2019-12-28 01:25:55 +03:00
|
|
|
#include "osd.h"
|
|
|
|
|
|
|
|
void osd_t::read_requests()
|
|
|
|
{
|
|
|
|
for (int i = 0; i < read_ready_clients.size(); i++)
|
|
|
|
{
|
|
|
|
int peer_fd = read_ready_clients[i];
|
|
|
|
auto & cl = clients[peer_fd];
|
|
|
|
io_uring_sqe* sqe = ringloop->get_sqe();
|
|
|
|
if (!sqe)
|
|
|
|
{
|
|
|
|
read_ready_clients.erase(read_ready_clients.begin(), read_ready_clients.begin() + i);
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
ring_data_t* data = ((ring_data_t*)sqe->user_data);
|
|
|
|
if (!cl.read_buf)
|
|
|
|
{
|
|
|
|
// no reads in progress
|
|
|
|
// so this is either a new command or a reply to a previously sent command
|
|
|
|
if (!cl.read_op)
|
|
|
|
{
|
|
|
|
cl.read_op = new osd_op_t;
|
|
|
|
cl.read_op->peer_fd = peer_fd;
|
|
|
|
}
|
|
|
|
cl.read_op->op_type = OSD_OP_IN;
|
2020-02-23 19:03:06 +03:00
|
|
|
cl.read_buf = &cl.read_op->req.buf;
|
2019-12-28 01:25:55 +03:00
|
|
|
cl.read_remaining = OSD_PACKET_SIZE;
|
|
|
|
cl.read_state = CL_READ_OP;
|
|
|
|
}
|
|
|
|
cl.read_iov.iov_base = cl.read_buf;
|
|
|
|
cl.read_iov.iov_len = cl.read_remaining;
|
|
|
|
cl.read_msg.msg_iov = &cl.read_iov;
|
|
|
|
cl.read_msg.msg_iovlen = 1;
|
|
|
|
data->callback = [this, peer_fd](ring_data_t *data) { handle_read(data, peer_fd); };
|
2020-03-03 17:22:40 +03:00
|
|
|
my_uring_prep_recvmsg(sqe, cl.peer_fd_index, &cl.read_msg, 0);
|
|
|
|
sqe->flags |= IOSQE_FIXED_FILE;
|
2019-12-28 01:25:55 +03:00
|
|
|
}
|
|
|
|
read_ready_clients.clear();
|
|
|
|
}
|
|
|
|
|
|
|
|
void osd_t::handle_read(ring_data_t *data, int peer_fd)
|
|
|
|
{
|
2020-03-03 17:22:40 +03:00
|
|
|
int res = data->res;
|
2019-12-28 01:25:55 +03:00
|
|
|
auto cl_it = clients.find(peer_fd);
|
|
|
|
if (cl_it != clients.end())
|
|
|
|
{
|
|
|
|
auto & cl = cl_it->second;
|
2020-03-03 17:22:40 +03:00
|
|
|
if (res == -EAGAIN)
|
2020-02-29 01:46:12 +03:00
|
|
|
{
|
|
|
|
cl.read_ready--;
|
|
|
|
if (cl.read_ready > 0)
|
|
|
|
read_ready_clients.push_back(peer_fd);
|
|
|
|
return;
|
|
|
|
}
|
2020-03-03 17:22:40 +03:00
|
|
|
else if (res < 0)
|
2019-12-28 01:25:55 +03:00
|
|
|
{
|
|
|
|
// this is a client socket, so don't panic. just disconnect it
|
2020-03-03 17:22:40 +03:00
|
|
|
printf("Client %d socket read error: %d (%s). Disconnecting client\n", peer_fd, -res, strerror(-res));
|
2019-12-28 01:25:55 +03:00
|
|
|
stop_client(peer_fd);
|
|
|
|
return;
|
|
|
|
}
|
2020-02-29 01:46:12 +03:00
|
|
|
read_ready_clients.push_back(peer_fd);
|
2020-03-03 17:22:40 +03:00
|
|
|
if (res > 0)
|
2019-12-28 01:25:55 +03:00
|
|
|
{
|
2020-03-03 17:22:40 +03:00
|
|
|
cl.read_remaining -= res;
|
|
|
|
cl.read_buf += res;
|
2019-12-28 01:25:55 +03:00
|
|
|
if (cl.read_remaining <= 0)
|
|
|
|
{
|
|
|
|
cl.read_buf = NULL;
|
|
|
|
if (cl.read_state == CL_READ_OP)
|
|
|
|
{
|
2020-02-23 19:03:06 +03:00
|
|
|
if (cl.read_op->req.hdr.magic == SECONDARY_OSD_REPLY_MAGIC)
|
2019-12-28 01:25:55 +03:00
|
|
|
{
|
2020-02-25 01:20:45 +03:00
|
|
|
handle_reply_hdr(&cl);
|
2019-12-28 01:25:55 +03:00
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
2020-02-25 01:20:45 +03:00
|
|
|
handle_op_hdr(&cl);
|
2019-12-28 01:25:55 +03:00
|
|
|
}
|
|
|
|
}
|
|
|
|
else if (cl.read_state == CL_READ_DATA)
|
|
|
|
{
|
|
|
|
// Operation is ready
|
|
|
|
exec_op(cl.read_op);
|
|
|
|
cl.read_op = NULL;
|
|
|
|
cl.read_state = 0;
|
|
|
|
}
|
|
|
|
else if (cl.read_state == CL_READ_REPLY_DATA)
|
|
|
|
{
|
|
|
|
// Reply is ready
|
|
|
|
auto req_it = cl.sent_ops.find(cl.read_reply_id);
|
|
|
|
osd_op_t *request = req_it->second;
|
|
|
|
cl.sent_ops.erase(req_it);
|
|
|
|
cl.read_reply_id = 0;
|
|
|
|
cl.read_state = 0;
|
2020-02-28 12:25:54 +03:00
|
|
|
// Measure subop latency
|
2020-03-03 00:54:42 +03:00
|
|
|
timespec tv_end;
|
|
|
|
clock_gettime(CLOCK_REALTIME, &tv_end);
|
2020-02-28 12:25:54 +03:00
|
|
|
subop_stat_count[request->req.hdr.opcode]++;
|
|
|
|
subop_stat_sum[request->req.hdr.opcode] += (
|
|
|
|
(tv_end.tv_sec - request->tv_begin.tv_sec)*1000000 +
|
2020-03-03 00:54:42 +03:00
|
|
|
(tv_end.tv_nsec - request->tv_begin.tv_nsec)/1000
|
2020-02-28 12:25:54 +03:00
|
|
|
);
|
2020-01-22 02:36:14 +03:00
|
|
|
request->callback(request);
|
2019-12-28 01:25:55 +03:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2020-02-25 01:20:45 +03:00
|
|
|
void osd_t::handle_op_hdr(osd_client_t *cl)
|
2019-12-28 01:25:55 +03:00
|
|
|
{
|
|
|
|
osd_op_t *cur_op = cl->read_op;
|
2020-02-25 01:20:45 +03:00
|
|
|
if (cur_op->req.hdr.opcode == OSD_OP_SECONDARY_READ)
|
|
|
|
{
|
|
|
|
if (cur_op->req.sec_rw.len > 0)
|
|
|
|
cur_op->buf = memalign(512, cur_op->req.sec_rw.len);
|
|
|
|
cl->read_remaining = 0;
|
|
|
|
}
|
2020-02-25 22:52:03 +03:00
|
|
|
else if (cur_op->req.hdr.opcode == OSD_OP_SECONDARY_WRITE)
|
2019-12-28 01:25:55 +03:00
|
|
|
{
|
2020-02-25 01:20:45 +03:00
|
|
|
if (cur_op->req.sec_rw.len > 0)
|
|
|
|
cur_op->buf = memalign(512, cur_op->req.sec_rw.len);
|
|
|
|
cl->read_remaining = cur_op->req.sec_rw.len;
|
2019-12-28 01:25:55 +03:00
|
|
|
}
|
2020-02-25 22:52:03 +03:00
|
|
|
else if (cur_op->req.hdr.opcode == OSD_OP_SECONDARY_STABILIZE ||
|
|
|
|
cur_op->req.hdr.opcode == OSD_OP_SECONDARY_ROLLBACK)
|
|
|
|
{
|
|
|
|
if (cur_op->req.sec_stab.len > 0)
|
|
|
|
cur_op->buf = memalign(512, cur_op->req.sec_stab.len);
|
|
|
|
cl->read_remaining = cur_op->req.sec_stab.len;
|
|
|
|
}
|
2020-02-25 01:20:45 +03:00
|
|
|
else if (cur_op->req.hdr.opcode == OSD_OP_READ)
|
2019-12-28 01:25:55 +03:00
|
|
|
{
|
2020-02-25 01:20:45 +03:00
|
|
|
if (cur_op->req.rw.len > 0)
|
|
|
|
cur_op->buf = memalign(512, cur_op->req.rw.len);
|
|
|
|
cl->read_remaining = 0;
|
2019-12-28 01:25:55 +03:00
|
|
|
}
|
2020-02-25 01:20:45 +03:00
|
|
|
else if (cur_op->req.hdr.opcode == OSD_OP_WRITE)
|
|
|
|
{
|
|
|
|
if (cur_op->req.rw.len > 0)
|
|
|
|
cur_op->buf = memalign(512, cur_op->req.rw.len);
|
|
|
|
cl->read_remaining = cur_op->req.rw.len;
|
|
|
|
}
|
|
|
|
if (cl->read_remaining > 0)
|
2019-12-28 01:25:55 +03:00
|
|
|
{
|
|
|
|
// Read data
|
|
|
|
cl->read_buf = cur_op->buf;
|
|
|
|
cl->read_state = CL_READ_DATA;
|
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
|
|
|
// Operation is ready
|
|
|
|
cl->read_op = NULL;
|
|
|
|
cl->read_state = 0;
|
|
|
|
exec_op(cur_op);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2020-02-25 01:20:45 +03:00
|
|
|
void osd_t::handle_reply_hdr(osd_client_t *cl)
|
2019-12-28 01:25:55 +03:00
|
|
|
{
|
|
|
|
osd_op_t *cur_op = cl->read_op;
|
2020-02-23 19:03:06 +03:00
|
|
|
auto req_it = cl->sent_ops.find(cur_op->req.hdr.id);
|
2019-12-28 01:25:55 +03:00
|
|
|
if (req_it == cl->sent_ops.end())
|
|
|
|
{
|
|
|
|
// Command out of sync. Drop connection
|
2020-02-25 22:52:03 +03:00
|
|
|
printf("Client %d command out of sync: id %lu\n", cl->peer_fd, cur_op->req.hdr.id);
|
2019-12-28 01:25:55 +03:00
|
|
|
stop_client(cl->peer_fd);
|
|
|
|
return;
|
|
|
|
}
|
2020-02-23 19:03:06 +03:00
|
|
|
osd_op_t *op = req_it->second;
|
|
|
|
memcpy(op->reply.buf, cur_op->req.buf, OSD_PACKET_SIZE);
|
|
|
|
if (op->reply.hdr.opcode == OSD_OP_SECONDARY_READ &&
|
|
|
|
op->reply.hdr.retval > 0)
|
2019-12-28 01:25:55 +03:00
|
|
|
{
|
2020-02-24 01:01:34 +03:00
|
|
|
// Read data. In this case we assume that the buffer is preallocated by the caller (!)
|
|
|
|
assert(op->buf);
|
2019-12-28 01:25:55 +03:00
|
|
|
cl->read_state = CL_READ_REPLY_DATA;
|
2020-02-23 19:03:06 +03:00
|
|
|
cl->read_reply_id = op->req.hdr.id;
|
|
|
|
cl->read_buf = op->buf;
|
|
|
|
cl->read_remaining = op->reply.hdr.retval;
|
2019-12-28 01:25:55 +03:00
|
|
|
}
|
2020-02-23 19:03:06 +03:00
|
|
|
else if (op->reply.hdr.opcode == OSD_OP_SECONDARY_LIST &&
|
|
|
|
op->reply.hdr.retval > 0)
|
2019-12-28 01:25:55 +03:00
|
|
|
{
|
2020-02-23 19:03:06 +03:00
|
|
|
op->buf = memalign(512, sizeof(obj_ver_id) * op->reply.hdr.retval);
|
2019-12-28 01:25:55 +03:00
|
|
|
cl->read_state = CL_READ_REPLY_DATA;
|
2020-02-23 19:03:06 +03:00
|
|
|
cl->read_reply_id = op->req.hdr.id;
|
|
|
|
cl->read_buf = op->buf;
|
|
|
|
cl->read_remaining = sizeof(obj_ver_id) * op->reply.hdr.retval;
|
2019-12-28 01:25:55 +03:00
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
|
|
|
cl->read_state = 0;
|
|
|
|
cl->sent_ops.erase(req_it);
|
2020-02-28 12:25:54 +03:00
|
|
|
// Measure subop latency
|
2020-03-03 00:54:42 +03:00
|
|
|
timespec tv_end;
|
|
|
|
clock_gettime(CLOCK_REALTIME, &tv_end);
|
2020-02-28 12:25:54 +03:00
|
|
|
subop_stat_count[op->req.hdr.opcode]++;
|
|
|
|
subop_stat_sum[op->req.hdr.opcode] += (
|
|
|
|
(tv_end.tv_sec - op->tv_begin.tv_sec)*1000000 +
|
2020-03-03 00:54:42 +03:00
|
|
|
(tv_end.tv_nsec - op->tv_begin.tv_nsec)/1000
|
2020-02-28 12:25:54 +03:00
|
|
|
);
|
2020-02-23 19:03:06 +03:00
|
|
|
op->callback(op);
|
2019-12-28 01:25:55 +03:00
|
|
|
}
|
|
|
|
}
|