11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * linux/fs/namei.c 31da177e4SLinus Torvalds * 41da177e4SLinus Torvalds * Copyright (C) 1991, 1992 Linus Torvalds 51da177e4SLinus Torvalds */ 61da177e4SLinus Torvalds 71da177e4SLinus Torvalds /* 81da177e4SLinus Torvalds * Some corrections by tytso. 91da177e4SLinus Torvalds */ 101da177e4SLinus Torvalds 111da177e4SLinus Torvalds /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname 121da177e4SLinus Torvalds * lookup logic. 131da177e4SLinus Torvalds */ 141da177e4SLinus Torvalds /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture. 151da177e4SLinus Torvalds */ 161da177e4SLinus Torvalds 171da177e4SLinus Torvalds #include <linux/init.h> 18630d9c47SPaul Gortmaker #include <linux/export.h> 1944696908SDavid S. Miller #include <linux/kernel.h> 201da177e4SLinus Torvalds #include <linux/slab.h> 211da177e4SLinus Torvalds #include <linux/fs.h> 221da177e4SLinus Torvalds #include <linux/namei.h> 231da177e4SLinus Torvalds #include <linux/pagemap.h> 240eeca283SRobert Love #include <linux/fsnotify.h> 251da177e4SLinus Torvalds #include <linux/personality.h> 261da177e4SLinus Torvalds #include <linux/security.h> 276146f0d5SMimi Zohar #include <linux/ima.h> 281da177e4SLinus Torvalds #include <linux/syscalls.h> 291da177e4SLinus Torvalds #include <linux/mount.h> 301da177e4SLinus Torvalds #include <linux/audit.h> 3116f7e0feSRandy Dunlap #include <linux/capability.h> 32834f2a4aSTrond Myklebust #include <linux/file.h> 335590ff0dSUlrich Drepper #include <linux/fcntl.h> 3408ce5f16SSerge E. Hallyn #include <linux/device_cgroup.h> 355ad4e53bSAl Viro #include <linux/fs_struct.h> 36e77819e5SLinus Torvalds #include <linux/posix_acl.h> 371da177e4SLinus Torvalds #include <asm/uaccess.h> 381da177e4SLinus Torvalds 39e81e3f4dSEric Paris #include "internal.h" 40c7105365SAl Viro #include "mount.h" 41e81e3f4dSEric Paris 421da177e4SLinus Torvalds /* [Feb-1997 T. Schoebel-Theuer] 431da177e4SLinus Torvalds * Fundamental changes in the pathname lookup mechanisms (namei) 441da177e4SLinus Torvalds * were necessary because of omirr. The reason is that omirr needs 451da177e4SLinus Torvalds * to know the _real_ pathname, not the user-supplied one, in case 461da177e4SLinus Torvalds * of symlinks (and also when transname replacements occur). 471da177e4SLinus Torvalds * 481da177e4SLinus Torvalds * The new code replaces the old recursive symlink resolution with 491da177e4SLinus Torvalds * an iterative one (in case of non-nested symlink chains). It does 501da177e4SLinus Torvalds * this with calls to <fs>_follow_link(). 511da177e4SLinus Torvalds * As a side effect, dir_namei(), _namei() and follow_link() are now 521da177e4SLinus Torvalds * replaced with a single function lookup_dentry() that can handle all 531da177e4SLinus Torvalds * the special cases of the former code. 541da177e4SLinus Torvalds * 551da177e4SLinus Torvalds * With the new dcache, the pathname is stored at each inode, at least as 561da177e4SLinus Torvalds * long as the refcount of the inode is positive. As a side effect, the 571da177e4SLinus Torvalds * size of the dcache depends on the inode cache and thus is dynamic. 581da177e4SLinus Torvalds * 591da177e4SLinus Torvalds * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink 601da177e4SLinus Torvalds * resolution to correspond with current state of the code. 611da177e4SLinus Torvalds * 621da177e4SLinus Torvalds * Note that the symlink resolution is not *completely* iterative. 631da177e4SLinus Torvalds * There is still a significant amount of tail- and mid- recursion in 641da177e4SLinus Torvalds * the algorithm. Also, note that <fs>_readlink() is not used in 651da177e4SLinus Torvalds * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink() 661da177e4SLinus Torvalds * may return different results than <fs>_follow_link(). Many virtual 671da177e4SLinus Torvalds * filesystems (including /proc) exhibit this behavior. 681da177e4SLinus Torvalds */ 691da177e4SLinus Torvalds 701da177e4SLinus Torvalds /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation: 711da177e4SLinus Torvalds * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL 721da177e4SLinus Torvalds * and the name already exists in form of a symlink, try to create the new 731da177e4SLinus Torvalds * name indicated by the symlink. The old code always complained that the 741da177e4SLinus Torvalds * name already exists, due to not following the symlink even if its target 751da177e4SLinus Torvalds * is nonexistent. The new semantics affects also mknod() and link() when 7625985edcSLucas De Marchi * the name is a symlink pointing to a non-existent name. 771da177e4SLinus Torvalds * 781da177e4SLinus Torvalds * I don't know which semantics is the right one, since I have no access 791da177e4SLinus Torvalds * to standards. But I found by trial that HP-UX 9.0 has the full "new" 801da177e4SLinus Torvalds * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the 811da177e4SLinus Torvalds * "old" one. Personally, I think the new semantics is much more logical. 821da177e4SLinus Torvalds * Note that "ln old new" where "new" is a symlink pointing to a non-existing 831da177e4SLinus Torvalds * file does succeed in both HP-UX and SunOs, but not in Solaris 841da177e4SLinus Torvalds * and in the old Linux semantics. 851da177e4SLinus Torvalds */ 861da177e4SLinus Torvalds 871da177e4SLinus Torvalds /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink 881da177e4SLinus Torvalds * semantics. See the comments in "open_namei" and "do_link" below. 891da177e4SLinus Torvalds * 901da177e4SLinus Torvalds * [10-Sep-98 Alan Modra] Another symlink change. 911da177e4SLinus Torvalds */ 921da177e4SLinus Torvalds 931da177e4SLinus Torvalds /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks: 941da177e4SLinus Torvalds * inside the path - always follow. 951da177e4SLinus Torvalds * in the last component in creation/removal/renaming - never follow. 961da177e4SLinus Torvalds * if LOOKUP_FOLLOW passed - follow. 971da177e4SLinus Torvalds * if the pathname has trailing slashes - follow. 981da177e4SLinus Torvalds * otherwise - don't follow. 991da177e4SLinus Torvalds * (applied in that order). 1001da177e4SLinus Torvalds * 1011da177e4SLinus Torvalds * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT 1021da177e4SLinus Torvalds * restored for 2.4. This is the last surviving part of old 4.2BSD bug. 1031da177e4SLinus Torvalds * During the 2.4 we need to fix the userland stuff depending on it - 1041da177e4SLinus Torvalds * hopefully we will be able to get rid of that wart in 2.5. So far only 1051da177e4SLinus Torvalds * XEmacs seems to be relying on it... 1061da177e4SLinus Torvalds */ 1071da177e4SLinus Torvalds /* 1081da177e4SLinus Torvalds * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland) 109a11f3a05SArjan van de Ven * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives 1101da177e4SLinus Torvalds * any extra contention... 1111da177e4SLinus Torvalds */ 1121da177e4SLinus Torvalds 1131da177e4SLinus Torvalds /* In order to reduce some races, while at the same time doing additional 1141da177e4SLinus Torvalds * checking and hopefully speeding things up, we copy filenames to the 1151da177e4SLinus Torvalds * kernel data space before using them.. 1161da177e4SLinus Torvalds * 1171da177e4SLinus Torvalds * POSIX.1 2.4: an empty pathname is invalid (ENOENT). 1181da177e4SLinus Torvalds * PATH_MAX includes the nul terminator --RR. 1191da177e4SLinus Torvalds */ 1201fa1e7f6SAndy Whitcroft static char *getname_flags(const char __user *filename, int flags, int *empty) 1211da177e4SLinus Torvalds { 1223f9f0aa6SLinus Torvalds char *result = __getname(), *err; 1233f9f0aa6SLinus Torvalds int len; 1241da177e4SLinus Torvalds 1253f9f0aa6SLinus Torvalds if (unlikely(!result)) 1264043cde8SEric Paris return ERR_PTR(-ENOMEM); 1271da177e4SLinus Torvalds 1283f9f0aa6SLinus Torvalds len = strncpy_from_user(result, filename, PATH_MAX); 1293f9f0aa6SLinus Torvalds err = ERR_PTR(len); 1303f9f0aa6SLinus Torvalds if (unlikely(len < 0)) 1313f9f0aa6SLinus Torvalds goto error; 1323f9f0aa6SLinus Torvalds 1333f9f0aa6SLinus Torvalds /* The empty path is special. */ 1343f9f0aa6SLinus Torvalds if (unlikely(!len)) { 1353f9f0aa6SLinus Torvalds if (empty) 1361fa1e7f6SAndy Whitcroft *empty = 1; 1373f9f0aa6SLinus Torvalds err = ERR_PTR(-ENOENT); 1383f9f0aa6SLinus Torvalds if (!(flags & LOOKUP_EMPTY)) 1393f9f0aa6SLinus Torvalds goto error; 1401da177e4SLinus Torvalds } 1413f9f0aa6SLinus Torvalds 1423f9f0aa6SLinus Torvalds err = ERR_PTR(-ENAMETOOLONG); 1433f9f0aa6SLinus Torvalds if (likely(len < PATH_MAX)) { 1441da177e4SLinus Torvalds audit_getname(result); 1451da177e4SLinus Torvalds return result; 1461da177e4SLinus Torvalds } 1471da177e4SLinus Torvalds 1483f9f0aa6SLinus Torvalds error: 1493f9f0aa6SLinus Torvalds __putname(result); 1503f9f0aa6SLinus Torvalds return err; 1513f9f0aa6SLinus Torvalds } 1523f9f0aa6SLinus Torvalds 153f52e0c11SAl Viro char *getname(const char __user * filename) 154f52e0c11SAl Viro { 155f7493e5dSLinus Torvalds return getname_flags(filename, 0, NULL); 156f52e0c11SAl Viro } 157f52e0c11SAl Viro 1581da177e4SLinus Torvalds #ifdef CONFIG_AUDITSYSCALL 1591da177e4SLinus Torvalds void putname(const char *name) 1601da177e4SLinus Torvalds { 1615ac3a9c2SAl Viro if (unlikely(!audit_dummy_context())) 1621da177e4SLinus Torvalds audit_putname(name); 1631da177e4SLinus Torvalds else 1641da177e4SLinus Torvalds __putname(name); 1651da177e4SLinus Torvalds } 1661da177e4SLinus Torvalds EXPORT_SYMBOL(putname); 1671da177e4SLinus Torvalds #endif 1681da177e4SLinus Torvalds 169e77819e5SLinus Torvalds static int check_acl(struct inode *inode, int mask) 170e77819e5SLinus Torvalds { 17184635d68SLinus Torvalds #ifdef CONFIG_FS_POSIX_ACL 172e77819e5SLinus Torvalds struct posix_acl *acl; 173e77819e5SLinus Torvalds 174e77819e5SLinus Torvalds if (mask & MAY_NOT_BLOCK) { 1753567866bSAl Viro acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS); 1763567866bSAl Viro if (!acl) 177e77819e5SLinus Torvalds return -EAGAIN; 1783567866bSAl Viro /* no ->get_acl() calls in RCU mode... */ 1793567866bSAl Viro if (acl == ACL_NOT_CACHED) 180e77819e5SLinus Torvalds return -ECHILD; 181206b1d09SAri Savolainen return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK); 182e77819e5SLinus Torvalds } 183e77819e5SLinus Torvalds 184e77819e5SLinus Torvalds acl = get_cached_acl(inode, ACL_TYPE_ACCESS); 185e77819e5SLinus Torvalds 186e77819e5SLinus Torvalds /* 1874e34e719SChristoph Hellwig * A filesystem can force a ACL callback by just never filling the 1884e34e719SChristoph Hellwig * ACL cache. But normally you'd fill the cache either at inode 1894e34e719SChristoph Hellwig * instantiation time, or on the first ->get_acl call. 190e77819e5SLinus Torvalds * 1914e34e719SChristoph Hellwig * If the filesystem doesn't have a get_acl() function at all, we'll 1924e34e719SChristoph Hellwig * just create the negative cache entry. 193e77819e5SLinus Torvalds */ 194e77819e5SLinus Torvalds if (acl == ACL_NOT_CACHED) { 1954e34e719SChristoph Hellwig if (inode->i_op->get_acl) { 1964e34e719SChristoph Hellwig acl = inode->i_op->get_acl(inode, ACL_TYPE_ACCESS); 1974e34e719SChristoph Hellwig if (IS_ERR(acl)) 1984e34e719SChristoph Hellwig return PTR_ERR(acl); 1994e34e719SChristoph Hellwig } else { 200e77819e5SLinus Torvalds set_cached_acl(inode, ACL_TYPE_ACCESS, NULL); 201e77819e5SLinus Torvalds return -EAGAIN; 202e77819e5SLinus Torvalds } 2034e34e719SChristoph Hellwig } 204e77819e5SLinus Torvalds 205e77819e5SLinus Torvalds if (acl) { 206e77819e5SLinus Torvalds int error = posix_acl_permission(inode, acl, mask); 207e77819e5SLinus Torvalds posix_acl_release(acl); 208e77819e5SLinus Torvalds return error; 209e77819e5SLinus Torvalds } 21084635d68SLinus Torvalds #endif 211e77819e5SLinus Torvalds 212e77819e5SLinus Torvalds return -EAGAIN; 213e77819e5SLinus Torvalds } 214e77819e5SLinus Torvalds 2155909ccaaSLinus Torvalds /* 216948409c7SAndreas Gruenbacher * This does the basic permission checking 2175909ccaaSLinus Torvalds */ 2187e40145eSAl Viro static int acl_permission_check(struct inode *inode, int mask) 2195909ccaaSLinus Torvalds { 22026cf46beSLinus Torvalds unsigned int mode = inode->i_mode; 2215909ccaaSLinus Torvalds 2228e96e3b7SEric W. Biederman if (likely(uid_eq(current_fsuid(), inode->i_uid))) 2235909ccaaSLinus Torvalds mode >>= 6; 2245909ccaaSLinus Torvalds else { 225e77819e5SLinus Torvalds if (IS_POSIXACL(inode) && (mode & S_IRWXG)) { 2267e40145eSAl Viro int error = check_acl(inode, mask); 2275909ccaaSLinus Torvalds if (error != -EAGAIN) 2285909ccaaSLinus Torvalds return error; 2295909ccaaSLinus Torvalds } 2305909ccaaSLinus Torvalds 2315909ccaaSLinus Torvalds if (in_group_p(inode->i_gid)) 2325909ccaaSLinus Torvalds mode >>= 3; 2335909ccaaSLinus Torvalds } 2345909ccaaSLinus Torvalds 2355909ccaaSLinus Torvalds /* 2365909ccaaSLinus Torvalds * If the DACs are ok we don't need any capability check. 2375909ccaaSLinus Torvalds */ 2389c2c7039SAl Viro if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) 2395909ccaaSLinus Torvalds return 0; 2405909ccaaSLinus Torvalds return -EACCES; 2415909ccaaSLinus Torvalds } 2421da177e4SLinus Torvalds 2431da177e4SLinus Torvalds /** 2441da177e4SLinus Torvalds * generic_permission - check for access rights on a Posix-like filesystem 2451da177e4SLinus Torvalds * @inode: inode to check access rights for 2468fd90c8dSAndreas Gruenbacher * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...) 2471da177e4SLinus Torvalds * 2481da177e4SLinus Torvalds * Used to check for read/write/execute permissions on a file. 2491da177e4SLinus Torvalds * We use "fsuid" for this, letting us set arbitrary permissions 2501da177e4SLinus Torvalds * for filesystem access without changing the "normal" uids which 251b74c79e9SNick Piggin * are used for other things. 252b74c79e9SNick Piggin * 253b74c79e9SNick Piggin * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk 254b74c79e9SNick Piggin * request cannot be satisfied (eg. requires blocking or too much complexity). 255b74c79e9SNick Piggin * It would then be called again in ref-walk mode. 2561da177e4SLinus Torvalds */ 2572830ba7fSAl Viro int generic_permission(struct inode *inode, int mask) 2581da177e4SLinus Torvalds { 2595909ccaaSLinus Torvalds int ret; 2601da177e4SLinus Torvalds 2611da177e4SLinus Torvalds /* 262948409c7SAndreas Gruenbacher * Do the basic permission checks. 2631da177e4SLinus Torvalds */ 2647e40145eSAl Viro ret = acl_permission_check(inode, mask); 2655909ccaaSLinus Torvalds if (ret != -EACCES) 2665909ccaaSLinus Torvalds return ret; 2671da177e4SLinus Torvalds 268d594e7ecSAl Viro if (S_ISDIR(inode->i_mode)) { 269d594e7ecSAl Viro /* DACs are overridable for directories */ 2701a48e2acSEric W. Biederman if (inode_capable(inode, CAP_DAC_OVERRIDE)) 271d594e7ecSAl Viro return 0; 272d594e7ecSAl Viro if (!(mask & MAY_WRITE)) 2731a48e2acSEric W. Biederman if (inode_capable(inode, CAP_DAC_READ_SEARCH)) 274d594e7ecSAl Viro return 0; 275d594e7ecSAl Viro return -EACCES; 276d594e7ecSAl Viro } 2771da177e4SLinus Torvalds /* 2781da177e4SLinus Torvalds * Read/write DACs are always overridable. 279d594e7ecSAl Viro * Executable DACs are overridable when there is 280d594e7ecSAl Viro * at least one exec bit set. 2811da177e4SLinus Torvalds */ 282d594e7ecSAl Viro if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO)) 2831a48e2acSEric W. Biederman if (inode_capable(inode, CAP_DAC_OVERRIDE)) 2841da177e4SLinus Torvalds return 0; 2851da177e4SLinus Torvalds 2861da177e4SLinus Torvalds /* 2871da177e4SLinus Torvalds * Searching includes executable on directories, else just read. 2881da177e4SLinus Torvalds */ 2897ea66001SSerge E. Hallyn mask &= MAY_READ | MAY_WRITE | MAY_EXEC; 290d594e7ecSAl Viro if (mask == MAY_READ) 2911a48e2acSEric W. Biederman if (inode_capable(inode, CAP_DAC_READ_SEARCH)) 2921da177e4SLinus Torvalds return 0; 2931da177e4SLinus Torvalds 2941da177e4SLinus Torvalds return -EACCES; 2951da177e4SLinus Torvalds } 2961da177e4SLinus Torvalds 2973ddcd056SLinus Torvalds /* 2983ddcd056SLinus Torvalds * We _really_ want to just do "generic_permission()" without 2993ddcd056SLinus Torvalds * even looking at the inode->i_op values. So we keep a cache 3003ddcd056SLinus Torvalds * flag in inode->i_opflags, that says "this has not special 3013ddcd056SLinus Torvalds * permission function, use the fast case". 3023ddcd056SLinus Torvalds */ 3033ddcd056SLinus Torvalds static inline int do_inode_permission(struct inode *inode, int mask) 3043ddcd056SLinus Torvalds { 3053ddcd056SLinus Torvalds if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) { 3063ddcd056SLinus Torvalds if (likely(inode->i_op->permission)) 3073ddcd056SLinus Torvalds return inode->i_op->permission(inode, mask); 3083ddcd056SLinus Torvalds 3093ddcd056SLinus Torvalds /* This gets set once for the inode lifetime */ 3103ddcd056SLinus Torvalds spin_lock(&inode->i_lock); 3113ddcd056SLinus Torvalds inode->i_opflags |= IOP_FASTPERM; 3123ddcd056SLinus Torvalds spin_unlock(&inode->i_lock); 3133ddcd056SLinus Torvalds } 3143ddcd056SLinus Torvalds return generic_permission(inode, mask); 3153ddcd056SLinus Torvalds } 3163ddcd056SLinus Torvalds 317cb23beb5SChristoph Hellwig /** 318cb23beb5SChristoph Hellwig * inode_permission - check for access rights to a given inode 319cb23beb5SChristoph Hellwig * @inode: inode to check permission on 3208fd90c8dSAndreas Gruenbacher * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...) 321cb23beb5SChristoph Hellwig * 322cb23beb5SChristoph Hellwig * Used to check for read/write/execute permissions on an inode. 323cb23beb5SChristoph Hellwig * We use "fsuid" for this, letting us set arbitrary permissions 324cb23beb5SChristoph Hellwig * for filesystem access without changing the "normal" uids which 325cb23beb5SChristoph Hellwig * are used for other things. 326948409c7SAndreas Gruenbacher * 327948409c7SAndreas Gruenbacher * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. 328cb23beb5SChristoph Hellwig */ 329f419a2e3SAl Viro int inode_permission(struct inode *inode, int mask) 3301da177e4SLinus Torvalds { 331e6305c43SAl Viro int retval; 3321da177e4SLinus Torvalds 3333ddcd056SLinus Torvalds if (unlikely(mask & MAY_WRITE)) { 33422590e41SMiklos Szeredi umode_t mode = inode->i_mode; 3351da177e4SLinus Torvalds 3361da177e4SLinus Torvalds /* 3371da177e4SLinus Torvalds * Nobody gets write access to a read-only fs. 3381da177e4SLinus Torvalds */ 3391da177e4SLinus Torvalds if (IS_RDONLY(inode) && 3401da177e4SLinus Torvalds (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) 3411da177e4SLinus Torvalds return -EROFS; 3421da177e4SLinus Torvalds 3431da177e4SLinus Torvalds /* 3441da177e4SLinus Torvalds * Nobody gets write access to an immutable file. 3451da177e4SLinus Torvalds */ 3461da177e4SLinus Torvalds if (IS_IMMUTABLE(inode)) 3471da177e4SLinus Torvalds return -EACCES; 3481da177e4SLinus Torvalds } 3491da177e4SLinus Torvalds 3503ddcd056SLinus Torvalds retval = do_inode_permission(inode, mask); 3511da177e4SLinus Torvalds if (retval) 3521da177e4SLinus Torvalds return retval; 3531da177e4SLinus Torvalds 35408ce5f16SSerge E. Hallyn retval = devcgroup_inode_permission(inode, mask); 35508ce5f16SSerge E. Hallyn if (retval) 35608ce5f16SSerge E. Hallyn return retval; 35708ce5f16SSerge E. Hallyn 358d09ca739SEric Paris return security_inode_permission(inode, mask); 3591da177e4SLinus Torvalds } 3601da177e4SLinus Torvalds 361f4d6ff89SAl Viro /** 3625dd784d0SJan Blunck * path_get - get a reference to a path 3635dd784d0SJan Blunck * @path: path to get the reference to 3645dd784d0SJan Blunck * 3655dd784d0SJan Blunck * Given a path increment the reference count to the dentry and the vfsmount. 3665dd784d0SJan Blunck */ 3675dd784d0SJan Blunck void path_get(struct path *path) 3685dd784d0SJan Blunck { 3695dd784d0SJan Blunck mntget(path->mnt); 3705dd784d0SJan Blunck dget(path->dentry); 3715dd784d0SJan Blunck } 3725dd784d0SJan Blunck EXPORT_SYMBOL(path_get); 3735dd784d0SJan Blunck 3745dd784d0SJan Blunck /** 3751d957f9bSJan Blunck * path_put - put a reference to a path 3761d957f9bSJan Blunck * @path: path to put the reference to 3771d957f9bSJan Blunck * 3781d957f9bSJan Blunck * Given a path decrement the reference count to the dentry and the vfsmount. 3791d957f9bSJan Blunck */ 3801d957f9bSJan Blunck void path_put(struct path *path) 3811da177e4SLinus Torvalds { 3821d957f9bSJan Blunck dput(path->dentry); 3831d957f9bSJan Blunck mntput(path->mnt); 3841da177e4SLinus Torvalds } 3851d957f9bSJan Blunck EXPORT_SYMBOL(path_put); 3861da177e4SLinus Torvalds 38719660af7SAl Viro /* 38831e6b01fSNick Piggin * Path walking has 2 modes, rcu-walk and ref-walk (see 38919660af7SAl Viro * Documentation/filesystems/path-lookup.txt). In situations when we can't 39019660af7SAl Viro * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab 39119660af7SAl Viro * normal reference counts on dentries and vfsmounts to transition to rcu-walk 39219660af7SAl Viro * mode. Refcounts are grabbed at the last known good point before rcu-walk 39319660af7SAl Viro * got stuck, so ref-walk may continue from there. If this is not successful 39419660af7SAl Viro * (eg. a seqcount has changed), then failure is returned and it's up to caller 39519660af7SAl Viro * to restart the path walk from the beginning in ref-walk mode. 39631e6b01fSNick Piggin */ 39731e6b01fSNick Piggin 39831e6b01fSNick Piggin /** 39919660af7SAl Viro * unlazy_walk - try to switch to ref-walk mode. 40019660af7SAl Viro * @nd: nameidata pathwalk data 40119660af7SAl Viro * @dentry: child of nd->path.dentry or NULL 40239191628SRandy Dunlap * Returns: 0 on success, -ECHILD on failure 40331e6b01fSNick Piggin * 40419660af7SAl Viro * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry 40519660af7SAl Viro * for ref-walk mode. @dentry must be a path found by a do_lookup call on 40619660af7SAl Viro * @nd or NULL. Must be called from rcu-walk context. 40731e6b01fSNick Piggin */ 40819660af7SAl Viro static int unlazy_walk(struct nameidata *nd, struct dentry *dentry) 40931e6b01fSNick Piggin { 41031e6b01fSNick Piggin struct fs_struct *fs = current->fs; 41131e6b01fSNick Piggin struct dentry *parent = nd->path.dentry; 4125b6ca027SAl Viro int want_root = 0; 41331e6b01fSNick Piggin 41431e6b01fSNick Piggin BUG_ON(!(nd->flags & LOOKUP_RCU)); 4155b6ca027SAl Viro if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 4165b6ca027SAl Viro want_root = 1; 41731e6b01fSNick Piggin spin_lock(&fs->lock); 41831e6b01fSNick Piggin if (nd->root.mnt != fs->root.mnt || 41931e6b01fSNick Piggin nd->root.dentry != fs->root.dentry) 42031e6b01fSNick Piggin goto err_root; 42131e6b01fSNick Piggin } 42231e6b01fSNick Piggin spin_lock(&parent->d_lock); 42319660af7SAl Viro if (!dentry) { 42419660af7SAl Viro if (!__d_rcu_to_refcount(parent, nd->seq)) 42519660af7SAl Viro goto err_parent; 42619660af7SAl Viro BUG_ON(nd->inode != parent->d_inode); 42719660af7SAl Viro } else { 42894c0d4ecSAl Viro if (dentry->d_parent != parent) 42994c0d4ecSAl Viro goto err_parent; 43031e6b01fSNick Piggin spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); 43131e6b01fSNick Piggin if (!__d_rcu_to_refcount(dentry, nd->seq)) 43219660af7SAl Viro goto err_child; 43331e6b01fSNick Piggin /* 43419660af7SAl Viro * If the sequence check on the child dentry passed, then 43519660af7SAl Viro * the child has not been removed from its parent. This 43619660af7SAl Viro * means the parent dentry must be valid and able to take 43719660af7SAl Viro * a reference at this point. 43831e6b01fSNick Piggin */ 43931e6b01fSNick Piggin BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent); 44031e6b01fSNick Piggin BUG_ON(!parent->d_count); 44131e6b01fSNick Piggin parent->d_count++; 44231e6b01fSNick Piggin spin_unlock(&dentry->d_lock); 44319660af7SAl Viro } 44431e6b01fSNick Piggin spin_unlock(&parent->d_lock); 4455b6ca027SAl Viro if (want_root) { 44631e6b01fSNick Piggin path_get(&nd->root); 44731e6b01fSNick Piggin spin_unlock(&fs->lock); 44831e6b01fSNick Piggin } 44931e6b01fSNick Piggin mntget(nd->path.mnt); 45031e6b01fSNick Piggin 45131e6b01fSNick Piggin rcu_read_unlock(); 452962830dfSAndi Kleen br_read_unlock(&vfsmount_lock); 45331e6b01fSNick Piggin nd->flags &= ~LOOKUP_RCU; 45431e6b01fSNick Piggin return 0; 45519660af7SAl Viro 45619660af7SAl Viro err_child: 45731e6b01fSNick Piggin spin_unlock(&dentry->d_lock); 45819660af7SAl Viro err_parent: 45931e6b01fSNick Piggin spin_unlock(&parent->d_lock); 46031e6b01fSNick Piggin err_root: 4615b6ca027SAl Viro if (want_root) 46231e6b01fSNick Piggin spin_unlock(&fs->lock); 46331e6b01fSNick Piggin return -ECHILD; 46431e6b01fSNick Piggin } 46531e6b01fSNick Piggin 4664ce16ef3SAl Viro static inline int d_revalidate(struct dentry *dentry, unsigned int flags) 46734286d66SNick Piggin { 4684ce16ef3SAl Viro return dentry->d_op->d_revalidate(dentry, flags); 46934286d66SNick Piggin } 47034286d66SNick Piggin 4719f1fafeeSAl Viro /** 4729f1fafeeSAl Viro * complete_walk - successful completion of path walk 4739f1fafeeSAl Viro * @nd: pointer nameidata 47439159de2SJeff Layton * 4759f1fafeeSAl Viro * If we had been in RCU mode, drop out of it and legitimize nd->path. 4769f1fafeeSAl Viro * Revalidate the final result, unless we'd already done that during 4779f1fafeeSAl Viro * the path walk or the filesystem doesn't ask for it. Return 0 on 4789f1fafeeSAl Viro * success, -error on failure. In case of failure caller does not 4799f1fafeeSAl Viro * need to drop nd->path. 48039159de2SJeff Layton */ 4819f1fafeeSAl Viro static int complete_walk(struct nameidata *nd) 48239159de2SJeff Layton { 48316c2cd71SAl Viro struct dentry *dentry = nd->path.dentry; 48439159de2SJeff Layton int status; 48539159de2SJeff Layton 4869f1fafeeSAl Viro if (nd->flags & LOOKUP_RCU) { 4879f1fafeeSAl Viro nd->flags &= ~LOOKUP_RCU; 4889f1fafeeSAl Viro if (!(nd->flags & LOOKUP_ROOT)) 4899f1fafeeSAl Viro nd->root.mnt = NULL; 4909f1fafeeSAl Viro spin_lock(&dentry->d_lock); 4919f1fafeeSAl Viro if (unlikely(!__d_rcu_to_refcount(dentry, nd->seq))) { 4929f1fafeeSAl Viro spin_unlock(&dentry->d_lock); 4939f1fafeeSAl Viro rcu_read_unlock(); 494962830dfSAndi Kleen br_read_unlock(&vfsmount_lock); 4959f1fafeeSAl Viro return -ECHILD; 4969f1fafeeSAl Viro } 4979f1fafeeSAl Viro BUG_ON(nd->inode != dentry->d_inode); 4989f1fafeeSAl Viro spin_unlock(&dentry->d_lock); 4999f1fafeeSAl Viro mntget(nd->path.mnt); 5009f1fafeeSAl Viro rcu_read_unlock(); 501962830dfSAndi Kleen br_read_unlock(&vfsmount_lock); 5029f1fafeeSAl Viro } 5039f1fafeeSAl Viro 50416c2cd71SAl Viro if (likely(!(nd->flags & LOOKUP_JUMPED))) 50539159de2SJeff Layton return 0; 50639159de2SJeff Layton 50716c2cd71SAl Viro if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE))) 50816c2cd71SAl Viro return 0; 50916c2cd71SAl Viro 51016c2cd71SAl Viro if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT))) 51116c2cd71SAl Viro return 0; 51216c2cd71SAl Viro 51316c2cd71SAl Viro /* Note: we do not d_invalidate() */ 5144ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 51539159de2SJeff Layton if (status > 0) 51639159de2SJeff Layton return 0; 51739159de2SJeff Layton 51816c2cd71SAl Viro if (!status) 51939159de2SJeff Layton status = -ESTALE; 52016c2cd71SAl Viro 5219f1fafeeSAl Viro path_put(&nd->path); 52239159de2SJeff Layton return status; 52339159de2SJeff Layton } 52439159de2SJeff Layton 5252a737871SAl Viro static __always_inline void set_root(struct nameidata *nd) 5262a737871SAl Viro { 527f7ad3c6bSMiklos Szeredi if (!nd->root.mnt) 528f7ad3c6bSMiklos Szeredi get_fs_root(current->fs, &nd->root); 5292a737871SAl Viro } 5302a737871SAl Viro 5316de88d72SAl Viro static int link_path_walk(const char *, struct nameidata *); 5326de88d72SAl Viro 53331e6b01fSNick Piggin static __always_inline void set_root_rcu(struct nameidata *nd) 53431e6b01fSNick Piggin { 53531e6b01fSNick Piggin if (!nd->root.mnt) { 53631e6b01fSNick Piggin struct fs_struct *fs = current->fs; 537c28cc364SNick Piggin unsigned seq; 538c28cc364SNick Piggin 539c28cc364SNick Piggin do { 540c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 54131e6b01fSNick Piggin nd->root = fs->root; 542c1530019STim Chen nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq); 543c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 54431e6b01fSNick Piggin } 54531e6b01fSNick Piggin } 54631e6b01fSNick Piggin 547f1662356SArjan van de Ven static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link) 5481da177e4SLinus Torvalds { 54931e6b01fSNick Piggin int ret; 55031e6b01fSNick Piggin 5511da177e4SLinus Torvalds if (IS_ERR(link)) 5521da177e4SLinus Torvalds goto fail; 5531da177e4SLinus Torvalds 5541da177e4SLinus Torvalds if (*link == '/') { 5552a737871SAl Viro set_root(nd); 5561d957f9bSJan Blunck path_put(&nd->path); 5572a737871SAl Viro nd->path = nd->root; 5582a737871SAl Viro path_get(&nd->root); 55916c2cd71SAl Viro nd->flags |= LOOKUP_JUMPED; 5601da177e4SLinus Torvalds } 56131e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 562b4091d5fSChristoph Hellwig 56331e6b01fSNick Piggin ret = link_path_walk(link, nd); 56431e6b01fSNick Piggin return ret; 5651da177e4SLinus Torvalds fail: 5661d957f9bSJan Blunck path_put(&nd->path); 5671da177e4SLinus Torvalds return PTR_ERR(link); 5681da177e4SLinus Torvalds } 5691da177e4SLinus Torvalds 5701d957f9bSJan Blunck static void path_put_conditional(struct path *path, struct nameidata *nd) 571051d3812SIan Kent { 572051d3812SIan Kent dput(path->dentry); 5734ac91378SJan Blunck if (path->mnt != nd->path.mnt) 574051d3812SIan Kent mntput(path->mnt); 575051d3812SIan Kent } 576051d3812SIan Kent 5777b9337aaSNick Piggin static inline void path_to_nameidata(const struct path *path, 5787b9337aaSNick Piggin struct nameidata *nd) 579051d3812SIan Kent { 58031e6b01fSNick Piggin if (!(nd->flags & LOOKUP_RCU)) { 5814ac91378SJan Blunck dput(nd->path.dentry); 58231e6b01fSNick Piggin if (nd->path.mnt != path->mnt) 5834ac91378SJan Blunck mntput(nd->path.mnt); 5849a229683SHuang Shijie } 58531e6b01fSNick Piggin nd->path.mnt = path->mnt; 5864ac91378SJan Blunck nd->path.dentry = path->dentry; 587051d3812SIan Kent } 588051d3812SIan Kent 589b5fb63c1SChristoph Hellwig /* 590b5fb63c1SChristoph Hellwig * Helper to directly jump to a known parsed path from ->follow_link, 591b5fb63c1SChristoph Hellwig * caller must have taken a reference to path beforehand. 592b5fb63c1SChristoph Hellwig */ 593b5fb63c1SChristoph Hellwig void nd_jump_link(struct nameidata *nd, struct path *path) 594b5fb63c1SChristoph Hellwig { 595b5fb63c1SChristoph Hellwig path_put(&nd->path); 596b5fb63c1SChristoph Hellwig 597b5fb63c1SChristoph Hellwig nd->path = *path; 598b5fb63c1SChristoph Hellwig nd->inode = nd->path.dentry->d_inode; 599b5fb63c1SChristoph Hellwig nd->flags |= LOOKUP_JUMPED; 600b5fb63c1SChristoph Hellwig 601b5fb63c1SChristoph Hellwig BUG_ON(nd->inode->i_op->follow_link); 602b5fb63c1SChristoph Hellwig } 603b5fb63c1SChristoph Hellwig 604574197e0SAl Viro static inline void put_link(struct nameidata *nd, struct path *link, void *cookie) 605574197e0SAl Viro { 606574197e0SAl Viro struct inode *inode = link->dentry->d_inode; 6076d7b5aaeSAl Viro if (inode->i_op->put_link) 608574197e0SAl Viro inode->i_op->put_link(link->dentry, nd, cookie); 609574197e0SAl Viro path_put(link); 610574197e0SAl Viro } 611574197e0SAl Viro 612def4af30SAl Viro static __always_inline int 613574197e0SAl Viro follow_link(struct path *link, struct nameidata *nd, void **p) 6141da177e4SLinus Torvalds { 6157b9337aaSNick Piggin struct dentry *dentry = link->dentry; 6166d7b5aaeSAl Viro int error; 6176d7b5aaeSAl Viro char *s; 6181da177e4SLinus Torvalds 619844a3917SAl Viro BUG_ON(nd->flags & LOOKUP_RCU); 620844a3917SAl Viro 6210e794589SAl Viro if (link->mnt == nd->path.mnt) 6220e794589SAl Viro mntget(link->mnt); 6230e794589SAl Viro 6246d7b5aaeSAl Viro error = -ELOOP; 6256d7b5aaeSAl Viro if (unlikely(current->total_link_count >= 40)) 6266d7b5aaeSAl Viro goto out_put_nd_path; 6276d7b5aaeSAl Viro 628574197e0SAl Viro cond_resched(); 629574197e0SAl Viro current->total_link_count++; 630574197e0SAl Viro 63168ac1234SAl Viro touch_atime(link); 6321da177e4SLinus Torvalds nd_set_link(nd, NULL); 633cd4e91d3SAl Viro 63436f3b4f6SAl Viro error = security_inode_follow_link(link->dentry, nd); 6356d7b5aaeSAl Viro if (error) 6366d7b5aaeSAl Viro goto out_put_nd_path; 63736f3b4f6SAl Viro 63886acdca1SAl Viro nd->last_type = LAST_BIND; 639def4af30SAl Viro *p = dentry->d_inode->i_op->follow_link(dentry, nd); 640def4af30SAl Viro error = PTR_ERR(*p); 6416d7b5aaeSAl Viro if (IS_ERR(*p)) 642408ef013SChristoph Hellwig goto out_put_nd_path; 6436d7b5aaeSAl Viro 644cc314eefSLinus Torvalds error = 0; 6456d7b5aaeSAl Viro s = nd_get_link(nd); 6466d7b5aaeSAl Viro if (s) { 6471da177e4SLinus Torvalds error = __vfs_follow_link(nd, s); 6486d7b5aaeSAl Viro if (unlikely(error)) 6496d7b5aaeSAl Viro put_link(nd, link, *p); 650b5fb63c1SChristoph Hellwig } 6516d7b5aaeSAl Viro 6526d7b5aaeSAl Viro return error; 6536d7b5aaeSAl Viro 6546d7b5aaeSAl Viro out_put_nd_path: 6556d7b5aaeSAl Viro path_put(&nd->path); 6566d7b5aaeSAl Viro path_put(link); 6571da177e4SLinus Torvalds return error; 6581da177e4SLinus Torvalds } 6591da177e4SLinus Torvalds 66031e6b01fSNick Piggin static int follow_up_rcu(struct path *path) 66131e6b01fSNick Piggin { 6620714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 6630714a533SAl Viro struct mount *parent; 66431e6b01fSNick Piggin struct dentry *mountpoint; 66531e6b01fSNick Piggin 6660714a533SAl Viro parent = mnt->mnt_parent; 6670714a533SAl Viro if (&parent->mnt == path->mnt) 66831e6b01fSNick Piggin return 0; 669a73324daSAl Viro mountpoint = mnt->mnt_mountpoint; 67031e6b01fSNick Piggin path->dentry = mountpoint; 6710714a533SAl Viro path->mnt = &parent->mnt; 67231e6b01fSNick Piggin return 1; 67331e6b01fSNick Piggin } 67431e6b01fSNick Piggin 675f015f126SDavid Howells /* 676f015f126SDavid Howells * follow_up - Find the mountpoint of path's vfsmount 677f015f126SDavid Howells * 678f015f126SDavid Howells * Given a path, find the mountpoint of its source file system. 679f015f126SDavid Howells * Replace @path with the path of the mountpoint in the parent mount. 680f015f126SDavid Howells * Up is towards /. 681f015f126SDavid Howells * 682f015f126SDavid Howells * Return 1 if we went up a level and 0 if we were already at the 683f015f126SDavid Howells * root. 684f015f126SDavid Howells */ 685bab77ebfSAl Viro int follow_up(struct path *path) 6861da177e4SLinus Torvalds { 6870714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 6880714a533SAl Viro struct mount *parent; 6891da177e4SLinus Torvalds struct dentry *mountpoint; 69099b7db7bSNick Piggin 691962830dfSAndi Kleen br_read_lock(&vfsmount_lock); 6920714a533SAl Viro parent = mnt->mnt_parent; 6930714a533SAl Viro if (&parent->mnt == path->mnt) { 694962830dfSAndi Kleen br_read_unlock(&vfsmount_lock); 6951da177e4SLinus Torvalds return 0; 6961da177e4SLinus Torvalds } 6970714a533SAl Viro mntget(&parent->mnt); 698a73324daSAl Viro mountpoint = dget(mnt->mnt_mountpoint); 699962830dfSAndi Kleen br_read_unlock(&vfsmount_lock); 700bab77ebfSAl Viro dput(path->dentry); 701bab77ebfSAl Viro path->dentry = mountpoint; 702bab77ebfSAl Viro mntput(path->mnt); 7030714a533SAl Viro path->mnt = &parent->mnt; 7041da177e4SLinus Torvalds return 1; 7051da177e4SLinus Torvalds } 7061da177e4SLinus Torvalds 707b5c84bf6SNick Piggin /* 7089875cf80SDavid Howells * Perform an automount 7099875cf80SDavid Howells * - return -EISDIR to tell follow_managed() to stop and return the path we 7109875cf80SDavid Howells * were called with. 7111da177e4SLinus Torvalds */ 7129875cf80SDavid Howells static int follow_automount(struct path *path, unsigned flags, 7139875cf80SDavid Howells bool *need_mntput) 71431e6b01fSNick Piggin { 7159875cf80SDavid Howells struct vfsmount *mnt; 716ea5b778aSDavid Howells int err; 7179875cf80SDavid Howells 7189875cf80SDavid Howells if (!path->dentry->d_op || !path->dentry->d_op->d_automount) 7199875cf80SDavid Howells return -EREMOTE; 7209875cf80SDavid Howells 7210ec26fd0SMiklos Szeredi /* We don't want to mount if someone's just doing a stat - 7220ec26fd0SMiklos Szeredi * unless they're stat'ing a directory and appended a '/' to 7230ec26fd0SMiklos Szeredi * the name. 7240ec26fd0SMiklos Szeredi * 7250ec26fd0SMiklos Szeredi * We do, however, want to mount if someone wants to open or 7265a30d8a2SDavid Howells * create a file of any type under the mountpoint, wants to 7270ec26fd0SMiklos Szeredi * traverse through the mountpoint or wants to open the 7280ec26fd0SMiklos Szeredi * mounted directory. Also, autofs may mark negative dentries 7290ec26fd0SMiklos Szeredi * as being automount points. These will need the attentions 7300ec26fd0SMiklos Szeredi * of the daemon to instantiate them before they can be used. 7315a30d8a2SDavid Howells */ 7325a30d8a2SDavid Howells if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | 733d94c177bSLinus Torvalds LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && 7345a30d8a2SDavid Howells path->dentry->d_inode) 7359875cf80SDavid Howells return -EISDIR; 7360ec26fd0SMiklos Szeredi 7379875cf80SDavid Howells current->total_link_count++; 7389875cf80SDavid Howells if (current->total_link_count >= 40) 7399875cf80SDavid Howells return -ELOOP; 7409875cf80SDavid Howells 7419875cf80SDavid Howells mnt = path->dentry->d_op->d_automount(path); 7429875cf80SDavid Howells if (IS_ERR(mnt)) { 7439875cf80SDavid Howells /* 7449875cf80SDavid Howells * The filesystem is allowed to return -EISDIR here to indicate 7459875cf80SDavid Howells * it doesn't want to automount. For instance, autofs would do 7469875cf80SDavid Howells * this so that its userspace daemon can mount on this dentry. 7479875cf80SDavid Howells * 7489875cf80SDavid Howells * However, we can only permit this if it's a terminal point in 7499875cf80SDavid Howells * the path being looked up; if it wasn't then the remainder of 7509875cf80SDavid Howells * the path is inaccessible and we should say so. 7519875cf80SDavid Howells */ 75249084c3bSAl Viro if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT)) 7539875cf80SDavid Howells return -EREMOTE; 7549875cf80SDavid Howells return PTR_ERR(mnt); 75531e6b01fSNick Piggin } 756ea5b778aSDavid Howells 7579875cf80SDavid Howells if (!mnt) /* mount collision */ 7589875cf80SDavid Howells return 0; 7599875cf80SDavid Howells 7608aef1884SAl Viro if (!*need_mntput) { 7618aef1884SAl Viro /* lock_mount() may release path->mnt on error */ 7628aef1884SAl Viro mntget(path->mnt); 7638aef1884SAl Viro *need_mntput = true; 7648aef1884SAl Viro } 76519a167afSAl Viro err = finish_automount(mnt, path); 766ea5b778aSDavid Howells 767ea5b778aSDavid Howells switch (err) { 768ea5b778aSDavid Howells case -EBUSY: 769ea5b778aSDavid Howells /* Someone else made a mount here whilst we were busy */ 77019a167afSAl Viro return 0; 771ea5b778aSDavid Howells case 0: 7728aef1884SAl Viro path_put(path); 7739875cf80SDavid Howells path->mnt = mnt; 7749875cf80SDavid Howells path->dentry = dget(mnt->mnt_root); 7759875cf80SDavid Howells return 0; 77619a167afSAl Viro default: 77719a167afSAl Viro return err; 7789875cf80SDavid Howells } 77919a167afSAl Viro 780ea5b778aSDavid Howells } 7819875cf80SDavid Howells 7829875cf80SDavid Howells /* 7839875cf80SDavid Howells * Handle a dentry that is managed in some way. 784cc53ce53SDavid Howells * - Flagged for transit management (autofs) 7859875cf80SDavid Howells * - Flagged as mountpoint 7869875cf80SDavid Howells * - Flagged as automount point 7879875cf80SDavid Howells * 7889875cf80SDavid Howells * This may only be called in refwalk mode. 7899875cf80SDavid Howells * 7909875cf80SDavid Howells * Serialization is taken care of in namespace.c 7919875cf80SDavid Howells */ 7929875cf80SDavid Howells static int follow_managed(struct path *path, unsigned flags) 7939875cf80SDavid Howells { 7948aef1884SAl Viro struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */ 7959875cf80SDavid Howells unsigned managed; 7969875cf80SDavid Howells bool need_mntput = false; 7978aef1884SAl Viro int ret = 0; 7989875cf80SDavid Howells 7999875cf80SDavid Howells /* Given that we're not holding a lock here, we retain the value in a 8009875cf80SDavid Howells * local variable for each dentry as we look at it so that we don't see 8019875cf80SDavid Howells * the components of that value change under us */ 8029875cf80SDavid Howells while (managed = ACCESS_ONCE(path->dentry->d_flags), 8039875cf80SDavid Howells managed &= DCACHE_MANAGED_DENTRY, 8049875cf80SDavid Howells unlikely(managed != 0)) { 805cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 806cc53ce53SDavid Howells * being held. */ 807cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 808cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 809cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 8101aed3e42SAl Viro ret = path->dentry->d_op->d_manage(path->dentry, false); 811cc53ce53SDavid Howells if (ret < 0) 8128aef1884SAl Viro break; 813cc53ce53SDavid Howells } 814cc53ce53SDavid Howells 8159875cf80SDavid Howells /* Transit to a mounted filesystem. */ 8169875cf80SDavid Howells if (managed & DCACHE_MOUNTED) { 8179875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 8189875cf80SDavid Howells if (mounted) { 8199875cf80SDavid Howells dput(path->dentry); 8209875cf80SDavid Howells if (need_mntput) 821463ffb2eSAl Viro mntput(path->mnt); 822463ffb2eSAl Viro path->mnt = mounted; 823463ffb2eSAl Viro path->dentry = dget(mounted->mnt_root); 8249875cf80SDavid Howells need_mntput = true; 8259875cf80SDavid Howells continue; 826463ffb2eSAl Viro } 827463ffb2eSAl Viro 8289875cf80SDavid Howells /* Something is mounted on this dentry in another 8299875cf80SDavid Howells * namespace and/or whatever was mounted there in this 8309875cf80SDavid Howells * namespace got unmounted before we managed to get the 8319875cf80SDavid Howells * vfsmount_lock */ 8321da177e4SLinus Torvalds } 8339875cf80SDavid Howells 8349875cf80SDavid Howells /* Handle an automount point */ 8359875cf80SDavid Howells if (managed & DCACHE_NEED_AUTOMOUNT) { 8369875cf80SDavid Howells ret = follow_automount(path, flags, &need_mntput); 8379875cf80SDavid Howells if (ret < 0) 8388aef1884SAl Viro break; 8399875cf80SDavid Howells continue; 8409875cf80SDavid Howells } 8419875cf80SDavid Howells 8429875cf80SDavid Howells /* We didn't change the current path point */ 8439875cf80SDavid Howells break; 8449875cf80SDavid Howells } 8458aef1884SAl Viro 8468aef1884SAl Viro if (need_mntput && path->mnt == mnt) 8478aef1884SAl Viro mntput(path->mnt); 8488aef1884SAl Viro if (ret == -EISDIR) 8498aef1884SAl Viro ret = 0; 850a3fbbde7SAl Viro return ret < 0 ? ret : need_mntput; 8511da177e4SLinus Torvalds } 8521da177e4SLinus Torvalds 853cc53ce53SDavid Howells int follow_down_one(struct path *path) 8541da177e4SLinus Torvalds { 8551da177e4SLinus Torvalds struct vfsmount *mounted; 8561da177e4SLinus Torvalds 8571c755af4SAl Viro mounted = lookup_mnt(path); 8581da177e4SLinus Torvalds if (mounted) { 8599393bd07SAl Viro dput(path->dentry); 8609393bd07SAl Viro mntput(path->mnt); 8619393bd07SAl Viro path->mnt = mounted; 8629393bd07SAl Viro path->dentry = dget(mounted->mnt_root); 8631da177e4SLinus Torvalds return 1; 8641da177e4SLinus Torvalds } 8651da177e4SLinus Torvalds return 0; 8661da177e4SLinus Torvalds } 8671da177e4SLinus Torvalds 86862a7375eSIan Kent static inline bool managed_dentry_might_block(struct dentry *dentry) 86962a7375eSIan Kent { 87062a7375eSIan Kent return (dentry->d_flags & DCACHE_MANAGE_TRANSIT && 87162a7375eSIan Kent dentry->d_op->d_manage(dentry, true) < 0); 87262a7375eSIan Kent } 87362a7375eSIan Kent 8749875cf80SDavid Howells /* 875287548e4SAl Viro * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if 876287548e4SAl Viro * we meet a managed dentry that would need blocking. 8779875cf80SDavid Howells */ 8789875cf80SDavid Howells static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, 879287548e4SAl Viro struct inode **inode) 8809875cf80SDavid Howells { 88162a7375eSIan Kent for (;;) { 882c7105365SAl Viro struct mount *mounted; 88362a7375eSIan Kent /* 88462a7375eSIan Kent * Don't forget we might have a non-mountpoint managed dentry 88562a7375eSIan Kent * that wants to block transit. 88662a7375eSIan Kent */ 887287548e4SAl Viro if (unlikely(managed_dentry_might_block(path->dentry))) 888ab90911fSDavid Howells return false; 88962a7375eSIan Kent 89062a7375eSIan Kent if (!d_mountpoint(path->dentry)) 89162a7375eSIan Kent break; 89262a7375eSIan Kent 8939875cf80SDavid Howells mounted = __lookup_mnt(path->mnt, path->dentry, 1); 8949875cf80SDavid Howells if (!mounted) 8959875cf80SDavid Howells break; 896c7105365SAl Viro path->mnt = &mounted->mnt; 897c7105365SAl Viro path->dentry = mounted->mnt.mnt_root; 898a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 8999875cf80SDavid Howells nd->seq = read_seqcount_begin(&path->dentry->d_seq); 90059430262SLinus Torvalds /* 90159430262SLinus Torvalds * Update the inode too. We don't need to re-check the 90259430262SLinus Torvalds * dentry sequence number here after this d_inode read, 90359430262SLinus Torvalds * because a mount-point is always pinned. 90459430262SLinus Torvalds */ 90559430262SLinus Torvalds *inode = path->dentry->d_inode; 9069875cf80SDavid Howells } 9079875cf80SDavid Howells return true; 9089875cf80SDavid Howells } 9099875cf80SDavid Howells 910dea39376SAl Viro static void follow_mount_rcu(struct nameidata *nd) 911287548e4SAl Viro { 912dea39376SAl Viro while (d_mountpoint(nd->path.dentry)) { 913c7105365SAl Viro struct mount *mounted; 914dea39376SAl Viro mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1); 915287548e4SAl Viro if (!mounted) 916287548e4SAl Viro break; 917c7105365SAl Viro nd->path.mnt = &mounted->mnt; 918c7105365SAl Viro nd->path.dentry = mounted->mnt.mnt_root; 919dea39376SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 920287548e4SAl Viro } 921287548e4SAl Viro } 922287548e4SAl Viro 92331e6b01fSNick Piggin static int follow_dotdot_rcu(struct nameidata *nd) 92431e6b01fSNick Piggin { 92531e6b01fSNick Piggin set_root_rcu(nd); 92631e6b01fSNick Piggin 92731e6b01fSNick Piggin while (1) { 92831e6b01fSNick Piggin if (nd->path.dentry == nd->root.dentry && 92931e6b01fSNick Piggin nd->path.mnt == nd->root.mnt) { 93031e6b01fSNick Piggin break; 93131e6b01fSNick Piggin } 93231e6b01fSNick Piggin if (nd->path.dentry != nd->path.mnt->mnt_root) { 93331e6b01fSNick Piggin struct dentry *old = nd->path.dentry; 93431e6b01fSNick Piggin struct dentry *parent = old->d_parent; 93531e6b01fSNick Piggin unsigned seq; 93631e6b01fSNick Piggin 93731e6b01fSNick Piggin seq = read_seqcount_begin(&parent->d_seq); 93831e6b01fSNick Piggin if (read_seqcount_retry(&old->d_seq, nd->seq)) 939ef7562d5SAl Viro goto failed; 94031e6b01fSNick Piggin nd->path.dentry = parent; 94131e6b01fSNick Piggin nd->seq = seq; 94231e6b01fSNick Piggin break; 94331e6b01fSNick Piggin } 94431e6b01fSNick Piggin if (!follow_up_rcu(&nd->path)) 94531e6b01fSNick Piggin break; 94631e6b01fSNick Piggin nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 94731e6b01fSNick Piggin } 948dea39376SAl Viro follow_mount_rcu(nd); 949dea39376SAl Viro nd->inode = nd->path.dentry->d_inode; 95031e6b01fSNick Piggin return 0; 951ef7562d5SAl Viro 952ef7562d5SAl Viro failed: 953ef7562d5SAl Viro nd->flags &= ~LOOKUP_RCU; 9545b6ca027SAl Viro if (!(nd->flags & LOOKUP_ROOT)) 955ef7562d5SAl Viro nd->root.mnt = NULL; 956ef7562d5SAl Viro rcu_read_unlock(); 957962830dfSAndi Kleen br_read_unlock(&vfsmount_lock); 958ef7562d5SAl Viro return -ECHILD; 95931e6b01fSNick Piggin } 96031e6b01fSNick Piggin 9619875cf80SDavid Howells /* 962cc53ce53SDavid Howells * Follow down to the covering mount currently visible to userspace. At each 963cc53ce53SDavid Howells * point, the filesystem owning that dentry may be queried as to whether the 964cc53ce53SDavid Howells * caller is permitted to proceed or not. 965cc53ce53SDavid Howells */ 9667cc90cc3SAl Viro int follow_down(struct path *path) 967cc53ce53SDavid Howells { 968cc53ce53SDavid Howells unsigned managed; 969cc53ce53SDavid Howells int ret; 970cc53ce53SDavid Howells 971cc53ce53SDavid Howells while (managed = ACCESS_ONCE(path->dentry->d_flags), 972cc53ce53SDavid Howells unlikely(managed & DCACHE_MANAGED_DENTRY)) { 973cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 974cc53ce53SDavid Howells * being held. 975cc53ce53SDavid Howells * 976cc53ce53SDavid Howells * We indicate to the filesystem if someone is trying to mount 977cc53ce53SDavid Howells * something here. This gives autofs the chance to deny anyone 978cc53ce53SDavid Howells * other than its daemon the right to mount on its 979cc53ce53SDavid Howells * superstructure. 980cc53ce53SDavid Howells * 981cc53ce53SDavid Howells * The filesystem may sleep at this point. 982cc53ce53SDavid Howells */ 983cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 984cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 985cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 986ab90911fSDavid Howells ret = path->dentry->d_op->d_manage( 9871aed3e42SAl Viro path->dentry, false); 988cc53ce53SDavid Howells if (ret < 0) 989cc53ce53SDavid Howells return ret == -EISDIR ? 0 : ret; 990cc53ce53SDavid Howells } 991cc53ce53SDavid Howells 992cc53ce53SDavid Howells /* Transit to a mounted filesystem. */ 993cc53ce53SDavid Howells if (managed & DCACHE_MOUNTED) { 994cc53ce53SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 995cc53ce53SDavid Howells if (!mounted) 996cc53ce53SDavid Howells break; 997cc53ce53SDavid Howells dput(path->dentry); 998cc53ce53SDavid Howells mntput(path->mnt); 999cc53ce53SDavid Howells path->mnt = mounted; 1000cc53ce53SDavid Howells path->dentry = dget(mounted->mnt_root); 1001cc53ce53SDavid Howells continue; 1002cc53ce53SDavid Howells } 1003cc53ce53SDavid Howells 1004cc53ce53SDavid Howells /* Don't handle automount points here */ 1005cc53ce53SDavid Howells break; 1006cc53ce53SDavid Howells } 1007cc53ce53SDavid Howells return 0; 1008cc53ce53SDavid Howells } 1009cc53ce53SDavid Howells 1010cc53ce53SDavid Howells /* 10119875cf80SDavid Howells * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() 10129875cf80SDavid Howells */ 10139875cf80SDavid Howells static void follow_mount(struct path *path) 10149875cf80SDavid Howells { 10159875cf80SDavid Howells while (d_mountpoint(path->dentry)) { 10169875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 10179875cf80SDavid Howells if (!mounted) 10189875cf80SDavid Howells break; 10199875cf80SDavid Howells dput(path->dentry); 10209875cf80SDavid Howells mntput(path->mnt); 10219875cf80SDavid Howells path->mnt = mounted; 10229875cf80SDavid Howells path->dentry = dget(mounted->mnt_root); 10239875cf80SDavid Howells } 10249875cf80SDavid Howells } 10259875cf80SDavid Howells 102631e6b01fSNick Piggin static void follow_dotdot(struct nameidata *nd) 10271da177e4SLinus Torvalds { 10282a737871SAl Viro set_root(nd); 1029e518ddb7SAndreas Mohr 10301da177e4SLinus Torvalds while(1) { 10314ac91378SJan Blunck struct dentry *old = nd->path.dentry; 10321da177e4SLinus Torvalds 10332a737871SAl Viro if (nd->path.dentry == nd->root.dentry && 10342a737871SAl Viro nd->path.mnt == nd->root.mnt) { 10351da177e4SLinus Torvalds break; 10361da177e4SLinus Torvalds } 10374ac91378SJan Blunck if (nd->path.dentry != nd->path.mnt->mnt_root) { 10383088dd70SAl Viro /* rare case of legitimate dget_parent()... */ 10393088dd70SAl Viro nd->path.dentry = dget_parent(nd->path.dentry); 10401da177e4SLinus Torvalds dput(old); 10411da177e4SLinus Torvalds break; 10421da177e4SLinus Torvalds } 10433088dd70SAl Viro if (!follow_up(&nd->path)) 10441da177e4SLinus Torvalds break; 10451da177e4SLinus Torvalds } 104679ed0226SAl Viro follow_mount(&nd->path); 104731e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 10481da177e4SLinus Torvalds } 10491da177e4SLinus Torvalds 10501da177e4SLinus Torvalds /* 1051bad61189SMiklos Szeredi * This looks up the name in dcache, possibly revalidates the old dentry and 1052bad61189SMiklos Szeredi * allocates a new one if not found or not valid. In the need_lookup argument 1053bad61189SMiklos Szeredi * returns whether i_op->lookup is necessary. 1054bad61189SMiklos Szeredi * 1055bad61189SMiklos Szeredi * dir->d_inode->i_mutex must be held 1056baa03890SNick Piggin */ 1057bad61189SMiklos Szeredi static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir, 1058201f956eSAl Viro unsigned int flags, bool *need_lookup) 1059baa03890SNick Piggin { 1060baa03890SNick Piggin struct dentry *dentry; 1061bad61189SMiklos Szeredi int error; 1062baa03890SNick Piggin 1063bad61189SMiklos Szeredi *need_lookup = false; 1064bad61189SMiklos Szeredi dentry = d_lookup(dir, name); 1065bad61189SMiklos Szeredi if (dentry) { 1066bad61189SMiklos Szeredi if (d_need_lookup(dentry)) { 1067bad61189SMiklos Szeredi *need_lookup = true; 1068bad61189SMiklos Szeredi } else if (dentry->d_flags & DCACHE_OP_REVALIDATE) { 1069201f956eSAl Viro error = d_revalidate(dentry, flags); 1070bad61189SMiklos Szeredi if (unlikely(error <= 0)) { 1071bad61189SMiklos Szeredi if (error < 0) { 1072bad61189SMiklos Szeredi dput(dentry); 1073bad61189SMiklos Szeredi return ERR_PTR(error); 1074bad61189SMiklos Szeredi } else if (!d_invalidate(dentry)) { 1075bad61189SMiklos Szeredi dput(dentry); 1076bad61189SMiklos Szeredi dentry = NULL; 1077bad61189SMiklos Szeredi } 1078bad61189SMiklos Szeredi } 1079bad61189SMiklos Szeredi } 1080bad61189SMiklos Szeredi } 1081baa03890SNick Piggin 1082bad61189SMiklos Szeredi if (!dentry) { 1083bad61189SMiklos Szeredi dentry = d_alloc(dir, name); 1084baa03890SNick Piggin if (unlikely(!dentry)) 1085baa03890SNick Piggin return ERR_PTR(-ENOMEM); 1086baa03890SNick Piggin 1087bad61189SMiklos Szeredi *need_lookup = true; 1088baa03890SNick Piggin } 1089baa03890SNick Piggin return dentry; 1090baa03890SNick Piggin } 1091baa03890SNick Piggin 1092baa03890SNick Piggin /* 1093bad61189SMiklos Szeredi * Call i_op->lookup on the dentry. The dentry must be negative but may be 1094bad61189SMiklos Szeredi * hashed if it was pouplated with DCACHE_NEED_LOOKUP. 1095bad61189SMiklos Szeredi * 1096bad61189SMiklos Szeredi * dir->d_inode->i_mutex must be held 109744396f4bSJosef Bacik */ 1098bad61189SMiklos Szeredi static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry, 109972bd866aSAl Viro unsigned int flags) 110044396f4bSJosef Bacik { 110144396f4bSJosef Bacik struct dentry *old; 110244396f4bSJosef Bacik 110344396f4bSJosef Bacik /* Don't create child dentry for a dead directory. */ 1104bad61189SMiklos Szeredi if (unlikely(IS_DEADDIR(dir))) { 1105e188dc02SMiklos Szeredi dput(dentry); 110644396f4bSJosef Bacik return ERR_PTR(-ENOENT); 1107e188dc02SMiklos Szeredi } 110844396f4bSJosef Bacik 110972bd866aSAl Viro old = dir->i_op->lookup(dir, dentry, flags); 111044396f4bSJosef Bacik if (unlikely(old)) { 111144396f4bSJosef Bacik dput(dentry); 111244396f4bSJosef Bacik dentry = old; 111344396f4bSJosef Bacik } 111444396f4bSJosef Bacik return dentry; 111544396f4bSJosef Bacik } 111644396f4bSJosef Bacik 1117a3255546SAl Viro static struct dentry *__lookup_hash(struct qstr *name, 111872bd866aSAl Viro struct dentry *base, unsigned int flags) 1119a3255546SAl Viro { 1120bad61189SMiklos Szeredi bool need_lookup; 1121a3255546SAl Viro struct dentry *dentry; 1122a3255546SAl Viro 112372bd866aSAl Viro dentry = lookup_dcache(name, base, flags, &need_lookup); 1124bad61189SMiklos Szeredi if (!need_lookup) 1125a3255546SAl Viro return dentry; 1126bad61189SMiklos Szeredi 112772bd866aSAl Viro return lookup_real(base->d_inode, dentry, flags); 1128a3255546SAl Viro } 1129a3255546SAl Viro 113044396f4bSJosef Bacik /* 11311da177e4SLinus Torvalds * It's more convoluted than I'd like it to be, but... it's still fairly 11321da177e4SLinus Torvalds * small and for now I'd prefer to have fast path as straight as possible. 11331da177e4SLinus Torvalds * It _is_ time-critical. 11341da177e4SLinus Torvalds */ 1135697f514dSMiklos Szeredi static int lookup_fast(struct nameidata *nd, struct qstr *name, 113631e6b01fSNick Piggin struct path *path, struct inode **inode) 11371da177e4SLinus Torvalds { 11384ac91378SJan Blunck struct vfsmount *mnt = nd->path.mnt; 113931e6b01fSNick Piggin struct dentry *dentry, *parent = nd->path.dentry; 11405a18fff2SAl Viro int need_reval = 1; 11415a18fff2SAl Viro int status = 1; 11429875cf80SDavid Howells int err; 11439875cf80SDavid Howells 11443cac260aSAl Viro /* 1145b04f784eSNick Piggin * Rename seqlock is not required here because in the off chance 1146b04f784eSNick Piggin * of a false negative due to a concurrent rename, we're going to 1147b04f784eSNick Piggin * do the non-racy lookup, below. 1148b04f784eSNick Piggin */ 114931e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) { 115031e6b01fSNick Piggin unsigned seq; 115112f8ad4bSLinus Torvalds dentry = __d_lookup_rcu(parent, name, &seq, nd->inode); 11525a18fff2SAl Viro if (!dentry) 11535a18fff2SAl Viro goto unlazy; 11545a18fff2SAl Viro 115512f8ad4bSLinus Torvalds /* 115612f8ad4bSLinus Torvalds * This sequence count validates that the inode matches 115712f8ad4bSLinus Torvalds * the dentry name information from lookup. 115812f8ad4bSLinus Torvalds */ 115912f8ad4bSLinus Torvalds *inode = dentry->d_inode; 116012f8ad4bSLinus Torvalds if (read_seqcount_retry(&dentry->d_seq, seq)) 116112f8ad4bSLinus Torvalds return -ECHILD; 116212f8ad4bSLinus Torvalds 116312f8ad4bSLinus Torvalds /* 116412f8ad4bSLinus Torvalds * This sequence count validates that the parent had no 116512f8ad4bSLinus Torvalds * changes while we did the lookup of the dentry above. 116612f8ad4bSLinus Torvalds * 116712f8ad4bSLinus Torvalds * The memory barrier in read_seqcount_begin of child is 116812f8ad4bSLinus Torvalds * enough, we can use __read_seqcount_retry here. 116912f8ad4bSLinus Torvalds */ 117031e6b01fSNick Piggin if (__read_seqcount_retry(&parent->d_seq, nd->seq)) 117131e6b01fSNick Piggin return -ECHILD; 117231e6b01fSNick Piggin nd->seq = seq; 11735a18fff2SAl Viro 1174fa4ee159SMiklos Szeredi if (unlikely(d_need_lookup(dentry))) 1175fa4ee159SMiklos Szeredi goto unlazy; 117624643087SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) { 11774ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 11785a18fff2SAl Viro if (unlikely(status <= 0)) { 11795a18fff2SAl Viro if (status != -ECHILD) 11805a18fff2SAl Viro need_reval = 0; 11815a18fff2SAl Viro goto unlazy; 11825a18fff2SAl Viro } 118324643087SAl Viro } 118431e6b01fSNick Piggin path->mnt = mnt; 118531e6b01fSNick Piggin path->dentry = dentry; 1186d6e9bd25SAl Viro if (unlikely(!__follow_mount_rcu(nd, path, inode))) 1187d6e9bd25SAl Viro goto unlazy; 1188d6e9bd25SAl Viro if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT)) 1189d6e9bd25SAl Viro goto unlazy; 11909875cf80SDavid Howells return 0; 11915a18fff2SAl Viro unlazy: 119219660af7SAl Viro if (unlazy_walk(nd, dentry)) 11935a18fff2SAl Viro return -ECHILD; 11945a18fff2SAl Viro } else { 119531e6b01fSNick Piggin dentry = __d_lookup(parent, name); 119624643087SAl Viro } 11975a18fff2SAl Viro 119881e6f520SAl Viro if (unlikely(!dentry)) 119981e6f520SAl Viro goto need_lookup; 12005a18fff2SAl Viro 120181e6f520SAl Viro if (unlikely(d_need_lookup(dentry))) { 120281e6f520SAl Viro dput(dentry); 120381e6f520SAl Viro goto need_lookup; 12045a18fff2SAl Viro } 120581e6f520SAl Viro 12065a18fff2SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval) 12074ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 12085a18fff2SAl Viro if (unlikely(status <= 0)) { 12095a18fff2SAl Viro if (status < 0) { 12105a18fff2SAl Viro dput(dentry); 12115a18fff2SAl Viro return status; 12125a18fff2SAl Viro } 12135a18fff2SAl Viro if (!d_invalidate(dentry)) { 12145a18fff2SAl Viro dput(dentry); 121581e6f520SAl Viro goto need_lookup; 12165a18fff2SAl Viro } 12175a18fff2SAl Viro } 1218697f514dSMiklos Szeredi 12191da177e4SLinus Torvalds path->mnt = mnt; 12201da177e4SLinus Torvalds path->dentry = dentry; 12219875cf80SDavid Howells err = follow_managed(path, nd->flags); 122289312214SIan Kent if (unlikely(err < 0)) { 122389312214SIan Kent path_put_conditional(path, nd); 12249875cf80SDavid Howells return err; 122589312214SIan Kent } 1226a3fbbde7SAl Viro if (err) 1227a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 122831e6b01fSNick Piggin *inode = path->dentry->d_inode; 12291da177e4SLinus Torvalds return 0; 123081e6f520SAl Viro 123181e6f520SAl Viro need_lookup: 1232697f514dSMiklos Szeredi return 1; 1233697f514dSMiklos Szeredi } 1234697f514dSMiklos Szeredi 1235697f514dSMiklos Szeredi /* Fast lookup failed, do it the slow way */ 1236697f514dSMiklos Szeredi static int lookup_slow(struct nameidata *nd, struct qstr *name, 1237697f514dSMiklos Szeredi struct path *path) 1238697f514dSMiklos Szeredi { 1239697f514dSMiklos Szeredi struct dentry *dentry, *parent; 1240697f514dSMiklos Szeredi int err; 1241697f514dSMiklos Szeredi 1242697f514dSMiklos Szeredi parent = nd->path.dentry; 124381e6f520SAl Viro BUG_ON(nd->inode != parent->d_inode); 124481e6f520SAl Viro 124581e6f520SAl Viro mutex_lock(&parent->d_inode->i_mutex); 124672bd866aSAl Viro dentry = __lookup_hash(name, parent, nd->flags); 124781e6f520SAl Viro mutex_unlock(&parent->d_inode->i_mutex); 124881e6f520SAl Viro if (IS_ERR(dentry)) 124981e6f520SAl Viro return PTR_ERR(dentry); 1250697f514dSMiklos Szeredi path->mnt = nd->path.mnt; 1251697f514dSMiklos Szeredi path->dentry = dentry; 1252697f514dSMiklos Szeredi err = follow_managed(path, nd->flags); 1253697f514dSMiklos Szeredi if (unlikely(err < 0)) { 1254697f514dSMiklos Szeredi path_put_conditional(path, nd); 1255697f514dSMiklos Szeredi return err; 1256697f514dSMiklos Szeredi } 1257697f514dSMiklos Szeredi if (err) 1258697f514dSMiklos Szeredi nd->flags |= LOOKUP_JUMPED; 1259697f514dSMiklos Szeredi return 0; 12601da177e4SLinus Torvalds } 12611da177e4SLinus Torvalds 126252094c8aSAl Viro static inline int may_lookup(struct nameidata *nd) 126352094c8aSAl Viro { 126452094c8aSAl Viro if (nd->flags & LOOKUP_RCU) { 12654ad5abb3SAl Viro int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); 126652094c8aSAl Viro if (err != -ECHILD) 126752094c8aSAl Viro return err; 126819660af7SAl Viro if (unlazy_walk(nd, NULL)) 126952094c8aSAl Viro return -ECHILD; 127052094c8aSAl Viro } 12714ad5abb3SAl Viro return inode_permission(nd->inode, MAY_EXEC); 127252094c8aSAl Viro } 127352094c8aSAl Viro 12749856fa1bSAl Viro static inline int handle_dots(struct nameidata *nd, int type) 12759856fa1bSAl Viro { 12769856fa1bSAl Viro if (type == LAST_DOTDOT) { 12779856fa1bSAl Viro if (nd->flags & LOOKUP_RCU) { 12789856fa1bSAl Viro if (follow_dotdot_rcu(nd)) 12799856fa1bSAl Viro return -ECHILD; 12809856fa1bSAl Viro } else 12819856fa1bSAl Viro follow_dotdot(nd); 12829856fa1bSAl Viro } 12839856fa1bSAl Viro return 0; 12849856fa1bSAl Viro } 12859856fa1bSAl Viro 1286951361f9SAl Viro static void terminate_walk(struct nameidata *nd) 1287951361f9SAl Viro { 1288951361f9SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1289951361f9SAl Viro path_put(&nd->path); 1290951361f9SAl Viro } else { 1291951361f9SAl Viro nd->flags &= ~LOOKUP_RCU; 12925b6ca027SAl Viro if (!(nd->flags & LOOKUP_ROOT)) 1293951361f9SAl Viro nd->root.mnt = NULL; 1294951361f9SAl Viro rcu_read_unlock(); 1295962830dfSAndi Kleen br_read_unlock(&vfsmount_lock); 1296951361f9SAl Viro } 1297951361f9SAl Viro } 1298951361f9SAl Viro 12993ddcd056SLinus Torvalds /* 13003ddcd056SLinus Torvalds * Do we need to follow links? We _really_ want to be able 13013ddcd056SLinus Torvalds * to do this check without having to look at inode->i_op, 13023ddcd056SLinus Torvalds * so we keep a cache of "no, this doesn't need follow_link" 13033ddcd056SLinus Torvalds * for the common case. 13043ddcd056SLinus Torvalds */ 13057813b94aSLinus Torvalds static inline int should_follow_link(struct inode *inode, int follow) 13063ddcd056SLinus Torvalds { 13073ddcd056SLinus Torvalds if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) { 13083ddcd056SLinus Torvalds if (likely(inode->i_op->follow_link)) 13093ddcd056SLinus Torvalds return follow; 13103ddcd056SLinus Torvalds 13113ddcd056SLinus Torvalds /* This gets set once for the inode lifetime */ 13123ddcd056SLinus Torvalds spin_lock(&inode->i_lock); 13133ddcd056SLinus Torvalds inode->i_opflags |= IOP_NOFOLLOW; 13143ddcd056SLinus Torvalds spin_unlock(&inode->i_lock); 13153ddcd056SLinus Torvalds } 13163ddcd056SLinus Torvalds return 0; 13173ddcd056SLinus Torvalds } 13183ddcd056SLinus Torvalds 1319ce57dfc1SAl Viro static inline int walk_component(struct nameidata *nd, struct path *path, 1320ce57dfc1SAl Viro struct qstr *name, int type, int follow) 1321ce57dfc1SAl Viro { 1322ce57dfc1SAl Viro struct inode *inode; 1323ce57dfc1SAl Viro int err; 1324ce57dfc1SAl Viro /* 1325ce57dfc1SAl Viro * "." and ".." are special - ".." especially so because it has 1326ce57dfc1SAl Viro * to be able to know about the current root directory and 1327ce57dfc1SAl Viro * parent relationships. 1328ce57dfc1SAl Viro */ 1329ce57dfc1SAl Viro if (unlikely(type != LAST_NORM)) 1330ce57dfc1SAl Viro return handle_dots(nd, type); 1331697f514dSMiklos Szeredi err = lookup_fast(nd, name, path, &inode); 1332ce57dfc1SAl Viro if (unlikely(err)) { 1333697f514dSMiklos Szeredi if (err < 0) 1334697f514dSMiklos Szeredi goto out_err; 1335697f514dSMiklos Szeredi 1336697f514dSMiklos Szeredi err = lookup_slow(nd, name, path); 1337697f514dSMiklos Szeredi if (err < 0) 1338697f514dSMiklos Szeredi goto out_err; 1339697f514dSMiklos Szeredi 1340697f514dSMiklos Szeredi inode = path->dentry->d_inode; 1341ce57dfc1SAl Viro } 1342697f514dSMiklos Szeredi err = -ENOENT; 1343697f514dSMiklos Szeredi if (!inode) 1344697f514dSMiklos Szeredi goto out_path_put; 1345697f514dSMiklos Szeredi 13467813b94aSLinus Torvalds if (should_follow_link(inode, follow)) { 134719660af7SAl Viro if (nd->flags & LOOKUP_RCU) { 134819660af7SAl Viro if (unlikely(unlazy_walk(nd, path->dentry))) { 1349697f514dSMiklos Szeredi err = -ECHILD; 1350697f514dSMiklos Szeredi goto out_err; 135119660af7SAl Viro } 135219660af7SAl Viro } 1353ce57dfc1SAl Viro BUG_ON(inode != path->dentry->d_inode); 1354ce57dfc1SAl Viro return 1; 1355ce57dfc1SAl Viro } 1356ce57dfc1SAl Viro path_to_nameidata(path, nd); 1357ce57dfc1SAl Viro nd->inode = inode; 1358ce57dfc1SAl Viro return 0; 1359697f514dSMiklos Szeredi 1360697f514dSMiklos Szeredi out_path_put: 1361697f514dSMiklos Szeredi path_to_nameidata(path, nd); 1362697f514dSMiklos Szeredi out_err: 1363697f514dSMiklos Szeredi terminate_walk(nd); 1364697f514dSMiklos Szeredi return err; 1365ce57dfc1SAl Viro } 1366ce57dfc1SAl Viro 13671da177e4SLinus Torvalds /* 1368b356379aSAl Viro * This limits recursive symlink follows to 8, while 1369b356379aSAl Viro * limiting consecutive symlinks to 40. 1370b356379aSAl Viro * 1371b356379aSAl Viro * Without that kind of total limit, nasty chains of consecutive 1372b356379aSAl Viro * symlinks can cause almost arbitrarily long lookups. 1373b356379aSAl Viro */ 1374b356379aSAl Viro static inline int nested_symlink(struct path *path, struct nameidata *nd) 1375b356379aSAl Viro { 1376b356379aSAl Viro int res; 1377b356379aSAl Viro 1378b356379aSAl Viro if (unlikely(current->link_count >= MAX_NESTED_LINKS)) { 1379b356379aSAl Viro path_put_conditional(path, nd); 1380b356379aSAl Viro path_put(&nd->path); 1381b356379aSAl Viro return -ELOOP; 1382b356379aSAl Viro } 13831a4022f8SErez Zadok BUG_ON(nd->depth >= MAX_NESTED_LINKS); 1384b356379aSAl Viro 1385b356379aSAl Viro nd->depth++; 1386b356379aSAl Viro current->link_count++; 1387b356379aSAl Viro 1388b356379aSAl Viro do { 1389b356379aSAl Viro struct path link = *path; 1390b356379aSAl Viro void *cookie; 1391574197e0SAl Viro 1392574197e0SAl Viro res = follow_link(&link, nd, &cookie); 13936d7b5aaeSAl Viro if (res) 13946d7b5aaeSAl Viro break; 1395b356379aSAl Viro res = walk_component(nd, path, &nd->last, 1396b356379aSAl Viro nd->last_type, LOOKUP_FOLLOW); 1397574197e0SAl Viro put_link(nd, &link, cookie); 1398b356379aSAl Viro } while (res > 0); 1399b356379aSAl Viro 1400b356379aSAl Viro current->link_count--; 1401b356379aSAl Viro nd->depth--; 1402b356379aSAl Viro return res; 1403b356379aSAl Viro } 1404b356379aSAl Viro 1405b356379aSAl Viro /* 14063ddcd056SLinus Torvalds * We really don't want to look at inode->i_op->lookup 14073ddcd056SLinus Torvalds * when we don't have to. So we keep a cache bit in 14083ddcd056SLinus Torvalds * the inode ->i_opflags field that says "yes, we can 14093ddcd056SLinus Torvalds * do lookup on this inode". 14103ddcd056SLinus Torvalds */ 14113ddcd056SLinus Torvalds static inline int can_lookup(struct inode *inode) 14123ddcd056SLinus Torvalds { 14133ddcd056SLinus Torvalds if (likely(inode->i_opflags & IOP_LOOKUP)) 14143ddcd056SLinus Torvalds return 1; 14153ddcd056SLinus Torvalds if (likely(!inode->i_op->lookup)) 14163ddcd056SLinus Torvalds return 0; 14173ddcd056SLinus Torvalds 14183ddcd056SLinus Torvalds /* We do this once for the lifetime of the inode */ 14193ddcd056SLinus Torvalds spin_lock(&inode->i_lock); 14203ddcd056SLinus Torvalds inode->i_opflags |= IOP_LOOKUP; 14213ddcd056SLinus Torvalds spin_unlock(&inode->i_lock); 14223ddcd056SLinus Torvalds return 1; 14233ddcd056SLinus Torvalds } 14243ddcd056SLinus Torvalds 1425bfcfaa77SLinus Torvalds /* 1426bfcfaa77SLinus Torvalds * We can do the critical dentry name comparison and hashing 1427bfcfaa77SLinus Torvalds * operations one word at a time, but we are limited to: 1428bfcfaa77SLinus Torvalds * 1429bfcfaa77SLinus Torvalds * - Architectures with fast unaligned word accesses. We could 1430bfcfaa77SLinus Torvalds * do a "get_unaligned()" if this helps and is sufficiently 1431bfcfaa77SLinus Torvalds * fast. 1432bfcfaa77SLinus Torvalds * 1433bfcfaa77SLinus Torvalds * - Little-endian machines (so that we can generate the mask 1434bfcfaa77SLinus Torvalds * of low bytes efficiently). Again, we *could* do a byte 1435bfcfaa77SLinus Torvalds * swapping load on big-endian architectures if that is not 1436bfcfaa77SLinus Torvalds * expensive enough to make the optimization worthless. 1437bfcfaa77SLinus Torvalds * 1438bfcfaa77SLinus Torvalds * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we 1439bfcfaa77SLinus Torvalds * do not trap on the (extremely unlikely) case of a page 1440bfcfaa77SLinus Torvalds * crossing operation. 1441bfcfaa77SLinus Torvalds * 1442bfcfaa77SLinus Torvalds * - Furthermore, we need an efficient 64-bit compile for the 1443bfcfaa77SLinus Torvalds * 64-bit case in order to generate the "number of bytes in 1444bfcfaa77SLinus Torvalds * the final mask". Again, that could be replaced with a 1445bfcfaa77SLinus Torvalds * efficient population count instruction or similar. 1446bfcfaa77SLinus Torvalds */ 1447bfcfaa77SLinus Torvalds #ifdef CONFIG_DCACHE_WORD_ACCESS 1448bfcfaa77SLinus Torvalds 1449f68e556eSLinus Torvalds #include <asm/word-at-a-time.h> 1450bfcfaa77SLinus Torvalds 1451f68e556eSLinus Torvalds #ifdef CONFIG_64BIT 1452bfcfaa77SLinus Torvalds 1453bfcfaa77SLinus Torvalds static inline unsigned int fold_hash(unsigned long hash) 1454bfcfaa77SLinus Torvalds { 1455bfcfaa77SLinus Torvalds hash += hash >> (8*sizeof(int)); 1456bfcfaa77SLinus Torvalds return hash; 1457bfcfaa77SLinus Torvalds } 1458bfcfaa77SLinus Torvalds 1459bfcfaa77SLinus Torvalds #else /* 32-bit case */ 1460bfcfaa77SLinus Torvalds 1461bfcfaa77SLinus Torvalds #define fold_hash(x) (x) 1462bfcfaa77SLinus Torvalds 1463bfcfaa77SLinus Torvalds #endif 1464bfcfaa77SLinus Torvalds 1465bfcfaa77SLinus Torvalds unsigned int full_name_hash(const unsigned char *name, unsigned int len) 1466bfcfaa77SLinus Torvalds { 1467bfcfaa77SLinus Torvalds unsigned long a, mask; 1468bfcfaa77SLinus Torvalds unsigned long hash = 0; 1469bfcfaa77SLinus Torvalds 1470bfcfaa77SLinus Torvalds for (;;) { 1471e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name); 1472bfcfaa77SLinus Torvalds if (len < sizeof(unsigned long)) 1473bfcfaa77SLinus Torvalds break; 1474bfcfaa77SLinus Torvalds hash += a; 1475f132c5beSAl Viro hash *= 9; 1476bfcfaa77SLinus Torvalds name += sizeof(unsigned long); 1477bfcfaa77SLinus Torvalds len -= sizeof(unsigned long); 1478bfcfaa77SLinus Torvalds if (!len) 1479bfcfaa77SLinus Torvalds goto done; 1480bfcfaa77SLinus Torvalds } 1481bfcfaa77SLinus Torvalds mask = ~(~0ul << len*8); 1482bfcfaa77SLinus Torvalds hash += mask & a; 1483bfcfaa77SLinus Torvalds done: 1484bfcfaa77SLinus Torvalds return fold_hash(hash); 1485bfcfaa77SLinus Torvalds } 1486bfcfaa77SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 1487bfcfaa77SLinus Torvalds 1488bfcfaa77SLinus Torvalds /* 1489bfcfaa77SLinus Torvalds * Calculate the length and hash of the path component, and 1490bfcfaa77SLinus Torvalds * return the length of the component; 1491bfcfaa77SLinus Torvalds */ 1492bfcfaa77SLinus Torvalds static inline unsigned long hash_name(const char *name, unsigned int *hashp) 1493bfcfaa77SLinus Torvalds { 149436126f8fSLinus Torvalds unsigned long a, b, adata, bdata, mask, hash, len; 149536126f8fSLinus Torvalds const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 1496bfcfaa77SLinus Torvalds 1497bfcfaa77SLinus Torvalds hash = a = 0; 1498bfcfaa77SLinus Torvalds len = -sizeof(unsigned long); 1499bfcfaa77SLinus Torvalds do { 1500bfcfaa77SLinus Torvalds hash = (hash + a) * 9; 1501bfcfaa77SLinus Torvalds len += sizeof(unsigned long); 1502e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name+len); 150336126f8fSLinus Torvalds b = a ^ REPEAT_BYTE('/'); 150436126f8fSLinus Torvalds } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); 1505bfcfaa77SLinus Torvalds 150636126f8fSLinus Torvalds adata = prep_zero_mask(a, adata, &constants); 150736126f8fSLinus Torvalds bdata = prep_zero_mask(b, bdata, &constants); 150836126f8fSLinus Torvalds 150936126f8fSLinus Torvalds mask = create_zero_mask(adata | bdata); 151036126f8fSLinus Torvalds 151136126f8fSLinus Torvalds hash += a & zero_bytemask(mask); 1512bfcfaa77SLinus Torvalds *hashp = fold_hash(hash); 1513bfcfaa77SLinus Torvalds 151436126f8fSLinus Torvalds return len + find_zero(mask); 1515bfcfaa77SLinus Torvalds } 1516bfcfaa77SLinus Torvalds 1517bfcfaa77SLinus Torvalds #else 1518bfcfaa77SLinus Torvalds 15190145acc2SLinus Torvalds unsigned int full_name_hash(const unsigned char *name, unsigned int len) 15200145acc2SLinus Torvalds { 15210145acc2SLinus Torvalds unsigned long hash = init_name_hash(); 15220145acc2SLinus Torvalds while (len--) 15230145acc2SLinus Torvalds hash = partial_name_hash(*name++, hash); 15240145acc2SLinus Torvalds return end_name_hash(hash); 15250145acc2SLinus Torvalds } 1526ae942ae7SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 15270145acc2SLinus Torvalds 15283ddcd056SLinus Torvalds /* 1529200e9ef7SLinus Torvalds * We know there's a real path component here of at least 1530200e9ef7SLinus Torvalds * one character. 1531200e9ef7SLinus Torvalds */ 1532200e9ef7SLinus Torvalds static inline unsigned long hash_name(const char *name, unsigned int *hashp) 1533200e9ef7SLinus Torvalds { 1534200e9ef7SLinus Torvalds unsigned long hash = init_name_hash(); 1535200e9ef7SLinus Torvalds unsigned long len = 0, c; 1536200e9ef7SLinus Torvalds 1537200e9ef7SLinus Torvalds c = (unsigned char)*name; 1538200e9ef7SLinus Torvalds do { 1539200e9ef7SLinus Torvalds len++; 1540200e9ef7SLinus Torvalds hash = partial_name_hash(c, hash); 1541200e9ef7SLinus Torvalds c = (unsigned char)name[len]; 1542200e9ef7SLinus Torvalds } while (c && c != '/'); 1543200e9ef7SLinus Torvalds *hashp = end_name_hash(hash); 1544200e9ef7SLinus Torvalds return len; 1545200e9ef7SLinus Torvalds } 1546200e9ef7SLinus Torvalds 1547bfcfaa77SLinus Torvalds #endif 1548bfcfaa77SLinus Torvalds 1549200e9ef7SLinus Torvalds /* 15501da177e4SLinus Torvalds * Name resolution. 1551ea3834d9SPrasanna Meda * This is the basic name resolution function, turning a pathname into 1552ea3834d9SPrasanna Meda * the final dentry. We expect 'base' to be positive and a directory. 15531da177e4SLinus Torvalds * 1554ea3834d9SPrasanna Meda * Returns 0 and nd will have valid dentry and mnt on success. 1555ea3834d9SPrasanna Meda * Returns error and drops reference to input namei data on failure. 15561da177e4SLinus Torvalds */ 15576de88d72SAl Viro static int link_path_walk(const char *name, struct nameidata *nd) 15581da177e4SLinus Torvalds { 15591da177e4SLinus Torvalds struct path next; 15601da177e4SLinus Torvalds int err; 15611da177e4SLinus Torvalds 15621da177e4SLinus Torvalds while (*name=='/') 15631da177e4SLinus Torvalds name++; 15641da177e4SLinus Torvalds if (!*name) 1565086e183aSAl Viro return 0; 15661da177e4SLinus Torvalds 15671da177e4SLinus Torvalds /* At this point we know we have a real path component. */ 15681da177e4SLinus Torvalds for(;;) { 15691da177e4SLinus Torvalds struct qstr this; 1570200e9ef7SLinus Torvalds long len; 1571fe479a58SAl Viro int type; 15721da177e4SLinus Torvalds 157352094c8aSAl Viro err = may_lookup(nd); 15741da177e4SLinus Torvalds if (err) 15751da177e4SLinus Torvalds break; 15761da177e4SLinus Torvalds 1577200e9ef7SLinus Torvalds len = hash_name(name, &this.hash); 15781da177e4SLinus Torvalds this.name = name; 1579200e9ef7SLinus Torvalds this.len = len; 15801da177e4SLinus Torvalds 1581fe479a58SAl Viro type = LAST_NORM; 1582200e9ef7SLinus Torvalds if (name[0] == '.') switch (len) { 1583fe479a58SAl Viro case 2: 1584200e9ef7SLinus Torvalds if (name[1] == '.') { 1585fe479a58SAl Viro type = LAST_DOTDOT; 158616c2cd71SAl Viro nd->flags |= LOOKUP_JUMPED; 158716c2cd71SAl Viro } 1588fe479a58SAl Viro break; 1589fe479a58SAl Viro case 1: 1590fe479a58SAl Viro type = LAST_DOT; 1591fe479a58SAl Viro } 15925a202bcdSAl Viro if (likely(type == LAST_NORM)) { 15935a202bcdSAl Viro struct dentry *parent = nd->path.dentry; 159416c2cd71SAl Viro nd->flags &= ~LOOKUP_JUMPED; 15955a202bcdSAl Viro if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { 15965a202bcdSAl Viro err = parent->d_op->d_hash(parent, nd->inode, 15975a202bcdSAl Viro &this); 15985a202bcdSAl Viro if (err < 0) 15995a202bcdSAl Viro break; 16005a202bcdSAl Viro } 16015a202bcdSAl Viro } 1602fe479a58SAl Viro 1603200e9ef7SLinus Torvalds if (!name[len]) 16041da177e4SLinus Torvalds goto last_component; 1605200e9ef7SLinus Torvalds /* 1606200e9ef7SLinus Torvalds * If it wasn't NUL, we know it was '/'. Skip that 1607200e9ef7SLinus Torvalds * slash, and continue until no more slashes. 1608200e9ef7SLinus Torvalds */ 1609200e9ef7SLinus Torvalds do { 1610200e9ef7SLinus Torvalds len++; 1611200e9ef7SLinus Torvalds } while (unlikely(name[len] == '/')); 1612200e9ef7SLinus Torvalds if (!name[len]) 1613b356379aSAl Viro goto last_component; 1614200e9ef7SLinus Torvalds name += len; 16151da177e4SLinus Torvalds 1616ce57dfc1SAl Viro err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW); 1617ce57dfc1SAl Viro if (err < 0) 1618ce57dfc1SAl Viro return err; 1619fe479a58SAl Viro 1620ce57dfc1SAl Viro if (err) { 1621b356379aSAl Viro err = nested_symlink(&next, nd); 16221da177e4SLinus Torvalds if (err) 1623a7472babSAl Viro return err; 162431e6b01fSNick Piggin } 16253ddcd056SLinus Torvalds if (can_lookup(nd->inode)) 16261da177e4SLinus Torvalds continue; 16273ddcd056SLinus Torvalds err = -ENOTDIR; 16283ddcd056SLinus Torvalds break; 16291da177e4SLinus Torvalds /* here ends the main loop */ 16301da177e4SLinus Torvalds 16311da177e4SLinus Torvalds last_component: 1632ce57dfc1SAl Viro nd->last = this; 1633ce57dfc1SAl Viro nd->last_type = type; 1634ce57dfc1SAl Viro return 0; 1635ce57dfc1SAl Viro } 1636951361f9SAl Viro terminate_walk(nd); 16371da177e4SLinus Torvalds return err; 16381da177e4SLinus Torvalds } 16391da177e4SLinus Torvalds 164070e9b357SAl Viro static int path_init(int dfd, const char *name, unsigned int flags, 164170e9b357SAl Viro struct nameidata *nd, struct file **fp) 164231e6b01fSNick Piggin { 164331e6b01fSNick Piggin int retval = 0; 164431e6b01fSNick Piggin int fput_needed; 164531e6b01fSNick Piggin struct file *file; 164631e6b01fSNick Piggin 164731e6b01fSNick Piggin nd->last_type = LAST_ROOT; /* if there are only slashes... */ 164816c2cd71SAl Viro nd->flags = flags | LOOKUP_JUMPED; 164931e6b01fSNick Piggin nd->depth = 0; 16505b6ca027SAl Viro if (flags & LOOKUP_ROOT) { 16515b6ca027SAl Viro struct inode *inode = nd->root.dentry->d_inode; 165273d049a4SAl Viro if (*name) { 16535b6ca027SAl Viro if (!inode->i_op->lookup) 16545b6ca027SAl Viro return -ENOTDIR; 16555b6ca027SAl Viro retval = inode_permission(inode, MAY_EXEC); 16565b6ca027SAl Viro if (retval) 16575b6ca027SAl Viro return retval; 165873d049a4SAl Viro } 16595b6ca027SAl Viro nd->path = nd->root; 16605b6ca027SAl Viro nd->inode = inode; 16615b6ca027SAl Viro if (flags & LOOKUP_RCU) { 1662962830dfSAndi Kleen br_read_lock(&vfsmount_lock); 16635b6ca027SAl Viro rcu_read_lock(); 16645b6ca027SAl Viro nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 16655b6ca027SAl Viro } else { 16665b6ca027SAl Viro path_get(&nd->path); 16675b6ca027SAl Viro } 16685b6ca027SAl Viro return 0; 16695b6ca027SAl Viro } 16705b6ca027SAl Viro 167131e6b01fSNick Piggin nd->root.mnt = NULL; 167231e6b01fSNick Piggin 167331e6b01fSNick Piggin if (*name=='/') { 1674e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 1675962830dfSAndi Kleen br_read_lock(&vfsmount_lock); 167631e6b01fSNick Piggin rcu_read_lock(); 1677e41f7d4eSAl Viro set_root_rcu(nd); 1678e41f7d4eSAl Viro } else { 1679e41f7d4eSAl Viro set_root(nd); 1680e41f7d4eSAl Viro path_get(&nd->root); 1681e41f7d4eSAl Viro } 168231e6b01fSNick Piggin nd->path = nd->root; 168331e6b01fSNick Piggin } else if (dfd == AT_FDCWD) { 1684e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 168531e6b01fSNick Piggin struct fs_struct *fs = current->fs; 1686c28cc364SNick Piggin unsigned seq; 168731e6b01fSNick Piggin 1688962830dfSAndi Kleen br_read_lock(&vfsmount_lock); 168931e6b01fSNick Piggin rcu_read_lock(); 169031e6b01fSNick Piggin 1691c28cc364SNick Piggin do { 1692c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 169331e6b01fSNick Piggin nd->path = fs->pwd; 1694c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 1695c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 1696e41f7d4eSAl Viro } else { 1697e41f7d4eSAl Viro get_fs_pwd(current->fs, &nd->path); 1698e41f7d4eSAl Viro } 169931e6b01fSNick Piggin } else { 170031e6b01fSNick Piggin struct dentry *dentry; 170131e6b01fSNick Piggin 17021abf0c71SAl Viro file = fget_raw_light(dfd, &fput_needed); 170331e6b01fSNick Piggin retval = -EBADF; 170431e6b01fSNick Piggin if (!file) 170531e6b01fSNick Piggin goto out_fail; 170631e6b01fSNick Piggin 170731e6b01fSNick Piggin dentry = file->f_path.dentry; 170831e6b01fSNick Piggin 1709f52e0c11SAl Viro if (*name) { 171031e6b01fSNick Piggin retval = -ENOTDIR; 171131e6b01fSNick Piggin if (!S_ISDIR(dentry->d_inode->i_mode)) 171231e6b01fSNick Piggin goto fput_fail; 171331e6b01fSNick Piggin 17144ad5abb3SAl Viro retval = inode_permission(dentry->d_inode, MAY_EXEC); 171531e6b01fSNick Piggin if (retval) 171631e6b01fSNick Piggin goto fput_fail; 1717f52e0c11SAl Viro } 171831e6b01fSNick Piggin 171931e6b01fSNick Piggin nd->path = file->f_path; 1720e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 172131e6b01fSNick Piggin if (fput_needed) 172270e9b357SAl Viro *fp = file; 1723c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 1724962830dfSAndi Kleen br_read_lock(&vfsmount_lock); 172531e6b01fSNick Piggin rcu_read_lock(); 17265590ff0dSUlrich Drepper } else { 17275dd784d0SJan Blunck path_get(&file->f_path); 17285590ff0dSUlrich Drepper fput_light(file, fput_needed); 17291da177e4SLinus Torvalds } 1730e41f7d4eSAl Viro } 1731e41f7d4eSAl Viro 173231e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 17339b4a9b14SAl Viro return 0; 17342dfdd266SJosef 'Jeff' Sipek 17359b4a9b14SAl Viro fput_fail: 17369b4a9b14SAl Viro fput_light(file, fput_needed); 17379b4a9b14SAl Viro out_fail: 17389b4a9b14SAl Viro return retval; 17399b4a9b14SAl Viro } 17409b4a9b14SAl Viro 1741bd92d7feSAl Viro static inline int lookup_last(struct nameidata *nd, struct path *path) 1742bd92d7feSAl Viro { 1743bd92d7feSAl Viro if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) 1744bd92d7feSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 1745bd92d7feSAl Viro 1746bd92d7feSAl Viro nd->flags &= ~LOOKUP_PARENT; 1747bd92d7feSAl Viro return walk_component(nd, path, &nd->last, nd->last_type, 1748bd92d7feSAl Viro nd->flags & LOOKUP_FOLLOW); 1749bd92d7feSAl Viro } 1750bd92d7feSAl Viro 17519b4a9b14SAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 1752ee0827cdSAl Viro static int path_lookupat(int dfd, const char *name, 17539b4a9b14SAl Viro unsigned int flags, struct nameidata *nd) 17549b4a9b14SAl Viro { 175570e9b357SAl Viro struct file *base = NULL; 1756bd92d7feSAl Viro struct path path; 1757bd92d7feSAl Viro int err; 175831e6b01fSNick Piggin 175931e6b01fSNick Piggin /* 176031e6b01fSNick Piggin * Path walking is largely split up into 2 different synchronisation 176131e6b01fSNick Piggin * schemes, rcu-walk and ref-walk (explained in 176231e6b01fSNick Piggin * Documentation/filesystems/path-lookup.txt). These share much of the 176331e6b01fSNick Piggin * path walk code, but some things particularly setup, cleanup, and 176431e6b01fSNick Piggin * following mounts are sufficiently divergent that functions are 176531e6b01fSNick Piggin * duplicated. Typically there is a function foo(), and its RCU 176631e6b01fSNick Piggin * analogue, foo_rcu(). 176731e6b01fSNick Piggin * 176831e6b01fSNick Piggin * -ECHILD is the error number of choice (just to avoid clashes) that 176931e6b01fSNick Piggin * is returned if some aspect of an rcu-walk fails. Such an error must 177031e6b01fSNick Piggin * be handled by restarting a traditional ref-walk (which will always 177131e6b01fSNick Piggin * be able to complete). 177231e6b01fSNick Piggin */ 1773bd92d7feSAl Viro err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base); 1774ee0827cdSAl Viro 1775bd92d7feSAl Viro if (unlikely(err)) 1776bd92d7feSAl Viro return err; 1777ee0827cdSAl Viro 1778ee0827cdSAl Viro current->total_link_count = 0; 1779bd92d7feSAl Viro err = link_path_walk(name, nd); 1780bd92d7feSAl Viro 1781bd92d7feSAl Viro if (!err && !(flags & LOOKUP_PARENT)) { 1782bd92d7feSAl Viro err = lookup_last(nd, &path); 1783bd92d7feSAl Viro while (err > 0) { 1784bd92d7feSAl Viro void *cookie; 1785bd92d7feSAl Viro struct path link = path; 1786bd92d7feSAl Viro nd->flags |= LOOKUP_PARENT; 1787574197e0SAl Viro err = follow_link(&link, nd, &cookie); 17886d7b5aaeSAl Viro if (err) 17896d7b5aaeSAl Viro break; 1790bd92d7feSAl Viro err = lookup_last(nd, &path); 1791574197e0SAl Viro put_link(nd, &link, cookie); 1792bd92d7feSAl Viro } 1793bd92d7feSAl Viro } 1794ee0827cdSAl Viro 17959f1fafeeSAl Viro if (!err) 17969f1fafeeSAl Viro err = complete_walk(nd); 1797bd92d7feSAl Viro 1798bd92d7feSAl Viro if (!err && nd->flags & LOOKUP_DIRECTORY) { 1799bd92d7feSAl Viro if (!nd->inode->i_op->lookup) { 1800bd92d7feSAl Viro path_put(&nd->path); 1801bd23a539SAl Viro err = -ENOTDIR; 1802bd92d7feSAl Viro } 1803bd92d7feSAl Viro } 180416c2cd71SAl Viro 180570e9b357SAl Viro if (base) 180670e9b357SAl Viro fput(base); 1807ee0827cdSAl Viro 18085b6ca027SAl Viro if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 180931e6b01fSNick Piggin path_put(&nd->root); 181031e6b01fSNick Piggin nd->root.mnt = NULL; 181131e6b01fSNick Piggin } 1812bd92d7feSAl Viro return err; 181331e6b01fSNick Piggin } 181431e6b01fSNick Piggin 1815ee0827cdSAl Viro static int do_path_lookup(int dfd, const char *name, 1816ee0827cdSAl Viro unsigned int flags, struct nameidata *nd) 1817ee0827cdSAl Viro { 1818ee0827cdSAl Viro int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd); 1819ee0827cdSAl Viro if (unlikely(retval == -ECHILD)) 1820ee0827cdSAl Viro retval = path_lookupat(dfd, name, flags, nd); 1821ee0827cdSAl Viro if (unlikely(retval == -ESTALE)) 1822ee0827cdSAl Viro retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd); 1823ee0827cdSAl Viro 182431e6b01fSNick Piggin if (likely(!retval)) { 182531e6b01fSNick Piggin if (unlikely(!audit_dummy_context())) { 182631e6b01fSNick Piggin if (nd->path.dentry && nd->inode) 182731e6b01fSNick Piggin audit_inode(name, nd->path.dentry); 182831e6b01fSNick Piggin } 182931e6b01fSNick Piggin } 1830170aa3d0SUlrich Drepper return retval; 18311da177e4SLinus Torvalds } 18321da177e4SLinus Torvalds 183379714f72SAl Viro /* does lookup, returns the object with parent locked */ 183479714f72SAl Viro struct dentry *kern_path_locked(const char *name, struct path *path) 18355590ff0dSUlrich Drepper { 183679714f72SAl Viro struct nameidata nd; 183779714f72SAl Viro struct dentry *d; 183879714f72SAl Viro int err = do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, &nd); 183979714f72SAl Viro if (err) 184079714f72SAl Viro return ERR_PTR(err); 184179714f72SAl Viro if (nd.last_type != LAST_NORM) { 184279714f72SAl Viro path_put(&nd.path); 184379714f72SAl Viro return ERR_PTR(-EINVAL); 184479714f72SAl Viro } 184579714f72SAl Viro mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 184679714f72SAl Viro d = lookup_one_len(nd.last.name, nd.path.dentry, nd.last.len); 184779714f72SAl Viro if (IS_ERR(d)) { 184879714f72SAl Viro mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 184979714f72SAl Viro path_put(&nd.path); 185079714f72SAl Viro return d; 185179714f72SAl Viro } 185279714f72SAl Viro *path = nd.path; 185379714f72SAl Viro return d; 18545590ff0dSUlrich Drepper } 18555590ff0dSUlrich Drepper 1856d1811465SAl Viro int kern_path(const char *name, unsigned int flags, struct path *path) 1857d1811465SAl Viro { 1858d1811465SAl Viro struct nameidata nd; 1859d1811465SAl Viro int res = do_path_lookup(AT_FDCWD, name, flags, &nd); 1860d1811465SAl Viro if (!res) 1861d1811465SAl Viro *path = nd.path; 1862d1811465SAl Viro return res; 1863d1811465SAl Viro } 1864d1811465SAl Viro 186516f18200SJosef 'Jeff' Sipek /** 186616f18200SJosef 'Jeff' Sipek * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair 186716f18200SJosef 'Jeff' Sipek * @dentry: pointer to dentry of the base directory 186816f18200SJosef 'Jeff' Sipek * @mnt: pointer to vfs mount of the base directory 186916f18200SJosef 'Jeff' Sipek * @name: pointer to file name 187016f18200SJosef 'Jeff' Sipek * @flags: lookup flags 1871e0a01249SAl Viro * @path: pointer to struct path to fill 187216f18200SJosef 'Jeff' Sipek */ 187316f18200SJosef 'Jeff' Sipek int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, 187416f18200SJosef 'Jeff' Sipek const char *name, unsigned int flags, 1875e0a01249SAl Viro struct path *path) 187616f18200SJosef 'Jeff' Sipek { 1877e0a01249SAl Viro struct nameidata nd; 1878e0a01249SAl Viro int err; 1879e0a01249SAl Viro nd.root.dentry = dentry; 1880e0a01249SAl Viro nd.root.mnt = mnt; 1881e0a01249SAl Viro BUG_ON(flags & LOOKUP_PARENT); 18825b6ca027SAl Viro /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */ 1883e0a01249SAl Viro err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd); 1884e0a01249SAl Viro if (!err) 1885e0a01249SAl Viro *path = nd.path; 1886e0a01249SAl Viro return err; 188716f18200SJosef 'Jeff' Sipek } 188816f18200SJosef 'Jeff' Sipek 1889057f6c01SJames Morris /* 1890057f6c01SJames Morris * Restricted form of lookup. Doesn't follow links, single-component only, 1891057f6c01SJames Morris * needs parent already locked. Doesn't follow mounts. 1892057f6c01SJames Morris * SMP-safe. 1893057f6c01SJames Morris */ 1894a244e169SAdrian Bunk static struct dentry *lookup_hash(struct nameidata *nd) 18951da177e4SLinus Torvalds { 189672bd866aSAl Viro return __lookup_hash(&nd->last, nd->path.dentry, nd->flags); 18971da177e4SLinus Torvalds } 18981da177e4SLinus Torvalds 1899eead1911SChristoph Hellwig /** 1900a6b91919SRandy Dunlap * lookup_one_len - filesystem helper to lookup single pathname component 1901eead1911SChristoph Hellwig * @name: pathname component to lookup 1902eead1911SChristoph Hellwig * @base: base directory to lookup from 1903eead1911SChristoph Hellwig * @len: maximum length @len should be interpreted to 1904eead1911SChristoph Hellwig * 1905a6b91919SRandy Dunlap * Note that this routine is purely a helper for filesystem usage and should 1906a6b91919SRandy Dunlap * not be called by generic code. Also note that by using this function the 1907eead1911SChristoph Hellwig * nameidata argument is passed to the filesystem methods and a filesystem 1908eead1911SChristoph Hellwig * using this helper needs to be prepared for that. 1909eead1911SChristoph Hellwig */ 1910057f6c01SJames Morris struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 1911057f6c01SJames Morris { 1912057f6c01SJames Morris struct qstr this; 19136a96ba54SAl Viro unsigned int c; 1914cda309deSMiklos Szeredi int err; 1915057f6c01SJames Morris 19162f9092e1SDavid Woodhouse WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex)); 19172f9092e1SDavid Woodhouse 19186a96ba54SAl Viro this.name = name; 19196a96ba54SAl Viro this.len = len; 19200145acc2SLinus Torvalds this.hash = full_name_hash(name, len); 19216a96ba54SAl Viro if (!len) 19226a96ba54SAl Viro return ERR_PTR(-EACCES); 19236a96ba54SAl Viro 19246a96ba54SAl Viro while (len--) { 19256a96ba54SAl Viro c = *(const unsigned char *)name++; 19266a96ba54SAl Viro if (c == '/' || c == '\0') 19276a96ba54SAl Viro return ERR_PTR(-EACCES); 19286a96ba54SAl Viro } 19295a202bcdSAl Viro /* 19305a202bcdSAl Viro * See if the low-level filesystem might want 19315a202bcdSAl Viro * to use its own hash.. 19325a202bcdSAl Viro */ 19335a202bcdSAl Viro if (base->d_flags & DCACHE_OP_HASH) { 19345a202bcdSAl Viro int err = base->d_op->d_hash(base, base->d_inode, &this); 19355a202bcdSAl Viro if (err < 0) 19365a202bcdSAl Viro return ERR_PTR(err); 19375a202bcdSAl Viro } 1938eead1911SChristoph Hellwig 1939cda309deSMiklos Szeredi err = inode_permission(base->d_inode, MAY_EXEC); 1940cda309deSMiklos Szeredi if (err) 1941cda309deSMiklos Szeredi return ERR_PTR(err); 1942cda309deSMiklos Szeredi 194372bd866aSAl Viro return __lookup_hash(&this, base, 0); 1944057f6c01SJames Morris } 1945057f6c01SJames Morris 19461fa1e7f6SAndy Whitcroft int user_path_at_empty(int dfd, const char __user *name, unsigned flags, 19471fa1e7f6SAndy Whitcroft struct path *path, int *empty) 19481da177e4SLinus Torvalds { 19492d8f3038SAl Viro struct nameidata nd; 19501fa1e7f6SAndy Whitcroft char *tmp = getname_flags(name, flags, empty); 19511da177e4SLinus Torvalds int err = PTR_ERR(tmp); 19521da177e4SLinus Torvalds if (!IS_ERR(tmp)) { 19532d8f3038SAl Viro 19542d8f3038SAl Viro BUG_ON(flags & LOOKUP_PARENT); 19552d8f3038SAl Viro 19562d8f3038SAl Viro err = do_path_lookup(dfd, tmp, flags, &nd); 19571da177e4SLinus Torvalds putname(tmp); 19582d8f3038SAl Viro if (!err) 19592d8f3038SAl Viro *path = nd.path; 19601da177e4SLinus Torvalds } 19611da177e4SLinus Torvalds return err; 19621da177e4SLinus Torvalds } 19631da177e4SLinus Torvalds 19641fa1e7f6SAndy Whitcroft int user_path_at(int dfd, const char __user *name, unsigned flags, 19651fa1e7f6SAndy Whitcroft struct path *path) 19661fa1e7f6SAndy Whitcroft { 1967f7493e5dSLinus Torvalds return user_path_at_empty(dfd, name, flags, path, NULL); 19681fa1e7f6SAndy Whitcroft } 19691fa1e7f6SAndy Whitcroft 19702ad94ae6SAl Viro static int user_path_parent(int dfd, const char __user *path, 19712ad94ae6SAl Viro struct nameidata *nd, char **name) 19722ad94ae6SAl Viro { 19732ad94ae6SAl Viro char *s = getname(path); 19742ad94ae6SAl Viro int error; 19752ad94ae6SAl Viro 19762ad94ae6SAl Viro if (IS_ERR(s)) 19772ad94ae6SAl Viro return PTR_ERR(s); 19782ad94ae6SAl Viro 19792ad94ae6SAl Viro error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd); 19802ad94ae6SAl Viro if (error) 19812ad94ae6SAl Viro putname(s); 19822ad94ae6SAl Viro else 19832ad94ae6SAl Viro *name = s; 19842ad94ae6SAl Viro 19852ad94ae6SAl Viro return error; 19862ad94ae6SAl Viro } 19872ad94ae6SAl Viro 19881da177e4SLinus Torvalds /* 19891da177e4SLinus Torvalds * It's inline, so penalty for filesystems that don't use sticky bit is 19901da177e4SLinus Torvalds * minimal. 19911da177e4SLinus Torvalds */ 19921da177e4SLinus Torvalds static inline int check_sticky(struct inode *dir, struct inode *inode) 19931da177e4SLinus Torvalds { 19948e96e3b7SEric W. Biederman kuid_t fsuid = current_fsuid(); 1995da9592edSDavid Howells 19961da177e4SLinus Torvalds if (!(dir->i_mode & S_ISVTX)) 19971da177e4SLinus Torvalds return 0; 19988e96e3b7SEric W. Biederman if (uid_eq(inode->i_uid, fsuid)) 19991da177e4SLinus Torvalds return 0; 20008e96e3b7SEric W. Biederman if (uid_eq(dir->i_uid, fsuid)) 20011da177e4SLinus Torvalds return 0; 20021a48e2acSEric W. Biederman return !inode_capable(inode, CAP_FOWNER); 20031da177e4SLinus Torvalds } 20041da177e4SLinus Torvalds 20051da177e4SLinus Torvalds /* 20061da177e4SLinus Torvalds * Check whether we can remove a link victim from directory dir, check 20071da177e4SLinus Torvalds * whether the type of victim is right. 20081da177e4SLinus Torvalds * 1. We can't do it if dir is read-only (done in permission()) 20091da177e4SLinus Torvalds * 2. We should have write and exec permissions on dir 20101da177e4SLinus Torvalds * 3. We can't remove anything from append-only dir 20111da177e4SLinus Torvalds * 4. We can't do anything with immutable dir (done in permission()) 20121da177e4SLinus Torvalds * 5. If the sticky bit on dir is set we should either 20131da177e4SLinus Torvalds * a. be owner of dir, or 20141da177e4SLinus Torvalds * b. be owner of victim, or 20151da177e4SLinus Torvalds * c. have CAP_FOWNER capability 20161da177e4SLinus Torvalds * 6. If the victim is append-only or immutable we can't do antyhing with 20171da177e4SLinus Torvalds * links pointing to it. 20181da177e4SLinus Torvalds * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. 20191da177e4SLinus Torvalds * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. 20201da177e4SLinus Torvalds * 9. We can't remove a root or mountpoint. 20211da177e4SLinus Torvalds * 10. We don't allow removal of NFS sillyrenamed files; it's handled by 20221da177e4SLinus Torvalds * nfs_async_unlink(). 20231da177e4SLinus Torvalds */ 2024858119e1SArjan van de Ven static int may_delete(struct inode *dir,struct dentry *victim,int isdir) 20251da177e4SLinus Torvalds { 20261da177e4SLinus Torvalds int error; 20271da177e4SLinus Torvalds 20281da177e4SLinus Torvalds if (!victim->d_inode) 20291da177e4SLinus Torvalds return -ENOENT; 20301da177e4SLinus Torvalds 20311da177e4SLinus Torvalds BUG_ON(victim->d_parent->d_inode != dir); 2032cccc6bbaSAl Viro audit_inode_child(victim, dir); 20331da177e4SLinus Torvalds 2034f419a2e3SAl Viro error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 20351da177e4SLinus Torvalds if (error) 20361da177e4SLinus Torvalds return error; 20371da177e4SLinus Torvalds if (IS_APPEND(dir)) 20381da177e4SLinus Torvalds return -EPERM; 20391da177e4SLinus Torvalds if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)|| 2040f9454548SHugh Dickins IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode)) 20411da177e4SLinus Torvalds return -EPERM; 20421da177e4SLinus Torvalds if (isdir) { 20431da177e4SLinus Torvalds if (!S_ISDIR(victim->d_inode->i_mode)) 20441da177e4SLinus Torvalds return -ENOTDIR; 20451da177e4SLinus Torvalds if (IS_ROOT(victim)) 20461da177e4SLinus Torvalds return -EBUSY; 20471da177e4SLinus Torvalds } else if (S_ISDIR(victim->d_inode->i_mode)) 20481da177e4SLinus Torvalds return -EISDIR; 20491da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 20501da177e4SLinus Torvalds return -ENOENT; 20511da177e4SLinus Torvalds if (victim->d_flags & DCACHE_NFSFS_RENAMED) 20521da177e4SLinus Torvalds return -EBUSY; 20531da177e4SLinus Torvalds return 0; 20541da177e4SLinus Torvalds } 20551da177e4SLinus Torvalds 20561da177e4SLinus Torvalds /* Check whether we can create an object with dentry child in directory 20571da177e4SLinus Torvalds * dir. 20581da177e4SLinus Torvalds * 1. We can't do it if child already exists (open has special treatment for 20591da177e4SLinus Torvalds * this case, but since we are inlined it's OK) 20601da177e4SLinus Torvalds * 2. We can't do it if dir is read-only (done in permission()) 20611da177e4SLinus Torvalds * 3. We should have write and exec permissions on dir 20621da177e4SLinus Torvalds * 4. We can't do it if dir is immutable (done in permission()) 20631da177e4SLinus Torvalds */ 2064a95164d9SMiklos Szeredi static inline int may_create(struct inode *dir, struct dentry *child) 20651da177e4SLinus Torvalds { 20661da177e4SLinus Torvalds if (child->d_inode) 20671da177e4SLinus Torvalds return -EEXIST; 20681da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 20691da177e4SLinus Torvalds return -ENOENT; 2070f419a2e3SAl Viro return inode_permission(dir, MAY_WRITE | MAY_EXEC); 20711da177e4SLinus Torvalds } 20721da177e4SLinus Torvalds 20731da177e4SLinus Torvalds /* 20741da177e4SLinus Torvalds * p1 and p2 should be directories on the same fs. 20751da177e4SLinus Torvalds */ 20761da177e4SLinus Torvalds struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 20771da177e4SLinus Torvalds { 20781da177e4SLinus Torvalds struct dentry *p; 20791da177e4SLinus Torvalds 20801da177e4SLinus Torvalds if (p1 == p2) { 2081f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 20821da177e4SLinus Torvalds return NULL; 20831da177e4SLinus Torvalds } 20841da177e4SLinus Torvalds 2085a11f3a05SArjan van de Ven mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 20861da177e4SLinus Torvalds 2087e2761a11SOGAWA Hirofumi p = d_ancestor(p2, p1); 2088e2761a11SOGAWA Hirofumi if (p) { 2089f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT); 2090f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD); 20911da177e4SLinus Torvalds return p; 20921da177e4SLinus Torvalds } 20931da177e4SLinus Torvalds 2094e2761a11SOGAWA Hirofumi p = d_ancestor(p1, p2); 2095e2761a11SOGAWA Hirofumi if (p) { 2096f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 2097f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); 20981da177e4SLinus Torvalds return p; 20991da177e4SLinus Torvalds } 21001da177e4SLinus Torvalds 2101f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 2102f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); 21031da177e4SLinus Torvalds return NULL; 21041da177e4SLinus Torvalds } 21051da177e4SLinus Torvalds 21061da177e4SLinus Torvalds void unlock_rename(struct dentry *p1, struct dentry *p2) 21071da177e4SLinus Torvalds { 21081b1dcc1bSJes Sorensen mutex_unlock(&p1->d_inode->i_mutex); 21091da177e4SLinus Torvalds if (p1 != p2) { 21101b1dcc1bSJes Sorensen mutex_unlock(&p2->d_inode->i_mutex); 2111a11f3a05SArjan van de Ven mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 21121da177e4SLinus Torvalds } 21131da177e4SLinus Torvalds } 21141da177e4SLinus Torvalds 21154acdaf27SAl Viro int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, 2116312b63fbSAl Viro bool want_excl) 21171da177e4SLinus Torvalds { 2118a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 21191da177e4SLinus Torvalds if (error) 21201da177e4SLinus Torvalds return error; 21211da177e4SLinus Torvalds 2122acfa4380SAl Viro if (!dir->i_op->create) 21231da177e4SLinus Torvalds return -EACCES; /* shouldn't it be ENOSYS? */ 21241da177e4SLinus Torvalds mode &= S_IALLUGO; 21251da177e4SLinus Torvalds mode |= S_IFREG; 21261da177e4SLinus Torvalds error = security_inode_create(dir, dentry, mode); 21271da177e4SLinus Torvalds if (error) 21281da177e4SLinus Torvalds return error; 2129312b63fbSAl Viro error = dir->i_op->create(dir, dentry, mode, want_excl); 2130a74574aaSStephen Smalley if (!error) 2131f38aa942SAmy Griffis fsnotify_create(dir, dentry); 21321da177e4SLinus Torvalds return error; 21331da177e4SLinus Torvalds } 21341da177e4SLinus Torvalds 213573d049a4SAl Viro static int may_open(struct path *path, int acc_mode, int flag) 21361da177e4SLinus Torvalds { 21373fb64190SChristoph Hellwig struct dentry *dentry = path->dentry; 21381da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 21391da177e4SLinus Torvalds int error; 21401da177e4SLinus Torvalds 2141bcda7652SAl Viro /* O_PATH? */ 2142bcda7652SAl Viro if (!acc_mode) 2143bcda7652SAl Viro return 0; 2144bcda7652SAl Viro 21451da177e4SLinus Torvalds if (!inode) 21461da177e4SLinus Torvalds return -ENOENT; 21471da177e4SLinus Torvalds 2148c8fe8f30SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 2149c8fe8f30SChristoph Hellwig case S_IFLNK: 21501da177e4SLinus Torvalds return -ELOOP; 2151c8fe8f30SChristoph Hellwig case S_IFDIR: 2152c8fe8f30SChristoph Hellwig if (acc_mode & MAY_WRITE) 21531da177e4SLinus Torvalds return -EISDIR; 2154c8fe8f30SChristoph Hellwig break; 2155c8fe8f30SChristoph Hellwig case S_IFBLK: 2156c8fe8f30SChristoph Hellwig case S_IFCHR: 21573fb64190SChristoph Hellwig if (path->mnt->mnt_flags & MNT_NODEV) 21581da177e4SLinus Torvalds return -EACCES; 2159c8fe8f30SChristoph Hellwig /*FALLTHRU*/ 2160c8fe8f30SChristoph Hellwig case S_IFIFO: 2161c8fe8f30SChristoph Hellwig case S_IFSOCK: 21621da177e4SLinus Torvalds flag &= ~O_TRUNC; 2163c8fe8f30SChristoph Hellwig break; 21644a3fd211SDave Hansen } 2165b41572e9SDave Hansen 21663fb64190SChristoph Hellwig error = inode_permission(inode, acc_mode); 2167b41572e9SDave Hansen if (error) 2168b41572e9SDave Hansen return error; 21696146f0d5SMimi Zohar 21701da177e4SLinus Torvalds /* 21711da177e4SLinus Torvalds * An append-only file must be opened in append mode for writing. 21721da177e4SLinus Torvalds */ 21731da177e4SLinus Torvalds if (IS_APPEND(inode)) { 21748737c930SAl Viro if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) 21757715b521SAl Viro return -EPERM; 21761da177e4SLinus Torvalds if (flag & O_TRUNC) 21777715b521SAl Viro return -EPERM; 21781da177e4SLinus Torvalds } 21791da177e4SLinus Torvalds 21801da177e4SLinus Torvalds /* O_NOATIME can only be set by the owner or superuser */ 21812e149670SSerge E. Hallyn if (flag & O_NOATIME && !inode_owner_or_capable(inode)) 21827715b521SAl Viro return -EPERM; 21831da177e4SLinus Torvalds 2184f3c7691eSJ. Bruce Fields return 0; 21857715b521SAl Viro } 21867715b521SAl Viro 2187e1181ee6SJeff Layton static int handle_truncate(struct file *filp) 21887715b521SAl Viro { 2189e1181ee6SJeff Layton struct path *path = &filp->f_path; 21907715b521SAl Viro struct inode *inode = path->dentry->d_inode; 21917715b521SAl Viro int error = get_write_access(inode); 21921da177e4SLinus Torvalds if (error) 21937715b521SAl Viro return error; 21941da177e4SLinus Torvalds /* 21951da177e4SLinus Torvalds * Refuse to truncate files with mandatory locks held on them. 21961da177e4SLinus Torvalds */ 21971da177e4SLinus Torvalds error = locks_verify_locked(inode); 2198be6d3e56SKentaro Takeda if (!error) 2199ea0d3ab2STetsuo Handa error = security_path_truncate(path); 22001da177e4SLinus Torvalds if (!error) { 22017715b521SAl Viro error = do_truncate(path->dentry, 0, 2202d139d7ffSMiklos Szeredi ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, 2203e1181ee6SJeff Layton filp); 22041da177e4SLinus Torvalds } 22051da177e4SLinus Torvalds put_write_access(inode); 2206acd0c935SMimi Zohar return error; 22071da177e4SLinus Torvalds } 22081da177e4SLinus Torvalds 2209d57999e1SDave Hansen static inline int open_to_namei_flags(int flag) 2210d57999e1SDave Hansen { 22118a5e929dSAl Viro if ((flag & O_ACCMODE) == 3) 22128a5e929dSAl Viro flag--; 2213d57999e1SDave Hansen return flag; 2214d57999e1SDave Hansen } 2215d57999e1SDave Hansen 2216d18e9008SMiklos Szeredi static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode) 2217d18e9008SMiklos Szeredi { 2218d18e9008SMiklos Szeredi int error = security_path_mknod(dir, dentry, mode, 0); 2219d18e9008SMiklos Szeredi if (error) 2220d18e9008SMiklos Szeredi return error; 2221d18e9008SMiklos Szeredi 2222d18e9008SMiklos Szeredi error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC); 2223d18e9008SMiklos Szeredi if (error) 2224d18e9008SMiklos Szeredi return error; 2225d18e9008SMiklos Szeredi 2226d18e9008SMiklos Szeredi return security_inode_create(dir->dentry->d_inode, dentry, mode); 2227d18e9008SMiklos Szeredi } 2228d18e9008SMiklos Szeredi 22291acf0af9SDavid Howells /* 22301acf0af9SDavid Howells * Attempt to atomically look up, create and open a file from a negative 22311acf0af9SDavid Howells * dentry. 22321acf0af9SDavid Howells * 22331acf0af9SDavid Howells * Returns 0 if successful. The file will have been created and attached to 22341acf0af9SDavid Howells * @file by the filesystem calling finish_open(). 22351acf0af9SDavid Howells * 22361acf0af9SDavid Howells * Returns 1 if the file was looked up only or didn't need creating. The 22371acf0af9SDavid Howells * caller will need to perform the open themselves. @path will have been 22381acf0af9SDavid Howells * updated to point to the new dentry. This may be negative. 22391acf0af9SDavid Howells * 22401acf0af9SDavid Howells * Returns an error code otherwise. 22411acf0af9SDavid Howells */ 22422675a4ebSAl Viro static int atomic_open(struct nameidata *nd, struct dentry *dentry, 224330d90494SAl Viro struct path *path, struct file *file, 2244015c3bbcSMiklos Szeredi const struct open_flags *op, 224577d660a8SMiklos Szeredi bool *want_write, bool need_lookup, 224647237687SAl Viro int *opened) 2247d18e9008SMiklos Szeredi { 2248d18e9008SMiklos Szeredi struct inode *dir = nd->path.dentry->d_inode; 2249d18e9008SMiklos Szeredi unsigned open_flag = open_to_namei_flags(op->open_flag); 2250d18e9008SMiklos Szeredi umode_t mode; 2251d18e9008SMiklos Szeredi int error; 2252d18e9008SMiklos Szeredi int acc_mode; 2253d18e9008SMiklos Szeredi int create_error = 0; 2254d18e9008SMiklos Szeredi struct dentry *const DENTRY_NOT_SET = (void *) -1UL; 2255d18e9008SMiklos Szeredi 2256d18e9008SMiklos Szeredi BUG_ON(dentry->d_inode); 2257d18e9008SMiklos Szeredi 2258d18e9008SMiklos Szeredi /* Don't create child dentry for a dead directory. */ 2259d18e9008SMiklos Szeredi if (unlikely(IS_DEADDIR(dir))) { 22602675a4ebSAl Viro error = -ENOENT; 2261d18e9008SMiklos Szeredi goto out; 2262d18e9008SMiklos Szeredi } 2263d18e9008SMiklos Szeredi 2264d18e9008SMiklos Szeredi mode = op->mode & S_IALLUGO; 2265d18e9008SMiklos Szeredi if ((open_flag & O_CREAT) && !IS_POSIXACL(dir)) 2266d18e9008SMiklos Szeredi mode &= ~current_umask(); 2267d18e9008SMiklos Szeredi 2268d18e9008SMiklos Szeredi if (open_flag & O_EXCL) { 2269d18e9008SMiklos Szeredi open_flag &= ~O_TRUNC; 227047237687SAl Viro *opened |= FILE_CREATED; 2271d18e9008SMiklos Szeredi } 2272d18e9008SMiklos Szeredi 2273d18e9008SMiklos Szeredi /* 2274d18e9008SMiklos Szeredi * Checking write permission is tricky, bacuse we don't know if we are 2275d18e9008SMiklos Szeredi * going to actually need it: O_CREAT opens should work as long as the 2276d18e9008SMiklos Szeredi * file exists. But checking existence breaks atomicity. The trick is 2277d18e9008SMiklos Szeredi * to check access and if not granted clear O_CREAT from the flags. 2278d18e9008SMiklos Szeredi * 2279d18e9008SMiklos Szeredi * Another problem is returing the "right" error value (e.g. for an 2280d18e9008SMiklos Szeredi * O_EXCL open we want to return EEXIST not EROFS). 2281d18e9008SMiklos Szeredi */ 2282d18e9008SMiklos Szeredi if ((open_flag & (O_CREAT | O_TRUNC)) || 2283d18e9008SMiklos Szeredi (open_flag & O_ACCMODE) != O_RDONLY) { 2284d18e9008SMiklos Szeredi error = mnt_want_write(nd->path.mnt); 2285d18e9008SMiklos Szeredi if (!error) { 228677d660a8SMiklos Szeredi *want_write = true; 2287d18e9008SMiklos Szeredi } else if (!(open_flag & O_CREAT)) { 2288d18e9008SMiklos Szeredi /* 2289d18e9008SMiklos Szeredi * No O_CREATE -> atomicity not a requirement -> fall 2290d18e9008SMiklos Szeredi * back to lookup + open 2291d18e9008SMiklos Szeredi */ 2292d18e9008SMiklos Szeredi goto no_open; 2293d18e9008SMiklos Szeredi } else if (open_flag & (O_EXCL | O_TRUNC)) { 2294d18e9008SMiklos Szeredi /* Fall back and fail with the right error */ 2295d18e9008SMiklos Szeredi create_error = error; 2296d18e9008SMiklos Szeredi goto no_open; 2297d18e9008SMiklos Szeredi } else { 2298d18e9008SMiklos Szeredi /* No side effects, safe to clear O_CREAT */ 2299d18e9008SMiklos Szeredi create_error = error; 2300d18e9008SMiklos Szeredi open_flag &= ~O_CREAT; 2301d18e9008SMiklos Szeredi } 2302d18e9008SMiklos Szeredi } 2303d18e9008SMiklos Szeredi 2304d18e9008SMiklos Szeredi if (open_flag & O_CREAT) { 2305d18e9008SMiklos Szeredi error = may_o_create(&nd->path, dentry, op->mode); 2306d18e9008SMiklos Szeredi if (error) { 2307d18e9008SMiklos Szeredi create_error = error; 2308d18e9008SMiklos Szeredi if (open_flag & O_EXCL) 2309d18e9008SMiklos Szeredi goto no_open; 2310d18e9008SMiklos Szeredi open_flag &= ~O_CREAT; 2311d18e9008SMiklos Szeredi } 2312d18e9008SMiklos Szeredi } 2313d18e9008SMiklos Szeredi 2314d18e9008SMiklos Szeredi if (nd->flags & LOOKUP_DIRECTORY) 2315d18e9008SMiklos Szeredi open_flag |= O_DIRECTORY; 2316d18e9008SMiklos Szeredi 231730d90494SAl Viro file->f_path.dentry = DENTRY_NOT_SET; 231830d90494SAl Viro file->f_path.mnt = nd->path.mnt; 231930d90494SAl Viro error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode, 232047237687SAl Viro opened); 2321d9585277SAl Viro if (error < 0) { 2322d9585277SAl Viro if (create_error && error == -ENOENT) 2323d9585277SAl Viro error = create_error; 2324d18e9008SMiklos Szeredi goto out; 2325d18e9008SMiklos Szeredi } 2326d18e9008SMiklos Szeredi 2327d18e9008SMiklos Szeredi acc_mode = op->acc_mode; 232847237687SAl Viro if (*opened & FILE_CREATED) { 2329d18e9008SMiklos Szeredi fsnotify_create(dir, dentry); 2330d18e9008SMiklos Szeredi acc_mode = MAY_OPEN; 2331d18e9008SMiklos Szeredi } 2332d18e9008SMiklos Szeredi 2333d9585277SAl Viro if (error) { /* returned 1, that is */ 233430d90494SAl Viro if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { 23352675a4ebSAl Viro error = -EIO; 2336d18e9008SMiklos Szeredi goto out; 2337d18e9008SMiklos Szeredi } 233830d90494SAl Viro if (file->f_path.dentry) { 2339d18e9008SMiklos Szeredi dput(dentry); 234030d90494SAl Viro dentry = file->f_path.dentry; 2341d18e9008SMiklos Szeredi } 2342d18e9008SMiklos Szeredi goto looked_up; 2343d18e9008SMiklos Szeredi } 2344d18e9008SMiklos Szeredi 2345d18e9008SMiklos Szeredi /* 2346d18e9008SMiklos Szeredi * We didn't have the inode before the open, so check open permission 2347d18e9008SMiklos Szeredi * here. 2348d18e9008SMiklos Szeredi */ 23492675a4ebSAl Viro error = may_open(&file->f_path, acc_mode, open_flag); 23502675a4ebSAl Viro if (error) 23512675a4ebSAl Viro fput(file); 2352d18e9008SMiklos Szeredi 2353d18e9008SMiklos Szeredi out: 2354d18e9008SMiklos Szeredi dput(dentry); 23552675a4ebSAl Viro return error; 2356d18e9008SMiklos Szeredi 2357d18e9008SMiklos Szeredi no_open: 2358d18e9008SMiklos Szeredi if (need_lookup) { 235972bd866aSAl Viro dentry = lookup_real(dir, dentry, nd->flags); 2360d18e9008SMiklos Szeredi if (IS_ERR(dentry)) 23612675a4ebSAl Viro return PTR_ERR(dentry); 2362d18e9008SMiklos Szeredi 2363d18e9008SMiklos Szeredi if (create_error) { 2364d18e9008SMiklos Szeredi int open_flag = op->open_flag; 2365d18e9008SMiklos Szeredi 23662675a4ebSAl Viro error = create_error; 2367d18e9008SMiklos Szeredi if ((open_flag & O_EXCL)) { 2368d18e9008SMiklos Szeredi if (!dentry->d_inode) 2369d18e9008SMiklos Szeredi goto out; 2370d18e9008SMiklos Szeredi } else if (!dentry->d_inode) { 2371d18e9008SMiklos Szeredi goto out; 2372d18e9008SMiklos Szeredi } else if ((open_flag & O_TRUNC) && 2373d18e9008SMiklos Szeredi S_ISREG(dentry->d_inode->i_mode)) { 2374d18e9008SMiklos Szeredi goto out; 2375d18e9008SMiklos Szeredi } 2376d18e9008SMiklos Szeredi /* will fail later, go on to get the right error */ 2377d18e9008SMiklos Szeredi } 2378d18e9008SMiklos Szeredi } 2379d18e9008SMiklos Szeredi looked_up: 2380d18e9008SMiklos Szeredi path->dentry = dentry; 2381d18e9008SMiklos Szeredi path->mnt = nd->path.mnt; 23822675a4ebSAl Viro return 1; 2383d18e9008SMiklos Szeredi } 2384d18e9008SMiklos Szeredi 238531e6b01fSNick Piggin /* 23861acf0af9SDavid Howells * Look up and maybe create and open the last component. 2387d58ffd35SMiklos Szeredi * 2388d58ffd35SMiklos Szeredi * Must be called with i_mutex held on parent. 2389d58ffd35SMiklos Szeredi * 23901acf0af9SDavid Howells * Returns 0 if the file was successfully atomically created (if necessary) and 23911acf0af9SDavid Howells * opened. In this case the file will be returned attached to @file. 23921acf0af9SDavid Howells * 23931acf0af9SDavid Howells * Returns 1 if the file was not completely opened at this time, though lookups 23941acf0af9SDavid Howells * and creations will have been performed and the dentry returned in @path will 23951acf0af9SDavid Howells * be positive upon return if O_CREAT was specified. If O_CREAT wasn't 23961acf0af9SDavid Howells * specified then a negative dentry may be returned. 23971acf0af9SDavid Howells * 23981acf0af9SDavid Howells * An error code is returned otherwise. 23991acf0af9SDavid Howells * 24001acf0af9SDavid Howells * FILE_CREATE will be set in @*opened if the dentry was created and will be 24011acf0af9SDavid Howells * cleared otherwise prior to returning. 2402d58ffd35SMiklos Szeredi */ 24032675a4ebSAl Viro static int lookup_open(struct nameidata *nd, struct path *path, 240430d90494SAl Viro struct file *file, 2405d58ffd35SMiklos Szeredi const struct open_flags *op, 240647237687SAl Viro bool *want_write, int *opened) 2407d58ffd35SMiklos Szeredi { 2408d58ffd35SMiklos Szeredi struct dentry *dir = nd->path.dentry; 240954ef4872SMiklos Szeredi struct inode *dir_inode = dir->d_inode; 2410d58ffd35SMiklos Szeredi struct dentry *dentry; 2411d58ffd35SMiklos Szeredi int error; 241254ef4872SMiklos Szeredi bool need_lookup; 2413d58ffd35SMiklos Szeredi 241447237687SAl Viro *opened &= ~FILE_CREATED; 2415201f956eSAl Viro dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup); 2416d58ffd35SMiklos Szeredi if (IS_ERR(dentry)) 24172675a4ebSAl Viro return PTR_ERR(dentry); 2418d58ffd35SMiklos Szeredi 2419d18e9008SMiklos Szeredi /* Cached positive dentry: will open in f_op->open */ 2420d18e9008SMiklos Szeredi if (!need_lookup && dentry->d_inode) 2421d18e9008SMiklos Szeredi goto out_no_open; 2422d18e9008SMiklos Szeredi 2423d18e9008SMiklos Szeredi if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) { 242430d90494SAl Viro return atomic_open(nd, dentry, path, file, op, want_write, 242547237687SAl Viro need_lookup, opened); 2426d18e9008SMiklos Szeredi } 2427d18e9008SMiklos Szeredi 242854ef4872SMiklos Szeredi if (need_lookup) { 242954ef4872SMiklos Szeredi BUG_ON(dentry->d_inode); 243054ef4872SMiklos Szeredi 243172bd866aSAl Viro dentry = lookup_real(dir_inode, dentry, nd->flags); 243254ef4872SMiklos Szeredi if (IS_ERR(dentry)) 24332675a4ebSAl Viro return PTR_ERR(dentry); 243454ef4872SMiklos Szeredi } 243554ef4872SMiklos Szeredi 2436d58ffd35SMiklos Szeredi /* Negative dentry, just create the file */ 2437d58ffd35SMiklos Szeredi if (!dentry->d_inode && (op->open_flag & O_CREAT)) { 2438d58ffd35SMiklos Szeredi umode_t mode = op->mode; 2439d58ffd35SMiklos Szeredi if (!IS_POSIXACL(dir->d_inode)) 2440d58ffd35SMiklos Szeredi mode &= ~current_umask(); 2441d58ffd35SMiklos Szeredi /* 2442d58ffd35SMiklos Szeredi * This write is needed to ensure that a 2443d58ffd35SMiklos Szeredi * rw->ro transition does not occur between 2444d58ffd35SMiklos Szeredi * the time when the file is created and when 2445d58ffd35SMiklos Szeredi * a permanent write count is taken through 2446015c3bbcSMiklos Szeredi * the 'struct file' in finish_open(). 2447d58ffd35SMiklos Szeredi */ 2448d58ffd35SMiklos Szeredi error = mnt_want_write(nd->path.mnt); 2449d58ffd35SMiklos Szeredi if (error) 2450d58ffd35SMiklos Szeredi goto out_dput; 245177d660a8SMiklos Szeredi *want_write = true; 245247237687SAl Viro *opened |= FILE_CREATED; 2453d58ffd35SMiklos Szeredi error = security_path_mknod(&nd->path, dentry, mode, 0); 2454d58ffd35SMiklos Szeredi if (error) 2455d58ffd35SMiklos Szeredi goto out_dput; 2456312b63fbSAl Viro error = vfs_create(dir->d_inode, dentry, mode, 2457312b63fbSAl Viro nd->flags & LOOKUP_EXCL); 2458d58ffd35SMiklos Szeredi if (error) 2459d58ffd35SMiklos Szeredi goto out_dput; 2460d58ffd35SMiklos Szeredi } 2461d18e9008SMiklos Szeredi out_no_open: 2462d58ffd35SMiklos Szeredi path->dentry = dentry; 2463d58ffd35SMiklos Szeredi path->mnt = nd->path.mnt; 24642675a4ebSAl Viro return 1; 2465d58ffd35SMiklos Szeredi 2466d58ffd35SMiklos Szeredi out_dput: 2467d58ffd35SMiklos Szeredi dput(dentry); 24682675a4ebSAl Viro return error; 2469d58ffd35SMiklos Szeredi } 2470d58ffd35SMiklos Szeredi 2471d58ffd35SMiklos Szeredi /* 2472fe2d35ffSAl Viro * Handle the last step of open() 247331e6b01fSNick Piggin */ 24742675a4ebSAl Viro static int do_last(struct nameidata *nd, struct path *path, 247530d90494SAl Viro struct file *file, const struct open_flags *op, 247647237687SAl Viro int *opened, const char *pathname) 2477fb1cc555SAl Viro { 2478a1e28038SAl Viro struct dentry *dir = nd->path.dentry; 2479ca344a89SAl Viro int open_flag = op->open_flag; 248077d660a8SMiklos Szeredi bool will_truncate = (open_flag & O_TRUNC) != 0; 248177d660a8SMiklos Szeredi bool want_write = false; 2482bcda7652SAl Viro int acc_mode = op->acc_mode; 2483a1eb3315SMiklos Szeredi struct inode *inode; 248477d660a8SMiklos Szeredi bool symlink_ok = false; 248516b1c1cdSMiklos Szeredi struct path save_parent = { .dentry = NULL, .mnt = NULL }; 248616b1c1cdSMiklos Szeredi bool retried = false; 248716c2cd71SAl Viro int error; 2488fb1cc555SAl Viro 2489c3e380b0SAl Viro nd->flags &= ~LOOKUP_PARENT; 2490c3e380b0SAl Viro nd->flags |= op->intent; 2491c3e380b0SAl Viro 24921f36f774SAl Viro switch (nd->last_type) { 24931f36f774SAl Viro case LAST_DOTDOT: 2494176306f5SNeil Brown case LAST_DOT: 2495fe2d35ffSAl Viro error = handle_dots(nd, nd->last_type); 2496fe2d35ffSAl Viro if (error) 24972675a4ebSAl Viro return error; 24981f36f774SAl Viro /* fallthrough */ 24991f36f774SAl Viro case LAST_ROOT: 25009f1fafeeSAl Viro error = complete_walk(nd); 250116c2cd71SAl Viro if (error) 25022675a4ebSAl Viro return error; 2503fe2d35ffSAl Viro audit_inode(pathname, nd->path.dentry); 2504ca344a89SAl Viro if (open_flag & O_CREAT) { 250516c2cd71SAl Viro error = -EISDIR; 25062675a4ebSAl Viro goto out; 2507fe2d35ffSAl Viro } 2508e83db167SMiklos Szeredi goto finish_open; 25091f36f774SAl Viro case LAST_BIND: 25109f1fafeeSAl Viro error = complete_walk(nd); 251116c2cd71SAl Viro if (error) 25122675a4ebSAl Viro return error; 25131f36f774SAl Viro audit_inode(pathname, dir); 2514e83db167SMiklos Szeredi goto finish_open; 25151f36f774SAl Viro } 2516a2c36b45SAl Viro 2517ca344a89SAl Viro if (!(open_flag & O_CREAT)) { 2518fe2d35ffSAl Viro if (nd->last.name[nd->last.len]) 2519fe2d35ffSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 2520bcda7652SAl Viro if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW)) 252177d660a8SMiklos Szeredi symlink_ok = true; 2522fe2d35ffSAl Viro /* we _can_ be in RCU mode here */ 2523a1eb3315SMiklos Szeredi error = lookup_fast(nd, &nd->last, path, &inode); 252471574865SMiklos Szeredi if (likely(!error)) 252571574865SMiklos Szeredi goto finish_lookup; 252671574865SMiklos Szeredi 2527ce57dfc1SAl Viro if (error < 0) 25282675a4ebSAl Viro goto out; 2529a1eb3315SMiklos Szeredi 253037d7fffcSMiklos Szeredi BUG_ON(nd->inode != dir->d_inode); 2531b6183df7SMiklos Szeredi } else { 2532fe2d35ffSAl Viro /* create side of things */ 2533a3fbbde7SAl Viro /* 2534b6183df7SMiklos Szeredi * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED 2535b6183df7SMiklos Szeredi * has been cleared when we got to the last component we are 2536b6183df7SMiklos Szeredi * about to look up 2537a3fbbde7SAl Viro */ 25389f1fafeeSAl Viro error = complete_walk(nd); 25399f1fafeeSAl Viro if (error) 25402675a4ebSAl Viro return error; 2541fe2d35ffSAl Viro 2542fe2d35ffSAl Viro audit_inode(pathname, dir); 254316c2cd71SAl Viro error = -EISDIR; 25441f36f774SAl Viro /* trailing slashes? */ 254531e6b01fSNick Piggin if (nd->last.name[nd->last.len]) 25462675a4ebSAl Viro goto out; 2547b6183df7SMiklos Szeredi } 25481f36f774SAl Viro 254916b1c1cdSMiklos Szeredi retry_lookup: 2550a1e28038SAl Viro mutex_lock(&dir->d_inode->i_mutex); 25512675a4ebSAl Viro error = lookup_open(nd, path, file, op, &want_write, opened); 2552fb1cc555SAl Viro mutex_unlock(&dir->d_inode->i_mutex); 2553fb1cc555SAl Viro 25542675a4ebSAl Viro if (error <= 0) { 25552675a4ebSAl Viro if (error) 2556d58ffd35SMiklos Szeredi goto out; 25576c0d46c4SAl Viro 255847237687SAl Viro if ((*opened & FILE_CREATED) || 25592675a4ebSAl Viro !S_ISREG(file->f_path.dentry->d_inode->i_mode)) 256077d660a8SMiklos Szeredi will_truncate = false; 2561d18e9008SMiklos Szeredi 25622675a4ebSAl Viro audit_inode(pathname, file->f_path.dentry); 2563d18e9008SMiklos Szeredi goto opened; 2564d18e9008SMiklos Szeredi } 2565d18e9008SMiklos Szeredi 256647237687SAl Viro if (*opened & FILE_CREATED) { 25679b44f1b3SAl Viro /* Don't check for write permission, don't truncate */ 2568ca344a89SAl Viro open_flag &= ~O_TRUNC; 256977d660a8SMiklos Szeredi will_truncate = false; 2570bcda7652SAl Viro acc_mode = MAY_OPEN; 2571d58ffd35SMiklos Szeredi path_to_nameidata(path, nd); 2572e83db167SMiklos Szeredi goto finish_open_created; 2573fb1cc555SAl Viro } 2574fb1cc555SAl Viro 2575fb1cc555SAl Viro /* 2576fb1cc555SAl Viro * It already exists. 2577fb1cc555SAl Viro */ 2578fb1cc555SAl Viro audit_inode(pathname, path->dentry); 2579fb1cc555SAl Viro 2580d18e9008SMiklos Szeredi /* 2581d18e9008SMiklos Szeredi * If atomic_open() acquired write access it is dropped now due to 2582d18e9008SMiklos Szeredi * possible mount and symlink following (this might be optimized away if 2583d18e9008SMiklos Szeredi * necessary...) 2584d18e9008SMiklos Szeredi */ 2585d18e9008SMiklos Szeredi if (want_write) { 2586d18e9008SMiklos Szeredi mnt_drop_write(nd->path.mnt); 258777d660a8SMiklos Szeredi want_write = false; 2588d18e9008SMiklos Szeredi } 2589d18e9008SMiklos Szeredi 2590fb1cc555SAl Viro error = -EEXIST; 2591ca344a89SAl Viro if (open_flag & O_EXCL) 2592fb1cc555SAl Viro goto exit_dput; 2593fb1cc555SAl Viro 25949875cf80SDavid Howells error = follow_managed(path, nd->flags); 25959875cf80SDavid Howells if (error < 0) 2596fb1cc555SAl Viro goto exit_dput; 2597fb1cc555SAl Viro 2598a3fbbde7SAl Viro if (error) 2599a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 2600a3fbbde7SAl Viro 2601decf3400SMiklos Szeredi BUG_ON(nd->flags & LOOKUP_RCU); 2602decf3400SMiklos Szeredi inode = path->dentry->d_inode; 26035f5daac1SMiklos Szeredi finish_lookup: 26045f5daac1SMiklos Szeredi /* we _can_ be in RCU mode here */ 2605fb1cc555SAl Viro error = -ENOENT; 260654c33e7fSMiklos Szeredi if (!inode) { 260754c33e7fSMiklos Szeredi path_to_nameidata(path, nd); 26082675a4ebSAl Viro goto out; 260954c33e7fSMiklos Szeredi } 26109e67f361SAl Viro 2611d45ea867SMiklos Szeredi if (should_follow_link(inode, !symlink_ok)) { 2612d45ea867SMiklos Szeredi if (nd->flags & LOOKUP_RCU) { 2613d45ea867SMiklos Szeredi if (unlikely(unlazy_walk(nd, path->dentry))) { 2614d45ea867SMiklos Szeredi error = -ECHILD; 26152675a4ebSAl Viro goto out; 2616d45ea867SMiklos Szeredi } 2617d45ea867SMiklos Szeredi } 2618d45ea867SMiklos Szeredi BUG_ON(inode != path->dentry->d_inode); 26192675a4ebSAl Viro return 1; 2620d45ea867SMiklos Szeredi } 2621fb1cc555SAl Viro 262216b1c1cdSMiklos Szeredi if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) { 2623fb1cc555SAl Viro path_to_nameidata(path, nd); 262416b1c1cdSMiklos Szeredi } else { 262516b1c1cdSMiklos Szeredi save_parent.dentry = nd->path.dentry; 262616b1c1cdSMiklos Szeredi save_parent.mnt = mntget(path->mnt); 262716b1c1cdSMiklos Szeredi nd->path.dentry = path->dentry; 262816b1c1cdSMiklos Szeredi 262916b1c1cdSMiklos Szeredi } 2630decf3400SMiklos Szeredi nd->inode = inode; 2631a3fbbde7SAl Viro /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ 2632a3fbbde7SAl Viro error = complete_walk(nd); 263316b1c1cdSMiklos Szeredi if (error) { 263416b1c1cdSMiklos Szeredi path_put(&save_parent); 26352675a4ebSAl Viro return error; 263616b1c1cdSMiklos Szeredi } 2637fb1cc555SAl Viro error = -EISDIR; 2638050ac841SMiklos Szeredi if ((open_flag & O_CREAT) && S_ISDIR(nd->inode->i_mode)) 26392675a4ebSAl Viro goto out; 2640af2f5542SMiklos Szeredi error = -ENOTDIR; 2641af2f5542SMiklos Szeredi if ((nd->flags & LOOKUP_DIRECTORY) && !nd->inode->i_op->lookup) 26422675a4ebSAl Viro goto out; 2643d7fdd7f6SMiklos Szeredi audit_inode(pathname, nd->path.dentry); 2644e83db167SMiklos Szeredi finish_open: 26456c0d46c4SAl Viro if (!S_ISREG(nd->inode->i_mode)) 264677d660a8SMiklos Szeredi will_truncate = false; 26476c0d46c4SAl Viro 26480f9d1a10SAl Viro if (will_truncate) { 26490f9d1a10SAl Viro error = mnt_want_write(nd->path.mnt); 26500f9d1a10SAl Viro if (error) 26512675a4ebSAl Viro goto out; 265277d660a8SMiklos Szeredi want_write = true; 26530f9d1a10SAl Viro } 2654e83db167SMiklos Szeredi finish_open_created: 2655bcda7652SAl Viro error = may_open(&nd->path, acc_mode, open_flag); 2656ca344a89SAl Viro if (error) 26572675a4ebSAl Viro goto out; 265830d90494SAl Viro file->f_path.mnt = nd->path.mnt; 265930d90494SAl Viro error = finish_open(file, nd->path.dentry, NULL, opened); 266030d90494SAl Viro if (error) { 266130d90494SAl Viro if (error == -EOPENSTALE) 2662f60dc3dbSMiklos Szeredi goto stale_open; 2663015c3bbcSMiklos Szeredi goto out; 2664f60dc3dbSMiklos Szeredi } 2665a8277b9bSMiklos Szeredi opened: 26662675a4ebSAl Viro error = open_check_o_direct(file); 2667015c3bbcSMiklos Szeredi if (error) 2668015c3bbcSMiklos Szeredi goto exit_fput; 26692675a4ebSAl Viro error = ima_file_check(file, op->acc_mode); 2670aa4caadbSMiklos Szeredi if (error) 2671aa4caadbSMiklos Szeredi goto exit_fput; 2672aa4caadbSMiklos Szeredi 26730f9d1a10SAl Viro if (will_truncate) { 26742675a4ebSAl Viro error = handle_truncate(file); 2675aa4caadbSMiklos Szeredi if (error) 2676aa4caadbSMiklos Szeredi goto exit_fput; 26770f9d1a10SAl Viro } 2678ca344a89SAl Viro out: 2679ca344a89SAl Viro if (want_write) 26800f9d1a10SAl Viro mnt_drop_write(nd->path.mnt); 268116b1c1cdSMiklos Szeredi path_put(&save_parent); 2682e276ae67SMiklos Szeredi terminate_walk(nd); 26832675a4ebSAl Viro return error; 2684fb1cc555SAl Viro 2685fb1cc555SAl Viro exit_dput: 2686fb1cc555SAl Viro path_put_conditional(path, nd); 2687ca344a89SAl Viro goto out; 2688015c3bbcSMiklos Szeredi exit_fput: 26892675a4ebSAl Viro fput(file); 26902675a4ebSAl Viro goto out; 2691015c3bbcSMiklos Szeredi 2692f60dc3dbSMiklos Szeredi stale_open: 2693f60dc3dbSMiklos Szeredi /* If no saved parent or already retried then can't retry */ 2694f60dc3dbSMiklos Szeredi if (!save_parent.dentry || retried) 2695f60dc3dbSMiklos Szeredi goto out; 2696f60dc3dbSMiklos Szeredi 2697f60dc3dbSMiklos Szeredi BUG_ON(save_parent.dentry != dir); 2698f60dc3dbSMiklos Szeredi path_put(&nd->path); 2699f60dc3dbSMiklos Szeredi nd->path = save_parent; 2700f60dc3dbSMiklos Szeredi nd->inode = dir->d_inode; 2701f60dc3dbSMiklos Szeredi save_parent.mnt = NULL; 2702f60dc3dbSMiklos Szeredi save_parent.dentry = NULL; 2703f60dc3dbSMiklos Szeredi if (want_write) { 2704f60dc3dbSMiklos Szeredi mnt_drop_write(nd->path.mnt); 2705f60dc3dbSMiklos Szeredi want_write = false; 2706f60dc3dbSMiklos Szeredi } 2707f60dc3dbSMiklos Szeredi retried = true; 2708f60dc3dbSMiklos Szeredi goto retry_lookup; 2709fb1cc555SAl Viro } 2710fb1cc555SAl Viro 271113aab428SAl Viro static struct file *path_openat(int dfd, const char *pathname, 271273d049a4SAl Viro struct nameidata *nd, const struct open_flags *op, int flags) 27131da177e4SLinus Torvalds { 2714fe2d35ffSAl Viro struct file *base = NULL; 271530d90494SAl Viro struct file *file; 27169850c056SAl Viro struct path path; 271747237687SAl Viro int opened = 0; 271813aab428SAl Viro int error; 271931e6b01fSNick Piggin 272030d90494SAl Viro file = get_empty_filp(); 272130d90494SAl Viro if (!file) 272231e6b01fSNick Piggin return ERR_PTR(-ENFILE); 272331e6b01fSNick Piggin 272430d90494SAl Viro file->f_flags = op->open_flag; 272531e6b01fSNick Piggin 272673d049a4SAl Viro error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base); 272731e6b01fSNick Piggin if (unlikely(error)) 27282675a4ebSAl Viro goto out; 272931e6b01fSNick Piggin 2730fe2d35ffSAl Viro current->total_link_count = 0; 273173d049a4SAl Viro error = link_path_walk(pathname, nd); 273231e6b01fSNick Piggin if (unlikely(error)) 27332675a4ebSAl Viro goto out; 27341da177e4SLinus Torvalds 27352675a4ebSAl Viro error = do_last(nd, &path, file, op, &opened, pathname); 27362675a4ebSAl Viro while (unlikely(error > 0)) { /* trailing symlink */ 27377b9337aaSNick Piggin struct path link = path; 2738def4af30SAl Viro void *cookie; 2739574197e0SAl Viro if (!(nd->flags & LOOKUP_FOLLOW)) { 274073d049a4SAl Viro path_put_conditional(&path, nd); 274173d049a4SAl Viro path_put(&nd->path); 27422675a4ebSAl Viro error = -ELOOP; 274340b39136SAl Viro break; 274440b39136SAl Viro } 274573d049a4SAl Viro nd->flags |= LOOKUP_PARENT; 274673d049a4SAl Viro nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); 2747574197e0SAl Viro error = follow_link(&link, nd, &cookie); 2748c3e380b0SAl Viro if (unlikely(error)) 27492675a4ebSAl Viro break; 27502675a4ebSAl Viro error = do_last(nd, &path, file, op, &opened, pathname); 2751574197e0SAl Viro put_link(nd, &link, cookie); 2752806b681cSAl Viro } 275310fa8e62SAl Viro out: 275473d049a4SAl Viro if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) 275573d049a4SAl Viro path_put(&nd->root); 2756fe2d35ffSAl Viro if (base) 2757fe2d35ffSAl Viro fput(base); 27582675a4ebSAl Viro if (!(opened & FILE_OPENED)) { 27592675a4ebSAl Viro BUG_ON(!error); 276030d90494SAl Viro put_filp(file); 2761015c3bbcSMiklos Szeredi } 27622675a4ebSAl Viro if (unlikely(error)) { 27632675a4ebSAl Viro if (error == -EOPENSTALE) { 27642675a4ebSAl Viro if (flags & LOOKUP_RCU) 27652675a4ebSAl Viro error = -ECHILD; 27662675a4ebSAl Viro else 27672675a4ebSAl Viro error = -ESTALE; 27682675a4ebSAl Viro } 27692675a4ebSAl Viro file = ERR_PTR(error); 27702675a4ebSAl Viro } 27712675a4ebSAl Viro return file; 2772de459215SKirill Korotaev } 27731da177e4SLinus Torvalds 277413aab428SAl Viro struct file *do_filp_open(int dfd, const char *pathname, 277513aab428SAl Viro const struct open_flags *op, int flags) 277613aab428SAl Viro { 277773d049a4SAl Viro struct nameidata nd; 277813aab428SAl Viro struct file *filp; 277913aab428SAl Viro 278073d049a4SAl Viro filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU); 278113aab428SAl Viro if (unlikely(filp == ERR_PTR(-ECHILD))) 278273d049a4SAl Viro filp = path_openat(dfd, pathname, &nd, op, flags); 278313aab428SAl Viro if (unlikely(filp == ERR_PTR(-ESTALE))) 278473d049a4SAl Viro filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL); 278513aab428SAl Viro return filp; 278613aab428SAl Viro } 278713aab428SAl Viro 278873d049a4SAl Viro struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, 278973d049a4SAl Viro const char *name, const struct open_flags *op, int flags) 279073d049a4SAl Viro { 279173d049a4SAl Viro struct nameidata nd; 279273d049a4SAl Viro struct file *file; 279373d049a4SAl Viro 279473d049a4SAl Viro nd.root.mnt = mnt; 279573d049a4SAl Viro nd.root.dentry = dentry; 279673d049a4SAl Viro 279773d049a4SAl Viro flags |= LOOKUP_ROOT; 279873d049a4SAl Viro 2799bcda7652SAl Viro if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN) 280073d049a4SAl Viro return ERR_PTR(-ELOOP); 280173d049a4SAl Viro 280273d049a4SAl Viro file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU); 280373d049a4SAl Viro if (unlikely(file == ERR_PTR(-ECHILD))) 280473d049a4SAl Viro file = path_openat(-1, name, &nd, op, flags); 280573d049a4SAl Viro if (unlikely(file == ERR_PTR(-ESTALE))) 280673d049a4SAl Viro file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL); 280773d049a4SAl Viro return file; 280873d049a4SAl Viro } 280973d049a4SAl Viro 2810ed75e95dSAl Viro struct dentry *kern_path_create(int dfd, const char *pathname, struct path *path, int is_dir) 28111da177e4SLinus Torvalds { 2812c663e5d8SChristoph Hellwig struct dentry *dentry = ERR_PTR(-EEXIST); 2813ed75e95dSAl Viro struct nameidata nd; 2814ed75e95dSAl Viro int error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd); 2815ed75e95dSAl Viro if (error) 2816ed75e95dSAl Viro return ERR_PTR(error); 28171da177e4SLinus Torvalds 2818c663e5d8SChristoph Hellwig /* 2819c663e5d8SChristoph Hellwig * Yucky last component or no last component at all? 2820c663e5d8SChristoph Hellwig * (foo/., foo/.., /////) 2821c663e5d8SChristoph Hellwig */ 2822ed75e95dSAl Viro if (nd.last_type != LAST_NORM) 2823ed75e95dSAl Viro goto out; 2824ed75e95dSAl Viro nd.flags &= ~LOOKUP_PARENT; 2825ed75e95dSAl Viro nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL; 2826c663e5d8SChristoph Hellwig 2827c663e5d8SChristoph Hellwig /* 2828c663e5d8SChristoph Hellwig * Do the final lookup. 2829c663e5d8SChristoph Hellwig */ 2830ed75e95dSAl Viro mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 2831ed75e95dSAl Viro dentry = lookup_hash(&nd); 28321da177e4SLinus Torvalds if (IS_ERR(dentry)) 28331da177e4SLinus Torvalds goto fail; 2834c663e5d8SChristoph Hellwig 2835e9baf6e5SAl Viro if (dentry->d_inode) 2836e9baf6e5SAl Viro goto eexist; 2837c663e5d8SChristoph Hellwig /* 2838c663e5d8SChristoph Hellwig * Special case - lookup gave negative, but... we had foo/bar/ 2839c663e5d8SChristoph Hellwig * From the vfs_mknod() POV we just have a negative dentry - 2840c663e5d8SChristoph Hellwig * all is fine. Let's be bastards - you had / on the end, you've 2841c663e5d8SChristoph Hellwig * been asking for (non-existent) directory. -ENOENT for you. 2842c663e5d8SChristoph Hellwig */ 2843ed75e95dSAl Viro if (unlikely(!is_dir && nd.last.name[nd.last.len])) { 28441da177e4SLinus Torvalds dput(dentry); 28451da177e4SLinus Torvalds dentry = ERR_PTR(-ENOENT); 2846ed75e95dSAl Viro goto fail; 2847e9baf6e5SAl Viro } 2848ed75e95dSAl Viro *path = nd.path; 2849e9baf6e5SAl Viro return dentry; 2850e9baf6e5SAl Viro eexist: 2851e9baf6e5SAl Viro dput(dentry); 2852e9baf6e5SAl Viro dentry = ERR_PTR(-EEXIST); 28531da177e4SLinus Torvalds fail: 2854dae6ad8fSAl Viro mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 2855ed75e95dSAl Viro out: 2856dae6ad8fSAl Viro path_put(&nd.path); 2857ed75e95dSAl Viro return dentry; 2858dae6ad8fSAl Viro } 2859dae6ad8fSAl Viro EXPORT_SYMBOL(kern_path_create); 2860dae6ad8fSAl Viro 2861dae6ad8fSAl Viro struct dentry *user_path_create(int dfd, const char __user *pathname, struct path *path, int is_dir) 2862dae6ad8fSAl Viro { 2863dae6ad8fSAl Viro char *tmp = getname(pathname); 2864dae6ad8fSAl Viro struct dentry *res; 2865dae6ad8fSAl Viro if (IS_ERR(tmp)) 2866dae6ad8fSAl Viro return ERR_CAST(tmp); 2867dae6ad8fSAl Viro res = kern_path_create(dfd, tmp, path, is_dir); 2868dae6ad8fSAl Viro putname(tmp); 2869dae6ad8fSAl Viro return res; 2870dae6ad8fSAl Viro } 2871dae6ad8fSAl Viro EXPORT_SYMBOL(user_path_create); 2872dae6ad8fSAl Viro 28731a67aafbSAl Viro int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) 28741da177e4SLinus Torvalds { 2875a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 28761da177e4SLinus Torvalds 28771da177e4SLinus Torvalds if (error) 28781da177e4SLinus Torvalds return error; 28791da177e4SLinus Torvalds 2880975d6b39SEric W. Biederman if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) 28811da177e4SLinus Torvalds return -EPERM; 28821da177e4SLinus Torvalds 2883acfa4380SAl Viro if (!dir->i_op->mknod) 28841da177e4SLinus Torvalds return -EPERM; 28851da177e4SLinus Torvalds 288608ce5f16SSerge E. Hallyn error = devcgroup_inode_mknod(mode, dev); 288708ce5f16SSerge E. Hallyn if (error) 288808ce5f16SSerge E. Hallyn return error; 288908ce5f16SSerge E. Hallyn 28901da177e4SLinus Torvalds error = security_inode_mknod(dir, dentry, mode, dev); 28911da177e4SLinus Torvalds if (error) 28921da177e4SLinus Torvalds return error; 28931da177e4SLinus Torvalds 28941da177e4SLinus Torvalds error = dir->i_op->mknod(dir, dentry, mode, dev); 2895a74574aaSStephen Smalley if (!error) 2896f38aa942SAmy Griffis fsnotify_create(dir, dentry); 28971da177e4SLinus Torvalds return error; 28981da177e4SLinus Torvalds } 28991da177e4SLinus Torvalds 2900f69aac00SAl Viro static int may_mknod(umode_t mode) 2901463c3197SDave Hansen { 2902463c3197SDave Hansen switch (mode & S_IFMT) { 2903463c3197SDave Hansen case S_IFREG: 2904463c3197SDave Hansen case S_IFCHR: 2905463c3197SDave Hansen case S_IFBLK: 2906463c3197SDave Hansen case S_IFIFO: 2907463c3197SDave Hansen case S_IFSOCK: 2908463c3197SDave Hansen case 0: /* zero mode translates to S_IFREG */ 2909463c3197SDave Hansen return 0; 2910463c3197SDave Hansen case S_IFDIR: 2911463c3197SDave Hansen return -EPERM; 2912463c3197SDave Hansen default: 2913463c3197SDave Hansen return -EINVAL; 2914463c3197SDave Hansen } 2915463c3197SDave Hansen } 2916463c3197SDave Hansen 29178208a22bSAl Viro SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, 29182e4d0924SHeiko Carstens unsigned, dev) 29191da177e4SLinus Torvalds { 29201da177e4SLinus Torvalds struct dentry *dentry; 2921dae6ad8fSAl Viro struct path path; 2922dae6ad8fSAl Viro int error; 29231da177e4SLinus Torvalds 29241da177e4SLinus Torvalds if (S_ISDIR(mode)) 29251da177e4SLinus Torvalds return -EPERM; 29261da177e4SLinus Torvalds 2927dae6ad8fSAl Viro dentry = user_path_create(dfd, filename, &path, 0); 2928dae6ad8fSAl Viro if (IS_ERR(dentry)) 2929dae6ad8fSAl Viro return PTR_ERR(dentry); 29302ad94ae6SAl Viro 2931dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 2932ce3b0f8dSAl Viro mode &= ~current_umask(); 2933463c3197SDave Hansen error = may_mknod(mode); 2934463c3197SDave Hansen if (error) 2935463c3197SDave Hansen goto out_dput; 2936dae6ad8fSAl Viro error = mnt_want_write(path.mnt); 2937463c3197SDave Hansen if (error) 2938463c3197SDave Hansen goto out_dput; 2939dae6ad8fSAl Viro error = security_path_mknod(&path, dentry, mode, dev); 2940be6d3e56SKentaro Takeda if (error) 2941be6d3e56SKentaro Takeda goto out_drop_write; 29421da177e4SLinus Torvalds switch (mode & S_IFMT) { 29431da177e4SLinus Torvalds case 0: case S_IFREG: 2944312b63fbSAl Viro error = vfs_create(path.dentry->d_inode,dentry,mode,true); 29451da177e4SLinus Torvalds break; 29461da177e4SLinus Torvalds case S_IFCHR: case S_IFBLK: 2947dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode, 29481da177e4SLinus Torvalds new_decode_dev(dev)); 29491da177e4SLinus Torvalds break; 29501da177e4SLinus Torvalds case S_IFIFO: case S_IFSOCK: 2951dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); 29521da177e4SLinus Torvalds break; 29531da177e4SLinus Torvalds } 2954be6d3e56SKentaro Takeda out_drop_write: 2955dae6ad8fSAl Viro mnt_drop_write(path.mnt); 2956463c3197SDave Hansen out_dput: 29571da177e4SLinus Torvalds dput(dentry); 2958dae6ad8fSAl Viro mutex_unlock(&path.dentry->d_inode->i_mutex); 2959dae6ad8fSAl Viro path_put(&path); 29601da177e4SLinus Torvalds 29611da177e4SLinus Torvalds return error; 29621da177e4SLinus Torvalds } 29631da177e4SLinus Torvalds 29648208a22bSAl Viro SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) 29655590ff0dSUlrich Drepper { 29665590ff0dSUlrich Drepper return sys_mknodat(AT_FDCWD, filename, mode, dev); 29675590ff0dSUlrich Drepper } 29685590ff0dSUlrich Drepper 296918bb1db3SAl Viro int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 29701da177e4SLinus Torvalds { 2971a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 29728de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 29731da177e4SLinus Torvalds 29741da177e4SLinus Torvalds if (error) 29751da177e4SLinus Torvalds return error; 29761da177e4SLinus Torvalds 2977acfa4380SAl Viro if (!dir->i_op->mkdir) 29781da177e4SLinus Torvalds return -EPERM; 29791da177e4SLinus Torvalds 29801da177e4SLinus Torvalds mode &= (S_IRWXUGO|S_ISVTX); 29811da177e4SLinus Torvalds error = security_inode_mkdir(dir, dentry, mode); 29821da177e4SLinus Torvalds if (error) 29831da177e4SLinus Torvalds return error; 29841da177e4SLinus Torvalds 29858de52778SAl Viro if (max_links && dir->i_nlink >= max_links) 29868de52778SAl Viro return -EMLINK; 29878de52778SAl Viro 29881da177e4SLinus Torvalds error = dir->i_op->mkdir(dir, dentry, mode); 2989a74574aaSStephen Smalley if (!error) 2990f38aa942SAmy Griffis fsnotify_mkdir(dir, dentry); 29911da177e4SLinus Torvalds return error; 29921da177e4SLinus Torvalds } 29931da177e4SLinus Torvalds 2994a218d0fdSAl Viro SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) 29951da177e4SLinus Torvalds { 29966902d925SDave Hansen struct dentry *dentry; 2997dae6ad8fSAl Viro struct path path; 2998dae6ad8fSAl Viro int error; 29991da177e4SLinus Torvalds 3000dae6ad8fSAl Viro dentry = user_path_create(dfd, pathname, &path, 1); 30016902d925SDave Hansen if (IS_ERR(dentry)) 3002dae6ad8fSAl Viro return PTR_ERR(dentry); 30036902d925SDave Hansen 3004dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3005ce3b0f8dSAl Viro mode &= ~current_umask(); 3006dae6ad8fSAl Viro error = mnt_want_write(path.mnt); 3007463c3197SDave Hansen if (error) 3008463c3197SDave Hansen goto out_dput; 3009dae6ad8fSAl Viro error = security_path_mkdir(&path, dentry, mode); 3010be6d3e56SKentaro Takeda if (error) 3011be6d3e56SKentaro Takeda goto out_drop_write; 3012dae6ad8fSAl Viro error = vfs_mkdir(path.dentry->d_inode, dentry, mode); 3013be6d3e56SKentaro Takeda out_drop_write: 3014dae6ad8fSAl Viro mnt_drop_write(path.mnt); 3015463c3197SDave Hansen out_dput: 30161da177e4SLinus Torvalds dput(dentry); 3017dae6ad8fSAl Viro mutex_unlock(&path.dentry->d_inode->i_mutex); 3018dae6ad8fSAl Viro path_put(&path); 30191da177e4SLinus Torvalds return error; 30201da177e4SLinus Torvalds } 30211da177e4SLinus Torvalds 3022a218d0fdSAl Viro SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) 30235590ff0dSUlrich Drepper { 30245590ff0dSUlrich Drepper return sys_mkdirat(AT_FDCWD, pathname, mode); 30255590ff0dSUlrich Drepper } 30265590ff0dSUlrich Drepper 30271da177e4SLinus Torvalds /* 3028a71905f0SSage Weil * The dentry_unhash() helper will try to drop the dentry early: we 3029c0d02594SJ. Bruce Fields * should have a usage count of 1 if we're the only user of this 3030a71905f0SSage Weil * dentry, and if that is true (possibly after pruning the dcache), 3031a71905f0SSage Weil * then we drop the dentry now. 30321da177e4SLinus Torvalds * 30331da177e4SLinus Torvalds * A low-level filesystem can, if it choses, legally 30341da177e4SLinus Torvalds * do a 30351da177e4SLinus Torvalds * 30361da177e4SLinus Torvalds * if (!d_unhashed(dentry)) 30371da177e4SLinus Torvalds * return -EBUSY; 30381da177e4SLinus Torvalds * 30391da177e4SLinus Torvalds * if it cannot handle the case of removing a directory 30401da177e4SLinus Torvalds * that is still in use by something else.. 30411da177e4SLinus Torvalds */ 30421da177e4SLinus Torvalds void dentry_unhash(struct dentry *dentry) 30431da177e4SLinus Torvalds { 30441da177e4SLinus Torvalds shrink_dcache_parent(dentry); 30451da177e4SLinus Torvalds spin_lock(&dentry->d_lock); 304664252c75SSage Weil if (dentry->d_count == 1) 30471da177e4SLinus Torvalds __d_drop(dentry); 30481da177e4SLinus Torvalds spin_unlock(&dentry->d_lock); 30491da177e4SLinus Torvalds } 30501da177e4SLinus Torvalds 30511da177e4SLinus Torvalds int vfs_rmdir(struct inode *dir, struct dentry *dentry) 30521da177e4SLinus Torvalds { 30531da177e4SLinus Torvalds int error = may_delete(dir, dentry, 1); 30541da177e4SLinus Torvalds 30551da177e4SLinus Torvalds if (error) 30561da177e4SLinus Torvalds return error; 30571da177e4SLinus Torvalds 3058acfa4380SAl Viro if (!dir->i_op->rmdir) 30591da177e4SLinus Torvalds return -EPERM; 30601da177e4SLinus Torvalds 30611d2ef590SAl Viro dget(dentry); 30621b1dcc1bSJes Sorensen mutex_lock(&dentry->d_inode->i_mutex); 3063912dbc15SSage Weil 30641da177e4SLinus Torvalds error = -EBUSY; 3065912dbc15SSage Weil if (d_mountpoint(dentry)) 3066912dbc15SSage Weil goto out; 3067912dbc15SSage Weil 30681da177e4SLinus Torvalds error = security_inode_rmdir(dir, dentry); 3069912dbc15SSage Weil if (error) 3070912dbc15SSage Weil goto out; 3071912dbc15SSage Weil 30723cebde24SSage Weil shrink_dcache_parent(dentry); 30731da177e4SLinus Torvalds error = dir->i_op->rmdir(dir, dentry); 3074912dbc15SSage Weil if (error) 3075912dbc15SSage Weil goto out; 3076912dbc15SSage Weil 30771da177e4SLinus Torvalds dentry->d_inode->i_flags |= S_DEAD; 3078d83c49f3SAl Viro dont_mount(dentry); 30791da177e4SLinus Torvalds 3080912dbc15SSage Weil out: 3081912dbc15SSage Weil mutex_unlock(&dentry->d_inode->i_mutex); 30821d2ef590SAl Viro dput(dentry); 3083912dbc15SSage Weil if (!error) 3084912dbc15SSage Weil d_delete(dentry); 30851da177e4SLinus Torvalds return error; 30861da177e4SLinus Torvalds } 30871da177e4SLinus Torvalds 30885590ff0dSUlrich Drepper static long do_rmdir(int dfd, const char __user *pathname) 30891da177e4SLinus Torvalds { 30901da177e4SLinus Torvalds int error = 0; 30911da177e4SLinus Torvalds char * name; 30921da177e4SLinus Torvalds struct dentry *dentry; 30931da177e4SLinus Torvalds struct nameidata nd; 30941da177e4SLinus Torvalds 30952ad94ae6SAl Viro error = user_path_parent(dfd, pathname, &nd, &name); 30961da177e4SLinus Torvalds if (error) 30972ad94ae6SAl Viro return error; 30981da177e4SLinus Torvalds 30991da177e4SLinus Torvalds switch(nd.last_type) { 31001da177e4SLinus Torvalds case LAST_DOTDOT: 31011da177e4SLinus Torvalds error = -ENOTEMPTY; 31021da177e4SLinus Torvalds goto exit1; 31031da177e4SLinus Torvalds case LAST_DOT: 31041da177e4SLinus Torvalds error = -EINVAL; 31051da177e4SLinus Torvalds goto exit1; 31061da177e4SLinus Torvalds case LAST_ROOT: 31071da177e4SLinus Torvalds error = -EBUSY; 31081da177e4SLinus Torvalds goto exit1; 31091da177e4SLinus Torvalds } 31100612d9fbSOGAWA Hirofumi 31110612d9fbSOGAWA Hirofumi nd.flags &= ~LOOKUP_PARENT; 31120612d9fbSOGAWA Hirofumi 31134ac91378SJan Blunck mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 311449705b77SChristoph Hellwig dentry = lookup_hash(&nd); 31151da177e4SLinus Torvalds error = PTR_ERR(dentry); 31166902d925SDave Hansen if (IS_ERR(dentry)) 31176902d925SDave Hansen goto exit2; 3118e6bc45d6STheodore Ts'o if (!dentry->d_inode) { 3119e6bc45d6STheodore Ts'o error = -ENOENT; 3120e6bc45d6STheodore Ts'o goto exit3; 3121e6bc45d6STheodore Ts'o } 31220622753bSDave Hansen error = mnt_want_write(nd.path.mnt); 31230622753bSDave Hansen if (error) 31240622753bSDave Hansen goto exit3; 3125be6d3e56SKentaro Takeda error = security_path_rmdir(&nd.path, dentry); 3126be6d3e56SKentaro Takeda if (error) 3127be6d3e56SKentaro Takeda goto exit4; 31284ac91378SJan Blunck error = vfs_rmdir(nd.path.dentry->d_inode, dentry); 3129be6d3e56SKentaro Takeda exit4: 31300622753bSDave Hansen mnt_drop_write(nd.path.mnt); 31310622753bSDave Hansen exit3: 31321da177e4SLinus Torvalds dput(dentry); 31336902d925SDave Hansen exit2: 31344ac91378SJan Blunck mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 31351da177e4SLinus Torvalds exit1: 31361d957f9bSJan Blunck path_put(&nd.path); 31371da177e4SLinus Torvalds putname(name); 31381da177e4SLinus Torvalds return error; 31391da177e4SLinus Torvalds } 31401da177e4SLinus Torvalds 31413cdad428SHeiko Carstens SYSCALL_DEFINE1(rmdir, const char __user *, pathname) 31425590ff0dSUlrich Drepper { 31435590ff0dSUlrich Drepper return do_rmdir(AT_FDCWD, pathname); 31445590ff0dSUlrich Drepper } 31455590ff0dSUlrich Drepper 31461da177e4SLinus Torvalds int vfs_unlink(struct inode *dir, struct dentry *dentry) 31471da177e4SLinus Torvalds { 31481da177e4SLinus Torvalds int error = may_delete(dir, dentry, 0); 31491da177e4SLinus Torvalds 31501da177e4SLinus Torvalds if (error) 31511da177e4SLinus Torvalds return error; 31521da177e4SLinus Torvalds 3153acfa4380SAl Viro if (!dir->i_op->unlink) 31541da177e4SLinus Torvalds return -EPERM; 31551da177e4SLinus Torvalds 31561b1dcc1bSJes Sorensen mutex_lock(&dentry->d_inode->i_mutex); 31571da177e4SLinus Torvalds if (d_mountpoint(dentry)) 31581da177e4SLinus Torvalds error = -EBUSY; 31591da177e4SLinus Torvalds else { 31601da177e4SLinus Torvalds error = security_inode_unlink(dir, dentry); 3161bec1052eSAl Viro if (!error) { 31621da177e4SLinus Torvalds error = dir->i_op->unlink(dir, dentry); 3163bec1052eSAl Viro if (!error) 3164d83c49f3SAl Viro dont_mount(dentry); 3165bec1052eSAl Viro } 31661da177e4SLinus Torvalds } 31671b1dcc1bSJes Sorensen mutex_unlock(&dentry->d_inode->i_mutex); 31681da177e4SLinus Torvalds 31691da177e4SLinus Torvalds /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 31701da177e4SLinus Torvalds if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { 3171ece95912SJan Kara fsnotify_link_count(dentry->d_inode); 31721da177e4SLinus Torvalds d_delete(dentry); 31731da177e4SLinus Torvalds } 31740eeca283SRobert Love 31751da177e4SLinus Torvalds return error; 31761da177e4SLinus Torvalds } 31771da177e4SLinus Torvalds 31781da177e4SLinus Torvalds /* 31791da177e4SLinus Torvalds * Make sure that the actual truncation of the file will occur outside its 31801b1dcc1bSJes Sorensen * directory's i_mutex. Truncate can take a long time if there is a lot of 31811da177e4SLinus Torvalds * writeout happening, and we don't want to prevent access to the directory 31821da177e4SLinus Torvalds * while waiting on the I/O. 31831da177e4SLinus Torvalds */ 31845590ff0dSUlrich Drepper static long do_unlinkat(int dfd, const char __user *pathname) 31851da177e4SLinus Torvalds { 31862ad94ae6SAl Viro int error; 31871da177e4SLinus Torvalds char *name; 31881da177e4SLinus Torvalds struct dentry *dentry; 31891da177e4SLinus Torvalds struct nameidata nd; 31901da177e4SLinus Torvalds struct inode *inode = NULL; 31911da177e4SLinus Torvalds 31922ad94ae6SAl Viro error = user_path_parent(dfd, pathname, &nd, &name); 31931da177e4SLinus Torvalds if (error) 31942ad94ae6SAl Viro return error; 31952ad94ae6SAl Viro 31961da177e4SLinus Torvalds error = -EISDIR; 31971da177e4SLinus Torvalds if (nd.last_type != LAST_NORM) 31981da177e4SLinus Torvalds goto exit1; 31990612d9fbSOGAWA Hirofumi 32000612d9fbSOGAWA Hirofumi nd.flags &= ~LOOKUP_PARENT; 32010612d9fbSOGAWA Hirofumi 32024ac91378SJan Blunck mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 320349705b77SChristoph Hellwig dentry = lookup_hash(&nd); 32041da177e4SLinus Torvalds error = PTR_ERR(dentry); 32051da177e4SLinus Torvalds if (!IS_ERR(dentry)) { 32061da177e4SLinus Torvalds /* Why not before? Because we want correct error value */ 320750338b88STörök Edwin if (nd.last.name[nd.last.len]) 320850338b88STörök Edwin goto slashes; 32091da177e4SLinus Torvalds inode = dentry->d_inode; 321050338b88STörök Edwin if (!inode) 3211e6bc45d6STheodore Ts'o goto slashes; 32127de9c6eeSAl Viro ihold(inode); 32130622753bSDave Hansen error = mnt_want_write(nd.path.mnt); 32140622753bSDave Hansen if (error) 32150622753bSDave Hansen goto exit2; 3216be6d3e56SKentaro Takeda error = security_path_unlink(&nd.path, dentry); 3217be6d3e56SKentaro Takeda if (error) 3218be6d3e56SKentaro Takeda goto exit3; 32194ac91378SJan Blunck error = vfs_unlink(nd.path.dentry->d_inode, dentry); 3220be6d3e56SKentaro Takeda exit3: 32210622753bSDave Hansen mnt_drop_write(nd.path.mnt); 32221da177e4SLinus Torvalds exit2: 32231da177e4SLinus Torvalds dput(dentry); 32241da177e4SLinus Torvalds } 32254ac91378SJan Blunck mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 32261da177e4SLinus Torvalds if (inode) 32271da177e4SLinus Torvalds iput(inode); /* truncate the inode here */ 32281da177e4SLinus Torvalds exit1: 32291d957f9bSJan Blunck path_put(&nd.path); 32301da177e4SLinus Torvalds putname(name); 32311da177e4SLinus Torvalds return error; 32321da177e4SLinus Torvalds 32331da177e4SLinus Torvalds slashes: 32341da177e4SLinus Torvalds error = !dentry->d_inode ? -ENOENT : 32351da177e4SLinus Torvalds S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR; 32361da177e4SLinus Torvalds goto exit2; 32371da177e4SLinus Torvalds } 32381da177e4SLinus Torvalds 32392e4d0924SHeiko Carstens SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) 32405590ff0dSUlrich Drepper { 32415590ff0dSUlrich Drepper if ((flag & ~AT_REMOVEDIR) != 0) 32425590ff0dSUlrich Drepper return -EINVAL; 32435590ff0dSUlrich Drepper 32445590ff0dSUlrich Drepper if (flag & AT_REMOVEDIR) 32455590ff0dSUlrich Drepper return do_rmdir(dfd, pathname); 32465590ff0dSUlrich Drepper 32475590ff0dSUlrich Drepper return do_unlinkat(dfd, pathname); 32485590ff0dSUlrich Drepper } 32495590ff0dSUlrich Drepper 32503480b257SHeiko Carstens SYSCALL_DEFINE1(unlink, const char __user *, pathname) 32515590ff0dSUlrich Drepper { 32525590ff0dSUlrich Drepper return do_unlinkat(AT_FDCWD, pathname); 32535590ff0dSUlrich Drepper } 32545590ff0dSUlrich Drepper 3255db2e747bSMiklos Szeredi int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) 32561da177e4SLinus Torvalds { 3257a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 32581da177e4SLinus Torvalds 32591da177e4SLinus Torvalds if (error) 32601da177e4SLinus Torvalds return error; 32611da177e4SLinus Torvalds 3262acfa4380SAl Viro if (!dir->i_op->symlink) 32631da177e4SLinus Torvalds return -EPERM; 32641da177e4SLinus Torvalds 32651da177e4SLinus Torvalds error = security_inode_symlink(dir, dentry, oldname); 32661da177e4SLinus Torvalds if (error) 32671da177e4SLinus Torvalds return error; 32681da177e4SLinus Torvalds 32691da177e4SLinus Torvalds error = dir->i_op->symlink(dir, dentry, oldname); 3270a74574aaSStephen Smalley if (!error) 3271f38aa942SAmy Griffis fsnotify_create(dir, dentry); 32721da177e4SLinus Torvalds return error; 32731da177e4SLinus Torvalds } 32741da177e4SLinus Torvalds 32752e4d0924SHeiko Carstens SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, 32762e4d0924SHeiko Carstens int, newdfd, const char __user *, newname) 32771da177e4SLinus Torvalds { 32782ad94ae6SAl Viro int error; 32791da177e4SLinus Torvalds char *from; 32806902d925SDave Hansen struct dentry *dentry; 3281dae6ad8fSAl Viro struct path path; 32821da177e4SLinus Torvalds 32831da177e4SLinus Torvalds from = getname(oldname); 32841da177e4SLinus Torvalds if (IS_ERR(from)) 32851da177e4SLinus Torvalds return PTR_ERR(from); 32862ad94ae6SAl Viro 3287dae6ad8fSAl Viro dentry = user_path_create(newdfd, newname, &path, 0); 32881da177e4SLinus Torvalds error = PTR_ERR(dentry); 32896902d925SDave Hansen if (IS_ERR(dentry)) 3290dae6ad8fSAl Viro goto out_putname; 32916902d925SDave Hansen 3292dae6ad8fSAl Viro error = mnt_want_write(path.mnt); 329375c3f29dSDave Hansen if (error) 329475c3f29dSDave Hansen goto out_dput; 3295dae6ad8fSAl Viro error = security_path_symlink(&path, dentry, from); 3296be6d3e56SKentaro Takeda if (error) 3297be6d3e56SKentaro Takeda goto out_drop_write; 3298dae6ad8fSAl Viro error = vfs_symlink(path.dentry->d_inode, dentry, from); 3299be6d3e56SKentaro Takeda out_drop_write: 3300dae6ad8fSAl Viro mnt_drop_write(path.mnt); 330175c3f29dSDave Hansen out_dput: 33021da177e4SLinus Torvalds dput(dentry); 3303dae6ad8fSAl Viro mutex_unlock(&path.dentry->d_inode->i_mutex); 3304dae6ad8fSAl Viro path_put(&path); 33056902d925SDave Hansen out_putname: 33061da177e4SLinus Torvalds putname(from); 33071da177e4SLinus Torvalds return error; 33081da177e4SLinus Torvalds } 33091da177e4SLinus Torvalds 33103480b257SHeiko Carstens SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) 33115590ff0dSUlrich Drepper { 33125590ff0dSUlrich Drepper return sys_symlinkat(oldname, AT_FDCWD, newname); 33135590ff0dSUlrich Drepper } 33145590ff0dSUlrich Drepper 33151da177e4SLinus Torvalds int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry) 33161da177e4SLinus Torvalds { 33171da177e4SLinus Torvalds struct inode *inode = old_dentry->d_inode; 33188de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 33191da177e4SLinus Torvalds int error; 33201da177e4SLinus Torvalds 33211da177e4SLinus Torvalds if (!inode) 33221da177e4SLinus Torvalds return -ENOENT; 33231da177e4SLinus Torvalds 3324a95164d9SMiklos Szeredi error = may_create(dir, new_dentry); 33251da177e4SLinus Torvalds if (error) 33261da177e4SLinus Torvalds return error; 33271da177e4SLinus Torvalds 33281da177e4SLinus Torvalds if (dir->i_sb != inode->i_sb) 33291da177e4SLinus Torvalds return -EXDEV; 33301da177e4SLinus Torvalds 33311da177e4SLinus Torvalds /* 33321da177e4SLinus Torvalds * A link to an append-only or immutable file cannot be created. 33331da177e4SLinus Torvalds */ 33341da177e4SLinus Torvalds if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 33351da177e4SLinus Torvalds return -EPERM; 3336acfa4380SAl Viro if (!dir->i_op->link) 33371da177e4SLinus Torvalds return -EPERM; 33387e79eedbSTetsuo Handa if (S_ISDIR(inode->i_mode)) 33391da177e4SLinus Torvalds return -EPERM; 33401da177e4SLinus Torvalds 33411da177e4SLinus Torvalds error = security_inode_link(old_dentry, dir, new_dentry); 33421da177e4SLinus Torvalds if (error) 33431da177e4SLinus Torvalds return error; 33441da177e4SLinus Torvalds 33457e79eedbSTetsuo Handa mutex_lock(&inode->i_mutex); 3346aae8a97dSAneesh Kumar K.V /* Make sure we don't allow creating hardlink to an unlinked file */ 3347aae8a97dSAneesh Kumar K.V if (inode->i_nlink == 0) 3348aae8a97dSAneesh Kumar K.V error = -ENOENT; 33498de52778SAl Viro else if (max_links && inode->i_nlink >= max_links) 33508de52778SAl Viro error = -EMLINK; 3351aae8a97dSAneesh Kumar K.V else 33521da177e4SLinus Torvalds error = dir->i_op->link(old_dentry, dir, new_dentry); 33537e79eedbSTetsuo Handa mutex_unlock(&inode->i_mutex); 3354e31e14ecSStephen Smalley if (!error) 33557e79eedbSTetsuo Handa fsnotify_link(dir, inode, new_dentry); 33561da177e4SLinus Torvalds return error; 33571da177e4SLinus Torvalds } 33581da177e4SLinus Torvalds 33591da177e4SLinus Torvalds /* 33601da177e4SLinus Torvalds * Hardlinks are often used in delicate situations. We avoid 33611da177e4SLinus Torvalds * security-related surprises by not following symlinks on the 33621da177e4SLinus Torvalds * newname. --KAB 33631da177e4SLinus Torvalds * 33641da177e4SLinus Torvalds * We don't follow them on the oldname either to be compatible 33651da177e4SLinus Torvalds * with linux 2.0, and to avoid hard-linking to directories 33661da177e4SLinus Torvalds * and other special files. --ADM 33671da177e4SLinus Torvalds */ 33682e4d0924SHeiko Carstens SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, 33692e4d0924SHeiko Carstens int, newdfd, const char __user *, newname, int, flags) 33701da177e4SLinus Torvalds { 33711da177e4SLinus Torvalds struct dentry *new_dentry; 3372dae6ad8fSAl Viro struct path old_path, new_path; 337311a7b371SAneesh Kumar K.V int how = 0; 33741da177e4SLinus Torvalds int error; 33751da177e4SLinus Torvalds 337611a7b371SAneesh Kumar K.V if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) 3377c04030e1SUlrich Drepper return -EINVAL; 337811a7b371SAneesh Kumar K.V /* 337911a7b371SAneesh Kumar K.V * To use null names we require CAP_DAC_READ_SEARCH 338011a7b371SAneesh Kumar K.V * This ensures that not everyone will be able to create 338111a7b371SAneesh Kumar K.V * handlink using the passed filedescriptor. 338211a7b371SAneesh Kumar K.V */ 338311a7b371SAneesh Kumar K.V if (flags & AT_EMPTY_PATH) { 338411a7b371SAneesh Kumar K.V if (!capable(CAP_DAC_READ_SEARCH)) 338511a7b371SAneesh Kumar K.V return -ENOENT; 338611a7b371SAneesh Kumar K.V how = LOOKUP_EMPTY; 338711a7b371SAneesh Kumar K.V } 3388c04030e1SUlrich Drepper 338911a7b371SAneesh Kumar K.V if (flags & AT_SYMLINK_FOLLOW) 339011a7b371SAneesh Kumar K.V how |= LOOKUP_FOLLOW; 339111a7b371SAneesh Kumar K.V 339211a7b371SAneesh Kumar K.V error = user_path_at(olddfd, oldname, how, &old_path); 33931da177e4SLinus Torvalds if (error) 33942ad94ae6SAl Viro return error; 33952ad94ae6SAl Viro 3396dae6ad8fSAl Viro new_dentry = user_path_create(newdfd, newname, &new_path, 0); 33971da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 33986902d925SDave Hansen if (IS_ERR(new_dentry)) 3399dae6ad8fSAl Viro goto out; 3400dae6ad8fSAl Viro 3401dae6ad8fSAl Viro error = -EXDEV; 3402dae6ad8fSAl Viro if (old_path.mnt != new_path.mnt) 3403dae6ad8fSAl Viro goto out_dput; 3404dae6ad8fSAl Viro error = mnt_want_write(new_path.mnt); 340575c3f29dSDave Hansen if (error) 340675c3f29dSDave Hansen goto out_dput; 3407dae6ad8fSAl Viro error = security_path_link(old_path.dentry, &new_path, new_dentry); 3408be6d3e56SKentaro Takeda if (error) 3409be6d3e56SKentaro Takeda goto out_drop_write; 3410dae6ad8fSAl Viro error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry); 3411be6d3e56SKentaro Takeda out_drop_write: 3412dae6ad8fSAl Viro mnt_drop_write(new_path.mnt); 341375c3f29dSDave Hansen out_dput: 34141da177e4SLinus Torvalds dput(new_dentry); 3415dae6ad8fSAl Viro mutex_unlock(&new_path.dentry->d_inode->i_mutex); 3416dae6ad8fSAl Viro path_put(&new_path); 34171da177e4SLinus Torvalds out: 34182d8f3038SAl Viro path_put(&old_path); 34191da177e4SLinus Torvalds 34201da177e4SLinus Torvalds return error; 34211da177e4SLinus Torvalds } 34221da177e4SLinus Torvalds 34233480b257SHeiko Carstens SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) 34245590ff0dSUlrich Drepper { 3425c04030e1SUlrich Drepper return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 34265590ff0dSUlrich Drepper } 34275590ff0dSUlrich Drepper 34281da177e4SLinus Torvalds /* 34291da177e4SLinus Torvalds * The worst of all namespace operations - renaming directory. "Perverted" 34301da177e4SLinus Torvalds * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 34311da177e4SLinus Torvalds * Problems: 34321da177e4SLinus Torvalds * a) we can get into loop creation. Check is done in is_subdir(). 34331da177e4SLinus Torvalds * b) race potential - two innocent renames can create a loop together. 34341da177e4SLinus Torvalds * That's where 4.4 screws up. Current fix: serialization on 3435a11f3a05SArjan van de Ven * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 34361da177e4SLinus Torvalds * story. 34371da177e4SLinus Torvalds * c) we have to lock _three_ objects - parents and victim (if it exists). 34381b1dcc1bSJes Sorensen * And that - after we got ->i_mutex on parents (until then we don't know 34391da177e4SLinus Torvalds * whether the target exists). Solution: try to be smart with locking 34401da177e4SLinus Torvalds * order for inodes. We rely on the fact that tree topology may change 3441a11f3a05SArjan van de Ven * only under ->s_vfs_rename_mutex _and_ that parent of the object we 34421da177e4SLinus Torvalds * move will be locked. Thus we can rank directories by the tree 34431da177e4SLinus Torvalds * (ancestors first) and rank all non-directories after them. 34441da177e4SLinus Torvalds * That works since everybody except rename does "lock parent, lookup, 3445a11f3a05SArjan van de Ven * lock child" and rename is under ->s_vfs_rename_mutex. 34461da177e4SLinus Torvalds * HOWEVER, it relies on the assumption that any object with ->lookup() 34471da177e4SLinus Torvalds * has no more than 1 dentry. If "hybrid" objects will ever appear, 34481da177e4SLinus Torvalds * we'd better make sure that there's no link(2) for them. 3449e4eaac06SSage Weil * d) conversion from fhandle to dentry may come in the wrong moment - when 34501b1dcc1bSJes Sorensen * we are removing the target. Solution: we will have to grab ->i_mutex 34511da177e4SLinus Torvalds * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 3452c41b20e7SAdam Buchbinder * ->i_mutex on parents, which works but leads to some truly excessive 34531da177e4SLinus Torvalds * locking]. 34541da177e4SLinus Torvalds */ 345575c96f85SAdrian Bunk static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry, 34561da177e4SLinus Torvalds struct inode *new_dir, struct dentry *new_dentry) 34571da177e4SLinus Torvalds { 34581da177e4SLinus Torvalds int error = 0; 34599055cba7SSage Weil struct inode *target = new_dentry->d_inode; 34608de52778SAl Viro unsigned max_links = new_dir->i_sb->s_max_links; 34611da177e4SLinus Torvalds 34621da177e4SLinus Torvalds /* 34631da177e4SLinus Torvalds * If we are going to change the parent - check write permissions, 34641da177e4SLinus Torvalds * we'll need to flip '..'. 34651da177e4SLinus Torvalds */ 34661da177e4SLinus Torvalds if (new_dir != old_dir) { 3467f419a2e3SAl Viro error = inode_permission(old_dentry->d_inode, MAY_WRITE); 34681da177e4SLinus Torvalds if (error) 34691da177e4SLinus Torvalds return error; 34701da177e4SLinus Torvalds } 34711da177e4SLinus Torvalds 34721da177e4SLinus Torvalds error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); 34731da177e4SLinus Torvalds if (error) 34741da177e4SLinus Torvalds return error; 34751da177e4SLinus Torvalds 34761d2ef590SAl Viro dget(new_dentry); 3477d83c49f3SAl Viro if (target) 34781b1dcc1bSJes Sorensen mutex_lock(&target->i_mutex); 34799055cba7SSage Weil 34801da177e4SLinus Torvalds error = -EBUSY; 34819055cba7SSage Weil if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry)) 34829055cba7SSage Weil goto out; 34839055cba7SSage Weil 34848de52778SAl Viro error = -EMLINK; 34858de52778SAl Viro if (max_links && !target && new_dir != old_dir && 34868de52778SAl Viro new_dir->i_nlink >= max_links) 34878de52778SAl Viro goto out; 34888de52778SAl Viro 34893cebde24SSage Weil if (target) 34903cebde24SSage Weil shrink_dcache_parent(new_dentry); 34911da177e4SLinus Torvalds error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); 34929055cba7SSage Weil if (error) 34939055cba7SSage Weil goto out; 34949055cba7SSage Weil 34951da177e4SLinus Torvalds if (target) { 34961da177e4SLinus Torvalds target->i_flags |= S_DEAD; 3497d83c49f3SAl Viro dont_mount(new_dentry); 3498d83c49f3SAl Viro } 34999055cba7SSage Weil out: 35009055cba7SSage Weil if (target) 35011b1dcc1bSJes Sorensen mutex_unlock(&target->i_mutex); 35021d2ef590SAl Viro dput(new_dentry); 3503e31e14ecSStephen Smalley if (!error) 3504349457ccSMark Fasheh if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) 35051da177e4SLinus Torvalds d_move(old_dentry,new_dentry); 35061da177e4SLinus Torvalds return error; 35071da177e4SLinus Torvalds } 35081da177e4SLinus Torvalds 350975c96f85SAdrian Bunk static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry, 35101da177e4SLinus Torvalds struct inode *new_dir, struct dentry *new_dentry) 35111da177e4SLinus Torvalds { 351251892bbbSSage Weil struct inode *target = new_dentry->d_inode; 35131da177e4SLinus Torvalds int error; 35141da177e4SLinus Torvalds 35151da177e4SLinus Torvalds error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); 35161da177e4SLinus Torvalds if (error) 35171da177e4SLinus Torvalds return error; 35181da177e4SLinus Torvalds 35191da177e4SLinus Torvalds dget(new_dentry); 35201da177e4SLinus Torvalds if (target) 35211b1dcc1bSJes Sorensen mutex_lock(&target->i_mutex); 352251892bbbSSage Weil 35231da177e4SLinus Torvalds error = -EBUSY; 352451892bbbSSage Weil if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry)) 352551892bbbSSage Weil goto out; 352651892bbbSSage Weil 35271da177e4SLinus Torvalds error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); 352851892bbbSSage Weil if (error) 352951892bbbSSage Weil goto out; 353051892bbbSSage Weil 3531bec1052eSAl Viro if (target) 3532d83c49f3SAl Viro dont_mount(new_dentry); 3533349457ccSMark Fasheh if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) 35341da177e4SLinus Torvalds d_move(old_dentry, new_dentry); 353551892bbbSSage Weil out: 35361da177e4SLinus Torvalds if (target) 35371b1dcc1bSJes Sorensen mutex_unlock(&target->i_mutex); 35381da177e4SLinus Torvalds dput(new_dentry); 35391da177e4SLinus Torvalds return error; 35401da177e4SLinus Torvalds } 35411da177e4SLinus Torvalds 35421da177e4SLinus Torvalds int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, 35431da177e4SLinus Torvalds struct inode *new_dir, struct dentry *new_dentry) 35441da177e4SLinus Torvalds { 35451da177e4SLinus Torvalds int error; 35461da177e4SLinus Torvalds int is_dir = S_ISDIR(old_dentry->d_inode->i_mode); 354759b0df21SEric Paris const unsigned char *old_name; 35481da177e4SLinus Torvalds 35491da177e4SLinus Torvalds if (old_dentry->d_inode == new_dentry->d_inode) 35501da177e4SLinus Torvalds return 0; 35511da177e4SLinus Torvalds 35521da177e4SLinus Torvalds error = may_delete(old_dir, old_dentry, is_dir); 35531da177e4SLinus Torvalds if (error) 35541da177e4SLinus Torvalds return error; 35551da177e4SLinus Torvalds 35561da177e4SLinus Torvalds if (!new_dentry->d_inode) 3557a95164d9SMiklos Szeredi error = may_create(new_dir, new_dentry); 35581da177e4SLinus Torvalds else 35591da177e4SLinus Torvalds error = may_delete(new_dir, new_dentry, is_dir); 35601da177e4SLinus Torvalds if (error) 35611da177e4SLinus Torvalds return error; 35621da177e4SLinus Torvalds 3563acfa4380SAl Viro if (!old_dir->i_op->rename) 35641da177e4SLinus Torvalds return -EPERM; 35651da177e4SLinus Torvalds 35660eeca283SRobert Love old_name = fsnotify_oldname_init(old_dentry->d_name.name); 35670eeca283SRobert Love 35681da177e4SLinus Torvalds if (is_dir) 35691da177e4SLinus Torvalds error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry); 35701da177e4SLinus Torvalds else 35711da177e4SLinus Torvalds error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry); 3572123df294SAl Viro if (!error) 3573123df294SAl Viro fsnotify_move(old_dir, new_dir, old_name, is_dir, 35745a190ae6SAl Viro new_dentry->d_inode, old_dentry); 35750eeca283SRobert Love fsnotify_oldname_free(old_name); 35760eeca283SRobert Love 35771da177e4SLinus Torvalds return error; 35781da177e4SLinus Torvalds } 35791da177e4SLinus Torvalds 35802e4d0924SHeiko Carstens SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, 35812e4d0924SHeiko Carstens int, newdfd, const char __user *, newname) 35821da177e4SLinus Torvalds { 35831da177e4SLinus Torvalds struct dentry *old_dir, *new_dir; 35841da177e4SLinus Torvalds struct dentry *old_dentry, *new_dentry; 35851da177e4SLinus Torvalds struct dentry *trap; 35861da177e4SLinus Torvalds struct nameidata oldnd, newnd; 35872ad94ae6SAl Viro char *from; 35882ad94ae6SAl Viro char *to; 35892ad94ae6SAl Viro int error; 35901da177e4SLinus Torvalds 35912ad94ae6SAl Viro error = user_path_parent(olddfd, oldname, &oldnd, &from); 35921da177e4SLinus Torvalds if (error) 35931da177e4SLinus Torvalds goto exit; 35941da177e4SLinus Torvalds 35952ad94ae6SAl Viro error = user_path_parent(newdfd, newname, &newnd, &to); 35961da177e4SLinus Torvalds if (error) 35971da177e4SLinus Torvalds goto exit1; 35981da177e4SLinus Torvalds 35991da177e4SLinus Torvalds error = -EXDEV; 36004ac91378SJan Blunck if (oldnd.path.mnt != newnd.path.mnt) 36011da177e4SLinus Torvalds goto exit2; 36021da177e4SLinus Torvalds 36034ac91378SJan Blunck old_dir = oldnd.path.dentry; 36041da177e4SLinus Torvalds error = -EBUSY; 36051da177e4SLinus Torvalds if (oldnd.last_type != LAST_NORM) 36061da177e4SLinus Torvalds goto exit2; 36071da177e4SLinus Torvalds 36084ac91378SJan Blunck new_dir = newnd.path.dentry; 36091da177e4SLinus Torvalds if (newnd.last_type != LAST_NORM) 36101da177e4SLinus Torvalds goto exit2; 36111da177e4SLinus Torvalds 36120612d9fbSOGAWA Hirofumi oldnd.flags &= ~LOOKUP_PARENT; 36130612d9fbSOGAWA Hirofumi newnd.flags &= ~LOOKUP_PARENT; 36144e9ed2f8SOGAWA Hirofumi newnd.flags |= LOOKUP_RENAME_TARGET; 36150612d9fbSOGAWA Hirofumi 36161da177e4SLinus Torvalds trap = lock_rename(new_dir, old_dir); 36171da177e4SLinus Torvalds 361849705b77SChristoph Hellwig old_dentry = lookup_hash(&oldnd); 36191da177e4SLinus Torvalds error = PTR_ERR(old_dentry); 36201da177e4SLinus Torvalds if (IS_ERR(old_dentry)) 36211da177e4SLinus Torvalds goto exit3; 36221da177e4SLinus Torvalds /* source must exist */ 36231da177e4SLinus Torvalds error = -ENOENT; 36241da177e4SLinus Torvalds if (!old_dentry->d_inode) 36251da177e4SLinus Torvalds goto exit4; 36261da177e4SLinus Torvalds /* unless the source is a directory trailing slashes give -ENOTDIR */ 36271da177e4SLinus Torvalds if (!S_ISDIR(old_dentry->d_inode->i_mode)) { 36281da177e4SLinus Torvalds error = -ENOTDIR; 36291da177e4SLinus Torvalds if (oldnd.last.name[oldnd.last.len]) 36301da177e4SLinus Torvalds goto exit4; 36311da177e4SLinus Torvalds if (newnd.last.name[newnd.last.len]) 36321da177e4SLinus Torvalds goto exit4; 36331da177e4SLinus Torvalds } 36341da177e4SLinus Torvalds /* source should not be ancestor of target */ 36351da177e4SLinus Torvalds error = -EINVAL; 36361da177e4SLinus Torvalds if (old_dentry == trap) 36371da177e4SLinus Torvalds goto exit4; 363849705b77SChristoph Hellwig new_dentry = lookup_hash(&newnd); 36391da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 36401da177e4SLinus Torvalds if (IS_ERR(new_dentry)) 36411da177e4SLinus Torvalds goto exit4; 36421da177e4SLinus Torvalds /* target should not be an ancestor of source */ 36431da177e4SLinus Torvalds error = -ENOTEMPTY; 36441da177e4SLinus Torvalds if (new_dentry == trap) 36451da177e4SLinus Torvalds goto exit5; 36461da177e4SLinus Torvalds 36479079b1ebSDave Hansen error = mnt_want_write(oldnd.path.mnt); 36489079b1ebSDave Hansen if (error) 36499079b1ebSDave Hansen goto exit5; 3650be6d3e56SKentaro Takeda error = security_path_rename(&oldnd.path, old_dentry, 3651be6d3e56SKentaro Takeda &newnd.path, new_dentry); 3652be6d3e56SKentaro Takeda if (error) 3653be6d3e56SKentaro Takeda goto exit6; 36541da177e4SLinus Torvalds error = vfs_rename(old_dir->d_inode, old_dentry, 36551da177e4SLinus Torvalds new_dir->d_inode, new_dentry); 3656be6d3e56SKentaro Takeda exit6: 36579079b1ebSDave Hansen mnt_drop_write(oldnd.path.mnt); 36581da177e4SLinus Torvalds exit5: 36591da177e4SLinus Torvalds dput(new_dentry); 36601da177e4SLinus Torvalds exit4: 36611da177e4SLinus Torvalds dput(old_dentry); 36621da177e4SLinus Torvalds exit3: 36631da177e4SLinus Torvalds unlock_rename(new_dir, old_dir); 36641da177e4SLinus Torvalds exit2: 36651d957f9bSJan Blunck path_put(&newnd.path); 36662ad94ae6SAl Viro putname(to); 36671da177e4SLinus Torvalds exit1: 36681d957f9bSJan Blunck path_put(&oldnd.path); 36691da177e4SLinus Torvalds putname(from); 36702ad94ae6SAl Viro exit: 36711da177e4SLinus Torvalds return error; 36721da177e4SLinus Torvalds } 36731da177e4SLinus Torvalds 3674a26eab24SHeiko Carstens SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) 36755590ff0dSUlrich Drepper { 36765590ff0dSUlrich Drepper return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname); 36775590ff0dSUlrich Drepper } 36785590ff0dSUlrich Drepper 36791da177e4SLinus Torvalds int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link) 36801da177e4SLinus Torvalds { 36811da177e4SLinus Torvalds int len; 36821da177e4SLinus Torvalds 36831da177e4SLinus Torvalds len = PTR_ERR(link); 36841da177e4SLinus Torvalds if (IS_ERR(link)) 36851da177e4SLinus Torvalds goto out; 36861da177e4SLinus Torvalds 36871da177e4SLinus Torvalds len = strlen(link); 36881da177e4SLinus Torvalds if (len > (unsigned) buflen) 36891da177e4SLinus Torvalds len = buflen; 36901da177e4SLinus Torvalds if (copy_to_user(buffer, link, len)) 36911da177e4SLinus Torvalds len = -EFAULT; 36921da177e4SLinus Torvalds out: 36931da177e4SLinus Torvalds return len; 36941da177e4SLinus Torvalds } 36951da177e4SLinus Torvalds 36961da177e4SLinus Torvalds /* 36971da177e4SLinus Torvalds * A helper for ->readlink(). This should be used *ONLY* for symlinks that 36981da177e4SLinus Torvalds * have ->follow_link() touching nd only in nd_set_link(). Using (or not 36991da177e4SLinus Torvalds * using) it for any given inode is up to filesystem. 37001da177e4SLinus Torvalds */ 37011da177e4SLinus Torvalds int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen) 37021da177e4SLinus Torvalds { 37031da177e4SLinus Torvalds struct nameidata nd; 3704cc314eefSLinus Torvalds void *cookie; 3705694a1764SMarcin Slusarz int res; 3706cc314eefSLinus Torvalds 37071da177e4SLinus Torvalds nd.depth = 0; 3708cc314eefSLinus Torvalds cookie = dentry->d_inode->i_op->follow_link(dentry, &nd); 3709694a1764SMarcin Slusarz if (IS_ERR(cookie)) 3710694a1764SMarcin Slusarz return PTR_ERR(cookie); 3711694a1764SMarcin Slusarz 3712694a1764SMarcin Slusarz res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd)); 37131da177e4SLinus Torvalds if (dentry->d_inode->i_op->put_link) 3714cc314eefSLinus Torvalds dentry->d_inode->i_op->put_link(dentry, &nd, cookie); 3715694a1764SMarcin Slusarz return res; 37161da177e4SLinus Torvalds } 37171da177e4SLinus Torvalds 37181da177e4SLinus Torvalds int vfs_follow_link(struct nameidata *nd, const char *link) 37191da177e4SLinus Torvalds { 37201da177e4SLinus Torvalds return __vfs_follow_link(nd, link); 37211da177e4SLinus Torvalds } 37221da177e4SLinus Torvalds 37231da177e4SLinus Torvalds /* get the link contents into pagecache */ 37241da177e4SLinus Torvalds static char *page_getlink(struct dentry * dentry, struct page **ppage) 37251da177e4SLinus Torvalds { 3726ebd09abbSDuane Griffin char *kaddr; 37271da177e4SLinus Torvalds struct page *page; 37281da177e4SLinus Torvalds struct address_space *mapping = dentry->d_inode->i_mapping; 3729090d2b18SPekka Enberg page = read_mapping_page(mapping, 0, NULL); 37301da177e4SLinus Torvalds if (IS_ERR(page)) 37316fe6900eSNick Piggin return (char*)page; 37321da177e4SLinus Torvalds *ppage = page; 3733ebd09abbSDuane Griffin kaddr = kmap(page); 3734ebd09abbSDuane Griffin nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1); 3735ebd09abbSDuane Griffin return kaddr; 37361da177e4SLinus Torvalds } 37371da177e4SLinus Torvalds 37381da177e4SLinus Torvalds int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 37391da177e4SLinus Torvalds { 37401da177e4SLinus Torvalds struct page *page = NULL; 37411da177e4SLinus Torvalds char *s = page_getlink(dentry, &page); 37421da177e4SLinus Torvalds int res = vfs_readlink(dentry,buffer,buflen,s); 37431da177e4SLinus Torvalds if (page) { 37441da177e4SLinus Torvalds kunmap(page); 37451da177e4SLinus Torvalds page_cache_release(page); 37461da177e4SLinus Torvalds } 37471da177e4SLinus Torvalds return res; 37481da177e4SLinus Torvalds } 37491da177e4SLinus Torvalds 3750cc314eefSLinus Torvalds void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd) 37511da177e4SLinus Torvalds { 3752cc314eefSLinus Torvalds struct page *page = NULL; 37531da177e4SLinus Torvalds nd_set_link(nd, page_getlink(dentry, &page)); 3754cc314eefSLinus Torvalds return page; 37551da177e4SLinus Torvalds } 37561da177e4SLinus Torvalds 3757cc314eefSLinus Torvalds void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) 37581da177e4SLinus Torvalds { 3759cc314eefSLinus Torvalds struct page *page = cookie; 3760cc314eefSLinus Torvalds 3761cc314eefSLinus Torvalds if (page) { 37621da177e4SLinus Torvalds kunmap(page); 37631da177e4SLinus Torvalds page_cache_release(page); 37641da177e4SLinus Torvalds } 37651da177e4SLinus Torvalds } 37661da177e4SLinus Torvalds 376754566b2cSNick Piggin /* 376854566b2cSNick Piggin * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS 376954566b2cSNick Piggin */ 377054566b2cSNick Piggin int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) 37711da177e4SLinus Torvalds { 37721da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 37730adb25d2SKirill Korotaev struct page *page; 3774afddba49SNick Piggin void *fsdata; 3775beb497abSDmitriy Monakhov int err; 37761da177e4SLinus Torvalds char *kaddr; 377754566b2cSNick Piggin unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE; 377854566b2cSNick Piggin if (nofs) 377954566b2cSNick Piggin flags |= AOP_FLAG_NOFS; 37801da177e4SLinus Torvalds 37817e53cac4SNeilBrown retry: 3782afddba49SNick Piggin err = pagecache_write_begin(NULL, mapping, 0, len-1, 378354566b2cSNick Piggin flags, &page, &fsdata); 37841da177e4SLinus Torvalds if (err) 3785afddba49SNick Piggin goto fail; 3786afddba49SNick Piggin 3787e8e3c3d6SCong Wang kaddr = kmap_atomic(page); 37881da177e4SLinus Torvalds memcpy(kaddr, symname, len-1); 3789e8e3c3d6SCong Wang kunmap_atomic(kaddr); 3790afddba49SNick Piggin 3791afddba49SNick Piggin err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, 3792afddba49SNick Piggin page, fsdata); 37931da177e4SLinus Torvalds if (err < 0) 37941da177e4SLinus Torvalds goto fail; 3795afddba49SNick Piggin if (err < len-1) 3796afddba49SNick Piggin goto retry; 3797afddba49SNick Piggin 37981da177e4SLinus Torvalds mark_inode_dirty(inode); 37991da177e4SLinus Torvalds return 0; 38001da177e4SLinus Torvalds fail: 38011da177e4SLinus Torvalds return err; 38021da177e4SLinus Torvalds } 38031da177e4SLinus Torvalds 38040adb25d2SKirill Korotaev int page_symlink(struct inode *inode, const char *symname, int len) 38050adb25d2SKirill Korotaev { 38060adb25d2SKirill Korotaev return __page_symlink(inode, symname, len, 380754566b2cSNick Piggin !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS)); 38080adb25d2SKirill Korotaev } 38090adb25d2SKirill Korotaev 381092e1d5beSArjan van de Ven const struct inode_operations page_symlink_inode_operations = { 38111da177e4SLinus Torvalds .readlink = generic_readlink, 38121da177e4SLinus Torvalds .follow_link = page_follow_link_light, 38131da177e4SLinus Torvalds .put_link = page_put_link, 38141da177e4SLinus Torvalds }; 38151da177e4SLinus Torvalds 38162d8f3038SAl Viro EXPORT_SYMBOL(user_path_at); 3817cc53ce53SDavid Howells EXPORT_SYMBOL(follow_down_one); 38181da177e4SLinus Torvalds EXPORT_SYMBOL(follow_down); 38191da177e4SLinus Torvalds EXPORT_SYMBOL(follow_up); 38201da177e4SLinus Torvalds EXPORT_SYMBOL(get_write_access); /* binfmt_aout */ 38211da177e4SLinus Torvalds EXPORT_SYMBOL(getname); 38221da177e4SLinus Torvalds EXPORT_SYMBOL(lock_rename); 38231da177e4SLinus Torvalds EXPORT_SYMBOL(lookup_one_len); 38241da177e4SLinus Torvalds EXPORT_SYMBOL(page_follow_link_light); 38251da177e4SLinus Torvalds EXPORT_SYMBOL(page_put_link); 38261da177e4SLinus Torvalds EXPORT_SYMBOL(page_readlink); 38270adb25d2SKirill Korotaev EXPORT_SYMBOL(__page_symlink); 38281da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink); 38291da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink_inode_operations); 3830d1811465SAl Viro EXPORT_SYMBOL(kern_path); 383116f18200SJosef 'Jeff' Sipek EXPORT_SYMBOL(vfs_path_lookup); 3832f419a2e3SAl Viro EXPORT_SYMBOL(inode_permission); 38331da177e4SLinus Torvalds EXPORT_SYMBOL(unlock_rename); 38341da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_create); 38351da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_follow_link); 38361da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_link); 38371da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_mkdir); 38381da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_mknod); 38391da177e4SLinus Torvalds EXPORT_SYMBOL(generic_permission); 38401da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_readlink); 38411da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_rename); 38421da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_rmdir); 38431da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_symlink); 38441da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_unlink); 38451da177e4SLinus Torvalds EXPORT_SYMBOL(dentry_unhash); 38461da177e4SLinus Torvalds EXPORT_SYMBOL(generic_readlink); 3847