linux/fs/namei.c
<<
>>
Prefs
   1/*
   2 *  linux/fs/namei.c
   3 *
   4 *  Copyright (C) 1991, 1992  Linus Torvalds
   5 */
   6
   7/*
   8 * Some corrections by tytso.
   9 */
  10
  11/* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
  12 * lookup logic.
  13 */
  14/* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
  15 */
  16
  17#include <linux/init.h>
  18#include <linux/export.h>
  19#include <linux/kernel.h>
  20#include <linux/slab.h>
  21#include <linux/fs.h>
  22#include <linux/namei.h>
  23#include <linux/pagemap.h>
  24#include <linux/fsnotify.h>
  25#include <linux/personality.h>
  26#include <linux/security.h>
  27#include <linux/ima.h>
  28#include <linux/syscalls.h>
  29#include <linux/mount.h>
  30#include <linux/audit.h>
  31#include <linux/capability.h>
  32#include <linux/file.h>
  33#include <linux/fcntl.h>
  34#include <linux/device_cgroup.h>
  35#include <linux/fs_struct.h>
  36#include <linux/posix_acl.h>
  37#include <asm/uaccess.h>
  38
  39#include "internal.h"
  40#include "mount.h"
  41
  42/* [Feb-1997 T. Schoebel-Theuer]
  43 * Fundamental changes in the pathname lookup mechanisms (namei)
  44 * were necessary because of omirr.  The reason is that omirr needs
  45 * to know the _real_ pathname, not the user-supplied one, in case
  46 * of symlinks (and also when transname replacements occur).
  47 *
  48 * The new code replaces the old recursive symlink resolution with
  49 * an iterative one (in case of non-nested symlink chains).  It does
  50 * this with calls to <fs>_follow_link().
  51 * As a side effect, dir_namei(), _namei() and follow_link() are now 
  52 * replaced with a single function lookup_dentry() that can handle all 
  53 * the special cases of the former code.
  54 *
  55 * With the new dcache, the pathname is stored at each inode, at least as
  56 * long as the refcount of the inode is positive.  As a side effect, the
  57 * size of the dcache depends on the inode cache and thus is dynamic.
  58 *
  59 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
  60 * resolution to correspond with current state of the code.
  61 *
  62 * Note that the symlink resolution is not *completely* iterative.
  63 * There is still a significant amount of tail- and mid- recursion in
  64 * the algorithm.  Also, note that <fs>_readlink() is not used in
  65 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
  66 * may return different results than <fs>_follow_link().  Many virtual
  67 * filesystems (including /proc) exhibit this behavior.
  68 */
  69
  70/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
  71 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
  72 * and the name already exists in form of a symlink, try to create the new
  73 * name indicated by the symlink. The old code always complained that the
  74 * name already exists, due to not following the symlink even if its target
  75 * is nonexistent.  The new semantics affects also mknod() and link() when
  76 * the name is a symlink pointing to a non-existent name.
  77 *
  78 * I don't know which semantics is the right one, since I have no access
  79 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
  80 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
  81 * "old" one. Personally, I think the new semantics is much more logical.
  82 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
  83 * file does succeed in both HP-UX and SunOs, but not in Solaris
  84 * and in the old Linux semantics.
  85 */
  86
  87/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
  88 * semantics.  See the comments in "open_namei" and "do_link" below.
  89 *
  90 * [10-Sep-98 Alan Modra] Another symlink change.
  91 */
  92
  93/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
  94 *      inside the path - always follow.
  95 *      in the last component in creation/removal/renaming - never follow.
  96 *      if LOOKUP_FOLLOW passed - follow.
  97 *      if the pathname has trailing slashes - follow.
  98 *      otherwise - don't follow.
  99 * (applied in that order).
 100 *
 101 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
 102 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
 103 * During the 2.4 we need to fix the userland stuff depending on it -
 104 * hopefully we will be able to get rid of that wart in 2.5. So far only
 105 * XEmacs seems to be relying on it...
 106 */
 107/*
 108 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
 109 * implemented.  Let's see if raised priority of ->s_vfs_rename_mutex gives
 110 * any extra contention...
 111 */
 112
 113/* In order to reduce some races, while at the same time doing additional
 114 * checking and hopefully speeding things up, we copy filenames to the
 115 * kernel data space before using them..
 116 *
 117 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
 118 * PATH_MAX includes the nul terminator --RR.
 119 */
 120void final_putname(struct filename *name)
 121{
 122        if (name->separate) {
 123                __putname(name->name);
 124                kfree(name);
 125        } else {
 126                __putname(name);
 127        }
 128}
 129
 130#define EMBEDDED_NAME_MAX       (PATH_MAX - sizeof(struct filename))
 131
 132static struct filename *
 133getname_flags(const char __user *filename, int flags, int *empty)
 134{
 135        struct filename *result, *err;
 136        int len;
 137        long max;
 138        char *kname;
 139
 140        result = audit_reusename(filename);
 141        if (result)
 142                return result;
 143
 144        result = __getname();
 145        if (unlikely(!result))
 146                return ERR_PTR(-ENOMEM);
 147
 148        /*
 149         * First, try to embed the struct filename inside the names_cache
 150         * allocation
 151         */
 152        kname = (char *)result + sizeof(*result);
 153        result->name = kname;
 154        result->separate = false;
 155        max = EMBEDDED_NAME_MAX;
 156
 157recopy:
 158        len = strncpy_from_user(kname, filename, max);
 159        if (unlikely(len < 0)) {
 160                err = ERR_PTR(len);
 161                goto error;
 162        }
 163
 164        /*
 165         * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
 166         * separate struct filename so we can dedicate the entire
 167         * names_cache allocation for the pathname, and re-do the copy from
 168         * userland.
 169         */
 170        if (len == EMBEDDED_NAME_MAX && max == EMBEDDED_NAME_MAX) {
 171                kname = (char *)result;
 172
 173                result = kzalloc(sizeof(*result), GFP_KERNEL);
 174                if (!result) {
 175                        err = ERR_PTR(-ENOMEM);
 176                        result = (struct filename *)kname;
 177                        goto error;
 178                }
 179                result->name = kname;
 180                result->separate = true;
 181                max = PATH_MAX;
 182                goto recopy;
 183        }
 184
 185        /* The empty path is special. */
 186        if (unlikely(!len)) {
 187                if (empty)
 188                        *empty = 1;
 189                err = ERR_PTR(-ENOENT);
 190                if (!(flags & LOOKUP_EMPTY))
 191                        goto error;
 192        }
 193
 194        err = ERR_PTR(-ENAMETOOLONG);
 195        if (unlikely(len >= PATH_MAX))
 196                goto error;
 197
 198        result->uptr = filename;
 199        audit_getname(result);
 200        return result;
 201
 202error:
 203        final_putname(result);
 204        return err;
 205}
 206
 207struct filename *
 208getname(const char __user * filename)
 209{
 210        return getname_flags(filename, 0, NULL);
 211}
 212EXPORT_SYMBOL(getname);
 213
 214#ifdef CONFIG_AUDITSYSCALL
 215void putname(struct filename *name)
 216{
 217        if (unlikely(!audit_dummy_context()))
 218                return audit_putname(name);
 219        final_putname(name);
 220}
 221#endif
 222
 223static int check_acl(struct inode *inode, int mask)
 224{
 225#ifdef CONFIG_FS_POSIX_ACL
 226        struct posix_acl *acl;
 227
 228        if (mask & MAY_NOT_BLOCK) {
 229                acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
 230                if (!acl)
 231                        return -EAGAIN;
 232                /* no ->get_acl() calls in RCU mode... */
 233                if (acl == ACL_NOT_CACHED)
 234                        return -ECHILD;
 235                return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
 236        }
 237
 238        acl = get_cached_acl(inode, ACL_TYPE_ACCESS);
 239
 240        /*
 241         * A filesystem can force a ACL callback by just never filling the
 242         * ACL cache. But normally you'd fill the cache either at inode
 243         * instantiation time, or on the first ->get_acl call.
 244         *
 245         * If the filesystem doesn't have a get_acl() function at all, we'll
 246         * just create the negative cache entry.
 247         */
 248        if (acl == ACL_NOT_CACHED) {
 249                if (inode->i_op->get_acl) {
 250                        acl = inode->i_op->get_acl(inode, ACL_TYPE_ACCESS);
 251                        if (IS_ERR(acl))
 252                                return PTR_ERR(acl);
 253                } else {
 254                        set_cached_acl(inode, ACL_TYPE_ACCESS, NULL);
 255                        return -EAGAIN;
 256                }
 257        }
 258
 259        if (acl) {
 260                int error = posix_acl_permission(inode, acl, mask);
 261                posix_acl_release(acl);
 262                return error;
 263        }
 264#endif
 265
 266        return -EAGAIN;
 267}
 268
 269/*
 270 * This does the basic permission checking
 271 */
 272static int acl_permission_check(struct inode *inode, int mask)
 273{
 274        unsigned int mode = inode->i_mode;
 275
 276        if (likely(uid_eq(current_fsuid(), inode->i_uid)))
 277                mode >>= 6;
 278        else {
 279                if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
 280                        int error = check_acl(inode, mask);
 281                        if (error != -EAGAIN)
 282                                return error;
 283                }
 284
 285                if (in_group_p(inode->i_gid))
 286                        mode >>= 3;
 287        }
 288
 289        /*
 290         * If the DACs are ok we don't need any capability check.
 291         */
 292        if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
 293                return 0;
 294        return -EACCES;
 295}
 296
 297/**
 298 * generic_permission -  check for access rights on a Posix-like filesystem
 299 * @inode:      inode to check access rights for
 300 * @mask:       right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
 301 *
 302 * Used to check for read/write/execute permissions on a file.
 303 * We use "fsuid" for this, letting us set arbitrary permissions
 304 * for filesystem access without changing the "normal" uids which
 305 * are used for other things.
 306 *
 307 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
 308 * request cannot be satisfied (eg. requires blocking or too much complexity).
 309 * It would then be called again in ref-walk mode.
 310 */
 311int generic_permission(struct inode *inode, int mask)
 312{
 313        int ret;
 314
 315        /*
 316         * Do the basic permission checks.
 317         */
 318        ret = acl_permission_check(inode, mask);
 319        if (ret != -EACCES)
 320                return ret;
 321
 322        if (S_ISDIR(inode->i_mode)) {
 323                /* DACs are overridable for directories */
 324                if (inode_capable(inode, CAP_DAC_OVERRIDE))
 325                        return 0;
 326                if (!(mask & MAY_WRITE))
 327                        if (inode_capable(inode, CAP_DAC_READ_SEARCH))
 328                                return 0;
 329                return -EACCES;
 330        }
 331        /*
 332         * Read/write DACs are always overridable.
 333         * Executable DACs are overridable when there is
 334         * at least one exec bit set.
 335         */
 336        if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
 337                if (inode_capable(inode, CAP_DAC_OVERRIDE))
 338                        return 0;
 339
 340        /*
 341         * Searching includes executable on directories, else just read.
 342         */
 343        mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
 344        if (mask == MAY_READ)
 345                if (inode_capable(inode, CAP_DAC_READ_SEARCH))
 346                        return 0;
 347
 348        return -EACCES;
 349}
 350
 351/*
 352 * We _really_ want to just do "generic_permission()" without
 353 * even looking at the inode->i_op values. So we keep a cache
 354 * flag in inode->i_opflags, that says "this has not special
 355 * permission function, use the fast case".
 356 */
 357static inline int do_inode_permission(struct inode *inode, int mask)
 358{
 359        if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
 360                if (likely(inode->i_op->permission))
 361                        return inode->i_op->permission(inode, mask);
 362
 363                /* This gets set once for the inode lifetime */
 364                spin_lock(&inode->i_lock);
 365                inode->i_opflags |= IOP_FASTPERM;
 366                spin_unlock(&inode->i_lock);
 367        }
 368        return generic_permission(inode, mask);
 369}
 370
 371/**
 372 * __inode_permission - Check for access rights to a given inode
 373 * @inode: Inode to check permission on
 374 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
 375 *
 376 * Check for read/write/execute permissions on an inode.
 377 *
 378 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
 379 *
 380 * This does not check for a read-only file system.  You probably want
 381 * inode_permission().
 382 */
 383int __inode_permission(struct inode *inode, int mask)
 384{
 385        int retval;
 386
 387        if (unlikely(mask & MAY_WRITE)) {
 388                /*
 389                 * Nobody gets write access to an immutable file.
 390                 */
 391                if (IS_IMMUTABLE(inode))
 392                        return -EACCES;
 393        }
 394
 395        retval = do_inode_permission(inode, mask);
 396        if (retval)
 397                return retval;
 398
 399        retval = devcgroup_inode_permission(inode, mask);
 400        if (retval)
 401                return retval;
 402
 403        return security_inode_permission(inode, mask);
 404}
 405
 406/**
 407 * sb_permission - Check superblock-level permissions
 408 * @sb: Superblock of inode to check permission on
 409 * @inode: Inode to check permission on
 410 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
 411 *
 412 * Separate out file-system wide checks from inode-specific permission checks.
 413 */
 414static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
 415{
 416        if (unlikely(mask & MAY_WRITE)) {
 417                umode_t mode = inode->i_mode;
 418
 419                /* Nobody gets write access to a read-only fs. */
 420                if ((sb->s_flags & MS_RDONLY) &&
 421                    (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
 422                        return -EROFS;
 423        }
 424        return 0;
 425}
 426
 427/**
 428 * inode_permission - Check for access rights to a given inode
 429 * @inode: Inode to check permission on
 430 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
 431 *
 432 * Check for read/write/execute permissions on an inode.  We use fs[ug]id for
 433 * this, letting us set arbitrary permissions for filesystem access without
 434 * changing the "normal" UIDs which are used for other things.
 435 *
 436 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
 437 */
 438int inode_permission(struct inode *inode, int mask)
 439{
 440        int retval;
 441
 442        retval = sb_permission(inode->i_sb, inode, mask);
 443        if (retval)
 444                return retval;
 445        return __inode_permission(inode, mask);
 446}
 447
 448/**
 449 * path_get - get a reference to a path
 450 * @path: path to get the reference to
 451 *
 452 * Given a path increment the reference count to the dentry and the vfsmount.
 453 */
 454void path_get(const struct path *path)
 455{
 456        mntget(path->mnt);
 457        dget(path->dentry);
 458}
 459EXPORT_SYMBOL(path_get);
 460
 461/**
 462 * path_put - put a reference to a path
 463 * @path: path to put the reference to
 464 *
 465 * Given a path decrement the reference count to the dentry and the vfsmount.
 466 */
 467void path_put(const struct path *path)
 468{
 469        dput(path->dentry);
 470        mntput(path->mnt);
 471}
 472EXPORT_SYMBOL(path_put);
 473
 474/*
 475 * Path walking has 2 modes, rcu-walk and ref-walk (see
 476 * Documentation/filesystems/path-lookup.txt).  In situations when we can't
 477 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
 478 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
 479 * mode.  Refcounts are grabbed at the last known good point before rcu-walk
 480 * got stuck, so ref-walk may continue from there. If this is not successful
 481 * (eg. a seqcount has changed), then failure is returned and it's up to caller
 482 * to restart the path walk from the beginning in ref-walk mode.
 483 */
 484
 485static inline void lock_rcu_walk(void)
 486{
 487        br_read_lock(&vfsmount_lock);
 488        rcu_read_lock();
 489}
 490
 491static inline void unlock_rcu_walk(void)
 492{
 493        rcu_read_unlock();
 494        br_read_unlock(&vfsmount_lock);
 495}
 496
 497/**
 498 * unlazy_walk - try to switch to ref-walk mode.
 499 * @nd: nameidata pathwalk data
 500 * @dentry: child of nd->path.dentry or NULL
 501 * Returns: 0 on success, -ECHILD on failure
 502 *
 503 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
 504 * for ref-walk mode.  @dentry must be a path found by a do_lookup call on
 505 * @nd or NULL.  Must be called from rcu-walk context.
 506 */
 507static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
 508{
 509        struct fs_struct *fs = current->fs;
 510        struct dentry *parent = nd->path.dentry;
 511        int want_root = 0;
 512
 513        BUG_ON(!(nd->flags & LOOKUP_RCU));
 514        if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
 515                want_root = 1;
 516                spin_lock(&fs->lock);
 517                if (nd->root.mnt != fs->root.mnt ||
 518                                nd->root.dentry != fs->root.dentry)
 519                        goto err_root;
 520        }
 521        spin_lock(&parent->d_lock);
 522        if (!dentry) {
 523                if (!__d_rcu_to_refcount(parent, nd->seq))
 524                        goto err_parent;
 525                BUG_ON(nd->inode != parent->d_inode);
 526        } else {
 527                if (dentry->d_parent != parent)
 528                        goto err_parent;
 529                spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
 530                if (!__d_rcu_to_refcount(dentry, nd->seq))
 531                        goto err_child;
 532                /*
 533                 * If the sequence check on the child dentry passed, then
 534                 * the child has not been removed from its parent. This
 535                 * means the parent dentry must be valid and able to take
 536                 * a reference at this point.
 537                 */
 538                BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent);
 539                BUG_ON(!parent->d_lockref.count);
 540                parent->d_lockref.count++;
 541                spin_unlock(&dentry->d_lock);
 542        }
 543        spin_unlock(&parent->d_lock);
 544        if (want_root) {
 545                path_get(&nd->root);
 546                spin_unlock(&fs->lock);
 547        }
 548        mntget(nd->path.mnt);
 549
 550        unlock_rcu_walk();
 551        nd->flags &= ~LOOKUP_RCU;
 552        return 0;
 553
 554err_child:
 555        spin_unlock(&dentry->d_lock);
 556err_parent:
 557        spin_unlock(&parent->d_lock);
 558err_root:
 559        if (want_root)
 560                spin_unlock(&fs->lock);
 561        return -ECHILD;
 562}
 563
 564static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
 565{
 566        return dentry->d_op->d_revalidate(dentry, flags);
 567}
 568
 569/**
 570 * complete_walk - successful completion of path walk
 571 * @nd:  pointer nameidata
 572 *
 573 * If we had been in RCU mode, drop out of it and legitimize nd->path.
 574 * Revalidate the final result, unless we'd already done that during
 575 * the path walk or the filesystem doesn't ask for it.  Return 0 on
 576 * success, -error on failure.  In case of failure caller does not
 577 * need to drop nd->path.
 578 */
 579static int complete_walk(struct nameidata *nd)
 580{
 581        struct dentry *dentry = nd->path.dentry;
 582        int status;
 583
 584        if (nd->flags & LOOKUP_RCU) {
 585                nd->flags &= ~LOOKUP_RCU;
 586                if (!(nd->flags & LOOKUP_ROOT))
 587                        nd->root.mnt = NULL;
 588                spin_lock(&dentry->d_lock);
 589                if (unlikely(!__d_rcu_to_refcount(dentry, nd->seq))) {
 590                        spin_unlock(&dentry->d_lock);
 591                        unlock_rcu_walk();
 592                        return -ECHILD;
 593                }
 594                BUG_ON(nd->inode != dentry->d_inode);
 595                spin_unlock(&dentry->d_lock);
 596                mntget(nd->path.mnt);
 597                unlock_rcu_walk();
 598        }
 599
 600        if (likely(!(nd->flags & LOOKUP_JUMPED)))
 601                return 0;
 602
 603        if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
 604                return 0;
 605
 606        status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
 607        if (status > 0)
 608                return 0;
 609
 610        if (!status)
 611                status = -ESTALE;
 612
 613        path_put(&nd->path);
 614        return status;
 615}
 616
 617static __always_inline void set_root(struct nameidata *nd)
 618{
 619        if (!nd->root.mnt)
 620                get_fs_root(current->fs, &nd->root);
 621}
 622
 623static int link_path_walk(const char *, struct nameidata *);
 624
 625static __always_inline void set_root_rcu(struct nameidata *nd)
 626{
 627        if (!nd->root.mnt) {
 628                struct fs_struct *fs = current->fs;
 629                unsigned seq;
 630
 631                do {
 632                        seq = read_seqcount_begin(&fs->seq);
 633                        nd->root = fs->root;
 634                        nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
 635                } while (read_seqcount_retry(&fs->seq, seq));
 636        }
 637}
 638
 639static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
 640{
 641        int ret;
 642
 643        if (IS_ERR(link))
 644                goto fail;
 645
 646        if (*link == '/') {
 647                set_root(nd);
 648                path_put(&nd->path);
 649                nd->path = nd->root;
 650                path_get(&nd->root);
 651                nd->flags |= LOOKUP_JUMPED;
 652        }
 653        nd->inode = nd->path.dentry->d_inode;
 654
 655        ret = link_path_walk(link, nd);
 656        return ret;
 657fail:
 658        path_put(&nd->path);
 659        return PTR_ERR(link);
 660}
 661
 662static void path_put_conditional(struct path *path, struct nameidata *nd)
 663{
 664        dput(path->dentry);
 665        if (path->mnt != nd->path.mnt)
 666                mntput(path->mnt);
 667}
 668
 669static inline void path_to_nameidata(const struct path *path,
 670                                        struct nameidata *nd)
 671{
 672        if (!(nd->flags & LOOKUP_RCU)) {
 673                dput(nd->path.dentry);
 674                if (nd->path.mnt != path->mnt)
 675                        mntput(nd->path.mnt);
 676        }
 677        nd->path.mnt = path->mnt;
 678        nd->path.dentry = path->dentry;
 679}
 680
 681/*
 682 * Helper to directly jump to a known parsed path from ->follow_link,
 683 * caller must have taken a reference to path beforehand.
 684 */
 685void nd_jump_link(struct nameidata *nd, struct path *path)
 686{
 687        path_put(&nd->path);
 688
 689        nd->path = *path;
 690        nd->inode = nd->path.dentry->d_inode;
 691        nd->flags |= LOOKUP_JUMPED;
 692}
 693
 694static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
 695{
 696        struct inode *inode = link->dentry->d_inode;
 697        if (inode->i_op->put_link)
 698                inode->i_op->put_link(link->dentry, nd, cookie);
 699        path_put(link);
 700}
 701
 702int sysctl_protected_symlinks __read_mostly = 0;
 703int sysctl_protected_hardlinks __read_mostly = 0;
 704
 705/**
 706 * may_follow_link - Check symlink following for unsafe situations
 707 * @link: The path of the symlink
 708 * @nd: nameidata pathwalk data
 709 *
 710 * In the case of the sysctl_protected_symlinks sysctl being enabled,
 711 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
 712 * in a sticky world-writable directory. This is to protect privileged
 713 * processes from failing races against path names that may change out
 714 * from under them by way of other users creating malicious symlinks.
 715 * It will permit symlinks to be followed only when outside a sticky
 716 * world-writable directory, or when the uid of the symlink and follower
 717 * match, or when the directory owner matches the symlink's owner.
 718 *
 719 * Returns 0 if following the symlink is allowed, -ve on error.
 720 */
 721static inline int may_follow_link(struct path *link, struct nameidata *nd)
 722{
 723        const struct inode *inode;
 724        const struct inode *parent;
 725
 726        if (!sysctl_protected_symlinks)
 727                return 0;
 728
 729        /* Allowed if owner and follower match. */
 730        inode = link->dentry->d_inode;
 731        if (uid_eq(current_cred()->fsuid, inode->i_uid))
 732                return 0;
 733
 734        /* Allowed if parent directory not sticky and world-writable. */
 735        parent = nd->path.dentry->d_inode;
 736        if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
 737                return 0;
 738
 739        /* Allowed if parent directory and link owner match. */
 740        if (uid_eq(parent->i_uid, inode->i_uid))
 741                return 0;
 742
 743        audit_log_link_denied("follow_link", link);
 744        path_put_conditional(link, nd);
 745        path_put(&nd->path);
 746        return -EACCES;
 747}
 748
 749/**
 750 * safe_hardlink_source - Check for safe hardlink conditions
 751 * @inode: the source inode to hardlink from
 752 *
 753 * Return false if at least one of the following conditions:
 754 *    - inode is not a regular file
 755 *    - inode is setuid
 756 *    - inode is setgid and group-exec
 757 *    - access failure for read and write
 758 *
 759 * Otherwise returns true.
 760 */
 761static bool safe_hardlink_source(struct inode *inode)
 762{
 763        umode_t mode = inode->i_mode;
 764
 765        /* Special files should not get pinned to the filesystem. */
 766        if (!S_ISREG(mode))
 767                return false;
 768
 769        /* Setuid files should not get pinned to the filesystem. */
 770        if (mode & S_ISUID)
 771                return false;
 772
 773        /* Executable setgid files should not get pinned to the filesystem. */
 774        if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
 775                return false;
 776
 777        /* Hardlinking to unreadable or unwritable sources is dangerous. */
 778        if (inode_permission(inode, MAY_READ | MAY_WRITE))
 779                return false;
 780
 781        return true;
 782}
 783
 784/**
 785 * may_linkat - Check permissions for creating a hardlink
 786 * @link: the source to hardlink from
 787 *
 788 * Block hardlink when all of:
 789 *  - sysctl_protected_hardlinks enabled
 790 *  - fsuid does not match inode
 791 *  - hardlink source is unsafe (see safe_hardlink_source() above)
 792 *  - not CAP_FOWNER
 793 *
 794 * Returns 0 if successful, -ve on error.
 795 */
 796static int may_linkat(struct path *link)
 797{
 798        const struct cred *cred;
 799        struct inode *inode;
 800
 801        if (!sysctl_protected_hardlinks)
 802                return 0;
 803
 804        cred = current_cred();
 805        inode = link->dentry->d_inode;
 806
 807        /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
 808         * otherwise, it must be a safe source.
 809         */
 810        if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) ||
 811            capable(CAP_FOWNER))
 812                return 0;
 813
 814        audit_log_link_denied("linkat", link);
 815        return -EPERM;
 816}
 817
 818static __always_inline int
 819follow_link(struct path *link, struct nameidata *nd, void **p)
 820{
 821        struct dentry *dentry = link->dentry;
 822        int error;
 823        char *s;
 824
 825        BUG_ON(nd->flags & LOOKUP_RCU);
 826
 827        if (link->mnt == nd->path.mnt)
 828                mntget(link->mnt);
 829
 830        error = -ELOOP;
 831        if (unlikely(current->total_link_count >= 40))
 832                goto out_put_nd_path;
 833
 834        cond_resched();
 835        current->total_link_count++;
 836
 837        touch_atime(link);
 838        nd_set_link(nd, NULL);
 839
 840        error = security_inode_follow_link(link->dentry, nd);
 841        if (error)
 842                goto out_put_nd_path;
 843
 844        nd->last_type = LAST_BIND;
 845        *p = dentry->d_inode->i_op->follow_link(dentry, nd);
 846        error = PTR_ERR(*p);
 847        if (IS_ERR(*p))
 848                goto out_put_nd_path;
 849
 850        error = 0;
 851        s = nd_get_link(nd);
 852        if (s) {
 853                error = __vfs_follow_link(nd, s);
 854                if (unlikely(error))
 855                        put_link(nd, link, *p);
 856        }
 857
 858        return error;
 859
 860out_put_nd_path:
 861        *p = NULL;
 862        path_put(&nd->path);
 863        path_put(link);
 864        return error;
 865}
 866
 867static int follow_up_rcu(struct path *path)
 868{
 869        struct mount *mnt = real_mount(path->mnt);
 870        struct mount *parent;
 871        struct dentry *mountpoint;
 872
 873        parent = mnt->mnt_parent;
 874        if (&parent->mnt == path->mnt)
 875                return 0;
 876        mountpoint = mnt->mnt_mountpoint;
 877        path->dentry = mountpoint;
 878        path->mnt = &parent->mnt;
 879        return 1;
 880}
 881
 882/*
 883 * follow_up - Find the mountpoint of path's vfsmount
 884 *
 885 * Given a path, find the mountpoint of its source file system.
 886 * Replace @path with the path of the mountpoint in the parent mount.
 887 * Up is towards /.
 888 *
 889 * Return 1 if we went up a level and 0 if we were already at the
 890 * root.
 891 */
 892int follow_up(struct path *path)
 893{
 894        struct mount *mnt = real_mount(path->mnt);
 895        struct mount *parent;
 896        struct dentry *mountpoint;
 897
 898        br_read_lock(&vfsmount_lock);
 899        parent = mnt->mnt_parent;
 900        if (parent == mnt) {
 901                br_read_unlock(&vfsmount_lock);
 902                return 0;
 903        }
 904        mntget(&parent->mnt);
 905        mountpoint = dget(mnt->mnt_mountpoint);
 906        br_read_unlock(&vfsmount_lock);
 907        dput(path->dentry);
 908        path->dentry = mountpoint;
 909        mntput(path->mnt);
 910        path->mnt = &parent->mnt;
 911        return 1;
 912}
 913
 914/*
 915 * Perform an automount
 916 * - return -EISDIR to tell follow_managed() to stop and return the path we
 917 *   were called with.
 918 */
 919static int follow_automount(struct path *path, unsigned flags,
 920                            bool *need_mntput)
 921{
 922        struct vfsmount *mnt;
 923        int err;
 924
 925        if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
 926                return -EREMOTE;
 927
 928        /* We don't want to mount if someone's just doing a stat -
 929         * unless they're stat'ing a directory and appended a '/' to
 930         * the name.
 931         *
 932         * We do, however, want to mount if someone wants to open or
 933         * create a file of any type under the mountpoint, wants to
 934         * traverse through the mountpoint or wants to open the
 935         * mounted directory.  Also, autofs may mark negative dentries
 936         * as being automount points.  These will need the attentions
 937         * of the daemon to instantiate them before they can be used.
 938         */
 939        if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
 940                     LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
 941            path->dentry->d_inode)
 942                return -EISDIR;
 943
 944        current->total_link_count++;
 945        if (current->total_link_count >= 40)
 946                return -ELOOP;
 947
 948        mnt = path->dentry->d_op->d_automount(path);
 949        if (IS_ERR(mnt)) {
 950                /*
 951                 * The filesystem is allowed to return -EISDIR here to indicate
 952                 * it doesn't want to automount.  For instance, autofs would do
 953                 * this so that its userspace daemon can mount on this dentry.
 954                 *
 955                 * However, we can only permit this if it's a terminal point in
 956                 * the path being looked up; if it wasn't then the remainder of
 957                 * the path is inaccessible and we should say so.
 958                 */
 959                if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
 960                        return -EREMOTE;
 961                return PTR_ERR(mnt);
 962        }
 963
 964        if (!mnt) /* mount collision */
 965                return 0;
 966
 967        if (!*need_mntput) {
 968                /* lock_mount() may release path->mnt on error */
 969                mntget(path->mnt);
 970                *need_mntput = true;
 971        }
 972        err = finish_automount(mnt, path);
 973
 974        switch (err) {
 975        case -EBUSY:
 976                /* Someone else made a mount here whilst we were busy */
 977                return 0;
 978        case 0:
 979                path_put(path);
 980                path->mnt = mnt;
 981                path->dentry = dget(mnt->mnt_root);
 982                return 0;
 983        default:
 984                return err;
 985        }
 986
 987}
 988
 989/*
 990 * Handle a dentry that is managed in some way.
 991 * - Flagged for transit management (autofs)
 992 * - Flagged as mountpoint
 993 * - Flagged as automount point
 994 *
 995 * This may only be called in refwalk mode.
 996 *
 997 * Serialization is taken care of in namespace.c
 998 */
 999static int follow_managed(struct path *path, unsigned flags)
1000{
1001        struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1002        unsigned managed;
1003        bool need_mntput = false;
1004        int ret = 0;
1005
1006        /* Given that we're not holding a lock here, we retain the value in a
1007         * local variable for each dentry as we look at it so that we don't see
1008         * the components of that value change under us */
1009        while (managed = ACCESS_ONCE(path->dentry->d_flags),
1010               managed &= DCACHE_MANAGED_DENTRY,
1011               unlikely(managed != 0)) {
1012                /* Allow the filesystem to manage the transit without i_mutex
1013                 * being held. */
1014                if (managed & DCACHE_MANAGE_TRANSIT) {
1015                        BUG_ON(!path->dentry->d_op);
1016                        BUG_ON(!path->dentry->d_op->d_manage);
1017                        ret = path->dentry->d_op->d_manage(path->dentry, false);
1018                        if (ret < 0)
1019                                break;
1020                }
1021
1022                /* Transit to a mounted filesystem. */
1023                if (managed & DCACHE_MOUNTED) {
1024                        struct vfsmount *mounted = lookup_mnt(path);
1025                        if (mounted) {
1026                                dput(path->dentry);
1027                                if (need_mntput)
1028                                        mntput(path->mnt);
1029                                path->mnt = mounted;
1030                                path->dentry = dget(mounted->mnt_root);
1031                                need_mntput = true;
1032                                continue;
1033                        }
1034
1035                        /* Something is mounted on this dentry in another
1036                         * namespace and/or whatever was mounted there in this
1037                         * namespace got unmounted before we managed to get the
1038                         * vfsmount_lock */
1039                }
1040
1041                /* Handle an automount point */
1042                if (managed & DCACHE_NEED_AUTOMOUNT) {
1043                        ret = follow_automount(path, flags, &need_mntput);
1044                        if (ret < 0)
1045                                break;
1046                        continue;
1047                }
1048
1049                /* We didn't change the current path point */
1050                break;
1051        }
1052
1053        if (need_mntput && path->mnt == mnt)
1054                mntput(path->mnt);
1055        if (ret == -EISDIR)
1056                ret = 0;
1057        return ret < 0 ? ret : need_mntput;
1058}
1059
1060int follow_down_one(struct path *path)
1061{
1062        struct vfsmount *mounted;
1063
1064        mounted = lookup_mnt(path);
1065        if (mounted) {
1066                dput(path->dentry);
1067                mntput(path->mnt);
1068                path->mnt = mounted;
1069                path->dentry = dget(mounted->mnt_root);
1070                return 1;
1071        }
1072        return 0;
1073}
1074
1075static inline bool managed_dentry_might_block(struct dentry *dentry)
1076{
1077        return (dentry->d_flags & DCACHE_MANAGE_TRANSIT &&
1078                dentry->d_op->d_manage(dentry, true) < 0);
1079}
1080
1081/*
1082 * Try to skip to top of mountpoint pile in rcuwalk mode.  Fail if
1083 * we meet a managed dentry that would need blocking.
1084 */
1085static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1086                               struct inode **inode)
1087{
1088        for (;;) {
1089                struct mount *mounted;
1090                /*
1091                 * Don't forget we might have a non-mountpoint managed dentry
1092                 * that wants to block transit.
1093                 */
1094                if (unlikely(managed_dentry_might_block(path->dentry)))
1095                        return false;
1096
1097                if (!d_mountpoint(path->dentry))
1098                        break;
1099
1100                mounted = __lookup_mnt(path->mnt, path->dentry, 1);
1101                if (!mounted)
1102                        break;
1103                path->mnt = &mounted->mnt;
1104                path->dentry = mounted->mnt.mnt_root;
1105                nd->flags |= LOOKUP_JUMPED;
1106                nd->seq = read_seqcount_begin(&path->dentry->d_seq);
1107                /*
1108                 * Update the inode too. We don't need to re-check the
1109                 * dentry sequence number here after this d_inode read,
1110                 * because a mount-point is always pinned.
1111                 */
1112                *inode = path->dentry->d_inode;
1113        }
1114        return true;
1115}
1116
1117static void follow_mount_rcu(struct nameidata *nd)
1118{
1119        while (d_mountpoint(nd->path.dentry)) {
1120                struct mount *mounted;
1121                mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1);
1122                if (!mounted)
1123                        break;
1124                nd->path.mnt = &mounted->mnt;
1125                nd->path.dentry = mounted->mnt.mnt_root;
1126                nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1127        }
1128}
1129
1130static int follow_dotdot_rcu(struct nameidata *nd)
1131{
1132        set_root_rcu(nd);
1133
1134        while (1) {
1135                if (nd->path.dentry == nd->root.dentry &&
1136                    nd->path.mnt == nd->root.mnt) {
1137                        break;
1138                }
1139                if (nd->path.dentry != nd->path.mnt->mnt_root) {
1140                        struct dentry *old = nd->path.dentry;
1141                        struct dentry *parent = old->d_parent;
1142                        unsigned seq;
1143
1144                        seq = read_seqcount_begin(&parent->d_seq);
1145                        if (read_seqcount_retry(&old->d_seq, nd->seq))
1146                                goto failed;
1147                        nd->path.dentry = parent;
1148                        nd->seq = seq;
1149                        break;
1150                }
1151                if (!follow_up_rcu(&nd->path))
1152                        break;
1153                nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1154        }
1155        follow_mount_rcu(nd);
1156        nd->inode = nd->path.dentry->d_inode;
1157        return 0;
1158
1159failed:
1160        nd->flags &= ~LOOKUP_RCU;
1161        if (!(nd->flags & LOOKUP_ROOT))
1162                nd->root.mnt = NULL;
1163        unlock_rcu_walk();
1164        return -ECHILD;
1165}
1166
1167/*
1168 * Follow down to the covering mount currently visible to userspace.  At each
1169 * point, the filesystem owning that dentry may be queried as to whether the
1170 * caller is permitted to proceed or not.
1171 */
1172int follow_down(struct path *path)
1173{
1174        unsigned managed;
1175        int ret;
1176
1177        while (managed = ACCESS_ONCE(path->dentry->d_flags),
1178               unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1179                /* Allow the filesystem to manage the transit without i_mutex
1180                 * being held.
1181                 *
1182                 * We indicate to the filesystem if someone is trying to mount
1183                 * something here.  This gives autofs the chance to deny anyone
1184                 * other than its daemon the right to mount on its
1185                 * superstructure.
1186                 *
1187                 * The filesystem may sleep at this point.
1188                 */
1189                if (managed & DCACHE_MANAGE_TRANSIT) {
1190                        BUG_ON(!path->dentry->d_op);
1191                        BUG_ON(!path->dentry->d_op->d_manage);
1192                        ret = path->dentry->d_op->d_manage(
1193                                path->dentry, false);
1194                        if (ret < 0)
1195                                return ret == -EISDIR ? 0 : ret;
1196                }
1197
1198                /* Transit to a mounted filesystem. */
1199                if (managed & DCACHE_MOUNTED) {
1200                        struct vfsmount *mounted = lookup_mnt(path);
1201                        if (!mounted)
1202                                break;
1203                        dput(path->dentry);
1204                        mntput(path->mnt);
1205                        path->mnt = mounted;
1206                        path->dentry = dget(mounted->mnt_root);
1207                        continue;
1208                }
1209
1210                /* Don't handle automount points here */
1211                break;
1212        }
1213        return 0;
1214}
1215
1216/*
1217 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1218 */
1219static void follow_mount(struct path *path)
1220{
1221        while (d_mountpoint(path->dentry)) {
1222                struct vfsmount *mounted = lookup_mnt(path);
1223                if (!mounted)
1224                        break;
1225                dput(path->dentry);
1226                mntput(path->mnt);
1227                path->mnt = mounted;
1228                path->dentry = dget(mounted->mnt_root);
1229        }
1230}
1231
1232static void follow_dotdot(struct nameidata *nd)
1233{
1234        set_root(nd);
1235
1236        while(1) {
1237                struct dentry *old = nd->path.dentry;
1238
1239                if (nd->path.dentry == nd->root.dentry &&
1240                    nd->path.mnt == nd->root.mnt) {
1241                        break;
1242                }
1243                if (nd->path.dentry != nd->path.mnt->mnt_root) {
1244                        /* rare case of legitimate dget_parent()... */
1245                        nd->path.dentry = dget_parent(nd->path.dentry);
1246                        dput(old);
1247                        break;
1248                }
1249                if (!follow_up(&nd->path))
1250                        break;
1251        }
1252        follow_mount(&nd->path);
1253        nd->inode = nd->path.dentry->d_inode;
1254}
1255
1256/*
1257 * This looks up the name in dcache, possibly revalidates the old dentry and
1258 * allocates a new one if not found or not valid.  In the need_lookup argument
1259 * returns whether i_op->lookup is necessary.
1260 *
1261 * dir->d_inode->i_mutex must be held
1262 */
1263static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
1264                                    unsigned int flags, bool *need_lookup)
1265{
1266        struct dentry *dentry;
1267        int error;
1268
1269        *need_lookup = false;
1270        dentry = d_lookup(dir, name);
1271        if (dentry) {
1272                if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1273                        error = d_revalidate(dentry, flags);
1274                        if (unlikely(error <= 0)) {
1275                                if (error < 0) {
1276                                        dput(dentry);
1277                                        return ERR_PTR(error);
1278                                } else if (!d_invalidate(dentry)) {
1279                                        dput(dentry);
1280                                        dentry = NULL;
1281                                }
1282                        }
1283                }
1284        }
1285
1286        if (!dentry) {
1287                dentry = d_alloc(dir, name);
1288                if (unlikely(!dentry))
1289                        return ERR_PTR(-ENOMEM);
1290
1291                *need_lookup = true;
1292        }
1293        return dentry;
1294}
1295
1296/*
1297 * Call i_op->lookup on the dentry.  The dentry must be negative but may be
1298 * hashed if it was pouplated with DCACHE_NEED_LOOKUP.
1299 *
1300 * dir->d_inode->i_mutex must be held
1301 */
1302static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1303                                  unsigned int flags)
1304{
1305        struct dentry *old;
1306
1307        /* Don't create child dentry for a dead directory. */
1308        if (unlikely(IS_DEADDIR(dir))) {
1309                dput(dentry);
1310                return ERR_PTR(-ENOENT);
1311        }
1312
1313        old = dir->i_op->lookup(dir, dentry, flags);
1314        if (unlikely(old)) {
1315                dput(dentry);
1316                dentry = old;
1317        }
1318        return dentry;
1319}
1320
1321static struct dentry *__lookup_hash(struct qstr *name,
1322                struct dentry *base, unsigned int flags)
1323{
1324        bool need_lookup;
1325        struct dentry *dentry;
1326
1327        dentry = lookup_dcache(name, base, flags, &need_lookup);
1328        if (!need_lookup)
1329                return dentry;
1330
1331        return lookup_real(base->d_inode, dentry, flags);
1332}
1333
1334/*
1335 *  It's more convoluted than I'd like it to be, but... it's still fairly
1336 *  small and for now I'd prefer to have fast path as straight as possible.
1337 *  It _is_ time-critical.
1338 */
1339static int lookup_fast(struct nameidata *nd,
1340                       struct path *path, struct inode **inode)
1341{
1342        struct vfsmount *mnt = nd->path.mnt;
1343        struct dentry *dentry, *parent = nd->path.dentry;
1344        int need_reval = 1;
1345        int status = 1;
1346        int err;
1347
1348        /*
1349         * Rename seqlock is not required here because in the off chance
1350         * of a false negative due to a concurrent rename, we're going to
1351         * do the non-racy lookup, below.
1352         */
1353        if (nd->flags & LOOKUP_RCU) {
1354                unsigned seq;
1355                dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1356                if (!dentry)
1357                        goto unlazy;
1358
1359                /*
1360                 * This sequence count validates that the inode matches
1361                 * the dentry name information from lookup.
1362                 */
1363                *inode = dentry->d_inode;
1364                if (read_seqcount_retry(&dentry->d_seq, seq))
1365                        return -ECHILD;
1366
1367                /*
1368                 * This sequence count validates that the parent had no
1369                 * changes while we did the lookup of the dentry above.
1370                 *
1371                 * The memory barrier in read_seqcount_begin of child is
1372                 *  enough, we can use __read_seqcount_retry here.
1373                 */
1374                if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1375                        return -ECHILD;
1376                nd->seq = seq;
1377
1378                if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1379                        status = d_revalidate(dentry, nd->flags);
1380                        if (unlikely(status <= 0)) {
1381                                if (status != -ECHILD)
1382                                        need_reval = 0;
1383                                goto unlazy;
1384                        }
1385                }
1386                path->mnt = mnt;
1387                path->dentry = dentry;
1388                if (unlikely(!__follow_mount_rcu(nd, path, inode)))
1389                        goto unlazy;
1390                if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1391                        goto unlazy;
1392                return 0;
1393unlazy:
1394                if (unlazy_walk(nd, dentry))
1395                        return -ECHILD;
1396        } else {
1397                dentry = __d_lookup(parent, &nd->last);
1398        }
1399
1400        if (unlikely(!dentry))
1401                goto need_lookup;
1402
1403        if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1404                status = d_revalidate(dentry, nd->flags);
1405        if (unlikely(status <= 0)) {
1406                if (status < 0) {
1407                        dput(dentry);
1408                        return status;
1409                }
1410                if (!d_invalidate(dentry)) {
1411                        dput(dentry);
1412                        goto need_lookup;
1413                }
1414        }
1415
1416        path->mnt = mnt;
1417        path->dentry = dentry;
1418        err = follow_managed(path, nd->flags);
1419        if (unlikely(err < 0)) {
1420                path_put_conditional(path, nd);
1421                return err;
1422        }
1423        if (err)
1424                nd->flags |= LOOKUP_JUMPED;
1425        *inode = path->dentry->d_inode;
1426        return 0;
1427
1428need_lookup:
1429        return 1;
1430}
1431
1432/* Fast lookup failed, do it the slow way */
1433static int lookup_slow(struct nameidata *nd, struct path *path)
1434{
1435        struct dentry *dentry, *parent;
1436        int err;
1437
1438        parent = nd->path.dentry;
1439        BUG_ON(nd->inode != parent->d_inode);
1440
1441        mutex_lock(&parent->d_inode->i_mutex);
1442        dentry = __lookup_hash(&nd->last, parent, nd->flags);
1443        mutex_unlock(&parent->d_inode->i_mutex);
1444        if (IS_ERR(dentry))
1445                return PTR_ERR(dentry);
1446        path->mnt = nd->path.mnt;
1447        path->dentry = dentry;
1448        err = follow_managed(path, nd->flags);
1449        if (unlikely(err < 0)) {
1450                path_put_conditional(path, nd);
1451                return err;
1452        }
1453        if (err)
1454                nd->flags |= LOOKUP_JUMPED;
1455        return 0;
1456}
1457
1458static inline int may_lookup(struct nameidata *nd)
1459{
1460        if (nd->flags & LOOKUP_RCU) {
1461                int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1462                if (err != -ECHILD)
1463                        return err;
1464                if (unlazy_walk(nd, NULL))
1465                        return -ECHILD;
1466        }
1467        return inode_permission(nd->inode, MAY_EXEC);
1468}
1469
1470static inline int handle_dots(struct nameidata *nd, int type)
1471{
1472        if (type == LAST_DOTDOT) {
1473                if (nd->flags & LOOKUP_RCU) {
1474                        if (follow_dotdot_rcu(nd))
1475                                return -ECHILD;
1476                } else
1477                        follow_dotdot(nd);
1478        }
1479        return 0;
1480}
1481
1482static void terminate_walk(struct nameidata *nd)
1483{
1484        if (!(nd->flags & LOOKUP_RCU)) {
1485                path_put(&nd->path);
1486        } else {
1487                nd->flags &= ~LOOKUP_RCU;
1488                if (!(nd->flags & LOOKUP_ROOT))
1489                        nd->root.mnt = NULL;
1490                unlock_rcu_walk();
1491        }
1492}
1493
1494/*
1495 * Do we need to follow links? We _really_ want to be able
1496 * to do this check without having to look at inode->i_op,
1497 * so we keep a cache of "no, this doesn't need follow_link"
1498 * for the common case.
1499 */
1500static inline int should_follow_link(struct inode *inode, int follow)
1501{
1502        if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1503                if (likely(inode->i_op->follow_link))
1504                        return follow;
1505
1506                /* This gets set once for the inode lifetime */
1507                spin_lock(&inode->i_lock);
1508                inode->i_opflags |= IOP_NOFOLLOW;
1509                spin_unlock(&inode->i_lock);
1510        }
1511        return 0;
1512}
1513
1514static inline int walk_component(struct nameidata *nd, struct path *path,
1515                int follow)
1516{
1517        struct inode *inode;
1518        int err;
1519        /*
1520         * "." and ".." are special - ".." especially so because it has
1521         * to be able to know about the current root directory and
1522         * parent relationships.
1523         */
1524        if (unlikely(nd->last_type != LAST_NORM))
1525                return handle_dots(nd, nd->last_type);
1526        err = lookup_fast(nd, path, &inode);
1527        if (unlikely(err)) {
1528                if (err < 0)
1529                        goto out_err;
1530
1531                err = lookup_slow(nd, path);
1532                if (err < 0)
1533                        goto out_err;
1534
1535                inode = path->dentry->d_inode;
1536        }
1537        err = -ENOENT;
1538        if (!inode)
1539                goto out_path_put;
1540
1541        if (should_follow_link(inode, follow)) {
1542                if (nd->flags & LOOKUP_RCU) {
1543                        if (unlikely(unlazy_walk(nd, path->dentry))) {
1544                                err = -ECHILD;
1545                                goto out_err;
1546                        }
1547                }
1548                BUG_ON(inode != path->dentry->d_inode);
1549                return 1;
1550        }
1551        path_to_nameidata(path, nd);
1552        nd->inode = inode;
1553        return 0;
1554
1555out_path_put:
1556        path_to_nameidata(path, nd);
1557out_err:
1558        terminate_walk(nd);
1559        return err;
1560}
1561
1562/*
1563 * This limits recursive symlink follows to 8, while
1564 * limiting consecutive symlinks to 40.
1565 *
1566 * Without that kind of total limit, nasty chains of consecutive
1567 * symlinks can cause almost arbitrarily long lookups.
1568 */
1569static inline int nested_symlink(struct path *path, struct nameidata *nd)
1570{
1571        int res;
1572
1573        if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1574                path_put_conditional(path, nd);
1575                path_put(&nd->path);
1576                return -ELOOP;
1577        }
1578        BUG_ON(nd->depth >= MAX_NESTED_LINKS);
1579
1580        nd->depth++;
1581        current->link_count++;
1582
1583        do {
1584                struct path link = *path;
1585                void *cookie;
1586
1587                res = follow_link(&link, nd, &cookie);
1588                if (res)
1589                        break;
1590                res = walk_component(nd, path, LOOKUP_FOLLOW);
1591                put_link(nd, &link, cookie);
1592        } while (res > 0);
1593
1594        current->link_count--;
1595        nd->depth--;
1596        return res;
1597}
1598
1599/*
1600 * We really don't want to look at inode->i_op->lookup
1601 * when we don't have to. So we keep a cache bit in
1602 * the inode ->i_opflags field that says "yes, we can
1603 * do lookup on this inode".
1604 */
1605static inline int can_lookup(struct inode *inode)
1606{
1607        if (likely(inode->i_opflags & IOP_LOOKUP))
1608                return 1;
1609        if (likely(!inode->i_op->lookup))
1610                return 0;
1611
1612        /* We do this once for the lifetime of the inode */
1613        spin_lock(&inode->i_lock);
1614        inode->i_opflags |= IOP_LOOKUP;
1615        spin_unlock(&inode->i_lock);
1616        return 1;
1617}
1618
1619/*
1620 * We can do the critical dentry name comparison and hashing
1621 * operations one word at a time, but we are limited to:
1622 *
1623 * - Architectures with fast unaligned word accesses. We could
1624 *   do a "get_unaligned()" if this helps and is sufficiently
1625 *   fast.
1626 *
1627 * - Little-endian machines (so that we can generate the mask
1628 *   of low bytes efficiently). Again, we *could* do a byte
1629 *   swapping load on big-endian architectures if that is not
1630 *   expensive enough to make the optimization worthless.
1631 *
1632 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1633 *   do not trap on the (extremely unlikely) case of a page
1634 *   crossing operation.
1635 *
1636 * - Furthermore, we need an efficient 64-bit compile for the
1637 *   64-bit case in order to generate the "number of bytes in
1638 *   the final mask". Again, that could be replaced with a
1639 *   efficient population count instruction or similar.
1640 */
1641#ifdef CONFIG_DCACHE_WORD_ACCESS
1642
1643#include <asm/word-at-a-time.h>
1644
1645#ifdef CONFIG_64BIT
1646
1647static inline unsigned int fold_hash(unsigned long hash)
1648{
1649        hash += hash >> (8*sizeof(int));
1650        return hash;
1651}
1652
1653#else   /* 32-bit case */
1654
1655#define fold_hash(x) (x)
1656
1657#endif
1658
1659unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1660{
1661        unsigned long a, mask;
1662        unsigned long hash = 0;
1663
1664        for (;;) {
1665                a = load_unaligned_zeropad(name);
1666                if (len < sizeof(unsigned long))
1667                        break;
1668                hash += a;
1669                hash *= 9;
1670                name += sizeof(unsigned long);
1671                len -= sizeof(unsigned long);
1672                if (!len)
1673                        goto done;
1674        }
1675        mask = ~(~0ul << len*8);
1676        hash += mask & a;
1677done:
1678        return fold_hash(hash);
1679}
1680EXPORT_SYMBOL(full_name_hash);
1681
1682/*
1683 * Calculate the length and hash of the path component, and
1684 * return the length of the component;
1685 */
1686static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1687{
1688        unsigned long a, b, adata, bdata, mask, hash, len;
1689        const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1690
1691        hash = a = 0;
1692        len = -sizeof(unsigned long);
1693        do {
1694                hash = (hash + a) * 9;
1695                len += sizeof(unsigned long);
1696                a = load_unaligned_zeropad(name+len);
1697                b = a ^ REPEAT_BYTE('/');
1698        } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1699
1700        adata = prep_zero_mask(a, adata, &constants);
1701        bdata = prep_zero_mask(b, bdata, &constants);
1702
1703        mask = create_zero_mask(adata | bdata);
1704
1705        hash += a & zero_bytemask(mask);
1706        *hashp = fold_hash(hash);
1707
1708        return len + find_zero(mask);
1709}
1710
1711#else
1712
1713unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1714{
1715        unsigned long hash = init_name_hash();
1716        while (len--)
1717                hash = partial_name_hash(*name++, hash);
1718        return end_name_hash(hash);
1719}
1720EXPORT_SYMBOL(full_name_hash);
1721
1722/*
1723 * We know there's a real path component here of at least
1724 * one character.
1725 */
1726static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1727{
1728        unsigned long hash = init_name_hash();
1729        unsigned long len = 0, c;
1730
1731        c = (unsigned char)*name;
1732        do {
1733                len++;
1734                hash = partial_name_hash(c, hash);
1735                c = (unsigned char)name[len];
1736        } while (c && c != '/');
1737        *hashp = end_name_hash(hash);
1738        return len;
1739}
1740
1741#endif
1742
1743/*
1744 * Name resolution.
1745 * This is the basic name resolution function, turning a pathname into
1746 * the final dentry. We expect 'base' to be positive and a directory.
1747 *
1748 * Returns 0 and nd will have valid dentry and mnt on success.
1749 * Returns error and drops reference to input namei data on failure.
1750 */
1751static int link_path_walk(const char *name, struct nameidata *nd)
1752{
1753        struct path next;
1754        int err;
1755        
1756        while (*name=='/')
1757                name++;
1758        if (!*name)
1759                return 0;
1760
1761        /* At this point we know we have a real path component. */
1762        for(;;) {
1763                struct qstr this;
1764                long len;
1765                int type;
1766
1767                err = may_lookup(nd);
1768                if (err)
1769                        break;
1770
1771                len = hash_name(name, &this.hash);
1772                this.name = name;
1773                this.len = len;
1774
1775                type = LAST_NORM;
1776                if (name[0] == '.') switch (len) {
1777                        case 2:
1778                                if (name[1] == '.') {
1779                                        type = LAST_DOTDOT;
1780                                        nd->flags |= LOOKUP_JUMPED;
1781                                }
1782                                break;
1783                        case 1:
1784                                type = LAST_DOT;
1785                }
1786                if (likely(type == LAST_NORM)) {
1787                        struct dentry *parent = nd->path.dentry;
1788                        nd->flags &= ~LOOKUP_JUMPED;
1789                        if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1790                                err = parent->d_op->d_hash(parent, &this);
1791                                if (err < 0)
1792                                        break;
1793                        }
1794                }
1795
1796                nd->last = this;
1797                nd->last_type = type;
1798
1799                if (!name[len])
1800                        return 0;
1801                /*
1802                 * If it wasn't NUL, we know it was '/'. Skip that
1803                 * slash, and continue until no more slashes.
1804                 */
1805                do {
1806                        len++;
1807                } while (unlikely(name[len] == '/'));
1808                if (!name[len])
1809                        return 0;
1810
1811                name += len;
1812
1813                err = walk_component(nd, &next, LOOKUP_FOLLOW);
1814                if (err < 0)
1815                        return err;
1816
1817                if (err) {
1818                        err = nested_symlink(&next, nd);
1819                        if (err)
1820                                return err;
1821                }
1822                if (!can_lookup(nd->inode)) {
1823                        err = -ENOTDIR; 
1824                        break;
1825                }
1826        }
1827        terminate_walk(nd);
1828        return err;
1829}
1830
1831static int path_init(int dfd, const char *name, unsigned int flags,
1832                     struct nameidata *nd, struct file **fp)
1833{
1834        int retval = 0;
1835
1836        nd->last_type = LAST_ROOT; /* if there are only slashes... */
1837        nd->flags = flags | LOOKUP_JUMPED;
1838        nd->depth = 0;
1839        if (flags & LOOKUP_ROOT) {
1840                struct inode *inode = nd->root.dentry->d_inode;
1841                if (*name) {
1842                        if (!can_lookup(inode))
1843                                return -ENOTDIR;
1844                        retval = inode_permission(inode, MAY_EXEC);
1845                        if (retval)
1846                                return retval;
1847                }
1848                nd->path = nd->root;
1849                nd->inode = inode;
1850                if (flags & LOOKUP_RCU) {
1851                        lock_rcu_walk();
1852                        nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1853                } else {
1854                        path_get(&nd->path);
1855                }
1856                return 0;
1857        }
1858
1859        nd->root.mnt = NULL;
1860
1861        if (*name=='/') {
1862                if (flags & LOOKUP_RCU) {
1863                        lock_rcu_walk();
1864                        set_root_rcu(nd);
1865                } else {
1866                        set_root(nd);
1867                        path_get(&nd->root);
1868                }
1869                nd->path = nd->root;
1870        } else if (dfd == AT_FDCWD) {
1871                if (flags & LOOKUP_RCU) {
1872                        struct fs_struct *fs = current->fs;
1873                        unsigned seq;
1874
1875                        lock_rcu_walk();
1876
1877                        do {
1878                                seq = read_seqcount_begin(&fs->seq);
1879                                nd->path = fs->pwd;
1880                                nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1881                        } while (read_seqcount_retry(&fs->seq, seq));
1882                } else {
1883                        get_fs_pwd(current->fs, &nd->path);
1884                }
1885        } else {
1886                /* Caller must check execute permissions on the starting path component */
1887                struct fd f = fdget_raw(dfd);
1888                struct dentry *dentry;
1889
1890                if (!f.file)
1891                        return -EBADF;
1892
1893                dentry = f.file->f_path.dentry;
1894
1895                if (*name) {
1896                        if (!can_lookup(dentry->d_inode)) {
1897                                fdput(f);
1898                                return -ENOTDIR;
1899                        }
1900                }
1901
1902                nd->path = f.file->f_path;
1903                if (flags & LOOKUP_RCU) {
1904                        if (f.need_put)
1905                                *fp = f.file;
1906                        nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1907                        lock_rcu_walk();
1908                } else {
1909                        path_get(&nd->path);
1910                        fdput(f);
1911                }
1912        }
1913
1914        nd->inode = nd->path.dentry->d_inode;
1915        return 0;
1916}
1917
1918static inline int lookup_last(struct nameidata *nd, struct path *path)
1919{
1920        if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1921                nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1922
1923        nd->flags &= ~LOOKUP_PARENT;
1924        return walk_component(nd, path, nd->flags & LOOKUP_FOLLOW);
1925}
1926
1927/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1928static int path_lookupat(int dfd, const char *name,
1929                                unsigned int flags, struct nameidata *nd)
1930{
1931        struct file *base = NULL;
1932        struct path path;
1933        int err;
1934
1935        /*
1936         * Path walking is largely split up into 2 different synchronisation
1937         * schemes, rcu-walk and ref-walk (explained in
1938         * Documentation/filesystems/path-lookup.txt). These share much of the
1939         * path walk code, but some things particularly setup, cleanup, and
1940         * following mounts are sufficiently divergent that functions are
1941         * duplicated. Typically there is a function foo(), and its RCU
1942         * analogue, foo_rcu().
1943         *
1944         * -ECHILD is the error number of choice (just to avoid clashes) that
1945         * is returned if some aspect of an rcu-walk fails. Such an error must
1946         * be handled by restarting a traditional ref-walk (which will always
1947         * be able to complete).
1948         */
1949        err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
1950
1951        if (unlikely(err))
1952                return err;
1953
1954        current->total_link_count = 0;
1955        err = link_path_walk(name, nd);
1956
1957        if (!err && !(flags & LOOKUP_PARENT)) {
1958                err = lookup_last(nd, &path);
1959                while (err > 0) {
1960                        void *cookie;
1961                        struct path link = path;
1962                        err = may_follow_link(&link, nd);
1963                        if (unlikely(err))
1964                                break;
1965                        nd->flags |= LOOKUP_PARENT;
1966                        err = follow_link(&link, nd, &cookie);
1967                        if (err)
1968                                break;
1969                        err = lookup_last(nd, &path);
1970                        put_link(nd, &link, cookie);
1971                }
1972        }
1973
1974        if (!err)
1975                err = complete_walk(nd);
1976
1977        if (!err && nd->flags & LOOKUP_DIRECTORY) {
1978                if (!can_lookup(nd->inode)) {
1979                        path_put(&nd->path);
1980                        err = -ENOTDIR;
1981                }
1982        }
1983
1984        if (base)
1985                fput(base);
1986
1987        if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
1988                path_put(&nd->root);
1989                nd->root.mnt = NULL;
1990        }
1991        return err;
1992}
1993
1994static int filename_lookup(int dfd, struct filename *name,
1995                                unsigned int flags, struct nameidata *nd)
1996{
1997        int retval = path_lookupat(dfd, name->name, flags | LOOKUP_RCU, nd);
1998        if (unlikely(retval == -ECHILD))
1999                retval = path_lookupat(dfd, name->name, flags, nd);
2000        if (unlikely(retval == -ESTALE))
2001                retval = path_lookupat(dfd, name->name,
2002                                                flags | LOOKUP_REVAL, nd);
2003
2004        if (likely(!retval))
2005                audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT);
2006        return retval;
2007}
2008
2009static int do_path_lookup(int dfd, const char *name,
2010                                unsigned int flags, struct nameidata *nd)
2011{
2012        struct filename filename = { .name = name };
2013
2014        return filename_lookup(dfd, &filename, flags, nd);
2015}
2016
2017/* does lookup, returns the object with parent locked */
2018struct dentry *kern_path_locked(const char *name, struct path *path)
2019{
2020        struct nameidata nd;
2021        struct dentry *d;
2022        int err = do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, &nd);
2023        if (err)
2024                return ERR_PTR(err);
2025        if (nd.last_type != LAST_NORM) {
2026                path_put(&nd.path);
2027                return ERR_PTR(-EINVAL);
2028        }
2029        mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2030        d = __lookup_hash(&nd.last, nd.path.dentry, 0);
2031        if (IS_ERR(d)) {
2032                mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2033                path_put(&nd.path);
2034                return d;
2035        }
2036        *path = nd.path;
2037        return d;
2038}
2039
2040int kern_path(const char *name, unsigned int flags, struct path *path)
2041{
2042        struct nameidata nd;
2043        int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
2044        if (!res)
2045                *path = nd.path;
2046        return res;
2047}
2048
2049/**
2050 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2051 * @dentry:  pointer to dentry of the base directory
2052 * @mnt: pointer to vfs mount of the base directory
2053 * @name: pointer to file name
2054 * @flags: lookup flags
2055 * @path: pointer to struct path to fill
2056 */
2057int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2058                    const char *name, unsigned int flags,
2059                    struct path *path)
2060{
2061        struct nameidata nd;
2062        int err;
2063        nd.root.dentry = dentry;
2064        nd.root.mnt = mnt;
2065        BUG_ON(flags & LOOKUP_PARENT);
2066        /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
2067        err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
2068        if (!err)
2069                *path = nd.path;
2070        return err;
2071}
2072
2073/*
2074 * Restricted form of lookup. Doesn't follow links, single-component only,
2075 * needs parent already locked. Doesn't follow mounts.
2076 * SMP-safe.
2077 */
2078static struct dentry *lookup_hash(struct nameidata *nd)
2079{
2080        return __lookup_hash(&nd->last, nd->path.dentry, nd->flags);
2081}
2082
2083/**
2084 * lookup_one_len - filesystem helper to lookup single pathname component
2085 * @name:       pathname component to lookup
2086 * @base:       base directory to lookup from
2087 * @len:        maximum length @len should be interpreted to
2088 *
2089 * Note that this routine is purely a helper for filesystem usage and should
2090 * not be called by generic code.  Also note that by using this function the
2091 * nameidata argument is passed to the filesystem methods and a filesystem
2092 * using this helper needs to be prepared for that.
2093 */
2094struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2095{
2096        struct qstr this;
2097        unsigned int c;
2098        int err;
2099
2100        WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
2101
2102        this.name = name;
2103        this.len = len;
2104        this.hash = full_name_hash(name, len);
2105        if (!len)
2106                return ERR_PTR(-EACCES);
2107
2108        if (unlikely(name[0] == '.')) {
2109                if (len < 2 || (len == 2 && name[1] == '.'))
2110                        return ERR_PTR(-EACCES);
2111        }
2112
2113        while (len--) {
2114                c = *(const unsigned char *)name++;
2115                if (c == '/' || c == '\0')
2116                        return ERR_PTR(-EACCES);
2117        }
2118        /*
2119         * See if the low-level filesystem might want
2120         * to use its own hash..
2121         */
2122        if (base->d_flags & DCACHE_OP_HASH) {
2123                int err = base->d_op->d_hash(base, &this);
2124                if (err < 0)
2125                        return ERR_PTR(err);
2126        }
2127
2128        err = inode_permission(base->d_inode, MAY_EXEC);
2129        if (err)
2130                return ERR_PTR(err);
2131
2132        return __lookup_hash(&this, base, 0);
2133}
2134
2135int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2136                 struct path *path, int *empty)
2137{
2138        struct nameidata nd;
2139        struct filename *tmp = getname_flags(name, flags, empty);
2140        int err = PTR_ERR(tmp);
2141        if (!IS_ERR(tmp)) {
2142
2143                BUG_ON(flags & LOOKUP_PARENT);
2144
2145                err = filename_lookup(dfd, tmp, flags, &nd);
2146                putname(tmp);
2147                if (!err)
2148                        *path = nd.path;
2149        }
2150        return err;
2151}
2152
2153int user_path_at(int dfd, const char __user *name, unsigned flags,
2154                 struct path *path)
2155{
2156        return user_path_at_empty(dfd, name, flags, path, NULL);
2157}
2158
2159/*
2160 * NB: most callers don't do anything directly with the reference to the
2161 *     to struct filename, but the nd->last pointer points into the name string
2162 *     allocated by getname. So we must hold the reference to it until all
2163 *     path-walking is complete.
2164 */
2165static struct filename *
2166user_path_parent(int dfd, const char __user *path, struct nameidata *nd,
2167                 unsigned int flags)
2168{
2169        struct filename *s = getname(path);
2170        int error;
2171
2172        /* only LOOKUP_REVAL is allowed in extra flags */
2173        flags &= LOOKUP_REVAL;
2174
2175        if (IS_ERR(s))
2176                return s;
2177
2178        error = filename_lookup(dfd, s, flags | LOOKUP_PARENT, nd);
2179        if (error) {
2180                putname(s);
2181                return ERR_PTR(error);
2182        }
2183
2184        return s;
2185}
2186
2187/*
2188 * It's inline, so penalty for filesystems that don't use sticky bit is
2189 * minimal.
2190 */
2191static inline int check_sticky(struct inode *dir, struct inode *inode)
2192{
2193        kuid_t fsuid = current_fsuid();
2194
2195        if (!(dir->i_mode & S_ISVTX))
2196                return 0;
2197        if (uid_eq(inode->i_uid, fsuid))
2198                return 0;
2199        if (uid_eq(dir->i_uid, fsuid))
2200                return 0;
2201        return !inode_capable(inode, CAP_FOWNER);
2202}
2203
2204/*
2205 *      Check whether we can remove a link victim from directory dir, check
2206 *  whether the type of victim is right.
2207 *  1. We can't do it if dir is read-only (done in permission())
2208 *  2. We should have write and exec permissions on dir
2209 *  3. We can't remove anything from append-only dir
2210 *  4. We can't do anything with immutable dir (done in permission())
2211 *  5. If the sticky bit on dir is set we should either
2212 *      a. be owner of dir, or
2213 *      b. be owner of victim, or
2214 *      c. have CAP_FOWNER capability
2215 *  6. If the victim is append-only or immutable we can't do antyhing with
2216 *     links pointing to it.
2217 *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2218 *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2219 *  9. We can't remove a root or mountpoint.
2220 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2221 *     nfs_async_unlink().
2222 */
2223static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
2224{
2225        int error;
2226
2227        if (!victim->d_inode)
2228                return -ENOENT;
2229
2230        BUG_ON(victim->d_parent->d_inode != dir);
2231        audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2232
2233        error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2234        if (error)
2235                return error;
2236        if (IS_APPEND(dir))
2237                return -EPERM;
2238        if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
2239            IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
2240                return -EPERM;
2241        if (isdir) {
2242                if (!S_ISDIR(victim->d_inode->i_mode))
2243                        return -ENOTDIR;
2244                if (IS_ROOT(victim))
2245                        return -EBUSY;
2246        } else if (S_ISDIR(victim->d_inode->i_mode))
2247                return -EISDIR;
2248        if (IS_DEADDIR(dir))
2249                return -ENOENT;
2250        if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2251                return -EBUSY;
2252        return 0;
2253}
2254
2255/*      Check whether we can create an object with dentry child in directory
2256 *  dir.
2257 *  1. We can't do it if child already exists (open has special treatment for
2258 *     this case, but since we are inlined it's OK)
2259 *  2. We can't do it if dir is read-only (done in permission())
2260 *  3. We should have write and exec permissions on dir
2261 *  4. We can't do it if dir is immutable (done in permission())
2262 */
2263static inline int may_create(struct inode *dir, struct dentry *child)
2264{
2265        if (child->d_inode)
2266                return -EEXIST;
2267        if (IS_DEADDIR(dir))
2268                return -ENOENT;
2269        return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2270}
2271
2272/*
2273 * p1 and p2 should be directories on the same fs.
2274 */
2275struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2276{
2277        struct dentry *p;
2278
2279        if (p1 == p2) {
2280                mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2281                return NULL;
2282        }
2283
2284        mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2285
2286        p = d_ancestor(p2, p1);
2287        if (p) {
2288                mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2289                mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2290                return p;
2291        }
2292
2293        p = d_ancestor(p1, p2);
2294        if (p) {
2295                mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2296                mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2297                return p;
2298        }
2299
2300        mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2301        mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2302        return NULL;
2303}
2304
2305void unlock_rename(struct dentry *p1, struct dentry *p2)
2306{
2307        mutex_unlock(&p1->d_inode->i_mutex);
2308        if (p1 != p2) {
2309                mutex_unlock(&p2->d_inode->i_mutex);
2310                mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2311        }
2312}
2313
2314int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2315                bool want_excl)
2316{
2317        int error = may_create(dir, dentry);
2318        if (error)
2319                return error;
2320
2321        if (!dir->i_op->create)
2322                return -EACCES; /* shouldn't it be ENOSYS? */
2323        mode &= S_IALLUGO;
2324        mode |= S_IFREG;
2325        error = security_inode_create(dir, dentry, mode);
2326        if (error)
2327                return error;
2328        error = dir->i_op->create(dir, dentry, mode, want_excl);
2329        if (!error)
2330                fsnotify_create(dir, dentry);
2331        return error;
2332}
2333
2334static int may_open(struct path *path, int acc_mode, int flag)
2335{
2336        struct dentry *dentry = path->dentry;
2337        struct inode *inode = dentry->d_inode;
2338        int error;
2339
2340        /* O_PATH? */
2341        if (!acc_mode)
2342                return 0;
2343
2344        if (!inode)
2345                return -ENOENT;
2346
2347        switch (inode->i_mode & S_IFMT) {
2348        case S_IFLNK:
2349                return -ELOOP;
2350        case S_IFDIR:
2351                if (acc_mode & MAY_WRITE)
2352                        return -EISDIR;
2353                break;
2354        case S_IFBLK:
2355        case S_IFCHR:
2356                if (path->mnt->mnt_flags & MNT_NODEV)
2357                        return -EACCES;
2358                /*FALLTHRU*/
2359        case S_IFIFO:
2360        case S_IFSOCK:
2361                flag &= ~O_TRUNC;
2362                break;
2363        }
2364
2365        error = inode_permission(inode, acc_mode);
2366        if (error)
2367                return error;
2368
2369        /*
2370         * An append-only file must be opened in append mode for writing.
2371         */
2372        if (IS_APPEND(inode)) {
2373                if  ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2374                        return -EPERM;
2375                if (flag & O_TRUNC)
2376                        return -EPERM;
2377        }
2378
2379        /* O_NOATIME can only be set by the owner or superuser */
2380        if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2381                return -EPERM;
2382
2383        return 0;
2384}
2385
2386static int handle_truncate(struct file *filp)
2387{
2388        struct path *path = &filp->f_path;
2389        struct inode *inode = path->dentry->d_inode;
2390        int error = get_write_access(inode);
2391        if (error)
2392                return error;
2393        /*
2394         * Refuse to truncate files with mandatory locks held on them.
2395         */
2396        error = locks_verify_locked(inode);
2397        if (!error)
2398                error = security_path_truncate(path);
2399        if (!error) {
2400                error = do_truncate(path->dentry, 0,
2401                                    ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2402                                    filp);
2403        }
2404        put_write_access(inode);
2405        return error;
2406}
2407
2408static inline int open_to_namei_flags(int flag)
2409{
2410        if ((flag & O_ACCMODE) == 3)
2411                flag--;
2412        return flag;
2413}
2414
2415static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2416{
2417        int error = security_path_mknod(dir, dentry, mode, 0);
2418        if (error)
2419                return error;
2420
2421        error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2422        if (error)
2423                return error;
2424
2425        return security_inode_create(dir->dentry->d_inode, dentry, mode);
2426}
2427
2428/*
2429 * Attempt to atomically look up, create and open a file from a negative
2430 * dentry.
2431 *
2432 * Returns 0 if successful.  The file will have been created and attached to
2433 * @file by the filesystem calling finish_open().
2434 *
2435 * Returns 1 if the file was looked up only or didn't need creating.  The
2436 * caller will need to perform the open themselves.  @path will have been
2437 * updated to point to the new dentry.  This may be negative.
2438 *
2439 * Returns an error code otherwise.
2440 */
2441static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2442                        struct path *path, struct file *file,
2443                        const struct open_flags *op,
2444                        bool got_write, bool need_lookup,
2445                        int *opened)
2446{
2447        struct inode *dir =  nd->path.dentry->d_inode;
2448        unsigned open_flag = open_to_namei_flags(op->open_flag);
2449        umode_t mode;
2450        int error;
2451        int acc_mode;
2452        int create_error = 0;
2453        struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2454
2455        BUG_ON(dentry->d_inode);
2456
2457        /* Don't create child dentry for a dead directory. */
2458        if (unlikely(IS_DEADDIR(dir))) {
2459                error = -ENOENT;
2460                goto out;
2461        }
2462
2463        mode = op->mode;
2464        if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2465                mode &= ~current_umask();
2466
2467        if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT)) {
2468                open_flag &= ~O_TRUNC;
2469                *opened |= FILE_CREATED;
2470        }
2471
2472        /*
2473         * Checking write permission is tricky, bacuse we don't know if we are
2474         * going to actually need it: O_CREAT opens should work as long as the
2475         * file exists.  But checking existence breaks atomicity.  The trick is
2476         * to check access and if not granted clear O_CREAT from the flags.
2477         *
2478         * Another problem is returing the "right" error value (e.g. for an
2479         * O_EXCL open we want to return EEXIST not EROFS).
2480         */
2481        if (((open_flag & (O_CREAT | O_TRUNC)) ||
2482            (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
2483                if (!(open_flag & O_CREAT)) {
2484                        /*
2485                         * No O_CREATE -> atomicity not a requirement -> fall
2486                         * back to lookup + open
2487                         */
2488                        goto no_open;
2489                } else if (open_flag & (O_EXCL | O_TRUNC)) {
2490                        /* Fall back and fail with the right error */
2491                        create_error = -EROFS;
2492                        goto no_open;
2493                } else {
2494                        /* No side effects, safe to clear O_CREAT */
2495                        create_error = -EROFS;
2496                        open_flag &= ~O_CREAT;
2497                }
2498        }
2499
2500        if (open_flag & O_CREAT) {
2501                error = may_o_create(&nd->path, dentry, mode);
2502                if (error) {
2503                        create_error = error;
2504                        if (open_flag & O_EXCL)
2505                                goto no_open;
2506                        open_flag &= ~O_CREAT;
2507                }
2508        }
2509
2510        if (nd->flags & LOOKUP_DIRECTORY)
2511                open_flag |= O_DIRECTORY;
2512
2513        file->f_path.dentry = DENTRY_NOT_SET;
2514        file->f_path.mnt = nd->path.mnt;
2515        error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
2516                                      opened);
2517        if (error < 0) {
2518                if (create_error && error == -ENOENT)
2519                        error = create_error;
2520                goto out;
2521        }
2522
2523        acc_mode = op->acc_mode;
2524        if (*opened & FILE_CREATED) {
2525                fsnotify_create(dir, dentry);
2526                acc_mode = MAY_OPEN;
2527        }
2528
2529        if (error) {    /* returned 1, that is */
2530                if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
2531                        error = -EIO;
2532                        goto out;
2533                }
2534                if (file->f_path.dentry) {
2535                        dput(dentry);
2536                        dentry = file->f_path.dentry;
2537                }
2538                if (create_error && dentry->d_inode == NULL) {
2539                        error = create_error;
2540                        goto out;
2541                }
2542                goto looked_up;
2543        }
2544
2545        /*
2546         * We didn't have the inode before the open, so check open permission
2547         * here.
2548         */
2549        error = may_open(&file->f_path, acc_mode, open_flag);
2550        if (error)
2551                fput(file);
2552
2553out:
2554        dput(dentry);
2555        return error;
2556
2557no_open:
2558        if (need_lookup) {
2559                dentry = lookup_real(dir, dentry, nd->flags);
2560                if (IS_ERR(dentry))
2561                        return PTR_ERR(dentry);
2562
2563                if (create_error) {
2564                        int open_flag = op->open_flag;
2565
2566                        error = create_error;
2567                        if ((open_flag & O_EXCL)) {
2568                                if (!dentry->d_inode)
2569                                        goto out;
2570                        } else if (!dentry->d_inode) {
2571                                goto out;
2572                        } else if ((open_flag & O_TRUNC) &&
2573                                   S_ISREG(dentry->d_inode->i_mode)) {
2574                                goto out;
2575                        }
2576                        /* will fail later, go on to get the right error */
2577                }
2578        }
2579looked_up:
2580        path->dentry = dentry;
2581        path->mnt = nd->path.mnt;
2582        return 1;
2583}
2584
2585/*
2586 * Look up and maybe create and open the last component.
2587 *
2588 * Must be called with i_mutex held on parent.
2589 *
2590 * Returns 0 if the file was successfully atomically created (if necessary) and
2591 * opened.  In this case the file will be returned attached to @file.
2592 *
2593 * Returns 1 if the file was not completely opened at this time, though lookups
2594 * and creations will have been performed and the dentry returned in @path will
2595 * be positive upon return if O_CREAT was specified.  If O_CREAT wasn't
2596 * specified then a negative dentry may be returned.
2597 *
2598 * An error code is returned otherwise.
2599 *
2600 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2601 * cleared otherwise prior to returning.
2602 */
2603static int lookup_open(struct nameidata *nd, struct path *path,
2604                        struct file *file,
2605                        const struct open_flags *op,
2606                        bool got_write, int *opened)
2607{
2608        struct dentry *dir = nd->path.dentry;
2609        struct inode *dir_inode = dir->d_inode;
2610        struct dentry *dentry;
2611        int error;
2612        bool need_lookup;
2613
2614        *opened &= ~FILE_CREATED;
2615        dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
2616        if (IS_ERR(dentry))
2617                return PTR_ERR(dentry);
2618
2619        /* Cached positive dentry: will open in f_op->open */
2620        if (!need_lookup && dentry->d_inode)
2621                goto out_no_open;
2622
2623        if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
2624                return atomic_open(nd, dentry, path, file, op, got_write,
2625                                   need_lookup, opened);
2626        }
2627
2628        if (need_lookup) {
2629                BUG_ON(dentry->d_inode);
2630
2631                dentry = lookup_real(dir_inode, dentry, nd->flags);
2632                if (IS_ERR(dentry))
2633                        return PTR_ERR(dentry);
2634        }
2635
2636        /* Negative dentry, just create the file */
2637        if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
2638                umode_t mode = op->mode;
2639                if (!IS_POSIXACL(dir->d_inode))
2640                        mode &= ~current_umask();
2641                /*
2642                 * This write is needed to ensure that a
2643                 * rw->ro transition does not occur between
2644                 * the time when the file is created and when
2645                 * a permanent write count is taken through
2646                 * the 'struct file' in finish_open().
2647                 */
2648                if (!got_write) {
2649                        error = -EROFS;
2650                        goto out_dput;
2651                }
2652                *opened |= FILE_CREATED;
2653                error = security_path_mknod(&nd->path, dentry, mode, 0);
2654                if (error)
2655                        goto out_dput;
2656                error = vfs_create(dir->d_inode, dentry, mode,
2657                                   nd->flags & LOOKUP_EXCL);
2658                if (error)
2659                        goto out_dput;
2660        }
2661out_no_open:
2662        path->dentry = dentry;
2663        path->mnt = nd->path.mnt;
2664        return 1;
2665
2666out_dput:
2667        dput(dentry);
2668        return error;
2669}
2670
2671/*
2672 * Handle the last step of open()
2673 */
2674static int do_last(struct nameidata *nd, struct path *path,
2675                   struct file *file, const struct open_flags *op,
2676                   int *opened, struct filename *name)
2677{
2678        struct dentry *dir = nd->path.dentry;
2679        int open_flag = op->open_flag;
2680        bool will_truncate = (open_flag & O_TRUNC) != 0;
2681        bool got_write = false;
2682        int acc_mode = op->acc_mode;
2683        struct inode *inode;
2684        bool symlink_ok = false;
2685        struct path save_parent = { .dentry = NULL, .mnt = NULL };
2686        bool retried = false;
2687        int error;
2688
2689        nd->flags &= ~LOOKUP_PARENT;
2690        nd->flags |= op->intent;
2691
2692        if (nd->last_type != LAST_NORM) {
2693                error = handle_dots(nd, nd->last_type);
2694                if (error)
2695                        return error;
2696                goto finish_open;
2697        }
2698
2699        if (!(open_flag & O_CREAT)) {
2700                if (nd->last.name[nd->last.len])
2701                        nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2702                if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2703                        symlink_ok = true;
2704                /* we _can_ be in RCU mode here */
2705                error = lookup_fast(nd, path, &inode);
2706                if (likely(!error))
2707                        goto finish_lookup;
2708
2709                if (error < 0)
2710                        goto out;
2711
2712                BUG_ON(nd->inode != dir->d_inode);
2713        } else {
2714                /* create side of things */
2715                /*
2716                 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2717                 * has been cleared when we got to the last component we are
2718                 * about to look up
2719                 */
2720                error = complete_walk(nd);
2721                if (error)
2722                        return error;
2723
2724                audit_inode(name, dir, LOOKUP_PARENT);
2725                error = -EISDIR;
2726                /* trailing slashes? */
2727                if (nd->last.name[nd->last.len])
2728                        goto out;
2729        }
2730
2731retry_lookup:
2732        if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
2733                error = mnt_want_write(nd->path.mnt);
2734                if (!error)
2735                        got_write = true;
2736                /*
2737                 * do _not_ fail yet - we might not need that or fail with
2738                 * a different error; let lookup_open() decide; we'll be
2739                 * dropping this one anyway.
2740                 */
2741        }
2742        mutex_lock(&dir->d_inode->i_mutex);
2743        error = lookup_open(nd, path, file, op, got_write, opened);
2744        mutex_unlock(&dir->d_inode->i_mutex);
2745
2746        if (error <= 0) {
2747                if (error)
2748                        goto out;
2749
2750                if ((*opened & FILE_CREATED) ||
2751                    !S_ISREG(file_inode(file)->i_mode))
2752                        will_truncate = false;
2753
2754                audit_inode(name, file->f_path.dentry, 0);
2755                goto opened;
2756        }
2757
2758        if (*opened & FILE_CREATED) {
2759                /* Don't check for write permission, don't truncate */
2760                open_flag &= ~O_TRUNC;
2761                will_truncate = false;
2762                acc_mode = MAY_OPEN;
2763                path_to_nameidata(path, nd);
2764                goto finish_open_created;
2765        }
2766
2767        /*
2768         * create/update audit record if it already exists.
2769         */
2770        if (path->dentry->d_inode)
2771                audit_inode(name, path->dentry, 0);
2772
2773        /*
2774         * If atomic_open() acquired write access it is dropped now due to
2775         * possible mount and symlink following (this might be optimized away if
2776         * necessary...)
2777         */
2778        if (got_write) {
2779                mnt_drop_write(nd->path.mnt);
2780                got_write = false;
2781        }
2782
2783        error = -EEXIST;
2784        if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))
2785                goto exit_dput;
2786
2787        error = follow_managed(path, nd->flags);
2788        if (error < 0)
2789                goto exit_dput;
2790
2791        if (error)
2792                nd->flags |= LOOKUP_JUMPED;
2793
2794        BUG_ON(nd->flags & LOOKUP_RCU);
2795        inode = path->dentry->d_inode;
2796finish_lookup:
2797        /* we _can_ be in RCU mode here */
2798        error = -ENOENT;
2799        if (!inode) {
2800                path_to_nameidata(path, nd);
2801                goto out;
2802        }
2803
2804        if (should_follow_link(inode, !symlink_ok)) {
2805                if (nd->flags & LOOKUP_RCU) {
2806                        if (unlikely(unlazy_walk(nd, path->dentry))) {
2807                                error = -ECHILD;
2808                                goto out;
2809                        }
2810                }
2811                BUG_ON(inode != path->dentry->d_inode);
2812                return 1;
2813        }
2814
2815        if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) {
2816                path_to_nameidata(path, nd);
2817        } else {
2818                save_parent.dentry = nd->path.dentry;
2819                save_parent.mnt = mntget(path->mnt);
2820                nd->path.dentry = path->dentry;
2821
2822        }
2823        nd->inode = inode;
2824        /* Why this, you ask?  _Now_ we might have grown LOOKUP_JUMPED... */
2825finish_open:
2826        error = complete_walk(nd);
2827        if (error) {
2828                path_put(&save_parent);
2829                return error;
2830        }
2831        audit_inode(name, nd->path.dentry, 0);
2832        error = -EISDIR;
2833        if ((open_flag & O_CREAT) && S_ISDIR(nd->inode->i_mode))
2834                goto out;
2835        error = -ENOTDIR;
2836        if ((nd->flags & LOOKUP_DIRECTORY) && !can_lookup(nd->inode))
2837                goto out;
2838        if (!S_ISREG(nd->inode->i_mode))
2839                will_truncate = false;
2840
2841        if (will_truncate) {
2842                error = mnt_want_write(nd->path.mnt);
2843                if (error)
2844                        goto out;
2845                got_write = true;
2846        }
2847finish_open_created:
2848        error = may_open(&nd->path, acc_mode, open_flag);
2849        if (error)
2850                goto out;
2851        file->f_path.mnt = nd->path.mnt;
2852        error = finish_open(file, nd->path.dentry, NULL, opened);
2853        if (error) {
2854                if (error == -EOPENSTALE)
2855                        goto stale_open;
2856                goto out;
2857        }
2858opened:
2859        error = open_check_o_direct(file);
2860        if (error)
2861                goto exit_fput;
2862        error = ima_file_check(file, op->acc_mode);
2863        if (error)
2864                goto exit_fput;
2865
2866        if (will_truncate) {
2867                error = handle_truncate(file);
2868                if (error)
2869                        goto exit_fput;
2870        }
2871out:
2872        if (got_write)
2873                mnt_drop_write(nd->path.mnt);
2874        path_put(&save_parent);
2875        terminate_walk(nd);
2876        return error;
2877
2878exit_dput:
2879        path_put_conditional(path, nd);
2880        goto out;
2881exit_fput:
2882        fput(file);
2883        goto out;
2884
2885stale_open:
2886        /* If no saved parent or already retried then can't retry */
2887        if (!save_parent.dentry || retried)
2888                goto out;
2889
2890        BUG_ON(save_parent.dentry != dir);
2891        path_put(&nd->path);
2892        nd->path = save_parent;
2893        nd->inode = dir->d_inode;
2894        save_parent.mnt = NULL;
2895        save_parent.dentry = NULL;
2896        if (got_write) {
2897                mnt_drop_write(nd->path.mnt);
2898                got_write = false;
2899        }
2900        retried = true;
2901        goto retry_lookup;
2902}
2903
2904static int do_tmpfile(int dfd, struct filename *pathname,
2905                struct nameidata *nd, int flags,
2906                const struct open_flags *op,
2907                struct file *file, int *opened)
2908{
2909        static const struct qstr name = QSTR_INIT("/", 1);
2910        struct dentry *dentry, *child;
2911        struct inode *dir;
2912        int error = path_lookupat(dfd, pathname->name,
2913                                  flags | LOOKUP_DIRECTORY, nd);
2914        if (unlikely(error))
2915                return error;
2916        error = mnt_want_write(nd->path.mnt);
2917        if (unlikely(error))
2918                goto out;
2919        /* we want directory to be writable */
2920        error = inode_permission(nd->inode, MAY_WRITE | MAY_EXEC);
2921        if (error)
2922                goto out2;
2923        dentry = nd->path.dentry;
2924        dir = dentry->d_inode;
2925        if (!dir->i_op->tmpfile) {
2926                error = -EOPNOTSUPP;
2927                goto out2;
2928        }
2929        child = d_alloc(dentry, &name);
2930        if (unlikely(!child)) {
2931                error = -ENOMEM;
2932                goto out2;
2933        }
2934        nd->flags &= ~LOOKUP_DIRECTORY;
2935        nd->flags |= op->intent;
2936        dput(nd->path.dentry);
2937        nd->path.dentry = child;
2938        error = dir->i_op->tmpfile(dir, nd->path.dentry, op->mode);
2939        if (error)
2940                goto out2;
2941        audit_inode(pathname, nd->path.dentry, 0);
2942        error = may_open(&nd->path, op->acc_mode, op->open_flag);
2943        if (error)
2944                goto out2;
2945        file->f_path.mnt = nd->path.mnt;
2946        error = finish_open(file, nd->path.dentry, NULL, opened);
2947        if (error)
2948                goto out2;
2949        error = open_check_o_direct(file);
2950        if (error) {
2951                fput(file);
2952        } else if (!(op->open_flag & O_EXCL)) {
2953                struct inode *inode = file_inode(file);
2954                spin_lock(&inode->i_lock);
2955                inode->i_state |= I_LINKABLE;
2956                spin_unlock(&inode->i_lock);
2957        }
2958out2:
2959        mnt_drop_write(nd->path.mnt);
2960out:
2961        path_put(&nd->path);
2962        return error;
2963}
2964
2965static struct file *path_openat(int dfd, struct filename *pathname,
2966                struct nameidata *nd, const struct open_flags *op, int flags)
2967{
2968        struct file *base = NULL;
2969        struct file *file;
2970        struct path path;
2971        int opened = 0;
2972        int error;
2973
2974        file = get_empty_filp();
2975        if (IS_ERR(file))
2976                return file;
2977
2978        file->f_flags = op->open_flag;
2979
2980        if (unlikely(file->f_flags & __O_TMPFILE)) {
2981                error = do_tmpfile(dfd, pathname, nd, flags, op, file, &opened);
2982                goto out;
2983        }
2984
2985        error = path_init(dfd, pathname->name, flags | LOOKUP_PARENT, nd, &base);
2986        if (unlikely(error))
2987                goto out;
2988
2989        current->total_link_count = 0;
2990        error = link_path_walk(pathname->name, nd);
2991        if (unlikely(error))
2992                goto out;
2993
2994        error = do_last(nd, &path, file, op, &opened, pathname);
2995        while (unlikely(error > 0)) { /* trailing symlink */
2996                struct path link = path;
2997                void *cookie;
2998                if (!(nd->flags & LOOKUP_FOLLOW)) {
2999                        path_put_conditional(&path, nd);
3000                        path_put(&nd->path);
3001                        error = -ELOOP;
3002                        break;
3003                }
3004                error = may_follow_link(&link, nd);
3005                if (unlikely(error))
3006                        break;
3007                nd->flags |= LOOKUP_PARENT;
3008                nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3009                error = follow_link(&link, nd, &cookie);
3010                if (unlikely(error))
3011                        break;
3012                error = do_last(nd, &path, file, op, &opened, pathname);
3013                put_link(nd, &link, cookie);
3014        }
3015out:
3016        if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
3017                path_put(&nd->root);
3018        if (base)
3019                fput(base);
3020        if (!(opened & FILE_OPENED)) {
3021                BUG_ON(!error);
3022                put_filp(file);
3023        }
3024        if (unlikely(error)) {
3025                if (error == -EOPENSTALE) {
3026                        if (flags & LOOKUP_RCU)
3027                                error = -ECHILD;
3028                        else
3029                                error = -ESTALE;
3030                }
3031                file = ERR_PTR(error);
3032        }
3033        return file;
3034}
3035
3036struct file *do_filp_open(int dfd, struct filename *pathname,
3037                const struct open_flags *op)
3038{
3039        struct nameidata nd;
3040        int flags = op->lookup_flags;
3041        struct file *filp;
3042
3043        filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
3044        if (unlikely(filp == ERR_PTR(-ECHILD)))
3045                filp = path_openat(dfd, pathname, &nd, op, flags);
3046        if (unlikely(filp == ERR_PTR(-ESTALE)))
3047                filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
3048        return filp;
3049}
3050
3051struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3052                const char *name, const struct open_flags *op)
3053{
3054        struct nameidata nd;
3055        struct file *file;
3056        struct filename filename = { .name = name };
3057        int flags = op->lookup_flags | LOOKUP_ROOT;
3058
3059        nd.root.mnt = mnt;
3060        nd.root.dentry = dentry;
3061
3062        if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
3063                return ERR_PTR(-ELOOP);
3064
3065        file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_RCU);
3066        if (unlikely(file == ERR_PTR(-ECHILD)))
3067                file = path_openat(-1, &filename, &nd, op, flags);
3068        if (unlikely(file == ERR_PTR(-ESTALE)))
3069                file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_REVAL);
3070        return file;
3071}
3072
3073struct dentry *kern_path_create(int dfd, const char *pathname,
3074                                struct path *path, unsigned int lookup_flags)
3075{
3076        struct dentry *dentry = ERR_PTR(-EEXIST);
3077        struct nameidata nd;
3078        int err2;
3079        int error;
3080        bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3081
3082        /*
3083         * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3084         * other flags passed in are ignored!
3085         */
3086        lookup_flags &= LOOKUP_REVAL;
3087
3088        error = do_path_lookup(dfd, pathname, LOOKUP_PARENT|lookup_flags, &nd);
3089        if (error)
3090                return ERR_PTR(error);
3091
3092        /*
3093         * Yucky last component or no last component at all?
3094         * (foo/., foo/.., /////)
3095         */
3096        if (nd.last_type != LAST_NORM)
3097                goto out;
3098        nd.flags &= ~LOOKUP_PARENT;
3099        nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3100
3101        /* don't fail immediately if it's r/o, at least try to report other errors */
3102        err2 = mnt_want_write(nd.path.mnt);
3103        /*
3104         * Do the final lookup.
3105         */
3106        mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3107        dentry = lookup_hash(&nd);
3108        if (IS_ERR(dentry))
3109                goto unlock;
3110
3111        error = -EEXIST;
3112        if (dentry->d_inode)
3113                goto fail;
3114        /*
3115         * Special case - lookup gave negative, but... we had foo/bar/
3116         * From the vfs_mknod() POV we just have a negative dentry -
3117         * all is fine. Let's be bastards - you had / on the end, you've
3118         * been asking for (non-existent) directory. -ENOENT for you.
3119         */
3120        if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
3121                error = -ENOENT;
3122                goto fail;
3123        }
3124        if (unlikely(err2)) {
3125                error = err2;
3126                goto fail;
3127        }
3128        *path = nd.path;
3129        return dentry;
3130fail:
3131        dput(dentry);
3132        dentry = ERR_PTR(error);
3133unlock:
3134        mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3135        if (!err2)
3136                mnt_drop_write(nd.path.mnt);
3137out:
3138        path_put(&nd.path);
3139        return dentry;
3140}
3141EXPORT_SYMBOL(kern_path_create);
3142
3143void done_path_create(struct path *path, struct dentry *dentry)
3144{
3145        dput(dentry);
3146        mutex_unlock(&path->dentry->d_inode->i_mutex);
3147        mnt_drop_write(path->mnt);
3148        path_put(path);
3149}
3150EXPORT_SYMBOL(done_path_create);
3151
3152struct dentry *user_path_create(int dfd, const char __user *pathname,
3153                                struct path *path, unsigned int lookup_flags)
3154{
3155        struct filename *tmp = getname(pathname);
3156        struct dentry *res;
3157        if (IS_ERR(tmp))
3158                return ERR_CAST(tmp);
3159        res = kern_path_create(dfd, tmp->name, path, lookup_flags);
3160        putname(tmp);
3161        return res;
3162}
3163EXPORT_SYMBOL(user_path_create);
3164
3165int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3166{
3167        int error = may_create(dir, dentry);
3168
3169        if (error)
3170                return error;
3171
3172        if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3173                return -EPERM;
3174
3175        if (!dir->i_op->mknod)
3176                return -EPERM;
3177
3178        error = devcgroup_inode_mknod(mode, dev);
3179        if (error)
3180                return error;
3181
3182        error = security_inode_mknod(dir, dentry, mode, dev);
3183        if (error)
3184                return error;
3185
3186        error = dir->i_op->mknod(dir, dentry, mode, dev);
3187        if (!error)
3188                fsnotify_create(dir, dentry);
3189        return error;
3190}
3191
3192static int may_mknod(umode_t mode)
3193{
3194        switch (mode & S_IFMT) {
3195        case S_IFREG:
3196        case S_IFCHR:
3197        case S_IFBLK:
3198        case S_IFIFO:
3199        case S_IFSOCK:
3200        case 0: /* zero mode translates to S_IFREG */
3201                return 0;
3202        case S_IFDIR:
3203                return -EPERM;
3204        default:
3205                return -EINVAL;
3206        }
3207}
3208
3209SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3210                unsigned, dev)
3211{
3212        struct dentry *dentry;
3213        struct path path;
3214        int error;
3215        unsigned int lookup_flags = 0;
3216
3217        error = may_mknod(mode);
3218        if (error)
3219                return error;
3220retry:
3221        dentry = user_path_create(dfd, filename, &path, lookup_flags);
3222        if (IS_ERR(dentry))
3223                return PTR_ERR(dentry);
3224
3225        if (!IS_POSIXACL(path.dentry->d_inode))
3226                mode &= ~current_umask();
3227        error = security_path_mknod(&path, dentry, mode, dev);
3228        if (error)
3229                goto out;
3230        switch (mode & S_IFMT) {
3231                case 0: case S_IFREG:
3232                        error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3233                        break;
3234                case S_IFCHR: case S_IFBLK:
3235                        error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3236                                        new_decode_dev(dev));
3237                        break;
3238                case S_IFIFO: case S_IFSOCK:
3239                        error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3240                        break;
3241        }
3242out:
3243        done_path_create(&path, dentry);
3244        if (retry_estale(error, lookup_flags)) {
3245                lookup_flags |= LOOKUP_REVAL;
3246                goto retry;
3247        }
3248        return error;
3249}
3250
3251SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3252{
3253        return sys_mknodat(AT_FDCWD, filename, mode, dev);
3254}
3255
3256int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3257{
3258        int error = may_create(dir, dentry);
3259        unsigned max_links = dir->i_sb->s_max_links;
3260
3261        if (error)
3262                return error;
3263
3264        if (!dir->i_op->mkdir)
3265                return -EPERM;
3266
3267        mode &= (S_IRWXUGO|S_ISVTX);
3268        error = security_inode_mkdir(dir, dentry, mode);
3269        if (error)
3270                return error;
3271
3272        if (max_links && dir->i_nlink >= max_links)
3273                return -EMLINK;
3274
3275        error = dir->i_op->mkdir(dir, dentry, mode);
3276        if (!error)
3277                fsnotify_mkdir(dir, dentry);
3278        return error;
3279}
3280
3281SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3282{
3283        struct dentry *dentry;
3284        struct path path;
3285        int error;
3286        unsigned int lookup_flags = LOOKUP_DIRECTORY;
3287
3288retry:
3289        dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3290        if (IS_ERR(dentry))
3291                return PTR_ERR(dentry);
3292
3293        if (!IS_POSIXACL(path.dentry->d_inode))
3294                mode &= ~current_umask();
3295        error = security_path_mkdir(&path, dentry, mode);
3296        if (!error)
3297                error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3298        done_path_create(&path, dentry);
3299        if (retry_estale(error, lookup_flags)) {
3300                lookup_flags |= LOOKUP_REVAL;
3301                goto retry;
3302        }
3303        return error;
3304}
3305
3306SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3307{
3308        return sys_mkdirat(AT_FDCWD, pathname, mode);
3309}
3310
3311/*
3312 * The dentry_unhash() helper will try to drop the dentry early: we
3313 * should have a usage count of 1 if we're the only user of this
3314 * dentry, and if that is true (possibly after pruning the dcache),
3315 * then we drop the dentry now.
3316 *
3317 * A low-level filesystem can, if it choses, legally
3318 * do a
3319 *
3320 *      if (!d_unhashed(dentry))
3321 *              return -EBUSY;
3322 *
3323 * if it cannot handle the case of removing a directory
3324 * that is still in use by something else..
3325 */
3326void dentry_unhash(struct dentry *dentry)
3327{
3328        shrink_dcache_parent(dentry);
3329        spin_lock(&dentry->d_lock);
3330        if (dentry->d_lockref.count == 1)
3331                __d_drop(dentry);
3332        spin_unlock(&dentry->d_lock);
3333}
3334
3335int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3336{
3337        int error = may_delete(dir, dentry, 1);
3338
3339        if (error)
3340                return error;
3341
3342        if (!dir->i_op->rmdir)
3343                return -EPERM;
3344
3345        dget(dentry);
3346        mutex_lock(&dentry->d_inode->i_mutex);
3347
3348        error = -EBUSY;
3349        if (d_mountpoint(dentry))
3350                goto out;
3351
3352        error = security_inode_rmdir(dir, dentry);
3353        if (error)
3354                goto out;
3355
3356        shrink_dcache_parent(dentry);
3357        error = dir->i_op->rmdir(dir, dentry);
3358        if (error)
3359                goto out;
3360
3361        dentry->d_inode->i_flags |= S_DEAD;
3362        dont_mount(dentry);
3363
3364out:
3365        mutex_unlock(&dentry->d_inode->i_mutex);
3366        dput(dentry);
3367        if (!error)
3368                d_delete(dentry);
3369        return error;
3370}
3371
3372static long do_rmdir(int dfd, const char __user *pathname)
3373{
3374        int error = 0;
3375        struct filename *name;
3376        struct dentry *dentry;
3377        struct nameidata nd;
3378        unsigned int lookup_flags = 0;
3379retry:
3380        name = user_path_parent(dfd, pathname, &nd, lookup_flags);
3381        if (IS_ERR(name))
3382                return PTR_ERR(name);
3383
3384        switch(nd.last_type) {
3385        case LAST_DOTDOT:
3386                error = -ENOTEMPTY;
3387                goto exit1;
3388        case LAST_DOT:
3389                error = -EINVAL;
3390                goto exit1;
3391        case LAST_ROOT:
3392                error = -EBUSY;
3393                goto exit1;
3394        }
3395
3396        nd.flags &= ~LOOKUP_PARENT;
3397        error = mnt_want_write(nd.path.mnt);
3398        if (error)
3399                goto exit1;
3400
3401        mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3402        dentry = lookup_hash(&nd);
3403        error = PTR_ERR(dentry);
3404        if (IS_ERR(dentry))
3405                goto exit2;
3406        if (!dentry->d_inode) {
3407                error = -ENOENT;
3408                goto exit3;
3409        }
3410        error = security_path_rmdir(&nd.path, dentry);
3411        if (error)
3412                goto exit3;
3413        error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
3414exit3:
3415        dput(dentry);
3416exit2:
3417        mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3418        mnt_drop_write(nd.path.mnt);
3419exit1:
3420        path_put(&nd.path);
3421        putname(name);
3422        if (retry_estale(error, lookup_flags)) {
3423                lookup_flags |= LOOKUP_REVAL;
3424                goto retry;
3425        }
3426        return error;
3427}
3428
3429SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3430{
3431        return do_rmdir(AT_FDCWD, pathname);
3432}
3433
3434int vfs_unlink(struct inode *dir, struct dentry *dentry)
3435{
3436        int error = may_delete(dir, dentry, 0);
3437
3438        if (error)
3439                return error;
3440
3441        if (!dir->i_op->unlink)
3442                return -EPERM;
3443
3444        mutex_lock(&dentry->d_inode->i_mutex);
3445        if (d_mountpoint(dentry))
3446                error = -EBUSY;
3447        else {
3448                error = security_inode_unlink(dir, dentry);
3449                if (!error) {
3450                        error = dir->i_op->unlink(dir, dentry);
3451                        if (!error)
3452                                dont_mount(dentry);
3453                }
3454        }
3455        mutex_unlock(&dentry->d_inode->i_mutex);
3456
3457        /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3458        if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3459                fsnotify_link_count(dentry->d_inode);
3460                d_delete(dentry);
3461        }
3462
3463        return error;
3464}
3465
3466/*
3467 * Make sure that the actual truncation of the file will occur outside its
3468 * directory's i_mutex.  Truncate can take a long time if there is a lot of
3469 * writeout happening, and we don't want to prevent access to the directory
3470 * while waiting on the I/O.
3471 */
3472static long do_unlinkat(int dfd, const char __user *pathname)
3473{
3474        int error;
3475        struct filename *name;
3476        struct dentry *dentry;
3477        struct nameidata nd;
3478        struct inode *inode = NULL;
3479        unsigned int lookup_flags = 0;
3480retry:
3481        name = user_path_parent(dfd, pathname, &nd, lookup_flags);
3482        if (IS_ERR(name))
3483                return PTR_ERR(name);
3484
3485        error = -EISDIR;
3486        if (nd.last_type != LAST_NORM)
3487                goto exit1;
3488
3489        nd.flags &= ~LOOKUP_PARENT;
3490        error = mnt_want_write(nd.path.mnt);
3491        if (error)
3492                goto exit1;
3493
3494        mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3495        dentry = lookup_hash(&nd);
3496        error = PTR_ERR(dentry);
3497        if (!IS_ERR(dentry)) {
3498                /* Why not before? Because we want correct error value */
3499                if (nd.last.name[nd.last.len])
3500                        goto slashes;
3501                inode = dentry->d_inode;
3502                if (!inode)
3503                        goto slashes;
3504                ihold(inode);
3505                error = security_path_unlink(&nd.path, dentry);
3506                if (error)
3507                        goto exit2;
3508                error = vfs_unlink(nd.path.dentry->d_inode, dentry);
3509exit2:
3510                dput(dentry);
3511        }
3512        mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3513        if (inode)
3514                iput(inode);    /* truncate the inode here */
3515        mnt_drop_write(nd.path.mnt);
3516exit1:
3517        path_put(&nd.path);
3518        putname(name);
3519        if (retry_estale(error, lookup_flags)) {
3520                lookup_flags |= LOOKUP_REVAL;
3521                inode = NULL;
3522                goto retry;
3523        }
3524        return error;
3525
3526slashes:
3527        error = !dentry->d_inode ? -ENOENT :
3528                S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
3529        goto exit2;
3530}
3531
3532SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
3533{
3534        if ((flag & ~AT_REMOVEDIR) != 0)
3535                return -EINVAL;
3536
3537        if (flag & AT_REMOVEDIR)
3538                return do_rmdir(dfd, pathname);
3539
3540        return do_unlinkat(dfd, pathname);
3541}
3542
3543SYSCALL_DEFINE1(unlink, const char __user *, pathname)
3544{
3545        return do_unlinkat(AT_FDCWD, pathname);
3546}
3547
3548int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
3549{
3550        int error = may_create(dir, dentry);
3551
3552        if (error)
3553                return error;
3554
3555        if (!dir->i_op->symlink)
3556                return -EPERM;
3557
3558        error = security_inode_symlink(dir, dentry, oldname);
3559        if (error)
3560                return error;
3561
3562        error = dir->i_op->symlink(dir, dentry, oldname);
3563        if (!error)
3564                fsnotify_create(dir, dentry);
3565        return error;
3566}
3567
3568SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3569                int, newdfd, const char __user *, newname)
3570{
3571        int error;
3572        struct filename *from;
3573        struct dentry *dentry;
3574        struct path path;
3575        unsigned int lookup_flags = 0;
3576
3577        from = getname(oldname);
3578        if (IS_ERR(from))
3579                return PTR_ERR(from);
3580retry:
3581        dentry = user_path_create(newdfd, newname, &path, lookup_flags);
3582        error = PTR_ERR(dentry);
3583        if (IS_ERR(dentry))
3584                goto out_putname;
3585
3586        error = security_path_symlink(&path, dentry, from->name);
3587        if (!error)
3588                error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
3589        done_path_create(&path, dentry);
3590        if (retry_estale(error, lookup_flags)) {
3591                lookup_flags |= LOOKUP_REVAL;
3592                goto retry;
3593        }
3594out_putname:
3595        putname(from);
3596        return error;
3597}
3598
3599SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
3600{
3601        return sys_symlinkat(oldname, AT_FDCWD, newname);
3602}
3603
3604int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
3605{
3606        struct inode *inode = old_dentry->d_inode;
3607        unsigned max_links = dir->i_sb->s_max_links;
3608        int error;
3609
3610        if (!inode)
3611                return -ENOENT;
3612
3613        error = may_create(dir, new_dentry);
3614        if (error)
3615                return error;
3616
3617        if (dir->i_sb != inode->i_sb)
3618                return -EXDEV;
3619
3620        /*
3621         * A link to an append-only or immutable file cannot be created.
3622         */
3623        if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3624                return -EPERM;
3625        if (!dir->i_op->link)
3626                return -EPERM;
3627        if (S_ISDIR(inode->i_mode))
3628                return -EPERM;
3629
3630        error = security_inode_link(old_dentry, dir, new_dentry);
3631        if (error)
3632                return error;
3633
3634        mutex_lock(&inode->i_mutex);
3635        /* Make sure we don't allow creating hardlink to an unlinked file */
3636        if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
3637                error =  -ENOENT;
3638        else if (max_links && inode->i_nlink >= max_links)
3639                error = -EMLINK;
3640        else
3641                error = dir->i_op->link(old_dentry, dir, new_dentry);
3642
3643        if (!error && (inode->i_state & I_LINKABLE)) {
3644                spin_lock(&inode->i_lock);
3645                inode->i_state &= ~I_LINKABLE;
3646                spin_unlock(&inode->i_lock);
3647        }
3648        mutex_unlock(&inode->i_mutex);
3649        if (!error)
3650                fsnotify_link(dir, inode, new_dentry);
3651        return error;
3652}
3653
3654/*
3655 * Hardlinks are often used in delicate situations.  We avoid
3656 * security-related surprises by not following symlinks on the
3657 * newname.  --KAB
3658 *
3659 * We don't follow them on the oldname either to be compatible
3660 * with linux 2.0, and to avoid hard-linking to directories
3661 * and other special files.  --ADM
3662 */
3663SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
3664                int, newdfd, const char __user *, newname, int, flags)
3665{
3666        struct dentry *new_dentry;
3667        struct path old_path, new_path;
3668        int how = 0;
3669        int error;
3670
3671        if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
3672                return -EINVAL;
3673        /*
3674         * To use null names we require CAP_DAC_READ_SEARCH
3675         * This ensures that not everyone will be able to create
3676         * handlink using the passed filedescriptor.
3677         */
3678        if (flags & AT_EMPTY_PATH) {
3679                if (!capable(CAP_DAC_READ_SEARCH))
3680                        return -ENOENT;
3681                how = LOOKUP_EMPTY;
3682        }
3683
3684        if (flags & AT_SYMLINK_FOLLOW)
3685                how |= LOOKUP_FOLLOW;
3686retry:
3687        error = user_path_at(olddfd, oldname, how, &old_path);
3688        if (error)
3689                return error;
3690
3691        new_dentry = user_path_create(newdfd, newname, &new_path,
3692                                        (how & LOOKUP_REVAL));
3693        error = PTR_ERR(new_dentry);
3694        if (IS_ERR(new_dentry))
3695                goto out;
3696
3697        error = -EXDEV;
3698        if (old_path.mnt != new_path.mnt)
3699                goto out_dput;
3700        error = may_linkat(&old_path);
3701        if (unlikely(error))
3702                goto out_dput;
3703        error = security_path_link(old_path.dentry, &new_path, new_dentry);
3704        if (error)
3705                goto out_dput;
3706        error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry);
3707out_dput:
3708        done_path_create(&new_path, new_dentry);
3709        if (retry_estale(error, how)) {
3710                how |= LOOKUP_REVAL;
3711                goto retry;
3712        }
3713out:
3714        path_put(&old_path);
3715
3716        return error;
3717}
3718
3719SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
3720{
3721        return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
3722}
3723
3724/*
3725 * The worst of all namespace operations - renaming directory. "Perverted"
3726 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3727 * Problems:
3728 *      a) we can get into loop creation. Check is done in is_subdir().
3729 *      b) race potential - two innocent renames can create a loop together.
3730 *         That's where 4.4 screws up. Current fix: serialization on
3731 *         sb->s_vfs_rename_mutex. We might be more accurate, but that's another
3732 *         story.
3733 *      c) we have to lock _three_ objects - parents and victim (if it exists).
3734 *         And that - after we got ->i_mutex on parents (until then we don't know
3735 *         whether the target exists).  Solution: try to be smart with locking
3736 *         order for inodes.  We rely on the fact that tree topology may change
3737 *         only under ->s_vfs_rename_mutex _and_ that parent of the object we
3738 *         move will be locked.  Thus we can rank directories by the tree
3739 *         (ancestors first) and rank all non-directories after them.
3740 *         That works since everybody except rename does "lock parent, lookup,
3741 *         lock child" and rename is under ->s_vfs_rename_mutex.
3742 *         HOWEVER, it relies on the assumption that any object with ->lookup()
3743 *         has no more than 1 dentry.  If "hybrid" objects will ever appear,
3744 *         we'd better make sure that there's no link(2) for them.
3745 *      d) conversion from fhandle to dentry may come in the wrong moment - when
3746 *         we are removing the target. Solution: we will have to grab ->i_mutex
3747 *         in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3748 *         ->i_mutex on parents, which works but leads to some truly excessive
3749 *         locking].
3750 */
3751static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3752                          struct inode *new_dir, struct dentry *new_dentry)
3753{
3754        int error = 0;
3755        struct inode *target = new_dentry->d_inode;
3756        unsigned max_links = new_dir->i_sb->s_max_links;
3757
3758        /*
3759         * If we are going to change the parent - check write permissions,
3760         * we'll need to flip '..'.
3761         */
3762        if (new_dir != old_dir) {
3763                error = inode_permission(old_dentry->d_inode, MAY_WRITE);
3764                if (error)
3765                        return error;
3766        }
3767
3768        error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3769        if (error)
3770                return error;
3771
3772        dget(new_dentry);
3773        if (target)
3774                mutex_lock(&target->i_mutex);
3775
3776        error = -EBUSY;
3777        if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
3778                goto out;
3779
3780        error = -EMLINK;
3781        if (max_links && !target && new_dir != old_dir &&
3782            new_dir->i_nlink >= max_links)
3783                goto out;
3784
3785        if (target)
3786                shrink_dcache_parent(new_dentry);
3787        error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3788        if (error)
3789                goto out;
3790
3791        if (target) {
3792                target->i_flags |= S_DEAD;
3793                dont_mount(new_dentry);
3794        }
3795out:
3796        if (target)
3797                mutex_unlock(&target->i_mutex);
3798        dput(new_dentry);
3799        if (!error)
3800                if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3801                        d_move(old_dentry,new_dentry);
3802        return error;
3803}
3804
3805static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3806                            struct inode *new_dir, struct dentry *new_dentry)
3807{
3808        struct inode *target = new_dentry->d_inode;
3809        int error;
3810
3811        error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3812        if (error)
3813                return error;
3814
3815        dget(new_dentry);
3816        if (target)
3817                mutex_lock(&target->i_mutex);
3818
3819        error = -EBUSY;
3820        if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3821                goto out;
3822
3823        error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3824        if (error)
3825                goto out;
3826
3827        if (target)
3828                dont_mount(new_dentry);
3829        if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3830                d_move(old_dentry, new_dentry);
3831out:
3832        if (target)
3833                mutex_unlock(&target->i_mutex);
3834        dput(new_dentry);
3835        return error;
3836}
3837
3838int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3839               struct inode *new_dir, struct dentry *new_dentry)
3840{
3841        int error;
3842        int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
3843        const unsigned char *old_name;
3844
3845        if (old_dentry->d_inode == new_dentry->d_inode)
3846                return 0;
3847 
3848        error = may_delete(old_dir, old_dentry, is_dir);
3849        if (error)
3850                return error;
3851
3852        if (!new_dentry->d_inode)
3853                error = may_create(new_dir, new_dentry);
3854        else
3855                error = may_delete(new_dir, new_dentry, is_dir);
3856        if (error)
3857                return error;
3858
3859        if (!old_dir->i_op->rename)
3860                return -EPERM;
3861
3862        old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3863
3864        if (is_dir)
3865                error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3866        else
3867                error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
3868        if (!error)
3869                fsnotify_move(old_dir, new_dir, old_name, is_dir,
3870                              new_dentry->d_inode, old_dentry);
3871        fsnotify_oldname_free(old_name);
3872
3873        return error;
3874}
3875
3876SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3877                int, newdfd, const char __user *, newname)
3878{
3879        struct dentry *old_dir, *new_dir;
3880        struct dentry *old_dentry, *new_dentry;
3881        struct dentry *trap;
3882        struct nameidata oldnd, newnd;
3883        struct filename *from;
3884        struct filename *to;
3885        unsigned int lookup_flags = 0;
3886        bool should_retry = false;
3887        int error;
3888retry:
3889        from = user_path_parent(olddfd, oldname, &oldnd, lookup_flags);
3890        if (IS_ERR(from)) {
3891                error = PTR_ERR(from);
3892                goto exit;
3893        }
3894
3895        to = user_path_parent(newdfd, newname, &newnd, lookup_flags);
3896        if (IS_ERR(to)) {
3897                error = PTR_ERR(to);
3898                goto exit1;
3899        }
3900
3901        error = -EXDEV;
3902        if (oldnd.path.mnt != newnd.path.mnt)
3903                goto exit2;
3904
3905        old_dir = oldnd.path.dentry;
3906        error = -EBUSY;
3907        if (oldnd.last_type != LAST_NORM)
3908                goto exit2;
3909
3910        new_dir = newnd.path.dentry;
3911        if (newnd.last_type != LAST_NORM)
3912                goto exit2;
3913
3914        error = mnt_want_write(oldnd.path.mnt);
3915        if (error)
3916                goto exit2;
3917
3918        oldnd.flags &= ~LOOKUP_PARENT;
3919        newnd.flags &= ~LOOKUP_PARENT;
3920        newnd.flags |= LOOKUP_RENAME_TARGET;
3921
3922        trap = lock_rename(new_dir, old_dir);
3923
3924        old_dentry = lookup_hash(&oldnd);
3925        error = PTR_ERR(old_dentry);
3926        if (IS_ERR(old_dentry))
3927                goto exit3;
3928        /* source must exist */
3929        error = -ENOENT;
3930        if (!old_dentry->d_inode)
3931                goto exit4;
3932        /* unless the source is a directory trailing slashes give -ENOTDIR */
3933        if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3934                error = -ENOTDIR;
3935                if (oldnd.last.name[oldnd.last.len])
3936                        goto exit4;
3937                if (newnd.last.name[newnd.last.len])
3938                        goto exit4;
3939        }
3940        /* source should not be ancestor of target */
3941        error = -EINVAL;
3942        if (old_dentry == trap)
3943                goto exit4;
3944        new_dentry = lookup_hash(&newnd);
3945        error = PTR_ERR(new_dentry);
3946        if (IS_ERR(new_dentry))
3947                goto exit4;
3948        /* target should not be an ancestor of source */
3949        error = -ENOTEMPTY;
3950        if (new_dentry == trap)
3951                goto exit5;
3952
3953        error = security_path_rename(&oldnd.path, old_dentry,
3954                                     &newnd.path, new_dentry);
3955        if (error)
3956                goto exit5;
3957        error = vfs_rename(old_dir->d_inode, old_dentry,
3958                                   new_dir->d_inode, new_dentry);
3959exit5:
3960        dput(new_dentry);
3961exit4:
3962        dput(old_dentry);
3963exit3:
3964        unlock_rename(new_dir, old_dir);
3965        mnt_drop_write(oldnd.path.mnt);
3966exit2:
3967        if (retry_estale(error, lookup_flags))
3968                should_retry = true;
3969        path_put(&newnd.path);
3970        putname(to);
3971exit1:
3972        path_put(&oldnd.path);
3973        putname(from);
3974        if (should_retry) {
3975                should_retry = false;
3976                lookup_flags |= LOOKUP_REVAL;
3977                goto retry;
3978        }
3979exit:
3980        return error;
3981}
3982
3983SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
3984{
3985        return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3986}
3987
3988int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3989{
3990        int len;
3991
3992        len = PTR_ERR(link);
3993        if (IS_ERR(link))
3994                goto out;
3995
3996        len = strlen(link);
3997        if (len > (unsigned) buflen)
3998                len = buflen;
3999        if (copy_to_user(buffer, link, len))
4000                len = -EFAULT;
4001out:
4002        return len;
4003}
4004
4005/*
4006 * A helper for ->readlink().  This should be used *ONLY* for symlinks that
4007 * have ->follow_link() touching nd only in nd_set_link().  Using (or not
4008 * using) it for any given inode is up to filesystem.
4009 */
4010int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4011{
4012        struct nameidata nd;
4013        void *cookie;
4014        int res;
4015
4016        nd.depth = 0;
4017        cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
4018        if (IS_ERR(cookie))
4019                return PTR_ERR(cookie);
4020
4021        res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
4022        if (dentry->d_inode->i_op->put_link)
4023                dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
4024        return res;
4025}
4026
4027int vfs_follow_link(struct nameidata *nd, const char *link)
4028{
4029        return __vfs_follow_link(nd, link);
4030}
4031
4032/* get the link contents into pagecache */
4033static char *page_getlink(struct dentry * dentry, struct page **ppage)
4034{
4035        char *kaddr;
4036        struct page *page;
4037        struct address_space *mapping = dentry->d_inode->i_mapping;
4038        page = read_mapping_page(mapping, 0, NULL);
4039        if (IS_ERR(page))
4040                return (char*)page;
4041        *ppage = page;
4042        kaddr = kmap(page);
4043        nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
4044        return kaddr;
4045}
4046
4047int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4048{
4049        struct page *page = NULL;
4050        char *s = page_getlink(dentry, &page);
4051        int res = vfs_readlink(dentry,buffer,buflen,s);
4052        if (page) {
4053                kunmap(page);
4054                page_cache_release(page);
4055        }
4056        return res;
4057}
4058
4059void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
4060{
4061        struct page *page = NULL;
4062        nd_set_link(nd, page_getlink(dentry, &page));
4063        return page;
4064}
4065
4066void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
4067{
4068        struct page *page = cookie;
4069
4070        if (page) {
4071                kunmap(page);
4072                page_cache_release(page);
4073        }
4074}
4075
4076/*
4077 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4078 */
4079int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4080{
4081        struct address_space *mapping = inode->i_mapping;
4082        struct page *page;
4083        void *fsdata;
4084        int err;
4085        char *kaddr;
4086        unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4087        if (nofs)
4088                flags |= AOP_FLAG_NOFS;
4089
4090retry:
4091        err = pagecache_write_begin(NULL, mapping, 0, len-1,
4092                                flags, &page, &fsdata);
4093        if (err)
4094                goto fail;
4095
4096        kaddr = kmap_atomic(page);
4097        memcpy(kaddr, symname, len-1);
4098        kunmap_atomic(kaddr);
4099
4100        err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4101                                                        page, fsdata);
4102        if (err < 0)
4103                goto fail;
4104        if (err < len-1)
4105                goto retry;
4106
4107        mark_inode_dirty(inode);
4108        return 0;
4109fail:
4110        return err;
4111}
4112
4113int page_symlink(struct inode *inode, const char *symname, int len)
4114{
4115        return __page_symlink(inode, symname, len,
4116                        !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
4117}
4118
4119const struct inode_operations page_symlink_inode_operations = {
4120        .readlink       = generic_readlink,
4121        .follow_link    = page_follow_link_light,
4122        .put_link       = page_put_link,
4123};
4124
4125EXPORT_SYMBOL(user_path_at);
4126EXPORT_SYMBOL(follow_down_one);
4127EXPORT_SYMBOL(follow_down);
4128EXPORT_SYMBOL(follow_up);
4129EXPORT_SYMBOL(get_write_access); /* nfsd */
4130EXPORT_SYMBOL(lock_rename);
4131EXPORT_SYMBOL(lookup_one_len);
4132EXPORT_SYMBOL(page_follow_link_light);
4133EXPORT_SYMBOL(page_put_link);
4134EXPORT_SYMBOL(page_readlink);
4135EXPORT_SYMBOL(__page_symlink);
4136EXPORT_SYMBOL(page_symlink);
4137EXPORT_SYMBOL(page_symlink_inode_operations);
4138EXPORT_SYMBOL(kern_path);
4139EXPORT_SYMBOL(vfs_path_lookup);
4140EXPORT_SYMBOL(inode_permission);
4141EXPORT_SYMBOL(unlock_rename);
4142EXPORT_SYMBOL(vfs_create);
4143EXPORT_SYMBOL(vfs_follow_link);
4144EXPORT_SYMBOL(vfs_link);
4145EXPORT_SYMBOL(vfs_mkdir);
4146EXPORT_SYMBOL(vfs_mknod);
4147EXPORT_SYMBOL(generic_permission);
4148EXPORT_SYMBOL(vfs_readlink);
4149EXPORT_SYMBOL(vfs_rename);
4150EXPORT_SYMBOL(vfs_rmdir);
4151EXPORT_SYMBOL(vfs_symlink);
4152EXPORT_SYMBOL(vfs_unlink);
4153EXPORT_SYMBOL(dentry_unhash);
4154EXPORT_SYMBOL(generic_readlink);
4155
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.