+++ /dev/null
-/* Blue Sky: File Systems in the Cloud
- *
- * Copyright (C) 2009 The Regents of the University of California
- * Written by Michael Vrable <mvrable@cs.ucsd.edu>
- *
- * TODO: Licensing
- */
-
-#include <stdint.h>
-#include <glib.h>
-#include <string.h>
-
-#include "bluesky.h"
-
-/* Core filesystem. Different proxies, such as the NFSv3 one, interface to
- * this, but the core actually tracks the data which is stored. So far we just
- * implement an in-memory filesystem, but eventually this will be state which
- * is persisted to the cloud. */
-
-/* Return the current time, in microseconds since the epoch. */
-int64_t bluesky_get_current_time()
-{
- GTimeVal t;
- g_get_current_time(&t);
- return (int64_t)t.tv_sec * 1000000 + t.tv_usec;
-}
-
-/* Update an inode to indicate that a modification was made. This increases
- * the change counter, updates the ctime to the current time, and optionally
- * updates the mtime. */
-void bluesky_inode_update_ctime(BlueSkyInode *inode, gboolean update_mtime)
-{
- int64_t now = bluesky_get_current_time();
- inode->change_count++;
- inode->ctime = now;
- if (update_mtime)
- inode->mtime = now;
-}
-
-/* Compute the HMAC keyed-hash function using the given hash algorithm, data,
- * and key. */
-void compute_hmac(GChecksumType algo,
- const guchar *data, gsize data_len,
- const guchar *key, gsize key_len,
- guint8 *buffer, gsize *digest_len)
-{
- int block_size;
-
- switch (algo) {
- case G_CHECKSUM_MD5:
- case G_CHECKSUM_SHA1:
- case G_CHECKSUM_SHA256:
- block_size = 64;
- break;
- default:
- g_error("Unknown hash algorithm for HMAC: %d\n", algo);
- }
-
- gsize digest_size = g_checksum_type_get_length(algo);
-
- guchar keybuf[block_size];
- memset(keybuf, 0, block_size);
- memcpy(keybuf, key, MIN(block_size, key_len));
- for (int i = 0; i < block_size; i++)
- keybuf[i] ^= 0x36;
-
- GChecksum *csum1 = g_checksum_new(algo);
- g_checksum_update(csum1, keybuf, block_size);
- g_checksum_update(csum1, data, data_len);
- guint8 digest[digest_size];
- g_checksum_get_digest(csum1, digest, &digest_size);
-
- memset(keybuf, 0, block_size);
- memcpy(keybuf, key, MIN(block_size, key_len));
- for (int i = 0; i < block_size; i++)
- keybuf[i] ^= 0x5c;
-
- GChecksum *csum2 = g_checksum_new(algo);
- g_checksum_update(csum2, keybuf, block_size);
- g_checksum_update(csum2, digest, digest_size);
-
- g_checksum_get_digest(csum2, buffer, digest_len);
-
- g_checksum_free(csum1);
- g_checksum_free(csum2);
-}
-
-/* Unfortunately a glib hash table is only guaranteed to be able to store
- * 32-bit keys if we use the key directly. If we want 64-bit inode numbers,
- * we'll have to allocate memory to store the 64-bit inumber, and use a pointer
- * to it. Rather than allocate the memory for the key, we'll just include a
- * pointer to the 64-bit inum stored in the inode itself, so that we don't need
- * to do any more memory management. */
-static guint bluesky_fs_key_hash_func(gconstpointer key)
-{
- uint64_t inum = *(const uint64_t *)key;
- return (guint)inum;
-}
-
-static gboolean bluesky_fs_key_equal_func(gconstpointer a, gconstpointer b)
-{
- uint64_t i1 = *(const uint64_t *)a;
- uint64_t i2 = *(const uint64_t *)b;
- return i1 == i2;
-}
-
-/* Filesystem-level operations. A filesystem is like a directory tree that we
- * are willing to export. */
-BlueSkyFS *bluesky_new_fs(gchar *name)
-{
- BlueSkyFS *fs = g_new0(BlueSkyFS, 1);
- fs->lock = g_mutex_new();
- fs->name = g_strdup(name);
- fs->inodes = g_hash_table_new(bluesky_fs_key_hash_func,
- bluesky_fs_key_equal_func);
- fs->next_inum = BLUESKY_ROOT_INUM + 1;
- fs->store = s3store_new();
-
- return fs;
-}
-
-/* Allocate a fresh inode number which has not been used before within a
- * filesystem. */
-uint64_t bluesky_fs_alloc_inode(BlueSkyFS *fs)
-{
- uint64_t inum;
-
- g_mutex_lock(fs->lock);
- inum = fs->next_inum;
- fs->next_inum++;
- g_mutex_unlock(fs->lock);
-
- return inum;
-}
-
-BlueSkyInode *bluesky_new_inode(uint64_t inum, BlueSkyFS *fs,
- BlueSkyFileType type)
-{
- BlueSkyInode *i = g_new0(BlueSkyInode, 1);
-
- i->lock = g_mutex_new();
- i->type = type;
- i->fs = fs;
- i->inum = inum;
-
- switch (type) {
- case BLUESKY_REGULAR:
- i->blocks = g_array_new(FALSE, TRUE, sizeof(BlueSkyBlock));
- break;
- case BLUESKY_DIRECTORY:
- i->dirents = g_sequence_new(bluesky_dirent_destroy);
- i->dirhash = g_hash_table_new(g_str_hash, g_str_equal);
- break;
- case BLUESKY_BLOCK:
- case BLUESKY_CHARACTER:
- case BLUESKY_SYMLINK:
- case BLUESKY_SOCKET:
- case BLUESKY_FIFO:
- break;
- }
-
- return i;
-}
-
-/* Retrieve an inode from the filesystem. Eventually this will be a cache and
- * so we might need to go fetch the inode from elsewhere; for now all
- * filesystem state is stored here. */
-BlueSkyInode *bluesky_get_inode(BlueSkyFS *fs, uint64_t inum)
-{
- BlueSkyInode *inode = NULL;
-
- g_mutex_lock(fs->lock);
- inode = (BlueSkyInode *)g_hash_table_lookup(fs->inodes, &inum);
- g_mutex_unlock(fs->lock);
-
- return inode;
-}
-
-/* Insert an inode into the filesystem inode cache. */
-void bluesky_insert_inode(BlueSkyFS *fs, BlueSkyInode *inode)
-{
- g_mutex_lock(fs->lock);
- g_hash_table_insert(fs->inodes, &inode->inum, inode);
- g_mutex_unlock(fs->lock);
-}
-
-/* Mark a given block dirty and make sure that data is faulted in so that it
- * can be written to. */
-void bluesky_block_touch(BlueSkyInode *inode, uint64_t i)
-{
- g_return_if_fail(i < inode->blocks->len);
- BlueSkyBlock *block = &g_array_index(inode->blocks, BlueSkyBlock, i);
-
- switch (block->type) {
- case BLUESKY_BLOCK_ZERO:
- block->data = bluesky_string_new(g_malloc0(BLUESKY_BLOCK_SIZE),
- BLUESKY_BLOCK_SIZE);
- break;
- case BLUESKY_BLOCK_REF:
- bluesky_block_fetch(inode->fs, block);
- g_assert(block->type == BLUESKY_BLOCK_CACHED);
- /* Fall through */
- case BLUESKY_BLOCK_CACHED:
- case BLUESKY_BLOCK_DIRTY:
- block->data = bluesky_string_dup(block->data);
- break;
- }
-
- block->type = BLUESKY_BLOCK_DIRTY;
-}
-
-/* Set the size of a file. This will truncate or extend the file as needed.
- * Newly-allocated bytes are zeroed. */
-void bluesky_file_truncate(BlueSkyInode *inode, uint64_t size)
-{
- g_return_if_fail(size <= BLUESKY_MAX_FILE_SIZE);
-
- if (size == inode->size)
- return;
-
- uint64_t blocks = (size + BLUESKY_BLOCK_SIZE - 1) / BLUESKY_BLOCK_SIZE;
-
- if (blocks > inode->blocks->len) {
- /* Need to add new blocks to the end of a file. New block structures
- * are automatically zeroed, which initializes them to be pointers to
- * zero blocks so we don't need to do any more work. */
- g_array_set_size(inode->blocks, blocks);
- } else if (blocks < inode->blocks->len) {
- /* Delete blocks from a file. Must reclaim memory. */
- for (guint i = inode->blocks->len; i < blocks; i++) {
- BlueSkyBlock *b = &g_array_index(inode->blocks, BlueSkyBlock, i);
- g_free(b->ref);
- bluesky_string_unref(b->data);
- }
- g_array_set_size(inode->blocks, blocks);
- }
-
- /* If the file size is being decreased, ensure that any trailing data in
- * the last block is zeroed. */
- if (size < inode->size) {
- BlueSkyBlock *b = &g_array_index(inode->blocks, BlueSkyBlock,
- blocks - 1);
- if (b->type != BLUESKY_BLOCK_ZERO) {
- bluesky_block_touch(inode, blocks - 1);
- int end_offset = size % BLUESKY_BLOCK_SIZE;
- if (end_offset > 0) {
- memset(&b->data->data[end_offset], 0,
- BLUESKY_BLOCK_SIZE - end_offset);
- }
- }
- }
-
- inode->size = size;
- bluesky_inode_update_ctime(inode, 1);
-}
-
-void bluesky_file_write(BlueSkyInode *inode, uint64_t offset,
- const char *data, gint len)
-{
- g_return_if_fail(inode->type == BLUESKY_REGULAR);
- g_return_if_fail(offset < inode->size);
- g_return_if_fail(len <= inode->size - offset);
-
- if (len == 0)
- return;
-
- while (len > 0) {
- uint64_t block_num = offset / BLUESKY_BLOCK_SIZE;
- gint block_offset = offset % BLUESKY_BLOCK_SIZE;
- gint bytes = MIN(BLUESKY_BLOCK_SIZE - block_offset, len);
-
- bluesky_block_touch(inode, block_num);
- BlueSkyBlock *b = &g_array_index(inode->blocks, BlueSkyBlock,
- block_num);
- memcpy(&b->data->data[block_offset], data, bytes);
- bluesky_block_flush(inode->fs, b);
-
- offset += bytes;
- data += bytes;
- len -= bytes;
- }
-
- bluesky_inode_update_ctime(inode, 1);
-}
-
-void bluesky_file_read(BlueSkyInode *inode, uint64_t offset,
- char *buf, gint len)
-{
- g_return_if_fail(inode->type == BLUESKY_REGULAR);
- g_return_if_fail(offset < inode->size);
- g_return_if_fail(len <= inode->size - offset);
-
- while (len > 0) {
- uint64_t block_num = offset / BLUESKY_BLOCK_SIZE;
- gint block_offset = offset % BLUESKY_BLOCK_SIZE;
- gint bytes = MIN(BLUESKY_BLOCK_SIZE - block_offset, len);
-
- BlueSkyBlock *b = &g_array_index(inode->blocks, BlueSkyBlock,
- block_num);
- switch (b->type) {
- case BLUESKY_BLOCK_ZERO:
- memset(buf, 0, bytes);
- break;
- case BLUESKY_BLOCK_REF:
- bluesky_block_fetch(inode->fs, b);
- /* Fall through */
- case BLUESKY_BLOCK_CACHED:
- case BLUESKY_BLOCK_DIRTY:
- memcpy(buf, &b->data->data[block_offset], bytes);
- break;
- }
-
- offset += bytes;
- buf += bytes;
- len -= bytes;
- }
-}
-
-/* Read the given block from cloud-backed storage if the data is not already
- * cached. */
-void bluesky_block_fetch(BlueSkyFS *fs, BlueSkyBlock *block)
-{
- if (block->type != BLUESKY_BLOCK_REF)
- return;
-
- g_print("Fetching block from %s\n", block->ref);
- BlueSkyRCStr *string = s3store_get(fs->store, block->ref);
-
- bluesky_string_unref(block->data);
- block->data = string;
- block->type = BLUESKY_BLOCK_CACHED;
-}
-
-/* Write the given block to cloud-backed storage and mark it clean. */
-void bluesky_block_flush(BlueSkyFS *fs, BlueSkyBlock *block)
-{
- if (block->type != BLUESKY_BLOCK_DIRTY)
- return;
-
- GChecksum *csum = g_checksum_new(G_CHECKSUM_SHA256);
- g_checksum_update(csum, block->data->data, block->data->len);
- gchar *name = g_strdup(g_checksum_get_string(csum));
-
- g_print("Flushing block as %s\n", name);
- s3store_put(fs->store, name, block->data);
- g_free(block->ref);
- block->ref = name;
-
- /* block->type = BLUESKY_BLOCK_CACHED; */
- bluesky_string_unref(block->data);
- block->data = NULL;
- block->type = BLUESKY_BLOCK_REF;
-
- g_checksum_free(csum);
-}
projects / bluesky.git / blobdiff