11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * linux/fs/namei.c 31da177e4SLinus Torvalds * 41da177e4SLinus Torvalds * Copyright (C) 1991, 1992 Linus Torvalds 51da177e4SLinus Torvalds */ 61da177e4SLinus Torvalds 71da177e4SLinus Torvalds /* 81da177e4SLinus Torvalds * Some corrections by tytso. 91da177e4SLinus Torvalds */ 101da177e4SLinus Torvalds 111da177e4SLinus Torvalds /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname 121da177e4SLinus Torvalds * lookup logic. 131da177e4SLinus Torvalds */ 141da177e4SLinus Torvalds /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture. 151da177e4SLinus Torvalds */ 161da177e4SLinus Torvalds 171da177e4SLinus Torvalds #include <linux/init.h> 18630d9c47SPaul Gortmaker #include <linux/export.h> 1944696908SDavid S. Miller #include <linux/kernel.h> 201da177e4SLinus Torvalds #include <linux/slab.h> 211da177e4SLinus Torvalds #include <linux/fs.h> 221da177e4SLinus Torvalds #include <linux/namei.h> 231da177e4SLinus Torvalds #include <linux/pagemap.h> 240eeca283SRobert Love #include <linux/fsnotify.h> 251da177e4SLinus Torvalds #include <linux/personality.h> 261da177e4SLinus Torvalds #include <linux/security.h> 276146f0d5SMimi Zohar #include <linux/ima.h> 281da177e4SLinus Torvalds #include <linux/syscalls.h> 291da177e4SLinus Torvalds #include <linux/mount.h> 301da177e4SLinus Torvalds #include <linux/audit.h> 3116f7e0feSRandy Dunlap #include <linux/capability.h> 32834f2a4aSTrond Myklebust #include <linux/file.h> 335590ff0dSUlrich Drepper #include <linux/fcntl.h> 3408ce5f16SSerge E. Hallyn #include <linux/device_cgroup.h> 355ad4e53bSAl Viro #include <linux/fs_struct.h> 36e77819e5SLinus Torvalds #include <linux/posix_acl.h> 371da177e4SLinus Torvalds #include <asm/uaccess.h> 381da177e4SLinus Torvalds 39e81e3f4dSEric Paris #include "internal.h" 40c7105365SAl Viro #include "mount.h" 41e81e3f4dSEric Paris 421da177e4SLinus Torvalds /* [Feb-1997 T. Schoebel-Theuer] 431da177e4SLinus Torvalds * Fundamental changes in the pathname lookup mechanisms (namei) 441da177e4SLinus Torvalds * were necessary because of omirr. The reason is that omirr needs 451da177e4SLinus Torvalds * to know the _real_ pathname, not the user-supplied one, in case 461da177e4SLinus Torvalds * of symlinks (and also when transname replacements occur). 471da177e4SLinus Torvalds * 481da177e4SLinus Torvalds * The new code replaces the old recursive symlink resolution with 491da177e4SLinus Torvalds * an iterative one (in case of non-nested symlink chains). It does 501da177e4SLinus Torvalds * this with calls to <fs>_follow_link(). 511da177e4SLinus Torvalds * As a side effect, dir_namei(), _namei() and follow_link() are now 521da177e4SLinus Torvalds * replaced with a single function lookup_dentry() that can handle all 531da177e4SLinus Torvalds * the special cases of the former code. 541da177e4SLinus Torvalds * 551da177e4SLinus Torvalds * With the new dcache, the pathname is stored at each inode, at least as 561da177e4SLinus Torvalds * long as the refcount of the inode is positive. As a side effect, the 571da177e4SLinus Torvalds * size of the dcache depends on the inode cache and thus is dynamic. 581da177e4SLinus Torvalds * 591da177e4SLinus Torvalds * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink 601da177e4SLinus Torvalds * resolution to correspond with current state of the code. 611da177e4SLinus Torvalds * 621da177e4SLinus Torvalds * Note that the symlink resolution is not *completely* iterative. 631da177e4SLinus Torvalds * There is still a significant amount of tail- and mid- recursion in 641da177e4SLinus Torvalds * the algorithm. Also, note that <fs>_readlink() is not used in 651da177e4SLinus Torvalds * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink() 661da177e4SLinus Torvalds * may return different results than <fs>_follow_link(). Many virtual 671da177e4SLinus Torvalds * filesystems (including /proc) exhibit this behavior. 681da177e4SLinus Torvalds */ 691da177e4SLinus Torvalds 701da177e4SLinus Torvalds /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation: 711da177e4SLinus Torvalds * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL 721da177e4SLinus Torvalds * and the name already exists in form of a symlink, try to create the new 731da177e4SLinus Torvalds * name indicated by the symlink. The old code always complained that the 741da177e4SLinus Torvalds * name already exists, due to not following the symlink even if its target 751da177e4SLinus Torvalds * is nonexistent. The new semantics affects also mknod() and link() when 7625985edcSLucas De Marchi * the name is a symlink pointing to a non-existent name. 771da177e4SLinus Torvalds * 781da177e4SLinus Torvalds * I don't know which semantics is the right one, since I have no access 791da177e4SLinus Torvalds * to standards. But I found by trial that HP-UX 9.0 has the full "new" 801da177e4SLinus Torvalds * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the 811da177e4SLinus Torvalds * "old" one. Personally, I think the new semantics is much more logical. 821da177e4SLinus Torvalds * Note that "ln old new" where "new" is a symlink pointing to a non-existing 831da177e4SLinus Torvalds * file does succeed in both HP-UX and SunOs, but not in Solaris 841da177e4SLinus Torvalds * and in the old Linux semantics. 851da177e4SLinus Torvalds */ 861da177e4SLinus Torvalds 871da177e4SLinus Torvalds /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink 881da177e4SLinus Torvalds * semantics. See the comments in "open_namei" and "do_link" below. 891da177e4SLinus Torvalds * 901da177e4SLinus Torvalds * [10-Sep-98 Alan Modra] Another symlink change. 911da177e4SLinus Torvalds */ 921da177e4SLinus Torvalds 931da177e4SLinus Torvalds /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks: 941da177e4SLinus Torvalds * inside the path - always follow. 951da177e4SLinus Torvalds * in the last component in creation/removal/renaming - never follow. 961da177e4SLinus Torvalds * if LOOKUP_FOLLOW passed - follow. 971da177e4SLinus Torvalds * if the pathname has trailing slashes - follow. 981da177e4SLinus Torvalds * otherwise - don't follow. 991da177e4SLinus Torvalds * (applied in that order). 1001da177e4SLinus Torvalds * 1011da177e4SLinus Torvalds * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT 1021da177e4SLinus Torvalds * restored for 2.4. This is the last surviving part of old 4.2BSD bug. 1031da177e4SLinus Torvalds * During the 2.4 we need to fix the userland stuff depending on it - 1041da177e4SLinus Torvalds * hopefully we will be able to get rid of that wart in 2.5. So far only 1051da177e4SLinus Torvalds * XEmacs seems to be relying on it... 1061da177e4SLinus Torvalds */ 1071da177e4SLinus Torvalds /* 1081da177e4SLinus Torvalds * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland) 109a11f3a05SArjan van de Ven * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives 1101da177e4SLinus Torvalds * any extra contention... 1111da177e4SLinus Torvalds */ 1121da177e4SLinus Torvalds 1131da177e4SLinus Torvalds /* In order to reduce some races, while at the same time doing additional 1141da177e4SLinus Torvalds * checking and hopefully speeding things up, we copy filenames to the 1151da177e4SLinus Torvalds * kernel data space before using them.. 1161da177e4SLinus Torvalds * 1171da177e4SLinus Torvalds * POSIX.1 2.4: an empty pathname is invalid (ENOENT). 1181da177e4SLinus Torvalds * PATH_MAX includes the nul terminator --RR. 1191da177e4SLinus Torvalds */ 12091a27b2aSJeff Layton void final_putname(struct filename *name) 1211da177e4SLinus Torvalds { 1227950e385SJeff Layton if (name->separate) { 12391a27b2aSJeff Layton __putname(name->name); 12491a27b2aSJeff Layton kfree(name); 1257950e385SJeff Layton } else { 1267950e385SJeff Layton __putname(name); 12791a27b2aSJeff Layton } 1287950e385SJeff Layton } 1297950e385SJeff Layton 1307950e385SJeff Layton #define EMBEDDED_NAME_MAX (PATH_MAX - sizeof(struct filename)) 13191a27b2aSJeff Layton 13291a27b2aSJeff Layton static struct filename * 13391a27b2aSJeff Layton getname_flags(const char __user *filename, int flags, int *empty) 13491a27b2aSJeff Layton { 13591a27b2aSJeff Layton struct filename *result, *err; 1363f9f0aa6SLinus Torvalds int len; 1377950e385SJeff Layton long max; 1387950e385SJeff Layton char *kname; 1391da177e4SLinus Torvalds 1407ac86265SJeff Layton result = audit_reusename(filename); 1417ac86265SJeff Layton if (result) 1427ac86265SJeff Layton return result; 1437ac86265SJeff Layton 1447950e385SJeff Layton result = __getname(); 1453f9f0aa6SLinus Torvalds if (unlikely(!result)) 1464043cde8SEric Paris return ERR_PTR(-ENOMEM); 1471da177e4SLinus Torvalds 1487950e385SJeff Layton /* 1497950e385SJeff Layton * First, try to embed the struct filename inside the names_cache 1507950e385SJeff Layton * allocation 1517950e385SJeff Layton */ 1527950e385SJeff Layton kname = (char *)result + sizeof(*result); 15391a27b2aSJeff Layton result->name = kname; 1547950e385SJeff Layton result->separate = false; 1557950e385SJeff Layton max = EMBEDDED_NAME_MAX; 1567950e385SJeff Layton 1577950e385SJeff Layton recopy: 1587950e385SJeff Layton len = strncpy_from_user(kname, filename, max); 15991a27b2aSJeff Layton if (unlikely(len < 0)) { 1603f9f0aa6SLinus Torvalds err = ERR_PTR(len); 1613f9f0aa6SLinus Torvalds goto error; 16291a27b2aSJeff Layton } 1633f9f0aa6SLinus Torvalds 1647950e385SJeff Layton /* 1657950e385SJeff Layton * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a 1667950e385SJeff Layton * separate struct filename so we can dedicate the entire 1677950e385SJeff Layton * names_cache allocation for the pathname, and re-do the copy from 1687950e385SJeff Layton * userland. 1697950e385SJeff Layton */ 1707950e385SJeff Layton if (len == EMBEDDED_NAME_MAX && max == EMBEDDED_NAME_MAX) { 1717950e385SJeff Layton kname = (char *)result; 1727950e385SJeff Layton 1737950e385SJeff Layton result = kzalloc(sizeof(*result), GFP_KERNEL); 1747950e385SJeff Layton if (!result) { 1757950e385SJeff Layton err = ERR_PTR(-ENOMEM); 1767950e385SJeff Layton result = (struct filename *)kname; 1777950e385SJeff Layton goto error; 1787950e385SJeff Layton } 1797950e385SJeff Layton result->name = kname; 1807950e385SJeff Layton result->separate = true; 1817950e385SJeff Layton max = PATH_MAX; 1827950e385SJeff Layton goto recopy; 1837950e385SJeff Layton } 1847950e385SJeff Layton 1853f9f0aa6SLinus Torvalds /* The empty path is special. */ 1863f9f0aa6SLinus Torvalds if (unlikely(!len)) { 1873f9f0aa6SLinus Torvalds if (empty) 1881fa1e7f6SAndy Whitcroft *empty = 1; 1893f9f0aa6SLinus Torvalds err = ERR_PTR(-ENOENT); 1903f9f0aa6SLinus Torvalds if (!(flags & LOOKUP_EMPTY)) 1913f9f0aa6SLinus Torvalds goto error; 1921da177e4SLinus Torvalds } 1933f9f0aa6SLinus Torvalds 1943f9f0aa6SLinus Torvalds err = ERR_PTR(-ENAMETOOLONG); 1957950e385SJeff Layton if (unlikely(len >= PATH_MAX)) 1967950e385SJeff Layton goto error; 1977950e385SJeff Layton 1987950e385SJeff Layton result->uptr = filename; 199c4ad8f98SLinus Torvalds result->aname = NULL; 2001da177e4SLinus Torvalds audit_getname(result); 2011da177e4SLinus Torvalds return result; 2021da177e4SLinus Torvalds 2033f9f0aa6SLinus Torvalds error: 2047950e385SJeff Layton final_putname(result); 2053f9f0aa6SLinus Torvalds return err; 2063f9f0aa6SLinus Torvalds } 2073f9f0aa6SLinus Torvalds 20891a27b2aSJeff Layton struct filename * 20991a27b2aSJeff Layton getname(const char __user * filename) 210f52e0c11SAl Viro { 211f7493e5dSLinus Torvalds return getname_flags(filename, 0, NULL); 212f52e0c11SAl Viro } 213f52e0c11SAl Viro 214c4ad8f98SLinus Torvalds /* 215c4ad8f98SLinus Torvalds * The "getname_kernel()" interface doesn't do pathnames longer 216c4ad8f98SLinus Torvalds * than EMBEDDED_NAME_MAX. Deal with it - you're a kernel user. 217c4ad8f98SLinus Torvalds */ 218c4ad8f98SLinus Torvalds struct filename * 219c4ad8f98SLinus Torvalds getname_kernel(const char * filename) 220c4ad8f98SLinus Torvalds { 221c4ad8f98SLinus Torvalds struct filename *result; 222c4ad8f98SLinus Torvalds char *kname; 223c4ad8f98SLinus Torvalds int len; 224c4ad8f98SLinus Torvalds 225c4ad8f98SLinus Torvalds len = strlen(filename); 226c4ad8f98SLinus Torvalds if (len >= EMBEDDED_NAME_MAX) 227c4ad8f98SLinus Torvalds return ERR_PTR(-ENAMETOOLONG); 228c4ad8f98SLinus Torvalds 229c4ad8f98SLinus Torvalds result = __getname(); 230c4ad8f98SLinus Torvalds if (unlikely(!result)) 231c4ad8f98SLinus Torvalds return ERR_PTR(-ENOMEM); 232c4ad8f98SLinus Torvalds 233c4ad8f98SLinus Torvalds kname = (char *)result + sizeof(*result); 234c4ad8f98SLinus Torvalds result->name = kname; 235c4ad8f98SLinus Torvalds result->uptr = NULL; 236c4ad8f98SLinus Torvalds result->aname = NULL; 237c4ad8f98SLinus Torvalds result->separate = false; 238c4ad8f98SLinus Torvalds 239c4ad8f98SLinus Torvalds strlcpy(kname, filename, EMBEDDED_NAME_MAX); 240c4ad8f98SLinus Torvalds return result; 241c4ad8f98SLinus Torvalds } 242c4ad8f98SLinus Torvalds 2431da177e4SLinus Torvalds #ifdef CONFIG_AUDITSYSCALL 24491a27b2aSJeff Layton void putname(struct filename *name) 2451da177e4SLinus Torvalds { 2465ac3a9c2SAl Viro if (unlikely(!audit_dummy_context())) 24791a27b2aSJeff Layton return audit_putname(name); 24891a27b2aSJeff Layton final_putname(name); 2491da177e4SLinus Torvalds } 2501da177e4SLinus Torvalds #endif 2511da177e4SLinus Torvalds 252e77819e5SLinus Torvalds static int check_acl(struct inode *inode, int mask) 253e77819e5SLinus Torvalds { 25484635d68SLinus Torvalds #ifdef CONFIG_FS_POSIX_ACL 255e77819e5SLinus Torvalds struct posix_acl *acl; 256e77819e5SLinus Torvalds 257e77819e5SLinus Torvalds if (mask & MAY_NOT_BLOCK) { 2583567866bSAl Viro acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS); 2593567866bSAl Viro if (!acl) 260e77819e5SLinus Torvalds return -EAGAIN; 2613567866bSAl Viro /* no ->get_acl() calls in RCU mode... */ 2623567866bSAl Viro if (acl == ACL_NOT_CACHED) 263e77819e5SLinus Torvalds return -ECHILD; 264206b1d09SAri Savolainen return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK); 265e77819e5SLinus Torvalds } 266e77819e5SLinus Torvalds 2672982baa2SChristoph Hellwig acl = get_acl(inode, ACL_TYPE_ACCESS); 2684e34e719SChristoph Hellwig if (IS_ERR(acl)) 2694e34e719SChristoph Hellwig return PTR_ERR(acl); 270e77819e5SLinus Torvalds if (acl) { 271e77819e5SLinus Torvalds int error = posix_acl_permission(inode, acl, mask); 272e77819e5SLinus Torvalds posix_acl_release(acl); 273e77819e5SLinus Torvalds return error; 274e77819e5SLinus Torvalds } 27584635d68SLinus Torvalds #endif 276e77819e5SLinus Torvalds 277e77819e5SLinus Torvalds return -EAGAIN; 278e77819e5SLinus Torvalds } 279e77819e5SLinus Torvalds 2805909ccaaSLinus Torvalds /* 281948409c7SAndreas Gruenbacher * This does the basic permission checking 2825909ccaaSLinus Torvalds */ 2837e40145eSAl Viro static int acl_permission_check(struct inode *inode, int mask) 2845909ccaaSLinus Torvalds { 28526cf46beSLinus Torvalds unsigned int mode = inode->i_mode; 2865909ccaaSLinus Torvalds 2878e96e3b7SEric W. Biederman if (likely(uid_eq(current_fsuid(), inode->i_uid))) 2885909ccaaSLinus Torvalds mode >>= 6; 2895909ccaaSLinus Torvalds else { 290e77819e5SLinus Torvalds if (IS_POSIXACL(inode) && (mode & S_IRWXG)) { 2917e40145eSAl Viro int error = check_acl(inode, mask); 2925909ccaaSLinus Torvalds if (error != -EAGAIN) 2935909ccaaSLinus Torvalds return error; 2945909ccaaSLinus Torvalds } 2955909ccaaSLinus Torvalds 2965909ccaaSLinus Torvalds if (in_group_p(inode->i_gid)) 2975909ccaaSLinus Torvalds mode >>= 3; 2985909ccaaSLinus Torvalds } 2995909ccaaSLinus Torvalds 3005909ccaaSLinus Torvalds /* 3015909ccaaSLinus Torvalds * If the DACs are ok we don't need any capability check. 3025909ccaaSLinus Torvalds */ 3039c2c7039SAl Viro if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) 3045909ccaaSLinus Torvalds return 0; 3055909ccaaSLinus Torvalds return -EACCES; 3065909ccaaSLinus Torvalds } 3071da177e4SLinus Torvalds 3081da177e4SLinus Torvalds /** 3091da177e4SLinus Torvalds * generic_permission - check for access rights on a Posix-like filesystem 3101da177e4SLinus Torvalds * @inode: inode to check access rights for 3118fd90c8dSAndreas Gruenbacher * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...) 3121da177e4SLinus Torvalds * 3131da177e4SLinus Torvalds * Used to check for read/write/execute permissions on a file. 3141da177e4SLinus Torvalds * We use "fsuid" for this, letting us set arbitrary permissions 3151da177e4SLinus Torvalds * for filesystem access without changing the "normal" uids which 316b74c79e9SNick Piggin * are used for other things. 317b74c79e9SNick Piggin * 318b74c79e9SNick Piggin * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk 319b74c79e9SNick Piggin * request cannot be satisfied (eg. requires blocking or too much complexity). 320b74c79e9SNick Piggin * It would then be called again in ref-walk mode. 3211da177e4SLinus Torvalds */ 3222830ba7fSAl Viro int generic_permission(struct inode *inode, int mask) 3231da177e4SLinus Torvalds { 3245909ccaaSLinus Torvalds int ret; 3251da177e4SLinus Torvalds 3261da177e4SLinus Torvalds /* 327948409c7SAndreas Gruenbacher * Do the basic permission checks. 3281da177e4SLinus Torvalds */ 3297e40145eSAl Viro ret = acl_permission_check(inode, mask); 3305909ccaaSLinus Torvalds if (ret != -EACCES) 3315909ccaaSLinus Torvalds return ret; 3321da177e4SLinus Torvalds 333d594e7ecSAl Viro if (S_ISDIR(inode->i_mode)) { 334d594e7ecSAl Viro /* DACs are overridable for directories */ 3351a48e2acSEric W. Biederman if (inode_capable(inode, CAP_DAC_OVERRIDE)) 336d594e7ecSAl Viro return 0; 337d594e7ecSAl Viro if (!(mask & MAY_WRITE)) 3381a48e2acSEric W. Biederman if (inode_capable(inode, CAP_DAC_READ_SEARCH)) 339d594e7ecSAl Viro return 0; 340d594e7ecSAl Viro return -EACCES; 341d594e7ecSAl Viro } 3421da177e4SLinus Torvalds /* 3431da177e4SLinus Torvalds * Read/write DACs are always overridable. 344d594e7ecSAl Viro * Executable DACs are overridable when there is 345d594e7ecSAl Viro * at least one exec bit set. 3461da177e4SLinus Torvalds */ 347d594e7ecSAl Viro if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO)) 3481a48e2acSEric W. Biederman if (inode_capable(inode, CAP_DAC_OVERRIDE)) 3491da177e4SLinus Torvalds return 0; 3501da177e4SLinus Torvalds 3511da177e4SLinus Torvalds /* 3521da177e4SLinus Torvalds * Searching includes executable on directories, else just read. 3531da177e4SLinus Torvalds */ 3547ea66001SSerge E. Hallyn mask &= MAY_READ | MAY_WRITE | MAY_EXEC; 355d594e7ecSAl Viro if (mask == MAY_READ) 3561a48e2acSEric W. Biederman if (inode_capable(inode, CAP_DAC_READ_SEARCH)) 3571da177e4SLinus Torvalds return 0; 3581da177e4SLinus Torvalds 3591da177e4SLinus Torvalds return -EACCES; 3601da177e4SLinus Torvalds } 3611da177e4SLinus Torvalds 3623ddcd056SLinus Torvalds /* 3633ddcd056SLinus Torvalds * We _really_ want to just do "generic_permission()" without 3643ddcd056SLinus Torvalds * even looking at the inode->i_op values. So we keep a cache 3653ddcd056SLinus Torvalds * flag in inode->i_opflags, that says "this has not special 3663ddcd056SLinus Torvalds * permission function, use the fast case". 3673ddcd056SLinus Torvalds */ 3683ddcd056SLinus Torvalds static inline int do_inode_permission(struct inode *inode, int mask) 3693ddcd056SLinus Torvalds { 3703ddcd056SLinus Torvalds if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) { 3713ddcd056SLinus Torvalds if (likely(inode->i_op->permission)) 3723ddcd056SLinus Torvalds return inode->i_op->permission(inode, mask); 3733ddcd056SLinus Torvalds 3743ddcd056SLinus Torvalds /* This gets set once for the inode lifetime */ 3753ddcd056SLinus Torvalds spin_lock(&inode->i_lock); 3763ddcd056SLinus Torvalds inode->i_opflags |= IOP_FASTPERM; 3773ddcd056SLinus Torvalds spin_unlock(&inode->i_lock); 3783ddcd056SLinus Torvalds } 3793ddcd056SLinus Torvalds return generic_permission(inode, mask); 3803ddcd056SLinus Torvalds } 3813ddcd056SLinus Torvalds 382cb23beb5SChristoph Hellwig /** 3830bdaea90SDavid Howells * __inode_permission - Check for access rights to a given inode 3840bdaea90SDavid Howells * @inode: Inode to check permission on 3850bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 386cb23beb5SChristoph Hellwig * 3870bdaea90SDavid Howells * Check for read/write/execute permissions on an inode. 388948409c7SAndreas Gruenbacher * 389948409c7SAndreas Gruenbacher * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. 3900bdaea90SDavid Howells * 3910bdaea90SDavid Howells * This does not check for a read-only file system. You probably want 3920bdaea90SDavid Howells * inode_permission(). 393cb23beb5SChristoph Hellwig */ 3940bdaea90SDavid Howells int __inode_permission(struct inode *inode, int mask) 3951da177e4SLinus Torvalds { 396e6305c43SAl Viro int retval; 3971da177e4SLinus Torvalds 3983ddcd056SLinus Torvalds if (unlikely(mask & MAY_WRITE)) { 3991da177e4SLinus Torvalds /* 4001da177e4SLinus Torvalds * Nobody gets write access to an immutable file. 4011da177e4SLinus Torvalds */ 4021da177e4SLinus Torvalds if (IS_IMMUTABLE(inode)) 4031da177e4SLinus Torvalds return -EACCES; 4041da177e4SLinus Torvalds } 4051da177e4SLinus Torvalds 4063ddcd056SLinus Torvalds retval = do_inode_permission(inode, mask); 4071da177e4SLinus Torvalds if (retval) 4081da177e4SLinus Torvalds return retval; 4091da177e4SLinus Torvalds 41008ce5f16SSerge E. Hallyn retval = devcgroup_inode_permission(inode, mask); 41108ce5f16SSerge E. Hallyn if (retval) 41208ce5f16SSerge E. Hallyn return retval; 41308ce5f16SSerge E. Hallyn 414d09ca739SEric Paris return security_inode_permission(inode, mask); 4151da177e4SLinus Torvalds } 4161da177e4SLinus Torvalds 417f4d6ff89SAl Viro /** 4180bdaea90SDavid Howells * sb_permission - Check superblock-level permissions 4190bdaea90SDavid Howells * @sb: Superblock of inode to check permission on 42055852635SRandy Dunlap * @inode: Inode to check permission on 4210bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 4220bdaea90SDavid Howells * 4230bdaea90SDavid Howells * Separate out file-system wide checks from inode-specific permission checks. 4240bdaea90SDavid Howells */ 4250bdaea90SDavid Howells static int sb_permission(struct super_block *sb, struct inode *inode, int mask) 4260bdaea90SDavid Howells { 4270bdaea90SDavid Howells if (unlikely(mask & MAY_WRITE)) { 4280bdaea90SDavid Howells umode_t mode = inode->i_mode; 4290bdaea90SDavid Howells 4300bdaea90SDavid Howells /* Nobody gets write access to a read-only fs. */ 4310bdaea90SDavid Howells if ((sb->s_flags & MS_RDONLY) && 4320bdaea90SDavid Howells (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) 4330bdaea90SDavid Howells return -EROFS; 4340bdaea90SDavid Howells } 4350bdaea90SDavid Howells return 0; 4360bdaea90SDavid Howells } 4370bdaea90SDavid Howells 4380bdaea90SDavid Howells /** 4390bdaea90SDavid Howells * inode_permission - Check for access rights to a given inode 4400bdaea90SDavid Howells * @inode: Inode to check permission on 4410bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 4420bdaea90SDavid Howells * 4430bdaea90SDavid Howells * Check for read/write/execute permissions on an inode. We use fs[ug]id for 4440bdaea90SDavid Howells * this, letting us set arbitrary permissions for filesystem access without 4450bdaea90SDavid Howells * changing the "normal" UIDs which are used for other things. 4460bdaea90SDavid Howells * 4470bdaea90SDavid Howells * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. 4480bdaea90SDavid Howells */ 4490bdaea90SDavid Howells int inode_permission(struct inode *inode, int mask) 4500bdaea90SDavid Howells { 4510bdaea90SDavid Howells int retval; 4520bdaea90SDavid Howells 4530bdaea90SDavid Howells retval = sb_permission(inode->i_sb, inode, mask); 4540bdaea90SDavid Howells if (retval) 4550bdaea90SDavid Howells return retval; 4560bdaea90SDavid Howells return __inode_permission(inode, mask); 4570bdaea90SDavid Howells } 4580bdaea90SDavid Howells 4590bdaea90SDavid Howells /** 4605dd784d0SJan Blunck * path_get - get a reference to a path 4615dd784d0SJan Blunck * @path: path to get the reference to 4625dd784d0SJan Blunck * 4635dd784d0SJan Blunck * Given a path increment the reference count to the dentry and the vfsmount. 4645dd784d0SJan Blunck */ 465dcf787f3SAl Viro void path_get(const struct path *path) 4665dd784d0SJan Blunck { 4675dd784d0SJan Blunck mntget(path->mnt); 4685dd784d0SJan Blunck dget(path->dentry); 4695dd784d0SJan Blunck } 4705dd784d0SJan Blunck EXPORT_SYMBOL(path_get); 4715dd784d0SJan Blunck 4725dd784d0SJan Blunck /** 4731d957f9bSJan Blunck * path_put - put a reference to a path 4741d957f9bSJan Blunck * @path: path to put the reference to 4751d957f9bSJan Blunck * 4761d957f9bSJan Blunck * Given a path decrement the reference count to the dentry and the vfsmount. 4771d957f9bSJan Blunck */ 478dcf787f3SAl Viro void path_put(const struct path *path) 4791da177e4SLinus Torvalds { 4801d957f9bSJan Blunck dput(path->dentry); 4811d957f9bSJan Blunck mntput(path->mnt); 4821da177e4SLinus Torvalds } 4831d957f9bSJan Blunck EXPORT_SYMBOL(path_put); 4841da177e4SLinus Torvalds 48519660af7SAl Viro /* 48631e6b01fSNick Piggin * Path walking has 2 modes, rcu-walk and ref-walk (see 48719660af7SAl Viro * Documentation/filesystems/path-lookup.txt). In situations when we can't 48819660af7SAl Viro * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab 48919660af7SAl Viro * normal reference counts on dentries and vfsmounts to transition to rcu-walk 49019660af7SAl Viro * mode. Refcounts are grabbed at the last known good point before rcu-walk 49119660af7SAl Viro * got stuck, so ref-walk may continue from there. If this is not successful 49219660af7SAl Viro * (eg. a seqcount has changed), then failure is returned and it's up to caller 49319660af7SAl Viro * to restart the path walk from the beginning in ref-walk mode. 49431e6b01fSNick Piggin */ 49531e6b01fSNick Piggin 49631e6b01fSNick Piggin /** 49719660af7SAl Viro * unlazy_walk - try to switch to ref-walk mode. 49819660af7SAl Viro * @nd: nameidata pathwalk data 49919660af7SAl Viro * @dentry: child of nd->path.dentry or NULL 50039191628SRandy Dunlap * Returns: 0 on success, -ECHILD on failure 50131e6b01fSNick Piggin * 50219660af7SAl Viro * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry 50319660af7SAl Viro * for ref-walk mode. @dentry must be a path found by a do_lookup call on 50419660af7SAl Viro * @nd or NULL. Must be called from rcu-walk context. 50531e6b01fSNick Piggin */ 50619660af7SAl Viro static int unlazy_walk(struct nameidata *nd, struct dentry *dentry) 50731e6b01fSNick Piggin { 50831e6b01fSNick Piggin struct fs_struct *fs = current->fs; 50931e6b01fSNick Piggin struct dentry *parent = nd->path.dentry; 51031e6b01fSNick Piggin 51131e6b01fSNick Piggin BUG_ON(!(nd->flags & LOOKUP_RCU)); 512e5c832d5SLinus Torvalds 513e5c832d5SLinus Torvalds /* 51448a066e7SAl Viro * After legitimizing the bastards, terminate_walk() 51548a066e7SAl Viro * will do the right thing for non-RCU mode, and all our 51648a066e7SAl Viro * subsequent exit cases should rcu_read_unlock() 51748a066e7SAl Viro * before returning. Do vfsmount first; if dentry 51848a066e7SAl Viro * can't be legitimized, just set nd->path.dentry to NULL 51948a066e7SAl Viro * and rely on dput(NULL) being a no-op. 520e5c832d5SLinus Torvalds */ 52148a066e7SAl Viro if (!legitimize_mnt(nd->path.mnt, nd->m_seq)) 522e5c832d5SLinus Torvalds return -ECHILD; 523e5c832d5SLinus Torvalds nd->flags &= ~LOOKUP_RCU; 52415570086SLinus Torvalds 52548a066e7SAl Viro if (!lockref_get_not_dead(&parent->d_lockref)) { 52648a066e7SAl Viro nd->path.dentry = NULL; 527d870b4a1SAl Viro goto out; 52848a066e7SAl Viro } 52948a066e7SAl Viro 53015570086SLinus Torvalds /* 53115570086SLinus Torvalds * For a negative lookup, the lookup sequence point is the parents 53215570086SLinus Torvalds * sequence point, and it only needs to revalidate the parent dentry. 53315570086SLinus Torvalds * 53415570086SLinus Torvalds * For a positive lookup, we need to move both the parent and the 53515570086SLinus Torvalds * dentry from the RCU domain to be properly refcounted. And the 53615570086SLinus Torvalds * sequence number in the dentry validates *both* dentry counters, 53715570086SLinus Torvalds * since we checked the sequence number of the parent after we got 53815570086SLinus Torvalds * the child sequence number. So we know the parent must still 53915570086SLinus Torvalds * be valid if the child sequence number is still valid. 54015570086SLinus Torvalds */ 54119660af7SAl Viro if (!dentry) { 542e5c832d5SLinus Torvalds if (read_seqcount_retry(&parent->d_seq, nd->seq)) 543e5c832d5SLinus Torvalds goto out; 54419660af7SAl Viro BUG_ON(nd->inode != parent->d_inode); 54519660af7SAl Viro } else { 546e5c832d5SLinus Torvalds if (!lockref_get_not_dead(&dentry->d_lockref)) 547e5c832d5SLinus Torvalds goto out; 548e5c832d5SLinus Torvalds if (read_seqcount_retry(&dentry->d_seq, nd->seq)) 549e5c832d5SLinus Torvalds goto drop_dentry; 55019660af7SAl Viro } 551e5c832d5SLinus Torvalds 552e5c832d5SLinus Torvalds /* 553e5c832d5SLinus Torvalds * Sequence counts matched. Now make sure that the root is 554e5c832d5SLinus Torvalds * still valid and get it if required. 555e5c832d5SLinus Torvalds */ 556e5c832d5SLinus Torvalds if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 557e5c832d5SLinus Torvalds spin_lock(&fs->lock); 558e5c832d5SLinus Torvalds if (nd->root.mnt != fs->root.mnt || nd->root.dentry != fs->root.dentry) 559e5c832d5SLinus Torvalds goto unlock_and_drop_dentry; 56031e6b01fSNick Piggin path_get(&nd->root); 56131e6b01fSNick Piggin spin_unlock(&fs->lock); 56231e6b01fSNick Piggin } 56331e6b01fSNick Piggin 5648b61e74fSAl Viro rcu_read_unlock(); 56531e6b01fSNick Piggin return 0; 56619660af7SAl Viro 567e5c832d5SLinus Torvalds unlock_and_drop_dentry: 56831e6b01fSNick Piggin spin_unlock(&fs->lock); 569e5c832d5SLinus Torvalds drop_dentry: 5708b61e74fSAl Viro rcu_read_unlock(); 571e5c832d5SLinus Torvalds dput(dentry); 572d0d27277SLinus Torvalds goto drop_root_mnt; 573e5c832d5SLinus Torvalds out: 5748b61e74fSAl Viro rcu_read_unlock(); 575d0d27277SLinus Torvalds drop_root_mnt: 576d0d27277SLinus Torvalds if (!(nd->flags & LOOKUP_ROOT)) 577d0d27277SLinus Torvalds nd->root.mnt = NULL; 57831e6b01fSNick Piggin return -ECHILD; 57931e6b01fSNick Piggin } 58031e6b01fSNick Piggin 5814ce16ef3SAl Viro static inline int d_revalidate(struct dentry *dentry, unsigned int flags) 58234286d66SNick Piggin { 5834ce16ef3SAl Viro return dentry->d_op->d_revalidate(dentry, flags); 58434286d66SNick Piggin } 58534286d66SNick Piggin 5869f1fafeeSAl Viro /** 5879f1fafeeSAl Viro * complete_walk - successful completion of path walk 5889f1fafeeSAl Viro * @nd: pointer nameidata 58939159de2SJeff Layton * 5909f1fafeeSAl Viro * If we had been in RCU mode, drop out of it and legitimize nd->path. 5919f1fafeeSAl Viro * Revalidate the final result, unless we'd already done that during 5929f1fafeeSAl Viro * the path walk or the filesystem doesn't ask for it. Return 0 on 5939f1fafeeSAl Viro * success, -error on failure. In case of failure caller does not 5949f1fafeeSAl Viro * need to drop nd->path. 59539159de2SJeff Layton */ 5969f1fafeeSAl Viro static int complete_walk(struct nameidata *nd) 59739159de2SJeff Layton { 59816c2cd71SAl Viro struct dentry *dentry = nd->path.dentry; 59939159de2SJeff Layton int status; 60039159de2SJeff Layton 6019f1fafeeSAl Viro if (nd->flags & LOOKUP_RCU) { 6029f1fafeeSAl Viro nd->flags &= ~LOOKUP_RCU; 6039f1fafeeSAl Viro if (!(nd->flags & LOOKUP_ROOT)) 6049f1fafeeSAl Viro nd->root.mnt = NULL; 60515570086SLinus Torvalds 60648a066e7SAl Viro if (!legitimize_mnt(nd->path.mnt, nd->m_seq)) { 6078b61e74fSAl Viro rcu_read_unlock(); 60848a066e7SAl Viro return -ECHILD; 60948a066e7SAl Viro } 610e5c832d5SLinus Torvalds if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) { 6118b61e74fSAl Viro rcu_read_unlock(); 61248a066e7SAl Viro mntput(nd->path.mnt); 6139f1fafeeSAl Viro return -ECHILD; 6149f1fafeeSAl Viro } 615e5c832d5SLinus Torvalds if (read_seqcount_retry(&dentry->d_seq, nd->seq)) { 6168b61e74fSAl Viro rcu_read_unlock(); 617e5c832d5SLinus Torvalds dput(dentry); 61848a066e7SAl Viro mntput(nd->path.mnt); 619e5c832d5SLinus Torvalds return -ECHILD; 620e5c832d5SLinus Torvalds } 6218b61e74fSAl Viro rcu_read_unlock(); 6229f1fafeeSAl Viro } 6239f1fafeeSAl Viro 62416c2cd71SAl Viro if (likely(!(nd->flags & LOOKUP_JUMPED))) 62539159de2SJeff Layton return 0; 62639159de2SJeff Layton 627ecf3d1f1SJeff Layton if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) 62816c2cd71SAl Viro return 0; 62916c2cd71SAl Viro 630ecf3d1f1SJeff Layton status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); 63139159de2SJeff Layton if (status > 0) 63239159de2SJeff Layton return 0; 63339159de2SJeff Layton 63416c2cd71SAl Viro if (!status) 63539159de2SJeff Layton status = -ESTALE; 63616c2cd71SAl Viro 6379f1fafeeSAl Viro path_put(&nd->path); 63839159de2SJeff Layton return status; 63939159de2SJeff Layton } 64039159de2SJeff Layton 6412a737871SAl Viro static __always_inline void set_root(struct nameidata *nd) 6422a737871SAl Viro { 643f7ad3c6bSMiklos Szeredi if (!nd->root.mnt) 644f7ad3c6bSMiklos Szeredi get_fs_root(current->fs, &nd->root); 6452a737871SAl Viro } 6462a737871SAl Viro 6476de88d72SAl Viro static int link_path_walk(const char *, struct nameidata *); 6486de88d72SAl Viro 64931e6b01fSNick Piggin static __always_inline void set_root_rcu(struct nameidata *nd) 65031e6b01fSNick Piggin { 65131e6b01fSNick Piggin if (!nd->root.mnt) { 65231e6b01fSNick Piggin struct fs_struct *fs = current->fs; 653c28cc364SNick Piggin unsigned seq; 654c28cc364SNick Piggin 655c28cc364SNick Piggin do { 656c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 65731e6b01fSNick Piggin nd->root = fs->root; 658c1530019STim Chen nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq); 659c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 66031e6b01fSNick Piggin } 66131e6b01fSNick Piggin } 66231e6b01fSNick Piggin 6631d957f9bSJan Blunck static void path_put_conditional(struct path *path, struct nameidata *nd) 664051d3812SIan Kent { 665051d3812SIan Kent dput(path->dentry); 6664ac91378SJan Blunck if (path->mnt != nd->path.mnt) 667051d3812SIan Kent mntput(path->mnt); 668051d3812SIan Kent } 669051d3812SIan Kent 6707b9337aaSNick Piggin static inline void path_to_nameidata(const struct path *path, 6717b9337aaSNick Piggin struct nameidata *nd) 672051d3812SIan Kent { 67331e6b01fSNick Piggin if (!(nd->flags & LOOKUP_RCU)) { 6744ac91378SJan Blunck dput(nd->path.dentry); 67531e6b01fSNick Piggin if (nd->path.mnt != path->mnt) 6764ac91378SJan Blunck mntput(nd->path.mnt); 6779a229683SHuang Shijie } 67831e6b01fSNick Piggin nd->path.mnt = path->mnt; 6794ac91378SJan Blunck nd->path.dentry = path->dentry; 680051d3812SIan Kent } 681051d3812SIan Kent 682b5fb63c1SChristoph Hellwig /* 683b5fb63c1SChristoph Hellwig * Helper to directly jump to a known parsed path from ->follow_link, 684b5fb63c1SChristoph Hellwig * caller must have taken a reference to path beforehand. 685b5fb63c1SChristoph Hellwig */ 686b5fb63c1SChristoph Hellwig void nd_jump_link(struct nameidata *nd, struct path *path) 687b5fb63c1SChristoph Hellwig { 688b5fb63c1SChristoph Hellwig path_put(&nd->path); 689b5fb63c1SChristoph Hellwig 690b5fb63c1SChristoph Hellwig nd->path = *path; 691b5fb63c1SChristoph Hellwig nd->inode = nd->path.dentry->d_inode; 692b5fb63c1SChristoph Hellwig nd->flags |= LOOKUP_JUMPED; 693b5fb63c1SChristoph Hellwig } 694b5fb63c1SChristoph Hellwig 695574197e0SAl Viro static inline void put_link(struct nameidata *nd, struct path *link, void *cookie) 696574197e0SAl Viro { 697574197e0SAl Viro struct inode *inode = link->dentry->d_inode; 6986d7b5aaeSAl Viro if (inode->i_op->put_link) 699574197e0SAl Viro inode->i_op->put_link(link->dentry, nd, cookie); 700574197e0SAl Viro path_put(link); 701574197e0SAl Viro } 702574197e0SAl Viro 703561ec64aSLinus Torvalds int sysctl_protected_symlinks __read_mostly = 0; 704561ec64aSLinus Torvalds int sysctl_protected_hardlinks __read_mostly = 0; 705800179c9SKees Cook 706800179c9SKees Cook /** 707800179c9SKees Cook * may_follow_link - Check symlink following for unsafe situations 708800179c9SKees Cook * @link: The path of the symlink 70955852635SRandy Dunlap * @nd: nameidata pathwalk data 710800179c9SKees Cook * 711800179c9SKees Cook * In the case of the sysctl_protected_symlinks sysctl being enabled, 712800179c9SKees Cook * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is 713800179c9SKees Cook * in a sticky world-writable directory. This is to protect privileged 714800179c9SKees Cook * processes from failing races against path names that may change out 715800179c9SKees Cook * from under them by way of other users creating malicious symlinks. 716800179c9SKees Cook * It will permit symlinks to be followed only when outside a sticky 717800179c9SKees Cook * world-writable directory, or when the uid of the symlink and follower 718800179c9SKees Cook * match, or when the directory owner matches the symlink's owner. 719800179c9SKees Cook * 720800179c9SKees Cook * Returns 0 if following the symlink is allowed, -ve on error. 721800179c9SKees Cook */ 722800179c9SKees Cook static inline int may_follow_link(struct path *link, struct nameidata *nd) 723800179c9SKees Cook { 724800179c9SKees Cook const struct inode *inode; 725800179c9SKees Cook const struct inode *parent; 726800179c9SKees Cook 727800179c9SKees Cook if (!sysctl_protected_symlinks) 728800179c9SKees Cook return 0; 729800179c9SKees Cook 730800179c9SKees Cook /* Allowed if owner and follower match. */ 731800179c9SKees Cook inode = link->dentry->d_inode; 73281abe27bSEric W. Biederman if (uid_eq(current_cred()->fsuid, inode->i_uid)) 733800179c9SKees Cook return 0; 734800179c9SKees Cook 735800179c9SKees Cook /* Allowed if parent directory not sticky and world-writable. */ 736800179c9SKees Cook parent = nd->path.dentry->d_inode; 737800179c9SKees Cook if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) 738800179c9SKees Cook return 0; 739800179c9SKees Cook 740800179c9SKees Cook /* Allowed if parent directory and link owner match. */ 74181abe27bSEric W. Biederman if (uid_eq(parent->i_uid, inode->i_uid)) 742800179c9SKees Cook return 0; 743800179c9SKees Cook 744ffd8d101SSasha Levin audit_log_link_denied("follow_link", link); 745800179c9SKees Cook path_put_conditional(link, nd); 746800179c9SKees Cook path_put(&nd->path); 747800179c9SKees Cook return -EACCES; 748800179c9SKees Cook } 749800179c9SKees Cook 750800179c9SKees Cook /** 751800179c9SKees Cook * safe_hardlink_source - Check for safe hardlink conditions 752800179c9SKees Cook * @inode: the source inode to hardlink from 753800179c9SKees Cook * 754800179c9SKees Cook * Return false if at least one of the following conditions: 755800179c9SKees Cook * - inode is not a regular file 756800179c9SKees Cook * - inode is setuid 757800179c9SKees Cook * - inode is setgid and group-exec 758800179c9SKees Cook * - access failure for read and write 759800179c9SKees Cook * 760800179c9SKees Cook * Otherwise returns true. 761800179c9SKees Cook */ 762800179c9SKees Cook static bool safe_hardlink_source(struct inode *inode) 763800179c9SKees Cook { 764800179c9SKees Cook umode_t mode = inode->i_mode; 765800179c9SKees Cook 766800179c9SKees Cook /* Special files should not get pinned to the filesystem. */ 767800179c9SKees Cook if (!S_ISREG(mode)) 768800179c9SKees Cook return false; 769800179c9SKees Cook 770800179c9SKees Cook /* Setuid files should not get pinned to the filesystem. */ 771800179c9SKees Cook if (mode & S_ISUID) 772800179c9SKees Cook return false; 773800179c9SKees Cook 774800179c9SKees Cook /* Executable setgid files should not get pinned to the filesystem. */ 775800179c9SKees Cook if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) 776800179c9SKees Cook return false; 777800179c9SKees Cook 778800179c9SKees Cook /* Hardlinking to unreadable or unwritable sources is dangerous. */ 779800179c9SKees Cook if (inode_permission(inode, MAY_READ | MAY_WRITE)) 780800179c9SKees Cook return false; 781800179c9SKees Cook 782800179c9SKees Cook return true; 783800179c9SKees Cook } 784800179c9SKees Cook 785800179c9SKees Cook /** 786800179c9SKees Cook * may_linkat - Check permissions for creating a hardlink 787800179c9SKees Cook * @link: the source to hardlink from 788800179c9SKees Cook * 789800179c9SKees Cook * Block hardlink when all of: 790800179c9SKees Cook * - sysctl_protected_hardlinks enabled 791800179c9SKees Cook * - fsuid does not match inode 792800179c9SKees Cook * - hardlink source is unsafe (see safe_hardlink_source() above) 793800179c9SKees Cook * - not CAP_FOWNER 794800179c9SKees Cook * 795800179c9SKees Cook * Returns 0 if successful, -ve on error. 796800179c9SKees Cook */ 797800179c9SKees Cook static int may_linkat(struct path *link) 798800179c9SKees Cook { 799800179c9SKees Cook const struct cred *cred; 800800179c9SKees Cook struct inode *inode; 801800179c9SKees Cook 802800179c9SKees Cook if (!sysctl_protected_hardlinks) 803800179c9SKees Cook return 0; 804800179c9SKees Cook 805800179c9SKees Cook cred = current_cred(); 806800179c9SKees Cook inode = link->dentry->d_inode; 807800179c9SKees Cook 808800179c9SKees Cook /* Source inode owner (or CAP_FOWNER) can hardlink all they like, 809800179c9SKees Cook * otherwise, it must be a safe source. 810800179c9SKees Cook */ 81181abe27bSEric W. Biederman if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) || 812800179c9SKees Cook capable(CAP_FOWNER)) 813800179c9SKees Cook return 0; 814800179c9SKees Cook 815a51d9eaaSKees Cook audit_log_link_denied("linkat", link); 816800179c9SKees Cook return -EPERM; 817800179c9SKees Cook } 818800179c9SKees Cook 819def4af30SAl Viro static __always_inline int 820574197e0SAl Viro follow_link(struct path *link, struct nameidata *nd, void **p) 8211da177e4SLinus Torvalds { 8227b9337aaSNick Piggin struct dentry *dentry = link->dentry; 8236d7b5aaeSAl Viro int error; 8246d7b5aaeSAl Viro char *s; 8251da177e4SLinus Torvalds 826844a3917SAl Viro BUG_ON(nd->flags & LOOKUP_RCU); 827844a3917SAl Viro 8280e794589SAl Viro if (link->mnt == nd->path.mnt) 8290e794589SAl Viro mntget(link->mnt); 8300e794589SAl Viro 8316d7b5aaeSAl Viro error = -ELOOP; 8326d7b5aaeSAl Viro if (unlikely(current->total_link_count >= 40)) 8336d7b5aaeSAl Viro goto out_put_nd_path; 8346d7b5aaeSAl Viro 835574197e0SAl Viro cond_resched(); 836574197e0SAl Viro current->total_link_count++; 837574197e0SAl Viro 83868ac1234SAl Viro touch_atime(link); 8391da177e4SLinus Torvalds nd_set_link(nd, NULL); 840cd4e91d3SAl Viro 84136f3b4f6SAl Viro error = security_inode_follow_link(link->dentry, nd); 8426d7b5aaeSAl Viro if (error) 8436d7b5aaeSAl Viro goto out_put_nd_path; 84436f3b4f6SAl Viro 84586acdca1SAl Viro nd->last_type = LAST_BIND; 846def4af30SAl Viro *p = dentry->d_inode->i_op->follow_link(dentry, nd); 847def4af30SAl Viro error = PTR_ERR(*p); 8486d7b5aaeSAl Viro if (IS_ERR(*p)) 849408ef013SChristoph Hellwig goto out_put_nd_path; 8506d7b5aaeSAl Viro 851cc314eefSLinus Torvalds error = 0; 8526d7b5aaeSAl Viro s = nd_get_link(nd); 8536d7b5aaeSAl Viro if (s) { 854443ed254SAl Viro if (unlikely(IS_ERR(s))) { 855443ed254SAl Viro path_put(&nd->path); 856443ed254SAl Viro put_link(nd, link, *p); 857443ed254SAl Viro return PTR_ERR(s); 858443ed254SAl Viro } 859443ed254SAl Viro if (*s == '/') { 860443ed254SAl Viro set_root(nd); 861443ed254SAl Viro path_put(&nd->path); 862443ed254SAl Viro nd->path = nd->root; 863443ed254SAl Viro path_get(&nd->root); 864443ed254SAl Viro nd->flags |= LOOKUP_JUMPED; 865443ed254SAl Viro } 866443ed254SAl Viro nd->inode = nd->path.dentry->d_inode; 867443ed254SAl Viro error = link_path_walk(s, nd); 8686d7b5aaeSAl Viro if (unlikely(error)) 8696d7b5aaeSAl Viro put_link(nd, link, *p); 870b5fb63c1SChristoph Hellwig } 8716d7b5aaeSAl Viro 8726d7b5aaeSAl Viro return error; 8736d7b5aaeSAl Viro 8746d7b5aaeSAl Viro out_put_nd_path: 87598f6ef64SArnd Bergmann *p = NULL; 8766d7b5aaeSAl Viro path_put(&nd->path); 8776d7b5aaeSAl Viro path_put(link); 8781da177e4SLinus Torvalds return error; 8791da177e4SLinus Torvalds } 8801da177e4SLinus Torvalds 88131e6b01fSNick Piggin static int follow_up_rcu(struct path *path) 88231e6b01fSNick Piggin { 8830714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 8840714a533SAl Viro struct mount *parent; 88531e6b01fSNick Piggin struct dentry *mountpoint; 88631e6b01fSNick Piggin 8870714a533SAl Viro parent = mnt->mnt_parent; 8880714a533SAl Viro if (&parent->mnt == path->mnt) 88931e6b01fSNick Piggin return 0; 890a73324daSAl Viro mountpoint = mnt->mnt_mountpoint; 89131e6b01fSNick Piggin path->dentry = mountpoint; 8920714a533SAl Viro path->mnt = &parent->mnt; 89331e6b01fSNick Piggin return 1; 89431e6b01fSNick Piggin } 89531e6b01fSNick Piggin 896f015f126SDavid Howells /* 897f015f126SDavid Howells * follow_up - Find the mountpoint of path's vfsmount 898f015f126SDavid Howells * 899f015f126SDavid Howells * Given a path, find the mountpoint of its source file system. 900f015f126SDavid Howells * Replace @path with the path of the mountpoint in the parent mount. 901f015f126SDavid Howells * Up is towards /. 902f015f126SDavid Howells * 903f015f126SDavid Howells * Return 1 if we went up a level and 0 if we were already at the 904f015f126SDavid Howells * root. 905f015f126SDavid Howells */ 906bab77ebfSAl Viro int follow_up(struct path *path) 9071da177e4SLinus Torvalds { 9080714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 9090714a533SAl Viro struct mount *parent; 9101da177e4SLinus Torvalds struct dentry *mountpoint; 91199b7db7bSNick Piggin 91248a066e7SAl Viro read_seqlock_excl(&mount_lock); 9130714a533SAl Viro parent = mnt->mnt_parent; 9143c0a6163SAl Viro if (parent == mnt) { 91548a066e7SAl Viro read_sequnlock_excl(&mount_lock); 9161da177e4SLinus Torvalds return 0; 9171da177e4SLinus Torvalds } 9180714a533SAl Viro mntget(&parent->mnt); 919a73324daSAl Viro mountpoint = dget(mnt->mnt_mountpoint); 92048a066e7SAl Viro read_sequnlock_excl(&mount_lock); 921bab77ebfSAl Viro dput(path->dentry); 922bab77ebfSAl Viro path->dentry = mountpoint; 923bab77ebfSAl Viro mntput(path->mnt); 9240714a533SAl Viro path->mnt = &parent->mnt; 9251da177e4SLinus Torvalds return 1; 9261da177e4SLinus Torvalds } 9271da177e4SLinus Torvalds 928b5c84bf6SNick Piggin /* 9299875cf80SDavid Howells * Perform an automount 9309875cf80SDavid Howells * - return -EISDIR to tell follow_managed() to stop and return the path we 9319875cf80SDavid Howells * were called with. 9321da177e4SLinus Torvalds */ 9339875cf80SDavid Howells static int follow_automount(struct path *path, unsigned flags, 9349875cf80SDavid Howells bool *need_mntput) 93531e6b01fSNick Piggin { 9369875cf80SDavid Howells struct vfsmount *mnt; 937ea5b778aSDavid Howells int err; 9389875cf80SDavid Howells 9399875cf80SDavid Howells if (!path->dentry->d_op || !path->dentry->d_op->d_automount) 9409875cf80SDavid Howells return -EREMOTE; 9419875cf80SDavid Howells 9420ec26fd0SMiklos Szeredi /* We don't want to mount if someone's just doing a stat - 9430ec26fd0SMiklos Szeredi * unless they're stat'ing a directory and appended a '/' to 9440ec26fd0SMiklos Szeredi * the name. 9450ec26fd0SMiklos Szeredi * 9460ec26fd0SMiklos Szeredi * We do, however, want to mount if someone wants to open or 9475a30d8a2SDavid Howells * create a file of any type under the mountpoint, wants to 9480ec26fd0SMiklos Szeredi * traverse through the mountpoint or wants to open the 9490ec26fd0SMiklos Szeredi * mounted directory. Also, autofs may mark negative dentries 9500ec26fd0SMiklos Szeredi * as being automount points. These will need the attentions 9510ec26fd0SMiklos Szeredi * of the daemon to instantiate them before they can be used. 9525a30d8a2SDavid Howells */ 9535a30d8a2SDavid Howells if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | 954d94c177bSLinus Torvalds LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && 9555a30d8a2SDavid Howells path->dentry->d_inode) 9569875cf80SDavid Howells return -EISDIR; 9570ec26fd0SMiklos Szeredi 9589875cf80SDavid Howells current->total_link_count++; 9599875cf80SDavid Howells if (current->total_link_count >= 40) 9609875cf80SDavid Howells return -ELOOP; 9619875cf80SDavid Howells 9629875cf80SDavid Howells mnt = path->dentry->d_op->d_automount(path); 9639875cf80SDavid Howells if (IS_ERR(mnt)) { 9649875cf80SDavid Howells /* 9659875cf80SDavid Howells * The filesystem is allowed to return -EISDIR here to indicate 9669875cf80SDavid Howells * it doesn't want to automount. For instance, autofs would do 9679875cf80SDavid Howells * this so that its userspace daemon can mount on this dentry. 9689875cf80SDavid Howells * 9699875cf80SDavid Howells * However, we can only permit this if it's a terminal point in 9709875cf80SDavid Howells * the path being looked up; if it wasn't then the remainder of 9719875cf80SDavid Howells * the path is inaccessible and we should say so. 9729875cf80SDavid Howells */ 97349084c3bSAl Viro if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT)) 9749875cf80SDavid Howells return -EREMOTE; 9759875cf80SDavid Howells return PTR_ERR(mnt); 97631e6b01fSNick Piggin } 977ea5b778aSDavid Howells 9789875cf80SDavid Howells if (!mnt) /* mount collision */ 9799875cf80SDavid Howells return 0; 9809875cf80SDavid Howells 9818aef1884SAl Viro if (!*need_mntput) { 9828aef1884SAl Viro /* lock_mount() may release path->mnt on error */ 9838aef1884SAl Viro mntget(path->mnt); 9848aef1884SAl Viro *need_mntput = true; 9858aef1884SAl Viro } 98619a167afSAl Viro err = finish_automount(mnt, path); 987ea5b778aSDavid Howells 988ea5b778aSDavid Howells switch (err) { 989ea5b778aSDavid Howells case -EBUSY: 990ea5b778aSDavid Howells /* Someone else made a mount here whilst we were busy */ 99119a167afSAl Viro return 0; 992ea5b778aSDavid Howells case 0: 9938aef1884SAl Viro path_put(path); 9949875cf80SDavid Howells path->mnt = mnt; 9959875cf80SDavid Howells path->dentry = dget(mnt->mnt_root); 9969875cf80SDavid Howells return 0; 99719a167afSAl Viro default: 99819a167afSAl Viro return err; 9999875cf80SDavid Howells } 100019a167afSAl Viro 1001ea5b778aSDavid Howells } 10029875cf80SDavid Howells 10039875cf80SDavid Howells /* 10049875cf80SDavid Howells * Handle a dentry that is managed in some way. 1005cc53ce53SDavid Howells * - Flagged for transit management (autofs) 10069875cf80SDavid Howells * - Flagged as mountpoint 10079875cf80SDavid Howells * - Flagged as automount point 10089875cf80SDavid Howells * 10099875cf80SDavid Howells * This may only be called in refwalk mode. 10109875cf80SDavid Howells * 10119875cf80SDavid Howells * Serialization is taken care of in namespace.c 10129875cf80SDavid Howells */ 10139875cf80SDavid Howells static int follow_managed(struct path *path, unsigned flags) 10149875cf80SDavid Howells { 10158aef1884SAl Viro struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */ 10169875cf80SDavid Howells unsigned managed; 10179875cf80SDavid Howells bool need_mntput = false; 10188aef1884SAl Viro int ret = 0; 10199875cf80SDavid Howells 10209875cf80SDavid Howells /* Given that we're not holding a lock here, we retain the value in a 10219875cf80SDavid Howells * local variable for each dentry as we look at it so that we don't see 10229875cf80SDavid Howells * the components of that value change under us */ 10239875cf80SDavid Howells while (managed = ACCESS_ONCE(path->dentry->d_flags), 10249875cf80SDavid Howells managed &= DCACHE_MANAGED_DENTRY, 10259875cf80SDavid Howells unlikely(managed != 0)) { 1026cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1027cc53ce53SDavid Howells * being held. */ 1028cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 1029cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1030cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 10311aed3e42SAl Viro ret = path->dentry->d_op->d_manage(path->dentry, false); 1032cc53ce53SDavid Howells if (ret < 0) 10338aef1884SAl Viro break; 1034cc53ce53SDavid Howells } 1035cc53ce53SDavid Howells 10369875cf80SDavid Howells /* Transit to a mounted filesystem. */ 10379875cf80SDavid Howells if (managed & DCACHE_MOUNTED) { 10389875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 10399875cf80SDavid Howells if (mounted) { 10409875cf80SDavid Howells dput(path->dentry); 10419875cf80SDavid Howells if (need_mntput) 1042463ffb2eSAl Viro mntput(path->mnt); 1043463ffb2eSAl Viro path->mnt = mounted; 1044463ffb2eSAl Viro path->dentry = dget(mounted->mnt_root); 10459875cf80SDavid Howells need_mntput = true; 10469875cf80SDavid Howells continue; 1047463ffb2eSAl Viro } 1048463ffb2eSAl Viro 10499875cf80SDavid Howells /* Something is mounted on this dentry in another 10509875cf80SDavid Howells * namespace and/or whatever was mounted there in this 105148a066e7SAl Viro * namespace got unmounted before lookup_mnt() could 105248a066e7SAl Viro * get it */ 10531da177e4SLinus Torvalds } 10549875cf80SDavid Howells 10559875cf80SDavid Howells /* Handle an automount point */ 10569875cf80SDavid Howells if (managed & DCACHE_NEED_AUTOMOUNT) { 10579875cf80SDavid Howells ret = follow_automount(path, flags, &need_mntput); 10589875cf80SDavid Howells if (ret < 0) 10598aef1884SAl Viro break; 10609875cf80SDavid Howells continue; 10619875cf80SDavid Howells } 10629875cf80SDavid Howells 10639875cf80SDavid Howells /* We didn't change the current path point */ 10649875cf80SDavid Howells break; 10659875cf80SDavid Howells } 10668aef1884SAl Viro 10678aef1884SAl Viro if (need_mntput && path->mnt == mnt) 10688aef1884SAl Viro mntput(path->mnt); 10698aef1884SAl Viro if (ret == -EISDIR) 10708aef1884SAl Viro ret = 0; 1071a3fbbde7SAl Viro return ret < 0 ? ret : need_mntput; 10721da177e4SLinus Torvalds } 10731da177e4SLinus Torvalds 1074cc53ce53SDavid Howells int follow_down_one(struct path *path) 10751da177e4SLinus Torvalds { 10761da177e4SLinus Torvalds struct vfsmount *mounted; 10771da177e4SLinus Torvalds 10781c755af4SAl Viro mounted = lookup_mnt(path); 10791da177e4SLinus Torvalds if (mounted) { 10809393bd07SAl Viro dput(path->dentry); 10819393bd07SAl Viro mntput(path->mnt); 10829393bd07SAl Viro path->mnt = mounted; 10839393bd07SAl Viro path->dentry = dget(mounted->mnt_root); 10841da177e4SLinus Torvalds return 1; 10851da177e4SLinus Torvalds } 10861da177e4SLinus Torvalds return 0; 10871da177e4SLinus Torvalds } 10881da177e4SLinus Torvalds 108962a7375eSIan Kent static inline bool managed_dentry_might_block(struct dentry *dentry) 109062a7375eSIan Kent { 109162a7375eSIan Kent return (dentry->d_flags & DCACHE_MANAGE_TRANSIT && 109262a7375eSIan Kent dentry->d_op->d_manage(dentry, true) < 0); 109362a7375eSIan Kent } 109462a7375eSIan Kent 10959875cf80SDavid Howells /* 1096287548e4SAl Viro * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if 1097287548e4SAl Viro * we meet a managed dentry that would need blocking. 10989875cf80SDavid Howells */ 10999875cf80SDavid Howells static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, 1100287548e4SAl Viro struct inode **inode) 11019875cf80SDavid Howells { 110262a7375eSIan Kent for (;;) { 1103c7105365SAl Viro struct mount *mounted; 110462a7375eSIan Kent /* 110562a7375eSIan Kent * Don't forget we might have a non-mountpoint managed dentry 110662a7375eSIan Kent * that wants to block transit. 110762a7375eSIan Kent */ 1108287548e4SAl Viro if (unlikely(managed_dentry_might_block(path->dentry))) 1109ab90911fSDavid Howells return false; 111062a7375eSIan Kent 111162a7375eSIan Kent if (!d_mountpoint(path->dentry)) 1112b37199e6SAl Viro return true; 111362a7375eSIan Kent 1114474279dcSAl Viro mounted = __lookup_mnt(path->mnt, path->dentry); 11159875cf80SDavid Howells if (!mounted) 11169875cf80SDavid Howells break; 1117c7105365SAl Viro path->mnt = &mounted->mnt; 1118c7105365SAl Viro path->dentry = mounted->mnt.mnt_root; 1119a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 11209875cf80SDavid Howells nd->seq = read_seqcount_begin(&path->dentry->d_seq); 112159430262SLinus Torvalds /* 112259430262SLinus Torvalds * Update the inode too. We don't need to re-check the 112359430262SLinus Torvalds * dentry sequence number here after this d_inode read, 112459430262SLinus Torvalds * because a mount-point is always pinned. 112559430262SLinus Torvalds */ 112659430262SLinus Torvalds *inode = path->dentry->d_inode; 11279875cf80SDavid Howells } 1128b37199e6SAl Viro return read_seqretry(&mount_lock, nd->m_seq); 1129287548e4SAl Viro } 1130287548e4SAl Viro 113131e6b01fSNick Piggin static int follow_dotdot_rcu(struct nameidata *nd) 113231e6b01fSNick Piggin { 113331e6b01fSNick Piggin set_root_rcu(nd); 113431e6b01fSNick Piggin 113531e6b01fSNick Piggin while (1) { 113631e6b01fSNick Piggin if (nd->path.dentry == nd->root.dentry && 113731e6b01fSNick Piggin nd->path.mnt == nd->root.mnt) { 113831e6b01fSNick Piggin break; 113931e6b01fSNick Piggin } 114031e6b01fSNick Piggin if (nd->path.dentry != nd->path.mnt->mnt_root) { 114131e6b01fSNick Piggin struct dentry *old = nd->path.dentry; 114231e6b01fSNick Piggin struct dentry *parent = old->d_parent; 114331e6b01fSNick Piggin unsigned seq; 114431e6b01fSNick Piggin 114531e6b01fSNick Piggin seq = read_seqcount_begin(&parent->d_seq); 114631e6b01fSNick Piggin if (read_seqcount_retry(&old->d_seq, nd->seq)) 1147ef7562d5SAl Viro goto failed; 114831e6b01fSNick Piggin nd->path.dentry = parent; 114931e6b01fSNick Piggin nd->seq = seq; 115031e6b01fSNick Piggin break; 115131e6b01fSNick Piggin } 115231e6b01fSNick Piggin if (!follow_up_rcu(&nd->path)) 115331e6b01fSNick Piggin break; 115431e6b01fSNick Piggin nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 115531e6b01fSNick Piggin } 1156b37199e6SAl Viro while (d_mountpoint(nd->path.dentry)) { 1157b37199e6SAl Viro struct mount *mounted; 1158b37199e6SAl Viro mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry); 1159b37199e6SAl Viro if (!mounted) 1160b37199e6SAl Viro break; 1161b37199e6SAl Viro nd->path.mnt = &mounted->mnt; 1162b37199e6SAl Viro nd->path.dentry = mounted->mnt.mnt_root; 1163b37199e6SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 1164b37199e6SAl Viro if (!read_seqretry(&mount_lock, nd->m_seq)) 1165b37199e6SAl Viro goto failed; 1166b37199e6SAl Viro } 1167dea39376SAl Viro nd->inode = nd->path.dentry->d_inode; 116831e6b01fSNick Piggin return 0; 1169ef7562d5SAl Viro 1170ef7562d5SAl Viro failed: 1171ef7562d5SAl Viro nd->flags &= ~LOOKUP_RCU; 11725b6ca027SAl Viro if (!(nd->flags & LOOKUP_ROOT)) 1173ef7562d5SAl Viro nd->root.mnt = NULL; 11748b61e74fSAl Viro rcu_read_unlock(); 1175ef7562d5SAl Viro return -ECHILD; 117631e6b01fSNick Piggin } 117731e6b01fSNick Piggin 11789875cf80SDavid Howells /* 1179cc53ce53SDavid Howells * Follow down to the covering mount currently visible to userspace. At each 1180cc53ce53SDavid Howells * point, the filesystem owning that dentry may be queried as to whether the 1181cc53ce53SDavid Howells * caller is permitted to proceed or not. 1182cc53ce53SDavid Howells */ 11837cc90cc3SAl Viro int follow_down(struct path *path) 1184cc53ce53SDavid Howells { 1185cc53ce53SDavid Howells unsigned managed; 1186cc53ce53SDavid Howells int ret; 1187cc53ce53SDavid Howells 1188cc53ce53SDavid Howells while (managed = ACCESS_ONCE(path->dentry->d_flags), 1189cc53ce53SDavid Howells unlikely(managed & DCACHE_MANAGED_DENTRY)) { 1190cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1191cc53ce53SDavid Howells * being held. 1192cc53ce53SDavid Howells * 1193cc53ce53SDavid Howells * We indicate to the filesystem if someone is trying to mount 1194cc53ce53SDavid Howells * something here. This gives autofs the chance to deny anyone 1195cc53ce53SDavid Howells * other than its daemon the right to mount on its 1196cc53ce53SDavid Howells * superstructure. 1197cc53ce53SDavid Howells * 1198cc53ce53SDavid Howells * The filesystem may sleep at this point. 1199cc53ce53SDavid Howells */ 1200cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 1201cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1202cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 1203ab90911fSDavid Howells ret = path->dentry->d_op->d_manage( 12041aed3e42SAl Viro path->dentry, false); 1205cc53ce53SDavid Howells if (ret < 0) 1206cc53ce53SDavid Howells return ret == -EISDIR ? 0 : ret; 1207cc53ce53SDavid Howells } 1208cc53ce53SDavid Howells 1209cc53ce53SDavid Howells /* Transit to a mounted filesystem. */ 1210cc53ce53SDavid Howells if (managed & DCACHE_MOUNTED) { 1211cc53ce53SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 1212cc53ce53SDavid Howells if (!mounted) 1213cc53ce53SDavid Howells break; 1214cc53ce53SDavid Howells dput(path->dentry); 1215cc53ce53SDavid Howells mntput(path->mnt); 1216cc53ce53SDavid Howells path->mnt = mounted; 1217cc53ce53SDavid Howells path->dentry = dget(mounted->mnt_root); 1218cc53ce53SDavid Howells continue; 1219cc53ce53SDavid Howells } 1220cc53ce53SDavid Howells 1221cc53ce53SDavid Howells /* Don't handle automount points here */ 1222cc53ce53SDavid Howells break; 1223cc53ce53SDavid Howells } 1224cc53ce53SDavid Howells return 0; 1225cc53ce53SDavid Howells } 1226cc53ce53SDavid Howells 1227cc53ce53SDavid Howells /* 12289875cf80SDavid Howells * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() 12299875cf80SDavid Howells */ 12309875cf80SDavid Howells static void follow_mount(struct path *path) 12319875cf80SDavid Howells { 12329875cf80SDavid Howells while (d_mountpoint(path->dentry)) { 12339875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 12349875cf80SDavid Howells if (!mounted) 12359875cf80SDavid Howells break; 12369875cf80SDavid Howells dput(path->dentry); 12379875cf80SDavid Howells mntput(path->mnt); 12389875cf80SDavid Howells path->mnt = mounted; 12399875cf80SDavid Howells path->dentry = dget(mounted->mnt_root); 12409875cf80SDavid Howells } 12419875cf80SDavid Howells } 12429875cf80SDavid Howells 124331e6b01fSNick Piggin static void follow_dotdot(struct nameidata *nd) 12441da177e4SLinus Torvalds { 12452a737871SAl Viro set_root(nd); 1246e518ddb7SAndreas Mohr 12471da177e4SLinus Torvalds while(1) { 12484ac91378SJan Blunck struct dentry *old = nd->path.dentry; 12491da177e4SLinus Torvalds 12502a737871SAl Viro if (nd->path.dentry == nd->root.dentry && 12512a737871SAl Viro nd->path.mnt == nd->root.mnt) { 12521da177e4SLinus Torvalds break; 12531da177e4SLinus Torvalds } 12544ac91378SJan Blunck if (nd->path.dentry != nd->path.mnt->mnt_root) { 12553088dd70SAl Viro /* rare case of legitimate dget_parent()... */ 12563088dd70SAl Viro nd->path.dentry = dget_parent(nd->path.dentry); 12571da177e4SLinus Torvalds dput(old); 12581da177e4SLinus Torvalds break; 12591da177e4SLinus Torvalds } 12603088dd70SAl Viro if (!follow_up(&nd->path)) 12611da177e4SLinus Torvalds break; 12621da177e4SLinus Torvalds } 126379ed0226SAl Viro follow_mount(&nd->path); 126431e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 12651da177e4SLinus Torvalds } 12661da177e4SLinus Torvalds 12671da177e4SLinus Torvalds /* 1268bad61189SMiklos Szeredi * This looks up the name in dcache, possibly revalidates the old dentry and 1269bad61189SMiklos Szeredi * allocates a new one if not found or not valid. In the need_lookup argument 1270bad61189SMiklos Szeredi * returns whether i_op->lookup is necessary. 1271bad61189SMiklos Szeredi * 1272bad61189SMiklos Szeredi * dir->d_inode->i_mutex must be held 1273baa03890SNick Piggin */ 1274bad61189SMiklos Szeredi static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir, 1275201f956eSAl Viro unsigned int flags, bool *need_lookup) 1276baa03890SNick Piggin { 1277baa03890SNick Piggin struct dentry *dentry; 1278bad61189SMiklos Szeredi int error; 1279baa03890SNick Piggin 1280bad61189SMiklos Szeredi *need_lookup = false; 1281bad61189SMiklos Szeredi dentry = d_lookup(dir, name); 1282bad61189SMiklos Szeredi if (dentry) { 128339e3c955SJeff Layton if (dentry->d_flags & DCACHE_OP_REVALIDATE) { 1284201f956eSAl Viro error = d_revalidate(dentry, flags); 1285bad61189SMiklos Szeredi if (unlikely(error <= 0)) { 1286bad61189SMiklos Szeredi if (error < 0) { 1287bad61189SMiklos Szeredi dput(dentry); 1288bad61189SMiklos Szeredi return ERR_PTR(error); 1289bad61189SMiklos Szeredi } else if (!d_invalidate(dentry)) { 1290bad61189SMiklos Szeredi dput(dentry); 1291bad61189SMiklos Szeredi dentry = NULL; 1292bad61189SMiklos Szeredi } 1293bad61189SMiklos Szeredi } 1294bad61189SMiklos Szeredi } 1295bad61189SMiklos Szeredi } 1296baa03890SNick Piggin 1297bad61189SMiklos Szeredi if (!dentry) { 1298bad61189SMiklos Szeredi dentry = d_alloc(dir, name); 1299baa03890SNick Piggin if (unlikely(!dentry)) 1300baa03890SNick Piggin return ERR_PTR(-ENOMEM); 1301baa03890SNick Piggin 1302bad61189SMiklos Szeredi *need_lookup = true; 1303baa03890SNick Piggin } 1304baa03890SNick Piggin return dentry; 1305baa03890SNick Piggin } 1306baa03890SNick Piggin 1307baa03890SNick Piggin /* 130813a2c3beSJ. Bruce Fields * Call i_op->lookup on the dentry. The dentry must be negative and 130913a2c3beSJ. Bruce Fields * unhashed. 1310bad61189SMiklos Szeredi * 1311bad61189SMiklos Szeredi * dir->d_inode->i_mutex must be held 131244396f4bSJosef Bacik */ 1313bad61189SMiklos Szeredi static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry, 131472bd866aSAl Viro unsigned int flags) 131544396f4bSJosef Bacik { 131644396f4bSJosef Bacik struct dentry *old; 131744396f4bSJosef Bacik 131844396f4bSJosef Bacik /* Don't create child dentry for a dead directory. */ 1319bad61189SMiklos Szeredi if (unlikely(IS_DEADDIR(dir))) { 1320e188dc02SMiklos Szeredi dput(dentry); 132144396f4bSJosef Bacik return ERR_PTR(-ENOENT); 1322e188dc02SMiklos Szeredi } 132344396f4bSJosef Bacik 132472bd866aSAl Viro old = dir->i_op->lookup(dir, dentry, flags); 132544396f4bSJosef Bacik if (unlikely(old)) { 132644396f4bSJosef Bacik dput(dentry); 132744396f4bSJosef Bacik dentry = old; 132844396f4bSJosef Bacik } 132944396f4bSJosef Bacik return dentry; 133044396f4bSJosef Bacik } 133144396f4bSJosef Bacik 1332a3255546SAl Viro static struct dentry *__lookup_hash(struct qstr *name, 133372bd866aSAl Viro struct dentry *base, unsigned int flags) 1334a3255546SAl Viro { 1335bad61189SMiklos Szeredi bool need_lookup; 1336a3255546SAl Viro struct dentry *dentry; 1337a3255546SAl Viro 133872bd866aSAl Viro dentry = lookup_dcache(name, base, flags, &need_lookup); 1339bad61189SMiklos Szeredi if (!need_lookup) 1340a3255546SAl Viro return dentry; 1341bad61189SMiklos Szeredi 134272bd866aSAl Viro return lookup_real(base->d_inode, dentry, flags); 1343a3255546SAl Viro } 1344a3255546SAl Viro 134544396f4bSJosef Bacik /* 13461da177e4SLinus Torvalds * It's more convoluted than I'd like it to be, but... it's still fairly 13471da177e4SLinus Torvalds * small and for now I'd prefer to have fast path as straight as possible. 13481da177e4SLinus Torvalds * It _is_ time-critical. 13491da177e4SLinus Torvalds */ 1350e97cdc87SAl Viro static int lookup_fast(struct nameidata *nd, 135131e6b01fSNick Piggin struct path *path, struct inode **inode) 13521da177e4SLinus Torvalds { 13534ac91378SJan Blunck struct vfsmount *mnt = nd->path.mnt; 135431e6b01fSNick Piggin struct dentry *dentry, *parent = nd->path.dentry; 13555a18fff2SAl Viro int need_reval = 1; 13565a18fff2SAl Viro int status = 1; 13579875cf80SDavid Howells int err; 13589875cf80SDavid Howells 13593cac260aSAl Viro /* 1360b04f784eSNick Piggin * Rename seqlock is not required here because in the off chance 1361b04f784eSNick Piggin * of a false negative due to a concurrent rename, we're going to 1362b04f784eSNick Piggin * do the non-racy lookup, below. 1363b04f784eSNick Piggin */ 136431e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) { 136531e6b01fSNick Piggin unsigned seq; 1366da53be12SLinus Torvalds dentry = __d_lookup_rcu(parent, &nd->last, &seq); 13675a18fff2SAl Viro if (!dentry) 13685a18fff2SAl Viro goto unlazy; 13695a18fff2SAl Viro 137012f8ad4bSLinus Torvalds /* 137112f8ad4bSLinus Torvalds * This sequence count validates that the inode matches 137212f8ad4bSLinus Torvalds * the dentry name information from lookup. 137312f8ad4bSLinus Torvalds */ 137412f8ad4bSLinus Torvalds *inode = dentry->d_inode; 137512f8ad4bSLinus Torvalds if (read_seqcount_retry(&dentry->d_seq, seq)) 137612f8ad4bSLinus Torvalds return -ECHILD; 137712f8ad4bSLinus Torvalds 137812f8ad4bSLinus Torvalds /* 137912f8ad4bSLinus Torvalds * This sequence count validates that the parent had no 138012f8ad4bSLinus Torvalds * changes while we did the lookup of the dentry above. 138112f8ad4bSLinus Torvalds * 138212f8ad4bSLinus Torvalds * The memory barrier in read_seqcount_begin of child is 138312f8ad4bSLinus Torvalds * enough, we can use __read_seqcount_retry here. 138412f8ad4bSLinus Torvalds */ 138531e6b01fSNick Piggin if (__read_seqcount_retry(&parent->d_seq, nd->seq)) 138631e6b01fSNick Piggin return -ECHILD; 138731e6b01fSNick Piggin nd->seq = seq; 13885a18fff2SAl Viro 138924643087SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) { 13904ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 13915a18fff2SAl Viro if (unlikely(status <= 0)) { 13925a18fff2SAl Viro if (status != -ECHILD) 13935a18fff2SAl Viro need_reval = 0; 13945a18fff2SAl Viro goto unlazy; 13955a18fff2SAl Viro } 139624643087SAl Viro } 139731e6b01fSNick Piggin path->mnt = mnt; 139831e6b01fSNick Piggin path->dentry = dentry; 1399d6e9bd25SAl Viro if (unlikely(!__follow_mount_rcu(nd, path, inode))) 1400d6e9bd25SAl Viro goto unlazy; 1401d6e9bd25SAl Viro if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT)) 1402d6e9bd25SAl Viro goto unlazy; 14039875cf80SDavid Howells return 0; 14045a18fff2SAl Viro unlazy: 140519660af7SAl Viro if (unlazy_walk(nd, dentry)) 14065a18fff2SAl Viro return -ECHILD; 14075a18fff2SAl Viro } else { 1408e97cdc87SAl Viro dentry = __d_lookup(parent, &nd->last); 140924643087SAl Viro } 14105a18fff2SAl Viro 141181e6f520SAl Viro if (unlikely(!dentry)) 141281e6f520SAl Viro goto need_lookup; 14135a18fff2SAl Viro 14145a18fff2SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval) 14154ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 14165a18fff2SAl Viro if (unlikely(status <= 0)) { 14175a18fff2SAl Viro if (status < 0) { 14185a18fff2SAl Viro dput(dentry); 14195a18fff2SAl Viro return status; 14205a18fff2SAl Viro } 14215a18fff2SAl Viro if (!d_invalidate(dentry)) { 14225a18fff2SAl Viro dput(dentry); 142381e6f520SAl Viro goto need_lookup; 14245a18fff2SAl Viro } 14255a18fff2SAl Viro } 1426697f514dSMiklos Szeredi 14271da177e4SLinus Torvalds path->mnt = mnt; 14281da177e4SLinus Torvalds path->dentry = dentry; 14299875cf80SDavid Howells err = follow_managed(path, nd->flags); 143089312214SIan Kent if (unlikely(err < 0)) { 143189312214SIan Kent path_put_conditional(path, nd); 14329875cf80SDavid Howells return err; 143389312214SIan Kent } 1434a3fbbde7SAl Viro if (err) 1435a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 143631e6b01fSNick Piggin *inode = path->dentry->d_inode; 14371da177e4SLinus Torvalds return 0; 143881e6f520SAl Viro 143981e6f520SAl Viro need_lookup: 1440697f514dSMiklos Szeredi return 1; 1441697f514dSMiklos Szeredi } 1442697f514dSMiklos Szeredi 1443697f514dSMiklos Szeredi /* Fast lookup failed, do it the slow way */ 1444cc2a5271SAl Viro static int lookup_slow(struct nameidata *nd, struct path *path) 1445697f514dSMiklos Szeredi { 1446697f514dSMiklos Szeredi struct dentry *dentry, *parent; 1447697f514dSMiklos Szeredi int err; 1448697f514dSMiklos Szeredi 1449697f514dSMiklos Szeredi parent = nd->path.dentry; 145081e6f520SAl Viro BUG_ON(nd->inode != parent->d_inode); 145181e6f520SAl Viro 145281e6f520SAl Viro mutex_lock(&parent->d_inode->i_mutex); 1453cc2a5271SAl Viro dentry = __lookup_hash(&nd->last, parent, nd->flags); 145481e6f520SAl Viro mutex_unlock(&parent->d_inode->i_mutex); 145581e6f520SAl Viro if (IS_ERR(dentry)) 145681e6f520SAl Viro return PTR_ERR(dentry); 1457697f514dSMiklos Szeredi path->mnt = nd->path.mnt; 1458697f514dSMiklos Szeredi path->dentry = dentry; 1459697f514dSMiklos Szeredi err = follow_managed(path, nd->flags); 1460697f514dSMiklos Szeredi if (unlikely(err < 0)) { 1461697f514dSMiklos Szeredi path_put_conditional(path, nd); 1462697f514dSMiklos Szeredi return err; 1463697f514dSMiklos Szeredi } 1464697f514dSMiklos Szeredi if (err) 1465697f514dSMiklos Szeredi nd->flags |= LOOKUP_JUMPED; 1466697f514dSMiklos Szeredi return 0; 14671da177e4SLinus Torvalds } 14681da177e4SLinus Torvalds 146952094c8aSAl Viro static inline int may_lookup(struct nameidata *nd) 147052094c8aSAl Viro { 147152094c8aSAl Viro if (nd->flags & LOOKUP_RCU) { 14724ad5abb3SAl Viro int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); 147352094c8aSAl Viro if (err != -ECHILD) 147452094c8aSAl Viro return err; 147519660af7SAl Viro if (unlazy_walk(nd, NULL)) 147652094c8aSAl Viro return -ECHILD; 147752094c8aSAl Viro } 14784ad5abb3SAl Viro return inode_permission(nd->inode, MAY_EXEC); 147952094c8aSAl Viro } 148052094c8aSAl Viro 14819856fa1bSAl Viro static inline int handle_dots(struct nameidata *nd, int type) 14829856fa1bSAl Viro { 14839856fa1bSAl Viro if (type == LAST_DOTDOT) { 14849856fa1bSAl Viro if (nd->flags & LOOKUP_RCU) { 14859856fa1bSAl Viro if (follow_dotdot_rcu(nd)) 14869856fa1bSAl Viro return -ECHILD; 14879856fa1bSAl Viro } else 14889856fa1bSAl Viro follow_dotdot(nd); 14899856fa1bSAl Viro } 14909856fa1bSAl Viro return 0; 14919856fa1bSAl Viro } 14929856fa1bSAl Viro 1493951361f9SAl Viro static void terminate_walk(struct nameidata *nd) 1494951361f9SAl Viro { 1495951361f9SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1496951361f9SAl Viro path_put(&nd->path); 1497951361f9SAl Viro } else { 1498951361f9SAl Viro nd->flags &= ~LOOKUP_RCU; 14995b6ca027SAl Viro if (!(nd->flags & LOOKUP_ROOT)) 1500951361f9SAl Viro nd->root.mnt = NULL; 15018b61e74fSAl Viro rcu_read_unlock(); 1502951361f9SAl Viro } 1503951361f9SAl Viro } 1504951361f9SAl Viro 15053ddcd056SLinus Torvalds /* 15063ddcd056SLinus Torvalds * Do we need to follow links? We _really_ want to be able 15073ddcd056SLinus Torvalds * to do this check without having to look at inode->i_op, 15083ddcd056SLinus Torvalds * so we keep a cache of "no, this doesn't need follow_link" 15093ddcd056SLinus Torvalds * for the common case. 15103ddcd056SLinus Torvalds */ 1511b18825a7SDavid Howells static inline int should_follow_link(struct dentry *dentry, int follow) 15123ddcd056SLinus Torvalds { 1513b18825a7SDavid Howells return unlikely(d_is_symlink(dentry)) ? follow : 0; 15143ddcd056SLinus Torvalds } 15153ddcd056SLinus Torvalds 1516ce57dfc1SAl Viro static inline int walk_component(struct nameidata *nd, struct path *path, 151721b9b073SAl Viro int follow) 1518ce57dfc1SAl Viro { 1519ce57dfc1SAl Viro struct inode *inode; 1520ce57dfc1SAl Viro int err; 1521ce57dfc1SAl Viro /* 1522ce57dfc1SAl Viro * "." and ".." are special - ".." especially so because it has 1523ce57dfc1SAl Viro * to be able to know about the current root directory and 1524ce57dfc1SAl Viro * parent relationships. 1525ce57dfc1SAl Viro */ 152621b9b073SAl Viro if (unlikely(nd->last_type != LAST_NORM)) 152721b9b073SAl Viro return handle_dots(nd, nd->last_type); 1528e97cdc87SAl Viro err = lookup_fast(nd, path, &inode); 1529ce57dfc1SAl Viro if (unlikely(err)) { 1530697f514dSMiklos Szeredi if (err < 0) 1531697f514dSMiklos Szeredi goto out_err; 1532697f514dSMiklos Szeredi 1533cc2a5271SAl Viro err = lookup_slow(nd, path); 1534697f514dSMiklos Szeredi if (err < 0) 1535697f514dSMiklos Szeredi goto out_err; 1536697f514dSMiklos Szeredi 1537697f514dSMiklos Szeredi inode = path->dentry->d_inode; 1538ce57dfc1SAl Viro } 1539697f514dSMiklos Szeredi err = -ENOENT; 1540697f514dSMiklos Szeredi if (!inode) 1541697f514dSMiklos Szeredi goto out_path_put; 1542697f514dSMiklos Szeredi 1543b18825a7SDavid Howells if (should_follow_link(path->dentry, follow)) { 154419660af7SAl Viro if (nd->flags & LOOKUP_RCU) { 154519660af7SAl Viro if (unlikely(unlazy_walk(nd, path->dentry))) { 1546697f514dSMiklos Szeredi err = -ECHILD; 1547697f514dSMiklos Szeredi goto out_err; 154819660af7SAl Viro } 154919660af7SAl Viro } 1550ce57dfc1SAl Viro BUG_ON(inode != path->dentry->d_inode); 1551ce57dfc1SAl Viro return 1; 1552ce57dfc1SAl Viro } 1553ce57dfc1SAl Viro path_to_nameidata(path, nd); 1554ce57dfc1SAl Viro nd->inode = inode; 1555ce57dfc1SAl Viro return 0; 1556697f514dSMiklos Szeredi 1557697f514dSMiklos Szeredi out_path_put: 1558697f514dSMiklos Szeredi path_to_nameidata(path, nd); 1559697f514dSMiklos Szeredi out_err: 1560697f514dSMiklos Szeredi terminate_walk(nd); 1561697f514dSMiklos Szeredi return err; 1562ce57dfc1SAl Viro } 1563ce57dfc1SAl Viro 15641da177e4SLinus Torvalds /* 1565b356379aSAl Viro * This limits recursive symlink follows to 8, while 1566b356379aSAl Viro * limiting consecutive symlinks to 40. 1567b356379aSAl Viro * 1568b356379aSAl Viro * Without that kind of total limit, nasty chains of consecutive 1569b356379aSAl Viro * symlinks can cause almost arbitrarily long lookups. 1570b356379aSAl Viro */ 1571b356379aSAl Viro static inline int nested_symlink(struct path *path, struct nameidata *nd) 1572b356379aSAl Viro { 1573b356379aSAl Viro int res; 1574b356379aSAl Viro 1575b356379aSAl Viro if (unlikely(current->link_count >= MAX_NESTED_LINKS)) { 1576b356379aSAl Viro path_put_conditional(path, nd); 1577b356379aSAl Viro path_put(&nd->path); 1578b356379aSAl Viro return -ELOOP; 1579b356379aSAl Viro } 15801a4022f8SErez Zadok BUG_ON(nd->depth >= MAX_NESTED_LINKS); 1581b356379aSAl Viro 1582b356379aSAl Viro nd->depth++; 1583b356379aSAl Viro current->link_count++; 1584b356379aSAl Viro 1585b356379aSAl Viro do { 1586b356379aSAl Viro struct path link = *path; 1587b356379aSAl Viro void *cookie; 1588574197e0SAl Viro 1589574197e0SAl Viro res = follow_link(&link, nd, &cookie); 15906d7b5aaeSAl Viro if (res) 15916d7b5aaeSAl Viro break; 159221b9b073SAl Viro res = walk_component(nd, path, LOOKUP_FOLLOW); 1593574197e0SAl Viro put_link(nd, &link, cookie); 1594b356379aSAl Viro } while (res > 0); 1595b356379aSAl Viro 1596b356379aSAl Viro current->link_count--; 1597b356379aSAl Viro nd->depth--; 1598b356379aSAl Viro return res; 1599b356379aSAl Viro } 1600b356379aSAl Viro 1601b356379aSAl Viro /* 1602bfcfaa77SLinus Torvalds * We can do the critical dentry name comparison and hashing 1603bfcfaa77SLinus Torvalds * operations one word at a time, but we are limited to: 1604bfcfaa77SLinus Torvalds * 1605bfcfaa77SLinus Torvalds * - Architectures with fast unaligned word accesses. We could 1606bfcfaa77SLinus Torvalds * do a "get_unaligned()" if this helps and is sufficiently 1607bfcfaa77SLinus Torvalds * fast. 1608bfcfaa77SLinus Torvalds * 1609bfcfaa77SLinus Torvalds * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we 1610bfcfaa77SLinus Torvalds * do not trap on the (extremely unlikely) case of a page 1611bfcfaa77SLinus Torvalds * crossing operation. 1612bfcfaa77SLinus Torvalds * 1613bfcfaa77SLinus Torvalds * - Furthermore, we need an efficient 64-bit compile for the 1614bfcfaa77SLinus Torvalds * 64-bit case in order to generate the "number of bytes in 1615bfcfaa77SLinus Torvalds * the final mask". Again, that could be replaced with a 1616bfcfaa77SLinus Torvalds * efficient population count instruction or similar. 1617bfcfaa77SLinus Torvalds */ 1618bfcfaa77SLinus Torvalds #ifdef CONFIG_DCACHE_WORD_ACCESS 1619bfcfaa77SLinus Torvalds 1620f68e556eSLinus Torvalds #include <asm/word-at-a-time.h> 1621bfcfaa77SLinus Torvalds 1622f68e556eSLinus Torvalds #ifdef CONFIG_64BIT 1623bfcfaa77SLinus Torvalds 1624bfcfaa77SLinus Torvalds static inline unsigned int fold_hash(unsigned long hash) 1625bfcfaa77SLinus Torvalds { 1626bfcfaa77SLinus Torvalds hash += hash >> (8*sizeof(int)); 1627bfcfaa77SLinus Torvalds return hash; 1628bfcfaa77SLinus Torvalds } 1629bfcfaa77SLinus Torvalds 1630bfcfaa77SLinus Torvalds #else /* 32-bit case */ 1631bfcfaa77SLinus Torvalds 1632bfcfaa77SLinus Torvalds #define fold_hash(x) (x) 1633bfcfaa77SLinus Torvalds 1634bfcfaa77SLinus Torvalds #endif 1635bfcfaa77SLinus Torvalds 1636bfcfaa77SLinus Torvalds unsigned int full_name_hash(const unsigned char *name, unsigned int len) 1637bfcfaa77SLinus Torvalds { 1638bfcfaa77SLinus Torvalds unsigned long a, mask; 1639bfcfaa77SLinus Torvalds unsigned long hash = 0; 1640bfcfaa77SLinus Torvalds 1641bfcfaa77SLinus Torvalds for (;;) { 1642e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name); 1643bfcfaa77SLinus Torvalds if (len < sizeof(unsigned long)) 1644bfcfaa77SLinus Torvalds break; 1645bfcfaa77SLinus Torvalds hash += a; 1646f132c5beSAl Viro hash *= 9; 1647bfcfaa77SLinus Torvalds name += sizeof(unsigned long); 1648bfcfaa77SLinus Torvalds len -= sizeof(unsigned long); 1649bfcfaa77SLinus Torvalds if (!len) 1650bfcfaa77SLinus Torvalds goto done; 1651bfcfaa77SLinus Torvalds } 1652a5c21dceSWill Deacon mask = bytemask_from_count(len); 1653bfcfaa77SLinus Torvalds hash += mask & a; 1654bfcfaa77SLinus Torvalds done: 1655bfcfaa77SLinus Torvalds return fold_hash(hash); 1656bfcfaa77SLinus Torvalds } 1657bfcfaa77SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 1658bfcfaa77SLinus Torvalds 1659bfcfaa77SLinus Torvalds /* 1660bfcfaa77SLinus Torvalds * Calculate the length and hash of the path component, and 1661bfcfaa77SLinus Torvalds * return the length of the component; 1662bfcfaa77SLinus Torvalds */ 1663bfcfaa77SLinus Torvalds static inline unsigned long hash_name(const char *name, unsigned int *hashp) 1664bfcfaa77SLinus Torvalds { 166536126f8fSLinus Torvalds unsigned long a, b, adata, bdata, mask, hash, len; 166636126f8fSLinus Torvalds const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 1667bfcfaa77SLinus Torvalds 1668bfcfaa77SLinus Torvalds hash = a = 0; 1669bfcfaa77SLinus Torvalds len = -sizeof(unsigned long); 1670bfcfaa77SLinus Torvalds do { 1671bfcfaa77SLinus Torvalds hash = (hash + a) * 9; 1672bfcfaa77SLinus Torvalds len += sizeof(unsigned long); 1673e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name+len); 167436126f8fSLinus Torvalds b = a ^ REPEAT_BYTE('/'); 167536126f8fSLinus Torvalds } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); 1676bfcfaa77SLinus Torvalds 167736126f8fSLinus Torvalds adata = prep_zero_mask(a, adata, &constants); 167836126f8fSLinus Torvalds bdata = prep_zero_mask(b, bdata, &constants); 167936126f8fSLinus Torvalds 168036126f8fSLinus Torvalds mask = create_zero_mask(adata | bdata); 168136126f8fSLinus Torvalds 168236126f8fSLinus Torvalds hash += a & zero_bytemask(mask); 1683bfcfaa77SLinus Torvalds *hashp = fold_hash(hash); 1684bfcfaa77SLinus Torvalds 168536126f8fSLinus Torvalds return len + find_zero(mask); 1686bfcfaa77SLinus Torvalds } 1687bfcfaa77SLinus Torvalds 1688bfcfaa77SLinus Torvalds #else 1689bfcfaa77SLinus Torvalds 16900145acc2SLinus Torvalds unsigned int full_name_hash(const unsigned char *name, unsigned int len) 16910145acc2SLinus Torvalds { 16920145acc2SLinus Torvalds unsigned long hash = init_name_hash(); 16930145acc2SLinus Torvalds while (len--) 16940145acc2SLinus Torvalds hash = partial_name_hash(*name++, hash); 16950145acc2SLinus Torvalds return end_name_hash(hash); 16960145acc2SLinus Torvalds } 1697ae942ae7SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 16980145acc2SLinus Torvalds 16993ddcd056SLinus Torvalds /* 1700200e9ef7SLinus Torvalds * We know there's a real path component here of at least 1701200e9ef7SLinus Torvalds * one character. 1702200e9ef7SLinus Torvalds */ 1703200e9ef7SLinus Torvalds static inline unsigned long hash_name(const char *name, unsigned int *hashp) 1704200e9ef7SLinus Torvalds { 1705200e9ef7SLinus Torvalds unsigned long hash = init_name_hash(); 1706200e9ef7SLinus Torvalds unsigned long len = 0, c; 1707200e9ef7SLinus Torvalds 1708200e9ef7SLinus Torvalds c = (unsigned char)*name; 1709200e9ef7SLinus Torvalds do { 1710200e9ef7SLinus Torvalds len++; 1711200e9ef7SLinus Torvalds hash = partial_name_hash(c, hash); 1712200e9ef7SLinus Torvalds c = (unsigned char)name[len]; 1713200e9ef7SLinus Torvalds } while (c && c != '/'); 1714200e9ef7SLinus Torvalds *hashp = end_name_hash(hash); 1715200e9ef7SLinus Torvalds return len; 1716200e9ef7SLinus Torvalds } 1717200e9ef7SLinus Torvalds 1718bfcfaa77SLinus Torvalds #endif 1719bfcfaa77SLinus Torvalds 1720200e9ef7SLinus Torvalds /* 17211da177e4SLinus Torvalds * Name resolution. 1722ea3834d9SPrasanna Meda * This is the basic name resolution function, turning a pathname into 1723ea3834d9SPrasanna Meda * the final dentry. We expect 'base' to be positive and a directory. 17241da177e4SLinus Torvalds * 1725ea3834d9SPrasanna Meda * Returns 0 and nd will have valid dentry and mnt on success. 1726ea3834d9SPrasanna Meda * Returns error and drops reference to input namei data on failure. 17271da177e4SLinus Torvalds */ 17286de88d72SAl Viro static int link_path_walk(const char *name, struct nameidata *nd) 17291da177e4SLinus Torvalds { 17301da177e4SLinus Torvalds struct path next; 17311da177e4SLinus Torvalds int err; 17321da177e4SLinus Torvalds 17331da177e4SLinus Torvalds while (*name=='/') 17341da177e4SLinus Torvalds name++; 17351da177e4SLinus Torvalds if (!*name) 1736086e183aSAl Viro return 0; 17371da177e4SLinus Torvalds 17381da177e4SLinus Torvalds /* At this point we know we have a real path component. */ 17391da177e4SLinus Torvalds for(;;) { 17401da177e4SLinus Torvalds struct qstr this; 1741200e9ef7SLinus Torvalds long len; 1742fe479a58SAl Viro int type; 17431da177e4SLinus Torvalds 174452094c8aSAl Viro err = may_lookup(nd); 17451da177e4SLinus Torvalds if (err) 17461da177e4SLinus Torvalds break; 17471da177e4SLinus Torvalds 1748200e9ef7SLinus Torvalds len = hash_name(name, &this.hash); 17491da177e4SLinus Torvalds this.name = name; 1750200e9ef7SLinus Torvalds this.len = len; 17511da177e4SLinus Torvalds 1752fe479a58SAl Viro type = LAST_NORM; 1753200e9ef7SLinus Torvalds if (name[0] == '.') switch (len) { 1754fe479a58SAl Viro case 2: 1755200e9ef7SLinus Torvalds if (name[1] == '.') { 1756fe479a58SAl Viro type = LAST_DOTDOT; 175716c2cd71SAl Viro nd->flags |= LOOKUP_JUMPED; 175816c2cd71SAl Viro } 1759fe479a58SAl Viro break; 1760fe479a58SAl Viro case 1: 1761fe479a58SAl Viro type = LAST_DOT; 1762fe479a58SAl Viro } 17635a202bcdSAl Viro if (likely(type == LAST_NORM)) { 17645a202bcdSAl Viro struct dentry *parent = nd->path.dentry; 176516c2cd71SAl Viro nd->flags &= ~LOOKUP_JUMPED; 17665a202bcdSAl Viro if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { 1767da53be12SLinus Torvalds err = parent->d_op->d_hash(parent, &this); 17685a202bcdSAl Viro if (err < 0) 17695a202bcdSAl Viro break; 17705a202bcdSAl Viro } 17715a202bcdSAl Viro } 1772fe479a58SAl Viro 17735f4a6a69SAl Viro nd->last = this; 17745f4a6a69SAl Viro nd->last_type = type; 17755f4a6a69SAl Viro 1776200e9ef7SLinus Torvalds if (!name[len]) 17775f4a6a69SAl Viro return 0; 1778200e9ef7SLinus Torvalds /* 1779200e9ef7SLinus Torvalds * If it wasn't NUL, we know it was '/'. Skip that 1780200e9ef7SLinus Torvalds * slash, and continue until no more slashes. 1781200e9ef7SLinus Torvalds */ 1782200e9ef7SLinus Torvalds do { 1783200e9ef7SLinus Torvalds len++; 1784200e9ef7SLinus Torvalds } while (unlikely(name[len] == '/')); 1785200e9ef7SLinus Torvalds if (!name[len]) 17865f4a6a69SAl Viro return 0; 17875f4a6a69SAl Viro 1788200e9ef7SLinus Torvalds name += len; 17891da177e4SLinus Torvalds 179021b9b073SAl Viro err = walk_component(nd, &next, LOOKUP_FOLLOW); 1791ce57dfc1SAl Viro if (err < 0) 1792ce57dfc1SAl Viro return err; 1793fe479a58SAl Viro 1794ce57dfc1SAl Viro if (err) { 1795b356379aSAl Viro err = nested_symlink(&next, nd); 17961da177e4SLinus Torvalds if (err) 1797a7472babSAl Viro return err; 179831e6b01fSNick Piggin } 179944b1d530SMiklos Szeredi if (!d_can_lookup(nd->path.dentry)) { 18003ddcd056SLinus Torvalds err = -ENOTDIR; 18013ddcd056SLinus Torvalds break; 18025f4a6a69SAl Viro } 1803ce57dfc1SAl Viro } 1804951361f9SAl Viro terminate_walk(nd); 18051da177e4SLinus Torvalds return err; 18061da177e4SLinus Torvalds } 18071da177e4SLinus Torvalds 180870e9b357SAl Viro static int path_init(int dfd, const char *name, unsigned int flags, 180970e9b357SAl Viro struct nameidata *nd, struct file **fp) 181031e6b01fSNick Piggin { 181131e6b01fSNick Piggin int retval = 0; 181231e6b01fSNick Piggin 181331e6b01fSNick Piggin nd->last_type = LAST_ROOT; /* if there are only slashes... */ 181416c2cd71SAl Viro nd->flags = flags | LOOKUP_JUMPED; 181531e6b01fSNick Piggin nd->depth = 0; 18165b6ca027SAl Viro if (flags & LOOKUP_ROOT) { 1817b18825a7SDavid Howells struct dentry *root = nd->root.dentry; 1818b18825a7SDavid Howells struct inode *inode = root->d_inode; 181973d049a4SAl Viro if (*name) { 182044b1d530SMiklos Szeredi if (!d_can_lookup(root)) 18215b6ca027SAl Viro return -ENOTDIR; 18225b6ca027SAl Viro retval = inode_permission(inode, MAY_EXEC); 18235b6ca027SAl Viro if (retval) 18245b6ca027SAl Viro return retval; 182573d049a4SAl Viro } 18265b6ca027SAl Viro nd->path = nd->root; 18275b6ca027SAl Viro nd->inode = inode; 18285b6ca027SAl Viro if (flags & LOOKUP_RCU) { 18298b61e74fSAl Viro rcu_read_lock(); 18305b6ca027SAl Viro nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 183148a066e7SAl Viro nd->m_seq = read_seqbegin(&mount_lock); 18325b6ca027SAl Viro } else { 18335b6ca027SAl Viro path_get(&nd->path); 18345b6ca027SAl Viro } 18355b6ca027SAl Viro return 0; 18365b6ca027SAl Viro } 18375b6ca027SAl Viro 183831e6b01fSNick Piggin nd->root.mnt = NULL; 183931e6b01fSNick Piggin 184048a066e7SAl Viro nd->m_seq = read_seqbegin(&mount_lock); 184131e6b01fSNick Piggin if (*name=='/') { 1842e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 18438b61e74fSAl Viro rcu_read_lock(); 1844e41f7d4eSAl Viro set_root_rcu(nd); 1845e41f7d4eSAl Viro } else { 1846e41f7d4eSAl Viro set_root(nd); 1847e41f7d4eSAl Viro path_get(&nd->root); 1848e41f7d4eSAl Viro } 184931e6b01fSNick Piggin nd->path = nd->root; 185031e6b01fSNick Piggin } else if (dfd == AT_FDCWD) { 1851e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 185231e6b01fSNick Piggin struct fs_struct *fs = current->fs; 1853c28cc364SNick Piggin unsigned seq; 185431e6b01fSNick Piggin 18558b61e74fSAl Viro rcu_read_lock(); 185631e6b01fSNick Piggin 1857c28cc364SNick Piggin do { 1858c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 185931e6b01fSNick Piggin nd->path = fs->pwd; 1860c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 1861c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 1862e41f7d4eSAl Viro } else { 1863e41f7d4eSAl Viro get_fs_pwd(current->fs, &nd->path); 1864e41f7d4eSAl Viro } 186531e6b01fSNick Piggin } else { 1866582aa64aSJeff Layton /* Caller must check execute permissions on the starting path component */ 18672903ff01SAl Viro struct fd f = fdget_raw(dfd); 186831e6b01fSNick Piggin struct dentry *dentry; 186931e6b01fSNick Piggin 18702903ff01SAl Viro if (!f.file) 18712903ff01SAl Viro return -EBADF; 187231e6b01fSNick Piggin 18732903ff01SAl Viro dentry = f.file->f_path.dentry; 187431e6b01fSNick Piggin 1875f52e0c11SAl Viro if (*name) { 187644b1d530SMiklos Szeredi if (!d_can_lookup(dentry)) { 18772903ff01SAl Viro fdput(f); 18782903ff01SAl Viro return -ENOTDIR; 1879f52e0c11SAl Viro } 18802903ff01SAl Viro } 18812903ff01SAl Viro 18822903ff01SAl Viro nd->path = f.file->f_path; 1883e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 18849c225f26SLinus Torvalds if (f.flags & FDPUT_FPUT) 18852903ff01SAl Viro *fp = f.file; 1886c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 18878b61e74fSAl Viro rcu_read_lock(); 18885590ff0dSUlrich Drepper } else { 18892903ff01SAl Viro path_get(&nd->path); 18902903ff01SAl Viro fdput(f); 18911da177e4SLinus Torvalds } 1892e41f7d4eSAl Viro } 1893e41f7d4eSAl Viro 189431e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 18959b4a9b14SAl Viro return 0; 18969b4a9b14SAl Viro } 18979b4a9b14SAl Viro 1898bd92d7feSAl Viro static inline int lookup_last(struct nameidata *nd, struct path *path) 1899bd92d7feSAl Viro { 1900bd92d7feSAl Viro if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) 1901bd92d7feSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 1902bd92d7feSAl Viro 1903bd92d7feSAl Viro nd->flags &= ~LOOKUP_PARENT; 190421b9b073SAl Viro return walk_component(nd, path, nd->flags & LOOKUP_FOLLOW); 1905bd92d7feSAl Viro } 1906bd92d7feSAl Viro 19079b4a9b14SAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 1908ee0827cdSAl Viro static int path_lookupat(int dfd, const char *name, 19099b4a9b14SAl Viro unsigned int flags, struct nameidata *nd) 19109b4a9b14SAl Viro { 191170e9b357SAl Viro struct file *base = NULL; 1912bd92d7feSAl Viro struct path path; 1913bd92d7feSAl Viro int err; 191431e6b01fSNick Piggin 191531e6b01fSNick Piggin /* 191631e6b01fSNick Piggin * Path walking is largely split up into 2 different synchronisation 191731e6b01fSNick Piggin * schemes, rcu-walk and ref-walk (explained in 191831e6b01fSNick Piggin * Documentation/filesystems/path-lookup.txt). These share much of the 191931e6b01fSNick Piggin * path walk code, but some things particularly setup, cleanup, and 192031e6b01fSNick Piggin * following mounts are sufficiently divergent that functions are 192131e6b01fSNick Piggin * duplicated. Typically there is a function foo(), and its RCU 192231e6b01fSNick Piggin * analogue, foo_rcu(). 192331e6b01fSNick Piggin * 192431e6b01fSNick Piggin * -ECHILD is the error number of choice (just to avoid clashes) that 192531e6b01fSNick Piggin * is returned if some aspect of an rcu-walk fails. Such an error must 192631e6b01fSNick Piggin * be handled by restarting a traditional ref-walk (which will always 192731e6b01fSNick Piggin * be able to complete). 192831e6b01fSNick Piggin */ 1929bd92d7feSAl Viro err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base); 1930ee0827cdSAl Viro 1931bd92d7feSAl Viro if (unlikely(err)) 1932bd92d7feSAl Viro return err; 1933ee0827cdSAl Viro 1934ee0827cdSAl Viro current->total_link_count = 0; 1935bd92d7feSAl Viro err = link_path_walk(name, nd); 1936bd92d7feSAl Viro 1937bd92d7feSAl Viro if (!err && !(flags & LOOKUP_PARENT)) { 1938bd92d7feSAl Viro err = lookup_last(nd, &path); 1939bd92d7feSAl Viro while (err > 0) { 1940bd92d7feSAl Viro void *cookie; 1941bd92d7feSAl Viro struct path link = path; 1942800179c9SKees Cook err = may_follow_link(&link, nd); 1943800179c9SKees Cook if (unlikely(err)) 1944800179c9SKees Cook break; 1945bd92d7feSAl Viro nd->flags |= LOOKUP_PARENT; 1946574197e0SAl Viro err = follow_link(&link, nd, &cookie); 19476d7b5aaeSAl Viro if (err) 19486d7b5aaeSAl Viro break; 1949bd92d7feSAl Viro err = lookup_last(nd, &path); 1950574197e0SAl Viro put_link(nd, &link, cookie); 1951bd92d7feSAl Viro } 1952bd92d7feSAl Viro } 1953ee0827cdSAl Viro 19549f1fafeeSAl Viro if (!err) 19559f1fafeeSAl Viro err = complete_walk(nd); 1956bd92d7feSAl Viro 1957bd92d7feSAl Viro if (!err && nd->flags & LOOKUP_DIRECTORY) { 195844b1d530SMiklos Szeredi if (!d_can_lookup(nd->path.dentry)) { 1959bd92d7feSAl Viro path_put(&nd->path); 1960bd23a539SAl Viro err = -ENOTDIR; 1961bd92d7feSAl Viro } 1962bd92d7feSAl Viro } 196316c2cd71SAl Viro 196470e9b357SAl Viro if (base) 196570e9b357SAl Viro fput(base); 1966ee0827cdSAl Viro 19675b6ca027SAl Viro if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 196831e6b01fSNick Piggin path_put(&nd->root); 196931e6b01fSNick Piggin nd->root.mnt = NULL; 197031e6b01fSNick Piggin } 1971bd92d7feSAl Viro return err; 197231e6b01fSNick Piggin } 197331e6b01fSNick Piggin 1974873f1eedSJeff Layton static int filename_lookup(int dfd, struct filename *name, 1975873f1eedSJeff Layton unsigned int flags, struct nameidata *nd) 1976873f1eedSJeff Layton { 1977873f1eedSJeff Layton int retval = path_lookupat(dfd, name->name, flags | LOOKUP_RCU, nd); 1978873f1eedSJeff Layton if (unlikely(retval == -ECHILD)) 1979873f1eedSJeff Layton retval = path_lookupat(dfd, name->name, flags, nd); 1980873f1eedSJeff Layton if (unlikely(retval == -ESTALE)) 1981873f1eedSJeff Layton retval = path_lookupat(dfd, name->name, 1982873f1eedSJeff Layton flags | LOOKUP_REVAL, nd); 1983873f1eedSJeff Layton 1984873f1eedSJeff Layton if (likely(!retval)) 1985adb5c247SJeff Layton audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT); 1986873f1eedSJeff Layton return retval; 1987873f1eedSJeff Layton } 1988873f1eedSJeff Layton 1989ee0827cdSAl Viro static int do_path_lookup(int dfd, const char *name, 1990ee0827cdSAl Viro unsigned int flags, struct nameidata *nd) 1991ee0827cdSAl Viro { 1992873f1eedSJeff Layton struct filename filename = { .name = name }; 1993ee0827cdSAl Viro 1994873f1eedSJeff Layton return filename_lookup(dfd, &filename, flags, nd); 19951da177e4SLinus Torvalds } 19961da177e4SLinus Torvalds 199779714f72SAl Viro /* does lookup, returns the object with parent locked */ 199879714f72SAl Viro struct dentry *kern_path_locked(const char *name, struct path *path) 19995590ff0dSUlrich Drepper { 200079714f72SAl Viro struct nameidata nd; 200179714f72SAl Viro struct dentry *d; 200279714f72SAl Viro int err = do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, &nd); 200379714f72SAl Viro if (err) 200479714f72SAl Viro return ERR_PTR(err); 200579714f72SAl Viro if (nd.last_type != LAST_NORM) { 200679714f72SAl Viro path_put(&nd.path); 200779714f72SAl Viro return ERR_PTR(-EINVAL); 200879714f72SAl Viro } 200979714f72SAl Viro mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 20101e0ea001SAl Viro d = __lookup_hash(&nd.last, nd.path.dentry, 0); 201179714f72SAl Viro if (IS_ERR(d)) { 201279714f72SAl Viro mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 201379714f72SAl Viro path_put(&nd.path); 201479714f72SAl Viro return d; 201579714f72SAl Viro } 201679714f72SAl Viro *path = nd.path; 201779714f72SAl Viro return d; 20185590ff0dSUlrich Drepper } 20195590ff0dSUlrich Drepper 2020d1811465SAl Viro int kern_path(const char *name, unsigned int flags, struct path *path) 2021d1811465SAl Viro { 2022d1811465SAl Viro struct nameidata nd; 2023d1811465SAl Viro int res = do_path_lookup(AT_FDCWD, name, flags, &nd); 2024d1811465SAl Viro if (!res) 2025d1811465SAl Viro *path = nd.path; 2026d1811465SAl Viro return res; 2027d1811465SAl Viro } 2028d1811465SAl Viro 202916f18200SJosef 'Jeff' Sipek /** 203016f18200SJosef 'Jeff' Sipek * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair 203116f18200SJosef 'Jeff' Sipek * @dentry: pointer to dentry of the base directory 203216f18200SJosef 'Jeff' Sipek * @mnt: pointer to vfs mount of the base directory 203316f18200SJosef 'Jeff' Sipek * @name: pointer to file name 203416f18200SJosef 'Jeff' Sipek * @flags: lookup flags 2035e0a01249SAl Viro * @path: pointer to struct path to fill 203616f18200SJosef 'Jeff' Sipek */ 203716f18200SJosef 'Jeff' Sipek int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, 203816f18200SJosef 'Jeff' Sipek const char *name, unsigned int flags, 2039e0a01249SAl Viro struct path *path) 204016f18200SJosef 'Jeff' Sipek { 2041e0a01249SAl Viro struct nameidata nd; 2042e0a01249SAl Viro int err; 2043e0a01249SAl Viro nd.root.dentry = dentry; 2044e0a01249SAl Viro nd.root.mnt = mnt; 2045e0a01249SAl Viro BUG_ON(flags & LOOKUP_PARENT); 20465b6ca027SAl Viro /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */ 2047e0a01249SAl Viro err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd); 2048e0a01249SAl Viro if (!err) 2049e0a01249SAl Viro *path = nd.path; 2050e0a01249SAl Viro return err; 205116f18200SJosef 'Jeff' Sipek } 205216f18200SJosef 'Jeff' Sipek 2053057f6c01SJames Morris /* 2054057f6c01SJames Morris * Restricted form of lookup. Doesn't follow links, single-component only, 2055057f6c01SJames Morris * needs parent already locked. Doesn't follow mounts. 2056057f6c01SJames Morris * SMP-safe. 2057057f6c01SJames Morris */ 2058a244e169SAdrian Bunk static struct dentry *lookup_hash(struct nameidata *nd) 20591da177e4SLinus Torvalds { 206072bd866aSAl Viro return __lookup_hash(&nd->last, nd->path.dentry, nd->flags); 20611da177e4SLinus Torvalds } 20621da177e4SLinus Torvalds 2063eead1911SChristoph Hellwig /** 2064a6b91919SRandy Dunlap * lookup_one_len - filesystem helper to lookup single pathname component 2065eead1911SChristoph Hellwig * @name: pathname component to lookup 2066eead1911SChristoph Hellwig * @base: base directory to lookup from 2067eead1911SChristoph Hellwig * @len: maximum length @len should be interpreted to 2068eead1911SChristoph Hellwig * 2069a6b91919SRandy Dunlap * Note that this routine is purely a helper for filesystem usage and should 2070a6b91919SRandy Dunlap * not be called by generic code. Also note that by using this function the 2071eead1911SChristoph Hellwig * nameidata argument is passed to the filesystem methods and a filesystem 2072eead1911SChristoph Hellwig * using this helper needs to be prepared for that. 2073eead1911SChristoph Hellwig */ 2074057f6c01SJames Morris struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 2075057f6c01SJames Morris { 2076057f6c01SJames Morris struct qstr this; 20776a96ba54SAl Viro unsigned int c; 2078cda309deSMiklos Szeredi int err; 2079057f6c01SJames Morris 20802f9092e1SDavid Woodhouse WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex)); 20812f9092e1SDavid Woodhouse 20826a96ba54SAl Viro this.name = name; 20836a96ba54SAl Viro this.len = len; 20840145acc2SLinus Torvalds this.hash = full_name_hash(name, len); 20856a96ba54SAl Viro if (!len) 20866a96ba54SAl Viro return ERR_PTR(-EACCES); 20876a96ba54SAl Viro 208821d8a15aSAl Viro if (unlikely(name[0] == '.')) { 208921d8a15aSAl Viro if (len < 2 || (len == 2 && name[1] == '.')) 209021d8a15aSAl Viro return ERR_PTR(-EACCES); 209121d8a15aSAl Viro } 209221d8a15aSAl Viro 20936a96ba54SAl Viro while (len--) { 20946a96ba54SAl Viro c = *(const unsigned char *)name++; 20956a96ba54SAl Viro if (c == '/' || c == '\0') 20966a96ba54SAl Viro return ERR_PTR(-EACCES); 20976a96ba54SAl Viro } 20985a202bcdSAl Viro /* 20995a202bcdSAl Viro * See if the low-level filesystem might want 21005a202bcdSAl Viro * to use its own hash.. 21015a202bcdSAl Viro */ 21025a202bcdSAl Viro if (base->d_flags & DCACHE_OP_HASH) { 2103da53be12SLinus Torvalds int err = base->d_op->d_hash(base, &this); 21045a202bcdSAl Viro if (err < 0) 21055a202bcdSAl Viro return ERR_PTR(err); 21065a202bcdSAl Viro } 2107eead1911SChristoph Hellwig 2108cda309deSMiklos Szeredi err = inode_permission(base->d_inode, MAY_EXEC); 2109cda309deSMiklos Szeredi if (err) 2110cda309deSMiklos Szeredi return ERR_PTR(err); 2111cda309deSMiklos Szeredi 211272bd866aSAl Viro return __lookup_hash(&this, base, 0); 2113057f6c01SJames Morris } 2114057f6c01SJames Morris 21151fa1e7f6SAndy Whitcroft int user_path_at_empty(int dfd, const char __user *name, unsigned flags, 21161fa1e7f6SAndy Whitcroft struct path *path, int *empty) 21171da177e4SLinus Torvalds { 21182d8f3038SAl Viro struct nameidata nd; 211991a27b2aSJeff Layton struct filename *tmp = getname_flags(name, flags, empty); 21201da177e4SLinus Torvalds int err = PTR_ERR(tmp); 21211da177e4SLinus Torvalds if (!IS_ERR(tmp)) { 21222d8f3038SAl Viro 21232d8f3038SAl Viro BUG_ON(flags & LOOKUP_PARENT); 21242d8f3038SAl Viro 2125873f1eedSJeff Layton err = filename_lookup(dfd, tmp, flags, &nd); 21261da177e4SLinus Torvalds putname(tmp); 21272d8f3038SAl Viro if (!err) 21282d8f3038SAl Viro *path = nd.path; 21291da177e4SLinus Torvalds } 21301da177e4SLinus Torvalds return err; 21311da177e4SLinus Torvalds } 21321da177e4SLinus Torvalds 21331fa1e7f6SAndy Whitcroft int user_path_at(int dfd, const char __user *name, unsigned flags, 21341fa1e7f6SAndy Whitcroft struct path *path) 21351fa1e7f6SAndy Whitcroft { 2136f7493e5dSLinus Torvalds return user_path_at_empty(dfd, name, flags, path, NULL); 21371fa1e7f6SAndy Whitcroft } 21381fa1e7f6SAndy Whitcroft 2139873f1eedSJeff Layton /* 2140873f1eedSJeff Layton * NB: most callers don't do anything directly with the reference to the 2141873f1eedSJeff Layton * to struct filename, but the nd->last pointer points into the name string 2142873f1eedSJeff Layton * allocated by getname. So we must hold the reference to it until all 2143873f1eedSJeff Layton * path-walking is complete. 2144873f1eedSJeff Layton */ 214591a27b2aSJeff Layton static struct filename * 21469e790bd6SJeff Layton user_path_parent(int dfd, const char __user *path, struct nameidata *nd, 21479e790bd6SJeff Layton unsigned int flags) 21482ad94ae6SAl Viro { 214991a27b2aSJeff Layton struct filename *s = getname(path); 21502ad94ae6SAl Viro int error; 21512ad94ae6SAl Viro 21529e790bd6SJeff Layton /* only LOOKUP_REVAL is allowed in extra flags */ 21539e790bd6SJeff Layton flags &= LOOKUP_REVAL; 21549e790bd6SJeff Layton 21552ad94ae6SAl Viro if (IS_ERR(s)) 215691a27b2aSJeff Layton return s; 21572ad94ae6SAl Viro 21589e790bd6SJeff Layton error = filename_lookup(dfd, s, flags | LOOKUP_PARENT, nd); 215991a27b2aSJeff Layton if (error) { 21602ad94ae6SAl Viro putname(s); 216191a27b2aSJeff Layton return ERR_PTR(error); 216291a27b2aSJeff Layton } 21632ad94ae6SAl Viro 216491a27b2aSJeff Layton return s; 21652ad94ae6SAl Viro } 21662ad94ae6SAl Viro 21678033426eSJeff Layton /** 2168197df04cSAl Viro * mountpoint_last - look up last component for umount 21698033426eSJeff Layton * @nd: pathwalk nameidata - currently pointing at parent directory of "last" 21708033426eSJeff Layton * @path: pointer to container for result 21718033426eSJeff Layton * 21728033426eSJeff Layton * This is a special lookup_last function just for umount. In this case, we 21738033426eSJeff Layton * need to resolve the path without doing any revalidation. 21748033426eSJeff Layton * 21758033426eSJeff Layton * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since 21768033426eSJeff Layton * mountpoints are always pinned in the dcache, their ancestors are too. Thus, 21778033426eSJeff Layton * in almost all cases, this lookup will be served out of the dcache. The only 21788033426eSJeff Layton * cases where it won't are if nd->last refers to a symlink or the path is 21798033426eSJeff Layton * bogus and it doesn't exist. 21808033426eSJeff Layton * 21818033426eSJeff Layton * Returns: 21828033426eSJeff Layton * -error: if there was an error during lookup. This includes -ENOENT if the 21838033426eSJeff Layton * lookup found a negative dentry. The nd->path reference will also be 21848033426eSJeff Layton * put in this case. 21858033426eSJeff Layton * 21868033426eSJeff Layton * 0: if we successfully resolved nd->path and found it to not to be a 21878033426eSJeff Layton * symlink that needs to be followed. "path" will also be populated. 21888033426eSJeff Layton * The nd->path reference will also be put. 21898033426eSJeff Layton * 21908033426eSJeff Layton * 1: if we successfully resolved nd->last and found it to be a symlink 21918033426eSJeff Layton * that needs to be followed. "path" will be populated with the path 21928033426eSJeff Layton * to the link, and nd->path will *not* be put. 21938033426eSJeff Layton */ 21948033426eSJeff Layton static int 2195197df04cSAl Viro mountpoint_last(struct nameidata *nd, struct path *path) 21968033426eSJeff Layton { 21978033426eSJeff Layton int error = 0; 21988033426eSJeff Layton struct dentry *dentry; 21998033426eSJeff Layton struct dentry *dir = nd->path.dentry; 22008033426eSJeff Layton 220135759521SAl Viro /* If we're in rcuwalk, drop out of it to handle last component */ 220235759521SAl Viro if (nd->flags & LOOKUP_RCU) { 220335759521SAl Viro if (unlazy_walk(nd, NULL)) { 22048033426eSJeff Layton error = -ECHILD; 220535759521SAl Viro goto out; 220635759521SAl Viro } 22078033426eSJeff Layton } 22088033426eSJeff Layton 22098033426eSJeff Layton nd->flags &= ~LOOKUP_PARENT; 22108033426eSJeff Layton 22118033426eSJeff Layton if (unlikely(nd->last_type != LAST_NORM)) { 22128033426eSJeff Layton error = handle_dots(nd, nd->last_type); 221335759521SAl Viro if (error) 221435759521SAl Viro goto out; 22158033426eSJeff Layton dentry = dget(nd->path.dentry); 221635759521SAl Viro goto done; 22178033426eSJeff Layton } 22188033426eSJeff Layton 22198033426eSJeff Layton mutex_lock(&dir->d_inode->i_mutex); 22208033426eSJeff Layton dentry = d_lookup(dir, &nd->last); 22218033426eSJeff Layton if (!dentry) { 22228033426eSJeff Layton /* 22238033426eSJeff Layton * No cached dentry. Mounted dentries are pinned in the cache, 22248033426eSJeff Layton * so that means that this dentry is probably a symlink or the 22258033426eSJeff Layton * path doesn't actually point to a mounted dentry. 22268033426eSJeff Layton */ 22278033426eSJeff Layton dentry = d_alloc(dir, &nd->last); 22288033426eSJeff Layton if (!dentry) { 22298033426eSJeff Layton error = -ENOMEM; 2230bcceeebaSDave Jones mutex_unlock(&dir->d_inode->i_mutex); 223135759521SAl Viro goto out; 22328033426eSJeff Layton } 223335759521SAl Viro dentry = lookup_real(dir->d_inode, dentry, nd->flags); 223435759521SAl Viro error = PTR_ERR(dentry); 2235bcceeebaSDave Jones if (IS_ERR(dentry)) { 2236bcceeebaSDave Jones mutex_unlock(&dir->d_inode->i_mutex); 223735759521SAl Viro goto out; 22388033426eSJeff Layton } 2239bcceeebaSDave Jones } 22408033426eSJeff Layton mutex_unlock(&dir->d_inode->i_mutex); 22418033426eSJeff Layton 224235759521SAl Viro done: 22438033426eSJeff Layton if (!dentry->d_inode) { 22448033426eSJeff Layton error = -ENOENT; 22458033426eSJeff Layton dput(dentry); 224635759521SAl Viro goto out; 224735759521SAl Viro } 22488033426eSJeff Layton path->dentry = dentry; 22498033426eSJeff Layton path->mnt = mntget(nd->path.mnt); 2250b18825a7SDavid Howells if (should_follow_link(dentry, nd->flags & LOOKUP_FOLLOW)) 22518033426eSJeff Layton return 1; 22528033426eSJeff Layton follow_mount(path); 225335759521SAl Viro error = 0; 225435759521SAl Viro out: 22558033426eSJeff Layton terminate_walk(nd); 22568033426eSJeff Layton return error; 22578033426eSJeff Layton } 22588033426eSJeff Layton 22598033426eSJeff Layton /** 2260197df04cSAl Viro * path_mountpoint - look up a path to be umounted 22618033426eSJeff Layton * @dfd: directory file descriptor to start walk from 22628033426eSJeff Layton * @name: full pathname to walk 2263606d6fe3SRandy Dunlap * @path: pointer to container for result 22648033426eSJeff Layton * @flags: lookup flags 22658033426eSJeff Layton * 22668033426eSJeff Layton * Look up the given name, but don't attempt to revalidate the last component. 2267606d6fe3SRandy Dunlap * Returns 0 and "path" will be valid on success; Returns error otherwise. 22688033426eSJeff Layton */ 22698033426eSJeff Layton static int 2270197df04cSAl Viro path_mountpoint(int dfd, const char *name, struct path *path, unsigned int flags) 22718033426eSJeff Layton { 22728033426eSJeff Layton struct file *base = NULL; 22738033426eSJeff Layton struct nameidata nd; 22748033426eSJeff Layton int err; 22758033426eSJeff Layton 22768033426eSJeff Layton err = path_init(dfd, name, flags | LOOKUP_PARENT, &nd, &base); 22778033426eSJeff Layton if (unlikely(err)) 22788033426eSJeff Layton return err; 22798033426eSJeff Layton 22808033426eSJeff Layton current->total_link_count = 0; 22818033426eSJeff Layton err = link_path_walk(name, &nd); 22828033426eSJeff Layton if (err) 22838033426eSJeff Layton goto out; 22848033426eSJeff Layton 2285197df04cSAl Viro err = mountpoint_last(&nd, path); 22868033426eSJeff Layton while (err > 0) { 22878033426eSJeff Layton void *cookie; 22888033426eSJeff Layton struct path link = *path; 22898033426eSJeff Layton err = may_follow_link(&link, &nd); 22908033426eSJeff Layton if (unlikely(err)) 22918033426eSJeff Layton break; 22928033426eSJeff Layton nd.flags |= LOOKUP_PARENT; 22938033426eSJeff Layton err = follow_link(&link, &nd, &cookie); 22948033426eSJeff Layton if (err) 22958033426eSJeff Layton break; 2296197df04cSAl Viro err = mountpoint_last(&nd, path); 22978033426eSJeff Layton put_link(&nd, &link, cookie); 22988033426eSJeff Layton } 22998033426eSJeff Layton out: 23008033426eSJeff Layton if (base) 23018033426eSJeff Layton fput(base); 23028033426eSJeff Layton 23038033426eSJeff Layton if (nd.root.mnt && !(nd.flags & LOOKUP_ROOT)) 23048033426eSJeff Layton path_put(&nd.root); 23058033426eSJeff Layton 23068033426eSJeff Layton return err; 23078033426eSJeff Layton } 23088033426eSJeff Layton 23092d864651SAl Viro static int 23102d864651SAl Viro filename_mountpoint(int dfd, struct filename *s, struct path *path, 23112d864651SAl Viro unsigned int flags) 23122d864651SAl Viro { 23132d864651SAl Viro int error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_RCU); 23142d864651SAl Viro if (unlikely(error == -ECHILD)) 23152d864651SAl Viro error = path_mountpoint(dfd, s->name, path, flags); 23162d864651SAl Viro if (unlikely(error == -ESTALE)) 23172d864651SAl Viro error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_REVAL); 23182d864651SAl Viro if (likely(!error)) 23192d864651SAl Viro audit_inode(s, path->dentry, 0); 23202d864651SAl Viro return error; 23212d864651SAl Viro } 23222d864651SAl Viro 23238033426eSJeff Layton /** 2324197df04cSAl Viro * user_path_mountpoint_at - lookup a path from userland in order to umount it 23258033426eSJeff Layton * @dfd: directory file descriptor 23268033426eSJeff Layton * @name: pathname from userland 23278033426eSJeff Layton * @flags: lookup flags 23288033426eSJeff Layton * @path: pointer to container to hold result 23298033426eSJeff Layton * 23308033426eSJeff Layton * A umount is a special case for path walking. We're not actually interested 23318033426eSJeff Layton * in the inode in this situation, and ESTALE errors can be a problem. We 23328033426eSJeff Layton * simply want track down the dentry and vfsmount attached at the mountpoint 23338033426eSJeff Layton * and avoid revalidating the last component. 23348033426eSJeff Layton * 23358033426eSJeff Layton * Returns 0 and populates "path" on success. 23368033426eSJeff Layton */ 23378033426eSJeff Layton int 2338197df04cSAl Viro user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags, 23398033426eSJeff Layton struct path *path) 23408033426eSJeff Layton { 23418033426eSJeff Layton struct filename *s = getname(name); 23428033426eSJeff Layton int error; 23438033426eSJeff Layton if (IS_ERR(s)) 23448033426eSJeff Layton return PTR_ERR(s); 23452d864651SAl Viro error = filename_mountpoint(dfd, s, path, flags); 23468033426eSJeff Layton putname(s); 23478033426eSJeff Layton return error; 23488033426eSJeff Layton } 23498033426eSJeff Layton 23502d864651SAl Viro int 23512d864651SAl Viro kern_path_mountpoint(int dfd, const char *name, struct path *path, 23522d864651SAl Viro unsigned int flags) 23532d864651SAl Viro { 23542d864651SAl Viro struct filename s = {.name = name}; 23552d864651SAl Viro return filename_mountpoint(dfd, &s, path, flags); 23562d864651SAl Viro } 23572d864651SAl Viro EXPORT_SYMBOL(kern_path_mountpoint); 23582d864651SAl Viro 23591da177e4SLinus Torvalds /* 23601da177e4SLinus Torvalds * It's inline, so penalty for filesystems that don't use sticky bit is 23611da177e4SLinus Torvalds * minimal. 23621da177e4SLinus Torvalds */ 23631da177e4SLinus Torvalds static inline int check_sticky(struct inode *dir, struct inode *inode) 23641da177e4SLinus Torvalds { 23658e96e3b7SEric W. Biederman kuid_t fsuid = current_fsuid(); 2366da9592edSDavid Howells 23671da177e4SLinus Torvalds if (!(dir->i_mode & S_ISVTX)) 23681da177e4SLinus Torvalds return 0; 23698e96e3b7SEric W. Biederman if (uid_eq(inode->i_uid, fsuid)) 23701da177e4SLinus Torvalds return 0; 23718e96e3b7SEric W. Biederman if (uid_eq(dir->i_uid, fsuid)) 23721da177e4SLinus Torvalds return 0; 23731a48e2acSEric W. Biederman return !inode_capable(inode, CAP_FOWNER); 23741da177e4SLinus Torvalds } 23751da177e4SLinus Torvalds 23761da177e4SLinus Torvalds /* 23771da177e4SLinus Torvalds * Check whether we can remove a link victim from directory dir, check 23781da177e4SLinus Torvalds * whether the type of victim is right. 23791da177e4SLinus Torvalds * 1. We can't do it if dir is read-only (done in permission()) 23801da177e4SLinus Torvalds * 2. We should have write and exec permissions on dir 23811da177e4SLinus Torvalds * 3. We can't remove anything from append-only dir 23821da177e4SLinus Torvalds * 4. We can't do anything with immutable dir (done in permission()) 23831da177e4SLinus Torvalds * 5. If the sticky bit on dir is set we should either 23841da177e4SLinus Torvalds * a. be owner of dir, or 23851da177e4SLinus Torvalds * b. be owner of victim, or 23861da177e4SLinus Torvalds * c. have CAP_FOWNER capability 23871da177e4SLinus Torvalds * 6. If the victim is append-only or immutable we can't do antyhing with 23881da177e4SLinus Torvalds * links pointing to it. 23891da177e4SLinus Torvalds * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. 23901da177e4SLinus Torvalds * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. 23911da177e4SLinus Torvalds * 9. We can't remove a root or mountpoint. 23921da177e4SLinus Torvalds * 10. We don't allow removal of NFS sillyrenamed files; it's handled by 23931da177e4SLinus Torvalds * nfs_async_unlink(). 23941da177e4SLinus Torvalds */ 2395b18825a7SDavid Howells static int may_delete(struct inode *dir, struct dentry *victim, bool isdir) 23961da177e4SLinus Torvalds { 2397b18825a7SDavid Howells struct inode *inode = victim->d_inode; 23981da177e4SLinus Torvalds int error; 23991da177e4SLinus Torvalds 2400b18825a7SDavid Howells if (d_is_negative(victim)) 24011da177e4SLinus Torvalds return -ENOENT; 2402b18825a7SDavid Howells BUG_ON(!inode); 24031da177e4SLinus Torvalds 24041da177e4SLinus Torvalds BUG_ON(victim->d_parent->d_inode != dir); 24054fa6b5ecSJeff Layton audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); 24061da177e4SLinus Torvalds 2407f419a2e3SAl Viro error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 24081da177e4SLinus Torvalds if (error) 24091da177e4SLinus Torvalds return error; 24101da177e4SLinus Torvalds if (IS_APPEND(dir)) 24111da177e4SLinus Torvalds return -EPERM; 2412b18825a7SDavid Howells 2413b18825a7SDavid Howells if (check_sticky(dir, inode) || IS_APPEND(inode) || 2414b18825a7SDavid Howells IS_IMMUTABLE(inode) || IS_SWAPFILE(inode)) 24151da177e4SLinus Torvalds return -EPERM; 24161da177e4SLinus Torvalds if (isdir) { 241744b1d530SMiklos Szeredi if (!d_is_dir(victim)) 24181da177e4SLinus Torvalds return -ENOTDIR; 24191da177e4SLinus Torvalds if (IS_ROOT(victim)) 24201da177e4SLinus Torvalds return -EBUSY; 242144b1d530SMiklos Szeredi } else if (d_is_dir(victim)) 24221da177e4SLinus Torvalds return -EISDIR; 24231da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 24241da177e4SLinus Torvalds return -ENOENT; 24251da177e4SLinus Torvalds if (victim->d_flags & DCACHE_NFSFS_RENAMED) 24261da177e4SLinus Torvalds return -EBUSY; 24271da177e4SLinus Torvalds return 0; 24281da177e4SLinus Torvalds } 24291da177e4SLinus Torvalds 24301da177e4SLinus Torvalds /* Check whether we can create an object with dentry child in directory 24311da177e4SLinus Torvalds * dir. 24321da177e4SLinus Torvalds * 1. We can't do it if child already exists (open has special treatment for 24331da177e4SLinus Torvalds * this case, but since we are inlined it's OK) 24341da177e4SLinus Torvalds * 2. We can't do it if dir is read-only (done in permission()) 24351da177e4SLinus Torvalds * 3. We should have write and exec permissions on dir 24361da177e4SLinus Torvalds * 4. We can't do it if dir is immutable (done in permission()) 24371da177e4SLinus Torvalds */ 2438a95164d9SMiklos Szeredi static inline int may_create(struct inode *dir, struct dentry *child) 24391da177e4SLinus Torvalds { 244014e972b4SJeff Layton audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); 24411da177e4SLinus Torvalds if (child->d_inode) 24421da177e4SLinus Torvalds return -EEXIST; 24431da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 24441da177e4SLinus Torvalds return -ENOENT; 2445f419a2e3SAl Viro return inode_permission(dir, MAY_WRITE | MAY_EXEC); 24461da177e4SLinus Torvalds } 24471da177e4SLinus Torvalds 24481da177e4SLinus Torvalds /* 24491da177e4SLinus Torvalds * p1 and p2 should be directories on the same fs. 24501da177e4SLinus Torvalds */ 24511da177e4SLinus Torvalds struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 24521da177e4SLinus Torvalds { 24531da177e4SLinus Torvalds struct dentry *p; 24541da177e4SLinus Torvalds 24551da177e4SLinus Torvalds if (p1 == p2) { 2456f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 24571da177e4SLinus Torvalds return NULL; 24581da177e4SLinus Torvalds } 24591da177e4SLinus Torvalds 2460a11f3a05SArjan van de Ven mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 24611da177e4SLinus Torvalds 2462e2761a11SOGAWA Hirofumi p = d_ancestor(p2, p1); 2463e2761a11SOGAWA Hirofumi if (p) { 2464f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT); 2465f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD); 24661da177e4SLinus Torvalds return p; 24671da177e4SLinus Torvalds } 24681da177e4SLinus Torvalds 2469e2761a11SOGAWA Hirofumi p = d_ancestor(p1, p2); 2470e2761a11SOGAWA Hirofumi if (p) { 2471f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 2472f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); 24731da177e4SLinus Torvalds return p; 24741da177e4SLinus Torvalds } 24751da177e4SLinus Torvalds 2476f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 2477f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); 24781da177e4SLinus Torvalds return NULL; 24791da177e4SLinus Torvalds } 24801da177e4SLinus Torvalds 24811da177e4SLinus Torvalds void unlock_rename(struct dentry *p1, struct dentry *p2) 24821da177e4SLinus Torvalds { 24831b1dcc1bSJes Sorensen mutex_unlock(&p1->d_inode->i_mutex); 24841da177e4SLinus Torvalds if (p1 != p2) { 24851b1dcc1bSJes Sorensen mutex_unlock(&p2->d_inode->i_mutex); 2486a11f3a05SArjan van de Ven mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 24871da177e4SLinus Torvalds } 24881da177e4SLinus Torvalds } 24891da177e4SLinus Torvalds 24904acdaf27SAl Viro int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, 2491312b63fbSAl Viro bool want_excl) 24921da177e4SLinus Torvalds { 2493a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 24941da177e4SLinus Torvalds if (error) 24951da177e4SLinus Torvalds return error; 24961da177e4SLinus Torvalds 2497acfa4380SAl Viro if (!dir->i_op->create) 24981da177e4SLinus Torvalds return -EACCES; /* shouldn't it be ENOSYS? */ 24991da177e4SLinus Torvalds mode &= S_IALLUGO; 25001da177e4SLinus Torvalds mode |= S_IFREG; 25011da177e4SLinus Torvalds error = security_inode_create(dir, dentry, mode); 25021da177e4SLinus Torvalds if (error) 25031da177e4SLinus Torvalds return error; 2504312b63fbSAl Viro error = dir->i_op->create(dir, dentry, mode, want_excl); 2505a74574aaSStephen Smalley if (!error) 2506f38aa942SAmy Griffis fsnotify_create(dir, dentry); 25071da177e4SLinus Torvalds return error; 25081da177e4SLinus Torvalds } 25091da177e4SLinus Torvalds 251073d049a4SAl Viro static int may_open(struct path *path, int acc_mode, int flag) 25111da177e4SLinus Torvalds { 25123fb64190SChristoph Hellwig struct dentry *dentry = path->dentry; 25131da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 25141da177e4SLinus Torvalds int error; 25151da177e4SLinus Torvalds 2516bcda7652SAl Viro /* O_PATH? */ 2517bcda7652SAl Viro if (!acc_mode) 2518bcda7652SAl Viro return 0; 2519bcda7652SAl Viro 25201da177e4SLinus Torvalds if (!inode) 25211da177e4SLinus Torvalds return -ENOENT; 25221da177e4SLinus Torvalds 2523c8fe8f30SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 2524c8fe8f30SChristoph Hellwig case S_IFLNK: 25251da177e4SLinus Torvalds return -ELOOP; 2526c8fe8f30SChristoph Hellwig case S_IFDIR: 2527c8fe8f30SChristoph Hellwig if (acc_mode & MAY_WRITE) 25281da177e4SLinus Torvalds return -EISDIR; 2529c8fe8f30SChristoph Hellwig break; 2530c8fe8f30SChristoph Hellwig case S_IFBLK: 2531c8fe8f30SChristoph Hellwig case S_IFCHR: 25323fb64190SChristoph Hellwig if (path->mnt->mnt_flags & MNT_NODEV) 25331da177e4SLinus Torvalds return -EACCES; 2534c8fe8f30SChristoph Hellwig /*FALLTHRU*/ 2535c8fe8f30SChristoph Hellwig case S_IFIFO: 2536c8fe8f30SChristoph Hellwig case S_IFSOCK: 25371da177e4SLinus Torvalds flag &= ~O_TRUNC; 2538c8fe8f30SChristoph Hellwig break; 25394a3fd211SDave Hansen } 2540b41572e9SDave Hansen 25413fb64190SChristoph Hellwig error = inode_permission(inode, acc_mode); 2542b41572e9SDave Hansen if (error) 2543b41572e9SDave Hansen return error; 25446146f0d5SMimi Zohar 25451da177e4SLinus Torvalds /* 25461da177e4SLinus Torvalds * An append-only file must be opened in append mode for writing. 25471da177e4SLinus Torvalds */ 25481da177e4SLinus Torvalds if (IS_APPEND(inode)) { 25498737c930SAl Viro if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) 25507715b521SAl Viro return -EPERM; 25511da177e4SLinus Torvalds if (flag & O_TRUNC) 25527715b521SAl Viro return -EPERM; 25531da177e4SLinus Torvalds } 25541da177e4SLinus Torvalds 25551da177e4SLinus Torvalds /* O_NOATIME can only be set by the owner or superuser */ 25562e149670SSerge E. Hallyn if (flag & O_NOATIME && !inode_owner_or_capable(inode)) 25577715b521SAl Viro return -EPERM; 25581da177e4SLinus Torvalds 2559f3c7691eSJ. Bruce Fields return 0; 25607715b521SAl Viro } 25617715b521SAl Viro 2562e1181ee6SJeff Layton static int handle_truncate(struct file *filp) 25637715b521SAl Viro { 2564e1181ee6SJeff Layton struct path *path = &filp->f_path; 25657715b521SAl Viro struct inode *inode = path->dentry->d_inode; 25667715b521SAl Viro int error = get_write_access(inode); 25671da177e4SLinus Torvalds if (error) 25687715b521SAl Viro return error; 25691da177e4SLinus Torvalds /* 25701da177e4SLinus Torvalds * Refuse to truncate files with mandatory locks held on them. 25711da177e4SLinus Torvalds */ 25721da177e4SLinus Torvalds error = locks_verify_locked(inode); 2573be6d3e56SKentaro Takeda if (!error) 2574ea0d3ab2STetsuo Handa error = security_path_truncate(path); 25751da177e4SLinus Torvalds if (!error) { 25767715b521SAl Viro error = do_truncate(path->dentry, 0, 2577d139d7ffSMiklos Szeredi ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, 2578e1181ee6SJeff Layton filp); 25791da177e4SLinus Torvalds } 25801da177e4SLinus Torvalds put_write_access(inode); 2581acd0c935SMimi Zohar return error; 25821da177e4SLinus Torvalds } 25831da177e4SLinus Torvalds 2584d57999e1SDave Hansen static inline int open_to_namei_flags(int flag) 2585d57999e1SDave Hansen { 25868a5e929dSAl Viro if ((flag & O_ACCMODE) == 3) 25878a5e929dSAl Viro flag--; 2588d57999e1SDave Hansen return flag; 2589d57999e1SDave Hansen } 2590d57999e1SDave Hansen 2591d18e9008SMiklos Szeredi static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode) 2592d18e9008SMiklos Szeredi { 2593d18e9008SMiklos Szeredi int error = security_path_mknod(dir, dentry, mode, 0); 2594d18e9008SMiklos Szeredi if (error) 2595d18e9008SMiklos Szeredi return error; 2596d18e9008SMiklos Szeredi 2597d18e9008SMiklos Szeredi error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC); 2598d18e9008SMiklos Szeredi if (error) 2599d18e9008SMiklos Szeredi return error; 2600d18e9008SMiklos Szeredi 2601d18e9008SMiklos Szeredi return security_inode_create(dir->dentry->d_inode, dentry, mode); 2602d18e9008SMiklos Szeredi } 2603d18e9008SMiklos Szeredi 26041acf0af9SDavid Howells /* 26051acf0af9SDavid Howells * Attempt to atomically look up, create and open a file from a negative 26061acf0af9SDavid Howells * dentry. 26071acf0af9SDavid Howells * 26081acf0af9SDavid Howells * Returns 0 if successful. The file will have been created and attached to 26091acf0af9SDavid Howells * @file by the filesystem calling finish_open(). 26101acf0af9SDavid Howells * 26111acf0af9SDavid Howells * Returns 1 if the file was looked up only or didn't need creating. The 26121acf0af9SDavid Howells * caller will need to perform the open themselves. @path will have been 26131acf0af9SDavid Howells * updated to point to the new dentry. This may be negative. 26141acf0af9SDavid Howells * 26151acf0af9SDavid Howells * Returns an error code otherwise. 26161acf0af9SDavid Howells */ 26172675a4ebSAl Viro static int atomic_open(struct nameidata *nd, struct dentry *dentry, 261830d90494SAl Viro struct path *path, struct file *file, 2619015c3bbcSMiklos Szeredi const struct open_flags *op, 262064894cf8SAl Viro bool got_write, bool need_lookup, 262147237687SAl Viro int *opened) 2622d18e9008SMiklos Szeredi { 2623d18e9008SMiklos Szeredi struct inode *dir = nd->path.dentry->d_inode; 2624d18e9008SMiklos Szeredi unsigned open_flag = open_to_namei_flags(op->open_flag); 2625d18e9008SMiklos Szeredi umode_t mode; 2626d18e9008SMiklos Szeredi int error; 2627d18e9008SMiklos Szeredi int acc_mode; 2628d18e9008SMiklos Szeredi int create_error = 0; 2629d18e9008SMiklos Szeredi struct dentry *const DENTRY_NOT_SET = (void *) -1UL; 2630116cc022SMiklos Szeredi bool excl; 2631d18e9008SMiklos Szeredi 2632d18e9008SMiklos Szeredi BUG_ON(dentry->d_inode); 2633d18e9008SMiklos Szeredi 2634d18e9008SMiklos Szeredi /* Don't create child dentry for a dead directory. */ 2635d18e9008SMiklos Szeredi if (unlikely(IS_DEADDIR(dir))) { 26362675a4ebSAl Viro error = -ENOENT; 2637d18e9008SMiklos Szeredi goto out; 2638d18e9008SMiklos Szeredi } 2639d18e9008SMiklos Szeredi 264062b259d8SMiklos Szeredi mode = op->mode; 2641d18e9008SMiklos Szeredi if ((open_flag & O_CREAT) && !IS_POSIXACL(dir)) 2642d18e9008SMiklos Szeredi mode &= ~current_umask(); 2643d18e9008SMiklos Szeredi 2644116cc022SMiklos Szeredi excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT); 2645116cc022SMiklos Szeredi if (excl) 2646d18e9008SMiklos Szeredi open_flag &= ~O_TRUNC; 2647d18e9008SMiklos Szeredi 2648d18e9008SMiklos Szeredi /* 2649d18e9008SMiklos Szeredi * Checking write permission is tricky, bacuse we don't know if we are 2650d18e9008SMiklos Szeredi * going to actually need it: O_CREAT opens should work as long as the 2651d18e9008SMiklos Szeredi * file exists. But checking existence breaks atomicity. The trick is 2652d18e9008SMiklos Szeredi * to check access and if not granted clear O_CREAT from the flags. 2653d18e9008SMiklos Szeredi * 2654d18e9008SMiklos Szeredi * Another problem is returing the "right" error value (e.g. for an 2655d18e9008SMiklos Szeredi * O_EXCL open we want to return EEXIST not EROFS). 2656d18e9008SMiklos Szeredi */ 265764894cf8SAl Viro if (((open_flag & (O_CREAT | O_TRUNC)) || 265864894cf8SAl Viro (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) { 265964894cf8SAl Viro if (!(open_flag & O_CREAT)) { 2660d18e9008SMiklos Szeredi /* 2661d18e9008SMiklos Szeredi * No O_CREATE -> atomicity not a requirement -> fall 2662d18e9008SMiklos Szeredi * back to lookup + open 2663d18e9008SMiklos Szeredi */ 2664d18e9008SMiklos Szeredi goto no_open; 2665d18e9008SMiklos Szeredi } else if (open_flag & (O_EXCL | O_TRUNC)) { 2666d18e9008SMiklos Szeredi /* Fall back and fail with the right error */ 266764894cf8SAl Viro create_error = -EROFS; 2668d18e9008SMiklos Szeredi goto no_open; 2669d18e9008SMiklos Szeredi } else { 2670d18e9008SMiklos Szeredi /* No side effects, safe to clear O_CREAT */ 267164894cf8SAl Viro create_error = -EROFS; 2672d18e9008SMiklos Szeredi open_flag &= ~O_CREAT; 2673d18e9008SMiklos Szeredi } 2674d18e9008SMiklos Szeredi } 2675d18e9008SMiklos Szeredi 2676d18e9008SMiklos Szeredi if (open_flag & O_CREAT) { 267738227f78SMiklos Szeredi error = may_o_create(&nd->path, dentry, mode); 2678d18e9008SMiklos Szeredi if (error) { 2679d18e9008SMiklos Szeredi create_error = error; 2680d18e9008SMiklos Szeredi if (open_flag & O_EXCL) 2681d18e9008SMiklos Szeredi goto no_open; 2682d18e9008SMiklos Szeredi open_flag &= ~O_CREAT; 2683d18e9008SMiklos Szeredi } 2684d18e9008SMiklos Szeredi } 2685d18e9008SMiklos Szeredi 2686d18e9008SMiklos Szeredi if (nd->flags & LOOKUP_DIRECTORY) 2687d18e9008SMiklos Szeredi open_flag |= O_DIRECTORY; 2688d18e9008SMiklos Szeredi 268930d90494SAl Viro file->f_path.dentry = DENTRY_NOT_SET; 269030d90494SAl Viro file->f_path.mnt = nd->path.mnt; 269130d90494SAl Viro error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode, 269247237687SAl Viro opened); 2693d9585277SAl Viro if (error < 0) { 2694d9585277SAl Viro if (create_error && error == -ENOENT) 2695d9585277SAl Viro error = create_error; 2696d18e9008SMiklos Szeredi goto out; 2697d18e9008SMiklos Szeredi } 2698d18e9008SMiklos Szeredi 2699d9585277SAl Viro if (error) { /* returned 1, that is */ 270030d90494SAl Viro if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { 27012675a4ebSAl Viro error = -EIO; 2702d18e9008SMiklos Szeredi goto out; 2703d18e9008SMiklos Szeredi } 270430d90494SAl Viro if (file->f_path.dentry) { 2705d18e9008SMiklos Szeredi dput(dentry); 270630d90494SAl Viro dentry = file->f_path.dentry; 2707d18e9008SMiklos Szeredi } 270803da633aSAl Viro if (*opened & FILE_CREATED) 270903da633aSAl Viro fsnotify_create(dir, dentry); 271003da633aSAl Viro if (!dentry->d_inode) { 271103da633aSAl Viro WARN_ON(*opened & FILE_CREATED); 271203da633aSAl Viro if (create_error) { 271362b2ce96SSage Weil error = create_error; 271462b2ce96SSage Weil goto out; 271562b2ce96SSage Weil } 271603da633aSAl Viro } else { 271703da633aSAl Viro if (excl && !(*opened & FILE_CREATED)) { 271803da633aSAl Viro error = -EEXIST; 271903da633aSAl Viro goto out; 272003da633aSAl Viro } 272103da633aSAl Viro } 2722d18e9008SMiklos Szeredi goto looked_up; 2723d18e9008SMiklos Szeredi } 2724d18e9008SMiklos Szeredi 2725d18e9008SMiklos Szeredi /* 2726d18e9008SMiklos Szeredi * We didn't have the inode before the open, so check open permission 2727d18e9008SMiklos Szeredi * here. 2728d18e9008SMiklos Szeredi */ 272903da633aSAl Viro acc_mode = op->acc_mode; 273003da633aSAl Viro if (*opened & FILE_CREATED) { 273103da633aSAl Viro WARN_ON(!(open_flag & O_CREAT)); 273203da633aSAl Viro fsnotify_create(dir, dentry); 273303da633aSAl Viro acc_mode = MAY_OPEN; 273403da633aSAl Viro } 27352675a4ebSAl Viro error = may_open(&file->f_path, acc_mode, open_flag); 27362675a4ebSAl Viro if (error) 27372675a4ebSAl Viro fput(file); 2738d18e9008SMiklos Szeredi 2739d18e9008SMiklos Szeredi out: 2740d18e9008SMiklos Szeredi dput(dentry); 27412675a4ebSAl Viro return error; 2742d18e9008SMiklos Szeredi 2743d18e9008SMiklos Szeredi no_open: 2744d18e9008SMiklos Szeredi if (need_lookup) { 274572bd866aSAl Viro dentry = lookup_real(dir, dentry, nd->flags); 2746d18e9008SMiklos Szeredi if (IS_ERR(dentry)) 27472675a4ebSAl Viro return PTR_ERR(dentry); 2748d18e9008SMiklos Szeredi 2749d18e9008SMiklos Szeredi if (create_error) { 2750d18e9008SMiklos Szeredi int open_flag = op->open_flag; 2751d18e9008SMiklos Szeredi 27522675a4ebSAl Viro error = create_error; 2753d18e9008SMiklos Szeredi if ((open_flag & O_EXCL)) { 2754d18e9008SMiklos Szeredi if (!dentry->d_inode) 2755d18e9008SMiklos Szeredi goto out; 2756d18e9008SMiklos Szeredi } else if (!dentry->d_inode) { 2757d18e9008SMiklos Szeredi goto out; 2758d18e9008SMiklos Szeredi } else if ((open_flag & O_TRUNC) && 2759d18e9008SMiklos Szeredi S_ISREG(dentry->d_inode->i_mode)) { 2760d18e9008SMiklos Szeredi goto out; 2761d18e9008SMiklos Szeredi } 2762d18e9008SMiklos Szeredi /* will fail later, go on to get the right error */ 2763d18e9008SMiklos Szeredi } 2764d18e9008SMiklos Szeredi } 2765d18e9008SMiklos Szeredi looked_up: 2766d18e9008SMiklos Szeredi path->dentry = dentry; 2767d18e9008SMiklos Szeredi path->mnt = nd->path.mnt; 27682675a4ebSAl Viro return 1; 2769d18e9008SMiklos Szeredi } 2770d18e9008SMiklos Szeredi 277131e6b01fSNick Piggin /* 27721acf0af9SDavid Howells * Look up and maybe create and open the last component. 2773d58ffd35SMiklos Szeredi * 2774d58ffd35SMiklos Szeredi * Must be called with i_mutex held on parent. 2775d58ffd35SMiklos Szeredi * 27761acf0af9SDavid Howells * Returns 0 if the file was successfully atomically created (if necessary) and 27771acf0af9SDavid Howells * opened. In this case the file will be returned attached to @file. 27781acf0af9SDavid Howells * 27791acf0af9SDavid Howells * Returns 1 if the file was not completely opened at this time, though lookups 27801acf0af9SDavid Howells * and creations will have been performed and the dentry returned in @path will 27811acf0af9SDavid Howells * be positive upon return if O_CREAT was specified. If O_CREAT wasn't 27821acf0af9SDavid Howells * specified then a negative dentry may be returned. 27831acf0af9SDavid Howells * 27841acf0af9SDavid Howells * An error code is returned otherwise. 27851acf0af9SDavid Howells * 27861acf0af9SDavid Howells * FILE_CREATE will be set in @*opened if the dentry was created and will be 27871acf0af9SDavid Howells * cleared otherwise prior to returning. 2788d58ffd35SMiklos Szeredi */ 27892675a4ebSAl Viro static int lookup_open(struct nameidata *nd, struct path *path, 279030d90494SAl Viro struct file *file, 2791d58ffd35SMiklos Szeredi const struct open_flags *op, 279264894cf8SAl Viro bool got_write, int *opened) 2793d58ffd35SMiklos Szeredi { 2794d58ffd35SMiklos Szeredi struct dentry *dir = nd->path.dentry; 279554ef4872SMiklos Szeredi struct inode *dir_inode = dir->d_inode; 2796d58ffd35SMiklos Szeredi struct dentry *dentry; 2797d58ffd35SMiklos Szeredi int error; 279854ef4872SMiklos Szeredi bool need_lookup; 2799d58ffd35SMiklos Szeredi 280047237687SAl Viro *opened &= ~FILE_CREATED; 2801201f956eSAl Viro dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup); 2802d58ffd35SMiklos Szeredi if (IS_ERR(dentry)) 28032675a4ebSAl Viro return PTR_ERR(dentry); 2804d58ffd35SMiklos Szeredi 2805d18e9008SMiklos Szeredi /* Cached positive dentry: will open in f_op->open */ 2806d18e9008SMiklos Szeredi if (!need_lookup && dentry->d_inode) 2807d18e9008SMiklos Szeredi goto out_no_open; 2808d18e9008SMiklos Szeredi 2809d18e9008SMiklos Szeredi if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) { 281064894cf8SAl Viro return atomic_open(nd, dentry, path, file, op, got_write, 281147237687SAl Viro need_lookup, opened); 2812d18e9008SMiklos Szeredi } 2813d18e9008SMiklos Szeredi 281454ef4872SMiklos Szeredi if (need_lookup) { 281554ef4872SMiklos Szeredi BUG_ON(dentry->d_inode); 281654ef4872SMiklos Szeredi 281772bd866aSAl Viro dentry = lookup_real(dir_inode, dentry, nd->flags); 281854ef4872SMiklos Szeredi if (IS_ERR(dentry)) 28192675a4ebSAl Viro return PTR_ERR(dentry); 282054ef4872SMiklos Szeredi } 282154ef4872SMiklos Szeredi 2822d58ffd35SMiklos Szeredi /* Negative dentry, just create the file */ 2823d58ffd35SMiklos Szeredi if (!dentry->d_inode && (op->open_flag & O_CREAT)) { 2824d58ffd35SMiklos Szeredi umode_t mode = op->mode; 2825d58ffd35SMiklos Szeredi if (!IS_POSIXACL(dir->d_inode)) 2826d58ffd35SMiklos Szeredi mode &= ~current_umask(); 2827d58ffd35SMiklos Szeredi /* 2828d58ffd35SMiklos Szeredi * This write is needed to ensure that a 2829d58ffd35SMiklos Szeredi * rw->ro transition does not occur between 2830d58ffd35SMiklos Szeredi * the time when the file is created and when 2831d58ffd35SMiklos Szeredi * a permanent write count is taken through 2832015c3bbcSMiklos Szeredi * the 'struct file' in finish_open(). 2833d58ffd35SMiklos Szeredi */ 283464894cf8SAl Viro if (!got_write) { 283564894cf8SAl Viro error = -EROFS; 2836d58ffd35SMiklos Szeredi goto out_dput; 283764894cf8SAl Viro } 283847237687SAl Viro *opened |= FILE_CREATED; 2839d58ffd35SMiklos Szeredi error = security_path_mknod(&nd->path, dentry, mode, 0); 2840d58ffd35SMiklos Szeredi if (error) 2841d58ffd35SMiklos Szeredi goto out_dput; 2842312b63fbSAl Viro error = vfs_create(dir->d_inode, dentry, mode, 2843312b63fbSAl Viro nd->flags & LOOKUP_EXCL); 2844d58ffd35SMiklos Szeredi if (error) 2845d58ffd35SMiklos Szeredi goto out_dput; 2846d58ffd35SMiklos Szeredi } 2847d18e9008SMiklos Szeredi out_no_open: 2848d58ffd35SMiklos Szeredi path->dentry = dentry; 2849d58ffd35SMiklos Szeredi path->mnt = nd->path.mnt; 28502675a4ebSAl Viro return 1; 2851d58ffd35SMiklos Szeredi 2852d58ffd35SMiklos Szeredi out_dput: 2853d58ffd35SMiklos Szeredi dput(dentry); 28542675a4ebSAl Viro return error; 2855d58ffd35SMiklos Szeredi } 2856d58ffd35SMiklos Szeredi 2857d58ffd35SMiklos Szeredi /* 2858fe2d35ffSAl Viro * Handle the last step of open() 285931e6b01fSNick Piggin */ 28602675a4ebSAl Viro static int do_last(struct nameidata *nd, struct path *path, 286130d90494SAl Viro struct file *file, const struct open_flags *op, 2862669abf4eSJeff Layton int *opened, struct filename *name) 2863fb1cc555SAl Viro { 2864a1e28038SAl Viro struct dentry *dir = nd->path.dentry; 2865ca344a89SAl Viro int open_flag = op->open_flag; 286677d660a8SMiklos Szeredi bool will_truncate = (open_flag & O_TRUNC) != 0; 286764894cf8SAl Viro bool got_write = false; 2868bcda7652SAl Viro int acc_mode = op->acc_mode; 2869a1eb3315SMiklos Szeredi struct inode *inode; 287077d660a8SMiklos Szeredi bool symlink_ok = false; 287116b1c1cdSMiklos Szeredi struct path save_parent = { .dentry = NULL, .mnt = NULL }; 287216b1c1cdSMiklos Szeredi bool retried = false; 287316c2cd71SAl Viro int error; 2874fb1cc555SAl Viro 2875c3e380b0SAl Viro nd->flags &= ~LOOKUP_PARENT; 2876c3e380b0SAl Viro nd->flags |= op->intent; 2877c3e380b0SAl Viro 2878bc77daa7SAl Viro if (nd->last_type != LAST_NORM) { 2879fe2d35ffSAl Viro error = handle_dots(nd, nd->last_type); 2880fe2d35ffSAl Viro if (error) 28812675a4ebSAl Viro return error; 2882e83db167SMiklos Szeredi goto finish_open; 28831f36f774SAl Viro } 2884a2c36b45SAl Viro 2885ca344a89SAl Viro if (!(open_flag & O_CREAT)) { 2886fe2d35ffSAl Viro if (nd->last.name[nd->last.len]) 2887fe2d35ffSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 2888bcda7652SAl Viro if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW)) 288977d660a8SMiklos Szeredi symlink_ok = true; 2890fe2d35ffSAl Viro /* we _can_ be in RCU mode here */ 2891e97cdc87SAl Viro error = lookup_fast(nd, path, &inode); 289271574865SMiklos Szeredi if (likely(!error)) 289371574865SMiklos Szeredi goto finish_lookup; 289471574865SMiklos Szeredi 2895ce57dfc1SAl Viro if (error < 0) 28962675a4ebSAl Viro goto out; 2897a1eb3315SMiklos Szeredi 289837d7fffcSMiklos Szeredi BUG_ON(nd->inode != dir->d_inode); 2899b6183df7SMiklos Szeredi } else { 2900fe2d35ffSAl Viro /* create side of things */ 2901a3fbbde7SAl Viro /* 2902b6183df7SMiklos Szeredi * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED 2903b6183df7SMiklos Szeredi * has been cleared when we got to the last component we are 2904b6183df7SMiklos Szeredi * about to look up 2905a3fbbde7SAl Viro */ 29069f1fafeeSAl Viro error = complete_walk(nd); 29079f1fafeeSAl Viro if (error) 29082675a4ebSAl Viro return error; 2909fe2d35ffSAl Viro 291033e2208aSJeff Layton audit_inode(name, dir, LOOKUP_PARENT); 291116c2cd71SAl Viro error = -EISDIR; 29121f36f774SAl Viro /* trailing slashes? */ 291331e6b01fSNick Piggin if (nd->last.name[nd->last.len]) 29142675a4ebSAl Viro goto out; 2915b6183df7SMiklos Szeredi } 29161f36f774SAl Viro 291716b1c1cdSMiklos Szeredi retry_lookup: 291864894cf8SAl Viro if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { 291964894cf8SAl Viro error = mnt_want_write(nd->path.mnt); 292064894cf8SAl Viro if (!error) 292164894cf8SAl Viro got_write = true; 292264894cf8SAl Viro /* 292364894cf8SAl Viro * do _not_ fail yet - we might not need that or fail with 292464894cf8SAl Viro * a different error; let lookup_open() decide; we'll be 292564894cf8SAl Viro * dropping this one anyway. 292664894cf8SAl Viro */ 292764894cf8SAl Viro } 2928a1e28038SAl Viro mutex_lock(&dir->d_inode->i_mutex); 292964894cf8SAl Viro error = lookup_open(nd, path, file, op, got_write, opened); 2930fb1cc555SAl Viro mutex_unlock(&dir->d_inode->i_mutex); 2931fb1cc555SAl Viro 29322675a4ebSAl Viro if (error <= 0) { 29332675a4ebSAl Viro if (error) 2934d58ffd35SMiklos Szeredi goto out; 29356c0d46c4SAl Viro 293647237687SAl Viro if ((*opened & FILE_CREATED) || 2937496ad9aaSAl Viro !S_ISREG(file_inode(file)->i_mode)) 293877d660a8SMiklos Szeredi will_truncate = false; 2939d18e9008SMiklos Szeredi 2940adb5c247SJeff Layton audit_inode(name, file->f_path.dentry, 0); 2941d18e9008SMiklos Szeredi goto opened; 2942d18e9008SMiklos Szeredi } 2943d18e9008SMiklos Szeredi 294447237687SAl Viro if (*opened & FILE_CREATED) { 29459b44f1b3SAl Viro /* Don't check for write permission, don't truncate */ 2946ca344a89SAl Viro open_flag &= ~O_TRUNC; 294777d660a8SMiklos Szeredi will_truncate = false; 2948bcda7652SAl Viro acc_mode = MAY_OPEN; 2949d58ffd35SMiklos Szeredi path_to_nameidata(path, nd); 2950e83db167SMiklos Szeredi goto finish_open_created; 2951fb1cc555SAl Viro } 2952fb1cc555SAl Viro 2953fb1cc555SAl Viro /* 29543134f37eSJeff Layton * create/update audit record if it already exists. 2955fb1cc555SAl Viro */ 2956b18825a7SDavid Howells if (d_is_positive(path->dentry)) 2957adb5c247SJeff Layton audit_inode(name, path->dentry, 0); 2958fb1cc555SAl Viro 2959d18e9008SMiklos Szeredi /* 2960d18e9008SMiklos Szeredi * If atomic_open() acquired write access it is dropped now due to 2961d18e9008SMiklos Szeredi * possible mount and symlink following (this might be optimized away if 2962d18e9008SMiklos Szeredi * necessary...) 2963d18e9008SMiklos Szeredi */ 296464894cf8SAl Viro if (got_write) { 2965d18e9008SMiklos Szeredi mnt_drop_write(nd->path.mnt); 296664894cf8SAl Viro got_write = false; 2967d18e9008SMiklos Szeredi } 2968d18e9008SMiklos Szeredi 2969fb1cc555SAl Viro error = -EEXIST; 2970f8310c59SAl Viro if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT)) 2971fb1cc555SAl Viro goto exit_dput; 2972fb1cc555SAl Viro 29739875cf80SDavid Howells error = follow_managed(path, nd->flags); 29749875cf80SDavid Howells if (error < 0) 2975fb1cc555SAl Viro goto exit_dput; 2976fb1cc555SAl Viro 2977a3fbbde7SAl Viro if (error) 2978a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 2979a3fbbde7SAl Viro 2980decf3400SMiklos Szeredi BUG_ON(nd->flags & LOOKUP_RCU); 2981decf3400SMiklos Szeredi inode = path->dentry->d_inode; 29825f5daac1SMiklos Szeredi finish_lookup: 29835f5daac1SMiklos Szeredi /* we _can_ be in RCU mode here */ 2984fb1cc555SAl Viro error = -ENOENT; 2985b18825a7SDavid Howells if (d_is_negative(path->dentry)) { 298654c33e7fSMiklos Szeredi path_to_nameidata(path, nd); 29872675a4ebSAl Viro goto out; 298854c33e7fSMiklos Szeredi } 29899e67f361SAl Viro 2990b18825a7SDavid Howells if (should_follow_link(path->dentry, !symlink_ok)) { 2991d45ea867SMiklos Szeredi if (nd->flags & LOOKUP_RCU) { 2992d45ea867SMiklos Szeredi if (unlikely(unlazy_walk(nd, path->dentry))) { 2993d45ea867SMiklos Szeredi error = -ECHILD; 29942675a4ebSAl Viro goto out; 2995d45ea867SMiklos Szeredi } 2996d45ea867SMiklos Szeredi } 2997d45ea867SMiklos Szeredi BUG_ON(inode != path->dentry->d_inode); 29982675a4ebSAl Viro return 1; 2999d45ea867SMiklos Szeredi } 3000fb1cc555SAl Viro 300116b1c1cdSMiklos Szeredi if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) { 3002fb1cc555SAl Viro path_to_nameidata(path, nd); 300316b1c1cdSMiklos Szeredi } else { 300416b1c1cdSMiklos Szeredi save_parent.dentry = nd->path.dentry; 300516b1c1cdSMiklos Szeredi save_parent.mnt = mntget(path->mnt); 300616b1c1cdSMiklos Szeredi nd->path.dentry = path->dentry; 300716b1c1cdSMiklos Szeredi 300816b1c1cdSMiklos Szeredi } 3009decf3400SMiklos Szeredi nd->inode = inode; 3010a3fbbde7SAl Viro /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ 3011bc77daa7SAl Viro finish_open: 3012a3fbbde7SAl Viro error = complete_walk(nd); 301316b1c1cdSMiklos Szeredi if (error) { 301416b1c1cdSMiklos Szeredi path_put(&save_parent); 30152675a4ebSAl Viro return error; 301616b1c1cdSMiklos Szeredi } 3017bc77daa7SAl Viro audit_inode(name, nd->path.dentry, 0); 3018fb1cc555SAl Viro error = -EISDIR; 301944b1d530SMiklos Szeredi if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry)) 30202675a4ebSAl Viro goto out; 3021af2f5542SMiklos Szeredi error = -ENOTDIR; 302244b1d530SMiklos Szeredi if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry)) 30232675a4ebSAl Viro goto out; 30246c0d46c4SAl Viro if (!S_ISREG(nd->inode->i_mode)) 302577d660a8SMiklos Szeredi will_truncate = false; 30266c0d46c4SAl Viro 30270f9d1a10SAl Viro if (will_truncate) { 30280f9d1a10SAl Viro error = mnt_want_write(nd->path.mnt); 30290f9d1a10SAl Viro if (error) 30302675a4ebSAl Viro goto out; 303164894cf8SAl Viro got_write = true; 30320f9d1a10SAl Viro } 3033e83db167SMiklos Szeredi finish_open_created: 3034bcda7652SAl Viro error = may_open(&nd->path, acc_mode, open_flag); 3035ca344a89SAl Viro if (error) 30362675a4ebSAl Viro goto out; 303730d90494SAl Viro file->f_path.mnt = nd->path.mnt; 303830d90494SAl Viro error = finish_open(file, nd->path.dentry, NULL, opened); 303930d90494SAl Viro if (error) { 304030d90494SAl Viro if (error == -EOPENSTALE) 3041f60dc3dbSMiklos Szeredi goto stale_open; 3042015c3bbcSMiklos Szeredi goto out; 3043f60dc3dbSMiklos Szeredi } 3044a8277b9bSMiklos Szeredi opened: 30452675a4ebSAl Viro error = open_check_o_direct(file); 3046015c3bbcSMiklos Szeredi if (error) 3047015c3bbcSMiklos Szeredi goto exit_fput; 30482675a4ebSAl Viro error = ima_file_check(file, op->acc_mode); 3049aa4caadbSMiklos Szeredi if (error) 3050aa4caadbSMiklos Szeredi goto exit_fput; 3051aa4caadbSMiklos Szeredi 30520f9d1a10SAl Viro if (will_truncate) { 30532675a4ebSAl Viro error = handle_truncate(file); 3054aa4caadbSMiklos Szeredi if (error) 3055aa4caadbSMiklos Szeredi goto exit_fput; 30560f9d1a10SAl Viro } 3057ca344a89SAl Viro out: 305864894cf8SAl Viro if (got_write) 30590f9d1a10SAl Viro mnt_drop_write(nd->path.mnt); 306016b1c1cdSMiklos Szeredi path_put(&save_parent); 3061e276ae67SMiklos Szeredi terminate_walk(nd); 30622675a4ebSAl Viro return error; 3063fb1cc555SAl Viro 3064fb1cc555SAl Viro exit_dput: 3065fb1cc555SAl Viro path_put_conditional(path, nd); 3066ca344a89SAl Viro goto out; 3067015c3bbcSMiklos Szeredi exit_fput: 30682675a4ebSAl Viro fput(file); 30692675a4ebSAl Viro goto out; 3070015c3bbcSMiklos Szeredi 3071f60dc3dbSMiklos Szeredi stale_open: 3072f60dc3dbSMiklos Szeredi /* If no saved parent or already retried then can't retry */ 3073f60dc3dbSMiklos Szeredi if (!save_parent.dentry || retried) 3074f60dc3dbSMiklos Szeredi goto out; 3075f60dc3dbSMiklos Szeredi 3076f60dc3dbSMiklos Szeredi BUG_ON(save_parent.dentry != dir); 3077f60dc3dbSMiklos Szeredi path_put(&nd->path); 3078f60dc3dbSMiklos Szeredi nd->path = save_parent; 3079f60dc3dbSMiklos Szeredi nd->inode = dir->d_inode; 3080f60dc3dbSMiklos Szeredi save_parent.mnt = NULL; 3081f60dc3dbSMiklos Szeredi save_parent.dentry = NULL; 308264894cf8SAl Viro if (got_write) { 3083f60dc3dbSMiklos Szeredi mnt_drop_write(nd->path.mnt); 308464894cf8SAl Viro got_write = false; 3085f60dc3dbSMiklos Szeredi } 3086f60dc3dbSMiklos Szeredi retried = true; 3087f60dc3dbSMiklos Szeredi goto retry_lookup; 3088fb1cc555SAl Viro } 3089fb1cc555SAl Viro 309060545d0dSAl Viro static int do_tmpfile(int dfd, struct filename *pathname, 309160545d0dSAl Viro struct nameidata *nd, int flags, 309260545d0dSAl Viro const struct open_flags *op, 309360545d0dSAl Viro struct file *file, int *opened) 309460545d0dSAl Viro { 309560545d0dSAl Viro static const struct qstr name = QSTR_INIT("/", 1); 309660545d0dSAl Viro struct dentry *dentry, *child; 309760545d0dSAl Viro struct inode *dir; 309860545d0dSAl Viro int error = path_lookupat(dfd, pathname->name, 309960545d0dSAl Viro flags | LOOKUP_DIRECTORY, nd); 310060545d0dSAl Viro if (unlikely(error)) 310160545d0dSAl Viro return error; 310260545d0dSAl Viro error = mnt_want_write(nd->path.mnt); 310360545d0dSAl Viro if (unlikely(error)) 310460545d0dSAl Viro goto out; 310560545d0dSAl Viro /* we want directory to be writable */ 310660545d0dSAl Viro error = inode_permission(nd->inode, MAY_WRITE | MAY_EXEC); 310760545d0dSAl Viro if (error) 310860545d0dSAl Viro goto out2; 310960545d0dSAl Viro dentry = nd->path.dentry; 311060545d0dSAl Viro dir = dentry->d_inode; 311160545d0dSAl Viro if (!dir->i_op->tmpfile) { 311260545d0dSAl Viro error = -EOPNOTSUPP; 311360545d0dSAl Viro goto out2; 311460545d0dSAl Viro } 311560545d0dSAl Viro child = d_alloc(dentry, &name); 311660545d0dSAl Viro if (unlikely(!child)) { 311760545d0dSAl Viro error = -ENOMEM; 311860545d0dSAl Viro goto out2; 311960545d0dSAl Viro } 312060545d0dSAl Viro nd->flags &= ~LOOKUP_DIRECTORY; 312160545d0dSAl Viro nd->flags |= op->intent; 312260545d0dSAl Viro dput(nd->path.dentry); 312360545d0dSAl Viro nd->path.dentry = child; 312460545d0dSAl Viro error = dir->i_op->tmpfile(dir, nd->path.dentry, op->mode); 312560545d0dSAl Viro if (error) 312660545d0dSAl Viro goto out2; 312760545d0dSAl Viro audit_inode(pathname, nd->path.dentry, 0); 312860545d0dSAl Viro error = may_open(&nd->path, op->acc_mode, op->open_flag); 312960545d0dSAl Viro if (error) 313060545d0dSAl Viro goto out2; 313160545d0dSAl Viro file->f_path.mnt = nd->path.mnt; 313260545d0dSAl Viro error = finish_open(file, nd->path.dentry, NULL, opened); 313360545d0dSAl Viro if (error) 313460545d0dSAl Viro goto out2; 313560545d0dSAl Viro error = open_check_o_direct(file); 3136f4e0c30cSAl Viro if (error) { 313760545d0dSAl Viro fput(file); 3138f4e0c30cSAl Viro } else if (!(op->open_flag & O_EXCL)) { 3139f4e0c30cSAl Viro struct inode *inode = file_inode(file); 3140f4e0c30cSAl Viro spin_lock(&inode->i_lock); 3141f4e0c30cSAl Viro inode->i_state |= I_LINKABLE; 3142f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 3143f4e0c30cSAl Viro } 314460545d0dSAl Viro out2: 314560545d0dSAl Viro mnt_drop_write(nd->path.mnt); 314660545d0dSAl Viro out: 314760545d0dSAl Viro path_put(&nd->path); 314860545d0dSAl Viro return error; 314960545d0dSAl Viro } 315060545d0dSAl Viro 3151669abf4eSJeff Layton static struct file *path_openat(int dfd, struct filename *pathname, 315273d049a4SAl Viro struct nameidata *nd, const struct open_flags *op, int flags) 31531da177e4SLinus Torvalds { 3154fe2d35ffSAl Viro struct file *base = NULL; 315530d90494SAl Viro struct file *file; 31569850c056SAl Viro struct path path; 315747237687SAl Viro int opened = 0; 315813aab428SAl Viro int error; 315931e6b01fSNick Piggin 316030d90494SAl Viro file = get_empty_filp(); 31611afc99beSAl Viro if (IS_ERR(file)) 31621afc99beSAl Viro return file; 316331e6b01fSNick Piggin 316430d90494SAl Viro file->f_flags = op->open_flag; 316531e6b01fSNick Piggin 3166bb458c64SAl Viro if (unlikely(file->f_flags & __O_TMPFILE)) { 316760545d0dSAl Viro error = do_tmpfile(dfd, pathname, nd, flags, op, file, &opened); 316860545d0dSAl Viro goto out; 316960545d0dSAl Viro } 317060545d0dSAl Viro 3171669abf4eSJeff Layton error = path_init(dfd, pathname->name, flags | LOOKUP_PARENT, nd, &base); 317231e6b01fSNick Piggin if (unlikely(error)) 31732675a4ebSAl Viro goto out; 317431e6b01fSNick Piggin 3175fe2d35ffSAl Viro current->total_link_count = 0; 3176669abf4eSJeff Layton error = link_path_walk(pathname->name, nd); 317731e6b01fSNick Piggin if (unlikely(error)) 31782675a4ebSAl Viro goto out; 31791da177e4SLinus Torvalds 31802675a4ebSAl Viro error = do_last(nd, &path, file, op, &opened, pathname); 31812675a4ebSAl Viro while (unlikely(error > 0)) { /* trailing symlink */ 31827b9337aaSNick Piggin struct path link = path; 3183def4af30SAl Viro void *cookie; 3184574197e0SAl Viro if (!(nd->flags & LOOKUP_FOLLOW)) { 318573d049a4SAl Viro path_put_conditional(&path, nd); 318673d049a4SAl Viro path_put(&nd->path); 31872675a4ebSAl Viro error = -ELOOP; 318840b39136SAl Viro break; 318940b39136SAl Viro } 3190800179c9SKees Cook error = may_follow_link(&link, nd); 3191800179c9SKees Cook if (unlikely(error)) 3192800179c9SKees Cook break; 319373d049a4SAl Viro nd->flags |= LOOKUP_PARENT; 319473d049a4SAl Viro nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); 3195574197e0SAl Viro error = follow_link(&link, nd, &cookie); 3196c3e380b0SAl Viro if (unlikely(error)) 31972675a4ebSAl Viro break; 31982675a4ebSAl Viro error = do_last(nd, &path, file, op, &opened, pathname); 3199574197e0SAl Viro put_link(nd, &link, cookie); 3200806b681cSAl Viro } 320110fa8e62SAl Viro out: 320273d049a4SAl Viro if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) 320373d049a4SAl Viro path_put(&nd->root); 3204fe2d35ffSAl Viro if (base) 3205fe2d35ffSAl Viro fput(base); 32062675a4ebSAl Viro if (!(opened & FILE_OPENED)) { 32072675a4ebSAl Viro BUG_ON(!error); 320830d90494SAl Viro put_filp(file); 3209015c3bbcSMiklos Szeredi } 32102675a4ebSAl Viro if (unlikely(error)) { 32112675a4ebSAl Viro if (error == -EOPENSTALE) { 32122675a4ebSAl Viro if (flags & LOOKUP_RCU) 32132675a4ebSAl Viro error = -ECHILD; 32142675a4ebSAl Viro else 32152675a4ebSAl Viro error = -ESTALE; 32162675a4ebSAl Viro } 32172675a4ebSAl Viro file = ERR_PTR(error); 32182675a4ebSAl Viro } 32192675a4ebSAl Viro return file; 3220de459215SKirill Korotaev } 32211da177e4SLinus Torvalds 3222669abf4eSJeff Layton struct file *do_filp_open(int dfd, struct filename *pathname, 3223f9652e10SAl Viro const struct open_flags *op) 322413aab428SAl Viro { 322573d049a4SAl Viro struct nameidata nd; 3226f9652e10SAl Viro int flags = op->lookup_flags; 322713aab428SAl Viro struct file *filp; 322813aab428SAl Viro 322973d049a4SAl Viro filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU); 323013aab428SAl Viro if (unlikely(filp == ERR_PTR(-ECHILD))) 323173d049a4SAl Viro filp = path_openat(dfd, pathname, &nd, op, flags); 323213aab428SAl Viro if (unlikely(filp == ERR_PTR(-ESTALE))) 323373d049a4SAl Viro filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL); 323413aab428SAl Viro return filp; 323513aab428SAl Viro } 323613aab428SAl Viro 323773d049a4SAl Viro struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, 3238f9652e10SAl Viro const char *name, const struct open_flags *op) 323973d049a4SAl Viro { 324073d049a4SAl Viro struct nameidata nd; 324173d049a4SAl Viro struct file *file; 3242669abf4eSJeff Layton struct filename filename = { .name = name }; 3243f9652e10SAl Viro int flags = op->lookup_flags | LOOKUP_ROOT; 324473d049a4SAl Viro 324573d049a4SAl Viro nd.root.mnt = mnt; 324673d049a4SAl Viro nd.root.dentry = dentry; 324773d049a4SAl Viro 3248b18825a7SDavid Howells if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN) 324973d049a4SAl Viro return ERR_PTR(-ELOOP); 325073d049a4SAl Viro 3251669abf4eSJeff Layton file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_RCU); 325273d049a4SAl Viro if (unlikely(file == ERR_PTR(-ECHILD))) 3253669abf4eSJeff Layton file = path_openat(-1, &filename, &nd, op, flags); 325473d049a4SAl Viro if (unlikely(file == ERR_PTR(-ESTALE))) 3255669abf4eSJeff Layton file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_REVAL); 325673d049a4SAl Viro return file; 325773d049a4SAl Viro } 325873d049a4SAl Viro 32591ac12b4bSJeff Layton struct dentry *kern_path_create(int dfd, const char *pathname, 32601ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 32611da177e4SLinus Torvalds { 3262c663e5d8SChristoph Hellwig struct dentry *dentry = ERR_PTR(-EEXIST); 3263ed75e95dSAl Viro struct nameidata nd; 3264c30dabfeSJan Kara int err2; 32651ac12b4bSJeff Layton int error; 32661ac12b4bSJeff Layton bool is_dir = (lookup_flags & LOOKUP_DIRECTORY); 32671ac12b4bSJeff Layton 32681ac12b4bSJeff Layton /* 32691ac12b4bSJeff Layton * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any 32701ac12b4bSJeff Layton * other flags passed in are ignored! 32711ac12b4bSJeff Layton */ 32721ac12b4bSJeff Layton lookup_flags &= LOOKUP_REVAL; 32731ac12b4bSJeff Layton 32741ac12b4bSJeff Layton error = do_path_lookup(dfd, pathname, LOOKUP_PARENT|lookup_flags, &nd); 3275ed75e95dSAl Viro if (error) 3276ed75e95dSAl Viro return ERR_PTR(error); 32771da177e4SLinus Torvalds 3278c663e5d8SChristoph Hellwig /* 3279c663e5d8SChristoph Hellwig * Yucky last component or no last component at all? 3280c663e5d8SChristoph Hellwig * (foo/., foo/.., /////) 3281c663e5d8SChristoph Hellwig */ 3282ed75e95dSAl Viro if (nd.last_type != LAST_NORM) 3283ed75e95dSAl Viro goto out; 3284ed75e95dSAl Viro nd.flags &= ~LOOKUP_PARENT; 3285ed75e95dSAl Viro nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL; 3286c663e5d8SChristoph Hellwig 3287c30dabfeSJan Kara /* don't fail immediately if it's r/o, at least try to report other errors */ 3288c30dabfeSJan Kara err2 = mnt_want_write(nd.path.mnt); 3289c663e5d8SChristoph Hellwig /* 3290c663e5d8SChristoph Hellwig * Do the final lookup. 3291c663e5d8SChristoph Hellwig */ 3292ed75e95dSAl Viro mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 3293ed75e95dSAl Viro dentry = lookup_hash(&nd); 32941da177e4SLinus Torvalds if (IS_ERR(dentry)) 3295a8104a9fSAl Viro goto unlock; 3296c663e5d8SChristoph Hellwig 3297a8104a9fSAl Viro error = -EEXIST; 3298b18825a7SDavid Howells if (d_is_positive(dentry)) 3299a8104a9fSAl Viro goto fail; 3300b18825a7SDavid Howells 3301c663e5d8SChristoph Hellwig /* 3302c663e5d8SChristoph Hellwig * Special case - lookup gave negative, but... we had foo/bar/ 3303c663e5d8SChristoph Hellwig * From the vfs_mknod() POV we just have a negative dentry - 3304c663e5d8SChristoph Hellwig * all is fine. Let's be bastards - you had / on the end, you've 3305c663e5d8SChristoph Hellwig * been asking for (non-existent) directory. -ENOENT for you. 3306c663e5d8SChristoph Hellwig */ 3307ed75e95dSAl Viro if (unlikely(!is_dir && nd.last.name[nd.last.len])) { 3308a8104a9fSAl Viro error = -ENOENT; 3309ed75e95dSAl Viro goto fail; 3310e9baf6e5SAl Viro } 3311c30dabfeSJan Kara if (unlikely(err2)) { 3312c30dabfeSJan Kara error = err2; 3313a8104a9fSAl Viro goto fail; 3314c30dabfeSJan Kara } 3315ed75e95dSAl Viro *path = nd.path; 3316e9baf6e5SAl Viro return dentry; 33171da177e4SLinus Torvalds fail: 3318a8104a9fSAl Viro dput(dentry); 3319a8104a9fSAl Viro dentry = ERR_PTR(error); 3320a8104a9fSAl Viro unlock: 3321dae6ad8fSAl Viro mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 3322c30dabfeSJan Kara if (!err2) 3323c30dabfeSJan Kara mnt_drop_write(nd.path.mnt); 3324ed75e95dSAl Viro out: 3325dae6ad8fSAl Viro path_put(&nd.path); 3326ed75e95dSAl Viro return dentry; 3327dae6ad8fSAl Viro } 3328dae6ad8fSAl Viro EXPORT_SYMBOL(kern_path_create); 3329dae6ad8fSAl Viro 3330921a1650SAl Viro void done_path_create(struct path *path, struct dentry *dentry) 3331921a1650SAl Viro { 3332921a1650SAl Viro dput(dentry); 3333921a1650SAl Viro mutex_unlock(&path->dentry->d_inode->i_mutex); 3334a8104a9fSAl Viro mnt_drop_write(path->mnt); 3335921a1650SAl Viro path_put(path); 3336921a1650SAl Viro } 3337921a1650SAl Viro EXPORT_SYMBOL(done_path_create); 3338921a1650SAl Viro 33391ac12b4bSJeff Layton struct dentry *user_path_create(int dfd, const char __user *pathname, 33401ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 3341dae6ad8fSAl Viro { 334291a27b2aSJeff Layton struct filename *tmp = getname(pathname); 3343dae6ad8fSAl Viro struct dentry *res; 3344dae6ad8fSAl Viro if (IS_ERR(tmp)) 3345dae6ad8fSAl Viro return ERR_CAST(tmp); 33461ac12b4bSJeff Layton res = kern_path_create(dfd, tmp->name, path, lookup_flags); 3347dae6ad8fSAl Viro putname(tmp); 3348dae6ad8fSAl Viro return res; 3349dae6ad8fSAl Viro } 3350dae6ad8fSAl Viro EXPORT_SYMBOL(user_path_create); 3351dae6ad8fSAl Viro 33521a67aafbSAl Viro int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) 33531da177e4SLinus Torvalds { 3354a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 33551da177e4SLinus Torvalds 33561da177e4SLinus Torvalds if (error) 33571da177e4SLinus Torvalds return error; 33581da177e4SLinus Torvalds 3359975d6b39SEric W. Biederman if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) 33601da177e4SLinus Torvalds return -EPERM; 33611da177e4SLinus Torvalds 3362acfa4380SAl Viro if (!dir->i_op->mknod) 33631da177e4SLinus Torvalds return -EPERM; 33641da177e4SLinus Torvalds 336508ce5f16SSerge E. Hallyn error = devcgroup_inode_mknod(mode, dev); 336608ce5f16SSerge E. Hallyn if (error) 336708ce5f16SSerge E. Hallyn return error; 336808ce5f16SSerge E. Hallyn 33691da177e4SLinus Torvalds error = security_inode_mknod(dir, dentry, mode, dev); 33701da177e4SLinus Torvalds if (error) 33711da177e4SLinus Torvalds return error; 33721da177e4SLinus Torvalds 33731da177e4SLinus Torvalds error = dir->i_op->mknod(dir, dentry, mode, dev); 3374a74574aaSStephen Smalley if (!error) 3375f38aa942SAmy Griffis fsnotify_create(dir, dentry); 33761da177e4SLinus Torvalds return error; 33771da177e4SLinus Torvalds } 33781da177e4SLinus Torvalds 3379f69aac00SAl Viro static int may_mknod(umode_t mode) 3380463c3197SDave Hansen { 3381463c3197SDave Hansen switch (mode & S_IFMT) { 3382463c3197SDave Hansen case S_IFREG: 3383463c3197SDave Hansen case S_IFCHR: 3384463c3197SDave Hansen case S_IFBLK: 3385463c3197SDave Hansen case S_IFIFO: 3386463c3197SDave Hansen case S_IFSOCK: 3387463c3197SDave Hansen case 0: /* zero mode translates to S_IFREG */ 3388463c3197SDave Hansen return 0; 3389463c3197SDave Hansen case S_IFDIR: 3390463c3197SDave Hansen return -EPERM; 3391463c3197SDave Hansen default: 3392463c3197SDave Hansen return -EINVAL; 3393463c3197SDave Hansen } 3394463c3197SDave Hansen } 3395463c3197SDave Hansen 33968208a22bSAl Viro SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, 33972e4d0924SHeiko Carstens unsigned, dev) 33981da177e4SLinus Torvalds { 33991da177e4SLinus Torvalds struct dentry *dentry; 3400dae6ad8fSAl Viro struct path path; 3401dae6ad8fSAl Viro int error; 3402972567f1SJeff Layton unsigned int lookup_flags = 0; 34031da177e4SLinus Torvalds 34048e4bfca1SAl Viro error = may_mknod(mode); 34058e4bfca1SAl Viro if (error) 34068e4bfca1SAl Viro return error; 3407972567f1SJeff Layton retry: 3408972567f1SJeff Layton dentry = user_path_create(dfd, filename, &path, lookup_flags); 3409dae6ad8fSAl Viro if (IS_ERR(dentry)) 3410dae6ad8fSAl Viro return PTR_ERR(dentry); 34112ad94ae6SAl Viro 3412dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3413ce3b0f8dSAl Viro mode &= ~current_umask(); 3414dae6ad8fSAl Viro error = security_path_mknod(&path, dentry, mode, dev); 3415be6d3e56SKentaro Takeda if (error) 3416a8104a9fSAl Viro goto out; 34171da177e4SLinus Torvalds switch (mode & S_IFMT) { 34181da177e4SLinus Torvalds case 0: case S_IFREG: 3419312b63fbSAl Viro error = vfs_create(path.dentry->d_inode,dentry,mode,true); 34201da177e4SLinus Torvalds break; 34211da177e4SLinus Torvalds case S_IFCHR: case S_IFBLK: 3422dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode, 34231da177e4SLinus Torvalds new_decode_dev(dev)); 34241da177e4SLinus Torvalds break; 34251da177e4SLinus Torvalds case S_IFIFO: case S_IFSOCK: 3426dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); 34271da177e4SLinus Torvalds break; 34281da177e4SLinus Torvalds } 3429a8104a9fSAl Viro out: 3430921a1650SAl Viro done_path_create(&path, dentry); 3431972567f1SJeff Layton if (retry_estale(error, lookup_flags)) { 3432972567f1SJeff Layton lookup_flags |= LOOKUP_REVAL; 3433972567f1SJeff Layton goto retry; 3434972567f1SJeff Layton } 34351da177e4SLinus Torvalds return error; 34361da177e4SLinus Torvalds } 34371da177e4SLinus Torvalds 34388208a22bSAl Viro SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) 34395590ff0dSUlrich Drepper { 34405590ff0dSUlrich Drepper return sys_mknodat(AT_FDCWD, filename, mode, dev); 34415590ff0dSUlrich Drepper } 34425590ff0dSUlrich Drepper 344318bb1db3SAl Viro int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 34441da177e4SLinus Torvalds { 3445a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 34468de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 34471da177e4SLinus Torvalds 34481da177e4SLinus Torvalds if (error) 34491da177e4SLinus Torvalds return error; 34501da177e4SLinus Torvalds 3451acfa4380SAl Viro if (!dir->i_op->mkdir) 34521da177e4SLinus Torvalds return -EPERM; 34531da177e4SLinus Torvalds 34541da177e4SLinus Torvalds mode &= (S_IRWXUGO|S_ISVTX); 34551da177e4SLinus Torvalds error = security_inode_mkdir(dir, dentry, mode); 34561da177e4SLinus Torvalds if (error) 34571da177e4SLinus Torvalds return error; 34581da177e4SLinus Torvalds 34598de52778SAl Viro if (max_links && dir->i_nlink >= max_links) 34608de52778SAl Viro return -EMLINK; 34618de52778SAl Viro 34621da177e4SLinus Torvalds error = dir->i_op->mkdir(dir, dentry, mode); 3463a74574aaSStephen Smalley if (!error) 3464f38aa942SAmy Griffis fsnotify_mkdir(dir, dentry); 34651da177e4SLinus Torvalds return error; 34661da177e4SLinus Torvalds } 34671da177e4SLinus Torvalds 3468a218d0fdSAl Viro SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) 34691da177e4SLinus Torvalds { 34706902d925SDave Hansen struct dentry *dentry; 3471dae6ad8fSAl Viro struct path path; 3472dae6ad8fSAl Viro int error; 3473b76d8b82SJeff Layton unsigned int lookup_flags = LOOKUP_DIRECTORY; 34741da177e4SLinus Torvalds 3475b76d8b82SJeff Layton retry: 3476b76d8b82SJeff Layton dentry = user_path_create(dfd, pathname, &path, lookup_flags); 34776902d925SDave Hansen if (IS_ERR(dentry)) 3478dae6ad8fSAl Viro return PTR_ERR(dentry); 34796902d925SDave Hansen 3480dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3481ce3b0f8dSAl Viro mode &= ~current_umask(); 3482dae6ad8fSAl Viro error = security_path_mkdir(&path, dentry, mode); 3483a8104a9fSAl Viro if (!error) 3484dae6ad8fSAl Viro error = vfs_mkdir(path.dentry->d_inode, dentry, mode); 3485921a1650SAl Viro done_path_create(&path, dentry); 3486b76d8b82SJeff Layton if (retry_estale(error, lookup_flags)) { 3487b76d8b82SJeff Layton lookup_flags |= LOOKUP_REVAL; 3488b76d8b82SJeff Layton goto retry; 3489b76d8b82SJeff Layton } 34901da177e4SLinus Torvalds return error; 34911da177e4SLinus Torvalds } 34921da177e4SLinus Torvalds 3493a218d0fdSAl Viro SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) 34945590ff0dSUlrich Drepper { 34955590ff0dSUlrich Drepper return sys_mkdirat(AT_FDCWD, pathname, mode); 34965590ff0dSUlrich Drepper } 34975590ff0dSUlrich Drepper 34981da177e4SLinus Torvalds /* 3499a71905f0SSage Weil * The dentry_unhash() helper will try to drop the dentry early: we 3500c0d02594SJ. Bruce Fields * should have a usage count of 1 if we're the only user of this 3501a71905f0SSage Weil * dentry, and if that is true (possibly after pruning the dcache), 3502a71905f0SSage Weil * then we drop the dentry now. 35031da177e4SLinus Torvalds * 35041da177e4SLinus Torvalds * A low-level filesystem can, if it choses, legally 35051da177e4SLinus Torvalds * do a 35061da177e4SLinus Torvalds * 35071da177e4SLinus Torvalds * if (!d_unhashed(dentry)) 35081da177e4SLinus Torvalds * return -EBUSY; 35091da177e4SLinus Torvalds * 35101da177e4SLinus Torvalds * if it cannot handle the case of removing a directory 35111da177e4SLinus Torvalds * that is still in use by something else.. 35121da177e4SLinus Torvalds */ 35131da177e4SLinus Torvalds void dentry_unhash(struct dentry *dentry) 35141da177e4SLinus Torvalds { 35151da177e4SLinus Torvalds shrink_dcache_parent(dentry); 35161da177e4SLinus Torvalds spin_lock(&dentry->d_lock); 351798474236SWaiman Long if (dentry->d_lockref.count == 1) 35181da177e4SLinus Torvalds __d_drop(dentry); 35191da177e4SLinus Torvalds spin_unlock(&dentry->d_lock); 35201da177e4SLinus Torvalds } 35211da177e4SLinus Torvalds 35221da177e4SLinus Torvalds int vfs_rmdir(struct inode *dir, struct dentry *dentry) 35231da177e4SLinus Torvalds { 35241da177e4SLinus Torvalds int error = may_delete(dir, dentry, 1); 35251da177e4SLinus Torvalds 35261da177e4SLinus Torvalds if (error) 35271da177e4SLinus Torvalds return error; 35281da177e4SLinus Torvalds 3529acfa4380SAl Viro if (!dir->i_op->rmdir) 35301da177e4SLinus Torvalds return -EPERM; 35311da177e4SLinus Torvalds 35321d2ef590SAl Viro dget(dentry); 35331b1dcc1bSJes Sorensen mutex_lock(&dentry->d_inode->i_mutex); 3534912dbc15SSage Weil 35351da177e4SLinus Torvalds error = -EBUSY; 3536912dbc15SSage Weil if (d_mountpoint(dentry)) 3537912dbc15SSage Weil goto out; 3538912dbc15SSage Weil 35391da177e4SLinus Torvalds error = security_inode_rmdir(dir, dentry); 3540912dbc15SSage Weil if (error) 3541912dbc15SSage Weil goto out; 3542912dbc15SSage Weil 35433cebde24SSage Weil shrink_dcache_parent(dentry); 35441da177e4SLinus Torvalds error = dir->i_op->rmdir(dir, dentry); 3545912dbc15SSage Weil if (error) 3546912dbc15SSage Weil goto out; 3547912dbc15SSage Weil 35481da177e4SLinus Torvalds dentry->d_inode->i_flags |= S_DEAD; 3549d83c49f3SAl Viro dont_mount(dentry); 35501da177e4SLinus Torvalds 3551912dbc15SSage Weil out: 3552912dbc15SSage Weil mutex_unlock(&dentry->d_inode->i_mutex); 35531d2ef590SAl Viro dput(dentry); 3554912dbc15SSage Weil if (!error) 3555912dbc15SSage Weil d_delete(dentry); 35561da177e4SLinus Torvalds return error; 35571da177e4SLinus Torvalds } 35581da177e4SLinus Torvalds 35595590ff0dSUlrich Drepper static long do_rmdir(int dfd, const char __user *pathname) 35601da177e4SLinus Torvalds { 35611da177e4SLinus Torvalds int error = 0; 356291a27b2aSJeff Layton struct filename *name; 35631da177e4SLinus Torvalds struct dentry *dentry; 35641da177e4SLinus Torvalds struct nameidata nd; 3565c6ee9206SJeff Layton unsigned int lookup_flags = 0; 3566c6ee9206SJeff Layton retry: 3567c6ee9206SJeff Layton name = user_path_parent(dfd, pathname, &nd, lookup_flags); 356891a27b2aSJeff Layton if (IS_ERR(name)) 356991a27b2aSJeff Layton return PTR_ERR(name); 35701da177e4SLinus Torvalds 35711da177e4SLinus Torvalds switch(nd.last_type) { 35721da177e4SLinus Torvalds case LAST_DOTDOT: 35731da177e4SLinus Torvalds error = -ENOTEMPTY; 35741da177e4SLinus Torvalds goto exit1; 35751da177e4SLinus Torvalds case LAST_DOT: 35761da177e4SLinus Torvalds error = -EINVAL; 35771da177e4SLinus Torvalds goto exit1; 35781da177e4SLinus Torvalds case LAST_ROOT: 35791da177e4SLinus Torvalds error = -EBUSY; 35801da177e4SLinus Torvalds goto exit1; 35811da177e4SLinus Torvalds } 35820612d9fbSOGAWA Hirofumi 35830612d9fbSOGAWA Hirofumi nd.flags &= ~LOOKUP_PARENT; 3584c30dabfeSJan Kara error = mnt_want_write(nd.path.mnt); 3585c30dabfeSJan Kara if (error) 3586c30dabfeSJan Kara goto exit1; 35870612d9fbSOGAWA Hirofumi 35884ac91378SJan Blunck mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 358949705b77SChristoph Hellwig dentry = lookup_hash(&nd); 35901da177e4SLinus Torvalds error = PTR_ERR(dentry); 35916902d925SDave Hansen if (IS_ERR(dentry)) 35926902d925SDave Hansen goto exit2; 3593e6bc45d6STheodore Ts'o if (!dentry->d_inode) { 3594e6bc45d6STheodore Ts'o error = -ENOENT; 3595e6bc45d6STheodore Ts'o goto exit3; 3596e6bc45d6STheodore Ts'o } 3597be6d3e56SKentaro Takeda error = security_path_rmdir(&nd.path, dentry); 3598be6d3e56SKentaro Takeda if (error) 3599c30dabfeSJan Kara goto exit3; 36004ac91378SJan Blunck error = vfs_rmdir(nd.path.dentry->d_inode, dentry); 36010622753bSDave Hansen exit3: 36021da177e4SLinus Torvalds dput(dentry); 36036902d925SDave Hansen exit2: 36044ac91378SJan Blunck mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 3605c30dabfeSJan Kara mnt_drop_write(nd.path.mnt); 36061da177e4SLinus Torvalds exit1: 36071d957f9bSJan Blunck path_put(&nd.path); 36081da177e4SLinus Torvalds putname(name); 3609c6ee9206SJeff Layton if (retry_estale(error, lookup_flags)) { 3610c6ee9206SJeff Layton lookup_flags |= LOOKUP_REVAL; 3611c6ee9206SJeff Layton goto retry; 3612c6ee9206SJeff Layton } 36131da177e4SLinus Torvalds return error; 36141da177e4SLinus Torvalds } 36151da177e4SLinus Torvalds 36163cdad428SHeiko Carstens SYSCALL_DEFINE1(rmdir, const char __user *, pathname) 36175590ff0dSUlrich Drepper { 36185590ff0dSUlrich Drepper return do_rmdir(AT_FDCWD, pathname); 36195590ff0dSUlrich Drepper } 36205590ff0dSUlrich Drepper 3621b21996e3SJ. Bruce Fields /** 3622b21996e3SJ. Bruce Fields * vfs_unlink - unlink a filesystem object 3623b21996e3SJ. Bruce Fields * @dir: parent directory 3624b21996e3SJ. Bruce Fields * @dentry: victim 3625b21996e3SJ. Bruce Fields * @delegated_inode: returns victim inode, if the inode is delegated. 3626b21996e3SJ. Bruce Fields * 3627b21996e3SJ. Bruce Fields * The caller must hold dir->i_mutex. 3628b21996e3SJ. Bruce Fields * 3629b21996e3SJ. Bruce Fields * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and 3630b21996e3SJ. Bruce Fields * return a reference to the inode in delegated_inode. The caller 3631b21996e3SJ. Bruce Fields * should then break the delegation on that inode and retry. Because 3632b21996e3SJ. Bruce Fields * breaking a delegation may take a long time, the caller should drop 3633b21996e3SJ. Bruce Fields * dir->i_mutex before doing so. 3634b21996e3SJ. Bruce Fields * 3635b21996e3SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 3636b21996e3SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 3637b21996e3SJ. Bruce Fields * to be NFS exported. 3638b21996e3SJ. Bruce Fields */ 3639b21996e3SJ. Bruce Fields int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode) 36401da177e4SLinus Torvalds { 36419accbb97SJ. Bruce Fields struct inode *target = dentry->d_inode; 36421da177e4SLinus Torvalds int error = may_delete(dir, dentry, 0); 36431da177e4SLinus Torvalds 36441da177e4SLinus Torvalds if (error) 36451da177e4SLinus Torvalds return error; 36461da177e4SLinus Torvalds 3647acfa4380SAl Viro if (!dir->i_op->unlink) 36481da177e4SLinus Torvalds return -EPERM; 36491da177e4SLinus Torvalds 36509accbb97SJ. Bruce Fields mutex_lock(&target->i_mutex); 36511da177e4SLinus Torvalds if (d_mountpoint(dentry)) 36521da177e4SLinus Torvalds error = -EBUSY; 36531da177e4SLinus Torvalds else { 36541da177e4SLinus Torvalds error = security_inode_unlink(dir, dentry); 3655bec1052eSAl Viro if (!error) { 36565a14696cSJ. Bruce Fields error = try_break_deleg(target, delegated_inode); 36575a14696cSJ. Bruce Fields if (error) 3658b21996e3SJ. Bruce Fields goto out; 36591da177e4SLinus Torvalds error = dir->i_op->unlink(dir, dentry); 3660bec1052eSAl Viro if (!error) 3661d83c49f3SAl Viro dont_mount(dentry); 3662bec1052eSAl Viro } 36631da177e4SLinus Torvalds } 3664b21996e3SJ. Bruce Fields out: 36659accbb97SJ. Bruce Fields mutex_unlock(&target->i_mutex); 36661da177e4SLinus Torvalds 36671da177e4SLinus Torvalds /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 36681da177e4SLinus Torvalds if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { 36699accbb97SJ. Bruce Fields fsnotify_link_count(target); 36701da177e4SLinus Torvalds d_delete(dentry); 36711da177e4SLinus Torvalds } 36720eeca283SRobert Love 36731da177e4SLinus Torvalds return error; 36741da177e4SLinus Torvalds } 36751da177e4SLinus Torvalds 36761da177e4SLinus Torvalds /* 36771da177e4SLinus Torvalds * Make sure that the actual truncation of the file will occur outside its 36781b1dcc1bSJes Sorensen * directory's i_mutex. Truncate can take a long time if there is a lot of 36791da177e4SLinus Torvalds * writeout happening, and we don't want to prevent access to the directory 36801da177e4SLinus Torvalds * while waiting on the I/O. 36811da177e4SLinus Torvalds */ 36825590ff0dSUlrich Drepper static long do_unlinkat(int dfd, const char __user *pathname) 36831da177e4SLinus Torvalds { 36842ad94ae6SAl Viro int error; 368591a27b2aSJeff Layton struct filename *name; 36861da177e4SLinus Torvalds struct dentry *dentry; 36871da177e4SLinus Torvalds struct nameidata nd; 36881da177e4SLinus Torvalds struct inode *inode = NULL; 3689b21996e3SJ. Bruce Fields struct inode *delegated_inode = NULL; 36905d18f813SJeff Layton unsigned int lookup_flags = 0; 36915d18f813SJeff Layton retry: 36925d18f813SJeff Layton name = user_path_parent(dfd, pathname, &nd, lookup_flags); 369391a27b2aSJeff Layton if (IS_ERR(name)) 369491a27b2aSJeff Layton return PTR_ERR(name); 36952ad94ae6SAl Viro 36961da177e4SLinus Torvalds error = -EISDIR; 36971da177e4SLinus Torvalds if (nd.last_type != LAST_NORM) 36981da177e4SLinus Torvalds goto exit1; 36990612d9fbSOGAWA Hirofumi 37000612d9fbSOGAWA Hirofumi nd.flags &= ~LOOKUP_PARENT; 3701c30dabfeSJan Kara error = mnt_want_write(nd.path.mnt); 3702c30dabfeSJan Kara if (error) 3703c30dabfeSJan Kara goto exit1; 3704b21996e3SJ. Bruce Fields retry_deleg: 37054ac91378SJan Blunck mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 370649705b77SChristoph Hellwig dentry = lookup_hash(&nd); 37071da177e4SLinus Torvalds error = PTR_ERR(dentry); 37081da177e4SLinus Torvalds if (!IS_ERR(dentry)) { 37091da177e4SLinus Torvalds /* Why not before? Because we want correct error value */ 371050338b88STörök Edwin if (nd.last.name[nd.last.len]) 371150338b88STörök Edwin goto slashes; 37121da177e4SLinus Torvalds inode = dentry->d_inode; 3713b18825a7SDavid Howells if (d_is_negative(dentry)) 3714e6bc45d6STheodore Ts'o goto slashes; 37157de9c6eeSAl Viro ihold(inode); 3716be6d3e56SKentaro Takeda error = security_path_unlink(&nd.path, dentry); 3717be6d3e56SKentaro Takeda if (error) 3718c30dabfeSJan Kara goto exit2; 3719b21996e3SJ. Bruce Fields error = vfs_unlink(nd.path.dentry->d_inode, dentry, &delegated_inode); 37201da177e4SLinus Torvalds exit2: 37211da177e4SLinus Torvalds dput(dentry); 37221da177e4SLinus Torvalds } 37234ac91378SJan Blunck mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 37241da177e4SLinus Torvalds if (inode) 37251da177e4SLinus Torvalds iput(inode); /* truncate the inode here */ 3726b21996e3SJ. Bruce Fields inode = NULL; 3727b21996e3SJ. Bruce Fields if (delegated_inode) { 37285a14696cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 3729b21996e3SJ. Bruce Fields if (!error) 3730b21996e3SJ. Bruce Fields goto retry_deleg; 3731b21996e3SJ. Bruce Fields } 3732c30dabfeSJan Kara mnt_drop_write(nd.path.mnt); 37331da177e4SLinus Torvalds exit1: 37341d957f9bSJan Blunck path_put(&nd.path); 37351da177e4SLinus Torvalds putname(name); 37365d18f813SJeff Layton if (retry_estale(error, lookup_flags)) { 37375d18f813SJeff Layton lookup_flags |= LOOKUP_REVAL; 37385d18f813SJeff Layton inode = NULL; 37395d18f813SJeff Layton goto retry; 37405d18f813SJeff Layton } 37411da177e4SLinus Torvalds return error; 37421da177e4SLinus Torvalds 37431da177e4SLinus Torvalds slashes: 3744b18825a7SDavid Howells if (d_is_negative(dentry)) 3745b18825a7SDavid Howells error = -ENOENT; 374644b1d530SMiklos Szeredi else if (d_is_dir(dentry)) 3747b18825a7SDavid Howells error = -EISDIR; 3748b18825a7SDavid Howells else 3749b18825a7SDavid Howells error = -ENOTDIR; 37501da177e4SLinus Torvalds goto exit2; 37511da177e4SLinus Torvalds } 37521da177e4SLinus Torvalds 37532e4d0924SHeiko Carstens SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) 37545590ff0dSUlrich Drepper { 37555590ff0dSUlrich Drepper if ((flag & ~AT_REMOVEDIR) != 0) 37565590ff0dSUlrich Drepper return -EINVAL; 37575590ff0dSUlrich Drepper 37585590ff0dSUlrich Drepper if (flag & AT_REMOVEDIR) 37595590ff0dSUlrich Drepper return do_rmdir(dfd, pathname); 37605590ff0dSUlrich Drepper 37615590ff0dSUlrich Drepper return do_unlinkat(dfd, pathname); 37625590ff0dSUlrich Drepper } 37635590ff0dSUlrich Drepper 37643480b257SHeiko Carstens SYSCALL_DEFINE1(unlink, const char __user *, pathname) 37655590ff0dSUlrich Drepper { 37665590ff0dSUlrich Drepper return do_unlinkat(AT_FDCWD, pathname); 37675590ff0dSUlrich Drepper } 37685590ff0dSUlrich Drepper 3769db2e747bSMiklos Szeredi int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) 37701da177e4SLinus Torvalds { 3771a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 37721da177e4SLinus Torvalds 37731da177e4SLinus Torvalds if (error) 37741da177e4SLinus Torvalds return error; 37751da177e4SLinus Torvalds 3776acfa4380SAl Viro if (!dir->i_op->symlink) 37771da177e4SLinus Torvalds return -EPERM; 37781da177e4SLinus Torvalds 37791da177e4SLinus Torvalds error = security_inode_symlink(dir, dentry, oldname); 37801da177e4SLinus Torvalds if (error) 37811da177e4SLinus Torvalds return error; 37821da177e4SLinus Torvalds 37831da177e4SLinus Torvalds error = dir->i_op->symlink(dir, dentry, oldname); 3784a74574aaSStephen Smalley if (!error) 3785f38aa942SAmy Griffis fsnotify_create(dir, dentry); 37861da177e4SLinus Torvalds return error; 37871da177e4SLinus Torvalds } 37881da177e4SLinus Torvalds 37892e4d0924SHeiko Carstens SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, 37902e4d0924SHeiko Carstens int, newdfd, const char __user *, newname) 37911da177e4SLinus Torvalds { 37922ad94ae6SAl Viro int error; 379391a27b2aSJeff Layton struct filename *from; 37946902d925SDave Hansen struct dentry *dentry; 3795dae6ad8fSAl Viro struct path path; 3796f46d3567SJeff Layton unsigned int lookup_flags = 0; 37971da177e4SLinus Torvalds 37981da177e4SLinus Torvalds from = getname(oldname); 37991da177e4SLinus Torvalds if (IS_ERR(from)) 38001da177e4SLinus Torvalds return PTR_ERR(from); 3801f46d3567SJeff Layton retry: 3802f46d3567SJeff Layton dentry = user_path_create(newdfd, newname, &path, lookup_flags); 38031da177e4SLinus Torvalds error = PTR_ERR(dentry); 38046902d925SDave Hansen if (IS_ERR(dentry)) 3805dae6ad8fSAl Viro goto out_putname; 38066902d925SDave Hansen 380791a27b2aSJeff Layton error = security_path_symlink(&path, dentry, from->name); 3808a8104a9fSAl Viro if (!error) 380991a27b2aSJeff Layton error = vfs_symlink(path.dentry->d_inode, dentry, from->name); 3810921a1650SAl Viro done_path_create(&path, dentry); 3811f46d3567SJeff Layton if (retry_estale(error, lookup_flags)) { 3812f46d3567SJeff Layton lookup_flags |= LOOKUP_REVAL; 3813f46d3567SJeff Layton goto retry; 3814f46d3567SJeff Layton } 38156902d925SDave Hansen out_putname: 38161da177e4SLinus Torvalds putname(from); 38171da177e4SLinus Torvalds return error; 38181da177e4SLinus Torvalds } 38191da177e4SLinus Torvalds 38203480b257SHeiko Carstens SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) 38215590ff0dSUlrich Drepper { 38225590ff0dSUlrich Drepper return sys_symlinkat(oldname, AT_FDCWD, newname); 38235590ff0dSUlrich Drepper } 38245590ff0dSUlrich Drepper 3825146a8595SJ. Bruce Fields /** 3826146a8595SJ. Bruce Fields * vfs_link - create a new link 3827146a8595SJ. Bruce Fields * @old_dentry: object to be linked 3828146a8595SJ. Bruce Fields * @dir: new parent 3829146a8595SJ. Bruce Fields * @new_dentry: where to create the new link 3830146a8595SJ. Bruce Fields * @delegated_inode: returns inode needing a delegation break 3831146a8595SJ. Bruce Fields * 3832146a8595SJ. Bruce Fields * The caller must hold dir->i_mutex 3833146a8595SJ. Bruce Fields * 3834146a8595SJ. Bruce Fields * If vfs_link discovers a delegation on the to-be-linked file in need 3835146a8595SJ. Bruce Fields * of breaking, it will return -EWOULDBLOCK and return a reference to the 3836146a8595SJ. Bruce Fields * inode in delegated_inode. The caller should then break the delegation 3837146a8595SJ. Bruce Fields * and retry. Because breaking a delegation may take a long time, the 3838146a8595SJ. Bruce Fields * caller should drop the i_mutex before doing so. 3839146a8595SJ. Bruce Fields * 3840146a8595SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 3841146a8595SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 3842146a8595SJ. Bruce Fields * to be NFS exported. 3843146a8595SJ. Bruce Fields */ 3844146a8595SJ. Bruce Fields int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode) 38451da177e4SLinus Torvalds { 38461da177e4SLinus Torvalds struct inode *inode = old_dentry->d_inode; 38478de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 38481da177e4SLinus Torvalds int error; 38491da177e4SLinus Torvalds 38501da177e4SLinus Torvalds if (!inode) 38511da177e4SLinus Torvalds return -ENOENT; 38521da177e4SLinus Torvalds 3853a95164d9SMiklos Szeredi error = may_create(dir, new_dentry); 38541da177e4SLinus Torvalds if (error) 38551da177e4SLinus Torvalds return error; 38561da177e4SLinus Torvalds 38571da177e4SLinus Torvalds if (dir->i_sb != inode->i_sb) 38581da177e4SLinus Torvalds return -EXDEV; 38591da177e4SLinus Torvalds 38601da177e4SLinus Torvalds /* 38611da177e4SLinus Torvalds * A link to an append-only or immutable file cannot be created. 38621da177e4SLinus Torvalds */ 38631da177e4SLinus Torvalds if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 38641da177e4SLinus Torvalds return -EPERM; 3865acfa4380SAl Viro if (!dir->i_op->link) 38661da177e4SLinus Torvalds return -EPERM; 38677e79eedbSTetsuo Handa if (S_ISDIR(inode->i_mode)) 38681da177e4SLinus Torvalds return -EPERM; 38691da177e4SLinus Torvalds 38701da177e4SLinus Torvalds error = security_inode_link(old_dentry, dir, new_dentry); 38711da177e4SLinus Torvalds if (error) 38721da177e4SLinus Torvalds return error; 38731da177e4SLinus Torvalds 38747e79eedbSTetsuo Handa mutex_lock(&inode->i_mutex); 3875aae8a97dSAneesh Kumar K.V /* Make sure we don't allow creating hardlink to an unlinked file */ 3876f4e0c30cSAl Viro if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) 3877aae8a97dSAneesh Kumar K.V error = -ENOENT; 38788de52778SAl Viro else if (max_links && inode->i_nlink >= max_links) 38798de52778SAl Viro error = -EMLINK; 3880146a8595SJ. Bruce Fields else { 3881146a8595SJ. Bruce Fields error = try_break_deleg(inode, delegated_inode); 3882146a8595SJ. Bruce Fields if (!error) 38831da177e4SLinus Torvalds error = dir->i_op->link(old_dentry, dir, new_dentry); 3884146a8595SJ. Bruce Fields } 3885f4e0c30cSAl Viro 3886f4e0c30cSAl Viro if (!error && (inode->i_state & I_LINKABLE)) { 3887f4e0c30cSAl Viro spin_lock(&inode->i_lock); 3888f4e0c30cSAl Viro inode->i_state &= ~I_LINKABLE; 3889f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 3890f4e0c30cSAl Viro } 38917e79eedbSTetsuo Handa mutex_unlock(&inode->i_mutex); 3892e31e14ecSStephen Smalley if (!error) 38937e79eedbSTetsuo Handa fsnotify_link(dir, inode, new_dentry); 38941da177e4SLinus Torvalds return error; 38951da177e4SLinus Torvalds } 38961da177e4SLinus Torvalds 38971da177e4SLinus Torvalds /* 38981da177e4SLinus Torvalds * Hardlinks are often used in delicate situations. We avoid 38991da177e4SLinus Torvalds * security-related surprises by not following symlinks on the 39001da177e4SLinus Torvalds * newname. --KAB 39011da177e4SLinus Torvalds * 39021da177e4SLinus Torvalds * We don't follow them on the oldname either to be compatible 39031da177e4SLinus Torvalds * with linux 2.0, and to avoid hard-linking to directories 39041da177e4SLinus Torvalds * and other special files. --ADM 39051da177e4SLinus Torvalds */ 39062e4d0924SHeiko Carstens SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, 39072e4d0924SHeiko Carstens int, newdfd, const char __user *, newname, int, flags) 39081da177e4SLinus Torvalds { 39091da177e4SLinus Torvalds struct dentry *new_dentry; 3910dae6ad8fSAl Viro struct path old_path, new_path; 3911146a8595SJ. Bruce Fields struct inode *delegated_inode = NULL; 391211a7b371SAneesh Kumar K.V int how = 0; 39131da177e4SLinus Torvalds int error; 39141da177e4SLinus Torvalds 391511a7b371SAneesh Kumar K.V if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) 3916c04030e1SUlrich Drepper return -EINVAL; 391711a7b371SAneesh Kumar K.V /* 3918f0cc6ffbSLinus Torvalds * To use null names we require CAP_DAC_READ_SEARCH 3919f0cc6ffbSLinus Torvalds * This ensures that not everyone will be able to create 3920f0cc6ffbSLinus Torvalds * handlink using the passed filedescriptor. 392111a7b371SAneesh Kumar K.V */ 3922f0cc6ffbSLinus Torvalds if (flags & AT_EMPTY_PATH) { 3923f0cc6ffbSLinus Torvalds if (!capable(CAP_DAC_READ_SEARCH)) 3924f0cc6ffbSLinus Torvalds return -ENOENT; 392511a7b371SAneesh Kumar K.V how = LOOKUP_EMPTY; 3926f0cc6ffbSLinus Torvalds } 3927c04030e1SUlrich Drepper 392811a7b371SAneesh Kumar K.V if (flags & AT_SYMLINK_FOLLOW) 392911a7b371SAneesh Kumar K.V how |= LOOKUP_FOLLOW; 3930442e31caSJeff Layton retry: 393111a7b371SAneesh Kumar K.V error = user_path_at(olddfd, oldname, how, &old_path); 39321da177e4SLinus Torvalds if (error) 39332ad94ae6SAl Viro return error; 39342ad94ae6SAl Viro 3935442e31caSJeff Layton new_dentry = user_path_create(newdfd, newname, &new_path, 3936442e31caSJeff Layton (how & LOOKUP_REVAL)); 39371da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 39386902d925SDave Hansen if (IS_ERR(new_dentry)) 3939dae6ad8fSAl Viro goto out; 3940dae6ad8fSAl Viro 3941dae6ad8fSAl Viro error = -EXDEV; 3942dae6ad8fSAl Viro if (old_path.mnt != new_path.mnt) 3943dae6ad8fSAl Viro goto out_dput; 3944800179c9SKees Cook error = may_linkat(&old_path); 3945800179c9SKees Cook if (unlikely(error)) 3946800179c9SKees Cook goto out_dput; 3947dae6ad8fSAl Viro error = security_path_link(old_path.dentry, &new_path, new_dentry); 3948be6d3e56SKentaro Takeda if (error) 3949a8104a9fSAl Viro goto out_dput; 3950146a8595SJ. Bruce Fields error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode); 395175c3f29dSDave Hansen out_dput: 3952921a1650SAl Viro done_path_create(&new_path, new_dentry); 3953146a8595SJ. Bruce Fields if (delegated_inode) { 3954146a8595SJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 3955d22e6338SOleg Drokin if (!error) { 3956d22e6338SOleg Drokin path_put(&old_path); 3957146a8595SJ. Bruce Fields goto retry; 3958146a8595SJ. Bruce Fields } 3959d22e6338SOleg Drokin } 3960442e31caSJeff Layton if (retry_estale(error, how)) { 3961d22e6338SOleg Drokin path_put(&old_path); 3962442e31caSJeff Layton how |= LOOKUP_REVAL; 3963442e31caSJeff Layton goto retry; 3964442e31caSJeff Layton } 39651da177e4SLinus Torvalds out: 39662d8f3038SAl Viro path_put(&old_path); 39671da177e4SLinus Torvalds 39681da177e4SLinus Torvalds return error; 39691da177e4SLinus Torvalds } 39701da177e4SLinus Torvalds 39713480b257SHeiko Carstens SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) 39725590ff0dSUlrich Drepper { 3973c04030e1SUlrich Drepper return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 39745590ff0dSUlrich Drepper } 39755590ff0dSUlrich Drepper 3976bc27027aSMiklos Szeredi /** 3977bc27027aSMiklos Szeredi * vfs_rename - rename a filesystem object 3978bc27027aSMiklos Szeredi * @old_dir: parent of source 3979bc27027aSMiklos Szeredi * @old_dentry: source 3980bc27027aSMiklos Szeredi * @new_dir: parent of destination 3981bc27027aSMiklos Szeredi * @new_dentry: destination 3982bc27027aSMiklos Szeredi * @delegated_inode: returns an inode needing a delegation break 3983520c8b16SMiklos Szeredi * @flags: rename flags 3984bc27027aSMiklos Szeredi * 3985bc27027aSMiklos Szeredi * The caller must hold multiple mutexes--see lock_rename()). 3986bc27027aSMiklos Szeredi * 3987bc27027aSMiklos Szeredi * If vfs_rename discovers a delegation in need of breaking at either 3988bc27027aSMiklos Szeredi * the source or destination, it will return -EWOULDBLOCK and return a 3989bc27027aSMiklos Szeredi * reference to the inode in delegated_inode. The caller should then 3990bc27027aSMiklos Szeredi * break the delegation and retry. Because breaking a delegation may 3991bc27027aSMiklos Szeredi * take a long time, the caller should drop all locks before doing 3992bc27027aSMiklos Szeredi * so. 3993bc27027aSMiklos Szeredi * 3994bc27027aSMiklos Szeredi * Alternatively, a caller may pass NULL for delegated_inode. This may 3995bc27027aSMiklos Szeredi * be appropriate for callers that expect the underlying filesystem not 3996bc27027aSMiklos Szeredi * to be NFS exported. 3997bc27027aSMiklos Szeredi * 39981da177e4SLinus Torvalds * The worst of all namespace operations - renaming directory. "Perverted" 39991da177e4SLinus Torvalds * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 40001da177e4SLinus Torvalds * Problems: 40011da177e4SLinus Torvalds * a) we can get into loop creation. Check is done in is_subdir(). 40021da177e4SLinus Torvalds * b) race potential - two innocent renames can create a loop together. 40031da177e4SLinus Torvalds * That's where 4.4 screws up. Current fix: serialization on 4004a11f3a05SArjan van de Ven * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 40051da177e4SLinus Torvalds * story. 40066cedba89SJ. Bruce Fields * c) we have to lock _four_ objects - parents and victim (if it exists), 40076cedba89SJ. Bruce Fields * and source (if it is not a directory). 40081b1dcc1bSJes Sorensen * And that - after we got ->i_mutex on parents (until then we don't know 40091da177e4SLinus Torvalds * whether the target exists). Solution: try to be smart with locking 40101da177e4SLinus Torvalds * order for inodes. We rely on the fact that tree topology may change 4011a11f3a05SArjan van de Ven * only under ->s_vfs_rename_mutex _and_ that parent of the object we 40121da177e4SLinus Torvalds * move will be locked. Thus we can rank directories by the tree 40131da177e4SLinus Torvalds * (ancestors first) and rank all non-directories after them. 40141da177e4SLinus Torvalds * That works since everybody except rename does "lock parent, lookup, 4015a11f3a05SArjan van de Ven * lock child" and rename is under ->s_vfs_rename_mutex. 40161da177e4SLinus Torvalds * HOWEVER, it relies on the assumption that any object with ->lookup() 40171da177e4SLinus Torvalds * has no more than 1 dentry. If "hybrid" objects will ever appear, 40181da177e4SLinus Torvalds * we'd better make sure that there's no link(2) for them. 4019e4eaac06SSage Weil * d) conversion from fhandle to dentry may come in the wrong moment - when 40201b1dcc1bSJes Sorensen * we are removing the target. Solution: we will have to grab ->i_mutex 40211da177e4SLinus Torvalds * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 4022c41b20e7SAdam Buchbinder * ->i_mutex on parents, which works but leads to some truly excessive 40231da177e4SLinus Torvalds * locking]. 40241da177e4SLinus Torvalds */ 40251da177e4SLinus Torvalds int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, 40268e6d782cSJ. Bruce Fields struct inode *new_dir, struct dentry *new_dentry, 4027520c8b16SMiklos Szeredi struct inode **delegated_inode, unsigned int flags) 40281da177e4SLinus Torvalds { 40291da177e4SLinus Torvalds int error; 4030bc27027aSMiklos Szeredi bool is_dir = d_is_dir(old_dentry); 403159b0df21SEric Paris const unsigned char *old_name; 4032bc27027aSMiklos Szeredi struct inode *source = old_dentry->d_inode; 4033bc27027aSMiklos Szeredi struct inode *target = new_dentry->d_inode; 40341da177e4SLinus Torvalds 4035bc27027aSMiklos Szeredi if (source == target) 40361da177e4SLinus Torvalds return 0; 40371da177e4SLinus Torvalds 40381da177e4SLinus Torvalds error = may_delete(old_dir, old_dentry, is_dir); 40391da177e4SLinus Torvalds if (error) 40401da177e4SLinus Torvalds return error; 40411da177e4SLinus Torvalds 4042bc27027aSMiklos Szeredi if (!target) 4043a95164d9SMiklos Szeredi error = may_create(new_dir, new_dentry); 40441da177e4SLinus Torvalds else 40451da177e4SLinus Torvalds error = may_delete(new_dir, new_dentry, is_dir); 40461da177e4SLinus Torvalds if (error) 40471da177e4SLinus Torvalds return error; 40481da177e4SLinus Torvalds 4049acfa4380SAl Viro if (!old_dir->i_op->rename) 40501da177e4SLinus Torvalds return -EPERM; 40511da177e4SLinus Torvalds 4052520c8b16SMiklos Szeredi if (flags && !old_dir->i_op->rename2) 4053520c8b16SMiklos Szeredi return -EINVAL; 4054520c8b16SMiklos Szeredi 4055bc27027aSMiklos Szeredi /* 4056bc27027aSMiklos Szeredi * If we are going to change the parent - check write permissions, 4057bc27027aSMiklos Szeredi * we'll need to flip '..'. 4058bc27027aSMiklos Szeredi */ 4059bc27027aSMiklos Szeredi if (is_dir && new_dir != old_dir) { 4060bc27027aSMiklos Szeredi error = inode_permission(source, MAY_WRITE); 4061bc27027aSMiklos Szeredi if (error) 4062bc27027aSMiklos Szeredi return error; 4063bc27027aSMiklos Szeredi } 40640eeca283SRobert Love 4065bc27027aSMiklos Szeredi error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); 4066bc27027aSMiklos Szeredi if (error) 4067bc27027aSMiklos Szeredi return error; 4068bc27027aSMiklos Szeredi 4069bc27027aSMiklos Szeredi old_name = fsnotify_oldname_init(old_dentry->d_name.name); 4070bc27027aSMiklos Szeredi dget(new_dentry); 4071bc27027aSMiklos Szeredi if (!is_dir) 4072bc27027aSMiklos Szeredi lock_two_nondirectories(source, target); 4073bc27027aSMiklos Szeredi else if (target) 4074bc27027aSMiklos Szeredi mutex_lock(&target->i_mutex); 4075bc27027aSMiklos Szeredi 4076bc27027aSMiklos Szeredi error = -EBUSY; 4077bc27027aSMiklos Szeredi if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry)) 4078bc27027aSMiklos Szeredi goto out; 4079bc27027aSMiklos Szeredi 4080bc27027aSMiklos Szeredi if (is_dir) { 4081bc27027aSMiklos Szeredi unsigned max_links = new_dir->i_sb->s_max_links; 4082bc27027aSMiklos Szeredi 4083bc27027aSMiklos Szeredi error = -EMLINK; 4084bc27027aSMiklos Szeredi if (max_links && !target && new_dir != old_dir && 4085bc27027aSMiklos Szeredi new_dir->i_nlink >= max_links) 4086bc27027aSMiklos Szeredi goto out; 4087bc27027aSMiklos Szeredi 4088bc27027aSMiklos Szeredi if (target) 4089bc27027aSMiklos Szeredi shrink_dcache_parent(new_dentry); 4090bc27027aSMiklos Szeredi } else { 4091bc27027aSMiklos Szeredi error = try_break_deleg(source, delegated_inode); 4092bc27027aSMiklos Szeredi if (error) 4093bc27027aSMiklos Szeredi goto out; 4094bc27027aSMiklos Szeredi if (target) { 4095bc27027aSMiklos Szeredi error = try_break_deleg(target, delegated_inode); 4096bc27027aSMiklos Szeredi if (error) 4097bc27027aSMiklos Szeredi goto out; 4098bc27027aSMiklos Szeredi } 4099bc27027aSMiklos Szeredi } 4100520c8b16SMiklos Szeredi if (!flags) { 4101520c8b16SMiklos Szeredi error = old_dir->i_op->rename(old_dir, old_dentry, 4102520c8b16SMiklos Szeredi new_dir, new_dentry); 4103520c8b16SMiklos Szeredi } else { 4104520c8b16SMiklos Szeredi error = old_dir->i_op->rename2(old_dir, old_dentry, 4105520c8b16SMiklos Szeredi new_dir, new_dentry, flags); 4106520c8b16SMiklos Szeredi } 4107bc27027aSMiklos Szeredi if (error) 4108bc27027aSMiklos Szeredi goto out; 4109bc27027aSMiklos Szeredi 4110bc27027aSMiklos Szeredi if (target) { 41111da177e4SLinus Torvalds if (is_dir) 4112bc27027aSMiklos Szeredi target->i_flags |= S_DEAD; 4113bc27027aSMiklos Szeredi dont_mount(new_dentry); 4114bc27027aSMiklos Szeredi } 4115bc27027aSMiklos Szeredi if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) 4116bc27027aSMiklos Szeredi d_move(old_dentry, new_dentry); 4117bc27027aSMiklos Szeredi out: 4118bc27027aSMiklos Szeredi if (!is_dir) 4119bc27027aSMiklos Szeredi unlock_two_nondirectories(source, target); 4120bc27027aSMiklos Szeredi else if (target) 4121bc27027aSMiklos Szeredi mutex_unlock(&target->i_mutex); 4122bc27027aSMiklos Szeredi dput(new_dentry); 4123123df294SAl Viro if (!error) 4124123df294SAl Viro fsnotify_move(old_dir, new_dir, old_name, is_dir, 4125bc27027aSMiklos Szeredi target, old_dentry); 41260eeca283SRobert Love fsnotify_oldname_free(old_name); 41270eeca283SRobert Love 41281da177e4SLinus Torvalds return error; 41291da177e4SLinus Torvalds } 41301da177e4SLinus Torvalds 4131520c8b16SMiklos Szeredi SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname, 4132520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname, unsigned int, flags) 41331da177e4SLinus Torvalds { 41341da177e4SLinus Torvalds struct dentry *old_dir, *new_dir; 41351da177e4SLinus Torvalds struct dentry *old_dentry, *new_dentry; 41361da177e4SLinus Torvalds struct dentry *trap; 41371da177e4SLinus Torvalds struct nameidata oldnd, newnd; 41388e6d782cSJ. Bruce Fields struct inode *delegated_inode = NULL; 413991a27b2aSJeff Layton struct filename *from; 414091a27b2aSJeff Layton struct filename *to; 4141c6a94284SJeff Layton unsigned int lookup_flags = 0; 4142c6a94284SJeff Layton bool should_retry = false; 41432ad94ae6SAl Viro int error; 4144520c8b16SMiklos Szeredi 4145520c8b16SMiklos Szeredi if (flags) 4146520c8b16SMiklos Szeredi return -EINVAL; 4147520c8b16SMiklos Szeredi 4148c6a94284SJeff Layton retry: 4149c6a94284SJeff Layton from = user_path_parent(olddfd, oldname, &oldnd, lookup_flags); 415091a27b2aSJeff Layton if (IS_ERR(from)) { 415191a27b2aSJeff Layton error = PTR_ERR(from); 41521da177e4SLinus Torvalds goto exit; 415391a27b2aSJeff Layton } 41541da177e4SLinus Torvalds 4155c6a94284SJeff Layton to = user_path_parent(newdfd, newname, &newnd, lookup_flags); 415691a27b2aSJeff Layton if (IS_ERR(to)) { 415791a27b2aSJeff Layton error = PTR_ERR(to); 41581da177e4SLinus Torvalds goto exit1; 415991a27b2aSJeff Layton } 41601da177e4SLinus Torvalds 41611da177e4SLinus Torvalds error = -EXDEV; 41624ac91378SJan Blunck if (oldnd.path.mnt != newnd.path.mnt) 41631da177e4SLinus Torvalds goto exit2; 41641da177e4SLinus Torvalds 41654ac91378SJan Blunck old_dir = oldnd.path.dentry; 41661da177e4SLinus Torvalds error = -EBUSY; 41671da177e4SLinus Torvalds if (oldnd.last_type != LAST_NORM) 41681da177e4SLinus Torvalds goto exit2; 41691da177e4SLinus Torvalds 41704ac91378SJan Blunck new_dir = newnd.path.dentry; 41711da177e4SLinus Torvalds if (newnd.last_type != LAST_NORM) 41721da177e4SLinus Torvalds goto exit2; 41731da177e4SLinus Torvalds 4174c30dabfeSJan Kara error = mnt_want_write(oldnd.path.mnt); 4175c30dabfeSJan Kara if (error) 4176c30dabfeSJan Kara goto exit2; 4177c30dabfeSJan Kara 41780612d9fbSOGAWA Hirofumi oldnd.flags &= ~LOOKUP_PARENT; 41790612d9fbSOGAWA Hirofumi newnd.flags &= ~LOOKUP_PARENT; 41804e9ed2f8SOGAWA Hirofumi newnd.flags |= LOOKUP_RENAME_TARGET; 41810612d9fbSOGAWA Hirofumi 41828e6d782cSJ. Bruce Fields retry_deleg: 41831da177e4SLinus Torvalds trap = lock_rename(new_dir, old_dir); 41841da177e4SLinus Torvalds 418549705b77SChristoph Hellwig old_dentry = lookup_hash(&oldnd); 41861da177e4SLinus Torvalds error = PTR_ERR(old_dentry); 41871da177e4SLinus Torvalds if (IS_ERR(old_dentry)) 41881da177e4SLinus Torvalds goto exit3; 41891da177e4SLinus Torvalds /* source must exist */ 41901da177e4SLinus Torvalds error = -ENOENT; 4191b18825a7SDavid Howells if (d_is_negative(old_dentry)) 41921da177e4SLinus Torvalds goto exit4; 41931da177e4SLinus Torvalds /* unless the source is a directory trailing slashes give -ENOTDIR */ 419444b1d530SMiklos Szeredi if (!d_is_dir(old_dentry)) { 41951da177e4SLinus Torvalds error = -ENOTDIR; 41961da177e4SLinus Torvalds if (oldnd.last.name[oldnd.last.len]) 41971da177e4SLinus Torvalds goto exit4; 41981da177e4SLinus Torvalds if (newnd.last.name[newnd.last.len]) 41991da177e4SLinus Torvalds goto exit4; 42001da177e4SLinus Torvalds } 42011da177e4SLinus Torvalds /* source should not be ancestor of target */ 42021da177e4SLinus Torvalds error = -EINVAL; 42031da177e4SLinus Torvalds if (old_dentry == trap) 42041da177e4SLinus Torvalds goto exit4; 420549705b77SChristoph Hellwig new_dentry = lookup_hash(&newnd); 42061da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 42071da177e4SLinus Torvalds if (IS_ERR(new_dentry)) 42081da177e4SLinus Torvalds goto exit4; 42091da177e4SLinus Torvalds /* target should not be an ancestor of source */ 42101da177e4SLinus Torvalds error = -ENOTEMPTY; 42111da177e4SLinus Torvalds if (new_dentry == trap) 42121da177e4SLinus Torvalds goto exit5; 42131da177e4SLinus Torvalds 4214be6d3e56SKentaro Takeda error = security_path_rename(&oldnd.path, old_dentry, 4215be6d3e56SKentaro Takeda &newnd.path, new_dentry); 4216be6d3e56SKentaro Takeda if (error) 4217c30dabfeSJan Kara goto exit5; 42181da177e4SLinus Torvalds error = vfs_rename(old_dir->d_inode, old_dentry, 42198e6d782cSJ. Bruce Fields new_dir->d_inode, new_dentry, 4220520c8b16SMiklos Szeredi &delegated_inode, flags); 42211da177e4SLinus Torvalds exit5: 42221da177e4SLinus Torvalds dput(new_dentry); 42231da177e4SLinus Torvalds exit4: 42241da177e4SLinus Torvalds dput(old_dentry); 42251da177e4SLinus Torvalds exit3: 42261da177e4SLinus Torvalds unlock_rename(new_dir, old_dir); 42278e6d782cSJ. Bruce Fields if (delegated_inode) { 42288e6d782cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 42298e6d782cSJ. Bruce Fields if (!error) 42308e6d782cSJ. Bruce Fields goto retry_deleg; 42318e6d782cSJ. Bruce Fields } 4232c30dabfeSJan Kara mnt_drop_write(oldnd.path.mnt); 42331da177e4SLinus Torvalds exit2: 4234c6a94284SJeff Layton if (retry_estale(error, lookup_flags)) 4235c6a94284SJeff Layton should_retry = true; 42361d957f9bSJan Blunck path_put(&newnd.path); 42372ad94ae6SAl Viro putname(to); 42381da177e4SLinus Torvalds exit1: 42391d957f9bSJan Blunck path_put(&oldnd.path); 42401da177e4SLinus Torvalds putname(from); 4241c6a94284SJeff Layton if (should_retry) { 4242c6a94284SJeff Layton should_retry = false; 4243c6a94284SJeff Layton lookup_flags |= LOOKUP_REVAL; 4244c6a94284SJeff Layton goto retry; 4245c6a94284SJeff Layton } 42462ad94ae6SAl Viro exit: 42471da177e4SLinus Torvalds return error; 42481da177e4SLinus Torvalds } 42491da177e4SLinus Torvalds 4250520c8b16SMiklos Szeredi SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, 4251520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname) 4252520c8b16SMiklos Szeredi { 4253520c8b16SMiklos Szeredi return sys_renameat2(olddfd, oldname, newdfd, newname, 0); 4254520c8b16SMiklos Szeredi } 4255520c8b16SMiklos Szeredi 4256a26eab24SHeiko Carstens SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) 42575590ff0dSUlrich Drepper { 4258520c8b16SMiklos Szeredi return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 42595590ff0dSUlrich Drepper } 42605590ff0dSUlrich Drepper 42611da177e4SLinus Torvalds int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link) 42621da177e4SLinus Torvalds { 42631da177e4SLinus Torvalds int len; 42641da177e4SLinus Torvalds 42651da177e4SLinus Torvalds len = PTR_ERR(link); 42661da177e4SLinus Torvalds if (IS_ERR(link)) 42671da177e4SLinus Torvalds goto out; 42681da177e4SLinus Torvalds 42691da177e4SLinus Torvalds len = strlen(link); 42701da177e4SLinus Torvalds if (len > (unsigned) buflen) 42711da177e4SLinus Torvalds len = buflen; 42721da177e4SLinus Torvalds if (copy_to_user(buffer, link, len)) 42731da177e4SLinus Torvalds len = -EFAULT; 42741da177e4SLinus Torvalds out: 42751da177e4SLinus Torvalds return len; 42761da177e4SLinus Torvalds } 42771da177e4SLinus Torvalds 42781da177e4SLinus Torvalds /* 42791da177e4SLinus Torvalds * A helper for ->readlink(). This should be used *ONLY* for symlinks that 42801da177e4SLinus Torvalds * have ->follow_link() touching nd only in nd_set_link(). Using (or not 42811da177e4SLinus Torvalds * using) it for any given inode is up to filesystem. 42821da177e4SLinus Torvalds */ 42831da177e4SLinus Torvalds int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen) 42841da177e4SLinus Torvalds { 42851da177e4SLinus Torvalds struct nameidata nd; 4286cc314eefSLinus Torvalds void *cookie; 4287694a1764SMarcin Slusarz int res; 4288cc314eefSLinus Torvalds 42891da177e4SLinus Torvalds nd.depth = 0; 4290cc314eefSLinus Torvalds cookie = dentry->d_inode->i_op->follow_link(dentry, &nd); 4291694a1764SMarcin Slusarz if (IS_ERR(cookie)) 4292694a1764SMarcin Slusarz return PTR_ERR(cookie); 4293694a1764SMarcin Slusarz 4294694a1764SMarcin Slusarz res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd)); 42951da177e4SLinus Torvalds if (dentry->d_inode->i_op->put_link) 4296cc314eefSLinus Torvalds dentry->d_inode->i_op->put_link(dentry, &nd, cookie); 4297694a1764SMarcin Slusarz return res; 42981da177e4SLinus Torvalds } 42991da177e4SLinus Torvalds 43001da177e4SLinus Torvalds /* get the link contents into pagecache */ 43011da177e4SLinus Torvalds static char *page_getlink(struct dentry * dentry, struct page **ppage) 43021da177e4SLinus Torvalds { 4303ebd09abbSDuane Griffin char *kaddr; 43041da177e4SLinus Torvalds struct page *page; 43051da177e4SLinus Torvalds struct address_space *mapping = dentry->d_inode->i_mapping; 4306090d2b18SPekka Enberg page = read_mapping_page(mapping, 0, NULL); 43071da177e4SLinus Torvalds if (IS_ERR(page)) 43086fe6900eSNick Piggin return (char*)page; 43091da177e4SLinus Torvalds *ppage = page; 4310ebd09abbSDuane Griffin kaddr = kmap(page); 4311ebd09abbSDuane Griffin nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1); 4312ebd09abbSDuane Griffin return kaddr; 43131da177e4SLinus Torvalds } 43141da177e4SLinus Torvalds 43151da177e4SLinus Torvalds int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 43161da177e4SLinus Torvalds { 43171da177e4SLinus Torvalds struct page *page = NULL; 43181da177e4SLinus Torvalds char *s = page_getlink(dentry, &page); 43191da177e4SLinus Torvalds int res = vfs_readlink(dentry,buffer,buflen,s); 43201da177e4SLinus Torvalds if (page) { 43211da177e4SLinus Torvalds kunmap(page); 43221da177e4SLinus Torvalds page_cache_release(page); 43231da177e4SLinus Torvalds } 43241da177e4SLinus Torvalds return res; 43251da177e4SLinus Torvalds } 43261da177e4SLinus Torvalds 4327cc314eefSLinus Torvalds void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd) 43281da177e4SLinus Torvalds { 4329cc314eefSLinus Torvalds struct page *page = NULL; 43301da177e4SLinus Torvalds nd_set_link(nd, page_getlink(dentry, &page)); 4331cc314eefSLinus Torvalds return page; 43321da177e4SLinus Torvalds } 43331da177e4SLinus Torvalds 4334cc314eefSLinus Torvalds void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) 43351da177e4SLinus Torvalds { 4336cc314eefSLinus Torvalds struct page *page = cookie; 4337cc314eefSLinus Torvalds 4338cc314eefSLinus Torvalds if (page) { 43391da177e4SLinus Torvalds kunmap(page); 43401da177e4SLinus Torvalds page_cache_release(page); 43411da177e4SLinus Torvalds } 43421da177e4SLinus Torvalds } 43431da177e4SLinus Torvalds 434454566b2cSNick Piggin /* 434554566b2cSNick Piggin * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS 434654566b2cSNick Piggin */ 434754566b2cSNick Piggin int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) 43481da177e4SLinus Torvalds { 43491da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 43500adb25d2SKirill Korotaev struct page *page; 4351afddba49SNick Piggin void *fsdata; 4352beb497abSDmitriy Monakhov int err; 43531da177e4SLinus Torvalds char *kaddr; 435454566b2cSNick Piggin unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE; 435554566b2cSNick Piggin if (nofs) 435654566b2cSNick Piggin flags |= AOP_FLAG_NOFS; 43571da177e4SLinus Torvalds 43587e53cac4SNeilBrown retry: 4359afddba49SNick Piggin err = pagecache_write_begin(NULL, mapping, 0, len-1, 436054566b2cSNick Piggin flags, &page, &fsdata); 43611da177e4SLinus Torvalds if (err) 4362afddba49SNick Piggin goto fail; 4363afddba49SNick Piggin 4364e8e3c3d6SCong Wang kaddr = kmap_atomic(page); 43651da177e4SLinus Torvalds memcpy(kaddr, symname, len-1); 4366e8e3c3d6SCong Wang kunmap_atomic(kaddr); 4367afddba49SNick Piggin 4368afddba49SNick Piggin err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, 4369afddba49SNick Piggin page, fsdata); 43701da177e4SLinus Torvalds if (err < 0) 43711da177e4SLinus Torvalds goto fail; 4372afddba49SNick Piggin if (err < len-1) 4373afddba49SNick Piggin goto retry; 4374afddba49SNick Piggin 43751da177e4SLinus Torvalds mark_inode_dirty(inode); 43761da177e4SLinus Torvalds return 0; 43771da177e4SLinus Torvalds fail: 43781da177e4SLinus Torvalds return err; 43791da177e4SLinus Torvalds } 43801da177e4SLinus Torvalds 43810adb25d2SKirill Korotaev int page_symlink(struct inode *inode, const char *symname, int len) 43820adb25d2SKirill Korotaev { 43830adb25d2SKirill Korotaev return __page_symlink(inode, symname, len, 438454566b2cSNick Piggin !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS)); 43850adb25d2SKirill Korotaev } 43860adb25d2SKirill Korotaev 438792e1d5beSArjan van de Ven const struct inode_operations page_symlink_inode_operations = { 43881da177e4SLinus Torvalds .readlink = generic_readlink, 43891da177e4SLinus Torvalds .follow_link = page_follow_link_light, 43901da177e4SLinus Torvalds .put_link = page_put_link, 43911da177e4SLinus Torvalds }; 43921da177e4SLinus Torvalds 43932d8f3038SAl Viro EXPORT_SYMBOL(user_path_at); 4394cc53ce53SDavid Howells EXPORT_SYMBOL(follow_down_one); 43951da177e4SLinus Torvalds EXPORT_SYMBOL(follow_down); 43961da177e4SLinus Torvalds EXPORT_SYMBOL(follow_up); 4397f6d2ac5cSAl Viro EXPORT_SYMBOL(get_write_access); /* nfsd */ 43981da177e4SLinus Torvalds EXPORT_SYMBOL(lock_rename); 43991da177e4SLinus Torvalds EXPORT_SYMBOL(lookup_one_len); 44001da177e4SLinus Torvalds EXPORT_SYMBOL(page_follow_link_light); 44011da177e4SLinus Torvalds EXPORT_SYMBOL(page_put_link); 44021da177e4SLinus Torvalds EXPORT_SYMBOL(page_readlink); 44030adb25d2SKirill Korotaev EXPORT_SYMBOL(__page_symlink); 44041da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink); 44051da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink_inode_operations); 4406d1811465SAl Viro EXPORT_SYMBOL(kern_path); 440716f18200SJosef 'Jeff' Sipek EXPORT_SYMBOL(vfs_path_lookup); 4408f419a2e3SAl Viro EXPORT_SYMBOL(inode_permission); 44091da177e4SLinus Torvalds EXPORT_SYMBOL(unlock_rename); 44101da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_create); 44111da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_link); 44121da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_mkdir); 44131da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_mknod); 44141da177e4SLinus Torvalds EXPORT_SYMBOL(generic_permission); 44151da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_readlink); 44161da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_rename); 44171da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_rmdir); 44181da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_symlink); 44191da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_unlink); 44201da177e4SLinus Torvalds EXPORT_SYMBOL(dentry_unhash); 44211da177e4SLinus Torvalds EXPORT_SYMBOL(generic_readlink); 4422