1 /* Blue Sky: File Systems in the Cloud
3 * Copyright (C) 2009 The Regents of the University of California
4 * Written by Michael Vrable <mvrable@cs.ucsd.edu>
9 /* Declarations internal to the BlueSky library. This header file should not
10 * be included by any users of the library (such as any filesystem
11 * proxy)--external users should only include bluesky.h. */
13 #ifndef _BLUESKY_PRIVATE_H
14 #define _BLUESKY_PRIVATE_H
22 extern int bluesky_verbose;
24 /* Target cache size levels. */
25 extern int bluesky_watermark_low_dirty;
26 extern int bluesky_watermark_medium_dirty;
27 extern int bluesky_watermark_high_dirty;
29 extern int bluesky_watermark_low_total;
30 extern int bluesky_watermark_medium_total;
31 extern int bluesky_watermark_high_total;
33 /* TODO: Make this go away entirely. */
34 BlueSkyFS *bluesky_new_fs(gchar *name);
36 void bluesky_inode_free_resources(BlueSkyInode *inode);
38 /* Linked list update functions for LRU lists. */
39 void bluesky_list_unlink(GList *head, GList *item);
40 GList *bluesky_list_prepend(GList *head, BlueSkyInode *inode);
41 GList *bluesky_list_append(GList *head, BlueSkyInode *inode);
42 BlueSkyInode *bluesky_list_head(GList *head);
43 BlueSkyInode *bluesky_list_tail(GList *head);
45 /* Serialization and deserialization of filesystem data for storing to
46 * persistent storage. */
47 void bluesky_serialize_superblock(GString *out, BlueSkyFS *fs);
48 BlueSkyFS *bluesky_deserialize_superblock(const gchar *buf);
49 BlueSkyCloudLog *bluesky_serialize_inode(BlueSkyInode *inode);
50 gboolean bluesky_deserialize_inode(BlueSkyInode *inode, BlueSkyCloudLog *item);
52 void bluesky_deserialize_cloudlog(BlueSkyCloudLog *item,
56 void bluesky_serialize_cloudlog(BlueSkyCloudLog *log,
58 GString *authenticated,
61 /* Cryptographic operations. */
62 #define CRYPTO_BLOCK_SIZE 16 /* 128-bit AES */
63 #define CRYPTO_KEY_SIZE 16
64 #define CRYPTO_HASH_SIZE 32 /* SHA-256 */
66 typedef struct BlueSkyCryptKeys {
67 uint8_t encryption_key[CRYPTO_KEY_SIZE];
68 uint8_t authentication_key[CRYPTO_HASH_SIZE];
71 void bluesky_crypt_init();
72 void bluesky_crypt_hash_key(const char *keystr, uint8_t *out);
73 void bluesky_crypt_random_bytes(guchar *buf, gint len);
74 void bluesky_crypt_derive_keys(BlueSkyCryptKeys *keys, const gchar *master);
75 BlueSkyRCStr *bluesky_crypt_encrypt(BlueSkyRCStr *in, const uint8_t *key);
76 BlueSkyRCStr *bluesky_crypt_decrypt(BlueSkyRCStr *in, const uint8_t *key);
78 /* Storage layer. Requests can be performed asynchronously, so these objects
79 * help keep track of operations in progress. */
85 STORE_OP_BARRIER, // Waits for other selected operations to complete
89 ASYNC_NEW, // Operation not yet submitted to storage layer
90 ASYNC_PENDING, // Submitted to storage layer
91 ASYNC_RUNNING, // Operation is in progress
92 ASYNC_COMPLETE, // Operation finished, results available
95 struct BlueSkyNotifierList;
96 typedef struct _BlueSkyStoreAsync BlueSkyStoreAsync;
97 struct _BlueSkyStoreAsync {
101 GCond *completion_cond; /* Used to wait for operation to complete. */
103 gint refcount; /* Reference count for destruction. */
105 BlueSkyAsyncStatus status;
108 gchar *key; /* Key to read/write */
109 BlueSkyRCStr *data; /* Data read/to write */
111 int result; /* Result code; 0 for success. */
112 struct BlueSkyNotifierList *notifiers;
115 /* The barrier waiting on this operation. Support for more than one
116 * barrier for a single async is not well-supported and should be avoided
118 BlueSkyStoreAsync *barrier;
120 bluesky_time_hires start_time; /* Time operation was submitted. */
121 bluesky_time_hires exec_time; /* Time processing started on operation. */
123 gpointer store_private; /* For use by the storage implementation */
126 /* Support for notification lists. These are lists of one-shot functions which
127 * can be called when certain events--primarily, competed storage
128 * events--occur. Multiple notifiers can be added, but no particular order is
129 * guaranteed for the notification functions to be called. */
130 struct BlueSkyNotifierList {
131 struct BlueSkyNotifierList *next;
133 BlueSkyStoreAsync *async;
134 gpointer user_data; // Passed to the function when called
137 /* The abstraction layer for storage, allowing multiple implementations. */
139 /* Create a new store instance and return a handle to it. */
140 gpointer (*create)(const gchar *path);
142 /* Clean up any resources used by this store. */
143 void (*destroy)(gpointer store);
145 /* Submit an operation (get/put/delete) to the storage layer to be
146 * performed asynchronously. */
147 void (*submit)(gpointer store, BlueSkyStoreAsync *async);
149 /* Clean up any implementation-private data in a BlueSkyStoreAsync. */
150 void (*cleanup)(gpointer store, BlueSkyStoreAsync *async);
152 /* Find the lexicographically-largest file starting with the specified
154 char * (*lookup_last)(gpointer store, const gchar *prefix);
155 } BlueSkyStoreImplementation;
157 void bluesky_store_register(const BlueSkyStoreImplementation *impl,
160 char *bluesky_store_lookup_last(BlueSkyStore *store, const char *prefix);
161 BlueSkyStoreAsync *bluesky_store_async_new(BlueSkyStore *store);
162 gpointer bluesky_store_async_get_handle(BlueSkyStoreAsync *async);
163 void bluesky_store_async_ref(BlueSkyStoreAsync *async);
164 void bluesky_store_async_unref(BlueSkyStoreAsync *async);
165 void bluesky_store_async_wait(BlueSkyStoreAsync *async);
166 void bluesky_store_async_add_notifier(BlueSkyStoreAsync *async,
167 GFunc func, gpointer user_data);
168 void bluesky_store_async_mark_complete(BlueSkyStoreAsync *async);
169 void bluesky_store_async_submit(BlueSkyStoreAsync *async);
170 void bluesky_store_sync(BlueSkyStore *store);
172 void bluesky_store_add_barrier(BlueSkyStoreAsync *barrier,
173 BlueSkyStoreAsync *async);
175 void bluesky_inode_start_sync(BlueSkyInode *inode);
177 void bluesky_block_touch(BlueSkyInode *inode, uint64_t i);
178 void bluesky_block_fetch(BlueSkyInode *inode, BlueSkyBlock *block,
179 BlueSkyStoreAsync *barrier);
180 void bluesky_block_flush(BlueSkyInode *inode, BlueSkyBlock *block,
182 void bluesky_file_flush(BlueSkyInode *inode, GList **log_items);
183 void bluesky_file_drop_cached(BlueSkyInode *inode);
185 /* Writing of data to the cloud in log segments and tracking the location of
186 * various pieces of data (both where in the cloud and where cached locally).
198 } BlueSkyCloudPointer;
201 LOGTYPE_INVALID = -1,
205 LOGTYPE_INODE_MAP = 3,
206 LOGTYPE_CHECKPOINT = 4,
208 /* Used only as metadata in the local journal, not loaded as a
209 * BlueSkyCloudLogState nor stored in the cloud */
210 LOGTYPE_JOURNAL_MARKER = 16,
211 LOGTYPE_JOURNAL_CHECKPOINT = 17,
212 } BlueSkyCloudLogType;
214 /* Headers that go on items in local log segments and cloud log segments. */
216 uint32_t magic; // HEADER_MAGIC
217 uint8_t type; // Object type + '0'
218 uint32_t offset; // Starting byte offset of the log header
219 uint32_t size1; // Size of the data item (bytes)
222 uint64_t inum; // Inode which owns this data, if any
223 BlueSkyCloudID id; // Object identifier
224 } __attribute__((packed));
227 uint32_t magic; // FOOTER_MAGIC
228 uint32_t crc; // Computed from log_header to log_footer.magic
229 } __attribute__((packed));
231 struct cloudlog_header {
233 uint8_t crypt_auth[CRYPTO_HASH_SIZE];
234 uint8_t crypt_iv[CRYPTO_BLOCK_SIZE];
238 uint32_t size1, size2, size3;
239 } __attribute__((packed));
241 #define JOURNAL_MAGIC "\nLog"
242 #define CLOUDLOG_MAGIC "AgI-"
244 /* A record which tracks an object which has been written to a local log,
245 * cached, locally, and/or written to the cloud. */
246 #define CLOUDLOG_JOURNAL 0x01
247 #define CLOUDLOG_CLOUD 0x02
248 #define CLOUDLOG_CACHE 0x04
249 #define CLOUDLOG_UNCOMMITTED 0x10
250 struct _BlueSkyCloudLog {
257 BlueSkyCloudLogType type;
259 // Bitmask of CLOUDLOG_* flags indicating where the object exists.
261 int pending_read, pending_write;
263 // A stable identifier for the object (only changes when authenticated data
264 // is written out, but stays the same when the in-cloud cleaner relocates
268 // The inode which owns this data, if any, and an offset.
272 // The size of encrypted object data, not including any headers
275 // The location of the object in the cloud, if available.
276 BlueSkyCloudPointer location;
278 // TODO: Location in journal/cache
279 int log_seq, log_offset, log_size;
281 // Pointers to other objects. Each link counts towards the reference count
282 // of the pointed-to object. To avoid memory leaks there should be no
283 // cycles in the reference graph.
286 // Serialized data, if available in memory (otherwise NULL), and a lock
287 // count which tracks if there are users that require the data to be kept
293 /* Serialize objects into a log segment to be written to the cloud. */
294 struct _BlueSkyCloudLogState {
296 BlueSkyCloudPointer location;
298 GSList *writeback_list; // Items which are being serialized right now
299 GList *pending_segments; // Segments which are being uploaded now
302 gboolean bluesky_cloudlog_equal(gconstpointer a, gconstpointer b);
303 guint bluesky_cloudlog_hash(gconstpointer a);
304 BlueSkyCloudLog *bluesky_cloudlog_new(BlueSkyFS *fs, const BlueSkyCloudID *id);
305 gchar *bluesky_cloudlog_id_to_string(BlueSkyCloudID id);
306 BlueSkyCloudID bluesky_cloudlog_id_from_string(const gchar *idstr);
307 void bluesky_cloudlog_threads_init(BlueSkyFS *fs);
308 void bluesky_cloudlog_ref(BlueSkyCloudLog *log);
309 void bluesky_cloudlog_unref(BlueSkyCloudLog *log);
310 void bluesky_cloudlog_unref_delayed(BlueSkyCloudLog *log);
311 void bluesky_cloudlog_erase(BlueSkyCloudLog *log);
312 void bluesky_cloudlog_stats_update(BlueSkyCloudLog *log, int type);
313 void bluesky_cloudlog_sync(BlueSkyCloudLog *log);
314 void bluesky_cloudlog_insert(BlueSkyCloudLog *log);
315 void bluesky_cloudlog_insert_locked(BlueSkyCloudLog *log);
316 BlueSkyCloudLog *bluesky_cloudlog_get(BlueSkyFS *fs, BlueSkyCloudID id);
317 void bluesky_cloudlog_fetch(BlueSkyCloudLog *log);
318 BlueSkyCloudPointer bluesky_cloudlog_serialize(BlueSkyCloudLog *log,
320 void bluesky_cloudlog_flush(BlueSkyFS *fs);
322 /* Logging infrastructure for ensuring operations are persistently recorded to
324 #define BLUESKY_CRC32C_SEED (~(uint32_t)0)
325 #define BLUESKY_CRC32C_VALIDATOR ((uint32_t)0xb798b438UL)
326 uint32_t crc32c(uint32_t crc, const char *buf, unsigned int length);
327 uint32_t crc32c_finalize(uint32_t crc);
337 /* The currently-open log file. */
338 BlueSkyCacheFile *current_log;
340 /* Cache of log segments which have been memory-mapped. */
342 GHashTable *mmap_cache;
344 /* A count of the disk space consumed (in 1024-byte units) by all files
345 * tracked by mmap_cache (whether mapped or not, actually). */
348 /* The smallest journal sequence number which may still contain data that
349 * must be preserved (since it it not yet in the cloud). */
350 int journal_watermark;
353 /* An object for tracking log files which are stored locally--either the
354 * journal for filesystem consistency or log segments which have been fetched
355 * back from cloud storage. */
356 struct _BlueSkyCacheFile {
360 int type; // Only one of CLOUDLOG_{JOURNAL,CLOUD}
363 char *filename; // Local filename, relateive to log directory
364 gint mapcount; // References to the mmaped data
365 const char *addr; // May be null if data is not mapped in memory
369 gboolean fetching, ready; // Cloud data: downloading or ready for use
370 int64_t atime; // Access time, for cache management
373 BlueSkyLog *bluesky_log_new(const char *log_directory);
374 void bluesky_log_item_submit(BlueSkyCloudLog *item, BlueSkyLog *log);
375 void bluesky_log_finish_all(GList *log_items);
376 BlueSkyCloudLog *bluesky_log_get_commit_point(BlueSkyFS *fs);
377 void bluesky_log_write_commit_point(BlueSkyFS *fs, BlueSkyCloudLog *marker);
379 BlueSkyRCStr *bluesky_log_map_object(BlueSkyFS *fs, int log_dir, int log_seq,
380 int log_offset, int log_size);
381 void bluesky_mmap_unref(BlueSkyCacheFile *mmap);
382 void bluesky_cachefile_unref(BlueSkyCacheFile *cachefile);
384 BlueSkyCacheFile *bluesky_cachefile_lookup(BlueSkyFS *fs,
385 int clouddir, int log_seq);
386 void bluesky_cachefile_gc(BlueSkyFS *fs);
388 void bluesky_replay(BlueSkyFS *fs);
390 /* Used to track log segments that are being written to the cloud. */
399 /***** Inode map management *****/
401 /* Mapping information for a single inode number. These are grouped together
402 * into InodeMapRange objects. */
406 /* A pointer to the cloud log entry for this inode. This may or may not
407 * actually have data loaded (it might just contain pointers to the data
408 * location, and in fact this will likely often be the case). */
409 BlueSkyCloudLog *item;
413 /* Starting and ending inode number values that fall in this section.
414 * Endpoint values are inclusive. */
417 /* A sorted list (by inode number) of InodeMapEntry objects. */
418 GSequence *map_entries;
420 /* The serialized version of the inode map data. */
421 BlueSkyCloudLog *serialized;
423 /* Have there been changes that require writing this section out again? */
427 InodeMapEntry *bluesky_inode_map_lookup(GSequence *inode_map, uint64_t inum,
429 BlueSkyCloudLog *bluesky_inode_map_serialize(BlueSkyFS *fs);
430 void bluesky_inode_map_minimize(BlueSkyFS *fs);
432 gboolean bluesky_checkpoint_load(BlueSkyFS *fs);