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_init_reval(dentry, newdentry);
 273
 274         if (!hardlink) {
 275                 /*
 276                  * ovl_obtain_alias() can be called after ovl_create_real()
 277                  * and before we get here, so we may get an inode from cache
 278                  * with the same real upperdentry that is not the inode we
 279                  * pre-allocated.  In this case we will use the cached inode
 280                  * to instantiate the new dentry.
 281                  *
 282                  * XXX: if we ever use ovl_obtain_alias() to decode directory
 283                  * file handles, need to use ovl_get_inode_locked() and
 284                  * d_instantiate_new() here to prevent from creating two
 285                  * hashed directory inode aliases.
 286                  */
 287                 inode = ovl_get_inode(dentry->d_sb, &oip);
 288                 if (IS_ERR(inode))
 289                         return PTR_ERR(inode);
 290                 if (inode == oip.newinode)
 291                         ovl_set_flag(OVL_UPPERDATA, inode);
 292         } else {
 293                 WARN_ON(ovl_inode_real(inode) != d_inode(newdentry));
 294                 dput(newdentry);
 295                 inc_nlink(inode);
 296         }
 297
 298         d_instantiate(dentry, inode);
 299         if (inode != oip.newinode) {
 300                 pr_warn_ratelimited("newly created inode found in cache (%pd2)\n",
 301                                     dentry);
 302         }
 303
 304         /* Force lookup of new upper hardlink to find its lower */
 305         if (hardlink)
 306                 d_drop(dentry);
 307
 308         return 0;
 309 }
 310
 311 static bool ovl_type_merge(struct dentry *dentry)
 312 {
 313         return OVL_TYPE_MERGE(ovl_path_type(dentry));
 314 }
 315
 316 static bool ovl_type_origin(struct dentry *dentry)
 317 {
 318         return OVL_TYPE_ORIGIN(ovl_path_type(dentry));
 319 }
 320
 321 static int ovl_create_upper(struct dentry *dentry, struct inode *inode,
 322                             struct ovl_cattr *attr)
 323 {
 324         struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
 325         struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
 326         struct inode *udir = upperdir->d_inode;
 327         struct dentry *newdentry;
 328         int err;
 329
 330         if (!attr->hardlink && !IS_POSIXACL(udir))
 331                 attr->mode &= ~current_umask();
 332
 333         inode_lock_nested(udir, I_MUTEX_PARENT);
 334         newdentry = ovl_create_real(ofs, udir,
 335                                     ovl_lookup_upper(ofs, dentry->d_name.name,
 336                                                      upperdir, 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(ofs, 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 ovl_fs *ofs = OVL_FS(dentry->d_sb);
 365         struct dentry *workdir = ovl_workdir(dentry);
 366         struct inode *wdir = workdir->d_inode;
 367         struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
 368         struct inode *udir = upperdir->d_inode;
 369         struct path upperpath;
 370         struct dentry *upper;
 371         struct dentry *opaquedir;
 372         struct kstat stat;
 373         int err;
 374
 375         if (WARN_ON(!workdir))
 376                 return ERR_PTR(-EROFS);
 377
 378         err = ovl_lock_rename_workdir(workdir, upperdir);
 379         if (err)
 380                 goto out;
 381
 382         ovl_path_upper(dentry, &upperpath);
 383         err = vfs_getattr(&upperpath, &stat,
 384                           STATX_BASIC_STATS, AT_STATX_SYNC_AS_STAT);
 385         if (err)
 386                 goto out_unlock;
 387
 388         err = -ESTALE;
 389         if (!S_ISDIR(stat.mode))
 390                 goto out_unlock;
 391         upper = upperpath.dentry;
 392         if (upper->d_parent->d_inode != udir)
 393                 goto out_unlock;
 394
 395         opaquedir = ovl_create_temp(ofs, workdir, OVL_CATTR(stat.mode));
 396         err = PTR_ERR(opaquedir);
 397         if (IS_ERR(opaquedir))
 398                 goto out_unlock;
 399
 400         err = ovl_copy_xattr(dentry->d_sb, &upperpath, opaquedir);
 401         if (err)
 402                 goto out_cleanup;
 403
 404         err = ovl_set_opaque(dentry, opaquedir);
 405         if (err)
 406                 goto out_cleanup;
 407
 408         inode_lock(opaquedir->d_inode);
 409         err = ovl_set_attr(ofs, opaquedir, &stat);
 410         inode_unlock(opaquedir->d_inode);
 411         if (err)
 412                 goto out_cleanup;
 413
 414         err = ovl_do_rename(ofs, wdir, opaquedir, udir, upper, RENAME_EXCHANGE);
 415         if (err)
 416                 goto out_cleanup;
 417
 418         ovl_cleanup_whiteouts(ofs, upper, list);
 419         ovl_cleanup(ofs, wdir, upper);
 420         unlock_rename(workdir, upperdir);
 421
 422         /* dentry's upper doesn't match now, get rid of it */
 423         d_drop(dentry);
 424
 425         return opaquedir;
 426
 427 out_cleanup:
 428         ovl_cleanup(ofs, wdir, opaquedir);
 429         dput(opaquedir);
 430 out_unlock:
 431         unlock_rename(workdir, upperdir);
 432 out:
 433         return ERR_PTR(err);
 434 }
 435
 436 static int ovl_set_upper_acl(struct ovl_fs *ofs, struct dentry *upperdentry,
 437                              const char *name, const struct posix_acl *acl)
 438 {
 439         void *buffer;
 440         size_t size;
 441         int err;
 442
 443         if (!IS_ENABLED(CONFIG_FS_POSIX_ACL) || !acl)
 444                 return 0;
 445
 446         size = posix_acl_xattr_size(acl->a_count);
 447         buffer = kmalloc(size, GFP_KERNEL);
 448         if (!buffer)
 449                 return -ENOMEM;
 450
 451         err = posix_acl_to_xattr(&init_user_ns, acl, buffer, size);
 452         if (err < 0)
 453                 goto out_free;
 454
 455         err = ovl_do_setxattr(ofs, upperdentry, name, buffer, size, XATTR_CREATE);
 456 out_free:
 457         kfree(buffer);
 458         return err;
 459 }
 460
 461 static int ovl_create_over_whiteout(struct dentry *dentry, struct inode *inode,
 462                                     struct ovl_cattr *cattr)
 463 {
 464         struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
 465         struct dentry *workdir = ovl_workdir(dentry);
 466         struct inode *wdir = workdir->d_inode;
 467         struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
 468         struct inode *udir = upperdir->d_inode;
 469         struct dentry *upper;
 470         struct dentry *newdentry;
 471         int err;
 472         struct posix_acl *acl, *default_acl;
 473         bool hardlink = !!cattr->hardlink;
 474
 475         if (WARN_ON(!workdir))
 476                 return -EROFS;
 477
 478         if (!hardlink) {
 479                 err = posix_acl_create(dentry->d_parent->d_inode,
 480                                        &cattr->mode, &default_acl, &acl);
 481                 if (err)
 482                         return err;
 483         }
 484
 485         err = ovl_lock_rename_workdir(workdir, upperdir);
 486         if (err)
 487                 goto out;
 488
 489         upper = ovl_lookup_upper(ofs, dentry->d_name.name, upperdir,
 490                                  dentry->d_name.len);
 491         err = PTR_ERR(upper);
 492         if (IS_ERR(upper))
 493                 goto out_unlock;
 494
 495         err = -ESTALE;
 496         if (d_is_negative(upper) || !IS_WHITEOUT(d_inode(upper)))
 497                 goto out_dput;
 498
 499         newdentry = ovl_create_temp(ofs, workdir, cattr);
 500         err = PTR_ERR(newdentry);
 501         if (IS_ERR(newdentry))
 502                 goto out_dput;
 503
 504         /*
 505          * mode could have been mutilated due to umask (e.g. sgid directory)
 506          */
 507         if (!hardlink &&
 508             !S_ISLNK(cattr->mode) &&
 509             newdentry->d_inode->i_mode != cattr->mode) {
 510                 struct iattr attr = {
 511                         .ia_valid = ATTR_MODE,
 512                         .ia_mode = cattr->mode,
 513                 };
 514                 inode_lock(newdentry->d_inode);
 515                 err = ovl_do_notify_change(ofs, newdentry, &attr);
 516                 inode_unlock(newdentry->d_inode);
 517                 if (err)
 518                         goto out_cleanup;
 519         }
 520         if (!hardlink) {
 521                 err = ovl_set_upper_acl(ofs, newdentry,
 522                                         XATTR_NAME_POSIX_ACL_ACCESS, acl);
 523                 if (err)
 524                         goto out_cleanup;
 525
 526                 err = ovl_set_upper_acl(ofs, newdentry,
 527                                         XATTR_NAME_POSIX_ACL_DEFAULT, default_acl);
 528                 if (err)
 529                         goto out_cleanup;
 530         }
 531
 532         if (!hardlink && S_ISDIR(cattr->mode)) {
 533                 err = ovl_set_opaque(dentry, newdentry);
 534                 if (err)
 535                         goto out_cleanup;
 536
 537                 err = ovl_do_rename(ofs, wdir, newdentry, udir, upper,
 538                                     RENAME_EXCHANGE);
 539                 if (err)
 540                         goto out_cleanup;
 541
 542                 ovl_cleanup(ofs, wdir, upper);
 543         } else {
 544                 err = ovl_do_rename(ofs, wdir, newdentry, udir, upper, 0);
 545                 if (err)
 546                         goto out_cleanup;
 547         }
 548         err = ovl_instantiate(dentry, inode, newdentry, hardlink);
 549         if (err) {
 550                 ovl_cleanup(ofs, udir, newdentry);
 551                 dput(newdentry);
 552         }
 553 out_dput:
 554         dput(upper);
 555 out_unlock:
 556         unlock_rename(workdir, upperdir);
 557 out:
 558         if (!hardlink) {
 559                 posix_acl_release(acl);
 560                 posix_acl_release(default_acl);
 561         }
 562         return err;
 563
 564 out_cleanup:
 565         ovl_cleanup(ofs, wdir, newdentry);
 566         dput(newdentry);
 567         goto out_dput;
 568 }
 569
 570 static int ovl_create_or_link(struct dentry *dentry, struct inode *inode,
 571                               struct ovl_cattr *attr, bool origin)
 572 {
 573         int err;
 574         const struct cred *old_cred;
 575         struct cred *override_cred;
 576         struct dentry *parent = dentry->d_parent;
 577
 578         err = ovl_copy_up(parent);
 579         if (err)
 580                 return err;
 581
 582         old_cred = ovl_override_creds(dentry->d_sb);
 583
 584         /*
 585          * When linking a file with copy up origin into a new parent, mark the
 586          * new parent dir "impure".
 587          */
 588         if (origin) {
 589                 err = ovl_set_impure(parent, ovl_dentry_upper(parent));
 590                 if (err)
 591                         goto out_revert_creds;
 592         }
 593
 594         if (!attr->hardlink) {
 595                 err = -ENOMEM;
 596                 override_cred = prepare_creds();
 597                 if (!override_cred)
 598                         goto out_revert_creds;
 599                 /*
 600                  * In the creation cases(create, mkdir, mknod, symlink),
 601                  * ovl should transfer current's fs{u,g}id to underlying
 602                  * fs. Because underlying fs want to initialize its new
 603                  * inode owner using current's fs{u,g}id. And in this
 604                  * case, the @inode is a new inode that is initialized
 605                  * in inode_init_owner() to current's fs{u,g}id. So use
 606                  * the inode's i_{u,g}id to override the cred's fs{u,g}id.
 607                  *
 608                  * But in the other hardlink case, ovl_link() does not
 609                  * create a new inode, so just use the ovl mounter's
 610                  * fs{u,g}id.
 611                  */
 612                 override_cred->fsuid = inode->i_uid;
 613                 override_cred->fsgid = inode->i_gid;
 614                 err = security_dentry_create_files_as(dentry,
 615                                 attr->mode, &dentry->d_name, old_cred,
 616                                 override_cred);
 617                 if (err) {
 618                         put_cred(override_cred);
 619                         goto out_revert_creds;
 620                 }
 621                 put_cred(override_creds(override_cred));
 622                 put_cred(override_cred);
 623         }
 624
 625         if (!ovl_dentry_is_whiteout(dentry))
 626                 err = ovl_create_upper(dentry, inode, attr);
 627         else
 628                 err = ovl_create_over_whiteout(dentry, inode, attr);
 629
 630 out_revert_creds:
 631         revert_creds(old_cred);
 632         return err;
 633 }
 634
 635 static int ovl_create_object(struct dentry *dentry, int mode, dev_t rdev,
 636                              const char *link)
 637 {
 638         int err;
 639         struct inode *inode;
 640         struct ovl_cattr attr = {
 641                 .rdev = rdev,
 642                 .link = link,
 643         };
 644
 645         err = ovl_want_write(dentry);
 646         if (err)
 647                 goto out;
 648
 649         /* Preallocate inode to be used by ovl_get_inode() */
 650         err = -ENOMEM;
 651         inode = ovl_new_inode(dentry->d_sb, mode, rdev);
 652         if (!inode)
 653                 goto out_drop_write;
 654
 655         spin_lock(&inode->i_lock);
 656         inode->i_state |= I_CREATING;
 657         spin_unlock(&inode->i_lock);
 658
 659         inode_init_owner(&init_user_ns, inode, dentry->d_parent->d_inode, mode);
 660         attr.mode = inode->i_mode;
 661
 662         err = ovl_create_or_link(dentry, inode, &attr, false);
 663         /* Did we end up using the preallocated inode? */
 664         if (inode != d_inode(dentry))
 665                 iput(inode);
 666
 667 out_drop_write:
 668         ovl_drop_write(dentry);
 669 out:
 670         return err;
 671 }
 672
 673 static int ovl_create(struct user_namespace *mnt_userns, struct inode *dir,
 674                       struct dentry *dentry, umode_t mode, bool excl)
 675 {
 676         return ovl_create_object(dentry, (mode & 07777) | S_IFREG, 0, NULL);
 677 }
 678
 679 static int ovl_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
 680                      struct dentry *dentry, umode_t mode)
 681 {
 682         return ovl_create_object(dentry, (mode & 07777) | S_IFDIR, 0, NULL);
 683 }
 684
 685 static int ovl_mknod(struct user_namespace *mnt_userns, struct inode *dir,
 686                      struct dentry *dentry, umode_t mode, dev_t rdev)
 687 {
 688         /* Don't allow creation of "whiteout" on overlay */
 689         if (S_ISCHR(mode) && rdev == WHITEOUT_DEV)
 690                 return -EPERM;
 691
 692         return ovl_create_object(dentry, mode, rdev, NULL);
 693 }
 694
 695 static int ovl_symlink(struct user_namespace *mnt_userns, struct inode *dir,
 696                        struct dentry *dentry, const char *link)
 697 {
 698         return ovl_create_object(dentry, S_IFLNK, 0, link);
 699 }
 700
 701 static int ovl_set_link_redirect(struct dentry *dentry)
 702 {
 703         const struct cred *old_cred;
 704         int err;
 705
 706         old_cred = ovl_override_creds(dentry->d_sb);
 707         err = ovl_set_redirect(dentry, false);
 708         revert_creds(old_cred);
 709
 710         return err;
 711 }
 712
 713 static int ovl_link(struct dentry *old, struct inode *newdir,
 714                     struct dentry *new)
 715 {
 716         int err;
 717         struct inode *inode;
 718
 719         err = ovl_want_write(old);
 720         if (err)
 721                 goto out;
 722
 723         err = ovl_copy_up(old);
 724         if (err)
 725                 goto out_drop_write;
 726
 727         err = ovl_copy_up(new->d_parent);
 728         if (err)
 729                 goto out_drop_write;
 730
 731         if (ovl_is_metacopy_dentry(old)) {
 732                 err = ovl_set_link_redirect(old);
 733                 if (err)
 734                         goto out_drop_write;
 735         }
 736
 737         err = ovl_nlink_start(old);
 738         if (err)
 739                 goto out_drop_write;
 740
 741         inode = d_inode(old);
 742         ihold(inode);
 743
 744         err = ovl_create_or_link(new, inode,
 745                         &(struct ovl_cattr) {.hardlink = ovl_dentry_upper(old)},
 746                         ovl_type_origin(old));
 747         if (err)
 748                 iput(inode);
 749
 750         ovl_nlink_end(old);
 751 out_drop_write:
 752         ovl_drop_write(old);
 753 out:
 754         return err;
 755 }
 756
 757 static bool ovl_matches_upper(struct dentry *dentry, struct dentry *upper)
 758 {
 759         return d_inode(ovl_dentry_upper(dentry)) == d_inode(upper);
 760 }
 761
 762 static int ovl_remove_and_whiteout(struct dentry *dentry,
 763                                    struct list_head *list)
 764 {
 765         struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
 766         struct dentry *workdir = ovl_workdir(dentry);
 767         struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
 768         struct dentry *upper;
 769         struct dentry *opaquedir = NULL;
 770         int err;
 771
 772         if (WARN_ON(!workdir))
 773                 return -EROFS;
 774
 775         if (!list_empty(list)) {
 776                 opaquedir = ovl_clear_empty(dentry, list);
 777                 err = PTR_ERR(opaquedir);
 778                 if (IS_ERR(opaquedir))
 779                         goto out;
 780         }
 781
 782         err = ovl_lock_rename_workdir(workdir, upperdir);
 783         if (err)
 784                 goto out_dput;
 785
 786         upper = ovl_lookup_upper(ofs, dentry->d_name.name, upperdir,
 787                                  dentry->d_name.len);
 788         err = PTR_ERR(upper);
 789         if (IS_ERR(upper))
 790                 goto out_unlock;
 791
 792         err = -ESTALE;
 793         if ((opaquedir && upper != opaquedir) ||
 794             (!opaquedir && ovl_dentry_upper(dentry) &&
 795              !ovl_matches_upper(dentry, upper))) {
 796                 goto out_dput_upper;
 797         }
 798
 799         err = ovl_cleanup_and_whiteout(ofs, d_inode(upperdir), upper);
 800         if (err)
 801                 goto out_d_drop;
 802
 803         ovl_dir_modified(dentry->d_parent, true);
 804 out_d_drop:
 805         d_drop(dentry);
 806 out_dput_upper:
 807         dput(upper);
 808 out_unlock:
 809         unlock_rename(workdir, upperdir);
 810 out_dput:
 811         dput(opaquedir);
 812 out:
 813         return err;
 814 }
 815
 816 static int ovl_remove_upper(struct dentry *dentry, bool is_dir,
 817                             struct list_head *list)
 818 {
 819         struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
 820         struct dentry *upperdir = ovl_dentry_upper(dentry->d_parent);
 821         struct inode *dir = upperdir->d_inode;
 822         struct dentry *upper;
 823         struct dentry *opaquedir = NULL;
 824         int err;
 825
 826         if (!list_empty(list)) {
 827                 opaquedir = ovl_clear_empty(dentry, list);
 828                 err = PTR_ERR(opaquedir);
 829                 if (IS_ERR(opaquedir))
 830                         goto out;
 831         }
 832
 833         inode_lock_nested(dir, I_MUTEX_PARENT);
 834         upper = ovl_lookup_upper(ofs, dentry->d_name.name, upperdir,
 835                                  dentry->d_name.len);
 836         err = PTR_ERR(upper);
 837         if (IS_ERR(upper))
 838                 goto out_unlock;
 839
 840         err = -ESTALE;
 841         if ((opaquedir && upper != opaquedir) ||
 842             (!opaquedir && !ovl_matches_upper(dentry, upper)))
 843                 goto out_dput_upper;
 844
 845         if (is_dir)
 846                 err = ovl_do_rmdir(ofs, dir, upper);
 847         else
 848                 err = ovl_do_unlink(ofs, dir, upper);
 849         ovl_dir_modified(dentry->d_parent, ovl_type_origin(dentry));
 850
 851         /*
 852          * Keeping this dentry hashed would mean having to release
 853          * upperpath/lowerpath, which could only be done if we are the
 854          * sole user of this dentry.  Too tricky...  Just unhash for
 855          * now.
 856          */
 857         if (!err)
 858                 d_drop(dentry);
 859 out_dput_upper:
 860         dput(upper);
 861 out_unlock:
 862         inode_unlock(dir);
 863         dput(opaquedir);
 864 out:
 865         return err;
 866 }
 867
 868 static bool ovl_pure_upper(struct dentry *dentry)
 869 {
 870         return !ovl_dentry_lower(dentry) &&
 871                !ovl_test_flag(OVL_WHITEOUTS, d_inode(dentry));
 872 }
 873
 874 static void ovl_drop_nlink(struct dentry *dentry)
 875 {
 876         struct inode *inode = d_inode(dentry);
 877         struct dentry *alias;
 878
 879         /* Try to find another, hashed alias */
 880         spin_lock(&inode->i_lock);
 881         hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
 882                 if (alias != dentry && !d_unhashed(alias))
 883                         break;
 884         }
 885         spin_unlock(&inode->i_lock);
 886
 887         /*
 888          * Changes to underlying layers may cause i_nlink to lose sync with
 889          * reality.  In this case prevent the link count from going to zero
 890          * prematurely.
 891          */
 892         if (inode->i_nlink > !!alias)
 893                 drop_nlink(inode);
 894 }
 895
 896 static int ovl_do_remove(struct dentry *dentry, bool is_dir)
 897 {
 898         int err;
 899         const struct cred *old_cred;
 900         bool lower_positive = ovl_lower_positive(dentry);
 901         LIST_HEAD(list);
 902
 903         /* No need to clean pure upper removed by vfs_rmdir() */
 904         if (is_dir && (lower_positive || !ovl_pure_upper(dentry))) {
 905                 err = ovl_check_empty_dir(dentry, &list);
 906                 if (err)
 907                         goto out;
 908         }
 909
 910         err = ovl_want_write(dentry);
 911         if (err)
 912                 goto out;
 913
 914         err = ovl_copy_up(dentry->d_parent);
 915         if (err)
 916                 goto out_drop_write;
 917
 918         err = ovl_nlink_start(dentry);
 919         if (err)
 920                 goto out_drop_write;
 921
 922         old_cred = ovl_override_creds(dentry->d_sb);
 923         if (!lower_positive)
 924                 err = ovl_remove_upper(dentry, is_dir, &list);
 925         else
 926                 err = ovl_remove_and_whiteout(dentry, &list);
 927         revert_creds(old_cred);
 928         if (!err) {
 929                 if (is_dir)
 930                         clear_nlink(dentry->d_inode);
 931                 else
 932                         ovl_drop_nlink(dentry);
 933         }
 934         ovl_nlink_end(dentry);
 935
 936         /*
 937          * Copy ctime
 938          *
 939          * Note: we fail to update ctime if there was no copy-up, only a
 940          * whiteout
 941          */
 942         if (ovl_dentry_upper(dentry))
 943                 ovl_copyattr(d_inode(dentry));
 944
 945 out_drop_write:
 946         ovl_drop_write(dentry);
 947 out:
 948         ovl_cache_free(&list);
 949         return err;
 950 }
 951
 952 static int ovl_unlink(struct inode *dir, struct dentry *dentry)
 953 {
 954         return ovl_do_remove(dentry, false);
 955 }
 956
 957 static int ovl_rmdir(struct inode *dir, struct dentry *dentry)
 958 {
 959         return ovl_do_remove(dentry, true);
 960 }
 961
 962 static bool ovl_type_merge_or_lower(struct dentry *dentry)
 963 {
 964         enum ovl_path_type type = ovl_path_type(dentry);
 965
 966         return OVL_TYPE_MERGE(type) || !OVL_TYPE_UPPER(type);
 967 }
 968
 969 static bool ovl_can_move(struct dentry *dentry)
 970 {
 971         return ovl_redirect_dir(dentry->d_sb) ||
 972                 !d_is_dir(dentry) || !ovl_type_merge_or_lower(dentry);
 973 }
 974
 975 static char *ovl_get_redirect(struct dentry *dentry, bool abs_redirect)
 976 {
 977         char *buf, *ret;
 978         struct dentry *d, *tmp;
 979         int buflen = ovl_redirect_max + 1;
 980
 981         if (!abs_redirect) {
 982                 ret = kstrndup(dentry->d_name.name, dentry->d_name.len,
 983                                GFP_KERNEL);
 984                 goto out;
 985         }
 986
 987         buf = ret = kmalloc(buflen, GFP_KERNEL);
 988         if (!buf)
 989                 goto out;
 990
 991         buflen--;
 992         buf[buflen] = '\0';
 993         for (d = dget(dentry); !IS_ROOT(d);) {
 994                 const char *name;
 995                 int thislen;
 996
 997                 spin_lock(&d->d_lock);
 998                 name = ovl_dentry_get_redirect(d);
 999                 if (name) {
1000                         thislen = strlen(name);
1001                 } else {
1002                         name = d->d_name.name;
1003                         thislen = d->d_name.len;
1004                 }
1005
1006                 /* If path is too long, fall back to userspace move */
1007                 if (thislen + (name[0] != '/') > buflen) {
1008                         ret = ERR_PTR(-EXDEV);
1009                         spin_unlock(&d->d_lock);
1010                         goto out_put;
1011                 }
1012
1013                 buflen -= thislen;
1014                 memcpy(&buf[buflen], name, thislen);
1015                 spin_unlock(&d->d_lock);
1016                 tmp = dget_parent(d);
1017
1018                 dput(d);
1019                 d = tmp;
1020
1021                 /* Absolute redirect: finished */
1022                 if (buf[buflen] == '/')
1023                         break;
1024                 buflen--;
1025                 buf[buflen] = '/';
1026         }
1027         ret = kstrdup(&buf[buflen], GFP_KERNEL);
1028 out_put:
1029         dput(d);
1030         kfree(buf);
1031 out:
1032         return ret ? ret : ERR_PTR(-ENOMEM);
1033 }
1034
1035 static bool ovl_need_absolute_redirect(struct dentry *dentry, bool samedir)
1036 {
1037         struct dentry *lowerdentry;
1038
1039         if (!samedir)
1040                 return true;
1041
1042         if (d_is_dir(dentry))
1043                 return false;
1044
1045         /*
1046          * For non-dir hardlinked files, we need absolute redirects
1047          * in general as two upper hardlinks could be in different
1048          * dirs. We could put a relative redirect now and convert
1049          * it to absolute redirect later. But when nlink > 1 and
1050          * indexing is on, that means relative redirect needs to be
1051          * converted to absolute during copy up of another lower
1052          * hardllink as well.
1053          *
1054          * So without optimizing too much, just check if lower is
1055          * a hard link or not. If lower is hard link, put absolute
1056          * redirect.
1057          */
1058         lowerdentry = ovl_dentry_lower(dentry);
1059         return (d_inode(lowerdentry)->i_nlink > 1);
1060 }
1061
1062 static int ovl_set_redirect(struct dentry *dentry, bool samedir)
1063 {
1064         int err;
1065         struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
1066         const char *redirect = ovl_dentry_get_redirect(dentry);
1067         bool absolute_redirect = ovl_need_absolute_redirect(dentry, samedir);
1068
1069         if (redirect && (!absolute_redirect || redirect[0] == '/'))
1070                 return 0;
1071
1072         redirect = ovl_get_redirect(dentry, absolute_redirect);
1073         if (IS_ERR(redirect))
1074                 return PTR_ERR(redirect);
1075
1076         err = ovl_check_setxattr(ofs, ovl_dentry_upper(dentry),
1077                                  OVL_XATTR_REDIRECT,
1078                                  redirect, strlen(redirect), -EXDEV);
1079         if (!err) {
1080                 spin_lock(&dentry->d_lock);
1081                 ovl_dentry_set_redirect(dentry, redirect);
1082                 spin_unlock(&dentry->d_lock);
1083         } else {
1084                 kfree(redirect);
1085                 pr_warn_ratelimited("failed to set redirect (%i)\n",
1086                                     err);
1087                 /* Fall back to userspace copy-up */
1088                 err = -EXDEV;
1089         }
1090         return err;
1091 }
1092
1093 static int ovl_rename(struct user_namespace *mnt_userns, struct inode *olddir,
1094                       struct dentry *old, struct inode *newdir,
1095                       struct dentry *new, unsigned int flags)
1096 {
1097         int err;
1098         struct dentry *old_upperdir;
1099         struct dentry *new_upperdir;
1100         struct dentry *olddentry;
1101         struct dentry *newdentry;
1102         struct dentry *trap;
1103         bool old_opaque;
1104         bool new_opaque;
1105         bool cleanup_whiteout = false;
1106         bool update_nlink = false;
1107         bool overwrite = !(flags & RENAME_EXCHANGE);
1108         bool is_dir = d_is_dir(old);
1109         bool new_is_dir = d_is_dir(new);
1110         bool samedir = olddir == newdir;
1111         struct dentry *opaquedir = NULL;
1112         const struct cred *old_cred = NULL;
1113         struct ovl_fs *ofs = OVL_FS(old->d_sb);
1114         LIST_HEAD(list);
1115
1116         err = -EINVAL;
1117         if (flags & ~(RENAME_EXCHANGE | RENAME_NOREPLACE))
1118                 goto out;
1119
1120         flags &= ~RENAME_NOREPLACE;
1121
1122         /* Don't copy up directory trees */
1123         err = -EXDEV;
1124         if (!ovl_can_move(old))
1125                 goto out;
1126         if (!overwrite && !ovl_can_move(new))
1127                 goto out;
1128
1129         if (overwrite && new_is_dir && !ovl_pure_upper(new)) {
1130                 err = ovl_check_empty_dir(new, &list);
1131                 if (err)
1132                         goto out;
1133         }
1134
1135         if (overwrite) {
1136                 if (ovl_lower_positive(old)) {
1137                         if (!ovl_dentry_is_whiteout(new)) {
1138                                 /* Whiteout source */
1139                                 flags |= RENAME_WHITEOUT;
1140                         } else {
1141                                 /* Switch whiteouts */
1142                                 flags |= RENAME_EXCHANGE;
1143                         }
1144                 } else if (is_dir && ovl_dentry_is_whiteout(new)) {
1145                         flags |= RENAME_EXCHANGE;
1146                         cleanup_whiteout = true;
1147                 }
1148         }
1149
1150         err = ovl_want_write(old);
1151         if (err)
1152                 goto out;
1153
1154         err = ovl_copy_up(old);
1155         if (err)
1156                 goto out_drop_write;
1157
1158         err = ovl_copy_up(new->d_parent);
1159         if (err)
1160                 goto out_drop_write;
1161         if (!overwrite) {
1162                 err = ovl_copy_up(new);
1163                 if (err)
1164                         goto out_drop_write;
1165         } else if (d_inode(new)) {
1166                 err = ovl_nlink_start(new);
1167                 if (err)
1168                         goto out_drop_write;
1169
1170                 update_nlink = true;
1171         }
1172
1173         old_cred = ovl_override_creds(old->d_sb);
1174
1175         if (!list_empty(&list)) {
1176                 opaquedir = ovl_clear_empty(new, &list);
1177                 err = PTR_ERR(opaquedir);
1178                 if (IS_ERR(opaquedir)) {
1179                         opaquedir = NULL;
1180                         goto out_revert_creds;
1181                 }
1182         }
1183
1184         old_upperdir = ovl_dentry_upper(old->d_parent);
1185         new_upperdir = ovl_dentry_upper(new->d_parent);
1186
1187         if (!samedir) {
1188                 /*
1189                  * When moving a merge dir or non-dir with copy up origin into
1190                  * a new parent, we are marking the new parent dir "impure".
1191                  * When ovl_iterate() iterates an "impure" upper dir, it will
1192                  * lookup the origin inodes of the entries to fill d_ino.
1193                  */
1194                 if (ovl_type_origin(old)) {
1195                         err = ovl_set_impure(new->d_parent, new_upperdir);
1196                         if (err)
1197                                 goto out_revert_creds;
1198                 }
1199                 if (!overwrite && ovl_type_origin(new)) {
1200                         err = ovl_set_impure(old->d_parent, old_upperdir);
1201                         if (err)
1202                                 goto out_revert_creds;
1203                 }
1204         }
1205
1206         trap = lock_rename(new_upperdir, old_upperdir);
1207
1208         olddentry = ovl_lookup_upper(ofs, old->d_name.name, old_upperdir,
1209                                      old->d_name.len);
1210         err = PTR_ERR(olddentry);
1211         if (IS_ERR(olddentry))
1212                 goto out_unlock;
1213
1214         err = -ESTALE;
1215         if (!ovl_matches_upper(old, olddentry))
1216                 goto out_dput_old;
1217
1218         newdentry = ovl_lookup_upper(ofs, new->d_name.name, new_upperdir,
1219                                      new->d_name.len);
1220         err = PTR_ERR(newdentry);
1221         if (IS_ERR(newdentry))
1222                 goto out_dput_old;
1223
1224         old_opaque = ovl_dentry_is_opaque(old);
1225         new_opaque = ovl_dentry_is_opaque(new);
1226
1227         err = -ESTALE;
1228         if (d_inode(new) && ovl_dentry_upper(new)) {
1229                 if (opaquedir) {
1230                         if (newdentry != opaquedir)
1231                                 goto out_dput;
1232                 } else {
1233                         if (!ovl_matches_upper(new, newdentry))
1234                                 goto out_dput;
1235                 }
1236         } else {
1237                 if (!d_is_negative(newdentry)) {
1238                         if (!new_opaque || !ovl_is_whiteout(newdentry))
1239                                 goto out_dput;
1240                 } else {
1241                         if (flags & RENAME_EXCHANGE)
1242                                 goto out_dput;
1243                 }
1244         }
1245
1246         if (olddentry == trap)
1247                 goto out_dput;
1248         if (newdentry == trap)
1249                 goto out_dput;
1250
1251         if (olddentry->d_inode == newdentry->d_inode)
1252                 goto out_dput;
1253
1254         err = 0;
1255         if (ovl_type_merge_or_lower(old))
1256                 err = ovl_set_redirect(old, samedir);
1257         else if (is_dir && !old_opaque && ovl_type_merge(new->d_parent))
1258                 err = ovl_set_opaque_xerr(old, olddentry, -EXDEV);
1259         if (err)
1260                 goto out_dput;
1261
1262         if (!overwrite && ovl_type_merge_or_lower(new))
1263                 err = ovl_set_redirect(new, samedir);
1264         else if (!overwrite && new_is_dir && !new_opaque &&
1265                  ovl_type_merge(old->d_parent))
1266                 err = ovl_set_opaque_xerr(new, newdentry, -EXDEV);
1267         if (err)
1268                 goto out_dput;
1269
1270         err = ovl_do_rename(ofs, old_upperdir->d_inode, olddentry,
1271                             new_upperdir->d_inode, newdentry, flags);
1272         if (err)
1273                 goto out_dput;
1274
1275         if (cleanup_whiteout)
1276                 ovl_cleanup(ofs, old_upperdir->d_inode, newdentry);
1277
1278         if (overwrite && d_inode(new)) {
1279                 if (new_is_dir)
1280                         clear_nlink(d_inode(new));
1281                 else
1282                         ovl_drop_nlink(new);
1283         }
1284
1285         ovl_dir_modified(old->d_parent, ovl_type_origin(old) ||
1286                          (!overwrite && ovl_type_origin(new)));
1287         ovl_dir_modified(new->d_parent, ovl_type_origin(old) ||
1288                          (d_inode(new) && ovl_type_origin(new)));
1289
1290         /* copy ctime: */
1291         ovl_copyattr(d_inode(old));
1292         if (d_inode(new) && ovl_dentry_upper(new))
1293                 ovl_copyattr(d_inode(new));
1294
1295 out_dput:
1296         dput(newdentry);
1297 out_dput_old:
1298         dput(olddentry);
1299 out_unlock:
1300         unlock_rename(new_upperdir, old_upperdir);
1301 out_revert_creds:
1302         revert_creds(old_cred);
1303         if (update_nlink)
1304                 ovl_nlink_end(new);
1305 out_drop_write:
1306         ovl_drop_write(old);
1307 out:
1308         dput(opaquedir);
1309         ovl_cache_free(&list);
1310         return err;
1311 }
1312
1313 const struct inode_operations ovl_dir_inode_operations = {
1314         .lookup         = ovl_lookup,
1315         .mkdir          = ovl_mkdir,
1316         .symlink        = ovl_symlink,
1317         .unlink         = ovl_unlink,
1318         .rmdir          = ovl_rmdir,
1319         .rename         = ovl_rename,
1320         .link           = ovl_link,
1321         .setattr        = ovl_setattr,
1322         .create         = ovl_create,
1323         .mknod          = ovl_mknod,
1324         .permission     = ovl_permission,
1325         .getattr        = ovl_getattr,
1326         .listxattr      = ovl_listxattr,
1327         .get_acl        = ovl_get_acl,
1328         .update_time    = ovl_update_time,
1329         .fileattr_get   = ovl_fileattr_get,
1330         .fileattr_set   = ovl_fileattr_set,
1331 };