sbin/hammer/ondisk.c

   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 check_volume(struct volume_info *vol);
  41 static void get_buffer_readahead(struct buffer_info *base);
  42 static __inline int readhammervol(struct volume_info *vol);
  43 static __inline int readhammerbuf(struct buffer_info *buf);
  44 static __inline int writehammervol(struct volume_info *vol);
  45 static __inline int writehammerbuf(struct buffer_info *buf);
  46
  47 uuid_t Hammer_FSType;
  48 uuid_t Hammer_FSId;
  49 int UseReadBehind = -4;
  50 int UseReadAhead = 4;
  51 int DebugOpt;
  52
  53 TAILQ_HEAD(volume_list, volume_info);
  54 static struct volume_list VolList = TAILQ_HEAD_INITIALIZER(VolList);
  55 static int valid_hammer_volumes;
  56
  57 static __inline
  58 int
  59 buffer_hash(hammer_off_t buf_offset)
  60 {
  61         int hi;
  62
  63         hi = (int)(buf_offset / HAMMER_BUFSIZE) & HAMMER_BUFLISTMASK;
  64         return(hi);
  65 }
  66
  67 static struct buffer_info*
  68 find_buffer(struct volume_info *volume, hammer_off_t buf_offset)
  69 {
  70         int hi;
  71         struct buffer_info *buf;
  72
  73         hi = buffer_hash(buf_offset);
  74         TAILQ_FOREACH(buf, &volume->buffer_lists[hi], entry)
  75                 if (buf->buf_offset == buf_offset)
  76                         return(buf);
  77         return(NULL);
  78 }
  79
  80 static
  81 struct volume_info *
  82 __alloc_volume(const char *volname, int oflags)
  83 {
  84         struct volume_info *vol;
  85         int i;
  86
  87         vol = malloc(sizeof(*vol));
  88         if (vol == NULL)
  89                 err(1, "alloc_volume");
  90         bzero(vol, sizeof(*vol));
  91
  92         vol->vol_no = -1;
  93         vol->rdonly = (oflags == O_RDONLY);
  94         vol->name = strdup(volname);
  95         vol->fd = open(vol->name, oflags);
  96         if (vol->fd < 0)
  97                 err(1, "alloc_volume: Failed to open %s", vol->name);
  98         check_volume(vol);
  99
 100         vol->ondisk = malloc(HAMMER_BUFSIZE);
 101         if (vol->ondisk == NULL)
 102                 err(1, "alloc_volume");
 103         bzero(vol->ondisk, HAMMER_BUFSIZE);
 104
 105         for (i = 0; i < HAMMER_BUFLISTS; ++i)
 106                 TAILQ_INIT(&vol->buffer_lists[i]);
 107
 108         return(vol);
 109 }
 110
 111 static void
 112 __add_volume(struct volume_info *vol)
 113 {
 114         struct volume_info *scan;
 115         struct stat st1, st2;
 116
 117         if (fstat(vol->fd, &st1) != 0)
 118                 errx(1, "add_volume: %s: Failed to stat", vol->name);
 119
 120         TAILQ_FOREACH(scan, &VolList, entry) {
 121                 if (scan->vol_no == vol->vol_no) {
 122                         errx(1, "add_volume: %s: Duplicate volume number %d "
 123                                 "against %s",
 124                                 vol->name, vol->vol_no, scan->name);
 125                 }
 126                 if (fstat(scan->fd, &st2) != 0) {
 127                         errx(1, "add_volume: %s: Failed to stat %s",
 128                                 vol->name, scan->name);
 129                 }
 130                 if ((st1.st_ino == st2.st_ino) && (st1.st_dev == st2.st_dev)) {
 131                         errx(1, "add_volume: %s: Specified more than once",
 132                                 vol->name);
 133                 }
 134         }
 135
 136         TAILQ_INSERT_TAIL(&VolList, vol, entry);
 137 }
 138
 139 static void
 140 __verify_volume(struct volume_info *vol)
 141 {
 142         hammer_volume_ondisk_t ondisk = vol->ondisk;
 143
 144         if (ondisk->vol_signature != HAMMER_FSBUF_VOLUME) {
 145                 errx(1, "verify_volume: Invalid volume signature %016jx",
 146                         ondisk->vol_signature);
 147         }
 148         if (ondisk->vol_rootvol != HAMMER_ROOT_VOLNO) {
 149                 errx(1, "verify_volume: Invalid root volume# %d",
 150                         ondisk->vol_rootvol);
 151         }
 152         if (bcmp(&Hammer_FSType, &ondisk->vol_fstype, sizeof(Hammer_FSType))) {
 153                 errx(1, "verify_volume: %s: Header does not indicate "
 154                         "that this is a HAMMER volume", vol->name);
 155         }
 156         if (bcmp(&Hammer_FSId, &ondisk->vol_fsid, sizeof(Hammer_FSId))) {
 157                 errx(1, "verify_volume: %s: FSId does not match other volumes!",
 158                         vol->name);
 159         }
 160 }
 161
 162 /*
 163  * Initialize a volume structure and ondisk vol_no field.
 164  */
 165 struct volume_info *
 166 init_volume(const char *filename, int oflags, int32_t vol_no)
 167 {
 168         struct volume_info *vol;
 169
 170         vol = __alloc_volume(filename, oflags);
 171         vol->vol_no = vol->ondisk->vol_no = vol_no;
 172
 173         __add_volume(vol);
 174
 175         return(vol);
 176 }
 177
 178 /*
 179  * Initialize a volume structure and read ondisk volume header.
 180  */
 181 struct volume_info*
 182 load_volume(const char *filename, int oflags, int verify)
 183 {
 184         struct volume_info *vol;
 185         int n;
 186
 187         vol = __alloc_volume(filename, oflags);
 188
 189         n = readhammervol(vol);
 190         if (n == -1) {
 191                 err(1, "load_volume: %s: Read failed at offset 0", vol->name);
 192         }
 193         vol->vol_no = vol->ondisk->vol_no;
 194
 195         if (valid_hammer_volumes++ == 0)
 196                 Hammer_FSId = vol->ondisk->vol_fsid;
 197         if (verify)
 198                 __verify_volume(vol);
 199
 200         __add_volume(vol);
 201
 202         return(vol);
 203 }
 204
 205 /*
 206  * Check basic volume characteristics.
 207  */
 208 static void
 209 check_volume(struct volume_info *vol)
 210 {
 211         struct partinfo pinfo;
 212         struct stat st;
 213
 214         /*
 215          * Get basic information about the volume
 216          */
 217         if (ioctl(vol->fd, DIOCGPART, &pinfo) < 0) {
 218                 /*
 219                  * Allow the formatting of regular files as HAMMER volumes
 220                  */
 221                 if (fstat(vol->fd, &st) < 0)
 222                         err(1, "Unable to stat %s", vol->name);
 223                 vol->size = st.st_size;
 224                 vol->type = "REGFILE";
 225         } else {
 226                 /*
 227                  * When formatting a block device as a HAMMER volume the
 228                  * sector size must be compatible.  HAMMER uses 16384 byte
 229                  * filesystem buffers.
 230                  */
 231                 if (pinfo.reserved_blocks) {
 232                         errx(1, "HAMMER cannot be placed in a partition "
 233                                 "which overlaps the disklabel or MBR");
 234                 }
 235                 if (pinfo.media_blksize > HAMMER_BUFSIZE ||
 236                     HAMMER_BUFSIZE % pinfo.media_blksize) {
 237                         errx(1, "A media sector size of %d is not supported",
 238                              pinfo.media_blksize);
 239                 }
 240
 241                 vol->size = pinfo.media_size;
 242                 vol->device_offset = pinfo.media_offset;
 243                 vol->type = "DEVICE";
 244         }
 245 }
 246
 247 void
 248 assert_volume_offset(struct volume_info *vol)
 249 {
 250         assert(hammer_is_zone_raw_buffer(vol->vol_free_off));
 251         assert(hammer_is_zone_raw_buffer(vol->vol_free_end));
 252         if (vol->vol_free_off >= vol->vol_free_end)
 253                 errx(1, "Ran out of room, filesystem too small");
 254 }
 255
 256 struct volume_info *
 257 get_volume(int32_t vol_no)
 258 {
 259         struct volume_info *vol;
 260
 261         TAILQ_FOREACH(vol, &VolList, entry) {
 262                 if (vol->vol_no == vol_no)
 263                         break;
 264         }
 265
 266         return(vol);
 267 }
 268
 269 struct volume_info *
 270 get_root_volume(void)
 271 {
 272         return(get_volume(HAMMER_ROOT_VOLNO));
 273 }
 274
 275 /*
 276  * Acquire the specified buffer.  isnew is -1 only when called
 277  * via get_buffer_readahead() to prevent another readahead.
 278  */
 279 static struct buffer_info *
 280 get_buffer(hammer_off_t buf_offset, int isnew)
 281 {
 282         struct buffer_info *buf;
 283         struct volume_info *volume;
 284         int vol_no;
 285         int zone;
 286         int hi;
 287         int dora = 0;
 288         int error = 0;
 289
 290         zone = HAMMER_ZONE_DECODE(buf_offset);
 291         if (zone > HAMMER_ZONE_RAW_BUFFER_INDEX)
 292                 buf_offset = blockmap_lookup(buf_offset, NULL, NULL, &error);
 293         if (error || buf_offset == HAMMER_OFF_BAD)
 294                 return(NULL);
 295         assert(hammer_is_zone_raw_buffer(buf_offset));
 296
 297         vol_no = HAMMER_VOL_DECODE(buf_offset);
 298         volume = get_volume(vol_no);
 299         assert(volume != NULL);
 300
 301         buf_offset &= ~HAMMER_BUFMASK64;
 302         buf = find_buffer(volume, buf_offset);
 303
 304         if (buf == NULL) {
 305                 buf = malloc(sizeof(*buf));
 306                 bzero(buf, sizeof(*buf));
 307                 buf->buf_offset = buf_offset;
 308                 buf->raw_offset = hammer_xlate_to_phys(volume->ondisk,
 309                                                         buf_offset);
 310                 buf->volume = volume;
 311                 buf->ondisk = malloc(HAMMER_BUFSIZE);
 312                 if (isnew <= 0) {
 313                         if (readhammerbuf(buf) == -1) {
 314                                 err(1, "get_buffer: %s:%016jx "
 315                                     "Read failed at offset %016jx",
 316                                     volume->name,
 317                                     (intmax_t)buf->buf_offset,
 318                                     (intmax_t)buf->raw_offset);
 319                         }
 320                 }
 321
 322                 hi = buffer_hash(buf_offset);
 323                 TAILQ_INSERT_TAIL(&volume->buffer_lists[hi], buf, entry);
 324                 hammer_cache_add(&buf->cache);
 325                 dora = (isnew == 0);
 326         } else {
 327                 assert(buf->ondisk != NULL);
 328                 assert(isnew != -1);
 329                 hammer_cache_used(&buf->cache);
 330         }
 331
 332         ++buf->cache.refs;
 333         hammer_cache_flush();
 334
 335         if (isnew > 0) {
 336                 assert(buf->cache.modified == 0);
 337                 bzero(buf->ondisk, HAMMER_BUFSIZE);
 338                 buf->cache.modified = 1;
 339         }
 340         if (dora)
 341                 get_buffer_readahead(buf);
 342         return(buf);
 343 }
 344
 345 static void
 346 get_buffer_readahead(struct buffer_info *base)
 347 {
 348         struct buffer_info *buf;
 349         struct volume_info *vol;
 350         hammer_off_t buf_offset;
 351         int64_t raw_offset;
 352         int ri = UseReadBehind;
 353         int re = UseReadAhead;
 354
 355         raw_offset = base->raw_offset + ri * HAMMER_BUFSIZE;
 356         vol = base->volume;
 357
 358         while (ri < re) {
 359                 if (raw_offset >= vol->ondisk->vol_buf_end)
 360                         break;
 361                 if (raw_offset < vol->ondisk->vol_buf_beg || ri == 0) {
 362                         ++ri;
 363                         raw_offset += HAMMER_BUFSIZE;
 364                         continue;
 365                 }
 366                 buf_offset = HAMMER_ENCODE_RAW_BUFFER(vol->vol_no,
 367                         raw_offset - vol->ondisk->vol_buf_beg);
 368                 buf = find_buffer(vol, buf_offset);
 369                 if (buf == NULL) {
 370                         buf = get_buffer(buf_offset, -1);
 371                         rel_buffer(buf);
 372                 }
 373                 ++ri;
 374                 raw_offset += HAMMER_BUFSIZE;
 375         }
 376 }
 377
 378 void
 379 rel_buffer(struct buffer_info *buffer)
 380 {
 381         struct volume_info *volume;
 382         int hi;
 383
 384         if (buffer == NULL)
 385                 return;
 386         assert(buffer->cache.refs > 0);
 387         if (--buffer->cache.refs == 0) {
 388                 if (buffer->cache.delete) {
 389                         hi = buffer_hash(buffer->buf_offset);
 390                         volume = buffer->volume;
 391                         if (buffer->cache.modified)
 392                                 flush_buffer(buffer);
 393                         TAILQ_REMOVE(&volume->buffer_lists[hi], buffer, entry);
 394                         hammer_cache_del(&buffer->cache);
 395                         free(buffer->ondisk);
 396                         free(buffer);
 397                 }
 398         }
 399 }
 400
 401 /*
 402  * Retrieve a pointer to a buffer data given a buffer offset.  The underlying
 403  * bufferp is freed if isnew or the offset is out of range of the cached data.
 404  * If bufferp is freed a referenced buffer is loaded into it.
 405  */
 406 void *
 407 get_buffer_data(hammer_off_t buf_offset, struct buffer_info **bufferp,
 408                 int isnew)
 409 {
 410         if (*bufferp != NULL) {
 411                 if (isnew > 0 ||
 412                     (((*bufferp)->buf_offset ^ buf_offset) & ~HAMMER_BUFMASK64)) {
 413                         rel_buffer(*bufferp);
 414                         *bufferp = NULL;
 415                 }
 416         }
 417
 418         if (*bufferp == NULL) {
 419                 *bufferp = get_buffer(buf_offset, isnew);
 420                 if (*bufferp == NULL)
 421                         return(NULL);
 422         }
 423
 424         return(((char *)(*bufferp)->ondisk) +
 425                 ((int32_t)buf_offset & HAMMER_BUFMASK));
 426 }
 427
 428 /*
 429  * Allocate HAMMER elements - B-Tree nodes
 430  */
 431 hammer_node_ondisk_t
 432 alloc_btree_node(hammer_off_t *offp, struct buffer_info **data_bufferp)
 433 {
 434         hammer_node_ondisk_t node;
 435
 436         node = alloc_blockmap(HAMMER_ZONE_BTREE_INDEX, sizeof(*node),
 437                               offp, data_bufferp);
 438         bzero(node, sizeof(*node));
 439         return(node);
 440 }
 441
 442 /*
 443  * Allocate HAMMER elements - meta data (inode, direntry, PFS, etc)
 444  */
 445 void *
 446 alloc_meta_element(hammer_off_t *offp, int32_t data_len,
 447                    struct buffer_info **data_bufferp)
 448 {
 449         void *data;
 450
 451         data = alloc_blockmap(HAMMER_ZONE_META_INDEX, data_len,
 452                               offp, data_bufferp);
 453         bzero(data, data_len);
 454         return(data);
 455 }
 456
 457 /*
 458  * Format a new blockmap.  This is mostly a degenerate case because
 459  * all allocations are now actually done from the freemap.
 460  */
 461 void
 462 format_blockmap(struct volume_info *root_vol, int zone, hammer_off_t offset)
 463 {
 464         hammer_blockmap_t blockmap;
 465         hammer_off_t zone_base;
 466
 467         /* Only root volume needs formatting */
 468         assert(root_vol->vol_no == HAMMER_ROOT_VOLNO);
 469
 470         assert(hammer_is_zone2_mapped_index(zone));
 471
 472         blockmap = &root_vol->ondisk->vol0_blockmap[zone];
 473         zone_base = HAMMER_ZONE_ENCODE(zone, offset);
 474
 475         bzero(blockmap, sizeof(*blockmap));
 476         blockmap->phys_offset = 0;
 477         blockmap->first_offset = zone_base;
 478         blockmap->next_offset = zone_base;
 479         blockmap->alloc_offset = HAMMER_ENCODE(zone, 255, -1);
 480         hammer_crc_set_blockmap(blockmap);
 481 }
 482
 483 /*
 484  * Format a new freemap.  Set all layer1 entries to UNAVAIL.  The initialize
 485  * code will load each volume's freemap.
 486  */
 487 void
 488 format_freemap(struct volume_info *root_vol)
 489 {
 490         struct buffer_info *buffer = NULL;
 491         hammer_off_t layer1_offset;
 492         hammer_blockmap_t blockmap;
 493         hammer_blockmap_layer1_t layer1;
 494         int i, isnew;
 495
 496         /* Only root volume needs formatting */
 497         assert(root_vol->vol_no == HAMMER_ROOT_VOLNO);
 498
 499         layer1_offset = bootstrap_bigblock(root_vol);
 500         for (i = 0; i < HAMMER_BIGBLOCK_SIZE; i += sizeof(*layer1)) {
 501                 isnew = ((i % HAMMER_BUFSIZE) == 0);
 502                 layer1 = get_buffer_data(layer1_offset + i, &buffer, isnew);
 503                 bzero(layer1, sizeof(*layer1));
 504                 layer1->phys_offset = HAMMER_BLOCKMAP_UNAVAIL;
 505                 layer1->blocks_free = 0;
 506                 hammer_crc_set_layer1(layer1);
 507         }
 508         assert(i == HAMMER_BIGBLOCK_SIZE);
 509         rel_buffer(buffer);
 510
 511         blockmap = &root_vol->ondisk->vol0_blockmap[HAMMER_ZONE_FREEMAP_INDEX];
 512         bzero(blockmap, sizeof(*blockmap));
 513         blockmap->phys_offset = layer1_offset;
 514         blockmap->first_offset = 0;
 515         blockmap->next_offset = HAMMER_ENCODE_RAW_BUFFER(0, 0);
 516         blockmap->alloc_offset = HAMMER_ENCODE_RAW_BUFFER(255, -1);
 517         hammer_crc_set_blockmap(blockmap);
 518 }
 519
 520 /*
 521  * Load the volume's remaining free space into the freemap.
 522  *
 523  * Returns the number of big-blocks available.
 524  */
 525 int64_t
 526 initialize_freemap(struct volume_info *vol)
 527 {
 528         struct volume_info *root_vol;
 529         struct buffer_info *buffer1 = NULL;
 530         struct buffer_info *buffer2 = NULL;
 531         hammer_blockmap_layer1_t layer1;
 532         hammer_blockmap_layer2_t layer2;
 533         hammer_off_t layer1_offset;
 534         hammer_off_t layer2_offset;
 535         hammer_off_t phys_offset;
 536         hammer_off_t block_offset;
 537         hammer_off_t aligned_vol_free_end;
 538         hammer_blockmap_t freemap;
 539         int64_t count = 0;
 540         int64_t layer1_count = 0;
 541
 542         root_vol = get_root_volume();
 543
 544         assert_volume_offset(vol);
 545         aligned_vol_free_end = HAMMER_BLOCKMAP_LAYER2_DOALIGN(vol->vol_free_end);
 546
 547         printf("initialize freemap volume %d\n", vol->vol_no);
 548
 549         /*
 550          * Initialize the freemap.  First preallocate the big-blocks required
 551          * to implement layer2.   This preallocation is a bootstrap allocation
 552          * using blocks from the target volume.
 553          */
 554         freemap = &root_vol->ondisk->vol0_blockmap[HAMMER_ZONE_FREEMAP_INDEX];
 555
 556         for (phys_offset = HAMMER_ENCODE_RAW_BUFFER(vol->vol_no, 0);
 557              phys_offset < aligned_vol_free_end;
 558              phys_offset += HAMMER_BLOCKMAP_LAYER2) {
 559                 layer1_offset = freemap->phys_offset +
 560                                 HAMMER_BLOCKMAP_LAYER1_OFFSET(phys_offset);
 561                 layer1 = get_buffer_data(layer1_offset, &buffer1, 0);
 562                 if (layer1->phys_offset == HAMMER_BLOCKMAP_UNAVAIL) {
 563                         layer1->phys_offset = bootstrap_bigblock(vol);
 564                         layer1->blocks_free = 0;
 565                         buffer1->cache.modified = 1;
 566                         hammer_crc_set_layer1(layer1);
 567                 }
 568         }
 569
 570         /*
 571          * Now fill everything in.
 572          */
 573         for (phys_offset = HAMMER_ENCODE_RAW_BUFFER(vol->vol_no, 0);
 574              phys_offset < aligned_vol_free_end;
 575              phys_offset += HAMMER_BLOCKMAP_LAYER2) {
 576                 layer1_count = 0;
 577                 layer1_offset = freemap->phys_offset +
 578                                 HAMMER_BLOCKMAP_LAYER1_OFFSET(phys_offset);
 579                 layer1 = get_buffer_data(layer1_offset, &buffer1, 0);
 580                 assert(layer1->phys_offset != HAMMER_BLOCKMAP_UNAVAIL);
 581
 582                 for (block_offset = 0;
 583                      block_offset < HAMMER_BLOCKMAP_LAYER2;
 584                      block_offset += HAMMER_BIGBLOCK_SIZE) {
 585                         layer2_offset = layer1->phys_offset +
 586                                         HAMMER_BLOCKMAP_LAYER2_OFFSET(block_offset);
 587                         layer2 = get_buffer_data(layer2_offset, &buffer2, 0);
 588                         bzero(layer2, sizeof(*layer2));
 589
 590                         if (phys_offset + block_offset < vol->vol_free_off) {
 591                                 /*
 592                                  * Big-blocks already allocated as part
 593                                  * of the freemap bootstrap.
 594                                  */
 595                                 layer2->zone = HAMMER_ZONE_FREEMAP_INDEX;
 596                                 layer2->append_off = HAMMER_BIGBLOCK_SIZE;
 597                                 layer2->bytes_free = 0;
 598                         } else if (phys_offset + block_offset < vol->vol_free_end) {
 599                                 layer2->zone = 0;
 600                                 layer2->append_off = 0;
 601                                 layer2->bytes_free = HAMMER_BIGBLOCK_SIZE;
 602                                 ++count;
 603                                 ++layer1_count;
 604                         } else {
 605                                 layer2->zone = HAMMER_ZONE_UNAVAIL_INDEX;
 606                                 layer2->append_off = HAMMER_BIGBLOCK_SIZE;
 607                                 layer2->bytes_free = 0;
 608                         }
 609                         hammer_crc_set_layer2(layer2);
 610                         buffer2->cache.modified = 1;
 611                 }
 612
 613                 layer1->blocks_free += layer1_count;
 614                 hammer_crc_set_layer1(layer1);
 615                 buffer1->cache.modified = 1;
 616         }
 617
 618         rel_buffer(buffer1);
 619         rel_buffer(buffer2);
 620         return(count);
 621 }
 622
 623 /*
 624  * Returns the number of big-blocks available for filesystem data and undos
 625  * without formatting.
 626  */
 627 int64_t
 628 count_freemap(struct volume_info *vol)
 629 {
 630         hammer_off_t phys_offset;
 631         hammer_off_t vol_free_off;
 632         hammer_off_t aligned_vol_free_end;
 633         int64_t count = 0;
 634
 635         vol_free_off = HAMMER_ENCODE_RAW_BUFFER(vol->vol_no, 0);
 636
 637         assert_volume_offset(vol);
 638         aligned_vol_free_end = HAMMER_BLOCKMAP_LAYER2_DOALIGN(vol->vol_free_end);
 639
 640         if (vol->vol_no == HAMMER_ROOT_VOLNO)
 641                 vol_free_off += HAMMER_BIGBLOCK_SIZE;
 642
 643         for (phys_offset = HAMMER_ENCODE_RAW_BUFFER(vol->vol_no, 0);
 644              phys_offset < aligned_vol_free_end;
 645              phys_offset += HAMMER_BLOCKMAP_LAYER2) {
 646                 vol_free_off += HAMMER_BIGBLOCK_SIZE;
 647         }
 648
 649         for (phys_offset = HAMMER_ENCODE_RAW_BUFFER(vol->vol_no, 0);
 650              phys_offset < aligned_vol_free_end;
 651              phys_offset += HAMMER_BIGBLOCK_SIZE) {
 652                 if (phys_offset < vol_free_off) {
 653                         ;
 654                 } else if (phys_offset < vol->vol_free_end) {
 655                         ++count;
 656                 }
 657         }
 658
 659         return(count);
 660 }
 661
 662 /*
 663  * Format the undomap for the root volume.
 664  */
 665 void
 666 format_undomap(struct volume_info *root_vol, int64_t *undo_buffer_size)
 667 {
 668         hammer_off_t undo_limit;
 669         hammer_blockmap_t blockmap;
 670         hammer_volume_ondisk_t ondisk;
 671         struct buffer_info *buffer = NULL;
 672         hammer_off_t scan;
 673         int n;
 674         int limit_index;
 675         uint32_t seqno;
 676
 677         /* Only root volume needs formatting */
 678         assert(root_vol->vol_no == HAMMER_ROOT_VOLNO);
 679         ondisk = root_vol->ondisk;
 680
 681         /*
 682          * Size the undo buffer in multiples of HAMMER_BIGBLOCK_SIZE,
 683          * up to HAMMER_MAX_UNDO_BIGBLOCKS big-blocks.
 684          * Size to approximately 0.1% of the disk.
 685          *
 686          * The minimum UNDO fifo size is 512MB, or approximately 1% of
 687          * the recommended 50G disk.
 688          *
 689          * Changing this minimum is rather dangerous as complex filesystem
 690          * operations can cause the UNDO FIFO to fill up otherwise.
 691          */
 692         undo_limit = *undo_buffer_size;
 693         if (undo_limit == 0) {
 694                 undo_limit = HAMMER_VOL_BUF_SIZE(ondisk) / 1000;
 695                 if (undo_limit < HAMMER_BIGBLOCK_SIZE * HAMMER_MIN_UNDO_BIGBLOCKS)
 696                         undo_limit = HAMMER_BIGBLOCK_SIZE * HAMMER_MIN_UNDO_BIGBLOCKS;
 697         }
 698         undo_limit = HAMMER_BIGBLOCK_DOALIGN(undo_limit);
 699         if (undo_limit < HAMMER_BIGBLOCK_SIZE)
 700                 undo_limit = HAMMER_BIGBLOCK_SIZE;
 701         if (undo_limit > HAMMER_BIGBLOCK_SIZE * HAMMER_MAX_UNDO_BIGBLOCKS)
 702                 undo_limit = HAMMER_BIGBLOCK_SIZE * HAMMER_MAX_UNDO_BIGBLOCKS;
 703         *undo_buffer_size = undo_limit;
 704
 705         blockmap = &ondisk->vol0_blockmap[HAMMER_ZONE_UNDO_INDEX];
 706         bzero(blockmap, sizeof(*blockmap));
 707         blockmap->phys_offset = HAMMER_BLOCKMAP_UNAVAIL;
 708         blockmap->first_offset = HAMMER_ENCODE_UNDO(0);
 709         blockmap->next_offset = blockmap->first_offset;
 710         blockmap->alloc_offset = HAMMER_ENCODE_UNDO(undo_limit);
 711         hammer_crc_set_blockmap(blockmap);
 712
 713         limit_index = undo_limit / HAMMER_BIGBLOCK_SIZE;
 714         assert(limit_index <= HAMMER_MAX_UNDO_BIGBLOCKS);
 715
 716         for (n = 0; n < limit_index; ++n) {
 717                 ondisk->vol0_undo_array[n] = alloc_undo_bigblock(root_vol);
 718         }
 719         while (n < HAMMER_MAX_UNDO_BIGBLOCKS) {
 720                 ondisk->vol0_undo_array[n++] = HAMMER_BLOCKMAP_UNAVAIL;
 721         }
 722
 723         /*
 724          * Pre-initialize the UNDO blocks (HAMMER version 4+)
 725          */
 726         printf("initializing the undo map (%jd MB)\n",
 727                 (intmax_t)HAMMER_OFF_LONG_ENCODE(blockmap->alloc_offset) /
 728                 (1024 * 1024));
 729
 730         scan = blockmap->first_offset;
 731         seqno = 0;
 732
 733         while (scan < blockmap->alloc_offset) {
 734                 hammer_fifo_head_t head;
 735                 hammer_fifo_tail_t tail;
 736                 int isnew;
 737                 int bytes = HAMMER_UNDO_ALIGN;
 738
 739                 isnew = ((scan & HAMMER_BUFMASK64) == 0);
 740                 head = get_buffer_data(scan, &buffer, isnew);
 741                 buffer->cache.modified = 1;
 742                 tail = (void *)((char *)head + bytes - sizeof(*tail));
 743
 744                 bzero(head, bytes);
 745                 head->hdr_signature = HAMMER_HEAD_SIGNATURE;
 746                 head->hdr_type = HAMMER_HEAD_TYPE_DUMMY;
 747                 head->hdr_size = bytes;
 748                 head->hdr_seq = seqno++;
 749
 750                 tail->tail_signature = HAMMER_TAIL_SIGNATURE;
 751                 tail->tail_type = HAMMER_HEAD_TYPE_DUMMY;
 752                 tail->tail_size = bytes;
 753
 754                 hammer_crc_set_fifo_head(head, bytes);
 755
 756                 scan += bytes;
 757         }
 758         rel_buffer(buffer);
 759 }
 760
 761 const char *zone_labels[] = {
 762         "",             /* 0 */
 763         "raw_volume",   /* 1 */
 764         "raw_buffer",   /* 2 */
 765         "undo",         /* 3 */
 766         "freemap",      /* 4 */
 767         "",             /* 5 */
 768         "",             /* 6 */
 769         "",             /* 7 */
 770         "btree",        /* 8 */
 771         "meta",         /* 9 */
 772         "large_data",   /* 10 */
 773         "small_data",   /* 11 */
 774         "",             /* 12 */
 775         "",             /* 13 */
 776         "",             /* 14 */
 777         "unavail",      /* 15 */
 778 };
 779
 780 void
 781 print_blockmap(const struct volume_info *vol)
 782 {
 783         hammer_blockmap_t blockmap;
 784         hammer_volume_ondisk_t ondisk;
 785         int64_t size, used;
 786         int i;
 787 #define INDENT ""
 788
 789         ondisk = vol->ondisk;
 790         printf(INDENT"vol_label\t%s\n", ondisk->vol_label);
 791         printf(INDENT"vol_count\t%d\n", ondisk->vol_count);
 792         printf(INDENT"vol_bot_beg\t%s\n", sizetostr(ondisk->vol_bot_beg));
 793         printf(INDENT"vol_mem_beg\t%s\n", sizetostr(ondisk->vol_mem_beg));
 794         printf(INDENT"vol_buf_beg\t%s\n", sizetostr(ondisk->vol_buf_beg));
 795         printf(INDENT"vol_buf_end\t%s\n", sizetostr(ondisk->vol_buf_end));
 796         printf(INDENT"vol0_next_tid\t%016jx\n",
 797                (uintmax_t)ondisk->vol0_next_tid);
 798
 799         blockmap = &ondisk->vol0_blockmap[HAMMER_ZONE_UNDO_INDEX];
 800         size = HAMMER_OFF_LONG_ENCODE(blockmap->alloc_offset);
 801         if (blockmap->first_offset <= blockmap->next_offset)
 802                 used = blockmap->next_offset - blockmap->first_offset;
 803         else
 804                 used = blockmap->alloc_offset - blockmap->first_offset +
 805                         HAMMER_OFF_LONG_ENCODE(blockmap->next_offset);
 806         printf(INDENT"undo_size\t%s\n", sizetostr(size));
 807         printf(INDENT"undo_used\t%s\n", sizetostr(used));
 808
 809         printf(INDENT"zone #             "
 810                "phys             first            next             alloc\n");
 811         for (i = 0; i < HAMMER_MAX_ZONES; i++) {
 812                 blockmap = &ondisk->vol0_blockmap[i];
 813                 printf(INDENT"zone %-2d %-10s %016jx %016jx %016jx %016jx\n",
 814                         i, zone_labels[i],
 815                         (uintmax_t)blockmap->phys_offset,
 816                         (uintmax_t)blockmap->first_offset,
 817                         (uintmax_t)blockmap->next_offset,
 818                         (uintmax_t)blockmap->alloc_offset);
 819         }
 820 }
 821
 822 /*
 823  * Flush various tracking structures to disk
 824  */
 825 void
 826 flush_all_volumes(void)
 827 {
 828         struct volume_info *vol;
 829
 830         TAILQ_FOREACH(vol, &VolList, entry)
 831                 flush_volume(vol);
 832 }
 833
 834 void
 835 flush_volume(struct volume_info *volume)
 836 {
 837         struct buffer_info *buffer;
 838         int i;
 839
 840         for (i = 0; i < HAMMER_BUFLISTS; ++i) {
 841                 TAILQ_FOREACH(buffer, &volume->buffer_lists[i], entry)
 842                         flush_buffer(buffer);
 843         }
 844         if (writehammervol(volume) == -1)
 845                 err(1, "Write volume %d (%s)", volume->vol_no, volume->name);
 846 }
 847
 848 void
 849 flush_buffer(struct buffer_info *buffer)
 850 {
 851         struct volume_info *vol;
 852
 853         vol = buffer->volume;
 854         if (writehammerbuf(buffer) == -1)
 855                 err(1, "Write volume %d (%s)", vol->vol_no, vol->name);
 856         buffer->cache.modified = 0;
 857 }
 858
 859 /*
 860  * Core I/O operations
 861  */
 862 static int
 863 __read(struct volume_info *vol, void *data, int64_t offset, int size)
 864 {
 865         ssize_t n;
 866
 867         n = pread(vol->fd, data, size, offset);
 868         if (n != size)
 869                 return(-1);
 870         return(0);
 871 }
 872
 873 static __inline int
 874 readhammervol(struct volume_info *vol)
 875 {
 876         return(__read(vol, vol->ondisk, 0, HAMMER_BUFSIZE));
 877 }
 878
 879 static __inline int
 880 readhammerbuf(struct buffer_info *buf)
 881 {
 882         return(__read(buf->volume, buf->ondisk, buf->raw_offset, HAMMER_BUFSIZE));
 883 }
 884
 885 static int
 886 __write(struct volume_info *vol, const void *data, int64_t offset, int size)
 887 {
 888         ssize_t n;
 889
 890         if (vol->rdonly)
 891                 return(0);
 892
 893         n = pwrite(vol->fd, data, size, offset);
 894         if (n != size)
 895                 return(-1);
 896         return(0);
 897 }
 898
 899 static __inline int
 900 writehammervol(struct volume_info *vol)
 901 {
 902         return(__write(vol, vol->ondisk, 0, HAMMER_BUFSIZE));
 903 }
 904
 905 static __inline int
 906 writehammerbuf(struct buffer_info *buf)
 907 {
 908         return(__write(buf->volume, buf->ondisk, buf->raw_offset, HAMMER_BUFSIZE));
 909 }
 910
 911 int64_t init_boot_area_size(int64_t value, off_t avg_vol_size)
 912 {
 913         if (value == 0) {
 914                 value = HAMMER_BOOT_NOMBYTES;
 915                 while (value > avg_vol_size / HAMMER_MAX_VOLUMES)
 916                         value >>= 1;
 917         }
 918
 919         if (value < HAMMER_BOOT_MINBYTES) {
 920                 value = HAMMER_BOOT_MINBYTES;
 921         } else if (value > HAMMER_BOOT_MAXBYTES) {
 922                 value = HAMMER_BOOT_MAXBYTES;
 923         }
 924
 925         return(value);
 926 }
 927
 928 int64_t init_memory_log_size(int64_t value, off_t avg_vol_size)
 929 {
 930         if (value == 0) {
 931                 value = HAMMER_MEM_NOMBYTES;
 932                 while (value > avg_vol_size / HAMMER_MAX_VOLUMES)
 933                         value >>= 1;
 934         }
 935
 936         if (value < HAMMER_MEM_MINBYTES) {
 937                 value = HAMMER_MEM_MINBYTES;
 938         } else if (value > HAMMER_MEM_MAXBYTES) {
 939                 value = HAMMER_MEM_MAXBYTES;
 940         }
 941
 942         return(value);
 943 }