X-Git-Url: http://git.vrable.net/?a=blobdiff_plain;f=bluesky%2Flog.c;h=88b95df335e0f7beae8f81433d4bfda7b10e7d6b;hb=72de763405222cc8f9bc5c75fd163362ec0f2266;hp=38606c1f69bb7d0020dee31cc21086fe715670ee;hpb=9f0e2f7de7d919d6a83944f0e7fdbd07cb6c4189;p=bluesky.git diff --git a/bluesky/log.c b/bluesky/log.c index 38606c1..88b95df 100644 --- a/bluesky/log.c +++ b/bluesky/log.c @@ -110,7 +110,8 @@ static gboolean log_open(BlueSkyLog *log) } } - log->current_log = bluesky_cachefile_lookup(log->fs, -1, log->seq_num); + log->current_log = bluesky_cachefile_lookup(log->fs, -1, log->seq_num, + FALSE); g_assert(log->current_log != NULL); g_mutex_unlock(log->current_log->lock); @@ -354,59 +355,13 @@ void bluesky_cachefile_unref(BlueSkyCacheFile *cachefile) g_atomic_int_add(&cachefile->refcount, -1); } -static void cloudlog_fetch_complete(BlueSkyStoreAsync *async, - BlueSkyCacheFile *cachefile); - -static void cloudlog_fetch_start(BlueSkyCacheFile *cachefile) -{ - g_atomic_int_inc(&cachefile->refcount); - cachefile->fetching = TRUE; - g_print("Starting fetch of %s from cloud\n", cachefile->filename); - BlueSkyStoreAsync *async = bluesky_store_async_new(cachefile->fs->store); - async->op = STORE_OP_GET; - async->key = g_strdup(cachefile->filename); - bluesky_store_async_add_notifier(async, - (GFunc)cloudlog_fetch_complete, - cachefile); - bluesky_store_async_submit(async); - bluesky_store_async_unref(async); -} - -static void cloudlog_fetch_complete(BlueSkyStoreAsync *async, - BlueSkyCacheFile *cachefile) -{ - g_print("Fetch of %s from cloud complete, status = %d\n", - async->key, async->result); - - g_mutex_lock(cachefile->lock); - if (async->result >= 0) { - char *pathname = g_strdup_printf("%s/%s", - cachefile->log->log_directory, - cachefile->filename); - async->data = bluesky_string_dup(async->data); - bluesky_cloudlog_decrypt(async->data->data, async->data->len, - cachefile->fs->keys, cachefile->items); - if (!g_file_set_contents(pathname, async->data->data, async->data->len, - NULL)) - g_print("Error writing out fetched file to cache!\n"); - g_free(pathname); - - cachefile->fetching = FALSE; - cachefile->ready = TRUE; - } else { - g_print("Error fetching from cloud, retrying...\n"); - cloudlog_fetch_start(cachefile); - } - - bluesky_cachefile_unref(cachefile); - g_cond_broadcast(cachefile->cond); - g_mutex_unlock(cachefile->lock); -} +static void cloudlog_fetch_start(BlueSkyCacheFile *cachefile); /* Find the BlueSkyCacheFile object for the given journal or cloud log segment. * Returns the object in the locked state and with a reference taken. */ BlueSkyCacheFile *bluesky_cachefile_lookup(BlueSkyFS *fs, - int clouddir, int log_seq) + int clouddir, int log_seq, + gboolean start_fetch) { if (page_size == 0) { page_size = getpagesize(); @@ -455,8 +410,14 @@ BlueSkyCacheFile *bluesky_cachefile_lookup(BlueSkyFS *fs, g_hash_table_insert(log->mmap_cache, map->filename, map); + int fd = openat(log->dirfd, logname, O_WRONLY | O_CREAT, 0600); + if (fd >= 0) { + ftruncate(fd, 5 << 20); // FIXME + close(fd); + } + // If the log file is stored in the cloud, we may need to fetch it - if (clouddir >= 0) + if (clouddir >= 0 && start_fetch) cloudlog_fetch_start(map); } else { g_mutex_lock(map->lock); @@ -468,55 +429,164 @@ BlueSkyCacheFile *bluesky_cachefile_lookup(BlueSkyFS *fs, return map; } +static void robust_pwrite(int fd, const char *buf, ssize_t count, off_t offset) +{ + while (count > 0) { + ssize_t written = pwrite(fd, buf, count, offset); + if (written < 0) { + if (errno == EINTR) + continue; + g_warning("pwrite failure: %m"); + return; + } + buf += written; + count -= written; + offset += written; + } +} + +static void cloudlog_partial_fetch_complete(BlueSkyStoreAsync *async, + BlueSkyCacheFile *cachefile); + +static void cloudlog_partial_fetch_start(BlueSkyCacheFile *cachefile, + size_t offset, size_t length) +{ + g_atomic_int_inc(&cachefile->refcount); + g_print("Starting fetch of %s from cloud\n", cachefile->filename); + BlueSkyStoreAsync *async = bluesky_store_async_new(cachefile->fs->store); + async->op = STORE_OP_GET; + async->key = g_strdup(cachefile->filename); + async->start = offset; + async->len = length; + bluesky_store_async_add_notifier(async, + (GFunc)cloudlog_partial_fetch_complete, + cachefile); + bluesky_store_async_submit(async); + bluesky_store_async_unref(async); +} + +static void cloudlog_partial_fetch_complete(BlueSkyStoreAsync *async, + BlueSkyCacheFile *cachefile) +{ + g_print("Partial fetch of %s from cloud complete, status = %d\n", + async->key, async->result); + + g_mutex_lock(cachefile->lock); + if (async->result >= 0) { + /* Descrypt items fetched and write valid items out to the local log, + * but only if they do not overlap existing objects. This will protect + * against an attack by the cloud provider where one valid object is + * moved to another offset and used to overwrite data that we already + * have fetched. */ + BlueSkyRangeset *items = bluesky_rangeset_new(); + int fd = openat(cachefile->log->dirfd, cachefile->filename, O_WRONLY); + if (fd >= 0) { + async->data = bluesky_string_dup(async->data); + bluesky_cloudlog_decrypt(async->data->data, async->data->len, + cachefile->fs->keys, items); + uint64_t item_offset = 0; + while (TRUE) { + const BlueSkyRangesetItem *item; + item = bluesky_rangeset_lookup_next(items, item_offset); + if (item == NULL) + break; + g_print(" item offset from range request: %d\n", + (int)(item->start + async->start)); + if (bluesky_rangeset_insert(cachefile->items, + async->start + item->start, + item->length, item->data)) + { + robust_pwrite(fd, async->data->data + item->start, + item->length, async->start + item->start); + } else { + g_print(" item overlaps existing data!\n"); + } + item_offset = item->start + 1; + } + /* TODO: Iterate over items and merge into cached file. */ + close(fd); + } else { + g_warning("Unable to open and write to cache file %s: %m", + cachefile->filename); + } + } else { + g_print("Error fetching from cloud, retrying...\n"); + cloudlog_partial_fetch_start(cachefile, async->start, async->len); + } + + bluesky_cachefile_unref(cachefile); + g_cond_broadcast(cachefile->cond); + g_mutex_unlock(cachefile->lock); +} + +static void cloudlog_fetch_start(BlueSkyCacheFile *cachefile) +{ + g_atomic_int_inc(&cachefile->refcount); + cachefile->fetching = TRUE; + g_print("Starting fetch of %s from cloud\n", cachefile->filename); + BlueSkyStoreAsync *async = bluesky_store_async_new(cachefile->fs->store); + async->op = STORE_OP_GET; + async->key = g_strdup(cachefile->filename); + bluesky_store_async_add_notifier(async, + (GFunc)cloudlog_partial_fetch_complete, + cachefile); + bluesky_store_async_submit(async); + bluesky_store_async_unref(async); +} + /* The arguments are mostly straightforward. log_dir is -1 for access from the * journal, and non-negative for access to a cloud log segment. map_data * should be TRUE for the case that are mapping just the data of an item where * we have already parsed the item headers; this surpresses the error when the * access is not to the first bytes of the item. */ -BlueSkyRCStr *bluesky_log_map_object(BlueSkyFS *fs, int log_dir, - int log_seq, int log_offset, int log_size, - gboolean map_data) +BlueSkyRCStr *bluesky_log_map_object(BlueSkyCloudLog *item, gboolean map_data) { + BlueSkyFS *fs = item->fs; + BlueSkyLog *log = fs->log; + BlueSkyCacheFile *map = NULL; + BlueSkyRCStr *str = NULL; + int location = 0; + size_t file_offset = 0, file_size = 0; + gboolean range_request = TRUE; + if (page_size == 0) { page_size = getpagesize(); } - BlueSkyLog *log = fs->log; - BlueSkyCacheFile *map = bluesky_cachefile_lookup(fs, log_dir, log_seq); + bluesky_cloudlog_stats_update(item, -1); - if (map == NULL) { - return NULL; + /* First, check to see if the journal still contains a copy of the item and + * if so use that. */ + if ((item->location_flags | item->pending_write) & CLOUDLOG_JOURNAL) { + map = bluesky_cachefile_lookup(fs, -1, item->log_seq, TRUE); + if (map != NULL) { + location = CLOUDLOG_JOURNAL; + file_offset = item->log_offset; + file_size = item->log_size; + } } - /* Log segments fetched from the cloud might only be partially-fetched. - * Check whether the object we are interested in is available. */ - if (log_dir >= 0) { - const BlueSkyRangesetItem *rangeitem; - rangeitem = bluesky_rangeset_lookup(map->items, log_offset); - if (rangeitem == NULL || rangeitem->start != log_offset) { - g_warning("log-%d: Item at offset %d does not seem to be available\n", log_seq, log_offset); - } - if (map_data && rangeitem != NULL - && log_offset > rangeitem->start - && log_size <= rangeitem->length - (log_offset - rangeitem->start)) - { - g_warning(" ...allowing access to middle of log item"); + if (location == 0 && (item->location_flags & CLOUDLOG_CLOUD)) { + item->location_flags &= ~CLOUDLOG_JOURNAL; + map = bluesky_cachefile_lookup(fs, + item->location.directory, + item->location.sequence, + !range_request); + if (map == NULL) { + g_warning("Unable to remap cloud log segment!"); + goto exit1; } + location = CLOUDLOG_CLOUD; + file_offset = item->location.offset; + file_size = item->location.size; } if (map->addr == NULL) { - while (!map->ready && map->fetching) { - g_print("Waiting for log segment to be fetched from cloud...\n"); - g_cond_wait(map->cond, map->lock); - } - int fd = openat(log->dirfd, map->filename, O_RDONLY); if (fd < 0) { fprintf(stderr, "Error opening logfile %s: %m\n", map->filename); - bluesky_cachefile_unref(map); - g_mutex_unlock(map->lock); - return NULL; + goto exit2; } off_t length = lseek(fd, 0, SEEK_END); @@ -526,17 +596,55 @@ BlueSkyRCStr *bluesky_log_map_object(BlueSkyFS *fs, int log_dir, map->len = length; g_atomic_int_add(&log->disk_used, map->len / 1024); - g_print("Re-mapped log segment %d...\n", log_seq); g_atomic_int_inc(&map->refcount); close(fd); } - BlueSkyRCStr *str; + /* Log segments fetched from the cloud might only be partially-fetched. + * Check whether the object we are interested in is available. */ + if (location == CLOUDLOG_CLOUD) { + while (TRUE) { + const BlueSkyRangesetItem *rangeitem; + rangeitem = bluesky_rangeset_lookup(map->items, file_offset); + if (rangeitem != NULL && (rangeitem->start != file_offset + || rangeitem->length != file_size)) { + g_warning("log-%d: Item offset %zd seems to be invalid!", + (int)item->location.sequence, file_offset); + goto exit2; + } + if (rangeitem == NULL) { + g_print("Item at offset 0x%zx not available, need to fetch.\n", + file_offset); + if (range_request) + cloudlog_partial_fetch_start(map, file_offset, file_size); + g_cond_wait(map->cond, map->lock); + } else if (rangeitem->start == file_offset + && rangeitem->length == file_size) { + g_print("Item now available.\n"); + break; + } + } + } + + if (map_data) { + if (location == CLOUDLOG_JOURNAL) + file_offset += sizeof(struct log_header); + else + file_offset += sizeof(struct cloudlog_header); + + file_size = item->data_size; + } + str = bluesky_string_new_from_mmap(map, file_offset, file_size); map->atime = bluesky_get_current_time(); - str = bluesky_string_new_from_mmap(map, log_offset, log_size); + + g_print("Returning item at offset 0x%zx.\n", file_offset); + +exit2: bluesky_cachefile_unref(map); g_mutex_unlock(map->lock); +exit1: + bluesky_cloudlog_stats_update(item, 1); return str; }