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 (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1279                        error = d_revalidate(dentry, flags);
1280                        if (unlikely(error <= 0)) {
1281                                if (error < 0) {
1282                                        dput(dentry);
1283                                        return ERR_PTR(error);
1284                                } else if (!d_invalidate(dentry)) {
1285                                        dput(dentry);
1286                                        dentry = NULL;
1287                                }
1288                        }
1289                }
1290        }
1291
1292        if (!dentry) {
1293                dentry = d_alloc(dir, name);
1294                if (unlikely(!dentry))
1295                        return ERR_PTR(-ENOMEM);
1296
1297                *need_lookup = true;
1298        }
1299        return dentry;
1300}
1301
1302/*
1303 * Call i_op->lookup on the dentry.  The dentry must be negative but may be
1304 * hashed if it was pouplated with DCACHE_NEED_LOOKUP.
1305 *
1306 * dir->d_inode->i_mutex must be held
1307 */
1308static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1309                                  unsigned int flags)
1310{
1311        struct dentry *old;
1312
1313        /* Don't create child dentry for a dead directory. */
1314        if (unlikely(IS_DEADDIR(dir))) {
1315                dput(dentry);
1316                return ERR_PTR(-ENOENT);
1317        }
1318
1319        old = dir->i_op->lookup(dir, dentry, flags);
1320        if (unlikely(old)) {
1321                dput(dentry);
1322                dentry = old;
1323        }
1324        return dentry;
1325}
1326
1327static struct dentry *__lookup_hash(struct qstr *name,
1328                struct dentry *base, unsigned int flags)
1329{
1330        bool need_lookup;
1331        struct dentry *dentry;
1332
1333        dentry = lookup_dcache(name, base, flags, &need_lookup);
1334        if (!need_lookup)
1335                return dentry;
1336
1337        return lookup_real(base->d_inode, dentry, flags);
1338}
1339
1340/*
1341 *  It's more convoluted than I'd like it to be, but... it's still fairly
1342 *  small and for now I'd prefer to have fast path as straight as possible.
1343 *  It _is_ time-critical.
1344 */
1345static int lookup_fast(struct nameidata *nd, struct qstr *name,
1346                       struct path *path, struct inode **inode)
1347{
1348        struct vfsmount *mnt = nd->path.mnt;
1349        struct dentry *dentry, *parent = nd->path.dentry;
1350        int need_reval = 1;
1351        int status = 1;
1352        int err;
1353
1354        /*
1355         * Rename seqlock is not required here because in the off chance
1356         * of a false negative due to a concurrent rename, we're going to
1357         * do the non-racy lookup, below.
1358         */
1359        if (nd->flags & LOOKUP_RCU) {
1360                unsigned seq;
1361                dentry = __d_lookup_rcu(parent, name, &seq, nd->inode);
1362                if (!dentry)
1363                        goto unlazy;
1364
1365                /*
1366                 * This sequence count validates that the inode matches
1367                 * the dentry name information from lookup.
1368                 */
1369                *inode = dentry->d_inode;
1370                if (read_seqcount_retry(&dentry->d_seq, seq))
1371                        return -ECHILD;
1372
1373                /*
1374                 * This sequence count validates that the parent had no
1375                 * changes while we did the lookup of the dentry above.
1376                 *
1377                 * The memory barrier in read_seqcount_begin of child is
1378                 *  enough, we can use __read_seqcount_retry here.
1379                 */
1380                if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1381                        return -ECHILD;
1382                nd->seq = seq;
1383
1384                if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1385                        status = d_revalidate(dentry, nd->flags);
1386                        if (unlikely(status <= 0)) {
1387                                if (status != -ECHILD)
1388                                        need_reval = 0;
1389                                goto unlazy;
1390                        }
1391                }
1392                path->mnt = mnt;
1393                path->dentry = dentry;
1394                if (unlikely(!__follow_mount_rcu(nd, path, inode)))
1395                        goto unlazy;
1396                if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1397                        goto unlazy;
1398                return 0;
1399unlazy:
1400                if (unlazy_walk(nd, dentry))
1401                        return -ECHILD;
1402        } else {
1403                dentry = __d_lookup(parent, name);
1404        }
1405
1406        if (unlikely(!dentry))
1407                goto need_lookup;
1408
1409        if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1410                status = d_revalidate(dentry, nd->flags);
1411        if (unlikely(status <= 0)) {
1412                if (status < 0) {
1413                        dput(dentry);
1414                        return status;
1415                }
1416                if (!d_invalidate(dentry)) {
1417                        dput(dentry);
1418                        goto need_lookup;
1419                }
1420        }
1421
1422        path->mnt = mnt;
1423        path->dentry = dentry;
1424        err = follow_managed(path, nd->flags);
1425        if (unlikely(err < 0)) {
1426                path_put_conditional(path, nd);
1427                return err;
1428        }
1429        if (err)
1430                nd->flags |= LOOKUP_JUMPED;
1431        *inode = path->dentry->d_inode;
1432        return 0;
1433
1434need_lookup:
1435        return 1;
1436}
1437
1438/* Fast lookup failed, do it the slow way */
1439static int lookup_slow(struct nameidata *nd, struct qstr *name,
1440                       struct path *path)
1441{
1442        struct dentry *dentry, *parent;
1443        int err;
1444
1445        parent = nd->path.dentry;
1446        BUG_ON(nd->inode != parent->d_inode);
1447
1448        mutex_lock(&parent->d_inode->i_mutex);
1449        dentry = __lookup_hash(name, parent, nd->flags);
1450        mutex_unlock(&parent->d_inode->i_mutex);
1451        if (IS_ERR(dentry))
1452                return PTR_ERR(dentry);
1453        path->mnt = nd->path.mnt;
1454        path->dentry = dentry;
1455        err = follow_managed(path, nd->flags);
1456        if (unlikely(err < 0)) {
1457                path_put_conditional(path, nd);
1458                return err;
1459        }
1460        if (err)
1461                nd->flags |= LOOKUP_JUMPED;
1462        return 0;
1463}
1464
1465static inline int may_lookup(struct nameidata *nd)
1466{
1467        if (nd->flags & LOOKUP_RCU) {
1468                int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1469                if (err != -ECHILD)
1470                        return err;
1471                if (unlazy_walk(nd, NULL))
1472                        return -ECHILD;
1473        }
1474        return inode_permission(nd->inode, MAY_EXEC);
1475}
1476
1477static inline int handle_dots(struct nameidata *nd, int type)
1478{
1479        if (type == LAST_DOTDOT) {
1480                if (nd->flags & LOOKUP_RCU) {
1481                        if (follow_dotdot_rcu(nd))
1482                                return -ECHILD;
1483                } else
1484                        follow_dotdot(nd);
1485        }
1486        return 0;
1487}
1488
1489static void terminate_walk(struct nameidata *nd)
1490{
1491        if (!(nd->flags & LOOKUP_RCU)) {
1492                path_put(&nd->path);
1493        } else {
1494                nd->flags &= ~LOOKUP_RCU;
1495                if (!(nd->flags & LOOKUP_ROOT))
1496                        nd->root.mnt = NULL;
1497                unlock_rcu_walk();
1498        }
1499}
1500
1501/*
1502 * Do we need to follow links? We _really_ want to be able
1503 * to do this check without having to look at inode->i_op,
1504 * so we keep a cache of "no, this doesn't need follow_link"
1505 * for the common case.
1506 */
1507static inline int should_follow_link(struct inode *inode, int follow)
1508{
1509        if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1510                if (likely(inode->i_op->follow_link))
1511                        return follow;
1512
1513                /* This gets set once for the inode lifetime */
1514                spin_lock(&inode->i_lock);
1515                inode->i_opflags |= IOP_NOFOLLOW;
1516                spin_unlock(&inode->i_lock);
1517        }
1518        return 0;
1519}
1520
1521static inline int walk_component(struct nameidata *nd, struct path *path,
1522                struct qstr *name, int type, int follow)
1523{
1524        struct inode *inode;
1525        int err;
1526        /*
1527         * "." and ".." are special - ".." especially so because it has
1528         * to be able to know about the current root directory and
1529         * parent relationships.
1530         */
1531        if (unlikely(type != LAST_NORM))
1532                return handle_dots(nd, type);
1533        err = lookup_fast(nd, name, path, &inode);
1534        if (unlikely(err)) {
1535                if (err < 0)
1536                        goto out_err;
1537
1538                err = lookup_slow(nd, name, path);
1539                if (err < 0)
1540                        goto out_err;
1541
1542                inode = path->dentry->d_inode;
1543        }
1544        err = -ENOENT;
1545        if (!inode)
1546                goto out_path_put;
1547
1548        if (should_follow_link(inode, follow)) {
1549                if (nd->flags & LOOKUP_RCU) {
1550                        if (unlikely(unlazy_walk(nd, path->dentry))) {
1551                                err = -ECHILD;
1552                                goto out_err;
1553                        }
1554                }
1555                BUG_ON(inode != path->dentry->d_inode);
1556                return 1;
1557        }
1558        path_to_nameidata(path, nd);
1559        nd->inode = inode;
1560        return 0;
1561
1562out_path_put:
1563        path_to_nameidata(path, nd);
1564out_err:
1565        terminate_walk(nd);
1566        return err;
1567}
1568
1569/*
1570 * This limits recursive symlink follows to 8, while
1571 * limiting consecutive symlinks to 40.
1572 *
1573 * Without that kind of total limit, nasty chains of consecutive
1574 * symlinks can cause almost arbitrarily long lookups.
1575 */
1576static inline int nested_symlink(struct path *path, struct nameidata *nd)
1577{
1578        int res;
1579
1580        if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1581                path_put_conditional(path, nd);
1582                path_put(&nd->path);
1583                return -ELOOP;
1584        }
1585        BUG_ON(nd->depth >= MAX_NESTED_LINKS);
1586
1587        nd->depth++;
1588        current->link_count++;
1589
1590        do {
1591                struct path link = *path;
1592                void *cookie;
1593
1594                res = follow_link(&link, nd, &cookie);
1595                if (res)
1596                        break;
1597                res = walk_component(nd, path, &nd->last,
1598                                     nd->last_type, LOOKUP_FOLLOW);
1599                put_link(nd, &link, cookie);
1600        } while (res > 0);
1601
1602        current->link_count--;
1603        nd->depth--;
1604        return res;
1605}
1606
1607/*
1608 * We really don't want to look at inode->i_op->lookup
1609 * when we don't have to. So we keep a cache bit in
1610 * the inode ->i_opflags field that says "yes, we can
1611 * do lookup on this inode".
1612 */
1613static inline int can_lookup(struct inode *inode)
1614{
1615        if (likely(inode->i_opflags & IOP_LOOKUP))
1616                return 1;
1617        if (likely(!inode->i_op->lookup))
1618                return 0;
1619
1620        /* We do this once for the lifetime of the inode */
1621        spin_lock(&inode->i_lock);
1622        inode->i_opflags |= IOP_LOOKUP;
1623        spin_unlock(&inode->i_lock);
1624        return 1;
1625}
1626
1627/*
1628 * We can do the critical dentry name comparison and hashing
1629 * operations one word at a time, but we are limited to:
1630 *
1631 * - Architectures with fast unaligned word accesses. We could
1632 *   do a "get_unaligned()" if this helps and is sufficiently
1633 *   fast.
1634 *
1635 * - Little-endian machines (so that we can generate the mask
1636 *   of low bytes efficiently). Again, we *could* do a byte
1637 *   swapping load on big-endian architectures if that is not
1638 *   expensive enough to make the optimization worthless.
1639 *
1640 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1641 *   do not trap on the (extremely unlikely) case of a page
1642 *   crossing operation.
1643 *
1644 * - Furthermore, we need an efficient 64-bit compile for the
1645 *   64-bit case in order to generate the "number of bytes in
1646 *   the final mask". Again, that could be replaced with a
1647 *   efficient population count instruction or similar.
1648 */
1649#ifdef CONFIG_DCACHE_WORD_ACCESS
1650
1651#include <asm/word-at-a-time.h>
1652
1653#ifdef CONFIG_64BIT
1654
1655static inline unsigned int fold_hash(unsigned long hash)
1656{
1657        hash += hash >> (8*sizeof(int));
1658        return hash;
1659}
1660
1661#else   /* 32-bit case */
1662
1663#define fold_hash(x) (x)
1664
1665#endif
1666
1667unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1668{
1669        unsigned long a, mask;
1670        unsigned long hash = 0;
1671
1672        for (;;) {
1673                a = load_unaligned_zeropad(name);
1674                if (len < sizeof(unsigned long))
1675                        break;
1676                hash += a;
1677                hash *= 9;
1678                name += sizeof(unsigned long);
1679                len -= sizeof(unsigned long);
1680                if (!len)
1681                        goto done;
1682        }
1683        mask = ~(~0ul << len*8);
1684        hash += mask & a;
1685done:
1686        return fold_hash(hash);
1687}
1688EXPORT_SYMBOL(full_name_hash);
1689
1690/*
1691 * Calculate the length and hash of the path component, and
1692 * return the length of the component;
1693 */
1694static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1695{
1696        unsigned long a, b, adata, bdata, mask, hash, len;
1697        const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1698
1699        hash = a = 0;
1700        len = -sizeof(unsigned long);
1701        do {
1702                hash = (hash + a) * 9;
1703                len += sizeof(unsigned long);
1704                a = load_unaligned_zeropad(name+len);
1705                b = a ^ REPEAT_BYTE('/');
1706        } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1707
1708        adata = prep_zero_mask(a, adata, &constants);
1709        bdata = prep_zero_mask(b, bdata, &constants);
1710
1711        mask = create_zero_mask(adata | bdata);
1712
1713        hash += a & zero_bytemask(mask);
1714        *hashp = fold_hash(hash);
1715
1716        return len + find_zero(mask);
1717}
1718
1719#else
1720
1721unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1722{
1723        unsigned long hash = init_name_hash();
1724        while (len--)
1725                hash = partial_name_hash(*name++, hash);
1726        return end_name_hash(hash);
1727}
1728EXPORT_SYMBOL(full_name_hash);
1729
1730/*
1731 * We know there's a real path component here of at least
1732 * one character.
1733 */
1734static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1735{
1736        unsigned long hash = init_name_hash();
1737        unsigned long len = 0, c;
1738
1739        c = (unsigned char)*name;
1740        do {
1741                len++;
1742                hash = partial_name_hash(c, hash);
1743                c = (unsigned char)name[len];
1744        } while (c && c != '/');
1745        *hashp = end_name_hash(hash);
1746        return len;
1747}
1748
1749#endif
1750
1751/*
1752 * Name resolution.
1753 * This is the basic name resolution function, turning a pathname into
1754 * the final dentry. We expect 'base' to be positive and a directory.
1755 *
1756 * Returns 0 and nd will have valid dentry and mnt on success.
1757 * Returns error and drops reference to input namei data on failure.
1758 */
1759static int link_path_walk(const char *name, struct nameidata *nd)
1760{
1761        struct path next;
1762        int err;
1763        
1764        while (*name=='/')
1765                name++;
1766        if (!*name)
1767                return 0;
1768
1769        /* At this point we know we have a real path component. */
1770        for(;;) {
1771                struct qstr this;
1772                long len;
1773                int type;
1774
1775                err = may_lookup(nd);
1776                if (err)
1777                        break;
1778
1779                len = hash_name(name, &this.hash);
1780                this.name = name;
1781                this.len = len;
1782
1783                type = LAST_NORM;
1784                if (name[0] == '.') switch (len) {
1785                        case 2:
1786                                if (name[1] == '.') {
1787                                        type = LAST_DOTDOT;
1788                                        nd->flags |= LOOKUP_JUMPED;
1789                                }
1790                                break;
1791                        case 1:
1792                                type = LAST_DOT;
1793                }
1794                if (likely(type == LAST_NORM)) {
1795                        struct dentry *parent = nd->path.dentry;
1796                        nd->flags &= ~LOOKUP_JUMPED;
1797                        if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1798                                err = parent->d_op->d_hash(parent, nd->inode,
1799                                                           &this);
1800                                if (err < 0)
1801                                        break;
1802                        }
1803                }
1804
1805                if (!name[len])
1806                        goto last_component;
1807                /*
1808                 * If it wasn't NUL, we know it was '/'. Skip that
1809                 * slash, and continue until no more slashes.
1810                 */
1811                do {
1812                        len++;
1813                } while (unlikely(name[len] == '/'));
1814                if (!name[len])
1815                        goto last_component;
1816                name += len;
1817
1818                err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW);
1819                if (err < 0)
1820                        return err;
1821
1822                if (err) {
1823                        err = nested_symlink(&next, nd);
1824                        if (err)
1825                                return err;
1826                }
1827                if (can_lookup(nd->inode))
1828                        continue;
1829                err = -ENOTDIR; 
1830                break;
1831                /* here ends the main loop */
1832
1833last_component:
1834                nd->last = this;
1835                nd->last_type = type;
1836                return 0;
1837        }
1838        terminate_walk(nd);
1839        return err;
1840}
1841
1842static int path_init(int dfd, const char *name, unsigned int flags,
1843                     struct nameidata *nd, struct file **fp)
1844{
1845        int retval = 0;
1846
1847        nd->last_type = LAST_ROOT; /* if there are only slashes... */
1848        nd->flags = flags | LOOKUP_JUMPED;
1849        nd->depth = 0;
1850        if (flags & LOOKUP_ROOT) {
1851                struct inode *inode = nd->root.dentry->d_inode;
1852                if (*name) {
1853                        if (!can_lookup(inode))
1854                                return -ENOTDIR;
1855                        retval = inode_permission(inode, MAY_EXEC);
1856                        if (retval)
1857                                return retval;
1858                }
1859                nd->path = nd->root;
1860                nd->inode = inode;
1861                if (flags & LOOKUP_RCU) {
1862                        lock_rcu_walk();
1863                        nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1864                } else {
1865                        path_get(&nd->path);
1866                }
1867                return 0;
1868        }
1869
1870        nd->root.mnt = NULL;
1871
1872        if (*name=='/') {
1873                if (flags & LOOKUP_RCU) {
1874                        lock_rcu_walk();
1875                        set_root_rcu(nd);
1876                } else {
1877                        set_root(nd);
1878                        path_get(&nd->root);
1879                }
1880                nd->path = nd->root;
1881        } else if (dfd == AT_FDCWD) {
1882                if (flags & LOOKUP_RCU) {
1883                        struct fs_struct *fs = current->fs;
1884                        unsigned seq;
1885
1886                        lock_rcu_walk();
1887
1888                        do {
1889                                seq = read_seqcount_begin(&fs->seq);
1890                                nd->path = fs->pwd;
1891                                nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1892                        } while (read_seqcount_retry(&fs->seq, seq));
1893                } else {
1894                        get_fs_pwd(current->fs, &nd->path);
1895                }
1896        } else {
1897                /* Caller must check execute permissions on the starting path component */
1898                struct fd f = fdget_raw(dfd);
1899                struct dentry *dentry;
1900
1901                if (!f.file)
1902                        return -EBADF;
1903
1904                dentry = f.file->f_path.dentry;
1905
1906                if (*name) {
1907                        if (!can_lookup(dentry->d_inode)) {
1908                                fdput(f);
1909                                return -ENOTDIR;
1910                        }
1911                }
1912
1913                nd->path = f.file->f_path;
1914                if (flags & LOOKUP_RCU) {
1915                        if (f.need_put)
1916                                *fp = f.file;
1917                        nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1918                        lock_rcu_walk();
1919                } else {
1920                        path_get(&nd->path);
1921                        fdput(f);
1922                }
1923        }
1924
1925        nd->inode = nd->path.dentry->d_inode;
1926        return 0;
1927}
1928
1929static inline int lookup_last(struct nameidata *nd, struct path *path)
1930{
1931        if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1932                nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1933
1934        nd->flags &= ~LOOKUP_PARENT;
1935        return walk_component(nd, path, &nd->last, nd->last_type,
1936                                        nd->flags & LOOKUP_FOLLOW);
1937}
1938
1939/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1940static int path_lookupat(int dfd, const char *name,
1941                                unsigned int flags, struct nameidata *nd)
1942{
1943        struct file *base = NULL;
1944        struct path path;
1945        int err;
1946
1947        /*
1948         * Path walking is largely split up into 2 different synchronisation
1949         * schemes, rcu-walk and ref-walk (explained in
1950         * Documentation/filesystems/path-lookup.txt). These share much of the
1951         * path walk code, but some things particularly setup, cleanup, and
1952         * following mounts are sufficiently divergent that functions are
1953         * duplicated. Typically there is a function foo(), and its RCU
1954         * analogue, foo_rcu().
1955         *
1956         * -ECHILD is the error number of choice (just to avoid clashes) that
1957         * is returned if some aspect of an rcu-walk fails. Such an error must
1958         * be handled by restarting a traditional ref-walk (which will always
1959         * be able to complete).
1960         */
1961        err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
1962
1963        if (unlikely(err))
1964                return err;
1965
1966        current->total_link_count = 0;
1967        err = link_path_walk(name, nd);
1968
1969        if (!err && !(flags & LOOKUP_PARENT)) {
1970                err = lookup_last(nd, &path);
1971                while (err > 0) {
1972                        void *cookie;
1973                        struct path link = path;
1974                        err = may_follow_link(&link, nd);
1975                        if (unlikely(err))
1976                                break;
1977                        nd->flags |= LOOKUP_PARENT;
1978                        err = follow_link(&link, nd, &cookie);
1979                        if (err)
1980                                break;
1981                        err = lookup_last(nd, &path);
1982                        put_link(nd, &link, cookie);
1983                }
1984        }
1985
1986        if (!err)
1987                err = complete_walk(nd);
1988
1989        if (!err && nd->flags & LOOKUP_DIRECTORY) {
1990                if (!nd->inode->i_op->lookup) {
1991                        path_put(&nd->path);
1992                        err = -ENOTDIR;
1993                }
1994        }
1995
1996        if (base)
1997                fput(base);
1998
1999        if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
2000                path_put(&nd->root);
2001                nd->root.mnt = NULL;
2002        }
2003        return err;
2004}
2005
2006static int filename_lookup(int dfd, struct filename *name,
2007                                unsigned int flags, struct nameidata *nd)
2008{
2009        int retval = path_lookupat(dfd, name->name, flags | LOOKUP_RCU, nd);
2010        if (unlikely(retval == -ECHILD))
2011                retval = path_lookupat(dfd, name->name, flags, nd);
2012        if (unlikely(retval == -ESTALE))
2013                retval = path_lookupat(dfd, name->name,
2014                                                flags | LOOKUP_REVAL, nd);
2015
2016        if (likely(!retval))
2017                audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT);
2018        return retval;
2019}
2020
2021static int do_path_lookup(int dfd, const char *name,
2022                                unsigned int flags, struct nameidata *nd)
2023{
2024        struct filename filename = { .name = name };
2025
2026        return filename_lookup(dfd, &filename, flags, nd);
2027}
2028
2029/* does lookup, returns the object with parent locked */
2030struct dentry *kern_path_locked(const char *name, struct path *path)
2031{
2032        struct nameidata nd;
2033        struct dentry *d;
2034        int err = do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, &nd);
2035        if (err)
2036                return ERR_PTR(err);
2037        if (nd.last_type != LAST_NORM) {
2038                path_put(&nd.path);
2039                return ERR_PTR(-EINVAL);
2040        }
2041        mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2042        d = __lookup_hash(&nd.last, nd.path.dentry, 0);
2043        if (IS_ERR(d)) {
2044                mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2045                path_put(&nd.path);
2046                return d;
2047        }
2048        *path = nd.path;
2049        return d;
2050}
2051
2052int kern_path(const char *name, unsigned int flags, struct path *path)
2053{
2054        struct nameidata nd;
2055        int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
2056        if (!res)
2057                *path = nd.path;
2058        return res;
2059}
2060
2061/**
2062 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2063 * @dentry:  pointer to dentry of the base directory
2064 * @mnt: pointer to vfs mount of the base directory
2065 * @name: pointer to file name
2066 * @flags: lookup flags
2067 * @path: pointer to struct path to fill
2068 */
2069int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2070                    const char *name, unsigned int flags,
2071                    struct path *path)
2072{
2073        struct nameidata nd;
2074        int err;
2075        nd.root.dentry = dentry;
2076        nd.root.mnt = mnt;
2077        BUG_ON(flags & LOOKUP_PARENT);
2078        /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
2079        err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
2080        if (!err)
2081                *path = nd.path;
2082        return err;
2083}
2084
2085/*
2086 * Restricted form of lookup. Doesn't follow links, single-component only,
2087 * needs parent already locked. Doesn't follow mounts.
2088 * SMP-safe.
2089 */
2090static struct dentry *lookup_hash(struct nameidata *nd)
2091{
2092        return __lookup_hash(&nd->last, nd->path.dentry, nd->flags);
2093}
2094
2095/**
2096 * lookup_one_len - filesystem helper to lookup single pathname component
2097 * @name:       pathname component to lookup
2098 * @base:       base directory to lookup from
2099 * @len:        maximum length @len should be interpreted to
2100 *
2101 * Note that this routine is purely a helper for filesystem usage and should
2102 * not be called by generic code.  Also note that by using this function the
2103 * nameidata argument is passed to the filesystem methods and a filesystem
2104 * using this helper needs to be prepared for that.
2105 */
2106struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2107{
2108        struct qstr this;
2109        unsigned int c;
2110        int err;
2111
2112        WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
2113
2114        this.name = name;
2115        this.len = len;
2116        this.hash = full_name_hash(name, len);
2117        if (!len)
2118                return ERR_PTR(-EACCES);
2119
2120        if (unlikely(name[0] == '.')) {
2121                if (len < 2 || (len == 2 && name[1] == '.'))
2122                        return ERR_PTR(-EACCES);
2123        }
2124
2125        while (len--) {
2126                c = *(const unsigned char *)name++;
2127                if (c == '/' || c == '\0')
2128                        return ERR_PTR(-EACCES);
2129        }
2130        /*
2131         * See if the low-level filesystem might want
2132         * to use its own hash..
2133         */
2134        if (base->d_flags & DCACHE_OP_HASH) {
2135                int err = base->d_op->d_hash(base, base->d_inode, &this);
2136                if (err < 0)
2137                        return ERR_PTR(err);
2138        }
2139
2140        err = inode_permission(base->d_inode, MAY_EXEC);
2141        if (err)
2142                return ERR_PTR(err);
2143
2144        return __lookup_hash(&this, base, 0);
2145}
2146
2147int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2148                 struct path *path, int *empty)
2149{
2150        struct nameidata nd;
2151        struct filename *tmp = getname_flags(name, flags, empty);
2152        int err = PTR_ERR(tmp);
2153        if (!IS_ERR(tmp)) {
2154
2155                BUG_ON(flags & LOOKUP_PARENT);
2156
2157                err = filename_lookup(dfd, tmp, flags, &nd);
2158                putname(tmp);
2159                if (!err)
2160                        *path = nd.path;
2161        }
2162        return err;
2163}
2164
2165int user_path_at(int dfd, const char __user *name, unsigned flags,
2166                 struct path *path)
2167{
2168        return user_path_at_empty(dfd, name, flags, path, NULL);
2169}
2170
2171/*
2172 * NB: most callers don't do anything directly with the reference to the
2173 *     to struct filename, but the nd->last pointer points into the name string
2174 *     allocated by getname. So we must hold the reference to it until all
2175 *     path-walking is complete.
2176 */
2177static struct filename *
2178user_path_parent(int dfd, const char __user *path, struct nameidata *nd,
2179                 unsigned int flags)
2180{
2181        struct filename *s = getname(path);
2182        int error;
2183
2184        /* only LOOKUP_REVAL is allowed in extra flags */
2185        flags &= LOOKUP_REVAL;
2186
2187        if (IS_ERR(s))
2188                return s;
2189
2190        error = filename_lookup(dfd, s, flags | LOOKUP_PARENT, nd);
2191        if (error) {
2192                putname(s);
2193                return ERR_PTR(error);
2194        }
2195
2196        return s;
2197}
2198
2199/*
2200 * It's inline, so penalty for filesystems that don't use sticky bit is
2201 * minimal.
2202 */
2203static inline int check_sticky(struct inode *dir, struct inode *inode)
2204{
2205        kuid_t fsuid = current_fsuid();
2206
2207        if (!(dir->i_mode & S_ISVTX))
2208                return 0;
2209        if (uid_eq(inode->i_uid, fsuid))
2210                return 0;
2211        if (uid_eq(dir->i_uid, fsuid))
2212                return 0;
2213        return !inode_capable(inode, CAP_FOWNER);
2214}
2215
2216/*
2217 *      Check whether we can remove a link victim from directory dir, check
2218 *  whether the type of victim is right.
2219 *  1. We can't do it if dir is read-only (done in permission())
2220 *  2. We should have write and exec permissions on dir
2221 *  3. We can't remove anything from append-only dir
2222 *  4. We can't do anything with immutable dir (done in permission())
2223 *  5. If the sticky bit on dir is set we should either
2224 *      a. be owner of dir, or
2225 *      b. be owner of victim, or
2226 *      c. have CAP_FOWNER capability
2227 *  6. If the victim is append-only or immutable we can't do antyhing with
2228 *     links pointing to it.
2229 *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2230 *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2231 *  9. We can't remove a root or mountpoint.
2232 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2233 *     nfs_async_unlink().
2234 */
2235static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
2236{
2237        int error;
2238
2239        if (!victim->d_inode)
2240                return -ENOENT;
2241
2242        BUG_ON(victim->d_parent->d_inode != dir);
2243        audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2244
2245        error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2246        if (error)
2247                return error;
2248        if (IS_APPEND(dir))
2249                return -EPERM;
2250        if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
2251            IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
2252                return -EPERM;
2253        if (isdir) {
2254                if (!S_ISDIR(victim->d_inode->i_mode))
2255                        return -ENOTDIR;
2256                if (IS_ROOT(victim))
2257                        return -EBUSY;
2258        } else if (S_ISDIR(victim->d_inode->i_mode))
2259                return -EISDIR;
2260        if (IS_DEADDIR(dir))
2261                return -ENOENT;
2262        if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2263                return -EBUSY;
2264        return 0;
2265}
2266
2267/*      Check whether we can create an object with dentry child in directory
2268 *  dir.
2269 *  1. We can't do it if child already exists (open has special treatment for
2270 *     this case, but since we are inlined it's OK)
2271 *  2. We can't do it if dir is read-only (done in permission())
2272 *  3. We should have write and exec permissions on dir
2273 *  4. We can't do it if dir is immutable (done in permission())
2274 */
2275static inline int may_create(struct inode *dir, struct dentry *child)
2276{
2277        if (child->d_inode)
2278                return -EEXIST;
2279        if (IS_DEADDIR(dir))
2280                return -ENOENT;
2281        return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2282}
2283
2284/*
2285 * p1 and p2 should be directories on the same fs.
2286 */
2287struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2288{
2289        struct dentry *p;
2290
2291        if (p1 == p2) {
2292                mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2293                return NULL;
2294        }
2295
2296        mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2297
2298        p = d_ancestor(p2, p1);
2299        if (p) {
2300                mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2301                mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2302                return p;
2303        }
2304
2305        p = d_ancestor(p1, p2);
2306        if (p) {
2307                mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2308                mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2309                return p;
2310        }
2311
2312        mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2313        mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2314        return NULL;
2315}
2316
2317void unlock_rename(struct dentry *p1, struct dentry *p2)
2318{
2319        mutex_unlock(&p1->d_inode->i_mutex);
2320        if (p1 != p2) {
2321                mutex_unlock(&p2->d_inode->i_mutex);
2322                mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2323        }
2324}
2325
2326int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2327                bool want_excl)
2328{
2329        int error = may_create(dir, dentry);
2330        if (error)
2331                return error;
2332
2333        if (!dir->i_op->create)
2334                return -EACCES; /* shouldn't it be ENOSYS? */
2335        mode &= S_IALLUGO;
2336        mode |= S_IFREG;
2337        error = security_inode_create(dir, dentry, mode);
2338        if (error)
2339                return error;
2340        error = dir->i_op->create(dir, dentry, mode, want_excl);
2341        if (!error)
2342                fsnotify_create(dir, dentry);
2343        return error;
2344}
2345
2346static int may_open(struct path *path, int acc_mode, int flag)
2347{
2348        struct dentry *dentry = path->dentry;
2349        struct inode *inode = dentry->d_inode;
2350        int error;
2351
2352        /* O_PATH? */
2353        if (!acc_mode)
2354                return 0;
2355
2356        if (!inode)
2357                return -ENOENT;
2358
2359        switch (inode->i_mode & S_IFMT) {
2360        case S_IFLNK:
2361                return -ELOOP;
2362        case S_IFDIR:
2363                if (acc_mode & MAY_WRITE)
2364                        return -EISDIR;
2365                break;
2366        case S_IFBLK:
2367        case S_IFCHR:
2368                if (path->mnt->mnt_flags & MNT_NODEV)
2369                        return -EACCES;
2370                /*FALLTHRU*/
2371        case S_IFIFO:
2372        case S_IFSOCK:
2373                flag &= ~O_TRUNC;
2374                break;
2375        }
2376
2377        error = inode_permission(inode, acc_mode);
2378        if (error)
2379                return error;
2380
2381        /*
2382         * An append-only file must be opened in append mode for writing.
2383         */
2384        if (IS_APPEND(inode)) {
2385                if  ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2386                        return -EPERM;
2387                if (flag & O_TRUNC)
2388                        return -EPERM;
2389        }
2390
2391        /* O_NOATIME can only be set by the owner or superuser */
2392        if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2393                return -EPERM;
2394
2395        return 0;
2396}
2397
2398static int handle_truncate(struct file *filp)
2399{
2400        struct path *path = &filp->f_path;
2401        struct inode *inode = path->dentry->d_inode;
2402        int error = get_write_access(inode);
2403        if (error)
2404                return error;
2405        /*
2406         * Refuse to truncate files with mandatory locks held on them.
2407         */
2408        error = locks_verify_locked(inode);
2409        if (!error)
2410                error = security_path_truncate(path);
2411        if (!error) {
2412                error = do_truncate(path->dentry, 0,
2413                                    ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2414                                    filp);
2415        }
2416        put_write_access(inode);
2417        return error;
2418}
2419
2420static inline int open_to_namei_flags(int flag)
2421{
2422        if ((flag & O_ACCMODE) == 3)
2423                flag--;
2424        return flag;
2425}
2426
2427static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2428{
2429        int error = security_path_mknod(dir, dentry, mode, 0);
2430        if (error)
2431                return error;
2432
2433        error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2434        if (error)
2435                return error;
2436
2437        return security_inode_create(dir->dentry->d_inode, dentry, mode);
2438}
2439
2440/*
2441 * Attempt to atomically look up, create and open a file from a negative
2442 * dentry.
2443 *
2444 * Returns 0 if successful.  The file will have been created and attached to
2445 * @file by the filesystem calling finish_open().
2446 *
2447 * Returns 1 if the file was looked up only or didn't need creating.  The
2448 * caller will need to perform the open themselves.  @path will have been
2449 * updated to point to the new dentry.  This may be negative.
2450 *
2451 * Returns an error code otherwise.
2452 */
2453static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2454                        struct path *path, struct file *file,
2455                        const struct open_flags *op,
2456                        bool got_write, bool need_lookup,
2457                        int *opened)
2458{
2459        struct inode *dir =  nd->path.dentry->d_inode;
2460        unsigned open_flag = open_to_namei_flags(op->open_flag);
2461        umode_t mode;
2462        int error;
2463        int acc_mode;
2464        int create_error = 0;
2465        struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2466
2467        BUG_ON(dentry->d_inode);
2468
2469        /* Don't create child dentry for a dead directory. */
2470        if (unlikely(IS_DEADDIR(dir))) {
2471                error = -ENOENT;
2472                goto out;
2473        }
2474
2475        mode = op->mode;
2476        if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2477                mode &= ~current_umask();
2478
2479        if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT)) {
2480                open_flag &= ~O_TRUNC;
2481                *opened |= FILE_CREATED;
2482        }
2483
2484        /*
2485         * Checking write permission is tricky, bacuse we don't know if we are
2486         * going to actually need it: O_CREAT opens should work as long as the
2487         * file exists.  But checking existence breaks atomicity.  The trick is
2488         * to check access and if not granted clear O_CREAT from the flags.
2489         *
2490         * Another problem is returing the "right" error value (e.g. for an
2491         * O_EXCL open we want to return EEXIST not EROFS).
2492         */
2493        if (((open_flag & (O_CREAT | O_TRUNC)) ||
2494            (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
2495                if (!(open_flag & O_CREAT)) {
2496                        /*
2497                         * No O_CREATE -> atomicity not a requirement -> fall
2498                         * back to lookup + open
2499                         */
2500                        goto no_open;
2501                } else if (open_flag & (O_EXCL | O_TRUNC)) {
2502                        /* Fall back and fail with the right error */
2503                        create_error = -EROFS;
2504                        goto no_open;
2505                } else {
2506                        /* No side effects, safe to clear O_CREAT */
2507                        create_error = -EROFS;
2508                        open_flag &= ~O_CREAT;
2509                }
2510        }
2511
2512        if (open_flag & O_CREAT) {
2513                error = may_o_create(&nd->path, dentry, mode);
2514                if (error) {
2515                        create_error = error;
2516                        if (open_flag & O_EXCL)
2517                                goto no_open;
2518                        open_flag &= ~O_CREAT;
2519                }
2520        }
2521
2522        if (nd->flags & LOOKUP_DIRECTORY)
2523                open_flag |= O_DIRECTORY;
2524
2525        file->f_path.dentry = DENTRY_NOT_SET;
2526        file->f_path.mnt = nd->path.mnt;
2527        error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
2528                                      opened);
2529        if (error < 0) {
2530                if (create_error && error == -ENOENT)
2531                        error = create_error;
2532                goto out;
2533        }
2534
2535        acc_mode = op->acc_mode;
2536        if (*opened & FILE_CREATED) {
2537                fsnotify_create(dir, dentry);
2538                acc_mode = MAY_OPEN;
2539        }
2540
2541        if (error) {    /* returned 1, that is */
2542                if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
2543                        error = -EIO;
2544                        goto out;
2545                }
2546                if (file->f_path.dentry) {
2547                        dput(dentry);
2548                        dentry = file->f_path.dentry;
2549                }
2550                if (create_error && dentry->d_inode == NULL) {
2551                        error = create_error;
2552                        goto out;
2553                }
2554                goto looked_up;
2555        }
2556
2557        /*
2558         * We didn't have the inode before the open, so check open permission
2559         * here.
2560         */
2561        error = may_open(&file->f_path, acc_mode, open_flag);
2562        if (error)
2563                fput(file);
2564
2565out:
2566        dput(dentry);
2567        return error;
2568
2569no_open:
2570        if (need_lookup) {
2571                dentry = lookup_real(dir, dentry, nd->flags);
2572                if (IS_ERR(dentry))
2573                        return PTR_ERR(dentry);
2574
2575                if (create_error) {
2576                        int open_flag = op->open_flag;
2577
2578                        error = create_error;
2579                        if ((open_flag & O_EXCL)) {
2580                                if (!dentry->d_inode)
2581                                        goto out;
2582                        } else if (!dentry->d_inode) {
2583                                goto out;
2584                        } else if ((open_flag & O_TRUNC) &&
2585                                   S_ISREG(dentry->d_inode->i_mode)) {
2586                                goto out;
2587                        }
2588                        /* will fail later, go on to get the right error */
2589                }
2590        }
2591looked_up:
2592        path->dentry = dentry;
2593        path->mnt = nd->path.mnt;
2594        return 1;
2595}
2596
2597/*
2598 * Look up and maybe create and open the last component.
2599 *
2600 * Must be called with i_mutex held on parent.
2601 *
2602 * Returns 0 if the file was successfully atomically created (if necessary) and
2603 * opened.  In this case the file will be returned attached to @file.
2604 *
2605 * Returns 1 if the file was not completely opened at this time, though lookups
2606 * and creations will have been performed and the dentry returned in @path will
2607 * be positive upon return if O_CREAT was specified.  If O_CREAT wasn't
2608 * specified then a negative dentry may be returned.
2609 *
2610 * An error code is returned otherwise.
2611 *
2612 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2613 * cleared otherwise prior to returning.
2614 */
2615static int lookup_open(struct nameidata *nd, struct path *path,
2616                        struct file *file,
2617                        const struct open_flags *op,
2618                        bool got_write, int *opened)
2619{
2620        struct dentry *dir = nd->path.dentry;
2621        struct inode *dir_inode = dir->d_inode;
2622        struct dentry *dentry;
2623        int error;
2624        bool need_lookup;
2625
2626        *opened &= ~FILE_CREATED;
2627        dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
2628        if (IS_ERR(dentry))
2629                return PTR_ERR(dentry);
2630
2631        /* Cached positive dentry: will open in f_op->open */
2632        if (!need_lookup && dentry->d_inode)
2633                goto out_no_open;
2634
2635        if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
2636                return atomic_open(nd, dentry, path, file, op, got_write,
2637                                   need_lookup, opened);
2638        }
2639
2640        if (need_lookup) {
2641                BUG_ON(dentry->d_inode);
2642
2643                dentry = lookup_real(dir_inode, dentry, nd->flags);
2644                if (IS_ERR(dentry))
2645                        return PTR_ERR(dentry);
2646        }
2647
2648        /* Negative dentry, just create the file */
2649        if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
2650                umode_t mode = op->mode;
2651                if (!IS_POSIXACL(dir->d_inode))
2652                        mode &= ~current_umask();
2653                /*
2654                 * This write is needed to ensure that a
2655                 * rw->ro transition does not occur between
2656                 * the time when the file is created and when
2657                 * a permanent write count is taken through
2658                 * the 'struct file' in finish_open().
2659                 */
2660                if (!got_write) {
2661                        error = -EROFS;
2662                        goto out_dput;
2663                }
2664                *opened |= FILE_CREATED;
2665                error = security_path_mknod(&nd->path, dentry, mode, 0);
2666                if (error)
2667                        goto out_dput;
2668                error = vfs_create(dir->d_inode, dentry, mode,
2669                                   nd->flags & LOOKUP_EXCL);
2670                if (error)
2671                        goto out_dput;
2672        }
2673out_no_open:
2674        path->dentry = dentry;
2675        path->mnt = nd->path.mnt;
2676        return 1;
2677
2678out_dput:
2679        dput(dentry);
2680        return error;
2681}
2682
2683/*
2684 * Handle the last step of open()
2685 */
2686static int do_last(struct nameidata *nd, struct path *path,
2687                   struct file *file, const struct open_flags *op,
2688                   int *opened, struct filename *name)
2689{
2690        struct dentry *dir = nd->path.dentry;
2691        int open_flag = op->open_flag;
2692        bool will_truncate = (open_flag & O_TRUNC) != 0;
2693        bool got_write = false;
2694        int acc_mode = op->acc_mode;
2695        struct inode *inode;
2696        bool symlink_ok = false;
2697        struct path save_parent = { .dentry = NULL, .mnt = NULL };
2698        bool retried = false;
2699        int error;
2700
2701        nd->flags &= ~LOOKUP_PARENT;
2702        nd->flags |= op->intent;
2703
2704        switch (nd->last_type) {
2705        case LAST_DOTDOT:
2706        case LAST_DOT:
2707                error = handle_dots(nd, nd->last_type);
2708                if (error)
2709                        return error;
2710                /* fallthrough */
2711        case LAST_ROOT:
2712                error = complete_walk(nd);
2713                if (error)
2714                        return error;
2715                audit_inode(name, nd->path.dentry, 0);
2716                if (open_flag & O_CREAT) {
2717                        error = -EISDIR;
2718                        goto out;
2719                }
2720                goto finish_open;
2721        case LAST_BIND:
2722                error = complete_walk(nd);
2723                if (error)
2724                        return error;
2725                audit_inode(name, dir, 0);
2726                goto finish_open;
2727        }
2728
2729        if (!(open_flag & O_CREAT)) {
2730                if (nd->last.name[nd->last.len])
2731                        nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2732                if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2733                        symlink_ok = true;
2734                /* we _can_ be in RCU mode here */
2735                error = lookup_fast(nd, &nd->last, path, &inode);
2736                if (likely(!error))
2737                        goto finish_lookup;
2738
2739                if (error < 0)
2740                        goto out;
2741
2742                BUG_ON(nd->inode != dir->d_inode);
2743        } else {
2744                /* create side of things */
2745                /*
2746                 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2747                 * has been cleared when we got to the last component we are
2748                 * about to look up
2749                 */
2750                error = complete_walk(nd);
2751                if (error)
2752                        return error;
2753
2754                audit_inode(name, dir, 0);
2755                error = -EISDIR;
2756                /* trailing slashes? */
2757                if (nd->last.name[nd->last.len])
2758                        goto out;
2759        }
2760
2761retry_lookup:
2762        if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
2763                error = mnt_want_write(nd->path.mnt);
2764                if (!error)
2765                        got_write = true;
2766                /*
2767                 * do _not_ fail yet - we might not need that or fail with
2768                 * a different error; let lookup_open() decide; we'll be
2769                 * dropping this one anyway.
2770                 */
2771        }
2772        mutex_lock(&dir->d_inode->i_mutex);
2773        error = lookup_open(nd, path, file, op, got_write, opened);
2774        mutex_unlock(&dir->d_inode->i_mutex);
2775
2776        if (error <= 0) {
2777                if (error)
2778                        goto out;
2779
2780                if ((*opened & FILE_CREATED) ||
2781                    !S_ISREG(file->f_path.dentry->d_inode->i_mode))
2782                        will_truncate = false;
2783
2784                audit_inode(name, file->f_path.dentry, 0);
2785                goto opened;
2786        }
2787
2788        if (*opened & FILE_CREATED) {
2789                /* Don't check for write permission, don't truncate */
2790                open_flag &= ~O_TRUNC;
2791                will_truncate = false;
2792                acc_mode = MAY_OPEN;
2793                path_to_nameidata(path, nd);
2794                goto finish_open_created;
2795        }
2796
2797        /*
2798         * create/update audit record if it already exists.
2799         */
2800        if (path->dentry->d_inode)
2801                audit_inode(name, path->dentry, 0);
2802
2803        /*
2804         * If atomic_open() acquired write access it is dropped now due to
2805         * possible mount and symlink following (this might be optimized away if
2806         * necessary...)
2807         */
2808        if (got_write) {
2809                mnt_drop_write(nd->path.mnt);
2810                got_write = false;
2811        }
2812
2813        error = -EEXIST;
2814        if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))
2815                goto exit_dput;
2816
2817        error = follow_managed(path, nd->flags);
2818        if (error < 0)
2819                goto exit_dput;
2820
2821        if (error)
2822                nd->flags |= LOOKUP_JUMPED;
2823
2824        BUG_ON(nd->flags & LOOKUP_RCU);
2825        inode = path->dentry->d_inode;
2826finish_lookup:
2827        /* we _can_ be in RCU mode here */
2828        error = -ENOENT;
2829        if (!inode) {
2830                path_to_nameidata(path, nd);
2831                goto out;
2832        }
2833
2834        if (should_follow_link(inode, !symlink_ok)) {
2835                if (nd->flags & LOOKUP_RCU) {
2836                        if (unlikely(unlazy_walk(nd, path->dentry))) {
2837                                error = -ECHILD;
2838                                goto out;
2839                        }
2840                }
2841                BUG_ON(inode != path->dentry->d_inode);
2842                return 1;
2843        }
2844
2845        if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) {
2846                path_to_nameidata(path, nd);
2847        } else {
2848                save_parent.dentry = nd->path.dentry;
2849                save_parent.mnt = mntget(path->mnt);
2850                nd->path.dentry = path->dentry;
2851
2852        }
2853        nd->inode = inode;
2854        /* Why this, you ask?  _Now_ we might have grown LOOKUP_JUMPED... */
2855        error = complete_walk(nd);
2856        if (error) {
2857                path_put(&save_parent);
2858                return error;
2859        }
2860        error = -EISDIR;
2861        if ((open_flag & O_CREAT) && S_ISDIR(nd->inode->i_mode))
2862                goto out;
2863        error = -ENOTDIR;
2864        if ((nd->flags & LOOKUP_DIRECTORY) && !nd->inode->i_op->lookup)
2865                goto out;
2866        audit_inode(name, nd->path.dentry, 0);
2867finish_open:
2868        if (!S_ISREG(nd->inode->i_mode))
2869                will_truncate = false;
2870
2871        if (will_truncate) {
2872                error = mnt_want_write(nd->path.mnt);
2873                if (error)
2874                        goto out;
2875                got_write = true;
2876        }
2877finish_open_created:
2878        error = may_open(&nd->path, acc_mode, open_flag);
2879        if (error)
2880                goto out;
2881        file->f_path.mnt = nd->path.mnt;
2882        error = finish_open(file, nd->path.dentry, NULL, opened);
2883        if (error) {
2884                if (error == -EOPENSTALE)
2885                        goto stale_open;
2886                goto out;
2887        }
2888opened:
2889        error = open_check_o_direct(file);
2890        if (error)
2891                goto exit_fput;
2892        error = ima_file_check(file, op->acc_mode);
2893        if (error)
2894                goto exit_fput;
2895
2896        if (will_truncate) {
2897                error = handle_truncate(file);
2898                if (error)
2899                        goto exit_fput;
2900        }
2901out:
2902        if (got_write)
2903                mnt_drop_write(nd->path.mnt);
2904        path_put(&save_parent);
2905        terminate_walk(nd);
2906        return error;
2907
2908exit_dput:
2909        path_put_conditional(path, nd);
2910        goto out;
2911exit_fput:
2912        fput(file);
2913        goto out;
2914
2915stale_open:
2916        /* If no saved parent or already retried then can't retry */
2917        if (!save_parent.dentry || retried)
2918                goto out;
2919
2920        BUG_ON(save_parent.dentry != dir);
2921        path_put(&nd->path);
2922        nd->path = save_parent;
2923        nd->inode = dir->d_inode;
2924        save_parent.mnt = NULL;
2925        save_parent.dentry = NULL;
2926        if (got_write) {
2927                mnt_drop_write(nd->path.mnt);
2928                got_write = false;
2929        }
2930        retried = true;
2931        goto retry_lookup;
2932}
2933
2934static struct file *path_openat(int dfd, struct filename *pathname,
2935                struct nameidata *nd, const struct open_flags *op, int flags)
2936{
2937        struct file *base = NULL;
2938        struct file *file;
2939        struct path path;
2940        int opened = 0;
2941        int error;
2942
2943        file = get_empty_filp();
2944        if (!file)
2945                return ERR_PTR(-ENFILE);
2946
2947        file->f_flags = op->open_flag;
2948
2949        error = path_init(dfd, pathname->name, flags | LOOKUP_PARENT, nd, &base);
2950        if (unlikely(error))
2951                goto out;
2952
2953        current->total_link_count = 0;
2954        error = link_path_walk(pathname->name, nd);
2955        if (unlikely(error))
2956                goto out;
2957
2958        error = do_last(nd, &path, file, op, &opened, pathname);
2959        while (unlikely(error > 0)) { /* trailing symlink */
2960                struct path link = path;
2961                void *cookie;
2962                if (!(nd->flags & LOOKUP_FOLLOW)) {
2963                        path_put_conditional(&path, nd);
2964                        path_put(&nd->path);
2965                        error = -ELOOP;
2966                        break;
2967                }
2968                error = may_follow_link(&link, nd);
2969                if (unlikely(error))
2970                        break;
2971                nd->flags |= LOOKUP_PARENT;
2972                nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
2973                error = follow_link(&link, nd, &cookie);
2974                if (unlikely(error))
2975                        break;
2976                error = do_last(nd, &path, file, op, &opened, pathname);
2977                put_link(nd, &link, cookie);
2978        }
2979out:
2980        if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2981                path_put(&nd->root);
2982        if (base)
2983                fput(base);
2984        if (!(opened & FILE_OPENED)) {
2985                BUG_ON(!error);
2986                put_filp(file);
2987        }
2988        if (unlikely(error)) {
2989                if (error == -EOPENSTALE) {
2990                        if (flags & LOOKUP_RCU)
2991                                error = -ECHILD;
2992                        else
2993                                error = -ESTALE;
2994                }
2995                file = ERR_PTR(error);
2996        }
2997        return file;
2998}
2999
3000struct file *do_filp_open(int dfd, struct filename *pathname,
3001                const struct open_flags *op, int flags)
3002{
3003        struct nameidata nd;
3004        struct file *filp;
3005
3006        filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
3007        if (unlikely(filp == ERR_PTR(-ECHILD)))
3008                filp = path_openat(dfd, pathname, &nd, op, flags);
3009        if (unlikely(filp == ERR_PTR(-ESTALE)))
3010                filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
3011        return filp;
3012}
3013
3014struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3015                const char *name, const struct open_flags *op, int flags)
3016{
3017        struct nameidata nd;
3018        struct file *file;
3019        struct filename filename = { .name = name };
3020
3021        nd.root.mnt = mnt;
3022        nd.root.dentry = dentry;
3023
3024        flags |= LOOKUP_ROOT;
3025
3026        if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
3027                return ERR_PTR(-ELOOP);
3028
3029        file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_RCU);
3030        if (unlikely(file == ERR_PTR(-ECHILD)))
3031                file = path_openat(-1, &filename, &nd, op, flags);
3032        if (unlikely(file == ERR_PTR(-ESTALE)))
3033                file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_REVAL);
3034        return file;
3035}
3036
3037struct dentry *kern_path_create(int dfd, const char *pathname,
3038                                struct path *path, unsigned int lookup_flags)
3039{
3040        struct dentry *dentry = ERR_PTR(-EEXIST);
3041        struct nameidata nd;
3042        int err2;
3043        int error;
3044        bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3045
3046        /*
3047         * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3048         * other flags passed in are ignored!
3049         */
3050        lookup_flags &= LOOKUP_REVAL;
3051
3052        error = do_path_lookup(dfd, pathname, LOOKUP_PARENT|lookup_flags, &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,
3117                                struct path *path, unsigned int lookup_flags)
3118{
3119        struct filename *tmp = getname(pathname);
3120        struct dentry *res;
3121        if (IS_ERR(tmp))
3122                return ERR_CAST(tmp);
3123        res = kern_path_create(dfd, tmp->name, path, lookup_flags);
3124        putname(tmp);
3125        return res;
3126}
3127EXPORT_SYMBOL(user_path_create);
3128
3129int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3130{
3131        int error = may_create(dir, dentry);
3132
3133        if (error)
3134                return error;
3135
3136        if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3137                return -EPERM;
3138
3139        if (!dir->i_op->mknod)
3140                return -EPERM;
3141
3142        error = devcgroup_inode_mknod(mode, dev);
3143        if (error)
3144                return error;
3145
3146        error = security_inode_mknod(dir, dentry, mode, dev);
3147        if (error)
3148                return error;
3149
3150        error = dir->i_op->mknod(dir, dentry, mode, dev);
3151        if (!error)
3152                fsnotify_create(dir, dentry);
3153        return error;
3154}
3155
3156static int may_mknod(umode_t mode)
3157{
3158        switch (mode & S_IFMT) {
3159        case S_IFREG:
3160        case S_IFCHR:
3161        case S_IFBLK:
3162        case S_IFIFO:
3163        case S_IFSOCK:
3164        case 0: /* zero mode translates to S_IFREG */
3165                return 0;
3166        case S_IFDIR:
3167                return -EPERM;
3168        default:
3169                return -EINVAL;
3170        }
3171}
3172
3173SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3174                unsigned, dev)
3175{
3176        struct dentry *dentry;
3177        struct path path;
3178        int error;
3179        unsigned int lookup_flags = 0;
3180
3181        error = may_mknod(mode);
3182        if (error)
3183                return error;
3184retry:
3185        dentry = user_path_create(dfd, filename, &path, lookup_flags);
3186        if (IS_ERR(dentry))
3187                return PTR_ERR(dentry);
3188
3189        if (!IS_POSIXACL(path.dentry->d_inode))
3190                mode &= ~current_umask();
3191        error = security_path_mknod(&path, dentry, mode, dev);
3192        if (error)
3193                goto out;
3194        switch (mode & S_IFMT) {
3195                case 0: case S_IFREG:
3196                        error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3197                        break;
3198                case S_IFCHR: case S_IFBLK:
3199                        error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3200                                        new_decode_dev(dev));
3201                        break;
3202                case S_IFIFO: case S_IFSOCK:
3203                        error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3204                        break;
3205        }
3206out:
3207        done_path_create(&path, dentry);
3208        if (retry_estale(error, lookup_flags)) {
3209                lookup_flags |= LOOKUP_REVAL;
3210                goto retry;
3211        }
3212        return error;
3213}
3214
3215SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3216{
3217        return sys_mknodat(AT_FDCWD, filename, mode, dev);
3218}
3219
3220int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3221{
3222        int error = may_create(dir, dentry);
3223        unsigned max_links = dir->i_sb->s_max_links;
3224
3225        if (error)
3226                return error;
3227
3228        if (!dir->i_op->mkdir)
3229                return -EPERM;
3230
3231        mode &= (S_IRWXUGO|S_ISVTX);
3232        error = security_inode_mkdir(dir, dentry, mode);
3233        if (error)
3234                return error;
3235
3236        if (max_links && dir->i_nlink >= max_links)
3237                return -EMLINK;
3238
3239        error = dir->i_op->mkdir(dir, dentry, mode);
3240        if (!error)
3241                fsnotify_mkdir(dir, dentry);
3242        return error;
3243}
3244
3245SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3246{
3247        struct dentry *dentry;
3248        struct path path;
3249        int error;
3250        unsigned int lookup_flags = LOOKUP_DIRECTORY;
3251
3252retry:
3253        dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3254        if (IS_ERR(dentry))
3255                return PTR_ERR(dentry);
3256
3257        if (!IS_POSIXACL(path.dentry->d_inode))
3258                mode &= ~current_umask();
3259        error = security_path_mkdir(&path, dentry, mode);
3260        if (!error)
3261                error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3262        done_path_create(&path, dentry);
3263        if (retry_estale(error, lookup_flags)) {
3264                lookup_flags |= LOOKUP_REVAL;
3265                goto retry;
3266        }
3267        return error;
3268}
3269
3270SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3271{
3272        return sys_mkdirat(AT_FDCWD, pathname, mode);
3273}
3274
3275/*
3276 * The dentry_unhash() helper will try to drop the dentry early: we
3277 * should have a usage count of 1 if we're the only user of this
3278 * dentry, and if that is true (possibly after pruning the dcache),
3279 * then we drop the dentry now.
3280 *
3281 * A low-level filesystem can, if it choses, legally
3282 * do a
3283 *
3284 *      if (!d_unhashed(dentry))
3285 *              return -EBUSY;
3286 *
3287 * if it cannot handle the case of removing a directory
3288 * that is still in use by something else..
3289 */
3290void dentry_unhash(struct dentry *dentry)
3291{
3292        shrink_dcache_parent(dentry);
3293        spin_lock(&dentry->d_lock);
3294        if (dentry->d_count == 1)
3295                __d_drop(dentry);
3296        spin_unlock(&dentry->d_lock);
3297}
3298
3299int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3300{
3301        int error = may_delete(dir, dentry, 1);
3302
3303        if (error)
3304                return error;
3305
3306        if (!dir->i_op->rmdir)
3307                return -EPERM;
3308
3309        dget(dentry);
3310        mutex_lock(&dentry->d_inode->i_mutex);
3311
3312        error = -EBUSY;
3313        if (d_mountpoint(dentry))
3314                goto out;
3315
3316        error = security_inode_rmdir(dir, dentry);
3317        if (error)
3318                goto out;
3319
3320        shrink_dcache_parent(dentry);
3321        error = dir->i_op->rmdir(dir, dentry);
3322        if (error)
3323                goto out;
3324
3325        dentry->d_inode->i_flags |= S_DEAD;
3326        dont_mount(dentry);
3327
3328out:
3329        mutex_unlock(&dentry->d_inode->i_mutex);
3330        dput(dentry);
3331        if (!error)
3332                d_delete(dentry);
3333        return error;
3334}
3335
3336static long do_rmdir(int dfd, const char __user *pathname)
3337{
3338        int error = 0;
3339        struct filename *name;
3340        struct dentry *dentry;
3341        struct nameidata nd;
3342        unsigned int lookup_flags = 0;
3343retry:
3344        name = user_path_parent(dfd, pathname, &nd, lookup_flags);
3345        if (IS_ERR(name))
3346                return PTR_ERR(name);
3347
3348        switch(nd.last_type) {
3349        case LAST_DOTDOT:
3350                error = -ENOTEMPTY;
3351                goto exit1;
3352        case LAST_DOT:
3353                error = -EINVAL;
3354                goto exit1;
3355        case LAST_ROOT:
3356                error = -EBUSY;
3357                goto exit1;
3358        }
3359
3360        nd.flags &= ~LOOKUP_PARENT;
3361        error = mnt_want_write(nd.path.mnt);
3362        if (error)
3363                goto exit1;
3364
3365        mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3366        dentry = lookup_hash(&nd);
3367        error = PTR_ERR(dentry);
3368        if (IS_ERR(dentry))
3369                goto exit2;
3370        if (!dentry->d_inode) {
3371                error = -ENOENT;
3372                goto exit3;
3373        }
3374        error = security_path_rmdir(&nd.path, dentry);
3375        if (error)
3376                goto exit3;
3377        error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
3378exit3:
3379        dput(dentry);
3380exit2:
3381        mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3382        mnt_drop_write(nd.path.mnt);
3383exit1:
3384        path_put(&nd.path);
3385        putname(name);
3386        if (retry_estale(error, lookup_flags)) {
3387                lookup_flags |= LOOKUP_REVAL;
3388                goto retry;
3389        }
3390        return error;
3391}
3392
3393SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3394{
3395        return do_rmdir(AT_FDCWD, pathname);
3396}
3397
3398int vfs_unlink(struct inode *dir, struct dentry *dentry)
3399{
3400        int error = may_delete(dir, dentry, 0);
3401
3402        if (error)
3403                return error;
3404
3405        if (!dir->i_op->unlink)
3406                return -EPERM;
3407
3408        mutex_lock(&dentry->d_inode->i_mutex);
3409        if (d_mountpoint(dentry))
3410                error = -EBUSY;
3411        else {
3412                error = security_inode_unlink(dir, dentry);
3413                if (!error) {
3414                        error = dir->i_op->unlink(dir, dentry);
3415                        if (!error)
3416                                dont_mount(dentry);
3417                }
3418        }
3419        mutex_unlock(&dentry->d_inode->i_mutex);
3420
3421        /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3422        if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3423                fsnotify_link_count(dentry->d_inode);
3424                d_delete(dentry);
3425        }
3426
3427        return error;
3428}
3429
3430/*
3431 * Make sure that the actual truncation of the file will occur outside its
3432 * directory's i_mutex.  Truncate can take a long time if there is a lot of
3433 * writeout happening, and we don't want to prevent access to the directory
3434 * while waiting on the I/O.
3435 */
3436static long do_unlinkat(int dfd, const char __user *pathname)
3437{
3438        int error;
3439        struct filename *name;
3440        struct dentry *dentry;
3441        struct nameidata nd;
3442        struct inode *inode = NULL;
3443        unsigned int lookup_flags = 0;
3444retry:
3445        name = user_path_parent(dfd, pathname, &nd, lookup_flags);
3446        if (IS_ERR(name))
3447                return PTR_ERR(name);
3448
3449        error = -EISDIR;
3450        if (nd.last_type != LAST_NORM)
3451                goto exit1;
3452
3453        nd.flags &= ~LOOKUP_PARENT;
3454        error = mnt_want_write(nd.path.mnt);
3455        if (error)
3456                goto exit1;
3457
3458        mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3459        dentry = lookup_hash(&nd);
3460        error = PTR_ERR(dentry);
3461        if (!IS_ERR(dentry)) {
3462                /* Why not before? Because we want correct error value */
3463                if (nd.last.name[nd.last.len])
3464                        goto slashes;
3465                inode = dentry->d_inode;
3466                if (!inode)
3467                        goto slashes;
3468                ihold(inode);
3469                error = security_path_unlink(&nd.path, dentry);
3470                if (error)
3471                        goto exit2;
3472                error = vfs_unlink(nd.path.dentry->d_inode, dentry);
3473exit2:
3474                dput(dentry);
3475        }
3476        mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3477        if (inode)
3478                iput(inode);    /* truncate the inode here */
3479        mnt_drop_write(nd.path.mnt);
3480exit1:
3481        path_put(&nd.path);
3482        putname(name);
3483        if (retry_estale(error, lookup_flags)) {
3484                lookup_flags |= LOOKUP_REVAL;
3485                inode = NULL;
3486                goto retry;
3487        }
3488        return error;
3489
3490slashes:
3491        error = !dentry->d_inode ? -ENOENT :
3492                S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
3493        goto exit2;
3494}
3495
3496SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
3497{
3498        if ((flag & ~AT_REMOVEDIR) != 0)
3499                return -EINVAL;
3500
3501        if (flag & AT_REMOVEDIR)
3502                return do_rmdir(dfd, pathname);
3503
3504        return do_unlinkat(dfd, pathname);
3505}
3506
3507SYSCALL_DEFINE1(unlink, const char __user *, pathname)
3508{
3509        return do_unlinkat(AT_FDCWD, pathname);
3510}
3511
3512int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
3513{
3514        int error = may_create(dir, dentry);
3515
3516        if (error)
3517                return error;
3518
3519        if (!dir->i_op->symlink)
3520                return -EPERM;
3521
3522        error = security_inode_symlink(dir, dentry, oldname);
3523        if (error)
3524                return error;
3525
3526        error = dir->i_op->symlink(dir, dentry, oldname);
3527        if (!error)
3528                fsnotify_create(dir, dentry);
3529        return error;
3530}
3531
3532SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3533                int, newdfd, const char __user *, newname)
3534{
3535        int error;
3536        struct filename *from;
3537        struct dentry *dentry;
3538        struct path path;
3539        unsigned int lookup_flags = 0;
3540
3541        from = getname(oldname);
3542        if (IS_ERR(from))
3543                return PTR_ERR(from);
3544retry:
3545        dentry = user_path_create(newdfd, newname, &path, lookup_flags);
3546        error = PTR_ERR(dentry);
3547        if (IS_ERR(dentry))
3548                goto out_putname;
3549
3550        error = security_path_symlink(&path, dentry, from->name);
3551        if (!error)
3552                error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
3553        done_path_create(&path, dentry);
3554        if (retry_estale(error, lookup_flags)) {
3555                lookup_flags |= LOOKUP_REVAL;
3556                goto retry;
3557        }
3558out_putname:
3559        putname(from);
3560        return error;
3561}
3562
3563SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
3564{
3565        return sys_symlinkat(oldname, AT_FDCWD, newname);
3566}
3567
3568int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
3569{
3570        struct inode *inode = old_dentry->d_inode;
3571        unsigned max_links = dir->i_sb->s_max_links;
3572        int error;
3573
3574        if (!inode)
3575                return -ENOENT;
3576
3577        error = may_create(dir, new_dentry);
3578        if (error)
3579                return error;
3580
3581        if (dir->i_sb != inode->i_sb)
3582                return -EXDEV;
3583
3584        /*
3585         * A link to an append-only or immutable file cannot be created.
3586         */
3587        if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3588                return -EPERM;
3589        if (!dir->i_op->link)
3590                return -EPERM;
3591        if (S_ISDIR(inode->i_mode))
3592                return -EPERM;
3593
3594        error = security_inode_link(old_dentry, dir, new_dentry);
3595        if (error)
3596                return error;
3597
3598        mutex_lock(&inode->i_mutex);
3599        /* Make sure we don't allow creating hardlink to an unlinked file */
3600        if (inode->i_nlink == 0)
3601                error =  -ENOENT;
3602        else if (max_links && inode->i_nlink >= max_links)
3603                error = -EMLINK;
3604        else
3605                error = dir->i_op->link(old_dentry, dir, new_dentry);
3606        mutex_unlock(&inode->i_mutex);
3607        if (!error)
3608                fsnotify_link(dir, inode, new_dentry);
3609        return error;
3610}
3611
3612/*
3613 * Hardlinks are often used in delicate situations.  We avoid
3614 * security-related surprises by not following symlinks on the
3615 * newname.  --KAB
3616 *
3617 * We don't follow them on the oldname either to be compatible
3618 * with linux 2.0, and to avoid hard-linking to directories
3619 * and other special files.  --ADM
3620 */
3621SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
3622                int, newdfd, const char __user *, newname, int, flags)
3623{
3624        struct dentry *new_dentry;
3625        struct path old_path, new_path;
3626        int how = 0;
3627        int error;
3628
3629        if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
3630                return -EINVAL;
3631        /*
3632         * To use null names we require CAP_DAC_READ_SEARCH
3633         * This ensures that not everyone will be able to create
3634         * handlink using the passed filedescriptor.
3635         */
3636        if (flags & AT_EMPTY_PATH) {
3637                if (!capable(CAP_DAC_READ_SEARCH))
3638                        return -ENOENT;
3639                how = LOOKUP_EMPTY;
3640        }
3641
3642        if (flags & AT_SYMLINK_FOLLOW)
3643                how |= LOOKUP_FOLLOW;
3644retry:
3645        error = user_path_at(olddfd, oldname, how, &old_path);
3646        if (error)
3647                return error;
3648
3649        new_dentry = user_path_create(newdfd, newname, &new_path,
3650                                        (how & LOOKUP_REVAL));
3651        error = PTR_ERR(new_dentry);
3652        if (IS_ERR(new_dentry))
3653                goto out;
3654
3655        error = -EXDEV;
3656        if (old_path.mnt != new_path.mnt)
3657                goto out_dput;
3658        error = may_linkat(&old_path);
3659        if (unlikely(error))
3660                goto out_dput;
3661        error = security_path_link(old_path.dentry, &new_path, new_dentry);
3662        if (error)
3663                goto out_dput;
3664        error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry);
3665out_dput:
3666        done_path_create(&new_path, new_dentry);
3667        if (retry_estale(error, how)) {
3668                how |= LOOKUP_REVAL;
3669                goto retry;
3670        }
3671out:
3672        path_put(&old_path);
3673
3674        return error;
3675}
3676
3677SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
3678{
3679        return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
3680}
3681
3682/*
3683 * The worst of all namespace operations - renaming directory. "Perverted"
3684 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3685 * Problems:
3686 *      a) we can get into loop creation. Check is done in is_subdir().
3687 *      b) race potential - two innocent renames can create a loop together.
3688 *         That's where 4.4 screws up. Current fix: serialization on
3689 *         sb->s_vfs_rename_mutex. We might be more accurate, but that's another
3690 *         story.
3691 *      c) we have to lock _three_ objects - parents and victim (if it exists).
3692 *         And that - after we got ->i_mutex on parents (until then we don't know
3693 *         whether the target exists).  Solution: try to be smart with locking
3694 *         order for inodes.  We rely on the fact that tree topology may change
3695 *         only under ->s_vfs_rename_mutex _and_ that parent of the object we
3696 *         move will be locked.  Thus we can rank directories by the tree
3697 *         (ancestors first) and rank all non-directories after them.
3698 *         That works since everybody except rename does "lock parent, lookup,
3699 *         lock child" and rename is under ->s_vfs_rename_mutex.
3700 *         HOWEVER, it relies on the assumption that any object with ->lookup()
3701 *         has no more than 1 dentry.  If "hybrid" objects will ever appear,
3702 *         we'd better make sure that there's no link(2) for them.
3703 *      d) conversion from fhandle to dentry may come in the wrong moment - when
3704 *         we are removing the target. Solution: we will have to grab ->i_mutex
3705 *         in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3706 *         ->i_mutex on parents, which works but leads to some truly excessive
3707 *         locking].
3708 */
3709static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3710                          struct inode *new_dir, struct dentry *new_dentry)
3711{
3712        int error = 0;
3713        struct inode *target = new_dentry->d_inode;
3714        unsigned max_links = new_dir->i_sb->s_max_links;
3715
3716        /*
3717         * If we are going to change the parent - check write permissions,
3718         * we'll need to flip '..'.
3719         */
3720        if (new_dir != old_dir) {
3721                error = inode_permission(old_dentry->d_inode, MAY_WRITE);
3722                if (error)
3723                        return error;
3724        }
3725
3726        error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3727        if (error)
3728                return error;
3729
3730        dget(new_dentry);
3731        if (target)
3732                mutex_lock(&target->i_mutex);
3733
3734        error = -EBUSY;
3735        if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
3736                goto out;
3737
3738        error = -EMLINK;
3739        if (max_links && !target && new_dir != old_dir &&
3740            new_dir->i_nlink >= max_links)
3741                goto out;
3742
3743        if (target)
3744                shrink_dcache_parent(new_dentry);
3745        error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3746        if (error)
3747                goto out;
3748
3749        if (target) {
3750                target->i_flags |= S_DEAD;
3751                dont_mount(new_dentry);
3752        }
3753out:
3754        if (target)
3755                mutex_unlock(&target->i_mutex);
3756        dput(new_dentry);
3757        if (!error)
3758                if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3759                        d_move(old_dentry,new_dentry);
3760        return error;
3761}
3762
3763static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3764                            struct inode *new_dir, struct dentry *new_dentry)
3765{
3766        struct inode *target = new_dentry->d_inode;
3767        int error;
3768
3769        error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3770        if (error)
3771                return error;
3772
3773        dget(new_dentry);
3774        if (target)
3775                mutex_lock(&target->i_mutex);
3776
3777        error = -EBUSY;
3778        if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3779                goto out;
3780
3781        error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3782        if (error)
3783                goto out;
3784
3785        if (target)
3786                dont_mount(new_dentry);
3787        if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3788                d_move(old_dentry, new_dentry);
3789out:
3790        if (target)
3791                mutex_unlock(&target->i_mutex);
3792        dput(new_dentry);
3793        return error;
3794}
3795
3796int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3797               struct inode *new_dir, struct dentry *new_dentry)
3798{
3799        int error;
3800        int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
3801        const unsigned char *old_name;
3802
3803        if (old_dentry->d_inode == new_dentry->d_inode)
3804                return 0;
3805 
3806        error = may_delete(old_dir, old_dentry, is_dir);
3807        if (error)
3808                return error;
3809
3810        if (!new_dentry->d_inode)
3811                error = may_create(new_dir, new_dentry);
3812        else
3813                error = may_delete(new_dir, new_dentry, is_dir);
3814        if (error)
3815                return error;
3816
3817        if (!old_dir->i_op->rename)
3818                return -EPERM;
3819
3820        old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3821
3822        if (is_dir)
3823                error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3824        else
3825                error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
3826        if (!error)
3827                fsnotify_move(old_dir, new_dir, old_name, is_dir,
3828                              new_dentry->d_inode, old_dentry);
3829        fsnotify_oldname_free(old_name);
3830
3831        return error;
3832}
3833
3834SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3835                int, newdfd, const char __user *, newname)
3836{
3837        struct dentry *old_dir, *new_dir;
3838        struct dentry *old_dentry, *new_dentry;
3839        struct dentry *trap;
3840        struct nameidata oldnd, newnd;
3841        struct filename *from;
3842        struct filename *to;
3843        unsigned int lookup_flags = 0;
3844        bool should_retry = false;
3845        int error;
3846retry:
3847        from = user_path_parent(olddfd, oldname, &oldnd, lookup_flags);
3848        if (IS_ERR(from)) {
3849                error = PTR_ERR(from);
3850                goto exit;
3851        }
3852
3853        to = user_path_parent(newdfd, newname, &newnd, lookup_flags);
3854        if (IS_ERR(to)) {
3855                error = PTR_ERR(to);
3856                goto exit1;
3857        }
3858
3859        error = -EXDEV;
3860        if (oldnd.path.mnt != newnd.path.mnt)
3861                goto exit2;
3862
3863        old_dir = oldnd.path.dentry;
3864        error = -EBUSY;
3865        if (oldnd.last_type != LAST_NORM)
3866                goto exit2;
3867
3868        new_dir = newnd.path.dentry;
3869        if (newnd.last_type != LAST_NORM)
3870                goto exit2;
3871
3872        error = mnt_want_write(oldnd.path.mnt);
3873        if (error)
3874                goto exit2;
3875
3876        oldnd.flags &= ~LOOKUP_PARENT;
3877        newnd.flags &= ~LOOKUP_PARENT;
3878        newnd.flags |= LOOKUP_RENAME_TARGET;
3879
3880        trap = lock_rename(new_dir, old_dir);
3881
3882        old_dentry = lookup_hash(&oldnd);
3883        error = PTR_ERR(old_dentry);
3884        if (IS_ERR(old_dentry))
3885                goto exit3;
3886        /* source must exist */
3887        error = -ENOENT;
3888        if (!old_dentry->d_inode)
3889                goto exit4;
3890        /* unless the source is a directory trailing slashes give -ENOTDIR */
3891        if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3892                error = -ENOTDIR;
3893                if (oldnd.last.name[oldnd.last.len])
3894                        goto exit4;
3895                if (newnd.last.name[newnd.last.len])
3896                        goto exit4;
3897        }
3898        /* source should not be ancestor of target */
3899        error = -EINVAL;
3900        if (old_dentry == trap)
3901                goto exit4;
3902        new_dentry = lookup_hash(&newnd);
3903        error = PTR_ERR(new_dentry);
3904        if (IS_ERR(new_dentry))
3905                goto exit4;
3906        /* target should not be an ancestor of source */
3907        error = -ENOTEMPTY;
3908        if (new_dentry == trap)
3909                goto exit5;
3910
3911        error = security_path_rename(&oldnd.path, old_dentry,
3912                                     &newnd.path, new_dentry);
3913        if (error)
3914                goto exit5;
3915        error = vfs_rename(old_dir->d_inode, old_dentry,
3916                                   new_dir->d_inode, new_dentry);
3917exit5:
3918        dput(new_dentry);
3919exit4:
3920        dput(old_dentry);
3921exit3:
3922        unlock_rename(new_dir, old_dir);
3923        mnt_drop_write(oldnd.path.mnt);
3924exit2:
3925        if (retry_estale(error, lookup_flags))
3926                should_retry = true;
3927        path_put(&newnd.path);
3928        putname(to);
3929exit1:
3930        path_put(&oldnd.path);
3931        putname(from);
3932        if (should_retry) {
3933                should_retry = false;
3934                lookup_flags |= LOOKUP_REVAL;
3935                goto retry;
3936        }
3937exit:
3938        return error;
3939}
3940
3941SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
3942{
3943        return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3944}
3945
3946int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3947{
3948        int len;
3949
3950        len = PTR_ERR(link);
3951        if (IS_ERR(link))
3952                goto out;
3953
3954        len = strlen(link);
3955        if (len > (unsigned) buflen)
3956                len = buflen;
3957        if (copy_to_user(buffer, link, len))
3958                len = -EFAULT;
3959out:
3960        return len;
3961}
3962
3963/*
3964 * A helper for ->readlink().  This should be used *ONLY* for symlinks that
3965 * have ->follow_link() touching nd only in nd_set_link().  Using (or not
3966 * using) it for any given inode is up to filesystem.
3967 */
3968int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3969{
3970        struct nameidata nd;
3971        void *cookie;
3972        int res;
3973
3974        nd.depth = 0;
3975        cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
3976        if (IS_ERR(cookie))
3977                return PTR_ERR(cookie);
3978
3979        res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3980        if (dentry->d_inode->i_op->put_link)
3981                dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3982        return res;
3983}
3984
3985int vfs_follow_link(struct nameidata *nd, const char *link)
3986{
3987        return __vfs_follow_link(nd, link);
3988}
3989
3990/* get the link contents into pagecache */
3991static char *page_getlink(struct dentry * dentry, struct page **ppage)
3992{
3993        char *kaddr;
3994        struct page *page;
3995        struct address_space *mapping = dentry->d_inode->i_mapping;
3996        page = read_mapping_page(mapping, 0, NULL);
3997        if (IS_ERR(page))
3998                return (char*)page;
3999        *ppage = page;
4000        kaddr = kmap(page);
4001        nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
4002        return kaddr;
4003}
4004
4005int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4006{
4007        struct page *page = NULL;
4008        char *s = page_getlink(dentry, &page);
4009        int res = vfs_readlink(dentry,buffer,buflen,s);
4010        if (page) {
4011                kunmap(page);
4012                page_cache_release(page);
4013        }
4014        return res;
4015}
4016
4017void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
4018{
4019        struct page *page = NULL;
4020        nd_set_link(nd, page_getlink(dentry, &page));
4021        return page;
4022}
4023
4024void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
4025{
4026        struct page *page = cookie;
4027
4028        if (page) {
4029                kunmap(page);
4030                page_cache_release(page);
4031        }
4032}
4033
4034/*
4035 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4036 */
4037int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4038{
4039        struct address_space *mapping = inode->i_mapping;
4040        struct page *page;
4041        void *fsdata;
4042        int err;
4043        char *kaddr;
4044        unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4045        if (nofs)
4046                flags |= AOP_FLAG_NOFS;
4047
4048retry:
4049        err = pagecache_write_begin(NULL, mapping, 0, len-1,
4050                                flags, &page, &fsdata);
4051        if (err)
4052                goto fail;
4053
4054        kaddr = kmap_atomic(page);
4055        memcpy(kaddr, symname, len-1);
4056        kunmap_atomic(kaddr);
4057
4058        err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4059                                                        page, fsdata);
4060        if (err < 0)
4061                goto fail;
4062        if (err < len-1)
4063                goto retry;
4064
4065        mark_inode_dirty(inode);
4066        return 0;
4067fail:
4068        return err;
4069}
4070
4071int page_symlink(struct inode *inode, const char *symname, int len)
4072{
4073        return __page_symlink(inode, symname, len,
4074                        !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
4075}
4076
4077const struct inode_operations page_symlink_inode_operations = {
4078        .readlink       = generic_readlink,
4079        .follow_link    = page_follow_link_light,
4080        .put_link       = page_put_link,
4081};
4082
4083EXPORT_SYMBOL(user_path_at);
4084EXPORT_SYMBOL(follow_down_one);
4085EXPORT_SYMBOL(follow_down);
4086EXPORT_SYMBOL(follow_up);
4087EXPORT_SYMBOL(get_write_access); /* nfsd */
4088EXPORT_SYMBOL(lock_rename);
4089EXPORT_SYMBOL(lookup_one_len);
4090EXPORT_SYMBOL(page_follow_link_light);
4091EXPORT_SYMBOL(page_put_link);
4092EXPORT_SYMBOL(page_readlink);
4093EXPORT_SYMBOL(__page_symlink);
4094EXPORT_SYMBOL(page_symlink);
4095EXPORT_SYMBOL(page_symlink_inode_operations);
4096EXPORT_SYMBOL(kern_path);
4097EXPORT_SYMBOL(vfs_path_lookup);
4098EXPORT_SYMBOL(inode_permission);
4099EXPORT_SYMBOL(unlock_rename);
4100EXPORT_SYMBOL(vfs_create);
4101EXPORT_SYMBOL(vfs_follow_link);
4102EXPORT_SYMBOL(vfs_link);
4103EXPORT_SYMBOL(vfs_mkdir);
4104EXPORT_SYMBOL(vfs_mknod);
4105EXPORT_SYMBOL(generic_permission);
4106EXPORT_SYMBOL(vfs_readlink);
4107EXPORT_SYMBOL(vfs_rename);
4108EXPORT_SYMBOL(vfs_rmdir);
4109EXPORT_SYMBOL(vfs_symlink);
4110EXPORT_SYMBOL(vfs_unlink);
4111EXPORT_SYMBOL(dentry_unhash);
4112EXPORT_SYMBOL(generic_readlink);
4113
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.