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 /* Various options to tweak for performance benchmarking purposes. */
22 /* Perform all get/put operations synchronously. */
23 int synchronous_stores;
26 extern BlueSkyOptions bluesky_options;
28 /* BlueSky status and error codes. Various frontends should translate these to
29 * the appropriate error code for whatever protocol they implement. */
31 BSTATUS_OK = 0, /* No error */
32 BSTATUS_IOERR, /* I/O error of some form */
33 BSTATUS_NOENT, /* File does not exist */
36 void bluesky_init(void);
38 gchar *bluesky_lowercase(const gchar *s);
40 /* Reference-counted blocks of memory, used for passing data in and out of
41 * storage backends and in other places. */
48 BlueSkyRCStr *bluesky_string_new(gpointer data, gsize len);
49 BlueSkyRCStr *bluesky_string_new_from_gstring(GString *s);
50 void bluesky_string_ref(BlueSkyRCStr *string);
51 void bluesky_string_unref(BlueSkyRCStr *string);
52 BlueSkyRCStr *bluesky_string_dup(BlueSkyRCStr *string);
53 void bluesky_string_resize(BlueSkyRCStr *string, gsize len);
55 /* Cryptographic operations. */
56 #define CRYPTO_BLOCK_SIZE 16 /* 128-bit AES */
57 #define CRYPTO_KEY_SIZE 16
59 void bluesky_crypt_init();
60 void bluesky_crypt_hash_key(const char *keystr, uint8_t *out);
61 void bluesky_crypt_random_bytes(guchar *buf, gint len);
62 BlueSkyRCStr *bluesky_crypt_encrypt(BlueSkyRCStr *in, const uint8_t *key);
63 BlueSkyRCStr *bluesky_crypt_decrypt(BlueSkyRCStr *in, const uint8_t *key);
65 /* Storage interface. This presents a key-value store abstraction, and can
66 * have multiple implementations: in-memory, on-disk, in-cloud. */
68 typedef struct _BlueSkyStore BlueSkyStore;
70 void bluesky_store_init();
71 BlueSkyStore *bluesky_store_new(const gchar *type);
72 void bluesky_store_free(BlueSkyStore *store);
73 BlueSkyRCStr *bluesky_store_get(BlueSkyStore *store, const gchar *key);
74 void bluesky_store_put(BlueSkyStore *store,
75 const gchar *key, BlueSkyRCStr *val);
77 /* File types. The numeric values are chosen to match with those used in
81 BLUESKY_DIRECTORY = 2,
83 BLUESKY_CHARACTER = 4,
89 /* Filesystem state. Each filesystem which is exported is represented by a
90 * single bluesky_fs structure in memory. */
94 gchar *name; /* Descriptive name for the filesystem */
95 GHashTable *inodes; /* Cached inodes */
96 uint64_t next_inum; /* Next available inode for allocation */
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 /* Timestamp for controlling when modified data is flushed to stable
145 * storage. When an inode is first modified from a clean state, this is
146 * set to the current time. If the inode is clean, it is set to zero. */
149 /* Additional state for tracking cache writeback status. */
150 uint64_t change_pending; /* change_count version currently being committed to storage */
152 int64_t atime; /* Microseconds since the Unix epoch */
155 int64_t ntime; /* "new time": time object was created */
157 /* File-specific fields */
161 /* Directory-specific fields */
162 GSequence *dirents; /* List of entries for READDIR */
163 GHashTable *dirhash; /* Hash table by name for LOOKUP */
164 GHashTable *dirhash_folded; /* As above, but case-folded */
165 uint64_t parent_inum; /* inode for ".."; 0 if the root directory */
168 /* A directory entry. The name is UTF-8 and is a freshly-allocated string.
169 * Each directory entry is listed in two indices: dirents is indexed by cookie
170 * and dirhash by name. The cookie is a randomly-assigned 32-bit value, unique
171 * within the directory, that remains unchanged until the entry is deleted. It
172 * is used to provide a stable key for restarting a READDIR call. */
175 gchar *name_folded; /* Name, folded for case-insensitive lookup */
180 /* File data is divided into fixed-size blocks (except the last block which may
181 * be short?). These blocks are backed by storage in a key/value store, but
182 * may also be dirty if modifications have been made in-memory that have not
184 #define BLUESKY_BLOCK_SIZE 32768ULL
185 #define BLUESKY_MAX_FILE_SIZE (BLUESKY_BLOCK_SIZE << 24)
187 BLUESKY_BLOCK_ZERO = 0, /* Data is all zeroes, not explicitly stored */
188 BLUESKY_BLOCK_REF = 1, /* Reference to key/value store, not cached */
189 BLUESKY_BLOCK_CACHED = 2, /* Data is cached in memory, clean */
190 BLUESKY_BLOCK_DIRTY = 3, /* Data needs to be committed to store */
194 BlueSkyBlockType type;
195 gchar *ref; /* Name of data block in the backing store */
196 BlueSkyRCStr *data; /* Pointer to data in memory if cached */
199 BlueSkyFS *bluesky_init_fs(gchar *name, BlueSkyStore *store);
200 void bluesky_superblock_flush(BlueSkyFS *fs);
202 int64_t bluesky_get_current_time();
203 void bluesky_inode_update_ctime(BlueSkyInode *inode, gboolean update_mtime);
204 uint64_t bluesky_fs_alloc_inode(BlueSkyFS *fs);
205 BlueSkyInode *bluesky_new_inode(uint64_t inum, BlueSkyFS *fs, BlueSkyFileType type);
207 BlueSkyInode *bluesky_get_inode(BlueSkyFS *fs, uint64_t inum);
208 void bluesky_insert_inode(BlueSkyFS *fs, BlueSkyInode *inode);
210 void bluesky_dirent_destroy(gpointer dirent);
211 uint64_t bluesky_directory_lookup(BlueSkyInode *inode, gchar *name);
212 uint64_t bluesky_directory_ilookup(BlueSkyInode *inode, gchar *name);
213 BlueSkyDirent *bluesky_directory_read(BlueSkyInode *dir, uint32_t cookie);
214 gboolean bluesky_directory_insert(BlueSkyInode *dir, const gchar *name,
216 void bluesky_directory_dump(BlueSkyInode *dir);
218 void bluesky_block_touch(BlueSkyInode *inode, uint64_t i);
219 void bluesky_block_fetch(BlueSkyFS *fs, BlueSkyBlock *block);
220 void bluesky_block_flush(BlueSkyFS *fs, BlueSkyBlock *block);
221 void bluesky_file_flush(BlueSkyInode *inode);
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,
234 void bluesky_flushd_invoke(BlueSkyFS *fs);