linux/fs/overlayfs/namei.c
<<
>>
Prefs
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * Copyright (C) 2011 Novell Inc.
   4 * Copyright (C) 2016 Red Hat, Inc.
   5 */
   6
   7#include <linux/fs.h>
   8#include <linux/cred.h>
   9#include <linux/ctype.h>
  10#include <linux/namei.h>
  11#include <linux/xattr.h>
  12#include <linux/ratelimit.h>
  13#include <linux/mount.h>
  14#include <linux/exportfs.h>
  15#include "overlayfs.h"
  16
  17struct ovl_lookup_data {
  18        struct super_block *sb;
  19        struct qstr name;
  20        bool is_dir;
  21        bool opaque;
  22        bool stop;
  23        bool last;
  24        char *redirect;
  25        bool metacopy;
  26};
  27
  28static int ovl_check_redirect(struct dentry *dentry, struct ovl_lookup_data *d,
  29                              size_t prelen, const char *post)
  30{
  31        int res;
  32        char *buf;
  33        struct ovl_fs *ofs = OVL_FS(d->sb);
  34
  35        buf = ovl_get_redirect_xattr(ofs, dentry, prelen + strlen(post));
  36        if (IS_ERR_OR_NULL(buf))
  37                return PTR_ERR(buf);
  38
  39        if (buf[0] == '/') {
  40                /*
  41                 * One of the ancestor path elements in an absolute path
  42                 * lookup in ovl_lookup_layer() could have been opaque and
  43                 * that will stop further lookup in lower layers (d->stop=true)
  44                 * But we have found an absolute redirect in decendant path
  45                 * element and that should force continue lookup in lower
  46                 * layers (reset d->stop).
  47                 */
  48                d->stop = false;
  49        } else {
  50                res = strlen(buf) + 1;
  51                memmove(buf + prelen, buf, res);
  52                memcpy(buf, d->name.name, prelen);
  53        }
  54
  55        strcat(buf, post);
  56        kfree(d->redirect);
  57        d->redirect = buf;
  58        d->name.name = d->redirect;
  59        d->name.len = strlen(d->redirect);
  60
  61        return 0;
  62}
  63
  64static int ovl_acceptable(void *ctx, struct dentry *dentry)
  65{
  66        /*
  67         * A non-dir origin may be disconnected, which is fine, because
  68         * we only need it for its unique inode number.
  69         */
  70        if (!d_is_dir(dentry))
  71                return 1;
  72
  73        /* Don't decode a deleted empty directory */
  74        if (d_unhashed(dentry))
  75                return 0;
  76
  77        /* Check if directory belongs to the layer we are decoding from */
  78        return is_subdir(dentry, ((struct vfsmount *)ctx)->mnt_root);
  79}
  80
  81/*
  82 * Check validity of an overlay file handle buffer.
  83 *
  84 * Return 0 for a valid file handle.
  85 * Return -ENODATA for "origin unknown".
  86 * Return <0 for an invalid file handle.
  87 */
  88int ovl_check_fb_len(struct ovl_fb *fb, int fb_len)
  89{
  90        if (fb_len < sizeof(struct ovl_fb) || fb_len < fb->len)
  91                return -EINVAL;
  92
  93        if (fb->magic != OVL_FH_MAGIC)
  94                return -EINVAL;
  95
  96        /* Treat larger version and unknown flags as "origin unknown" */
  97        if (fb->version > OVL_FH_VERSION || fb->flags & ~OVL_FH_FLAG_ALL)
  98                return -ENODATA;
  99
 100        /* Treat endianness mismatch as "origin unknown" */
 101        if (!(fb->flags & OVL_FH_FLAG_ANY_ENDIAN) &&
 102            (fb->flags & OVL_FH_FLAG_BIG_ENDIAN) != OVL_FH_FLAG_CPU_ENDIAN)
 103                return -ENODATA;
 104
 105        return 0;
 106}
 107
 108static struct ovl_fh *ovl_get_fh(struct ovl_fs *ofs, struct dentry *dentry,
 109                                 enum ovl_xattr ox)
 110{
 111        int res, err;
 112        struct ovl_fh *fh = NULL;
 113
 114        res = ovl_do_getxattr(ofs, dentry, ox, NULL, 0);
 115        if (res < 0) {
 116                if (res == -ENODATA || res == -EOPNOTSUPP)
 117                        return NULL;
 118                goto fail;
 119        }
 120        /* Zero size value means "copied up but origin unknown" */
 121        if (res == 0)
 122                return NULL;
 123
 124        fh = kzalloc(res + OVL_FH_WIRE_OFFSET, GFP_KERNEL);
 125        if (!fh)
 126                return ERR_PTR(-ENOMEM);
 127
 128        res = ovl_do_getxattr(ofs, dentry, ox, fh->buf, res);
 129        if (res < 0)
 130                goto fail;
 131
 132        err = ovl_check_fb_len(&fh->fb, res);
 133        if (err < 0) {
 134                if (err == -ENODATA)
 135                        goto out;
 136                goto invalid;
 137        }
 138
 139        return fh;
 140
 141out:
 142        kfree(fh);
 143        return NULL;
 144
 145fail:
 146        pr_warn_ratelimited("failed to get origin (%i)\n", res);
 147        goto out;
 148invalid:
 149        pr_warn_ratelimited("invalid origin (%*phN)\n", res, fh);
 150        goto out;
 151}
 152
 153struct dentry *ovl_decode_real_fh(struct ovl_fh *fh, struct vfsmount *mnt,
 154                                  bool connected)
 155{
 156        struct dentry *real;
 157        int bytes;
 158
 159        /*
 160         * Make sure that the stored uuid matches the uuid of the lower
 161         * layer where file handle will be decoded.
 162         */
 163        if (!uuid_equal(&fh->fb.uuid, &mnt->mnt_sb->s_uuid))
 164                return NULL;
 165
 166        bytes = (fh->fb.len - offsetof(struct ovl_fb, fid));
 167        real = exportfs_decode_fh(mnt, (struct fid *)fh->fb.fid,
 168                                  bytes >> 2, (int)fh->fb.type,
 169                                  connected ? ovl_acceptable : NULL, mnt);
 170        if (IS_ERR(real)) {
 171                /*
 172                 * Treat stale file handle to lower file as "origin unknown".
 173                 * upper file handle could become stale when upper file is
 174                 * unlinked and this information is needed to handle stale
 175                 * index entries correctly.
 176                 */
 177                if (real == ERR_PTR(-ESTALE) &&
 178                    !(fh->fb.flags & OVL_FH_FLAG_PATH_UPPER))
 179                        real = NULL;
 180                return real;
 181        }
 182
 183        if (ovl_dentry_weird(real)) {
 184                dput(real);
 185                return NULL;
 186        }
 187
 188        return real;
 189}
 190
 191static bool ovl_is_opaquedir(struct super_block *sb, struct dentry *dentry)
 192{
 193        return ovl_check_dir_xattr(sb, dentry, OVL_XATTR_OPAQUE);
 194}
 195
 196static struct dentry *ovl_lookup_positive_unlocked(const char *name,
 197                                                   struct dentry *base, int len,
 198                                                   bool drop_negative)
 199{
 200        struct dentry *ret = lookup_one_len_unlocked(name, base, len);
 201
 202        if (!IS_ERR(ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) {
 203                if (drop_negative && ret->d_lockref.count == 1) {
 204                        spin_lock(&ret->d_lock);
 205                        /* Recheck condition under lock */
 206                        if (d_is_negative(ret) && ret->d_lockref.count == 1)
 207                                __d_drop(ret);
 208                        spin_unlock(&ret->d_lock);
 209                }
 210                dput(ret);
 211                ret = ERR_PTR(-ENOENT);
 212        }
 213        return ret;
 214}
 215
 216static int ovl_lookup_single(struct dentry *base, struct ovl_lookup_data *d,
 217                             const char *name, unsigned int namelen,
 218                             size_t prelen, const char *post,
 219                             struct dentry **ret, bool drop_negative)
 220{
 221        struct dentry *this;
 222        int err;
 223        bool last_element = !post[0];
 224
 225        this = ovl_lookup_positive_unlocked(name, base, namelen, drop_negative);
 226        if (IS_ERR(this)) {
 227                err = PTR_ERR(this);
 228                this = NULL;
 229                if (err == -ENOENT || err == -ENAMETOOLONG)
 230                        goto out;
 231                goto out_err;
 232        }
 233
 234        if (ovl_dentry_weird(this)) {
 235                /* Don't support traversing automounts and other weirdness */
 236                err = -EREMOTE;
 237                goto out_err;
 238        }
 239        if (ovl_is_whiteout(this)) {
 240                d->stop = d->opaque = true;
 241                goto put_and_out;
 242        }
 243        /*
 244         * This dentry should be a regular file if previous layer lookup
 245         * found a metacopy dentry.
 246         */
 247        if (last_element && d->metacopy && !d_is_reg(this)) {
 248                d->stop = true;
 249                goto put_and_out;
 250        }
 251        if (!d_can_lookup(this)) {
 252                if (d->is_dir || !last_element) {
 253                        d->stop = true;
 254                        goto put_and_out;
 255                }
 256                err = ovl_check_metacopy_xattr(OVL_FS(d->sb), this);
 257                if (err < 0)
 258                        goto out_err;
 259
 260                d->metacopy = err;
 261                d->stop = !d->metacopy;
 262                if (!d->metacopy || d->last)
 263                        goto out;
 264        } else {
 265                if (ovl_lookup_trap_inode(d->sb, this)) {
 266                        /* Caught in a trap of overlapping layers */
 267                        err = -ELOOP;
 268                        goto out_err;
 269                }
 270
 271                if (last_element)
 272                        d->is_dir = true;
 273                if (d->last)
 274                        goto out;
 275
 276                if (ovl_is_opaquedir(d->sb, this)) {
 277                        d->stop = true;
 278                        if (last_element)
 279                                d->opaque = true;
 280                        goto out;
 281                }
 282        }
 283        err = ovl_check_redirect(this, d, prelen, post);
 284        if (err)
 285                goto out_err;
 286out:
 287        *ret = this;
 288        return 0;
 289
 290put_and_out:
 291        dput(this);
 292        this = NULL;
 293        goto out;
 294
 295out_err:
 296        dput(this);
 297        return err;
 298}
 299
 300static int ovl_lookup_layer(struct dentry *base, struct ovl_lookup_data *d,
 301                            struct dentry **ret, bool drop_negative)
 302{
 303        /* Counting down from the end, since the prefix can change */
 304        size_t rem = d->name.len - 1;
 305        struct dentry *dentry = NULL;
 306        int err;
 307
 308        if (d->name.name[0] != '/')
 309                return ovl_lookup_single(base, d, d->name.name, d->name.len,
 310                                         0, "", ret, drop_negative);
 311
 312        while (!IS_ERR_OR_NULL(base) && d_can_lookup(base)) {
 313                const char *s = d->name.name + d->name.len - rem;
 314                const char *next = strchrnul(s, '/');
 315                size_t thislen = next - s;
 316                bool end = !next[0];
 317
 318                /* Verify we did not go off the rails */
 319                if (WARN_ON(s[-1] != '/'))
 320                        return -EIO;
 321
 322                err = ovl_lookup_single(base, d, s, thislen,
 323                                        d->name.len - rem, next, &base,
 324                                        drop_negative);
 325                dput(dentry);
 326                if (err)
 327                        return err;
 328                dentry = base;
 329                if (end)
 330                        break;
 331
 332                rem -= thislen + 1;
 333
 334                if (WARN_ON(rem >= d->name.len))
 335                        return -EIO;
 336        }
 337        *ret = dentry;
 338        return 0;
 339}
 340
 341
 342int ovl_check_origin_fh(struct ovl_fs *ofs, struct ovl_fh *fh, bool connected,
 343                        struct dentry *upperdentry, struct ovl_path **stackp)
 344{
 345        struct dentry *origin = NULL;
 346        int i;
 347
 348        for (i = 1; i < ofs->numlayer; i++) {
 349                /*
 350                 * If lower fs uuid is not unique among lower fs we cannot match
 351                 * fh->uuid to layer.
 352                 */
 353                if (ofs->layers[i].fsid &&
 354                    ofs->layers[i].fs->bad_uuid)
 355                        continue;
 356
 357                origin = ovl_decode_real_fh(fh, ofs->layers[i].mnt,
 358                                            connected);
 359                if (origin)
 360                        break;
 361        }
 362
 363        if (!origin)
 364                return -ESTALE;
 365        else if (IS_ERR(origin))
 366                return PTR_ERR(origin);
 367
 368        if (upperdentry && !ovl_is_whiteout(upperdentry) &&
 369            ((d_inode(origin)->i_mode ^ d_inode(upperdentry)->i_mode) & S_IFMT))
 370                goto invalid;
 371
 372        if (!*stackp)
 373                *stackp = kmalloc(sizeof(struct ovl_path), GFP_KERNEL);
 374        if (!*stackp) {
 375                dput(origin);
 376                return -ENOMEM;
 377        }
 378        **stackp = (struct ovl_path){
 379                .dentry = origin,
 380                .layer = &ofs->layers[i]
 381        };
 382
 383        return 0;
 384
 385invalid:
 386        pr_warn_ratelimited("invalid origin (%pd2, ftype=%x, origin ftype=%x).\n",
 387                            upperdentry, d_inode(upperdentry)->i_mode & S_IFMT,
 388                            d_inode(origin)->i_mode & S_IFMT);
 389        dput(origin);
 390        return -EIO;
 391}
 392
 393static int ovl_check_origin(struct ovl_fs *ofs, struct dentry *upperdentry,
 394                            struct ovl_path **stackp)
 395{
 396        struct ovl_fh *fh = ovl_get_fh(ofs, upperdentry, OVL_XATTR_ORIGIN);
 397        int err;
 398
 399        if (IS_ERR_OR_NULL(fh))
 400                return PTR_ERR(fh);
 401
 402        err = ovl_check_origin_fh(ofs, fh, false, upperdentry, stackp);
 403        kfree(fh);
 404
 405        if (err) {
 406                if (err == -ESTALE)
 407                        return 0;
 408                return err;
 409        }
 410
 411        return 0;
 412}
 413
 414/*
 415 * Verify that @fh matches the file handle stored in xattr @name.
 416 * Return 0 on match, -ESTALE on mismatch, < 0 on error.
 417 */
 418static int ovl_verify_fh(struct ovl_fs *ofs, struct dentry *dentry,
 419                         enum ovl_xattr ox, const struct ovl_fh *fh)
 420{
 421        struct ovl_fh *ofh = ovl_get_fh(ofs, dentry, ox);
 422        int err = 0;
 423
 424        if (!ofh)
 425                return -ENODATA;
 426
 427        if (IS_ERR(ofh))
 428                return PTR_ERR(ofh);
 429
 430        if (fh->fb.len != ofh->fb.len || memcmp(&fh->fb, &ofh->fb, fh->fb.len))
 431                err = -ESTALE;
 432
 433        kfree(ofh);
 434        return err;
 435}
 436
 437/*
 438 * Verify that @real dentry matches the file handle stored in xattr @name.
 439 *
 440 * If @set is true and there is no stored file handle, encode @real and store
 441 * file handle in xattr @name.
 442 *
 443 * Return 0 on match, -ESTALE on mismatch, -ENODATA on no xattr, < 0 on error.
 444 */
 445int ovl_verify_set_fh(struct ovl_fs *ofs, struct dentry *dentry,
 446                      enum ovl_xattr ox, struct dentry *real, bool is_upper,
 447                      bool set)
 448{
 449        struct inode *inode;
 450        struct ovl_fh *fh;
 451        int err;
 452
 453        fh = ovl_encode_real_fh(real, is_upper);
 454        err = PTR_ERR(fh);
 455        if (IS_ERR(fh)) {
 456                fh = NULL;
 457                goto fail;
 458        }
 459
 460        err = ovl_verify_fh(ofs, dentry, ox, fh);
 461        if (set && err == -ENODATA)
 462                err = ovl_do_setxattr(ofs, dentry, ox, fh->buf, fh->fb.len);
 463        if (err)
 464                goto fail;
 465
 466out:
 467        kfree(fh);
 468        return err;
 469
 470fail:
 471        inode = d_inode(real);
 472        pr_warn_ratelimited("failed to verify %s (%pd2, ino=%lu, err=%i)\n",
 473                            is_upper ? "upper" : "origin", real,
 474                            inode ? inode->i_ino : 0, err);
 475        goto out;
 476}
 477
 478/* Get upper dentry from index */
 479struct dentry *ovl_index_upper(struct ovl_fs *ofs, struct dentry *index)
 480{
 481        struct ovl_fh *fh;
 482        struct dentry *upper;
 483
 484        if (!d_is_dir(index))
 485                return dget(index);
 486
 487        fh = ovl_get_fh(ofs, index, OVL_XATTR_UPPER);
 488        if (IS_ERR_OR_NULL(fh))
 489                return ERR_CAST(fh);
 490
 491        upper = ovl_decode_real_fh(fh, ovl_upper_mnt(ofs), true);
 492        kfree(fh);
 493
 494        if (IS_ERR_OR_NULL(upper))
 495                return upper ?: ERR_PTR(-ESTALE);
 496
 497        if (!d_is_dir(upper)) {
 498                pr_warn_ratelimited("invalid index upper (%pd2, upper=%pd2).\n",
 499                                    index, upper);
 500                dput(upper);
 501                return ERR_PTR(-EIO);
 502        }
 503
 504        return upper;
 505}
 506
 507/*
 508 * Verify that an index entry name matches the origin file handle stored in
 509 * OVL_XATTR_ORIGIN and that origin file handle can be decoded to lower path.
 510 * Return 0 on match, -ESTALE on mismatch or stale origin, < 0 on error.
 511 */
 512int ovl_verify_index(struct ovl_fs *ofs, struct dentry *index)
 513{
 514        struct ovl_fh *fh = NULL;
 515        size_t len;
 516        struct ovl_path origin = { };
 517        struct ovl_path *stack = &origin;
 518        struct dentry *upper = NULL;
 519        int err;
 520
 521        if (!d_inode(index))
 522                return 0;
 523
 524        err = -EINVAL;
 525        if (index->d_name.len < sizeof(struct ovl_fb)*2)
 526                goto fail;
 527
 528        err = -ENOMEM;
 529        len = index->d_name.len / 2;
 530        fh = kzalloc(len + OVL_FH_WIRE_OFFSET, GFP_KERNEL);
 531        if (!fh)
 532                goto fail;
 533
 534        err = -EINVAL;
 535        if (hex2bin(fh->buf, index->d_name.name, len))
 536                goto fail;
 537
 538        err = ovl_check_fb_len(&fh->fb, len);
 539        if (err)
 540                goto fail;
 541
 542        /*
 543         * Whiteout index entries are used as an indication that an exported
 544         * overlay file handle should be treated as stale (i.e. after unlink
 545         * of the overlay inode). These entries contain no origin xattr.
 546         */
 547        if (ovl_is_whiteout(index))
 548                goto out;
 549
 550        /*
 551         * Verifying directory index entries are not stale is expensive, so
 552         * only verify stale dir index if NFS export is enabled.
 553         */
 554        if (d_is_dir(index) && !ofs->config.nfs_export)
 555                goto out;
 556
 557        /*
 558         * Directory index entries should have 'upper' xattr pointing to the
 559         * real upper dir. Non-dir index entries are hardlinks to the upper
 560         * real inode. For non-dir index, we can read the copy up origin xattr
 561         * directly from the index dentry, but for dir index we first need to
 562         * decode the upper directory.
 563         */
 564        upper = ovl_index_upper(ofs, index);
 565        if (IS_ERR_OR_NULL(upper)) {
 566                err = PTR_ERR(upper);
 567                /*
 568                 * Directory index entries with no 'upper' xattr need to be
 569                 * removed. When dir index entry has a stale 'upper' xattr,
 570                 * we assume that upper dir was removed and we treat the dir
 571                 * index as orphan entry that needs to be whited out.
 572                 */
 573                if (err == -ESTALE)
 574                        goto orphan;
 575                else if (!err)
 576                        err = -ESTALE;
 577                goto fail;
 578        }
 579
 580        err = ovl_verify_fh(ofs, upper, OVL_XATTR_ORIGIN, fh);
 581        dput(upper);
 582        if (err)
 583                goto fail;
 584
 585        /* Check if non-dir index is orphan and don't warn before cleaning it */
 586        if (!d_is_dir(index) && d_inode(index)->i_nlink == 1) {
 587                err = ovl_check_origin_fh(ofs, fh, false, index, &stack);
 588                if (err)
 589                        goto fail;
 590
 591                if (ovl_get_nlink(ofs, origin.dentry, index, 0) == 0)
 592                        goto orphan;
 593        }
 594
 595out:
 596        dput(origin.dentry);
 597        kfree(fh);
 598        return err;
 599
 600fail:
 601        pr_warn_ratelimited("failed to verify index (%pd2, ftype=%x, err=%i)\n",
 602                            index, d_inode(index)->i_mode & S_IFMT, err);
 603        goto out;
 604
 605orphan:
 606        pr_warn_ratelimited("orphan index entry (%pd2, ftype=%x, nlink=%u)\n",
 607                            index, d_inode(index)->i_mode & S_IFMT,
 608                            d_inode(index)->i_nlink);
 609        err = -ENOENT;
 610        goto out;
 611}
 612
 613static int ovl_get_index_name_fh(struct ovl_fh *fh, struct qstr *name)
 614{
 615        char *n, *s;
 616
 617        n = kcalloc(fh->fb.len, 2, GFP_KERNEL);
 618        if (!n)
 619                return -ENOMEM;
 620
 621        s  = bin2hex(n, fh->buf, fh->fb.len);
 622        *name = (struct qstr) QSTR_INIT(n, s - n);
 623
 624        return 0;
 625
 626}
 627
 628/*
 629 * Lookup in indexdir for the index entry of a lower real inode or a copy up
 630 * origin inode. The index entry name is the hex representation of the lower
 631 * inode file handle.
 632 *
 633 * If the index dentry in negative, then either no lower aliases have been
 634 * copied up yet, or aliases have been copied up in older kernels and are
 635 * not indexed.
 636 *
 637 * If the index dentry for a copy up origin inode is positive, but points
 638 * to an inode different than the upper inode, then either the upper inode
 639 * has been copied up and not indexed or it was indexed, but since then
 640 * index dir was cleared. Either way, that index cannot be used to indentify
 641 * the overlay inode.
 642 */
 643int ovl_get_index_name(struct dentry *origin, struct qstr *name)
 644{
 645        struct ovl_fh *fh;
 646        int err;
 647
 648        fh = ovl_encode_real_fh(origin, false);
 649        if (IS_ERR(fh))
 650                return PTR_ERR(fh);
 651
 652        err = ovl_get_index_name_fh(fh, name);
 653
 654        kfree(fh);
 655        return err;
 656}
 657
 658/* Lookup index by file handle for NFS export */
 659struct dentry *ovl_get_index_fh(struct ovl_fs *ofs, struct ovl_fh *fh)
 660{
 661        struct dentry *index;
 662        struct qstr name;
 663        int err;
 664
 665        err = ovl_get_index_name_fh(fh, &name);
 666        if (err)
 667                return ERR_PTR(err);
 668
 669        index = lookup_positive_unlocked(name.name, ofs->indexdir, name.len);
 670        kfree(name.name);
 671        if (IS_ERR(index)) {
 672                if (PTR_ERR(index) == -ENOENT)
 673                        index = NULL;
 674                return index;
 675        }
 676
 677        if (ovl_is_whiteout(index))
 678                err = -ESTALE;
 679        else if (ovl_dentry_weird(index))
 680                err = -EIO;
 681        else
 682                return index;
 683
 684        dput(index);
 685        return ERR_PTR(err);
 686}
 687
 688struct dentry *ovl_lookup_index(struct ovl_fs *ofs, struct dentry *upper,
 689                                struct dentry *origin, bool verify)
 690{
 691        struct dentry *index;
 692        struct inode *inode;
 693        struct qstr name;
 694        bool is_dir = d_is_dir(origin);
 695        int err;
 696
 697        err = ovl_get_index_name(origin, &name);
 698        if (err)
 699                return ERR_PTR(err);
 700
 701        index = lookup_positive_unlocked(name.name, ofs->indexdir, name.len);
 702        if (IS_ERR(index)) {
 703                err = PTR_ERR(index);
 704                if (err == -ENOENT) {
 705                        index = NULL;
 706                        goto out;
 707                }
 708                pr_warn_ratelimited("failed inode index lookup (ino=%lu, key=%.*s, err=%i);\n"
 709                                    "overlayfs: mount with '-o index=off' to disable inodes index.\n",
 710                                    d_inode(origin)->i_ino, name.len, name.name,
 711                                    err);
 712                goto out;
 713        }
 714
 715        inode = d_inode(index);
 716        if (ovl_is_whiteout(index) && !verify) {
 717                /*
 718                 * When index lookup is called with !verify for decoding an
 719                 * overlay file handle, a whiteout index implies that decode
 720                 * should treat file handle as stale and no need to print a
 721                 * warning about it.
 722                 */
 723                dput(index);
 724                index = ERR_PTR(-ESTALE);
 725                goto out;
 726        } else if (ovl_dentry_weird(index) || ovl_is_whiteout(index) ||
 727                   ((inode->i_mode ^ d_inode(origin)->i_mode) & S_IFMT)) {
 728                /*
 729                 * Index should always be of the same file type as origin
 730                 * except for the case of a whiteout index. A whiteout
 731                 * index should only exist if all lower aliases have been
 732                 * unlinked, which means that finding a lower origin on lookup
 733                 * whose index is a whiteout should be treated as an error.
 734                 */
 735                pr_warn_ratelimited("bad index found (index=%pd2, ftype=%x, origin ftype=%x).\n",
 736                                    index, d_inode(index)->i_mode & S_IFMT,
 737                                    d_inode(origin)->i_mode & S_IFMT);
 738                goto fail;
 739        } else if (is_dir && verify) {
 740                if (!upper) {
 741                        pr_warn_ratelimited("suspected uncovered redirected dir found (origin=%pd2, index=%pd2).\n",
 742                                            origin, index);
 743                        goto fail;
 744                }
 745
 746                /* Verify that dir index 'upper' xattr points to upper dir */
 747                err = ovl_verify_upper(ofs, index, upper, false);
 748                if (err) {
 749                        if (err == -ESTALE) {
 750                                pr_warn_ratelimited("suspected multiply redirected dir found (upper=%pd2, origin=%pd2, index=%pd2).\n",
 751                                                    upper, origin, index);
 752                        }
 753                        goto fail;
 754                }
 755        } else if (upper && d_inode(upper) != inode) {
 756                goto out_dput;
 757        }
 758out:
 759        kfree(name.name);
 760        return index;
 761
 762out_dput:
 763        dput(index);
 764        index = NULL;
 765        goto out;
 766
 767fail:
 768        dput(index);
 769        index = ERR_PTR(-EIO);
 770        goto out;
 771}
 772
 773/*
 774 * Returns next layer in stack starting from top.
 775 * Returns -1 if this is the last layer.
 776 */
 777int ovl_path_next(int idx, struct dentry *dentry, struct path *path)
 778{
 779        struct ovl_entry *oe = dentry->d_fsdata;
 780
 781        BUG_ON(idx < 0);
 782        if (idx == 0) {
 783                ovl_path_upper(dentry, path);
 784                if (path->dentry)
 785                        return oe->numlower ? 1 : -1;
 786                idx++;
 787        }
 788        BUG_ON(idx > oe->numlower);
 789        path->dentry = oe->lowerstack[idx - 1].dentry;
 790        path->mnt = oe->lowerstack[idx - 1].layer->mnt;
 791
 792        return (idx < oe->numlower) ? idx + 1 : -1;
 793}
 794
 795/* Fix missing 'origin' xattr */
 796static int ovl_fix_origin(struct ovl_fs *ofs, struct dentry *dentry,
 797                          struct dentry *lower, struct dentry *upper)
 798{
 799        int err;
 800
 801        if (ovl_check_origin_xattr(ofs, upper))
 802                return 0;
 803
 804        err = ovl_want_write(dentry);
 805        if (err)
 806                return err;
 807
 808        err = ovl_set_origin(dentry, lower, upper);
 809        if (!err)
 810                err = ovl_set_impure(dentry->d_parent, upper->d_parent);
 811
 812        ovl_drop_write(dentry);
 813        return err;
 814}
 815
 816struct dentry *ovl_lookup(struct inode *dir, struct dentry *dentry,
 817                          unsigned int flags)
 818{
 819        struct ovl_entry *oe;
 820        const struct cred *old_cred;
 821        struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 822        struct ovl_entry *poe = dentry->d_parent->d_fsdata;
 823        struct ovl_entry *roe = dentry->d_sb->s_root->d_fsdata;
 824        struct ovl_path *stack = NULL, *origin_path = NULL;
 825        struct dentry *upperdir, *upperdentry = NULL;
 826        struct dentry *origin = NULL;
 827        struct dentry *index = NULL;
 828        unsigned int ctr = 0;
 829        struct inode *inode = NULL;
 830        bool upperopaque = false;
 831        char *upperredirect = NULL;
 832        struct dentry *this;
 833        unsigned int i;
 834        int err;
 835        bool uppermetacopy = false;
 836        struct ovl_lookup_data d = {
 837                .sb = dentry->d_sb,
 838                .name = dentry->d_name,
 839                .is_dir = false,
 840                .opaque = false,
 841                .stop = false,
 842                .last = ofs->config.redirect_follow ? false : !poe->numlower,
 843                .redirect = NULL,
 844                .metacopy = false,
 845        };
 846
 847        if (dentry->d_name.len > ofs->namelen)
 848                return ERR_PTR(-ENAMETOOLONG);
 849
 850        old_cred = ovl_override_creds(dentry->d_sb);
 851        upperdir = ovl_dentry_upper(dentry->d_parent);
 852        if (upperdir) {
 853                err = ovl_lookup_layer(upperdir, &d, &upperdentry, true);
 854                if (err)
 855                        goto out;
 856
 857                if (upperdentry && upperdentry->d_flags & DCACHE_OP_REAL) {
 858                        dput(upperdentry);
 859                        err = -EREMOTE;
 860                        goto out;
 861                }
 862                if (upperdentry && !d.is_dir) {
 863                        /*
 864                         * Lookup copy up origin by decoding origin file handle.
 865                         * We may get a disconnected dentry, which is fine,
 866                         * because we only need to hold the origin inode in
 867                         * cache and use its inode number.  We may even get a
 868                         * connected dentry, that is not under any of the lower
 869                         * layers root.  That is also fine for using it's inode
 870                         * number - it's the same as if we held a reference
 871                         * to a dentry in lower layer that was moved under us.
 872                         */
 873                        err = ovl_check_origin(ofs, upperdentry, &origin_path);
 874                        if (err)
 875                                goto out_put_upper;
 876
 877                        if (d.metacopy)
 878                                uppermetacopy = true;
 879                }
 880
 881                if (d.redirect) {
 882                        err = -ENOMEM;
 883                        upperredirect = kstrdup(d.redirect, GFP_KERNEL);
 884                        if (!upperredirect)
 885                                goto out_put_upper;
 886                        if (d.redirect[0] == '/')
 887                                poe = roe;
 888                }
 889                upperopaque = d.opaque;
 890        }
 891
 892        if (!d.stop && poe->numlower) {
 893                err = -ENOMEM;
 894                stack = kcalloc(ofs->numlayer - 1, sizeof(struct ovl_path),
 895                                GFP_KERNEL);
 896                if (!stack)
 897                        goto out_put_upper;
 898        }
 899
 900        for (i = 0; !d.stop && i < poe->numlower; i++) {
 901                struct ovl_path lower = poe->lowerstack[i];
 902
 903                if (!ofs->config.redirect_follow)
 904                        d.last = i == poe->numlower - 1;
 905                else
 906                        d.last = lower.layer->idx == roe->numlower;
 907
 908                err = ovl_lookup_layer(lower.dentry, &d, &this, false);
 909                if (err)
 910                        goto out_put;
 911
 912                if (!this)
 913                        continue;
 914
 915                if ((uppermetacopy || d.metacopy) && !ofs->config.metacopy) {
 916                        err = -EPERM;
 917                        pr_warn_ratelimited("refusing to follow metacopy origin for (%pd2)\n", dentry);
 918                        goto out_put;
 919                }
 920
 921                /*
 922                 * If no origin fh is stored in upper of a merge dir, store fh
 923                 * of lower dir and set upper parent "impure".
 924                 */
 925                if (upperdentry && !ctr && !ofs->noxattr && d.is_dir) {
 926                        err = ovl_fix_origin(ofs, dentry, this, upperdentry);
 927                        if (err) {
 928                                dput(this);
 929                                goto out_put;
 930                        }
 931                }
 932
 933                /*
 934                 * When "verify_lower" feature is enabled, do not merge with a
 935                 * lower dir that does not match a stored origin xattr. In any
 936                 * case, only verified origin is used for index lookup.
 937                 *
 938                 * For non-dir dentry, if index=on, then ensure origin
 939                 * matches the dentry found using path based lookup,
 940                 * otherwise error out.
 941                 */
 942                if (upperdentry && !ctr &&
 943                    ((d.is_dir && ovl_verify_lower(dentry->d_sb)) ||
 944                     (!d.is_dir && ofs->config.index && origin_path))) {
 945                        err = ovl_verify_origin(ofs, upperdentry, this, false);
 946                        if (err) {
 947                                dput(this);
 948                                if (d.is_dir)
 949                                        break;
 950                                goto out_put;
 951                        }
 952                        origin = this;
 953                }
 954
 955                if (d.metacopy && ctr) {
 956                        /*
 957                         * Do not store intermediate metacopy dentries in
 958                         * lower chain, except top most lower metacopy dentry.
 959                         * Continue the loop so that if there is an absolute
 960                         * redirect on this dentry, poe can be reset to roe.
 961                         */
 962                        dput(this);
 963                        this = NULL;
 964                } else {
 965                        stack[ctr].dentry = this;
 966                        stack[ctr].layer = lower.layer;
 967                        ctr++;
 968                }
 969
 970                /*
 971                 * Following redirects can have security consequences: it's like
 972                 * a symlink into the lower layer without the permission checks.
 973                 * This is only a problem if the upper layer is untrusted (e.g
 974                 * comes from an USB drive).  This can allow a non-readable file
 975                 * or directory to become readable.
 976                 *
 977                 * Only following redirects when redirects are enabled disables
 978                 * this attack vector when not necessary.
 979                 */
 980                err = -EPERM;
 981                if (d.redirect && !ofs->config.redirect_follow) {
 982                        pr_warn_ratelimited("refusing to follow redirect for (%pd2)\n",
 983                                            dentry);
 984                        goto out_put;
 985                }
 986
 987                if (d.stop)
 988                        break;
 989
 990                if (d.redirect && d.redirect[0] == '/' && poe != roe) {
 991                        poe = roe;
 992                        /* Find the current layer on the root dentry */
 993                        i = lower.layer->idx - 1;
 994                }
 995        }
 996
 997        /*
 998         * For regular non-metacopy upper dentries, there is no lower
 999         * path based lookup, hence ctr will be zero. If a dentry is found
1000         * using ORIGIN xattr on upper, install it in stack.
1001         *
1002         * For metacopy dentry, path based lookup will find lower dentries.
1003         * Just make sure a corresponding data dentry has been found.
1004         */
1005        if (d.metacopy || (uppermetacopy && !ctr)) {
1006                err = -EIO;
1007                goto out_put;
1008        } else if (!d.is_dir && upperdentry && !ctr && origin_path) {
1009                if (WARN_ON(stack != NULL)) {
1010                        err = -EIO;
1011                        goto out_put;
1012                }
1013                stack = origin_path;
1014                ctr = 1;
1015                origin = origin_path->dentry;
1016                origin_path = NULL;
1017        }
1018
1019        /*
1020         * Always lookup index if there is no-upperdentry.
1021         *
1022         * For the case of upperdentry, we have set origin by now if it
1023         * needed to be set. There are basically three cases.
1024         *
1025         * For directories, lookup index by lower inode and verify it matches
1026         * upper inode. We only trust dir index if we verified that lower dir
1027         * matches origin, otherwise dir index entries may be inconsistent
1028         * and we ignore them.
1029         *
1030         * For regular upper, we already set origin if upper had ORIGIN
1031         * xattr. There is no verification though as there is no path
1032         * based dentry lookup in lower in this case.
1033         *
1034         * For metacopy upper, we set a verified origin already if index
1035         * is enabled and if upper had an ORIGIN xattr.
1036         *
1037         */
1038        if (!upperdentry && ctr)
1039                origin = stack[0].dentry;
1040
1041        if (origin && ovl_indexdir(dentry->d_sb) &&
1042            (!d.is_dir || ovl_index_all(dentry->d_sb))) {
1043                index = ovl_lookup_index(ofs, upperdentry, origin, true);
1044                if (IS_ERR(index)) {
1045                        err = PTR_ERR(index);
1046                        index = NULL;
1047                        goto out_put;
1048                }
1049        }
1050
1051        oe = ovl_alloc_entry(ctr);
1052        err = -ENOMEM;
1053        if (!oe)
1054                goto out_put;
1055
1056        memcpy(oe->lowerstack, stack, sizeof(struct ovl_path) * ctr);
1057        dentry->d_fsdata = oe;
1058
1059        if (upperopaque)
1060                ovl_dentry_set_opaque(dentry);
1061
1062        if (upperdentry)
1063                ovl_dentry_set_upper_alias(dentry);
1064        else if (index) {
1065                upperdentry = dget(index);
1066                upperredirect = ovl_get_redirect_xattr(ofs, upperdentry, 0);
1067                if (IS_ERR(upperredirect)) {
1068                        err = PTR_ERR(upperredirect);
1069                        upperredirect = NULL;
1070                        goto out_free_oe;
1071                }
1072                err = ovl_check_metacopy_xattr(ofs, upperdentry);
1073                if (err < 0)
1074                        goto out_free_oe;
1075                uppermetacopy = err;
1076        }
1077
1078        if (upperdentry || ctr) {
1079                struct ovl_inode_params oip = {
1080                        .upperdentry = upperdentry,
1081                        .lowerpath = stack,
1082                        .index = index,
1083                        .numlower = ctr,
1084                        .redirect = upperredirect,
1085                        .lowerdata = (ctr > 1 && !d.is_dir) ?
1086                                      stack[ctr - 1].dentry : NULL,
1087                };
1088
1089                inode = ovl_get_inode(dentry->d_sb, &oip);
1090                err = PTR_ERR(inode);
1091                if (IS_ERR(inode))
1092                        goto out_free_oe;
1093                if (upperdentry && !uppermetacopy)
1094                        ovl_set_flag(OVL_UPPERDATA, inode);
1095        }
1096
1097        ovl_dentry_update_reval(dentry, upperdentry,
1098                        DCACHE_OP_REVALIDATE | DCACHE_OP_WEAK_REVALIDATE);
1099
1100        revert_creds(old_cred);
1101        if (origin_path) {
1102                dput(origin_path->dentry);
1103                kfree(origin_path);
1104        }
1105        dput(index);
1106        kfree(stack);
1107        kfree(d.redirect);
1108        return d_splice_alias(inode, dentry);
1109
1110out_free_oe:
1111        dentry->d_fsdata = NULL;
1112        kfree(oe);
1113out_put:
1114        dput(index);
1115        for (i = 0; i < ctr; i++)
1116                dput(stack[i].dentry);
1117        kfree(stack);
1118out_put_upper:
1119        if (origin_path) {
1120                dput(origin_path->dentry);
1121                kfree(origin_path);
1122        }
1123        dput(upperdentry);
1124        kfree(upperredirect);
1125out:
1126        kfree(d.redirect);
1127        revert_creds(old_cred);
1128        return ERR_PTR(err);
1129}
1130
1131bool ovl_lower_positive(struct dentry *dentry)
1132{
1133        struct ovl_entry *poe = dentry->d_parent->d_fsdata;
1134        const struct qstr *name = &dentry->d_name;
1135        const struct cred *old_cred;
1136        unsigned int i;
1137        bool positive = false;
1138        bool done = false;
1139
1140        /*
1141         * If dentry is negative, then lower is positive iff this is a
1142         * whiteout.
1143         */
1144        if (!dentry->d_inode)
1145                return ovl_dentry_is_opaque(dentry);
1146
1147        /* Negative upper -> positive lower */
1148        if (!ovl_dentry_upper(dentry))
1149                return true;
1150
1151        old_cred = ovl_override_creds(dentry->d_sb);
1152        /* Positive upper -> have to look up lower to see whether it exists */
1153        for (i = 0; !done && !positive && i < poe->numlower; i++) {
1154                struct dentry *this;
1155                struct dentry *lowerdir = poe->lowerstack[i].dentry;
1156
1157                this = lookup_positive_unlocked(name->name, lowerdir,
1158                                               name->len);
1159                if (IS_ERR(this)) {
1160                        switch (PTR_ERR(this)) {
1161                        case -ENOENT:
1162                        case -ENAMETOOLONG:
1163                                break;
1164
1165                        default:
1166                                /*
1167                                 * Assume something is there, we just couldn't
1168                                 * access it.
1169                                 */
1170                                positive = true;
1171                                break;
1172                        }
1173                } else {
1174                        positive = !ovl_is_whiteout(this);
1175                        done = true;
1176                        dput(this);
1177                }
1178        }
1179        revert_creds(old_cred);
1180
1181        return positive;
1182}
1183