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(hammer_off_t buf_offset)
  69 {
  70         struct volume_info *volume;
  71         struct buffer_info *buf;
  72         int hi;
  73
  74         volume = get_volume(HAMMER_VOL_DECODE(buf_offset));
  75         assert(volume);
  76
  77         hi = buffer_hash(buf_offset);
  78         TAILQ_FOREACH(buf, &volume->buffer_lists[hi], entry)
  79                 if (buf->buf_offset == buf_offset)
  80                         return(buf);
  81         return(NULL);
  82 }
  83
  84 static
  85 struct volume_info *
  86 __alloc_volume(const char *volname, int oflags)
  87 {
  88         struct volume_info *vol;
  89         int i;
  90
  91         vol = malloc(sizeof(*vol));
  92         if (vol == NULL)
  93                 err(1, "alloc_volume");
  94         bzero(vol, sizeof(*vol));
  95
  96         vol->vol_no = -1;
  97         vol->rdonly = (oflags == O_RDONLY);
  98         vol->name = strdup(volname);
  99         vol->fd = open(vol->name, oflags);
 100         if (vol->fd < 0)
 101                 err(1, "alloc_volume: Failed to open %s", vol->name);
 102         check_volume(vol);
 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 static void
 144 __verify_volume(struct volume_info *vol)
 145 {
 146         hammer_volume_ondisk_t ondisk = vol->ondisk;
 147
 148         if (ondisk->vol_signature != HAMMER_FSBUF_VOLUME) {
 149                 errx(1, "verify_volume: Invalid volume signature %016jx",
 150                         ondisk->vol_signature);
 151         }
 152         if (ondisk->vol_rootvol != HAMMER_ROOT_VOLNO) {
 153                 errx(1, "verify_volume: Invalid root volume# %d",
 154                         ondisk->vol_rootvol);
 155         }
 156         if (bcmp(&Hammer_FSType, &ondisk->vol_fstype, sizeof(Hammer_FSType))) {
 157                 errx(1, "verify_volume: %s: Header does not indicate "
 158                         "that this is a HAMMER volume", vol->name);
 159         }
 160         if (bcmp(&Hammer_FSId, &ondisk->vol_fsid, sizeof(Hammer_FSId))) {
 161                 errx(1, "verify_volume: %s: FSId does not match other volumes!",
 162                         vol->name);
 163         }
 164 }
 165
 166 /*
 167  * Initialize a volume structure and ondisk vol_no field.
 168  */
 169 struct volume_info *
 170 init_volume(const char *filename, int oflags, int32_t vol_no)
 171 {
 172         struct volume_info *vol;
 173
 174         vol = __alloc_volume(filename, oflags);
 175         vol->vol_no = vol->ondisk->vol_no = vol_no;
 176
 177         __add_volume(vol);
 178
 179         return(vol);
 180 }
 181
 182 /*
 183  * Initialize a volume structure and read ondisk volume header.
 184  */
 185 struct volume_info*
 186 load_volume(const char *filename, int oflags, int verify)
 187 {
 188         struct volume_info *vol;
 189         int n;
 190
 191         vol = __alloc_volume(filename, oflags);
 192
 193         n = readhammervol(vol);
 194         if (n == -1) {
 195                 err(1, "load_volume: %s: Read failed at offset 0", vol->name);
 196         }
 197         vol->vol_no = vol->ondisk->vol_no;
 198
 199         if (valid_hammer_volumes++ == 0)
 200                 Hammer_FSId = vol->ondisk->vol_fsid;
 201         if (verify)
 202                 __verify_volume(vol);
 203
 204         __add_volume(vol);
 205
 206         return(vol);
 207 }
 208
 209 /*
 210  * Check basic volume characteristics.
 211  */
 212 static void
 213 check_volume(struct volume_info *vol)
 214 {
 215         struct partinfo pinfo;
 216         struct stat st;
 217
 218         /*
 219          * Get basic information about the volume
 220          */
 221         if (ioctl(vol->fd, DIOCGPART, &pinfo) < 0) {
 222                 /*
 223                  * Allow the formatting of regular files as HAMMER volumes
 224                  */
 225                 if (fstat(vol->fd, &st) < 0)
 226                         err(1, "Unable to stat %s", vol->name);
 227                 vol->size = st.st_size;
 228                 vol->type = "REGFILE";
 229         } else {
 230                 /*
 231                  * When formatting a block device as a HAMMER volume the
 232                  * sector size must be compatible.  HAMMER uses 16384 byte
 233                  * filesystem buffers.
 234                  */
 235                 if (pinfo.reserved_blocks) {
 236                         errx(1, "HAMMER cannot be placed in a partition "
 237                                 "which overlaps the disklabel or MBR");
 238                 }
 239                 if (pinfo.media_blksize > HAMMER_BUFSIZE ||
 240                     HAMMER_BUFSIZE % pinfo.media_blksize) {
 241                         errx(1, "A media sector size of %d is not supported",
 242                              pinfo.media_blksize);
 243                 }
 244
 245                 vol->size = pinfo.media_size;
 246                 vol->device_offset = pinfo.media_offset;
 247                 vol->type = "DEVICE";
 248         }
 249 }
 250
 251 void
 252 assert_volume_offset(struct volume_info *vol)
 253 {
 254         assert(hammer_is_zone_raw_buffer(vol->vol_free_off));
 255         assert(hammer_is_zone_raw_buffer(vol->vol_free_end));
 256         if (vol->vol_free_off >= vol->vol_free_end)
 257                 errx(1, "Ran out of room, filesystem too small");
 258 }
 259
 260 struct volume_info *
 261 get_volume(int32_t vol_no)
 262 {
 263         struct volume_info *vol;
 264
 265         TAILQ_FOREACH(vol, &VolList, entry) {
 266                 if (vol->vol_no == vol_no)
 267                         break;
 268         }
 269
 270         return(vol);
 271 }
 272
 273 struct volume_info *
 274 get_root_volume(void)
 275 {
 276         return(get_volume(HAMMER_ROOT_VOLNO));
 277 }
 278
 279 /*
 280  * Acquire the specified buffer.  isnew is -1 only when called
 281  * via get_buffer_readahead() to prevent another readahead.
 282  */
 283 static struct buffer_info *
 284 get_buffer(hammer_off_t buf_offset, int isnew)
 285 {
 286         struct buffer_info *buf;
 287         struct volume_info *volume;
 288         int vol_no;
 289         int zone;
 290         int hi;
 291         int dora = 0;
 292         int error = 0;
 293
 294         zone = HAMMER_ZONE_DECODE(buf_offset);
 295         if (zone > HAMMER_ZONE_RAW_BUFFER_INDEX)
 296                 buf_offset = blockmap_lookup(buf_offset, NULL, NULL, &error);
 297         if (error || buf_offset == HAMMER_OFF_BAD)
 298                 return(NULL);
 299         assert(hammer_is_zone_raw_buffer(buf_offset));
 300
 301         vol_no = HAMMER_VOL_DECODE(buf_offset);
 302         volume = get_volume(vol_no);
 303         assert(volume != NULL);
 304
 305         buf_offset &= ~HAMMER_BUFMASK64;
 306         buf = find_buffer(buf_offset);
 307
 308         if (buf == NULL) {
 309                 buf = malloc(sizeof(*buf));
 310                 bzero(buf, sizeof(*buf));
 311                 buf->buf_offset = buf_offset;
 312                 buf->raw_offset = hammer_xlate_to_phys(volume->ondisk,
 313                                                         buf_offset);
 314                 buf->volume = volume;
 315                 buf->ondisk = malloc(HAMMER_BUFSIZE);
 316                 if (isnew <= 0) {
 317                         if (readhammerbuf(buf) == -1) {
 318                                 err(1, "get_buffer: %s:%016jx "
 319                                     "Read failed at offset %016jx",
 320                                     volume->name,
 321                                     (intmax_t)buf->buf_offset,
 322                                     (intmax_t)buf->raw_offset);
 323                         }
 324                 }
 325
 326                 hi = buffer_hash(buf_offset);
 327                 TAILQ_INSERT_TAIL(&volume->buffer_lists[hi], buf, entry);
 328                 hammer_cache_add(&buf->cache);
 329                 dora = (isnew == 0);
 330         } else {
 331                 assert(buf->ondisk != NULL);
 332                 assert(isnew != -1);
 333                 hammer_cache_used(&buf->cache);
 334         }
 335
 336         ++buf->cache.refs;
 337         hammer_cache_flush();
 338
 339         if (isnew > 0) {
 340                 assert(buf->cache.modified == 0);
 341                 bzero(buf->ondisk, HAMMER_BUFSIZE);
 342                 buf->cache.modified = 1;
 343         }
 344         if (dora)
 345                 get_buffer_readahead(buf);
 346         return(buf);
 347 }
 348
 349 static void
 350 get_buffer_readahead(struct buffer_info *base)
 351 {
 352         struct buffer_info *buf;
 353         struct volume_info *vol;
 354         hammer_off_t buf_offset;
 355         int64_t raw_offset;
 356         int ri = UseReadBehind;
 357         int re = UseReadAhead;
 358
 359         raw_offset = base->raw_offset + ri * HAMMER_BUFSIZE;
 360         vol = base->volume;
 361
 362         while (ri < re) {
 363                 if (raw_offset >= vol->ondisk->vol_buf_end)
 364                         break;
 365                 if (raw_offset < vol->ondisk->vol_buf_beg || ri == 0) {
 366                         ++ri;
 367                         raw_offset += HAMMER_BUFSIZE;
 368                         continue;
 369                 }
 370                 buf_offset = HAMMER_ENCODE_RAW_BUFFER(vol->vol_no,
 371                         raw_offset - vol->ondisk->vol_buf_beg);
 372                 buf = find_buffer(buf_offset);
 373                 if (buf == NULL) {
 374                         buf = get_buffer(buf_offset, -1);
 375                         rel_buffer(buf);
 376                 }
 377                 ++ri;
 378                 raw_offset += HAMMER_BUFSIZE;
 379         }
 380 }
 381
 382 void
 383 rel_buffer(struct buffer_info *buffer)
 384 {
 385         struct volume_info *volume;
 386         int hi;
 387
 388         if (buffer == NULL)
 389                 return;
 390         assert(buffer->cache.refs > 0);
 391         if (--buffer->cache.refs == 0) {
 392                 if (buffer->cache.delete) {
 393                         hi = buffer_hash(buffer->buf_offset);
 394                         volume = buffer->volume;
 395                         if (buffer->cache.modified)
 396                                 flush_buffer(buffer);
 397                         TAILQ_REMOVE(&volume->buffer_lists[hi], buffer, entry);
 398                         hammer_cache_del(&buffer->cache);
 399                         free(buffer->ondisk);
 400                         free(buffer);
 401                 }
 402         }
 403 }
 404
 405 /*
 406  * Retrieve a pointer to a buffer data given a buffer offset.  The underlying
 407  * bufferp is freed if isnew or the offset is out of range of the cached data.
 408  * If bufferp is freed a referenced buffer is loaded into it.
 409  */
 410 void *
 411 get_buffer_data(hammer_off_t buf_offset, struct buffer_info **bufferp,
 412                 int isnew)
 413 {
 414         if (*bufferp != NULL) {
 415                 if (isnew > 0 ||
 416                     (((*bufferp)->buf_offset ^ buf_offset) & ~HAMMER_BUFMASK64)) {
 417                         rel_buffer(*bufferp);
 418                         *bufferp = NULL;
 419                 }
 420         }
 421
 422         if (*bufferp == NULL) {
 423                 *bufferp = get_buffer(buf_offset, isnew);
 424                 if (*bufferp == NULL)
 425                         return(NULL);
 426         }
 427
 428         return(((char *)(*bufferp)->ondisk) +
 429                 ((int32_t)buf_offset & HAMMER_BUFMASK));
 430 }
 431
 432 /*
 433  * Allocate HAMMER elements - B-Tree nodes
 434  */
 435 hammer_node_ondisk_t
 436 alloc_btree_node(hammer_off_t *offp, struct buffer_info **data_bufferp)
 437 {
 438         hammer_node_ondisk_t node;
 439
 440         node = alloc_blockmap(HAMMER_ZONE_BTREE_INDEX, sizeof(*node),
 441                               offp, data_bufferp);
 442         bzero(node, sizeof(*node));
 443         return(node);
 444 }
 445
 446 /*
 447  * Allocate HAMMER elements - meta data (inode, direntry, PFS, etc)
 448  */
 449 void *
 450 alloc_meta_element(hammer_off_t *offp, int32_t data_len,
 451                    struct buffer_info **data_bufferp)
 452 {
 453         void *data;
 454
 455         data = alloc_blockmap(HAMMER_ZONE_META_INDEX, data_len,
 456                               offp, data_bufferp);
 457         bzero(data, data_len);
 458         return(data);
 459 }
 460
 461 /*
 462  * Format a new blockmap.  This is mostly a degenerate case because
 463  * all allocations are now actually done from the freemap.
 464  */
 465 void
 466 format_blockmap(struct volume_info *root_vol, int zone, hammer_off_t offset)
 467 {
 468         hammer_blockmap_t blockmap;
 469         hammer_off_t zone_base;
 470
 471         /* Only root volume needs formatting */
 472         assert(root_vol->vol_no == HAMMER_ROOT_VOLNO);
 473
 474         assert(hammer_is_zone2_mapped_index(zone));
 475
 476         blockmap = &root_vol->ondisk->vol0_blockmap[zone];
 477         zone_base = HAMMER_ZONE_ENCODE(zone, offset);
 478
 479         bzero(blockmap, sizeof(*blockmap));
 480         blockmap->phys_offset = 0;
 481         blockmap->first_offset = zone_base;
 482         blockmap->next_offset = zone_base;
 483         blockmap->alloc_offset = HAMMER_ENCODE(zone, 255, -1);
 484         hammer_crc_set_blockmap(blockmap);
 485 }
 486
 487 /*
 488  * Format a new freemap.  Set all layer1 entries to UNAVAIL.  The initialize
 489  * code will load each volume's freemap.
 490  */
 491 void
 492 format_freemap(struct volume_info *root_vol)
 493 {
 494         struct buffer_info *buffer = NULL;
 495         hammer_off_t layer1_offset;
 496         hammer_blockmap_t blockmap;
 497         hammer_blockmap_layer1_t layer1;
 498         int i, isnew;
 499
 500         /* Only root volume needs formatting */
 501         assert(root_vol->vol_no == HAMMER_ROOT_VOLNO);
 502
 503         layer1_offset = bootstrap_bigblock(root_vol);
 504         for (i = 0; i < HAMMER_BIGBLOCK_SIZE; i += sizeof(*layer1)) {
 505                 isnew = ((i % HAMMER_BUFSIZE) == 0);
 506                 layer1 = get_buffer_data(layer1_offset + i, &buffer, isnew);
 507                 bzero(layer1, sizeof(*layer1));
 508                 layer1->phys_offset = HAMMER_BLOCKMAP_UNAVAIL;
 509                 layer1->blocks_free = 0;
 510                 hammer_crc_set_layer1(layer1);
 511         }
 512         assert(i == HAMMER_BIGBLOCK_SIZE);
 513         rel_buffer(buffer);
 514
 515         blockmap = &root_vol->ondisk->vol0_blockmap[HAMMER_ZONE_FREEMAP_INDEX];
 516         bzero(blockmap, sizeof(*blockmap));
 517         blockmap->phys_offset = layer1_offset;
 518         blockmap->first_offset = 0;
 519         blockmap->next_offset = HAMMER_ENCODE_RAW_BUFFER(0, 0);
 520         blockmap->alloc_offset = HAMMER_ENCODE_RAW_BUFFER(255, -1);
 521         hammer_crc_set_blockmap(blockmap);
 522 }
 523
 524 /*
 525  * Load the volume's remaining free space into the freemap.
 526  *
 527  * Returns the number of big-blocks available.
 528  */
 529 int64_t
 530 initialize_freemap(struct volume_info *vol)
 531 {
 532         struct volume_info *root_vol;
 533         struct buffer_info *buffer1 = NULL;
 534         struct buffer_info *buffer2 = NULL;
 535         hammer_blockmap_layer1_t layer1;
 536         hammer_blockmap_layer2_t layer2;
 537         hammer_off_t layer1_offset;
 538         hammer_off_t layer2_offset;
 539         hammer_off_t phys_offset;
 540         hammer_off_t block_offset;
 541         hammer_off_t aligned_vol_free_end;
 542         hammer_blockmap_t freemap;
 543         int64_t count = 0;
 544         int64_t layer1_count = 0;
 545
 546         root_vol = get_root_volume();
 547
 548         assert_volume_offset(vol);
 549         aligned_vol_free_end = HAMMER_BLOCKMAP_LAYER2_DOALIGN(vol->vol_free_end);
 550
 551         printf("initialize freemap volume %d\n", vol->vol_no);
 552
 553         /*
 554          * Initialize the freemap.  First preallocate the big-blocks required
 555          * to implement layer2.   This preallocation is a bootstrap allocation
 556          * using blocks from the target volume.
 557          */
 558         freemap = &root_vol->ondisk->vol0_blockmap[HAMMER_ZONE_FREEMAP_INDEX];
 559
 560         for (phys_offset = HAMMER_ENCODE_RAW_BUFFER(vol->vol_no, 0);
 561              phys_offset < aligned_vol_free_end;
 562              phys_offset += HAMMER_BLOCKMAP_LAYER2) {
 563                 layer1_offset = freemap->phys_offset +
 564                                 HAMMER_BLOCKMAP_LAYER1_OFFSET(phys_offset);
 565                 layer1 = get_buffer_data(layer1_offset, &buffer1, 0);
 566                 if (layer1->phys_offset == HAMMER_BLOCKMAP_UNAVAIL) {
 567                         layer1->phys_offset = bootstrap_bigblock(vol);
 568                         layer1->blocks_free = 0;
 569                         buffer1->cache.modified = 1;
 570                         hammer_crc_set_layer1(layer1);
 571                 }
 572         }
 573
 574         /*
 575          * Now fill everything in.
 576          */
 577         for (phys_offset = HAMMER_ENCODE_RAW_BUFFER(vol->vol_no, 0);
 578              phys_offset < aligned_vol_free_end;
 579              phys_offset += HAMMER_BLOCKMAP_LAYER2) {
 580                 layer1_count = 0;
 581                 layer1_offset = freemap->phys_offset +
 582                                 HAMMER_BLOCKMAP_LAYER1_OFFSET(phys_offset);
 583                 layer1 = get_buffer_data(layer1_offset, &buffer1, 0);
 584                 assert(layer1->phys_offset != HAMMER_BLOCKMAP_UNAVAIL);
 585
 586                 for (block_offset = 0;
 587                      block_offset < HAMMER_BLOCKMAP_LAYER2;
 588                      block_offset += HAMMER_BIGBLOCK_SIZE) {
 589                         layer2_offset = layer1->phys_offset +
 590                                         HAMMER_BLOCKMAP_LAYER2_OFFSET(block_offset);
 591                         layer2 = get_buffer_data(layer2_offset, &buffer2, 0);
 592                         bzero(layer2, sizeof(*layer2));
 593
 594                         if (phys_offset + block_offset < vol->vol_free_off) {
 595                                 /*
 596                                  * Big-blocks already allocated as part
 597                                  * of the freemap bootstrap.
 598                                  */
 599                                 layer2->zone = HAMMER_ZONE_FREEMAP_INDEX;
 600                                 layer2->append_off = HAMMER_BIGBLOCK_SIZE;
 601                                 layer2->bytes_free = 0;
 602                         } else if (phys_offset + block_offset < vol->vol_free_end) {
 603                                 layer2->zone = 0;
 604                                 layer2->append_off = 0;
 605                                 layer2->bytes_free = HAMMER_BIGBLOCK_SIZE;
 606                                 ++count;
 607                                 ++layer1_count;
 608                         } else {
 609                                 layer2->zone = HAMMER_ZONE_UNAVAIL_INDEX;
 610                                 layer2->append_off = HAMMER_BIGBLOCK_SIZE;
 611                                 layer2->bytes_free = 0;
 612                         }
 613                         hammer_crc_set_layer2(layer2);
 614                         buffer2->cache.modified = 1;
 615                 }
 616
 617                 layer1->blocks_free += layer1_count;
 618                 hammer_crc_set_layer1(layer1);
 619                 buffer1->cache.modified = 1;
 620         }
 621
 622         rel_buffer(buffer1);
 623         rel_buffer(buffer2);
 624         return(count);
 625 }
 626
 627 /*
 628  * Returns the number of big-blocks available for filesystem data and undos
 629  * without formatting.
 630  */
 631 int64_t
 632 count_freemap(struct volume_info *vol)
 633 {
 634         hammer_off_t phys_offset;
 635         hammer_off_t vol_free_off;
 636         hammer_off_t aligned_vol_free_end;
 637         int64_t count = 0;
 638
 639         vol_free_off = HAMMER_ENCODE_RAW_BUFFER(vol->vol_no, 0);
 640
 641         assert_volume_offset(vol);
 642         aligned_vol_free_end = HAMMER_BLOCKMAP_LAYER2_DOALIGN(vol->vol_free_end);
 643
 644         if (vol->vol_no == HAMMER_ROOT_VOLNO)
 645                 vol_free_off += HAMMER_BIGBLOCK_SIZE;
 646
 647         for (phys_offset = HAMMER_ENCODE_RAW_BUFFER(vol->vol_no, 0);
 648              phys_offset < aligned_vol_free_end;
 649              phys_offset += HAMMER_BLOCKMAP_LAYER2) {
 650                 vol_free_off += HAMMER_BIGBLOCK_SIZE;
 651         }
 652
 653         for (phys_offset = HAMMER_ENCODE_RAW_BUFFER(vol->vol_no, 0);
 654              phys_offset < aligned_vol_free_end;
 655              phys_offset += HAMMER_BIGBLOCK_SIZE) {
 656                 if (phys_offset < vol_free_off) {
 657                         ;
 658                 } else if (phys_offset < vol->vol_free_end) {
 659                         ++count;
 660                 }
 661         }
 662
 663         return(count);
 664 }
 665
 666 /*
 667  * Format the undomap for the root volume.
 668  */
 669 void
 670 format_undomap(struct volume_info *root_vol, int64_t *undo_buffer_size)
 671 {
 672         hammer_off_t undo_limit;
 673         hammer_blockmap_t blockmap;
 674         hammer_volume_ondisk_t ondisk;
 675         struct buffer_info *buffer = NULL;
 676         hammer_off_t scan;
 677         int n;
 678         int limit_index;
 679         uint32_t seqno;
 680
 681         /* Only root volume needs formatting */
 682         assert(root_vol->vol_no == HAMMER_ROOT_VOLNO);
 683         ondisk = root_vol->ondisk;
 684
 685         /*
 686          * Size the undo buffer in multiples of HAMMER_BIGBLOCK_SIZE,
 687          * up to HAMMER_MAX_UNDO_BIGBLOCKS big-blocks.
 688          * Size to approximately 0.1% of the disk.
 689          *
 690          * The minimum UNDO fifo size is 512MB, or approximately 1% of
 691          * the recommended 50G disk.
 692          *
 693          * Changing this minimum is rather dangerous as complex filesystem
 694          * operations can cause the UNDO FIFO to fill up otherwise.
 695          */
 696         undo_limit = *undo_buffer_size;
 697         if (undo_limit == 0) {
 698                 undo_limit = HAMMER_VOL_BUF_SIZE(ondisk) / 1000;
 699                 if (undo_limit < HAMMER_BIGBLOCK_SIZE * HAMMER_MIN_UNDO_BIGBLOCKS)
 700                         undo_limit = HAMMER_BIGBLOCK_SIZE * HAMMER_MIN_UNDO_BIGBLOCKS;
 701         }
 702         undo_limit = HAMMER_BIGBLOCK_DOALIGN(undo_limit);
 703         if (undo_limit < HAMMER_BIGBLOCK_SIZE)
 704                 undo_limit = HAMMER_BIGBLOCK_SIZE;
 705         if (undo_limit > HAMMER_BIGBLOCK_SIZE * HAMMER_MAX_UNDO_BIGBLOCKS)
 706                 undo_limit = HAMMER_BIGBLOCK_SIZE * HAMMER_MAX_UNDO_BIGBLOCKS;
 707         *undo_buffer_size = undo_limit;
 708
 709         blockmap = &ondisk->vol0_blockmap[HAMMER_ZONE_UNDO_INDEX];
 710         bzero(blockmap, sizeof(*blockmap));
 711         blockmap->phys_offset = HAMMER_BLOCKMAP_UNAVAIL;
 712         blockmap->first_offset = HAMMER_ENCODE_UNDO(0);
 713         blockmap->next_offset = blockmap->first_offset;
 714         blockmap->alloc_offset = HAMMER_ENCODE_UNDO(undo_limit);
 715         hammer_crc_set_blockmap(blockmap);
 716
 717         limit_index = undo_limit / HAMMER_BIGBLOCK_SIZE;
 718         assert(limit_index <= HAMMER_MAX_UNDO_BIGBLOCKS);
 719
 720         for (n = 0; n < limit_index; ++n) {
 721                 ondisk->vol0_undo_array[n] = alloc_undo_bigblock(root_vol);
 722         }
 723         while (n < HAMMER_MAX_UNDO_BIGBLOCKS) {
 724                 ondisk->vol0_undo_array[n++] = HAMMER_BLOCKMAP_UNAVAIL;
 725         }
 726
 727         /*
 728          * Pre-initialize the UNDO blocks (HAMMER version 4+)
 729          */
 730         printf("initializing the undo map (%jd MB)\n",
 731                 (intmax_t)HAMMER_OFF_LONG_ENCODE(blockmap->alloc_offset) /
 732                 (1024 * 1024));
 733
 734         scan = blockmap->first_offset;
 735         seqno = 0;
 736
 737         while (scan < blockmap->alloc_offset) {
 738                 hammer_fifo_head_t head;
 739                 hammer_fifo_tail_t tail;
 740                 int isnew;
 741                 int bytes = HAMMER_UNDO_ALIGN;
 742
 743                 isnew = ((scan & HAMMER_BUFMASK64) == 0);
 744                 head = get_buffer_data(scan, &buffer, isnew);
 745                 buffer->cache.modified = 1;
 746                 tail = (void *)((char *)head + bytes - sizeof(*tail));
 747
 748                 bzero(head, bytes);
 749                 head->hdr_signature = HAMMER_HEAD_SIGNATURE;
 750                 head->hdr_type = HAMMER_HEAD_TYPE_DUMMY;
 751                 head->hdr_size = bytes;
 752                 head->hdr_seq = seqno++;
 753
 754                 tail->tail_signature = HAMMER_TAIL_SIGNATURE;
 755                 tail->tail_type = HAMMER_HEAD_TYPE_DUMMY;
 756                 tail->tail_size = bytes;
 757
 758                 hammer_crc_set_fifo_head(head, bytes);
 759
 760                 scan += bytes;
 761         }
 762         rel_buffer(buffer);
 763 }
 764
 765 const char *zone_labels[] = {
 766         "",             /* 0 */
 767         "raw_volume",   /* 1 */
 768         "raw_buffer",   /* 2 */
 769         "undo",         /* 3 */
 770         "freemap",      /* 4 */
 771         "",             /* 5 */
 772         "",             /* 6 */
 773         "",             /* 7 */
 774         "btree",        /* 8 */
 775         "meta",         /* 9 */
 776         "large_data",   /* 10 */
 777         "small_data",   /* 11 */
 778         "",             /* 12 */
 779         "",             /* 13 */
 780         "",             /* 14 */
 781         "unavail",      /* 15 */
 782 };
 783
 784 void
 785 print_blockmap(const struct volume_info *vol)
 786 {
 787         hammer_blockmap_t blockmap;
 788         hammer_volume_ondisk_t ondisk;
 789         int64_t size, used;
 790         int i;
 791 #define INDENT ""
 792
 793         ondisk = vol->ondisk;
 794         printf(INDENT"vol_label\t%s\n", ondisk->vol_label);
 795         printf(INDENT"vol_count\t%d\n", ondisk->vol_count);
 796         printf(INDENT"vol_bot_beg\t%s\n", sizetostr(ondisk->vol_bot_beg));
 797         printf(INDENT"vol_mem_beg\t%s\n", sizetostr(ondisk->vol_mem_beg));
 798         printf(INDENT"vol_buf_beg\t%s\n", sizetostr(ondisk->vol_buf_beg));
 799         printf(INDENT"vol_buf_end\t%s\n", sizetostr(ondisk->vol_buf_end));
 800         printf(INDENT"vol0_next_tid\t%016jx\n",
 801                (uintmax_t)ondisk->vol0_next_tid);
 802
 803         blockmap = &ondisk->vol0_blockmap[HAMMER_ZONE_UNDO_INDEX];
 804         size = HAMMER_OFF_LONG_ENCODE(blockmap->alloc_offset);
 805         if (blockmap->first_offset <= blockmap->next_offset)
 806                 used = blockmap->next_offset - blockmap->first_offset;
 807         else
 808                 used = blockmap->alloc_offset - blockmap->first_offset +
 809                         HAMMER_OFF_LONG_ENCODE(blockmap->next_offset);
 810         printf(INDENT"undo_size\t%s\n", sizetostr(size));
 811         printf(INDENT"undo_used\t%s\n", sizetostr(used));
 812
 813         printf(INDENT"zone #             "
 814                "phys             first            next             alloc\n");
 815         for (i = 0; i < HAMMER_MAX_ZONES; i++) {
 816                 blockmap = &ondisk->vol0_blockmap[i];
 817                 printf(INDENT"zone %-2d %-10s %016jx %016jx %016jx %016jx\n",
 818                         i, zone_labels[i],
 819                         (uintmax_t)blockmap->phys_offset,
 820                         (uintmax_t)blockmap->first_offset,
 821                         (uintmax_t)blockmap->next_offset,
 822                         (uintmax_t)blockmap->alloc_offset);
 823         }
 824 }
 825
 826 /*
 827  * Flush various tracking structures to disk
 828  */
 829 void
 830 flush_all_volumes(void)
 831 {
 832         struct volume_info *vol;
 833
 834         TAILQ_FOREACH(vol, &VolList, entry)
 835                 flush_volume(vol);
 836 }
 837
 838 void
 839 flush_volume(struct volume_info *volume)
 840 {
 841         struct buffer_info *buffer;
 842         int i;
 843
 844         for (i = 0; i < HAMMER_BUFLISTS; ++i) {
 845                 TAILQ_FOREACH(buffer, &volume->buffer_lists[i], entry)
 846                         flush_buffer(buffer);
 847         }
 848         if (writehammervol(volume) == -1)
 849                 err(1, "Write volume %d (%s)", volume->vol_no, volume->name);
 850 }
 851
 852 void
 853 flush_buffer(struct buffer_info *buffer)
 854 {
 855         struct volume_info *vol;
 856
 857         vol = buffer->volume;
 858         if (writehammerbuf(buffer) == -1)
 859                 err(1, "Write volume %d (%s)", vol->vol_no, vol->name);
 860         buffer->cache.modified = 0;
 861 }
 862
 863 /*
 864  * Core I/O operations
 865  */
 866 static int
 867 __read(struct volume_info *vol, void *data, int64_t offset, int size)
 868 {
 869         ssize_t n;
 870
 871         n = pread(vol->fd, data, size, offset);
 872         if (n != size)
 873                 return(-1);
 874         return(0);
 875 }
 876
 877 static __inline int
 878 readhammervol(struct volume_info *vol)
 879 {
 880         return(__read(vol, vol->ondisk, 0, HAMMER_BUFSIZE));
 881 }
 882
 883 static __inline int
 884 readhammerbuf(struct buffer_info *buf)
 885 {
 886         return(__read(buf->volume, buf->ondisk, buf->raw_offset, HAMMER_BUFSIZE));
 887 }
 888
 889 static int
 890 __write(struct volume_info *vol, const void *data, int64_t offset, int size)
 891 {
 892         ssize_t n;
 893
 894         if (vol->rdonly)
 895                 return(0);
 896
 897         n = pwrite(vol->fd, data, size, offset);
 898         if (n != size)
 899                 return(-1);
 900         return(0);
 901 }
 902
 903 static __inline int
 904 writehammervol(struct volume_info *vol)
 905 {
 906         return(__write(vol, vol->ondisk, 0, HAMMER_BUFSIZE));
 907 }
 908
 909 static __inline int
 910 writehammerbuf(struct buffer_info *buf)
 911 {
 912         return(__write(buf->volume, buf->ondisk, buf->raw_offset, HAMMER_BUFSIZE));
 913 }
 914
 915 int64_t init_boot_area_size(int64_t value, off_t avg_vol_size)
 916 {
 917         if (value == 0) {
 918                 value = HAMMER_BOOT_NOMBYTES;
 919                 while (value > avg_vol_size / HAMMER_MAX_VOLUMES)
 920                         value >>= 1;
 921         }
 922
 923         if (value < HAMMER_BOOT_MINBYTES) {
 924                 value = HAMMER_BOOT_MINBYTES;
 925         } else if (value > HAMMER_BOOT_MAXBYTES) {
 926                 value = HAMMER_BOOT_MAXBYTES;
 927         }
 928
 929         return(value);
 930 }
 931
 932 int64_t init_memory_log_size(int64_t value, off_t avg_vol_size)
 933 {
 934         if (value == 0) {
 935                 value = HAMMER_MEM_NOMBYTES;
 936                 while (value > avg_vol_size / HAMMER_MAX_VOLUMES)
 937                         value >>= 1;
 938         }
 939
 940         if (value < HAMMER_MEM_MINBYTES) {
 941                 value = HAMMER_MEM_MINBYTES;
 942         } else if (value > HAMMER_MEM_MAXBYTES) {
 943                 value = HAMMER_MEM_MAXBYTES;
 944         }
 945
 946         return(value);
 947 }