diff options
| -rw-r--r-- | src/kernel/block.c | 14 | ||||
| -rw-r--r-- | src/kernel/block.h | 7 | ||||
| -rw-r--r-- | src/kernel/core.c | 4 | ||||
| -rw-r--r-- | src/kernel/dnbd3.h | 24 | ||||
| -rw-r--r-- | src/kernel/mq.c | 186 |
5 files changed, 212 insertions, 23 deletions
diff --git a/src/kernel/block.c b/src/kernel/block.c index 4eb52c5..26caa25 100644 --- a/src/kernel/block.c +++ b/src/kernel/block.c @@ -13,13 +13,7 @@ #define dnbd3_sock_create(af,type,proto,sock) sock_create_kern(&init_net, (af) == HOST_IP4 ? AF_INET : AF_INET6, type, proto, sock) -#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 int dnbd3_open(struct block_device *bdev, fmode_t mode) { @@ -222,6 +216,8 @@ int dnbd3_net_connect(dnbd3_device_t *dev) dev->panic = 0; dev->panic_count = 0; + refcount_inc(&dev->config_refs); + // Enqueue request to request_queue_send for a fresh list of alt servers //TODO refresh alt server list @@ -337,7 +333,7 @@ static int dnbd3_ioctl(struct block_device *bdev, fmode_t mode, unsigned int cmd dev->imgname = imgname; dev->rid = msg->rid; dev->use_server_provided_alts = msg->use_server_provided_alts; - // Forget all alt servers on explicit connect, set first al server to initial server + // Forget all alt servers on explicit connect, set first alt server to initial server memset(dev->alt_servers, 0, sizeof(dev->alt_servers[0])*NUMBER_SERVERS); memcpy(dev->alt_servers, &dev->initial_server, sizeof(dev->alt_servers[0])); //#if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 11, 0) @@ -397,7 +393,7 @@ static int dnbd3_ioctl(struct block_device *bdev, fmode_t mode, unsigned int cmd break; default: - printk(KERN_DEBUG "dnbd3: ioctl unhandled cmd\n"); + printk(KERN_DEBUG "dnbd3: ioctl unhandled cmd %d\n", cmd); result = -EIO; break; } diff --git a/src/kernel/block.h b/src/kernel/block.h index af99108..07bf950 100644 --- a/src/kernel/block.h +++ b/src/kernel/block.h @@ -8,6 +8,13 @@ #ifndef SRC_KERNEL_BLOCK_H_ #define SRC_KERNEL_BLOCK_H_ +#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) int dnbd3_net_connect(dnbd3_device_t *dev); diff --git a/src/kernel/core.c b/src/kernel/core.c index 3b30a11..731c094 100644 --- a/src/kernel/core.c +++ b/src/kernel/core.c @@ -53,8 +53,8 @@ #include "block.h" #include "mq.h" -static DEFINE_IDR(dnbd3_index_idr); -static DEFINE_MUTEX(dnbd3_index_mutex); +DEFINE_IDR(dnbd3_index_idr); +DEFINE_MUTEX(dnbd3_index_mutex); static unsigned int max_devs = NUMBER_DEVICES; static dnbd3_device_t *dnbd3_device; diff --git a/src/kernel/dnbd3.h b/src/kernel/dnbd3.h index 3fb625a..af9bdff 100644 --- a/src/kernel/dnbd3.h +++ b/src/kernel/dnbd3.h @@ -30,6 +30,8 @@ #include "types.h" #include "serialize.h" + + typedef struct { dnbd3_host_t host; @@ -57,6 +59,8 @@ typedef struct dnbd3_device_t { // network char *imgname; struct socket *sock; + struct mutex socket_lock; + struct request *pending; dnbd3_server_t cur_server, initial_server; uint64_t cur_rtt; serialized_buffer_t payload_buffer; @@ -95,15 +99,15 @@ typedef struct dnbd3_cmd { unsigned long flags; uint32_t cmd_cookie; } dnbd3_cmd; - -typedef struct dnbd3_sock { - struct socket *sock; - struct mutex tx_lock; - struct request *pending; - int sent; - bool dead; - int fallback_index; - int cookie; -} dnbd3_sock; +// +//typedef struct dnbd3_sock { +// struct socket *sock; +// struct mutex tx_lock; +// struct request *pending; +// int sent; +// bool dead; +// int fallback_index; +// int cookie; +//} dnbd3_sock; #endif /* DNBD_H_ */ diff --git a/src/kernel/mq.c b/src/kernel/mq.c index ee60574..e3407ab 100644 --- a/src/kernel/mq.c +++ b/src/kernel/mq.c @@ -5,24 +5,206 @@ * Author: fred */ +#include <net/sock.h> #include "mq.h" +#include "block.h" #define DNBD3_CMD_REQUEUED 1 +//static void dnbd3_config_put(struct dnbd3_device_t *dev) +//{ +// if (refcount_dec_and_mutex_lock(&dev->config_refs, +// &dev->config_lock)) { +// +// dev->tag_set.timeout = 0; +// dev->disk->queue->limits.discard_granularity = 0; +// dev->disk->queue->limits.discard_alignment = 0; +// blk_queue_max_discard_sectors(dev->disk->queue, UINT_MAX); +// blk_queue_flag_clear(QUEUE_FLAG_DISCARD, dev->disk->queue); +// +// mutex_unlock(&dev->config_lock); +// nbd_put(dev); +// module_put(THIS_MODULE); +// } +//} + +#define dnbd3_priv_to_cmd(req) ((req)->cmd_flags >> REQ_FLAG_BITS) + +static int dnbd3_send_cmd(struct dnbd3_device_t *dev, struct dnbd3_cmd *cmd, int index) +{ + struct request *req = blk_mq_rq_from_pdu(cmd); + dnbd3_request_t dnbd3_request; + dnbd3_reply_t dnbd3_reply; + 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; + int result; + + init_msghdr(msg); + + dnbd3_request.magic = dnbd3_packet_magic; + + switch (req_op(req)) { + case REQ_OP_DISCARD: + printk(KERN_DEBUG "dnbd3: request operation discard on device %d\n", dev->minor); + break; + case REQ_OP_FLUSH: + printk(KERN_DEBUG "dnbd3: request operation flush on device %d\n", dev->minor); + break; + case REQ_OP_WRITE: + printk(KERN_DEBUG "dnbd3: request operation write on device %d\n", dev->minor); + break; + case REQ_OP_READ: + printk(KERN_DEBUG "dnbd3: request operation read on device %d\n", dev->minor); + 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 REQ_OP_DRV_IN: + printk(KERN_DEBUG "dnbd3: request operation driver in on device %d\n", dev->minor); + dnbd3_request.cmd = dnbd3_priv_to_cmd(req); + dnbd3_request.size = 0; + break; + + default: + return -EIO; + } + + dnbd3_request.handle = (uint64_t)(uintptr_t)req; // Double cast to prevent warning on 32bit + fixup_request(dnbd3_request); + iov.iov_base = &dnbd3_request; + iov.iov_len = sizeof(dnbd3_request); + if (kernel_sendmsg(dev->sock, &msg, &iov, 1, sizeof(dnbd3_request)) != sizeof(dnbd3_request)) { + printk(KERN_ERR "dnbd3: connection to server lost\n"); + result = -EAGAIN; + goto error; + } + + // receive net reply + iov.iov_base = &dnbd3_reply; + iov.iov_len = sizeof(dnbd3_reply); + result = kernel_recvmsg(dev->sock, &msg, &iov, 1, sizeof(dnbd3_reply), msg.msg_flags); + if (!result) { + printk(KERN_ERR "dnbd3: connection to server lost\n"); + result = -EIO; + 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; +// } + + if (dnbd3_reply.cmd != CMD_GET_BLOCK) { + printk(KERN_ERR "dnbd3: command was %d\n", dnbd3_reply.cmd); + result = -EIO; + goto error; + } + + + + 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; + if (kernel_recvmsg(dev->sock, &msg, &iov, 1, bvec->bv_len, msg.msg_flags) != 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"); + goto error; + } + kunmap(bvec->bv_page); + + sigprocmask(SIG_SETMASK, &oldset, NULL ); + } +// +// __blk_end_request_all(req, 0); + blk_mq_end_request(req, 0); + dev->pending = NULL; +error: + return result; +} + +static void dnbd3_requeue_cmd(struct dnbd3_cmd *cmd) +{ + struct request *req = blk_mq_rq_from_pdu(cmd); + + if (!test_and_set_bit(DNBD3_CMD_REQUEUED, &cmd->flags)) + blk_mq_requeue_request(req, true); +} static int dnbd3_handle_cmd(struct dnbd3_cmd *cmd, int index) { struct request *req = blk_mq_rq_from_pdu(cmd); struct dnbd3_device_t *dev = cmd->dnbd3; - struct dnbd3_sock *nsock; int ret = -1; - printk(KERN_DEBUG "dnbd3: handle command %i device %i\n", cmd->cmd_cookie, dev->minor); + printk(KERN_DEBUG "dnbd3: handle request at position %lu and size %d, device %i\n", blk_rq_pos(req), blk_rq_bytes(req), dev->minor); + + if (!refcount_inc_not_zero(&dev->config_refs)) { + dev_err_ratelimited(disk_to_dev(dev->disk), "socks array is empty\n"); + blk_mq_start_request(req); + return -EINVAL; + } + + if (index >= 1) { // TODO use next server with good rtt for this request + printk(KERN_INFO "dnbd3: index is %d", index); + dev_err_ratelimited(disk_to_dev(dev->disk), "attempted send on invalid socket\n"); +// nbd_config_put(nbd); +// TODO what to do here? + blk_mq_start_request(req); + return -EINVAL; + } + cmd->status = BLK_STS_OK; + +again: + mutex_lock(&dev->socket_lock); + if (!dev->sock) { + mutex_unlock(&dev->socket_lock); + printk(KERN_DEBUG "dnbd3: not connected, try to reconnect\n"); + if (!dnbd3_net_connect(dev)) { + printk(KERN_ERR "dnbd3: failed to reconnect\n"); + blk_mq_start_request(req); + return -EIO; + } + goto again; + } + blk_mq_start_request(req); + if (unlikely(dev->pending && dev->pending != req)) { + dnbd3_requeue_cmd(cmd); + ret = 0; + goto out; + } + ret = dnbd3_send_cmd(dev, cmd, index); + if (ret == -EAGAIN) { + dev_err_ratelimited(disk_to_dev(dev->disk), "request send failed, requeueing\n"); + dnbd3_requeue_cmd(cmd); + ret = 0; + } +out: + mutex_unlock(&dev->socket_lock); +// nbd_config_put(nbd); return ret; } |