bitmap.c

   1 /*
   2  * mdadm - manage Linux "md" devices aka RAID arrays.
   3  *
   4  * Copyright (C) 2004 Paul Clements, SteelEye Technology, Inc.
   5  *
   6  *    This program is free software; you can redistribute it and/or modify
   7  *    it under the terms of the GNU General Public License as published by
   8  *    the Free Software Foundation; either version 2 of the License, or
   9  *    (at your option) any later version.
  10  *
  11  *    This program is distributed in the hope that it will be useful,
  12  *    but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  *    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  14  *    GNU General Public License for more details.
  15  *
  16  *    You should have received a copy of the GNU General Public License
  17  *    along with this program; if not, write to the Free Software
  18  *    Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  19  */
  20
  21 #include "mdadm.h"
  22
  23 static inline void sb_le_to_cpu(bitmap_super_t *sb)
  24 {
  25         sb->magic = __le32_to_cpu(sb->magic);
  26         sb->version = __le32_to_cpu(sb->version);
  27         /* uuid gets no translation */
  28         sb->events = __le64_to_cpu(sb->events);
  29         sb->events_cleared = __le64_to_cpu(sb->events_cleared);
  30         sb->state = __le32_to_cpu(sb->state);
  31         sb->chunksize = __le32_to_cpu(sb->chunksize);
  32         sb->daemon_sleep = __le32_to_cpu(sb->daemon_sleep);
  33         sb->sync_size = __le64_to_cpu(sb->sync_size);
  34         sb->write_behind = __le32_to_cpu(sb->write_behind);
  35         sb->nodes = __le32_to_cpu(sb->nodes);
  36         sb->sectors_reserved = __le32_to_cpu(sb->sectors_reserved);
  37 }
  38
  39 static inline void sb_cpu_to_le(bitmap_super_t *sb)
  40 {
  41         sb_le_to_cpu(sb); /* these are really the same thing */
  42 }
  43
  44 mapping_t bitmap_states[] = {
  45         { "OK", 0 },
  46         { "Out of date", 2 },
  47         { NULL, -1 }
  48 };
  49
  50 static const char *bitmap_state(int state_num)
  51 {
  52         char *state = map_num(bitmap_states, state_num);
  53         return state ? state : "Unknown";
  54 }
  55
  56 static const char *human_chunksize(unsigned long bytes)
  57 {
  58         static char buf[16];
  59         char *suffixes[] = { "B", "KB", "MB", "GB", "TB", NULL };
  60         int i = 0;
  61
  62         while (bytes >> 10) {
  63                 bytes >>= 10;
  64                 i++;
  65         }
  66
  67         snprintf(buf, sizeof(buf), "%lu %s", bytes, suffixes[i]);
  68
  69         return buf;
  70 }
  71
  72 typedef struct bitmap_info_s {
  73         bitmap_super_t sb;
  74         unsigned long long total_bits;
  75         unsigned long long dirty_bits;
  76 } bitmap_info_t;
  77
  78 /* count the dirty bits in the first num_bits of byte */
  79 static inline int count_dirty_bits_byte(char byte, int num_bits)
  80 {
  81         int num = 0;
  82
  83         switch (num_bits) { /* fall through... */
  84                 case 8: if (byte & 128) num++;
  85                 case 7: if (byte &  64) num++;
  86                 case 6: if (byte &  32) num++;
  87                 case 5: if (byte &  16) num++;
  88                 case 4: if (byte &   8) num++;
  89                 case 3: if (byte &   4) num++;
  90                 case 2: if (byte &   2) num++;
  91                 case 1: if (byte &   1) num++;
  92                 default: break;
  93         }
  94
  95         return num;
  96 }
  97
  98 static int count_dirty_bits(char *buf, int num_bits)
  99 {
 100         int i, num = 0;
 101
 102         for (i = 0; i < num_bits / 8; i++)
 103                 num += count_dirty_bits_byte(buf[i], 8);
 104
 105         if (num_bits % 8) /* not an even byte boundary */
 106                 num += count_dirty_bits_byte(buf[i], num_bits % 8);
 107
 108         return num;
 109 }
 110
 111 static bitmap_info_t *bitmap_fd_read(int fd, int brief)
 112 {
 113         /* Note: fd might be open O_DIRECT, so we must be
 114          * careful to align reads properly
 115          */
 116         unsigned long long total_bits = 0, read_bits = 0, dirty_bits = 0;
 117         bitmap_info_t *info;
 118         void *buf;
 119         unsigned int n, skip;
 120
 121         if (posix_memalign(&buf, 4096, 8192) != 0) {
 122                 pr_err("failed to allocate 8192 bytes\n");
 123                 return NULL;
 124         }
 125         n = read(fd, buf, 8192);
 126
 127         info = xmalloc(sizeof(*info));
 128
 129         if (n < sizeof(info->sb)) {
 130                 pr_err("failed to read superblock of bitmap file: %s\n", strerror(errno));
 131                 free(info);
 132                 free(buf);
 133                 return NULL;
 134         }
 135         memcpy(&info->sb, buf, sizeof(info->sb));
 136         skip = sizeof(info->sb);
 137
 138         sb_le_to_cpu(&info->sb); /* convert superblock to CPU byte ordering */
 139
 140         if (brief || info->sb.sync_size == 0 || info->sb.chunksize == 0)
 141                 goto out;
 142
 143         /* read the rest of the file counting total bits and dirty bits --
 144          * we stop when either:
 145          * 1) we hit EOF, in which case we assume the rest of the bits (if any)
 146          *    are dirty
 147          * 2) we've read the full bitmap, in which case we ignore any trailing
 148          *    data in the file
 149          */
 150         total_bits = bitmap_bits(info->sb.sync_size, info->sb.chunksize);
 151
 152         while(read_bits < total_bits) {
 153                 unsigned long long remaining = total_bits - read_bits;
 154
 155                 if (n == 0) {
 156                         n = read(fd, buf, 8192);
 157                         skip = 0;
 158                         if (n <= 0)
 159                                 break;
 160                 }
 161                 if (remaining > (n-skip) * 8) /* we want the full buffer */
 162                         remaining = (n-skip) * 8;
 163
 164                 dirty_bits += count_dirty_bits(buf+skip, remaining);
 165
 166                 read_bits += remaining;
 167                 n = 0;
 168         }
 169
 170         if (read_bits < total_bits) { /* file truncated... */
 171                 pr_err("WARNING: bitmap file is not large enough for array size %llu!\n\n",
 172                         (unsigned long long)info->sb.sync_size);
 173                 total_bits = read_bits;
 174         }
 175 out:
 176         free(buf);
 177         info->total_bits = total_bits;
 178         info->dirty_bits = dirty_bits;
 179         return info;
 180 }
 181
 182 static int
 183 bitmap_file_open(char *filename, struct supertype **stp, int node_num)
 184 {
 185         int fd;
 186         struct stat stb;
 187         struct supertype *st = *stp;
 188
 189         fd = open(filename, O_RDONLY|O_DIRECT);
 190         if (fd < 0) {
 191                 pr_err("failed to open bitmap file %s: %s\n",
 192                        filename, strerror(errno));
 193                 return -1;
 194         }
 195
 196         if (fstat(fd, &stb) < 0) {
 197                 pr_err("failed to determine bitmap file/device type: %s\n",
 198                         strerror(errno));
 199                 close(fd);
 200                 return -1;
 201         }
 202
 203         if ((stb.st_mode & S_IFMT) == S_IFBLK) {
 204                 /* block device, so we are probably after an internal bitmap */
 205                 if (!st)
 206                         st = guess_super(fd);
 207                 if (!st) {
 208                         /* just look at device... */
 209                         lseek(fd, 0, 0);
 210                 } else if (!st->ss->locate_bitmap) {
 211                         pr_err("No bitmap possible with %s metadata\n",
 212                                 st->ss->name);
 213                         close(fd);
 214                         return -1;
 215                 } else {
 216                         if (st->ss->locate_bitmap(st, fd, node_num)) {
 217                                 pr_err("%s doesn't have bitmap\n", filename);
 218                                 close(fd);
 219                                 fd = -1;
 220                         }
 221                 }
 222
 223                 *stp = st;
 224         }
 225
 226         return fd;
 227 }
 228
 229 static __u32 swapl(__u32 l)
 230 {
 231         char *c = (char*)&l;
 232         char t= c[0];
 233         c[0] = c[3];
 234         c[3] = t;
 235
 236         t = c[1];
 237         c[1] = c[2];
 238         c[2] = t;
 239         return l;
 240 }
 241 int ExamineBitmap(char *filename, int brief, struct supertype *st)
 242 {
 243         /*
 244          * Read the bitmap file and display its contents
 245          */
 246
 247         bitmap_super_t *sb;
 248         bitmap_info_t *info;
 249         int rv = 1;
 250         char buf[64];
 251         int swap;
 252         int fd, i;
 253         __u32 uuid32[4];
 254
 255         fd = bitmap_file_open(filename, &st, 0);
 256         if (fd < 0)
 257                 return rv;
 258
 259         info = bitmap_fd_read(fd, brief);
 260         if (!info)
 261                 return rv;
 262         sb = &info->sb;
 263         if (sb->magic != BITMAP_MAGIC && md_get_version(fd) > 0) {
 264                 pr_err("This is an md array.  To view a bitmap you need to examine\n");
 265                 pr_err("a member device, not the array.\n");
 266                 pr_err("Reporting bitmap that would be used if this array were used\n");
 267                 pr_err("as a member of some other array\n");
 268         }
 269         close(fd);
 270         printf("        Filename : %s\n", filename);
 271         printf("           Magic : %08x\n", sb->magic);
 272         if (sb->magic != BITMAP_MAGIC) {
 273                 pr_err("invalid bitmap magic 0x%x, the bitmap file appears\n",
 274                        sb->magic);
 275                 pr_err("to be corrupted or missing.\n");
 276         }
 277         printf("         Version : %d\n", sb->version);
 278         if (sb->version < BITMAP_MAJOR_LO ||
 279             sb->version > BITMAP_MAJOR_CLUSTERED) {
 280                 pr_err("unknown bitmap version %d, either the bitmap file\n",
 281                        sb->version);
 282                 pr_err("is corrupted or you need to upgrade your tools\n");
 283                 goto free_info;
 284         }
 285
 286         rv = 0;
 287         if (st)
 288                 swap = st->ss->swapuuid;
 289         else
 290 #if __BYTE_ORDER == BIG_ENDIAN
 291                 swap = 0;
 292 #else
 293                 swap = 1;
 294 #endif
 295         memcpy(uuid32, sb->uuid, 16);
 296         if (swap)
 297                 printf("            UUID : %08x:%08x:%08x:%08x\n",
 298                        swapl(uuid32[0]),
 299                        swapl(uuid32[1]),
 300                        swapl(uuid32[2]),
 301                        swapl(uuid32[3]));
 302         else
 303                 printf("            UUID : %08x:%08x:%08x:%08x\n",
 304                        uuid32[0],
 305                        uuid32[1],
 306                        uuid32[2],
 307                        uuid32[3]);
 308
 309         if (sb->nodes == 0) {
 310                 printf("          Events : %llu\n", (unsigned long long)sb->events);
 311                 printf("  Events Cleared : %llu\n", (unsigned long long)sb->events_cleared);
 312                 printf("           State : %s\n", bitmap_state(sb->state));
 313
 314         }
 315
 316         printf("       Chunksize : %s\n", human_chunksize(sb->chunksize));
 317         printf("          Daemon : %ds flush period\n", sb->daemon_sleep);
 318         if (sb->write_behind)
 319                 sprintf(buf, "Allow write behind, max %d", sb->write_behind);
 320         else
 321                 sprintf(buf, "Normal");
 322         printf("      Write Mode : %s\n", buf);
 323         printf("       Sync Size : %llu%s\n", (unsigned long long)sb->sync_size/2,
 324                                         human_size(sb->sync_size * 512));
 325
 326         if (sb->nodes == 0) {
 327                 if (brief)
 328                         goto free_info;
 329                 printf("          Bitmap : %llu bits (chunks), %llu dirty (%2.1f%%)\n",
 330                        info->total_bits, info->dirty_bits,
 331                        100.0 * info->dirty_bits / (info->total_bits?:1));
 332         } else {
 333                 printf("   Cluster nodes : %d\n", sb->nodes);
 334                 printf("    Cluster name : %-64s\n", sb->cluster_name);
 335                 for (i = 0; i < (int)sb->nodes; i++) {
 336                         st = NULL;
 337                         free(info);
 338                         fd = bitmap_file_open(filename, &st, i);
 339                         if (fd < 0) {
 340                                 printf("   Unable to open bitmap file on node: %i\n", i);
 341
 342                                 continue;
 343                         }
 344                         info = bitmap_fd_read(fd, brief);
 345                         if (!info) {
 346                                 close(fd);
 347                                 printf("   Unable to read bitmap on node: %i\n", i);
 348                                 continue;
 349                         }
 350                         sb = &info->sb;
 351                         if (sb->magic != BITMAP_MAGIC)
 352                                 pr_err("invalid bitmap magic 0x%x, the bitmap file appears to be corrupted\n", sb->magic);
 353
 354                         printf("       Node Slot : %d\n", i);
 355                         printf("          Events : %llu\n",
 356                                (unsigned long long)sb->events);
 357                         printf("  Events Cleared : %llu\n",
 358                                (unsigned long long)sb->events_cleared);
 359                         printf("           State : %s\n", bitmap_state(sb->state));
 360                         if (brief)
 361                                 continue;
 362                         printf("          Bitmap : %llu bits (chunks), %llu dirty (%2.1f%%)\n",
 363                                info->total_bits, info->dirty_bits,
 364                                100.0 * info->dirty_bits / (info->total_bits?:1));
 365                          close(fd);
 366                 }
 367         }
 368
 369 free_info:
 370         free(info);
 371         return rv;
 372 }
 373
 374 int CreateBitmap(char *filename, int force, char uuid[16],
 375                  unsigned long chunksize, unsigned long daemon_sleep,
 376                  unsigned long write_behind,
 377                  unsigned long long array_size /* sectors */,
 378                  int major)
 379 {
 380         /*
 381          * Create a bitmap file with a superblock and (optionally) a full bitmap
 382          */
 383
 384         FILE *fp;
 385         int rv = 1;
 386         char block[512];
 387         bitmap_super_t sb;
 388         long long bytes, filesize;
 389
 390         if (!force && access(filename, F_OK) == 0) {
 391                 pr_err("bitmap file %s already exists, use --force to overwrite\n", filename);
 392                 return rv;
 393         }
 394
 395         fp = fopen(filename, "w");
 396         if (fp == NULL) {
 397                 pr_err("failed to open bitmap file %s: %s\n",
 398                         filename, strerror(errno));
 399                 return rv;
 400         }
 401
 402         if (chunksize == UnSet) {
 403                 /* We don't want more than 2^21 chunks, as 2^11 fill up one
 404                  * 4K page (2 bytes per chunk), and 2^10 address of those
 405                  * fill up a 4K indexing page.  2^20 might be safer, especially
 406                  * on 64bit hosts, so use that.
 407                  */
 408                 chunksize = DEFAULT_BITMAP_CHUNK;
 409                 /* <<20 for 2^20 chunks, >>9 to convert bytes to sectors */
 410                 while (array_size > ((unsigned long long)chunksize << (20-9)))
 411                         chunksize <<= 1;
 412         }
 413
 414         memset(&sb, 0, sizeof(sb));
 415         sb.magic = BITMAP_MAGIC;
 416         sb.version = major;
 417         if (uuid != NULL)
 418                 memcpy(sb.uuid, uuid, 16);
 419         sb.chunksize = chunksize;
 420         sb.daemon_sleep = daemon_sleep;
 421         sb.write_behind = write_behind;
 422         sb.sync_size = array_size;
 423
 424         sb_cpu_to_le(&sb); /* convert to on-disk byte ordering */
 425
 426         if (fwrite(&sb, sizeof(sb), 1, fp) != 1) {
 427                 pr_err("failed to write superblock to bitmap file %s: %s\n", filename, strerror(errno));
 428                 goto out;
 429         }
 430
 431         /* calculate the size of the bitmap and write it to disk */
 432         bytes = (bitmap_bits(array_size, chunksize) + 7) / 8;
 433         if (!bytes) {
 434                 rv = 0;
 435                 goto out;
 436         }
 437
 438         filesize = bytes + sizeof(sb);
 439
 440         memset(block, 0xff, sizeof(block));
 441
 442         while (bytes > 0) {
 443                 if (fwrite(block, sizeof(block), 1, fp) != 1) {
 444                         pr_err("failed to write bitmap file %s: %s\n", filename, strerror(errno));
 445                         goto out;
 446                 }
 447                 bytes -= sizeof(block);
 448         }
 449
 450         rv = 0;
 451         fflush(fp);
 452         /* make the file be the right size (well, to the nearest byte) */
 453         if (ftruncate(fileno(fp), filesize))
 454                 perror("ftrunace");
 455 out:
 456         fclose(fp);
 457         if (rv)
 458                 unlink(filename); /* possibly corrupted, better get rid of it */
 459         return rv;
 460 }
 461
 462 int bitmap_update_uuid(int fd, int *uuid, int swap)
 463 {
 464         struct bitmap_super_s bm;
 465         if (lseek(fd, 0, 0) != 0)
 466                 return 1;
 467         if (read(fd, &bm, sizeof(bm)) != sizeof(bm))
 468                 return 1;
 469         if (bm.magic != __cpu_to_le32(BITMAP_MAGIC))
 470                 return 1;
 471         copy_uuid(bm.uuid, uuid, swap);
 472         if (lseek(fd, 0, 0) != 0)
 473                 return 2;
 474         if (write(fd, &bm, sizeof(bm)) != sizeof(bm)) {
 475                 lseek(fd, 0, 0);
 476                 return 2;
 477         }
 478         lseek(fd, 0, 0);
 479         return 0;
 480 }