X-Git-Url: http://git.vrable.net/?a=blobdiff_plain;f=bluesky%2Flog.c;h=a3acf104239fe56e7a24b5d44c39d94d1aac5dc0;hb=d27b934a06369794d21a3eeaf86c55f942518956;hp=1459ee4e76d90918228cb134eb787e407e462121;hpb=b73912811d8a272e4fca4e042020ef1cc3592caf;p=bluesky.git diff --git a/bluesky/log.c b/bluesky/log.c index 1459ee4..a3acf10 100644 --- a/bluesky/log.c +++ b/bluesky/log.c @@ -569,6 +569,26 @@ void bluesky_cachefile_gc(BlueSkyFS *fs) * also been committed to the cloud. Then, we read in all data in the log past * that point. */ +static GList *directory_contents(const char *dirname) +{ + GList *contents = NULL; + GDir *dir = g_dir_open(dirname, 0, NULL); + if (dir == NULL) { + g_warning("Unable to open journal directory: %s", dirname); + return NULL; + } + + const gchar *file; + while ((file = g_dir_read_name(dir)) != NULL) { + if (strncmp(file, "journal-", 8) == 0) + contents = g_list_prepend(contents, g_strdup(file)); + } + g_dir_close(dir); + + contents = g_list_sort(contents, (GCompareFunc)strcmp); + + return contents; +} static gboolean validate_journal_item(const char *buf, size_t len, off_t offset) { @@ -596,8 +616,11 @@ static gboolean validate_journal_item(const char *buf, size_t len, off_t offset) uint32_t crc = crc32c(BLUESKY_CRC32C_SEED, buf + offset, sizeof(struct log_header) + sizeof(struct log_footer) + size); - if (crc != 0) + if (crc != BLUESKY_CRC32C_VALIDATOR) { + g_warning("Journal entry failed to validate: CRC %08x != %08x", + crc, BLUESKY_CRC32C_VALIDATOR); return FALSE; + } return TRUE; } @@ -611,10 +634,45 @@ static void bluesky_replay_scan_journal(const char *buf, size_t len) while (validate_journal_item(buf, len, offset)) { header = (const struct log_header *)(buf + offset); - g_print("In replay found valid item at offset %zd\n", offset); + size_t size = GUINT32_FROM_LE(header->size); + offset += sizeof(struct log_header) + size + sizeof(struct log_footer); + } +} + +static void bluesky_replay_scan_journal2(BlueSkyFS *fs, GList **objects, + int log_seq, + const char *buf, size_t len) +{ + const struct log_header *header; + off_t offset = 0; + while (validate_journal_item(buf, len, offset)) { + header = (const struct log_header *)(buf + offset); + g_print("In replay found valid item at offset %zd\n", offset); size_t size = GUINT32_FROM_LE(header->size); + g_mutex_lock(fs->lock); + BlueSkyCloudLog *log_item; + log_item = g_hash_table_lookup(fs->locations, &header->id); + if (log_item == NULL) { + log_item = bluesky_cloudlog_new(fs, &header->id); + g_hash_table_insert(fs->locations, &log_item->id, log_item); + g_mutex_lock(log_item->lock); + } else { + bluesky_cloudlog_ref(log_item); + g_mutex_lock(log_item->lock); + } + g_mutex_unlock(fs->lock); + *objects = g_list_prepend(*objects, log_item); + + bluesky_string_unref(log_item->data); + log_item->location_flags = CLOUDLOG_JOURNAL; + log_item->data = NULL; + log_item->log_seq = log_seq; + log_item->log_offset = offset + sizeof(struct log_header); + log_item->log_size = header->size; + g_mutex_unlock(log_item->lock); + offset += sizeof(struct log_header) + size + sizeof(struct log_footer); } } @@ -622,21 +680,58 @@ static void bluesky_replay_scan_journal(const char *buf, size_t len) void bluesky_replay(BlueSkyFS *fs) { BlueSkyLog *log = fs->log; - char logname[64]; - int seq_num = 0; - int fd; - - g_snprintf(logname, sizeof(logname), "journal-%08d", seq_num); - fd = openat(log->dirfd, logname, O_RDONLY); - if (fd < 0) - return; + GList *logfiles = directory_contents(log->log_directory); + + /* Scan through log files in reverse order to find the most recent commit + * record. */ + logfiles = g_list_reverse(logfiles); + while (logfiles != NULL) { + char *filename = g_strdup_printf("%s/%s", log->log_directory, + (char *)logfiles->data); + g_print("Scanning file %s\n", filename); + GMappedFile *map = g_mapped_file_new(filename, FALSE, NULL); + if (map == NULL) { + g_warning("Mapping logfile %s failed!\n", filename); + } else { + bluesky_replay_scan_journal(g_mapped_file_get_contents(map), + g_mapped_file_get_length(map)); + g_mapped_file_unref(map); + } + g_free(filename); - off_t length = lseek(fd, 0, SEEK_END); - void *addr = mmap(NULL, length, PROT_READ, MAP_SHARED, fd, 0); - if (addr == NULL) - return; + g_free(logfiles->data); + logfiles = g_list_delete_link(logfiles, logfiles); + } + g_list_foreach(logfiles, (GFunc)g_free, NULL); + g_list_free(logfiles); + + /* Now, scan forward starting from the given point in the log to + * reconstruct all filesystem state. As we reload objects we hold a + * reference to each loaded object. At the end we free all these + * references, so that any objects which were not linked into persistent + * filesystem data structures are freed. */ + GList *objects = NULL; + int seq_num = 0; + while (TRUE) { + char *filename = g_strdup_printf("%s/journal-%08d", + log->log_directory, seq_num); + g_print("Replaying file %s\n", filename); + GMappedFile *map = g_mapped_file_new(filename, FALSE, NULL); + g_free(filename); + if (map == NULL) { + g_warning("Mapping logfile failed, assuming end of journal\n"); + break; + } - bluesky_replay_scan_journal((const char *)addr, length); + bluesky_replay_scan_journal2(fs, &objects, seq_num, + g_mapped_file_get_contents(map), + g_mapped_file_get_length(map)); + g_mapped_file_unref(map); + seq_num++; + } - munmap(addr, length); + while (objects != NULL) { + bluesky_cloudlog_unref((BlueSkyCloudLog *)objects->data); + objects = g_list_delete_link(objects, objects); + } }