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;
sd_debug("%"PRIx64, oid);
get_store_path(oid, iocb->ec_index, path);
get_store_tmp_path(oid, iocb->ec_index, tmp_path);
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);
ret = prealloc(fd, obj_size);
if (ret < 0) {
ret = err_to_sderr(path, oid, errno);
goto out;
}
ret = xpwrite(fd, iocb->buf, len, iocb->offset);
if (ret != len) {
sd_err("failed to write object. %m");
ret = err_to_sderr(path, oid, errno);
goto out;
}
/*
* Modern FS like ext4, xfs defaults to automatic syncing of files after
* replace-via-rename and replace-via-truncate operations. So rename
* without fsync() is actually safe.
*/
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) != 0)
sd_err("failed to unlink %s: %m", 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], *dir;
int flags = prepare_iocb(oid, iocb, true);
int ret, fd;
uint32_t len = iocb->length;
uint32_t object_size = 0;
size_t obj_size;
uint64_t offset = iocb->offset;
sd_debug("%016"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_SYNC;
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);
object_size = get_vdi_object_size(oid_to_vid(oid));
if (offset != 0 || len != get_objsize(oid, object_size)) {
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;
}
close(fd);
if (uatomic_is_true(&sys->use_journal) || sys->nosync == true) {
objlist_cache_insert(oid);
return SD_RES_SUCCESS;
}
pstrcpy(tmp_path, sizeof(tmp_path), path);
dir = dirname(tmp_path);
fd = open(dir, O_DIRECTORY | O_RDONLY);
if (fd < 0) {
sd_err("failed to open directory %s: %m", dir);
return err_to_sderr(path, oid, errno);
}
if (fsync(fd) != 0) {
sd_err("failed to write directory %s: %m", dir);
ret = err_to_sderr(path, oid, errno);
close(fd);
if (unlink(path) != 0)
sd_err("failed to unlink %s: %m", path);
return ret;
}
close(fd);
objlist_cache_insert(oid);
return SD_RES_SUCCESS;
out:
if (unlink(tmp_path) != 0)
sd_err("failed to unlink %s: %m", tmp_path);
close(fd);
return ret;
}
