#define HEADER_MAGIC 0x676f4c0a
#define FOOTER_MAGIC 0x2e435243
-struct log_header {
- uint32_t magic; // HEADER_MAGIC
- uint8_t type; // Object type + '0'
- uint32_t offset; // Starting byte offset of the log header
- uint32_t size1; // Size of the data item (bytes)
- uint32_t size2; //
- uint32_t size3; //
- uint64_t inum; // Inode which owns this data, if any
- BlueSkyCloudID id; // Object identifier
-} __attribute__((packed));
-
-struct log_footer {
- uint32_t magic; // FOOTER_MAGIC
- uint32_t crc; // Computed from log_header to log_footer.magic
-} __attribute__((packed));
-
static void writebuf(int fd, const char *buf, size_t len)
{
while (len > 0) {
g_mutex_lock(item->lock);
bluesky_cloudlog_stats_update(item, -1);
item->pending_write &= ~CLOUDLOG_JOURNAL;
- item->location_flags |= CLOUDLOG_JOURNAL;
+ item->location_flags
+ = (item->location_flags & ~CLOUDLOG_UNCOMMITTED) | CLOUDLOG_JOURNAL;
bluesky_cloudlog_stats_update(item, 1);
g_cond_signal(item->cond);
g_mutex_unlock(item->lock);
writebuf(log->fd, (const char *)&footer, sizeof(footer));
item->log_seq = log->seq_num;
- item->log_offset = offset + sizeof(header);
- item->log_size = item->data->len;
+ item->log_offset = offset;
+ item->log_size = size;
+ item->data_size = item->data->len;
- offset += sizeof(header) + sizeof(footer) + item->data->len;
+ offset += size;
g_string_free(data1, TRUE);
g_string_free(data2, TRUE);
item->data = NULL;
bluesky_cloudlog_fetch(item);
- log->committed = g_slist_prepend(log->committed, item);
+ log->committed = g_slist_prepend(log->committed, item);
g_atomic_int_add(&item->data_lock_count, -1);
g_mutex_unlock(item->lock);
log->mmap_lock = g_mutex_new();
log->mmap_cache = g_hash_table_new(g_str_hash, g_str_equal);
+ /* Determine the highest-numbered log file, so that we can start writing
+ * out new journal entries at the next sequence number. */
+ GDir *dir = g_dir_open(log_directory, 0, NULL);
+ if (dir != NULL) {
+ const gchar *file;
+ while ((file = g_dir_read_name(dir)) != NULL) {
+ if (strncmp(file, "journal-", 8) == 0) {
+ log->seq_num = MAX(log->seq_num, atoi(&file[8]) + 1);
+ }
+ }
+ g_dir_close(dir);
+ g_print("Starting journal at sequence number %d\n", log->seq_num);
+ }
+
log->dirfd = open(log->log_directory, O_DIRECTORY);
if (log->dirfd < 0) {
fprintf(stderr, "Unable to open logging directory: %m\n");
void bluesky_log_item_submit(BlueSkyCloudLog *item, BlueSkyLog *log)
{
- bluesky_cloudlog_ref(item);
- g_atomic_int_add(&item->data_lock_count, 1);
- g_async_queue_push(log->queue, item);
+ if (!(item->location_flags & CLOUDLOG_JOURNAL)) {
+ bluesky_cloudlog_ref(item);
+ item->location_flags |= CLOUDLOG_UNCOMMITTED;
+ g_atomic_int_add(&item->data_lock_count, 1);
+ g_async_queue_push(log->queue, item);
+ }
}
void bluesky_log_finish_all(GList *log_items)
BlueSkyCloudLog *item = (BlueSkyCloudLog *)log_items->data;
g_mutex_lock(item->lock);
- while ((item->pending_write & CLOUDLOG_JOURNAL))
+ while ((item->location_flags & CLOUDLOG_UNCOMMITTED))
g_cond_wait(item->cond, item->lock);
g_mutex_unlock(item->lock);
bluesky_cloudlog_unref(item);
}
}
+/* Return a committed cloud log record that can be used as a watermark for how
+ * much of the journal has been written. */
+BlueSkyCloudLog *bluesky_log_get_commit_point(BlueSkyFS *fs)
+{
+ BlueSkyCloudLog *marker = bluesky_cloudlog_new(fs, NULL);
+ marker->type = LOGTYPE_JOURNAL_MARKER;
+ marker->data = bluesky_string_new(g_strdup(""), 0);
+ bluesky_cloudlog_sync(marker);
+
+ g_mutex_lock(marker->lock);
+ while ((marker->pending_write & CLOUDLOG_JOURNAL))
+ g_cond_wait(marker->cond, marker->lock);
+ g_mutex_unlock(marker->lock);
+
+ return marker;
+}
+
+void bluesky_log_write_commit_point(BlueSkyFS *fs, BlueSkyCloudLog *marker)
+{
+ BlueSkyCloudLog *commit = bluesky_cloudlog_new(fs, NULL);
+ commit->type = LOGTYPE_JOURNAL_CHECKPOINT;
+
+ uint32_t seq, offset;
+ seq = GUINT32_TO_LE(marker->log_seq);
+ offset = GUINT32_TO_LE(marker->log_offset);
+ GString *loc = g_string_new("");
+ g_string_append_len(loc, (const gchar *)&seq, sizeof(seq));
+ g_string_append_len(loc, (const gchar *)&offset, sizeof(offset));
+ commit->data = bluesky_string_new_from_gstring(loc);
+ bluesky_cloudlog_sync(commit);
+
+ g_mutex_lock(commit->lock);
+ while ((commit->location_flags & CLOUDLOG_UNCOMMITTED))
+ g_cond_wait(commit->cond, commit->lock);
+ g_mutex_unlock(commit->lock);
+
+ bluesky_cloudlog_unref(marker);
+ bluesky_cloudlog_unref(commit);
+}
+
/* Memory-map the given log object into memory (read-only) and return a pointer
* to it. */
static int page_size = 0;
char *pathname = g_strdup_printf("%s/%s",
cachefile->log->log_directory,
cachefile->filename);
+ async->data = bluesky_string_dup(async->data);
+ bluesky_cloudlog_decrypt(async->data->data, async->data->len,
+ cachefile->fs->keys);
if (!g_file_set_contents(pathname, async->data->data, async->data->len,
NULL))
g_print("Error writing out fetched file to cache!\n");
close(fd);
}
- g_mutex_unlock(log->mmap_lock);
-
BlueSkyRCStr *str;
map->atime = bluesky_get_current_time();
str = bluesky_string_new_from_mmap(map, log_offset, log_size);
/* Scan through a journal segment to extract correctly-written items (those
* that pass sanity checks and have a valid checksum). */
-static void bluesky_replay_scan_journal(const char *buf, size_t len)
+static void bluesky_replay_scan_journal(const char *buf, size_t len,
+ uint32_t *seq, uint32_t *start_offset)
{
const struct log_header *header;
off_t offset = 0;
size_t size = GUINT32_FROM_LE(header->size1)
+ GUINT32_FROM_LE(header->size2)
+ GUINT32_FROM_LE(header->size3);
+
+ if (header->type - '0' == LOGTYPE_JOURNAL_CHECKPOINT) {
+ const uint32_t *data = (const uint32_t *)((const char *)header + sizeof(struct log_header));
+ *seq = GUINT32_FROM_LE(data[0]);
+ *start_offset = GUINT32_FROM_LE(data[1]);
+ }
+
offset += sizeof(struct log_header) + size + sizeof(struct log_footer);
}
}
}
static void bluesky_replay_scan_journal2(BlueSkyFS *fs, GList **objects,
- int log_seq,
+ int log_seq, int start_offset,
const char *buf, size_t len)
{
const struct log_header *header;
- off_t offset = 0;
+ off_t offset = start_offset;
while (validate_journal_item(buf, len, offset)) {
header = (const struct log_header *)(buf + offset);
+ GUINT32_FROM_LE(header->size2)
+ GUINT32_FROM_LE(header->size3);
- 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);
+ BlueSkyCloudLog *log_item = bluesky_cloudlog_get(fs, header->id);
+ g_mutex_lock(log_item->lock);
*objects = g_list_prepend(*objects, log_item);
log_item->inum = GUINT64_FROM_LE(header->inum);
if (!bluesky_deserialize_inode(inode, log_item))
g_print("Error deserializing inode %"PRIu64"\n", inum);
fs->next_inum = MAX(fs->next_inum, inum + 1);
+ bluesky_list_unlink(&fs->accessed_list, inode->accessed_list);
+ inode->accessed_list = bluesky_list_prepend(&fs->accessed_list, inode);
+ bluesky_list_unlink(&fs->dirty_list, inode->dirty_list);
+ inode->dirty_list = bluesky_list_prepend(&fs->dirty_list, inode);
+ bluesky_list_unlink(&fs->unlogged_list, inode->unlogged_list);
+ inode->unlogged_list = NULL;
+ inode->change_cloud = inode->change_commit;
+ bluesky_cloudlog_ref(log_item);
+ bluesky_cloudlog_unref(inode->committed_item);
+ inode->committed_item = log_item;
g_mutex_unlock(inode->lock);
g_mutex_unlock(fs->lock);
}
/* Scan through log files in reverse order to find the most recent commit
* record. */
logfiles = g_list_reverse(logfiles);
+ uint32_t seq_num = 0, start_offset = 0;
while (logfiles != NULL) {
char *filename = g_strdup_printf("%s/%s", log->log_directory,
(char *)logfiles->data);
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_get_length(map),
+ &seq_num, &start_offset);
g_mapped_file_unref(map);
}
g_free(filename);
g_free(logfiles->data);
logfiles = g_list_delete_link(logfiles, logfiles);
+ if (seq_num != 0 || start_offset != 0)
+ break;
}
g_list_foreach(logfiles, (GFunc)g_free, NULL);
g_list_free(logfiles);
* 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);
+ g_print("Replaying file %s from offset %d\n", filename, start_offset);
GMappedFile *map = g_mapped_file_new(filename, FALSE, NULL);
g_free(filename);
if (map == NULL) {
break;
}
- bluesky_replay_scan_journal2(fs, &objects, seq_num,
+ bluesky_replay_scan_journal2(fs, &objects, seq_num, start_offset,
g_mapped_file_get_contents(map),
g_mapped_file_get_length(map));
g_mapped_file_unref(map);
seq_num++;
+ start_offset = 0;
}
while (objects != NULL) {
projects / bluesky.git / blobdiff