1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 3 * 4 * Copyright (c) 2022 The FreeBSD Foundation 5 * 6 * This software was developed by Mark Johnston under sponsorship from 7 * the FreeBSD Foundation. 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions are 11 * met: 12 * 1. Redistributions of source code must retain the above copyright 13 * notice, this list of conditions and the following disclaimer. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in 16 * the documentation and/or other materials provided with the distribution. 17 * 18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 21 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 28 * SUCH DAMAGE. 29 */ 30 31 #include <sys/dirent.h> 32 #include <sys/stat.h> 33 34 #include <assert.h> 35 #include <fcntl.h> 36 #include <string.h> 37 #include <unistd.h> 38 39 #include <util.h> 40 41 #include "makefs.h" 42 #include "zfs.h" 43 44 typedef struct { 45 const char *name; 46 unsigned int id; 47 uint16_t size; 48 sa_bswap_type_t bs; 49 } zfs_sattr_t; 50 51 typedef struct zfs_fs { 52 zfs_objset_t *os; 53 54 /* Offset table for system attributes, indexed by a zpl_attr_t. */ 55 uint16_t *saoffs; 56 size_t sacnt; 57 const zfs_sattr_t *satab; 58 } zfs_fs_t; 59 60 /* 61 * The order of the attributes doesn't matter, this is simply the one hard-coded 62 * by OpenZFS, based on a zdb dump of the SA_REGISTRY table. 63 */ 64 typedef enum zpl_attr { 65 ZPL_ATIME, 66 ZPL_MTIME, 67 ZPL_CTIME, 68 ZPL_CRTIME, 69 ZPL_GEN, 70 ZPL_MODE, 71 ZPL_SIZE, 72 ZPL_PARENT, 73 ZPL_LINKS, 74 ZPL_XATTR, 75 ZPL_RDEV, 76 ZPL_FLAGS, 77 ZPL_UID, 78 ZPL_GID, 79 ZPL_PAD, 80 ZPL_ZNODE_ACL, 81 ZPL_DACL_COUNT, 82 ZPL_SYMLINK, 83 ZPL_SCANSTAMP, 84 ZPL_DACL_ACES, 85 ZPL_DXATTR, 86 ZPL_PROJID, 87 } zpl_attr_t; 88 89 /* 90 * This table must be kept in sync with zpl_attr_layout[] and zpl_attr_t. 91 */ 92 static const zfs_sattr_t zpl_attrs[] = { 93 #define _ZPL_ATTR(n, s, b) { .name = #n, .id = n, .size = s, .bs = b } 94 _ZPL_ATTR(ZPL_ATIME, sizeof(uint64_t) * 2, SA_UINT64_ARRAY), 95 _ZPL_ATTR(ZPL_MTIME, sizeof(uint64_t) * 2, SA_UINT64_ARRAY), 96 _ZPL_ATTR(ZPL_CTIME, sizeof(uint64_t) * 2, SA_UINT64_ARRAY), 97 _ZPL_ATTR(ZPL_CRTIME, sizeof(uint64_t) * 2, SA_UINT64_ARRAY), 98 _ZPL_ATTR(ZPL_GEN, sizeof(uint64_t), SA_UINT64_ARRAY), 99 _ZPL_ATTR(ZPL_MODE, sizeof(uint64_t), SA_UINT64_ARRAY), 100 _ZPL_ATTR(ZPL_SIZE, sizeof(uint64_t), SA_UINT64_ARRAY), 101 _ZPL_ATTR(ZPL_PARENT, sizeof(uint64_t), SA_UINT64_ARRAY), 102 _ZPL_ATTR(ZPL_LINKS, sizeof(uint64_t), SA_UINT64_ARRAY), 103 _ZPL_ATTR(ZPL_XATTR, sizeof(uint64_t), SA_UINT64_ARRAY), 104 _ZPL_ATTR(ZPL_RDEV, sizeof(uint64_t), SA_UINT64_ARRAY), 105 _ZPL_ATTR(ZPL_FLAGS, sizeof(uint64_t), SA_UINT64_ARRAY), 106 _ZPL_ATTR(ZPL_UID, sizeof(uint64_t), SA_UINT64_ARRAY), 107 _ZPL_ATTR(ZPL_GID, sizeof(uint64_t), SA_UINT64_ARRAY), 108 _ZPL_ATTR(ZPL_PAD, sizeof(uint64_t), SA_UINT64_ARRAY), 109 _ZPL_ATTR(ZPL_ZNODE_ACL, 88, SA_UINT64_ARRAY), 110 _ZPL_ATTR(ZPL_DACL_COUNT, sizeof(uint64_t), SA_UINT64_ARRAY), 111 _ZPL_ATTR(ZPL_SYMLINK, 0, SA_UINT8_ARRAY), 112 _ZPL_ATTR(ZPL_SCANSTAMP, sizeof(uint64_t) * 4, SA_UINT8_ARRAY), 113 _ZPL_ATTR(ZPL_DACL_ACES, 0, SA_ACL), 114 _ZPL_ATTR(ZPL_DXATTR, 0, SA_UINT8_ARRAY), 115 _ZPL_ATTR(ZPL_PROJID, sizeof(uint64_t), SA_UINT64_ARRAY), 116 #undef ZPL_ATTR 117 }; 118 119 /* 120 * This layout matches that of a filesystem created using OpenZFS on FreeBSD. 121 * It need not match in general, but FreeBSD's loader doesn't bother parsing the 122 * layout and just hard-codes attribute offsets. 123 */ 124 static const sa_attr_type_t zpl_attr_layout[] = { 125 ZPL_MODE, 126 ZPL_SIZE, 127 ZPL_GEN, 128 ZPL_UID, 129 ZPL_GID, 130 ZPL_PARENT, 131 ZPL_FLAGS, 132 ZPL_ATIME, 133 ZPL_MTIME, 134 ZPL_CTIME, 135 ZPL_CRTIME, 136 ZPL_LINKS, 137 ZPL_DACL_COUNT, 138 ZPL_DACL_ACES, 139 ZPL_SYMLINK, 140 }; 141 142 /* 143 * Keys for the ZPL attribute tables in the SA layout ZAP. The first two 144 * indices are reserved for legacy attribute encoding. 145 */ 146 #define SA_LAYOUT_INDEX_DEFAULT 2 147 #define SA_LAYOUT_INDEX_SYMLINK 3 148 149 struct fs_populate_dir { 150 SLIST_ENTRY(fs_populate_dir) next; 151 int dirfd; 152 uint64_t objid; 153 zfs_zap_t *zap; 154 }; 155 156 struct fs_populate_arg { 157 zfs_opt_t *zfs; 158 zfs_fs_t *fs; /* owning filesystem */ 159 int dirfd; /* current directory fd */ 160 uint64_t rootdirid; /* root directory dnode ID */ 161 SLIST_HEAD(, fs_populate_dir) dirs; /* stack of directories */ 162 }; 163 164 static void fs_build_one(zfs_opt_t *, zfs_dsl_dir_t *, fsnode *, int); 165 166 static bool 167 fsnode_isroot(const fsnode *cur) 168 { 169 return (strcmp(cur->name, ".") == 0); 170 } 171 172 /* 173 * Visit each node in a directory hierarchy, in pre-order depth-first order. 174 */ 175 static void 176 fsnode_foreach(fsnode *root, int (*cb)(fsnode *, void *), void *arg) 177 { 178 assert(root->type == S_IFDIR); 179 180 for (fsnode *cur = root; cur != NULL; cur = cur->next) { 181 assert(cur->type == S_IFREG || cur->type == S_IFDIR || 182 cur->type == S_IFLNK); 183 184 if (cb(cur, arg) == 0) 185 continue; 186 if (cur->type == S_IFDIR && cur->child != NULL) 187 fsnode_foreach(cur->child, cb, arg); 188 } 189 } 190 191 static void 192 fs_populate_dirent(struct fs_populate_arg *arg, fsnode *cur, uint64_t dnid) 193 { 194 struct fs_populate_dir *dir; 195 uint64_t type; 196 197 switch (cur->type) { 198 case S_IFREG: 199 type = DT_REG; 200 break; 201 case S_IFDIR: 202 type = DT_DIR; 203 break; 204 case S_IFLNK: 205 type = DT_LNK; 206 break; 207 default: 208 assert(0); 209 } 210 211 dir = SLIST_FIRST(&arg->dirs); 212 zap_add_uint64(dir->zap, cur->name, ZFS_DIRENT_MAKE(type, dnid)); 213 } 214 215 static void 216 fs_populate_attr(zfs_fs_t *fs, char *attrbuf, const void *val, uint16_t ind, 217 size_t *szp) 218 { 219 assert(ind < fs->sacnt); 220 assert(fs->saoffs[ind] != 0xffff); 221 222 memcpy(attrbuf + fs->saoffs[ind], val, fs->satab[ind].size); 223 *szp += fs->satab[ind].size; 224 } 225 226 static void 227 fs_populate_varszattr(zfs_fs_t *fs, char *attrbuf, const void *val, 228 size_t valsz, size_t varoff, uint16_t ind, size_t *szp) 229 { 230 assert(ind < fs->sacnt); 231 assert(fs->saoffs[ind] != 0xffff); 232 assert(fs->satab[ind].size == 0); 233 234 memcpy(attrbuf + fs->saoffs[ind] + varoff, val, valsz); 235 *szp += valsz; 236 } 237 238 static void 239 fs_populate_sattrs(struct fs_populate_arg *arg, const fsnode *cur, 240 dnode_phys_t *dnode) 241 { 242 char target[PATH_MAX]; 243 zfs_fs_t *fs; 244 zfs_ace_hdr_t aces[3]; 245 struct stat *sb; 246 sa_hdr_phys_t *sahdr; 247 uint64_t daclcount, flags, gen, gid, links, mode, parent, objsize, uid; 248 char *attrbuf; 249 size_t bonussz, hdrsz; 250 int layout; 251 252 assert(dnode->dn_bonustype == DMU_OT_SA); 253 assert(dnode->dn_nblkptr == 1); 254 255 fs = arg->fs; 256 sb = &cur->inode->st; 257 258 switch (cur->type) { 259 case S_IFREG: 260 layout = SA_LAYOUT_INDEX_DEFAULT; 261 links = cur->inode->nlink; 262 objsize = sb->st_size; 263 parent = SLIST_FIRST(&arg->dirs)->objid; 264 break; 265 case S_IFDIR: 266 layout = SA_LAYOUT_INDEX_DEFAULT; 267 links = 1; /* .. */ 268 objsize = 1; /* .. */ 269 270 /* 271 * The size of a ZPL directory is the number of entries 272 * (including "." and ".."), and the link count is the number of 273 * entries which are directories (including "." and ".."). 274 */ 275 for (fsnode *c = fsnode_isroot(cur) ? cur->next : cur->child; 276 c != NULL; c = c->next) { 277 if (c->type == S_IFDIR) 278 links++; 279 objsize++; 280 } 281 282 /* The root directory is its own parent. */ 283 parent = SLIST_EMPTY(&arg->dirs) ? 284 arg->rootdirid : SLIST_FIRST(&arg->dirs)->objid; 285 break; 286 case S_IFLNK: { 287 ssize_t n; 288 289 if ((n = readlinkat(SLIST_FIRST(&arg->dirs)->dirfd, cur->name, 290 target, sizeof(target) - 1)) == -1) 291 err(1, "readlinkat(%s)", cur->name); 292 target[n] = '\0'; 293 294 layout = SA_LAYOUT_INDEX_SYMLINK; 295 links = 1; 296 objsize = strlen(target); 297 parent = SLIST_FIRST(&arg->dirs)->objid; 298 break; 299 } 300 default: 301 assert(0); 302 } 303 304 daclcount = nitems(aces); 305 flags = ZFS_ACL_TRIVIAL | ZFS_ACL_AUTO_INHERIT | ZFS_NO_EXECS_DENIED | 306 ZFS_ARCHIVE | ZFS_AV_MODIFIED; /* XXX-MJ */ 307 gen = 1; 308 gid = sb->st_gid; 309 mode = sb->st_mode; 310 uid = sb->st_uid; 311 312 memset(aces, 0, sizeof(aces)); 313 aces[0].z_flags = ACE_OWNER; 314 aces[0].z_type = ACE_ACCESS_ALLOWED_ACE_TYPE; 315 aces[0].z_access_mask = ACE_WRITE_ATTRIBUTES | ACE_WRITE_OWNER | 316 ACE_WRITE_ACL | ACE_WRITE_NAMED_ATTRS | ACE_READ_ACL | 317 ACE_READ_ATTRIBUTES | ACE_READ_NAMED_ATTRS | ACE_SYNCHRONIZE; 318 if ((mode & S_IRUSR) != 0) 319 aces[0].z_access_mask |= ACE_READ_DATA; 320 if ((mode & S_IWUSR) != 0) 321 aces[0].z_access_mask |= ACE_WRITE_DATA | ACE_APPEND_DATA; 322 if ((mode & S_IXUSR) != 0) 323 aces[0].z_access_mask |= ACE_EXECUTE; 324 325 aces[1].z_flags = ACE_GROUP | ACE_IDENTIFIER_GROUP; 326 aces[1].z_type = ACE_ACCESS_ALLOWED_ACE_TYPE; 327 aces[1].z_access_mask = ACE_READ_ACL | ACE_READ_ATTRIBUTES | 328 ACE_READ_NAMED_ATTRS | ACE_SYNCHRONIZE; 329 if ((mode & S_IRGRP) != 0) 330 aces[1].z_access_mask |= ACE_READ_DATA; 331 if ((mode & S_IWGRP) != 0) 332 aces[1].z_access_mask |= ACE_WRITE_DATA | ACE_APPEND_DATA; 333 if ((mode & S_IXGRP) != 0) 334 aces[1].z_access_mask |= ACE_EXECUTE; 335 336 aces[2].z_flags = ACE_EVERYONE; 337 aces[2].z_type = ACE_ACCESS_ALLOWED_ACE_TYPE; 338 aces[2].z_access_mask = ACE_READ_ACL | ACE_READ_ATTRIBUTES | 339 ACE_READ_NAMED_ATTRS | ACE_SYNCHRONIZE; 340 if ((mode & S_IROTH) != 0) 341 aces[2].z_access_mask |= ACE_READ_DATA; 342 if ((mode & S_IWOTH) != 0) 343 aces[2].z_access_mask |= ACE_WRITE_DATA | ACE_APPEND_DATA; 344 if ((mode & S_IXOTH) != 0) 345 aces[2].z_access_mask |= ACE_EXECUTE; 346 347 switch (layout) { 348 case SA_LAYOUT_INDEX_DEFAULT: 349 /* At most one variable-length attribute. */ 350 hdrsz = sizeof(uint64_t); 351 break; 352 case SA_LAYOUT_INDEX_SYMLINK: 353 /* At most five variable-length attributes. */ 354 hdrsz = sizeof(uint64_t) * 2; 355 break; 356 default: 357 assert(0); 358 } 359 360 sahdr = (sa_hdr_phys_t *)DN_BONUS(dnode); 361 sahdr->sa_magic = SA_MAGIC; 362 SA_HDR_LAYOUT_INFO_ENCODE(sahdr->sa_layout_info, layout, hdrsz); 363 364 bonussz = SA_HDR_SIZE(sahdr); 365 attrbuf = (char *)sahdr + SA_HDR_SIZE(sahdr); 366 367 fs_populate_attr(fs, attrbuf, &daclcount, ZPL_DACL_COUNT, &bonussz); 368 fs_populate_attr(fs, attrbuf, &flags, ZPL_FLAGS, &bonussz); 369 fs_populate_attr(fs, attrbuf, &gen, ZPL_GEN, &bonussz); 370 fs_populate_attr(fs, attrbuf, &gid, ZPL_GID, &bonussz); 371 fs_populate_attr(fs, attrbuf, &links, ZPL_LINKS, &bonussz); 372 fs_populate_attr(fs, attrbuf, &mode, ZPL_MODE, &bonussz); 373 fs_populate_attr(fs, attrbuf, &parent, ZPL_PARENT, &bonussz); 374 fs_populate_attr(fs, attrbuf, &objsize, ZPL_SIZE, &bonussz); 375 fs_populate_attr(fs, attrbuf, &uid, ZPL_UID, &bonussz); 376 377 /* 378 * We deliberately set atime = mtime here to ensure that images are 379 * reproducible. 380 */ 381 assert(sizeof(sb->st_mtim) == fs->satab[ZPL_ATIME].size); 382 fs_populate_attr(fs, attrbuf, &sb->st_mtim, ZPL_ATIME, &bonussz); 383 assert(sizeof(sb->st_ctim) == fs->satab[ZPL_CTIME].size); 384 fs_populate_attr(fs, attrbuf, &sb->st_ctim, ZPL_CTIME, &bonussz); 385 assert(sizeof(sb->st_mtim) == fs->satab[ZPL_MTIME].size); 386 fs_populate_attr(fs, attrbuf, &sb->st_mtim, ZPL_MTIME, &bonussz); 387 assert(sizeof(sb->st_birthtim) == fs->satab[ZPL_CRTIME].size); 388 fs_populate_attr(fs, attrbuf, &sb->st_birthtim, ZPL_CRTIME, &bonussz); 389 390 fs_populate_varszattr(fs, attrbuf, aces, sizeof(aces), 0, 391 ZPL_DACL_ACES, &bonussz); 392 sahdr->sa_lengths[0] = sizeof(aces); 393 394 if (cur->type == S_IFLNK) { 395 assert(layout == SA_LAYOUT_INDEX_SYMLINK); 396 /* Need to use a spill block pointer if the target is long. */ 397 assert(bonussz + objsize <= DN_OLD_MAX_BONUSLEN); 398 fs_populate_varszattr(fs, attrbuf, target, objsize, 399 sahdr->sa_lengths[0], ZPL_SYMLINK, &bonussz); 400 sahdr->sa_lengths[1] = (uint16_t)objsize; 401 } 402 403 dnode->dn_bonuslen = bonussz; 404 } 405 406 static void 407 fs_populate_file(fsnode *cur, struct fs_populate_arg *arg) 408 { 409 struct dnode_cursor *c; 410 dnode_phys_t *dnode; 411 zfs_opt_t *zfs; 412 char *buf; 413 uint64_t dnid; 414 ssize_t n; 415 size_t bufsz; 416 off_t size, target; 417 int fd; 418 419 assert(cur->type == S_IFREG); 420 assert((cur->inode->flags & FI_ROOT) == 0); 421 422 zfs = arg->zfs; 423 424 assert(cur->inode->ino != 0); 425 if ((cur->inode->flags & FI_ALLOCATED) != 0) { 426 /* 427 * This is a hard link of an existing file. 428 * 429 * XXX-MJ need to check whether it crosses datasets, add a test 430 * case for that 431 */ 432 fs_populate_dirent(arg, cur, cur->inode->ino); 433 return; 434 } 435 436 dnode = objset_dnode_bonus_alloc(arg->fs->os, 437 DMU_OT_PLAIN_FILE_CONTENTS, DMU_OT_SA, 0, &dnid); 438 cur->inode->ino = dnid; 439 cur->inode->flags |= FI_ALLOCATED; 440 441 fd = openat(SLIST_FIRST(&arg->dirs)->dirfd, cur->name, O_RDONLY); 442 if (fd == -1) 443 err(1, "openat(%s)", cur->name); 444 445 buf = zfs->filebuf; 446 bufsz = sizeof(zfs->filebuf); 447 size = cur->inode->st.st_size; 448 c = dnode_cursor_init(zfs, arg->fs->os, dnode, size, 0); 449 for (off_t foff = 0; foff < size; foff += target) { 450 off_t loc, sofar; 451 452 /* 453 * Fill up our buffer, handling partial reads. 454 * 455 * It might be profitable to use copy_file_range(2) here. 456 */ 457 sofar = 0; 458 target = MIN(size - foff, (off_t)bufsz); 459 do { 460 n = read(fd, buf + sofar, target); 461 if (n < 0) 462 err(1, "reading from '%s'", cur->name); 463 if (n == 0) 464 errx(1, "unexpected EOF reading '%s'", 465 cur->name); 466 sofar += n; 467 } while (sofar < target); 468 469 if (target < (off_t)bufsz) 470 memset(buf + target, 0, bufsz - target); 471 472 loc = objset_space_alloc(zfs, arg->fs->os, &target); 473 vdev_pwrite_dnode_indir(zfs, dnode, 0, 1, buf, target, loc, 474 dnode_cursor_next(zfs, c, foff)); 475 } 476 if (close(fd) != 0) 477 err(1, "close"); 478 dnode_cursor_finish(zfs, c); 479 480 fs_populate_sattrs(arg, cur, dnode); 481 fs_populate_dirent(arg, cur, dnid); 482 } 483 484 static void 485 fs_populate_dir(fsnode *cur, struct fs_populate_arg *arg) 486 { 487 dnode_phys_t *dnode; 488 zfs_objset_t *os; 489 uint64_t dnid; 490 int dirfd; 491 492 assert(cur->type == S_IFDIR); 493 assert((cur->inode->flags & FI_ALLOCATED) == 0); 494 495 os = arg->fs->os; 496 497 dnode = objset_dnode_bonus_alloc(os, DMU_OT_DIRECTORY_CONTENTS, 498 DMU_OT_SA, 0, &dnid); 499 500 /* 501 * Add an entry to the parent directory and open this directory. 502 */ 503 if (!SLIST_EMPTY(&arg->dirs)) { 504 fs_populate_dirent(arg, cur, dnid); 505 dirfd = openat(SLIST_FIRST(&arg->dirs)->dirfd, cur->name, 506 O_DIRECTORY); 507 if (dirfd < 0) 508 err(1, "open(%s)", cur->name); 509 } else { 510 arg->rootdirid = dnid; 511 dirfd = arg->dirfd; 512 } 513 514 /* 515 * Set ZPL attributes. 516 */ 517 fs_populate_sattrs(arg, cur, dnode); 518 519 /* 520 * If this is a root directory, then its children belong to a different 521 * dataset and this directory remains empty in the current objset. 522 */ 523 if ((cur->inode->flags & FI_ROOT) == 0) { 524 struct fs_populate_dir *dir; 525 526 dir = ecalloc(1, sizeof(*dir)); 527 dir->dirfd = dirfd; 528 dir->objid = dnid; 529 dir->zap = zap_alloc(os, dnode); 530 SLIST_INSERT_HEAD(&arg->dirs, dir, next); 531 } else { 532 zap_write(arg->zfs, zap_alloc(os, dnode)); 533 fs_build_one(arg->zfs, cur->inode->param, cur->child, dirfd); 534 } 535 } 536 537 static void 538 fs_populate_symlink(fsnode *cur, struct fs_populate_arg *arg) 539 { 540 dnode_phys_t *dnode; 541 uint64_t dnid; 542 543 assert(cur->type == S_IFLNK); 544 assert((cur->inode->flags & (FI_ALLOCATED | FI_ROOT)) == 0); 545 546 dnode = objset_dnode_bonus_alloc(arg->fs->os, 547 DMU_OT_PLAIN_FILE_CONTENTS, DMU_OT_SA, 0, &dnid); 548 549 fs_populate_dirent(arg, cur, dnid); 550 551 fs_populate_sattrs(arg, cur, dnode); 552 } 553 554 static int 555 fs_foreach_populate(fsnode *cur, void *_arg) 556 { 557 struct fs_populate_arg *arg; 558 struct fs_populate_dir *dir; 559 int ret; 560 561 arg = _arg; 562 switch (cur->type) { 563 case S_IFREG: 564 fs_populate_file(cur, arg); 565 break; 566 case S_IFDIR: 567 if (fsnode_isroot(cur)) 568 break; 569 fs_populate_dir(cur, arg); 570 break; 571 case S_IFLNK: 572 fs_populate_symlink(cur, arg); 573 break; 574 default: 575 assert(0); 576 } 577 578 ret = (cur->inode->flags & FI_ROOT) != 0 ? 0 : 1; 579 580 if (cur->next == NULL && 581 (cur->child == NULL || (cur->inode->flags & FI_ROOT) != 0)) { 582 /* 583 * We reached a terminal node in a subtree. Walk back up and 584 * write out directories. We're done once we hit the root of a 585 * dataset or find a level where we're not on the edge of the 586 * tree. 587 */ 588 do { 589 dir = SLIST_FIRST(&arg->dirs); 590 SLIST_REMOVE_HEAD(&arg->dirs, next); 591 zap_write(arg->zfs, dir->zap); 592 if (dir->dirfd != -1 && close(dir->dirfd) != 0) 593 err(1, "close"); 594 free(dir); 595 cur = cur->parent; 596 } while (cur != NULL && cur->next == NULL && 597 (cur->inode->flags & FI_ROOT) == 0); 598 } 599 600 return (ret); 601 } 602 603 static void 604 fs_add_zpl_attr_layout(zfs_zap_t *zap, unsigned int index, 605 const sa_attr_type_t layout[], size_t sacnt) 606 { 607 char ti[16]; 608 609 assert(sizeof(layout[0]) == 2); 610 611 snprintf(ti, sizeof(ti), "%u", index); 612 zap_add(zap, ti, sizeof(sa_attr_type_t), sacnt, 613 (const uint8_t *)layout); 614 } 615 616 /* 617 * Initialize system attribute tables. 618 * 619 * There are two elements to this. First, we write the zpl_attrs[] and 620 * zpl_attr_layout[] tables to disk. Then we create a lookup table which 621 * allows us to set file attributes quickly. 622 */ 623 static uint64_t 624 fs_set_zpl_attrs(zfs_opt_t *zfs, zfs_fs_t *fs) 625 { 626 zfs_zap_t *sazap, *salzap, *sarzap; 627 zfs_objset_t *os; 628 dnode_phys_t *saobj, *salobj, *sarobj; 629 uint64_t saobjid, salobjid, sarobjid; 630 uint16_t offset; 631 632 os = fs->os; 633 634 /* 635 * The on-disk tables are stored in two ZAP objects, the registry object 636 * and the layout object. Individual attributes are described by 637 * entries in the registry object; for example, the value for the 638 * "ZPL_SIZE" key gives the size and encoding of the ZPL_SIZE attribute. 639 * The attributes of a file are ordered according to one of the layouts 640 * defined in the layout object. The master node object is simply used 641 * to locate the registry and layout objects. 642 */ 643 saobj = objset_dnode_alloc(os, DMU_OT_SA_MASTER_NODE, &saobjid); 644 salobj = objset_dnode_alloc(os, DMU_OT_SA_ATTR_LAYOUTS, &salobjid); 645 sarobj = objset_dnode_alloc(os, DMU_OT_SA_ATTR_REGISTRATION, &sarobjid); 646 647 sarzap = zap_alloc(os, sarobj); 648 for (size_t i = 0; i < nitems(zpl_attrs); i++) { 649 const zfs_sattr_t *sa; 650 uint64_t attr; 651 652 attr = 0; 653 sa = &zpl_attrs[i]; 654 SA_ATTR_ENCODE(attr, (uint64_t)i, sa->size, sa->bs); 655 zap_add_uint64(sarzap, sa->name, attr); 656 } 657 zap_write(zfs, sarzap); 658 659 /* 660 * Layouts are arrays of indices into the registry. We define two 661 * layouts for use by the ZPL, one for non-symlinks and one for 662 * symlinks. They are identical except that the symlink layout includes 663 * ZPL_SYMLINK as its final attribute. 664 */ 665 salzap = zap_alloc(os, salobj); 666 assert(zpl_attr_layout[nitems(zpl_attr_layout) - 1] == ZPL_SYMLINK); 667 fs_add_zpl_attr_layout(salzap, SA_LAYOUT_INDEX_DEFAULT, 668 zpl_attr_layout, nitems(zpl_attr_layout) - 1); 669 fs_add_zpl_attr_layout(salzap, SA_LAYOUT_INDEX_SYMLINK, 670 zpl_attr_layout, nitems(zpl_attr_layout)); 671 zap_write(zfs, salzap); 672 673 sazap = zap_alloc(os, saobj); 674 zap_add_uint64(sazap, SA_LAYOUTS, salobjid); 675 zap_add_uint64(sazap, SA_REGISTRY, sarobjid); 676 zap_write(zfs, sazap); 677 678 /* Sanity check. */ 679 for (size_t i = 0; i < nitems(zpl_attrs); i++) 680 assert(i == zpl_attrs[i].id); 681 682 /* 683 * Build the offset table used when setting file attributes. File 684 * attributes are stored in the object's bonus buffer; this table 685 * provides the buffer offset of attributes referenced by the layout 686 * table. 687 */ 688 fs->sacnt = nitems(zpl_attrs); 689 fs->saoffs = ecalloc(fs->sacnt, sizeof(*fs->saoffs)); 690 for (size_t i = 0; i < fs->sacnt; i++) 691 fs->saoffs[i] = 0xffff; 692 offset = 0; 693 for (size_t i = 0; i < nitems(zpl_attr_layout); i++) { 694 uint16_t size; 695 696 assert(zpl_attr_layout[i] < fs->sacnt); 697 698 fs->saoffs[zpl_attr_layout[i]] = offset; 699 size = zpl_attrs[zpl_attr_layout[i]].size; 700 offset += size; 701 } 702 fs->satab = zpl_attrs; 703 704 return (saobjid); 705 } 706 707 static void 708 fs_layout_one(zfs_opt_t *zfs, zfs_dsl_dir_t *dsldir, void *arg) 709 { 710 char *mountpoint, *origmountpoint, *name, *next; 711 fsnode *cur, *root; 712 uint64_t canmount; 713 714 if (!dsl_dir_has_dataset(dsldir)) 715 return; 716 717 if (dsl_dir_get_canmount(dsldir, &canmount) == 0 && canmount == 0) 718 return; 719 mountpoint = dsl_dir_get_mountpoint(zfs, dsldir); 720 if (mountpoint == NULL) 721 return; 722 723 /* 724 * If we were asked to specify a bootfs, set it here. 725 */ 726 if (zfs->bootfs != NULL && strcmp(zfs->bootfs, 727 dsl_dir_fullname(dsldir)) == 0) { 728 zap_add_uint64(zfs->poolprops, "bootfs", 729 dsl_dir_dataset_id(dsldir)); 730 } 731 732 origmountpoint = mountpoint; 733 734 /* 735 * Figure out which fsnode corresponds to our mountpoint. 736 */ 737 root = arg; 738 cur = root; 739 if (strcmp(mountpoint, zfs->rootpath) != 0) { 740 mountpoint += strlen(zfs->rootpath); 741 742 /* 743 * Look up the directory in the staged tree. For example, if 744 * the dataset's mount point is /foo/bar/baz, we'll search the 745 * root directory for "foo", search "foo" for "baz", and so on. 746 * Each intermediate name must refer to a directory; the final 747 * component need not exist. 748 */ 749 cur = root; 750 for (next = name = mountpoint; next != NULL;) { 751 for (; *next == '/'; next++) 752 ; 753 name = strsep(&next, "/"); 754 755 for (; cur != NULL && strcmp(cur->name, name) != 0; 756 cur = cur->next) 757 ; 758 if (cur == NULL) { 759 if (next == NULL) 760 break; 761 errx(1, "missing mountpoint directory for `%s'", 762 dsl_dir_fullname(dsldir)); 763 } 764 if (cur->type != S_IFDIR) { 765 errx(1, 766 "mountpoint for `%s' is not a directory", 767 dsl_dir_fullname(dsldir)); 768 } 769 if (next != NULL) 770 cur = cur->child; 771 } 772 } 773 774 if (cur != NULL) { 775 assert(cur->type == S_IFDIR); 776 777 /* 778 * Multiple datasets shouldn't share a mountpoint. It's 779 * technically allowed, but it's not clear what makefs should do 780 * in that case. 781 */ 782 assert((cur->inode->flags & FI_ROOT) == 0); 783 if (cur != root) 784 cur->inode->flags |= FI_ROOT; 785 assert(cur->inode->param == NULL); 786 cur->inode->param = dsldir; 787 } 788 789 free(origmountpoint); 790 } 791 792 static int 793 fs_foreach_mark(fsnode *cur, void *arg) 794 { 795 uint64_t *countp; 796 797 countp = arg; 798 if (cur->type == S_IFDIR && fsnode_isroot(cur)) 799 return (1); 800 801 if (cur->inode->ino == 0) { 802 cur->inode->ino = ++(*countp); 803 cur->inode->nlink = 1; 804 } else { 805 cur->inode->nlink++; 806 } 807 808 return ((cur->inode->flags & FI_ROOT) != 0 ? 0 : 1); 809 } 810 811 /* 812 * Create a filesystem dataset. More specifically: 813 * - create an object set for the dataset, 814 * - add required metadata (SA tables, property definitions, etc.) to that 815 * object set, 816 * - optionally populate the object set with file objects, using "root" as the 817 * root directory. 818 * 819 * "dirfd" is a directory descriptor for the directory referenced by "root". It 820 * is closed before returning. 821 */ 822 static void 823 fs_build_one(zfs_opt_t *zfs, zfs_dsl_dir_t *dsldir, fsnode *root, int dirfd) 824 { 825 struct fs_populate_arg arg; 826 zfs_fs_t fs; 827 zfs_zap_t *masterzap; 828 zfs_objset_t *os; 829 dnode_phys_t *deleteq, *masterobj; 830 uint64_t deleteqid, dnodecount, moid, rootdirid, saobjid; 831 bool fakedroot; 832 833 /* 834 * This dataset's mountpoint doesn't exist in the staging tree, or the 835 * dataset doesn't have a mountpoint at all. In either case we still 836 * need a root directory. Fake up a root fsnode to handle this case. 837 */ 838 fakedroot = root == NULL; 839 if (fakedroot) { 840 struct stat *stp; 841 842 assert(dirfd == -1); 843 844 root = ecalloc(1, sizeof(*root)); 845 root->inode = ecalloc(1, sizeof(*root->inode)); 846 root->name = estrdup("."); 847 root->type = S_IFDIR; 848 849 stp = &root->inode->st; 850 stp->st_uid = 0; 851 stp->st_gid = 0; 852 stp->st_mode = S_IFDIR | 0755; 853 } 854 assert(root->type == S_IFDIR); 855 assert(fsnode_isroot(root)); 856 857 /* 858 * Initialize the object set for this dataset. 859 */ 860 os = objset_alloc(zfs, DMU_OST_ZFS); 861 masterobj = objset_dnode_alloc(os, DMU_OT_MASTER_NODE, &moid); 862 assert(moid == MASTER_NODE_OBJ); 863 864 memset(&fs, 0, sizeof(fs)); 865 fs.os = os; 866 867 /* 868 * Create the ZAP SA layout now since filesystem object dnodes will 869 * refer to those attributes. 870 */ 871 saobjid = fs_set_zpl_attrs(zfs, &fs); 872 873 /* 874 * Make a pass over the staged directory to detect hard links and assign 875 * virtual dnode numbers. 876 */ 877 dnodecount = 1; /* root directory */ 878 fsnode_foreach(root, fs_foreach_mark, &dnodecount); 879 880 /* 881 * Make a second pass to populate the dataset with files from the 882 * staged directory. Most of our runtime is spent here. 883 */ 884 arg.dirfd = dirfd; 885 arg.zfs = zfs; 886 arg.fs = &fs; 887 SLIST_INIT(&arg.dirs); 888 fs_populate_dir(root, &arg); 889 assert(!SLIST_EMPTY(&arg.dirs)); 890 fsnode_foreach(root, fs_foreach_populate, &arg); 891 assert(SLIST_EMPTY(&arg.dirs)); 892 rootdirid = arg.rootdirid; 893 894 /* 895 * Create an empty delete queue. We don't do anything with it, but 896 * OpenZFS will refuse to mount filesystems that don't have one. 897 */ 898 deleteq = objset_dnode_alloc(os, DMU_OT_UNLINKED_SET, &deleteqid); 899 zap_write(zfs, zap_alloc(os, deleteq)); 900 901 /* 902 * Populate and write the master node object. This is a ZAP object 903 * containing various dataset properties and the object IDs of the root 904 * directory and delete queue. 905 */ 906 masterzap = zap_alloc(os, masterobj); 907 zap_add_uint64(masterzap, ZFS_ROOT_OBJ, rootdirid); 908 zap_add_uint64(masterzap, ZFS_UNLINKED_SET, deleteqid); 909 zap_add_uint64(masterzap, ZFS_SA_ATTRS, saobjid); 910 zap_add_uint64(masterzap, ZPL_VERSION_OBJ, 5 /* ZPL_VERSION_SA */); 911 zap_add_uint64(masterzap, "normalization", 0 /* off */); 912 zap_add_uint64(masterzap, "utf8only", 0 /* off */); 913 zap_add_uint64(masterzap, "casesensitivity", 0 /* case sensitive */); 914 zap_add_uint64(masterzap, "acltype", 2 /* NFSv4 */); 915 zap_write(zfs, masterzap); 916 917 /* 918 * All finished with this object set, we may as well write it now. 919 * The DSL layer will sum up the bytes consumed by each dataset using 920 * information stored in the object set, so it can't be freed just yet. 921 */ 922 dsl_dir_dataset_write(zfs, os, dsldir); 923 924 if (fakedroot) { 925 free(root->inode); 926 free(root->name); 927 free(root); 928 } 929 free(fs.saoffs); 930 } 931 932 /* 933 * Create an object set for each DSL directory which has a dataset and doesn't 934 * already have an object set. 935 */ 936 static void 937 fs_build_unmounted(zfs_opt_t *zfs, zfs_dsl_dir_t *dsldir, void *arg __unused) 938 { 939 if (dsl_dir_has_dataset(dsldir) && !dsl_dir_dataset_has_objset(dsldir)) 940 fs_build_one(zfs, dsldir, NULL, -1); 941 } 942 943 /* 944 * Create our datasets and populate them with files. 945 */ 946 void 947 fs_build(zfs_opt_t *zfs, int dirfd, fsnode *root) 948 { 949 /* 950 * Run through our datasets and find the root fsnode for each one. Each 951 * root fsnode is flagged so that we can figure out which dataset it 952 * belongs to. 953 */ 954 dsl_dir_foreach(zfs, zfs->rootdsldir, fs_layout_one, root); 955 956 /* 957 * Did we find our boot filesystem? 958 */ 959 if (zfs->bootfs != NULL && !zap_entry_exists(zfs->poolprops, "bootfs")) 960 errx(1, "no mounted dataset matches bootfs property `%s'", 961 zfs->bootfs); 962 963 /* 964 * Traverse the file hierarchy starting from the root fsnode. One 965 * dataset, not necessarily the root dataset, must "own" the root 966 * directory by having its mountpoint be equal to the root path. 967 * 968 * As roots of other datasets are encountered during the traversal, 969 * fs_build_one() recursively creates the corresponding object sets and 970 * populates them. Once this function has returned, all datasets will 971 * have been fully populated. 972 */ 973 fs_build_one(zfs, root->inode->param, root, dirfd); 974 975 /* 976 * Now create object sets for datasets whose mountpoints weren't found 977 * in the staging directory, either because there is no mountpoint, or 978 * because the mountpoint doesn't correspond to an existing directory. 979 */ 980 dsl_dir_foreach(zfs, zfs->rootdsldir, fs_build_unmounted, NULL); 981 } 982