1 /* 2 * Copyright (c) 2007 The DragonFly Project. All rights reserved. 3 * 4 * This code is derived from software contributed to The DragonFly Project 5 * by Matthew Dillon <dillon@backplane.com> 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in 15 * the documentation and/or other materials provided with the 16 * distribution. 17 * 3. Neither the name of The DragonFly Project nor the names of its 18 * contributors may be used to endorse or promote products derived 19 * from this software without specific, prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 24 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 25 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 26 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING, 27 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 28 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED 29 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT 31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 */ 34 35 #include <sys/diskslice.h> 36 #include <sys/diskmbr.h> 37 38 #include "hammer_util.h" 39 40 static void get_buffer_readahead(struct buffer_info *base); 41 static void *get_ondisk(hammer_off_t buf_offset, struct buffer_info **bufferp, 42 int isnew); 43 static __inline int readhammervol(struct volume_info *vol); 44 static __inline int readhammerbuf(struct buffer_info *buf); 45 static __inline int writehammervol(struct volume_info *vol); 46 static __inline int writehammerbuf(struct buffer_info *buf); 47 48 int DebugOpt; 49 50 uuid_t Hammer_FSType; 51 uuid_t Hammer_FSId; 52 int UseReadBehind = -4; 53 int UseReadAhead = 4; 54 int AssertOnFailure = 1; 55 struct volume_list VolList = TAILQ_HEAD_INITIALIZER(VolList); 56 static int valid_hammer_volumes; 57 58 static __inline 59 int 60 buffer_hash(hammer_off_t buf_offset) 61 { 62 int hi; 63 64 hi = (int)(buf_offset / HAMMER_BUFSIZE) & HAMMER_BUFLISTMASK; 65 return(hi); 66 } 67 68 static struct buffer_info* 69 find_buffer(struct volume_info *volume, hammer_off_t buf_offset) 70 { 71 int hi; 72 struct buffer_info *buf; 73 74 hi = buffer_hash(buf_offset); 75 TAILQ_FOREACH(buf, &volume->buffer_lists[hi], entry) 76 if (buf->buf_offset == buf_offset) 77 return(buf); 78 return(NULL); 79 } 80 81 static 82 struct volume_info * 83 __alloc_volume(const char *volname, int oflags) 84 { 85 struct volume_info *vol; 86 int i; 87 88 vol = malloc(sizeof(*vol)); 89 if (vol == NULL) 90 err(1, "alloc_volume"); 91 bzero(vol, sizeof(*vol)); 92 93 vol->vol_no = -1; 94 vol->rdonly = (oflags == O_RDONLY); 95 vol->name = strdup(volname); 96 vol->fd = open(vol->name, oflags); 97 if (vol->fd < 0) 98 err(1, "alloc_volume: Failed to open %s", vol->name); 99 100 vol->size = 0; 101 vol->device_offset = 0; 102 vol->type = NULL; 103 104 vol->ondisk = malloc(HAMMER_BUFSIZE); 105 if (vol->ondisk == NULL) 106 err(1, "alloc_volume"); 107 bzero(vol->ondisk, HAMMER_BUFSIZE); 108 109 for (i = 0; i < HAMMER_BUFLISTS; ++i) 110 TAILQ_INIT(&vol->buffer_lists[i]); 111 112 return(vol); 113 } 114 115 static void 116 __add_volume(struct volume_info *vol) 117 { 118 struct volume_info *scan; 119 struct stat st1, st2; 120 121 if (fstat(vol->fd, &st1) != 0) 122 errx(1, "add_volume: %s: Failed to stat", vol->name); 123 124 TAILQ_FOREACH(scan, &VolList, entry) { 125 if (scan->vol_no == vol->vol_no) { 126 errx(1, "add_volume: %s: Duplicate volume number %d " 127 "against %s", 128 vol->name, vol->vol_no, scan->name); 129 } 130 if (fstat(scan->fd, &st2) != 0) { 131 errx(1, "add_volume: %s: Failed to stat %s", 132 vol->name, scan->name); 133 } 134 if ((st1.st_ino == st2.st_ino) && (st1.st_dev == st2.st_dev)) { 135 errx(1, "add_volume: %s: Specified more than once", 136 vol->name); 137 } 138 } 139 140 TAILQ_INSERT_TAIL(&VolList, vol, entry); 141 } 142 143 /* 144 * Initialize a volume structure and ondisk vol_no field. 145 */ 146 struct volume_info * 147 init_volume(int32_t vol_no, const char *filename, int oflags) 148 { 149 struct volume_info *vol; 150 151 vol = __alloc_volume(filename, oflags); 152 vol->vol_no = vol->ondisk->vol_no = vol_no; 153 154 __add_volume(vol); 155 156 return(vol); 157 } 158 159 /* 160 * Initialize a volume structure and read ondisk volume header. 161 */ 162 struct volume_info* 163 load_volume(const char *filename, int oflags) 164 { 165 struct volume_info *vol; 166 struct hammer_volume_ondisk *ondisk; 167 int n; 168 169 vol = __alloc_volume(filename, oflags); 170 171 n = readhammervol(vol); 172 if (n == -1) { 173 err(1, "load_volume: %s: Read failed at offset 0", vol->name); 174 } 175 ondisk = vol->ondisk; 176 vol->vol_no = ondisk->vol_no; 177 178 if (ondisk->vol_rootvol != HAMMER_ROOT_VOLNO) { 179 errx(1, "load_volume: Invalid root volume# %d", 180 ondisk->vol_rootvol); 181 } 182 183 if (bcmp(&Hammer_FSType, &ondisk->vol_fstype, sizeof(Hammer_FSType))) { 184 errx(1, "load_volume: %s: Header does not indicate " 185 "that this is a hammer volume", vol->name); 186 } 187 188 if (valid_hammer_volumes++ == 0) { 189 Hammer_FSId = ondisk->vol_fsid; 190 } else if (bcmp(&Hammer_FSId, &ondisk->vol_fsid, sizeof(Hammer_FSId))) { 191 errx(1, "load_volume: %s: FSId does match other volumes!", 192 vol->name); 193 } 194 195 __add_volume(vol); 196 197 return(vol); 198 } 199 200 /* 201 * Check basic volume characteristics. 202 */ 203 void 204 check_volume(struct volume_info *vol) 205 { 206 struct partinfo pinfo; 207 struct stat st; 208 209 /* 210 * Get basic information about the volume 211 */ 212 if (ioctl(vol->fd, DIOCGPART, &pinfo) < 0) { 213 /* 214 * Allow the formatting of regular files as HAMMER volumes 215 */ 216 if (fstat(vol->fd, &st) < 0) 217 err(1, "Unable to stat %s", vol->name); 218 vol->size = st.st_size; 219 vol->type = "REGFILE"; 220 } else { 221 /* 222 * When formatting a block device as a HAMMER volume the 223 * sector size must be compatible. HAMMER uses 16384 byte 224 * filesystem buffers. 225 */ 226 if (pinfo.reserved_blocks) { 227 errx(1, "HAMMER cannot be placed in a partition " 228 "which overlaps the disklabel or MBR"); 229 } 230 if (pinfo.media_blksize > HAMMER_BUFSIZE || 231 HAMMER_BUFSIZE % pinfo.media_blksize) { 232 errx(1, "A media sector size of %d is not supported", 233 pinfo.media_blksize); 234 } 235 236 vol->size = pinfo.media_size; 237 vol->device_offset = pinfo.media_offset; 238 vol->type = "DEVICE"; 239 } 240 } 241 242 struct volume_info * 243 get_volume(int32_t vol_no) 244 { 245 struct volume_info *vol; 246 247 TAILQ_FOREACH(vol, &VolList, entry) { 248 if (vol->vol_no == vol_no) 249 break; 250 } 251 if (vol == NULL) 252 errx(1, "get_volume: Volume %d does not exist!", vol_no); 253 254 /* not added to or removed from hammer cache */ 255 return(vol); 256 } 257 258 struct volume_info * 259 get_root_volume(void) 260 { 261 return(get_volume(HAMMER_ROOT_VOLNO)); 262 } 263 264 /* 265 * Acquire the specified buffer. isnew is -1 only when called 266 * via get_buffer_readahead() to prevent another readahead. 267 */ 268 struct buffer_info * 269 get_buffer(hammer_off_t buf_offset, int isnew) 270 { 271 struct buffer_info *buf; 272 struct volume_info *volume; 273 int vol_no; 274 int zone; 275 int hi; 276 int dora = 0; 277 278 zone = HAMMER_ZONE_DECODE(buf_offset); 279 if (zone > HAMMER_ZONE_RAW_BUFFER_INDEX) { 280 buf_offset = blockmap_lookup(buf_offset, NULL, NULL, NULL); 281 } 282 if (buf_offset == HAMMER_OFF_BAD) 283 return(NULL); 284 assert(hammer_is_zone_raw_buffer(buf_offset)); 285 286 vol_no = HAMMER_VOL_DECODE(buf_offset); 287 volume = get_volume(vol_no); 288 289 buf_offset &= ~HAMMER_BUFMASK64; 290 buf = find_buffer(volume, buf_offset); 291 292 if (buf == NULL) { 293 buf = malloc(sizeof(*buf)); 294 bzero(buf, sizeof(*buf)); 295 buf->buf_offset = buf_offset; 296 buf->raw_offset = hammer_xlate_to_phys(volume->ondisk, 297 buf_offset); 298 buf->volume = volume; 299 buf->ondisk = malloc(HAMMER_BUFSIZE); 300 if (isnew <= 0) { 301 if (readhammerbuf(buf) == -1) { 302 err(1, "get_buffer: %s:%016jx " 303 "Read failed at offset %016jx", 304 volume->name, 305 (intmax_t)buf->buf_offset, 306 (intmax_t)buf->raw_offset); 307 } 308 } 309 310 hi = buffer_hash(buf_offset); 311 TAILQ_INSERT_TAIL(&volume->buffer_lists[hi], buf, entry); 312 buf->cache.buffer = buf; 313 hammer_cache_add(&buf->cache); 314 dora = (isnew == 0); 315 } else { 316 assert(buf->ondisk != NULL); 317 assert(isnew != -1); 318 hammer_cache_used(&buf->cache); 319 } 320 321 ++buf->cache.refs; 322 hammer_cache_flush(); 323 324 if (isnew > 0) { 325 assert(buf->cache.modified == 0); 326 bzero(buf->ondisk, HAMMER_BUFSIZE); 327 buf->cache.modified = 1; 328 } 329 if (dora) 330 get_buffer_readahead(buf); 331 return(buf); 332 } 333 334 static void 335 get_buffer_readahead(struct buffer_info *base) 336 { 337 struct buffer_info *buf; 338 struct volume_info *vol; 339 hammer_off_t buf_offset; 340 int64_t raw_offset; 341 int ri = UseReadBehind; 342 int re = UseReadAhead; 343 344 raw_offset = base->raw_offset + ri * HAMMER_BUFSIZE; 345 vol = base->volume; 346 347 while (ri < re) { 348 if (raw_offset >= vol->ondisk->vol_buf_end) 349 break; 350 if (raw_offset < vol->ondisk->vol_buf_beg || ri == 0) { 351 ++ri; 352 raw_offset += HAMMER_BUFSIZE; 353 continue; 354 } 355 buf_offset = HAMMER_ENCODE_RAW_BUFFER(vol->vol_no, 356 raw_offset - vol->ondisk->vol_buf_beg); 357 buf = find_buffer(vol, buf_offset); 358 if (buf == NULL) { 359 buf = get_buffer(buf_offset, -1); 360 rel_buffer(buf); 361 } 362 ++ri; 363 raw_offset += HAMMER_BUFSIZE; 364 } 365 } 366 367 void 368 rel_buffer(struct buffer_info *buffer) 369 { 370 struct volume_info *volume; 371 int hi; 372 373 if (buffer == NULL) 374 return; 375 assert(buffer->cache.refs > 0); 376 if (--buffer->cache.refs == 0) { 377 if (buffer->cache.delete) { 378 hi = buffer_hash(buffer->buf_offset); 379 volume = buffer->volume; 380 if (buffer->cache.modified) 381 flush_buffer(buffer); 382 TAILQ_REMOVE(&volume->buffer_lists[hi], buffer, entry); 383 hammer_cache_del(&buffer->cache); 384 free(buffer->ondisk); 385 free(buffer); 386 } 387 } 388 } 389 390 /* 391 * Retrieve a pointer to a buffer data given a buffer offset. The underlying 392 * bufferp is freed if isnew or the offset is out of range of the cached data. 393 * If bufferp is freed a referenced buffer is loaded into it. 394 */ 395 void * 396 get_buffer_data(hammer_off_t buf_offset, struct buffer_info **bufferp, 397 int isnew) 398 { 399 if (*bufferp != NULL) { 400 if (isnew > 0 || 401 (((*bufferp)->buf_offset ^ buf_offset) & ~HAMMER_BUFMASK64)) { 402 rel_buffer(*bufferp); 403 *bufferp = NULL; 404 } 405 } 406 return(get_ondisk(buf_offset, bufferp, isnew)); 407 } 408 409 /* 410 * Retrieve a pointer to a B-Tree node given a zone offset. The underlying 411 * bufferp is freed if non-NULL and a referenced buffer is loaded into it. 412 */ 413 hammer_node_ondisk_t 414 get_node(hammer_off_t node_offset, struct buffer_info **bufferp) 415 { 416 if (*bufferp != NULL) { 417 rel_buffer(*bufferp); 418 *bufferp = NULL; 419 } 420 return(get_ondisk(node_offset, bufferp, 0)); 421 } 422 423 /* 424 * Return a pointer to a buffer data given a buffer offset. 425 * If *bufferp is NULL acquire the buffer otherwise use that buffer. 426 */ 427 static void * 428 get_ondisk(hammer_off_t buf_offset, struct buffer_info **bufferp, int isnew) 429 { 430 struct buffer_info *buffer; 431 432 buffer = *bufferp; 433 if (buffer == NULL) { 434 buffer = *bufferp = get_buffer(buf_offset, isnew); 435 if (buffer == NULL) 436 return(NULL); 437 } 438 439 return((char *)buffer->ondisk + 440 ((int32_t)buf_offset & HAMMER_BUFMASK)); 441 } 442 443 /* 444 * Allocate HAMMER elements - B-Tree nodes 445 */ 446 void * 447 alloc_btree_element(hammer_off_t *offp, struct buffer_info **data_bufferp) 448 { 449 hammer_node_ondisk_t node; 450 451 node = alloc_blockmap(HAMMER_ZONE_BTREE_INDEX, sizeof(*node), 452 offp, data_bufferp); 453 bzero(node, sizeof(*node)); 454 return (node); 455 } 456 457 /* 458 * Allocate HAMMER elements - meta data (inode, direntry, PFS, etc) 459 */ 460 void * 461 alloc_meta_element(hammer_off_t *offp, int32_t data_len, 462 struct buffer_info **data_bufferp) 463 { 464 void *data; 465 466 data = alloc_blockmap(HAMMER_ZONE_META_INDEX, data_len, 467 offp, data_bufferp); 468 bzero(data, data_len); 469 return (data); 470 } 471 472 /* 473 * Allocate HAMMER elements - data storage 474 * 475 * The only data_len supported by HAMMER userspace for large data zone 476 * (zone 10) is HAMMER_BUFSIZE which is 16KB. >16KB data does not fit 477 * in a buffer allocated by get_buffer(). Also alloc_blockmap() does 478 * not consider >16KB buffer size. 479 */ 480 void * 481 alloc_data_element(hammer_off_t *offp, int32_t data_len, 482 struct buffer_info **data_bufferp) 483 { 484 void *data; 485 int zone; 486 487 if (data_len == 0) 488 return(NULL); 489 490 zone = hammer_data_zone_index(data_len); 491 assert(data_len <= HAMMER_BUFSIZE); /* just one buffer */ 492 assert(zone == HAMMER_ZONE_LARGE_DATA_INDEX || 493 zone == HAMMER_ZONE_SMALL_DATA_INDEX); 494 495 data = alloc_blockmap(zone, data_len, offp, data_bufferp); 496 bzero(data, data_len); 497 return(data); 498 } 499 500 /* 501 * Format a new blockmap. This is mostly a degenerate case because 502 * all allocations are now actually done from the freemap. 503 */ 504 void 505 format_blockmap(struct volume_info *root_vol, int zone, hammer_off_t offset) 506 { 507 hammer_blockmap_t blockmap; 508 hammer_off_t zone_base; 509 510 /* Only root volume needs formatting */ 511 assert(root_vol->vol_no == HAMMER_ROOT_VOLNO); 512 513 assert(hammer_is_zone2_mapped_index(zone)); 514 515 blockmap = &root_vol->ondisk->vol0_blockmap[zone]; 516 zone_base = HAMMER_ZONE_ENCODE(zone, offset); 517 518 bzero(blockmap, sizeof(*blockmap)); 519 blockmap->phys_offset = 0; 520 blockmap->first_offset = zone_base; 521 blockmap->next_offset = zone_base; 522 blockmap->alloc_offset = HAMMER_ENCODE(zone, 255, -1); 523 blockmap->entry_crc = crc32(blockmap, HAMMER_BLOCKMAP_CRCSIZE); 524 } 525 526 /* 527 * Format a new freemap. Set all layer1 entries to UNAVAIL. The initialize 528 * code will load each volume's freemap. 529 */ 530 void 531 format_freemap(struct volume_info *root_vol) 532 { 533 struct buffer_info *buffer = NULL; 534 hammer_off_t layer1_offset; 535 hammer_blockmap_t blockmap; 536 struct hammer_blockmap_layer1 *layer1; 537 int i, isnew; 538 539 /* Only root volume needs formatting */ 540 assert(root_vol->vol_no == HAMMER_ROOT_VOLNO); 541 542 layer1_offset = alloc_bigblock(root_vol, HAMMER_ZONE_FREEMAP_INDEX); 543 for (i = 0; i < HAMMER_BIGBLOCK_SIZE; i += sizeof(*layer1)) { 544 isnew = ((i % HAMMER_BUFSIZE) == 0); 545 layer1 = get_buffer_data(layer1_offset + i, &buffer, isnew); 546 bzero(layer1, sizeof(*layer1)); 547 layer1->phys_offset = HAMMER_BLOCKMAP_UNAVAIL; 548 layer1->blocks_free = 0; 549 layer1->layer1_crc = crc32(layer1, HAMMER_LAYER1_CRCSIZE); 550 } 551 assert(i == HAMMER_BIGBLOCK_SIZE); 552 rel_buffer(buffer); 553 554 blockmap = &root_vol->ondisk->vol0_blockmap[HAMMER_ZONE_FREEMAP_INDEX]; 555 bzero(blockmap, sizeof(*blockmap)); 556 blockmap->phys_offset = layer1_offset; 557 blockmap->first_offset = 0; 558 blockmap->next_offset = HAMMER_ENCODE_RAW_BUFFER(0, 0); 559 blockmap->alloc_offset = HAMMER_ENCODE_RAW_BUFFER(255, -1); 560 blockmap->entry_crc = crc32(blockmap, HAMMER_BLOCKMAP_CRCSIZE); 561 } 562 563 /* 564 * Load the volume's remaining free space into the freemap. 565 * 566 * Returns the number of big-blocks available. 567 */ 568 int64_t 569 initialize_freemap(struct volume_info *vol) 570 { 571 struct volume_info *root_vol; 572 struct buffer_info *buffer1 = NULL; 573 struct buffer_info *buffer2 = NULL; 574 struct hammer_blockmap_layer1 *layer1; 575 struct hammer_blockmap_layer2 *layer2; 576 hammer_off_t layer1_offset; 577 hammer_off_t layer2_offset; 578 hammer_off_t phys_offset; 579 hammer_off_t block_offset; 580 hammer_off_t aligned_vol_free_end; 581 hammer_blockmap_t freemap; 582 int64_t count = 0; 583 int64_t layer1_count = 0; 584 585 root_vol = get_root_volume(); 586 aligned_vol_free_end = (vol->vol_free_end + HAMMER_BLOCKMAP_LAYER2_MASK) 587 & ~HAMMER_BLOCKMAP_LAYER2_MASK; 588 589 printf("initialize freemap volume %d\n", vol->vol_no); 590 591 /* 592 * Initialize the freemap. First preallocate the big-blocks required 593 * to implement layer2. This preallocation is a bootstrap allocation 594 * using blocks from the target volume. 595 */ 596 freemap = &root_vol->ondisk->vol0_blockmap[HAMMER_ZONE_FREEMAP_INDEX]; 597 598 for (phys_offset = HAMMER_ENCODE_RAW_BUFFER(vol->vol_no, 0); 599 phys_offset < aligned_vol_free_end; 600 phys_offset += HAMMER_BLOCKMAP_LAYER2) { 601 layer1_offset = freemap->phys_offset + 602 HAMMER_BLOCKMAP_LAYER1_OFFSET(phys_offset); 603 layer1 = get_buffer_data(layer1_offset, &buffer1, 0); 604 if (layer1->phys_offset == HAMMER_BLOCKMAP_UNAVAIL) { 605 layer1->phys_offset = alloc_bigblock(vol, 606 HAMMER_ZONE_FREEMAP_INDEX); 607 layer1->blocks_free = 0; 608 buffer1->cache.modified = 1; 609 layer1->layer1_crc = crc32(layer1, 610 HAMMER_LAYER1_CRCSIZE); 611 } 612 } 613 614 /* 615 * Now fill everything in. 616 */ 617 for (phys_offset = HAMMER_ENCODE_RAW_BUFFER(vol->vol_no, 0); 618 phys_offset < aligned_vol_free_end; 619 phys_offset += HAMMER_BLOCKMAP_LAYER2) { 620 layer1_count = 0; 621 layer1_offset = freemap->phys_offset + 622 HAMMER_BLOCKMAP_LAYER1_OFFSET(phys_offset); 623 layer1 = get_buffer_data(layer1_offset, &buffer1, 0); 624 assert(layer1->phys_offset != HAMMER_BLOCKMAP_UNAVAIL); 625 626 for (block_offset = 0; 627 block_offset < HAMMER_BLOCKMAP_LAYER2; 628 block_offset += HAMMER_BIGBLOCK_SIZE) { 629 layer2_offset = layer1->phys_offset + 630 HAMMER_BLOCKMAP_LAYER2_OFFSET(block_offset); 631 layer2 = get_buffer_data(layer2_offset, &buffer2, 0); 632 bzero(layer2, sizeof(*layer2)); 633 634 if (phys_offset + block_offset < vol->vol_free_off) { 635 /* 636 * Fixups XXX - big-blocks already allocated as part 637 * of the freemap bootstrap. 638 */ 639 layer2->zone = HAMMER_ZONE_FREEMAP_INDEX; 640 layer2->append_off = HAMMER_BIGBLOCK_SIZE; 641 layer2->bytes_free = 0; 642 } else if (phys_offset + block_offset < vol->vol_free_end) { 643 layer2->zone = 0; 644 layer2->append_off = 0; 645 layer2->bytes_free = HAMMER_BIGBLOCK_SIZE; 646 ++count; 647 ++layer1_count; 648 } else { 649 layer2->zone = HAMMER_ZONE_UNAVAIL_INDEX; 650 layer2->append_off = HAMMER_BIGBLOCK_SIZE; 651 layer2->bytes_free = 0; 652 } 653 layer2->entry_crc = crc32(layer2, HAMMER_LAYER2_CRCSIZE); 654 buffer2->cache.modified = 1; 655 } 656 657 layer1->blocks_free += layer1_count; 658 layer1->layer1_crc = crc32(layer1, HAMMER_LAYER1_CRCSIZE); 659 buffer1->cache.modified = 1; 660 } 661 662 rel_buffer(buffer1); 663 rel_buffer(buffer2); 664 return(count); 665 } 666 667 /* 668 * Returns the number of big-blocks available for filesystem data and undos 669 * without formatting. 670 */ 671 int64_t 672 count_freemap(struct volume_info *vol) 673 { 674 hammer_off_t phys_offset; 675 hammer_off_t vol_free_off; 676 hammer_off_t aligned_vol_free_end; 677 int64_t count = 0; 678 679 vol_free_off = HAMMER_ENCODE_RAW_BUFFER(vol->vol_no, 0); 680 aligned_vol_free_end = (vol->vol_free_end + HAMMER_BLOCKMAP_LAYER2_MASK) 681 & ~HAMMER_BLOCKMAP_LAYER2_MASK; 682 683 if (vol->vol_no == HAMMER_ROOT_VOLNO) 684 vol_free_off += HAMMER_BIGBLOCK_SIZE; 685 686 for (phys_offset = HAMMER_ENCODE_RAW_BUFFER(vol->vol_no, 0); 687 phys_offset < aligned_vol_free_end; 688 phys_offset += HAMMER_BLOCKMAP_LAYER2) { 689 vol_free_off += HAMMER_BIGBLOCK_SIZE; 690 } 691 692 for (phys_offset = HAMMER_ENCODE_RAW_BUFFER(vol->vol_no, 0); 693 phys_offset < aligned_vol_free_end; 694 phys_offset += HAMMER_BIGBLOCK_SIZE) { 695 if (phys_offset < vol_free_off) { 696 ; 697 } else if (phys_offset < vol->vol_free_end) { 698 ++count; 699 } 700 } 701 702 return(count); 703 } 704 705 /* 706 * Format the undomap for the root volume. 707 */ 708 void 709 format_undomap(struct volume_info *root_vol, int64_t *undo_buffer_size) 710 { 711 const int undo_zone = HAMMER_ZONE_UNDO_INDEX; 712 hammer_off_t undo_limit; 713 hammer_blockmap_t blockmap; 714 struct hammer_volume_ondisk *ondisk; 715 struct buffer_info *buffer = NULL; 716 hammer_off_t scan; 717 int n; 718 int limit_index; 719 uint32_t seqno; 720 721 /* Only root volume needs formatting */ 722 assert(root_vol->vol_no == HAMMER_ROOT_VOLNO); 723 ondisk = root_vol->ondisk; 724 725 /* 726 * Size the undo buffer in multiples of HAMMER_BIGBLOCK_SIZE, 727 * up to HAMMER_UNDO_LAYER2 big-blocks. Size to approximately 728 * 0.1% of the disk. 729 * 730 * The minimum UNDO fifo size is 500MB, or approximately 1% of 731 * the recommended 50G disk. 732 * 733 * Changing this minimum is rather dangerous as complex filesystem 734 * operations can cause the UNDO FIFO to fill up otherwise. 735 */ 736 undo_limit = *undo_buffer_size; 737 if (undo_limit == 0) { 738 undo_limit = HAMMER_VOL_BUF_SIZE(ondisk) / 1000; 739 if (undo_limit < 500*1024*1024) 740 undo_limit = 500*1024*1024; 741 } 742 undo_limit = (undo_limit + HAMMER_BIGBLOCK_MASK64) & 743 ~HAMMER_BIGBLOCK_MASK64; 744 if (undo_limit < HAMMER_BIGBLOCK_SIZE) 745 undo_limit = HAMMER_BIGBLOCK_SIZE; 746 if (undo_limit > HAMMER_BIGBLOCK_SIZE * HAMMER_UNDO_LAYER2) 747 undo_limit = HAMMER_BIGBLOCK_SIZE * HAMMER_UNDO_LAYER2; 748 *undo_buffer_size = undo_limit; 749 750 blockmap = &ondisk->vol0_blockmap[undo_zone]; 751 bzero(blockmap, sizeof(*blockmap)); 752 blockmap->phys_offset = HAMMER_BLOCKMAP_UNAVAIL; 753 blockmap->first_offset = HAMMER_ZONE_ENCODE(undo_zone, 0); 754 blockmap->next_offset = blockmap->first_offset; 755 blockmap->alloc_offset = HAMMER_ZONE_ENCODE(undo_zone, undo_limit); 756 blockmap->entry_crc = crc32(blockmap, HAMMER_BLOCKMAP_CRCSIZE); 757 758 limit_index = undo_limit / HAMMER_BIGBLOCK_SIZE; 759 assert(limit_index <= HAMMER_UNDO_LAYER2); 760 761 for (n = 0; n < limit_index; ++n) { 762 ondisk->vol0_undo_array[n] = alloc_bigblock(root_vol, 763 HAMMER_ZONE_UNDO_INDEX); 764 } 765 while (n < HAMMER_UNDO_LAYER2) { 766 ondisk->vol0_undo_array[n++] = HAMMER_BLOCKMAP_UNAVAIL; 767 } 768 769 /* 770 * Pre-initialize the UNDO blocks (HAMMER version 4+) 771 */ 772 printf("initializing the undo map (%jd MB)\n", 773 (intmax_t)(blockmap->alloc_offset & HAMMER_OFF_LONG_MASK) / 774 (1024 * 1024)); 775 776 scan = blockmap->first_offset; 777 seqno = 0; 778 779 while (scan < blockmap->alloc_offset) { 780 hammer_fifo_head_t head; 781 hammer_fifo_tail_t tail; 782 int isnew; 783 int bytes = HAMMER_UNDO_ALIGN; 784 785 isnew = ((scan & HAMMER_BUFMASK64) == 0); 786 head = get_buffer_data(scan, &buffer, isnew); 787 buffer->cache.modified = 1; 788 tail = (void *)((char *)head + bytes - sizeof(*tail)); 789 790 bzero(head, bytes); 791 head->hdr_signature = HAMMER_HEAD_SIGNATURE; 792 head->hdr_type = HAMMER_HEAD_TYPE_DUMMY; 793 head->hdr_size = bytes; 794 head->hdr_seq = seqno++; 795 796 tail->tail_signature = HAMMER_TAIL_SIGNATURE; 797 tail->tail_type = HAMMER_HEAD_TYPE_DUMMY; 798 tail->tail_size = bytes; 799 800 head->hdr_crc = crc32(head, HAMMER_FIFO_HEAD_CRCOFF) ^ 801 crc32(head + 1, bytes - sizeof(*head)); 802 803 scan += bytes; 804 } 805 rel_buffer(buffer); 806 } 807 808 /* 809 * Flush various tracking structures to disk 810 */ 811 void 812 flush_all_volumes(void) 813 { 814 struct volume_info *vol; 815 816 TAILQ_FOREACH(vol, &VolList, entry) 817 flush_volume(vol); 818 } 819 820 void 821 flush_volume(struct volume_info *volume) 822 { 823 struct buffer_info *buffer; 824 int i; 825 826 for (i = 0; i < HAMMER_BUFLISTS; ++i) { 827 TAILQ_FOREACH(buffer, &volume->buffer_lists[i], entry) 828 flush_buffer(buffer); 829 } 830 if (writehammervol(volume) == -1) 831 err(1, "Write volume %d (%s)", volume->vol_no, volume->name); 832 } 833 834 void 835 flush_buffer(struct buffer_info *buffer) 836 { 837 struct volume_info *vol; 838 839 vol = buffer->volume; 840 if (writehammerbuf(buffer) == -1) 841 err(1, "Write volume %d (%s)", vol->vol_no, vol->name); 842 buffer->cache.modified = 0; 843 } 844 845 /* 846 * Core I/O operations 847 */ 848 static int 849 __read(struct volume_info *vol, void *data, int64_t offset, int size) 850 { 851 ssize_t n; 852 853 n = pread(vol->fd, data, size, offset); 854 if (n != size) 855 return(-1); 856 return(0); 857 } 858 859 static __inline int 860 readhammervol(struct volume_info *vol) 861 { 862 return(__read(vol, vol->ondisk, 0, HAMMER_BUFSIZE)); 863 } 864 865 static __inline int 866 readhammerbuf(struct buffer_info *buf) 867 { 868 return(__read(buf->volume, buf->ondisk, buf->raw_offset, HAMMER_BUFSIZE)); 869 } 870 871 static int 872 __write(struct volume_info *vol, const void *data, int64_t offset, int size) 873 { 874 ssize_t n; 875 876 if (vol->rdonly) 877 return(0); 878 879 n = pwrite(vol->fd, data, size, offset); 880 if (n != size) 881 return(-1); 882 return(0); 883 } 884 885 static __inline int 886 writehammervol(struct volume_info *vol) 887 { 888 return(__write(vol, vol->ondisk, 0, HAMMER_BUFSIZE)); 889 } 890 891 static __inline int 892 writehammerbuf(struct buffer_info *buf) 893 { 894 return(__write(buf->volume, buf->ondisk, buf->raw_offset, HAMMER_BUFSIZE)); 895 } 896 897 int64_t init_boot_area_size(int64_t value, off_t avg_vol_size) 898 { 899 if (value == 0) { 900 value = HAMMER_BOOT_NOMBYTES; 901 while (value > avg_vol_size / HAMMER_MAX_VOLUMES) 902 value >>= 1; 903 if (value < HAMMER_BOOT_MINBYTES) 904 value = 0; 905 } else if (value < HAMMER_BOOT_MINBYTES) { 906 value = HAMMER_BOOT_MINBYTES; 907 } 908 909 return(value); 910 } 911 912 int64_t init_mem_area_size(int64_t value, off_t avg_vol_size) 913 { 914 if (value == 0) { 915 value = HAMMER_MEM_NOMBYTES; 916 while (value > avg_vol_size / HAMMER_MAX_VOLUMES) 917 value >>= 1; 918 if (value < HAMMER_MEM_MINBYTES) 919 value = 0; 920 } else if (value < HAMMER_MEM_MINBYTES) { 921 value = HAMMER_MEM_MINBYTES; 922 } 923 924 return(value); 925 } 926