sbin/newfs_hammer/newfs_hammer.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/newfs_hammer/newfs_hammer.c,v 1.29 2008/06/08 17:19:09 dillon Exp $
  35  */
  36
  37 #include "newfs_hammer.h"
  38
  39 static int64_t getsize(const char *str, int64_t minval, int64_t maxval, int pw);
  40 static const char *sizetostr(off_t size);
  41 static void check_volume(struct volume_info *vol);
  42 static void format_volume(struct volume_info *vol, int nvols,const char *label,
  43                         off_t total_size);
  44 static hammer_off_t format_root(void);
  45 static void usage(void);
  46
  47 int
  48 main(int ac, char **av)
  49 {
  50         u_int32_t status;
  51         off_t total;
  52         int ch;
  53         int i;
  54         const char *label = NULL;
  55         struct volume_info *vol;
  56         hammer_off_t vol0_zone_limit;
  57
  58         /*
  59          * Sanity check basic filesystem structures.  No cookies for us
  60          * if it gets broken!
  61          */
  62         assert(sizeof(struct hammer_volume_ondisk) <= HAMMER_BUFSIZE);
  63         assert(sizeof(struct hammer_blockmap_layer1) == 32);
  64         assert(sizeof(struct hammer_blockmap_layer2) == 16);
  65
  66         /*
  67          * Generate a filesysem id and lookup the filesystem type
  68          */
  69         uuidgen(&Hammer_FSId, 1);
  70         uuid_name_lookup(&Hammer_FSType, "DragonFly HAMMER", &status);
  71         if (status != uuid_s_ok) {
  72                 errx(1, "uuids file does not have the DragonFly "
  73                         "HAMMER filesystem type");
  74         }
  75
  76         /*
  77          * Parse arguments
  78          */
  79         while ((ch = getopt(ac, av, "L:b:m:u:")) != -1) {
  80                 switch(ch) {
  81                 case 'L':
  82                         label = optarg;
  83                         break;
  84                 case 'b':
  85                         BootAreaSize = getsize(optarg,
  86                                          HAMMER_BUFSIZE,
  87                                          HAMMER_BOOT_MAXBYTES, 2);
  88                         break;
  89                 case 'm':
  90                         MemAreaSize = getsize(optarg,
  91                                          HAMMER_BUFSIZE,
  92                                          HAMMER_MEM_MAXBYTES, 2);
  93                         break;
  94                 case 'u':
  95                         UndoBufferSize = getsize(optarg,
  96                                          HAMMER_LARGEBLOCK_SIZE,
  97                                          HAMMER_LARGEBLOCK_SIZE *
  98                                          HAMMER_UNDO_LAYER2, 2);
  99                         break;
 100                 default:
 101                         usage();
 102                         break;
 103                 }
 104         }
 105
 106         if (label == NULL) {
 107                 fprintf(stderr,
 108                         "newfs_hammer: A filesystem label must be specified\n");
 109                 exit(1);
 110         }
 111
 112         /*
 113          * Collect volume information
 114          */
 115         ac -= optind;
 116         av += optind;
 117         NumVolumes = ac;
 118         RootVolNo = 0;
 119
 120         total = 0;
 121         for (i = 0; i < NumVolumes; ++i) {
 122                 vol = setup_volume(i, av[i], 1, O_RDWR);
 123
 124                 /*
 125                  * Load up information on the volume and initialize
 126                  * its remaining fields.
 127                  */
 128                 check_volume(vol);
 129                 total += vol->size;
 130         }
 131
 132         /*
 133          * Calculate defaults for the boot and memory area sizes.
 134          */
 135         if (BootAreaSize == 0) {
 136                 BootAreaSize = HAMMER_BOOT_NOMBYTES;
 137                 while (BootAreaSize > total / NumVolumes / 256)
 138                         BootAreaSize >>= 1;
 139                 if (BootAreaSize < HAMMER_BOOT_MINBYTES)
 140                         BootAreaSize = 0;
 141         } else if (BootAreaSize < HAMMER_BOOT_MINBYTES) {
 142                 BootAreaSize = HAMMER_BOOT_MINBYTES;
 143         }
 144         if (MemAreaSize == 0) {
 145                 MemAreaSize = HAMMER_MEM_NOMBYTES;
 146                 while (MemAreaSize > total / NumVolumes / 256)
 147                         MemAreaSize >>= 1;
 148                 if (MemAreaSize < HAMMER_MEM_MINBYTES)
 149                         MemAreaSize = 0;
 150         } else if (MemAreaSize < HAMMER_MEM_MINBYTES) {
 151                 MemAreaSize = HAMMER_MEM_MINBYTES;
 152         }
 153
 154         /*
 155          * Format the volumes.  Format the root volume first so we can
 156          * bootstrap the freemap.
 157          */
 158         format_volume(get_volume(RootVolNo), NumVolumes, label, total);
 159         for (i = 0; i < NumVolumes; ++i) {
 160                 if (i != RootVolNo)
 161                         format_volume(get_volume(i), NumVolumes, label, total);
 162         }
 163
 164         /*
 165          * Pre-size the blockmap layer1/layer2 infrastructure to the zone
 166          * limit.  If we do this the filesystem does not have to allocate
 167          * new layer2 blocks which reduces the chances of the reblocker
 168          * having to fallback to an extremely inefficient algorithm.
 169          */
 170         vol = get_volume(RootVolNo);
 171         vol0_zone_limit = vol->ondisk->vol0_zone_limit;
 172         printf("Initializing B-Tree blockmap\n");
 173         presize_blockmap(&vol->ondisk->vol0_blockmap[HAMMER_ZONE_BTREE_INDEX],
 174                          HAMMER_ZONE_BTREE, vol0_zone_limit);
 175         printf("Initializing Large-Data blockmap\n");
 176         presize_blockmap(&vol->ondisk->vol0_blockmap[HAMMER_ZONE_LARGE_DATA_INDEX],
 177                          HAMMER_ZONE_LARGE_DATA, vol0_zone_limit);
 178         printf("Initializing Small-Data blockmap\n");
 179         presize_blockmap(&vol->ondisk->vol0_blockmap[HAMMER_ZONE_SMALL_DATA_INDEX],
 180                          HAMMER_ZONE_SMALL_DATA, vol0_zone_limit);
 181         vol->ondisk->vol0_stat_bigblocks = vol->ondisk->vol0_stat_freebigblocks;
 182         vol->cache.modified = 1;
 183
 184         printf("---------------------------------------------\n");
 185         printf("%d volume%s total size %s\n",
 186                 NumVolumes, (NumVolumes == 1 ? "" : "s"), sizetostr(total));
 187         printf("boot-area-size:      %s\n", sizetostr(BootAreaSize));
 188         printf("memory-log-size:     %s\n", sizetostr(MemAreaSize));
 189         printf("undo-buffer-size:    %s\n", sizetostr(UndoBufferSize));
 190
 191         printf("zone-limit:          %s\n", sizetostr(vol0_zone_limit));
 192         printf("total-pre-allocated: %s\n",
 193                 sizetostr(vol->vol_free_off & HAMMER_OFF_SHORT_MASK));
 194         printf("\n");
 195
 196         flush_all_volumes();
 197         return(0);
 198 }
 199
 200 static
 201 void
 202 usage(void)
 203 {
 204         fprintf(stderr, "newfs_hammer vol0 [vol1 ...]\n");
 205         exit(1);
 206 }
 207
 208 /*
 209  * Convert the size in bytes to a human readable string.
 210  */
 211 static
 212 const char *
 213 sizetostr(off_t size)
 214 {
 215         static char buf[32];
 216
 217         if (size < 1024 / 2) {
 218                 snprintf(buf, sizeof(buf), "%6.2f", (double)size);
 219         } else if (size < 1024 * 1024 / 2) {
 220                 snprintf(buf, sizeof(buf), "%6.2fKB",
 221                         (double)size / 1024);
 222         } else if (size < 1024 * 1024 * 1024LL / 2) {
 223                 snprintf(buf, sizeof(buf), "%6.2fMB",
 224                         (double)size / (1024 * 1024));
 225         } else if (size < 1024 * 1024 * 1024LL * 1024LL / 2) {
 226                 snprintf(buf, sizeof(buf), "%6.2fGB",
 227                         (double)size / (1024 * 1024 * 1024LL));
 228         } else {
 229                 snprintf(buf, sizeof(buf), "%6.2fTB",
 230                         (double)size / (1024 * 1024 * 1024LL * 1024LL));
 231         }
 232         return(buf);
 233 }
 234
 235 /*
 236  * Convert a string to a 64 bit signed integer with various requirements.
 237  */
 238 static int64_t
 239 getsize(const char *str, int64_t minval, int64_t maxval, int powerof2)
 240 {
 241         int64_t val;
 242         char *ptr;
 243
 244         val = strtoll(str, &ptr, 0);
 245         switch(*ptr) {
 246         case 't':
 247         case 'T':
 248                 val *= 1024;
 249                 /* fall through */
 250         case 'g':
 251         case 'G':
 252                 val *= 1024;
 253                 /* fall through */
 254         case 'm':
 255         case 'M':
 256                 val *= 1024;
 257                 /* fall through */
 258         case 'k':
 259         case 'K':
 260                 val *= 1024;
 261                 break;
 262         default:
 263                 errx(1, "Unknown suffix in number '%s'\n", str);
 264                 /* not reached */
 265         }
 266         if (ptr[1]) {
 267                 errx(1, "Unknown suffix in number '%s'\n", str);
 268                 /* not reached */
 269         }
 270         if (val < minval) {
 271                 errx(1, "Value too small: %s, min is %s\n",
 272                      str, sizetostr(minval));
 273                 /* not reached */
 274         }
 275         if (val > maxval) {
 276                 errx(1, "Value too large: %s, max is %s\n",
 277                      str, sizetostr(maxval));
 278                 /* not reached */
 279         }
 280         if ((powerof2 & 1) && (val ^ (val - 1)) != ((val << 1) - 1)) {
 281                 errx(1, "Value not power of 2: %s\n", str);
 282                 /* not reached */
 283         }
 284         if ((powerof2 & 2) && (val & HAMMER_BUFMASK)) {
 285                 errx(1, "Value not an integral multiple of %dK: %s",
 286                      HAMMER_BUFSIZE / 1024, str);
 287                 /* not reached */
 288         }
 289         return(val);
 290 }
 291
 292 /*
 293  * Generate a transaction id
 294  */
 295 static hammer_tid_t
 296 createtid(void)
 297 {
 298         static hammer_tid_t lasttid;
 299         struct timeval tv;
 300
 301         if (lasttid == 0) {
 302                 gettimeofday(&tv, NULL);
 303                 lasttid = tv.tv_sec * 1000000000LL +
 304                           tv.tv_usec * 1000LL;
 305         }
 306         return(lasttid++);
 307 }
 308
 309 /*
 310  * Check basic volume characteristics.  HAMMER filesystems use a minimum
 311  * of a 16KB filesystem buffer size.
 312  */
 313 static
 314 void
 315 check_volume(struct volume_info *vol)
 316 {
 317         struct partinfo pinfo;
 318         struct stat st;
 319
 320         /*
 321          * Get basic information about the volume
 322          */
 323         vol->fd = open(vol->name, O_RDWR);
 324         if (vol->fd < 0)
 325                 err(1, "Unable to open %s R+W", vol->name);
 326         if (ioctl(vol->fd, DIOCGPART, &pinfo) < 0) {
 327                 /*
 328                  * Allow the formatting of regular filews as HAMMER volumes
 329                  */
 330                 if (fstat(vol->fd, &st) < 0)
 331                         err(1, "Unable to stat %s", vol->name);
 332                 vol->size = st.st_size;
 333                 vol->type = "REGFILE";
 334         } else {
 335                 /*
 336                  * When formatting a block device as a HAMMER volume the
 337                  * sector size must be compatible.  HAMMER uses 16384 byte
 338                  * filesystem buffers.
 339                  */
 340                 if (pinfo.reserved_blocks) {
 341                         errx(1, "HAMMER cannot be placed in a partition "
 342                                 "which overlaps the disklabel or MBR");
 343                 }
 344                 if (pinfo.media_blksize > 16384 ||
 345                     16384 % pinfo.media_blksize) {
 346                         errx(1, "A media sector size of %d is not supported",
 347                              pinfo.media_blksize);
 348                 }
 349
 350                 vol->size = pinfo.media_size;
 351                 vol->type = "DEVICE";
 352         }
 353         printf("Volume %d %s %-15s size %s\n",
 354                vol->vol_no, vol->type, vol->name,
 355                sizetostr(vol->size));
 356
 357         /*
 358          * Reserve space for (future) header junk, setup our poor-man's
 359          * bigblock allocator.
 360          */
 361         vol->vol_alloc = HAMMER_BUFSIZE * 16;
 362 }
 363
 364 /*
 365  * Format a HAMMER volume.  Cluster 0 will be initially placed in volume 0.
 366  */
 367 static
 368 void
 369 format_volume(struct volume_info *vol, int nvols, const char *label,
 370               off_t total_size)
 371 {
 372         struct volume_info *root_vol;
 373         struct hammer_volume_ondisk *ondisk;
 374         int64_t freeblks;
 375
 376         /*
 377          * Initialize basic information in the on-disk volume structure.
 378          */
 379         ondisk = vol->ondisk;
 380
 381         ondisk->vol_fsid = Hammer_FSId;
 382         ondisk->vol_fstype = Hammer_FSType;
 383         snprintf(ondisk->vol_name, sizeof(ondisk->vol_name), "%s", label);
 384         ondisk->vol_no = vol->vol_no;
 385         ondisk->vol_count = nvols;
 386         ondisk->vol_version = 1;
 387
 388         ondisk->vol_bot_beg = vol->vol_alloc;
 389         vol->vol_alloc += BootAreaSize;
 390         ondisk->vol_mem_beg = vol->vol_alloc;
 391         vol->vol_alloc += MemAreaSize;
 392
 393         /*
 394          * The remaining area is the zone 2 buffer allocation area.  These
 395          * buffers
 396          */
 397         ondisk->vol_buf_beg = vol->vol_alloc;
 398         ondisk->vol_buf_end = vol->size & ~(int64_t)HAMMER_BUFMASK;
 399
 400         if (ondisk->vol_buf_end < ondisk->vol_buf_beg) {
 401                 errx(1, "volume %d %s is too small to hold the volume header",
 402                      vol->vol_no, vol->name);
 403         }
 404
 405         ondisk->vol_nblocks = (ondisk->vol_buf_end - ondisk->vol_buf_beg) /
 406                               HAMMER_BUFSIZE;
 407         ondisk->vol_blocksize = HAMMER_BUFSIZE;
 408
 409         ondisk->vol_rootvol = RootVolNo;
 410         ondisk->vol_signature = HAMMER_FSBUF_VOLUME;
 411
 412         vol->vol_free_off = HAMMER_ENCODE_RAW_BUFFER(vol->vol_no, 0);
 413         vol->vol_free_end = HAMMER_ENCODE_RAW_BUFFER(vol->vol_no, (ondisk->vol_buf_end - ondisk->vol_buf_beg) & ~HAMMER_LARGEBLOCK_MASK64);
 414
 415         /*
 416          * Format the root volume.
 417          */
 418         if (vol->vol_no == RootVolNo) {
 419                 hammer_off_t zone_limit;
 420
 421                 /*
 422                  * Starting TID
 423                  */
 424                 ondisk->vol0_next_tid = createtid();
 425
 426                 /*
 427                  * Set the default zone limit to 4x the size of the
 428                  * filesystem, allowing the filesystem to survive a
 429                  * worse-case of 75% fragmentation per zone.  The limit
 430                  * can be expanded at any time if the filesystem is
 431                  * made larger.
 432                  *
 433                  * We do not want to create a huge zone limit for tiny
 434                  * filesystems because the blockmap could wind up getting
 435                  * fragmented and eating a large chunk of the disk space.
 436                  */
 437                 zone_limit = (hammer_off_t)total_size * 4;
 438                 zone_limit = (zone_limit + HAMMER_BLOCKMAP_LAYER2_MASK) &
 439                              ~HAMMER_BLOCKMAP_LAYER2_MASK;
 440                 if (zone_limit < (hammer_off_t)total_size ||
 441                     zone_limit > HAMMER_ZONE_LIMIT) {
 442                         zone_limit = HAMMER_ZONE_LIMIT;
 443                 }
 444                 ondisk->vol0_zone_limit = zone_limit;
 445
 446                 format_freemap(vol,
 447                         &ondisk->vol0_blockmap[HAMMER_ZONE_FREEMAP_INDEX]);
 448
 449                 freeblks = initialize_freemap(vol);
 450                 ondisk->vol0_stat_freebigblocks = freeblks;
 451
 452                 format_blockmap(
 453                         &ondisk->vol0_blockmap[HAMMER_ZONE_BTREE_INDEX],
 454                         HAMMER_ZONE_BTREE);
 455 #if 0
 456                 format_blockmap(
 457                         &ondisk->vol0_blockmap[HAMMER_ZONE_RECORD_INDEX],
 458                         HAMMER_ZONE_RECORD);
 459 #endif
 460                 format_blockmap(
 461                         &ondisk->vol0_blockmap[HAMMER_ZONE_LARGE_DATA_INDEX],
 462                         HAMMER_ZONE_LARGE_DATA);
 463                 format_blockmap(
 464                         &ondisk->vol0_blockmap[HAMMER_ZONE_SMALL_DATA_INDEX],
 465                         HAMMER_ZONE_SMALL_DATA);
 466
 467                 format_undomap(ondisk);
 468
 469                 ondisk->vol0_btree_root = format_root();
 470                 ++ondisk->vol0_stat_inodes;     /* root inode */
 471         } else {
 472                 freeblks = initialize_freemap(vol);
 473                 root_vol = get_volume(RootVolNo);
 474                 root_vol->cache.modified = 1;
 475                 root_vol->ondisk->vol0_stat_freebigblocks += freeblks;
 476                 root_vol->ondisk->vol0_stat_bigblocks += freeblks;
 477                 rel_volume(root_vol);
 478         }
 479 }
 480
 481 /*
 482  * Format the root directory.
 483  */
 484 static
 485 hammer_off_t
 486 format_root(void)
 487 {
 488         hammer_off_t btree_off;
 489         hammer_off_t data_off;
 490         hammer_tid_t create_tid;
 491         hammer_node_ondisk_t bnode;
 492         struct hammer_inode_data *idata;
 493         struct buffer_info *data_buffer = NULL;
 494         hammer_btree_elm_t elm;
 495
 496         bnode = alloc_btree_element(&btree_off);
 497         idata = alloc_data_element(&data_off, sizeof(*idata), &data_buffer);
 498         create_tid = createtid();
 499
 500         /*
 501          * Populate the inode data and inode record for the root directory.
 502          */
 503         idata->version = HAMMER_INODE_DATA_VERSION;
 504         idata->mode = 0755;
 505         idata->mtime = create_tid;
 506         idata->obj_type = HAMMER_OBJTYPE_DIRECTORY;
 507         idata->size = 0;
 508         idata->nlinks = 1;
 509
 510         /*
 511          * Create the root of the B-Tree.  The root is a leaf node so we
 512          * do not have to worry about boundary elements.
 513          */
 514         bnode->signature = HAMMER_BTREE_SIGNATURE_GOOD;
 515         bnode->count = 1;
 516         bnode->type = HAMMER_BTREE_TYPE_LEAF;
 517
 518         elm = &bnode->elms[0];
 519         elm->leaf.base.btype = HAMMER_BTREE_TYPE_RECORD;
 520         elm->leaf.base.localization = HAMMER_LOCALIZE_INODE;
 521         elm->leaf.base.obj_id = HAMMER_OBJID_ROOT;
 522         elm->leaf.base.key = 0;
 523         elm->leaf.base.create_tid = create_tid;
 524         elm->leaf.base.delete_tid = 0;
 525         elm->leaf.base.rec_type = HAMMER_RECTYPE_INODE;
 526         elm->leaf.base.obj_type = HAMMER_OBJTYPE_DIRECTORY;
 527
 528         elm->leaf.atime = create_tid;
 529         elm->leaf.data_offset = data_off;
 530         elm->leaf.data_len = sizeof(*idata);
 531         elm->leaf.data_crc = crc32(idata, sizeof(*idata));
 532
 533         bnode->crc = crc32(&bnode->crc + 1, HAMMER_BTREE_CRCSIZE);
 534
 535         return(btree_off);
 536 }
 537