fs/overlayfs/dir.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  *
   4  * Copyright (C) 2011 Novell Inc.
   5  */
   6
   7 #include <linux/fs.h>
   8 #include <linux/namei.h>
   9 #include <linux/xattr.h>
  10 #include <linux/security.h>
  11 #include <linux/cred.h>
  12 #include <linux/module.h>
  13 #include <linux/posix_acl.h>
  14 #include <linux/posix_acl_xattr.h>
  15 #include <linux/atomic.h>
  16 #include <linux/ratelimit.h>
  17 #include "overlayfs.h"
  18
  19 static unsigned short ovl_redirect_max = 256;
  20 module_param_named(redirect_max, ovl_redirect_max, ushort, 0644);
  21 MODULE_PARM_DESC(redirect_max,
  22                  "Maximum length of absolute redirect xattr value");
  23
  24 static int ovl_set_redirect(struct dentry *dentry, bool samedir);
  25
  26 int ovl_cleanup(struct inode *wdir, struct dentry *wdentry)
  27 {
  28         int err;
  29
  30         dget(wdentry);
  31         if (d_is_dir(wdentry))
  32                 err = ovl_do_rmdir(wdir, wdentry);
  33         else
  34                 err = ovl_do_unlink(wdir, wdentry);
  35         dput(wdentry);
  36
  37         if (err) {
  38                 pr_err("overlayfs: cleanup of '%pd2' failed (%i)\n",
  39                        wdentry, err);
  40         }
  41
  42         return err;
  43 }
  44
  45 static struct dentry *ovl_lookup_temp(struct dentry *workdir)
  46 {
  47         struct dentry *temp;
  48         char name[20];
  49         static atomic_t temp_id = ATOMIC_INIT(0);
  50
  51         /* counter is allowed to wrap, since temp dentries are ephemeral */
  52         snprintf(name, sizeof(name), "#%x", atomic_inc_return(&temp_id));
  53
  54         temp = lookup_one_len(name, workdir, strlen(name));
  55         if (!IS_ERR(temp) && temp->d_inode) {
  56                 pr_err("overlayfs: workdir/%s already exists\n", name);
  57                 dput(temp);
  58                 temp = ERR_PTR(-EIO);
  59         }
  60
  61         return temp;
  62 }
  63
  64 /* caller holds i_mutex on workdir */
  65 static struct dentry *ovl_whiteout(struct dentry *workdir)
  66 {
  67         int err;
  68         struct dentry *whiteout;
  69         struct inode *wdir = workdir->d_inode;
  70
  71         whiteout = ovl_lookup_temp(workdir);
  72         if (IS_ERR(whiteout))
  73                 return whiteout;
  74
  75         err = ovl_do_whiteout(wdir, whiteout);
  76         if (err) {
  77                 dput(whiteout);
  78                 whiteout = ERR_PTR(err);
  79         }
  80
  81         return whiteout;
  82 }
  83
  84 /* Caller must hold i_mutex on both workdir and dir */
  85 int ovl_cleanup_and_whiteout(struct dentry *workdir, struct inode *dir,
  86                              struct dentry *dentry)
  87 {
  88         struct inode *wdir = workdir->d_inode;
  89         struct dentry *whiteout;
  90         int err;
  91         int flags = 0;
  92
  93         whiteout = ovl_whiteout(workdir);
  94         err = PTR_ERR(whiteout);
  95         if (IS_ERR(whiteout))
  96                 return err;
  97
  98         if (d_is_dir(dentry))
  99                 flags = RENAME_EXCHANGE;
 100
 101         err = ovl_do_rename(wdir, whiteout, dir, dentry, flags);
 102         if (err)
 103                 goto kill_whiteout;
 104         if (flags)
 105                 ovl_cleanup(wdir, dentry);
 106
 107 out:
 108         dput(whiteout);
 109         return err;
 110
 111 kill_whiteout:
 112         ovl_cleanup(wdir, whiteout);
 113         goto out;
 114 }
 115
 116 static int ovl_mkdir_real(struct inode *dir, struct dentry **newdentry,
 117                           umode_t mode)
 118 {
 119         int err;
 120         struct dentry *d, *dentry = *newdentry;
 121
 122         err = ovl_do_mkdir(dir, dentry, mode);
 123         if (err)
 124                 return err;
 125
 126         if (likely(!d_unhashed(dentry)))
 127                 return 0;
 128
 129         /*
 130          * vfs_mkdir() may succeed and leave the dentry passed
 131          * to it unhashed and negative. If that happens, try to
 132          * lookup a new hashed and positive dentry.
 133          */
 134         d = lookup_one_len(dentry->d_name.name, dentry->d_parent,
 135                            dentry->d_name.len);
 136         if (IS_ERR(d)) {
 137                 pr_warn("overlayfs: failed lookup after mkdir (%pd2, err=%i).\n",
 138                         dentry, err);
 139                 return PTR_ERR(d);
 140         }
 141         dput(dentry);
 142         *newdentry = d;
 143
 144         return 0;
 145 }
 146
 147 struct dentry *ovl_create_real(struct inode *dir, struct dentry *newdentry,
 148                                struct ovl_cattr *attr)
 149 {
 150         int err;
 151
 152         if (IS_ERR(newdentry))
 153                 return newdentry;
 154
 155         err = -ESTALE;
 156         if (newdentry->d_inode)
 157                 goto out;
 158
 159         if (attr->hardlink) {
 160                 err = ovl_do_link(attr->hardlink, dir, newdentry);
 161         } else {
 162                 switch (attr->mode & S_IFMT) {
 163                 case S_IFREG:
 164                         err = ovl_do_create(dir, newdentry, attr->mode);
 165                         break;
 166
 167                 case S_IFDIR:
 168                         /* mkdir is special... */
 169                         err =  ovl_mkdir_real(dir, &newdentry, attr->mode);
 170                         break;
 171
 172                 case S_IFCHR:
 173                 case S_IFBLK:
 174                 case S_IFIFO:
 175                 case S_IFSOCK:
 176                         err = ovl_do_mknod(dir, newdentry, attr->mode,
 177                                            attr->rdev);
 178                         break;
 179
 180                 case S_IFLNK:
 181                         err = ovl_do_symlink(dir, newdentry, attr->link);
 182                         break;
 183
 184                 default:
 185                         err = -EPERM;
 186                 }
 187         }
 188         if (!err && WARN_ON(!newdentry->d_inode)) {
 189                 /*
 190                  * Not quite sure if non-instantiated dentry is legal or not.
 191                  * VFS doesn't seem to care so check and warn here.
 192                  */
 193                 err = -EIO;
 194         }
 195 out:
 196         if (err) {
 197                 dput(newdentry);
 198                 return ERR_PTR(err);
 199         }
 200         return newdentry;
 201 }
 202
 203 struct dentry *ovl_create_temp(struct dentry *workdir, struct ovl_cattr *attr)
 204 {
 205         return ovl_create_real(d_inode(workdir), ovl_lookup_temp(workdir),
 206                                attr);
 207 }
 208
 209 static int ovl_set_opaque_xerr(struct dentry *dentry, struct dentry *upper,
 210                                int xerr)
 211 {
 212         int err;
 213
 214         err = ovl_check_setxattr(dentry, upper, OVL_XATTR_OPAQUE, "y", 1, xerr);
 215         if (!err)
 216                 ovl_dentry_set_opaque(dentry);
 217
 218         return err;
 219 }
 220
 221 static int ovl_set_opaque(struct dentry *dentry, struct dentry *upperdentry)
 222 {
 223         /*
 224          * Fail with -EIO when trying to create opaque dir and upper doesn't
 225          * support xattrs. ovl_rename() calls ovl_set_opaque_xerr(-EXDEV) to
 226          * return a specific error for noxattr case.
 227          */
 228         return ovl_set_opaque_xerr(dentry, upperdentry, -EIO);
 229 }
 230
 231 /*
 232  * Common operations required to be done after creation of file on upper.
 233  * If @hardlink is false, then @inode is a pre-allocated inode, we may or
 234  * may not use to instantiate the new dentry.
 235  */
 236 static int ovl_instantiate(struct dentry *dentry, struct inode *inode,
 237                            struct dentry *newdentry, bool hardlink)
 238 {
 239         struct ovl_inode_params oip = {
 240                 .upperdentry = newdentry,
 241                 .newinode = inode,
 242         };
 243
 244         ovl_dir_modified(dentry->d_parent, false);
 245         ovl_dentry_set_upper_alias(dentry);
 246         if (!hardlink) {
 247                 /*
 248                  * ovl_obtain_alias() can be called after ovl_create_real()
 249                  * and before we get here, so we may get an inode from cache
 250                  * with the same real upperdentry that is not the inode we
 251                  * pre-allocated.  In this case we will use the cached inode
 252                  * to instantiate the new dentry.
 253                  *
 254                  * XXX: if we ever use ovl_obtain_alias() to decode directory
 255                  * file handles, need to use ovl_get_inode_locked() and
 256                  * d_instantiate_new() here to prevent from creating two
 257                  * hashed directory inode aliases.
 258                  */
 259                 inode = ovl_get_inode(dentry->d_sb, &oip);
 260                 if (IS_ERR(inode))
 261                         return PTR_ERR(inode);
 262         } else {
 263                 WARN_ON(ovl_inode_real(inode) != d_inode(newdentry));
 264                 dput(newdentry);
 265                 inc_nlink(inode);
 266         }
 267
 268         d_instantiate(dentry, inode);
 269         if (inode != oip.newinode) {
 270                 pr_warn_ratelimited("overlayfs: newly created inode found in cache (%pd2)\n",
 271                                     dentry);
 272         }
 273
 274         /* Force lookup of new upper hardlink to find its lower */
 275         if (hardlink)
 276                 d_drop(dentry);
 277
 278         return 0;
 279 }
 280
 281 static bool ovl_type_merge(struct dentry *dentry)
 282 {
 283         return OVL_TYPE_MERGE(ovl_path_type(dentry));
 284 }
 285
 286 static bool ovl_type_origin(struct dentry *dentry)
 287 {
 288         return OVL_TYPE_ORIGIN(ovl_path_type(dentry));
 289 }
 290
 291 static int ovl_create_upper(struct dentry *dentry, struct inode *inode,
 292                             struct ovl_cattr *attr)
 293 {
 294         struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
 295         struct inode *udir = upperdir->d_inode;
 296         struct dentry *newdentry;
 297         int err;
 298
 299         if (!attr->hardlink && !IS_POSIXACL(udir))
 300                 attr->mode &= ~current_umask();
 301
 302         inode_lock_nested(udir, I_MUTEX_PARENT);
 303         newdentry = ovl_create_real(udir,
 304                                     lookup_one_len(dentry->d_name.name,
 305                                                    upperdir,
 306                                                    dentry->d_name.len),
 307                                     attr);
 308         err = PTR_ERR(newdentry);
 309         if (IS_ERR(newdentry))
 310                 goto out_unlock;
 311
 312         if (ovl_type_merge(dentry->d_parent) && d_is_dir(newdentry)) {
 313                 /* Setting opaque here is just an optimization, allow to fail */
 314                 ovl_set_opaque(dentry, newdentry);
 315         }
 316
 317         err = ovl_instantiate(dentry, inode, newdentry, !!attr->hardlink);
 318         if (err)
 319                 goto out_cleanup;
 320 out_unlock:
 321         inode_unlock(udir);
 322         return err;
 323
 324 out_cleanup:
 325         ovl_cleanup(udir, newdentry);
 326         dput(newdentry);
 327         goto out_unlock;
 328 }
 329
 330 static struct dentry *ovl_clear_empty(struct dentry *dentry,
 331                                       struct list_head *list)
 332 {
 333         struct dentry *workdir = ovl_workdir(dentry);
 334         struct inode *wdir = workdir->d_inode;
 335         struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
 336         struct inode *udir = upperdir->d_inode;
 337         struct path upperpath;
 338         struct dentry *upper;
 339         struct dentry *opaquedir;
 340         struct kstat stat;
 341         int err;
 342
 343         if (WARN_ON(!workdir))
 344                 return ERR_PTR(-EROFS);
 345
 346         err = ovl_lock_rename_workdir(workdir, upperdir);
 347         if (err)
 348                 goto out;
 349
 350         ovl_path_upper(dentry, &upperpath);
 351         err = vfs_getattr(&upperpath, &stat,
 352                           STATX_BASIC_STATS, AT_STATX_SYNC_AS_STAT);
 353         if (err)
 354                 goto out_unlock;
 355
 356         err = -ESTALE;
 357         if (!S_ISDIR(stat.mode))
 358                 goto out_unlock;
 359         upper = upperpath.dentry;
 360         if (upper->d_parent->d_inode != udir)
 361                 goto out_unlock;
 362
 363         opaquedir = ovl_create_temp(workdir, OVL_CATTR(stat.mode));
 364         err = PTR_ERR(opaquedir);
 365         if (IS_ERR(opaquedir))
 366                 goto out_unlock;
 367
 368         err = ovl_copy_xattr(upper, opaquedir);
 369         if (err)
 370                 goto out_cleanup;
 371
 372         err = ovl_set_opaque(dentry, opaquedir);
 373         if (err)
 374                 goto out_cleanup;
 375
 376         inode_lock(opaquedir->d_inode);
 377         err = ovl_set_attr(opaquedir, &stat);
 378         inode_unlock(opaquedir->d_inode);
 379         if (err)
 380                 goto out_cleanup;
 381
 382         err = ovl_do_rename(wdir, opaquedir, udir, upper, RENAME_EXCHANGE);
 383         if (err)
 384                 goto out_cleanup;
 385
 386         ovl_cleanup_whiteouts(upper, list);
 387         ovl_cleanup(wdir, upper);
 388         unlock_rename(workdir, upperdir);
 389
 390         /* dentry's upper doesn't match now, get rid of it */
 391         d_drop(dentry);
 392
 393         return opaquedir;
 394
 395 out_cleanup:
 396         ovl_cleanup(wdir, opaquedir);
 397         dput(opaquedir);
 398 out_unlock:
 399         unlock_rename(workdir, upperdir);
 400 out:
 401         return ERR_PTR(err);
 402 }
 403
 404 static int ovl_set_upper_acl(struct dentry *upperdentry, const char *name,
 405                              const struct posix_acl *acl)
 406 {
 407         void *buffer;
 408         size_t size;
 409         int err;
 410
 411         if (!IS_ENABLED(CONFIG_FS_POSIX_ACL) || !acl)
 412                 return 0;
 413
 414         size = posix_acl_xattr_size(acl->a_count);
 415         buffer = kmalloc(size, GFP_KERNEL);
 416         if (!buffer)
 417                 return -ENOMEM;
 418
 419         err = posix_acl_to_xattr(&init_user_ns, acl, buffer, size);
 420         if (err < 0)
 421                 goto out_free;
 422
 423         err = vfs_setxattr(upperdentry, name, buffer, size, XATTR_CREATE);
 424 out_free:
 425         kfree(buffer);
 426         return err;
 427 }
 428
 429 static int ovl_create_over_whiteout(struct dentry *dentry, struct inode *inode,
 430                                     struct ovl_cattr *cattr)
 431 {
 432         struct dentry *workdir = ovl_workdir(dentry);
 433         struct inode *wdir = workdir->d_inode;
 434         struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
 435         struct inode *udir = upperdir->d_inode;
 436         struct dentry *upper;
 437         struct dentry *newdentry;
 438         int err;
 439         struct posix_acl *acl, *default_acl;
 440         bool hardlink = !!cattr->hardlink;
 441
 442         if (WARN_ON(!workdir))
 443                 return -EROFS;
 444
 445         if (!hardlink) {
 446                 err = posix_acl_create(dentry->d_parent->d_inode,
 447                                        &cattr->mode, &default_acl, &acl);
 448                 if (err)
 449                         return err;
 450         }
 451
 452         err = ovl_lock_rename_workdir(workdir, upperdir);
 453         if (err)
 454                 goto out;
 455
 456         upper = lookup_one_len(dentry->d_name.name, upperdir,
 457                                dentry->d_name.len);
 458         err = PTR_ERR(upper);
 459         if (IS_ERR(upper))
 460                 goto out_unlock;
 461
 462         err = -ESTALE;
 463         if (d_is_negative(upper) || !IS_WHITEOUT(d_inode(upper)))
 464                 goto out_dput;
 465
 466         newdentry = ovl_create_temp(workdir, cattr);
 467         err = PTR_ERR(newdentry);
 468         if (IS_ERR(newdentry))
 469                 goto out_dput;
 470
 471         /*
 472          * mode could have been mutilated due to umask (e.g. sgid directory)
 473          */
 474         if (!hardlink &&
 475             !S_ISLNK(cattr->mode) &&
 476             newdentry->d_inode->i_mode != cattr->mode) {
 477                 struct iattr attr = {
 478                         .ia_valid = ATTR_MODE,
 479                         .ia_mode = cattr->mode,
 480                 };
 481                 inode_lock(newdentry->d_inode);
 482                 err = notify_change(newdentry, &attr, NULL);
 483                 inode_unlock(newdentry->d_inode);
 484                 if (err)
 485                         goto out_cleanup;
 486         }
 487         if (!hardlink) {
 488                 err = ovl_set_upper_acl(newdentry, XATTR_NAME_POSIX_ACL_ACCESS,
 489                                         acl);
 490                 if (err)
 491                         goto out_cleanup;
 492
 493                 err = ovl_set_upper_acl(newdentry, XATTR_NAME_POSIX_ACL_DEFAULT,
 494                                         default_acl);
 495                 if (err)
 496                         goto out_cleanup;
 497         }
 498
 499         if (!hardlink && S_ISDIR(cattr->mode)) {
 500                 err = ovl_set_opaque(dentry, newdentry);
 501                 if (err)
 502                         goto out_cleanup;
 503
 504                 err = ovl_do_rename(wdir, newdentry, udir, upper,
 505                                     RENAME_EXCHANGE);
 506                 if (err)
 507                         goto out_cleanup;
 508
 509                 ovl_cleanup(wdir, upper);
 510         } else {
 511                 err = ovl_do_rename(wdir, newdentry, udir, upper, 0);
 512                 if (err)
 513                         goto out_cleanup;
 514         }
 515         err = ovl_instantiate(dentry, inode, newdentry, hardlink);
 516         if (err)
 517                 goto out_cleanup;
 518 out_dput:
 519         dput(upper);
 520 out_unlock:
 521         unlock_rename(workdir, upperdir);
 522 out:
 523         if (!hardlink) {
 524                 posix_acl_release(acl);
 525                 posix_acl_release(default_acl);
 526         }
 527         return err;
 528
 529 out_cleanup:
 530         ovl_cleanup(wdir, newdentry);
 531         dput(newdentry);
 532         goto out_dput;
 533 }
 534
 535 static int ovl_create_or_link(struct dentry *dentry, struct inode *inode,
 536                               struct ovl_cattr *attr, bool origin)
 537 {
 538         int err;
 539         const struct cred *old_cred;
 540         struct cred *override_cred;
 541         struct dentry *parent = dentry->d_parent;
 542
 543         err = ovl_copy_up(parent);
 544         if (err)
 545                 return err;
 546
 547         old_cred = ovl_override_creds(dentry->d_sb);
 548
 549         /*
 550          * When linking a file with copy up origin into a new parent, mark the
 551          * new parent dir "impure".
 552          */
 553         if (origin) {
 554                 err = ovl_set_impure(parent, ovl_dentry_upper(parent));
 555                 if (err)
 556                         goto out_revert_creds;
 557         }
 558
 559         err = -ENOMEM;
 560         override_cred = prepare_creds();
 561         if (override_cred) {
 562                 override_cred->fsuid = inode->i_uid;
 563                 override_cred->fsgid = inode->i_gid;
 564                 if (!attr->hardlink) {
 565                         err = security_dentry_create_files_as(dentry,
 566                                         attr->mode, &dentry->d_name, old_cred,
 567                                         override_cred);
 568                         if (err) {
 569                                 put_cred(override_cred);
 570                                 goto out_revert_creds;
 571                         }
 572                 }
 573                 put_cred(override_creds(override_cred));
 574                 put_cred(override_cred);
 575
 576                 if (!ovl_dentry_is_whiteout(dentry))
 577                         err = ovl_create_upper(dentry, inode, attr);
 578                 else
 579                         err = ovl_create_over_whiteout(dentry, inode, attr);
 580         }
 581 out_revert_creds:
 582         revert_creds(old_cred);
 583         return err;
 584 }
 585
 586 static int ovl_create_object(struct dentry *dentry, int mode, dev_t rdev,
 587                              const char *link)
 588 {
 589         int err;
 590         struct inode *inode;
 591         struct ovl_cattr attr = {
 592                 .rdev = rdev,
 593                 .link = link,
 594         };
 595
 596         err = ovl_want_write(dentry);
 597         if (err)
 598                 goto out;
 599
 600         /* Preallocate inode to be used by ovl_get_inode() */
 601         err = -ENOMEM;
 602         inode = ovl_new_inode(dentry->d_sb, mode, rdev);
 603         if (!inode)
 604                 goto out_drop_write;
 605
 606         spin_lock(&inode->i_lock);
 607         inode->i_state |= I_CREATING;
 608         spin_unlock(&inode->i_lock);
 609
 610         inode_init_owner(inode, dentry->d_parent->d_inode, mode);
 611         attr.mode = inode->i_mode;
 612
 613         err = ovl_create_or_link(dentry, inode, &attr, false);
 614         /* Did we end up using the preallocated inode? */
 615         if (inode != d_inode(dentry))
 616                 iput(inode);
 617
 618 out_drop_write:
 619         ovl_drop_write(dentry);
 620 out:
 621         return err;
 622 }
 623
 624 static int ovl_create(struct inode *dir, struct dentry *dentry, umode_t mode,
 625                       bool excl)
 626 {
 627         return ovl_create_object(dentry, (mode & 07777) | S_IFREG, 0, NULL);
 628 }
 629
 630 static int ovl_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
 631 {
 632         return ovl_create_object(dentry, (mode & 07777) | S_IFDIR, 0, NULL);
 633 }
 634
 635 static int ovl_mknod(struct inode *dir, struct dentry *dentry, umode_t mode,
 636                      dev_t rdev)
 637 {
 638         /* Don't allow creation of "whiteout" on overlay */
 639         if (S_ISCHR(mode) && rdev == WHITEOUT_DEV)
 640                 return -EPERM;
 641
 642         return ovl_create_object(dentry, mode, rdev, NULL);
 643 }
 644
 645 static int ovl_symlink(struct inode *dir, struct dentry *dentry,
 646                        const char *link)
 647 {
 648         return ovl_create_object(dentry, S_IFLNK, 0, link);
 649 }
 650
 651 static int ovl_set_link_redirect(struct dentry *dentry)
 652 {
 653         const struct cred *old_cred;
 654         int err;
 655
 656         old_cred = ovl_override_creds(dentry->d_sb);
 657         err = ovl_set_redirect(dentry, false);
 658         revert_creds(old_cred);
 659
 660         return err;
 661 }
 662
 663 static int ovl_link(struct dentry *old, struct inode *newdir,
 664                     struct dentry *new)
 665 {
 666         int err;
 667         struct inode *inode;
 668
 669         err = ovl_want_write(old);
 670         if (err)
 671                 goto out;
 672
 673         err = ovl_copy_up(old);
 674         if (err)
 675                 goto out_drop_write;
 676
 677         err = ovl_copy_up(new->d_parent);
 678         if (err)
 679                 goto out_drop_write;
 680
 681         if (ovl_is_metacopy_dentry(old)) {
 682                 err = ovl_set_link_redirect(old);
 683                 if (err)
 684                         goto out_drop_write;
 685         }
 686
 687         err = ovl_nlink_start(old);
 688         if (err)
 689                 goto out_drop_write;
 690
 691         inode = d_inode(old);
 692         ihold(inode);
 693
 694         err = ovl_create_or_link(new, inode,
 695                         &(struct ovl_cattr) {.hardlink = ovl_dentry_upper(old)},
 696                         ovl_type_origin(old));
 697         if (err)
 698                 iput(inode);
 699
 700         ovl_nlink_end(old);
 701 out_drop_write:
 702         ovl_drop_write(old);
 703 out:
 704         return err;
 705 }
 706
 707 static bool ovl_matches_upper(struct dentry *dentry, struct dentry *upper)
 708 {
 709         return d_inode(ovl_dentry_upper(dentry)) == d_inode(upper);
 710 }
 711
 712 static int ovl_remove_and_whiteout(struct dentry *dentry,
 713                                    struct list_head *list)
 714 {
 715         struct dentry *workdir = ovl_workdir(dentry);
 716         struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
 717         struct dentry *upper;
 718         struct dentry *opaquedir = NULL;
 719         int err;
 720
 721         if (WARN_ON(!workdir))
 722                 return -EROFS;
 723
 724         if (!list_empty(list)) {
 725                 opaquedir = ovl_clear_empty(dentry, list);
 726                 err = PTR_ERR(opaquedir);
 727                 if (IS_ERR(opaquedir))
 728                         goto out;
 729         }
 730
 731         err = ovl_lock_rename_workdir(workdir, upperdir);
 732         if (err)
 733                 goto out_dput;
 734
 735         upper = lookup_one_len(dentry->d_name.name, upperdir,
 736                                dentry->d_name.len);
 737         err = PTR_ERR(upper);
 738         if (IS_ERR(upper))
 739                 goto out_unlock;
 740
 741         err = -ESTALE;
 742         if ((opaquedir && upper != opaquedir) ||
 743             (!opaquedir && ovl_dentry_upper(dentry) &&
 744              !ovl_matches_upper(dentry, upper))) {
 745                 goto out_dput_upper;
 746         }
 747
 748         err = ovl_cleanup_and_whiteout(workdir, d_inode(upperdir), upper);
 749         if (err)
 750                 goto out_d_drop;
 751
 752         ovl_dir_modified(dentry->d_parent, true);
 753 out_d_drop:
 754         d_drop(dentry);
 755 out_dput_upper:
 756         dput(upper);
 757 out_unlock:
 758         unlock_rename(workdir, upperdir);
 759 out_dput:
 760         dput(opaquedir);
 761 out:
 762         return err;
 763 }
 764
 765 static int ovl_remove_upper(struct dentry *dentry, bool is_dir,
 766                             struct list_head *list)
 767 {
 768         struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
 769         struct inode *dir = upperdir->d_inode;
 770         struct dentry *upper;
 771         struct dentry *opaquedir = NULL;
 772         int err;
 773
 774         if (!list_empty(list)) {
 775                 opaquedir = ovl_clear_empty(dentry, list);
 776                 err = PTR_ERR(opaquedir);
 777                 if (IS_ERR(opaquedir))
 778                         goto out;
 779         }
 780
 781         inode_lock_nested(dir, I_MUTEX_PARENT);
 782         upper = lookup_one_len(dentry->d_name.name, upperdir,
 783                                dentry->d_name.len);
 784         err = PTR_ERR(upper);
 785         if (IS_ERR(upper))
 786                 goto out_unlock;
 787
 788         err = -ESTALE;
 789         if ((opaquedir && upper != opaquedir) ||
 790             (!opaquedir && !ovl_matches_upper(dentry, upper)))
 791                 goto out_dput_upper;
 792
 793         if (is_dir)
 794                 err = vfs_rmdir(dir, upper);
 795         else
 796                 err = vfs_unlink(dir, upper, NULL);
 797         ovl_dir_modified(dentry->d_parent, ovl_type_origin(dentry));
 798
 799         /*
 800          * Keeping this dentry hashed would mean having to release
 801          * upperpath/lowerpath, which could only be done if we are the
 802          * sole user of this dentry.  Too tricky...  Just unhash for
 803          * now.
 804          */
 805         if (!err)
 806                 d_drop(dentry);
 807 out_dput_upper:
 808         dput(upper);
 809 out_unlock:
 810         inode_unlock(dir);
 811         dput(opaquedir);
 812 out:
 813         return err;
 814 }
 815
 816 static bool ovl_pure_upper(struct dentry *dentry)
 817 {
 818         return !ovl_dentry_lower(dentry) &&
 819                !ovl_test_flag(OVL_WHITEOUTS, d_inode(dentry));
 820 }
 821
 822 static int ovl_do_remove(struct dentry *dentry, bool is_dir)
 823 {
 824         int err;
 825         const struct cred *old_cred;
 826         struct dentry *upperdentry;
 827         bool lower_positive = ovl_lower_positive(dentry);
 828         LIST_HEAD(list);
 829
 830         /* No need to clean pure upper removed by vfs_rmdir() */
 831         if (is_dir && (lower_positive || !ovl_pure_upper(dentry))) {
 832                 err = ovl_check_empty_dir(dentry, &list);
 833                 if (err)
 834                         goto out;
 835         }
 836
 837         err = ovl_want_write(dentry);
 838         if (err)
 839                 goto out;
 840
 841         err = ovl_copy_up(dentry->d_parent);
 842         if (err)
 843                 goto out_drop_write;
 844
 845         err = ovl_nlink_start(dentry);
 846         if (err)
 847                 goto out_drop_write;
 848
 849         old_cred = ovl_override_creds(dentry->d_sb);
 850         if (!lower_positive)
 851                 err = ovl_remove_upper(dentry, is_dir, &list);
 852         else
 853                 err = ovl_remove_and_whiteout(dentry, &list);
 854         revert_creds(old_cred);
 855         if (!err) {
 856                 if (is_dir)
 857                         clear_nlink(dentry->d_inode);
 858                 else
 859                         drop_nlink(dentry->d_inode);
 860         }
 861         ovl_nlink_end(dentry);
 862
 863         /*
 864          * Copy ctime
 865          *
 866          * Note: we fail to update ctime if there was no copy-up, only a
 867          * whiteout
 868          */
 869         upperdentry = ovl_dentry_upper(dentry);
 870         if (upperdentry)
 871                 ovl_copyattr(d_inode(upperdentry), d_inode(dentry));
 872
 873 out_drop_write:
 874         ovl_drop_write(dentry);
 875 out:
 876         ovl_cache_free(&list);
 877         return err;
 878 }
 879
 880 static int ovl_unlink(struct inode *dir, struct dentry *dentry)
 881 {
 882         return ovl_do_remove(dentry, false);
 883 }
 884
 885 static int ovl_rmdir(struct inode *dir, struct dentry *dentry)
 886 {
 887         return ovl_do_remove(dentry, true);
 888 }
 889
 890 static bool ovl_type_merge_or_lower(struct dentry *dentry)
 891 {
 892         enum ovl_path_type type = ovl_path_type(dentry);
 893
 894         return OVL_TYPE_MERGE(type) || !OVL_TYPE_UPPER(type);
 895 }
 896
 897 static bool ovl_can_move(struct dentry *dentry)
 898 {
 899         return ovl_redirect_dir(dentry->d_sb) ||
 900                 !d_is_dir(dentry) || !ovl_type_merge_or_lower(dentry);
 901 }
 902
 903 static char *ovl_get_redirect(struct dentry *dentry, bool abs_redirect)
 904 {
 905         char *buf, *ret;
 906         struct dentry *d, *tmp;
 907         int buflen = ovl_redirect_max + 1;
 908
 909         if (!abs_redirect) {
 910                 ret = kstrndup(dentry->d_name.name, dentry->d_name.len,
 911                                GFP_KERNEL);
 912                 goto out;
 913         }
 914
 915         buf = ret = kmalloc(buflen, GFP_KERNEL);
 916         if (!buf)
 917                 goto out;
 918
 919         buflen--;
 920         buf[buflen] = '\0';
 921         for (d = dget(dentry); !IS_ROOT(d);) {
 922                 const char *name;
 923                 int thislen;
 924
 925                 spin_lock(&d->d_lock);
 926                 name = ovl_dentry_get_redirect(d);
 927                 if (name) {
 928                         thislen = strlen(name);
 929                 } else {
 930                         name = d->d_name.name;
 931                         thislen = d->d_name.len;
 932                 }
 933
 934                 /* If path is too long, fall back to userspace move */
 935                 if (thislen + (name[0] != '/') > buflen) {
 936                         ret = ERR_PTR(-EXDEV);
 937                         spin_unlock(&d->d_lock);
 938                         goto out_put;
 939                 }
 940
 941                 buflen -= thislen;
 942                 memcpy(&buf[buflen], name, thislen);
 943                 tmp = dget_dlock(d->d_parent);
 944                 spin_unlock(&d->d_lock);
 945
 946                 dput(d);
 947                 d = tmp;
 948
 949                 /* Absolute redirect: finished */
 950                 if (buf[buflen] == '/')
 951                         break;
 952                 buflen--;
 953                 buf[buflen] = '/';
 954         }
 955         ret = kstrdup(&buf[buflen], GFP_KERNEL);
 956 out_put:
 957         dput(d);
 958         kfree(buf);
 959 out:
 960         return ret ? ret : ERR_PTR(-ENOMEM);
 961 }
 962
 963 static bool ovl_need_absolute_redirect(struct dentry *dentry, bool samedir)
 964 {
 965         struct dentry *lowerdentry;
 966
 967         if (!samedir)
 968                 return true;
 969
 970         if (d_is_dir(dentry))
 971                 return false;
 972
 973         /*
 974          * For non-dir hardlinked files, we need absolute redirects
 975          * in general as two upper hardlinks could be in different
 976          * dirs. We could put a relative redirect now and convert
 977          * it to absolute redirect later. But when nlink > 1 and
 978          * indexing is on, that means relative redirect needs to be
 979          * converted to absolute during copy up of another lower
 980          * hardllink as well.
 981          *
 982          * So without optimizing too much, just check if lower is
 983          * a hard link or not. If lower is hard link, put absolute
 984          * redirect.
 985          */
 986         lowerdentry = ovl_dentry_lower(dentry);
 987         return (d_inode(lowerdentry)->i_nlink > 1);
 988 }
 989
 990 static int ovl_set_redirect(struct dentry *dentry, bool samedir)
 991 {
 992         int err;
 993         const char *redirect = ovl_dentry_get_redirect(dentry);
 994         bool absolute_redirect = ovl_need_absolute_redirect(dentry, samedir);
 995
 996         if (redirect && (!absolute_redirect || redirect[0] == '/'))
 997                 return 0;
 998
 999         redirect = ovl_get_redirect(dentry, absolute_redirect);
1000         if (IS_ERR(redirect))
1001                 return PTR_ERR(redirect);
1002
1003         err = ovl_check_setxattr(dentry, ovl_dentry_upper(dentry),
1004                                  OVL_XATTR_REDIRECT,
1005                                  redirect, strlen(redirect), -EXDEV);
1006         if (!err) {
1007                 spin_lock(&dentry->d_lock);
1008                 ovl_dentry_set_redirect(dentry, redirect);
1009                 spin_unlock(&dentry->d_lock);
1010         } else {
1011                 kfree(redirect);
1012                 pr_warn_ratelimited("overlayfs: failed to set redirect (%i)\n",
1013                                     err);
1014                 /* Fall back to userspace copy-up */
1015                 err = -EXDEV;
1016         }
1017         return err;
1018 }
1019
1020 static int ovl_rename(struct inode *olddir, struct dentry *old,
1021                       struct inode *newdir, struct dentry *new,
1022                       unsigned int flags)
1023 {
1024         int err;
1025         struct dentry *old_upperdir;
1026         struct dentry *new_upperdir;
1027         struct dentry *olddentry;
1028         struct dentry *newdentry;
1029         struct dentry *trap;
1030         bool old_opaque;
1031         bool new_opaque;
1032         bool cleanup_whiteout = false;
1033         bool update_nlink = false;
1034         bool overwrite = !(flags & RENAME_EXCHANGE);
1035         bool is_dir = d_is_dir(old);
1036         bool new_is_dir = d_is_dir(new);
1037         bool samedir = olddir == newdir;
1038         struct dentry *opaquedir = NULL;
1039         const struct cred *old_cred = NULL;
1040         LIST_HEAD(list);
1041
1042         err = -EINVAL;
1043         if (flags & ~(RENAME_EXCHANGE | RENAME_NOREPLACE))
1044                 goto out;
1045
1046         flags &= ~RENAME_NOREPLACE;
1047
1048         /* Don't copy up directory trees */
1049         err = -EXDEV;
1050         if (!ovl_can_move(old))
1051                 goto out;
1052         if (!overwrite && !ovl_can_move(new))
1053                 goto out;
1054
1055         if (overwrite && new_is_dir && !ovl_pure_upper(new)) {
1056                 err = ovl_check_empty_dir(new, &list);
1057                 if (err)
1058                         goto out;
1059         }
1060
1061         if (overwrite) {
1062                 if (ovl_lower_positive(old)) {
1063                         if (!ovl_dentry_is_whiteout(new)) {
1064                                 /* Whiteout source */
1065                                 flags |= RENAME_WHITEOUT;
1066                         } else {
1067                                 /* Switch whiteouts */
1068                                 flags |= RENAME_EXCHANGE;
1069                         }
1070                 } else if (is_dir && ovl_dentry_is_whiteout(new)) {
1071                         flags |= RENAME_EXCHANGE;
1072                         cleanup_whiteout = true;
1073                 }
1074         }
1075
1076         err = ovl_want_write(old);
1077         if (err)
1078                 goto out;
1079
1080         err = ovl_copy_up(old);
1081         if (err)
1082                 goto out_drop_write;
1083
1084         err = ovl_copy_up(new->d_parent);
1085         if (err)
1086                 goto out_drop_write;
1087         if (!overwrite) {
1088                 err = ovl_copy_up(new);
1089                 if (err)
1090                         goto out_drop_write;
1091         } else if (d_inode(new)) {
1092                 err = ovl_nlink_start(new);
1093                 if (err)
1094                         goto out_drop_write;
1095
1096                 update_nlink = true;
1097         }
1098
1099         old_cred = ovl_override_creds(old->d_sb);
1100
1101         if (!list_empty(&list)) {
1102                 opaquedir = ovl_clear_empty(new, &list);
1103                 err = PTR_ERR(opaquedir);
1104                 if (IS_ERR(opaquedir)) {
1105                         opaquedir = NULL;
1106                         goto out_revert_creds;
1107                 }
1108         }
1109
1110         old_upperdir = ovl_dentry_upper(old->d_parent);
1111         new_upperdir = ovl_dentry_upper(new->d_parent);
1112
1113         if (!samedir) {
1114                 /*
1115                  * When moving a merge dir or non-dir with copy up origin into
1116                  * a new parent, we are marking the new parent dir "impure".
1117                  * When ovl_iterate() iterates an "impure" upper dir, it will
1118                  * lookup the origin inodes of the entries to fill d_ino.
1119                  */
1120                 if (ovl_type_origin(old)) {
1121                         err = ovl_set_impure(new->d_parent, new_upperdir);
1122                         if (err)
1123                                 goto out_revert_creds;
1124                 }
1125                 if (!overwrite && ovl_type_origin(new)) {
1126                         err = ovl_set_impure(old->d_parent, old_upperdir);
1127                         if (err)
1128                                 goto out_revert_creds;
1129                 }
1130         }
1131
1132         trap = lock_rename(new_upperdir, old_upperdir);
1133
1134         olddentry = lookup_one_len(old->d_name.name, old_upperdir,
1135                                    old->d_name.len);
1136         err = PTR_ERR(olddentry);
1137         if (IS_ERR(olddentry))
1138                 goto out_unlock;
1139
1140         err = -ESTALE;
1141         if (!ovl_matches_upper(old, olddentry))
1142                 goto out_dput_old;
1143
1144         newdentry = lookup_one_len(new->d_name.name, new_upperdir,
1145                                    new->d_name.len);
1146         err = PTR_ERR(newdentry);
1147         if (IS_ERR(newdentry))
1148                 goto out_dput_old;
1149
1150         old_opaque = ovl_dentry_is_opaque(old);
1151         new_opaque = ovl_dentry_is_opaque(new);
1152
1153         err = -ESTALE;
1154         if (d_inode(new) && ovl_dentry_upper(new)) {
1155                 if (opaquedir) {
1156                         if (newdentry != opaquedir)
1157                                 goto out_dput;
1158                 } else {
1159                         if (!ovl_matches_upper(new, newdentry))
1160                                 goto out_dput;
1161                 }
1162         } else {
1163                 if (!d_is_negative(newdentry) &&
1164                     (!new_opaque || !ovl_is_whiteout(newdentry)))
1165                         goto out_dput;
1166         }
1167
1168         if (olddentry == trap)
1169                 goto out_dput;
1170         if (newdentry == trap)
1171                 goto out_dput;
1172
1173         if (olddentry->d_inode == newdentry->d_inode)
1174                 goto out_dput;
1175
1176         err = 0;
1177         if (ovl_type_merge_or_lower(old))
1178                 err = ovl_set_redirect(old, samedir);
1179         else if (is_dir && !old_opaque && ovl_type_merge(new->d_parent))
1180                 err = ovl_set_opaque_xerr(old, olddentry, -EXDEV);
1181         if (err)
1182                 goto out_dput;
1183
1184         if (!overwrite && ovl_type_merge_or_lower(new))
1185                 err = ovl_set_redirect(new, samedir);
1186         else if (!overwrite && new_is_dir && !new_opaque &&
1187                  ovl_type_merge(old->d_parent))
1188                 err = ovl_set_opaque_xerr(new, newdentry, -EXDEV);
1189         if (err)
1190                 goto out_dput;
1191
1192         err = ovl_do_rename(old_upperdir->d_inode, olddentry,
1193                             new_upperdir->d_inode, newdentry, flags);
1194         if (err)
1195                 goto out_dput;
1196
1197         if (cleanup_whiteout)
1198                 ovl_cleanup(old_upperdir->d_inode, newdentry);
1199
1200         if (overwrite && d_inode(new)) {
1201                 if (new_is_dir)
1202                         clear_nlink(d_inode(new));
1203                 else
1204                         drop_nlink(d_inode(new));
1205         }
1206
1207         ovl_dir_modified(old->d_parent, ovl_type_origin(old) ||
1208                          (!overwrite && ovl_type_origin(new)));
1209         ovl_dir_modified(new->d_parent, ovl_type_origin(old) ||
1210                          (d_inode(new) && ovl_type_origin(new)));
1211
1212         /* copy ctime: */
1213         ovl_copyattr(d_inode(olddentry), d_inode(old));
1214         if (d_inode(new) && ovl_dentry_upper(new))
1215                 ovl_copyattr(d_inode(newdentry), d_inode(new));
1216
1217 out_dput:
1218         dput(newdentry);
1219 out_dput_old:
1220         dput(olddentry);
1221 out_unlock:
1222         unlock_rename(new_upperdir, old_upperdir);
1223 out_revert_creds:
1224         revert_creds(old_cred);
1225         if (update_nlink)
1226                 ovl_nlink_end(new);
1227 out_drop_write:
1228         ovl_drop_write(old);
1229 out:
1230         dput(opaquedir);
1231         ovl_cache_free(&list);
1232         return err;
1233 }
1234
1235 const struct inode_operations ovl_dir_inode_operations = {
1236         .lookup         = ovl_lookup,
1237         .mkdir          = ovl_mkdir,
1238         .symlink        = ovl_symlink,
1239         .unlink         = ovl_unlink,
1240         .rmdir          = ovl_rmdir,
1241         .rename         = ovl_rename,
1242         .link           = ovl_link,
1243         .setattr        = ovl_setattr,
1244         .create         = ovl_create,
1245         .mknod          = ovl_mknod,
1246         .permission     = ovl_permission,
1247         .getattr        = ovl_getattr,
1248         .listxattr      = ovl_listxattr,
1249         .get_acl        = ovl_get_acl,
1250         .update_time    = ovl_update_time,
1251 };