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/stat.h> 32 33 #include <assert.h> 34 #include <dirent.h> 35 #include <fcntl.h> 36 #include <stdlib.h> 37 #include <string.h> 38 #include <unistd.h> 39 40 #include <util.h> 41 42 #include "makefs.h" 43 #include "zfs.h" 44 45 typedef struct { 46 const char *name; 47 unsigned int id; 48 uint16_t size; 49 sa_bswap_type_t bs; 50 } zfs_sattr_t; 51 52 typedef struct zfs_fs { 53 zfs_objset_t *os; 54 55 /* Offset table for system attributes, indexed by a zpl_attr_t. */ 56 uint16_t *saoffs; 57 size_t sacnt; 58 const zfs_sattr_t *satab; 59 } zfs_fs_t; 60 61 /* 62 * The order of the attributes doesn't matter, this is simply the one hard-coded 63 * by OpenZFS, based on a zdb dump of the SA_REGISTRY table. 64 */ 65 typedef enum zpl_attr { 66 ZPL_ATIME, 67 ZPL_MTIME, 68 ZPL_CTIME, 69 ZPL_CRTIME, 70 ZPL_GEN, 71 ZPL_MODE, 72 ZPL_SIZE, 73 ZPL_PARENT, 74 ZPL_LINKS, 75 ZPL_XATTR, 76 ZPL_RDEV, 77 ZPL_FLAGS, 78 ZPL_UID, 79 ZPL_GID, 80 ZPL_PAD, 81 ZPL_ZNODE_ACL, 82 ZPL_DACL_COUNT, 83 ZPL_SYMLINK, 84 ZPL_SCANSTAMP, 85 ZPL_DACL_ACES, 86 ZPL_DXATTR, 87 ZPL_PROJID, 88 } zpl_attr_t; 89 90 /* 91 * This table must be kept in sync with zpl_attr_layout[] and zpl_attr_t. 92 */ 93 static const zfs_sattr_t zpl_attrs[] = { 94 #define _ZPL_ATTR(n, s, b) { .name = #n, .id = n, .size = s, .bs = b } 95 _ZPL_ATTR(ZPL_ATIME, sizeof(uint64_t) * 2, SA_UINT64_ARRAY), 96 _ZPL_ATTR(ZPL_MTIME, sizeof(uint64_t) * 2, SA_UINT64_ARRAY), 97 _ZPL_ATTR(ZPL_CTIME, sizeof(uint64_t) * 2, SA_UINT64_ARRAY), 98 _ZPL_ATTR(ZPL_CRTIME, sizeof(uint64_t) * 2, SA_UINT64_ARRAY), 99 _ZPL_ATTR(ZPL_GEN, sizeof(uint64_t), SA_UINT64_ARRAY), 100 _ZPL_ATTR(ZPL_MODE, sizeof(uint64_t), SA_UINT64_ARRAY), 101 _ZPL_ATTR(ZPL_SIZE, sizeof(uint64_t), SA_UINT64_ARRAY), 102 _ZPL_ATTR(ZPL_PARENT, sizeof(uint64_t), SA_UINT64_ARRAY), 103 _ZPL_ATTR(ZPL_LINKS, sizeof(uint64_t), SA_UINT64_ARRAY), 104 _ZPL_ATTR(ZPL_XATTR, sizeof(uint64_t), SA_UINT64_ARRAY), 105 _ZPL_ATTR(ZPL_RDEV, sizeof(uint64_t), SA_UINT64_ARRAY), 106 _ZPL_ATTR(ZPL_FLAGS, sizeof(uint64_t), SA_UINT64_ARRAY), 107 _ZPL_ATTR(ZPL_UID, sizeof(uint64_t), SA_UINT64_ARRAY), 108 _ZPL_ATTR(ZPL_GID, sizeof(uint64_t), SA_UINT64_ARRAY), 109 _ZPL_ATTR(ZPL_PAD, sizeof(uint64_t), SA_UINT64_ARRAY), 110 _ZPL_ATTR(ZPL_ZNODE_ACL, 88, SA_UINT64_ARRAY), 111 _ZPL_ATTR(ZPL_DACL_COUNT, sizeof(uint64_t), SA_UINT64_ARRAY), 112 _ZPL_ATTR(ZPL_SYMLINK, 0, SA_UINT8_ARRAY), 113 _ZPL_ATTR(ZPL_SCANSTAMP, sizeof(uint64_t) * 4, SA_UINT8_ARRAY), 114 _ZPL_ATTR(ZPL_DACL_ACES, 0, SA_ACL), 115 _ZPL_ATTR(ZPL_DXATTR, 0, SA_UINT8_ARRAY), 116 _ZPL_ATTR(ZPL_PROJID, sizeof(uint64_t), SA_UINT64_ARRAY), 117 #undef ZPL_ATTR 118 }; 119 120 /* 121 * This layout matches that of a filesystem created using OpenZFS on FreeBSD. 122 * It need not match in general, but FreeBSD's loader doesn't bother parsing the 123 * layout and just hard-codes attribute offsets. 124 */ 125 static const sa_attr_type_t zpl_attr_layout[] = { 126 ZPL_MODE, 127 ZPL_SIZE, 128 ZPL_GEN, 129 ZPL_UID, 130 ZPL_GID, 131 ZPL_PARENT, 132 ZPL_FLAGS, 133 ZPL_ATIME, 134 ZPL_MTIME, 135 ZPL_CTIME, 136 ZPL_CRTIME, 137 ZPL_LINKS, 138 ZPL_DACL_COUNT, 139 ZPL_DACL_ACES, 140 ZPL_SYMLINK, 141 }; 142 143 /* 144 * Keys for the ZPL attribute tables in the SA layout ZAP. The first two 145 * indices are reserved for legacy attribute encoding. 146 */ 147 #define SA_LAYOUT_INDEX_DEFAULT 2 148 #define SA_LAYOUT_INDEX_SYMLINK 3 149 150 struct fs_populate_dir { 151 SLIST_ENTRY(fs_populate_dir) next; 152 int dirfd; 153 uint64_t objid; 154 zfs_zap_t *zap; 155 }; 156 157 struct fs_populate_arg { 158 zfs_opt_t *zfs; 159 zfs_fs_t *fs; /* owning filesystem */ 160 uint64_t rootdirid; /* root directory dnode ID */ 161 int rootdirfd; /* root directory fd */ 162 SLIST_HEAD(, fs_populate_dir) dirs; /* stack of directories */ 163 }; 164 165 static void fs_build_one(zfs_opt_t *, zfs_dsl_dir_t *, fsnode *, int); 166 167 static void 168 eclose(int fd) 169 { 170 if (close(fd) != 0) 171 err(1, "close"); 172 } 173 174 static bool 175 fsnode_isroot(const fsnode *cur) 176 { 177 return (strcmp(cur->name, ".") == 0); 178 } 179 180 /* 181 * Visit each node in a directory hierarchy, in pre-order depth-first order. 182 */ 183 static void 184 fsnode_foreach(fsnode *root, int (*cb)(fsnode *, void *), void *arg) 185 { 186 assert(root->type == S_IFDIR); 187 188 for (fsnode *cur = root; cur != NULL; cur = cur->next) { 189 assert(cur->type == S_IFREG || cur->type == S_IFDIR || 190 cur->type == S_IFLNK); 191 192 if (cb(cur, arg) == 0) 193 continue; 194 if (cur->type == S_IFDIR && cur->child != NULL) 195 fsnode_foreach(cur->child, cb, arg); 196 } 197 } 198 199 static void 200 fs_populate_dirent(struct fs_populate_arg *arg, fsnode *cur, uint64_t dnid) 201 { 202 struct fs_populate_dir *dir; 203 uint64_t type; 204 205 switch (cur->type) { 206 case S_IFREG: 207 type = DT_REG; 208 break; 209 case S_IFDIR: 210 type = DT_DIR; 211 break; 212 case S_IFLNK: 213 type = DT_LNK; 214 break; 215 default: 216 assert(0); 217 } 218 219 dir = SLIST_FIRST(&arg->dirs); 220 zap_add_uint64(dir->zap, cur->name, ZFS_DIRENT_MAKE(type, dnid)); 221 } 222 223 static void 224 fs_populate_attr(zfs_fs_t *fs, char *attrbuf, const void *val, uint16_t ind, 225 size_t *szp) 226 { 227 assert(ind < fs->sacnt); 228 assert(fs->saoffs[ind] != 0xffff); 229 230 memcpy(attrbuf + fs->saoffs[ind], val, fs->satab[ind].size); 231 *szp += fs->satab[ind].size; 232 } 233 234 static void 235 fs_populate_varszattr(zfs_fs_t *fs, char *attrbuf, const void *val, 236 size_t valsz, size_t varoff, uint16_t ind, size_t *szp) 237 { 238 assert(ind < fs->sacnt); 239 assert(fs->saoffs[ind] != 0xffff); 240 assert(fs->satab[ind].size == 0); 241 242 memcpy(attrbuf + fs->saoffs[ind] + varoff, val, valsz); 243 *szp += valsz; 244 } 245 246 /* 247 * Derive the relative fd/path combo needed to access a file. Ideally we'd 248 * always be able to use relative lookups (i.e., use the *at() system calls), 249 * since they require less path translation and are more amenable to sandboxing, 250 * but the handling of multiple staging directories makes that difficult. To 251 * make matters worse, we have no choice but to use relative lookups when 252 * dealing with an mtree manifest, so both mechanisms are implemented. 253 */ 254 static void 255 fs_populate_path(const fsnode *cur, struct fs_populate_arg *arg, 256 char *path, size_t sz, int *dirfdp) 257 { 258 if (cur->root == NULL) { 259 size_t n; 260 261 *dirfdp = SLIST_FIRST(&arg->dirs)->dirfd; 262 n = strlcpy(path, cur->name, sz); 263 assert(n < sz); 264 } else { 265 int n; 266 267 *dirfdp = AT_FDCWD; 268 n = snprintf(path, sz, "%s/%s/%s", 269 cur->root, cur->path, cur->name); 270 assert(n >= 0); 271 assert((size_t)n < sz); 272 } 273 } 274 275 static int 276 fs_open(const fsnode *cur, struct fs_populate_arg *arg, int flags) 277 { 278 char path[PATH_MAX]; 279 int fd; 280 281 fs_populate_path(cur, arg, path, sizeof(path), &fd); 282 283 fd = openat(fd, path, flags); 284 if (fd < 0) 285 err(1, "openat(%s)", path); 286 return (fd); 287 } 288 289 static void 290 fs_readlink(const fsnode *cur, struct fs_populate_arg *arg, 291 char *buf, size_t bufsz) 292 { 293 char path[PATH_MAX]; 294 ssize_t n; 295 int fd; 296 297 fs_populate_path(cur, arg, path, sizeof(path), &fd); 298 299 n = readlinkat(fd, path, buf, bufsz - 1); 300 if (n == -1) 301 err(1, "readlinkat(%s)", cur->name); 302 buf[n] = '\0'; 303 } 304 305 static void 306 fs_populate_time(zfs_fs_t *fs, char *attrbuf, struct timespec *ts, 307 uint16_t ind, size_t *szp) 308 { 309 uint64_t timebuf[2]; 310 311 assert(ind < fs->sacnt); 312 assert(fs->saoffs[ind] != 0xffff); 313 assert(fs->satab[ind].size == sizeof(timebuf)); 314 315 timebuf[0] = ts->tv_sec; 316 timebuf[1] = ts->tv_nsec; 317 fs_populate_attr(fs, attrbuf, timebuf, ind, szp); 318 } 319 320 static void 321 fs_populate_sattrs(struct fs_populate_arg *arg, const fsnode *cur, 322 dnode_phys_t *dnode) 323 { 324 char target[PATH_MAX]; 325 zfs_fs_t *fs; 326 zfs_ace_hdr_t aces[3]; 327 struct stat *sb; 328 sa_hdr_phys_t *sahdr; 329 uint64_t daclcount, flags, gen, gid, links, mode, parent, objsize, uid; 330 char *attrbuf; 331 size_t bonussz, hdrsz; 332 int layout; 333 334 assert(dnode->dn_bonustype == DMU_OT_SA); 335 assert(dnode->dn_nblkptr == 1); 336 337 fs = arg->fs; 338 sb = &cur->inode->st; 339 340 switch (cur->type) { 341 case S_IFREG: 342 layout = SA_LAYOUT_INDEX_DEFAULT; 343 links = cur->inode->nlink; 344 objsize = sb->st_size; 345 parent = SLIST_FIRST(&arg->dirs)->objid; 346 break; 347 case S_IFDIR: 348 layout = SA_LAYOUT_INDEX_DEFAULT; 349 links = 1; /* .. */ 350 objsize = 1; /* .. */ 351 352 /* 353 * The size of a ZPL directory is the number of entries 354 * (including "." and ".."), and the link count is the number of 355 * entries which are directories (including "." and ".."). 356 */ 357 for (fsnode *c = fsnode_isroot(cur) ? cur->next : cur->child; 358 c != NULL; c = c->next) { 359 if (c->type == S_IFDIR) 360 links++; 361 objsize++; 362 } 363 364 /* The root directory is its own parent. */ 365 parent = SLIST_EMPTY(&arg->dirs) ? 366 arg->rootdirid : SLIST_FIRST(&arg->dirs)->objid; 367 break; 368 case S_IFLNK: 369 fs_readlink(cur, arg, target, sizeof(target)); 370 371 layout = SA_LAYOUT_INDEX_SYMLINK; 372 links = 1; 373 objsize = strlen(target); 374 parent = SLIST_FIRST(&arg->dirs)->objid; 375 break; 376 default: 377 assert(0); 378 } 379 380 daclcount = nitems(aces); 381 flags = ZFS_ACL_TRIVIAL | ZFS_ACL_AUTO_INHERIT | ZFS_NO_EXECS_DENIED | 382 ZFS_ARCHIVE | ZFS_AV_MODIFIED; /* XXX-MJ */ 383 gen = 1; 384 gid = sb->st_gid; 385 mode = sb->st_mode; 386 uid = sb->st_uid; 387 388 memset(aces, 0, sizeof(aces)); 389 aces[0].z_flags = ACE_OWNER; 390 aces[0].z_type = ACE_ACCESS_ALLOWED_ACE_TYPE; 391 aces[0].z_access_mask = ACE_WRITE_ATTRIBUTES | ACE_WRITE_OWNER | 392 ACE_WRITE_ACL | ACE_WRITE_NAMED_ATTRS | ACE_READ_ACL | 393 ACE_READ_ATTRIBUTES | ACE_READ_NAMED_ATTRS | ACE_SYNCHRONIZE; 394 if ((mode & S_IRUSR) != 0) 395 aces[0].z_access_mask |= ACE_READ_DATA; 396 if ((mode & S_IWUSR) != 0) 397 aces[0].z_access_mask |= ACE_WRITE_DATA | ACE_APPEND_DATA; 398 if ((mode & S_IXUSR) != 0) 399 aces[0].z_access_mask |= ACE_EXECUTE; 400 401 aces[1].z_flags = ACE_GROUP | ACE_IDENTIFIER_GROUP; 402 aces[1].z_type = ACE_ACCESS_ALLOWED_ACE_TYPE; 403 aces[1].z_access_mask = ACE_READ_ACL | ACE_READ_ATTRIBUTES | 404 ACE_READ_NAMED_ATTRS | ACE_SYNCHRONIZE; 405 if ((mode & S_IRGRP) != 0) 406 aces[1].z_access_mask |= ACE_READ_DATA; 407 if ((mode & S_IWGRP) != 0) 408 aces[1].z_access_mask |= ACE_WRITE_DATA | ACE_APPEND_DATA; 409 if ((mode & S_IXGRP) != 0) 410 aces[1].z_access_mask |= ACE_EXECUTE; 411 412 aces[2].z_flags = ACE_EVERYONE; 413 aces[2].z_type = ACE_ACCESS_ALLOWED_ACE_TYPE; 414 aces[2].z_access_mask = ACE_READ_ACL | ACE_READ_ATTRIBUTES | 415 ACE_READ_NAMED_ATTRS | ACE_SYNCHRONIZE; 416 if ((mode & S_IROTH) != 0) 417 aces[2].z_access_mask |= ACE_READ_DATA; 418 if ((mode & S_IWOTH) != 0) 419 aces[2].z_access_mask |= ACE_WRITE_DATA | ACE_APPEND_DATA; 420 if ((mode & S_IXOTH) != 0) 421 aces[2].z_access_mask |= ACE_EXECUTE; 422 423 switch (layout) { 424 case SA_LAYOUT_INDEX_DEFAULT: 425 /* At most one variable-length attribute. */ 426 hdrsz = sizeof(uint64_t); 427 break; 428 case SA_LAYOUT_INDEX_SYMLINK: 429 /* At most five variable-length attributes. */ 430 hdrsz = sizeof(uint64_t) * 2; 431 break; 432 default: 433 assert(0); 434 } 435 436 sahdr = (sa_hdr_phys_t *)DN_BONUS(dnode); 437 sahdr->sa_magic = SA_MAGIC; 438 SA_HDR_LAYOUT_INFO_ENCODE(sahdr->sa_layout_info, layout, hdrsz); 439 440 bonussz = SA_HDR_SIZE(sahdr); 441 attrbuf = (char *)sahdr + SA_HDR_SIZE(sahdr); 442 443 fs_populate_attr(fs, attrbuf, &daclcount, ZPL_DACL_COUNT, &bonussz); 444 fs_populate_attr(fs, attrbuf, &flags, ZPL_FLAGS, &bonussz); 445 fs_populate_attr(fs, attrbuf, &gen, ZPL_GEN, &bonussz); 446 fs_populate_attr(fs, attrbuf, &gid, ZPL_GID, &bonussz); 447 fs_populate_attr(fs, attrbuf, &links, ZPL_LINKS, &bonussz); 448 fs_populate_attr(fs, attrbuf, &mode, ZPL_MODE, &bonussz); 449 fs_populate_attr(fs, attrbuf, &parent, ZPL_PARENT, &bonussz); 450 fs_populate_attr(fs, attrbuf, &objsize, ZPL_SIZE, &bonussz); 451 fs_populate_attr(fs, attrbuf, &uid, ZPL_UID, &bonussz); 452 453 /* 454 * We deliberately set atime = mtime here to ensure that images are 455 * reproducible. 456 */ 457 fs_populate_time(fs, attrbuf, &sb->st_mtim, ZPL_ATIME, &bonussz); 458 fs_populate_time(fs, attrbuf, &sb->st_ctim, ZPL_CTIME, &bonussz); 459 fs_populate_time(fs, attrbuf, &sb->st_mtim, ZPL_MTIME, &bonussz); 460 #ifdef __linux__ 461 /* Linux has no st_birthtim; approximate with st_ctim */ 462 fs_populate_time(fs, attrbuf, &sb->st_ctim, ZPL_CRTIME, &bonussz); 463 #else 464 fs_populate_time(fs, attrbuf, &sb->st_birthtim, ZPL_CRTIME, &bonussz); 465 #endif 466 467 fs_populate_varszattr(fs, attrbuf, aces, sizeof(aces), 0, 468 ZPL_DACL_ACES, &bonussz); 469 sahdr->sa_lengths[0] = sizeof(aces); 470 471 if (cur->type == S_IFLNK) { 472 assert(layout == SA_LAYOUT_INDEX_SYMLINK); 473 /* Need to use a spill block pointer if the target is long. */ 474 assert(bonussz + objsize <= DN_OLD_MAX_BONUSLEN); 475 fs_populate_varszattr(fs, attrbuf, target, objsize, 476 sahdr->sa_lengths[0], ZPL_SYMLINK, &bonussz); 477 sahdr->sa_lengths[1] = (uint16_t)objsize; 478 } 479 480 dnode->dn_bonuslen = bonussz; 481 } 482 483 static void 484 fs_populate_file(fsnode *cur, struct fs_populate_arg *arg) 485 { 486 struct dnode_cursor *c; 487 dnode_phys_t *dnode; 488 zfs_opt_t *zfs; 489 char *buf; 490 uint64_t dnid; 491 ssize_t n; 492 size_t bufsz; 493 off_t size, target; 494 int fd; 495 496 assert(cur->type == S_IFREG); 497 assert((cur->inode->flags & FI_ROOT) == 0); 498 499 zfs = arg->zfs; 500 501 assert(cur->inode->ino != 0); 502 if ((cur->inode->flags & FI_ALLOCATED) != 0) { 503 /* 504 * This is a hard link of an existing file. 505 * 506 * XXX-MJ need to check whether it crosses datasets, add a test 507 * case for that 508 */ 509 fs_populate_dirent(arg, cur, cur->inode->ino); 510 return; 511 } 512 513 dnode = objset_dnode_bonus_alloc(arg->fs->os, 514 DMU_OT_PLAIN_FILE_CONTENTS, DMU_OT_SA, 0, &dnid); 515 cur->inode->ino = dnid; 516 cur->inode->flags |= FI_ALLOCATED; 517 518 fd = fs_open(cur, arg, O_RDONLY); 519 520 buf = zfs->filebuf; 521 bufsz = sizeof(zfs->filebuf); 522 size = cur->inode->st.st_size; 523 c = dnode_cursor_init(zfs, arg->fs->os, dnode, size, 0); 524 for (off_t foff = 0; foff < size; foff += target) { 525 off_t loc, sofar; 526 527 /* 528 * Fill up our buffer, handling partial reads. 529 * 530 * It might be profitable to use copy_file_range(2) here. 531 */ 532 sofar = 0; 533 target = MIN(size - foff, (off_t)bufsz); 534 do { 535 n = read(fd, buf + sofar, target); 536 if (n < 0) 537 err(1, "reading from '%s'", cur->name); 538 if (n == 0) 539 errx(1, "unexpected EOF reading '%s'", 540 cur->name); 541 sofar += n; 542 } while (sofar < target); 543 544 if (target < (off_t)bufsz) 545 memset(buf + target, 0, bufsz - target); 546 547 loc = objset_space_alloc(zfs, arg->fs->os, &target); 548 vdev_pwrite_dnode_indir(zfs, dnode, 0, 1, buf, target, loc, 549 dnode_cursor_next(zfs, c, foff)); 550 } 551 eclose(fd); 552 dnode_cursor_finish(zfs, c); 553 554 fs_populate_sattrs(arg, cur, dnode); 555 fs_populate_dirent(arg, cur, dnid); 556 } 557 558 static void 559 fs_populate_dir(fsnode *cur, struct fs_populate_arg *arg) 560 { 561 dnode_phys_t *dnode; 562 zfs_objset_t *os; 563 uint64_t dnid; 564 int dirfd; 565 566 assert(cur->type == S_IFDIR); 567 assert((cur->inode->flags & FI_ALLOCATED) == 0); 568 569 os = arg->fs->os; 570 571 dnode = objset_dnode_bonus_alloc(os, DMU_OT_DIRECTORY_CONTENTS, 572 DMU_OT_SA, 0, &dnid); 573 574 /* 575 * Add an entry to the parent directory and open this directory. 576 */ 577 if (!SLIST_EMPTY(&arg->dirs)) { 578 fs_populate_dirent(arg, cur, dnid); 579 dirfd = fs_open(cur, arg, O_DIRECTORY | O_RDONLY); 580 } else { 581 arg->rootdirid = dnid; 582 dirfd = arg->rootdirfd; 583 arg->rootdirfd = -1; 584 } 585 586 /* 587 * Set ZPL attributes. 588 */ 589 fs_populate_sattrs(arg, cur, dnode); 590 591 /* 592 * If this is a root directory, then its children belong to a different 593 * dataset and this directory remains empty in the current objset. 594 */ 595 if ((cur->inode->flags & FI_ROOT) == 0) { 596 struct fs_populate_dir *dir; 597 598 dir = ecalloc(1, sizeof(*dir)); 599 dir->dirfd = dirfd; 600 dir->objid = dnid; 601 dir->zap = zap_alloc(os, dnode); 602 SLIST_INSERT_HEAD(&arg->dirs, dir, next); 603 } else { 604 zap_write(arg->zfs, zap_alloc(os, dnode)); 605 fs_build_one(arg->zfs, cur->inode->param, cur->child, dirfd); 606 } 607 } 608 609 static void 610 fs_populate_symlink(fsnode *cur, struct fs_populate_arg *arg) 611 { 612 dnode_phys_t *dnode; 613 uint64_t dnid; 614 615 assert(cur->type == S_IFLNK); 616 assert((cur->inode->flags & (FI_ALLOCATED | FI_ROOT)) == 0); 617 618 dnode = objset_dnode_bonus_alloc(arg->fs->os, 619 DMU_OT_PLAIN_FILE_CONTENTS, DMU_OT_SA, 0, &dnid); 620 621 fs_populate_dirent(arg, cur, dnid); 622 623 fs_populate_sattrs(arg, cur, dnode); 624 } 625 626 static int 627 fs_foreach_populate(fsnode *cur, void *_arg) 628 { 629 struct fs_populate_arg *arg; 630 struct fs_populate_dir *dir; 631 int ret; 632 633 arg = _arg; 634 switch (cur->type) { 635 case S_IFREG: 636 fs_populate_file(cur, arg); 637 break; 638 case S_IFDIR: 639 if (fsnode_isroot(cur)) 640 break; 641 fs_populate_dir(cur, arg); 642 break; 643 case S_IFLNK: 644 fs_populate_symlink(cur, arg); 645 break; 646 default: 647 assert(0); 648 } 649 650 ret = (cur->inode->flags & FI_ROOT) != 0 ? 0 : 1; 651 652 if (cur->next == NULL && 653 (cur->child == NULL || (cur->inode->flags & FI_ROOT) != 0)) { 654 /* 655 * We reached a terminal node in a subtree. Walk back up and 656 * write out directories. We're done once we hit the root of a 657 * dataset or find a level where we're not on the edge of the 658 * tree. 659 */ 660 do { 661 dir = SLIST_FIRST(&arg->dirs); 662 SLIST_REMOVE_HEAD(&arg->dirs, next); 663 zap_write(arg->zfs, dir->zap); 664 if (dir->dirfd != -1) 665 eclose(dir->dirfd); 666 free(dir); 667 cur = cur->parent; 668 } while (cur != NULL && cur->next == NULL && 669 (cur->inode->flags & FI_ROOT) == 0); 670 } 671 672 return (ret); 673 } 674 675 static void 676 fs_add_zpl_attr_layout(zfs_zap_t *zap, unsigned int index, 677 const sa_attr_type_t layout[], size_t sacnt) 678 { 679 char ti[16]; 680 681 assert(sizeof(layout[0]) == 2); 682 683 snprintf(ti, sizeof(ti), "%u", index); 684 zap_add(zap, ti, sizeof(sa_attr_type_t), sacnt, 685 (const uint8_t *)layout); 686 } 687 688 /* 689 * Initialize system attribute tables. 690 * 691 * There are two elements to this. First, we write the zpl_attrs[] and 692 * zpl_attr_layout[] tables to disk. Then we create a lookup table which 693 * allows us to set file attributes quickly. 694 */ 695 static uint64_t 696 fs_set_zpl_attrs(zfs_opt_t *zfs, zfs_fs_t *fs) 697 { 698 zfs_zap_t *sazap, *salzap, *sarzap; 699 zfs_objset_t *os; 700 dnode_phys_t *saobj, *salobj, *sarobj; 701 uint64_t saobjid, salobjid, sarobjid; 702 uint16_t offset; 703 704 os = fs->os; 705 706 /* 707 * The on-disk tables are stored in two ZAP objects, the registry object 708 * and the layout object. Individual attributes are described by 709 * entries in the registry object; for example, the value for the 710 * "ZPL_SIZE" key gives the size and encoding of the ZPL_SIZE attribute. 711 * The attributes of a file are ordered according to one of the layouts 712 * defined in the layout object. The master node object is simply used 713 * to locate the registry and layout objects. 714 */ 715 saobj = objset_dnode_alloc(os, DMU_OT_SA_MASTER_NODE, &saobjid); 716 salobj = objset_dnode_alloc(os, DMU_OT_SA_ATTR_LAYOUTS, &salobjid); 717 sarobj = objset_dnode_alloc(os, DMU_OT_SA_ATTR_REGISTRATION, &sarobjid); 718 719 sarzap = zap_alloc(os, sarobj); 720 for (size_t i = 0; i < nitems(zpl_attrs); i++) { 721 const zfs_sattr_t *sa; 722 uint64_t attr; 723 724 attr = 0; 725 sa = &zpl_attrs[i]; 726 SA_ATTR_ENCODE(attr, (uint64_t)i, sa->size, sa->bs); 727 zap_add_uint64(sarzap, sa->name, attr); 728 } 729 zap_write(zfs, sarzap); 730 731 /* 732 * Layouts are arrays of indices into the registry. We define two 733 * layouts for use by the ZPL, one for non-symlinks and one for 734 * symlinks. They are identical except that the symlink layout includes 735 * ZPL_SYMLINK as its final attribute. 736 */ 737 salzap = zap_alloc(os, salobj); 738 assert(zpl_attr_layout[nitems(zpl_attr_layout) - 1] == ZPL_SYMLINK); 739 fs_add_zpl_attr_layout(salzap, SA_LAYOUT_INDEX_DEFAULT, 740 zpl_attr_layout, nitems(zpl_attr_layout) - 1); 741 fs_add_zpl_attr_layout(salzap, SA_LAYOUT_INDEX_SYMLINK, 742 zpl_attr_layout, nitems(zpl_attr_layout)); 743 zap_write(zfs, salzap); 744 745 sazap = zap_alloc(os, saobj); 746 zap_add_uint64(sazap, SA_LAYOUTS, salobjid); 747 zap_add_uint64(sazap, SA_REGISTRY, sarobjid); 748 zap_write(zfs, sazap); 749 750 /* Sanity check. */ 751 for (size_t i = 0; i < nitems(zpl_attrs); i++) 752 assert(i == zpl_attrs[i].id); 753 754 /* 755 * Build the offset table used when setting file attributes. File 756 * attributes are stored in the object's bonus buffer; this table 757 * provides the buffer offset of attributes referenced by the layout 758 * table. 759 */ 760 fs->sacnt = nitems(zpl_attrs); 761 fs->saoffs = ecalloc(fs->sacnt, sizeof(*fs->saoffs)); 762 for (size_t i = 0; i < fs->sacnt; i++) 763 fs->saoffs[i] = 0xffff; 764 offset = 0; 765 for (size_t i = 0; i < nitems(zpl_attr_layout); i++) { 766 uint16_t size; 767 768 assert(zpl_attr_layout[i] < fs->sacnt); 769 770 fs->saoffs[zpl_attr_layout[i]] = offset; 771 size = zpl_attrs[zpl_attr_layout[i]].size; 772 offset += size; 773 } 774 fs->satab = zpl_attrs; 775 776 return (saobjid); 777 } 778 779 static void 780 fs_layout_one(zfs_opt_t *zfs, zfs_dsl_dir_t *dsldir, void *arg) 781 { 782 char *mountpoint, *origmountpoint, *name, *next; 783 fsnode *cur, *root; 784 uint64_t canmount; 785 786 if (!dsl_dir_has_dataset(dsldir)) 787 return; 788 789 if (dsl_dir_get_canmount(dsldir, &canmount) == 0 && canmount == 0) 790 return; 791 mountpoint = dsl_dir_get_mountpoint(zfs, dsldir); 792 if (mountpoint == NULL) 793 return; 794 795 /* 796 * If we were asked to specify a bootfs, set it here. 797 */ 798 if (zfs->bootfs != NULL && strcmp(zfs->bootfs, 799 dsl_dir_fullname(dsldir)) == 0) { 800 zap_add_uint64(zfs->poolprops, "bootfs", 801 dsl_dir_dataset_id(dsldir)); 802 } 803 804 origmountpoint = mountpoint; 805 806 /* 807 * Figure out which fsnode corresponds to our mountpoint. 808 */ 809 root = arg; 810 cur = root; 811 if (strcmp(mountpoint, zfs->rootpath) != 0) { 812 mountpoint += strlen(zfs->rootpath); 813 814 /* 815 * Look up the directory in the staged tree. For example, if 816 * the dataset's mount point is /foo/bar/baz, we'll search the 817 * root directory for "foo", search "foo" for "baz", and so on. 818 * Each intermediate name must refer to a directory; the final 819 * component need not exist. 820 */ 821 cur = root; 822 for (next = name = mountpoint; next != NULL;) { 823 for (; *next == '/'; next++) 824 ; 825 name = strsep(&next, "/"); 826 827 for (; cur != NULL && strcmp(cur->name, name) != 0; 828 cur = cur->next) 829 ; 830 if (cur == NULL) { 831 if (next == NULL) 832 break; 833 errx(1, "missing mountpoint directory for `%s'", 834 dsl_dir_fullname(dsldir)); 835 } 836 if (cur->type != S_IFDIR) { 837 errx(1, 838 "mountpoint for `%s' is not a directory", 839 dsl_dir_fullname(dsldir)); 840 } 841 if (next != NULL) 842 cur = cur->child; 843 } 844 } 845 846 if (cur != NULL) { 847 assert(cur->type == S_IFDIR); 848 849 /* 850 * Multiple datasets shouldn't share a mountpoint. It's 851 * technically allowed, but it's not clear what makefs should do 852 * in that case. 853 */ 854 assert((cur->inode->flags & FI_ROOT) == 0); 855 if (cur != root) 856 cur->inode->flags |= FI_ROOT; 857 assert(cur->inode->param == NULL); 858 cur->inode->param = dsldir; 859 } 860 861 free(origmountpoint); 862 } 863 864 static int 865 fs_foreach_mark(fsnode *cur, void *arg) 866 { 867 uint64_t *countp; 868 869 countp = arg; 870 if (cur->type == S_IFDIR && fsnode_isroot(cur)) 871 return (1); 872 873 if (cur->inode->ino == 0) { 874 cur->inode->ino = ++(*countp); 875 cur->inode->nlink = 1; 876 } else { 877 cur->inode->nlink++; 878 } 879 880 return ((cur->inode->flags & FI_ROOT) != 0 ? 0 : 1); 881 } 882 883 /* 884 * Create a filesystem dataset. More specifically: 885 * - create an object set for the dataset, 886 * - add required metadata (SA tables, property definitions, etc.) to that 887 * object set, 888 * - optionally populate the object set with file objects, using "root" as the 889 * root directory. 890 * 891 * "dirfd" is a directory descriptor for the directory referenced by "root". It 892 * is closed before returning. 893 */ 894 static void 895 fs_build_one(zfs_opt_t *zfs, zfs_dsl_dir_t *dsldir, fsnode *root, int dirfd) 896 { 897 struct fs_populate_arg arg; 898 zfs_fs_t fs; 899 zfs_zap_t *masterzap; 900 zfs_objset_t *os; 901 dnode_phys_t *deleteq, *masterobj; 902 uint64_t deleteqid, dnodecount, moid, rootdirid, saobjid; 903 bool fakedroot; 904 905 /* 906 * This dataset's mountpoint doesn't exist in the staging tree, or the 907 * dataset doesn't have a mountpoint at all. In either case we still 908 * need a root directory. Fake up a root fsnode to handle this case. 909 */ 910 fakedroot = root == NULL; 911 if (fakedroot) { 912 struct stat *stp; 913 914 assert(dirfd == -1); 915 916 root = ecalloc(1, sizeof(*root)); 917 root->inode = ecalloc(1, sizeof(*root->inode)); 918 root->name = estrdup("."); 919 root->type = S_IFDIR; 920 921 stp = &root->inode->st; 922 stp->st_uid = 0; 923 stp->st_gid = 0; 924 stp->st_mode = S_IFDIR | 0755; 925 } 926 assert(root->type == S_IFDIR); 927 assert(fsnode_isroot(root)); 928 929 /* 930 * Initialize the object set for this dataset. 931 */ 932 os = objset_alloc(zfs, DMU_OST_ZFS); 933 masterobj = objset_dnode_alloc(os, DMU_OT_MASTER_NODE, &moid); 934 assert(moid == MASTER_NODE_OBJ); 935 936 memset(&fs, 0, sizeof(fs)); 937 fs.os = os; 938 939 /* 940 * Create the ZAP SA layout now since filesystem object dnodes will 941 * refer to those attributes. 942 */ 943 saobjid = fs_set_zpl_attrs(zfs, &fs); 944 945 /* 946 * Make a pass over the staged directory to detect hard links and assign 947 * virtual dnode numbers. 948 */ 949 dnodecount = 1; /* root directory */ 950 fsnode_foreach(root, fs_foreach_mark, &dnodecount); 951 952 /* 953 * Make a second pass to populate the dataset with files from the 954 * staged directory. Most of our runtime is spent here. 955 */ 956 arg.rootdirfd = dirfd; 957 arg.zfs = zfs; 958 arg.fs = &fs; 959 SLIST_INIT(&arg.dirs); 960 fs_populate_dir(root, &arg); 961 assert(!SLIST_EMPTY(&arg.dirs)); 962 fsnode_foreach(root, fs_foreach_populate, &arg); 963 assert(SLIST_EMPTY(&arg.dirs)); 964 rootdirid = arg.rootdirid; 965 966 /* 967 * Create an empty delete queue. We don't do anything with it, but 968 * OpenZFS will refuse to mount filesystems that don't have one. 969 */ 970 deleteq = objset_dnode_alloc(os, DMU_OT_UNLINKED_SET, &deleteqid); 971 zap_write(zfs, zap_alloc(os, deleteq)); 972 973 /* 974 * Populate and write the master node object. This is a ZAP object 975 * containing various dataset properties and the object IDs of the root 976 * directory and delete queue. 977 */ 978 masterzap = zap_alloc(os, masterobj); 979 zap_add_uint64(masterzap, ZFS_ROOT_OBJ, rootdirid); 980 zap_add_uint64(masterzap, ZFS_UNLINKED_SET, deleteqid); 981 zap_add_uint64(masterzap, ZFS_SA_ATTRS, saobjid); 982 zap_add_uint64(masterzap, ZPL_VERSION_OBJ, 5 /* ZPL_VERSION_SA */); 983 zap_add_uint64(masterzap, "normalization", 0 /* off */); 984 zap_add_uint64(masterzap, "utf8only", 0 /* off */); 985 zap_add_uint64(masterzap, "casesensitivity", 0 /* case sensitive */); 986 zap_add_uint64(masterzap, "acltype", 2 /* NFSv4 */); 987 zap_write(zfs, masterzap); 988 989 /* 990 * All finished with this object set, we may as well write it now. 991 * The DSL layer will sum up the bytes consumed by each dataset using 992 * information stored in the object set, so it can't be freed just yet. 993 */ 994 dsl_dir_dataset_write(zfs, os, dsldir); 995 996 if (fakedroot) { 997 free(root->inode); 998 free(root->name); 999 free(root); 1000 } 1001 free(fs.saoffs); 1002 } 1003 1004 /* 1005 * Create an object set for each DSL directory which has a dataset and doesn't 1006 * already have an object set. 1007 */ 1008 static void 1009 fs_build_unmounted(zfs_opt_t *zfs, zfs_dsl_dir_t *dsldir, void *arg __unused) 1010 { 1011 if (dsl_dir_has_dataset(dsldir) && !dsl_dir_dataset_has_objset(dsldir)) 1012 fs_build_one(zfs, dsldir, NULL, -1); 1013 } 1014 1015 /* 1016 * Create our datasets and populate them with files. 1017 */ 1018 void 1019 fs_build(zfs_opt_t *zfs, int dirfd, fsnode *root) 1020 { 1021 /* 1022 * Run through our datasets and find the root fsnode for each one. Each 1023 * root fsnode is flagged so that we can figure out which dataset it 1024 * belongs to. 1025 */ 1026 dsl_dir_foreach(zfs, zfs->rootdsldir, fs_layout_one, root); 1027 1028 /* 1029 * Did we find our boot filesystem? 1030 */ 1031 if (zfs->bootfs != NULL && !zap_entry_exists(zfs->poolprops, "bootfs")) 1032 errx(1, "no mounted dataset matches bootfs property `%s'", 1033 zfs->bootfs); 1034 1035 /* 1036 * Traverse the file hierarchy starting from the root fsnode. One 1037 * dataset, not necessarily the root dataset, must "own" the root 1038 * directory by having its mountpoint be equal to the root path. 1039 * 1040 * As roots of other datasets are encountered during the traversal, 1041 * fs_build_one() recursively creates the corresponding object sets and 1042 * populates them. Once this function has returned, all datasets will 1043 * have been fully populated. 1044 */ 1045 fs_build_one(zfs, root->inode->param, root, dirfd); 1046 1047 /* 1048 * Now create object sets for datasets whose mountpoints weren't found 1049 * in the staging directory, either because there is no mountpoint, or 1050 * because the mountpoint doesn't correspond to an existing directory. 1051 */ 1052 dsl_dir_foreach(zfs, zfs->rootdsldir, fs_build_unmounted, NULL); 1053 } 1054