Fix crashes on multiple OSD reconnects
Identify clients by pointers instead of peer_fd as peer may be dropped and reconnected between callbacks Yeah maybe I need some Rust, but ... maybe in the future :)
This commit is contained in:
+112
-107
@@ -8,21 +8,22 @@ void osd_messenger_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];
|
||||
if (cl.read_remaining < receive_buffer_size)
|
||||
osd_client_t *cl = clients[peer_fd];
|
||||
if (cl->read_remaining < receive_buffer_size)
|
||||
{
|
||||
cl.read_iov.iov_base = cl.in_buf;
|
||||
cl.read_iov.iov_len = receive_buffer_size;
|
||||
cl.read_msg.msg_iov = &cl.read_iov;
|
||||
cl.read_msg.msg_iovlen = 1;
|
||||
cl->read_iov.iov_base = cl->in_buf;
|
||||
cl->read_iov.iov_len = receive_buffer_size;
|
||||
cl->read_msg.msg_iov = &cl->read_iov;
|
||||
cl->read_msg.msg_iovlen = 1;
|
||||
}
|
||||
else
|
||||
{
|
||||
cl.read_iov.iov_base = 0;
|
||||
cl.read_iov.iov_len = cl.read_remaining;
|
||||
cl.read_msg.msg_iov = cl.recv_list.get_iovec();
|
||||
cl.read_msg.msg_iovlen = cl.recv_list.get_size();
|
||||
cl->read_iov.iov_base = 0;
|
||||
cl->read_iov.iov_len = cl->read_remaining;
|
||||
cl->read_msg.msg_iov = cl->recv_list.get_iovec();
|
||||
cl->read_msg.msg_iovlen = cl->recv_list.get_size();
|
||||
}
|
||||
cl->refs++;
|
||||
if (ringloop && !use_sync_send_recv)
|
||||
{
|
||||
io_uring_sqe* sqe = ringloop->get_sqe();
|
||||
@@ -32,111 +33,115 @@ void osd_messenger_t::read_requests()
|
||||
return;
|
||||
}
|
||||
ring_data_t* data = ((ring_data_t*)sqe->user_data);
|
||||
data->callback = [this, peer_fd](ring_data_t *data) { handle_read(data->res, peer_fd); };
|
||||
my_uring_prep_recvmsg(sqe, peer_fd, &cl.read_msg, 0);
|
||||
data->callback = [this, cl](ring_data_t *data) { handle_read(data->res, cl); };
|
||||
my_uring_prep_recvmsg(sqe, peer_fd, &cl->read_msg, 0);
|
||||
}
|
||||
else
|
||||
{
|
||||
int result = recvmsg(peer_fd, &cl.read_msg, 0);
|
||||
int result = recvmsg(peer_fd, &cl->read_msg, 0);
|
||||
if (result < 0)
|
||||
{
|
||||
result = -errno;
|
||||
}
|
||||
handle_read(result, peer_fd);
|
||||
handle_read(result, cl);
|
||||
}
|
||||
}
|
||||
read_ready_clients.clear();
|
||||
}
|
||||
|
||||
bool osd_messenger_t::handle_read(int result, int peer_fd)
|
||||
bool osd_messenger_t::handle_read(int result, osd_client_t *cl)
|
||||
{
|
||||
bool ret = false;
|
||||
auto cl_it = clients.find(peer_fd);
|
||||
if (cl_it != clients.end())
|
||||
cl->refs--;
|
||||
if (cl->peer_state == PEER_STOPPED)
|
||||
{
|
||||
auto & cl = cl_it->second;
|
||||
if (result <= 0 && result != -EAGAIN)
|
||||
if (cl->refs <= 0)
|
||||
{
|
||||
// this is a client socket, so don't panic on error. just disconnect it
|
||||
if (result != 0)
|
||||
{
|
||||
printf("Client %d socket read error: %d (%s). Disconnecting client\n", peer_fd, -result, strerror(-result));
|
||||
}
|
||||
stop_client(peer_fd);
|
||||
return false;
|
||||
delete cl;
|
||||
}
|
||||
if (result == -EAGAIN || result < cl.read_iov.iov_len)
|
||||
return false;
|
||||
}
|
||||
if (result <= 0 && result != -EAGAIN)
|
||||
{
|
||||
// this is a client socket, so don't panic on error. just disconnect it
|
||||
if (result != 0)
|
||||
{
|
||||
cl.read_ready--;
|
||||
if (cl.read_ready > 0)
|
||||
read_ready_clients.push_back(peer_fd);
|
||||
printf("Client %d socket read error: %d (%s). Disconnecting client\n", cl->peer_fd, -result, strerror(-result));
|
||||
}
|
||||
stop_client(cl->peer_fd);
|
||||
return false;
|
||||
}
|
||||
if (result == -EAGAIN || result < cl->read_iov.iov_len)
|
||||
{
|
||||
cl->read_ready--;
|
||||
if (cl->read_ready > 0)
|
||||
read_ready_clients.push_back(cl->peer_fd);
|
||||
}
|
||||
else
|
||||
{
|
||||
read_ready_clients.push_back(cl->peer_fd);
|
||||
}
|
||||
if (result > 0)
|
||||
{
|
||||
if (cl->read_iov.iov_base == cl->in_buf)
|
||||
{
|
||||
// Compose operation(s) from the buffer
|
||||
int remain = result;
|
||||
void *curbuf = cl->in_buf;
|
||||
while (remain > 0)
|
||||
{
|
||||
if (!cl->read_op)
|
||||
{
|
||||
cl->read_op = new osd_op_t;
|
||||
cl->read_op->peer_fd = cl->peer_fd;
|
||||
cl->read_op->op_type = OSD_OP_IN;
|
||||
cl->recv_list.push_back(cl->read_op->req.buf, OSD_PACKET_SIZE);
|
||||
cl->read_remaining = OSD_PACKET_SIZE;
|
||||
cl->read_state = CL_READ_HDR;
|
||||
}
|
||||
while (cl->recv_list.done < cl->recv_list.count && remain > 0)
|
||||
{
|
||||
iovec* cur = cl->recv_list.get_iovec();
|
||||
if (cur->iov_len > remain)
|
||||
{
|
||||
memcpy(cur->iov_base, curbuf, remain);
|
||||
cl->read_remaining -= remain;
|
||||
cur->iov_len -= remain;
|
||||
cur->iov_base += remain;
|
||||
remain = 0;
|
||||
}
|
||||
else
|
||||
{
|
||||
memcpy(cur->iov_base, curbuf, cur->iov_len);
|
||||
curbuf += cur->iov_len;
|
||||
cl->read_remaining -= cur->iov_len;
|
||||
remain -= cur->iov_len;
|
||||
cur->iov_len = 0;
|
||||
cl->recv_list.done++;
|
||||
}
|
||||
}
|
||||
if (cl->recv_list.done >= cl->recv_list.count)
|
||||
{
|
||||
if (!handle_finished_read(cl))
|
||||
{
|
||||
goto fin;
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
else
|
||||
{
|
||||
read_ready_clients.push_back(peer_fd);
|
||||
// Long data
|
||||
cl->read_remaining -= result;
|
||||
cl->recv_list.eat(result);
|
||||
if (cl->recv_list.done >= cl->recv_list.count)
|
||||
{
|
||||
handle_finished_read(cl);
|
||||
}
|
||||
}
|
||||
if (result > 0)
|
||||
if (result >= cl->read_iov.iov_len)
|
||||
{
|
||||
if (cl.read_iov.iov_base == cl.in_buf)
|
||||
{
|
||||
// Compose operation(s) from the buffer
|
||||
int remain = result;
|
||||
void *curbuf = cl.in_buf;
|
||||
while (remain > 0)
|
||||
{
|
||||
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;
|
||||
cl.recv_list.push_back(cl.read_op->req.buf, OSD_PACKET_SIZE);
|
||||
cl.read_remaining = OSD_PACKET_SIZE;
|
||||
cl.read_state = CL_READ_HDR;
|
||||
}
|
||||
while (cl.recv_list.done < cl.recv_list.count && remain > 0)
|
||||
{
|
||||
iovec* cur = cl.recv_list.get_iovec();
|
||||
if (cur->iov_len > remain)
|
||||
{
|
||||
memcpy(cur->iov_base, curbuf, remain);
|
||||
cl.read_remaining -= remain;
|
||||
cur->iov_len -= remain;
|
||||
cur->iov_base += remain;
|
||||
remain = 0;
|
||||
}
|
||||
else
|
||||
{
|
||||
memcpy(cur->iov_base, curbuf, cur->iov_len);
|
||||
curbuf += cur->iov_len;
|
||||
cl.read_remaining -= cur->iov_len;
|
||||
remain -= cur->iov_len;
|
||||
cur->iov_len = 0;
|
||||
cl.recv_list.done++;
|
||||
}
|
||||
}
|
||||
if (cl.recv_list.done >= cl.recv_list.count)
|
||||
{
|
||||
if (!handle_finished_read(cl))
|
||||
{
|
||||
goto fin;
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
else
|
||||
{
|
||||
// Long data
|
||||
cl.read_remaining -= result;
|
||||
cl.recv_list.eat(result);
|
||||
if (cl.recv_list.done >= cl.recv_list.count)
|
||||
{
|
||||
handle_finished_read(cl);
|
||||
}
|
||||
}
|
||||
if (result >= cl.read_iov.iov_len)
|
||||
{
|
||||
ret = true;
|
||||
}
|
||||
ret = true;
|
||||
}
|
||||
}
|
||||
fin:
|
||||
@@ -148,30 +153,30 @@ fin:
|
||||
return ret;
|
||||
}
|
||||
|
||||
bool osd_messenger_t::handle_finished_read(osd_client_t & cl)
|
||||
bool osd_messenger_t::handle_finished_read(osd_client_t *cl)
|
||||
{
|
||||
cl.recv_list.reset();
|
||||
if (cl.read_state == CL_READ_HDR)
|
||||
cl->recv_list.reset();
|
||||
if (cl->read_state == CL_READ_HDR)
|
||||
{
|
||||
if (cl.read_op->req.hdr.magic == SECONDARY_OSD_REPLY_MAGIC)
|
||||
return handle_reply_hdr(&cl);
|
||||
if (cl->read_op->req.hdr.magic == SECONDARY_OSD_REPLY_MAGIC)
|
||||
return handle_reply_hdr(cl);
|
||||
else
|
||||
handle_op_hdr(&cl);
|
||||
handle_op_hdr(cl);
|
||||
}
|
||||
else if (cl.read_state == CL_READ_DATA)
|
||||
else if (cl->read_state == CL_READ_DATA)
|
||||
{
|
||||
// Operation is ready
|
||||
cl.received_ops.push_back(cl.read_op);
|
||||
set_immediate.push_back([this, op = cl.read_op]() { exec_op(op); });
|
||||
cl.read_op = NULL;
|
||||
cl.read_state = 0;
|
||||
cl->received_ops.push_back(cl->read_op);
|
||||
set_immediate.push_back([this, op = cl->read_op]() { exec_op(op); });
|
||||
cl->read_op = NULL;
|
||||
cl->read_state = 0;
|
||||
}
|
||||
else if (cl.read_state == CL_READ_REPLY_DATA)
|
||||
else if (cl->read_state == CL_READ_REPLY_DATA)
|
||||
{
|
||||
// Reply is ready
|
||||
handle_reply_ready(cl.read_op);
|
||||
cl.read_op = NULL;
|
||||
cl.read_state = 0;
|
||||
handle_reply_ready(cl->read_op);
|
||||
cl->read_op = NULL;
|
||||
cl->read_state = 0;
|
||||
}
|
||||
else
|
||||
{
|
||||
|
||||
Reference in New Issue
Block a user