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>
21 /* Reference-counted blocks of memory, used for passing data in and out of
22 * storage backends and in other places. */
29 BlueSkyRCStr *bluesky_string_new(gpointer data, gsize len);
30 void bluesky_string_ref(BlueSkyRCStr *string);
31 void bluesky_string_unref(BlueSkyRCStr *string);
32 BlueSkyRCStr *bluesky_string_dup(BlueSkyRCStr *string);
34 /* Cryptographic operations. */
35 void bluesky_crypt_init();
36 void bluesky_crypt_random_bytes(guchar *buf, gint len);
37 BlueSkyRCStr *bluesky_crypt_encrypt(BlueSkyRCStr *in, const uint8_t *key);
38 BlueSkyRCStr *bluesky_crypt_decrypt(BlueSkyRCStr *in, const uint8_t *key);
40 /* File types. The numeric values are chosen to match with those used in
44 BLUESKY_DIRECTORY = 2,
46 BLUESKY_CHARACTER = 4,
52 /* Filesystem state. Each filesystem which is exported is represented by a
53 * single bluesky_fs structure in memory. */
57 gchar *name; /* Descriptive name for the filesystem */
58 GHashTable *inodes; /* Cached inodes */
59 uint64_t next_inum; /* Next available inode for allocation */
61 struct S3Store *store;
63 uint8_t *encryption_key;
66 /* Inode number of the root directory. */
67 #define BLUESKY_ROOT_INUM 1
69 /* Timestamp, measured in microseconds since the Unix epoch. */
70 typedef int64_t bluesky_time;
72 /* In-memory representation of an inode within a Blue Sky server. This
73 * corresponds roughly with information that is committed to persistent
74 * storage. Locking/refcounting rules:
75 * - To access or modify any data fields, the lock must be held. This
76 * includes file blocks.
77 * - One reference is held by the BlueSkyFS inode hash table. If that is the
78 * only reference (and the inode is unlocked), the inode is subject to
79 * dropping from the cache.
80 * - Any pending operations should hold extra references to the inode as
81 * appropriate to keep it available until the operation completes.
94 /* Rather than track an inode number and generation number, we will simply
95 * never re-use a fileid after a file is deleted. 64 bits should be enough
96 * that we don't exhaust the identifier space. */
99 uint64_t change_count; /* Incremented each with each change made */
100 int64_t atime; /* Microseconds since the Unix epoch */
103 int64_t ntime; /* "new time": time object was created */
105 /* File-specific fields */
109 /* Directory-specific fields */
110 GSequence *dirents; /* List of entries for READDIR */
111 GHashTable *dirhash; /* Hash table by name for LOOKUP */
112 uint64_t parent_inum; /* inode for ".."; 0 if the root directory */
115 /* A directory entry. The name is UTF-8 and is a freshly-allocated string.
116 * Each directory entry is listed in two indices: dirents is indexed by cookie
117 * and dirhash by name. The cookie is a randomly-assigned 32-bit value, unique
118 * within the directory, that remains unchanged until the entry is deleted. It
119 * is used to provide a stable key for restarting a READDIR call. */
126 /* File data is divided into fixed-size blocks (except the last block which may
127 * be short?). These blocks are backed by storage in a key/value store, but
128 * may also be dirty if modifications have been made in-memory that have not
130 #define BLUESKY_BLOCK_SIZE 32768ULL
131 #define BLUESKY_MAX_FILE_SIZE (BLUESKY_BLOCK_SIZE << 24)
133 BLUESKY_BLOCK_ZERO = 0, /* Data is all zeroes, not explicitly stored */
134 BLUESKY_BLOCK_REF = 1, /* Reference to key/value store, not cached */
135 BLUESKY_BLOCK_CACHED = 2, /* Data is cached in memory, clean */
136 BLUESKY_BLOCK_DIRTY = 3, /* Data needs to be committed to store */
140 BlueSkyBlockType type;
141 gchar *ref; /* Name of data block in the backing store */
142 BlueSkyRCStr *data; /* Pointer to data in memory if cached */
145 BlueSkyFS *bluesky_new_fs(gchar *name);
146 int64_t bluesky_get_current_time();
147 void bluesky_inode_update_ctime(BlueSkyInode *inode, gboolean update_mtime);
148 uint64_t bluesky_fs_alloc_inode(BlueSkyFS *fs);
149 BlueSkyInode *bluesky_new_inode(uint64_t inum, BlueSkyFS *fs, BlueSkyFileType type);
151 BlueSkyInode *bluesky_get_inode(BlueSkyFS *fs, uint64_t inum);
152 void bluesky_insert_inode(BlueSkyFS *fs, BlueSkyInode *inode);
154 void bluesky_dirent_destroy(gpointer dirent);
155 uint64_t bluesky_directory_hash(gchar *name);
156 uint64_t bluesky_directory_lookup(BlueSkyInode *inode, gchar *name);
157 gboolean bluesky_directory_insert(BlueSkyInode *dir, gchar *name,
159 void bluesky_directory_dump(BlueSkyInode *dir);
161 void bluesky_block_touch(BlueSkyInode *inode, uint64_t i);
162 void bluesky_block_fetch(BlueSkyFS *fs, BlueSkyBlock *block);
163 void bluesky_block_flush(BlueSkyFS *fs, BlueSkyBlock *block);
164 void bluesky_file_truncate(BlueSkyInode *inode, uint64_t size);
165 void bluesky_file_write(BlueSkyInode *inode, uint64_t offset,
166 const char *data, gint len);
167 void bluesky_file_read(BlueSkyInode *inode, uint64_t offset,
168 char *buf, gint len);
170 struct S3Store *s3store_new();
171 BlueSkyRCStr *s3store_get(struct S3Store *store, const gchar *key);
172 void s3store_put(struct S3Store *store, const gchar *key, BlueSkyRCStr *val);