/*
* This file is part of the Distributed Network Block Device 3
*
* Copyright(c) 2019 Frederic Robra <frederic@robra.org>
* Parts copyright 2011-2012 Johann Latocha <johann@latocha.de>
*
* This file may be licensed under the terms of of the
* GNU General Public License Version 2 (the ``GPL'').
*
* Software distributed under the License is distributed
* on an ``AS IS'' basis, WITHOUT WARRANTY OF ANY KIND, either
* express or implied. See the GPL for the specific language
* governing rights and limitations.
*
* You should have received a copy of the GPL along with this
* program. If not, go to http://www.gnu.org/licenses/gpl.html
* or write to the Free Software Foundation, Inc.,
* 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
*
*/
#include <net/sock.h>
#include <linux/wait.h>
#include "dnbd3.h"
#include "clientconfig.h"
#define DNBD3_REQ_OP_SPECIAL REQ_OP_DRV_IN
#define DNBD3_REQ_OP_CONNECT REQ_OP_DRV_OUT
#define dnbd3_cmd_to_priv(req, cmd) (req)->cmd_flags = DNBD3_REQ_OP_SPECIAL | ((cmd) << REQ_FLAG_BITS)
#define dnbd3_connect(req) (req)->cmd_flags = DNBD3_REQ_OP_CONNECT | ((CMD_SELECT_IMAGE) << REQ_FLAG_BITS)
#define dnbd3_priv_to_cmd(req) ((req)->cmd_flags >> REQ_FLAG_BITS)
#define dnbd3_sock_create(af,type,proto,sock) sock_create_kern(&init_net, (af) == HOST_IP4 ? AF_INET : AF_INET6, type, proto, sock)
#define KEEPALIVE_TIMER (jiffies + (HZ * TIMER_INTERVAL_KEEPALIVE_PACKET))
#define DISCOVERY_TIMER (jiffies + (HZ * TIMER_INTERVAL_PROBE_NORMAL))
#define REQUEST_TIMEOUT (jiffies + (HZ * SOCKET_TIMEOUT_CLIENT_DATA))
#define init_msghdr(h) do { \
h.msg_name = NULL; \
h.msg_namelen = 0; \
h.msg_control = NULL; \
h.msg_controllen = 0; \
h.msg_flags = MSG_WAITALL | MSG_NOSIGNAL; \
} while (0)
static DECLARE_WAIT_QUEUE_HEAD(send_wq);
static volatile uint64_t send_wq_signal;
static int dnbd3_socket_connect(struct dnbd3_device *dev, struct dnbd3_server *server);
static int __dnbd3_socket_connect(struct dnbd3_server * server, struct dnbd3_sock *sock);
static int dnbd3_socket_disconnect(struct dnbd3_device *dev, struct dnbd3_server *server, struct dnbd3_sock *sock, bool wait_for_receiver);
static void dnbd3_print_host(struct dnbd3_host_t *host, char *msg)
{
if (host->type == HOST_IP4) {
printk(KERN_INFO "dnbd3: %s %pI4:%d\n", msg, host->addr, host->port);
} else {
printk(KERN_INFO "dnbd3: %s [%pI6]:%d\n", msg, host->addr, host->port);
}
}
static void dnbd3_print_server_list(struct dnbd3_device *dev)
{
int i;
dnbd3_print_host(&dev->initial_server.host, "initial server is");
for (i = 0; i < NUMBER_SERVERS; i++) {
if (dev->alt_servers[i].host.addr[0] != 0) {
dnbd3_print_host(&dev->alt_servers[i].host, "alternative server is");
}
}
}
static inline uint64_t dnbd3_to_wq_signal(int minor, uint16_t dnbd3_cmd, uint16_t sock_nr) {
return ((uint64_t) minor << 32) | ((uint32_t) dnbd3_cmd << 16) | sock_nr;
}
static uint64_t dnbd3_to_handle(uint32_t arg0, uint32_t arg1) {
return ((uint64_t) arg0 << 32) | arg1;
}
static uint32_t dnbd3_arg0_from_handle(uint64_t handle) {
return (uint32_t)(handle >> 32);
}
static uint32_t dnbd3_arg1_from_handle(uint64_t handle) {
return (uint32_t) handle;
}
int dnbd3_send_request(struct dnbd3_sock *sock, struct request *req, struct dnbd3_cmd *cmd)
{
dnbd3_request_t dnbd3_request;
struct msghdr msg;
struct kvec iov[2];
size_t iov_num = 1;
size_t send_len;
int result;
uint32_t tag;
uint64_t handle;
serialized_buffer_t payload_buffer;
sock->pending = req;
init_msghdr(msg);
dnbd3_request.magic = dnbd3_packet_magic;
switch (req_op(req)) {
case REQ_OP_READ:
printk(KERN_DEBUG "dnbd3: request operation read\n");
dnbd3_request.cmd = CMD_GET_BLOCK;
dnbd3_request.offset = blk_rq_pos(req) << 9; // *512
dnbd3_request.size = blk_rq_bytes(req); // bytes left to complete entire request
break;
case DNBD3_REQ_OP_SPECIAL:
printk(KERN_DEBUG "dnbd3: request operation special\n");
dnbd3_request.cmd = dnbd3_priv_to_cmd(req);
dnbd3_request.size = 0;
break;
case DNBD3_REQ_OP_CONNECT:
printk(KERN_DEBUG "dnbd3: request operation connect to %s\n", sock->device->imgname);
dnbd3_request.cmd = CMD_SELECT_IMAGE;
serializer_reset_write(&payload_buffer);
serializer_put_uint16(&payload_buffer, PROTOCOL_VERSION);
serializer_put_string(&payload_buffer, sock->device->imgname);
serializer_put_uint16(&payload_buffer, sock->device->rid);
serializer_put_uint8(&payload_buffer, 0); // is_server = false
iov[1].iov_base = &payload_buffer;
dnbd3_request.size = iov[1].iov_len = serializer_get_written_length(&payload_buffer);
iov_num = 2;
break;
default:
return -EIO;
}
sock->cookie++;
if (cmd != NULL) {
cmd->cookie = sock->cookie;
tag = blk_mq_unique_tag(req);
handle = dnbd3_to_handle(tag, sock->cookie);// ((uint64_t) tag << 32) | sock->cookie;
} else {
handle = sock->cookie;
}
memcpy(&dnbd3_request.handle, &handle, sizeof(handle));
fixup_request(dnbd3_request);
iov[0].iov_base = &dnbd3_request;
iov[0].iov_len = sizeof(dnbd3_request);
send_len = iov_num == 1 ? sizeof(dnbd3_request) : iov[0].iov_len + iov[1].iov_len;
if ((result = kernel_sendmsg(sock->sock, &msg, iov, iov_num, send_len)) != send_len) {
printk(KERN_ERR "dnbd3: connection to server lost\n");
goto error;
}
sock->pending = NULL;
result = 0;
error:
return result;
}
int dnbd3_send_request_blocking(struct dnbd3_sock *sock, int dnbd3_cmd)
{
int result = 0;
uint64_t handle;
struct request *req = kmalloc(sizeof(struct request), GFP_KERNEL);
printk(KERN_DEBUG "dnbd3: starting blocking request\n");
if (!req) {
printk(KERN_ERR "dnbd3: kmalloc failed\n");
goto error;
}
switch (dnbd3_cmd) {
case CMD_KEEPALIVE:
case CMD_GET_SERVERS:
dnbd3_cmd_to_priv(req, dnbd3_cmd);
break;
case CMD_SELECT_IMAGE:
dnbd3_connect(req);
break;
default:
printk(KERN_WARNING "dnbd3: unsupported command for blocking %d\n", dnbd3_cmd);
result = -EINVAL;
goto error;
}
mutex_lock(&sock->lock);
result = dnbd3_send_request(sock, req, NULL);
if (result) {
mutex_unlock(&sock->lock);
goto error;
}
send_wq_signal = 0;
handle = dnbd3_to_wq_signal(sock->device->minor, dnbd3_cmd, sock->sock_nr);
mutex_unlock(&sock->lock);
printk(KERN_DEBUG "dnbd3: blocking request going to sleep wait for handle %llu\n", handle);
if (wait_event_interruptible_timeout(send_wq, handle == send_wq_signal, 10 /* REQUEST_TIMEOUT*/) <= 0) { // timeout or interrupt
printk(KERN_WARNING "dndbd3: request timed out\n");
result = -EIO;
goto error;
}
printk(KERN_DEBUG "dnbd3: blocking request woke up with handle %llu\n", handle);
error:
if (req) {
kfree(req);
}
return result;
}
static void dnbd3_receive_worker(struct work_struct *work)
{
struct dnbd3_sock *sock = container_of(work, struct dnbd3_sock, receive);
struct dnbd3_device *dev = sock->device;
struct request *req = NULL;
dnbd3_reply_t dnbd3_reply;
struct dnbd3_cmd *cmd;
struct msghdr msg;
struct kvec iov;
struct req_iterator iter;
struct bio_vec bvec_inst;
struct bio_vec *bvec = &bvec_inst;
sigset_t blocked, oldset;
void *kaddr;
uint32_t tag, cookie;
uint16_t hwq, rid;
int result, count, remaining;
uint64_t reported_size, handle;
char *name;
serialized_buffer_t payload_buffer;
init_msghdr(msg);
while(sock->sock && sock->server) {
iov.iov_base = &dnbd3_reply;
iov.iov_len = sizeof(dnbd3_reply);
result = kernel_recvmsg(sock->sock, &msg, &iov, 1, sizeof(dnbd3_reply), msg.msg_flags);
if (result == -EAGAIN) {
continue;
} else if (result <= 0) {
printk(KERN_ERR "dnbd3: connection to server lost %d\n", result);
goto error;
}
fixup_reply(dnbd3_reply);
// check error
if (dnbd3_reply.magic != dnbd3_packet_magic) {
printk(KERN_ERR "dnbd3: wrong magic packet\n");
result = -EIO;
goto error;
}
if (dnbd3_reply.cmd == 0) {
printk(KERN_ERR "dnbd3: command was 0\n");
result = -EIO;
goto error;
}
switch (dnbd3_reply.cmd) {
case CMD_GET_BLOCK:
printk(KERN_DEBUG "dnbd3: handle is %llu\n", dnbd3_reply.handle);
memcpy(&handle, &dnbd3_reply.handle, sizeof(handle));
cookie = dnbd3_arg1_from_handle(handle);
tag = dnbd3_arg0_from_handle(handle);
hwq = blk_mq_unique_tag_to_hwq(tag);
if (hwq < dev->tag_set.nr_hw_queues)
req = blk_mq_tag_to_rq(dev->tag_set.tags[hwq], blk_mq_unique_tag_to_tag(tag));
if (!req || !blk_mq_request_started(req)) {
dev_err(disk_to_dev(dev->disk), "Unexpected reply (%d) %p\n", tag, req);
continue;
}
cmd = blk_mq_rq_to_pdu(req);
mutex_lock(&cmd->lock);
if (cmd->cookie != cookie) {
dev_err(disk_to_dev(dev->disk), "Double reply on req %p, cookie %u, handle cookie %u\n",
req, cmd->cookie, cookie);
mutex_unlock(&cmd->lock);
continue;
}
rq_for_each_segment(bvec_inst, req, iter) {
siginitsetinv(&blocked, sigmask(SIGKILL));
sigprocmask(SIG_SETMASK, &blocked, &oldset);
kaddr = kmap(bvec->bv_page) + bvec->bv_offset;
iov.iov_base = kaddr;
iov.iov_len = bvec->bv_len;
result = kernel_recvmsg(sock->sock, &msg, &iov, 1, bvec->bv_len, msg.msg_flags);
if (result != bvec->bv_len) {
kunmap(bvec->bv_page);
sigprocmask(SIG_SETMASK, &oldset, NULL );
printk(KERN_ERR "dnbd3: could not receive form net to block layer\n");
mutex_unlock(&cmd->lock);
continue;
}
kunmap(bvec->bv_page);
sigprocmask(SIG_SETMASK, &oldset, NULL );
}
mutex_unlock(&cmd->lock);
blk_mq_end_request(req, 0);
break;
case CMD_GET_SERVERS:
printk(KERN_DEBUG "dnbd3: get servers received\n");
mutex_lock(&dev->device_lock);
if (!dev->use_server_provided_alts) {
remaining = dnbd3_reply.size;
goto consume_payload;
}
dev->new_servers_num = 0;
count = MIN(NUMBER_SERVERS, dnbd3_reply.size / sizeof(dnbd3_server_entry_t));
if (count != 0) {
iov.iov_base = dev->new_servers;
iov.iov_len = count * sizeof(dnbd3_server_entry_t);
result = kernel_recvmsg(sock->sock, &msg, &iov, 1, (count * sizeof(dnbd3_server_entry_t)), msg.msg_flags);
if (result != (count * sizeof(dnbd3_server_entry_t))) {
printk(KERN_ERR "dnbd3: failed to get servers\n");
mutex_unlock(&dev->device_lock);
goto error;
}
dev->new_servers_num = count;
}
// If there were more servers than accepted, remove the remaining data from the socket buffer
remaining = dnbd3_reply.size - (count * sizeof(dnbd3_server_entry_t));
consume_payload:
while (remaining > 0) {
count = MIN(sizeof(dnbd3_reply), remaining); // Abuse the reply struct as the receive buffer
iov.iov_base = &dnbd3_reply;
iov.iov_len = count;
result = kernel_recvmsg(sock->sock, &msg, &iov, 1, count, msg.msg_flags);
if (result <= 0) {
printk(KERN_ERR "dnbd3: failed to receive payload from get servers\n");
mutex_unlock(&dev->device_lock);
goto error;
}
}
mutex_unlock(&dev->device_lock);
break;
case CMD_LATEST_RID:
if (dnbd3_reply.size != 2) {
printk(KERN_ERR "dnbd3: failed to get latest rid, wrong size\n");
goto error;
}
printk(KERN_DEBUG "dnbd3: latest rid received\n");
iov.iov_base = &rid;
iov.iov_len = sizeof(rid);
result = kernel_recvmsg(sock->sock, &msg, &iov, 1, iov.iov_len, msg.msg_flags);
if (result <= 0) {
printk(KERN_ERR "dnbd3: failed to get latest rid\n");
goto error;
}
rid = net_order_16(rid);
printk("Latest rid of %s is %d (currently using %d)\n", dev->imgname, (int)rid, (int)dev->rid);
dev->update_available = (rid > dev->rid ? 1 : 0);
break;
case CMD_KEEPALIVE:
if (dnbd3_reply.size != 0) {
printk(KERN_ERR "dnbd3: got keep alive packet with payload\n");
goto error;
}
printk(KERN_DEBUG "dnbd3: keep alive received\n");
break;
case CMD_SELECT_IMAGE:
printk(KERN_DEBUG "dnbd3: select image received\n");
// receive reply payload
iov.iov_base = &payload_buffer;
iov.iov_len = dnbd3_reply.size;
result = kernel_recvmsg(sock->sock, &msg, &iov, 1, dnbd3_reply.size, msg.msg_flags);
if (result != dnbd3_reply.size) {
printk(KERN_ERR "dnbd3: could not read CMD_SELECT_IMAGE payload on handshake, size is %d and should be%d\n",
result, dnbd3_reply.size);
goto error;
}
// handle/check reply payload
serializer_reset_read(&payload_buffer, dnbd3_reply.size);
sock->server->protocol_version = serializer_get_uint16(&payload_buffer);
if (sock->server->protocol_version < MIN_SUPPORTED_SERVER) {
printk(KERN_ERR "dnbd3: server version is lower than min supported version\n");
goto error;
}
name = serializer_get_string(&payload_buffer);
rid = serializer_get_uint16(&payload_buffer);
if (dev->rid != rid && strcmp(name, dev->imgname) != 0) {
printk(KERN_ERR "dnbd3: server offers image '%s', requested '%s'\n", name, dev->imgname);
goto error;
}
reported_size = serializer_get_uint64(&payload_buffer);
if (!dev->reported_size) {
if (reported_size < 4096) {
printk(KERN_ERR "dnbd3: reported size by server is < 4096\n");
goto error;
}
dev->reported_size = reported_size;
set_capacity(dev->disk, dev->reported_size >> 9); /* 512 Byte blocks */
} else if (dev->reported_size != reported_size) {
printk(KERN_ERR "dnbd3: reported size by server is %llu but should be %llu\n", reported_size, dev->reported_size);
}
break;
default:
printk(KERN_WARNING "dnbd3: Unknown command (Receive)\n");
break;
}
error:
handle = dnbd3_to_wq_signal(dev->minor, dnbd3_reply.cmd, sock->sock_nr);
printk(KERN_DEBUG "dnbd3: try to wake up queue with handle %llu\n", handle);
send_wq_signal = handle;
wake_up_interruptible(&send_wq);
if (result == 0) {
printk(KERN_INFO "dnbd3: result is 0, socket seems to be down\n");
// dnbd3_socket_disconnect(dev, NULL, sock, false);//TODO use panic or something or start worker to reconnect?
break; //the socket seems to be down
} else if (result < 0) {
sock->server->failures++; // discovery takes care of to many failures
printk(KERN_WARNING "dnbd3: receive error happened %d, total failures %d\n", result, sock->server->failures);
}
printk(KERN_DEBUG "dnbd3: receive completed, waiting for next receive\n");
}
printk(KERN_DEBUG "dnbd3: receive work queue is stopped\n");
}
void dnbd3_keepalive_timer(struct timer_list *arg)
{
struct dnbd3_sock *sock = container_of(arg, struct dnbd3_sock, keepalive_timer);
queue_work(dnbd3_wq, &sock->keepalive);
sock->keepalive_timer.expires = KEEPALIVE_TIMER;
add_timer(&sock->keepalive_timer);
}
static void dnbd3_keepalive_worker(struct work_struct *work)
{
struct dnbd3_sock *sock = container_of(work, struct dnbd3_sock, keepalive);
printk(KERN_DEBUG "dnbd3: starting keepalive worker\n");
dnbd3_send_request_blocking(sock, CMD_KEEPALIVE);
sock->heartbeat_count++;
}
static void dnbd3_discovery_timer(struct timer_list *arg)
{
struct dnbd3_device *dev = container_of(arg, struct dnbd3_device, discovery_timer);
queue_work(dnbd3_wq, &dev->discovery);
dev->discovery_timer.expires = DISCOVERY_TIMER;
add_timer(&dev->discovery_timer);
}
static void dnbd3_discovery_worker(struct work_struct *work)
{
struct dnbd3_device *dev = container_of(work, struct dnbd3_device, discovery);
struct dnbd3_sock *sock = NULL;
int i, j;
struct dnbd3_server *existing_server, *free_server, *failed_server;
dnbd3_server_entry_t *new_server;
struct kvec iov;
struct timeval start, end;
dnbd3_request_t dnbd3_request;
dnbd3_reply_t dnbd3_reply;
struct msghdr msg;
char *buf, *name;
struct request *req = NULL;
uint64_t rtt;
serialized_buffer_t *payload;
uint64_t filesize;
uint16_t rid;
printk(KERN_DEBUG "dnbd3: starting discovery worker\n");
dnbd3_send_request_blocking(&dev->socks[0], CMD_GET_SERVERS);
printk(KERN_DEBUG "dnbd3: new server num is %d\n", dev->new_servers_num);
if (dev->new_servers_num) {
mutex_lock(&dev->device_lock);
for (i = 0; i < dev->new_servers_num; i++) {
new_server = &dev->new_servers[i];
if (new_server->host.type == HOST_IP4 || new_server->host.type == HOST_IP6) {
existing_server = NULL;
free_server = NULL;
failed_server = NULL;
// find servers in alt servers
for (j = 0; j < NUMBER_SERVERS; j++) {
if ((new_server->host.type == dev->alt_servers[j].host.type)
&& (new_server->host.port == dev->alt_servers[j].host.port)
&& (0 == memcmp(new_server->host.addr, dev->alt_servers[j].host.addr,
(new_server->host.type == HOST_IP4 ? 4 : 16)))) {
existing_server = &dev->alt_servers[j];
} else if (dev->alt_servers[j].host.type == 0) {
free_server = &dev->alt_servers[j];
} else if (dev->alt_servers[j].failures > 20) {
failed_server = &dev->alt_servers[j];
}
}
if (existing_server) {
if (new_server->failures == 1) { // remove is requested
dnbd3_print_host(&existing_server->host, "remove server");
dnbd3_socket_disconnect(dev, existing_server, NULL, true); // TODO what to do when only one connection?
existing_server->host.type = 0;
}
// ADD, so just reset fail counter
// existing_server->failures = 0; makes no sense?
continue;
} else if (free_server) {
free_server->host = new_server->host;
} else if (failed_server) {
failed_server->host = new_server->host;
free_server = failed_server;
} else {
//no server found to replace
continue;
}
dnbd3_print_host(&free_server->host, "got new alt server");
free_server->failures = 0;
free_server->protocol_version = 0;
free_server->rtts[0] = free_server->rtts[1] = free_server->rtts[2] = free_server->rtts[3] = RTT_UNREACHABLE;
}
}
dev->new_servers_num = 0;
mutex_unlock(&dev->device_lock);
}
buf = kmalloc(RTT_BLOCK_SIZE, GFP_KERNEL);
if (!buf) {
printk(KERN_ERR "dnbd3: kmalloc failed\n");
goto error;
}
payload = (serialized_buffer_t *)buf;
req = kmalloc(sizeof(struct request), GFP_KERNEL);
if (!req) {
printk(KERN_ERR "dnbd3: kmalloc failed\n");
goto error;
}
sock = kmalloc(sizeof(struct dnbd3_sock), GFP_KERNEL);
if (!sock) {
printk(KERN_ERR "dnbd3: kmalloc failed\n");
goto error;
}
// measure rtt for all alt servers
for (i = 0; i < NUMBER_SERVERS; i++) {
existing_server = &dev->alt_servers[i];
if (existing_server->host.type) {
sock->sock = NULL;
sock->device = dev;
init_msghdr(msg);
if (__dnbd3_socket_connect(existing_server, sock)) {
printk(KERN_ERR "dnbd3: socket connect failed in rtt measurement\n");
goto rtt_error;
}
dnbd3_connect(req);
if (dnbd3_send_request(sock, req, NULL)) {
printk(KERN_ERR "dnbd3: request select image failed in rtt measurement\n");
goto rtt_error;
}
iov.iov_base = &dnbd3_reply;
iov.iov_len = sizeof(dnbd3_reply);
if (kernel_recvmsg(sock->sock, &msg, &iov, 1, sizeof(dnbd3_reply), msg.msg_flags) != sizeof(dnbd3_reply)) {
printk(KERN_ERR "dnbd3: receive select image failed in rtt measurement %d\n", j);
goto rtt_error;
}
fixup_reply(dnbd3_reply);
if (dnbd3_reply.magic != dnbd3_packet_magic || dnbd3_reply.cmd != CMD_SELECT_IMAGE || dnbd3_reply.size < 4) {
printk(KERN_ERR "dnbd3: receive select image wrong header in rtt measurement\n");
goto rtt_error;
}
// receive data
iov.iov_base = payload;
iov.iov_len = dnbd3_reply.size;
if (kernel_recvmsg(sock->sock, &msg, &iov, 1, dnbd3_reply.size, msg.msg_flags) != dnbd3_reply.size) {
printk(KERN_ERR "dnbd3: receive data select image failed in rtt measurement\n");
goto rtt_error;
}
serializer_reset_read(payload, dnbd3_reply.size);
existing_server->protocol_version = serializer_get_uint16(payload);
if (existing_server->protocol_version < MIN_SUPPORTED_SERVER) {
printk(KERN_ERR "dnbd3: server version to old in rtt measurement\n");
goto rtt_error;
}
name = serializer_get_string(payload);
if (name == NULL ) {
printk(KERN_ERR "dnbd3: server did not supply an image name in rtt measurement\n");
goto rtt_error;
}
if (strcmp(name, dev->imgname) != 0) {
printk(KERN_ERR "dnbd3: image name %s does not match requested %s in rtt measurement\n", name, dev->imgname);
goto rtt_error;
}
// TODO rid is 1 but dev->rid is 0
rid = serializer_get_uint16(payload);
// if (rid != dev->rid) {
// printk(KERN_ERR "dnbd3: rid %d does not match requested %d in rtt measurement\n", rid, dev->rid);
// goto rtt_error;
// }
filesize = serializer_get_uint64(payload);
if (filesize != dev->reported_size) {
printk(KERN_ERR "dnbd3: image size %llu does not match requested %llu in rtt measurement\n", filesize, dev->reported_size);
goto rtt_error;
}
// Request block
dnbd3_request.cmd = CMD_GET_BLOCK;
// Do *NOT* pick a random block as it has proven to cause severe
// cache thrashing on the server
dnbd3_request.offset = 0;
dnbd3_request.size = RTT_BLOCK_SIZE;
fixup_request(dnbd3_request);
iov.iov_base = &dnbd3_request;
iov.iov_len = sizeof(dnbd3_request);
init_msghdr(msg);
// start rtt measurement
do_gettimeofday(&start);
if (kernel_sendmsg(sock->sock, &msg, &iov, 1, sizeof(dnbd3_request)) <= 0) {
printk(KERN_ERR "dnbd3: request test block failed in rtt measurement\n");
goto rtt_error;
}
// receive net reply
iov.iov_base = &dnbd3_reply;
iov.iov_len = sizeof(dnbd3_reply);
if ((j = kernel_recvmsg(sock->sock, &msg, &iov, 1, sizeof(dnbd3_reply), msg.msg_flags)) != sizeof(dnbd3_reply)) {
printk(KERN_ERR "dnbd3: receive header test block failed in rtt measurement %d %ld\n", j, sizeof(dnbd3_reply));
goto rtt_error;
}
fixup_reply(dnbd3_reply);
if (dnbd3_reply.magic != dnbd3_packet_magic|| dnbd3_reply.cmd != CMD_GET_BLOCK || dnbd3_reply.size != RTT_BLOCK_SIZE) {
printk(KERN_ERR "dnbd3: receive header cmd test block failed in rtt measurement\n");
goto rtt_error;
}
// receive data
iov.iov_base = buf;
iov.iov_len = RTT_BLOCK_SIZE;
if (kernel_recvmsg(sock->sock, &msg, &iov, 1, dnbd3_reply.size, msg.msg_flags) != RTT_BLOCK_SIZE) {
printk(KERN_ERR "dnbd3: receive test block failed in rtt measurement\n");
goto rtt_error;
}
do_gettimeofday(&end); // end rtt measurement
rtt = (uint64_t)((end.tv_sec - start.tv_sec) * 1000000ull + (end.tv_usec - start.tv_usec));
printk(KERN_DEBUG "dnbd3: new rrt for %pI4 is %llu\n", existing_server->host.addr, rtt);
rtt_error:
if (sock->sock) {
kernel_sock_shutdown(sock->sock, SHUT_RDWR);
sock->server = NULL;
}
if (sock->sock) {
sock_release(sock->sock);
sock->sock = NULL;
}
}
}
error:
if (buf) {
kfree(buf);
buf = NULL;
}
if (req) {
kfree(req);
req = NULL;
}
if (sock) {
kfree(sock);
sock = NULL;
}
}
static int __dnbd3_socket_connect(struct dnbd3_server *server, struct dnbd3_sock *sock)
{
int result = 0;
struct timeval timeout;
if (server->host.port == 0 || server->host.type == 0) {
printk(KERN_ERR "dnbd3: host or port not set\n");
return -EIO;
}
if (sock->sock) {
printk(KERN_WARNING "dnbd3: socket already connected\n");
return -EIO;
}
timeout.tv_sec = SOCKET_TIMEOUT_CLIENT_DATA;
timeout.tv_usec = 0;
if ((result = dnbd3_sock_create(server->host.type, SOCK_STREAM, IPPROTO_TCP, &sock->sock)) < 0) {
printk(KERN_ERR "dnbd3: could not create socket\n");
goto error;
}
kernel_setsockopt(sock->sock, SOL_SOCKET, SO_SNDTIMEO, (char *)&timeout, sizeof(timeout));
kernel_setsockopt(sock->sock, SOL_SOCKET, SO_RCVTIMEO, (char *)&timeout, sizeof(timeout));
sock->sock->sk->sk_allocation = GFP_NOIO;
if (server->host.type == HOST_IP4) {
struct sockaddr_in sin;
memset(&sin, 0, sizeof(sin));
sin.sin_family = AF_INET;
memcpy(&(sin.sin_addr), server->host.addr, 4);
sin.sin_port = server->host.port;
if ((result = kernel_connect(sock->sock, (struct sockaddr *)&sin, sizeof(sin), 0)) != 0) {
printk(KERN_ERR "dnbd3: connection to host failed (ipv4)\n");
goto error;
}
} else {
struct sockaddr_in6 sin;
memset(&sin, 0, sizeof(sin));
sin.sin6_family = AF_INET6;
memcpy(&(sin.sin6_addr), server->host.addr, 16);
sin.sin6_port = server->host.port;
if ((result = kernel_connect(sock->sock, (struct sockaddr *)&sin, sizeof(sin), 0)) != 0){
printk(KERN_ERR "dnbd3: connection to host failed (ipv6)\n");
goto error;
}
}
return 0;
error:
if (sock->sock) {
sock_release(sock->sock);
sock->sock = NULL;
}
return result;
}
static int dnbd3_socket_connect(struct dnbd3_device *dev, struct dnbd3_server *server)
{
int i;
int result = -EIO;
struct dnbd3_sock *sock = NULL;
for (i = 0; i < NUMBER_CONNECTIONS; i++) {
if (!dev->socks[i].sock) {
sock = &dev->socks[i];
break;
}
}
if (sock == NULL) {
printk(KERN_WARNING "dnbd3: could not connect to socket, to many connections\n");
return -EIO;
}
sock->server = server;
printk(KERN_DEBUG "dnbd3: socket connect device %i\n", dev->minor);
mutex_init(&sock->lock);
mutex_lock(&sock->lock);
__dnbd3_socket_connect(server, sock);
mutex_unlock(&sock->lock);
if (!sock->sock) {
printk(KERN_DEBUG "dnbd3: socket is not connected\n");
result = -EIO;
goto error;
}
// start the receiver
INIT_WORK(&sock->receive, dnbd3_receive_worker);
queue_work(dnbd3_wq, &sock->receive);
result = dnbd3_send_request_blocking(sock, CMD_SELECT_IMAGE);
if (result) {
printk(KERN_ERR "dnbd3: connection to image %s failed\n", dev->imgname);
goto error;
}
printk(KERN_DEBUG "dnbd3: connected to image %s, filesize %llu\n", dev->imgname, dev->reported_size);
// add heartbeat timer and scheduler for the command
INIT_WORK(&sock->keepalive, dnbd3_keepalive_worker);
sock->heartbeat_count = 0;
timer_setup(&sock->keepalive_timer, dnbd3_keepalive_timer, 0);
sock->keepalive_timer.expires = KEEPALIVE_TIMER;
add_timer(&sock->keepalive_timer);
// kfree(req);
return 0;
error:
if (sock->sock) {
sock_release(sock->sock);
sock->sock = NULL;
}
return result;
}
static int dnbd3_socket_disconnect(struct dnbd3_device *dev, struct dnbd3_server *server, struct dnbd3_sock *sock, bool wait_for_receiver)
{
int i;
int sock_alive = 0;
for (i = 0; i < NUMBER_CONNECTIONS; i++) {
if (sock == NULL && dev->socks[i].server == server) {
sock = &dev->socks[i];
}
if (dev->socks[i].sock) {
sock_alive++;
}
}
if (!sock || !sock->sock) {
printk(KERN_WARNING "dnbd3: could not find socket to disconnect\n");
return -EIO;
}
if (sock_alive <= 1) {
printk(KERN_INFO "dnbd3: shutting down last socket and stopping discovery\n");
del_timer_sync(&dev->discovery_timer);
cancel_work_sync(&dev->discovery);
}
printk(KERN_DEBUG "dnbd3: stopping keepalive\n");
del_timer_sync(&sock->keepalive_timer);
cancel_work_sync(&sock->keepalive);
printk(KERN_DEBUG "dnbd3: socket disconnect device %i\n", dev->minor);
mutex_lock(&sock->lock);
/*
* Important sequence to shut down socket
* 1. kernel_sock_shutdown
* socket shutdown, receiver which hangs in kernel_recvmsg returns 0
* 2. cancel_work_sync(receiver)
* wait for the receiver to finish, so the socket is not usesd anymore
* 3. sock_release
* release the socket and set to NULL
*/
if (sock->sock) {
kernel_sock_shutdown(sock->sock, SHUT_RDWR);
sock->server = NULL;
}
mutex_unlock(&sock->lock);
mutex_destroy(&sock->lock);
if (wait_for_receiver) {
printk(KERN_DEBUG "dnbd3: cancel receiver work device %i\n", dev->minor);
cancel_work_sync(&sock->receive);
}
if (sock->sock) {
sock_release(sock->sock);
sock->sock = NULL;
}
return 0;
}
int dnbd3_net_disconnect(struct dnbd3_device *dev)
{
int i;
int result = 0;
for (i = 0; i < NUMBER_CONNECTIONS; i++) {
if (dev->socks[i].sock) {
if (dnbd3_socket_disconnect(dev, NULL, &dev->socks[i], true)) {
result = -EIO;
}
}
}
return result;
}
int dnbd3_net_connect(struct dnbd3_device *dev)
{
// TODO decide which socket to connect
int result;
if (dnbd3_socket_connect(dev, &dev->alt_servers[0]) == 0) {
dnbd3_print_server_list(dev);
INIT_WORK(&dev->discovery, dnbd3_discovery_worker);
timer_setup(&dev->discovery_timer, dnbd3_discovery_timer, 0);
dev->discovery_timer.expires = DISCOVERY_TIMER;
add_timer(&dev->discovery_timer);
// let it discover alt servers
queue_work(dnbd3_wq, &dev->discovery);
result = 0;
} else {
printk(KERN_ERR "dnbd3: failed to connect to initial server\n");
result = -ENOENT;
dev->imgname = NULL;
dev->socks[0].server = NULL;
}
return result;
}