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