int default_create_and_write(uint64_t oid, struct siocb *iocb)
{
char path[PATH_MAX], tmp_path[PATH_MAX];
int flags = get_open_flags(oid, true);
int ret, fd;
uint32_t len = iocb->length;
get_obj_path(oid, path);
get_tmp_obj_path(oid, tmp_path);
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 != SD_RES_SUCCESS)
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;
}
/* Trim zero blocks of the beginning and end of the object. */
static int default_trim(int fd, uint64_t oid, const struct siocb *iocb,
uint64_t *poffset, uint32_t *plen)
{
trim_zero_blocks(iocb->buf, poffset, plen);
if (iocb->offset < *poffset) {
sd_debug("discard between %d, %ld, %" PRIx64, iocb->offset,
*poffset, oid);
if (discard(fd, iocb->offset, *poffset) < 0)
return -1;
}
if (*poffset + *plen < iocb->offset + iocb->length) {
uint64_t end = iocb->offset + iocb->length;
if (end == get_objsize(oid))
/* This is necessary to punch the last block */
end = round_up(end, BLOCK_SIZE);
sd_debug("discard between %ld, %ld, %" PRIx64, *poffset + *plen,
end, oid);
if (discard(fd, *poffset + *plen, end) < 0)
return -1;
}
return 0;
}
size_t get_store_objsize(uint64_t oid)
{
if (is_erasure_oid(oid)) {
uint8_t policy = get_vdi_copy_policy(oid_to_vid(oid));
int d;
ec_policy_to_dp(policy, &d, NULL);
return SD_DATA_OBJ_SIZE / d;
}
return get_objsize(oid);
}
static int farm_write(uint64_t oid, struct siocb *iocb, int create)
{
int flags = def_open_flags, fd, ret = SD_RES_SUCCESS;
char path[PATH_MAX];
ssize_t size;
if (iocb->epoch < sys_epoch()) {
dprintf("%"PRIu32" sys %"PRIu32"\n", iocb->epoch, sys_epoch());
return SD_RES_OLD_NODE_VER;
}
if (!is_data_obj(oid))
flags &= ~O_DIRECT;
if (create)
flags |= O_CREAT | O_TRUNC;
sprintf(path, "%s%016"PRIx64, obj_path, oid);
fd = open(path, flags, def_fmode);
if (fd < 0)
return err_to_sderr(oid, errno);
if (flock(fd, LOCK_EX) < 0) {
ret = SD_RES_EIO;
eprintf("%m\n");
goto out;
}
if (create && !(iocb->flags & SD_FLAG_CMD_COW)) {
ret = prealloc(fd, get_objsize(oid));
if (ret != SD_RES_SUCCESS) {
if (flock(fd, LOCK_UN) < 0) {
ret = SD_RES_EIO;
eprintf("%m\n");
goto out;
}
goto out;
}
}
size = xpwrite(fd, iocb->buf, iocb->length, iocb->offset);
if (flock(fd, LOCK_UN) < 0) {
ret = SD_RES_EIO;
eprintf("%m\n");
goto out;
}
if (size != iocb->length) {
eprintf("%m\n");
ret = SD_RES_EIO;
goto out;
}
trunk_update_entry(oid);
out:
close(fd);
return ret;
}
size_t get_store_objsize(uint8_t copy_policy, uint64_t oid)
{
if (is_vdi_obj(oid))
return SD_INODE_SIZE;
if (copy_policy != 0) {
int d;
ec_policy_to_dp(copy_policy, &d, NULL);
return SD_DATA_OBJ_SIZE / d;
}
return get_objsize(oid);
}
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;
}
int default_get_hash(uint64_t oid, uint32_t epoch, uint8_t *sha1)
{
int ret;
void *buf;
struct siocb iocb = {};
uint32_t length;
bool is_readonly_obj = oid_is_readonly(oid);
char path[PATH_MAX];
ret = get_object_path(oid, epoch, path);
if (ret != SD_RES_SUCCESS)
return ret;
if (is_readonly_obj) {
if (get_object_sha1(path, sha1) == 0) {
sd_debug("use cached sha1 digest %s",
sha1_to_hex(sha1));
return SD_RES_SUCCESS;
}
}
length = get_objsize(oid);
buf = valloc(length);
if (buf == NULL)
return SD_RES_NO_MEM;
iocb.epoch = epoch;
iocb.buf = buf;
iocb.length = length;
ret = default_read_from_path(oid, path, &iocb);
if (ret != SD_RES_SUCCESS) {
free(buf);
return ret;
}
sha1_from_buffer(buf, length, sha1);
free(buf);
sd_debug("the message digest of %"PRIx64" at epoch %d is %s", oid,
epoch, sha1_to_hex(sha1));
if (is_readonly_obj)
set_object_sha1(path, sha1);
return ret;
}
size_t get_store_objsize(uint8_t copy_policy, uint8_t block_size_shift,
uint64_t oid)
{
if (is_vdi_obj(oid))
return SD_INODE_SIZE;
if (is_vdi_btree_obj(oid))
return SD_INODE_DATA_INDEX_SIZE;
uint32_t object_size = (UINT32_C(1) << block_size_shift);
if (copy_policy != 0) {
int d;
ec_policy_to_dp(copy_policy, &d, NULL);
return object_size / d;
}
return get_objsize(oid, object_size);
}
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 = 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) {
sd_eprintf("turn off journaling");
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.
*/
sd_dprintf("%s exists", tmp_path);
return SD_RES_SUCCESS;
}
sd_eprintf("failed to open %s: %m", 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) {
sd_eprintf("failed to write object. %m");
ret = err_to_sderr(oid, errno);
goto out;
}
ret = rename(tmp_path, path);
if (ret < 0) {
sd_eprintf("failed to rename %s to %s: %m", tmp_path, path);
ret = err_to_sderr(oid, errno);
goto out;
}
sd_dprintf("%"PRIx64, oid);
ret = SD_RES_SUCCESS;
out:
if (ret != SD_RES_SUCCESS)
unlink(tmp_path);
close(fd);
return ret;
}
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
obj_size = get_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;
}
ret = rename(tmp_path, path);