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 /* BlueSky status and error codes. Various frontends should translate these to
21 * the appropriate error code for whatever protocol they implement. */
23 BSTATUS_OK = 0, /* No error */
24 BSTATUS_IOERR, /* I/O error of some form */
25 BSTATUS_NOENT, /* File does not exist */
28 void bluesky_init(void);
30 gchar *bluesky_lowercase(const gchar *s);
32 /* Reference-counted blocks of memory, used for passing data in and out of
33 * storage backends and in other places. */
40 BlueSkyRCStr *bluesky_string_new(gpointer data, gsize len);
41 BlueSkyRCStr *bluesky_string_new_from_gstring(GString *s);
42 void bluesky_string_ref(BlueSkyRCStr *string);
43 void bluesky_string_unref(BlueSkyRCStr *string);
44 BlueSkyRCStr *bluesky_string_dup(BlueSkyRCStr *string);
45 void bluesky_string_resize(BlueSkyRCStr *string, gsize len);
47 /* Cryptographic operations. */
48 #define CRYPTO_BLOCK_SIZE 16 /* 128-bit AES */
49 #define CRYPTO_KEY_SIZE 16
51 void bluesky_crypt_init();
52 void bluesky_crypt_hash_key(const char *keystr, uint8_t *out);
53 void bluesky_crypt_random_bytes(guchar *buf, gint len);
54 BlueSkyRCStr *bluesky_crypt_encrypt(BlueSkyRCStr *in, const uint8_t *key);
55 BlueSkyRCStr *bluesky_crypt_decrypt(BlueSkyRCStr *in, const uint8_t *key);
57 /* Storage interface. This presents a key-value store abstraction, and can
58 * have multiple implementations: in-memory, on-disk, in-cloud. */
60 /* Create a new store instance and return a handle to it. */
63 /* Clean up any resources used by this store. */
64 void (*destroy)(gpointer store);
66 /* Fetch an item with the given name, or return NULL if not found. */
67 BlueSkyRCStr * (*get)(gpointer store, const gchar *key);
69 /* Store an item to the given key name. */
70 void (*put)(gpointer store, const gchar *key, BlueSkyRCStr *val);
71 } BlueSkyStoreImplementation;
73 void bluesky_store_register(const BlueSkyStoreImplementation *impl,
77 typedef struct _BlueSkyStore BlueSkyStore;
79 void bluesky_store_init();
80 BlueSkyStore *bluesky_store_new(const gchar *type);
81 void bluesky_store_free(BlueSkyStore *store);
82 BlueSkyRCStr *bluesky_store_get(BlueSkyStore *store, const gchar *key);
83 void bluesky_store_put(BlueSkyStore *store,
84 const gchar *key, BlueSkyRCStr *val);
86 /* File types. The numeric values are chosen to match with those used in
90 BLUESKY_DIRECTORY = 2,
92 BLUESKY_CHARACTER = 4,
98 /* Filesystem state. Each filesystem which is exported is represented by a
99 * single bluesky_fs structure in memory. */
103 gchar *name; /* Descriptive name for the filesystem */
104 GHashTable *inodes; /* Cached inodes */
105 uint64_t next_inum; /* Next available inode for allocation */
110 /* Inode number of the root directory. */
111 #define BLUESKY_ROOT_INUM 1
113 /* Timestamp, measured in microseconds since the Unix epoch. */
114 typedef int64_t bluesky_time;
116 /* High-resolution timer, measured in nanoseconds. */
117 typedef int64_t bluesky_time_hires;
118 bluesky_time_hires bluesky_now_hires();
120 /* In-memory representation of an inode within a Blue Sky server. This
121 * corresponds roughly with information that is committed to persistent
122 * storage. Locking/refcounting rules:
123 * - To access or modify any data fields, the lock must be held. This
124 * includes file blocks.
125 * - One reference is held by the BlueSkyFS inode hash table. If that is the
126 * only reference (and the inode is unlocked), the inode is subject to
127 * dropping from the cache.
128 * - Any pending operations should hold extra references to the inode as
129 * appropriate to keep it available until the operation completes.
137 BlueSkyFileType type;
142 /* Rather than track an inode number and generation number, we will simply
143 * never re-use a fileid after a file is deleted. 64 bits should be enough
144 * that we don't exhaust the identifier space. */
147 /* change_count is increased with every operation which modifies the inode,
148 * and can be used to determine if cached data is still valid.
149 * change_commit is the value of change_count when the inode was last
150 * committed to stable storage. */
151 uint64_t change_count, change_commit;
153 int64_t atime; /* Microseconds since the Unix epoch */
156 int64_t ntime; /* "new time": time object was created */
158 /* File-specific fields */
162 /* Directory-specific fields */
163 GSequence *dirents; /* List of entries for READDIR */
164 GHashTable *dirhash; /* Hash table by name for LOOKUP */
165 GHashTable *dirhash_folded; /* As above, but case-folded */
166 uint64_t parent_inum; /* inode for ".."; 0 if the root directory */
169 /* A directory entry. The name is UTF-8 and is a freshly-allocated string.
170 * Each directory entry is listed in two indices: dirents is indexed by cookie
171 * and dirhash by name. The cookie is a randomly-assigned 32-bit value, unique
172 * within the directory, that remains unchanged until the entry is deleted. It
173 * is used to provide a stable key for restarting a READDIR call. */
176 gchar *name_folded; /* Name, folded for case-insensitive lookup */
181 /* File data is divided into fixed-size blocks (except the last block which may
182 * be short?). These blocks are backed by storage in a key/value store, but
183 * may also be dirty if modifications have been made in-memory that have not
185 #define BLUESKY_BLOCK_SIZE 32768ULL
186 #define BLUESKY_MAX_FILE_SIZE (BLUESKY_BLOCK_SIZE << 24)
188 BLUESKY_BLOCK_ZERO = 0, /* Data is all zeroes, not explicitly stored */
189 BLUESKY_BLOCK_REF = 1, /* Reference to key/value store, not cached */
190 BLUESKY_BLOCK_CACHED = 2, /* Data is cached in memory, clean */
191 BLUESKY_BLOCK_DIRTY = 3, /* Data needs to be committed to store */
195 BlueSkyBlockType type;
196 gchar *ref; /* Name of data block in the backing store */
197 BlueSkyRCStr *data; /* Pointer to data in memory if cached */
200 BlueSkyFS *bluesky_init_fs(gchar *name, BlueSkyStore *store);
201 void bluesky_superblock_flush(BlueSkyFS *fs);
203 int64_t bluesky_get_current_time();
204 void bluesky_inode_update_ctime(BlueSkyInode *inode, gboolean update_mtime);
205 uint64_t bluesky_fs_alloc_inode(BlueSkyFS *fs);
206 BlueSkyInode *bluesky_new_inode(uint64_t inum, BlueSkyFS *fs, BlueSkyFileType type);
208 BlueSkyInode *bluesky_get_inode(BlueSkyFS *fs, uint64_t inum);
209 void bluesky_insert_inode(BlueSkyFS *fs, BlueSkyInode *inode);
211 void bluesky_dirent_destroy(gpointer dirent);
212 uint64_t bluesky_directory_lookup(BlueSkyInode *inode, gchar *name);
213 uint64_t bluesky_directory_ilookup(BlueSkyInode *inode, gchar *name);
214 BlueSkyDirent *bluesky_directory_read(BlueSkyInode *dir, uint32_t cookie);
215 gboolean bluesky_directory_insert(BlueSkyInode *dir, const gchar *name,
217 void bluesky_directory_dump(BlueSkyInode *dir);
219 void bluesky_block_touch(BlueSkyInode *inode, uint64_t i);
220 void bluesky_block_fetch(BlueSkyFS *fs, BlueSkyBlock *block);
221 void bluesky_block_flush(BlueSkyFS *fs, BlueSkyBlock *block);
222 void bluesky_file_truncate(BlueSkyInode *inode, uint64_t size);
223 void bluesky_file_write(BlueSkyInode *inode, uint64_t offset,
224 const char *data, gint len);
225 void bluesky_file_read(BlueSkyInode *inode, uint64_t offset,
226 char *buf, gint len);
228 void bluesky_inode_flush(BlueSkyFS *fs, BlueSkyInode *inode);
229 void bluesky_inode_fetch(BlueSkyFS *fs, uint64_t inum);
231 gint bluesky_dirent_compare(gconstpointer a, gconstpointer b,