src/log-writes/log-writes.c

   1 #include <linux/fs.h>
   2 #include <sys/types.h>
   3 #include <sys/stat.h>
   4 #include <sys/ioctl.h>
   5 #include <fcntl.h>
   6 #include <stdio.h>
   7 #include <stdlib.h>
   8 #include <errno.h>
   9 #include <unistd.h>
  10 #include <string.h>
  11 #include "log-writes.h"
  12
  13 int log_writes_verbose = 0;
  14
  15 /*
  16  * @log: the log to free.
  17  *
  18  * This will close any open fd's the log has and free up its memory.
  19  */
  20 void log_free(struct log *log)
  21 {
  22         if (log->replayfd >= 0)
  23                 close(log->replayfd);
  24         if (log->logfd >= 0)
  25                 close(log->logfd);
  26         free(log);
  27 }
  28
  29 static int discard_range(struct log *log, u64 start, u64 len)
  30 {
  31         u64 range[2] = { start, len };
  32
  33         if (ioctl(log->replayfd, BLKDISCARD, &range) < 0) {
  34                 if (log_writes_verbose)
  35                         printf("replay device doesn't support discard, "
  36                                "switching to writing zeros\n");
  37                 log->flags |= LOG_DISCARD_NOT_SUPP;
  38         }
  39         return 0;
  40 }
  41
  42 static int zero_range(struct log *log, u64 start, u64 len)
  43 {
  44         u64 bufsize = len;
  45         ssize_t ret;
  46         char *buf = NULL;
  47
  48         if (log->max_zero_size < len) {
  49                 if (log_writes_verbose)
  50                         printf("discard len %llu larger than max %llu\n",
  51                                (unsigned long long)len,
  52                                (unsigned long long)log->max_zero_size);
  53                 return 0;
  54         }
  55
  56         while (!buf) {
  57                 buf = malloc(bufsize);
  58                 if (!buf)
  59                         bufsize >>= 1;
  60                 if (!bufsize) {
  61                         fprintf(stderr, "Couldn't allocate zero buffer");
  62                         return -1;
  63                 }
  64         }
  65
  66         memset(buf, 0, bufsize);
  67         while (len) {
  68                 ret = pwrite(log->replayfd, buf, bufsize, start);
  69                 if (ret != bufsize) {
  70                         fprintf(stderr, "Error zeroing file: %d\n", errno);
  71                         free(buf);
  72                         return -1;
  73                 }
  74                 len -= ret;
  75                 start += ret;
  76         }
  77         free(buf);
  78         return 0;
  79 }
  80
  81 /*
  82  * @log: the log we are replaying.
  83  * @entry: the discard entry.
  84  *
  85  * Discard the given length.  If the device supports discard we will call that
  86  * ioctl, otherwise we will write 0's to emulate discard.  If the discard size
  87  * is larger than log->max_zero_size then we will simply skip the zero'ing if
  88  * the drive doesn't support discard.
  89  */
  90 int log_discard(struct log *log, struct log_write_entry *entry)
  91 {
  92         u64 start = le64_to_cpu(entry->sector) * log->sectorsize;
  93         u64 size = le64_to_cpu(entry->nr_sectors) * log->sectorsize;
  94         u64 max_chunk = 1 * 1024 * 1024 * 1024;
  95
  96         if (log->flags & LOG_IGNORE_DISCARD)
  97                 return 0;
  98
  99         while (size) {
 100                 u64 len = size > max_chunk ? max_chunk : size;
 101                 int ret;
 102
 103                 /*
 104                  * Do this check first in case it is our first discard, that way
 105                  * if we return EOPNOTSUPP we will fall back to the 0 method
 106                  * automatically.
 107                  */
 108                 if (!(log->flags & LOG_DISCARD_NOT_SUPP))
 109                         ret = discard_range(log, start, len);
 110                 if (log->flags & LOG_DISCARD_NOT_SUPP)
 111                         ret = zero_range(log, start, len);
 112                 if (ret)
 113                         return -1;
 114                 size -= len;
 115                 start += len;
 116         }
 117         return 0;
 118 }
 119
 120 /*
 121  * @entry: entry to be replayed.
 122  *
 123  * @return: 1 if the entry is sane, 0 if it is invalid.
 124  *
 125  * Check if this is a sane log entry.
 126  */
 127 int log_entry_valid(struct log_write_entry *entry)
 128 {
 129         u64 flags = le64_to_cpu(entry->flags);
 130
 131         /* Suspect all zeroes entry */
 132         if (!flags && !entry->nr_sectors)
 133                 return 0;
 134         /* Suspect non zero padded entry */
 135         if (flags != LOG_MARK_FLAG && entry->data[0] != 0)
 136                 return 0;
 137         return 1;
 138 }
 139
 140 /*
 141  * @log: the log we are replaying.
 142  * @entry: where we put the entry.
 143  * @read_data: read the entry data as well, entry must be log->sectorsize sized
 144  * if this is set.
 145  *
 146  * @return: 0 if we replayed, 1 if we are at the end, -1 if there was an error.
 147  *
 148  * Replay the next entry in our log onto the replay device.
 149  */
 150 int log_replay_next_entry(struct log *log, struct log_write_entry *entry,
 151                           int read_data)
 152 {
 153         u64 size;
 154         u64 flags;
 155         size_t read_size = read_data ? log->sectorsize :
 156                 sizeof(struct log_write_entry);
 157         char *buf;
 158         ssize_t ret;
 159         off_t offset;
 160
 161         if (log->cur_entry >= log->nr_entries)
 162                 return 1;
 163
 164         ret = read(log->logfd, entry, read_size);
 165         if (ret != read_size) {
 166                 fprintf(stderr, "Error reading entry: %d\n", errno);
 167                 return -1;
 168         }
 169         if (!log_entry_valid(entry)) {
 170                 fprintf(stderr, "Malformed entry @%llu\n",
 171                                 log->cur_pos / log->sectorsize);
 172                 return -1;
 173         }
 174         log->cur_entry++;
 175
 176         size = le64_to_cpu(entry->nr_sectors) * log->sectorsize;
 177         if (read_size < log->sectorsize) {
 178                 log->cur_pos = lseek(log->logfd,
 179                         log->sectorsize - sizeof(struct log_write_entry), SEEK_CUR);
 180                 if (log->cur_pos == (off_t)-1) {
 181                         fprintf(stderr, "Error seeking in log: %d\n", errno);
 182                         return -1;
 183                 }
 184         } else {
 185                 log->cur_pos += read_size;
 186         }
 187
 188         if (log_writes_verbose) {
 189                 printf("replaying %d@%llu: sector %llu, size %llu, flags %llu\n",
 190                        (int)log->cur_entry - 1, log->cur_pos / log->sectorsize,
 191                        (unsigned long long)le64_to_cpu(entry->sector),
 192                        (unsigned long long)size,
 193                        (unsigned long long)le64_to_cpu(entry->flags));
 194         }
 195         if (!size)
 196                 return 0;
 197
 198         flags = le64_to_cpu(entry->flags);
 199         if (flags & LOG_DISCARD_FLAG)
 200                 return log_discard(log, entry);
 201
 202         buf = malloc(size);
 203         if (!buf) {
 204                 fprintf(stderr, "Error allocating buffer %llu entry %llu\n", (unsigned long long)size, (unsigned long long)log->cur_entry - 1);
 205                 return -1;
 206         }
 207
 208         ret = read(log->logfd, buf, size);
 209         if (ret != size) {
 210                 fprintf(stderr, "Error reading data: %d\n", errno);
 211                 free(buf);
 212                 return -1;
 213         }
 214         log->cur_pos += size;
 215
 216         offset = le64_to_cpu(entry->sector) * log->sectorsize;
 217         ret = pwrite(log->replayfd, buf, size, offset);
 218         free(buf);
 219         if (ret != size) {
 220                 fprintf(stderr, "Error writing data: %d\n", errno);
 221                 return -1;
 222         }
 223
 224         return 0;
 225 }
 226
 227 /*
 228  * @log: the log we are manipulating.
 229  * @entry_num: the entry we want.
 230  *
 231  * Seek to the given entry in the log, starting at 0 and ending at
 232  * log->nr_entries - 1.
 233  */
 234 int log_seek_entry(struct log *log, u64 entry_num)
 235 {
 236         u64 i = 0;
 237
 238         if (entry_num >= log->nr_entries) {
 239                 fprintf(stderr, "Invalid entry number\n");
 240                 return -1;
 241         }
 242
 243         /* Skip the first sector containing the log super block */
 244         log->cur_pos = lseek(log->logfd, log->sectorsize, SEEK_SET);
 245         if (log->cur_pos == (off_t)-1) {
 246                 fprintf(stderr, "Error seeking in file: %d\n", errno);
 247                 return -1;
 248         }
 249
 250         log->cur_entry = 0;
 251         for (i = 0; i < entry_num; i++) {
 252                 struct log_write_entry entry;
 253                 ssize_t ret;
 254                 off_t seek_size;
 255                 u64 flags;
 256
 257                 ret = read(log->logfd, &entry, sizeof(entry));
 258                 if (ret != sizeof(entry)) {
 259                         fprintf(stderr, "Error reading entry: %d\n", errno);
 260                         return -1;
 261                 }
 262                 if (!log_entry_valid(&entry)) {
 263                         fprintf(stderr, "Malformed entry @%llu\n",
 264                                         log->cur_pos / log->sectorsize);
 265                         return -1;
 266                 }
 267                 if (log_writes_verbose > 1)
 268                         printf("seek entry %d@%llu: %llu, size %llu, flags %llu\n",
 269                                (int)i, log->cur_pos / log->sectorsize,
 270                                (unsigned long long)le64_to_cpu(entry.sector),
 271                                (unsigned long long)le64_to_cpu(entry.nr_sectors),
 272                                (unsigned long long)le64_to_cpu(entry.flags));
 273                 flags = le64_to_cpu(entry.flags);
 274                 seek_size = log->sectorsize - sizeof(entry);
 275                 if (!(flags & LOG_DISCARD_FLAG))
 276                         seek_size += le64_to_cpu(entry.nr_sectors) *
 277                                 log->sectorsize;
 278                 log->cur_pos = lseek(log->logfd, seek_size, SEEK_CUR);
 279                 if (log->cur_pos == (off_t)-1) {
 280                         fprintf(stderr, "Error seeking in file: %d\n", errno);
 281                         return -1;
 282                 }
 283                 log->cur_entry++;
 284         }
 285
 286         return 0;
 287 }
 288
 289 /*
 290  * @log: the log we are manipulating.
 291  * @entry: the entry we read.
 292  * @read_data: read the extra data for the entry, your entry must be
 293  * log->sectorsize large.
 294  *
 295  * @return: 1 if we hit the end of the log, 0 we got the next entry, < 0 if
 296  * there was an error.
 297  *
 298  * Seek to the next entry in the log.
 299  */
 300 int log_seek_next_entry(struct log *log, struct log_write_entry *entry,
 301                         int read_data)
 302 {
 303         size_t read_size = read_data ? log->sectorsize :
 304                 sizeof(struct log_write_entry);
 305         u64 flags;
 306         ssize_t ret;
 307
 308         if (log->cur_entry >= log->nr_entries)
 309                 return 1;
 310
 311         ret = read(log->logfd, entry, read_size);
 312         if (ret != read_size) {
 313                 fprintf(stderr, "Error reading entry: %d\n", errno);
 314                 return -1;
 315         }
 316         if (!log_entry_valid(entry)) {
 317                 fprintf(stderr, "Malformed entry @%llu\n",
 318                                 log->cur_pos / log->sectorsize);
 319                 return -1;
 320         }
 321         log->cur_entry++;
 322
 323         if (read_size < log->sectorsize) {
 324                 log->cur_pos = lseek(log->logfd,
 325                         log->sectorsize - sizeof(struct log_write_entry), SEEK_CUR);
 326                 if (log->cur_pos == (off_t)-1) {
 327                         fprintf(stderr, "Error seeking in log: %d\n", errno);
 328                         return -1;
 329                 }
 330         } else {
 331                 log->cur_pos += read_size;
 332         }
 333         if (log_writes_verbose > 1)
 334                 printf("seek entry %d@%llu: %llu, size %llu, flags %llu\n",
 335                        (int)log->cur_entry - 1, log->cur_pos / log->sectorsize,
 336                        (unsigned long long)le64_to_cpu(entry->sector),
 337                        (unsigned long long)le64_to_cpu(entry->nr_sectors),
 338                        (unsigned long long)le64_to_cpu(entry->flags));
 339
 340         flags = le64_to_cpu(entry->flags);
 341         read_size = le64_to_cpu(entry->nr_sectors) * log->sectorsize;
 342         if (!read_size || (flags & LOG_DISCARD_FLAG))
 343                 return 0;
 344
 345         log->cur_pos = lseek(log->logfd, read_size, SEEK_CUR);
 346         if (log->cur_pos == (off_t)-1) {
 347                 fprintf(stderr, "Error seeking in log: %d\n", errno);
 348                 return -1;
 349         }
 350
 351         return 0;
 352 }
 353
 354 /*
 355  * @logfile: the file that contains the write log.
 356  * @replayfile: the file/device to replay onto, can be NULL.
 357  *
 358  * Opens a logfile and makes sure it is valid and returns a struct log.
 359  */
 360 struct log *log_open(char *logfile, char *replayfile)
 361 {
 362         struct log *log;
 363         struct log_write_super super;
 364         ssize_t ret;
 365
 366         log = malloc(sizeof(struct log));
 367         if (!log) {
 368                 fprintf(stderr, "Couldn't alloc log\n");
 369                 return NULL;
 370         }
 371
 372         log->replayfd = -1;
 373
 374         log->logfd = open(logfile, O_RDONLY);
 375         if (log->logfd < 0) {
 376                 fprintf(stderr, "Couldn't open log %s: %d\n", logfile,
 377                         errno);
 378                 log_free(log);
 379                 return NULL;
 380         }
 381
 382         if (replayfile) {
 383                 log->replayfd = open(replayfile, O_WRONLY);
 384                 if (log->replayfd < 0) {
 385                         fprintf(stderr, "Couldn't open replay file %s: %d\n",
 386                                 replayfile, errno);
 387                         log_free(log);
 388                         return NULL;
 389                 }
 390         }
 391
 392         ret = read(log->logfd, &super, sizeof(struct log_write_super));
 393         if (ret < sizeof(struct log_write_super)) {
 394                 fprintf(stderr, "Error reading super: %d\n", errno);
 395                 log_free(log);
 396                 return NULL;
 397         }
 398
 399         if (le64_to_cpu(super.magic) != WRITE_LOG_MAGIC) {
 400                 fprintf(stderr, "Magic doesn't match\n");
 401                 log_free(log);
 402                 return NULL;
 403         }
 404
 405         if (le64_to_cpu(super.version) != WRITE_LOG_VERSION) {
 406                 fprintf(stderr, "Version mismatch, wanted %d, have %d\n",
 407                         WRITE_LOG_VERSION, (int)le64_to_cpu(super.version));
 408                 log_free(log);
 409                 return NULL;
 410         }
 411
 412         log->sectorsize = le32_to_cpu(super.sectorsize);
 413         log->nr_entries = le64_to_cpu(super.nr_entries);
 414         log->max_zero_size = 128 * 1024 * 1024;
 415
 416         log->cur_pos = lseek(log->logfd, log->sectorsize - sizeof(super), SEEK_CUR);
 417         if (log->cur_pos == (off_t) -1) {
 418                 fprintf(stderr, "Error seeking to first entry: %d\n", errno);
 419                 log_free(log);
 420                 return NULL;
 421         }
 422         log->cur_entry = 0;
 423
 424         return log;
 425 }