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_xattr_size(acl->a_count);
 418         buffer = kmalloc(size, GFP_KERNEL);
 419         if (!buffer)
 420                 return -ENOMEM;
 421
 422         err = posix_acl_to_xattr(&init_user_ns, acl, buffer, size);
 423         if (err < 0)
 424                 goto out_free;
 425
 426         err = vfs_setxattr(upperdentry, name, buffer, size, XATTR_CREATE);
 427 out_free:
 428         kfree(buffer);
 429         return err;
 430 }
 431
 432 static int ovl_create_over_whiteout(struct dentry *dentry, struct inode *inode,
 433                                     struct ovl_cattr *cattr)
 434 {
 435         struct dentry *workdir = ovl_workdir(dentry);
 436         struct inode *wdir = workdir->d_inode;
 437         struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
 438         struct inode *udir = upperdir->d_inode;
 439         struct dentry *upper;
 440         struct dentry *newdentry;
 441         int err;
 442         struct posix_acl *acl, *default_acl;
 443         bool hardlink = !!cattr->hardlink;
 444
 445         if (WARN_ON(!workdir))
 446                 return -EROFS;
 447
 448         if (!hardlink) {
 449                 err = posix_acl_create(dentry->d_parent->d_inode,
 450                                        &cattr->mode, &default_acl, &acl);
 451                 if (err)
 452                         return err;
 453         }
 454
 455         err = ovl_lock_rename_workdir(workdir, upperdir);
 456         if (err)
 457                 goto out;
 458
 459         upper = lookup_one_len(dentry->d_name.name, upperdir,
 460                                dentry->d_name.len);
 461         err = PTR_ERR(upper);
 462         if (IS_ERR(upper))
 463                 goto out_unlock;
 464
 465         newdentry = ovl_create_temp(workdir, cattr);
 466         err = PTR_ERR(newdentry);
 467         if (IS_ERR(newdentry))
 468                 goto out_dput;
 469
 470         /*
 471          * mode could have been mutilated due to umask (e.g. sgid directory)
 472          */
 473         if (!hardlink &&
 474             !S_ISLNK(cattr->mode) &&
 475             newdentry->d_inode->i_mode != cattr->mode) {
 476                 struct iattr attr = {
 477                         .ia_valid = ATTR_MODE,
 478                         .ia_mode = cattr->mode,
 479                 };
 480                 inode_lock(newdentry->d_inode);
 481                 err = notify_change(newdentry, &attr, NULL);
 482                 inode_unlock(newdentry->d_inode);
 483                 if (err)
 484                         goto out_cleanup;
 485         }
 486         if (!hardlink) {
 487                 err = ovl_set_upper_acl(newdentry, XATTR_NAME_POSIX_ACL_ACCESS,
 488                                         acl);
 489                 if (err)
 490                         goto out_cleanup;
 491
 492                 err = ovl_set_upper_acl(newdentry, XATTR_NAME_POSIX_ACL_DEFAULT,
 493                                         default_acl);
 494                 if (err)
 495                         goto out_cleanup;
 496         }
 497
 498         if (!hardlink && S_ISDIR(cattr->mode)) {
 499                 err = ovl_set_opaque(dentry, newdentry);
 500                 if (err)
 501                         goto out_cleanup;
 502
 503                 err = ovl_do_rename(wdir, newdentry, udir, upper,
 504                                     RENAME_EXCHANGE);
 505                 if (err)
 506                         goto out_cleanup;
 507
 508                 ovl_cleanup(wdir, upper);
 509         } else {
 510                 err = ovl_do_rename(wdir, newdentry, udir, upper, 0);
 511                 if (err)
 512                         goto out_cleanup;
 513         }
 514         err = ovl_instantiate(dentry, inode, newdentry, hardlink);
 515         if (err)
 516                 goto out_cleanup;
 517 out_dput:
 518         dput(upper);
 519 out_unlock:
 520         unlock_rename(workdir, upperdir);
 521 out:
 522         if (!hardlink) {
 523                 posix_acl_release(acl);
 524                 posix_acl_release(default_acl);
 525         }
 526         return err;
 527
 528 out_cleanup:
 529         ovl_cleanup(wdir, newdentry);
 530         dput(newdentry);
 531         goto out_dput;
 532 }
 533
 534 static int ovl_create_or_link(struct dentry *dentry, struct inode *inode,
 535                               struct ovl_cattr *attr, bool origin)
 536 {
 537         int err;
 538         const struct cred *old_cred;
 539         struct cred *override_cred;
 540         struct dentry *parent = dentry->d_parent;
 541
 542         err = ovl_copy_up(parent);
 543         if (err)
 544                 return err;
 545
 546         old_cred = ovl_override_creds(dentry->d_sb);
 547
 548         /*
 549          * When linking a file with copy up origin into a new parent, mark the
 550          * new parent dir "impure".
 551          */
 552         if (origin) {
 553                 err = ovl_set_impure(parent, ovl_dentry_upper(parent));
 554                 if (err)
 555                         goto out_revert_creds;
 556         }
 557
 558         err = -ENOMEM;
 559         override_cred = prepare_creds();
 560         if (override_cred) {
 561                 override_cred->fsuid = inode->i_uid;
 562                 override_cred->fsgid = inode->i_gid;
 563                 if (!attr->hardlink) {
 564                         err = security_dentry_create_files_as(dentry,
 565                                         attr->mode, &dentry->d_name, old_cred,
 566                                         override_cred);
 567                         if (err) {
 568                                 put_cred(override_cred);
 569                                 goto out_revert_creds;
 570                         }
 571                 }
 572                 put_cred(override_creds(override_cred));
 573                 put_cred(override_cred);
 574
 575                 if (!ovl_dentry_is_whiteout(dentry))
 576                         err = ovl_create_upper(dentry, inode, attr);
 577                 else
 578                         err = ovl_create_over_whiteout(dentry, inode, attr);
 579         }
 580 out_revert_creds:
 581         revert_creds(old_cred);
 582         return err;
 583 }
 584
 585 static int ovl_create_object(struct dentry *dentry, int mode, dev_t rdev,
 586                              const char *link)
 587 {
 588         int err;
 589         struct inode *inode;
 590         struct ovl_cattr attr = {
 591                 .rdev = rdev,
 592                 .link = link,
 593         };
 594
 595         err = ovl_want_write(dentry);
 596         if (err)
 597                 goto out;
 598
 599         /* Preallocate inode to be used by ovl_get_inode() */
 600         err = -ENOMEM;
 601         inode = ovl_new_inode(dentry->d_sb, mode, rdev);
 602         if (!inode)
 603                 goto out_drop_write;
 604
 605         spin_lock(&inode->i_lock);
 606         inode->i_state |= I_CREATING;
 607         spin_unlock(&inode->i_lock);
 608
 609         inode_init_owner(inode, dentry->d_parent->d_inode, mode);
 610         attr.mode = inode->i_mode;
 611
 612         err = ovl_create_or_link(dentry, inode, &attr, false);
 613         /* Did we end up using the preallocated inode? */
 614         if (inode != d_inode(dentry))
 615                 iput(inode);
 616
 617 out_drop_write:
 618         ovl_drop_write(dentry);
 619 out:
 620         return err;
 621 }
 622
 623 static int ovl_create(struct inode *dir, struct dentry *dentry, umode_t mode,
 624                       bool excl)
 625 {
 626         return ovl_create_object(dentry, (mode & 07777) | S_IFREG, 0, NULL);
 627 }
 628
 629 static int ovl_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
 630 {
 631         return ovl_create_object(dentry, (mode & 07777) | S_IFDIR, 0, NULL);
 632 }
 633
 634 static int ovl_mknod(struct inode *dir, struct dentry *dentry, umode_t mode,
 635                      dev_t rdev)
 636 {
 637         /* Don't allow creation of "whiteout" on overlay */
 638         if (S_ISCHR(mode) && rdev == WHITEOUT_DEV)
 639                 return -EPERM;
 640
 641         return ovl_create_object(dentry, mode, rdev, NULL);
 642 }
 643
 644 static int ovl_symlink(struct inode *dir, struct dentry *dentry,
 645                        const char *link)
 646 {
 647         return ovl_create_object(dentry, S_IFLNK, 0, link);
 648 }
 649
 650 static int ovl_link(struct dentry *old, struct inode *newdir,
 651                     struct dentry *new)
 652 {
 653         int err;
 654         struct inode *inode;
 655
 656         err = ovl_want_write(old);
 657         if (err)
 658                 goto out;
 659
 660         err = ovl_copy_up(old);
 661         if (err)
 662                 goto out_drop_write;
 663
 664         err = ovl_copy_up(new->d_parent);
 665         if (err)
 666                 goto out_drop_write;
 667
 668         if (ovl_is_metacopy_dentry(old)) {
 669                 err = ovl_set_redirect(old, false);
 670                 if (err)
 671                         goto out_drop_write;
 672         }
 673
 674         err = ovl_nlink_start(old);
 675         if (err)
 676                 goto out_drop_write;
 677
 678         inode = d_inode(old);
 679         ihold(inode);
 680
 681         err = ovl_create_or_link(new, inode,
 682                         &(struct ovl_cattr) {.hardlink = ovl_dentry_upper(old)},
 683                         ovl_type_origin(old));
 684         if (err)
 685                 iput(inode);
 686
 687         ovl_nlink_end(old);
 688 out_drop_write:
 689         ovl_drop_write(old);
 690 out:
 691         return err;
 692 }
 693
 694 static bool ovl_matches_upper(struct dentry *dentry, struct dentry *upper)
 695 {
 696         return d_inode(ovl_dentry_upper(dentry)) == d_inode(upper);
 697 }
 698
 699 static int ovl_remove_and_whiteout(struct dentry *dentry,
 700                                    struct list_head *list)
 701 {
 702         struct dentry *workdir = ovl_workdir(dentry);
 703         struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
 704         struct dentry *upper;
 705         struct dentry *opaquedir = NULL;
 706         int err;
 707
 708         if (WARN_ON(!workdir))
 709                 return -EROFS;
 710
 711         if (!list_empty(list)) {
 712                 opaquedir = ovl_clear_empty(dentry, list);
 713                 err = PTR_ERR(opaquedir);
 714                 if (IS_ERR(opaquedir))
 715                         goto out;
 716         }
 717
 718         err = ovl_lock_rename_workdir(workdir, upperdir);
 719         if (err)
 720                 goto out_dput;
 721
 722         upper = lookup_one_len(dentry->d_name.name, upperdir,
 723                                dentry->d_name.len);
 724         err = PTR_ERR(upper);
 725         if (IS_ERR(upper))
 726                 goto out_unlock;
 727
 728         err = -ESTALE;
 729         if ((opaquedir && upper != opaquedir) ||
 730             (!opaquedir && ovl_dentry_upper(dentry) &&
 731              !ovl_matches_upper(dentry, upper))) {
 732                 goto out_dput_upper;
 733         }
 734
 735         err = ovl_cleanup_and_whiteout(workdir, d_inode(upperdir), upper);
 736         if (err)
 737                 goto out_d_drop;
 738
 739         ovl_dir_modified(dentry->d_parent, true);
 740 out_d_drop:
 741         d_drop(dentry);
 742 out_dput_upper:
 743         dput(upper);
 744 out_unlock:
 745         unlock_rename(workdir, upperdir);
 746 out_dput:
 747         dput(opaquedir);
 748 out:
 749         return err;
 750 }
 751
 752 static int ovl_remove_upper(struct dentry *dentry, bool is_dir,
 753                             struct list_head *list)
 754 {
 755         struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
 756         struct inode *dir = upperdir->d_inode;
 757         struct dentry *upper;
 758         struct dentry *opaquedir = NULL;
 759         int err;
 760
 761         if (!list_empty(list)) {
 762                 opaquedir = ovl_clear_empty(dentry, list);
 763                 err = PTR_ERR(opaquedir);
 764                 if (IS_ERR(opaquedir))
 765                         goto out;
 766         }
 767
 768         inode_lock_nested(dir, I_MUTEX_PARENT);
 769         upper = lookup_one_len(dentry->d_name.name, upperdir,
 770                                dentry->d_name.len);
 771         err = PTR_ERR(upper);
 772         if (IS_ERR(upper))
 773                 goto out_unlock;
 774
 775         err = -ESTALE;
 776         if ((opaquedir && upper != opaquedir) ||
 777             (!opaquedir && !ovl_matches_upper(dentry, upper)))
 778                 goto out_dput_upper;
 779
 780         if (is_dir)
 781                 err = vfs_rmdir(dir, upper);
 782         else
 783                 err = vfs_unlink(dir, upper, NULL);
 784         ovl_dir_modified(dentry->d_parent, ovl_type_origin(dentry));
 785
 786         /*
 787          * Keeping this dentry hashed would mean having to release
 788          * upperpath/lowerpath, which could only be done if we are the
 789          * sole user of this dentry.  Too tricky...  Just unhash for
 790          * now.
 791          */
 792         if (!err)
 793                 d_drop(dentry);
 794 out_dput_upper:
 795         dput(upper);
 796 out_unlock:
 797         inode_unlock(dir);
 798         dput(opaquedir);
 799 out:
 800         return err;
 801 }
 802
 803 static bool ovl_pure_upper(struct dentry *dentry)
 804 {
 805         return !ovl_dentry_lower(dentry) &&
 806                !ovl_test_flag(OVL_WHITEOUTS, d_inode(dentry));
 807 }
 808
 809 static int ovl_do_remove(struct dentry *dentry, bool is_dir)
 810 {
 811         int err;
 812         const struct cred *old_cred;
 813         struct dentry *upperdentry;
 814         bool lower_positive = ovl_lower_positive(dentry);
 815         LIST_HEAD(list);
 816
 817         /* No need to clean pure upper removed by vfs_rmdir() */
 818         if (is_dir && (lower_positive || !ovl_pure_upper(dentry))) {
 819                 err = ovl_check_empty_dir(dentry, &list);
 820                 if (err)
 821                         goto out;
 822         }
 823
 824         err = ovl_want_write(dentry);
 825         if (err)
 826                 goto out;
 827
 828         err = ovl_copy_up(dentry->d_parent);
 829         if (err)
 830                 goto out_drop_write;
 831
 832         err = ovl_nlink_start(dentry);
 833         if (err)
 834                 goto out_drop_write;
 835
 836         old_cred = ovl_override_creds(dentry->d_sb);
 837         if (!lower_positive)
 838                 err = ovl_remove_upper(dentry, is_dir, &list);
 839         else
 840                 err = ovl_remove_and_whiteout(dentry, &list);
 841         revert_creds(old_cred);
 842         if (!err) {
 843                 if (is_dir)
 844                         clear_nlink(dentry->d_inode);
 845                 else
 846                         drop_nlink(dentry->d_inode);
 847         }
 848         ovl_nlink_end(dentry);
 849
 850         /*
 851          * Copy ctime
 852          *
 853          * Note: we fail to update ctime if there was no copy-up, only a
 854          * whiteout
 855          */
 856         upperdentry = ovl_dentry_upper(dentry);
 857         if (upperdentry)
 858                 ovl_copyattr(d_inode(upperdentry), d_inode(dentry));
 859
 860 out_drop_write:
 861         ovl_drop_write(dentry);
 862 out:
 863         ovl_cache_free(&list);
 864         return err;
 865 }
 866
 867 static int ovl_unlink(struct inode *dir, struct dentry *dentry)
 868 {
 869         return ovl_do_remove(dentry, false);
 870 }
 871
 872 static int ovl_rmdir(struct inode *dir, struct dentry *dentry)
 873 {
 874         return ovl_do_remove(dentry, true);
 875 }
 876
 877 static bool ovl_type_merge_or_lower(struct dentry *dentry)
 878 {
 879         enum ovl_path_type type = ovl_path_type(dentry);
 880
 881         return OVL_TYPE_MERGE(type) || !OVL_TYPE_UPPER(type);
 882 }
 883
 884 static bool ovl_can_move(struct dentry *dentry)
 885 {
 886         return ovl_redirect_dir(dentry->d_sb) ||
 887                 !d_is_dir(dentry) || !ovl_type_merge_or_lower(dentry);
 888 }
 889
 890 static char *ovl_get_redirect(struct dentry *dentry, bool abs_redirect)
 891 {
 892         char *buf, *ret;
 893         struct dentry *d, *tmp;
 894         int buflen = ovl_redirect_max + 1;
 895
 896         if (!abs_redirect) {
 897                 ret = kstrndup(dentry->d_name.name, dentry->d_name.len,
 898                                GFP_KERNEL);
 899                 goto out;
 900         }
 901
 902         buf = ret = kmalloc(buflen, GFP_KERNEL);
 903         if (!buf)
 904                 goto out;
 905
 906         buflen--;
 907         buf[buflen] = '\0';
 908         for (d = dget(dentry); !IS_ROOT(d);) {
 909                 const char *name;
 910                 int thislen;
 911
 912                 spin_lock(&d->d_lock);
 913                 name = ovl_dentry_get_redirect(d);
 914                 if (name) {
 915                         thislen = strlen(name);
 916                 } else {
 917                         name = d->d_name.name;
 918                         thislen = d->d_name.len;
 919                 }
 920
 921                 /* If path is too long, fall back to userspace move */
 922                 if (thislen + (name[0] != '/') > buflen) {
 923                         ret = ERR_PTR(-EXDEV);
 924                         spin_unlock(&d->d_lock);
 925                         goto out_put;
 926                 }
 927
 928                 buflen -= thislen;
 929                 memcpy(&buf[buflen], name, thislen);
 930                 tmp = dget_dlock(d->d_parent);
 931                 spin_unlock(&d->d_lock);
 932
 933                 dput(d);
 934                 d = tmp;
 935
 936                 /* Absolute redirect: finished */
 937                 if (buf[buflen] == '/')
 938                         break;
 939                 buflen--;
 940                 buf[buflen] = '/';
 941         }
 942         ret = kstrdup(&buf[buflen], GFP_KERNEL);
 943 out_put:
 944         dput(d);
 945         kfree(buf);
 946 out:
 947         return ret ? ret : ERR_PTR(-ENOMEM);
 948 }
 949
 950 static bool ovl_need_absolute_redirect(struct dentry *dentry, bool samedir)
 951 {
 952         struct dentry *lowerdentry;
 953
 954         if (!samedir)
 955                 return true;
 956
 957         if (d_is_dir(dentry))
 958                 return false;
 959
 960         /*
 961          * For non-dir hardlinked files, we need absolute redirects
 962          * in general as two upper hardlinks could be in different
 963          * dirs. We could put a relative redirect now and convert
 964          * it to absolute redirect later. But when nlink > 1 and
 965          * indexing is on, that means relative redirect needs to be
 966          * converted to absolute during copy up of another lower
 967          * hardllink as well.
 968          *
 969          * So without optimizing too much, just check if lower is
 970          * a hard link or not. If lower is hard link, put absolute
 971          * redirect.
 972          */
 973         lowerdentry = ovl_dentry_lower(dentry);
 974         return (d_inode(lowerdentry)->i_nlink > 1);
 975 }
 976
 977 static int ovl_set_redirect(struct dentry *dentry, bool samedir)
 978 {
 979         int err;
 980         const char *redirect = ovl_dentry_get_redirect(dentry);
 981         bool absolute_redirect = ovl_need_absolute_redirect(dentry, samedir);
 982
 983         if (redirect && (!absolute_redirect || redirect[0] == '/'))
 984                 return 0;
 985
 986         redirect = ovl_get_redirect(dentry, absolute_redirect);
 987         if (IS_ERR(redirect))
 988                 return PTR_ERR(redirect);
 989
 990         err = ovl_check_setxattr(dentry, ovl_dentry_upper(dentry),
 991                                  OVL_XATTR_REDIRECT,
 992                                  redirect, strlen(redirect), -EXDEV);
 993         if (!err) {
 994                 spin_lock(&dentry->d_lock);
 995                 ovl_dentry_set_redirect(dentry, redirect);
 996                 spin_unlock(&dentry->d_lock);
 997         } else {
 998                 kfree(redirect);
 999                 pr_warn_ratelimited("overlayfs: failed to set redirect (%i)\n",
1000                                     err);
1001                 /* Fall back to userspace copy-up */
1002                 err = -EXDEV;
1003         }
1004         return err;
1005 }
1006
1007 static int ovl_rename(struct inode *olddir, struct dentry *old,
1008                       struct inode *newdir, struct dentry *new,
1009                       unsigned int flags)
1010 {
1011         int err;
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 update_nlink = false;
1021         bool overwrite = !(flags & RENAME_EXCHANGE);
1022         bool is_dir = d_is_dir(old);
1023         bool new_is_dir = d_is_dir(new);
1024         bool samedir = olddir == newdir;
1025         struct dentry *opaquedir = NULL;
1026         const struct cred *old_cred = NULL;
1027         LIST_HEAD(list);
1028
1029         err = -EINVAL;
1030         if (flags & ~(RENAME_EXCHANGE | RENAME_NOREPLACE))
1031                 goto out;
1032
1033         flags &= ~RENAME_NOREPLACE;
1034
1035         /* Don't copy up directory trees */
1036         err = -EXDEV;
1037         if (!ovl_can_move(old))
1038                 goto out;
1039         if (!overwrite && !ovl_can_move(new))
1040                 goto out;
1041
1042         if (overwrite && new_is_dir && !ovl_pure_upper(new)) {
1043                 err = ovl_check_empty_dir(new, &list);
1044                 if (err)
1045                         goto out;
1046         }
1047
1048         if (overwrite) {
1049                 if (ovl_lower_positive(old)) {
1050                         if (!ovl_dentry_is_whiteout(new)) {
1051                                 /* Whiteout source */
1052                                 flags |= RENAME_WHITEOUT;
1053                         } else {
1054                                 /* Switch whiteouts */
1055                                 flags |= RENAME_EXCHANGE;
1056                         }
1057                 } else if (is_dir && ovl_dentry_is_whiteout(new)) {
1058                         flags |= RENAME_EXCHANGE;
1059                         cleanup_whiteout = true;
1060                 }
1061         }
1062
1063         err = ovl_want_write(old);
1064         if (err)
1065                 goto out;
1066
1067         err = ovl_copy_up(old);
1068         if (err)
1069                 goto out_drop_write;
1070
1071         err = ovl_copy_up(new->d_parent);
1072         if (err)
1073                 goto out_drop_write;
1074         if (!overwrite) {
1075                 err = ovl_copy_up(new);
1076                 if (err)
1077                         goto out_drop_write;
1078         } else if (d_inode(new)) {
1079                 err = ovl_nlink_start(new);
1080                 if (err)
1081                         goto out_drop_write;
1082
1083                 update_nlink = true;
1084         }
1085
1086         old_cred = ovl_override_creds(old->d_sb);
1087
1088         if (!list_empty(&list)) {
1089                 opaquedir = ovl_clear_empty(new, &list);
1090                 err = PTR_ERR(opaquedir);
1091                 if (IS_ERR(opaquedir)) {
1092                         opaquedir = NULL;
1093                         goto out_revert_creds;
1094                 }
1095         }
1096
1097         old_upperdir = ovl_dentry_upper(old->d_parent);
1098         new_upperdir = ovl_dentry_upper(new->d_parent);
1099
1100         if (!samedir) {
1101                 /*
1102                  * When moving a merge dir or non-dir with copy up origin into
1103                  * a new parent, we are marking the new parent dir "impure".
1104                  * When ovl_iterate() iterates an "impure" upper dir, it will
1105                  * lookup the origin inodes of the entries to fill d_ino.
1106                  */
1107                 if (ovl_type_origin(old)) {
1108                         err = ovl_set_impure(new->d_parent, new_upperdir);
1109                         if (err)
1110                                 goto out_revert_creds;
1111                 }
1112                 if (!overwrite && ovl_type_origin(new)) {
1113                         err = ovl_set_impure(old->d_parent, old_upperdir);
1114                         if (err)
1115                                 goto out_revert_creds;
1116                 }
1117         }
1118
1119         trap = lock_rename(new_upperdir, old_upperdir);
1120
1121         olddentry = lookup_one_len(old->d_name.name, old_upperdir,
1122                                    old->d_name.len);
1123         err = PTR_ERR(olddentry);
1124         if (IS_ERR(olddentry))
1125                 goto out_unlock;
1126
1127         err = -ESTALE;
1128         if (!ovl_matches_upper(old, olddentry))
1129                 goto out_dput_old;
1130
1131         newdentry = lookup_one_len(new->d_name.name, new_upperdir,
1132                                    new->d_name.len);
1133         err = PTR_ERR(newdentry);
1134         if (IS_ERR(newdentry))
1135                 goto out_dput_old;
1136
1137         old_opaque = ovl_dentry_is_opaque(old);
1138         new_opaque = ovl_dentry_is_opaque(new);
1139
1140         err = -ESTALE;
1141         if (d_inode(new) && ovl_dentry_upper(new)) {
1142                 if (opaquedir) {
1143                         if (newdentry != opaquedir)
1144                                 goto out_dput;
1145                 } else {
1146                         if (!ovl_matches_upper(new, newdentry))
1147                                 goto out_dput;
1148                 }
1149         } else {
1150                 if (!d_is_negative(newdentry) &&
1151                     (!new_opaque || !ovl_is_whiteout(newdentry)))
1152                         goto out_dput;
1153         }
1154
1155         if (olddentry == trap)
1156                 goto out_dput;
1157         if (newdentry == trap)
1158                 goto out_dput;
1159
1160         if (WARN_ON(olddentry->d_inode == newdentry->d_inode))
1161                 goto out_dput;
1162
1163         err = 0;
1164         if (ovl_type_merge_or_lower(old))
1165                 err = ovl_set_redirect(old, samedir);
1166         else if (is_dir && !old_opaque && ovl_type_merge(new->d_parent))
1167                 err = ovl_set_opaque_xerr(old, olddentry, -EXDEV);
1168         if (err)
1169                 goto out_dput;
1170
1171         if (!overwrite && ovl_type_merge_or_lower(new))
1172                 err = ovl_set_redirect(new, samedir);
1173         else if (!overwrite && new_is_dir && !new_opaque &&
1174                  ovl_type_merge(old->d_parent))
1175                 err = ovl_set_opaque_xerr(new, newdentry, -EXDEV);
1176         if (err)
1177                 goto out_dput;
1178
1179         err = ovl_do_rename(old_upperdir->d_inode, olddentry,
1180                             new_upperdir->d_inode, newdentry, flags);
1181         if (err)
1182                 goto out_dput;
1183
1184         if (cleanup_whiteout)
1185                 ovl_cleanup(old_upperdir->d_inode, newdentry);
1186
1187         if (overwrite && d_inode(new)) {
1188                 if (new_is_dir)
1189                         clear_nlink(d_inode(new));
1190                 else
1191                         drop_nlink(d_inode(new));
1192         }
1193
1194         ovl_dir_modified(old->d_parent, ovl_type_origin(old) ||
1195                          (!overwrite && ovl_type_origin(new)));
1196         ovl_dir_modified(new->d_parent, ovl_type_origin(old) ||
1197                          (d_inode(new) && ovl_type_origin(new)));
1198
1199         /* copy ctime: */
1200         ovl_copyattr(d_inode(olddentry), d_inode(old));
1201         if (d_inode(new) && ovl_dentry_upper(new))
1202                 ovl_copyattr(d_inode(newdentry), d_inode(new));
1203
1204 out_dput:
1205         dput(newdentry);
1206 out_dput_old:
1207         dput(olddentry);
1208 out_unlock:
1209         unlock_rename(new_upperdir, old_upperdir);
1210 out_revert_creds:
1211         revert_creds(old_cred);
1212         if (update_nlink)
1213                 ovl_nlink_end(new);
1214 out_drop_write:
1215         ovl_drop_write(old);
1216 out:
1217         dput(opaquedir);
1218         ovl_cache_free(&list);
1219         return err;
1220 }
1221
1222 const struct inode_operations ovl_dir_inode_operations = {
1223         .lookup         = ovl_lookup,
1224         .mkdir          = ovl_mkdir,
1225         .symlink        = ovl_symlink,
1226         .unlink         = ovl_unlink,
1227         .rmdir          = ovl_rmdir,
1228         .rename         = ovl_rename,
1229         .link           = ovl_link,
1230         .setattr        = ovl_setattr,
1231         .create         = ovl_create,
1232         .mknod          = ovl_mknod,
1233         .permission     = ovl_permission,
1234         .getattr        = ovl_getattr,
1235         .listxattr      = ovl_listxattr,
1236         .get_acl        = ovl_get_acl,
1237         .update_time    = ovl_update_time,
1238 };