int default_write(uint64_t oid, const struct siocb *iocb) { int flags = prepare_iocb(oid, iocb, false), fd, ret = SD_RES_SUCCESS; char path[PATH_MAX]; ssize_t size; uint32_t len = iocb->length; uint64_t offset = iocb->offset; static bool trim_is_supported = true; if (iocb->epoch < sys_epoch()) { sd_debug("%"PRIu32" sys %"PRIu32, iocb->epoch, sys_epoch()); return SD_RES_OLD_NODE_VER; } if (uatomic_is_true(&sys->use_journal) && unlikely(journal_write_store(oid, iocb->buf, iocb->length, iocb->offset, false)) != SD_RES_SUCCESS) { sd_err("turn off journaling"); uatomic_set_false(&sys->use_journal); flags |= O_DSYNC; sync(); } get_store_path(oid, iocb->ec_index, path); /* * Make sure oid is in the right place because oid might be misplaced * in a wrong place, due to 'shutdown/restart with less/more disks' or * any bugs. We need call err_to_sderr() to return EIO if disk is broken */ if (!default_exist(oid, iocb->ec_index)) return err_to_sderr(path, oid, ENOENT); fd = open(path, flags, sd_def_fmode); if (unlikely(fd < 0)) return err_to_sderr(path, oid, errno); if (trim_is_supported && is_sparse_object(oid)) { if (default_trim(fd, oid, iocb, &offset, &len) < 0) { trim_is_supported = false; offset = iocb->offset; len = iocb->length; } } size = xpwrite(fd, iocb->buf, len, offset); if (unlikely(size != len)) { sd_err("failed to write object %"PRIx64", path=%s, offset=%" PRId32", size=%"PRId32", result=%zd, %m", oid, path, iocb->offset, iocb->length, size); ret = err_to_sderr(path, oid, errno); goto out; } out: close(fd); return ret; }
int default_create_and_write(uint64_t oid, const struct siocb *iocb) { char path[PATH_MAX], tmp_path[PATH_MAX]; int flags = prepare_iocb(oid, iocb, true); int ret, fd; uint32_t len = iocb->length; bool ec = is_erasure_obj(oid, iocb->copy_policy); size_t obj_size; sd_debug("%"PRIx64, oid); get_obj_path(oid, path, sizeof(path)); get_tmp_obj_path(oid, tmp_path, sizeof(tmp_path)); if (uatomic_is_true(&sys->use_journal) && journal_write_store(oid, iocb->buf, iocb->length, iocb->offset, true) != SD_RES_SUCCESS) { sd_err("turn off journaling"); uatomic_set_false(&sys->use_journal); flags |= O_DSYNC; sync(); } fd = open(tmp_path, flags, sd_def_fmode); if (fd < 0) { if (errno == EEXIST) { /* * This happens if node membership changes during object * creation; while gateway retries a CREATE request, * recovery process could also recover the object at the * same time. They should try to write the same date, * so it is okay to simply return success here. */ sd_debug("%s exists", tmp_path); return SD_RES_SUCCESS; } sd_err("failed to open %s: %m", tmp_path); return err_to_sderr(path, oid, errno); } if (ec) { uint8_t policy = iocb->copy_policy ?: get_vdi_copy_policy(oid_to_vid(oid)); int d; ec_policy_to_dp(policy, &d, NULL); obj_size = SD_DATA_OBJ_SIZE / d; } else
static int get_open_flags(uint64_t oid, bool create, int fl) { int flags = O_DSYNC | O_RDWR; if (uatomic_is_true(&sys->use_journal) || sys->nosync == true) flags &= ~O_DSYNC; /* We can not use DIO for inode object because it is not 512B aligned */ if (sys->backend_dio && is_data_obj(oid)) flags |= O_DIRECT; if (create) flags |= O_CREAT | O_EXCL; return flags; }
static int prepare_iocb(uint64_t oid, const struct siocb *iocb, bool create) { int flags = O_DSYNC | O_RDWR; if (uatomic_is_true(&sys->use_journal) || sys->nosync == true) flags &= ~O_DSYNC; if (sys->backend_dio && iocb_is_aligned(iocb)) { assert(is_aligned_to_pagesize(iocb->buf)); flags |= O_DIRECT; } if (create) flags |= O_CREAT | O_EXCL; return flags; }
int default_remove_object(uint64_t oid, uint8_t ec_index) { char path[PATH_MAX]; if (uatomic_is_true(&sys->use_journal)) journal_remove_object(oid); get_store_path(oid, ec_index, path); if (unlink(path) < 0) { if (errno == ENOENT) return SD_RES_NO_OBJ; sd_err("failed, %s, %m", path); return SD_RES_EIO; } return SD_RES_SUCCESS; }
int default_remove_object(uint64_t oid) { char path[PATH_MAX]; if (uatomic_is_true(&sys->use_journal)) journal_remove_object(oid); get_obj_path(oid, path); if (unlink(path) < 0) { if (errno == ENOENT) return SD_RES_NO_OBJ; sd_err("failed to remove object %"PRIx64", %m", oid); return SD_RES_EIO; } return SD_RES_SUCCESS; }
int default_write(uint64_t oid, const struct siocb *iocb) { int flags = get_open_flags(oid, false, iocb->flags), fd, ret = SD_RES_SUCCESS; char path[PATH_MAX]; ssize_t size; if (iocb->epoch < sys_epoch()) { sd_dprintf("%"PRIu32" sys %"PRIu32"\n", iocb->epoch, sys_epoch()); return SD_RES_OLD_NODE_VER; } get_obj_path(oid, path); if (uatomic_is_true(&sys->use_journal) && journal_file_write(oid, iocb->buf, iocb->length, iocb->offset, false) != SD_RES_SUCCESS) { sd_eprintf("turn off journaling\n"); uatomic_set_false(&sys->use_journal); flags |= O_DSYNC; sync(); } fd = open(path, flags, def_fmode); if (fd < 0) return err_to_sderr(oid, errno); size = xpwrite(fd, iocb->buf, iocb->length, iocb->offset); if (size != iocb->length) { sd_eprintf("failed to write object %"PRIx64", path=%s, offset=%" PRId64", size=%"PRId32", result=%zd, %m\n", oid, path, iocb->offset, iocb->length, size); ret = err_to_sderr(oid, errno); goto out; } out: close(fd); return ret; }
int default_write(uint64_t oid, const struct siocb *iocb) { int flags = prepare_iocb(oid, iocb, false), fd, ret = SD_RES_SUCCESS; char path[PATH_MAX]; ssize_t size; if (iocb->epoch < sys_epoch()) { sd_debug("%"PRIu32" sys %"PRIu32, iocb->epoch, sys_epoch()); return SD_RES_OLD_NODE_VER; } if (uatomic_is_true(&sys->use_journal) && unlikely(journal_write_store(oid, iocb->buf, iocb->length, iocb->offset, false)) != SD_RES_SUCCESS) { sd_err("turn off journaling"); uatomic_set_false(&sys->use_journal); flags |= O_DSYNC; sync(); } get_obj_path(oid, path); fd = open(path, flags, sd_def_fmode); if (unlikely(fd < 0)) return err_to_sderr(path, oid, errno); size = xpwrite(fd, iocb->buf, iocb->length, iocb->offset); if (unlikely(size != iocb->length)) { sd_err("failed to write object %"PRIx64", path=%s, offset=%" PRId64", size=%"PRId32", result=%zd, %m", oid, path, iocb->offset, iocb->length, size); ret = err_to_sderr(path, oid, errno); goto out; } out: close(fd); return ret; }
static int get_open_flags(uint64_t oid, bool create, int fl) { int flags = O_RDWR; if (!sys->no_sync) flags |= O_DSYNC; if ((fl & SD_FLAG_CMD_CACHE && is_disk_cache_enabled()) || uatomic_is_true(&sys->use_journal)) flags &= ~O_DSYNC; /* * We can not use DIO for inode object because it is not 512B aligned. */ if (sys->backend_dio && is_data_obj(oid)) flags |= O_DIRECT; if (create) flags |= O_CREAT | O_EXCL; return flags; }
int default_create_and_write(uint64_t oid, const struct siocb *iocb) { char path[PATH_MAX], tmp_path[PATH_MAX]; int flags = prepare_iocb(oid, iocb, true); int ret, fd; uint32_t len = iocb->length; size_t obj_size; uint64_t offset = iocb->offset; sd_debug("%"PRIx64, oid); get_store_path(oid, iocb->ec_index, path); get_store_tmp_path(oid, iocb->ec_index, tmp_path); if (uatomic_is_true(&sys->use_journal) && journal_write_store(oid, iocb->buf, iocb->length, iocb->offset, true) != SD_RES_SUCCESS) { sd_err("turn off journaling"); uatomic_set_false(&sys->use_journal); flags |= O_DSYNC; sync(); } fd = open(tmp_path, flags, sd_def_fmode); if (fd < 0) { if (errno == EEXIST) { /* * This happens if node membership changes during object * creation; while gateway retries a CREATE request, * recovery process could also recover the object at the * same time. They should try to write the same date, * so it is okay to simply return success here. */ sd_debug("%s exists", tmp_path); return SD_RES_SUCCESS; } sd_err("failed to open %s: %m", tmp_path); return err_to_sderr(path, oid, errno); } obj_size = get_store_objsize(oid); trim_zero_blocks(iocb->buf, &offset, &len); if (offset != 0 || len != get_objsize(oid)) { if (is_sparse_object(oid)) ret = xftruncate(fd, obj_size); else ret = prealloc(fd, obj_size); if (ret < 0) { ret = err_to_sderr(path, oid, errno); goto out; } } ret = xpwrite(fd, iocb->buf, len, offset); if (ret != len) { sd_err("failed to write object. %m"); ret = err_to_sderr(path, oid, errno); goto out; } ret = rename(tmp_path, path); if (ret < 0) { sd_err("failed to rename %s to %s: %m", tmp_path, path); ret = err_to_sderr(path, oid, errno); goto out; } ret = SD_RES_SUCCESS; objlist_cache_insert(oid); out: if (ret != SD_RES_SUCCESS) unlink(tmp_path); close(fd); return ret; }
int default_create_and_write(uint64_t oid, const struct siocb *iocb) { char path[PATH_MAX], tmp_path[PATH_MAX]; int flags = get_open_flags(oid, true, iocb->flags); int ret, fd; uint32_t len = iocb->length; get_obj_path(oid, path); get_tmp_obj_path(oid, tmp_path); if (uatomic_is_true(&sys->use_journal) && journal_file_write(oid, iocb->buf, iocb->length, iocb->offset, true) != SD_RES_SUCCESS) { eprintf("turn off journaling\n"); uatomic_set_false(&sys->use_journal); flags |= O_DSYNC; sync(); } fd = open(tmp_path, flags, def_fmode); if (fd < 0) { if (errno == EEXIST) { /* This happens if node membership changes during object * creation; while gateway retries a CREATE request, * recovery process could also recover the object at the * same time. They should try to write the same date, * so it is okay to simply return success here. */ dprintf("%s exists\n", tmp_path); return SD_RES_SUCCESS; } eprintf("failed to open %s: %m\n", tmp_path); return err_to_sderr(oid, errno); } if (iocb->offset != 0 || iocb->length != get_objsize(oid)) { ret = prealloc(fd, get_objsize(oid)); if (ret < 0) { ret = err_to_sderr(oid, errno); goto out; } } ret = xpwrite(fd, iocb->buf, len, iocb->offset); if (ret != len) { eprintf("failed to write object. %m\n"); ret = err_to_sderr(oid, errno); goto out; } ret = rename(tmp_path, path); if (ret < 0) { eprintf("failed to rename %s to %s: %m\n", tmp_path, path); ret = err_to_sderr(oid, errno); goto out; } dprintf("%"PRIx64"\n", oid); ret = SD_RES_SUCCESS; out: if (ret != SD_RES_SUCCESS) unlink(tmp_path); close(fd); return ret; }