summaryrefslogtreecommitdiffstats
path: root/src/kernel/net.c
diff options
context:
space:
mode:
Diffstat (limited to 'src/kernel/net.c')
-rw-r--r--src/kernel/net.c608
1 files changed, 10 insertions, 598 deletions
diff --git a/src/kernel/net.c b/src/kernel/net.c
index 1207d0a..cb4e76c 100644
--- a/src/kernel/net.c
+++ b/src/kernel/net.c
@@ -25,613 +25,26 @@
#include <linux/sort.h>
#include "net.h"
+#include "net-txrx.h"
#include "utils.h"
#include "clientconfig.h"
#include "mq.h"
-#define DNBD3_REQ_OP_SPECIAL REQ_OP_DRV_IN
-#define DNBD3_REQ_OP_CONNECT REQ_OP_DRV_OUT
-
-#define dnbd3_cmd_to_op_special(req, cmd) \
- (req)->cmd_flags = DNBD3_REQ_OP_SPECIAL | ((cmd) << REQ_FLAG_BITS)
-
-#define dnbd3_op_special_to_cmd(req) \
- ((req)->cmd_flags >> REQ_FLAG_BITS)
-
-#define dnbd3_connect_to_req(req) \
- (req)->cmd_flags = DNBD3_REQ_OP_CONNECT \
- | ((CMD_SELECT_IMAGE) << REQ_FLAG_BITS)
-
-#define dnbd3_test_block_to_req(req) \
- do { \
- (req)->cmd_flags = REQ_OP_READ; \
- (req)->__data_len = RTT_BLOCK_SIZE; \
- (req)->__sector = 0; \
- } while (0)
-
#define dnbd3_sock_create(af,type,proto,sock) \
sock_create_kern(&init_net, (af) == HOST_IP4 ? AF_INET : AF_INET6, \
type, proto, sock)
-#define REQUEST_TIMEOUT \
- (HZ * SOCKET_TIMEOUT_CLIENT_DATA)
+#define dnbd3_is_sock_alive(s) ((s).sock && (s).server)
-#define dnbd3_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 int __dnbd3_socket_connect(struct dnbd3_sock *sock,
- struct dnbd3_server * server);
+ struct dnbd3_server *server);
static int dnbd3_socket_connect(struct dnbd3_sock *sock,
- struct dnbd3_server * server);
+ struct dnbd3_server *server);
static int dnbd3_socket_disconnect(struct dnbd3_sock *sock);
-static uint64_t dnbd3_average_rtt(struct dnbd3_server *server)
-{
- int i, j = 0;
- uint64_t avg = 0;
- for (i = 0; i < 4; i++) {
- avg += server->rtts[i];
- j += server->rtts[i] == 0 ? 0 : 1;
- }
- if (avg == 0) {
- return RTT_UNKNOWN;
- } else {
- avg = avg / j;
- avg += server->failures * avg / 10;
- return avg;
- }
-}
-
-/*
- * Methods for request and receive commands
- */
-
-
-/**
- * dnbd3_to_handle - convert tag and cookie to handle
- * @tag: the tag to convert
- * @cookie: the cookie to convert
- */
-static inline uint64_t dnbd3_to_handle(uint32_t tag, uint32_t cookie) {
- return ((uint64_t) tag << 32) | cookie;
-}
-
-/**
- * dnbd3_tag_from_handle - get tag from handle
- * @handle: the handle
- */
-static inline uint32_t dnbd3_tag_from_handle(uint64_t handle) {
- return (uint32_t)(handle >> 32);
-}
-
-/**
- * dnbd3_cookie_from_handle - get cookie from handle
- * @handle: the handle
- */
-static inline uint32_t dnbd3_cookie_from_handle(uint64_t handle) {
- return (uint32_t) handle;
-}
-
-/**
- * dnbd3_send_request - send a request
- * @sock: the socket where the request is send
- * @req: the request to send
- * @cmd: optional - the dnbd3_cmd from mq
- *
- * the tx_lock of the socket must be held
- */
-int dnbd3_send_request(struct dnbd3_sock *sock, struct request *req,
- struct dnbd3_cmd *cmd)
-{
- dnbd3_request_t request;
- struct msghdr msg;
- struct kvec iov[2];
- size_t iov_num = 1;
- size_t lng;
- int result;
- uint32_t tag;
- uint64_t handle;
- serialized_buffer_t payload_buffer;
- sock->pending = req;
- dnbd3_init_msghdr(msg);
-
- request.magic = dnbd3_packet_magic;
-
- switch (req_op(req)) {
- case REQ_OP_READ:
- debug_sock(sock, "request operation read");
- request.cmd = CMD_GET_BLOCK;
- request.offset = blk_rq_pos(req) << 9; // * 512
- request.size = blk_rq_bytes(req);
- break;
- case DNBD3_REQ_OP_SPECIAL:
- debug_sock(sock, "request operation special");
- request.cmd = dnbd3_op_special_to_cmd(req);
- request.size = 0;
- break;
- case DNBD3_REQ_OP_CONNECT:
- debug_sock(sock, "request operation connect to %s",
- sock->device->imgname);
- 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;
- request.size = serializer_get_written_length(&payload_buffer);
- iov[1].iov_len = request.size;
- 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);
- } else {
- handle = sock->cookie;
- }
- memcpy(&request.handle, &handle, sizeof(handle));
-
- fixup_request(request);
- iov[0].iov_base = &request;
- iov[0].iov_len = sizeof(request);
- lng = iov_num == 1 ? iov[0].iov_len : iov[0].iov_len + iov[1].iov_len;
- result = kernel_sendmsg(sock->sock, &msg, iov, iov_num, lng);
- if (result != lng) {
- error_sock(sock, "connection to server lost");
- if (cmd) {
- dnbd3_requeue_cmd(cmd);
- }
- sock->panic = true;
- sock->server->failures++;
- goto error;
- }
-
-error:
- sock->pending = NULL;
- return result;
-}
-
-
-/**
- * dnbd3_send_request_cmd - send a dndb3 cmd
- * @sock: the socket where the request is send
- * @dnbd3_cmd: the dnbd3 cmd to send
- */
-static int dnbd3_send_request_cmd(struct dnbd3_sock *sock, uint16_t dnbd3_cmd)
-{
- int result;
- struct request *req = kmalloc(sizeof(struct request), GFP_KERNEL);
- if (!req) {
- error_sock(sock, "kmalloc failed");
- result = -EIO;
- goto error;
- }
-
- switch (dnbd3_cmd) {
- case CMD_KEEPALIVE:
- case CMD_GET_SERVERS:
- dnbd3_cmd_to_op_special(req, dnbd3_cmd);
- break;
- case CMD_SELECT_IMAGE:
- dnbd3_connect_to_req(req);
- break;
- case CMD_GET_BLOCK:
- dnbd3_test_block_to_req(req);
- break;
- default:
- warn_sock(sock, "unsupported command %d", dnbd3_cmd);
- result = -EINVAL;
- goto error;
- }
-
- mutex_lock(&sock->tx_lock);
- sock->pending = req;
- result = dnbd3_send_request(sock, req, NULL);
- mutex_unlock(&sock->tx_lock);
-
-error:
- if (req) {
- kfree(req);
- }
- return result;
-}
-
-/**
- * dnbd3_receive_cmd - receive a command
- * @sock: the socket where the request is received
- * @reply: an unused reply will be filled with the reply of the server
- *
- * this method should be called directly after the dnbd3_send_request_ method
- */
-static int dnbd3_receive_cmd(struct dnbd3_sock *sock, dnbd3_reply_t *reply)
-{
- int result;
- struct msghdr msg;
- struct kvec iov;
- dnbd3_init_msghdr(msg);
- iov.iov_base = reply;
- iov.iov_len = sizeof(dnbd3_reply_t);
- result = kernel_recvmsg(sock->sock, &msg, &iov, 1, iov.iov_len,
- msg.msg_flags);
- if (result <= 0) {
- return result;
- }
- fixup_reply(dnbd3_reply);
-
- if (reply->magic != dnbd3_packet_magic) {
- error_sock(sock, "receive cmd wrong magic packet");
- return -EIO;
- }
-
- if (reply->cmd == 0) {
- error_sock(sock, "receive command was 0");
- return -EIO;
- }
- return result;
-}
-
-static int dnbd3_clear_socket(struct dnbd3_sock *sock, dnbd3_reply_t *reply,
- int remaining)
-{
- int result = 0;
- char *buf;
- struct kvec iov;
- struct msghdr msg;
- dnbd3_init_msghdr(msg);
- warn_sock(sock, "clearing socket %d bytes", remaining);
- buf = kmalloc(RTT_BLOCK_SIZE, GFP_KERNEL);
- if (!buf) {
- error_sock(sock, "kmalloc failed");
- return -EIO;
- }
- /* hold the tx_lock so no new requests are send */
- mutex_lock(&sock->tx_lock);
- iov.iov_base = buf;
- iov.iov_len = RTT_BLOCK_SIZE;
- while (remaining > 0) {
- result = kernel_recvmsg(sock->sock, &msg, &iov, 1, iov.iov_len,
- msg.msg_flags);
- if (result <= 0) {
- goto error;
- }
- remaining -= result;
- }
-
- debug_sock(sock, "cleared socket");
-error:
- mutex_unlock(&sock->tx_lock);
- if (buf) {
- kfree(buf);
- }
- return result;
-}
-
-/**
- * dnbd3_receive_cmd_get_block_mq - receive a block for mq
- * @sock: the socket where the request is received
- * @reply: the reply initialized by dnbd3_receive_cmd
- *
- * this method should be called directly after the dnbd3_receive_cmd method
- *
- * this method copies the data to user space according to the request which is
- * encoded in the handle by the send request method and decoded here.
- */
-static int dnbd3_receive_cmd_get_block_mq(struct dnbd3_sock *sock,
- dnbd3_reply_t *reply)
-{
- struct dnbd3_cmd *cmd;
- struct msghdr msg;
- struct request *req = NULL;
- struct kvec iov;
- struct req_iterator iter;
- struct bio_vec bvec_inst;
- struct dnbd3_device *dev = sock->device;
- struct bio_vec *bvec = &bvec_inst;
- sigset_t blocked, oldset;
- void *kaddr;
- uint32_t tag, cookie;
- uint16_t hwq;
- uint32_t remaining = reply->size;
- int result = 0;
- uint64_t handle;
- dnbd3_init_msghdr(msg);
-
- memcpy(&handle, &reply->handle, sizeof(handle));
- cookie = dnbd3_cookie_from_handle(handle);
- tag = dnbd3_tag_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)) {
- error_sock(sock, "unexpected reply (%d) %p", tag, req);
- if (req) {
- debug_sock(sock, "requeue request");
- dnbd3_requeue_cmd(blk_mq_rq_to_pdu(req));
- }
- dnbd3_clear_socket(sock, reply, remaining);
- return -EIO;
- }
- cmd = blk_mq_rq_to_pdu(req);
-
- mutex_lock(&cmd->lock);
- if (cmd->cookie != cookie) {
- error_sock(sock, "double reply on req %p, cookie %u, handle cookie %u",
- req, cmd->cookie, cookie);
- mutex_unlock(&cmd->lock);
- dnbd3_clear_socket(sock, reply, remaining);
- return -EIO;
- }
-
- 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);
- remaining -= result;
- if (result != bvec->bv_len) {
- kunmap(bvec->bv_page);
- sigprocmask(SIG_SETMASK, &oldset, NULL );
- error_sock(sock, "could not receive from net to block layer");
- dnbd3_requeue_cmd(cmd);
- mutex_unlock(&cmd->lock);
- if (result >= 0) {
- dnbd3_clear_socket(sock, reply, remaining);
- return -EIO;
- } else {
- return result;
- }
- }
- kunmap(bvec->bv_page);
-
- sigprocmask(SIG_SETMASK, &oldset, NULL );
- }
- mutex_unlock(&cmd->lock);
- dnbd3_end_cmd(cmd, 0);
- return result;
-}
-
-
-
-/**
- * dnbd3_receive_cmd_get_block_test - receive a test block
- * @sock: the socket where the request is received
- * @reply: the reply initialized by dnbd3_receive_cmd
- *
- * this method should be called directly after the dnbd3_receive_cmd method
- *
- * the received data is just thrown away
- */
-static int dnbd3_receive_cmd_get_block_test(struct dnbd3_sock *sock,
- dnbd3_reply_t *reply)
-{
- struct msghdr msg;
- struct kvec iov;
- int result = 0;
- char *buf = kmalloc(reply->size, GFP_KERNEL);
- if (!buf) {
- error_sock(sock, "kmalloc failed");
- goto error;
- }
-
- dnbd3_init_msghdr(msg);
- iov.iov_base = buf;
- iov.iov_len = reply->size;
- result = kernel_recvmsg(sock->sock, &msg, &iov, 1, reply->size,
- msg.msg_flags);
- if (result != RTT_BLOCK_SIZE) {
- error_sock(sock, "receive test block failed");
- goto error;
- }
-
-error:
- if (buf) {
- kfree(buf);
- }
- return result;
-}
-
-/**
- * dnbd3_receive_cmd_get_servers - receive new servers
- * @sock: the socket where the request is received
- * @reply: the reply initialized by dnbd3_receive_cmd
- *
- * this method should be called directly after the dnbd3_receive_cmd method
- *
- * the new servers are copied to dnbd3_device.new_servers and
- * dnbd3_device.new_server_num is set accordingly
- */
-static int dnbd3_receive_cmd_get_servers(struct dnbd3_sock *sock,
- dnbd3_reply_t *reply)
-{
- struct msghdr msg;
- struct kvec iov;
- struct dnbd3_device *dev = sock->device;
- int result = 1;
- int count, remaining;
- dnbd3_init_msghdr(msg);
-
- debug_sock(sock, "get servers received");
- mutex_lock(&dev->device_lock);
- if (!dev->use_server_provided_alts) {
- remaining = reply->size;
- goto consume_payload;
- }
- dev->new_servers_num = 0;
- count = MIN(NUMBER_SERVERS, 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, iov.iov_len,
- msg.msg_flags);
- if (result <= 0) {
- error_sock(sock, "failed to receive get servers %d",
- result);
- mutex_unlock(&dev->device_lock);
- return result;
- } else if (result != (count * sizeof(dnbd3_server_entry_t))) {
- error_sock(sock, "failed to get servers");
- mutex_unlock(&dev->device_lock);
- return -EIO;
- }
- dev->new_servers_num = count;
- }
- /*
- * if there were more servers than accepted, remove the remaining data
- * from the socket buffer
- * abuse the reply struct as the receive buffer
- */
- remaining = reply->size - (count * sizeof(dnbd3_server_entry_t));
-consume_payload:
- while (remaining > 0) {
- count = MIN(sizeof(dnbd3_reply_t), remaining);
- iov.iov_base = reply;
- iov.iov_len = count;
- result = kernel_recvmsg(sock->sock, &msg, &iov, 1, count,
- msg.msg_flags);
- if (result <= 0) {
- error_sock(sock, "failed to receive payload from get servers");
- mutex_unlock(&dev->device_lock);
- return result;
- }
- remaining -= result;
- }
- mutex_unlock(&dev->device_lock);
- return result;
-}
-
-/**
- * dnbd3_receive_cmd_latest_rid - receive latest rid
- * @sock: the socket where the request is received
- * @reply: the reply initialized by dnbd3_receive_cmd
- *
- * this method should be called directly after the dnbd3_receive_cmd method
- *
- * dnbd3_device.update_available is set if a new RID is received
- */
-static int dnbd3_receive_cmd_latest_rid(struct dnbd3_sock *sock,
- dnbd3_reply_t *reply)
-{
- struct kvec iov;
- uint16_t rid;
- int result;
- struct msghdr msg;
- struct dnbd3_device *dev = sock->device;
- dnbd3_init_msghdr(msg);
- debug_sock(sock, "latest rid received");
-
- if (reply->size != 2) {
- error_sock(sock, "failed to get latest rid, wrong size");
- return -EIO;
- }
- 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) {
- error_sock(sock, "failed to receive latest rid");
- return result;
- }
- rid = net_order_16(rid);
- debug_sock(sock, "latest rid of %s is %d (currently using %d)",
- dev->imgname, (int)rid, (int)dev->rid);
- dev->update_available = (rid > dev->rid ? true : false);
- return result;
-}
-
-
-/**
- * dnbd3_receive_cmd_latest_rid - select the image
- * @sock: the socket where the request is received
- * @reply: the reply initialized by dnbd3_receive_cmd
- *
- * this method should be called directly after the dnbd3_receive_cmd method
- *
- * if this is the first connection the image name, file size and rid will be set
- * if this is a further connection image name, file size and rid will be checked
- */
-static int dnbd3_receive_cmd_select_image(struct dnbd3_sock *sock,
- dnbd3_reply_t *reply)
-{
- struct kvec iov;
- uint16_t rid;
- char *name;
- int result;
- struct msghdr msg;
- serialized_buffer_t payload_buffer;
- uint64_t reported_size;
- struct dnbd3_device *dev = sock->device;
- dnbd3_init_msghdr(msg);
- debug_sock(sock, "select image received");
- iov.iov_base = &payload_buffer;
- iov.iov_len = reply->size;
- result = kernel_recvmsg(sock->sock, &msg, &iov, 1, iov.iov_len,
- msg.msg_flags);
- if (result <= 0) {
- error_sock(sock, "failed to receive select image %d", result);
- return result;
- } else if (result != reply->size) {
- error_sock(sock, "could not read CMD_SELECT_IMAGE payload on handshake, size is %d and should be %d",
- result, reply->size);
- return -EIO;
- }
-
- /* handle/check reply payload */
- serializer_reset_read(&payload_buffer, reply->size);
- sock->server->protocol_version = serializer_get_uint16(&payload_buffer);
- if (sock->server->protocol_version < MIN_SUPPORTED_SERVER) {
- error_sock(sock, "server version is lower than min supported version");
- return -EIO;
- }
-
- //TODO compare RID
-
- name = serializer_get_string(&payload_buffer);
- rid = serializer_get_uint16(&payload_buffer);
- if (dev->rid != rid && strcmp(name, dev->imgname) != 0) {
- error_sock(sock, "server offers image '%s', requested '%s'",
- name, dev->imgname);
- return -EIO;
- }
-
- reported_size = serializer_get_uint64(&payload_buffer);
- if (!dev->reported_size) {
- if (reported_size < 4096) {
- error_sock(sock, "reported size by server is < 4096");
- return -EIO;
- }
- dev->reported_size = reported_size;
- set_capacity(dev->disk, dev->reported_size >> 9); /* 512 Byte */
- } else if (dev->reported_size != reported_size) {
- error_sock(sock, "reported size by server is %llu but should be %llu",
- reported_size, dev->reported_size);
- return -EIO;
- }
- return result;
-
-}
-
-
/*
* Timer and workers
*/
@@ -838,7 +251,7 @@ static int dnbd3_panic_connect(struct dnbd3_device *dev)
}
}
if (working == NULL) {
- return -EIO;
+ return -ENOENT;
}
for (i = 0; i < NUMBER_CONNECTIONS; i++) {
if (dev->socks[i].server != working) {
@@ -925,7 +338,8 @@ static void dnbd3_panic_worker(struct work_struct *work)
int sock_alive = 0;
dev = container_of(work, struct dnbd3_device, panic_worker);
for (i = 0; i < NUMBER_CONNECTIONS; i++) {
- if (dev->socks[i].panic) {
+ if (dev->socks[i].panic
+ || dev->socks[i].server->failures > 1000) {
panic = true;
dnbd3_set_rtt_unreachable(dev->socks[i].server);
dnbd3_socket_disconnect(&dev->socks[i]);
@@ -962,7 +376,6 @@ static int dnbd3_meassure_rtt(struct dnbd3_device *dev,
{
struct timeval start, end;
dnbd3_reply_t reply;
- struct request req;
int result;
uint64_t rtt = RTT_UNREACHABLE;
struct dnbd3_sock sock = {
@@ -977,7 +390,6 @@ static int dnbd3_meassure_rtt(struct dnbd3_device *dev,
error_sock(&sock, "socket connect failed in rtt measurement");
goto error;
}
- dnbd3_connect_to_req(&req);
result = dnbd3_send_request_cmd(&sock, CMD_SELECT_IMAGE);
if (result <= 0) {
error_sock(&sock, "request select image failed in rtt measurement");
@@ -1138,8 +550,10 @@ static void dnbd3_discovery_worker(struct work_struct *work)
for (i = 0; i < NUMBER_SERVERS; i++) {
server = &dev->alt_servers[i];
if (server->host.type) {
+ /* failure count is divided*/
+ server->failures = server->failures / 2;
if (dnbd3_meassure_rtt(dev, server) <= 0) {
- server->failures++;
+ server->failures = server->failures * 2;
warn_server(dev, server,
"failed to meassure rtt");
}
@@ -1316,8 +730,6 @@ static int dnbd3_socket_connect(struct dnbd3_sock *sock,
if (dnbd3_send_request_cmd(sock, CMD_GET_SERVERS) <= 0) {
error_sock(sock, "failed to get servers in discovery");
}
- /* failure count is divided on each successful connect */
- server->failures = server->failures / 2;
return 0;
error: