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         err = -ESTALE;
 466         if (d_is_negative(upper) || !IS_WHITEOUT(d_inode(upper)))
 467                 goto out_dput;
 468
 469         newdentry = ovl_create_temp(workdir, cattr);
 470         err = PTR_ERR(newdentry);
 471         if (IS_ERR(newdentry))
 472                 goto out_dput;
 473
 474         /*
 475          * mode could have been mutilated due to umask (e.g. sgid directory)
 476          */
 477         if (!hardlink &&
 478             !S_ISLNK(cattr->mode) &&
 479             newdentry->d_inode->i_mode != cattr->mode) {
 480                 struct iattr attr = {
 481                         .ia_valid = ATTR_MODE,
 482                         .ia_mode = cattr->mode,
 483                 };
 484                 inode_lock(newdentry->d_inode);
 485                 err = notify_change(newdentry, &attr, NULL);
 486                 inode_unlock(newdentry->d_inode);
 487                 if (err)
 488                         goto out_cleanup;
 489         }
 490         if (!hardlink) {
 491                 err = ovl_set_upper_acl(newdentry, XATTR_NAME_POSIX_ACL_ACCESS,
 492                                         acl);
 493                 if (err)
 494                         goto out_cleanup;
 495
 496                 err = ovl_set_upper_acl(newdentry, XATTR_NAME_POSIX_ACL_DEFAULT,
 497                                         default_acl);
 498                 if (err)
 499                         goto out_cleanup;
 500         }
 501
 502         if (!hardlink && S_ISDIR(cattr->mode)) {
 503                 err = ovl_set_opaque(dentry, newdentry);
 504                 if (err)
 505                         goto out_cleanup;
 506
 507                 err = ovl_do_rename(wdir, newdentry, udir, upper,
 508                                     RENAME_EXCHANGE);
 509                 if (err)
 510                         goto out_cleanup;
 511
 512                 ovl_cleanup(wdir, upper);
 513         } else {
 514                 err = ovl_do_rename(wdir, newdentry, udir, upper, 0);
 515                 if (err)
 516                         goto out_cleanup;
 517         }
 518         err = ovl_instantiate(dentry, inode, newdentry, hardlink);
 519         if (err)
 520                 goto out_cleanup;
 521 out_dput:
 522         dput(upper);
 523 out_unlock:
 524         unlock_rename(workdir, upperdir);
 525 out:
 526         if (!hardlink) {
 527                 posix_acl_release(acl);
 528                 posix_acl_release(default_acl);
 529         }
 530         return err;
 531
 532 out_cleanup:
 533         ovl_cleanup(wdir, newdentry);
 534         dput(newdentry);
 535         goto out_dput;
 536 }
 537
 538 static int ovl_create_or_link(struct dentry *dentry, struct inode *inode,
 539                               struct ovl_cattr *attr, bool origin)
 540 {
 541         int err;
 542         const struct cred *old_cred;
 543         struct cred *override_cred;
 544         struct dentry *parent = dentry->d_parent;
 545
 546         err = ovl_copy_up(parent);
 547         if (err)
 548                 return err;
 549
 550         old_cred = ovl_override_creds(dentry->d_sb);
 551
 552         /*
 553          * When linking a file with copy up origin into a new parent, mark the
 554          * new parent dir "impure".
 555          */
 556         if (origin) {
 557                 err = ovl_set_impure(parent, ovl_dentry_upper(parent));
 558                 if (err)
 559                         goto out_revert_creds;
 560         }
 561
 562         err = -ENOMEM;
 563         override_cred = prepare_creds();
 564         if (override_cred) {
 565                 override_cred->fsuid = inode->i_uid;
 566                 override_cred->fsgid = inode->i_gid;
 567                 if (!attr->hardlink) {
 568                         err = security_dentry_create_files_as(dentry,
 569                                         attr->mode, &dentry->d_name, old_cred,
 570                                         override_cred);
 571                         if (err) {
 572                                 put_cred(override_cred);
 573                                 goto out_revert_creds;
 574                         }
 575                 }
 576                 put_cred(override_creds(override_cred));
 577                 put_cred(override_cred);
 578
 579                 if (!ovl_dentry_is_whiteout(dentry))
 580                         err = ovl_create_upper(dentry, inode, attr);
 581                 else
 582                         err = ovl_create_over_whiteout(dentry, inode, attr);
 583         }
 584 out_revert_creds:
 585         revert_creds(old_cred);
 586         return err;
 587 }
 588
 589 static int ovl_create_object(struct dentry *dentry, int mode, dev_t rdev,
 590                              const char *link)
 591 {
 592         int err;
 593         struct inode *inode;
 594         struct ovl_cattr attr = {
 595                 .rdev = rdev,
 596                 .link = link,
 597         };
 598
 599         err = ovl_want_write(dentry);
 600         if (err)
 601                 goto out;
 602
 603         /* Preallocate inode to be used by ovl_get_inode() */
 604         err = -ENOMEM;
 605         inode = ovl_new_inode(dentry->d_sb, mode, rdev);
 606         if (!inode)
 607                 goto out_drop_write;
 608
 609         spin_lock(&inode->i_lock);
 610         inode->i_state |= I_CREATING;
 611         spin_unlock(&inode->i_lock);
 612
 613         inode_init_owner(inode, dentry->d_parent->d_inode, mode);
 614         attr.mode = inode->i_mode;
 615
 616         err = ovl_create_or_link(dentry, inode, &attr, false);
 617         /* Did we end up using the preallocated inode? */
 618         if (inode != d_inode(dentry))
 619                 iput(inode);
 620
 621 out_drop_write:
 622         ovl_drop_write(dentry);
 623 out:
 624         return err;
 625 }
 626
 627 static int ovl_create(struct inode *dir, struct dentry *dentry, umode_t mode,
 628                       bool excl)
 629 {
 630         return ovl_create_object(dentry, (mode & 07777) | S_IFREG, 0, NULL);
 631 }
 632
 633 static int ovl_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
 634 {
 635         return ovl_create_object(dentry, (mode & 07777) | S_IFDIR, 0, NULL);
 636 }
 637
 638 static int ovl_mknod(struct inode *dir, struct dentry *dentry, umode_t mode,
 639                      dev_t rdev)
 640 {
 641         /* Don't allow creation of "whiteout" on overlay */
 642         if (S_ISCHR(mode) && rdev == WHITEOUT_DEV)
 643                 return -EPERM;
 644
 645         return ovl_create_object(dentry, mode, rdev, NULL);
 646 }
 647
 648 static int ovl_symlink(struct inode *dir, struct dentry *dentry,
 649                        const char *link)
 650 {
 651         return ovl_create_object(dentry, S_IFLNK, 0, link);
 652 }
 653
 654 static int ovl_set_link_redirect(struct dentry *dentry)
 655 {
 656         const struct cred *old_cred;
 657         int err;
 658
 659         old_cred = ovl_override_creds(dentry->d_sb);
 660         err = ovl_set_redirect(dentry, false);
 661         revert_creds(old_cred);
 662
 663         return err;
 664 }
 665
 666 static int ovl_link(struct dentry *old, struct inode *newdir,
 667                     struct dentry *new)
 668 {
 669         int err;
 670         struct inode *inode;
 671
 672         err = ovl_want_write(old);
 673         if (err)
 674                 goto out;
 675
 676         err = ovl_copy_up(old);
 677         if (err)
 678                 goto out_drop_write;
 679
 680         err = ovl_copy_up(new->d_parent);
 681         if (err)
 682                 goto out_drop_write;
 683
 684         if (ovl_is_metacopy_dentry(old)) {
 685                 err = ovl_set_link_redirect(old);
 686                 if (err)
 687                         goto out_drop_write;
 688         }
 689
 690         err = ovl_nlink_start(old);
 691         if (err)
 692                 goto out_drop_write;
 693
 694         inode = d_inode(old);
 695         ihold(inode);
 696
 697         err = ovl_create_or_link(new, inode,
 698                         &(struct ovl_cattr) {.hardlink = ovl_dentry_upper(old)},
 699                         ovl_type_origin(old));
 700         if (err)
 701                 iput(inode);
 702
 703         ovl_nlink_end(old);
 704 out_drop_write:
 705         ovl_drop_write(old);
 706 out:
 707         return err;
 708 }
 709
 710 static bool ovl_matches_upper(struct dentry *dentry, struct dentry *upper)
 711 {
 712         return d_inode(ovl_dentry_upper(dentry)) == d_inode(upper);
 713 }
 714
 715 static int ovl_remove_and_whiteout(struct dentry *dentry,
 716                                    struct list_head *list)
 717 {
 718         struct dentry *workdir = ovl_workdir(dentry);
 719         struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
 720         struct dentry *upper;
 721         struct dentry *opaquedir = NULL;
 722         int err;
 723
 724         if (WARN_ON(!workdir))
 725                 return -EROFS;
 726
 727         if (!list_empty(list)) {
 728                 opaquedir = ovl_clear_empty(dentry, list);
 729                 err = PTR_ERR(opaquedir);
 730                 if (IS_ERR(opaquedir))
 731                         goto out;
 732         }
 733
 734         err = ovl_lock_rename_workdir(workdir, upperdir);
 735         if (err)
 736                 goto out_dput;
 737
 738         upper = lookup_one_len(dentry->d_name.name, upperdir,
 739                                dentry->d_name.len);
 740         err = PTR_ERR(upper);
 741         if (IS_ERR(upper))
 742                 goto out_unlock;
 743
 744         err = -ESTALE;
 745         if ((opaquedir && upper != opaquedir) ||
 746             (!opaquedir && ovl_dentry_upper(dentry) &&
 747              !ovl_matches_upper(dentry, upper))) {
 748                 goto out_dput_upper;
 749         }
 750
 751         err = ovl_cleanup_and_whiteout(workdir, d_inode(upperdir), upper);
 752         if (err)
 753                 goto out_d_drop;
 754
 755         ovl_dir_modified(dentry->d_parent, true);
 756 out_d_drop:
 757         d_drop(dentry);
 758 out_dput_upper:
 759         dput(upper);
 760 out_unlock:
 761         unlock_rename(workdir, upperdir);
 762 out_dput:
 763         dput(opaquedir);
 764 out:
 765         return err;
 766 }
 767
 768 static int ovl_remove_upper(struct dentry *dentry, bool is_dir,
 769                             struct list_head *list)
 770 {
 771         struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
 772         struct inode *dir = upperdir->d_inode;
 773         struct dentry *upper;
 774         struct dentry *opaquedir = NULL;
 775         int err;
 776
 777         if (!list_empty(list)) {
 778                 opaquedir = ovl_clear_empty(dentry, list);
 779                 err = PTR_ERR(opaquedir);
 780                 if (IS_ERR(opaquedir))
 781                         goto out;
 782         }
 783
 784         inode_lock_nested(dir, I_MUTEX_PARENT);
 785         upper = lookup_one_len(dentry->d_name.name, upperdir,
 786                                dentry->d_name.len);
 787         err = PTR_ERR(upper);
 788         if (IS_ERR(upper))
 789                 goto out_unlock;
 790
 791         err = -ESTALE;
 792         if ((opaquedir && upper != opaquedir) ||
 793             (!opaquedir && !ovl_matches_upper(dentry, upper)))
 794                 goto out_dput_upper;
 795
 796         if (is_dir)
 797                 err = vfs_rmdir(dir, upper);
 798         else
 799                 err = vfs_unlink(dir, upper, NULL);
 800         ovl_dir_modified(dentry->d_parent, ovl_type_origin(dentry));
 801
 802         /*
 803          * Keeping this dentry hashed would mean having to release
 804          * upperpath/lowerpath, which could only be done if we are the
 805          * sole user of this dentry.  Too tricky...  Just unhash for
 806          * now.
 807          */
 808         if (!err)
 809                 d_drop(dentry);
 810 out_dput_upper:
 811         dput(upper);
 812 out_unlock:
 813         inode_unlock(dir);
 814         dput(opaquedir);
 815 out:
 816         return err;
 817 }
 818
 819 static bool ovl_pure_upper(struct dentry *dentry)
 820 {
 821         return !ovl_dentry_lower(dentry) &&
 822                !ovl_test_flag(OVL_WHITEOUTS, d_inode(dentry));
 823 }
 824
 825 static int ovl_do_remove(struct dentry *dentry, bool is_dir)
 826 {
 827         int err;
 828         const struct cred *old_cred;
 829         struct dentry *upperdentry;
 830         bool lower_positive = ovl_lower_positive(dentry);
 831         LIST_HEAD(list);
 832
 833         /* No need to clean pure upper removed by vfs_rmdir() */
 834         if (is_dir && (lower_positive || !ovl_pure_upper(dentry))) {
 835                 err = ovl_check_empty_dir(dentry, &list);
 836                 if (err)
 837                         goto out;
 838         }
 839
 840         err = ovl_want_write(dentry);
 841         if (err)
 842                 goto out;
 843
 844         err = ovl_copy_up(dentry->d_parent);
 845         if (err)
 846                 goto out_drop_write;
 847
 848         err = ovl_nlink_start(dentry);
 849         if (err)
 850                 goto out_drop_write;
 851
 852         old_cred = ovl_override_creds(dentry->d_sb);
 853         if (!lower_positive)
 854                 err = ovl_remove_upper(dentry, is_dir, &list);
 855         else
 856                 err = ovl_remove_and_whiteout(dentry, &list);
 857         revert_creds(old_cred);
 858         if (!err) {
 859                 if (is_dir)
 860                         clear_nlink(dentry->d_inode);
 861                 else
 862                         drop_nlink(dentry->d_inode);
 863         }
 864         ovl_nlink_end(dentry);
 865
 866         /*
 867          * Copy ctime
 868          *
 869          * Note: we fail to update ctime if there was no copy-up, only a
 870          * whiteout
 871          */
 872         upperdentry = ovl_dentry_upper(dentry);
 873         if (upperdentry)
 874                 ovl_copyattr(d_inode(upperdentry), d_inode(dentry));
 875
 876 out_drop_write:
 877         ovl_drop_write(dentry);
 878 out:
 879         ovl_cache_free(&list);
 880         return err;
 881 }
 882
 883 static int ovl_unlink(struct inode *dir, struct dentry *dentry)
 884 {
 885         return ovl_do_remove(dentry, false);
 886 }
 887
 888 static int ovl_rmdir(struct inode *dir, struct dentry *dentry)
 889 {
 890         return ovl_do_remove(dentry, true);
 891 }
 892
 893 static bool ovl_type_merge_or_lower(struct dentry *dentry)
 894 {
 895         enum ovl_path_type type = ovl_path_type(dentry);
 896
 897         return OVL_TYPE_MERGE(type) || !OVL_TYPE_UPPER(type);
 898 }
 899
 900 static bool ovl_can_move(struct dentry *dentry)
 901 {
 902         return ovl_redirect_dir(dentry->d_sb) ||
 903                 !d_is_dir(dentry) || !ovl_type_merge_or_lower(dentry);
 904 }
 905
 906 static char *ovl_get_redirect(struct dentry *dentry, bool abs_redirect)
 907 {
 908         char *buf, *ret;
 909         struct dentry *d, *tmp;
 910         int buflen = ovl_redirect_max + 1;
 911
 912         if (!abs_redirect) {
 913                 ret = kstrndup(dentry->d_name.name, dentry->d_name.len,
 914                                GFP_KERNEL);
 915                 goto out;
 916         }
 917
 918         buf = ret = kmalloc(buflen, GFP_KERNEL);
 919         if (!buf)
 920                 goto out;
 921
 922         buflen--;
 923         buf[buflen] = '\0';
 924         for (d = dget(dentry); !IS_ROOT(d);) {
 925                 const char *name;
 926                 int thislen;
 927
 928                 spin_lock(&d->d_lock);
 929                 name = ovl_dentry_get_redirect(d);
 930                 if (name) {
 931                         thislen = strlen(name);
 932                 } else {
 933                         name = d->d_name.name;
 934                         thislen = d->d_name.len;
 935                 }
 936
 937                 /* If path is too long, fall back to userspace move */
 938                 if (thislen + (name[0] != '/') > buflen) {
 939                         ret = ERR_PTR(-EXDEV);
 940                         spin_unlock(&d->d_lock);
 941                         goto out_put;
 942                 }
 943
 944                 buflen -= thislen;
 945                 memcpy(&buf[buflen], name, thislen);
 946                 tmp = dget_dlock(d->d_parent);
 947                 spin_unlock(&d->d_lock);
 948
 949                 dput(d);
 950                 d = tmp;
 951
 952                 /* Absolute redirect: finished */
 953                 if (buf[buflen] == '/')
 954                         break;
 955                 buflen--;
 956                 buf[buflen] = '/';
 957         }
 958         ret = kstrdup(&buf[buflen], GFP_KERNEL);
 959 out_put:
 960         dput(d);
 961         kfree(buf);
 962 out:
 963         return ret ? ret : ERR_PTR(-ENOMEM);
 964 }
 965
 966 static bool ovl_need_absolute_redirect(struct dentry *dentry, bool samedir)
 967 {
 968         struct dentry *lowerdentry;
 969
 970         if (!samedir)
 971                 return true;
 972
 973         if (d_is_dir(dentry))
 974                 return false;
 975
 976         /*
 977          * For non-dir hardlinked files, we need absolute redirects
 978          * in general as two upper hardlinks could be in different
 979          * dirs. We could put a relative redirect now and convert
 980          * it to absolute redirect later. But when nlink > 1 and
 981          * indexing is on, that means relative redirect needs to be
 982          * converted to absolute during copy up of another lower
 983          * hardllink as well.
 984          *
 985          * So without optimizing too much, just check if lower is
 986          * a hard link or not. If lower is hard link, put absolute
 987          * redirect.
 988          */
 989         lowerdentry = ovl_dentry_lower(dentry);
 990         return (d_inode(lowerdentry)->i_nlink > 1);
 991 }
 992
 993 static int ovl_set_redirect(struct dentry *dentry, bool samedir)
 994 {
 995         int err;
 996         const char *redirect = ovl_dentry_get_redirect(dentry);
 997         bool absolute_redirect = ovl_need_absolute_redirect(dentry, samedir);
 998
 999         if (redirect && (!absolute_redirect || redirect[0] == '/'))
1000                 return 0;
1001
1002         redirect = ovl_get_redirect(dentry, absolute_redirect);
1003         if (IS_ERR(redirect))
1004                 return PTR_ERR(redirect);
1005
1006         err = ovl_check_setxattr(dentry, ovl_dentry_upper(dentry),
1007                                  OVL_XATTR_REDIRECT,
1008                                  redirect, strlen(redirect), -EXDEV);
1009         if (!err) {
1010                 spin_lock(&dentry->d_lock);
1011                 ovl_dentry_set_redirect(dentry, redirect);
1012                 spin_unlock(&dentry->d_lock);
1013         } else {
1014                 kfree(redirect);
1015                 pr_warn_ratelimited("overlayfs: failed to set redirect (%i)\n",
1016                                     err);
1017                 /* Fall back to userspace copy-up */
1018                 err = -EXDEV;
1019         }
1020         return err;
1021 }
1022
1023 static int ovl_rename(struct inode *olddir, struct dentry *old,
1024                       struct inode *newdir, struct dentry *new,
1025                       unsigned int flags)
1026 {
1027         int err;
1028         struct dentry *old_upperdir;
1029         struct dentry *new_upperdir;
1030         struct dentry *olddentry;
1031         struct dentry *newdentry;
1032         struct dentry *trap;
1033         bool old_opaque;
1034         bool new_opaque;
1035         bool cleanup_whiteout = false;
1036         bool update_nlink = false;
1037         bool overwrite = !(flags & RENAME_EXCHANGE);
1038         bool is_dir = d_is_dir(old);
1039         bool new_is_dir = d_is_dir(new);
1040         bool samedir = olddir == newdir;
1041         struct dentry *opaquedir = NULL;
1042         const struct cred *old_cred = NULL;
1043         LIST_HEAD(list);
1044
1045         err = -EINVAL;
1046         if (flags & ~(RENAME_EXCHANGE | RENAME_NOREPLACE))
1047                 goto out;
1048
1049         flags &= ~RENAME_NOREPLACE;
1050
1051         /* Don't copy up directory trees */
1052         err = -EXDEV;
1053         if (!ovl_can_move(old))
1054                 goto out;
1055         if (!overwrite && !ovl_can_move(new))
1056                 goto out;
1057
1058         if (overwrite && new_is_dir && !ovl_pure_upper(new)) {
1059                 err = ovl_check_empty_dir(new, &list);
1060                 if (err)
1061                         goto out;
1062         }
1063
1064         if (overwrite) {
1065                 if (ovl_lower_positive(old)) {
1066                         if (!ovl_dentry_is_whiteout(new)) {
1067                                 /* Whiteout source */
1068                                 flags |= RENAME_WHITEOUT;
1069                         } else {
1070                                 /* Switch whiteouts */
1071                                 flags |= RENAME_EXCHANGE;
1072                         }
1073                 } else if (is_dir && ovl_dentry_is_whiteout(new)) {
1074                         flags |= RENAME_EXCHANGE;
1075                         cleanup_whiteout = true;
1076                 }
1077         }
1078
1079         err = ovl_want_write(old);
1080         if (err)
1081                 goto out;
1082
1083         err = ovl_copy_up(old);
1084         if (err)
1085                 goto out_drop_write;
1086
1087         err = ovl_copy_up(new->d_parent);
1088         if (err)
1089                 goto out_drop_write;
1090         if (!overwrite) {
1091                 err = ovl_copy_up(new);
1092                 if (err)
1093                         goto out_drop_write;
1094         } else if (d_inode(new)) {
1095                 err = ovl_nlink_start(new);
1096                 if (err)
1097                         goto out_drop_write;
1098
1099                 update_nlink = true;
1100         }
1101
1102         old_cred = ovl_override_creds(old->d_sb);
1103
1104         if (!list_empty(&list)) {
1105                 opaquedir = ovl_clear_empty(new, &list);
1106                 err = PTR_ERR(opaquedir);
1107                 if (IS_ERR(opaquedir)) {
1108                         opaquedir = NULL;
1109                         goto out_revert_creds;
1110                 }
1111         }
1112
1113         old_upperdir = ovl_dentry_upper(old->d_parent);
1114         new_upperdir = ovl_dentry_upper(new->d_parent);
1115
1116         if (!samedir) {
1117                 /*
1118                  * When moving a merge dir or non-dir with copy up origin into
1119                  * a new parent, we are marking the new parent dir "impure".
1120                  * When ovl_iterate() iterates an "impure" upper dir, it will
1121                  * lookup the origin inodes of the entries to fill d_ino.
1122                  */
1123                 if (ovl_type_origin(old)) {
1124                         err = ovl_set_impure(new->d_parent, new_upperdir);
1125                         if (err)
1126                                 goto out_revert_creds;
1127                 }
1128                 if (!overwrite && ovl_type_origin(new)) {
1129                         err = ovl_set_impure(old->d_parent, old_upperdir);
1130                         if (err)
1131                                 goto out_revert_creds;
1132                 }
1133         }
1134
1135         trap = lock_rename(new_upperdir, old_upperdir);
1136
1137         olddentry = lookup_one_len(old->d_name.name, old_upperdir,
1138                                    old->d_name.len);
1139         err = PTR_ERR(olddentry);
1140         if (IS_ERR(olddentry))
1141                 goto out_unlock;
1142
1143         err = -ESTALE;
1144         if (!ovl_matches_upper(old, olddentry))
1145                 goto out_dput_old;
1146
1147         newdentry = lookup_one_len(new->d_name.name, new_upperdir,
1148                                    new->d_name.len);
1149         err = PTR_ERR(newdentry);
1150         if (IS_ERR(newdentry))
1151                 goto out_dput_old;
1152
1153         old_opaque = ovl_dentry_is_opaque(old);
1154         new_opaque = ovl_dentry_is_opaque(new);
1155
1156         err = -ESTALE;
1157         if (d_inode(new) && ovl_dentry_upper(new)) {
1158                 if (opaquedir) {
1159                         if (newdentry != opaquedir)
1160                                 goto out_dput;
1161                 } else {
1162                         if (!ovl_matches_upper(new, newdentry))
1163                                 goto out_dput;
1164                 }
1165         } else {
1166                 if (!d_is_negative(newdentry) &&
1167                     (!new_opaque || !ovl_is_whiteout(newdentry)))
1168                         goto out_dput;
1169         }
1170
1171         if (olddentry == trap)
1172                 goto out_dput;
1173         if (newdentry == trap)
1174                 goto out_dput;
1175
1176         if (WARN_ON(olddentry->d_inode == newdentry->d_inode))
1177                 goto out_dput;
1178
1179         err = 0;
1180         if (ovl_type_merge_or_lower(old))
1181                 err = ovl_set_redirect(old, samedir);
1182         else if (is_dir && !old_opaque && ovl_type_merge(new->d_parent))
1183                 err = ovl_set_opaque_xerr(old, olddentry, -EXDEV);
1184         if (err)
1185                 goto out_dput;
1186
1187         if (!overwrite && ovl_type_merge_or_lower(new))
1188                 err = ovl_set_redirect(new, samedir);
1189         else if (!overwrite && new_is_dir && !new_opaque &&
1190                  ovl_type_merge(old->d_parent))
1191                 err = ovl_set_opaque_xerr(new, newdentry, -EXDEV);
1192         if (err)
1193                 goto out_dput;
1194
1195         err = ovl_do_rename(old_upperdir->d_inode, olddentry,
1196                             new_upperdir->d_inode, newdentry, flags);
1197         if (err)
1198                 goto out_dput;
1199
1200         if (cleanup_whiteout)
1201                 ovl_cleanup(old_upperdir->d_inode, newdentry);
1202
1203         if (overwrite && d_inode(new)) {
1204                 if (new_is_dir)
1205                         clear_nlink(d_inode(new));
1206                 else
1207                         drop_nlink(d_inode(new));
1208         }
1209
1210         ovl_dir_modified(old->d_parent, ovl_type_origin(old) ||
1211                          (!overwrite && ovl_type_origin(new)));
1212         ovl_dir_modified(new->d_parent, ovl_type_origin(old) ||
1213                          (d_inode(new) && ovl_type_origin(new)));
1214
1215         /* copy ctime: */
1216         ovl_copyattr(d_inode(olddentry), d_inode(old));
1217         if (d_inode(new) && ovl_dentry_upper(new))
1218                 ovl_copyattr(d_inode(newdentry), d_inode(new));
1219
1220 out_dput:
1221         dput(newdentry);
1222 out_dput_old:
1223         dput(olddentry);
1224 out_unlock:
1225         unlock_rename(new_upperdir, old_upperdir);
1226 out_revert_creds:
1227         revert_creds(old_cred);
1228         if (update_nlink)
1229                 ovl_nlink_end(new);
1230 out_drop_write:
1231         ovl_drop_write(old);
1232 out:
1233         dput(opaquedir);
1234         ovl_cache_free(&list);
1235         return err;
1236 }
1237
1238 const struct inode_operations ovl_dir_inode_operations = {
1239         .lookup         = ovl_lookup,
1240         .mkdir          = ovl_mkdir,
1241         .symlink        = ovl_symlink,
1242         .unlink         = ovl_unlink,
1243         .rmdir          = ovl_rmdir,
1244         .rename         = ovl_rename,
1245         .link           = ovl_link,
1246         .setattr        = ovl_setattr,
1247         .create         = ovl_create,
1248         .mknod          = ovl_mknod,
1249         .permission     = ovl_permission,
1250         .getattr        = ovl_getattr,
1251         .listxattr      = ovl_listxattr,
1252         .get_acl        = ovl_get_acl,
1253         .update_time    = ovl_update_time,
1254 };