Linux Audio

Check our new training course

Loading...
   1/*
   2 *  linux/fs/namei.c
   3 *
   4 *  Copyright (C) 1991, 1992  Linus Torvalds
   5 */
   6
   7/*
   8 * Some corrections by tytso.
   9 */
  10
  11/* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
  12 * lookup logic.
  13 */
  14/* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
  15 */
  16
  17#include <linux/init.h>
  18#include <linux/export.h>
  19#include <linux/kernel.h>
  20#include <linux/slab.h>
  21#include <linux/fs.h>
  22#include <linux/namei.h>
  23#include <linux/pagemap.h>
  24#include <linux/fsnotify.h>
  25#include <linux/personality.h>
  26#include <linux/security.h>
  27#include <linux/ima.h>
  28#include <linux/syscalls.h>
  29#include <linux/mount.h>
  30#include <linux/audit.h>
  31#include <linux/capability.h>
  32#include <linux/file.h>
  33#include <linux/fcntl.h>
  34#include <linux/device_cgroup.h>
  35#include <linux/fs_struct.h>
  36#include <linux/posix_acl.h>
  37#include <linux/hash.h>
  38#include <asm/uaccess.h>
  39
  40#include "internal.h"
  41#include "mount.h"
  42
  43/* [Feb-1997 T. Schoebel-Theuer]
  44 * Fundamental changes in the pathname lookup mechanisms (namei)
  45 * were necessary because of omirr.  The reason is that omirr needs
  46 * to know the _real_ pathname, not the user-supplied one, in case
  47 * of symlinks (and also when transname replacements occur).
  48 *
  49 * The new code replaces the old recursive symlink resolution with
  50 * an iterative one (in case of non-nested symlink chains).  It does
  51 * this with calls to <fs>_follow_link().
  52 * As a side effect, dir_namei(), _namei() and follow_link() are now 
  53 * replaced with a single function lookup_dentry() that can handle all 
  54 * the special cases of the former code.
  55 *
  56 * With the new dcache, the pathname is stored at each inode, at least as
  57 * long as the refcount of the inode is positive.  As a side effect, the
  58 * size of the dcache depends on the inode cache and thus is dynamic.
  59 *
  60 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
  61 * resolution to correspond with current state of the code.
  62 *
  63 * Note that the symlink resolution is not *completely* iterative.
  64 * There is still a significant amount of tail- and mid- recursion in
  65 * the algorithm.  Also, note that <fs>_readlink() is not used in
  66 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
  67 * may return different results than <fs>_follow_link().  Many virtual
  68 * filesystems (including /proc) exhibit this behavior.
  69 */
  70
  71/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
  72 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
  73 * and the name already exists in form of a symlink, try to create the new
  74 * name indicated by the symlink. The old code always complained that the
  75 * name already exists, due to not following the symlink even if its target
  76 * is nonexistent.  The new semantics affects also mknod() and link() when
  77 * the name is a symlink pointing to a non-existent name.
  78 *
  79 * I don't know which semantics is the right one, since I have no access
  80 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
  81 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
  82 * "old" one. Personally, I think the new semantics is much more logical.
  83 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
  84 * file does succeed in both HP-UX and SunOs, but not in Solaris
  85 * and in the old Linux semantics.
  86 */
  87
  88/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
  89 * semantics.  See the comments in "open_namei" and "do_link" below.
  90 *
  91 * [10-Sep-98 Alan Modra] Another symlink change.
  92 */
  93
  94/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
  95 *	inside the path - always follow.
  96 *	in the last component in creation/removal/renaming - never follow.
  97 *	if LOOKUP_FOLLOW passed - follow.
  98 *	if the pathname has trailing slashes - follow.
  99 *	otherwise - don't follow.
 100 * (applied in that order).
 101 *
 102 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
 103 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
 104 * During the 2.4 we need to fix the userland stuff depending on it -
 105 * hopefully we will be able to get rid of that wart in 2.5. So far only
 106 * XEmacs seems to be relying on it...
 107 */
 108/*
 109 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
 110 * implemented.  Let's see if raised priority of ->s_vfs_rename_mutex gives
 111 * any extra contention...
 112 */
 113
 114/* In order to reduce some races, while at the same time doing additional
 115 * checking and hopefully speeding things up, we copy filenames to the
 116 * kernel data space before using them..
 117 *
 118 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
 119 * PATH_MAX includes the nul terminator --RR.
 120 */
 121
 122#define EMBEDDED_NAME_MAX	(PATH_MAX - offsetof(struct filename, iname))
 123
 124struct filename *
 125getname_flags(const char __user *filename, int flags, int *empty)
 126{
 127	struct filename *result;
 128	char *kname;
 129	int len;
 130
 131	result = audit_reusename(filename);
 132	if (result)
 133		return result;
 134
 135	result = __getname();
 136	if (unlikely(!result))
 137		return ERR_PTR(-ENOMEM);
 138
 139	/*
 140	 * First, try to embed the struct filename inside the names_cache
 141	 * allocation
 142	 */
 143	kname = (char *)result->iname;
 144	result->name = kname;
 145
 146	len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX);
 147	if (unlikely(len < 0)) {
 148		__putname(result);
 149		return ERR_PTR(len);
 150	}
 151
 152	/*
 153	 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
 154	 * separate struct filename so we can dedicate the entire
 155	 * names_cache allocation for the pathname, and re-do the copy from
 156	 * userland.
 157	 */
 158	if (unlikely(len == EMBEDDED_NAME_MAX)) {
 159		const size_t size = offsetof(struct filename, iname[1]);
 160		kname = (char *)result;
 161
 162		/*
 163		 * size is chosen that way we to guarantee that
 164		 * result->iname[0] is within the same object and that
 165		 * kname can't be equal to result->iname, no matter what.
 166		 */
 167		result = kzalloc(size, GFP_KERNEL);
 168		if (unlikely(!result)) {
 169			__putname(kname);
 170			return ERR_PTR(-ENOMEM);
 171		}
 172		result->name = kname;
 173		len = strncpy_from_user(kname, filename, PATH_MAX);
 174		if (unlikely(len < 0)) {
 175			__putname(kname);
 176			kfree(result);
 177			return ERR_PTR(len);
 178		}
 179		if (unlikely(len == PATH_MAX)) {
 180			__putname(kname);
 181			kfree(result);
 182			return ERR_PTR(-ENAMETOOLONG);
 183		}
 184	}
 185
 186	result->refcnt = 1;
 187	/* The empty path is special. */
 188	if (unlikely(!len)) {
 189		if (empty)
 190			*empty = 1;
 191		if (!(flags & LOOKUP_EMPTY)) {
 192			putname(result);
 193			return ERR_PTR(-ENOENT);
 194		}
 195	}
 196
 197	result->uptr = filename;
 198	result->aname = NULL;
 199	audit_getname(result);
 200	return result;
 201}
 202
 203struct filename *
 204getname(const char __user * filename)
 205{
 206	return getname_flags(filename, 0, NULL);
 207}
 208
 209struct filename *
 210getname_kernel(const char * filename)
 211{
 212	struct filename *result;
 213	int len = strlen(filename) + 1;
 214
 215	result = __getname();
 216	if (unlikely(!result))
 217		return ERR_PTR(-ENOMEM);
 218
 219	if (len <= EMBEDDED_NAME_MAX) {
 220		result->name = (char *)result->iname;
 221	} else if (len <= PATH_MAX) {
 222		struct filename *tmp;
 223
 224		tmp = kmalloc(sizeof(*tmp), GFP_KERNEL);
 225		if (unlikely(!tmp)) {
 226			__putname(result);
 227			return ERR_PTR(-ENOMEM);
 228		}
 229		tmp->name = (char *)result;
 230		result = tmp;
 231	} else {
 232		__putname(result);
 233		return ERR_PTR(-ENAMETOOLONG);
 234	}
 235	memcpy((char *)result->name, filename, len);
 236	result->uptr = NULL;
 237	result->aname = NULL;
 238	result->refcnt = 1;
 239	audit_getname(result);
 240
 241	return result;
 242}
 243
 244void putname(struct filename *name)
 245{
 246	BUG_ON(name->refcnt <= 0);
 247
 248	if (--name->refcnt > 0)
 249		return;
 250
 251	if (name->name != name->iname) {
 252		__putname(name->name);
 253		kfree(name);
 254	} else
 255		__putname(name);
 256}
 257
 258static int check_acl(struct inode *inode, int mask)
 259{
 260#ifdef CONFIG_FS_POSIX_ACL
 261	struct posix_acl *acl;
 262
 263	if (mask & MAY_NOT_BLOCK) {
 264		acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
 265	        if (!acl)
 266	                return -EAGAIN;
 267		/* no ->get_acl() calls in RCU mode... */
 268		if (acl == ACL_NOT_CACHED)
 269			return -ECHILD;
 270	        return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
 271	}
 272
 273	acl = get_acl(inode, ACL_TYPE_ACCESS);
 274	if (IS_ERR(acl))
 275		return PTR_ERR(acl);
 276	if (acl) {
 277	        int error = posix_acl_permission(inode, acl, mask);
 278	        posix_acl_release(acl);
 279	        return error;
 280	}
 281#endif
 282
 283	return -EAGAIN;
 284}
 285
 286/*
 287 * This does the basic permission checking
 288 */
 289static int acl_permission_check(struct inode *inode, int mask)
 290{
 291	unsigned int mode = inode->i_mode;
 292
 293	if (likely(uid_eq(current_fsuid(), inode->i_uid)))
 294		mode >>= 6;
 295	else {
 296		if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
 297			int error = check_acl(inode, mask);
 298			if (error != -EAGAIN)
 299				return error;
 300		}
 301
 302		if (in_group_p(inode->i_gid))
 303			mode >>= 3;
 304	}
 305
 306	/*
 307	 * If the DACs are ok we don't need any capability check.
 308	 */
 309	if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
 310		return 0;
 311	return -EACCES;
 312}
 313
 314/**
 315 * generic_permission -  check for access rights on a Posix-like filesystem
 316 * @inode:	inode to check access rights for
 317 * @mask:	right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
 318 *
 319 * Used to check for read/write/execute permissions on a file.
 320 * We use "fsuid" for this, letting us set arbitrary permissions
 321 * for filesystem access without changing the "normal" uids which
 322 * are used for other things.
 323 *
 324 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
 325 * request cannot be satisfied (eg. requires blocking or too much complexity).
 326 * It would then be called again in ref-walk mode.
 327 */
 328int generic_permission(struct inode *inode, int mask)
 329{
 330	int ret;
 331
 332	/*
 333	 * Do the basic permission checks.
 334	 */
 335	ret = acl_permission_check(inode, mask);
 336	if (ret != -EACCES)
 337		return ret;
 338
 339	if (S_ISDIR(inode->i_mode)) {
 340		/* DACs are overridable for directories */
 341		if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
 342			return 0;
 343		if (!(mask & MAY_WRITE))
 344			if (capable_wrt_inode_uidgid(inode,
 345						     CAP_DAC_READ_SEARCH))
 346				return 0;
 347		return -EACCES;
 348	}
 349	/*
 350	 * Read/write DACs are always overridable.
 351	 * Executable DACs are overridable when there is
 352	 * at least one exec bit set.
 353	 */
 354	if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
 355		if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
 356			return 0;
 357
 358	/*
 359	 * Searching includes executable on directories, else just read.
 360	 */
 361	mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
 362	if (mask == MAY_READ)
 363		if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH))
 364			return 0;
 365
 366	return -EACCES;
 367}
 368EXPORT_SYMBOL(generic_permission);
 369
 370/*
 371 * We _really_ want to just do "generic_permission()" without
 372 * even looking at the inode->i_op values. So we keep a cache
 373 * flag in inode->i_opflags, that says "this has not special
 374 * permission function, use the fast case".
 375 */
 376static inline int do_inode_permission(struct inode *inode, int mask)
 377{
 378	if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
 379		if (likely(inode->i_op->permission))
 380			return inode->i_op->permission(inode, mask);
 381
 382		/* This gets set once for the inode lifetime */
 383		spin_lock(&inode->i_lock);
 384		inode->i_opflags |= IOP_FASTPERM;
 385		spin_unlock(&inode->i_lock);
 386	}
 387	return generic_permission(inode, mask);
 388}
 389
 390/**
 391 * __inode_permission - Check for access rights to a given inode
 392 * @inode: Inode to check permission on
 393 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
 394 *
 395 * Check for read/write/execute permissions on an inode.
 396 *
 397 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
 398 *
 399 * This does not check for a read-only file system.  You probably want
 400 * inode_permission().
 401 */
 402int __inode_permission(struct inode *inode, int mask)
 403{
 404	int retval;
 405
 406	if (unlikely(mask & MAY_WRITE)) {
 407		/*
 408		 * Nobody gets write access to an immutable file.
 409		 */
 410		if (IS_IMMUTABLE(inode))
 411			return -EACCES;
 412	}
 413
 414	retval = do_inode_permission(inode, mask);
 415	if (retval)
 416		return retval;
 417
 418	retval = devcgroup_inode_permission(inode, mask);
 419	if (retval)
 420		return retval;
 421
 422	return security_inode_permission(inode, mask);
 423}
 424EXPORT_SYMBOL(__inode_permission);
 425
 426/**
 427 * sb_permission - Check superblock-level permissions
 428 * @sb: Superblock of inode to check permission on
 429 * @inode: Inode to check permission on
 430 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
 431 *
 432 * Separate out file-system wide checks from inode-specific permission checks.
 433 */
 434static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
 435{
 436	if (unlikely(mask & MAY_WRITE)) {
 437		umode_t mode = inode->i_mode;
 438
 439		/* Nobody gets write access to a read-only fs. */
 440		if ((sb->s_flags & MS_RDONLY) &&
 441		    (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
 442			return -EROFS;
 443	}
 444	return 0;
 445}
 446
 447/**
 448 * inode_permission - Check for access rights to a given inode
 449 * @inode: Inode to check permission on
 450 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
 451 *
 452 * Check for read/write/execute permissions on an inode.  We use fs[ug]id for
 453 * this, letting us set arbitrary permissions for filesystem access without
 454 * changing the "normal" UIDs which are used for other things.
 455 *
 456 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
 457 */
 458int inode_permission(struct inode *inode, int mask)
 459{
 460	int retval;
 461
 462	retval = sb_permission(inode->i_sb, inode, mask);
 463	if (retval)
 464		return retval;
 465	return __inode_permission(inode, mask);
 466}
 467EXPORT_SYMBOL(inode_permission);
 468
 469/**
 470 * path_get - get a reference to a path
 471 * @path: path to get the reference to
 472 *
 473 * Given a path increment the reference count to the dentry and the vfsmount.
 474 */
 475void path_get(const struct path *path)
 476{
 477	mntget(path->mnt);
 478	dget(path->dentry);
 479}
 480EXPORT_SYMBOL(path_get);
 481
 482/**
 483 * path_put - put a reference to a path
 484 * @path: path to put the reference to
 485 *
 486 * Given a path decrement the reference count to the dentry and the vfsmount.
 487 */
 488void path_put(const struct path *path)
 489{
 490	dput(path->dentry);
 491	mntput(path->mnt);
 492}
 493EXPORT_SYMBOL(path_put);
 494
 495#define EMBEDDED_LEVELS 2
 496struct nameidata {
 497	struct path	path;
 498	struct qstr	last;
 499	struct path	root;
 500	struct inode	*inode; /* path.dentry.d_inode */
 501	unsigned int	flags;
 502	unsigned	seq, m_seq;
 503	int		last_type;
 504	unsigned	depth;
 505	int		total_link_count;
 506	struct saved {
 507		struct path link;
 508		struct delayed_call done;
 509		const char *name;
 510		unsigned seq;
 511	} *stack, internal[EMBEDDED_LEVELS];
 512	struct filename	*name;
 513	struct nameidata *saved;
 514	struct inode	*link_inode;
 515	unsigned	root_seq;
 516	int		dfd;
 517};
 518
 519static void set_nameidata(struct nameidata *p, int dfd, struct filename *name)
 520{
 521	struct nameidata *old = current->nameidata;
 522	p->stack = p->internal;
 523	p->dfd = dfd;
 524	p->name = name;
 525	p->total_link_count = old ? old->total_link_count : 0;
 526	p->saved = old;
 527	current->nameidata = p;
 528}
 529
 530static void restore_nameidata(void)
 531{
 532	struct nameidata *now = current->nameidata, *old = now->saved;
 533
 534	current->nameidata = old;
 535	if (old)
 536		old->total_link_count = now->total_link_count;
 537	if (now->stack != now->internal)
 538		kfree(now->stack);
 539}
 540
 541static int __nd_alloc_stack(struct nameidata *nd)
 542{
 543	struct saved *p;
 544
 545	if (nd->flags & LOOKUP_RCU) {
 546		p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
 547				  GFP_ATOMIC);
 548		if (unlikely(!p))
 549			return -ECHILD;
 550	} else {
 551		p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
 552				  GFP_KERNEL);
 553		if (unlikely(!p))
 554			return -ENOMEM;
 555	}
 556	memcpy(p, nd->internal, sizeof(nd->internal));
 557	nd->stack = p;
 558	return 0;
 559}
 560
 561/**
 562 * path_connected - Verify that a path->dentry is below path->mnt.mnt_root
 563 * @path: nameidate to verify
 564 *
 565 * Rename can sometimes move a file or directory outside of a bind
 566 * mount, path_connected allows those cases to be detected.
 567 */
 568static bool path_connected(const struct path *path)
 569{
 570	struct vfsmount *mnt = path->mnt;
 571
 572	/* Only bind mounts can have disconnected paths */
 573	if (mnt->mnt_root == mnt->mnt_sb->s_root)
 574		return true;
 575
 576	return is_subdir(path->dentry, mnt->mnt_root);
 577}
 578
 579static inline int nd_alloc_stack(struct nameidata *nd)
 580{
 581	if (likely(nd->depth != EMBEDDED_LEVELS))
 582		return 0;
 583	if (likely(nd->stack != nd->internal))
 584		return 0;
 585	return __nd_alloc_stack(nd);
 586}
 587
 588static void drop_links(struct nameidata *nd)
 589{
 590	int i = nd->depth;
 591	while (i--) {
 592		struct saved *last = nd->stack + i;
 593		do_delayed_call(&last->done);
 594		clear_delayed_call(&last->done);
 595	}
 596}
 597
 598static void terminate_walk(struct nameidata *nd)
 599{
 600	drop_links(nd);
 601	if (!(nd->flags & LOOKUP_RCU)) {
 602		int i;
 603		path_put(&nd->path);
 604		for (i = 0; i < nd->depth; i++)
 605			path_put(&nd->stack[i].link);
 606		if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
 607			path_put(&nd->root);
 608			nd->root.mnt = NULL;
 609		}
 610	} else {
 611		nd->flags &= ~LOOKUP_RCU;
 612		if (!(nd->flags & LOOKUP_ROOT))
 613			nd->root.mnt = NULL;
 614		rcu_read_unlock();
 615	}
 616	nd->depth = 0;
 617}
 618
 619/* path_put is needed afterwards regardless of success or failure */
 620static bool legitimize_path(struct nameidata *nd,
 621			    struct path *path, unsigned seq)
 622{
 623	int res = __legitimize_mnt(path->mnt, nd->m_seq);
 624	if (unlikely(res)) {
 625		if (res > 0)
 626			path->mnt = NULL;
 627		path->dentry = NULL;
 628		return false;
 629	}
 630	if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
 631		path->dentry = NULL;
 632		return false;
 633	}
 634	return !read_seqcount_retry(&path->dentry->d_seq, seq);
 635}
 636
 637static bool legitimize_links(struct nameidata *nd)
 638{
 639	int i;
 640	for (i = 0; i < nd->depth; i++) {
 641		struct saved *last = nd->stack + i;
 642		if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
 643			drop_links(nd);
 644			nd->depth = i + 1;
 645			return false;
 646		}
 647	}
 648	return true;
 649}
 650
 651/*
 652 * Path walking has 2 modes, rcu-walk and ref-walk (see
 653 * Documentation/filesystems/path-lookup.txt).  In situations when we can't
 654 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
 655 * normal reference counts on dentries and vfsmounts to transition to ref-walk
 656 * mode.  Refcounts are grabbed at the last known good point before rcu-walk
 657 * got stuck, so ref-walk may continue from there. If this is not successful
 658 * (eg. a seqcount has changed), then failure is returned and it's up to caller
 659 * to restart the path walk from the beginning in ref-walk mode.
 660 */
 661
 662/**
 663 * unlazy_walk - try to switch to ref-walk mode.
 664 * @nd: nameidata pathwalk data
 665 * @dentry: child of nd->path.dentry or NULL
 666 * @seq: seq number to check dentry against
 667 * Returns: 0 on success, -ECHILD on failure
 668 *
 669 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
 670 * for ref-walk mode.  @dentry must be a path found by a do_lookup call on
 671 * @nd or NULL.  Must be called from rcu-walk context.
 672 * Nothing should touch nameidata between unlazy_walk() failure and
 673 * terminate_walk().
 674 */
 675static int unlazy_walk(struct nameidata *nd, struct dentry *dentry, unsigned seq)
 676{
 677	struct dentry *parent = nd->path.dentry;
 678
 679	BUG_ON(!(nd->flags & LOOKUP_RCU));
 680
 681	nd->flags &= ~LOOKUP_RCU;
 682	if (unlikely(!legitimize_links(nd)))
 683		goto out2;
 684	if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq)))
 685		goto out2;
 686	if (unlikely(!lockref_get_not_dead(&parent->d_lockref)))
 687		goto out1;
 688
 689	/*
 690	 * For a negative lookup, the lookup sequence point is the parents
 691	 * sequence point, and it only needs to revalidate the parent dentry.
 692	 *
 693	 * For a positive lookup, we need to move both the parent and the
 694	 * dentry from the RCU domain to be properly refcounted. And the
 695	 * sequence number in the dentry validates *both* dentry counters,
 696	 * since we checked the sequence number of the parent after we got
 697	 * the child sequence number. So we know the parent must still
 698	 * be valid if the child sequence number is still valid.
 699	 */
 700	if (!dentry) {
 701		if (read_seqcount_retry(&parent->d_seq, nd->seq))
 702			goto out;
 703		BUG_ON(nd->inode != parent->d_inode);
 704	} else {
 705		if (!lockref_get_not_dead(&dentry->d_lockref))
 706			goto out;
 707		if (read_seqcount_retry(&dentry->d_seq, seq))
 708			goto drop_dentry;
 709	}
 710
 711	/*
 712	 * Sequence counts matched. Now make sure that the root is
 713	 * still valid and get it if required.
 714	 */
 715	if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
 716		if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) {
 717			rcu_read_unlock();
 718			dput(dentry);
 719			return -ECHILD;
 720		}
 721	}
 722
 723	rcu_read_unlock();
 724	return 0;
 725
 726drop_dentry:
 727	rcu_read_unlock();
 728	dput(dentry);
 729	goto drop_root_mnt;
 730out2:
 731	nd->path.mnt = NULL;
 732out1:
 733	nd->path.dentry = NULL;
 734out:
 735	rcu_read_unlock();
 736drop_root_mnt:
 737	if (!(nd->flags & LOOKUP_ROOT))
 738		nd->root.mnt = NULL;
 739	return -ECHILD;
 740}
 741
 742static int unlazy_link(struct nameidata *nd, struct path *link, unsigned seq)
 743{
 744	if (unlikely(!legitimize_path(nd, link, seq))) {
 745		drop_links(nd);
 746		nd->depth = 0;
 747		nd->flags &= ~LOOKUP_RCU;
 748		nd->path.mnt = NULL;
 749		nd->path.dentry = NULL;
 750		if (!(nd->flags & LOOKUP_ROOT))
 751			nd->root.mnt = NULL;
 752		rcu_read_unlock();
 753	} else if (likely(unlazy_walk(nd, NULL, 0)) == 0) {
 754		return 0;
 755	}
 756	path_put(link);
 757	return -ECHILD;
 758}
 759
 760static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
 761{
 762	return dentry->d_op->d_revalidate(dentry, flags);
 763}
 764
 765/**
 766 * complete_walk - successful completion of path walk
 767 * @nd:  pointer nameidata
 768 *
 769 * If we had been in RCU mode, drop out of it and legitimize nd->path.
 770 * Revalidate the final result, unless we'd already done that during
 771 * the path walk or the filesystem doesn't ask for it.  Return 0 on
 772 * success, -error on failure.  In case of failure caller does not
 773 * need to drop nd->path.
 774 */
 775static int complete_walk(struct nameidata *nd)
 776{
 777	struct dentry *dentry = nd->path.dentry;
 778	int status;
 779
 780	if (nd->flags & LOOKUP_RCU) {
 781		if (!(nd->flags & LOOKUP_ROOT))
 782			nd->root.mnt = NULL;
 783		if (unlikely(unlazy_walk(nd, NULL, 0)))
 784			return -ECHILD;
 785	}
 786
 787	if (likely(!(nd->flags & LOOKUP_JUMPED)))
 788		return 0;
 789
 790	if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
 791		return 0;
 792
 793	status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
 794	if (status > 0)
 795		return 0;
 796
 797	if (!status)
 798		status = -ESTALE;
 799
 800	return status;
 801}
 802
 803static void set_root(struct nameidata *nd)
 804{
 805	struct fs_struct *fs = current->fs;
 806
 807	if (nd->flags & LOOKUP_RCU) {
 808		unsigned seq;
 809
 810		do {
 811			seq = read_seqcount_begin(&fs->seq);
 812			nd->root = fs->root;
 813			nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
 814		} while (read_seqcount_retry(&fs->seq, seq));
 815	} else {
 816		get_fs_root(fs, &nd->root);
 817	}
 818}
 819
 820static void path_put_conditional(struct path *path, struct nameidata *nd)
 821{
 822	dput(path->dentry);
 823	if (path->mnt != nd->path.mnt)
 824		mntput(path->mnt);
 825}
 826
 827static inline void path_to_nameidata(const struct path *path,
 828					struct nameidata *nd)
 829{
 830	if (!(nd->flags & LOOKUP_RCU)) {
 831		dput(nd->path.dentry);
 832		if (nd->path.mnt != path->mnt)
 833			mntput(nd->path.mnt);
 834	}
 835	nd->path.mnt = path->mnt;
 836	nd->path.dentry = path->dentry;
 837}
 838
 839static int nd_jump_root(struct nameidata *nd)
 840{
 841	if (nd->flags & LOOKUP_RCU) {
 842		struct dentry *d;
 843		nd->path = nd->root;
 844		d = nd->path.dentry;
 845		nd->inode = d->d_inode;
 846		nd->seq = nd->root_seq;
 847		if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq)))
 848			return -ECHILD;
 849	} else {
 850		path_put(&nd->path);
 851		nd->path = nd->root;
 852		path_get(&nd->path);
 853		nd->inode = nd->path.dentry->d_inode;
 854	}
 855	nd->flags |= LOOKUP_JUMPED;
 856	return 0;
 857}
 858
 859/*
 860 * Helper to directly jump to a known parsed path from ->get_link,
 861 * caller must have taken a reference to path beforehand.
 862 */
 863void nd_jump_link(struct path *path)
 864{
 865	struct nameidata *nd = current->nameidata;
 866	path_put(&nd->path);
 867
 868	nd->path = *path;
 869	nd->inode = nd->path.dentry->d_inode;
 870	nd->flags |= LOOKUP_JUMPED;
 871}
 872
 873static inline void put_link(struct nameidata *nd)
 874{
 875	struct saved *last = nd->stack + --nd->depth;
 876	do_delayed_call(&last->done);
 877	if (!(nd->flags & LOOKUP_RCU))
 878		path_put(&last->link);
 879}
 880
 881int sysctl_protected_symlinks __read_mostly = 0;
 882int sysctl_protected_hardlinks __read_mostly = 0;
 883
 884/**
 885 * may_follow_link - Check symlink following for unsafe situations
 886 * @nd: nameidata pathwalk data
 887 *
 888 * In the case of the sysctl_protected_symlinks sysctl being enabled,
 889 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
 890 * in a sticky world-writable directory. This is to protect privileged
 891 * processes from failing races against path names that may change out
 892 * from under them by way of other users creating malicious symlinks.
 893 * It will permit symlinks to be followed only when outside a sticky
 894 * world-writable directory, or when the uid of the symlink and follower
 895 * match, or when the directory owner matches the symlink's owner.
 896 *
 897 * Returns 0 if following the symlink is allowed, -ve on error.
 898 */
 899static inline int may_follow_link(struct nameidata *nd)
 900{
 901	const struct inode *inode;
 902	const struct inode *parent;
 903
 904	if (!sysctl_protected_symlinks)
 905		return 0;
 906
 907	/* Allowed if owner and follower match. */
 908	inode = nd->link_inode;
 909	if (uid_eq(current_cred()->fsuid, inode->i_uid))
 910		return 0;
 911
 912	/* Allowed if parent directory not sticky and world-writable. */
 913	parent = nd->inode;
 914	if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
 915		return 0;
 916
 917	/* Allowed if parent directory and link owner match. */
 918	if (uid_eq(parent->i_uid, inode->i_uid))
 919		return 0;
 920
 921	if (nd->flags & LOOKUP_RCU)
 922		return -ECHILD;
 923
 924	audit_log_link_denied("follow_link", &nd->stack[0].link);
 925	return -EACCES;
 926}
 927
 928/**
 929 * safe_hardlink_source - Check for safe hardlink conditions
 930 * @inode: the source inode to hardlink from
 931 *
 932 * Return false if at least one of the following conditions:
 933 *    - inode is not a regular file
 934 *    - inode is setuid
 935 *    - inode is setgid and group-exec
 936 *    - access failure for read and write
 937 *
 938 * Otherwise returns true.
 939 */
 940static bool safe_hardlink_source(struct inode *inode)
 941{
 942	umode_t mode = inode->i_mode;
 943
 944	/* Special files should not get pinned to the filesystem. */
 945	if (!S_ISREG(mode))
 946		return false;
 947
 948	/* Setuid files should not get pinned to the filesystem. */
 949	if (mode & S_ISUID)
 950		return false;
 951
 952	/* Executable setgid files should not get pinned to the filesystem. */
 953	if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
 954		return false;
 955
 956	/* Hardlinking to unreadable or unwritable sources is dangerous. */
 957	if (inode_permission(inode, MAY_READ | MAY_WRITE))
 958		return false;
 959
 960	return true;
 961}
 962
 963/**
 964 * may_linkat - Check permissions for creating a hardlink
 965 * @link: the source to hardlink from
 966 *
 967 * Block hardlink when all of:
 968 *  - sysctl_protected_hardlinks enabled
 969 *  - fsuid does not match inode
 970 *  - hardlink source is unsafe (see safe_hardlink_source() above)
 971 *  - not CAP_FOWNER in a namespace with the inode owner uid mapped
 972 *
 973 * Returns 0 if successful, -ve on error.
 974 */
 975static int may_linkat(struct path *link)
 976{
 977	struct inode *inode;
 978
 979	if (!sysctl_protected_hardlinks)
 980		return 0;
 981
 982	inode = link->dentry->d_inode;
 983
 984	/* Source inode owner (or CAP_FOWNER) can hardlink all they like,
 985	 * otherwise, it must be a safe source.
 986	 */
 987	if (inode_owner_or_capable(inode) || safe_hardlink_source(inode))
 988		return 0;
 989
 990	audit_log_link_denied("linkat", link);
 991	return -EPERM;
 992}
 993
 994static __always_inline
 995const char *get_link(struct nameidata *nd)
 996{
 997	struct saved *last = nd->stack + nd->depth - 1;
 998	struct dentry *dentry = last->link.dentry;
 999	struct inode *inode = nd->link_inode;
1000	int error;
1001	const char *res;
1002
1003	if (!(nd->flags & LOOKUP_RCU)) {
1004		touch_atime(&last->link);
1005		cond_resched();
1006	} else if (atime_needs_update(&last->link, inode)) {
1007		if (unlikely(unlazy_walk(nd, NULL, 0)))
1008			return ERR_PTR(-ECHILD);
1009		touch_atime(&last->link);
1010	}
1011
1012	error = security_inode_follow_link(dentry, inode,
1013					   nd->flags & LOOKUP_RCU);
1014	if (unlikely(error))
1015		return ERR_PTR(error);
1016
1017	nd->last_type = LAST_BIND;
1018	res = inode->i_link;
1019	if (!res) {
1020		const char * (*get)(struct dentry *, struct inode *,
1021				struct delayed_call *);
1022		get = inode->i_op->get_link;
1023		if (nd->flags & LOOKUP_RCU) {
1024			res = get(NULL, inode, &last->done);
1025			if (res == ERR_PTR(-ECHILD)) {
1026				if (unlikely(unlazy_walk(nd, NULL, 0)))
1027					return ERR_PTR(-ECHILD);
1028				res = get(dentry, inode, &last->done);
1029			}
1030		} else {
1031			res = get(dentry, inode, &last->done);
1032		}
1033		if (IS_ERR_OR_NULL(res))
1034			return res;
1035	}
1036	if (*res == '/') {
1037		if (!nd->root.mnt)
1038			set_root(nd);
1039		if (unlikely(nd_jump_root(nd)))
1040			return ERR_PTR(-ECHILD);
1041		while (unlikely(*++res == '/'))
1042			;
1043	}
1044	if (!*res)
1045		res = NULL;
1046	return res;
1047}
1048
1049/*
1050 * follow_up - Find the mountpoint of path's vfsmount
1051 *
1052 * Given a path, find the mountpoint of its source file system.
1053 * Replace @path with the path of the mountpoint in the parent mount.
1054 * Up is towards /.
1055 *
1056 * Return 1 if we went up a level and 0 if we were already at the
1057 * root.
1058 */
1059int follow_up(struct path *path)
1060{
1061	struct mount *mnt = real_mount(path->mnt);
1062	struct mount *parent;
1063	struct dentry *mountpoint;
1064
1065	read_seqlock_excl(&mount_lock);
1066	parent = mnt->mnt_parent;
1067	if (parent == mnt) {
1068		read_sequnlock_excl(&mount_lock);
1069		return 0;
1070	}
1071	mntget(&parent->mnt);
1072	mountpoint = dget(mnt->mnt_mountpoint);
1073	read_sequnlock_excl(&mount_lock);
1074	dput(path->dentry);
1075	path->dentry = mountpoint;
1076	mntput(path->mnt);
1077	path->mnt = &parent->mnt;
1078	return 1;
1079}
1080EXPORT_SYMBOL(follow_up);
1081
1082/*
1083 * Perform an automount
1084 * - return -EISDIR to tell follow_managed() to stop and return the path we
1085 *   were called with.
1086 */
1087static int follow_automount(struct path *path, struct nameidata *nd,
1088			    bool *need_mntput)
1089{
1090	struct vfsmount *mnt;
1091	int err;
1092
1093	if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
1094		return -EREMOTE;
1095
1096	/* We don't want to mount if someone's just doing a stat -
1097	 * unless they're stat'ing a directory and appended a '/' to
1098	 * the name.
1099	 *
1100	 * We do, however, want to mount if someone wants to open or
1101	 * create a file of any type under the mountpoint, wants to
1102	 * traverse through the mountpoint or wants to open the
1103	 * mounted directory.  Also, autofs may mark negative dentries
1104	 * as being automount points.  These will need the attentions
1105	 * of the daemon to instantiate them before they can be used.
1106	 */
1107	if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1108			   LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1109	    path->dentry->d_inode)
1110		return -EISDIR;
1111
1112	nd->total_link_count++;
1113	if (nd->total_link_count >= 40)
1114		return -ELOOP;
1115
1116	mnt = path->dentry->d_op->d_automount(path);
1117	if (IS_ERR(mnt)) {
1118		/*
1119		 * The filesystem is allowed to return -EISDIR here to indicate
1120		 * it doesn't want to automount.  For instance, autofs would do
1121		 * this so that its userspace daemon can mount on this dentry.
1122		 *
1123		 * However, we can only permit this if it's a terminal point in
1124		 * the path being looked up; if it wasn't then the remainder of
1125		 * the path is inaccessible and we should say so.
1126		 */
1127		if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT))
1128			return -EREMOTE;
1129		return PTR_ERR(mnt);
1130	}
1131
1132	if (!mnt) /* mount collision */
1133		return 0;
1134
1135	if (!*need_mntput) {
1136		/* lock_mount() may release path->mnt on error */
1137		mntget(path->mnt);
1138		*need_mntput = true;
1139	}
1140	err = finish_automount(mnt, path);
1141
1142	switch (err) {
1143	case -EBUSY:
1144		/* Someone else made a mount here whilst we were busy */
1145		return 0;
1146	case 0:
1147		path_put(path);
1148		path->mnt = mnt;
1149		path->dentry = dget(mnt->mnt_root);
1150		return 0;
1151	default:
1152		return err;
1153	}
1154
1155}
1156
1157/*
1158 * Handle a dentry that is managed in some way.
1159 * - Flagged for transit management (autofs)
1160 * - Flagged as mountpoint
1161 * - Flagged as automount point
1162 *
1163 * This may only be called in refwalk mode.
1164 *
1165 * Serialization is taken care of in namespace.c
1166 */
1167static int follow_managed(struct path *path, struct nameidata *nd)
1168{
1169	struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1170	unsigned managed;
1171	bool need_mntput = false;
1172	int ret = 0;
1173
1174	/* Given that we're not holding a lock here, we retain the value in a
1175	 * local variable for each dentry as we look at it so that we don't see
1176	 * the components of that value change under us */
1177	while (managed = ACCESS_ONCE(path->dentry->d_flags),
1178	       managed &= DCACHE_MANAGED_DENTRY,
1179	       unlikely(managed != 0)) {
1180		/* Allow the filesystem to manage the transit without i_mutex
1181		 * being held. */
1182		if (managed & DCACHE_MANAGE_TRANSIT) {
1183			BUG_ON(!path->dentry->d_op);
1184			BUG_ON(!path->dentry->d_op->d_manage);
1185			ret = path->dentry->d_op->d_manage(path->dentry, false);
1186			if (ret < 0)
1187				break;
1188		}
1189
1190		/* Transit to a mounted filesystem. */
1191		if (managed & DCACHE_MOUNTED) {
1192			struct vfsmount *mounted = lookup_mnt(path);
1193			if (mounted) {
1194				dput(path->dentry);
1195				if (need_mntput)
1196					mntput(path->mnt);
1197				path->mnt = mounted;
1198				path->dentry = dget(mounted->mnt_root);
1199				need_mntput = true;
1200				continue;
1201			}
1202
1203			/* Something is mounted on this dentry in another
1204			 * namespace and/or whatever was mounted there in this
1205			 * namespace got unmounted before lookup_mnt() could
1206			 * get it */
1207		}
1208
1209		/* Handle an automount point */
1210		if (managed & DCACHE_NEED_AUTOMOUNT) {
1211			ret = follow_automount(path, nd, &need_mntput);
1212			if (ret < 0)
1213				break;
1214			continue;
1215		}
1216
1217		/* We didn't change the current path point */
1218		break;
1219	}
1220
1221	if (need_mntput && path->mnt == mnt)
1222		mntput(path->mnt);
1223	if (ret == -EISDIR || !ret)
1224		ret = 1;
1225	if (need_mntput)
1226		nd->flags |= LOOKUP_JUMPED;
1227	if (unlikely(ret < 0))
1228		path_put_conditional(path, nd);
1229	return ret;
1230}
1231
1232int follow_down_one(struct path *path)
1233{
1234	struct vfsmount *mounted;
1235
1236	mounted = lookup_mnt(path);
1237	if (mounted) {
1238		dput(path->dentry);
1239		mntput(path->mnt);
1240		path->mnt = mounted;
1241		path->dentry = dget(mounted->mnt_root);
1242		return 1;
1243	}
1244	return 0;
1245}
1246EXPORT_SYMBOL(follow_down_one);
1247
1248static inline int managed_dentry_rcu(struct dentry *dentry)
1249{
1250	return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1251		dentry->d_op->d_manage(dentry, true) : 0;
1252}
1253
1254/*
1255 * Try to skip to top of mountpoint pile in rcuwalk mode.  Fail if
1256 * we meet a managed dentry that would need blocking.
1257 */
1258static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1259			       struct inode **inode, unsigned *seqp)
1260{
1261	for (;;) {
1262		struct mount *mounted;
1263		/*
1264		 * Don't forget we might have a non-mountpoint managed dentry
1265		 * that wants to block transit.
1266		 */
1267		switch (managed_dentry_rcu(path->dentry)) {
1268		case -ECHILD:
1269		default:
1270			return false;
1271		case -EISDIR:
1272			return true;
1273		case 0:
1274			break;
1275		}
1276
1277		if (!d_mountpoint(path->dentry))
1278			return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1279
1280		mounted = __lookup_mnt(path->mnt, path->dentry);
1281		if (!mounted)
1282			break;
1283		path->mnt = &mounted->mnt;
1284		path->dentry = mounted->mnt.mnt_root;
1285		nd->flags |= LOOKUP_JUMPED;
1286		*seqp = read_seqcount_begin(&path->dentry->d_seq);
1287		/*
1288		 * Update the inode too. We don't need to re-check the
1289		 * dentry sequence number here after this d_inode read,
1290		 * because a mount-point is always pinned.
1291		 */
1292		*inode = path->dentry->d_inode;
1293	}
1294	return !read_seqretry(&mount_lock, nd->m_seq) &&
1295		!(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1296}
1297
1298static int follow_dotdot_rcu(struct nameidata *nd)
1299{
1300	struct inode *inode = nd->inode;
1301
1302	while (1) {
1303		if (path_equal(&nd->path, &nd->root))
1304			break;
1305		if (nd->path.dentry != nd->path.mnt->mnt_root) {
1306			struct dentry *old = nd->path.dentry;
1307			struct dentry *parent = old->d_parent;
1308			unsigned seq;
1309
1310			inode = parent->d_inode;
1311			seq = read_seqcount_begin(&parent->d_seq);
1312			if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq)))
1313				return -ECHILD;
1314			nd->path.dentry = parent;
1315			nd->seq = seq;
1316			if (unlikely(!path_connected(&nd->path)))
1317				return -ENOENT;
1318			break;
1319		} else {
1320			struct mount *mnt = real_mount(nd->path.mnt);
1321			struct mount *mparent = mnt->mnt_parent;
1322			struct dentry *mountpoint = mnt->mnt_mountpoint;
1323			struct inode *inode2 = mountpoint->d_inode;
1324			unsigned seq = read_seqcount_begin(&mountpoint->d_seq);
1325			if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1326				return -ECHILD;
1327			if (&mparent->mnt == nd->path.mnt)
1328				break;
1329			/* we know that mountpoint was pinned */
1330			nd->path.dentry = mountpoint;
1331			nd->path.mnt = &mparent->mnt;
1332			inode = inode2;
1333			nd->seq = seq;
1334		}
1335	}
1336	while (unlikely(d_mountpoint(nd->path.dentry))) {
1337		struct mount *mounted;
1338		mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1339		if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1340			return -ECHILD;
1341		if (!mounted)
1342			break;
1343		nd->path.mnt = &mounted->mnt;
1344		nd->path.dentry = mounted->mnt.mnt_root;
1345		inode = nd->path.dentry->d_inode;
1346		nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1347	}
1348	nd->inode = inode;
1349	return 0;
1350}
1351
1352/*
1353 * Follow down to the covering mount currently visible to userspace.  At each
1354 * point, the filesystem owning that dentry may be queried as to whether the
1355 * caller is permitted to proceed or not.
1356 */
1357int follow_down(struct path *path)
1358{
1359	unsigned managed;
1360	int ret;
1361
1362	while (managed = ACCESS_ONCE(path->dentry->d_flags),
1363	       unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1364		/* Allow the filesystem to manage the transit without i_mutex
1365		 * being held.
1366		 *
1367		 * We indicate to the filesystem if someone is trying to mount
1368		 * something here.  This gives autofs the chance to deny anyone
1369		 * other than its daemon the right to mount on its
1370		 * superstructure.
1371		 *
1372		 * The filesystem may sleep at this point.
1373		 */
1374		if (managed & DCACHE_MANAGE_TRANSIT) {
1375			BUG_ON(!path->dentry->d_op);
1376			BUG_ON(!path->dentry->d_op->d_manage);
1377			ret = path->dentry->d_op->d_manage(
1378				path->dentry, false);
1379			if (ret < 0)
1380				return ret == -EISDIR ? 0 : ret;
1381		}
1382
1383		/* Transit to a mounted filesystem. */
1384		if (managed & DCACHE_MOUNTED) {
1385			struct vfsmount *mounted = lookup_mnt(path);
1386			if (!mounted)
1387				break;
1388			dput(path->dentry);
1389			mntput(path->mnt);
1390			path->mnt = mounted;
1391			path->dentry = dget(mounted->mnt_root);
1392			continue;
1393		}
1394
1395		/* Don't handle automount points here */
1396		break;
1397	}
1398	return 0;
1399}
1400EXPORT_SYMBOL(follow_down);
1401
1402/*
1403 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1404 */
1405static void follow_mount(struct path *path)
1406{
1407	while (d_mountpoint(path->dentry)) {
1408		struct vfsmount *mounted = lookup_mnt(path);
1409		if (!mounted)
1410			break;
1411		dput(path->dentry);
1412		mntput(path->mnt);
1413		path->mnt = mounted;
1414		path->dentry = dget(mounted->mnt_root);
1415	}
1416}
1417
1418static int follow_dotdot(struct nameidata *nd)
1419{
1420	while(1) {
1421		struct dentry *old = nd->path.dentry;
1422
1423		if (nd->path.dentry == nd->root.dentry &&
1424		    nd->path.mnt == nd->root.mnt) {
1425			break;
1426		}
1427		if (nd->path.dentry != nd->path.mnt->mnt_root) {
1428			/* rare case of legitimate dget_parent()... */
1429			nd->path.dentry = dget_parent(nd->path.dentry);
1430			dput(old);
1431			if (unlikely(!path_connected(&nd->path)))
1432				return -ENOENT;
1433			break;
1434		}
1435		if (!follow_up(&nd->path))
1436			break;
1437	}
1438	follow_mount(&nd->path);
1439	nd->inode = nd->path.dentry->d_inode;
1440	return 0;
1441}
1442
1443/*
1444 * This looks up the name in dcache, possibly revalidates the old dentry and
1445 * allocates a new one if not found or not valid.  In the need_lookup argument
1446 * returns whether i_op->lookup is necessary.
1447 */
1448static struct dentry *lookup_dcache(const struct qstr *name,
1449				    struct dentry *dir,
1450				    unsigned int flags)
1451{
1452	struct dentry *dentry;
1453	int error;
1454
1455	dentry = d_lookup(dir, name);
1456	if (dentry) {
1457		if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1458			error = d_revalidate(dentry, flags);
1459			if (unlikely(error <= 0)) {
1460				if (!error)
1461					d_invalidate(dentry);
1462				dput(dentry);
1463				return ERR_PTR(error);
1464			}
1465		}
1466	}
1467	return dentry;
1468}
1469
1470/*
1471 * Call i_op->lookup on the dentry.  The dentry must be negative and
1472 * unhashed.
1473 *
1474 * dir->d_inode->i_mutex must be held
1475 */
1476static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1477				  unsigned int flags)
1478{
1479	struct dentry *old;
1480
1481	/* Don't create child dentry for a dead directory. */
1482	if (unlikely(IS_DEADDIR(dir))) {
1483		dput(dentry);
1484		return ERR_PTR(-ENOENT);
1485	}
1486
1487	old = dir->i_op->lookup(dir, dentry, flags);
1488	if (unlikely(old)) {
1489		dput(dentry);
1490		dentry = old;
1491	}
1492	return dentry;
1493}
1494
1495static struct dentry *__lookup_hash(const struct qstr *name,
1496		struct dentry *base, unsigned int flags)
1497{
1498	struct dentry *dentry = lookup_dcache(name, base, flags);
1499
1500	if (dentry)
1501		return dentry;
1502
1503	dentry = d_alloc(base, name);
1504	if (unlikely(!dentry))
1505		return ERR_PTR(-ENOMEM);
1506
1507	return lookup_real(base->d_inode, dentry, flags);
1508}
1509
1510static int lookup_fast(struct nameidata *nd,
1511		       struct path *path, struct inode **inode,
1512		       unsigned *seqp)
1513{
1514	struct vfsmount *mnt = nd->path.mnt;
1515	struct dentry *dentry, *parent = nd->path.dentry;
1516	int status = 1;
1517	int err;
1518
1519	/*
1520	 * Rename seqlock is not required here because in the off chance
1521	 * of a false negative due to a concurrent rename, the caller is
1522	 * going to fall back to non-racy lookup.
1523	 */
1524	if (nd->flags & LOOKUP_RCU) {
1525		unsigned seq;
1526		bool negative;
1527		dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1528		if (unlikely(!dentry)) {
1529			if (unlazy_walk(nd, NULL, 0))
1530				return -ECHILD;
1531			return 0;
1532		}
1533
1534		/*
1535		 * This sequence count validates that the inode matches
1536		 * the dentry name information from lookup.
1537		 */
1538		*inode = d_backing_inode(dentry);
1539		negative = d_is_negative(dentry);
1540		if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
1541			return -ECHILD;
1542
1543		/*
1544		 * This sequence count validates that the parent had no
1545		 * changes while we did the lookup of the dentry above.
1546		 *
1547		 * The memory barrier in read_seqcount_begin of child is
1548		 *  enough, we can use __read_seqcount_retry here.
1549		 */
1550		if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq)))
1551			return -ECHILD;
1552
1553		*seqp = seq;
1554		if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
1555			status = d_revalidate(dentry, nd->flags);
1556		if (unlikely(status <= 0)) {
1557			if (unlazy_walk(nd, dentry, seq))
1558				return -ECHILD;
1559			if (status == -ECHILD)
1560				status = d_revalidate(dentry, nd->flags);
1561		} else {
1562			/*
1563			 * Note: do negative dentry check after revalidation in
1564			 * case that drops it.
1565			 */
1566			if (unlikely(negative))
1567				return -ENOENT;
1568			path->mnt = mnt;
1569			path->dentry = dentry;
1570			if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
1571				return 1;
1572			if (unlazy_walk(nd, dentry, seq))
1573				return -ECHILD;
1574		}
1575	} else {
1576		dentry = __d_lookup(parent, &nd->last);
1577		if (unlikely(!dentry))
1578			return 0;
1579		if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
1580			status = d_revalidate(dentry, nd->flags);
1581	}
1582	if (unlikely(status <= 0)) {
1583		if (!status)
1584			d_invalidate(dentry);
1585		dput(dentry);
1586		return status;
1587	}
1588	if (unlikely(d_is_negative(dentry))) {
1589		dput(dentry);
1590		return -ENOENT;
1591	}
1592
1593	path->mnt = mnt;
1594	path->dentry = dentry;
1595	err = follow_managed(path, nd);
1596	if (likely(err > 0))
1597		*inode = d_backing_inode(path->dentry);
1598	return err;
1599}
1600
1601/* Fast lookup failed, do it the slow way */
1602static struct dentry *lookup_slow(const struct qstr *name,
1603				  struct dentry *dir,
1604				  unsigned int flags)
1605{
1606	struct dentry *dentry;
1607	inode_lock(dir->d_inode);
1608	dentry = d_lookup(dir, name);
1609	if (unlikely(dentry)) {
1610		if ((dentry->d_flags & DCACHE_OP_REVALIDATE) &&
1611		    !(flags & LOOKUP_NO_REVAL)) {
1612			int error = d_revalidate(dentry, flags);
1613			if (unlikely(error <= 0)) {
1614				if (!error)
1615					d_invalidate(dentry);
1616				dput(dentry);
1617				dentry = ERR_PTR(error);
1618			}
1619		}
1620		if (dentry) {
1621			inode_unlock(dir->d_inode);
1622			return dentry;
1623		}
1624	}
1625	dentry = d_alloc(dir, name);
1626	if (unlikely(!dentry)) {
1627		inode_unlock(dir->d_inode);
1628		return ERR_PTR(-ENOMEM);
1629	}
1630	dentry = lookup_real(dir->d_inode, dentry, flags);
1631	inode_unlock(dir->d_inode);
1632	return dentry;
1633}
1634
1635static inline int may_lookup(struct nameidata *nd)
1636{
1637	if (nd->flags & LOOKUP_RCU) {
1638		int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1639		if (err != -ECHILD)
1640			return err;
1641		if (unlazy_walk(nd, NULL, 0))
1642			return -ECHILD;
1643	}
1644	return inode_permission(nd->inode, MAY_EXEC);
1645}
1646
1647static inline int handle_dots(struct nameidata *nd, int type)
1648{
1649	if (type == LAST_DOTDOT) {
1650		if (!nd->root.mnt)
1651			set_root(nd);
1652		if (nd->flags & LOOKUP_RCU) {
1653			return follow_dotdot_rcu(nd);
1654		} else
1655			return follow_dotdot(nd);
1656	}
1657	return 0;
1658}
1659
1660static int pick_link(struct nameidata *nd, struct path *link,
1661		     struct inode *inode, unsigned seq)
1662{
1663	int error;
1664	struct saved *last;
1665	if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) {
1666		path_to_nameidata(link, nd);
1667		return -ELOOP;
1668	}
1669	if (!(nd->flags & LOOKUP_RCU)) {
1670		if (link->mnt == nd->path.mnt)
1671			mntget(link->mnt);
1672	}
1673	error = nd_alloc_stack(nd);
1674	if (unlikely(error)) {
1675		if (error == -ECHILD) {
1676			if (unlikely(unlazy_link(nd, link, seq)))
1677				return -ECHILD;
1678			error = nd_alloc_stack(nd);
1679		}
1680		if (error) {
1681			path_put(link);
1682			return error;
1683		}
1684	}
1685
1686	last = nd->stack + nd->depth++;
1687	last->link = *link;
1688	clear_delayed_call(&last->done);
1689	nd->link_inode = inode;
1690	last->seq = seq;
1691	return 1;
1692}
1693
1694/*
1695 * Do we need to follow links? We _really_ want to be able
1696 * to do this check without having to look at inode->i_op,
1697 * so we keep a cache of "no, this doesn't need follow_link"
1698 * for the common case.
1699 */
1700static inline int should_follow_link(struct nameidata *nd, struct path *link,
1701				     int follow,
1702				     struct inode *inode, unsigned seq)
1703{
1704	if (likely(!d_is_symlink(link->dentry)))
1705		return 0;
1706	if (!follow)
1707		return 0;
1708	/* make sure that d_is_symlink above matches inode */
1709	if (nd->flags & LOOKUP_RCU) {
1710		if (read_seqcount_retry(&link->dentry->d_seq, seq))
1711			return -ECHILD;
1712	}
1713	return pick_link(nd, link, inode, seq);
1714}
1715
1716enum {WALK_GET = 1, WALK_PUT = 2};
1717
1718static int walk_component(struct nameidata *nd, int flags)
1719{
1720	struct path path;
1721	struct inode *inode;
1722	unsigned seq;
1723	int err;
1724	/*
1725	 * "." and ".." are special - ".." especially so because it has
1726	 * to be able to know about the current root directory and
1727	 * parent relationships.
1728	 */
1729	if (unlikely(nd->last_type != LAST_NORM)) {
1730		err = handle_dots(nd, nd->last_type);
1731		if (flags & WALK_PUT)
1732			put_link(nd);
1733		return err;
1734	}
1735	err = lookup_fast(nd, &path, &inode, &seq);
1736	if (unlikely(err <= 0)) {
1737		if (err < 0)
1738			return err;
1739		path.dentry = lookup_slow(&nd->last, nd->path.dentry,
1740					  nd->flags);
1741		if (IS_ERR(path.dentry))
1742			return PTR_ERR(path.dentry);
1743
1744		path.mnt = nd->path.mnt;
1745		err = follow_managed(&path, nd);
1746		if (unlikely(err < 0))
1747			return err;
1748
1749		if (unlikely(d_is_negative(path.dentry))) {
1750			path_to_nameidata(&path, nd);
1751			return -ENOENT;
1752		}
1753
1754		seq = 0;	/* we are already out of RCU mode */
1755		inode = d_backing_inode(path.dentry);
1756	}
1757
1758	if (flags & WALK_PUT)
1759		put_link(nd);
1760	err = should_follow_link(nd, &path, flags & WALK_GET, inode, seq);
1761	if (unlikely(err))
1762		return err;
1763	path_to_nameidata(&path, nd);
1764	nd->inode = inode;
1765	nd->seq = seq;
1766	return 0;
1767}
1768
1769/*
1770 * We can do the critical dentry name comparison and hashing
1771 * operations one word at a time, but we are limited to:
1772 *
1773 * - Architectures with fast unaligned word accesses. We could
1774 *   do a "get_unaligned()" if this helps and is sufficiently
1775 *   fast.
1776 *
1777 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1778 *   do not trap on the (extremely unlikely) case of a page
1779 *   crossing operation.
1780 *
1781 * - Furthermore, we need an efficient 64-bit compile for the
1782 *   64-bit case in order to generate the "number of bytes in
1783 *   the final mask". Again, that could be replaced with a
1784 *   efficient population count instruction or similar.
1785 */
1786#ifdef CONFIG_DCACHE_WORD_ACCESS
1787
1788#include <asm/word-at-a-time.h>
1789
1790#ifdef CONFIG_64BIT
1791
1792static inline unsigned int fold_hash(unsigned long hash)
1793{
1794	return hash_64(hash, 32);
1795}
1796
1797#else	/* 32-bit case */
1798
1799#define fold_hash(x) (x)
1800
1801#endif
1802
1803unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1804{
1805	unsigned long a, mask;
1806	unsigned long hash = 0;
1807
1808	for (;;) {
1809		a = load_unaligned_zeropad(name);
1810		if (len < sizeof(unsigned long))
1811			break;
1812		hash += a;
1813		hash *= 9;
1814		name += sizeof(unsigned long);
1815		len -= sizeof(unsigned long);
1816		if (!len)
1817			goto done;
1818	}
1819	mask = bytemask_from_count(len);
1820	hash += mask & a;
1821done:
1822	return fold_hash(hash);
1823}
1824EXPORT_SYMBOL(full_name_hash);
1825
1826/*
1827 * Calculate the length and hash of the path component, and
1828 * return the "hash_len" as the result.
1829 */
1830static inline u64 hash_name(const char *name)
1831{
1832	unsigned long a, b, adata, bdata, mask, hash, len;
1833	const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1834
1835	hash = a = 0;
1836	len = -sizeof(unsigned long);
1837	do {
1838		hash = (hash + a) * 9;
1839		len += sizeof(unsigned long);
1840		a = load_unaligned_zeropad(name+len);
1841		b = a ^ REPEAT_BYTE('/');
1842	} while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1843
1844	adata = prep_zero_mask(a, adata, &constants);
1845	bdata = prep_zero_mask(b, bdata, &constants);
1846
1847	mask = create_zero_mask(adata | bdata);
1848
1849	hash += a & zero_bytemask(mask);
1850	len += find_zero(mask);
1851	return hashlen_create(fold_hash(hash), len);
1852}
1853
1854#else
1855
1856unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1857{
1858	unsigned long hash = init_name_hash();
1859	while (len--)
1860		hash = partial_name_hash(*name++, hash);
1861	return end_name_hash(hash);
1862}
1863EXPORT_SYMBOL(full_name_hash);
1864
1865/*
1866 * We know there's a real path component here of at least
1867 * one character.
1868 */
1869static inline u64 hash_name(const char *name)
1870{
1871	unsigned long hash = init_name_hash();
1872	unsigned long len = 0, c;
1873
1874	c = (unsigned char)*name;
1875	do {
1876		len++;
1877		hash = partial_name_hash(c, hash);
1878		c = (unsigned char)name[len];
1879	} while (c && c != '/');
1880	return hashlen_create(end_name_hash(hash), len);
1881}
1882
1883#endif
1884
1885/*
1886 * Name resolution.
1887 * This is the basic name resolution function, turning a pathname into
1888 * the final dentry. We expect 'base' to be positive and a directory.
1889 *
1890 * Returns 0 and nd will have valid dentry and mnt on success.
1891 * Returns error and drops reference to input namei data on failure.
1892 */
1893static int link_path_walk(const char *name, struct nameidata *nd)
1894{
1895	int err;
1896
1897	while (*name=='/')
1898		name++;
1899	if (!*name)
1900		return 0;
1901
1902	/* At this point we know we have a real path component. */
1903	for(;;) {
1904		u64 hash_len;
1905		int type;
1906
1907		err = may_lookup(nd);
1908 		if (err)
1909			return err;
1910
1911		hash_len = hash_name(name);
1912
1913		type = LAST_NORM;
1914		if (name[0] == '.') switch (hashlen_len(hash_len)) {
1915			case 2:
1916				if (name[1] == '.') {
1917					type = LAST_DOTDOT;
1918					nd->flags |= LOOKUP_JUMPED;
1919				}
1920				break;
1921			case 1:
1922				type = LAST_DOT;
1923		}
1924		if (likely(type == LAST_NORM)) {
1925			struct dentry *parent = nd->path.dentry;
1926			nd->flags &= ~LOOKUP_JUMPED;
1927			if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1928				struct qstr this = { { .hash_len = hash_len }, .name = name };
1929				err = parent->d_op->d_hash(parent, &this);
1930				if (err < 0)
1931					return err;
1932				hash_len = this.hash_len;
1933				name = this.name;
1934			}
1935		}
1936
1937		nd->last.hash_len = hash_len;
1938		nd->last.name = name;
1939		nd->last_type = type;
1940
1941		name += hashlen_len(hash_len);
1942		if (!*name)
1943			goto OK;
1944		/*
1945		 * If it wasn't NUL, we know it was '/'. Skip that
1946		 * slash, and continue until no more slashes.
1947		 */
1948		do {
1949			name++;
1950		} while (unlikely(*name == '/'));
1951		if (unlikely(!*name)) {
1952OK:
1953			/* pathname body, done */
1954			if (!nd->depth)
1955				return 0;
1956			name = nd->stack[nd->depth - 1].name;
1957			/* trailing symlink, done */
1958			if (!name)
1959				return 0;
1960			/* last component of nested symlink */
1961			err = walk_component(nd, WALK_GET | WALK_PUT);
1962		} else {
1963			err = walk_component(nd, WALK_GET);
1964		}
1965		if (err < 0)
1966			return err;
1967
1968		if (err) {
1969			const char *s = get_link(nd);
1970
1971			if (IS_ERR(s))
1972				return PTR_ERR(s);
1973			err = 0;
1974			if (unlikely(!s)) {
1975				/* jumped */
1976				put_link(nd);
1977			} else {
1978				nd->stack[nd->depth - 1].name = name;
1979				name = s;
1980				continue;
1981			}
1982		}
1983		if (unlikely(!d_can_lookup(nd->path.dentry))) {
1984			if (nd->flags & LOOKUP_RCU) {
1985				if (unlazy_walk(nd, NULL, 0))
1986					return -ECHILD;
1987			}
1988			return -ENOTDIR;
1989		}
1990	}
1991}
1992
1993static const char *path_init(struct nameidata *nd, unsigned flags)
1994{
1995	int retval = 0;
1996	const char *s = nd->name->name;
1997
1998	nd->last_type = LAST_ROOT; /* if there are only slashes... */
1999	nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
2000	nd->depth = 0;
2001	if (flags & LOOKUP_ROOT) {
2002		struct dentry *root = nd->root.dentry;
2003		struct inode *inode = root->d_inode;
2004		if (*s) {
2005			if (!d_can_lookup(root))
2006				return ERR_PTR(-ENOTDIR);
2007			retval = inode_permission(inode, MAY_EXEC);
2008			if (retval)
2009				return ERR_PTR(retval);
2010		}
2011		nd->path = nd->root;
2012		nd->inode = inode;
2013		if (flags & LOOKUP_RCU) {
2014			rcu_read_lock();
2015			nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2016			nd->root_seq = nd->seq;
2017			nd->m_seq = read_seqbegin(&mount_lock);
2018		} else {
2019			path_get(&nd->path);
2020		}
2021		return s;
2022	}
2023
2024	nd->root.mnt = NULL;
2025	nd->path.mnt = NULL;
2026	nd->path.dentry = NULL;
2027
2028	nd->m_seq = read_seqbegin(&mount_lock);
2029	if (*s == '/') {
2030		if (flags & LOOKUP_RCU)
2031			rcu_read_lock();
2032		set_root(nd);
2033		if (likely(!nd_jump_root(nd)))
2034			return s;
2035		nd->root.mnt = NULL;
2036		rcu_read_unlock();
2037		return ERR_PTR(-ECHILD);
2038	} else if (nd->dfd == AT_FDCWD) {
2039		if (flags & LOOKUP_RCU) {
2040			struct fs_struct *fs = current->fs;
2041			unsigned seq;
2042
2043			rcu_read_lock();
2044
2045			do {
2046				seq = read_seqcount_begin(&fs->seq);
2047				nd->path = fs->pwd;
2048				nd->inode = nd->path.dentry->d_inode;
2049				nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2050			} while (read_seqcount_retry(&fs->seq, seq));
2051		} else {
2052			get_fs_pwd(current->fs, &nd->path);
2053			nd->inode = nd->path.dentry->d_inode;
2054		}
2055		return s;
2056	} else {
2057		/* Caller must check execute permissions on the starting path component */
2058		struct fd f = fdget_raw(nd->dfd);
2059		struct dentry *dentry;
2060
2061		if (!f.file)
2062			return ERR_PTR(-EBADF);
2063
2064		dentry = f.file->f_path.dentry;
2065
2066		if (*s) {
2067			if (!d_can_lookup(dentry)) {
2068				fdput(f);
2069				return ERR_PTR(-ENOTDIR);
2070			}
2071		}
2072
2073		nd->path = f.file->f_path;
2074		if (flags & LOOKUP_RCU) {
2075			rcu_read_lock();
2076			nd->inode = nd->path.dentry->d_inode;
2077			nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2078		} else {
2079			path_get(&nd->path);
2080			nd->inode = nd->path.dentry->d_inode;
2081		}
2082		fdput(f);
2083		return s;
2084	}
2085}
2086
2087static const char *trailing_symlink(struct nameidata *nd)
2088{
2089	const char *s;
2090	int error = may_follow_link(nd);
2091	if (unlikely(error))
2092		return ERR_PTR(error);
2093	nd->flags |= LOOKUP_PARENT;
2094	nd->stack[0].name = NULL;
2095	s = get_link(nd);
2096	return s ? s : "";
2097}
2098
2099static inline int lookup_last(struct nameidata *nd)
2100{
2101	if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2102		nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2103
2104	nd->flags &= ~LOOKUP_PARENT;
2105	return walk_component(nd,
2106			nd->flags & LOOKUP_FOLLOW
2107				? nd->depth
2108					? WALK_PUT | WALK_GET
2109					: WALK_GET
2110				: 0);
2111}
2112
2113/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2114static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2115{
2116	const char *s = path_init(nd, flags);
2117	int err;
2118
2119	if (IS_ERR(s))
2120		return PTR_ERR(s);
2121	while (!(err = link_path_walk(s, nd))
2122		&& ((err = lookup_last(nd)) > 0)) {
2123		s = trailing_symlink(nd);
2124		if (IS_ERR(s)) {
2125			err = PTR_ERR(s);
2126			break;
2127		}
2128	}
2129	if (!err)
2130		err = complete_walk(nd);
2131
2132	if (!err && nd->flags & LOOKUP_DIRECTORY)
2133		if (!d_can_lookup(nd->path.dentry))
2134			err = -ENOTDIR;
2135	if (!err) {
2136		*path = nd->path;
2137		nd->path.mnt = NULL;
2138		nd->path.dentry = NULL;
2139	}
2140	terminate_walk(nd);
2141	return err;
2142}
2143
2144static int filename_lookup(int dfd, struct filename *name, unsigned flags,
2145			   struct path *path, struct path *root)
2146{
2147	int retval;
2148	struct nameidata nd;
2149	if (IS_ERR(name))
2150		return PTR_ERR(name);
2151	if (unlikely(root)) {
2152		nd.root = *root;
2153		flags |= LOOKUP_ROOT;
2154	}
2155	set_nameidata(&nd, dfd, name);
2156	retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
2157	if (unlikely(retval == -ECHILD))
2158		retval = path_lookupat(&nd, flags, path);
2159	if (unlikely(retval == -ESTALE))
2160		retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
2161
2162	if (likely(!retval))
2163		audit_inode(name, path->dentry, flags & LOOKUP_PARENT);
2164	restore_nameidata();
2165	putname(name);
2166	return retval;
2167}
2168
2169/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2170static int path_parentat(struct nameidata *nd, unsigned flags,
2171				struct path *parent)
2172{
2173	const char *s = path_init(nd, flags);
2174	int err;
2175	if (IS_ERR(s))
2176		return PTR_ERR(s);
2177	err = link_path_walk(s, nd);
2178	if (!err)
2179		err = complete_walk(nd);
2180	if (!err) {
2181		*parent = nd->path;
2182		nd->path.mnt = NULL;
2183		nd->path.dentry = NULL;
2184	}
2185	terminate_walk(nd);
2186	return err;
2187}
2188
2189static struct filename *filename_parentat(int dfd, struct filename *name,
2190				unsigned int flags, struct path *parent,
2191				struct qstr *last, int *type)
2192{
2193	int retval;
2194	struct nameidata nd;
2195
2196	if (IS_ERR(name))
2197		return name;
2198	set_nameidata(&nd, dfd, name);
2199	retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
2200	if (unlikely(retval == -ECHILD))
2201		retval = path_parentat(&nd, flags, parent);
2202	if (unlikely(retval == -ESTALE))
2203		retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
2204	if (likely(!retval)) {
2205		*last = nd.last;
2206		*type = nd.last_type;
2207		audit_inode(name, parent->dentry, LOOKUP_PARENT);
2208	} else {
2209		putname(name);
2210		name = ERR_PTR(retval);
2211	}
2212	restore_nameidata();
2213	return name;
2214}
2215
2216/* does lookup, returns the object with parent locked */
2217struct dentry *kern_path_locked(const char *name, struct path *path)
2218{
2219	struct filename *filename;
2220	struct dentry *d;
2221	struct qstr last;
2222	int type;
2223
2224	filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path,
2225				    &last, &type);
2226	if (IS_ERR(filename))
2227		return ERR_CAST(filename);
2228	if (unlikely(type != LAST_NORM)) {
2229		path_put(path);
2230		putname(filename);
2231		return ERR_PTR(-EINVAL);
2232	}
2233	inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
2234	d = __lookup_hash(&last, path->dentry, 0);
2235	if (IS_ERR(d)) {
2236		inode_unlock(path->dentry->d_inode);
2237		path_put(path);
2238	}
2239	putname(filename);
2240	return d;
2241}
2242
2243int kern_path(const char *name, unsigned int flags, struct path *path)
2244{
2245	return filename_lookup(AT_FDCWD, getname_kernel(name),
2246			       flags, path, NULL);
2247}
2248EXPORT_SYMBOL(kern_path);
2249
2250/**
2251 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2252 * @dentry:  pointer to dentry of the base directory
2253 * @mnt: pointer to vfs mount of the base directory
2254 * @name: pointer to file name
2255 * @flags: lookup flags
2256 * @path: pointer to struct path to fill
2257 */
2258int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2259		    const char *name, unsigned int flags,
2260		    struct path *path)
2261{
2262	struct path root = {.mnt = mnt, .dentry = dentry};
2263	/* the first argument of filename_lookup() is ignored with root */
2264	return filename_lookup(AT_FDCWD, getname_kernel(name),
2265			       flags , path, &root);
2266}
2267EXPORT_SYMBOL(vfs_path_lookup);
2268
2269/**
2270 * lookup_hash - lookup single pathname component on already hashed name
2271 * @name:	name and hash to lookup
2272 * @base:	base directory to lookup from
2273 *
2274 * The name must have been verified and hashed (see lookup_one_len()).  Using
2275 * this after just full_name_hash() is unsafe.
2276 *
2277 * This function also doesn't check for search permission on base directory.
2278 *
2279 * Use lookup_one_len_unlocked() instead, unless you really know what you are
2280 * doing.
2281 *
2282 * Do not hold i_mutex; this helper takes i_mutex if necessary.
2283 */
2284struct dentry *lookup_hash(const struct qstr *name, struct dentry *base)
2285{
2286	struct dentry *ret;
2287
2288	ret = lookup_dcache(name, base, 0);
2289	if (!ret)
2290		ret = lookup_slow(name, base, 0);
2291
2292	return ret;
2293}
2294EXPORT_SYMBOL(lookup_hash);
2295
2296/**
2297 * lookup_one_len - filesystem helper to lookup single pathname component
2298 * @name:	pathname component to lookup
2299 * @base:	base directory to lookup from
2300 * @len:	maximum length @len should be interpreted to
2301 *
2302 * Note that this routine is purely a helper for filesystem usage and should
2303 * not be called by generic code.
2304 *
2305 * The caller must hold base->i_mutex.
2306 */
2307struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2308{
2309	struct qstr this;
2310	unsigned int c;
2311	int err;
2312
2313	WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2314
2315	this.name = name;
2316	this.len = len;
2317	this.hash = full_name_hash(name, len);
2318	if (!len)
2319		return ERR_PTR(-EACCES);
2320
2321	if (unlikely(name[0] == '.')) {
2322		if (len < 2 || (len == 2 && name[1] == '.'))
2323			return ERR_PTR(-EACCES);
2324	}
2325
2326	while (len--) {
2327		c = *(const unsigned char *)name++;
2328		if (c == '/' || c == '\0')
2329			return ERR_PTR(-EACCES);
2330	}
2331	/*
2332	 * See if the low-level filesystem might want
2333	 * to use its own hash..
2334	 */
2335	if (base->d_flags & DCACHE_OP_HASH) {
2336		int err = base->d_op->d_hash(base, &this);
2337		if (err < 0)
2338			return ERR_PTR(err);
2339	}
2340
2341	err = inode_permission(base->d_inode, MAY_EXEC);
2342	if (err)
2343		return ERR_PTR(err);
2344
2345	return __lookup_hash(&this, base, 0);
2346}
2347EXPORT_SYMBOL(lookup_one_len);
2348
2349/**
2350 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2351 * @name:	pathname component to lookup
2352 * @base:	base directory to lookup from
2353 * @len:	maximum length @len should be interpreted to
2354 *
2355 * Note that this routine is purely a helper for filesystem usage and should
2356 * not be called by generic code.
2357 *
2358 * Unlike lookup_one_len, it should be called without the parent
2359 * i_mutex held, and will take the i_mutex itself if necessary.
2360 */
2361struct dentry *lookup_one_len_unlocked(const char *name,
2362				       struct dentry *base, int len)
2363{
2364	struct qstr this;
2365	unsigned int c;
2366	int err;
2367
2368	this.name = name;
2369	this.len = len;
2370	this.hash = full_name_hash(name, len);
2371	if (!len)
2372		return ERR_PTR(-EACCES);
2373
2374	if (unlikely(name[0] == '.')) {
2375		if (len < 2 || (len == 2 && name[1] == '.'))
2376			return ERR_PTR(-EACCES);
2377	}
2378
2379	while (len--) {
2380		c = *(const unsigned char *)name++;
2381		if (c == '/' || c == '\0')
2382			return ERR_PTR(-EACCES);
2383	}
2384	/*
2385	 * See if the low-level filesystem might want
2386	 * to use its own hash..
2387	 */
2388	if (base->d_flags & DCACHE_OP_HASH) {
2389		int err = base->d_op->d_hash(base, &this);
2390		if (err < 0)
2391			return ERR_PTR(err);
2392	}
2393
2394	err = inode_permission(base->d_inode, MAY_EXEC);
2395	if (err)
2396		return ERR_PTR(err);
2397
2398	return lookup_hash(&this, base);
2399}
2400EXPORT_SYMBOL(lookup_one_len_unlocked);
2401
2402int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2403		 struct path *path, int *empty)
2404{
2405	return filename_lookup(dfd, getname_flags(name, flags, empty),
2406			       flags, path, NULL);
2407}
2408EXPORT_SYMBOL(user_path_at_empty);
2409
2410/*
2411 * NB: most callers don't do anything directly with the reference to the
2412 *     to struct filename, but the nd->last pointer points into the name string
2413 *     allocated by getname. So we must hold the reference to it until all
2414 *     path-walking is complete.
2415 */
2416static inline struct filename *
2417user_path_parent(int dfd, const char __user *path,
2418		 struct path *parent,
2419		 struct qstr *last,
2420		 int *type,
2421		 unsigned int flags)
2422{
2423	/* only LOOKUP_REVAL is allowed in extra flags */
2424	return filename_parentat(dfd, getname(path), flags & LOOKUP_REVAL,
2425				 parent, last, type);
2426}
2427
2428/**
2429 * mountpoint_last - look up last component for umount
2430 * @nd:   pathwalk nameidata - currently pointing at parent directory of "last"
2431 * @path: pointer to container for result
2432 *
2433 * This is a special lookup_last function just for umount. In this case, we
2434 * need to resolve the path without doing any revalidation.
2435 *
2436 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2437 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2438 * in almost all cases, this lookup will be served out of the dcache. The only
2439 * cases where it won't are if nd->last refers to a symlink or the path is
2440 * bogus and it doesn't exist.
2441 *
2442 * Returns:
2443 * -error: if there was an error during lookup. This includes -ENOENT if the
2444 *         lookup found a negative dentry. The nd->path reference will also be
2445 *         put in this case.
2446 *
2447 * 0:      if we successfully resolved nd->path and found it to not to be a
2448 *         symlink that needs to be followed. "path" will also be populated.
2449 *         The nd->path reference will also be put.
2450 *
2451 * 1:      if we successfully resolved nd->last and found it to be a symlink
2452 *         that needs to be followed. "path" will be populated with the path
2453 *         to the link, and nd->path will *not* be put.
2454 */
2455static int
2456mountpoint_last(struct nameidata *nd, struct path *path)
2457{
2458	int error = 0;
2459	struct dentry *dentry;
2460	struct dentry *dir = nd->path.dentry;
2461
2462	/* If we're in rcuwalk, drop out of it to handle last component */
2463	if (nd->flags & LOOKUP_RCU) {
2464		if (unlazy_walk(nd, NULL, 0))
2465			return -ECHILD;
2466	}
2467
2468	nd->flags &= ~LOOKUP_PARENT;
2469
2470	if (unlikely(nd->last_type != LAST_NORM)) {
2471		error = handle_dots(nd, nd->last_type);
2472		if (error)
2473			return error;
2474		dentry = dget(nd->path.dentry);
2475	} else {
2476		dentry = d_lookup(dir, &nd->last);
2477		if (!dentry) {
2478			/*
2479			 * No cached dentry. Mounted dentries are pinned in the
2480			 * cache, so that means that this dentry is probably
2481			 * a symlink or the path doesn't actually point
2482			 * to a mounted dentry.
2483			 */
2484			dentry = lookup_slow(&nd->last, dir,
2485					     nd->flags | LOOKUP_NO_REVAL);
2486			if (IS_ERR(dentry))
2487				return PTR_ERR(dentry);
2488		}
2489	}
2490	if (d_is_negative(dentry)) {
2491		dput(dentry);
2492		return -ENOENT;
2493	}
2494	if (nd->depth)
2495		put_link(nd);
2496	path->dentry = dentry;
2497	path->mnt = nd->path.mnt;
2498	error = should_follow_link(nd, path, nd->flags & LOOKUP_FOLLOW,
2499				   d_backing_inode(dentry), 0);
2500	if (unlikely(error))
2501		return error;
2502	mntget(path->mnt);
2503	follow_mount(path);
2504	return 0;
2505}
2506
2507/**
2508 * path_mountpoint - look up a path to be umounted
2509 * @nd:		lookup context
2510 * @flags:	lookup flags
2511 * @path:	pointer to container for result
2512 *
2513 * Look up the given name, but don't attempt to revalidate the last component.
2514 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2515 */
2516static int
2517path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path)
2518{
2519	const char *s = path_init(nd, flags);
2520	int err;
2521	if (IS_ERR(s))
2522		return PTR_ERR(s);
2523	while (!(err = link_path_walk(s, nd)) &&
2524		(err = mountpoint_last(nd, path)) > 0) {
2525		s = trailing_symlink(nd);
2526		if (IS_ERR(s)) {
2527			err = PTR_ERR(s);
2528			break;
2529		}
2530	}
2531	terminate_walk(nd);
2532	return err;
2533}
2534
2535static int
2536filename_mountpoint(int dfd, struct filename *name, struct path *path,
2537			unsigned int flags)
2538{
2539	struct nameidata nd;
2540	int error;
2541	if (IS_ERR(name))
2542		return PTR_ERR(name);
2543	set_nameidata(&nd, dfd, name);
2544	error = path_mountpoint(&nd, flags | LOOKUP_RCU, path);
2545	if (unlikely(error == -ECHILD))
2546		error = path_mountpoint(&nd, flags, path);
2547	if (unlikely(error == -ESTALE))
2548		error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path);
2549	if (likely(!error))
2550		audit_inode(name, path->dentry, 0);
2551	restore_nameidata();
2552	putname(name);
2553	return error;
2554}
2555
2556/**
2557 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2558 * @dfd:	directory file descriptor
2559 * @name:	pathname from userland
2560 * @flags:	lookup flags
2561 * @path:	pointer to container to hold result
2562 *
2563 * A umount is a special case for path walking. We're not actually interested
2564 * in the inode in this situation, and ESTALE errors can be a problem. We
2565 * simply want track down the dentry and vfsmount attached at the mountpoint
2566 * and avoid revalidating the last component.
2567 *
2568 * Returns 0 and populates "path" on success.
2569 */
2570int
2571user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2572			struct path *path)
2573{
2574	return filename_mountpoint(dfd, getname(name), path, flags);
2575}
2576
2577int
2578kern_path_mountpoint(int dfd, const char *name, struct path *path,
2579			unsigned int flags)
2580{
2581	return filename_mountpoint(dfd, getname_kernel(name), path, flags);
2582}
2583EXPORT_SYMBOL(kern_path_mountpoint);
2584
2585int __check_sticky(struct inode *dir, struct inode *inode)
2586{
2587	kuid_t fsuid = current_fsuid();
2588
2589	if (uid_eq(inode->i_uid, fsuid))
2590		return 0;
2591	if (uid_eq(dir->i_uid, fsuid))
2592		return 0;
2593	return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2594}
2595EXPORT_SYMBOL(__check_sticky);
2596
2597/*
2598 *	Check whether we can remove a link victim from directory dir, check
2599 *  whether the type of victim is right.
2600 *  1. We can't do it if dir is read-only (done in permission())
2601 *  2. We should have write and exec permissions on dir
2602 *  3. We can't remove anything from append-only dir
2603 *  4. We can't do anything with immutable dir (done in permission())
2604 *  5. If the sticky bit on dir is set we should either
2605 *	a. be owner of dir, or
2606 *	b. be owner of victim, or
2607 *	c. have CAP_FOWNER capability
2608 *  6. If the victim is append-only or immutable we can't do antyhing with
2609 *     links pointing to it.
2610 *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2611 *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2612 *  9. We can't remove a root or mountpoint.
2613 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2614 *     nfs_async_unlink().
2615 */
2616static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2617{
2618	struct inode *inode = d_backing_inode(victim);
2619	int error;
2620
2621	if (d_is_negative(victim))
2622		return -ENOENT;
2623	BUG_ON(!inode);
2624
2625	BUG_ON(victim->d_parent->d_inode != dir);
2626	audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2627
2628	error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2629	if (error)
2630		return error;
2631	if (IS_APPEND(dir))
2632		return -EPERM;
2633
2634	if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2635	    IS_IMMUTABLE(inode) || IS_SWAPFILE(inode))
2636		return -EPERM;
2637	if (isdir) {
2638		if (!d_is_dir(victim))
2639			return -ENOTDIR;
2640		if (IS_ROOT(victim))
2641			return -EBUSY;
2642	} else if (d_is_dir(victim))
2643		return -EISDIR;
2644	if (IS_DEADDIR(dir))
2645		return -ENOENT;
2646	if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2647		return -EBUSY;
2648	return 0;
2649}
2650
2651/*	Check whether we can create an object with dentry child in directory
2652 *  dir.
2653 *  1. We can't do it if child already exists (open has special treatment for
2654 *     this case, but since we are inlined it's OK)
2655 *  2. We can't do it if dir is read-only (done in permission())
2656 *  3. We should have write and exec permissions on dir
2657 *  4. We can't do it if dir is immutable (done in permission())
2658 */
2659static inline int may_create(struct inode *dir, struct dentry *child)
2660{
2661	audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2662	if (child->d_inode)
2663		return -EEXIST;
2664	if (IS_DEADDIR(dir))
2665		return -ENOENT;
2666	return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2667}
2668
2669/*
2670 * p1 and p2 should be directories on the same fs.
2671 */
2672struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2673{
2674	struct dentry *p;
2675
2676	if (p1 == p2) {
2677		inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2678		return NULL;
2679	}
2680
2681	mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2682
2683	p = d_ancestor(p2, p1);
2684	if (p) {
2685		inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
2686		inode_lock_nested(p1->d_inode, I_MUTEX_CHILD);
2687		return p;
2688	}
2689
2690	p = d_ancestor(p1, p2);
2691	if (p) {
2692		inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2693		inode_lock_nested(p2->d_inode, I_MUTEX_CHILD);
2694		return p;
2695	}
2696
2697	inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2698	inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
2699	return NULL;
2700}
2701EXPORT_SYMBOL(lock_rename);
2702
2703void unlock_rename(struct dentry *p1, struct dentry *p2)
2704{
2705	inode_unlock(p1->d_inode);
2706	if (p1 != p2) {
2707		inode_unlock(p2->d_inode);
2708		mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2709	}
2710}
2711EXPORT_SYMBOL(unlock_rename);
2712
2713int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2714		bool want_excl)
2715{
2716	int error = may_create(dir, dentry);
2717	if (error)
2718		return error;
2719
2720	if (!dir->i_op->create)
2721		return -EACCES;	/* shouldn't it be ENOSYS? */
2722	mode &= S_IALLUGO;
2723	mode |= S_IFREG;
2724	error = security_inode_create(dir, dentry, mode);
2725	if (error)
2726		return error;
2727	error = dir->i_op->create(dir, dentry, mode, want_excl);
2728	if (!error)
2729		fsnotify_create(dir, dentry);
2730	return error;
2731}
2732EXPORT_SYMBOL(vfs_create);
2733
2734static int may_open(struct path *path, int acc_mode, int flag)
2735{
2736	struct dentry *dentry = path->dentry;
2737	struct inode *inode = dentry->d_inode;
2738	int error;
2739
2740	if (!inode)
2741		return -ENOENT;
2742
2743	switch (inode->i_mode & S_IFMT) {
2744	case S_IFLNK:
2745		return -ELOOP;
2746	case S_IFDIR:
2747		if (acc_mode & MAY_WRITE)
2748			return -EISDIR;
2749		break;
2750	case S_IFBLK:
2751	case S_IFCHR:
2752		if (path->mnt->mnt_flags & MNT_NODEV)
2753			return -EACCES;
2754		/*FALLTHRU*/
2755	case S_IFIFO:
2756	case S_IFSOCK:
2757		flag &= ~O_TRUNC;
2758		break;
2759	}
2760
2761	error = inode_permission(inode, MAY_OPEN | acc_mode);
2762	if (error)
2763		return error;
2764
2765	/*
2766	 * An append-only file must be opened in append mode for writing.
2767	 */
2768	if (IS_APPEND(inode)) {
2769		if  ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2770			return -EPERM;
2771		if (flag & O_TRUNC)
2772			return -EPERM;
2773	}
2774
2775	/* O_NOATIME can only be set by the owner or superuser */
2776	if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2777		return -EPERM;
2778
2779	return 0;
2780}
2781
2782static int handle_truncate(struct file *filp)
2783{
2784	struct path *path = &filp->f_path;
2785	struct inode *inode = path->dentry->d_inode;
2786	int error = get_write_access(inode);
2787	if (error)
2788		return error;
2789	/*
2790	 * Refuse to truncate files with mandatory locks held on them.
2791	 */
2792	error = locks_verify_locked(filp);
2793	if (!error)
2794		error = security_path_truncate(path);
2795	if (!error) {
2796		error = do_truncate(path->dentry, 0,
2797				    ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2798				    filp);
2799	}
2800	put_write_access(inode);
2801	return error;
2802}
2803
2804static inline int open_to_namei_flags(int flag)
2805{
2806	if ((flag & O_ACCMODE) == 3)
2807		flag--;
2808	return flag;
2809}
2810
2811static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2812{
2813	int error = security_path_mknod(dir, dentry, mode, 0);
2814	if (error)
2815		return error;
2816
2817	error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2818	if (error)
2819		return error;
2820
2821	return security_inode_create(dir->dentry->d_inode, dentry, mode);
2822}
2823
2824/*
2825 * Attempt to atomically look up, create and open a file from a negative
2826 * dentry.
2827 *
2828 * Returns 0 if successful.  The file will have been created and attached to
2829 * @file by the filesystem calling finish_open().
2830 *
2831 * Returns 1 if the file was looked up only or didn't need creating.  The
2832 * caller will need to perform the open themselves.  @path will have been
2833 * updated to point to the new dentry.  This may be negative.
2834 *
2835 * Returns an error code otherwise.
2836 */
2837static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2838			struct path *path, struct file *file,
2839			const struct open_flags *op,
2840			bool got_write, bool need_lookup,
2841			int *opened)
2842{
2843	struct inode *dir =  nd->path.dentry->d_inode;
2844	unsigned open_flag = open_to_namei_flags(op->open_flag);
2845	umode_t mode;
2846	int error;
2847	int acc_mode;
2848	int create_error = 0;
2849	struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2850	bool excl;
2851
2852	BUG_ON(dentry->d_inode);
2853
2854	/* Don't create child dentry for a dead directory. */
2855	if (unlikely(IS_DEADDIR(dir))) {
2856		error = -ENOENT;
2857		goto out;
2858	}
2859
2860	mode = op->mode;
2861	if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2862		mode &= ~current_umask();
2863
2864	excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT);
2865	if (excl)
2866		open_flag &= ~O_TRUNC;
2867
2868	/*
2869	 * Checking write permission is tricky, bacuse we don't know if we are
2870	 * going to actually need it: O_CREAT opens should work as long as the
2871	 * file exists.  But checking existence breaks atomicity.  The trick is
2872	 * to check access and if not granted clear O_CREAT from the flags.
2873	 *
2874	 * Another problem is returing the "right" error value (e.g. for an
2875	 * O_EXCL open we want to return EEXIST not EROFS).
2876	 */
2877	if (((open_flag & (O_CREAT | O_TRUNC)) ||
2878	    (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
2879		if (!(open_flag & O_CREAT)) {
2880			/*
2881			 * No O_CREATE -> atomicity not a requirement -> fall
2882			 * back to lookup + open
2883			 */
2884			goto no_open;
2885		} else if (open_flag & (O_EXCL | O_TRUNC)) {
2886			/* Fall back and fail with the right error */
2887			create_error = -EROFS;
2888			goto no_open;
2889		} else {
2890			/* No side effects, safe to clear O_CREAT */
2891			create_error = -EROFS;
2892			open_flag &= ~O_CREAT;
2893		}
2894	}
2895
2896	if (open_flag & O_CREAT) {
2897		error = may_o_create(&nd->path, dentry, mode);
2898		if (error) {
2899			create_error = error;
2900			if (open_flag & O_EXCL)
2901				goto no_open;
2902			open_flag &= ~O_CREAT;
2903		}
2904	}
2905
2906	if (nd->flags & LOOKUP_DIRECTORY)
2907		open_flag |= O_DIRECTORY;
2908
2909	file->f_path.dentry = DENTRY_NOT_SET;
2910	file->f_path.mnt = nd->path.mnt;
2911	error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
2912				      opened);
2913	if (error < 0) {
2914		if (create_error && error == -ENOENT)
2915			error = create_error;
2916		goto out;
2917	}
2918
2919	if (error) {	/* returned 1, that is */
2920		if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
2921			error = -EIO;
2922			goto out;
2923		}
2924		if (file->f_path.dentry) {
2925			dput(dentry);
2926			dentry = file->f_path.dentry;
2927		}
2928		if (*opened & FILE_CREATED)
2929			fsnotify_create(dir, dentry);
2930		if (!dentry->d_inode) {
2931			WARN_ON(*opened & FILE_CREATED);
2932			if (create_error) {
2933				error = create_error;
2934				goto out;
2935			}
2936		} else {
2937			if (excl && !(*opened & FILE_CREATED)) {
2938				error = -EEXIST;
2939				goto out;
2940			}
2941		}
2942		goto looked_up;
2943	}
2944
2945	/*
2946	 * We didn't have the inode before the open, so check open permission
2947	 * here.
2948	 */
2949	acc_mode = op->acc_mode;
2950	if (*opened & FILE_CREATED) {
2951		WARN_ON(!(open_flag & O_CREAT));
2952		fsnotify_create(dir, dentry);
2953		acc_mode = 0;
2954	}
2955	error = may_open(&file->f_path, acc_mode, open_flag);
2956	if (error)
2957		fput(file);
2958
2959out:
2960	dput(dentry);
2961	return error;
2962
2963no_open:
2964	if (need_lookup) {
2965		dentry = lookup_real(dir, dentry, nd->flags);
2966		if (IS_ERR(dentry))
2967			return PTR_ERR(dentry);
2968	}
2969	if (create_error && !dentry->d_inode) {
2970		error = create_error;
2971		goto out;
2972	}
2973looked_up:
2974	path->dentry = dentry;
2975	path->mnt = nd->path.mnt;
2976	return 1;
2977}
2978
2979/*
2980 * Look up and maybe create and open the last component.
2981 *
2982 * Must be called with i_mutex held on parent.
2983 *
2984 * Returns 0 if the file was successfully atomically created (if necessary) and
2985 * opened.  In this case the file will be returned attached to @file.
2986 *
2987 * Returns 1 if the file was not completely opened at this time, though lookups
2988 * and creations will have been performed and the dentry returned in @path will
2989 * be positive upon return if O_CREAT was specified.  If O_CREAT wasn't
2990 * specified then a negative dentry may be returned.
2991 *
2992 * An error code is returned otherwise.
2993 *
2994 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2995 * cleared otherwise prior to returning.
2996 */
2997static int lookup_open(struct nameidata *nd, struct path *path,
2998			struct file *file,
2999			const struct open_flags *op,
3000			bool got_write, int *opened)
3001{
3002	struct dentry *dir = nd->path.dentry;
3003	struct inode *dir_inode = dir->d_inode;
3004	struct dentry *dentry;
3005	int error;
3006	bool need_lookup = false;
3007
3008	*opened &= ~FILE_CREATED;
3009	dentry = lookup_dcache(&nd->last, dir, nd->flags);
3010	if (IS_ERR(dentry))
3011		return PTR_ERR(dentry);
3012
3013	if (!dentry) {
3014		dentry = d_alloc(dir, &nd->last);
3015		if (unlikely(!dentry))
3016			return -ENOMEM;
3017		need_lookup = true;
3018	} else if (dentry->d_inode) {
3019		/* Cached positive dentry: will open in f_op->open */
3020		goto out_no_open;
3021	}
3022
3023	if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
3024		return atomic_open(nd, dentry, path, file, op, got_write,
3025				   need_lookup, opened);
3026	}
3027
3028	if (need_lookup) {
3029		BUG_ON(dentry->d_inode);
3030
3031		dentry = lookup_real(dir_inode, dentry, nd->flags);
3032		if (IS_ERR(dentry))
3033			return PTR_ERR(dentry);
3034	}
3035
3036	/* Negative dentry, just create the file */
3037	if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
3038		umode_t mode = op->mode;
3039		if (!IS_POSIXACL(dir->d_inode))
3040			mode &= ~current_umask();
3041		/*
3042		 * This write is needed to ensure that a
3043		 * rw->ro transition does not occur between
3044		 * the time when the file is created and when
3045		 * a permanent write count is taken through
3046		 * the 'struct file' in finish_open().
3047		 */
3048		if (!got_write) {
3049			error = -EROFS;
3050			goto out_dput;
3051		}
3052		*opened |= FILE_CREATED;
3053		error = security_path_mknod(&nd->path, dentry, mode, 0);
3054		if (error)
3055			goto out_dput;
3056		error = vfs_create(dir->d_inode, dentry, mode,
3057				   nd->flags & LOOKUP_EXCL);
3058		if (error)
3059			goto out_dput;
3060	}
3061out_no_open:
3062	path->dentry = dentry;
3063	path->mnt = nd->path.mnt;
3064	return 1;
3065
3066out_dput:
3067	dput(dentry);
3068	return error;
3069}
3070
3071/*
3072 * Handle the last step of open()
3073 */
3074static int do_last(struct nameidata *nd,
3075		   struct file *file, const struct open_flags *op,
3076		   int *opened)
3077{
3078	struct dentry *dir = nd->path.dentry;
3079	int open_flag = op->open_flag;
3080	bool will_truncate = (open_flag & O_TRUNC) != 0;
3081	bool got_write = false;
3082	int acc_mode = op->acc_mode;
3083	unsigned seq;
3084	struct inode *inode;
3085	struct path save_parent = { .dentry = NULL, .mnt = NULL };
3086	struct path path;
3087	bool retried = false;
3088	int error;
3089
3090	nd->flags &= ~LOOKUP_PARENT;
3091	nd->flags |= op->intent;
3092
3093	if (nd->last_type != LAST_NORM) {
3094		error = handle_dots(nd, nd->last_type);
3095		if (unlikely(error))
3096			return error;
3097		goto finish_open;
3098	}
3099
3100	if (!(open_flag & O_CREAT)) {
3101		if (nd->last.name[nd->last.len])
3102			nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3103		/* we _can_ be in RCU mode here */
3104		error = lookup_fast(nd, &path, &inode, &seq);
3105		if (likely(error > 0))
3106			goto finish_lookup;
3107
3108		if (error < 0)
3109			return error;
3110
3111		BUG_ON(nd->inode != dir->d_inode);
3112		BUG_ON(nd->flags & LOOKUP_RCU);
3113	} else {
3114		/* create side of things */
3115		/*
3116		 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3117		 * has been cleared when we got to the last component we are
3118		 * about to look up
3119		 */
3120		error = complete_walk(nd);
3121		if (error)
3122			return error;
3123
3124		audit_inode(nd->name, dir, LOOKUP_PARENT);
3125		/* trailing slashes? */
3126		if (unlikely(nd->last.name[nd->last.len]))
3127			return -EISDIR;
3128	}
3129
3130retry_lookup:
3131	if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3132		error = mnt_want_write(nd->path.mnt);
3133		if (!error)
3134			got_write = true;
3135		/*
3136		 * do _not_ fail yet - we might not need that or fail with
3137		 * a different error; let lookup_open() decide; we'll be
3138		 * dropping this one anyway.
3139		 */
3140	}
3141	inode_lock(dir->d_inode);
3142	error = lookup_open(nd, &path, file, op, got_write, opened);
3143	inode_unlock(dir->d_inode);
3144
3145	if (error <= 0) {
3146		if (error)
3147			goto out;
3148
3149		if ((*opened & FILE_CREATED) ||
3150		    !S_ISREG(file_inode(file)->i_mode))
3151			will_truncate = false;
3152
3153		audit_inode(nd->name, file->f_path.dentry, 0);
3154		goto opened;
3155	}
3156
3157	if (*opened & FILE_CREATED) {
3158		/* Don't check for write permission, don't truncate */
3159		open_flag &= ~O_TRUNC;
3160		will_truncate = false;
3161		acc_mode = 0;
3162		path_to_nameidata(&path, nd);
3163		goto finish_open_created;
3164	}
3165
3166	/*
3167	 * If atomic_open() acquired write access it is dropped now due to
3168	 * possible mount and symlink following (this might be optimized away if
3169	 * necessary...)
3170	 */
3171	if (got_write) {
3172		mnt_drop_write(nd->path.mnt);
3173		got_write = false;
3174	}
3175
3176	if (unlikely(d_is_negative(path.dentry))) {
3177		path_to_nameidata(&path, nd);
3178		return -ENOENT;
3179	}
3180
3181	/*
3182	 * create/update audit record if it already exists.
3183	 */
3184	audit_inode(nd->name, path.dentry, 0);
3185
3186	if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) {
3187		path_to_nameidata(&path, nd);
3188		return -EEXIST;
3189	}
3190
3191	error = follow_managed(&path, nd);
3192	if (unlikely(error < 0))
3193		return error;
3194
3195	seq = 0;	/* out of RCU mode, so the value doesn't matter */
3196	inode = d_backing_inode(path.dentry);
3197finish_lookup:
3198	if (nd->depth)
3199		put_link(nd);
3200	error = should_follow_link(nd, &path, nd->flags & LOOKUP_FOLLOW,
3201				   inode, seq);
3202	if (unlikely(error))
3203		return error;
3204
3205	if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path.mnt) {
3206		path_to_nameidata(&path, nd);
3207	} else {
3208		save_parent.dentry = nd->path.dentry;
3209		save_parent.mnt = mntget(path.mnt);
3210		nd->path.dentry = path.dentry;
3211
3212	}
3213	nd->inode = inode;
3214	nd->seq = seq;
3215	/* Why this, you ask?  _Now_ we might have grown LOOKUP_JUMPED... */
3216finish_open:
3217	error = complete_walk(nd);
3218	if (error) {
3219		path_put(&save_parent);
3220		return error;
3221	}
3222	audit_inode(nd->name, nd->path.dentry, 0);
3223	if (unlikely(d_is_symlink(nd->path.dentry)) && !(open_flag & O_PATH)) {
3224		error = -ELOOP;
3225		goto out;
3226	}
3227	error = -EISDIR;
3228	if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
3229		goto out;
3230	error = -ENOTDIR;
3231	if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3232		goto out;
3233	if (!d_is_reg(nd->path.dentry))
3234		will_truncate = false;
3235
3236	if (will_truncate) {
3237		error = mnt_want_write(nd->path.mnt);
3238		if (error)
3239			goto out;
3240		got_write = true;
3241	}
3242finish_open_created:
3243	if (likely(!(open_flag & O_PATH))) {
3244		error = may_open(&nd->path, acc_mode, open_flag);
3245		if (error)
3246			goto out;
3247	}
3248	BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
3249	error = vfs_open(&nd->path, file, current_cred());
3250	if (!error) {
3251		*opened |= FILE_OPENED;
3252	} else {
3253		if (error == -EOPENSTALE)
3254			goto stale_open;
3255		goto out;
3256	}
3257opened:
3258	error = open_check_o_direct(file);
3259	if (error)
3260		goto exit_fput;
3261	error = ima_file_check(file, op->acc_mode, *opened);
3262	if (error)
3263		goto exit_fput;
3264
3265	if (will_truncate) {
3266		error = handle_truncate(file);
3267		if (error)
3268			goto exit_fput;
3269	}
3270out:
3271	if (unlikely(error > 0)) {
3272		WARN_ON(1);
3273		error = -EINVAL;
3274	}
3275	if (got_write)
3276		mnt_drop_write(nd->path.mnt);
3277	path_put(&save_parent);
3278	return error;
3279
3280exit_fput:
3281	fput(file);
3282	goto out;
3283
3284stale_open:
3285	/* If no saved parent or already retried then can't retry */
3286	if (!save_parent.dentry || retried)
3287		goto out;
3288
3289	BUG_ON(save_parent.dentry != dir);
3290	path_put(&nd->path);
3291	nd->path = save_parent;
3292	nd->inode = dir->d_inode;
3293	save_parent.mnt = NULL;
3294	save_parent.dentry = NULL;
3295	if (got_write) {
3296		mnt_drop_write(nd->path.mnt);
3297		got_write = false;
3298	}
3299	retried = true;
3300	goto retry_lookup;
3301}
3302
3303static int do_tmpfile(struct nameidata *nd, unsigned flags,
3304		const struct open_flags *op,
3305		struct file *file, int *opened)
3306{
3307	static const struct qstr name = QSTR_INIT("/", 1);
3308	struct dentry *child;
3309	struct inode *dir;
3310	struct path path;
3311	int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
3312	if (unlikely(error))
3313		return error;
3314	error = mnt_want_write(path.mnt);
3315	if (unlikely(error))
3316		goto out;
3317	dir = path.dentry->d_inode;
3318	/* we want directory to be writable */
3319	error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
3320	if (error)
3321		goto out2;
3322	if (!dir->i_op->tmpfile) {
3323		error = -EOPNOTSUPP;
3324		goto out2;
3325	}
3326	child = d_alloc(path.dentry, &name);
3327	if (unlikely(!child)) {
3328		error = -ENOMEM;
3329		goto out2;
3330	}
3331	dput(path.dentry);
3332	path.dentry = child;
3333	error = dir->i_op->tmpfile(dir, child, op->mode);
3334	if (error)
3335		goto out2;
3336	audit_inode(nd->name, child, 0);
3337	/* Don't check for other permissions, the inode was just created */
3338	error = may_open(&path, 0, op->open_flag);
3339	if (error)
3340		goto out2;
3341	file->f_path.mnt = path.mnt;
3342	error = finish_open(file, child, NULL, opened);
3343	if (error)
3344		goto out2;
3345	error = open_check_o_direct(file);
3346	if (error) {
3347		fput(file);
3348	} else if (!(op->open_flag & O_EXCL)) {
3349		struct inode *inode = file_inode(file);
3350		spin_lock(&inode->i_lock);
3351		inode->i_state |= I_LINKABLE;
3352		spin_unlock(&inode->i_lock);
3353	}
3354out2:
3355	mnt_drop_write(path.mnt);
3356out:
3357	path_put(&path);
3358	return error;
3359}
3360
3361static struct file *path_openat(struct nameidata *nd,
3362			const struct open_flags *op, unsigned flags)
3363{
3364	const char *s;
3365	struct file *file;
3366	int opened = 0;
3367	int error;
3368
3369	file = get_empty_filp();
3370	if (IS_ERR(file))
3371		return file;
3372
3373	file->f_flags = op->open_flag;
3374
3375	if (unlikely(file->f_flags & __O_TMPFILE)) {
3376		error = do_tmpfile(nd, flags, op, file, &opened);
3377		goto out2;
3378	}
3379
3380	s = path_init(nd, flags);
3381	if (IS_ERR(s)) {
3382		put_filp(file);
3383		return ERR_CAST(s);
3384	}
3385	while (!(error = link_path_walk(s, nd)) &&
3386		(error = do_last(nd, file, op, &opened)) > 0) {
3387		nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3388		s = trailing_symlink(nd);
3389		if (IS_ERR(s)) {
3390			error = PTR_ERR(s);
3391			break;
3392		}
3393	}
3394	terminate_walk(nd);
3395out2:
3396	if (!(opened & FILE_OPENED)) {
3397		BUG_ON(!error);
3398		put_filp(file);
3399	}
3400	if (unlikely(error)) {
3401		if (error == -EOPENSTALE) {
3402			if (flags & LOOKUP_RCU)
3403				error = -ECHILD;
3404			else
3405				error = -ESTALE;
3406		}
3407		file = ERR_PTR(error);
3408	}
3409	return file;
3410}
3411
3412struct file *do_filp_open(int dfd, struct filename *pathname,
3413		const struct open_flags *op)
3414{
3415	struct nameidata nd;
3416	int flags = op->lookup_flags;
3417	struct file *filp;
3418
3419	set_nameidata(&nd, dfd, pathname);
3420	filp = path_openat(&nd, op, flags | LOOKUP_RCU);
3421	if (unlikely(filp == ERR_PTR(-ECHILD)))
3422		filp = path_openat(&nd, op, flags);
3423	if (unlikely(filp == ERR_PTR(-ESTALE)))
3424		filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
3425	restore_nameidata();
3426	return filp;
3427}
3428
3429struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3430		const char *name, const struct open_flags *op)
3431{
3432	struct nameidata nd;
3433	struct file *file;
3434	struct filename *filename;
3435	int flags = op->lookup_flags | LOOKUP_ROOT;
3436
3437	nd.root.mnt = mnt;
3438	nd.root.dentry = dentry;
3439
3440	if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3441		return ERR_PTR(-ELOOP);
3442
3443	filename = getname_kernel(name);
3444	if (IS_ERR(filename))
3445		return ERR_CAST(filename);
3446
3447	set_nameidata(&nd, -1, filename);
3448	file = path_openat(&nd, op, flags | LOOKUP_RCU);
3449	if (unlikely(file == ERR_PTR(-ECHILD)))
3450		file = path_openat(&nd, op, flags);
3451	if (unlikely(file == ERR_PTR(-ESTALE)))
3452		file = path_openat(&nd, op, flags | LOOKUP_REVAL);
3453	restore_nameidata();
3454	putname(filename);
3455	return file;
3456}
3457
3458static struct dentry *filename_create(int dfd, struct filename *name,
3459				struct path *path, unsigned int lookup_flags)
3460{
3461	struct dentry *dentry = ERR_PTR(-EEXIST);
3462	struct qstr last;
3463	int type;
3464	int err2;
3465	int error;
3466	bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3467
3468	/*
3469	 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3470	 * other flags passed in are ignored!
3471	 */
3472	lookup_flags &= LOOKUP_REVAL;
3473
3474	name = filename_parentat(dfd, name, lookup_flags, path, &last, &type);
3475	if (IS_ERR(name))
3476		return ERR_CAST(name);
3477
3478	/*
3479	 * Yucky last component or no last component at all?
3480	 * (foo/., foo/.., /////)
3481	 */
3482	if (unlikely(type != LAST_NORM))
3483		goto out;
3484
3485	/* don't fail immediately if it's r/o, at least try to report other errors */
3486	err2 = mnt_want_write(path->mnt);
3487	/*
3488	 * Do the final lookup.
3489	 */
3490	lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3491	inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
3492	dentry = __lookup_hash(&last, path->dentry, lookup_flags);
3493	if (IS_ERR(dentry))
3494		goto unlock;
3495
3496	error = -EEXIST;
3497	if (d_is_positive(dentry))
3498		goto fail;
3499
3500	/*
3501	 * Special case - lookup gave negative, but... we had foo/bar/
3502	 * From the vfs_mknod() POV we just have a negative dentry -
3503	 * all is fine. Let's be bastards - you had / on the end, you've
3504	 * been asking for (non-existent) directory. -ENOENT for you.
3505	 */
3506	if (unlikely(!is_dir && last.name[last.len])) {
3507		error = -ENOENT;
3508		goto fail;
3509	}
3510	if (unlikely(err2)) {
3511		error = err2;
3512		goto fail;
3513	}
3514	putname(name);
3515	return dentry;
3516fail:
3517	dput(dentry);
3518	dentry = ERR_PTR(error);
3519unlock:
3520	inode_unlock(path->dentry->d_inode);
3521	if (!err2)
3522		mnt_drop_write(path->mnt);
3523out:
3524	path_put(path);
3525	putname(name);
3526	return dentry;
3527}
3528
3529struct dentry *kern_path_create(int dfd, const char *pathname,
3530				struct path *path, unsigned int lookup_flags)
3531{
3532	return filename_create(dfd, getname_kernel(pathname),
3533				path, lookup_flags);
3534}
3535EXPORT_SYMBOL(kern_path_create);
3536
3537void done_path_create(struct path *path, struct dentry *dentry)
3538{
3539	dput(dentry);
3540	inode_unlock(path->dentry->d_inode);
3541	mnt_drop_write(path->mnt);
3542	path_put(path);
3543}
3544EXPORT_SYMBOL(done_path_create);
3545
3546inline struct dentry *user_path_create(int dfd, const char __user *pathname,
3547				struct path *path, unsigned int lookup_flags)
3548{
3549	return filename_create(dfd, getname(pathname), path, lookup_flags);
3550}
3551EXPORT_SYMBOL(user_path_create);
3552
3553int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3554{
3555	int error = may_create(dir, dentry);
3556
3557	if (error)
3558		return error;
3559
3560	if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3561		return -EPERM;
3562
3563	if (!dir->i_op->mknod)
3564		return -EPERM;
3565
3566	error = devcgroup_inode_mknod(mode, dev);
3567	if (error)
3568		return error;
3569
3570	error = security_inode_mknod(dir, dentry, mode, dev);
3571	if (error)
3572		return error;
3573
3574	error = dir->i_op->mknod(dir, dentry, mode, dev);
3575	if (!error)
3576		fsnotify_create(dir, dentry);
3577	return error;
3578}
3579EXPORT_SYMBOL(vfs_mknod);
3580
3581static int may_mknod(umode_t mode)
3582{
3583	switch (mode & S_IFMT) {
3584	case S_IFREG:
3585	case S_IFCHR:
3586	case S_IFBLK:
3587	case S_IFIFO:
3588	case S_IFSOCK:
3589	case 0: /* zero mode translates to S_IFREG */
3590		return 0;
3591	case S_IFDIR:
3592		return -EPERM;
3593	default:
3594		return -EINVAL;
3595	}
3596}
3597
3598SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3599		unsigned, dev)
3600{
3601	struct dentry *dentry;
3602	struct path path;
3603	int error;
3604	unsigned int lookup_flags = 0;
3605
3606	error = may_mknod(mode);
3607	if (error)
3608		return error;
3609retry:
3610	dentry = user_path_create(dfd, filename, &path, lookup_flags);
3611	if (IS_ERR(dentry))
3612		return PTR_ERR(dentry);
3613
3614	if (!IS_POSIXACL(path.dentry->d_inode))
3615		mode &= ~current_umask();
3616	error = security_path_mknod(&path, dentry, mode, dev);
3617	if (error)
3618		goto out;
3619	switch (mode & S_IFMT) {
3620		case 0: case S_IFREG:
3621			error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3622			break;
3623		case S_IFCHR: case S_IFBLK:
3624			error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3625					new_decode_dev(dev));
3626			break;
3627		case S_IFIFO: case S_IFSOCK:
3628			error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3629			break;
3630	}
3631out:
3632	done_path_create(&path, dentry);
3633	if (retry_estale(error, lookup_flags)) {
3634		lookup_flags |= LOOKUP_REVAL;
3635		goto retry;
3636	}
3637	return error;
3638}
3639
3640SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3641{
3642	return sys_mknodat(AT_FDCWD, filename, mode, dev);
3643}
3644
3645int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3646{
3647	int error = may_create(dir, dentry);
3648	unsigned max_links = dir->i_sb->s_max_links;
3649
3650	if (error)
3651		return error;
3652
3653	if (!dir->i_op->mkdir)
3654		return -EPERM;
3655
3656	mode &= (S_IRWXUGO|S_ISVTX);
3657	error = security_inode_mkdir(dir, dentry, mode);
3658	if (error)
3659		return error;
3660
3661	if (max_links && dir->i_nlink >= max_links)
3662		return -EMLINK;
3663
3664	error = dir->i_op->mkdir(dir, dentry, mode);
3665	if (!error)
3666		fsnotify_mkdir(dir, dentry);
3667	return error;
3668}
3669EXPORT_SYMBOL(vfs_mkdir);
3670
3671SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3672{
3673	struct dentry *dentry;
3674	struct path path;
3675	int error;
3676	unsigned int lookup_flags = LOOKUP_DIRECTORY;
3677
3678retry:
3679	dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3680	if (IS_ERR(dentry))
3681		return PTR_ERR(dentry);
3682
3683	if (!IS_POSIXACL(path.dentry->d_inode))
3684		mode &= ~current_umask();
3685	error = security_path_mkdir(&path, dentry, mode);
3686	if (!error)
3687		error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3688	done_path_create(&path, dentry);
3689	if (retry_estale(error, lookup_flags)) {
3690		lookup_flags |= LOOKUP_REVAL;
3691		goto retry;
3692	}
3693	return error;
3694}
3695
3696SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3697{
3698	return sys_mkdirat(AT_FDCWD, pathname, mode);
3699}
3700
3701int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3702{
3703	int error = may_delete(dir, dentry, 1);
3704
3705	if (error)
3706		return error;
3707
3708	if (!dir->i_op->rmdir)
3709		return -EPERM;
3710
3711	dget(dentry);
3712	inode_lock(dentry->d_inode);
3713
3714	error = -EBUSY;
3715	if (is_local_mountpoint(dentry))
3716		goto out;
3717
3718	error = security_inode_rmdir(dir, dentry);
3719	if (error)
3720		goto out;
3721
3722	shrink_dcache_parent(dentry);
3723	error = dir->i_op->rmdir(dir, dentry);
3724	if (error)
3725		goto out;
3726
3727	dentry->d_inode->i_flags |= S_DEAD;
3728	dont_mount(dentry);
3729	detach_mounts(dentry);
3730
3731out:
3732	inode_unlock(dentry->d_inode);
3733	dput(dentry);
3734	if (!error)
3735		d_delete(dentry);
3736	return error;
3737}
3738EXPORT_SYMBOL(vfs_rmdir);
3739
3740static long do_rmdir(int dfd, const char __user *pathname)
3741{
3742	int error = 0;
3743	struct filename *name;
3744	struct dentry *dentry;
3745	struct path path;
3746	struct qstr last;
3747	int type;
3748	unsigned int lookup_flags = 0;
3749retry:
3750	name = user_path_parent(dfd, pathname,
3751				&path, &last, &type, lookup_flags);
3752	if (IS_ERR(name))
3753		return PTR_ERR(name);
3754
3755	switch (type) {
3756	case LAST_DOTDOT:
3757		error = -ENOTEMPTY;
3758		goto exit1;
3759	case LAST_DOT:
3760		error = -EINVAL;
3761		goto exit1;
3762	case LAST_ROOT:
3763		error = -EBUSY;
3764		goto exit1;
3765	}
3766
3767	error = mnt_want_write(path.mnt);
3768	if (error)
3769		goto exit1;
3770
3771	inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
3772	dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3773	error = PTR_ERR(dentry);
3774	if (IS_ERR(dentry))
3775		goto exit2;
3776	if (!dentry->d_inode) {
3777		error = -ENOENT;
3778		goto exit3;
3779	}
3780	error = security_path_rmdir(&path, dentry);
3781	if (error)
3782		goto exit3;
3783	error = vfs_rmdir(path.dentry->d_inode, dentry);
3784exit3:
3785	dput(dentry);
3786exit2:
3787	inode_unlock(path.dentry->d_inode);
3788	mnt_drop_write(path.mnt);
3789exit1:
3790	path_put(&path);
3791	putname(name);
3792	if (retry_estale(error, lookup_flags)) {
3793		lookup_flags |= LOOKUP_REVAL;
3794		goto retry;
3795	}
3796	return error;
3797}
3798
3799SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3800{
3801	return do_rmdir(AT_FDCWD, pathname);
3802}
3803
3804/**
3805 * vfs_unlink - unlink a filesystem object
3806 * @dir:	parent directory
3807 * @dentry:	victim
3808 * @delegated_inode: returns victim inode, if the inode is delegated.
3809 *
3810 * The caller must hold dir->i_mutex.
3811 *
3812 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3813 * return a reference to the inode in delegated_inode.  The caller
3814 * should then break the delegation on that inode and retry.  Because
3815 * breaking a delegation may take a long time, the caller should drop
3816 * dir->i_mutex before doing so.
3817 *
3818 * Alternatively, a caller may pass NULL for delegated_inode.  This may
3819 * be appropriate for callers that expect the underlying filesystem not
3820 * to be NFS exported.
3821 */
3822int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3823{
3824	struct inode *target = dentry->d_inode;
3825	int error = may_delete(dir, dentry, 0);
3826
3827	if (error)
3828		return error;
3829
3830	if (!dir->i_op->unlink)
3831		return -EPERM;
3832
3833	inode_lock(target);
3834	if (is_local_mountpoint(dentry))
3835		error = -EBUSY;
3836	else {
3837		error = security_inode_unlink(dir, dentry);
3838		if (!error) {
3839			error = try_break_deleg(target, delegated_inode);
3840			if (error)
3841				goto out;
3842			error = dir->i_op->unlink(dir, dentry);
3843			if (!error) {
3844				dont_mount(dentry);
3845				detach_mounts(dentry);
3846			}
3847		}
3848	}
3849out:
3850	inode_unlock(target);
3851
3852	/* We don't d_delete() NFS sillyrenamed files--they still exist. */
3853	if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3854		fsnotify_link_count(target);
3855		d_delete(dentry);
3856	}
3857
3858	return error;
3859}
3860EXPORT_SYMBOL(vfs_unlink);
3861
3862/*
3863 * Make sure that the actual truncation of the file will occur outside its
3864 * directory's i_mutex.  Truncate can take a long time if there is a lot of
3865 * writeout happening, and we don't want to prevent access to the directory
3866 * while waiting on the I/O.
3867 */
3868static long do_unlinkat(int dfd, const char __user *pathname)
3869{
3870	int error;
3871	struct filename *name;
3872	struct dentry *dentry;
3873	struct path path;
3874	struct qstr last;
3875	int type;
3876	struct inode *inode = NULL;
3877	struct inode *delegated_inode = NULL;
3878	unsigned int lookup_flags = 0;
3879retry:
3880	name = user_path_parent(dfd, pathname,
3881				&path, &last, &type, lookup_flags);
3882	if (IS_ERR(name))
3883		return PTR_ERR(name);
3884
3885	error = -EISDIR;
3886	if (type != LAST_NORM)
3887		goto exit1;
3888
3889	error = mnt_want_write(path.mnt);
3890	if (error)
3891		goto exit1;
3892retry_deleg:
3893	inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
3894	dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3895	error = PTR_ERR(dentry);
3896	if (!IS_ERR(dentry)) {
3897		/* Why not before? Because we want correct error value */
3898		if (last.name[last.len])
3899			goto slashes;
3900		inode = dentry->d_inode;
3901		if (d_is_negative(dentry))
3902			goto slashes;
3903		ihold(inode);
3904		error = security_path_unlink(&path, dentry);
3905		if (error)
3906			goto exit2;
3907		error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
3908exit2:
3909		dput(dentry);
3910	}
3911	inode_unlock(path.dentry->d_inode);
3912	if (inode)
3913		iput(inode);	/* truncate the inode here */
3914	inode = NULL;
3915	if (delegated_inode) {
3916		error = break_deleg_wait(&delegated_inode);
3917		if (!error)
3918			goto retry_deleg;
3919	}
3920	mnt_drop_write(path.mnt);
3921exit1:
3922	path_put(&path);
3923	putname(name);
3924	if (retry_estale(error, lookup_flags)) {
3925		lookup_flags |= LOOKUP_REVAL;
3926		inode = NULL;
3927		goto retry;
3928	}
3929	return error;
3930
3931slashes:
3932	if (d_is_negative(dentry))
3933		error = -ENOENT;
3934	else if (d_is_dir(dentry))
3935		error = -EISDIR;
3936	else
3937		error = -ENOTDIR;
3938	goto exit2;
3939}
3940
3941SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
3942{
3943	if ((flag & ~AT_REMOVEDIR) != 0)
3944		return -EINVAL;
3945
3946	if (flag & AT_REMOVEDIR)
3947		return do_rmdir(dfd, pathname);
3948
3949	return do_unlinkat(dfd, pathname);
3950}
3951
3952SYSCALL_DEFINE1(unlink, const char __user *, pathname)
3953{
3954	return do_unlinkat(AT_FDCWD, pathname);
3955}
3956
3957int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
3958{
3959	int error = may_create(dir, dentry);
3960
3961	if (error)
3962		return error;
3963
3964	if (!dir->i_op->symlink)
3965		return -EPERM;
3966
3967	error = security_inode_symlink(dir, dentry, oldname);
3968	if (error)
3969		return error;
3970
3971	error = dir->i_op->symlink(dir, dentry, oldname);
3972	if (!error)
3973		fsnotify_create(dir, dentry);
3974	return error;
3975}
3976EXPORT_SYMBOL(vfs_symlink);
3977
3978SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3979		int, newdfd, const char __user *, newname)
3980{
3981	int error;
3982	struct filename *from;
3983	struct dentry *dentry;
3984	struct path path;
3985	unsigned int lookup_flags = 0;
3986
3987	from = getname(oldname);
3988	if (IS_ERR(from))
3989		return PTR_ERR(from);
3990retry:
3991	dentry = user_path_create(newdfd, newname, &path, lookup_flags);
3992	error = PTR_ERR(dentry);
3993	if (IS_ERR(dentry))
3994		goto out_putname;
3995
3996	error = security_path_symlink(&path, dentry, from->name);
3997	if (!error)
3998		error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
3999	done_path_create(&path, dentry);
4000	if (retry_estale(error, lookup_flags)) {
4001		lookup_flags |= LOOKUP_REVAL;
4002		goto retry;
4003	}
4004out_putname:
4005	putname(from);
4006	return error;
4007}
4008
4009SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
4010{
4011	return sys_symlinkat(oldname, AT_FDCWD, newname);
4012}
4013
4014/**
4015 * vfs_link - create a new link
4016 * @old_dentry:	object to be linked
4017 * @dir:	new parent
4018 * @new_dentry:	where to create the new link
4019 * @delegated_inode: returns inode needing a delegation break
4020 *
4021 * The caller must hold dir->i_mutex
4022 *
4023 * If vfs_link discovers a delegation on the to-be-linked file in need
4024 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4025 * inode in delegated_inode.  The caller should then break the delegation
4026 * and retry.  Because breaking a delegation may take a long time, the
4027 * caller should drop the i_mutex before doing so.
4028 *
4029 * Alternatively, a caller may pass NULL for delegated_inode.  This may
4030 * be appropriate for callers that expect the underlying filesystem not
4031 * to be NFS exported.
4032 */
4033int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
4034{
4035	struct inode *inode = old_dentry->d_inode;
4036	unsigned max_links = dir->i_sb->s_max_links;
4037	int error;
4038
4039	if (!inode)
4040		return -ENOENT;
4041
4042	error = may_create(dir, new_dentry);
4043	if (error)
4044		return error;
4045
4046	if (dir->i_sb != inode->i_sb)
4047		return -EXDEV;
4048
4049	/*
4050	 * A link to an append-only or immutable file cannot be created.
4051	 */
4052	if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4053		return -EPERM;
4054	if (!dir->i_op->link)
4055		return -EPERM;
4056	if (S_ISDIR(inode->i_mode))
4057		return -EPERM;
4058
4059	error = security_inode_link(old_dentry, dir, new_dentry);
4060	if (error)
4061		return error;
4062
4063	inode_lock(inode);
4064	/* Make sure we don't allow creating hardlink to an unlinked file */
4065	if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4066		error =  -ENOENT;
4067	else if (max_links && inode->i_nlink >= max_links)
4068		error = -EMLINK;
4069	else {
4070		error = try_break_deleg(inode, delegated_inode);
4071		if (!error)
4072			error = dir->i_op->link(old_dentry, dir, new_dentry);
4073	}
4074
4075	if (!error && (inode->i_state & I_LINKABLE)) {
4076		spin_lock(&inode->i_lock);
4077		inode->i_state &= ~I_LINKABLE;
4078		spin_unlock(&inode->i_lock);
4079	}
4080	inode_unlock(inode);
4081	if (!error)
4082		fsnotify_link(dir, inode, new_dentry);
4083	return error;
4084}
4085EXPORT_SYMBOL(vfs_link);
4086
4087/*
4088 * Hardlinks are often used in delicate situations.  We avoid
4089 * security-related surprises by not following symlinks on the
4090 * newname.  --KAB
4091 *
4092 * We don't follow them on the oldname either to be compatible
4093 * with linux 2.0, and to avoid hard-linking to directories
4094 * and other special files.  --ADM
4095 */
4096SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4097		int, newdfd, const char __user *, newname, int, flags)
4098{
4099	struct dentry *new_dentry;
4100	struct path old_path, new_path;
4101	struct inode *delegated_inode = NULL;
4102	int how = 0;
4103	int error;
4104
4105	if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
4106		return -EINVAL;
4107	/*
4108	 * To use null names we require CAP_DAC_READ_SEARCH
4109	 * This ensures that not everyone will be able to create
4110	 * handlink using the passed filedescriptor.
4111	 */
4112	if (flags & AT_EMPTY_PATH) {
4113		if (!capable(CAP_DAC_READ_SEARCH))
4114			return -ENOENT;
4115		how = LOOKUP_EMPTY;
4116	}
4117
4118	if (flags & AT_SYMLINK_FOLLOW)
4119		how |= LOOKUP_FOLLOW;
4120retry:
4121	error = user_path_at(olddfd, oldname, how, &old_path);
4122	if (error)
4123		return error;
4124
4125	new_dentry = user_path_create(newdfd, newname, &new_path,
4126					(how & LOOKUP_REVAL));
4127	error = PTR_ERR(new_dentry);
4128	if (IS_ERR(new_dentry))
4129		goto out;
4130
4131	error = -EXDEV;
4132	if (old_path.mnt != new_path.mnt)
4133		goto out_dput;
4134	error = may_linkat(&old_path);
4135	if (unlikely(error))
4136		goto out_dput;
4137	error = security_path_link(old_path.dentry, &new_path, new_dentry);
4138	if (error)
4139		goto out_dput;
4140	error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
4141out_dput:
4142	done_path_create(&new_path, new_dentry);
4143	if (delegated_inode) {
4144		error = break_deleg_wait(&delegated_inode);
4145		if (!error) {
4146			path_put(&old_path);
4147			goto retry;
4148		}
4149	}
4150	if (retry_estale(error, how)) {
4151		path_put(&old_path);
4152		how |= LOOKUP_REVAL;
4153		goto retry;
4154	}
4155out:
4156	path_put(&old_path);
4157
4158	return error;
4159}
4160
4161SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4162{
4163	return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4164}
4165
4166/**
4167 * vfs_rename - rename a filesystem object
4168 * @old_dir:	parent of source
4169 * @old_dentry:	source
4170 * @new_dir:	parent of destination
4171 * @new_dentry:	destination
4172 * @delegated_inode: returns an inode needing a delegation break
4173 * @flags:	rename flags
4174 *
4175 * The caller must hold multiple mutexes--see lock_rename()).
4176 *
4177 * If vfs_rename discovers a delegation in need of breaking at either
4178 * the source or destination, it will return -EWOULDBLOCK and return a
4179 * reference to the inode in delegated_inode.  The caller should then
4180 * break the delegation and retry.  Because breaking a delegation may
4181 * take a long time, the caller should drop all locks before doing
4182 * so.
4183 *
4184 * Alternatively, a caller may pass NULL for delegated_inode.  This may
4185 * be appropriate for callers that expect the underlying filesystem not
4186 * to be NFS exported.
4187 *
4188 * The worst of all namespace operations - renaming directory. "Perverted"
4189 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4190 * Problems:
4191 *	a) we can get into loop creation.
4192 *	b) race potential - two innocent renames can create a loop together.
4193 *	   That's where 4.4 screws up. Current fix: serialization on
4194 *	   sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4195 *	   story.
4196 *	c) we have to lock _four_ objects - parents and victim (if it exists),
4197 *	   and source (if it is not a directory).
4198 *	   And that - after we got ->i_mutex on parents (until then we don't know
4199 *	   whether the target exists).  Solution: try to be smart with locking
4200 *	   order for inodes.  We rely on the fact that tree topology may change
4201 *	   only under ->s_vfs_rename_mutex _and_ that parent of the object we
4202 *	   move will be locked.  Thus we can rank directories by the tree
4203 *	   (ancestors first) and rank all non-directories after them.
4204 *	   That works since everybody except rename does "lock parent, lookup,
4205 *	   lock child" and rename is under ->s_vfs_rename_mutex.
4206 *	   HOWEVER, it relies on the assumption that any object with ->lookup()
4207 *	   has no more than 1 dentry.  If "hybrid" objects will ever appear,
4208 *	   we'd better make sure that there's no link(2) for them.
4209 *	d) conversion from fhandle to dentry may come in the wrong moment - when
4210 *	   we are removing the target. Solution: we will have to grab ->i_mutex
4211 *	   in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4212 *	   ->i_mutex on parents, which works but leads to some truly excessive
4213 *	   locking].
4214 */
4215int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4216	       struct inode *new_dir, struct dentry *new_dentry,
4217	       struct inode **delegated_inode, unsigned int flags)
4218{
4219	int error;
4220	bool is_dir = d_is_dir(old_dentry);
4221	const unsigned char *old_name;
4222	struct inode *source = old_dentry->d_inode;
4223	struct inode *target = new_dentry->d_inode;
4224	bool new_is_dir = false;
4225	unsigned max_links = new_dir->i_sb->s_max_links;
4226
4227	/*
4228	 * Check source == target.
4229	 * On overlayfs need to look at underlying inodes.
4230	 */
4231	if (vfs_select_inode(old_dentry, 0) == vfs_select_inode(new_dentry, 0))
4232		return 0;
4233
4234	error = may_delete(old_dir, old_dentry, is_dir);
4235	if (error)
4236		return error;
4237
4238	if (!target) {
4239		error = may_create(new_dir, new_dentry);
4240	} else {
4241		new_is_dir = d_is_dir(new_dentry);
4242
4243		if (!(flags & RENAME_EXCHANGE))
4244			error = may_delete(new_dir, new_dentry, is_dir);
4245		else
4246			error = may_delete(new_dir, new_dentry, new_is_dir);
4247	}
4248	if (error)
4249		return error;
4250
4251	if (!old_dir->i_op->rename && !old_dir->i_op->rename2)
4252		return -EPERM;
4253
4254	if (flags && !old_dir->i_op->rename2)
4255		return -EINVAL;
4256
4257	/*
4258	 * If we are going to change the parent - check write permissions,
4259	 * we'll need to flip '..'.
4260	 */
4261	if (new_dir != old_dir) {
4262		if (is_dir) {
4263			error = inode_permission(source, MAY_WRITE);
4264			if (error)
4265				return error;
4266		}
4267		if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4268			error = inode_permission(target, MAY_WRITE);
4269			if (error)
4270				return error;
4271		}
4272	}
4273
4274	error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4275				      flags);
4276	if (error)
4277		return error;
4278
4279	old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4280	dget(new_dentry);
4281	if (!is_dir || (flags & RENAME_EXCHANGE))
4282		lock_two_nondirectories(source, target);
4283	else if (target)
4284		inode_lock(target);
4285
4286	error = -EBUSY;
4287	if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4288		goto out;
4289
4290	if (max_links && new_dir != old_dir) {
4291		error = -EMLINK;
4292		if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4293			goto out;
4294		if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4295		    old_dir->i_nlink >= max_links)
4296			goto out;
4297	}
4298	if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4299		shrink_dcache_parent(new_dentry);
4300	if (!is_dir) {
4301		error = try_break_deleg(source, delegated_inode);
4302		if (error)
4303			goto out;
4304	}
4305	if (target && !new_is_dir) {
4306		error = try_break_deleg(target, delegated_inode);
4307		if (error)
4308			goto out;
4309	}
4310	if (!old_dir->i_op->rename2) {
4311		error = old_dir->i_op->rename(old_dir, old_dentry,
4312					      new_dir, new_dentry);
4313	} else {
4314		WARN_ON(old_dir->i_op->rename != NULL);
4315		error = old_dir->i_op->rename2(old_dir, old_dentry,
4316					       new_dir, new_dentry, flags);
4317	}
4318	if (error)
4319		goto out;
4320
4321	if (!(flags & RENAME_EXCHANGE) && target) {
4322		if (is_dir)
4323			target->i_flags |= S_DEAD;
4324		dont_mount(new_dentry);
4325		detach_mounts(new_dentry);
4326	}
4327	if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4328		if (!(flags & RENAME_EXCHANGE))
4329			d_move(old_dentry, new_dentry);
4330		else
4331			d_exchange(old_dentry, new_dentry);
4332	}
4333out:
4334	if (!is_dir || (flags & RENAME_EXCHANGE))
4335		unlock_two_nondirectories(source, target);
4336	else if (target)
4337		inode_unlock(target);
4338	dput(new_dentry);
4339	if (!error) {
4340		fsnotify_move(old_dir, new_dir, old_name, is_dir,
4341			      !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4342		if (flags & RENAME_EXCHANGE) {
4343			fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4344				      new_is_dir, NULL, new_dentry);
4345		}
4346	}
4347	fsnotify_oldname_free(old_name);
4348
4349	return error;
4350}
4351EXPORT_SYMBOL(vfs_rename);
4352
4353SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4354		int, newdfd, const char __user *, newname, unsigned int, flags)
4355{
4356	struct dentry *old_dentry, *new_dentry;
4357	struct dentry *trap;
4358	struct path old_path, new_path;
4359	struct qstr old_last, new_last;
4360	int old_type, new_type;
4361	struct inode *delegated_inode = NULL;
4362	struct filename *from;
4363	struct filename *to;
4364	unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4365	bool should_retry = false;
4366	int error;
4367
4368	if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4369		return -EINVAL;
4370
4371	if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4372	    (flags & RENAME_EXCHANGE))
4373		return -EINVAL;
4374
4375	if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4376		return -EPERM;
4377
4378	if (flags & RENAME_EXCHANGE)
4379		target_flags = 0;
4380
4381retry:
4382	from = user_path_parent(olddfd, oldname,
4383				&old_path, &old_last, &old_type, lookup_flags);
4384	if (IS_ERR(from)) {
4385		error = PTR_ERR(from);
4386		goto exit;
4387	}
4388
4389	to = user_path_parent(newdfd, newname,
4390				&new_path, &new_last, &new_type, lookup_flags);
4391	if (IS_ERR(to)) {
4392		error = PTR_ERR(to);
4393		goto exit1;
4394	}
4395
4396	error = -EXDEV;
4397	if (old_path.mnt != new_path.mnt)
4398		goto exit2;
4399
4400	error = -EBUSY;
4401	if (old_type != LAST_NORM)
4402		goto exit2;
4403
4404	if (flags & RENAME_NOREPLACE)
4405		error = -EEXIST;
4406	if (new_type != LAST_NORM)
4407		goto exit2;
4408
4409	error = mnt_want_write(old_path.mnt);
4410	if (error)
4411		goto exit2;
4412
4413retry_deleg:
4414	trap = lock_rename(new_path.dentry, old_path.dentry);
4415
4416	old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4417	error = PTR_ERR(old_dentry);
4418	if (IS_ERR(old_dentry))
4419		goto exit3;
4420	/* source must exist */
4421	error = -ENOENT;
4422	if (d_is_negative(old_dentry))
4423		goto exit4;
4424	new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4425	error = PTR_ERR(new_dentry);
4426	if (IS_ERR(new_dentry))
4427		goto exit4;
4428	error = -EEXIST;
4429	if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4430		goto exit5;
4431	if (flags & RENAME_EXCHANGE) {
4432		error = -ENOENT;
4433		if (d_is_negative(new_dentry))
4434			goto exit5;
4435
4436		if (!d_is_dir(new_dentry)) {
4437			error = -ENOTDIR;
4438			if (new_last.name[new_last.len])
4439				goto exit5;
4440		}
4441	}
4442	/* unless the source is a directory trailing slashes give -ENOTDIR */
4443	if (!d_is_dir(old_dentry)) {
4444		error = -ENOTDIR;
4445		if (old_last.name[old_last.len])
4446			goto exit5;
4447		if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4448			goto exit5;
4449	}
4450	/* source should not be ancestor of target */
4451	error = -EINVAL;
4452	if (old_dentry == trap)
4453		goto exit5;
4454	/* target should not be an ancestor of source */
4455	if (!(flags & RENAME_EXCHANGE))
4456		error = -ENOTEMPTY;
4457	if (new_dentry == trap)
4458		goto exit5;
4459
4460	error = security_path_rename(&old_path, old_dentry,
4461				     &new_path, new_dentry, flags);
4462	if (error)
4463		goto exit5;
4464	error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4465			   new_path.dentry->d_inode, new_dentry,
4466			   &delegated_inode, flags);
4467exit5:
4468	dput(new_dentry);
4469exit4:
4470	dput(old_dentry);
4471exit3:
4472	unlock_rename(new_path.dentry, old_path.dentry);
4473	if (delegated_inode) {
4474		error = break_deleg_wait(&delegated_inode);
4475		if (!error)
4476			goto retry_deleg;
4477	}
4478	mnt_drop_write(old_path.mnt);
4479exit2:
4480	if (retry_estale(error, lookup_flags))
4481		should_retry = true;
4482	path_put(&new_path);
4483	putname(to);
4484exit1:
4485	path_put(&old_path);
4486	putname(from);
4487	if (should_retry) {
4488		should_retry = false;
4489		lookup_flags |= LOOKUP_REVAL;
4490		goto retry;
4491	}
4492exit:
4493	return error;
4494}
4495
4496SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4497		int, newdfd, const char __user *, newname)
4498{
4499	return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4500}
4501
4502SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4503{
4504	return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4505}
4506
4507int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4508{
4509	int error = may_create(dir, dentry);
4510	if (error)
4511		return error;
4512
4513	if (!dir->i_op->mknod)
4514		return -EPERM;
4515
4516	return dir->i_op->mknod(dir, dentry,
4517				S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4518}
4519EXPORT_SYMBOL(vfs_whiteout);
4520
4521int readlink_copy(char __user *buffer, int buflen, const char *link)
4522{
4523	int len = PTR_ERR(link);
4524	if (IS_ERR(link))
4525		goto out;
4526
4527	len = strlen(link);
4528	if (len > (unsigned) buflen)
4529		len = buflen;
4530	if (copy_to_user(buffer, link, len))
4531		len = -EFAULT;
4532out:
4533	return len;
4534}
4535EXPORT_SYMBOL(readlink_copy);
4536
4537/*
4538 * A helper for ->readlink().  This should be used *ONLY* for symlinks that
4539 * have ->get_link() not calling nd_jump_link().  Using (or not using) it
4540 * for any given inode is up to filesystem.
4541 */
4542int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4543{
4544	DEFINE_DELAYED_CALL(done);
4545	struct inode *inode = d_inode(dentry);
4546	const char *link = inode->i_link;
4547	int res;
4548
4549	if (!link) {
4550		link = inode->i_op->get_link(dentry, inode, &done);
4551		if (IS_ERR(link))
4552			return PTR_ERR(link);
4553	}
4554	res = readlink_copy(buffer, buflen, link);
4555	do_delayed_call(&done);
4556	return res;
4557}
4558EXPORT_SYMBOL(generic_readlink);
4559
4560/* get the link contents into pagecache */
4561const char *page_get_link(struct dentry *dentry, struct inode *inode,
4562			  struct delayed_call *callback)
4563{
4564	char *kaddr;
4565	struct page *page;
4566	struct address_space *mapping = inode->i_mapping;
4567
4568	if (!dentry) {
4569		page = find_get_page(mapping, 0);
4570		if (!page)
4571			return ERR_PTR(-ECHILD);
4572		if (!PageUptodate(page)) {
4573			put_page(page);
4574			return ERR_PTR(-ECHILD);
4575		}
4576	} else {
4577		page = read_mapping_page(mapping, 0, NULL);
4578		if (IS_ERR(page))
4579			return (char*)page;
4580	}
4581	set_delayed_call(callback, page_put_link, page);
4582	BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
4583	kaddr = page_address(page);
4584	nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
4585	return kaddr;
4586}
4587
4588EXPORT_SYMBOL(page_get_link);
4589
4590void page_put_link(void *arg)
4591{
4592	put_page(arg);
4593}
4594EXPORT_SYMBOL(page_put_link);
4595
4596int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4597{
4598	DEFINE_DELAYED_CALL(done);
4599	int res = readlink_copy(buffer, buflen,
4600				page_get_link(dentry, d_inode(dentry),
4601					      &done));
4602	do_delayed_call(&done);
4603	return res;
4604}
4605EXPORT_SYMBOL(page_readlink);
4606
4607/*
4608 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4609 */
4610int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4611{
4612	struct address_space *mapping = inode->i_mapping;
4613	struct page *page;
4614	void *fsdata;
4615	int err;
4616	unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4617	if (nofs)
4618		flags |= AOP_FLAG_NOFS;
4619
4620retry:
4621	err = pagecache_write_begin(NULL, mapping, 0, len-1,
4622				flags, &page, &fsdata);
4623	if (err)
4624		goto fail;
4625
4626	memcpy(page_address(page), symname, len-1);
4627
4628	err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4629							page, fsdata);
4630	if (err < 0)
4631		goto fail;
4632	if (err < len-1)
4633		goto retry;
4634
4635	mark_inode_dirty(inode);
4636	return 0;
4637fail:
4638	return err;
4639}
4640EXPORT_SYMBOL(__page_symlink);
4641
4642int page_symlink(struct inode *inode, const char *symname, int len)
4643{
4644	return __page_symlink(inode, symname, len,
4645			!mapping_gfp_constraint(inode->i_mapping, __GFP_FS));
4646}
4647EXPORT_SYMBOL(page_symlink);
4648
4649const struct inode_operations page_symlink_inode_operations = {
4650	.readlink	= generic_readlink,
4651	.get_link	= page_get_link,
4652};
4653EXPORT_SYMBOL(page_symlink_inode_operations);
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 *  linux/fs/namei.c
   4 *
   5 *  Copyright (C) 1991, 1992  Linus Torvalds
   6 */
   7
   8/*
   9 * Some corrections by tytso.
  10 */
  11
  12/* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
  13 * lookup logic.
  14 */
  15/* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
  16 */
  17
  18#include <linux/init.h>
  19#include <linux/export.h>
  20#include <linux/kernel.h>
  21#include <linux/slab.h>
  22#include <linux/fs.h>
  23#include <linux/namei.h>
  24#include <linux/pagemap.h>
  25#include <linux/sched/mm.h>
  26#include <linux/fsnotify.h>
  27#include <linux/personality.h>
  28#include <linux/security.h>
  29#include <linux/ima.h>
  30#include <linux/syscalls.h>
  31#include <linux/mount.h>
  32#include <linux/audit.h>
  33#include <linux/capability.h>
  34#include <linux/file.h>
  35#include <linux/fcntl.h>
  36#include <linux/device_cgroup.h>
  37#include <linux/fs_struct.h>
  38#include <linux/posix_acl.h>
  39#include <linux/hash.h>
  40#include <linux/bitops.h>
  41#include <linux/init_task.h>
  42#include <linux/uaccess.h>
  43
  44#include "internal.h"
  45#include "mount.h"
  46
  47/* [Feb-1997 T. Schoebel-Theuer]
  48 * Fundamental changes in the pathname lookup mechanisms (namei)
  49 * were necessary because of omirr.  The reason is that omirr needs
  50 * to know the _real_ pathname, not the user-supplied one, in case
  51 * of symlinks (and also when transname replacements occur).
  52 *
  53 * The new code replaces the old recursive symlink resolution with
  54 * an iterative one (in case of non-nested symlink chains).  It does
  55 * this with calls to <fs>_follow_link().
  56 * As a side effect, dir_namei(), _namei() and follow_link() are now 
  57 * replaced with a single function lookup_dentry() that can handle all 
  58 * the special cases of the former code.
  59 *
  60 * With the new dcache, the pathname is stored at each inode, at least as
  61 * long as the refcount of the inode is positive.  As a side effect, the
  62 * size of the dcache depends on the inode cache and thus is dynamic.
  63 *
  64 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
  65 * resolution to correspond with current state of the code.
  66 *
  67 * Note that the symlink resolution is not *completely* iterative.
  68 * There is still a significant amount of tail- and mid- recursion in
  69 * the algorithm.  Also, note that <fs>_readlink() is not used in
  70 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
  71 * may return different results than <fs>_follow_link().  Many virtual
  72 * filesystems (including /proc) exhibit this behavior.
  73 */
  74
  75/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
  76 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
  77 * and the name already exists in form of a symlink, try to create the new
  78 * name indicated by the symlink. The old code always complained that the
  79 * name already exists, due to not following the symlink even if its target
  80 * is nonexistent.  The new semantics affects also mknod() and link() when
  81 * the name is a symlink pointing to a non-existent name.
  82 *
  83 * I don't know which semantics is the right one, since I have no access
  84 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
  85 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
  86 * "old" one. Personally, I think the new semantics is much more logical.
  87 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
  88 * file does succeed in both HP-UX and SunOs, but not in Solaris
  89 * and in the old Linux semantics.
  90 */
  91
  92/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
  93 * semantics.  See the comments in "open_namei" and "do_link" below.
  94 *
  95 * [10-Sep-98 Alan Modra] Another symlink change.
  96 */
  97
  98/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
  99 *	inside the path - always follow.
 100 *	in the last component in creation/removal/renaming - never follow.
 101 *	if LOOKUP_FOLLOW passed - follow.
 102 *	if the pathname has trailing slashes - follow.
 103 *	otherwise - don't follow.
 104 * (applied in that order).
 105 *
 106 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
 107 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
 108 * During the 2.4 we need to fix the userland stuff depending on it -
 109 * hopefully we will be able to get rid of that wart in 2.5. So far only
 110 * XEmacs seems to be relying on it...
 111 */
 112/*
 113 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
 114 * implemented.  Let's see if raised priority of ->s_vfs_rename_mutex gives
 115 * any extra contention...
 116 */
 117
 118/* In order to reduce some races, while at the same time doing additional
 119 * checking and hopefully speeding things up, we copy filenames to the
 120 * kernel data space before using them..
 121 *
 122 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
 123 * PATH_MAX includes the nul terminator --RR.
 124 */
 125
 126#define EMBEDDED_NAME_MAX	(PATH_MAX - offsetof(struct filename, iname))
 127
 128struct filename *
 129getname_flags(const char __user *filename, int flags, int *empty)
 130{
 131	struct filename *result;
 132	char *kname;
 133	int len;
 134
 135	result = audit_reusename(filename);
 136	if (result)
 137		return result;
 138
 139	result = __getname();
 140	if (unlikely(!result))
 141		return ERR_PTR(-ENOMEM);
 142
 143	/*
 144	 * First, try to embed the struct filename inside the names_cache
 145	 * allocation
 146	 */
 147	kname = (char *)result->iname;
 148	result->name = kname;
 149
 150	len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX);
 151	if (unlikely(len < 0)) {
 152		__putname(result);
 153		return ERR_PTR(len);
 154	}
 155
 156	/*
 157	 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
 158	 * separate struct filename so we can dedicate the entire
 159	 * names_cache allocation for the pathname, and re-do the copy from
 160	 * userland.
 161	 */
 162	if (unlikely(len == EMBEDDED_NAME_MAX)) {
 163		const size_t size = offsetof(struct filename, iname[1]);
 164		kname = (char *)result;
 165
 166		/*
 167		 * size is chosen that way we to guarantee that
 168		 * result->iname[0] is within the same object and that
 169		 * kname can't be equal to result->iname, no matter what.
 170		 */
 171		result = kzalloc(size, GFP_KERNEL);
 172		if (unlikely(!result)) {
 173			__putname(kname);
 174			return ERR_PTR(-ENOMEM);
 175		}
 176		result->name = kname;
 177		len = strncpy_from_user(kname, filename, PATH_MAX);
 178		if (unlikely(len < 0)) {
 179			__putname(kname);
 180			kfree(result);
 181			return ERR_PTR(len);
 182		}
 183		if (unlikely(len == PATH_MAX)) {
 184			__putname(kname);
 185			kfree(result);
 186			return ERR_PTR(-ENAMETOOLONG);
 187		}
 188	}
 189
 190	result->refcnt = 1;
 191	/* The empty path is special. */
 192	if (unlikely(!len)) {
 193		if (empty)
 194			*empty = 1;
 195		if (!(flags & LOOKUP_EMPTY)) {
 196			putname(result);
 197			return ERR_PTR(-ENOENT);
 198		}
 199	}
 200
 201	result->uptr = filename;
 202	result->aname = NULL;
 203	audit_getname(result);
 204	return result;
 205}
 206
 207struct filename *
 208getname_uflags(const char __user *filename, int uflags)
 209{
 210	int flags = (uflags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
 211
 212	return getname_flags(filename, flags, NULL);
 213}
 214
 215struct filename *
 216getname(const char __user * filename)
 217{
 218	return getname_flags(filename, 0, NULL);
 219}
 220
 221struct filename *
 222getname_kernel(const char * filename)
 223{
 224	struct filename *result;
 225	int len = strlen(filename) + 1;
 226
 227	result = __getname();
 228	if (unlikely(!result))
 229		return ERR_PTR(-ENOMEM);
 230
 231	if (len <= EMBEDDED_NAME_MAX) {
 232		result->name = (char *)result->iname;
 233	} else if (len <= PATH_MAX) {
 234		const size_t size = offsetof(struct filename, iname[1]);
 235		struct filename *tmp;
 236
 237		tmp = kmalloc(size, GFP_KERNEL);
 238		if (unlikely(!tmp)) {
 239			__putname(result);
 240			return ERR_PTR(-ENOMEM);
 241		}
 242		tmp->name = (char *)result;
 243		result = tmp;
 244	} else {
 245		__putname(result);
 246		return ERR_PTR(-ENAMETOOLONG);
 247	}
 248	memcpy((char *)result->name, filename, len);
 249	result->uptr = NULL;
 250	result->aname = NULL;
 251	result->refcnt = 1;
 252	audit_getname(result);
 253
 254	return result;
 255}
 256
 257void putname(struct filename *name)
 258{
 259	if (IS_ERR(name))
 260		return;
 261
 262	BUG_ON(name->refcnt <= 0);
 263
 264	if (--name->refcnt > 0)
 265		return;
 266
 267	if (name->name != name->iname) {
 268		__putname(name->name);
 269		kfree(name);
 270	} else
 271		__putname(name);
 272}
 273
 274/**
 275 * check_acl - perform ACL permission checking
 276 * @mnt_userns:	user namespace of the mount the inode was found from
 277 * @inode:	inode to check permissions on
 278 * @mask:	right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
 279 *
 280 * This function performs the ACL permission checking. Since this function
 281 * retrieve POSIX acls it needs to know whether it is called from a blocking or
 282 * non-blocking context and thus cares about the MAY_NOT_BLOCK bit.
 283 *
 284 * If the inode has been found through an idmapped mount the user namespace of
 285 * the vfsmount must be passed through @mnt_userns. This function will then take
 286 * care to map the inode according to @mnt_userns before checking permissions.
 287 * On non-idmapped mounts or if permission checking is to be performed on the
 288 * raw inode simply passs init_user_ns.
 289 */
 290static int check_acl(struct user_namespace *mnt_userns,
 291		     struct inode *inode, int mask)
 292{
 293#ifdef CONFIG_FS_POSIX_ACL
 294	struct posix_acl *acl;
 295
 296	if (mask & MAY_NOT_BLOCK) {
 297		acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
 298	        if (!acl)
 299	                return -EAGAIN;
 300		/* no ->get_inode_acl() calls in RCU mode... */
 301		if (is_uncached_acl(acl))
 302			return -ECHILD;
 303	        return posix_acl_permission(mnt_userns, inode, acl, mask);
 304	}
 305
 306	acl = get_inode_acl(inode, ACL_TYPE_ACCESS);
 307	if (IS_ERR(acl))
 308		return PTR_ERR(acl);
 309	if (acl) {
 310	        int error = posix_acl_permission(mnt_userns, inode, acl, mask);
 311	        posix_acl_release(acl);
 312	        return error;
 313	}
 314#endif
 315
 316	return -EAGAIN;
 317}
 318
 319/**
 320 * acl_permission_check - perform basic UNIX permission checking
 321 * @mnt_userns:	user namespace of the mount the inode was found from
 322 * @inode:	inode to check permissions on
 323 * @mask:	right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
 324 *
 325 * This function performs the basic UNIX permission checking. Since this
 326 * function may retrieve POSIX acls it needs to know whether it is called from a
 327 * blocking or non-blocking context and thus cares about the MAY_NOT_BLOCK bit.
 328 *
 329 * If the inode has been found through an idmapped mount the user namespace of
 330 * the vfsmount must be passed through @mnt_userns. This function will then take
 331 * care to map the inode according to @mnt_userns before checking permissions.
 332 * On non-idmapped mounts or if permission checking is to be performed on the
 333 * raw inode simply passs init_user_ns.
 334 */
 335static int acl_permission_check(struct user_namespace *mnt_userns,
 336				struct inode *inode, int mask)
 337{
 338	unsigned int mode = inode->i_mode;
 339	vfsuid_t vfsuid;
 340
 341	/* Are we the owner? If so, ACL's don't matter */
 342	vfsuid = i_uid_into_vfsuid(mnt_userns, inode);
 343	if (likely(vfsuid_eq_kuid(vfsuid, current_fsuid()))) {
 344		mask &= 7;
 345		mode >>= 6;
 346		return (mask & ~mode) ? -EACCES : 0;
 347	}
 348
 349	/* Do we have ACL's? */
 350	if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
 351		int error = check_acl(mnt_userns, inode, mask);
 352		if (error != -EAGAIN)
 353			return error;
 354	}
 355
 356	/* Only RWX matters for group/other mode bits */
 357	mask &= 7;
 358
 359	/*
 360	 * Are the group permissions different from
 361	 * the other permissions in the bits we care
 362	 * about? Need to check group ownership if so.
 363	 */
 364	if (mask & (mode ^ (mode >> 3))) {
 365		vfsgid_t vfsgid = i_gid_into_vfsgid(mnt_userns, inode);
 366		if (vfsgid_in_group_p(vfsgid))
 367			mode >>= 3;
 368	}
 369
 370	/* Bits in 'mode' clear that we require? */
 371	return (mask & ~mode) ? -EACCES : 0;
 372}
 373
 374/**
 375 * generic_permission -  check for access rights on a Posix-like filesystem
 376 * @mnt_userns:	user namespace of the mount the inode was found from
 377 * @inode:	inode to check access rights for
 378 * @mask:	right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC,
 379 *		%MAY_NOT_BLOCK ...)
 380 *
 381 * Used to check for read/write/execute permissions on a file.
 382 * We use "fsuid" for this, letting us set arbitrary permissions
 383 * for filesystem access without changing the "normal" uids which
 384 * are used for other things.
 385 *
 386 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
 387 * request cannot be satisfied (eg. requires blocking or too much complexity).
 388 * It would then be called again in ref-walk mode.
 389 *
 390 * If the inode has been found through an idmapped mount the user namespace of
 391 * the vfsmount must be passed through @mnt_userns. This function will then take
 392 * care to map the inode according to @mnt_userns before checking permissions.
 393 * On non-idmapped mounts or if permission checking is to be performed on the
 394 * raw inode simply passs init_user_ns.
 395 */
 396int generic_permission(struct user_namespace *mnt_userns, struct inode *inode,
 397		       int mask)
 398{
 399	int ret;
 400
 401	/*
 402	 * Do the basic permission checks.
 403	 */
 404	ret = acl_permission_check(mnt_userns, inode, mask);
 405	if (ret != -EACCES)
 406		return ret;
 407
 408	if (S_ISDIR(inode->i_mode)) {
 409		/* DACs are overridable for directories */
 410		if (!(mask & MAY_WRITE))
 411			if (capable_wrt_inode_uidgid(mnt_userns, inode,
 412						     CAP_DAC_READ_SEARCH))
 413				return 0;
 414		if (capable_wrt_inode_uidgid(mnt_userns, inode,
 415					     CAP_DAC_OVERRIDE))
 416			return 0;
 417		return -EACCES;
 418	}
 419
 420	/*
 421	 * Searching includes executable on directories, else just read.
 422	 */
 423	mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
 424	if (mask == MAY_READ)
 425		if (capable_wrt_inode_uidgid(mnt_userns, inode,
 426					     CAP_DAC_READ_SEARCH))
 427			return 0;
 428	/*
 429	 * Read/write DACs are always overridable.
 430	 * Executable DACs are overridable when there is
 431	 * at least one exec bit set.
 432	 */
 433	if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
 434		if (capable_wrt_inode_uidgid(mnt_userns, inode,
 435					     CAP_DAC_OVERRIDE))
 436			return 0;
 437
 438	return -EACCES;
 439}
 440EXPORT_SYMBOL(generic_permission);
 441
 442/**
 443 * do_inode_permission - UNIX permission checking
 444 * @mnt_userns:	user namespace of the mount the inode was found from
 445 * @inode:	inode to check permissions on
 446 * @mask:	right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
 447 *
 448 * We _really_ want to just do "generic_permission()" without
 449 * even looking at the inode->i_op values. So we keep a cache
 450 * flag in inode->i_opflags, that says "this has not special
 451 * permission function, use the fast case".
 452 */
 453static inline int do_inode_permission(struct user_namespace *mnt_userns,
 454				      struct inode *inode, int mask)
 455{
 456	if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
 457		if (likely(inode->i_op->permission))
 458			return inode->i_op->permission(mnt_userns, inode, mask);
 459
 460		/* This gets set once for the inode lifetime */
 461		spin_lock(&inode->i_lock);
 462		inode->i_opflags |= IOP_FASTPERM;
 463		spin_unlock(&inode->i_lock);
 464	}
 465	return generic_permission(mnt_userns, inode, mask);
 466}
 467
 468/**
 469 * sb_permission - Check superblock-level permissions
 470 * @sb: Superblock of inode to check permission on
 471 * @inode: Inode to check permission on
 472 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
 473 *
 474 * Separate out file-system wide checks from inode-specific permission checks.
 475 */
 476static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
 477{
 478	if (unlikely(mask & MAY_WRITE)) {
 479		umode_t mode = inode->i_mode;
 480
 481		/* Nobody gets write access to a read-only fs. */
 482		if (sb_rdonly(sb) && (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
 483			return -EROFS;
 484	}
 485	return 0;
 486}
 487
 488/**
 489 * inode_permission - Check for access rights to a given inode
 490 * @mnt_userns:	User namespace of the mount the inode was found from
 491 * @inode:	Inode to check permission on
 492 * @mask:	Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
 493 *
 494 * Check for read/write/execute permissions on an inode.  We use fs[ug]id for
 495 * this, letting us set arbitrary permissions for filesystem access without
 496 * changing the "normal" UIDs which are used for other things.
 497 *
 498 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
 499 */
 500int inode_permission(struct user_namespace *mnt_userns,
 501		     struct inode *inode, int mask)
 502{
 503	int retval;
 504
 505	retval = sb_permission(inode->i_sb, inode, mask);
 506	if (retval)
 507		return retval;
 508
 509	if (unlikely(mask & MAY_WRITE)) {
 510		/*
 511		 * Nobody gets write access to an immutable file.
 512		 */
 513		if (IS_IMMUTABLE(inode))
 514			return -EPERM;
 515
 516		/*
 517		 * Updating mtime will likely cause i_uid and i_gid to be
 518		 * written back improperly if their true value is unknown
 519		 * to the vfs.
 520		 */
 521		if (HAS_UNMAPPED_ID(mnt_userns, inode))
 522			return -EACCES;
 523	}
 524
 525	retval = do_inode_permission(mnt_userns, inode, mask);
 526	if (retval)
 527		return retval;
 528
 529	retval = devcgroup_inode_permission(inode, mask);
 530	if (retval)
 531		return retval;
 532
 533	return security_inode_permission(inode, mask);
 534}
 535EXPORT_SYMBOL(inode_permission);
 536
 537/**
 538 * path_get - get a reference to a path
 539 * @path: path to get the reference to
 540 *
 541 * Given a path increment the reference count to the dentry and the vfsmount.
 542 */
 543void path_get(const struct path *path)
 544{
 545	mntget(path->mnt);
 546	dget(path->dentry);
 547}
 548EXPORT_SYMBOL(path_get);
 549
 550/**
 551 * path_put - put a reference to a path
 552 * @path: path to put the reference to
 553 *
 554 * Given a path decrement the reference count to the dentry and the vfsmount.
 555 */
 556void path_put(const struct path *path)
 557{
 558	dput(path->dentry);
 559	mntput(path->mnt);
 560}
 561EXPORT_SYMBOL(path_put);
 562
 563#define EMBEDDED_LEVELS 2
 564struct nameidata {
 565	struct path	path;
 566	struct qstr	last;
 567	struct path	root;
 568	struct inode	*inode; /* path.dentry.d_inode */
 569	unsigned int	flags, state;
 570	unsigned	seq, next_seq, m_seq, r_seq;
 571	int		last_type;
 572	unsigned	depth;
 573	int		total_link_count;
 574	struct saved {
 575		struct path link;
 576		struct delayed_call done;
 577		const char *name;
 578		unsigned seq;
 579	} *stack, internal[EMBEDDED_LEVELS];
 580	struct filename	*name;
 581	struct nameidata *saved;
 582	unsigned	root_seq;
 583	int		dfd;
 584	vfsuid_t	dir_vfsuid;
 585	umode_t		dir_mode;
 586} __randomize_layout;
 587
 588#define ND_ROOT_PRESET 1
 589#define ND_ROOT_GRABBED 2
 590#define ND_JUMPED 4
 591
 592static void __set_nameidata(struct nameidata *p, int dfd, struct filename *name)
 593{
 594	struct nameidata *old = current->nameidata;
 595	p->stack = p->internal;
 596	p->depth = 0;
 597	p->dfd = dfd;
 598	p->name = name;
 599	p->path.mnt = NULL;
 600	p->path.dentry = NULL;
 601	p->total_link_count = old ? old->total_link_count : 0;
 602	p->saved = old;
 603	current->nameidata = p;
 604}
 605
 606static inline void set_nameidata(struct nameidata *p, int dfd, struct filename *name,
 607			  const struct path *root)
 608{
 609	__set_nameidata(p, dfd, name);
 610	p->state = 0;
 611	if (unlikely(root)) {
 612		p->state = ND_ROOT_PRESET;
 613		p->root = *root;
 614	}
 615}
 616
 617static void restore_nameidata(void)
 618{
 619	struct nameidata *now = current->nameidata, *old = now->saved;
 620
 621	current->nameidata = old;
 622	if (old)
 623		old->total_link_count = now->total_link_count;
 624	if (now->stack != now->internal)
 625		kfree(now->stack);
 626}
 627
 628static bool nd_alloc_stack(struct nameidata *nd)
 629{
 630	struct saved *p;
 631
 632	p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved),
 633			 nd->flags & LOOKUP_RCU ? GFP_ATOMIC : GFP_KERNEL);
 634	if (unlikely(!p))
 635		return false;
 636	memcpy(p, nd->internal, sizeof(nd->internal));
 637	nd->stack = p;
 638	return true;
 639}
 640
 641/**
 642 * path_connected - Verify that a dentry is below mnt.mnt_root
 643 *
 644 * Rename can sometimes move a file or directory outside of a bind
 645 * mount, path_connected allows those cases to be detected.
 646 */
 647static bool path_connected(struct vfsmount *mnt, struct dentry *dentry)
 648{
 649	struct super_block *sb = mnt->mnt_sb;
 650
 651	/* Bind mounts can have disconnected paths */
 652	if (mnt->mnt_root == sb->s_root)
 653		return true;
 654
 655	return is_subdir(dentry, mnt->mnt_root);
 656}
 657
 658static void drop_links(struct nameidata *nd)
 659{
 660	int i = nd->depth;
 661	while (i--) {
 662		struct saved *last = nd->stack + i;
 663		do_delayed_call(&last->done);
 664		clear_delayed_call(&last->done);
 665	}
 666}
 667
 668static void leave_rcu(struct nameidata *nd)
 669{
 670	nd->flags &= ~LOOKUP_RCU;
 671	nd->seq = nd->next_seq = 0;
 672	rcu_read_unlock();
 673}
 674
 675static void terminate_walk(struct nameidata *nd)
 676{
 677	drop_links(nd);
 678	if (!(nd->flags & LOOKUP_RCU)) {
 679		int i;
 680		path_put(&nd->path);
 681		for (i = 0; i < nd->depth; i++)
 682			path_put(&nd->stack[i].link);
 683		if (nd->state & ND_ROOT_GRABBED) {
 684			path_put(&nd->root);
 685			nd->state &= ~ND_ROOT_GRABBED;
 686		}
 687	} else {
 688		leave_rcu(nd);
 689	}
 690	nd->depth = 0;
 691	nd->path.mnt = NULL;
 692	nd->path.dentry = NULL;
 693}
 694
 695/* path_put is needed afterwards regardless of success or failure */
 696static bool __legitimize_path(struct path *path, unsigned seq, unsigned mseq)
 697{
 698	int res = __legitimize_mnt(path->mnt, mseq);
 699	if (unlikely(res)) {
 700		if (res > 0)
 701			path->mnt = NULL;
 702		path->dentry = NULL;
 703		return false;
 704	}
 705	if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
 706		path->dentry = NULL;
 707		return false;
 708	}
 709	return !read_seqcount_retry(&path->dentry->d_seq, seq);
 710}
 711
 712static inline bool legitimize_path(struct nameidata *nd,
 713			    struct path *path, unsigned seq)
 714{
 715	return __legitimize_path(path, seq, nd->m_seq);
 716}
 717
 718static bool legitimize_links(struct nameidata *nd)
 719{
 720	int i;
 721	if (unlikely(nd->flags & LOOKUP_CACHED)) {
 722		drop_links(nd);
 723		nd->depth = 0;
 724		return false;
 725	}
 726	for (i = 0; i < nd->depth; i++) {
 727		struct saved *last = nd->stack + i;
 728		if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
 729			drop_links(nd);
 730			nd->depth = i + 1;
 731			return false;
 732		}
 733	}
 734	return true;
 735}
 736
 737static bool legitimize_root(struct nameidata *nd)
 738{
 739	/* Nothing to do if nd->root is zero or is managed by the VFS user. */
 740	if (!nd->root.mnt || (nd->state & ND_ROOT_PRESET))
 741		return true;
 742	nd->state |= ND_ROOT_GRABBED;
 743	return legitimize_path(nd, &nd->root, nd->root_seq);
 744}
 745
 746/*
 747 * Path walking has 2 modes, rcu-walk and ref-walk (see
 748 * Documentation/filesystems/path-lookup.txt).  In situations when we can't
 749 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
 750 * normal reference counts on dentries and vfsmounts to transition to ref-walk
 751 * mode.  Refcounts are grabbed at the last known good point before rcu-walk
 752 * got stuck, so ref-walk may continue from there. If this is not successful
 753 * (eg. a seqcount has changed), then failure is returned and it's up to caller
 754 * to restart the path walk from the beginning in ref-walk mode.
 755 */
 756
 757/**
 758 * try_to_unlazy - try to switch to ref-walk mode.
 759 * @nd: nameidata pathwalk data
 760 * Returns: true on success, false on failure
 761 *
 762 * try_to_unlazy attempts to legitimize the current nd->path and nd->root
 763 * for ref-walk mode.
 764 * Must be called from rcu-walk context.
 765 * Nothing should touch nameidata between try_to_unlazy() failure and
 766 * terminate_walk().
 767 */
 768static bool try_to_unlazy(struct nameidata *nd)
 769{
 770	struct dentry *parent = nd->path.dentry;
 771
 772	BUG_ON(!(nd->flags & LOOKUP_RCU));
 773
 774	if (unlikely(!legitimize_links(nd)))
 775		goto out1;
 776	if (unlikely(!legitimize_path(nd, &nd->path, nd->seq)))
 777		goto out;
 778	if (unlikely(!legitimize_root(nd)))
 779		goto out;
 780	leave_rcu(nd);
 781	BUG_ON(nd->inode != parent->d_inode);
 782	return true;
 783
 784out1:
 785	nd->path.mnt = NULL;
 786	nd->path.dentry = NULL;
 787out:
 788	leave_rcu(nd);
 789	return false;
 790}
 791
 792/**
 793 * try_to_unlazy_next - try to switch to ref-walk mode.
 794 * @nd: nameidata pathwalk data
 795 * @dentry: next dentry to step into
 796 * Returns: true on success, false on failure
 797 *
 798 * Similar to try_to_unlazy(), but here we have the next dentry already
 799 * picked by rcu-walk and want to legitimize that in addition to the current
 800 * nd->path and nd->root for ref-walk mode.  Must be called from rcu-walk context.
 801 * Nothing should touch nameidata between try_to_unlazy_next() failure and
 802 * terminate_walk().
 803 */
 804static bool try_to_unlazy_next(struct nameidata *nd, struct dentry *dentry)
 805{
 806	int res;
 807	BUG_ON(!(nd->flags & LOOKUP_RCU));
 808
 809	if (unlikely(!legitimize_links(nd)))
 810		goto out2;
 811	res = __legitimize_mnt(nd->path.mnt, nd->m_seq);
 812	if (unlikely(res)) {
 813		if (res > 0)
 814			goto out2;
 815		goto out1;
 816	}
 817	if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref)))
 818		goto out1;
 819
 820	/*
 821	 * We need to move both the parent and the dentry from the RCU domain
 822	 * to be properly refcounted. And the sequence number in the dentry
 823	 * validates *both* dentry counters, since we checked the sequence
 824	 * number of the parent after we got the child sequence number. So we
 825	 * know the parent must still be valid if the child sequence number is
 826	 */
 827	if (unlikely(!lockref_get_not_dead(&dentry->d_lockref)))
 828		goto out;
 829	if (read_seqcount_retry(&dentry->d_seq, nd->next_seq))
 830		goto out_dput;
 831	/*
 832	 * Sequence counts matched. Now make sure that the root is
 833	 * still valid and get it if required.
 834	 */
 835	if (unlikely(!legitimize_root(nd)))
 836		goto out_dput;
 837	leave_rcu(nd);
 838	return true;
 839
 840out2:
 841	nd->path.mnt = NULL;
 842out1:
 843	nd->path.dentry = NULL;
 844out:
 845	leave_rcu(nd);
 846	return false;
 847out_dput:
 848	leave_rcu(nd);
 849	dput(dentry);
 850	return false;
 851}
 852
 853static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
 854{
 855	if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
 856		return dentry->d_op->d_revalidate(dentry, flags);
 857	else
 858		return 1;
 859}
 860
 861/**
 862 * complete_walk - successful completion of path walk
 863 * @nd:  pointer nameidata
 864 *
 865 * If we had been in RCU mode, drop out of it and legitimize nd->path.
 866 * Revalidate the final result, unless we'd already done that during
 867 * the path walk or the filesystem doesn't ask for it.  Return 0 on
 868 * success, -error on failure.  In case of failure caller does not
 869 * need to drop nd->path.
 870 */
 871static int complete_walk(struct nameidata *nd)
 872{
 873	struct dentry *dentry = nd->path.dentry;
 874	int status;
 875
 876	if (nd->flags & LOOKUP_RCU) {
 877		/*
 878		 * We don't want to zero nd->root for scoped-lookups or
 879		 * externally-managed nd->root.
 880		 */
 881		if (!(nd->state & ND_ROOT_PRESET))
 882			if (!(nd->flags & LOOKUP_IS_SCOPED))
 883				nd->root.mnt = NULL;
 884		nd->flags &= ~LOOKUP_CACHED;
 885		if (!try_to_unlazy(nd))
 886			return -ECHILD;
 887	}
 888
 889	if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) {
 890		/*
 891		 * While the guarantee of LOOKUP_IS_SCOPED is (roughly) "don't
 892		 * ever step outside the root during lookup" and should already
 893		 * be guaranteed by the rest of namei, we want to avoid a namei
 894		 * BUG resulting in userspace being given a path that was not
 895		 * scoped within the root at some point during the lookup.
 896		 *
 897		 * So, do a final sanity-check to make sure that in the
 898		 * worst-case scenario (a complete bypass of LOOKUP_IS_SCOPED)
 899		 * we won't silently return an fd completely outside of the
 900		 * requested root to userspace.
 901		 *
 902		 * Userspace could move the path outside the root after this
 903		 * check, but as discussed elsewhere this is not a concern (the
 904		 * resolved file was inside the root at some point).
 905		 */
 906		if (!path_is_under(&nd->path, &nd->root))
 907			return -EXDEV;
 908	}
 909
 910	if (likely(!(nd->state & ND_JUMPED)))
 911		return 0;
 912
 913	if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
 914		return 0;
 915
 916	status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
 917	if (status > 0)
 918		return 0;
 919
 920	if (!status)
 921		status = -ESTALE;
 922
 923	return status;
 924}
 925
 926static int set_root(struct nameidata *nd)
 927{
 928	struct fs_struct *fs = current->fs;
 929
 930	/*
 931	 * Jumping to the real root in a scoped-lookup is a BUG in namei, but we
 932	 * still have to ensure it doesn't happen because it will cause a breakout
 933	 * from the dirfd.
 934	 */
 935	if (WARN_ON(nd->flags & LOOKUP_IS_SCOPED))
 936		return -ENOTRECOVERABLE;
 937
 938	if (nd->flags & LOOKUP_RCU) {
 939		unsigned seq;
 940
 941		do {
 942			seq = read_seqcount_begin(&fs->seq);
 943			nd->root = fs->root;
 944			nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
 945		} while (read_seqcount_retry(&fs->seq, seq));
 946	} else {
 947		get_fs_root(fs, &nd->root);
 948		nd->state |= ND_ROOT_GRABBED;
 949	}
 950	return 0;
 951}
 952
 953static int nd_jump_root(struct nameidata *nd)
 954{
 955	if (unlikely(nd->flags & LOOKUP_BENEATH))
 956		return -EXDEV;
 957	if (unlikely(nd->flags & LOOKUP_NO_XDEV)) {
 958		/* Absolute path arguments to path_init() are allowed. */
 959		if (nd->path.mnt != NULL && nd->path.mnt != nd->root.mnt)
 960			return -EXDEV;
 961	}
 962	if (!nd->root.mnt) {
 963		int error = set_root(nd);
 964		if (error)
 965			return error;
 966	}
 967	if (nd->flags & LOOKUP_RCU) {
 968		struct dentry *d;
 969		nd->path = nd->root;
 970		d = nd->path.dentry;
 971		nd->inode = d->d_inode;
 972		nd->seq = nd->root_seq;
 973		if (read_seqcount_retry(&d->d_seq, nd->seq))
 974			return -ECHILD;
 975	} else {
 976		path_put(&nd->path);
 977		nd->path = nd->root;
 978		path_get(&nd->path);
 979		nd->inode = nd->path.dentry->d_inode;
 980	}
 981	nd->state |= ND_JUMPED;
 982	return 0;
 983}
 984
 985/*
 986 * Helper to directly jump to a known parsed path from ->get_link,
 987 * caller must have taken a reference to path beforehand.
 988 */
 989int nd_jump_link(const struct path *path)
 990{
 991	int error = -ELOOP;
 992	struct nameidata *nd = current->nameidata;
 993
 994	if (unlikely(nd->flags & LOOKUP_NO_MAGICLINKS))
 995		goto err;
 996
 997	error = -EXDEV;
 998	if (unlikely(nd->flags & LOOKUP_NO_XDEV)) {
 999		if (nd->path.mnt != path->mnt)
1000			goto err;
1001	}
1002	/* Not currently safe for scoped-lookups. */
1003	if (unlikely(nd->flags & LOOKUP_IS_SCOPED))
1004		goto err;
1005
1006	path_put(&nd->path);
1007	nd->path = *path;
1008	nd->inode = nd->path.dentry->d_inode;
1009	nd->state |= ND_JUMPED;
1010	return 0;
1011
1012err:
1013	path_put(path);
1014	return error;
1015}
1016
1017static inline void put_link(struct nameidata *nd)
1018{
1019	struct saved *last = nd->stack + --nd->depth;
1020	do_delayed_call(&last->done);
1021	if (!(nd->flags & LOOKUP_RCU))
1022		path_put(&last->link);
1023}
1024
1025static int sysctl_protected_symlinks __read_mostly;
1026static int sysctl_protected_hardlinks __read_mostly;
1027static int sysctl_protected_fifos __read_mostly;
1028static int sysctl_protected_regular __read_mostly;
1029
1030#ifdef CONFIG_SYSCTL
1031static struct ctl_table namei_sysctls[] = {
1032	{
1033		.procname	= "protected_symlinks",
1034		.data		= &sysctl_protected_symlinks,
1035		.maxlen		= sizeof(int),
1036		.mode		= 0644,
1037		.proc_handler	= proc_dointvec_minmax,
1038		.extra1		= SYSCTL_ZERO,
1039		.extra2		= SYSCTL_ONE,
1040	},
1041	{
1042		.procname	= "protected_hardlinks",
1043		.data		= &sysctl_protected_hardlinks,
1044		.maxlen		= sizeof(int),
1045		.mode		= 0644,
1046		.proc_handler	= proc_dointvec_minmax,
1047		.extra1		= SYSCTL_ZERO,
1048		.extra2		= SYSCTL_ONE,
1049	},
1050	{
1051		.procname	= "protected_fifos",
1052		.data		= &sysctl_protected_fifos,
1053		.maxlen		= sizeof(int),
1054		.mode		= 0644,
1055		.proc_handler	= proc_dointvec_minmax,
1056		.extra1		= SYSCTL_ZERO,
1057		.extra2		= SYSCTL_TWO,
1058	},
1059	{
1060		.procname	= "protected_regular",
1061		.data		= &sysctl_protected_regular,
1062		.maxlen		= sizeof(int),
1063		.mode		= 0644,
1064		.proc_handler	= proc_dointvec_minmax,
1065		.extra1		= SYSCTL_ZERO,
1066		.extra2		= SYSCTL_TWO,
1067	},
1068	{ }
1069};
1070
1071static int __init init_fs_namei_sysctls(void)
1072{
1073	register_sysctl_init("fs", namei_sysctls);
1074	return 0;
1075}
1076fs_initcall(init_fs_namei_sysctls);
1077
1078#endif /* CONFIG_SYSCTL */
1079
1080/**
1081 * may_follow_link - Check symlink following for unsafe situations
1082 * @nd: nameidata pathwalk data
1083 *
1084 * In the case of the sysctl_protected_symlinks sysctl being enabled,
1085 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
1086 * in a sticky world-writable directory. This is to protect privileged
1087 * processes from failing races against path names that may change out
1088 * from under them by way of other users creating malicious symlinks.
1089 * It will permit symlinks to be followed only when outside a sticky
1090 * world-writable directory, or when the uid of the symlink and follower
1091 * match, or when the directory owner matches the symlink's owner.
1092 *
1093 * Returns 0 if following the symlink is allowed, -ve on error.
1094 */
1095static inline int may_follow_link(struct nameidata *nd, const struct inode *inode)
1096{
1097	struct user_namespace *mnt_userns;
1098	vfsuid_t vfsuid;
1099
1100	if (!sysctl_protected_symlinks)
1101		return 0;
1102
1103	mnt_userns = mnt_user_ns(nd->path.mnt);
1104	vfsuid = i_uid_into_vfsuid(mnt_userns, inode);
1105	/* Allowed if owner and follower match. */
1106	if (vfsuid_eq_kuid(vfsuid, current_fsuid()))
1107		return 0;
1108
1109	/* Allowed if parent directory not sticky and world-writable. */
1110	if ((nd->dir_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
1111		return 0;
1112
1113	/* Allowed if parent directory and link owner match. */
1114	if (vfsuid_valid(nd->dir_vfsuid) && vfsuid_eq(nd->dir_vfsuid, vfsuid))
1115		return 0;
1116
1117	if (nd->flags & LOOKUP_RCU)
1118		return -ECHILD;
1119
1120	audit_inode(nd->name, nd->stack[0].link.dentry, 0);
1121	audit_log_path_denied(AUDIT_ANOM_LINK, "follow_link");
1122	return -EACCES;
1123}
1124
1125/**
1126 * safe_hardlink_source - Check for safe hardlink conditions
1127 * @mnt_userns:	user namespace of the mount the inode was found from
1128 * @inode: the source inode to hardlink from
1129 *
1130 * Return false if at least one of the following conditions:
1131 *    - inode is not a regular file
1132 *    - inode is setuid
1133 *    - inode is setgid and group-exec
1134 *    - access failure for read and write
1135 *
1136 * Otherwise returns true.
1137 */
1138static bool safe_hardlink_source(struct user_namespace *mnt_userns,
1139				 struct inode *inode)
1140{
1141	umode_t mode = inode->i_mode;
1142
1143	/* Special files should not get pinned to the filesystem. */
1144	if (!S_ISREG(mode))
1145		return false;
1146
1147	/* Setuid files should not get pinned to the filesystem. */
1148	if (mode & S_ISUID)
1149		return false;
1150
1151	/* Executable setgid files should not get pinned to the filesystem. */
1152	if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
1153		return false;
1154
1155	/* Hardlinking to unreadable or unwritable sources is dangerous. */
1156	if (inode_permission(mnt_userns, inode, MAY_READ | MAY_WRITE))
1157		return false;
1158
1159	return true;
1160}
1161
1162/**
1163 * may_linkat - Check permissions for creating a hardlink
1164 * @mnt_userns:	user namespace of the mount the inode was found from
1165 * @link: the source to hardlink from
1166 *
1167 * Block hardlink when all of:
1168 *  - sysctl_protected_hardlinks enabled
1169 *  - fsuid does not match inode
1170 *  - hardlink source is unsafe (see safe_hardlink_source() above)
1171 *  - not CAP_FOWNER in a namespace with the inode owner uid mapped
1172 *
1173 * If the inode has been found through an idmapped mount the user namespace of
1174 * the vfsmount must be passed through @mnt_userns. This function will then take
1175 * care to map the inode according to @mnt_userns before checking permissions.
1176 * On non-idmapped mounts or if permission checking is to be performed on the
1177 * raw inode simply passs init_user_ns.
1178 *
1179 * Returns 0 if successful, -ve on error.
1180 */
1181int may_linkat(struct user_namespace *mnt_userns, const struct path *link)
1182{
1183	struct inode *inode = link->dentry->d_inode;
1184
1185	/* Inode writeback is not safe when the uid or gid are invalid. */
1186	if (!vfsuid_valid(i_uid_into_vfsuid(mnt_userns, inode)) ||
1187	    !vfsgid_valid(i_gid_into_vfsgid(mnt_userns, inode)))
1188		return -EOVERFLOW;
1189
1190	if (!sysctl_protected_hardlinks)
1191		return 0;
1192
1193	/* Source inode owner (or CAP_FOWNER) can hardlink all they like,
1194	 * otherwise, it must be a safe source.
1195	 */
1196	if (safe_hardlink_source(mnt_userns, inode) ||
1197	    inode_owner_or_capable(mnt_userns, inode))
1198		return 0;
1199
1200	audit_log_path_denied(AUDIT_ANOM_LINK, "linkat");
1201	return -EPERM;
1202}
1203
1204/**
1205 * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory
1206 *			  should be allowed, or not, on files that already
1207 *			  exist.
1208 * @mnt_userns:	user namespace of the mount the inode was found from
1209 * @nd: nameidata pathwalk data
1210 * @inode: the inode of the file to open
1211 *
1212 * Block an O_CREAT open of a FIFO (or a regular file) when:
1213 *   - sysctl_protected_fifos (or sysctl_protected_regular) is enabled
1214 *   - the file already exists
1215 *   - we are in a sticky directory
1216 *   - we don't own the file
1217 *   - the owner of the directory doesn't own the file
1218 *   - the directory is world writable
1219 * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2
1220 * the directory doesn't have to be world writable: being group writable will
1221 * be enough.
1222 *
1223 * If the inode has been found through an idmapped mount the user namespace of
1224 * the vfsmount must be passed through @mnt_userns. This function will then take
1225 * care to map the inode according to @mnt_userns before checking permissions.
1226 * On non-idmapped mounts or if permission checking is to be performed on the
1227 * raw inode simply passs init_user_ns.
1228 *
1229 * Returns 0 if the open is allowed, -ve on error.
1230 */
1231static int may_create_in_sticky(struct user_namespace *mnt_userns,
1232				struct nameidata *nd, struct inode *const inode)
1233{
1234	umode_t dir_mode = nd->dir_mode;
1235	vfsuid_t dir_vfsuid = nd->dir_vfsuid;
1236
1237	if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) ||
1238	    (!sysctl_protected_regular && S_ISREG(inode->i_mode)) ||
1239	    likely(!(dir_mode & S_ISVTX)) ||
1240	    vfsuid_eq(i_uid_into_vfsuid(mnt_userns, inode), dir_vfsuid) ||
1241	    vfsuid_eq_kuid(i_uid_into_vfsuid(mnt_userns, inode), current_fsuid()))
1242		return 0;
1243
1244	if (likely(dir_mode & 0002) ||
1245	    (dir_mode & 0020 &&
1246	     ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) ||
1247	      (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) {
1248		const char *operation = S_ISFIFO(inode->i_mode) ?
1249					"sticky_create_fifo" :
1250					"sticky_create_regular";
1251		audit_log_path_denied(AUDIT_ANOM_CREAT, operation);
1252		return -EACCES;
1253	}
1254	return 0;
1255}
1256
1257/*
1258 * follow_up - Find the mountpoint of path's vfsmount
1259 *
1260 * Given a path, find the mountpoint of its source file system.
1261 * Replace @path with the path of the mountpoint in the parent mount.
1262 * Up is towards /.
1263 *
1264 * Return 1 if we went up a level and 0 if we were already at the
1265 * root.
1266 */
1267int follow_up(struct path *path)
1268{
1269	struct mount *mnt = real_mount(path->mnt);
1270	struct mount *parent;
1271	struct dentry *mountpoint;
1272
1273	read_seqlock_excl(&mount_lock);
1274	parent = mnt->mnt_parent;
1275	if (parent == mnt) {
1276		read_sequnlock_excl(&mount_lock);
1277		return 0;
1278	}
1279	mntget(&parent->mnt);
1280	mountpoint = dget(mnt->mnt_mountpoint);
1281	read_sequnlock_excl(&mount_lock);
1282	dput(path->dentry);
1283	path->dentry = mountpoint;
1284	mntput(path->mnt);
1285	path->mnt = &parent->mnt;
1286	return 1;
1287}
1288EXPORT_SYMBOL(follow_up);
1289
1290static bool choose_mountpoint_rcu(struct mount *m, const struct path *root,
1291				  struct path *path, unsigned *seqp)
1292{
1293	while (mnt_has_parent(m)) {
1294		struct dentry *mountpoint = m->mnt_mountpoint;
1295
1296		m = m->mnt_parent;
1297		if (unlikely(root->dentry == mountpoint &&
1298			     root->mnt == &m->mnt))
1299			break;
1300		if (mountpoint != m->mnt.mnt_root) {
1301			path->mnt = &m->mnt;
1302			path->dentry = mountpoint;
1303			*seqp = read_seqcount_begin(&mountpoint->d_seq);
1304			return true;
1305		}
1306	}
1307	return false;
1308}
1309
1310static bool choose_mountpoint(struct mount *m, const struct path *root,
1311			      struct path *path)
1312{
1313	bool found;
1314
1315	rcu_read_lock();
1316	while (1) {
1317		unsigned seq, mseq = read_seqbegin(&mount_lock);
1318
1319		found = choose_mountpoint_rcu(m, root, path, &seq);
1320		if (unlikely(!found)) {
1321			if (!read_seqretry(&mount_lock, mseq))
1322				break;
1323		} else {
1324			if (likely(__legitimize_path(path, seq, mseq)))
1325				break;
1326			rcu_read_unlock();
1327			path_put(path);
1328			rcu_read_lock();
1329		}
1330	}
1331	rcu_read_unlock();
1332	return found;
1333}
1334
1335/*
1336 * Perform an automount
1337 * - return -EISDIR to tell follow_managed() to stop and return the path we
1338 *   were called with.
1339 */
1340static int follow_automount(struct path *path, int *count, unsigned lookup_flags)
1341{
1342	struct dentry *dentry = path->dentry;
1343
1344	/* We don't want to mount if someone's just doing a stat -
1345	 * unless they're stat'ing a directory and appended a '/' to
1346	 * the name.
1347	 *
1348	 * We do, however, want to mount if someone wants to open or
1349	 * create a file of any type under the mountpoint, wants to
1350	 * traverse through the mountpoint or wants to open the
1351	 * mounted directory.  Also, autofs may mark negative dentries
1352	 * as being automount points.  These will need the attentions
1353	 * of the daemon to instantiate them before they can be used.
1354	 */
1355	if (!(lookup_flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1356			   LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1357	    dentry->d_inode)
1358		return -EISDIR;
1359
1360	if (count && (*count)++ >= MAXSYMLINKS)
1361		return -ELOOP;
1362
1363	return finish_automount(dentry->d_op->d_automount(path), path);
1364}
1365
1366/*
1367 * mount traversal - out-of-line part.  One note on ->d_flags accesses -
1368 * dentries are pinned but not locked here, so negative dentry can go
1369 * positive right under us.  Use of smp_load_acquire() provides a barrier
1370 * sufficient for ->d_inode and ->d_flags consistency.
1371 */
1372static int __traverse_mounts(struct path *path, unsigned flags, bool *jumped,
1373			     int *count, unsigned lookup_flags)
1374{
1375	struct vfsmount *mnt = path->mnt;
1376	bool need_mntput = false;
1377	int ret = 0;
1378
1379	while (flags & DCACHE_MANAGED_DENTRY) {
1380		/* Allow the filesystem to manage the transit without i_mutex
1381		 * being held. */
1382		if (flags & DCACHE_MANAGE_TRANSIT) {
1383			ret = path->dentry->d_op->d_manage(path, false);
1384			flags = smp_load_acquire(&path->dentry->d_flags);
1385			if (ret < 0)
1386				break;
1387		}
1388
1389		if (flags & DCACHE_MOUNTED) {	// something's mounted on it..
1390			struct vfsmount *mounted = lookup_mnt(path);
1391			if (mounted) {		// ... in our namespace
1392				dput(path->dentry);
1393				if (need_mntput)
1394					mntput(path->mnt);
1395				path->mnt = mounted;
1396				path->dentry = dget(mounted->mnt_root);
1397				// here we know it's positive
1398				flags = path->dentry->d_flags;
1399				need_mntput = true;
1400				continue;
1401			}
1402		}
1403
1404		if (!(flags & DCACHE_NEED_AUTOMOUNT))
1405			break;
1406
1407		// uncovered automount point
1408		ret = follow_automount(path, count, lookup_flags);
1409		flags = smp_load_acquire(&path->dentry->d_flags);
1410		if (ret < 0)
1411			break;
1412	}
1413
1414	if (ret == -EISDIR)
1415		ret = 0;
1416	// possible if you race with several mount --move
1417	if (need_mntput && path->mnt == mnt)
1418		mntput(path->mnt);
1419	if (!ret && unlikely(d_flags_negative(flags)))
1420		ret = -ENOENT;
1421	*jumped = need_mntput;
1422	return ret;
1423}
1424
1425static inline int traverse_mounts(struct path *path, bool *jumped,
1426				  int *count, unsigned lookup_flags)
1427{
1428	unsigned flags = smp_load_acquire(&path->dentry->d_flags);
1429
1430	/* fastpath */
1431	if (likely(!(flags & DCACHE_MANAGED_DENTRY))) {
1432		*jumped = false;
1433		if (unlikely(d_flags_negative(flags)))
1434			return -ENOENT;
1435		return 0;
1436	}
1437	return __traverse_mounts(path, flags, jumped, count, lookup_flags);
1438}
1439
1440int follow_down_one(struct path *path)
1441{
1442	struct vfsmount *mounted;
1443
1444	mounted = lookup_mnt(path);
1445	if (mounted) {
1446		dput(path->dentry);
1447		mntput(path->mnt);
1448		path->mnt = mounted;
1449		path->dentry = dget(mounted->mnt_root);
1450		return 1;
1451	}
1452	return 0;
1453}
1454EXPORT_SYMBOL(follow_down_one);
1455
1456/*
1457 * Follow down to the covering mount currently visible to userspace.  At each
1458 * point, the filesystem owning that dentry may be queried as to whether the
1459 * caller is permitted to proceed or not.
1460 */
1461int follow_down(struct path *path)
1462{
1463	struct vfsmount *mnt = path->mnt;
1464	bool jumped;
1465	int ret = traverse_mounts(path, &jumped, NULL, 0);
1466
1467	if (path->mnt != mnt)
1468		mntput(mnt);
1469	return ret;
1470}
1471EXPORT_SYMBOL(follow_down);
1472
1473/*
1474 * Try to skip to top of mountpoint pile in rcuwalk mode.  Fail if
1475 * we meet a managed dentry that would need blocking.
1476 */
1477static bool __follow_mount_rcu(struct nameidata *nd, struct path *path)
1478{
1479	struct dentry *dentry = path->dentry;
1480	unsigned int flags = dentry->d_flags;
1481
1482	if (likely(!(flags & DCACHE_MANAGED_DENTRY)))
1483		return true;
1484
1485	if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1486		return false;
1487
1488	for (;;) {
1489		/*
1490		 * Don't forget we might have a non-mountpoint managed dentry
1491		 * that wants to block transit.
1492		 */
1493		if (unlikely(flags & DCACHE_MANAGE_TRANSIT)) {
1494			int res = dentry->d_op->d_manage(path, true);
1495			if (res)
1496				return res == -EISDIR;
1497			flags = dentry->d_flags;
1498		}
1499
1500		if (flags & DCACHE_MOUNTED) {
1501			struct mount *mounted = __lookup_mnt(path->mnt, dentry);
1502			if (mounted) {
1503				path->mnt = &mounted->mnt;
1504				dentry = path->dentry = mounted->mnt.mnt_root;
1505				nd->state |= ND_JUMPED;
1506				nd->next_seq = read_seqcount_begin(&dentry->d_seq);
1507				flags = dentry->d_flags;
1508				// makes sure that non-RCU pathwalk could reach
1509				// this state.
1510				if (read_seqretry(&mount_lock, nd->m_seq))
1511					return false;
1512				continue;
1513			}
1514			if (read_seqretry(&mount_lock, nd->m_seq))
1515				return false;
1516		}
1517		return !(flags & DCACHE_NEED_AUTOMOUNT);
1518	}
1519}
1520
1521static inline int handle_mounts(struct nameidata *nd, struct dentry *dentry,
1522			  struct path *path)
1523{
1524	bool jumped;
1525	int ret;
1526
1527	path->mnt = nd->path.mnt;
1528	path->dentry = dentry;
1529	if (nd->flags & LOOKUP_RCU) {
1530		unsigned int seq = nd->next_seq;
1531		if (likely(__follow_mount_rcu(nd, path)))
1532			return 0;
1533		// *path and nd->next_seq might've been clobbered
1534		path->mnt = nd->path.mnt;
1535		path->dentry = dentry;
1536		nd->next_seq = seq;
1537		if (!try_to_unlazy_next(nd, dentry))
1538			return -ECHILD;
1539	}
1540	ret = traverse_mounts(path, &jumped, &nd->total_link_count, nd->flags);
1541	if (jumped) {
1542		if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1543			ret = -EXDEV;
1544		else
1545			nd->state |= ND_JUMPED;
1546	}
1547	if (unlikely(ret)) {
1548		dput(path->dentry);
1549		if (path->mnt != nd->path.mnt)
1550			mntput(path->mnt);
1551	}
1552	return ret;
1553}
1554
1555/*
1556 * This looks up the name in dcache and possibly revalidates the found dentry.
1557 * NULL is returned if the dentry does not exist in the cache.
1558 */
1559static struct dentry *lookup_dcache(const struct qstr *name,
1560				    struct dentry *dir,
1561				    unsigned int flags)
1562{
1563	struct dentry *dentry = d_lookup(dir, name);
1564	if (dentry) {
1565		int error = d_revalidate(dentry, flags);
1566		if (unlikely(error <= 0)) {
1567			if (!error)
1568				d_invalidate(dentry);
1569			dput(dentry);
1570			return ERR_PTR(error);
1571		}
1572	}
1573	return dentry;
1574}
1575
1576/*
1577 * Parent directory has inode locked exclusive.  This is one
1578 * and only case when ->lookup() gets called on non in-lookup
1579 * dentries - as the matter of fact, this only gets called
1580 * when directory is guaranteed to have no in-lookup children
1581 * at all.
1582 */
1583static struct dentry *__lookup_hash(const struct qstr *name,
1584		struct dentry *base, unsigned int flags)
1585{
1586	struct dentry *dentry = lookup_dcache(name, base, flags);
1587	struct dentry *old;
1588	struct inode *dir = base->d_inode;
1589
1590	if (dentry)
1591		return dentry;
1592
1593	/* Don't create child dentry for a dead directory. */
1594	if (unlikely(IS_DEADDIR(dir)))
1595		return ERR_PTR(-ENOENT);
1596
1597	dentry = d_alloc(base, name);
1598	if (unlikely(!dentry))
1599		return ERR_PTR(-ENOMEM);
1600
1601	old = dir->i_op->lookup(dir, dentry, flags);
1602	if (unlikely(old)) {
1603		dput(dentry);
1604		dentry = old;
1605	}
1606	return dentry;
1607}
1608
1609static struct dentry *lookup_fast(struct nameidata *nd)
1610{
1611	struct dentry *dentry, *parent = nd->path.dentry;
1612	int status = 1;
1613
1614	/*
1615	 * Rename seqlock is not required here because in the off chance
1616	 * of a false negative due to a concurrent rename, the caller is
1617	 * going to fall back to non-racy lookup.
1618	 */
1619	if (nd->flags & LOOKUP_RCU) {
1620		dentry = __d_lookup_rcu(parent, &nd->last, &nd->next_seq);
1621		if (unlikely(!dentry)) {
1622			if (!try_to_unlazy(nd))
1623				return ERR_PTR(-ECHILD);
1624			return NULL;
1625		}
1626
1627		/*
1628		 * This sequence count validates that the parent had no
1629		 * changes while we did the lookup of the dentry above.
1630		 */
1631		if (read_seqcount_retry(&parent->d_seq, nd->seq))
1632			return ERR_PTR(-ECHILD);
1633
1634		status = d_revalidate(dentry, nd->flags);
1635		if (likely(status > 0))
1636			return dentry;
1637		if (!try_to_unlazy_next(nd, dentry))
1638			return ERR_PTR(-ECHILD);
1639		if (status == -ECHILD)
1640			/* we'd been told to redo it in non-rcu mode */
1641			status = d_revalidate(dentry, nd->flags);
1642	} else {
1643		dentry = __d_lookup(parent, &nd->last);
1644		if (unlikely(!dentry))
1645			return NULL;
1646		status = d_revalidate(dentry, nd->flags);
1647	}
1648	if (unlikely(status <= 0)) {
1649		if (!status)
1650			d_invalidate(dentry);
1651		dput(dentry);
1652		return ERR_PTR(status);
1653	}
1654	return dentry;
1655}
1656
1657/* Fast lookup failed, do it the slow way */
1658static struct dentry *__lookup_slow(const struct qstr *name,
1659				    struct dentry *dir,
1660				    unsigned int flags)
1661{
1662	struct dentry *dentry, *old;
1663	struct inode *inode = dir->d_inode;
1664	DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1665
1666	/* Don't go there if it's already dead */
1667	if (unlikely(IS_DEADDIR(inode)))
1668		return ERR_PTR(-ENOENT);
1669again:
1670	dentry = d_alloc_parallel(dir, name, &wq);
1671	if (IS_ERR(dentry))
1672		return dentry;
1673	if (unlikely(!d_in_lookup(dentry))) {
1674		int error = d_revalidate(dentry, flags);
1675		if (unlikely(error <= 0)) {
1676			if (!error) {
1677				d_invalidate(dentry);
1678				dput(dentry);
1679				goto again;
1680			}
1681			dput(dentry);
1682			dentry = ERR_PTR(error);
1683		}
1684	} else {
1685		old = inode->i_op->lookup(inode, dentry, flags);
1686		d_lookup_done(dentry);
1687		if (unlikely(old)) {
1688			dput(dentry);
1689			dentry = old;
1690		}
1691	}
1692	return dentry;
1693}
1694
1695static struct dentry *lookup_slow(const struct qstr *name,
1696				  struct dentry *dir,
1697				  unsigned int flags)
1698{
1699	struct inode *inode = dir->d_inode;
1700	struct dentry *res;
1701	inode_lock_shared(inode);
1702	res = __lookup_slow(name, dir, flags);
1703	inode_unlock_shared(inode);
1704	return res;
1705}
1706
1707static inline int may_lookup(struct user_namespace *mnt_userns,
1708			     struct nameidata *nd)
1709{
1710	if (nd->flags & LOOKUP_RCU) {
1711		int err = inode_permission(mnt_userns, nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1712		if (err != -ECHILD || !try_to_unlazy(nd))
1713			return err;
1714	}
1715	return inode_permission(mnt_userns, nd->inode, MAY_EXEC);
1716}
1717
1718static int reserve_stack(struct nameidata *nd, struct path *link)
1719{
1720	if (unlikely(nd->total_link_count++ >= MAXSYMLINKS))
1721		return -ELOOP;
1722
1723	if (likely(nd->depth != EMBEDDED_LEVELS))
1724		return 0;
1725	if (likely(nd->stack != nd->internal))
1726		return 0;
1727	if (likely(nd_alloc_stack(nd)))
1728		return 0;
1729
1730	if (nd->flags & LOOKUP_RCU) {
1731		// we need to grab link before we do unlazy.  And we can't skip
1732		// unlazy even if we fail to grab the link - cleanup needs it
1733		bool grabbed_link = legitimize_path(nd, link, nd->next_seq);
1734
1735		if (!try_to_unlazy(nd) || !grabbed_link)
1736			return -ECHILD;
1737
1738		if (nd_alloc_stack(nd))
1739			return 0;
1740	}
1741	return -ENOMEM;
1742}
1743
1744enum {WALK_TRAILING = 1, WALK_MORE = 2, WALK_NOFOLLOW = 4};
1745
1746static const char *pick_link(struct nameidata *nd, struct path *link,
1747		     struct inode *inode, int flags)
1748{
1749	struct saved *last;
1750	const char *res;
1751	int error = reserve_stack(nd, link);
1752
1753	if (unlikely(error)) {
1754		if (!(nd->flags & LOOKUP_RCU))
1755			path_put(link);
1756		return ERR_PTR(error);
1757	}
1758	last = nd->stack + nd->depth++;
1759	last->link = *link;
1760	clear_delayed_call(&last->done);
1761	last->seq = nd->next_seq;
1762
1763	if (flags & WALK_TRAILING) {
1764		error = may_follow_link(nd, inode);
1765		if (unlikely(error))
1766			return ERR_PTR(error);
1767	}
1768
1769	if (unlikely(nd->flags & LOOKUP_NO_SYMLINKS) ||
1770			unlikely(link->mnt->mnt_flags & MNT_NOSYMFOLLOW))
1771		return ERR_PTR(-ELOOP);
1772
1773	if (!(nd->flags & LOOKUP_RCU)) {
1774		touch_atime(&last->link);
1775		cond_resched();
1776	} else if (atime_needs_update(&last->link, inode)) {
1777		if (!try_to_unlazy(nd))
1778			return ERR_PTR(-ECHILD);
1779		touch_atime(&last->link);
1780	}
1781
1782	error = security_inode_follow_link(link->dentry, inode,
1783					   nd->flags & LOOKUP_RCU);
1784	if (unlikely(error))
1785		return ERR_PTR(error);
1786
1787	res = READ_ONCE(inode->i_link);
1788	if (!res) {
1789		const char * (*get)(struct dentry *, struct inode *,
1790				struct delayed_call *);
1791		get = inode->i_op->get_link;
1792		if (nd->flags & LOOKUP_RCU) {
1793			res = get(NULL, inode, &last->done);
1794			if (res == ERR_PTR(-ECHILD) && try_to_unlazy(nd))
1795				res = get(link->dentry, inode, &last->done);
1796		} else {
1797			res = get(link->dentry, inode, &last->done);
1798		}
1799		if (!res)
1800			goto all_done;
1801		if (IS_ERR(res))
1802			return res;
1803	}
1804	if (*res == '/') {
1805		error = nd_jump_root(nd);
1806		if (unlikely(error))
1807			return ERR_PTR(error);
1808		while (unlikely(*++res == '/'))
1809			;
1810	}
1811	if (*res)
1812		return res;
1813all_done: // pure jump
1814	put_link(nd);
1815	return NULL;
1816}
1817
1818/*
1819 * Do we need to follow links? We _really_ want to be able
1820 * to do this check without having to look at inode->i_op,
1821 * so we keep a cache of "no, this doesn't need follow_link"
1822 * for the common case.
1823 *
1824 * NOTE: dentry must be what nd->next_seq had been sampled from.
1825 */
1826static const char *step_into(struct nameidata *nd, int flags,
1827		     struct dentry *dentry)
1828{
1829	struct path path;
1830	struct inode *inode;
1831	int err = handle_mounts(nd, dentry, &path);
1832
1833	if (err < 0)
1834		return ERR_PTR(err);
1835	inode = path.dentry->d_inode;
1836	if (likely(!d_is_symlink(path.dentry)) ||
1837	   ((flags & WALK_TRAILING) && !(nd->flags & LOOKUP_FOLLOW)) ||
1838	   (flags & WALK_NOFOLLOW)) {
1839		/* not a symlink or should not follow */
1840		if (nd->flags & LOOKUP_RCU) {
1841			if (read_seqcount_retry(&path.dentry->d_seq, nd->next_seq))
1842				return ERR_PTR(-ECHILD);
1843			if (unlikely(!inode))
1844				return ERR_PTR(-ENOENT);
1845		} else {
1846			dput(nd->path.dentry);
1847			if (nd->path.mnt != path.mnt)
1848				mntput(nd->path.mnt);
1849		}
1850		nd->path = path;
1851		nd->inode = inode;
1852		nd->seq = nd->next_seq;
1853		return NULL;
1854	}
1855	if (nd->flags & LOOKUP_RCU) {
1856		/* make sure that d_is_symlink above matches inode */
1857		if (read_seqcount_retry(&path.dentry->d_seq, nd->next_seq))
1858			return ERR_PTR(-ECHILD);
1859	} else {
1860		if (path.mnt == nd->path.mnt)
1861			mntget(path.mnt);
1862	}
1863	return pick_link(nd, &path, inode, flags);
1864}
1865
1866static struct dentry *follow_dotdot_rcu(struct nameidata *nd)
1867{
1868	struct dentry *parent, *old;
1869
1870	if (path_equal(&nd->path, &nd->root))
1871		goto in_root;
1872	if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) {
1873		struct path path;
1874		unsigned seq;
1875		if (!choose_mountpoint_rcu(real_mount(nd->path.mnt),
1876					   &nd->root, &path, &seq))
1877			goto in_root;
1878		if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1879			return ERR_PTR(-ECHILD);
1880		nd->path = path;
1881		nd->inode = path.dentry->d_inode;
1882		nd->seq = seq;
1883		// makes sure that non-RCU pathwalk could reach this state
1884		if (read_seqretry(&mount_lock, nd->m_seq))
1885			return ERR_PTR(-ECHILD);
1886		/* we know that mountpoint was pinned */
1887	}
1888	old = nd->path.dentry;
1889	parent = old->d_parent;
1890	nd->next_seq = read_seqcount_begin(&parent->d_seq);
1891	// makes sure that non-RCU pathwalk could reach this state
1892	if (read_seqcount_retry(&old->d_seq, nd->seq))
1893		return ERR_PTR(-ECHILD);
1894	if (unlikely(!path_connected(nd->path.mnt, parent)))
1895		return ERR_PTR(-ECHILD);
1896	return parent;
1897in_root:
1898	if (read_seqretry(&mount_lock, nd->m_seq))
1899		return ERR_PTR(-ECHILD);
1900	if (unlikely(nd->flags & LOOKUP_BENEATH))
1901		return ERR_PTR(-ECHILD);
1902	nd->next_seq = nd->seq;
1903	return nd->path.dentry;
1904}
1905
1906static struct dentry *follow_dotdot(struct nameidata *nd)
1907{
1908	struct dentry *parent;
1909
1910	if (path_equal(&nd->path, &nd->root))
1911		goto in_root;
1912	if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) {
1913		struct path path;
1914
1915		if (!choose_mountpoint(real_mount(nd->path.mnt),
1916				       &nd->root, &path))
1917			goto in_root;
1918		path_put(&nd->path);
1919		nd->path = path;
1920		nd->inode = path.dentry->d_inode;
1921		if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1922			return ERR_PTR(-EXDEV);
1923	}
1924	/* rare case of legitimate dget_parent()... */
1925	parent = dget_parent(nd->path.dentry);
1926	if (unlikely(!path_connected(nd->path.mnt, parent))) {
1927		dput(parent);
1928		return ERR_PTR(-ENOENT);
1929	}
1930	return parent;
1931
1932in_root:
1933	if (unlikely(nd->flags & LOOKUP_BENEATH))
1934		return ERR_PTR(-EXDEV);
1935	return dget(nd->path.dentry);
1936}
1937
1938static const char *handle_dots(struct nameidata *nd, int type)
1939{
1940	if (type == LAST_DOTDOT) {
1941		const char *error = NULL;
1942		struct dentry *parent;
1943
1944		if (!nd->root.mnt) {
1945			error = ERR_PTR(set_root(nd));
1946			if (error)
1947				return error;
1948		}
1949		if (nd->flags & LOOKUP_RCU)
1950			parent = follow_dotdot_rcu(nd);
1951		else
1952			parent = follow_dotdot(nd);
1953		if (IS_ERR(parent))
1954			return ERR_CAST(parent);
1955		error = step_into(nd, WALK_NOFOLLOW, parent);
1956		if (unlikely(error))
1957			return error;
1958
1959		if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) {
1960			/*
1961			 * If there was a racing rename or mount along our
1962			 * path, then we can't be sure that ".." hasn't jumped
1963			 * above nd->root (and so userspace should retry or use
1964			 * some fallback).
1965			 */
1966			smp_rmb();
1967			if (__read_seqcount_retry(&mount_lock.seqcount, nd->m_seq))
1968				return ERR_PTR(-EAGAIN);
1969			if (__read_seqcount_retry(&rename_lock.seqcount, nd->r_seq))
1970				return ERR_PTR(-EAGAIN);
1971		}
1972	}
1973	return NULL;
1974}
1975
1976static const char *walk_component(struct nameidata *nd, int flags)
1977{
1978	struct dentry *dentry;
1979	/*
1980	 * "." and ".." are special - ".." especially so because it has
1981	 * to be able to know about the current root directory and
1982	 * parent relationships.
1983	 */
1984	if (unlikely(nd->last_type != LAST_NORM)) {
1985		if (!(flags & WALK_MORE) && nd->depth)
1986			put_link(nd);
1987		return handle_dots(nd, nd->last_type);
1988	}
1989	dentry = lookup_fast(nd);
1990	if (IS_ERR(dentry))
1991		return ERR_CAST(dentry);
1992	if (unlikely(!dentry)) {
1993		dentry = lookup_slow(&nd->last, nd->path.dentry, nd->flags);
1994		if (IS_ERR(dentry))
1995			return ERR_CAST(dentry);
1996	}
1997	if (!(flags & WALK_MORE) && nd->depth)
1998		put_link(nd);
1999	return step_into(nd, flags, dentry);
2000}
2001
2002/*
2003 * We can do the critical dentry name comparison and hashing
2004 * operations one word at a time, but we are limited to:
2005 *
2006 * - Architectures with fast unaligned word accesses. We could
2007 *   do a "get_unaligned()" if this helps and is sufficiently
2008 *   fast.
2009 *
2010 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
2011 *   do not trap on the (extremely unlikely) case of a page
2012 *   crossing operation.
2013 *
2014 * - Furthermore, we need an efficient 64-bit compile for the
2015 *   64-bit case in order to generate the "number of bytes in
2016 *   the final mask". Again, that could be replaced with a
2017 *   efficient population count instruction or similar.
2018 */
2019#ifdef CONFIG_DCACHE_WORD_ACCESS
2020
2021#include <asm/word-at-a-time.h>
2022
2023#ifdef HASH_MIX
2024
2025/* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
2026
2027#elif defined(CONFIG_64BIT)
2028/*
2029 * Register pressure in the mixing function is an issue, particularly
2030 * on 32-bit x86, but almost any function requires one state value and
2031 * one temporary.  Instead, use a function designed for two state values
2032 * and no temporaries.
2033 *
2034 * This function cannot create a collision in only two iterations, so
2035 * we have two iterations to achieve avalanche.  In those two iterations,
2036 * we have six layers of mixing, which is enough to spread one bit's
2037 * influence out to 2^6 = 64 state bits.
2038 *
2039 * Rotate constants are scored by considering either 64 one-bit input
2040 * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
2041 * probability of that delta causing a change to each of the 128 output
2042 * bits, using a sample of random initial states.
2043 *
2044 * The Shannon entropy of the computed probabilities is then summed
2045 * to produce a score.  Ideally, any input change has a 50% chance of
2046 * toggling any given output bit.
2047 *
2048 * Mixing scores (in bits) for (12,45):
2049 * Input delta: 1-bit      2-bit
2050 * 1 round:     713.3    42542.6
2051 * 2 rounds:   2753.7   140389.8
2052 * 3 rounds:   5954.1   233458.2
2053 * 4 rounds:   7862.6   256672.2
2054 * Perfect:    8192     258048
2055 *            (64*128) (64*63/2 * 128)
2056 */
2057#define HASH_MIX(x, y, a)	\
2058	(	x ^= (a),	\
2059	y ^= x,	x = rol64(x,12),\
2060	x += y,	y = rol64(y,45),\
2061	y *= 9			)
2062
2063/*
2064 * Fold two longs into one 32-bit hash value.  This must be fast, but
2065 * latency isn't quite as critical, as there is a fair bit of additional
2066 * work done before the hash value is used.
2067 */
2068static inline unsigned int fold_hash(unsigned long x, unsigned long y)
2069{
2070	y ^= x * GOLDEN_RATIO_64;
2071	y *= GOLDEN_RATIO_64;
2072	return y >> 32;
2073}
2074
2075#else	/* 32-bit case */
2076
2077/*
2078 * Mixing scores (in bits) for (7,20):
2079 * Input delta: 1-bit      2-bit
2080 * 1 round:     330.3     9201.6
2081 * 2 rounds:   1246.4    25475.4
2082 * 3 rounds:   1907.1    31295.1
2083 * 4 rounds:   2042.3    31718.6
2084 * Perfect:    2048      31744
2085 *            (32*64)   (32*31/2 * 64)
2086 */
2087#define HASH_MIX(x, y, a)	\
2088	(	x ^= (a),	\
2089	y ^= x,	x = rol32(x, 7),\
2090	x += y,	y = rol32(y,20),\
2091	y *= 9			)
2092
2093static inline unsigned int fold_hash(unsigned long x, unsigned long y)
2094{
2095	/* Use arch-optimized multiply if one exists */
2096	return __hash_32(y ^ __hash_32(x));
2097}
2098
2099#endif
2100
2101/*
2102 * Return the hash of a string of known length.  This is carfully
2103 * designed to match hash_name(), which is the more critical function.
2104 * In particular, we must end by hashing a final word containing 0..7
2105 * payload bytes, to match the way that hash_name() iterates until it
2106 * finds the delimiter after the name.
2107 */
2108unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2109{
2110	unsigned long a, x = 0, y = (unsigned long)salt;
2111
2112	for (;;) {
2113		if (!len)
2114			goto done;
2115		a = load_unaligned_zeropad(name);
2116		if (len < sizeof(unsigned long))
2117			break;
2118		HASH_MIX(x, y, a);
2119		name += sizeof(unsigned long);
2120		len -= sizeof(unsigned long);
2121	}
2122	x ^= a & bytemask_from_count(len);
2123done:
2124	return fold_hash(x, y);
2125}
2126EXPORT_SYMBOL(full_name_hash);
2127
2128/* Return the "hash_len" (hash and length) of a null-terminated string */
2129u64 hashlen_string(const void *salt, const char *name)
2130{
2131	unsigned long a = 0, x = 0, y = (unsigned long)salt;
2132	unsigned long adata, mask, len;
2133	const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2134
2135	len = 0;
2136	goto inside;
2137
2138	do {
2139		HASH_MIX(x, y, a);
2140		len += sizeof(unsigned long);
2141inside:
2142		a = load_unaligned_zeropad(name+len);
2143	} while (!has_zero(a, &adata, &constants));
2144
2145	adata = prep_zero_mask(a, adata, &constants);
2146	mask = create_zero_mask(adata);
2147	x ^= a & zero_bytemask(mask);
2148
2149	return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2150}
2151EXPORT_SYMBOL(hashlen_string);
2152
2153/*
2154 * Calculate the length and hash of the path component, and
2155 * return the "hash_len" as the result.
2156 */
2157static inline u64 hash_name(const void *salt, const char *name)
2158{
2159	unsigned long a = 0, b, x = 0, y = (unsigned long)salt;
2160	unsigned long adata, bdata, mask, len;
2161	const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2162
2163	len = 0;
2164	goto inside;
2165
2166	do {
2167		HASH_MIX(x, y, a);
2168		len += sizeof(unsigned long);
2169inside:
2170		a = load_unaligned_zeropad(name+len);
2171		b = a ^ REPEAT_BYTE('/');
2172	} while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
2173
2174	adata = prep_zero_mask(a, adata, &constants);
2175	bdata = prep_zero_mask(b, bdata, &constants);
2176	mask = create_zero_mask(adata | bdata);
2177	x ^= a & zero_bytemask(mask);
2178
2179	return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2180}
2181
2182#else	/* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
2183
2184/* Return the hash of a string of known length */
2185unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2186{
2187	unsigned long hash = init_name_hash(salt);
2188	while (len--)
2189		hash = partial_name_hash((unsigned char)*name++, hash);
2190	return end_name_hash(hash);
2191}
2192EXPORT_SYMBOL(full_name_hash);
2193
2194/* Return the "hash_len" (hash and length) of a null-terminated string */
2195u64 hashlen_string(const void *salt, const char *name)
2196{
2197	unsigned long hash = init_name_hash(salt);
2198	unsigned long len = 0, c;
2199
2200	c = (unsigned char)*name;
2201	while (c) {
2202		len++;
2203		hash = partial_name_hash(c, hash);
2204		c = (unsigned char)name[len];
2205	}
2206	return hashlen_create(end_name_hash(hash), len);
2207}
2208EXPORT_SYMBOL(hashlen_string);
2209
2210/*
2211 * We know there's a real path component here of at least
2212 * one character.
2213 */
2214static inline u64 hash_name(const void *salt, const char *name)
2215{
2216	unsigned long hash = init_name_hash(salt);
2217	unsigned long len = 0, c;
2218
2219	c = (unsigned char)*name;
2220	do {
2221		len++;
2222		hash = partial_name_hash(c, hash);
2223		c = (unsigned char)name[len];
2224	} while (c && c != '/');
2225	return hashlen_create(end_name_hash(hash), len);
2226}
2227
2228#endif
2229
2230/*
2231 * Name resolution.
2232 * This is the basic name resolution function, turning a pathname into
2233 * the final dentry. We expect 'base' to be positive and a directory.
2234 *
2235 * Returns 0 and nd will have valid dentry and mnt on success.
2236 * Returns error and drops reference to input namei data on failure.
2237 */
2238static int link_path_walk(const char *name, struct nameidata *nd)
2239{
2240	int depth = 0; // depth <= nd->depth
2241	int err;
2242
2243	nd->last_type = LAST_ROOT;
2244	nd->flags |= LOOKUP_PARENT;
2245	if (IS_ERR(name))
2246		return PTR_ERR(name);
2247	while (*name=='/')
2248		name++;
2249	if (!*name) {
2250		nd->dir_mode = 0; // short-circuit the 'hardening' idiocy
2251		return 0;
2252	}
2253
2254	/* At this point we know we have a real path component. */
2255	for(;;) {
2256		struct user_namespace *mnt_userns;
2257		const char *link;
2258		u64 hash_len;
2259		int type;
2260
2261		mnt_userns = mnt_user_ns(nd->path.mnt);
2262		err = may_lookup(mnt_userns, nd);
2263		if (err)
2264			return err;
2265
2266		hash_len = hash_name(nd->path.dentry, name);
2267
2268		type = LAST_NORM;
2269		if (name[0] == '.') switch (hashlen_len(hash_len)) {
2270			case 2:
2271				if (name[1] == '.') {
2272					type = LAST_DOTDOT;
2273					nd->state |= ND_JUMPED;
2274				}
2275				break;
2276			case 1:
2277				type = LAST_DOT;
2278		}
2279		if (likely(type == LAST_NORM)) {
2280			struct dentry *parent = nd->path.dentry;
2281			nd->state &= ~ND_JUMPED;
2282			if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
2283				struct qstr this = { { .hash_len = hash_len }, .name = name };
2284				err = parent->d_op->d_hash(parent, &this);
2285				if (err < 0)
2286					return err;
2287				hash_len = this.hash_len;
2288				name = this.name;
2289			}
2290		}
2291
2292		nd->last.hash_len = hash_len;
2293		nd->last.name = name;
2294		nd->last_type = type;
2295
2296		name += hashlen_len(hash_len);
2297		if (!*name)
2298			goto OK;
2299		/*
2300		 * If it wasn't NUL, we know it was '/'. Skip that
2301		 * slash, and continue until no more slashes.
2302		 */
2303		do {
2304			name++;
2305		} while (unlikely(*name == '/'));
2306		if (unlikely(!*name)) {
2307OK:
2308			/* pathname or trailing symlink, done */
2309			if (!depth) {
2310				nd->dir_vfsuid = i_uid_into_vfsuid(mnt_userns, nd->inode);
2311				nd->dir_mode = nd->inode->i_mode;
2312				nd->flags &= ~LOOKUP_PARENT;
2313				return 0;
2314			}
2315			/* last component of nested symlink */
2316			name = nd->stack[--depth].name;
2317			link = walk_component(nd, 0);
2318		} else {
2319			/* not the last component */
2320			link = walk_component(nd, WALK_MORE);
2321		}
2322		if (unlikely(link)) {
2323			if (IS_ERR(link))
2324				return PTR_ERR(link);
2325			/* a symlink to follow */
2326			nd->stack[depth++].name = name;
2327			name = link;
2328			continue;
2329		}
2330		if (unlikely(!d_can_lookup(nd->path.dentry))) {
2331			if (nd->flags & LOOKUP_RCU) {
2332				if (!try_to_unlazy(nd))
2333					return -ECHILD;
2334			}
2335			return -ENOTDIR;
2336		}
2337	}
2338}
2339
2340/* must be paired with terminate_walk() */
2341static const char *path_init(struct nameidata *nd, unsigned flags)
2342{
2343	int error;
2344	const char *s = nd->name->name;
2345
2346	/* LOOKUP_CACHED requires RCU, ask caller to retry */
2347	if ((flags & (LOOKUP_RCU | LOOKUP_CACHED)) == LOOKUP_CACHED)
2348		return ERR_PTR(-EAGAIN);
2349
2350	if (!*s)
2351		flags &= ~LOOKUP_RCU;
2352	if (flags & LOOKUP_RCU)
2353		rcu_read_lock();
2354	else
2355		nd->seq = nd->next_seq = 0;
2356
2357	nd->flags = flags;
2358	nd->state |= ND_JUMPED;
2359
2360	nd->m_seq = __read_seqcount_begin(&mount_lock.seqcount);
2361	nd->r_seq = __read_seqcount_begin(&rename_lock.seqcount);
2362	smp_rmb();
2363
2364	if (nd->state & ND_ROOT_PRESET) {
2365		struct dentry *root = nd->root.dentry;
2366		struct inode *inode = root->d_inode;
2367		if (*s && unlikely(!d_can_lookup(root)))
2368			return ERR_PTR(-ENOTDIR);
2369		nd->path = nd->root;
2370		nd->inode = inode;
2371		if (flags & LOOKUP_RCU) {
2372			nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2373			nd->root_seq = nd->seq;
2374		} else {
2375			path_get(&nd->path);
2376		}
2377		return s;
2378	}
2379
2380	nd->root.mnt = NULL;
2381
2382	/* Absolute pathname -- fetch the root (LOOKUP_IN_ROOT uses nd->dfd). */
2383	if (*s == '/' && !(flags & LOOKUP_IN_ROOT)) {
2384		error = nd_jump_root(nd);
2385		if (unlikely(error))
2386			return ERR_PTR(error);
2387		return s;
2388	}
2389
2390	/* Relative pathname -- get the starting-point it is relative to. */
2391	if (nd->dfd == AT_FDCWD) {
2392		if (flags & LOOKUP_RCU) {
2393			struct fs_struct *fs = current->fs;
2394			unsigned seq;
2395
2396			do {
2397				seq = read_seqcount_begin(&fs->seq);
2398				nd->path = fs->pwd;
2399				nd->inode = nd->path.dentry->d_inode;
2400				nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2401			} while (read_seqcount_retry(&fs->seq, seq));
2402		} else {
2403			get_fs_pwd(current->fs, &nd->path);
2404			nd->inode = nd->path.dentry->d_inode;
2405		}
2406	} else {
2407		/* Caller must check execute permissions on the starting path component */
2408		struct fd f = fdget_raw(nd->dfd);
2409		struct dentry *dentry;
2410
2411		if (!f.file)
2412			return ERR_PTR(-EBADF);
2413
2414		dentry = f.file->f_path.dentry;
2415
2416		if (*s && unlikely(!d_can_lookup(dentry))) {
2417			fdput(f);
2418			return ERR_PTR(-ENOTDIR);
2419		}
2420
2421		nd->path = f.file->f_path;
2422		if (flags & LOOKUP_RCU) {
2423			nd->inode = nd->path.dentry->d_inode;
2424			nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2425		} else {
2426			path_get(&nd->path);
2427			nd->inode = nd->path.dentry->d_inode;
2428		}
2429		fdput(f);
2430	}
2431
2432	/* For scoped-lookups we need to set the root to the dirfd as well. */
2433	if (flags & LOOKUP_IS_SCOPED) {
2434		nd->root = nd->path;
2435		if (flags & LOOKUP_RCU) {
2436			nd->root_seq = nd->seq;
2437		} else {
2438			path_get(&nd->root);
2439			nd->state |= ND_ROOT_GRABBED;
2440		}
2441	}
2442	return s;
2443}
2444
2445static inline const char *lookup_last(struct nameidata *nd)
2446{
2447	if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2448		nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2449
2450	return walk_component(nd, WALK_TRAILING);
2451}
2452
2453static int handle_lookup_down(struct nameidata *nd)
2454{
2455	if (!(nd->flags & LOOKUP_RCU))
2456		dget(nd->path.dentry);
2457	nd->next_seq = nd->seq;
2458	return PTR_ERR(step_into(nd, WALK_NOFOLLOW, nd->path.dentry));
2459}
2460
2461/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2462static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2463{
2464	const char *s = path_init(nd, flags);
2465	int err;
2466
2467	if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) {
2468		err = handle_lookup_down(nd);
2469		if (unlikely(err < 0))
2470			s = ERR_PTR(err);
2471	}
2472
2473	while (!(err = link_path_walk(s, nd)) &&
2474	       (s = lookup_last(nd)) != NULL)
2475		;
2476	if (!err && unlikely(nd->flags & LOOKUP_MOUNTPOINT)) {
2477		err = handle_lookup_down(nd);
2478		nd->state &= ~ND_JUMPED; // no d_weak_revalidate(), please...
2479	}
2480	if (!err)
2481		err = complete_walk(nd);
2482
2483	if (!err && nd->flags & LOOKUP_DIRECTORY)
2484		if (!d_can_lookup(nd->path.dentry))
2485			err = -ENOTDIR;
2486	if (!err) {
2487		*path = nd->path;
2488		nd->path.mnt = NULL;
2489		nd->path.dentry = NULL;
2490	}
2491	terminate_walk(nd);
2492	return err;
2493}
2494
2495int filename_lookup(int dfd, struct filename *name, unsigned flags,
2496		    struct path *path, struct path *root)
2497{
2498	int retval;
2499	struct nameidata nd;
2500	if (IS_ERR(name))
2501		return PTR_ERR(name);
2502	set_nameidata(&nd, dfd, name, root);
2503	retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
2504	if (unlikely(retval == -ECHILD))
2505		retval = path_lookupat(&nd, flags, path);
2506	if (unlikely(retval == -ESTALE))
2507		retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
2508
2509	if (likely(!retval))
2510		audit_inode(name, path->dentry,
2511			    flags & LOOKUP_MOUNTPOINT ? AUDIT_INODE_NOEVAL : 0);
2512	restore_nameidata();
2513	return retval;
2514}
2515
2516/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2517static int path_parentat(struct nameidata *nd, unsigned flags,
2518				struct path *parent)
2519{
2520	const char *s = path_init(nd, flags);
2521	int err = link_path_walk(s, nd);
2522	if (!err)
2523		err = complete_walk(nd);
2524	if (!err) {
2525		*parent = nd->path;
2526		nd->path.mnt = NULL;
2527		nd->path.dentry = NULL;
2528	}
2529	terminate_walk(nd);
2530	return err;
2531}
2532
2533/* Note: this does not consume "name" */
2534static int filename_parentat(int dfd, struct filename *name,
2535			     unsigned int flags, struct path *parent,
2536			     struct qstr *last, int *type)
2537{
2538	int retval;
2539	struct nameidata nd;
2540
2541	if (IS_ERR(name))
2542		return PTR_ERR(name);
2543	set_nameidata(&nd, dfd, name, NULL);
2544	retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
2545	if (unlikely(retval == -ECHILD))
2546		retval = path_parentat(&nd, flags, parent);
2547	if (unlikely(retval == -ESTALE))
2548		retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
2549	if (likely(!retval)) {
2550		*last = nd.last;
2551		*type = nd.last_type;
2552		audit_inode(name, parent->dentry, AUDIT_INODE_PARENT);
2553	}
2554	restore_nameidata();
2555	return retval;
2556}
2557
2558/* does lookup, returns the object with parent locked */
2559static struct dentry *__kern_path_locked(struct filename *name, struct path *path)
2560{
2561	struct dentry *d;
2562	struct qstr last;
2563	int type, error;
2564
2565	error = filename_parentat(AT_FDCWD, name, 0, path, &last, &type);
2566	if (error)
2567		return ERR_PTR(error);
2568	if (unlikely(type != LAST_NORM)) {
2569		path_put(path);
2570		return ERR_PTR(-EINVAL);
2571	}
2572	inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
2573	d = __lookup_hash(&last, path->dentry, 0);
2574	if (IS_ERR(d)) {
2575		inode_unlock(path->dentry->d_inode);
2576		path_put(path);
2577	}
2578	return d;
2579}
2580
2581struct dentry *kern_path_locked(const char *name, struct path *path)
2582{
2583	struct filename *filename = getname_kernel(name);
2584	struct dentry *res = __kern_path_locked(filename, path);
2585
2586	putname(filename);
2587	return res;
2588}
2589
2590int kern_path(const char *name, unsigned int flags, struct path *path)
2591{
2592	struct filename *filename = getname_kernel(name);
2593	int ret = filename_lookup(AT_FDCWD, filename, flags, path, NULL);
2594
2595	putname(filename);
2596	return ret;
2597
2598}
2599EXPORT_SYMBOL(kern_path);
2600
2601/**
2602 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2603 * @dentry:  pointer to dentry of the base directory
2604 * @mnt: pointer to vfs mount of the base directory
2605 * @name: pointer to file name
2606 * @flags: lookup flags
2607 * @path: pointer to struct path to fill
2608 */
2609int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2610		    const char *name, unsigned int flags,
2611		    struct path *path)
2612{
2613	struct filename *filename;
2614	struct path root = {.mnt = mnt, .dentry = dentry};
2615	int ret;
2616
2617	filename = getname_kernel(name);
2618	/* the first argument of filename_lookup() is ignored with root */
2619	ret = filename_lookup(AT_FDCWD, filename, flags, path, &root);
2620	putname(filename);
2621	return ret;
2622}
2623EXPORT_SYMBOL(vfs_path_lookup);
2624
2625static int lookup_one_common(struct user_namespace *mnt_userns,
2626			     const char *name, struct dentry *base, int len,
2627			     struct qstr *this)
2628{
2629	this->name = name;
2630	this->len = len;
2631	this->hash = full_name_hash(base, name, len);
2632	if (!len)
2633		return -EACCES;
2634
2635	if (unlikely(name[0] == '.')) {
2636		if (len < 2 || (len == 2 && name[1] == '.'))
2637			return -EACCES;
2638	}
2639
2640	while (len--) {
2641		unsigned int c = *(const unsigned char *)name++;
2642		if (c == '/' || c == '\0')
2643			return -EACCES;
2644	}
2645	/*
2646	 * See if the low-level filesystem might want
2647	 * to use its own hash..
2648	 */
2649	if (base->d_flags & DCACHE_OP_HASH) {
2650		int err = base->d_op->d_hash(base, this);
2651		if (err < 0)
2652			return err;
2653	}
2654
2655	return inode_permission(mnt_userns, base->d_inode, MAY_EXEC);
2656}
2657
2658/**
2659 * try_lookup_one_len - filesystem helper to lookup single pathname component
2660 * @name:	pathname component to lookup
2661 * @base:	base directory to lookup from
2662 * @len:	maximum length @len should be interpreted to
2663 *
2664 * Look up a dentry by name in the dcache, returning NULL if it does not
2665 * currently exist.  The function does not try to create a dentry.
2666 *
2667 * Note that this routine is purely a helper for filesystem usage and should
2668 * not be called by generic code.
2669 *
2670 * The caller must hold base->i_mutex.
2671 */
2672struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len)
2673{
2674	struct qstr this;
2675	int err;
2676
2677	WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2678
2679	err = lookup_one_common(&init_user_ns, name, base, len, &this);
2680	if (err)
2681		return ERR_PTR(err);
2682
2683	return lookup_dcache(&this, base, 0);
2684}
2685EXPORT_SYMBOL(try_lookup_one_len);
2686
2687/**
2688 * lookup_one_len - filesystem helper to lookup single pathname component
2689 * @name:	pathname component to lookup
2690 * @base:	base directory to lookup from
2691 * @len:	maximum length @len should be interpreted to
2692 *
2693 * Note that this routine is purely a helper for filesystem usage and should
2694 * not be called by generic code.
2695 *
2696 * The caller must hold base->i_mutex.
2697 */
2698struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2699{
2700	struct dentry *dentry;
2701	struct qstr this;
2702	int err;
2703
2704	WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2705
2706	err = lookup_one_common(&init_user_ns, name, base, len, &this);
2707	if (err)
2708		return ERR_PTR(err);
2709
2710	dentry = lookup_dcache(&this, base, 0);
2711	return dentry ? dentry : __lookup_slow(&this, base, 0);
2712}
2713EXPORT_SYMBOL(lookup_one_len);
2714
2715/**
2716 * lookup_one - filesystem helper to lookup single pathname component
2717 * @mnt_userns:	user namespace of the mount the lookup is performed from
2718 * @name:	pathname component to lookup
2719 * @base:	base directory to lookup from
2720 * @len:	maximum length @len should be interpreted to
2721 *
2722 * Note that this routine is purely a helper for filesystem usage and should
2723 * not be called by generic code.
2724 *
2725 * The caller must hold base->i_mutex.
2726 */
2727struct dentry *lookup_one(struct user_namespace *mnt_userns, const char *name,
2728			  struct dentry *base, int len)
2729{
2730	struct dentry *dentry;
2731	struct qstr this;
2732	int err;
2733
2734	WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2735
2736	err = lookup_one_common(mnt_userns, name, base, len, &this);
2737	if (err)
2738		return ERR_PTR(err);
2739
2740	dentry = lookup_dcache(&this, base, 0);
2741	return dentry ? dentry : __lookup_slow(&this, base, 0);
2742}
2743EXPORT_SYMBOL(lookup_one);
2744
2745/**
2746 * lookup_one_unlocked - filesystem helper to lookup single pathname component
2747 * @mnt_userns:	idmapping of the mount the lookup is performed from
2748 * @name:	pathname component to lookup
2749 * @base:	base directory to lookup from
2750 * @len:	maximum length @len should be interpreted to
2751 *
2752 * Note that this routine is purely a helper for filesystem usage and should
2753 * not be called by generic code.
2754 *
2755 * Unlike lookup_one_len, it should be called without the parent
2756 * i_mutex held, and will take the i_mutex itself if necessary.
2757 */
2758struct dentry *lookup_one_unlocked(struct user_namespace *mnt_userns,
2759				   const char *name, struct dentry *base,
2760				   int len)
2761{
2762	struct qstr this;
2763	int err;
2764	struct dentry *ret;
2765
2766	err = lookup_one_common(mnt_userns, name, base, len, &this);
2767	if (err)
2768		return ERR_PTR(err);
2769
2770	ret = lookup_dcache(&this, base, 0);
2771	if (!ret)
2772		ret = lookup_slow(&this, base, 0);
2773	return ret;
2774}
2775EXPORT_SYMBOL(lookup_one_unlocked);
2776
2777/**
2778 * lookup_one_positive_unlocked - filesystem helper to lookup single
2779 *				  pathname component
2780 * @mnt_userns:	idmapping of the mount the lookup is performed from
2781 * @name:	pathname component to lookup
2782 * @base:	base directory to lookup from
2783 * @len:	maximum length @len should be interpreted to
2784 *
2785 * This helper will yield ERR_PTR(-ENOENT) on negatives. The helper returns
2786 * known positive or ERR_PTR(). This is what most of the users want.
2787 *
2788 * Note that pinned negative with unlocked parent _can_ become positive at any
2789 * time, so callers of lookup_one_unlocked() need to be very careful; pinned
2790 * positives have >d_inode stable, so this one avoids such problems.
2791 *
2792 * Note that this routine is purely a helper for filesystem usage and should
2793 * not be called by generic code.
2794 *
2795 * The helper should be called without i_mutex held.
2796 */
2797struct dentry *lookup_one_positive_unlocked(struct user_namespace *mnt_userns,
2798					    const char *name,
2799					    struct dentry *base, int len)
2800{
2801	struct dentry *ret = lookup_one_unlocked(mnt_userns, name, base, len);
2802
2803	if (!IS_ERR(ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) {
2804		dput(ret);
2805		ret = ERR_PTR(-ENOENT);
2806	}
2807	return ret;
2808}
2809EXPORT_SYMBOL(lookup_one_positive_unlocked);
2810
2811/**
2812 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2813 * @name:	pathname component to lookup
2814 * @base:	base directory to lookup from
2815 * @len:	maximum length @len should be interpreted to
2816 *
2817 * Note that this routine is purely a helper for filesystem usage and should
2818 * not be called by generic code.
2819 *
2820 * Unlike lookup_one_len, it should be called without the parent
2821 * i_mutex held, and will take the i_mutex itself if necessary.
2822 */
2823struct dentry *lookup_one_len_unlocked(const char *name,
2824				       struct dentry *base, int len)
2825{
2826	return lookup_one_unlocked(&init_user_ns, name, base, len);
2827}
2828EXPORT_SYMBOL(lookup_one_len_unlocked);
2829
2830/*
2831 * Like lookup_one_len_unlocked(), except that it yields ERR_PTR(-ENOENT)
2832 * on negatives.  Returns known positive or ERR_PTR(); that's what
2833 * most of the users want.  Note that pinned negative with unlocked parent
2834 * _can_ become positive at any time, so callers of lookup_one_len_unlocked()
2835 * need to be very careful; pinned positives have ->d_inode stable, so
2836 * this one avoids such problems.
2837 */
2838struct dentry *lookup_positive_unlocked(const char *name,
2839				       struct dentry *base, int len)
2840{
2841	return lookup_one_positive_unlocked(&init_user_ns, name, base, len);
2842}
2843EXPORT_SYMBOL(lookup_positive_unlocked);
2844
2845#ifdef CONFIG_UNIX98_PTYS
2846int path_pts(struct path *path)
2847{
2848	/* Find something mounted on "pts" in the same directory as
2849	 * the input path.
2850	 */
2851	struct dentry *parent = dget_parent(path->dentry);
2852	struct dentry *child;
2853	struct qstr this = QSTR_INIT("pts", 3);
2854
2855	if (unlikely(!path_connected(path->mnt, parent))) {
2856		dput(parent);
2857		return -ENOENT;
2858	}
2859	dput(path->dentry);
2860	path->dentry = parent;
2861	child = d_hash_and_lookup(parent, &this);
2862	if (!child)
2863		return -ENOENT;
2864
2865	path->dentry = child;
2866	dput(parent);
2867	follow_down(path);
2868	return 0;
2869}
2870#endif
2871
2872int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2873		 struct path *path, int *empty)
2874{
2875	struct filename *filename = getname_flags(name, flags, empty);
2876	int ret = filename_lookup(dfd, filename, flags, path, NULL);
2877
2878	putname(filename);
2879	return ret;
2880}
2881EXPORT_SYMBOL(user_path_at_empty);
2882
2883int __check_sticky(struct user_namespace *mnt_userns, struct inode *dir,
2884		   struct inode *inode)
2885{
2886	kuid_t fsuid = current_fsuid();
2887
2888	if (vfsuid_eq_kuid(i_uid_into_vfsuid(mnt_userns, inode), fsuid))
2889		return 0;
2890	if (vfsuid_eq_kuid(i_uid_into_vfsuid(mnt_userns, dir), fsuid))
2891		return 0;
2892	return !capable_wrt_inode_uidgid(mnt_userns, inode, CAP_FOWNER);
2893}
2894EXPORT_SYMBOL(__check_sticky);
2895
2896/*
2897 *	Check whether we can remove a link victim from directory dir, check
2898 *  whether the type of victim is right.
2899 *  1. We can't do it if dir is read-only (done in permission())
2900 *  2. We should have write and exec permissions on dir
2901 *  3. We can't remove anything from append-only dir
2902 *  4. We can't do anything with immutable dir (done in permission())
2903 *  5. If the sticky bit on dir is set we should either
2904 *	a. be owner of dir, or
2905 *	b. be owner of victim, or
2906 *	c. have CAP_FOWNER capability
2907 *  6. If the victim is append-only or immutable we can't do antyhing with
2908 *     links pointing to it.
2909 *  7. If the victim has an unknown uid or gid we can't change the inode.
2910 *  8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2911 *  9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2912 * 10. We can't remove a root or mountpoint.
2913 * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
2914 *     nfs_async_unlink().
2915 */
2916static int may_delete(struct user_namespace *mnt_userns, struct inode *dir,
2917		      struct dentry *victim, bool isdir)
2918{
2919	struct inode *inode = d_backing_inode(victim);
2920	int error;
2921
2922	if (d_is_negative(victim))
2923		return -ENOENT;
2924	BUG_ON(!inode);
2925
2926	BUG_ON(victim->d_parent->d_inode != dir);
2927
2928	/* Inode writeback is not safe when the uid or gid are invalid. */
2929	if (!vfsuid_valid(i_uid_into_vfsuid(mnt_userns, inode)) ||
2930	    !vfsgid_valid(i_gid_into_vfsgid(mnt_userns, inode)))
2931		return -EOVERFLOW;
2932
2933	audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2934
2935	error = inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC);
2936	if (error)
2937		return error;
2938	if (IS_APPEND(dir))
2939		return -EPERM;
2940
2941	if (check_sticky(mnt_userns, dir, inode) || IS_APPEND(inode) ||
2942	    IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) ||
2943	    HAS_UNMAPPED_ID(mnt_userns, inode))
2944		return -EPERM;
2945	if (isdir) {
2946		if (!d_is_dir(victim))
2947			return -ENOTDIR;
2948		if (IS_ROOT(victim))
2949			return -EBUSY;
2950	} else if (d_is_dir(victim))
2951		return -EISDIR;
2952	if (IS_DEADDIR(dir))
2953		return -ENOENT;
2954	if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2955		return -EBUSY;
2956	return 0;
2957}
2958
2959/*	Check whether we can create an object with dentry child in directory
2960 *  dir.
2961 *  1. We can't do it if child already exists (open has special treatment for
2962 *     this case, but since we are inlined it's OK)
2963 *  2. We can't do it if dir is read-only (done in permission())
2964 *  3. We can't do it if the fs can't represent the fsuid or fsgid.
2965 *  4. We should have write and exec permissions on dir
2966 *  5. We can't do it if dir is immutable (done in permission())
2967 */
2968static inline int may_create(struct user_namespace *mnt_userns,
2969			     struct inode *dir, struct dentry *child)
2970{
2971	audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2972	if (child->d_inode)
2973		return -EEXIST;
2974	if (IS_DEADDIR(dir))
2975		return -ENOENT;
2976	if (!fsuidgid_has_mapping(dir->i_sb, mnt_userns))
2977		return -EOVERFLOW;
2978
2979	return inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC);
2980}
2981
2982/*
2983 * p1 and p2 should be directories on the same fs.
2984 */
2985struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2986{
2987	struct dentry *p;
2988
2989	if (p1 == p2) {
2990		inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2991		return NULL;
2992	}
2993
2994	mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
2995
2996	p = d_ancestor(p2, p1);
2997	if (p) {
2998		inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
2999		inode_lock_nested(p1->d_inode, I_MUTEX_CHILD);
3000		return p;
3001	}
3002
3003	p = d_ancestor(p1, p2);
3004	if (p) {
3005		inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
3006		inode_lock_nested(p2->d_inode, I_MUTEX_CHILD);
3007		return p;
3008	}
3009
3010	inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
3011	inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
3012	return NULL;
3013}
3014EXPORT_SYMBOL(lock_rename);
3015
3016void unlock_rename(struct dentry *p1, struct dentry *p2)
3017{
3018	inode_unlock(p1->d_inode);
3019	if (p1 != p2) {
3020		inode_unlock(p2->d_inode);
3021		mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
3022	}
3023}
3024EXPORT_SYMBOL(unlock_rename);
3025
3026/**
3027 * mode_strip_umask - handle vfs umask stripping
3028 * @dir:	parent directory of the new inode
3029 * @mode:	mode of the new inode to be created in @dir
3030 *
3031 * Umask stripping depends on whether or not the filesystem supports POSIX
3032 * ACLs. If the filesystem doesn't support it umask stripping is done directly
3033 * in here. If the filesystem does support POSIX ACLs umask stripping is
3034 * deferred until the filesystem calls posix_acl_create().
3035 *
3036 * Returns: mode
3037 */
3038static inline umode_t mode_strip_umask(const struct inode *dir, umode_t mode)
3039{
3040	if (!IS_POSIXACL(dir))
3041		mode &= ~current_umask();
3042	return mode;
3043}
3044
3045/**
3046 * vfs_prepare_mode - prepare the mode to be used for a new inode
3047 * @mnt_userns:		user namespace of the mount the inode was found from
3048 * @dir:	parent directory of the new inode
3049 * @mode:	mode of the new inode
3050 * @mask_perms:	allowed permission by the vfs
3051 * @type:	type of file to be created
3052 *
3053 * This helper consolidates and enforces vfs restrictions on the @mode of a new
3054 * object to be created.
3055 *
3056 * Umask stripping depends on whether the filesystem supports POSIX ACLs (see
3057 * the kernel documentation for mode_strip_umask()). Moving umask stripping
3058 * after setgid stripping allows the same ordering for both non-POSIX ACL and
3059 * POSIX ACL supporting filesystems.
3060 *
3061 * Note that it's currently valid for @type to be 0 if a directory is created.
3062 * Filesystems raise that flag individually and we need to check whether each
3063 * filesystem can deal with receiving S_IFDIR from the vfs before we enforce a
3064 * non-zero type.
3065 *
3066 * Returns: mode to be passed to the filesystem
3067 */
3068static inline umode_t vfs_prepare_mode(struct user_namespace *mnt_userns,
3069				       const struct inode *dir, umode_t mode,
3070				       umode_t mask_perms, umode_t type)
3071{
3072	mode = mode_strip_sgid(mnt_userns, dir, mode);
3073	mode = mode_strip_umask(dir, mode);
3074
3075	/*
3076	 * Apply the vfs mandated allowed permission mask and set the type of
3077	 * file to be created before we call into the filesystem.
3078	 */
3079	mode &= (mask_perms & ~S_IFMT);
3080	mode |= (type & S_IFMT);
3081
3082	return mode;
3083}
3084
3085/**
3086 * vfs_create - create new file
3087 * @mnt_userns:	user namespace of the mount the inode was found from
3088 * @dir:	inode of @dentry
3089 * @dentry:	pointer to dentry of the base directory
3090 * @mode:	mode of the new file
3091 * @want_excl:	whether the file must not yet exist
3092 *
3093 * Create a new file.
3094 *
3095 * If the inode has been found through an idmapped mount the user namespace of
3096 * the vfsmount must be passed through @mnt_userns. This function will then take
3097 * care to map the inode according to @mnt_userns before checking permissions.
3098 * On non-idmapped mounts or if permission checking is to be performed on the
3099 * raw inode simply passs init_user_ns.
3100 */
3101int vfs_create(struct user_namespace *mnt_userns, struct inode *dir,
3102	       struct dentry *dentry, umode_t mode, bool want_excl)
3103{
3104	int error = may_create(mnt_userns, dir, dentry);
3105	if (error)
3106		return error;
3107
3108	if (!dir->i_op->create)
3109		return -EACCES;	/* shouldn't it be ENOSYS? */
3110
3111	mode = vfs_prepare_mode(mnt_userns, dir, mode, S_IALLUGO, S_IFREG);
3112	error = security_inode_create(dir, dentry, mode);
3113	if (error)
3114		return error;
3115	error = dir->i_op->create(mnt_userns, dir, dentry, mode, want_excl);
3116	if (!error)
3117		fsnotify_create(dir, dentry);
3118	return error;
3119}
3120EXPORT_SYMBOL(vfs_create);
3121
3122int vfs_mkobj(struct dentry *dentry, umode_t mode,
3123		int (*f)(struct dentry *, umode_t, void *),
3124		void *arg)
3125{
3126	struct inode *dir = dentry->d_parent->d_inode;
3127	int error = may_create(&init_user_ns, dir, dentry);
3128	if (error)
3129		return error;
3130
3131	mode &= S_IALLUGO;
3132	mode |= S_IFREG;
3133	error = security_inode_create(dir, dentry, mode);
3134	if (error)
3135		return error;
3136	error = f(dentry, mode, arg);
3137	if (!error)
3138		fsnotify_create(dir, dentry);
3139	return error;
3140}
3141EXPORT_SYMBOL(vfs_mkobj);
3142
3143bool may_open_dev(const struct path *path)
3144{
3145	return !(path->mnt->mnt_flags & MNT_NODEV) &&
3146		!(path->mnt->mnt_sb->s_iflags & SB_I_NODEV);
3147}
3148
3149static int may_open(struct user_namespace *mnt_userns, const struct path *path,
3150		    int acc_mode, int flag)
3151{
3152	struct dentry *dentry = path->dentry;
3153	struct inode *inode = dentry->d_inode;
3154	int error;
3155
3156	if (!inode)
3157		return -ENOENT;
3158
3159	switch (inode->i_mode & S_IFMT) {
3160	case S_IFLNK:
3161		return -ELOOP;
3162	case S_IFDIR:
3163		if (acc_mode & MAY_WRITE)
3164			return -EISDIR;
3165		if (acc_mode & MAY_EXEC)
3166			return -EACCES;
3167		break;
3168	case S_IFBLK:
3169	case S_IFCHR:
3170		if (!may_open_dev(path))
3171			return -EACCES;
3172		fallthrough;
3173	case S_IFIFO:
3174	case S_IFSOCK:
3175		if (acc_mode & MAY_EXEC)
3176			return -EACCES;
3177		flag &= ~O_TRUNC;
3178		break;
3179	case S_IFREG:
3180		if ((acc_mode & MAY_EXEC) && path_noexec(path))
3181			return -EACCES;
3182		break;
3183	}
3184
3185	error = inode_permission(mnt_userns, inode, MAY_OPEN | acc_mode);
3186	if (error)
3187		return error;
3188
3189	/*
3190	 * An append-only file must be opened in append mode for writing.
3191	 */
3192	if (IS_APPEND(inode)) {
3193		if  ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
3194			return -EPERM;
3195		if (flag & O_TRUNC)
3196			return -EPERM;
3197	}
3198
3199	/* O_NOATIME can only be set by the owner or superuser */
3200	if (flag & O_NOATIME && !inode_owner_or_capable(mnt_userns, inode))
3201		return -EPERM;
3202
3203	return 0;
3204}
3205
3206static int handle_truncate(struct user_namespace *mnt_userns, struct file *filp)
3207{
3208	const struct path *path = &filp->f_path;
3209	struct inode *inode = path->dentry->d_inode;
3210	int error = get_write_access(inode);
3211	if (error)
3212		return error;
3213
3214	error = security_file_truncate(filp);
3215	if (!error) {
3216		error = do_truncate(mnt_userns, path->dentry, 0,
3217				    ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
3218				    filp);
3219	}
3220	put_write_access(inode);
3221	return error;
3222}
3223
3224static inline int open_to_namei_flags(int flag)
3225{
3226	if ((flag & O_ACCMODE) == 3)
3227		flag--;
3228	return flag;
3229}
3230
3231static int may_o_create(struct user_namespace *mnt_userns,
3232			const struct path *dir, struct dentry *dentry,
3233			umode_t mode)
3234{
3235	int error = security_path_mknod(dir, dentry, mode, 0);
3236	if (error)
3237		return error;
3238
3239	if (!fsuidgid_has_mapping(dir->dentry->d_sb, mnt_userns))
3240		return -EOVERFLOW;
3241
3242	error = inode_permission(mnt_userns, dir->dentry->d_inode,
3243				 MAY_WRITE | MAY_EXEC);
3244	if (error)
3245		return error;
3246
3247	return security_inode_create(dir->dentry->d_inode, dentry, mode);
3248}
3249
3250/*
3251 * Attempt to atomically look up, create and open a file from a negative
3252 * dentry.
3253 *
3254 * Returns 0 if successful.  The file will have been created and attached to
3255 * @file by the filesystem calling finish_open().
3256 *
3257 * If the file was looked up only or didn't need creating, FMODE_OPENED won't
3258 * be set.  The caller will need to perform the open themselves.  @path will
3259 * have been updated to point to the new dentry.  This may be negative.
3260 *
3261 * Returns an error code otherwise.
3262 */
3263static struct dentry *atomic_open(struct nameidata *nd, struct dentry *dentry,
3264				  struct file *file,
3265				  int open_flag, umode_t mode)
3266{
3267	struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
3268	struct inode *dir =  nd->path.dentry->d_inode;
3269	int error;
3270
3271	if (nd->flags & LOOKUP_DIRECTORY)
3272		open_flag |= O_DIRECTORY;
3273
3274	file->f_path.dentry = DENTRY_NOT_SET;
3275	file->f_path.mnt = nd->path.mnt;
3276	error = dir->i_op->atomic_open(dir, dentry, file,
3277				       open_to_namei_flags(open_flag), mode);
3278	d_lookup_done(dentry);
3279	if (!error) {
3280		if (file->f_mode & FMODE_OPENED) {
3281			if (unlikely(dentry != file->f_path.dentry)) {
3282				dput(dentry);
3283				dentry = dget(file->f_path.dentry);
3284			}
3285		} else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
3286			error = -EIO;
3287		} else {
3288			if (file->f_path.dentry) {
3289				dput(dentry);
3290				dentry = file->f_path.dentry;
3291			}
3292			if (unlikely(d_is_negative(dentry)))
3293				error = -ENOENT;
3294		}
3295	}
3296	if (error) {
3297		dput(dentry);
3298		dentry = ERR_PTR(error);
3299	}
3300	return dentry;
3301}
3302
3303/*
3304 * Look up and maybe create and open the last component.
3305 *
3306 * Must be called with parent locked (exclusive in O_CREAT case).
3307 *
3308 * Returns 0 on success, that is, if
3309 *  the file was successfully atomically created (if necessary) and opened, or
3310 *  the file was not completely opened at this time, though lookups and
3311 *  creations were performed.
3312 * These case are distinguished by presence of FMODE_OPENED on file->f_mode.
3313 * In the latter case dentry returned in @path might be negative if O_CREAT
3314 * hadn't been specified.
3315 *
3316 * An error code is returned on failure.
3317 */
3318static struct dentry *lookup_open(struct nameidata *nd, struct file *file,
3319				  const struct open_flags *op,
3320				  bool got_write)
3321{
3322	struct user_namespace *mnt_userns;
3323	struct dentry *dir = nd->path.dentry;
3324	struct inode *dir_inode = dir->d_inode;
3325	int open_flag = op->open_flag;
3326	struct dentry *dentry;
3327	int error, create_error = 0;
3328	umode_t mode = op->mode;
3329	DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
3330
3331	if (unlikely(IS_DEADDIR(dir_inode)))
3332		return ERR_PTR(-ENOENT);
3333
3334	file->f_mode &= ~FMODE_CREATED;
3335	dentry = d_lookup(dir, &nd->last);
3336	for (;;) {
3337		if (!dentry) {
3338			dentry = d_alloc_parallel(dir, &nd->last, &wq);
3339			if (IS_ERR(dentry))
3340				return dentry;
3341		}
3342		if (d_in_lookup(dentry))
3343			break;
3344
3345		error = d_revalidate(dentry, nd->flags);
3346		if (likely(error > 0))
3347			break;
3348		if (error)
3349			goto out_dput;
3350		d_invalidate(dentry);
3351		dput(dentry);
3352		dentry = NULL;
3353	}
3354	if (dentry->d_inode) {
3355		/* Cached positive dentry: will open in f_op->open */
3356		return dentry;
3357	}
3358
3359	/*
3360	 * Checking write permission is tricky, bacuse we don't know if we are
3361	 * going to actually need it: O_CREAT opens should work as long as the
3362	 * file exists.  But checking existence breaks atomicity.  The trick is
3363	 * to check access and if not granted clear O_CREAT from the flags.
3364	 *
3365	 * Another problem is returing the "right" error value (e.g. for an
3366	 * O_EXCL open we want to return EEXIST not EROFS).
3367	 */
3368	if (unlikely(!got_write))
3369		open_flag &= ~O_TRUNC;
3370	mnt_userns = mnt_user_ns(nd->path.mnt);
3371	if (open_flag & O_CREAT) {
3372		if (open_flag & O_EXCL)
3373			open_flag &= ~O_TRUNC;
3374		mode = vfs_prepare_mode(mnt_userns, dir->d_inode, mode, mode, mode);
3375		if (likely(got_write))
3376			create_error = may_o_create(mnt_userns, &nd->path,
3377						    dentry, mode);
3378		else
3379			create_error = -EROFS;
3380	}
3381	if (create_error)
3382		open_flag &= ~O_CREAT;
3383	if (dir_inode->i_op->atomic_open) {
3384		dentry = atomic_open(nd, dentry, file, open_flag, mode);
3385		if (unlikely(create_error) && dentry == ERR_PTR(-ENOENT))
3386			dentry = ERR_PTR(create_error);
3387		return dentry;
3388	}
3389
3390	if (d_in_lookup(dentry)) {
3391		struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3392							     nd->flags);
3393		d_lookup_done(dentry);
3394		if (unlikely(res)) {
3395			if (IS_ERR(res)) {
3396				error = PTR_ERR(res);
3397				goto out_dput;
3398			}
3399			dput(dentry);
3400			dentry = res;
3401		}
3402	}
3403
3404	/* Negative dentry, just create the file */
3405	if (!dentry->d_inode && (open_flag & O_CREAT)) {
3406		file->f_mode |= FMODE_CREATED;
3407		audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
3408		if (!dir_inode->i_op->create) {
3409			error = -EACCES;
3410			goto out_dput;
3411		}
3412
3413		error = dir_inode->i_op->create(mnt_userns, dir_inode, dentry,
3414						mode, open_flag & O_EXCL);
3415		if (error)
3416			goto out_dput;
3417	}
3418	if (unlikely(create_error) && !dentry->d_inode) {
3419		error = create_error;
3420		goto out_dput;
3421	}
3422	return dentry;
3423
3424out_dput:
3425	dput(dentry);
3426	return ERR_PTR(error);
3427}
3428
3429static const char *open_last_lookups(struct nameidata *nd,
3430		   struct file *file, const struct open_flags *op)
3431{
3432	struct dentry *dir = nd->path.dentry;
3433	int open_flag = op->open_flag;
3434	bool got_write = false;
3435	struct dentry *dentry;
3436	const char *res;
3437
3438	nd->flags |= op->intent;
3439
3440	if (nd->last_type != LAST_NORM) {
3441		if (nd->depth)
3442			put_link(nd);
3443		return handle_dots(nd, nd->last_type);
3444	}
3445
3446	if (!(open_flag & O_CREAT)) {
3447		if (nd->last.name[nd->last.len])
3448			nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3449		/* we _can_ be in RCU mode here */
3450		dentry = lookup_fast(nd);
3451		if (IS_ERR(dentry))
3452			return ERR_CAST(dentry);
3453		if (likely(dentry))
3454			goto finish_lookup;
3455
3456		BUG_ON(nd->flags & LOOKUP_RCU);
3457	} else {
3458		/* create side of things */
3459		if (nd->flags & LOOKUP_RCU) {
3460			if (!try_to_unlazy(nd))
3461				return ERR_PTR(-ECHILD);
3462		}
3463		audit_inode(nd->name, dir, AUDIT_INODE_PARENT);
3464		/* trailing slashes? */
3465		if (unlikely(nd->last.name[nd->last.len]))
3466			return ERR_PTR(-EISDIR);
3467	}
3468
3469	if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3470		got_write = !mnt_want_write(nd->path.mnt);
3471		/*
3472		 * do _not_ fail yet - we might not need that or fail with
3473		 * a different error; let lookup_open() decide; we'll be
3474		 * dropping this one anyway.
3475		 */
3476	}
3477	if (open_flag & O_CREAT)
3478		inode_lock(dir->d_inode);
3479	else
3480		inode_lock_shared(dir->d_inode);
3481	dentry = lookup_open(nd, file, op, got_write);
3482	if (!IS_ERR(dentry) && (file->f_mode & FMODE_CREATED))
3483		fsnotify_create(dir->d_inode, dentry);
3484	if (open_flag & O_CREAT)
3485		inode_unlock(dir->d_inode);
3486	else
3487		inode_unlock_shared(dir->d_inode);
3488
3489	if (got_write)
3490		mnt_drop_write(nd->path.mnt);
3491
3492	if (IS_ERR(dentry))
3493		return ERR_CAST(dentry);
3494
3495	if (file->f_mode & (FMODE_OPENED | FMODE_CREATED)) {
3496		dput(nd->path.dentry);
3497		nd->path.dentry = dentry;
3498		return NULL;
3499	}
3500
3501finish_lookup:
3502	if (nd->depth)
3503		put_link(nd);
3504	res = step_into(nd, WALK_TRAILING, dentry);
3505	if (unlikely(res))
3506		nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3507	return res;
3508}
3509
3510/*
3511 * Handle the last step of open()
3512 */
3513static int do_open(struct nameidata *nd,
3514		   struct file *file, const struct open_flags *op)
3515{
3516	struct user_namespace *mnt_userns;
3517	int open_flag = op->open_flag;
3518	bool do_truncate;
3519	int acc_mode;
3520	int error;
3521
3522	if (!(file->f_mode & (FMODE_OPENED | FMODE_CREATED))) {
3523		error = complete_walk(nd);
3524		if (error)
3525			return error;
3526	}
3527	if (!(file->f_mode & FMODE_CREATED))
3528		audit_inode(nd->name, nd->path.dentry, 0);
3529	mnt_userns = mnt_user_ns(nd->path.mnt);
3530	if (open_flag & O_CREAT) {
3531		if ((open_flag & O_EXCL) && !(file->f_mode & FMODE_CREATED))
3532			return -EEXIST;
3533		if (d_is_dir(nd->path.dentry))
3534			return -EISDIR;
3535		error = may_create_in_sticky(mnt_userns, nd,
3536					     d_backing_inode(nd->path.dentry));
3537		if (unlikely(error))
3538			return error;
3539	}
3540	if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3541		return -ENOTDIR;
3542
3543	do_truncate = false;
3544	acc_mode = op->acc_mode;
3545	if (file->f_mode & FMODE_CREATED) {
3546		/* Don't check for write permission, don't truncate */
3547		open_flag &= ~O_TRUNC;
3548		acc_mode = 0;
3549	} else if (d_is_reg(nd->path.dentry) && open_flag & O_TRUNC) {
3550		error = mnt_want_write(nd->path.mnt);
3551		if (error)
3552			return error;
3553		do_truncate = true;
3554	}
3555	error = may_open(mnt_userns, &nd->path, acc_mode, open_flag);
3556	if (!error && !(file->f_mode & FMODE_OPENED))
3557		error = vfs_open(&nd->path, file);
3558	if (!error)
3559		error = ima_file_check(file, op->acc_mode);
3560	if (!error && do_truncate)
3561		error = handle_truncate(mnt_userns, file);
3562	if (unlikely(error > 0)) {
3563		WARN_ON(1);
3564		error = -EINVAL;
3565	}
3566	if (do_truncate)
3567		mnt_drop_write(nd->path.mnt);
3568	return error;
3569}
3570
3571/**
3572 * vfs_tmpfile - create tmpfile
3573 * @mnt_userns:	user namespace of the mount the inode was found from
3574 * @dentry:	pointer to dentry of the base directory
3575 * @mode:	mode of the new tmpfile
3576 * @open_flag:	flags
3577 *
3578 * Create a temporary file.
3579 *
3580 * If the inode has been found through an idmapped mount the user namespace of
3581 * the vfsmount must be passed through @mnt_userns. This function will then take
3582 * care to map the inode according to @mnt_userns before checking permissions.
3583 * On non-idmapped mounts or if permission checking is to be performed on the
3584 * raw inode simply passs init_user_ns.
3585 */
3586static int vfs_tmpfile(struct user_namespace *mnt_userns,
3587		       const struct path *parentpath,
3588		       struct file *file, umode_t mode)
3589{
3590	struct dentry *child;
3591	struct inode *dir = d_inode(parentpath->dentry);
3592	struct inode *inode;
3593	int error;
3594	int open_flag = file->f_flags;
3595
3596	/* we want directory to be writable */
3597	error = inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC);
3598	if (error)
3599		return error;
3600	if (!dir->i_op->tmpfile)
3601		return -EOPNOTSUPP;
3602	child = d_alloc(parentpath->dentry, &slash_name);
3603	if (unlikely(!child))
3604		return -ENOMEM;
3605	file->f_path.mnt = parentpath->mnt;
3606	file->f_path.dentry = child;
3607	mode = vfs_prepare_mode(mnt_userns, dir, mode, mode, mode);
3608	error = dir->i_op->tmpfile(mnt_userns, dir, file, mode);
3609	dput(child);
3610	if (error)
3611		return error;
3612	/* Don't check for other permissions, the inode was just created */
3613	error = may_open(mnt_userns, &file->f_path, 0, file->f_flags);
3614	if (error)
3615		return error;
3616	inode = file_inode(file);
3617	if (!(open_flag & O_EXCL)) {
3618		spin_lock(&inode->i_lock);
3619		inode->i_state |= I_LINKABLE;
3620		spin_unlock(&inode->i_lock);
3621	}
3622	ima_post_create_tmpfile(mnt_userns, inode);
3623	return 0;
3624}
3625
3626/**
3627 * vfs_tmpfile_open - open a tmpfile for kernel internal use
3628 * @mnt_userns:	user namespace of the mount the inode was found from
3629 * @parentpath:	path of the base directory
3630 * @mode:	mode of the new tmpfile
3631 * @open_flag:	flags
3632 * @cred:	credentials for open
3633 *
3634 * Create and open a temporary file.  The file is not accounted in nr_files,
3635 * hence this is only for kernel internal use, and must not be installed into
3636 * file tables or such.
3637 */
3638struct file *vfs_tmpfile_open(struct user_namespace *mnt_userns,
3639			  const struct path *parentpath,
3640			  umode_t mode, int open_flag, const struct cred *cred)
3641{
3642	struct file *file;
3643	int error;
3644
3645	file = alloc_empty_file_noaccount(open_flag, cred);
3646	if (!IS_ERR(file)) {
3647		error = vfs_tmpfile(mnt_userns, parentpath, file, mode);
3648		if (error) {
3649			fput(file);
3650			file = ERR_PTR(error);
3651		}
3652	}
3653	return file;
3654}
3655EXPORT_SYMBOL(vfs_tmpfile_open);
3656
3657static int do_tmpfile(struct nameidata *nd, unsigned flags,
3658		const struct open_flags *op,
3659		struct file *file)
3660{
3661	struct user_namespace *mnt_userns;
3662	struct path path;
3663	int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
3664
3665	if (unlikely(error))
3666		return error;
3667	error = mnt_want_write(path.mnt);
3668	if (unlikely(error))
3669		goto out;
3670	mnt_userns = mnt_user_ns(path.mnt);
3671	error = vfs_tmpfile(mnt_userns, &path, file, op->mode);
3672	if (error)
3673		goto out2;
3674	audit_inode(nd->name, file->f_path.dentry, 0);
3675out2:
3676	mnt_drop_write(path.mnt);
3677out:
3678	path_put(&path);
3679	return error;
3680}
3681
3682static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file)
3683{
3684	struct path path;
3685	int error = path_lookupat(nd, flags, &path);
3686	if (!error) {
3687		audit_inode(nd->name, path.dentry, 0);
3688		error = vfs_open(&path, file);
3689		path_put(&path);
3690	}
3691	return error;
3692}
3693
3694static struct file *path_openat(struct nameidata *nd,
3695			const struct open_flags *op, unsigned flags)
3696{
3697	struct file *file;
3698	int error;
3699
3700	file = alloc_empty_file(op->open_flag, current_cred());
3701	if (IS_ERR(file))
3702		return file;
3703
3704	if (unlikely(file->f_flags & __O_TMPFILE)) {
3705		error = do_tmpfile(nd, flags, op, file);
3706	} else if (unlikely(file->f_flags & O_PATH)) {
3707		error = do_o_path(nd, flags, file);
3708	} else {
3709		const char *s = path_init(nd, flags);
3710		while (!(error = link_path_walk(s, nd)) &&
3711		       (s = open_last_lookups(nd, file, op)) != NULL)
3712			;
3713		if (!error)
3714			error = do_open(nd, file, op);
3715		terminate_walk(nd);
3716	}
3717	if (likely(!error)) {
3718		if (likely(file->f_mode & FMODE_OPENED))
3719			return file;
3720		WARN_ON(1);
3721		error = -EINVAL;
3722	}
3723	fput(file);
3724	if (error == -EOPENSTALE) {
3725		if (flags & LOOKUP_RCU)
3726			error = -ECHILD;
3727		else
3728			error = -ESTALE;
3729	}
3730	return ERR_PTR(error);
3731}
3732
3733struct file *do_filp_open(int dfd, struct filename *pathname,
3734		const struct open_flags *op)
3735{
3736	struct nameidata nd;
3737	int flags = op->lookup_flags;
3738	struct file *filp;
3739
3740	set_nameidata(&nd, dfd, pathname, NULL);
3741	filp = path_openat(&nd, op, flags | LOOKUP_RCU);
3742	if (unlikely(filp == ERR_PTR(-ECHILD)))
3743		filp = path_openat(&nd, op, flags);
3744	if (unlikely(filp == ERR_PTR(-ESTALE)))
3745		filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
3746	restore_nameidata();
3747	return filp;
3748}
3749
3750struct file *do_file_open_root(const struct path *root,
3751		const char *name, const struct open_flags *op)
3752{
3753	struct nameidata nd;
3754	struct file *file;
3755	struct filename *filename;
3756	int flags = op->lookup_flags;
3757
3758	if (d_is_symlink(root->dentry) && op->intent & LOOKUP_OPEN)
3759		return ERR_PTR(-ELOOP);
3760
3761	filename = getname_kernel(name);
3762	if (IS_ERR(filename))
3763		return ERR_CAST(filename);
3764
3765	set_nameidata(&nd, -1, filename, root);
3766	file = path_openat(&nd, op, flags | LOOKUP_RCU);
3767	if (unlikely(file == ERR_PTR(-ECHILD)))
3768		file = path_openat(&nd, op, flags);
3769	if (unlikely(file == ERR_PTR(-ESTALE)))
3770		file = path_openat(&nd, op, flags | LOOKUP_REVAL);
3771	restore_nameidata();
3772	putname(filename);
3773	return file;
3774}
3775
3776static struct dentry *filename_create(int dfd, struct filename *name,
3777				      struct path *path, unsigned int lookup_flags)
3778{
3779	struct dentry *dentry = ERR_PTR(-EEXIST);
3780	struct qstr last;
3781	bool want_dir = lookup_flags & LOOKUP_DIRECTORY;
3782	unsigned int reval_flag = lookup_flags & LOOKUP_REVAL;
3783	unsigned int create_flags = LOOKUP_CREATE | LOOKUP_EXCL;
3784	int type;
3785	int err2;
3786	int error;
3787
3788	error = filename_parentat(dfd, name, reval_flag, path, &last, &type);
3789	if (error)
3790		return ERR_PTR(error);
3791
3792	/*
3793	 * Yucky last component or no last component at all?
3794	 * (foo/., foo/.., /////)
3795	 */
3796	if (unlikely(type != LAST_NORM))
3797		goto out;
3798
3799	/* don't fail immediately if it's r/o, at least try to report other errors */
3800	err2 = mnt_want_write(path->mnt);
3801	/*
3802	 * Do the final lookup.  Suppress 'create' if there is a trailing
3803	 * '/', and a directory wasn't requested.
3804	 */
3805	if (last.name[last.len] && !want_dir)
3806		create_flags = 0;
3807	inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
3808	dentry = __lookup_hash(&last, path->dentry, reval_flag | create_flags);
3809	if (IS_ERR(dentry))
3810		goto unlock;
3811
3812	error = -EEXIST;
3813	if (d_is_positive(dentry))
3814		goto fail;
3815
3816	/*
3817	 * Special case - lookup gave negative, but... we had foo/bar/
3818	 * From the vfs_mknod() POV we just have a negative dentry -
3819	 * all is fine. Let's be bastards - you had / on the end, you've
3820	 * been asking for (non-existent) directory. -ENOENT for you.
3821	 */
3822	if (unlikely(!create_flags)) {
3823		error = -ENOENT;
3824		goto fail;
3825	}
3826	if (unlikely(err2)) {
3827		error = err2;
3828		goto fail;
3829	}
3830	return dentry;
3831fail:
3832	dput(dentry);
3833	dentry = ERR_PTR(error);
3834unlock:
3835	inode_unlock(path->dentry->d_inode);
3836	if (!err2)
3837		mnt_drop_write(path->mnt);
3838out:
3839	path_put(path);
3840	return dentry;
3841}
3842
3843struct dentry *kern_path_create(int dfd, const char *pathname,
3844				struct path *path, unsigned int lookup_flags)
3845{
3846	struct filename *filename = getname_kernel(pathname);
3847	struct dentry *res = filename_create(dfd, filename, path, lookup_flags);
3848
3849	putname(filename);
3850	return res;
3851}
3852EXPORT_SYMBOL(kern_path_create);
3853
3854void done_path_create(struct path *path, struct dentry *dentry)
3855{
3856	dput(dentry);
3857	inode_unlock(path->dentry->d_inode);
3858	mnt_drop_write(path->mnt);
3859	path_put(path);
3860}
3861EXPORT_SYMBOL(done_path_create);
3862
3863inline struct dentry *user_path_create(int dfd, const char __user *pathname,
3864				struct path *path, unsigned int lookup_flags)
3865{
3866	struct filename *filename = getname(pathname);
3867	struct dentry *res = filename_create(dfd, filename, path, lookup_flags);
3868
3869	putname(filename);
3870	return res;
3871}
3872EXPORT_SYMBOL(user_path_create);
3873
3874/**
3875 * vfs_mknod - create device node or file
3876 * @mnt_userns:	user namespace of the mount the inode was found from
3877 * @dir:	inode of @dentry
3878 * @dentry:	pointer to dentry of the base directory
3879 * @mode:	mode of the new device node or file
3880 * @dev:	device number of device to create
3881 *
3882 * Create a device node or file.
3883 *
3884 * If the inode has been found through an idmapped mount the user namespace of
3885 * the vfsmount must be passed through @mnt_userns. This function will then take
3886 * care to map the inode according to @mnt_userns before checking permissions.
3887 * On non-idmapped mounts or if permission checking is to be performed on the
3888 * raw inode simply passs init_user_ns.
3889 */
3890int vfs_mknod(struct user_namespace *mnt_userns, struct inode *dir,
3891	      struct dentry *dentry, umode_t mode, dev_t dev)
3892{
3893	bool is_whiteout = S_ISCHR(mode) && dev == WHITEOUT_DEV;
3894	int error = may_create(mnt_userns, dir, dentry);
3895
3896	if (error)
3897		return error;
3898
3899	if ((S_ISCHR(mode) || S_ISBLK(mode)) && !is_whiteout &&
3900	    !capable(CAP_MKNOD))
3901		return -EPERM;
3902
3903	if (!dir->i_op->mknod)
3904		return -EPERM;
3905
3906	mode = vfs_prepare_mode(mnt_userns, dir, mode, mode, mode);
3907	error = devcgroup_inode_mknod(mode, dev);
3908	if (error)
3909		return error;
3910
3911	error = security_inode_mknod(dir, dentry, mode, dev);
3912	if (error)
3913		return error;
3914
3915	error = dir->i_op->mknod(mnt_userns, dir, dentry, mode, dev);
3916	if (!error)
3917		fsnotify_create(dir, dentry);
3918	return error;
3919}
3920EXPORT_SYMBOL(vfs_mknod);
3921
3922static int may_mknod(umode_t mode)
3923{
3924	switch (mode & S_IFMT) {
3925	case S_IFREG:
3926	case S_IFCHR:
3927	case S_IFBLK:
3928	case S_IFIFO:
3929	case S_IFSOCK:
3930	case 0: /* zero mode translates to S_IFREG */
3931		return 0;
3932	case S_IFDIR:
3933		return -EPERM;
3934	default:
3935		return -EINVAL;
3936	}
3937}
3938
3939static int do_mknodat(int dfd, struct filename *name, umode_t mode,
3940		unsigned int dev)
3941{
3942	struct user_namespace *mnt_userns;
3943	struct dentry *dentry;
3944	struct path path;
3945	int error;
3946	unsigned int lookup_flags = 0;
3947
3948	error = may_mknod(mode);
3949	if (error)
3950		goto out1;
3951retry:
3952	dentry = filename_create(dfd, name, &path, lookup_flags);
3953	error = PTR_ERR(dentry);
3954	if (IS_ERR(dentry))
3955		goto out1;
3956
3957	error = security_path_mknod(&path, dentry,
3958			mode_strip_umask(path.dentry->d_inode, mode), dev);
3959	if (error)
3960		goto out2;
3961
3962	mnt_userns = mnt_user_ns(path.mnt);
3963	switch (mode & S_IFMT) {
3964		case 0: case S_IFREG:
3965			error = vfs_create(mnt_userns, path.dentry->d_inode,
3966					   dentry, mode, true);
3967			if (!error)
3968				ima_post_path_mknod(mnt_userns, dentry);
3969			break;
3970		case S_IFCHR: case S_IFBLK:
3971			error = vfs_mknod(mnt_userns, path.dentry->d_inode,
3972					  dentry, mode, new_decode_dev(dev));
3973			break;
3974		case S_IFIFO: case S_IFSOCK:
3975			error = vfs_mknod(mnt_userns, path.dentry->d_inode,
3976					  dentry, mode, 0);
3977			break;
3978	}
3979out2:
3980	done_path_create(&path, dentry);
3981	if (retry_estale(error, lookup_flags)) {
3982		lookup_flags |= LOOKUP_REVAL;
3983		goto retry;
3984	}
3985out1:
3986	putname(name);
3987	return error;
3988}
3989
3990SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3991		unsigned int, dev)
3992{
3993	return do_mknodat(dfd, getname(filename), mode, dev);
3994}
3995
3996SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3997{
3998	return do_mknodat(AT_FDCWD, getname(filename), mode, dev);
3999}
4000
4001/**
4002 * vfs_mkdir - create directory
4003 * @mnt_userns:	user namespace of the mount the inode was found from
4004 * @dir:	inode of @dentry
4005 * @dentry:	pointer to dentry of the base directory
4006 * @mode:	mode of the new directory
4007 *
4008 * Create a directory.
4009 *
4010 * If the inode has been found through an idmapped mount the user namespace of
4011 * the vfsmount must be passed through @mnt_userns. This function will then take
4012 * care to map the inode according to @mnt_userns before checking permissions.
4013 * On non-idmapped mounts or if permission checking is to be performed on the
4014 * raw inode simply passs init_user_ns.
4015 */
4016int vfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
4017	      struct dentry *dentry, umode_t mode)
4018{
4019	int error = may_create(mnt_userns, dir, dentry);
4020	unsigned max_links = dir->i_sb->s_max_links;
4021
4022	if (error)
4023		return error;
4024
4025	if (!dir->i_op->mkdir)
4026		return -EPERM;
4027
4028	mode = vfs_prepare_mode(mnt_userns, dir, mode, S_IRWXUGO | S_ISVTX, 0);
4029	error = security_inode_mkdir(dir, dentry, mode);
4030	if (error)
4031		return error;
4032
4033	if (max_links && dir->i_nlink >= max_links)
4034		return -EMLINK;
4035
4036	error = dir->i_op->mkdir(mnt_userns, dir, dentry, mode);
4037	if (!error)
4038		fsnotify_mkdir(dir, dentry);
4039	return error;
4040}
4041EXPORT_SYMBOL(vfs_mkdir);
4042
4043int do_mkdirat(int dfd, struct filename *name, umode_t mode)
4044{
4045	struct dentry *dentry;
4046	struct path path;
4047	int error;
4048	unsigned int lookup_flags = LOOKUP_DIRECTORY;
4049
4050retry:
4051	dentry = filename_create(dfd, name, &path, lookup_flags);
4052	error = PTR_ERR(dentry);
4053	if (IS_ERR(dentry))
4054		goto out_putname;
4055
4056	error = security_path_mkdir(&path, dentry,
4057			mode_strip_umask(path.dentry->d_inode, mode));
4058	if (!error) {
4059		struct user_namespace *mnt_userns;
4060		mnt_userns = mnt_user_ns(path.mnt);
4061		error = vfs_mkdir(mnt_userns, path.dentry->d_inode, dentry,
4062				  mode);
4063	}
4064	done_path_create(&path, dentry);
4065	if (retry_estale(error, lookup_flags)) {
4066		lookup_flags |= LOOKUP_REVAL;
4067		goto retry;
4068	}
4069out_putname:
4070	putname(name);
4071	return error;
4072}
4073
4074SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
4075{
4076	return do_mkdirat(dfd, getname(pathname), mode);
4077}
4078
4079SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
4080{
4081	return do_mkdirat(AT_FDCWD, getname(pathname), mode);
4082}
4083
4084/**
4085 * vfs_rmdir - remove directory
4086 * @mnt_userns:	user namespace of the mount the inode was found from
4087 * @dir:	inode of @dentry
4088 * @dentry:	pointer to dentry of the base directory
4089 *
4090 * Remove a directory.
4091 *
4092 * If the inode has been found through an idmapped mount the user namespace of
4093 * the vfsmount must be passed through @mnt_userns. This function will then take
4094 * care to map the inode according to @mnt_userns before checking permissions.
4095 * On non-idmapped mounts or if permission checking is to be performed on the
4096 * raw inode simply passs init_user_ns.
4097 */
4098int vfs_rmdir(struct user_namespace *mnt_userns, struct inode *dir,
4099		     struct dentry *dentry)
4100{
4101	int error = may_delete(mnt_userns, dir, dentry, 1);
4102
4103	if (error)
4104		return error;
4105
4106	if (!dir->i_op->rmdir)
4107		return -EPERM;
4108
4109	dget(dentry);
4110	inode_lock(dentry->d_inode);
4111
4112	error = -EBUSY;
4113	if (is_local_mountpoint(dentry) ||
4114	    (dentry->d_inode->i_flags & S_KERNEL_FILE))
4115		goto out;
4116
4117	error = security_inode_rmdir(dir, dentry);
4118	if (error)
4119		goto out;
4120
4121	error = dir->i_op->rmdir(dir, dentry);
4122	if (error)
4123		goto out;
4124
4125	shrink_dcache_parent(dentry);
4126	dentry->d_inode->i_flags |= S_DEAD;
4127	dont_mount(dentry);
4128	detach_mounts(dentry);
4129
4130out:
4131	inode_unlock(dentry->d_inode);
4132	dput(dentry);
4133	if (!error)
4134		d_delete_notify(dir, dentry);
4135	return error;
4136}
4137EXPORT_SYMBOL(vfs_rmdir);
4138
4139int do_rmdir(int dfd, struct filename *name)
4140{
4141	struct user_namespace *mnt_userns;
4142	int error;
4143	struct dentry *dentry;
4144	struct path path;
4145	struct qstr last;
4146	int type;
4147	unsigned int lookup_flags = 0;
4148retry:
4149	error = filename_parentat(dfd, name, lookup_flags, &path, &last, &type);
4150	if (error)
4151		goto exit1;
4152
4153	switch (type) {
4154	case LAST_DOTDOT:
4155		error = -ENOTEMPTY;
4156		goto exit2;
4157	case LAST_DOT:
4158		error = -EINVAL;
4159		goto exit2;
4160	case LAST_ROOT:
4161		error = -EBUSY;
4162		goto exit2;
4163	}
4164
4165	error = mnt_want_write(path.mnt);
4166	if (error)
4167		goto exit2;
4168
4169	inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
4170	dentry = __lookup_hash(&last, path.dentry, lookup_flags);
4171	error = PTR_ERR(dentry);
4172	if (IS_ERR(dentry))
4173		goto exit3;
4174	if (!dentry->d_inode) {
4175		error = -ENOENT;
4176		goto exit4;
4177	}
4178	error = security_path_rmdir(&path, dentry);
4179	if (error)
4180		goto exit4;
4181	mnt_userns = mnt_user_ns(path.mnt);
4182	error = vfs_rmdir(mnt_userns, path.dentry->d_inode, dentry);
4183exit4:
4184	dput(dentry);
4185exit3:
4186	inode_unlock(path.dentry->d_inode);
4187	mnt_drop_write(path.mnt);
4188exit2:
4189	path_put(&path);
4190	if (retry_estale(error, lookup_flags)) {
4191		lookup_flags |= LOOKUP_REVAL;
4192		goto retry;
4193	}
4194exit1:
4195	putname(name);
4196	return error;
4197}
4198
4199SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
4200{
4201	return do_rmdir(AT_FDCWD, getname(pathname));
4202}
4203
4204/**
4205 * vfs_unlink - unlink a filesystem object
4206 * @mnt_userns:	user namespace of the mount the inode was found from
4207 * @dir:	parent directory
4208 * @dentry:	victim
4209 * @delegated_inode: returns victim inode, if the inode is delegated.
4210 *
4211 * The caller must hold dir->i_mutex.
4212 *
4213 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
4214 * return a reference to the inode in delegated_inode.  The caller
4215 * should then break the delegation on that inode and retry.  Because
4216 * breaking a delegation may take a long time, the caller should drop
4217 * dir->i_mutex before doing so.
4218 *
4219 * Alternatively, a caller may pass NULL for delegated_inode.  This may
4220 * be appropriate for callers that expect the underlying filesystem not
4221 * to be NFS exported.
4222 *
4223 * If the inode has been found through an idmapped mount the user namespace of
4224 * the vfsmount must be passed through @mnt_userns. This function will then take
4225 * care to map the inode according to @mnt_userns before checking permissions.
4226 * On non-idmapped mounts or if permission checking is to be performed on the
4227 * raw inode simply passs init_user_ns.
4228 */
4229int vfs_unlink(struct user_namespace *mnt_userns, struct inode *dir,
4230	       struct dentry *dentry, struct inode **delegated_inode)
4231{
4232	struct inode *target = dentry->d_inode;
4233	int error = may_delete(mnt_userns, dir, dentry, 0);
4234
4235	if (error)
4236		return error;
4237
4238	if (!dir->i_op->unlink)
4239		return -EPERM;
4240
4241	inode_lock(target);
4242	if (IS_SWAPFILE(target))
4243		error = -EPERM;
4244	else if (is_local_mountpoint(dentry))
4245		error = -EBUSY;
4246	else {
4247		error = security_inode_unlink(dir, dentry);
4248		if (!error) {
4249			error = try_break_deleg(target, delegated_inode);
4250			if (error)
4251				goto out;
4252			error = dir->i_op->unlink(dir, dentry);
4253			if (!error) {
4254				dont_mount(dentry);
4255				detach_mounts(dentry);
4256			}
4257		}
4258	}
4259out:
4260	inode_unlock(target);
4261
4262	/* We don't d_delete() NFS sillyrenamed files--they still exist. */
4263	if (!error && dentry->d_flags & DCACHE_NFSFS_RENAMED) {
4264		fsnotify_unlink(dir, dentry);
4265	} else if (!error) {
4266		fsnotify_link_count(target);
4267		d_delete_notify(dir, dentry);
4268	}
4269
4270	return error;
4271}
4272EXPORT_SYMBOL(vfs_unlink);
4273
4274/*
4275 * Make sure that the actual truncation of the file will occur outside its
4276 * directory's i_mutex.  Truncate can take a long time if there is a lot of
4277 * writeout happening, and we don't want to prevent access to the directory
4278 * while waiting on the I/O.
4279 */
4280int do_unlinkat(int dfd, struct filename *name)
4281{
4282	int error;
4283	struct dentry *dentry;
4284	struct path path;
4285	struct qstr last;
4286	int type;
4287	struct inode *inode = NULL;
4288	struct inode *delegated_inode = NULL;
4289	unsigned int lookup_flags = 0;
4290retry:
4291	error = filename_parentat(dfd, name, lookup_flags, &path, &last, &type);
4292	if (error)
4293		goto exit1;
4294
4295	error = -EISDIR;
4296	if (type != LAST_NORM)
4297		goto exit2;
4298
4299	error = mnt_want_write(path.mnt);
4300	if (error)
4301		goto exit2;
4302retry_deleg:
4303	inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
4304	dentry = __lookup_hash(&last, path.dentry, lookup_flags);
4305	error = PTR_ERR(dentry);
4306	if (!IS_ERR(dentry)) {
4307		struct user_namespace *mnt_userns;
4308
4309		/* Why not before? Because we want correct error value */
4310		if (last.name[last.len])
4311			goto slashes;
4312		inode = dentry->d_inode;
4313		if (d_is_negative(dentry))
4314			goto slashes;
4315		ihold(inode);
4316		error = security_path_unlink(&path, dentry);
4317		if (error)
4318			goto exit3;
4319		mnt_userns = mnt_user_ns(path.mnt);
4320		error = vfs_unlink(mnt_userns, path.dentry->d_inode, dentry,
4321				   &delegated_inode);
4322exit3:
4323		dput(dentry);
4324	}
4325	inode_unlock(path.dentry->d_inode);
4326	if (inode)
4327		iput(inode);	/* truncate the inode here */
4328	inode = NULL;
4329	if (delegated_inode) {
4330		error = break_deleg_wait(&delegated_inode);
4331		if (!error)
4332			goto retry_deleg;
4333	}
4334	mnt_drop_write(path.mnt);
4335exit2:
4336	path_put(&path);
4337	if (retry_estale(error, lookup_flags)) {
4338		lookup_flags |= LOOKUP_REVAL;
4339		inode = NULL;
4340		goto retry;
4341	}
4342exit1:
4343	putname(name);
4344	return error;
4345
4346slashes:
4347	if (d_is_negative(dentry))
4348		error = -ENOENT;
4349	else if (d_is_dir(dentry))
4350		error = -EISDIR;
4351	else
4352		error = -ENOTDIR;
4353	goto exit3;
4354}
4355
4356SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
4357{
4358	if ((flag & ~AT_REMOVEDIR) != 0)
4359		return -EINVAL;
4360
4361	if (flag & AT_REMOVEDIR)
4362		return do_rmdir(dfd, getname(pathname));
4363	return do_unlinkat(dfd, getname(pathname));
4364}
4365
4366SYSCALL_DEFINE1(unlink, const char __user *, pathname)
4367{
4368	return do_unlinkat(AT_FDCWD, getname(pathname));
4369}
4370
4371/**
4372 * vfs_symlink - create symlink
4373 * @mnt_userns:	user namespace of the mount the inode was found from
4374 * @dir:	inode of @dentry
4375 * @dentry:	pointer to dentry of the base directory
4376 * @oldname:	name of the file to link to
4377 *
4378 * Create a symlink.
4379 *
4380 * If the inode has been found through an idmapped mount the user namespace of
4381 * the vfsmount must be passed through @mnt_userns. This function will then take
4382 * care to map the inode according to @mnt_userns before checking permissions.
4383 * On non-idmapped mounts or if permission checking is to be performed on the
4384 * raw inode simply passs init_user_ns.
4385 */
4386int vfs_symlink(struct user_namespace *mnt_userns, struct inode *dir,
4387		struct dentry *dentry, const char *oldname)
4388{
4389	int error = may_create(mnt_userns, dir, dentry);
4390
4391	if (error)
4392		return error;
4393
4394	if (!dir->i_op->symlink)
4395		return -EPERM;
4396
4397	error = security_inode_symlink(dir, dentry, oldname);
4398	if (error)
4399		return error;
4400
4401	error = dir->i_op->symlink(mnt_userns, dir, dentry, oldname);
4402	if (!error)
4403		fsnotify_create(dir, dentry);
4404	return error;
4405}
4406EXPORT_SYMBOL(vfs_symlink);
4407
4408int do_symlinkat(struct filename *from, int newdfd, struct filename *to)
4409{
4410	int error;
4411	struct dentry *dentry;
4412	struct path path;
4413	unsigned int lookup_flags = 0;
4414
4415	if (IS_ERR(from)) {
4416		error = PTR_ERR(from);
4417		goto out_putnames;
4418	}
4419retry:
4420	dentry = filename_create(newdfd, to, &path, lookup_flags);
4421	error = PTR_ERR(dentry);
4422	if (IS_ERR(dentry))
4423		goto out_putnames;
4424
4425	error = security_path_symlink(&path, dentry, from->name);
4426	if (!error) {
4427		struct user_namespace *mnt_userns;
4428
4429		mnt_userns = mnt_user_ns(path.mnt);
4430		error = vfs_symlink(mnt_userns, path.dentry->d_inode, dentry,
4431				    from->name);
4432	}
4433	done_path_create(&path, dentry);
4434	if (retry_estale(error, lookup_flags)) {
4435		lookup_flags |= LOOKUP_REVAL;
4436		goto retry;
4437	}
4438out_putnames:
4439	putname(to);
4440	putname(from);
4441	return error;
4442}
4443
4444SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
4445		int, newdfd, const char __user *, newname)
4446{
4447	return do_symlinkat(getname(oldname), newdfd, getname(newname));
4448}
4449
4450SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
4451{
4452	return do_symlinkat(getname(oldname), AT_FDCWD, getname(newname));
4453}
4454
4455/**
4456 * vfs_link - create a new link
4457 * @old_dentry:	object to be linked
4458 * @mnt_userns:	the user namespace of the mount
4459 * @dir:	new parent
4460 * @new_dentry:	where to create the new link
4461 * @delegated_inode: returns inode needing a delegation break
4462 *
4463 * The caller must hold dir->i_mutex
4464 *
4465 * If vfs_link discovers a delegation on the to-be-linked file in need
4466 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4467 * inode in delegated_inode.  The caller should then break the delegation
4468 * and retry.  Because breaking a delegation may take a long time, the
4469 * caller should drop the i_mutex before doing so.
4470 *
4471 * Alternatively, a caller may pass NULL for delegated_inode.  This may
4472 * be appropriate for callers that expect the underlying filesystem not
4473 * to be NFS exported.
4474 *
4475 * If the inode has been found through an idmapped mount the user namespace of
4476 * the vfsmount must be passed through @mnt_userns. This function will then take
4477 * care to map the inode according to @mnt_userns before checking permissions.
4478 * On non-idmapped mounts or if permission checking is to be performed on the
4479 * raw inode simply passs init_user_ns.
4480 */
4481int vfs_link(struct dentry *old_dentry, struct user_namespace *mnt_userns,
4482	     struct inode *dir, struct dentry *new_dentry,
4483	     struct inode **delegated_inode)
4484{
4485	struct inode *inode = old_dentry->d_inode;
4486	unsigned max_links = dir->i_sb->s_max_links;
4487	int error;
4488
4489	if (!inode)
4490		return -ENOENT;
4491
4492	error = may_create(mnt_userns, dir, new_dentry);
4493	if (error)
4494		return error;
4495
4496	if (dir->i_sb != inode->i_sb)
4497		return -EXDEV;
4498
4499	/*
4500	 * A link to an append-only or immutable file cannot be created.
4501	 */
4502	if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4503		return -EPERM;
4504	/*
4505	 * Updating the link count will likely cause i_uid and i_gid to
4506	 * be writen back improperly if their true value is unknown to
4507	 * the vfs.
4508	 */
4509	if (HAS_UNMAPPED_ID(mnt_userns, inode))
4510		return -EPERM;
4511	if (!dir->i_op->link)
4512		return -EPERM;
4513	if (S_ISDIR(inode->i_mode))
4514		return -EPERM;
4515
4516	error = security_inode_link(old_dentry, dir, new_dentry);
4517	if (error)
4518		return error;
4519
4520	inode_lock(inode);
4521	/* Make sure we don't allow creating hardlink to an unlinked file */
4522	if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4523		error =  -ENOENT;
4524	else if (max_links && inode->i_nlink >= max_links)
4525		error = -EMLINK;
4526	else {
4527		error = try_break_deleg(inode, delegated_inode);
4528		if (!error)
4529			error = dir->i_op->link(old_dentry, dir, new_dentry);
4530	}
4531
4532	if (!error && (inode->i_state & I_LINKABLE)) {
4533		spin_lock(&inode->i_lock);
4534		inode->i_state &= ~I_LINKABLE;
4535		spin_unlock(&inode->i_lock);
4536	}
4537	inode_unlock(inode);
4538	if (!error)
4539		fsnotify_link(dir, inode, new_dentry);
4540	return error;
4541}
4542EXPORT_SYMBOL(vfs_link);
4543
4544/*
4545 * Hardlinks are often used in delicate situations.  We avoid
4546 * security-related surprises by not following symlinks on the
4547 * newname.  --KAB
4548 *
4549 * We don't follow them on the oldname either to be compatible
4550 * with linux 2.0, and to avoid hard-linking to directories
4551 * and other special files.  --ADM
4552 */
4553int do_linkat(int olddfd, struct filename *old, int newdfd,
4554	      struct filename *new, int flags)
4555{
4556	struct user_namespace *mnt_userns;
4557	struct dentry *new_dentry;
4558	struct path old_path, new_path;
4559	struct inode *delegated_inode = NULL;
4560	int how = 0;
4561	int error;
4562
4563	if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) {
4564		error = -EINVAL;
4565		goto out_putnames;
4566	}
4567	/*
4568	 * To use null names we require CAP_DAC_READ_SEARCH
4569	 * This ensures that not everyone will be able to create
4570	 * handlink using the passed filedescriptor.
4571	 */
4572	if (flags & AT_EMPTY_PATH && !capable(CAP_DAC_READ_SEARCH)) {
4573		error = -ENOENT;
4574		goto out_putnames;
4575	}
4576
4577	if (flags & AT_SYMLINK_FOLLOW)
4578		how |= LOOKUP_FOLLOW;
4579retry:
4580	error = filename_lookup(olddfd, old, how, &old_path, NULL);
4581	if (error)
4582		goto out_putnames;
4583
4584	new_dentry = filename_create(newdfd, new, &new_path,
4585					(how & LOOKUP_REVAL));
4586	error = PTR_ERR(new_dentry);
4587	if (IS_ERR(new_dentry))
4588		goto out_putpath;
4589
4590	error = -EXDEV;
4591	if (old_path.mnt != new_path.mnt)
4592		goto out_dput;
4593	mnt_userns = mnt_user_ns(new_path.mnt);
4594	error = may_linkat(mnt_userns, &old_path);
4595	if (unlikely(error))
4596		goto out_dput;
4597	error = security_path_link(old_path.dentry, &new_path, new_dentry);
4598	if (error)
4599		goto out_dput;
4600	error = vfs_link(old_path.dentry, mnt_userns, new_path.dentry->d_inode,
4601			 new_dentry, &delegated_inode);
4602out_dput:
4603	done_path_create(&new_path, new_dentry);
4604	if (delegated_inode) {
4605		error = break_deleg_wait(&delegated_inode);
4606		if (!error) {
4607			path_put(&old_path);
4608			goto retry;
4609		}
4610	}
4611	if (retry_estale(error, how)) {
4612		path_put(&old_path);
4613		how |= LOOKUP_REVAL;
4614		goto retry;
4615	}
4616out_putpath:
4617	path_put(&old_path);
4618out_putnames:
4619	putname(old);
4620	putname(new);
4621
4622	return error;
4623}
4624
4625SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4626		int, newdfd, const char __user *, newname, int, flags)
4627{
4628	return do_linkat(olddfd, getname_uflags(oldname, flags),
4629		newdfd, getname(newname), flags);
4630}
4631
4632SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4633{
4634	return do_linkat(AT_FDCWD, getname(oldname), AT_FDCWD, getname(newname), 0);
4635}
4636
4637/**
4638 * vfs_rename - rename a filesystem object
4639 * @rd:		pointer to &struct renamedata info
4640 *
4641 * The caller must hold multiple mutexes--see lock_rename()).
4642 *
4643 * If vfs_rename discovers a delegation in need of breaking at either
4644 * the source or destination, it will return -EWOULDBLOCK and return a
4645 * reference to the inode in delegated_inode.  The caller should then
4646 * break the delegation and retry.  Because breaking a delegation may
4647 * take a long time, the caller should drop all locks before doing
4648 * so.
4649 *
4650 * Alternatively, a caller may pass NULL for delegated_inode.  This may
4651 * be appropriate for callers that expect the underlying filesystem not
4652 * to be NFS exported.
4653 *
4654 * The worst of all namespace operations - renaming directory. "Perverted"
4655 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4656 * Problems:
4657 *
4658 *	a) we can get into loop creation.
4659 *	b) race potential - two innocent renames can create a loop together.
4660 *	   That's where 4.4 screws up. Current fix: serialization on
4661 *	   sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4662 *	   story.
4663 *	c) we have to lock _four_ objects - parents and victim (if it exists),
4664 *	   and source (if it is not a directory).
4665 *	   And that - after we got ->i_mutex on parents (until then we don't know
4666 *	   whether the target exists).  Solution: try to be smart with locking
4667 *	   order for inodes.  We rely on the fact that tree topology may change
4668 *	   only under ->s_vfs_rename_mutex _and_ that parent of the object we
4669 *	   move will be locked.  Thus we can rank directories by the tree
4670 *	   (ancestors first) and rank all non-directories after them.
4671 *	   That works since everybody except rename does "lock parent, lookup,
4672 *	   lock child" and rename is under ->s_vfs_rename_mutex.
4673 *	   HOWEVER, it relies on the assumption that any object with ->lookup()
4674 *	   has no more than 1 dentry.  If "hybrid" objects will ever appear,
4675 *	   we'd better make sure that there's no link(2) for them.
4676 *	d) conversion from fhandle to dentry may come in the wrong moment - when
4677 *	   we are removing the target. Solution: we will have to grab ->i_mutex
4678 *	   in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4679 *	   ->i_mutex on parents, which works but leads to some truly excessive
4680 *	   locking].
4681 */
4682int vfs_rename(struct renamedata *rd)
4683{
4684	int error;
4685	struct inode *old_dir = rd->old_dir, *new_dir = rd->new_dir;
4686	struct dentry *old_dentry = rd->old_dentry;
4687	struct dentry *new_dentry = rd->new_dentry;
4688	struct inode **delegated_inode = rd->delegated_inode;
4689	unsigned int flags = rd->flags;
4690	bool is_dir = d_is_dir(old_dentry);
4691	struct inode *source = old_dentry->d_inode;
4692	struct inode *target = new_dentry->d_inode;
4693	bool new_is_dir = false;
4694	unsigned max_links = new_dir->i_sb->s_max_links;
4695	struct name_snapshot old_name;
4696
4697	if (source == target)
4698		return 0;
4699
4700	error = may_delete(rd->old_mnt_userns, old_dir, old_dentry, is_dir);
4701	if (error)
4702		return error;
4703
4704	if (!target) {
4705		error = may_create(rd->new_mnt_userns, new_dir, new_dentry);
4706	} else {
4707		new_is_dir = d_is_dir(new_dentry);
4708
4709		if (!(flags & RENAME_EXCHANGE))
4710			error = may_delete(rd->new_mnt_userns, new_dir,
4711					   new_dentry, is_dir);
4712		else
4713			error = may_delete(rd->new_mnt_userns, new_dir,
4714					   new_dentry, new_is_dir);
4715	}
4716	if (error)
4717		return error;
4718
4719	if (!old_dir->i_op->rename)
4720		return -EPERM;
4721
4722	/*
4723	 * If we are going to change the parent - check write permissions,
4724	 * we'll need to flip '..'.
4725	 */
4726	if (new_dir != old_dir) {
4727		if (is_dir) {
4728			error = inode_permission(rd->old_mnt_userns, source,
4729						 MAY_WRITE);
4730			if (error)
4731				return error;
4732		}
4733		if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4734			error = inode_permission(rd->new_mnt_userns, target,
4735						 MAY_WRITE);
4736			if (error)
4737				return error;
4738		}
4739	}
4740
4741	error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4742				      flags);
4743	if (error)
4744		return error;
4745
4746	take_dentry_name_snapshot(&old_name, old_dentry);
4747	dget(new_dentry);
4748	if (!is_dir || (flags & RENAME_EXCHANGE))
4749		lock_two_nondirectories(source, target);
4750	else if (target)
4751		inode_lock(target);
4752
4753	error = -EPERM;
4754	if (IS_SWAPFILE(source) || (target && IS_SWAPFILE(target)))
4755		goto out;
4756
4757	error = -EBUSY;
4758	if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4759		goto out;
4760
4761	if (max_links && new_dir != old_dir) {
4762		error = -EMLINK;
4763		if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4764			goto out;
4765		if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4766		    old_dir->i_nlink >= max_links)
4767			goto out;
4768	}
4769	if (!is_dir) {
4770		error = try_break_deleg(source, delegated_inode);
4771		if (error)
4772			goto out;
4773	}
4774	if (target && !new_is_dir) {
4775		error = try_break_deleg(target, delegated_inode);
4776		if (error)
4777			goto out;
4778	}
4779	error = old_dir->i_op->rename(rd->new_mnt_userns, old_dir, old_dentry,
4780				      new_dir, new_dentry, flags);
4781	if (error)
4782		goto out;
4783
4784	if (!(flags & RENAME_EXCHANGE) && target) {
4785		if (is_dir) {
4786			shrink_dcache_parent(new_dentry);
4787			target->i_flags |= S_DEAD;
4788		}
4789		dont_mount(new_dentry);
4790		detach_mounts(new_dentry);
4791	}
4792	if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4793		if (!(flags & RENAME_EXCHANGE))
4794			d_move(old_dentry, new_dentry);
4795		else
4796			d_exchange(old_dentry, new_dentry);
4797	}
4798out:
4799	if (!is_dir || (flags & RENAME_EXCHANGE))
4800		unlock_two_nondirectories(source, target);
4801	else if (target)
4802		inode_unlock(target);
4803	dput(new_dentry);
4804	if (!error) {
4805		fsnotify_move(old_dir, new_dir, &old_name.name, is_dir,
4806			      !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4807		if (flags & RENAME_EXCHANGE) {
4808			fsnotify_move(new_dir, old_dir, &old_dentry->d_name,
4809				      new_is_dir, NULL, new_dentry);
4810		}
4811	}
4812	release_dentry_name_snapshot(&old_name);
4813
4814	return error;
4815}
4816EXPORT_SYMBOL(vfs_rename);
4817
4818int do_renameat2(int olddfd, struct filename *from, int newdfd,
4819		 struct filename *to, unsigned int flags)
4820{
4821	struct renamedata rd;
4822	struct dentry *old_dentry, *new_dentry;
4823	struct dentry *trap;
4824	struct path old_path, new_path;
4825	struct qstr old_last, new_last;
4826	int old_type, new_type;
4827	struct inode *delegated_inode = NULL;
4828	unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4829	bool should_retry = false;
4830	int error = -EINVAL;
4831
4832	if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4833		goto put_names;
4834
4835	if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4836	    (flags & RENAME_EXCHANGE))
4837		goto put_names;
4838
4839	if (flags & RENAME_EXCHANGE)
4840		target_flags = 0;
4841
4842retry:
4843	error = filename_parentat(olddfd, from, lookup_flags, &old_path,
4844				  &old_last, &old_type);
4845	if (error)
4846		goto put_names;
4847
4848	error = filename_parentat(newdfd, to, lookup_flags, &new_path, &new_last,
4849				  &new_type);
4850	if (error)
4851		goto exit1;
4852
4853	error = -EXDEV;
4854	if (old_path.mnt != new_path.mnt)
4855		goto exit2;
4856
4857	error = -EBUSY;
4858	if (old_type != LAST_NORM)
4859		goto exit2;
4860
4861	if (flags & RENAME_NOREPLACE)
4862		error = -EEXIST;
4863	if (new_type != LAST_NORM)
4864		goto exit2;
4865
4866	error = mnt_want_write(old_path.mnt);
4867	if (error)
4868		goto exit2;
4869
4870retry_deleg:
4871	trap = lock_rename(new_path.dentry, old_path.dentry);
4872
4873	old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4874	error = PTR_ERR(old_dentry);
4875	if (IS_ERR(old_dentry))
4876		goto exit3;
4877	/* source must exist */
4878	error = -ENOENT;
4879	if (d_is_negative(old_dentry))
4880		goto exit4;
4881	new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4882	error = PTR_ERR(new_dentry);
4883	if (IS_ERR(new_dentry))
4884		goto exit4;
4885	error = -EEXIST;
4886	if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4887		goto exit5;
4888	if (flags & RENAME_EXCHANGE) {
4889		error = -ENOENT;
4890		if (d_is_negative(new_dentry))
4891			goto exit5;
4892
4893		if (!d_is_dir(new_dentry)) {
4894			error = -ENOTDIR;
4895			if (new_last.name[new_last.len])
4896				goto exit5;
4897		}
4898	}
4899	/* unless the source is a directory trailing slashes give -ENOTDIR */
4900	if (!d_is_dir(old_dentry)) {
4901		error = -ENOTDIR;
4902		if (old_last.name[old_last.len])
4903			goto exit5;
4904		if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4905			goto exit5;
4906	}
4907	/* source should not be ancestor of target */
4908	error = -EINVAL;
4909	if (old_dentry == trap)
4910		goto exit5;
4911	/* target should not be an ancestor of source */
4912	if (!(flags & RENAME_EXCHANGE))
4913		error = -ENOTEMPTY;
4914	if (new_dentry == trap)
4915		goto exit5;
4916
4917	error = security_path_rename(&old_path, old_dentry,
4918				     &new_path, new_dentry, flags);
4919	if (error)
4920		goto exit5;
4921
4922	rd.old_dir	   = old_path.dentry->d_inode;
4923	rd.old_dentry	   = old_dentry;
4924	rd.old_mnt_userns  = mnt_user_ns(old_path.mnt);
4925	rd.new_dir	   = new_path.dentry->d_inode;
4926	rd.new_dentry	   = new_dentry;
4927	rd.new_mnt_userns  = mnt_user_ns(new_path.mnt);
4928	rd.delegated_inode = &delegated_inode;
4929	rd.flags	   = flags;
4930	error = vfs_rename(&rd);
4931exit5:
4932	dput(new_dentry);
4933exit4:
4934	dput(old_dentry);
4935exit3:
4936	unlock_rename(new_path.dentry, old_path.dentry);
4937	if (delegated_inode) {
4938		error = break_deleg_wait(&delegated_inode);
4939		if (!error)
4940			goto retry_deleg;
4941	}
4942	mnt_drop_write(old_path.mnt);
4943exit2:
4944	if (retry_estale(error, lookup_flags))
4945		should_retry = true;
4946	path_put(&new_path);
4947exit1:
4948	path_put(&old_path);
4949	if (should_retry) {
4950		should_retry = false;
4951		lookup_flags |= LOOKUP_REVAL;
4952		goto retry;
4953	}
4954put_names:
4955	putname(from);
4956	putname(to);
4957	return error;
4958}
4959
4960SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4961		int, newdfd, const char __user *, newname, unsigned int, flags)
4962{
4963	return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname),
4964				flags);
4965}
4966
4967SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4968		int, newdfd, const char __user *, newname)
4969{
4970	return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname),
4971				0);
4972}
4973
4974SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4975{
4976	return do_renameat2(AT_FDCWD, getname(oldname), AT_FDCWD,
4977				getname(newname), 0);
4978}
4979
4980int readlink_copy(char __user *buffer, int buflen, const char *link)
4981{
4982	int len = PTR_ERR(link);
4983	if (IS_ERR(link))
4984		goto out;
4985
4986	len = strlen(link);
4987	if (len > (unsigned) buflen)
4988		len = buflen;
4989	if (copy_to_user(buffer, link, len))
4990		len = -EFAULT;
4991out:
4992	return len;
4993}
4994
4995/**
4996 * vfs_readlink - copy symlink body into userspace buffer
4997 * @dentry: dentry on which to get symbolic link
4998 * @buffer: user memory pointer
4999 * @buflen: size of buffer
5000 *
5001 * Does not touch atime.  That's up to the caller if necessary
5002 *
5003 * Does not call security hook.
5004 */
5005int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen)
5006{
5007	struct inode *inode = d_inode(dentry);
5008	DEFINE_DELAYED_CALL(done);
5009	const char *link;
5010	int res;
5011
5012	if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) {
5013		if (unlikely(inode->i_op->readlink))
5014			return inode->i_op->readlink(dentry, buffer, buflen);
5015
5016		if (!d_is_symlink(dentry))
5017			return -EINVAL;
5018
5019		spin_lock(&inode->i_lock);
5020		inode->i_opflags |= IOP_DEFAULT_READLINK;
5021		spin_unlock(&inode->i_lock);
5022	}
5023
5024	link = READ_ONCE(inode->i_link);
5025	if (!link) {
5026		link = inode->i_op->get_link(dentry, inode, &done);
5027		if (IS_ERR(link))
5028			return PTR_ERR(link);
5029	}
5030	res = readlink_copy(buffer, buflen, link);
5031	do_delayed_call(&done);
5032	return res;
5033}
5034EXPORT_SYMBOL(vfs_readlink);
5035
5036/**
5037 * vfs_get_link - get symlink body
5038 * @dentry: dentry on which to get symbolic link
5039 * @done: caller needs to free returned data with this
5040 *
5041 * Calls security hook and i_op->get_link() on the supplied inode.
5042 *
5043 * It does not touch atime.  That's up to the caller if necessary.
5044 *
5045 * Does not work on "special" symlinks like /proc/$$/fd/N
5046 */
5047const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done)
5048{
5049	const char *res = ERR_PTR(-EINVAL);
5050	struct inode *inode = d_inode(dentry);
5051
5052	if (d_is_symlink(dentry)) {
5053		res = ERR_PTR(security_inode_readlink(dentry));
5054		if (!res)
5055			res = inode->i_op->get_link(dentry, inode, done);
5056	}
5057	return res;
5058}
5059EXPORT_SYMBOL(vfs_get_link);
5060
5061/* get the link contents into pagecache */
5062const char *page_get_link(struct dentry *dentry, struct inode *inode,
5063			  struct delayed_call *callback)
5064{
5065	char *kaddr;
5066	struct page *page;
5067	struct address_space *mapping = inode->i_mapping;
5068
5069	if (!dentry) {
5070		page = find_get_page(mapping, 0);
5071		if (!page)
5072			return ERR_PTR(-ECHILD);
5073		if (!PageUptodate(page)) {
5074			put_page(page);
5075			return ERR_PTR(-ECHILD);
5076		}
5077	} else {
5078		page = read_mapping_page(mapping, 0, NULL);
5079		if (IS_ERR(page))
5080			return (char*)page;
5081	}
5082	set_delayed_call(callback, page_put_link, page);
5083	BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
5084	kaddr = page_address(page);
5085	nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
5086	return kaddr;
5087}
5088
5089EXPORT_SYMBOL(page_get_link);
5090
5091void page_put_link(void *arg)
5092{
5093	put_page(arg);
5094}
5095EXPORT_SYMBOL(page_put_link);
5096
5097int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
5098{
5099	DEFINE_DELAYED_CALL(done);
5100	int res = readlink_copy(buffer, buflen,
5101				page_get_link(dentry, d_inode(dentry),
5102					      &done));
5103	do_delayed_call(&done);
5104	return res;
5105}
5106EXPORT_SYMBOL(page_readlink);
5107
5108int page_symlink(struct inode *inode, const char *symname, int len)
5109{
5110	struct address_space *mapping = inode->i_mapping;
5111	const struct address_space_operations *aops = mapping->a_ops;
5112	bool nofs = !mapping_gfp_constraint(mapping, __GFP_FS);
5113	struct page *page;
5114	void *fsdata = NULL;
5115	int err;
5116	unsigned int flags;
5117
5118retry:
5119	if (nofs)
5120		flags = memalloc_nofs_save();
5121	err = aops->write_begin(NULL, mapping, 0, len-1, &page, &fsdata);
5122	if (nofs)
5123		memalloc_nofs_restore(flags);
5124	if (err)
5125		goto fail;
5126
5127	memcpy(page_address(page), symname, len-1);
5128
5129	err = aops->write_end(NULL, mapping, 0, len-1, len-1,
5130							page, fsdata);
5131	if (err < 0)
5132		goto fail;
5133	if (err < len-1)
5134		goto retry;
5135
5136	mark_inode_dirty(inode);
5137	return 0;
5138fail:
5139	return err;
5140}
5141EXPORT_SYMBOL(page_symlink);
5142
5143const struct inode_operations page_symlink_inode_operations = {
5144	.get_link	= page_get_link,
5145};
5146EXPORT_SYMBOL(page_symlink_inode_operations);