X-Git-Url: http://git.vrable.net/?a=blobdiff_plain;f=bluesky%2Fbluesky-private.h;h=74b4e06784d6a32663b1315c2ea8c86f6b4faab4;hb=52a6a600573a0483ea1aa99ec018e649299ea151;hp=baeb14196e916c29c2f04f294b999d3943a73f8d;hpb=810fdc7bdf0cd13aaa7c88d3c5af5aae24e77254;p=bluesky.git diff --git a/bluesky/bluesky-private.h b/bluesky/bluesky-private.h index baeb141..74b4e06 100644 --- a/bluesky/bluesky-private.h +++ b/bluesky/bluesky-private.h @@ -33,6 +33,8 @@ extern int bluesky_watermark_high_total; /* TODO: Make this go away entirely. */ BlueSkyFS *bluesky_new_fs(gchar *name); +void bluesky_inode_free_resources(BlueSkyInode *inode); + /* Linked list update functions for LRU lists. */ void bluesky_list_unlink(GList *head, GList *item); GList *bluesky_list_prepend(GList *head, BlueSkyInode *inode); @@ -44,8 +46,34 @@ BlueSkyInode *bluesky_list_tail(GList *head); * persistent storage. */ void bluesky_serialize_superblock(GString *out, BlueSkyFS *fs); BlueSkyFS *bluesky_deserialize_superblock(const gchar *buf); -void bluesky_serialize_inode(GString *out, BlueSkyInode *inode); -gboolean bluesky_deserialize_inode(BlueSkyInode *inode, const gchar *buf); +BlueSkyCloudLog *bluesky_serialize_inode(BlueSkyInode *inode); +gboolean bluesky_deserialize_inode(BlueSkyInode *inode, BlueSkyCloudLog *item); + +void bluesky_deserialize_cloudlog(BlueSkyCloudLog *item, + const char *data, + size_t len); + +void bluesky_serialize_cloudlog(BlueSkyCloudLog *log, + GString *encrypted, + GString *authenticated, + GString *writable); + +/* Cryptographic operations. */ +#define CRYPTO_BLOCK_SIZE 16 /* 128-bit AES */ +#define CRYPTO_KEY_SIZE 16 +#define CRYPTO_HASH_SIZE 32 /* SHA-256 */ + +typedef struct BlueSkyCryptKeys { + uint8_t encryption_key[CRYPTO_KEY_SIZE]; + uint8_t authentication_key[CRYPTO_HASH_SIZE]; +} BlueSkyCryptKeys; + +void bluesky_crypt_init(); +void bluesky_crypt_hash_key(const char *keystr, uint8_t *out); +void bluesky_crypt_random_bytes(guchar *buf, gint len); +void bluesky_crypt_derive_keys(BlueSkyCryptKeys *keys, const gchar *master); +BlueSkyRCStr *bluesky_crypt_encrypt(BlueSkyRCStr *in, const uint8_t *key); +BlueSkyRCStr *bluesky_crypt_decrypt(BlueSkyRCStr *in, const uint8_t *key); /* Storage layer. Requests can be performed asynchronously, so these objects * help keep track of operations in progress. */ @@ -120,11 +148,16 @@ typedef struct { /* Clean up any implementation-private data in a BlueSkyStoreAsync. */ void (*cleanup)(gpointer store, BlueSkyStoreAsync *async); + + /* Find the lexicographically-largest file starting with the specified + * prefix. */ + char * (*lookup_last)(gpointer store, const gchar *prefix); } BlueSkyStoreImplementation; void bluesky_store_register(const BlueSkyStoreImplementation *impl, const gchar *name); +char *bluesky_store_lookup_last(BlueSkyStore *store, const char *prefix); BlueSkyStoreAsync *bluesky_store_async_new(BlueSkyStore *store); gpointer bluesky_store_async_get_handle(BlueSkyStoreAsync *async); void bluesky_store_async_ref(BlueSkyStoreAsync *async); @@ -139,43 +172,16 @@ void bluesky_store_sync(BlueSkyStore *store); void bluesky_store_add_barrier(BlueSkyStoreAsync *barrier, BlueSkyStoreAsync *async); -void bluesky_inode_start_sync(BlueSkyInode *inode, BlueSkyStoreAsync *barrier); +void bluesky_inode_start_sync(BlueSkyInode *inode); void bluesky_block_touch(BlueSkyInode *inode, uint64_t i); -void bluesky_block_fetch(BlueSkyFS *fs, BlueSkyBlock *block, +void bluesky_block_fetch(BlueSkyInode *inode, BlueSkyBlock *block, BlueSkyStoreAsync *barrier); -void bluesky_block_flush(BlueSkyFS *fs, BlueSkyBlock *block, - BlueSkyStoreAsync *barrier, GList **log_items); -void bluesky_file_flush(BlueSkyInode *inode, BlueSkyStoreAsync *barrier, - GList **log_items); +void bluesky_block_flush(BlueSkyInode *inode, BlueSkyBlock *block, + GList **log_items); +void bluesky_file_flush(BlueSkyInode *inode, GList **log_items); void bluesky_file_drop_cached(BlueSkyInode *inode); -/* Logging infrastructure for ensuring operations are persistently recorded to - * disk. */ -#define BLUESKY_CRC32C_SEED (~(uint32_t)0) -uint32_t crc32c(uint32_t crc, const char *buf, unsigned int length); -uint32_t crc32c_finalize(uint32_t crc); - -struct _BlueSkyLog { - char *log_directory; - GAsyncQueue *queue; - int fd; - int seq_num; -}; - -typedef struct { - gboolean committed; - GMutex *lock; - GCond *cond; - char *key; - BlueSkyRCStr *data; -} BlueSkyLogItem; - -BlueSkyLog *bluesky_log_new(const char *log_directory); -BlueSkyLogItem *bluesky_log_item_new(); -void bluesky_log_item_submit(BlueSkyLogItem *item, BlueSkyLog *log); -void bluesky_log_item_finish(BlueSkyLogItem *item); - /* Writing of data to the cloud in log segments and tracking the location of * various pieces of data (both where in the cloud and where cached locally). * */ @@ -192,21 +198,59 @@ typedef struct { } BlueSkyCloudPointer; typedef enum { + LOGTYPE_INVALID = -1, LOGTYPE_UNKNOWN = 0, LOGTYPE_DATA = 1, LOGTYPE_INODE = 2, LOGTYPE_INODE_MAP = 3, LOGTYPE_CHECKPOINT = 4, - LOGTYPE_CHECKPOINT_PTR = 5, + + /* Used only as metadata in the local journal, not loaded as a + * BlueSkyCloudLogState nor stored in the cloud */ + LOGTYPE_JOURNAL_MARKER = 16, + LOGTYPE_JOURNAL_CHECKPOINT = 17, } BlueSkyCloudLogType; +/* Headers that go on items in local log segments and cloud log segments. */ +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)); + +struct cloudlog_header { + char magic[4]; + uint8_t crypt_auth[CRYPTO_HASH_SIZE]; + uint8_t crypt_iv[CRYPTO_BLOCK_SIZE]; + uint8_t type; + BlueSkyCloudID id; + uint64_t inum; + uint32_t size1, size2, size3; +} __attribute__((packed)); + +#define JOURNAL_MAGIC "\nLog" +#define CLOUDLOG_MAGIC "AgI-" + /* A record which tracks an object which has been written to a local log, * cached, locally, and/or written to the cloud. */ #define CLOUDLOG_JOURNAL 0x01 -#define CLOUDLOG_CACHE 0x02 -#define CLOUDLOG_CLOUD 0x04 -typedef struct { +#define CLOUDLOG_CLOUD 0x02 +#define CLOUDLOG_CACHE 0x04 +#define CLOUDLOG_UNCOMMITTED 0x10 +struct _BlueSkyCloudLog { gint refcount; + GMutex *lock; + GCond *cond; BlueSkyFS *fs; @@ -214,6 +258,7 @@ typedef struct { // Bitmask of CLOUDLOG_* flags indicating where the object exists. int location_flags; + int pending_read, pending_write; // A stable identifier for the object (only changes when authenticated data // is written out, but stays the same when the in-cloud cleaner relocates @@ -224,22 +269,167 @@ typedef struct { uint64_t inum; int32_t inum_offset; + // The size of encrypted object data, not including any headers + int data_size; + // The location of the object in the cloud, if available. BlueSkyCloudPointer location; // TODO: Location in journal/cache + int log_seq, log_offset, log_size; - // Serialized data, if available in memory (otherwise NULL). + // Pointers to other objects. Each link counts towards the reference count + // of the pointed-to object. To avoid memory leaks there should be no + // cycles in the reference graph. + GArray *links; + + // Serialized data, if available in memory (otherwise NULL), and a lock + // count which tracks if there are users that require the data to be kept + // around. BlueSkyRCStr *data; -} BlueSkyCloudLog; + int data_lock_count; +}; + +/* Serialize objects into a log segment to be written to the cloud. */ +struct _BlueSkyCloudLogState { + GString *data; + BlueSkyCloudPointer location; + GList *inode_list; + GSList *writeback_list; // Items which are being serialized right now + GList *pending_segments; // Segments which are being uploaded now +}; gboolean bluesky_cloudlog_equal(gconstpointer a, gconstpointer b); guint bluesky_cloudlog_hash(gconstpointer a); -BlueSkyCloudLog *bluesky_cloudlog_new(BlueSkyFS *fs); +BlueSkyCloudLog *bluesky_cloudlog_new(BlueSkyFS *fs, const BlueSkyCloudID *id); +gchar *bluesky_cloudlog_id_to_string(BlueSkyCloudID id); +BlueSkyCloudID bluesky_cloudlog_id_from_string(const gchar *idstr); +void bluesky_cloudlog_threads_init(BlueSkyFS *fs); void bluesky_cloudlog_ref(BlueSkyCloudLog *log); void bluesky_cloudlog_unref(BlueSkyCloudLog *log); -BlueSkyLogItem *bluesky_cloudlog_sync(BlueSkyCloudLog *log); +void bluesky_cloudlog_unref_delayed(BlueSkyCloudLog *log); +void bluesky_cloudlog_erase(BlueSkyCloudLog *log); +void bluesky_cloudlog_stats_update(BlueSkyCloudLog *log, int type); +void bluesky_cloudlog_sync(BlueSkyCloudLog *log); void bluesky_cloudlog_insert(BlueSkyCloudLog *log); +void bluesky_cloudlog_insert_locked(BlueSkyCloudLog *log); +BlueSkyCloudLog *bluesky_cloudlog_get(BlueSkyFS *fs, BlueSkyCloudID id); +void bluesky_cloudlog_fetch(BlueSkyCloudLog *log); +BlueSkyCloudPointer bluesky_cloudlog_serialize(BlueSkyCloudLog *log, + BlueSkyFS *fs); +void bluesky_cloudlog_flush(BlueSkyFS *fs); + +/* Logging infrastructure for ensuring operations are persistently recorded to + * disk. */ +#define BLUESKY_CRC32C_SEED (~(uint32_t)0) +#define BLUESKY_CRC32C_VALIDATOR ((uint32_t)0xb798b438UL) +uint32_t crc32c(uint32_t crc, const char *buf, unsigned int length); +uint32_t crc32c_finalize(uint32_t crc); + +struct _BlueSkyLog { + BlueSkyFS *fs; + char *log_directory; + GAsyncQueue *queue; + int fd, dirfd; + int seq_num; + GSList *committed; + + /* The currently-open log file. */ + BlueSkyCacheFile *current_log; + + /* Cache of log segments which have been memory-mapped. */ + GMutex *mmap_lock; + GHashTable *mmap_cache; + + /* A count of the disk space consumed (in 1024-byte units) by all files + * tracked by mmap_cache (whether mapped or not, actually). */ + gint disk_used; + + /* The smallest journal sequence number which may still contain data that + * must be preserved (since it it not yet in the cloud). */ + int journal_watermark; +}; + +/* An object for tracking log files which are stored locally--either the + * journal for filesystem consistency or log segments which have been fetched + * back from cloud storage. */ +struct _BlueSkyCacheFile { + GMutex *lock; + GCond *cond; + gint refcount; + int type; // Only one of CLOUDLOG_{JOURNAL,CLOUD} + int log_dir; + int log_seq; + char *filename; // Local filename, relateive to log directory + gint mapcount; // References to the mmaped data + const char *addr; // May be null if data is not mapped in memory + size_t len; + BlueSkyFS *fs; + BlueSkyLog *log; + gboolean fetching, ready; // Cloud data: downloading or ready for use + int64_t atime; // Access time, for cache management +}; + +BlueSkyLog *bluesky_log_new(const char *log_directory); +void bluesky_log_item_submit(BlueSkyCloudLog *item, BlueSkyLog *log); +void bluesky_log_finish_all(GList *log_items); +BlueSkyCloudLog *bluesky_log_get_commit_point(BlueSkyFS *fs); +void bluesky_log_write_commit_point(BlueSkyFS *fs, BlueSkyCloudLog *marker); + +BlueSkyRCStr *bluesky_log_map_object(BlueSkyFS *fs, int log_dir, int log_seq, + int log_offset, int log_size); +void bluesky_mmap_unref(BlueSkyCacheFile *mmap); +void bluesky_cachefile_unref(BlueSkyCacheFile *cachefile); + +BlueSkyCacheFile *bluesky_cachefile_lookup(BlueSkyFS *fs, + int clouddir, int log_seq); +void bluesky_cachefile_gc(BlueSkyFS *fs); + +void bluesky_replay(BlueSkyFS *fs); + +/* Used to track log segments that are being written to the cloud. */ +typedef struct { + BlueSkyRCStr *data; + GSList *items; + GMutex *lock; + GCond *cond; + gboolean complete; +} SerializedRecord; + +/***** Inode map management *****/ + +/* Mapping information for a single inode number. These are grouped together + * into InodeMapRange objects. */ +typedef struct { + uint64_t inum; + + /* A pointer to the cloud log entry for this inode. This may or may not + * actually have data loaded (it might just contain pointers to the data + * location, and in fact this will likely often be the case). */ + BlueSkyCloudLog *item; +} InodeMapEntry; + +typedef struct { + /* Starting and ending inode number values that fall in this section. + * Endpoint values are inclusive. */ + uint64_t start, end; + + /* A sorted list (by inode number) of InodeMapEntry objects. */ + GSequence *map_entries; + + /* The serialized version of the inode map data. */ + BlueSkyCloudLog *serialized; + + /* Have there been changes that require writing this section out again? */ + gboolean dirty; +} InodeMapRange; + +InodeMapEntry *bluesky_inode_map_lookup(GSequence *inode_map, uint64_t inum, + int action); +BlueSkyCloudLog *bluesky_inode_map_serialize(BlueSkyFS *fs); +void bluesky_inode_map_minimize(BlueSkyFS *fs); + +gboolean bluesky_checkpoint_load(BlueSkyFS *fs); #ifdef __cplusplus }