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
  34        buf = ovl_get_redirect_xattr(dentry, prelen + strlen(post));
  35        if (IS_ERR_OR_NULL(buf))
  36                return PTR_ERR(buf);
  37
  38        if (buf[0] == '/') {
  39                /*
  40                 * One of the ancestor path elements in an absolute path
  41                 * lookup in ovl_lookup_layer() could have been opaque and
  42                 * that will stop further lookup in lower layers (d->stop=true)
  43                 * But we have found an absolute redirect in decendant path
  44                 * element and that should force continue lookup in lower
  45                 * layers (reset d->stop).
  46                 */
  47                d->stop = false;
  48        } else {
  49                res = strlen(buf) + 1;
  50                memmove(buf + prelen, buf, res);
  51                memcpy(buf, d->name.name, prelen);
  52        }
  53
  54        strcat(buf, post);
  55        kfree(d->redirect);
  56        d->redirect = buf;
  57        d->name.name = d->redirect;
  58        d->name.len = strlen(d->redirect);
  59
  60        return 0;
  61}
  62
  63static int ovl_acceptable(void *ctx, struct dentry *dentry)
  64{
  65        /*
  66         * A non-dir origin may be disconnected, which is fine, because
  67         * we only need it for its unique inode number.
  68         */
  69        if (!d_is_dir(dentry))
  70                return 1;
  71
  72        /* Don't decode a deleted empty directory */
  73        if (d_unhashed(dentry))
  74                return 0;
  75
  76        /* Check if directory belongs to the layer we are decoding from */
  77        return is_subdir(dentry, ((struct vfsmount *)ctx)->mnt_root);
  78}
  79
  80/*
  81 * Check validity of an overlay file handle buffer.
  82 *
  83 * Return 0 for a valid file handle.
  84 * Return -ENODATA for "origin unknown".
  85 * Return <0 for an invalid file handle.
  86 */
  87int ovl_check_fh_len(struct ovl_fh *fh, int fh_len)
  88{
  89        if (fh_len < sizeof(struct ovl_fh) || fh_len < fh->len)
  90                return -EINVAL;
  91
  92        if (fh->magic != OVL_FH_MAGIC)
  93                return -EINVAL;
  94
  95        /* Treat larger version and unknown flags as "origin unknown" */
  96        if (fh->version > OVL_FH_VERSION || fh->flags & ~OVL_FH_FLAG_ALL)
  97                return -ENODATA;
  98
  99        /* Treat endianness mismatch as "origin unknown" */
 100        if (!(fh->flags & OVL_FH_FLAG_ANY_ENDIAN) &&
 101            (fh->flags & OVL_FH_FLAG_BIG_ENDIAN) != OVL_FH_FLAG_CPU_ENDIAN)
 102                return -ENODATA;
 103
 104        return 0;
 105}
 106
 107static struct ovl_fh *ovl_get_fh(struct dentry *dentry, const char *name)
 108{
 109        int res, err;
 110        struct ovl_fh *fh = NULL;
 111
 112        res = vfs_getxattr(dentry, name, NULL, 0);
 113        if (res < 0) {
 114                if (res == -ENODATA || res == -EOPNOTSUPP)
 115                        return NULL;
 116                goto fail;
 117        }
 118        /* Zero size value means "copied up but origin unknown" */
 119        if (res == 0)
 120                return NULL;
 121
 122        fh = kzalloc(res, GFP_KERNEL);
 123        if (!fh)
 124                return ERR_PTR(-ENOMEM);
 125
 126        res = vfs_getxattr(dentry, name, fh, res);
 127        if (res < 0)
 128                goto fail;
 129
 130        err = ovl_check_fh_len(fh, res);
 131        if (err < 0) {
 132                if (err == -ENODATA)
 133                        goto out;
 134                goto invalid;
 135        }
 136
 137        return fh;
 138
 139out:
 140        kfree(fh);
 141        return NULL;
 142
 143fail:
 144        pr_warn_ratelimited("overlayfs: failed to get origin (%i)\n", res);
 145        goto out;
 146invalid:
 147        pr_warn_ratelimited("overlayfs: invalid origin (%*phN)\n", res, fh);
 148        goto out;
 149}
 150
 151struct dentry *ovl_decode_real_fh(struct ovl_fh *fh, struct vfsmount *mnt,
 152                                  bool connected)
 153{
 154        struct dentry *real;
 155        int bytes;
 156
 157        /*
 158         * Make sure that the stored uuid matches the uuid of the lower
 159         * layer where file handle will be decoded.
 160         */
 161        if (!uuid_equal(&fh->uuid, &mnt->mnt_sb->s_uuid))
 162                return NULL;
 163
 164        bytes = (fh->len - offsetof(struct ovl_fh, fid));
 165        real = exportfs_decode_fh(mnt, (struct fid *)fh->fid,
 166                                  bytes >> 2, (int)fh->type,
 167                                  connected ? ovl_acceptable : NULL, mnt);
 168        if (IS_ERR(real)) {
 169                /*
 170                 * Treat stale file handle to lower file as "origin unknown".
 171                 * upper file handle could become stale when upper file is
 172                 * unlinked and this information is needed to handle stale
 173                 * index entries correctly.
 174                 */
 175                if (real == ERR_PTR(-ESTALE) &&
 176                    !(fh->flags & OVL_FH_FLAG_PATH_UPPER))
 177                        real = NULL;
 178                return real;
 179        }
 180
 181        if (ovl_dentry_weird(real)) {
 182                dput(real);
 183                return NULL;
 184        }
 185
 186        return real;
 187}
 188
 189static bool ovl_is_opaquedir(struct dentry *dentry)
 190{
 191        return ovl_check_dir_xattr(dentry, OVL_XATTR_OPAQUE);
 192}
 193
 194static int ovl_lookup_single(struct dentry *base, struct ovl_lookup_data *d,
 195                             const char *name, unsigned int namelen,
 196                             size_t prelen, const char *post,
 197                             struct dentry **ret)
 198{
 199        struct dentry *this;
 200        int err;
 201        bool last_element = !post[0];
 202
 203        this = lookup_one_len_unlocked(name, base, namelen);
 204        if (IS_ERR(this)) {
 205                err = PTR_ERR(this);
 206                this = NULL;
 207                if (err == -ENOENT || err == -ENAMETOOLONG)
 208                        goto out;
 209                goto out_err;
 210        }
 211        if (!this->d_inode)
 212                goto put_and_out;
 213
 214        if (ovl_dentry_weird(this)) {
 215                /* Don't support traversing automounts and other weirdness */
 216                err = -EREMOTE;
 217                goto out_err;
 218        }
 219        if (ovl_is_whiteout(this)) {
 220                d->stop = d->opaque = true;
 221                goto put_and_out;
 222        }
 223        /*
 224         * This dentry should be a regular file if previous layer lookup
 225         * found a metacopy dentry.
 226         */
 227        if (last_element && d->metacopy && !d_is_reg(this)) {
 228                d->stop = true;
 229                goto put_and_out;
 230        }
 231        if (!d_can_lookup(this)) {
 232                if (d->is_dir || !last_element) {
 233                        d->stop = true;
 234                        goto put_and_out;
 235                }
 236                err = ovl_check_metacopy_xattr(this);
 237                if (err < 0)
 238                        goto out_err;
 239
 240                d->metacopy = err;
 241                d->stop = !d->metacopy;
 242                if (!d->metacopy || d->last)
 243                        goto out;
 244        } else {
 245                if (ovl_lookup_trap_inode(d->sb, this)) {
 246                        /* Caught in a trap of overlapping layers */
 247                        err = -ELOOP;
 248                        goto out_err;
 249                }
 250
 251                if (last_element)
 252                        d->is_dir = true;
 253                if (d->last)
 254                        goto out;
 255
 256                if (ovl_is_opaquedir(this)) {
 257                        d->stop = true;
 258                        if (last_element)
 259                                d->opaque = true;
 260                        goto out;
 261                }
 262        }
 263        err = ovl_check_redirect(this, d, prelen, post);
 264        if (err)
 265                goto out_err;
 266out:
 267        *ret = this;
 268        return 0;
 269
 270put_and_out:
 271        dput(this);
 272        this = NULL;
 273        goto out;
 274
 275out_err:
 276        dput(this);
 277        return err;
 278}
 279
 280static int ovl_lookup_layer(struct dentry *base, struct ovl_lookup_data *d,
 281                            struct dentry **ret)
 282{
 283        /* Counting down from the end, since the prefix can change */
 284        size_t rem = d->name.len - 1;
 285        struct dentry *dentry = NULL;
 286        int err;
 287
 288        if (d->name.name[0] != '/')
 289                return ovl_lookup_single(base, d, d->name.name, d->name.len,
 290                                         0, "", ret);
 291
 292        while (!IS_ERR_OR_NULL(base) && d_can_lookup(base)) {
 293                const char *s = d->name.name + d->name.len - rem;
 294                const char *next = strchrnul(s, '/');
 295                size_t thislen = next - s;
 296                bool end = !next[0];
 297
 298                /* Verify we did not go off the rails */
 299                if (WARN_ON(s[-1] != '/'))
 300                        return -EIO;
 301
 302                err = ovl_lookup_single(base, d, s, thislen,
 303                                        d->name.len - rem, next, &base);
 304                dput(dentry);
 305                if (err)
 306                        return err;
 307                dentry = base;
 308                if (end)
 309                        break;
 310
 311                rem -= thislen + 1;
 312
 313                if (WARN_ON(rem >= d->name.len))
 314                        return -EIO;
 315        }
 316        *ret = dentry;
 317        return 0;
 318}
 319
 320
 321int ovl_check_origin_fh(struct ovl_fs *ofs, struct ovl_fh *fh, bool connected,
 322                        struct dentry *upperdentry, struct ovl_path **stackp)
 323{
 324        struct dentry *origin = NULL;
 325        int i;
 326
 327        for (i = 0; i < ofs->numlower; i++) {
 328                origin = ovl_decode_real_fh(fh, ofs->lower_layers[i].mnt,
 329                                            connected);
 330                if (origin)
 331                        break;
 332        }
 333
 334        if (!origin)
 335                return -ESTALE;
 336        else if (IS_ERR(origin))
 337                return PTR_ERR(origin);
 338
 339        if (upperdentry && !ovl_is_whiteout(upperdentry) &&
 340            ((d_inode(origin)->i_mode ^ d_inode(upperdentry)->i_mode) & S_IFMT))
 341                goto invalid;
 342
 343        if (!*stackp)
 344                *stackp = kmalloc(sizeof(struct ovl_path), GFP_KERNEL);
 345        if (!*stackp) {
 346                dput(origin);
 347                return -ENOMEM;
 348        }
 349        **stackp = (struct ovl_path){
 350                .dentry = origin,
 351                .layer = &ofs->lower_layers[i]
 352        };
 353
 354        return 0;
 355
 356invalid:
 357        pr_warn_ratelimited("overlayfs: invalid origin (%pd2, ftype=%x, origin ftype=%x).\n",
 358                            upperdentry, d_inode(upperdentry)->i_mode & S_IFMT,
 359                            d_inode(origin)->i_mode & S_IFMT);
 360        dput(origin);
 361        return -EIO;
 362}
 363
 364static int ovl_check_origin(struct ovl_fs *ofs, struct dentry *upperdentry,
 365                            struct ovl_path **stackp, unsigned int *ctrp)
 366{
 367        struct ovl_fh *fh = ovl_get_fh(upperdentry, OVL_XATTR_ORIGIN);
 368        int err;
 369
 370        if (IS_ERR_OR_NULL(fh))
 371                return PTR_ERR(fh);
 372
 373        err = ovl_check_origin_fh(ofs, fh, false, upperdentry, stackp);
 374        kfree(fh);
 375
 376        if (err) {
 377                if (err == -ESTALE)
 378                        return 0;
 379                return err;
 380        }
 381
 382        if (WARN_ON(*ctrp))
 383                return -EIO;
 384
 385        *ctrp = 1;
 386        return 0;
 387}
 388
 389/*
 390 * Verify that @fh matches the file handle stored in xattr @name.
 391 * Return 0 on match, -ESTALE on mismatch, < 0 on error.
 392 */
 393static int ovl_verify_fh(struct dentry *dentry, const char *name,
 394                         const struct ovl_fh *fh)
 395{
 396        struct ovl_fh *ofh = ovl_get_fh(dentry, name);
 397        int err = 0;
 398
 399        if (!ofh)
 400                return -ENODATA;
 401
 402        if (IS_ERR(ofh))
 403                return PTR_ERR(ofh);
 404
 405        if (fh->len != ofh->len || memcmp(fh, ofh, fh->len))
 406                err = -ESTALE;
 407
 408        kfree(ofh);
 409        return err;
 410}
 411
 412/*
 413 * Verify that @real dentry matches the file handle stored in xattr @name.
 414 *
 415 * If @set is true and there is no stored file handle, encode @real and store
 416 * file handle in xattr @name.
 417 *
 418 * Return 0 on match, -ESTALE on mismatch, -ENODATA on no xattr, < 0 on error.
 419 */
 420int ovl_verify_set_fh(struct dentry *dentry, const char *name,
 421                      struct dentry *real, bool is_upper, bool set)
 422{
 423        struct inode *inode;
 424        struct ovl_fh *fh;
 425        int err;
 426
 427        fh = ovl_encode_real_fh(real, is_upper);
 428        err = PTR_ERR(fh);
 429        if (IS_ERR(fh)) {
 430                fh = NULL;
 431                goto fail;
 432        }
 433
 434        err = ovl_verify_fh(dentry, name, fh);
 435        if (set && err == -ENODATA)
 436                err = ovl_do_setxattr(dentry, name, fh, fh->len, 0);
 437        if (err)
 438                goto fail;
 439
 440out:
 441        kfree(fh);
 442        return err;
 443
 444fail:
 445        inode = d_inode(real);
 446        pr_warn_ratelimited("overlayfs: failed to verify %s (%pd2, ino=%lu, err=%i)\n",
 447                            is_upper ? "upper" : "origin", real,
 448                            inode ? inode->i_ino : 0, err);
 449        goto out;
 450}
 451
 452/* Get upper dentry from index */
 453struct dentry *ovl_index_upper(struct ovl_fs *ofs, struct dentry *index)
 454{
 455        struct ovl_fh *fh;
 456        struct dentry *upper;
 457
 458        if (!d_is_dir(index))
 459                return dget(index);
 460
 461        fh = ovl_get_fh(index, OVL_XATTR_UPPER);
 462        if (IS_ERR_OR_NULL(fh))
 463                return ERR_CAST(fh);
 464
 465        upper = ovl_decode_real_fh(fh, ofs->upper_mnt, true);
 466        kfree(fh);
 467
 468        if (IS_ERR_OR_NULL(upper))
 469                return upper ?: ERR_PTR(-ESTALE);
 470
 471        if (!d_is_dir(upper)) {
 472                pr_warn_ratelimited("overlayfs: invalid index upper (%pd2, upper=%pd2).\n",
 473                                    index, upper);
 474                dput(upper);
 475                return ERR_PTR(-EIO);
 476        }
 477
 478        return upper;
 479}
 480
 481/* Is this a leftover from create/whiteout of directory index entry? */
 482static bool ovl_is_temp_index(struct dentry *index)
 483{
 484        return index->d_name.name[0] == '#';
 485}
 486
 487/*
 488 * Verify that an index entry name matches the origin file handle stored in
 489 * OVL_XATTR_ORIGIN and that origin file handle can be decoded to lower path.
 490 * Return 0 on match, -ESTALE on mismatch or stale origin, < 0 on error.
 491 */
 492int ovl_verify_index(struct ovl_fs *ofs, struct dentry *index)
 493{
 494        struct ovl_fh *fh = NULL;
 495        size_t len;
 496        struct ovl_path origin = { };
 497        struct ovl_path *stack = &origin;
 498        struct dentry *upper = NULL;
 499        int err;
 500
 501        if (!d_inode(index))
 502                return 0;
 503
 504        /* Cleanup leftover from index create/cleanup attempt */
 505        err = -ESTALE;
 506        if (ovl_is_temp_index(index))
 507                goto fail;
 508
 509        err = -EINVAL;
 510        if (index->d_name.len < sizeof(struct ovl_fh)*2)
 511                goto fail;
 512
 513        err = -ENOMEM;
 514        len = index->d_name.len / 2;
 515        fh = kzalloc(len, GFP_KERNEL);
 516        if (!fh)
 517                goto fail;
 518
 519        err = -EINVAL;
 520        if (hex2bin((u8 *)fh, index->d_name.name, len))
 521                goto fail;
 522
 523        err = ovl_check_fh_len(fh, len);
 524        if (err)
 525                goto fail;
 526
 527        /*
 528         * Whiteout index entries are used as an indication that an exported
 529         * overlay file handle should be treated as stale (i.e. after unlink
 530         * of the overlay inode). These entries contain no origin xattr.
 531         */
 532        if (ovl_is_whiteout(index))
 533                goto out;
 534
 535        /*
 536         * Verifying directory index entries are not stale is expensive, so
 537         * only verify stale dir index if NFS export is enabled.
 538         */
 539        if (d_is_dir(index) && !ofs->config.nfs_export)
 540                goto out;
 541
 542        /*
 543         * Directory index entries should have 'upper' xattr pointing to the
 544         * real upper dir. Non-dir index entries are hardlinks to the upper
 545         * real inode. For non-dir index, we can read the copy up origin xattr
 546         * directly from the index dentry, but for dir index we first need to
 547         * decode the upper directory.
 548         */
 549        upper = ovl_index_upper(ofs, index);
 550        if (IS_ERR_OR_NULL(upper)) {
 551                err = PTR_ERR(upper);
 552                /*
 553                 * Directory index entries with no 'upper' xattr need to be
 554                 * removed. When dir index entry has a stale 'upper' xattr,
 555                 * we assume that upper dir was removed and we treat the dir
 556                 * index as orphan entry that needs to be whited out.
 557                 */
 558                if (err == -ESTALE)
 559                        goto orphan;
 560                else if (!err)
 561                        err = -ESTALE;
 562                goto fail;
 563        }
 564
 565        err = ovl_verify_fh(upper, OVL_XATTR_ORIGIN, fh);
 566        dput(upper);
 567        if (err)
 568                goto fail;
 569
 570        /* Check if non-dir index is orphan and don't warn before cleaning it */
 571        if (!d_is_dir(index) && d_inode(index)->i_nlink == 1) {
 572                err = ovl_check_origin_fh(ofs, fh, false, index, &stack);
 573                if (err)
 574                        goto fail;
 575
 576                if (ovl_get_nlink(origin.dentry, index, 0) == 0)
 577                        goto orphan;
 578        }
 579
 580out:
 581        dput(origin.dentry);
 582        kfree(fh);
 583        return err;
 584
 585fail:
 586        pr_warn_ratelimited("overlayfs: failed to verify index (%pd2, ftype=%x, err=%i)\n",
 587                            index, d_inode(index)->i_mode & S_IFMT, err);
 588        goto out;
 589
 590orphan:
 591        pr_warn_ratelimited("overlayfs: orphan index entry (%pd2, ftype=%x, nlink=%u)\n",
 592                            index, d_inode(index)->i_mode & S_IFMT,
 593                            d_inode(index)->i_nlink);
 594        err = -ENOENT;
 595        goto out;
 596}
 597
 598static int ovl_get_index_name_fh(struct ovl_fh *fh, struct qstr *name)
 599{
 600        char *n, *s;
 601
 602        n = kcalloc(fh->len, 2, GFP_KERNEL);
 603        if (!n)
 604                return -ENOMEM;
 605
 606        s  = bin2hex(n, fh, fh->len);
 607        *name = (struct qstr) QSTR_INIT(n, s - n);
 608
 609        return 0;
 610
 611}
 612
 613/*
 614 * Lookup in indexdir for the index entry of a lower real inode or a copy up
 615 * origin inode. The index entry name is the hex representation of the lower
 616 * inode file handle.
 617 *
 618 * If the index dentry in negative, then either no lower aliases have been
 619 * copied up yet, or aliases have been copied up in older kernels and are
 620 * not indexed.
 621 *
 622 * If the index dentry for a copy up origin inode is positive, but points
 623 * to an inode different than the upper inode, then either the upper inode
 624 * has been copied up and not indexed or it was indexed, but since then
 625 * index dir was cleared. Either way, that index cannot be used to indentify
 626 * the overlay inode.
 627 */
 628int ovl_get_index_name(struct dentry *origin, struct qstr *name)
 629{
 630        struct ovl_fh *fh;
 631        int err;
 632
 633        fh = ovl_encode_real_fh(origin, false);
 634        if (IS_ERR(fh))
 635                return PTR_ERR(fh);
 636
 637        err = ovl_get_index_name_fh(fh, name);
 638
 639        kfree(fh);
 640        return err;
 641}
 642
 643/* Lookup index by file handle for NFS export */
 644struct dentry *ovl_get_index_fh(struct ovl_fs *ofs, struct ovl_fh *fh)
 645{
 646        struct dentry *index;
 647        struct qstr name;
 648        int err;
 649
 650        err = ovl_get_index_name_fh(fh, &name);
 651        if (err)
 652                return ERR_PTR(err);
 653
 654        index = lookup_one_len_unlocked(name.name, ofs->indexdir, name.len);
 655        kfree(name.name);
 656        if (IS_ERR(index)) {
 657                if (PTR_ERR(index) == -ENOENT)
 658                        index = NULL;
 659                return index;
 660        }
 661
 662        if (d_is_negative(index))
 663                err = 0;
 664        else if (ovl_is_whiteout(index))
 665                err = -ESTALE;
 666        else if (ovl_dentry_weird(index))
 667                err = -EIO;
 668        else
 669                return index;
 670
 671        dput(index);
 672        return ERR_PTR(err);
 673}
 674
 675struct dentry *ovl_lookup_index(struct ovl_fs *ofs, struct dentry *upper,
 676                                struct dentry *origin, bool verify)
 677{
 678        struct dentry *index;
 679        struct inode *inode;
 680        struct qstr name;
 681        bool is_dir = d_is_dir(origin);
 682        int err;
 683
 684        err = ovl_get_index_name(origin, &name);
 685        if (err)
 686                return ERR_PTR(err);
 687
 688        index = lookup_one_len_unlocked(name.name, ofs->indexdir, name.len);
 689        if (IS_ERR(index)) {
 690                err = PTR_ERR(index);
 691                if (err == -ENOENT) {
 692                        index = NULL;
 693                        goto out;
 694                }
 695                pr_warn_ratelimited("overlayfs: failed inode index lookup (ino=%lu, key=%.*s, err=%i);\n"
 696                                    "overlayfs: mount with '-o index=off' to disable inodes index.\n",
 697                                    d_inode(origin)->i_ino, name.len, name.name,
 698                                    err);
 699                goto out;
 700        }
 701
 702        inode = d_inode(index);
 703        if (d_is_negative(index)) {
 704                goto out_dput;
 705        } else if (ovl_is_whiteout(index) && !verify) {
 706                /*
 707                 * When index lookup is called with !verify for decoding an
 708                 * overlay file handle, a whiteout index implies that decode
 709                 * should treat file handle as stale and no need to print a
 710                 * warning about it.
 711                 */
 712                dput(index);
 713                index = ERR_PTR(-ESTALE);
 714                goto out;
 715        } else if (ovl_dentry_weird(index) || ovl_is_whiteout(index) ||
 716                   ((inode->i_mode ^ d_inode(origin)->i_mode) & S_IFMT)) {
 717                /*
 718                 * Index should always be of the same file type as origin
 719                 * except for the case of a whiteout index. A whiteout
 720                 * index should only exist if all lower aliases have been
 721                 * unlinked, which means that finding a lower origin on lookup
 722                 * whose index is a whiteout should be treated as an error.
 723                 */
 724                pr_warn_ratelimited("overlayfs: bad index found (index=%pd2, ftype=%x, origin ftype=%x).\n",
 725                                    index, d_inode(index)->i_mode & S_IFMT,
 726                                    d_inode(origin)->i_mode & S_IFMT);
 727                goto fail;
 728        } else if (is_dir && verify) {
 729                if (!upper) {
 730                        pr_warn_ratelimited("overlayfs: suspected uncovered redirected dir found (origin=%pd2, index=%pd2).\n",
 731                                            origin, index);
 732                        goto fail;
 733                }
 734
 735                /* Verify that dir index 'upper' xattr points to upper dir */
 736                err = ovl_verify_upper(index, upper, false);
 737                if (err) {
 738                        if (err == -ESTALE) {
 739                                pr_warn_ratelimited("overlayfs: suspected multiply redirected dir found (upper=%pd2, origin=%pd2, index=%pd2).\n",
 740                                                    upper, origin, index);
 741                        }
 742                        goto fail;
 743                }
 744        } else if (upper && d_inode(upper) != inode) {
 745                goto out_dput;
 746        }
 747out:
 748        kfree(name.name);
 749        return index;
 750
 751out_dput:
 752        dput(index);
 753        index = NULL;
 754        goto out;
 755
 756fail:
 757        dput(index);
 758        index = ERR_PTR(-EIO);
 759        goto out;
 760}
 761
 762/*
 763 * Returns next layer in stack starting from top.
 764 * Returns -1 if this is the last layer.
 765 */
 766int ovl_path_next(int idx, struct dentry *dentry, struct path *path)
 767{
 768        struct ovl_entry *oe = dentry->d_fsdata;
 769
 770        BUG_ON(idx < 0);
 771        if (idx == 0) {
 772                ovl_path_upper(dentry, path);
 773                if (path->dentry)
 774                        return oe->numlower ? 1 : -1;
 775                idx++;
 776        }
 777        BUG_ON(idx > oe->numlower);
 778        path->dentry = oe->lowerstack[idx - 1].dentry;
 779        path->mnt = oe->lowerstack[idx - 1].layer->mnt;
 780
 781        return (idx < oe->numlower) ? idx + 1 : -1;
 782}
 783
 784/* Fix missing 'origin' xattr */
 785static int ovl_fix_origin(struct dentry *dentry, struct dentry *lower,
 786                          struct dentry *upper)
 787{
 788        int err;
 789
 790        if (ovl_check_origin_xattr(upper))
 791                return 0;
 792
 793        err = ovl_want_write(dentry);
 794        if (err)
 795                return err;
 796
 797        err = ovl_set_origin(dentry, lower, upper);
 798        if (!err)
 799                err = ovl_set_impure(dentry->d_parent, upper->d_parent);
 800
 801        ovl_drop_write(dentry);
 802        return err;
 803}
 804
 805struct dentry *ovl_lookup(struct inode *dir, struct dentry *dentry,
 806                          unsigned int flags)
 807{
 808        struct ovl_entry *oe;
 809        const struct cred *old_cred;
 810        struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 811        struct ovl_entry *poe = dentry->d_parent->d_fsdata;
 812        struct ovl_entry *roe = dentry->d_sb->s_root->d_fsdata;
 813        struct ovl_path *stack = NULL, *origin_path = NULL;
 814        struct dentry *upperdir, *upperdentry = NULL;
 815        struct dentry *origin = NULL;
 816        struct dentry *index = NULL;
 817        unsigned int ctr = 0;
 818        struct inode *inode = NULL;
 819        bool upperopaque = false;
 820        char *upperredirect = NULL;
 821        struct dentry *this;
 822        unsigned int i;
 823        int err;
 824        bool metacopy = false;
 825        struct ovl_lookup_data d = {
 826                .sb = dentry->d_sb,
 827                .name = dentry->d_name,
 828                .is_dir = false,
 829                .opaque = false,
 830                .stop = false,
 831                .last = ofs->config.redirect_follow ? false : !poe->numlower,
 832                .redirect = NULL,
 833                .metacopy = false,
 834        };
 835
 836        if (dentry->d_name.len > ofs->namelen)
 837                return ERR_PTR(-ENAMETOOLONG);
 838
 839        old_cred = ovl_override_creds(dentry->d_sb);
 840        upperdir = ovl_dentry_upper(dentry->d_parent);
 841        if (upperdir) {
 842                err = ovl_lookup_layer(upperdir, &d, &upperdentry);
 843                if (err)
 844                        goto out;
 845
 846                if (upperdentry && unlikely(ovl_dentry_remote(upperdentry))) {
 847                        dput(upperdentry);
 848                        err = -EREMOTE;
 849                        goto out;
 850                }
 851                if (upperdentry && !d.is_dir) {
 852                        unsigned int origin_ctr = 0;
 853
 854                        /*
 855                         * Lookup copy up origin by decoding origin file handle.
 856                         * We may get a disconnected dentry, which is fine,
 857                         * because we only need to hold the origin inode in
 858                         * cache and use its inode number.  We may even get a
 859                         * connected dentry, that is not under any of the lower
 860                         * layers root.  That is also fine for using it's inode
 861                         * number - it's the same as if we held a reference
 862                         * to a dentry in lower layer that was moved under us.
 863                         */
 864                        err = ovl_check_origin(ofs, upperdentry, &origin_path,
 865                                               &origin_ctr);
 866                        if (err)
 867                                goto out_put_upper;
 868
 869                        if (d.metacopy)
 870                                metacopy = true;
 871                }
 872
 873                if (d.redirect) {
 874                        err = -ENOMEM;
 875                        upperredirect = kstrdup(d.redirect, GFP_KERNEL);
 876                        if (!upperredirect)
 877                                goto out_put_upper;
 878                        if (d.redirect[0] == '/')
 879                                poe = roe;
 880                }
 881                upperopaque = d.opaque;
 882        }
 883
 884        if (!d.stop && poe->numlower) {
 885                err = -ENOMEM;
 886                stack = kcalloc(ofs->numlower, sizeof(struct ovl_path),
 887                                GFP_KERNEL);
 888                if (!stack)
 889                        goto out_put_upper;
 890        }
 891
 892        for (i = 0; !d.stop && i < poe->numlower; i++) {
 893                struct ovl_path lower = poe->lowerstack[i];
 894
 895                if (!ofs->config.redirect_follow)
 896                        d.last = i == poe->numlower - 1;
 897                else
 898                        d.last = lower.layer->idx == roe->numlower;
 899
 900                err = ovl_lookup_layer(lower.dentry, &d, &this);
 901                if (err)
 902                        goto out_put;
 903
 904                if (!this)
 905                        continue;
 906
 907                /*
 908                 * If no origin fh is stored in upper of a merge dir, store fh
 909                 * of lower dir and set upper parent "impure".
 910                 */
 911                if (upperdentry && !ctr && !ofs->noxattr && d.is_dir) {
 912                        err = ovl_fix_origin(dentry, this, upperdentry);
 913                        if (err) {
 914                                dput(this);
 915                                goto out_put;
 916                        }
 917                }
 918
 919                /*
 920                 * When "verify_lower" feature is enabled, do not merge with a
 921                 * lower dir that does not match a stored origin xattr. In any
 922                 * case, only verified origin is used for index lookup.
 923                 *
 924                 * For non-dir dentry, if index=on, then ensure origin
 925                 * matches the dentry found using path based lookup,
 926                 * otherwise error out.
 927                 */
 928                if (upperdentry && !ctr &&
 929                    ((d.is_dir && ovl_verify_lower(dentry->d_sb)) ||
 930                     (!d.is_dir && ofs->config.index && origin_path))) {
 931                        err = ovl_verify_origin(upperdentry, this, false);
 932                        if (err) {
 933                                dput(this);
 934                                if (d.is_dir)
 935                                        break;
 936                                goto out_put;
 937                        }
 938                        origin = this;
 939                }
 940
 941                if (d.metacopy)
 942                        metacopy = true;
 943                /*
 944                 * Do not store intermediate metacopy dentries in chain,
 945                 * except top most lower metacopy dentry
 946                 */
 947                if (d.metacopy && ctr) {
 948                        dput(this);
 949                        continue;
 950                }
 951
 952                stack[ctr].dentry = this;
 953                stack[ctr].layer = lower.layer;
 954                ctr++;
 955
 956                /*
 957                 * Following redirects can have security consequences: it's like
 958                 * a symlink into the lower layer without the permission checks.
 959                 * This is only a problem if the upper layer is untrusted (e.g
 960                 * comes from an USB drive).  This can allow a non-readable file
 961                 * or directory to become readable.
 962                 *
 963                 * Only following redirects when redirects are enabled disables
 964                 * this attack vector when not necessary.
 965                 */
 966                err = -EPERM;
 967                if (d.redirect && !ofs->config.redirect_follow) {
 968                        pr_warn_ratelimited("overlayfs: refusing to follow redirect for (%pd2)\n",
 969                                            dentry);
 970                        goto out_put;
 971                }
 972
 973                if (d.stop)
 974                        break;
 975
 976                if (d.redirect && d.redirect[0] == '/' && poe != roe) {
 977                        poe = roe;
 978                        /* Find the current layer on the root dentry */
 979                        i = lower.layer->idx - 1;
 980                }
 981        }
 982
 983        if (metacopy) {
 984                /*
 985                 * Found a metacopy dentry but did not find corresponding
 986                 * data dentry
 987                 */
 988                if (d.metacopy) {
 989                        err = -EIO;
 990                        goto out_put;
 991                }
 992
 993                err = -EPERM;
 994                if (!ofs->config.metacopy) {
 995                        pr_warn_ratelimited("overlay: refusing to follow metacopy origin for (%pd2)\n",
 996                                            dentry);
 997                        goto out_put;
 998                }
 999        } else if (!d.is_dir && upperdentry && !ctr && origin_path) {
1000                if (WARN_ON(stack != NULL)) {
1001                        err = -EIO;
1002                        goto out_put;
1003                }
1004                stack = origin_path;
1005                ctr = 1;
1006                origin_path = NULL;
1007        }
1008
1009        /*
1010         * Lookup index by lower inode and verify it matches upper inode.
1011         * We only trust dir index if we verified that lower dir matches
1012         * origin, otherwise dir index entries may be inconsistent and we
1013         * ignore them.
1014         *
1015         * For non-dir upper metacopy dentry, we already set "origin" if we
1016         * verified that lower matched upper origin. If upper origin was
1017         * not present (because lower layer did not support fh encode/decode),
1018         * or indexing is not enabled, do not set "origin" and skip looking up
1019         * index. This case should be handled in same way as a non-dir upper
1020         * without ORIGIN is handled.
1021         *
1022         * Always lookup index of non-dir non-metacopy and non-upper.
1023         */
1024        if (ctr && (!upperdentry || (!d.is_dir && !metacopy)))
1025                origin = stack[0].dentry;
1026
1027        if (origin && ovl_indexdir(dentry->d_sb) &&
1028            (!d.is_dir || ovl_index_all(dentry->d_sb))) {
1029                index = ovl_lookup_index(ofs, upperdentry, origin, true);
1030                if (IS_ERR(index)) {
1031                        err = PTR_ERR(index);
1032                        index = NULL;
1033                        goto out_put;
1034                }
1035        }
1036
1037        oe = ovl_alloc_entry(ctr);
1038        err = -ENOMEM;
1039        if (!oe)
1040                goto out_put;
1041
1042        memcpy(oe->lowerstack, stack, sizeof(struct ovl_path) * ctr);
1043        dentry->d_fsdata = oe;
1044
1045        if (upperopaque)
1046                ovl_dentry_set_opaque(dentry);
1047
1048        if (upperdentry)
1049                ovl_dentry_set_upper_alias(dentry);
1050        else if (index) {
1051                upperdentry = dget(index);
1052                upperredirect = ovl_get_redirect_xattr(upperdentry, 0);
1053                if (IS_ERR(upperredirect)) {
1054                        err = PTR_ERR(upperredirect);
1055                        upperredirect = NULL;
1056                        goto out_free_oe;
1057                }
1058        }
1059
1060        if (upperdentry || ctr) {
1061                struct ovl_inode_params oip = {
1062                        .upperdentry = upperdentry,
1063                        .lowerpath = stack,
1064                        .index = index,
1065                        .numlower = ctr,
1066                        .redirect = upperredirect,
1067                        .lowerdata = (ctr > 1 && !d.is_dir) ?
1068                                      stack[ctr - 1].dentry : NULL,
1069                };
1070
1071                inode = ovl_get_inode(dentry->d_sb, &oip);
1072                err = PTR_ERR(inode);
1073                if (IS_ERR(inode))
1074                        goto out_free_oe;
1075        }
1076
1077        revert_creds(old_cred);
1078        if (origin_path) {
1079                dput(origin_path->dentry);
1080                kfree(origin_path);
1081        }
1082        dput(index);
1083        kfree(stack);
1084        kfree(d.redirect);
1085        return d_splice_alias(inode, dentry);
1086
1087out_free_oe:
1088        dentry->d_fsdata = NULL;
1089        kfree(oe);
1090out_put:
1091        dput(index);
1092        for (i = 0; i < ctr; i++)
1093                dput(stack[i].dentry);
1094        kfree(stack);
1095out_put_upper:
1096        if (origin_path) {
1097                dput(origin_path->dentry);
1098                kfree(origin_path);
1099        }
1100        dput(upperdentry);
1101        kfree(upperredirect);
1102out:
1103        kfree(d.redirect);
1104        revert_creds(old_cred);
1105        return ERR_PTR(err);
1106}
1107
1108bool ovl_lower_positive(struct dentry *dentry)
1109{
1110        struct ovl_entry *poe = dentry->d_parent->d_fsdata;
1111        const struct qstr *name = &dentry->d_name;
1112        const struct cred *old_cred;
1113        unsigned int i;
1114        bool positive = false;
1115        bool done = false;
1116
1117        /*
1118         * If dentry is negative, then lower is positive iff this is a
1119         * whiteout.
1120         */
1121        if (!dentry->d_inode)
1122                return ovl_dentry_is_opaque(dentry);
1123
1124        /* Negative upper -> positive lower */
1125        if (!ovl_dentry_upper(dentry))
1126                return true;
1127
1128        old_cred = ovl_override_creds(dentry->d_sb);
1129        /* Positive upper -> have to look up lower to see whether it exists */
1130        for (i = 0; !done && !positive && i < poe->numlower; i++) {
1131                struct dentry *this;
1132                struct dentry *lowerdir = poe->lowerstack[i].dentry;
1133
1134                this = lookup_one_len_unlocked(name->name, lowerdir,
1135                                               name->len);
1136                if (IS_ERR(this)) {
1137                        switch (PTR_ERR(this)) {
1138                        case -ENOENT:
1139                        case -ENAMETOOLONG:
1140                                break;
1141
1142                        default:
1143                                /*
1144                                 * Assume something is there, we just couldn't
1145                                 * access it.
1146                                 */
1147                                positive = true;
1148                                break;
1149                        }
1150                } else {
1151                        if (this->d_inode) {
1152                                positive = !ovl_is_whiteout(this);
1153                                done = true;
1154                        }
1155                        dput(this);
1156                }
1157        }
1158        revert_creds(old_cred);
1159
1160        return positive;
1161}
1162