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>
19 void bluesky_init(void);
21 gchar *bluesky_lowercase(const gchar *s);
23 /* Reference-counted blocks of memory, used for passing data in and out of
24 * storage backends and in other places. */
31 BlueSkyRCStr *bluesky_string_new(gpointer data, gsize len);
32 void bluesky_string_ref(BlueSkyRCStr *string);
33 void bluesky_string_unref(BlueSkyRCStr *string);
34 BlueSkyRCStr *bluesky_string_dup(BlueSkyRCStr *string);
36 /* Cryptographic operations. */
37 #define CRYPTO_BLOCK_SIZE 16 /* 128-bit AES */
38 #define CRYPTO_KEY_SIZE 16
40 void bluesky_crypt_init();
41 void bluesky_crypt_random_bytes(guchar *buf, gint len);
42 BlueSkyRCStr *bluesky_crypt_encrypt(BlueSkyRCStr *in, const uint8_t *key);
43 BlueSkyRCStr *bluesky_crypt_decrypt(BlueSkyRCStr *in, const uint8_t *key);
45 /* Storage interface. This presents a key-value store abstraction, and can
46 * have multiple implementations: in-memory, on-disk, in-cloud. */
48 /* Create a new store instance and return a handle to it. */
51 /* Clean up any resources used by this store. */
52 void (*destroy)(gpointer store);
54 /* Fetch an item with the given name, or return NULL if not found. */
55 BlueSkyRCStr * (*get)(gpointer store, const gchar *key);
57 /* Store an item to the given key name. */
58 void (*put)(gpointer store, const gchar *key, BlueSkyRCStr *val);
59 } BlueSkyStoreImplementation;
61 void bluesky_store_register(const BlueSkyStoreImplementation *impl,
65 typedef struct _BlueSkyStore BlueSkyStore;
67 void bluesky_store_init();
68 BlueSkyStore *bluesky_store_new(const gchar *type);
69 void bluesky_store_free(BlueSkyStore *store);
70 BlueSkyRCStr *bluesky_store_get(BlueSkyStore *store, const gchar *key);
71 void bluesky_store_put(BlueSkyStore *store,
72 const gchar *key, BlueSkyRCStr *val);
74 /* File types. The numeric values are chosen to match with those used in
78 BLUESKY_DIRECTORY = 2,
80 BLUESKY_CHARACTER = 4,
86 /* Filesystem state. Each filesystem which is exported is represented by a
87 * single bluesky_fs structure in memory. */
91 gchar *name; /* Descriptive name for the filesystem */
92 GHashTable *inodes; /* Cached inodes */
93 uint64_t next_inum; /* Next available inode for allocation */
97 uint8_t *encryption_key;
100 /* Inode number of the root directory. */
101 #define BLUESKY_ROOT_INUM 1
103 /* Timestamp, measured in microseconds since the Unix epoch. */
104 typedef int64_t bluesky_time;
106 /* In-memory representation of an inode within a Blue Sky server. This
107 * corresponds roughly with information that is committed to persistent
108 * storage. Locking/refcounting rules:
109 * - To access or modify any data fields, the lock must be held. This
110 * includes file blocks.
111 * - One reference is held by the BlueSkyFS inode hash table. If that is the
112 * only reference (and the inode is unlocked), the inode is subject to
113 * dropping from the cache.
114 * - Any pending operations should hold extra references to the inode as
115 * appropriate to keep it available until the operation completes.
123 BlueSkyFileType type;
128 /* Rather than track an inode number and generation number, we will simply
129 * never re-use a fileid after a file is deleted. 64 bits should be enough
130 * that we don't exhaust the identifier space. */
133 /* change_count is increased with every operation which modifies the inode,
134 * and can be used to determine if cached data is still valid.
135 * change_commit is the value of change_count when the inode was last
136 * committed to stable storage. */
137 uint64_t change_count, change_commit;
139 int64_t atime; /* Microseconds since the Unix epoch */
142 int64_t ntime; /* "new time": time object was created */
144 /* File-specific fields */
148 /* Directory-specific fields */
149 GSequence *dirents; /* List of entries for READDIR */
150 GHashTable *dirhash; /* Hash table by name for LOOKUP */
151 uint64_t parent_inum; /* inode for ".."; 0 if the root directory */
154 /* A directory entry. The name is UTF-8 and is a freshly-allocated string.
155 * Each directory entry is listed in two indices: dirents is indexed by cookie
156 * and dirhash by name. The cookie is a randomly-assigned 32-bit value, unique
157 * within the directory, that remains unchanged until the entry is deleted. It
158 * is used to provide a stable key for restarting a READDIR call. */
165 /* File data is divided into fixed-size blocks (except the last block which may
166 * be short?). These blocks are backed by storage in a key/value store, but
167 * may also be dirty if modifications have been made in-memory that have not
169 #define BLUESKY_BLOCK_SIZE 32768ULL
170 #define BLUESKY_MAX_FILE_SIZE (BLUESKY_BLOCK_SIZE << 24)
172 BLUESKY_BLOCK_ZERO = 0, /* Data is all zeroes, not explicitly stored */
173 BLUESKY_BLOCK_REF = 1, /* Reference to key/value store, not cached */
174 BLUESKY_BLOCK_CACHED = 2, /* Data is cached in memory, clean */
175 BLUESKY_BLOCK_DIRTY = 3, /* Data needs to be committed to store */
179 BlueSkyBlockType type;
180 gchar *ref; /* Name of data block in the backing store */
181 BlueSkyRCStr *data; /* Pointer to data in memory if cached */
184 BlueSkyFS *bluesky_new_fs(gchar *name);
185 int64_t bluesky_get_current_time();
186 void bluesky_inode_update_ctime(BlueSkyInode *inode, gboolean update_mtime);
187 uint64_t bluesky_fs_alloc_inode(BlueSkyFS *fs);
188 BlueSkyInode *bluesky_new_inode(uint64_t inum, BlueSkyFS *fs, BlueSkyFileType type);
190 BlueSkyInode *bluesky_get_inode(BlueSkyFS *fs, uint64_t inum);
191 void bluesky_insert_inode(BlueSkyFS *fs, BlueSkyInode *inode);
193 void bluesky_dirent_destroy(gpointer dirent);
194 uint64_t bluesky_directory_hash(gchar *name);
195 uint64_t bluesky_directory_lookup(BlueSkyInode *inode, gchar *name);
196 gboolean bluesky_directory_insert(BlueSkyInode *dir, gchar *name,
198 void bluesky_directory_dump(BlueSkyInode *dir);
200 void bluesky_block_touch(BlueSkyInode *inode, uint64_t i);
201 void bluesky_block_fetch(BlueSkyFS *fs, BlueSkyBlock *block);
202 void bluesky_block_flush(BlueSkyFS *fs, BlueSkyBlock *block);
203 void bluesky_file_truncate(BlueSkyInode *inode, uint64_t size);
204 void bluesky_file_write(BlueSkyInode *inode, uint64_t offset,
205 const char *data, gint len);
206 void bluesky_file_read(BlueSkyInode *inode, uint64_t offset,
207 char *buf, gint len);
209 void bluesky_inode_flush(BlueSkyFS *fs, BlueSkyInode *inode);
210 void bluesky_inode_fetch(BlueSkyFS *fs, uint64_t inum);
211 void bluesky_serialize_inode(GString *out, BlueSkyInode *inode);
212 BlueSkyInode *bluesky_deserialize_inode(BlueSkyFS *fs, const gchar *buf);
214 gint bluesky_dirent_compare(gconstpointer a, gconstpointer b,