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>
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
9 * 1. Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
14 * 3. Neither the name of the University nor the names of its contributors
15 * may be used to endorse or promote products derived from this software
16 * without specific prior written permission.
18 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
37 #include "bluesky-private.h"
39 static void inode_fetch_task(gpointer a, gpointer b);
41 /* Core filesystem. Different proxies, such as the NFSv3 one, interface to
42 * this, but the core actually tracks the data which is stored. So far we just
43 * implement an in-memory filesystem, but eventually this will be state which
44 * is persisted to the cloud. */
46 /* Return the current time, in microseconds since the epoch. */
47 int64_t bluesky_get_current_time()
50 g_get_current_time(&t);
51 return (int64_t)t.tv_sec * 1000000 + t.tv_usec;
54 /* Update an inode to indicate that a modification was made. This increases
55 * the change counter, updates the ctime to the current time, and optionally
56 * updates the mtime. This also makes the inode contents subject to writeback
57 * to storage in the future. inode must already be locked. */
58 void bluesky_inode_update_ctime(BlueSkyInode *inode, gboolean update_mtime)
60 int64_t now = bluesky_get_current_time();
61 inode->change_count++;
66 if (inode->change_time == 0)
67 inode->change_time = now;
70 if (bluesky_options.writethrough_cache)
71 bluesky_file_flush(inode, NULL);
74 g_mutex_lock(inode->fs->lock);
75 bluesky_list_unlink(&inode->fs->unlogged_list, inode->unlogged_list);
76 inode->unlogged_list = bluesky_list_prepend(&inode->fs->unlogged_list, inode);
77 bluesky_list_unlink(&inode->fs->accessed_list, inode->accessed_list);
78 inode->accessed_list = bluesky_list_prepend(&inode->fs->accessed_list, inode);
79 g_mutex_unlock(inode->fs->lock);
82 /* Unfortunately a glib hash table is only guaranteed to be able to store
83 * 32-bit keys if we use the key directly. If we want 64-bit inode numbers,
84 * we'll have to allocate memory to store the 64-bit inumber, and use a pointer
85 * to it. Rather than allocate the memory for the key, we'll just include a
86 * pointer to the 64-bit inum stored in the inode itself, so that we don't need
87 * to do any more memory management. */
88 static guint bluesky_fs_key_hash_func(gconstpointer key)
90 uint64_t inum = *(const uint64_t *)key;
94 static gboolean bluesky_fs_key_equal_func(gconstpointer a, gconstpointer b)
96 uint64_t i1 = *(const uint64_t *)a;
97 uint64_t i2 = *(const uint64_t *)b;
101 /* Filesystem-level operations. A filesystem is like a directory tree that we
102 * are willing to export. */
103 BlueSkyFS *bluesky_new_fs(gchar *name)
105 BlueSkyFS *fs = g_new0(BlueSkyFS, 1);
106 fs->lock = g_mutex_new();
107 fs->name = g_strdup(name);
108 fs->inodes = g_hash_table_new(bluesky_fs_key_hash_func,
109 bluesky_fs_key_equal_func);
110 fs->next_inum = BLUESKY_ROOT_INUM + 1;
111 fs->store = bluesky_store_new("file");
112 fs->flushd_lock = g_mutex_new();
113 fs->flushd_cond = g_cond_new();
114 fs->locations = g_hash_table_new(bluesky_cloudlog_hash,
115 bluesky_cloudlog_equal);
116 fs->inode_map = g_sequence_new(NULL);
118 fs->log_state = g_new0(BlueSkyCloudLogState, 1);
119 fs->log_state->data = g_string_new("");
120 fs->log_state->latest_cleaner_seq_seen = -1;
121 fs->log_state->uploads_pending_lock = g_mutex_new();
122 fs->log_state->uploads_pending_cond = g_cond_new();
124 bluesky_cloudlog_threads_init(fs);
125 fs->inode_fetch_thread_pool = g_thread_pool_new(inode_fetch_task, NULL,
132 BlueSkyFS *bluesky_init_fs(gchar *name, BlueSkyStore *store,
133 const gchar *master_key)
135 BlueSkyFS *fs = bluesky_new_fs(name);
136 fs->master_key = g_strdup(master_key);
137 fs->keys = g_new(BlueSkyCryptKeys, 1);
138 bluesky_crypt_derive_keys(fs->keys, master_key);
140 fs->log = bluesky_log_new("journal");
143 if (bluesky_checkpoint_load(fs)) {
144 g_print("Filesystem checkpoint loaded, starting journal replay...\n");
146 g_print("Journal replay complete, filesystem ready.\n");
148 /* Initialize a fresh filesystem */
149 g_print("Initializing new filesystem...\n");
150 BlueSkyInode *root = bluesky_new_inode(BLUESKY_ROOT_INUM, fs,
154 bluesky_insert_inode(fs, root);
155 bluesky_inode_update_ctime(root, TRUE);
156 bluesky_inode_do_sync(root);
159 bluesky_cleaner_thread_launch(fs);
164 /* Inode reference counting. */
165 void bluesky_inode_ref(BlueSkyInode *inode)
167 g_atomic_int_inc(&inode->refcount);
170 /* Free most of the resources used by an inode structure, but do not free the
171 * inode itself. Can be used if the inode data will be reloaded from
172 * serialized form to clear out old information first. */
173 void bluesky_inode_free_resources(BlueSkyInode *inode)
175 switch (inode->type) {
176 case BLUESKY_REGULAR:
177 if (inode->blocks != NULL) {
178 for (int i = 0; i < inode->blocks->len; i++) {
179 BlueSkyBlock *b = &g_array_index(inode->blocks,
181 if (b->type == BLUESKY_BLOCK_DIRTY) {
182 g_error("Deleting an inode with dirty file data!");
184 bluesky_cloudlog_unref(b->ref);
185 bluesky_string_unref(b->dirty);
187 g_array_unref(inode->blocks);
188 inode->blocks = NULL;
192 case BLUESKY_DIRECTORY:
193 if (inode->dirhash != NULL)
194 g_hash_table_destroy(inode->dirhash);
195 inode->dirhash = NULL;
196 if (inode->dirhash_folded != NULL)
197 g_hash_table_destroy(inode->dirhash_folded);
198 inode->dirhash_folded = NULL;
199 if (inode->dirents != NULL)
200 g_sequence_free(inode->dirents);
201 inode->dirents = NULL;
204 case BLUESKY_SYMLINK:
205 g_free(inode->symlink_contents);
206 inode->symlink_contents = NULL;
214 void bluesky_inode_unref(BlueSkyInode *inode)
216 if (g_atomic_int_dec_and_test(&inode->refcount)) {
217 if (bluesky_verbose) {
218 g_log("bluesky/inode", G_LOG_LEVEL_DEBUG,
219 "Reference count for inode %"PRIu64" dropped to zero.",
223 /* Sanity check: Is the inode clean? */
224 if (inode->change_commit < inode->change_count
225 || inode->accessed_list != NULL
226 || inode->unlogged_list != NULL
227 || inode->dirty_list != NULL) {
228 g_warning("Dropping inode which is not clean (commit %"PRIi64" < change %"PRIi64"; accessed_list = %p; dirty_list = %p)\n", inode->change_commit, inode->change_count, inode->accessed_list, inode->dirty_list);
231 /* These shouldn't be needed, but in case the above warning fires and
232 * we delete the inode anyway, we ought to be sure the inode is not on
234 g_mutex_lock(inode->fs->lock);
235 bluesky_list_unlink(&inode->fs->accessed_list, inode->accessed_list);
236 bluesky_list_unlink(&inode->fs->dirty_list, inode->dirty_list);
237 bluesky_list_unlink(&inode->fs->unlogged_list, inode->unlogged_list);
238 g_mutex_unlock(inode->fs->lock);
240 bluesky_inode_free_resources(inode);
242 g_mutex_free(inode->lock);
247 /* Allocate a fresh inode number which has not been used before within a
248 * filesystem. fs must already be locked. */
249 uint64_t bluesky_fs_alloc_inode(BlueSkyFS *fs)
253 inum = fs->next_inum;
259 /* Perform type-specification initialization of an inode. Normally performed
260 * in bluesky_new_inode, but can be separated if an inode is created first,
261 * then deserialized. */
262 void bluesky_init_inode(BlueSkyInode *i, BlueSkyFileType type)
267 case BLUESKY_REGULAR:
268 i->blocks = g_array_new(FALSE, TRUE, sizeof(BlueSkyBlock));
270 case BLUESKY_DIRECTORY:
271 i->dirents = g_sequence_new(bluesky_dirent_destroy);
272 i->dirhash = g_hash_table_new(g_str_hash, g_str_equal);
273 i->dirhash_folded = g_hash_table_new(g_str_hash, g_str_equal);
280 BlueSkyInode *bluesky_new_inode(uint64_t inum, BlueSkyFS *fs,
281 BlueSkyFileType type)
283 BlueSkyInode *i = g_new0(BlueSkyInode, 1);
285 i->lock = g_mutex_new();
290 bluesky_init_inode(i, type);
295 /* Issue a prefetch hint for an inode. This signals that the inode may be
296 * needed soon. Does not return any useful data. */
297 void bluesky_inode_prefetch(BlueSkyFS *fs, uint64_t inum)
299 BlueSkyInode *inode = NULL;
301 g_mutex_lock(fs->lock);
302 inode = (BlueSkyInode *)g_hash_table_lookup(fs->inodes, &inum);
305 /* Inode is already available, no need for any prefetching... */
306 g_mutex_unlock(fs->lock);
310 InodeMapEntry *entry = bluesky_inode_map_lookup(fs->inode_map, inum, 0);
312 bluesky_cloudlog_prefetch(entry->item);
315 g_mutex_unlock(fs->lock);
319 /* Retrieve an inode from the filesystem. Eventually this will be a cache and
320 * so we might need to go fetch the inode from elsewhere; for now all
321 * filesystem state is stored here. inode is returned with a reference held
323 BlueSkyInode *bluesky_get_inode(BlueSkyFS *fs, uint64_t inum)
325 BlueSkyInode *inode = NULL;
331 g_mutex_lock(fs->lock);
332 inode = (BlueSkyInode *)g_hash_table_lookup(fs->inodes, &inum);
335 bluesky_inode_fetch(fs, inum);
336 inode = (BlueSkyInode *)g_hash_table_lookup(fs->inodes, &inum);
340 bluesky_inode_ref(inode);
342 /* FIXME: We assume we can atomically update the in-memory access time
344 inode->access_time = bluesky_get_current_time();
347 g_mutex_unlock(fs->lock);
352 /* Insert an inode into the filesystem inode cache. fs should be locked. */
353 void bluesky_insert_inode(BlueSkyFS *fs, BlueSkyInode *inode)
355 g_hash_table_insert(fs->inodes, &inode->inum, inode);
358 /* Start writeback of an inode and all associated data. */
359 void bluesky_inode_start_sync(BlueSkyInode *inode)
361 GList *log_items = NULL;
363 if (inode->type == BLUESKY_REGULAR)
364 bluesky_file_flush(inode, &log_items);
366 BlueSkyCloudLog *cloudlog = bluesky_serialize_inode(inode);
368 bluesky_cloudlog_unref(inode->committed_item);
369 inode->committed_item = cloudlog;
371 bluesky_cloudlog_sync(cloudlog);
372 bluesky_cloudlog_ref(cloudlog);
373 log_items = g_list_prepend(log_items, cloudlog);
375 /* Wait for all log items to be committed to disk. */
376 bluesky_log_finish_all(log_items);
378 /* Mark the inode as clean */
379 inode->change_commit = inode->change_count;
380 inode->change_time = 0;
381 g_mutex_lock(inode->fs->lock);
382 bluesky_list_unlink(&inode->fs->unlogged_list, inode->unlogged_list);
383 inode->unlogged_list = NULL;
385 /* Since a new version of the inode has been written to the log, also
386 * schedule a future flush of the new data to cloud storage. */
387 bluesky_list_unlink(&inode->fs->dirty_list, inode->dirty_list);
388 inode->dirty_list = bluesky_list_prepend(&inode->fs->dirty_list, inode);
389 inode->change_cloud = inode->change_count;
391 g_mutex_unlock(inode->fs->lock);
394 /* Write back an inode and all associated data and wait for completion. Inode
395 * should already be locked. */
396 void bluesky_inode_do_sync(BlueSkyInode *inode)
398 if (bluesky_verbose) {
399 g_log("bluesky/inode", G_LOG_LEVEL_DEBUG,
400 "Synchronous writeback for inode %"PRIu64"...", inode->inum);
402 bluesky_inode_start_sync(inode);
403 if (bluesky_verbose) {
404 g_log("bluesky/inode", G_LOG_LEVEL_DEBUG,
405 "Writeback for inode %"PRIu64" complete", inode->inum);
409 static void inode_fetch_task(gpointer a, gpointer b)
411 BlueSkyInode *inode = (BlueSkyInode *)a;
413 bluesky_profile_set((BlueSkyProfile *)inode->private_data);
415 BlueSkyCloudLog *item = inode->committed_item;
416 inode->committed_item = NULL;
417 g_print("Completing fetch of inode %"PRIu64"...\n", inode->inum);
419 g_mutex_lock(item->lock);
420 bluesky_cloudlog_fetch(item);
421 if (!bluesky_deserialize_inode(inode, item))
422 g_print("Error deserializing inode %"PRIu64"\n", inode->inum);
423 g_mutex_unlock(item->lock);
425 inode->access_time = bluesky_get_current_time();
426 g_mutex_lock(inode->fs->lock);
427 bluesky_list_unlink(&inode->fs->accessed_list, inode->accessed_list);
428 inode->accessed_list = bluesky_list_prepend(&inode->fs->accessed_list, inode);
429 g_mutex_unlock(inode->fs->lock);
431 g_mutex_unlock(inode->lock);
432 bluesky_cloudlog_unref(item);
433 bluesky_inode_unref(inode);
436 /* Fetch an inode from stable storage. The fetch can be performed
437 * asynchronously: the in-memory inode is allocated, but not filled with data
438 * immediately. It is kept locked until it has been filled in, so any users
439 * should try to acquire the lock on the inode before accessing any data. The
440 * fs lock must be held. */
441 void bluesky_inode_fetch(BlueSkyFS *fs, uint64_t inum)
443 InodeMapEntry *entry = bluesky_inode_map_lookup(fs->inode_map, inum, 0);
447 /* Non-portable behavior: We take the inode lock here, and release it in
448 * the fetching thread. This works with the default Linux pthreads
449 * implementation but is not guaranteed. */
451 BlueSkyInode *inode = bluesky_new_inode(inum, fs, BLUESKY_PENDING);
452 inode->change_count = 0;
453 bluesky_inode_ref(inode); // Extra ref held by fetching process
454 g_mutex_lock(inode->lock);
456 inode->committed_item = entry->item;
457 bluesky_cloudlog_ref(entry->item);
458 bluesky_insert_inode(fs, inode);
460 inode->private_data = bluesky_profile_get();
461 g_thread_pool_push(fs->inode_fetch_thread_pool, inode, NULL);