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