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 ovl_fs *ofs, 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(ofs, wdir, wdentry);
  33         else
  34                 err = ovl_do_unlink(ofs, 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 ovl_fs *ofs, 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 = ovl_lookup_upper(ofs, 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(ofs, workdir);
  74                 if (IS_ERR(whiteout))
  75                         goto out;
  76
  77                 err = ovl_do_whiteout(ofs, 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(ofs, workdir);
  88                 if (IS_ERR(whiteout))
  89                         goto out;
  90
  91                 err = ovl_do_link(ofs, 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(ofs, wdir, whiteout, dir, dentry, flags);
 126         if (err)
 127                 goto kill_whiteout;
 128         if (flags)
 129                 ovl_cleanup(ofs, wdir, dentry);
 130
 131 out:
 132         dput(whiteout);
 133         return err;
 134
 135 kill_whiteout:
 136         ovl_cleanup(ofs, wdir, whiteout);
 137         goto out;
 138 }
 139
 140 int ovl_mkdir_real(struct ovl_fs *ofs, struct inode *dir,
 141                    struct dentry **newdentry, umode_t mode)
 142 {
 143         int err;
 144         struct dentry *d, *dentry = *newdentry;
 145
 146         err = ovl_do_mkdir(ofs, dir, dentry, mode);
 147         if (err)
 148                 return err;
 149
 150         if (likely(!d_unhashed(dentry)))
 151                 return 0;
 152
 153         /*
 154          * vfs_mkdir() may succeed and leave the dentry passed
 155          * to it unhashed and negative. If that happens, try to
 156          * lookup a new hashed and positive dentry.
 157          */
 158         d = ovl_lookup_upper(ofs, dentry->d_name.name, dentry->d_parent,
 159                              dentry->d_name.len);
 160         if (IS_ERR(d)) {
 161                 pr_warn("failed lookup after mkdir (%pd2, err=%i).\n",
 162                         dentry, err);
 163                 return PTR_ERR(d);
 164         }
 165         dput(dentry);
 166         *newdentry = d;
 167
 168         return 0;
 169 }
 170
 171 struct dentry *ovl_create_real(struct ovl_fs *ofs, struct inode *dir,
 172                                struct dentry *newdentry, struct ovl_cattr *attr)
 173 {
 174         int err;
 175
 176         if (IS_ERR(newdentry))
 177                 return newdentry;
 178
 179         err = -ESTALE;
 180         if (newdentry->d_inode)
 181                 goto out;
 182
 183         if (attr->hardlink) {
 184                 err = ovl_do_link(ofs, attr->hardlink, dir, newdentry);
 185         } else {
 186                 switch (attr->mode & S_IFMT) {
 187                 case S_IFREG:
 188                         err = ovl_do_create(ofs, dir, newdentry, attr->mode);
 189                         break;
 190
 191                 case S_IFDIR:
 192                         /* mkdir is special... */
 193                         err =  ovl_mkdir_real(ofs, dir, &newdentry, attr->mode);
 194                         break;
 195
 196                 case S_IFCHR:
 197                 case S_IFBLK:
 198                 case S_IFIFO:
 199                 case S_IFSOCK:
 200                         err = ovl_do_mknod(ofs, dir, newdentry, attr->mode,
 201                                            attr->rdev);
 202                         break;
 203
 204                 case S_IFLNK:
 205                         err = ovl_do_symlink(ofs, dir, newdentry, attr->link);
 206                         break;
 207
 208                 default:
 209                         err = -EPERM;
 210                 }
 211         }
 212         if (!err && WARN_ON(!newdentry->d_inode)) {
 213                 /*
 214                  * Not quite sure if non-instantiated dentry is legal or not.
 215                  * VFS doesn't seem to care so check and warn here.
 216                  */
 217                 err = -EIO;
 218         }
 219 out:
 220         if (err) {
 221                 dput(newdentry);
 222                 return ERR_PTR(err);
 223         }
 224         return newdentry;
 225 }
 226
 227 struct dentry *ovl_create_temp(struct ovl_fs *ofs, struct dentry *workdir,
 228                                struct ovl_cattr *attr)
 229 {
 230         return ovl_create_real(ofs, d_inode(workdir),
 231                                ovl_lookup_temp(ofs, workdir), attr);
 232 }
 233
 234 static int ovl_set_opaque_xerr(struct dentry *dentry, struct dentry *upper,
 235                                int xerr)
 236 {
 237         struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
 238         int err;
 239
 240         err = ovl_check_setxattr(ofs, upper, OVL_XATTR_OPAQUE, "y", 1, xerr);
 241         if (!err)
 242                 ovl_dentry_set_opaque(dentry);
 243
 244         return err;
 245 }
 246
 247 static int ovl_set_opaque(struct dentry *dentry, struct dentry *upperdentry)
 248 {
 249         /*
 250          * Fail with -EIO when trying to create opaque dir and upper doesn't
 251          * support xattrs. ovl_rename() calls ovl_set_opaque_xerr(-EXDEV) to
 252          * return a specific error for noxattr case.
 253          */
 254         return ovl_set_opaque_xerr(dentry, upperdentry, -EIO);
 255 }
 256
 257 /*
 258  * Common operations required to be done after creation of file on upper.
 259  * If @hardlink is false, then @inode is a pre-allocated inode, we may or
 260  * may not use to instantiate the new dentry.
 261  */
 262 static int ovl_instantiate(struct dentry *dentry, struct inode *inode,
 263                            struct dentry *newdentry, bool hardlink)
 264 {
 265         struct ovl_inode_params oip = {
 266                 .upperdentry = newdentry,
 267                 .newinode = inode,
 268         };
 269
 270         ovl_dir_modified(dentry->d_parent, false);
 271         ovl_dentry_set_upper_alias(dentry);
 272         ovl_dentry_update_reval(dentry, newdentry,
 273                         DCACHE_OP_REVALIDATE | DCACHE_OP_WEAK_REVALIDATE);
 274
 275         if (!hardlink) {
 276                 /*
 277                  * ovl_obtain_alias() can be called after ovl_create_real()
 278                  * and before we get here, so we may get an inode from cache
 279                  * with the same real upperdentry that is not the inode we
 280                  * pre-allocated.  In this case we will use the cached inode
 281                  * to instantiate the new dentry.
 282                  *
 283                  * XXX: if we ever use ovl_obtain_alias() to decode directory
 284                  * file handles, need to use ovl_get_inode_locked() and
 285                  * d_instantiate_new() here to prevent from creating two
 286                  * hashed directory inode aliases.
 287                  */
 288                 inode = ovl_get_inode(dentry->d_sb, &oip);
 289                 if (IS_ERR(inode))
 290                         return PTR_ERR(inode);
 291                 if (inode == oip.newinode)
 292                         ovl_set_flag(OVL_UPPERDATA, inode);
 293         } else {
 294                 WARN_ON(ovl_inode_real(inode) != d_inode(newdentry));
 295                 dput(newdentry);
 296                 inc_nlink(inode);
 297         }
 298
 299         d_instantiate(dentry, inode);
 300         if (inode != oip.newinode) {
 301                 pr_warn_ratelimited("newly created inode found in cache (%pd2)\n",
 302                                     dentry);
 303         }
 304
 305         /* Force lookup of new upper hardlink to find its lower */
 306         if (hardlink)
 307                 d_drop(dentry);
 308
 309         return 0;
 310 }
 311
 312 static bool ovl_type_merge(struct dentry *dentry)
 313 {
 314         return OVL_TYPE_MERGE(ovl_path_type(dentry));
 315 }
 316
 317 static bool ovl_type_origin(struct dentry *dentry)
 318 {
 319         return OVL_TYPE_ORIGIN(ovl_path_type(dentry));
 320 }
 321
 322 static int ovl_create_upper(struct dentry *dentry, struct inode *inode,
 323                             struct ovl_cattr *attr)
 324 {
 325         struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
 326         struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
 327         struct inode *udir = upperdir->d_inode;
 328         struct dentry *newdentry;
 329         int err;
 330
 331         if (!attr->hardlink && !IS_POSIXACL(udir))
 332                 attr->mode &= ~current_umask();
 333
 334         inode_lock_nested(udir, I_MUTEX_PARENT);
 335         newdentry = ovl_create_real(ofs, udir,
 336                                     ovl_lookup_upper(ofs, dentry->d_name.name,
 337                                                      upperdir, dentry->d_name.len),
 338                                     attr);
 339         err = PTR_ERR(newdentry);
 340         if (IS_ERR(newdentry))
 341                 goto out_unlock;
 342
 343         if (ovl_type_merge(dentry->d_parent) && d_is_dir(newdentry) &&
 344             !ovl_allow_offline_changes(ofs)) {
 345                 /* Setting opaque here is just an optimization, allow to fail */
 346                 ovl_set_opaque(dentry, newdentry);
 347         }
 348
 349         err = ovl_instantiate(dentry, inode, newdentry, !!attr->hardlink);
 350         if (err)
 351                 goto out_cleanup;
 352 out_unlock:
 353         inode_unlock(udir);
 354         return err;
 355
 356 out_cleanup:
 357         ovl_cleanup(ofs, udir, newdentry);
 358         dput(newdentry);
 359         goto out_unlock;
 360 }
 361
 362 static struct dentry *ovl_clear_empty(struct dentry *dentry,
 363                                       struct list_head *list)
 364 {
 365         struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
 366         struct dentry *workdir = ovl_workdir(dentry);
 367         struct inode *wdir = workdir->d_inode;
 368         struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
 369         struct inode *udir = upperdir->d_inode;
 370         struct path upperpath;
 371         struct dentry *upper;
 372         struct dentry *opaquedir;
 373         struct kstat stat;
 374         int err;
 375
 376         if (WARN_ON(!workdir))
 377                 return ERR_PTR(-EROFS);
 378
 379         err = ovl_lock_rename_workdir(workdir, upperdir);
 380         if (err)
 381                 goto out;
 382
 383         ovl_path_upper(dentry, &upperpath);
 384         err = vfs_getattr(&upperpath, &stat,
 385                           STATX_BASIC_STATS, AT_STATX_SYNC_AS_STAT);
 386         if (err)
 387                 goto out_unlock;
 388
 389         err = -ESTALE;
 390         if (!S_ISDIR(stat.mode))
 391                 goto out_unlock;
 392         upper = upperpath.dentry;
 393         if (upper->d_parent->d_inode != udir)
 394                 goto out_unlock;
 395
 396         opaquedir = ovl_create_temp(ofs, workdir, OVL_CATTR(stat.mode));
 397         err = PTR_ERR(opaquedir);
 398         if (IS_ERR(opaquedir))
 399                 goto out_unlock;
 400
 401         err = ovl_copy_xattr(dentry->d_sb, &upperpath, opaquedir);
 402         if (err)
 403                 goto out_cleanup;
 404
 405         err = ovl_set_opaque(dentry, opaquedir);
 406         if (err)
 407                 goto out_cleanup;
 408
 409         inode_lock(opaquedir->d_inode);
 410         err = ovl_set_attr(ofs, opaquedir, &stat);
 411         inode_unlock(opaquedir->d_inode);
 412         if (err)
 413                 goto out_cleanup;
 414
 415         err = ovl_do_rename(ofs, wdir, opaquedir, udir, upper, RENAME_EXCHANGE);
 416         if (err)
 417                 goto out_cleanup;
 418
 419         ovl_cleanup_whiteouts(ofs, upper, list);
 420         ovl_cleanup(ofs, wdir, upper);
 421         unlock_rename(workdir, upperdir);
 422
 423         /* dentry's upper doesn't match now, get rid of it */
 424         d_drop(dentry);
 425
 426         return opaquedir;
 427
 428 out_cleanup:
 429         ovl_cleanup(ofs, wdir, opaquedir);
 430         dput(opaquedir);
 431 out_unlock:
 432         unlock_rename(workdir, upperdir);
 433 out:
 434         return ERR_PTR(err);
 435 }
 436
 437 static int ovl_set_upper_acl(struct ovl_fs *ofs, struct dentry *upperdentry,
 438                              const char *name, const struct posix_acl *acl)
 439 {
 440         void *buffer;
 441         size_t size;
 442         int err;
 443
 444         if (!IS_ENABLED(CONFIG_FS_POSIX_ACL) || !acl)
 445                 return 0;
 446
 447         size = posix_acl_xattr_size(acl->a_count);
 448         buffer = kmalloc(size, GFP_KERNEL);
 449         if (!buffer)
 450                 return -ENOMEM;
 451
 452         err = posix_acl_to_xattr(&init_user_ns, acl, buffer, size);
 453         if (err < 0)
 454                 goto out_free;
 455
 456         err = ovl_do_setxattr(ofs, upperdentry, name, buffer, size, XATTR_CREATE);
 457 out_free:
 458         kfree(buffer);
 459         return err;
 460 }
 461
 462 static int ovl_create_over_whiteout(struct dentry *dentry, struct inode *inode,
 463                                     struct ovl_cattr *cattr)
 464 {
 465         struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
 466         struct dentry *workdir = ovl_workdir(dentry);
 467         struct inode *wdir = workdir->d_inode;
 468         struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
 469         struct inode *udir = upperdir->d_inode;
 470         struct dentry *upper;
 471         struct dentry *newdentry;
 472         int err;
 473         struct posix_acl *acl, *default_acl;
 474         bool hardlink = !!cattr->hardlink;
 475
 476         if (WARN_ON(!workdir))
 477                 return -EROFS;
 478
 479         if (!hardlink) {
 480                 err = posix_acl_create(dentry->d_parent->d_inode,
 481                                        &cattr->mode, &default_acl, &acl);
 482                 if (err)
 483                         return err;
 484         }
 485
 486         err = ovl_lock_rename_workdir(workdir, upperdir);
 487         if (err)
 488                 goto out;
 489
 490         upper = ovl_lookup_upper(ofs, dentry->d_name.name, upperdir,
 491                                  dentry->d_name.len);
 492         err = PTR_ERR(upper);
 493         if (IS_ERR(upper))
 494                 goto out_unlock;
 495
 496         err = -ESTALE;
 497         if (d_is_negative(upper) || !IS_WHITEOUT(d_inode(upper)))
 498                 goto out_dput;
 499
 500         newdentry = ovl_create_temp(ofs, workdir, cattr);
 501         err = PTR_ERR(newdentry);
 502         if (IS_ERR(newdentry))
 503                 goto out_dput;
 504
 505         /*
 506          * mode could have been mutilated due to umask (e.g. sgid directory)
 507          */
 508         if (!hardlink &&
 509             !S_ISLNK(cattr->mode) &&
 510             newdentry->d_inode->i_mode != cattr->mode) {
 511                 struct iattr attr = {
 512                         .ia_valid = ATTR_MODE,
 513                         .ia_mode = cattr->mode,
 514                 };
 515                 inode_lock(newdentry->d_inode);
 516                 err = ovl_do_notify_change(ofs, newdentry, &attr);
 517                 inode_unlock(newdentry->d_inode);
 518                 if (err)
 519                         goto out_cleanup;
 520         }
 521         if (!hardlink) {
 522                 err = ovl_set_upper_acl(ofs, newdentry,
 523                                         XATTR_NAME_POSIX_ACL_ACCESS, acl);
 524                 if (err)
 525                         goto out_cleanup;
 526
 527                 err = ovl_set_upper_acl(ofs, newdentry,
 528                                         XATTR_NAME_POSIX_ACL_DEFAULT, default_acl);
 529                 if (err)
 530                         goto out_cleanup;
 531         }
 532
 533         if (!hardlink && S_ISDIR(cattr->mode)) {
 534                 err = ovl_set_opaque(dentry, newdentry);
 535                 if (err)
 536                         goto out_cleanup;
 537
 538                 err = ovl_do_rename(ofs, wdir, newdentry, udir, upper,
 539                                     RENAME_EXCHANGE);
 540                 if (err)
 541                         goto out_cleanup;
 542
 543                 ovl_cleanup(ofs, wdir, upper);
 544         } else {
 545                 err = ovl_do_rename(ofs, wdir, newdentry, udir, upper, 0);
 546                 if (err)
 547                         goto out_cleanup;
 548         }
 549         err = ovl_instantiate(dentry, inode, newdentry, hardlink);
 550         if (err) {
 551                 ovl_cleanup(ofs, udir, newdentry);
 552                 dput(newdentry);
 553         }
 554 out_dput:
 555         dput(upper);
 556 out_unlock:
 557         unlock_rename(workdir, upperdir);
 558 out:
 559         if (!hardlink) {
 560                 posix_acl_release(acl);
 561                 posix_acl_release(default_acl);
 562         }
 563         return err;
 564
 565 out_cleanup:
 566         ovl_cleanup(ofs, wdir, newdentry);
 567         dput(newdentry);
 568         goto out_dput;
 569 }
 570
 571 static int ovl_create_or_link(struct dentry *dentry, struct inode *inode,
 572                               struct ovl_cattr *attr, bool origin)
 573 {
 574         int err;
 575         const struct cred *old_cred;
 576         struct cred *override_cred;
 577         struct dentry *parent = dentry->d_parent;
 578
 579         err = ovl_copy_up(parent);
 580         if (err)
 581                 return err;
 582
 583         old_cred = ovl_override_creds(dentry->d_sb);
 584
 585         /*
 586          * When linking a file with copy up origin into a new parent, mark the
 587          * new parent dir "impure".
 588          */
 589         if (origin) {
 590                 err = ovl_set_impure(parent, ovl_dentry_upper(parent));
 591                 if (err)
 592                         goto out_revert_creds;
 593         }
 594
 595         err = -ENOMEM;
 596         override_cred = prepare_creds();
 597         if (override_cred) {
 598                 override_cred->fsuid = inode->i_uid;
 599                 override_cred->fsgid = inode->i_gid;
 600                 if (!attr->hardlink) {
 601                         err = security_dentry_create_files_as(dentry,
 602                                         attr->mode, &dentry->d_name, old_cred,
 603                                         override_cred);
 604                         if (err) {
 605                                 put_cred(override_cred);
 606                                 goto out_revert_creds;
 607                         }
 608                 }
 609                 put_cred(override_creds(override_cred));
 610                 put_cred(override_cred);
 611
 612                 if (!ovl_dentry_is_whiteout(dentry))
 613                         err = ovl_create_upper(dentry, inode, attr);
 614                 else
 615                         err = ovl_create_over_whiteout(dentry, inode, attr);
 616         }
 617 out_revert_creds:
 618         revert_creds(old_cred);
 619         return err;
 620 }
 621
 622 static int ovl_create_object(struct dentry *dentry, int mode, dev_t rdev,
 623                              const char *link)
 624 {
 625         int err;
 626         struct inode *inode;
 627         struct ovl_cattr attr = {
 628                 .rdev = rdev,
 629                 .link = link,
 630         };
 631
 632         err = ovl_want_write(dentry);
 633         if (err)
 634                 goto out;
 635
 636         /* Preallocate inode to be used by ovl_get_inode() */
 637         err = -ENOMEM;
 638         inode = ovl_new_inode(dentry->d_sb, mode, rdev);
 639         if (!inode)
 640                 goto out_drop_write;
 641
 642         spin_lock(&inode->i_lock);
 643         inode->i_state |= I_CREATING;
 644         spin_unlock(&inode->i_lock);
 645
 646         inode_init_owner(&init_user_ns, inode, dentry->d_parent->d_inode, mode);
 647         attr.mode = inode->i_mode;
 648
 649         err = ovl_create_or_link(dentry, inode, &attr, false);
 650         /* Did we end up using the preallocated inode? */
 651         if (inode != d_inode(dentry))
 652                 iput(inode);
 653
 654 out_drop_write:
 655         ovl_drop_write(dentry);
 656 out:
 657         return err;
 658 }
 659
 660 static int ovl_create(struct user_namespace *mnt_userns, struct inode *dir,
 661                       struct dentry *dentry, umode_t mode, bool excl)
 662 {
 663         return ovl_create_object(dentry, (mode & 07777) | S_IFREG, 0, NULL);
 664 }
 665
 666 static int ovl_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
 667                      struct dentry *dentry, umode_t mode)
 668 {
 669         return ovl_create_object(dentry, (mode & 07777) | S_IFDIR, 0, NULL);
 670 }
 671
 672 static int ovl_mknod(struct user_namespace *mnt_userns, struct inode *dir,
 673                      struct dentry *dentry, umode_t mode, dev_t rdev)
 674 {
 675         /* Don't allow creation of "whiteout" on overlay */
 676         if (S_ISCHR(mode) && rdev == WHITEOUT_DEV)
 677                 return -EPERM;
 678
 679         return ovl_create_object(dentry, mode, rdev, NULL);
 680 }
 681
 682 static int ovl_symlink(struct user_namespace *mnt_userns, struct inode *dir,
 683                        struct dentry *dentry, const char *link)
 684 {
 685         return ovl_create_object(dentry, S_IFLNK, 0, link);
 686 }
 687
 688 static int ovl_set_link_redirect(struct dentry *dentry)
 689 {
 690         const struct cred *old_cred;
 691         int err;
 692
 693         old_cred = ovl_override_creds(dentry->d_sb);
 694         err = ovl_set_redirect(dentry, false);
 695         revert_creds(old_cred);
 696
 697         return err;
 698 }
 699
 700 static int ovl_link(struct dentry *old, struct inode *newdir,
 701                     struct dentry *new)
 702 {
 703         int err;
 704         struct inode *inode;
 705
 706         err = ovl_want_write(old);
 707         if (err)
 708                 goto out;
 709
 710         err = ovl_copy_up(old);
 711         if (err)
 712                 goto out_drop_write;
 713
 714         err = ovl_copy_up(new->d_parent);
 715         if (err)
 716                 goto out_drop_write;
 717
 718         if (ovl_is_metacopy_dentry(old)) {
 719                 err = ovl_set_link_redirect(old);
 720                 if (err)
 721                         goto out_drop_write;
 722         }
 723
 724         err = ovl_nlink_start(old);
 725         if (err)
 726                 goto out_drop_write;
 727
 728         inode = d_inode(old);
 729         ihold(inode);
 730
 731         err = ovl_create_or_link(new, inode,
 732                         &(struct ovl_cattr) {.hardlink = ovl_dentry_upper(old)},
 733                         ovl_type_origin(old));
 734         if (err)
 735                 iput(inode);
 736
 737         ovl_nlink_end(old);
 738 out_drop_write:
 739         ovl_drop_write(old);
 740 out:
 741         return err;
 742 }
 743
 744 static bool ovl_matches_upper(struct dentry *dentry, struct dentry *upper)
 745 {
 746         return d_inode(ovl_dentry_upper(dentry)) == d_inode(upper);
 747 }
 748
 749 static int ovl_remove_and_whiteout(struct dentry *dentry,
 750                                    struct list_head *list)
 751 {
 752         struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
 753         struct dentry *workdir = ovl_workdir(dentry);
 754         struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
 755         struct dentry *upper;
 756         struct dentry *opaquedir = NULL;
 757         int err;
 758
 759         if (WARN_ON(!workdir))
 760                 return -EROFS;
 761
 762         if (!list_empty(list)) {
 763                 opaquedir = ovl_clear_empty(dentry, list);
 764                 err = PTR_ERR(opaquedir);
 765                 if (IS_ERR(opaquedir))
 766                         goto out;
 767         }
 768
 769         err = ovl_lock_rename_workdir(workdir, upperdir);
 770         if (err)
 771                 goto out_dput;
 772
 773         upper = ovl_lookup_upper(ofs, dentry->d_name.name, upperdir,
 774                                  dentry->d_name.len);
 775         err = PTR_ERR(upper);
 776         if (IS_ERR(upper))
 777                 goto out_unlock;
 778
 779         err = -ESTALE;
 780         if ((opaquedir && upper != opaquedir) ||
 781             (!opaquedir && ovl_dentry_upper(dentry) &&
 782              !ovl_matches_upper(dentry, upper))) {
 783                 goto out_dput_upper;
 784         }
 785
 786         err = ovl_cleanup_and_whiteout(ofs, d_inode(upperdir), upper);
 787         if (err)
 788                 goto out_d_drop;
 789
 790         ovl_dir_modified(dentry->d_parent, true);
 791 out_d_drop:
 792         d_drop(dentry);
 793 out_dput_upper:
 794         dput(upper);
 795 out_unlock:
 796         unlock_rename(workdir, upperdir);
 797 out_dput:
 798         dput(opaquedir);
 799 out:
 800         return err;
 801 }
 802
 803 static int ovl_remove_upper(struct dentry *dentry, bool is_dir,
 804                             struct list_head *list)
 805 {
 806         struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
 807         struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
 808         struct inode *dir = upperdir->d_inode;
 809         struct dentry *upper;
 810         struct dentry *opaquedir = NULL;
 811         int err;
 812
 813         if (!list_empty(list)) {
 814                 opaquedir = ovl_clear_empty(dentry, list);
 815                 err = PTR_ERR(opaquedir);
 816                 if (IS_ERR(opaquedir))
 817                         goto out;
 818         }
 819
 820         inode_lock_nested(dir, I_MUTEX_PARENT);
 821         upper = ovl_lookup_upper(ofs, dentry->d_name.name, upperdir,
 822                                  dentry->d_name.len);
 823         err = PTR_ERR(upper);
 824         if (IS_ERR(upper))
 825                 goto out_unlock;
 826
 827         err = -ESTALE;
 828         if ((opaquedir && upper != opaquedir) ||
 829             (!opaquedir && !ovl_matches_upper(dentry, upper)))
 830                 goto out_dput_upper;
 831
 832         if (is_dir)
 833                 err = ovl_do_rmdir(ofs, dir, upper);
 834         else
 835                 err = ovl_do_unlink(ofs, dir, upper);
 836         ovl_dir_modified(dentry->d_parent, ovl_type_origin(dentry));
 837
 838         /*
 839          * Keeping this dentry hashed would mean having to release
 840          * upperpath/lowerpath, which could only be done if we are the
 841          * sole user of this dentry.  Too tricky...  Just unhash for
 842          * now.
 843          */
 844         if (!err)
 845                 d_drop(dentry);
 846 out_dput_upper:
 847         dput(upper);
 848 out_unlock:
 849         inode_unlock(dir);
 850         dput(opaquedir);
 851 out:
 852         return err;
 853 }
 854
 855 static bool ovl_pure_upper(struct dentry *dentry)
 856 {
 857         return !ovl_dentry_lower(dentry) &&
 858                !ovl_test_flag(OVL_WHITEOUTS, d_inode(dentry));
 859 }
 860
 861 static void ovl_drop_nlink(struct dentry *dentry)
 862 {
 863         struct inode *inode = d_inode(dentry);
 864         struct dentry *alias;
 865
 866         /* Try to find another, hashed alias */
 867         spin_lock(&inode->i_lock);
 868         hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
 869                 if (alias != dentry && !d_unhashed(alias))
 870                         break;
 871         }
 872         spin_unlock(&inode->i_lock);
 873
 874         /*
 875          * Changes to underlying layers may cause i_nlink to lose sync with
 876          * reality.  In this case prevent the link count from going to zero
 877          * prematurely.
 878          */
 879         if (inode->i_nlink > !!alias)
 880                 drop_nlink(inode);
 881 }
 882
 883 static int ovl_do_remove(struct dentry *dentry, bool is_dir)
 884 {
 885         int err;
 886         const struct cred *old_cred;
 887         bool lower_positive = ovl_lower_positive(dentry);
 888         LIST_HEAD(list);
 889
 890         /* No need to clean pure upper removed by vfs_rmdir() */
 891         if (is_dir && (lower_positive || !ovl_pure_upper(dentry))) {
 892                 err = ovl_check_empty_dir(dentry, &list);
 893                 if (err)
 894                         goto out;
 895         }
 896
 897         err = ovl_want_write(dentry);
 898         if (err)
 899                 goto out;
 900
 901         err = ovl_copy_up(dentry->d_parent);
 902         if (err)
 903                 goto out_drop_write;
 904
 905         err = ovl_nlink_start(dentry);
 906         if (err)
 907                 goto out_drop_write;
 908
 909         old_cred = ovl_override_creds(dentry->d_sb);
 910         if (!lower_positive)
 911                 err = ovl_remove_upper(dentry, is_dir, &list);
 912         else
 913                 err = ovl_remove_and_whiteout(dentry, &list);
 914         revert_creds(old_cred);
 915         if (!err) {
 916                 if (is_dir)
 917                         clear_nlink(dentry->d_inode);
 918                 else
 919                         ovl_drop_nlink(dentry);
 920         }
 921         ovl_nlink_end(dentry);
 922
 923         /*
 924          * Copy ctime
 925          *
 926          * Note: we fail to update ctime if there was no copy-up, only a
 927          * whiteout
 928          */
 929         if (ovl_dentry_upper(dentry))
 930                 ovl_copyattr(d_inode(dentry));
 931
 932 out_drop_write:
 933         ovl_drop_write(dentry);
 934 out:
 935         ovl_cache_free(&list);
 936         return err;
 937 }
 938
 939 static int ovl_unlink(struct inode *dir, struct dentry *dentry)
 940 {
 941         return ovl_do_remove(dentry, false);
 942 }
 943
 944 static int ovl_rmdir(struct inode *dir, struct dentry *dentry)
 945 {
 946         return ovl_do_remove(dentry, true);
 947 }
 948
 949 static bool ovl_type_merge_or_lower(struct dentry *dentry)
 950 {
 951         enum ovl_path_type type = ovl_path_type(dentry);
 952
 953         return OVL_TYPE_MERGE(type) || !OVL_TYPE_UPPER(type);
 954 }
 955
 956 static bool ovl_can_move(struct dentry *dentry)
 957 {
 958         return ovl_redirect_dir(dentry->d_sb) ||
 959                 !d_is_dir(dentry) || !ovl_type_merge_or_lower(dentry);
 960 }
 961
 962 static char *ovl_get_redirect(struct dentry *dentry, bool abs_redirect)
 963 {
 964         char *buf, *ret;
 965         struct dentry *d, *tmp;
 966         int buflen = ovl_redirect_max + 1;
 967
 968         if (!abs_redirect) {
 969                 ret = kstrndup(dentry->d_name.name, dentry->d_name.len,
 970                                GFP_KERNEL);
 971                 goto out;
 972         }
 973
 974         buf = ret = kmalloc(buflen, GFP_KERNEL);
 975         if (!buf)
 976                 goto out;
 977
 978         buflen--;
 979         buf[buflen] = '\0';
 980         for (d = dget(dentry); !IS_ROOT(d);) {
 981                 const char *name;
 982                 int thislen;
 983
 984                 spin_lock(&d->d_lock);
 985                 name = ovl_dentry_get_redirect(d);
 986                 if (name) {
 987                         thislen = strlen(name);
 988                 } else {
 989                         name = d->d_name.name;
 990                         thislen = d->d_name.len;
 991                 }
 992
 993                 /* If path is too long, fall back to userspace move */
 994                 if (thislen + (name[0] != '/') > buflen) {
 995                         ret = ERR_PTR(-EXDEV);
 996                         spin_unlock(&d->d_lock);
 997                         goto out_put;
 998                 }
 999
1000                 buflen -= thislen;
1001                 memcpy(&buf[buflen], name, thislen);
1002                 spin_unlock(&d->d_lock);
1003                 tmp = dget_parent(d);
1004
1005                 dput(d);
1006                 d = tmp;
1007
1008                 /* Absolute redirect: finished */
1009                 if (buf[buflen] == '/')
1010                         break;
1011                 buflen--;
1012                 buf[buflen] = '/';
1013         }
1014         ret = kstrdup(&buf[buflen], GFP_KERNEL);
1015 out_put:
1016         dput(d);
1017         kfree(buf);
1018 out:
1019         return ret ? ret : ERR_PTR(-ENOMEM);
1020 }
1021
1022 static bool ovl_need_absolute_redirect(struct dentry *dentry, bool samedir)
1023 {
1024         struct dentry *lowerdentry;
1025
1026         if (!samedir)
1027                 return true;
1028
1029         if (d_is_dir(dentry))
1030                 return false;
1031
1032         /*
1033          * For non-dir hardlinked files, we need absolute redirects
1034          * in general as two upper hardlinks could be in different
1035          * dirs. We could put a relative redirect now and convert
1036          * it to absolute redirect later. But when nlink > 1 and
1037          * indexing is on, that means relative redirect needs to be
1038          * converted to absolute during copy up of another lower
1039          * hardllink as well.
1040          *
1041          * So without optimizing too much, just check if lower is
1042          * a hard link or not. If lower is hard link, put absolute
1043          * redirect.
1044          */
1045         lowerdentry = ovl_dentry_lower(dentry);
1046         return (d_inode(lowerdentry)->i_nlink > 1);
1047 }
1048
1049 static int ovl_set_redirect(struct dentry *dentry, bool samedir)
1050 {
1051         int err;
1052         struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
1053         const char *redirect = ovl_dentry_get_redirect(dentry);
1054         bool absolute_redirect = ovl_need_absolute_redirect(dentry, samedir);
1055
1056         if (redirect && (!absolute_redirect || redirect[0] == '/'))
1057                 return 0;
1058
1059         redirect = ovl_get_redirect(dentry, absolute_redirect);
1060         if (IS_ERR(redirect))
1061                 return PTR_ERR(redirect);
1062
1063         err = ovl_check_setxattr(ofs, ovl_dentry_upper(dentry),
1064                                  OVL_XATTR_REDIRECT,
1065                                  redirect, strlen(redirect), -EXDEV);
1066         if (!err) {
1067                 spin_lock(&dentry->d_lock);
1068                 ovl_dentry_set_redirect(dentry, redirect);
1069                 spin_unlock(&dentry->d_lock);
1070         } else {
1071                 kfree(redirect);
1072                 pr_warn_ratelimited("failed to set redirect (%i)\n",
1073                                     err);
1074                 /* Fall back to userspace copy-up */
1075                 err = -EXDEV;
1076         }
1077         return err;
1078 }
1079
1080 static int ovl_rename(struct user_namespace *mnt_userns, struct inode *olddir,
1081                       struct dentry *old, struct inode *newdir,
1082                       struct dentry *new, unsigned int flags)
1083 {
1084         int err;
1085         struct dentry *old_upperdir;
1086         struct dentry *new_upperdir;
1087         struct dentry *olddentry;
1088         struct dentry *newdentry;
1089         struct dentry *trap;
1090         bool old_opaque;
1091         bool new_opaque;
1092         bool cleanup_whiteout = false;
1093         bool update_nlink = false;
1094         bool overwrite = !(flags & RENAME_EXCHANGE);
1095         bool is_dir = d_is_dir(old);
1096         bool new_is_dir = d_is_dir(new);
1097         bool samedir = olddir == newdir;
1098         struct dentry *opaquedir = NULL;
1099         const struct cred *old_cred = NULL;
1100         struct ovl_fs *ofs = OVL_FS(old->d_sb);
1101         LIST_HEAD(list);
1102
1103         err = -EINVAL;
1104         if (flags & ~(RENAME_EXCHANGE | RENAME_NOREPLACE))
1105                 goto out;
1106
1107         flags &= ~RENAME_NOREPLACE;
1108
1109         /* Don't copy up directory trees */
1110         err = -EXDEV;
1111         if (!ovl_can_move(old))
1112                 goto out;
1113         if (!overwrite && !ovl_can_move(new))
1114                 goto out;
1115
1116         if (overwrite && new_is_dir && !ovl_pure_upper(new)) {
1117                 err = ovl_check_empty_dir(new, &list);
1118                 if (err)
1119                         goto out;
1120         }
1121
1122         if (overwrite) {
1123                 if (ovl_lower_positive(old)) {
1124                         if (!ovl_dentry_is_whiteout(new)) {
1125                                 /* Whiteout source */
1126                                 flags |= RENAME_WHITEOUT;
1127                         } else {
1128                                 /* Switch whiteouts */
1129                                 flags |= RENAME_EXCHANGE;
1130                         }
1131                 } else if (is_dir && ovl_dentry_is_whiteout(new)) {
1132                         flags |= RENAME_EXCHANGE;
1133                         cleanup_whiteout = true;
1134                 }
1135         }
1136
1137         err = ovl_want_write(old);
1138         if (err)
1139                 goto out;
1140
1141         err = ovl_copy_up(old);
1142         if (err)
1143                 goto out_drop_write;
1144
1145         err = ovl_copy_up(new->d_parent);
1146         if (err)
1147                 goto out_drop_write;
1148         if (!overwrite) {
1149                 err = ovl_copy_up(new);
1150                 if (err)
1151                         goto out_drop_write;
1152         } else if (d_inode(new)) {
1153                 err = ovl_nlink_start(new);
1154                 if (err)
1155                         goto out_drop_write;
1156
1157                 update_nlink = true;
1158         }
1159
1160         old_cred = ovl_override_creds(old->d_sb);
1161
1162         if (!list_empty(&list)) {
1163                 opaquedir = ovl_clear_empty(new, &list);
1164                 err = PTR_ERR(opaquedir);
1165                 if (IS_ERR(opaquedir)) {
1166                         opaquedir = NULL;
1167                         goto out_revert_creds;
1168                 }
1169         }
1170
1171         old_upperdir = ovl_dentry_upper(old->d_parent);
1172         new_upperdir = ovl_dentry_upper(new->d_parent);
1173
1174         if (!samedir) {
1175                 /*
1176                  * When moving a merge dir or non-dir with copy up origin into
1177                  * a new parent, we are marking the new parent dir "impure".
1178                  * When ovl_iterate() iterates an "impure" upper dir, it will
1179                  * lookup the origin inodes of the entries to fill d_ino.
1180                  */
1181                 if (ovl_type_origin(old)) {
1182                         err = ovl_set_impure(new->d_parent, new_upperdir);
1183                         if (err)
1184                                 goto out_revert_creds;
1185                 }
1186                 if (!overwrite && ovl_type_origin(new)) {
1187                         err = ovl_set_impure(old->d_parent, old_upperdir);
1188                         if (err)
1189                                 goto out_revert_creds;
1190                 }
1191         }
1192
1193         trap = lock_rename(new_upperdir, old_upperdir);
1194
1195         olddentry = ovl_lookup_upper(ofs, old->d_name.name, old_upperdir,
1196                                      old->d_name.len);
1197         err = PTR_ERR(olddentry);
1198         if (IS_ERR(olddentry))
1199                 goto out_unlock;
1200
1201         err = -ESTALE;
1202         if (!ovl_matches_upper(old, olddentry))
1203                 goto out_dput_old;
1204
1205         newdentry = ovl_lookup_upper(ofs, new->d_name.name, new_upperdir,
1206                                      new->d_name.len);
1207         err = PTR_ERR(newdentry);
1208         if (IS_ERR(newdentry))
1209                 goto out_dput_old;
1210
1211         old_opaque = ovl_dentry_is_opaque(old);
1212         new_opaque = ovl_dentry_is_opaque(new);
1213
1214         err = -ESTALE;
1215         if (d_inode(new) && ovl_dentry_upper(new)) {
1216                 if (opaquedir) {
1217                         if (newdentry != opaquedir)
1218                                 goto out_dput;
1219                 } else {
1220                         if (!ovl_matches_upper(new, newdentry))
1221                                 goto out_dput;
1222                 }
1223         } else {
1224                 if (!d_is_negative(newdentry)) {
1225                         if (!new_opaque || !ovl_is_whiteout(newdentry))
1226                                 goto out_dput;
1227                 } else {
1228                         if (flags & RENAME_EXCHANGE)
1229                                 goto out_dput;
1230                 }
1231         }
1232
1233         if (olddentry == trap)
1234                 goto out_dput;
1235         if (newdentry == trap)
1236                 goto out_dput;
1237
1238         if (olddentry->d_inode == newdentry->d_inode)
1239                 goto out_dput;
1240
1241         err = 0;
1242         if (ovl_type_merge_or_lower(old))
1243                 err = ovl_set_redirect(old, samedir);
1244         else if (is_dir && !old_opaque && ovl_type_merge(new->d_parent))
1245                 err = ovl_set_opaque_xerr(old, olddentry, -EXDEV);
1246         if (err)
1247                 goto out_dput;
1248
1249         if (!overwrite && ovl_type_merge_or_lower(new))
1250                 err = ovl_set_redirect(new, samedir);
1251         else if (!overwrite && new_is_dir && !new_opaque &&
1252                  ovl_type_merge(old->d_parent))
1253                 err = ovl_set_opaque_xerr(new, newdentry, -EXDEV);
1254         if (err)
1255                 goto out_dput;
1256
1257         err = ovl_do_rename(ofs, old_upperdir->d_inode, olddentry,
1258                             new_upperdir->d_inode, newdentry, flags);
1259         if (err)
1260                 goto out_dput;
1261
1262         if (cleanup_whiteout)
1263                 ovl_cleanup(ofs, old_upperdir->d_inode, newdentry);
1264
1265         if (overwrite && d_inode(new)) {
1266                 if (new_is_dir)
1267                         clear_nlink(d_inode(new));
1268                 else
1269                         ovl_drop_nlink(new);
1270         }
1271
1272         ovl_dir_modified(old->d_parent, ovl_type_origin(old) ||
1273                          (!overwrite && ovl_type_origin(new)));
1274         ovl_dir_modified(new->d_parent, ovl_type_origin(old) ||
1275                          (d_inode(new) && ovl_type_origin(new)));
1276
1277         /* copy ctime: */
1278         ovl_copyattr(d_inode(old));
1279         if (d_inode(new) && ovl_dentry_upper(new))
1280                 ovl_copyattr(d_inode(new));
1281
1282 out_dput:
1283         dput(newdentry);
1284 out_dput_old:
1285         dput(olddentry);
1286 out_unlock:
1287         unlock_rename(new_upperdir, old_upperdir);
1288 out_revert_creds:
1289         revert_creds(old_cred);
1290         if (update_nlink)
1291                 ovl_nlink_end(new);
1292 out_drop_write:
1293         ovl_drop_write(old);
1294 out:
1295         dput(opaquedir);
1296         ovl_cache_free(&list);
1297         return err;
1298 }
1299
1300 const struct inode_operations ovl_dir_inode_operations = {
1301         .lookup         = ovl_lookup,
1302         .mkdir          = ovl_mkdir,
1303         .symlink        = ovl_symlink,
1304         .unlink         = ovl_unlink,
1305         .rmdir          = ovl_rmdir,
1306         .rename         = ovl_rename,
1307         .link           = ovl_link,
1308         .setattr        = ovl_setattr,
1309         .create         = ovl_create,
1310         .mknod          = ovl_mknod,
1311         .permission     = ovl_permission,
1312         .getattr        = ovl_getattr,
1313         .listxattr      = ovl_listxattr,
1314         .get_acl        = ovl_get_acl,
1315         .update_time    = ovl_update_time,
1316         .fileattr_get   = ovl_fileattr_get,
1317         .fileattr_set   = ovl_fileattr_set,
1318 };