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> 181da177e4SLinus Torvalds #include <linux/module.h> 191da177e4SLinus Torvalds #include <linux/slab.h> 201da177e4SLinus Torvalds #include <linux/fs.h> 211da177e4SLinus Torvalds #include <linux/namei.h> 221da177e4SLinus Torvalds #include <linux/quotaops.h> 231da177e4SLinus Torvalds #include <linux/pagemap.h> 240eeca283SRobert Love #include <linux/fsnotify.h> 251da177e4SLinus Torvalds #include <linux/smp_lock.h> 261da177e4SLinus Torvalds #include <linux/personality.h> 271da177e4SLinus Torvalds #include <linux/security.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> 345590ff0dSUlrich Drepper #include <linux/namei.h> 351da177e4SLinus Torvalds #include <asm/namei.h> 361da177e4SLinus Torvalds #include <asm/uaccess.h> 371da177e4SLinus Torvalds 381da177e4SLinus Torvalds #define ACC_MODE(x) ("\000\004\002\006"[(x)&O_ACCMODE]) 391da177e4SLinus Torvalds 401da177e4SLinus Torvalds /* [Feb-1997 T. Schoebel-Theuer] 411da177e4SLinus Torvalds * Fundamental changes in the pathname lookup mechanisms (namei) 421da177e4SLinus Torvalds * were necessary because of omirr. The reason is that omirr needs 431da177e4SLinus Torvalds * to know the _real_ pathname, not the user-supplied one, in case 441da177e4SLinus Torvalds * of symlinks (and also when transname replacements occur). 451da177e4SLinus Torvalds * 461da177e4SLinus Torvalds * The new code replaces the old recursive symlink resolution with 471da177e4SLinus Torvalds * an iterative one (in case of non-nested symlink chains). It does 481da177e4SLinus Torvalds * this with calls to <fs>_follow_link(). 491da177e4SLinus Torvalds * As a side effect, dir_namei(), _namei() and follow_link() are now 501da177e4SLinus Torvalds * replaced with a single function lookup_dentry() that can handle all 511da177e4SLinus Torvalds * the special cases of the former code. 521da177e4SLinus Torvalds * 531da177e4SLinus Torvalds * With the new dcache, the pathname is stored at each inode, at least as 541da177e4SLinus Torvalds * long as the refcount of the inode is positive. As a side effect, the 551da177e4SLinus Torvalds * size of the dcache depends on the inode cache and thus is dynamic. 561da177e4SLinus Torvalds * 571da177e4SLinus Torvalds * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink 581da177e4SLinus Torvalds * resolution to correspond with current state of the code. 591da177e4SLinus Torvalds * 601da177e4SLinus Torvalds * Note that the symlink resolution is not *completely* iterative. 611da177e4SLinus Torvalds * There is still a significant amount of tail- and mid- recursion in 621da177e4SLinus Torvalds * the algorithm. Also, note that <fs>_readlink() is not used in 631da177e4SLinus Torvalds * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink() 641da177e4SLinus Torvalds * may return different results than <fs>_follow_link(). Many virtual 651da177e4SLinus Torvalds * filesystems (including /proc) exhibit this behavior. 661da177e4SLinus Torvalds */ 671da177e4SLinus Torvalds 681da177e4SLinus Torvalds /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation: 691da177e4SLinus Torvalds * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL 701da177e4SLinus Torvalds * and the name already exists in form of a symlink, try to create the new 711da177e4SLinus Torvalds * name indicated by the symlink. The old code always complained that the 721da177e4SLinus Torvalds * name already exists, due to not following the symlink even if its target 731da177e4SLinus Torvalds * is nonexistent. The new semantics affects also mknod() and link() when 741da177e4SLinus Torvalds * the name is a symlink pointing to a non-existant name. 751da177e4SLinus Torvalds * 761da177e4SLinus Torvalds * I don't know which semantics is the right one, since I have no access 771da177e4SLinus Torvalds * to standards. But I found by trial that HP-UX 9.0 has the full "new" 781da177e4SLinus Torvalds * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the 791da177e4SLinus Torvalds * "old" one. Personally, I think the new semantics is much more logical. 801da177e4SLinus Torvalds * Note that "ln old new" where "new" is a symlink pointing to a non-existing 811da177e4SLinus Torvalds * file does succeed in both HP-UX and SunOs, but not in Solaris 821da177e4SLinus Torvalds * and in the old Linux semantics. 831da177e4SLinus Torvalds */ 841da177e4SLinus Torvalds 851da177e4SLinus Torvalds /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink 861da177e4SLinus Torvalds * semantics. See the comments in "open_namei" and "do_link" below. 871da177e4SLinus Torvalds * 881da177e4SLinus Torvalds * [10-Sep-98 Alan Modra] Another symlink change. 891da177e4SLinus Torvalds */ 901da177e4SLinus Torvalds 911da177e4SLinus Torvalds /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks: 921da177e4SLinus Torvalds * inside the path - always follow. 931da177e4SLinus Torvalds * in the last component in creation/removal/renaming - never follow. 941da177e4SLinus Torvalds * if LOOKUP_FOLLOW passed - follow. 951da177e4SLinus Torvalds * if the pathname has trailing slashes - follow. 961da177e4SLinus Torvalds * otherwise - don't follow. 971da177e4SLinus Torvalds * (applied in that order). 981da177e4SLinus Torvalds * 991da177e4SLinus Torvalds * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT 1001da177e4SLinus Torvalds * restored for 2.4. This is the last surviving part of old 4.2BSD bug. 1011da177e4SLinus Torvalds * During the 2.4 we need to fix the userland stuff depending on it - 1021da177e4SLinus Torvalds * hopefully we will be able to get rid of that wart in 2.5. So far only 1031da177e4SLinus Torvalds * XEmacs seems to be relying on it... 1041da177e4SLinus Torvalds */ 1051da177e4SLinus Torvalds /* 1061da177e4SLinus Torvalds * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland) 107a11f3a05SArjan van de Ven * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives 1081da177e4SLinus Torvalds * any extra contention... 1091da177e4SLinus Torvalds */ 1101da177e4SLinus Torvalds 1111da177e4SLinus Torvalds /* In order to reduce some races, while at the same time doing additional 1121da177e4SLinus Torvalds * checking and hopefully speeding things up, we copy filenames to the 1131da177e4SLinus Torvalds * kernel data space before using them.. 1141da177e4SLinus Torvalds * 1151da177e4SLinus Torvalds * POSIX.1 2.4: an empty pathname is invalid (ENOENT). 1161da177e4SLinus Torvalds * PATH_MAX includes the nul terminator --RR. 1171da177e4SLinus Torvalds */ 118858119e1SArjan van de Ven static int do_getname(const char __user *filename, char *page) 1191da177e4SLinus Torvalds { 1201da177e4SLinus Torvalds int retval; 1211da177e4SLinus Torvalds unsigned long len = PATH_MAX; 1221da177e4SLinus Torvalds 1231da177e4SLinus Torvalds if (!segment_eq(get_fs(), KERNEL_DS)) { 1241da177e4SLinus Torvalds if ((unsigned long) filename >= TASK_SIZE) 1251da177e4SLinus Torvalds return -EFAULT; 1261da177e4SLinus Torvalds if (TASK_SIZE - (unsigned long) filename < PATH_MAX) 1271da177e4SLinus Torvalds len = TASK_SIZE - (unsigned long) filename; 1281da177e4SLinus Torvalds } 1291da177e4SLinus Torvalds 1301da177e4SLinus Torvalds retval = strncpy_from_user(page, filename, len); 1311da177e4SLinus Torvalds if (retval > 0) { 1321da177e4SLinus Torvalds if (retval < len) 1331da177e4SLinus Torvalds return 0; 1341da177e4SLinus Torvalds return -ENAMETOOLONG; 1351da177e4SLinus Torvalds } else if (!retval) 1361da177e4SLinus Torvalds retval = -ENOENT; 1371da177e4SLinus Torvalds return retval; 1381da177e4SLinus Torvalds } 1391da177e4SLinus Torvalds 1401da177e4SLinus Torvalds char * getname(const char __user * filename) 1411da177e4SLinus Torvalds { 1421da177e4SLinus Torvalds char *tmp, *result; 1431da177e4SLinus Torvalds 1441da177e4SLinus Torvalds result = ERR_PTR(-ENOMEM); 1451da177e4SLinus Torvalds tmp = __getname(); 1461da177e4SLinus Torvalds if (tmp) { 1471da177e4SLinus Torvalds int retval = do_getname(filename, tmp); 1481da177e4SLinus Torvalds 1491da177e4SLinus Torvalds result = tmp; 1501da177e4SLinus Torvalds if (retval < 0) { 1511da177e4SLinus Torvalds __putname(tmp); 1521da177e4SLinus Torvalds result = ERR_PTR(retval); 1531da177e4SLinus Torvalds } 1541da177e4SLinus Torvalds } 1551da177e4SLinus Torvalds audit_getname(result); 1561da177e4SLinus Torvalds return result; 1571da177e4SLinus Torvalds } 1581da177e4SLinus Torvalds 1591da177e4SLinus Torvalds #ifdef CONFIG_AUDITSYSCALL 1601da177e4SLinus Torvalds void putname(const char *name) 1611da177e4SLinus Torvalds { 1625ac3a9c2SAl Viro if (unlikely(!audit_dummy_context())) 1631da177e4SLinus Torvalds audit_putname(name); 1641da177e4SLinus Torvalds else 1651da177e4SLinus Torvalds __putname(name); 1661da177e4SLinus Torvalds } 1671da177e4SLinus Torvalds EXPORT_SYMBOL(putname); 1681da177e4SLinus Torvalds #endif 1691da177e4SLinus Torvalds 1701da177e4SLinus Torvalds 1711da177e4SLinus Torvalds /** 1721da177e4SLinus Torvalds * generic_permission - check for access rights on a Posix-like filesystem 1731da177e4SLinus Torvalds * @inode: inode to check access rights for 1741da177e4SLinus Torvalds * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 1751da177e4SLinus Torvalds * @check_acl: optional callback to check for Posix ACLs 1761da177e4SLinus Torvalds * 1771da177e4SLinus Torvalds * Used to check for read/write/execute permissions on a file. 1781da177e4SLinus Torvalds * We use "fsuid" for this, letting us set arbitrary permissions 1791da177e4SLinus Torvalds * for filesystem access without changing the "normal" uids which 1801da177e4SLinus Torvalds * are used for other things.. 1811da177e4SLinus Torvalds */ 1821da177e4SLinus Torvalds int generic_permission(struct inode *inode, int mask, 1831da177e4SLinus Torvalds int (*check_acl)(struct inode *inode, int mask)) 1841da177e4SLinus Torvalds { 1851da177e4SLinus Torvalds umode_t mode = inode->i_mode; 1861da177e4SLinus Torvalds 1871da177e4SLinus Torvalds if (current->fsuid == inode->i_uid) 1881da177e4SLinus Torvalds mode >>= 6; 1891da177e4SLinus Torvalds else { 1901da177e4SLinus Torvalds if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) { 1911da177e4SLinus Torvalds int error = check_acl(inode, mask); 1921da177e4SLinus Torvalds if (error == -EACCES) 1931da177e4SLinus Torvalds goto check_capabilities; 1941da177e4SLinus Torvalds else if (error != -EAGAIN) 1951da177e4SLinus Torvalds return error; 1961da177e4SLinus Torvalds } 1971da177e4SLinus Torvalds 1981da177e4SLinus Torvalds if (in_group_p(inode->i_gid)) 1991da177e4SLinus Torvalds mode >>= 3; 2001da177e4SLinus Torvalds } 2011da177e4SLinus Torvalds 2021da177e4SLinus Torvalds /* 2031da177e4SLinus Torvalds * If the DACs are ok we don't need any capability check. 2041da177e4SLinus Torvalds */ 2051da177e4SLinus Torvalds if (((mode & mask & (MAY_READ|MAY_WRITE|MAY_EXEC)) == mask)) 2061da177e4SLinus Torvalds return 0; 2071da177e4SLinus Torvalds 2081da177e4SLinus Torvalds check_capabilities: 2091da177e4SLinus Torvalds /* 2101da177e4SLinus Torvalds * Read/write DACs are always overridable. 2111da177e4SLinus Torvalds * Executable DACs are overridable if at least one exec bit is set. 2121da177e4SLinus Torvalds */ 2131da177e4SLinus Torvalds if (!(mask & MAY_EXEC) || 2141da177e4SLinus Torvalds (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode)) 2151da177e4SLinus Torvalds if (capable(CAP_DAC_OVERRIDE)) 2161da177e4SLinus Torvalds return 0; 2171da177e4SLinus Torvalds 2181da177e4SLinus Torvalds /* 2191da177e4SLinus Torvalds * Searching includes executable on directories, else just read. 2201da177e4SLinus Torvalds */ 2211da177e4SLinus Torvalds if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE))) 2221da177e4SLinus Torvalds if (capable(CAP_DAC_READ_SEARCH)) 2231da177e4SLinus Torvalds return 0; 2241da177e4SLinus Torvalds 2251da177e4SLinus Torvalds return -EACCES; 2261da177e4SLinus Torvalds } 2271da177e4SLinus Torvalds 2281da177e4SLinus Torvalds int permission(struct inode *inode, int mask, struct nameidata *nd) 2291da177e4SLinus Torvalds { 230a343bb77STrond Myklebust umode_t mode = inode->i_mode; 2311da177e4SLinus Torvalds int retval, submask; 2321da177e4SLinus Torvalds 2331da177e4SLinus Torvalds if (mask & MAY_WRITE) { 2341da177e4SLinus Torvalds 2351da177e4SLinus Torvalds /* 2361da177e4SLinus Torvalds * Nobody gets write access to a read-only fs. 2371da177e4SLinus Torvalds */ 2381da177e4SLinus Torvalds if (IS_RDONLY(inode) && 2391da177e4SLinus Torvalds (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) 2401da177e4SLinus Torvalds return -EROFS; 2411da177e4SLinus Torvalds 2421da177e4SLinus Torvalds /* 2431da177e4SLinus Torvalds * Nobody gets write access to an immutable file. 2441da177e4SLinus Torvalds */ 2451da177e4SLinus Torvalds if (IS_IMMUTABLE(inode)) 2461da177e4SLinus Torvalds return -EACCES; 2471da177e4SLinus Torvalds } 2481da177e4SLinus Torvalds 2491da177e4SLinus Torvalds 250a343bb77STrond Myklebust /* 251a343bb77STrond Myklebust * MAY_EXEC on regular files requires special handling: We override 252317a40acSStas Sergeev * filesystem execute permissions if the mode bits aren't set or 253317a40acSStas Sergeev * the fs is mounted with the "noexec" flag. 254a343bb77STrond Myklebust */ 255317a40acSStas Sergeev if ((mask & MAY_EXEC) && S_ISREG(mode) && (!(mode & S_IXUGO) || 256317a40acSStas Sergeev (nd && nd->mnt && (nd->mnt->mnt_flags & MNT_NOEXEC)))) 257a343bb77STrond Myklebust return -EACCES; 258a343bb77STrond Myklebust 2591da177e4SLinus Torvalds /* Ordinary permission routines do not understand MAY_APPEND. */ 2601da177e4SLinus Torvalds submask = mask & ~MAY_APPEND; 2611da177e4SLinus Torvalds if (inode->i_op && inode->i_op->permission) 2621da177e4SLinus Torvalds retval = inode->i_op->permission(inode, submask, nd); 2631da177e4SLinus Torvalds else 2641da177e4SLinus Torvalds retval = generic_permission(inode, submask, NULL); 2651da177e4SLinus Torvalds if (retval) 2661da177e4SLinus Torvalds return retval; 2671da177e4SLinus Torvalds 2681da177e4SLinus Torvalds return security_inode_permission(inode, mask, nd); 2691da177e4SLinus Torvalds } 2701da177e4SLinus Torvalds 271e4543eddSChristoph Hellwig /** 272e4543eddSChristoph Hellwig * vfs_permission - check for access rights to a given path 273e4543eddSChristoph Hellwig * @nd: lookup result that describes the path 274e4543eddSChristoph Hellwig * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 275e4543eddSChristoph Hellwig * 276e4543eddSChristoph Hellwig * Used to check for read/write/execute permissions on a path. 277e4543eddSChristoph Hellwig * We use "fsuid" for this, letting us set arbitrary permissions 278e4543eddSChristoph Hellwig * for filesystem access without changing the "normal" uids which 279e4543eddSChristoph Hellwig * are used for other things. 280e4543eddSChristoph Hellwig */ 281e4543eddSChristoph Hellwig int vfs_permission(struct nameidata *nd, int mask) 282e4543eddSChristoph Hellwig { 283e4543eddSChristoph Hellwig return permission(nd->dentry->d_inode, mask, nd); 284e4543eddSChristoph Hellwig } 285e4543eddSChristoph Hellwig 2868c744fb8SChristoph Hellwig /** 2878c744fb8SChristoph Hellwig * file_permission - check for additional access rights to a given file 2888c744fb8SChristoph Hellwig * @file: file to check access rights for 2898c744fb8SChristoph Hellwig * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 2908c744fb8SChristoph Hellwig * 2918c744fb8SChristoph Hellwig * Used to check for read/write/execute permissions on an already opened 2928c744fb8SChristoph Hellwig * file. 2938c744fb8SChristoph Hellwig * 2948c744fb8SChristoph Hellwig * Note: 2958c744fb8SChristoph Hellwig * Do not use this function in new code. All access checks should 2968c744fb8SChristoph Hellwig * be done using vfs_permission(). 2978c744fb8SChristoph Hellwig */ 2988c744fb8SChristoph Hellwig int file_permission(struct file *file, int mask) 2998c744fb8SChristoph Hellwig { 3000f7fc9e4SJosef "Jeff" Sipek return permission(file->f_path.dentry->d_inode, mask, NULL); 3018c744fb8SChristoph Hellwig } 3028c744fb8SChristoph Hellwig 3031da177e4SLinus Torvalds /* 3041da177e4SLinus Torvalds * get_write_access() gets write permission for a file. 3051da177e4SLinus Torvalds * put_write_access() releases this write permission. 3061da177e4SLinus Torvalds * This is used for regular files. 3071da177e4SLinus Torvalds * We cannot support write (and maybe mmap read-write shared) accesses and 3081da177e4SLinus Torvalds * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode 3091da177e4SLinus Torvalds * can have the following values: 3101da177e4SLinus Torvalds * 0: no writers, no VM_DENYWRITE mappings 3111da177e4SLinus Torvalds * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist 3121da177e4SLinus Torvalds * > 0: (i_writecount) users are writing to the file. 3131da177e4SLinus Torvalds * 3141da177e4SLinus Torvalds * Normally we operate on that counter with atomic_{inc,dec} and it's safe 3151da177e4SLinus Torvalds * except for the cases where we don't hold i_writecount yet. Then we need to 3161da177e4SLinus Torvalds * use {get,deny}_write_access() - these functions check the sign and refuse 3171da177e4SLinus Torvalds * to do the change if sign is wrong. Exclusion between them is provided by 3181da177e4SLinus Torvalds * the inode->i_lock spinlock. 3191da177e4SLinus Torvalds */ 3201da177e4SLinus Torvalds 3211da177e4SLinus Torvalds int get_write_access(struct inode * inode) 3221da177e4SLinus Torvalds { 3231da177e4SLinus Torvalds spin_lock(&inode->i_lock); 3241da177e4SLinus Torvalds if (atomic_read(&inode->i_writecount) < 0) { 3251da177e4SLinus Torvalds spin_unlock(&inode->i_lock); 3261da177e4SLinus Torvalds return -ETXTBSY; 3271da177e4SLinus Torvalds } 3281da177e4SLinus Torvalds atomic_inc(&inode->i_writecount); 3291da177e4SLinus Torvalds spin_unlock(&inode->i_lock); 3301da177e4SLinus Torvalds 3311da177e4SLinus Torvalds return 0; 3321da177e4SLinus Torvalds } 3331da177e4SLinus Torvalds 3341da177e4SLinus Torvalds int deny_write_access(struct file * file) 3351da177e4SLinus Torvalds { 3360f7fc9e4SJosef "Jeff" Sipek struct inode *inode = file->f_path.dentry->d_inode; 3371da177e4SLinus Torvalds 3381da177e4SLinus Torvalds spin_lock(&inode->i_lock); 3391da177e4SLinus Torvalds if (atomic_read(&inode->i_writecount) > 0) { 3401da177e4SLinus Torvalds spin_unlock(&inode->i_lock); 3411da177e4SLinus Torvalds return -ETXTBSY; 3421da177e4SLinus Torvalds } 3431da177e4SLinus Torvalds atomic_dec(&inode->i_writecount); 3441da177e4SLinus Torvalds spin_unlock(&inode->i_lock); 3451da177e4SLinus Torvalds 3461da177e4SLinus Torvalds return 0; 3471da177e4SLinus Torvalds } 3481da177e4SLinus Torvalds 3491da177e4SLinus Torvalds void path_release(struct nameidata *nd) 3501da177e4SLinus Torvalds { 3511da177e4SLinus Torvalds dput(nd->dentry); 3521da177e4SLinus Torvalds mntput(nd->mnt); 3531da177e4SLinus Torvalds } 3541da177e4SLinus Torvalds 3551da177e4SLinus Torvalds /* 3561da177e4SLinus Torvalds * umount() mustn't call path_release()/mntput() as that would clear 3571da177e4SLinus Torvalds * mnt_expiry_mark 3581da177e4SLinus Torvalds */ 3591da177e4SLinus Torvalds void path_release_on_umount(struct nameidata *nd) 3601da177e4SLinus Torvalds { 3611da177e4SLinus Torvalds dput(nd->dentry); 362751c404bSMiklos Szeredi mntput_no_expire(nd->mnt); 3631da177e4SLinus Torvalds } 3641da177e4SLinus Torvalds 365834f2a4aSTrond Myklebust /** 366834f2a4aSTrond Myklebust * release_open_intent - free up open intent resources 367834f2a4aSTrond Myklebust * @nd: pointer to nameidata 368834f2a4aSTrond Myklebust */ 369834f2a4aSTrond Myklebust void release_open_intent(struct nameidata *nd) 370834f2a4aSTrond Myklebust { 3710f7fc9e4SJosef "Jeff" Sipek if (nd->intent.open.file->f_path.dentry == NULL) 372834f2a4aSTrond Myklebust put_filp(nd->intent.open.file); 373834f2a4aSTrond Myklebust else 374834f2a4aSTrond Myklebust fput(nd->intent.open.file); 375834f2a4aSTrond Myklebust } 376834f2a4aSTrond Myklebust 377bcdc5e01SIan Kent static inline struct dentry * 378bcdc5e01SIan Kent do_revalidate(struct dentry *dentry, struct nameidata *nd) 379bcdc5e01SIan Kent { 380bcdc5e01SIan Kent int status = dentry->d_op->d_revalidate(dentry, nd); 381bcdc5e01SIan Kent if (unlikely(status <= 0)) { 382bcdc5e01SIan Kent /* 383bcdc5e01SIan Kent * The dentry failed validation. 384bcdc5e01SIan Kent * If d_revalidate returned 0 attempt to invalidate 385bcdc5e01SIan Kent * the dentry otherwise d_revalidate is asking us 386bcdc5e01SIan Kent * to return a fail status. 387bcdc5e01SIan Kent */ 388bcdc5e01SIan Kent if (!status) { 389bcdc5e01SIan Kent if (!d_invalidate(dentry)) { 390bcdc5e01SIan Kent dput(dentry); 391bcdc5e01SIan Kent dentry = NULL; 392bcdc5e01SIan Kent } 393bcdc5e01SIan Kent } else { 394bcdc5e01SIan Kent dput(dentry); 395bcdc5e01SIan Kent dentry = ERR_PTR(status); 396bcdc5e01SIan Kent } 397bcdc5e01SIan Kent } 398bcdc5e01SIan Kent return dentry; 399bcdc5e01SIan Kent } 400bcdc5e01SIan Kent 4011da177e4SLinus Torvalds /* 4021da177e4SLinus Torvalds * Internal lookup() using the new generic dcache. 4031da177e4SLinus Torvalds * SMP-safe 4041da177e4SLinus Torvalds */ 4051da177e4SLinus Torvalds static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd) 4061da177e4SLinus Torvalds { 4071da177e4SLinus Torvalds struct dentry * dentry = __d_lookup(parent, name); 4081da177e4SLinus Torvalds 4091da177e4SLinus Torvalds /* lockess __d_lookup may fail due to concurrent d_move() 4101da177e4SLinus Torvalds * in some unrelated directory, so try with d_lookup 4111da177e4SLinus Torvalds */ 4121da177e4SLinus Torvalds if (!dentry) 4131da177e4SLinus Torvalds dentry = d_lookup(parent, name); 4141da177e4SLinus Torvalds 415bcdc5e01SIan Kent if (dentry && dentry->d_op && dentry->d_op->d_revalidate) 416bcdc5e01SIan Kent dentry = do_revalidate(dentry, nd); 417bcdc5e01SIan Kent 4181da177e4SLinus Torvalds return dentry; 4191da177e4SLinus Torvalds } 4201da177e4SLinus Torvalds 4211da177e4SLinus Torvalds /* 4221da177e4SLinus Torvalds * Short-cut version of permission(), for calling by 4231da177e4SLinus Torvalds * path_walk(), when dcache lock is held. Combines parts 4241da177e4SLinus Torvalds * of permission() and generic_permission(), and tests ONLY for 4251da177e4SLinus Torvalds * MAY_EXEC permission. 4261da177e4SLinus Torvalds * 4271da177e4SLinus Torvalds * If appropriate, check DAC only. If not appropriate, or 4281da177e4SLinus Torvalds * short-cut DAC fails, then call permission() to do more 4291da177e4SLinus Torvalds * complete permission check. 4301da177e4SLinus Torvalds */ 431858119e1SArjan van de Ven static int exec_permission_lite(struct inode *inode, 4321da177e4SLinus Torvalds struct nameidata *nd) 4331da177e4SLinus Torvalds { 4341da177e4SLinus Torvalds umode_t mode = inode->i_mode; 4351da177e4SLinus Torvalds 4361da177e4SLinus Torvalds if (inode->i_op && inode->i_op->permission) 4371da177e4SLinus Torvalds return -EAGAIN; 4381da177e4SLinus Torvalds 4391da177e4SLinus Torvalds if (current->fsuid == inode->i_uid) 4401da177e4SLinus Torvalds mode >>= 6; 4411da177e4SLinus Torvalds else if (in_group_p(inode->i_gid)) 4421da177e4SLinus Torvalds mode >>= 3; 4431da177e4SLinus Torvalds 4441da177e4SLinus Torvalds if (mode & MAY_EXEC) 4451da177e4SLinus Torvalds goto ok; 4461da177e4SLinus Torvalds 4471da177e4SLinus Torvalds if ((inode->i_mode & S_IXUGO) && capable(CAP_DAC_OVERRIDE)) 4481da177e4SLinus Torvalds goto ok; 4491da177e4SLinus Torvalds 4501da177e4SLinus Torvalds if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_OVERRIDE)) 4511da177e4SLinus Torvalds goto ok; 4521da177e4SLinus Torvalds 4531da177e4SLinus Torvalds if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_READ_SEARCH)) 4541da177e4SLinus Torvalds goto ok; 4551da177e4SLinus Torvalds 4561da177e4SLinus Torvalds return -EACCES; 4571da177e4SLinus Torvalds ok: 4581da177e4SLinus Torvalds return security_inode_permission(inode, MAY_EXEC, nd); 4591da177e4SLinus Torvalds } 4601da177e4SLinus Torvalds 4611da177e4SLinus Torvalds /* 4621da177e4SLinus Torvalds * This is called when everything else fails, and we actually have 4631da177e4SLinus Torvalds * to go to the low-level filesystem to find out what we should do.. 4641da177e4SLinus Torvalds * 4651da177e4SLinus Torvalds * We get the directory semaphore, and after getting that we also 4661da177e4SLinus Torvalds * make sure that nobody added the entry to the dcache in the meantime.. 4671da177e4SLinus Torvalds * SMP-safe 4681da177e4SLinus Torvalds */ 4691da177e4SLinus Torvalds static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd) 4701da177e4SLinus Torvalds { 4711da177e4SLinus Torvalds struct dentry * result; 4721da177e4SLinus Torvalds struct inode *dir = parent->d_inode; 4731da177e4SLinus Torvalds 4741b1dcc1bSJes Sorensen mutex_lock(&dir->i_mutex); 4751da177e4SLinus Torvalds /* 4761da177e4SLinus Torvalds * First re-do the cached lookup just in case it was created 4771da177e4SLinus Torvalds * while we waited for the directory semaphore.. 4781da177e4SLinus Torvalds * 4791da177e4SLinus Torvalds * FIXME! This could use version numbering or similar to 4801da177e4SLinus Torvalds * avoid unnecessary cache lookups. 4811da177e4SLinus Torvalds * 4821da177e4SLinus Torvalds * The "dcache_lock" is purely to protect the RCU list walker 4831da177e4SLinus Torvalds * from concurrent renames at this point (we mustn't get false 4841da177e4SLinus Torvalds * negatives from the RCU list walk here, unlike the optimistic 4851da177e4SLinus Torvalds * fast walk). 4861da177e4SLinus Torvalds * 4871da177e4SLinus Torvalds * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup 4881da177e4SLinus Torvalds */ 4891da177e4SLinus Torvalds result = d_lookup(parent, name); 4901da177e4SLinus Torvalds if (!result) { 4911da177e4SLinus Torvalds struct dentry * dentry = d_alloc(parent, name); 4921da177e4SLinus Torvalds result = ERR_PTR(-ENOMEM); 4931da177e4SLinus Torvalds if (dentry) { 4941da177e4SLinus Torvalds result = dir->i_op->lookup(dir, dentry, nd); 4951da177e4SLinus Torvalds if (result) 4961da177e4SLinus Torvalds dput(dentry); 4971da177e4SLinus Torvalds else 4981da177e4SLinus Torvalds result = dentry; 4991da177e4SLinus Torvalds } 5001b1dcc1bSJes Sorensen mutex_unlock(&dir->i_mutex); 5011da177e4SLinus Torvalds return result; 5021da177e4SLinus Torvalds } 5031da177e4SLinus Torvalds 5041da177e4SLinus Torvalds /* 5051da177e4SLinus Torvalds * Uhhuh! Nasty case: the cache was re-populated while 5061da177e4SLinus Torvalds * we waited on the semaphore. Need to revalidate. 5071da177e4SLinus Torvalds */ 5081b1dcc1bSJes Sorensen mutex_unlock(&dir->i_mutex); 5091da177e4SLinus Torvalds if (result->d_op && result->d_op->d_revalidate) { 510bcdc5e01SIan Kent result = do_revalidate(result, nd); 511bcdc5e01SIan Kent if (!result) 5121da177e4SLinus Torvalds result = ERR_PTR(-ENOENT); 5131da177e4SLinus Torvalds } 5141da177e4SLinus Torvalds return result; 5151da177e4SLinus Torvalds } 5161da177e4SLinus Torvalds 5171da177e4SLinus Torvalds static int __emul_lookup_dentry(const char *, struct nameidata *); 5181da177e4SLinus Torvalds 5191da177e4SLinus Torvalds /* SMP-safe */ 520f1662356SArjan van de Ven static __always_inline int 5211da177e4SLinus Torvalds walk_init_root(const char *name, struct nameidata *nd) 5221da177e4SLinus Torvalds { 523e518ddb7SAndreas Mohr struct fs_struct *fs = current->fs; 524e518ddb7SAndreas Mohr 525e518ddb7SAndreas Mohr read_lock(&fs->lock); 526e518ddb7SAndreas Mohr if (fs->altroot && !(nd->flags & LOOKUP_NOALT)) { 527e518ddb7SAndreas Mohr nd->mnt = mntget(fs->altrootmnt); 528e518ddb7SAndreas Mohr nd->dentry = dget(fs->altroot); 529e518ddb7SAndreas Mohr read_unlock(&fs->lock); 5301da177e4SLinus Torvalds if (__emul_lookup_dentry(name,nd)) 5311da177e4SLinus Torvalds return 0; 532e518ddb7SAndreas Mohr read_lock(&fs->lock); 5331da177e4SLinus Torvalds } 534e518ddb7SAndreas Mohr nd->mnt = mntget(fs->rootmnt); 535e518ddb7SAndreas Mohr nd->dentry = dget(fs->root); 536e518ddb7SAndreas Mohr read_unlock(&fs->lock); 5371da177e4SLinus Torvalds return 1; 5381da177e4SLinus Torvalds } 5391da177e4SLinus Torvalds 540f1662356SArjan van de Ven static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link) 5411da177e4SLinus Torvalds { 5421da177e4SLinus Torvalds int res = 0; 5431da177e4SLinus Torvalds char *name; 5441da177e4SLinus Torvalds if (IS_ERR(link)) 5451da177e4SLinus Torvalds goto fail; 5461da177e4SLinus Torvalds 5471da177e4SLinus Torvalds if (*link == '/') { 5481da177e4SLinus Torvalds path_release(nd); 5491da177e4SLinus Torvalds if (!walk_init_root(link, nd)) 5501da177e4SLinus Torvalds /* weird __emul_prefix() stuff did it */ 5511da177e4SLinus Torvalds goto out; 5521da177e4SLinus Torvalds } 5531da177e4SLinus Torvalds res = link_path_walk(link, nd); 5541da177e4SLinus Torvalds out: 5551da177e4SLinus Torvalds if (nd->depth || res || nd->last_type!=LAST_NORM) 5561da177e4SLinus Torvalds return res; 5571da177e4SLinus Torvalds /* 5581da177e4SLinus Torvalds * If it is an iterative symlinks resolution in open_namei() we 5591da177e4SLinus Torvalds * have to copy the last component. And all that crap because of 5601da177e4SLinus Torvalds * bloody create() on broken symlinks. Furrfu... 5611da177e4SLinus Torvalds */ 5621da177e4SLinus Torvalds name = __getname(); 5631da177e4SLinus Torvalds if (unlikely(!name)) { 5641da177e4SLinus Torvalds path_release(nd); 5651da177e4SLinus Torvalds return -ENOMEM; 5661da177e4SLinus Torvalds } 5671da177e4SLinus Torvalds strcpy(name, nd->last.name); 5681da177e4SLinus Torvalds nd->last.name = name; 5691da177e4SLinus Torvalds return 0; 5701da177e4SLinus Torvalds fail: 5711da177e4SLinus Torvalds path_release(nd); 5721da177e4SLinus Torvalds return PTR_ERR(link); 5731da177e4SLinus Torvalds } 5741da177e4SLinus Torvalds 575051d3812SIan Kent static inline void dput_path(struct path *path, struct nameidata *nd) 576051d3812SIan Kent { 577051d3812SIan Kent dput(path->dentry); 578051d3812SIan Kent if (path->mnt != nd->mnt) 579051d3812SIan Kent mntput(path->mnt); 580051d3812SIan Kent } 581051d3812SIan Kent 582051d3812SIan Kent static inline void path_to_nameidata(struct path *path, struct nameidata *nd) 583051d3812SIan Kent { 584051d3812SIan Kent dput(nd->dentry); 585051d3812SIan Kent if (nd->mnt != path->mnt) 586051d3812SIan Kent mntput(nd->mnt); 587051d3812SIan Kent nd->mnt = path->mnt; 588051d3812SIan Kent nd->dentry = path->dentry; 589051d3812SIan Kent } 590051d3812SIan Kent 591f1662356SArjan van de Ven static __always_inline int __do_follow_link(struct path *path, struct nameidata *nd) 5921da177e4SLinus Torvalds { 5931da177e4SLinus Torvalds int error; 594cc314eefSLinus Torvalds void *cookie; 595cd4e91d3SAl Viro struct dentry *dentry = path->dentry; 5961da177e4SLinus Torvalds 597d671a1cbSAl Viro touch_atime(path->mnt, dentry); 5981da177e4SLinus Torvalds nd_set_link(nd, NULL); 599cd4e91d3SAl Viro 600051d3812SIan Kent if (path->mnt != nd->mnt) { 601051d3812SIan Kent path_to_nameidata(path, nd); 602051d3812SIan Kent dget(dentry); 603051d3812SIan Kent } 604cd4e91d3SAl Viro mntget(path->mnt); 605cc314eefSLinus Torvalds cookie = dentry->d_inode->i_op->follow_link(dentry, nd); 606cc314eefSLinus Torvalds error = PTR_ERR(cookie); 607cc314eefSLinus Torvalds if (!IS_ERR(cookie)) { 6081da177e4SLinus Torvalds char *s = nd_get_link(nd); 609cc314eefSLinus Torvalds error = 0; 6101da177e4SLinus Torvalds if (s) 6111da177e4SLinus Torvalds error = __vfs_follow_link(nd, s); 6121da177e4SLinus Torvalds if (dentry->d_inode->i_op->put_link) 613cc314eefSLinus Torvalds dentry->d_inode->i_op->put_link(dentry, nd, cookie); 6141da177e4SLinus Torvalds } 615cd4e91d3SAl Viro dput(dentry); 616cd4e91d3SAl Viro mntput(path->mnt); 6171da177e4SLinus Torvalds 6181da177e4SLinus Torvalds return error; 6191da177e4SLinus Torvalds } 6201da177e4SLinus Torvalds 6211da177e4SLinus Torvalds /* 6221da177e4SLinus Torvalds * This limits recursive symlink follows to 8, while 6231da177e4SLinus Torvalds * limiting consecutive symlinks to 40. 6241da177e4SLinus Torvalds * 6251da177e4SLinus Torvalds * Without that kind of total limit, nasty chains of consecutive 6261da177e4SLinus Torvalds * symlinks can cause almost arbitrarily long lookups. 6271da177e4SLinus Torvalds */ 62890ebe565SAl Viro static inline int do_follow_link(struct path *path, struct nameidata *nd) 6291da177e4SLinus Torvalds { 6301da177e4SLinus Torvalds int err = -ELOOP; 6311da177e4SLinus Torvalds if (current->link_count >= MAX_NESTED_LINKS) 6321da177e4SLinus Torvalds goto loop; 6331da177e4SLinus Torvalds if (current->total_link_count >= 40) 6341da177e4SLinus Torvalds goto loop; 6351da177e4SLinus Torvalds BUG_ON(nd->depth >= MAX_NESTED_LINKS); 6361da177e4SLinus Torvalds cond_resched(); 63790ebe565SAl Viro err = security_inode_follow_link(path->dentry, nd); 6381da177e4SLinus Torvalds if (err) 6391da177e4SLinus Torvalds goto loop; 6401da177e4SLinus Torvalds current->link_count++; 6411da177e4SLinus Torvalds current->total_link_count++; 6421da177e4SLinus Torvalds nd->depth++; 643cd4e91d3SAl Viro err = __do_follow_link(path, nd); 6441da177e4SLinus Torvalds current->link_count--; 6451da177e4SLinus Torvalds nd->depth--; 6461da177e4SLinus Torvalds return err; 6471da177e4SLinus Torvalds loop: 64809dd17d3SMiklos Szeredi dput_path(path, nd); 649839d9f93SAl Viro path_release(nd); 6501da177e4SLinus Torvalds return err; 6511da177e4SLinus Torvalds } 6521da177e4SLinus Torvalds 6531da177e4SLinus Torvalds int follow_up(struct vfsmount **mnt, struct dentry **dentry) 6541da177e4SLinus Torvalds { 6551da177e4SLinus Torvalds struct vfsmount *parent; 6561da177e4SLinus Torvalds struct dentry *mountpoint; 6571da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 6581da177e4SLinus Torvalds parent=(*mnt)->mnt_parent; 6591da177e4SLinus Torvalds if (parent == *mnt) { 6601da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 6611da177e4SLinus Torvalds return 0; 6621da177e4SLinus Torvalds } 6631da177e4SLinus Torvalds mntget(parent); 6641da177e4SLinus Torvalds mountpoint=dget((*mnt)->mnt_mountpoint); 6651da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 6661da177e4SLinus Torvalds dput(*dentry); 6671da177e4SLinus Torvalds *dentry = mountpoint; 6681da177e4SLinus Torvalds mntput(*mnt); 6691da177e4SLinus Torvalds *mnt = parent; 6701da177e4SLinus Torvalds return 1; 6711da177e4SLinus Torvalds } 6721da177e4SLinus Torvalds 6731da177e4SLinus Torvalds /* no need for dcache_lock, as serialization is taken care in 6741da177e4SLinus Torvalds * namespace.c 6751da177e4SLinus Torvalds */ 676463ffb2eSAl Viro static int __follow_mount(struct path *path) 677463ffb2eSAl Viro { 678463ffb2eSAl Viro int res = 0; 679463ffb2eSAl Viro while (d_mountpoint(path->dentry)) { 680463ffb2eSAl Viro struct vfsmount *mounted = lookup_mnt(path->mnt, path->dentry); 681463ffb2eSAl Viro if (!mounted) 682463ffb2eSAl Viro break; 683463ffb2eSAl Viro dput(path->dentry); 684463ffb2eSAl Viro if (res) 685463ffb2eSAl Viro mntput(path->mnt); 686463ffb2eSAl Viro path->mnt = mounted; 687463ffb2eSAl Viro path->dentry = dget(mounted->mnt_root); 688463ffb2eSAl Viro res = 1; 689463ffb2eSAl Viro } 690463ffb2eSAl Viro return res; 691463ffb2eSAl Viro } 692463ffb2eSAl Viro 69358c465ebSAl Viro static void follow_mount(struct vfsmount **mnt, struct dentry **dentry) 6941da177e4SLinus Torvalds { 6951da177e4SLinus Torvalds while (d_mountpoint(*dentry)) { 6961da177e4SLinus Torvalds struct vfsmount *mounted = lookup_mnt(*mnt, *dentry); 6971da177e4SLinus Torvalds if (!mounted) 6981da177e4SLinus Torvalds break; 69958c465ebSAl Viro dput(*dentry); 7001da177e4SLinus Torvalds mntput(*mnt); 7011da177e4SLinus Torvalds *mnt = mounted; 7021da177e4SLinus Torvalds *dentry = dget(mounted->mnt_root); 7031da177e4SLinus Torvalds } 7041da177e4SLinus Torvalds } 7051da177e4SLinus Torvalds 7061da177e4SLinus Torvalds /* no need for dcache_lock, as serialization is taken care in 7071da177e4SLinus Torvalds * namespace.c 7081da177e4SLinus Torvalds */ 709e13b210fSAl Viro int follow_down(struct vfsmount **mnt, struct dentry **dentry) 7101da177e4SLinus Torvalds { 7111da177e4SLinus Torvalds struct vfsmount *mounted; 7121da177e4SLinus Torvalds 7131da177e4SLinus Torvalds mounted = lookup_mnt(*mnt, *dentry); 7141da177e4SLinus Torvalds if (mounted) { 715e13b210fSAl Viro dput(*dentry); 7161da177e4SLinus Torvalds mntput(*mnt); 7171da177e4SLinus Torvalds *mnt = mounted; 7181da177e4SLinus Torvalds *dentry = dget(mounted->mnt_root); 7191da177e4SLinus Torvalds return 1; 7201da177e4SLinus Torvalds } 7211da177e4SLinus Torvalds return 0; 7221da177e4SLinus Torvalds } 7231da177e4SLinus Torvalds 724f1662356SArjan van de Ven static __always_inline void follow_dotdot(struct nameidata *nd) 7251da177e4SLinus Torvalds { 726e518ddb7SAndreas Mohr struct fs_struct *fs = current->fs; 727e518ddb7SAndreas Mohr 7281da177e4SLinus Torvalds while(1) { 7291da177e4SLinus Torvalds struct vfsmount *parent; 73058c465ebSAl Viro struct dentry *old = nd->dentry; 7311da177e4SLinus Torvalds 732e518ddb7SAndreas Mohr read_lock(&fs->lock); 733e518ddb7SAndreas Mohr if (nd->dentry == fs->root && 734e518ddb7SAndreas Mohr nd->mnt == fs->rootmnt) { 735e518ddb7SAndreas Mohr read_unlock(&fs->lock); 7361da177e4SLinus Torvalds break; 7371da177e4SLinus Torvalds } 738e518ddb7SAndreas Mohr read_unlock(&fs->lock); 7391da177e4SLinus Torvalds spin_lock(&dcache_lock); 74058c465ebSAl Viro if (nd->dentry != nd->mnt->mnt_root) { 74158c465ebSAl Viro nd->dentry = dget(nd->dentry->d_parent); 7421da177e4SLinus Torvalds spin_unlock(&dcache_lock); 7431da177e4SLinus Torvalds dput(old); 7441da177e4SLinus Torvalds break; 7451da177e4SLinus Torvalds } 7461da177e4SLinus Torvalds spin_unlock(&dcache_lock); 7471da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 74858c465ebSAl Viro parent = nd->mnt->mnt_parent; 74958c465ebSAl Viro if (parent == nd->mnt) { 7501da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 7511da177e4SLinus Torvalds break; 7521da177e4SLinus Torvalds } 7531da177e4SLinus Torvalds mntget(parent); 75458c465ebSAl Viro nd->dentry = dget(nd->mnt->mnt_mountpoint); 7551da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 7561da177e4SLinus Torvalds dput(old); 75758c465ebSAl Viro mntput(nd->mnt); 75858c465ebSAl Viro nd->mnt = parent; 7591da177e4SLinus Torvalds } 76058c465ebSAl Viro follow_mount(&nd->mnt, &nd->dentry); 7611da177e4SLinus Torvalds } 7621da177e4SLinus Torvalds 7631da177e4SLinus Torvalds /* 7641da177e4SLinus Torvalds * It's more convoluted than I'd like it to be, but... it's still fairly 7651da177e4SLinus Torvalds * small and for now I'd prefer to have fast path as straight as possible. 7661da177e4SLinus Torvalds * It _is_ time-critical. 7671da177e4SLinus Torvalds */ 7681da177e4SLinus Torvalds static int do_lookup(struct nameidata *nd, struct qstr *name, 7691da177e4SLinus Torvalds struct path *path) 7701da177e4SLinus Torvalds { 7711da177e4SLinus Torvalds struct vfsmount *mnt = nd->mnt; 7721da177e4SLinus Torvalds struct dentry *dentry = __d_lookup(nd->dentry, name); 7731da177e4SLinus Torvalds 7741da177e4SLinus Torvalds if (!dentry) 7751da177e4SLinus Torvalds goto need_lookup; 7761da177e4SLinus Torvalds if (dentry->d_op && dentry->d_op->d_revalidate) 7771da177e4SLinus Torvalds goto need_revalidate; 7781da177e4SLinus Torvalds done: 7791da177e4SLinus Torvalds path->mnt = mnt; 7801da177e4SLinus Torvalds path->dentry = dentry; 781634ee701SAl Viro __follow_mount(path); 7821da177e4SLinus Torvalds return 0; 7831da177e4SLinus Torvalds 7841da177e4SLinus Torvalds need_lookup: 7851da177e4SLinus Torvalds dentry = real_lookup(nd->dentry, name, nd); 7861da177e4SLinus Torvalds if (IS_ERR(dentry)) 7871da177e4SLinus Torvalds goto fail; 7881da177e4SLinus Torvalds goto done; 7891da177e4SLinus Torvalds 7901da177e4SLinus Torvalds need_revalidate: 791bcdc5e01SIan Kent dentry = do_revalidate(dentry, nd); 792bcdc5e01SIan Kent if (!dentry) 7931da177e4SLinus Torvalds goto need_lookup; 794bcdc5e01SIan Kent if (IS_ERR(dentry)) 795bcdc5e01SIan Kent goto fail; 796bcdc5e01SIan Kent goto done; 7971da177e4SLinus Torvalds 7981da177e4SLinus Torvalds fail: 7991da177e4SLinus Torvalds return PTR_ERR(dentry); 8001da177e4SLinus Torvalds } 8011da177e4SLinus Torvalds 8021da177e4SLinus Torvalds /* 8031da177e4SLinus Torvalds * Name resolution. 804ea3834d9SPrasanna Meda * This is the basic name resolution function, turning a pathname into 805ea3834d9SPrasanna Meda * the final dentry. We expect 'base' to be positive and a directory. 8061da177e4SLinus Torvalds * 807ea3834d9SPrasanna Meda * Returns 0 and nd will have valid dentry and mnt on success. 808ea3834d9SPrasanna Meda * Returns error and drops reference to input namei data on failure. 8091da177e4SLinus Torvalds */ 8101da177e4SLinus Torvalds static fastcall int __link_path_walk(const char * name, struct nameidata *nd) 8111da177e4SLinus Torvalds { 8121da177e4SLinus Torvalds struct path next; 8131da177e4SLinus Torvalds struct inode *inode; 8141da177e4SLinus Torvalds int err; 8151da177e4SLinus Torvalds unsigned int lookup_flags = nd->flags; 8161da177e4SLinus Torvalds 8171da177e4SLinus Torvalds while (*name=='/') 8181da177e4SLinus Torvalds name++; 8191da177e4SLinus Torvalds if (!*name) 8201da177e4SLinus Torvalds goto return_reval; 8211da177e4SLinus Torvalds 8221da177e4SLinus Torvalds inode = nd->dentry->d_inode; 8231da177e4SLinus Torvalds if (nd->depth) 824f55eab82STrond Myklebust lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE); 8251da177e4SLinus Torvalds 8261da177e4SLinus Torvalds /* At this point we know we have a real path component. */ 8271da177e4SLinus Torvalds for(;;) { 8281da177e4SLinus Torvalds unsigned long hash; 8291da177e4SLinus Torvalds struct qstr this; 8301da177e4SLinus Torvalds unsigned int c; 8311da177e4SLinus Torvalds 832cdce5d6bSTrond Myklebust nd->flags |= LOOKUP_CONTINUE; 8331da177e4SLinus Torvalds err = exec_permission_lite(inode, nd); 834e4543eddSChristoph Hellwig if (err == -EAGAIN) 835e4543eddSChristoph Hellwig err = vfs_permission(nd, MAY_EXEC); 8361da177e4SLinus Torvalds if (err) 8371da177e4SLinus Torvalds break; 8381da177e4SLinus Torvalds 8391da177e4SLinus Torvalds this.name = name; 8401da177e4SLinus Torvalds c = *(const unsigned char *)name; 8411da177e4SLinus Torvalds 8421da177e4SLinus Torvalds hash = init_name_hash(); 8431da177e4SLinus Torvalds do { 8441da177e4SLinus Torvalds name++; 8451da177e4SLinus Torvalds hash = partial_name_hash(c, hash); 8461da177e4SLinus Torvalds c = *(const unsigned char *)name; 8471da177e4SLinus Torvalds } while (c && (c != '/')); 8481da177e4SLinus Torvalds this.len = name - (const char *) this.name; 8491da177e4SLinus Torvalds this.hash = end_name_hash(hash); 8501da177e4SLinus Torvalds 8511da177e4SLinus Torvalds /* remove trailing slashes? */ 8521da177e4SLinus Torvalds if (!c) 8531da177e4SLinus Torvalds goto last_component; 8541da177e4SLinus Torvalds while (*++name == '/'); 8551da177e4SLinus Torvalds if (!*name) 8561da177e4SLinus Torvalds goto last_with_slashes; 8571da177e4SLinus Torvalds 8581da177e4SLinus Torvalds /* 8591da177e4SLinus Torvalds * "." and ".." are special - ".." especially so because it has 8601da177e4SLinus Torvalds * to be able to know about the current root directory and 8611da177e4SLinus Torvalds * parent relationships. 8621da177e4SLinus Torvalds */ 8631da177e4SLinus Torvalds if (this.name[0] == '.') switch (this.len) { 8641da177e4SLinus Torvalds default: 8651da177e4SLinus Torvalds break; 8661da177e4SLinus Torvalds case 2: 8671da177e4SLinus Torvalds if (this.name[1] != '.') 8681da177e4SLinus Torvalds break; 86958c465ebSAl Viro follow_dotdot(nd); 8701da177e4SLinus Torvalds inode = nd->dentry->d_inode; 8711da177e4SLinus Torvalds /* fallthrough */ 8721da177e4SLinus Torvalds case 1: 8731da177e4SLinus Torvalds continue; 8741da177e4SLinus Torvalds } 8751da177e4SLinus Torvalds /* 8761da177e4SLinus Torvalds * See if the low-level filesystem might want 8771da177e4SLinus Torvalds * to use its own hash.. 8781da177e4SLinus Torvalds */ 8791da177e4SLinus Torvalds if (nd->dentry->d_op && nd->dentry->d_op->d_hash) { 8801da177e4SLinus Torvalds err = nd->dentry->d_op->d_hash(nd->dentry, &this); 8811da177e4SLinus Torvalds if (err < 0) 8821da177e4SLinus Torvalds break; 8831da177e4SLinus Torvalds } 8841da177e4SLinus Torvalds /* This does the actual lookups.. */ 8851da177e4SLinus Torvalds err = do_lookup(nd, &this, &next); 8861da177e4SLinus Torvalds if (err) 8871da177e4SLinus Torvalds break; 8881da177e4SLinus Torvalds 8891da177e4SLinus Torvalds err = -ENOENT; 8901da177e4SLinus Torvalds inode = next.dentry->d_inode; 8911da177e4SLinus Torvalds if (!inode) 8921da177e4SLinus Torvalds goto out_dput; 8931da177e4SLinus Torvalds err = -ENOTDIR; 8941da177e4SLinus Torvalds if (!inode->i_op) 8951da177e4SLinus Torvalds goto out_dput; 8961da177e4SLinus Torvalds 8971da177e4SLinus Torvalds if (inode->i_op->follow_link) { 89890ebe565SAl Viro err = do_follow_link(&next, nd); 8991da177e4SLinus Torvalds if (err) 9001da177e4SLinus Torvalds goto return_err; 9011da177e4SLinus Torvalds err = -ENOENT; 9021da177e4SLinus Torvalds inode = nd->dentry->d_inode; 9031da177e4SLinus Torvalds if (!inode) 9041da177e4SLinus Torvalds break; 9051da177e4SLinus Torvalds err = -ENOTDIR; 9061da177e4SLinus Torvalds if (!inode->i_op) 9071da177e4SLinus Torvalds break; 90809dd17d3SMiklos Szeredi } else 90909dd17d3SMiklos Szeredi path_to_nameidata(&next, nd); 9101da177e4SLinus Torvalds err = -ENOTDIR; 9111da177e4SLinus Torvalds if (!inode->i_op->lookup) 9121da177e4SLinus Torvalds break; 9131da177e4SLinus Torvalds continue; 9141da177e4SLinus Torvalds /* here ends the main loop */ 9151da177e4SLinus Torvalds 9161da177e4SLinus Torvalds last_with_slashes: 9171da177e4SLinus Torvalds lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 9181da177e4SLinus Torvalds last_component: 919f55eab82STrond Myklebust /* Clear LOOKUP_CONTINUE iff it was previously unset */ 920f55eab82STrond Myklebust nd->flags &= lookup_flags | ~LOOKUP_CONTINUE; 9211da177e4SLinus Torvalds if (lookup_flags & LOOKUP_PARENT) 9221da177e4SLinus Torvalds goto lookup_parent; 9231da177e4SLinus Torvalds if (this.name[0] == '.') switch (this.len) { 9241da177e4SLinus Torvalds default: 9251da177e4SLinus Torvalds break; 9261da177e4SLinus Torvalds case 2: 9271da177e4SLinus Torvalds if (this.name[1] != '.') 9281da177e4SLinus Torvalds break; 92958c465ebSAl Viro follow_dotdot(nd); 9301da177e4SLinus Torvalds inode = nd->dentry->d_inode; 9311da177e4SLinus Torvalds /* fallthrough */ 9321da177e4SLinus Torvalds case 1: 9331da177e4SLinus Torvalds goto return_reval; 9341da177e4SLinus Torvalds } 9351da177e4SLinus Torvalds if (nd->dentry->d_op && nd->dentry->d_op->d_hash) { 9361da177e4SLinus Torvalds err = nd->dentry->d_op->d_hash(nd->dentry, &this); 9371da177e4SLinus Torvalds if (err < 0) 9381da177e4SLinus Torvalds break; 9391da177e4SLinus Torvalds } 9401da177e4SLinus Torvalds err = do_lookup(nd, &this, &next); 9411da177e4SLinus Torvalds if (err) 9421da177e4SLinus Torvalds break; 9431da177e4SLinus Torvalds inode = next.dentry->d_inode; 9441da177e4SLinus Torvalds if ((lookup_flags & LOOKUP_FOLLOW) 9451da177e4SLinus Torvalds && inode && inode->i_op && inode->i_op->follow_link) { 94690ebe565SAl Viro err = do_follow_link(&next, nd); 9471da177e4SLinus Torvalds if (err) 9481da177e4SLinus Torvalds goto return_err; 9491da177e4SLinus Torvalds inode = nd->dentry->d_inode; 95009dd17d3SMiklos Szeredi } else 95109dd17d3SMiklos Szeredi path_to_nameidata(&next, nd); 9521da177e4SLinus Torvalds err = -ENOENT; 9531da177e4SLinus Torvalds if (!inode) 9541da177e4SLinus Torvalds break; 9551da177e4SLinus Torvalds if (lookup_flags & LOOKUP_DIRECTORY) { 9561da177e4SLinus Torvalds err = -ENOTDIR; 9571da177e4SLinus Torvalds if (!inode->i_op || !inode->i_op->lookup) 9581da177e4SLinus Torvalds break; 9591da177e4SLinus Torvalds } 9601da177e4SLinus Torvalds goto return_base; 9611da177e4SLinus Torvalds lookup_parent: 9621da177e4SLinus Torvalds nd->last = this; 9631da177e4SLinus Torvalds nd->last_type = LAST_NORM; 9641da177e4SLinus Torvalds if (this.name[0] != '.') 9651da177e4SLinus Torvalds goto return_base; 9661da177e4SLinus Torvalds if (this.len == 1) 9671da177e4SLinus Torvalds nd->last_type = LAST_DOT; 9681da177e4SLinus Torvalds else if (this.len == 2 && this.name[1] == '.') 9691da177e4SLinus Torvalds nd->last_type = LAST_DOTDOT; 9701da177e4SLinus Torvalds else 9711da177e4SLinus Torvalds goto return_base; 9721da177e4SLinus Torvalds return_reval: 9731da177e4SLinus Torvalds /* 9741da177e4SLinus Torvalds * We bypassed the ordinary revalidation routines. 9751da177e4SLinus Torvalds * We may need to check the cached dentry for staleness. 9761da177e4SLinus Torvalds */ 9771da177e4SLinus Torvalds if (nd->dentry && nd->dentry->d_sb && 9781da177e4SLinus Torvalds (nd->dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) { 9791da177e4SLinus Torvalds err = -ESTALE; 9801da177e4SLinus Torvalds /* Note: we do not d_invalidate() */ 9811da177e4SLinus Torvalds if (!nd->dentry->d_op->d_revalidate(nd->dentry, nd)) 9821da177e4SLinus Torvalds break; 9831da177e4SLinus Torvalds } 9841da177e4SLinus Torvalds return_base: 9851da177e4SLinus Torvalds return 0; 9861da177e4SLinus Torvalds out_dput: 98709dd17d3SMiklos Szeredi dput_path(&next, nd); 9881da177e4SLinus Torvalds break; 9891da177e4SLinus Torvalds } 9901da177e4SLinus Torvalds path_release(nd); 9911da177e4SLinus Torvalds return_err: 9921da177e4SLinus Torvalds return err; 9931da177e4SLinus Torvalds } 9941da177e4SLinus Torvalds 9951da177e4SLinus Torvalds /* 9961da177e4SLinus Torvalds * Wrapper to retry pathname resolution whenever the underlying 9971da177e4SLinus Torvalds * file system returns an ESTALE. 9981da177e4SLinus Torvalds * 9991da177e4SLinus Torvalds * Retry the whole path once, forcing real lookup requests 10001da177e4SLinus Torvalds * instead of relying on the dcache. 10011da177e4SLinus Torvalds */ 10021da177e4SLinus Torvalds int fastcall link_path_walk(const char *name, struct nameidata *nd) 10031da177e4SLinus Torvalds { 10041da177e4SLinus Torvalds struct nameidata save = *nd; 10051da177e4SLinus Torvalds int result; 10061da177e4SLinus Torvalds 10071da177e4SLinus Torvalds /* make sure the stuff we saved doesn't go away */ 10081da177e4SLinus Torvalds dget(save.dentry); 10091da177e4SLinus Torvalds mntget(save.mnt); 10101da177e4SLinus Torvalds 10111da177e4SLinus Torvalds result = __link_path_walk(name, nd); 10121da177e4SLinus Torvalds if (result == -ESTALE) { 10131da177e4SLinus Torvalds *nd = save; 10141da177e4SLinus Torvalds dget(nd->dentry); 10151da177e4SLinus Torvalds mntget(nd->mnt); 10161da177e4SLinus Torvalds nd->flags |= LOOKUP_REVAL; 10171da177e4SLinus Torvalds result = __link_path_walk(name, nd); 10181da177e4SLinus Torvalds } 10191da177e4SLinus Torvalds 10201da177e4SLinus Torvalds dput(save.dentry); 10211da177e4SLinus Torvalds mntput(save.mnt); 10221da177e4SLinus Torvalds 10231da177e4SLinus Torvalds return result; 10241da177e4SLinus Torvalds } 10251da177e4SLinus Torvalds 10261da177e4SLinus Torvalds int fastcall path_walk(const char * name, struct nameidata *nd) 10271da177e4SLinus Torvalds { 10281da177e4SLinus Torvalds current->total_link_count = 0; 10291da177e4SLinus Torvalds return link_path_walk(name, nd); 10301da177e4SLinus Torvalds } 10311da177e4SLinus Torvalds 1032ea3834d9SPrasanna Meda /* 1033ea3834d9SPrasanna Meda * SMP-safe: Returns 1 and nd will have valid dentry and mnt, if 1034ea3834d9SPrasanna Meda * everything is done. Returns 0 and drops input nd, if lookup failed; 1035ea3834d9SPrasanna Meda */ 10361da177e4SLinus Torvalds static int __emul_lookup_dentry(const char *name, struct nameidata *nd) 10371da177e4SLinus Torvalds { 10381da177e4SLinus Torvalds if (path_walk(name, nd)) 10391da177e4SLinus Torvalds return 0; /* something went wrong... */ 10401da177e4SLinus Torvalds 10411da177e4SLinus Torvalds if (!nd->dentry->d_inode || S_ISDIR(nd->dentry->d_inode->i_mode)) { 10421da177e4SLinus Torvalds struct dentry *old_dentry = nd->dentry; 10431da177e4SLinus Torvalds struct vfsmount *old_mnt = nd->mnt; 10441da177e4SLinus Torvalds struct qstr last = nd->last; 10451da177e4SLinus Torvalds int last_type = nd->last_type; 1046e518ddb7SAndreas Mohr struct fs_struct *fs = current->fs; 1047e518ddb7SAndreas Mohr 10481da177e4SLinus Torvalds /* 1049e518ddb7SAndreas Mohr * NAME was not found in alternate root or it's a directory. 1050e518ddb7SAndreas Mohr * Try to find it in the normal root: 10511da177e4SLinus Torvalds */ 10521da177e4SLinus Torvalds nd->last_type = LAST_ROOT; 1053e518ddb7SAndreas Mohr read_lock(&fs->lock); 1054e518ddb7SAndreas Mohr nd->mnt = mntget(fs->rootmnt); 1055e518ddb7SAndreas Mohr nd->dentry = dget(fs->root); 1056e518ddb7SAndreas Mohr read_unlock(&fs->lock); 10571da177e4SLinus Torvalds if (path_walk(name, nd) == 0) { 10581da177e4SLinus Torvalds if (nd->dentry->d_inode) { 10591da177e4SLinus Torvalds dput(old_dentry); 10601da177e4SLinus Torvalds mntput(old_mnt); 10611da177e4SLinus Torvalds return 1; 10621da177e4SLinus Torvalds } 10631da177e4SLinus Torvalds path_release(nd); 10641da177e4SLinus Torvalds } 10651da177e4SLinus Torvalds nd->dentry = old_dentry; 10661da177e4SLinus Torvalds nd->mnt = old_mnt; 10671da177e4SLinus Torvalds nd->last = last; 10681da177e4SLinus Torvalds nd->last_type = last_type; 10691da177e4SLinus Torvalds } 10701da177e4SLinus Torvalds return 1; 10711da177e4SLinus Torvalds } 10721da177e4SLinus Torvalds 10731da177e4SLinus Torvalds void set_fs_altroot(void) 10741da177e4SLinus Torvalds { 10751da177e4SLinus Torvalds char *emul = __emul_prefix(); 10761da177e4SLinus Torvalds struct nameidata nd; 10771da177e4SLinus Torvalds struct vfsmount *mnt = NULL, *oldmnt; 10781da177e4SLinus Torvalds struct dentry *dentry = NULL, *olddentry; 10791da177e4SLinus Torvalds int err; 1080e518ddb7SAndreas Mohr struct fs_struct *fs = current->fs; 10811da177e4SLinus Torvalds 10821da177e4SLinus Torvalds if (!emul) 10831da177e4SLinus Torvalds goto set_it; 10841da177e4SLinus Torvalds err = path_lookup(emul, LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_NOALT, &nd); 10851da177e4SLinus Torvalds if (!err) { 10861da177e4SLinus Torvalds mnt = nd.mnt; 10871da177e4SLinus Torvalds dentry = nd.dentry; 10881da177e4SLinus Torvalds } 10891da177e4SLinus Torvalds set_it: 1090e518ddb7SAndreas Mohr write_lock(&fs->lock); 1091e518ddb7SAndreas Mohr oldmnt = fs->altrootmnt; 1092e518ddb7SAndreas Mohr olddentry = fs->altroot; 1093e518ddb7SAndreas Mohr fs->altrootmnt = mnt; 1094e518ddb7SAndreas Mohr fs->altroot = dentry; 1095e518ddb7SAndreas Mohr write_unlock(&fs->lock); 10961da177e4SLinus Torvalds if (olddentry) { 10971da177e4SLinus Torvalds dput(olddentry); 10981da177e4SLinus Torvalds mntput(oldmnt); 10991da177e4SLinus Torvalds } 11001da177e4SLinus Torvalds } 11011da177e4SLinus Torvalds 1102ea3834d9SPrasanna Meda /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 11035590ff0dSUlrich Drepper static int fastcall do_path_lookup(int dfd, const char *name, 11045590ff0dSUlrich Drepper unsigned int flags, struct nameidata *nd) 11051da177e4SLinus Torvalds { 1106ea3834d9SPrasanna Meda int retval = 0; 1107170aa3d0SUlrich Drepper int fput_needed; 1108170aa3d0SUlrich Drepper struct file *file; 1109e518ddb7SAndreas Mohr struct fs_struct *fs = current->fs; 11101da177e4SLinus Torvalds 11111da177e4SLinus Torvalds nd->last_type = LAST_ROOT; /* if there are only slashes... */ 11121da177e4SLinus Torvalds nd->flags = flags; 11131da177e4SLinus Torvalds nd->depth = 0; 11141da177e4SLinus Torvalds 11151da177e4SLinus Torvalds if (*name=='/') { 1116e518ddb7SAndreas Mohr read_lock(&fs->lock); 1117e518ddb7SAndreas Mohr if (fs->altroot && !(nd->flags & LOOKUP_NOALT)) { 1118e518ddb7SAndreas Mohr nd->mnt = mntget(fs->altrootmnt); 1119e518ddb7SAndreas Mohr nd->dentry = dget(fs->altroot); 1120e518ddb7SAndreas Mohr read_unlock(&fs->lock); 11211da177e4SLinus Torvalds if (__emul_lookup_dentry(name,nd)) 1122ea3834d9SPrasanna Meda goto out; /* found in altroot */ 1123e518ddb7SAndreas Mohr read_lock(&fs->lock); 11241da177e4SLinus Torvalds } 1125e518ddb7SAndreas Mohr nd->mnt = mntget(fs->rootmnt); 1126e518ddb7SAndreas Mohr nd->dentry = dget(fs->root); 1127e518ddb7SAndreas Mohr read_unlock(&fs->lock); 11285590ff0dSUlrich Drepper } else if (dfd == AT_FDCWD) { 1129e518ddb7SAndreas Mohr read_lock(&fs->lock); 1130e518ddb7SAndreas Mohr nd->mnt = mntget(fs->pwdmnt); 1131e518ddb7SAndreas Mohr nd->dentry = dget(fs->pwd); 1132e518ddb7SAndreas Mohr read_unlock(&fs->lock); 11335590ff0dSUlrich Drepper } else { 11345590ff0dSUlrich Drepper struct dentry *dentry; 11355590ff0dSUlrich Drepper 11365590ff0dSUlrich Drepper file = fget_light(dfd, &fput_needed); 11375590ff0dSUlrich Drepper retval = -EBADF; 1138170aa3d0SUlrich Drepper if (!file) 11396d09bb62STrond Myklebust goto out_fail; 11405590ff0dSUlrich Drepper 11410f7fc9e4SJosef "Jeff" Sipek dentry = file->f_path.dentry; 11425590ff0dSUlrich Drepper 11435590ff0dSUlrich Drepper retval = -ENOTDIR; 1144170aa3d0SUlrich Drepper if (!S_ISDIR(dentry->d_inode->i_mode)) 11456d09bb62STrond Myklebust goto fput_fail; 11465590ff0dSUlrich Drepper 11475590ff0dSUlrich Drepper retval = file_permission(file, MAY_EXEC); 1148170aa3d0SUlrich Drepper if (retval) 11496d09bb62STrond Myklebust goto fput_fail; 11505590ff0dSUlrich Drepper 11510f7fc9e4SJosef "Jeff" Sipek nd->mnt = mntget(file->f_path.mnt); 11525590ff0dSUlrich Drepper nd->dentry = dget(dentry); 11535590ff0dSUlrich Drepper 11545590ff0dSUlrich Drepper fput_light(file, fput_needed); 11551da177e4SLinus Torvalds } 11561da177e4SLinus Torvalds current->total_link_count = 0; 11571da177e4SLinus Torvalds retval = link_path_walk(name, nd); 1158ea3834d9SPrasanna Meda out: 11593bc8414bSSuzuki if (likely(retval == 0)) { 11605ac3a9c2SAl Viro if (unlikely(!audit_dummy_context() && nd && nd->dentry && 11613bc8414bSSuzuki nd->dentry->d_inode)) 11629c937dccSAmy Griffis audit_inode(name, nd->dentry->d_inode); 11633bc8414bSSuzuki } 11646d09bb62STrond Myklebust out_fail: 1165170aa3d0SUlrich Drepper return retval; 1166170aa3d0SUlrich Drepper 11676d09bb62STrond Myklebust fput_fail: 1168170aa3d0SUlrich Drepper fput_light(file, fput_needed); 11696d09bb62STrond Myklebust goto out_fail; 11701da177e4SLinus Torvalds } 11711da177e4SLinus Torvalds 11725590ff0dSUlrich Drepper int fastcall path_lookup(const char *name, unsigned int flags, 11735590ff0dSUlrich Drepper struct nameidata *nd) 11745590ff0dSUlrich Drepper { 11755590ff0dSUlrich Drepper return do_path_lookup(AT_FDCWD, name, flags, nd); 11765590ff0dSUlrich Drepper } 11775590ff0dSUlrich Drepper 11785590ff0dSUlrich Drepper static int __path_lookup_intent_open(int dfd, const char *name, 11795590ff0dSUlrich Drepper unsigned int lookup_flags, struct nameidata *nd, 11805590ff0dSUlrich Drepper int open_flags, int create_mode) 1181834f2a4aSTrond Myklebust { 1182834f2a4aSTrond Myklebust struct file *filp = get_empty_filp(); 1183834f2a4aSTrond Myklebust int err; 1184834f2a4aSTrond Myklebust 1185834f2a4aSTrond Myklebust if (filp == NULL) 1186834f2a4aSTrond Myklebust return -ENFILE; 1187834f2a4aSTrond Myklebust nd->intent.open.file = filp; 1188834f2a4aSTrond Myklebust nd->intent.open.flags = open_flags; 1189834f2a4aSTrond Myklebust nd->intent.open.create_mode = create_mode; 11905590ff0dSUlrich Drepper err = do_path_lookup(dfd, name, lookup_flags|LOOKUP_OPEN, nd); 1191834f2a4aSTrond Myklebust if (IS_ERR(nd->intent.open.file)) { 1192834f2a4aSTrond Myklebust if (err == 0) { 1193834f2a4aSTrond Myklebust err = PTR_ERR(nd->intent.open.file); 1194834f2a4aSTrond Myklebust path_release(nd); 1195834f2a4aSTrond Myklebust } 1196834f2a4aSTrond Myklebust } else if (err != 0) 1197834f2a4aSTrond Myklebust release_open_intent(nd); 1198834f2a4aSTrond Myklebust return err; 1199834f2a4aSTrond Myklebust } 1200834f2a4aSTrond Myklebust 1201834f2a4aSTrond Myklebust /** 1202834f2a4aSTrond Myklebust * path_lookup_open - lookup a file path with open intent 12037045f37bSMartin Waitz * @dfd: the directory to use as base, or AT_FDCWD 1204834f2a4aSTrond Myklebust * @name: pointer to file name 1205834f2a4aSTrond Myklebust * @lookup_flags: lookup intent flags 1206834f2a4aSTrond Myklebust * @nd: pointer to nameidata 1207834f2a4aSTrond Myklebust * @open_flags: open intent flags 1208834f2a4aSTrond Myklebust */ 12095590ff0dSUlrich Drepper int path_lookup_open(int dfd, const char *name, unsigned int lookup_flags, 1210834f2a4aSTrond Myklebust struct nameidata *nd, int open_flags) 1211834f2a4aSTrond Myklebust { 12125590ff0dSUlrich Drepper return __path_lookup_intent_open(dfd, name, lookup_flags, nd, 1213834f2a4aSTrond Myklebust open_flags, 0); 1214834f2a4aSTrond Myklebust } 1215834f2a4aSTrond Myklebust 1216834f2a4aSTrond Myklebust /** 1217834f2a4aSTrond Myklebust * path_lookup_create - lookup a file path with open + create intent 12187045f37bSMartin Waitz * @dfd: the directory to use as base, or AT_FDCWD 1219834f2a4aSTrond Myklebust * @name: pointer to file name 1220834f2a4aSTrond Myklebust * @lookup_flags: lookup intent flags 1221834f2a4aSTrond Myklebust * @nd: pointer to nameidata 1222834f2a4aSTrond Myklebust * @open_flags: open intent flags 1223834f2a4aSTrond Myklebust * @create_mode: create intent flags 1224834f2a4aSTrond Myklebust */ 12255590ff0dSUlrich Drepper static int path_lookup_create(int dfd, const char *name, 12265590ff0dSUlrich Drepper unsigned int lookup_flags, struct nameidata *nd, 12275590ff0dSUlrich Drepper int open_flags, int create_mode) 1228834f2a4aSTrond Myklebust { 12295590ff0dSUlrich Drepper return __path_lookup_intent_open(dfd, name, lookup_flags|LOOKUP_CREATE, 12305590ff0dSUlrich Drepper nd, open_flags, create_mode); 1231834f2a4aSTrond Myklebust } 1232834f2a4aSTrond Myklebust 1233834f2a4aSTrond Myklebust int __user_path_lookup_open(const char __user *name, unsigned int lookup_flags, 1234834f2a4aSTrond Myklebust struct nameidata *nd, int open_flags) 1235834f2a4aSTrond Myklebust { 1236834f2a4aSTrond Myklebust char *tmp = getname(name); 1237834f2a4aSTrond Myklebust int err = PTR_ERR(tmp); 1238834f2a4aSTrond Myklebust 1239834f2a4aSTrond Myklebust if (!IS_ERR(tmp)) { 12405590ff0dSUlrich Drepper err = __path_lookup_intent_open(AT_FDCWD, tmp, lookup_flags, nd, open_flags, 0); 1241834f2a4aSTrond Myklebust putname(tmp); 1242834f2a4aSTrond Myklebust } 1243834f2a4aSTrond Myklebust return err; 1244834f2a4aSTrond Myklebust } 1245834f2a4aSTrond Myklebust 1246057f6c01SJames Morris static inline struct dentry *__lookup_hash_kern(struct qstr *name, struct dentry *base, struct nameidata *nd) 12471da177e4SLinus Torvalds { 12481da177e4SLinus Torvalds struct dentry *dentry; 12491da177e4SLinus Torvalds struct inode *inode; 12501da177e4SLinus Torvalds int err; 12511da177e4SLinus Torvalds 12521da177e4SLinus Torvalds inode = base->d_inode; 12531da177e4SLinus Torvalds 12541da177e4SLinus Torvalds /* 12551da177e4SLinus Torvalds * See if the low-level filesystem might want 12561da177e4SLinus Torvalds * to use its own hash.. 12571da177e4SLinus Torvalds */ 12581da177e4SLinus Torvalds if (base->d_op && base->d_op->d_hash) { 12591da177e4SLinus Torvalds err = base->d_op->d_hash(base, name); 12601da177e4SLinus Torvalds dentry = ERR_PTR(err); 12611da177e4SLinus Torvalds if (err < 0) 12621da177e4SLinus Torvalds goto out; 12631da177e4SLinus Torvalds } 12641da177e4SLinus Torvalds 12651da177e4SLinus Torvalds dentry = cached_lookup(base, name, nd); 12661da177e4SLinus Torvalds if (!dentry) { 12671da177e4SLinus Torvalds struct dentry *new = d_alloc(base, name); 12681da177e4SLinus Torvalds dentry = ERR_PTR(-ENOMEM); 12691da177e4SLinus Torvalds if (!new) 12701da177e4SLinus Torvalds goto out; 12711da177e4SLinus Torvalds dentry = inode->i_op->lookup(inode, new, nd); 12721da177e4SLinus Torvalds if (!dentry) 12731da177e4SLinus Torvalds dentry = new; 12741da177e4SLinus Torvalds else 12751da177e4SLinus Torvalds dput(new); 12761da177e4SLinus Torvalds } 12771da177e4SLinus Torvalds out: 12781da177e4SLinus Torvalds return dentry; 12791da177e4SLinus Torvalds } 12801da177e4SLinus Torvalds 1281057f6c01SJames Morris /* 1282057f6c01SJames Morris * Restricted form of lookup. Doesn't follow links, single-component only, 1283057f6c01SJames Morris * needs parent already locked. Doesn't follow mounts. 1284057f6c01SJames Morris * SMP-safe. 1285057f6c01SJames Morris */ 1286057f6c01SJames Morris static inline struct dentry * __lookup_hash(struct qstr *name, struct dentry *base, struct nameidata *nd) 1287057f6c01SJames Morris { 1288057f6c01SJames Morris struct dentry *dentry; 1289057f6c01SJames Morris struct inode *inode; 1290057f6c01SJames Morris int err; 1291057f6c01SJames Morris 1292057f6c01SJames Morris inode = base->d_inode; 1293057f6c01SJames Morris 1294057f6c01SJames Morris err = permission(inode, MAY_EXEC, nd); 1295057f6c01SJames Morris dentry = ERR_PTR(err); 1296057f6c01SJames Morris if (err) 1297057f6c01SJames Morris goto out; 1298057f6c01SJames Morris 1299057f6c01SJames Morris dentry = __lookup_hash_kern(name, base, nd); 1300057f6c01SJames Morris out: 1301057f6c01SJames Morris return dentry; 1302057f6c01SJames Morris } 1303057f6c01SJames Morris 1304a244e169SAdrian Bunk static struct dentry *lookup_hash(struct nameidata *nd) 13051da177e4SLinus Torvalds { 130649705b77SChristoph Hellwig return __lookup_hash(&nd->last, nd->dentry, nd); 13071da177e4SLinus Torvalds } 13081da177e4SLinus Torvalds 13091da177e4SLinus Torvalds /* SMP-safe */ 1310057f6c01SJames Morris static inline int __lookup_one_len(const char *name, struct qstr *this, struct dentry *base, int len) 13111da177e4SLinus Torvalds { 13121da177e4SLinus Torvalds unsigned long hash; 13131da177e4SLinus Torvalds unsigned int c; 13141da177e4SLinus Torvalds 1315057f6c01SJames Morris this->name = name; 1316057f6c01SJames Morris this->len = len; 13171da177e4SLinus Torvalds if (!len) 1318057f6c01SJames Morris return -EACCES; 13191da177e4SLinus Torvalds 13201da177e4SLinus Torvalds hash = init_name_hash(); 13211da177e4SLinus Torvalds while (len--) { 13221da177e4SLinus Torvalds c = *(const unsigned char *)name++; 13231da177e4SLinus Torvalds if (c == '/' || c == '\0') 1324057f6c01SJames Morris return -EACCES; 13251da177e4SLinus Torvalds hash = partial_name_hash(c, hash); 13261da177e4SLinus Torvalds } 1327057f6c01SJames Morris this->hash = end_name_hash(hash); 1328057f6c01SJames Morris return 0; 1329057f6c01SJames Morris } 13301da177e4SLinus Torvalds 1331057f6c01SJames Morris struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 1332057f6c01SJames Morris { 1333057f6c01SJames Morris int err; 1334057f6c01SJames Morris struct qstr this; 1335057f6c01SJames Morris 1336057f6c01SJames Morris err = __lookup_one_len(name, &this, base, len); 1337057f6c01SJames Morris if (err) 1338057f6c01SJames Morris return ERR_PTR(err); 133949705b77SChristoph Hellwig return __lookup_hash(&this, base, NULL); 1340057f6c01SJames Morris } 1341057f6c01SJames Morris 1342057f6c01SJames Morris struct dentry *lookup_one_len_kern(const char *name, struct dentry *base, int len) 1343057f6c01SJames Morris { 1344057f6c01SJames Morris int err; 1345057f6c01SJames Morris struct qstr this; 1346057f6c01SJames Morris 1347057f6c01SJames Morris err = __lookup_one_len(name, &this, base, len); 1348057f6c01SJames Morris if (err) 1349057f6c01SJames Morris return ERR_PTR(err); 1350057f6c01SJames Morris return __lookup_hash_kern(&this, base, NULL); 13511da177e4SLinus Torvalds } 13521da177e4SLinus Torvalds 13531da177e4SLinus Torvalds /* 13541da177e4SLinus Torvalds * namei() 13551da177e4SLinus Torvalds * 13561da177e4SLinus Torvalds * is used by most simple commands to get the inode of a specified name. 13571da177e4SLinus Torvalds * Open, link etc use their own routines, but this is enough for things 13581da177e4SLinus Torvalds * like 'chmod' etc. 13591da177e4SLinus Torvalds * 13601da177e4SLinus Torvalds * namei exists in two versions: namei/lnamei. The only difference is 13611da177e4SLinus Torvalds * that namei follows links, while lnamei does not. 13621da177e4SLinus Torvalds * SMP-safe 13631da177e4SLinus Torvalds */ 13645590ff0dSUlrich Drepper int fastcall __user_walk_fd(int dfd, const char __user *name, unsigned flags, 13655590ff0dSUlrich Drepper struct nameidata *nd) 13661da177e4SLinus Torvalds { 13671da177e4SLinus Torvalds char *tmp = getname(name); 13681da177e4SLinus Torvalds int err = PTR_ERR(tmp); 13691da177e4SLinus Torvalds 13701da177e4SLinus Torvalds if (!IS_ERR(tmp)) { 13715590ff0dSUlrich Drepper err = do_path_lookup(dfd, tmp, flags, nd); 13721da177e4SLinus Torvalds putname(tmp); 13731da177e4SLinus Torvalds } 13741da177e4SLinus Torvalds return err; 13751da177e4SLinus Torvalds } 13761da177e4SLinus Torvalds 13775590ff0dSUlrich Drepper int fastcall __user_walk(const char __user *name, unsigned flags, struct nameidata *nd) 13785590ff0dSUlrich Drepper { 13795590ff0dSUlrich Drepper return __user_walk_fd(AT_FDCWD, name, flags, nd); 13805590ff0dSUlrich Drepper } 13815590ff0dSUlrich Drepper 13821da177e4SLinus Torvalds /* 13831da177e4SLinus Torvalds * It's inline, so penalty for filesystems that don't use sticky bit is 13841da177e4SLinus Torvalds * minimal. 13851da177e4SLinus Torvalds */ 13861da177e4SLinus Torvalds static inline int check_sticky(struct inode *dir, struct inode *inode) 13871da177e4SLinus Torvalds { 13881da177e4SLinus Torvalds if (!(dir->i_mode & S_ISVTX)) 13891da177e4SLinus Torvalds return 0; 13901da177e4SLinus Torvalds if (inode->i_uid == current->fsuid) 13911da177e4SLinus Torvalds return 0; 13921da177e4SLinus Torvalds if (dir->i_uid == current->fsuid) 13931da177e4SLinus Torvalds return 0; 13941da177e4SLinus Torvalds return !capable(CAP_FOWNER); 13951da177e4SLinus Torvalds } 13961da177e4SLinus Torvalds 13971da177e4SLinus Torvalds /* 13981da177e4SLinus Torvalds * Check whether we can remove a link victim from directory dir, check 13991da177e4SLinus Torvalds * whether the type of victim is right. 14001da177e4SLinus Torvalds * 1. We can't do it if dir is read-only (done in permission()) 14011da177e4SLinus Torvalds * 2. We should have write and exec permissions on dir 14021da177e4SLinus Torvalds * 3. We can't remove anything from append-only dir 14031da177e4SLinus Torvalds * 4. We can't do anything with immutable dir (done in permission()) 14041da177e4SLinus Torvalds * 5. If the sticky bit on dir is set we should either 14051da177e4SLinus Torvalds * a. be owner of dir, or 14061da177e4SLinus Torvalds * b. be owner of victim, or 14071da177e4SLinus Torvalds * c. have CAP_FOWNER capability 14081da177e4SLinus Torvalds * 6. If the victim is append-only or immutable we can't do antyhing with 14091da177e4SLinus Torvalds * links pointing to it. 14101da177e4SLinus Torvalds * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. 14111da177e4SLinus Torvalds * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. 14121da177e4SLinus Torvalds * 9. We can't remove a root or mountpoint. 14131da177e4SLinus Torvalds * 10. We don't allow removal of NFS sillyrenamed files; it's handled by 14141da177e4SLinus Torvalds * nfs_async_unlink(). 14151da177e4SLinus Torvalds */ 1416858119e1SArjan van de Ven static int may_delete(struct inode *dir,struct dentry *victim,int isdir) 14171da177e4SLinus Torvalds { 14181da177e4SLinus Torvalds int error; 14191da177e4SLinus Torvalds 14201da177e4SLinus Torvalds if (!victim->d_inode) 14211da177e4SLinus Torvalds return -ENOENT; 14221da177e4SLinus Torvalds 14231da177e4SLinus Torvalds BUG_ON(victim->d_parent->d_inode != dir); 142473d3ec5aSAmy Griffis audit_inode_child(victim->d_name.name, victim->d_inode, dir); 14251da177e4SLinus Torvalds 14261da177e4SLinus Torvalds error = permission(dir,MAY_WRITE | MAY_EXEC, NULL); 14271da177e4SLinus Torvalds if (error) 14281da177e4SLinus Torvalds return error; 14291da177e4SLinus Torvalds if (IS_APPEND(dir)) 14301da177e4SLinus Torvalds return -EPERM; 14311da177e4SLinus Torvalds if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)|| 14321da177e4SLinus Torvalds IS_IMMUTABLE(victim->d_inode)) 14331da177e4SLinus Torvalds return -EPERM; 14341da177e4SLinus Torvalds if (isdir) { 14351da177e4SLinus Torvalds if (!S_ISDIR(victim->d_inode->i_mode)) 14361da177e4SLinus Torvalds return -ENOTDIR; 14371da177e4SLinus Torvalds if (IS_ROOT(victim)) 14381da177e4SLinus Torvalds return -EBUSY; 14391da177e4SLinus Torvalds } else if (S_ISDIR(victim->d_inode->i_mode)) 14401da177e4SLinus Torvalds return -EISDIR; 14411da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 14421da177e4SLinus Torvalds return -ENOENT; 14431da177e4SLinus Torvalds if (victim->d_flags & DCACHE_NFSFS_RENAMED) 14441da177e4SLinus Torvalds return -EBUSY; 14451da177e4SLinus Torvalds return 0; 14461da177e4SLinus Torvalds } 14471da177e4SLinus Torvalds 14481da177e4SLinus Torvalds /* Check whether we can create an object with dentry child in directory 14491da177e4SLinus Torvalds * dir. 14501da177e4SLinus Torvalds * 1. We can't do it if child already exists (open has special treatment for 14511da177e4SLinus Torvalds * this case, but since we are inlined it's OK) 14521da177e4SLinus Torvalds * 2. We can't do it if dir is read-only (done in permission()) 14531da177e4SLinus Torvalds * 3. We should have write and exec permissions on dir 14541da177e4SLinus Torvalds * 4. We can't do it if dir is immutable (done in permission()) 14551da177e4SLinus Torvalds */ 14561da177e4SLinus Torvalds static inline int may_create(struct inode *dir, struct dentry *child, 14571da177e4SLinus Torvalds struct nameidata *nd) 14581da177e4SLinus Torvalds { 14591da177e4SLinus Torvalds if (child->d_inode) 14601da177e4SLinus Torvalds return -EEXIST; 14611da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 14621da177e4SLinus Torvalds return -ENOENT; 14631da177e4SLinus Torvalds return permission(dir,MAY_WRITE | MAY_EXEC, nd); 14641da177e4SLinus Torvalds } 14651da177e4SLinus Torvalds 14661da177e4SLinus Torvalds /* 14671da177e4SLinus Torvalds * O_DIRECTORY translates into forcing a directory lookup. 14681da177e4SLinus Torvalds */ 14691da177e4SLinus Torvalds static inline int lookup_flags(unsigned int f) 14701da177e4SLinus Torvalds { 14711da177e4SLinus Torvalds unsigned long retval = LOOKUP_FOLLOW; 14721da177e4SLinus Torvalds 14731da177e4SLinus Torvalds if (f & O_NOFOLLOW) 14741da177e4SLinus Torvalds retval &= ~LOOKUP_FOLLOW; 14751da177e4SLinus Torvalds 14761da177e4SLinus Torvalds if (f & O_DIRECTORY) 14771da177e4SLinus Torvalds retval |= LOOKUP_DIRECTORY; 14781da177e4SLinus Torvalds 14791da177e4SLinus Torvalds return retval; 14801da177e4SLinus Torvalds } 14811da177e4SLinus Torvalds 14821da177e4SLinus Torvalds /* 14831da177e4SLinus Torvalds * p1 and p2 should be directories on the same fs. 14841da177e4SLinus Torvalds */ 14851da177e4SLinus Torvalds struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 14861da177e4SLinus Torvalds { 14871da177e4SLinus Torvalds struct dentry *p; 14881da177e4SLinus Torvalds 14891da177e4SLinus Torvalds if (p1 == p2) { 1490f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 14911da177e4SLinus Torvalds return NULL; 14921da177e4SLinus Torvalds } 14931da177e4SLinus Torvalds 1494a11f3a05SArjan van de Ven mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 14951da177e4SLinus Torvalds 14961da177e4SLinus Torvalds for (p = p1; p->d_parent != p; p = p->d_parent) { 14971da177e4SLinus Torvalds if (p->d_parent == p2) { 1498f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT); 1499f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD); 15001da177e4SLinus Torvalds return p; 15011da177e4SLinus Torvalds } 15021da177e4SLinus Torvalds } 15031da177e4SLinus Torvalds 15041da177e4SLinus Torvalds for (p = p2; p->d_parent != p; p = p->d_parent) { 15051da177e4SLinus Torvalds if (p->d_parent == p1) { 1506f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 1507f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); 15081da177e4SLinus Torvalds return p; 15091da177e4SLinus Torvalds } 15101da177e4SLinus Torvalds } 15111da177e4SLinus Torvalds 1512f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 1513f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); 15141da177e4SLinus Torvalds return NULL; 15151da177e4SLinus Torvalds } 15161da177e4SLinus Torvalds 15171da177e4SLinus Torvalds void unlock_rename(struct dentry *p1, struct dentry *p2) 15181da177e4SLinus Torvalds { 15191b1dcc1bSJes Sorensen mutex_unlock(&p1->d_inode->i_mutex); 15201da177e4SLinus Torvalds if (p1 != p2) { 15211b1dcc1bSJes Sorensen mutex_unlock(&p2->d_inode->i_mutex); 1522a11f3a05SArjan van de Ven mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 15231da177e4SLinus Torvalds } 15241da177e4SLinus Torvalds } 15251da177e4SLinus Torvalds 15261da177e4SLinus Torvalds int vfs_create(struct inode *dir, struct dentry *dentry, int mode, 15271da177e4SLinus Torvalds struct nameidata *nd) 15281da177e4SLinus Torvalds { 15291da177e4SLinus Torvalds int error = may_create(dir, dentry, nd); 15301da177e4SLinus Torvalds 15311da177e4SLinus Torvalds if (error) 15321da177e4SLinus Torvalds return error; 15331da177e4SLinus Torvalds 15341da177e4SLinus Torvalds if (!dir->i_op || !dir->i_op->create) 15351da177e4SLinus Torvalds return -EACCES; /* shouldn't it be ENOSYS? */ 15361da177e4SLinus Torvalds mode &= S_IALLUGO; 15371da177e4SLinus Torvalds mode |= S_IFREG; 15381da177e4SLinus Torvalds error = security_inode_create(dir, dentry, mode); 15391da177e4SLinus Torvalds if (error) 15401da177e4SLinus Torvalds return error; 15411da177e4SLinus Torvalds DQUOT_INIT(dir); 15421da177e4SLinus Torvalds error = dir->i_op->create(dir, dentry, mode, nd); 1543a74574aaSStephen Smalley if (!error) 1544f38aa942SAmy Griffis fsnotify_create(dir, dentry); 15451da177e4SLinus Torvalds return error; 15461da177e4SLinus Torvalds } 15471da177e4SLinus Torvalds 15481da177e4SLinus Torvalds int may_open(struct nameidata *nd, int acc_mode, int flag) 15491da177e4SLinus Torvalds { 15501da177e4SLinus Torvalds struct dentry *dentry = nd->dentry; 15511da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 15521da177e4SLinus Torvalds int error; 15531da177e4SLinus Torvalds 15541da177e4SLinus Torvalds if (!inode) 15551da177e4SLinus Torvalds return -ENOENT; 15561da177e4SLinus Torvalds 15571da177e4SLinus Torvalds if (S_ISLNK(inode->i_mode)) 15581da177e4SLinus Torvalds return -ELOOP; 15591da177e4SLinus Torvalds 15601da177e4SLinus Torvalds if (S_ISDIR(inode->i_mode) && (flag & FMODE_WRITE)) 15611da177e4SLinus Torvalds return -EISDIR; 15621da177e4SLinus Torvalds 1563e4543eddSChristoph Hellwig error = vfs_permission(nd, acc_mode); 15641da177e4SLinus Torvalds if (error) 15651da177e4SLinus Torvalds return error; 15661da177e4SLinus Torvalds 15671da177e4SLinus Torvalds /* 15681da177e4SLinus Torvalds * FIFO's, sockets and device files are special: they don't 15691da177e4SLinus Torvalds * actually live on the filesystem itself, and as such you 15701da177e4SLinus Torvalds * can write to them even if the filesystem is read-only. 15711da177e4SLinus Torvalds */ 15721da177e4SLinus Torvalds if (S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) { 15731da177e4SLinus Torvalds flag &= ~O_TRUNC; 15741da177e4SLinus Torvalds } else if (S_ISBLK(inode->i_mode) || S_ISCHR(inode->i_mode)) { 15751da177e4SLinus Torvalds if (nd->mnt->mnt_flags & MNT_NODEV) 15761da177e4SLinus Torvalds return -EACCES; 15771da177e4SLinus Torvalds 15781da177e4SLinus Torvalds flag &= ~O_TRUNC; 15791da177e4SLinus Torvalds } else if (IS_RDONLY(inode) && (flag & FMODE_WRITE)) 15801da177e4SLinus Torvalds return -EROFS; 15811da177e4SLinus Torvalds /* 15821da177e4SLinus Torvalds * An append-only file must be opened in append mode for writing. 15831da177e4SLinus Torvalds */ 15841da177e4SLinus Torvalds if (IS_APPEND(inode)) { 15851da177e4SLinus Torvalds if ((flag & FMODE_WRITE) && !(flag & O_APPEND)) 15861da177e4SLinus Torvalds return -EPERM; 15871da177e4SLinus Torvalds if (flag & O_TRUNC) 15881da177e4SLinus Torvalds return -EPERM; 15891da177e4SLinus Torvalds } 15901da177e4SLinus Torvalds 15911da177e4SLinus Torvalds /* O_NOATIME can only be set by the owner or superuser */ 15921da177e4SLinus Torvalds if (flag & O_NOATIME) 15931da177e4SLinus Torvalds if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER)) 15941da177e4SLinus Torvalds return -EPERM; 15951da177e4SLinus Torvalds 15961da177e4SLinus Torvalds /* 15971da177e4SLinus Torvalds * Ensure there are no outstanding leases on the file. 15981da177e4SLinus Torvalds */ 15991da177e4SLinus Torvalds error = break_lease(inode, flag); 16001da177e4SLinus Torvalds if (error) 16011da177e4SLinus Torvalds return error; 16021da177e4SLinus Torvalds 16031da177e4SLinus Torvalds if (flag & O_TRUNC) { 16041da177e4SLinus Torvalds error = get_write_access(inode); 16051da177e4SLinus Torvalds if (error) 16061da177e4SLinus Torvalds return error; 16071da177e4SLinus Torvalds 16081da177e4SLinus Torvalds /* 16091da177e4SLinus Torvalds * Refuse to truncate files with mandatory locks held on them. 16101da177e4SLinus Torvalds */ 16111da177e4SLinus Torvalds error = locks_verify_locked(inode); 16121da177e4SLinus Torvalds if (!error) { 16131da177e4SLinus Torvalds DQUOT_INIT(inode); 16141da177e4SLinus Torvalds 16154a30131eSNeilBrown error = do_truncate(dentry, 0, ATTR_MTIME|ATTR_CTIME, NULL); 16161da177e4SLinus Torvalds } 16171da177e4SLinus Torvalds put_write_access(inode); 16181da177e4SLinus Torvalds if (error) 16191da177e4SLinus Torvalds return error; 16201da177e4SLinus Torvalds } else 16211da177e4SLinus Torvalds if (flag & FMODE_WRITE) 16221da177e4SLinus Torvalds DQUOT_INIT(inode); 16231da177e4SLinus Torvalds 16241da177e4SLinus Torvalds return 0; 16251da177e4SLinus Torvalds } 16261da177e4SLinus Torvalds 1627aab520e2SDave Hansen static int open_namei_create(struct nameidata *nd, struct path *path, 1628aab520e2SDave Hansen int flag, int mode) 1629aab520e2SDave Hansen { 1630aab520e2SDave Hansen int error; 1631aab520e2SDave Hansen struct dentry *dir = nd->dentry; 1632aab520e2SDave Hansen 1633aab520e2SDave Hansen if (!IS_POSIXACL(dir->d_inode)) 1634aab520e2SDave Hansen mode &= ~current->fs->umask; 1635aab520e2SDave Hansen error = vfs_create(dir->d_inode, path->dentry, mode, nd); 1636aab520e2SDave Hansen mutex_unlock(&dir->d_inode->i_mutex); 1637aab520e2SDave Hansen dput(nd->dentry); 1638aab520e2SDave Hansen nd->dentry = path->dentry; 1639aab520e2SDave Hansen if (error) 1640aab520e2SDave Hansen return error; 1641aab520e2SDave Hansen /* Don't check for write permission, don't truncate */ 1642aab520e2SDave Hansen return may_open(nd, 0, flag & ~O_TRUNC); 1643aab520e2SDave Hansen } 1644aab520e2SDave Hansen 16451da177e4SLinus Torvalds /* 16461da177e4SLinus Torvalds * open_namei() 16471da177e4SLinus Torvalds * 16481da177e4SLinus Torvalds * namei for open - this is in fact almost the whole open-routine. 16491da177e4SLinus Torvalds * 16501da177e4SLinus Torvalds * Note that the low bits of "flag" aren't the same as in the open 16511da177e4SLinus Torvalds * system call - they are 00 - no permissions needed 16521da177e4SLinus Torvalds * 01 - read permission needed 16531da177e4SLinus Torvalds * 10 - write permission needed 16541da177e4SLinus Torvalds * 11 - read/write permissions needed 16551da177e4SLinus Torvalds * which is a lot more logical, and also allows the "no perm" needed 16561da177e4SLinus Torvalds * for symlinks (where the permissions are checked later). 16571da177e4SLinus Torvalds * SMP-safe 16581da177e4SLinus Torvalds */ 16595590ff0dSUlrich Drepper int open_namei(int dfd, const char *pathname, int flag, 16605590ff0dSUlrich Drepper int mode, struct nameidata *nd) 16611da177e4SLinus Torvalds { 1662834f2a4aSTrond Myklebust int acc_mode, error; 16634e7506e4SAl Viro struct path path; 16641da177e4SLinus Torvalds struct dentry *dir; 16651da177e4SLinus Torvalds int count = 0; 16661da177e4SLinus Torvalds 16671da177e4SLinus Torvalds acc_mode = ACC_MODE(flag); 16681da177e4SLinus Torvalds 1669834f2a4aSTrond Myklebust /* O_TRUNC implies we need access checks for write permissions */ 1670834f2a4aSTrond Myklebust if (flag & O_TRUNC) 1671834f2a4aSTrond Myklebust acc_mode |= MAY_WRITE; 1672834f2a4aSTrond Myklebust 16731da177e4SLinus Torvalds /* Allow the LSM permission hook to distinguish append 16741da177e4SLinus Torvalds access from general write access. */ 16751da177e4SLinus Torvalds if (flag & O_APPEND) 16761da177e4SLinus Torvalds acc_mode |= MAY_APPEND; 16771da177e4SLinus Torvalds 16781da177e4SLinus Torvalds /* 16791da177e4SLinus Torvalds * The simplest case - just a plain lookup. 16801da177e4SLinus Torvalds */ 16811da177e4SLinus Torvalds if (!(flag & O_CREAT)) { 16825590ff0dSUlrich Drepper error = path_lookup_open(dfd, pathname, lookup_flags(flag), 16835590ff0dSUlrich Drepper nd, flag); 16841da177e4SLinus Torvalds if (error) 16851da177e4SLinus Torvalds return error; 16861da177e4SLinus Torvalds goto ok; 16871da177e4SLinus Torvalds } 16881da177e4SLinus Torvalds 16891da177e4SLinus Torvalds /* 16901da177e4SLinus Torvalds * Create - we need to know the parent. 16911da177e4SLinus Torvalds */ 16925590ff0dSUlrich Drepper error = path_lookup_create(dfd,pathname,LOOKUP_PARENT,nd,flag,mode); 16931da177e4SLinus Torvalds if (error) 16941da177e4SLinus Torvalds return error; 16951da177e4SLinus Torvalds 16961da177e4SLinus Torvalds /* 16971da177e4SLinus Torvalds * We have the parent and last component. First of all, check 16981da177e4SLinus Torvalds * that we are not asked to creat(2) an obvious directory - that 16991da177e4SLinus Torvalds * will not do. 17001da177e4SLinus Torvalds */ 17011da177e4SLinus Torvalds error = -EISDIR; 17021da177e4SLinus Torvalds if (nd->last_type != LAST_NORM || nd->last.name[nd->last.len]) 17031da177e4SLinus Torvalds goto exit; 17041da177e4SLinus Torvalds 17051da177e4SLinus Torvalds dir = nd->dentry; 17061da177e4SLinus Torvalds nd->flags &= ~LOOKUP_PARENT; 17071b1dcc1bSJes Sorensen mutex_lock(&dir->d_inode->i_mutex); 170849705b77SChristoph Hellwig path.dentry = lookup_hash(nd); 1709d73ffe16SAl Viro path.mnt = nd->mnt; 17101da177e4SLinus Torvalds 17111da177e4SLinus Torvalds do_last: 17124e7506e4SAl Viro error = PTR_ERR(path.dentry); 17134e7506e4SAl Viro if (IS_ERR(path.dentry)) { 17141b1dcc1bSJes Sorensen mutex_unlock(&dir->d_inode->i_mutex); 17151da177e4SLinus Torvalds goto exit; 17161da177e4SLinus Torvalds } 17171da177e4SLinus Torvalds 17184af4c52fSOleg Drokin if (IS_ERR(nd->intent.open.file)) { 17194af4c52fSOleg Drokin mutex_unlock(&dir->d_inode->i_mutex); 17204af4c52fSOleg Drokin error = PTR_ERR(nd->intent.open.file); 17214af4c52fSOleg Drokin goto exit_dput; 17224af4c52fSOleg Drokin } 17234af4c52fSOleg Drokin 17241da177e4SLinus Torvalds /* Negative dentry, just create the file */ 17254e7506e4SAl Viro if (!path.dentry->d_inode) { 1726aab520e2SDave Hansen error = open_namei_create(nd, &path, flag, mode); 17271da177e4SLinus Torvalds if (error) 17281da177e4SLinus Torvalds goto exit; 1729aab520e2SDave Hansen return 0; 17301da177e4SLinus Torvalds } 17311da177e4SLinus Torvalds 17321da177e4SLinus Torvalds /* 17331da177e4SLinus Torvalds * It already exists. 17341da177e4SLinus Torvalds */ 17351b1dcc1bSJes Sorensen mutex_unlock(&dir->d_inode->i_mutex); 17363e2efce0SAmy Griffis audit_inode_update(path.dentry->d_inode); 17371da177e4SLinus Torvalds 17381da177e4SLinus Torvalds error = -EEXIST; 17391da177e4SLinus Torvalds if (flag & O_EXCL) 17401da177e4SLinus Torvalds goto exit_dput; 17411da177e4SLinus Torvalds 1742e13b210fSAl Viro if (__follow_mount(&path)) { 17431da177e4SLinus Torvalds error = -ELOOP; 1744ba7a4c1aSAl Viro if (flag & O_NOFOLLOW) 1745ba7a4c1aSAl Viro goto exit_dput; 17461da177e4SLinus Torvalds } 17473e2efce0SAmy Griffis 17481da177e4SLinus Torvalds error = -ENOENT; 17494e7506e4SAl Viro if (!path.dentry->d_inode) 17501da177e4SLinus Torvalds goto exit_dput; 17514e7506e4SAl Viro if (path.dentry->d_inode->i_op && path.dentry->d_inode->i_op->follow_link) 17521da177e4SLinus Torvalds goto do_link; 17531da177e4SLinus Torvalds 175409dd17d3SMiklos Szeredi path_to_nameidata(&path, nd); 17551da177e4SLinus Torvalds error = -EISDIR; 17564e7506e4SAl Viro if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode)) 17571da177e4SLinus Torvalds goto exit; 17581da177e4SLinus Torvalds ok: 17591da177e4SLinus Torvalds error = may_open(nd, acc_mode, flag); 17601da177e4SLinus Torvalds if (error) 17611da177e4SLinus Torvalds goto exit; 17621da177e4SLinus Torvalds return 0; 17631da177e4SLinus Torvalds 17641da177e4SLinus Torvalds exit_dput: 176509dd17d3SMiklos Szeredi dput_path(&path, nd); 17661da177e4SLinus Torvalds exit: 1767834f2a4aSTrond Myklebust if (!IS_ERR(nd->intent.open.file)) 1768834f2a4aSTrond Myklebust release_open_intent(nd); 17691da177e4SLinus Torvalds path_release(nd); 17701da177e4SLinus Torvalds return error; 17711da177e4SLinus Torvalds 17721da177e4SLinus Torvalds do_link: 17731da177e4SLinus Torvalds error = -ELOOP; 17741da177e4SLinus Torvalds if (flag & O_NOFOLLOW) 17751da177e4SLinus Torvalds goto exit_dput; 17761da177e4SLinus Torvalds /* 17771da177e4SLinus Torvalds * This is subtle. Instead of calling do_follow_link() we do the 17781da177e4SLinus Torvalds * thing by hands. The reason is that this way we have zero link_count 17791da177e4SLinus Torvalds * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT. 17801da177e4SLinus Torvalds * After that we have the parent and last component, i.e. 17811da177e4SLinus Torvalds * we are in the same situation as after the first path_walk(). 17821da177e4SLinus Torvalds * Well, almost - if the last component is normal we get its copy 17831da177e4SLinus Torvalds * stored in nd->last.name and we will have to putname() it when we 17841da177e4SLinus Torvalds * are done. Procfs-like symlinks just set LAST_BIND. 17851da177e4SLinus Torvalds */ 17861da177e4SLinus Torvalds nd->flags |= LOOKUP_PARENT; 17874e7506e4SAl Viro error = security_inode_follow_link(path.dentry, nd); 17881da177e4SLinus Torvalds if (error) 17891da177e4SLinus Torvalds goto exit_dput; 1790cd4e91d3SAl Viro error = __do_follow_link(&path, nd); 1791de459215SKirill Korotaev if (error) { 1792de459215SKirill Korotaev /* Does someone understand code flow here? Or it is only 1793de459215SKirill Korotaev * me so stupid? Anathema to whoever designed this non-sense 1794de459215SKirill Korotaev * with "intent.open". 1795de459215SKirill Korotaev */ 1796de459215SKirill Korotaev release_open_intent(nd); 17971da177e4SLinus Torvalds return error; 1798de459215SKirill Korotaev } 17991da177e4SLinus Torvalds nd->flags &= ~LOOKUP_PARENT; 1800d671d5e5SAl Viro if (nd->last_type == LAST_BIND) 18011da177e4SLinus Torvalds goto ok; 18021da177e4SLinus Torvalds error = -EISDIR; 18031da177e4SLinus Torvalds if (nd->last_type != LAST_NORM) 18041da177e4SLinus Torvalds goto exit; 18051da177e4SLinus Torvalds if (nd->last.name[nd->last.len]) { 180682984114SLinus Torvalds __putname(nd->last.name); 18071da177e4SLinus Torvalds goto exit; 18081da177e4SLinus Torvalds } 18091da177e4SLinus Torvalds error = -ELOOP; 18101da177e4SLinus Torvalds if (count++==32) { 181182984114SLinus Torvalds __putname(nd->last.name); 18121da177e4SLinus Torvalds goto exit; 18131da177e4SLinus Torvalds } 18141da177e4SLinus Torvalds dir = nd->dentry; 18151b1dcc1bSJes Sorensen mutex_lock(&dir->d_inode->i_mutex); 181649705b77SChristoph Hellwig path.dentry = lookup_hash(nd); 1817d671d5e5SAl Viro path.mnt = nd->mnt; 181882984114SLinus Torvalds __putname(nd->last.name); 18191da177e4SLinus Torvalds goto do_last; 18201da177e4SLinus Torvalds } 18211da177e4SLinus Torvalds 18221da177e4SLinus Torvalds /** 18231da177e4SLinus Torvalds * lookup_create - lookup a dentry, creating it if it doesn't exist 18241da177e4SLinus Torvalds * @nd: nameidata info 18251da177e4SLinus Torvalds * @is_dir: directory flag 18261da177e4SLinus Torvalds * 18271da177e4SLinus Torvalds * Simple function to lookup and return a dentry and create it 18281da177e4SLinus Torvalds * if it doesn't exist. Is SMP-safe. 1829c663e5d8SChristoph Hellwig * 18301b1dcc1bSJes Sorensen * Returns with nd->dentry->d_inode->i_mutex locked. 18311da177e4SLinus Torvalds */ 18321da177e4SLinus Torvalds struct dentry *lookup_create(struct nameidata *nd, int is_dir) 18331da177e4SLinus Torvalds { 1834c663e5d8SChristoph Hellwig struct dentry *dentry = ERR_PTR(-EEXIST); 18351da177e4SLinus Torvalds 1836f2eace23SIngo Molnar mutex_lock_nested(&nd->dentry->d_inode->i_mutex, I_MUTEX_PARENT); 1837c663e5d8SChristoph Hellwig /* 1838c663e5d8SChristoph Hellwig * Yucky last component or no last component at all? 1839c663e5d8SChristoph Hellwig * (foo/., foo/.., /////) 1840c663e5d8SChristoph Hellwig */ 18411da177e4SLinus Torvalds if (nd->last_type != LAST_NORM) 18421da177e4SLinus Torvalds goto fail; 18431da177e4SLinus Torvalds nd->flags &= ~LOOKUP_PARENT; 1844a634904aSASANO Masahiro nd->flags |= LOOKUP_CREATE; 1845a634904aSASANO Masahiro nd->intent.open.flags = O_EXCL; 1846c663e5d8SChristoph Hellwig 1847c663e5d8SChristoph Hellwig /* 1848c663e5d8SChristoph Hellwig * Do the final lookup. 1849c663e5d8SChristoph Hellwig */ 185049705b77SChristoph Hellwig dentry = lookup_hash(nd); 18511da177e4SLinus Torvalds if (IS_ERR(dentry)) 18521da177e4SLinus Torvalds goto fail; 1853c663e5d8SChristoph Hellwig 1854c663e5d8SChristoph Hellwig /* 1855c663e5d8SChristoph Hellwig * Special case - lookup gave negative, but... we had foo/bar/ 1856c663e5d8SChristoph Hellwig * From the vfs_mknod() POV we just have a negative dentry - 1857c663e5d8SChristoph Hellwig * all is fine. Let's be bastards - you had / on the end, you've 1858c663e5d8SChristoph Hellwig * been asking for (non-existent) directory. -ENOENT for you. 1859c663e5d8SChristoph Hellwig */ 18601da177e4SLinus Torvalds if (!is_dir && nd->last.name[nd->last.len] && !dentry->d_inode) 18611da177e4SLinus Torvalds goto enoent; 18621da177e4SLinus Torvalds return dentry; 18631da177e4SLinus Torvalds enoent: 18641da177e4SLinus Torvalds dput(dentry); 18651da177e4SLinus Torvalds dentry = ERR_PTR(-ENOENT); 18661da177e4SLinus Torvalds fail: 18671da177e4SLinus Torvalds return dentry; 18681da177e4SLinus Torvalds } 1869f81a0bffSChristoph Hellwig EXPORT_SYMBOL_GPL(lookup_create); 18701da177e4SLinus Torvalds 18711da177e4SLinus Torvalds int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev) 18721da177e4SLinus Torvalds { 18731da177e4SLinus Torvalds int error = may_create(dir, dentry, NULL); 18741da177e4SLinus Torvalds 18751da177e4SLinus Torvalds if (error) 18761da177e4SLinus Torvalds return error; 18771da177e4SLinus Torvalds 18781da177e4SLinus Torvalds if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) 18791da177e4SLinus Torvalds return -EPERM; 18801da177e4SLinus Torvalds 18811da177e4SLinus Torvalds if (!dir->i_op || !dir->i_op->mknod) 18821da177e4SLinus Torvalds return -EPERM; 18831da177e4SLinus Torvalds 18841da177e4SLinus Torvalds error = security_inode_mknod(dir, dentry, mode, dev); 18851da177e4SLinus Torvalds if (error) 18861da177e4SLinus Torvalds return error; 18871da177e4SLinus Torvalds 18881da177e4SLinus Torvalds DQUOT_INIT(dir); 18891da177e4SLinus Torvalds error = dir->i_op->mknod(dir, dentry, mode, dev); 1890a74574aaSStephen Smalley if (!error) 1891f38aa942SAmy Griffis fsnotify_create(dir, dentry); 18921da177e4SLinus Torvalds return error; 18931da177e4SLinus Torvalds } 18941da177e4SLinus Torvalds 18955590ff0dSUlrich Drepper asmlinkage long sys_mknodat(int dfd, const char __user *filename, int mode, 18965590ff0dSUlrich Drepper unsigned dev) 18971da177e4SLinus Torvalds { 18981da177e4SLinus Torvalds int error = 0; 18991da177e4SLinus Torvalds char * tmp; 19001da177e4SLinus Torvalds struct dentry * dentry; 19011da177e4SLinus Torvalds struct nameidata nd; 19021da177e4SLinus Torvalds 19031da177e4SLinus Torvalds if (S_ISDIR(mode)) 19041da177e4SLinus Torvalds return -EPERM; 19051da177e4SLinus Torvalds tmp = getname(filename); 19061da177e4SLinus Torvalds if (IS_ERR(tmp)) 19071da177e4SLinus Torvalds return PTR_ERR(tmp); 19081da177e4SLinus Torvalds 19095590ff0dSUlrich Drepper error = do_path_lookup(dfd, tmp, LOOKUP_PARENT, &nd); 19101da177e4SLinus Torvalds if (error) 19111da177e4SLinus Torvalds goto out; 19121da177e4SLinus Torvalds dentry = lookup_create(&nd, 0); 19131da177e4SLinus Torvalds error = PTR_ERR(dentry); 19141da177e4SLinus Torvalds 19151da177e4SLinus Torvalds if (!IS_POSIXACL(nd.dentry->d_inode)) 19161da177e4SLinus Torvalds mode &= ~current->fs->umask; 19171da177e4SLinus Torvalds if (!IS_ERR(dentry)) { 19181da177e4SLinus Torvalds switch (mode & S_IFMT) { 19191da177e4SLinus Torvalds case 0: case S_IFREG: 19201da177e4SLinus Torvalds error = vfs_create(nd.dentry->d_inode,dentry,mode,&nd); 19211da177e4SLinus Torvalds break; 19221da177e4SLinus Torvalds case S_IFCHR: case S_IFBLK: 19231da177e4SLinus Torvalds error = vfs_mknod(nd.dentry->d_inode,dentry,mode, 19241da177e4SLinus Torvalds new_decode_dev(dev)); 19251da177e4SLinus Torvalds break; 19261da177e4SLinus Torvalds case S_IFIFO: case S_IFSOCK: 19271da177e4SLinus Torvalds error = vfs_mknod(nd.dentry->d_inode,dentry,mode,0); 19281da177e4SLinus Torvalds break; 19291da177e4SLinus Torvalds case S_IFDIR: 19301da177e4SLinus Torvalds error = -EPERM; 19311da177e4SLinus Torvalds break; 19321da177e4SLinus Torvalds default: 19331da177e4SLinus Torvalds error = -EINVAL; 19341da177e4SLinus Torvalds } 19351da177e4SLinus Torvalds dput(dentry); 19361da177e4SLinus Torvalds } 19371b1dcc1bSJes Sorensen mutex_unlock(&nd.dentry->d_inode->i_mutex); 19381da177e4SLinus Torvalds path_release(&nd); 19391da177e4SLinus Torvalds out: 19401da177e4SLinus Torvalds putname(tmp); 19411da177e4SLinus Torvalds 19421da177e4SLinus Torvalds return error; 19431da177e4SLinus Torvalds } 19441da177e4SLinus Torvalds 19455590ff0dSUlrich Drepper asmlinkage long sys_mknod(const char __user *filename, int mode, unsigned dev) 19465590ff0dSUlrich Drepper { 19475590ff0dSUlrich Drepper return sys_mknodat(AT_FDCWD, filename, mode, dev); 19485590ff0dSUlrich Drepper } 19495590ff0dSUlrich Drepper 19501da177e4SLinus Torvalds int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode) 19511da177e4SLinus Torvalds { 19521da177e4SLinus Torvalds int error = may_create(dir, dentry, NULL); 19531da177e4SLinus Torvalds 19541da177e4SLinus Torvalds if (error) 19551da177e4SLinus Torvalds return error; 19561da177e4SLinus Torvalds 19571da177e4SLinus Torvalds if (!dir->i_op || !dir->i_op->mkdir) 19581da177e4SLinus Torvalds return -EPERM; 19591da177e4SLinus Torvalds 19601da177e4SLinus Torvalds mode &= (S_IRWXUGO|S_ISVTX); 19611da177e4SLinus Torvalds error = security_inode_mkdir(dir, dentry, mode); 19621da177e4SLinus Torvalds if (error) 19631da177e4SLinus Torvalds return error; 19641da177e4SLinus Torvalds 19651da177e4SLinus Torvalds DQUOT_INIT(dir); 19661da177e4SLinus Torvalds error = dir->i_op->mkdir(dir, dentry, mode); 1967a74574aaSStephen Smalley if (!error) 1968f38aa942SAmy Griffis fsnotify_mkdir(dir, dentry); 19691da177e4SLinus Torvalds return error; 19701da177e4SLinus Torvalds } 19711da177e4SLinus Torvalds 19725590ff0dSUlrich Drepper asmlinkage long sys_mkdirat(int dfd, const char __user *pathname, int mode) 19731da177e4SLinus Torvalds { 19741da177e4SLinus Torvalds int error = 0; 19751da177e4SLinus Torvalds char * tmp; 19766902d925SDave Hansen struct dentry *dentry; 19776902d925SDave Hansen struct nameidata nd; 19781da177e4SLinus Torvalds 19791da177e4SLinus Torvalds tmp = getname(pathname); 19801da177e4SLinus Torvalds error = PTR_ERR(tmp); 19816902d925SDave Hansen if (IS_ERR(tmp)) 19826902d925SDave Hansen goto out_err; 19831da177e4SLinus Torvalds 19845590ff0dSUlrich Drepper error = do_path_lookup(dfd, tmp, LOOKUP_PARENT, &nd); 19851da177e4SLinus Torvalds if (error) 19861da177e4SLinus Torvalds goto out; 19871da177e4SLinus Torvalds dentry = lookup_create(&nd, 1); 19881da177e4SLinus Torvalds error = PTR_ERR(dentry); 19896902d925SDave Hansen if (IS_ERR(dentry)) 19906902d925SDave Hansen goto out_unlock; 19916902d925SDave Hansen 19921da177e4SLinus Torvalds if (!IS_POSIXACL(nd.dentry->d_inode)) 19931da177e4SLinus Torvalds mode &= ~current->fs->umask; 19941da177e4SLinus Torvalds error = vfs_mkdir(nd.dentry->d_inode, dentry, mode); 19951da177e4SLinus Torvalds dput(dentry); 19966902d925SDave Hansen out_unlock: 19971b1dcc1bSJes Sorensen mutex_unlock(&nd.dentry->d_inode->i_mutex); 19981da177e4SLinus Torvalds path_release(&nd); 19991da177e4SLinus Torvalds out: 20001da177e4SLinus Torvalds putname(tmp); 20016902d925SDave Hansen out_err: 20021da177e4SLinus Torvalds return error; 20031da177e4SLinus Torvalds } 20041da177e4SLinus Torvalds 20055590ff0dSUlrich Drepper asmlinkage long sys_mkdir(const char __user *pathname, int mode) 20065590ff0dSUlrich Drepper { 20075590ff0dSUlrich Drepper return sys_mkdirat(AT_FDCWD, pathname, mode); 20085590ff0dSUlrich Drepper } 20095590ff0dSUlrich Drepper 20101da177e4SLinus Torvalds /* 20111da177e4SLinus Torvalds * We try to drop the dentry early: we should have 20121da177e4SLinus Torvalds * a usage count of 2 if we're the only user of this 20131da177e4SLinus Torvalds * dentry, and if that is true (possibly after pruning 20141da177e4SLinus Torvalds * the dcache), then we drop the dentry now. 20151da177e4SLinus Torvalds * 20161da177e4SLinus Torvalds * A low-level filesystem can, if it choses, legally 20171da177e4SLinus Torvalds * do a 20181da177e4SLinus Torvalds * 20191da177e4SLinus Torvalds * if (!d_unhashed(dentry)) 20201da177e4SLinus Torvalds * return -EBUSY; 20211da177e4SLinus Torvalds * 20221da177e4SLinus Torvalds * if it cannot handle the case of removing a directory 20231da177e4SLinus Torvalds * that is still in use by something else.. 20241da177e4SLinus Torvalds */ 20251da177e4SLinus Torvalds void dentry_unhash(struct dentry *dentry) 20261da177e4SLinus Torvalds { 20271da177e4SLinus Torvalds dget(dentry); 20281da177e4SLinus Torvalds shrink_dcache_parent(dentry); 20291da177e4SLinus Torvalds spin_lock(&dcache_lock); 20301da177e4SLinus Torvalds spin_lock(&dentry->d_lock); 20311da177e4SLinus Torvalds if (atomic_read(&dentry->d_count) == 2) 20321da177e4SLinus Torvalds __d_drop(dentry); 20331da177e4SLinus Torvalds spin_unlock(&dentry->d_lock); 20341da177e4SLinus Torvalds spin_unlock(&dcache_lock); 20351da177e4SLinus Torvalds } 20361da177e4SLinus Torvalds 20371da177e4SLinus Torvalds int vfs_rmdir(struct inode *dir, struct dentry *dentry) 20381da177e4SLinus Torvalds { 20391da177e4SLinus Torvalds int error = may_delete(dir, dentry, 1); 20401da177e4SLinus Torvalds 20411da177e4SLinus Torvalds if (error) 20421da177e4SLinus Torvalds return error; 20431da177e4SLinus Torvalds 20441da177e4SLinus Torvalds if (!dir->i_op || !dir->i_op->rmdir) 20451da177e4SLinus Torvalds return -EPERM; 20461da177e4SLinus Torvalds 20471da177e4SLinus Torvalds DQUOT_INIT(dir); 20481da177e4SLinus Torvalds 20491b1dcc1bSJes Sorensen mutex_lock(&dentry->d_inode->i_mutex); 20501da177e4SLinus Torvalds dentry_unhash(dentry); 20511da177e4SLinus Torvalds if (d_mountpoint(dentry)) 20521da177e4SLinus Torvalds error = -EBUSY; 20531da177e4SLinus Torvalds else { 20541da177e4SLinus Torvalds error = security_inode_rmdir(dir, dentry); 20551da177e4SLinus Torvalds if (!error) { 20561da177e4SLinus Torvalds error = dir->i_op->rmdir(dir, dentry); 20571da177e4SLinus Torvalds if (!error) 20581da177e4SLinus Torvalds dentry->d_inode->i_flags |= S_DEAD; 20591da177e4SLinus Torvalds } 20601da177e4SLinus Torvalds } 20611b1dcc1bSJes Sorensen mutex_unlock(&dentry->d_inode->i_mutex); 20621da177e4SLinus Torvalds if (!error) { 20631da177e4SLinus Torvalds d_delete(dentry); 20641da177e4SLinus Torvalds } 20651da177e4SLinus Torvalds dput(dentry); 20661da177e4SLinus Torvalds 20671da177e4SLinus Torvalds return error; 20681da177e4SLinus Torvalds } 20691da177e4SLinus Torvalds 20705590ff0dSUlrich Drepper static long do_rmdir(int dfd, const char __user *pathname) 20711da177e4SLinus Torvalds { 20721da177e4SLinus Torvalds int error = 0; 20731da177e4SLinus Torvalds char * name; 20741da177e4SLinus Torvalds struct dentry *dentry; 20751da177e4SLinus Torvalds struct nameidata nd; 20761da177e4SLinus Torvalds 20771da177e4SLinus Torvalds name = getname(pathname); 20781da177e4SLinus Torvalds if(IS_ERR(name)) 20791da177e4SLinus Torvalds return PTR_ERR(name); 20801da177e4SLinus Torvalds 20815590ff0dSUlrich Drepper error = do_path_lookup(dfd, name, LOOKUP_PARENT, &nd); 20821da177e4SLinus Torvalds if (error) 20831da177e4SLinus Torvalds goto exit; 20841da177e4SLinus Torvalds 20851da177e4SLinus Torvalds switch(nd.last_type) { 20861da177e4SLinus Torvalds case LAST_DOTDOT: 20871da177e4SLinus Torvalds error = -ENOTEMPTY; 20881da177e4SLinus Torvalds goto exit1; 20891da177e4SLinus Torvalds case LAST_DOT: 20901da177e4SLinus Torvalds error = -EINVAL; 20911da177e4SLinus Torvalds goto exit1; 20921da177e4SLinus Torvalds case LAST_ROOT: 20931da177e4SLinus Torvalds error = -EBUSY; 20941da177e4SLinus Torvalds goto exit1; 20951da177e4SLinus Torvalds } 2096f2eace23SIngo Molnar mutex_lock_nested(&nd.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 209749705b77SChristoph Hellwig dentry = lookup_hash(&nd); 20981da177e4SLinus Torvalds error = PTR_ERR(dentry); 20996902d925SDave Hansen if (IS_ERR(dentry)) 21006902d925SDave Hansen goto exit2; 21011da177e4SLinus Torvalds error = vfs_rmdir(nd.dentry->d_inode, dentry); 21021da177e4SLinus Torvalds dput(dentry); 21036902d925SDave Hansen exit2: 21041b1dcc1bSJes Sorensen mutex_unlock(&nd.dentry->d_inode->i_mutex); 21051da177e4SLinus Torvalds exit1: 21061da177e4SLinus Torvalds path_release(&nd); 21071da177e4SLinus Torvalds exit: 21081da177e4SLinus Torvalds putname(name); 21091da177e4SLinus Torvalds return error; 21101da177e4SLinus Torvalds } 21111da177e4SLinus Torvalds 21125590ff0dSUlrich Drepper asmlinkage long sys_rmdir(const char __user *pathname) 21135590ff0dSUlrich Drepper { 21145590ff0dSUlrich Drepper return do_rmdir(AT_FDCWD, pathname); 21155590ff0dSUlrich Drepper } 21165590ff0dSUlrich Drepper 21171da177e4SLinus Torvalds int vfs_unlink(struct inode *dir, struct dentry *dentry) 21181da177e4SLinus Torvalds { 21191da177e4SLinus Torvalds int error = may_delete(dir, dentry, 0); 21201da177e4SLinus Torvalds 21211da177e4SLinus Torvalds if (error) 21221da177e4SLinus Torvalds return error; 21231da177e4SLinus Torvalds 21241da177e4SLinus Torvalds if (!dir->i_op || !dir->i_op->unlink) 21251da177e4SLinus Torvalds return -EPERM; 21261da177e4SLinus Torvalds 21271da177e4SLinus Torvalds DQUOT_INIT(dir); 21281da177e4SLinus Torvalds 21291b1dcc1bSJes Sorensen mutex_lock(&dentry->d_inode->i_mutex); 21301da177e4SLinus Torvalds if (d_mountpoint(dentry)) 21311da177e4SLinus Torvalds error = -EBUSY; 21321da177e4SLinus Torvalds else { 21331da177e4SLinus Torvalds error = security_inode_unlink(dir, dentry); 21341da177e4SLinus Torvalds if (!error) 21351da177e4SLinus Torvalds error = dir->i_op->unlink(dir, dentry); 21361da177e4SLinus Torvalds } 21371b1dcc1bSJes Sorensen mutex_unlock(&dentry->d_inode->i_mutex); 21381da177e4SLinus Torvalds 21391da177e4SLinus Torvalds /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 21401da177e4SLinus Torvalds if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { 21411da177e4SLinus Torvalds d_delete(dentry); 21421da177e4SLinus Torvalds } 21430eeca283SRobert Love 21441da177e4SLinus Torvalds return error; 21451da177e4SLinus Torvalds } 21461da177e4SLinus Torvalds 21471da177e4SLinus Torvalds /* 21481da177e4SLinus Torvalds * Make sure that the actual truncation of the file will occur outside its 21491b1dcc1bSJes Sorensen * directory's i_mutex. Truncate can take a long time if there is a lot of 21501da177e4SLinus Torvalds * writeout happening, and we don't want to prevent access to the directory 21511da177e4SLinus Torvalds * while waiting on the I/O. 21521da177e4SLinus Torvalds */ 21535590ff0dSUlrich Drepper static long do_unlinkat(int dfd, const char __user *pathname) 21541da177e4SLinus Torvalds { 21551da177e4SLinus Torvalds int error = 0; 21561da177e4SLinus Torvalds char * name; 21571da177e4SLinus Torvalds struct dentry *dentry; 21581da177e4SLinus Torvalds struct nameidata nd; 21591da177e4SLinus Torvalds struct inode *inode = NULL; 21601da177e4SLinus Torvalds 21611da177e4SLinus Torvalds name = getname(pathname); 21621da177e4SLinus Torvalds if(IS_ERR(name)) 21631da177e4SLinus Torvalds return PTR_ERR(name); 21641da177e4SLinus Torvalds 21655590ff0dSUlrich Drepper error = do_path_lookup(dfd, name, LOOKUP_PARENT, &nd); 21661da177e4SLinus Torvalds if (error) 21671da177e4SLinus Torvalds goto exit; 21681da177e4SLinus Torvalds error = -EISDIR; 21691da177e4SLinus Torvalds if (nd.last_type != LAST_NORM) 21701da177e4SLinus Torvalds goto exit1; 2171f2eace23SIngo Molnar mutex_lock_nested(&nd.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 217249705b77SChristoph Hellwig dentry = lookup_hash(&nd); 21731da177e4SLinus Torvalds error = PTR_ERR(dentry); 21741da177e4SLinus Torvalds if (!IS_ERR(dentry)) { 21751da177e4SLinus Torvalds /* Why not before? Because we want correct error value */ 21761da177e4SLinus Torvalds if (nd.last.name[nd.last.len]) 21771da177e4SLinus Torvalds goto slashes; 21781da177e4SLinus Torvalds inode = dentry->d_inode; 21791da177e4SLinus Torvalds if (inode) 21801da177e4SLinus Torvalds atomic_inc(&inode->i_count); 21811da177e4SLinus Torvalds error = vfs_unlink(nd.dentry->d_inode, dentry); 21821da177e4SLinus Torvalds exit2: 21831da177e4SLinus Torvalds dput(dentry); 21841da177e4SLinus Torvalds } 21851b1dcc1bSJes Sorensen mutex_unlock(&nd.dentry->d_inode->i_mutex); 21861da177e4SLinus Torvalds if (inode) 21871da177e4SLinus Torvalds iput(inode); /* truncate the inode here */ 21881da177e4SLinus Torvalds exit1: 21891da177e4SLinus Torvalds path_release(&nd); 21901da177e4SLinus Torvalds exit: 21911da177e4SLinus Torvalds putname(name); 21921da177e4SLinus Torvalds return error; 21931da177e4SLinus Torvalds 21941da177e4SLinus Torvalds slashes: 21951da177e4SLinus Torvalds error = !dentry->d_inode ? -ENOENT : 21961da177e4SLinus Torvalds S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR; 21971da177e4SLinus Torvalds goto exit2; 21981da177e4SLinus Torvalds } 21991da177e4SLinus Torvalds 22005590ff0dSUlrich Drepper asmlinkage long sys_unlinkat(int dfd, const char __user *pathname, int flag) 22015590ff0dSUlrich Drepper { 22025590ff0dSUlrich Drepper if ((flag & ~AT_REMOVEDIR) != 0) 22035590ff0dSUlrich Drepper return -EINVAL; 22045590ff0dSUlrich Drepper 22055590ff0dSUlrich Drepper if (flag & AT_REMOVEDIR) 22065590ff0dSUlrich Drepper return do_rmdir(dfd, pathname); 22075590ff0dSUlrich Drepper 22085590ff0dSUlrich Drepper return do_unlinkat(dfd, pathname); 22095590ff0dSUlrich Drepper } 22105590ff0dSUlrich Drepper 22115590ff0dSUlrich Drepper asmlinkage long sys_unlink(const char __user *pathname) 22125590ff0dSUlrich Drepper { 22135590ff0dSUlrich Drepper return do_unlinkat(AT_FDCWD, pathname); 22145590ff0dSUlrich Drepper } 22155590ff0dSUlrich Drepper 22161da177e4SLinus Torvalds int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname, int mode) 22171da177e4SLinus Torvalds { 22181da177e4SLinus Torvalds int error = may_create(dir, dentry, NULL); 22191da177e4SLinus Torvalds 22201da177e4SLinus Torvalds if (error) 22211da177e4SLinus Torvalds return error; 22221da177e4SLinus Torvalds 22231da177e4SLinus Torvalds if (!dir->i_op || !dir->i_op->symlink) 22241da177e4SLinus Torvalds return -EPERM; 22251da177e4SLinus Torvalds 22261da177e4SLinus Torvalds error = security_inode_symlink(dir, dentry, oldname); 22271da177e4SLinus Torvalds if (error) 22281da177e4SLinus Torvalds return error; 22291da177e4SLinus Torvalds 22301da177e4SLinus Torvalds DQUOT_INIT(dir); 22311da177e4SLinus Torvalds error = dir->i_op->symlink(dir, dentry, oldname); 2232a74574aaSStephen Smalley if (!error) 2233f38aa942SAmy Griffis fsnotify_create(dir, dentry); 22341da177e4SLinus Torvalds return error; 22351da177e4SLinus Torvalds } 22361da177e4SLinus Torvalds 22375590ff0dSUlrich Drepper asmlinkage long sys_symlinkat(const char __user *oldname, 22385590ff0dSUlrich Drepper int newdfd, const char __user *newname) 22391da177e4SLinus Torvalds { 22401da177e4SLinus Torvalds int error = 0; 22411da177e4SLinus Torvalds char * from; 22421da177e4SLinus Torvalds char * to; 22436902d925SDave Hansen struct dentry *dentry; 22446902d925SDave Hansen struct nameidata nd; 22451da177e4SLinus Torvalds 22461da177e4SLinus Torvalds from = getname(oldname); 22471da177e4SLinus Torvalds if(IS_ERR(from)) 22481da177e4SLinus Torvalds return PTR_ERR(from); 22491da177e4SLinus Torvalds to = getname(newname); 22501da177e4SLinus Torvalds error = PTR_ERR(to); 22516902d925SDave Hansen if (IS_ERR(to)) 22526902d925SDave Hansen goto out_putname; 22531da177e4SLinus Torvalds 22545590ff0dSUlrich Drepper error = do_path_lookup(newdfd, to, LOOKUP_PARENT, &nd); 22551da177e4SLinus Torvalds if (error) 22561da177e4SLinus Torvalds goto out; 22571da177e4SLinus Torvalds dentry = lookup_create(&nd, 0); 22581da177e4SLinus Torvalds error = PTR_ERR(dentry); 22596902d925SDave Hansen if (IS_ERR(dentry)) 22606902d925SDave Hansen goto out_unlock; 22616902d925SDave Hansen 22621da177e4SLinus Torvalds error = vfs_symlink(nd.dentry->d_inode, dentry, from, S_IALLUGO); 22631da177e4SLinus Torvalds dput(dentry); 22646902d925SDave Hansen out_unlock: 22651b1dcc1bSJes Sorensen mutex_unlock(&nd.dentry->d_inode->i_mutex); 22661da177e4SLinus Torvalds path_release(&nd); 22671da177e4SLinus Torvalds out: 22681da177e4SLinus Torvalds putname(to); 22696902d925SDave Hansen out_putname: 22701da177e4SLinus Torvalds putname(from); 22711da177e4SLinus Torvalds return error; 22721da177e4SLinus Torvalds } 22731da177e4SLinus Torvalds 22745590ff0dSUlrich Drepper asmlinkage long sys_symlink(const char __user *oldname, const char __user *newname) 22755590ff0dSUlrich Drepper { 22765590ff0dSUlrich Drepper return sys_symlinkat(oldname, AT_FDCWD, newname); 22775590ff0dSUlrich Drepper } 22785590ff0dSUlrich Drepper 22791da177e4SLinus Torvalds int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry) 22801da177e4SLinus Torvalds { 22811da177e4SLinus Torvalds struct inode *inode = old_dentry->d_inode; 22821da177e4SLinus Torvalds int error; 22831da177e4SLinus Torvalds 22841da177e4SLinus Torvalds if (!inode) 22851da177e4SLinus Torvalds return -ENOENT; 22861da177e4SLinus Torvalds 22871da177e4SLinus Torvalds error = may_create(dir, new_dentry, NULL); 22881da177e4SLinus Torvalds if (error) 22891da177e4SLinus Torvalds return error; 22901da177e4SLinus Torvalds 22911da177e4SLinus Torvalds if (dir->i_sb != inode->i_sb) 22921da177e4SLinus Torvalds return -EXDEV; 22931da177e4SLinus Torvalds 22941da177e4SLinus Torvalds /* 22951da177e4SLinus Torvalds * A link to an append-only or immutable file cannot be created. 22961da177e4SLinus Torvalds */ 22971da177e4SLinus Torvalds if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 22981da177e4SLinus Torvalds return -EPERM; 22991da177e4SLinus Torvalds if (!dir->i_op || !dir->i_op->link) 23001da177e4SLinus Torvalds return -EPERM; 23011da177e4SLinus Torvalds if (S_ISDIR(old_dentry->d_inode->i_mode)) 23021da177e4SLinus Torvalds return -EPERM; 23031da177e4SLinus Torvalds 23041da177e4SLinus Torvalds error = security_inode_link(old_dentry, dir, new_dentry); 23051da177e4SLinus Torvalds if (error) 23061da177e4SLinus Torvalds return error; 23071da177e4SLinus Torvalds 23081b1dcc1bSJes Sorensen mutex_lock(&old_dentry->d_inode->i_mutex); 23091da177e4SLinus Torvalds DQUOT_INIT(dir); 23101da177e4SLinus Torvalds error = dir->i_op->link(old_dentry, dir, new_dentry); 23111b1dcc1bSJes Sorensen mutex_unlock(&old_dentry->d_inode->i_mutex); 2312e31e14ecSStephen Smalley if (!error) 2313f38aa942SAmy Griffis fsnotify_create(dir, new_dentry); 23141da177e4SLinus Torvalds return error; 23151da177e4SLinus Torvalds } 23161da177e4SLinus Torvalds 23171da177e4SLinus Torvalds /* 23181da177e4SLinus Torvalds * Hardlinks are often used in delicate situations. We avoid 23191da177e4SLinus Torvalds * security-related surprises by not following symlinks on the 23201da177e4SLinus Torvalds * newname. --KAB 23211da177e4SLinus Torvalds * 23221da177e4SLinus Torvalds * We don't follow them on the oldname either to be compatible 23231da177e4SLinus Torvalds * with linux 2.0, and to avoid hard-linking to directories 23241da177e4SLinus Torvalds * and other special files. --ADM 23251da177e4SLinus Torvalds */ 23265590ff0dSUlrich Drepper asmlinkage long sys_linkat(int olddfd, const char __user *oldname, 2327c04030e1SUlrich Drepper int newdfd, const char __user *newname, 2328c04030e1SUlrich Drepper int flags) 23291da177e4SLinus Torvalds { 23301da177e4SLinus Torvalds struct dentry *new_dentry; 23311da177e4SLinus Torvalds struct nameidata nd, old_nd; 23321da177e4SLinus Torvalds int error; 23331da177e4SLinus Torvalds char * to; 23341da177e4SLinus Torvalds 233545c9b11aSUlrich Drepper if ((flags & ~AT_SYMLINK_FOLLOW) != 0) 2336c04030e1SUlrich Drepper return -EINVAL; 2337c04030e1SUlrich Drepper 23381da177e4SLinus Torvalds to = getname(newname); 23391da177e4SLinus Torvalds if (IS_ERR(to)) 23401da177e4SLinus Torvalds return PTR_ERR(to); 23411da177e4SLinus Torvalds 234245c9b11aSUlrich Drepper error = __user_walk_fd(olddfd, oldname, 234345c9b11aSUlrich Drepper flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0, 234445c9b11aSUlrich Drepper &old_nd); 23451da177e4SLinus Torvalds if (error) 23461da177e4SLinus Torvalds goto exit; 23475590ff0dSUlrich Drepper error = do_path_lookup(newdfd, to, LOOKUP_PARENT, &nd); 23481da177e4SLinus Torvalds if (error) 23491da177e4SLinus Torvalds goto out; 23501da177e4SLinus Torvalds error = -EXDEV; 23511da177e4SLinus Torvalds if (old_nd.mnt != nd.mnt) 23521da177e4SLinus Torvalds goto out_release; 23531da177e4SLinus Torvalds new_dentry = lookup_create(&nd, 0); 23541da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 23556902d925SDave Hansen if (IS_ERR(new_dentry)) 23566902d925SDave Hansen goto out_unlock; 23571da177e4SLinus Torvalds error = vfs_link(old_nd.dentry, nd.dentry->d_inode, new_dentry); 23581da177e4SLinus Torvalds dput(new_dentry); 23596902d925SDave Hansen out_unlock: 23601b1dcc1bSJes Sorensen mutex_unlock(&nd.dentry->d_inode->i_mutex); 23611da177e4SLinus Torvalds out_release: 23621da177e4SLinus Torvalds path_release(&nd); 23631da177e4SLinus Torvalds out: 23641da177e4SLinus Torvalds path_release(&old_nd); 23651da177e4SLinus Torvalds exit: 23661da177e4SLinus Torvalds putname(to); 23671da177e4SLinus Torvalds 23681da177e4SLinus Torvalds return error; 23691da177e4SLinus Torvalds } 23701da177e4SLinus Torvalds 23715590ff0dSUlrich Drepper asmlinkage long sys_link(const char __user *oldname, const char __user *newname) 23725590ff0dSUlrich Drepper { 2373c04030e1SUlrich Drepper return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 23745590ff0dSUlrich Drepper } 23755590ff0dSUlrich Drepper 23761da177e4SLinus Torvalds /* 23771da177e4SLinus Torvalds * The worst of all namespace operations - renaming directory. "Perverted" 23781da177e4SLinus Torvalds * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 23791da177e4SLinus Torvalds * Problems: 23801da177e4SLinus Torvalds * a) we can get into loop creation. Check is done in is_subdir(). 23811da177e4SLinus Torvalds * b) race potential - two innocent renames can create a loop together. 23821da177e4SLinus Torvalds * That's where 4.4 screws up. Current fix: serialization on 2383a11f3a05SArjan van de Ven * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 23841da177e4SLinus Torvalds * story. 23851da177e4SLinus Torvalds * c) we have to lock _three_ objects - parents and victim (if it exists). 23861b1dcc1bSJes Sorensen * And that - after we got ->i_mutex on parents (until then we don't know 23871da177e4SLinus Torvalds * whether the target exists). Solution: try to be smart with locking 23881da177e4SLinus Torvalds * order for inodes. We rely on the fact that tree topology may change 2389a11f3a05SArjan van de Ven * only under ->s_vfs_rename_mutex _and_ that parent of the object we 23901da177e4SLinus Torvalds * move will be locked. Thus we can rank directories by the tree 23911da177e4SLinus Torvalds * (ancestors first) and rank all non-directories after them. 23921da177e4SLinus Torvalds * That works since everybody except rename does "lock parent, lookup, 2393a11f3a05SArjan van de Ven * lock child" and rename is under ->s_vfs_rename_mutex. 23941da177e4SLinus Torvalds * HOWEVER, it relies on the assumption that any object with ->lookup() 23951da177e4SLinus Torvalds * has no more than 1 dentry. If "hybrid" objects will ever appear, 23961da177e4SLinus Torvalds * we'd better make sure that there's no link(2) for them. 23971da177e4SLinus Torvalds * d) some filesystems don't support opened-but-unlinked directories, 23981da177e4SLinus Torvalds * either because of layout or because they are not ready to deal with 23991da177e4SLinus Torvalds * all cases correctly. The latter will be fixed (taking this sort of 24001da177e4SLinus Torvalds * stuff into VFS), but the former is not going away. Solution: the same 24011da177e4SLinus Torvalds * trick as in rmdir(). 24021da177e4SLinus Torvalds * e) conversion from fhandle to dentry may come in the wrong moment - when 24031b1dcc1bSJes Sorensen * we are removing the target. Solution: we will have to grab ->i_mutex 24041da177e4SLinus Torvalds * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 24051b1dcc1bSJes Sorensen * ->i_mutex on parents, which works but leads to some truely excessive 24061da177e4SLinus Torvalds * locking]. 24071da177e4SLinus Torvalds */ 240875c96f85SAdrian Bunk static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry, 24091da177e4SLinus Torvalds struct inode *new_dir, struct dentry *new_dentry) 24101da177e4SLinus Torvalds { 24111da177e4SLinus Torvalds int error = 0; 24121da177e4SLinus Torvalds struct inode *target; 24131da177e4SLinus Torvalds 24141da177e4SLinus Torvalds /* 24151da177e4SLinus Torvalds * If we are going to change the parent - check write permissions, 24161da177e4SLinus Torvalds * we'll need to flip '..'. 24171da177e4SLinus Torvalds */ 24181da177e4SLinus Torvalds if (new_dir != old_dir) { 24191da177e4SLinus Torvalds error = permission(old_dentry->d_inode, MAY_WRITE, NULL); 24201da177e4SLinus Torvalds if (error) 24211da177e4SLinus Torvalds return error; 24221da177e4SLinus Torvalds } 24231da177e4SLinus Torvalds 24241da177e4SLinus Torvalds error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); 24251da177e4SLinus Torvalds if (error) 24261da177e4SLinus Torvalds return error; 24271da177e4SLinus Torvalds 24281da177e4SLinus Torvalds target = new_dentry->d_inode; 24291da177e4SLinus Torvalds if (target) { 24301b1dcc1bSJes Sorensen mutex_lock(&target->i_mutex); 24311da177e4SLinus Torvalds dentry_unhash(new_dentry); 24321da177e4SLinus Torvalds } 24331da177e4SLinus Torvalds if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry)) 24341da177e4SLinus Torvalds error = -EBUSY; 24351da177e4SLinus Torvalds else 24361da177e4SLinus Torvalds error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); 24371da177e4SLinus Torvalds if (target) { 24381da177e4SLinus Torvalds if (!error) 24391da177e4SLinus Torvalds target->i_flags |= S_DEAD; 24401b1dcc1bSJes Sorensen mutex_unlock(&target->i_mutex); 24411da177e4SLinus Torvalds if (d_unhashed(new_dentry)) 24421da177e4SLinus Torvalds d_rehash(new_dentry); 24431da177e4SLinus Torvalds dput(new_dentry); 24441da177e4SLinus Torvalds } 2445e31e14ecSStephen Smalley if (!error) 2446349457ccSMark Fasheh if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) 24471da177e4SLinus Torvalds d_move(old_dentry,new_dentry); 24481da177e4SLinus Torvalds return error; 24491da177e4SLinus Torvalds } 24501da177e4SLinus Torvalds 245175c96f85SAdrian Bunk static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry, 24521da177e4SLinus Torvalds struct inode *new_dir, struct dentry *new_dentry) 24531da177e4SLinus Torvalds { 24541da177e4SLinus Torvalds struct inode *target; 24551da177e4SLinus Torvalds int error; 24561da177e4SLinus Torvalds 24571da177e4SLinus Torvalds error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); 24581da177e4SLinus Torvalds if (error) 24591da177e4SLinus Torvalds return error; 24601da177e4SLinus Torvalds 24611da177e4SLinus Torvalds dget(new_dentry); 24621da177e4SLinus Torvalds target = new_dentry->d_inode; 24631da177e4SLinus Torvalds if (target) 24641b1dcc1bSJes Sorensen mutex_lock(&target->i_mutex); 24651da177e4SLinus Torvalds if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry)) 24661da177e4SLinus Torvalds error = -EBUSY; 24671da177e4SLinus Torvalds else 24681da177e4SLinus Torvalds error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); 24691da177e4SLinus Torvalds if (!error) { 2470349457ccSMark Fasheh if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) 24711da177e4SLinus Torvalds d_move(old_dentry, new_dentry); 24721da177e4SLinus Torvalds } 24731da177e4SLinus Torvalds if (target) 24741b1dcc1bSJes Sorensen mutex_unlock(&target->i_mutex); 24751da177e4SLinus Torvalds dput(new_dentry); 24761da177e4SLinus Torvalds return error; 24771da177e4SLinus Torvalds } 24781da177e4SLinus Torvalds 24791da177e4SLinus Torvalds int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, 24801da177e4SLinus Torvalds struct inode *new_dir, struct dentry *new_dentry) 24811da177e4SLinus Torvalds { 24821da177e4SLinus Torvalds int error; 24831da177e4SLinus Torvalds int is_dir = S_ISDIR(old_dentry->d_inode->i_mode); 24840eeca283SRobert Love const char *old_name; 24851da177e4SLinus Torvalds 24861da177e4SLinus Torvalds if (old_dentry->d_inode == new_dentry->d_inode) 24871da177e4SLinus Torvalds return 0; 24881da177e4SLinus Torvalds 24891da177e4SLinus Torvalds error = may_delete(old_dir, old_dentry, is_dir); 24901da177e4SLinus Torvalds if (error) 24911da177e4SLinus Torvalds return error; 24921da177e4SLinus Torvalds 24931da177e4SLinus Torvalds if (!new_dentry->d_inode) 24941da177e4SLinus Torvalds error = may_create(new_dir, new_dentry, NULL); 24951da177e4SLinus Torvalds else 24961da177e4SLinus Torvalds error = may_delete(new_dir, new_dentry, is_dir); 24971da177e4SLinus Torvalds if (error) 24981da177e4SLinus Torvalds return error; 24991da177e4SLinus Torvalds 25001da177e4SLinus Torvalds if (!old_dir->i_op || !old_dir->i_op->rename) 25011da177e4SLinus Torvalds return -EPERM; 25021da177e4SLinus Torvalds 25031da177e4SLinus Torvalds DQUOT_INIT(old_dir); 25041da177e4SLinus Torvalds DQUOT_INIT(new_dir); 25051da177e4SLinus Torvalds 25060eeca283SRobert Love old_name = fsnotify_oldname_init(old_dentry->d_name.name); 25070eeca283SRobert Love 25081da177e4SLinus Torvalds if (is_dir) 25091da177e4SLinus Torvalds error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry); 25101da177e4SLinus Torvalds else 25111da177e4SLinus Torvalds error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry); 25121da177e4SLinus Torvalds if (!error) { 25130eeca283SRobert Love const char *new_name = old_dentry->d_name.name; 251489204c40SJohn McCutchan fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir, 251589204c40SJohn McCutchan new_dentry->d_inode, old_dentry->d_inode); 25161da177e4SLinus Torvalds } 25170eeca283SRobert Love fsnotify_oldname_free(old_name); 25180eeca283SRobert Love 25191da177e4SLinus Torvalds return error; 25201da177e4SLinus Torvalds } 25211da177e4SLinus Torvalds 25225590ff0dSUlrich Drepper static int do_rename(int olddfd, const char *oldname, 25235590ff0dSUlrich Drepper int newdfd, const char *newname) 25241da177e4SLinus Torvalds { 25251da177e4SLinus Torvalds int error = 0; 25261da177e4SLinus Torvalds struct dentry * old_dir, * new_dir; 25271da177e4SLinus Torvalds struct dentry * old_dentry, *new_dentry; 25281da177e4SLinus Torvalds struct dentry * trap; 25291da177e4SLinus Torvalds struct nameidata oldnd, newnd; 25301da177e4SLinus Torvalds 25315590ff0dSUlrich Drepper error = do_path_lookup(olddfd, oldname, LOOKUP_PARENT, &oldnd); 25321da177e4SLinus Torvalds if (error) 25331da177e4SLinus Torvalds goto exit; 25341da177e4SLinus Torvalds 25355590ff0dSUlrich Drepper error = do_path_lookup(newdfd, newname, LOOKUP_PARENT, &newnd); 25361da177e4SLinus Torvalds if (error) 25371da177e4SLinus Torvalds goto exit1; 25381da177e4SLinus Torvalds 25391da177e4SLinus Torvalds error = -EXDEV; 25401da177e4SLinus Torvalds if (oldnd.mnt != newnd.mnt) 25411da177e4SLinus Torvalds goto exit2; 25421da177e4SLinus Torvalds 25431da177e4SLinus Torvalds old_dir = oldnd.dentry; 25441da177e4SLinus Torvalds error = -EBUSY; 25451da177e4SLinus Torvalds if (oldnd.last_type != LAST_NORM) 25461da177e4SLinus Torvalds goto exit2; 25471da177e4SLinus Torvalds 25481da177e4SLinus Torvalds new_dir = newnd.dentry; 25491da177e4SLinus Torvalds if (newnd.last_type != LAST_NORM) 25501da177e4SLinus Torvalds goto exit2; 25511da177e4SLinus Torvalds 25521da177e4SLinus Torvalds trap = lock_rename(new_dir, old_dir); 25531da177e4SLinus Torvalds 255449705b77SChristoph Hellwig old_dentry = lookup_hash(&oldnd); 25551da177e4SLinus Torvalds error = PTR_ERR(old_dentry); 25561da177e4SLinus Torvalds if (IS_ERR(old_dentry)) 25571da177e4SLinus Torvalds goto exit3; 25581da177e4SLinus Torvalds /* source must exist */ 25591da177e4SLinus Torvalds error = -ENOENT; 25601da177e4SLinus Torvalds if (!old_dentry->d_inode) 25611da177e4SLinus Torvalds goto exit4; 25621da177e4SLinus Torvalds /* unless the source is a directory trailing slashes give -ENOTDIR */ 25631da177e4SLinus Torvalds if (!S_ISDIR(old_dentry->d_inode->i_mode)) { 25641da177e4SLinus Torvalds error = -ENOTDIR; 25651da177e4SLinus Torvalds if (oldnd.last.name[oldnd.last.len]) 25661da177e4SLinus Torvalds goto exit4; 25671da177e4SLinus Torvalds if (newnd.last.name[newnd.last.len]) 25681da177e4SLinus Torvalds goto exit4; 25691da177e4SLinus Torvalds } 25701da177e4SLinus Torvalds /* source should not be ancestor of target */ 25711da177e4SLinus Torvalds error = -EINVAL; 25721da177e4SLinus Torvalds if (old_dentry == trap) 25731da177e4SLinus Torvalds goto exit4; 257449705b77SChristoph Hellwig new_dentry = lookup_hash(&newnd); 25751da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 25761da177e4SLinus Torvalds if (IS_ERR(new_dentry)) 25771da177e4SLinus Torvalds goto exit4; 25781da177e4SLinus Torvalds /* target should not be an ancestor of source */ 25791da177e4SLinus Torvalds error = -ENOTEMPTY; 25801da177e4SLinus Torvalds if (new_dentry == trap) 25811da177e4SLinus Torvalds goto exit5; 25821da177e4SLinus Torvalds 25831da177e4SLinus Torvalds error = vfs_rename(old_dir->d_inode, old_dentry, 25841da177e4SLinus Torvalds new_dir->d_inode, new_dentry); 25851da177e4SLinus Torvalds exit5: 25861da177e4SLinus Torvalds dput(new_dentry); 25871da177e4SLinus Torvalds exit4: 25881da177e4SLinus Torvalds dput(old_dentry); 25891da177e4SLinus Torvalds exit3: 25901da177e4SLinus Torvalds unlock_rename(new_dir, old_dir); 25911da177e4SLinus Torvalds exit2: 25921da177e4SLinus Torvalds path_release(&newnd); 25931da177e4SLinus Torvalds exit1: 25941da177e4SLinus Torvalds path_release(&oldnd); 25951da177e4SLinus Torvalds exit: 25961da177e4SLinus Torvalds return error; 25971da177e4SLinus Torvalds } 25981da177e4SLinus Torvalds 25995590ff0dSUlrich Drepper asmlinkage long sys_renameat(int olddfd, const char __user *oldname, 26005590ff0dSUlrich Drepper int newdfd, const char __user *newname) 26011da177e4SLinus Torvalds { 26021da177e4SLinus Torvalds int error; 26031da177e4SLinus Torvalds char * from; 26041da177e4SLinus Torvalds char * to; 26051da177e4SLinus Torvalds 26061da177e4SLinus Torvalds from = getname(oldname); 26071da177e4SLinus Torvalds if(IS_ERR(from)) 26081da177e4SLinus Torvalds return PTR_ERR(from); 26091da177e4SLinus Torvalds to = getname(newname); 26101da177e4SLinus Torvalds error = PTR_ERR(to); 26111da177e4SLinus Torvalds if (!IS_ERR(to)) { 26125590ff0dSUlrich Drepper error = do_rename(olddfd, from, newdfd, to); 26131da177e4SLinus Torvalds putname(to); 26141da177e4SLinus Torvalds } 26151da177e4SLinus Torvalds putname(from); 26161da177e4SLinus Torvalds return error; 26171da177e4SLinus Torvalds } 26181da177e4SLinus Torvalds 26195590ff0dSUlrich Drepper asmlinkage long sys_rename(const char __user *oldname, const char __user *newname) 26205590ff0dSUlrich Drepper { 26215590ff0dSUlrich Drepper return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname); 26225590ff0dSUlrich Drepper } 26235590ff0dSUlrich Drepper 26241da177e4SLinus Torvalds int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link) 26251da177e4SLinus Torvalds { 26261da177e4SLinus Torvalds int len; 26271da177e4SLinus Torvalds 26281da177e4SLinus Torvalds len = PTR_ERR(link); 26291da177e4SLinus Torvalds if (IS_ERR(link)) 26301da177e4SLinus Torvalds goto out; 26311da177e4SLinus Torvalds 26321da177e4SLinus Torvalds len = strlen(link); 26331da177e4SLinus Torvalds if (len > (unsigned) buflen) 26341da177e4SLinus Torvalds len = buflen; 26351da177e4SLinus Torvalds if (copy_to_user(buffer, link, len)) 26361da177e4SLinus Torvalds len = -EFAULT; 26371da177e4SLinus Torvalds out: 26381da177e4SLinus Torvalds return len; 26391da177e4SLinus Torvalds } 26401da177e4SLinus Torvalds 26411da177e4SLinus Torvalds /* 26421da177e4SLinus Torvalds * A helper for ->readlink(). This should be used *ONLY* for symlinks that 26431da177e4SLinus Torvalds * have ->follow_link() touching nd only in nd_set_link(). Using (or not 26441da177e4SLinus Torvalds * using) it for any given inode is up to filesystem. 26451da177e4SLinus Torvalds */ 26461da177e4SLinus Torvalds int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen) 26471da177e4SLinus Torvalds { 26481da177e4SLinus Torvalds struct nameidata nd; 2649cc314eefSLinus Torvalds void *cookie; 2650cc314eefSLinus Torvalds 26511da177e4SLinus Torvalds nd.depth = 0; 2652cc314eefSLinus Torvalds cookie = dentry->d_inode->i_op->follow_link(dentry, &nd); 2653cc314eefSLinus Torvalds if (!IS_ERR(cookie)) { 2654cc314eefSLinus Torvalds int res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd)); 26551da177e4SLinus Torvalds if (dentry->d_inode->i_op->put_link) 2656cc314eefSLinus Torvalds dentry->d_inode->i_op->put_link(dentry, &nd, cookie); 2657cc314eefSLinus Torvalds cookie = ERR_PTR(res); 26581da177e4SLinus Torvalds } 2659cc314eefSLinus Torvalds return PTR_ERR(cookie); 26601da177e4SLinus Torvalds } 26611da177e4SLinus Torvalds 26621da177e4SLinus Torvalds int vfs_follow_link(struct nameidata *nd, const char *link) 26631da177e4SLinus Torvalds { 26641da177e4SLinus Torvalds return __vfs_follow_link(nd, link); 26651da177e4SLinus Torvalds } 26661da177e4SLinus Torvalds 26671da177e4SLinus Torvalds /* get the link contents into pagecache */ 26681da177e4SLinus Torvalds static char *page_getlink(struct dentry * dentry, struct page **ppage) 26691da177e4SLinus Torvalds { 26701da177e4SLinus Torvalds struct page * page; 26711da177e4SLinus Torvalds struct address_space *mapping = dentry->d_inode->i_mapping; 2672090d2b18SPekka Enberg page = read_mapping_page(mapping, 0, NULL); 26731da177e4SLinus Torvalds if (IS_ERR(page)) 26746fe6900eSNick Piggin return (char*)page; 26751da177e4SLinus Torvalds *ppage = page; 26761da177e4SLinus Torvalds return kmap(page); 26771da177e4SLinus Torvalds } 26781da177e4SLinus Torvalds 26791da177e4SLinus Torvalds int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 26801da177e4SLinus Torvalds { 26811da177e4SLinus Torvalds struct page *page = NULL; 26821da177e4SLinus Torvalds char *s = page_getlink(dentry, &page); 26831da177e4SLinus Torvalds int res = vfs_readlink(dentry,buffer,buflen,s); 26841da177e4SLinus Torvalds if (page) { 26851da177e4SLinus Torvalds kunmap(page); 26861da177e4SLinus Torvalds page_cache_release(page); 26871da177e4SLinus Torvalds } 26881da177e4SLinus Torvalds return res; 26891da177e4SLinus Torvalds } 26901da177e4SLinus Torvalds 2691cc314eefSLinus Torvalds void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd) 26921da177e4SLinus Torvalds { 2693cc314eefSLinus Torvalds struct page *page = NULL; 26941da177e4SLinus Torvalds nd_set_link(nd, page_getlink(dentry, &page)); 2695cc314eefSLinus Torvalds return page; 26961da177e4SLinus Torvalds } 26971da177e4SLinus Torvalds 2698cc314eefSLinus Torvalds void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) 26991da177e4SLinus Torvalds { 2700cc314eefSLinus Torvalds struct page *page = cookie; 2701cc314eefSLinus Torvalds 2702cc314eefSLinus Torvalds if (page) { 27031da177e4SLinus Torvalds kunmap(page); 27041da177e4SLinus Torvalds page_cache_release(page); 27051da177e4SLinus Torvalds } 27061da177e4SLinus Torvalds } 27071da177e4SLinus Torvalds 27080adb25d2SKirill Korotaev int __page_symlink(struct inode *inode, const char *symname, int len, 27090adb25d2SKirill Korotaev gfp_t gfp_mask) 27101da177e4SLinus Torvalds { 27111da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 27120adb25d2SKirill Korotaev struct page *page; 2713beb497abSDmitriy Monakhov int err; 27141da177e4SLinus Torvalds char *kaddr; 27151da177e4SLinus Torvalds 27167e53cac4SNeilBrown retry: 2717beb497abSDmitriy Monakhov err = -ENOMEM; 27180adb25d2SKirill Korotaev page = find_or_create_page(mapping, 0, gfp_mask); 27191da177e4SLinus Torvalds if (!page) 27201da177e4SLinus Torvalds goto fail; 27211da177e4SLinus Torvalds err = mapping->a_ops->prepare_write(NULL, page, 0, len-1); 27227e53cac4SNeilBrown if (err == AOP_TRUNCATED_PAGE) { 27237e53cac4SNeilBrown page_cache_release(page); 27247e53cac4SNeilBrown goto retry; 27257e53cac4SNeilBrown } 27261da177e4SLinus Torvalds if (err) 27271da177e4SLinus Torvalds goto fail_map; 27281da177e4SLinus Torvalds kaddr = kmap_atomic(page, KM_USER0); 27291da177e4SLinus Torvalds memcpy(kaddr, symname, len-1); 27301da177e4SLinus Torvalds kunmap_atomic(kaddr, KM_USER0); 27317e53cac4SNeilBrown err = mapping->a_ops->commit_write(NULL, page, 0, len-1); 27327e53cac4SNeilBrown if (err == AOP_TRUNCATED_PAGE) { 27337e53cac4SNeilBrown page_cache_release(page); 27347e53cac4SNeilBrown goto retry; 27357e53cac4SNeilBrown } 27367e53cac4SNeilBrown if (err) 27377e53cac4SNeilBrown goto fail_map; 27381da177e4SLinus Torvalds /* 27391da177e4SLinus Torvalds * Notice that we are _not_ going to block here - end of page is 27401da177e4SLinus Torvalds * unmapped, so this will only try to map the rest of page, see 27411da177e4SLinus Torvalds * that it is unmapped (typically even will not look into inode - 27421da177e4SLinus Torvalds * ->i_size will be enough for everything) and zero it out. 27431da177e4SLinus Torvalds * OTOH it's obviously correct and should make the page up-to-date. 27441da177e4SLinus Torvalds */ 27451da177e4SLinus Torvalds if (!PageUptodate(page)) { 27461da177e4SLinus Torvalds err = mapping->a_ops->readpage(NULL, page); 27477e53cac4SNeilBrown if (err != AOP_TRUNCATED_PAGE) 27481da177e4SLinus Torvalds wait_on_page_locked(page); 27491da177e4SLinus Torvalds } else { 27501da177e4SLinus Torvalds unlock_page(page); 27511da177e4SLinus Torvalds } 27521da177e4SLinus Torvalds page_cache_release(page); 27531da177e4SLinus Torvalds if (err < 0) 27541da177e4SLinus Torvalds goto fail; 27551da177e4SLinus Torvalds mark_inode_dirty(inode); 27561da177e4SLinus Torvalds return 0; 27571da177e4SLinus Torvalds fail_map: 27581da177e4SLinus Torvalds unlock_page(page); 27591da177e4SLinus Torvalds page_cache_release(page); 27601da177e4SLinus Torvalds fail: 27611da177e4SLinus Torvalds return err; 27621da177e4SLinus Torvalds } 27631da177e4SLinus Torvalds 27640adb25d2SKirill Korotaev int page_symlink(struct inode *inode, const char *symname, int len) 27650adb25d2SKirill Korotaev { 27660adb25d2SKirill Korotaev return __page_symlink(inode, symname, len, 27670adb25d2SKirill Korotaev mapping_gfp_mask(inode->i_mapping)); 27680adb25d2SKirill Korotaev } 27690adb25d2SKirill Korotaev 277092e1d5beSArjan van de Ven const struct inode_operations page_symlink_inode_operations = { 27711da177e4SLinus Torvalds .readlink = generic_readlink, 27721da177e4SLinus Torvalds .follow_link = page_follow_link_light, 27731da177e4SLinus Torvalds .put_link = page_put_link, 27741da177e4SLinus Torvalds }; 27751da177e4SLinus Torvalds 27761da177e4SLinus Torvalds EXPORT_SYMBOL(__user_walk); 27775590ff0dSUlrich Drepper EXPORT_SYMBOL(__user_walk_fd); 27781da177e4SLinus Torvalds EXPORT_SYMBOL(follow_down); 27791da177e4SLinus Torvalds EXPORT_SYMBOL(follow_up); 27801da177e4SLinus Torvalds EXPORT_SYMBOL(get_write_access); /* binfmt_aout */ 27811da177e4SLinus Torvalds EXPORT_SYMBOL(getname); 27821da177e4SLinus Torvalds EXPORT_SYMBOL(lock_rename); 27831da177e4SLinus Torvalds EXPORT_SYMBOL(lookup_one_len); 27841da177e4SLinus Torvalds EXPORT_SYMBOL(page_follow_link_light); 27851da177e4SLinus Torvalds EXPORT_SYMBOL(page_put_link); 27861da177e4SLinus Torvalds EXPORT_SYMBOL(page_readlink); 27870adb25d2SKirill Korotaev EXPORT_SYMBOL(__page_symlink); 27881da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink); 27891da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink_inode_operations); 27901da177e4SLinus Torvalds EXPORT_SYMBOL(path_lookup); 27911da177e4SLinus Torvalds EXPORT_SYMBOL(path_release); 27921da177e4SLinus Torvalds EXPORT_SYMBOL(path_walk); 27931da177e4SLinus Torvalds EXPORT_SYMBOL(permission); 2794e4543eddSChristoph Hellwig EXPORT_SYMBOL(vfs_permission); 27958c744fb8SChristoph Hellwig EXPORT_SYMBOL(file_permission); 27961da177e4SLinus Torvalds EXPORT_SYMBOL(unlock_rename); 27971da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_create); 27981da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_follow_link); 27991da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_link); 28001da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_mkdir); 28011da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_mknod); 28021da177e4SLinus Torvalds EXPORT_SYMBOL(generic_permission); 28031da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_readlink); 28041da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_rename); 28051da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_rmdir); 28061da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_symlink); 28071da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_unlink); 28081da177e4SLinus Torvalds EXPORT_SYMBOL(dentry_unhash); 28091da177e4SLinus Torvalds EXPORT_SYMBOL(generic_readlink); 2810