1b2441318SGreg Kroah-Hartman // SPDX-License-Identifier: GPL-2.0 21da177e4SLinus Torvalds /* 31da177e4SLinus Torvalds * linux/fs/namei.c 41da177e4SLinus Torvalds * 51da177e4SLinus Torvalds * Copyright (C) 1991, 1992 Linus Torvalds 61da177e4SLinus Torvalds */ 71da177e4SLinus Torvalds 81da177e4SLinus Torvalds /* 91da177e4SLinus Torvalds * Some corrections by tytso. 101da177e4SLinus Torvalds */ 111da177e4SLinus Torvalds 121da177e4SLinus Torvalds /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname 131da177e4SLinus Torvalds * lookup logic. 141da177e4SLinus Torvalds */ 151da177e4SLinus Torvalds /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture. 161da177e4SLinus Torvalds */ 171da177e4SLinus Torvalds 181da177e4SLinus Torvalds #include <linux/init.h> 19630d9c47SPaul Gortmaker #include <linux/export.h> 2044696908SDavid S. Miller #include <linux/kernel.h> 211da177e4SLinus Torvalds #include <linux/slab.h> 221da177e4SLinus Torvalds #include <linux/fs.h> 231da177e4SLinus Torvalds #include <linux/namei.h> 241da177e4SLinus Torvalds #include <linux/pagemap.h> 250eeca283SRobert Love #include <linux/fsnotify.h> 261da177e4SLinus Torvalds #include <linux/personality.h> 271da177e4SLinus Torvalds #include <linux/security.h> 286146f0d5SMimi Zohar #include <linux/ima.h> 291da177e4SLinus Torvalds #include <linux/syscalls.h> 301da177e4SLinus Torvalds #include <linux/mount.h> 311da177e4SLinus Torvalds #include <linux/audit.h> 3216f7e0feSRandy Dunlap #include <linux/capability.h> 33834f2a4aSTrond Myklebust #include <linux/file.h> 345590ff0dSUlrich Drepper #include <linux/fcntl.h> 3508ce5f16SSerge E. Hallyn #include <linux/device_cgroup.h> 365ad4e53bSAl Viro #include <linux/fs_struct.h> 37e77819e5SLinus Torvalds #include <linux/posix_acl.h> 3899d263d4SLinus Torvalds #include <linux/hash.h> 392a18da7aSGeorge Spelvin #include <linux/bitops.h> 40aeaa4a79SEric W. Biederman #include <linux/init_task.h> 417c0f6ba6SLinus Torvalds #include <linux/uaccess.h> 421da177e4SLinus Torvalds 43e81e3f4dSEric Paris #include "internal.h" 44c7105365SAl Viro #include "mount.h" 45e81e3f4dSEric Paris 461da177e4SLinus Torvalds /* [Feb-1997 T. Schoebel-Theuer] 471da177e4SLinus Torvalds * Fundamental changes in the pathname lookup mechanisms (namei) 481da177e4SLinus Torvalds * were necessary because of omirr. The reason is that omirr needs 491da177e4SLinus Torvalds * to know the _real_ pathname, not the user-supplied one, in case 501da177e4SLinus Torvalds * of symlinks (and also when transname replacements occur). 511da177e4SLinus Torvalds * 521da177e4SLinus Torvalds * The new code replaces the old recursive symlink resolution with 531da177e4SLinus Torvalds * an iterative one (in case of non-nested symlink chains). It does 541da177e4SLinus Torvalds * this with calls to <fs>_follow_link(). 551da177e4SLinus Torvalds * As a side effect, dir_namei(), _namei() and follow_link() are now 561da177e4SLinus Torvalds * replaced with a single function lookup_dentry() that can handle all 571da177e4SLinus Torvalds * the special cases of the former code. 581da177e4SLinus Torvalds * 591da177e4SLinus Torvalds * With the new dcache, the pathname is stored at each inode, at least as 601da177e4SLinus Torvalds * long as the refcount of the inode is positive. As a side effect, the 611da177e4SLinus Torvalds * size of the dcache depends on the inode cache and thus is dynamic. 621da177e4SLinus Torvalds * 631da177e4SLinus Torvalds * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink 641da177e4SLinus Torvalds * resolution to correspond with current state of the code. 651da177e4SLinus Torvalds * 661da177e4SLinus Torvalds * Note that the symlink resolution is not *completely* iterative. 671da177e4SLinus Torvalds * There is still a significant amount of tail- and mid- recursion in 681da177e4SLinus Torvalds * the algorithm. Also, note that <fs>_readlink() is not used in 691da177e4SLinus Torvalds * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink() 701da177e4SLinus Torvalds * may return different results than <fs>_follow_link(). Many virtual 711da177e4SLinus Torvalds * filesystems (including /proc) exhibit this behavior. 721da177e4SLinus Torvalds */ 731da177e4SLinus Torvalds 741da177e4SLinus Torvalds /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation: 751da177e4SLinus Torvalds * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL 761da177e4SLinus Torvalds * and the name already exists in form of a symlink, try to create the new 771da177e4SLinus Torvalds * name indicated by the symlink. The old code always complained that the 781da177e4SLinus Torvalds * name already exists, due to not following the symlink even if its target 791da177e4SLinus Torvalds * is nonexistent. The new semantics affects also mknod() and link() when 8025985edcSLucas De Marchi * the name is a symlink pointing to a non-existent name. 811da177e4SLinus Torvalds * 821da177e4SLinus Torvalds * I don't know which semantics is the right one, since I have no access 831da177e4SLinus Torvalds * to standards. But I found by trial that HP-UX 9.0 has the full "new" 841da177e4SLinus Torvalds * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the 851da177e4SLinus Torvalds * "old" one. Personally, I think the new semantics is much more logical. 861da177e4SLinus Torvalds * Note that "ln old new" where "new" is a symlink pointing to a non-existing 871da177e4SLinus Torvalds * file does succeed in both HP-UX and SunOs, but not in Solaris 881da177e4SLinus Torvalds * and in the old Linux semantics. 891da177e4SLinus Torvalds */ 901da177e4SLinus Torvalds 911da177e4SLinus Torvalds /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink 921da177e4SLinus Torvalds * semantics. See the comments in "open_namei" and "do_link" below. 931da177e4SLinus Torvalds * 941da177e4SLinus Torvalds * [10-Sep-98 Alan Modra] Another symlink change. 951da177e4SLinus Torvalds */ 961da177e4SLinus Torvalds 971da177e4SLinus Torvalds /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks: 981da177e4SLinus Torvalds * inside the path - always follow. 991da177e4SLinus Torvalds * in the last component in creation/removal/renaming - never follow. 1001da177e4SLinus Torvalds * if LOOKUP_FOLLOW passed - follow. 1011da177e4SLinus Torvalds * if the pathname has trailing slashes - follow. 1021da177e4SLinus Torvalds * otherwise - don't follow. 1031da177e4SLinus Torvalds * (applied in that order). 1041da177e4SLinus Torvalds * 1051da177e4SLinus Torvalds * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT 1061da177e4SLinus Torvalds * restored for 2.4. This is the last surviving part of old 4.2BSD bug. 1071da177e4SLinus Torvalds * During the 2.4 we need to fix the userland stuff depending on it - 1081da177e4SLinus Torvalds * hopefully we will be able to get rid of that wart in 2.5. So far only 1091da177e4SLinus Torvalds * XEmacs seems to be relying on it... 1101da177e4SLinus Torvalds */ 1111da177e4SLinus Torvalds /* 1121da177e4SLinus Torvalds * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland) 113a11f3a05SArjan van de Ven * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives 1141da177e4SLinus Torvalds * any extra contention... 1151da177e4SLinus Torvalds */ 1161da177e4SLinus Torvalds 1171da177e4SLinus Torvalds /* In order to reduce some races, while at the same time doing additional 1181da177e4SLinus Torvalds * checking and hopefully speeding things up, we copy filenames to the 1191da177e4SLinus Torvalds * kernel data space before using them.. 1201da177e4SLinus Torvalds * 1211da177e4SLinus Torvalds * POSIX.1 2.4: an empty pathname is invalid (ENOENT). 1221da177e4SLinus Torvalds * PATH_MAX includes the nul terminator --RR. 1231da177e4SLinus Torvalds */ 1247950e385SJeff Layton 125fd2f7cb5SAl Viro #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname)) 12691a27b2aSJeff Layton 12751f39a1fSDavid Drysdale struct filename * 12891a27b2aSJeff Layton getname_flags(const char __user *filename, int flags, int *empty) 12991a27b2aSJeff Layton { 13094b5d262SAl Viro struct filename *result; 1317950e385SJeff Layton char *kname; 13294b5d262SAl Viro int len; 1331da177e4SLinus Torvalds 1347ac86265SJeff Layton result = audit_reusename(filename); 1357ac86265SJeff Layton if (result) 1367ac86265SJeff Layton return result; 1377ac86265SJeff Layton 1387950e385SJeff Layton result = __getname(); 1393f9f0aa6SLinus Torvalds if (unlikely(!result)) 1404043cde8SEric Paris return ERR_PTR(-ENOMEM); 1411da177e4SLinus Torvalds 1427950e385SJeff Layton /* 1437950e385SJeff Layton * First, try to embed the struct filename inside the names_cache 1447950e385SJeff Layton * allocation 1457950e385SJeff Layton */ 146fd2f7cb5SAl Viro kname = (char *)result->iname; 14791a27b2aSJeff Layton result->name = kname; 1487950e385SJeff Layton 14994b5d262SAl Viro len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX); 15091a27b2aSJeff Layton if (unlikely(len < 0)) { 15194b5d262SAl Viro __putname(result); 15294b5d262SAl Viro return ERR_PTR(len); 15391a27b2aSJeff Layton } 1543f9f0aa6SLinus Torvalds 1557950e385SJeff Layton /* 1567950e385SJeff Layton * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a 1577950e385SJeff Layton * separate struct filename so we can dedicate the entire 1587950e385SJeff Layton * names_cache allocation for the pathname, and re-do the copy from 1597950e385SJeff Layton * userland. 1607950e385SJeff Layton */ 16194b5d262SAl Viro if (unlikely(len == EMBEDDED_NAME_MAX)) { 162fd2f7cb5SAl Viro const size_t size = offsetof(struct filename, iname[1]); 1637950e385SJeff Layton kname = (char *)result; 1647950e385SJeff Layton 165fd2f7cb5SAl Viro /* 166fd2f7cb5SAl Viro * size is chosen that way we to guarantee that 167fd2f7cb5SAl Viro * result->iname[0] is within the same object and that 168fd2f7cb5SAl Viro * kname can't be equal to result->iname, no matter what. 169fd2f7cb5SAl Viro */ 170fd2f7cb5SAl Viro result = kzalloc(size, GFP_KERNEL); 17194b5d262SAl Viro if (unlikely(!result)) { 17294b5d262SAl Viro __putname(kname); 17394b5d262SAl Viro return ERR_PTR(-ENOMEM); 1747950e385SJeff Layton } 1757950e385SJeff Layton result->name = kname; 17694b5d262SAl Viro len = strncpy_from_user(kname, filename, PATH_MAX); 17794b5d262SAl Viro if (unlikely(len < 0)) { 17894b5d262SAl Viro __putname(kname); 17994b5d262SAl Viro kfree(result); 18094b5d262SAl Viro return ERR_PTR(len); 18194b5d262SAl Viro } 18294b5d262SAl Viro if (unlikely(len == PATH_MAX)) { 18394b5d262SAl Viro __putname(kname); 18494b5d262SAl Viro kfree(result); 18594b5d262SAl Viro return ERR_PTR(-ENAMETOOLONG); 18694b5d262SAl Viro } 1877950e385SJeff Layton } 1887950e385SJeff Layton 18994b5d262SAl Viro result->refcnt = 1; 1903f9f0aa6SLinus Torvalds /* The empty path is special. */ 1913f9f0aa6SLinus Torvalds if (unlikely(!len)) { 1923f9f0aa6SLinus Torvalds if (empty) 1931fa1e7f6SAndy Whitcroft *empty = 1; 19494b5d262SAl Viro if (!(flags & LOOKUP_EMPTY)) { 19594b5d262SAl Viro putname(result); 19694b5d262SAl Viro return ERR_PTR(-ENOENT); 1971da177e4SLinus Torvalds } 19894b5d262SAl Viro } 1997950e385SJeff Layton 2007950e385SJeff Layton result->uptr = filename; 201c4ad8f98SLinus Torvalds result->aname = NULL; 2021da177e4SLinus Torvalds audit_getname(result); 2031da177e4SLinus Torvalds return result; 2043f9f0aa6SLinus Torvalds } 2053f9f0aa6SLinus Torvalds 20691a27b2aSJeff Layton struct filename * 20791a27b2aSJeff Layton getname(const char __user * filename) 208f52e0c11SAl Viro { 209f7493e5dSLinus Torvalds return getname_flags(filename, 0, NULL); 210f52e0c11SAl Viro } 211f52e0c11SAl Viro 212c4ad8f98SLinus Torvalds struct filename * 213c4ad8f98SLinus Torvalds getname_kernel(const char * filename) 214c4ad8f98SLinus Torvalds { 215c4ad8f98SLinus Torvalds struct filename *result; 21608518549SPaul Moore int len = strlen(filename) + 1; 217c4ad8f98SLinus Torvalds 218c4ad8f98SLinus Torvalds result = __getname(); 219c4ad8f98SLinus Torvalds if (unlikely(!result)) 220c4ad8f98SLinus Torvalds return ERR_PTR(-ENOMEM); 221c4ad8f98SLinus Torvalds 22208518549SPaul Moore if (len <= EMBEDDED_NAME_MAX) { 223fd2f7cb5SAl Viro result->name = (char *)result->iname; 22408518549SPaul Moore } else if (len <= PATH_MAX) { 22530ce4d19SAl Viro const size_t size = offsetof(struct filename, iname[1]); 22608518549SPaul Moore struct filename *tmp; 22708518549SPaul Moore 22830ce4d19SAl Viro tmp = kmalloc(size, GFP_KERNEL); 22908518549SPaul Moore if (unlikely(!tmp)) { 23008518549SPaul Moore __putname(result); 23108518549SPaul Moore return ERR_PTR(-ENOMEM); 23208518549SPaul Moore } 23308518549SPaul Moore tmp->name = (char *)result; 23408518549SPaul Moore result = tmp; 23508518549SPaul Moore } else { 23608518549SPaul Moore __putname(result); 23708518549SPaul Moore return ERR_PTR(-ENAMETOOLONG); 23808518549SPaul Moore } 23908518549SPaul Moore memcpy((char *)result->name, filename, len); 240c4ad8f98SLinus Torvalds result->uptr = NULL; 241c4ad8f98SLinus Torvalds result->aname = NULL; 24255422d0bSPaul Moore result->refcnt = 1; 243fd3522fdSPaul Moore audit_getname(result); 244c4ad8f98SLinus Torvalds 245c4ad8f98SLinus Torvalds return result; 246c4ad8f98SLinus Torvalds } 247c4ad8f98SLinus Torvalds 24891a27b2aSJeff Layton void putname(struct filename *name) 2491da177e4SLinus Torvalds { 25055422d0bSPaul Moore BUG_ON(name->refcnt <= 0); 25155422d0bSPaul Moore 25255422d0bSPaul Moore if (--name->refcnt > 0) 25355422d0bSPaul Moore return; 25455422d0bSPaul Moore 255fd2f7cb5SAl Viro if (name->name != name->iname) { 25655422d0bSPaul Moore __putname(name->name); 25755422d0bSPaul Moore kfree(name); 25855422d0bSPaul Moore } else 25955422d0bSPaul Moore __putname(name); 2601da177e4SLinus Torvalds } 2611da177e4SLinus Torvalds 262e77819e5SLinus Torvalds static int check_acl(struct inode *inode, int mask) 263e77819e5SLinus Torvalds { 26484635d68SLinus Torvalds #ifdef CONFIG_FS_POSIX_ACL 265e77819e5SLinus Torvalds struct posix_acl *acl; 266e77819e5SLinus Torvalds 267e77819e5SLinus Torvalds if (mask & MAY_NOT_BLOCK) { 2683567866bSAl Viro acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS); 2693567866bSAl Viro if (!acl) 270e77819e5SLinus Torvalds return -EAGAIN; 2713567866bSAl Viro /* no ->get_acl() calls in RCU mode... */ 272b8a7a3a6SAndreas Gruenbacher if (is_uncached_acl(acl)) 273e77819e5SLinus Torvalds return -ECHILD; 27463d72b93SLinus Torvalds return posix_acl_permission(inode, acl, mask); 275e77819e5SLinus Torvalds } 276e77819e5SLinus Torvalds 2772982baa2SChristoph Hellwig acl = get_acl(inode, ACL_TYPE_ACCESS); 2784e34e719SChristoph Hellwig if (IS_ERR(acl)) 2794e34e719SChristoph Hellwig return PTR_ERR(acl); 280e77819e5SLinus Torvalds if (acl) { 281e77819e5SLinus Torvalds int error = posix_acl_permission(inode, acl, mask); 282e77819e5SLinus Torvalds posix_acl_release(acl); 283e77819e5SLinus Torvalds return error; 284e77819e5SLinus Torvalds } 28584635d68SLinus Torvalds #endif 286e77819e5SLinus Torvalds 287e77819e5SLinus Torvalds return -EAGAIN; 288e77819e5SLinus Torvalds } 289e77819e5SLinus Torvalds 2905909ccaaSLinus Torvalds /* 2915fc475b7SLinus Torvalds * This does the basic UNIX permission checking. 2925fc475b7SLinus Torvalds * 2935fc475b7SLinus Torvalds * Note that the POSIX ACL check cares about the MAY_NOT_BLOCK bit, 2945fc475b7SLinus Torvalds * for RCU walking. 2955909ccaaSLinus Torvalds */ 2967e40145eSAl Viro static int acl_permission_check(struct inode *inode, int mask) 2975909ccaaSLinus Torvalds { 29826cf46beSLinus Torvalds unsigned int mode = inode->i_mode; 2995909ccaaSLinus Torvalds 3005fc475b7SLinus Torvalds /* Are we the owner? If so, ACL's don't matter */ 3015fc475b7SLinus Torvalds if (likely(uid_eq(current_fsuid(), inode->i_uid))) { 3025fc475b7SLinus Torvalds mask &= 7; 3035909ccaaSLinus Torvalds mode >>= 6; 3045fc475b7SLinus Torvalds return (mask & ~mode) ? -EACCES : 0; 3055fc475b7SLinus Torvalds } 3065fc475b7SLinus Torvalds 3075fc475b7SLinus Torvalds /* Do we have ACL's? */ 308e77819e5SLinus Torvalds if (IS_POSIXACL(inode) && (mode & S_IRWXG)) { 3097e40145eSAl Viro int error = check_acl(inode, mask); 3105909ccaaSLinus Torvalds if (error != -EAGAIN) 3115909ccaaSLinus Torvalds return error; 3125909ccaaSLinus Torvalds } 3135909ccaaSLinus Torvalds 3145fc475b7SLinus Torvalds /* Only RWX matters for group/other mode bits */ 3155fc475b7SLinus Torvalds mask &= 7; 3165fc475b7SLinus Torvalds 3175fc475b7SLinus Torvalds /* 3185fc475b7SLinus Torvalds * Are the group permissions different from 3195fc475b7SLinus Torvalds * the other permissions in the bits we care 3205fc475b7SLinus Torvalds * about? Need to check group ownership if so. 3215fc475b7SLinus Torvalds */ 3225fc475b7SLinus Torvalds if (mask & (mode ^ (mode >> 3))) { 3235909ccaaSLinus Torvalds if (in_group_p(inode->i_gid)) 3245909ccaaSLinus Torvalds mode >>= 3; 3255909ccaaSLinus Torvalds } 3265909ccaaSLinus Torvalds 3275fc475b7SLinus Torvalds /* Bits in 'mode' clear that we require? */ 3285fc475b7SLinus Torvalds return (mask & ~mode) ? -EACCES : 0; 3295909ccaaSLinus Torvalds } 3301da177e4SLinus Torvalds 3311da177e4SLinus Torvalds /** 3321da177e4SLinus Torvalds * generic_permission - check for access rights on a Posix-like filesystem 3331da177e4SLinus Torvalds * @inode: inode to check access rights for 3345fc475b7SLinus Torvalds * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, 3355fc475b7SLinus Torvalds * %MAY_NOT_BLOCK ...) 3361da177e4SLinus Torvalds * 3371da177e4SLinus Torvalds * Used to check for read/write/execute permissions on a file. 3381da177e4SLinus Torvalds * We use "fsuid" for this, letting us set arbitrary permissions 3391da177e4SLinus Torvalds * for filesystem access without changing the "normal" uids which 340b74c79e9SNick Piggin * are used for other things. 341b74c79e9SNick Piggin * 342b74c79e9SNick Piggin * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk 343b74c79e9SNick Piggin * request cannot be satisfied (eg. requires blocking or too much complexity). 344b74c79e9SNick Piggin * It would then be called again in ref-walk mode. 3451da177e4SLinus Torvalds */ 3462830ba7fSAl Viro int generic_permission(struct inode *inode, int mask) 3471da177e4SLinus Torvalds { 3485909ccaaSLinus Torvalds int ret; 3491da177e4SLinus Torvalds 3501da177e4SLinus Torvalds /* 351948409c7SAndreas Gruenbacher * Do the basic permission checks. 3521da177e4SLinus Torvalds */ 3537e40145eSAl Viro ret = acl_permission_check(inode, mask); 3545909ccaaSLinus Torvalds if (ret != -EACCES) 3555909ccaaSLinus Torvalds return ret; 3561da177e4SLinus Torvalds 357d594e7ecSAl Viro if (S_ISDIR(inode->i_mode)) { 358d594e7ecSAl Viro /* DACs are overridable for directories */ 359d594e7ecSAl Viro if (!(mask & MAY_WRITE)) 36023adbe12SAndy Lutomirski if (capable_wrt_inode_uidgid(inode, 36123adbe12SAndy Lutomirski CAP_DAC_READ_SEARCH)) 362d594e7ecSAl Viro return 0; 36323adbe12SAndy Lutomirski if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE)) 3641da177e4SLinus Torvalds return 0; 3652a4c2242SStephen Smalley return -EACCES; 3662a4c2242SStephen Smalley } 3671da177e4SLinus Torvalds 3681da177e4SLinus Torvalds /* 3691da177e4SLinus Torvalds * Searching includes executable on directories, else just read. 3701da177e4SLinus Torvalds */ 3717ea66001SSerge E. Hallyn mask &= MAY_READ | MAY_WRITE | MAY_EXEC; 372d594e7ecSAl Viro if (mask == MAY_READ) 37323adbe12SAndy Lutomirski if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH)) 3741da177e4SLinus Torvalds return 0; 3752a4c2242SStephen Smalley /* 3762a4c2242SStephen Smalley * Read/write DACs are always overridable. 3772a4c2242SStephen Smalley * Executable DACs are overridable when there is 3782a4c2242SStephen Smalley * at least one exec bit set. 3792a4c2242SStephen Smalley */ 3802a4c2242SStephen Smalley if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO)) 3812a4c2242SStephen Smalley if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE)) 3822a4c2242SStephen Smalley return 0; 3831da177e4SLinus Torvalds 3841da177e4SLinus Torvalds return -EACCES; 3851da177e4SLinus Torvalds } 3864d359507SAl Viro EXPORT_SYMBOL(generic_permission); 3871da177e4SLinus Torvalds 3883ddcd056SLinus Torvalds /* 3893ddcd056SLinus Torvalds * We _really_ want to just do "generic_permission()" without 3903ddcd056SLinus Torvalds * even looking at the inode->i_op values. So we keep a cache 3913ddcd056SLinus Torvalds * flag in inode->i_opflags, that says "this has not special 3923ddcd056SLinus Torvalds * permission function, use the fast case". 3933ddcd056SLinus Torvalds */ 3943ddcd056SLinus Torvalds static inline int do_inode_permission(struct inode *inode, int mask) 3953ddcd056SLinus Torvalds { 3963ddcd056SLinus Torvalds if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) { 3973ddcd056SLinus Torvalds if (likely(inode->i_op->permission)) 3983ddcd056SLinus Torvalds return inode->i_op->permission(inode, mask); 3993ddcd056SLinus Torvalds 4003ddcd056SLinus Torvalds /* This gets set once for the inode lifetime */ 4013ddcd056SLinus Torvalds spin_lock(&inode->i_lock); 4023ddcd056SLinus Torvalds inode->i_opflags |= IOP_FASTPERM; 4033ddcd056SLinus Torvalds spin_unlock(&inode->i_lock); 4043ddcd056SLinus Torvalds } 4053ddcd056SLinus Torvalds return generic_permission(inode, mask); 4063ddcd056SLinus Torvalds } 4073ddcd056SLinus Torvalds 408cb23beb5SChristoph Hellwig /** 4090bdaea90SDavid Howells * sb_permission - Check superblock-level permissions 4100bdaea90SDavid Howells * @sb: Superblock of inode to check permission on 41155852635SRandy Dunlap * @inode: Inode to check permission on 4120bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 4130bdaea90SDavid Howells * 4140bdaea90SDavid Howells * Separate out file-system wide checks from inode-specific permission checks. 4150bdaea90SDavid Howells */ 4160bdaea90SDavid Howells static int sb_permission(struct super_block *sb, struct inode *inode, int mask) 4170bdaea90SDavid Howells { 4180bdaea90SDavid Howells if (unlikely(mask & MAY_WRITE)) { 4190bdaea90SDavid Howells umode_t mode = inode->i_mode; 4200bdaea90SDavid Howells 4210bdaea90SDavid Howells /* Nobody gets write access to a read-only fs. */ 422bc98a42cSDavid Howells if (sb_rdonly(sb) && (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) 4230bdaea90SDavid Howells return -EROFS; 4240bdaea90SDavid Howells } 4250bdaea90SDavid Howells return 0; 4260bdaea90SDavid Howells } 4270bdaea90SDavid Howells 4280bdaea90SDavid Howells /** 4290bdaea90SDavid Howells * inode_permission - Check for access rights to a given inode 4300bdaea90SDavid Howells * @inode: Inode to check permission on 4310bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 4320bdaea90SDavid Howells * 4330bdaea90SDavid Howells * Check for read/write/execute permissions on an inode. We use fs[ug]id for 4340bdaea90SDavid Howells * this, letting us set arbitrary permissions for filesystem access without 4350bdaea90SDavid Howells * changing the "normal" UIDs which are used for other things. 4360bdaea90SDavid Howells * 4370bdaea90SDavid Howells * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. 4380bdaea90SDavid Howells */ 4390bdaea90SDavid Howells int inode_permission(struct inode *inode, int mask) 4400bdaea90SDavid Howells { 4410bdaea90SDavid Howells int retval; 4420bdaea90SDavid Howells 4430bdaea90SDavid Howells retval = sb_permission(inode->i_sb, inode, mask); 4440bdaea90SDavid Howells if (retval) 4450bdaea90SDavid Howells return retval; 4464bfd054aSEric Biggers 4474bfd054aSEric Biggers if (unlikely(mask & MAY_WRITE)) { 4484bfd054aSEric Biggers /* 4494bfd054aSEric Biggers * Nobody gets write access to an immutable file. 4504bfd054aSEric Biggers */ 4514bfd054aSEric Biggers if (IS_IMMUTABLE(inode)) 4524bfd054aSEric Biggers return -EPERM; 4534bfd054aSEric Biggers 4544bfd054aSEric Biggers /* 4554bfd054aSEric Biggers * Updating mtime will likely cause i_uid and i_gid to be 4564bfd054aSEric Biggers * written back improperly if their true value is unknown 4574bfd054aSEric Biggers * to the vfs. 4584bfd054aSEric Biggers */ 4594bfd054aSEric Biggers if (HAS_UNMAPPED_ID(inode)) 4604bfd054aSEric Biggers return -EACCES; 4614bfd054aSEric Biggers } 4624bfd054aSEric Biggers 4634bfd054aSEric Biggers retval = do_inode_permission(inode, mask); 4644bfd054aSEric Biggers if (retval) 4654bfd054aSEric Biggers return retval; 4664bfd054aSEric Biggers 4674bfd054aSEric Biggers retval = devcgroup_inode_permission(inode, mask); 4684bfd054aSEric Biggers if (retval) 4694bfd054aSEric Biggers return retval; 4704bfd054aSEric Biggers 4714bfd054aSEric Biggers return security_inode_permission(inode, mask); 4720bdaea90SDavid Howells } 4734d359507SAl Viro EXPORT_SYMBOL(inode_permission); 4740bdaea90SDavid Howells 4750bdaea90SDavid Howells /** 4765dd784d0SJan Blunck * path_get - get a reference to a path 4775dd784d0SJan Blunck * @path: path to get the reference to 4785dd784d0SJan Blunck * 4795dd784d0SJan Blunck * Given a path increment the reference count to the dentry and the vfsmount. 4805dd784d0SJan Blunck */ 481dcf787f3SAl Viro void path_get(const struct path *path) 4825dd784d0SJan Blunck { 4835dd784d0SJan Blunck mntget(path->mnt); 4845dd784d0SJan Blunck dget(path->dentry); 4855dd784d0SJan Blunck } 4865dd784d0SJan Blunck EXPORT_SYMBOL(path_get); 4875dd784d0SJan Blunck 4885dd784d0SJan Blunck /** 4891d957f9bSJan Blunck * path_put - put a reference to a path 4901d957f9bSJan Blunck * @path: path to put the reference to 4911d957f9bSJan Blunck * 4921d957f9bSJan Blunck * Given a path decrement the reference count to the dentry and the vfsmount. 4931d957f9bSJan Blunck */ 494dcf787f3SAl Viro void path_put(const struct path *path) 4951da177e4SLinus Torvalds { 4961d957f9bSJan Blunck dput(path->dentry); 4971d957f9bSJan Blunck mntput(path->mnt); 4981da177e4SLinus Torvalds } 4991d957f9bSJan Blunck EXPORT_SYMBOL(path_put); 5001da177e4SLinus Torvalds 501894bc8c4SAl Viro #define EMBEDDED_LEVELS 2 5021f55a6ecSAl Viro struct nameidata { 5031f55a6ecSAl Viro struct path path; 5041f55a6ecSAl Viro struct qstr last; 5051f55a6ecSAl Viro struct path root; 5061f55a6ecSAl Viro struct inode *inode; /* path.dentry.d_inode */ 5071f55a6ecSAl Viro unsigned int flags; 508ab87f9a5SAleksa Sarai unsigned seq, m_seq, r_seq; 5091f55a6ecSAl Viro int last_type; 5101f55a6ecSAl Viro unsigned depth; 511756daf26SNeilBrown int total_link_count; 512697fc6caSAl Viro struct saved { 513697fc6caSAl Viro struct path link; 514fceef393SAl Viro struct delayed_call done; 515697fc6caSAl Viro const char *name; 5160450b2d1SAl Viro unsigned seq; 517894bc8c4SAl Viro } *stack, internal[EMBEDDED_LEVELS]; 5189883d185SAl Viro struct filename *name; 5199883d185SAl Viro struct nameidata *saved; 5209883d185SAl Viro unsigned root_seq; 5219883d185SAl Viro int dfd; 5220f705953SAl Viro kuid_t dir_uid; 5230f705953SAl Viro umode_t dir_mode; 5243859a271SKees Cook } __randomize_layout; 5251f55a6ecSAl Viro 5269883d185SAl Viro static void set_nameidata(struct nameidata *p, int dfd, struct filename *name) 527894bc8c4SAl Viro { 528756daf26SNeilBrown struct nameidata *old = current->nameidata; 529756daf26SNeilBrown p->stack = p->internal; 530c8a53ee5SAl Viro p->dfd = dfd; 531c8a53ee5SAl Viro p->name = name; 532756daf26SNeilBrown p->total_link_count = old ? old->total_link_count : 0; 5339883d185SAl Viro p->saved = old; 534756daf26SNeilBrown current->nameidata = p; 535894bc8c4SAl Viro } 536894bc8c4SAl Viro 5379883d185SAl Viro static void restore_nameidata(void) 538894bc8c4SAl Viro { 5399883d185SAl Viro struct nameidata *now = current->nameidata, *old = now->saved; 540756daf26SNeilBrown 541756daf26SNeilBrown current->nameidata = old; 542756daf26SNeilBrown if (old) 543756daf26SNeilBrown old->total_link_count = now->total_link_count; 544e1a63bbcSAl Viro if (now->stack != now->internal) 545756daf26SNeilBrown kfree(now->stack); 546894bc8c4SAl Viro } 547894bc8c4SAl Viro 54860ef60c7SAl Viro static bool nd_alloc_stack(struct nameidata *nd) 549894bc8c4SAl Viro { 550bc40aee0SAl Viro struct saved *p; 551bc40aee0SAl Viro 5526da2ec56SKees Cook p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved), 55360ef60c7SAl Viro nd->flags & LOOKUP_RCU ? GFP_ATOMIC : GFP_KERNEL); 554bc40aee0SAl Viro if (unlikely(!p)) 55560ef60c7SAl Viro return false; 556894bc8c4SAl Viro memcpy(p, nd->internal, sizeof(nd->internal)); 557894bc8c4SAl Viro nd->stack = p; 55860ef60c7SAl Viro return true; 559894bc8c4SAl Viro } 560894bc8c4SAl Viro 561397d425dSEric W. Biederman /** 5626b03f7edSAl Viro * path_connected - Verify that a dentry is below mnt.mnt_root 563397d425dSEric W. Biederman * 564397d425dSEric W. Biederman * Rename can sometimes move a file or directory outside of a bind 565397d425dSEric W. Biederman * mount, path_connected allows those cases to be detected. 566397d425dSEric W. Biederman */ 5676b03f7edSAl Viro static bool path_connected(struct vfsmount *mnt, struct dentry *dentry) 568397d425dSEric W. Biederman { 56995dd7758SEric W. Biederman struct super_block *sb = mnt->mnt_sb; 570397d425dSEric W. Biederman 571402dd2cfSChristoph Hellwig /* Bind mounts can have disconnected paths */ 572402dd2cfSChristoph Hellwig if (mnt->mnt_root == sb->s_root) 573397d425dSEric W. Biederman return true; 574397d425dSEric W. Biederman 5756b03f7edSAl Viro return is_subdir(dentry, mnt->mnt_root); 576397d425dSEric W. Biederman } 577397d425dSEric W. Biederman 5787973387aSAl Viro static void drop_links(struct nameidata *nd) 5797973387aSAl Viro { 5807973387aSAl Viro int i = nd->depth; 5817973387aSAl Viro while (i--) { 5827973387aSAl Viro struct saved *last = nd->stack + i; 583fceef393SAl Viro do_delayed_call(&last->done); 584fceef393SAl Viro clear_delayed_call(&last->done); 5857973387aSAl Viro } 5867973387aSAl Viro } 5877973387aSAl Viro 5887973387aSAl Viro static void terminate_walk(struct nameidata *nd) 5897973387aSAl Viro { 5907973387aSAl Viro drop_links(nd); 5917973387aSAl Viro if (!(nd->flags & LOOKUP_RCU)) { 5927973387aSAl Viro int i; 5937973387aSAl Viro path_put(&nd->path); 5947973387aSAl Viro for (i = 0; i < nd->depth; i++) 5957973387aSAl Viro path_put(&nd->stack[i].link); 59684a2bd39SAl Viro if (nd->flags & LOOKUP_ROOT_GRABBED) { 597102b8af2SAl Viro path_put(&nd->root); 59884a2bd39SAl Viro nd->flags &= ~LOOKUP_ROOT_GRABBED; 599102b8af2SAl Viro } 6007973387aSAl Viro } else { 6017973387aSAl Viro nd->flags &= ~LOOKUP_RCU; 6027973387aSAl Viro rcu_read_unlock(); 6037973387aSAl Viro } 6047973387aSAl Viro nd->depth = 0; 6057973387aSAl Viro } 6067973387aSAl Viro 6077973387aSAl Viro /* path_put is needed afterwards regardless of success or failure */ 6082aa38470SAl Viro static bool __legitimize_path(struct path *path, unsigned seq, unsigned mseq) 6097973387aSAl Viro { 6102aa38470SAl Viro int res = __legitimize_mnt(path->mnt, mseq); 6117973387aSAl Viro if (unlikely(res)) { 6127973387aSAl Viro if (res > 0) 6137973387aSAl Viro path->mnt = NULL; 6147973387aSAl Viro path->dentry = NULL; 6157973387aSAl Viro return false; 6167973387aSAl Viro } 6177973387aSAl Viro if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) { 6187973387aSAl Viro path->dentry = NULL; 6197973387aSAl Viro return false; 6207973387aSAl Viro } 6217973387aSAl Viro return !read_seqcount_retry(&path->dentry->d_seq, seq); 6227973387aSAl Viro } 6237973387aSAl Viro 6242aa38470SAl Viro static inline bool legitimize_path(struct nameidata *nd, 6252aa38470SAl Viro struct path *path, unsigned seq) 6262aa38470SAl Viro { 6275bd73286SAl Viro return __legitimize_path(path, seq, nd->m_seq); 6282aa38470SAl Viro } 6292aa38470SAl Viro 6307973387aSAl Viro static bool legitimize_links(struct nameidata *nd) 6317973387aSAl Viro { 6327973387aSAl Viro int i; 6337973387aSAl Viro for (i = 0; i < nd->depth; i++) { 6347973387aSAl Viro struct saved *last = nd->stack + i; 6357973387aSAl Viro if (unlikely(!legitimize_path(nd, &last->link, last->seq))) { 6367973387aSAl Viro drop_links(nd); 6377973387aSAl Viro nd->depth = i + 1; 6387973387aSAl Viro return false; 6397973387aSAl Viro } 6407973387aSAl Viro } 6417973387aSAl Viro return true; 6427973387aSAl Viro } 6437973387aSAl Viro 644ee594bffSAl Viro static bool legitimize_root(struct nameidata *nd) 645ee594bffSAl Viro { 646adb21d2bSAleksa Sarai /* 647adb21d2bSAleksa Sarai * For scoped-lookups (where nd->root has been zeroed), we need to 648adb21d2bSAleksa Sarai * restart the whole lookup from scratch -- because set_root() is wrong 649adb21d2bSAleksa Sarai * for these lookups (nd->dfd is the root, not the filesystem root). 650adb21d2bSAleksa Sarai */ 651adb21d2bSAleksa Sarai if (!nd->root.mnt && (nd->flags & LOOKUP_IS_SCOPED)) 652adb21d2bSAleksa Sarai return false; 653adb21d2bSAleksa Sarai /* Nothing to do if nd->root is zero or is managed by the VFS user. */ 654ee594bffSAl Viro if (!nd->root.mnt || (nd->flags & LOOKUP_ROOT)) 655ee594bffSAl Viro return true; 65684a2bd39SAl Viro nd->flags |= LOOKUP_ROOT_GRABBED; 657ee594bffSAl Viro return legitimize_path(nd, &nd->root, nd->root_seq); 658ee594bffSAl Viro } 659ee594bffSAl Viro 66019660af7SAl Viro /* 66131e6b01fSNick Piggin * Path walking has 2 modes, rcu-walk and ref-walk (see 66219660af7SAl Viro * Documentation/filesystems/path-lookup.txt). In situations when we can't 66319660af7SAl Viro * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab 66457e3715cSMike Marshall * normal reference counts on dentries and vfsmounts to transition to ref-walk 66519660af7SAl Viro * mode. Refcounts are grabbed at the last known good point before rcu-walk 66619660af7SAl Viro * got stuck, so ref-walk may continue from there. If this is not successful 66719660af7SAl Viro * (eg. a seqcount has changed), then failure is returned and it's up to caller 66819660af7SAl Viro * to restart the path walk from the beginning in ref-walk mode. 66931e6b01fSNick Piggin */ 67031e6b01fSNick Piggin 67131e6b01fSNick Piggin /** 67219660af7SAl Viro * unlazy_walk - try to switch to ref-walk mode. 67319660af7SAl Viro * @nd: nameidata pathwalk data 67439191628SRandy Dunlap * Returns: 0 on success, -ECHILD on failure 67531e6b01fSNick Piggin * 6764675ac39SAl Viro * unlazy_walk attempts to legitimize the current nd->path and nd->root 6774675ac39SAl Viro * for ref-walk mode. 6784675ac39SAl Viro * Must be called from rcu-walk context. 6797973387aSAl Viro * Nothing should touch nameidata between unlazy_walk() failure and 6807973387aSAl Viro * terminate_walk(). 68131e6b01fSNick Piggin */ 6824675ac39SAl Viro static int unlazy_walk(struct nameidata *nd) 68331e6b01fSNick Piggin { 68431e6b01fSNick Piggin struct dentry *parent = nd->path.dentry; 68531e6b01fSNick Piggin 68631e6b01fSNick Piggin BUG_ON(!(nd->flags & LOOKUP_RCU)); 687e5c832d5SLinus Torvalds 688e5c832d5SLinus Torvalds nd->flags &= ~LOOKUP_RCU; 6897973387aSAl Viro if (unlikely(!legitimize_links(nd))) 6904675ac39SAl Viro goto out1; 69184a2bd39SAl Viro if (unlikely(!legitimize_path(nd, &nd->path, nd->seq))) 69284a2bd39SAl Viro goto out; 693ee594bffSAl Viro if (unlikely(!legitimize_root(nd))) 6944675ac39SAl Viro goto out; 6954675ac39SAl Viro rcu_read_unlock(); 6964675ac39SAl Viro BUG_ON(nd->inode != parent->d_inode); 6974675ac39SAl Viro return 0; 6984675ac39SAl Viro 69984a2bd39SAl Viro out1: 7004675ac39SAl Viro nd->path.mnt = NULL; 7014675ac39SAl Viro nd->path.dentry = NULL; 7024675ac39SAl Viro out: 7034675ac39SAl Viro rcu_read_unlock(); 7044675ac39SAl Viro return -ECHILD; 7054675ac39SAl Viro } 7064675ac39SAl Viro 7074675ac39SAl Viro /** 7084675ac39SAl Viro * unlazy_child - try to switch to ref-walk mode. 7094675ac39SAl Viro * @nd: nameidata pathwalk data 7104675ac39SAl Viro * @dentry: child of nd->path.dentry 7114675ac39SAl Viro * @seq: seq number to check dentry against 7124675ac39SAl Viro * Returns: 0 on success, -ECHILD on failure 7134675ac39SAl Viro * 7144675ac39SAl Viro * unlazy_child attempts to legitimize the current nd->path, nd->root and dentry 7154675ac39SAl Viro * for ref-walk mode. @dentry must be a path found by a do_lookup call on 7164675ac39SAl Viro * @nd. Must be called from rcu-walk context. 7174675ac39SAl Viro * Nothing should touch nameidata between unlazy_child() failure and 7184675ac39SAl Viro * terminate_walk(). 7194675ac39SAl Viro */ 7204675ac39SAl Viro static int unlazy_child(struct nameidata *nd, struct dentry *dentry, unsigned seq) 7214675ac39SAl Viro { 7224675ac39SAl Viro BUG_ON(!(nd->flags & LOOKUP_RCU)); 7234675ac39SAl Viro 7244675ac39SAl Viro nd->flags &= ~LOOKUP_RCU; 7254675ac39SAl Viro if (unlikely(!legitimize_links(nd))) 7264675ac39SAl Viro goto out2; 7277973387aSAl Viro if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq))) 7287973387aSAl Viro goto out2; 7294675ac39SAl Viro if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref))) 7307973387aSAl Viro goto out1; 73148a066e7SAl Viro 73215570086SLinus Torvalds /* 7334675ac39SAl Viro * We need to move both the parent and the dentry from the RCU domain 7344675ac39SAl Viro * to be properly refcounted. And the sequence number in the dentry 7354675ac39SAl Viro * validates *both* dentry counters, since we checked the sequence 7364675ac39SAl Viro * number of the parent after we got the child sequence number. So we 7374675ac39SAl Viro * know the parent must still be valid if the child sequence number is 73815570086SLinus Torvalds */ 7394675ac39SAl Viro if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) 740e5c832d5SLinus Torvalds goto out; 74184a2bd39SAl Viro if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) 74284a2bd39SAl Viro goto out_dput; 743e5c832d5SLinus Torvalds /* 744e5c832d5SLinus Torvalds * Sequence counts matched. Now make sure that the root is 745e5c832d5SLinus Torvalds * still valid and get it if required. 746e5c832d5SLinus Torvalds */ 74784a2bd39SAl Viro if (unlikely(!legitimize_root(nd))) 74884a2bd39SAl Viro goto out_dput; 7498b61e74fSAl Viro rcu_read_unlock(); 75031e6b01fSNick Piggin return 0; 75119660af7SAl Viro 7527973387aSAl Viro out2: 7537973387aSAl Viro nd->path.mnt = NULL; 7547973387aSAl Viro out1: 7557973387aSAl Viro nd->path.dentry = NULL; 756e5c832d5SLinus Torvalds out: 7578b61e74fSAl Viro rcu_read_unlock(); 75884a2bd39SAl Viro return -ECHILD; 75984a2bd39SAl Viro out_dput: 76084a2bd39SAl Viro rcu_read_unlock(); 76184a2bd39SAl Viro dput(dentry); 76231e6b01fSNick Piggin return -ECHILD; 76331e6b01fSNick Piggin } 76431e6b01fSNick Piggin 7654ce16ef3SAl Viro static inline int d_revalidate(struct dentry *dentry, unsigned int flags) 76634286d66SNick Piggin { 767a89f8337SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) 7684ce16ef3SAl Viro return dentry->d_op->d_revalidate(dentry, flags); 769a89f8337SAl Viro else 770a89f8337SAl Viro return 1; 77134286d66SNick Piggin } 77234286d66SNick Piggin 7739f1fafeeSAl Viro /** 7749f1fafeeSAl Viro * complete_walk - successful completion of path walk 7759f1fafeeSAl Viro * @nd: pointer nameidata 77639159de2SJeff Layton * 7779f1fafeeSAl Viro * If we had been in RCU mode, drop out of it and legitimize nd->path. 7789f1fafeeSAl Viro * Revalidate the final result, unless we'd already done that during 7799f1fafeeSAl Viro * the path walk or the filesystem doesn't ask for it. Return 0 on 7809f1fafeeSAl Viro * success, -error on failure. In case of failure caller does not 7819f1fafeeSAl Viro * need to drop nd->path. 78239159de2SJeff Layton */ 7839f1fafeeSAl Viro static int complete_walk(struct nameidata *nd) 78439159de2SJeff Layton { 78516c2cd71SAl Viro struct dentry *dentry = nd->path.dentry; 78639159de2SJeff Layton int status; 78739159de2SJeff Layton 7889f1fafeeSAl Viro if (nd->flags & LOOKUP_RCU) { 789adb21d2bSAleksa Sarai /* 790adb21d2bSAleksa Sarai * We don't want to zero nd->root for scoped-lookups or 791adb21d2bSAleksa Sarai * externally-managed nd->root. 792adb21d2bSAleksa Sarai */ 793adb21d2bSAleksa Sarai if (!(nd->flags & (LOOKUP_ROOT | LOOKUP_IS_SCOPED))) 7949f1fafeeSAl Viro nd->root.mnt = NULL; 7954675ac39SAl Viro if (unlikely(unlazy_walk(nd))) 79648a066e7SAl Viro return -ECHILD; 79748a066e7SAl Viro } 7989f1fafeeSAl Viro 799adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) { 800adb21d2bSAleksa Sarai /* 801adb21d2bSAleksa Sarai * While the guarantee of LOOKUP_IS_SCOPED is (roughly) "don't 802adb21d2bSAleksa Sarai * ever step outside the root during lookup" and should already 803adb21d2bSAleksa Sarai * be guaranteed by the rest of namei, we want to avoid a namei 804adb21d2bSAleksa Sarai * BUG resulting in userspace being given a path that was not 805adb21d2bSAleksa Sarai * scoped within the root at some point during the lookup. 806adb21d2bSAleksa Sarai * 807adb21d2bSAleksa Sarai * So, do a final sanity-check to make sure that in the 808adb21d2bSAleksa Sarai * worst-case scenario (a complete bypass of LOOKUP_IS_SCOPED) 809adb21d2bSAleksa Sarai * we won't silently return an fd completely outside of the 810adb21d2bSAleksa Sarai * requested root to userspace. 811adb21d2bSAleksa Sarai * 812adb21d2bSAleksa Sarai * Userspace could move the path outside the root after this 813adb21d2bSAleksa Sarai * check, but as discussed elsewhere this is not a concern (the 814adb21d2bSAleksa Sarai * resolved file was inside the root at some point). 815adb21d2bSAleksa Sarai */ 816adb21d2bSAleksa Sarai if (!path_is_under(&nd->path, &nd->root)) 817adb21d2bSAleksa Sarai return -EXDEV; 818adb21d2bSAleksa Sarai } 819adb21d2bSAleksa Sarai 82016c2cd71SAl Viro if (likely(!(nd->flags & LOOKUP_JUMPED))) 82139159de2SJeff Layton return 0; 82239159de2SJeff Layton 823ecf3d1f1SJeff Layton if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) 82416c2cd71SAl Viro return 0; 82516c2cd71SAl Viro 826ecf3d1f1SJeff Layton status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); 82739159de2SJeff Layton if (status > 0) 82839159de2SJeff Layton return 0; 82939159de2SJeff Layton 83016c2cd71SAl Viro if (!status) 83139159de2SJeff Layton status = -ESTALE; 83216c2cd71SAl Viro 83339159de2SJeff Layton return status; 83439159de2SJeff Layton } 83539159de2SJeff Layton 836740a1678SAleksa Sarai static int set_root(struct nameidata *nd) 8372a737871SAl Viro { 83831e6b01fSNick Piggin struct fs_struct *fs = current->fs; 8399e6697e2SAl Viro 840adb21d2bSAleksa Sarai /* 841adb21d2bSAleksa Sarai * Jumping to the real root in a scoped-lookup is a BUG in namei, but we 842adb21d2bSAleksa Sarai * still have to ensure it doesn't happen because it will cause a breakout 843adb21d2bSAleksa Sarai * from the dirfd. 844adb21d2bSAleksa Sarai */ 845adb21d2bSAleksa Sarai if (WARN_ON(nd->flags & LOOKUP_IS_SCOPED)) 846adb21d2bSAleksa Sarai return -ENOTRECOVERABLE; 847adb21d2bSAleksa Sarai 8489e6697e2SAl Viro if (nd->flags & LOOKUP_RCU) { 8498f47a016SAl Viro unsigned seq; 850c28cc364SNick Piggin 851c28cc364SNick Piggin do { 852c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 85331e6b01fSNick Piggin nd->root = fs->root; 8548f47a016SAl Viro nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq); 855c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 8569e6697e2SAl Viro } else { 8579e6697e2SAl Viro get_fs_root(fs, &nd->root); 85884a2bd39SAl Viro nd->flags |= LOOKUP_ROOT_GRABBED; 8599e6697e2SAl Viro } 860740a1678SAleksa Sarai return 0; 86131e6b01fSNick Piggin } 86231e6b01fSNick Piggin 863248fb5b9SAl Viro static int nd_jump_root(struct nameidata *nd) 864248fb5b9SAl Viro { 865adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_BENEATH)) 866adb21d2bSAleksa Sarai return -EXDEV; 86772ba2929SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_XDEV)) { 86872ba2929SAleksa Sarai /* Absolute path arguments to path_init() are allowed. */ 86972ba2929SAleksa Sarai if (nd->path.mnt != NULL && nd->path.mnt != nd->root.mnt) 87072ba2929SAleksa Sarai return -EXDEV; 87172ba2929SAleksa Sarai } 872740a1678SAleksa Sarai if (!nd->root.mnt) { 873740a1678SAleksa Sarai int error = set_root(nd); 874740a1678SAleksa Sarai if (error) 875740a1678SAleksa Sarai return error; 876740a1678SAleksa Sarai } 877248fb5b9SAl Viro if (nd->flags & LOOKUP_RCU) { 878248fb5b9SAl Viro struct dentry *d; 879248fb5b9SAl Viro nd->path = nd->root; 880248fb5b9SAl Viro d = nd->path.dentry; 881248fb5b9SAl Viro nd->inode = d->d_inode; 882248fb5b9SAl Viro nd->seq = nd->root_seq; 883248fb5b9SAl Viro if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq))) 884248fb5b9SAl Viro return -ECHILD; 885248fb5b9SAl Viro } else { 886248fb5b9SAl Viro path_put(&nd->path); 887248fb5b9SAl Viro nd->path = nd->root; 888248fb5b9SAl Viro path_get(&nd->path); 889248fb5b9SAl Viro nd->inode = nd->path.dentry->d_inode; 890248fb5b9SAl Viro } 891248fb5b9SAl Viro nd->flags |= LOOKUP_JUMPED; 892248fb5b9SAl Viro return 0; 893248fb5b9SAl Viro } 894248fb5b9SAl Viro 895b5fb63c1SChristoph Hellwig /* 8966b255391SAl Viro * Helper to directly jump to a known parsed path from ->get_link, 897b5fb63c1SChristoph Hellwig * caller must have taken a reference to path beforehand. 898b5fb63c1SChristoph Hellwig */ 8991bc82070SAleksa Sarai int nd_jump_link(struct path *path) 900b5fb63c1SChristoph Hellwig { 9014b99d499SAleksa Sarai int error = -ELOOP; 9026e77137bSAl Viro struct nameidata *nd = current->nameidata; 903b5fb63c1SChristoph Hellwig 9044b99d499SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_MAGICLINKS)) 9054b99d499SAleksa Sarai goto err; 9064b99d499SAleksa Sarai 90772ba2929SAleksa Sarai error = -EXDEV; 90872ba2929SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_XDEV)) { 90972ba2929SAleksa Sarai if (nd->path.mnt != path->mnt) 91072ba2929SAleksa Sarai goto err; 91172ba2929SAleksa Sarai } 912adb21d2bSAleksa Sarai /* Not currently safe for scoped-lookups. */ 913adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) 914adb21d2bSAleksa Sarai goto err; 91572ba2929SAleksa Sarai 9164b99d499SAleksa Sarai path_put(&nd->path); 917b5fb63c1SChristoph Hellwig nd->path = *path; 918b5fb63c1SChristoph Hellwig nd->inode = nd->path.dentry->d_inode; 919b5fb63c1SChristoph Hellwig nd->flags |= LOOKUP_JUMPED; 9201bc82070SAleksa Sarai return 0; 9214b99d499SAleksa Sarai 9224b99d499SAleksa Sarai err: 9234b99d499SAleksa Sarai path_put(path); 9244b99d499SAleksa Sarai return error; 925b5fb63c1SChristoph Hellwig } 926b5fb63c1SChristoph Hellwig 927b9ff4429SAl Viro static inline void put_link(struct nameidata *nd) 928574197e0SAl Viro { 92921c3003dSAl Viro struct saved *last = nd->stack + --nd->depth; 930fceef393SAl Viro do_delayed_call(&last->done); 9316548fae2SAl Viro if (!(nd->flags & LOOKUP_RCU)) 932b9ff4429SAl Viro path_put(&last->link); 933574197e0SAl Viro } 934574197e0SAl Viro 935561ec64aSLinus Torvalds int sysctl_protected_symlinks __read_mostly = 0; 936561ec64aSLinus Torvalds int sysctl_protected_hardlinks __read_mostly = 0; 93730aba665SSalvatore Mesoraca int sysctl_protected_fifos __read_mostly; 93830aba665SSalvatore Mesoraca int sysctl_protected_regular __read_mostly; 939800179c9SKees Cook 940800179c9SKees Cook /** 941800179c9SKees Cook * may_follow_link - Check symlink following for unsafe situations 94255852635SRandy Dunlap * @nd: nameidata pathwalk data 943800179c9SKees Cook * 944800179c9SKees Cook * In the case of the sysctl_protected_symlinks sysctl being enabled, 945800179c9SKees Cook * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is 946800179c9SKees Cook * in a sticky world-writable directory. This is to protect privileged 947800179c9SKees Cook * processes from failing races against path names that may change out 948800179c9SKees Cook * from under them by way of other users creating malicious symlinks. 949800179c9SKees Cook * It will permit symlinks to be followed only when outside a sticky 950800179c9SKees Cook * world-writable directory, or when the uid of the symlink and follower 951800179c9SKees Cook * match, or when the directory owner matches the symlink's owner. 952800179c9SKees Cook * 953800179c9SKees Cook * Returns 0 if following the symlink is allowed, -ve on error. 954800179c9SKees Cook */ 955ad6cc4c3SAl Viro static inline int may_follow_link(struct nameidata *nd, const struct inode *inode) 956800179c9SKees Cook { 957800179c9SKees Cook if (!sysctl_protected_symlinks) 958800179c9SKees Cook return 0; 959800179c9SKees Cook 960800179c9SKees Cook /* Allowed if owner and follower match. */ 96181abe27bSEric W. Biederman if (uid_eq(current_cred()->fsuid, inode->i_uid)) 962800179c9SKees Cook return 0; 963800179c9SKees Cook 964800179c9SKees Cook /* Allowed if parent directory not sticky and world-writable. */ 9650f705953SAl Viro if ((nd->dir_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) 966800179c9SKees Cook return 0; 967800179c9SKees Cook 968800179c9SKees Cook /* Allowed if parent directory and link owner match. */ 9690f705953SAl Viro if (uid_valid(nd->dir_uid) && uid_eq(nd->dir_uid, inode->i_uid)) 970800179c9SKees Cook return 0; 971800179c9SKees Cook 97231956502SAl Viro if (nd->flags & LOOKUP_RCU) 97331956502SAl Viro return -ECHILD; 97431956502SAl Viro 975ea841bafSRichard Guy Briggs audit_inode(nd->name, nd->stack[0].link.dentry, 0); 976245d7369SKees Cook audit_log_path_denied(AUDIT_ANOM_LINK, "follow_link"); 977800179c9SKees Cook return -EACCES; 978800179c9SKees Cook } 979800179c9SKees Cook 980800179c9SKees Cook /** 981800179c9SKees Cook * safe_hardlink_source - Check for safe hardlink conditions 982800179c9SKees Cook * @inode: the source inode to hardlink from 983800179c9SKees Cook * 984800179c9SKees Cook * Return false if at least one of the following conditions: 985800179c9SKees Cook * - inode is not a regular file 986800179c9SKees Cook * - inode is setuid 987800179c9SKees Cook * - inode is setgid and group-exec 988800179c9SKees Cook * - access failure for read and write 989800179c9SKees Cook * 990800179c9SKees Cook * Otherwise returns true. 991800179c9SKees Cook */ 992800179c9SKees Cook static bool safe_hardlink_source(struct inode *inode) 993800179c9SKees Cook { 994800179c9SKees Cook umode_t mode = inode->i_mode; 995800179c9SKees Cook 996800179c9SKees Cook /* Special files should not get pinned to the filesystem. */ 997800179c9SKees Cook if (!S_ISREG(mode)) 998800179c9SKees Cook return false; 999800179c9SKees Cook 1000800179c9SKees Cook /* Setuid files should not get pinned to the filesystem. */ 1001800179c9SKees Cook if (mode & S_ISUID) 1002800179c9SKees Cook return false; 1003800179c9SKees Cook 1004800179c9SKees Cook /* Executable setgid files should not get pinned to the filesystem. */ 1005800179c9SKees Cook if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) 1006800179c9SKees Cook return false; 1007800179c9SKees Cook 1008800179c9SKees Cook /* Hardlinking to unreadable or unwritable sources is dangerous. */ 1009800179c9SKees Cook if (inode_permission(inode, MAY_READ | MAY_WRITE)) 1010800179c9SKees Cook return false; 1011800179c9SKees Cook 1012800179c9SKees Cook return true; 1013800179c9SKees Cook } 1014800179c9SKees Cook 1015800179c9SKees Cook /** 1016800179c9SKees Cook * may_linkat - Check permissions for creating a hardlink 1017800179c9SKees Cook * @link: the source to hardlink from 1018800179c9SKees Cook * 1019800179c9SKees Cook * Block hardlink when all of: 1020800179c9SKees Cook * - sysctl_protected_hardlinks enabled 1021800179c9SKees Cook * - fsuid does not match inode 1022800179c9SKees Cook * - hardlink source is unsafe (see safe_hardlink_source() above) 1023f2ca3796SDirk Steinmetz * - not CAP_FOWNER in a namespace with the inode owner uid mapped 1024800179c9SKees Cook * 1025800179c9SKees Cook * Returns 0 if successful, -ve on error. 1026800179c9SKees Cook */ 1027812931d6SChristoph Hellwig int may_linkat(struct path *link) 1028800179c9SKees Cook { 1029593d1ce8SEric W. Biederman struct inode *inode = link->dentry->d_inode; 1030593d1ce8SEric W. Biederman 1031593d1ce8SEric W. Biederman /* Inode writeback is not safe when the uid or gid are invalid. */ 1032593d1ce8SEric W. Biederman if (!uid_valid(inode->i_uid) || !gid_valid(inode->i_gid)) 1033593d1ce8SEric W. Biederman return -EOVERFLOW; 1034800179c9SKees Cook 1035800179c9SKees Cook if (!sysctl_protected_hardlinks) 1036800179c9SKees Cook return 0; 1037800179c9SKees Cook 1038800179c9SKees Cook /* Source inode owner (or CAP_FOWNER) can hardlink all they like, 1039800179c9SKees Cook * otherwise, it must be a safe source. 1040800179c9SKees Cook */ 1041cc658db4SKees Cook if (safe_hardlink_source(inode) || inode_owner_or_capable(inode)) 1042800179c9SKees Cook return 0; 1043800179c9SKees Cook 1044245d7369SKees Cook audit_log_path_denied(AUDIT_ANOM_LINK, "linkat"); 1045800179c9SKees Cook return -EPERM; 1046800179c9SKees Cook } 1047800179c9SKees Cook 104830aba665SSalvatore Mesoraca /** 104930aba665SSalvatore Mesoraca * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory 105030aba665SSalvatore Mesoraca * should be allowed, or not, on files that already 105130aba665SSalvatore Mesoraca * exist. 1052d0cb5018SAl Viro * @dir_mode: mode bits of directory 1053d0cb5018SAl Viro * @dir_uid: owner of directory 105430aba665SSalvatore Mesoraca * @inode: the inode of the file to open 105530aba665SSalvatore Mesoraca * 105630aba665SSalvatore Mesoraca * Block an O_CREAT open of a FIFO (or a regular file) when: 105730aba665SSalvatore Mesoraca * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled 105830aba665SSalvatore Mesoraca * - the file already exists 105930aba665SSalvatore Mesoraca * - we are in a sticky directory 106030aba665SSalvatore Mesoraca * - we don't own the file 106130aba665SSalvatore Mesoraca * - the owner of the directory doesn't own the file 106230aba665SSalvatore Mesoraca * - the directory is world writable 106330aba665SSalvatore Mesoraca * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2 106430aba665SSalvatore Mesoraca * the directory doesn't have to be world writable: being group writable will 106530aba665SSalvatore Mesoraca * be enough. 106630aba665SSalvatore Mesoraca * 106730aba665SSalvatore Mesoraca * Returns 0 if the open is allowed, -ve on error. 106830aba665SSalvatore Mesoraca */ 1069d0cb5018SAl Viro static int may_create_in_sticky(umode_t dir_mode, kuid_t dir_uid, 107030aba665SSalvatore Mesoraca struct inode * const inode) 107130aba665SSalvatore Mesoraca { 107230aba665SSalvatore Mesoraca if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) || 107330aba665SSalvatore Mesoraca (!sysctl_protected_regular && S_ISREG(inode->i_mode)) || 1074d0cb5018SAl Viro likely(!(dir_mode & S_ISVTX)) || 1075d0cb5018SAl Viro uid_eq(inode->i_uid, dir_uid) || 107630aba665SSalvatore Mesoraca uid_eq(current_fsuid(), inode->i_uid)) 107730aba665SSalvatore Mesoraca return 0; 107830aba665SSalvatore Mesoraca 1079d0cb5018SAl Viro if (likely(dir_mode & 0002) || 1080d0cb5018SAl Viro (dir_mode & 0020 && 108130aba665SSalvatore Mesoraca ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) || 108230aba665SSalvatore Mesoraca (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) { 1083245d7369SKees Cook const char *operation = S_ISFIFO(inode->i_mode) ? 1084245d7369SKees Cook "sticky_create_fifo" : 1085245d7369SKees Cook "sticky_create_regular"; 1086245d7369SKees Cook audit_log_path_denied(AUDIT_ANOM_CREAT, operation); 108730aba665SSalvatore Mesoraca return -EACCES; 108830aba665SSalvatore Mesoraca } 108930aba665SSalvatore Mesoraca return 0; 109030aba665SSalvatore Mesoraca } 109130aba665SSalvatore Mesoraca 1092f015f126SDavid Howells /* 1093f015f126SDavid Howells * follow_up - Find the mountpoint of path's vfsmount 1094f015f126SDavid Howells * 1095f015f126SDavid Howells * Given a path, find the mountpoint of its source file system. 1096f015f126SDavid Howells * Replace @path with the path of the mountpoint in the parent mount. 1097f015f126SDavid Howells * Up is towards /. 1098f015f126SDavid Howells * 1099f015f126SDavid Howells * Return 1 if we went up a level and 0 if we were already at the 1100f015f126SDavid Howells * root. 1101f015f126SDavid Howells */ 1102bab77ebfSAl Viro int follow_up(struct path *path) 11031da177e4SLinus Torvalds { 11040714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 11050714a533SAl Viro struct mount *parent; 11061da177e4SLinus Torvalds struct dentry *mountpoint; 110799b7db7bSNick Piggin 110848a066e7SAl Viro read_seqlock_excl(&mount_lock); 11090714a533SAl Viro parent = mnt->mnt_parent; 11103c0a6163SAl Viro if (parent == mnt) { 111148a066e7SAl Viro read_sequnlock_excl(&mount_lock); 11121da177e4SLinus Torvalds return 0; 11131da177e4SLinus Torvalds } 11140714a533SAl Viro mntget(&parent->mnt); 1115a73324daSAl Viro mountpoint = dget(mnt->mnt_mountpoint); 111648a066e7SAl Viro read_sequnlock_excl(&mount_lock); 1117bab77ebfSAl Viro dput(path->dentry); 1118bab77ebfSAl Viro path->dentry = mountpoint; 1119bab77ebfSAl Viro mntput(path->mnt); 11200714a533SAl Viro path->mnt = &parent->mnt; 11211da177e4SLinus Torvalds return 1; 11221da177e4SLinus Torvalds } 11234d359507SAl Viro EXPORT_SYMBOL(follow_up); 11241da177e4SLinus Torvalds 11257ef482faSAl Viro static bool choose_mountpoint_rcu(struct mount *m, const struct path *root, 11267ef482faSAl Viro struct path *path, unsigned *seqp) 11277ef482faSAl Viro { 11287ef482faSAl Viro while (mnt_has_parent(m)) { 11297ef482faSAl Viro struct dentry *mountpoint = m->mnt_mountpoint; 11307ef482faSAl Viro 11317ef482faSAl Viro m = m->mnt_parent; 11327ef482faSAl Viro if (unlikely(root->dentry == mountpoint && 11337ef482faSAl Viro root->mnt == &m->mnt)) 11347ef482faSAl Viro break; 11357ef482faSAl Viro if (mountpoint != m->mnt.mnt_root) { 11367ef482faSAl Viro path->mnt = &m->mnt; 11377ef482faSAl Viro path->dentry = mountpoint; 11387ef482faSAl Viro *seqp = read_seqcount_begin(&mountpoint->d_seq); 11397ef482faSAl Viro return true; 11407ef482faSAl Viro } 11417ef482faSAl Viro } 11427ef482faSAl Viro return false; 11437ef482faSAl Viro } 11447ef482faSAl Viro 11452aa38470SAl Viro static bool choose_mountpoint(struct mount *m, const struct path *root, 11462aa38470SAl Viro struct path *path) 11472aa38470SAl Viro { 11482aa38470SAl Viro bool found; 11492aa38470SAl Viro 11502aa38470SAl Viro rcu_read_lock(); 11512aa38470SAl Viro while (1) { 11522aa38470SAl Viro unsigned seq, mseq = read_seqbegin(&mount_lock); 11532aa38470SAl Viro 11542aa38470SAl Viro found = choose_mountpoint_rcu(m, root, path, &seq); 11552aa38470SAl Viro if (unlikely(!found)) { 11562aa38470SAl Viro if (!read_seqretry(&mount_lock, mseq)) 11572aa38470SAl Viro break; 11582aa38470SAl Viro } else { 11592aa38470SAl Viro if (likely(__legitimize_path(path, seq, mseq))) 11602aa38470SAl Viro break; 11612aa38470SAl Viro rcu_read_unlock(); 11622aa38470SAl Viro path_put(path); 11632aa38470SAl Viro rcu_read_lock(); 11642aa38470SAl Viro } 11652aa38470SAl Viro } 11662aa38470SAl Viro rcu_read_unlock(); 11672aa38470SAl Viro return found; 11682aa38470SAl Viro } 11692aa38470SAl Viro 1170b5c84bf6SNick Piggin /* 11719875cf80SDavid Howells * Perform an automount 11729875cf80SDavid Howells * - return -EISDIR to tell follow_managed() to stop and return the path we 11739875cf80SDavid Howells * were called with. 11741da177e4SLinus Torvalds */ 11751c9f5e06SAl Viro static int follow_automount(struct path *path, int *count, unsigned lookup_flags) 117631e6b01fSNick Piggin { 117725e195aaSAl Viro struct dentry *dentry = path->dentry; 11789875cf80SDavid Howells 11790ec26fd0SMiklos Szeredi /* We don't want to mount if someone's just doing a stat - 11800ec26fd0SMiklos Szeredi * unless they're stat'ing a directory and appended a '/' to 11810ec26fd0SMiklos Szeredi * the name. 11820ec26fd0SMiklos Szeredi * 11830ec26fd0SMiklos Szeredi * We do, however, want to mount if someone wants to open or 11845a30d8a2SDavid Howells * create a file of any type under the mountpoint, wants to 11850ec26fd0SMiklos Szeredi * traverse through the mountpoint or wants to open the 11860ec26fd0SMiklos Szeredi * mounted directory. Also, autofs may mark negative dentries 11870ec26fd0SMiklos Szeredi * as being automount points. These will need the attentions 11880ec26fd0SMiklos Szeredi * of the daemon to instantiate them before they can be used. 11895a30d8a2SDavid Howells */ 11901c9f5e06SAl Viro if (!(lookup_flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | 11915d38f049SIan Kent LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && 119225e195aaSAl Viro dentry->d_inode) 11939875cf80SDavid Howells return -EISDIR; 11940ec26fd0SMiklos Szeredi 11951c9f5e06SAl Viro if (count && (*count)++ >= MAXSYMLINKS) 11969875cf80SDavid Howells return -ELOOP; 11979875cf80SDavid Howells 119825e195aaSAl Viro return finish_automount(dentry->d_op->d_automount(path), path); 1199ea5b778aSDavid Howells } 12009875cf80SDavid Howells 12019875cf80SDavid Howells /* 12029deed3ebSAl Viro * mount traversal - out-of-line part. One note on ->d_flags accesses - 12039deed3ebSAl Viro * dentries are pinned but not locked here, so negative dentry can go 12049deed3ebSAl Viro * positive right under us. Use of smp_load_acquire() provides a barrier 12059deed3ebSAl Viro * sufficient for ->d_inode and ->d_flags consistency. 12069875cf80SDavid Howells */ 12079deed3ebSAl Viro static int __traverse_mounts(struct path *path, unsigned flags, bool *jumped, 12089deed3ebSAl Viro int *count, unsigned lookup_flags) 12099875cf80SDavid Howells { 12109deed3ebSAl Viro struct vfsmount *mnt = path->mnt; 12119875cf80SDavid Howells bool need_mntput = false; 12128aef1884SAl Viro int ret = 0; 12139875cf80SDavid Howells 12149deed3ebSAl Viro while (flags & DCACHE_MANAGED_DENTRY) { 1215cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1216cc53ce53SDavid Howells * being held. */ 1217d41efb52SAl Viro if (flags & DCACHE_MANAGE_TRANSIT) { 1218fb5f51c7SIan Kent ret = path->dentry->d_op->d_manage(path, false); 1219508c8772SAl Viro flags = smp_load_acquire(&path->dentry->d_flags); 1220cc53ce53SDavid Howells if (ret < 0) 12218aef1884SAl Viro break; 1222cc53ce53SDavid Howells } 1223cc53ce53SDavid Howells 12249deed3ebSAl Viro if (flags & DCACHE_MOUNTED) { // something's mounted on it.. 12259875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 12269deed3ebSAl Viro if (mounted) { // ... in our namespace 12279875cf80SDavid Howells dput(path->dentry); 12289875cf80SDavid Howells if (need_mntput) 1229463ffb2eSAl Viro mntput(path->mnt); 1230463ffb2eSAl Viro path->mnt = mounted; 1231463ffb2eSAl Viro path->dentry = dget(mounted->mnt_root); 12329deed3ebSAl Viro // here we know it's positive 12339deed3ebSAl Viro flags = path->dentry->d_flags; 12349875cf80SDavid Howells need_mntput = true; 12359875cf80SDavid Howells continue; 1236463ffb2eSAl Viro } 12371da177e4SLinus Torvalds } 12389875cf80SDavid Howells 12399deed3ebSAl Viro if (!(flags & DCACHE_NEED_AUTOMOUNT)) 12409deed3ebSAl Viro break; 12419deed3ebSAl Viro 12429deed3ebSAl Viro // uncovered automount point 12439deed3ebSAl Viro ret = follow_automount(path, count, lookup_flags); 12449deed3ebSAl Viro flags = smp_load_acquire(&path->dentry->d_flags); 12459875cf80SDavid Howells if (ret < 0) 12468aef1884SAl Viro break; 12479875cf80SDavid Howells } 12489875cf80SDavid Howells 12499deed3ebSAl Viro if (ret == -EISDIR) 12509deed3ebSAl Viro ret = 0; 12519deed3ebSAl Viro // possible if you race with several mount --move 12529deed3ebSAl Viro if (need_mntput && path->mnt == mnt) 12538aef1884SAl Viro mntput(path->mnt); 12549deed3ebSAl Viro if (!ret && unlikely(d_flags_negative(flags))) 1255d41efb52SAl Viro ret = -ENOENT; 12569deed3ebSAl Viro *jumped = need_mntput; 12578402752eSAl Viro return ret; 12581da177e4SLinus Torvalds } 12591da177e4SLinus Torvalds 12609deed3ebSAl Viro static inline int traverse_mounts(struct path *path, bool *jumped, 12619deed3ebSAl Viro int *count, unsigned lookup_flags) 12629deed3ebSAl Viro { 12639deed3ebSAl Viro unsigned flags = smp_load_acquire(&path->dentry->d_flags); 12649deed3ebSAl Viro 12659deed3ebSAl Viro /* fastpath */ 12669deed3ebSAl Viro if (likely(!(flags & DCACHE_MANAGED_DENTRY))) { 12679deed3ebSAl Viro *jumped = false; 12689deed3ebSAl Viro if (unlikely(d_flags_negative(flags))) 12699deed3ebSAl Viro return -ENOENT; 12709deed3ebSAl Viro return 0; 12719deed3ebSAl Viro } 12729deed3ebSAl Viro return __traverse_mounts(path, flags, jumped, count, lookup_flags); 12739deed3ebSAl Viro } 12749deed3ebSAl Viro 1275cc53ce53SDavid Howells int follow_down_one(struct path *path) 12761da177e4SLinus Torvalds { 12771da177e4SLinus Torvalds struct vfsmount *mounted; 12781da177e4SLinus Torvalds 12791c755af4SAl Viro mounted = lookup_mnt(path); 12801da177e4SLinus Torvalds if (mounted) { 12819393bd07SAl Viro dput(path->dentry); 12829393bd07SAl Viro mntput(path->mnt); 12839393bd07SAl Viro path->mnt = mounted; 12849393bd07SAl Viro path->dentry = dget(mounted->mnt_root); 12851da177e4SLinus Torvalds return 1; 12861da177e4SLinus Torvalds } 12871da177e4SLinus Torvalds return 0; 12881da177e4SLinus Torvalds } 12894d359507SAl Viro EXPORT_SYMBOL(follow_down_one); 12901da177e4SLinus Torvalds 12919875cf80SDavid Howells /* 12929deed3ebSAl Viro * Follow down to the covering mount currently visible to userspace. At each 12939deed3ebSAl Viro * point, the filesystem owning that dentry may be queried as to whether the 12949deed3ebSAl Viro * caller is permitted to proceed or not. 12959deed3ebSAl Viro */ 12969deed3ebSAl Viro int follow_down(struct path *path) 12979deed3ebSAl Viro { 12989deed3ebSAl Viro struct vfsmount *mnt = path->mnt; 12999deed3ebSAl Viro bool jumped; 13009deed3ebSAl Viro int ret = traverse_mounts(path, &jumped, NULL, 0); 13019deed3ebSAl Viro 13029deed3ebSAl Viro if (path->mnt != mnt) 13039deed3ebSAl Viro mntput(mnt); 13049deed3ebSAl Viro return ret; 13059deed3ebSAl Viro } 13069deed3ebSAl Viro EXPORT_SYMBOL(follow_down); 13079deed3ebSAl Viro 13089deed3ebSAl Viro /* 1309287548e4SAl Viro * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if 1310287548e4SAl Viro * we meet a managed dentry that would need blocking. 13119875cf80SDavid Howells */ 13129875cf80SDavid Howells static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, 1313254cf582SAl Viro struct inode **inode, unsigned *seqp) 13149875cf80SDavid Howells { 1315ea936aebSAl Viro struct dentry *dentry = path->dentry; 1316ea936aebSAl Viro unsigned int flags = dentry->d_flags; 1317ea936aebSAl Viro 1318ea936aebSAl Viro if (likely(!(flags & DCACHE_MANAGED_DENTRY))) 1319ea936aebSAl Viro return true; 1320ea936aebSAl Viro 1321ea936aebSAl Viro if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 1322ea936aebSAl Viro return false; 1323ea936aebSAl Viro 132462a7375eSIan Kent for (;;) { 132562a7375eSIan Kent /* 132662a7375eSIan Kent * Don't forget we might have a non-mountpoint managed dentry 132762a7375eSIan Kent * that wants to block transit. 132862a7375eSIan Kent */ 1329ea936aebSAl Viro if (unlikely(flags & DCACHE_MANAGE_TRANSIT)) { 1330ea936aebSAl Viro int res = dentry->d_op->d_manage(path, true); 1331ea936aebSAl Viro if (res) 1332ea936aebSAl Viro return res == -EISDIR; 1333ea936aebSAl Viro flags = dentry->d_flags; 1334b8faf035SNeilBrown } 133562a7375eSIan Kent 1336ea936aebSAl Viro if (flags & DCACHE_MOUNTED) { 1337ea936aebSAl Viro struct mount *mounted = __lookup_mnt(path->mnt, dentry); 1338ea936aebSAl Viro if (mounted) { 1339c7105365SAl Viro path->mnt = &mounted->mnt; 1340ea936aebSAl Viro dentry = path->dentry = mounted->mnt.mnt_root; 1341a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 1342ea936aebSAl Viro *seqp = read_seqcount_begin(&dentry->d_seq); 1343ea936aebSAl Viro *inode = dentry->d_inode; 134459430262SLinus Torvalds /* 1345ea936aebSAl Viro * We don't need to re-check ->d_seq after this 1346ea936aebSAl Viro * ->d_inode read - there will be an RCU delay 1347ea936aebSAl Viro * between mount hash removal and ->mnt_root 1348ea936aebSAl Viro * becoming unpinned. 134959430262SLinus Torvalds */ 1350ea936aebSAl Viro flags = dentry->d_flags; 1351ea936aebSAl Viro continue; 13529875cf80SDavid Howells } 1353ea936aebSAl Viro if (read_seqretry(&mount_lock, nd->m_seq)) 1354ea936aebSAl Viro return false; 1355ea936aebSAl Viro } 1356ea936aebSAl Viro return !(flags & DCACHE_NEED_AUTOMOUNT); 1357ea936aebSAl Viro } 1358287548e4SAl Viro } 1359287548e4SAl Viro 1360db3c9adeSAl Viro static inline int handle_mounts(struct nameidata *nd, struct dentry *dentry, 1361db3c9adeSAl Viro struct path *path, struct inode **inode, 1362db3c9adeSAl Viro unsigned int *seqp) 1363bd7c4b50SAl Viro { 13649deed3ebSAl Viro bool jumped; 1365db3c9adeSAl Viro int ret; 1366bd7c4b50SAl Viro 1367db3c9adeSAl Viro path->mnt = nd->path.mnt; 1368db3c9adeSAl Viro path->dentry = dentry; 1369c153007bSAl Viro if (nd->flags & LOOKUP_RCU) { 1370c153007bSAl Viro unsigned int seq = *seqp; 1371c153007bSAl Viro if (unlikely(!*inode)) 1372c153007bSAl Viro return -ENOENT; 1373c153007bSAl Viro if (likely(__follow_mount_rcu(nd, path, inode, seqp))) 13749deed3ebSAl Viro return 0; 1375c153007bSAl Viro if (unlazy_child(nd, dentry, seq)) 1376c153007bSAl Viro return -ECHILD; 1377c153007bSAl Viro // *path might've been clobbered by __follow_mount_rcu() 1378c153007bSAl Viro path->mnt = nd->path.mnt; 1379c153007bSAl Viro path->dentry = dentry; 1380c153007bSAl Viro } 13819deed3ebSAl Viro ret = traverse_mounts(path, &jumped, &nd->total_link_count, nd->flags); 13829deed3ebSAl Viro if (jumped) { 13839deed3ebSAl Viro if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 13849deed3ebSAl Viro ret = -EXDEV; 13859deed3ebSAl Viro else 13869deed3ebSAl Viro nd->flags |= LOOKUP_JUMPED; 13879deed3ebSAl Viro } 13889deed3ebSAl Viro if (unlikely(ret)) { 13899deed3ebSAl Viro dput(path->dentry); 13909deed3ebSAl Viro if (path->mnt != nd->path.mnt) 13919deed3ebSAl Viro mntput(path->mnt); 13929deed3ebSAl Viro } else { 1393bd7c4b50SAl Viro *inode = d_backing_inode(path->dentry); 1394bd7c4b50SAl Viro *seqp = 0; /* out of RCU mode, so the value doesn't matter */ 1395bd7c4b50SAl Viro } 1396bd7c4b50SAl Viro return ret; 1397bd7c4b50SAl Viro } 1398bd7c4b50SAl Viro 13999875cf80SDavid Howells /* 1400f4fdace9SOleg Drokin * This looks up the name in dcache and possibly revalidates the found dentry. 1401f4fdace9SOleg Drokin * NULL is returned if the dentry does not exist in the cache. 1402baa03890SNick Piggin */ 1403e3c13928SAl Viro static struct dentry *lookup_dcache(const struct qstr *name, 1404e3c13928SAl Viro struct dentry *dir, 14056c51e513SAl Viro unsigned int flags) 1406baa03890SNick Piggin { 1407a89f8337SAl Viro struct dentry *dentry = d_lookup(dir, name); 1408bad61189SMiklos Szeredi if (dentry) { 1409a89f8337SAl Viro int error = d_revalidate(dentry, flags); 1410bad61189SMiklos Szeredi if (unlikely(error <= 0)) { 141174ff0ffcSAl Viro if (!error) 14125542aa2fSEric W. Biederman d_invalidate(dentry); 1413bad61189SMiklos Szeredi dput(dentry); 141474ff0ffcSAl Viro return ERR_PTR(error); 1415bad61189SMiklos Szeredi } 1416bad61189SMiklos Szeredi } 1417baa03890SNick Piggin return dentry; 1418baa03890SNick Piggin } 1419baa03890SNick Piggin 1420baa03890SNick Piggin /* 1421a03ece5fSAl Viro * Parent directory has inode locked exclusive. This is one 1422a03ece5fSAl Viro * and only case when ->lookup() gets called on non in-lookup 1423a03ece5fSAl Viro * dentries - as the matter of fact, this only gets called 1424a03ece5fSAl Viro * when directory is guaranteed to have no in-lookup children 1425a03ece5fSAl Viro * at all. 142644396f4bSJosef Bacik */ 1427a03ece5fSAl Viro static struct dentry *__lookup_hash(const struct qstr *name, 1428a03ece5fSAl Viro struct dentry *base, unsigned int flags) 142944396f4bSJosef Bacik { 1430a03ece5fSAl Viro struct dentry *dentry = lookup_dcache(name, base, flags); 143144396f4bSJosef Bacik struct dentry *old; 1432a03ece5fSAl Viro struct inode *dir = base->d_inode; 1433a03ece5fSAl Viro 1434a03ece5fSAl Viro if (dentry) 1435a03ece5fSAl Viro return dentry; 143644396f4bSJosef Bacik 143744396f4bSJosef Bacik /* Don't create child dentry for a dead directory. */ 1438a03ece5fSAl Viro if (unlikely(IS_DEADDIR(dir))) 143944396f4bSJosef Bacik return ERR_PTR(-ENOENT); 1440a03ece5fSAl Viro 1441a03ece5fSAl Viro dentry = d_alloc(base, name); 1442a03ece5fSAl Viro if (unlikely(!dentry)) 1443a03ece5fSAl Viro return ERR_PTR(-ENOMEM); 144444396f4bSJosef Bacik 144572bd866aSAl Viro old = dir->i_op->lookup(dir, dentry, flags); 144644396f4bSJosef Bacik if (unlikely(old)) { 144744396f4bSJosef Bacik dput(dentry); 144844396f4bSJosef Bacik dentry = old; 144944396f4bSJosef Bacik } 145044396f4bSJosef Bacik return dentry; 145144396f4bSJosef Bacik } 145244396f4bSJosef Bacik 145320e34357SAl Viro static struct dentry *lookup_fast(struct nameidata *nd, 145420e34357SAl Viro struct inode **inode, 1455254cf582SAl Viro unsigned *seqp) 14561da177e4SLinus Torvalds { 145731e6b01fSNick Piggin struct dentry *dentry, *parent = nd->path.dentry; 14585a18fff2SAl Viro int status = 1; 14599875cf80SDavid Howells 14603cac260aSAl Viro /* 1461b04f784eSNick Piggin * Rename seqlock is not required here because in the off chance 14625d0f49c1SAl Viro * of a false negative due to a concurrent rename, the caller is 14635d0f49c1SAl Viro * going to fall back to non-racy lookup. 1464b04f784eSNick Piggin */ 146531e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) { 146631e6b01fSNick Piggin unsigned seq; 1467da53be12SLinus Torvalds dentry = __d_lookup_rcu(parent, &nd->last, &seq); 14685d0f49c1SAl Viro if (unlikely(!dentry)) { 14694675ac39SAl Viro if (unlazy_walk(nd)) 147020e34357SAl Viro return ERR_PTR(-ECHILD); 147120e34357SAl Viro return NULL; 14725d0f49c1SAl Viro } 14735a18fff2SAl Viro 147412f8ad4bSLinus Torvalds /* 147512f8ad4bSLinus Torvalds * This sequence count validates that the inode matches 147612f8ad4bSLinus Torvalds * the dentry name information from lookup. 147712f8ad4bSLinus Torvalds */ 147863afdfc7SDavid Howells *inode = d_backing_inode(dentry); 14795d0f49c1SAl Viro if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) 148020e34357SAl Viro return ERR_PTR(-ECHILD); 148112f8ad4bSLinus Torvalds 148212f8ad4bSLinus Torvalds /* 148312f8ad4bSLinus Torvalds * This sequence count validates that the parent had no 148412f8ad4bSLinus Torvalds * changes while we did the lookup of the dentry above. 148512f8ad4bSLinus Torvalds * 148612f8ad4bSLinus Torvalds * The memory barrier in read_seqcount_begin of child is 148712f8ad4bSLinus Torvalds * enough, we can use __read_seqcount_retry here. 148812f8ad4bSLinus Torvalds */ 14895d0f49c1SAl Viro if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq))) 149020e34357SAl Viro return ERR_PTR(-ECHILD); 14915a18fff2SAl Viro 1492254cf582SAl Viro *seqp = seq; 14934ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 1494c153007bSAl Viro if (likely(status > 0)) 149520e34357SAl Viro return dentry; 14964675ac39SAl Viro if (unlazy_child(nd, dentry, seq)) 149720e34357SAl Viro return ERR_PTR(-ECHILD); 1498209a7fb2SAl Viro if (unlikely(status == -ECHILD)) 1499209a7fb2SAl Viro /* we'd been told to redo it in non-rcu mode */ 1500209a7fb2SAl Viro status = d_revalidate(dentry, nd->flags); 15015a18fff2SAl Viro } else { 1502e97cdc87SAl Viro dentry = __d_lookup(parent, &nd->last); 150381e6f520SAl Viro if (unlikely(!dentry)) 150420e34357SAl Viro return NULL; 15054ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 15065d0f49c1SAl Viro } 15075a18fff2SAl Viro if (unlikely(status <= 0)) { 1508e9742b53SAl Viro if (!status) 15095d0f49c1SAl Viro d_invalidate(dentry); 15105a18fff2SAl Viro dput(dentry); 151120e34357SAl Viro return ERR_PTR(status); 15125a18fff2SAl Viro } 151320e34357SAl Viro return dentry; 1514697f514dSMiklos Szeredi } 1515697f514dSMiklos Szeredi 1516697f514dSMiklos Szeredi /* Fast lookup failed, do it the slow way */ 151788d8331aSAl Viro static struct dentry *__lookup_slow(const struct qstr *name, 1518e3c13928SAl Viro struct dentry *dir, 1519e3c13928SAl Viro unsigned int flags) 1520697f514dSMiklos Szeredi { 152188d8331aSAl Viro struct dentry *dentry, *old; 15221936386eSAl Viro struct inode *inode = dir->d_inode; 1523d9171b93SAl Viro DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 15241936386eSAl Viro 15251936386eSAl Viro /* Don't go there if it's already dead */ 152694bdd655SAl Viro if (unlikely(IS_DEADDIR(inode))) 152788d8331aSAl Viro return ERR_PTR(-ENOENT); 152894bdd655SAl Viro again: 1529d9171b93SAl Viro dentry = d_alloc_parallel(dir, name, &wq); 153094bdd655SAl Viro if (IS_ERR(dentry)) 153188d8331aSAl Viro return dentry; 153294bdd655SAl Viro if (unlikely(!d_in_lookup(dentry))) { 1533949a852eSAl Viro int error = d_revalidate(dentry, flags); 1534949a852eSAl Viro if (unlikely(error <= 0)) { 153594bdd655SAl Viro if (!error) { 1536949a852eSAl Viro d_invalidate(dentry); 1537949a852eSAl Viro dput(dentry); 153894bdd655SAl Viro goto again; 153994bdd655SAl Viro } 154094bdd655SAl Viro dput(dentry); 1541949a852eSAl Viro dentry = ERR_PTR(error); 1542949a852eSAl Viro } 154394bdd655SAl Viro } else { 15441936386eSAl Viro old = inode->i_op->lookup(inode, dentry, flags); 154585c7f810SAl Viro d_lookup_done(dentry); 15461936386eSAl Viro if (unlikely(old)) { 15471936386eSAl Viro dput(dentry); 15481936386eSAl Viro dentry = old; 1549949a852eSAl Viro } 1550949a852eSAl Viro } 1551e3c13928SAl Viro return dentry; 15521da177e4SLinus Torvalds } 15531da177e4SLinus Torvalds 155488d8331aSAl Viro static struct dentry *lookup_slow(const struct qstr *name, 155588d8331aSAl Viro struct dentry *dir, 155688d8331aSAl Viro unsigned int flags) 155788d8331aSAl Viro { 155888d8331aSAl Viro struct inode *inode = dir->d_inode; 155988d8331aSAl Viro struct dentry *res; 156088d8331aSAl Viro inode_lock_shared(inode); 156188d8331aSAl Viro res = __lookup_slow(name, dir, flags); 156288d8331aSAl Viro inode_unlock_shared(inode); 156388d8331aSAl Viro return res; 156488d8331aSAl Viro } 156588d8331aSAl Viro 156652094c8aSAl Viro static inline int may_lookup(struct nameidata *nd) 156752094c8aSAl Viro { 156852094c8aSAl Viro if (nd->flags & LOOKUP_RCU) { 15694ad5abb3SAl Viro int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); 157052094c8aSAl Viro if (err != -ECHILD) 157152094c8aSAl Viro return err; 15724675ac39SAl Viro if (unlazy_walk(nd)) 157352094c8aSAl Viro return -ECHILD; 157452094c8aSAl Viro } 15754ad5abb3SAl Viro return inode_permission(nd->inode, MAY_EXEC); 157652094c8aSAl Viro } 157752094c8aSAl Viro 157849055906SAl Viro static int reserve_stack(struct nameidata *nd, struct path *link, unsigned seq) 1579d63ff28fSAl Viro { 158049055906SAl Viro if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) 158149055906SAl Viro return -ELOOP; 15824542576bSAl Viro 15834542576bSAl Viro if (likely(nd->depth != EMBEDDED_LEVELS)) 15844542576bSAl Viro return 0; 15854542576bSAl Viro if (likely(nd->stack != nd->internal)) 15864542576bSAl Viro return 0; 158760ef60c7SAl Viro if (likely(nd_alloc_stack(nd))) 158849055906SAl Viro return 0; 158960ef60c7SAl Viro 159060ef60c7SAl Viro if (nd->flags & LOOKUP_RCU) { 159160ef60c7SAl Viro // we need to grab link before we do unlazy. And we can't skip 159260ef60c7SAl Viro // unlazy even if we fail to grab the link - cleanup needs it 1593aef9404dSAl Viro bool grabbed_link = legitimize_path(nd, link, seq); 159460ef60c7SAl Viro 159560ef60c7SAl Viro if (unlazy_walk(nd) != 0 || !grabbed_link) 159660ef60c7SAl Viro return -ECHILD; 159760ef60c7SAl Viro 159860ef60c7SAl Viro if (nd_alloc_stack(nd)) 159960ef60c7SAl Viro return 0; 1600bc40aee0SAl Viro } 160160ef60c7SAl Viro return -ENOMEM; 160249055906SAl Viro } 160349055906SAl Viro 160449055906SAl Viro enum {WALK_TRAILING = 1, WALK_MORE = 2, WALK_NOFOLLOW = 4}; 160549055906SAl Viro 160649055906SAl Viro static const char *pick_link(struct nameidata *nd, struct path *link, 160749055906SAl Viro struct inode *inode, unsigned seq, int flags) 160849055906SAl Viro { 160949055906SAl Viro struct saved *last; 161049055906SAl Viro const char *res; 161149055906SAl Viro int error = reserve_stack(nd, link, seq); 161249055906SAl Viro 161349055906SAl Viro if (unlikely(error)) { 161449055906SAl Viro if (!(nd->flags & LOOKUP_RCU)) 1615cd179f44SAl Viro path_put(link); 161606708adbSAl Viro return ERR_PTR(error); 1617626de996SAl Viro } 1618ab104923SAl Viro last = nd->stack + nd->depth++; 16191cf2665bSAl Viro last->link = *link; 1620fceef393SAl Viro clear_delayed_call(&last->done); 16210450b2d1SAl Viro last->seq = seq; 1622ad6cc4c3SAl Viro 1623b1a81972SAl Viro if (flags & WALK_TRAILING) { 1624ad6cc4c3SAl Viro error = may_follow_link(nd, inode); 1625ad6cc4c3SAl Viro if (unlikely(error)) 1626ad6cc4c3SAl Viro return ERR_PTR(error); 1627ad6cc4c3SAl Viro } 1628ad6cc4c3SAl Viro 1629ad6cc4c3SAl Viro if (unlikely(nd->flags & LOOKUP_NO_SYMLINKS)) 1630ad6cc4c3SAl Viro return ERR_PTR(-ELOOP); 1631ad6cc4c3SAl Viro 1632ad6cc4c3SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1633ad6cc4c3SAl Viro touch_atime(&last->link); 1634ad6cc4c3SAl Viro cond_resched(); 1635ad6cc4c3SAl Viro } else if (atime_needs_update(&last->link, inode)) { 1636ad6cc4c3SAl Viro if (unlikely(unlazy_walk(nd))) 1637ad6cc4c3SAl Viro return ERR_PTR(-ECHILD); 1638ad6cc4c3SAl Viro touch_atime(&last->link); 1639ad6cc4c3SAl Viro } 1640ad6cc4c3SAl Viro 1641ad6cc4c3SAl Viro error = security_inode_follow_link(link->dentry, inode, 1642ad6cc4c3SAl Viro nd->flags & LOOKUP_RCU); 1643ad6cc4c3SAl Viro if (unlikely(error)) 1644ad6cc4c3SAl Viro return ERR_PTR(error); 1645ad6cc4c3SAl Viro 1646ad6cc4c3SAl Viro res = READ_ONCE(inode->i_link); 1647ad6cc4c3SAl Viro if (!res) { 1648ad6cc4c3SAl Viro const char * (*get)(struct dentry *, struct inode *, 1649ad6cc4c3SAl Viro struct delayed_call *); 1650ad6cc4c3SAl Viro get = inode->i_op->get_link; 1651ad6cc4c3SAl Viro if (nd->flags & LOOKUP_RCU) { 1652ad6cc4c3SAl Viro res = get(NULL, inode, &last->done); 1653ad6cc4c3SAl Viro if (res == ERR_PTR(-ECHILD)) { 1654ad6cc4c3SAl Viro if (unlikely(unlazy_walk(nd))) 1655ad6cc4c3SAl Viro return ERR_PTR(-ECHILD); 1656ad6cc4c3SAl Viro res = get(link->dentry, inode, &last->done); 1657ad6cc4c3SAl Viro } 1658ad6cc4c3SAl Viro } else { 1659ad6cc4c3SAl Viro res = get(link->dentry, inode, &last->done); 1660ad6cc4c3SAl Viro } 1661ad6cc4c3SAl Viro if (!res) 1662ad6cc4c3SAl Viro goto all_done; 1663ad6cc4c3SAl Viro if (IS_ERR(res)) 1664ad6cc4c3SAl Viro return res; 1665ad6cc4c3SAl Viro } 1666ad6cc4c3SAl Viro if (*res == '/') { 1667ad6cc4c3SAl Viro error = nd_jump_root(nd); 1668ad6cc4c3SAl Viro if (unlikely(error)) 1669ad6cc4c3SAl Viro return ERR_PTR(error); 1670ad6cc4c3SAl Viro while (unlikely(*++res == '/')) 1671ad6cc4c3SAl Viro ; 1672ad6cc4c3SAl Viro } 1673ad6cc4c3SAl Viro if (*res) 1674ad6cc4c3SAl Viro return res; 1675ad6cc4c3SAl Viro all_done: // pure jump 1676ad6cc4c3SAl Viro put_link(nd); 1677ad6cc4c3SAl Viro return NULL; 1678d63ff28fSAl Viro } 1679d63ff28fSAl Viro 16803ddcd056SLinus Torvalds /* 16813ddcd056SLinus Torvalds * Do we need to follow links? We _really_ want to be able 16823ddcd056SLinus Torvalds * to do this check without having to look at inode->i_op, 16833ddcd056SLinus Torvalds * so we keep a cache of "no, this doesn't need follow_link" 16843ddcd056SLinus Torvalds * for the common case. 16853ddcd056SLinus Torvalds */ 1686b0417d2cSAl Viro static const char *step_into(struct nameidata *nd, int flags, 1687cbae4d12SAl Viro struct dentry *dentry, struct inode *inode, unsigned seq) 16883ddcd056SLinus Torvalds { 1689cbae4d12SAl Viro struct path path; 1690cbae4d12SAl Viro int err = handle_mounts(nd, dentry, &path, &inode, &seq); 1691cbae4d12SAl Viro 1692cbae4d12SAl Viro if (err < 0) 1693b0417d2cSAl Viro return ERR_PTR(err); 1694cbae4d12SAl Viro if (likely(!d_is_symlink(path.dentry)) || 16958c4efe22SAl Viro ((flags & WALK_TRAILING) && !(nd->flags & LOOKUP_FOLLOW)) || 1696aca2903eSAl Viro (flags & WALK_NOFOLLOW)) { 16978f64fb1cSAl Viro /* not a symlink or should not follow */ 1698c99687a0SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1699c99687a0SAl Viro dput(nd->path.dentry); 1700c99687a0SAl Viro if (nd->path.mnt != path.mnt) 1701c99687a0SAl Viro mntput(nd->path.mnt); 1702c99687a0SAl Viro } 1703c99687a0SAl Viro nd->path = path; 17048f64fb1cSAl Viro nd->inode = inode; 17058f64fb1cSAl Viro nd->seq = seq; 1706b0417d2cSAl Viro return NULL; 17078f64fb1cSAl Viro } 1708a7f77542SAl Viro if (nd->flags & LOOKUP_RCU) { 170984f0cd9eSAl Viro /* make sure that d_is_symlink above matches inode */ 1710cbae4d12SAl Viro if (read_seqcount_retry(&path.dentry->d_seq, seq)) 1711b0417d2cSAl Viro return ERR_PTR(-ECHILD); 171284f0cd9eSAl Viro } else { 171384f0cd9eSAl Viro if (path.mnt == nd->path.mnt) 171484f0cd9eSAl Viro mntget(path.mnt); 1715a7f77542SAl Viro } 1716b1a81972SAl Viro return pick_link(nd, &path, inode, seq, flags); 17173ddcd056SLinus Torvalds } 17183ddcd056SLinus Torvalds 1719c2df1968SAl Viro static struct dentry *follow_dotdot_rcu(struct nameidata *nd, 1720c2df1968SAl Viro struct inode **inodep, 1721c2df1968SAl Viro unsigned *seqp) 1722957dd41dSAl Viro { 172312487f30SAl Viro struct dentry *parent, *old; 1724957dd41dSAl Viro 172512487f30SAl Viro if (path_equal(&nd->path, &nd->root)) 172612487f30SAl Viro goto in_root; 172712487f30SAl Viro if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) { 17287ef482faSAl Viro struct path path; 1729efe772d6SAl Viro unsigned seq; 17307ef482faSAl Viro if (!choose_mountpoint_rcu(real_mount(nd->path.mnt), 17317ef482faSAl Viro &nd->root, &path, &seq)) 173212487f30SAl Viro goto in_root; 1733efe772d6SAl Viro if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 1734efe772d6SAl Viro return ERR_PTR(-ECHILD); 1735efe772d6SAl Viro nd->path = path; 1736efe772d6SAl Viro nd->inode = path.dentry->d_inode; 1737efe772d6SAl Viro nd->seq = seq; 1738efe772d6SAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1739efe772d6SAl Viro return ERR_PTR(-ECHILD); 1740efe772d6SAl Viro /* we know that mountpoint was pinned */ 1741957dd41dSAl Viro } 174212487f30SAl Viro old = nd->path.dentry; 174312487f30SAl Viro parent = old->d_parent; 174412487f30SAl Viro *inodep = parent->d_inode; 174512487f30SAl Viro *seqp = read_seqcount_begin(&parent->d_seq); 174612487f30SAl Viro if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq))) 174712487f30SAl Viro return ERR_PTR(-ECHILD); 174812487f30SAl Viro if (unlikely(!path_connected(nd->path.mnt, parent))) 174912487f30SAl Viro return ERR_PTR(-ECHILD); 175012487f30SAl Viro return parent; 175112487f30SAl Viro in_root: 1752efe772d6SAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1753efe772d6SAl Viro return ERR_PTR(-ECHILD); 1754957dd41dSAl Viro if (unlikely(nd->flags & LOOKUP_BENEATH)) 17557521f22bSAl Viro return ERR_PTR(-ECHILD); 1756c2df1968SAl Viro return NULL; 1757957dd41dSAl Viro } 1758957dd41dSAl Viro 1759c2df1968SAl Viro static struct dentry *follow_dotdot(struct nameidata *nd, 1760c2df1968SAl Viro struct inode **inodep, 1761c2df1968SAl Viro unsigned *seqp) 1762957dd41dSAl Viro { 176312487f30SAl Viro struct dentry *parent; 176412487f30SAl Viro 1765957dd41dSAl Viro if (path_equal(&nd->path, &nd->root)) 176612487f30SAl Viro goto in_root; 176712487f30SAl Viro if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) { 17682aa38470SAl Viro struct path path; 17692aa38470SAl Viro 17702aa38470SAl Viro if (!choose_mountpoint(real_mount(nd->path.mnt), 17712aa38470SAl Viro &nd->root, &path)) 177212487f30SAl Viro goto in_root; 1773165200d6SAl Viro path_put(&nd->path); 1774165200d6SAl Viro nd->path = path; 17752aa38470SAl Viro nd->inode = path.dentry->d_inode; 1776165200d6SAl Viro if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 1777165200d6SAl Viro return ERR_PTR(-EXDEV); 177812487f30SAl Viro } 1779957dd41dSAl Viro /* rare case of legitimate dget_parent()... */ 178012487f30SAl Viro parent = dget_parent(nd->path.dentry); 1781957dd41dSAl Viro if (unlikely(!path_connected(nd->path.mnt, parent))) { 1782957dd41dSAl Viro dput(parent); 17837521f22bSAl Viro return ERR_PTR(-ENOENT); 1784957dd41dSAl Viro } 1785c2df1968SAl Viro *seqp = 0; 1786c2df1968SAl Viro *inodep = parent->d_inode; 1787c2df1968SAl Viro return parent; 178812487f30SAl Viro 178912487f30SAl Viro in_root: 1790957dd41dSAl Viro if (unlikely(nd->flags & LOOKUP_BENEATH)) 17917521f22bSAl Viro return ERR_PTR(-EXDEV); 1792c2df1968SAl Viro dget(nd->path.dentry); 1793c2df1968SAl Viro return NULL; 1794957dd41dSAl Viro } 1795957dd41dSAl Viro 17967521f22bSAl Viro static const char *handle_dots(struct nameidata *nd, int type) 1797957dd41dSAl Viro { 1798957dd41dSAl Viro if (type == LAST_DOTDOT) { 17997521f22bSAl Viro const char *error = NULL; 1800c2df1968SAl Viro struct dentry *parent; 1801c2df1968SAl Viro struct inode *inode; 1802c2df1968SAl Viro unsigned seq; 1803957dd41dSAl Viro 1804957dd41dSAl Viro if (!nd->root.mnt) { 18057521f22bSAl Viro error = ERR_PTR(set_root(nd)); 1806957dd41dSAl Viro if (error) 1807957dd41dSAl Viro return error; 1808957dd41dSAl Viro } 1809957dd41dSAl Viro if (nd->flags & LOOKUP_RCU) 1810c2df1968SAl Viro parent = follow_dotdot_rcu(nd, &inode, &seq); 1811957dd41dSAl Viro else 1812c2df1968SAl Viro parent = follow_dotdot(nd, &inode, &seq); 1813c2df1968SAl Viro if (IS_ERR(parent)) 1814c2df1968SAl Viro return ERR_CAST(parent); 1815c2df1968SAl Viro if (unlikely(!parent)) 1816c2df1968SAl Viro error = step_into(nd, WALK_NOFOLLOW, 1817c2df1968SAl Viro nd->path.dentry, nd->inode, nd->seq); 1818c2df1968SAl Viro else 1819c2df1968SAl Viro error = step_into(nd, WALK_NOFOLLOW, 1820c2df1968SAl Viro parent, inode, seq); 1821c2df1968SAl Viro if (unlikely(error)) 1822957dd41dSAl Viro return error; 1823957dd41dSAl Viro 1824957dd41dSAl Viro if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) { 1825957dd41dSAl Viro /* 1826957dd41dSAl Viro * If there was a racing rename or mount along our 1827957dd41dSAl Viro * path, then we can't be sure that ".." hasn't jumped 1828957dd41dSAl Viro * above nd->root (and so userspace should retry or use 1829957dd41dSAl Viro * some fallback). 1830957dd41dSAl Viro */ 1831957dd41dSAl Viro smp_rmb(); 1832957dd41dSAl Viro if (unlikely(__read_seqcount_retry(&mount_lock.seqcount, nd->m_seq))) 18337521f22bSAl Viro return ERR_PTR(-EAGAIN); 1834957dd41dSAl Viro if (unlikely(__read_seqcount_retry(&rename_lock.seqcount, nd->r_seq))) 18357521f22bSAl Viro return ERR_PTR(-EAGAIN); 1836957dd41dSAl Viro } 1837957dd41dSAl Viro } 18387521f22bSAl Viro return NULL; 1839957dd41dSAl Viro } 1840957dd41dSAl Viro 184192d27016SAl Viro static const char *walk_component(struct nameidata *nd, int flags) 1842ce57dfc1SAl Viro { 1843db3c9adeSAl Viro struct dentry *dentry; 1844ce57dfc1SAl Viro struct inode *inode; 1845254cf582SAl Viro unsigned seq; 1846ce57dfc1SAl Viro /* 1847ce57dfc1SAl Viro * "." and ".." are special - ".." especially so because it has 1848ce57dfc1SAl Viro * to be able to know about the current root directory and 1849ce57dfc1SAl Viro * parent relationships. 1850ce57dfc1SAl Viro */ 18514693a547SAl Viro if (unlikely(nd->last_type != LAST_NORM)) { 18521c4ff1a8SAl Viro if (!(flags & WALK_MORE) && nd->depth) 18534693a547SAl Viro put_link(nd); 18547521f22bSAl Viro return handle_dots(nd, nd->last_type); 18554693a547SAl Viro } 185620e34357SAl Viro dentry = lookup_fast(nd, &inode, &seq); 185720e34357SAl Viro if (IS_ERR(dentry)) 185892d27016SAl Viro return ERR_CAST(dentry); 185920e34357SAl Viro if (unlikely(!dentry)) { 1860db3c9adeSAl Viro dentry = lookup_slow(&nd->last, nd->path.dentry, nd->flags); 1861db3c9adeSAl Viro if (IS_ERR(dentry)) 186292d27016SAl Viro return ERR_CAST(dentry); 186320e34357SAl Viro } 186456676ec3SAl Viro if (!(flags & WALK_MORE) && nd->depth) 186556676ec3SAl Viro put_link(nd); 1866b0417d2cSAl Viro return step_into(nd, flags, dentry, inode, seq); 1867ce57dfc1SAl Viro } 1868ce57dfc1SAl Viro 18691da177e4SLinus Torvalds /* 1870bfcfaa77SLinus Torvalds * We can do the critical dentry name comparison and hashing 1871bfcfaa77SLinus Torvalds * operations one word at a time, but we are limited to: 1872bfcfaa77SLinus Torvalds * 1873bfcfaa77SLinus Torvalds * - Architectures with fast unaligned word accesses. We could 1874bfcfaa77SLinus Torvalds * do a "get_unaligned()" if this helps and is sufficiently 1875bfcfaa77SLinus Torvalds * fast. 1876bfcfaa77SLinus Torvalds * 1877bfcfaa77SLinus Torvalds * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we 1878bfcfaa77SLinus Torvalds * do not trap on the (extremely unlikely) case of a page 1879bfcfaa77SLinus Torvalds * crossing operation. 1880bfcfaa77SLinus Torvalds * 1881bfcfaa77SLinus Torvalds * - Furthermore, we need an efficient 64-bit compile for the 1882bfcfaa77SLinus Torvalds * 64-bit case in order to generate the "number of bytes in 1883bfcfaa77SLinus Torvalds * the final mask". Again, that could be replaced with a 1884bfcfaa77SLinus Torvalds * efficient population count instruction or similar. 1885bfcfaa77SLinus Torvalds */ 1886bfcfaa77SLinus Torvalds #ifdef CONFIG_DCACHE_WORD_ACCESS 1887bfcfaa77SLinus Torvalds 1888f68e556eSLinus Torvalds #include <asm/word-at-a-time.h> 1889bfcfaa77SLinus Torvalds 1890468a9428SGeorge Spelvin #ifdef HASH_MIX 1891bfcfaa77SLinus Torvalds 1892468a9428SGeorge Spelvin /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */ 1893468a9428SGeorge Spelvin 1894468a9428SGeorge Spelvin #elif defined(CONFIG_64BIT) 18952a18da7aSGeorge Spelvin /* 18962a18da7aSGeorge Spelvin * Register pressure in the mixing function is an issue, particularly 18972a18da7aSGeorge Spelvin * on 32-bit x86, but almost any function requires one state value and 18982a18da7aSGeorge Spelvin * one temporary. Instead, use a function designed for two state values 18992a18da7aSGeorge Spelvin * and no temporaries. 19002a18da7aSGeorge Spelvin * 19012a18da7aSGeorge Spelvin * This function cannot create a collision in only two iterations, so 19022a18da7aSGeorge Spelvin * we have two iterations to achieve avalanche. In those two iterations, 19032a18da7aSGeorge Spelvin * we have six layers of mixing, which is enough to spread one bit's 19042a18da7aSGeorge Spelvin * influence out to 2^6 = 64 state bits. 19052a18da7aSGeorge Spelvin * 19062a18da7aSGeorge Spelvin * Rotate constants are scored by considering either 64 one-bit input 19072a18da7aSGeorge Spelvin * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the 19082a18da7aSGeorge Spelvin * probability of that delta causing a change to each of the 128 output 19092a18da7aSGeorge Spelvin * bits, using a sample of random initial states. 19102a18da7aSGeorge Spelvin * 19112a18da7aSGeorge Spelvin * The Shannon entropy of the computed probabilities is then summed 19122a18da7aSGeorge Spelvin * to produce a score. Ideally, any input change has a 50% chance of 19132a18da7aSGeorge Spelvin * toggling any given output bit. 19142a18da7aSGeorge Spelvin * 19152a18da7aSGeorge Spelvin * Mixing scores (in bits) for (12,45): 19162a18da7aSGeorge Spelvin * Input delta: 1-bit 2-bit 19172a18da7aSGeorge Spelvin * 1 round: 713.3 42542.6 19182a18da7aSGeorge Spelvin * 2 rounds: 2753.7 140389.8 19192a18da7aSGeorge Spelvin * 3 rounds: 5954.1 233458.2 19202a18da7aSGeorge Spelvin * 4 rounds: 7862.6 256672.2 19212a18da7aSGeorge Spelvin * Perfect: 8192 258048 19222a18da7aSGeorge Spelvin * (64*128) (64*63/2 * 128) 19232a18da7aSGeorge Spelvin */ 19242a18da7aSGeorge Spelvin #define HASH_MIX(x, y, a) \ 19252a18da7aSGeorge Spelvin ( x ^= (a), \ 19262a18da7aSGeorge Spelvin y ^= x, x = rol64(x,12),\ 19272a18da7aSGeorge Spelvin x += y, y = rol64(y,45),\ 19282a18da7aSGeorge Spelvin y *= 9 ) 1929bfcfaa77SLinus Torvalds 19300fed3ac8SGeorge Spelvin /* 19312a18da7aSGeorge Spelvin * Fold two longs into one 32-bit hash value. This must be fast, but 19322a18da7aSGeorge Spelvin * latency isn't quite as critical, as there is a fair bit of additional 19332a18da7aSGeorge Spelvin * work done before the hash value is used. 19340fed3ac8SGeorge Spelvin */ 19352a18da7aSGeorge Spelvin static inline unsigned int fold_hash(unsigned long x, unsigned long y) 19360fed3ac8SGeorge Spelvin { 19372a18da7aSGeorge Spelvin y ^= x * GOLDEN_RATIO_64; 19382a18da7aSGeorge Spelvin y *= GOLDEN_RATIO_64; 19392a18da7aSGeorge Spelvin return y >> 32; 19400fed3ac8SGeorge Spelvin } 19410fed3ac8SGeorge Spelvin 1942bfcfaa77SLinus Torvalds #else /* 32-bit case */ 1943bfcfaa77SLinus Torvalds 19442a18da7aSGeorge Spelvin /* 19452a18da7aSGeorge Spelvin * Mixing scores (in bits) for (7,20): 19462a18da7aSGeorge Spelvin * Input delta: 1-bit 2-bit 19472a18da7aSGeorge Spelvin * 1 round: 330.3 9201.6 19482a18da7aSGeorge Spelvin * 2 rounds: 1246.4 25475.4 19492a18da7aSGeorge Spelvin * 3 rounds: 1907.1 31295.1 19502a18da7aSGeorge Spelvin * 4 rounds: 2042.3 31718.6 19512a18da7aSGeorge Spelvin * Perfect: 2048 31744 19522a18da7aSGeorge Spelvin * (32*64) (32*31/2 * 64) 19532a18da7aSGeorge Spelvin */ 19542a18da7aSGeorge Spelvin #define HASH_MIX(x, y, a) \ 19552a18da7aSGeorge Spelvin ( x ^= (a), \ 19562a18da7aSGeorge Spelvin y ^= x, x = rol32(x, 7),\ 19572a18da7aSGeorge Spelvin x += y, y = rol32(y,20),\ 19582a18da7aSGeorge Spelvin y *= 9 ) 1959bfcfaa77SLinus Torvalds 19602a18da7aSGeorge Spelvin static inline unsigned int fold_hash(unsigned long x, unsigned long y) 19610fed3ac8SGeorge Spelvin { 19622a18da7aSGeorge Spelvin /* Use arch-optimized multiply if one exists */ 19632a18da7aSGeorge Spelvin return __hash_32(y ^ __hash_32(x)); 19640fed3ac8SGeorge Spelvin } 19650fed3ac8SGeorge Spelvin 1966bfcfaa77SLinus Torvalds #endif 1967bfcfaa77SLinus Torvalds 19682a18da7aSGeorge Spelvin /* 19692a18da7aSGeorge Spelvin * Return the hash of a string of known length. This is carfully 19702a18da7aSGeorge Spelvin * designed to match hash_name(), which is the more critical function. 19712a18da7aSGeorge Spelvin * In particular, we must end by hashing a final word containing 0..7 19722a18da7aSGeorge Spelvin * payload bytes, to match the way that hash_name() iterates until it 19732a18da7aSGeorge Spelvin * finds the delimiter after the name. 19742a18da7aSGeorge Spelvin */ 19758387ff25SLinus Torvalds unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) 1976bfcfaa77SLinus Torvalds { 19778387ff25SLinus Torvalds unsigned long a, x = 0, y = (unsigned long)salt; 1978bfcfaa77SLinus Torvalds 1979bfcfaa77SLinus Torvalds for (;;) { 1980fcfd2fbfSGeorge Spelvin if (!len) 1981fcfd2fbfSGeorge Spelvin goto done; 1982e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name); 1983bfcfaa77SLinus Torvalds if (len < sizeof(unsigned long)) 1984bfcfaa77SLinus Torvalds break; 19852a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 1986bfcfaa77SLinus Torvalds name += sizeof(unsigned long); 1987bfcfaa77SLinus Torvalds len -= sizeof(unsigned long); 1988bfcfaa77SLinus Torvalds } 19892a18da7aSGeorge Spelvin x ^= a & bytemask_from_count(len); 1990bfcfaa77SLinus Torvalds done: 19912a18da7aSGeorge Spelvin return fold_hash(x, y); 1992bfcfaa77SLinus Torvalds } 1993bfcfaa77SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 1994bfcfaa77SLinus Torvalds 1995fcfd2fbfSGeorge Spelvin /* Return the "hash_len" (hash and length) of a null-terminated string */ 19968387ff25SLinus Torvalds u64 hashlen_string(const void *salt, const char *name) 1997fcfd2fbfSGeorge Spelvin { 19988387ff25SLinus Torvalds unsigned long a = 0, x = 0, y = (unsigned long)salt; 19998387ff25SLinus Torvalds unsigned long adata, mask, len; 2000fcfd2fbfSGeorge Spelvin const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 2001fcfd2fbfSGeorge Spelvin 20028387ff25SLinus Torvalds len = 0; 20038387ff25SLinus Torvalds goto inside; 20048387ff25SLinus Torvalds 2005fcfd2fbfSGeorge Spelvin do { 20062a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 2007fcfd2fbfSGeorge Spelvin len += sizeof(unsigned long); 20088387ff25SLinus Torvalds inside: 2009fcfd2fbfSGeorge Spelvin a = load_unaligned_zeropad(name+len); 2010fcfd2fbfSGeorge Spelvin } while (!has_zero(a, &adata, &constants)); 2011fcfd2fbfSGeorge Spelvin 2012fcfd2fbfSGeorge Spelvin adata = prep_zero_mask(a, adata, &constants); 2013fcfd2fbfSGeorge Spelvin mask = create_zero_mask(adata); 20142a18da7aSGeorge Spelvin x ^= a & zero_bytemask(mask); 2015fcfd2fbfSGeorge Spelvin 20162a18da7aSGeorge Spelvin return hashlen_create(fold_hash(x, y), len + find_zero(mask)); 2017fcfd2fbfSGeorge Spelvin } 2018fcfd2fbfSGeorge Spelvin EXPORT_SYMBOL(hashlen_string); 2019fcfd2fbfSGeorge Spelvin 2020bfcfaa77SLinus Torvalds /* 2021bfcfaa77SLinus Torvalds * Calculate the length and hash of the path component, and 2022d6bb3e90SLinus Torvalds * return the "hash_len" as the result. 2023bfcfaa77SLinus Torvalds */ 20248387ff25SLinus Torvalds static inline u64 hash_name(const void *salt, const char *name) 2025bfcfaa77SLinus Torvalds { 20268387ff25SLinus Torvalds unsigned long a = 0, b, x = 0, y = (unsigned long)salt; 20278387ff25SLinus Torvalds unsigned long adata, bdata, mask, len; 202836126f8fSLinus Torvalds const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 2029bfcfaa77SLinus Torvalds 20308387ff25SLinus Torvalds len = 0; 20318387ff25SLinus Torvalds goto inside; 20328387ff25SLinus Torvalds 2033bfcfaa77SLinus Torvalds do { 20342a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 2035bfcfaa77SLinus Torvalds len += sizeof(unsigned long); 20368387ff25SLinus Torvalds inside: 2037e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name+len); 203836126f8fSLinus Torvalds b = a ^ REPEAT_BYTE('/'); 203936126f8fSLinus Torvalds } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); 2040bfcfaa77SLinus Torvalds 204136126f8fSLinus Torvalds adata = prep_zero_mask(a, adata, &constants); 204236126f8fSLinus Torvalds bdata = prep_zero_mask(b, bdata, &constants); 204336126f8fSLinus Torvalds mask = create_zero_mask(adata | bdata); 20442a18da7aSGeorge Spelvin x ^= a & zero_bytemask(mask); 204536126f8fSLinus Torvalds 20462a18da7aSGeorge Spelvin return hashlen_create(fold_hash(x, y), len + find_zero(mask)); 2047bfcfaa77SLinus Torvalds } 2048bfcfaa77SLinus Torvalds 20492a18da7aSGeorge Spelvin #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */ 2050bfcfaa77SLinus Torvalds 2051fcfd2fbfSGeorge Spelvin /* Return the hash of a string of known length */ 20528387ff25SLinus Torvalds unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) 20530145acc2SLinus Torvalds { 20548387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 20550145acc2SLinus Torvalds while (len--) 2056fcfd2fbfSGeorge Spelvin hash = partial_name_hash((unsigned char)*name++, hash); 20570145acc2SLinus Torvalds return end_name_hash(hash); 20580145acc2SLinus Torvalds } 2059ae942ae7SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 20600145acc2SLinus Torvalds 2061fcfd2fbfSGeorge Spelvin /* Return the "hash_len" (hash and length) of a null-terminated string */ 20628387ff25SLinus Torvalds u64 hashlen_string(const void *salt, const char *name) 2063fcfd2fbfSGeorge Spelvin { 20648387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 2065fcfd2fbfSGeorge Spelvin unsigned long len = 0, c; 2066fcfd2fbfSGeorge Spelvin 2067fcfd2fbfSGeorge Spelvin c = (unsigned char)*name; 2068e0ab7af9SGeorge Spelvin while (c) { 2069fcfd2fbfSGeorge Spelvin len++; 2070fcfd2fbfSGeorge Spelvin hash = partial_name_hash(c, hash); 2071fcfd2fbfSGeorge Spelvin c = (unsigned char)name[len]; 2072e0ab7af9SGeorge Spelvin } 2073fcfd2fbfSGeorge Spelvin return hashlen_create(end_name_hash(hash), len); 2074fcfd2fbfSGeorge Spelvin } 2075f2a031b6SGeorge Spelvin EXPORT_SYMBOL(hashlen_string); 2076fcfd2fbfSGeorge Spelvin 20773ddcd056SLinus Torvalds /* 2078200e9ef7SLinus Torvalds * We know there's a real path component here of at least 2079200e9ef7SLinus Torvalds * one character. 2080200e9ef7SLinus Torvalds */ 20818387ff25SLinus Torvalds static inline u64 hash_name(const void *salt, const char *name) 2082200e9ef7SLinus Torvalds { 20838387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 2084200e9ef7SLinus Torvalds unsigned long len = 0, c; 2085200e9ef7SLinus Torvalds 2086200e9ef7SLinus Torvalds c = (unsigned char)*name; 2087200e9ef7SLinus Torvalds do { 2088200e9ef7SLinus Torvalds len++; 2089200e9ef7SLinus Torvalds hash = partial_name_hash(c, hash); 2090200e9ef7SLinus Torvalds c = (unsigned char)name[len]; 2091200e9ef7SLinus Torvalds } while (c && c != '/'); 2092d6bb3e90SLinus Torvalds return hashlen_create(end_name_hash(hash), len); 2093200e9ef7SLinus Torvalds } 2094200e9ef7SLinus Torvalds 2095bfcfaa77SLinus Torvalds #endif 2096bfcfaa77SLinus Torvalds 2097200e9ef7SLinus Torvalds /* 20981da177e4SLinus Torvalds * Name resolution. 2099ea3834d9SPrasanna Meda * This is the basic name resolution function, turning a pathname into 2100ea3834d9SPrasanna Meda * the final dentry. We expect 'base' to be positive and a directory. 21011da177e4SLinus Torvalds * 2102ea3834d9SPrasanna Meda * Returns 0 and nd will have valid dentry and mnt on success. 2103ea3834d9SPrasanna Meda * Returns error and drops reference to input namei data on failure. 21041da177e4SLinus Torvalds */ 21056de88d72SAl Viro static int link_path_walk(const char *name, struct nameidata *nd) 21061da177e4SLinus Torvalds { 2107d8d4611aSAl Viro int depth = 0; // depth <= nd->depth 21081da177e4SLinus Torvalds int err; 21091da177e4SLinus Torvalds 2110b4c03536SAl Viro nd->last_type = LAST_ROOT; 2111c108837eSAl Viro nd->flags |= LOOKUP_PARENT; 21129b5858e9SAl Viro if (IS_ERR(name)) 21139b5858e9SAl Viro return PTR_ERR(name); 21141da177e4SLinus Torvalds while (*name=='/') 21151da177e4SLinus Torvalds name++; 21161da177e4SLinus Torvalds if (!*name) 21179e18f10aSAl Viro return 0; 21181da177e4SLinus Torvalds 21191da177e4SLinus Torvalds /* At this point we know we have a real path component. */ 21201da177e4SLinus Torvalds for(;;) { 212192d27016SAl Viro const char *link; 2122d6bb3e90SLinus Torvalds u64 hash_len; 2123fe479a58SAl Viro int type; 21241da177e4SLinus Torvalds 212552094c8aSAl Viro err = may_lookup(nd); 21261da177e4SLinus Torvalds if (err) 21273595e234SAl Viro return err; 21281da177e4SLinus Torvalds 21298387ff25SLinus Torvalds hash_len = hash_name(nd->path.dentry, name); 21301da177e4SLinus Torvalds 2131fe479a58SAl Viro type = LAST_NORM; 2132d6bb3e90SLinus Torvalds if (name[0] == '.') switch (hashlen_len(hash_len)) { 2133fe479a58SAl Viro case 2: 2134200e9ef7SLinus Torvalds if (name[1] == '.') { 2135fe479a58SAl Viro type = LAST_DOTDOT; 213616c2cd71SAl Viro nd->flags |= LOOKUP_JUMPED; 213716c2cd71SAl Viro } 2138fe479a58SAl Viro break; 2139fe479a58SAl Viro case 1: 2140fe479a58SAl Viro type = LAST_DOT; 2141fe479a58SAl Viro } 21425a202bcdSAl Viro if (likely(type == LAST_NORM)) { 21435a202bcdSAl Viro struct dentry *parent = nd->path.dentry; 214416c2cd71SAl Viro nd->flags &= ~LOOKUP_JUMPED; 21455a202bcdSAl Viro if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { 2146a060dc50SJames Hogan struct qstr this = { { .hash_len = hash_len }, .name = name }; 2147da53be12SLinus Torvalds err = parent->d_op->d_hash(parent, &this); 21485a202bcdSAl Viro if (err < 0) 21493595e234SAl Viro return err; 2150d6bb3e90SLinus Torvalds hash_len = this.hash_len; 2151d6bb3e90SLinus Torvalds name = this.name; 21525a202bcdSAl Viro } 21535a202bcdSAl Viro } 2154fe479a58SAl Viro 2155d6bb3e90SLinus Torvalds nd->last.hash_len = hash_len; 2156d6bb3e90SLinus Torvalds nd->last.name = name; 21575f4a6a69SAl Viro nd->last_type = type; 21585f4a6a69SAl Viro 2159d6bb3e90SLinus Torvalds name += hashlen_len(hash_len); 2160d6bb3e90SLinus Torvalds if (!*name) 2161bdf6cbf1SAl Viro goto OK; 2162200e9ef7SLinus Torvalds /* 2163200e9ef7SLinus Torvalds * If it wasn't NUL, we know it was '/'. Skip that 2164200e9ef7SLinus Torvalds * slash, and continue until no more slashes. 2165200e9ef7SLinus Torvalds */ 2166200e9ef7SLinus Torvalds do { 2167d6bb3e90SLinus Torvalds name++; 2168d6bb3e90SLinus Torvalds } while (unlikely(*name == '/')); 21698620c238SAl Viro if (unlikely(!*name)) { 21708620c238SAl Viro OK: 2171d8d4611aSAl Viro /* pathname or trailing symlink, done */ 2172c108837eSAl Viro if (!depth) { 21730f705953SAl Viro nd->dir_uid = nd->inode->i_uid; 21740f705953SAl Viro nd->dir_mode = nd->inode->i_mode; 2175c108837eSAl Viro nd->flags &= ~LOOKUP_PARENT; 21768620c238SAl Viro return 0; 2177c108837eSAl Viro } 21788620c238SAl Viro /* last component of nested symlink */ 2179d8d4611aSAl Viro name = nd->stack[--depth].name; 21808c4efe22SAl Viro link = walk_component(nd, 0); 21811c4ff1a8SAl Viro } else { 21821c4ff1a8SAl Viro /* not the last component */ 21838c4efe22SAl Viro link = walk_component(nd, WALK_MORE); 21848620c238SAl Viro } 218592d27016SAl Viro if (unlikely(link)) { 218692d27016SAl Viro if (IS_ERR(link)) 218792d27016SAl Viro return PTR_ERR(link); 218892d27016SAl Viro /* a symlink to follow */ 2189d8d4611aSAl Viro nd->stack[depth++].name = name; 219092d27016SAl Viro name = link; 21919e18f10aSAl Viro continue; 219248c8b0c5SAl Viro } 219397242f99SAl Viro if (unlikely(!d_can_lookup(nd->path.dentry))) { 219497242f99SAl Viro if (nd->flags & LOOKUP_RCU) { 21954675ac39SAl Viro if (unlazy_walk(nd)) 219697242f99SAl Viro return -ECHILD; 219797242f99SAl Viro } 21983595e234SAl Viro return -ENOTDIR; 21995f4a6a69SAl Viro } 2200ce57dfc1SAl Viro } 220197242f99SAl Viro } 22021da177e4SLinus Torvalds 2203edc2b1daSAl Viro /* must be paired with terminate_walk() */ 2204c8a53ee5SAl Viro static const char *path_init(struct nameidata *nd, unsigned flags) 220531e6b01fSNick Piggin { 2206740a1678SAleksa Sarai int error; 2207c8a53ee5SAl Viro const char *s = nd->name->name; 220831e6b01fSNick Piggin 2209c0eb027eSLinus Torvalds if (!*s) 2210c0eb027eSLinus Torvalds flags &= ~LOOKUP_RCU; 2211edc2b1daSAl Viro if (flags & LOOKUP_RCU) 2212edc2b1daSAl Viro rcu_read_lock(); 2213c0eb027eSLinus Torvalds 2214c108837eSAl Viro nd->flags = flags | LOOKUP_JUMPED; 221531e6b01fSNick Piggin nd->depth = 0; 2216ab87f9a5SAleksa Sarai 2217ab87f9a5SAleksa Sarai nd->m_seq = __read_seqcount_begin(&mount_lock.seqcount); 2218ab87f9a5SAleksa Sarai nd->r_seq = __read_seqcount_begin(&rename_lock.seqcount); 2219ab87f9a5SAleksa Sarai smp_rmb(); 2220ab87f9a5SAleksa Sarai 22215b6ca027SAl Viro if (flags & LOOKUP_ROOT) { 2222b18825a7SDavid Howells struct dentry *root = nd->root.dentry; 2223b18825a7SDavid Howells struct inode *inode = root->d_inode; 222493893862SAl Viro if (*s && unlikely(!d_can_lookup(root))) 2225368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 22265b6ca027SAl Viro nd->path = nd->root; 22275b6ca027SAl Viro nd->inode = inode; 22285b6ca027SAl Viro if (flags & LOOKUP_RCU) { 2229ab87f9a5SAleksa Sarai nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 22308f47a016SAl Viro nd->root_seq = nd->seq; 22315b6ca027SAl Viro } else { 22325b6ca027SAl Viro path_get(&nd->path); 22335b6ca027SAl Viro } 2234368ee9baSAl Viro return s; 22355b6ca027SAl Viro } 22365b6ca027SAl Viro 223731e6b01fSNick Piggin nd->root.mnt = NULL; 2238248fb5b9SAl Viro nd->path.mnt = NULL; 2239248fb5b9SAl Viro nd->path.dentry = NULL; 224031e6b01fSNick Piggin 22418db52c7eSAleksa Sarai /* Absolute pathname -- fetch the root (LOOKUP_IN_ROOT uses nd->dfd). */ 22428db52c7eSAleksa Sarai if (*s == '/' && !(flags & LOOKUP_IN_ROOT)) { 2243740a1678SAleksa Sarai error = nd_jump_root(nd); 2244740a1678SAleksa Sarai if (unlikely(error)) 2245740a1678SAleksa Sarai return ERR_PTR(error); 2246ef55d917SAl Viro return s; 22478db52c7eSAleksa Sarai } 22488db52c7eSAleksa Sarai 22498db52c7eSAleksa Sarai /* Relative pathname -- get the starting-point it is relative to. */ 22508db52c7eSAleksa Sarai if (nd->dfd == AT_FDCWD) { 2251e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 225231e6b01fSNick Piggin struct fs_struct *fs = current->fs; 2253c28cc364SNick Piggin unsigned seq; 225431e6b01fSNick Piggin 2255c28cc364SNick Piggin do { 2256c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 225731e6b01fSNick Piggin nd->path = fs->pwd; 2258ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2259c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 2260c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 2261e41f7d4eSAl Viro } else { 2262e41f7d4eSAl Viro get_fs_pwd(current->fs, &nd->path); 2263ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2264e41f7d4eSAl Viro } 226531e6b01fSNick Piggin } else { 2266582aa64aSJeff Layton /* Caller must check execute permissions on the starting path component */ 2267c8a53ee5SAl Viro struct fd f = fdget_raw(nd->dfd); 226831e6b01fSNick Piggin struct dentry *dentry; 226931e6b01fSNick Piggin 22702903ff01SAl Viro if (!f.file) 2271368ee9baSAl Viro return ERR_PTR(-EBADF); 227231e6b01fSNick Piggin 22732903ff01SAl Viro dentry = f.file->f_path.dentry; 227431e6b01fSNick Piggin 2275edc2b1daSAl Viro if (*s && unlikely(!d_can_lookup(dentry))) { 22762903ff01SAl Viro fdput(f); 2277368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 2278f52e0c11SAl Viro } 22792903ff01SAl Viro 22802903ff01SAl Viro nd->path = f.file->f_path; 2281e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 228234a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 228334a26b99SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 22845590ff0dSUlrich Drepper } else { 22852903ff01SAl Viro path_get(&nd->path); 228634a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 22871da177e4SLinus Torvalds } 228834a26b99SAl Viro fdput(f); 2289e41f7d4eSAl Viro } 22908db52c7eSAleksa Sarai 2291adb21d2bSAleksa Sarai /* For scoped-lookups we need to set the root to the dirfd as well. */ 2292adb21d2bSAleksa Sarai if (flags & LOOKUP_IS_SCOPED) { 2293adb21d2bSAleksa Sarai nd->root = nd->path; 2294adb21d2bSAleksa Sarai if (flags & LOOKUP_RCU) { 2295adb21d2bSAleksa Sarai nd->root_seq = nd->seq; 2296adb21d2bSAleksa Sarai } else { 2297adb21d2bSAleksa Sarai path_get(&nd->root); 2298adb21d2bSAleksa Sarai nd->flags |= LOOKUP_ROOT_GRABBED; 2299adb21d2bSAleksa Sarai } 2300adb21d2bSAleksa Sarai } 2301adb21d2bSAleksa Sarai return s; 23029b4a9b14SAl Viro } 23039b4a9b14SAl Viro 23041ccac622SAl Viro static inline const char *lookup_last(struct nameidata *nd) 230595fa25d9SAl Viro { 2306bd92d7feSAl Viro if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) 2307bd92d7feSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 2308bd92d7feSAl Viro 2309c108837eSAl Viro return walk_component(nd, WALK_TRAILING); 2310bd92d7feSAl Viro } 2311bd92d7feSAl Viro 23124f757f3cSAl Viro static int handle_lookup_down(struct nameidata *nd) 23134f757f3cSAl Viro { 2314c153007bSAl Viro if (!(nd->flags & LOOKUP_RCU)) 2315db3c9adeSAl Viro dget(nd->path.dentry); 2316b0417d2cSAl Viro return PTR_ERR(step_into(nd, WALK_NOFOLLOW, 2317b0417d2cSAl Viro nd->path.dentry, nd->inode, nd->seq)); 23184f757f3cSAl Viro } 23194f757f3cSAl Viro 23209b4a9b14SAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2321c8a53ee5SAl Viro static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path) 23229b4a9b14SAl Viro { 2323c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2324bd92d7feSAl Viro int err; 232531e6b01fSNick Piggin 23269b5858e9SAl Viro if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) { 23274f757f3cSAl Viro err = handle_lookup_down(nd); 23285f336e72SAl Viro if (unlikely(err < 0)) 23295f336e72SAl Viro s = ERR_PTR(err); 23304f757f3cSAl Viro } 23314f757f3cSAl Viro 23321ccac622SAl Viro while (!(err = link_path_walk(s, nd)) && 23331ccac622SAl Viro (s = lookup_last(nd)) != NULL) 23341ccac622SAl Viro ; 23359f1fafeeSAl Viro if (!err) 23369f1fafeeSAl Viro err = complete_walk(nd); 2337bd92d7feSAl Viro 2338deb106c6SAl Viro if (!err && nd->flags & LOOKUP_DIRECTORY) 2339deb106c6SAl Viro if (!d_can_lookup(nd->path.dentry)) 2340bd23a539SAl Viro err = -ENOTDIR; 2341161aff1dSAl Viro if (!err && unlikely(nd->flags & LOOKUP_MOUNTPOINT)) { 2342161aff1dSAl Viro err = handle_lookup_down(nd); 2343161aff1dSAl Viro nd->flags &= ~LOOKUP_JUMPED; // no d_weak_revalidate(), please... 2344161aff1dSAl Viro } 2345625b6d10SAl Viro if (!err) { 2346625b6d10SAl Viro *path = nd->path; 2347625b6d10SAl Viro nd->path.mnt = NULL; 2348625b6d10SAl Viro nd->path.dentry = NULL; 2349625b6d10SAl Viro } 2350deb106c6SAl Viro terminate_walk(nd); 2351bd92d7feSAl Viro return err; 235231e6b01fSNick Piggin } 235331e6b01fSNick Piggin 235431d921c7SDavid Howells int filename_lookup(int dfd, struct filename *name, unsigned flags, 23559ad1aaa6SAl Viro struct path *path, struct path *root) 2356873f1eedSJeff Layton { 2357894bc8c4SAl Viro int retval; 23589883d185SAl Viro struct nameidata nd; 2359abc9f5beSAl Viro if (IS_ERR(name)) 2360abc9f5beSAl Viro return PTR_ERR(name); 23619ad1aaa6SAl Viro if (unlikely(root)) { 23629ad1aaa6SAl Viro nd.root = *root; 23639ad1aaa6SAl Viro flags |= LOOKUP_ROOT; 23649ad1aaa6SAl Viro } 23659883d185SAl Viro set_nameidata(&nd, dfd, name); 2366c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_RCU, path); 2367873f1eedSJeff Layton if (unlikely(retval == -ECHILD)) 2368c8a53ee5SAl Viro retval = path_lookupat(&nd, flags, path); 2369873f1eedSJeff Layton if (unlikely(retval == -ESTALE)) 2370c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path); 2371873f1eedSJeff Layton 2372873f1eedSJeff Layton if (likely(!retval)) 2373161aff1dSAl Viro audit_inode(name, path->dentry, 2374161aff1dSAl Viro flags & LOOKUP_MOUNTPOINT ? AUDIT_INODE_NOEVAL : 0); 23759883d185SAl Viro restore_nameidata(); 2376e4bd1c1aSAl Viro putname(name); 2377873f1eedSJeff Layton return retval; 2378873f1eedSJeff Layton } 2379873f1eedSJeff Layton 23808bcb77faSAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2381c8a53ee5SAl Viro static int path_parentat(struct nameidata *nd, unsigned flags, 2382391172c4SAl Viro struct path *parent) 23838bcb77faSAl Viro { 2384c8a53ee5SAl Viro const char *s = path_init(nd, flags); 23859b5858e9SAl Viro int err = link_path_walk(s, nd); 23868bcb77faSAl Viro if (!err) 23878bcb77faSAl Viro err = complete_walk(nd); 2388391172c4SAl Viro if (!err) { 2389391172c4SAl Viro *parent = nd->path; 2390391172c4SAl Viro nd->path.mnt = NULL; 2391391172c4SAl Viro nd->path.dentry = NULL; 2392391172c4SAl Viro } 2393deb106c6SAl Viro terminate_walk(nd); 23948bcb77faSAl Viro return err; 23958bcb77faSAl Viro } 23968bcb77faSAl Viro 23975c31b6ceSAl Viro static struct filename *filename_parentat(int dfd, struct filename *name, 2398391172c4SAl Viro unsigned int flags, struct path *parent, 2399391172c4SAl Viro struct qstr *last, int *type) 24008bcb77faSAl Viro { 24018bcb77faSAl Viro int retval; 24029883d185SAl Viro struct nameidata nd; 24038bcb77faSAl Viro 24045c31b6ceSAl Viro if (IS_ERR(name)) 24055c31b6ceSAl Viro return name; 24069883d185SAl Viro set_nameidata(&nd, dfd, name); 2407c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_RCU, parent); 24088bcb77faSAl Viro if (unlikely(retval == -ECHILD)) 2409c8a53ee5SAl Viro retval = path_parentat(&nd, flags, parent); 24108bcb77faSAl Viro if (unlikely(retval == -ESTALE)) 2411c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent); 2412391172c4SAl Viro if (likely(!retval)) { 2413391172c4SAl Viro *last = nd.last; 2414391172c4SAl Viro *type = nd.last_type; 2415c9b07eabSAl Viro audit_inode(name, parent->dentry, AUDIT_INODE_PARENT); 24165c31b6ceSAl Viro } else { 24175c31b6ceSAl Viro putname(name); 24185c31b6ceSAl Viro name = ERR_PTR(retval); 2419391172c4SAl Viro } 24209883d185SAl Viro restore_nameidata(); 24215c31b6ceSAl Viro return name; 24228bcb77faSAl Viro } 24238bcb77faSAl Viro 242479714f72SAl Viro /* does lookup, returns the object with parent locked */ 242579714f72SAl Viro struct dentry *kern_path_locked(const char *name, struct path *path) 24265590ff0dSUlrich Drepper { 24275c31b6ceSAl Viro struct filename *filename; 24285c31b6ceSAl Viro struct dentry *d; 2429391172c4SAl Viro struct qstr last; 2430391172c4SAl Viro int type; 243151689104SPaul Moore 24325c31b6ceSAl Viro filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path, 24335c31b6ceSAl Viro &last, &type); 243451689104SPaul Moore if (IS_ERR(filename)) 243551689104SPaul Moore return ERR_CAST(filename); 24365c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) { 2437391172c4SAl Viro path_put(path); 24385c31b6ceSAl Viro putname(filename); 24395c31b6ceSAl Viro return ERR_PTR(-EINVAL); 244079714f72SAl Viro } 24415955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 2442391172c4SAl Viro d = __lookup_hash(&last, path->dentry, 0); 244379714f72SAl Viro if (IS_ERR(d)) { 24445955102cSAl Viro inode_unlock(path->dentry->d_inode); 2445391172c4SAl Viro path_put(path); 244679714f72SAl Viro } 244751689104SPaul Moore putname(filename); 244879714f72SAl Viro return d; 24495590ff0dSUlrich Drepper } 24505590ff0dSUlrich Drepper 2451d1811465SAl Viro int kern_path(const char *name, unsigned int flags, struct path *path) 2452d1811465SAl Viro { 2453abc9f5beSAl Viro return filename_lookup(AT_FDCWD, getname_kernel(name), 2454abc9f5beSAl Viro flags, path, NULL); 2455d1811465SAl Viro } 24564d359507SAl Viro EXPORT_SYMBOL(kern_path); 2457d1811465SAl Viro 245816f18200SJosef 'Jeff' Sipek /** 245916f18200SJosef 'Jeff' Sipek * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair 246016f18200SJosef 'Jeff' Sipek * @dentry: pointer to dentry of the base directory 246116f18200SJosef 'Jeff' Sipek * @mnt: pointer to vfs mount of the base directory 246216f18200SJosef 'Jeff' Sipek * @name: pointer to file name 246316f18200SJosef 'Jeff' Sipek * @flags: lookup flags 2464e0a01249SAl Viro * @path: pointer to struct path to fill 246516f18200SJosef 'Jeff' Sipek */ 246616f18200SJosef 'Jeff' Sipek int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, 246716f18200SJosef 'Jeff' Sipek const char *name, unsigned int flags, 2468e0a01249SAl Viro struct path *path) 246916f18200SJosef 'Jeff' Sipek { 24709ad1aaa6SAl Viro struct path root = {.mnt = mnt, .dentry = dentry}; 24719ad1aaa6SAl Viro /* the first argument of filename_lookup() is ignored with root */ 2472abc9f5beSAl Viro return filename_lookup(AT_FDCWD, getname_kernel(name), 2473abc9f5beSAl Viro flags , path, &root); 247416f18200SJosef 'Jeff' Sipek } 24754d359507SAl Viro EXPORT_SYMBOL(vfs_path_lookup); 247616f18200SJosef 'Jeff' Sipek 24773c95f0dcSAl Viro static int lookup_one_len_common(const char *name, struct dentry *base, 24783c95f0dcSAl Viro int len, struct qstr *this) 24793c95f0dcSAl Viro { 24803c95f0dcSAl Viro this->name = name; 24813c95f0dcSAl Viro this->len = len; 24823c95f0dcSAl Viro this->hash = full_name_hash(base, name, len); 24833c95f0dcSAl Viro if (!len) 24843c95f0dcSAl Viro return -EACCES; 24853c95f0dcSAl Viro 24863c95f0dcSAl Viro if (unlikely(name[0] == '.')) { 24873c95f0dcSAl Viro if (len < 2 || (len == 2 && name[1] == '.')) 24883c95f0dcSAl Viro return -EACCES; 24893c95f0dcSAl Viro } 24903c95f0dcSAl Viro 24913c95f0dcSAl Viro while (len--) { 24923c95f0dcSAl Viro unsigned int c = *(const unsigned char *)name++; 24933c95f0dcSAl Viro if (c == '/' || c == '\0') 24943c95f0dcSAl Viro return -EACCES; 24953c95f0dcSAl Viro } 24963c95f0dcSAl Viro /* 24973c95f0dcSAl Viro * See if the low-level filesystem might want 24983c95f0dcSAl Viro * to use its own hash.. 24993c95f0dcSAl Viro */ 25003c95f0dcSAl Viro if (base->d_flags & DCACHE_OP_HASH) { 25013c95f0dcSAl Viro int err = base->d_op->d_hash(base, this); 25023c95f0dcSAl Viro if (err < 0) 25033c95f0dcSAl Viro return err; 25043c95f0dcSAl Viro } 25053c95f0dcSAl Viro 25063c95f0dcSAl Viro return inode_permission(base->d_inode, MAY_EXEC); 25073c95f0dcSAl Viro } 25083c95f0dcSAl Viro 2509eead1911SChristoph Hellwig /** 25100da0b7fdSDavid Howells * try_lookup_one_len - filesystem helper to lookup single pathname component 25110da0b7fdSDavid Howells * @name: pathname component to lookup 25120da0b7fdSDavid Howells * @base: base directory to lookup from 25130da0b7fdSDavid Howells * @len: maximum length @len should be interpreted to 25140da0b7fdSDavid Howells * 25150da0b7fdSDavid Howells * Look up a dentry by name in the dcache, returning NULL if it does not 25160da0b7fdSDavid Howells * currently exist. The function does not try to create a dentry. 25170da0b7fdSDavid Howells * 25180da0b7fdSDavid Howells * Note that this routine is purely a helper for filesystem usage and should 25190da0b7fdSDavid Howells * not be called by generic code. 25200da0b7fdSDavid Howells * 25210da0b7fdSDavid Howells * The caller must hold base->i_mutex. 25220da0b7fdSDavid Howells */ 25230da0b7fdSDavid Howells struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len) 25240da0b7fdSDavid Howells { 25250da0b7fdSDavid Howells struct qstr this; 25260da0b7fdSDavid Howells int err; 25270da0b7fdSDavid Howells 25280da0b7fdSDavid Howells WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 25290da0b7fdSDavid Howells 25300da0b7fdSDavid Howells err = lookup_one_len_common(name, base, len, &this); 25310da0b7fdSDavid Howells if (err) 25320da0b7fdSDavid Howells return ERR_PTR(err); 25330da0b7fdSDavid Howells 25340da0b7fdSDavid Howells return lookup_dcache(&this, base, 0); 25350da0b7fdSDavid Howells } 25360da0b7fdSDavid Howells EXPORT_SYMBOL(try_lookup_one_len); 25370da0b7fdSDavid Howells 25380da0b7fdSDavid Howells /** 2539a6b91919SRandy Dunlap * lookup_one_len - filesystem helper to lookup single pathname component 2540eead1911SChristoph Hellwig * @name: pathname component to lookup 2541eead1911SChristoph Hellwig * @base: base directory to lookup from 2542eead1911SChristoph Hellwig * @len: maximum length @len should be interpreted to 2543eead1911SChristoph Hellwig * 2544a6b91919SRandy Dunlap * Note that this routine is purely a helper for filesystem usage and should 25459e7543e9SAl Viro * not be called by generic code. 2546bbddca8eSNeilBrown * 2547bbddca8eSNeilBrown * The caller must hold base->i_mutex. 2548eead1911SChristoph Hellwig */ 2549057f6c01SJames Morris struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 2550057f6c01SJames Morris { 25518613a209SAl Viro struct dentry *dentry; 2552057f6c01SJames Morris struct qstr this; 2553cda309deSMiklos Szeredi int err; 2554057f6c01SJames Morris 25555955102cSAl Viro WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 25562f9092e1SDavid Woodhouse 25573c95f0dcSAl Viro err = lookup_one_len_common(name, base, len, &this); 2558cda309deSMiklos Szeredi if (err) 2559cda309deSMiklos Szeredi return ERR_PTR(err); 2560cda309deSMiklos Szeredi 25618613a209SAl Viro dentry = lookup_dcache(&this, base, 0); 25628613a209SAl Viro return dentry ? dentry : __lookup_slow(&this, base, 0); 2563057f6c01SJames Morris } 25644d359507SAl Viro EXPORT_SYMBOL(lookup_one_len); 2565057f6c01SJames Morris 2566bbddca8eSNeilBrown /** 2567bbddca8eSNeilBrown * lookup_one_len_unlocked - filesystem helper to lookup single pathname component 2568bbddca8eSNeilBrown * @name: pathname component to lookup 2569bbddca8eSNeilBrown * @base: base directory to lookup from 2570bbddca8eSNeilBrown * @len: maximum length @len should be interpreted to 2571bbddca8eSNeilBrown * 2572bbddca8eSNeilBrown * Note that this routine is purely a helper for filesystem usage and should 2573bbddca8eSNeilBrown * not be called by generic code. 2574bbddca8eSNeilBrown * 2575bbddca8eSNeilBrown * Unlike lookup_one_len, it should be called without the parent 2576bbddca8eSNeilBrown * i_mutex held, and will take the i_mutex itself if necessary. 2577bbddca8eSNeilBrown */ 2578bbddca8eSNeilBrown struct dentry *lookup_one_len_unlocked(const char *name, 2579bbddca8eSNeilBrown struct dentry *base, int len) 2580bbddca8eSNeilBrown { 2581bbddca8eSNeilBrown struct qstr this; 2582bbddca8eSNeilBrown int err; 258320d00ee8SLinus Torvalds struct dentry *ret; 2584bbddca8eSNeilBrown 25853c95f0dcSAl Viro err = lookup_one_len_common(name, base, len, &this); 2586bbddca8eSNeilBrown if (err) 2587bbddca8eSNeilBrown return ERR_PTR(err); 2588bbddca8eSNeilBrown 258920d00ee8SLinus Torvalds ret = lookup_dcache(&this, base, 0); 259020d00ee8SLinus Torvalds if (!ret) 259120d00ee8SLinus Torvalds ret = lookup_slow(&this, base, 0); 259220d00ee8SLinus Torvalds return ret; 2593bbddca8eSNeilBrown } 2594bbddca8eSNeilBrown EXPORT_SYMBOL(lookup_one_len_unlocked); 2595bbddca8eSNeilBrown 25966c2d4798SAl Viro /* 25976c2d4798SAl Viro * Like lookup_one_len_unlocked(), except that it yields ERR_PTR(-ENOENT) 25986c2d4798SAl Viro * on negatives. Returns known positive or ERR_PTR(); that's what 25996c2d4798SAl Viro * most of the users want. Note that pinned negative with unlocked parent 26006c2d4798SAl Viro * _can_ become positive at any time, so callers of lookup_one_len_unlocked() 26016c2d4798SAl Viro * need to be very careful; pinned positives have ->d_inode stable, so 26026c2d4798SAl Viro * this one avoids such problems. 26036c2d4798SAl Viro */ 26046c2d4798SAl Viro struct dentry *lookup_positive_unlocked(const char *name, 26056c2d4798SAl Viro struct dentry *base, int len) 26066c2d4798SAl Viro { 26076c2d4798SAl Viro struct dentry *ret = lookup_one_len_unlocked(name, base, len); 26082fa6b1e0SAl Viro if (!IS_ERR(ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) { 26096c2d4798SAl Viro dput(ret); 26106c2d4798SAl Viro ret = ERR_PTR(-ENOENT); 26116c2d4798SAl Viro } 26126c2d4798SAl Viro return ret; 26136c2d4798SAl Viro } 26146c2d4798SAl Viro EXPORT_SYMBOL(lookup_positive_unlocked); 26156c2d4798SAl Viro 2616eedf265aSEric W. Biederman #ifdef CONFIG_UNIX98_PTYS 2617eedf265aSEric W. Biederman int path_pts(struct path *path) 2618eedf265aSEric W. Biederman { 2619eedf265aSEric W. Biederman /* Find something mounted on "pts" in the same directory as 2620eedf265aSEric W. Biederman * the input path. 2621eedf265aSEric W. Biederman */ 2622a6a7eb76SAl Viro struct dentry *parent = dget_parent(path->dentry); 2623a6a7eb76SAl Viro struct dentry *child; 262419f6028aSAl Viro struct qstr this = QSTR_INIT("pts", 3); 2625eedf265aSEric W. Biederman 2626a6a7eb76SAl Viro if (unlikely(!path_connected(path->mnt, parent))) { 2627a6a7eb76SAl Viro dput(parent); 262863b27720SAl Viro return -ENOENT; 2629a6a7eb76SAl Viro } 263063b27720SAl Viro dput(path->dentry); 263163b27720SAl Viro path->dentry = parent; 2632eedf265aSEric W. Biederman child = d_hash_and_lookup(parent, &this); 2633eedf265aSEric W. Biederman if (!child) 2634eedf265aSEric W. Biederman return -ENOENT; 2635eedf265aSEric W. Biederman 2636eedf265aSEric W. Biederman path->dentry = child; 2637eedf265aSEric W. Biederman dput(parent); 263819f6028aSAl Viro follow_down(path); 2639eedf265aSEric W. Biederman return 0; 2640eedf265aSEric W. Biederman } 2641eedf265aSEric W. Biederman #endif 2642eedf265aSEric W. Biederman 26431fa1e7f6SAndy Whitcroft int user_path_at_empty(int dfd, const char __user *name, unsigned flags, 26441fa1e7f6SAndy Whitcroft struct path *path, int *empty) 26451da177e4SLinus Torvalds { 2646abc9f5beSAl Viro return filename_lookup(dfd, getname_flags(name, flags, empty), 2647abc9f5beSAl Viro flags, path, NULL); 26481da177e4SLinus Torvalds } 2649b853a161SAl Viro EXPORT_SYMBOL(user_path_at_empty); 26501fa1e7f6SAndy Whitcroft 2651cbdf35bcSMiklos Szeredi int __check_sticky(struct inode *dir, struct inode *inode) 26521da177e4SLinus Torvalds { 26538e96e3b7SEric W. Biederman kuid_t fsuid = current_fsuid(); 2654da9592edSDavid Howells 26558e96e3b7SEric W. Biederman if (uid_eq(inode->i_uid, fsuid)) 26561da177e4SLinus Torvalds return 0; 26578e96e3b7SEric W. Biederman if (uid_eq(dir->i_uid, fsuid)) 26581da177e4SLinus Torvalds return 0; 265923adbe12SAndy Lutomirski return !capable_wrt_inode_uidgid(inode, CAP_FOWNER); 26601da177e4SLinus Torvalds } 2661cbdf35bcSMiklos Szeredi EXPORT_SYMBOL(__check_sticky); 26621da177e4SLinus Torvalds 26631da177e4SLinus Torvalds /* 26641da177e4SLinus Torvalds * Check whether we can remove a link victim from directory dir, check 26651da177e4SLinus Torvalds * whether the type of victim is right. 26661da177e4SLinus Torvalds * 1. We can't do it if dir is read-only (done in permission()) 26671da177e4SLinus Torvalds * 2. We should have write and exec permissions on dir 26681da177e4SLinus Torvalds * 3. We can't remove anything from append-only dir 26691da177e4SLinus Torvalds * 4. We can't do anything with immutable dir (done in permission()) 26701da177e4SLinus Torvalds * 5. If the sticky bit on dir is set we should either 26711da177e4SLinus Torvalds * a. be owner of dir, or 26721da177e4SLinus Torvalds * b. be owner of victim, or 26731da177e4SLinus Torvalds * c. have CAP_FOWNER capability 26741da177e4SLinus Torvalds * 6. If the victim is append-only or immutable we can't do antyhing with 26751da177e4SLinus Torvalds * links pointing to it. 26760bd23d09SEric W. Biederman * 7. If the victim has an unknown uid or gid we can't change the inode. 26770bd23d09SEric W. Biederman * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR. 26780bd23d09SEric W. Biederman * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR. 26790bd23d09SEric W. Biederman * 10. We can't remove a root or mountpoint. 26800bd23d09SEric W. Biederman * 11. We don't allow removal of NFS sillyrenamed files; it's handled by 26811da177e4SLinus Torvalds * nfs_async_unlink(). 26821da177e4SLinus Torvalds */ 2683b18825a7SDavid Howells static int may_delete(struct inode *dir, struct dentry *victim, bool isdir) 26841da177e4SLinus Torvalds { 268563afdfc7SDavid Howells struct inode *inode = d_backing_inode(victim); 26861da177e4SLinus Torvalds int error; 26871da177e4SLinus Torvalds 2688b18825a7SDavid Howells if (d_is_negative(victim)) 26891da177e4SLinus Torvalds return -ENOENT; 2690b18825a7SDavid Howells BUG_ON(!inode); 26911da177e4SLinus Torvalds 26921da177e4SLinus Torvalds BUG_ON(victim->d_parent->d_inode != dir); 2693593d1ce8SEric W. Biederman 2694593d1ce8SEric W. Biederman /* Inode writeback is not safe when the uid or gid are invalid. */ 2695593d1ce8SEric W. Biederman if (!uid_valid(inode->i_uid) || !gid_valid(inode->i_gid)) 2696593d1ce8SEric W. Biederman return -EOVERFLOW; 2697593d1ce8SEric W. Biederman 26984fa6b5ecSJeff Layton audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); 26991da177e4SLinus Torvalds 2700f419a2e3SAl Viro error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 27011da177e4SLinus Torvalds if (error) 27021da177e4SLinus Torvalds return error; 27031da177e4SLinus Torvalds if (IS_APPEND(dir)) 27041da177e4SLinus Torvalds return -EPERM; 2705b18825a7SDavid Howells 2706b18825a7SDavid Howells if (check_sticky(dir, inode) || IS_APPEND(inode) || 27070bd23d09SEric W. Biederman IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || HAS_UNMAPPED_ID(inode)) 27081da177e4SLinus Torvalds return -EPERM; 27091da177e4SLinus Torvalds if (isdir) { 271044b1d530SMiklos Szeredi if (!d_is_dir(victim)) 27111da177e4SLinus Torvalds return -ENOTDIR; 27121da177e4SLinus Torvalds if (IS_ROOT(victim)) 27131da177e4SLinus Torvalds return -EBUSY; 271444b1d530SMiklos Szeredi } else if (d_is_dir(victim)) 27151da177e4SLinus Torvalds return -EISDIR; 27161da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 27171da177e4SLinus Torvalds return -ENOENT; 27181da177e4SLinus Torvalds if (victim->d_flags & DCACHE_NFSFS_RENAMED) 27191da177e4SLinus Torvalds return -EBUSY; 27201da177e4SLinus Torvalds return 0; 27211da177e4SLinus Torvalds } 27221da177e4SLinus Torvalds 27231da177e4SLinus Torvalds /* Check whether we can create an object with dentry child in directory 27241da177e4SLinus Torvalds * dir. 27251da177e4SLinus Torvalds * 1. We can't do it if child already exists (open has special treatment for 27261da177e4SLinus Torvalds * this case, but since we are inlined it's OK) 27271da177e4SLinus Torvalds * 2. We can't do it if dir is read-only (done in permission()) 2728036d5236SEric W. Biederman * 3. We can't do it if the fs can't represent the fsuid or fsgid. 2729036d5236SEric W. Biederman * 4. We should have write and exec permissions on dir 2730036d5236SEric W. Biederman * 5. We can't do it if dir is immutable (done in permission()) 27311da177e4SLinus Torvalds */ 2732a95164d9SMiklos Szeredi static inline int may_create(struct inode *dir, struct dentry *child) 27331da177e4SLinus Torvalds { 2734036d5236SEric W. Biederman struct user_namespace *s_user_ns; 273514e972b4SJeff Layton audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); 27361da177e4SLinus Torvalds if (child->d_inode) 27371da177e4SLinus Torvalds return -EEXIST; 27381da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 27391da177e4SLinus Torvalds return -ENOENT; 2740036d5236SEric W. Biederman s_user_ns = dir->i_sb->s_user_ns; 2741036d5236SEric W. Biederman if (!kuid_has_mapping(s_user_ns, current_fsuid()) || 2742036d5236SEric W. Biederman !kgid_has_mapping(s_user_ns, current_fsgid())) 2743036d5236SEric W. Biederman return -EOVERFLOW; 2744f419a2e3SAl Viro return inode_permission(dir, MAY_WRITE | MAY_EXEC); 27451da177e4SLinus Torvalds } 27461da177e4SLinus Torvalds 27471da177e4SLinus Torvalds /* 27481da177e4SLinus Torvalds * p1 and p2 should be directories on the same fs. 27491da177e4SLinus Torvalds */ 27501da177e4SLinus Torvalds struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 27511da177e4SLinus Torvalds { 27521da177e4SLinus Torvalds struct dentry *p; 27531da177e4SLinus Torvalds 27541da177e4SLinus Torvalds if (p1 == p2) { 27555955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 27561da177e4SLinus Torvalds return NULL; 27571da177e4SLinus Torvalds } 27581da177e4SLinus Torvalds 2759fc64005cSAl Viro mutex_lock(&p1->d_sb->s_vfs_rename_mutex); 27601da177e4SLinus Torvalds 2761e2761a11SOGAWA Hirofumi p = d_ancestor(p2, p1); 2762e2761a11SOGAWA Hirofumi if (p) { 27635955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT); 27645955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_CHILD); 27651da177e4SLinus Torvalds return p; 27661da177e4SLinus Torvalds } 27671da177e4SLinus Torvalds 2768e2761a11SOGAWA Hirofumi p = d_ancestor(p1, p2); 2769e2761a11SOGAWA Hirofumi if (p) { 27705955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 27715955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_CHILD); 27721da177e4SLinus Torvalds return p; 27731da177e4SLinus Torvalds } 27741da177e4SLinus Torvalds 27755955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 27765955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2); 27771da177e4SLinus Torvalds return NULL; 27781da177e4SLinus Torvalds } 27794d359507SAl Viro EXPORT_SYMBOL(lock_rename); 27801da177e4SLinus Torvalds 27811da177e4SLinus Torvalds void unlock_rename(struct dentry *p1, struct dentry *p2) 27821da177e4SLinus Torvalds { 27835955102cSAl Viro inode_unlock(p1->d_inode); 27841da177e4SLinus Torvalds if (p1 != p2) { 27855955102cSAl Viro inode_unlock(p2->d_inode); 2786fc64005cSAl Viro mutex_unlock(&p1->d_sb->s_vfs_rename_mutex); 27871da177e4SLinus Torvalds } 27881da177e4SLinus Torvalds } 27894d359507SAl Viro EXPORT_SYMBOL(unlock_rename); 27901da177e4SLinus Torvalds 27914acdaf27SAl Viro int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, 2792312b63fbSAl Viro bool want_excl) 27931da177e4SLinus Torvalds { 2794a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 27951da177e4SLinus Torvalds if (error) 27961da177e4SLinus Torvalds return error; 27971da177e4SLinus Torvalds 2798acfa4380SAl Viro if (!dir->i_op->create) 27991da177e4SLinus Torvalds return -EACCES; /* shouldn't it be ENOSYS? */ 28001da177e4SLinus Torvalds mode &= S_IALLUGO; 28011da177e4SLinus Torvalds mode |= S_IFREG; 28021da177e4SLinus Torvalds error = security_inode_create(dir, dentry, mode); 28031da177e4SLinus Torvalds if (error) 28041da177e4SLinus Torvalds return error; 2805312b63fbSAl Viro error = dir->i_op->create(dir, dentry, mode, want_excl); 2806a74574aaSStephen Smalley if (!error) 2807f38aa942SAmy Griffis fsnotify_create(dir, dentry); 28081da177e4SLinus Torvalds return error; 28091da177e4SLinus Torvalds } 28104d359507SAl Viro EXPORT_SYMBOL(vfs_create); 28111da177e4SLinus Torvalds 28128e6c848eSAl Viro int vfs_mkobj(struct dentry *dentry, umode_t mode, 28138e6c848eSAl Viro int (*f)(struct dentry *, umode_t, void *), 28148e6c848eSAl Viro void *arg) 28158e6c848eSAl Viro { 28168e6c848eSAl Viro struct inode *dir = dentry->d_parent->d_inode; 28178e6c848eSAl Viro int error = may_create(dir, dentry); 28188e6c848eSAl Viro if (error) 28198e6c848eSAl Viro return error; 28208e6c848eSAl Viro 28218e6c848eSAl Viro mode &= S_IALLUGO; 28228e6c848eSAl Viro mode |= S_IFREG; 28238e6c848eSAl Viro error = security_inode_create(dir, dentry, mode); 28248e6c848eSAl Viro if (error) 28258e6c848eSAl Viro return error; 28268e6c848eSAl Viro error = f(dentry, mode, arg); 28278e6c848eSAl Viro if (!error) 28288e6c848eSAl Viro fsnotify_create(dir, dentry); 28298e6c848eSAl Viro return error; 28308e6c848eSAl Viro } 28318e6c848eSAl Viro EXPORT_SYMBOL(vfs_mkobj); 28328e6c848eSAl Viro 2833a2982cc9SEric W. Biederman bool may_open_dev(const struct path *path) 2834a2982cc9SEric W. Biederman { 2835a2982cc9SEric W. Biederman return !(path->mnt->mnt_flags & MNT_NODEV) && 2836a2982cc9SEric W. Biederman !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV); 2837a2982cc9SEric W. Biederman } 2838a2982cc9SEric W. Biederman 2839f0bb5aafSAl Viro static int may_open(const struct path *path, int acc_mode, int flag) 28401da177e4SLinus Torvalds { 28413fb64190SChristoph Hellwig struct dentry *dentry = path->dentry; 28421da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 28431da177e4SLinus Torvalds int error; 28441da177e4SLinus Torvalds 28451da177e4SLinus Torvalds if (!inode) 28461da177e4SLinus Torvalds return -ENOENT; 28471da177e4SLinus Torvalds 2848c8fe8f30SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 2849c8fe8f30SChristoph Hellwig case S_IFLNK: 28501da177e4SLinus Torvalds return -ELOOP; 2851c8fe8f30SChristoph Hellwig case S_IFDIR: 2852fc4177beSKees Cook if (acc_mode & MAY_WRITE) 28531da177e4SLinus Torvalds return -EISDIR; 2854fc4177beSKees Cook if (acc_mode & MAY_EXEC) 2855fc4177beSKees Cook return -EACCES; 2856c8fe8f30SChristoph Hellwig break; 2857c8fe8f30SChristoph Hellwig case S_IFBLK: 2858c8fe8f30SChristoph Hellwig case S_IFCHR: 2859a2982cc9SEric W. Biederman if (!may_open_dev(path)) 28601da177e4SLinus Torvalds return -EACCES; 2861633fb6acSKees Cook fallthrough; 2862c8fe8f30SChristoph Hellwig case S_IFIFO: 2863c8fe8f30SChristoph Hellwig case S_IFSOCK: 2864633fb6acSKees Cook if (acc_mode & MAY_EXEC) 2865633fb6acSKees Cook return -EACCES; 28661da177e4SLinus Torvalds flag &= ~O_TRUNC; 2867c8fe8f30SChristoph Hellwig break; 28680fd338b2SKees Cook case S_IFREG: 28690fd338b2SKees Cook if ((acc_mode & MAY_EXEC) && path_noexec(path)) 28700fd338b2SKees Cook return -EACCES; 28710fd338b2SKees Cook break; 28724a3fd211SDave Hansen } 2873b41572e9SDave Hansen 287462fb4a15SAl Viro error = inode_permission(inode, MAY_OPEN | acc_mode); 2875b41572e9SDave Hansen if (error) 2876b41572e9SDave Hansen return error; 28776146f0d5SMimi Zohar 28781da177e4SLinus Torvalds /* 28791da177e4SLinus Torvalds * An append-only file must be opened in append mode for writing. 28801da177e4SLinus Torvalds */ 28811da177e4SLinus Torvalds if (IS_APPEND(inode)) { 28828737c930SAl Viro if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) 28837715b521SAl Viro return -EPERM; 28841da177e4SLinus Torvalds if (flag & O_TRUNC) 28857715b521SAl Viro return -EPERM; 28861da177e4SLinus Torvalds } 28871da177e4SLinus Torvalds 28881da177e4SLinus Torvalds /* O_NOATIME can only be set by the owner or superuser */ 28892e149670SSerge E. Hallyn if (flag & O_NOATIME && !inode_owner_or_capable(inode)) 28907715b521SAl Viro return -EPERM; 28911da177e4SLinus Torvalds 2892f3c7691eSJ. Bruce Fields return 0; 28937715b521SAl Viro } 28947715b521SAl Viro 2895e1181ee6SJeff Layton static int handle_truncate(struct file *filp) 28967715b521SAl Viro { 2897f0bb5aafSAl Viro const struct path *path = &filp->f_path; 28987715b521SAl Viro struct inode *inode = path->dentry->d_inode; 28997715b521SAl Viro int error = get_write_access(inode); 29001da177e4SLinus Torvalds if (error) 29017715b521SAl Viro return error; 29021da177e4SLinus Torvalds /* 29031da177e4SLinus Torvalds * Refuse to truncate files with mandatory locks held on them. 29041da177e4SLinus Torvalds */ 2905d7a06983SJeff Layton error = locks_verify_locked(filp); 2906be6d3e56SKentaro Takeda if (!error) 2907ea0d3ab2STetsuo Handa error = security_path_truncate(path); 29081da177e4SLinus Torvalds if (!error) { 29097715b521SAl Viro error = do_truncate(path->dentry, 0, 2910d139d7ffSMiklos Szeredi ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, 2911e1181ee6SJeff Layton filp); 29121da177e4SLinus Torvalds } 29131da177e4SLinus Torvalds put_write_access(inode); 2914acd0c935SMimi Zohar return error; 29151da177e4SLinus Torvalds } 29161da177e4SLinus Torvalds 2917d57999e1SDave Hansen static inline int open_to_namei_flags(int flag) 2918d57999e1SDave Hansen { 29198a5e929dSAl Viro if ((flag & O_ACCMODE) == 3) 29208a5e929dSAl Viro flag--; 2921d57999e1SDave Hansen return flag; 2922d57999e1SDave Hansen } 2923d57999e1SDave Hansen 2924d3607752SAl Viro static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode) 2925d18e9008SMiklos Szeredi { 29261328c727SSeth Forshee struct user_namespace *s_user_ns; 2927d18e9008SMiklos Szeredi int error = security_path_mknod(dir, dentry, mode, 0); 2928d18e9008SMiklos Szeredi if (error) 2929d18e9008SMiklos Szeredi return error; 2930d18e9008SMiklos Szeredi 29311328c727SSeth Forshee s_user_ns = dir->dentry->d_sb->s_user_ns; 29321328c727SSeth Forshee if (!kuid_has_mapping(s_user_ns, current_fsuid()) || 29331328c727SSeth Forshee !kgid_has_mapping(s_user_ns, current_fsgid())) 29341328c727SSeth Forshee return -EOVERFLOW; 29351328c727SSeth Forshee 2936d18e9008SMiklos Szeredi error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC); 2937d18e9008SMiklos Szeredi if (error) 2938d18e9008SMiklos Szeredi return error; 2939d18e9008SMiklos Szeredi 2940d18e9008SMiklos Szeredi return security_inode_create(dir->dentry->d_inode, dentry, mode); 2941d18e9008SMiklos Szeredi } 2942d18e9008SMiklos Szeredi 29431acf0af9SDavid Howells /* 29441acf0af9SDavid Howells * Attempt to atomically look up, create and open a file from a negative 29451acf0af9SDavid Howells * dentry. 29461acf0af9SDavid Howells * 29471acf0af9SDavid Howells * Returns 0 if successful. The file will have been created and attached to 29481acf0af9SDavid Howells * @file by the filesystem calling finish_open(). 29491acf0af9SDavid Howells * 295000a07c15SAl Viro * If the file was looked up only or didn't need creating, FMODE_OPENED won't 295100a07c15SAl Viro * be set. The caller will need to perform the open themselves. @path will 295200a07c15SAl Viro * have been updated to point to the new dentry. This may be negative. 29531acf0af9SDavid Howells * 29541acf0af9SDavid Howells * Returns an error code otherwise. 29551acf0af9SDavid Howells */ 2956239eb983SAl Viro static struct dentry *atomic_open(struct nameidata *nd, struct dentry *dentry, 2957239eb983SAl Viro struct file *file, 29583ec2eef1SAl Viro int open_flag, umode_t mode) 2959d18e9008SMiklos Szeredi { 2960d18e9008SMiklos Szeredi struct dentry *const DENTRY_NOT_SET = (void *) -1UL; 2961d18e9008SMiklos Szeredi struct inode *dir = nd->path.dentry->d_inode; 2962d18e9008SMiklos Szeredi int error; 2963d18e9008SMiklos Szeredi 2964d18e9008SMiklos Szeredi if (nd->flags & LOOKUP_DIRECTORY) 2965d18e9008SMiklos Szeredi open_flag |= O_DIRECTORY; 2966d18e9008SMiklos Szeredi 296730d90494SAl Viro file->f_path.dentry = DENTRY_NOT_SET; 296830d90494SAl Viro file->f_path.mnt = nd->path.mnt; 29690fb1ea09SAl Viro error = dir->i_op->atomic_open(dir, dentry, file, 297044907d79SAl Viro open_to_namei_flags(open_flag), mode); 29716fbd0714SAl Viro d_lookup_done(dentry); 2972384f26e2SAl Viro if (!error) { 297364e1ac4dSAl Viro if (file->f_mode & FMODE_OPENED) { 29746fb968cdSAl Viro if (unlikely(dentry != file->f_path.dentry)) { 29756fb968cdSAl Viro dput(dentry); 29766fb968cdSAl Viro dentry = dget(file->f_path.dentry); 29776fb968cdSAl Viro } 297864e1ac4dSAl Viro } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { 29792675a4ebSAl Viro error = -EIO; 2980384f26e2SAl Viro } else { 298130d90494SAl Viro if (file->f_path.dentry) { 2982d18e9008SMiklos Szeredi dput(dentry); 2983d18e9008SMiklos Szeredi dentry = file->f_path.dentry; 298410c64ceaSAl Viro } 2985239eb983SAl Viro if (unlikely(d_is_negative(dentry))) 2986a01e718fSAl Viro error = -ENOENT; 2987d18e9008SMiklos Szeredi } 2988d18e9008SMiklos Szeredi } 2989239eb983SAl Viro if (error) { 2990d18e9008SMiklos Szeredi dput(dentry); 2991239eb983SAl Viro dentry = ERR_PTR(error); 2992239eb983SAl Viro } 2993239eb983SAl Viro return dentry; 2994d18e9008SMiklos Szeredi } 2995d18e9008SMiklos Szeredi 299631e6b01fSNick Piggin /* 29971acf0af9SDavid Howells * Look up and maybe create and open the last component. 2998d58ffd35SMiklos Szeredi * 299900a07c15SAl Viro * Must be called with parent locked (exclusive in O_CREAT case). 3000d58ffd35SMiklos Szeredi * 300100a07c15SAl Viro * Returns 0 on success, that is, if 300200a07c15SAl Viro * the file was successfully atomically created (if necessary) and opened, or 300300a07c15SAl Viro * the file was not completely opened at this time, though lookups and 300400a07c15SAl Viro * creations were performed. 300500a07c15SAl Viro * These case are distinguished by presence of FMODE_OPENED on file->f_mode. 300600a07c15SAl Viro * In the latter case dentry returned in @path might be negative if O_CREAT 300700a07c15SAl Viro * hadn't been specified. 30081acf0af9SDavid Howells * 300900a07c15SAl Viro * An error code is returned on failure. 3010d58ffd35SMiklos Szeredi */ 3011da5ebf5aSAl Viro static struct dentry *lookup_open(struct nameidata *nd, struct file *file, 3012d58ffd35SMiklos Szeredi const struct open_flags *op, 30133ec2eef1SAl Viro bool got_write) 3014d58ffd35SMiklos Szeredi { 3015d58ffd35SMiklos Szeredi struct dentry *dir = nd->path.dentry; 301654ef4872SMiklos Szeredi struct inode *dir_inode = dir->d_inode; 30171643b43fSAl Viro int open_flag = op->open_flag; 3018d58ffd35SMiklos Szeredi struct dentry *dentry; 30191643b43fSAl Viro int error, create_error = 0; 30201643b43fSAl Viro umode_t mode = op->mode; 30216fbd0714SAl Viro DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 3022d58ffd35SMiklos Szeredi 3023ce8644fcSAl Viro if (unlikely(IS_DEADDIR(dir_inode))) 3024da5ebf5aSAl Viro return ERR_PTR(-ENOENT); 3025d58ffd35SMiklos Szeredi 302673a09dd9SAl Viro file->f_mode &= ~FMODE_CREATED; 30276fbd0714SAl Viro dentry = d_lookup(dir, &nd->last); 30286fbd0714SAl Viro for (;;) { 30296fbd0714SAl Viro if (!dentry) { 30306fbd0714SAl Viro dentry = d_alloc_parallel(dir, &nd->last, &wq); 3031d58ffd35SMiklos Szeredi if (IS_ERR(dentry)) 3032da5ebf5aSAl Viro return dentry; 30336fbd0714SAl Viro } 30346fbd0714SAl Viro if (d_in_lookup(dentry)) 30356fbd0714SAl Viro break; 3036d58ffd35SMiklos Szeredi 30376fbd0714SAl Viro error = d_revalidate(dentry, nd->flags); 30386fbd0714SAl Viro if (likely(error > 0)) 30396fbd0714SAl Viro break; 30406fbd0714SAl Viro if (error) 30416fbd0714SAl Viro goto out_dput; 30426fbd0714SAl Viro d_invalidate(dentry); 30436fbd0714SAl Viro dput(dentry); 30446fbd0714SAl Viro dentry = NULL; 30456fbd0714SAl Viro } 30466fbd0714SAl Viro if (dentry->d_inode) { 3047d18e9008SMiklos Szeredi /* Cached positive dentry: will open in f_op->open */ 3048da5ebf5aSAl Viro return dentry; 30496c51e513SAl Viro } 3050d18e9008SMiklos Szeredi 30511643b43fSAl Viro /* 30521643b43fSAl Viro * Checking write permission is tricky, bacuse we don't know if we are 30531643b43fSAl Viro * going to actually need it: O_CREAT opens should work as long as the 30541643b43fSAl Viro * file exists. But checking existence breaks atomicity. The trick is 30551643b43fSAl Viro * to check access and if not granted clear O_CREAT from the flags. 30561643b43fSAl Viro * 30571643b43fSAl Viro * Another problem is returing the "right" error value (e.g. for an 30581643b43fSAl Viro * O_EXCL open we want to return EEXIST not EROFS). 30591643b43fSAl Viro */ 306099a4a90cSAl Viro if (unlikely(!got_write)) 306199a4a90cSAl Viro open_flag &= ~O_TRUNC; 30621643b43fSAl Viro if (open_flag & O_CREAT) { 306399a4a90cSAl Viro if (open_flag & O_EXCL) 306499a4a90cSAl Viro open_flag &= ~O_TRUNC; 30651643b43fSAl Viro if (!IS_POSIXACL(dir->d_inode)) 30661643b43fSAl Viro mode &= ~current_umask(); 306799a4a90cSAl Viro if (likely(got_write)) 30681643b43fSAl Viro create_error = may_o_create(&nd->path, dentry, mode); 306999a4a90cSAl Viro else 307099a4a90cSAl Viro create_error = -EROFS; 307199a4a90cSAl Viro } 307299a4a90cSAl Viro if (create_error) 30731643b43fSAl Viro open_flag &= ~O_CREAT; 30741643b43fSAl Viro if (dir_inode->i_op->atomic_open) { 3075d489cf9aSAl Viro dentry = atomic_open(nd, dentry, file, open_flag, mode); 3076da5ebf5aSAl Viro if (unlikely(create_error) && dentry == ERR_PTR(-ENOENT)) 3077da5ebf5aSAl Viro dentry = ERR_PTR(create_error); 3078da5ebf5aSAl Viro return dentry; 3079239eb983SAl Viro } 308054ef4872SMiklos Szeredi 30816fbd0714SAl Viro if (d_in_lookup(dentry)) { 308212fa5e24SAl Viro struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry, 308312fa5e24SAl Viro nd->flags); 30846fbd0714SAl Viro d_lookup_done(dentry); 308512fa5e24SAl Viro if (unlikely(res)) { 308612fa5e24SAl Viro if (IS_ERR(res)) { 308712fa5e24SAl Viro error = PTR_ERR(res); 308812fa5e24SAl Viro goto out_dput; 308912fa5e24SAl Viro } 309012fa5e24SAl Viro dput(dentry); 309112fa5e24SAl Viro dentry = res; 309212fa5e24SAl Viro } 309354ef4872SMiklos Szeredi } 309454ef4872SMiklos Szeredi 3095d58ffd35SMiklos Szeredi /* Negative dentry, just create the file */ 30961643b43fSAl Viro if (!dentry->d_inode && (open_flag & O_CREAT)) { 309773a09dd9SAl Viro file->f_mode |= FMODE_CREATED; 3098ce8644fcSAl Viro audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE); 3099ce8644fcSAl Viro if (!dir_inode->i_op->create) { 3100ce8644fcSAl Viro error = -EACCES; 3101d58ffd35SMiklos Szeredi goto out_dput; 310264894cf8SAl Viro } 3103ce8644fcSAl Viro error = dir_inode->i_op->create(dir_inode, dentry, mode, 31041643b43fSAl Viro open_flag & O_EXCL); 3105d58ffd35SMiklos Szeredi if (error) 3106d58ffd35SMiklos Szeredi goto out_dput; 3107d58ffd35SMiklos Szeredi } 31081643b43fSAl Viro if (unlikely(create_error) && !dentry->d_inode) { 31091643b43fSAl Viro error = create_error; 3110d58ffd35SMiklos Szeredi goto out_dput; 3111d58ffd35SMiklos Szeredi } 3112da5ebf5aSAl Viro return dentry; 3113d58ffd35SMiklos Szeredi 3114d58ffd35SMiklos Szeredi out_dput: 3115d58ffd35SMiklos Szeredi dput(dentry); 3116da5ebf5aSAl Viro return ERR_PTR(error); 3117d58ffd35SMiklos Szeredi } 3118d58ffd35SMiklos Szeredi 3119c981a482SAl Viro static const char *open_last_lookups(struct nameidata *nd, 31203ec2eef1SAl Viro struct file *file, const struct open_flags *op) 3121fb1cc555SAl Viro { 3122a1e28038SAl Viro struct dentry *dir = nd->path.dentry; 3123ca344a89SAl Viro int open_flag = op->open_flag; 312464894cf8SAl Viro bool got_write = false; 3125254cf582SAl Viro unsigned seq; 3126a1eb3315SMiklos Szeredi struct inode *inode; 3127da5ebf5aSAl Viro struct dentry *dentry; 3128b0417d2cSAl Viro const char *res; 312916c2cd71SAl Viro int error; 3130fb1cc555SAl Viro 3131c3e380b0SAl Viro nd->flags |= op->intent; 3132c3e380b0SAl Viro 3133bc77daa7SAl Viro if (nd->last_type != LAST_NORM) { 313456676ec3SAl Viro if (nd->depth) 313556676ec3SAl Viro put_link(nd); 3136ff326a32SAl Viro return handle_dots(nd, nd->last_type); 31371f36f774SAl Viro } 3138a2c36b45SAl Viro 3139ca344a89SAl Viro if (!(open_flag & O_CREAT)) { 3140fe2d35ffSAl Viro if (nd->last.name[nd->last.len]) 3141fe2d35ffSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 3142fe2d35ffSAl Viro /* we _can_ be in RCU mode here */ 314320e34357SAl Viro dentry = lookup_fast(nd, &inode, &seq); 314420e34357SAl Viro if (IS_ERR(dentry)) 31451ccac622SAl Viro return ERR_CAST(dentry); 314620e34357SAl Viro if (likely(dentry)) 314771574865SMiklos Szeredi goto finish_lookup; 314871574865SMiklos Szeredi 31496583fe22SAl Viro BUG_ON(nd->flags & LOOKUP_RCU); 3150b6183df7SMiklos Szeredi } else { 3151fe2d35ffSAl Viro /* create side of things */ 315272287417SAl Viro if (nd->flags & LOOKUP_RCU) { 315372287417SAl Viro error = unlazy_walk(nd); 3154c981a482SAl Viro if (unlikely(error)) 31551ccac622SAl Viro return ERR_PTR(error); 315672287417SAl Viro } 3157c9b07eabSAl Viro audit_inode(nd->name, dir, AUDIT_INODE_PARENT); 31581f36f774SAl Viro /* trailing slashes? */ 3159deb106c6SAl Viro if (unlikely(nd->last.name[nd->last.len])) 31601ccac622SAl Viro return ERR_PTR(-EISDIR); 3161b6183df7SMiklos Szeredi } 31621f36f774SAl Viro 31639cf843e3SAl Viro if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { 316464894cf8SAl Viro error = mnt_want_write(nd->path.mnt); 316564894cf8SAl Viro if (!error) 316664894cf8SAl Viro got_write = true; 316764894cf8SAl Viro /* 316864894cf8SAl Viro * do _not_ fail yet - we might not need that or fail with 316964894cf8SAl Viro * a different error; let lookup_open() decide; we'll be 317064894cf8SAl Viro * dropping this one anyway. 317164894cf8SAl Viro */ 317264894cf8SAl Viro } 31739cf843e3SAl Viro if (open_flag & O_CREAT) 31745955102cSAl Viro inode_lock(dir->d_inode); 31759cf843e3SAl Viro else 31769cf843e3SAl Viro inode_lock_shared(dir->d_inode); 3177da5ebf5aSAl Viro dentry = lookup_open(nd, file, op, got_write); 3178f7bb959dSAl Viro if (!IS_ERR(dentry) && (file->f_mode & FMODE_CREATED)) 3179f7bb959dSAl Viro fsnotify_create(dir->d_inode, dentry); 31809cf843e3SAl Viro if (open_flag & O_CREAT) 31815955102cSAl Viro inode_unlock(dir->d_inode); 31829cf843e3SAl Viro else 31839cf843e3SAl Viro inode_unlock_shared(dir->d_inode); 3184fb1cc555SAl Viro 3185c981a482SAl Viro if (got_write) 318659e96e65SAl Viro mnt_drop_write(nd->path.mnt); 31876c0d46c4SAl Viro 318859e96e65SAl Viro if (IS_ERR(dentry)) 318959e96e65SAl Viro return ERR_CAST(dentry); 319059e96e65SAl Viro 3191973d4b73SAl Viro if (file->f_mode & (FMODE_OPENED | FMODE_CREATED)) { 3192e73cabffSAl Viro dput(nd->path.dentry); 3193e73cabffSAl Viro nd->path.dentry = dentry; 3194c981a482SAl Viro return NULL; 3195fb1cc555SAl Viro } 3196fb1cc555SAl Viro 319720e34357SAl Viro finish_lookup: 319856676ec3SAl Viro if (nd->depth) 319956676ec3SAl Viro put_link(nd); 32008c4efe22SAl Viro res = step_into(nd, WALK_TRAILING, dentry, inode, seq); 3201ff326a32SAl Viro if (unlikely(res)) 32021ccac622SAl Viro nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); 3203b0417d2cSAl Viro return res; 32041ccac622SAl Viro } 320531d1726dSAl Viro 3206c981a482SAl Viro /* 3207c981a482SAl Viro * Handle the last step of open() 3208c981a482SAl Viro */ 3209c5971b8cSAl Viro static int do_open(struct nameidata *nd, 3210c981a482SAl Viro struct file *file, const struct open_flags *op) 3211c981a482SAl Viro { 3212c981a482SAl Viro int open_flag = op->open_flag; 3213c981a482SAl Viro bool do_truncate; 3214c981a482SAl Viro int acc_mode; 3215c981a482SAl Viro int error; 3216c981a482SAl Viro 3217ff326a32SAl Viro if (!(file->f_mode & (FMODE_OPENED | FMODE_CREATED))) { 3218ff326a32SAl Viro error = complete_walk(nd); 3219ff326a32SAl Viro if (error) 3220ff326a32SAl Viro return error; 3221ff326a32SAl Viro } 3222973d4b73SAl Viro if (!(file->f_mode & FMODE_CREATED)) 322376ae2a5aSAl Viro audit_inode(nd->name, nd->path.dentry, 0); 322430aba665SSalvatore Mesoraca if (open_flag & O_CREAT) { 3225b94e0b32SAl Viro if ((open_flag & O_EXCL) && !(file->f_mode & FMODE_CREATED)) 3226b94e0b32SAl Viro return -EEXIST; 322730aba665SSalvatore Mesoraca if (d_is_dir(nd->path.dentry)) 3228c5971b8cSAl Viro return -EISDIR; 32290f705953SAl Viro error = may_create_in_sticky(nd->dir_mode, nd->dir_uid, 323030aba665SSalvatore Mesoraca d_backing_inode(nd->path.dentry)); 323130aba665SSalvatore Mesoraca if (unlikely(error)) 3232c5971b8cSAl Viro return error; 323330aba665SSalvatore Mesoraca } 323444b1d530SMiklos Szeredi if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry)) 3235c5971b8cSAl Viro return -ENOTDIR; 32366c0d46c4SAl Viro 32378795e7d4SAl Viro do_truncate = false; 32388795e7d4SAl Viro acc_mode = op->acc_mode; 32395a2d3eddSAl Viro if (file->f_mode & FMODE_CREATED) { 32405a2d3eddSAl Viro /* Don't check for write permission, don't truncate */ 32415a2d3eddSAl Viro open_flag &= ~O_TRUNC; 32425a2d3eddSAl Viro acc_mode = 0; 32438795e7d4SAl Viro } else if (d_is_reg(nd->path.dentry) && open_flag & O_TRUNC) { 32440f9d1a10SAl Viro error = mnt_want_write(nd->path.mnt); 32450f9d1a10SAl Viro if (error) 3246c5971b8cSAl Viro return error; 32478795e7d4SAl Viro do_truncate = true; 32480f9d1a10SAl Viro } 3249bcda7652SAl Viro error = may_open(&nd->path, acc_mode, open_flag); 32508795e7d4SAl Viro if (!error && !(file->f_mode & FMODE_OPENED)) 3251ae2bb293SAl Viro error = vfs_open(&nd->path, file); 32528795e7d4SAl Viro if (!error) 32536035a27bSAl Viro error = ima_file_check(file, op->acc_mode); 32548795e7d4SAl Viro if (!error && do_truncate) 32552675a4ebSAl Viro error = handle_truncate(file); 3256c80567c8SAl Viro if (unlikely(error > 0)) { 3257c80567c8SAl Viro WARN_ON(1); 3258c80567c8SAl Viro error = -EINVAL; 3259c80567c8SAl Viro } 32608795e7d4SAl Viro if (do_truncate) 32610f9d1a10SAl Viro mnt_drop_write(nd->path.mnt); 3262c5971b8cSAl Viro return error; 3263fb1cc555SAl Viro } 3264fb1cc555SAl Viro 3265af7bd4dcSAmir Goldstein struct dentry *vfs_tmpfile(struct dentry *dentry, umode_t mode, int open_flag) 3266af7bd4dcSAmir Goldstein { 3267af7bd4dcSAmir Goldstein struct dentry *child = NULL; 3268af7bd4dcSAmir Goldstein struct inode *dir = dentry->d_inode; 3269af7bd4dcSAmir Goldstein struct inode *inode; 3270af7bd4dcSAmir Goldstein int error; 3271af7bd4dcSAmir Goldstein 3272af7bd4dcSAmir Goldstein /* we want directory to be writable */ 3273af7bd4dcSAmir Goldstein error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 3274af7bd4dcSAmir Goldstein if (error) 3275af7bd4dcSAmir Goldstein goto out_err; 3276af7bd4dcSAmir Goldstein error = -EOPNOTSUPP; 3277af7bd4dcSAmir Goldstein if (!dir->i_op->tmpfile) 3278af7bd4dcSAmir Goldstein goto out_err; 3279af7bd4dcSAmir Goldstein error = -ENOMEM; 3280cdf01226SDavid Howells child = d_alloc(dentry, &slash_name); 3281af7bd4dcSAmir Goldstein if (unlikely(!child)) 3282af7bd4dcSAmir Goldstein goto out_err; 3283af7bd4dcSAmir Goldstein error = dir->i_op->tmpfile(dir, child, mode); 3284af7bd4dcSAmir Goldstein if (error) 3285af7bd4dcSAmir Goldstein goto out_err; 3286af7bd4dcSAmir Goldstein error = -ENOENT; 3287af7bd4dcSAmir Goldstein inode = child->d_inode; 3288af7bd4dcSAmir Goldstein if (unlikely(!inode)) 3289af7bd4dcSAmir Goldstein goto out_err; 3290af7bd4dcSAmir Goldstein if (!(open_flag & O_EXCL)) { 3291af7bd4dcSAmir Goldstein spin_lock(&inode->i_lock); 3292af7bd4dcSAmir Goldstein inode->i_state |= I_LINKABLE; 3293af7bd4dcSAmir Goldstein spin_unlock(&inode->i_lock); 3294af7bd4dcSAmir Goldstein } 3295fdb2410fSMimi Zohar ima_post_create_tmpfile(inode); 3296af7bd4dcSAmir Goldstein return child; 3297af7bd4dcSAmir Goldstein 3298af7bd4dcSAmir Goldstein out_err: 3299af7bd4dcSAmir Goldstein dput(child); 3300af7bd4dcSAmir Goldstein return ERR_PTR(error); 3301af7bd4dcSAmir Goldstein } 3302af7bd4dcSAmir Goldstein EXPORT_SYMBOL(vfs_tmpfile); 3303af7bd4dcSAmir Goldstein 3304c8a53ee5SAl Viro static int do_tmpfile(struct nameidata *nd, unsigned flags, 330560545d0dSAl Viro const struct open_flags *op, 33063ec2eef1SAl Viro struct file *file) 330760545d0dSAl Viro { 3308625b6d10SAl Viro struct dentry *child; 3309625b6d10SAl Viro struct path path; 3310c8a53ee5SAl Viro int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path); 331160545d0dSAl Viro if (unlikely(error)) 331260545d0dSAl Viro return error; 3313625b6d10SAl Viro error = mnt_want_write(path.mnt); 331460545d0dSAl Viro if (unlikely(error)) 331560545d0dSAl Viro goto out; 3316af7bd4dcSAmir Goldstein child = vfs_tmpfile(path.dentry, op->mode, op->open_flag); 3317af7bd4dcSAmir Goldstein error = PTR_ERR(child); 3318684e73beSHirofumi Nakagawa if (IS_ERR(child)) 331960545d0dSAl Viro goto out2; 3320625b6d10SAl Viro dput(path.dentry); 3321625b6d10SAl Viro path.dentry = child; 3322c8a53ee5SAl Viro audit_inode(nd->name, child, 0); 332369a91c23SEric Rannaud /* Don't check for other permissions, the inode was just created */ 332462fb4a15SAl Viro error = may_open(&path, 0, op->open_flag); 332560545d0dSAl Viro if (error) 332660545d0dSAl Viro goto out2; 3327625b6d10SAl Viro file->f_path.mnt = path.mnt; 3328be12af3eSAl Viro error = finish_open(file, child, NULL); 332960545d0dSAl Viro out2: 3330625b6d10SAl Viro mnt_drop_write(path.mnt); 333160545d0dSAl Viro out: 3332625b6d10SAl Viro path_put(&path); 333360545d0dSAl Viro return error; 333460545d0dSAl Viro } 333560545d0dSAl Viro 33366ac08709SAl Viro static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file) 33376ac08709SAl Viro { 33386ac08709SAl Viro struct path path; 33396ac08709SAl Viro int error = path_lookupat(nd, flags, &path); 33406ac08709SAl Viro if (!error) { 33416ac08709SAl Viro audit_inode(nd->name, path.dentry, 0); 3342ae2bb293SAl Viro error = vfs_open(&path, file); 33436ac08709SAl Viro path_put(&path); 33446ac08709SAl Viro } 33456ac08709SAl Viro return error; 33466ac08709SAl Viro } 33476ac08709SAl Viro 3348c8a53ee5SAl Viro static struct file *path_openat(struct nameidata *nd, 3349c8a53ee5SAl Viro const struct open_flags *op, unsigned flags) 33501da177e4SLinus Torvalds { 335130d90494SAl Viro struct file *file; 335213aab428SAl Viro int error; 335331e6b01fSNick Piggin 3354ea73ea72SAl Viro file = alloc_empty_file(op->open_flag, current_cred()); 33551afc99beSAl Viro if (IS_ERR(file)) 33561afc99beSAl Viro return file; 335731e6b01fSNick Piggin 3358bb458c64SAl Viro if (unlikely(file->f_flags & __O_TMPFILE)) { 33593ec2eef1SAl Viro error = do_tmpfile(nd, flags, op, file); 33605f336e72SAl Viro } else if (unlikely(file->f_flags & O_PATH)) { 33616ac08709SAl Viro error = do_o_path(nd, flags, file); 33625f336e72SAl Viro } else { 33635f336e72SAl Viro const char *s = path_init(nd, flags); 33643bdba28bSAl Viro while (!(error = link_path_walk(s, nd)) && 3365c5971b8cSAl Viro (s = open_last_lookups(nd, file, op)) != NULL) 33661ccac622SAl Viro ; 3367c5971b8cSAl Viro if (!error) 3368c5971b8cSAl Viro error = do_open(nd, file, op); 3369deb106c6SAl Viro terminate_walk(nd); 33705f336e72SAl Viro } 33717c1c01ecSAl Viro if (likely(!error)) { 3372aad888f8SAl Viro if (likely(file->f_mode & FMODE_OPENED)) 33737c1c01ecSAl Viro return file; 33747c1c01ecSAl Viro WARN_ON(1); 33757c1c01ecSAl Viro error = -EINVAL; 3376015c3bbcSMiklos Szeredi } 33777c1c01ecSAl Viro fput(file); 33782675a4ebSAl Viro if (error == -EOPENSTALE) { 33792675a4ebSAl Viro if (flags & LOOKUP_RCU) 33802675a4ebSAl Viro error = -ECHILD; 33812675a4ebSAl Viro else 33822675a4ebSAl Viro error = -ESTALE; 33832675a4ebSAl Viro } 33847c1c01ecSAl Viro return ERR_PTR(error); 3385de459215SKirill Korotaev } 33861da177e4SLinus Torvalds 3387669abf4eSJeff Layton struct file *do_filp_open(int dfd, struct filename *pathname, 3388f9652e10SAl Viro const struct open_flags *op) 338913aab428SAl Viro { 33909883d185SAl Viro struct nameidata nd; 3391f9652e10SAl Viro int flags = op->lookup_flags; 339213aab428SAl Viro struct file *filp; 339313aab428SAl Viro 33949883d185SAl Viro set_nameidata(&nd, dfd, pathname); 3395c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_RCU); 339613aab428SAl Viro if (unlikely(filp == ERR_PTR(-ECHILD))) 3397c8a53ee5SAl Viro filp = path_openat(&nd, op, flags); 339813aab428SAl Viro if (unlikely(filp == ERR_PTR(-ESTALE))) 3399c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_REVAL); 34009883d185SAl Viro restore_nameidata(); 340113aab428SAl Viro return filp; 340213aab428SAl Viro } 340313aab428SAl Viro 340473d049a4SAl Viro struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, 3405f9652e10SAl Viro const char *name, const struct open_flags *op) 340673d049a4SAl Viro { 34079883d185SAl Viro struct nameidata nd; 340873d049a4SAl Viro struct file *file; 340951689104SPaul Moore struct filename *filename; 3410f9652e10SAl Viro int flags = op->lookup_flags | LOOKUP_ROOT; 341173d049a4SAl Viro 341273d049a4SAl Viro nd.root.mnt = mnt; 341373d049a4SAl Viro nd.root.dentry = dentry; 341473d049a4SAl Viro 3415b18825a7SDavid Howells if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN) 341673d049a4SAl Viro return ERR_PTR(-ELOOP); 341773d049a4SAl Viro 341851689104SPaul Moore filename = getname_kernel(name); 3419a1c83681SViresh Kumar if (IS_ERR(filename)) 342051689104SPaul Moore return ERR_CAST(filename); 342151689104SPaul Moore 34229883d185SAl Viro set_nameidata(&nd, -1, filename); 3423c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_RCU); 342473d049a4SAl Viro if (unlikely(file == ERR_PTR(-ECHILD))) 3425c8a53ee5SAl Viro file = path_openat(&nd, op, flags); 342673d049a4SAl Viro if (unlikely(file == ERR_PTR(-ESTALE))) 3427c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_REVAL); 34289883d185SAl Viro restore_nameidata(); 342951689104SPaul Moore putname(filename); 343073d049a4SAl Viro return file; 343173d049a4SAl Viro } 343273d049a4SAl Viro 3433fa14a0b8SAl Viro static struct dentry *filename_create(int dfd, struct filename *name, 34341ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 34351da177e4SLinus Torvalds { 3436c663e5d8SChristoph Hellwig struct dentry *dentry = ERR_PTR(-EEXIST); 3437391172c4SAl Viro struct qstr last; 3438391172c4SAl Viro int type; 3439c30dabfeSJan Kara int err2; 34401ac12b4bSJeff Layton int error; 34411ac12b4bSJeff Layton bool is_dir = (lookup_flags & LOOKUP_DIRECTORY); 34421ac12b4bSJeff Layton 34431ac12b4bSJeff Layton /* 34441ac12b4bSJeff Layton * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any 34451ac12b4bSJeff Layton * other flags passed in are ignored! 34461ac12b4bSJeff Layton */ 34471ac12b4bSJeff Layton lookup_flags &= LOOKUP_REVAL; 34481ac12b4bSJeff Layton 34495c31b6ceSAl Viro name = filename_parentat(dfd, name, lookup_flags, path, &last, &type); 34505c31b6ceSAl Viro if (IS_ERR(name)) 34515c31b6ceSAl Viro return ERR_CAST(name); 34521da177e4SLinus Torvalds 3453c663e5d8SChristoph Hellwig /* 3454c663e5d8SChristoph Hellwig * Yucky last component or no last component at all? 3455c663e5d8SChristoph Hellwig * (foo/., foo/.., /////) 3456c663e5d8SChristoph Hellwig */ 34575c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) 3458ed75e95dSAl Viro goto out; 3459c663e5d8SChristoph Hellwig 3460c30dabfeSJan Kara /* don't fail immediately if it's r/o, at least try to report other errors */ 3461391172c4SAl Viro err2 = mnt_want_write(path->mnt); 3462c663e5d8SChristoph Hellwig /* 3463c663e5d8SChristoph Hellwig * Do the final lookup. 3464c663e5d8SChristoph Hellwig */ 3465391172c4SAl Viro lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL; 34665955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 3467391172c4SAl Viro dentry = __lookup_hash(&last, path->dentry, lookup_flags); 34681da177e4SLinus Torvalds if (IS_ERR(dentry)) 3469a8104a9fSAl Viro goto unlock; 3470c663e5d8SChristoph Hellwig 3471a8104a9fSAl Viro error = -EEXIST; 3472b18825a7SDavid Howells if (d_is_positive(dentry)) 3473a8104a9fSAl Viro goto fail; 3474b18825a7SDavid Howells 3475c663e5d8SChristoph Hellwig /* 3476c663e5d8SChristoph Hellwig * Special case - lookup gave negative, but... we had foo/bar/ 3477c663e5d8SChristoph Hellwig * From the vfs_mknod() POV we just have a negative dentry - 3478c663e5d8SChristoph Hellwig * all is fine. Let's be bastards - you had / on the end, you've 3479c663e5d8SChristoph Hellwig * been asking for (non-existent) directory. -ENOENT for you. 3480c663e5d8SChristoph Hellwig */ 3481391172c4SAl Viro if (unlikely(!is_dir && last.name[last.len])) { 3482a8104a9fSAl Viro error = -ENOENT; 3483ed75e95dSAl Viro goto fail; 3484e9baf6e5SAl Viro } 3485c30dabfeSJan Kara if (unlikely(err2)) { 3486c30dabfeSJan Kara error = err2; 3487a8104a9fSAl Viro goto fail; 3488c30dabfeSJan Kara } 3489181c37b6SAl Viro putname(name); 3490e9baf6e5SAl Viro return dentry; 34911da177e4SLinus Torvalds fail: 3492a8104a9fSAl Viro dput(dentry); 3493a8104a9fSAl Viro dentry = ERR_PTR(error); 3494a8104a9fSAl Viro unlock: 34955955102cSAl Viro inode_unlock(path->dentry->d_inode); 3496c30dabfeSJan Kara if (!err2) 3497391172c4SAl Viro mnt_drop_write(path->mnt); 3498ed75e95dSAl Viro out: 3499391172c4SAl Viro path_put(path); 3500181c37b6SAl Viro putname(name); 3501ed75e95dSAl Viro return dentry; 3502dae6ad8fSAl Viro } 3503fa14a0b8SAl Viro 3504fa14a0b8SAl Viro struct dentry *kern_path_create(int dfd, const char *pathname, 3505fa14a0b8SAl Viro struct path *path, unsigned int lookup_flags) 3506fa14a0b8SAl Viro { 3507181c37b6SAl Viro return filename_create(dfd, getname_kernel(pathname), 3508181c37b6SAl Viro path, lookup_flags); 3509fa14a0b8SAl Viro } 3510dae6ad8fSAl Viro EXPORT_SYMBOL(kern_path_create); 3511dae6ad8fSAl Viro 3512921a1650SAl Viro void done_path_create(struct path *path, struct dentry *dentry) 3513921a1650SAl Viro { 3514921a1650SAl Viro dput(dentry); 35155955102cSAl Viro inode_unlock(path->dentry->d_inode); 3516a8104a9fSAl Viro mnt_drop_write(path->mnt); 3517921a1650SAl Viro path_put(path); 3518921a1650SAl Viro } 3519921a1650SAl Viro EXPORT_SYMBOL(done_path_create); 3520921a1650SAl Viro 3521520ae687SAl Viro inline struct dentry *user_path_create(int dfd, const char __user *pathname, 35221ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 3523dae6ad8fSAl Viro { 3524181c37b6SAl Viro return filename_create(dfd, getname(pathname), path, lookup_flags); 3525dae6ad8fSAl Viro } 3526dae6ad8fSAl Viro EXPORT_SYMBOL(user_path_create); 3527dae6ad8fSAl Viro 35281a67aafbSAl Viro int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) 35291da177e4SLinus Torvalds { 3530a3c751a5SMiklos Szeredi bool is_whiteout = S_ISCHR(mode) && dev == WHITEOUT_DEV; 3531a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 35321da177e4SLinus Torvalds 35331da177e4SLinus Torvalds if (error) 35341da177e4SLinus Torvalds return error; 35351da177e4SLinus Torvalds 3536a3c751a5SMiklos Szeredi if ((S_ISCHR(mode) || S_ISBLK(mode)) && !is_whiteout && 3537a3c751a5SMiklos Szeredi !capable(CAP_MKNOD)) 35381da177e4SLinus Torvalds return -EPERM; 35391da177e4SLinus Torvalds 3540acfa4380SAl Viro if (!dir->i_op->mknod) 35411da177e4SLinus Torvalds return -EPERM; 35421da177e4SLinus Torvalds 354308ce5f16SSerge E. Hallyn error = devcgroup_inode_mknod(mode, dev); 354408ce5f16SSerge E. Hallyn if (error) 354508ce5f16SSerge E. Hallyn return error; 354608ce5f16SSerge E. Hallyn 35471da177e4SLinus Torvalds error = security_inode_mknod(dir, dentry, mode, dev); 35481da177e4SLinus Torvalds if (error) 35491da177e4SLinus Torvalds return error; 35501da177e4SLinus Torvalds 35511da177e4SLinus Torvalds error = dir->i_op->mknod(dir, dentry, mode, dev); 3552a74574aaSStephen Smalley if (!error) 3553f38aa942SAmy Griffis fsnotify_create(dir, dentry); 35541da177e4SLinus Torvalds return error; 35551da177e4SLinus Torvalds } 35564d359507SAl Viro EXPORT_SYMBOL(vfs_mknod); 35571da177e4SLinus Torvalds 3558f69aac00SAl Viro static int may_mknod(umode_t mode) 3559463c3197SDave Hansen { 3560463c3197SDave Hansen switch (mode & S_IFMT) { 3561463c3197SDave Hansen case S_IFREG: 3562463c3197SDave Hansen case S_IFCHR: 3563463c3197SDave Hansen case S_IFBLK: 3564463c3197SDave Hansen case S_IFIFO: 3565463c3197SDave Hansen case S_IFSOCK: 3566463c3197SDave Hansen case 0: /* zero mode translates to S_IFREG */ 3567463c3197SDave Hansen return 0; 3568463c3197SDave Hansen case S_IFDIR: 3569463c3197SDave Hansen return -EPERM; 3570463c3197SDave Hansen default: 3571463c3197SDave Hansen return -EINVAL; 3572463c3197SDave Hansen } 3573463c3197SDave Hansen } 3574463c3197SDave Hansen 35755fee64fcSChristoph Hellwig static long do_mknodat(int dfd, const char __user *filename, umode_t mode, 357687c4e192SDominik Brodowski unsigned int dev) 35771da177e4SLinus Torvalds { 35781da177e4SLinus Torvalds struct dentry *dentry; 3579dae6ad8fSAl Viro struct path path; 3580dae6ad8fSAl Viro int error; 3581972567f1SJeff Layton unsigned int lookup_flags = 0; 35821da177e4SLinus Torvalds 35838e4bfca1SAl Viro error = may_mknod(mode); 35848e4bfca1SAl Viro if (error) 35858e4bfca1SAl Viro return error; 3586972567f1SJeff Layton retry: 3587972567f1SJeff Layton dentry = user_path_create(dfd, filename, &path, lookup_flags); 3588dae6ad8fSAl Viro if (IS_ERR(dentry)) 3589dae6ad8fSAl Viro return PTR_ERR(dentry); 35902ad94ae6SAl Viro 3591dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3592ce3b0f8dSAl Viro mode &= ~current_umask(); 3593dae6ad8fSAl Viro error = security_path_mknod(&path, dentry, mode, dev); 3594be6d3e56SKentaro Takeda if (error) 3595a8104a9fSAl Viro goto out; 35961da177e4SLinus Torvalds switch (mode & S_IFMT) { 35971da177e4SLinus Torvalds case 0: case S_IFREG: 3598312b63fbSAl Viro error = vfs_create(path.dentry->d_inode,dentry,mode,true); 359905d1a717SMimi Zohar if (!error) 360005d1a717SMimi Zohar ima_post_path_mknod(dentry); 36011da177e4SLinus Torvalds break; 36021da177e4SLinus Torvalds case S_IFCHR: case S_IFBLK: 3603dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode, 36041da177e4SLinus Torvalds new_decode_dev(dev)); 36051da177e4SLinus Torvalds break; 36061da177e4SLinus Torvalds case S_IFIFO: case S_IFSOCK: 3607dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); 36081da177e4SLinus Torvalds break; 36091da177e4SLinus Torvalds } 3610a8104a9fSAl Viro out: 3611921a1650SAl Viro done_path_create(&path, dentry); 3612972567f1SJeff Layton if (retry_estale(error, lookup_flags)) { 3613972567f1SJeff Layton lookup_flags |= LOOKUP_REVAL; 3614972567f1SJeff Layton goto retry; 3615972567f1SJeff Layton } 36161da177e4SLinus Torvalds return error; 36171da177e4SLinus Torvalds } 36181da177e4SLinus Torvalds 361987c4e192SDominik Brodowski SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, 362087c4e192SDominik Brodowski unsigned int, dev) 362187c4e192SDominik Brodowski { 362287c4e192SDominik Brodowski return do_mknodat(dfd, filename, mode, dev); 362387c4e192SDominik Brodowski } 362487c4e192SDominik Brodowski 36258208a22bSAl Viro SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) 36265590ff0dSUlrich Drepper { 362787c4e192SDominik Brodowski return do_mknodat(AT_FDCWD, filename, mode, dev); 36285590ff0dSUlrich Drepper } 36295590ff0dSUlrich Drepper 363018bb1db3SAl Viro int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 36311da177e4SLinus Torvalds { 3632a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 36338de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 36341da177e4SLinus Torvalds 36351da177e4SLinus Torvalds if (error) 36361da177e4SLinus Torvalds return error; 36371da177e4SLinus Torvalds 3638acfa4380SAl Viro if (!dir->i_op->mkdir) 36391da177e4SLinus Torvalds return -EPERM; 36401da177e4SLinus Torvalds 36411da177e4SLinus Torvalds mode &= (S_IRWXUGO|S_ISVTX); 36421da177e4SLinus Torvalds error = security_inode_mkdir(dir, dentry, mode); 36431da177e4SLinus Torvalds if (error) 36441da177e4SLinus Torvalds return error; 36451da177e4SLinus Torvalds 36468de52778SAl Viro if (max_links && dir->i_nlink >= max_links) 36478de52778SAl Viro return -EMLINK; 36488de52778SAl Viro 36491da177e4SLinus Torvalds error = dir->i_op->mkdir(dir, dentry, mode); 3650a74574aaSStephen Smalley if (!error) 3651f38aa942SAmy Griffis fsnotify_mkdir(dir, dentry); 36521da177e4SLinus Torvalds return error; 36531da177e4SLinus Torvalds } 36544d359507SAl Viro EXPORT_SYMBOL(vfs_mkdir); 36551da177e4SLinus Torvalds 365683ff98c3SChristoph Hellwig static long do_mkdirat(int dfd, const char __user *pathname, umode_t mode) 36571da177e4SLinus Torvalds { 36586902d925SDave Hansen struct dentry *dentry; 3659dae6ad8fSAl Viro struct path path; 3660dae6ad8fSAl Viro int error; 3661b76d8b82SJeff Layton unsigned int lookup_flags = LOOKUP_DIRECTORY; 36621da177e4SLinus Torvalds 3663b76d8b82SJeff Layton retry: 3664b76d8b82SJeff Layton dentry = user_path_create(dfd, pathname, &path, lookup_flags); 36656902d925SDave Hansen if (IS_ERR(dentry)) 3666dae6ad8fSAl Viro return PTR_ERR(dentry); 36676902d925SDave Hansen 3668dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3669ce3b0f8dSAl Viro mode &= ~current_umask(); 3670dae6ad8fSAl Viro error = security_path_mkdir(&path, dentry, mode); 3671a8104a9fSAl Viro if (!error) 3672dae6ad8fSAl Viro error = vfs_mkdir(path.dentry->d_inode, dentry, mode); 3673921a1650SAl Viro done_path_create(&path, dentry); 3674b76d8b82SJeff Layton if (retry_estale(error, lookup_flags)) { 3675b76d8b82SJeff Layton lookup_flags |= LOOKUP_REVAL; 3676b76d8b82SJeff Layton goto retry; 3677b76d8b82SJeff Layton } 36781da177e4SLinus Torvalds return error; 36791da177e4SLinus Torvalds } 36801da177e4SLinus Torvalds 36810101db7aSDominik Brodowski SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) 36820101db7aSDominik Brodowski { 36830101db7aSDominik Brodowski return do_mkdirat(dfd, pathname, mode); 36840101db7aSDominik Brodowski } 36850101db7aSDominik Brodowski 3686a218d0fdSAl Viro SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) 36875590ff0dSUlrich Drepper { 36880101db7aSDominik Brodowski return do_mkdirat(AT_FDCWD, pathname, mode); 36895590ff0dSUlrich Drepper } 36905590ff0dSUlrich Drepper 36911da177e4SLinus Torvalds int vfs_rmdir(struct inode *dir, struct dentry *dentry) 36921da177e4SLinus Torvalds { 36931da177e4SLinus Torvalds int error = may_delete(dir, dentry, 1); 36941da177e4SLinus Torvalds 36951da177e4SLinus Torvalds if (error) 36961da177e4SLinus Torvalds return error; 36971da177e4SLinus Torvalds 3698acfa4380SAl Viro if (!dir->i_op->rmdir) 36991da177e4SLinus Torvalds return -EPERM; 37001da177e4SLinus Torvalds 37011d2ef590SAl Viro dget(dentry); 37025955102cSAl Viro inode_lock(dentry->d_inode); 3703912dbc15SSage Weil 37041da177e4SLinus Torvalds error = -EBUSY; 37057af1364fSEric W. Biederman if (is_local_mountpoint(dentry)) 3706912dbc15SSage Weil goto out; 3707912dbc15SSage Weil 37081da177e4SLinus Torvalds error = security_inode_rmdir(dir, dentry); 3709912dbc15SSage Weil if (error) 3710912dbc15SSage Weil goto out; 3711912dbc15SSage Weil 37121da177e4SLinus Torvalds error = dir->i_op->rmdir(dir, dentry); 3713912dbc15SSage Weil if (error) 3714912dbc15SSage Weil goto out; 3715912dbc15SSage Weil 37168767712fSAl Viro shrink_dcache_parent(dentry); 37171da177e4SLinus Torvalds dentry->d_inode->i_flags |= S_DEAD; 3718d83c49f3SAl Viro dont_mount(dentry); 37198ed936b5SEric W. Biederman detach_mounts(dentry); 3720116b9731SAmir Goldstein fsnotify_rmdir(dir, dentry); 37211da177e4SLinus Torvalds 3722912dbc15SSage Weil out: 37235955102cSAl Viro inode_unlock(dentry->d_inode); 37241d2ef590SAl Viro dput(dentry); 3725912dbc15SSage Weil if (!error) 3726912dbc15SSage Weil d_delete(dentry); 37271da177e4SLinus Torvalds return error; 37281da177e4SLinus Torvalds } 37294d359507SAl Viro EXPORT_SYMBOL(vfs_rmdir); 37301da177e4SLinus Torvalds 3731e24ab0efSChristoph Hellwig long do_rmdir(int dfd, struct filename *name) 37321da177e4SLinus Torvalds { 37331da177e4SLinus Torvalds int error = 0; 37341da177e4SLinus Torvalds struct dentry *dentry; 3735f5beed75SAl Viro struct path path; 3736f5beed75SAl Viro struct qstr last; 3737f5beed75SAl Viro int type; 3738c6ee9206SJeff Layton unsigned int lookup_flags = 0; 3739c6ee9206SJeff Layton retry: 3740e24ab0efSChristoph Hellwig name = filename_parentat(dfd, name, lookup_flags, 3741c1d4dd27SAl Viro &path, &last, &type); 374291a27b2aSJeff Layton if (IS_ERR(name)) 374391a27b2aSJeff Layton return PTR_ERR(name); 37441da177e4SLinus Torvalds 3745f5beed75SAl Viro switch (type) { 37461da177e4SLinus Torvalds case LAST_DOTDOT: 37471da177e4SLinus Torvalds error = -ENOTEMPTY; 37481da177e4SLinus Torvalds goto exit1; 37491da177e4SLinus Torvalds case LAST_DOT: 37501da177e4SLinus Torvalds error = -EINVAL; 37511da177e4SLinus Torvalds goto exit1; 37521da177e4SLinus Torvalds case LAST_ROOT: 37531da177e4SLinus Torvalds error = -EBUSY; 37541da177e4SLinus Torvalds goto exit1; 37551da177e4SLinus Torvalds } 37560612d9fbSOGAWA Hirofumi 3757f5beed75SAl Viro error = mnt_want_write(path.mnt); 3758c30dabfeSJan Kara if (error) 3759c30dabfeSJan Kara goto exit1; 37600612d9fbSOGAWA Hirofumi 37615955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 3762f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 37631da177e4SLinus Torvalds error = PTR_ERR(dentry); 37646902d925SDave Hansen if (IS_ERR(dentry)) 37656902d925SDave Hansen goto exit2; 3766e6bc45d6STheodore Ts'o if (!dentry->d_inode) { 3767e6bc45d6STheodore Ts'o error = -ENOENT; 3768e6bc45d6STheodore Ts'o goto exit3; 3769e6bc45d6STheodore Ts'o } 3770f5beed75SAl Viro error = security_path_rmdir(&path, dentry); 3771be6d3e56SKentaro Takeda if (error) 3772c30dabfeSJan Kara goto exit3; 3773f5beed75SAl Viro error = vfs_rmdir(path.dentry->d_inode, dentry); 37740622753bSDave Hansen exit3: 37751da177e4SLinus Torvalds dput(dentry); 37766902d925SDave Hansen exit2: 37775955102cSAl Viro inode_unlock(path.dentry->d_inode); 3778f5beed75SAl Viro mnt_drop_write(path.mnt); 37791da177e4SLinus Torvalds exit1: 3780f5beed75SAl Viro path_put(&path); 3781c6ee9206SJeff Layton if (retry_estale(error, lookup_flags)) { 3782c6ee9206SJeff Layton lookup_flags |= LOOKUP_REVAL; 3783c6ee9206SJeff Layton goto retry; 3784c6ee9206SJeff Layton } 378524fb33d4SAl Viro putname(name); 37861da177e4SLinus Torvalds return error; 37871da177e4SLinus Torvalds } 37881da177e4SLinus Torvalds 37893cdad428SHeiko Carstens SYSCALL_DEFINE1(rmdir, const char __user *, pathname) 37905590ff0dSUlrich Drepper { 3791e24ab0efSChristoph Hellwig return do_rmdir(AT_FDCWD, getname(pathname)); 37925590ff0dSUlrich Drepper } 37935590ff0dSUlrich Drepper 3794b21996e3SJ. Bruce Fields /** 3795b21996e3SJ. Bruce Fields * vfs_unlink - unlink a filesystem object 3796b21996e3SJ. Bruce Fields * @dir: parent directory 3797b21996e3SJ. Bruce Fields * @dentry: victim 3798b21996e3SJ. Bruce Fields * @delegated_inode: returns victim inode, if the inode is delegated. 3799b21996e3SJ. Bruce Fields * 3800b21996e3SJ. Bruce Fields * The caller must hold dir->i_mutex. 3801b21996e3SJ. Bruce Fields * 3802b21996e3SJ. Bruce Fields * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and 3803b21996e3SJ. Bruce Fields * return a reference to the inode in delegated_inode. The caller 3804b21996e3SJ. Bruce Fields * should then break the delegation on that inode and retry. Because 3805b21996e3SJ. Bruce Fields * breaking a delegation may take a long time, the caller should drop 3806b21996e3SJ. Bruce Fields * dir->i_mutex before doing so. 3807b21996e3SJ. Bruce Fields * 3808b21996e3SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 3809b21996e3SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 3810b21996e3SJ. Bruce Fields * to be NFS exported. 3811b21996e3SJ. Bruce Fields */ 3812b21996e3SJ. Bruce Fields int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode) 38131da177e4SLinus Torvalds { 38149accbb97SJ. Bruce Fields struct inode *target = dentry->d_inode; 38151da177e4SLinus Torvalds int error = may_delete(dir, dentry, 0); 38161da177e4SLinus Torvalds 38171da177e4SLinus Torvalds if (error) 38181da177e4SLinus Torvalds return error; 38191da177e4SLinus Torvalds 3820acfa4380SAl Viro if (!dir->i_op->unlink) 38211da177e4SLinus Torvalds return -EPERM; 38221da177e4SLinus Torvalds 38235955102cSAl Viro inode_lock(target); 38248ed936b5SEric W. Biederman if (is_local_mountpoint(dentry)) 38251da177e4SLinus Torvalds error = -EBUSY; 38261da177e4SLinus Torvalds else { 38271da177e4SLinus Torvalds error = security_inode_unlink(dir, dentry); 3828bec1052eSAl Viro if (!error) { 38295a14696cSJ. Bruce Fields error = try_break_deleg(target, delegated_inode); 38305a14696cSJ. Bruce Fields if (error) 3831b21996e3SJ. Bruce Fields goto out; 38321da177e4SLinus Torvalds error = dir->i_op->unlink(dir, dentry); 38338ed936b5SEric W. Biederman if (!error) { 3834d83c49f3SAl Viro dont_mount(dentry); 38358ed936b5SEric W. Biederman detach_mounts(dentry); 3836116b9731SAmir Goldstein fsnotify_unlink(dir, dentry); 38378ed936b5SEric W. Biederman } 3838bec1052eSAl Viro } 38391da177e4SLinus Torvalds } 3840b21996e3SJ. Bruce Fields out: 38415955102cSAl Viro inode_unlock(target); 38421da177e4SLinus Torvalds 38431da177e4SLinus Torvalds /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 38441da177e4SLinus Torvalds if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { 38459accbb97SJ. Bruce Fields fsnotify_link_count(target); 38461da177e4SLinus Torvalds d_delete(dentry); 38471da177e4SLinus Torvalds } 38480eeca283SRobert Love 38491da177e4SLinus Torvalds return error; 38501da177e4SLinus Torvalds } 38514d359507SAl Viro EXPORT_SYMBOL(vfs_unlink); 38521da177e4SLinus Torvalds 38531da177e4SLinus Torvalds /* 38541da177e4SLinus Torvalds * Make sure that the actual truncation of the file will occur outside its 38551b1dcc1bSJes Sorensen * directory's i_mutex. Truncate can take a long time if there is a lot of 38561da177e4SLinus Torvalds * writeout happening, and we don't want to prevent access to the directory 38571da177e4SLinus Torvalds * while waiting on the I/O. 38581da177e4SLinus Torvalds */ 3859da2f1362SChristoph Hellwig long do_unlinkat(int dfd, struct filename *name) 38601da177e4SLinus Torvalds { 38612ad94ae6SAl Viro int error; 38621da177e4SLinus Torvalds struct dentry *dentry; 3863f5beed75SAl Viro struct path path; 3864f5beed75SAl Viro struct qstr last; 3865f5beed75SAl Viro int type; 38661da177e4SLinus Torvalds struct inode *inode = NULL; 3867b21996e3SJ. Bruce Fields struct inode *delegated_inode = NULL; 38685d18f813SJeff Layton unsigned int lookup_flags = 0; 38695d18f813SJeff Layton retry: 3870da2f1362SChristoph Hellwig name = filename_parentat(dfd, name, lookup_flags, &path, &last, &type); 387191a27b2aSJeff Layton if (IS_ERR(name)) 387291a27b2aSJeff Layton return PTR_ERR(name); 38732ad94ae6SAl Viro 38741da177e4SLinus Torvalds error = -EISDIR; 3875f5beed75SAl Viro if (type != LAST_NORM) 38761da177e4SLinus Torvalds goto exit1; 38770612d9fbSOGAWA Hirofumi 3878f5beed75SAl Viro error = mnt_want_write(path.mnt); 3879c30dabfeSJan Kara if (error) 3880c30dabfeSJan Kara goto exit1; 3881b21996e3SJ. Bruce Fields retry_deleg: 38825955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 3883f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 38841da177e4SLinus Torvalds error = PTR_ERR(dentry); 38851da177e4SLinus Torvalds if (!IS_ERR(dentry)) { 38861da177e4SLinus Torvalds /* Why not before? Because we want correct error value */ 3887f5beed75SAl Viro if (last.name[last.len]) 388850338b88STörök Edwin goto slashes; 38891da177e4SLinus Torvalds inode = dentry->d_inode; 3890b18825a7SDavid Howells if (d_is_negative(dentry)) 3891e6bc45d6STheodore Ts'o goto slashes; 38927de9c6eeSAl Viro ihold(inode); 3893f5beed75SAl Viro error = security_path_unlink(&path, dentry); 3894be6d3e56SKentaro Takeda if (error) 3895c30dabfeSJan Kara goto exit2; 3896f5beed75SAl Viro error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode); 38971da177e4SLinus Torvalds exit2: 38981da177e4SLinus Torvalds dput(dentry); 38991da177e4SLinus Torvalds } 39005955102cSAl Viro inode_unlock(path.dentry->d_inode); 39011da177e4SLinus Torvalds if (inode) 39021da177e4SLinus Torvalds iput(inode); /* truncate the inode here */ 3903b21996e3SJ. Bruce Fields inode = NULL; 3904b21996e3SJ. Bruce Fields if (delegated_inode) { 39055a14696cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 3906b21996e3SJ. Bruce Fields if (!error) 3907b21996e3SJ. Bruce Fields goto retry_deleg; 3908b21996e3SJ. Bruce Fields } 3909f5beed75SAl Viro mnt_drop_write(path.mnt); 39101da177e4SLinus Torvalds exit1: 3911f5beed75SAl Viro path_put(&path); 39125d18f813SJeff Layton if (retry_estale(error, lookup_flags)) { 39135d18f813SJeff Layton lookup_flags |= LOOKUP_REVAL; 39145d18f813SJeff Layton inode = NULL; 39155d18f813SJeff Layton goto retry; 39165d18f813SJeff Layton } 3917da2f1362SChristoph Hellwig putname(name); 39181da177e4SLinus Torvalds return error; 39191da177e4SLinus Torvalds 39201da177e4SLinus Torvalds slashes: 3921b18825a7SDavid Howells if (d_is_negative(dentry)) 3922b18825a7SDavid Howells error = -ENOENT; 392344b1d530SMiklos Szeredi else if (d_is_dir(dentry)) 3924b18825a7SDavid Howells error = -EISDIR; 3925b18825a7SDavid Howells else 3926b18825a7SDavid Howells error = -ENOTDIR; 39271da177e4SLinus Torvalds goto exit2; 39281da177e4SLinus Torvalds } 39291da177e4SLinus Torvalds 39302e4d0924SHeiko Carstens SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) 39315590ff0dSUlrich Drepper { 39325590ff0dSUlrich Drepper if ((flag & ~AT_REMOVEDIR) != 0) 39335590ff0dSUlrich Drepper return -EINVAL; 39345590ff0dSUlrich Drepper 39355590ff0dSUlrich Drepper if (flag & AT_REMOVEDIR) 3936e24ab0efSChristoph Hellwig return do_rmdir(dfd, getname(pathname)); 3937da2f1362SChristoph Hellwig return do_unlinkat(dfd, getname(pathname)); 39385590ff0dSUlrich Drepper } 39395590ff0dSUlrich Drepper 39403480b257SHeiko Carstens SYSCALL_DEFINE1(unlink, const char __user *, pathname) 39415590ff0dSUlrich Drepper { 3942da2f1362SChristoph Hellwig return do_unlinkat(AT_FDCWD, getname(pathname)); 39435590ff0dSUlrich Drepper } 39445590ff0dSUlrich Drepper 3945db2e747bSMiklos Szeredi int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) 39461da177e4SLinus Torvalds { 3947a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 39481da177e4SLinus Torvalds 39491da177e4SLinus Torvalds if (error) 39501da177e4SLinus Torvalds return error; 39511da177e4SLinus Torvalds 3952acfa4380SAl Viro if (!dir->i_op->symlink) 39531da177e4SLinus Torvalds return -EPERM; 39541da177e4SLinus Torvalds 39551da177e4SLinus Torvalds error = security_inode_symlink(dir, dentry, oldname); 39561da177e4SLinus Torvalds if (error) 39571da177e4SLinus Torvalds return error; 39581da177e4SLinus Torvalds 39591da177e4SLinus Torvalds error = dir->i_op->symlink(dir, dentry, oldname); 3960a74574aaSStephen Smalley if (!error) 3961f38aa942SAmy Griffis fsnotify_create(dir, dentry); 39621da177e4SLinus Torvalds return error; 39631da177e4SLinus Torvalds } 39644d359507SAl Viro EXPORT_SYMBOL(vfs_symlink); 39651da177e4SLinus Torvalds 3966cd3acb6aSChristoph Hellwig static long do_symlinkat(const char __user *oldname, int newdfd, 3967b724e846SDominik Brodowski const char __user *newname) 39681da177e4SLinus Torvalds { 39692ad94ae6SAl Viro int error; 397091a27b2aSJeff Layton struct filename *from; 39716902d925SDave Hansen struct dentry *dentry; 3972dae6ad8fSAl Viro struct path path; 3973f46d3567SJeff Layton unsigned int lookup_flags = 0; 39741da177e4SLinus Torvalds 39751da177e4SLinus Torvalds from = getname(oldname); 39761da177e4SLinus Torvalds if (IS_ERR(from)) 39771da177e4SLinus Torvalds return PTR_ERR(from); 3978f46d3567SJeff Layton retry: 3979f46d3567SJeff Layton dentry = user_path_create(newdfd, newname, &path, lookup_flags); 39801da177e4SLinus Torvalds error = PTR_ERR(dentry); 39816902d925SDave Hansen if (IS_ERR(dentry)) 3982dae6ad8fSAl Viro goto out_putname; 39836902d925SDave Hansen 398491a27b2aSJeff Layton error = security_path_symlink(&path, dentry, from->name); 3985a8104a9fSAl Viro if (!error) 398691a27b2aSJeff Layton error = vfs_symlink(path.dentry->d_inode, dentry, from->name); 3987921a1650SAl Viro done_path_create(&path, dentry); 3988f46d3567SJeff Layton if (retry_estale(error, lookup_flags)) { 3989f46d3567SJeff Layton lookup_flags |= LOOKUP_REVAL; 3990f46d3567SJeff Layton goto retry; 3991f46d3567SJeff Layton } 39926902d925SDave Hansen out_putname: 39931da177e4SLinus Torvalds putname(from); 39941da177e4SLinus Torvalds return error; 39951da177e4SLinus Torvalds } 39961da177e4SLinus Torvalds 3997b724e846SDominik Brodowski SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, 3998b724e846SDominik Brodowski int, newdfd, const char __user *, newname) 3999b724e846SDominik Brodowski { 4000b724e846SDominik Brodowski return do_symlinkat(oldname, newdfd, newname); 4001b724e846SDominik Brodowski } 4002b724e846SDominik Brodowski 40033480b257SHeiko Carstens SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) 40045590ff0dSUlrich Drepper { 4005b724e846SDominik Brodowski return do_symlinkat(oldname, AT_FDCWD, newname); 40065590ff0dSUlrich Drepper } 40075590ff0dSUlrich Drepper 4008146a8595SJ. Bruce Fields /** 4009146a8595SJ. Bruce Fields * vfs_link - create a new link 4010146a8595SJ. Bruce Fields * @old_dentry: object to be linked 4011146a8595SJ. Bruce Fields * @dir: new parent 4012146a8595SJ. Bruce Fields * @new_dentry: where to create the new link 4013146a8595SJ. Bruce Fields * @delegated_inode: returns inode needing a delegation break 4014146a8595SJ. Bruce Fields * 4015146a8595SJ. Bruce Fields * The caller must hold dir->i_mutex 4016146a8595SJ. Bruce Fields * 4017146a8595SJ. Bruce Fields * If vfs_link discovers a delegation on the to-be-linked file in need 4018146a8595SJ. Bruce Fields * of breaking, it will return -EWOULDBLOCK and return a reference to the 4019146a8595SJ. Bruce Fields * inode in delegated_inode. The caller should then break the delegation 4020146a8595SJ. Bruce Fields * and retry. Because breaking a delegation may take a long time, the 4021146a8595SJ. Bruce Fields * caller should drop the i_mutex before doing so. 4022146a8595SJ. Bruce Fields * 4023146a8595SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 4024146a8595SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 4025146a8595SJ. Bruce Fields * to be NFS exported. 4026146a8595SJ. Bruce Fields */ 4027146a8595SJ. Bruce Fields int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode) 40281da177e4SLinus Torvalds { 40291da177e4SLinus Torvalds struct inode *inode = old_dentry->d_inode; 40308de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 40311da177e4SLinus Torvalds int error; 40321da177e4SLinus Torvalds 40331da177e4SLinus Torvalds if (!inode) 40341da177e4SLinus Torvalds return -ENOENT; 40351da177e4SLinus Torvalds 4036a95164d9SMiklos Szeredi error = may_create(dir, new_dentry); 40371da177e4SLinus Torvalds if (error) 40381da177e4SLinus Torvalds return error; 40391da177e4SLinus Torvalds 40401da177e4SLinus Torvalds if (dir->i_sb != inode->i_sb) 40411da177e4SLinus Torvalds return -EXDEV; 40421da177e4SLinus Torvalds 40431da177e4SLinus Torvalds /* 40441da177e4SLinus Torvalds * A link to an append-only or immutable file cannot be created. 40451da177e4SLinus Torvalds */ 40461da177e4SLinus Torvalds if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 40471da177e4SLinus Torvalds return -EPERM; 40480bd23d09SEric W. Biederman /* 40490bd23d09SEric W. Biederman * Updating the link count will likely cause i_uid and i_gid to 40500bd23d09SEric W. Biederman * be writen back improperly if their true value is unknown to 40510bd23d09SEric W. Biederman * the vfs. 40520bd23d09SEric W. Biederman */ 40530bd23d09SEric W. Biederman if (HAS_UNMAPPED_ID(inode)) 40540bd23d09SEric W. Biederman return -EPERM; 4055acfa4380SAl Viro if (!dir->i_op->link) 40561da177e4SLinus Torvalds return -EPERM; 40577e79eedbSTetsuo Handa if (S_ISDIR(inode->i_mode)) 40581da177e4SLinus Torvalds return -EPERM; 40591da177e4SLinus Torvalds 40601da177e4SLinus Torvalds error = security_inode_link(old_dentry, dir, new_dentry); 40611da177e4SLinus Torvalds if (error) 40621da177e4SLinus Torvalds return error; 40631da177e4SLinus Torvalds 40645955102cSAl Viro inode_lock(inode); 4065aae8a97dSAneesh Kumar K.V /* Make sure we don't allow creating hardlink to an unlinked file */ 4066f4e0c30cSAl Viro if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) 4067aae8a97dSAneesh Kumar K.V error = -ENOENT; 40688de52778SAl Viro else if (max_links && inode->i_nlink >= max_links) 40698de52778SAl Viro error = -EMLINK; 4070146a8595SJ. Bruce Fields else { 4071146a8595SJ. Bruce Fields error = try_break_deleg(inode, delegated_inode); 4072146a8595SJ. Bruce Fields if (!error) 40731da177e4SLinus Torvalds error = dir->i_op->link(old_dentry, dir, new_dentry); 4074146a8595SJ. Bruce Fields } 4075f4e0c30cSAl Viro 4076f4e0c30cSAl Viro if (!error && (inode->i_state & I_LINKABLE)) { 4077f4e0c30cSAl Viro spin_lock(&inode->i_lock); 4078f4e0c30cSAl Viro inode->i_state &= ~I_LINKABLE; 4079f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 4080f4e0c30cSAl Viro } 40815955102cSAl Viro inode_unlock(inode); 4082e31e14ecSStephen Smalley if (!error) 40837e79eedbSTetsuo Handa fsnotify_link(dir, inode, new_dentry); 40841da177e4SLinus Torvalds return error; 40851da177e4SLinus Torvalds } 40864d359507SAl Viro EXPORT_SYMBOL(vfs_link); 40871da177e4SLinus Torvalds 40881da177e4SLinus Torvalds /* 40891da177e4SLinus Torvalds * Hardlinks are often used in delicate situations. We avoid 40901da177e4SLinus Torvalds * security-related surprises by not following symlinks on the 40911da177e4SLinus Torvalds * newname. --KAB 40921da177e4SLinus Torvalds * 40931da177e4SLinus Torvalds * We don't follow them on the oldname either to be compatible 40941da177e4SLinus Torvalds * with linux 2.0, and to avoid hard-linking to directories 40951da177e4SLinus Torvalds * and other special files. --ADM 40961da177e4SLinus Torvalds */ 4097812931d6SChristoph Hellwig static int do_linkat(int olddfd, const char __user *oldname, int newdfd, 409846ea89ebSDominik Brodowski const char __user *newname, int flags) 40991da177e4SLinus Torvalds { 41001da177e4SLinus Torvalds struct dentry *new_dentry; 4101dae6ad8fSAl Viro struct path old_path, new_path; 4102146a8595SJ. Bruce Fields struct inode *delegated_inode = NULL; 410311a7b371SAneesh Kumar K.V int how = 0; 41041da177e4SLinus Torvalds int error; 41051da177e4SLinus Torvalds 410611a7b371SAneesh Kumar K.V if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) 4107c04030e1SUlrich Drepper return -EINVAL; 410811a7b371SAneesh Kumar K.V /* 4109f0cc6ffbSLinus Torvalds * To use null names we require CAP_DAC_READ_SEARCH 4110f0cc6ffbSLinus Torvalds * This ensures that not everyone will be able to create 4111f0cc6ffbSLinus Torvalds * handlink using the passed filedescriptor. 411211a7b371SAneesh Kumar K.V */ 4113f0cc6ffbSLinus Torvalds if (flags & AT_EMPTY_PATH) { 4114f0cc6ffbSLinus Torvalds if (!capable(CAP_DAC_READ_SEARCH)) 4115f0cc6ffbSLinus Torvalds return -ENOENT; 411611a7b371SAneesh Kumar K.V how = LOOKUP_EMPTY; 4117f0cc6ffbSLinus Torvalds } 4118c04030e1SUlrich Drepper 411911a7b371SAneesh Kumar K.V if (flags & AT_SYMLINK_FOLLOW) 412011a7b371SAneesh Kumar K.V how |= LOOKUP_FOLLOW; 4121442e31caSJeff Layton retry: 412211a7b371SAneesh Kumar K.V error = user_path_at(olddfd, oldname, how, &old_path); 41231da177e4SLinus Torvalds if (error) 41242ad94ae6SAl Viro return error; 41252ad94ae6SAl Viro 4126442e31caSJeff Layton new_dentry = user_path_create(newdfd, newname, &new_path, 4127442e31caSJeff Layton (how & LOOKUP_REVAL)); 41281da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 41296902d925SDave Hansen if (IS_ERR(new_dentry)) 4130dae6ad8fSAl Viro goto out; 4131dae6ad8fSAl Viro 4132dae6ad8fSAl Viro error = -EXDEV; 4133dae6ad8fSAl Viro if (old_path.mnt != new_path.mnt) 4134dae6ad8fSAl Viro goto out_dput; 4135800179c9SKees Cook error = may_linkat(&old_path); 4136800179c9SKees Cook if (unlikely(error)) 4137800179c9SKees Cook goto out_dput; 4138dae6ad8fSAl Viro error = security_path_link(old_path.dentry, &new_path, new_dentry); 4139be6d3e56SKentaro Takeda if (error) 4140a8104a9fSAl Viro goto out_dput; 4141146a8595SJ. Bruce Fields error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode); 414275c3f29dSDave Hansen out_dput: 4143921a1650SAl Viro done_path_create(&new_path, new_dentry); 4144146a8595SJ. Bruce Fields if (delegated_inode) { 4145146a8595SJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4146d22e6338SOleg Drokin if (!error) { 4147d22e6338SOleg Drokin path_put(&old_path); 4148146a8595SJ. Bruce Fields goto retry; 4149146a8595SJ. Bruce Fields } 4150d22e6338SOleg Drokin } 4151442e31caSJeff Layton if (retry_estale(error, how)) { 4152d22e6338SOleg Drokin path_put(&old_path); 4153442e31caSJeff Layton how |= LOOKUP_REVAL; 4154442e31caSJeff Layton goto retry; 4155442e31caSJeff Layton } 41561da177e4SLinus Torvalds out: 41572d8f3038SAl Viro path_put(&old_path); 41581da177e4SLinus Torvalds 41591da177e4SLinus Torvalds return error; 41601da177e4SLinus Torvalds } 41611da177e4SLinus Torvalds 416246ea89ebSDominik Brodowski SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, 416346ea89ebSDominik Brodowski int, newdfd, const char __user *, newname, int, flags) 416446ea89ebSDominik Brodowski { 416546ea89ebSDominik Brodowski return do_linkat(olddfd, oldname, newdfd, newname, flags); 416646ea89ebSDominik Brodowski } 416746ea89ebSDominik Brodowski 41683480b257SHeiko Carstens SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) 41695590ff0dSUlrich Drepper { 417046ea89ebSDominik Brodowski return do_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 41715590ff0dSUlrich Drepper } 41725590ff0dSUlrich Drepper 4173bc27027aSMiklos Szeredi /** 4174bc27027aSMiklos Szeredi * vfs_rename - rename a filesystem object 4175bc27027aSMiklos Szeredi * @old_dir: parent of source 4176bc27027aSMiklos Szeredi * @old_dentry: source 4177bc27027aSMiklos Szeredi * @new_dir: parent of destination 4178bc27027aSMiklos Szeredi * @new_dentry: destination 4179bc27027aSMiklos Szeredi * @delegated_inode: returns an inode needing a delegation break 4180520c8b16SMiklos Szeredi * @flags: rename flags 4181bc27027aSMiklos Szeredi * 4182bc27027aSMiklos Szeredi * The caller must hold multiple mutexes--see lock_rename()). 4183bc27027aSMiklos Szeredi * 4184bc27027aSMiklos Szeredi * If vfs_rename discovers a delegation in need of breaking at either 4185bc27027aSMiklos Szeredi * the source or destination, it will return -EWOULDBLOCK and return a 4186bc27027aSMiklos Szeredi * reference to the inode in delegated_inode. The caller should then 4187bc27027aSMiklos Szeredi * break the delegation and retry. Because breaking a delegation may 4188bc27027aSMiklos Szeredi * take a long time, the caller should drop all locks before doing 4189bc27027aSMiklos Szeredi * so. 4190bc27027aSMiklos Szeredi * 4191bc27027aSMiklos Szeredi * Alternatively, a caller may pass NULL for delegated_inode. This may 4192bc27027aSMiklos Szeredi * be appropriate for callers that expect the underlying filesystem not 4193bc27027aSMiklos Szeredi * to be NFS exported. 4194bc27027aSMiklos Szeredi * 41951da177e4SLinus Torvalds * The worst of all namespace operations - renaming directory. "Perverted" 41961da177e4SLinus Torvalds * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 41971da177e4SLinus Torvalds * Problems: 41980117d427SMauro Carvalho Chehab * 4199d03b29a2SJ. Bruce Fields * a) we can get into loop creation. 42001da177e4SLinus Torvalds * b) race potential - two innocent renames can create a loop together. 42011da177e4SLinus Torvalds * That's where 4.4 screws up. Current fix: serialization on 4202a11f3a05SArjan van de Ven * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 42031da177e4SLinus Torvalds * story. 42046cedba89SJ. Bruce Fields * c) we have to lock _four_ objects - parents and victim (if it exists), 42056cedba89SJ. Bruce Fields * and source (if it is not a directory). 42061b1dcc1bSJes Sorensen * And that - after we got ->i_mutex on parents (until then we don't know 42071da177e4SLinus Torvalds * whether the target exists). Solution: try to be smart with locking 42081da177e4SLinus Torvalds * order for inodes. We rely on the fact that tree topology may change 4209a11f3a05SArjan van de Ven * only under ->s_vfs_rename_mutex _and_ that parent of the object we 42101da177e4SLinus Torvalds * move will be locked. Thus we can rank directories by the tree 42111da177e4SLinus Torvalds * (ancestors first) and rank all non-directories after them. 42121da177e4SLinus Torvalds * That works since everybody except rename does "lock parent, lookup, 4213a11f3a05SArjan van de Ven * lock child" and rename is under ->s_vfs_rename_mutex. 42141da177e4SLinus Torvalds * HOWEVER, it relies on the assumption that any object with ->lookup() 42151da177e4SLinus Torvalds * has no more than 1 dentry. If "hybrid" objects will ever appear, 42161da177e4SLinus Torvalds * we'd better make sure that there's no link(2) for them. 4217e4eaac06SSage Weil * d) conversion from fhandle to dentry may come in the wrong moment - when 42181b1dcc1bSJes Sorensen * we are removing the target. Solution: we will have to grab ->i_mutex 42191da177e4SLinus Torvalds * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 4220c41b20e7SAdam Buchbinder * ->i_mutex on parents, which works but leads to some truly excessive 42211da177e4SLinus Torvalds * locking]. 42221da177e4SLinus Torvalds */ 42231da177e4SLinus Torvalds int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, 42248e6d782cSJ. Bruce Fields struct inode *new_dir, struct dentry *new_dentry, 4225520c8b16SMiklos Szeredi struct inode **delegated_inode, unsigned int flags) 42261da177e4SLinus Torvalds { 42271da177e4SLinus Torvalds int error; 4228bc27027aSMiklos Szeredi bool is_dir = d_is_dir(old_dentry); 4229bc27027aSMiklos Szeredi struct inode *source = old_dentry->d_inode; 4230bc27027aSMiklos Szeredi struct inode *target = new_dentry->d_inode; 4231da1ce067SMiklos Szeredi bool new_is_dir = false; 4232da1ce067SMiklos Szeredi unsigned max_links = new_dir->i_sb->s_max_links; 423349d31c2fSAl Viro struct name_snapshot old_name; 42341da177e4SLinus Torvalds 42358d3e2936SMiklos Szeredi if (source == target) 42361da177e4SLinus Torvalds return 0; 42371da177e4SLinus Torvalds 42381da177e4SLinus Torvalds error = may_delete(old_dir, old_dentry, is_dir); 42391da177e4SLinus Torvalds if (error) 42401da177e4SLinus Torvalds return error; 42411da177e4SLinus Torvalds 4242da1ce067SMiklos Szeredi if (!target) { 4243a95164d9SMiklos Szeredi error = may_create(new_dir, new_dentry); 4244da1ce067SMiklos Szeredi } else { 4245da1ce067SMiklos Szeredi new_is_dir = d_is_dir(new_dentry); 4246da1ce067SMiklos Szeredi 4247da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 42481da177e4SLinus Torvalds error = may_delete(new_dir, new_dentry, is_dir); 4249da1ce067SMiklos Szeredi else 4250da1ce067SMiklos Szeredi error = may_delete(new_dir, new_dentry, new_is_dir); 4251da1ce067SMiklos Szeredi } 42521da177e4SLinus Torvalds if (error) 42531da177e4SLinus Torvalds return error; 42541da177e4SLinus Torvalds 42552773bf00SMiklos Szeredi if (!old_dir->i_op->rename) 42561da177e4SLinus Torvalds return -EPERM; 42571da177e4SLinus Torvalds 4258bc27027aSMiklos Szeredi /* 4259bc27027aSMiklos Szeredi * If we are going to change the parent - check write permissions, 4260bc27027aSMiklos Szeredi * we'll need to flip '..'. 4261bc27027aSMiklos Szeredi */ 4262da1ce067SMiklos Szeredi if (new_dir != old_dir) { 4263da1ce067SMiklos Szeredi if (is_dir) { 4264bc27027aSMiklos Szeredi error = inode_permission(source, MAY_WRITE); 4265bc27027aSMiklos Szeredi if (error) 4266bc27027aSMiklos Szeredi return error; 4267bc27027aSMiklos Szeredi } 4268da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && new_is_dir) { 4269da1ce067SMiklos Szeredi error = inode_permission(target, MAY_WRITE); 4270da1ce067SMiklos Szeredi if (error) 4271da1ce067SMiklos Szeredi return error; 4272da1ce067SMiklos Szeredi } 4273da1ce067SMiklos Szeredi } 42740eeca283SRobert Love 42750b3974ebSMiklos Szeredi error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry, 42760b3974ebSMiklos Szeredi flags); 4277bc27027aSMiklos Szeredi if (error) 4278bc27027aSMiklos Szeredi return error; 4279bc27027aSMiklos Szeredi 428049d31c2fSAl Viro take_dentry_name_snapshot(&old_name, old_dentry); 4281bc27027aSMiklos Szeredi dget(new_dentry); 4282da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4283bc27027aSMiklos Szeredi lock_two_nondirectories(source, target); 4284bc27027aSMiklos Szeredi else if (target) 42855955102cSAl Viro inode_lock(target); 4286bc27027aSMiklos Szeredi 4287bc27027aSMiklos Szeredi error = -EBUSY; 42887af1364fSEric W. Biederman if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry)) 4289bc27027aSMiklos Szeredi goto out; 4290bc27027aSMiklos Szeredi 4291da1ce067SMiklos Szeredi if (max_links && new_dir != old_dir) { 4292bc27027aSMiklos Szeredi error = -EMLINK; 4293da1ce067SMiklos Szeredi if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links) 4294bc27027aSMiklos Szeredi goto out; 4295da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir && 4296da1ce067SMiklos Szeredi old_dir->i_nlink >= max_links) 4297da1ce067SMiklos Szeredi goto out; 4298da1ce067SMiklos Szeredi } 4299da1ce067SMiklos Szeredi if (!is_dir) { 4300bc27027aSMiklos Szeredi error = try_break_deleg(source, delegated_inode); 4301bc27027aSMiklos Szeredi if (error) 4302bc27027aSMiklos Szeredi goto out; 4303da1ce067SMiklos Szeredi } 4304da1ce067SMiklos Szeredi if (target && !new_is_dir) { 4305bc27027aSMiklos Szeredi error = try_break_deleg(target, delegated_inode); 4306bc27027aSMiklos Szeredi if (error) 4307bc27027aSMiklos Szeredi goto out; 4308bc27027aSMiklos Szeredi } 4309520c8b16SMiklos Szeredi error = old_dir->i_op->rename(old_dir, old_dentry, 4310520c8b16SMiklos Szeredi new_dir, new_dentry, flags); 4311bc27027aSMiklos Szeredi if (error) 4312bc27027aSMiklos Szeredi goto out; 4313bc27027aSMiklos Szeredi 4314da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE) && target) { 43158767712fSAl Viro if (is_dir) { 43168767712fSAl Viro shrink_dcache_parent(new_dentry); 4317bc27027aSMiklos Szeredi target->i_flags |= S_DEAD; 43188767712fSAl Viro } 4319bc27027aSMiklos Szeredi dont_mount(new_dentry); 43208ed936b5SEric W. Biederman detach_mounts(new_dentry); 4321bc27027aSMiklos Szeredi } 4322da1ce067SMiklos Szeredi if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) { 4323da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 4324bc27027aSMiklos Szeredi d_move(old_dentry, new_dentry); 4325da1ce067SMiklos Szeredi else 4326da1ce067SMiklos Szeredi d_exchange(old_dentry, new_dentry); 4327da1ce067SMiklos Szeredi } 4328bc27027aSMiklos Szeredi out: 4329da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4330bc27027aSMiklos Szeredi unlock_two_nondirectories(source, target); 4331bc27027aSMiklos Szeredi else if (target) 43325955102cSAl Viro inode_unlock(target); 4333bc27027aSMiklos Szeredi dput(new_dentry); 4334da1ce067SMiklos Szeredi if (!error) { 4335f4ec3a3dSAl Viro fsnotify_move(old_dir, new_dir, &old_name.name, is_dir, 4336da1ce067SMiklos Szeredi !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry); 4337da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4338f4ec3a3dSAl Viro fsnotify_move(new_dir, old_dir, &old_dentry->d_name, 4339da1ce067SMiklos Szeredi new_is_dir, NULL, new_dentry); 4340da1ce067SMiklos Szeredi } 4341da1ce067SMiklos Szeredi } 434249d31c2fSAl Viro release_dentry_name_snapshot(&old_name); 43430eeca283SRobert Love 43441da177e4SLinus Torvalds return error; 43451da177e4SLinus Torvalds } 43464d359507SAl Viro EXPORT_SYMBOL(vfs_rename); 43471da177e4SLinus Torvalds 4348ee81feb6SDominik Brodowski static int do_renameat2(int olddfd, const char __user *oldname, int newdfd, 4349ee81feb6SDominik Brodowski const char __user *newname, unsigned int flags) 43501da177e4SLinus Torvalds { 43511da177e4SLinus Torvalds struct dentry *old_dentry, *new_dentry; 43521da177e4SLinus Torvalds struct dentry *trap; 4353f5beed75SAl Viro struct path old_path, new_path; 4354f5beed75SAl Viro struct qstr old_last, new_last; 4355f5beed75SAl Viro int old_type, new_type; 43568e6d782cSJ. Bruce Fields struct inode *delegated_inode = NULL; 435791a27b2aSJeff Layton struct filename *from; 435891a27b2aSJeff Layton struct filename *to; 4359f5beed75SAl Viro unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET; 4360c6a94284SJeff Layton bool should_retry = false; 43612ad94ae6SAl Viro int error; 4362520c8b16SMiklos Szeredi 43630d7a8555SMiklos Szeredi if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) 4364da1ce067SMiklos Szeredi return -EINVAL; 4365da1ce067SMiklos Szeredi 43660d7a8555SMiklos Szeredi if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) && 43670d7a8555SMiklos Szeredi (flags & RENAME_EXCHANGE)) 4368520c8b16SMiklos Szeredi return -EINVAL; 4369520c8b16SMiklos Szeredi 4370f5beed75SAl Viro if (flags & RENAME_EXCHANGE) 4371f5beed75SAl Viro target_flags = 0; 4372f5beed75SAl Viro 4373c6a94284SJeff Layton retry: 4374c1d4dd27SAl Viro from = filename_parentat(olddfd, getname(oldname), lookup_flags, 4375c1d4dd27SAl Viro &old_path, &old_last, &old_type); 437691a27b2aSJeff Layton if (IS_ERR(from)) { 437791a27b2aSJeff Layton error = PTR_ERR(from); 43781da177e4SLinus Torvalds goto exit; 437991a27b2aSJeff Layton } 43801da177e4SLinus Torvalds 4381c1d4dd27SAl Viro to = filename_parentat(newdfd, getname(newname), lookup_flags, 4382c1d4dd27SAl Viro &new_path, &new_last, &new_type); 438391a27b2aSJeff Layton if (IS_ERR(to)) { 438491a27b2aSJeff Layton error = PTR_ERR(to); 43851da177e4SLinus Torvalds goto exit1; 438691a27b2aSJeff Layton } 43871da177e4SLinus Torvalds 43881da177e4SLinus Torvalds error = -EXDEV; 4389f5beed75SAl Viro if (old_path.mnt != new_path.mnt) 43901da177e4SLinus Torvalds goto exit2; 43911da177e4SLinus Torvalds 43921da177e4SLinus Torvalds error = -EBUSY; 4393f5beed75SAl Viro if (old_type != LAST_NORM) 43941da177e4SLinus Torvalds goto exit2; 43951da177e4SLinus Torvalds 43960a7c3937SMiklos Szeredi if (flags & RENAME_NOREPLACE) 43970a7c3937SMiklos Szeredi error = -EEXIST; 4398f5beed75SAl Viro if (new_type != LAST_NORM) 43991da177e4SLinus Torvalds goto exit2; 44001da177e4SLinus Torvalds 4401f5beed75SAl Viro error = mnt_want_write(old_path.mnt); 4402c30dabfeSJan Kara if (error) 4403c30dabfeSJan Kara goto exit2; 4404c30dabfeSJan Kara 44058e6d782cSJ. Bruce Fields retry_deleg: 4406f5beed75SAl Viro trap = lock_rename(new_path.dentry, old_path.dentry); 44071da177e4SLinus Torvalds 4408f5beed75SAl Viro old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags); 44091da177e4SLinus Torvalds error = PTR_ERR(old_dentry); 44101da177e4SLinus Torvalds if (IS_ERR(old_dentry)) 44111da177e4SLinus Torvalds goto exit3; 44121da177e4SLinus Torvalds /* source must exist */ 44131da177e4SLinus Torvalds error = -ENOENT; 4414b18825a7SDavid Howells if (d_is_negative(old_dentry)) 44151da177e4SLinus Torvalds goto exit4; 4416f5beed75SAl Viro new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags); 44171da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 44181da177e4SLinus Torvalds if (IS_ERR(new_dentry)) 44191da177e4SLinus Torvalds goto exit4; 44200a7c3937SMiklos Szeredi error = -EEXIST; 44210a7c3937SMiklos Szeredi if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry)) 44220a7c3937SMiklos Szeredi goto exit5; 4423da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4424da1ce067SMiklos Szeredi error = -ENOENT; 4425da1ce067SMiklos Szeredi if (d_is_negative(new_dentry)) 4426da1ce067SMiklos Szeredi goto exit5; 4427da1ce067SMiklos Szeredi 4428da1ce067SMiklos Szeredi if (!d_is_dir(new_dentry)) { 4429da1ce067SMiklos Szeredi error = -ENOTDIR; 4430f5beed75SAl Viro if (new_last.name[new_last.len]) 4431da1ce067SMiklos Szeredi goto exit5; 4432da1ce067SMiklos Szeredi } 4433da1ce067SMiklos Szeredi } 44340a7c3937SMiklos Szeredi /* unless the source is a directory trailing slashes give -ENOTDIR */ 44350a7c3937SMiklos Szeredi if (!d_is_dir(old_dentry)) { 44360a7c3937SMiklos Szeredi error = -ENOTDIR; 4437f5beed75SAl Viro if (old_last.name[old_last.len]) 44380a7c3937SMiklos Szeredi goto exit5; 4439f5beed75SAl Viro if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len]) 44400a7c3937SMiklos Szeredi goto exit5; 44410a7c3937SMiklos Szeredi } 44420a7c3937SMiklos Szeredi /* source should not be ancestor of target */ 44430a7c3937SMiklos Szeredi error = -EINVAL; 44440a7c3937SMiklos Szeredi if (old_dentry == trap) 44450a7c3937SMiklos Szeredi goto exit5; 44461da177e4SLinus Torvalds /* target should not be an ancestor of source */ 4447da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 44481da177e4SLinus Torvalds error = -ENOTEMPTY; 44491da177e4SLinus Torvalds if (new_dentry == trap) 44501da177e4SLinus Torvalds goto exit5; 44511da177e4SLinus Torvalds 4452f5beed75SAl Viro error = security_path_rename(&old_path, old_dentry, 4453f5beed75SAl Viro &new_path, new_dentry, flags); 4454be6d3e56SKentaro Takeda if (error) 4455c30dabfeSJan Kara goto exit5; 4456f5beed75SAl Viro error = vfs_rename(old_path.dentry->d_inode, old_dentry, 4457f5beed75SAl Viro new_path.dentry->d_inode, new_dentry, 4458520c8b16SMiklos Szeredi &delegated_inode, flags); 44591da177e4SLinus Torvalds exit5: 44601da177e4SLinus Torvalds dput(new_dentry); 44611da177e4SLinus Torvalds exit4: 44621da177e4SLinus Torvalds dput(old_dentry); 44631da177e4SLinus Torvalds exit3: 4464f5beed75SAl Viro unlock_rename(new_path.dentry, old_path.dentry); 44658e6d782cSJ. Bruce Fields if (delegated_inode) { 44668e6d782cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 44678e6d782cSJ. Bruce Fields if (!error) 44688e6d782cSJ. Bruce Fields goto retry_deleg; 44698e6d782cSJ. Bruce Fields } 4470f5beed75SAl Viro mnt_drop_write(old_path.mnt); 44711da177e4SLinus Torvalds exit2: 4472c6a94284SJeff Layton if (retry_estale(error, lookup_flags)) 4473c6a94284SJeff Layton should_retry = true; 4474f5beed75SAl Viro path_put(&new_path); 44752ad94ae6SAl Viro putname(to); 44761da177e4SLinus Torvalds exit1: 4477f5beed75SAl Viro path_put(&old_path); 44781da177e4SLinus Torvalds putname(from); 4479c6a94284SJeff Layton if (should_retry) { 4480c6a94284SJeff Layton should_retry = false; 4481c6a94284SJeff Layton lookup_flags |= LOOKUP_REVAL; 4482c6a94284SJeff Layton goto retry; 4483c6a94284SJeff Layton } 44842ad94ae6SAl Viro exit: 44851da177e4SLinus Torvalds return error; 44861da177e4SLinus Torvalds } 44871da177e4SLinus Torvalds 4488ee81feb6SDominik Brodowski SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname, 4489ee81feb6SDominik Brodowski int, newdfd, const char __user *, newname, unsigned int, flags) 4490ee81feb6SDominik Brodowski { 4491ee81feb6SDominik Brodowski return do_renameat2(olddfd, oldname, newdfd, newname, flags); 4492ee81feb6SDominik Brodowski } 4493ee81feb6SDominik Brodowski 4494520c8b16SMiklos Szeredi SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, 4495520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname) 4496520c8b16SMiklos Szeredi { 4497ee81feb6SDominik Brodowski return do_renameat2(olddfd, oldname, newdfd, newname, 0); 4498520c8b16SMiklos Szeredi } 4499520c8b16SMiklos Szeredi 4500a26eab24SHeiko Carstens SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) 45015590ff0dSUlrich Drepper { 4502ee81feb6SDominik Brodowski return do_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 45035590ff0dSUlrich Drepper } 45045590ff0dSUlrich Drepper 45055d826c84SAl Viro int readlink_copy(char __user *buffer, int buflen, const char *link) 45061da177e4SLinus Torvalds { 45075d826c84SAl Viro int len = PTR_ERR(link); 45081da177e4SLinus Torvalds if (IS_ERR(link)) 45091da177e4SLinus Torvalds goto out; 45101da177e4SLinus Torvalds 45111da177e4SLinus Torvalds len = strlen(link); 45121da177e4SLinus Torvalds if (len > (unsigned) buflen) 45131da177e4SLinus Torvalds len = buflen; 45141da177e4SLinus Torvalds if (copy_to_user(buffer, link, len)) 45151da177e4SLinus Torvalds len = -EFAULT; 45161da177e4SLinus Torvalds out: 45171da177e4SLinus Torvalds return len; 45181da177e4SLinus Torvalds } 45191da177e4SLinus Torvalds 4520d60874cdSMiklos Szeredi /** 4521fd4a0edfSMiklos Szeredi * vfs_readlink - copy symlink body into userspace buffer 4522fd4a0edfSMiklos Szeredi * @dentry: dentry on which to get symbolic link 4523fd4a0edfSMiklos Szeredi * @buffer: user memory pointer 4524fd4a0edfSMiklos Szeredi * @buflen: size of buffer 4525fd4a0edfSMiklos Szeredi * 4526fd4a0edfSMiklos Szeredi * Does not touch atime. That's up to the caller if necessary 4527fd4a0edfSMiklos Szeredi * 4528fd4a0edfSMiklos Szeredi * Does not call security hook. 4529fd4a0edfSMiklos Szeredi */ 4530fd4a0edfSMiklos Szeredi int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen) 4531fd4a0edfSMiklos Szeredi { 4532fd4a0edfSMiklos Szeredi struct inode *inode = d_inode(dentry); 4533f2df5da6SAl Viro DEFINE_DELAYED_CALL(done); 4534f2df5da6SAl Viro const char *link; 4535f2df5da6SAl Viro int res; 4536fd4a0edfSMiklos Szeredi 453776fca90eSMiklos Szeredi if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) { 453876fca90eSMiklos Szeredi if (unlikely(inode->i_op->readlink)) 453976fca90eSMiklos Szeredi return inode->i_op->readlink(dentry, buffer, buflen); 454076fca90eSMiklos Szeredi 454176fca90eSMiklos Szeredi if (!d_is_symlink(dentry)) 4542fd4a0edfSMiklos Szeredi return -EINVAL; 4543fd4a0edfSMiklos Szeredi 454476fca90eSMiklos Szeredi spin_lock(&inode->i_lock); 454576fca90eSMiklos Szeredi inode->i_opflags |= IOP_DEFAULT_READLINK; 454676fca90eSMiklos Szeredi spin_unlock(&inode->i_lock); 454776fca90eSMiklos Szeredi } 454876fca90eSMiklos Szeredi 45494c4f7c19SEric Biggers link = READ_ONCE(inode->i_link); 4550f2df5da6SAl Viro if (!link) { 4551f2df5da6SAl Viro link = inode->i_op->get_link(dentry, inode, &done); 4552f2df5da6SAl Viro if (IS_ERR(link)) 4553f2df5da6SAl Viro return PTR_ERR(link); 4554f2df5da6SAl Viro } 4555f2df5da6SAl Viro res = readlink_copy(buffer, buflen, link); 4556f2df5da6SAl Viro do_delayed_call(&done); 4557f2df5da6SAl Viro return res; 4558fd4a0edfSMiklos Szeredi } 4559fd4a0edfSMiklos Szeredi EXPORT_SYMBOL(vfs_readlink); 45601da177e4SLinus Torvalds 4561d60874cdSMiklos Szeredi /** 4562d60874cdSMiklos Szeredi * vfs_get_link - get symlink body 4563d60874cdSMiklos Szeredi * @dentry: dentry on which to get symbolic link 4564d60874cdSMiklos Szeredi * @done: caller needs to free returned data with this 4565d60874cdSMiklos Szeredi * 4566d60874cdSMiklos Szeredi * Calls security hook and i_op->get_link() on the supplied inode. 4567d60874cdSMiklos Szeredi * 4568d60874cdSMiklos Szeredi * It does not touch atime. That's up to the caller if necessary. 4569d60874cdSMiklos Szeredi * 4570d60874cdSMiklos Szeredi * Does not work on "special" symlinks like /proc/$$/fd/N 4571d60874cdSMiklos Szeredi */ 4572d60874cdSMiklos Szeredi const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done) 4573d60874cdSMiklos Szeredi { 4574d60874cdSMiklos Szeredi const char *res = ERR_PTR(-EINVAL); 4575d60874cdSMiklos Szeredi struct inode *inode = d_inode(dentry); 4576d60874cdSMiklos Szeredi 4577d60874cdSMiklos Szeredi if (d_is_symlink(dentry)) { 4578d60874cdSMiklos Szeredi res = ERR_PTR(security_inode_readlink(dentry)); 4579d60874cdSMiklos Szeredi if (!res) 4580d60874cdSMiklos Szeredi res = inode->i_op->get_link(dentry, inode, done); 4581d60874cdSMiklos Szeredi } 4582d60874cdSMiklos Szeredi return res; 4583d60874cdSMiklos Szeredi } 4584d60874cdSMiklos Szeredi EXPORT_SYMBOL(vfs_get_link); 4585d60874cdSMiklos Szeredi 45861da177e4SLinus Torvalds /* get the link contents into pagecache */ 45876b255391SAl Viro const char *page_get_link(struct dentry *dentry, struct inode *inode, 4588fceef393SAl Viro struct delayed_call *callback) 45891da177e4SLinus Torvalds { 4590ebd09abbSDuane Griffin char *kaddr; 45911da177e4SLinus Torvalds struct page *page; 45926b255391SAl Viro struct address_space *mapping = inode->i_mapping; 45936b255391SAl Viro 4594d3883d4fSAl Viro if (!dentry) { 4595d3883d4fSAl Viro page = find_get_page(mapping, 0); 4596d3883d4fSAl Viro if (!page) 45976b255391SAl Viro return ERR_PTR(-ECHILD); 4598d3883d4fSAl Viro if (!PageUptodate(page)) { 4599d3883d4fSAl Viro put_page(page); 4600d3883d4fSAl Viro return ERR_PTR(-ECHILD); 4601d3883d4fSAl Viro } 4602d3883d4fSAl Viro } else { 4603090d2b18SPekka Enberg page = read_mapping_page(mapping, 0, NULL); 46041da177e4SLinus Torvalds if (IS_ERR(page)) 46056fe6900eSNick Piggin return (char*)page; 4606d3883d4fSAl Viro } 4607fceef393SAl Viro set_delayed_call(callback, page_put_link, page); 460821fc61c7SAl Viro BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM); 460921fc61c7SAl Viro kaddr = page_address(page); 46106b255391SAl Viro nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1); 4611ebd09abbSDuane Griffin return kaddr; 46121da177e4SLinus Torvalds } 46131da177e4SLinus Torvalds 46146b255391SAl Viro EXPORT_SYMBOL(page_get_link); 46151da177e4SLinus Torvalds 4616fceef393SAl Viro void page_put_link(void *arg) 46171da177e4SLinus Torvalds { 4618fceef393SAl Viro put_page(arg); 46191da177e4SLinus Torvalds } 46204d359507SAl Viro EXPORT_SYMBOL(page_put_link); 46211da177e4SLinus Torvalds 4622aa80deabSAl Viro int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 4623aa80deabSAl Viro { 4624fceef393SAl Viro DEFINE_DELAYED_CALL(done); 46256b255391SAl Viro int res = readlink_copy(buffer, buflen, 46266b255391SAl Viro page_get_link(dentry, d_inode(dentry), 4627fceef393SAl Viro &done)); 4628fceef393SAl Viro do_delayed_call(&done); 4629aa80deabSAl Viro return res; 4630aa80deabSAl Viro } 4631aa80deabSAl Viro EXPORT_SYMBOL(page_readlink); 4632aa80deabSAl Viro 463354566b2cSNick Piggin /* 463454566b2cSNick Piggin * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS 463554566b2cSNick Piggin */ 463654566b2cSNick Piggin int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) 46371da177e4SLinus Torvalds { 46381da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 46390adb25d2SKirill Korotaev struct page *page; 4640afddba49SNick Piggin void *fsdata; 4641beb497abSDmitriy Monakhov int err; 4642c718a975STetsuo Handa unsigned int flags = 0; 464354566b2cSNick Piggin if (nofs) 464454566b2cSNick Piggin flags |= AOP_FLAG_NOFS; 46451da177e4SLinus Torvalds 46467e53cac4SNeilBrown retry: 4647afddba49SNick Piggin err = pagecache_write_begin(NULL, mapping, 0, len-1, 464854566b2cSNick Piggin flags, &page, &fsdata); 46491da177e4SLinus Torvalds if (err) 4650afddba49SNick Piggin goto fail; 4651afddba49SNick Piggin 465221fc61c7SAl Viro memcpy(page_address(page), symname, len-1); 4653afddba49SNick Piggin 4654afddba49SNick Piggin err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, 4655afddba49SNick Piggin page, fsdata); 46561da177e4SLinus Torvalds if (err < 0) 46571da177e4SLinus Torvalds goto fail; 4658afddba49SNick Piggin if (err < len-1) 4659afddba49SNick Piggin goto retry; 4660afddba49SNick Piggin 46611da177e4SLinus Torvalds mark_inode_dirty(inode); 46621da177e4SLinus Torvalds return 0; 46631da177e4SLinus Torvalds fail: 46641da177e4SLinus Torvalds return err; 46651da177e4SLinus Torvalds } 46664d359507SAl Viro EXPORT_SYMBOL(__page_symlink); 46671da177e4SLinus Torvalds 46680adb25d2SKirill Korotaev int page_symlink(struct inode *inode, const char *symname, int len) 46690adb25d2SKirill Korotaev { 46700adb25d2SKirill Korotaev return __page_symlink(inode, symname, len, 4671c62d2555SMichal Hocko !mapping_gfp_constraint(inode->i_mapping, __GFP_FS)); 46720adb25d2SKirill Korotaev } 46734d359507SAl Viro EXPORT_SYMBOL(page_symlink); 46740adb25d2SKirill Korotaev 467592e1d5beSArjan van de Ven const struct inode_operations page_symlink_inode_operations = { 46766b255391SAl Viro .get_link = page_get_link, 46771da177e4SLinus Torvalds }; 46781da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink_inode_operations); 4679