// no absolute guarantees on the size of a log segment.
#define LOG_SEGMENT_SIZE (1 << 22)
-// Target amount of disk space to use for the journal and cache files, in
-// kilobytes.
-#define DISK_CACHE_SIZE_TARGET (64 * 1024)
-
#define HEADER_MAGIC 0x676f4c0a
#define FOOTER_MAGIC 0x2e435243
struct log_header {
uint32_t magic; // HEADER_MAGIC
- uint64_t offset; // Starting byte offset of the log header
+ uint8_t type; // Object type + '0'
+ uint32_t offset; // Starting byte offset of the log header
uint32_t size; // Size of the data item (bytes)
+ uint64_t inum; // Inode which owns this data, if any
BlueSkyCloudID id; // Object identifier
} __attribute__((packed));
}
header.magic = GUINT32_TO_LE(HEADER_MAGIC);
- header.offset = GUINT64_TO_LE(offset);
+ header.offset = GUINT32_TO_LE(offset);
header.size = GUINT32_TO_LE(item->data->len);
+ header.type = item->type + '0';
header.id = item->id;
+ header.inum = GUINT64_TO_LE(item->inum);
footer.magic = GUINT32_TO_LE(FOOTER_MAGIC);
uint32_t crc = BLUESKY_CRC32C_SEED;
offset += sizeof(header) + sizeof(footer) + item->data->len;
- /* Since we have just written a new dirty object to the journal,
- * increment the count of live dirty objects in that journal file. The
- * count will be decremented when objects are deleted or written to the
- * cloud. */
- if (!(item->location_flags & CLOUDLOG_CLOUD)) {
- g_atomic_int_add(&log->current_log->dirty_refs, 1);
- item->dirty_journal = log->current_log;
- }
-
/* Replace the log item's string data with a memory-mapped copy of the
* data, now that it has been written to the log file. (Even if it
* isn't yet on disk, it should at least be in the page cache and so
* then we'll just skip the file on this pass. */
if (g_mutex_trylock(cachefile->lock)) {
int64_t age = bluesky_get_current_time() - cachefile->atime;
- g_print("%s addr=%p mapcount=%d refcount=%d dirty=%d atime_age=%f",
+ g_print("%s addr=%p mapcount=%d refcount=%d atime_age=%f",
cachefile->filename, cachefile->addr, cachefile->mapcount,
- cachefile->refcount, cachefile->dirty_refs, age / 1e6);
+ cachefile->refcount, age / 1e6);
if (cachefile->fetching)
g_print(" (fetching)");
g_print("\n");
- if (g_atomic_int_get(&fs->log->disk_used) > DISK_CACHE_SIZE_TARGET
+ gboolean deletion_candidate = FALSE;
+ if (g_atomic_int_get(&fs->log->disk_used)
+ > bluesky_options.cache_size
&& g_atomic_int_get(&cachefile->refcount) == 0
- && g_atomic_int_get(&cachefile->mapcount) == 0
- && g_atomic_int_get(&cachefile->dirty_refs) == 0)
+ && g_atomic_int_get(&cachefile->mapcount) == 0)
+ {
+ deletion_candidate = TRUE;
+ }
+
+ /* Don't allow journal files to be reclaimed until all data is
+ * known to be durably stored in the cloud. */
+ if (cachefile->type == CLOUDLOG_JOURNAL
+ && cachefile->log_seq >= fs->log->journal_watermark)
{
+ deletion_candidate = FALSE;
+ }
+
+ if (deletion_candidate) {
g_print(" ...deleting\n");
if (unlinkat(fs->log->dirfd, cachefile->filename, 0) < 0) {
fprintf(stderr, "Unable to unlink journal %s: %m\n",
g_mutex_unlock(fs->log->mmap_lock);
}
+
+/******************************* JOURNAL REPLAY *******************************
+ * The journal replay code is used to recover filesystem state after a
+ * filesystem restart. We first look for the most recent commit record in the
+ * journal, which indicates the point before which all data in the journal has
+ * also been committed to the cloud. Then, we read in all data in the log past
+ * that point.
+ */
+
+static gboolean validate_journal_item(const char *buf, size_t len, off_t offset)
+{
+ const struct log_header *header;
+ const struct log_footer *footer;
+
+ if (offset + sizeof(struct log_header) + sizeof(struct log_footer) > len)
+ return FALSE;
+
+ header = (const struct log_header *)(buf + offset);
+ if (GUINT32_FROM_LE(header->magic) != HEADER_MAGIC)
+ return FALSE;
+ if (GUINT32_FROM_LE(header->offset) != offset)
+ return FALSE;
+ size_t size = GUINT32_FROM_LE(header->size);
+
+ off_t footer_offset = offset + sizeof(struct log_header) + size;
+ if (footer_offset + sizeof(struct log_footer) > len)
+ return FALSE;
+ footer = (const struct log_footer *)(buf + footer_offset);
+
+ if (GUINT32_FROM_LE(footer->magic) != FOOTER_MAGIC)
+ return FALSE;
+
+ uint32_t crc = crc32c(BLUESKY_CRC32C_SEED, buf + offset,
+ sizeof(struct log_header) + sizeof(struct log_footer)
+ + size);
+ if (crc != 0)
+ return FALSE;
+
+ return TRUE;
+}
+
+/* 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)
+{
+ 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);
+
+ offset += sizeof(struct log_header) + size + sizeof(struct log_footer);
+ }
+}
+
+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;
+
+ off_t length = lseek(fd, 0, SEEK_END);
+ void *addr = mmap(NULL, length, PROT_READ, MAP_SHARED, fd, 0);
+ if (addr == NULL)
+ return;
+
+ bluesky_replay_scan_journal((const char *)addr, length);
+
+ munmap(addr, length);
+}