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  * $DragonFly: src/sbin/hammer/ondisk.c,v 1.23.2.1 2008/07/24 05:41:56 dillon Exp $
  35  */
  36
  37 #include <sys/types.h>
  38 #include <assert.h>
  39 #include <stdio.h>
  40 #include <stdlib.h>
  41 #include <stdarg.h>
  42 #include <string.h>
  43 #include <unistd.h>
  44 #include <stddef.h>
  45 #include <err.h>
  46 #include <fcntl.h>
  47 #include "hammer_util.h"
  48
  49 static void *alloc_blockmap(int zone, int bytes, hammer_off_t *result_offp,
  50                         struct buffer_info **bufferp);
  51 static hammer_off_t alloc_bigblock(struct volume_info *volume, int zone);
  52 #if 0
  53 static void init_fifo_head(hammer_fifo_head_t head, u_int16_t hdr_type);
  54 static hammer_off_t hammer_alloc_fifo(int32_t base_bytes, int32_t ext_bytes,
  55                         struct buffer_info **bufp, u_int16_t hdr_type);
  56 static void readhammerbuf(struct volume_info *vol, void *data,
  57                         int64_t offset);
  58 #endif
  59 static void writehammerbuf(struct volume_info *vol, const void *data,
  60                         int64_t offset);
  61
  62 int DebugOpt;
  63
  64 uuid_t Hammer_FSType;
  65 uuid_t Hammer_FSId;
  66 int64_t BootAreaSize;
  67 int64_t MemAreaSize;
  68 int64_t UndoBufferSize;
  69 int     UsingSuperClusters;
  70 int     NumVolumes;
  71 int     RootVolNo = -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 /*
  85  * Lookup the requested information structure and related on-disk buffer.
  86  * Missing structures are created.
  87  */
  88 struct volume_info *
  89 setup_volume(int32_t vol_no, const char *filename, int isnew, int oflags)
  90 {
  91         struct volume_info *vol;
  92         struct volume_info *scan;
  93         struct hammer_volume_ondisk *ondisk;
  94         int i, n;
  95
  96         /*
  97          * Allocate the volume structure
  98          */
  99         vol = malloc(sizeof(*vol));
 100         bzero(vol, sizeof(*vol));
 101         for (i = 0; i < HAMMER_BUFLISTS; ++i)
 102                 TAILQ_INIT(&vol->buffer_lists[i]);
 103         vol->name = strdup(filename);
 104         vol->fd = open(filename, oflags);
 105         if (vol->fd < 0) {
 106                 free(vol->name);
 107                 free(vol);
 108                 err(1, "setup_volume: %s: Open failed", filename);
 109         }
 110
 111         /*
 112          * Read or initialize the volume header
 113          */
 114         vol->ondisk = ondisk = malloc(HAMMER_BUFSIZE);
 115         if (isnew) {
 116                 bzero(ondisk, HAMMER_BUFSIZE);
 117         } else {
 118                 n = pread(vol->fd, ondisk, HAMMER_BUFSIZE, 0);
 119                 if (n != HAMMER_BUFSIZE) {
 120                         err(1, "setup_volume: %s: Read failed at offset 0",
 121                             filename);
 122                 }
 123                 vol_no = ondisk->vol_no;
 124                 if (RootVolNo < 0) {
 125                         RootVolNo = ondisk->vol_rootvol;
 126                 } else if (RootVolNo != (int)ondisk->vol_rootvol) {
 127                         errx(1, "setup_volume: %s: root volume disagreement: "
 128                                 "%d vs %d",
 129                                 vol->name, RootVolNo, ondisk->vol_rootvol);
 130                 }
 131
 132                 if (bcmp(&Hammer_FSType, &ondisk->vol_fstype, sizeof(Hammer_FSType)) != 0) {
 133                         errx(1, "setup_volume: %s: Header does not indicate "
 134                                 "that this is a hammer volume", vol->name);
 135                 }
 136                 if (TAILQ_EMPTY(&VolList)) {
 137                         Hammer_FSId = vol->ondisk->vol_fsid;
 138                 } else if (bcmp(&Hammer_FSId, &ondisk->vol_fsid, sizeof(Hammer_FSId)) != 0) {
 139                         errx(1, "setup_volume: %s: FSId does match other "
 140                                 "volumes!", vol->name);
 141                 }
 142         }
 143         vol->vol_no = vol_no;
 144
 145         if (isnew) {
 146                 /*init_fifo_head(&ondisk->head, HAMMER_HEAD_TYPE_VOL);*/
 147                 vol->cache.modified = 1;
 148         }
 149
 150         /*
 151          * Link the volume structure in
 152          */
 153         TAILQ_FOREACH(scan, &VolList, entry) {
 154                 if (scan->vol_no == vol_no) {
 155                         errx(1, "setup_volume %s: Duplicate volume number %d "
 156                                 "against %s", filename, vol_no, scan->name);
 157                 }
 158         }
 159         TAILQ_INSERT_TAIL(&VolList, vol, entry);
 160         return(vol);
 161 }
 162
 163 struct volume_info *
 164 get_volume(int32_t vol_no)
 165 {
 166         struct volume_info *vol;
 167
 168         TAILQ_FOREACH(vol, &VolList, entry) {
 169                 if (vol->vol_no == vol_no)
 170                         break;
 171         }
 172         if (vol == NULL)
 173                 errx(1, "get_volume: Volume %d does not exist!", vol_no);
 174         ++vol->cache.refs;
 175         /* not added to or removed from hammer cache */
 176         return(vol);
 177 }
 178
 179 void
 180 rel_volume(struct volume_info *volume)
 181 {
 182         /* not added to or removed from hammer cache */
 183         --volume->cache.refs;
 184 }
 185
 186 /*
 187  * Acquire the specified buffer.
 188  */
 189 struct buffer_info *
 190 get_buffer(hammer_off_t buf_offset, int isnew)
 191 {
 192         void *ondisk;
 193         struct buffer_info *buf;
 194         struct volume_info *volume;
 195         hammer_off_t orig_offset = buf_offset;
 196         int vol_no;
 197         int zone;
 198         int hi, n;
 199
 200         zone = HAMMER_ZONE_DECODE(buf_offset);
 201         if (zone > HAMMER_ZONE_RAW_BUFFER_INDEX) {
 202                 buf_offset = blockmap_lookup(buf_offset, NULL, NULL);
 203         }
 204         assert((buf_offset & HAMMER_OFF_ZONE_MASK) == HAMMER_ZONE_RAW_BUFFER);
 205         vol_no = HAMMER_VOL_DECODE(buf_offset);
 206         volume = get_volume(vol_no);
 207         buf_offset &= ~HAMMER_BUFMASK64;
 208
 209         hi = buffer_hash(buf_offset);
 210
 211         TAILQ_FOREACH(buf, &volume->buffer_lists[hi], entry) {
 212                 if (buf->buf_offset == buf_offset)
 213                         break;
 214         }
 215         if (buf == NULL) {
 216                 buf = malloc(sizeof(*buf));
 217                 bzero(buf, sizeof(*buf));
 218                 if (DebugOpt) {
 219                         fprintf(stderr, "get_buffer %016llx %016llx\n",
 220                                 orig_offset, buf_offset);
 221                 }
 222                 buf->buf_offset = buf_offset;
 223                 buf->buf_disk_offset = volume->ondisk->vol_buf_beg +
 224                                         (buf_offset & HAMMER_OFF_SHORT_MASK);
 225                 buf->volume = volume;
 226                 TAILQ_INSERT_TAIL(&volume->buffer_lists[hi], buf, entry);
 227                 ++volume->cache.refs;
 228                 buf->cache.u.buffer = buf;
 229                 hammer_cache_add(&buf->cache, ISBUFFER);
 230         }
 231         ++buf->cache.refs;
 232         hammer_cache_flush();
 233         if ((ondisk = buf->ondisk) == NULL) {
 234                 buf->ondisk = ondisk = malloc(HAMMER_BUFSIZE);
 235                 if (isnew == 0) {
 236                         n = pread(volume->fd, ondisk, HAMMER_BUFSIZE,
 237                                   buf->buf_disk_offset);
 238                         if (n != HAMMER_BUFSIZE) {
 239                                 err(1, "get_buffer: %s:%016llx Read failed at "
 240                                        "offset %lld",
 241                                     volume->name, buf->buf_offset,
 242                                     buf->buf_disk_offset);
 243                         }
 244                 }
 245         }
 246         if (isnew) {
 247                 bzero(ondisk, HAMMER_BUFSIZE);
 248                 buf->cache.modified = 1;
 249         }
 250         return(buf);
 251 }
 252
 253 void
 254 rel_buffer(struct buffer_info *buffer)
 255 {
 256         struct volume_info *volume;
 257         int hi;
 258
 259         assert(buffer->cache.refs > 0);
 260         if (--buffer->cache.refs == 0) {
 261                 if (buffer->cache.delete) {
 262                         hi = buffer_hash(buffer->buf_offset);
 263                         volume = buffer->volume;
 264                         if (buffer->cache.modified)
 265                                 flush_buffer(buffer);
 266                         TAILQ_REMOVE(&volume->buffer_lists[hi], buffer, entry);
 267                         hammer_cache_del(&buffer->cache);
 268                         free(buffer->ondisk);
 269                         free(buffer);
 270                         rel_volume(volume);
 271                 }
 272         }
 273 }
 274
 275 void *
 276 get_buffer_data(hammer_off_t buf_offset, struct buffer_info **bufferp,
 277                 int isnew)
 278 {
 279         struct buffer_info *buffer;
 280
 281         if ((buffer = *bufferp) != NULL) {
 282                 if (isnew ||
 283                     ((buffer->buf_offset ^ buf_offset) & ~HAMMER_BUFMASK64)) {
 284                         rel_buffer(buffer);
 285                         buffer = *bufferp = NULL;
 286                 }
 287         }
 288         if (buffer == NULL)
 289                 buffer = *bufferp = get_buffer(buf_offset, isnew);
 290         return((char *)buffer->ondisk + ((int32_t)buf_offset & HAMMER_BUFMASK));
 291 }
 292
 293 /*
 294  * Retrieve a pointer to a B-Tree node given a cluster offset.  The underlying
 295  * bufp is freed if non-NULL and a referenced buffer is loaded into it.
 296  */
 297 hammer_node_ondisk_t
 298 get_node(hammer_off_t node_offset, struct buffer_info **bufp)
 299 {
 300         struct buffer_info *buf;
 301
 302         if (*bufp)
 303                 rel_buffer(*bufp);
 304         *bufp = buf = get_buffer(node_offset, 0);
 305         return((void *)((char *)buf->ondisk +
 306                         (int32_t)(node_offset & HAMMER_BUFMASK)));
 307 }
 308
 309 /*
 310  * Allocate HAMMER elements - btree nodes, data storage, and record elements
 311  *
 312  * NOTE: hammer_alloc_fifo() initializes the fifo header for the returned
 313  * item and zero's out the remainder, so don't bzero() it.
 314  */
 315 void *
 316 alloc_btree_element(hammer_off_t *offp)
 317 {
 318         struct buffer_info *buffer = NULL;
 319         hammer_node_ondisk_t node;
 320
 321         node = alloc_blockmap(HAMMER_ZONE_BTREE_INDEX, sizeof(*node),
 322                               offp, &buffer);
 323         bzero(node, sizeof(*node));
 324         /* XXX buffer not released, pointer remains valid */
 325         return(node);
 326 }
 327
 328 void *
 329 alloc_data_element(hammer_off_t *offp, int32_t data_len,
 330                    struct buffer_info **data_bufferp)
 331 {
 332         void *data;
 333
 334         if (data_len >= HAMMER_BUFSIZE) {
 335                 assert(data_len <= HAMMER_BUFSIZE); /* just one buffer */
 336                 data = alloc_blockmap(HAMMER_ZONE_LARGE_DATA_INDEX, data_len,
 337                                       offp, data_bufferp);
 338                 bzero(data, data_len);
 339         } else if (data_len) {
 340                 data = alloc_blockmap(HAMMER_ZONE_SMALL_DATA_INDEX, data_len,
 341                                       offp, data_bufferp);
 342                 bzero(data, data_len);
 343         } else {
 344                 data = NULL;
 345         }
 346         return (data);
 347 }
 348
 349 /*
 350  * Format a new freemap.  Set all layer1 entries to UNAVAIL.  The initialize
 351  * code will load each volume's freemap.
 352  */
 353 void
 354 format_freemap(struct volume_info *root_vol, hammer_blockmap_t blockmap)
 355 {
 356         struct buffer_info *buffer = NULL;
 357         hammer_off_t layer1_offset;
 358         struct hammer_blockmap_layer1 *layer1;
 359         int i, isnew;
 360
 361         layer1_offset = alloc_bigblock(root_vol, HAMMER_ZONE_FREEMAP_INDEX);
 362         for (i = 0; i < (int)HAMMER_BLOCKMAP_RADIX1; ++i) {
 363                 isnew = ((i % HAMMER_BLOCKMAP_RADIX1_PERBUFFER) == 0);
 364                 layer1 = get_buffer_data(layer1_offset + i * sizeof(*layer1),
 365                                          &buffer, isnew);
 366                 bzero(layer1, sizeof(*layer1));
 367                 layer1->phys_offset = HAMMER_BLOCKMAP_UNAVAIL;
 368                 layer1->layer1_crc = crc32(layer1, HAMMER_LAYER1_CRCSIZE);
 369                 layer1->blocks_free = 0;
 370         }
 371         rel_buffer(buffer);
 372
 373         blockmap = &root_vol->ondisk->vol0_blockmap[HAMMER_ZONE_FREEMAP_INDEX];
 374         blockmap->phys_offset = layer1_offset;
 375         blockmap->alloc_offset = HAMMER_ENCODE_RAW_BUFFER(255, -1);
 376         blockmap->next_offset = HAMMER_ENCODE_RAW_BUFFER(0, 0);
 377         blockmap->reserved01 = 0;
 378         blockmap->entry_crc = crc32(blockmap, HAMMER_BLOCKMAP_CRCSIZE);
 379         root_vol->cache.modified = 1;
 380 }
 381
 382 /*
 383  * Load the volume's remaining free space into the freemap.
 384  *
 385  * Returns the number of bigblocks available.
 386  */
 387 int64_t
 388 initialize_freemap(struct volume_info *vol)
 389 {
 390         struct volume_info *root_vol;
 391         struct buffer_info *buffer1 = NULL;
 392         struct buffer_info *buffer2 = NULL;
 393         struct hammer_blockmap_layer1 *layer1;
 394         struct hammer_blockmap_layer2 *layer2;
 395         hammer_off_t layer1_base;
 396         hammer_off_t layer1_offset;
 397         hammer_off_t layer2_offset;
 398         hammer_off_t phys_offset;
 399         hammer_off_t aligned_vol_free_end;
 400         int64_t count = 0;
 401         int modified1 = 0;
 402
 403         root_vol = get_volume(RootVolNo);
 404         aligned_vol_free_end = (vol->vol_free_end + HAMMER_BLOCKMAP_LAYER2_MASK)
 405                                 & ~HAMMER_BLOCKMAP_LAYER2_MASK;
 406
 407         printf("initialize freemap volume %d\n", vol->vol_no);
 408
 409         /*
 410          * Initialize the freemap.  First preallocate the bigblocks required
 411          * to implement layer2.   This preallocation is a bootstrap allocation
 412          * using blocks from the target volume.
 413          */
 414         layer1_base = root_vol->ondisk->vol0_blockmap[
 415                                         HAMMER_ZONE_FREEMAP_INDEX].phys_offset;
 416         for (phys_offset = HAMMER_ENCODE_RAW_BUFFER(vol->vol_no, 0);
 417              phys_offset < aligned_vol_free_end;
 418              phys_offset += HAMMER_BLOCKMAP_LAYER2) {
 419                 layer1_offset = layer1_base +
 420                                 HAMMER_BLOCKMAP_LAYER1_OFFSET(phys_offset);
 421                 layer1 = get_buffer_data(layer1_offset, &buffer1, 0);
 422                 if (layer1->phys_offset == HAMMER_BLOCKMAP_UNAVAIL) {
 423                         layer1->phys_offset = alloc_bigblock(vol,
 424                                                 HAMMER_ZONE_FREEMAP_INDEX);
 425                         layer1->blocks_free = 0;
 426                         buffer1->cache.modified = 1;
 427                         layer1->layer1_crc = crc32(layer1,
 428                                                    HAMMER_LAYER1_CRCSIZE);
 429                 }
 430         }
 431
 432         /*
 433          * Now fill everything in.
 434          */
 435         for (phys_offset = HAMMER_ENCODE_RAW_BUFFER(vol->vol_no, 0);
 436              phys_offset < aligned_vol_free_end;
 437              phys_offset += HAMMER_LARGEBLOCK_SIZE) {
 438                 modified1 = 0;
 439                 layer1_offset = layer1_base +
 440                                 HAMMER_BLOCKMAP_LAYER1_OFFSET(phys_offset);
 441                 layer1 = get_buffer_data(layer1_offset, &buffer1, 0);
 442
 443                 assert(layer1->phys_offset != HAMMER_BLOCKMAP_UNAVAIL);
 444                 layer2_offset = layer1->phys_offset +
 445                                 HAMMER_BLOCKMAP_LAYER2_OFFSET(phys_offset);
 446
 447                 layer2 = get_buffer_data(layer2_offset, &buffer2, 0);
 448                 bzero(layer2, sizeof(*layer2));
 449                 if (phys_offset < vol->vol_free_off) {
 450                         /*
 451                          * Fixups XXX - bigblocks already allocated as part
 452                          * of the freemap bootstrap.
 453                          */
 454                         if (layer2->zone == 0) {
 455                                 layer2->zone = HAMMER_ZONE_FREEMAP_INDEX;
 456                                 layer2->append_off = HAMMER_LARGEBLOCK_SIZE;
 457                                 layer2->bytes_free = 0;
 458                         }
 459                 } else if (phys_offset < vol->vol_free_end) {
 460                         ++layer1->blocks_free;
 461                         buffer1->cache.modified = 1;
 462                         layer2->zone = 0;
 463                         layer2->append_off = 0;
 464                         layer2->bytes_free = HAMMER_LARGEBLOCK_SIZE;
 465                         ++count;
 466                         modified1 = 1;
 467                 } else {
 468                         layer2->zone = HAMMER_ZONE_UNAVAIL_INDEX;
 469                         layer2->append_off = HAMMER_LARGEBLOCK_SIZE;
 470                         layer2->bytes_free = 0;
 471                 }
 472                 layer2->entry_crc = crc32(layer2, HAMMER_LAYER2_CRCSIZE);
 473                 buffer2->cache.modified = 1;
 474
 475                 /*
 476                  * Finish-up layer 1
 477                  */
 478                 if (modified1) {
 479                         layer1->layer1_crc = crc32(layer1,
 480                                                    HAMMER_LAYER1_CRCSIZE);
 481                         buffer1->cache.modified = 1;
 482                 }
 483         }
 484         rel_buffer(buffer1);
 485         rel_buffer(buffer2);
 486         rel_volume(root_vol);
 487         return(count);
 488 }
 489
 490 /*
 491  * Allocate big-blocks using our poor-man's volume->vol_free_off.
 492  *
 493  * If the zone is HAMMER_ZONE_FREEMAP_INDEX we are bootstrapping the freemap
 494  * itself and cannot update it yet.
 495  */
 496 hammer_off_t
 497 alloc_bigblock(struct volume_info *volume, int zone)
 498 {
 499         struct buffer_info *buffer = NULL;
 500         struct volume_info *root_vol;
 501         hammer_off_t result_offset;
 502         hammer_off_t layer_offset;
 503         struct hammer_blockmap_layer1 *layer1;
 504         struct hammer_blockmap_layer2 *layer2;
 505         int didget;
 506
 507         if (volume == NULL) {
 508                 volume = get_volume(RootVolNo);
 509                 didget = 1;
 510         } else {
 511                 didget = 0;
 512         }
 513         result_offset = volume->vol_free_off;
 514         if (result_offset >= volume->vol_free_end)
 515                 panic("alloc_bigblock: Ran out of room, filesystem too small");
 516         volume->vol_free_off += HAMMER_LARGEBLOCK_SIZE;
 517
 518         /*
 519          * Update the freemap.
 520          */
 521         if (zone != HAMMER_ZONE_FREEMAP_INDEX) {
 522                 root_vol = get_volume(RootVolNo);
 523                 layer_offset = root_vol->ondisk->vol0_blockmap[
 524                                         HAMMER_ZONE_FREEMAP_INDEX].phys_offset;
 525                 layer_offset += HAMMER_BLOCKMAP_LAYER1_OFFSET(result_offset);
 526                 layer1 = get_buffer_data(layer_offset, &buffer, 0);
 527                 assert(layer1->phys_offset != HAMMER_BLOCKMAP_UNAVAIL);
 528                 --layer1->blocks_free;
 529                 layer1->layer1_crc = crc32(layer1, HAMMER_LAYER1_CRCSIZE);
 530                 buffer->cache.modified = 1;
 531                 layer_offset = layer1->phys_offset +
 532                                HAMMER_BLOCKMAP_LAYER2_OFFSET(result_offset);
 533                 layer2 = get_buffer_data(layer_offset, &buffer, 0);
 534                 assert(layer2->zone == 0);
 535                 layer2->zone = zone;
 536                 layer2->append_off = HAMMER_LARGEBLOCK_SIZE;
 537                 layer2->bytes_free = 0;
 538                 layer2->entry_crc = crc32(layer2, HAMMER_LAYER2_CRCSIZE);
 539                 buffer->cache.modified = 1;
 540
 541                 --root_vol->ondisk->vol0_stat_freebigblocks;
 542                 root_vol->cache.modified = 1;
 543
 544                 rel_buffer(buffer);
 545                 rel_volume(root_vol);
 546         }
 547
 548         if (didget)
 549                 rel_volume(volume);
 550         return(result_offset);
 551 }
 552
 553 /*
 554  * Format the undo-map for the root volume.
 555  */
 556 void
 557 format_undomap(hammer_volume_ondisk_t ondisk)
 558 {
 559         const int undo_zone = HAMMER_ZONE_UNDO_INDEX;
 560         hammer_off_t undo_limit;
 561         hammer_blockmap_t blockmap;
 562         hammer_off_t scan;
 563         int n;
 564         int limit_index;
 565
 566         /*
 567          * Size the undo buffer in multiples of HAMMER_LARGEBLOCK_SIZE,
 568          * up to HAMMER_UNDO_LAYER2 large blocks.  Size to approximately
 569          * 0.1% of the disk.
 570          *
 571          * The minimum UNDO fifo size is 100MB.
 572          */
 573         undo_limit = UndoBufferSize;
 574         if (undo_limit == 0) {
 575                 undo_limit = (ondisk->vol_buf_end - ondisk->vol_buf_beg) / 1000;
 576                 if (undo_limit < 100*1024*1024)
 577                         undo_limit = 100*1024*1024;
 578         }
 579         undo_limit = (undo_limit + HAMMER_LARGEBLOCK_MASK64) &
 580                      ~HAMMER_LARGEBLOCK_MASK64;
 581         if (undo_limit < HAMMER_LARGEBLOCK_SIZE)
 582                 undo_limit = HAMMER_LARGEBLOCK_SIZE;
 583         if (undo_limit > HAMMER_LARGEBLOCK_SIZE * HAMMER_UNDO_LAYER2)
 584                 undo_limit = HAMMER_LARGEBLOCK_SIZE * HAMMER_UNDO_LAYER2;
 585         UndoBufferSize = undo_limit;
 586
 587         blockmap = &ondisk->vol0_blockmap[undo_zone];
 588         bzero(blockmap, sizeof(*blockmap));
 589         blockmap->phys_offset = HAMMER_BLOCKMAP_UNAVAIL;
 590         blockmap->first_offset = HAMMER_ZONE_ENCODE(undo_zone, 0);
 591         blockmap->next_offset = blockmap->first_offset;
 592         blockmap->alloc_offset = HAMMER_ZONE_ENCODE(undo_zone, undo_limit);
 593         blockmap->entry_crc = crc32(blockmap, HAMMER_BLOCKMAP_CRCSIZE);
 594
 595         n = 0;
 596         scan = blockmap->next_offset;
 597         limit_index = undo_limit / HAMMER_LARGEBLOCK_SIZE;
 598
 599         assert(limit_index <= HAMMER_UNDO_LAYER2);
 600
 601         for (n = 0; n < limit_index; ++n) {
 602                 ondisk->vol0_undo_array[n] = alloc_bigblock(NULL,
 603                                                         HAMMER_ZONE_UNDO_INDEX);
 604                 scan += HAMMER_LARGEBLOCK_SIZE;
 605         }
 606         while (n < HAMMER_UNDO_LAYER2) {
 607                 ondisk->vol0_undo_array[n] = HAMMER_BLOCKMAP_UNAVAIL;
 608                 ++n;
 609         }
 610 }
 611
 612 /*
 613  * Format a new blockmap.  This is mostly a degenerate case because
 614  * all allocations are now actually done from the freemap.
 615  */
 616 void
 617 format_blockmap(hammer_blockmap_t blockmap, hammer_off_t zone_base)
 618 {
 619         blockmap->phys_offset = 0;
 620         blockmap->alloc_offset = zone_base | HAMMER_VOL_ENCODE(255) |
 621                                  HAMMER_SHORT_OFF_ENCODE(-1);
 622         blockmap->first_offset = zone_base;
 623         blockmap->next_offset = zone_base;
 624         blockmap->entry_crc = crc32(blockmap, HAMMER_BLOCKMAP_CRCSIZE);
 625 }
 626
 627 /*
 628  * Allocate a chunk of data out of a blockmap.  This is a simplified
 629  * version which uses next_offset as a simple allocation iterator.
 630  */
 631 static
 632 void *
 633 alloc_blockmap(int zone, int bytes, hammer_off_t *result_offp,
 634                struct buffer_info **bufferp)
 635 {
 636         struct buffer_info *buffer1 = NULL;
 637         struct buffer_info *buffer2 = NULL;
 638         struct volume_info *volume;
 639         hammer_blockmap_t blockmap;
 640         hammer_blockmap_t freemap;
 641         struct hammer_blockmap_layer1 *layer1;
 642         struct hammer_blockmap_layer2 *layer2;
 643         hammer_off_t layer1_offset;
 644         hammer_off_t layer2_offset;
 645         hammer_off_t zone2_offset;
 646         void *ptr;
 647
 648         volume = get_volume(RootVolNo);
 649
 650         blockmap = &volume->ondisk->vol0_blockmap[zone];
 651         freemap = &volume->ondisk->vol0_blockmap[HAMMER_ZONE_FREEMAP_INDEX];
 652
 653         /*
 654          * Alignment and buffer-boundary issues.  If the allocation would
 655          * cross a buffer boundary we have to skip to the next buffer.
 656          */
 657         bytes = (bytes + 15) & ~15;
 658
 659 again:
 660         if ((blockmap->next_offset ^ (blockmap->next_offset + bytes - 1)) &
 661             ~HAMMER_BUFMASK64) {
 662                 volume->cache.modified = 1;
 663                 blockmap->next_offset = (blockmap->next_offset + bytes) &
 664                                         ~HAMMER_BUFMASK64;
 665         }
 666
 667         /*
 668          * Dive layer 1.  For now we can't allocate data outside of volume 0.
 669          */
 670         layer1_offset = freemap->phys_offset +
 671                         HAMMER_BLOCKMAP_LAYER1_OFFSET(blockmap->next_offset);
 672
 673         layer1 = get_buffer_data(layer1_offset, &buffer1, 0);
 674
 675         if (layer1->phys_offset == HAMMER_BLOCKMAP_UNAVAIL) {
 676                 fprintf(stderr, "alloc_blockmap: ran out of space!\n");
 677                 exit(1);
 678         }
 679
 680         /*
 681          * Dive layer 2
 682          */
 683         layer2_offset = layer1->phys_offset +
 684                         HAMMER_BLOCKMAP_LAYER2_OFFSET(blockmap->next_offset);
 685
 686         layer2 = get_buffer_data(layer2_offset, &buffer2, 0);
 687
 688         if (layer2->zone == HAMMER_ZONE_UNAVAIL_INDEX) {
 689                 fprintf(stderr, "alloc_blockmap: ran out of space!\n");
 690                 exit(1);
 691         }
 692
 693         /*
 694          * If we are entering a new bigblock assign ownership to our
 695          * zone.  If the bigblock is owned by another zone skip it.
 696          */
 697         if (layer2->zone == 0) {
 698                 --layer1->blocks_free;
 699                 layer2->zone = zone;
 700                 assert(layer2->bytes_free == HAMMER_LARGEBLOCK_SIZE);
 701                 assert(layer2->append_off == 0);
 702         }
 703         if (layer2->zone != zone) {
 704                 blockmap->next_offset = (blockmap->next_offset + HAMMER_LARGEBLOCK_SIZE) & ~HAMMER_LARGEBLOCK_MASK64;
 705                 goto again;
 706         }
 707
 708         buffer1->cache.modified = 1;
 709         buffer2->cache.modified = 1;
 710         volume->cache.modified = 1;
 711         assert(layer2->append_off ==
 712                (blockmap->next_offset & HAMMER_LARGEBLOCK_MASK));
 713         layer2->bytes_free -= bytes;
 714         *result_offp = blockmap->next_offset;
 715         blockmap->next_offset += bytes;
 716         layer2->append_off = (int)blockmap->next_offset &
 717                               HAMMER_LARGEBLOCK_MASK;
 718
 719         layer1->layer1_crc = crc32(layer1, HAMMER_LAYER1_CRCSIZE);
 720         layer2->entry_crc = crc32(layer2, HAMMER_LAYER2_CRCSIZE);
 721
 722         zone2_offset = (*result_offp & ~HAMMER_OFF_ZONE_MASK) |
 723                         HAMMER_ZONE_ENCODE(zone, 0);
 724
 725         ptr = get_buffer_data(zone2_offset, bufferp, 0);
 726         (*bufferp)->cache.modified = 1;
 727
 728         if (buffer1)
 729                 rel_buffer(buffer1);
 730         if (buffer2)
 731                 rel_buffer(buffer2);
 732
 733         rel_volume(volume);
 734         return(ptr);
 735 }
 736
 737 /*
 738  * Flush various tracking structures to disk
 739  */
 740
 741 /*
 742  * Flush various tracking structures to disk
 743  */
 744 void
 745 flush_all_volumes(void)
 746 {
 747         struct volume_info *vol;
 748
 749         TAILQ_FOREACH(vol, &VolList, entry)
 750                 flush_volume(vol);
 751 }
 752
 753 void
 754 flush_volume(struct volume_info *volume)
 755 {
 756         struct buffer_info *buffer;
 757         int i;
 758
 759         for (i = 0; i < HAMMER_BUFLISTS; ++i) {
 760                 TAILQ_FOREACH(buffer, &volume->buffer_lists[i], entry)
 761                         flush_buffer(buffer);
 762         }
 763         writehammerbuf(volume, volume->ondisk, 0);
 764         volume->cache.modified = 0;
 765 }
 766
 767 void
 768 flush_buffer(struct buffer_info *buffer)
 769 {
 770         writehammerbuf(buffer->volume, buffer->ondisk, buffer->buf_disk_offset);
 771         buffer->cache.modified = 0;
 772 }
 773
 774 #if 0
 775 /*
 776  * Generic buffer initialization
 777  */
 778 static void
 779 init_fifo_head(hammer_fifo_head_t head, u_int16_t hdr_type)
 780 {
 781         head->hdr_signature = HAMMER_HEAD_SIGNATURE;
 782         head->hdr_type = hdr_type;
 783         head->hdr_size = 0;
 784         head->hdr_crc = 0;
 785         head->hdr_seq = 0;
 786 }
 787
 788 #endif
 789
 790 #if 0
 791 /*
 792  * Core I/O operations
 793  */
 794 static void
 795 readhammerbuf(struct volume_info *vol, void *data, int64_t offset)
 796 {
 797         ssize_t n;
 798
 799         n = pread(vol->fd, data, HAMMER_BUFSIZE, offset);
 800         if (n != HAMMER_BUFSIZE)
 801                 err(1, "Read volume %d (%s)", vol->vol_no, vol->name);
 802 }
 803
 804 #endif
 805
 806 static void
 807 writehammerbuf(struct volume_info *vol, const void *data, int64_t offset)
 808 {
 809         ssize_t n;
 810
 811         n = pwrite(vol->fd, data, HAMMER_BUFSIZE, offset);
 812         if (n != HAMMER_BUFSIZE)
 813                 err(1, "Write volume %d (%s)", vol->vol_no, vol->name);
 814 }
 815
 816 void
 817 panic(const char *ctl, ...)
 818 {
 819         va_list va;
 820
 821         va_start(va, ctl);
 822         vfprintf(stderr, ctl, va);
 823         va_end(va);
 824         fprintf(stderr, "\n");
 825         exit(1);
 826 }
 827