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/types.h>
  36 #include <sys/stat.h>
  37 #include <assert.h>
  38 #include <stdio.h>
  39 #include <stdlib.h>
  40 #include <stdarg.h>
  41 #include <string.h>
  42 #include <unistd.h>
  43 #include <stddef.h>
  44 #include <err.h>
  45 #include <fcntl.h>
  46 #include "hammer_util.h"
  47
  48 static void *alloc_blockmap(int zone, int bytes, hammer_off_t *result_offp,
  49                         struct buffer_info **bufferp);
  50 static hammer_off_t alloc_bigblock(struct volume_info *volume, int zone);
  51 static void get_buffer_readahead(struct buffer_info *base);
  52 static __inline void *get_ondisk(hammer_off_t buf_offset,
  53                         struct buffer_info **bufferp, int isnew);
  54 static int readhammerbuf(struct volume_info *vol, void *data,
  55                         int64_t offset);
  56 static void writehammerbuf(struct volume_info *vol, const void *data,
  57                         int64_t offset);
  58
  59 int DebugOpt;
  60
  61 uuid_t Hammer_FSType;
  62 uuid_t Hammer_FSId;
  63 int64_t BootAreaSize;
  64 int64_t MemAreaSize;
  65 int64_t UndoBufferSize;
  66 int     UsingSuperClusters;
  67 int     NumVolumes;
  68 int     RootVolNo = -1;
  69 int     UseReadBehind = -4;
  70 int     UseReadAhead = 4;
  71 int     AssertOnFailure = 1;
  72 struct volume_list VolList = TAILQ_HEAD_INITIALIZER(VolList);
  73
  74 static __inline
  75 int
  76 buffer_hash(hammer_off_t buf_offset)
  77 {
  78         int hi;
  79
  80         hi = (int)(buf_offset / HAMMER_BUFSIZE) & HAMMER_BUFLISTMASK;
  81         return(hi);
  82 }
  83
  84 static struct buffer_info*
  85 find_buffer(struct volume_info *volume, hammer_off_t buf_offset)
  86 {
  87         int hi;
  88         struct buffer_info *buf;
  89
  90         hi = buffer_hash(buf_offset);
  91         TAILQ_FOREACH(buf, &volume->buffer_lists[hi], entry)
  92                 if (buf->buf_offset == buf_offset)
  93                         return(buf);
  94         return(NULL);
  95 }
  96
  97 /*
  98  * Lookup the requested information structure and related on-disk buffer.
  99  * Missing structures are created.
 100  */
 101 struct volume_info *
 102 setup_volume(int32_t vol_no, const char *filename, int isnew, int oflags)
 103 {
 104         struct volume_info *vol;
 105         struct volume_info *scan;
 106         struct hammer_volume_ondisk *ondisk;
 107         int i, n;
 108         struct stat st1, st2;
 109
 110         /*
 111          * Allocate the volume structure
 112          */
 113         vol = malloc(sizeof(*vol));
 114         bzero(vol, sizeof(*vol));
 115         for (i = 0; i < HAMMER_BUFLISTS; ++i)
 116                 TAILQ_INIT(&vol->buffer_lists[i]);
 117         vol->name = strdup(filename);
 118         vol->fd = open(vol->name, oflags);
 119         if (vol->fd < 0) {
 120                 err(1, "setup_volume: %s: Open failed", vol->name);
 121         }
 122
 123         /*
 124          * Read or initialize the volume header
 125          */
 126         vol->ondisk = ondisk = malloc(HAMMER_BUFSIZE);
 127         if (isnew > 0) {
 128                 bzero(ondisk, HAMMER_BUFSIZE);
 129         } else {
 130                 n = readhammerbuf(vol, ondisk, 0);
 131                 if (n == -1) {
 132                         err(1, "setup_volume: %s: Read failed at offset 0",
 133                             vol->name);
 134                 }
 135                 vol_no = ondisk->vol_no;
 136                 if (RootVolNo < 0) {
 137                         RootVolNo = ondisk->vol_rootvol;
 138                 } else if (RootVolNo != (int)ondisk->vol_rootvol) {
 139                         errx(1, "setup_volume: %s: root volume disagreement: "
 140                                 "%d vs %d",
 141                                 vol->name, RootVolNo, ondisk->vol_rootvol);
 142                 }
 143
 144                 if (bcmp(&Hammer_FSType, &ondisk->vol_fstype, sizeof(Hammer_FSType)) != 0) {
 145                         errx(1, "setup_volume: %s: Header does not indicate "
 146                                 "that this is a hammer volume", vol->name);
 147                 }
 148                 if (TAILQ_EMPTY(&VolList)) {
 149                         Hammer_FSId = vol->ondisk->vol_fsid;
 150                 } else if (bcmp(&Hammer_FSId, &ondisk->vol_fsid, sizeof(Hammer_FSId)) != 0) {
 151                         errx(1, "setup_volume: %s: FSId does match other "
 152                                 "volumes!", vol->name);
 153                 }
 154         }
 155         vol->vol_no = vol_no;
 156
 157         if (isnew > 0) {
 158                 vol->cache.modified = 1;
 159         }
 160
 161         if (fstat(vol->fd, &st1) != 0){
 162                 errx(1, "setup_volume: %s: Failed to stat", vol->name);
 163         }
 164
 165         /*
 166          * Link the volume structure in
 167          */
 168         TAILQ_FOREACH(scan, &VolList, entry) {
 169                 if (scan->vol_no == vol_no) {
 170                         errx(1, "setup_volume: %s: Duplicate volume number %d "
 171                                 "against %s", vol->name, vol_no, scan->name);
 172                 }
 173                 if (fstat(scan->fd, &st2) != 0){
 174                         errx(1, "setup_volume: %s: Failed to stat %s",
 175                                 vol->name, scan->name);
 176                 }
 177                 if ((st1.st_ino == st2.st_ino) && (st1.st_dev == st2.st_dev)) {
 178                         errx(1, "setup_volume: %s: Specified more than once",
 179                                 vol->name);
 180                 }
 181         }
 182         TAILQ_INSERT_TAIL(&VolList, vol, entry);
 183         return(vol);
 184 }
 185
 186 struct volume_info *
 187 get_volume(int32_t vol_no)
 188 {
 189         struct volume_info *vol;
 190
 191         TAILQ_FOREACH(vol, &VolList, entry) {
 192                 if (vol->vol_no == vol_no)
 193                         break;
 194         }
 195         if (vol == NULL) {
 196                 if (AssertOnFailure)
 197                         errx(1, "get_volume: Volume %d does not exist!",
 198                                 vol_no);
 199                 return(NULL);
 200         }
 201         ++vol->cache.refs;
 202         /* not added to or removed from hammer cache */
 203         return(vol);
 204 }
 205
 206 void
 207 rel_volume(struct volume_info *volume)
 208 {
 209         if (volume == NULL)
 210                 return;
 211         /* not added to or removed from hammer cache */
 212         --volume->cache.refs;
 213 }
 214
 215 /*
 216  * Acquire the specified buffer.  isnew is -1 only when called
 217  * via get_buffer_readahead() to prevent another readahead.
 218  */
 219 struct buffer_info *
 220 get_buffer(hammer_off_t buf_offset, int isnew)
 221 {
 222         void *ondisk;
 223         struct buffer_info *buf;
 224         struct volume_info *volume;
 225         hammer_off_t orig_offset = buf_offset;
 226         int vol_no;
 227         int zone;
 228         int hi, n;
 229         int dora = 0;
 230
 231         zone = HAMMER_ZONE_DECODE(buf_offset);
 232         if (zone > HAMMER_ZONE_RAW_BUFFER_INDEX) {
 233                 buf_offset = blockmap_lookup(buf_offset, NULL, NULL, NULL);
 234         }
 235         if (buf_offset == HAMMER_OFF_BAD)
 236                 return(NULL);
 237
 238         if (AssertOnFailure) {
 239                 assert((buf_offset & HAMMER_OFF_ZONE_MASK) ==
 240                        HAMMER_ZONE_RAW_BUFFER);
 241         }
 242         vol_no = HAMMER_VOL_DECODE(buf_offset);
 243         volume = get_volume(vol_no);
 244         if (volume == NULL)
 245                 return(NULL);
 246
 247         buf_offset &= ~HAMMER_BUFMASK64;
 248         buf = find_buffer(volume, buf_offset);
 249
 250         if (buf == NULL) {
 251                 buf = malloc(sizeof(*buf));
 252                 bzero(buf, sizeof(*buf));
 253                 if (DebugOpt > 1) {
 254                         fprintf(stderr, "get_buffer: %016llx %016llx at %p\n",
 255                                 (long long)orig_offset, (long long)buf_offset,
 256                                 buf);
 257                 }
 258                 buf->buf_offset = buf_offset;
 259                 buf->raw_offset = volume->ondisk->vol_buf_beg +
 260                                   (buf_offset & HAMMER_OFF_SHORT_MASK);
 261                 buf->volume = volume;
 262                 hi = buffer_hash(buf_offset);
 263                 TAILQ_INSERT_TAIL(&volume->buffer_lists[hi], buf, entry);
 264                 ++volume->cache.refs;
 265                 buf->cache.u.buffer = buf;
 266                 hammer_cache_add(&buf->cache, ISBUFFER);
 267                 dora = (isnew == 0);
 268         } else {
 269                 if (DebugOpt > 1) {
 270                         fprintf(stderr, "get_buffer: %016llx %016llx at %p *\n",
 271                                 (long long)orig_offset, (long long)buf_offset,
 272                                 buf);
 273                 }
 274                 hammer_cache_used(&buf->cache);
 275                 ++buf->use_count;
 276         }
 277         ++buf->cache.refs;
 278         hammer_cache_flush();
 279         if ((ondisk = buf->ondisk) == NULL) {
 280                 buf->ondisk = ondisk = malloc(HAMMER_BUFSIZE);
 281                 if (isnew <= 0) {
 282                         n = readhammerbuf(volume, ondisk, buf->raw_offset);
 283                         if (n == -1) {
 284                                 if (AssertOnFailure)
 285                                         err(1, "get_buffer: %s:%016llx "
 286                                             "Read failed at offset %016llx",
 287                                             volume->name,
 288                                             (long long)buf->buf_offset,
 289                                             (long long)buf->raw_offset);
 290                                 bzero(ondisk, HAMMER_BUFSIZE);
 291                         }
 292                 }
 293         }
 294         if (isnew > 0) {
 295                 bzero(ondisk, HAMMER_BUFSIZE);
 296                 buf->cache.modified = 1;
 297         }
 298         if (dora)
 299                 get_buffer_readahead(buf);
 300         return(buf);
 301 }
 302
 303 static void
 304 get_buffer_readahead(struct buffer_info *base)
 305 {
 306         struct buffer_info *buf;
 307         struct volume_info *vol;
 308         hammer_off_t buf_offset;
 309         int64_t raw_offset;
 310         int ri = UseReadBehind;
 311         int re = UseReadAhead;
 312
 313         raw_offset = base->raw_offset + ri * HAMMER_BUFSIZE;
 314         vol = base->volume;
 315
 316         while (ri < re) {
 317                 if (raw_offset >= vol->ondisk->vol_buf_end)
 318                         break;
 319                 if (raw_offset < vol->ondisk->vol_buf_beg || ri == 0) {
 320                         ++ri;
 321                         raw_offset += HAMMER_BUFSIZE;
 322                         continue;
 323                 }
 324                 buf_offset = HAMMER_ENCODE_RAW_BUFFER(vol->vol_no,
 325                         raw_offset - vol->ondisk->vol_buf_beg);
 326                 buf = find_buffer(vol, buf_offset);
 327                 if (buf == NULL) {
 328                         buf = get_buffer(buf_offset, -1);
 329                         rel_buffer(buf);
 330                 }
 331                 ++ri;
 332                 raw_offset += HAMMER_BUFSIZE;
 333         }
 334 }
 335
 336 void
 337 rel_buffer(struct buffer_info *buffer)
 338 {
 339         struct volume_info *volume;
 340         int hi;
 341
 342         if (buffer == NULL)
 343                 return;
 344         assert(buffer->cache.refs > 0);
 345         if (--buffer->cache.refs == 0) {
 346                 if (buffer->cache.delete) {
 347                         hi = buffer_hash(buffer->buf_offset);
 348                         volume = buffer->volume;
 349                         if (buffer->cache.modified)
 350                                 flush_buffer(buffer);
 351                         TAILQ_REMOVE(&volume->buffer_lists[hi], buffer, entry);
 352                         hammer_cache_del(&buffer->cache);
 353                         free(buffer->ondisk);
 354                         free(buffer);
 355                         rel_volume(volume);
 356                 }
 357         }
 358 }
 359
 360 /*
 361  * Retrieve a pointer to a buffer data given a buffer offset.  The underlying
 362  * bufferp is freed if isnew or the offset is out of range of the cached data.
 363  * If bufferp is freed a referenced buffer is loaded into it.
 364  */
 365 void *
 366 get_buffer_data(hammer_off_t buf_offset, struct buffer_info **bufferp,
 367                 int isnew)
 368 {
 369         if (*bufferp != NULL) {
 370                 if (isnew > 0 ||
 371                     (((*bufferp)->buf_offset ^ buf_offset) & ~HAMMER_BUFMASK64)) {
 372                         rel_buffer(*bufferp);
 373                         *bufferp = NULL;
 374                 }
 375         }
 376         return(get_ondisk(buf_offset, bufferp, isnew));
 377 }
 378
 379 /*
 380  * Retrieve a pointer to a B-Tree node given a cluster offset.  The underlying
 381  * bufferp is freed if non-NULL and a referenced buffer is loaded into it.
 382  */
 383 hammer_node_ondisk_t
 384 get_node(hammer_off_t node_offset, struct buffer_info **bufferp)
 385 {
 386         if (*bufferp != NULL) {
 387                 rel_buffer(*bufferp);
 388                 *bufferp = NULL;
 389         }
 390         return(get_ondisk(node_offset, bufferp, 0));
 391 }
 392
 393 /*
 394  * Return a pointer to a buffer data given a buffer offset.
 395  * If *bufferp is NULL acquire the buffer otherwise use that buffer.
 396  */
 397 static __inline
 398 void *
 399 get_ondisk(hammer_off_t buf_offset, struct buffer_info **bufferp,
 400         int isnew)
 401 {
 402         struct buffer_info *buffer;
 403
 404         buffer = *bufferp;
 405         if (buffer == NULL) {
 406                 buffer = *bufferp = get_buffer(buf_offset, isnew);
 407                 if (buffer == NULL)
 408                         return(NULL);
 409         }
 410
 411         return((char *)buffer->ondisk +
 412                 ((int32_t)buf_offset & HAMMER_BUFMASK));
 413 }
 414
 415 /*
 416  * Allocate HAMMER elements - btree nodes, meta data, data storage
 417  */
 418 void *
 419 alloc_btree_element(hammer_off_t *offp,
 420                     struct buffer_info **data_bufferp)
 421 {
 422         hammer_node_ondisk_t node;
 423
 424         node = alloc_blockmap(HAMMER_ZONE_BTREE_INDEX, sizeof(*node),
 425                               offp, data_bufferp);
 426         bzero(node, sizeof(*node));
 427         return (node);
 428 }
 429
 430 void *
 431 alloc_meta_element(hammer_off_t *offp, int32_t data_len,
 432                    struct buffer_info **data_bufferp)
 433 {
 434         void *data;
 435
 436         data = alloc_blockmap(HAMMER_ZONE_META_INDEX, data_len,
 437                               offp, data_bufferp);
 438         bzero(data, data_len);
 439         return (data);
 440 }
 441
 442 /*
 443  * The only data_len supported by HAMMER userspace for large data zone
 444  * (zone 10) is HAMMER_BUFSIZE which is 16KB.  >16KB data does not fit
 445  * in a buffer allocated by get_buffer().  Also alloc_blockmap() does
 446  * not consider >16KB buffer size.
 447  */
 448 void *
 449 alloc_data_element(hammer_off_t *offp, int32_t data_len,
 450                    struct buffer_info **data_bufferp)
 451 {
 452         void *data;
 453
 454         if (data_len >= HAMMER_BUFSIZE) {
 455                 assert(data_len == HAMMER_BUFSIZE); /* just one buffer */
 456                 data = alloc_blockmap(HAMMER_ZONE_LARGE_DATA_INDEX, data_len,
 457                                       offp, data_bufferp);
 458                 bzero(data, data_len);
 459         } else if (data_len) {
 460                 data = alloc_blockmap(HAMMER_ZONE_SMALL_DATA_INDEX, data_len,
 461                                       offp, data_bufferp);
 462                 bzero(data, data_len);
 463         } else {
 464                 data = NULL;
 465         }
 466         return (data);
 467 }
 468
 469 /*
 470  * Format a new freemap.  Set all layer1 entries to UNAVAIL.  The initialize
 471  * code will load each volume's freemap.
 472  */
 473 void
 474 format_freemap(struct volume_info *root_vol)
 475 {
 476         struct buffer_info *buffer = NULL;
 477         hammer_off_t layer1_offset;
 478         hammer_blockmap_t blockmap;
 479         struct hammer_blockmap_layer1 *layer1;
 480         int i, isnew;
 481
 482         /* Only root volume needs formatting */
 483         assert(root_vol->vol_no == RootVolNo);
 484
 485         layer1_offset = alloc_bigblock(root_vol, HAMMER_ZONE_FREEMAP_INDEX);
 486         for (i = 0; i < (int)HAMMER_BLOCKMAP_RADIX1; ++i) {
 487                 isnew = ((i % HAMMER_BLOCKMAP_RADIX1_PERBUFFER) == 0);
 488                 layer1 = get_buffer_data(layer1_offset + i * sizeof(*layer1),
 489                                          &buffer, isnew);
 490                 bzero(layer1, sizeof(*layer1));
 491                 layer1->phys_offset = HAMMER_BLOCKMAP_UNAVAIL;
 492                 layer1->blocks_free = 0;
 493                 layer1->layer1_crc = crc32(layer1, HAMMER_LAYER1_CRCSIZE);
 494         }
 495         rel_buffer(buffer);
 496
 497         blockmap = &root_vol->ondisk->vol0_blockmap[HAMMER_ZONE_FREEMAP_INDEX];
 498         bzero(blockmap, sizeof(*blockmap));
 499         blockmap->phys_offset = layer1_offset;
 500         blockmap->first_offset = 0;
 501         blockmap->next_offset = HAMMER_ENCODE_RAW_BUFFER(0, 0);
 502         blockmap->alloc_offset = HAMMER_ENCODE_RAW_BUFFER(255, -1);
 503         blockmap->entry_crc = crc32(blockmap, HAMMER_BLOCKMAP_CRCSIZE);
 504         root_vol->cache.modified = 1;
 505 }
 506
 507 /*
 508  * Load the volume's remaining free space into the freemap.
 509  *
 510  * Returns the number of big-blocks available.
 511  */
 512 int64_t
 513 initialize_freemap(struct volume_info *vol)
 514 {
 515         struct volume_info *root_vol;
 516         struct buffer_info *buffer1 = NULL;
 517         struct buffer_info *buffer2 = NULL;
 518         struct hammer_blockmap_layer1 *layer1;
 519         struct hammer_blockmap_layer2 *layer2;
 520         hammer_off_t layer1_base;
 521         hammer_off_t layer1_offset;
 522         hammer_off_t layer2_offset;
 523         hammer_off_t phys_offset;
 524         hammer_off_t aligned_vol_free_end;
 525         hammer_blockmap_t freemap;
 526         int64_t count = 0;
 527
 528         root_vol = get_volume(RootVolNo);
 529         aligned_vol_free_end = (vol->vol_free_end + HAMMER_BLOCKMAP_LAYER2_MASK)
 530                                 & ~HAMMER_BLOCKMAP_LAYER2_MASK;
 531
 532         printf("initialize freemap volume %d\n", vol->vol_no);
 533
 534         /*
 535          * Initialize the freemap.  First preallocate the big-blocks required
 536          * to implement layer2.   This preallocation is a bootstrap allocation
 537          * using blocks from the target volume.
 538          */
 539         freemap = &root_vol->ondisk->vol0_blockmap[HAMMER_ZONE_FREEMAP_INDEX];
 540         layer1_base = freemap->phys_offset;
 541
 542         for (phys_offset = HAMMER_ENCODE_RAW_BUFFER(vol->vol_no, 0);
 543              phys_offset < aligned_vol_free_end;
 544              phys_offset += HAMMER_BLOCKMAP_LAYER2) {
 545                 layer1_offset = layer1_base +
 546                                 HAMMER_BLOCKMAP_LAYER1_OFFSET(phys_offset);
 547                 layer1 = get_buffer_data(layer1_offset, &buffer1, 0);
 548                 if (layer1->phys_offset == HAMMER_BLOCKMAP_UNAVAIL) {
 549                         layer1->phys_offset = alloc_bigblock(vol,
 550                                                 HAMMER_ZONE_FREEMAP_INDEX);
 551                         layer1->blocks_free = 0;
 552                         buffer1->cache.modified = 1;
 553                         layer1->layer1_crc = crc32(layer1,
 554                                                    HAMMER_LAYER1_CRCSIZE);
 555                 }
 556         }
 557
 558         /*
 559          * Now fill everything in.
 560          */
 561         for (phys_offset = HAMMER_ENCODE_RAW_BUFFER(vol->vol_no, 0);
 562              phys_offset < aligned_vol_free_end;
 563              phys_offset += HAMMER_BIGBLOCK_SIZE) {
 564                 layer1_offset = layer1_base +
 565                                 HAMMER_BLOCKMAP_LAYER1_OFFSET(phys_offset);
 566                 layer1 = get_buffer_data(layer1_offset, &buffer1, 0);
 567
 568                 assert(layer1->phys_offset != HAMMER_BLOCKMAP_UNAVAIL);
 569                 layer2_offset = layer1->phys_offset +
 570                                 HAMMER_BLOCKMAP_LAYER2_OFFSET(phys_offset);
 571
 572                 layer2 = get_buffer_data(layer2_offset, &buffer2, 0);
 573                 bzero(layer2, sizeof(*layer2));
 574                 if (phys_offset < vol->vol_free_off) {
 575                         /*
 576                          * Fixups XXX - big-blocks already allocated as part
 577                          * of the freemap bootstrap.
 578                          */
 579                         if (layer2->zone == 0) {
 580                                 layer2->zone = HAMMER_ZONE_FREEMAP_INDEX;
 581                                 layer2->append_off = HAMMER_BIGBLOCK_SIZE;
 582                                 layer2->bytes_free = 0;
 583                         }
 584                 } else if (phys_offset < vol->vol_free_end) {
 585                         ++layer1->blocks_free;
 586                         layer1->layer1_crc = crc32(layer1,
 587                                                    HAMMER_LAYER1_CRCSIZE);
 588                         buffer1->cache.modified = 1;
 589                         layer2->zone = 0;
 590                         layer2->append_off = 0;
 591                         layer2->bytes_free = HAMMER_BIGBLOCK_SIZE;
 592                         ++count;
 593                 } else {
 594                         layer2->zone = HAMMER_ZONE_UNAVAIL_INDEX;
 595                         layer2->append_off = HAMMER_BIGBLOCK_SIZE;
 596                         layer2->bytes_free = 0;
 597                 }
 598                 layer2->entry_crc = crc32(layer2, HAMMER_LAYER2_CRCSIZE);
 599                 buffer2->cache.modified = 1;
 600         }
 601         rel_buffer(buffer1);
 602         rel_buffer(buffer2);
 603         rel_volume(root_vol);
 604         return(count);
 605 }
 606
 607 /*
 608  * Returns the number of big-blocks available for filesystem data and undos
 609  * without formatting.
 610  */
 611 int64_t
 612 count_freemap(struct volume_info *vol)
 613 {
 614         hammer_off_t phys_offset;
 615         hammer_off_t vol_free_off;
 616         hammer_off_t aligned_vol_free_end;
 617         int64_t count = 0;
 618
 619         vol_free_off = HAMMER_ENCODE_RAW_BUFFER(vol->vol_no, 0);
 620         aligned_vol_free_end = (vol->vol_free_end + HAMMER_BLOCKMAP_LAYER2_MASK)
 621                                 & ~HAMMER_BLOCKMAP_LAYER2_MASK;
 622
 623         if (vol->vol_no == RootVolNo)
 624                 vol_free_off += HAMMER_BIGBLOCK_SIZE;
 625
 626         for (phys_offset = HAMMER_ENCODE_RAW_BUFFER(vol->vol_no, 0);
 627              phys_offset < aligned_vol_free_end;
 628              phys_offset += HAMMER_BLOCKMAP_LAYER2) {
 629                 vol_free_off += HAMMER_BIGBLOCK_SIZE;
 630         }
 631
 632         for (phys_offset = HAMMER_ENCODE_RAW_BUFFER(vol->vol_no, 0);
 633              phys_offset < aligned_vol_free_end;
 634              phys_offset += HAMMER_BIGBLOCK_SIZE) {
 635                 if (phys_offset < vol_free_off) {
 636                         ;
 637                 } else if (phys_offset < vol->vol_free_end) {
 638                         ++count;
 639                 }
 640         }
 641
 642         return(count);
 643 }
 644
 645 /*
 646  * Allocate big-blocks using our poor-man's volume->vol_free_off.
 647  *
 648  * If the zone is HAMMER_ZONE_FREEMAP_INDEX we are bootstrapping the freemap
 649  * itself and cannot update it yet.
 650  */
 651 hammer_off_t
 652 alloc_bigblock(struct volume_info *volume, int zone)
 653 {
 654         struct buffer_info *buffer1 = NULL;
 655         struct buffer_info *buffer2 = NULL;
 656         struct volume_info *root_vol;
 657         hammer_off_t result_offset;
 658         hammer_off_t layer_offset;
 659         hammer_blockmap_t freemap;
 660         struct hammer_blockmap_layer1 *layer1;
 661         struct hammer_blockmap_layer2 *layer2;
 662
 663         if (volume == NULL)
 664                 volume = get_volume(RootVolNo);
 665
 666         result_offset = volume->vol_free_off;
 667         if (result_offset >= volume->vol_free_end)
 668                 errx(1, "alloc_bigblock: Ran out of room, filesystem too small");
 669         volume->vol_free_off += HAMMER_BIGBLOCK_SIZE;
 670
 671         /*
 672          * Update the freemap.
 673          */
 674         if (zone != HAMMER_ZONE_FREEMAP_INDEX) {
 675                 root_vol = get_volume(RootVolNo);
 676                 freemap = &root_vol->ondisk->vol0_blockmap[HAMMER_ZONE_FREEMAP_INDEX];
 677                 layer_offset = freemap->phys_offset +
 678                                HAMMER_BLOCKMAP_LAYER1_OFFSET(result_offset);
 679                 layer1 = get_buffer_data(layer_offset, &buffer1, 0);
 680                 assert(layer1->phys_offset != HAMMER_BLOCKMAP_UNAVAIL);
 681                 --layer1->blocks_free;
 682                 layer1->layer1_crc = crc32(layer1, HAMMER_LAYER1_CRCSIZE);
 683                 buffer1->cache.modified = 1;
 684                 layer_offset = layer1->phys_offset +
 685                                HAMMER_BLOCKMAP_LAYER2_OFFSET(result_offset);
 686                 layer2 = get_buffer_data(layer_offset, &buffer2, 0);
 687                 assert(layer2->zone == 0);
 688                 layer2->zone = zone;
 689                 layer2->append_off = HAMMER_BIGBLOCK_SIZE;
 690                 layer2->bytes_free = 0;
 691                 layer2->entry_crc = crc32(layer2, HAMMER_LAYER2_CRCSIZE);
 692                 buffer2->cache.modified = 1;
 693
 694                 --root_vol->ondisk->vol0_stat_freebigblocks;
 695                 root_vol->cache.modified = 1;
 696
 697                 rel_buffer(buffer1);
 698                 rel_buffer(buffer2);
 699                 rel_volume(root_vol);
 700         }
 701
 702         rel_volume(volume);
 703         return(result_offset);
 704 }
 705
 706 /*
 707  * Format the undomap for the root volume.
 708  */
 709 void
 710 format_undomap(struct volume_info *root_vol)
 711 {
 712         const int undo_zone = HAMMER_ZONE_UNDO_INDEX;
 713         hammer_off_t undo_limit;
 714         hammer_blockmap_t blockmap;
 715         struct hammer_volume_ondisk *ondisk;
 716         struct buffer_info *buffer = NULL;
 717         hammer_off_t scan;
 718         int n;
 719         int limit_index;
 720         u_int32_t seqno;
 721
 722         /* Only root volume needs formatting */
 723         assert(root_vol->vol_no == RootVolNo);
 724         ondisk = root_vol->ondisk;
 725
 726         /*
 727          * Size the undo buffer in multiples of HAMMER_BIGBLOCK_SIZE,
 728          * up to HAMMER_UNDO_LAYER2 big-blocks.  Size to approximately
 729          * 0.1% of the disk.
 730          *
 731          * The minimum UNDO fifo size is 500MB, or approximately 1% of
 732          * the recommended 50G disk.
 733          *
 734          * Changing this minimum is rather dangerous as complex filesystem
 735          * operations can cause the UNDO FIFO to fill up otherwise.
 736          */
 737         undo_limit = UndoBufferSize;
 738         if (undo_limit == 0) {
 739                 undo_limit = (ondisk->vol_buf_end - ondisk->vol_buf_beg) / 1000;
 740                 if (undo_limit < 500*1024*1024)
 741                         undo_limit = 500*1024*1024;
 742         }
 743         undo_limit = (undo_limit + HAMMER_BIGBLOCK_MASK64) &
 744                      ~HAMMER_BIGBLOCK_MASK64;
 745         if (undo_limit < HAMMER_BIGBLOCK_SIZE)
 746                 undo_limit = HAMMER_BIGBLOCK_SIZE;
 747         if (undo_limit > HAMMER_BIGBLOCK_SIZE * HAMMER_UNDO_LAYER2)
 748                 undo_limit = HAMMER_BIGBLOCK_SIZE * HAMMER_UNDO_LAYER2;
 749         UndoBufferSize = undo_limit;
 750
 751         blockmap = &ondisk->vol0_blockmap[undo_zone];
 752         bzero(blockmap, sizeof(*blockmap));
 753         blockmap->phys_offset = HAMMER_BLOCKMAP_UNAVAIL;
 754         blockmap->first_offset = HAMMER_ZONE_ENCODE(undo_zone, 0);
 755         blockmap->next_offset = blockmap->first_offset;
 756         blockmap->alloc_offset = HAMMER_ZONE_ENCODE(undo_zone, undo_limit);
 757         blockmap->entry_crc = crc32(blockmap, HAMMER_BLOCKMAP_CRCSIZE);
 758
 759         limit_index = undo_limit / HAMMER_BIGBLOCK_SIZE;
 760         assert(limit_index <= HAMMER_UNDO_LAYER2);
 761
 762         for (n = 0; n < limit_index; ++n) {
 763                 ondisk->vol0_undo_array[n] = alloc_bigblock(NULL,
 764                                                         HAMMER_ZONE_UNDO_INDEX);
 765         }
 766         while (n < HAMMER_UNDO_LAYER2) {
 767                 ondisk->vol0_undo_array[n++] = HAMMER_BLOCKMAP_UNAVAIL;
 768         }
 769
 770         /*
 771          * Pre-initialize the UNDO blocks (HAMMER version 4+)
 772          */
 773         printf("initializing the undo map (%jd MB)\n",
 774                 (intmax_t)(blockmap->alloc_offset & HAMMER_OFF_LONG_MASK) /
 775                 (1024 * 1024));
 776
 777         scan = blockmap->first_offset;
 778         seqno = 0;
 779
 780         while (scan < blockmap->alloc_offset) {
 781                 hammer_fifo_head_t head;
 782                 hammer_fifo_tail_t tail;
 783                 int isnew;
 784                 int bytes = HAMMER_UNDO_ALIGN;
 785
 786                 isnew = ((scan & HAMMER_BUFMASK64) == 0);
 787                 head = get_buffer_data(scan, &buffer, isnew);
 788                 buffer->cache.modified = 1;
 789                 tail = (void *)((char *)head + bytes - sizeof(*tail));
 790
 791                 bzero(head, bytes);
 792                 head->hdr_signature = HAMMER_HEAD_SIGNATURE;
 793                 head->hdr_type = HAMMER_HEAD_TYPE_DUMMY;
 794                 head->hdr_size = bytes;
 795                 head->hdr_seq = seqno++;
 796
 797                 tail->tail_signature = HAMMER_TAIL_SIGNATURE;
 798                 tail->tail_type = HAMMER_HEAD_TYPE_DUMMY;
 799                 tail->tail_size = bytes;
 800
 801                 head->hdr_crc = crc32(head, HAMMER_FIFO_HEAD_CRCOFF) ^
 802                                 crc32(head + 1, bytes - sizeof(*head));
 803
 804                 scan += bytes;
 805         }
 806         rel_buffer(buffer);
 807 }
 808
 809 /*
 810  * Format a new blockmap.  This is mostly a degenerate case because
 811  * all allocations are now actually done from the freemap.
 812  */
 813 void
 814 format_blockmap(hammer_blockmap_t blockmap, int zone, hammer_off_t offset)
 815 {
 816         hammer_off_t zone_base = HAMMER_ZONE_ENCODE(zone, offset);
 817
 818         bzero(blockmap, sizeof(*blockmap));
 819         blockmap->phys_offset = 0;
 820         blockmap->first_offset = zone_base;
 821         blockmap->next_offset = zone_base;
 822         blockmap->alloc_offset = HAMMER_ENCODE(zone, 255, -1);
 823         blockmap->entry_crc = crc32(blockmap, HAMMER_BLOCKMAP_CRCSIZE);
 824 }
 825
 826 /*
 827  * Allocate a chunk of data out of a blockmap.  This is a simplified
 828  * version which uses next_offset as a simple allocation iterator.
 829  */
 830 static
 831 void *
 832 alloc_blockmap(int zone, int bytes, hammer_off_t *result_offp,
 833                struct buffer_info **bufferp)
 834 {
 835         struct buffer_info *buffer1 = NULL;
 836         struct buffer_info *buffer2 = NULL;
 837         struct volume_info *volume;
 838         hammer_blockmap_t blockmap;
 839         hammer_blockmap_t freemap;
 840         struct hammer_blockmap_layer1 *layer1;
 841         struct hammer_blockmap_layer2 *layer2;
 842         hammer_off_t layer1_offset;
 843         hammer_off_t layer2_offset;
 844         hammer_off_t chunk_offset;
 845         void *ptr;
 846
 847         volume = get_volume(RootVolNo);
 848
 849         blockmap = &volume->ondisk->vol0_blockmap[zone];
 850         freemap = &volume->ondisk->vol0_blockmap[HAMMER_ZONE_FREEMAP_INDEX];
 851         assert(HAMMER_ZONE_DECODE(blockmap->next_offset) == zone);
 852
 853         /*
 854          * Alignment and buffer-boundary issues.  If the allocation would
 855          * cross a buffer boundary we have to skip to the next buffer.
 856          */
 857         bytes = (bytes + 15) & ~15;
 858         assert(bytes > 0 && bytes <= HAMMER_BUFSIZE);  /* not HAMMER_XBUFSIZE */
 859         assert(zone >= HAMMER_ZONE2_MAPPED_INDEX && zone < HAMMER_MAX_ZONES);
 860
 861 again:
 862         assert(blockmap->next_offset != HAMMER_ZONE_ENCODE(zone + 1, 0));
 863
 864         if ((blockmap->next_offset ^ (blockmap->next_offset + bytes - 1)) &
 865             ~HAMMER_BUFMASK64) {
 866                 volume->cache.modified = 1;
 867                 blockmap->next_offset = (blockmap->next_offset + bytes - 1) &
 868                                         ~HAMMER_BUFMASK64;
 869         }
 870         chunk_offset = blockmap->next_offset & HAMMER_BIGBLOCK_MASK;
 871
 872         /*
 873          * Dive layer 1.
 874          */
 875         layer1_offset = freemap->phys_offset +
 876                         HAMMER_BLOCKMAP_LAYER1_OFFSET(blockmap->next_offset);
 877
 878         layer1 = get_buffer_data(layer1_offset, &buffer1, 0);
 879         assert(!(chunk_offset == 0 && layer1->blocks_free == 0));
 880
 881         if (layer1->phys_offset == HAMMER_BLOCKMAP_UNAVAIL) {
 882                 fprintf(stderr, "alloc_blockmap: ran out of space!\n");
 883                 exit(1);
 884         }
 885
 886         /*
 887          * Dive layer 2, each entry represents a big-block.
 888          */
 889         layer2_offset = layer1->phys_offset +
 890                         HAMMER_BLOCKMAP_LAYER2_OFFSET(blockmap->next_offset);
 891
 892         layer2 = get_buffer_data(layer2_offset, &buffer2, 0);
 893
 894         if (layer2->zone == HAMMER_ZONE_UNAVAIL_INDEX) {
 895                 fprintf(stderr, "alloc_blockmap: ran out of space!\n");
 896                 exit(1);
 897         }
 898
 899         /*
 900          * If we are entering a new big-block assign ownership to our
 901          * zone.  If the big-block is owned by another zone skip it.
 902          */
 903         if (layer2->zone == 0) {
 904                 --layer1->blocks_free;
 905                 layer1->layer1_crc = crc32(layer1, HAMMER_LAYER1_CRCSIZE);
 906                 layer2->zone = zone;
 907                 --volume->ondisk->vol0_stat_freebigblocks;
 908                 assert(layer2->bytes_free == HAMMER_BIGBLOCK_SIZE);
 909                 assert(layer2->append_off == 0);
 910         }
 911         if (layer2->zone != zone) {
 912                 volume->cache.modified = 1;
 913                 blockmap->next_offset = (blockmap->next_offset + HAMMER_BIGBLOCK_SIZE) &
 914                                         ~HAMMER_BIGBLOCK_MASK64;
 915                 goto again;
 916         }
 917
 918         buffer1->cache.modified = 1;
 919         buffer2->cache.modified = 1;
 920         volume->cache.modified = 1;
 921         assert(layer2->append_off == chunk_offset);
 922         layer2->bytes_free -= bytes;
 923         *result_offp = blockmap->next_offset;
 924         blockmap->next_offset += bytes;
 925         layer2->append_off = (int)blockmap->next_offset &
 926                               HAMMER_BIGBLOCK_MASK;
 927
 928         layer2->entry_crc = crc32(layer2, HAMMER_LAYER2_CRCSIZE);
 929
 930         ptr = get_buffer_data(*result_offp, bufferp, 0);
 931         (*bufferp)->cache.modified = 1;
 932
 933         rel_buffer(buffer1);
 934         rel_buffer(buffer2);
 935         rel_volume(volume);
 936         return(ptr);
 937 }
 938
 939 /*
 940  * Flush various tracking structures to disk
 941  */
 942 void
 943 flush_all_volumes(void)
 944 {
 945         struct volume_info *vol;
 946
 947         TAILQ_FOREACH(vol, &VolList, entry)
 948                 flush_volume(vol);
 949 }
 950
 951 void
 952 flush_volume(struct volume_info *volume)
 953 {
 954         struct buffer_info *buffer;
 955         int i;
 956
 957         for (i = 0; i < HAMMER_BUFLISTS; ++i) {
 958                 TAILQ_FOREACH(buffer, &volume->buffer_lists[i], entry)
 959                         flush_buffer(buffer);
 960         }
 961         writehammerbuf(volume, volume->ondisk, 0);
 962         volume->cache.modified = 0;
 963 }
 964
 965 void
 966 flush_buffer(struct buffer_info *buffer)
 967 {
 968         writehammerbuf(buffer->volume, buffer->ondisk, buffer->raw_offset);
 969         buffer->cache.modified = 0;
 970 }
 971
 972 /*
 973  * Core I/O operations
 974  */
 975 static int
 976 readhammerbuf(struct volume_info *vol, void *data, int64_t offset)
 977 {
 978         ssize_t n;
 979
 980         n = pread(vol->fd, data, HAMMER_BUFSIZE, offset);
 981         if (n != HAMMER_BUFSIZE)
 982                 return(-1);
 983         return(0);
 984 }
 985
 986 static void
 987 writehammerbuf(struct volume_info *vol, const void *data, int64_t offset)
 988 {
 989         ssize_t n;
 990
 991         n = pwrite(vol->fd, data, HAMMER_BUFSIZE, offset);
 992         if (n != HAMMER_BUFSIZE)
 993                 err(1, "Write volume %d (%s)", vol->vol_no, vol->name);
 994 }