fs/overlayfs/util.c

   1 /*
   2  * Copyright (C) 2011 Novell Inc.
   3  * Copyright (C) 2016 Red Hat, Inc.
   4  *
   5  * This program is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 as published by
   7  * the Free Software Foundation.
   8  */
   9
  10 #include <linux/fs.h>
  11 #include <linux/mount.h>
  12 #include <linux/slab.h>
  13 #include <linux/cred.h>
  14 #include <linux/xattr.h>
  15 #include <linux/exportfs.h>
  16 #include <linux/uuid.h>
  17 #include <linux/namei.h>
  18 #include <linux/ratelimit.h>
  19 #include "overlayfs.h"
  20
  21 int ovl_want_write(struct dentry *dentry)
  22 {
  23         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
  24         return mnt_want_write(ofs->upper_mnt);
  25 }
  26
  27 void ovl_drop_write(struct dentry *dentry)
  28 {
  29         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
  30         mnt_drop_write(ofs->upper_mnt);
  31 }
  32
  33 struct dentry *ovl_workdir(struct dentry *dentry)
  34 {
  35         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
  36         return ofs->workdir;
  37 }
  38
  39 const struct cred *ovl_override_creds(struct super_block *sb)
  40 {
  41         struct ovl_fs *ofs = sb->s_fs_info;
  42
  43         return override_creds(ofs->creator_cred);
  44 }
  45
  46 struct super_block *ovl_same_sb(struct super_block *sb)
  47 {
  48         struct ovl_fs *ofs = sb->s_fs_info;
  49
  50         return ofs->same_sb;
  51 }
  52
  53 bool ovl_can_decode_fh(struct super_block *sb)
  54 {
  55         return (sb->s_export_op && sb->s_export_op->fh_to_dentry &&
  56                 !uuid_is_null(&sb->s_uuid));
  57 }
  58
  59 struct dentry *ovl_indexdir(struct super_block *sb)
  60 {
  61         struct ovl_fs *ofs = sb->s_fs_info;
  62
  63         return ofs->indexdir;
  64 }
  65
  66 struct ovl_entry *ovl_alloc_entry(unsigned int numlower)
  67 {
  68         size_t size = offsetof(struct ovl_entry, lowerstack[numlower]);
  69         struct ovl_entry *oe = kzalloc(size, GFP_KERNEL);
  70
  71         if (oe)
  72                 oe->numlower = numlower;
  73
  74         return oe;
  75 }
  76
  77 bool ovl_dentry_remote(struct dentry *dentry)
  78 {
  79         return dentry->d_flags &
  80                 (DCACHE_OP_REVALIDATE | DCACHE_OP_WEAK_REVALIDATE |
  81                  DCACHE_OP_REAL);
  82 }
  83
  84 bool ovl_dentry_weird(struct dentry *dentry)
  85 {
  86         return dentry->d_flags & (DCACHE_NEED_AUTOMOUNT |
  87                                   DCACHE_MANAGE_TRANSIT |
  88                                   DCACHE_OP_HASH |
  89                                   DCACHE_OP_COMPARE);
  90 }
  91
  92 enum ovl_path_type ovl_path_type(struct dentry *dentry)
  93 {
  94         struct ovl_entry *oe = dentry->d_fsdata;
  95         enum ovl_path_type type = 0;
  96
  97         if (ovl_dentry_upper(dentry)) {
  98                 type = __OVL_PATH_UPPER;
  99
 100                 /*
 101                  * Non-dir dentry can hold lower dentry of its copy up origin.
 102                  */
 103                 if (oe->numlower) {
 104                         type |= __OVL_PATH_ORIGIN;
 105                         if (d_is_dir(dentry))
 106                                 type |= __OVL_PATH_MERGE;
 107                 }
 108         } else {
 109                 if (oe->numlower > 1)
 110                         type |= __OVL_PATH_MERGE;
 111         }
 112         return type;
 113 }
 114
 115 void ovl_path_upper(struct dentry *dentry, struct path *path)
 116 {
 117         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 118
 119         path->mnt = ofs->upper_mnt;
 120         path->dentry = ovl_dentry_upper(dentry);
 121 }
 122
 123 void ovl_path_lower(struct dentry *dentry, struct path *path)
 124 {
 125         struct ovl_entry *oe = dentry->d_fsdata;
 126
 127         if (oe->numlower) {
 128                 path->mnt = oe->lowerstack[0].layer->mnt;
 129                 path->dentry = oe->lowerstack[0].dentry;
 130         } else {
 131                 *path = (struct path) { };
 132         }
 133 }
 134
 135 enum ovl_path_type ovl_path_real(struct dentry *dentry, struct path *path)
 136 {
 137         enum ovl_path_type type = ovl_path_type(dentry);
 138
 139         if (!OVL_TYPE_UPPER(type))
 140                 ovl_path_lower(dentry, path);
 141         else
 142                 ovl_path_upper(dentry, path);
 143
 144         return type;
 145 }
 146
 147 struct dentry *ovl_dentry_upper(struct dentry *dentry)
 148 {
 149         return ovl_upperdentry_dereference(OVL_I(d_inode(dentry)));
 150 }
 151
 152 struct dentry *ovl_dentry_lower(struct dentry *dentry)
 153 {
 154         struct ovl_entry *oe = dentry->d_fsdata;
 155
 156         return oe->numlower ? oe->lowerstack[0].dentry : NULL;
 157 }
 158
 159 struct dentry *ovl_dentry_real(struct dentry *dentry)
 160 {
 161         return ovl_dentry_upper(dentry) ?: ovl_dentry_lower(dentry);
 162 }
 163
 164 struct dentry *ovl_i_dentry_upper(struct inode *inode)
 165 {
 166         return ovl_upperdentry_dereference(OVL_I(inode));
 167 }
 168
 169 struct inode *ovl_inode_upper(struct inode *inode)
 170 {
 171         struct dentry *upperdentry = ovl_i_dentry_upper(inode);
 172
 173         return upperdentry ? d_inode(upperdentry) : NULL;
 174 }
 175
 176 struct inode *ovl_inode_lower(struct inode *inode)
 177 {
 178         return OVL_I(inode)->lower;
 179 }
 180
 181 struct inode *ovl_inode_real(struct inode *inode)
 182 {
 183         return ovl_inode_upper(inode) ?: ovl_inode_lower(inode);
 184 }
 185
 186
 187 struct ovl_dir_cache *ovl_dir_cache(struct inode *inode)
 188 {
 189         return OVL_I(inode)->cache;
 190 }
 191
 192 void ovl_set_dir_cache(struct inode *inode, struct ovl_dir_cache *cache)
 193 {
 194         OVL_I(inode)->cache = cache;
 195 }
 196
 197 bool ovl_dentry_is_opaque(struct dentry *dentry)
 198 {
 199         struct ovl_entry *oe = dentry->d_fsdata;
 200         return oe->opaque;
 201 }
 202
 203 bool ovl_dentry_is_whiteout(struct dentry *dentry)
 204 {
 205         return !dentry->d_inode && ovl_dentry_is_opaque(dentry);
 206 }
 207
 208 void ovl_dentry_set_opaque(struct dentry *dentry)
 209 {
 210         struct ovl_entry *oe = dentry->d_fsdata;
 211
 212         oe->opaque = true;
 213 }
 214
 215 /*
 216  * For hard links it's possible for ovl_dentry_upper() to return positive, while
 217  * there's no actual upper alias for the inode.  Copy up code needs to know
 218  * about the existence of the upper alias, so it can't use ovl_dentry_upper().
 219  */
 220 bool ovl_dentry_has_upper_alias(struct dentry *dentry)
 221 {
 222         struct ovl_entry *oe = dentry->d_fsdata;
 223
 224         return oe->has_upper;
 225 }
 226
 227 void ovl_dentry_set_upper_alias(struct dentry *dentry)
 228 {
 229         struct ovl_entry *oe = dentry->d_fsdata;
 230
 231         oe->has_upper = true;
 232 }
 233
 234 bool ovl_redirect_dir(struct super_block *sb)
 235 {
 236         struct ovl_fs *ofs = sb->s_fs_info;
 237
 238         return ofs->config.redirect_dir && !ofs->noxattr;
 239 }
 240
 241 const char *ovl_dentry_get_redirect(struct dentry *dentry)
 242 {
 243         return OVL_I(d_inode(dentry))->redirect;
 244 }
 245
 246 void ovl_dentry_set_redirect(struct dentry *dentry, const char *redirect)
 247 {
 248         struct ovl_inode *oi = OVL_I(d_inode(dentry));
 249
 250         kfree(oi->redirect);
 251         oi->redirect = redirect;
 252 }
 253
 254 void ovl_inode_init(struct inode *inode, struct dentry *upperdentry,
 255                     struct dentry *lowerdentry)
 256 {
 257         if (upperdentry)
 258                 OVL_I(inode)->__upperdentry = upperdentry;
 259         if (lowerdentry)
 260                 OVL_I(inode)->lower = igrab(d_inode(lowerdentry));
 261
 262         ovl_copyattr(d_inode(upperdentry ?: lowerdentry), inode);
 263 }
 264
 265 void ovl_inode_update(struct inode *inode, struct dentry *upperdentry)
 266 {
 267         struct inode *upperinode = d_inode(upperdentry);
 268
 269         WARN_ON(OVL_I(inode)->__upperdentry);
 270
 271         /*
 272          * Make sure upperdentry is consistent before making it visible
 273          */
 274         smp_wmb();
 275         OVL_I(inode)->__upperdentry = upperdentry;
 276         if (inode_unhashed(inode)) {
 277                 inode->i_private = upperinode;
 278                 __insert_inode_hash(inode, (unsigned long) upperinode);
 279         }
 280 }
 281
 282 void ovl_dentry_version_inc(struct dentry *dentry, bool impurity)
 283 {
 284         struct inode *inode = d_inode(dentry);
 285
 286         WARN_ON(!inode_is_locked(inode));
 287         /*
 288          * Version is used by readdir code to keep cache consistent.  For merge
 289          * dirs all changes need to be noted.  For non-merge dirs, cache only
 290          * contains impure (ones which have been copied up and have origins)
 291          * entries, so only need to note changes to impure entries.
 292          */
 293         if (OVL_TYPE_MERGE(ovl_path_type(dentry)) || impurity)
 294                 OVL_I(inode)->version++;
 295 }
 296
 297 u64 ovl_dentry_version_get(struct dentry *dentry)
 298 {
 299         struct inode *inode = d_inode(dentry);
 300
 301         WARN_ON(!inode_is_locked(inode));
 302         return OVL_I(inode)->version;
 303 }
 304
 305 bool ovl_is_whiteout(struct dentry *dentry)
 306 {
 307         struct inode *inode = dentry->d_inode;
 308
 309         return inode && IS_WHITEOUT(inode);
 310 }
 311
 312 struct file *ovl_path_open(struct path *path, int flags)
 313 {
 314         return dentry_open(path, flags | O_NOATIME, current_cred());
 315 }
 316
 317 int ovl_copy_up_start(struct dentry *dentry)
 318 {
 319         struct ovl_inode *oi = OVL_I(d_inode(dentry));
 320         int err;
 321
 322         err = mutex_lock_interruptible(&oi->lock);
 323         if (!err && ovl_dentry_has_upper_alias(dentry)) {
 324                 err = 1; /* Already copied up */
 325                 mutex_unlock(&oi->lock);
 326         }
 327
 328         return err;
 329 }
 330
 331 void ovl_copy_up_end(struct dentry *dentry)
 332 {
 333         mutex_unlock(&OVL_I(d_inode(dentry))->lock);
 334 }
 335
 336 bool ovl_check_origin_xattr(struct dentry *dentry)
 337 {
 338         int res;
 339
 340         res = vfs_getxattr(dentry, OVL_XATTR_ORIGIN, NULL, 0);
 341
 342         /* Zero size value means "copied up but origin unknown" */
 343         if (res >= 0)
 344                 return true;
 345
 346         return false;
 347 }
 348
 349 bool ovl_check_dir_xattr(struct dentry *dentry, const char *name)
 350 {
 351         int res;
 352         char val;
 353
 354         if (!d_is_dir(dentry))
 355                 return false;
 356
 357         res = vfs_getxattr(dentry, name, &val, 1);
 358         if (res == 1 && val == 'y')
 359                 return true;
 360
 361         return false;
 362 }
 363
 364 int ovl_check_setxattr(struct dentry *dentry, struct dentry *upperdentry,
 365                        const char *name, const void *value, size_t size,
 366                        int xerr)
 367 {
 368         int err;
 369         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 370
 371         if (ofs->noxattr)
 372                 return xerr;
 373
 374         err = ovl_do_setxattr(upperdentry, name, value, size, 0);
 375
 376         if (err == -EOPNOTSUPP) {
 377                 pr_warn("overlayfs: cannot set %s xattr on upper\n", name);
 378                 ofs->noxattr = true;
 379                 return xerr;
 380         }
 381
 382         return err;
 383 }
 384
 385 int ovl_set_impure(struct dentry *dentry, struct dentry *upperdentry)
 386 {
 387         int err;
 388
 389         if (ovl_test_flag(OVL_IMPURE, d_inode(dentry)))
 390                 return 0;
 391
 392         /*
 393          * Do not fail when upper doesn't support xattrs.
 394          * Upper inodes won't have origin nor redirect xattr anyway.
 395          */
 396         err = ovl_check_setxattr(dentry, upperdentry, OVL_XATTR_IMPURE,
 397                                  "y", 1, 0);
 398         if (!err)
 399                 ovl_set_flag(OVL_IMPURE, d_inode(dentry));
 400
 401         return err;
 402 }
 403
 404 void ovl_set_flag(unsigned long flag, struct inode *inode)
 405 {
 406         set_bit(flag, &OVL_I(inode)->flags);
 407 }
 408
 409 void ovl_clear_flag(unsigned long flag, struct inode *inode)
 410 {
 411         clear_bit(flag, &OVL_I(inode)->flags);
 412 }
 413
 414 bool ovl_test_flag(unsigned long flag, struct inode *inode)
 415 {
 416         return test_bit(flag, &OVL_I(inode)->flags);
 417 }
 418
 419 /**
 420  * Caller must hold a reference to inode to prevent it from being freed while
 421  * it is marked inuse.
 422  */
 423 bool ovl_inuse_trylock(struct dentry *dentry)
 424 {
 425         struct inode *inode = d_inode(dentry);
 426         bool locked = false;
 427
 428         spin_lock(&inode->i_lock);
 429         if (!(inode->i_state & I_OVL_INUSE)) {
 430                 inode->i_state |= I_OVL_INUSE;
 431                 locked = true;
 432         }
 433         spin_unlock(&inode->i_lock);
 434
 435         return locked;
 436 }
 437
 438 void ovl_inuse_unlock(struct dentry *dentry)
 439 {
 440         if (dentry) {
 441                 struct inode *inode = d_inode(dentry);
 442
 443                 spin_lock(&inode->i_lock);
 444                 WARN_ON(!(inode->i_state & I_OVL_INUSE));
 445                 inode->i_state &= ~I_OVL_INUSE;
 446                 spin_unlock(&inode->i_lock);
 447         }
 448 }
 449
 450 /* Caller must hold OVL_I(inode)->lock */
 451 static void ovl_cleanup_index(struct dentry *dentry)
 452 {
 453         struct inode *dir = ovl_indexdir(dentry->d_sb)->d_inode;
 454         struct dentry *lowerdentry = ovl_dentry_lower(dentry);
 455         struct dentry *upperdentry = ovl_dentry_upper(dentry);
 456         struct dentry *index = NULL;
 457         struct inode *inode;
 458         struct qstr name;
 459         int err;
 460
 461         err = ovl_get_index_name(lowerdentry, &name);
 462         if (err)
 463                 goto fail;
 464
 465         inode = d_inode(upperdentry);
 466         if (inode->i_nlink != 1) {
 467                 pr_warn_ratelimited("overlayfs: cleanup linked index (%pd2, ino=%lu, nlink=%u)\n",
 468                                     upperdentry, inode->i_ino, inode->i_nlink);
 469                 /*
 470                  * We either have a bug with persistent union nlink or a lower
 471                  * hardlink was added while overlay is mounted. Adding a lower
 472                  * hardlink and then unlinking all overlay hardlinks would drop
 473                  * overlay nlink to zero before all upper inodes are unlinked.
 474                  * As a safety measure, when that situation is detected, set
 475                  * the overlay nlink to the index inode nlink minus one for the
 476                  * index entry itself.
 477                  */
 478                 set_nlink(d_inode(dentry), inode->i_nlink - 1);
 479                 ovl_set_nlink_upper(dentry);
 480                 goto out;
 481         }
 482
 483         inode_lock_nested(dir, I_MUTEX_PARENT);
 484         /* TODO: whiteout instead of cleanup to block future open by handle */
 485         index = lookup_one_len(name.name, ovl_indexdir(dentry->d_sb), name.len);
 486         err = PTR_ERR(index);
 487         if (!IS_ERR(index))
 488                 err = ovl_cleanup(dir, index);
 489         else
 490                 index = NULL;
 491
 492         inode_unlock(dir);
 493         if (err)
 494                 goto fail;
 495
 496 out:
 497         dput(index);
 498         return;
 499
 500 fail:
 501         pr_err("overlayfs: cleanup index of '%pd2' failed (%i)\n", dentry, err);
 502         goto out;
 503 }
 504
 505 /*
 506  * Operations that change overlay inode and upper inode nlink need to be
 507  * synchronized with copy up for persistent nlink accounting.
 508  */
 509 int ovl_nlink_start(struct dentry *dentry, bool *locked)
 510 {
 511         struct ovl_inode *oi = OVL_I(d_inode(dentry));
 512         const struct cred *old_cred;
 513         int err;
 514
 515         if (!d_inode(dentry) || d_is_dir(dentry))
 516                 return 0;
 517
 518         /*
 519          * With inodes index is enabled, we store the union overlay nlink
 520          * in an xattr on the index inode. When whiting out lower hardlinks
 521          * we need to decrement the overlay persistent nlink, but before the
 522          * first copy up, we have no upper index inode to store the xattr.
 523          *
 524          * As a workaround, before whiteout/rename over of a lower hardlink,
 525          * copy up to create the upper index. Creating the upper index will
 526          * initialize the overlay nlink, so it could be dropped if unlink
 527          * or rename succeeds.
 528          *
 529          * TODO: implement metadata only index copy up when called with
 530          *       ovl_copy_up_flags(dentry, O_PATH).
 531          */
 532         if (ovl_indexdir(dentry->d_sb) && !ovl_dentry_has_upper_alias(dentry) &&
 533             d_inode(ovl_dentry_lower(dentry))->i_nlink > 1) {
 534                 err = ovl_copy_up(dentry);
 535                 if (err)
 536                         return err;
 537         }
 538
 539         err = mutex_lock_interruptible(&oi->lock);
 540         if (err)
 541                 return err;
 542
 543         if (!ovl_test_flag(OVL_INDEX, d_inode(dentry)))
 544                 goto out;
 545
 546         old_cred = ovl_override_creds(dentry->d_sb);
 547         /*
 548          * The overlay inode nlink should be incremented/decremented IFF the
 549          * upper operation succeeds, along with nlink change of upper inode.
 550          * Therefore, before link/unlink/rename, we store the union nlink
 551          * value relative to the upper inode nlink in an upper inode xattr.
 552          */
 553         err = ovl_set_nlink_upper(dentry);
 554         revert_creds(old_cred);
 555
 556 out:
 557         if (err)
 558                 mutex_unlock(&oi->lock);
 559         else
 560                 *locked = true;
 561
 562         return err;
 563 }
 564
 565 void ovl_nlink_end(struct dentry *dentry, bool locked)
 566 {
 567         if (locked) {
 568                 if (ovl_test_flag(OVL_INDEX, d_inode(dentry)) &&
 569                     d_inode(dentry)->i_nlink == 0) {
 570                         const struct cred *old_cred;
 571
 572                         old_cred = ovl_override_creds(dentry->d_sb);
 573                         ovl_cleanup_index(dentry);
 574                         revert_creds(old_cred);
 575                 }
 576
 577                 mutex_unlock(&OVL_I(d_inode(dentry))->lock);
 578         }
 579 }
 580
 581 int ovl_lock_rename_workdir(struct dentry *workdir, struct dentry *upperdir)
 582 {
 583         /* Workdir should not be the same as upperdir */
 584         if (workdir == upperdir)
 585                 goto err;
 586
 587         /* Workdir should not be subdir of upperdir and vice versa */
 588         if (lock_rename(workdir, upperdir) != NULL)
 589                 goto err_unlock;
 590
 591         return 0;
 592
 593 err_unlock:
 594         unlock_rename(workdir, upperdir);
 595 err:
 596         pr_err("overlayfs: failed to lock workdir+upperdir\n");
 597         return -EIO;
 598 }