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