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 35555852635SRandy Dunlap * @inode: Inode to check permission on 3560bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 3570bdaea90SDavid Howells * 3580bdaea90SDavid Howells * Separate out file-system wide checks from inode-specific permission checks. 3590bdaea90SDavid Howells */ 3600bdaea90SDavid Howells static int sb_permission(struct super_block *sb, struct inode *inode, int mask) 3610bdaea90SDavid Howells { 3620bdaea90SDavid Howells if (unlikely(mask & MAY_WRITE)) { 3630bdaea90SDavid Howells umode_t mode = inode->i_mode; 3640bdaea90SDavid Howells 3650bdaea90SDavid Howells /* Nobody gets write access to a read-only fs. */ 3660bdaea90SDavid Howells if ((sb->s_flags & MS_RDONLY) && 3670bdaea90SDavid Howells (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) 3680bdaea90SDavid Howells return -EROFS; 3690bdaea90SDavid Howells } 3700bdaea90SDavid Howells return 0; 3710bdaea90SDavid Howells } 3720bdaea90SDavid Howells 3730bdaea90SDavid Howells /** 3740bdaea90SDavid Howells * inode_permission - Check for access rights to a given inode 3750bdaea90SDavid Howells * @inode: Inode to check permission on 3760bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 3770bdaea90SDavid Howells * 3780bdaea90SDavid Howells * Check for read/write/execute permissions on an inode. We use fs[ug]id for 3790bdaea90SDavid Howells * this, letting us set arbitrary permissions for filesystem access without 3800bdaea90SDavid Howells * changing the "normal" UIDs which are used for other things. 3810bdaea90SDavid Howells * 3820bdaea90SDavid Howells * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. 3830bdaea90SDavid Howells */ 3840bdaea90SDavid Howells int inode_permission(struct inode *inode, int mask) 3850bdaea90SDavid Howells { 3860bdaea90SDavid Howells int retval; 3870bdaea90SDavid Howells 3880bdaea90SDavid Howells retval = sb_permission(inode->i_sb, inode, mask); 3890bdaea90SDavid Howells if (retval) 3900bdaea90SDavid Howells return retval; 3910bdaea90SDavid Howells return __inode_permission(inode, mask); 3920bdaea90SDavid Howells } 3930bdaea90SDavid Howells 3940bdaea90SDavid Howells /** 3955dd784d0SJan Blunck * path_get - get a reference to a path 3965dd784d0SJan Blunck * @path: path to get the reference to 3975dd784d0SJan Blunck * 3985dd784d0SJan Blunck * Given a path increment the reference count to the dentry and the vfsmount. 3995dd784d0SJan Blunck */ 4005dd784d0SJan Blunck void path_get(struct path *path) 4015dd784d0SJan Blunck { 4025dd784d0SJan Blunck mntget(path->mnt); 4035dd784d0SJan Blunck dget(path->dentry); 4045dd784d0SJan Blunck } 4055dd784d0SJan Blunck EXPORT_SYMBOL(path_get); 4065dd784d0SJan Blunck 4075dd784d0SJan Blunck /** 4081d957f9bSJan Blunck * path_put - put a reference to a path 4091d957f9bSJan Blunck * @path: path to put the reference to 4101d957f9bSJan Blunck * 4111d957f9bSJan Blunck * Given a path decrement the reference count to the dentry and the vfsmount. 4121d957f9bSJan Blunck */ 4131d957f9bSJan Blunck void path_put(struct path *path) 4141da177e4SLinus Torvalds { 4151d957f9bSJan Blunck dput(path->dentry); 4161d957f9bSJan Blunck mntput(path->mnt); 4171da177e4SLinus Torvalds } 4181d957f9bSJan Blunck EXPORT_SYMBOL(path_put); 4191da177e4SLinus Torvalds 42019660af7SAl Viro /* 42131e6b01fSNick Piggin * Path walking has 2 modes, rcu-walk and ref-walk (see 42219660af7SAl Viro * Documentation/filesystems/path-lookup.txt). In situations when we can't 42319660af7SAl Viro * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab 42419660af7SAl Viro * normal reference counts on dentries and vfsmounts to transition to rcu-walk 42519660af7SAl Viro * mode. Refcounts are grabbed at the last known good point before rcu-walk 42619660af7SAl Viro * got stuck, so ref-walk may continue from there. If this is not successful 42719660af7SAl Viro * (eg. a seqcount has changed), then failure is returned and it's up to caller 42819660af7SAl Viro * to restart the path walk from the beginning in ref-walk mode. 42931e6b01fSNick Piggin */ 43031e6b01fSNick Piggin 43132a7991bSAl Viro static inline void lock_rcu_walk(void) 43232a7991bSAl Viro { 43332a7991bSAl Viro br_read_lock(&vfsmount_lock); 43432a7991bSAl Viro rcu_read_lock(); 43532a7991bSAl Viro } 43632a7991bSAl Viro 43732a7991bSAl Viro static inline void unlock_rcu_walk(void) 43832a7991bSAl Viro { 43932a7991bSAl Viro rcu_read_unlock(); 44032a7991bSAl Viro br_read_unlock(&vfsmount_lock); 44132a7991bSAl Viro } 44232a7991bSAl Viro 44331e6b01fSNick Piggin /** 44419660af7SAl Viro * unlazy_walk - try to switch to ref-walk mode. 44519660af7SAl Viro * @nd: nameidata pathwalk data 44619660af7SAl Viro * @dentry: child of nd->path.dentry or NULL 44739191628SRandy Dunlap * Returns: 0 on success, -ECHILD on failure 44831e6b01fSNick Piggin * 44919660af7SAl Viro * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry 45019660af7SAl Viro * for ref-walk mode. @dentry must be a path found by a do_lookup call on 45119660af7SAl Viro * @nd or NULL. Must be called from rcu-walk context. 45231e6b01fSNick Piggin */ 45319660af7SAl Viro static int unlazy_walk(struct nameidata *nd, struct dentry *dentry) 45431e6b01fSNick Piggin { 45531e6b01fSNick Piggin struct fs_struct *fs = current->fs; 45631e6b01fSNick Piggin struct dentry *parent = nd->path.dentry; 4575b6ca027SAl Viro int want_root = 0; 45831e6b01fSNick Piggin 45931e6b01fSNick Piggin BUG_ON(!(nd->flags & LOOKUP_RCU)); 4605b6ca027SAl Viro if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 4615b6ca027SAl Viro want_root = 1; 46231e6b01fSNick Piggin spin_lock(&fs->lock); 46331e6b01fSNick Piggin if (nd->root.mnt != fs->root.mnt || 46431e6b01fSNick Piggin nd->root.dentry != fs->root.dentry) 46531e6b01fSNick Piggin goto err_root; 46631e6b01fSNick Piggin } 46731e6b01fSNick Piggin spin_lock(&parent->d_lock); 46819660af7SAl Viro if (!dentry) { 46919660af7SAl Viro if (!__d_rcu_to_refcount(parent, nd->seq)) 47019660af7SAl Viro goto err_parent; 47119660af7SAl Viro BUG_ON(nd->inode != parent->d_inode); 47219660af7SAl Viro } else { 47394c0d4ecSAl Viro if (dentry->d_parent != parent) 47494c0d4ecSAl Viro goto err_parent; 47531e6b01fSNick Piggin spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); 47631e6b01fSNick Piggin if (!__d_rcu_to_refcount(dentry, nd->seq)) 47719660af7SAl Viro goto err_child; 47831e6b01fSNick Piggin /* 47919660af7SAl Viro * If the sequence check on the child dentry passed, then 48019660af7SAl Viro * the child has not been removed from its parent. This 48119660af7SAl Viro * means the parent dentry must be valid and able to take 48219660af7SAl Viro * a reference at this point. 48331e6b01fSNick Piggin */ 48431e6b01fSNick Piggin BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent); 48531e6b01fSNick Piggin BUG_ON(!parent->d_count); 48631e6b01fSNick Piggin parent->d_count++; 48731e6b01fSNick Piggin spin_unlock(&dentry->d_lock); 48819660af7SAl Viro } 48931e6b01fSNick Piggin spin_unlock(&parent->d_lock); 4905b6ca027SAl Viro if (want_root) { 49131e6b01fSNick Piggin path_get(&nd->root); 49231e6b01fSNick Piggin spin_unlock(&fs->lock); 49331e6b01fSNick Piggin } 49431e6b01fSNick Piggin mntget(nd->path.mnt); 49531e6b01fSNick Piggin 49632a7991bSAl Viro unlock_rcu_walk(); 49731e6b01fSNick Piggin nd->flags &= ~LOOKUP_RCU; 49831e6b01fSNick Piggin return 0; 49919660af7SAl Viro 50019660af7SAl Viro err_child: 50131e6b01fSNick Piggin spin_unlock(&dentry->d_lock); 50219660af7SAl Viro err_parent: 50331e6b01fSNick Piggin spin_unlock(&parent->d_lock); 50431e6b01fSNick Piggin err_root: 5055b6ca027SAl Viro if (want_root) 50631e6b01fSNick Piggin spin_unlock(&fs->lock); 50731e6b01fSNick Piggin return -ECHILD; 50831e6b01fSNick Piggin } 50931e6b01fSNick Piggin 5104ce16ef3SAl Viro static inline int d_revalidate(struct dentry *dentry, unsigned int flags) 51134286d66SNick Piggin { 5124ce16ef3SAl Viro return dentry->d_op->d_revalidate(dentry, flags); 51334286d66SNick Piggin } 51434286d66SNick Piggin 5159f1fafeeSAl Viro /** 5169f1fafeeSAl Viro * complete_walk - successful completion of path walk 5179f1fafeeSAl Viro * @nd: pointer nameidata 51839159de2SJeff Layton * 5199f1fafeeSAl Viro * If we had been in RCU mode, drop out of it and legitimize nd->path. 5209f1fafeeSAl Viro * Revalidate the final result, unless we'd already done that during 5219f1fafeeSAl Viro * the path walk or the filesystem doesn't ask for it. Return 0 on 5229f1fafeeSAl Viro * success, -error on failure. In case of failure caller does not 5239f1fafeeSAl Viro * need to drop nd->path. 52439159de2SJeff Layton */ 5259f1fafeeSAl Viro static int complete_walk(struct nameidata *nd) 52639159de2SJeff Layton { 52716c2cd71SAl Viro struct dentry *dentry = nd->path.dentry; 52839159de2SJeff Layton int status; 52939159de2SJeff Layton 5309f1fafeeSAl Viro if (nd->flags & LOOKUP_RCU) { 5319f1fafeeSAl Viro nd->flags &= ~LOOKUP_RCU; 5329f1fafeeSAl Viro if (!(nd->flags & LOOKUP_ROOT)) 5339f1fafeeSAl Viro nd->root.mnt = NULL; 5349f1fafeeSAl Viro spin_lock(&dentry->d_lock); 5359f1fafeeSAl Viro if (unlikely(!__d_rcu_to_refcount(dentry, nd->seq))) { 5369f1fafeeSAl Viro spin_unlock(&dentry->d_lock); 53732a7991bSAl Viro unlock_rcu_walk(); 5389f1fafeeSAl Viro return -ECHILD; 5399f1fafeeSAl Viro } 5409f1fafeeSAl Viro BUG_ON(nd->inode != dentry->d_inode); 5419f1fafeeSAl Viro spin_unlock(&dentry->d_lock); 5429f1fafeeSAl Viro mntget(nd->path.mnt); 54332a7991bSAl Viro unlock_rcu_walk(); 5449f1fafeeSAl Viro } 5459f1fafeeSAl Viro 54616c2cd71SAl Viro if (likely(!(nd->flags & LOOKUP_JUMPED))) 54739159de2SJeff Layton return 0; 54839159de2SJeff Layton 54916c2cd71SAl Viro if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE))) 55016c2cd71SAl Viro return 0; 55116c2cd71SAl Viro 55216c2cd71SAl Viro if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT))) 55316c2cd71SAl Viro return 0; 55416c2cd71SAl Viro 55516c2cd71SAl Viro /* Note: we do not d_invalidate() */ 5564ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 55739159de2SJeff Layton if (status > 0) 55839159de2SJeff Layton return 0; 55939159de2SJeff Layton 56016c2cd71SAl Viro if (!status) 56139159de2SJeff Layton status = -ESTALE; 56216c2cd71SAl Viro 5639f1fafeeSAl Viro path_put(&nd->path); 56439159de2SJeff Layton return status; 56539159de2SJeff Layton } 56639159de2SJeff Layton 5672a737871SAl Viro static __always_inline void set_root(struct nameidata *nd) 5682a737871SAl Viro { 569f7ad3c6bSMiklos Szeredi if (!nd->root.mnt) 570f7ad3c6bSMiklos Szeredi get_fs_root(current->fs, &nd->root); 5712a737871SAl Viro } 5722a737871SAl Viro 5736de88d72SAl Viro static int link_path_walk(const char *, struct nameidata *); 5746de88d72SAl Viro 57531e6b01fSNick Piggin static __always_inline void set_root_rcu(struct nameidata *nd) 57631e6b01fSNick Piggin { 57731e6b01fSNick Piggin if (!nd->root.mnt) { 57831e6b01fSNick Piggin struct fs_struct *fs = current->fs; 579c28cc364SNick Piggin unsigned seq; 580c28cc364SNick Piggin 581c28cc364SNick Piggin do { 582c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 58331e6b01fSNick Piggin nd->root = fs->root; 584c1530019STim Chen nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq); 585c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 58631e6b01fSNick Piggin } 58731e6b01fSNick Piggin } 58831e6b01fSNick Piggin 589f1662356SArjan van de Ven static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link) 5901da177e4SLinus Torvalds { 59131e6b01fSNick Piggin int ret; 59231e6b01fSNick Piggin 5931da177e4SLinus Torvalds if (IS_ERR(link)) 5941da177e4SLinus Torvalds goto fail; 5951da177e4SLinus Torvalds 5961da177e4SLinus Torvalds if (*link == '/') { 5972a737871SAl Viro set_root(nd); 5981d957f9bSJan Blunck path_put(&nd->path); 5992a737871SAl Viro nd->path = nd->root; 6002a737871SAl Viro path_get(&nd->root); 60116c2cd71SAl Viro nd->flags |= LOOKUP_JUMPED; 6021da177e4SLinus Torvalds } 60331e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 604b4091d5fSChristoph Hellwig 60531e6b01fSNick Piggin ret = link_path_walk(link, nd); 60631e6b01fSNick Piggin return ret; 6071da177e4SLinus Torvalds fail: 6081d957f9bSJan Blunck path_put(&nd->path); 6091da177e4SLinus Torvalds return PTR_ERR(link); 6101da177e4SLinus Torvalds } 6111da177e4SLinus Torvalds 6121d957f9bSJan Blunck static void path_put_conditional(struct path *path, struct nameidata *nd) 613051d3812SIan Kent { 614051d3812SIan Kent dput(path->dentry); 6154ac91378SJan Blunck if (path->mnt != nd->path.mnt) 616051d3812SIan Kent mntput(path->mnt); 617051d3812SIan Kent } 618051d3812SIan Kent 6197b9337aaSNick Piggin static inline void path_to_nameidata(const struct path *path, 6207b9337aaSNick Piggin struct nameidata *nd) 621051d3812SIan Kent { 62231e6b01fSNick Piggin if (!(nd->flags & LOOKUP_RCU)) { 6234ac91378SJan Blunck dput(nd->path.dentry); 62431e6b01fSNick Piggin if (nd->path.mnt != path->mnt) 6254ac91378SJan Blunck mntput(nd->path.mnt); 6269a229683SHuang Shijie } 62731e6b01fSNick Piggin nd->path.mnt = path->mnt; 6284ac91378SJan Blunck nd->path.dentry = path->dentry; 629051d3812SIan Kent } 630051d3812SIan Kent 631b5fb63c1SChristoph Hellwig /* 632b5fb63c1SChristoph Hellwig * Helper to directly jump to a known parsed path from ->follow_link, 633b5fb63c1SChristoph Hellwig * caller must have taken a reference to path beforehand. 634b5fb63c1SChristoph Hellwig */ 635b5fb63c1SChristoph Hellwig void nd_jump_link(struct nameidata *nd, struct path *path) 636b5fb63c1SChristoph Hellwig { 637b5fb63c1SChristoph Hellwig path_put(&nd->path); 638b5fb63c1SChristoph Hellwig 639b5fb63c1SChristoph Hellwig nd->path = *path; 640b5fb63c1SChristoph Hellwig nd->inode = nd->path.dentry->d_inode; 641b5fb63c1SChristoph Hellwig nd->flags |= LOOKUP_JUMPED; 642b5fb63c1SChristoph Hellwig 643b5fb63c1SChristoph Hellwig BUG_ON(nd->inode->i_op->follow_link); 644b5fb63c1SChristoph Hellwig } 645b5fb63c1SChristoph Hellwig 646574197e0SAl Viro static inline void put_link(struct nameidata *nd, struct path *link, void *cookie) 647574197e0SAl Viro { 648574197e0SAl Viro struct inode *inode = link->dentry->d_inode; 6496d7b5aaeSAl Viro if (inode->i_op->put_link) 650574197e0SAl Viro inode->i_op->put_link(link->dentry, nd, cookie); 651574197e0SAl Viro path_put(link); 652574197e0SAl Viro } 653574197e0SAl Viro 654800179c9SKees Cook int sysctl_protected_symlinks __read_mostly = 1; 655800179c9SKees Cook int sysctl_protected_hardlinks __read_mostly = 1; 656800179c9SKees Cook 657800179c9SKees Cook /** 658800179c9SKees Cook * may_follow_link - Check symlink following for unsafe situations 659800179c9SKees Cook * @link: The path of the symlink 66055852635SRandy Dunlap * @nd: nameidata pathwalk data 661800179c9SKees Cook * 662800179c9SKees Cook * In the case of the sysctl_protected_symlinks sysctl being enabled, 663800179c9SKees Cook * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is 664800179c9SKees Cook * in a sticky world-writable directory. This is to protect privileged 665800179c9SKees Cook * processes from failing races against path names that may change out 666800179c9SKees Cook * from under them by way of other users creating malicious symlinks. 667800179c9SKees Cook * It will permit symlinks to be followed only when outside a sticky 668800179c9SKees Cook * world-writable directory, or when the uid of the symlink and follower 669800179c9SKees Cook * match, or when the directory owner matches the symlink's owner. 670800179c9SKees Cook * 671800179c9SKees Cook * Returns 0 if following the symlink is allowed, -ve on error. 672800179c9SKees Cook */ 673800179c9SKees Cook static inline int may_follow_link(struct path *link, struct nameidata *nd) 674800179c9SKees Cook { 675800179c9SKees Cook const struct inode *inode; 676800179c9SKees Cook const struct inode *parent; 677800179c9SKees Cook 678800179c9SKees Cook if (!sysctl_protected_symlinks) 679800179c9SKees Cook return 0; 680800179c9SKees Cook 681800179c9SKees Cook /* Allowed if owner and follower match. */ 682800179c9SKees Cook inode = link->dentry->d_inode; 68381abe27bSEric W. Biederman if (uid_eq(current_cred()->fsuid, inode->i_uid)) 684800179c9SKees Cook return 0; 685800179c9SKees Cook 686800179c9SKees Cook /* Allowed if parent directory not sticky and world-writable. */ 687800179c9SKees Cook parent = nd->path.dentry->d_inode; 688800179c9SKees Cook if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) 689800179c9SKees Cook return 0; 690800179c9SKees Cook 691800179c9SKees Cook /* Allowed if parent directory and link owner match. */ 69281abe27bSEric W. Biederman if (uid_eq(parent->i_uid, inode->i_uid)) 693800179c9SKees Cook return 0; 694800179c9SKees Cook 695ffd8d101SSasha Levin audit_log_link_denied("follow_link", link); 696800179c9SKees Cook path_put_conditional(link, nd); 697800179c9SKees Cook path_put(&nd->path); 698800179c9SKees Cook return -EACCES; 699800179c9SKees Cook } 700800179c9SKees Cook 701800179c9SKees Cook /** 702800179c9SKees Cook * safe_hardlink_source - Check for safe hardlink conditions 703800179c9SKees Cook * @inode: the source inode to hardlink from 704800179c9SKees Cook * 705800179c9SKees Cook * Return false if at least one of the following conditions: 706800179c9SKees Cook * - inode is not a regular file 707800179c9SKees Cook * - inode is setuid 708800179c9SKees Cook * - inode is setgid and group-exec 709800179c9SKees Cook * - access failure for read and write 710800179c9SKees Cook * 711800179c9SKees Cook * Otherwise returns true. 712800179c9SKees Cook */ 713800179c9SKees Cook static bool safe_hardlink_source(struct inode *inode) 714800179c9SKees Cook { 715800179c9SKees Cook umode_t mode = inode->i_mode; 716800179c9SKees Cook 717800179c9SKees Cook /* Special files should not get pinned to the filesystem. */ 718800179c9SKees Cook if (!S_ISREG(mode)) 719800179c9SKees Cook return false; 720800179c9SKees Cook 721800179c9SKees Cook /* Setuid files should not get pinned to the filesystem. */ 722800179c9SKees Cook if (mode & S_ISUID) 723800179c9SKees Cook return false; 724800179c9SKees Cook 725800179c9SKees Cook /* Executable setgid files should not get pinned to the filesystem. */ 726800179c9SKees Cook if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) 727800179c9SKees Cook return false; 728800179c9SKees Cook 729800179c9SKees Cook /* Hardlinking to unreadable or unwritable sources is dangerous. */ 730800179c9SKees Cook if (inode_permission(inode, MAY_READ | MAY_WRITE)) 731800179c9SKees Cook return false; 732800179c9SKees Cook 733800179c9SKees Cook return true; 734800179c9SKees Cook } 735800179c9SKees Cook 736800179c9SKees Cook /** 737800179c9SKees Cook * may_linkat - Check permissions for creating a hardlink 738800179c9SKees Cook * @link: the source to hardlink from 739800179c9SKees Cook * 740800179c9SKees Cook * Block hardlink when all of: 741800179c9SKees Cook * - sysctl_protected_hardlinks enabled 742800179c9SKees Cook * - fsuid does not match inode 743800179c9SKees Cook * - hardlink source is unsafe (see safe_hardlink_source() above) 744800179c9SKees Cook * - not CAP_FOWNER 745800179c9SKees Cook * 746800179c9SKees Cook * Returns 0 if successful, -ve on error. 747800179c9SKees Cook */ 748800179c9SKees Cook static int may_linkat(struct path *link) 749800179c9SKees Cook { 750800179c9SKees Cook const struct cred *cred; 751800179c9SKees Cook struct inode *inode; 752800179c9SKees Cook 753800179c9SKees Cook if (!sysctl_protected_hardlinks) 754800179c9SKees Cook return 0; 755800179c9SKees Cook 756800179c9SKees Cook cred = current_cred(); 757800179c9SKees Cook inode = link->dentry->d_inode; 758800179c9SKees Cook 759800179c9SKees Cook /* Source inode owner (or CAP_FOWNER) can hardlink all they like, 760800179c9SKees Cook * otherwise, it must be a safe source. 761800179c9SKees Cook */ 76281abe27bSEric W. Biederman if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) || 763800179c9SKees Cook capable(CAP_FOWNER)) 764800179c9SKees Cook return 0; 765800179c9SKees Cook 766a51d9eaaSKees Cook audit_log_link_denied("linkat", link); 767800179c9SKees Cook return -EPERM; 768800179c9SKees Cook } 769800179c9SKees Cook 770def4af30SAl Viro static __always_inline int 771574197e0SAl Viro follow_link(struct path *link, struct nameidata *nd, void **p) 7721da177e4SLinus Torvalds { 7737b9337aaSNick Piggin struct dentry *dentry = link->dentry; 7746d7b5aaeSAl Viro int error; 7756d7b5aaeSAl Viro char *s; 7761da177e4SLinus Torvalds 777844a3917SAl Viro BUG_ON(nd->flags & LOOKUP_RCU); 778844a3917SAl Viro 7790e794589SAl Viro if (link->mnt == nd->path.mnt) 7800e794589SAl Viro mntget(link->mnt); 7810e794589SAl Viro 7826d7b5aaeSAl Viro error = -ELOOP; 7836d7b5aaeSAl Viro if (unlikely(current->total_link_count >= 40)) 7846d7b5aaeSAl Viro goto out_put_nd_path; 7856d7b5aaeSAl Viro 786574197e0SAl Viro cond_resched(); 787574197e0SAl Viro current->total_link_count++; 788574197e0SAl Viro 78968ac1234SAl Viro touch_atime(link); 7901da177e4SLinus Torvalds nd_set_link(nd, NULL); 791cd4e91d3SAl Viro 79236f3b4f6SAl Viro error = security_inode_follow_link(link->dentry, nd); 7936d7b5aaeSAl Viro if (error) 7946d7b5aaeSAl Viro goto out_put_nd_path; 79536f3b4f6SAl Viro 79686acdca1SAl Viro nd->last_type = LAST_BIND; 797def4af30SAl Viro *p = dentry->d_inode->i_op->follow_link(dentry, nd); 798def4af30SAl Viro error = PTR_ERR(*p); 7996d7b5aaeSAl Viro if (IS_ERR(*p)) 800408ef013SChristoph Hellwig goto out_put_nd_path; 8016d7b5aaeSAl Viro 802cc314eefSLinus Torvalds error = 0; 8036d7b5aaeSAl Viro s = nd_get_link(nd); 8046d7b5aaeSAl Viro if (s) { 8051da177e4SLinus Torvalds error = __vfs_follow_link(nd, s); 8066d7b5aaeSAl Viro if (unlikely(error)) 8076d7b5aaeSAl Viro put_link(nd, link, *p); 808b5fb63c1SChristoph Hellwig } 8096d7b5aaeSAl Viro 8106d7b5aaeSAl Viro return error; 8116d7b5aaeSAl Viro 8126d7b5aaeSAl Viro out_put_nd_path: 81398f6ef64SArnd Bergmann *p = NULL; 8146d7b5aaeSAl Viro path_put(&nd->path); 8156d7b5aaeSAl Viro path_put(link); 8161da177e4SLinus Torvalds return error; 8171da177e4SLinus Torvalds } 8181da177e4SLinus Torvalds 81931e6b01fSNick Piggin static int follow_up_rcu(struct path *path) 82031e6b01fSNick Piggin { 8210714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 8220714a533SAl Viro struct mount *parent; 82331e6b01fSNick Piggin struct dentry *mountpoint; 82431e6b01fSNick Piggin 8250714a533SAl Viro parent = mnt->mnt_parent; 8260714a533SAl Viro if (&parent->mnt == path->mnt) 82731e6b01fSNick Piggin return 0; 828a73324daSAl Viro mountpoint = mnt->mnt_mountpoint; 82931e6b01fSNick Piggin path->dentry = mountpoint; 8300714a533SAl Viro path->mnt = &parent->mnt; 83131e6b01fSNick Piggin return 1; 83231e6b01fSNick Piggin } 83331e6b01fSNick Piggin 834f015f126SDavid Howells /* 835f015f126SDavid Howells * follow_up - Find the mountpoint of path's vfsmount 836f015f126SDavid Howells * 837f015f126SDavid Howells * Given a path, find the mountpoint of its source file system. 838f015f126SDavid Howells * Replace @path with the path of the mountpoint in the parent mount. 839f015f126SDavid Howells * Up is towards /. 840f015f126SDavid Howells * 841f015f126SDavid Howells * Return 1 if we went up a level and 0 if we were already at the 842f015f126SDavid Howells * root. 843f015f126SDavid Howells */ 844bab77ebfSAl Viro int follow_up(struct path *path) 8451da177e4SLinus Torvalds { 8460714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 8470714a533SAl Viro struct mount *parent; 8481da177e4SLinus Torvalds struct dentry *mountpoint; 84999b7db7bSNick Piggin 850962830dfSAndi Kleen br_read_lock(&vfsmount_lock); 8510714a533SAl Viro parent = mnt->mnt_parent; 8523c0a6163SAl Viro if (parent == mnt) { 853962830dfSAndi Kleen br_read_unlock(&vfsmount_lock); 8541da177e4SLinus Torvalds return 0; 8551da177e4SLinus Torvalds } 8560714a533SAl Viro mntget(&parent->mnt); 857a73324daSAl Viro mountpoint = dget(mnt->mnt_mountpoint); 858962830dfSAndi Kleen br_read_unlock(&vfsmount_lock); 859bab77ebfSAl Viro dput(path->dentry); 860bab77ebfSAl Viro path->dentry = mountpoint; 861bab77ebfSAl Viro mntput(path->mnt); 8620714a533SAl Viro path->mnt = &parent->mnt; 8631da177e4SLinus Torvalds return 1; 8641da177e4SLinus Torvalds } 8651da177e4SLinus Torvalds 866b5c84bf6SNick Piggin /* 8679875cf80SDavid Howells * Perform an automount 8689875cf80SDavid Howells * - return -EISDIR to tell follow_managed() to stop and return the path we 8699875cf80SDavid Howells * were called with. 8701da177e4SLinus Torvalds */ 8719875cf80SDavid Howells static int follow_automount(struct path *path, unsigned flags, 8729875cf80SDavid Howells bool *need_mntput) 87331e6b01fSNick Piggin { 8749875cf80SDavid Howells struct vfsmount *mnt; 875ea5b778aSDavid Howells int err; 8769875cf80SDavid Howells 8779875cf80SDavid Howells if (!path->dentry->d_op || !path->dentry->d_op->d_automount) 8789875cf80SDavid Howells return -EREMOTE; 8799875cf80SDavid Howells 8800ec26fd0SMiklos Szeredi /* We don't want to mount if someone's just doing a stat - 8810ec26fd0SMiklos Szeredi * unless they're stat'ing a directory and appended a '/' to 8820ec26fd0SMiklos Szeredi * the name. 8830ec26fd0SMiklos Szeredi * 8840ec26fd0SMiklos Szeredi * We do, however, want to mount if someone wants to open or 8855a30d8a2SDavid Howells * create a file of any type under the mountpoint, wants to 8860ec26fd0SMiklos Szeredi * traverse through the mountpoint or wants to open the 8870ec26fd0SMiklos Szeredi * mounted directory. Also, autofs may mark negative dentries 8880ec26fd0SMiklos Szeredi * as being automount points. These will need the attentions 8890ec26fd0SMiklos Szeredi * of the daemon to instantiate them before they can be used. 8905a30d8a2SDavid Howells */ 8915a30d8a2SDavid Howells if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | 892d94c177bSLinus Torvalds LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && 8935a30d8a2SDavid Howells path->dentry->d_inode) 8949875cf80SDavid Howells return -EISDIR; 8950ec26fd0SMiklos Szeredi 8969875cf80SDavid Howells current->total_link_count++; 8979875cf80SDavid Howells if (current->total_link_count >= 40) 8989875cf80SDavid Howells return -ELOOP; 8999875cf80SDavid Howells 9009875cf80SDavid Howells mnt = path->dentry->d_op->d_automount(path); 9019875cf80SDavid Howells if (IS_ERR(mnt)) { 9029875cf80SDavid Howells /* 9039875cf80SDavid Howells * The filesystem is allowed to return -EISDIR here to indicate 9049875cf80SDavid Howells * it doesn't want to automount. For instance, autofs would do 9059875cf80SDavid Howells * this so that its userspace daemon can mount on this dentry. 9069875cf80SDavid Howells * 9079875cf80SDavid Howells * However, we can only permit this if it's a terminal point in 9089875cf80SDavid Howells * the path being looked up; if it wasn't then the remainder of 9099875cf80SDavid Howells * the path is inaccessible and we should say so. 9109875cf80SDavid Howells */ 91149084c3bSAl Viro if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT)) 9129875cf80SDavid Howells return -EREMOTE; 9139875cf80SDavid Howells return PTR_ERR(mnt); 91431e6b01fSNick Piggin } 915ea5b778aSDavid Howells 9169875cf80SDavid Howells if (!mnt) /* mount collision */ 9179875cf80SDavid Howells return 0; 9189875cf80SDavid Howells 9198aef1884SAl Viro if (!*need_mntput) { 9208aef1884SAl Viro /* lock_mount() may release path->mnt on error */ 9218aef1884SAl Viro mntget(path->mnt); 9228aef1884SAl Viro *need_mntput = true; 9238aef1884SAl Viro } 92419a167afSAl Viro err = finish_automount(mnt, path); 925ea5b778aSDavid Howells 926ea5b778aSDavid Howells switch (err) { 927ea5b778aSDavid Howells case -EBUSY: 928ea5b778aSDavid Howells /* Someone else made a mount here whilst we were busy */ 92919a167afSAl Viro return 0; 930ea5b778aSDavid Howells case 0: 9318aef1884SAl Viro path_put(path); 9329875cf80SDavid Howells path->mnt = mnt; 9339875cf80SDavid Howells path->dentry = dget(mnt->mnt_root); 9349875cf80SDavid Howells return 0; 93519a167afSAl Viro default: 93619a167afSAl Viro return err; 9379875cf80SDavid Howells } 93819a167afSAl Viro 939ea5b778aSDavid Howells } 9409875cf80SDavid Howells 9419875cf80SDavid Howells /* 9429875cf80SDavid Howells * Handle a dentry that is managed in some way. 943cc53ce53SDavid Howells * - Flagged for transit management (autofs) 9449875cf80SDavid Howells * - Flagged as mountpoint 9459875cf80SDavid Howells * - Flagged as automount point 9469875cf80SDavid Howells * 9479875cf80SDavid Howells * This may only be called in refwalk mode. 9489875cf80SDavid Howells * 9499875cf80SDavid Howells * Serialization is taken care of in namespace.c 9509875cf80SDavid Howells */ 9519875cf80SDavid Howells static int follow_managed(struct path *path, unsigned flags) 9529875cf80SDavid Howells { 9538aef1884SAl Viro struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */ 9549875cf80SDavid Howells unsigned managed; 9559875cf80SDavid Howells bool need_mntput = false; 9568aef1884SAl Viro int ret = 0; 9579875cf80SDavid Howells 9589875cf80SDavid Howells /* Given that we're not holding a lock here, we retain the value in a 9599875cf80SDavid Howells * local variable for each dentry as we look at it so that we don't see 9609875cf80SDavid Howells * the components of that value change under us */ 9619875cf80SDavid Howells while (managed = ACCESS_ONCE(path->dentry->d_flags), 9629875cf80SDavid Howells managed &= DCACHE_MANAGED_DENTRY, 9639875cf80SDavid Howells unlikely(managed != 0)) { 964cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 965cc53ce53SDavid Howells * being held. */ 966cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 967cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 968cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 9691aed3e42SAl Viro ret = path->dentry->d_op->d_manage(path->dentry, false); 970cc53ce53SDavid Howells if (ret < 0) 9718aef1884SAl Viro break; 972cc53ce53SDavid Howells } 973cc53ce53SDavid Howells 9749875cf80SDavid Howells /* Transit to a mounted filesystem. */ 9759875cf80SDavid Howells if (managed & DCACHE_MOUNTED) { 9769875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 9779875cf80SDavid Howells if (mounted) { 9789875cf80SDavid Howells dput(path->dentry); 9799875cf80SDavid Howells if (need_mntput) 980463ffb2eSAl Viro mntput(path->mnt); 981463ffb2eSAl Viro path->mnt = mounted; 982463ffb2eSAl Viro path->dentry = dget(mounted->mnt_root); 9839875cf80SDavid Howells need_mntput = true; 9849875cf80SDavid Howells continue; 985463ffb2eSAl Viro } 986463ffb2eSAl Viro 9879875cf80SDavid Howells /* Something is mounted on this dentry in another 9889875cf80SDavid Howells * namespace and/or whatever was mounted there in this 9899875cf80SDavid Howells * namespace got unmounted before we managed to get the 9909875cf80SDavid Howells * vfsmount_lock */ 9911da177e4SLinus Torvalds } 9929875cf80SDavid Howells 9939875cf80SDavid Howells /* Handle an automount point */ 9949875cf80SDavid Howells if (managed & DCACHE_NEED_AUTOMOUNT) { 9959875cf80SDavid Howells ret = follow_automount(path, flags, &need_mntput); 9969875cf80SDavid Howells if (ret < 0) 9978aef1884SAl Viro break; 9989875cf80SDavid Howells continue; 9999875cf80SDavid Howells } 10009875cf80SDavid Howells 10019875cf80SDavid Howells /* We didn't change the current path point */ 10029875cf80SDavid Howells break; 10039875cf80SDavid Howells } 10048aef1884SAl Viro 10058aef1884SAl Viro if (need_mntput && path->mnt == mnt) 10068aef1884SAl Viro mntput(path->mnt); 10078aef1884SAl Viro if (ret == -EISDIR) 10088aef1884SAl Viro ret = 0; 1009a3fbbde7SAl Viro return ret < 0 ? ret : need_mntput; 10101da177e4SLinus Torvalds } 10111da177e4SLinus Torvalds 1012cc53ce53SDavid Howells int follow_down_one(struct path *path) 10131da177e4SLinus Torvalds { 10141da177e4SLinus Torvalds struct vfsmount *mounted; 10151da177e4SLinus Torvalds 10161c755af4SAl Viro mounted = lookup_mnt(path); 10171da177e4SLinus Torvalds if (mounted) { 10189393bd07SAl Viro dput(path->dentry); 10199393bd07SAl Viro mntput(path->mnt); 10209393bd07SAl Viro path->mnt = mounted; 10219393bd07SAl Viro path->dentry = dget(mounted->mnt_root); 10221da177e4SLinus Torvalds return 1; 10231da177e4SLinus Torvalds } 10241da177e4SLinus Torvalds return 0; 10251da177e4SLinus Torvalds } 10261da177e4SLinus Torvalds 102762a7375eSIan Kent static inline bool managed_dentry_might_block(struct dentry *dentry) 102862a7375eSIan Kent { 102962a7375eSIan Kent return (dentry->d_flags & DCACHE_MANAGE_TRANSIT && 103062a7375eSIan Kent dentry->d_op->d_manage(dentry, true) < 0); 103162a7375eSIan Kent } 103262a7375eSIan Kent 10339875cf80SDavid Howells /* 1034287548e4SAl Viro * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if 1035287548e4SAl Viro * we meet a managed dentry that would need blocking. 10369875cf80SDavid Howells */ 10379875cf80SDavid Howells static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, 1038287548e4SAl Viro struct inode **inode) 10399875cf80SDavid Howells { 104062a7375eSIan Kent for (;;) { 1041c7105365SAl Viro struct mount *mounted; 104262a7375eSIan Kent /* 104362a7375eSIan Kent * Don't forget we might have a non-mountpoint managed dentry 104462a7375eSIan Kent * that wants to block transit. 104562a7375eSIan Kent */ 1046287548e4SAl Viro if (unlikely(managed_dentry_might_block(path->dentry))) 1047ab90911fSDavid Howells return false; 104862a7375eSIan Kent 104962a7375eSIan Kent if (!d_mountpoint(path->dentry)) 105062a7375eSIan Kent break; 105162a7375eSIan Kent 10529875cf80SDavid Howells mounted = __lookup_mnt(path->mnt, path->dentry, 1); 10539875cf80SDavid Howells if (!mounted) 10549875cf80SDavid Howells break; 1055c7105365SAl Viro path->mnt = &mounted->mnt; 1056c7105365SAl Viro path->dentry = mounted->mnt.mnt_root; 1057a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 10589875cf80SDavid Howells nd->seq = read_seqcount_begin(&path->dentry->d_seq); 105959430262SLinus Torvalds /* 106059430262SLinus Torvalds * Update the inode too. We don't need to re-check the 106159430262SLinus Torvalds * dentry sequence number here after this d_inode read, 106259430262SLinus Torvalds * because a mount-point is always pinned. 106359430262SLinus Torvalds */ 106459430262SLinus Torvalds *inode = path->dentry->d_inode; 10659875cf80SDavid Howells } 10669875cf80SDavid Howells return true; 10679875cf80SDavid Howells } 10689875cf80SDavid Howells 1069dea39376SAl Viro static void follow_mount_rcu(struct nameidata *nd) 1070287548e4SAl Viro { 1071dea39376SAl Viro while (d_mountpoint(nd->path.dentry)) { 1072c7105365SAl Viro struct mount *mounted; 1073dea39376SAl Viro mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1); 1074287548e4SAl Viro if (!mounted) 1075287548e4SAl Viro break; 1076c7105365SAl Viro nd->path.mnt = &mounted->mnt; 1077c7105365SAl Viro nd->path.dentry = mounted->mnt.mnt_root; 1078dea39376SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 1079287548e4SAl Viro } 1080287548e4SAl Viro } 1081287548e4SAl Viro 108231e6b01fSNick Piggin static int follow_dotdot_rcu(struct nameidata *nd) 108331e6b01fSNick Piggin { 108431e6b01fSNick Piggin set_root_rcu(nd); 108531e6b01fSNick Piggin 108631e6b01fSNick Piggin while (1) { 108731e6b01fSNick Piggin if (nd->path.dentry == nd->root.dentry && 108831e6b01fSNick Piggin nd->path.mnt == nd->root.mnt) { 108931e6b01fSNick Piggin break; 109031e6b01fSNick Piggin } 109131e6b01fSNick Piggin if (nd->path.dentry != nd->path.mnt->mnt_root) { 109231e6b01fSNick Piggin struct dentry *old = nd->path.dentry; 109331e6b01fSNick Piggin struct dentry *parent = old->d_parent; 109431e6b01fSNick Piggin unsigned seq; 109531e6b01fSNick Piggin 109631e6b01fSNick Piggin seq = read_seqcount_begin(&parent->d_seq); 109731e6b01fSNick Piggin if (read_seqcount_retry(&old->d_seq, nd->seq)) 1098ef7562d5SAl Viro goto failed; 109931e6b01fSNick Piggin nd->path.dentry = parent; 110031e6b01fSNick Piggin nd->seq = seq; 110131e6b01fSNick Piggin break; 110231e6b01fSNick Piggin } 110331e6b01fSNick Piggin if (!follow_up_rcu(&nd->path)) 110431e6b01fSNick Piggin break; 110531e6b01fSNick Piggin nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 110631e6b01fSNick Piggin } 1107dea39376SAl Viro follow_mount_rcu(nd); 1108dea39376SAl Viro nd->inode = nd->path.dentry->d_inode; 110931e6b01fSNick Piggin return 0; 1110ef7562d5SAl Viro 1111ef7562d5SAl Viro failed: 1112ef7562d5SAl Viro nd->flags &= ~LOOKUP_RCU; 11135b6ca027SAl Viro if (!(nd->flags & LOOKUP_ROOT)) 1114ef7562d5SAl Viro nd->root.mnt = NULL; 111532a7991bSAl Viro unlock_rcu_walk(); 1116ef7562d5SAl Viro return -ECHILD; 111731e6b01fSNick Piggin } 111831e6b01fSNick Piggin 11199875cf80SDavid Howells /* 1120cc53ce53SDavid Howells * Follow down to the covering mount currently visible to userspace. At each 1121cc53ce53SDavid Howells * point, the filesystem owning that dentry may be queried as to whether the 1122cc53ce53SDavid Howells * caller is permitted to proceed or not. 1123cc53ce53SDavid Howells */ 11247cc90cc3SAl Viro int follow_down(struct path *path) 1125cc53ce53SDavid Howells { 1126cc53ce53SDavid Howells unsigned managed; 1127cc53ce53SDavid Howells int ret; 1128cc53ce53SDavid Howells 1129cc53ce53SDavid Howells while (managed = ACCESS_ONCE(path->dentry->d_flags), 1130cc53ce53SDavid Howells unlikely(managed & DCACHE_MANAGED_DENTRY)) { 1131cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1132cc53ce53SDavid Howells * being held. 1133cc53ce53SDavid Howells * 1134cc53ce53SDavid Howells * We indicate to the filesystem if someone is trying to mount 1135cc53ce53SDavid Howells * something here. This gives autofs the chance to deny anyone 1136cc53ce53SDavid Howells * other than its daemon the right to mount on its 1137cc53ce53SDavid Howells * superstructure. 1138cc53ce53SDavid Howells * 1139cc53ce53SDavid Howells * The filesystem may sleep at this point. 1140cc53ce53SDavid Howells */ 1141cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 1142cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1143cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 1144ab90911fSDavid Howells ret = path->dentry->d_op->d_manage( 11451aed3e42SAl Viro path->dentry, false); 1146cc53ce53SDavid Howells if (ret < 0) 1147cc53ce53SDavid Howells return ret == -EISDIR ? 0 : ret; 1148cc53ce53SDavid Howells } 1149cc53ce53SDavid Howells 1150cc53ce53SDavid Howells /* Transit to a mounted filesystem. */ 1151cc53ce53SDavid Howells if (managed & DCACHE_MOUNTED) { 1152cc53ce53SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 1153cc53ce53SDavid Howells if (!mounted) 1154cc53ce53SDavid Howells break; 1155cc53ce53SDavid Howells dput(path->dentry); 1156cc53ce53SDavid Howells mntput(path->mnt); 1157cc53ce53SDavid Howells path->mnt = mounted; 1158cc53ce53SDavid Howells path->dentry = dget(mounted->mnt_root); 1159cc53ce53SDavid Howells continue; 1160cc53ce53SDavid Howells } 1161cc53ce53SDavid Howells 1162cc53ce53SDavid Howells /* Don't handle automount points here */ 1163cc53ce53SDavid Howells break; 1164cc53ce53SDavid Howells } 1165cc53ce53SDavid Howells return 0; 1166cc53ce53SDavid Howells } 1167cc53ce53SDavid Howells 1168cc53ce53SDavid Howells /* 11699875cf80SDavid Howells * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() 11709875cf80SDavid Howells */ 11719875cf80SDavid Howells static void follow_mount(struct path *path) 11729875cf80SDavid Howells { 11739875cf80SDavid Howells while (d_mountpoint(path->dentry)) { 11749875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 11759875cf80SDavid Howells if (!mounted) 11769875cf80SDavid Howells break; 11779875cf80SDavid Howells dput(path->dentry); 11789875cf80SDavid Howells mntput(path->mnt); 11799875cf80SDavid Howells path->mnt = mounted; 11809875cf80SDavid Howells path->dentry = dget(mounted->mnt_root); 11819875cf80SDavid Howells } 11829875cf80SDavid Howells } 11839875cf80SDavid Howells 118431e6b01fSNick Piggin static void follow_dotdot(struct nameidata *nd) 11851da177e4SLinus Torvalds { 11862a737871SAl Viro set_root(nd); 1187e518ddb7SAndreas Mohr 11881da177e4SLinus Torvalds while(1) { 11894ac91378SJan Blunck struct dentry *old = nd->path.dentry; 11901da177e4SLinus Torvalds 11912a737871SAl Viro if (nd->path.dentry == nd->root.dentry && 11922a737871SAl Viro nd->path.mnt == nd->root.mnt) { 11931da177e4SLinus Torvalds break; 11941da177e4SLinus Torvalds } 11954ac91378SJan Blunck if (nd->path.dentry != nd->path.mnt->mnt_root) { 11963088dd70SAl Viro /* rare case of legitimate dget_parent()... */ 11973088dd70SAl Viro nd->path.dentry = dget_parent(nd->path.dentry); 11981da177e4SLinus Torvalds dput(old); 11991da177e4SLinus Torvalds break; 12001da177e4SLinus Torvalds } 12013088dd70SAl Viro if (!follow_up(&nd->path)) 12021da177e4SLinus Torvalds break; 12031da177e4SLinus Torvalds } 120479ed0226SAl Viro follow_mount(&nd->path); 120531e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 12061da177e4SLinus Torvalds } 12071da177e4SLinus Torvalds 12081da177e4SLinus Torvalds /* 1209bad61189SMiklos Szeredi * This looks up the name in dcache, possibly revalidates the old dentry and 1210bad61189SMiklos Szeredi * allocates a new one if not found or not valid. In the need_lookup argument 1211bad61189SMiklos Szeredi * returns whether i_op->lookup is necessary. 1212bad61189SMiklos Szeredi * 1213bad61189SMiklos Szeredi * dir->d_inode->i_mutex must be held 1214baa03890SNick Piggin */ 1215bad61189SMiklos Szeredi static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir, 1216201f956eSAl Viro unsigned int flags, bool *need_lookup) 1217baa03890SNick Piggin { 1218baa03890SNick Piggin struct dentry *dentry; 1219bad61189SMiklos Szeredi int error; 1220baa03890SNick Piggin 1221bad61189SMiklos Szeredi *need_lookup = false; 1222bad61189SMiklos Szeredi dentry = d_lookup(dir, name); 1223bad61189SMiklos Szeredi if (dentry) { 1224bad61189SMiklos Szeredi if (d_need_lookup(dentry)) { 1225bad61189SMiklos Szeredi *need_lookup = true; 1226bad61189SMiklos Szeredi } else if (dentry->d_flags & DCACHE_OP_REVALIDATE) { 1227201f956eSAl Viro error = d_revalidate(dentry, flags); 1228bad61189SMiklos Szeredi if (unlikely(error <= 0)) { 1229bad61189SMiklos Szeredi if (error < 0) { 1230bad61189SMiklos Szeredi dput(dentry); 1231bad61189SMiklos Szeredi return ERR_PTR(error); 1232bad61189SMiklos Szeredi } else if (!d_invalidate(dentry)) { 1233bad61189SMiklos Szeredi dput(dentry); 1234bad61189SMiklos Szeredi dentry = NULL; 1235bad61189SMiklos Szeredi } 1236bad61189SMiklos Szeredi } 1237bad61189SMiklos Szeredi } 1238bad61189SMiklos Szeredi } 1239baa03890SNick Piggin 1240bad61189SMiklos Szeredi if (!dentry) { 1241bad61189SMiklos Szeredi dentry = d_alloc(dir, name); 1242baa03890SNick Piggin if (unlikely(!dentry)) 1243baa03890SNick Piggin return ERR_PTR(-ENOMEM); 1244baa03890SNick Piggin 1245bad61189SMiklos Szeredi *need_lookup = true; 1246baa03890SNick Piggin } 1247baa03890SNick Piggin return dentry; 1248baa03890SNick Piggin } 1249baa03890SNick Piggin 1250baa03890SNick Piggin /* 1251bad61189SMiklos Szeredi * Call i_op->lookup on the dentry. The dentry must be negative but may be 1252bad61189SMiklos Szeredi * hashed if it was pouplated with DCACHE_NEED_LOOKUP. 1253bad61189SMiklos Szeredi * 1254bad61189SMiklos Szeredi * dir->d_inode->i_mutex must be held 125544396f4bSJosef Bacik */ 1256bad61189SMiklos Szeredi static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry, 125772bd866aSAl Viro unsigned int flags) 125844396f4bSJosef Bacik { 125944396f4bSJosef Bacik struct dentry *old; 126044396f4bSJosef Bacik 126144396f4bSJosef Bacik /* Don't create child dentry for a dead directory. */ 1262bad61189SMiklos Szeredi if (unlikely(IS_DEADDIR(dir))) { 1263e188dc02SMiklos Szeredi dput(dentry); 126444396f4bSJosef Bacik return ERR_PTR(-ENOENT); 1265e188dc02SMiklos Szeredi } 126644396f4bSJosef Bacik 126772bd866aSAl Viro old = dir->i_op->lookup(dir, dentry, flags); 126844396f4bSJosef Bacik if (unlikely(old)) { 126944396f4bSJosef Bacik dput(dentry); 127044396f4bSJosef Bacik dentry = old; 127144396f4bSJosef Bacik } 127244396f4bSJosef Bacik return dentry; 127344396f4bSJosef Bacik } 127444396f4bSJosef Bacik 1275a3255546SAl Viro static struct dentry *__lookup_hash(struct qstr *name, 127672bd866aSAl Viro struct dentry *base, unsigned int flags) 1277a3255546SAl Viro { 1278bad61189SMiklos Szeredi bool need_lookup; 1279a3255546SAl Viro struct dentry *dentry; 1280a3255546SAl Viro 128172bd866aSAl Viro dentry = lookup_dcache(name, base, flags, &need_lookup); 1282bad61189SMiklos Szeredi if (!need_lookup) 1283a3255546SAl Viro return dentry; 1284bad61189SMiklos Szeredi 128572bd866aSAl Viro return lookup_real(base->d_inode, dentry, flags); 1286a3255546SAl Viro } 1287a3255546SAl Viro 128844396f4bSJosef Bacik /* 12891da177e4SLinus Torvalds * It's more convoluted than I'd like it to be, but... it's still fairly 12901da177e4SLinus Torvalds * small and for now I'd prefer to have fast path as straight as possible. 12911da177e4SLinus Torvalds * It _is_ time-critical. 12921da177e4SLinus Torvalds */ 1293697f514dSMiklos Szeredi static int lookup_fast(struct nameidata *nd, struct qstr *name, 129431e6b01fSNick Piggin struct path *path, struct inode **inode) 12951da177e4SLinus Torvalds { 12964ac91378SJan Blunck struct vfsmount *mnt = nd->path.mnt; 129731e6b01fSNick Piggin struct dentry *dentry, *parent = nd->path.dentry; 12985a18fff2SAl Viro int need_reval = 1; 12995a18fff2SAl Viro int status = 1; 13009875cf80SDavid Howells int err; 13019875cf80SDavid Howells 13023cac260aSAl Viro /* 1303b04f784eSNick Piggin * Rename seqlock is not required here because in the off chance 1304b04f784eSNick Piggin * of a false negative due to a concurrent rename, we're going to 1305b04f784eSNick Piggin * do the non-racy lookup, below. 1306b04f784eSNick Piggin */ 130731e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) { 130831e6b01fSNick Piggin unsigned seq; 130912f8ad4bSLinus Torvalds dentry = __d_lookup_rcu(parent, name, &seq, nd->inode); 13105a18fff2SAl Viro if (!dentry) 13115a18fff2SAl Viro goto unlazy; 13125a18fff2SAl Viro 131312f8ad4bSLinus Torvalds /* 131412f8ad4bSLinus Torvalds * This sequence count validates that the inode matches 131512f8ad4bSLinus Torvalds * the dentry name information from lookup. 131612f8ad4bSLinus Torvalds */ 131712f8ad4bSLinus Torvalds *inode = dentry->d_inode; 131812f8ad4bSLinus Torvalds if (read_seqcount_retry(&dentry->d_seq, seq)) 131912f8ad4bSLinus Torvalds return -ECHILD; 132012f8ad4bSLinus Torvalds 132112f8ad4bSLinus Torvalds /* 132212f8ad4bSLinus Torvalds * This sequence count validates that the parent had no 132312f8ad4bSLinus Torvalds * changes while we did the lookup of the dentry above. 132412f8ad4bSLinus Torvalds * 132512f8ad4bSLinus Torvalds * The memory barrier in read_seqcount_begin of child is 132612f8ad4bSLinus Torvalds * enough, we can use __read_seqcount_retry here. 132712f8ad4bSLinus Torvalds */ 132831e6b01fSNick Piggin if (__read_seqcount_retry(&parent->d_seq, nd->seq)) 132931e6b01fSNick Piggin return -ECHILD; 133031e6b01fSNick Piggin nd->seq = seq; 13315a18fff2SAl Viro 1332fa4ee159SMiklos Szeredi if (unlikely(d_need_lookup(dentry))) 1333fa4ee159SMiklos Szeredi goto unlazy; 133424643087SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) { 13354ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 13365a18fff2SAl Viro if (unlikely(status <= 0)) { 13375a18fff2SAl Viro if (status != -ECHILD) 13385a18fff2SAl Viro need_reval = 0; 13395a18fff2SAl Viro goto unlazy; 13405a18fff2SAl Viro } 134124643087SAl Viro } 134231e6b01fSNick Piggin path->mnt = mnt; 134331e6b01fSNick Piggin path->dentry = dentry; 1344d6e9bd25SAl Viro if (unlikely(!__follow_mount_rcu(nd, path, inode))) 1345d6e9bd25SAl Viro goto unlazy; 1346d6e9bd25SAl Viro if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT)) 1347d6e9bd25SAl Viro goto unlazy; 13489875cf80SDavid Howells return 0; 13495a18fff2SAl Viro unlazy: 135019660af7SAl Viro if (unlazy_walk(nd, dentry)) 13515a18fff2SAl Viro return -ECHILD; 13525a18fff2SAl Viro } else { 135331e6b01fSNick Piggin dentry = __d_lookup(parent, name); 135424643087SAl Viro } 13555a18fff2SAl Viro 135681e6f520SAl Viro if (unlikely(!dentry)) 135781e6f520SAl Viro goto need_lookup; 13585a18fff2SAl Viro 135981e6f520SAl Viro if (unlikely(d_need_lookup(dentry))) { 136081e6f520SAl Viro dput(dentry); 136181e6f520SAl Viro goto need_lookup; 13625a18fff2SAl Viro } 136381e6f520SAl Viro 13645a18fff2SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval) 13654ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 13665a18fff2SAl Viro if (unlikely(status <= 0)) { 13675a18fff2SAl Viro if (status < 0) { 13685a18fff2SAl Viro dput(dentry); 13695a18fff2SAl Viro return status; 13705a18fff2SAl Viro } 13715a18fff2SAl Viro if (!d_invalidate(dentry)) { 13725a18fff2SAl Viro dput(dentry); 137381e6f520SAl Viro goto need_lookup; 13745a18fff2SAl Viro } 13755a18fff2SAl Viro } 1376697f514dSMiklos Szeredi 13771da177e4SLinus Torvalds path->mnt = mnt; 13781da177e4SLinus Torvalds path->dentry = dentry; 13799875cf80SDavid Howells err = follow_managed(path, nd->flags); 138089312214SIan Kent if (unlikely(err < 0)) { 138189312214SIan Kent path_put_conditional(path, nd); 13829875cf80SDavid Howells return err; 138389312214SIan Kent } 1384a3fbbde7SAl Viro if (err) 1385a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 138631e6b01fSNick Piggin *inode = path->dentry->d_inode; 13871da177e4SLinus Torvalds return 0; 138881e6f520SAl Viro 138981e6f520SAl Viro need_lookup: 1390697f514dSMiklos Szeredi return 1; 1391697f514dSMiklos Szeredi } 1392697f514dSMiklos Szeredi 1393697f514dSMiklos Szeredi /* Fast lookup failed, do it the slow way */ 1394697f514dSMiklos Szeredi static int lookup_slow(struct nameidata *nd, struct qstr *name, 1395697f514dSMiklos Szeredi struct path *path) 1396697f514dSMiklos Szeredi { 1397697f514dSMiklos Szeredi struct dentry *dentry, *parent; 1398697f514dSMiklos Szeredi int err; 1399697f514dSMiklos Szeredi 1400697f514dSMiklos Szeredi parent = nd->path.dentry; 140181e6f520SAl Viro BUG_ON(nd->inode != parent->d_inode); 140281e6f520SAl Viro 140381e6f520SAl Viro mutex_lock(&parent->d_inode->i_mutex); 140472bd866aSAl Viro dentry = __lookup_hash(name, parent, nd->flags); 140581e6f520SAl Viro mutex_unlock(&parent->d_inode->i_mutex); 140681e6f520SAl Viro if (IS_ERR(dentry)) 140781e6f520SAl Viro return PTR_ERR(dentry); 1408697f514dSMiklos Szeredi path->mnt = nd->path.mnt; 1409697f514dSMiklos Szeredi path->dentry = dentry; 1410697f514dSMiklos Szeredi err = follow_managed(path, nd->flags); 1411697f514dSMiklos Szeredi if (unlikely(err < 0)) { 1412697f514dSMiklos Szeredi path_put_conditional(path, nd); 1413697f514dSMiklos Szeredi return err; 1414697f514dSMiklos Szeredi } 1415697f514dSMiklos Szeredi if (err) 1416697f514dSMiklos Szeredi nd->flags |= LOOKUP_JUMPED; 1417697f514dSMiklos Szeredi return 0; 14181da177e4SLinus Torvalds } 14191da177e4SLinus Torvalds 142052094c8aSAl Viro static inline int may_lookup(struct nameidata *nd) 142152094c8aSAl Viro { 142252094c8aSAl Viro if (nd->flags & LOOKUP_RCU) { 14234ad5abb3SAl Viro int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); 142452094c8aSAl Viro if (err != -ECHILD) 142552094c8aSAl Viro return err; 142619660af7SAl Viro if (unlazy_walk(nd, NULL)) 142752094c8aSAl Viro return -ECHILD; 142852094c8aSAl Viro } 14294ad5abb3SAl Viro return inode_permission(nd->inode, MAY_EXEC); 143052094c8aSAl Viro } 143152094c8aSAl Viro 14329856fa1bSAl Viro static inline int handle_dots(struct nameidata *nd, int type) 14339856fa1bSAl Viro { 14349856fa1bSAl Viro if (type == LAST_DOTDOT) { 14359856fa1bSAl Viro if (nd->flags & LOOKUP_RCU) { 14369856fa1bSAl Viro if (follow_dotdot_rcu(nd)) 14379856fa1bSAl Viro return -ECHILD; 14389856fa1bSAl Viro } else 14399856fa1bSAl Viro follow_dotdot(nd); 14409856fa1bSAl Viro } 14419856fa1bSAl Viro return 0; 14429856fa1bSAl Viro } 14439856fa1bSAl Viro 1444951361f9SAl Viro static void terminate_walk(struct nameidata *nd) 1445951361f9SAl Viro { 1446951361f9SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1447951361f9SAl Viro path_put(&nd->path); 1448951361f9SAl Viro } else { 1449951361f9SAl Viro nd->flags &= ~LOOKUP_RCU; 14505b6ca027SAl Viro if (!(nd->flags & LOOKUP_ROOT)) 1451951361f9SAl Viro nd->root.mnt = NULL; 145232a7991bSAl Viro unlock_rcu_walk(); 1453951361f9SAl Viro } 1454951361f9SAl Viro } 1455951361f9SAl Viro 14563ddcd056SLinus Torvalds /* 14573ddcd056SLinus Torvalds * Do we need to follow links? We _really_ want to be able 14583ddcd056SLinus Torvalds * to do this check without having to look at inode->i_op, 14593ddcd056SLinus Torvalds * so we keep a cache of "no, this doesn't need follow_link" 14603ddcd056SLinus Torvalds * for the common case. 14613ddcd056SLinus Torvalds */ 14627813b94aSLinus Torvalds static inline int should_follow_link(struct inode *inode, int follow) 14633ddcd056SLinus Torvalds { 14643ddcd056SLinus Torvalds if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) { 14653ddcd056SLinus Torvalds if (likely(inode->i_op->follow_link)) 14663ddcd056SLinus Torvalds return follow; 14673ddcd056SLinus Torvalds 14683ddcd056SLinus Torvalds /* This gets set once for the inode lifetime */ 14693ddcd056SLinus Torvalds spin_lock(&inode->i_lock); 14703ddcd056SLinus Torvalds inode->i_opflags |= IOP_NOFOLLOW; 14713ddcd056SLinus Torvalds spin_unlock(&inode->i_lock); 14723ddcd056SLinus Torvalds } 14733ddcd056SLinus Torvalds return 0; 14743ddcd056SLinus Torvalds } 14753ddcd056SLinus Torvalds 1476ce57dfc1SAl Viro static inline int walk_component(struct nameidata *nd, struct path *path, 1477ce57dfc1SAl Viro struct qstr *name, int type, int follow) 1478ce57dfc1SAl Viro { 1479ce57dfc1SAl Viro struct inode *inode; 1480ce57dfc1SAl Viro int err; 1481ce57dfc1SAl Viro /* 1482ce57dfc1SAl Viro * "." and ".." are special - ".." especially so because it has 1483ce57dfc1SAl Viro * to be able to know about the current root directory and 1484ce57dfc1SAl Viro * parent relationships. 1485ce57dfc1SAl Viro */ 1486ce57dfc1SAl Viro if (unlikely(type != LAST_NORM)) 1487ce57dfc1SAl Viro return handle_dots(nd, type); 1488697f514dSMiklos Szeredi err = lookup_fast(nd, name, path, &inode); 1489ce57dfc1SAl Viro if (unlikely(err)) { 1490697f514dSMiklos Szeredi if (err < 0) 1491697f514dSMiklos Szeredi goto out_err; 1492697f514dSMiklos Szeredi 1493697f514dSMiklos Szeredi err = lookup_slow(nd, name, path); 1494697f514dSMiklos Szeredi if (err < 0) 1495697f514dSMiklos Szeredi goto out_err; 1496697f514dSMiklos Szeredi 1497697f514dSMiklos Szeredi inode = path->dentry->d_inode; 1498ce57dfc1SAl Viro } 1499697f514dSMiklos Szeredi err = -ENOENT; 1500697f514dSMiklos Szeredi if (!inode) 1501697f514dSMiklos Szeredi goto out_path_put; 1502697f514dSMiklos Szeredi 15037813b94aSLinus Torvalds if (should_follow_link(inode, follow)) { 150419660af7SAl Viro if (nd->flags & LOOKUP_RCU) { 150519660af7SAl Viro if (unlikely(unlazy_walk(nd, path->dentry))) { 1506697f514dSMiklos Szeredi err = -ECHILD; 1507697f514dSMiklos Szeredi goto out_err; 150819660af7SAl Viro } 150919660af7SAl Viro } 1510ce57dfc1SAl Viro BUG_ON(inode != path->dentry->d_inode); 1511ce57dfc1SAl Viro return 1; 1512ce57dfc1SAl Viro } 1513ce57dfc1SAl Viro path_to_nameidata(path, nd); 1514ce57dfc1SAl Viro nd->inode = inode; 1515ce57dfc1SAl Viro return 0; 1516697f514dSMiklos Szeredi 1517697f514dSMiklos Szeredi out_path_put: 1518697f514dSMiklos Szeredi path_to_nameidata(path, nd); 1519697f514dSMiklos Szeredi out_err: 1520697f514dSMiklos Szeredi terminate_walk(nd); 1521697f514dSMiklos Szeredi return err; 1522ce57dfc1SAl Viro } 1523ce57dfc1SAl Viro 15241da177e4SLinus Torvalds /* 1525b356379aSAl Viro * This limits recursive symlink follows to 8, while 1526b356379aSAl Viro * limiting consecutive symlinks to 40. 1527b356379aSAl Viro * 1528b356379aSAl Viro * Without that kind of total limit, nasty chains of consecutive 1529b356379aSAl Viro * symlinks can cause almost arbitrarily long lookups. 1530b356379aSAl Viro */ 1531b356379aSAl Viro static inline int nested_symlink(struct path *path, struct nameidata *nd) 1532b356379aSAl Viro { 1533b356379aSAl Viro int res; 1534b356379aSAl Viro 1535b356379aSAl Viro if (unlikely(current->link_count >= MAX_NESTED_LINKS)) { 1536b356379aSAl Viro path_put_conditional(path, nd); 1537b356379aSAl Viro path_put(&nd->path); 1538b356379aSAl Viro return -ELOOP; 1539b356379aSAl Viro } 15401a4022f8SErez Zadok BUG_ON(nd->depth >= MAX_NESTED_LINKS); 1541b356379aSAl Viro 1542b356379aSAl Viro nd->depth++; 1543b356379aSAl Viro current->link_count++; 1544b356379aSAl Viro 1545b356379aSAl Viro do { 1546b356379aSAl Viro struct path link = *path; 1547b356379aSAl Viro void *cookie; 1548574197e0SAl Viro 1549574197e0SAl Viro res = follow_link(&link, nd, &cookie); 15506d7b5aaeSAl Viro if (res) 15516d7b5aaeSAl Viro break; 1552b356379aSAl Viro res = walk_component(nd, path, &nd->last, 1553b356379aSAl Viro nd->last_type, LOOKUP_FOLLOW); 1554574197e0SAl Viro put_link(nd, &link, cookie); 1555b356379aSAl Viro } while (res > 0); 1556b356379aSAl Viro 1557b356379aSAl Viro current->link_count--; 1558b356379aSAl Viro nd->depth--; 1559b356379aSAl Viro return res; 1560b356379aSAl Viro } 1561b356379aSAl Viro 1562b356379aSAl Viro /* 15633ddcd056SLinus Torvalds * We really don't want to look at inode->i_op->lookup 15643ddcd056SLinus Torvalds * when we don't have to. So we keep a cache bit in 15653ddcd056SLinus Torvalds * the inode ->i_opflags field that says "yes, we can 15663ddcd056SLinus Torvalds * do lookup on this inode". 15673ddcd056SLinus Torvalds */ 15683ddcd056SLinus Torvalds static inline int can_lookup(struct inode *inode) 15693ddcd056SLinus Torvalds { 15703ddcd056SLinus Torvalds if (likely(inode->i_opflags & IOP_LOOKUP)) 15713ddcd056SLinus Torvalds return 1; 15723ddcd056SLinus Torvalds if (likely(!inode->i_op->lookup)) 15733ddcd056SLinus Torvalds return 0; 15743ddcd056SLinus Torvalds 15753ddcd056SLinus Torvalds /* We do this once for the lifetime of the inode */ 15763ddcd056SLinus Torvalds spin_lock(&inode->i_lock); 15773ddcd056SLinus Torvalds inode->i_opflags |= IOP_LOOKUP; 15783ddcd056SLinus Torvalds spin_unlock(&inode->i_lock); 15793ddcd056SLinus Torvalds return 1; 15803ddcd056SLinus Torvalds } 15813ddcd056SLinus Torvalds 1582bfcfaa77SLinus Torvalds /* 1583bfcfaa77SLinus Torvalds * We can do the critical dentry name comparison and hashing 1584bfcfaa77SLinus Torvalds * operations one word at a time, but we are limited to: 1585bfcfaa77SLinus Torvalds * 1586bfcfaa77SLinus Torvalds * - Architectures with fast unaligned word accesses. We could 1587bfcfaa77SLinus Torvalds * do a "get_unaligned()" if this helps and is sufficiently 1588bfcfaa77SLinus Torvalds * fast. 1589bfcfaa77SLinus Torvalds * 1590bfcfaa77SLinus Torvalds * - Little-endian machines (so that we can generate the mask 1591bfcfaa77SLinus Torvalds * of low bytes efficiently). Again, we *could* do a byte 1592bfcfaa77SLinus Torvalds * swapping load on big-endian architectures if that is not 1593bfcfaa77SLinus Torvalds * expensive enough to make the optimization worthless. 1594bfcfaa77SLinus Torvalds * 1595bfcfaa77SLinus Torvalds * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we 1596bfcfaa77SLinus Torvalds * do not trap on the (extremely unlikely) case of a page 1597bfcfaa77SLinus Torvalds * crossing operation. 1598bfcfaa77SLinus Torvalds * 1599bfcfaa77SLinus Torvalds * - Furthermore, we need an efficient 64-bit compile for the 1600bfcfaa77SLinus Torvalds * 64-bit case in order to generate the "number of bytes in 1601bfcfaa77SLinus Torvalds * the final mask". Again, that could be replaced with a 1602bfcfaa77SLinus Torvalds * efficient population count instruction or similar. 1603bfcfaa77SLinus Torvalds */ 1604bfcfaa77SLinus Torvalds #ifdef CONFIG_DCACHE_WORD_ACCESS 1605bfcfaa77SLinus Torvalds 1606f68e556eSLinus Torvalds #include <asm/word-at-a-time.h> 1607bfcfaa77SLinus Torvalds 1608f68e556eSLinus Torvalds #ifdef CONFIG_64BIT 1609bfcfaa77SLinus Torvalds 1610bfcfaa77SLinus Torvalds static inline unsigned int fold_hash(unsigned long hash) 1611bfcfaa77SLinus Torvalds { 1612bfcfaa77SLinus Torvalds hash += hash >> (8*sizeof(int)); 1613bfcfaa77SLinus Torvalds return hash; 1614bfcfaa77SLinus Torvalds } 1615bfcfaa77SLinus Torvalds 1616bfcfaa77SLinus Torvalds #else /* 32-bit case */ 1617bfcfaa77SLinus Torvalds 1618bfcfaa77SLinus Torvalds #define fold_hash(x) (x) 1619bfcfaa77SLinus Torvalds 1620bfcfaa77SLinus Torvalds #endif 1621bfcfaa77SLinus Torvalds 1622bfcfaa77SLinus Torvalds unsigned int full_name_hash(const unsigned char *name, unsigned int len) 1623bfcfaa77SLinus Torvalds { 1624bfcfaa77SLinus Torvalds unsigned long a, mask; 1625bfcfaa77SLinus Torvalds unsigned long hash = 0; 1626bfcfaa77SLinus Torvalds 1627bfcfaa77SLinus Torvalds for (;;) { 1628e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name); 1629bfcfaa77SLinus Torvalds if (len < sizeof(unsigned long)) 1630bfcfaa77SLinus Torvalds break; 1631bfcfaa77SLinus Torvalds hash += a; 1632f132c5beSAl Viro hash *= 9; 1633bfcfaa77SLinus Torvalds name += sizeof(unsigned long); 1634bfcfaa77SLinus Torvalds len -= sizeof(unsigned long); 1635bfcfaa77SLinus Torvalds if (!len) 1636bfcfaa77SLinus Torvalds goto done; 1637bfcfaa77SLinus Torvalds } 1638bfcfaa77SLinus Torvalds mask = ~(~0ul << len*8); 1639bfcfaa77SLinus Torvalds hash += mask & a; 1640bfcfaa77SLinus Torvalds done: 1641bfcfaa77SLinus Torvalds return fold_hash(hash); 1642bfcfaa77SLinus Torvalds } 1643bfcfaa77SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 1644bfcfaa77SLinus Torvalds 1645bfcfaa77SLinus Torvalds /* 1646bfcfaa77SLinus Torvalds * Calculate the length and hash of the path component, and 1647bfcfaa77SLinus Torvalds * return the length of the component; 1648bfcfaa77SLinus Torvalds */ 1649bfcfaa77SLinus Torvalds static inline unsigned long hash_name(const char *name, unsigned int *hashp) 1650bfcfaa77SLinus Torvalds { 165136126f8fSLinus Torvalds unsigned long a, b, adata, bdata, mask, hash, len; 165236126f8fSLinus Torvalds const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 1653bfcfaa77SLinus Torvalds 1654bfcfaa77SLinus Torvalds hash = a = 0; 1655bfcfaa77SLinus Torvalds len = -sizeof(unsigned long); 1656bfcfaa77SLinus Torvalds do { 1657bfcfaa77SLinus Torvalds hash = (hash + a) * 9; 1658bfcfaa77SLinus Torvalds len += sizeof(unsigned long); 1659e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name+len); 166036126f8fSLinus Torvalds b = a ^ REPEAT_BYTE('/'); 166136126f8fSLinus Torvalds } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); 1662bfcfaa77SLinus Torvalds 166336126f8fSLinus Torvalds adata = prep_zero_mask(a, adata, &constants); 166436126f8fSLinus Torvalds bdata = prep_zero_mask(b, bdata, &constants); 166536126f8fSLinus Torvalds 166636126f8fSLinus Torvalds mask = create_zero_mask(adata | bdata); 166736126f8fSLinus Torvalds 166836126f8fSLinus Torvalds hash += a & zero_bytemask(mask); 1669bfcfaa77SLinus Torvalds *hashp = fold_hash(hash); 1670bfcfaa77SLinus Torvalds 167136126f8fSLinus Torvalds return len + find_zero(mask); 1672bfcfaa77SLinus Torvalds } 1673bfcfaa77SLinus Torvalds 1674bfcfaa77SLinus Torvalds #else 1675bfcfaa77SLinus Torvalds 16760145acc2SLinus Torvalds unsigned int full_name_hash(const unsigned char *name, unsigned int len) 16770145acc2SLinus Torvalds { 16780145acc2SLinus Torvalds unsigned long hash = init_name_hash(); 16790145acc2SLinus Torvalds while (len--) 16800145acc2SLinus Torvalds hash = partial_name_hash(*name++, hash); 16810145acc2SLinus Torvalds return end_name_hash(hash); 16820145acc2SLinus Torvalds } 1683ae942ae7SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 16840145acc2SLinus Torvalds 16853ddcd056SLinus Torvalds /* 1686200e9ef7SLinus Torvalds * We know there's a real path component here of at least 1687200e9ef7SLinus Torvalds * one character. 1688200e9ef7SLinus Torvalds */ 1689200e9ef7SLinus Torvalds static inline unsigned long hash_name(const char *name, unsigned int *hashp) 1690200e9ef7SLinus Torvalds { 1691200e9ef7SLinus Torvalds unsigned long hash = init_name_hash(); 1692200e9ef7SLinus Torvalds unsigned long len = 0, c; 1693200e9ef7SLinus Torvalds 1694200e9ef7SLinus Torvalds c = (unsigned char)*name; 1695200e9ef7SLinus Torvalds do { 1696200e9ef7SLinus Torvalds len++; 1697200e9ef7SLinus Torvalds hash = partial_name_hash(c, hash); 1698200e9ef7SLinus Torvalds c = (unsigned char)name[len]; 1699200e9ef7SLinus Torvalds } while (c && c != '/'); 1700200e9ef7SLinus Torvalds *hashp = end_name_hash(hash); 1701200e9ef7SLinus Torvalds return len; 1702200e9ef7SLinus Torvalds } 1703200e9ef7SLinus Torvalds 1704bfcfaa77SLinus Torvalds #endif 1705bfcfaa77SLinus Torvalds 1706200e9ef7SLinus Torvalds /* 17071da177e4SLinus Torvalds * Name resolution. 1708ea3834d9SPrasanna Meda * This is the basic name resolution function, turning a pathname into 1709ea3834d9SPrasanna Meda * the final dentry. We expect 'base' to be positive and a directory. 17101da177e4SLinus Torvalds * 1711ea3834d9SPrasanna Meda * Returns 0 and nd will have valid dentry and mnt on success. 1712ea3834d9SPrasanna Meda * Returns error and drops reference to input namei data on failure. 17131da177e4SLinus Torvalds */ 17146de88d72SAl Viro static int link_path_walk(const char *name, struct nameidata *nd) 17151da177e4SLinus Torvalds { 17161da177e4SLinus Torvalds struct path next; 17171da177e4SLinus Torvalds int err; 17181da177e4SLinus Torvalds 17191da177e4SLinus Torvalds while (*name=='/') 17201da177e4SLinus Torvalds name++; 17211da177e4SLinus Torvalds if (!*name) 1722086e183aSAl Viro return 0; 17231da177e4SLinus Torvalds 17241da177e4SLinus Torvalds /* At this point we know we have a real path component. */ 17251da177e4SLinus Torvalds for(;;) { 17261da177e4SLinus Torvalds struct qstr this; 1727200e9ef7SLinus Torvalds long len; 1728fe479a58SAl Viro int type; 17291da177e4SLinus Torvalds 173052094c8aSAl Viro err = may_lookup(nd); 17311da177e4SLinus Torvalds if (err) 17321da177e4SLinus Torvalds break; 17331da177e4SLinus Torvalds 1734200e9ef7SLinus Torvalds len = hash_name(name, &this.hash); 17351da177e4SLinus Torvalds this.name = name; 1736200e9ef7SLinus Torvalds this.len = len; 17371da177e4SLinus Torvalds 1738fe479a58SAl Viro type = LAST_NORM; 1739200e9ef7SLinus Torvalds if (name[0] == '.') switch (len) { 1740fe479a58SAl Viro case 2: 1741200e9ef7SLinus Torvalds if (name[1] == '.') { 1742fe479a58SAl Viro type = LAST_DOTDOT; 174316c2cd71SAl Viro nd->flags |= LOOKUP_JUMPED; 174416c2cd71SAl Viro } 1745fe479a58SAl Viro break; 1746fe479a58SAl Viro case 1: 1747fe479a58SAl Viro type = LAST_DOT; 1748fe479a58SAl Viro } 17495a202bcdSAl Viro if (likely(type == LAST_NORM)) { 17505a202bcdSAl Viro struct dentry *parent = nd->path.dentry; 175116c2cd71SAl Viro nd->flags &= ~LOOKUP_JUMPED; 17525a202bcdSAl Viro if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { 17535a202bcdSAl Viro err = parent->d_op->d_hash(parent, nd->inode, 17545a202bcdSAl Viro &this); 17555a202bcdSAl Viro if (err < 0) 17565a202bcdSAl Viro break; 17575a202bcdSAl Viro } 17585a202bcdSAl Viro } 1759fe479a58SAl Viro 1760200e9ef7SLinus Torvalds if (!name[len]) 17611da177e4SLinus Torvalds goto last_component; 1762200e9ef7SLinus Torvalds /* 1763200e9ef7SLinus Torvalds * If it wasn't NUL, we know it was '/'. Skip that 1764200e9ef7SLinus Torvalds * slash, and continue until no more slashes. 1765200e9ef7SLinus Torvalds */ 1766200e9ef7SLinus Torvalds do { 1767200e9ef7SLinus Torvalds len++; 1768200e9ef7SLinus Torvalds } while (unlikely(name[len] == '/')); 1769200e9ef7SLinus Torvalds if (!name[len]) 1770b356379aSAl Viro goto last_component; 1771200e9ef7SLinus Torvalds name += len; 17721da177e4SLinus Torvalds 1773ce57dfc1SAl Viro err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW); 1774ce57dfc1SAl Viro if (err < 0) 1775ce57dfc1SAl Viro return err; 1776fe479a58SAl Viro 1777ce57dfc1SAl Viro if (err) { 1778b356379aSAl Viro err = nested_symlink(&next, nd); 17791da177e4SLinus Torvalds if (err) 1780a7472babSAl Viro return err; 178131e6b01fSNick Piggin } 17823ddcd056SLinus Torvalds if (can_lookup(nd->inode)) 17831da177e4SLinus Torvalds continue; 17843ddcd056SLinus Torvalds err = -ENOTDIR; 17853ddcd056SLinus Torvalds break; 17861da177e4SLinus Torvalds /* here ends the main loop */ 17871da177e4SLinus Torvalds 17881da177e4SLinus Torvalds last_component: 1789ce57dfc1SAl Viro nd->last = this; 1790ce57dfc1SAl Viro nd->last_type = type; 1791ce57dfc1SAl Viro return 0; 1792ce57dfc1SAl Viro } 1793951361f9SAl Viro terminate_walk(nd); 17941da177e4SLinus Torvalds return err; 17951da177e4SLinus Torvalds } 17961da177e4SLinus Torvalds 179770e9b357SAl Viro static int path_init(int dfd, const char *name, unsigned int flags, 179870e9b357SAl Viro struct nameidata *nd, struct file **fp) 179931e6b01fSNick Piggin { 180031e6b01fSNick Piggin int retval = 0; 180131e6b01fSNick Piggin 180231e6b01fSNick Piggin nd->last_type = LAST_ROOT; /* if there are only slashes... */ 180316c2cd71SAl Viro nd->flags = flags | LOOKUP_JUMPED; 180431e6b01fSNick Piggin nd->depth = 0; 18055b6ca027SAl Viro if (flags & LOOKUP_ROOT) { 18065b6ca027SAl Viro struct inode *inode = nd->root.dentry->d_inode; 180773d049a4SAl Viro if (*name) { 18085b6ca027SAl Viro if (!inode->i_op->lookup) 18095b6ca027SAl Viro return -ENOTDIR; 18105b6ca027SAl Viro retval = inode_permission(inode, MAY_EXEC); 18115b6ca027SAl Viro if (retval) 18125b6ca027SAl Viro return retval; 181373d049a4SAl Viro } 18145b6ca027SAl Viro nd->path = nd->root; 18155b6ca027SAl Viro nd->inode = inode; 18165b6ca027SAl Viro if (flags & LOOKUP_RCU) { 181732a7991bSAl Viro lock_rcu_walk(); 18185b6ca027SAl Viro nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 18195b6ca027SAl Viro } else { 18205b6ca027SAl Viro path_get(&nd->path); 18215b6ca027SAl Viro } 18225b6ca027SAl Viro return 0; 18235b6ca027SAl Viro } 18245b6ca027SAl Viro 182531e6b01fSNick Piggin nd->root.mnt = NULL; 182631e6b01fSNick Piggin 182731e6b01fSNick Piggin if (*name=='/') { 1828e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 182932a7991bSAl Viro lock_rcu_walk(); 1830e41f7d4eSAl Viro set_root_rcu(nd); 1831e41f7d4eSAl Viro } else { 1832e41f7d4eSAl Viro set_root(nd); 1833e41f7d4eSAl Viro path_get(&nd->root); 1834e41f7d4eSAl Viro } 183531e6b01fSNick Piggin nd->path = nd->root; 183631e6b01fSNick Piggin } else if (dfd == AT_FDCWD) { 1837e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 183831e6b01fSNick Piggin struct fs_struct *fs = current->fs; 1839c28cc364SNick Piggin unsigned seq; 184031e6b01fSNick Piggin 184132a7991bSAl Viro lock_rcu_walk(); 184231e6b01fSNick Piggin 1843c28cc364SNick Piggin do { 1844c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 184531e6b01fSNick Piggin nd->path = fs->pwd; 1846c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 1847c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 1848e41f7d4eSAl Viro } else { 1849e41f7d4eSAl Viro get_fs_pwd(current->fs, &nd->path); 1850e41f7d4eSAl Viro } 185131e6b01fSNick Piggin } else { 18522903ff01SAl Viro struct fd f = fdget_raw(dfd); 185331e6b01fSNick Piggin struct dentry *dentry; 185431e6b01fSNick Piggin 18552903ff01SAl Viro if (!f.file) 18562903ff01SAl Viro return -EBADF; 185731e6b01fSNick Piggin 18582903ff01SAl Viro dentry = f.file->f_path.dentry; 185931e6b01fSNick Piggin 1860f52e0c11SAl Viro if (*name) { 18612903ff01SAl Viro if (!S_ISDIR(dentry->d_inode->i_mode)) { 18622903ff01SAl Viro fdput(f); 18632903ff01SAl Viro return -ENOTDIR; 1864f52e0c11SAl Viro } 186531e6b01fSNick Piggin 18662903ff01SAl Viro retval = inode_permission(dentry->d_inode, MAY_EXEC); 18672903ff01SAl Viro if (retval) { 18682903ff01SAl Viro fdput(f); 18692903ff01SAl Viro return retval; 18702903ff01SAl Viro } 18712903ff01SAl Viro } 18722903ff01SAl Viro 18732903ff01SAl Viro nd->path = f.file->f_path; 1874e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 18752903ff01SAl Viro if (f.need_put) 18762903ff01SAl Viro *fp = f.file; 1877c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 187832a7991bSAl Viro lock_rcu_walk(); 18795590ff0dSUlrich Drepper } else { 18802903ff01SAl Viro path_get(&nd->path); 18812903ff01SAl Viro fdput(f); 18821da177e4SLinus Torvalds } 1883e41f7d4eSAl Viro } 1884e41f7d4eSAl Viro 188531e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 18869b4a9b14SAl Viro return 0; 18879b4a9b14SAl Viro } 18889b4a9b14SAl Viro 1889bd92d7feSAl Viro static inline int lookup_last(struct nameidata *nd, struct path *path) 1890bd92d7feSAl Viro { 1891bd92d7feSAl Viro if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) 1892bd92d7feSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 1893bd92d7feSAl Viro 1894bd92d7feSAl Viro nd->flags &= ~LOOKUP_PARENT; 1895bd92d7feSAl Viro return walk_component(nd, path, &nd->last, nd->last_type, 1896bd92d7feSAl Viro nd->flags & LOOKUP_FOLLOW); 1897bd92d7feSAl Viro } 1898bd92d7feSAl Viro 18999b4a9b14SAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 1900ee0827cdSAl Viro static int path_lookupat(int dfd, const char *name, 19019b4a9b14SAl Viro unsigned int flags, struct nameidata *nd) 19029b4a9b14SAl Viro { 190370e9b357SAl Viro struct file *base = NULL; 1904bd92d7feSAl Viro struct path path; 1905bd92d7feSAl Viro int err; 190631e6b01fSNick Piggin 190731e6b01fSNick Piggin /* 190831e6b01fSNick Piggin * Path walking is largely split up into 2 different synchronisation 190931e6b01fSNick Piggin * schemes, rcu-walk and ref-walk (explained in 191031e6b01fSNick Piggin * Documentation/filesystems/path-lookup.txt). These share much of the 191131e6b01fSNick Piggin * path walk code, but some things particularly setup, cleanup, and 191231e6b01fSNick Piggin * following mounts are sufficiently divergent that functions are 191331e6b01fSNick Piggin * duplicated. Typically there is a function foo(), and its RCU 191431e6b01fSNick Piggin * analogue, foo_rcu(). 191531e6b01fSNick Piggin * 191631e6b01fSNick Piggin * -ECHILD is the error number of choice (just to avoid clashes) that 191731e6b01fSNick Piggin * is returned if some aspect of an rcu-walk fails. Such an error must 191831e6b01fSNick Piggin * be handled by restarting a traditional ref-walk (which will always 191931e6b01fSNick Piggin * be able to complete). 192031e6b01fSNick Piggin */ 1921bd92d7feSAl Viro err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base); 1922ee0827cdSAl Viro 1923bd92d7feSAl Viro if (unlikely(err)) 1924bd92d7feSAl Viro return err; 1925ee0827cdSAl Viro 1926ee0827cdSAl Viro current->total_link_count = 0; 1927bd92d7feSAl Viro err = link_path_walk(name, nd); 1928bd92d7feSAl Viro 1929bd92d7feSAl Viro if (!err && !(flags & LOOKUP_PARENT)) { 1930bd92d7feSAl Viro err = lookup_last(nd, &path); 1931bd92d7feSAl Viro while (err > 0) { 1932bd92d7feSAl Viro void *cookie; 1933bd92d7feSAl Viro struct path link = path; 1934800179c9SKees Cook err = may_follow_link(&link, nd); 1935800179c9SKees Cook if (unlikely(err)) 1936800179c9SKees Cook break; 1937bd92d7feSAl Viro nd->flags |= LOOKUP_PARENT; 1938574197e0SAl Viro err = follow_link(&link, nd, &cookie); 19396d7b5aaeSAl Viro if (err) 19406d7b5aaeSAl Viro break; 1941bd92d7feSAl Viro err = lookup_last(nd, &path); 1942574197e0SAl Viro put_link(nd, &link, cookie); 1943bd92d7feSAl Viro } 1944bd92d7feSAl Viro } 1945ee0827cdSAl Viro 19469f1fafeeSAl Viro if (!err) 19479f1fafeeSAl Viro err = complete_walk(nd); 1948bd92d7feSAl Viro 1949bd92d7feSAl Viro if (!err && nd->flags & LOOKUP_DIRECTORY) { 1950bd92d7feSAl Viro if (!nd->inode->i_op->lookup) { 1951bd92d7feSAl Viro path_put(&nd->path); 1952bd23a539SAl Viro err = -ENOTDIR; 1953bd92d7feSAl Viro } 1954bd92d7feSAl Viro } 195516c2cd71SAl Viro 195670e9b357SAl Viro if (base) 195770e9b357SAl Viro fput(base); 1958ee0827cdSAl Viro 19595b6ca027SAl Viro if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 196031e6b01fSNick Piggin path_put(&nd->root); 196131e6b01fSNick Piggin nd->root.mnt = NULL; 196231e6b01fSNick Piggin } 1963bd92d7feSAl Viro return err; 196431e6b01fSNick Piggin } 196531e6b01fSNick Piggin 1966ee0827cdSAl Viro static int do_path_lookup(int dfd, const char *name, 1967ee0827cdSAl Viro unsigned int flags, struct nameidata *nd) 1968ee0827cdSAl Viro { 1969ee0827cdSAl Viro int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd); 1970ee0827cdSAl Viro if (unlikely(retval == -ECHILD)) 1971ee0827cdSAl Viro retval = path_lookupat(dfd, name, flags, nd); 1972ee0827cdSAl Viro if (unlikely(retval == -ESTALE)) 1973ee0827cdSAl Viro retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd); 1974ee0827cdSAl Viro 197531e6b01fSNick Piggin if (likely(!retval)) { 197631e6b01fSNick Piggin if (unlikely(!audit_dummy_context())) { 197731e6b01fSNick Piggin if (nd->path.dentry && nd->inode) 197831e6b01fSNick Piggin audit_inode(name, nd->path.dentry); 197931e6b01fSNick Piggin } 198031e6b01fSNick Piggin } 1981170aa3d0SUlrich Drepper return retval; 19821da177e4SLinus Torvalds } 19831da177e4SLinus Torvalds 198479714f72SAl Viro /* does lookup, returns the object with parent locked */ 198579714f72SAl Viro struct dentry *kern_path_locked(const char *name, struct path *path) 19865590ff0dSUlrich Drepper { 198779714f72SAl Viro struct nameidata nd; 198879714f72SAl Viro struct dentry *d; 198979714f72SAl Viro int err = do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, &nd); 199079714f72SAl Viro if (err) 199179714f72SAl Viro return ERR_PTR(err); 199279714f72SAl Viro if (nd.last_type != LAST_NORM) { 199379714f72SAl Viro path_put(&nd.path); 199479714f72SAl Viro return ERR_PTR(-EINVAL); 199579714f72SAl Viro } 199679714f72SAl Viro mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 19971e0ea001SAl Viro d = __lookup_hash(&nd.last, nd.path.dentry, 0); 199879714f72SAl Viro if (IS_ERR(d)) { 199979714f72SAl Viro mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 200079714f72SAl Viro path_put(&nd.path); 200179714f72SAl Viro return d; 200279714f72SAl Viro } 200379714f72SAl Viro *path = nd.path; 200479714f72SAl Viro return d; 20055590ff0dSUlrich Drepper } 20065590ff0dSUlrich Drepper 2007d1811465SAl Viro int kern_path(const char *name, unsigned int flags, struct path *path) 2008d1811465SAl Viro { 2009d1811465SAl Viro struct nameidata nd; 2010d1811465SAl Viro int res = do_path_lookup(AT_FDCWD, name, flags, &nd); 2011d1811465SAl Viro if (!res) 2012d1811465SAl Viro *path = nd.path; 2013d1811465SAl Viro return res; 2014d1811465SAl Viro } 2015d1811465SAl Viro 201616f18200SJosef 'Jeff' Sipek /** 201716f18200SJosef 'Jeff' Sipek * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair 201816f18200SJosef 'Jeff' Sipek * @dentry: pointer to dentry of the base directory 201916f18200SJosef 'Jeff' Sipek * @mnt: pointer to vfs mount of the base directory 202016f18200SJosef 'Jeff' Sipek * @name: pointer to file name 202116f18200SJosef 'Jeff' Sipek * @flags: lookup flags 2022e0a01249SAl Viro * @path: pointer to struct path to fill 202316f18200SJosef 'Jeff' Sipek */ 202416f18200SJosef 'Jeff' Sipek int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, 202516f18200SJosef 'Jeff' Sipek const char *name, unsigned int flags, 2026e0a01249SAl Viro struct path *path) 202716f18200SJosef 'Jeff' Sipek { 2028e0a01249SAl Viro struct nameidata nd; 2029e0a01249SAl Viro int err; 2030e0a01249SAl Viro nd.root.dentry = dentry; 2031e0a01249SAl Viro nd.root.mnt = mnt; 2032e0a01249SAl Viro BUG_ON(flags & LOOKUP_PARENT); 20335b6ca027SAl Viro /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */ 2034e0a01249SAl Viro err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd); 2035e0a01249SAl Viro if (!err) 2036e0a01249SAl Viro *path = nd.path; 2037e0a01249SAl Viro return err; 203816f18200SJosef 'Jeff' Sipek } 203916f18200SJosef 'Jeff' Sipek 2040057f6c01SJames Morris /* 2041057f6c01SJames Morris * Restricted form of lookup. Doesn't follow links, single-component only, 2042057f6c01SJames Morris * needs parent already locked. Doesn't follow mounts. 2043057f6c01SJames Morris * SMP-safe. 2044057f6c01SJames Morris */ 2045a244e169SAdrian Bunk static struct dentry *lookup_hash(struct nameidata *nd) 20461da177e4SLinus Torvalds { 204772bd866aSAl Viro return __lookup_hash(&nd->last, nd->path.dentry, nd->flags); 20481da177e4SLinus Torvalds } 20491da177e4SLinus Torvalds 2050eead1911SChristoph Hellwig /** 2051a6b91919SRandy Dunlap * lookup_one_len - filesystem helper to lookup single pathname component 2052eead1911SChristoph Hellwig * @name: pathname component to lookup 2053eead1911SChristoph Hellwig * @base: base directory to lookup from 2054eead1911SChristoph Hellwig * @len: maximum length @len should be interpreted to 2055eead1911SChristoph Hellwig * 2056a6b91919SRandy Dunlap * Note that this routine is purely a helper for filesystem usage and should 2057a6b91919SRandy Dunlap * not be called by generic code. Also note that by using this function the 2058eead1911SChristoph Hellwig * nameidata argument is passed to the filesystem methods and a filesystem 2059eead1911SChristoph Hellwig * using this helper needs to be prepared for that. 2060eead1911SChristoph Hellwig */ 2061057f6c01SJames Morris struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 2062057f6c01SJames Morris { 2063057f6c01SJames Morris struct qstr this; 20646a96ba54SAl Viro unsigned int c; 2065cda309deSMiklos Szeredi int err; 2066057f6c01SJames Morris 20672f9092e1SDavid Woodhouse WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex)); 20682f9092e1SDavid Woodhouse 20696a96ba54SAl Viro this.name = name; 20706a96ba54SAl Viro this.len = len; 20710145acc2SLinus Torvalds this.hash = full_name_hash(name, len); 20726a96ba54SAl Viro if (!len) 20736a96ba54SAl Viro return ERR_PTR(-EACCES); 20746a96ba54SAl Viro 20756a96ba54SAl Viro while (len--) { 20766a96ba54SAl Viro c = *(const unsigned char *)name++; 20776a96ba54SAl Viro if (c == '/' || c == '\0') 20786a96ba54SAl Viro return ERR_PTR(-EACCES); 20796a96ba54SAl Viro } 20805a202bcdSAl Viro /* 20815a202bcdSAl Viro * See if the low-level filesystem might want 20825a202bcdSAl Viro * to use its own hash.. 20835a202bcdSAl Viro */ 20845a202bcdSAl Viro if (base->d_flags & DCACHE_OP_HASH) { 20855a202bcdSAl Viro int err = base->d_op->d_hash(base, base->d_inode, &this); 20865a202bcdSAl Viro if (err < 0) 20875a202bcdSAl Viro return ERR_PTR(err); 20885a202bcdSAl Viro } 2089eead1911SChristoph Hellwig 2090cda309deSMiklos Szeredi err = inode_permission(base->d_inode, MAY_EXEC); 2091cda309deSMiklos Szeredi if (err) 2092cda309deSMiklos Szeredi return ERR_PTR(err); 2093cda309deSMiklos Szeredi 209472bd866aSAl Viro return __lookup_hash(&this, base, 0); 2095057f6c01SJames Morris } 2096057f6c01SJames Morris 20971fa1e7f6SAndy Whitcroft int user_path_at_empty(int dfd, const char __user *name, unsigned flags, 20981fa1e7f6SAndy Whitcroft struct path *path, int *empty) 20991da177e4SLinus Torvalds { 21002d8f3038SAl Viro struct nameidata nd; 21011fa1e7f6SAndy Whitcroft char *tmp = getname_flags(name, flags, empty); 21021da177e4SLinus Torvalds int err = PTR_ERR(tmp); 21031da177e4SLinus Torvalds if (!IS_ERR(tmp)) { 21042d8f3038SAl Viro 21052d8f3038SAl Viro BUG_ON(flags & LOOKUP_PARENT); 21062d8f3038SAl Viro 21072d8f3038SAl Viro err = do_path_lookup(dfd, tmp, flags, &nd); 21081da177e4SLinus Torvalds putname(tmp); 21092d8f3038SAl Viro if (!err) 21102d8f3038SAl Viro *path = nd.path; 21111da177e4SLinus Torvalds } 21121da177e4SLinus Torvalds return err; 21131da177e4SLinus Torvalds } 21141da177e4SLinus Torvalds 21151fa1e7f6SAndy Whitcroft int user_path_at(int dfd, const char __user *name, unsigned flags, 21161fa1e7f6SAndy Whitcroft struct path *path) 21171fa1e7f6SAndy Whitcroft { 2118f7493e5dSLinus Torvalds return user_path_at_empty(dfd, name, flags, path, NULL); 21191fa1e7f6SAndy Whitcroft } 21201fa1e7f6SAndy Whitcroft 21212ad94ae6SAl Viro static int user_path_parent(int dfd, const char __user *path, 21222ad94ae6SAl Viro struct nameidata *nd, char **name) 21232ad94ae6SAl Viro { 21242ad94ae6SAl Viro char *s = getname(path); 21252ad94ae6SAl Viro int error; 21262ad94ae6SAl Viro 21272ad94ae6SAl Viro if (IS_ERR(s)) 21282ad94ae6SAl Viro return PTR_ERR(s); 21292ad94ae6SAl Viro 21302ad94ae6SAl Viro error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd); 21312ad94ae6SAl Viro if (error) 21322ad94ae6SAl Viro putname(s); 21332ad94ae6SAl Viro else 21342ad94ae6SAl Viro *name = s; 21352ad94ae6SAl Viro 21362ad94ae6SAl Viro return error; 21372ad94ae6SAl Viro } 21382ad94ae6SAl Viro 21391da177e4SLinus Torvalds /* 21401da177e4SLinus Torvalds * It's inline, so penalty for filesystems that don't use sticky bit is 21411da177e4SLinus Torvalds * minimal. 21421da177e4SLinus Torvalds */ 21431da177e4SLinus Torvalds static inline int check_sticky(struct inode *dir, struct inode *inode) 21441da177e4SLinus Torvalds { 21458e96e3b7SEric W. Biederman kuid_t fsuid = current_fsuid(); 2146da9592edSDavid Howells 21471da177e4SLinus Torvalds if (!(dir->i_mode & S_ISVTX)) 21481da177e4SLinus Torvalds return 0; 21498e96e3b7SEric W. Biederman if (uid_eq(inode->i_uid, fsuid)) 21501da177e4SLinus Torvalds return 0; 21518e96e3b7SEric W. Biederman if (uid_eq(dir->i_uid, fsuid)) 21521da177e4SLinus Torvalds return 0; 21531a48e2acSEric W. Biederman return !inode_capable(inode, CAP_FOWNER); 21541da177e4SLinus Torvalds } 21551da177e4SLinus Torvalds 21561da177e4SLinus Torvalds /* 21571da177e4SLinus Torvalds * Check whether we can remove a link victim from directory dir, check 21581da177e4SLinus Torvalds * whether the type of victim is right. 21591da177e4SLinus Torvalds * 1. We can't do it if dir is read-only (done in permission()) 21601da177e4SLinus Torvalds * 2. We should have write and exec permissions on dir 21611da177e4SLinus Torvalds * 3. We can't remove anything from append-only dir 21621da177e4SLinus Torvalds * 4. We can't do anything with immutable dir (done in permission()) 21631da177e4SLinus Torvalds * 5. If the sticky bit on dir is set we should either 21641da177e4SLinus Torvalds * a. be owner of dir, or 21651da177e4SLinus Torvalds * b. be owner of victim, or 21661da177e4SLinus Torvalds * c. have CAP_FOWNER capability 21671da177e4SLinus Torvalds * 6. If the victim is append-only or immutable we can't do antyhing with 21681da177e4SLinus Torvalds * links pointing to it. 21691da177e4SLinus Torvalds * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. 21701da177e4SLinus Torvalds * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. 21711da177e4SLinus Torvalds * 9. We can't remove a root or mountpoint. 21721da177e4SLinus Torvalds * 10. We don't allow removal of NFS sillyrenamed files; it's handled by 21731da177e4SLinus Torvalds * nfs_async_unlink(). 21741da177e4SLinus Torvalds */ 2175858119e1SArjan van de Ven static int may_delete(struct inode *dir,struct dentry *victim,int isdir) 21761da177e4SLinus Torvalds { 21771da177e4SLinus Torvalds int error; 21781da177e4SLinus Torvalds 21791da177e4SLinus Torvalds if (!victim->d_inode) 21801da177e4SLinus Torvalds return -ENOENT; 21811da177e4SLinus Torvalds 21821da177e4SLinus Torvalds BUG_ON(victim->d_parent->d_inode != dir); 2183cccc6bbaSAl Viro audit_inode_child(victim, dir); 21841da177e4SLinus Torvalds 2185f419a2e3SAl Viro error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 21861da177e4SLinus Torvalds if (error) 21871da177e4SLinus Torvalds return error; 21881da177e4SLinus Torvalds if (IS_APPEND(dir)) 21891da177e4SLinus Torvalds return -EPERM; 21901da177e4SLinus Torvalds if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)|| 2191f9454548SHugh Dickins IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode)) 21921da177e4SLinus Torvalds return -EPERM; 21931da177e4SLinus Torvalds if (isdir) { 21941da177e4SLinus Torvalds if (!S_ISDIR(victim->d_inode->i_mode)) 21951da177e4SLinus Torvalds return -ENOTDIR; 21961da177e4SLinus Torvalds if (IS_ROOT(victim)) 21971da177e4SLinus Torvalds return -EBUSY; 21981da177e4SLinus Torvalds } else if (S_ISDIR(victim->d_inode->i_mode)) 21991da177e4SLinus Torvalds return -EISDIR; 22001da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 22011da177e4SLinus Torvalds return -ENOENT; 22021da177e4SLinus Torvalds if (victim->d_flags & DCACHE_NFSFS_RENAMED) 22031da177e4SLinus Torvalds return -EBUSY; 22041da177e4SLinus Torvalds return 0; 22051da177e4SLinus Torvalds } 22061da177e4SLinus Torvalds 22071da177e4SLinus Torvalds /* Check whether we can create an object with dentry child in directory 22081da177e4SLinus Torvalds * dir. 22091da177e4SLinus Torvalds * 1. We can't do it if child already exists (open has special treatment for 22101da177e4SLinus Torvalds * this case, but since we are inlined it's OK) 22111da177e4SLinus Torvalds * 2. We can't do it if dir is read-only (done in permission()) 22121da177e4SLinus Torvalds * 3. We should have write and exec permissions on dir 22131da177e4SLinus Torvalds * 4. We can't do it if dir is immutable (done in permission()) 22141da177e4SLinus Torvalds */ 2215a95164d9SMiklos Szeredi static inline int may_create(struct inode *dir, struct dentry *child) 22161da177e4SLinus Torvalds { 22171da177e4SLinus Torvalds if (child->d_inode) 22181da177e4SLinus Torvalds return -EEXIST; 22191da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 22201da177e4SLinus Torvalds return -ENOENT; 2221f419a2e3SAl Viro return inode_permission(dir, MAY_WRITE | MAY_EXEC); 22221da177e4SLinus Torvalds } 22231da177e4SLinus Torvalds 22241da177e4SLinus Torvalds /* 22251da177e4SLinus Torvalds * p1 and p2 should be directories on the same fs. 22261da177e4SLinus Torvalds */ 22271da177e4SLinus Torvalds struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 22281da177e4SLinus Torvalds { 22291da177e4SLinus Torvalds struct dentry *p; 22301da177e4SLinus Torvalds 22311da177e4SLinus Torvalds if (p1 == p2) { 2232f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 22331da177e4SLinus Torvalds return NULL; 22341da177e4SLinus Torvalds } 22351da177e4SLinus Torvalds 2236a11f3a05SArjan van de Ven mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 22371da177e4SLinus Torvalds 2238e2761a11SOGAWA Hirofumi p = d_ancestor(p2, p1); 2239e2761a11SOGAWA Hirofumi if (p) { 2240f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT); 2241f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD); 22421da177e4SLinus Torvalds return p; 22431da177e4SLinus Torvalds } 22441da177e4SLinus Torvalds 2245e2761a11SOGAWA Hirofumi p = d_ancestor(p1, p2); 2246e2761a11SOGAWA Hirofumi if (p) { 2247f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 2248f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); 22491da177e4SLinus Torvalds return p; 22501da177e4SLinus Torvalds } 22511da177e4SLinus Torvalds 2252f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 2253f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); 22541da177e4SLinus Torvalds return NULL; 22551da177e4SLinus Torvalds } 22561da177e4SLinus Torvalds 22571da177e4SLinus Torvalds void unlock_rename(struct dentry *p1, struct dentry *p2) 22581da177e4SLinus Torvalds { 22591b1dcc1bSJes Sorensen mutex_unlock(&p1->d_inode->i_mutex); 22601da177e4SLinus Torvalds if (p1 != p2) { 22611b1dcc1bSJes Sorensen mutex_unlock(&p2->d_inode->i_mutex); 2262a11f3a05SArjan van de Ven mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 22631da177e4SLinus Torvalds } 22641da177e4SLinus Torvalds } 22651da177e4SLinus Torvalds 22664acdaf27SAl Viro int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, 2267312b63fbSAl Viro bool want_excl) 22681da177e4SLinus Torvalds { 2269a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 22701da177e4SLinus Torvalds if (error) 22711da177e4SLinus Torvalds return error; 22721da177e4SLinus Torvalds 2273acfa4380SAl Viro if (!dir->i_op->create) 22741da177e4SLinus Torvalds return -EACCES; /* shouldn't it be ENOSYS? */ 22751da177e4SLinus Torvalds mode &= S_IALLUGO; 22761da177e4SLinus Torvalds mode |= S_IFREG; 22771da177e4SLinus Torvalds error = security_inode_create(dir, dentry, mode); 22781da177e4SLinus Torvalds if (error) 22791da177e4SLinus Torvalds return error; 2280312b63fbSAl Viro error = dir->i_op->create(dir, dentry, mode, want_excl); 2281a74574aaSStephen Smalley if (!error) 2282f38aa942SAmy Griffis fsnotify_create(dir, dentry); 22831da177e4SLinus Torvalds return error; 22841da177e4SLinus Torvalds } 22851da177e4SLinus Torvalds 228673d049a4SAl Viro static int may_open(struct path *path, int acc_mode, int flag) 22871da177e4SLinus Torvalds { 22883fb64190SChristoph Hellwig struct dentry *dentry = path->dentry; 22891da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 22901da177e4SLinus Torvalds int error; 22911da177e4SLinus Torvalds 2292bcda7652SAl Viro /* O_PATH? */ 2293bcda7652SAl Viro if (!acc_mode) 2294bcda7652SAl Viro return 0; 2295bcda7652SAl Viro 22961da177e4SLinus Torvalds if (!inode) 22971da177e4SLinus Torvalds return -ENOENT; 22981da177e4SLinus Torvalds 2299c8fe8f30SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 2300c8fe8f30SChristoph Hellwig case S_IFLNK: 23011da177e4SLinus Torvalds return -ELOOP; 2302c8fe8f30SChristoph Hellwig case S_IFDIR: 2303c8fe8f30SChristoph Hellwig if (acc_mode & MAY_WRITE) 23041da177e4SLinus Torvalds return -EISDIR; 2305c8fe8f30SChristoph Hellwig break; 2306c8fe8f30SChristoph Hellwig case S_IFBLK: 2307c8fe8f30SChristoph Hellwig case S_IFCHR: 23083fb64190SChristoph Hellwig if (path->mnt->mnt_flags & MNT_NODEV) 23091da177e4SLinus Torvalds return -EACCES; 2310c8fe8f30SChristoph Hellwig /*FALLTHRU*/ 2311c8fe8f30SChristoph Hellwig case S_IFIFO: 2312c8fe8f30SChristoph Hellwig case S_IFSOCK: 23131da177e4SLinus Torvalds flag &= ~O_TRUNC; 2314c8fe8f30SChristoph Hellwig break; 23154a3fd211SDave Hansen } 2316b41572e9SDave Hansen 23173fb64190SChristoph Hellwig error = inode_permission(inode, acc_mode); 2318b41572e9SDave Hansen if (error) 2319b41572e9SDave Hansen return error; 23206146f0d5SMimi Zohar 23211da177e4SLinus Torvalds /* 23221da177e4SLinus Torvalds * An append-only file must be opened in append mode for writing. 23231da177e4SLinus Torvalds */ 23241da177e4SLinus Torvalds if (IS_APPEND(inode)) { 23258737c930SAl Viro if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) 23267715b521SAl Viro return -EPERM; 23271da177e4SLinus Torvalds if (flag & O_TRUNC) 23287715b521SAl Viro return -EPERM; 23291da177e4SLinus Torvalds } 23301da177e4SLinus Torvalds 23311da177e4SLinus Torvalds /* O_NOATIME can only be set by the owner or superuser */ 23322e149670SSerge E. Hallyn if (flag & O_NOATIME && !inode_owner_or_capable(inode)) 23337715b521SAl Viro return -EPERM; 23341da177e4SLinus Torvalds 2335f3c7691eSJ. Bruce Fields return 0; 23367715b521SAl Viro } 23377715b521SAl Viro 2338e1181ee6SJeff Layton static int handle_truncate(struct file *filp) 23397715b521SAl Viro { 2340e1181ee6SJeff Layton struct path *path = &filp->f_path; 23417715b521SAl Viro struct inode *inode = path->dentry->d_inode; 23427715b521SAl Viro int error = get_write_access(inode); 23431da177e4SLinus Torvalds if (error) 23447715b521SAl Viro return error; 23451da177e4SLinus Torvalds /* 23461da177e4SLinus Torvalds * Refuse to truncate files with mandatory locks held on them. 23471da177e4SLinus Torvalds */ 23481da177e4SLinus Torvalds error = locks_verify_locked(inode); 2349be6d3e56SKentaro Takeda if (!error) 2350ea0d3ab2STetsuo Handa error = security_path_truncate(path); 23511da177e4SLinus Torvalds if (!error) { 23527715b521SAl Viro error = do_truncate(path->dentry, 0, 2353d139d7ffSMiklos Szeredi ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, 2354e1181ee6SJeff Layton filp); 23551da177e4SLinus Torvalds } 23561da177e4SLinus Torvalds put_write_access(inode); 2357acd0c935SMimi Zohar return error; 23581da177e4SLinus Torvalds } 23591da177e4SLinus Torvalds 2360d57999e1SDave Hansen static inline int open_to_namei_flags(int flag) 2361d57999e1SDave Hansen { 23628a5e929dSAl Viro if ((flag & O_ACCMODE) == 3) 23638a5e929dSAl Viro flag--; 2364d57999e1SDave Hansen return flag; 2365d57999e1SDave Hansen } 2366d57999e1SDave Hansen 2367d18e9008SMiklos Szeredi static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode) 2368d18e9008SMiklos Szeredi { 2369d18e9008SMiklos Szeredi int error = security_path_mknod(dir, dentry, mode, 0); 2370d18e9008SMiklos Szeredi if (error) 2371d18e9008SMiklos Szeredi return error; 2372d18e9008SMiklos Szeredi 2373d18e9008SMiklos Szeredi error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC); 2374d18e9008SMiklos Szeredi if (error) 2375d18e9008SMiklos Szeredi return error; 2376d18e9008SMiklos Szeredi 2377d18e9008SMiklos Szeredi return security_inode_create(dir->dentry->d_inode, dentry, mode); 2378d18e9008SMiklos Szeredi } 2379d18e9008SMiklos Szeredi 23801acf0af9SDavid Howells /* 23811acf0af9SDavid Howells * Attempt to atomically look up, create and open a file from a negative 23821acf0af9SDavid Howells * dentry. 23831acf0af9SDavid Howells * 23841acf0af9SDavid Howells * Returns 0 if successful. The file will have been created and attached to 23851acf0af9SDavid Howells * @file by the filesystem calling finish_open(). 23861acf0af9SDavid Howells * 23871acf0af9SDavid Howells * Returns 1 if the file was looked up only or didn't need creating. The 23881acf0af9SDavid Howells * caller will need to perform the open themselves. @path will have been 23891acf0af9SDavid Howells * updated to point to the new dentry. This may be negative. 23901acf0af9SDavid Howells * 23911acf0af9SDavid Howells * Returns an error code otherwise. 23921acf0af9SDavid Howells */ 23932675a4ebSAl Viro static int atomic_open(struct nameidata *nd, struct dentry *dentry, 239430d90494SAl Viro struct path *path, struct file *file, 2395015c3bbcSMiklos Szeredi const struct open_flags *op, 239664894cf8SAl Viro bool got_write, bool need_lookup, 239747237687SAl Viro int *opened) 2398d18e9008SMiklos Szeredi { 2399d18e9008SMiklos Szeredi struct inode *dir = nd->path.dentry->d_inode; 2400d18e9008SMiklos Szeredi unsigned open_flag = open_to_namei_flags(op->open_flag); 2401d18e9008SMiklos Szeredi umode_t mode; 2402d18e9008SMiklos Szeredi int error; 2403d18e9008SMiklos Szeredi int acc_mode; 2404d18e9008SMiklos Szeredi int create_error = 0; 2405d18e9008SMiklos Szeredi struct dentry *const DENTRY_NOT_SET = (void *) -1UL; 2406d18e9008SMiklos Szeredi 2407d18e9008SMiklos Szeredi BUG_ON(dentry->d_inode); 2408d18e9008SMiklos Szeredi 2409d18e9008SMiklos Szeredi /* Don't create child dentry for a dead directory. */ 2410d18e9008SMiklos Szeredi if (unlikely(IS_DEADDIR(dir))) { 24112675a4ebSAl Viro error = -ENOENT; 2412d18e9008SMiklos Szeredi goto out; 2413d18e9008SMiklos Szeredi } 2414d18e9008SMiklos Szeredi 241562b259d8SMiklos Szeredi mode = op->mode; 2416d18e9008SMiklos Szeredi if ((open_flag & O_CREAT) && !IS_POSIXACL(dir)) 2417d18e9008SMiklos Szeredi mode &= ~current_umask(); 2418d18e9008SMiklos Szeredi 2419f8310c59SAl Viro if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT)) { 2420d18e9008SMiklos Szeredi open_flag &= ~O_TRUNC; 242147237687SAl Viro *opened |= FILE_CREATED; 2422d18e9008SMiklos Szeredi } 2423d18e9008SMiklos Szeredi 2424d18e9008SMiklos Szeredi /* 2425d18e9008SMiklos Szeredi * Checking write permission is tricky, bacuse we don't know if we are 2426d18e9008SMiklos Szeredi * going to actually need it: O_CREAT opens should work as long as the 2427d18e9008SMiklos Szeredi * file exists. But checking existence breaks atomicity. The trick is 2428d18e9008SMiklos Szeredi * to check access and if not granted clear O_CREAT from the flags. 2429d18e9008SMiklos Szeredi * 2430d18e9008SMiklos Szeredi * Another problem is returing the "right" error value (e.g. for an 2431d18e9008SMiklos Szeredi * O_EXCL open we want to return EEXIST not EROFS). 2432d18e9008SMiklos Szeredi */ 243364894cf8SAl Viro if (((open_flag & (O_CREAT | O_TRUNC)) || 243464894cf8SAl Viro (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) { 243564894cf8SAl Viro if (!(open_flag & O_CREAT)) { 2436d18e9008SMiklos Szeredi /* 2437d18e9008SMiklos Szeredi * No O_CREATE -> atomicity not a requirement -> fall 2438d18e9008SMiklos Szeredi * back to lookup + open 2439d18e9008SMiklos Szeredi */ 2440d18e9008SMiklos Szeredi goto no_open; 2441d18e9008SMiklos Szeredi } else if (open_flag & (O_EXCL | O_TRUNC)) { 2442d18e9008SMiklos Szeredi /* Fall back and fail with the right error */ 244364894cf8SAl Viro create_error = -EROFS; 2444d18e9008SMiklos Szeredi goto no_open; 2445d18e9008SMiklos Szeredi } else { 2446d18e9008SMiklos Szeredi /* No side effects, safe to clear O_CREAT */ 244764894cf8SAl Viro create_error = -EROFS; 2448d18e9008SMiklos Szeredi open_flag &= ~O_CREAT; 2449d18e9008SMiklos Szeredi } 2450d18e9008SMiklos Szeredi } 2451d18e9008SMiklos Szeredi 2452d18e9008SMiklos Szeredi if (open_flag & O_CREAT) { 245338227f78SMiklos Szeredi error = may_o_create(&nd->path, dentry, mode); 2454d18e9008SMiklos Szeredi if (error) { 2455d18e9008SMiklos Szeredi create_error = error; 2456d18e9008SMiklos Szeredi if (open_flag & O_EXCL) 2457d18e9008SMiklos Szeredi goto no_open; 2458d18e9008SMiklos Szeredi open_flag &= ~O_CREAT; 2459d18e9008SMiklos Szeredi } 2460d18e9008SMiklos Szeredi } 2461d18e9008SMiklos Szeredi 2462d18e9008SMiklos Szeredi if (nd->flags & LOOKUP_DIRECTORY) 2463d18e9008SMiklos Szeredi open_flag |= O_DIRECTORY; 2464d18e9008SMiklos Szeredi 246530d90494SAl Viro file->f_path.dentry = DENTRY_NOT_SET; 246630d90494SAl Viro file->f_path.mnt = nd->path.mnt; 246730d90494SAl Viro error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode, 246847237687SAl Viro opened); 2469d9585277SAl Viro if (error < 0) { 2470d9585277SAl Viro if (create_error && error == -ENOENT) 2471d9585277SAl Viro error = create_error; 2472d18e9008SMiklos Szeredi goto out; 2473d18e9008SMiklos Szeredi } 2474d18e9008SMiklos Szeredi 2475d18e9008SMiklos Szeredi acc_mode = op->acc_mode; 247647237687SAl Viro if (*opened & FILE_CREATED) { 2477d18e9008SMiklos Szeredi fsnotify_create(dir, dentry); 2478d18e9008SMiklos Szeredi acc_mode = MAY_OPEN; 2479d18e9008SMiklos Szeredi } 2480d18e9008SMiklos Szeredi 2481d9585277SAl Viro if (error) { /* returned 1, that is */ 248230d90494SAl Viro if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { 24832675a4ebSAl Viro error = -EIO; 2484d18e9008SMiklos Szeredi goto out; 2485d18e9008SMiklos Szeredi } 248630d90494SAl Viro if (file->f_path.dentry) { 2487d18e9008SMiklos Szeredi dput(dentry); 248830d90494SAl Viro dentry = file->f_path.dentry; 2489d18e9008SMiklos Szeredi } 249062b2ce96SSage Weil if (create_error && dentry->d_inode == NULL) { 249162b2ce96SSage Weil error = create_error; 249262b2ce96SSage Weil goto out; 249362b2ce96SSage Weil } 2494d18e9008SMiklos Szeredi goto looked_up; 2495d18e9008SMiklos Szeredi } 2496d18e9008SMiklos Szeredi 2497d18e9008SMiklos Szeredi /* 2498d18e9008SMiklos Szeredi * We didn't have the inode before the open, so check open permission 2499d18e9008SMiklos Szeredi * here. 2500d18e9008SMiklos Szeredi */ 25012675a4ebSAl Viro error = may_open(&file->f_path, acc_mode, open_flag); 25022675a4ebSAl Viro if (error) 25032675a4ebSAl Viro fput(file); 2504d18e9008SMiklos Szeredi 2505d18e9008SMiklos Szeredi out: 2506d18e9008SMiklos Szeredi dput(dentry); 25072675a4ebSAl Viro return error; 2508d18e9008SMiklos Szeredi 2509d18e9008SMiklos Szeredi no_open: 2510d18e9008SMiklos Szeredi if (need_lookup) { 251172bd866aSAl Viro dentry = lookup_real(dir, dentry, nd->flags); 2512d18e9008SMiklos Szeredi if (IS_ERR(dentry)) 25132675a4ebSAl Viro return PTR_ERR(dentry); 2514d18e9008SMiklos Szeredi 2515d18e9008SMiklos Szeredi if (create_error) { 2516d18e9008SMiklos Szeredi int open_flag = op->open_flag; 2517d18e9008SMiklos Szeredi 25182675a4ebSAl Viro error = create_error; 2519d18e9008SMiklos Szeredi if ((open_flag & O_EXCL)) { 2520d18e9008SMiklos Szeredi if (!dentry->d_inode) 2521d18e9008SMiklos Szeredi goto out; 2522d18e9008SMiklos Szeredi } else if (!dentry->d_inode) { 2523d18e9008SMiklos Szeredi goto out; 2524d18e9008SMiklos Szeredi } else if ((open_flag & O_TRUNC) && 2525d18e9008SMiklos Szeredi S_ISREG(dentry->d_inode->i_mode)) { 2526d18e9008SMiklos Szeredi goto out; 2527d18e9008SMiklos Szeredi } 2528d18e9008SMiklos Szeredi /* will fail later, go on to get the right error */ 2529d18e9008SMiklos Szeredi } 2530d18e9008SMiklos Szeredi } 2531d18e9008SMiklos Szeredi looked_up: 2532d18e9008SMiklos Szeredi path->dentry = dentry; 2533d18e9008SMiklos Szeredi path->mnt = nd->path.mnt; 25342675a4ebSAl Viro return 1; 2535d18e9008SMiklos Szeredi } 2536d18e9008SMiklos Szeredi 253731e6b01fSNick Piggin /* 25381acf0af9SDavid Howells * Look up and maybe create and open the last component. 2539d58ffd35SMiklos Szeredi * 2540d58ffd35SMiklos Szeredi * Must be called with i_mutex held on parent. 2541d58ffd35SMiklos Szeredi * 25421acf0af9SDavid Howells * Returns 0 if the file was successfully atomically created (if necessary) and 25431acf0af9SDavid Howells * opened. In this case the file will be returned attached to @file. 25441acf0af9SDavid Howells * 25451acf0af9SDavid Howells * Returns 1 if the file was not completely opened at this time, though lookups 25461acf0af9SDavid Howells * and creations will have been performed and the dentry returned in @path will 25471acf0af9SDavid Howells * be positive upon return if O_CREAT was specified. If O_CREAT wasn't 25481acf0af9SDavid Howells * specified then a negative dentry may be returned. 25491acf0af9SDavid Howells * 25501acf0af9SDavid Howells * An error code is returned otherwise. 25511acf0af9SDavid Howells * 25521acf0af9SDavid Howells * FILE_CREATE will be set in @*opened if the dentry was created and will be 25531acf0af9SDavid Howells * cleared otherwise prior to returning. 2554d58ffd35SMiklos Szeredi */ 25552675a4ebSAl Viro static int lookup_open(struct nameidata *nd, struct path *path, 255630d90494SAl Viro struct file *file, 2557d58ffd35SMiklos Szeredi const struct open_flags *op, 255864894cf8SAl Viro bool got_write, int *opened) 2559d58ffd35SMiklos Szeredi { 2560d58ffd35SMiklos Szeredi struct dentry *dir = nd->path.dentry; 256154ef4872SMiklos Szeredi struct inode *dir_inode = dir->d_inode; 2562d58ffd35SMiklos Szeredi struct dentry *dentry; 2563d58ffd35SMiklos Szeredi int error; 256454ef4872SMiklos Szeredi bool need_lookup; 2565d58ffd35SMiklos Szeredi 256647237687SAl Viro *opened &= ~FILE_CREATED; 2567201f956eSAl Viro dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup); 2568d58ffd35SMiklos Szeredi if (IS_ERR(dentry)) 25692675a4ebSAl Viro return PTR_ERR(dentry); 2570d58ffd35SMiklos Szeredi 2571d18e9008SMiklos Szeredi /* Cached positive dentry: will open in f_op->open */ 2572d18e9008SMiklos Szeredi if (!need_lookup && dentry->d_inode) 2573d18e9008SMiklos Szeredi goto out_no_open; 2574d18e9008SMiklos Szeredi 2575d18e9008SMiklos Szeredi if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) { 257664894cf8SAl Viro return atomic_open(nd, dentry, path, file, op, got_write, 257747237687SAl Viro need_lookup, opened); 2578d18e9008SMiklos Szeredi } 2579d18e9008SMiklos Szeredi 258054ef4872SMiklos Szeredi if (need_lookup) { 258154ef4872SMiklos Szeredi BUG_ON(dentry->d_inode); 258254ef4872SMiklos Szeredi 258372bd866aSAl Viro dentry = lookup_real(dir_inode, dentry, nd->flags); 258454ef4872SMiklos Szeredi if (IS_ERR(dentry)) 25852675a4ebSAl Viro return PTR_ERR(dentry); 258654ef4872SMiklos Szeredi } 258754ef4872SMiklos Szeredi 2588d58ffd35SMiklos Szeredi /* Negative dentry, just create the file */ 2589d58ffd35SMiklos Szeredi if (!dentry->d_inode && (op->open_flag & O_CREAT)) { 2590d58ffd35SMiklos Szeredi umode_t mode = op->mode; 2591d58ffd35SMiklos Szeredi if (!IS_POSIXACL(dir->d_inode)) 2592d58ffd35SMiklos Szeredi mode &= ~current_umask(); 2593d58ffd35SMiklos Szeredi /* 2594d58ffd35SMiklos Szeredi * This write is needed to ensure that a 2595d58ffd35SMiklos Szeredi * rw->ro transition does not occur between 2596d58ffd35SMiklos Szeredi * the time when the file is created and when 2597d58ffd35SMiklos Szeredi * a permanent write count is taken through 2598015c3bbcSMiklos Szeredi * the 'struct file' in finish_open(). 2599d58ffd35SMiklos Szeredi */ 260064894cf8SAl Viro if (!got_write) { 260164894cf8SAl Viro error = -EROFS; 2602d58ffd35SMiklos Szeredi goto out_dput; 260364894cf8SAl Viro } 260447237687SAl Viro *opened |= FILE_CREATED; 2605d58ffd35SMiklos Szeredi error = security_path_mknod(&nd->path, dentry, mode, 0); 2606d58ffd35SMiklos Szeredi if (error) 2607d58ffd35SMiklos Szeredi goto out_dput; 2608312b63fbSAl Viro error = vfs_create(dir->d_inode, dentry, mode, 2609312b63fbSAl Viro nd->flags & LOOKUP_EXCL); 2610d58ffd35SMiklos Szeredi if (error) 2611d58ffd35SMiklos Szeredi goto out_dput; 2612d58ffd35SMiklos Szeredi } 2613d18e9008SMiklos Szeredi out_no_open: 2614d58ffd35SMiklos Szeredi path->dentry = dentry; 2615d58ffd35SMiklos Szeredi path->mnt = nd->path.mnt; 26162675a4ebSAl Viro return 1; 2617d58ffd35SMiklos Szeredi 2618d58ffd35SMiklos Szeredi out_dput: 2619d58ffd35SMiklos Szeredi dput(dentry); 26202675a4ebSAl Viro return error; 2621d58ffd35SMiklos Szeredi } 2622d58ffd35SMiklos Szeredi 2623d58ffd35SMiklos Szeredi /* 2624fe2d35ffSAl Viro * Handle the last step of open() 262531e6b01fSNick Piggin */ 26262675a4ebSAl Viro static int do_last(struct nameidata *nd, struct path *path, 262730d90494SAl Viro struct file *file, const struct open_flags *op, 262847237687SAl Viro int *opened, const char *pathname) 2629fb1cc555SAl Viro { 2630a1e28038SAl Viro struct dentry *dir = nd->path.dentry; 2631ca344a89SAl Viro int open_flag = op->open_flag; 263277d660a8SMiklos Szeredi bool will_truncate = (open_flag & O_TRUNC) != 0; 263364894cf8SAl Viro bool got_write = false; 2634bcda7652SAl Viro int acc_mode = op->acc_mode; 2635a1eb3315SMiklos Szeredi struct inode *inode; 263677d660a8SMiklos Szeredi bool symlink_ok = false; 263716b1c1cdSMiklos Szeredi struct path save_parent = { .dentry = NULL, .mnt = NULL }; 263816b1c1cdSMiklos Szeredi bool retried = false; 263916c2cd71SAl Viro int error; 2640fb1cc555SAl Viro 2641c3e380b0SAl Viro nd->flags &= ~LOOKUP_PARENT; 2642c3e380b0SAl Viro nd->flags |= op->intent; 2643c3e380b0SAl Viro 26441f36f774SAl Viro switch (nd->last_type) { 26451f36f774SAl Viro case LAST_DOTDOT: 2646176306f5SNeil Brown case LAST_DOT: 2647fe2d35ffSAl Viro error = handle_dots(nd, nd->last_type); 2648fe2d35ffSAl Viro if (error) 26492675a4ebSAl Viro return error; 26501f36f774SAl Viro /* fallthrough */ 26511f36f774SAl Viro case LAST_ROOT: 26529f1fafeeSAl Viro error = complete_walk(nd); 265316c2cd71SAl Viro if (error) 26542675a4ebSAl Viro return error; 2655fe2d35ffSAl Viro audit_inode(pathname, nd->path.dentry); 2656ca344a89SAl Viro if (open_flag & O_CREAT) { 265716c2cd71SAl Viro error = -EISDIR; 26582675a4ebSAl Viro goto out; 2659fe2d35ffSAl Viro } 2660e83db167SMiklos Szeredi goto finish_open; 26611f36f774SAl Viro case LAST_BIND: 26629f1fafeeSAl Viro error = complete_walk(nd); 266316c2cd71SAl Viro if (error) 26642675a4ebSAl Viro return error; 26651f36f774SAl Viro audit_inode(pathname, dir); 2666e83db167SMiklos Szeredi goto finish_open; 26671f36f774SAl Viro } 2668a2c36b45SAl Viro 2669ca344a89SAl Viro if (!(open_flag & O_CREAT)) { 2670fe2d35ffSAl Viro if (nd->last.name[nd->last.len]) 2671fe2d35ffSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 2672bcda7652SAl Viro if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW)) 267377d660a8SMiklos Szeredi symlink_ok = true; 2674fe2d35ffSAl Viro /* we _can_ be in RCU mode here */ 2675a1eb3315SMiklos Szeredi error = lookup_fast(nd, &nd->last, path, &inode); 267671574865SMiklos Szeredi if (likely(!error)) 267771574865SMiklos Szeredi goto finish_lookup; 267871574865SMiklos Szeredi 2679ce57dfc1SAl Viro if (error < 0) 26802675a4ebSAl Viro goto out; 2681a1eb3315SMiklos Szeredi 268237d7fffcSMiklos Szeredi BUG_ON(nd->inode != dir->d_inode); 2683b6183df7SMiklos Szeredi } else { 2684fe2d35ffSAl Viro /* create side of things */ 2685a3fbbde7SAl Viro /* 2686b6183df7SMiklos Szeredi * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED 2687b6183df7SMiklos Szeredi * has been cleared when we got to the last component we are 2688b6183df7SMiklos Szeredi * about to look up 2689a3fbbde7SAl Viro */ 26909f1fafeeSAl Viro error = complete_walk(nd); 26919f1fafeeSAl Viro if (error) 26922675a4ebSAl Viro return error; 2693fe2d35ffSAl Viro 2694fe2d35ffSAl Viro audit_inode(pathname, dir); 269516c2cd71SAl Viro error = -EISDIR; 26961f36f774SAl Viro /* trailing slashes? */ 269731e6b01fSNick Piggin if (nd->last.name[nd->last.len]) 26982675a4ebSAl Viro goto out; 2699b6183df7SMiklos Szeredi } 27001f36f774SAl Viro 270116b1c1cdSMiklos Szeredi retry_lookup: 270264894cf8SAl Viro if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { 270364894cf8SAl Viro error = mnt_want_write(nd->path.mnt); 270464894cf8SAl Viro if (!error) 270564894cf8SAl Viro got_write = true; 270664894cf8SAl Viro /* 270764894cf8SAl Viro * do _not_ fail yet - we might not need that or fail with 270864894cf8SAl Viro * a different error; let lookup_open() decide; we'll be 270964894cf8SAl Viro * dropping this one anyway. 271064894cf8SAl Viro */ 271164894cf8SAl Viro } 2712a1e28038SAl Viro mutex_lock(&dir->d_inode->i_mutex); 271364894cf8SAl Viro error = lookup_open(nd, path, file, op, got_write, opened); 2714fb1cc555SAl Viro mutex_unlock(&dir->d_inode->i_mutex); 2715fb1cc555SAl Viro 27162675a4ebSAl Viro if (error <= 0) { 27172675a4ebSAl Viro if (error) 2718d58ffd35SMiklos Szeredi goto out; 27196c0d46c4SAl Viro 272047237687SAl Viro if ((*opened & FILE_CREATED) || 27212675a4ebSAl Viro !S_ISREG(file->f_path.dentry->d_inode->i_mode)) 272277d660a8SMiklos Szeredi will_truncate = false; 2723d18e9008SMiklos Szeredi 27242675a4ebSAl Viro audit_inode(pathname, file->f_path.dentry); 2725d18e9008SMiklos Szeredi goto opened; 2726d18e9008SMiklos Szeredi } 2727d18e9008SMiklos Szeredi 272847237687SAl Viro if (*opened & FILE_CREATED) { 27299b44f1b3SAl Viro /* Don't check for write permission, don't truncate */ 2730ca344a89SAl Viro open_flag &= ~O_TRUNC; 273177d660a8SMiklos Szeredi will_truncate = false; 2732bcda7652SAl Viro acc_mode = MAY_OPEN; 2733d58ffd35SMiklos Szeredi path_to_nameidata(path, nd); 2734e83db167SMiklos Szeredi goto finish_open_created; 2735fb1cc555SAl Viro } 2736fb1cc555SAl Viro 2737fb1cc555SAl Viro /* 27383134f37eSJeff Layton * create/update audit record if it already exists. 2739fb1cc555SAl Viro */ 27403134f37eSJeff Layton if (path->dentry->d_inode) 2741fb1cc555SAl Viro audit_inode(pathname, path->dentry); 2742fb1cc555SAl Viro 2743d18e9008SMiklos Szeredi /* 2744d18e9008SMiklos Szeredi * If atomic_open() acquired write access it is dropped now due to 2745d18e9008SMiklos Szeredi * possible mount and symlink following (this might be optimized away if 2746d18e9008SMiklos Szeredi * necessary...) 2747d18e9008SMiklos Szeredi */ 274864894cf8SAl Viro if (got_write) { 2749d18e9008SMiklos Szeredi mnt_drop_write(nd->path.mnt); 275064894cf8SAl Viro got_write = false; 2751d18e9008SMiklos Szeredi } 2752d18e9008SMiklos Szeredi 2753fb1cc555SAl Viro error = -EEXIST; 2754f8310c59SAl Viro if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT)) 2755fb1cc555SAl Viro goto exit_dput; 2756fb1cc555SAl Viro 27579875cf80SDavid Howells error = follow_managed(path, nd->flags); 27589875cf80SDavid Howells if (error < 0) 2759fb1cc555SAl Viro goto exit_dput; 2760fb1cc555SAl Viro 2761a3fbbde7SAl Viro if (error) 2762a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 2763a3fbbde7SAl Viro 2764decf3400SMiklos Szeredi BUG_ON(nd->flags & LOOKUP_RCU); 2765decf3400SMiklos Szeredi inode = path->dentry->d_inode; 27665f5daac1SMiklos Szeredi finish_lookup: 27675f5daac1SMiklos Szeredi /* we _can_ be in RCU mode here */ 2768fb1cc555SAl Viro error = -ENOENT; 276954c33e7fSMiklos Szeredi if (!inode) { 277054c33e7fSMiklos Szeredi path_to_nameidata(path, nd); 27712675a4ebSAl Viro goto out; 277254c33e7fSMiklos Szeredi } 27739e67f361SAl Viro 2774d45ea867SMiklos Szeredi if (should_follow_link(inode, !symlink_ok)) { 2775d45ea867SMiklos Szeredi if (nd->flags & LOOKUP_RCU) { 2776d45ea867SMiklos Szeredi if (unlikely(unlazy_walk(nd, path->dentry))) { 2777d45ea867SMiklos Szeredi error = -ECHILD; 27782675a4ebSAl Viro goto out; 2779d45ea867SMiklos Szeredi } 2780d45ea867SMiklos Szeredi } 2781d45ea867SMiklos Szeredi BUG_ON(inode != path->dentry->d_inode); 27822675a4ebSAl Viro return 1; 2783d45ea867SMiklos Szeredi } 2784fb1cc555SAl Viro 278516b1c1cdSMiklos Szeredi if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) { 2786fb1cc555SAl Viro path_to_nameidata(path, nd); 278716b1c1cdSMiklos Szeredi } else { 278816b1c1cdSMiklos Szeredi save_parent.dentry = nd->path.dentry; 278916b1c1cdSMiklos Szeredi save_parent.mnt = mntget(path->mnt); 279016b1c1cdSMiklos Szeredi nd->path.dentry = path->dentry; 279116b1c1cdSMiklos Szeredi 279216b1c1cdSMiklos Szeredi } 2793decf3400SMiklos Szeredi nd->inode = inode; 2794a3fbbde7SAl Viro /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ 2795a3fbbde7SAl Viro error = complete_walk(nd); 279616b1c1cdSMiklos Szeredi if (error) { 279716b1c1cdSMiklos Szeredi path_put(&save_parent); 27982675a4ebSAl Viro return error; 279916b1c1cdSMiklos Szeredi } 2800fb1cc555SAl Viro error = -EISDIR; 2801050ac841SMiklos Szeredi if ((open_flag & O_CREAT) && S_ISDIR(nd->inode->i_mode)) 28022675a4ebSAl Viro goto out; 2803af2f5542SMiklos Szeredi error = -ENOTDIR; 2804af2f5542SMiklos Szeredi if ((nd->flags & LOOKUP_DIRECTORY) && !nd->inode->i_op->lookup) 28052675a4ebSAl Viro goto out; 2806d7fdd7f6SMiklos Szeredi audit_inode(pathname, nd->path.dentry); 2807e83db167SMiklos Szeredi finish_open: 28086c0d46c4SAl Viro if (!S_ISREG(nd->inode->i_mode)) 280977d660a8SMiklos Szeredi will_truncate = false; 28106c0d46c4SAl Viro 28110f9d1a10SAl Viro if (will_truncate) { 28120f9d1a10SAl Viro error = mnt_want_write(nd->path.mnt); 28130f9d1a10SAl Viro if (error) 28142675a4ebSAl Viro goto out; 281564894cf8SAl Viro got_write = true; 28160f9d1a10SAl Viro } 2817e83db167SMiklos Szeredi finish_open_created: 2818bcda7652SAl Viro error = may_open(&nd->path, acc_mode, open_flag); 2819ca344a89SAl Viro if (error) 28202675a4ebSAl Viro goto out; 282130d90494SAl Viro file->f_path.mnt = nd->path.mnt; 282230d90494SAl Viro error = finish_open(file, nd->path.dentry, NULL, opened); 282330d90494SAl Viro if (error) { 282430d90494SAl Viro if (error == -EOPENSTALE) 2825f60dc3dbSMiklos Szeredi goto stale_open; 2826015c3bbcSMiklos Szeredi goto out; 2827f60dc3dbSMiklos Szeredi } 2828a8277b9bSMiklos Szeredi opened: 28292675a4ebSAl Viro error = open_check_o_direct(file); 2830015c3bbcSMiklos Szeredi if (error) 2831015c3bbcSMiklos Szeredi goto exit_fput; 28322675a4ebSAl Viro error = ima_file_check(file, op->acc_mode); 2833aa4caadbSMiklos Szeredi if (error) 2834aa4caadbSMiklos Szeredi goto exit_fput; 2835aa4caadbSMiklos Szeredi 28360f9d1a10SAl Viro if (will_truncate) { 28372675a4ebSAl Viro error = handle_truncate(file); 2838aa4caadbSMiklos Szeredi if (error) 2839aa4caadbSMiklos Szeredi goto exit_fput; 28400f9d1a10SAl Viro } 2841ca344a89SAl Viro out: 284264894cf8SAl Viro if (got_write) 28430f9d1a10SAl Viro mnt_drop_write(nd->path.mnt); 284416b1c1cdSMiklos Szeredi path_put(&save_parent); 2845e276ae67SMiklos Szeredi terminate_walk(nd); 28462675a4ebSAl Viro return error; 2847fb1cc555SAl Viro 2848fb1cc555SAl Viro exit_dput: 2849fb1cc555SAl Viro path_put_conditional(path, nd); 2850ca344a89SAl Viro goto out; 2851015c3bbcSMiklos Szeredi exit_fput: 28522675a4ebSAl Viro fput(file); 28532675a4ebSAl Viro goto out; 2854015c3bbcSMiklos Szeredi 2855f60dc3dbSMiklos Szeredi stale_open: 2856f60dc3dbSMiklos Szeredi /* If no saved parent or already retried then can't retry */ 2857f60dc3dbSMiklos Szeredi if (!save_parent.dentry || retried) 2858f60dc3dbSMiklos Szeredi goto out; 2859f60dc3dbSMiklos Szeredi 2860f60dc3dbSMiklos Szeredi BUG_ON(save_parent.dentry != dir); 2861f60dc3dbSMiklos Szeredi path_put(&nd->path); 2862f60dc3dbSMiklos Szeredi nd->path = save_parent; 2863f60dc3dbSMiklos Szeredi nd->inode = dir->d_inode; 2864f60dc3dbSMiklos Szeredi save_parent.mnt = NULL; 2865f60dc3dbSMiklos Szeredi save_parent.dentry = NULL; 286664894cf8SAl Viro if (got_write) { 2867f60dc3dbSMiklos Szeredi mnt_drop_write(nd->path.mnt); 286864894cf8SAl Viro got_write = false; 2869f60dc3dbSMiklos Szeredi } 2870f60dc3dbSMiklos Szeredi retried = true; 2871f60dc3dbSMiklos Szeredi goto retry_lookup; 2872fb1cc555SAl Viro } 2873fb1cc555SAl Viro 287413aab428SAl Viro static struct file *path_openat(int dfd, const char *pathname, 287573d049a4SAl Viro struct nameidata *nd, const struct open_flags *op, int flags) 28761da177e4SLinus Torvalds { 2877fe2d35ffSAl Viro struct file *base = NULL; 287830d90494SAl Viro struct file *file; 28799850c056SAl Viro struct path path; 288047237687SAl Viro int opened = 0; 288113aab428SAl Viro int error; 288231e6b01fSNick Piggin 288330d90494SAl Viro file = get_empty_filp(); 288430d90494SAl Viro if (!file) 288531e6b01fSNick Piggin return ERR_PTR(-ENFILE); 288631e6b01fSNick Piggin 288730d90494SAl Viro file->f_flags = op->open_flag; 288831e6b01fSNick Piggin 288973d049a4SAl Viro error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base); 289031e6b01fSNick Piggin if (unlikely(error)) 28912675a4ebSAl Viro goto out; 289231e6b01fSNick Piggin 2893fe2d35ffSAl Viro current->total_link_count = 0; 289473d049a4SAl Viro error = link_path_walk(pathname, nd); 289531e6b01fSNick Piggin if (unlikely(error)) 28962675a4ebSAl Viro goto out; 28971da177e4SLinus Torvalds 28982675a4ebSAl Viro error = do_last(nd, &path, file, op, &opened, pathname); 28992675a4ebSAl Viro while (unlikely(error > 0)) { /* trailing symlink */ 29007b9337aaSNick Piggin struct path link = path; 2901def4af30SAl Viro void *cookie; 2902574197e0SAl Viro if (!(nd->flags & LOOKUP_FOLLOW)) { 290373d049a4SAl Viro path_put_conditional(&path, nd); 290473d049a4SAl Viro path_put(&nd->path); 29052675a4ebSAl Viro error = -ELOOP; 290640b39136SAl Viro break; 290740b39136SAl Viro } 2908800179c9SKees Cook error = may_follow_link(&link, nd); 2909800179c9SKees Cook if (unlikely(error)) 2910800179c9SKees Cook break; 291173d049a4SAl Viro nd->flags |= LOOKUP_PARENT; 291273d049a4SAl Viro nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); 2913574197e0SAl Viro error = follow_link(&link, nd, &cookie); 2914c3e380b0SAl Viro if (unlikely(error)) 29152675a4ebSAl Viro break; 29162675a4ebSAl Viro error = do_last(nd, &path, file, op, &opened, pathname); 2917574197e0SAl Viro put_link(nd, &link, cookie); 2918806b681cSAl Viro } 291910fa8e62SAl Viro out: 292073d049a4SAl Viro if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) 292173d049a4SAl Viro path_put(&nd->root); 2922fe2d35ffSAl Viro if (base) 2923fe2d35ffSAl Viro fput(base); 29242675a4ebSAl Viro if (!(opened & FILE_OPENED)) { 29252675a4ebSAl Viro BUG_ON(!error); 292630d90494SAl Viro put_filp(file); 2927015c3bbcSMiklos Szeredi } 29282675a4ebSAl Viro if (unlikely(error)) { 29292675a4ebSAl Viro if (error == -EOPENSTALE) { 29302675a4ebSAl Viro if (flags & LOOKUP_RCU) 29312675a4ebSAl Viro error = -ECHILD; 29322675a4ebSAl Viro else 29332675a4ebSAl Viro error = -ESTALE; 29342675a4ebSAl Viro } 29352675a4ebSAl Viro file = ERR_PTR(error); 29362675a4ebSAl Viro } 29372675a4ebSAl Viro return file; 2938de459215SKirill Korotaev } 29391da177e4SLinus Torvalds 294013aab428SAl Viro struct file *do_filp_open(int dfd, const char *pathname, 294113aab428SAl Viro const struct open_flags *op, int flags) 294213aab428SAl Viro { 294373d049a4SAl Viro struct nameidata nd; 294413aab428SAl Viro struct file *filp; 294513aab428SAl Viro 294673d049a4SAl Viro filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU); 294713aab428SAl Viro if (unlikely(filp == ERR_PTR(-ECHILD))) 294873d049a4SAl Viro filp = path_openat(dfd, pathname, &nd, op, flags); 294913aab428SAl Viro if (unlikely(filp == ERR_PTR(-ESTALE))) 295073d049a4SAl Viro filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL); 295113aab428SAl Viro return filp; 295213aab428SAl Viro } 295313aab428SAl Viro 295473d049a4SAl Viro struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, 295573d049a4SAl Viro const char *name, const struct open_flags *op, int flags) 295673d049a4SAl Viro { 295773d049a4SAl Viro struct nameidata nd; 295873d049a4SAl Viro struct file *file; 295973d049a4SAl Viro 296073d049a4SAl Viro nd.root.mnt = mnt; 296173d049a4SAl Viro nd.root.dentry = dentry; 296273d049a4SAl Viro 296373d049a4SAl Viro flags |= LOOKUP_ROOT; 296473d049a4SAl Viro 2965bcda7652SAl Viro if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN) 296673d049a4SAl Viro return ERR_PTR(-ELOOP); 296773d049a4SAl Viro 296873d049a4SAl Viro file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU); 296973d049a4SAl Viro if (unlikely(file == ERR_PTR(-ECHILD))) 297073d049a4SAl Viro file = path_openat(-1, name, &nd, op, flags); 297173d049a4SAl Viro if (unlikely(file == ERR_PTR(-ESTALE))) 297273d049a4SAl Viro file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL); 297373d049a4SAl Viro return file; 297473d049a4SAl Viro } 297573d049a4SAl Viro 2976ed75e95dSAl Viro struct dentry *kern_path_create(int dfd, const char *pathname, struct path *path, int is_dir) 29771da177e4SLinus Torvalds { 2978c663e5d8SChristoph Hellwig struct dentry *dentry = ERR_PTR(-EEXIST); 2979ed75e95dSAl Viro struct nameidata nd; 2980c30dabfeSJan Kara int err2; 2981ed75e95dSAl Viro int error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd); 2982ed75e95dSAl Viro if (error) 2983ed75e95dSAl Viro return ERR_PTR(error); 29841da177e4SLinus Torvalds 2985c663e5d8SChristoph Hellwig /* 2986c663e5d8SChristoph Hellwig * Yucky last component or no last component at all? 2987c663e5d8SChristoph Hellwig * (foo/., foo/.., /////) 2988c663e5d8SChristoph Hellwig */ 2989ed75e95dSAl Viro if (nd.last_type != LAST_NORM) 2990ed75e95dSAl Viro goto out; 2991ed75e95dSAl Viro nd.flags &= ~LOOKUP_PARENT; 2992ed75e95dSAl Viro nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL; 2993c663e5d8SChristoph Hellwig 2994c30dabfeSJan Kara /* don't fail immediately if it's r/o, at least try to report other errors */ 2995c30dabfeSJan Kara err2 = mnt_want_write(nd.path.mnt); 2996c663e5d8SChristoph Hellwig /* 2997c663e5d8SChristoph Hellwig * Do the final lookup. 2998c663e5d8SChristoph Hellwig */ 2999ed75e95dSAl Viro mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 3000ed75e95dSAl Viro dentry = lookup_hash(&nd); 30011da177e4SLinus Torvalds if (IS_ERR(dentry)) 3002a8104a9fSAl Viro goto unlock; 3003c663e5d8SChristoph Hellwig 3004a8104a9fSAl Viro error = -EEXIST; 3005e9baf6e5SAl Viro if (dentry->d_inode) 3006a8104a9fSAl Viro goto fail; 3007c663e5d8SChristoph Hellwig /* 3008c663e5d8SChristoph Hellwig * Special case - lookup gave negative, but... we had foo/bar/ 3009c663e5d8SChristoph Hellwig * From the vfs_mknod() POV we just have a negative dentry - 3010c663e5d8SChristoph Hellwig * all is fine. Let's be bastards - you had / on the end, you've 3011c663e5d8SChristoph Hellwig * been asking for (non-existent) directory. -ENOENT for you. 3012c663e5d8SChristoph Hellwig */ 3013ed75e95dSAl Viro if (unlikely(!is_dir && nd.last.name[nd.last.len])) { 3014a8104a9fSAl Viro error = -ENOENT; 3015ed75e95dSAl Viro goto fail; 3016e9baf6e5SAl Viro } 3017c30dabfeSJan Kara if (unlikely(err2)) { 3018c30dabfeSJan Kara error = err2; 3019a8104a9fSAl Viro goto fail; 3020c30dabfeSJan Kara } 3021ed75e95dSAl Viro *path = nd.path; 3022e9baf6e5SAl Viro return dentry; 30231da177e4SLinus Torvalds fail: 3024a8104a9fSAl Viro dput(dentry); 3025a8104a9fSAl Viro dentry = ERR_PTR(error); 3026a8104a9fSAl Viro unlock: 3027dae6ad8fSAl Viro mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 3028c30dabfeSJan Kara if (!err2) 3029c30dabfeSJan Kara mnt_drop_write(nd.path.mnt); 3030ed75e95dSAl Viro out: 3031dae6ad8fSAl Viro path_put(&nd.path); 3032ed75e95dSAl Viro return dentry; 3033dae6ad8fSAl Viro } 3034dae6ad8fSAl Viro EXPORT_SYMBOL(kern_path_create); 3035dae6ad8fSAl Viro 3036921a1650SAl Viro void done_path_create(struct path *path, struct dentry *dentry) 3037921a1650SAl Viro { 3038921a1650SAl Viro dput(dentry); 3039921a1650SAl Viro mutex_unlock(&path->dentry->d_inode->i_mutex); 3040a8104a9fSAl Viro mnt_drop_write(path->mnt); 3041921a1650SAl Viro path_put(path); 3042921a1650SAl Viro } 3043921a1650SAl Viro EXPORT_SYMBOL(done_path_create); 3044921a1650SAl Viro 3045dae6ad8fSAl Viro struct dentry *user_path_create(int dfd, const char __user *pathname, struct path *path, int is_dir) 3046dae6ad8fSAl Viro { 3047dae6ad8fSAl Viro char *tmp = getname(pathname); 3048dae6ad8fSAl Viro struct dentry *res; 3049dae6ad8fSAl Viro if (IS_ERR(tmp)) 3050dae6ad8fSAl Viro return ERR_CAST(tmp); 3051dae6ad8fSAl Viro res = kern_path_create(dfd, tmp, path, is_dir); 3052dae6ad8fSAl Viro putname(tmp); 3053dae6ad8fSAl Viro return res; 3054dae6ad8fSAl Viro } 3055dae6ad8fSAl Viro EXPORT_SYMBOL(user_path_create); 3056dae6ad8fSAl Viro 30571a67aafbSAl Viro int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) 30581da177e4SLinus Torvalds { 3059a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 30601da177e4SLinus Torvalds 30611da177e4SLinus Torvalds if (error) 30621da177e4SLinus Torvalds return error; 30631da177e4SLinus Torvalds 3064975d6b39SEric W. Biederman if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) 30651da177e4SLinus Torvalds return -EPERM; 30661da177e4SLinus Torvalds 3067acfa4380SAl Viro if (!dir->i_op->mknod) 30681da177e4SLinus Torvalds return -EPERM; 30691da177e4SLinus Torvalds 307008ce5f16SSerge E. Hallyn error = devcgroup_inode_mknod(mode, dev); 307108ce5f16SSerge E. Hallyn if (error) 307208ce5f16SSerge E. Hallyn return error; 307308ce5f16SSerge E. Hallyn 30741da177e4SLinus Torvalds error = security_inode_mknod(dir, dentry, mode, dev); 30751da177e4SLinus Torvalds if (error) 30761da177e4SLinus Torvalds return error; 30771da177e4SLinus Torvalds 30781da177e4SLinus Torvalds error = dir->i_op->mknod(dir, dentry, mode, dev); 3079a74574aaSStephen Smalley if (!error) 3080f38aa942SAmy Griffis fsnotify_create(dir, dentry); 30811da177e4SLinus Torvalds return error; 30821da177e4SLinus Torvalds } 30831da177e4SLinus Torvalds 3084f69aac00SAl Viro static int may_mknod(umode_t mode) 3085463c3197SDave Hansen { 3086463c3197SDave Hansen switch (mode & S_IFMT) { 3087463c3197SDave Hansen case S_IFREG: 3088463c3197SDave Hansen case S_IFCHR: 3089463c3197SDave Hansen case S_IFBLK: 3090463c3197SDave Hansen case S_IFIFO: 3091463c3197SDave Hansen case S_IFSOCK: 3092463c3197SDave Hansen case 0: /* zero mode translates to S_IFREG */ 3093463c3197SDave Hansen return 0; 3094463c3197SDave Hansen case S_IFDIR: 3095463c3197SDave Hansen return -EPERM; 3096463c3197SDave Hansen default: 3097463c3197SDave Hansen return -EINVAL; 3098463c3197SDave Hansen } 3099463c3197SDave Hansen } 3100463c3197SDave Hansen 31018208a22bSAl Viro SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, 31022e4d0924SHeiko Carstens unsigned, dev) 31031da177e4SLinus Torvalds { 31041da177e4SLinus Torvalds struct dentry *dentry; 3105dae6ad8fSAl Viro struct path path; 3106dae6ad8fSAl Viro int error; 31071da177e4SLinus Torvalds 31088e4bfca1SAl Viro error = may_mknod(mode); 31098e4bfca1SAl Viro if (error) 31108e4bfca1SAl Viro return error; 31111da177e4SLinus Torvalds 3112dae6ad8fSAl Viro dentry = user_path_create(dfd, filename, &path, 0); 3113dae6ad8fSAl Viro if (IS_ERR(dentry)) 3114dae6ad8fSAl Viro return PTR_ERR(dentry); 31152ad94ae6SAl Viro 3116dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3117ce3b0f8dSAl Viro mode &= ~current_umask(); 3118dae6ad8fSAl Viro error = security_path_mknod(&path, dentry, mode, dev); 3119be6d3e56SKentaro Takeda if (error) 3120a8104a9fSAl Viro goto out; 31211da177e4SLinus Torvalds switch (mode & S_IFMT) { 31221da177e4SLinus Torvalds case 0: case S_IFREG: 3123312b63fbSAl Viro error = vfs_create(path.dentry->d_inode,dentry,mode,true); 31241da177e4SLinus Torvalds break; 31251da177e4SLinus Torvalds case S_IFCHR: case S_IFBLK: 3126dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode, 31271da177e4SLinus Torvalds new_decode_dev(dev)); 31281da177e4SLinus Torvalds break; 31291da177e4SLinus Torvalds case S_IFIFO: case S_IFSOCK: 3130dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); 31311da177e4SLinus Torvalds break; 31321da177e4SLinus Torvalds } 3133a8104a9fSAl Viro out: 3134921a1650SAl Viro done_path_create(&path, dentry); 31351da177e4SLinus Torvalds return error; 31361da177e4SLinus Torvalds } 31371da177e4SLinus Torvalds 31388208a22bSAl Viro SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) 31395590ff0dSUlrich Drepper { 31405590ff0dSUlrich Drepper return sys_mknodat(AT_FDCWD, filename, mode, dev); 31415590ff0dSUlrich Drepper } 31425590ff0dSUlrich Drepper 314318bb1db3SAl Viro int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 31441da177e4SLinus Torvalds { 3145a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 31468de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 31471da177e4SLinus Torvalds 31481da177e4SLinus Torvalds if (error) 31491da177e4SLinus Torvalds return error; 31501da177e4SLinus Torvalds 3151acfa4380SAl Viro if (!dir->i_op->mkdir) 31521da177e4SLinus Torvalds return -EPERM; 31531da177e4SLinus Torvalds 31541da177e4SLinus Torvalds mode &= (S_IRWXUGO|S_ISVTX); 31551da177e4SLinus Torvalds error = security_inode_mkdir(dir, dentry, mode); 31561da177e4SLinus Torvalds if (error) 31571da177e4SLinus Torvalds return error; 31581da177e4SLinus Torvalds 31598de52778SAl Viro if (max_links && dir->i_nlink >= max_links) 31608de52778SAl Viro return -EMLINK; 31618de52778SAl Viro 31621da177e4SLinus Torvalds error = dir->i_op->mkdir(dir, dentry, mode); 3163a74574aaSStephen Smalley if (!error) 3164f38aa942SAmy Griffis fsnotify_mkdir(dir, dentry); 31651da177e4SLinus Torvalds return error; 31661da177e4SLinus Torvalds } 31671da177e4SLinus Torvalds 3168a218d0fdSAl Viro SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) 31691da177e4SLinus Torvalds { 31706902d925SDave Hansen struct dentry *dentry; 3171dae6ad8fSAl Viro struct path path; 3172dae6ad8fSAl Viro int error; 31731da177e4SLinus Torvalds 3174dae6ad8fSAl Viro dentry = user_path_create(dfd, pathname, &path, 1); 31756902d925SDave Hansen if (IS_ERR(dentry)) 3176dae6ad8fSAl Viro return PTR_ERR(dentry); 31776902d925SDave Hansen 3178dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3179ce3b0f8dSAl Viro mode &= ~current_umask(); 3180dae6ad8fSAl Viro error = security_path_mkdir(&path, dentry, mode); 3181a8104a9fSAl Viro if (!error) 3182dae6ad8fSAl Viro error = vfs_mkdir(path.dentry->d_inode, dentry, mode); 3183921a1650SAl Viro done_path_create(&path, dentry); 31841da177e4SLinus Torvalds return error; 31851da177e4SLinus Torvalds } 31861da177e4SLinus Torvalds 3187a218d0fdSAl Viro SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) 31885590ff0dSUlrich Drepper { 31895590ff0dSUlrich Drepper return sys_mkdirat(AT_FDCWD, pathname, mode); 31905590ff0dSUlrich Drepper } 31915590ff0dSUlrich Drepper 31921da177e4SLinus Torvalds /* 3193a71905f0SSage Weil * The dentry_unhash() helper will try to drop the dentry early: we 3194c0d02594SJ. Bruce Fields * should have a usage count of 1 if we're the only user of this 3195a71905f0SSage Weil * dentry, and if that is true (possibly after pruning the dcache), 3196a71905f0SSage Weil * then we drop the dentry now. 31971da177e4SLinus Torvalds * 31981da177e4SLinus Torvalds * A low-level filesystem can, if it choses, legally 31991da177e4SLinus Torvalds * do a 32001da177e4SLinus Torvalds * 32011da177e4SLinus Torvalds * if (!d_unhashed(dentry)) 32021da177e4SLinus Torvalds * return -EBUSY; 32031da177e4SLinus Torvalds * 32041da177e4SLinus Torvalds * if it cannot handle the case of removing a directory 32051da177e4SLinus Torvalds * that is still in use by something else.. 32061da177e4SLinus Torvalds */ 32071da177e4SLinus Torvalds void dentry_unhash(struct dentry *dentry) 32081da177e4SLinus Torvalds { 32091da177e4SLinus Torvalds shrink_dcache_parent(dentry); 32101da177e4SLinus Torvalds spin_lock(&dentry->d_lock); 321164252c75SSage Weil if (dentry->d_count == 1) 32121da177e4SLinus Torvalds __d_drop(dentry); 32131da177e4SLinus Torvalds spin_unlock(&dentry->d_lock); 32141da177e4SLinus Torvalds } 32151da177e4SLinus Torvalds 32161da177e4SLinus Torvalds int vfs_rmdir(struct inode *dir, struct dentry *dentry) 32171da177e4SLinus Torvalds { 32181da177e4SLinus Torvalds int error = may_delete(dir, dentry, 1); 32191da177e4SLinus Torvalds 32201da177e4SLinus Torvalds if (error) 32211da177e4SLinus Torvalds return error; 32221da177e4SLinus Torvalds 3223acfa4380SAl Viro if (!dir->i_op->rmdir) 32241da177e4SLinus Torvalds return -EPERM; 32251da177e4SLinus Torvalds 32261d2ef590SAl Viro dget(dentry); 32271b1dcc1bSJes Sorensen mutex_lock(&dentry->d_inode->i_mutex); 3228912dbc15SSage Weil 32291da177e4SLinus Torvalds error = -EBUSY; 3230912dbc15SSage Weil if (d_mountpoint(dentry)) 3231912dbc15SSage Weil goto out; 3232912dbc15SSage Weil 32331da177e4SLinus Torvalds error = security_inode_rmdir(dir, dentry); 3234912dbc15SSage Weil if (error) 3235912dbc15SSage Weil goto out; 3236912dbc15SSage Weil 32373cebde24SSage Weil shrink_dcache_parent(dentry); 32381da177e4SLinus Torvalds error = dir->i_op->rmdir(dir, dentry); 3239912dbc15SSage Weil if (error) 3240912dbc15SSage Weil goto out; 3241912dbc15SSage Weil 32421da177e4SLinus Torvalds dentry->d_inode->i_flags |= S_DEAD; 3243d83c49f3SAl Viro dont_mount(dentry); 32441da177e4SLinus Torvalds 3245912dbc15SSage Weil out: 3246912dbc15SSage Weil mutex_unlock(&dentry->d_inode->i_mutex); 32471d2ef590SAl Viro dput(dentry); 3248912dbc15SSage Weil if (!error) 3249912dbc15SSage Weil d_delete(dentry); 32501da177e4SLinus Torvalds return error; 32511da177e4SLinus Torvalds } 32521da177e4SLinus Torvalds 32535590ff0dSUlrich Drepper static long do_rmdir(int dfd, const char __user *pathname) 32541da177e4SLinus Torvalds { 32551da177e4SLinus Torvalds int error = 0; 32561da177e4SLinus Torvalds char * name; 32571da177e4SLinus Torvalds struct dentry *dentry; 32581da177e4SLinus Torvalds struct nameidata nd; 32591da177e4SLinus Torvalds 32602ad94ae6SAl Viro error = user_path_parent(dfd, pathname, &nd, &name); 32611da177e4SLinus Torvalds if (error) 32622ad94ae6SAl Viro return error; 32631da177e4SLinus Torvalds 32641da177e4SLinus Torvalds switch(nd.last_type) { 32651da177e4SLinus Torvalds case LAST_DOTDOT: 32661da177e4SLinus Torvalds error = -ENOTEMPTY; 32671da177e4SLinus Torvalds goto exit1; 32681da177e4SLinus Torvalds case LAST_DOT: 32691da177e4SLinus Torvalds error = -EINVAL; 32701da177e4SLinus Torvalds goto exit1; 32711da177e4SLinus Torvalds case LAST_ROOT: 32721da177e4SLinus Torvalds error = -EBUSY; 32731da177e4SLinus Torvalds goto exit1; 32741da177e4SLinus Torvalds } 32750612d9fbSOGAWA Hirofumi 32760612d9fbSOGAWA Hirofumi nd.flags &= ~LOOKUP_PARENT; 3277c30dabfeSJan Kara error = mnt_want_write(nd.path.mnt); 3278c30dabfeSJan Kara if (error) 3279c30dabfeSJan Kara goto exit1; 32800612d9fbSOGAWA Hirofumi 32814ac91378SJan Blunck mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 328249705b77SChristoph Hellwig dentry = lookup_hash(&nd); 32831da177e4SLinus Torvalds error = PTR_ERR(dentry); 32846902d925SDave Hansen if (IS_ERR(dentry)) 32856902d925SDave Hansen goto exit2; 3286e6bc45d6STheodore Ts'o if (!dentry->d_inode) { 3287e6bc45d6STheodore Ts'o error = -ENOENT; 3288e6bc45d6STheodore Ts'o goto exit3; 3289e6bc45d6STheodore Ts'o } 3290be6d3e56SKentaro Takeda error = security_path_rmdir(&nd.path, dentry); 3291be6d3e56SKentaro Takeda if (error) 3292c30dabfeSJan Kara goto exit3; 32934ac91378SJan Blunck error = vfs_rmdir(nd.path.dentry->d_inode, dentry); 32940622753bSDave Hansen exit3: 32951da177e4SLinus Torvalds dput(dentry); 32966902d925SDave Hansen exit2: 32974ac91378SJan Blunck mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 3298c30dabfeSJan Kara mnt_drop_write(nd.path.mnt); 32991da177e4SLinus Torvalds exit1: 33001d957f9bSJan Blunck path_put(&nd.path); 33011da177e4SLinus Torvalds putname(name); 33021da177e4SLinus Torvalds return error; 33031da177e4SLinus Torvalds } 33041da177e4SLinus Torvalds 33053cdad428SHeiko Carstens SYSCALL_DEFINE1(rmdir, const char __user *, pathname) 33065590ff0dSUlrich Drepper { 33075590ff0dSUlrich Drepper return do_rmdir(AT_FDCWD, pathname); 33085590ff0dSUlrich Drepper } 33095590ff0dSUlrich Drepper 33101da177e4SLinus Torvalds int vfs_unlink(struct inode *dir, struct dentry *dentry) 33111da177e4SLinus Torvalds { 33121da177e4SLinus Torvalds int error = may_delete(dir, dentry, 0); 33131da177e4SLinus Torvalds 33141da177e4SLinus Torvalds if (error) 33151da177e4SLinus Torvalds return error; 33161da177e4SLinus Torvalds 3317acfa4380SAl Viro if (!dir->i_op->unlink) 33181da177e4SLinus Torvalds return -EPERM; 33191da177e4SLinus Torvalds 33201b1dcc1bSJes Sorensen mutex_lock(&dentry->d_inode->i_mutex); 33211da177e4SLinus Torvalds if (d_mountpoint(dentry)) 33221da177e4SLinus Torvalds error = -EBUSY; 33231da177e4SLinus Torvalds else { 33241da177e4SLinus Torvalds error = security_inode_unlink(dir, dentry); 3325bec1052eSAl Viro if (!error) { 33261da177e4SLinus Torvalds error = dir->i_op->unlink(dir, dentry); 3327bec1052eSAl Viro if (!error) 3328d83c49f3SAl Viro dont_mount(dentry); 3329bec1052eSAl Viro } 33301da177e4SLinus Torvalds } 33311b1dcc1bSJes Sorensen mutex_unlock(&dentry->d_inode->i_mutex); 33321da177e4SLinus Torvalds 33331da177e4SLinus Torvalds /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 33341da177e4SLinus Torvalds if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { 3335ece95912SJan Kara fsnotify_link_count(dentry->d_inode); 33361da177e4SLinus Torvalds d_delete(dentry); 33371da177e4SLinus Torvalds } 33380eeca283SRobert Love 33391da177e4SLinus Torvalds return error; 33401da177e4SLinus Torvalds } 33411da177e4SLinus Torvalds 33421da177e4SLinus Torvalds /* 33431da177e4SLinus Torvalds * Make sure that the actual truncation of the file will occur outside its 33441b1dcc1bSJes Sorensen * directory's i_mutex. Truncate can take a long time if there is a lot of 33451da177e4SLinus Torvalds * writeout happening, and we don't want to prevent access to the directory 33461da177e4SLinus Torvalds * while waiting on the I/O. 33471da177e4SLinus Torvalds */ 33485590ff0dSUlrich Drepper static long do_unlinkat(int dfd, const char __user *pathname) 33491da177e4SLinus Torvalds { 33502ad94ae6SAl Viro int error; 33511da177e4SLinus Torvalds char *name; 33521da177e4SLinus Torvalds struct dentry *dentry; 33531da177e4SLinus Torvalds struct nameidata nd; 33541da177e4SLinus Torvalds struct inode *inode = NULL; 33551da177e4SLinus Torvalds 33562ad94ae6SAl Viro error = user_path_parent(dfd, pathname, &nd, &name); 33571da177e4SLinus Torvalds if (error) 33582ad94ae6SAl Viro return error; 33592ad94ae6SAl Viro 33601da177e4SLinus Torvalds error = -EISDIR; 33611da177e4SLinus Torvalds if (nd.last_type != LAST_NORM) 33621da177e4SLinus Torvalds goto exit1; 33630612d9fbSOGAWA Hirofumi 33640612d9fbSOGAWA Hirofumi nd.flags &= ~LOOKUP_PARENT; 3365c30dabfeSJan Kara error = mnt_want_write(nd.path.mnt); 3366c30dabfeSJan Kara if (error) 3367c30dabfeSJan Kara goto exit1; 33680612d9fbSOGAWA Hirofumi 33694ac91378SJan Blunck mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 337049705b77SChristoph Hellwig dentry = lookup_hash(&nd); 33711da177e4SLinus Torvalds error = PTR_ERR(dentry); 33721da177e4SLinus Torvalds if (!IS_ERR(dentry)) { 33731da177e4SLinus Torvalds /* Why not before? Because we want correct error value */ 337450338b88STörök Edwin if (nd.last.name[nd.last.len]) 337550338b88STörök Edwin goto slashes; 33761da177e4SLinus Torvalds inode = dentry->d_inode; 337750338b88STörök Edwin if (!inode) 3378e6bc45d6STheodore Ts'o goto slashes; 33797de9c6eeSAl Viro ihold(inode); 3380be6d3e56SKentaro Takeda error = security_path_unlink(&nd.path, dentry); 3381be6d3e56SKentaro Takeda if (error) 3382c30dabfeSJan Kara goto exit2; 33834ac91378SJan Blunck error = vfs_unlink(nd.path.dentry->d_inode, dentry); 33841da177e4SLinus Torvalds exit2: 33851da177e4SLinus Torvalds dput(dentry); 33861da177e4SLinus Torvalds } 33874ac91378SJan Blunck mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 33881da177e4SLinus Torvalds if (inode) 33891da177e4SLinus Torvalds iput(inode); /* truncate the inode here */ 3390c30dabfeSJan Kara mnt_drop_write(nd.path.mnt); 33911da177e4SLinus Torvalds exit1: 33921d957f9bSJan Blunck path_put(&nd.path); 33931da177e4SLinus Torvalds putname(name); 33941da177e4SLinus Torvalds return error; 33951da177e4SLinus Torvalds 33961da177e4SLinus Torvalds slashes: 33971da177e4SLinus Torvalds error = !dentry->d_inode ? -ENOENT : 33981da177e4SLinus Torvalds S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR; 33991da177e4SLinus Torvalds goto exit2; 34001da177e4SLinus Torvalds } 34011da177e4SLinus Torvalds 34022e4d0924SHeiko Carstens SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) 34035590ff0dSUlrich Drepper { 34045590ff0dSUlrich Drepper if ((flag & ~AT_REMOVEDIR) != 0) 34055590ff0dSUlrich Drepper return -EINVAL; 34065590ff0dSUlrich Drepper 34075590ff0dSUlrich Drepper if (flag & AT_REMOVEDIR) 34085590ff0dSUlrich Drepper return do_rmdir(dfd, pathname); 34095590ff0dSUlrich Drepper 34105590ff0dSUlrich Drepper return do_unlinkat(dfd, pathname); 34115590ff0dSUlrich Drepper } 34125590ff0dSUlrich Drepper 34133480b257SHeiko Carstens SYSCALL_DEFINE1(unlink, const char __user *, pathname) 34145590ff0dSUlrich Drepper { 34155590ff0dSUlrich Drepper return do_unlinkat(AT_FDCWD, pathname); 34165590ff0dSUlrich Drepper } 34175590ff0dSUlrich Drepper 3418db2e747bSMiklos Szeredi int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) 34191da177e4SLinus Torvalds { 3420a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 34211da177e4SLinus Torvalds 34221da177e4SLinus Torvalds if (error) 34231da177e4SLinus Torvalds return error; 34241da177e4SLinus Torvalds 3425acfa4380SAl Viro if (!dir->i_op->symlink) 34261da177e4SLinus Torvalds return -EPERM; 34271da177e4SLinus Torvalds 34281da177e4SLinus Torvalds error = security_inode_symlink(dir, dentry, oldname); 34291da177e4SLinus Torvalds if (error) 34301da177e4SLinus Torvalds return error; 34311da177e4SLinus Torvalds 34321da177e4SLinus Torvalds error = dir->i_op->symlink(dir, dentry, oldname); 3433a74574aaSStephen Smalley if (!error) 3434f38aa942SAmy Griffis fsnotify_create(dir, dentry); 34351da177e4SLinus Torvalds return error; 34361da177e4SLinus Torvalds } 34371da177e4SLinus Torvalds 34382e4d0924SHeiko Carstens SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, 34392e4d0924SHeiko Carstens int, newdfd, const char __user *, newname) 34401da177e4SLinus Torvalds { 34412ad94ae6SAl Viro int error; 34421da177e4SLinus Torvalds char *from; 34436902d925SDave Hansen struct dentry *dentry; 3444dae6ad8fSAl Viro struct path path; 34451da177e4SLinus Torvalds 34461da177e4SLinus Torvalds from = getname(oldname); 34471da177e4SLinus Torvalds if (IS_ERR(from)) 34481da177e4SLinus Torvalds return PTR_ERR(from); 34492ad94ae6SAl Viro 3450dae6ad8fSAl Viro dentry = user_path_create(newdfd, newname, &path, 0); 34511da177e4SLinus Torvalds error = PTR_ERR(dentry); 34526902d925SDave Hansen if (IS_ERR(dentry)) 3453dae6ad8fSAl Viro goto out_putname; 34546902d925SDave Hansen 3455dae6ad8fSAl Viro error = security_path_symlink(&path, dentry, from); 3456a8104a9fSAl Viro if (!error) 3457dae6ad8fSAl Viro error = vfs_symlink(path.dentry->d_inode, dentry, from); 3458921a1650SAl Viro done_path_create(&path, dentry); 34596902d925SDave Hansen out_putname: 34601da177e4SLinus Torvalds putname(from); 34611da177e4SLinus Torvalds return error; 34621da177e4SLinus Torvalds } 34631da177e4SLinus Torvalds 34643480b257SHeiko Carstens SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) 34655590ff0dSUlrich Drepper { 34665590ff0dSUlrich Drepper return sys_symlinkat(oldname, AT_FDCWD, newname); 34675590ff0dSUlrich Drepper } 34685590ff0dSUlrich Drepper 34691da177e4SLinus Torvalds int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry) 34701da177e4SLinus Torvalds { 34711da177e4SLinus Torvalds struct inode *inode = old_dentry->d_inode; 34728de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 34731da177e4SLinus Torvalds int error; 34741da177e4SLinus Torvalds 34751da177e4SLinus Torvalds if (!inode) 34761da177e4SLinus Torvalds return -ENOENT; 34771da177e4SLinus Torvalds 3478a95164d9SMiklos Szeredi error = may_create(dir, new_dentry); 34791da177e4SLinus Torvalds if (error) 34801da177e4SLinus Torvalds return error; 34811da177e4SLinus Torvalds 34821da177e4SLinus Torvalds if (dir->i_sb != inode->i_sb) 34831da177e4SLinus Torvalds return -EXDEV; 34841da177e4SLinus Torvalds 34851da177e4SLinus Torvalds /* 34861da177e4SLinus Torvalds * A link to an append-only or immutable file cannot be created. 34871da177e4SLinus Torvalds */ 34881da177e4SLinus Torvalds if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 34891da177e4SLinus Torvalds return -EPERM; 3490acfa4380SAl Viro if (!dir->i_op->link) 34911da177e4SLinus Torvalds return -EPERM; 34927e79eedbSTetsuo Handa if (S_ISDIR(inode->i_mode)) 34931da177e4SLinus Torvalds return -EPERM; 34941da177e4SLinus Torvalds 34951da177e4SLinus Torvalds error = security_inode_link(old_dentry, dir, new_dentry); 34961da177e4SLinus Torvalds if (error) 34971da177e4SLinus Torvalds return error; 34981da177e4SLinus Torvalds 34997e79eedbSTetsuo Handa mutex_lock(&inode->i_mutex); 3500aae8a97dSAneesh Kumar K.V /* Make sure we don't allow creating hardlink to an unlinked file */ 3501aae8a97dSAneesh Kumar K.V if (inode->i_nlink == 0) 3502aae8a97dSAneesh Kumar K.V error = -ENOENT; 35038de52778SAl Viro else if (max_links && inode->i_nlink >= max_links) 35048de52778SAl Viro error = -EMLINK; 3505aae8a97dSAneesh Kumar K.V else 35061da177e4SLinus Torvalds error = dir->i_op->link(old_dentry, dir, new_dentry); 35077e79eedbSTetsuo Handa mutex_unlock(&inode->i_mutex); 3508e31e14ecSStephen Smalley if (!error) 35097e79eedbSTetsuo Handa fsnotify_link(dir, inode, new_dentry); 35101da177e4SLinus Torvalds return error; 35111da177e4SLinus Torvalds } 35121da177e4SLinus Torvalds 35131da177e4SLinus Torvalds /* 35141da177e4SLinus Torvalds * Hardlinks are often used in delicate situations. We avoid 35151da177e4SLinus Torvalds * security-related surprises by not following symlinks on the 35161da177e4SLinus Torvalds * newname. --KAB 35171da177e4SLinus Torvalds * 35181da177e4SLinus Torvalds * We don't follow them on the oldname either to be compatible 35191da177e4SLinus Torvalds * with linux 2.0, and to avoid hard-linking to directories 35201da177e4SLinus Torvalds * and other special files. --ADM 35211da177e4SLinus Torvalds */ 35222e4d0924SHeiko Carstens SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, 35232e4d0924SHeiko Carstens int, newdfd, const char __user *, newname, int, flags) 35241da177e4SLinus Torvalds { 35251da177e4SLinus Torvalds struct dentry *new_dentry; 3526dae6ad8fSAl Viro struct path old_path, new_path; 352711a7b371SAneesh Kumar K.V int how = 0; 35281da177e4SLinus Torvalds int error; 35291da177e4SLinus Torvalds 353011a7b371SAneesh Kumar K.V if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) 3531c04030e1SUlrich Drepper return -EINVAL; 353211a7b371SAneesh Kumar K.V /* 353311a7b371SAneesh Kumar K.V * To use null names we require CAP_DAC_READ_SEARCH 353411a7b371SAneesh Kumar K.V * This ensures that not everyone will be able to create 353511a7b371SAneesh Kumar K.V * handlink using the passed filedescriptor. 353611a7b371SAneesh Kumar K.V */ 353711a7b371SAneesh Kumar K.V if (flags & AT_EMPTY_PATH) { 353811a7b371SAneesh Kumar K.V if (!capable(CAP_DAC_READ_SEARCH)) 353911a7b371SAneesh Kumar K.V return -ENOENT; 354011a7b371SAneesh Kumar K.V how = LOOKUP_EMPTY; 354111a7b371SAneesh Kumar K.V } 3542c04030e1SUlrich Drepper 354311a7b371SAneesh Kumar K.V if (flags & AT_SYMLINK_FOLLOW) 354411a7b371SAneesh Kumar K.V how |= LOOKUP_FOLLOW; 354511a7b371SAneesh Kumar K.V 354611a7b371SAneesh Kumar K.V error = user_path_at(olddfd, oldname, how, &old_path); 35471da177e4SLinus Torvalds if (error) 35482ad94ae6SAl Viro return error; 35492ad94ae6SAl Viro 3550dae6ad8fSAl Viro new_dentry = user_path_create(newdfd, newname, &new_path, 0); 35511da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 35526902d925SDave Hansen if (IS_ERR(new_dentry)) 3553dae6ad8fSAl Viro goto out; 3554dae6ad8fSAl Viro 3555dae6ad8fSAl Viro error = -EXDEV; 3556dae6ad8fSAl Viro if (old_path.mnt != new_path.mnt) 3557dae6ad8fSAl Viro goto out_dput; 3558800179c9SKees Cook error = may_linkat(&old_path); 3559800179c9SKees Cook if (unlikely(error)) 3560800179c9SKees Cook goto out_dput; 3561dae6ad8fSAl Viro error = security_path_link(old_path.dentry, &new_path, new_dentry); 3562be6d3e56SKentaro Takeda if (error) 3563a8104a9fSAl Viro goto out_dput; 3564dae6ad8fSAl Viro error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry); 356575c3f29dSDave Hansen out_dput: 3566921a1650SAl Viro done_path_create(&new_path, new_dentry); 35671da177e4SLinus Torvalds out: 35682d8f3038SAl Viro path_put(&old_path); 35691da177e4SLinus Torvalds 35701da177e4SLinus Torvalds return error; 35711da177e4SLinus Torvalds } 35721da177e4SLinus Torvalds 35733480b257SHeiko Carstens SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) 35745590ff0dSUlrich Drepper { 3575c04030e1SUlrich Drepper return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 35765590ff0dSUlrich Drepper } 35775590ff0dSUlrich Drepper 35781da177e4SLinus Torvalds /* 35791da177e4SLinus Torvalds * The worst of all namespace operations - renaming directory. "Perverted" 35801da177e4SLinus Torvalds * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 35811da177e4SLinus Torvalds * Problems: 35821da177e4SLinus Torvalds * a) we can get into loop creation. Check is done in is_subdir(). 35831da177e4SLinus Torvalds * b) race potential - two innocent renames can create a loop together. 35841da177e4SLinus Torvalds * That's where 4.4 screws up. Current fix: serialization on 3585a11f3a05SArjan van de Ven * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 35861da177e4SLinus Torvalds * story. 35871da177e4SLinus Torvalds * c) we have to lock _three_ objects - parents and victim (if it exists). 35881b1dcc1bSJes Sorensen * And that - after we got ->i_mutex on parents (until then we don't know 35891da177e4SLinus Torvalds * whether the target exists). Solution: try to be smart with locking 35901da177e4SLinus Torvalds * order for inodes. We rely on the fact that tree topology may change 3591a11f3a05SArjan van de Ven * only under ->s_vfs_rename_mutex _and_ that parent of the object we 35921da177e4SLinus Torvalds * move will be locked. Thus we can rank directories by the tree 35931da177e4SLinus Torvalds * (ancestors first) and rank all non-directories after them. 35941da177e4SLinus Torvalds * That works since everybody except rename does "lock parent, lookup, 3595a11f3a05SArjan van de Ven * lock child" and rename is under ->s_vfs_rename_mutex. 35961da177e4SLinus Torvalds * HOWEVER, it relies on the assumption that any object with ->lookup() 35971da177e4SLinus Torvalds * has no more than 1 dentry. If "hybrid" objects will ever appear, 35981da177e4SLinus Torvalds * we'd better make sure that there's no link(2) for them. 3599e4eaac06SSage Weil * d) conversion from fhandle to dentry may come in the wrong moment - when 36001b1dcc1bSJes Sorensen * we are removing the target. Solution: we will have to grab ->i_mutex 36011da177e4SLinus Torvalds * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 3602c41b20e7SAdam Buchbinder * ->i_mutex on parents, which works but leads to some truly excessive 36031da177e4SLinus Torvalds * locking]. 36041da177e4SLinus Torvalds */ 360575c96f85SAdrian Bunk static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry, 36061da177e4SLinus Torvalds struct inode *new_dir, struct dentry *new_dentry) 36071da177e4SLinus Torvalds { 36081da177e4SLinus Torvalds int error = 0; 36099055cba7SSage Weil struct inode *target = new_dentry->d_inode; 36108de52778SAl Viro unsigned max_links = new_dir->i_sb->s_max_links; 36111da177e4SLinus Torvalds 36121da177e4SLinus Torvalds /* 36131da177e4SLinus Torvalds * If we are going to change the parent - check write permissions, 36141da177e4SLinus Torvalds * we'll need to flip '..'. 36151da177e4SLinus Torvalds */ 36161da177e4SLinus Torvalds if (new_dir != old_dir) { 3617f419a2e3SAl Viro error = inode_permission(old_dentry->d_inode, MAY_WRITE); 36181da177e4SLinus Torvalds if (error) 36191da177e4SLinus Torvalds return error; 36201da177e4SLinus Torvalds } 36211da177e4SLinus Torvalds 36221da177e4SLinus Torvalds error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); 36231da177e4SLinus Torvalds if (error) 36241da177e4SLinus Torvalds return error; 36251da177e4SLinus Torvalds 36261d2ef590SAl Viro dget(new_dentry); 3627d83c49f3SAl Viro if (target) 36281b1dcc1bSJes Sorensen mutex_lock(&target->i_mutex); 36299055cba7SSage Weil 36301da177e4SLinus Torvalds error = -EBUSY; 36319055cba7SSage Weil if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry)) 36329055cba7SSage Weil goto out; 36339055cba7SSage Weil 36348de52778SAl Viro error = -EMLINK; 36358de52778SAl Viro if (max_links && !target && new_dir != old_dir && 36368de52778SAl Viro new_dir->i_nlink >= max_links) 36378de52778SAl Viro goto out; 36388de52778SAl Viro 36393cebde24SSage Weil if (target) 36403cebde24SSage Weil shrink_dcache_parent(new_dentry); 36411da177e4SLinus Torvalds error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); 36429055cba7SSage Weil if (error) 36439055cba7SSage Weil goto out; 36449055cba7SSage Weil 36451da177e4SLinus Torvalds if (target) { 36461da177e4SLinus Torvalds target->i_flags |= S_DEAD; 3647d83c49f3SAl Viro dont_mount(new_dentry); 3648d83c49f3SAl Viro } 36499055cba7SSage Weil out: 36509055cba7SSage Weil if (target) 36511b1dcc1bSJes Sorensen mutex_unlock(&target->i_mutex); 36521d2ef590SAl Viro dput(new_dentry); 3653e31e14ecSStephen Smalley if (!error) 3654349457ccSMark Fasheh if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) 36551da177e4SLinus Torvalds d_move(old_dentry,new_dentry); 36561da177e4SLinus Torvalds return error; 36571da177e4SLinus Torvalds } 36581da177e4SLinus Torvalds 365975c96f85SAdrian Bunk static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry, 36601da177e4SLinus Torvalds struct inode *new_dir, struct dentry *new_dentry) 36611da177e4SLinus Torvalds { 366251892bbbSSage Weil struct inode *target = new_dentry->d_inode; 36631da177e4SLinus Torvalds int error; 36641da177e4SLinus Torvalds 36651da177e4SLinus Torvalds error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); 36661da177e4SLinus Torvalds if (error) 36671da177e4SLinus Torvalds return error; 36681da177e4SLinus Torvalds 36691da177e4SLinus Torvalds dget(new_dentry); 36701da177e4SLinus Torvalds if (target) 36711b1dcc1bSJes Sorensen mutex_lock(&target->i_mutex); 367251892bbbSSage Weil 36731da177e4SLinus Torvalds error = -EBUSY; 367451892bbbSSage Weil if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry)) 367551892bbbSSage Weil goto out; 367651892bbbSSage Weil 36771da177e4SLinus Torvalds error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); 367851892bbbSSage Weil if (error) 367951892bbbSSage Weil goto out; 368051892bbbSSage Weil 3681bec1052eSAl Viro if (target) 3682d83c49f3SAl Viro dont_mount(new_dentry); 3683349457ccSMark Fasheh if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) 36841da177e4SLinus Torvalds d_move(old_dentry, new_dentry); 368551892bbbSSage Weil out: 36861da177e4SLinus Torvalds if (target) 36871b1dcc1bSJes Sorensen mutex_unlock(&target->i_mutex); 36881da177e4SLinus Torvalds dput(new_dentry); 36891da177e4SLinus Torvalds return error; 36901da177e4SLinus Torvalds } 36911da177e4SLinus Torvalds 36921da177e4SLinus Torvalds int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, 36931da177e4SLinus Torvalds struct inode *new_dir, struct dentry *new_dentry) 36941da177e4SLinus Torvalds { 36951da177e4SLinus Torvalds int error; 36961da177e4SLinus Torvalds int is_dir = S_ISDIR(old_dentry->d_inode->i_mode); 369759b0df21SEric Paris const unsigned char *old_name; 36981da177e4SLinus Torvalds 36991da177e4SLinus Torvalds if (old_dentry->d_inode == new_dentry->d_inode) 37001da177e4SLinus Torvalds return 0; 37011da177e4SLinus Torvalds 37021da177e4SLinus Torvalds error = may_delete(old_dir, old_dentry, is_dir); 37031da177e4SLinus Torvalds if (error) 37041da177e4SLinus Torvalds return error; 37051da177e4SLinus Torvalds 37061da177e4SLinus Torvalds if (!new_dentry->d_inode) 3707a95164d9SMiklos Szeredi error = may_create(new_dir, new_dentry); 37081da177e4SLinus Torvalds else 37091da177e4SLinus Torvalds error = may_delete(new_dir, new_dentry, is_dir); 37101da177e4SLinus Torvalds if (error) 37111da177e4SLinus Torvalds return error; 37121da177e4SLinus Torvalds 3713acfa4380SAl Viro if (!old_dir->i_op->rename) 37141da177e4SLinus Torvalds return -EPERM; 37151da177e4SLinus Torvalds 37160eeca283SRobert Love old_name = fsnotify_oldname_init(old_dentry->d_name.name); 37170eeca283SRobert Love 37181da177e4SLinus Torvalds if (is_dir) 37191da177e4SLinus Torvalds error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry); 37201da177e4SLinus Torvalds else 37211da177e4SLinus Torvalds error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry); 3722123df294SAl Viro if (!error) 3723123df294SAl Viro fsnotify_move(old_dir, new_dir, old_name, is_dir, 37245a190ae6SAl Viro new_dentry->d_inode, old_dentry); 37250eeca283SRobert Love fsnotify_oldname_free(old_name); 37260eeca283SRobert Love 37271da177e4SLinus Torvalds return error; 37281da177e4SLinus Torvalds } 37291da177e4SLinus Torvalds 37302e4d0924SHeiko Carstens SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, 37312e4d0924SHeiko Carstens int, newdfd, const char __user *, newname) 37321da177e4SLinus Torvalds { 37331da177e4SLinus Torvalds struct dentry *old_dir, *new_dir; 37341da177e4SLinus Torvalds struct dentry *old_dentry, *new_dentry; 37351da177e4SLinus Torvalds struct dentry *trap; 37361da177e4SLinus Torvalds struct nameidata oldnd, newnd; 37372ad94ae6SAl Viro char *from; 37382ad94ae6SAl Viro char *to; 37392ad94ae6SAl Viro int error; 37401da177e4SLinus Torvalds 37412ad94ae6SAl Viro error = user_path_parent(olddfd, oldname, &oldnd, &from); 37421da177e4SLinus Torvalds if (error) 37431da177e4SLinus Torvalds goto exit; 37441da177e4SLinus Torvalds 37452ad94ae6SAl Viro error = user_path_parent(newdfd, newname, &newnd, &to); 37461da177e4SLinus Torvalds if (error) 37471da177e4SLinus Torvalds goto exit1; 37481da177e4SLinus Torvalds 37491da177e4SLinus Torvalds error = -EXDEV; 37504ac91378SJan Blunck if (oldnd.path.mnt != newnd.path.mnt) 37511da177e4SLinus Torvalds goto exit2; 37521da177e4SLinus Torvalds 37534ac91378SJan Blunck old_dir = oldnd.path.dentry; 37541da177e4SLinus Torvalds error = -EBUSY; 37551da177e4SLinus Torvalds if (oldnd.last_type != LAST_NORM) 37561da177e4SLinus Torvalds goto exit2; 37571da177e4SLinus Torvalds 37584ac91378SJan Blunck new_dir = newnd.path.dentry; 37591da177e4SLinus Torvalds if (newnd.last_type != LAST_NORM) 37601da177e4SLinus Torvalds goto exit2; 37611da177e4SLinus Torvalds 3762c30dabfeSJan Kara error = mnt_want_write(oldnd.path.mnt); 3763c30dabfeSJan Kara if (error) 3764c30dabfeSJan Kara goto exit2; 3765c30dabfeSJan Kara 37660612d9fbSOGAWA Hirofumi oldnd.flags &= ~LOOKUP_PARENT; 37670612d9fbSOGAWA Hirofumi newnd.flags &= ~LOOKUP_PARENT; 37684e9ed2f8SOGAWA Hirofumi newnd.flags |= LOOKUP_RENAME_TARGET; 37690612d9fbSOGAWA Hirofumi 37701da177e4SLinus Torvalds trap = lock_rename(new_dir, old_dir); 37711da177e4SLinus Torvalds 377249705b77SChristoph Hellwig old_dentry = lookup_hash(&oldnd); 37731da177e4SLinus Torvalds error = PTR_ERR(old_dentry); 37741da177e4SLinus Torvalds if (IS_ERR(old_dentry)) 37751da177e4SLinus Torvalds goto exit3; 37761da177e4SLinus Torvalds /* source must exist */ 37771da177e4SLinus Torvalds error = -ENOENT; 37781da177e4SLinus Torvalds if (!old_dentry->d_inode) 37791da177e4SLinus Torvalds goto exit4; 37801da177e4SLinus Torvalds /* unless the source is a directory trailing slashes give -ENOTDIR */ 37811da177e4SLinus Torvalds if (!S_ISDIR(old_dentry->d_inode->i_mode)) { 37821da177e4SLinus Torvalds error = -ENOTDIR; 37831da177e4SLinus Torvalds if (oldnd.last.name[oldnd.last.len]) 37841da177e4SLinus Torvalds goto exit4; 37851da177e4SLinus Torvalds if (newnd.last.name[newnd.last.len]) 37861da177e4SLinus Torvalds goto exit4; 37871da177e4SLinus Torvalds } 37881da177e4SLinus Torvalds /* source should not be ancestor of target */ 37891da177e4SLinus Torvalds error = -EINVAL; 37901da177e4SLinus Torvalds if (old_dentry == trap) 37911da177e4SLinus Torvalds goto exit4; 379249705b77SChristoph Hellwig new_dentry = lookup_hash(&newnd); 37931da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 37941da177e4SLinus Torvalds if (IS_ERR(new_dentry)) 37951da177e4SLinus Torvalds goto exit4; 37961da177e4SLinus Torvalds /* target should not be an ancestor of source */ 37971da177e4SLinus Torvalds error = -ENOTEMPTY; 37981da177e4SLinus Torvalds if (new_dentry == trap) 37991da177e4SLinus Torvalds goto exit5; 38001da177e4SLinus Torvalds 3801be6d3e56SKentaro Takeda error = security_path_rename(&oldnd.path, old_dentry, 3802be6d3e56SKentaro Takeda &newnd.path, new_dentry); 3803be6d3e56SKentaro Takeda if (error) 3804c30dabfeSJan Kara goto exit5; 38051da177e4SLinus Torvalds error = vfs_rename(old_dir->d_inode, old_dentry, 38061da177e4SLinus Torvalds new_dir->d_inode, new_dentry); 38071da177e4SLinus Torvalds exit5: 38081da177e4SLinus Torvalds dput(new_dentry); 38091da177e4SLinus Torvalds exit4: 38101da177e4SLinus Torvalds dput(old_dentry); 38111da177e4SLinus Torvalds exit3: 38121da177e4SLinus Torvalds unlock_rename(new_dir, old_dir); 3813c30dabfeSJan Kara mnt_drop_write(oldnd.path.mnt); 38141da177e4SLinus Torvalds exit2: 38151d957f9bSJan Blunck path_put(&newnd.path); 38162ad94ae6SAl Viro putname(to); 38171da177e4SLinus Torvalds exit1: 38181d957f9bSJan Blunck path_put(&oldnd.path); 38191da177e4SLinus Torvalds putname(from); 38202ad94ae6SAl Viro exit: 38211da177e4SLinus Torvalds return error; 38221da177e4SLinus Torvalds } 38231da177e4SLinus Torvalds 3824a26eab24SHeiko Carstens SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) 38255590ff0dSUlrich Drepper { 38265590ff0dSUlrich Drepper return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname); 38275590ff0dSUlrich Drepper } 38285590ff0dSUlrich Drepper 38291da177e4SLinus Torvalds int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link) 38301da177e4SLinus Torvalds { 38311da177e4SLinus Torvalds int len; 38321da177e4SLinus Torvalds 38331da177e4SLinus Torvalds len = PTR_ERR(link); 38341da177e4SLinus Torvalds if (IS_ERR(link)) 38351da177e4SLinus Torvalds goto out; 38361da177e4SLinus Torvalds 38371da177e4SLinus Torvalds len = strlen(link); 38381da177e4SLinus Torvalds if (len > (unsigned) buflen) 38391da177e4SLinus Torvalds len = buflen; 38401da177e4SLinus Torvalds if (copy_to_user(buffer, link, len)) 38411da177e4SLinus Torvalds len = -EFAULT; 38421da177e4SLinus Torvalds out: 38431da177e4SLinus Torvalds return len; 38441da177e4SLinus Torvalds } 38451da177e4SLinus Torvalds 38461da177e4SLinus Torvalds /* 38471da177e4SLinus Torvalds * A helper for ->readlink(). This should be used *ONLY* for symlinks that 38481da177e4SLinus Torvalds * have ->follow_link() touching nd only in nd_set_link(). Using (or not 38491da177e4SLinus Torvalds * using) it for any given inode is up to filesystem. 38501da177e4SLinus Torvalds */ 38511da177e4SLinus Torvalds int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen) 38521da177e4SLinus Torvalds { 38531da177e4SLinus Torvalds struct nameidata nd; 3854cc314eefSLinus Torvalds void *cookie; 3855694a1764SMarcin Slusarz int res; 3856cc314eefSLinus Torvalds 38571da177e4SLinus Torvalds nd.depth = 0; 3858cc314eefSLinus Torvalds cookie = dentry->d_inode->i_op->follow_link(dentry, &nd); 3859694a1764SMarcin Slusarz if (IS_ERR(cookie)) 3860694a1764SMarcin Slusarz return PTR_ERR(cookie); 3861694a1764SMarcin Slusarz 3862694a1764SMarcin Slusarz res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd)); 38631da177e4SLinus Torvalds if (dentry->d_inode->i_op->put_link) 3864cc314eefSLinus Torvalds dentry->d_inode->i_op->put_link(dentry, &nd, cookie); 3865694a1764SMarcin Slusarz return res; 38661da177e4SLinus Torvalds } 38671da177e4SLinus Torvalds 38681da177e4SLinus Torvalds int vfs_follow_link(struct nameidata *nd, const char *link) 38691da177e4SLinus Torvalds { 38701da177e4SLinus Torvalds return __vfs_follow_link(nd, link); 38711da177e4SLinus Torvalds } 38721da177e4SLinus Torvalds 38731da177e4SLinus Torvalds /* get the link contents into pagecache */ 38741da177e4SLinus Torvalds static char *page_getlink(struct dentry * dentry, struct page **ppage) 38751da177e4SLinus Torvalds { 3876ebd09abbSDuane Griffin char *kaddr; 38771da177e4SLinus Torvalds struct page *page; 38781da177e4SLinus Torvalds struct address_space *mapping = dentry->d_inode->i_mapping; 3879090d2b18SPekka Enberg page = read_mapping_page(mapping, 0, NULL); 38801da177e4SLinus Torvalds if (IS_ERR(page)) 38816fe6900eSNick Piggin return (char*)page; 38821da177e4SLinus Torvalds *ppage = page; 3883ebd09abbSDuane Griffin kaddr = kmap(page); 3884ebd09abbSDuane Griffin nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1); 3885ebd09abbSDuane Griffin return kaddr; 38861da177e4SLinus Torvalds } 38871da177e4SLinus Torvalds 38881da177e4SLinus Torvalds int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 38891da177e4SLinus Torvalds { 38901da177e4SLinus Torvalds struct page *page = NULL; 38911da177e4SLinus Torvalds char *s = page_getlink(dentry, &page); 38921da177e4SLinus Torvalds int res = vfs_readlink(dentry,buffer,buflen,s); 38931da177e4SLinus Torvalds if (page) { 38941da177e4SLinus Torvalds kunmap(page); 38951da177e4SLinus Torvalds page_cache_release(page); 38961da177e4SLinus Torvalds } 38971da177e4SLinus Torvalds return res; 38981da177e4SLinus Torvalds } 38991da177e4SLinus Torvalds 3900cc314eefSLinus Torvalds void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd) 39011da177e4SLinus Torvalds { 3902cc314eefSLinus Torvalds struct page *page = NULL; 39031da177e4SLinus Torvalds nd_set_link(nd, page_getlink(dentry, &page)); 3904cc314eefSLinus Torvalds return page; 39051da177e4SLinus Torvalds } 39061da177e4SLinus Torvalds 3907cc314eefSLinus Torvalds void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) 39081da177e4SLinus Torvalds { 3909cc314eefSLinus Torvalds struct page *page = cookie; 3910cc314eefSLinus Torvalds 3911cc314eefSLinus Torvalds if (page) { 39121da177e4SLinus Torvalds kunmap(page); 39131da177e4SLinus Torvalds page_cache_release(page); 39141da177e4SLinus Torvalds } 39151da177e4SLinus Torvalds } 39161da177e4SLinus Torvalds 391754566b2cSNick Piggin /* 391854566b2cSNick Piggin * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS 391954566b2cSNick Piggin */ 392054566b2cSNick Piggin int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) 39211da177e4SLinus Torvalds { 39221da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 39230adb25d2SKirill Korotaev struct page *page; 3924afddba49SNick Piggin void *fsdata; 3925beb497abSDmitriy Monakhov int err; 39261da177e4SLinus Torvalds char *kaddr; 392754566b2cSNick Piggin unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE; 392854566b2cSNick Piggin if (nofs) 392954566b2cSNick Piggin flags |= AOP_FLAG_NOFS; 39301da177e4SLinus Torvalds 39317e53cac4SNeilBrown retry: 3932afddba49SNick Piggin err = pagecache_write_begin(NULL, mapping, 0, len-1, 393354566b2cSNick Piggin flags, &page, &fsdata); 39341da177e4SLinus Torvalds if (err) 3935afddba49SNick Piggin goto fail; 3936afddba49SNick Piggin 3937e8e3c3d6SCong Wang kaddr = kmap_atomic(page); 39381da177e4SLinus Torvalds memcpy(kaddr, symname, len-1); 3939e8e3c3d6SCong Wang kunmap_atomic(kaddr); 3940afddba49SNick Piggin 3941afddba49SNick Piggin err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, 3942afddba49SNick Piggin page, fsdata); 39431da177e4SLinus Torvalds if (err < 0) 39441da177e4SLinus Torvalds goto fail; 3945afddba49SNick Piggin if (err < len-1) 3946afddba49SNick Piggin goto retry; 3947afddba49SNick Piggin 39481da177e4SLinus Torvalds mark_inode_dirty(inode); 39491da177e4SLinus Torvalds return 0; 39501da177e4SLinus Torvalds fail: 39511da177e4SLinus Torvalds return err; 39521da177e4SLinus Torvalds } 39531da177e4SLinus Torvalds 39540adb25d2SKirill Korotaev int page_symlink(struct inode *inode, const char *symname, int len) 39550adb25d2SKirill Korotaev { 39560adb25d2SKirill Korotaev return __page_symlink(inode, symname, len, 395754566b2cSNick Piggin !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS)); 39580adb25d2SKirill Korotaev } 39590adb25d2SKirill Korotaev 396092e1d5beSArjan van de Ven const struct inode_operations page_symlink_inode_operations = { 39611da177e4SLinus Torvalds .readlink = generic_readlink, 39621da177e4SLinus Torvalds .follow_link = page_follow_link_light, 39631da177e4SLinus Torvalds .put_link = page_put_link, 39641da177e4SLinus Torvalds }; 39651da177e4SLinus Torvalds 39662d8f3038SAl Viro EXPORT_SYMBOL(user_path_at); 3967cc53ce53SDavid Howells EXPORT_SYMBOL(follow_down_one); 39681da177e4SLinus Torvalds EXPORT_SYMBOL(follow_down); 39691da177e4SLinus Torvalds EXPORT_SYMBOL(follow_up); 3970f6d2ac5cSAl Viro EXPORT_SYMBOL(get_write_access); /* nfsd */ 39711da177e4SLinus Torvalds EXPORT_SYMBOL(getname); 39721da177e4SLinus Torvalds EXPORT_SYMBOL(lock_rename); 39731da177e4SLinus Torvalds EXPORT_SYMBOL(lookup_one_len); 39741da177e4SLinus Torvalds EXPORT_SYMBOL(page_follow_link_light); 39751da177e4SLinus Torvalds EXPORT_SYMBOL(page_put_link); 39761da177e4SLinus Torvalds EXPORT_SYMBOL(page_readlink); 39770adb25d2SKirill Korotaev EXPORT_SYMBOL(__page_symlink); 39781da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink); 39791da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink_inode_operations); 3980d1811465SAl Viro EXPORT_SYMBOL(kern_path); 398116f18200SJosef 'Jeff' Sipek EXPORT_SYMBOL(vfs_path_lookup); 3982f419a2e3SAl Viro EXPORT_SYMBOL(inode_permission); 39831da177e4SLinus Torvalds EXPORT_SYMBOL(unlock_rename); 39841da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_create); 39851da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_follow_link); 39861da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_link); 39871da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_mkdir); 39881da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_mknod); 39891da177e4SLinus Torvalds EXPORT_SYMBOL(generic_permission); 39901da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_readlink); 39911da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_rename); 39921da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_rmdir); 39931da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_symlink); 39941da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_unlink); 39951da177e4SLinus Torvalds EXPORT_SYMBOL(dentry_unhash); 39961da177e4SLinus Torvalds EXPORT_SYMBOL(generic_readlink); 3997