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>
20 void bluesky_init(void);
22 gchar *bluesky_lowercase(const gchar *s);
24 /* Reference-counted blocks of memory, used for passing data in and out of
25 * storage backends and in other places. */
32 BlueSkyRCStr *bluesky_string_new(gpointer data, gsize len);
33 void bluesky_string_ref(BlueSkyRCStr *string);
34 void bluesky_string_unref(BlueSkyRCStr *string);
35 BlueSkyRCStr *bluesky_string_dup(BlueSkyRCStr *string);
37 /* Cryptographic operations. */
38 #define CRYPTO_BLOCK_SIZE 16 /* 128-bit AES */
39 #define CRYPTO_KEY_SIZE 16
41 void bluesky_crypt_init();
42 void bluesky_crypt_random_bytes(guchar *buf, gint len);
43 BlueSkyRCStr *bluesky_crypt_encrypt(BlueSkyRCStr *in, const uint8_t *key);
44 BlueSkyRCStr *bluesky_crypt_decrypt(BlueSkyRCStr *in, const uint8_t *key);
46 /* Storage interface. This presents a key-value store abstraction, and can
47 * have multiple implementations: in-memory, on-disk, in-cloud. */
49 /* Create a new store instance and return a handle to it. */
52 /* Clean up any resources used by this store. */
53 void (*destroy)(gpointer store);
55 /* Fetch an item with the given name, or return NULL if not found. */
56 BlueSkyRCStr * (*get)(gpointer store, const gchar *key);
58 /* Store an item to the given key name. */
59 void (*put)(gpointer store, const gchar *key, BlueSkyRCStr *val);
60 } BlueSkyStoreImplementation;
62 void bluesky_store_register(const BlueSkyStoreImplementation *impl,
66 typedef struct _BlueSkyStore BlueSkyStore;
68 void bluesky_store_init();
69 BlueSkyStore *bluesky_store_new(const gchar *type);
70 void bluesky_store_free(BlueSkyStore *store);
71 BlueSkyRCStr *bluesky_store_get(BlueSkyStore *store, const gchar *key);
72 void bluesky_store_put(BlueSkyStore *store,
73 const gchar *key, BlueSkyRCStr *val);
75 /* File types. The numeric values are chosen to match with those used in
79 BLUESKY_DIRECTORY = 2,
81 BLUESKY_CHARACTER = 4,
87 /* Filesystem state. Each filesystem which is exported is represented by a
88 * single bluesky_fs structure in memory. */
92 gchar *name; /* Descriptive name for the filesystem */
93 GHashTable *inodes; /* Cached inodes */
94 uint64_t next_inum; /* Next available inode for allocation */
98 uint8_t *encryption_key;
101 /* Inode number of the root directory. */
102 #define BLUESKY_ROOT_INUM 1
104 /* Timestamp, measured in microseconds since the Unix epoch. */
105 typedef int64_t bluesky_time;
107 /* High-resolution timer, measured in nanoseconds. */
108 typedef int64_t bluesky_time_hires;
109 bluesky_time_hires bluesky_now_hires();
111 /* In-memory representation of an inode within a Blue Sky server. This
112 * corresponds roughly with information that is committed to persistent
113 * storage. Locking/refcounting rules:
114 * - To access or modify any data fields, the lock must be held. This
115 * includes file blocks.
116 * - One reference is held by the BlueSkyFS inode hash table. If that is the
117 * only reference (and the inode is unlocked), the inode is subject to
118 * dropping from the cache.
119 * - Any pending operations should hold extra references to the inode as
120 * appropriate to keep it available until the operation completes.
128 BlueSkyFileType type;
133 /* Rather than track an inode number and generation number, we will simply
134 * never re-use a fileid after a file is deleted. 64 bits should be enough
135 * that we don't exhaust the identifier space. */
138 /* change_count is increased with every operation which modifies the inode,
139 * and can be used to determine if cached data is still valid.
140 * change_commit is the value of change_count when the inode was last
141 * committed to stable storage. */
142 uint64_t change_count, change_commit;
144 int64_t atime; /* Microseconds since the Unix epoch */
147 int64_t ntime; /* "new time": time object was created */
149 /* File-specific fields */
153 /* Directory-specific fields */
154 GSequence *dirents; /* List of entries for READDIR */
155 GHashTable *dirhash; /* Hash table by name for LOOKUP */
156 GHashTable *dirhash_folded; /* As above, but case-folded */
157 uint64_t parent_inum; /* inode for ".."; 0 if the root directory */
160 /* A directory entry. The name is UTF-8 and is a freshly-allocated string.
161 * Each directory entry is listed in two indices: dirents is indexed by cookie
162 * and dirhash by name. The cookie is a randomly-assigned 32-bit value, unique
163 * within the directory, that remains unchanged until the entry is deleted. It
164 * is used to provide a stable key for restarting a READDIR call. */
167 gchar *name_folded; /* Name, folded for case-insensitive lookup */
172 /* File data is divided into fixed-size blocks (except the last block which may
173 * be short?). These blocks are backed by storage in a key/value store, but
174 * may also be dirty if modifications have been made in-memory that have not
176 #define BLUESKY_BLOCK_SIZE 32768ULL
177 #define BLUESKY_MAX_FILE_SIZE (BLUESKY_BLOCK_SIZE << 24)
179 BLUESKY_BLOCK_ZERO = 0, /* Data is all zeroes, not explicitly stored */
180 BLUESKY_BLOCK_REF = 1, /* Reference to key/value store, not cached */
181 BLUESKY_BLOCK_CACHED = 2, /* Data is cached in memory, clean */
182 BLUESKY_BLOCK_DIRTY = 3, /* Data needs to be committed to store */
186 BlueSkyBlockType type;
187 gchar *ref; /* Name of data block in the backing store */
188 BlueSkyRCStr *data; /* Pointer to data in memory if cached */
191 BlueSkyFS *bluesky_new_fs(gchar *name);
192 BlueSkyFS *bluesky_init_fs(gchar *name, BlueSkyStore *store);
193 void bluesky_superblock_flush(BlueSkyFS *fs);
195 int64_t bluesky_get_current_time();
196 void bluesky_inode_update_ctime(BlueSkyInode *inode, gboolean update_mtime);
197 uint64_t bluesky_fs_alloc_inode(BlueSkyFS *fs);
198 BlueSkyInode *bluesky_new_inode(uint64_t inum, BlueSkyFS *fs, BlueSkyFileType type);
200 BlueSkyInode *bluesky_get_inode(BlueSkyFS *fs, uint64_t inum);
201 void bluesky_insert_inode(BlueSkyFS *fs, BlueSkyInode *inode);
203 void bluesky_dirent_destroy(gpointer dirent);
204 uint64_t bluesky_directory_hash(gchar *name);
205 uint64_t bluesky_directory_lookup(BlueSkyInode *inode, gchar *name);
206 uint64_t bluesky_directory_ilookup(BlueSkyInode *inode, gchar *name);
207 gboolean bluesky_directory_insert(BlueSkyInode *dir, const gchar *name,
209 void bluesky_directory_dump(BlueSkyInode *dir);
211 void bluesky_block_touch(BlueSkyInode *inode, uint64_t i);
212 void bluesky_block_fetch(BlueSkyFS *fs, BlueSkyBlock *block);
213 void bluesky_block_flush(BlueSkyFS *fs, BlueSkyBlock *block);
214 void bluesky_file_truncate(BlueSkyInode *inode, uint64_t size);
215 void bluesky_file_write(BlueSkyInode *inode, uint64_t offset,
216 const char *data, gint len);
217 void bluesky_file_read(BlueSkyInode *inode, uint64_t offset,
218 char *buf, gint len);
220 void bluesky_inode_flush(BlueSkyFS *fs, BlueSkyInode *inode);
221 void bluesky_inode_fetch(BlueSkyFS *fs, uint64_t inum);
223 void bluesky_serialize_superblock(GString *out, BlueSkyFS *fs);
224 BlueSkyFS *bluesky_deserialize_superblock(const gchar *buf);
225 void bluesky_serialize_inode(GString *out, BlueSkyInode *inode);
226 BlueSkyInode *bluesky_deserialize_inode(BlueSkyFS *fs, const gchar *buf);
228 gint bluesky_dirent_compare(gconstpointer a, gconstpointer b,