int trunk_init(void)
{
DIR *dir;
struct dirent *d;
uint64_t oid;
dir = opendir(obj_path);
if (!dir)
return -1;
while ((d = readdir(dir))) {
if (!strncmp(d->d_name, ".", 1))
continue;
oid = strtoull(d->d_name, NULL, 16);
if (oid == 0 || oid == ULLONG_MAX)
continue;
objlist_cache_insert(oid);
lookup_trunk_entry(oid, 1);
}
omap_tree = RB_ROOT;
closedir(dir);
return 0;
}
static int init_objlist_and_vdi_bitmap(uint64_t oid, char *wd, uint32_t epoch,
void *arg)
{
int ret;
objlist_cache_insert(oid);
if (is_vdi_obj(oid)) {
sd_debug("found the VDI object %" PRIx64, oid);
ret = init_vdi_state(oid, wd, epoch);
if (ret != SD_RES_SUCCESS)
return ret;
}
return SD_RES_SUCCESS;
}
static int init_objlist_and_vdi_bitmap(uint64_t oid, void *arg)
{
int ret;
objlist_cache_insert(oid);
if (is_vdi_obj(oid)) {
vprintf(SDOG_DEBUG, "found the VDI object %" PRIx64 "\n", oid);
set_bit(oid_to_vid(oid), sys->vdi_inuse);
ret = init_vdi_copy_number(oid);
if (ret != SD_RES_SUCCESS)
return ret;
}
return SD_RES_SUCCESS;
}
static int init_objlist_and_vdi_bitmap(uint64_t oid, const char *wd,
uint32_t epoch, uint8_t ec_index,
struct vnode_info *vinfo,
void *arg)
{
int ret = SD_RES_SUCCESS;
objlist_cache_insert(oid);
if (is_vdi_obj(oid)) {
sd_debug("found the VDI object %" PRIx64" epoch %"PRIu32
" at %s", oid, epoch, wd);
atomic_set_bit(oid_to_vid(oid), sys->vdi_inuse);
}
return ret;
}
static int recover_object_from_replica(uint64_t oid,
const struct sd_vnode *vnode,
uint32_t epoch, uint32_t tgt_epoch)
{
struct sd_req hdr;
struct sd_rsp *rsp = (struct sd_rsp *)&hdr;
unsigned rlen;
int ret = SD_RES_NO_MEM;
void *buf = NULL;
struct siocb iocb = { 0 };
if (vnode_is_local(vnode)) {
ret = sd_store->link(oid, tgt_epoch);
goto out;
}
rlen = get_objsize(oid);
buf = valloc(rlen);
if (!buf) {
sd_eprintf("%m");
goto out;
}
sd_init_req(&hdr, SD_OP_READ_PEER);
hdr.epoch = epoch;
hdr.flags = SD_FLAG_CMD_RECOVERY;
hdr.data_length = rlen;
hdr.obj.oid = oid;
hdr.obj.tgt_epoch = tgt_epoch;
ret = sheep_exec_req(&vnode->nid, &hdr, buf);
if (ret != SD_RES_SUCCESS)
goto out;
iocb.epoch = epoch;
iocb.length = rsp->data_length;
iocb.offset = rsp->obj.offset;
iocb.buf = buf;
ret = sd_store->create_and_write(oid, &iocb);
out:
if (ret == SD_RES_SUCCESS) {
sd_dprintf("recovered oid %"PRIx64" from %d to epoch %d", oid,
tgt_epoch, epoch);
objlist_cache_insert(oid);
}
free(buf);
return ret;
}
static int init_objlist_and_vdi_bitmap(uint64_t oid, const char *wd,
uint32_t epoch, uint8_t ec_index,
struct vnode_info *vinfo,
void *arg)
{
int ret;
objlist_cache_insert(oid);
if (is_vdi_obj(oid)) {
sd_debug("found the VDI object %" PRIx64" epoch %"PRIu32
" at %s", oid, epoch, wd);
ret = init_vdi_state(oid, wd, epoch);
if (ret != SD_RES_SUCCESS)
return ret;
}
return SD_RES_SUCCESS;
}
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;
}
static int recover_object_from_replica(uint64_t oid,
struct sd_vnode *entry,
uint32_t epoch, uint32_t tgt_epoch)
{
struct sd_req hdr;
struct sd_rsp *rsp = (struct sd_rsp *)&hdr;
char name[128];
unsigned wlen = 0, rlen;
int fd, ret = -1;
void *buf;
struct siocb iocb = { 0 };
rlen = get_objsize(oid);
buf = valloc(rlen);
if (!buf) {
eprintf("%m\n");
goto out;
}
if (vnode_is_local(entry)) {
iocb.epoch = epoch;
iocb.length = rlen;
ret = sd_store->link(oid, &iocb, tgt_epoch);
if (ret == SD_RES_SUCCESS) {
ret = 0;
goto done;
} else {
ret = -1;
goto out;
}
}
addr_to_str(name, sizeof(name), entry->nid.addr, 0);
fd = connect_to(name, entry->nid.port);
dprintf("%s, %d\n", name, entry->nid.port);
if (fd < 0) {
eprintf("failed to connect to %s:%"PRIu32"\n", name, entry->nid.port);
ret = -1;
goto out;
}
sd_init_req(&hdr, SD_OP_READ_PEER);
hdr.epoch = epoch;
hdr.flags = SD_FLAG_CMD_RECOVERY;
hdr.data_length = rlen;
hdr.obj.oid = oid;
hdr.obj.tgt_epoch = tgt_epoch;
ret = exec_req(fd, &hdr, buf, &wlen, &rlen);
close(fd);
if (ret != 0) {
eprintf("res: %"PRIx32"\n", rsp->result);
ret = -1;
goto out;
}
rsp = (struct sd_rsp *)&hdr;
if (rsp->result == SD_RES_SUCCESS) {
iocb.epoch = epoch;
iocb.length = rlen;
iocb.buf = buf;
ret = sd_store->atomic_put(oid, &iocb);
if (ret != SD_RES_SUCCESS) {
ret = -1;
goto out;
}
} else {
eprintf("failed, res: %"PRIx32"\n", rsp->result);
ret = rsp->result;
goto out;
}
done:
dprintf("recovered oid %"PRIx64" from %d to epoch %d\n", oid, tgt_epoch, epoch);
out:
if (ret == SD_RES_SUCCESS)
objlist_cache_insert(oid);
free(buf);
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;
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;
}