Linux Audio

Check our new training course

Loading...
v6.8
   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/filelock.h>
  24#include <linux/namei.h>
  25#include <linux/pagemap.h>
  26#include <linux/sched/mm.h>
  27#include <linux/fsnotify.h>
  28#include <linux/personality.h>
  29#include <linux/security.h>
  30#include <linux/ima.h>
  31#include <linux/syscalls.h>
  32#include <linux/mount.h>
  33#include <linux/audit.h>
  34#include <linux/capability.h>
  35#include <linux/file.h>
  36#include <linux/fcntl.h>
  37#include <linux/device_cgroup.h>
  38#include <linux/fs_struct.h>
  39#include <linux/posix_acl.h>
  40#include <linux/hash.h>
  41#include <linux/bitops.h>
  42#include <linux/init_task.h>
  43#include <linux/uaccess.h>
  44
  45#include "internal.h"
  46#include "mount.h"
  47
  48/* [Feb-1997 T. Schoebel-Theuer]
  49 * Fundamental changes in the pathname lookup mechanisms (namei)
  50 * were necessary because of omirr.  The reason is that omirr needs
  51 * to know the _real_ pathname, not the user-supplied one, in case
  52 * of symlinks (and also when transname replacements occur).
  53 *
  54 * The new code replaces the old recursive symlink resolution with
  55 * an iterative one (in case of non-nested symlink chains).  It does
  56 * this with calls to <fs>_follow_link().
  57 * As a side effect, dir_namei(), _namei() and follow_link() are now 
  58 * replaced with a single function lookup_dentry() that can handle all 
  59 * the special cases of the former code.
  60 *
  61 * With the new dcache, the pathname is stored at each inode, at least as
  62 * long as the refcount of the inode is positive.  As a side effect, the
  63 * size of the dcache depends on the inode cache and thus is dynamic.
  64 *
  65 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
  66 * resolution to correspond with current state of the code.
  67 *
  68 * Note that the symlink resolution is not *completely* iterative.
  69 * There is still a significant amount of tail- and mid- recursion in
  70 * the algorithm.  Also, note that <fs>_readlink() is not used in
  71 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
  72 * may return different results than <fs>_follow_link().  Many virtual
  73 * filesystems (including /proc) exhibit this behavior.
  74 */
  75
  76/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
  77 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
  78 * and the name already exists in form of a symlink, try to create the new
  79 * name indicated by the symlink. The old code always complained that the
  80 * name already exists, due to not following the symlink even if its target
  81 * is nonexistent.  The new semantics affects also mknod() and link() when
  82 * the name is a symlink pointing to a non-existent name.
  83 *
  84 * I don't know which semantics is the right one, since I have no access
  85 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
  86 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
  87 * "old" one. Personally, I think the new semantics is much more logical.
  88 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
  89 * file does succeed in both HP-UX and SunOs, but not in Solaris
  90 * and in the old Linux semantics.
  91 */
  92
  93/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
  94 * semantics.  See the comments in "open_namei" and "do_link" below.
  95 *
  96 * [10-Sep-98 Alan Modra] Another symlink change.
  97 */
  98
  99/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
 100 *	inside the path - always follow.
 101 *	in the last component in creation/removal/renaming - never follow.
 102 *	if LOOKUP_FOLLOW passed - follow.
 103 *	if the pathname has trailing slashes - follow.
 104 *	otherwise - don't follow.
 105 * (applied in that order).
 106 *
 107 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
 108 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
 109 * During the 2.4 we need to fix the userland stuff depending on it -
 110 * hopefully we will be able to get rid of that wart in 2.5. So far only
 111 * XEmacs seems to be relying on it...
 112 */
 113/*
 114 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
 115 * implemented.  Let's see if raised priority of ->s_vfs_rename_mutex gives
 116 * any extra contention...
 117 */
 118
 119/* In order to reduce some races, while at the same time doing additional
 120 * checking and hopefully speeding things up, we copy filenames to the
 121 * kernel data space before using them..
 122 *
 123 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
 124 * PATH_MAX includes the nul terminator --RR.
 125 */
 126
 127#define EMBEDDED_NAME_MAX	(PATH_MAX - offsetof(struct filename, iname))
 128
 129struct filename *
 130getname_flags(const char __user *filename, int flags, int *empty)
 131{
 132	struct filename *result;
 133	char *kname;
 134	int len;
 135
 136	result = audit_reusename(filename);
 137	if (result)
 138		return result;
 139
 140	result = __getname();
 141	if (unlikely(!result))
 142		return ERR_PTR(-ENOMEM);
 143
 144	/*
 145	 * First, try to embed the struct filename inside the names_cache
 146	 * allocation
 147	 */
 148	kname = (char *)result->iname;
 149	result->name = kname;
 150
 151	len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX);
 152	if (unlikely(len < 0)) {
 153		__putname(result);
 154		return ERR_PTR(len);
 155	}
 156
 157	/*
 158	 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
 159	 * separate struct filename so we can dedicate the entire
 160	 * names_cache allocation for the pathname, and re-do the copy from
 161	 * userland.
 162	 */
 163	if (unlikely(len == EMBEDDED_NAME_MAX)) {
 164		const size_t size = offsetof(struct filename, iname[1]);
 165		kname = (char *)result;
 166
 167		/*
 168		 * size is chosen that way we to guarantee that
 169		 * result->iname[0] is within the same object and that
 170		 * kname can't be equal to result->iname, no matter what.
 171		 */
 172		result = kzalloc(size, GFP_KERNEL);
 173		if (unlikely(!result)) {
 174			__putname(kname);
 175			return ERR_PTR(-ENOMEM);
 176		}
 177		result->name = kname;
 178		len = strncpy_from_user(kname, filename, PATH_MAX);
 179		if (unlikely(len < 0)) {
 180			__putname(kname);
 181			kfree(result);
 182			return ERR_PTR(len);
 183		}
 184		if (unlikely(len == PATH_MAX)) {
 185			__putname(kname);
 186			kfree(result);
 187			return ERR_PTR(-ENAMETOOLONG);
 188		}
 189	}
 190
 191	atomic_set(&result->refcnt, 1);
 192	/* The empty path is special. */
 193	if (unlikely(!len)) {
 194		if (empty)
 195			*empty = 1;
 196		if (!(flags & LOOKUP_EMPTY)) {
 197			putname(result);
 198			return ERR_PTR(-ENOENT);
 199		}
 200	}
 201
 202	result->uptr = filename;
 203	result->aname = NULL;
 204	audit_getname(result);
 205	return result;
 206}
 207
 208struct filename *
 209getname_uflags(const char __user *filename, int uflags)
 210{
 211	int flags = (uflags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
 212
 213	return getname_flags(filename, flags, NULL);
 214}
 215
 216struct filename *
 217getname(const char __user * filename)
 218{
 219	return getname_flags(filename, 0, NULL);
 220}
 221
 222struct filename *
 223getname_kernel(const char * filename)
 224{
 225	struct filename *result;
 226	int len = strlen(filename) + 1;
 227
 228	result = __getname();
 229	if (unlikely(!result))
 230		return ERR_PTR(-ENOMEM);
 231
 232	if (len <= EMBEDDED_NAME_MAX) {
 233		result->name = (char *)result->iname;
 234	} else if (len <= PATH_MAX) {
 235		const size_t size = offsetof(struct filename, iname[1]);
 236		struct filename *tmp;
 237
 238		tmp = kmalloc(size, GFP_KERNEL);
 239		if (unlikely(!tmp)) {
 240			__putname(result);
 241			return ERR_PTR(-ENOMEM);
 242		}
 243		tmp->name = (char *)result;
 244		result = tmp;
 245	} else {
 246		__putname(result);
 247		return ERR_PTR(-ENAMETOOLONG);
 248	}
 249	memcpy((char *)result->name, filename, len);
 250	result->uptr = NULL;
 251	result->aname = NULL;
 252	atomic_set(&result->refcnt, 1);
 253	audit_getname(result);
 254
 255	return result;
 256}
 257EXPORT_SYMBOL(getname_kernel);
 258
 259void putname(struct filename *name)
 260{
 261	if (IS_ERR(name))
 262		return;
 263
 264	if (WARN_ON_ONCE(!atomic_read(&name->refcnt)))
 265		return;
 266
 267	if (!atomic_dec_and_test(&name->refcnt))
 268		return;
 269
 270	if (name->name != name->iname) {
 271		__putname(name->name);
 272		kfree(name);
 273	} else
 274		__putname(name);
 275}
 276EXPORT_SYMBOL(putname);
 277
 278/**
 279 * check_acl - perform ACL permission checking
 280 * @idmap:	idmap of the mount the inode was found from
 281 * @inode:	inode to check permissions on
 282 * @mask:	right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
 283 *
 284 * This function performs the ACL permission checking. Since this function
 285 * retrieve POSIX acls it needs to know whether it is called from a blocking or
 286 * non-blocking context and thus cares about the MAY_NOT_BLOCK bit.
 287 *
 288 * If the inode has been found through an idmapped mount the idmap of
 289 * the vfsmount must be passed through @idmap. This function will then take
 290 * care to map the inode according to @idmap before checking permissions.
 291 * On non-idmapped mounts or if permission checking is to be performed on the
 292 * raw inode simply pass @nop_mnt_idmap.
 293 */
 294static int check_acl(struct mnt_idmap *idmap,
 295		     struct inode *inode, int mask)
 296{
 297#ifdef CONFIG_FS_POSIX_ACL
 298	struct posix_acl *acl;
 299
 300	if (mask & MAY_NOT_BLOCK) {
 301		acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
 302	        if (!acl)
 303	                return -EAGAIN;
 304		/* no ->get_inode_acl() calls in RCU mode... */
 305		if (is_uncached_acl(acl))
 306			return -ECHILD;
 307	        return posix_acl_permission(idmap, inode, acl, mask);
 308	}
 309
 310	acl = get_inode_acl(inode, ACL_TYPE_ACCESS);
 311	if (IS_ERR(acl))
 312		return PTR_ERR(acl);
 313	if (acl) {
 314	        int error = posix_acl_permission(idmap, inode, acl, mask);
 315	        posix_acl_release(acl);
 316	        return error;
 317	}
 318#endif
 319
 320	return -EAGAIN;
 321}
 322
 323/**
 324 * acl_permission_check - perform basic UNIX permission checking
 325 * @idmap:	idmap of the mount the inode was found from
 326 * @inode:	inode to check permissions on
 327 * @mask:	right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
 328 *
 329 * This function performs the basic UNIX permission checking. Since this
 330 * function may retrieve POSIX acls it needs to know whether it is called from a
 331 * blocking or non-blocking context and thus cares about the MAY_NOT_BLOCK bit.
 332 *
 333 * If the inode has been found through an idmapped mount the idmap of
 334 * the vfsmount must be passed through @idmap. This function will then take
 335 * care to map the inode according to @idmap before checking permissions.
 336 * On non-idmapped mounts or if permission checking is to be performed on the
 337 * raw inode simply pass @nop_mnt_idmap.
 338 */
 339static int acl_permission_check(struct mnt_idmap *idmap,
 340				struct inode *inode, int mask)
 341{
 342	unsigned int mode = inode->i_mode;
 343	vfsuid_t vfsuid;
 344
 345	/* Are we the owner? If so, ACL's don't matter */
 346	vfsuid = i_uid_into_vfsuid(idmap, inode);
 347	if (likely(vfsuid_eq_kuid(vfsuid, current_fsuid()))) {
 348		mask &= 7;
 349		mode >>= 6;
 350		return (mask & ~mode) ? -EACCES : 0;
 351	}
 352
 353	/* Do we have ACL's? */
 354	if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
 355		int error = check_acl(idmap, inode, mask);
 356		if (error != -EAGAIN)
 357			return error;
 358	}
 359
 360	/* Only RWX matters for group/other mode bits */
 361	mask &= 7;
 362
 363	/*
 364	 * Are the group permissions different from
 365	 * the other permissions in the bits we care
 366	 * about? Need to check group ownership if so.
 367	 */
 368	if (mask & (mode ^ (mode >> 3))) {
 369		vfsgid_t vfsgid = i_gid_into_vfsgid(idmap, inode);
 370		if (vfsgid_in_group_p(vfsgid))
 371			mode >>= 3;
 372	}
 373
 374	/* Bits in 'mode' clear that we require? */
 375	return (mask & ~mode) ? -EACCES : 0;
 376}
 377
 378/**
 379 * generic_permission -  check for access rights on a Posix-like filesystem
 380 * @idmap:	idmap of the mount the inode was found from
 381 * @inode:	inode to check access rights for
 382 * @mask:	right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC,
 383 *		%MAY_NOT_BLOCK ...)
 384 *
 385 * Used to check for read/write/execute permissions on a file.
 386 * We use "fsuid" for this, letting us set arbitrary permissions
 387 * for filesystem access without changing the "normal" uids which
 388 * are used for other things.
 389 *
 390 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
 391 * request cannot be satisfied (eg. requires blocking or too much complexity).
 392 * It would then be called again in ref-walk mode.
 393 *
 394 * If the inode has been found through an idmapped mount the idmap of
 395 * the vfsmount must be passed through @idmap. This function will then take
 396 * care to map the inode according to @idmap before checking permissions.
 397 * On non-idmapped mounts or if permission checking is to be performed on the
 398 * raw inode simply pass @nop_mnt_idmap.
 399 */
 400int generic_permission(struct mnt_idmap *idmap, struct inode *inode,
 401		       int mask)
 402{
 403	int ret;
 404
 405	/*
 406	 * Do the basic permission checks.
 407	 */
 408	ret = acl_permission_check(idmap, inode, mask);
 409	if (ret != -EACCES)
 410		return ret;
 411
 412	if (S_ISDIR(inode->i_mode)) {
 413		/* DACs are overridable for directories */
 414		if (!(mask & MAY_WRITE))
 415			if (capable_wrt_inode_uidgid(idmap, inode,
 416						     CAP_DAC_READ_SEARCH))
 417				return 0;
 418		if (capable_wrt_inode_uidgid(idmap, inode,
 419					     CAP_DAC_OVERRIDE))
 420			return 0;
 421		return -EACCES;
 422	}
 423
 424	/*
 425	 * Searching includes executable on directories, else just read.
 426	 */
 427	mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
 428	if (mask == MAY_READ)
 429		if (capable_wrt_inode_uidgid(idmap, inode,
 430					     CAP_DAC_READ_SEARCH))
 431			return 0;
 432	/*
 433	 * Read/write DACs are always overridable.
 434	 * Executable DACs are overridable when there is
 435	 * at least one exec bit set.
 436	 */
 437	if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
 438		if (capable_wrt_inode_uidgid(idmap, inode,
 439					     CAP_DAC_OVERRIDE))
 440			return 0;
 441
 442	return -EACCES;
 443}
 444EXPORT_SYMBOL(generic_permission);
 445
 446/**
 447 * do_inode_permission - UNIX permission checking
 448 * @idmap:	idmap of the mount the inode was found from
 449 * @inode:	inode to check permissions on
 450 * @mask:	right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
 451 *
 452 * We _really_ want to just do "generic_permission()" without
 453 * even looking at the inode->i_op values. So we keep a cache
 454 * flag in inode->i_opflags, that says "this has not special
 455 * permission function, use the fast case".
 456 */
 457static inline int do_inode_permission(struct mnt_idmap *idmap,
 458				      struct inode *inode, int mask)
 459{
 460	if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
 461		if (likely(inode->i_op->permission))
 462			return inode->i_op->permission(idmap, inode, mask);
 463
 464		/* This gets set once for the inode lifetime */
 465		spin_lock(&inode->i_lock);
 466		inode->i_opflags |= IOP_FASTPERM;
 467		spin_unlock(&inode->i_lock);
 468	}
 469	return generic_permission(idmap, inode, mask);
 470}
 471
 472/**
 473 * sb_permission - Check superblock-level permissions
 474 * @sb: Superblock of inode to check permission on
 475 * @inode: Inode to check permission on
 476 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
 477 *
 478 * Separate out file-system wide checks from inode-specific permission checks.
 479 */
 480static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
 481{
 482	if (unlikely(mask & MAY_WRITE)) {
 483		umode_t mode = inode->i_mode;
 484
 485		/* Nobody gets write access to a read-only fs. */
 486		if (sb_rdonly(sb) && (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
 487			return -EROFS;
 488	}
 489	return 0;
 490}
 491
 492/**
 493 * inode_permission - Check for access rights to a given inode
 494 * @idmap:	idmap of the mount the inode was found from
 495 * @inode:	Inode to check permission on
 496 * @mask:	Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
 497 *
 498 * Check for read/write/execute permissions on an inode.  We use fs[ug]id for
 499 * this, letting us set arbitrary permissions for filesystem access without
 500 * changing the "normal" UIDs which are used for other things.
 501 *
 502 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
 503 */
 504int inode_permission(struct mnt_idmap *idmap,
 505		     struct inode *inode, int mask)
 506{
 507	int retval;
 508
 509	retval = sb_permission(inode->i_sb, inode, mask);
 510	if (retval)
 511		return retval;
 512
 513	if (unlikely(mask & MAY_WRITE)) {
 514		/*
 515		 * Nobody gets write access to an immutable file.
 516		 */
 517		if (IS_IMMUTABLE(inode))
 518			return -EPERM;
 519
 520		/*
 521		 * Updating mtime will likely cause i_uid and i_gid to be
 522		 * written back improperly if their true value is unknown
 523		 * to the vfs.
 524		 */
 525		if (HAS_UNMAPPED_ID(idmap, inode))
 526			return -EACCES;
 527	}
 528
 529	retval = do_inode_permission(idmap, inode, mask);
 530	if (retval)
 531		return retval;
 532
 533	retval = devcgroup_inode_permission(inode, mask);
 534	if (retval)
 535		return retval;
 536
 537	return security_inode_permission(inode, mask);
 538}
 539EXPORT_SYMBOL(inode_permission);
 540
 541/**
 542 * path_get - get a reference to a path
 543 * @path: path to get the reference to
 544 *
 545 * Given a path increment the reference count to the dentry and the vfsmount.
 546 */
 547void path_get(const struct path *path)
 548{
 549	mntget(path->mnt);
 550	dget(path->dentry);
 551}
 552EXPORT_SYMBOL(path_get);
 553
 554/**
 555 * path_put - put a reference to a path
 556 * @path: path to put the reference to
 557 *
 558 * Given a path decrement the reference count to the dentry and the vfsmount.
 559 */
 560void path_put(const struct path *path)
 561{
 562	dput(path->dentry);
 563	mntput(path->mnt);
 564}
 565EXPORT_SYMBOL(path_put);
 566
 567#define EMBEDDED_LEVELS 2
 568struct nameidata {
 569	struct path	path;
 570	struct qstr	last;
 571	struct path	root;
 572	struct inode	*inode; /* path.dentry.d_inode */
 573	unsigned int	flags, state;
 574	unsigned	seq, next_seq, m_seq, r_seq;
 575	int		last_type;
 576	unsigned	depth;
 577	int		total_link_count;
 578	struct saved {
 579		struct path link;
 580		struct delayed_call done;
 581		const char *name;
 582		unsigned seq;
 583	} *stack, internal[EMBEDDED_LEVELS];
 584	struct filename	*name;
 585	struct nameidata *saved;
 586	unsigned	root_seq;
 587	int		dfd;
 588	vfsuid_t	dir_vfsuid;
 589	umode_t		dir_mode;
 590} __randomize_layout;
 591
 592#define ND_ROOT_PRESET 1
 593#define ND_ROOT_GRABBED 2
 594#define ND_JUMPED 4
 595
 596static void __set_nameidata(struct nameidata *p, int dfd, struct filename *name)
 597{
 598	struct nameidata *old = current->nameidata;
 599	p->stack = p->internal;
 600	p->depth = 0;
 601	p->dfd = dfd;
 602	p->name = name;
 603	p->path.mnt = NULL;
 604	p->path.dentry = NULL;
 605	p->total_link_count = old ? old->total_link_count : 0;
 606	p->saved = old;
 607	current->nameidata = p;
 608}
 609
 610static inline void set_nameidata(struct nameidata *p, int dfd, struct filename *name,
 611			  const struct path *root)
 612{
 613	__set_nameidata(p, dfd, name);
 614	p->state = 0;
 615	if (unlikely(root)) {
 616		p->state = ND_ROOT_PRESET;
 617		p->root = *root;
 618	}
 619}
 620
 621static void restore_nameidata(void)
 622{
 623	struct nameidata *now = current->nameidata, *old = now->saved;
 624
 625	current->nameidata = old;
 626	if (old)
 627		old->total_link_count = now->total_link_count;
 628	if (now->stack != now->internal)
 629		kfree(now->stack);
 630}
 631
 632static bool nd_alloc_stack(struct nameidata *nd)
 633{
 634	struct saved *p;
 635
 636	p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved),
 637			 nd->flags & LOOKUP_RCU ? GFP_ATOMIC : GFP_KERNEL);
 638	if (unlikely(!p))
 639		return false;
 640	memcpy(p, nd->internal, sizeof(nd->internal));
 641	nd->stack = p;
 642	return true;
 643}
 644
 645/**
 646 * path_connected - Verify that a dentry is below mnt.mnt_root
 647 * @mnt: The mountpoint to check.
 648 * @dentry: The dentry to check.
 649 *
 650 * Rename can sometimes move a file or directory outside of a bind
 651 * mount, path_connected allows those cases to be detected.
 652 */
 653static bool path_connected(struct vfsmount *mnt, struct dentry *dentry)
 654{
 655	struct super_block *sb = mnt->mnt_sb;
 656
 657	/* Bind mounts can have disconnected paths */
 658	if (mnt->mnt_root == sb->s_root)
 659		return true;
 660
 661	return is_subdir(dentry, mnt->mnt_root);
 662}
 663
 664static void drop_links(struct nameidata *nd)
 665{
 666	int i = nd->depth;
 667	while (i--) {
 668		struct saved *last = nd->stack + i;
 669		do_delayed_call(&last->done);
 670		clear_delayed_call(&last->done);
 671	}
 672}
 673
 674static void leave_rcu(struct nameidata *nd)
 675{
 676	nd->flags &= ~LOOKUP_RCU;
 677	nd->seq = nd->next_seq = 0;
 678	rcu_read_unlock();
 679}
 680
 681static void terminate_walk(struct nameidata *nd)
 682{
 683	drop_links(nd);
 684	if (!(nd->flags & LOOKUP_RCU)) {
 685		int i;
 686		path_put(&nd->path);
 687		for (i = 0; i < nd->depth; i++)
 688			path_put(&nd->stack[i].link);
 689		if (nd->state & ND_ROOT_GRABBED) {
 690			path_put(&nd->root);
 691			nd->state &= ~ND_ROOT_GRABBED;
 692		}
 693	} else {
 694		leave_rcu(nd);
 695	}
 696	nd->depth = 0;
 697	nd->path.mnt = NULL;
 698	nd->path.dentry = NULL;
 699}
 700
 701/* path_put is needed afterwards regardless of success or failure */
 702static bool __legitimize_path(struct path *path, unsigned seq, unsigned mseq)
 703{
 704	int res = __legitimize_mnt(path->mnt, mseq);
 705	if (unlikely(res)) {
 706		if (res > 0)
 707			path->mnt = NULL;
 708		path->dentry = NULL;
 709		return false;
 710	}
 711	if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
 712		path->dentry = NULL;
 713		return false;
 714	}
 715	return !read_seqcount_retry(&path->dentry->d_seq, seq);
 716}
 717
 718static inline bool legitimize_path(struct nameidata *nd,
 719			    struct path *path, unsigned seq)
 720{
 721	return __legitimize_path(path, seq, nd->m_seq);
 722}
 723
 724static bool legitimize_links(struct nameidata *nd)
 725{
 726	int i;
 727	if (unlikely(nd->flags & LOOKUP_CACHED)) {
 728		drop_links(nd);
 729		nd->depth = 0;
 730		return false;
 731	}
 732	for (i = 0; i < nd->depth; i++) {
 733		struct saved *last = nd->stack + i;
 734		if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
 735			drop_links(nd);
 736			nd->depth = i + 1;
 737			return false;
 738		}
 739	}
 740	return true;
 741}
 742
 743static bool legitimize_root(struct nameidata *nd)
 744{
 745	/* Nothing to do if nd->root is zero or is managed by the VFS user. */
 746	if (!nd->root.mnt || (nd->state & ND_ROOT_PRESET))
 747		return true;
 748	nd->state |= ND_ROOT_GRABBED;
 749	return legitimize_path(nd, &nd->root, nd->root_seq);
 750}
 751
 752/*
 753 * Path walking has 2 modes, rcu-walk and ref-walk (see
 754 * Documentation/filesystems/path-lookup.txt).  In situations when we can't
 755 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
 756 * normal reference counts on dentries and vfsmounts to transition to ref-walk
 757 * mode.  Refcounts are grabbed at the last known good point before rcu-walk
 758 * got stuck, so ref-walk may continue from there. If this is not successful
 759 * (eg. a seqcount has changed), then failure is returned and it's up to caller
 760 * to restart the path walk from the beginning in ref-walk mode.
 761 */
 762
 763/**
 764 * try_to_unlazy - try to switch to ref-walk mode.
 765 * @nd: nameidata pathwalk data
 766 * Returns: true on success, false on failure
 767 *
 768 * try_to_unlazy attempts to legitimize the current nd->path and nd->root
 769 * for ref-walk mode.
 770 * Must be called from rcu-walk context.
 771 * Nothing should touch nameidata between try_to_unlazy() failure and
 772 * terminate_walk().
 773 */
 774static bool try_to_unlazy(struct nameidata *nd)
 775{
 776	struct dentry *parent = nd->path.dentry;
 777
 778	BUG_ON(!(nd->flags & LOOKUP_RCU));
 779
 780	if (unlikely(!legitimize_links(nd)))
 781		goto out1;
 782	if (unlikely(!legitimize_path(nd, &nd->path, nd->seq)))
 783		goto out;
 784	if (unlikely(!legitimize_root(nd)))
 785		goto out;
 786	leave_rcu(nd);
 787	BUG_ON(nd->inode != parent->d_inode);
 788	return true;
 789
 790out1:
 791	nd->path.mnt = NULL;
 792	nd->path.dentry = NULL;
 793out:
 794	leave_rcu(nd);
 795	return false;
 796}
 797
 798/**
 799 * try_to_unlazy_next - try to switch to ref-walk mode.
 800 * @nd: nameidata pathwalk data
 801 * @dentry: next dentry to step into
 802 * Returns: true on success, false on failure
 803 *
 804 * Similar to try_to_unlazy(), but here we have the next dentry already
 805 * picked by rcu-walk and want to legitimize that in addition to the current
 806 * nd->path and nd->root for ref-walk mode.  Must be called from rcu-walk context.
 807 * Nothing should touch nameidata between try_to_unlazy_next() failure and
 808 * terminate_walk().
 809 */
 810static bool try_to_unlazy_next(struct nameidata *nd, struct dentry *dentry)
 811{
 812	int res;
 813	BUG_ON(!(nd->flags & LOOKUP_RCU));
 814
 815	if (unlikely(!legitimize_links(nd)))
 816		goto out2;
 817	res = __legitimize_mnt(nd->path.mnt, nd->m_seq);
 818	if (unlikely(res)) {
 819		if (res > 0)
 820			goto out2;
 821		goto out1;
 822	}
 823	if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref)))
 824		goto out1;
 825
 826	/*
 827	 * We need to move both the parent and the dentry from the RCU domain
 828	 * to be properly refcounted. And the sequence number in the dentry
 829	 * validates *both* dentry counters, since we checked the sequence
 830	 * number of the parent after we got the child sequence number. So we
 831	 * know the parent must still be valid if the child sequence number is
 832	 */
 833	if (unlikely(!lockref_get_not_dead(&dentry->d_lockref)))
 834		goto out;
 835	if (read_seqcount_retry(&dentry->d_seq, nd->next_seq))
 836		goto out_dput;
 837	/*
 838	 * Sequence counts matched. Now make sure that the root is
 839	 * still valid and get it if required.
 840	 */
 841	if (unlikely(!legitimize_root(nd)))
 842		goto out_dput;
 843	leave_rcu(nd);
 844	return true;
 845
 846out2:
 847	nd->path.mnt = NULL;
 848out1:
 849	nd->path.dentry = NULL;
 850out:
 851	leave_rcu(nd);
 852	return false;
 853out_dput:
 854	leave_rcu(nd);
 855	dput(dentry);
 856	return false;
 857}
 858
 859static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
 860{
 861	if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
 862		return dentry->d_op->d_revalidate(dentry, flags);
 863	else
 864		return 1;
 865}
 866
 867/**
 868 * complete_walk - successful completion of path walk
 869 * @nd:  pointer nameidata
 870 *
 871 * If we had been in RCU mode, drop out of it and legitimize nd->path.
 872 * Revalidate the final result, unless we'd already done that during
 873 * the path walk or the filesystem doesn't ask for it.  Return 0 on
 874 * success, -error on failure.  In case of failure caller does not
 875 * need to drop nd->path.
 876 */
 877static int complete_walk(struct nameidata *nd)
 878{
 879	struct dentry *dentry = nd->path.dentry;
 880	int status;
 881
 882	if (nd->flags & LOOKUP_RCU) {
 883		/*
 884		 * We don't want to zero nd->root for scoped-lookups or
 885		 * externally-managed nd->root.
 886		 */
 887		if (!(nd->state & ND_ROOT_PRESET))
 888			if (!(nd->flags & LOOKUP_IS_SCOPED))
 889				nd->root.mnt = NULL;
 890		nd->flags &= ~LOOKUP_CACHED;
 891		if (!try_to_unlazy(nd))
 892			return -ECHILD;
 893	}
 894
 895	if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) {
 896		/*
 897		 * While the guarantee of LOOKUP_IS_SCOPED is (roughly) "don't
 898		 * ever step outside the root during lookup" and should already
 899		 * be guaranteed by the rest of namei, we want to avoid a namei
 900		 * BUG resulting in userspace being given a path that was not
 901		 * scoped within the root at some point during the lookup.
 902		 *
 903		 * So, do a final sanity-check to make sure that in the
 904		 * worst-case scenario (a complete bypass of LOOKUP_IS_SCOPED)
 905		 * we won't silently return an fd completely outside of the
 906		 * requested root to userspace.
 907		 *
 908		 * Userspace could move the path outside the root after this
 909		 * check, but as discussed elsewhere this is not a concern (the
 910		 * resolved file was inside the root at some point).
 911		 */
 912		if (!path_is_under(&nd->path, &nd->root))
 913			return -EXDEV;
 914	}
 915
 916	if (likely(!(nd->state & ND_JUMPED)))
 917		return 0;
 918
 919	if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
 920		return 0;
 921
 922	status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
 923	if (status > 0)
 924		return 0;
 925
 926	if (!status)
 927		status = -ESTALE;
 928
 929	return status;
 930}
 931
 932static int set_root(struct nameidata *nd)
 933{
 934	struct fs_struct *fs = current->fs;
 935
 936	/*
 937	 * Jumping to the real root in a scoped-lookup is a BUG in namei, but we
 938	 * still have to ensure it doesn't happen because it will cause a breakout
 939	 * from the dirfd.
 940	 */
 941	if (WARN_ON(nd->flags & LOOKUP_IS_SCOPED))
 942		return -ENOTRECOVERABLE;
 943
 944	if (nd->flags & LOOKUP_RCU) {
 945		unsigned seq;
 946
 947		do {
 948			seq = read_seqcount_begin(&fs->seq);
 949			nd->root = fs->root;
 950			nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
 951		} while (read_seqcount_retry(&fs->seq, seq));
 952	} else {
 953		get_fs_root(fs, &nd->root);
 954		nd->state |= ND_ROOT_GRABBED;
 955	}
 956	return 0;
 957}
 958
 959static int nd_jump_root(struct nameidata *nd)
 960{
 961	if (unlikely(nd->flags & LOOKUP_BENEATH))
 962		return -EXDEV;
 963	if (unlikely(nd->flags & LOOKUP_NO_XDEV)) {
 964		/* Absolute path arguments to path_init() are allowed. */
 965		if (nd->path.mnt != NULL && nd->path.mnt != nd->root.mnt)
 966			return -EXDEV;
 967	}
 968	if (!nd->root.mnt) {
 969		int error = set_root(nd);
 970		if (error)
 971			return error;
 972	}
 973	if (nd->flags & LOOKUP_RCU) {
 974		struct dentry *d;
 975		nd->path = nd->root;
 976		d = nd->path.dentry;
 977		nd->inode = d->d_inode;
 978		nd->seq = nd->root_seq;
 979		if (read_seqcount_retry(&d->d_seq, nd->seq))
 980			return -ECHILD;
 981	} else {
 982		path_put(&nd->path);
 983		nd->path = nd->root;
 984		path_get(&nd->path);
 985		nd->inode = nd->path.dentry->d_inode;
 986	}
 987	nd->state |= ND_JUMPED;
 988	return 0;
 989}
 990
 991/*
 992 * Helper to directly jump to a known parsed path from ->get_link,
 993 * caller must have taken a reference to path beforehand.
 994 */
 995int nd_jump_link(const struct path *path)
 996{
 997	int error = -ELOOP;
 998	struct nameidata *nd = current->nameidata;
 999
1000	if (unlikely(nd->flags & LOOKUP_NO_MAGICLINKS))
1001		goto err;
1002
1003	error = -EXDEV;
1004	if (unlikely(nd->flags & LOOKUP_NO_XDEV)) {
1005		if (nd->path.mnt != path->mnt)
1006			goto err;
1007	}
1008	/* Not currently safe for scoped-lookups. */
1009	if (unlikely(nd->flags & LOOKUP_IS_SCOPED))
1010		goto err;
1011
1012	path_put(&nd->path);
1013	nd->path = *path;
1014	nd->inode = nd->path.dentry->d_inode;
1015	nd->state |= ND_JUMPED;
1016	return 0;
1017
1018err:
1019	path_put(path);
1020	return error;
1021}
1022
1023static inline void put_link(struct nameidata *nd)
1024{
1025	struct saved *last = nd->stack + --nd->depth;
1026	do_delayed_call(&last->done);
1027	if (!(nd->flags & LOOKUP_RCU))
1028		path_put(&last->link);
1029}
1030
1031static int sysctl_protected_symlinks __read_mostly;
1032static int sysctl_protected_hardlinks __read_mostly;
1033static int sysctl_protected_fifos __read_mostly;
1034static int sysctl_protected_regular __read_mostly;
1035
1036#ifdef CONFIG_SYSCTL
1037static struct ctl_table namei_sysctls[] = {
1038	{
1039		.procname	= "protected_symlinks",
1040		.data		= &sysctl_protected_symlinks,
1041		.maxlen		= sizeof(int),
1042		.mode		= 0644,
1043		.proc_handler	= proc_dointvec_minmax,
1044		.extra1		= SYSCTL_ZERO,
1045		.extra2		= SYSCTL_ONE,
1046	},
1047	{
1048		.procname	= "protected_hardlinks",
1049		.data		= &sysctl_protected_hardlinks,
1050		.maxlen		= sizeof(int),
1051		.mode		= 0644,
1052		.proc_handler	= proc_dointvec_minmax,
1053		.extra1		= SYSCTL_ZERO,
1054		.extra2		= SYSCTL_ONE,
1055	},
1056	{
1057		.procname	= "protected_fifos",
1058		.data		= &sysctl_protected_fifos,
1059		.maxlen		= sizeof(int),
1060		.mode		= 0644,
1061		.proc_handler	= proc_dointvec_minmax,
1062		.extra1		= SYSCTL_ZERO,
1063		.extra2		= SYSCTL_TWO,
1064	},
1065	{
1066		.procname	= "protected_regular",
1067		.data		= &sysctl_protected_regular,
1068		.maxlen		= sizeof(int),
1069		.mode		= 0644,
1070		.proc_handler	= proc_dointvec_minmax,
1071		.extra1		= SYSCTL_ZERO,
1072		.extra2		= SYSCTL_TWO,
1073	},
 
1074};
1075
1076static int __init init_fs_namei_sysctls(void)
1077{
1078	register_sysctl_init("fs", namei_sysctls);
1079	return 0;
1080}
1081fs_initcall(init_fs_namei_sysctls);
1082
1083#endif /* CONFIG_SYSCTL */
1084
1085/**
1086 * may_follow_link - Check symlink following for unsafe situations
1087 * @nd: nameidata pathwalk data
1088 * @inode: Used for idmapping.
1089 *
1090 * In the case of the sysctl_protected_symlinks sysctl being enabled,
1091 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
1092 * in a sticky world-writable directory. This is to protect privileged
1093 * processes from failing races against path names that may change out
1094 * from under them by way of other users creating malicious symlinks.
1095 * It will permit symlinks to be followed only when outside a sticky
1096 * world-writable directory, or when the uid of the symlink and follower
1097 * match, or when the directory owner matches the symlink's owner.
1098 *
1099 * Returns 0 if following the symlink is allowed, -ve on error.
1100 */
1101static inline int may_follow_link(struct nameidata *nd, const struct inode *inode)
1102{
1103	struct mnt_idmap *idmap;
1104	vfsuid_t vfsuid;
1105
1106	if (!sysctl_protected_symlinks)
1107		return 0;
1108
1109	idmap = mnt_idmap(nd->path.mnt);
1110	vfsuid = i_uid_into_vfsuid(idmap, inode);
1111	/* Allowed if owner and follower match. */
1112	if (vfsuid_eq_kuid(vfsuid, current_fsuid()))
1113		return 0;
1114
1115	/* Allowed if parent directory not sticky and world-writable. */
1116	if ((nd->dir_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
1117		return 0;
1118
1119	/* Allowed if parent directory and link owner match. */
1120	if (vfsuid_valid(nd->dir_vfsuid) && vfsuid_eq(nd->dir_vfsuid, vfsuid))
1121		return 0;
1122
1123	if (nd->flags & LOOKUP_RCU)
1124		return -ECHILD;
1125
1126	audit_inode(nd->name, nd->stack[0].link.dentry, 0);
1127	audit_log_path_denied(AUDIT_ANOM_LINK, "follow_link");
1128	return -EACCES;
1129}
1130
1131/**
1132 * safe_hardlink_source - Check for safe hardlink conditions
1133 * @idmap: idmap of the mount the inode was found from
1134 * @inode: the source inode to hardlink from
1135 *
1136 * Return false if at least one of the following conditions:
1137 *    - inode is not a regular file
1138 *    - inode is setuid
1139 *    - inode is setgid and group-exec
1140 *    - access failure for read and write
1141 *
1142 * Otherwise returns true.
1143 */
1144static bool safe_hardlink_source(struct mnt_idmap *idmap,
1145				 struct inode *inode)
1146{
1147	umode_t mode = inode->i_mode;
1148
1149	/* Special files should not get pinned to the filesystem. */
1150	if (!S_ISREG(mode))
1151		return false;
1152
1153	/* Setuid files should not get pinned to the filesystem. */
1154	if (mode & S_ISUID)
1155		return false;
1156
1157	/* Executable setgid files should not get pinned to the filesystem. */
1158	if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
1159		return false;
1160
1161	/* Hardlinking to unreadable or unwritable sources is dangerous. */
1162	if (inode_permission(idmap, inode, MAY_READ | MAY_WRITE))
1163		return false;
1164
1165	return true;
1166}
1167
1168/**
1169 * may_linkat - Check permissions for creating a hardlink
1170 * @idmap: idmap of the mount the inode was found from
1171 * @link:  the source to hardlink from
1172 *
1173 * Block hardlink when all of:
1174 *  - sysctl_protected_hardlinks enabled
1175 *  - fsuid does not match inode
1176 *  - hardlink source is unsafe (see safe_hardlink_source() above)
1177 *  - not CAP_FOWNER in a namespace with the inode owner uid mapped
1178 *
1179 * If the inode has been found through an idmapped mount the idmap of
1180 * the vfsmount must be passed through @idmap. This function will then take
1181 * care to map the inode according to @idmap before checking permissions.
1182 * On non-idmapped mounts or if permission checking is to be performed on the
1183 * raw inode simply pass @nop_mnt_idmap.
1184 *
1185 * Returns 0 if successful, -ve on error.
1186 */
1187int may_linkat(struct mnt_idmap *idmap, const struct path *link)
1188{
1189	struct inode *inode = link->dentry->d_inode;
1190
1191	/* Inode writeback is not safe when the uid or gid are invalid. */
1192	if (!vfsuid_valid(i_uid_into_vfsuid(idmap, inode)) ||
1193	    !vfsgid_valid(i_gid_into_vfsgid(idmap, inode)))
1194		return -EOVERFLOW;
1195
1196	if (!sysctl_protected_hardlinks)
1197		return 0;
1198
1199	/* Source inode owner (or CAP_FOWNER) can hardlink all they like,
1200	 * otherwise, it must be a safe source.
1201	 */
1202	if (safe_hardlink_source(idmap, inode) ||
1203	    inode_owner_or_capable(idmap, inode))
1204		return 0;
1205
1206	audit_log_path_denied(AUDIT_ANOM_LINK, "linkat");
1207	return -EPERM;
1208}
1209
1210/**
1211 * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory
1212 *			  should be allowed, or not, on files that already
1213 *			  exist.
1214 * @idmap: idmap of the mount the inode was found from
1215 * @nd: nameidata pathwalk data
1216 * @inode: the inode of the file to open
1217 *
1218 * Block an O_CREAT open of a FIFO (or a regular file) when:
1219 *   - sysctl_protected_fifos (or sysctl_protected_regular) is enabled
1220 *   - the file already exists
1221 *   - we are in a sticky directory
1222 *   - we don't own the file
1223 *   - the owner of the directory doesn't own the file
1224 *   - the directory is world writable
1225 * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2
1226 * the directory doesn't have to be world writable: being group writable will
1227 * be enough.
1228 *
1229 * If the inode has been found through an idmapped mount the idmap of
1230 * the vfsmount must be passed through @idmap. This function will then take
1231 * care to map the inode according to @idmap before checking permissions.
1232 * On non-idmapped mounts or if permission checking is to be performed on the
1233 * raw inode simply pass @nop_mnt_idmap.
1234 *
1235 * Returns 0 if the open is allowed, -ve on error.
1236 */
1237static int may_create_in_sticky(struct mnt_idmap *idmap,
1238				struct nameidata *nd, struct inode *const inode)
1239{
1240	umode_t dir_mode = nd->dir_mode;
1241	vfsuid_t dir_vfsuid = nd->dir_vfsuid;
1242
1243	if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) ||
1244	    (!sysctl_protected_regular && S_ISREG(inode->i_mode)) ||
1245	    likely(!(dir_mode & S_ISVTX)) ||
1246	    vfsuid_eq(i_uid_into_vfsuid(idmap, inode), dir_vfsuid) ||
1247	    vfsuid_eq_kuid(i_uid_into_vfsuid(idmap, inode), current_fsuid()))
1248		return 0;
1249
1250	if (likely(dir_mode & 0002) ||
1251	    (dir_mode & 0020 &&
1252	     ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) ||
1253	      (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) {
1254		const char *operation = S_ISFIFO(inode->i_mode) ?
1255					"sticky_create_fifo" :
1256					"sticky_create_regular";
1257		audit_log_path_denied(AUDIT_ANOM_CREAT, operation);
1258		return -EACCES;
1259	}
1260	return 0;
1261}
1262
1263/*
1264 * follow_up - Find the mountpoint of path's vfsmount
1265 *
1266 * Given a path, find the mountpoint of its source file system.
1267 * Replace @path with the path of the mountpoint in the parent mount.
1268 * Up is towards /.
1269 *
1270 * Return 1 if we went up a level and 0 if we were already at the
1271 * root.
1272 */
1273int follow_up(struct path *path)
1274{
1275	struct mount *mnt = real_mount(path->mnt);
1276	struct mount *parent;
1277	struct dentry *mountpoint;
1278
1279	read_seqlock_excl(&mount_lock);
1280	parent = mnt->mnt_parent;
1281	if (parent == mnt) {
1282		read_sequnlock_excl(&mount_lock);
1283		return 0;
1284	}
1285	mntget(&parent->mnt);
1286	mountpoint = dget(mnt->mnt_mountpoint);
1287	read_sequnlock_excl(&mount_lock);
1288	dput(path->dentry);
1289	path->dentry = mountpoint;
1290	mntput(path->mnt);
1291	path->mnt = &parent->mnt;
1292	return 1;
1293}
1294EXPORT_SYMBOL(follow_up);
1295
1296static bool choose_mountpoint_rcu(struct mount *m, const struct path *root,
1297				  struct path *path, unsigned *seqp)
1298{
1299	while (mnt_has_parent(m)) {
1300		struct dentry *mountpoint = m->mnt_mountpoint;
1301
1302		m = m->mnt_parent;
1303		if (unlikely(root->dentry == mountpoint &&
1304			     root->mnt == &m->mnt))
1305			break;
1306		if (mountpoint != m->mnt.mnt_root) {
1307			path->mnt = &m->mnt;
1308			path->dentry = mountpoint;
1309			*seqp = read_seqcount_begin(&mountpoint->d_seq);
1310			return true;
1311		}
1312	}
1313	return false;
1314}
1315
1316static bool choose_mountpoint(struct mount *m, const struct path *root,
1317			      struct path *path)
1318{
1319	bool found;
1320
1321	rcu_read_lock();
1322	while (1) {
1323		unsigned seq, mseq = read_seqbegin(&mount_lock);
1324
1325		found = choose_mountpoint_rcu(m, root, path, &seq);
1326		if (unlikely(!found)) {
1327			if (!read_seqretry(&mount_lock, mseq))
1328				break;
1329		} else {
1330			if (likely(__legitimize_path(path, seq, mseq)))
1331				break;
1332			rcu_read_unlock();
1333			path_put(path);
1334			rcu_read_lock();
1335		}
1336	}
1337	rcu_read_unlock();
1338	return found;
1339}
1340
1341/*
1342 * Perform an automount
1343 * - return -EISDIR to tell follow_managed() to stop and return the path we
1344 *   were called with.
1345 */
1346static int follow_automount(struct path *path, int *count, unsigned lookup_flags)
1347{
1348	struct dentry *dentry = path->dentry;
1349
1350	/* We don't want to mount if someone's just doing a stat -
1351	 * unless they're stat'ing a directory and appended a '/' to
1352	 * the name.
1353	 *
1354	 * We do, however, want to mount if someone wants to open or
1355	 * create a file of any type under the mountpoint, wants to
1356	 * traverse through the mountpoint or wants to open the
1357	 * mounted directory.  Also, autofs may mark negative dentries
1358	 * as being automount points.  These will need the attentions
1359	 * of the daemon to instantiate them before they can be used.
1360	 */
1361	if (!(lookup_flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1362			   LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1363	    dentry->d_inode)
1364		return -EISDIR;
1365
1366	if (count && (*count)++ >= MAXSYMLINKS)
1367		return -ELOOP;
1368
1369	return finish_automount(dentry->d_op->d_automount(path), path);
1370}
1371
1372/*
1373 * mount traversal - out-of-line part.  One note on ->d_flags accesses -
1374 * dentries are pinned but not locked here, so negative dentry can go
1375 * positive right under us.  Use of smp_load_acquire() provides a barrier
1376 * sufficient for ->d_inode and ->d_flags consistency.
1377 */
1378static int __traverse_mounts(struct path *path, unsigned flags, bool *jumped,
1379			     int *count, unsigned lookup_flags)
1380{
1381	struct vfsmount *mnt = path->mnt;
1382	bool need_mntput = false;
1383	int ret = 0;
1384
1385	while (flags & DCACHE_MANAGED_DENTRY) {
1386		/* Allow the filesystem to manage the transit without i_mutex
1387		 * being held. */
1388		if (flags & DCACHE_MANAGE_TRANSIT) {
1389			ret = path->dentry->d_op->d_manage(path, false);
1390			flags = smp_load_acquire(&path->dentry->d_flags);
1391			if (ret < 0)
1392				break;
1393		}
1394
1395		if (flags & DCACHE_MOUNTED) {	// something's mounted on it..
1396			struct vfsmount *mounted = lookup_mnt(path);
1397			if (mounted) {		// ... in our namespace
1398				dput(path->dentry);
1399				if (need_mntput)
1400					mntput(path->mnt);
1401				path->mnt = mounted;
1402				path->dentry = dget(mounted->mnt_root);
1403				// here we know it's positive
1404				flags = path->dentry->d_flags;
1405				need_mntput = true;
1406				continue;
1407			}
1408		}
1409
1410		if (!(flags & DCACHE_NEED_AUTOMOUNT))
1411			break;
1412
1413		// uncovered automount point
1414		ret = follow_automount(path, count, lookup_flags);
1415		flags = smp_load_acquire(&path->dentry->d_flags);
1416		if (ret < 0)
1417			break;
1418	}
1419
1420	if (ret == -EISDIR)
1421		ret = 0;
1422	// possible if you race with several mount --move
1423	if (need_mntput && path->mnt == mnt)
1424		mntput(path->mnt);
1425	if (!ret && unlikely(d_flags_negative(flags)))
1426		ret = -ENOENT;
1427	*jumped = need_mntput;
1428	return ret;
1429}
1430
1431static inline int traverse_mounts(struct path *path, bool *jumped,
1432				  int *count, unsigned lookup_flags)
1433{
1434	unsigned flags = smp_load_acquire(&path->dentry->d_flags);
1435
1436	/* fastpath */
1437	if (likely(!(flags & DCACHE_MANAGED_DENTRY))) {
1438		*jumped = false;
1439		if (unlikely(d_flags_negative(flags)))
1440			return -ENOENT;
1441		return 0;
1442	}
1443	return __traverse_mounts(path, flags, jumped, count, lookup_flags);
1444}
1445
1446int follow_down_one(struct path *path)
1447{
1448	struct vfsmount *mounted;
1449
1450	mounted = lookup_mnt(path);
1451	if (mounted) {
1452		dput(path->dentry);
1453		mntput(path->mnt);
1454		path->mnt = mounted;
1455		path->dentry = dget(mounted->mnt_root);
1456		return 1;
1457	}
1458	return 0;
1459}
1460EXPORT_SYMBOL(follow_down_one);
1461
1462/*
1463 * Follow down to the covering mount currently visible to userspace.  At each
1464 * point, the filesystem owning that dentry may be queried as to whether the
1465 * caller is permitted to proceed or not.
1466 */
1467int follow_down(struct path *path, unsigned int flags)
1468{
1469	struct vfsmount *mnt = path->mnt;
1470	bool jumped;
1471	int ret = traverse_mounts(path, &jumped, NULL, flags);
1472
1473	if (path->mnt != mnt)
1474		mntput(mnt);
1475	return ret;
1476}
1477EXPORT_SYMBOL(follow_down);
1478
1479/*
1480 * Try to skip to top of mountpoint pile in rcuwalk mode.  Fail if
1481 * we meet a managed dentry that would need blocking.
1482 */
1483static bool __follow_mount_rcu(struct nameidata *nd, struct path *path)
1484{
1485	struct dentry *dentry = path->dentry;
1486	unsigned int flags = dentry->d_flags;
1487
1488	if (likely(!(flags & DCACHE_MANAGED_DENTRY)))
1489		return true;
1490
1491	if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1492		return false;
1493
1494	for (;;) {
1495		/*
1496		 * Don't forget we might have a non-mountpoint managed dentry
1497		 * that wants to block transit.
1498		 */
1499		if (unlikely(flags & DCACHE_MANAGE_TRANSIT)) {
1500			int res = dentry->d_op->d_manage(path, true);
1501			if (res)
1502				return res == -EISDIR;
1503			flags = dentry->d_flags;
1504		}
1505
1506		if (flags & DCACHE_MOUNTED) {
1507			struct mount *mounted = __lookup_mnt(path->mnt, dentry);
1508			if (mounted) {
1509				path->mnt = &mounted->mnt;
1510				dentry = path->dentry = mounted->mnt.mnt_root;
1511				nd->state |= ND_JUMPED;
1512				nd->next_seq = read_seqcount_begin(&dentry->d_seq);
1513				flags = dentry->d_flags;
1514				// makes sure that non-RCU pathwalk could reach
1515				// this state.
1516				if (read_seqretry(&mount_lock, nd->m_seq))
1517					return false;
1518				continue;
1519			}
1520			if (read_seqretry(&mount_lock, nd->m_seq))
1521				return false;
1522		}
1523		return !(flags & DCACHE_NEED_AUTOMOUNT);
1524	}
1525}
1526
1527static inline int handle_mounts(struct nameidata *nd, struct dentry *dentry,
1528			  struct path *path)
1529{
1530	bool jumped;
1531	int ret;
1532
1533	path->mnt = nd->path.mnt;
1534	path->dentry = dentry;
1535	if (nd->flags & LOOKUP_RCU) {
1536		unsigned int seq = nd->next_seq;
1537		if (likely(__follow_mount_rcu(nd, path)))
1538			return 0;
1539		// *path and nd->next_seq might've been clobbered
1540		path->mnt = nd->path.mnt;
1541		path->dentry = dentry;
1542		nd->next_seq = seq;
1543		if (!try_to_unlazy_next(nd, dentry))
1544			return -ECHILD;
1545	}
1546	ret = traverse_mounts(path, &jumped, &nd->total_link_count, nd->flags);
1547	if (jumped) {
1548		if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1549			ret = -EXDEV;
1550		else
1551			nd->state |= ND_JUMPED;
1552	}
1553	if (unlikely(ret)) {
1554		dput(path->dentry);
1555		if (path->mnt != nd->path.mnt)
1556			mntput(path->mnt);
1557	}
1558	return ret;
1559}
1560
1561/*
1562 * This looks up the name in dcache and possibly revalidates the found dentry.
1563 * NULL is returned if the dentry does not exist in the cache.
1564 */
1565static struct dentry *lookup_dcache(const struct qstr *name,
1566				    struct dentry *dir,
1567				    unsigned int flags)
1568{
1569	struct dentry *dentry = d_lookup(dir, name);
1570	if (dentry) {
1571		int error = d_revalidate(dentry, flags);
1572		if (unlikely(error <= 0)) {
1573			if (!error)
1574				d_invalidate(dentry);
1575			dput(dentry);
1576			return ERR_PTR(error);
1577		}
1578	}
1579	return dentry;
1580}
1581
1582/*
1583 * Parent directory has inode locked exclusive.  This is one
1584 * and only case when ->lookup() gets called on non in-lookup
1585 * dentries - as the matter of fact, this only gets called
1586 * when directory is guaranteed to have no in-lookup children
1587 * at all.
1588 */
1589struct dentry *lookup_one_qstr_excl(const struct qstr *name,
1590				    struct dentry *base,
1591				    unsigned int flags)
1592{
1593	struct dentry *dentry = lookup_dcache(name, base, flags);
1594	struct dentry *old;
1595	struct inode *dir = base->d_inode;
1596
1597	if (dentry)
1598		return dentry;
1599
1600	/* Don't create child dentry for a dead directory. */
1601	if (unlikely(IS_DEADDIR(dir)))
1602		return ERR_PTR(-ENOENT);
1603
1604	dentry = d_alloc(base, name);
1605	if (unlikely(!dentry))
1606		return ERR_PTR(-ENOMEM);
1607
1608	old = dir->i_op->lookup(dir, dentry, flags);
1609	if (unlikely(old)) {
1610		dput(dentry);
1611		dentry = old;
1612	}
1613	return dentry;
1614}
1615EXPORT_SYMBOL(lookup_one_qstr_excl);
1616
1617static struct dentry *lookup_fast(struct nameidata *nd)
1618{
1619	struct dentry *dentry, *parent = nd->path.dentry;
1620	int status = 1;
1621
1622	/*
1623	 * Rename seqlock is not required here because in the off chance
1624	 * of a false negative due to a concurrent rename, the caller is
1625	 * going to fall back to non-racy lookup.
1626	 */
1627	if (nd->flags & LOOKUP_RCU) {
1628		dentry = __d_lookup_rcu(parent, &nd->last, &nd->next_seq);
1629		if (unlikely(!dentry)) {
1630			if (!try_to_unlazy(nd))
1631				return ERR_PTR(-ECHILD);
1632			return NULL;
1633		}
1634
1635		/*
1636		 * This sequence count validates that the parent had no
1637		 * changes while we did the lookup of the dentry above.
1638		 */
1639		if (read_seqcount_retry(&parent->d_seq, nd->seq))
1640			return ERR_PTR(-ECHILD);
1641
1642		status = d_revalidate(dentry, nd->flags);
1643		if (likely(status > 0))
1644			return dentry;
1645		if (!try_to_unlazy_next(nd, dentry))
1646			return ERR_PTR(-ECHILD);
1647		if (status == -ECHILD)
1648			/* we'd been told to redo it in non-rcu mode */
1649			status = d_revalidate(dentry, nd->flags);
1650	} else {
1651		dentry = __d_lookup(parent, &nd->last);
1652		if (unlikely(!dentry))
1653			return NULL;
1654		status = d_revalidate(dentry, nd->flags);
1655	}
1656	if (unlikely(status <= 0)) {
1657		if (!status)
1658			d_invalidate(dentry);
1659		dput(dentry);
1660		return ERR_PTR(status);
1661	}
1662	return dentry;
1663}
1664
1665/* Fast lookup failed, do it the slow way */
1666static struct dentry *__lookup_slow(const struct qstr *name,
1667				    struct dentry *dir,
1668				    unsigned int flags)
1669{
1670	struct dentry *dentry, *old;
1671	struct inode *inode = dir->d_inode;
1672	DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1673
1674	/* Don't go there if it's already dead */
1675	if (unlikely(IS_DEADDIR(inode)))
1676		return ERR_PTR(-ENOENT);
1677again:
1678	dentry = d_alloc_parallel(dir, name, &wq);
1679	if (IS_ERR(dentry))
1680		return dentry;
1681	if (unlikely(!d_in_lookup(dentry))) {
1682		int error = d_revalidate(dentry, flags);
1683		if (unlikely(error <= 0)) {
1684			if (!error) {
1685				d_invalidate(dentry);
1686				dput(dentry);
1687				goto again;
1688			}
1689			dput(dentry);
1690			dentry = ERR_PTR(error);
1691		}
1692	} else {
1693		old = inode->i_op->lookup(inode, dentry, flags);
1694		d_lookup_done(dentry);
1695		if (unlikely(old)) {
1696			dput(dentry);
1697			dentry = old;
1698		}
1699	}
1700	return dentry;
1701}
1702
1703static struct dentry *lookup_slow(const struct qstr *name,
1704				  struct dentry *dir,
1705				  unsigned int flags)
1706{
1707	struct inode *inode = dir->d_inode;
1708	struct dentry *res;
1709	inode_lock_shared(inode);
1710	res = __lookup_slow(name, dir, flags);
1711	inode_unlock_shared(inode);
1712	return res;
1713}
1714
1715static inline int may_lookup(struct mnt_idmap *idmap,
1716			     struct nameidata *nd)
1717{
1718	if (nd->flags & LOOKUP_RCU) {
1719		int err = inode_permission(idmap, nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1720		if (!err)		// success, keep going
1721			return 0;
1722		if (!try_to_unlazy(nd))
1723			return -ECHILD;	// redo it all non-lazy
1724		if (err != -ECHILD)	// hard error
1725			return err;
1726	}
1727	return inode_permission(idmap, nd->inode, MAY_EXEC);
1728}
1729
1730static int reserve_stack(struct nameidata *nd, struct path *link)
1731{
1732	if (unlikely(nd->total_link_count++ >= MAXSYMLINKS))
1733		return -ELOOP;
1734
1735	if (likely(nd->depth != EMBEDDED_LEVELS))
1736		return 0;
1737	if (likely(nd->stack != nd->internal))
1738		return 0;
1739	if (likely(nd_alloc_stack(nd)))
1740		return 0;
1741
1742	if (nd->flags & LOOKUP_RCU) {
1743		// we need to grab link before we do unlazy.  And we can't skip
1744		// unlazy even if we fail to grab the link - cleanup needs it
1745		bool grabbed_link = legitimize_path(nd, link, nd->next_seq);
1746
1747		if (!try_to_unlazy(nd) || !grabbed_link)
1748			return -ECHILD;
1749
1750		if (nd_alloc_stack(nd))
1751			return 0;
1752	}
1753	return -ENOMEM;
1754}
1755
1756enum {WALK_TRAILING = 1, WALK_MORE = 2, WALK_NOFOLLOW = 4};
1757
1758static const char *pick_link(struct nameidata *nd, struct path *link,
1759		     struct inode *inode, int flags)
1760{
1761	struct saved *last;
1762	const char *res;
1763	int error = reserve_stack(nd, link);
1764
1765	if (unlikely(error)) {
1766		if (!(nd->flags & LOOKUP_RCU))
1767			path_put(link);
1768		return ERR_PTR(error);
1769	}
1770	last = nd->stack + nd->depth++;
1771	last->link = *link;
1772	clear_delayed_call(&last->done);
1773	last->seq = nd->next_seq;
1774
1775	if (flags & WALK_TRAILING) {
1776		error = may_follow_link(nd, inode);
1777		if (unlikely(error))
1778			return ERR_PTR(error);
1779	}
1780
1781	if (unlikely(nd->flags & LOOKUP_NO_SYMLINKS) ||
1782			unlikely(link->mnt->mnt_flags & MNT_NOSYMFOLLOW))
1783		return ERR_PTR(-ELOOP);
1784
1785	if (!(nd->flags & LOOKUP_RCU)) {
1786		touch_atime(&last->link);
1787		cond_resched();
1788	} else if (atime_needs_update(&last->link, inode)) {
1789		if (!try_to_unlazy(nd))
1790			return ERR_PTR(-ECHILD);
1791		touch_atime(&last->link);
1792	}
1793
1794	error = security_inode_follow_link(link->dentry, inode,
1795					   nd->flags & LOOKUP_RCU);
1796	if (unlikely(error))
1797		return ERR_PTR(error);
1798
1799	res = READ_ONCE(inode->i_link);
1800	if (!res) {
1801		const char * (*get)(struct dentry *, struct inode *,
1802				struct delayed_call *);
1803		get = inode->i_op->get_link;
1804		if (nd->flags & LOOKUP_RCU) {
1805			res = get(NULL, inode, &last->done);
1806			if (res == ERR_PTR(-ECHILD) && try_to_unlazy(nd))
1807				res = get(link->dentry, inode, &last->done);
1808		} else {
1809			res = get(link->dentry, inode, &last->done);
1810		}
1811		if (!res)
1812			goto all_done;
1813		if (IS_ERR(res))
1814			return res;
1815	}
1816	if (*res == '/') {
1817		error = nd_jump_root(nd);
1818		if (unlikely(error))
1819			return ERR_PTR(error);
1820		while (unlikely(*++res == '/'))
1821			;
1822	}
1823	if (*res)
1824		return res;
1825all_done: // pure jump
1826	put_link(nd);
1827	return NULL;
1828}
1829
1830/*
1831 * Do we need to follow links? We _really_ want to be able
1832 * to do this check without having to look at inode->i_op,
1833 * so we keep a cache of "no, this doesn't need follow_link"
1834 * for the common case.
1835 *
1836 * NOTE: dentry must be what nd->next_seq had been sampled from.
1837 */
1838static const char *step_into(struct nameidata *nd, int flags,
1839		     struct dentry *dentry)
1840{
1841	struct path path;
1842	struct inode *inode;
1843	int err = handle_mounts(nd, dentry, &path);
1844
1845	if (err < 0)
1846		return ERR_PTR(err);
1847	inode = path.dentry->d_inode;
1848	if (likely(!d_is_symlink(path.dentry)) ||
1849	   ((flags & WALK_TRAILING) && !(nd->flags & LOOKUP_FOLLOW)) ||
1850	   (flags & WALK_NOFOLLOW)) {
1851		/* not a symlink or should not follow */
1852		if (nd->flags & LOOKUP_RCU) {
1853			if (read_seqcount_retry(&path.dentry->d_seq, nd->next_seq))
1854				return ERR_PTR(-ECHILD);
1855			if (unlikely(!inode))
1856				return ERR_PTR(-ENOENT);
1857		} else {
1858			dput(nd->path.dentry);
1859			if (nd->path.mnt != path.mnt)
1860				mntput(nd->path.mnt);
1861		}
1862		nd->path = path;
1863		nd->inode = inode;
1864		nd->seq = nd->next_seq;
1865		return NULL;
1866	}
1867	if (nd->flags & LOOKUP_RCU) {
1868		/* make sure that d_is_symlink above matches inode */
1869		if (read_seqcount_retry(&path.dentry->d_seq, nd->next_seq))
1870			return ERR_PTR(-ECHILD);
1871	} else {
1872		if (path.mnt == nd->path.mnt)
1873			mntget(path.mnt);
1874	}
1875	return pick_link(nd, &path, inode, flags);
1876}
1877
1878static struct dentry *follow_dotdot_rcu(struct nameidata *nd)
1879{
1880	struct dentry *parent, *old;
1881
1882	if (path_equal(&nd->path, &nd->root))
1883		goto in_root;
1884	if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) {
1885		struct path path;
1886		unsigned seq;
1887		if (!choose_mountpoint_rcu(real_mount(nd->path.mnt),
1888					   &nd->root, &path, &seq))
1889			goto in_root;
1890		if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1891			return ERR_PTR(-ECHILD);
1892		nd->path = path;
1893		nd->inode = path.dentry->d_inode;
1894		nd->seq = seq;
1895		// makes sure that non-RCU pathwalk could reach this state
1896		if (read_seqretry(&mount_lock, nd->m_seq))
1897			return ERR_PTR(-ECHILD);
1898		/* we know that mountpoint was pinned */
1899	}
1900	old = nd->path.dentry;
1901	parent = old->d_parent;
1902	nd->next_seq = read_seqcount_begin(&parent->d_seq);
1903	// makes sure that non-RCU pathwalk could reach this state
1904	if (read_seqcount_retry(&old->d_seq, nd->seq))
1905		return ERR_PTR(-ECHILD);
1906	if (unlikely(!path_connected(nd->path.mnt, parent)))
1907		return ERR_PTR(-ECHILD);
1908	return parent;
1909in_root:
1910	if (read_seqretry(&mount_lock, nd->m_seq))
1911		return ERR_PTR(-ECHILD);
1912	if (unlikely(nd->flags & LOOKUP_BENEATH))
1913		return ERR_PTR(-ECHILD);
1914	nd->next_seq = nd->seq;
1915	return nd->path.dentry;
1916}
1917
1918static struct dentry *follow_dotdot(struct nameidata *nd)
1919{
1920	struct dentry *parent;
1921
1922	if (path_equal(&nd->path, &nd->root))
1923		goto in_root;
1924	if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) {
1925		struct path path;
1926
1927		if (!choose_mountpoint(real_mount(nd->path.mnt),
1928				       &nd->root, &path))
1929			goto in_root;
1930		path_put(&nd->path);
1931		nd->path = path;
1932		nd->inode = path.dentry->d_inode;
1933		if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1934			return ERR_PTR(-EXDEV);
1935	}
1936	/* rare case of legitimate dget_parent()... */
1937	parent = dget_parent(nd->path.dentry);
1938	if (unlikely(!path_connected(nd->path.mnt, parent))) {
1939		dput(parent);
1940		return ERR_PTR(-ENOENT);
1941	}
1942	return parent;
1943
1944in_root:
1945	if (unlikely(nd->flags & LOOKUP_BENEATH))
1946		return ERR_PTR(-EXDEV);
1947	return dget(nd->path.dentry);
1948}
1949
1950static const char *handle_dots(struct nameidata *nd, int type)
1951{
1952	if (type == LAST_DOTDOT) {
1953		const char *error = NULL;
1954		struct dentry *parent;
1955
1956		if (!nd->root.mnt) {
1957			error = ERR_PTR(set_root(nd));
1958			if (error)
1959				return error;
1960		}
1961		if (nd->flags & LOOKUP_RCU)
1962			parent = follow_dotdot_rcu(nd);
1963		else
1964			parent = follow_dotdot(nd);
1965		if (IS_ERR(parent))
1966			return ERR_CAST(parent);
1967		error = step_into(nd, WALK_NOFOLLOW, parent);
1968		if (unlikely(error))
1969			return error;
1970
1971		if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) {
1972			/*
1973			 * If there was a racing rename or mount along our
1974			 * path, then we can't be sure that ".." hasn't jumped
1975			 * above nd->root (and so userspace should retry or use
1976			 * some fallback).
1977			 */
1978			smp_rmb();
1979			if (__read_seqcount_retry(&mount_lock.seqcount, nd->m_seq))
1980				return ERR_PTR(-EAGAIN);
1981			if (__read_seqcount_retry(&rename_lock.seqcount, nd->r_seq))
1982				return ERR_PTR(-EAGAIN);
1983		}
1984	}
1985	return NULL;
1986}
1987
1988static const char *walk_component(struct nameidata *nd, int flags)
1989{
1990	struct dentry *dentry;
1991	/*
1992	 * "." and ".." are special - ".." especially so because it has
1993	 * to be able to know about the current root directory and
1994	 * parent relationships.
1995	 */
1996	if (unlikely(nd->last_type != LAST_NORM)) {
1997		if (!(flags & WALK_MORE) && nd->depth)
1998			put_link(nd);
1999		return handle_dots(nd, nd->last_type);
2000	}
2001	dentry = lookup_fast(nd);
2002	if (IS_ERR(dentry))
2003		return ERR_CAST(dentry);
2004	if (unlikely(!dentry)) {
2005		dentry = lookup_slow(&nd->last, nd->path.dentry, nd->flags);
2006		if (IS_ERR(dentry))
2007			return ERR_CAST(dentry);
2008	}
2009	if (!(flags & WALK_MORE) && nd->depth)
2010		put_link(nd);
2011	return step_into(nd, flags, dentry);
2012}
2013
2014/*
2015 * We can do the critical dentry name comparison and hashing
2016 * operations one word at a time, but we are limited to:
2017 *
2018 * - Architectures with fast unaligned word accesses. We could
2019 *   do a "get_unaligned()" if this helps and is sufficiently
2020 *   fast.
2021 *
2022 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
2023 *   do not trap on the (extremely unlikely) case of a page
2024 *   crossing operation.
2025 *
2026 * - Furthermore, we need an efficient 64-bit compile for the
2027 *   64-bit case in order to generate the "number of bytes in
2028 *   the final mask". Again, that could be replaced with a
2029 *   efficient population count instruction or similar.
2030 */
2031#ifdef CONFIG_DCACHE_WORD_ACCESS
2032
2033#include <asm/word-at-a-time.h>
2034
2035#ifdef HASH_MIX
2036
2037/* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
2038
2039#elif defined(CONFIG_64BIT)
2040/*
2041 * Register pressure in the mixing function is an issue, particularly
2042 * on 32-bit x86, but almost any function requires one state value and
2043 * one temporary.  Instead, use a function designed for two state values
2044 * and no temporaries.
2045 *
2046 * This function cannot create a collision in only two iterations, so
2047 * we have two iterations to achieve avalanche.  In those two iterations,
2048 * we have six layers of mixing, which is enough to spread one bit's
2049 * influence out to 2^6 = 64 state bits.
2050 *
2051 * Rotate constants are scored by considering either 64 one-bit input
2052 * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
2053 * probability of that delta causing a change to each of the 128 output
2054 * bits, using a sample of random initial states.
2055 *
2056 * The Shannon entropy of the computed probabilities is then summed
2057 * to produce a score.  Ideally, any input change has a 50% chance of
2058 * toggling any given output bit.
2059 *
2060 * Mixing scores (in bits) for (12,45):
2061 * Input delta: 1-bit      2-bit
2062 * 1 round:     713.3    42542.6
2063 * 2 rounds:   2753.7   140389.8
2064 * 3 rounds:   5954.1   233458.2
2065 * 4 rounds:   7862.6   256672.2
2066 * Perfect:    8192     258048
2067 *            (64*128) (64*63/2 * 128)
2068 */
2069#define HASH_MIX(x, y, a)	\
2070	(	x ^= (a),	\
2071	y ^= x,	x = rol64(x,12),\
2072	x += y,	y = rol64(y,45),\
2073	y *= 9			)
2074
2075/*
2076 * Fold two longs into one 32-bit hash value.  This must be fast, but
2077 * latency isn't quite as critical, as there is a fair bit of additional
2078 * work done before the hash value is used.
2079 */
2080static inline unsigned int fold_hash(unsigned long x, unsigned long y)
2081{
2082	y ^= x * GOLDEN_RATIO_64;
2083	y *= GOLDEN_RATIO_64;
2084	return y >> 32;
2085}
2086
2087#else	/* 32-bit case */
2088
2089/*
2090 * Mixing scores (in bits) for (7,20):
2091 * Input delta: 1-bit      2-bit
2092 * 1 round:     330.3     9201.6
2093 * 2 rounds:   1246.4    25475.4
2094 * 3 rounds:   1907.1    31295.1
2095 * 4 rounds:   2042.3    31718.6
2096 * Perfect:    2048      31744
2097 *            (32*64)   (32*31/2 * 64)
2098 */
2099#define HASH_MIX(x, y, a)	\
2100	(	x ^= (a),	\
2101	y ^= x,	x = rol32(x, 7),\
2102	x += y,	y = rol32(y,20),\
2103	y *= 9			)
2104
2105static inline unsigned int fold_hash(unsigned long x, unsigned long y)
2106{
2107	/* Use arch-optimized multiply if one exists */
2108	return __hash_32(y ^ __hash_32(x));
2109}
2110
2111#endif
2112
2113/*
2114 * Return the hash of a string of known length.  This is carfully
2115 * designed to match hash_name(), which is the more critical function.
2116 * In particular, we must end by hashing a final word containing 0..7
2117 * payload bytes, to match the way that hash_name() iterates until it
2118 * finds the delimiter after the name.
2119 */
2120unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2121{
2122	unsigned long a, x = 0, y = (unsigned long)salt;
2123
2124	for (;;) {
2125		if (!len)
2126			goto done;
2127		a = load_unaligned_zeropad(name);
2128		if (len < sizeof(unsigned long))
2129			break;
2130		HASH_MIX(x, y, a);
2131		name += sizeof(unsigned long);
2132		len -= sizeof(unsigned long);
2133	}
2134	x ^= a & bytemask_from_count(len);
2135done:
2136	return fold_hash(x, y);
2137}
2138EXPORT_SYMBOL(full_name_hash);
2139
2140/* Return the "hash_len" (hash and length) of a null-terminated string */
2141u64 hashlen_string(const void *salt, const char *name)
2142{
2143	unsigned long a = 0, x = 0, y = (unsigned long)salt;
2144	unsigned long adata, mask, len;
2145	const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2146
2147	len = 0;
2148	goto inside;
2149
2150	do {
2151		HASH_MIX(x, y, a);
2152		len += sizeof(unsigned long);
2153inside:
2154		a = load_unaligned_zeropad(name+len);
2155	} while (!has_zero(a, &adata, &constants));
2156
2157	adata = prep_zero_mask(a, adata, &constants);
2158	mask = create_zero_mask(adata);
2159	x ^= a & zero_bytemask(mask);
2160
2161	return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2162}
2163EXPORT_SYMBOL(hashlen_string);
2164
2165/*
2166 * Calculate the length and hash of the path component, and
2167 * return the "hash_len" as the result.
2168 */
2169static inline u64 hash_name(const void *salt, const char *name)
2170{
2171	unsigned long a = 0, b, x = 0, y = (unsigned long)salt;
2172	unsigned long adata, bdata, mask, len;
2173	const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2174
2175	len = 0;
2176	goto inside;
2177
2178	do {
2179		HASH_MIX(x, y, a);
2180		len += sizeof(unsigned long);
2181inside:
2182		a = load_unaligned_zeropad(name+len);
2183		b = a ^ REPEAT_BYTE('/');
2184	} while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
2185
2186	adata = prep_zero_mask(a, adata, &constants);
2187	bdata = prep_zero_mask(b, bdata, &constants);
2188	mask = create_zero_mask(adata | bdata);
2189	x ^= a & zero_bytemask(mask);
2190
2191	return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2192}
2193
2194#else	/* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
2195
2196/* Return the hash of a string of known length */
2197unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2198{
2199	unsigned long hash = init_name_hash(salt);
2200	while (len--)
2201		hash = partial_name_hash((unsigned char)*name++, hash);
2202	return end_name_hash(hash);
2203}
2204EXPORT_SYMBOL(full_name_hash);
2205
2206/* Return the "hash_len" (hash and length) of a null-terminated string */
2207u64 hashlen_string(const void *salt, const char *name)
2208{
2209	unsigned long hash = init_name_hash(salt);
2210	unsigned long len = 0, c;
2211
2212	c = (unsigned char)*name;
2213	while (c) {
2214		len++;
2215		hash = partial_name_hash(c, hash);
2216		c = (unsigned char)name[len];
2217	}
2218	return hashlen_create(end_name_hash(hash), len);
2219}
2220EXPORT_SYMBOL(hashlen_string);
2221
2222/*
2223 * We know there's a real path component here of at least
2224 * one character.
2225 */
2226static inline u64 hash_name(const void *salt, const char *name)
2227{
2228	unsigned long hash = init_name_hash(salt);
2229	unsigned long len = 0, c;
2230
2231	c = (unsigned char)*name;
2232	do {
2233		len++;
2234		hash = partial_name_hash(c, hash);
2235		c = (unsigned char)name[len];
2236	} while (c && c != '/');
2237	return hashlen_create(end_name_hash(hash), len);
2238}
2239
2240#endif
2241
2242/*
2243 * Name resolution.
2244 * This is the basic name resolution function, turning a pathname into
2245 * the final dentry. We expect 'base' to be positive and a directory.
2246 *
2247 * Returns 0 and nd will have valid dentry and mnt on success.
2248 * Returns error and drops reference to input namei data on failure.
2249 */
2250static int link_path_walk(const char *name, struct nameidata *nd)
2251{
2252	int depth = 0; // depth <= nd->depth
2253	int err;
2254
2255	nd->last_type = LAST_ROOT;
2256	nd->flags |= LOOKUP_PARENT;
2257	if (IS_ERR(name))
2258		return PTR_ERR(name);
2259	while (*name=='/')
2260		name++;
2261	if (!*name) {
2262		nd->dir_mode = 0; // short-circuit the 'hardening' idiocy
2263		return 0;
2264	}
2265
2266	/* At this point we know we have a real path component. */
2267	for(;;) {
2268		struct mnt_idmap *idmap;
2269		const char *link;
2270		u64 hash_len;
2271		int type;
2272
2273		idmap = mnt_idmap(nd->path.mnt);
2274		err = may_lookup(idmap, nd);
2275		if (err)
2276			return err;
2277
2278		hash_len = hash_name(nd->path.dentry, name);
2279
2280		type = LAST_NORM;
2281		if (name[0] == '.') switch (hashlen_len(hash_len)) {
2282			case 2:
2283				if (name[1] == '.') {
2284					type = LAST_DOTDOT;
2285					nd->state |= ND_JUMPED;
2286				}
2287				break;
2288			case 1:
2289				type = LAST_DOT;
2290		}
2291		if (likely(type == LAST_NORM)) {
2292			struct dentry *parent = nd->path.dentry;
2293			nd->state &= ~ND_JUMPED;
2294			if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
2295				struct qstr this = { { .hash_len = hash_len }, .name = name };
2296				err = parent->d_op->d_hash(parent, &this);
2297				if (err < 0)
2298					return err;
2299				hash_len = this.hash_len;
2300				name = this.name;
2301			}
2302		}
2303
2304		nd->last.hash_len = hash_len;
2305		nd->last.name = name;
2306		nd->last_type = type;
2307
2308		name += hashlen_len(hash_len);
2309		if (!*name)
2310			goto OK;
2311		/*
2312		 * If it wasn't NUL, we know it was '/'. Skip that
2313		 * slash, and continue until no more slashes.
2314		 */
2315		do {
2316			name++;
2317		} while (unlikely(*name == '/'));
2318		if (unlikely(!*name)) {
2319OK:
2320			/* pathname or trailing symlink, done */
2321			if (!depth) {
2322				nd->dir_vfsuid = i_uid_into_vfsuid(idmap, nd->inode);
2323				nd->dir_mode = nd->inode->i_mode;
2324				nd->flags &= ~LOOKUP_PARENT;
2325				return 0;
2326			}
2327			/* last component of nested symlink */
2328			name = nd->stack[--depth].name;
2329			link = walk_component(nd, 0);
2330		} else {
2331			/* not the last component */
2332			link = walk_component(nd, WALK_MORE);
2333		}
2334		if (unlikely(link)) {
2335			if (IS_ERR(link))
2336				return PTR_ERR(link);
2337			/* a symlink to follow */
2338			nd->stack[depth++].name = name;
2339			name = link;
2340			continue;
2341		}
2342		if (unlikely(!d_can_lookup(nd->path.dentry))) {
2343			if (nd->flags & LOOKUP_RCU) {
2344				if (!try_to_unlazy(nd))
2345					return -ECHILD;
2346			}
2347			return -ENOTDIR;
2348		}
2349	}
2350}
2351
2352/* must be paired with terminate_walk() */
2353static const char *path_init(struct nameidata *nd, unsigned flags)
2354{
2355	int error;
2356	const char *s = nd->name->name;
2357
2358	/* LOOKUP_CACHED requires RCU, ask caller to retry */
2359	if ((flags & (LOOKUP_RCU | LOOKUP_CACHED)) == LOOKUP_CACHED)
2360		return ERR_PTR(-EAGAIN);
2361
2362	if (!*s)
2363		flags &= ~LOOKUP_RCU;
2364	if (flags & LOOKUP_RCU)
2365		rcu_read_lock();
2366	else
2367		nd->seq = nd->next_seq = 0;
2368
2369	nd->flags = flags;
2370	nd->state |= ND_JUMPED;
2371
2372	nd->m_seq = __read_seqcount_begin(&mount_lock.seqcount);
2373	nd->r_seq = __read_seqcount_begin(&rename_lock.seqcount);
2374	smp_rmb();
2375
2376	if (nd->state & ND_ROOT_PRESET) {
2377		struct dentry *root = nd->root.dentry;
2378		struct inode *inode = root->d_inode;
2379		if (*s && unlikely(!d_can_lookup(root)))
2380			return ERR_PTR(-ENOTDIR);
2381		nd->path = nd->root;
2382		nd->inode = inode;
2383		if (flags & LOOKUP_RCU) {
2384			nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2385			nd->root_seq = nd->seq;
2386		} else {
2387			path_get(&nd->path);
2388		}
2389		return s;
2390	}
2391
2392	nd->root.mnt = NULL;
2393
2394	/* Absolute pathname -- fetch the root (LOOKUP_IN_ROOT uses nd->dfd). */
2395	if (*s == '/' && !(flags & LOOKUP_IN_ROOT)) {
2396		error = nd_jump_root(nd);
2397		if (unlikely(error))
2398			return ERR_PTR(error);
2399		return s;
2400	}
2401
2402	/* Relative pathname -- get the starting-point it is relative to. */
2403	if (nd->dfd == AT_FDCWD) {
2404		if (flags & LOOKUP_RCU) {
2405			struct fs_struct *fs = current->fs;
2406			unsigned seq;
2407
2408			do {
2409				seq = read_seqcount_begin(&fs->seq);
2410				nd->path = fs->pwd;
2411				nd->inode = nd->path.dentry->d_inode;
2412				nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2413			} while (read_seqcount_retry(&fs->seq, seq));
2414		} else {
2415			get_fs_pwd(current->fs, &nd->path);
2416			nd->inode = nd->path.dentry->d_inode;
2417		}
2418	} else {
2419		/* Caller must check execute permissions on the starting path component */
2420		struct fd f = fdget_raw(nd->dfd);
2421		struct dentry *dentry;
2422
2423		if (!f.file)
2424			return ERR_PTR(-EBADF);
2425
2426		dentry = f.file->f_path.dentry;
2427
2428		if (*s && unlikely(!d_can_lookup(dentry))) {
2429			fdput(f);
2430			return ERR_PTR(-ENOTDIR);
2431		}
2432
2433		nd->path = f.file->f_path;
2434		if (flags & LOOKUP_RCU) {
2435			nd->inode = nd->path.dentry->d_inode;
2436			nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2437		} else {
2438			path_get(&nd->path);
2439			nd->inode = nd->path.dentry->d_inode;
2440		}
2441		fdput(f);
2442	}
2443
2444	/* For scoped-lookups we need to set the root to the dirfd as well. */
2445	if (flags & LOOKUP_IS_SCOPED) {
2446		nd->root = nd->path;
2447		if (flags & LOOKUP_RCU) {
2448			nd->root_seq = nd->seq;
2449		} else {
2450			path_get(&nd->root);
2451			nd->state |= ND_ROOT_GRABBED;
2452		}
2453	}
2454	return s;
2455}
2456
2457static inline const char *lookup_last(struct nameidata *nd)
2458{
2459	if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2460		nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2461
2462	return walk_component(nd, WALK_TRAILING);
2463}
2464
2465static int handle_lookup_down(struct nameidata *nd)
2466{
2467	if (!(nd->flags & LOOKUP_RCU))
2468		dget(nd->path.dentry);
2469	nd->next_seq = nd->seq;
2470	return PTR_ERR(step_into(nd, WALK_NOFOLLOW, nd->path.dentry));
2471}
2472
2473/* Returns 0 and nd will be valid on success; Returns error, otherwise. */
2474static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2475{
2476	const char *s = path_init(nd, flags);
2477	int err;
2478
2479	if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) {
2480		err = handle_lookup_down(nd);
2481		if (unlikely(err < 0))
2482			s = ERR_PTR(err);
2483	}
2484
2485	while (!(err = link_path_walk(s, nd)) &&
2486	       (s = lookup_last(nd)) != NULL)
2487		;
2488	if (!err && unlikely(nd->flags & LOOKUP_MOUNTPOINT)) {
2489		err = handle_lookup_down(nd);
2490		nd->state &= ~ND_JUMPED; // no d_weak_revalidate(), please...
2491	}
2492	if (!err)
2493		err = complete_walk(nd);
2494
2495	if (!err && nd->flags & LOOKUP_DIRECTORY)
2496		if (!d_can_lookup(nd->path.dentry))
2497			err = -ENOTDIR;
2498	if (!err) {
2499		*path = nd->path;
2500		nd->path.mnt = NULL;
2501		nd->path.dentry = NULL;
2502	}
2503	terminate_walk(nd);
2504	return err;
2505}
2506
2507int filename_lookup(int dfd, struct filename *name, unsigned flags,
2508		    struct path *path, struct path *root)
2509{
2510	int retval;
2511	struct nameidata nd;
2512	if (IS_ERR(name))
2513		return PTR_ERR(name);
2514	set_nameidata(&nd, dfd, name, root);
2515	retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
2516	if (unlikely(retval == -ECHILD))
2517		retval = path_lookupat(&nd, flags, path);
2518	if (unlikely(retval == -ESTALE))
2519		retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
2520
2521	if (likely(!retval))
2522		audit_inode(name, path->dentry,
2523			    flags & LOOKUP_MOUNTPOINT ? AUDIT_INODE_NOEVAL : 0);
2524	restore_nameidata();
2525	return retval;
2526}
2527
2528/* Returns 0 and nd will be valid on success; Returns error, otherwise. */
2529static int path_parentat(struct nameidata *nd, unsigned flags,
2530				struct path *parent)
2531{
2532	const char *s = path_init(nd, flags);
2533	int err = link_path_walk(s, nd);
2534	if (!err)
2535		err = complete_walk(nd);
2536	if (!err) {
2537		*parent = nd->path;
2538		nd->path.mnt = NULL;
2539		nd->path.dentry = NULL;
2540	}
2541	terminate_walk(nd);
2542	return err;
2543}
2544
2545/* Note: this does not consume "name" */
2546static int __filename_parentat(int dfd, struct filename *name,
2547			       unsigned int flags, struct path *parent,
2548			       struct qstr *last, int *type,
2549			       const struct path *root)
2550{
2551	int retval;
2552	struct nameidata nd;
2553
2554	if (IS_ERR(name))
2555		return PTR_ERR(name);
2556	set_nameidata(&nd, dfd, name, root);
2557	retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
2558	if (unlikely(retval == -ECHILD))
2559		retval = path_parentat(&nd, flags, parent);
2560	if (unlikely(retval == -ESTALE))
2561		retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
2562	if (likely(!retval)) {
2563		*last = nd.last;
2564		*type = nd.last_type;
2565		audit_inode(name, parent->dentry, AUDIT_INODE_PARENT);
2566	}
2567	restore_nameidata();
2568	return retval;
2569}
2570
2571static int filename_parentat(int dfd, struct filename *name,
2572			     unsigned int flags, struct path *parent,
2573			     struct qstr *last, int *type)
2574{
2575	return __filename_parentat(dfd, name, flags, parent, last, type, NULL);
2576}
2577
2578/* does lookup, returns the object with parent locked */
2579static struct dentry *__kern_path_locked(int dfd, struct filename *name, struct path *path)
2580{
2581	struct dentry *d;
2582	struct qstr last;
2583	int type, error;
2584
2585	error = filename_parentat(dfd, name, 0, path, &last, &type);
2586	if (error)
2587		return ERR_PTR(error);
2588	if (unlikely(type != LAST_NORM)) {
2589		path_put(path);
2590		return ERR_PTR(-EINVAL);
2591	}
2592	inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
2593	d = lookup_one_qstr_excl(&last, path->dentry, 0);
2594	if (IS_ERR(d)) {
2595		inode_unlock(path->dentry->d_inode);
2596		path_put(path);
2597	}
2598	return d;
2599}
2600
2601struct dentry *kern_path_locked(const char *name, struct path *path)
2602{
2603	struct filename *filename = getname_kernel(name);
2604	struct dentry *res = __kern_path_locked(AT_FDCWD, filename, path);
2605
2606	putname(filename);
2607	return res;
2608}
2609
2610struct dentry *user_path_locked_at(int dfd, const char __user *name, struct path *path)
2611{
2612	struct filename *filename = getname(name);
2613	struct dentry *res = __kern_path_locked(dfd, filename, path);
2614
2615	putname(filename);
2616	return res;
2617}
2618EXPORT_SYMBOL(user_path_locked_at);
2619
2620int kern_path(const char *name, unsigned int flags, struct path *path)
2621{
2622	struct filename *filename = getname_kernel(name);
2623	int ret = filename_lookup(AT_FDCWD, filename, flags, path, NULL);
2624
2625	putname(filename);
2626	return ret;
2627
2628}
2629EXPORT_SYMBOL(kern_path);
2630
2631/**
2632 * vfs_path_parent_lookup - lookup a parent path relative to a dentry-vfsmount pair
2633 * @filename: filename structure
2634 * @flags: lookup flags
2635 * @parent: pointer to struct path to fill
2636 * @last: last component
2637 * @type: type of the last component
2638 * @root: pointer to struct path of the base directory
2639 */
2640int vfs_path_parent_lookup(struct filename *filename, unsigned int flags,
2641			   struct path *parent, struct qstr *last, int *type,
2642			   const struct path *root)
2643{
2644	return  __filename_parentat(AT_FDCWD, filename, flags, parent, last,
2645				    type, root);
2646}
2647EXPORT_SYMBOL(vfs_path_parent_lookup);
2648
2649/**
2650 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2651 * @dentry:  pointer to dentry of the base directory
2652 * @mnt: pointer to vfs mount of the base directory
2653 * @name: pointer to file name
2654 * @flags: lookup flags
2655 * @path: pointer to struct path to fill
2656 */
2657int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2658		    const char *name, unsigned int flags,
2659		    struct path *path)
2660{
2661	struct filename *filename;
2662	struct path root = {.mnt = mnt, .dentry = dentry};
2663	int ret;
2664
2665	filename = getname_kernel(name);
2666	/* the first argument of filename_lookup() is ignored with root */
2667	ret = filename_lookup(AT_FDCWD, filename, flags, path, &root);
2668	putname(filename);
2669	return ret;
2670}
2671EXPORT_SYMBOL(vfs_path_lookup);
2672
2673static int lookup_one_common(struct mnt_idmap *idmap,
2674			     const char *name, struct dentry *base, int len,
2675			     struct qstr *this)
2676{
2677	this->name = name;
2678	this->len = len;
2679	this->hash = full_name_hash(base, name, len);
2680	if (!len)
2681		return -EACCES;
2682
2683	if (unlikely(name[0] == '.')) {
2684		if (len < 2 || (len == 2 && name[1] == '.'))
2685			return -EACCES;
2686	}
2687
2688	while (len--) {
2689		unsigned int c = *(const unsigned char *)name++;
2690		if (c == '/' || c == '\0')
2691			return -EACCES;
2692	}
2693	/*
2694	 * See if the low-level filesystem might want
2695	 * to use its own hash..
2696	 */
2697	if (base->d_flags & DCACHE_OP_HASH) {
2698		int err = base->d_op->d_hash(base, this);
2699		if (err < 0)
2700			return err;
2701	}
2702
2703	return inode_permission(idmap, base->d_inode, MAY_EXEC);
2704}
2705
2706/**
2707 * try_lookup_one_len - filesystem helper to lookup single pathname component
2708 * @name:	pathname component to lookup
2709 * @base:	base directory to lookup from
2710 * @len:	maximum length @len should be interpreted to
2711 *
2712 * Look up a dentry by name in the dcache, returning NULL if it does not
2713 * currently exist.  The function does not try to create a dentry.
2714 *
2715 * Note that this routine is purely a helper for filesystem usage and should
2716 * not be called by generic code.
2717 *
2718 * The caller must hold base->i_mutex.
2719 */
2720struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len)
2721{
2722	struct qstr this;
2723	int err;
2724
2725	WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2726
2727	err = lookup_one_common(&nop_mnt_idmap, name, base, len, &this);
2728	if (err)
2729		return ERR_PTR(err);
2730
2731	return lookup_dcache(&this, base, 0);
2732}
2733EXPORT_SYMBOL(try_lookup_one_len);
2734
2735/**
2736 * lookup_one_len - filesystem helper to lookup single pathname component
2737 * @name:	pathname component to lookup
2738 * @base:	base directory to lookup from
2739 * @len:	maximum length @len should be interpreted to
2740 *
2741 * Note that this routine is purely a helper for filesystem usage and should
2742 * not be called by generic code.
2743 *
2744 * The caller must hold base->i_mutex.
2745 */
2746struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2747{
2748	struct dentry *dentry;
2749	struct qstr this;
2750	int err;
2751
2752	WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2753
2754	err = lookup_one_common(&nop_mnt_idmap, name, base, len, &this);
2755	if (err)
2756		return ERR_PTR(err);
2757
2758	dentry = lookup_dcache(&this, base, 0);
2759	return dentry ? dentry : __lookup_slow(&this, base, 0);
2760}
2761EXPORT_SYMBOL(lookup_one_len);
2762
2763/**
2764 * lookup_one - filesystem helper to lookup single pathname component
2765 * @idmap:	idmap of the mount the lookup is performed from
2766 * @name:	pathname component to lookup
2767 * @base:	base directory to lookup from
2768 * @len:	maximum length @len should be interpreted to
2769 *
2770 * Note that this routine is purely a helper for filesystem usage and should
2771 * not be called by generic code.
2772 *
2773 * The caller must hold base->i_mutex.
2774 */
2775struct dentry *lookup_one(struct mnt_idmap *idmap, const char *name,
2776			  struct dentry *base, int len)
2777{
2778	struct dentry *dentry;
2779	struct qstr this;
2780	int err;
2781
2782	WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2783
2784	err = lookup_one_common(idmap, name, base, len, &this);
2785	if (err)
2786		return ERR_PTR(err);
2787
2788	dentry = lookup_dcache(&this, base, 0);
2789	return dentry ? dentry : __lookup_slow(&this, base, 0);
2790}
2791EXPORT_SYMBOL(lookup_one);
2792
2793/**
2794 * lookup_one_unlocked - filesystem helper to lookup single pathname component
2795 * @idmap:	idmap of the mount the lookup is performed from
2796 * @name:	pathname component to lookup
2797 * @base:	base directory to lookup from
2798 * @len:	maximum length @len should be interpreted to
2799 *
2800 * Note that this routine is purely a helper for filesystem usage and should
2801 * not be called by generic code.
2802 *
2803 * Unlike lookup_one_len, it should be called without the parent
2804 * i_mutex held, and will take the i_mutex itself if necessary.
2805 */
2806struct dentry *lookup_one_unlocked(struct mnt_idmap *idmap,
2807				   const char *name, struct dentry *base,
2808				   int len)
2809{
2810	struct qstr this;
2811	int err;
2812	struct dentry *ret;
2813
2814	err = lookup_one_common(idmap, name, base, len, &this);
2815	if (err)
2816		return ERR_PTR(err);
2817
2818	ret = lookup_dcache(&this, base, 0);
2819	if (!ret)
2820		ret = lookup_slow(&this, base, 0);
2821	return ret;
2822}
2823EXPORT_SYMBOL(lookup_one_unlocked);
2824
2825/**
2826 * lookup_one_positive_unlocked - filesystem helper to lookup single
2827 *				  pathname component
2828 * @idmap:	idmap of the mount the lookup is performed from
2829 * @name:	pathname component to lookup
2830 * @base:	base directory to lookup from
2831 * @len:	maximum length @len should be interpreted to
2832 *
2833 * This helper will yield ERR_PTR(-ENOENT) on negatives. The helper returns
2834 * known positive or ERR_PTR(). This is what most of the users want.
2835 *
2836 * Note that pinned negative with unlocked parent _can_ become positive at any
2837 * time, so callers of lookup_one_unlocked() need to be very careful; pinned
2838 * positives have >d_inode stable, so this one avoids such problems.
2839 *
2840 * Note that this routine is purely a helper for filesystem usage and should
2841 * not be called by generic code.
2842 *
2843 * The helper should be called without i_mutex held.
2844 */
2845struct dentry *lookup_one_positive_unlocked(struct mnt_idmap *idmap,
2846					    const char *name,
2847					    struct dentry *base, int len)
2848{
2849	struct dentry *ret = lookup_one_unlocked(idmap, name, base, len);
2850
2851	if (!IS_ERR(ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) {
2852		dput(ret);
2853		ret = ERR_PTR(-ENOENT);
2854	}
2855	return ret;
2856}
2857EXPORT_SYMBOL(lookup_one_positive_unlocked);
2858
2859/**
2860 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2861 * @name:	pathname component to lookup
2862 * @base:	base directory to lookup from
2863 * @len:	maximum length @len should be interpreted to
2864 *
2865 * Note that this routine is purely a helper for filesystem usage and should
2866 * not be called by generic code.
2867 *
2868 * Unlike lookup_one_len, it should be called without the parent
2869 * i_mutex held, and will take the i_mutex itself if necessary.
2870 */
2871struct dentry *lookup_one_len_unlocked(const char *name,
2872				       struct dentry *base, int len)
2873{
2874	return lookup_one_unlocked(&nop_mnt_idmap, name, base, len);
2875}
2876EXPORT_SYMBOL(lookup_one_len_unlocked);
2877
2878/*
2879 * Like lookup_one_len_unlocked(), except that it yields ERR_PTR(-ENOENT)
2880 * on negatives.  Returns known positive or ERR_PTR(); that's what
2881 * most of the users want.  Note that pinned negative with unlocked parent
2882 * _can_ become positive at any time, so callers of lookup_one_len_unlocked()
2883 * need to be very careful; pinned positives have ->d_inode stable, so
2884 * this one avoids such problems.
2885 */
2886struct dentry *lookup_positive_unlocked(const char *name,
2887				       struct dentry *base, int len)
2888{
2889	return lookup_one_positive_unlocked(&nop_mnt_idmap, name, base, len);
2890}
2891EXPORT_SYMBOL(lookup_positive_unlocked);
2892
2893#ifdef CONFIG_UNIX98_PTYS
2894int path_pts(struct path *path)
2895{
2896	/* Find something mounted on "pts" in the same directory as
2897	 * the input path.
2898	 */
2899	struct dentry *parent = dget_parent(path->dentry);
2900	struct dentry *child;
2901	struct qstr this = QSTR_INIT("pts", 3);
2902
2903	if (unlikely(!path_connected(path->mnt, parent))) {
2904		dput(parent);
2905		return -ENOENT;
2906	}
2907	dput(path->dentry);
2908	path->dentry = parent;
2909	child = d_hash_and_lookup(parent, &this);
2910	if (IS_ERR_OR_NULL(child))
2911		return -ENOENT;
2912
2913	path->dentry = child;
2914	dput(parent);
2915	follow_down(path, 0);
2916	return 0;
2917}
2918#endif
2919
2920int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2921		 struct path *path, int *empty)
2922{
2923	struct filename *filename = getname_flags(name, flags, empty);
2924	int ret = filename_lookup(dfd, filename, flags, path, NULL);
2925
2926	putname(filename);
2927	return ret;
2928}
2929EXPORT_SYMBOL(user_path_at_empty);
2930
2931int __check_sticky(struct mnt_idmap *idmap, struct inode *dir,
2932		   struct inode *inode)
2933{
2934	kuid_t fsuid = current_fsuid();
2935
2936	if (vfsuid_eq_kuid(i_uid_into_vfsuid(idmap, inode), fsuid))
2937		return 0;
2938	if (vfsuid_eq_kuid(i_uid_into_vfsuid(idmap, dir), fsuid))
2939		return 0;
2940	return !capable_wrt_inode_uidgid(idmap, inode, CAP_FOWNER);
2941}
2942EXPORT_SYMBOL(__check_sticky);
2943
2944/*
2945 *	Check whether we can remove a link victim from directory dir, check
2946 *  whether the type of victim is right.
2947 *  1. We can't do it if dir is read-only (done in permission())
2948 *  2. We should have write and exec permissions on dir
2949 *  3. We can't remove anything from append-only dir
2950 *  4. We can't do anything with immutable dir (done in permission())
2951 *  5. If the sticky bit on dir is set we should either
2952 *	a. be owner of dir, or
2953 *	b. be owner of victim, or
2954 *	c. have CAP_FOWNER capability
2955 *  6. If the victim is append-only or immutable we can't do antyhing with
2956 *     links pointing to it.
2957 *  7. If the victim has an unknown uid or gid we can't change the inode.
2958 *  8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2959 *  9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2960 * 10. We can't remove a root or mountpoint.
2961 * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
2962 *     nfs_async_unlink().
2963 */
2964static int may_delete(struct mnt_idmap *idmap, struct inode *dir,
2965		      struct dentry *victim, bool isdir)
2966{
2967	struct inode *inode = d_backing_inode(victim);
2968	int error;
2969
2970	if (d_is_negative(victim))
2971		return -ENOENT;
2972	BUG_ON(!inode);
2973
2974	BUG_ON(victim->d_parent->d_inode != dir);
2975
2976	/* Inode writeback is not safe when the uid or gid are invalid. */
2977	if (!vfsuid_valid(i_uid_into_vfsuid(idmap, inode)) ||
2978	    !vfsgid_valid(i_gid_into_vfsgid(idmap, inode)))
2979		return -EOVERFLOW;
2980
2981	audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2982
2983	error = inode_permission(idmap, dir, MAY_WRITE | MAY_EXEC);
2984	if (error)
2985		return error;
2986	if (IS_APPEND(dir))
2987		return -EPERM;
2988
2989	if (check_sticky(idmap, dir, inode) || IS_APPEND(inode) ||
2990	    IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) ||
2991	    HAS_UNMAPPED_ID(idmap, inode))
2992		return -EPERM;
2993	if (isdir) {
2994		if (!d_is_dir(victim))
2995			return -ENOTDIR;
2996		if (IS_ROOT(victim))
2997			return -EBUSY;
2998	} else if (d_is_dir(victim))
2999		return -EISDIR;
3000	if (IS_DEADDIR(dir))
3001		return -ENOENT;
3002	if (victim->d_flags & DCACHE_NFSFS_RENAMED)
3003		return -EBUSY;
3004	return 0;
3005}
3006
3007/*	Check whether we can create an object with dentry child in directory
3008 *  dir.
3009 *  1. We can't do it if child already exists (open has special treatment for
3010 *     this case, but since we are inlined it's OK)
3011 *  2. We can't do it if dir is read-only (done in permission())
3012 *  3. We can't do it if the fs can't represent the fsuid or fsgid.
3013 *  4. We should have write and exec permissions on dir
3014 *  5. We can't do it if dir is immutable (done in permission())
3015 */
3016static inline int may_create(struct mnt_idmap *idmap,
3017			     struct inode *dir, struct dentry *child)
3018{
3019	audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
3020	if (child->d_inode)
3021		return -EEXIST;
3022	if (IS_DEADDIR(dir))
3023		return -ENOENT;
3024	if (!fsuidgid_has_mapping(dir->i_sb, idmap))
3025		return -EOVERFLOW;
3026
3027	return inode_permission(idmap, dir, MAY_WRITE | MAY_EXEC);
3028}
3029
3030// p1 != p2, both are on the same filesystem, ->s_vfs_rename_mutex is held
3031static struct dentry *lock_two_directories(struct dentry *p1, struct dentry *p2)
3032{
3033	struct dentry *p = p1, *q = p2, *r;
3034
3035	while ((r = p->d_parent) != p2 && r != p)
3036		p = r;
3037	if (r == p2) {
3038		// p is a child of p2 and an ancestor of p1 or p1 itself
3039		inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
3040		inode_lock_nested(p1->d_inode, I_MUTEX_PARENT2);
3041		return p;
3042	}
3043	// p is the root of connected component that contains p1
3044	// p2 does not occur on the path from p to p1
3045	while ((r = q->d_parent) != p1 && r != p && r != q)
3046		q = r;
3047	if (r == p1) {
3048		// q is a child of p1 and an ancestor of p2 or p2 itself
3049		inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
3050		inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
3051		return q;
3052	} else if (likely(r == p)) {
3053		// both p2 and p1 are descendents of p
3054		inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
3055		inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
3056		return NULL;
3057	} else { // no common ancestor at the time we'd been called
3058		mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
3059		return ERR_PTR(-EXDEV);
3060	}
3061}
3062
3063/*
3064 * p1 and p2 should be directories on the same fs.
3065 */
3066struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
3067{
 
 
3068	if (p1 == p2) {
3069		inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
3070		return NULL;
3071	}
3072
3073	mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
3074	return lock_two_directories(p1, p2);
3075}
3076EXPORT_SYMBOL(lock_rename);
3077
3078/*
3079 * c1 and p2 should be on the same fs.
3080 */
3081struct dentry *lock_rename_child(struct dentry *c1, struct dentry *p2)
3082{
3083	if (READ_ONCE(c1->d_parent) == p2) {
3084		/*
3085		 * hopefully won't need to touch ->s_vfs_rename_mutex at all.
3086		 */
3087		inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
3088		/*
3089		 * now that p2 is locked, nobody can move in or out of it,
3090		 * so the test below is safe.
3091		 */
3092		if (likely(c1->d_parent == p2))
3093			return NULL;
3094
3095		/*
3096		 * c1 got moved out of p2 while we'd been taking locks;
3097		 * unlock and fall back to slow case.
3098		 */
3099		inode_unlock(p2->d_inode);
3100	}
3101
3102	mutex_lock(&c1->d_sb->s_vfs_rename_mutex);
3103	/*
3104	 * nobody can move out of any directories on this fs.
3105	 */
3106	if (likely(c1->d_parent != p2))
3107		return lock_two_directories(c1->d_parent, p2);
3108
3109	/*
3110	 * c1 got moved into p2 while we were taking locks;
3111	 * we need p2 locked and ->s_vfs_rename_mutex unlocked,
3112	 * for consistency with lock_rename().
3113	 */
3114	inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
3115	mutex_unlock(&c1->d_sb->s_vfs_rename_mutex);
3116	return NULL;
3117}
3118EXPORT_SYMBOL(lock_rename_child);
3119
3120void unlock_rename(struct dentry *p1, struct dentry *p2)
3121{
3122	inode_unlock(p1->d_inode);
3123	if (p1 != p2) {
3124		inode_unlock(p2->d_inode);
3125		mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
3126	}
3127}
3128EXPORT_SYMBOL(unlock_rename);
3129
3130/**
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3131 * vfs_prepare_mode - prepare the mode to be used for a new inode
3132 * @idmap:	idmap of the mount the inode was found from
3133 * @dir:	parent directory of the new inode
3134 * @mode:	mode of the new inode
3135 * @mask_perms:	allowed permission by the vfs
3136 * @type:	type of file to be created
3137 *
3138 * This helper consolidates and enforces vfs restrictions on the @mode of a new
3139 * object to be created.
3140 *
3141 * Umask stripping depends on whether the filesystem supports POSIX ACLs (see
3142 * the kernel documentation for mode_strip_umask()). Moving umask stripping
3143 * after setgid stripping allows the same ordering for both non-POSIX ACL and
3144 * POSIX ACL supporting filesystems.
3145 *
3146 * Note that it's currently valid for @type to be 0 if a directory is created.
3147 * Filesystems raise that flag individually and we need to check whether each
3148 * filesystem can deal with receiving S_IFDIR from the vfs before we enforce a
3149 * non-zero type.
3150 *
3151 * Returns: mode to be passed to the filesystem
3152 */
3153static inline umode_t vfs_prepare_mode(struct mnt_idmap *idmap,
3154				       const struct inode *dir, umode_t mode,
3155				       umode_t mask_perms, umode_t type)
3156{
3157	mode = mode_strip_sgid(idmap, dir, mode);
3158	mode = mode_strip_umask(dir, mode);
3159
3160	/*
3161	 * Apply the vfs mandated allowed permission mask and set the type of
3162	 * file to be created before we call into the filesystem.
3163	 */
3164	mode &= (mask_perms & ~S_IFMT);
3165	mode |= (type & S_IFMT);
3166
3167	return mode;
3168}
3169
3170/**
3171 * vfs_create - create new file
3172 * @idmap:	idmap of the mount the inode was found from
3173 * @dir:	inode of @dentry
3174 * @dentry:	pointer to dentry of the base directory
3175 * @mode:	mode of the new file
3176 * @want_excl:	whether the file must not yet exist
3177 *
3178 * Create a new file.
3179 *
3180 * If the inode has been found through an idmapped mount the idmap of
3181 * the vfsmount must be passed through @idmap. This function will then take
3182 * care to map the inode according to @idmap before checking permissions.
3183 * On non-idmapped mounts or if permission checking is to be performed on the
3184 * raw inode simply pass @nop_mnt_idmap.
3185 */
3186int vfs_create(struct mnt_idmap *idmap, struct inode *dir,
3187	       struct dentry *dentry, umode_t mode, bool want_excl)
3188{
3189	int error;
3190
3191	error = may_create(idmap, dir, dentry);
3192	if (error)
3193		return error;
3194
3195	if (!dir->i_op->create)
3196		return -EACCES;	/* shouldn't it be ENOSYS? */
3197
3198	mode = vfs_prepare_mode(idmap, dir, mode, S_IALLUGO, S_IFREG);
3199	error = security_inode_create(dir, dentry, mode);
3200	if (error)
3201		return error;
3202	error = dir->i_op->create(idmap, dir, dentry, mode, want_excl);
3203	if (!error)
3204		fsnotify_create(dir, dentry);
3205	return error;
3206}
3207EXPORT_SYMBOL(vfs_create);
3208
3209int vfs_mkobj(struct dentry *dentry, umode_t mode,
3210		int (*f)(struct dentry *, umode_t, void *),
3211		void *arg)
3212{
3213	struct inode *dir = dentry->d_parent->d_inode;
3214	int error = may_create(&nop_mnt_idmap, dir, dentry);
3215	if (error)
3216		return error;
3217
3218	mode &= S_IALLUGO;
3219	mode |= S_IFREG;
3220	error = security_inode_create(dir, dentry, mode);
3221	if (error)
3222		return error;
3223	error = f(dentry, mode, arg);
3224	if (!error)
3225		fsnotify_create(dir, dentry);
3226	return error;
3227}
3228EXPORT_SYMBOL(vfs_mkobj);
3229
3230bool may_open_dev(const struct path *path)
3231{
3232	return !(path->mnt->mnt_flags & MNT_NODEV) &&
3233		!(path->mnt->mnt_sb->s_iflags & SB_I_NODEV);
3234}
3235
3236static int may_open(struct mnt_idmap *idmap, const struct path *path,
3237		    int acc_mode, int flag)
3238{
3239	struct dentry *dentry = path->dentry;
3240	struct inode *inode = dentry->d_inode;
3241	int error;
3242
3243	if (!inode)
3244		return -ENOENT;
3245
3246	switch (inode->i_mode & S_IFMT) {
3247	case S_IFLNK:
3248		return -ELOOP;
3249	case S_IFDIR:
3250		if (acc_mode & MAY_WRITE)
3251			return -EISDIR;
3252		if (acc_mode & MAY_EXEC)
3253			return -EACCES;
3254		break;
3255	case S_IFBLK:
3256	case S_IFCHR:
3257		if (!may_open_dev(path))
3258			return -EACCES;
3259		fallthrough;
3260	case S_IFIFO:
3261	case S_IFSOCK:
3262		if (acc_mode & MAY_EXEC)
3263			return -EACCES;
3264		flag &= ~O_TRUNC;
3265		break;
3266	case S_IFREG:
3267		if ((acc_mode & MAY_EXEC) && path_noexec(path))
3268			return -EACCES;
3269		break;
3270	}
3271
3272	error = inode_permission(idmap, inode, MAY_OPEN | acc_mode);
3273	if (error)
3274		return error;
3275
3276	/*
3277	 * An append-only file must be opened in append mode for writing.
3278	 */
3279	if (IS_APPEND(inode)) {
3280		if  ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
3281			return -EPERM;
3282		if (flag & O_TRUNC)
3283			return -EPERM;
3284	}
3285
3286	/* O_NOATIME can only be set by the owner or superuser */
3287	if (flag & O_NOATIME && !inode_owner_or_capable(idmap, inode))
3288		return -EPERM;
3289
3290	return 0;
3291}
3292
3293static int handle_truncate(struct mnt_idmap *idmap, struct file *filp)
3294{
3295	const struct path *path = &filp->f_path;
3296	struct inode *inode = path->dentry->d_inode;
3297	int error = get_write_access(inode);
3298	if (error)
3299		return error;
3300
3301	error = security_file_truncate(filp);
3302	if (!error) {
3303		error = do_truncate(idmap, path->dentry, 0,
3304				    ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
3305				    filp);
3306	}
3307	put_write_access(inode);
3308	return error;
3309}
3310
3311static inline int open_to_namei_flags(int flag)
3312{
3313	if ((flag & O_ACCMODE) == 3)
3314		flag--;
3315	return flag;
3316}
3317
3318static int may_o_create(struct mnt_idmap *idmap,
3319			const struct path *dir, struct dentry *dentry,
3320			umode_t mode)
3321{
3322	int error = security_path_mknod(dir, dentry, mode, 0);
3323	if (error)
3324		return error;
3325
3326	if (!fsuidgid_has_mapping(dir->dentry->d_sb, idmap))
3327		return -EOVERFLOW;
3328
3329	error = inode_permission(idmap, dir->dentry->d_inode,
3330				 MAY_WRITE | MAY_EXEC);
3331	if (error)
3332		return error;
3333
3334	return security_inode_create(dir->dentry->d_inode, dentry, mode);
3335}
3336
3337/*
3338 * Attempt to atomically look up, create and open a file from a negative
3339 * dentry.
3340 *
3341 * Returns 0 if successful.  The file will have been created and attached to
3342 * @file by the filesystem calling finish_open().
3343 *
3344 * If the file was looked up only or didn't need creating, FMODE_OPENED won't
3345 * be set.  The caller will need to perform the open themselves.  @path will
3346 * have been updated to point to the new dentry.  This may be negative.
3347 *
3348 * Returns an error code otherwise.
3349 */
3350static struct dentry *atomic_open(struct nameidata *nd, struct dentry *dentry,
3351				  struct file *file,
3352				  int open_flag, umode_t mode)
3353{
3354	struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
3355	struct inode *dir =  nd->path.dentry->d_inode;
3356	int error;
3357
3358	if (nd->flags & LOOKUP_DIRECTORY)
3359		open_flag |= O_DIRECTORY;
3360
3361	file->f_path.dentry = DENTRY_NOT_SET;
3362	file->f_path.mnt = nd->path.mnt;
3363	error = dir->i_op->atomic_open(dir, dentry, file,
3364				       open_to_namei_flags(open_flag), mode);
3365	d_lookup_done(dentry);
3366	if (!error) {
3367		if (file->f_mode & FMODE_OPENED) {
3368			if (unlikely(dentry != file->f_path.dentry)) {
3369				dput(dentry);
3370				dentry = dget(file->f_path.dentry);
3371			}
3372		} else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
3373			error = -EIO;
3374		} else {
3375			if (file->f_path.dentry) {
3376				dput(dentry);
3377				dentry = file->f_path.dentry;
3378			}
3379			if (unlikely(d_is_negative(dentry)))
3380				error = -ENOENT;
3381		}
3382	}
3383	if (error) {
3384		dput(dentry);
3385		dentry = ERR_PTR(error);
3386	}
3387	return dentry;
3388}
3389
3390/*
3391 * Look up and maybe create and open the last component.
3392 *
3393 * Must be called with parent locked (exclusive in O_CREAT case).
3394 *
3395 * Returns 0 on success, that is, if
3396 *  the file was successfully atomically created (if necessary) and opened, or
3397 *  the file was not completely opened at this time, though lookups and
3398 *  creations were performed.
3399 * These case are distinguished by presence of FMODE_OPENED on file->f_mode.
3400 * In the latter case dentry returned in @path might be negative if O_CREAT
3401 * hadn't been specified.
3402 *
3403 * An error code is returned on failure.
3404 */
3405static struct dentry *lookup_open(struct nameidata *nd, struct file *file,
3406				  const struct open_flags *op,
3407				  bool got_write)
3408{
3409	struct mnt_idmap *idmap;
3410	struct dentry *dir = nd->path.dentry;
3411	struct inode *dir_inode = dir->d_inode;
3412	int open_flag = op->open_flag;
3413	struct dentry *dentry;
3414	int error, create_error = 0;
3415	umode_t mode = op->mode;
3416	DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
3417
3418	if (unlikely(IS_DEADDIR(dir_inode)))
3419		return ERR_PTR(-ENOENT);
3420
3421	file->f_mode &= ~FMODE_CREATED;
3422	dentry = d_lookup(dir, &nd->last);
3423	for (;;) {
3424		if (!dentry) {
3425			dentry = d_alloc_parallel(dir, &nd->last, &wq);
3426			if (IS_ERR(dentry))
3427				return dentry;
3428		}
3429		if (d_in_lookup(dentry))
3430			break;
3431
3432		error = d_revalidate(dentry, nd->flags);
3433		if (likely(error > 0))
3434			break;
3435		if (error)
3436			goto out_dput;
3437		d_invalidate(dentry);
3438		dput(dentry);
3439		dentry = NULL;
3440	}
3441	if (dentry->d_inode) {
3442		/* Cached positive dentry: will open in f_op->open */
3443		return dentry;
3444	}
3445
3446	/*
3447	 * Checking write permission is tricky, bacuse we don't know if we are
3448	 * going to actually need it: O_CREAT opens should work as long as the
3449	 * file exists.  But checking existence breaks atomicity.  The trick is
3450	 * to check access and if not granted clear O_CREAT from the flags.
3451	 *
3452	 * Another problem is returing the "right" error value (e.g. for an
3453	 * O_EXCL open we want to return EEXIST not EROFS).
3454	 */
3455	if (unlikely(!got_write))
3456		open_flag &= ~O_TRUNC;
3457	idmap = mnt_idmap(nd->path.mnt);
3458	if (open_flag & O_CREAT) {
3459		if (open_flag & O_EXCL)
3460			open_flag &= ~O_TRUNC;
3461		mode = vfs_prepare_mode(idmap, dir->d_inode, mode, mode, mode);
3462		if (likely(got_write))
3463			create_error = may_o_create(idmap, &nd->path,
3464						    dentry, mode);
3465		else
3466			create_error = -EROFS;
3467	}
3468	if (create_error)
3469		open_flag &= ~O_CREAT;
3470	if (dir_inode->i_op->atomic_open) {
3471		dentry = atomic_open(nd, dentry, file, open_flag, mode);
3472		if (unlikely(create_error) && dentry == ERR_PTR(-ENOENT))
3473			dentry = ERR_PTR(create_error);
3474		return dentry;
3475	}
3476
3477	if (d_in_lookup(dentry)) {
3478		struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3479							     nd->flags);
3480		d_lookup_done(dentry);
3481		if (unlikely(res)) {
3482			if (IS_ERR(res)) {
3483				error = PTR_ERR(res);
3484				goto out_dput;
3485			}
3486			dput(dentry);
3487			dentry = res;
3488		}
3489	}
3490
3491	/* Negative dentry, just create the file */
3492	if (!dentry->d_inode && (open_flag & O_CREAT)) {
3493		file->f_mode |= FMODE_CREATED;
3494		audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
3495		if (!dir_inode->i_op->create) {
3496			error = -EACCES;
3497			goto out_dput;
3498		}
3499
3500		error = dir_inode->i_op->create(idmap, dir_inode, dentry,
3501						mode, open_flag & O_EXCL);
3502		if (error)
3503			goto out_dput;
3504	}
3505	if (unlikely(create_error) && !dentry->d_inode) {
3506		error = create_error;
3507		goto out_dput;
3508	}
3509	return dentry;
3510
3511out_dput:
3512	dput(dentry);
3513	return ERR_PTR(error);
3514}
3515
3516static const char *open_last_lookups(struct nameidata *nd,
3517		   struct file *file, const struct open_flags *op)
3518{
3519	struct dentry *dir = nd->path.dentry;
3520	int open_flag = op->open_flag;
3521	bool got_write = false;
3522	struct dentry *dentry;
3523	const char *res;
3524
3525	nd->flags |= op->intent;
3526
3527	if (nd->last_type != LAST_NORM) {
3528		if (nd->depth)
3529			put_link(nd);
3530		return handle_dots(nd, nd->last_type);
3531	}
3532
3533	if (!(open_flag & O_CREAT)) {
3534		if (nd->last.name[nd->last.len])
3535			nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3536		/* we _can_ be in RCU mode here */
3537		dentry = lookup_fast(nd);
3538		if (IS_ERR(dentry))
3539			return ERR_CAST(dentry);
3540		if (likely(dentry))
3541			goto finish_lookup;
3542
3543		if (WARN_ON_ONCE(nd->flags & LOOKUP_RCU))
3544			return ERR_PTR(-ECHILD);
3545	} else {
3546		/* create side of things */
3547		if (nd->flags & LOOKUP_RCU) {
3548			if (!try_to_unlazy(nd))
3549				return ERR_PTR(-ECHILD);
3550		}
3551		audit_inode(nd->name, dir, AUDIT_INODE_PARENT);
3552		/* trailing slashes? */
3553		if (unlikely(nd->last.name[nd->last.len]))
3554			return ERR_PTR(-EISDIR);
3555	}
3556
3557	if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3558		got_write = !mnt_want_write(nd->path.mnt);
3559		/*
3560		 * do _not_ fail yet - we might not need that or fail with
3561		 * a different error; let lookup_open() decide; we'll be
3562		 * dropping this one anyway.
3563		 */
3564	}
3565	if (open_flag & O_CREAT)
3566		inode_lock(dir->d_inode);
3567	else
3568		inode_lock_shared(dir->d_inode);
3569	dentry = lookup_open(nd, file, op, got_write);
3570	if (!IS_ERR(dentry) && (file->f_mode & FMODE_CREATED))
3571		fsnotify_create(dir->d_inode, dentry);
3572	if (open_flag & O_CREAT)
3573		inode_unlock(dir->d_inode);
3574	else
3575		inode_unlock_shared(dir->d_inode);
3576
3577	if (got_write)
3578		mnt_drop_write(nd->path.mnt);
3579
3580	if (IS_ERR(dentry))
3581		return ERR_CAST(dentry);
3582
3583	if (file->f_mode & (FMODE_OPENED | FMODE_CREATED)) {
3584		dput(nd->path.dentry);
3585		nd->path.dentry = dentry;
3586		return NULL;
3587	}
3588
3589finish_lookup:
3590	if (nd->depth)
3591		put_link(nd);
3592	res = step_into(nd, WALK_TRAILING, dentry);
3593	if (unlikely(res))
3594		nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3595	return res;
3596}
3597
3598/*
3599 * Handle the last step of open()
3600 */
3601static int do_open(struct nameidata *nd,
3602		   struct file *file, const struct open_flags *op)
3603{
3604	struct mnt_idmap *idmap;
3605	int open_flag = op->open_flag;
3606	bool do_truncate;
3607	int acc_mode;
3608	int error;
3609
3610	if (!(file->f_mode & (FMODE_OPENED | FMODE_CREATED))) {
3611		error = complete_walk(nd);
3612		if (error)
3613			return error;
3614	}
3615	if (!(file->f_mode & FMODE_CREATED))
3616		audit_inode(nd->name, nd->path.dentry, 0);
3617	idmap = mnt_idmap(nd->path.mnt);
3618	if (open_flag & O_CREAT) {
3619		if ((open_flag & O_EXCL) && !(file->f_mode & FMODE_CREATED))
3620			return -EEXIST;
3621		if (d_is_dir(nd->path.dentry))
3622			return -EISDIR;
3623		error = may_create_in_sticky(idmap, nd,
3624					     d_backing_inode(nd->path.dentry));
3625		if (unlikely(error))
3626			return error;
3627	}
3628	if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3629		return -ENOTDIR;
3630
3631	do_truncate = false;
3632	acc_mode = op->acc_mode;
3633	if (file->f_mode & FMODE_CREATED) {
3634		/* Don't check for write permission, don't truncate */
3635		open_flag &= ~O_TRUNC;
3636		acc_mode = 0;
3637	} else if (d_is_reg(nd->path.dentry) && open_flag & O_TRUNC) {
3638		error = mnt_want_write(nd->path.mnt);
3639		if (error)
3640			return error;
3641		do_truncate = true;
3642	}
3643	error = may_open(idmap, &nd->path, acc_mode, open_flag);
3644	if (!error && !(file->f_mode & FMODE_OPENED))
3645		error = vfs_open(&nd->path, file);
3646	if (!error)
3647		error = ima_file_check(file, op->acc_mode);
3648	if (!error && do_truncate)
3649		error = handle_truncate(idmap, file);
3650	if (unlikely(error > 0)) {
3651		WARN_ON(1);
3652		error = -EINVAL;
3653	}
3654	if (do_truncate)
3655		mnt_drop_write(nd->path.mnt);
3656	return error;
3657}
3658
3659/**
3660 * vfs_tmpfile - create tmpfile
3661 * @idmap:	idmap of the mount the inode was found from
3662 * @parentpath:	pointer to the path of the base directory
3663 * @file:	file descriptor of the new tmpfile
3664 * @mode:	mode of the new tmpfile
 
3665 *
3666 * Create a temporary file.
3667 *
3668 * If the inode has been found through an idmapped mount the idmap of
3669 * the vfsmount must be passed through @idmap. This function will then take
3670 * care to map the inode according to @idmap before checking permissions.
3671 * On non-idmapped mounts or if permission checking is to be performed on the
3672 * raw inode simply pass @nop_mnt_idmap.
3673 */
3674static int vfs_tmpfile(struct mnt_idmap *idmap,
3675		       const struct path *parentpath,
3676		       struct file *file, umode_t mode)
3677{
3678	struct dentry *child;
3679	struct inode *dir = d_inode(parentpath->dentry);
3680	struct inode *inode;
3681	int error;
3682	int open_flag = file->f_flags;
3683
3684	/* we want directory to be writable */
3685	error = inode_permission(idmap, dir, MAY_WRITE | MAY_EXEC);
3686	if (error)
3687		return error;
3688	if (!dir->i_op->tmpfile)
3689		return -EOPNOTSUPP;
3690	child = d_alloc(parentpath->dentry, &slash_name);
3691	if (unlikely(!child))
3692		return -ENOMEM;
3693	file->f_path.mnt = parentpath->mnt;
3694	file->f_path.dentry = child;
3695	mode = vfs_prepare_mode(idmap, dir, mode, mode, mode);
3696	error = dir->i_op->tmpfile(idmap, dir, file, mode);
3697	dput(child);
3698	if (error)
3699		return error;
3700	/* Don't check for other permissions, the inode was just created */
3701	error = may_open(idmap, &file->f_path, 0, file->f_flags);
3702	if (error)
3703		return error;
3704	inode = file_inode(file);
3705	if (!(open_flag & O_EXCL)) {
3706		spin_lock(&inode->i_lock);
3707		inode->i_state |= I_LINKABLE;
3708		spin_unlock(&inode->i_lock);
3709	}
3710	ima_post_create_tmpfile(idmap, inode);
3711	return 0;
3712}
3713
3714/**
3715 * kernel_tmpfile_open - open a tmpfile for kernel internal use
3716 * @idmap:	idmap of the mount the inode was found from
3717 * @parentpath:	path of the base directory
3718 * @mode:	mode of the new tmpfile
3719 * @open_flag:	flags
3720 * @cred:	credentials for open
3721 *
3722 * Create and open a temporary file.  The file is not accounted in nr_files,
3723 * hence this is only for kernel internal use, and must not be installed into
3724 * file tables or such.
3725 */
3726struct file *kernel_tmpfile_open(struct mnt_idmap *idmap,
3727				 const struct path *parentpath,
3728				 umode_t mode, int open_flag,
3729				 const struct cred *cred)
3730{
3731	struct file *file;
3732	int error;
3733
3734	file = alloc_empty_file_noaccount(open_flag, cred);
3735	if (IS_ERR(file))
3736		return file;
3737
3738	error = vfs_tmpfile(idmap, parentpath, file, mode);
3739	if (error) {
3740		fput(file);
3741		file = ERR_PTR(error);
3742	}
3743	return file;
3744}
3745EXPORT_SYMBOL(kernel_tmpfile_open);
3746
3747static int do_tmpfile(struct nameidata *nd, unsigned flags,
3748		const struct open_flags *op,
3749		struct file *file)
3750{
 
3751	struct path path;
3752	int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
3753
3754	if (unlikely(error))
3755		return error;
3756	error = mnt_want_write(path.mnt);
3757	if (unlikely(error))
3758		goto out;
3759	error = vfs_tmpfile(mnt_idmap(path.mnt), &path, file, op->mode);
 
3760	if (error)
3761		goto out2;
3762	audit_inode(nd->name, file->f_path.dentry, 0);
3763out2:
3764	mnt_drop_write(path.mnt);
3765out:
3766	path_put(&path);
3767	return error;
3768}
3769
3770static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file)
3771{
3772	struct path path;
3773	int error = path_lookupat(nd, flags, &path);
3774	if (!error) {
3775		audit_inode(nd->name, path.dentry, 0);
3776		error = vfs_open(&path, file);
3777		path_put(&path);
3778	}
3779	return error;
3780}
3781
3782static struct file *path_openat(struct nameidata *nd,
3783			const struct open_flags *op, unsigned flags)
3784{
3785	struct file *file;
3786	int error;
3787
3788	file = alloc_empty_file(op->open_flag, current_cred());
3789	if (IS_ERR(file))
3790		return file;
3791
3792	if (unlikely(file->f_flags & __O_TMPFILE)) {
3793		error = do_tmpfile(nd, flags, op, file);
3794	} else if (unlikely(file->f_flags & O_PATH)) {
3795		error = do_o_path(nd, flags, file);
3796	} else {
3797		const char *s = path_init(nd, flags);
3798		while (!(error = link_path_walk(s, nd)) &&
3799		       (s = open_last_lookups(nd, file, op)) != NULL)
3800			;
3801		if (!error)
3802			error = do_open(nd, file, op);
3803		terminate_walk(nd);
3804	}
3805	if (likely(!error)) {
3806		if (likely(file->f_mode & FMODE_OPENED))
3807			return file;
3808		WARN_ON(1);
3809		error = -EINVAL;
3810	}
3811	fput(file);
3812	if (error == -EOPENSTALE) {
3813		if (flags & LOOKUP_RCU)
3814			error = -ECHILD;
3815		else
3816			error = -ESTALE;
3817	}
3818	return ERR_PTR(error);
3819}
3820
3821struct file *do_filp_open(int dfd, struct filename *pathname,
3822		const struct open_flags *op)
3823{
3824	struct nameidata nd;
3825	int flags = op->lookup_flags;
3826	struct file *filp;
3827
3828	set_nameidata(&nd, dfd, pathname, NULL);
3829	filp = path_openat(&nd, op, flags | LOOKUP_RCU);
3830	if (unlikely(filp == ERR_PTR(-ECHILD)))
3831		filp = path_openat(&nd, op, flags);
3832	if (unlikely(filp == ERR_PTR(-ESTALE)))
3833		filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
3834	restore_nameidata();
3835	return filp;
3836}
3837
3838struct file *do_file_open_root(const struct path *root,
3839		const char *name, const struct open_flags *op)
3840{
3841	struct nameidata nd;
3842	struct file *file;
3843	struct filename *filename;
3844	int flags = op->lookup_flags;
3845
3846	if (d_is_symlink(root->dentry) && op->intent & LOOKUP_OPEN)
3847		return ERR_PTR(-ELOOP);
3848
3849	filename = getname_kernel(name);
3850	if (IS_ERR(filename))
3851		return ERR_CAST(filename);
3852
3853	set_nameidata(&nd, -1, filename, root);
3854	file = path_openat(&nd, op, flags | LOOKUP_RCU);
3855	if (unlikely(file == ERR_PTR(-ECHILD)))
3856		file = path_openat(&nd, op, flags);
3857	if (unlikely(file == ERR_PTR(-ESTALE)))
3858		file = path_openat(&nd, op, flags | LOOKUP_REVAL);
3859	restore_nameidata();
3860	putname(filename);
3861	return file;
3862}
3863
3864static struct dentry *filename_create(int dfd, struct filename *name,
3865				      struct path *path, unsigned int lookup_flags)
3866{
3867	struct dentry *dentry = ERR_PTR(-EEXIST);
3868	struct qstr last;
3869	bool want_dir = lookup_flags & LOOKUP_DIRECTORY;
3870	unsigned int reval_flag = lookup_flags & LOOKUP_REVAL;
3871	unsigned int create_flags = LOOKUP_CREATE | LOOKUP_EXCL;
3872	int type;
3873	int err2;
3874	int error;
3875
3876	error = filename_parentat(dfd, name, reval_flag, path, &last, &type);
3877	if (error)
3878		return ERR_PTR(error);
3879
3880	/*
3881	 * Yucky last component or no last component at all?
3882	 * (foo/., foo/.., /////)
3883	 */
3884	if (unlikely(type != LAST_NORM))
3885		goto out;
3886
3887	/* don't fail immediately if it's r/o, at least try to report other errors */
3888	err2 = mnt_want_write(path->mnt);
3889	/*
3890	 * Do the final lookup.  Suppress 'create' if there is a trailing
3891	 * '/', and a directory wasn't requested.
3892	 */
3893	if (last.name[last.len] && !want_dir)
3894		create_flags = 0;
3895	inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
3896	dentry = lookup_one_qstr_excl(&last, path->dentry,
3897				      reval_flag | create_flags);
3898	if (IS_ERR(dentry))
3899		goto unlock;
3900
3901	error = -EEXIST;
3902	if (d_is_positive(dentry))
3903		goto fail;
3904
3905	/*
3906	 * Special case - lookup gave negative, but... we had foo/bar/
3907	 * From the vfs_mknod() POV we just have a negative dentry -
3908	 * all is fine. Let's be bastards - you had / on the end, you've
3909	 * been asking for (non-existent) directory. -ENOENT for you.
3910	 */
3911	if (unlikely(!create_flags)) {
3912		error = -ENOENT;
3913		goto fail;
3914	}
3915	if (unlikely(err2)) {
3916		error = err2;
3917		goto fail;
3918	}
3919	return dentry;
3920fail:
3921	dput(dentry);
3922	dentry = ERR_PTR(error);
3923unlock:
3924	inode_unlock(path->dentry->d_inode);
3925	if (!err2)
3926		mnt_drop_write(path->mnt);
3927out:
3928	path_put(path);
3929	return dentry;
3930}
3931
3932struct dentry *kern_path_create(int dfd, const char *pathname,
3933				struct path *path, unsigned int lookup_flags)
3934{
3935	struct filename *filename = getname_kernel(pathname);
3936	struct dentry *res = filename_create(dfd, filename, path, lookup_flags);
3937
3938	putname(filename);
3939	return res;
3940}
3941EXPORT_SYMBOL(kern_path_create);
3942
3943void done_path_create(struct path *path, struct dentry *dentry)
3944{
3945	dput(dentry);
3946	inode_unlock(path->dentry->d_inode);
3947	mnt_drop_write(path->mnt);
3948	path_put(path);
3949}
3950EXPORT_SYMBOL(done_path_create);
3951
3952inline struct dentry *user_path_create(int dfd, const char __user *pathname,
3953				struct path *path, unsigned int lookup_flags)
3954{
3955	struct filename *filename = getname(pathname);
3956	struct dentry *res = filename_create(dfd, filename, path, lookup_flags);
3957
3958	putname(filename);
3959	return res;
3960}
3961EXPORT_SYMBOL(user_path_create);
3962
3963/**
3964 * vfs_mknod - create device node or file
3965 * @idmap:	idmap of the mount the inode was found from
3966 * @dir:	inode of @dentry
3967 * @dentry:	pointer to dentry of the base directory
3968 * @mode:	mode of the new device node or file
3969 * @dev:	device number of device to create
3970 *
3971 * Create a device node or file.
3972 *
3973 * If the inode has been found through an idmapped mount the idmap of
3974 * the vfsmount must be passed through @idmap. This function will then take
3975 * care to map the inode according to @idmap before checking permissions.
3976 * On non-idmapped mounts or if permission checking is to be performed on the
3977 * raw inode simply pass @nop_mnt_idmap.
3978 */
3979int vfs_mknod(struct mnt_idmap *idmap, struct inode *dir,
3980	      struct dentry *dentry, umode_t mode, dev_t dev)
3981{
3982	bool is_whiteout = S_ISCHR(mode) && dev == WHITEOUT_DEV;
3983	int error = may_create(idmap, dir, dentry);
3984
3985	if (error)
3986		return error;
3987
3988	if ((S_ISCHR(mode) || S_ISBLK(mode)) && !is_whiteout &&
3989	    !capable(CAP_MKNOD))
3990		return -EPERM;
3991
3992	if (!dir->i_op->mknod)
3993		return -EPERM;
3994
3995	mode = vfs_prepare_mode(idmap, dir, mode, mode, mode);
3996	error = devcgroup_inode_mknod(mode, dev);
3997	if (error)
3998		return error;
3999
4000	error = security_inode_mknod(dir, dentry, mode, dev);
4001	if (error)
4002		return error;
4003
4004	error = dir->i_op->mknod(idmap, dir, dentry, mode, dev);
4005	if (!error)
4006		fsnotify_create(dir, dentry);
4007	return error;
4008}
4009EXPORT_SYMBOL(vfs_mknod);
4010
4011static int may_mknod(umode_t mode)
4012{
4013	switch (mode & S_IFMT) {
4014	case S_IFREG:
4015	case S_IFCHR:
4016	case S_IFBLK:
4017	case S_IFIFO:
4018	case S_IFSOCK:
4019	case 0: /* zero mode translates to S_IFREG */
4020		return 0;
4021	case S_IFDIR:
4022		return -EPERM;
4023	default:
4024		return -EINVAL;
4025	}
4026}
4027
4028static int do_mknodat(int dfd, struct filename *name, umode_t mode,
4029		unsigned int dev)
4030{
4031	struct mnt_idmap *idmap;
4032	struct dentry *dentry;
4033	struct path path;
4034	int error;
4035	unsigned int lookup_flags = 0;
4036
4037	error = may_mknod(mode);
4038	if (error)
4039		goto out1;
4040retry:
4041	dentry = filename_create(dfd, name, &path, lookup_flags);
4042	error = PTR_ERR(dentry);
4043	if (IS_ERR(dentry))
4044		goto out1;
4045
4046	error = security_path_mknod(&path, dentry,
4047			mode_strip_umask(path.dentry->d_inode, mode), dev);
4048	if (error)
4049		goto out2;
4050
4051	idmap = mnt_idmap(path.mnt);
4052	switch (mode & S_IFMT) {
4053		case 0: case S_IFREG:
4054			error = vfs_create(idmap, path.dentry->d_inode,
4055					   dentry, mode, true);
4056			if (!error)
4057				ima_post_path_mknod(idmap, dentry);
4058			break;
4059		case S_IFCHR: case S_IFBLK:
4060			error = vfs_mknod(idmap, path.dentry->d_inode,
4061					  dentry, mode, new_decode_dev(dev));
4062			break;
4063		case S_IFIFO: case S_IFSOCK:
4064			error = vfs_mknod(idmap, path.dentry->d_inode,
4065					  dentry, mode, 0);
4066			break;
4067	}
4068out2:
4069	done_path_create(&path, dentry);
4070	if (retry_estale(error, lookup_flags)) {
4071		lookup_flags |= LOOKUP_REVAL;
4072		goto retry;
4073	}
4074out1:
4075	putname(name);
4076	return error;
4077}
4078
4079SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
4080		unsigned int, dev)
4081{
4082	return do_mknodat(dfd, getname(filename), mode, dev);
4083}
4084
4085SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
4086{
4087	return do_mknodat(AT_FDCWD, getname(filename), mode, dev);
4088}
4089
4090/**
4091 * vfs_mkdir - create directory
4092 * @idmap:	idmap of the mount the inode was found from
4093 * @dir:	inode of @dentry
4094 * @dentry:	pointer to dentry of the base directory
4095 * @mode:	mode of the new directory
4096 *
4097 * Create a directory.
4098 *
4099 * If the inode has been found through an idmapped mount the idmap of
4100 * the vfsmount must be passed through @idmap. This function will then take
4101 * care to map the inode according to @idmap before checking permissions.
4102 * On non-idmapped mounts or if permission checking is to be performed on the
4103 * raw inode simply pass @nop_mnt_idmap.
4104 */
4105int vfs_mkdir(struct mnt_idmap *idmap, struct inode *dir,
4106	      struct dentry *dentry, umode_t mode)
4107{
4108	int error;
4109	unsigned max_links = dir->i_sb->s_max_links;
4110
4111	error = may_create(idmap, dir, dentry);
4112	if (error)
4113		return error;
4114
4115	if (!dir->i_op->mkdir)
4116		return -EPERM;
4117
4118	mode = vfs_prepare_mode(idmap, dir, mode, S_IRWXUGO | S_ISVTX, 0);
4119	error = security_inode_mkdir(dir, dentry, mode);
4120	if (error)
4121		return error;
4122
4123	if (max_links && dir->i_nlink >= max_links)
4124		return -EMLINK;
4125
4126	error = dir->i_op->mkdir(idmap, dir, dentry, mode);
4127	if (!error)
4128		fsnotify_mkdir(dir, dentry);
4129	return error;
4130}
4131EXPORT_SYMBOL(vfs_mkdir);
4132
4133int do_mkdirat(int dfd, struct filename *name, umode_t mode)
4134{
4135	struct dentry *dentry;
4136	struct path path;
4137	int error;
4138	unsigned int lookup_flags = LOOKUP_DIRECTORY;
4139
4140retry:
4141	dentry = filename_create(dfd, name, &path, lookup_flags);
4142	error = PTR_ERR(dentry);
4143	if (IS_ERR(dentry))
4144		goto out_putname;
4145
4146	error = security_path_mkdir(&path, dentry,
4147			mode_strip_umask(path.dentry->d_inode, mode));
4148	if (!error) {
4149		error = vfs_mkdir(mnt_idmap(path.mnt), path.dentry->d_inode,
4150				  dentry, mode);
 
 
4151	}
4152	done_path_create(&path, dentry);
4153	if (retry_estale(error, lookup_flags)) {
4154		lookup_flags |= LOOKUP_REVAL;
4155		goto retry;
4156	}
4157out_putname:
4158	putname(name);
4159	return error;
4160}
4161
4162SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
4163{
4164	return do_mkdirat(dfd, getname(pathname), mode);
4165}
4166
4167SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
4168{
4169	return do_mkdirat(AT_FDCWD, getname(pathname), mode);
4170}
4171
4172/**
4173 * vfs_rmdir - remove directory
4174 * @idmap:	idmap of the mount the inode was found from
4175 * @dir:	inode of @dentry
4176 * @dentry:	pointer to dentry of the base directory
4177 *
4178 * Remove a directory.
4179 *
4180 * If the inode has been found through an idmapped mount the idmap of
4181 * the vfsmount must be passed through @idmap. This function will then take
4182 * care to map the inode according to @idmap before checking permissions.
4183 * On non-idmapped mounts or if permission checking is to be performed on the
4184 * raw inode simply pass @nop_mnt_idmap.
4185 */
4186int vfs_rmdir(struct mnt_idmap *idmap, struct inode *dir,
4187		     struct dentry *dentry)
4188{
4189	int error = may_delete(idmap, dir, dentry, 1);
4190
4191	if (error)
4192		return error;
4193
4194	if (!dir->i_op->rmdir)
4195		return -EPERM;
4196
4197	dget(dentry);
4198	inode_lock(dentry->d_inode);
4199
4200	error = -EBUSY;
4201	if (is_local_mountpoint(dentry) ||
4202	    (dentry->d_inode->i_flags & S_KERNEL_FILE))
4203		goto out;
4204
4205	error = security_inode_rmdir(dir, dentry);
4206	if (error)
4207		goto out;
4208
4209	error = dir->i_op->rmdir(dir, dentry);
4210	if (error)
4211		goto out;
4212
4213	shrink_dcache_parent(dentry);
4214	dentry->d_inode->i_flags |= S_DEAD;
4215	dont_mount(dentry);
4216	detach_mounts(dentry);
4217
4218out:
4219	inode_unlock(dentry->d_inode);
4220	dput(dentry);
4221	if (!error)
4222		d_delete_notify(dir, dentry);
4223	return error;
4224}
4225EXPORT_SYMBOL(vfs_rmdir);
4226
4227int do_rmdir(int dfd, struct filename *name)
4228{
 
4229	int error;
4230	struct dentry *dentry;
4231	struct path path;
4232	struct qstr last;
4233	int type;
4234	unsigned int lookup_flags = 0;
4235retry:
4236	error = filename_parentat(dfd, name, lookup_flags, &path, &last, &type);
4237	if (error)
4238		goto exit1;
4239
4240	switch (type) {
4241	case LAST_DOTDOT:
4242		error = -ENOTEMPTY;
4243		goto exit2;
4244	case LAST_DOT:
4245		error = -EINVAL;
4246		goto exit2;
4247	case LAST_ROOT:
4248		error = -EBUSY;
4249		goto exit2;
4250	}
4251
4252	error = mnt_want_write(path.mnt);
4253	if (error)
4254		goto exit2;
4255
4256	inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
4257	dentry = lookup_one_qstr_excl(&last, path.dentry, lookup_flags);
4258	error = PTR_ERR(dentry);
4259	if (IS_ERR(dentry))
4260		goto exit3;
4261	if (!dentry->d_inode) {
4262		error = -ENOENT;
4263		goto exit4;
4264	}
4265	error = security_path_rmdir(&path, dentry);
4266	if (error)
4267		goto exit4;
4268	error = vfs_rmdir(mnt_idmap(path.mnt), path.dentry->d_inode, dentry);
 
4269exit4:
4270	dput(dentry);
4271exit3:
4272	inode_unlock(path.dentry->d_inode);
4273	mnt_drop_write(path.mnt);
4274exit2:
4275	path_put(&path);
4276	if (retry_estale(error, lookup_flags)) {
4277		lookup_flags |= LOOKUP_REVAL;
4278		goto retry;
4279	}
4280exit1:
4281	putname(name);
4282	return error;
4283}
4284
4285SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
4286{
4287	return do_rmdir(AT_FDCWD, getname(pathname));
4288}
4289
4290/**
4291 * vfs_unlink - unlink a filesystem object
4292 * @idmap:	idmap of the mount the inode was found from
4293 * @dir:	parent directory
4294 * @dentry:	victim
4295 * @delegated_inode: returns victim inode, if the inode is delegated.
4296 *
4297 * The caller must hold dir->i_mutex.
4298 *
4299 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
4300 * return a reference to the inode in delegated_inode.  The caller
4301 * should then break the delegation on that inode and retry.  Because
4302 * breaking a delegation may take a long time, the caller should drop
4303 * dir->i_mutex before doing so.
4304 *
4305 * Alternatively, a caller may pass NULL for delegated_inode.  This may
4306 * be appropriate for callers that expect the underlying filesystem not
4307 * to be NFS exported.
4308 *
4309 * If the inode has been found through an idmapped mount the idmap of
4310 * the vfsmount must be passed through @idmap. This function will then take
4311 * care to map the inode according to @idmap before checking permissions.
4312 * On non-idmapped mounts or if permission checking is to be performed on the
4313 * raw inode simply pass @nop_mnt_idmap.
4314 */
4315int vfs_unlink(struct mnt_idmap *idmap, struct inode *dir,
4316	       struct dentry *dentry, struct inode **delegated_inode)
4317{
4318	struct inode *target = dentry->d_inode;
4319	int error = may_delete(idmap, dir, dentry, 0);
4320
4321	if (error)
4322		return error;
4323
4324	if (!dir->i_op->unlink)
4325		return -EPERM;
4326
4327	inode_lock(target);
4328	if (IS_SWAPFILE(target))
4329		error = -EPERM;
4330	else if (is_local_mountpoint(dentry))
4331		error = -EBUSY;
4332	else {
4333		error = security_inode_unlink(dir, dentry);
4334		if (!error) {
4335			error = try_break_deleg(target, delegated_inode);
4336			if (error)
4337				goto out;
4338			error = dir->i_op->unlink(dir, dentry);
4339			if (!error) {
4340				dont_mount(dentry);
4341				detach_mounts(dentry);
4342			}
4343		}
4344	}
4345out:
4346	inode_unlock(target);
4347
4348	/* We don't d_delete() NFS sillyrenamed files--they still exist. */
4349	if (!error && dentry->d_flags & DCACHE_NFSFS_RENAMED) {
4350		fsnotify_unlink(dir, dentry);
4351	} else if (!error) {
4352		fsnotify_link_count(target);
4353		d_delete_notify(dir, dentry);
4354	}
4355
4356	return error;
4357}
4358EXPORT_SYMBOL(vfs_unlink);
4359
4360/*
4361 * Make sure that the actual truncation of the file will occur outside its
4362 * directory's i_mutex.  Truncate can take a long time if there is a lot of
4363 * writeout happening, and we don't want to prevent access to the directory
4364 * while waiting on the I/O.
4365 */
4366int do_unlinkat(int dfd, struct filename *name)
4367{
4368	int error;
4369	struct dentry *dentry;
4370	struct path path;
4371	struct qstr last;
4372	int type;
4373	struct inode *inode = NULL;
4374	struct inode *delegated_inode = NULL;
4375	unsigned int lookup_flags = 0;
4376retry:
4377	error = filename_parentat(dfd, name, lookup_flags, &path, &last, &type);
4378	if (error)
4379		goto exit1;
4380
4381	error = -EISDIR;
4382	if (type != LAST_NORM)
4383		goto exit2;
4384
4385	error = mnt_want_write(path.mnt);
4386	if (error)
4387		goto exit2;
4388retry_deleg:
4389	inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
4390	dentry = lookup_one_qstr_excl(&last, path.dentry, lookup_flags);
4391	error = PTR_ERR(dentry);
4392	if (!IS_ERR(dentry)) {
 
4393
4394		/* Why not before? Because we want correct error value */
4395		if (last.name[last.len] || d_is_negative(dentry))
4396			goto slashes;
4397		inode = dentry->d_inode;
 
 
4398		ihold(inode);
4399		error = security_path_unlink(&path, dentry);
4400		if (error)
4401			goto exit3;
4402		error = vfs_unlink(mnt_idmap(path.mnt), path.dentry->d_inode,
4403				   dentry, &delegated_inode);
 
4404exit3:
4405		dput(dentry);
4406	}
4407	inode_unlock(path.dentry->d_inode);
4408	if (inode)
4409		iput(inode);	/* truncate the inode here */
4410	inode = NULL;
4411	if (delegated_inode) {
4412		error = break_deleg_wait(&delegated_inode);
4413		if (!error)
4414			goto retry_deleg;
4415	}
4416	mnt_drop_write(path.mnt);
4417exit2:
4418	path_put(&path);
4419	if (retry_estale(error, lookup_flags)) {
4420		lookup_flags |= LOOKUP_REVAL;
4421		inode = NULL;
4422		goto retry;
4423	}
4424exit1:
4425	putname(name);
4426	return error;
4427
4428slashes:
4429	if (d_is_negative(dentry))
4430		error = -ENOENT;
4431	else if (d_is_dir(dentry))
4432		error = -EISDIR;
4433	else
4434		error = -ENOTDIR;
4435	goto exit3;
4436}
4437
4438SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
4439{
4440	if ((flag & ~AT_REMOVEDIR) != 0)
4441		return -EINVAL;
4442
4443	if (flag & AT_REMOVEDIR)
4444		return do_rmdir(dfd, getname(pathname));
4445	return do_unlinkat(dfd, getname(pathname));
4446}
4447
4448SYSCALL_DEFINE1(unlink, const char __user *, pathname)
4449{
4450	return do_unlinkat(AT_FDCWD, getname(pathname));
4451}
4452
4453/**
4454 * vfs_symlink - create symlink
4455 * @idmap:	idmap of the mount the inode was found from
4456 * @dir:	inode of @dentry
4457 * @dentry:	pointer to dentry of the base directory
4458 * @oldname:	name of the file to link to
4459 *
4460 * Create a symlink.
4461 *
4462 * If the inode has been found through an idmapped mount the idmap of
4463 * the vfsmount must be passed through @idmap. This function will then take
4464 * care to map the inode according to @idmap before checking permissions.
4465 * On non-idmapped mounts or if permission checking is to be performed on the
4466 * raw inode simply pass @nop_mnt_idmap.
4467 */
4468int vfs_symlink(struct mnt_idmap *idmap, struct inode *dir,
4469		struct dentry *dentry, const char *oldname)
4470{
4471	int error;
4472
4473	error = may_create(idmap, dir, dentry);
4474	if (error)
4475		return error;
4476
4477	if (!dir->i_op->symlink)
4478		return -EPERM;
4479
4480	error = security_inode_symlink(dir, dentry, oldname);
4481	if (error)
4482		return error;
4483
4484	error = dir->i_op->symlink(idmap, dir, dentry, oldname);
4485	if (!error)
4486		fsnotify_create(dir, dentry);
4487	return error;
4488}
4489EXPORT_SYMBOL(vfs_symlink);
4490
4491int do_symlinkat(struct filename *from, int newdfd, struct filename *to)
4492{
4493	int error;
4494	struct dentry *dentry;
4495	struct path path;
4496	unsigned int lookup_flags = 0;
4497
4498	if (IS_ERR(from)) {
4499		error = PTR_ERR(from);
4500		goto out_putnames;
4501	}
4502retry:
4503	dentry = filename_create(newdfd, to, &path, lookup_flags);
4504	error = PTR_ERR(dentry);
4505	if (IS_ERR(dentry))
4506		goto out_putnames;
4507
4508	error = security_path_symlink(&path, dentry, from->name);
4509	if (!error)
4510		error = vfs_symlink(mnt_idmap(path.mnt), path.dentry->d_inode,
4511				    dentry, from->name);
 
 
 
 
4512	done_path_create(&path, dentry);
4513	if (retry_estale(error, lookup_flags)) {
4514		lookup_flags |= LOOKUP_REVAL;
4515		goto retry;
4516	}
4517out_putnames:
4518	putname(to);
4519	putname(from);
4520	return error;
4521}
4522
4523SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
4524		int, newdfd, const char __user *, newname)
4525{
4526	return do_symlinkat(getname(oldname), newdfd, getname(newname));
4527}
4528
4529SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
4530{
4531	return do_symlinkat(getname(oldname), AT_FDCWD, getname(newname));
4532}
4533
4534/**
4535 * vfs_link - create a new link
4536 * @old_dentry:	object to be linked
4537 * @idmap:	idmap of the mount
4538 * @dir:	new parent
4539 * @new_dentry:	where to create the new link
4540 * @delegated_inode: returns inode needing a delegation break
4541 *
4542 * The caller must hold dir->i_mutex
4543 *
4544 * If vfs_link discovers a delegation on the to-be-linked file in need
4545 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4546 * inode in delegated_inode.  The caller should then break the delegation
4547 * and retry.  Because breaking a delegation may take a long time, the
4548 * caller should drop the i_mutex before doing so.
4549 *
4550 * Alternatively, a caller may pass NULL for delegated_inode.  This may
4551 * be appropriate for callers that expect the underlying filesystem not
4552 * to be NFS exported.
4553 *
4554 * If the inode has been found through an idmapped mount the idmap of
4555 * the vfsmount must be passed through @idmap. This function will then take
4556 * care to map the inode according to @idmap before checking permissions.
4557 * On non-idmapped mounts or if permission checking is to be performed on the
4558 * raw inode simply pass @nop_mnt_idmap.
4559 */
4560int vfs_link(struct dentry *old_dentry, struct mnt_idmap *idmap,
4561	     struct inode *dir, struct dentry *new_dentry,
4562	     struct inode **delegated_inode)
4563{
4564	struct inode *inode = old_dentry->d_inode;
4565	unsigned max_links = dir->i_sb->s_max_links;
4566	int error;
4567
4568	if (!inode)
4569		return -ENOENT;
4570
4571	error = may_create(idmap, dir, new_dentry);
4572	if (error)
4573		return error;
4574
4575	if (dir->i_sb != inode->i_sb)
4576		return -EXDEV;
4577
4578	/*
4579	 * A link to an append-only or immutable file cannot be created.
4580	 */
4581	if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4582		return -EPERM;
4583	/*
4584	 * Updating the link count will likely cause i_uid and i_gid to
4585	 * be writen back improperly if their true value is unknown to
4586	 * the vfs.
4587	 */
4588	if (HAS_UNMAPPED_ID(idmap, inode))
4589		return -EPERM;
4590	if (!dir->i_op->link)
4591		return -EPERM;
4592	if (S_ISDIR(inode->i_mode))
4593		return -EPERM;
4594
4595	error = security_inode_link(old_dentry, dir, new_dentry);
4596	if (error)
4597		return error;
4598
4599	inode_lock(inode);
4600	/* Make sure we don't allow creating hardlink to an unlinked file */
4601	if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4602		error =  -ENOENT;
4603	else if (max_links && inode->i_nlink >= max_links)
4604		error = -EMLINK;
4605	else {
4606		error = try_break_deleg(inode, delegated_inode);
4607		if (!error)
4608			error = dir->i_op->link(old_dentry, dir, new_dentry);
4609	}
4610
4611	if (!error && (inode->i_state & I_LINKABLE)) {
4612		spin_lock(&inode->i_lock);
4613		inode->i_state &= ~I_LINKABLE;
4614		spin_unlock(&inode->i_lock);
4615	}
4616	inode_unlock(inode);
4617	if (!error)
4618		fsnotify_link(dir, inode, new_dentry);
4619	return error;
4620}
4621EXPORT_SYMBOL(vfs_link);
4622
4623/*
4624 * Hardlinks are often used in delicate situations.  We avoid
4625 * security-related surprises by not following symlinks on the
4626 * newname.  --KAB
4627 *
4628 * We don't follow them on the oldname either to be compatible
4629 * with linux 2.0, and to avoid hard-linking to directories
4630 * and other special files.  --ADM
4631 */
4632int do_linkat(int olddfd, struct filename *old, int newdfd,
4633	      struct filename *new, int flags)
4634{
4635	struct mnt_idmap *idmap;
4636	struct dentry *new_dentry;
4637	struct path old_path, new_path;
4638	struct inode *delegated_inode = NULL;
4639	int how = 0;
4640	int error;
4641
4642	if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) {
4643		error = -EINVAL;
4644		goto out_putnames;
4645	}
4646	/*
4647	 * To use null names we require CAP_DAC_READ_SEARCH
4648	 * This ensures that not everyone will be able to create
4649	 * handlink using the passed filedescriptor.
4650	 */
4651	if (flags & AT_EMPTY_PATH && !capable(CAP_DAC_READ_SEARCH)) {
4652		error = -ENOENT;
4653		goto out_putnames;
4654	}
4655
4656	if (flags & AT_SYMLINK_FOLLOW)
4657		how |= LOOKUP_FOLLOW;
4658retry:
4659	error = filename_lookup(olddfd, old, how, &old_path, NULL);
4660	if (error)
4661		goto out_putnames;
4662
4663	new_dentry = filename_create(newdfd, new, &new_path,
4664					(how & LOOKUP_REVAL));
4665	error = PTR_ERR(new_dentry);
4666	if (IS_ERR(new_dentry))
4667		goto out_putpath;
4668
4669	error = -EXDEV;
4670	if (old_path.mnt != new_path.mnt)
4671		goto out_dput;
4672	idmap = mnt_idmap(new_path.mnt);
4673	error = may_linkat(idmap, &old_path);
4674	if (unlikely(error))
4675		goto out_dput;
4676	error = security_path_link(old_path.dentry, &new_path, new_dentry);
4677	if (error)
4678		goto out_dput;
4679	error = vfs_link(old_path.dentry, idmap, new_path.dentry->d_inode,
4680			 new_dentry, &delegated_inode);
4681out_dput:
4682	done_path_create(&new_path, new_dentry);
4683	if (delegated_inode) {
4684		error = break_deleg_wait(&delegated_inode);
4685		if (!error) {
4686			path_put(&old_path);
4687			goto retry;
4688		}
4689	}
4690	if (retry_estale(error, how)) {
4691		path_put(&old_path);
4692		how |= LOOKUP_REVAL;
4693		goto retry;
4694	}
4695out_putpath:
4696	path_put(&old_path);
4697out_putnames:
4698	putname(old);
4699	putname(new);
4700
4701	return error;
4702}
4703
4704SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4705		int, newdfd, const char __user *, newname, int, flags)
4706{
4707	return do_linkat(olddfd, getname_uflags(oldname, flags),
4708		newdfd, getname(newname), flags);
4709}
4710
4711SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4712{
4713	return do_linkat(AT_FDCWD, getname(oldname), AT_FDCWD, getname(newname), 0);
4714}
4715
4716/**
4717 * vfs_rename - rename a filesystem object
4718 * @rd:		pointer to &struct renamedata info
4719 *
4720 * The caller must hold multiple mutexes--see lock_rename()).
4721 *
4722 * If vfs_rename discovers a delegation in need of breaking at either
4723 * the source or destination, it will return -EWOULDBLOCK and return a
4724 * reference to the inode in delegated_inode.  The caller should then
4725 * break the delegation and retry.  Because breaking a delegation may
4726 * take a long time, the caller should drop all locks before doing
4727 * so.
4728 *
4729 * Alternatively, a caller may pass NULL for delegated_inode.  This may
4730 * be appropriate for callers that expect the underlying filesystem not
4731 * to be NFS exported.
4732 *
4733 * The worst of all namespace operations - renaming directory. "Perverted"
4734 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4735 * Problems:
4736 *
4737 *	a) we can get into loop creation.
4738 *	b) race potential - two innocent renames can create a loop together.
4739 *	   That's where 4.4BSD screws up. Current fix: serialization on
4740 *	   sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4741 *	   story.
4742 *	c) we may have to lock up to _four_ objects - parents and victim (if it exists),
4743 *	   and source (if it's a non-directory or a subdirectory that moves to
4744 *	   different parent).
4745 *	   And that - after we got ->i_mutex on parents (until then we don't know
4746 *	   whether the target exists).  Solution: try to be smart with locking
4747 *	   order for inodes.  We rely on the fact that tree topology may change
4748 *	   only under ->s_vfs_rename_mutex _and_ that parent of the object we
4749 *	   move will be locked.  Thus we can rank directories by the tree
4750 *	   (ancestors first) and rank all non-directories after them.
4751 *	   That works since everybody except rename does "lock parent, lookup,
4752 *	   lock child" and rename is under ->s_vfs_rename_mutex.
4753 *	   HOWEVER, it relies on the assumption that any object with ->lookup()
4754 *	   has no more than 1 dentry.  If "hybrid" objects will ever appear,
4755 *	   we'd better make sure that there's no link(2) for them.
4756 *	d) conversion from fhandle to dentry may come in the wrong moment - when
4757 *	   we are removing the target. Solution: we will have to grab ->i_mutex
4758 *	   in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4759 *	   ->i_mutex on parents, which works but leads to some truly excessive
4760 *	   locking].
4761 */
4762int vfs_rename(struct renamedata *rd)
4763{
4764	int error;
4765	struct inode *old_dir = rd->old_dir, *new_dir = rd->new_dir;
4766	struct dentry *old_dentry = rd->old_dentry;
4767	struct dentry *new_dentry = rd->new_dentry;
4768	struct inode **delegated_inode = rd->delegated_inode;
4769	unsigned int flags = rd->flags;
4770	bool is_dir = d_is_dir(old_dentry);
4771	struct inode *source = old_dentry->d_inode;
4772	struct inode *target = new_dentry->d_inode;
4773	bool new_is_dir = false;
4774	unsigned max_links = new_dir->i_sb->s_max_links;
4775	struct name_snapshot old_name;
4776	bool lock_old_subdir, lock_new_subdir;
4777
4778	if (source == target)
4779		return 0;
4780
4781	error = may_delete(rd->old_mnt_idmap, old_dir, old_dentry, is_dir);
4782	if (error)
4783		return error;
4784
4785	if (!target) {
4786		error = may_create(rd->new_mnt_idmap, new_dir, new_dentry);
4787	} else {
4788		new_is_dir = d_is_dir(new_dentry);
4789
4790		if (!(flags & RENAME_EXCHANGE))
4791			error = may_delete(rd->new_mnt_idmap, new_dir,
4792					   new_dentry, is_dir);
4793		else
4794			error = may_delete(rd->new_mnt_idmap, new_dir,
4795					   new_dentry, new_is_dir);
4796	}
4797	if (error)
4798		return error;
4799
4800	if (!old_dir->i_op->rename)
4801		return -EPERM;
4802
4803	/*
4804	 * If we are going to change the parent - check write permissions,
4805	 * we'll need to flip '..'.
4806	 */
4807	if (new_dir != old_dir) {
4808		if (is_dir) {
4809			error = inode_permission(rd->old_mnt_idmap, source,
4810						 MAY_WRITE);
4811			if (error)
4812				return error;
4813		}
4814		if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4815			error = inode_permission(rd->new_mnt_idmap, target,
4816						 MAY_WRITE);
4817			if (error)
4818				return error;
4819		}
4820	}
4821
4822	error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4823				      flags);
4824	if (error)
4825		return error;
4826
4827	take_dentry_name_snapshot(&old_name, old_dentry);
4828	dget(new_dentry);
4829	/*
4830	 * Lock children.
4831	 * The source subdirectory needs to be locked on cross-directory
4832	 * rename or cross-directory exchange since its parent changes.
4833	 * The target subdirectory needs to be locked on cross-directory
4834	 * exchange due to parent change and on any rename due to becoming
4835	 * a victim.
4836	 * Non-directories need locking in all cases (for NFS reasons);
4837	 * they get locked after any subdirectories (in inode address order).
4838	 *
4839	 * NOTE: WE ONLY LOCK UNRELATED DIRECTORIES IN CROSS-DIRECTORY CASE.
4840	 * NEVER, EVER DO THAT WITHOUT ->s_vfs_rename_mutex.
4841	 */
4842	lock_old_subdir = new_dir != old_dir;
4843	lock_new_subdir = new_dir != old_dir || !(flags & RENAME_EXCHANGE);
4844	if (is_dir) {
4845		if (lock_old_subdir)
4846			inode_lock_nested(source, I_MUTEX_CHILD);
4847		if (target && (!new_is_dir || lock_new_subdir))
4848			inode_lock(target);
4849	} else if (new_is_dir) {
4850		if (lock_new_subdir)
4851			inode_lock_nested(target, I_MUTEX_CHILD);
4852		inode_lock(source);
4853	} else {
4854		lock_two_nondirectories(source, target);
4855	}
 
4856
4857	error = -EPERM;
4858	if (IS_SWAPFILE(source) || (target && IS_SWAPFILE(target)))
4859		goto out;
4860
4861	error = -EBUSY;
4862	if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4863		goto out;
4864
4865	if (max_links && new_dir != old_dir) {
4866		error = -EMLINK;
4867		if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4868			goto out;
4869		if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4870		    old_dir->i_nlink >= max_links)
4871			goto out;
4872	}
4873	if (!is_dir) {
4874		error = try_break_deleg(source, delegated_inode);
4875		if (error)
4876			goto out;
4877	}
4878	if (target && !new_is_dir) {
4879		error = try_break_deleg(target, delegated_inode);
4880		if (error)
4881			goto out;
4882	}
4883	error = old_dir->i_op->rename(rd->new_mnt_idmap, old_dir, old_dentry,
4884				      new_dir, new_dentry, flags);
4885	if (error)
4886		goto out;
4887
4888	if (!(flags & RENAME_EXCHANGE) && target) {
4889		if (is_dir) {
4890			shrink_dcache_parent(new_dentry);
4891			target->i_flags |= S_DEAD;
4892		}
4893		dont_mount(new_dentry);
4894		detach_mounts(new_dentry);
4895	}
4896	if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4897		if (!(flags & RENAME_EXCHANGE))
4898			d_move(old_dentry, new_dentry);
4899		else
4900			d_exchange(old_dentry, new_dentry);
4901	}
4902out:
4903	if (!is_dir || lock_old_subdir)
4904		inode_unlock(source);
4905	if (target && (!new_is_dir || lock_new_subdir))
4906		inode_unlock(target);
4907	dput(new_dentry);
4908	if (!error) {
4909		fsnotify_move(old_dir, new_dir, &old_name.name, is_dir,
4910			      !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4911		if (flags & RENAME_EXCHANGE) {
4912			fsnotify_move(new_dir, old_dir, &old_dentry->d_name,
4913				      new_is_dir, NULL, new_dentry);
4914		}
4915	}
4916	release_dentry_name_snapshot(&old_name);
4917
4918	return error;
4919}
4920EXPORT_SYMBOL(vfs_rename);
4921
4922int do_renameat2(int olddfd, struct filename *from, int newdfd,
4923		 struct filename *to, unsigned int flags)
4924{
4925	struct renamedata rd;
4926	struct dentry *old_dentry, *new_dentry;
4927	struct dentry *trap;
4928	struct path old_path, new_path;
4929	struct qstr old_last, new_last;
4930	int old_type, new_type;
4931	struct inode *delegated_inode = NULL;
4932	unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4933	bool should_retry = false;
4934	int error = -EINVAL;
4935
4936	if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4937		goto put_names;
4938
4939	if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4940	    (flags & RENAME_EXCHANGE))
4941		goto put_names;
4942
4943	if (flags & RENAME_EXCHANGE)
4944		target_flags = 0;
4945
4946retry:
4947	error = filename_parentat(olddfd, from, lookup_flags, &old_path,
4948				  &old_last, &old_type);
4949	if (error)
4950		goto put_names;
4951
4952	error = filename_parentat(newdfd, to, lookup_flags, &new_path, &new_last,
4953				  &new_type);
4954	if (error)
4955		goto exit1;
4956
4957	error = -EXDEV;
4958	if (old_path.mnt != new_path.mnt)
4959		goto exit2;
4960
4961	error = -EBUSY;
4962	if (old_type != LAST_NORM)
4963		goto exit2;
4964
4965	if (flags & RENAME_NOREPLACE)
4966		error = -EEXIST;
4967	if (new_type != LAST_NORM)
4968		goto exit2;
4969
4970	error = mnt_want_write(old_path.mnt);
4971	if (error)
4972		goto exit2;
4973
4974retry_deleg:
4975	trap = lock_rename(new_path.dentry, old_path.dentry);
4976	if (IS_ERR(trap)) {
4977		error = PTR_ERR(trap);
4978		goto exit_lock_rename;
4979	}
4980
4981	old_dentry = lookup_one_qstr_excl(&old_last, old_path.dentry,
4982					  lookup_flags);
4983	error = PTR_ERR(old_dentry);
4984	if (IS_ERR(old_dentry))
4985		goto exit3;
4986	/* source must exist */
4987	error = -ENOENT;
4988	if (d_is_negative(old_dentry))
4989		goto exit4;
4990	new_dentry = lookup_one_qstr_excl(&new_last, new_path.dentry,
4991					  lookup_flags | target_flags);
4992	error = PTR_ERR(new_dentry);
4993	if (IS_ERR(new_dentry))
4994		goto exit4;
4995	error = -EEXIST;
4996	if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4997		goto exit5;
4998	if (flags & RENAME_EXCHANGE) {
4999		error = -ENOENT;
5000		if (d_is_negative(new_dentry))
5001			goto exit5;
5002
5003		if (!d_is_dir(new_dentry)) {
5004			error = -ENOTDIR;
5005			if (new_last.name[new_last.len])
5006				goto exit5;
5007		}
5008	}
5009	/* unless the source is a directory trailing slashes give -ENOTDIR */
5010	if (!d_is_dir(old_dentry)) {
5011		error = -ENOTDIR;
5012		if (old_last.name[old_last.len])
5013			goto exit5;
5014		if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
5015			goto exit5;
5016	}
5017	/* source should not be ancestor of target */
5018	error = -EINVAL;
5019	if (old_dentry == trap)
5020		goto exit5;
5021	/* target should not be an ancestor of source */
5022	if (!(flags & RENAME_EXCHANGE))
5023		error = -ENOTEMPTY;
5024	if (new_dentry == trap)
5025		goto exit5;
5026
5027	error = security_path_rename(&old_path, old_dentry,
5028				     &new_path, new_dentry, flags);
5029	if (error)
5030		goto exit5;
5031
5032	rd.old_dir	   = old_path.dentry->d_inode;
5033	rd.old_dentry	   = old_dentry;
5034	rd.old_mnt_idmap   = mnt_idmap(old_path.mnt);
5035	rd.new_dir	   = new_path.dentry->d_inode;
5036	rd.new_dentry	   = new_dentry;
5037	rd.new_mnt_idmap   = mnt_idmap(new_path.mnt);
5038	rd.delegated_inode = &delegated_inode;
5039	rd.flags	   = flags;
5040	error = vfs_rename(&rd);
5041exit5:
5042	dput(new_dentry);
5043exit4:
5044	dput(old_dentry);
5045exit3:
5046	unlock_rename(new_path.dentry, old_path.dentry);
5047exit_lock_rename:
5048	if (delegated_inode) {
5049		error = break_deleg_wait(&delegated_inode);
5050		if (!error)
5051			goto retry_deleg;
5052	}
5053	mnt_drop_write(old_path.mnt);
5054exit2:
5055	if (retry_estale(error, lookup_flags))
5056		should_retry = true;
5057	path_put(&new_path);
5058exit1:
5059	path_put(&old_path);
5060	if (should_retry) {
5061		should_retry = false;
5062		lookup_flags |= LOOKUP_REVAL;
5063		goto retry;
5064	}
5065put_names:
5066	putname(from);
5067	putname(to);
5068	return error;
5069}
5070
5071SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
5072		int, newdfd, const char __user *, newname, unsigned int, flags)
5073{
5074	return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname),
5075				flags);
5076}
5077
5078SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
5079		int, newdfd, const char __user *, newname)
5080{
5081	return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname),
5082				0);
5083}
5084
5085SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
5086{
5087	return do_renameat2(AT_FDCWD, getname(oldname), AT_FDCWD,
5088				getname(newname), 0);
5089}
5090
5091int readlink_copy(char __user *buffer, int buflen, const char *link)
5092{
5093	int len = PTR_ERR(link);
5094	if (IS_ERR(link))
5095		goto out;
5096
5097	len = strlen(link);
5098	if (len > (unsigned) buflen)
5099		len = buflen;
5100	if (copy_to_user(buffer, link, len))
5101		len = -EFAULT;
5102out:
5103	return len;
5104}
5105
5106/**
5107 * vfs_readlink - copy symlink body into userspace buffer
5108 * @dentry: dentry on which to get symbolic link
5109 * @buffer: user memory pointer
5110 * @buflen: size of buffer
5111 *
5112 * Does not touch atime.  That's up to the caller if necessary
5113 *
5114 * Does not call security hook.
5115 */
5116int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen)
5117{
5118	struct inode *inode = d_inode(dentry);
5119	DEFINE_DELAYED_CALL(done);
5120	const char *link;
5121	int res;
5122
5123	if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) {
5124		if (unlikely(inode->i_op->readlink))
5125			return inode->i_op->readlink(dentry, buffer, buflen);
5126
5127		if (!d_is_symlink(dentry))
5128			return -EINVAL;
5129
5130		spin_lock(&inode->i_lock);
5131		inode->i_opflags |= IOP_DEFAULT_READLINK;
5132		spin_unlock(&inode->i_lock);
5133	}
5134
5135	link = READ_ONCE(inode->i_link);
5136	if (!link) {
5137		link = inode->i_op->get_link(dentry, inode, &done);
5138		if (IS_ERR(link))
5139			return PTR_ERR(link);
5140	}
5141	res = readlink_copy(buffer, buflen, link);
5142	do_delayed_call(&done);
5143	return res;
5144}
5145EXPORT_SYMBOL(vfs_readlink);
5146
5147/**
5148 * vfs_get_link - get symlink body
5149 * @dentry: dentry on which to get symbolic link
5150 * @done: caller needs to free returned data with this
5151 *
5152 * Calls security hook and i_op->get_link() on the supplied inode.
5153 *
5154 * It does not touch atime.  That's up to the caller if necessary.
5155 *
5156 * Does not work on "special" symlinks like /proc/$$/fd/N
5157 */
5158const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done)
5159{
5160	const char *res = ERR_PTR(-EINVAL);
5161	struct inode *inode = d_inode(dentry);
5162
5163	if (d_is_symlink(dentry)) {
5164		res = ERR_PTR(security_inode_readlink(dentry));
5165		if (!res)
5166			res = inode->i_op->get_link(dentry, inode, done);
5167	}
5168	return res;
5169}
5170EXPORT_SYMBOL(vfs_get_link);
5171
5172/* get the link contents into pagecache */
5173const char *page_get_link(struct dentry *dentry, struct inode *inode,
5174			  struct delayed_call *callback)
5175{
5176	char *kaddr;
5177	struct page *page;
5178	struct address_space *mapping = inode->i_mapping;
5179
5180	if (!dentry) {
5181		page = find_get_page(mapping, 0);
5182		if (!page)
5183			return ERR_PTR(-ECHILD);
5184		if (!PageUptodate(page)) {
5185			put_page(page);
5186			return ERR_PTR(-ECHILD);
5187		}
5188	} else {
5189		page = read_mapping_page(mapping, 0, NULL);
5190		if (IS_ERR(page))
5191			return (char*)page;
5192	}
5193	set_delayed_call(callback, page_put_link, page);
5194	BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
5195	kaddr = page_address(page);
5196	nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
5197	return kaddr;
5198}
5199
5200EXPORT_SYMBOL(page_get_link);
5201
5202void page_put_link(void *arg)
5203{
5204	put_page(arg);
5205}
5206EXPORT_SYMBOL(page_put_link);
5207
5208int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
5209{
5210	DEFINE_DELAYED_CALL(done);
5211	int res = readlink_copy(buffer, buflen,
5212				page_get_link(dentry, d_inode(dentry),
5213					      &done));
5214	do_delayed_call(&done);
5215	return res;
5216}
5217EXPORT_SYMBOL(page_readlink);
5218
5219int page_symlink(struct inode *inode, const char *symname, int len)
5220{
5221	struct address_space *mapping = inode->i_mapping;
5222	const struct address_space_operations *aops = mapping->a_ops;
5223	bool nofs = !mapping_gfp_constraint(mapping, __GFP_FS);
5224	struct page *page;
5225	void *fsdata = NULL;
5226	int err;
5227	unsigned int flags;
5228
5229retry:
5230	if (nofs)
5231		flags = memalloc_nofs_save();
5232	err = aops->write_begin(NULL, mapping, 0, len-1, &page, &fsdata);
5233	if (nofs)
5234		memalloc_nofs_restore(flags);
5235	if (err)
5236		goto fail;
5237
5238	memcpy(page_address(page), symname, len-1);
5239
5240	err = aops->write_end(NULL, mapping, 0, len-1, len-1,
5241							page, fsdata);
5242	if (err < 0)
5243		goto fail;
5244	if (err < len-1)
5245		goto retry;
5246
5247	mark_inode_dirty(inode);
5248	return 0;
5249fail:
5250	return err;
5251}
5252EXPORT_SYMBOL(page_symlink);
5253
5254const struct inode_operations page_symlink_inode_operations = {
5255	.get_link	= page_get_link,
5256};
5257EXPORT_SYMBOL(page_symlink_inode_operations);
v6.2
   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);