fs/overlayfs/dir.c

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