fs/overlayfs/dir.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  *
   4  * Copyright (C) 2011 Novell Inc.
   5  */
   6
   7 #include <linux/fs.h>
   8 #include <linux/namei.h>
   9 #include <linux/xattr.h>
  10 #include <linux/security.h>
  11 #include <linux/cred.h>
  12 #include <linux/module.h>
  13 #include <linux/posix_acl.h>
  14 #include <linux/posix_acl_xattr.h>
  15 #include <linux/atomic.h>
  16 #include <linux/ratelimit.h>
  17 #include "overlayfs.h"
  18
  19 static unsigned short ovl_redirect_max = 256;
  20 module_param_named(redirect_max, ovl_redirect_max, ushort, 0644);
  21 MODULE_PARM_DESC(redirect_max,
  22                  "Maximum length of absolute redirect xattr value");
  23
  24 static int ovl_set_redirect(struct dentry *dentry, bool samedir);
  25
  26 int ovl_cleanup(struct inode *wdir, struct dentry *wdentry)
  27 {
  28         int err;
  29
  30         dget(wdentry);
  31         if (d_is_dir(wdentry))
  32                 err = ovl_do_rmdir(wdir, wdentry);
  33         else
  34                 err = ovl_do_unlink(wdir, wdentry);
  35         dput(wdentry);
  36
  37         if (err) {
  38                 pr_err("cleanup of '%pd2' failed (%i)\n",
  39                        wdentry, err);
  40         }
  41
  42         return err;
  43 }
  44
  45 struct dentry *ovl_lookup_temp(struct dentry *workdir)
  46 {
  47         struct dentry *temp;
  48         char name[20];
  49         static atomic_t temp_id = ATOMIC_INIT(0);
  50
  51         /* counter is allowed to wrap, since temp dentries are ephemeral */
  52         snprintf(name, sizeof(name), "#%x", atomic_inc_return(&temp_id));
  53
  54         temp = lookup_one_len(name, workdir, strlen(name));
  55         if (!IS_ERR(temp) && temp->d_inode) {
  56                 pr_err("workdir/%s already exists\n", name);
  57                 dput(temp);
  58                 temp = ERR_PTR(-EIO);
  59         }
  60
  61         return temp;
  62 }
  63
  64 /* caller holds i_mutex on workdir */
  65 static struct dentry *ovl_whiteout(struct ovl_fs *ofs)
  66 {
  67         int err;
  68         struct dentry *whiteout;
  69         struct dentry *workdir = ofs->workdir;
  70         struct inode *wdir = workdir->d_inode;
  71
  72         if (!ofs->whiteout) {
  73                 whiteout = ovl_lookup_temp(workdir);
  74                 if (IS_ERR(whiteout))
  75                         goto out;
  76
  77                 err = ovl_do_whiteout(wdir, whiteout);
  78                 if (err) {
  79                         dput(whiteout);
  80                         whiteout = ERR_PTR(err);
  81                         goto out;
  82                 }
  83                 ofs->whiteout = whiteout;
  84         }
  85
  86         if (ofs->share_whiteout) {
  87                 whiteout = ovl_lookup_temp(workdir);
  88                 if (IS_ERR(whiteout))
  89                         goto out;
  90
  91                 err = ovl_do_link(ofs->whiteout, wdir, whiteout);
  92                 if (!err)
  93                         goto out;
  94
  95                 if (err != -EMLINK) {
  96                         pr_warn("Failed to link whiteout - disabling whiteout inode sharing(nlink=%u, err=%i)\n",
  97                                 ofs->whiteout->d_inode->i_nlink, err);
  98                         ofs->share_whiteout = false;
  99                 }
 100                 dput(whiteout);
 101         }
 102         whiteout = ofs->whiteout;
 103         ofs->whiteout = NULL;
 104 out:
 105         return whiteout;
 106 }
 107
 108 /* Caller must hold i_mutex on both workdir and dir */
 109 int ovl_cleanup_and_whiteout(struct ovl_fs *ofs, struct inode *dir,
 110                              struct dentry *dentry)
 111 {
 112         struct inode *wdir = ofs->workdir->d_inode;
 113         struct dentry *whiteout;
 114         int err;
 115         int flags = 0;
 116
 117         whiteout = ovl_whiteout(ofs);
 118         err = PTR_ERR(whiteout);
 119         if (IS_ERR(whiteout))
 120                 return err;
 121
 122         if (d_is_dir(dentry))
 123                 flags = RENAME_EXCHANGE;
 124
 125         err = ovl_do_rename(wdir, whiteout, dir, dentry, flags);
 126         if (err)
 127                 goto kill_whiteout;
 128         if (flags)
 129                 ovl_cleanup(wdir, dentry);
 130
 131 out:
 132         dput(whiteout);
 133         return err;
 134
 135 kill_whiteout:
 136         ovl_cleanup(wdir, whiteout);
 137         goto out;
 138 }
 139
 140 int ovl_mkdir_real(struct inode *dir, struct dentry **newdentry, umode_t mode)
 141 {
 142         int err;
 143         struct dentry *d, *dentry = *newdentry;
 144
 145         err = ovl_do_mkdir(dir, dentry, mode);
 146         if (err)
 147                 return err;
 148
 149         if (likely(!d_unhashed(dentry)))
 150                 return 0;
 151
 152         /*
 153          * vfs_mkdir() may succeed and leave the dentry passed
 154          * to it unhashed and negative. If that happens, try to
 155          * lookup a new hashed and positive dentry.
 156          */
 157         d = lookup_one_len(dentry->d_name.name, dentry->d_parent,
 158                            dentry->d_name.len);
 159         if (IS_ERR(d)) {
 160                 pr_warn("failed lookup after mkdir (%pd2, err=%i).\n",
 161                         dentry, err);
 162                 return PTR_ERR(d);
 163         }
 164         dput(dentry);
 165         *newdentry = d;
 166
 167         return 0;
 168 }
 169
 170 struct dentry *ovl_create_real(struct inode *dir, struct dentry *newdentry,
 171                                struct ovl_cattr *attr)
 172 {
 173         int err;
 174
 175         if (IS_ERR(newdentry))
 176                 return newdentry;
 177
 178         err = -ESTALE;
 179         if (newdentry->d_inode)
 180                 goto out;
 181
 182         if (attr->hardlink) {
 183                 err = ovl_do_link(attr->hardlink, dir, newdentry);
 184         } else {
 185                 switch (attr->mode & S_IFMT) {
 186                 case S_IFREG:
 187                         err = ovl_do_create(dir, newdentry, attr->mode);
 188                         break;
 189
 190                 case S_IFDIR:
 191                         /* mkdir is special... */
 192                         err =  ovl_mkdir_real(dir, &newdentry, attr->mode);
 193                         break;
 194
 195                 case S_IFCHR:
 196                 case S_IFBLK:
 197                 case S_IFIFO:
 198                 case S_IFSOCK:
 199                         err = ovl_do_mknod(dir, newdentry, attr->mode,
 200                                            attr->rdev);
 201                         break;
 202
 203                 case S_IFLNK:
 204                         err = ovl_do_symlink(dir, newdentry, attr->link);
 205                         break;
 206
 207                 default:
 208                         err = -EPERM;
 209                 }
 210         }
 211         if (!err && WARN_ON(!newdentry->d_inode)) {
 212                 /*
 213                  * Not quite sure if non-instantiated dentry is legal or not.
 214                  * VFS doesn't seem to care so check and warn here.
 215                  */
 216                 err = -EIO;
 217         }
 218 out:
 219         if (err) {
 220                 dput(newdentry);
 221                 return ERR_PTR(err);
 222         }
 223         return newdentry;
 224 }
 225
 226 struct dentry *ovl_create_temp(struct dentry *workdir, struct ovl_cattr *attr)
 227 {
 228         return ovl_create_real(d_inode(workdir), ovl_lookup_temp(workdir),
 229                                attr);
 230 }
 231
 232 static int ovl_set_opaque_xerr(struct dentry *dentry, struct dentry *upper,
 233                                int xerr)
 234 {
 235         struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
 236         int err;
 237
 238         err = ovl_check_setxattr(ofs, upper, OVL_XATTR_OPAQUE, "y", 1, xerr);
 239         if (!err)
 240                 ovl_dentry_set_opaque(dentry);
 241
 242         return err;
 243 }
 244
 245 static int ovl_set_opaque(struct dentry *dentry, struct dentry *upperdentry)
 246 {
 247         /*
 248          * Fail with -EIO when trying to create opaque dir and upper doesn't
 249          * support xattrs. ovl_rename() calls ovl_set_opaque_xerr(-EXDEV) to
 250          * return a specific error for noxattr case.
 251          */
 252         return ovl_set_opaque_xerr(dentry, upperdentry, -EIO);
 253 }
 254
 255 /*
 256  * Common operations required to be done after creation of file on upper.
 257  * If @hardlink is false, then @inode is a pre-allocated inode, we may or
 258  * may not use to instantiate the new dentry.
 259  */
 260 static int ovl_instantiate(struct dentry *dentry, struct inode *inode,
 261                            struct dentry *newdentry, bool hardlink)
 262 {
 263         struct ovl_inode_params oip = {
 264                 .upperdentry = newdentry,
 265                 .newinode = inode,
 266         };
 267
 268         ovl_dir_modified(dentry->d_parent, false);
 269         ovl_dentry_set_upper_alias(dentry);
 270         ovl_dentry_update_reval(dentry, newdentry,
 271                         DCACHE_OP_REVALIDATE | DCACHE_OP_WEAK_REVALIDATE);
 272
 273         if (!hardlink) {
 274                 /*
 275                  * ovl_obtain_alias() can be called after ovl_create_real()
 276                  * and before we get here, so we may get an inode from cache
 277                  * with the same real upperdentry that is not the inode we
 278                  * pre-allocated.  In this case we will use the cached inode
 279                  * to instantiate the new dentry.
 280                  *
 281                  * XXX: if we ever use ovl_obtain_alias() to decode directory
 282                  * file handles, need to use ovl_get_inode_locked() and
 283                  * d_instantiate_new() here to prevent from creating two
 284                  * hashed directory inode aliases.
 285                  */
 286                 inode = ovl_get_inode(dentry->d_sb, &oip);
 287                 if (IS_ERR(inode))
 288                         return PTR_ERR(inode);
 289                 if (inode == oip.newinode)
 290                         ovl_set_flag(OVL_UPPERDATA, inode);
 291         } else {
 292                 WARN_ON(ovl_inode_real(inode) != d_inode(newdentry));
 293                 dput(newdentry);
 294                 inc_nlink(inode);
 295         }
 296
 297         d_instantiate(dentry, inode);
 298         if (inode != oip.newinode) {
 299                 pr_warn_ratelimited("newly created inode found in cache (%pd2)\n",
 300                                     dentry);
 301         }
 302
 303         /* Force lookup of new upper hardlink to find its lower */
 304         if (hardlink)
 305                 d_drop(dentry);
 306
 307         return 0;
 308 }
 309
 310 static bool ovl_type_merge(struct dentry *dentry)
 311 {
 312         return OVL_TYPE_MERGE(ovl_path_type(dentry));
 313 }
 314
 315 static bool ovl_type_origin(struct dentry *dentry)
 316 {
 317         return OVL_TYPE_ORIGIN(ovl_path_type(dentry));
 318 }
 319
 320 static int ovl_create_upper(struct dentry *dentry, struct inode *inode,
 321                             struct ovl_cattr *attr)
 322 {
 323         struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
 324         struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
 325         struct inode *udir = upperdir->d_inode;
 326         struct dentry *newdentry;
 327         int err;
 328
 329         if (!attr->hardlink && !IS_POSIXACL(udir))
 330                 attr->mode &= ~current_umask();
 331
 332         inode_lock_nested(udir, I_MUTEX_PARENT);
 333         newdentry = ovl_create_real(udir,
 334                                     lookup_one_len(dentry->d_name.name,
 335                                                    upperdir,
 336                                                    dentry->d_name.len),
 337                                     attr);
 338         err = PTR_ERR(newdentry);
 339         if (IS_ERR(newdentry))
 340                 goto out_unlock;
 341
 342         if (ovl_type_merge(dentry->d_parent) && d_is_dir(newdentry) &&
 343             !ovl_allow_offline_changes(ofs)) {
 344                 /* Setting opaque here is just an optimization, allow to fail */
 345                 ovl_set_opaque(dentry, newdentry);
 346         }
 347
 348         err = ovl_instantiate(dentry, inode, newdentry, !!attr->hardlink);
 349         if (err)
 350                 goto out_cleanup;
 351 out_unlock:
 352         inode_unlock(udir);
 353         return err;
 354
 355 out_cleanup:
 356         ovl_cleanup(udir, newdentry);
 357         dput(newdentry);
 358         goto out_unlock;
 359 }
 360
 361 static struct dentry *ovl_clear_empty(struct dentry *dentry,
 362                                       struct list_head *list)
 363 {
 364         struct dentry *workdir = ovl_workdir(dentry);
 365         struct inode *wdir = workdir->d_inode;
 366         struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
 367         struct inode *udir = upperdir->d_inode;
 368         struct path upperpath;
 369         struct dentry *upper;
 370         struct dentry *opaquedir;
 371         struct kstat stat;
 372         int err;
 373
 374         if (WARN_ON(!workdir))
 375                 return ERR_PTR(-EROFS);
 376
 377         err = ovl_lock_rename_workdir(workdir, upperdir);
 378         if (err)
 379                 goto out;
 380
 381         ovl_path_upper(dentry, &upperpath);
 382         err = vfs_getattr(&upperpath, &stat,
 383                           STATX_BASIC_STATS, AT_STATX_SYNC_AS_STAT);
 384         if (err)
 385                 goto out_unlock;
 386
 387         err = -ESTALE;
 388         if (!S_ISDIR(stat.mode))
 389                 goto out_unlock;
 390         upper = upperpath.dentry;
 391         if (upper->d_parent->d_inode != udir)
 392                 goto out_unlock;
 393
 394         opaquedir = ovl_create_temp(workdir, OVL_CATTR(stat.mode));
 395         err = PTR_ERR(opaquedir);
 396         if (IS_ERR(opaquedir))
 397                 goto out_unlock;
 398
 399         err = ovl_copy_xattr(dentry->d_sb, upper, opaquedir);
 400         if (err)
 401                 goto out_cleanup;
 402
 403         err = ovl_set_opaque(dentry, opaquedir);
 404         if (err)
 405                 goto out_cleanup;
 406
 407         inode_lock(opaquedir->d_inode);
 408         err = ovl_set_attr(opaquedir, &stat);
 409         inode_unlock(opaquedir->d_inode);
 410         if (err)
 411                 goto out_cleanup;
 412
 413         err = ovl_do_rename(wdir, opaquedir, udir, upper, RENAME_EXCHANGE);
 414         if (err)
 415                 goto out_cleanup;
 416
 417         ovl_cleanup_whiteouts(upper, list);
 418         ovl_cleanup(wdir, upper);
 419         unlock_rename(workdir, upperdir);
 420
 421         /* dentry's upper doesn't match now, get rid of it */
 422         d_drop(dentry);
 423
 424         return opaquedir;
 425
 426 out_cleanup:
 427         ovl_cleanup(wdir, opaquedir);
 428         dput(opaquedir);
 429 out_unlock:
 430         unlock_rename(workdir, upperdir);
 431 out:
 432         return ERR_PTR(err);
 433 }
 434
 435 static int ovl_set_upper_acl(struct dentry *upperdentry, const char *name,
 436                              const struct posix_acl *acl)
 437 {
 438         void *buffer;
 439         size_t size;
 440         int err;
 441
 442         if (!IS_ENABLED(CONFIG_FS_POSIX_ACL) || !acl)
 443                 return 0;
 444
 445         size = posix_acl_xattr_size(acl->a_count);
 446         buffer = kmalloc(size, GFP_KERNEL);
 447         if (!buffer)
 448                 return -ENOMEM;
 449
 450         err = posix_acl_to_xattr(&init_user_ns, acl, buffer, size);
 451         if (err < 0)
 452                 goto out_free;
 453
 454         err = vfs_setxattr(&init_user_ns, upperdentry, name, buffer, size, XATTR_CREATE);
 455 out_free:
 456         kfree(buffer);
 457         return err;
 458 }
 459
 460 static int ovl_create_over_whiteout(struct dentry *dentry, struct inode *inode,
 461                                     struct ovl_cattr *cattr)
 462 {
 463         struct dentry *workdir = ovl_workdir(dentry);
 464         struct inode *wdir = workdir->d_inode;
 465         struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
 466         struct inode *udir = upperdir->d_inode;
 467         struct dentry *upper;
 468         struct dentry *newdentry;
 469         int err;
 470         struct posix_acl *acl, *default_acl;
 471         bool hardlink = !!cattr->hardlink;
 472
 473         if (WARN_ON(!workdir))
 474                 return -EROFS;
 475
 476         if (!hardlink) {
 477                 err = posix_acl_create(dentry->d_parent->d_inode,
 478                                        &cattr->mode, &default_acl, &acl);
 479                 if (err)
 480                         return err;
 481         }
 482
 483         err = ovl_lock_rename_workdir(workdir, upperdir);
 484         if (err)
 485                 goto out;
 486
 487         upper = lookup_one_len(dentry->d_name.name, upperdir,
 488                                dentry->d_name.len);
 489         err = PTR_ERR(upper);
 490         if (IS_ERR(upper))
 491                 goto out_unlock;
 492
 493         err = -ESTALE;
 494         if (d_is_negative(upper) || !IS_WHITEOUT(d_inode(upper)))
 495                 goto out_dput;
 496
 497         newdentry = ovl_create_temp(workdir, cattr);
 498         err = PTR_ERR(newdentry);
 499         if (IS_ERR(newdentry))
 500                 goto out_dput;
 501
 502         /*
 503          * mode could have been mutilated due to umask (e.g. sgid directory)
 504          */
 505         if (!hardlink &&
 506             !S_ISLNK(cattr->mode) &&
 507             newdentry->d_inode->i_mode != cattr->mode) {
 508                 struct iattr attr = {
 509                         .ia_valid = ATTR_MODE,
 510                         .ia_mode = cattr->mode,
 511                 };
 512                 inode_lock(newdentry->d_inode);
 513                 err = notify_change(&init_user_ns, newdentry, &attr, NULL);
 514                 inode_unlock(newdentry->d_inode);
 515                 if (err)
 516                         goto out_cleanup;
 517         }
 518         if (!hardlink) {
 519                 err = ovl_set_upper_acl(newdentry, XATTR_NAME_POSIX_ACL_ACCESS,
 520                                         acl);
 521                 if (err)
 522                         goto out_cleanup;
 523
 524                 err = ovl_set_upper_acl(newdentry, XATTR_NAME_POSIX_ACL_DEFAULT,
 525                                         default_acl);
 526                 if (err)
 527                         goto out_cleanup;
 528         }
 529
 530         if (!hardlink && S_ISDIR(cattr->mode)) {
 531                 err = ovl_set_opaque(dentry, newdentry);
 532                 if (err)
 533                         goto out_cleanup;
 534
 535                 err = ovl_do_rename(wdir, newdentry, udir, upper,
 536                                     RENAME_EXCHANGE);
 537                 if (err)
 538                         goto out_cleanup;
 539
 540                 ovl_cleanup(wdir, upper);
 541         } else {
 542                 err = ovl_do_rename(wdir, newdentry, udir, upper, 0);
 543                 if (err)
 544                         goto out_cleanup;
 545         }
 546         err = ovl_instantiate(dentry, inode, newdentry, hardlink);
 547         if (err) {
 548                 ovl_cleanup(udir, newdentry);
 549                 dput(newdentry);
 550         }
 551 out_dput:
 552         dput(upper);
 553 out_unlock:
 554         unlock_rename(workdir, upperdir);
 555 out:
 556         if (!hardlink) {
 557                 posix_acl_release(acl);
 558                 posix_acl_release(default_acl);
 559         }
 560         return err;
 561
 562 out_cleanup:
 563         ovl_cleanup(wdir, newdentry);
 564         dput(newdentry);
 565         goto out_dput;
 566 }
 567
 568 static int ovl_create_or_link(struct dentry *dentry, struct inode *inode,
 569                               struct ovl_cattr *attr, bool origin)
 570 {
 571         int err;
 572         const struct cred *old_cred;
 573         struct cred *override_cred;
 574         struct dentry *parent = dentry->d_parent;
 575
 576         err = ovl_copy_up(parent);
 577         if (err)
 578                 return err;
 579
 580         old_cred = ovl_override_creds(dentry->d_sb);
 581
 582         /*
 583          * When linking a file with copy up origin into a new parent, mark the
 584          * new parent dir "impure".
 585          */
 586         if (origin) {
 587                 err = ovl_set_impure(parent, ovl_dentry_upper(parent));
 588                 if (err)
 589                         goto out_revert_creds;
 590         }
 591
 592         err = -ENOMEM;
 593         override_cred = prepare_creds();
 594         if (override_cred) {
 595                 override_cred->fsuid = inode->i_uid;
 596                 override_cred->fsgid = inode->i_gid;
 597                 if (!attr->hardlink) {
 598                         err = security_dentry_create_files_as(dentry,
 599                                         attr->mode, &dentry->d_name, old_cred,
 600                                         override_cred);
 601                         if (err) {
 602                                 put_cred(override_cred);
 603                                 goto out_revert_creds;
 604                         }
 605                 }
 606                 put_cred(override_creds(override_cred));
 607                 put_cred(override_cred);
 608
 609                 if (!ovl_dentry_is_whiteout(dentry))
 610                         err = ovl_create_upper(dentry, inode, attr);
 611                 else
 612                         err = ovl_create_over_whiteout(dentry, inode, attr);
 613         }
 614 out_revert_creds:
 615         revert_creds(old_cred);
 616         return err;
 617 }
 618
 619 static int ovl_create_object(struct dentry *dentry, int mode, dev_t rdev,
 620                              const char *link)
 621 {
 622         int err;
 623         struct inode *inode;
 624         struct ovl_cattr attr = {
 625                 .rdev = rdev,
 626                 .link = link,
 627         };
 628
 629         err = ovl_want_write(dentry);
 630         if (err)
 631                 goto out;
 632
 633         /* Preallocate inode to be used by ovl_get_inode() */
 634         err = -ENOMEM;
 635         inode = ovl_new_inode(dentry->d_sb, mode, rdev);
 636         if (!inode)
 637                 goto out_drop_write;
 638
 639         spin_lock(&inode->i_lock);
 640         inode->i_state |= I_CREATING;
 641         spin_unlock(&inode->i_lock);
 642
 643         inode_init_owner(&init_user_ns, inode, dentry->d_parent->d_inode, mode);
 644         attr.mode = inode->i_mode;
 645
 646         err = ovl_create_or_link(dentry, inode, &attr, false);
 647         /* Did we end up using the preallocated inode? */
 648         if (inode != d_inode(dentry))
 649                 iput(inode);
 650
 651 out_drop_write:
 652         ovl_drop_write(dentry);
 653 out:
 654         return err;
 655 }
 656
 657 static int ovl_create(struct user_namespace *mnt_userns, struct inode *dir,
 658                       struct dentry *dentry, umode_t mode, bool excl)
 659 {
 660         return ovl_create_object(dentry, (mode & 07777) | S_IFREG, 0, NULL);
 661 }
 662
 663 static int ovl_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
 664                      struct dentry *dentry, umode_t mode)
 665 {
 666         return ovl_create_object(dentry, (mode & 07777) | S_IFDIR, 0, NULL);
 667 }
 668
 669 static int ovl_mknod(struct user_namespace *mnt_userns, struct inode *dir,
 670                      struct dentry *dentry, umode_t mode, dev_t rdev)
 671 {
 672         /* Don't allow creation of "whiteout" on overlay */
 673         if (S_ISCHR(mode) && rdev == WHITEOUT_DEV)
 674                 return -EPERM;
 675
 676         return ovl_create_object(dentry, mode, rdev, NULL);
 677 }
 678
 679 static int ovl_symlink(struct user_namespace *mnt_userns, struct inode *dir,
 680                        struct dentry *dentry, const char *link)
 681 {
 682         return ovl_create_object(dentry, S_IFLNK, 0, link);
 683 }
 684
 685 static int ovl_set_link_redirect(struct dentry *dentry)
 686 {
 687         const struct cred *old_cred;
 688         int err;
 689
 690         old_cred = ovl_override_creds(dentry->d_sb);
 691         err = ovl_set_redirect(dentry, false);
 692         revert_creds(old_cred);
 693
 694         return err;
 695 }
 696
 697 static int ovl_link(struct dentry *old, struct inode *newdir,
 698                     struct dentry *new)
 699 {
 700         int err;
 701         struct inode *inode;
 702
 703         err = ovl_want_write(old);
 704         if (err)
 705                 goto out;
 706
 707         err = ovl_copy_up(old);
 708         if (err)
 709                 goto out_drop_write;
 710
 711         err = ovl_copy_up(new->d_parent);
 712         if (err)
 713                 goto out_drop_write;
 714
 715         if (ovl_is_metacopy_dentry(old)) {
 716                 err = ovl_set_link_redirect(old);
 717                 if (err)
 718                         goto out_drop_write;
 719         }
 720
 721         err = ovl_nlink_start(old);
 722         if (err)
 723                 goto out_drop_write;
 724
 725         inode = d_inode(old);
 726         ihold(inode);
 727
 728         err = ovl_create_or_link(new, inode,
 729                         &(struct ovl_cattr) {.hardlink = ovl_dentry_upper(old)},
 730                         ovl_type_origin(old));
 731         if (err)
 732                 iput(inode);
 733
 734         ovl_nlink_end(old);
 735 out_drop_write:
 736         ovl_drop_write(old);
 737 out:
 738         return err;
 739 }
 740
 741 static bool ovl_matches_upper(struct dentry *dentry, struct dentry *upper)
 742 {
 743         return d_inode(ovl_dentry_upper(dentry)) == d_inode(upper);
 744 }
 745
 746 static int ovl_remove_and_whiteout(struct dentry *dentry,
 747                                    struct list_head *list)
 748 {
 749         struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
 750         struct dentry *workdir = ovl_workdir(dentry);
 751         struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
 752         struct dentry *upper;
 753         struct dentry *opaquedir = NULL;
 754         int err;
 755
 756         if (WARN_ON(!workdir))
 757                 return -EROFS;
 758
 759         if (!list_empty(list)) {
 760                 opaquedir = ovl_clear_empty(dentry, list);
 761                 err = PTR_ERR(opaquedir);
 762                 if (IS_ERR(opaquedir))
 763                         goto out;
 764         }
 765
 766         err = ovl_lock_rename_workdir(workdir, upperdir);
 767         if (err)
 768                 goto out_dput;
 769
 770         upper = lookup_one_len(dentry->d_name.name, upperdir,
 771                                dentry->d_name.len);
 772         err = PTR_ERR(upper);
 773         if (IS_ERR(upper))
 774                 goto out_unlock;
 775
 776         err = -ESTALE;
 777         if ((opaquedir && upper != opaquedir) ||
 778             (!opaquedir && ovl_dentry_upper(dentry) &&
 779              !ovl_matches_upper(dentry, upper))) {
 780                 goto out_dput_upper;
 781         }
 782
 783         err = ovl_cleanup_and_whiteout(ofs, d_inode(upperdir), upper);
 784         if (err)
 785                 goto out_d_drop;
 786
 787         ovl_dir_modified(dentry->d_parent, true);
 788 out_d_drop:
 789         d_drop(dentry);
 790 out_dput_upper:
 791         dput(upper);
 792 out_unlock:
 793         unlock_rename(workdir, upperdir);
 794 out_dput:
 795         dput(opaquedir);
 796 out:
 797         return err;
 798 }
 799
 800 static int ovl_remove_upper(struct dentry *dentry, bool is_dir,
 801                             struct list_head *list)
 802 {
 803         struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
 804         struct inode *dir = upperdir->d_inode;
 805         struct dentry *upper;
 806         struct dentry *opaquedir = NULL;
 807         int err;
 808
 809         if (!list_empty(list)) {
 810                 opaquedir = ovl_clear_empty(dentry, list);
 811                 err = PTR_ERR(opaquedir);
 812                 if (IS_ERR(opaquedir))
 813                         goto out;
 814         }
 815
 816         inode_lock_nested(dir, I_MUTEX_PARENT);
 817         upper = lookup_one_len(dentry->d_name.name, upperdir,
 818                                dentry->d_name.len);
 819         err = PTR_ERR(upper);
 820         if (IS_ERR(upper))
 821                 goto out_unlock;
 822
 823         err = -ESTALE;
 824         if ((opaquedir && upper != opaquedir) ||
 825             (!opaquedir && !ovl_matches_upper(dentry, upper)))
 826                 goto out_dput_upper;
 827
 828         if (is_dir)
 829                 err = vfs_rmdir(&init_user_ns, dir, upper);
 830         else
 831                 err = vfs_unlink(&init_user_ns, dir, upper, NULL);
 832         ovl_dir_modified(dentry->d_parent, ovl_type_origin(dentry));
 833
 834         /*
 835          * Keeping this dentry hashed would mean having to release
 836          * upperpath/lowerpath, which could only be done if we are the
 837          * sole user of this dentry.  Too tricky...  Just unhash for
 838          * now.
 839          */
 840         if (!err)
 841                 d_drop(dentry);
 842 out_dput_upper:
 843         dput(upper);
 844 out_unlock:
 845         inode_unlock(dir);
 846         dput(opaquedir);
 847 out:
 848         return err;
 849 }
 850
 851 static bool ovl_pure_upper(struct dentry *dentry)
 852 {
 853         return !ovl_dentry_lower(dentry) &&
 854                !ovl_test_flag(OVL_WHITEOUTS, d_inode(dentry));
 855 }
 856
 857 static void ovl_drop_nlink(struct dentry *dentry)
 858 {
 859         struct inode *inode = d_inode(dentry);
 860         struct dentry *alias;
 861
 862         /* Try to find another, hashed alias */
 863         spin_lock(&inode->i_lock);
 864         hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
 865                 if (alias != dentry && !d_unhashed(alias))
 866                         break;
 867         }
 868         spin_unlock(&inode->i_lock);
 869
 870         /*
 871          * Changes to underlying layers may cause i_nlink to lose sync with
 872          * reality.  In this case prevent the link count from going to zero
 873          * prematurely.
 874          */
 875         if (inode->i_nlink > !!alias)
 876                 drop_nlink(inode);
 877 }
 878
 879 static int ovl_do_remove(struct dentry *dentry, bool is_dir)
 880 {
 881         int err;
 882         const struct cred *old_cred;
 883         struct dentry *upperdentry;
 884         bool lower_positive = ovl_lower_positive(dentry);
 885         LIST_HEAD(list);
 886
 887         /* No need to clean pure upper removed by vfs_rmdir() */
 888         if (is_dir && (lower_positive || !ovl_pure_upper(dentry))) {
 889                 err = ovl_check_empty_dir(dentry, &list);
 890                 if (err)
 891                         goto out;
 892         }
 893
 894         err = ovl_want_write(dentry);
 895         if (err)
 896                 goto out;
 897
 898         err = ovl_copy_up(dentry->d_parent);
 899         if (err)
 900                 goto out_drop_write;
 901
 902         err = ovl_nlink_start(dentry);
 903         if (err)
 904                 goto out_drop_write;
 905
 906         old_cred = ovl_override_creds(dentry->d_sb);
 907         if (!lower_positive)
 908                 err = ovl_remove_upper(dentry, is_dir, &list);
 909         else
 910                 err = ovl_remove_and_whiteout(dentry, &list);
 911         revert_creds(old_cred);
 912         if (!err) {
 913                 if (is_dir)
 914                         clear_nlink(dentry->d_inode);
 915                 else
 916                         ovl_drop_nlink(dentry);
 917         }
 918         ovl_nlink_end(dentry);
 919
 920         /*
 921          * Copy ctime
 922          *
 923          * Note: we fail to update ctime if there was no copy-up, only a
 924          * whiteout
 925          */
 926         upperdentry = ovl_dentry_upper(dentry);
 927         if (upperdentry)
 928                 ovl_copyattr(d_inode(upperdentry), d_inode(dentry));
 929
 930 out_drop_write:
 931         ovl_drop_write(dentry);
 932 out:
 933         ovl_cache_free(&list);
 934         return err;
 935 }
 936
 937 static int ovl_unlink(struct inode *dir, struct dentry *dentry)
 938 {
 939         return ovl_do_remove(dentry, false);
 940 }
 941
 942 static int ovl_rmdir(struct inode *dir, struct dentry *dentry)
 943 {
 944         return ovl_do_remove(dentry, true);
 945 }
 946
 947 static bool ovl_type_merge_or_lower(struct dentry *dentry)
 948 {
 949         enum ovl_path_type type = ovl_path_type(dentry);
 950
 951         return OVL_TYPE_MERGE(type) || !OVL_TYPE_UPPER(type);
 952 }
 953
 954 static bool ovl_can_move(struct dentry *dentry)
 955 {
 956         return ovl_redirect_dir(dentry->d_sb) ||
 957                 !d_is_dir(dentry) || !ovl_type_merge_or_lower(dentry);
 958 }
 959
 960 static char *ovl_get_redirect(struct dentry *dentry, bool abs_redirect)
 961 {
 962         char *buf, *ret;
 963         struct dentry *d, *tmp;
 964         int buflen = ovl_redirect_max + 1;
 965
 966         if (!abs_redirect) {
 967                 ret = kstrndup(dentry->d_name.name, dentry->d_name.len,
 968                                GFP_KERNEL);
 969                 goto out;
 970         }
 971
 972         buf = ret = kmalloc(buflen, GFP_KERNEL);
 973         if (!buf)
 974                 goto out;
 975
 976         buflen--;
 977         buf[buflen] = '\0';
 978         for (d = dget(dentry); !IS_ROOT(d);) {
 979                 const char *name;
 980                 int thislen;
 981
 982                 spin_lock(&d->d_lock);
 983                 name = ovl_dentry_get_redirect(d);
 984                 if (name) {
 985                         thislen = strlen(name);
 986                 } else {
 987                         name = d->d_name.name;
 988                         thislen = d->d_name.len;
 989                 }
 990
 991                 /* If path is too long, fall back to userspace move */
 992                 if (thislen + (name[0] != '/') > buflen) {
 993                         ret = ERR_PTR(-EXDEV);
 994                         spin_unlock(&d->d_lock);
 995                         goto out_put;
 996                 }
 997
 998                 buflen -= thislen;
 999                 memcpy(&buf[buflen], name, thislen);
1000                 spin_unlock(&d->d_lock);
1001                 tmp = dget_parent(d);
1002
1003                 dput(d);
1004                 d = tmp;
1005
1006                 /* Absolute redirect: finished */
1007                 if (buf[buflen] == '/')
1008                         break;
1009                 buflen--;
1010                 buf[buflen] = '/';
1011         }
1012         ret = kstrdup(&buf[buflen], GFP_KERNEL);
1013 out_put:
1014         dput(d);
1015         kfree(buf);
1016 out:
1017         return ret ? ret : ERR_PTR(-ENOMEM);
1018 }
1019
1020 static bool ovl_need_absolute_redirect(struct dentry *dentry, bool samedir)
1021 {
1022         struct dentry *lowerdentry;
1023
1024         if (!samedir)
1025                 return true;
1026
1027         if (d_is_dir(dentry))
1028                 return false;
1029
1030         /*
1031          * For non-dir hardlinked files, we need absolute redirects
1032          * in general as two upper hardlinks could be in different
1033          * dirs. We could put a relative redirect now and convert
1034          * it to absolute redirect later. But when nlink > 1 and
1035          * indexing is on, that means relative redirect needs to be
1036          * converted to absolute during copy up of another lower
1037          * hardllink as well.
1038          *
1039          * So without optimizing too much, just check if lower is
1040          * a hard link or not. If lower is hard link, put absolute
1041          * redirect.
1042          */
1043         lowerdentry = ovl_dentry_lower(dentry);
1044         return (d_inode(lowerdentry)->i_nlink > 1);
1045 }
1046
1047 static int ovl_set_redirect(struct dentry *dentry, bool samedir)
1048 {
1049         int err;
1050         struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
1051         const char *redirect = ovl_dentry_get_redirect(dentry);
1052         bool absolute_redirect = ovl_need_absolute_redirect(dentry, samedir);
1053
1054         if (redirect && (!absolute_redirect || redirect[0] == '/'))
1055                 return 0;
1056
1057         redirect = ovl_get_redirect(dentry, absolute_redirect);
1058         if (IS_ERR(redirect))
1059                 return PTR_ERR(redirect);
1060
1061         err = ovl_check_setxattr(ofs, ovl_dentry_upper(dentry),
1062                                  OVL_XATTR_REDIRECT,
1063                                  redirect, strlen(redirect), -EXDEV);
1064         if (!err) {
1065                 spin_lock(&dentry->d_lock);
1066                 ovl_dentry_set_redirect(dentry, redirect);
1067                 spin_unlock(&dentry->d_lock);
1068         } else {
1069                 kfree(redirect);
1070                 pr_warn_ratelimited("failed to set redirect (%i)\n",
1071                                     err);
1072                 /* Fall back to userspace copy-up */
1073                 err = -EXDEV;
1074         }
1075         return err;
1076 }
1077
1078 static int ovl_rename(struct user_namespace *mnt_userns, struct inode *olddir,
1079                       struct dentry *old, struct inode *newdir,
1080                       struct dentry *new, unsigned int flags)
1081 {
1082         int err;
1083         struct dentry *old_upperdir;
1084         struct dentry *new_upperdir;
1085         struct dentry *olddentry;
1086         struct dentry *newdentry;
1087         struct dentry *trap;
1088         bool old_opaque;
1089         bool new_opaque;
1090         bool cleanup_whiteout = false;
1091         bool update_nlink = false;
1092         bool overwrite = !(flags & RENAME_EXCHANGE);
1093         bool is_dir = d_is_dir(old);
1094         bool new_is_dir = d_is_dir(new);
1095         bool samedir = olddir == newdir;
1096         struct dentry *opaquedir = NULL;
1097         const struct cred *old_cred = NULL;
1098         LIST_HEAD(list);
1099
1100         err = -EINVAL;
1101         if (flags & ~(RENAME_EXCHANGE | RENAME_NOREPLACE))
1102                 goto out;
1103
1104         flags &= ~RENAME_NOREPLACE;
1105
1106         /* Don't copy up directory trees */
1107         err = -EXDEV;
1108         if (!ovl_can_move(old))
1109                 goto out;
1110         if (!overwrite && !ovl_can_move(new))
1111                 goto out;
1112
1113         if (overwrite && new_is_dir && !ovl_pure_upper(new)) {
1114                 err = ovl_check_empty_dir(new, &list);
1115                 if (err)
1116                         goto out;
1117         }
1118
1119         if (overwrite) {
1120                 if (ovl_lower_positive(old)) {
1121                         if (!ovl_dentry_is_whiteout(new)) {
1122                                 /* Whiteout source */
1123                                 flags |= RENAME_WHITEOUT;
1124                         } else {
1125                                 /* Switch whiteouts */
1126                                 flags |= RENAME_EXCHANGE;
1127                         }
1128                 } else if (is_dir && ovl_dentry_is_whiteout(new)) {
1129                         flags |= RENAME_EXCHANGE;
1130                         cleanup_whiteout = true;
1131                 }
1132         }
1133
1134         err = ovl_want_write(old);
1135         if (err)
1136                 goto out;
1137
1138         err = ovl_copy_up(old);
1139         if (err)
1140                 goto out_drop_write;
1141
1142         err = ovl_copy_up(new->d_parent);
1143         if (err)
1144                 goto out_drop_write;
1145         if (!overwrite) {
1146                 err = ovl_copy_up(new);
1147                 if (err)
1148                         goto out_drop_write;
1149         } else if (d_inode(new)) {
1150                 err = ovl_nlink_start(new);
1151                 if (err)
1152                         goto out_drop_write;
1153
1154                 update_nlink = true;
1155         }
1156
1157         old_cred = ovl_override_creds(old->d_sb);
1158
1159         if (!list_empty(&list)) {
1160                 opaquedir = ovl_clear_empty(new, &list);
1161                 err = PTR_ERR(opaquedir);
1162                 if (IS_ERR(opaquedir)) {
1163                         opaquedir = NULL;
1164                         goto out_revert_creds;
1165                 }
1166         }
1167
1168         old_upperdir = ovl_dentry_upper(old->d_parent);
1169         new_upperdir = ovl_dentry_upper(new->d_parent);
1170
1171         if (!samedir) {
1172                 /*
1173                  * When moving a merge dir or non-dir with copy up origin into
1174                  * a new parent, we are marking the new parent dir "impure".
1175                  * When ovl_iterate() iterates an "impure" upper dir, it will
1176                  * lookup the origin inodes of the entries to fill d_ino.
1177                  */
1178                 if (ovl_type_origin(old)) {
1179                         err = ovl_set_impure(new->d_parent, new_upperdir);
1180                         if (err)
1181                                 goto out_revert_creds;
1182                 }
1183                 if (!overwrite && ovl_type_origin(new)) {
1184                         err = ovl_set_impure(old->d_parent, old_upperdir);
1185                         if (err)
1186                                 goto out_revert_creds;
1187                 }
1188         }
1189
1190         trap = lock_rename(new_upperdir, old_upperdir);
1191
1192         olddentry = lookup_one_len(old->d_name.name, old_upperdir,
1193                                    old->d_name.len);
1194         err = PTR_ERR(olddentry);
1195         if (IS_ERR(olddentry))
1196                 goto out_unlock;
1197
1198         err = -ESTALE;
1199         if (!ovl_matches_upper(old, olddentry))
1200                 goto out_dput_old;
1201
1202         newdentry = lookup_one_len(new->d_name.name, new_upperdir,
1203                                    new->d_name.len);
1204         err = PTR_ERR(newdentry);
1205         if (IS_ERR(newdentry))
1206                 goto out_dput_old;
1207
1208         old_opaque = ovl_dentry_is_opaque(old);
1209         new_opaque = ovl_dentry_is_opaque(new);
1210
1211         err = -ESTALE;
1212         if (d_inode(new) && ovl_dentry_upper(new)) {
1213                 if (opaquedir) {
1214                         if (newdentry != opaquedir)
1215                                 goto out_dput;
1216                 } else {
1217                         if (!ovl_matches_upper(new, newdentry))
1218                                 goto out_dput;
1219                 }
1220         } else {
1221                 if (!d_is_negative(newdentry)) {
1222                         if (!new_opaque || !ovl_is_whiteout(newdentry))
1223                                 goto out_dput;
1224                 } else {
1225                         if (flags & RENAME_EXCHANGE)
1226                                 goto out_dput;
1227                 }
1228         }
1229
1230         if (olddentry == trap)
1231                 goto out_dput;
1232         if (newdentry == trap)
1233                 goto out_dput;
1234
1235         if (olddentry->d_inode == newdentry->d_inode)
1236                 goto out_dput;
1237
1238         err = 0;
1239         if (ovl_type_merge_or_lower(old))
1240                 err = ovl_set_redirect(old, samedir);
1241         else if (is_dir && !old_opaque && ovl_type_merge(new->d_parent))
1242                 err = ovl_set_opaque_xerr(old, olddentry, -EXDEV);
1243         if (err)
1244                 goto out_dput;
1245
1246         if (!overwrite && ovl_type_merge_or_lower(new))
1247                 err = ovl_set_redirect(new, samedir);
1248         else if (!overwrite && new_is_dir && !new_opaque &&
1249                  ovl_type_merge(old->d_parent))
1250                 err = ovl_set_opaque_xerr(new, newdentry, -EXDEV);
1251         if (err)
1252                 goto out_dput;
1253
1254         err = ovl_do_rename(old_upperdir->d_inode, olddentry,
1255                             new_upperdir->d_inode, newdentry, flags);
1256         if (err)
1257                 goto out_dput;
1258
1259         if (cleanup_whiteout)
1260                 ovl_cleanup(old_upperdir->d_inode, newdentry);
1261
1262         if (overwrite && d_inode(new)) {
1263                 if (new_is_dir)
1264                         clear_nlink(d_inode(new));
1265                 else
1266                         ovl_drop_nlink(new);
1267         }
1268
1269         ovl_dir_modified(old->d_parent, ovl_type_origin(old) ||
1270                          (!overwrite && ovl_type_origin(new)));
1271         ovl_dir_modified(new->d_parent, ovl_type_origin(old) ||
1272                          (d_inode(new) && ovl_type_origin(new)));
1273
1274         /* copy ctime: */
1275         ovl_copyattr(d_inode(olddentry), d_inode(old));
1276         if (d_inode(new) && ovl_dentry_upper(new))
1277                 ovl_copyattr(d_inode(newdentry), d_inode(new));
1278
1279 out_dput:
1280         dput(newdentry);
1281 out_dput_old:
1282         dput(olddentry);
1283 out_unlock:
1284         unlock_rename(new_upperdir, old_upperdir);
1285 out_revert_creds:
1286         revert_creds(old_cred);
1287         if (update_nlink)
1288                 ovl_nlink_end(new);
1289 out_drop_write:
1290         ovl_drop_write(old);
1291 out:
1292         dput(opaquedir);
1293         ovl_cache_free(&list);
1294         return err;
1295 }
1296
1297 const struct inode_operations ovl_dir_inode_operations = {
1298         .lookup         = ovl_lookup,
1299         .mkdir          = ovl_mkdir,
1300         .symlink        = ovl_symlink,
1301         .unlink         = ovl_unlink,
1302         .rmdir          = ovl_rmdir,
1303         .rename         = ovl_rename,
1304         .link           = ovl_link,
1305         .setattr        = ovl_setattr,
1306         .create         = ovl_create,
1307         .mknod          = ovl_mknod,
1308         .permission     = ovl_permission,
1309         .getattr        = ovl_getattr,
1310         .listxattr      = ovl_listxattr,
1311         .get_acl        = ovl_get_acl,
1312         .update_time    = ovl_update_time,
1313         .fileattr_get   = ovl_fileattr_get,
1314         .fileattr_set   = ovl_fileattr_set,
1315 };