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 /* Linked list update functions for LRU lists. */
37 void bluesky_list_unlink(GList *head, GList *item);
38 GList *bluesky_list_prepend(GList *head, BlueSkyInode *inode);
39 GList *bluesky_list_append(GList *head, BlueSkyInode *inode);
40 BlueSkyInode *bluesky_list_head(GList *head);
41 BlueSkyInode *bluesky_list_tail(GList *head);
43 /* Serialization and deserialization of filesystem data for storing to
44 * persistent storage. */
45 void bluesky_serialize_superblock(GString *out, BlueSkyFS *fs);
46 BlueSkyFS *bluesky_deserialize_superblock(const gchar *buf);
47 BlueSkyCloudLog *bluesky_serialize_inode(BlueSkyInode *inode);
48 gboolean bluesky_deserialize_inode(BlueSkyInode *inode, BlueSkyCloudLog *item);
50 void bluesky_serialize_cloudlog(BlueSkyCloudLog *log,
52 GString *authenticated,
55 /* Storage layer. Requests can be performed asynchronously, so these objects
56 * help keep track of operations in progress. */
62 STORE_OP_BARRIER, // Waits for other selected operations to complete
66 ASYNC_NEW, // Operation not yet submitted to storage layer
67 ASYNC_PENDING, // Submitted to storage layer
68 ASYNC_RUNNING, // Operation is in progress
69 ASYNC_COMPLETE, // Operation finished, results available
72 struct BlueSkyNotifierList;
73 typedef struct _BlueSkyStoreAsync BlueSkyStoreAsync;
74 struct _BlueSkyStoreAsync {
78 GCond *completion_cond; /* Used to wait for operation to complete. */
80 gint refcount; /* Reference count for destruction. */
82 BlueSkyAsyncStatus status;
85 gchar *key; /* Key to read/write */
86 BlueSkyRCStr *data; /* Data read/to write */
88 int result; /* Result code; 0 for success. */
89 struct BlueSkyNotifierList *notifiers;
92 /* The barrier waiting on this operation. Support for more than one
93 * barrier for a single async is not well-supported and should be avoided
95 BlueSkyStoreAsync *barrier;
97 bluesky_time_hires start_time; /* Time operation was submitted. */
98 bluesky_time_hires exec_time; /* Time processing started on operation. */
100 gpointer store_private; /* For use by the storage implementation */
103 /* Support for notification lists. These are lists of one-shot functions which
104 * can be called when certain events--primarily, competed storage
105 * events--occur. Multiple notifiers can be added, but no particular order is
106 * guaranteed for the notification functions to be called. */
107 struct BlueSkyNotifierList {
108 struct BlueSkyNotifierList *next;
110 BlueSkyStoreAsync *async;
111 gpointer user_data; // Passed to the function when called
114 /* The abstraction layer for storage, allowing multiple implementations. */
116 /* Create a new store instance and return a handle to it. */
117 gpointer (*create)(const gchar *path);
119 /* Clean up any resources used by this store. */
120 void (*destroy)(gpointer store);
122 /* Submit an operation (get/put/delete) to the storage layer to be
123 * performed asynchronously. */
124 void (*submit)(gpointer store, BlueSkyStoreAsync *async);
126 /* Clean up any implementation-private data in a BlueSkyStoreAsync. */
127 void (*cleanup)(gpointer store, BlueSkyStoreAsync *async);
128 } BlueSkyStoreImplementation;
130 void bluesky_store_register(const BlueSkyStoreImplementation *impl,
133 BlueSkyStoreAsync *bluesky_store_async_new(BlueSkyStore *store);
134 gpointer bluesky_store_async_get_handle(BlueSkyStoreAsync *async);
135 void bluesky_store_async_ref(BlueSkyStoreAsync *async);
136 void bluesky_store_async_unref(BlueSkyStoreAsync *async);
137 void bluesky_store_async_wait(BlueSkyStoreAsync *async);
138 void bluesky_store_async_add_notifier(BlueSkyStoreAsync *async,
139 GFunc func, gpointer user_data);
140 void bluesky_store_async_mark_complete(BlueSkyStoreAsync *async);
141 void bluesky_store_async_submit(BlueSkyStoreAsync *async);
142 void bluesky_store_sync(BlueSkyStore *store);
144 void bluesky_store_add_barrier(BlueSkyStoreAsync *barrier,
145 BlueSkyStoreAsync *async);
147 void bluesky_inode_start_sync(BlueSkyInode *inode);
149 void bluesky_block_touch(BlueSkyInode *inode, uint64_t i);
150 void bluesky_block_fetch(BlueSkyInode *inode, BlueSkyBlock *block,
151 BlueSkyStoreAsync *barrier);
152 void bluesky_block_flush(BlueSkyInode *inode, BlueSkyBlock *block,
154 void bluesky_file_flush(BlueSkyInode *inode, GList **log_items);
155 void bluesky_file_drop_cached(BlueSkyInode *inode);
157 /* Writing of data to the cloud in log segments and tracking the location of
158 * various pieces of data (both where in the cloud and where cached locally).
170 } BlueSkyCloudPointer;
173 LOGTYPE_INVALID = -1,
177 LOGTYPE_INODE_MAP = 3,
178 LOGTYPE_CHECKPOINT = 4,
179 LOGTYPE_CHECKPOINT_PTR = 5,
180 } BlueSkyCloudLogType;
182 /* A record which tracks an object which has been written to a local log,
183 * cached, locally, and/or written to the cloud. */
184 #define CLOUDLOG_JOURNAL 0x01
185 #define CLOUDLOG_CLOUD 0x02
186 #define CLOUDLOG_CACHE 0x04
187 struct _BlueSkyCloudLog {
194 BlueSkyCloudLogType type;
196 // Bitmask of CLOUDLOG_* flags indicating where the object exists.
198 int pending_read, pending_write;
200 // A stable identifier for the object (only changes when authenticated data
201 // is written out, but stays the same when the in-cloud cleaner relocates
205 // The inode which owns this data, if any, and an offset.
209 // The location of the object in the cloud, if available.
210 BlueSkyCloudPointer location;
212 // TODO: Location in journal/cache
213 int log_seq, log_offset, log_size;
215 // Pointers to other objects. Each link counts towards the reference count
216 // of the pointed-to object. To avoid memory leaks there should be no
217 // cycles in the reference graph.
220 // Serialized data, if available in memory (otherwise NULL), and a lock
221 // count which tracks if there are users that require the data to be kept
227 /* Serialize objects into a log segment to be written to the cloud. */
228 struct _BlueSkyCloudLogState {
230 BlueSkyCloudPointer location;
232 GSList *writeback_list; // Items which are being serialized right now
233 GList *pending_segments; // Segments which are being uploaded now
236 gboolean bluesky_cloudlog_equal(gconstpointer a, gconstpointer b);
237 guint bluesky_cloudlog_hash(gconstpointer a);
238 BlueSkyCloudLog *bluesky_cloudlog_new(BlueSkyFS *fs, const BlueSkyCloudID *id);
239 gchar *bluesky_cloudlog_id_to_string(BlueSkyCloudID id);
240 BlueSkyCloudID bluesky_cloudlog_id_from_string(const gchar *idstr);
241 void bluesky_cloudlog_ref(BlueSkyCloudLog *log);
242 void bluesky_cloudlog_unref(BlueSkyCloudLog *log);
243 void bluesky_cloudlog_stats_update(BlueSkyCloudLog *log, int type);
244 void bluesky_cloudlog_sync(BlueSkyCloudLog *log);
245 void bluesky_cloudlog_insert(BlueSkyCloudLog *log);
246 void bluesky_cloudlog_fetch(BlueSkyCloudLog *log);
247 BlueSkyCloudPointer bluesky_cloudlog_serialize(BlueSkyCloudLog *log,
249 void bluesky_cloudlog_flush(BlueSkyFS *fs);
251 /* Logging infrastructure for ensuring operations are persistently recorded to
253 #define BLUESKY_CRC32C_SEED (~(uint32_t)0)
254 #define BLUESKY_CRC32C_VALIDATOR ((uint32_t)0xb798b438UL)
255 uint32_t crc32c(uint32_t crc, const char *buf, unsigned int length);
256 uint32_t crc32c_finalize(uint32_t crc);
266 /* The currently-open log file. */
267 BlueSkyCacheFile *current_log;
269 /* Cache of log segments which have been memory-mapped. */
271 GHashTable *mmap_cache;
273 /* A count of the disk space consumed (in 1024-byte units) by all files
274 * tracked by mmap_cache (whether mapped or not, actually). */
277 /* The smallest journal sequence number which may still contain data that
278 * must be preserved (since it it not yet in the cloud). */
279 int journal_watermark;
282 /* An object for tracking log files which are stored locally--either the
283 * journal for filesystem consistency or log segments which have been fetched
284 * back from cloud storage. */
285 struct _BlueSkyCacheFile {
289 int type; // Only one of CLOUDLOG_{JOURNAL,CLOUD}
292 char *filename; // Local filename, relateive to log directory
293 gint mapcount; // References to the mmaped data
294 const char *addr; // May be null if data is not mapped in memory
298 gboolean fetching, ready; // Cloud data: downloading or ready for use
299 int64_t atime; // Access time, for cache management
302 BlueSkyLog *bluesky_log_new(const char *log_directory);
303 void bluesky_log_item_submit(BlueSkyCloudLog *item, BlueSkyLog *log);
304 void bluesky_log_finish_all(GList *log_items);
305 BlueSkyRCStr *bluesky_log_map_object(BlueSkyFS *fs, int log_dir, int log_seq,
306 int log_offset, int log_size);
307 void bluesky_mmap_unref(BlueSkyCacheFile *mmap);
308 void bluesky_cachefile_unref(BlueSkyCacheFile *cachefile);
310 BlueSkyCacheFile *bluesky_cachefile_lookup(BlueSkyFS *fs,
311 int clouddir, int log_seq);
312 void bluesky_cachefile_gc(BlueSkyFS *fs);
314 void bluesky_replay(BlueSkyFS *fs);
316 /* Used to track log segments that are being written to the cloud. */
325 /***** Inode map management *****/
327 /* Mapping information for a single inode number. These are grouped together
328 * into InodeMapRange objects. */
332 /* The ID of the most recent version of the inode. */
335 /* The location where that version is written in the cloud. */
336 BlueSkyCloudPointer location;
338 /* If the cloud log entry exists in memory, then a pointer to it, otherwise
340 BlueSkyCloudLog *item;
344 /* Starting and ending inode number values that fall in this section.
345 * Endpoint values are inclusive. */
348 /* A sorted list (by inode number) of InodeMapEntry objects. */
349 GSequence *map_entries;
351 /* The location where this inode map section is stored in the cloud. */
352 BlueSkyCloudPointer location;
354 /* Have there been changes that require writing this section out again? */
358 InodeMapEntry *bluesky_inode_map_lookup(GSequence *inode_map, uint64_t inum,