus Torvalds spin_lock(&inode->i_lock); 14523ddcd056SLinus Torvalds inode->i_opflags |= IOP_LOOKUP; 14533ddcd056SLinus Torvalds spin_unlock(&inode->i_lock); 14543ddcd056SLinus Torvalds return 1; 14553ddcd056SLinus Torvalds } 14563ddcd056SLinus Torvalds 1457bfcfaa77SLinus Torvalds /* 1458bfcfaa77SLinus Torvalds * We can do the critical dentry name comparison and hashing 1459bfcfaa77SLinus Torvalds * operations one word at a time, but we are limited to: 1460bfcfaa77SLinus Torvalds * 1461bfcfaa77SLinus Torvalds * - Architectures with fast unaligned word accesses. We could 1462bfcfaa77SLinus Torvalds * do a "get_unaligned()" if this helps and is sufficiently 1463bfcfaa77SLinus Torvalds * fast. 1464bfcfaa77SLinus Torvalds * 1465bfcfaa77SLinus Torvalds * - Little-endian machines (so that we can generate the mask 1466bfcfaa77SLinus Torvalds * of low bytes efficiently). Again, we *could* do a byte 1467bfcfaa77SLinus Torvalds * swapping load on big-endian architectures if that is not 1468bfcfaa77SLinus Torvalds * expensive enough to make the optimization worthless. 1469bfcfaa77SLinus Torvalds * 1470bfcfaa77SLinus Torvalds * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we 1471bfcfaa77SLinus Torvalds * do not trap on the (extremely unlikely) case of a page 1472bfcfaa77SLinus Torvalds * crossing operation. 1473bfcfaa77SLinus Torvalds * 1474bfcfaa77SLinus Torvalds * - Furthermore, we need an efficient 64-bit compile for the 1475bfcfaa77SLinus Torvalds * 64-bit case in order to generate the "number of bytes in 1476bfcfaa77SLinus Torvalds * the final mask". Again, that could be replaced with a 1477bfcfaa77SLinus Torvalds * efficient population count instruction or similar. 1478bfcfaa77SLinus Torvalds */ 1479bfcfaa77SLinus Torvalds #ifdef CONFIG_DCACHE_WORD_ACCESS 1480bfcfaa77SLinus Torvalds 1481f68e556eSLinus Torvalds #include <asm/word-at-a-time.h> 1482bfcfaa77SLinus Torvalds 1483f68e556eSLinus Torvalds #ifdef CONFIG_64BIT 1484bfcfaa77SLinus Torvalds 1485bfcfaa77SLinus Torvalds static inline unsigned int fold_hash(unsigned long hash) 1486bfcfaa77SLinus Torvalds { 1487bfcfaa77SLinus Torvalds hash += hash >> (8*sizeof(int)); 1488bfcfaa77SLinus Torvalds return hash; 1489bfcfaa77SLinus Torvalds } 1490bfcfaa77SLinus Torvalds 1491bfcfaa77SLinus Torvalds #else /* 32-bit case */ 1492bfcfaa77SLinus Torvalds 1493bfcfaa77SLinus Torvalds #define fold_hash(x) (x) 1494bfcfaa77SLinus Torvalds 1495bfcfaa77SLinus Torvalds #endif 1496bfcfaa77SLinus Torvalds 1497bfcfaa77SLinus Torvalds unsigned int full_name_hash(const unsigned char *name, unsigned int len) 1498bfcfaa77SLinus Torvalds { 1499bfcfaa77SLinus Torvalds unsigned long a, mask; 1500bfcfaa77SLinus Torvalds unsigned long hash = 0; 1501bfcfaa77SLinus Torvalds 1502bfcfaa77SLinus Torvalds for (;;) { 1503e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name); 1504bfcfaa77SLinus Torvalds if (len < sizeof(unsigned long)) 1505bfcfaa77SLinus Torvalds break; 1506bfcfaa77SLinus Torvalds hash += a; 1507f132c5beSAl Viro hash *= 9; 1508bfcfaa77SLinus Torvalds name += sizeof(unsigned long); 1509bfcfaa77SLinus Torvalds len -= sizeof(unsigned long); 1510bfcfaa77SLinus Torvalds if (!len) 1511bfcfaa77SLinus Torvalds goto done; 1512bfcfaa77SLinus Torvalds } 1513bfcfaa77SLinus Torvalds mask = ~(~0ul << len*8); 1514bfcfaa77SLinus Torvalds hash += mask & a; 1515bfcfaa77SLinus Torvalds done: 1516bfcfaa77SLinus Torvalds return fold_hash(hash); 1517bfcfaa77SLinus Torvalds } 1518bfcfaa77SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 1519bfcfaa77SLinus Torvalds 1520bfcfaa77SLinus Torvalds /* 1521bfcfaa77SLinus Torvalds * Calculate the length and hash of the path component, and 1522bfcfaa77SLinus Torvalds * return the length of the component; 1523bfcfaa77SLinus Torvalds */ 1524bfcfaa77SLinus Torvalds static inline unsigned long hash_name(const char *name, unsigned int *hashp) 1525bfcfaa77SLinus Torvalds { 152636126f8fSLinus Torvalds unsigned long a, b, adata, bdata, mask, hash, len; 152736126f8fSLinus Torvalds const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 1528bfcfaa77SLinus Torvalds 1529bfcfaa77SLinus Torvalds hash = a = 0; 1530bfcfaa77SLinus Torvalds len = -sizeof(unsigned long); 1531bfcfaa77SLinus Torvalds do { 1532bfcfaa77SLinus Torvalds hash = (hash + a) * 9; 1533bfcfaa77SLinus Torvalds len += sizeof(unsigned long); 1534e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name+len); 153536126f8fSLinus Torvalds b = a ^ REPEAT_BYTE('/'); 153636126f8fSLinus Torvalds } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); 1537bfcfaa77SLinus Torvalds 153836126f8fSLinus Torvalds adata = prep_zero_mask(a, adata, &constants); 153936126f8fSLinus Torvalds bdata = prep_zero_mask(b, bdata, &constants); 154036126f8fSLinus Torvalds 154136126f8fSLinus Torvalds mask = create_zero_mask(adata | bdata); 154236126f8fSLinus Torvalds 154336126f8fSLinus Torvalds hash += a & zero_bytemask(mask); 1544bfcfaa77SLinus Torvalds *hashp = fold_hash(hash); 1545bfcfaa77SLinus Torvalds 154636126f8fSLinus Torvalds return len + find_zero(mask); 1547bfcfaa77SLinus Torvalds } 1548bfcfaa77SLinus Torvalds 1549bfcfaa77SLinus Torvalds #else 1550bfcfaa77SLinus Torvalds 15510145acc2SLinus Torvalds unsigned int full_name_hash(const unsigned char *name, unsigned int len) 15520145acc2SLinus Torvalds { 15530145acc2SLinus Torvalds unsigned long hash = init_name_hash(); 15540145acc2SLinus Torvalds while (len--) 15550145acc2SLinus Torvalds hash = partial_name_hash(*name++, hash); 15560145acc2SLinus Torvalds return end_name_hash(hash); 15570145acc2SLinus Torvalds } 1558ae942ae7SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 15590145acc2SLinus Torvalds 15603ddcd056SLinus Torvalds /* 1561200e9ef7SLinus Torvalds * We know there's a real path component here of at least 1562200e9ef7SLinus Torvalds * one character. 1563200e9ef7SLinus Torvalds */ 1564200e9ef7SLinus Torvalds static inline unsigned long hash_name(const char *name, unsigned int *hashp) 1565200e9ef7SLinus Torvalds { 1566200e9ef7SLinus Torvalds unsigned long hash = init_name_hash(); 1567200e9ef7SLinus Torvalds unsigned long len = 0, c; 1568200e9ef7SLinus Torvalds 1569200e9ef7SLinus Torvalds c = (unsigned char)*name; 1570200e9ef7SLinus Torvalds do { 1571200e9ef7SLinus Torvalds len++; 1572200e9ef7SLinus Torvalds hash = partial_name_hash(c, hash); 1573200e9ef7SLinus Torvalds c = (unsigned char)name[len]; 1574200e9ef7SLinus Torvalds } while (c && c != '/'); 1575200e9ef7SLinus Torvalds *hashp = end_name_hash(hash); 1576200e9ef7SLinus Torvalds return len; 1577200e9ef7SLinus Torvalds } 1578200e9ef7SLinus Torvalds 1579bfcfaa77SLinus Torvalds #endif 1580bfcfaa77SLinus Torvalds 1581200e9ef7SLinus Torvalds /* 15821da177e4SLinus Torvalds * Name resolution. 1583ea3834d9SPrasanna Meda * This is the basic name resolution function, turning a pathname into 1584ea3834d9SPrasanna Meda * the final dentry. We expect 'base' to be positive and a directory. 15851da177e4SLinus Torvalds * 1586ea3834d9SPrasanna Meda * Returns 0 and nd will have valid dentry and mnt on success. 1587ea3834d9SPrasanna Meda * Returns error and drops reference to input namei data on failure. 15881da177e4SLinus Torvalds */ 15896de88d72SAl Viro static int link_path_walk(const char *name, struct nameidata *nd) 15901da177e4SLinus Torvalds { 15911da177e4SLinus Torvalds struct path next; 15921da177e4SLinus Torvalds int err; 15931da177e4SLinus Torvalds 15941da177e4SLinus Torvalds while (*name=='/') 15951da177e4SLinus Torvalds name++; 15961da177e4SLinus Torvalds if (!*name) 1597086e183aSAl Viro return 0; 15981da177e4SLinus Torvalds 15991da177e4SLinus Torvalds /* At this point we know we have a real path component. */ 16001da177e4SLinus Torvalds for(;;) { 16011da177e4SLinus Torvalds struct qstr this; 1602200e9ef7SLinus Torvalds long len; 1603fe479a58SAl Viro int type; 16041da177e4SLinus Torvalds 160552094c8aSAl Viro err = may_lookup(nd); 16061da177e4SLinus Torvalds if (err) 16071da177e4SLinus Torvalds break; 16081da177e4SLinus Torvalds 1609200e9ef7SLinus Torvalds len = hash_name(name, &this.hash); 16101da177e4SLinus Torvalds this.name = name; 1611200e9ef7SLinus Torvalds this.len = len; 16121da177e4SLinus Torvalds 1613fe479a58SAl Viro type = LAST_NORM; 1614200e9ef7SLinus Torvalds if (name[0] == '.') switch (len) { 1615fe479a58SAl Viro case 2: 1616200e9ef7SLinus Torvalds if (name[1] == '.') { 1617fe479a58SAl Viro type = LAST_DOTDOT; 161816c2cd71SAl Viro nd->flags |= LOOKUP_JUMPED; 161916c2cd71SAl Viro } 1620fe479a58SAl Viro break; 1621fe479a58SAl Viro case 1: 1622fe479a58SAl Viro type = LAST_DOT; 1623fe479a58SAl Viro } 16245a202bcdSAl Viro if (likely(type == LAST_NORM)) { 16255a202bcdSAl Viro struct dentry *parent = nd->path.dentry; 162616c2cd71SAl Viro nd->flags &= ~LOOKUP_JUMPED; 16275a202bcdSAl Viro if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { 16285a202bcdSAl Viro err = parent->d_op->d_hash(parent, nd->inode, 16295a202bcdSAl Viro &this); 16305a202bcdSAl Viro if (err < 0) 16315a202bcdSAl Viro break; 16325a202bcdSAl Viro } 16335a202bcdSAl Viro } 1634fe479a58SAl Viro 1635200e9ef7SLinus Torvalds if (!name[len]) 16361da177e4SLinus Torvalds goto last_component; 1637200e9ef7SLinus Torvalds /* 1638200e9ef7SLinus Torvalds * If it wasn't NUL, we know it was '/'. Skip that 1639200e9ef7SLinus Torvalds * slash, and continue until no more slashes. 1640200e9ef7SLinus Torvalds */ 1641200e9ef7SLinus Torvalds do { 1642200e9ef7SLinus Torvalds len++; 1643200e9ef7SLinus Torvalds } while (unlikely(name[len] == '/')); 1644200e9ef7SLinus Torvalds if (!name[len]) 1645b356379aSAl Viro goto last_component; 1646200e9ef7SLinus Torvalds name += len; 16471da177e4SLinus Torvalds 1648ce57dfc1SAl Viro err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW); 1649ce57dfc1SAl Viro if (err < 0) 1650ce57dfc1SAl Viro return err; 1651fe479a58SAl Viro 1652ce57dfc1SAl Viro if (err) { 1653b356379aSAl Viro err = nested_symlink(&next, nd); 16541da177e4SLinus Torvalds if (err) 1655a7472babSAl Viro return err; 165631e6b01fSNick Piggin } 16573ddcd056SLinus Torvalds if (can_lookup(nd->inode)) 16581da177e4SLinus Torvalds continue; 16593ddcd056SLinus Torvalds err = -ENOTDIR; 16603ddcd056SLinus Torvalds break; 16611da177e4SLinus Torvalds /* here ends the main loop */ 16621da177e4SLinus Torvalds 16631da177e4SLinus Torvalds last_component: 1664ce57dfc1SAl Viro nd->last = this; 1665ce57dfc1SAl Viro nd->last_type = type; 1666ce57dfc1SAl Viro return 0; 1667ce57dfc1SAl Viro } 1668951361f9SAl Viro terminate_walk(nd); 16691da177e4SLinus Torvalds return err; 16701da177e4SLinus Torvalds } 16711da177e4SLinus Torvalds 167270e9b357SAl Viro static int path_init(int dfd, const char *name, unsigned int flags, 167370e9b357SAl Viro struct nameidata *nd, struct file **fp) 167431e6b01fSNick Piggin { 167531e6b01fSNick Piggin int retval = 0; 167631e6b01fSNick Piggin int fput_needed; 167731e6b01fSNick Piggin struct file *file; 167831e6b01fSNick Piggin 167931e6b01fSNick Piggin nd->last_type = LAST_ROOT; /* if there are only slashes... */ 168016c2cd71SAl Viro nd->flags = flags | LOOKUP_JUMPED; 168131e6b01fSNick Piggin nd->depth = 0; 16825b6ca027SAl Viro if (flags & LOOKUP_ROOT) { 16835b6ca027SAl Viro struct inode *inode = nd->root.dentry->d_inode; 168473d049a4SAl Viro if (*name) { 16855b6ca027SAl Viro if (!inode->i_op->lookup) 16865b6ca027SAl Viro return -ENOTDIR; 16875b6ca027SAl Viro retval = inode_permission(inode, MAY_EXEC); 16885b6ca027SAl Viro if (retval) 16895b6ca027SAl Viro return retval; 169073d049a4SAl Viro } 16915b6ca027SAl Viro nd->path = nd->root; 16925b6ca027SAl Viro nd->inode = inode; 16935b6ca027SAl Viro if (flags & LOOKUP_RCU) { 1694962830dfSAndi Kleen br_read_lock(&vfsmount_lock); 16955b6ca027SAl Viro rcu_read_lock(); 16965b6ca027SAl Viro nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 16975b6ca027SAl Viro } else { 16985b6ca027SAl Viro path_get(&nd->path); 16995b6ca027SAl Viro } 17005b6ca027SAl Viro return 0; 17015b6ca027SAl Viro } 17025b6ca027SAl Viro 170331e6b01fSNick Piggin nd->root.mnt = NULL; 170431e6b01fSNick Piggin 170531e6b01fSNick Piggin if (*name=='/') { 1706e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 1707962830dfSAndi Kleen br_read_lock(&vfsmount_lock); 170831e6b01fSNick Piggin rcu_read_lock(); 1709e41f7d4eSAl Viro set_root_rcu(nd); 1710e41f7d4eSAl Viro } else { 1711e41f7d4eSAl Viro set_root(nd); 1712e41f7d4eSAl Viro path_get(&nd->root); 1713e41f7d4eSAl Viro } 171431e6b01fSNick Piggin nd->path = nd->root; 171531e6b01fSNick Piggin } else if (dfd == AT_FDCWD) { 1716e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 171731e6b01fSNick Piggin struct fs_struct *fs = current->fs; 1718c28cc364SNick Piggin unsigned seq; 171931e6b01fSNick Piggin 1720962830dfSAndi Kleen br_read_lock(&vfsmount_lock); 172131e6b01fSNick Piggin rcu_read_lock(); 172231e6b01fSNick Piggin 1723c28cc364SNick Piggin do { 1724c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 172531e6b01fSNick Piggin nd->path = fs->pwd; 1726c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 1727c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 1728e41f7d4eSAl Viro } else { 1729e41f7d4eSAl Viro get_fs_pwd(current->fs, &nd->path); 1730e41f7d4eSAl Viro } 173131e6b01fSNick Piggin } else { 173231e6b01fSNick Piggin struct dentry *dentry; 173331e6b01fSNick Piggin 17341abf0c71SAl Viro file = fget_raw_light(dfd, &fput_needed); 173531e6b01fSNick Piggin retval = -EBADF; 173631e6b01fSNick Piggin if (!file) 173731e6b01fSNick Piggin goto out_fail; 173831e6b01fSNick Piggin 173931e6b01fSNick Piggin dentry = file->f_path.dentry; 174031e6b01fSNick Piggin 1741f52e0c11SAl Viro if (*name) { 174231e6b01fSNick Piggin retval = -ENOTDIR; 174331e6b01fSNick Piggin if (!S_ISDIR(dentry->d_inode->i_mode)) 174431e6b01fSNick Piggin goto fput_fail; 174531e6b01fSNick Piggin 17464ad5abb3SAl Viro retval = inode_permission(dentry->d_inode, MAY_EXEC); 174731e6b01fSNick Piggin if (retval) 174831e6b01fSNick Piggin goto fput_fail; 1749f52e0c11SAl Viro } 175031e6b01fSNick Piggin 175131e6b01fSNick Piggin nd->path = file->f_path; 1752e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 175331e6b01fSNick Piggin if (fput_needed) 175470e9b357SAl Viro *fp = file; 1755c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 1756962830dfSAndi Kleen br_read_lock(&vfsmount_lock); 175731e6b01fSNick Piggin rcu_read_lock(); 17585590ff0dSUlrich Drepper } else { 17595dd784d0SJan Blunck path_get(&file->f_path); 17605590ff0dSUlrich Drepper fput_light(file, fput_needed); 17611da177e4SLinus Torvalds } 1762e41f7d4eSAl Viro } 1763e41f7d4eSAl Viro 176431e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 17659b4a9b14SAl Viro return 0; 17662dfdd266SJosef 'Jeff' Sipek 17679b4a9b14SAl Viro fput_fail: 17689b4a9b14SAl Viro fput_light(file, fput_needed); 17699b4a9b14SAl Viro out_fail: 17709b4a9b14SAl Viro return retval; 17719b4a9b14SAl Viro } 17729b4a9b14SAl Viro 1773bd92d7feSAl Viro static inline int lookup_last(struct nameidata *nd, struct path *path) 1774bd92d7feSAl Viro { 1775bd92d7feSAl Viro if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) 1776bd92d7feSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 1777bd92d7feSAl Viro 1778bd92d7feSAl Viro nd->flags &= ~LOOKUP_PARENT; 1779bd92d7feSAl Viro return walk_component(nd, path, &nd->last, nd->last_type, 1780bd92d7feSAl Viro nd->flags & LOOKUP_FOLLOW); 1781bd92d7feSAl Viro } 1782bd92d7feSAl Viro 17839b4a9b14SAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 1784ee0827cdSAl Viro static int path_lookupat(int dfd, const char *name, 17859b4a9b14SAl Viro unsigned int flags, struct nameidata *nd) 17869b4a9b14SAl Viro { 178770e9b357SAl Viro struct file *base = NULL; 1788bd92d7feSAl Viro struct path path; 1789bd92d7feSAl Viro int err; 179031e6b01fSNick Piggin 179131e6b01fSNick Piggin /* 179231e6b01fSNick Piggin * Path walking is largely split up into 2 different synchronisation 179331e6b01fSNick Piggin * schemes, rcu-walk and ref-walk (explained in 179431e6b01fSNick Piggin * Documentation/filesystems/path-lookup.txt). These share much of the 179531e6b01fSNick Piggin * path walk code, but some things particularly setup, cleanup, and 179631e6b01fSNick Piggin * following mounts are sufficiently divergent that functions are 179731e6b01fSNick Piggin * duplicated. Typically there is a function foo(), and its RCU 179831e6b01fSNick Piggin * analogue, foo_rcu(). 179931e6b01fSNick Piggin * 180031e6b01fSNick Piggin * -ECHILD is the error number of choice (just to avoid clashes) that 180131e6b01fSNick Piggin * is returned if some aspect of an rcu-walk fails. Such an error must 180231e6b01fSNick Piggin * be handled by restarting a traditional ref-walk (which will always 180331e6b01fSNick Piggin * be able to complete). 180431e6b01fSNick Piggin */ 1805bd92d7feSAl Viro err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base); 1806ee0827cdSAl Viro 1807bd92d7feSAl Viro if (unlikely(err)) 1808bd92d7feSAl Viro return err; 1809ee0827cdSAl Viro 1810ee0827cdSAl Viro current->total_link_count = 0; 1811bd92d7feSAl Viro err = link_path_walk(name, nd); 1812bd92d7feSAl Viro 1813bd92d7feSAl Viro if (!err && !(flags & LOOKUP_PARENT)) { 1814bd92d7feSAl Viro err = lookup_last(nd, &path); 1815bd92d7feSAl Viro while (err > 0) { 1816bd92d7feSAl Viro void *cookie; 1817bd92d7feSAl Viro struct path link = path; 1818bd92d7feSAl Viro nd->flags |= LOOKUP_PARENT; 1819574197e0SAl Viro err = follow_link(&link, nd, &cookie); 18206d7b5aaeSAl Viro if (err) 18216d7b5aaeSAl Viro break; 1822bd92d7feSAl Viro err = lookup_last(nd, &path); 1823574197e0SAl Viro put_link(nd, &link, cookie); 1824bd92d7feSAl Viro } 1825bd92d7feSAl Viro } 1826ee0827cdSAl Viro 18279f1fafeeSAl Viro if (!err) 18289f1fafeeSAl Viro err = complete_walk(nd); 1829bd92d7feSAl Viro 1830bd92d7feSAl Viro if (!err && nd->flags & LOOKUP_DIRECTORY) { 1831bd92d7feSAl Viro if (!nd->inode->i_op->lookup) { 1832bd92d7feSAl Viro path_put(&nd->path); 1833bd23a539SAl Viro err = -ENOTDIR; 1834bd92d7feSAl Viro } 1835bd92d7feSAl Viro } 183616c2cd71SAl Viro 183770e9b357SAl Viro if (base) 183870e9b357SAl Viro fput(base); 1839ee0827cdSAl Viro 18405b6ca027SAl Viro if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 184131e6b01fSNick Piggin path_put(&nd->root); 184231e6b01fSNick Piggin nd->root.mnt = NULL; 184331e6b01fSNick Piggin } 1844bd92d7feSAl Viro return err; 184531e6b01fSNick Piggin } 184631e6b01fSNick Piggin 1847ee0827cdSAl Viro static int do_path_lookup(int dfd, const char *name, 1848ee0827cdSAl Viro unsigned int flags, struct nameidata *nd) 1849ee0827cdSAl Viro { 1850ee0827cdSAl Viro int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd); 1851ee0827cdSAl Viro if (unlikely(retval == -ECHILD)) 1852ee0827cdSAl Viro retval = path_lookupat(dfd, name, flags, nd); 1853ee0827cdSAl Viro if (unlikely(retval == -ESTALE)) 1854ee0827cdSAl Viro retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd); 1855ee0827cdSAl Viro 185631e6b01fSNick Piggin if (likely(!retval)) { 185731e6b01fSNick Piggin if (unlikely(!audit_dummy_context())) { 185831e6b01fSNick Piggin if (nd->path.dentry && nd->inode) 185931e6b01fSNick Piggin audit_inode(name, nd->path.dentry); 186031e6b01fSNick Piggin } 186131e6b01fSNick Piggin } 1862170aa3d0SUlrich Drepper return retval; 18631da177e4SLinus Torvalds } 18641da177e4SLinus Torvalds 186579714f72SAl Viro /* does lookup, returns the object with parent locked */ 186679714f72SAl Viro struct dentry *kern_path_locked(const char *name, struct path *path) 18675590ff0dSUlrich Drepper { 186879714f72SAl Viro struct nameidata nd; 186979714f72SAl Viro struct dentry *d; 187079714f72SAl Viro int err = do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, &nd); 187179714f72SAl Viro if (err) 187279714f72SAl Viro return ERR_PTR(err); 187379714f72SAl Viro if (nd.last_type != LAST_NORM) { 187479714f72SAl Viro path_put(&nd.path); 187579714f72SAl Viro return ERR_PTR(-EINVAL); 187679714f72SAl Viro } 187779714f72SAl Viro mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 187879714f72SAl Viro d = lookup_one_len(nd.last.name, nd.path.dentry, nd.last.len); 187979714f72SAl Viro if (IS_ERR(d)) { 188079714f72SAl Viro mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 188179714f72SAl Viro path_put(&nd.path); 188279714f72SAl Viro return d; 188379714f72SAl Viro } 188479714f72SAl Viro *path = nd.path; 188579714f72SAl Viro return d; 18865590ff0dSUlrich Drepper } 18875590ff0dSUlrich Drepper 1888d1811465SAl Viro int kern_path(const char *name, unsigned int flags, struct path *path) 1889d1811465SAl Viro { 1890d1811465SAl Viro struct nameidata nd; 1891d1811465SAl Viro int res = do_path_lookup(AT_FDCWD, name, flags, &nd); 1892d1811465SAl Viro if (!res) 1893d1811465SAl Viro *path = nd.path; 1894d1811465SAl Viro return res; 1895d1811465SAl Viro } 1896d1811465SAl Viro 189716f18200SJosef 'Jeff' Sipek /** 189816f18200SJosef 'Jeff' Sipek * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair 189916f18200SJosef 'Jeff' Sipek * @dentry: pointer to dentry of the base directory 190016f18200SJosef 'Jeff' Sipek * @mnt: pointer to vfs mount of the base directory 190116f18200SJosef 'Jeff' Sipek * @name: pointer to file name 190216f18200SJosef 'Jeff' Sipek * @flags: lookup flags 1903e0a01249SAl Viro * @path: pointer to struct path to fill 190416f18200SJosef 'Jeff' Sipek */ 190516f18200SJosef 'Jeff' Sipek int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, 190616f18200SJosef 'Jeff' Sipek const char *name, unsigned int flags, 1907e0a01249SAl Viro struct path *path) 190816f18200SJosef 'Jeff' Sipek { 1909e0a01249SAl Viro struct nameidata nd; 1910e0a01249SAl Viro int err; 1911e0a01249SAl Viro nd.root.dentry = dentry; 1912e0a01249SAl Viro nd.root.mnt = mnt; 1913e0a01249SAl Viro BUG_ON(flags & LOOKUP_PARENT); 19145b6ca027SAl Viro /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */ 1915e0a01249SAl Viro err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd); 1916e0a01249SAl Viro if (!err) 1917e0a01249SAl Viro *path = nd.path; 1918e0a01249SAl Viro return err; 191916f18200SJosef 'Jeff' Sipek } 192016f18200SJosef 'Jeff' Sipek 1921057f6c01SJames Morris /* 1922057f6c01SJames Morris * Restricted form of lookup. Doesn't follow links, single-component only, 1923057f6c01SJames Morris * needs parent already locked. Doesn't follow mounts. 1924057f6c01SJames Morris * SMP-safe. 1925057f6c01SJames Morris */ 1926a244e169SAdrian Bunk static struct dentry *lookup_hash(struct nameidata *nd) 19271da177e4SLinus Torvalds { 192872bd866aSAl Viro return __lookup_hash(&nd->last, nd->path.dentry, nd->flags); 19291da177e4SLinus Torvalds } 19301da177e4SLinus Torvalds 1931eead1911SChristoph Hellwig /** 1932a6b91919SRandy Dunlap * lookup_one_len - filesystem helper to lookup single pathname component 1933eead1911SChristoph Hellwig * @name: pathname component to lookup 1934eead1911SChristoph Hellwig * @base: base directory to lookup from 1935eead1911SChristoph Hellwig * @len: maximum length @len should be interpreted to 1936eead1911SChristoph Hellwig * 1937a6b91919SRandy Dunlap * Note that this routine is purely a helper for filesystem usage and should 1938a6b91919SRandy Dunlap * not be called by generic code. Also note that by using this function the 1939eead1911SChristoph Hellwig * nameidata argument is passed to the filesystem methods and a filesystem 1940eead1911SChristoph Hellwig * using this helper needs to be prepared for that. 1941eead1911SChristoph Hellwig */ 1942057f6c01SJames Morris struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 1943057f6c01SJames Morris { 1944057f6c01SJames Morris struct qstr this; 19456a96ba54SAl Viro unsigned int c; 1946cda309deSMiklos Szeredi int err; 1947057f6c01SJames Morris 19482f9092e1SDavid Woodhouse WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex)); 19492f9092e1SDavid Woodhouse 19506a96ba54SAl Viro this.name = name; 19516a96ba54SAl Viro this.len = len; 19520145acc2SLinus Torvalds this.hash = full_name_hash(name, len); 19536a96ba54SAl Viro if (!len) 19546a96ba54SAl Viro return ERR_PTR(-EACCES); 19556a96ba54SAl Viro 19566a96ba54SAl Viro while (len--) { 19576a96ba54SAl Viro c = *(const unsigned char *)name++; 19586a96ba54SAl Viro if (c == '/' || c == '\0') 19596a96ba54SAl Viro return ERR_PTR(-EACCES); 19606a96ba54SAl Viro } 19615a202bcdSAl Viro /* 19625a202bcdSAl Viro * See if the low-level filesystem might want 19635a202bcdSAl Viro * to use its own hash.. 19645a202bcdSAl Viro */ 19655a202bcdSAl Viro if (base->d_flags & DCACHE_OP_HASH) { 19665a202bcdSAl Viro int err = base->d_op->d_hash(base, base->d_inode, &this); 19675a202bcdSAl Viro if (err < 0) 19685a202bcdSAl Viro return ERR_PTR(err); 19695a202bcdSAl Viro } 1970eead1911SChristoph Hellwig 1971cda309deSMiklos Szeredi err = inode_permission(base->d_inode, MAY_EXEC); 1972cda309deSMiklos Szeredi if (err) 1973cda309deSMiklos Szeredi return ERR_PTR(err); 1974cda309deSMiklos Szeredi 197572bd866aSAl Viro return __lookup_hash(&this, base, 0); 1976057f6c01SJames Morris } 1977057f6c01SJames Morris 19781fa1e7f6SAndy Whitcroft int user_path_at_empty(int dfd, const char __user *name, unsigned flags, 19791fa1e7f6SAndy Whitcroft struct path *path, int *empty) 19801da177e4SLinus Torvalds { 19812d8f3038SAl Viro struct nameidata nd; 19821fa1e7f6SAndy Whitcroft char *tmp = getname_flags(name, flags, empty); 19831da177e4SLinus Torvalds int err = PTR_ERR(tmp); 19841da177e4SLinus Torvalds if (!IS_ERR(tmp)) { 19852d8f3038SAl Viro 19862d8f3038SAl Viro BUG_ON(flags & LOOKUP_PARENT); 19872d8f3038SAl Viro 19882d8f3038SAl Viro err = do_path_lookup(dfd, tmp, flags, &nd); 19891da177e4SLinus Torvalds putname(tmp); 19902d8f3038SAl Viro if (!err) 19912d8f3038SAl Viro *path = nd.path; 19921da177e4SLinus Torvalds } 19931da177e4SLinus Torvalds return err; 19941da177e4SLinus Torvalds } 19951da177e4SLinus Torvalds 19961fa1e7f6SAndy Whitcroft int user_path_at(int dfd, const char __user *name, unsigned flags, 19971fa1e7f6SAndy Whitcroft struct path *path) 19981fa1e7f6SAndy Whitcroft { 1999f7493e5dSLinus Torvalds return user_path_at_empty(dfd, name, flags, path, NULL); 20001fa1e7f6SAndy Whitcroft } 20011fa1e7f6SAndy Whitcroft 20022ad94ae6SAl Viro static int user_path_parent(int dfd, const char __user *path, 20032ad94ae6SAl Viro struct nameidata *nd, char **name) 20042ad94ae6SAl Viro { 20052ad94ae6SAl Viro char *s = getname(path); 20062ad94ae6SAl Viro int error; 20072ad94ae6SAl Viro 20082ad94ae6SAl Viro if (IS_ERR(s)) 20092ad94ae6SAl Viro return PTR_ERR(s); 20102ad94ae6SAl Viro 20112ad94ae6SAl Viro error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd); 20122ad94ae6SAl Viro if (error) 20132ad94ae6SAl Viro putname(s); 20142ad94ae6SAl Viro else 20152ad94ae6SAl Viro *name = s; 20162ad94ae6SAl Viro 20172ad94ae6SAl Viro return error; 20182ad94ae6SAl Viro } 20192ad94ae6SAl Viro 20201da177e4SLinus Torvalds /* 20211da177e4SLinus Torvalds * It's inline, so penalty for filesystems that don't use sticky bit is 20221da177e4SLinus Torvalds * minimal. 20231da177e4SLinus Torvalds */ 20241da177e4SLinus Torvalds static inline int check_sticky(struct inode *dir, struct inode *inode) 20251da177e4SLinus Torvalds { 20268e96e3b7SEric W. Biederman kuid_t fsuid = current_fsuid(); 2027da9592edSDavid Howells 20281da177e4SLinus Torvalds if (!(dir->i_mode & S_ISVTX)) 20291da177e4SLinus Torvalds return 0; 20308e96e3b7SEric W. Biederman if (uid_eq(inode->i_uid, fsuid)) 20311da177e4SLinus Torvalds return 0; 20328e96e3b7SEric W. Biederman if (uid_eq(dir->i_uid, fsuid)) 20331da177e4SLinus Torvalds return 0; 20341a48e2acSEric W. Biederman return !inode_capable(inode, CAP_FOWNER); 20351da177e4SLinus Torvalds } 20361da177e4SLinus Torvalds 20371da177e4SLinus Torvalds /* 20381da177e4SLinus Torvalds * Check whether we can remove a link victim from directory dir, check 20391da177e4SLinus Torvalds * whether the type of victim is right. 20401da177e4SLinus Torvalds * 1. We can't do it if dir is read-only (done in permission()) 20411da177e4SLinus Torvalds * 2. We should have write and exec permissions on dir 20421da177e4SLinus Torvalds * 3. We can't remove anything from append-only dir 20431da177e4SLinus Torvalds * 4. We can't do anything with immutable dir (done in permission()) 20441da177e4SLinus Torvalds * 5. If the sticky bit on dir is set we should either 20451da177e4SLinus Torvalds * a. be owner of dir, or 20461da177e4SLinus Torvalds * b. be owner of victim, or 20471da177e4SLinus Torvalds * c. have CAP_FOWNER capability 20481da177e4SLinus Torvalds * 6. If the victim is append-only or immutable we can't do antyhing with 20491da177e4SLinus Torvalds * links pointing to it. 20501da177e4SLinus Torvalds * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. 20511da177e4SLinus Torvalds * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. 20521da177e4SLinus Torvalds * 9. We can't remove a root or mountpoint. 20531da177e4SLinus Torvalds * 10. We don't allow removal of NFS sillyrenamed files; it's handled by 20541da177e4SLinus Torvalds * nfs_async_unlink(). 20551da177e4SLinus Torvalds */ 2056858119e1SArjan van de Ven static int may_delete(struct inode *dir,struct dentry *victim,int isdir) 20571da177e4SLinus Torvalds { 20581da177e4SLinus Torvalds int error; 20591da177e4SLinus Torvalds 20601da177e4SLinus Torvalds if (!victim->d_inode) 20611da177e4SLinus Torvalds return -ENOENT; 20621da177e4SLinus Torvalds 20631da177e4SLinus Torvalds BUG_ON(victim->d_parent->d_inode != dir); 2064cccc6bbaSAl Viro audit_inode_child(victim, dir); 20651da177e4SLinus Torvalds 2066f419a2e3SAl Viro error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 20671da177e4SLinus Torvalds if (error) 20681da177e4SLinus Torvalds return error; 20691da177e4SLinus Torvalds if (IS_APPEND(dir)) 20701da177e4SLinus Torvalds return -EPERM; 20711da177e4SLinus Torvalds if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)|| 2072f9454548SHugh Dickins IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode)) 20731da177e4SLinus Torvalds return -EPERM; 20741da177e4SLinus Torvalds if (isdir) { 20751da177e4SLinus Torvalds if (!S_ISDIR(victim->d_inode->i_mode)) 20761da177e4SLinus Torvalds return -ENOTDIR; 20771da177e4SLinus Torvalds if (IS_ROOT(victim)) 20781da177e4SLinus Torvalds return -EBUSY; 20791da177e4SLinus Torvalds } else if (S_ISDIR(victim->d_inode->i_mode)) 20801da177e4SLinus Torvalds return -EISDIR; 20811da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 20821da177e4SLinus Torvalds return -ENOENT; 20831da177e4SLinus Torvalds if (victim->d_flags & DCACHE_NFSFS_RENAMED) 20841da177e4SLinus Torvalds return -EBUSY; 20851da177e4SLinus Torvalds return 0; 20861da177e4SLinus Torvalds } 20871da177e4SLinus Torvalds 20881da177e4SLinus Torvalds /* Check whether we can create an object with dentry child in directory 20891da177e4SLinus Torvalds * dir. 20901da177e4SLinus Torvalds * 1. We can't do it if child already exists (open has special treatment for 20911da177e4SLinus Torvalds * this case, but since we are inlined it's OK) 20921da177e4SLinus Torvalds * 2. We can't do it if dir is read-only (done in permission()) 20931da177e4SLinus Torvalds * 3. We should have write and exec permissions on dir 20941da177e4SLinus Torvalds * 4. We can't do it if dir is immutable (done in permission()) 20951da177e4SLinus Torvalds */ 2096a95164d9SMiklos Szeredi static inline int may_create(struct inode *dir, struct dentry *child) 20971da177e4SLinus Torvalds { 20981da177e4SLinus Torvalds if (child->d_inode) 20991da177e4SLinus Torvalds return -EEXIST; 21001da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 21011da177e4SLinus Torvalds return -ENOENT; 2102f419a2e3SAl Viro return inode_permission(dir, MAY_WRITE | MAY_EXEC); 21031da177e4SLinus Torvalds } 21041da177e4SLinus Torvalds 21051da177e4SLinus Torvalds /* 21061da177e4SLinus Torvalds * p1 and p2 should be directories on the same fs. 21071da177e4SLinus Torvalds */ 21081da177e4SLinus Torvalds struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 21091da177e4SLinus Torvalds { 21101da177e4SLinus Torvalds struct dentry *p; 21111da177e4SLinus Torvalds 21121da177e4SLinus Torvalds if (p1 == p2) { 2113f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 21141da177e4SLinus Torvalds return NULL; 21151da177e4SLinus Torvalds } 21161da177e4SLinus Torvalds 2117a11f3a05SArjan van de Ven mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 21181da177e4SLinus Torvalds 2119e2761a11SOGAWA Hirofumi p = d_ancestor(p2, p1); 2120e2761a11SOGAWA Hirofumi if (p) { 2121f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT); 2122f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD); 21231da177e4SLinus Torvalds return p; 21241da177e4SLinus Torvalds } 21251da177e4SLinus Torvalds 2126e2761a11SOGAWA Hirofumi p = d_ancestor(p1, p2); 2127e2761a11SOGAWA Hirofumi if (p) { 2128f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 2129f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); 21301da177e4SLinus Torvalds return p; 21311da177e4SLinus Torvalds } 21321da177e4SLinus Torvalds 2133f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 2134f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); 21351da177e4SLinus Torvalds return NULL; 21361da177e4SLinus Torvalds } 21371da177e4SLinus Torvalds 21381da177e4SLinus Torvalds void unlock_rename(struct dentry *p1, struct dentry *p2) 21391da177e4SLinus Torvalds { 21401b1dcc1bSJes Sorensen mutex_unlock(&p1->d_inode->i_mutex); 21411da177e4SLinus Torvalds if (p1 != p2) { 21421b1dcc1bSJes Sorensen mutex_unlock(&p2->d_inode->i_mutex); 2143a11f3a05SArjan van de Ven mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 21441da177e4SLinus Torvalds } 21451da177e4SLinus Torvalds } 21461da177e4SLinus Torvalds 21474acdaf27SAl Viro int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, 2148312b63fbSAl Viro bool want_excl) 21491da177e4SLinus Torvalds { 2150a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 21511da177e4SLinus Torvalds if (error) 21521da177e4SLinus Torvalds return error; 21531da177e4SLinus Torvalds 2154acfa4380SAl Viro if (!dir->i_op->create) 21551da177e4SLinus Torvalds return -EACCES; /* shouldn't it be ENOSYS? */ 21561da177e4SLinus Torvalds mode &= S_IALLUGO; 21571da177e4SLinus Torvalds mode |= S_IFREG; 21581da177e4SLinus Torvalds error = security_inode_create(dir, dentry, mode); 21591da177e4SLinus Torvalds if (error) 21601da177e4SLinus Torvalds return error; 2161312b63fbSAl Viro error = dir->i_op->create(dir, dentry, mode, want_excl); 2162a74574aaSStephen Smalley if (!error) 2163f38aa942SAmy Griffis fsnotify_create(dir, dentry); 21641da177e4SLinus Torvalds return error; 21651da177e4SLinus Torvalds } 21661da177e4SLinus Torvalds 216773d049a4SAl Viro static int may_open(struct path *path, int acc_mode, int flag) 21681da177e4SLinus Torvalds { 21693fb64190SChristoph Hellwig struct dentry *dentry = path->dentry; 21701da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 21711da177e4SLinus Torvalds int error; 21721da177e4SLinus Torvalds 2173bcda7652SAl Viro /* O_PATH? */ 2174bcda7652SAl Viro if (!acc_mode) 2175bcda7652SAl Viro return 0; 2176bcda7652SAl Viro 21771da177e4SLinus Torvalds if (!inode) 21781da177e4SLinus Torvalds return -ENOENT; 21791da177e4SLinus Torvalds 2180c8fe8f30SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 2181c8fe8f30SChristoph Hellwig case S_IFLNK: 21821da177e4SLinus Torvalds return -ELOOP; 2183c8fe8f30SChristoph Hellwig case S_IFDIR: 2184c8fe8f30SChristoph Hellwig if (acc_mode & MAY_WRITE) 21851da177e4SLinus Torvalds return -EISDIR; 2186c8fe8f30SChristoph Hellwig break; 2187c8fe8f30SChristoph Hellwig case S_IFBLK: 2188c8fe8f30SChristoph Hellwig case S_IFCHR: 21893fb64190SChristoph Hellwig if (path->mnt->mnt_flags & MNT_NODEV) 21901da177e4SLinus Torvalds return -EACCES; 2191c8fe8f30SChristoph Hellwig /*FALLTHRU*/ 2192c8fe8f30SChristoph Hellwig case S_IFIFO: 2193c8fe8f30SChristoph Hellwig case S_IFSOCK: 21941da177e4SLinus Torvalds flag &= ~O_TRUNC; 2195c8fe8f30SChristoph Hellwig break; 21964a3fd211SDave Hansen } 2197b41572e9SDave Hansen 21983fb64190SChristoph Hellwig error = inode_permission(inode, acc_mode); 2199b41572e9SDave Hansen if (error) 2200b41572e9SDave Hansen return error; 22016146f0d5SMimi Zohar 22021da177e4SLinus Torvalds /* 22031da177e4SLinus Torvalds * An append-only file must be opened in append mode for writing. 22041da177e4SLinus Torvalds */ 22051da177e4SLinus Torvalds if (IS_APPEND(inode)) { 22068737c930SAl Viro if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) 22077715b521SAl Viro return -EPERM; 22081da177e4SLinus Torvalds if (flag & O_TRUNC) 22097715b521SAl Viro return -EPERM; 22101da177e4SLinus Torvalds } 22111da177e4SLinus Torvalds 22121da177e4SLinus Torvalds /* O_NOATIME can only be set by the owner or superuser */ 22132e149670SSerge E. Hallyn if (flag & O_NOATIME && !inode_owner_or_capable(inode)) 22147715b521SAl Viro return -EPERM; 22151da177e4SLinus Torvalds 2216f3c7691eSJ. Bruce Fields return 0; 22177715b521SAl Viro } 22187715b521SAl Viro 2219e1181ee6SJeff Layton static int handle_truncate(struct file *filp) 22207715b521SAl Viro { 2221e1181ee6SJeff Layton struct path *path = &filp->f_path; 22227715b521SAl Viro struct inode *inode = path->dentry->d_inode; 22237715b521SAl Viro int error = get_write_access(inode); 22241da177e4SLinus Torvalds if (error) 22257715b521SAl Viro return error; 22261da177e4SLinus Torvalds /* 22271da177e4SLinus Torvalds * Refuse to truncate files with mandatory locks held on them. 22281da177e4SLinus Torvalds */ 22291da177e4SLinus Torvalds error = locks_verify_locked(inode); 2230be6d3e56SKentaro Takeda if (!error) 2231ea0d3ab2STetsuo Handa error = security_path_truncate(path); 22321da177e4SLinus Torvalds if (!error) { 22337715b521SAl Viro error = do_truncate(path->dentry, 0, 2234d139d7ffSMiklos Szeredi ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, 2235e1181ee6SJeff Layton filp); 22361da177e4SLinus Torvalds } 22371da177e4SLinus Torvalds put_write_access(inode); 2238acd0c935SMimi Zohar return error; 22391da177e4SLinus Torvalds } 22401da177e4SLinus Torvalds 2241d57999e1SDave Hansen static inline int open_to_namei_flags(int flag) 2242d57999e1SDave Hansen { 22438a5e929dSAl Viro if ((flag & O_ACCMODE) == 3) 22448a5e929dSAl Viro flag--; 2245d57999e1SDave Hansen return flag; 2246d57999e1SDave Hansen } 2247d57999e1SDave Hansen 2248d18e9008SMiklos Szeredi static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode) 2249d18e9008SMiklos Szeredi { 2250d18e9008SMiklos Szeredi int error = security_path_mknod(dir, dentry, mode, 0); 2251d18e9008SMiklos Szeredi if (error) 2252d18e9008SMiklos Szeredi return error; 2253d18e9008SMiklos Szeredi 2254d18e9008SMiklos Szeredi error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC); 2255d18e9008SMiklos Szeredi if (error) 2256d18e9008SMiklos Szeredi return error; 2257d18e9008SMiklos Szeredi 2258d18e9008SMiklos Szeredi return security_inode_create(dir->dentry->d_inode, dentry, mode); 2259d18e9008SMiklos Szeredi } 2260d18e9008SMiklos Szeredi 22611acf0af9SDavid Howells /* 22621acf0af9SDavid Howells * Attempt to atomically look up, create and open a file from a negative 22631acf0af9SDavid Howells * dentry. 22641acf0af9SDavid Howells * 22651acf0af9SDavid Howells * Returns 0 if successful. The file will have been created and attached to 22661acf0af9SDavid Howells * @file by the filesystem calling finish_open(). 22671acf0af9SDavid Howells * 22681acf0af9SDavid Howells * Returns 1 if the file was looked up only or didn't need creating. The 22691acf0af9SDavid Howells * caller will need to perform the open themselves. @path will have been 22701acf0af9SDavid Howells * updated to point to the new dentry. This may be negative. 22711acf0af9SDavid Howells * 22721acf0af9SDavid Howells * Returns an error code otherwise. 22731acf0af9SDavid Howells */ 22742675a4ebSAl Viro static int atomic_open(struct nameidata *nd, struct dentry *dentry, 227530d90494SAl Viro struct path *path, struct file *file, 2276015c3bbcSMiklos Szeredi const struct open_flags *op, 227777d660a8SMiklos Szeredi bool *want_write, bool need_lookup, 227847237687SAl Viro int *opened) 2279d18e9008SMiklos Szeredi { 2280d18e9008SMiklos Szeredi struct inode *dir = nd->path.dentry->d_inode; 2281d18e9008SMiklos Szeredi unsigned open_flag = open_to_namei_flags(op->open_flag); 2282d18e9008SMiklos Szeredi umode_t mode; 2283d18e9008SMiklos Szeredi int error; 2284d18e9008SMiklos Szeredi int acc_mode; 2285d18e9008SMiklos Szeredi int create_error = 0; 2286d18e9008SMiklos Szeredi struct dentry *const DENTRY_NOT_SET = (void *) -1UL; 2287d18e9008SMiklos Szeredi 2288d18e9008SMiklos Szeredi BUG_ON(dentry->d_inode); 2289d18e9008SMiklos Szeredi 2290d18e9008SMiklos Szeredi /* Don't create child dentry for a dead directory. */ 2291d18e9008SMiklos Szeredi if (unlikely(IS_DEADDIR(dir))) { 22922675a4ebSAl Viro error = -ENOENT; 2293d18e9008SMiklos Szeredi goto out; 2294d18e9008SMiklos Szeredi } 2295d18e9008SMiklos Szeredi 2296d18e9008SMiklos Szeredi mode = op->mode & S_IALLUGO; 2297d18e9008SMiklos Szeredi if ((open_flag & O_CREAT) && !IS_POSIXACL(dir)) 2298d18e9008SMiklos Szeredi mode &= ~current_umask(); 2299d18e9008SMiklos Szeredi 2300d18e9008SMiklos Szeredi if (open_flag & O_EXCL) { 2301d18e9008SMiklos Szeredi open_flag &= ~O_TRUNC; 230247237687SAl Viro *opened |= FILE_CREATED; 2303d18e9008SMiklos Szeredi } 2304d18e9008SMiklos Szeredi 2305d18e9008SMiklos Szeredi /* 2306d18e9008SMiklos Szeredi * Checking write permission is tricky, bacuse we don't know if we are 2307d18e9008SMiklos Szeredi * going to actually need it: O_CREAT opens should work as long as the 2308d18e9008SMiklos Szeredi * file exists. But checking existence breaks atomicity. The trick is 2309d18e9008SMiklos Szeredi * to check access and if not granted clear O_CREAT from the flags. 2310d18e9008SMiklos Szeredi * 2311d18e9008SMiklos Szeredi * Another problem is returing the "right" error value (e.g. for an 2312d18e9008SMiklos Szeredi * O_EXCL open we want to return EEXIST not EROFS). 2313d18e9008SMiklos Szeredi */ 2314d18e9008SMiklos Szeredi if ((open_flag & (O_CREAT | O_TRUNC)) || 2315d18e9008SMiklos Szeredi (open_flag & O_ACCMODE) != O_RDONLY) { 2316d18e9008SMiklos Szeredi error = mnt_want_write(nd->path.mnt); 2317d18e9008SMiklos Szeredi if (!error) { 231877d660a8SMiklos Szeredi *want_write = true; 2319d18e9008SMiklos Szeredi } else if (!(open_flag & O_CREAT)) { 2320d18e9008SMiklos Szeredi /* 2321d18e9008SMiklos Szeredi * No O_CREATE -> atomicity not a requirement -> fall 2322d18e9008SMiklos Szeredi * back to lookup + open 2323d18e9008SMiklos Szeredi */ 2324d18e9008SMiklos Szeredi goto no_open; 2325d18e9008SMiklos Szeredi } else if (open_flag & (O_EXCL | O_TRUNC)) { 2326d18e9008SMiklos Szeredi /* Fall back and fail with the right error */ 2327d18e9008SMiklos Szeredi create_error = error; 2328d18e9008SMiklos Szeredi goto no_open; 2329d18e9008SMiklos Szeredi } else { 2330d18e9008SMiklos Szeredi /* No side effects, safe to clear O_CREAT */ 2331d18e9008SMiklos Szeredi create_error = error; 2332d18e9008SMiklos Szeredi open_flag &= ~O_CREAT; 2333d18e9008SMiklos Szeredi } 2334d18e9008SMiklos Szeredi } 2335d18e9008SMiklos Szeredi 2336d18e9008SMiklos Szeredi if (open_flag & O_CREAT) { 2337d18e9008SMiklos Szeredi error = may_o_create(&nd->path, dentry, op->mode); 2338d18e9008SMiklos Szeredi if (error) { 2339d18e9008SMiklos Szeredi create_error = error; 2340d18e9008SMiklos Szeredi if (open_flag & O_EXCL) 2341d18e9008SMiklos Szeredi goto no_open; 2342d18e9008SMiklos Szeredi open_flag &= ~O_CREAT; 2343d18e9008SMiklos Szeredi } 2344d18e9008SMiklos Szeredi } 2345d18e9008SMiklos Szeredi 2346d18e9008SMiklos Szeredi if (nd->flags & LOOKUP_DIRECTORY) 2347d18e9008SMiklos Szeredi open_flag |= O_DIRECTORY; 2348d18e9008SMiklos Szeredi 234930d90494SAl Viro file->f_path.dentry = DENTRY_NOT_SET; 235030d90494SAl Viro file->f_path.mnt = nd->path.mnt; 235130d90494SAl Viro error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode, 235247237687SAl Viro opened); 2353d9585277SAl Viro if (error < 0) { 2354d9585277SAl Viro if (create_error && error == -ENOENT) 2355d9585277SAl Viro error = create_error; 2356d18e9008SMiklos Szeredi goto out; 2357d18e9008SMiklos Szeredi } 2358d18e9008SMiklos Szeredi 2359d18e9008SMiklos Szeredi acc_mode = op->acc_mode; 236047237687SAl Viro if (*opened & FILE_CREATED) { 2361d18e9008SMiklos Szeredi fsnotify_create(dir, dentry); 2362d18e9008SMiklos Szeredi acc_mode = MAY_OPEN; 2363d18e9008SMiklos Szeredi } 2364d18e9008SMiklos Szeredi 2365d9585277SAl Viro if (error) { /* returned 1, that is */ 236630d90494SAl Viro if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { 23672675a4ebSAl Viro error = -EIO; 2368d18e9008SMiklos Szeredi goto out; 2369d18e9008SMiklos Szeredi } 237030d90494SAl Viro if (file->f_path.dentry) { 2371d18e9008SMiklos Szeredi dput(dentry); 237230d90494SAl Viro dentry = file->f_path.dentry; 2373d18e9008SMiklos Szeredi } 2374d18e9008SMiklos Szeredi goto looked_up; 2375d18e9008SMiklos Szeredi } 2376d18e9008SMiklos Szeredi 2377d18e9008SMiklos Szeredi /* 2378d18e9008SMiklos Szeredi * We didn't have the inode before the open, so check open permission 2379d18e9008SMiklos Szeredi * here. 2380d18e9008SMiklos Szeredi */ 23812675a4ebSAl Viro error = may_open(&file->f_path, acc_mode, open_flag); 23822675a4ebSAl Viro if (error) 23832675a4ebSAl Viro fput(file); 2384d18e9008SMiklos Szeredi 2385d18e9008SMiklos Szeredi out: 2386d18e9008SMiklos Szeredi dput(dentry); 23872675a4ebSAl Viro return error; 2388d18e9008SMiklos Szeredi 2389d18e9008SMiklos Szeredi no_open: 2390d18e9008SMiklos Szeredi if (need_lookup) { 239172bd866aSAl Viro dentry = lookup_real(dir, dentry, nd->flags); 2392d18e9008SMiklos Szeredi if (IS_ERR(dentry)) 23932675a4ebSAl Viro return PTR_ERR(dentry); 2394d18e9008SMiklos Szeredi 2395d18e9008SMiklos Szeredi if (create_error) { 2396d18e9008SMiklos Szeredi int open_flag = op->open_flag; 2397d18e9008SMiklos Szeredi 23982675a4ebSAl Viro error = create_error; 2399d18e9008SMiklos Szeredi if ((open_flag & O_EXCL)) { 2400d18e9008SMiklos Szeredi if (!dentry->d_inode) 2401d18e9008SMiklos Szeredi goto out; 2402d18e9008SMiklos Szeredi } else if (!dentry->d_inode) { 2403d18e9008SMiklos Szeredi goto out; 2404d18e9008SMiklos Szeredi } else if ((open_flag & O_TRUNC) && 2405d18e9008SMiklos Szeredi S_ISREG(dentry->d_inode->i_mode)) { 2406d18e9008SMiklos Szeredi goto out; 2407d18e9008SMiklos Szeredi } 2408d18e9008SMiklos Szeredi /* will fail later, go on to get the right error */ 2409d18e9008SMiklos Szeredi } 2410d18e9008SMiklos Szeredi } 2411d18e9008SMiklos Szeredi looked_up: 2412d18e9008SMiklos Szeredi path->dentry = dentry; 2413d18e9008SMiklos Szeredi path->mnt = nd->path.mnt; 24142675a4ebSAl Viro return 1; 2415d18e9008SMiklos Szeredi } 2416d18e9008SMiklos Szeredi 241731e6b01fSNick Piggin /* 24181acf0af9SDavid Howells * Look up and maybe create and open the last component. 2419d58ffd35SMiklos Szeredi * 2420d58ffd35SMiklos Szeredi * Must be called with i_mutex held on parent. 2421d58ffd35SMiklos Szeredi * 24221acf0af9SDavid Howells * Returns 0 if the file was successfully atomically created (if necessary) and 24231acf0af9SDavid Howells * opened. In this case the file will be returned attached to @file. 24241acf0af9SDavid Howells * 24251acf0af9SDavid Howells * Returns 1 if the file was not completely opened at this time, though lookups 24261acf0af9SDavid Howells * and creations will have been performed and the dentry returned in @path will 24271acf0af9SDavid Howells * be positive upon return if O_CREAT was specified. If O_CREAT wasn't 24281acf0af9SDavid Howells * specified then a negative dentry may be returned. 24291acf0af9SDavid Howells * 24301acf0af9SDavid Howells * An error code is returned otherwise. 24311acf0af9SDavid Howells * 24321acf0af9SDavid Howells * FILE_CREATE will be set in @*opened if the dentry was created and will be 24331acf0af9SDavid Howells * cleared otherwise prior to returning. 2434d58ffd35SMiklos Szeredi */ 24352675a4ebSAl Viro static int lookup_open(struct nameidata *nd, struct path *path, 243630d90494SAl Viro struct file *file, 2437d58ffd35SMiklos Szeredi const struct open_flags *op, 243847237687SAl Viro bool *want_write, int *opened) 2439d58ffd35SMiklos Szeredi { 2440d58ffd35SMiklos Szeredi struct dentry *dir = nd->path.dentry; 244154ef4872SMiklos Szeredi struct inode *dir_inode = dir->d_inode; 2442d58ffd35SMiklos Szeredi struct dentry *dentry; 2443d58ffd35SMiklos Szeredi int error; 244454ef4872SMiklos Szeredi bool need_lookup; 2445d58ffd35SMiklos Szeredi 244647237687SAl Viro *opened &= ~FILE_CREATED; 2447201f956eSAl Viro dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup); 2448d58ffd35SMiklos Szeredi if (IS_ERR(dentry)) 24492675a4ebSAl Viro return PTR_ERR(dentry); 2450d58ffd35SMiklos Szeredi 2451d18e9008SMiklos Szeredi /* Cached positive dentry: will open in f_op->open */ 2452d18e9008SMiklos Szeredi if (!need_lookup && dentry->d_inode) 2453d18e9008SMiklos Szeredi goto out_no_open; 2454d18e9008SMiklos Szeredi 2455d18e9008SMiklos Szeredi if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) { 245630d90494SAl Viro return atomic_open(nd, dentry, path, file, op, want_write, 245747237687SAl Viro need_lookup, opened); 2458d18e9008SMiklos Szeredi } 2459d18e9008SMiklos Szeredi 246054ef4872SMiklos Szeredi if (need_lookup) { 246154ef4872SMiklos Szeredi BUG_ON(dentry->d_inode); 246254ef4872SMiklos Szeredi 246372bd866aSAl Viro dentry = lookup_real(dir_inode, dentry, nd->flags); 246454ef4872SMiklos Szeredi if (IS_ERR(dentry)) 24652675a4ebSAl Viro return PTR_ERR(dentry); 246654ef4872SMiklos Szeredi } 246754ef4872SMiklos Szeredi 2468d58ffd35SMiklos Szeredi /* Negative dentry, just create the file */ 2469d58ffd35SMiklos Szeredi if (!dentry->d_inode && (op->open_flag & O_CREAT)) { 2470d58ffd35SMiklos Szeredi umode_t mode = op->mode; 2471d58ffd35SMiklos Szeredi if (!IS_POSIXACL(dir->d_inode)) 2472d58ffd35SMiklos Szeredi mode &= ~current_umask(); 2473d58ffd35SMiklos Szeredi /* 2474d58ffd35SMiklos Szeredi * This write is needed to ensure that a 2475d58ffd35SMiklos Szeredi * rw->ro transition does not occur between 2476d58ffd35SMiklos Szeredi * the time when the file is created and when 2477d58ffd35SMiklos Szeredi * a permanent write count is taken through 2478015c3bbcSMiklos Szeredi * the 'struct file' in finish_open(). 2479d58ffd35SMiklos Szeredi */ 2480d58ffd35SMiklos Szeredi error = mnt_want_write(nd->path.mnt); 2481d58ffd35SMiklos Szeredi if (error) 2482d58ffd35SMiklos Szeredi goto out_dput; 248377d660a8SMiklos Szeredi *want_write = true; 248447237687SAl Viro *opened |= FILE_CREATED; 2485d58ffd35SMiklos Szeredi error = security_path_mknod(&nd->path, dentry, mode, 0); 2486d58ffd35SMiklos Szeredi if (error) 2487d58ffd35SMiklos Szeredi goto out_dput; 2488312b63fbSAl Viro error = vfs_create(dir->d_inode, dentry, mode, 2489312b63fbSAl Viro nd->flags & LOOKUP_EXCL); 2490d58ffd35SMiklos Szeredi if (error) 2491d58ffd35SMiklos Szeredi goto out_dput; 2492d58ffd35SMiklos Szeredi } 2493d18e9008SMiklos Szeredi out_no_open: 2494d58ffd35SMiklos Szeredi path->dentry = dentry; 2495d58ffd35SMiklos Szeredi path->mnt = nd->path.mnt; 24962675a4ebSAl Viro return 1; 2497d58ffd35SMiklos Szeredi 2498d58ffd35SMiklos Szeredi out_dput: 2499d58ffd35SMiklos Szeredi dput(dentry); 25002675a4ebSAl Viro return error; 2501d58ffd35SMiklos Szeredi } 2502d58ffd35SMiklos Szeredi 2503d58ffd35SMiklos Szeredi /* 2504fe2d35ffSAl Viro * Handle the last step of open() 250531e6b01fSNick Piggin */ 25062675a4ebSAl Viro static int do_last(struct nameidata *nd, struct path *path, 250730d90494SAl Viro struct file *file, const struct open_flags *op, 250847237687SAl Viro int *opened, const char *pathname) 2509fb1cc555SAl Viro { 2510a1e28038SAl Viro struct dentry *dir = nd->path.dentry; 2511ca344a89SAl Viro int open_flag = op->open_flag; 251277d660a8SMiklos Szeredi bool will_truncate = (open_flag & O_TRUNC) != 0; 251377d660a8SMiklos Szeredi bool want_write = false; 2514bcda7652SAl Viro int acc_mode = op->acc_mode; 2515a1eb3315SMiklos Szeredi struct inode *inode; 251677d660a8SMiklos Szeredi bool symlink_ok = false; 251716b1c1cdSMiklos Szeredi struct path save_parent = { .dentry = NULL, .mnt = NULL }; 251816b1c1cdSMiklos Szeredi bool retried = false; 251916c2cd71SAl Viro int error; 2520fb1cc555SAl Viro 2521c3e380b0SAl Viro nd->flags &= ~LOOKUP_PARENT; 2522c3e380b0SAl Viro nd->flags |= op->intent; 2523c3e380b0SAl Viro 25241f36f774SAl Viro switch (nd->last_type) { 25251f36f774SAl Viro case LAST_DOTDOT: 2526176306f5SNeil Brown case LAST_DOT: 2527fe2d35ffSAl Viro error = handle_dots(nd, nd->last_type); 2528fe2d35ffSAl Viro if (error) 25292675a4ebSAl Viro return error; 25301f36f774SAl Viro /* fallthrough */ 25311f36f774SAl Viro case LAST_ROOT: 25329f1fafeeSAl Viro error = complete_walk(nd); 253316c2cd71SAl Viro if (error) 25342675a4ebSAl Viro return error; 2535fe2d35ffSAl Viro audit_inode(pathname, nd->path.dentry); 2536ca344a89SAl Viro if (open_flag & O_CREAT) { 253716c2cd71SAl Viro error = -EISDIR; 25382675a4ebSAl Viro goto out; 2539fe2d35ffSAl Viro } 2540e83db167SMiklos Szeredi goto finish_open; 25411f36f774SAl Viro case LAST_BIND: 25429f1fafeeSAl Viro error = complete_walk(nd); 254316c2cd71SAl Viro if (error) 25442675a4ebSAl Viro return error; 25451f36f774SAl Viro audit_inode(pathname, dir); 2546e83db167SMiklos Szeredi goto finish_open; 25471f36f774SAl Viro } 2548a2c36b45SAl Viro 2549ca344a89SAl Viro if (!(open_flag & O_CREAT)) { 2550fe2d35ffSAl Viro if (nd->last.name[nd->last.len]) 2551fe2d35ffSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 2552bcda7652SAl Viro if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW)) 255377d660a8SMiklos Szeredi symlink_ok = true; 2554fe2d35ffSAl Viro /* we _can_ be in RCU mode here */ 2555a1eb3315SMiklos Szeredi error = lookup_fast(nd, &nd->last, path, &inode); 255671574865SMiklos Szeredi if (likely(!error)) 255771574865SMiklos Szeredi goto finish_lookup; 255871574865SMiklos Szeredi 2559ce57dfc1SAl Viro if (error < 0) 25602675a4ebSAl Viro goto out; 2561a1eb3315SMiklos Szeredi 256237d7fffcSMiklos Szeredi BUG_ON(nd->inode != dir->d_inode); 2563b6183df7SMiklos Szeredi } else { 2564fe2d35ffSAl Viro /* create side of things */ 2565a3fbbde7SAl Viro /* 2566b6183df7SMiklos Szeredi * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED 2567b6183df7SMiklos Szeredi * has been cleared when we got to the last component we are 2568b6183df7SMiklos Szeredi * about to look up 2569a3fbbde7SAl Viro */ 25709f1fafeeSAl Viro error = complete_walk(nd); 25719f1fafeeSAl Viro if (error) 25722675a4ebSAl Viro return error; 2573fe2d35ffSAl Viro 2574fe2d35ffSAl Viro audit_inode(pathname, dir); 257516c2cd71SAl Viro error = -EISDIR; 25761f36f774SAl Viro /* trailing slashes? */ 257731e6b01fSNick Piggin if (nd->last.name[nd->last.len]) 25782675a4ebSAl Viro goto out; 2579b6183df7SMiklos Szeredi } 25801f36f774SAl Viro 258116b1c1cdSMiklos Szeredi retry_lookup: 2582a1e28038SAl Viro mutex_lock(&dir->d_inode->i_mutex); 25832675a4ebSAl Viro error = lookup_open(nd, path, file, op, &want_write, opened); 2584fb1cc555SAl Viro mutex_unlock(&dir->d_inode->i_mutex); 2585fb1cc555SAl Viro 25862675a4ebSAl Viro if (error <= 0) { 25872675a4ebSAl Viro if (error) 2588d58ffd35SMiklos Szeredi goto out; 25896c0d46c4SAl Viro 259047237687SAl Viro if ((*opened & FILE_CREATED) || 25912675a4ebSAl Viro !S_ISREG(file->f_path.dentry->d_inode->i_mode)) 259277d660a8SMiklos Szeredi will_truncate = false; 2593d18e9008SMiklos Szeredi 25942675a4ebSAl Viro audit_inode(pathname, file->f_path.dentry); 2595d18e9008SMiklos Szeredi goto opened; 2596d18e9008SMiklos Szeredi } 2597d18e9008SMiklos Szeredi 259847237687SAl Viro if (*opened & FILE_CREATED) { 25999b44f1b3SAl Viro /* Don't check for write permission, don't truncate */ 2600ca344a89SAl Viro open_flag &= ~O_TRUNC; 260177d660a8SMiklos Szeredi will_truncate = false; 2602bcda7652SAl Viro acc_mode = MAY_OPEN; 2603d58ffd35SMiklos Szeredi path_to_nameidata(path, nd); 2604e83db167SMiklos Szeredi goto finish_open_created; 2605fb1cc555SAl Viro } 2606fb1cc555SAl Viro 2607fb1cc555SAl Viro /* 2608fb1cc555SAl Viro * It already exists. 2609fb1cc555SAl Viro */ 2610fb1cc555SAl Viro audit_inode(pathname, path->dentry); 2611fb1cc555SAl Viro 2612d18e9008SMiklos Szeredi /* 2613d18e9008SMiklos Szeredi * If atomic_open() acquired write access it is dropped now due to 2614d18e9008SMiklos Szeredi * possible mount and symlink following (this might be optimized away if 2615d18e9008SMiklos Szeredi * necessary...) 2616d18e9008SMiklos Szeredi */ 2617d18e9008SMiklos Szeredi if (want_write) { 2618d18e9008SMiklos Szeredi mnt_drop_write(nd->path.mnt); 261977d660a8SMiklos Szeredi want_write = false; 2620d18e9008SMiklos Szeredi } 2621d18e9008SMiklos Szeredi 2622fb1cc555SAl Viro error = -EEXIST; 2623ca344a89SAl Viro if (open_flag & O_EXCL) 2624fb1cc555SAl Viro goto exit_dput; 2625fb1cc555SAl Viro 26269875cf80SDavid Howells error = follow_managed(path, nd->flags); 26279875cf80SDavid Howells if (error < 0) 2628fb1cc555SAl Viro goto exit_dput; 2629fb1cc555SAl Viro 2630a3fbbde7SAl Viro if (error) 2631a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 2632a3fbbde7SAl Viro 2633decf3400SMiklos Szeredi BUG_ON(nd->flags & LOOKUP_RCU); 2634decf3400SMiklos Szeredi inode = path->dentry->d_inode; 26355f5daac1SMiklos Szeredi finish_lookup: 26365f5daac1SMiklos Szeredi /* we _can_ be in RCU mode here */ 2637fb1cc555SAl Viro error = -ENOENT; 263854c33e7fSMiklos Szeredi if (!inode) { 263954c33e7fSMiklos Szeredi path_to_nameidata(path, nd); 26402675a4ebSAl Viro goto out; 264154c33e7fSMiklos Szeredi } 26429e67f361SAl Viro 2643d45ea867SMiklos Szeredi if (should_follow_link(inode, !symlink_ok)) { 2644d45ea867SMiklos Szeredi if (nd->flags & LOOKUP_RCU) { 2645d45ea867SMiklos Szeredi if (unlikely(unlazy_walk(nd, path->dentry))) { 2646d45ea867SMiklos Szeredi error = -ECHILD; 26472675a4ebSAl Viro goto out; 2648d45ea867SMiklos Szeredi } 2649d45ea867SMiklos Szeredi } 2650d45ea867SMiklos Szeredi BUG_ON(inode != path->dentry->d_inode); 26512675a4ebSAl Viro return 1; 2652d45ea867SMiklos Szeredi } 2653fb1cc555SAl Viro 265416b1c1cdSMiklos Szeredi if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) { 2655fb1cc555SAl Viro path_to_nameidata(path, nd); 265616b1c1cdSMiklos Szeredi } else { 265716b1c1cdSMiklos Szeredi save_parent.dentry = nd->path.dentry; 265816b1c1cdSMiklos Szeredi save_parent.mnt = mntget(path->mnt); 265916b1c1cdSMiklos Szeredi nd->path.dentry = path->dentry; 266016b1c1cdSMiklos Szeredi 266116b1c1cdSMiklos Szeredi } 2662decf3400SMiklos Szeredi nd->inode = inode; 2663a3fbbde7SAl Viro /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ 2664a3fbbde7SAl Viro error = complete_walk(nd); 266516b1c1cdSMiklos Szeredi if (error) { 266616b1c1cdSMiklos Szeredi path_put(&save_parent); 26672675a4ebSAl Viro return error; 266816b1c1cdSMiklos Szeredi } 2669fb1cc555SAl Viro error = -EISDIR; 2670050ac841SMiklos Szeredi if ((open_flag & O_CREAT) && S_ISDIR(nd->inode->i_mode)) 26712675a4ebSAl Viro goto out; 2672af2f5542SMiklos Szeredi error = -ENOTDIR; 2673af2f5542SMiklos Szeredi if ((nd->flags & LOOKUP_DIRECTORY) && !nd->inode->i_op->lookup) 26742675a4ebSAl Viro goto out; 2675d7fdd7f6SMiklos Szeredi audit_inode(pathname, nd->path.dentry); 2676e83db167SMiklos Szeredi finish_open: 26776c0d46c4SAl Viro if (!S_ISREG(nd->inode->i_mode)) 267877d660a8SMiklos Szeredi will_truncate = false; 26796c0d46c4SAl Viro 26800f9d1a10SAl Viro if (will_truncate) { 26810f9d1a10SAl Viro error = mnt_want_write(nd->path.mnt); 26820f9d1a10SAl Viro if (error) 26832675a4ebSAl Viro goto out; 268477d660a8SMiklos Szeredi want_write = true; 26850f9d1a10SAl Viro } 2686e83db167SMiklos Szeredi finish_open_created: 2687bcda7652SAl Viro error = may_open(&nd->path, acc_mode, open_flag); 2688ca344a89SAl Viro if (error) 26892675a4ebSAl Viro goto out; 269030d90494SAl Viro file->f_path.mnt = nd->path.mnt; 269130d90494SAl Viro error = finish_open(file, nd->path.dentry, NULL, opened); 269230d90494SAl Viro if (error) { 269330d90494SAl Viro if (error == -EOPENSTALE) 2694f60dc3dbSMiklos Szeredi goto stale_open; 2695015c3bbcSMiklos Szeredi goto out; 2696f60dc3dbSMiklos Szeredi } 2697a8277b9bSMiklos Szeredi opened: 26982675a4ebSAl Viro error = open_check_o_direct(file); 2699015c3bbcSMiklos Szeredi if (error) 2700015c3bbcSMiklos Szeredi goto exit_fput; 27012675a4ebSAl Viro error = ima_file_check(file, op->acc_mode); 2702aa4caadbSMiklos Szeredi if (error) 2703aa4caadbSMiklos Szeredi goto exit_fput; 2704aa4caadbSMiklos Szeredi 27050f9d1a10SAl Viro if (will_truncate) { 27062675a4ebSAl Viro error = handle_truncate(file); 2707aa4caadbSMiklos Szeredi if (error) 2708aa4caadbSMiklos Szeredi goto exit_fput; 27090f9d1a10SAl Viro } 2710ca344a89SAl Viro out: 2711ca344a89SAl Viro if (want_write) 27120f9d1a10SAl Viro mnt_drop_write(nd->path.mnt); 271316b1c1cdSMiklos Szeredi path_put(&save_parent); 2714e276ae67SMiklos Szeredi terminate_walk(nd); 27152675a4ebSAl Viro return error; 2716fb1cc555SAl Viro 2717fb1cc555SAl Viro exit_dput: 2718fb1cc555SAl Viro path_put_conditional(path, nd); 2719ca344a89SAl Viro goto out; 2720015c3bbcSMiklos Szeredi exit_fput: 27212675a4ebSAl Viro fput(file); 27222675a4ebSAl Viro goto out; 2723015c3bbcSMiklos Szeredi 2724f60dc3dbSMiklos Szeredi stale_open: 2725f60dc3dbSMiklos Szeredi /* If no saved parent or already retried then can't retry */ 2726f60dc3dbSMiklos Szeredi if (!save_parent.dentry || retried) 2727f60dc3dbSMiklos Szeredi goto out; 2728f60dc3dbSMiklos Szeredi 2729f60dc3dbSMiklos Szeredi BUG_ON(save_parent.dentry != dir); 2730f60dc3dbSMiklos Szeredi path_put(&nd->path); 2731f60dc3dbSMiklos Szeredi nd->path = save_parent; 2732f60dc3dbSMiklos Szeredi nd->inode = dir->d_inode; 2733f60dc3dbSMiklos Szeredi save_parent.mnt = NULL; 2734f60dc3dbSMiklos Szeredi save_parent.dentry = NULL; 2735f60dc3dbSMiklos Szeredi if (want_write) { 2736f60dc3dbSMiklos Szeredi mnt_drop_write(nd->path.mnt); 2737f60dc3dbSMiklos Szeredi want_write = false; 2738f60dc3dbSMiklos Szeredi } 2739f60dc3dbSMiklos Szeredi retried = true; 2740f60dc3dbSMiklos Szeredi goto retry_lookup; 2741fb1cc555SAl Viro } 2742fb1cc555SAl Viro 274313aab428SAl Viro static struct file *path_openat(int dfd, const char *pathname, 274473d049a4SAl Viro struct nameidata *nd, const struct open_flags *op, int flags) 27451da177e4SLinus Torvalds { 2746fe2d35ffSAl Viro struct file *base = NULL; 274730d90494SAl Viro struct file *file; 27489850c056SAl Viro struct path path; 274947237687SAl Viro int opened = 0; 275013aab428SAl Viro int error; 275131e6b01fSNick Piggin 275230d90494SAl Viro file = get_empty_filp(); 275330d90494SAl Viro if (!file) 275431e6b01fSNick Piggin return ERR_PTR(-ENFILE); 275531e6b01fSNick Piggin 275630d90494SAl Viro file->f_flags = op->open_flag; 275731e6b01fSNick Piggin 275873d049a4SAl Viro error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base); 275931e6b01fSNick Piggin if (unlikely(error)) 27602675a4ebSAl Viro goto out; 276131e6b01fSNick Piggin 2762fe2d35ffSAl Viro current->total_link_count = 0; 276373d049a4SAl Viro error = link_path_walk(pathname, nd); 276431e6b01fSNick Piggin if (unlikely(error)) 27652675a4ebSAl Viro goto out; 27661da177e4SLinus Torvalds 27672675a4ebSAl Viro error = do_last(nd, &path, file, op, &opened, pathname); 27682675a4ebSAl Viro while (unlikely(error > 0)) { /* trailing symlink */ 27697b9337aaSNick Piggin struct path link = path; 2770def4af30SAl Viro void *cookie; 2771574197e0SAl Viro if (!(nd->flags & LOOKUP_FOLLOW)) { 277273d049a4SAl Viro path_put_conditional(&path, nd); 277373d049a4SAl Viro path_put(&nd->path); 27742675a4ebSAl Viro error = -ELOOP; 277540b39136SAl Viro break; 277640b39136SAl Viro } 277773d049a4SAl Viro nd->flags |= LOOKUP_PARENT; 277873d049a4SAl Viro nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); 2779574197e0SAl Viro error = follow_link(&link, nd, &cookie); 2780c3e380b0SAl Viro if (unlikely(error)) 27812675a4ebSAl Viro break; 27822675a4ebSAl Viro error = do_last(nd, &path, file, op, &opened, pathname); 2783574197e0SAl Viro put_link(nd, &link, cookie); 2784806b681cSAl Viro } 278510fa8e62SAl Viro out: 278673d049a4SAl Viro if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) 278773d049a4SAl Viro path_put(&nd->root); 2788fe2d35ffSAl Viro if (base) 2789fe2d35ffSAl Viro fput(base); 27902675a4ebSAl Viro if (!(opened & FILE_OPENED)) { 27912675a4ebSAl Viro BUG_ON(!error); 279230d90494SAl Viro put_filp(file); 2793015c3bbcSMiklos Szeredi } 27942675a4ebSAl Viro if (unlikely(error)) { 27952675a4ebSAl Viro if (error == -EOPENSTALE) { 27962675a4ebSAl Viro if (flags & LOOKUP_RCU) 27972675a4ebSAl Viro error = -ECHILD; 27982675a4ebSAl Viro else 27992675a4ebSAl Viro error = -ESTALE; 28002675a4ebSAl Viro } 28012675a4ebSAl Viro file = ERR_PTR(error); 28022675a4ebSAl Viro } 28032675a4ebSAl Viro return file; 2804de459215SKirill Korotaev } 28051da177e4SLinus Torvalds 280613aab428SAl Viro struct file *do_filp_open(int dfd, const char *pathname, 280713aab428SAl Viro const struct open_flags *op, int flags) 280813aab428SAl Viro { 280973d049a4SAl Viro struct nameidata nd; 281013aab428SAl Viro struct file *filp; 281113aab428SAl Viro 281273d049a4SAl Viro filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU); 281313aab428SAl Viro if (unlikely(filp == ERR_PTR(-ECHILD))) 281473d049a4SAl Viro filp = path_openat(dfd, pathname, &nd, op, flags); 281513aab428SAl Viro if (unlikely(filp == ERR_PTR(-ESTALE))) 281673d049a4SAl Viro filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL); 281713aab428SAl Viro return filp; 281813aab428SAl Viro } 281913aab428SAl Viro 282073d049a4SAl Viro struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, 282173d049a4SAl Viro const char *name, const struct open_flags *op, int flags) 282273d049a4SAl Viro { 282373d049a4SAl Viro struct nameidata nd; 282473d049a4SAl Viro struct file *file; 282573d049a4SAl Viro 282673d049a4SAl Viro nd.root.mnt = mnt; 282773d049a4SAl Viro nd.root.dentry = dentry; 282873d049a4SAl Viro 282973d049a4SAl Viro flags |= LOOKUP_ROOT; 283073d049a4SAl Viro 2831bcda7652SAl Viro if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN) 283273d049a4SAl Viro return ERR_PTR(-ELOOP); 283373d049a4SAl Viro 283473d049a4SAl Viro file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU); 283573d049a4SAl Viro if (unlikely(file == ERR_PTR(-ECHILD))) 283673d049a4SAl Viro file = path_openat(-1, name, &nd, op, flags); 283773d049a4SAl Viro if (unlikely(file == ERR_PTR(-ESTALE))) 283873d049a4SAl Viro file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL); 283973d049a4SAl Viro return file; 284073d049a4SAl Viro } 284173d049a4SAl Viro 2842ed75e95dSAl Viro struct dentry *kern_path_create(int dfd, const char *pathname, struct path *path, int is_dir) 28431da177e4SLinus Torvalds { 2844c663e5d8SChristoph Hellwig struct dentry *dentry = ERR_PTR(-EEXIST); 2845ed75e95dSAl Viro struct nameidata nd; 2846ed75e95dSAl Viro int error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd); 2847ed75e95dSAl Viro if (error) 2848ed75e95dSAl Viro return ERR_PTR(error); 28491da177e4SLinus Torvalds 2850c663e5d8SChristoph Hellwig /* 2851c663e5d8SChristoph Hellwig * Yucky last component or no last component at all? 2852c663e5d8SChristoph Hellwig * (foo/., foo/.., /////) 2853c663e5d8SChristoph Hellwig */ 2854ed75e95dSAl Viro if (nd.last_type != LAST_NORM) 2855ed75e95dSAl Viro goto out; 2856ed75e95dSAl Viro nd.flags &= ~LOOKUP_PARENT; 2857ed75e95dSAl Viro nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL; 2858c663e5d8SChristoph Hellwig 2859c663e5d8SChristoph Hellwig /* 2860c663e5d8SChristoph Hellwig * Do the final lookup. 2861c663e5d8SChristoph Hellwig */ 2862ed75e95dSAl Viro mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 2863ed75e95dSAl Viro dentry = lookup_hash(&nd); 28641da177e4SLinus Torvalds if (IS_ERR(dentry)) 28651da177e4SLinus Torvalds goto fail; 2866c663e5d8SChristoph Hellwig 2867e9baf6e5SAl Viro if (dentry->d_inode) 2868e9baf6e5SAl Viro goto eexist; 2869c663e5d8SChristoph Hellwig /* 2870c663e5d8SChristoph Hellwig * Special case - lookup gave negative, but... we had foo/bar/ 2871c663e5d8SChristoph Hellwig * From the vfs_mknod() POV we just have a negative dentry - 2872c663e5d8SChristoph Hellwig * all is fine. Let's be bastards - you had / on the end, you've 2873c663e5d8SChristoph Hellwig * been asking for (non-existent) directory. -ENOENT for you. 2874c663e5d8SChristoph Hellwig */ 2875ed75e95dSAl Viro if (unlikely(!is_dir && nd.last.name[nd.last.len])) { 28761da177e4SLinus Torvalds dput(dentry); 28771da177e4SLinus Torvalds dentry = ERR_PTR(-ENOENT); 2878ed75e95dSAl Viro goto fail; 2879e9baf6e5SAl Viro } 2880ed75e95dSAl Viro *path = nd.path; 2881e9baf6e5SAl Viro return dentry; 2882e9baf6e5SAl Viro eexist: 2883e9baf6e5SAl Viro dput(dentry); 2884e9baf6e5SAl Viro dentry = ERR_PTR(-EEXIST); 28851da177e4SLinus Torvalds fail: 2886dae6ad8fSAl Viro mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 2887ed75e95dSAl Viro out: 2888dae6ad8fSAl Viro path_put(&nd.path); 2889ed75e95dSAl Viro return dentry; 2890dae6ad8fSAl Viro } 2891dae6ad8fSAl Viro EXPORT_SYMBOL(kern_path_create); 2892dae6ad8fSAl Viro 2893dae6ad8fSAl Viro struct dentry *user_path_create(int dfd, const char __user *pathname, struct path *path, int is_dir) 2894dae6ad8fSAl Viro { 2895dae6ad8fSAl Viro char *tmp = getname(pathname); 2896dae6ad8fSAl Viro struct dentry *res; 2897dae6ad8fSAl Viro if (IS_ERR(tmp)) 2898dae6ad8fSAl Viro return ERR_CAST(tmp); 2899dae6ad8fSAl Viro res = kern_path_create(dfd, tmp, path, is_dir); 2900dae6ad8fSAl Viro putname(tmp); 2901dae6ad8fSAl Viro return res; 2902dae6ad8fSAl Viro } 2903dae6ad8fSAl Viro EXPORT_SYMBOL(user_path_create); 2904dae6ad8fSAl Viro 29051a67aafbSAl Viro int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) 29061da177e4SLinus Torvalds { 2907a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 29081da177e4SLinus Torvalds 29091da177e4SLinus Torvalds if (error) 29101da177e4SLinus Torvalds return error; 29111da177e4SLinus Torvalds 2912975d6b39SEric W. Biederman if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) 29131da177e4SLinus Torvalds return -EPERM; 29141da177e4SLinus Torvalds 2915acfa4380SAl Viro if (!dir->i_op->mknod) 29161da177e4SLinus Torvalds return -EPERM; 29171da177e4SLinus Torvalds 291808ce5f16SSerge E. Hallyn error = devcgroup_inode_mknod(mode, dev); 291908ce5f16SSerge E. Hallyn if (error) 292008ce5f16SSerge E. Hallyn return error; 292108ce5f16SSerge E. Hallyn 29221da177e4SLinus Torvalds error = security_inode_mknod(dir, dentry, mode, dev); 29231da177e4SLinus Torvalds if (error) 29241da177e4SLinus Torvalds return error; 29251da177e4SLinus Torvalds 29261da177e4SLinus Torvalds error = dir->i_op->mknod(dir, dentry, mode, dev); 2927a74574aaSStephen Smalley if (!error) 2928f38aa942SAmy Griffis fsnotify_create(dir, dentry); 29291da177e4SLinus Torvalds return error; 29301da177e4SLinus Torvalds } 29311da177e4SLinus Torvalds 2932f69aac00SAl Viro static int may_mknod(umode_t mode) 2933463c3197SDave Hansen { 2934463c3197SDave Hansen switch (mode & S_IFMT) { 2935463c3197SDave Hansen case S_IFREG: 2936463c3197SDave Hansen case S_IFCHR: 2937463c3197SDave Hansen case S_IFBLK: 2938463c3197SDave Hansen case S_IFIFO: 2939463c3197SDave Hansen case S_IFSOCK: 2940463c3197SDave Hansen case 0: /* zero mode translates to S_IFREG */ 2941463c3197SDave Hansen return 0; 2942463c3197SDave Hansen case S_IFDIR: 2943463c3197SDave Hansen return -EPERM; 2944463c3197SDave Hansen default: 2945463c3197SDave Hansen return -EINVAL; 2946463c3197SDave Hansen } 2947463c3197SDave Hansen } 2948463c3197SDave Hansen 29498208a22bSAl Viro SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, 29502e4d0924SHeiko Carstens unsigned, dev) 29511da177e4SLinus Torvalds { 29521da177e4SLinus Torvalds struct dentry *dentry; 2953dae6ad8fSAl Viro struct path path; 2954dae6ad8fSAl Viro int error; 29551da177e4SLinus Torvalds 29561da177e4SLinus Torvalds if (S_ISDIR(mode)) 29571da177e4SLinus Torvalds return -EPERM; 29581da177e4SLinus Torvalds 2959dae6ad8fSAl Viro dentry = user_path_create(dfd, filename, &path, 0); 2960dae6ad8fSAl Viro if (IS_ERR(dentry)) 2961dae6ad8fSAl Viro return PTR_ERR(dentry); 29622ad94ae6SAl Viro 2963dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 2964ce3b0f8dSAl Viro mode &= ~current_umask(); 2965463c3197SDave Hansen error = may_mknod(mode); 2966463c3197SDave Hansen if (error) 2967463c3197SDave Hansen goto out_dput; 2968dae6ad8fSAl Viro error = mnt_want_write(path.mnt); 2969463c3197SDave Hansen if (error) 2970463c3197SDave Hansen goto out_dput; 2971dae6ad8fSAl Viro error = security_path_mknod(&path, dentry, mode, dev); 2972be6d3e56SKentaro Takeda if (error) 2973be6d3e56SKentaro Takeda goto out_drop_write; 29741da177e4SLinus Torvalds switch (mode & S_IFMT) { 29751da177e4SLinus Torvalds case 0: case S_IFREG: 2976312b63fbSAl Viro error = vfs_create(path.dentry->d_inode,dentry,mode,true); 29771da177e4SLinus Torvalds break; 29781da177e4SLinus Torvalds case S_IFCHR: case S_IFBLK: 2979dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode, 29801da177e4SLinus Torvalds new_decode_dev(dev)); 29811da177e4SLinus Torvalds break; 29821da177e4SLinus Torvalds case S_IFIFO: case S_IFSOCK: 2983dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); 29841da177e4SLinus Torvalds break; 29851da177e4SLinus Torvalds } 2986be6d3e56SKentaro Takeda out_drop_write: 2987dae6ad8fSAl Viro mnt_drop_write(path.mnt); 2988463c3197SDave Hansen out_dput: 29891da177e4SLinus Torvalds dput(dentry); 2990dae6ad8fSAl Viro mutex_unlock(&path.dentry->d_inode->i_mutex); 2991dae6ad8fSAl Viro path_put(&path); 29921da177e4SLinus Torvalds 29931da177e4SLinus Torvalds return error; 29941da177e4SLinus Torvalds } 29951da177e4SLinus Torvalds 29968208a22bSAl Viro SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) 29975590ff0dSUlrich Drepper { 29985590ff0dSUlrich Drepper return sys_mknodat(AT_FDCWD, filename, mode, dev); 29995590ff0dSUlrich Drepper } 30005590ff0dSUlrich Drepper 300118bb1db3SAl Viro int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 30021da177e4SLinus Torvalds { 3003a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 30048de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 30051da177e4SLinus Torvalds 30061da177e4SLinus Torvalds if (error) 30071da177e4SLinus Torvalds return error; 30081da177e4SLinus Torvalds 3009acfa4380SAl Viro if (!dir->i_op->mkdir) 30101da177e4SLinus Torvalds return -EPERM; 30111da177e4SLinus Torvalds 30121da177e4SLinus Torvalds mode &= (S_IRWXUGO|S_ISVTX); 30131da177e4SLinus Torvalds error = security_inode_mkdir(dir, dentry, mode); 30141da177e4SLinus Torvalds if (error) 30151da177e4SLinus Torvalds return error; 30161da177e4SLinus Torvalds 30178de52778SAl Viro if (max_links && dir->i_nlink >= max_links) 30188de52778SAl Viro return -EMLINK; 30198de52778SAl Viro 30201da177e4SLinus Torvalds error = dir->i_op->mkdir(dir, dentry, mode); 3021a74574aaSStephen Smalley if (!error) 3022f38aa942SAmy Griffis fsnotify_mkdir(dir, dentry); 30231da177e4SLinus Torvalds return error; 30241da177e4SLinus Torvalds } 30251da177e4SLinus Torvalds 3026a218d0fdSAl Viro SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) 30271da177e4SLinus Torvalds { 30286902d925SDave Hansen struct dentry *dentry; 3029dae6ad8fSAl Viro struct path path; 3030dae6ad8fSAl Viro int error; 30311da177e4SLinus Torvalds 3032dae6ad8fSAl Viro dentry = user_path_create(dfd, pathname, &path, 1); 30336902d925SDave Hansen if (IS_ERR(dentry)) 3034dae6ad8fSAl Viro return PTR_ERR(dentry); 30356902d925SDave Hansen 3036dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3037ce3b0f8dSAl Viro mode &= ~current_umask(); 3038dae6ad8fSAl Viro error = mnt_want_write(path.mnt); 3039463c3197SDave Hansen if (error) 3040463c3197SDave Hansen goto out_dput; 3041dae6ad8fSAl Viro error = security_path_mkdir(&path, dentry, mode); 3042be6d3e56SKentaro Takeda if (error) 3043be6d3e56SKentaro Takeda goto out_drop_write; 3044dae6ad8fSAl Viro error = vfs_mkdir(path.dentry->d_inode, dentry, mode); 3045be6d3e56SKentaro Takeda out_drop_write: 3046dae6ad8fSAl Viro mnt_drop_write(path.mnt); 3047463c3197SDave Hansen out_dput: 30481da177e4SLinus Torvalds dput(dentry); 3049dae6ad8fSAl Viro mutex_unlock(&path.dentry->d_inode->i_mutex); 3050dae6ad8fSAl Viro path_put(&path); 30511da177e4SLinus Torvalds return error; 30521da177e4SLinus Torvalds } 30531da177e4SLinus Torvalds 3054a218d0fdSAl Viro SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) 30555590ff0dSUlrich Drepper { 30565590ff0dSUlrich Drepper return sys_mkdirat(AT_FDCWD, pathname, mode); 30575590ff0dSUlrich Drepper } 30585590ff0dSUlrich Drepper 30591da177e4SLinus Torvalds /* 3060a71905f0SSage Weil * The dentry_unhash() helper will try to drop the dentry early: we 3061c0d02594SJ. Bruce Fields * should have a usage count of 1 if we're the only user of this 3062a71905f0SSage Weil * dentry, and if that is true (possibly after pruning the dcache), 3063a71905f0SSage Weil * then we drop the dentry now. 30641da177e4SLinus Torvalds * 30651da177e4SLinus Torvalds * A low-level filesystem can, if it choses, legally 30661da177e4SLinus Torvalds * do a 30671da177e4SLinus Torvalds * 30681da177e4SLinus Torvalds * if (!d_unhashed(dentry)) 30691da177e4SLinus Torvalds * return -EBUSY; 30701da177e4SLinus Torvalds * 30711da177e4SLinus Torvalds * if it cannot handle the case of removing a directory 30721da177e4SLinus Torvalds * that is still in use by something else.. 30731da177e4SLinus Torvalds */ 30741da177e4SLinus Torvalds void dentry_unhash(struct dentry *dentry) 30751da177e4SLinus Torvalds { 30761da177e4SLinus Torvalds shrink_dcache_parent(dentry); 30771da177e4SLinus Torvalds spin_lock(&dentry->d_lock); 307864252c75SSage Weil if (dentry->d_count == 1) 30791da177e4SLinus Torvalds __d_drop(dentry); 30801da177e4SLinus Torvalds spin_unlock(&dentry->d_lock); 30811da177e4SLinus Torvalds } 30821da177e4SLinus Torvalds 30831da177e4SLinus Torvalds int vfs_rmdir(struct inode *dir, struct dentry *dentry) 30841da177e4SLinus Torvalds { 30851da177e4SLinus Torvalds int error = may_delete(dir, dentry, 1); 30861da177e4SLinus Torvalds 30871da177e4SLinus Torvalds if (error) 30881da177e4SLinus Torvalds return error; 30891da177e4SLinus Torvalds 3090acfa4380SAl Viro if (!dir->i_op->rmdir) 30911da177e4SLinus Torvalds return -EPERM; 30921da177e4SLinus Torvalds 30931d2ef590SAl Viro dget(dentry); 30941b1dcc1bSJes Sorensen mutex_lock(&dentry->d_inode->i_mutex); 3095912dbc15SSage Weil 30961da177e4SLinus Torvalds error = -EBUSY; 3097912dbc15SSage Weil if (d_mountpoint(dentry)) 3098912dbc15SSage Weil goto out; 3099912dbc15SSage Weil 31001da177e4SLinus Torvalds error = security_inode_rmdir(dir, dentry); 3101912dbc15SSage Weil if (error) 3102912dbc15SSage Weil goto out; 3103912dbc15SSage Weil 31043cebde24SSage Weil shrink_dcache_parent(dentry); 31051da177e4SLinus Torvalds error = dir->i_op->rmdir(dir, dentry); 3106912dbc15SSage Weil if (error) 3107912dbc15SSage Weil goto out; 3108912dbc15SSage Weil 31091da177e4SLinus Torvalds dentry->d_inode->i_flags |= S_DEAD; 3110d83c49f3SAl Viro dont_mount(dentry); 31111da177e4SLinus Torvalds 3112912dbc15SSage Weil out: 3113912dbc15SSage Weil mutex_unlock(&dentry->d_inode->i_mutex); 31141d2ef590SAl Viro dput(dentry); 3115912dbc15SSage Weil if (!error) 3116912dbc15SSage Weil d_delete(dentry); 31171da177e4SLinus Torvalds return error; 31181da177e4SLinus Torvalds } 31191da177e4SLinus Torvalds 31205590ff0dSUlrich Drepper static long do_rmdir(int dfd, const char __user *pathname) 31211da177e4SLinus Torvalds { 31221da177e4SLinus Torvalds int error = 0; 31231da177e4SLinus Torvalds char * name; 31241da177e4SLinus Torvalds struct dentry *dentry; 31251da177e4SLinus Torvalds struct nameidata nd; 31261da177e4SLinus Torvalds 31272ad94ae6SAl Viro error = user_path_parent(dfd, pathname, &nd, &name); 31281da177e4SLinus Torvalds if (error) 31292ad94ae6SAl Viro return error; 31301da177e4SLinus Torvalds 31311da177e4SLinus Torvalds switch(nd.last_type) { 31321da177e4SLinus Torvalds case LAST_DOTDOT: 31331da177e4SLinus Torvalds error = -ENOTEMPTY; 31341da177e4SLinus Torvalds goto exit1; 31351da177e4SLinus Torvalds case LAST_DOT: 31361da177e4SLinus Torvalds error = -EINVAL; 31371da177e4SLinus Torvalds goto exit1; 31381da177e4SLinus Torvalds case LAST_ROOT: 31391da177e4SLinus Torvalds error = -EBUSY; 31401da177e4SLinus Torvalds goto exit1; 31411da177e4SLinus Torvalds } 31420612d9fbSOGAWA Hirofumi 31430612d9fbSOGAWA Hirofumi nd.flags &= ~LOOKUP_PARENT; 31440612d9fbSOGAWA Hirofumi 31454ac91378SJan Blunck mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 314649705b77SChristoph Hellwig dentry = lookup_hash(&nd); 31471da177e4SLinus Torvalds error = PTR_ERR(dentry); 31486902d925SDave Hansen if (IS_ERR(dentry)) 31496902d925SDave Hansen goto exit2; 3150e6bc45d6STheodore Ts'o if (!dentry->d_inode) { 3151e6bc45d6STheodore Ts'o error = -ENOENT; 3152e6bc45d6STheodore Ts'o goto exit3; 3153e6bc45d6STheodore Ts'o } 31540622753bSDave Hansen error = mnt_want_write(nd.path.mnt); 31550622753bSDave Hansen if (error) 31560622753bSDave Hansen goto exit3; 3157be6d3e56SKentaro Takeda error = security_path_rmdir(&nd.path, dentry); 3158be6d3e56SKentaro Takeda if (error) 3159be6d3e56SKentaro Takeda goto exit4; 31604ac91378SJan Blunck error = vfs_rmdir(nd.path.dentry->d_inode, dentry); 3161be6d3e56SKentaro Takeda exit4: 31620622753bSDave Hansen mnt_drop_write(nd.path.mnt); 31630622753bSDave Hansen exit3: 31641da177e4SLinus Torvalds dput(dentry); 31656902d925SDave Hansen exit2: 31664ac91378SJan Blunck mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 31671da177e4SLinus Torvalds exit1: 31681d957f9bSJan Blunck path_put(&nd.path); 31691da177e4SLinus Torvalds putname(name); 31701da177e4SLinus Torvalds return error; 31711da177e4SLinus Torvalds } 31721da177e4SLinus Torvalds 31733cdad428SHeiko Carstens SYSCALL_DEFINE1(rmdir, const char __user *, pathname) 31745590ff0dSUlrich Drepper { 31755590ff0dSUlrich Drepper return do_rmdir(AT_FDCWD, pathname); 31765590ff0dSUlrich Drepper } 31775590ff0dSUlrich Drepper 31781da177e4SLinus Torvalds int vfs_unlink(struct inode *dir, struct dentry *dentry) 31791da177e4SLinus Torvalds { 31801da177e4SLinus Torvalds int error = may_delete(dir, dentry, 0); 31811da177e4SLinus Torvalds 31821da177e4SLinus Torvalds if (error) 31831da177e4SLinus Torvalds return error; 31841da177e4SLinus Torvalds 3185acfa4380SAl Viro if (!dir->i_op->unlink) 31861da177e4SLinus Torvalds return -EPERM; 31871da177e4SLinus Torvalds 31881b1dcc1bSJes Sorensen mutex_lock(&dentry->d_inode->i_mutex); 31891da177e4SLinus Torvalds if (d_mountpoint(dentry)) 31901da177e4SLinus Torvalds error = -EBUSY; 31911da177e4SLinus Torvalds else { 31921da177e4SLinus Torvalds error = security_inode_unlink(dir, dentry); 3193bec1052eSAl Viro if (!error) { 31941da177e4SLinus Torvalds error = dir->i_op->unlink(dir, dentry); 3195bec1052eSAl Viro if (!error) 3196d83c49f3SAl Viro dont_mount(dentry); 3197bec1052eSAl Viro } 31981da177e4SLinus Torvalds } 31991b1dcc1bSJes Sorensen mutex_unlock(&dentry->d_inode->i_mutex); 32001da177e4SLinus Torvalds 32011da177e4SLinus Torvalds /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 32021da177e4SLinus Torvalds if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { 3203ece95912SJan Kara fsnotify_link_count(dentry->d_inode); 32041da177e4SLinus Torvalds d_delete(dentry); 32051da177e4SLinus Torvalds } 32060eeca283SRobert Love 32071da177e4SLinus Torvalds return error; 32081da177e4SLinus Torvalds } 32091da177e4SLinus Torvalds 32101da177e4SLinus Torvalds /* 32111da177e4SLinus Torvalds * Make sure that the actual truncation of the file will occur outside its 32121b1dcc1bSJes Sorensen * directory's i_mutex. Truncate can take a long time if there is a lot of 32131da177e4SLinus Torvalds * writeout happening, and we don't want to prevent access to the directory 32141da177e4SLinus Torvalds * while waiting on the I/O. 32151da177e4SLinus Torvalds */ 32165590ff0dSUlrich Drepper static long do_unlinkat(int dfd, const char __user *pathname) 32171da177e4SLinus Torvalds { 32182ad94ae6SAl Viro int error; 32191da177e4SLinus Torvalds char *name; 32201da177e4SLinus Torvalds struct dentry *dentry; 32211da177e4SLinus Torvalds struct nameidata nd; 32221da177e4SLinus Torvalds struct inode *inode = NULL; 32231da177e4SLinus Torvalds 32242ad94ae6SAl Viro error = user_path_parent(dfd, pathname, &nd, &name); 32251da177e4SLinus Torvalds if (error) 32262ad94ae6SAl Viro return error; 32272ad94ae6SAl Viro 32281da177e4SLinus Torvalds error = -EISDIR; 32291da177e4SLinus Torvalds if (nd.last_type != LAST_NORM) 32301da177e4SLinus Torvalds goto exit1; 32310612d9fbSOGAWA Hirofumi 32320612d9fbSOGAWA Hirofumi nd.flags &= ~LOOKUP_PARENT; 32330612d9fbSOGAWA Hirofumi 32344ac91378SJan Blunck mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 323549705b77SChristoph Hellwig dentry = lookup_hash(&nd); 32361da177e4SLinus Torvalds error = PTR_ERR(dentry); 32371da177e4SLinus Torvalds if (!IS_ERR(dentry)) { 32381da177e4SLinus Torvalds /* Why not before? Because we want correct error value */ 323950338b88STörök Edwin if (nd.last.name[nd.last.len]) 324050338b88STörök Edwin goto slashes; 32411da177e4SLinus Torvalds inode = dentry->d_inode; 324250338b88STörök Edwin if (!inode) 3243e6bc45d6STheodore Ts'o goto slashes; 32447de9c6eeSAl Viro ihold(inode); 32450622753bSDave Hansen error = mnt_want_write(nd.path.mnt); 32460622753bSDave Hansen if (error) 32470622753bSDave Hansen goto exit2; 3248be6d3e56SKentaro Takeda error = security_path_unlink(&nd.path, dentry); 3249be6d3e56SKentaro Takeda if (error) 3250be6d3e56SKentaro Takeda goto exit3; 32514ac91378SJan Blunck error = vfs_unlink(nd.path.dentry->d_inode, dentry); 3252be6d3e56SKentaro Takeda exit3: 32530622753bSDave Hansen mnt_drop_write(nd.path.mnt); 32541da177e4SLinus Torvalds exit2: 32551da177e4SLinus Torvalds dput(dentry); 32561da177e4SLinus Torvalds } 32574ac91378SJan Blunck mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 32581da177e4SLinus Torvalds if (inode) 32591da177e4SLinus Torvalds iput(inode); /* truncate the inode here */ 32601da177e4SLinus Torvalds exit1: 32611d957f9bSJan Blunck path_put(&nd.path); 32621da177e4SLinus Torvalds putname(name); 32631da177e4SLinus Torvalds return error; 32641da177e4SLinus Torvalds 32651da177e4SLinus Torvalds slashes: 32661da177e4SLinus Torvalds error = !dentry->d_inode ? -ENOENT : 32671da177e4SLinus Torvalds S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR; 32681da177e4SLinus Torvalds goto exit2; 32691da177e4SLinus Torvalds } 32701da177e4SLinus Torvalds 32712e4d0924SHeiko Carstens SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) 32725590ff0dSUlrich Drepper { 32735590ff0dSUlrich Drepper if ((flag & ~AT_REMOVEDIR) != 0) 32745590ff0dSUlrich Drepper return -EINVAL; 32755590ff0dSUlrich Drepper 32765590ff0dSUlrich Drepper if (flag & AT_REMOVEDIR) 32775590ff0dSUlrich Drepper return do_rmdir(dfd, pathname); 32785590ff0dSUlrich Drepper 32795590ff0dSUlrich Drepper return do_unlinkat(dfd, pathname); 32805590ff0dSUlrich Drepper } 32815590ff0dSUlrich Drepper 32823480b257SHeiko Carstens SYSCALL_DEFINE1(unlink, const char __user *, pathname) 32835590ff0dSUlrich Drepper { 32845590ff0dSUlrich Drepper return do_unlinkat(AT_FDCWD, pathname); 32855590ff0dSUlrich Drepper } 32865590ff0dSUlrich Drepper 3287db2e747bSMiklos Szeredi int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) 32881da177e4SLinus Torvalds { 3289a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 32901da177e4SLinus Torvalds 32911da177e4SLinus Torvalds if (error) 32921da177e4SLinus Torvalds return error; 32931da177e4SLinus Torvalds 3294acfa4380SAl Viro if (!dir->i_op->symlink) 32951da177e4SLinus Torvalds return -EPERM; 32961da177e4SLinus Torvalds 32971da177e4SLinus Torvalds error = security_inode_symlink(dir, dentry, oldname); 32981da177e4SLinus Torvalds if (error) 32991da177e4SLinus Torvalds return error; 33001da177e4SLinus Torvalds 33011da177e4SLinus Torvalds error = dir->i_op->symlink(dir, dentry, oldname); 3302a74574aaSStephen Smalley if (!error) 3303f38aa942SAmy Griffis fsnotify_create(dir, dentry); 33041da177e4SLinus Torvalds return error; 33051da177e4SLinus Torvalds } 33061da177e4SLinus Torvalds 33072e4d0924SHeiko Carstens SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, 33082e4d0924SHeiko Carstens int, newdfd, const char __user *, newname) 33091da177e4SLinus Torvalds { 33102ad94ae6SAl Viro int error; 33111da177e4SLinus Torvalds char *from; 33126902d925SDave Hansen struct dentry *dentry; 3313dae6ad8fSAl Viro struct path path; 33141da177e4SLinus Torvalds 33151da177e4SLinus Torvalds from = getname(oldname); 33161da177e4SLinus Torvalds if (IS_ERR(from)) 33171da177e4SLinus Torvalds return PTR_ERR(from); 33182ad94ae6SAl Viro 3319dae6ad8fSAl Viro dentry = user_path_create(newdfd, newname, &path, 0); 33201da177e4SLinus Torvalds error = PTR_ERR(dentry); 33216902d925SDave Hansen if (IS_ERR(dentry)) 3322dae6ad8fSAl Viro goto out_putname; 33236902d925SDave Hansen 3324dae6ad8fSAl Viro error = mnt_want_write(path.mnt); 332575c3f29dSDave Hansen if (error) 332675c3f29dSDave Hansen goto out_dput; 3327dae6ad8fSAl Viro error = security_path_symlink(&path, dentry, from); 3328be6d3e56SKentaro Takeda if (error) 3329be6d3e56SKentaro Takeda goto out_drop_write; 3330dae6ad8fSAl Viro error = vfs_symlink(path.dentry->d_inode, dentry, from); 3331be6d3e56SKentaro Takeda out_drop_write: 3332dae6ad8fSAl Viro mnt_drop_write(path.mnt); 333375c3f29dSDave Hansen out_dput: 33341da177e4SLinus Torvalds dput(dentry); 3335dae6ad8fSAl Viro mutex_unlock(&path.dentry->d_inode->i_mutex); 3336dae6ad8fSAl Viro path_put(&path); 33376902d925SDave Hansen out_putname: 33381da177e4SLinus Torvalds putname(from); 33391da177e4SLinus Torvalds return error; 33401da177e4SLinus Torvalds } 33411da177e4SLinus Torvalds 33423480b257SHeiko Carstens SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) 33435590ff0dSUlrich Drepper { 33445590ff0dSUlrich Drepper return sys_symlinkat(oldname, AT_FDCWD, newname); 33455590ff0dSUlrich Drepper } 33465590ff0dSUlrich Drepper 33471da177e4SLinus Torvalds int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry) 33481da177e4SLinus Torvalds { 33491da177e4SLinus Torvalds struct inode *inode = old_dentry->d_inode; 33508de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 33511da177e4SLinus Torvalds int error; 33521da177e4SLinus Torvalds 33531da177e4SLinus Torvalds if (!inode) 33541da177e4SLinus Torvalds return -ENOENT; 33551da177e4SLinus Torvalds 3356a95164d9SMiklos Szeredi error = may_create(dir, new_dentry); 33571da177e4SLinus Torvalds if (error) 33581da177e4SLinus Torvalds return error; 33591da177e4SLinus Torvalds 33601da177e4SLinus Torvalds if (dir->i_sb != inode->i_sb) 33611da177e4SLinus Torvalds return -EXDEV; 33621da177e4SLinus Torvalds 33631da177e4SLinus Torvalds /* 33641da177e4SLinus Torvalds * A link to an append-only or immutable file cannot be created. 33651da177e4SLinus Torvalds */ 33661da177e4SLinus Torvalds if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 33671da177e4SLinus Torvalds return -EPERM; 3368acfa4380SAl Viro if (!dir->i_op->link) 33691da177e4SLinus Torvalds return -EPERM; 33707e79eedbSTetsuo Handa if (S_ISDIR(inode->i_mode)) 33711da177e4SLinus Torvalds return -EPERM; 33721da177e4SLinus Torvalds 33731da177e4SLinus Torvalds error = security_inode_link(old_dentry, dir, new_dentry); 33741da177e4SLinus Torvalds if (error) 33751da177e4SLinus Torvalds return error; 33761da177e4SLinus Torvalds 33777e79eedbSTetsuo Handa mutex_lock(&inode->i_mutex); 3378aae8a97dSAneesh Kumar K.V /* Make sure we don't allow creating hardlink to an unlinked file */ 3379aae8a97dSAneesh Kumar K.V if (inode->i_nlink == 0) 3380aae8a97dSAneesh Kumar K.V error = -ENOENT; 33818de52778SAl Viro else if (max_links && inode->i_nlink >= max_links) 33828de52778SAl Viro error = -EMLINK; 3383aae8a97dSAneesh Kumar K.V else 33841da177e4SLinus Torvalds error = dir->i_op->link(old_dentry, dir, new_dentry); 33857e79eedbSTetsuo Handa mutex_unlock(&inode->i_mutex); 3386e31e14ecSStephen Smalley if (!error) 33877e79eedbSTetsuo Handa fsnotify_link(dir, inode, new_dentry); 33881da177e4SLinus Torvalds return error; 33891da177e4SLinus Torvalds } 33901da177e4SLinus Torvalds 33911da177e4SLinus Torvalds /* 33921da177e4SLinus Torvalds * Hardlinks are often used in delicate situations. We avoid 33931da177e4SLinus Torvalds * security-related surprises by not following symlinks on the 33941da177e4SLinus Torvalds * newname. --KAB 33951da177e4SLinus Torvalds * 33961da177e4SLinus Torvalds * We don't follow them on the oldname either to be compatible 33971da177e4SLinus Torvalds * with linux 2.0, and to avoid hard-linking to directories 33981da177e4SLinus Torvalds * and other special files. --ADM 33991da177e4SLinus Torvalds */ 34002e4d0924SHeiko Carstens SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, 34012e4d0924SHeiko Carstens int, newdfd, const char __user *, newname, int, flags) 34021da177e4SLinus Torvalds { 34031da177e4SLinus Torvalds struct dentry *new_dentry; 3404dae6ad8fSAl Viro struct path old_path, new_path; 340511a7b371SAneesh Kumar K.V int how = 0; 34061da177e4SLinus Torvalds int error; 34071da177e4SLinus Torvalds 340811a7b371SAneesh Kumar K.V if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) 3409c04030e1SUlrich Drepper return -EINVAL; 341011a7b371SAneesh Kumar K.V /* 341111a7b371SAneesh Kumar K.V * To use null names we require CAP_DAC_READ_SEARCH 341211a7b371SAneesh Kumar K.V * This ensures that not everyone will be able to create 341311a7b371SAneesh Kumar K.V * handlink using the passed filedescriptor. 341411a7b371SAneesh Kumar K.V */ 341511a7b371SAneesh Kumar K.V if (flags & AT_EMPTY_PATH) { 341611a7b371SAneesh Kumar K.V if (!capable(CAP_DAC_READ_SEARCH)) 341711a7b371SAneesh Kumar K.V return -ENOENT; 341811a7b371SAneesh Kumar K.V how = LOOKUP_EMPTY; 341911a7b371SAneesh Kumar K.V } 3420c04030e1SUlrich Drepper 342111a7b371SAneesh Kumar K.V if (flags & AT_SYMLINK_FOLLOW) 342211a7b371SAneesh Kumar K.V how |= LOOKUP_FOLLOW; 342311a7b371SAneesh Kumar K.V 342411a7b371SAneesh Kumar K.V error = user_path_at(olddfd, oldname, how, &old_path); 34251da177e4SLinus Torvalds if (error) 34262ad94ae6SAl Viro return error; 34272ad94ae6SAl Viro 3428dae6ad8fSAl Viro new_dentry = user_path_create(newdfd, newname, &new_path, 0); 34291da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 34306902d925SDave Hansen if (IS_ERR(new_dentry)) 3431dae6ad8fSAl Viro goto out; 3432dae6ad8fSAl Viro 3433dae6ad8fSAl Viro error = -EXDEV; 3434dae6ad8fSAl Viro if (old_path.mnt != new_path.mnt) 3435dae6ad8fSAl Viro goto out_dput; 3436dae6ad8fSAl Viro error = mnt_want_write(new_path.mnt); 343775c3f29dSDave Hansen if (error) 343875c3f29dSDave Hansen goto out_dput; 3439dae6ad8fSAl Viro error = security_path_link(old_path.dentry, &new_path, new_dentry); 3440be6d3e56SKentaro Takeda if (error) 3441be6d3e56SKentaro Takeda goto out_drop_write; 3442dae6ad8fSAl Viro error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry); 3443be6d3e56SKentaro Takeda out_drop_write: 3444dae6ad8fSAl Viro mnt_drop_write(new_path.mnt); 344575c3f29dSDave Hansen out_dput: 34461da177e4SLinus Torvalds dput(new_dentry); 3447dae6ad8fSAl Viro mutex_unlock(&new_path.dentry->d_inode->i_mutex); 3448dae6ad8fSAl Viro path_put(&new_path); 34491da177e4SLinus Torvalds out: 34502d8f3038SAl Viro path_put(&old_path); 34511da177e4SLinus Torvalds 34521da177e4SLinus Torvalds return error; 34531da177e4SLinus Torvalds } 34541da177e4SLinus Torvalds 34553480b257SHeiko Carstens SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) 34565590ff0dSUlrich Drepper { 3457c04030e1SUlrich Drepper return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 34585590ff0dSUlrich Drepper } 34595590ff0dSUlrich Drepper 34601da177e4SLinus Torvalds /* 34611da177e4SLinus Torvalds * The worst of all namespace operations - renaming directory. "Perverted" 34621da177e4SLinus Torvalds * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 34631da177e4SLinus Torvalds * Problems: 34641da177e4SLinus Torvalds * a) we can get into loop creation. Check is done in is_subdir(). 34651da177e4SLinus Torvalds * b) race potential - two innocent renames can create a loop together. 34661da177e4SLinus Torvalds * That's where 4.4 screws up. Current fix: serialization on 3467a11f3a05SArjan van de Ven * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 34681da177e4SLinus Torvalds * story. 34691da177e4SLinus Torvalds * c) we have to lock _three_ objects - parents and victim (if it exists). 34701b1dcc1bSJes Sorensen * And that - after we got ->i_mutex on parents (until then we don't know 34711da177e4SLinus Torvalds * whether the target exists). Solution: try to be smart with locking 34721da177e4SLinus Torvalds * order for inodes. We rely on the fact that tree topology may change 3473a11f3a05SArjan van de Ven * only under ->s_vfs_rename_mutex _and_ that parent of the object we 34741da177e4SLinus Torvalds * move will be locked. Thus we can rank directories by the tree 34751da177e4SLinus Torvalds * (ancestors first) and rank all non-directories after them. 34761da177e4SLinus Torvalds * That works since everybody except rename does "lock parent, lookup, 3477a11f3a05SArjan van de Ven * lock child" and rename is under ->s_vfs_rename_mutex. 34781da177e4SLinus Torvalds * HOWEVER, it relies on the assumption that any object with ->lookup() 34791da177e4SLinus Torvalds * has no more than 1 dentry. If "hybrid" objects will ever appear, 34801da177e4SLinus Torvalds * we'd better make sure that there's no link(2) for them. 3481e4eaac06SSage Weil * d) conversion from fhandle to dentry may come in the wrong moment - when 34821b1dcc1bSJes Sorensen * we are removing the target. Solution: we will have to grab ->i_mutex 34831da177e4SLinus Torvalds * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 3484c41b20e7SAdam Buchbinder * ->i_mutex on parents, which works but leads to some truly excessive 34851da177e4SLinus Torvalds * locking]. 34861da177e4SLinus Torvalds */ 348775c96f85SAdrian Bunk static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry, 34881da177e4SLinus Torvalds struct inode *new_dir, struct dentry *new_dentry) 34891da177e4SLinus Torvalds { 34901da177e4SLinus Torvalds int error = 0; 34919055cba7SSage Weil struct inode *target = new_dentry->d_inode; 34928de52778SAl Viro unsigned max_links = new_dir->i_sb->s_max_links; 34931da177e4SLinus Torvalds 34941da177e4SLinus Torvalds /* 34951da177e4SLinus Torvalds * If we are going to change the parent - check write permissions, 34961da177e4SLinus Torvalds * we'll need to flip '..'. 34971da177e4SLinus Torvalds */ 34981da177e4SLinus Torvalds if (new_dir != old_dir) { 3499f419a2e3SAl Viro error = inode_permission(old_dentry->d_inode, MAY_WRITE); 35001da177e4SLinus Torvalds if (error) 35011da177e4SLinus Torvalds return error; 35021da177e4SLinus Torvalds } 35031da177e4SLinus Torvalds 35041da177e4SLinus Torvalds error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); 35051da177e4SLinus Torvalds if (error) 35061da177e4SLinus Torvalds return error; 35071da177e4SLinus Torvalds 35081d2ef590SAl Viro dget(new_dentry); 3509d83c49f3SAl Viro if (target) 35101b1dcc1bSJes Sorensen mutex_lock(&target->i_mutex); 35119055cba7SSage Weil 35121da177e4SLinus Torvalds error = -EBUSY; 35139055cba7SSage Weil if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry)) 35149055cba7SSage Weil goto out; 35159055cba7SSage Weil 35168de52778SAl Viro error = -EMLINK; 35178de52778SAl Viro if (max_links && !target && new_dir != old_dir && 35188de52778SAl Viro new_dir->i_nlink >= max_links) 35198de52778SAl Viro goto out; 35208de52778SAl Viro 35213cebde24SSage Weil if (target) 35223cebde24SSage Weil shrink_dcache_parent(new_dentry); 35231da177e4SLinus Torvalds error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); 35249055cba7SSage Weil if (error) 35259055cba7SSage Weil goto out; 35269055cba7SSage Weil 35271da177e4SLinus Torvalds if (target) { 35281da177e4SLinus Torvalds target->i_flags |= S_DEAD; 3529d83c49f3SAl Viro dont_mount(new_dentry); 3530d83c49f3SAl Viro } 35319055cba7SSage Weil out: 35329055cba7SSage Weil if (target) 35331b1dcc1bSJes Sorensen mutex_unlock(&target->i_mutex); 35341d2ef590SAl Viro dput(new_dentry); 3535e31e14ecSStephen Smalley if (!error) 3536349457ccSMark Fasheh if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) 35371da177e4SLinus Torvalds d_move(old_dentry,new_dentry); 35381da177e4SLinus Torvalds return error; 35391da177e4SLinus Torvalds } 35401da177e4SLinus Torvalds 354175c96f85SAdrian Bunk static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry, 35421da177e4SLinus Torvalds struct inode *new_dir, struct dentry *new_dentry) 35431da177e4SLinus Torvalds { 354451892bbbSSage Weil struct inode *target = new_dentry->d_inode; 35451da177e4SLinus Torvalds int error; 35461da177e4SLinus Torvalds 35471da177e4SLinus Torvalds error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); 35481da177e4SLinus Torvalds if (error) 35491da177e4SLinus Torvalds return error; 35501da177e4SLinus Torvalds 35511da177e4SLinus Torvalds dget(new_dentry); 35521da177e4SLinus Torvalds if (target) 35531b1dcc1bSJes Sorensen mutex_lock(&target->i_mutex); 355451892bbbSSage Weil 35551da177e4SLinus Torvalds error = -EBUSY; 355651892bbbSSage Weil if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry)) 355751892bbbSSage Weil goto out; 355851892bbbSSage Weil 35591da177e4SLinus Torvalds error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); 356051892bbbSSage Weil if (error) 356151892bbbSSage Weil goto out; 356251892bbbSSage Weil 3563bec1052eSAl Viro if (target) 3564d83c49f3SAl Viro dont_mount(new_dentry); 3565349457ccSMark Fasheh if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) 35661da177e4SLinus Torvalds d_move(old_dentry, new_dentry); 356751892bbbSSage Weil out: 35681da177e4SLinus Torvalds if (target) 35691b1dcc1bSJes Sorensen mutex_unlock(&target->i_mutex); 35701da177e4SLinus Torvalds dput(new_dentry); 35711da177e4SLinus Torvalds return error; 35721da177e4SLinus Torvalds } 35731da177e4SLinus Torvalds 35741da177e4SLinus Torvalds int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, 35751da177e4SLinus Torvalds struct inode *new_dir, struct dentry *new_dentry) 35761da177e4SLinus Torvalds { 35771da177e4SLinus Torvalds int error; 35781da177e4SLinus Torvalds int is_dir = S_ISDIR(old_dentry->d_inode->i_mode); 357959b0df21SEric Paris const unsigned char *old_name; 35801da177e4SLinus Torvalds 35811da177e4SLinus Torvalds if (old_dentry->d_inode == new_dentry->d_inode) 35821da177e4SLinus Torvalds return 0; 35831da177e4SLinus Torvalds 35841da177e4SLinus Torvalds error = may_delete(old_dir, old_dentry, is_dir); 35851da177e4SLinus Torvalds if (error) 35861da177e4SLinus Torvalds return error; 35871da177e4SLinus Torvalds 35881da177e4SLinus Torvalds if (!new_dentry->d_inode) 3589a95164d9SMiklos Szeredi error = may_create(new_dir, new_dentry); 35901da177e4SLinus Torvalds else 35911da177e4SLinus Torvalds error = may_delete(new_dir, new_dentry, is_dir); 35921da177e4SLinus Torvalds if (error) 35931da177e4SLinus Torvalds return error; 35941da177e4SLinus Torvalds 3595acfa4380SAl Viro if (!old_dir->i_op->rename) 35961da177e4SLinus Torvalds return -EPERM; 35971da177e4SLinus Torvalds 35980eeca283SRobert Love old_name = fsnotify_oldname_init(old_dentry->d_name.name); 35990eeca283SRobert Love 36001da177e4SLinus Torvalds if (is_dir) 36011da177e4SLinus Torvalds error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry); 36021da177e4SLinus Torvalds else 36031da177e4SLinus Torvalds error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry); 3604123df294SAl Viro if (!error) 3605123df294SAl Viro fsnotify_move(old_dir, new_dir, old_name, is_dir, 36065a190ae6SAl Viro new_dentry->d_inode, old_dentry); 36070eeca283SRobert Love fsnotify_oldname_free(old_name); 36080eeca283SRobert Love 36091da177e4SLinus Torvalds return error; 36101da177e4SLinus Torvalds } 36111da177e4SLinus Torvalds 36122e4d0924SHeiko Carstens SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, 36132e4d0924SHeiko Carstens int, newdfd, const char __user *, newname) 36141da177e4SLinus Torvalds { 36151da177e4SLinus Torvalds struct dentry *old_dir, *new_dir; 36161da177e4SLinus Torvalds struct dentry *old_dentry, *new_dentry; 36171da177e4SLinus Torvalds struct dentry *trap; 36181da177e4SLinus Torvalds struct nameidata oldnd, newnd; 36192ad94ae6SAl Viro char *from; 36202ad94ae6SAl Viro char *to; 36212ad94ae6SAl Viro int error; 36221da177e4SLinus Torvalds 36232ad94ae6SAl Viro error = user_path_parent(olddfd, oldname, &oldnd, &from); 36241da177e4SLinus Torvalds if (error) 36251da177e4SLinus Torvalds goto exit; 36261da177e4SLinus Torvalds 36272ad94ae6SAl Viro error = user_path_parent(newdfd, newname, &newnd, &to); 36281da177e4SLinus Torvalds if (error) 36291da177e4SLinus Torvalds goto exit1; 36301da177e4SLinus Torvalds 36311da177e4SLinus Torvalds error = -EXDEV; 36324ac91378SJan Blunck if (oldnd.path.mnt != newnd.path.mnt) 36331da177e4SLinus Torvalds goto exit2; 36341da177e4SLinus Torvalds 36354ac91378SJan Blunck old_dir = oldnd.path.dentry; 36361da177e4SLinus Torvalds error = -EBUSY; 36371da177e4SLinus Torvalds if (oldnd.last_type != LAST_NORM) 36381da177e4SLinus Torvalds goto exit2; 36391da177e4SLinus Torvalds 36404ac91378SJan Blunck new_dir = newnd.path.dentry; 36411da177e4SLinus Torvalds if (newnd.last_type != LAST_NORM) 36421da177e4SLinus Torvalds goto exit2; 36431da177e4SLinus Torvalds 36440612d9fbSOGAWA Hirofumi oldnd.flags &= ~LOOKUP_PARENT; 36450612d9fbSOGAWA Hirofumi newnd.flags &= ~LOOKUP_PARENT; 36464e9ed2f8SOGAWA Hirofumi newnd.flags |= LOOKUP_RENAME_TARGET; 36470612d9fbSOGAWA Hirofumi 36481da177e4SLinus Torvalds trap = lock_rename(new_dir, old_dir); 36491da177e4SLinus Torvalds 365049705b77SChristoph Hellwig old_dentry = lookup_hash(&oldnd); 36511da177e4SLinus Torvalds error = PTR_ERR(old_dentry); 36521da177e4SLinus Torvalds if (IS_ERR(old_dentry)) 36531da177e4SLinus Torvalds goto exit3; 36541da177e4SLinus Torvalds /* source must exist */ 36551da177e4SLinus Torvalds error = -ENOENT; 36561da177e4SLinus Torvalds if (!old_dentry->d_inode) 36571da177e4SLinus Torvalds goto exit4; 36581da177e4SLinus Torvalds /* unless the source is a directory trailing slashes give -ENOTDIR */ 36591da177e4SLinus Torvalds if (!S_ISDIR(old_dentry->d_inode->i_mode)) { 36601da177e4SLinus Torvalds error = -ENOTDIR; 36611da177e4SLinus Torvalds if (oldnd.last.name[oldnd.last.len]) 36621da177e4SLinus Torvalds goto exit4; 36631da177e4SLinus Torvalds if (newnd.last.name[newnd.last.len]) 36641da177e4SLinus Torvalds goto exit4; 36651da177e4SLinus Torvalds } 36661da177e4SLinus Torvalds /* source should not be ancestor of target */ 36671da177e4SLinus Torvalds error = -EINVAL; 36681da177e4SLinus Torvalds if (old_dentry == trap) 36691da177e4SLinus Torvalds goto exit4; 367049705b77SChristoph Hellwig new_dentry = lookup_hash(&newnd); 36711da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 36721da177e4SLinus Torvalds if (IS_ERR(new_dentry)) 36731da177e4SLinus Torvalds goto exit4; 36741da177e4SLinus Torvalds /* target should not be an ancestor of source */ 36751da177e4SLinus Torvalds error = -ENOTEMPTY; 36761da177e4SLinus Torvalds if (new_dentry == trap) 36771da177e4SLinus Torvalds goto exit5; 36781da177e4SLinus Torvalds 36799079b1ebSDave Hansen error = mnt_want_write(oldnd.path.mnt); 36809079b1ebSDave Hansen if (error) 36819079b1ebSDave Hansen goto exit5; 3682be6d3e56SKentaro Takeda error = security_path_rename(&oldnd.path, old_dentry, 3683be6d3e56SKentaro Takeda &newnd.path, new_dentry); 3684be6d3e56SKentaro Takeda if (error) 3685be6d3e56SKentaro Takeda goto exit6; 36861da177e4SLinus Torvalds error = vfs_rename(old_dir->d_inode, old_dentry, 36871da177e4SLinus Torvalds new_dir->d_inode, new_dentry); 3688be6d3e56SKentaro Takeda exit6: 36899079b1ebSDave Hansen mnt_drop_write(oldnd.path.mnt); 36901da177e4SLinus Torvalds exit5: 36911da177e4SLinus Torvalds dput(new_dentry); 36921da177e4SLinus Torvalds exit4: 36931da177e4SLinus Torvalds dput(old_dentry); 36941da177e4SLinus Torvalds exit3: 36951da177e4SLinus Torvalds unlock_rename(new_dir, old_dir); 36961da177e4SLinus Torvalds exit2: 36971d957f9bSJan Blunck path_put(&newnd.path); 36982ad94ae6SAl Viro putname(to); 36991da177e4SLinus Torvalds exit1: 37001d957f9bSJan Blunck path_put(&oldnd.path); 37011da177e4SLinus Torvalds putname(from); 37022ad94ae6SAl Viro exit: 37031da177e4SLinus Torvalds return error; 37041da177e4SLinus Torvalds } 37051da177e4SLinus Torvalds 3706a26eab24SHeiko Carstens SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) 37075590ff0dSUlrich Drepper { 37085590ff0dSUlrich Drepper return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname); 37095590ff0dSUlrich Drepper } 37105590ff0dSUlrich Drepper 37111da177e4SLinus Torvalds int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link) 37121da177e4SLinus Torvalds { 37131da177e4SLinus Torvalds int len; 37141da177e4SLinus Torvalds 37151da177e4SLinus Torvalds len = PTR_ERR(link); 37161da177e4SLinus Torvalds if (IS_ERR(link)) 37171da177e4SLinus Torvalds goto out; 37181da177e4SLinus Torvalds 37191da177e4SLinus Torvalds len = strlen(link); 37201da177e4SLinus Torvalds if (len > (unsigned) buflen) 37211da177e4SLinus Torvalds len = buflen; 37221da177e4SLinus Torvalds if (copy_to_user(buffer, link, len)) 37231da177e4SLinus Torvalds len = -EFAULT; 37241da177e4SLinus Torvalds out: 37251da177e4SLinus Torvalds return len; 37261da177e4SLinus Torvalds } 37271da177e4SLinus Torvalds 37281da177e4SLinus Torvalds /* 37291da177e4SLinus Torvalds * A helper for ->readlink(). This should be used *ONLY* for symlinks that 37301da177e4SLinus Torvalds * have ->follow_link() touching nd only in nd_set_link(). Using (or not 37311da177e4SLinus Torvalds * using) it for any given inode is up to filesystem. 37321da177e4SLinus Torvalds */ 37331da177e4SLinus Torvalds int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen) 37341da177e4SLinus Torvalds { 37351da177e4SLinus Torvalds struct nameidata nd; 3736cc314eefSLinus Torvalds void *cookie; 3737694a1764SMarcin Slusarz int res; 3738cc314eefSLinus Torvalds 37391da177e4SLinus Torvalds nd.depth = 0; 3740cc314eefSLinus Torvalds cookie = dentry->d_inode->i_op->follow_link(dentry, &nd); 3741694a1764SMarcin Slusarz if (IS_ERR(cookie)) 3742694a1764SMarcin Slusarz return PTR_ERR(cookie); 3743694a1764SMarcin Slusarz 3744694a1764SMarcin Slusarz res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd)); 37451da177e4SLinus Torvalds if (dentry->d_inode->i_op->put_link) 3746cc314eefSLinus Torvalds dentry->d_inode->i_op->put_link(dentry, &nd, cookie); 3747694a1764SMarcin Slusarz return res; 37481da177e4SLinus Torvalds } 37491da177e4SLinus Torvalds 37501da177e4SLinus Torvalds int vfs_follow_link(struct nameidata *nd, const char *link) 37511da177e4SLinus Torvalds { 37521da177e4SLinus Torvalds return __vfs_follow_link(nd, link); 37531da177e4SLinus Torvalds } 37541da177e4SLinus Torvalds 37551da177e4SLinus Torvalds /* get the link contents into pagecache */ 37561da177e4SLinus Torvalds static char *page_getlink(struct dentry * dentry, struct page **ppage) 37571da177e4SLinus Torvalds { 3758ebd09abbSDuane Griffin char *kaddr; 37591da177e4SLinus Torvalds struct page *page; 37601da177e4SLinus Torvalds struct address_space *mapping = dentry->d_inode->i_mapping; 3761090d2b18SPekka Enberg page = read_mapping_page(mapping, 0, NULL); 37621da177e4SLinus Torvalds if (IS_ERR(page)) 37636fe6900eSNick Piggin return (char*)page; 37641da177e4SLinus Torvalds *ppage = page; 3765ebd09abbSDuane Griffin kaddr = kmap(page); 3766ebd09abbSDuane Griffin nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1); 3767ebd09abbSDuane Griffin return kaddr; 37681da177e4SLinus Torvalds } 37691da177e4SLinus Torvalds 37701da177e4SLinus Torvalds int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 37711da177e4SLinus Torvalds { 37721da177e4SLinus Torvalds struct page *page = NULL; 37731da177e4SLinus Torvalds char *s = page_getlink(dentry, &page); 37741da177e4SLinus Torvalds int res = vfs_readlink(dentry,buffer,buflen,s); 37751da177e4SLinus Torvalds if (page) { 37761da177e4SLinus Torvalds kunmap(page); 37771da177e4SLinus Torvalds page_cache_release(page); 37781da177e4SLinus Torvalds } 37791da177e4SLinus Torvalds return res; 37801da177e4SLinus Torvalds } 37811da177e4SLinus Torvalds 3782cc314eefSLinus Torvalds void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd) 37831da177e4SLinus Torvalds { 3784cc314eefSLinus Torvalds struct page *page = NULL; 37851da177e4SLinus Torvalds nd_set_link(nd, page_getlink(dentry, &page)); 3786cc314eefSLinus Torvalds return page; 37871da177e4SLinus Torvalds } 37881da177e4SLinus Torvalds 3789cc314eefSLinus Torvalds void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) 37901da177e4SLinus Torvalds { 3791cc314eefSLinus Torvalds struct page *page = cookie; 3792cc314eefSLinus Torvalds 3793cc314eefSLinus Torvalds if (page) { 37941da177e4SLinus Torvalds kunmap(page); 37951da177e4SLinus Torvalds page_cache_release(page); 37961da177e4SLinus Torvalds } 37971da177e4SLinus Torvalds } 37981da177e4SLinus Torvalds 379954566b2cSNick Piggin /* 380054566b2cSNick Piggin * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS 380154566b2cSNick Piggin */ 380254566b2cSNick Piggin int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) 38031da177e4SLinus Torvalds { 38041da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 38050adb25d2SKirill Korotaev struct page *page; 3806afddba49SNick Piggin void *fsdata; 3807beb497abSDmitriy Monakhov int err; 38081da177e4SLinus Torvalds char *kaddr; 380954566b2cSNick Piggin unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE; 381054566b2cSNick Piggin if (nofs) 381154566b2cSNick Piggin flags |= AOP_FLAG_NOFS; 38121da177e4SLinus Torvalds 38137e53cac4SNeilBrown retry: 3814afddba49SNick Piggin err = pagecache_write_begin(NULL, mapping, 0, len-1, 381554566b2cSNick Piggin flags, &page, &fsdata); 38161da177e4SLinus Torvalds if (err) 3817afddba49SNick Piggin goto fail; 3818afddba49SNick Piggin 3819e8e3c3d6SCong Wang kaddr = kmap_atomic(page); 38201da177e4SLinus Torvalds memcpy(kaddr, symname, len-1); 3821e8e3c3d6SCong Wang kunmap_atomic(kaddr); 3822afddba49SNick Piggin 3823afddba49SNick Piggin err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, 3824afddba49SNick Piggin page, fsdata); 38251da177e4SLinus Torvalds if (err < 0) 38261da177e4SLinus Torvalds goto fail; 3827afddba49SNick Piggin if (err < len-1) 3828afddba49SNick Piggin goto retry; 3829afddba49SNick Piggin 38301da177e4SLinus Torvalds mark_inode_dirty(inode); 38311da177e4SLinus Torvalds return 0; 38321da177e4SLinus Torvalds fail: 38331da177e4SLinus Torvalds return err; 38341da177e4SLinus Torvalds } 38351da177e4SLinus Torvalds 38360adb25d2SKirill Korotaev int page_symlink(struct inode *inode, const char *symname, int len) 38370adb25d2SKirill Korotaev { 38380adb25d2SKirill Korotaev return __page_symlink(inode, symname, len, 383954566b2cSNick Piggin !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS)); 38400adb25d2SKirill Korotaev } 38410adb25d2SKirill Korotaev 384292e1d5beSArjan van de Ven const struct inode_operations page_symlink_inode_operations = { 38431da177e4SLinus Torvalds .readlink = generic_readlink, 38441da177e4SLinus Torvalds .follow_link = page_follow_link_light, 38451da177e4SLinus Torvalds .put_link = page_put_link, 38461da177e4SLinus Torvalds }; 38471da177e4SLinus Torvalds 38482d8f3038SAl Viro EXPORT_SYMBOL(user_path_at); 3849cc53ce53SDavid Howells EXPORT_SYMBOL(follow_down_one); 38501da177e4SLinus Torvalds EXPORT_SYMBOL(follow_down); 38511da177e4SLinus Torvalds EXPORT_SYMBOL(follow_up); 38521da177e4SLinus Torvalds EXPORT_SYMBOL(get_write_access); /* binfmt_aout */ 38531da177e4SLinus Torvalds EXPORT_SYMBOL(getname); 38541da177e4SLinus Torvalds EXPORT_SYMBOL(lock_rename); 38551da177e4SLinus Torvalds EXPORT_SYMBOL(lookup_one_len); 38561da177e4SLinus Torvalds EXPORT_SYMBOL(page_follow_link_light); 38571da177e4SLinus Torvalds EXPORT_SYMBOL(page_put_link); 38581da177e4SLinus Torvalds EXPORT_SYMBOL(page_readlink); 38590adb25d2SKirill Korotaev EXPORT_SYMBOL(__page_symlink); 38601da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink); 38611da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink_inode_operations); 3862d1811465SAl Viro EXPORT_SYMBOL(kern_path); 386316f18200SJosef 'Jeff' Sipek EXPORT_SYMBOL(vfs_path_lookup); 3864f419a2e3SAl Viro EXPORT_SYMBOL(inode_permission); 38651da177e4SLinus Torvalds EXPORT_SYMBOL(unlock_rename); 38661da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_create); 38671da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_follow_link); 38681da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_link); 38691da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_mkdir); 38701da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_mknod); 38711da177e4SLinus Torvalds EXPORT_SYMBOL(generic_permission); 38721da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_readlink); 38731da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_rename); 38741da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_rmdir); 38751da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_symlink); 38761da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_unlink); 38771da177e4SLinus Torvalds EXPORT_SYMBOL(dentry_unhash); 38781da177e4SLinus Torvalds EXPORT_SYMBOL(generic_readlink); 3879