12025cf9eSThomas Gleixner // SPDX-License-Identifier: GPL-2.0-only 2ae0078fcSDexuan Cui /* 3ae0078fcSDexuan Cui * Hyper-V transport for vsock 4ae0078fcSDexuan Cui * 5ae0078fcSDexuan Cui * Hyper-V Sockets supplies a byte-stream based communication mechanism 6ae0078fcSDexuan Cui * between the host and the VM. This driver implements the necessary 7ae0078fcSDexuan Cui * support in the VM by introducing the new vsock transport. 8ae0078fcSDexuan Cui * 9ae0078fcSDexuan Cui * Copyright (c) 2017, Microsoft Corporation. 10ae0078fcSDexuan Cui */ 11ae0078fcSDexuan Cui #include <linux/module.h> 12ae0078fcSDexuan Cui #include <linux/vmalloc.h> 13ae0078fcSDexuan Cui #include <linux/hyperv.h> 14ae0078fcSDexuan Cui #include <net/sock.h> 15ae0078fcSDexuan Cui #include <net/af_vsock.h> 1677ffe333SHimadri Pandya #include <asm/hyperv-tlfs.h> 17ae0078fcSDexuan Cui 18ac383f58SSunil Muthuswamy /* Older (VMBUS version 'VERSION_WIN10' or before) Windows hosts have some 1977ffe333SHimadri Pandya * stricter requirements on the hv_sock ring buffer size of six 4K pages. 2077ffe333SHimadri Pandya * hyperv-tlfs defines HV_HYP_PAGE_SIZE as 4K. Newer hosts don't have this 2177ffe333SHimadri Pandya * limitation; but, keep the defaults the same for compat. 22ae0078fcSDexuan Cui */ 2377ffe333SHimadri Pandya #define RINGBUFFER_HVS_RCV_SIZE (HV_HYP_PAGE_SIZE * 6) 2477ffe333SHimadri Pandya #define RINGBUFFER_HVS_SND_SIZE (HV_HYP_PAGE_SIZE * 6) 2577ffe333SHimadri Pandya #define RINGBUFFER_HVS_MAX_SIZE (HV_HYP_PAGE_SIZE * 64) 26ae0078fcSDexuan Cui 27ae0078fcSDexuan Cui /* The MTU is 16KB per the host side's design */ 28ae0078fcSDexuan Cui #define HVS_MTU_SIZE (1024 * 16) 29ae0078fcSDexuan Cui 30a9eeb998SSunil Muthuswamy /* How long to wait for graceful shutdown of a connection */ 31a9eeb998SSunil Muthuswamy #define HVS_CLOSE_TIMEOUT (8 * HZ) 32a9eeb998SSunil Muthuswamy 33ae0078fcSDexuan Cui struct vmpipe_proto_header { 34ae0078fcSDexuan Cui u32 pkt_type; 35ae0078fcSDexuan Cui u32 data_size; 36ae0078fcSDexuan Cui }; 37ae0078fcSDexuan Cui 38ae0078fcSDexuan Cui /* For recv, we use the VMBus in-place packet iterator APIs to directly copy 39ae0078fcSDexuan Cui * data from the ringbuffer into the userspace buffer. 40ae0078fcSDexuan Cui */ 41ae0078fcSDexuan Cui struct hvs_recv_buf { 42ae0078fcSDexuan Cui /* The header before the payload data */ 43ae0078fcSDexuan Cui struct vmpipe_proto_header hdr; 44ae0078fcSDexuan Cui 45ae0078fcSDexuan Cui /* The payload */ 46ae0078fcSDexuan Cui u8 data[HVS_MTU_SIZE]; 47ae0078fcSDexuan Cui }; 48ae0078fcSDexuan Cui 49ae0078fcSDexuan Cui /* We can send up to HVS_MTU_SIZE bytes of payload to the host, but let's use 5014a1eaa8SSunil Muthuswamy * a smaller size, i.e. HVS_SEND_BUF_SIZE, to maximize concurrency between the 5114a1eaa8SSunil Muthuswamy * guest and the host processing as one VMBUS packet is the smallest processing 5214a1eaa8SSunil Muthuswamy * unit. 53ae0078fcSDexuan Cui * 54ae0078fcSDexuan Cui * Note: the buffer can be eliminated in the future when we add new VMBus 55ae0078fcSDexuan Cui * ringbuffer APIs that allow us to directly copy data from userspace buffer 56ae0078fcSDexuan Cui * to VMBus ringbuffer. 57ae0078fcSDexuan Cui */ 5877ffe333SHimadri Pandya #define HVS_SEND_BUF_SIZE \ 5977ffe333SHimadri Pandya (HV_HYP_PAGE_SIZE - sizeof(struct vmpipe_proto_header)) 60ae0078fcSDexuan Cui 61ae0078fcSDexuan Cui struct hvs_send_buf { 62ae0078fcSDexuan Cui /* The header before the payload data */ 63ae0078fcSDexuan Cui struct vmpipe_proto_header hdr; 64ae0078fcSDexuan Cui 65ae0078fcSDexuan Cui /* The payload */ 66ae0078fcSDexuan Cui u8 data[HVS_SEND_BUF_SIZE]; 67ae0078fcSDexuan Cui }; 68ae0078fcSDexuan Cui 69ae0078fcSDexuan Cui #define HVS_HEADER_LEN (sizeof(struct vmpacket_descriptor) + \ 70ae0078fcSDexuan Cui sizeof(struct vmpipe_proto_header)) 71ae0078fcSDexuan Cui 72ae0078fcSDexuan Cui /* See 'prev_indices' in hv_ringbuffer_read(), hv_ringbuffer_write(), and 73ae0078fcSDexuan Cui * __hv_pkt_iter_next(). 74ae0078fcSDexuan Cui */ 75ae0078fcSDexuan Cui #define VMBUS_PKT_TRAILER_SIZE (sizeof(u64)) 76ae0078fcSDexuan Cui 77ae0078fcSDexuan Cui #define HVS_PKT_LEN(payload_len) (HVS_HEADER_LEN + \ 78ae0078fcSDexuan Cui ALIGN((payload_len), 8) + \ 79ae0078fcSDexuan Cui VMBUS_PKT_TRAILER_SIZE) 80ae0078fcSDexuan Cui 81ae0078fcSDexuan Cui union hvs_service_id { 82ce103204SAndy Shevchenko guid_t srv_id; 83ae0078fcSDexuan Cui 84ae0078fcSDexuan Cui struct { 85ae0078fcSDexuan Cui unsigned int svm_port; 86ce103204SAndy Shevchenko unsigned char b[sizeof(guid_t) - sizeof(unsigned int)]; 87ae0078fcSDexuan Cui }; 88ae0078fcSDexuan Cui }; 89ae0078fcSDexuan Cui 90ae0078fcSDexuan Cui /* Per-socket state (accessed via vsk->trans) */ 91ae0078fcSDexuan Cui struct hvsock { 92ae0078fcSDexuan Cui struct vsock_sock *vsk; 93ae0078fcSDexuan Cui 94ce103204SAndy Shevchenko guid_t vm_srv_id; 95ce103204SAndy Shevchenko guid_t host_srv_id; 96ae0078fcSDexuan Cui 97ae0078fcSDexuan Cui struct vmbus_channel *chan; 98ae0078fcSDexuan Cui struct vmpacket_descriptor *recv_desc; 99ae0078fcSDexuan Cui 100ae0078fcSDexuan Cui /* The length of the payload not delivered to userland yet */ 101ae0078fcSDexuan Cui u32 recv_data_len; 102ae0078fcSDexuan Cui /* The offset of the payload */ 103ae0078fcSDexuan Cui u32 recv_data_off; 104ae0078fcSDexuan Cui 105ae0078fcSDexuan Cui /* Have we sent the zero-length packet (FIN)? */ 106ae0078fcSDexuan Cui bool fin_sent; 107ae0078fcSDexuan Cui }; 108ae0078fcSDexuan Cui 109ae0078fcSDexuan Cui /* In the VM, we support Hyper-V Sockets with AF_VSOCK, and the endpoint is 110ae0078fcSDexuan Cui * <cid, port> (see struct sockaddr_vm). Note: cid is not really used here: 111ae0078fcSDexuan Cui * when we write apps to connect to the host, we can only use VMADDR_CID_ANY 112ae0078fcSDexuan Cui * or VMADDR_CID_HOST (both are equivalent) as the remote cid, and when we 113ae0078fcSDexuan Cui * write apps to bind() & listen() in the VM, we can only use VMADDR_CID_ANY 114ae0078fcSDexuan Cui * as the local cid. 115ae0078fcSDexuan Cui * 116ae0078fcSDexuan Cui * On the host, Hyper-V Sockets are supported by Winsock AF_HYPERV: 117ae0078fcSDexuan Cui * https://docs.microsoft.com/en-us/virtualization/hyper-v-on-windows/user- 118ae0078fcSDexuan Cui * guide/make-integration-service, and the endpoint is <VmID, ServiceId> with 119ae0078fcSDexuan Cui * the below sockaddr: 120ae0078fcSDexuan Cui * 121ae0078fcSDexuan Cui * struct SOCKADDR_HV 122ae0078fcSDexuan Cui * { 123ae0078fcSDexuan Cui * ADDRESS_FAMILY Family; 124ae0078fcSDexuan Cui * USHORT Reserved; 125ae0078fcSDexuan Cui * GUID VmId; 126ae0078fcSDexuan Cui * GUID ServiceId; 127ae0078fcSDexuan Cui * }; 128ae0078fcSDexuan Cui * Note: VmID is not used by Linux VM and actually it isn't transmitted via 129ae0078fcSDexuan Cui * VMBus, because here it's obvious the host and the VM can easily identify 130ae0078fcSDexuan Cui * each other. Though the VmID is useful on the host, especially in the case 131ae0078fcSDexuan Cui * of Windows container, Linux VM doesn't need it at all. 132ae0078fcSDexuan Cui * 133ae0078fcSDexuan Cui * To make use of the AF_VSOCK infrastructure in Linux VM, we have to limit 134ae0078fcSDexuan Cui * the available GUID space of SOCKADDR_HV so that we can create a mapping 135ae0078fcSDexuan Cui * between AF_VSOCK port and SOCKADDR_HV Service GUID. The rule of writing 136ae0078fcSDexuan Cui * Hyper-V Sockets apps on the host and in Linux VM is: 137ae0078fcSDexuan Cui * 138ae0078fcSDexuan Cui **************************************************************************** 139ae0078fcSDexuan Cui * The only valid Service GUIDs, from the perspectives of both the host and * 140ae0078fcSDexuan Cui * Linux VM, that can be connected by the other end, must conform to this * 141ae0078fcSDexuan Cui * format: <port>-facb-11e6-bd58-64006a7986d3, and the "port" must be in * 142ae0078fcSDexuan Cui * this range [0, 0x7FFFFFFF]. * 143ae0078fcSDexuan Cui **************************************************************************** 144ae0078fcSDexuan Cui * 145ae0078fcSDexuan Cui * When we write apps on the host to connect(), the GUID ServiceID is used. 146ae0078fcSDexuan Cui * When we write apps in Linux VM to connect(), we only need to specify the 147ae0078fcSDexuan Cui * port and the driver will form the GUID and use that to request the host. 148ae0078fcSDexuan Cui * 149ae0078fcSDexuan Cui * From the perspective of Linux VM: 150ae0078fcSDexuan Cui * 1. the local ephemeral port (i.e. the local auto-bound port when we call 151ae0078fcSDexuan Cui * connect() without explicit bind()) is generated by __vsock_bind_stream(), 152ae0078fcSDexuan Cui * and the range is [1024, 0xFFFFFFFF). 153ae0078fcSDexuan Cui * 2. the remote ephemeral port (i.e. the auto-generated remote port for 154ae0078fcSDexuan Cui * a connect request initiated by the host's connect()) is generated by 155ae0078fcSDexuan Cui * hvs_remote_addr_init() and the range is [0x80000000, 0xFFFFFFFF). 156ae0078fcSDexuan Cui */ 157ae0078fcSDexuan Cui 158ae0078fcSDexuan Cui #define MAX_LISTEN_PORT ((u32)0x7FFFFFFF) 159ae0078fcSDexuan Cui #define MAX_VM_LISTEN_PORT MAX_LISTEN_PORT 160ae0078fcSDexuan Cui #define MAX_HOST_LISTEN_PORT MAX_LISTEN_PORT 161ae0078fcSDexuan Cui #define MIN_HOST_EPHEMERAL_PORT (MAX_HOST_LISTEN_PORT + 1) 162ae0078fcSDexuan Cui 163ae0078fcSDexuan Cui /* 00000000-facb-11e6-bd58-64006a7986d3 */ 164ce103204SAndy Shevchenko static const guid_t srv_id_template = 165ce103204SAndy Shevchenko GUID_INIT(0x00000000, 0xfacb, 0x11e6, 0xbd, 0x58, 166ae0078fcSDexuan Cui 0x64, 0x00, 0x6a, 0x79, 0x86, 0xd3); 167ae0078fcSDexuan Cui 168c0cfa2d8SStefano Garzarella static bool hvs_check_transport(struct vsock_sock *vsk); 169c0cfa2d8SStefano Garzarella 170ce103204SAndy Shevchenko static bool is_valid_srv_id(const guid_t *id) 171ae0078fcSDexuan Cui { 172ce103204SAndy Shevchenko return !memcmp(&id->b[4], &srv_id_template.b[4], sizeof(guid_t) - 4); 173ae0078fcSDexuan Cui } 174ae0078fcSDexuan Cui 175ce103204SAndy Shevchenko static unsigned int get_port_by_srv_id(const guid_t *svr_id) 176ae0078fcSDexuan Cui { 177ae0078fcSDexuan Cui return *((unsigned int *)svr_id); 178ae0078fcSDexuan Cui } 179ae0078fcSDexuan Cui 180ce103204SAndy Shevchenko static void hvs_addr_init(struct sockaddr_vm *addr, const guid_t *svr_id) 181ae0078fcSDexuan Cui { 182ae0078fcSDexuan Cui unsigned int port = get_port_by_srv_id(svr_id); 183ae0078fcSDexuan Cui 184ae0078fcSDexuan Cui vsock_addr_init(addr, VMADDR_CID_ANY, port); 185ae0078fcSDexuan Cui } 186ae0078fcSDexuan Cui 187ae0078fcSDexuan Cui static void hvs_remote_addr_init(struct sockaddr_vm *remote, 188ae0078fcSDexuan Cui struct sockaddr_vm *local) 189ae0078fcSDexuan Cui { 190ae0078fcSDexuan Cui static u32 host_ephemeral_port = MIN_HOST_EPHEMERAL_PORT; 191ae0078fcSDexuan Cui struct sock *sk; 192ae0078fcSDexuan Cui 19303964257SStefano Garzarella /* Remote peer is always the host */ 19403964257SStefano Garzarella vsock_addr_init(remote, VMADDR_CID_HOST, VMADDR_PORT_ANY); 195ae0078fcSDexuan Cui 196ae0078fcSDexuan Cui while (1) { 197ae0078fcSDexuan Cui /* Wrap around ? */ 198ae0078fcSDexuan Cui if (host_ephemeral_port < MIN_HOST_EPHEMERAL_PORT || 199ae0078fcSDexuan Cui host_ephemeral_port == VMADDR_PORT_ANY) 200ae0078fcSDexuan Cui host_ephemeral_port = MIN_HOST_EPHEMERAL_PORT; 201ae0078fcSDexuan Cui 202ae0078fcSDexuan Cui remote->svm_port = host_ephemeral_port++; 203ae0078fcSDexuan Cui 204ae0078fcSDexuan Cui sk = vsock_find_connected_socket(remote, local); 205ae0078fcSDexuan Cui if (!sk) { 206ae0078fcSDexuan Cui /* Found an available ephemeral port */ 207ae0078fcSDexuan Cui return; 208ae0078fcSDexuan Cui } 209ae0078fcSDexuan Cui 210ae0078fcSDexuan Cui /* Release refcnt got in vsock_find_connected_socket */ 211ae0078fcSDexuan Cui sock_put(sk); 212ae0078fcSDexuan Cui } 213ae0078fcSDexuan Cui } 214ae0078fcSDexuan Cui 215ae0078fcSDexuan Cui static void hvs_set_channel_pending_send_size(struct vmbus_channel *chan) 216ae0078fcSDexuan Cui { 217ae0078fcSDexuan Cui set_channel_pending_send_size(chan, 218ae0078fcSDexuan Cui HVS_PKT_LEN(HVS_SEND_BUF_SIZE)); 219ae0078fcSDexuan Cui 220ae0078fcSDexuan Cui virt_mb(); 221ae0078fcSDexuan Cui } 222ae0078fcSDexuan Cui 223ae0078fcSDexuan Cui static bool hvs_channel_readable(struct vmbus_channel *chan) 224ae0078fcSDexuan Cui { 225ae0078fcSDexuan Cui u32 readable = hv_get_bytes_to_read(&chan->inbound); 226ae0078fcSDexuan Cui 227ae0078fcSDexuan Cui /* 0-size payload means FIN */ 228ae0078fcSDexuan Cui return readable >= HVS_PKT_LEN(0); 229ae0078fcSDexuan Cui } 230ae0078fcSDexuan Cui 231ae0078fcSDexuan Cui static int hvs_channel_readable_payload(struct vmbus_channel *chan) 232ae0078fcSDexuan Cui { 233ae0078fcSDexuan Cui u32 readable = hv_get_bytes_to_read(&chan->inbound); 234ae0078fcSDexuan Cui 235ae0078fcSDexuan Cui if (readable > HVS_PKT_LEN(0)) { 236ae0078fcSDexuan Cui /* At least we have 1 byte to read. We don't need to return 237ae0078fcSDexuan Cui * the exact readable bytes: see vsock_stream_recvmsg() -> 238ae0078fcSDexuan Cui * vsock_stream_has_data(). 239ae0078fcSDexuan Cui */ 240ae0078fcSDexuan Cui return 1; 241ae0078fcSDexuan Cui } 242ae0078fcSDexuan Cui 243ae0078fcSDexuan Cui if (readable == HVS_PKT_LEN(0)) { 244ae0078fcSDexuan Cui /* 0-size payload means FIN */ 245ae0078fcSDexuan Cui return 0; 246ae0078fcSDexuan Cui } 247ae0078fcSDexuan Cui 248ae0078fcSDexuan Cui /* No payload or FIN */ 249ae0078fcSDexuan Cui return -1; 250ae0078fcSDexuan Cui } 251ae0078fcSDexuan Cui 252ae0078fcSDexuan Cui static size_t hvs_channel_writable_bytes(struct vmbus_channel *chan) 253ae0078fcSDexuan Cui { 254ae0078fcSDexuan Cui u32 writeable = hv_get_bytes_to_write(&chan->outbound); 255ae0078fcSDexuan Cui size_t ret; 256ae0078fcSDexuan Cui 257ae0078fcSDexuan Cui /* The ringbuffer mustn't be 100% full, and we should reserve a 258ae0078fcSDexuan Cui * zero-length-payload packet for the FIN: see hv_ringbuffer_write() 259ae0078fcSDexuan Cui * and hvs_shutdown(). 260ae0078fcSDexuan Cui */ 261ae0078fcSDexuan Cui if (writeable <= HVS_PKT_LEN(1) + HVS_PKT_LEN(0)) 262ae0078fcSDexuan Cui return 0; 263ae0078fcSDexuan Cui 264ae0078fcSDexuan Cui ret = writeable - HVS_PKT_LEN(1) - HVS_PKT_LEN(0); 265ae0078fcSDexuan Cui 266ae0078fcSDexuan Cui return round_down(ret, 8); 267ae0078fcSDexuan Cui } 268ae0078fcSDexuan Cui 269ae0078fcSDexuan Cui static int hvs_send_data(struct vmbus_channel *chan, 270ae0078fcSDexuan Cui struct hvs_send_buf *send_buf, size_t to_write) 271ae0078fcSDexuan Cui { 272ae0078fcSDexuan Cui send_buf->hdr.pkt_type = 1; 273ae0078fcSDexuan Cui send_buf->hdr.data_size = to_write; 274ae0078fcSDexuan Cui return vmbus_sendpacket(chan, &send_buf->hdr, 275ae0078fcSDexuan Cui sizeof(send_buf->hdr) + to_write, 276ae0078fcSDexuan Cui 0, VM_PKT_DATA_INBAND, 0); 277ae0078fcSDexuan Cui } 278ae0078fcSDexuan Cui 279ae0078fcSDexuan Cui static void hvs_channel_cb(void *ctx) 280ae0078fcSDexuan Cui { 281ae0078fcSDexuan Cui struct sock *sk = (struct sock *)ctx; 282ae0078fcSDexuan Cui struct vsock_sock *vsk = vsock_sk(sk); 283ae0078fcSDexuan Cui struct hvsock *hvs = vsk->trans; 284ae0078fcSDexuan Cui struct vmbus_channel *chan = hvs->chan; 285ae0078fcSDexuan Cui 286ae0078fcSDexuan Cui if (hvs_channel_readable(chan)) 287ae0078fcSDexuan Cui sk->sk_data_ready(sk); 288ae0078fcSDexuan Cui 289ae0078fcSDexuan Cui if (hv_get_bytes_to_write(&chan->outbound) > 0) 290ae0078fcSDexuan Cui sk->sk_write_space(sk); 291ae0078fcSDexuan Cui } 292ae0078fcSDexuan Cui 293a9eeb998SSunil Muthuswamy static void hvs_do_close_lock_held(struct vsock_sock *vsk, 294a9eeb998SSunil Muthuswamy bool cancel_timeout) 295a9eeb998SSunil Muthuswamy { 296a9eeb998SSunil Muthuswamy struct sock *sk = sk_vsock(vsk); 297a9eeb998SSunil Muthuswamy 298a9eeb998SSunil Muthuswamy sock_set_flag(sk, SOCK_DONE); 299a9eeb998SSunil Muthuswamy vsk->peer_shutdown = SHUTDOWN_MASK; 300a9eeb998SSunil Muthuswamy if (vsock_stream_has_data(vsk) <= 0) 301a9eeb998SSunil Muthuswamy sk->sk_state = TCP_CLOSING; 302a9eeb998SSunil Muthuswamy sk->sk_state_change(sk); 303a9eeb998SSunil Muthuswamy if (vsk->close_work_scheduled && 304a9eeb998SSunil Muthuswamy (!cancel_timeout || cancel_delayed_work(&vsk->close_work))) { 305a9eeb998SSunil Muthuswamy vsk->close_work_scheduled = false; 306a9eeb998SSunil Muthuswamy vsock_remove_sock(vsk); 307a9eeb998SSunil Muthuswamy 308a9eeb998SSunil Muthuswamy /* Release the reference taken while scheduling the timeout */ 309a9eeb998SSunil Muthuswamy sock_put(sk); 310a9eeb998SSunil Muthuswamy } 311a9eeb998SSunil Muthuswamy } 312a9eeb998SSunil Muthuswamy 313ae0078fcSDexuan Cui static void hvs_close_connection(struct vmbus_channel *chan) 314ae0078fcSDexuan Cui { 315ae0078fcSDexuan Cui struct sock *sk = get_per_channel_state(chan); 316ae0078fcSDexuan Cui 317b4562ca7SDexuan Cui lock_sock(sk); 318a9eeb998SSunil Muthuswamy hvs_do_close_lock_held(vsock_sk(sk), true); 319b4562ca7SDexuan Cui release_sock(sk); 320685703b4SDexuan Cui 321685703b4SDexuan Cui /* Release the refcnt for the channel that's opened in 322685703b4SDexuan Cui * hvs_open_connection(). 323685703b4SDexuan Cui */ 324685703b4SDexuan Cui sock_put(sk); 325ae0078fcSDexuan Cui } 326ae0078fcSDexuan Cui 327ae0078fcSDexuan Cui static void hvs_open_connection(struct vmbus_channel *chan) 328ae0078fcSDexuan Cui { 329ce103204SAndy Shevchenko guid_t *if_instance, *if_type; 330ae0078fcSDexuan Cui unsigned char conn_from_host; 331ae0078fcSDexuan Cui 332ae0078fcSDexuan Cui struct sockaddr_vm addr; 333ae0078fcSDexuan Cui struct sock *sk, *new = NULL; 334ac383f58SSunil Muthuswamy struct vsock_sock *vnew = NULL; 335ac383f58SSunil Muthuswamy struct hvsock *hvs = NULL; 336ac383f58SSunil Muthuswamy struct hvsock *hvs_new = NULL; 337ac383f58SSunil Muthuswamy int rcvbuf; 338ae0078fcSDexuan Cui int ret; 339ac383f58SSunil Muthuswamy int sndbuf; 340ae0078fcSDexuan Cui 341ae0078fcSDexuan Cui if_type = &chan->offermsg.offer.if_type; 342ae0078fcSDexuan Cui if_instance = &chan->offermsg.offer.if_instance; 343ae0078fcSDexuan Cui conn_from_host = chan->offermsg.offer.u.pipe.user_def[0]; 344ae0078fcSDexuan Cui 345ae0078fcSDexuan Cui /* The host or the VM should only listen on a port in 346ae0078fcSDexuan Cui * [0, MAX_LISTEN_PORT] 347ae0078fcSDexuan Cui */ 348ae0078fcSDexuan Cui if (!is_valid_srv_id(if_type) || 349ae0078fcSDexuan Cui get_port_by_srv_id(if_type) > MAX_LISTEN_PORT) 350ae0078fcSDexuan Cui return; 351ae0078fcSDexuan Cui 352ae0078fcSDexuan Cui hvs_addr_init(&addr, conn_from_host ? if_type : if_instance); 353ae0078fcSDexuan Cui sk = vsock_find_bound_socket(&addr); 354ae0078fcSDexuan Cui if (!sk) 355ae0078fcSDexuan Cui return; 356ae0078fcSDexuan Cui 357b4562ca7SDexuan Cui lock_sock(sk); 3583b4477d2SStefan Hajnoczi if ((conn_from_host && sk->sk_state != TCP_LISTEN) || 3593b4477d2SStefan Hajnoczi (!conn_from_host && sk->sk_state != TCP_SYN_SENT)) 360ae0078fcSDexuan Cui goto out; 361ae0078fcSDexuan Cui 362ae0078fcSDexuan Cui if (conn_from_host) { 363ae0078fcSDexuan Cui if (sk->sk_ack_backlog >= sk->sk_max_ack_backlog) 364ae0078fcSDexuan Cui goto out; 365ae0078fcSDexuan Cui 366b9ca2f5fSStefano Garzarella new = vsock_create_connected(sk); 367ae0078fcSDexuan Cui if (!new) 368ae0078fcSDexuan Cui goto out; 369ae0078fcSDexuan Cui 3703b4477d2SStefan Hajnoczi new->sk_state = TCP_SYN_SENT; 371ae0078fcSDexuan Cui vnew = vsock_sk(new); 372c0cfa2d8SStefano Garzarella 373c0cfa2d8SStefano Garzarella hvs_addr_init(&vnew->local_addr, if_type); 374c0cfa2d8SStefano Garzarella hvs_remote_addr_init(&vnew->remote_addr, &vnew->local_addr); 375c0cfa2d8SStefano Garzarella 376c0cfa2d8SStefano Garzarella ret = vsock_assign_transport(vnew, vsock_sk(sk)); 377c0cfa2d8SStefano Garzarella /* Transport assigned (looking at remote_addr) must be the 378c0cfa2d8SStefano Garzarella * same where we received the request. 379c0cfa2d8SStefano Garzarella */ 380c0cfa2d8SStefano Garzarella if (ret || !hvs_check_transport(vnew)) { 381c0cfa2d8SStefano Garzarella sock_put(new); 382c0cfa2d8SStefano Garzarella goto out; 383c0cfa2d8SStefano Garzarella } 384ae0078fcSDexuan Cui hvs_new = vnew->trans; 385ae0078fcSDexuan Cui hvs_new->chan = chan; 386ae0078fcSDexuan Cui } else { 387ae0078fcSDexuan Cui hvs = vsock_sk(sk)->trans; 388ae0078fcSDexuan Cui hvs->chan = chan; 389ae0078fcSDexuan Cui } 390ae0078fcSDexuan Cui 391ae0078fcSDexuan Cui set_channel_read_mode(chan, HV_CALL_DIRECT); 392ac383f58SSunil Muthuswamy 393ac383f58SSunil Muthuswamy /* Use the socket buffer sizes as hints for the VMBUS ring size. For 394ac383f58SSunil Muthuswamy * server side sockets, 'sk' is the parent socket and thus, this will 395ac383f58SSunil Muthuswamy * allow the child sockets to inherit the size from the parent. Keep 396ac383f58SSunil Muthuswamy * the mins to the default value and align to page size as per VMBUS 397ac383f58SSunil Muthuswamy * requirements. 398ac383f58SSunil Muthuswamy * For the max, the socket core library will limit the socket buffer 399ac383f58SSunil Muthuswamy * size that can be set by the user, but, since currently, the hv_sock 400ac383f58SSunil Muthuswamy * VMBUS ring buffer is physically contiguous allocation, restrict it 401ac383f58SSunil Muthuswamy * further. 402ac383f58SSunil Muthuswamy * Older versions of hv_sock host side code cannot handle bigger VMBUS 403ac383f58SSunil Muthuswamy * ring buffer size. Use the version number to limit the change to newer 404ac383f58SSunil Muthuswamy * versions. 405ac383f58SSunil Muthuswamy */ 406ac383f58SSunil Muthuswamy if (vmbus_proto_version < VERSION_WIN10_V5) { 407ac383f58SSunil Muthuswamy sndbuf = RINGBUFFER_HVS_SND_SIZE; 408ac383f58SSunil Muthuswamy rcvbuf = RINGBUFFER_HVS_RCV_SIZE; 409ac383f58SSunil Muthuswamy } else { 410ac383f58SSunil Muthuswamy sndbuf = max_t(int, sk->sk_sndbuf, RINGBUFFER_HVS_SND_SIZE); 411ac383f58SSunil Muthuswamy sndbuf = min_t(int, sndbuf, RINGBUFFER_HVS_MAX_SIZE); 41277ffe333SHimadri Pandya sndbuf = ALIGN(sndbuf, HV_HYP_PAGE_SIZE); 413ac383f58SSunil Muthuswamy rcvbuf = max_t(int, sk->sk_rcvbuf, RINGBUFFER_HVS_RCV_SIZE); 414ac383f58SSunil Muthuswamy rcvbuf = min_t(int, rcvbuf, RINGBUFFER_HVS_MAX_SIZE); 41577ffe333SHimadri Pandya rcvbuf = ALIGN(rcvbuf, HV_HYP_PAGE_SIZE); 416ac383f58SSunil Muthuswamy } 417ac383f58SSunil Muthuswamy 418ac383f58SSunil Muthuswamy ret = vmbus_open(chan, sndbuf, rcvbuf, NULL, 0, hvs_channel_cb, 419ac383f58SSunil Muthuswamy conn_from_host ? new : sk); 420ae0078fcSDexuan Cui if (ret != 0) { 421ae0078fcSDexuan Cui if (conn_from_host) { 422ae0078fcSDexuan Cui hvs_new->chan = NULL; 423ae0078fcSDexuan Cui sock_put(new); 424ae0078fcSDexuan Cui } else { 425ae0078fcSDexuan Cui hvs->chan = NULL; 426ae0078fcSDexuan Cui } 427ae0078fcSDexuan Cui goto out; 428ae0078fcSDexuan Cui } 429ae0078fcSDexuan Cui 430ae0078fcSDexuan Cui set_per_channel_state(chan, conn_from_host ? new : sk); 431685703b4SDexuan Cui 432685703b4SDexuan Cui /* This reference will be dropped by hvs_close_connection(). */ 433685703b4SDexuan Cui sock_hold(conn_from_host ? new : sk); 434ae0078fcSDexuan Cui vmbus_set_chn_rescind_callback(chan, hvs_close_connection); 435ae0078fcSDexuan Cui 436cb359b60SSunil Muthuswamy /* Set the pending send size to max packet size to always get 437cb359b60SSunil Muthuswamy * notifications from the host when there is enough writable space. 438cb359b60SSunil Muthuswamy * The host is optimized to send notifications only when the pending 439cb359b60SSunil Muthuswamy * size boundary is crossed, and not always. 440cb359b60SSunil Muthuswamy */ 441cb359b60SSunil Muthuswamy hvs_set_channel_pending_send_size(chan); 442cb359b60SSunil Muthuswamy 443ae0078fcSDexuan Cui if (conn_from_host) { 4443b4477d2SStefan Hajnoczi new->sk_state = TCP_ESTABLISHED; 4457976a11bSEric Dumazet sk_acceptq_added(sk); 446ae0078fcSDexuan Cui 447ae0078fcSDexuan Cui hvs_new->vm_srv_id = *if_type; 448ae0078fcSDexuan Cui hvs_new->host_srv_id = *if_instance; 449ae0078fcSDexuan Cui 450ae0078fcSDexuan Cui vsock_insert_connected(vnew); 451ae0078fcSDexuan Cui 452ae0078fcSDexuan Cui vsock_enqueue_accept(sk, new); 453ae0078fcSDexuan Cui } else { 4543b4477d2SStefan Hajnoczi sk->sk_state = TCP_ESTABLISHED; 455ae0078fcSDexuan Cui sk->sk_socket->state = SS_CONNECTED; 456ae0078fcSDexuan Cui 457ae0078fcSDexuan Cui vsock_insert_connected(vsock_sk(sk)); 458ae0078fcSDexuan Cui } 459ae0078fcSDexuan Cui 460ae0078fcSDexuan Cui sk->sk_state_change(sk); 461ae0078fcSDexuan Cui 462ae0078fcSDexuan Cui out: 463ae0078fcSDexuan Cui /* Release refcnt obtained when we called vsock_find_bound_socket() */ 464ae0078fcSDexuan Cui sock_put(sk); 465b4562ca7SDexuan Cui 466b4562ca7SDexuan Cui release_sock(sk); 467ae0078fcSDexuan Cui } 468ae0078fcSDexuan Cui 469ae0078fcSDexuan Cui static u32 hvs_get_local_cid(void) 470ae0078fcSDexuan Cui { 471ae0078fcSDexuan Cui return VMADDR_CID_ANY; 472ae0078fcSDexuan Cui } 473ae0078fcSDexuan Cui 474ae0078fcSDexuan Cui static int hvs_sock_init(struct vsock_sock *vsk, struct vsock_sock *psk) 475ae0078fcSDexuan Cui { 476ae0078fcSDexuan Cui struct hvsock *hvs; 477ac383f58SSunil Muthuswamy struct sock *sk = sk_vsock(vsk); 478ae0078fcSDexuan Cui 479ae0078fcSDexuan Cui hvs = kzalloc(sizeof(*hvs), GFP_KERNEL); 480ae0078fcSDexuan Cui if (!hvs) 481ae0078fcSDexuan Cui return -ENOMEM; 482ae0078fcSDexuan Cui 483ae0078fcSDexuan Cui vsk->trans = hvs; 484ae0078fcSDexuan Cui hvs->vsk = vsk; 485ac383f58SSunil Muthuswamy sk->sk_sndbuf = RINGBUFFER_HVS_SND_SIZE; 486ac383f58SSunil Muthuswamy sk->sk_rcvbuf = RINGBUFFER_HVS_RCV_SIZE; 487ae0078fcSDexuan Cui return 0; 488ae0078fcSDexuan Cui } 489ae0078fcSDexuan Cui 490ae0078fcSDexuan Cui static int hvs_connect(struct vsock_sock *vsk) 491ae0078fcSDexuan Cui { 492ae0078fcSDexuan Cui union hvs_service_id vm, host; 493ae0078fcSDexuan Cui struct hvsock *h = vsk->trans; 494ae0078fcSDexuan Cui 495ae0078fcSDexuan Cui vm.srv_id = srv_id_template; 496ae0078fcSDexuan Cui vm.svm_port = vsk->local_addr.svm_port; 497ae0078fcSDexuan Cui h->vm_srv_id = vm.srv_id; 498ae0078fcSDexuan Cui 499ae0078fcSDexuan Cui host.srv_id = srv_id_template; 500ae0078fcSDexuan Cui host.svm_port = vsk->remote_addr.svm_port; 501ae0078fcSDexuan Cui h->host_srv_id = host.srv_id; 502ae0078fcSDexuan Cui 503ae0078fcSDexuan Cui return vmbus_send_tl_connect_request(&h->vm_srv_id, &h->host_srv_id); 504ae0078fcSDexuan Cui } 505ae0078fcSDexuan Cui 506a9eeb998SSunil Muthuswamy static void hvs_shutdown_lock_held(struct hvsock *hvs, int mode) 507a9eeb998SSunil Muthuswamy { 508a9eeb998SSunil Muthuswamy struct vmpipe_proto_header hdr; 509a9eeb998SSunil Muthuswamy 510a9eeb998SSunil Muthuswamy if (hvs->fin_sent || !hvs->chan) 511a9eeb998SSunil Muthuswamy return; 512a9eeb998SSunil Muthuswamy 513a9eeb998SSunil Muthuswamy /* It can't fail: see hvs_channel_writable_bytes(). */ 514a9eeb998SSunil Muthuswamy (void)hvs_send_data(hvs->chan, (struct hvs_send_buf *)&hdr, 0); 515a9eeb998SSunil Muthuswamy hvs->fin_sent = true; 516a9eeb998SSunil Muthuswamy } 517a9eeb998SSunil Muthuswamy 518ae0078fcSDexuan Cui static int hvs_shutdown(struct vsock_sock *vsk, int mode) 519ae0078fcSDexuan Cui { 520ae0078fcSDexuan Cui struct sock *sk = sk_vsock(vsk); 521ae0078fcSDexuan Cui 522ae0078fcSDexuan Cui if (!(mode & SEND_SHUTDOWN)) 523ae0078fcSDexuan Cui return 0; 524ae0078fcSDexuan Cui 525ae0078fcSDexuan Cui lock_sock(sk); 526a9eeb998SSunil Muthuswamy hvs_shutdown_lock_held(vsk->trans, mode); 527ae0078fcSDexuan Cui release_sock(sk); 528ae0078fcSDexuan Cui return 0; 529ae0078fcSDexuan Cui } 530ae0078fcSDexuan Cui 531a9eeb998SSunil Muthuswamy static void hvs_close_timeout(struct work_struct *work) 532a9eeb998SSunil Muthuswamy { 533a9eeb998SSunil Muthuswamy struct vsock_sock *vsk = 534a9eeb998SSunil Muthuswamy container_of(work, struct vsock_sock, close_work.work); 535a9eeb998SSunil Muthuswamy struct sock *sk = sk_vsock(vsk); 536a9eeb998SSunil Muthuswamy 537a9eeb998SSunil Muthuswamy sock_hold(sk); 538a9eeb998SSunil Muthuswamy lock_sock(sk); 539a9eeb998SSunil Muthuswamy if (!sock_flag(sk, SOCK_DONE)) 540a9eeb998SSunil Muthuswamy hvs_do_close_lock_held(vsk, false); 541a9eeb998SSunil Muthuswamy 542a9eeb998SSunil Muthuswamy vsk->close_work_scheduled = false; 543a9eeb998SSunil Muthuswamy release_sock(sk); 544a9eeb998SSunil Muthuswamy sock_put(sk); 545a9eeb998SSunil Muthuswamy } 546a9eeb998SSunil Muthuswamy 547a9eeb998SSunil Muthuswamy /* Returns true, if it is safe to remove socket; false otherwise */ 548a9eeb998SSunil Muthuswamy static bool hvs_close_lock_held(struct vsock_sock *vsk) 549a9eeb998SSunil Muthuswamy { 550a9eeb998SSunil Muthuswamy struct sock *sk = sk_vsock(vsk); 551a9eeb998SSunil Muthuswamy 552a9eeb998SSunil Muthuswamy if (!(sk->sk_state == TCP_ESTABLISHED || 553a9eeb998SSunil Muthuswamy sk->sk_state == TCP_CLOSING)) 554a9eeb998SSunil Muthuswamy return true; 555a9eeb998SSunil Muthuswamy 556a9eeb998SSunil Muthuswamy if ((sk->sk_shutdown & SHUTDOWN_MASK) != SHUTDOWN_MASK) 557a9eeb998SSunil Muthuswamy hvs_shutdown_lock_held(vsk->trans, SHUTDOWN_MASK); 558a9eeb998SSunil Muthuswamy 559a9eeb998SSunil Muthuswamy if (sock_flag(sk, SOCK_DONE)) 560a9eeb998SSunil Muthuswamy return true; 561a9eeb998SSunil Muthuswamy 562a9eeb998SSunil Muthuswamy /* This reference will be dropped by the delayed close routine */ 563a9eeb998SSunil Muthuswamy sock_hold(sk); 564a9eeb998SSunil Muthuswamy INIT_DELAYED_WORK(&vsk->close_work, hvs_close_timeout); 565a9eeb998SSunil Muthuswamy vsk->close_work_scheduled = true; 566a9eeb998SSunil Muthuswamy schedule_delayed_work(&vsk->close_work, HVS_CLOSE_TIMEOUT); 567a9eeb998SSunil Muthuswamy return false; 568a9eeb998SSunil Muthuswamy } 569a9eeb998SSunil Muthuswamy 570ae0078fcSDexuan Cui static void hvs_release(struct vsock_sock *vsk) 571ae0078fcSDexuan Cui { 572b4562ca7SDexuan Cui struct sock *sk = sk_vsock(vsk); 573a9eeb998SSunil Muthuswamy bool remove_sock; 574ae0078fcSDexuan Cui 5750d9138ffSDexuan Cui lock_sock_nested(sk, SINGLE_DEPTH_NESTING); 576a9eeb998SSunil Muthuswamy remove_sock = hvs_close_lock_held(vsk); 577b4562ca7SDexuan Cui release_sock(sk); 578a9eeb998SSunil Muthuswamy if (remove_sock) 579a9eeb998SSunil Muthuswamy vsock_remove_sock(vsk); 580ae0078fcSDexuan Cui } 581ae0078fcSDexuan Cui 582ae0078fcSDexuan Cui static void hvs_destruct(struct vsock_sock *vsk) 583ae0078fcSDexuan Cui { 584ae0078fcSDexuan Cui struct hvsock *hvs = vsk->trans; 585ae0078fcSDexuan Cui struct vmbus_channel *chan = hvs->chan; 586ae0078fcSDexuan Cui 587ae0078fcSDexuan Cui if (chan) 588ae0078fcSDexuan Cui vmbus_hvsock_device_unregister(chan); 589ae0078fcSDexuan Cui 590ae0078fcSDexuan Cui kfree(hvs); 591ae0078fcSDexuan Cui } 592ae0078fcSDexuan Cui 593ae0078fcSDexuan Cui static int hvs_dgram_bind(struct vsock_sock *vsk, struct sockaddr_vm *addr) 594ae0078fcSDexuan Cui { 595ae0078fcSDexuan Cui return -EOPNOTSUPP; 596ae0078fcSDexuan Cui } 597ae0078fcSDexuan Cui 598ae0078fcSDexuan Cui static int hvs_dgram_dequeue(struct vsock_sock *vsk, struct msghdr *msg, 599ae0078fcSDexuan Cui size_t len, int flags) 600ae0078fcSDexuan Cui { 601ae0078fcSDexuan Cui return -EOPNOTSUPP; 602ae0078fcSDexuan Cui } 603ae0078fcSDexuan Cui 604ae0078fcSDexuan Cui static int hvs_dgram_enqueue(struct vsock_sock *vsk, 605ae0078fcSDexuan Cui struct sockaddr_vm *remote, struct msghdr *msg, 606ae0078fcSDexuan Cui size_t dgram_len) 607ae0078fcSDexuan Cui { 608ae0078fcSDexuan Cui return -EOPNOTSUPP; 609ae0078fcSDexuan Cui } 610ae0078fcSDexuan Cui 611ae0078fcSDexuan Cui static bool hvs_dgram_allow(u32 cid, u32 port) 612ae0078fcSDexuan Cui { 613ae0078fcSDexuan Cui return false; 614ae0078fcSDexuan Cui } 615ae0078fcSDexuan Cui 616ae0078fcSDexuan Cui static int hvs_update_recv_data(struct hvsock *hvs) 617ae0078fcSDexuan Cui { 618ae0078fcSDexuan Cui struct hvs_recv_buf *recv_buf; 619ae0078fcSDexuan Cui u32 payload_len; 620ae0078fcSDexuan Cui 621ae0078fcSDexuan Cui recv_buf = (struct hvs_recv_buf *)(hvs->recv_desc + 1); 622ae0078fcSDexuan Cui payload_len = recv_buf->hdr.data_size; 623ae0078fcSDexuan Cui 624ae0078fcSDexuan Cui if (payload_len > HVS_MTU_SIZE) 625ae0078fcSDexuan Cui return -EIO; 626ae0078fcSDexuan Cui 627ae0078fcSDexuan Cui if (payload_len == 0) 628ae0078fcSDexuan Cui hvs->vsk->peer_shutdown |= SEND_SHUTDOWN; 629ae0078fcSDexuan Cui 630ae0078fcSDexuan Cui hvs->recv_data_len = payload_len; 631ae0078fcSDexuan Cui hvs->recv_data_off = 0; 632ae0078fcSDexuan Cui 633ae0078fcSDexuan Cui return 0; 634ae0078fcSDexuan Cui } 635ae0078fcSDexuan Cui 636ae0078fcSDexuan Cui static ssize_t hvs_stream_dequeue(struct vsock_sock *vsk, struct msghdr *msg, 637ae0078fcSDexuan Cui size_t len, int flags) 638ae0078fcSDexuan Cui { 639ae0078fcSDexuan Cui struct hvsock *hvs = vsk->trans; 640ae0078fcSDexuan Cui bool need_refill = !hvs->recv_desc; 641ae0078fcSDexuan Cui struct hvs_recv_buf *recv_buf; 642ae0078fcSDexuan Cui u32 to_read; 643ae0078fcSDexuan Cui int ret; 644ae0078fcSDexuan Cui 645ae0078fcSDexuan Cui if (flags & MSG_PEEK) 646ae0078fcSDexuan Cui return -EOPNOTSUPP; 647ae0078fcSDexuan Cui 648ae0078fcSDexuan Cui if (need_refill) { 649ae0078fcSDexuan Cui hvs->recv_desc = hv_pkt_iter_first(hvs->chan); 650ae0078fcSDexuan Cui ret = hvs_update_recv_data(hvs); 651ae0078fcSDexuan Cui if (ret) 652ae0078fcSDexuan Cui return ret; 653ae0078fcSDexuan Cui } 654ae0078fcSDexuan Cui 655ae0078fcSDexuan Cui recv_buf = (struct hvs_recv_buf *)(hvs->recv_desc + 1); 656ae0078fcSDexuan Cui to_read = min_t(u32, len, hvs->recv_data_len); 657ae0078fcSDexuan Cui ret = memcpy_to_msg(msg, recv_buf->data + hvs->recv_data_off, to_read); 658ae0078fcSDexuan Cui if (ret != 0) 659ae0078fcSDexuan Cui return ret; 660ae0078fcSDexuan Cui 661ae0078fcSDexuan Cui hvs->recv_data_len -= to_read; 662ae0078fcSDexuan Cui if (hvs->recv_data_len == 0) { 663ae0078fcSDexuan Cui hvs->recv_desc = hv_pkt_iter_next(hvs->chan, hvs->recv_desc); 664ae0078fcSDexuan Cui if (hvs->recv_desc) { 665ae0078fcSDexuan Cui ret = hvs_update_recv_data(hvs); 666ae0078fcSDexuan Cui if (ret) 667ae0078fcSDexuan Cui return ret; 668ae0078fcSDexuan Cui } 669ae0078fcSDexuan Cui } else { 670ae0078fcSDexuan Cui hvs->recv_data_off += to_read; 671ae0078fcSDexuan Cui } 672ae0078fcSDexuan Cui 673ae0078fcSDexuan Cui return to_read; 674ae0078fcSDexuan Cui } 675ae0078fcSDexuan Cui 676ae0078fcSDexuan Cui static ssize_t hvs_stream_enqueue(struct vsock_sock *vsk, struct msghdr *msg, 677ae0078fcSDexuan Cui size_t len) 678ae0078fcSDexuan Cui { 679ae0078fcSDexuan Cui struct hvsock *hvs = vsk->trans; 680ae0078fcSDexuan Cui struct vmbus_channel *chan = hvs->chan; 681ae0078fcSDexuan Cui struct hvs_send_buf *send_buf; 68214a1eaa8SSunil Muthuswamy ssize_t to_write, max_writable; 68314a1eaa8SSunil Muthuswamy ssize_t ret = 0; 68414a1eaa8SSunil Muthuswamy ssize_t bytes_written = 0; 685ae0078fcSDexuan Cui 68677ffe333SHimadri Pandya BUILD_BUG_ON(sizeof(*send_buf) != HV_HYP_PAGE_SIZE); 687ae0078fcSDexuan Cui 688ae0078fcSDexuan Cui send_buf = kmalloc(sizeof(*send_buf), GFP_KERNEL); 689ae0078fcSDexuan Cui if (!send_buf) 690ae0078fcSDexuan Cui return -ENOMEM; 691ae0078fcSDexuan Cui 69214a1eaa8SSunil Muthuswamy /* Reader(s) could be draining data from the channel as we write. 69314a1eaa8SSunil Muthuswamy * Maximize bandwidth, by iterating until the channel is found to be 69414a1eaa8SSunil Muthuswamy * full. 69514a1eaa8SSunil Muthuswamy */ 69614a1eaa8SSunil Muthuswamy while (len) { 697ae0078fcSDexuan Cui max_writable = hvs_channel_writable_bytes(chan); 69814a1eaa8SSunil Muthuswamy if (!max_writable) 69914a1eaa8SSunil Muthuswamy break; 700ae0078fcSDexuan Cui to_write = min_t(ssize_t, len, max_writable); 701ae0078fcSDexuan Cui to_write = min_t(ssize_t, to_write, HVS_SEND_BUF_SIZE); 70214a1eaa8SSunil Muthuswamy /* memcpy_from_msg is safe for loop as it advances the offsets 70314a1eaa8SSunil Muthuswamy * within the message iterator. 70414a1eaa8SSunil Muthuswamy */ 705ae0078fcSDexuan Cui ret = memcpy_from_msg(send_buf->data, msg, to_write); 706ae0078fcSDexuan Cui if (ret < 0) 707ae0078fcSDexuan Cui goto out; 708ae0078fcSDexuan Cui 709ae0078fcSDexuan Cui ret = hvs_send_data(hvs->chan, send_buf, to_write); 710ae0078fcSDexuan Cui if (ret < 0) 711ae0078fcSDexuan Cui goto out; 712ae0078fcSDexuan Cui 71314a1eaa8SSunil Muthuswamy bytes_written += to_write; 71414a1eaa8SSunil Muthuswamy len -= to_write; 71514a1eaa8SSunil Muthuswamy } 716ae0078fcSDexuan Cui out: 71714a1eaa8SSunil Muthuswamy /* If any data has been sent, return that */ 71814a1eaa8SSunil Muthuswamy if (bytes_written) 71914a1eaa8SSunil Muthuswamy ret = bytes_written; 720ae0078fcSDexuan Cui kfree(send_buf); 721ae0078fcSDexuan Cui return ret; 722ae0078fcSDexuan Cui } 723ae0078fcSDexuan Cui 724ae0078fcSDexuan Cui static s64 hvs_stream_has_data(struct vsock_sock *vsk) 725ae0078fcSDexuan Cui { 726ae0078fcSDexuan Cui struct hvsock *hvs = vsk->trans; 727ae0078fcSDexuan Cui s64 ret; 728ae0078fcSDexuan Cui 729ae0078fcSDexuan Cui if (hvs->recv_data_len > 0) 730ae0078fcSDexuan Cui return 1; 731ae0078fcSDexuan Cui 732ae0078fcSDexuan Cui switch (hvs_channel_readable_payload(hvs->chan)) { 733ae0078fcSDexuan Cui case 1: 734ae0078fcSDexuan Cui ret = 1; 735ae0078fcSDexuan Cui break; 736ae0078fcSDexuan Cui case 0: 737ae0078fcSDexuan Cui vsk->peer_shutdown |= SEND_SHUTDOWN; 738ae0078fcSDexuan Cui ret = 0; 739ae0078fcSDexuan Cui break; 740ae0078fcSDexuan Cui default: /* -1 */ 741ae0078fcSDexuan Cui ret = 0; 742ae0078fcSDexuan Cui break; 743ae0078fcSDexuan Cui } 744ae0078fcSDexuan Cui 745ae0078fcSDexuan Cui return ret; 746ae0078fcSDexuan Cui } 747ae0078fcSDexuan Cui 748ae0078fcSDexuan Cui static s64 hvs_stream_has_space(struct vsock_sock *vsk) 749ae0078fcSDexuan Cui { 750ae0078fcSDexuan Cui struct hvsock *hvs = vsk->trans; 751ae0078fcSDexuan Cui 752cb359b60SSunil Muthuswamy return hvs_channel_writable_bytes(hvs->chan); 753ae0078fcSDexuan Cui } 754ae0078fcSDexuan Cui 755ae0078fcSDexuan Cui static u64 hvs_stream_rcvhiwat(struct vsock_sock *vsk) 756ae0078fcSDexuan Cui { 757ae0078fcSDexuan Cui return HVS_MTU_SIZE + 1; 758ae0078fcSDexuan Cui } 759ae0078fcSDexuan Cui 760ae0078fcSDexuan Cui static bool hvs_stream_is_active(struct vsock_sock *vsk) 761ae0078fcSDexuan Cui { 762ae0078fcSDexuan Cui struct hvsock *hvs = vsk->trans; 763ae0078fcSDexuan Cui 764ae0078fcSDexuan Cui return hvs->chan != NULL; 765ae0078fcSDexuan Cui } 766ae0078fcSDexuan Cui 767ae0078fcSDexuan Cui static bool hvs_stream_allow(u32 cid, u32 port) 768ae0078fcSDexuan Cui { 769ae0078fcSDexuan Cui /* The host's port range [MIN_HOST_EPHEMERAL_PORT, 0xFFFFFFFF) is 770ae0078fcSDexuan Cui * reserved as ephemeral ports, which are used as the host's ports 771ae0078fcSDexuan Cui * when the host initiates connections. 772ae0078fcSDexuan Cui * 773ae0078fcSDexuan Cui * Perform this check in the guest so an immediate error is produced 774ae0078fcSDexuan Cui * instead of a timeout. 775ae0078fcSDexuan Cui */ 776ae0078fcSDexuan Cui if (port > MAX_HOST_LISTEN_PORT) 777ae0078fcSDexuan Cui return false; 778ae0078fcSDexuan Cui 779ae0078fcSDexuan Cui if (cid == VMADDR_CID_HOST) 780ae0078fcSDexuan Cui return true; 781ae0078fcSDexuan Cui 782ae0078fcSDexuan Cui return false; 783ae0078fcSDexuan Cui } 784ae0078fcSDexuan Cui 785ae0078fcSDexuan Cui static 786ae0078fcSDexuan Cui int hvs_notify_poll_in(struct vsock_sock *vsk, size_t target, bool *readable) 787ae0078fcSDexuan Cui { 788ae0078fcSDexuan Cui struct hvsock *hvs = vsk->trans; 789ae0078fcSDexuan Cui 790ae0078fcSDexuan Cui *readable = hvs_channel_readable(hvs->chan); 791ae0078fcSDexuan Cui return 0; 792ae0078fcSDexuan Cui } 793ae0078fcSDexuan Cui 794ae0078fcSDexuan Cui static 795ae0078fcSDexuan Cui int hvs_notify_poll_out(struct vsock_sock *vsk, size_t target, bool *writable) 796ae0078fcSDexuan Cui { 797ae0078fcSDexuan Cui *writable = hvs_stream_has_space(vsk) > 0; 798ae0078fcSDexuan Cui 799ae0078fcSDexuan Cui return 0; 800ae0078fcSDexuan Cui } 801ae0078fcSDexuan Cui 802ae0078fcSDexuan Cui static 803ae0078fcSDexuan Cui int hvs_notify_recv_init(struct vsock_sock *vsk, size_t target, 804ae0078fcSDexuan Cui struct vsock_transport_recv_notify_data *d) 805ae0078fcSDexuan Cui { 806ae0078fcSDexuan Cui return 0; 807ae0078fcSDexuan Cui } 808ae0078fcSDexuan Cui 809ae0078fcSDexuan Cui static 810ae0078fcSDexuan Cui int hvs_notify_recv_pre_block(struct vsock_sock *vsk, size_t target, 811ae0078fcSDexuan Cui struct vsock_transport_recv_notify_data *d) 812ae0078fcSDexuan Cui { 813ae0078fcSDexuan Cui return 0; 814ae0078fcSDexuan Cui } 815ae0078fcSDexuan Cui 816ae0078fcSDexuan Cui static 817ae0078fcSDexuan Cui int hvs_notify_recv_pre_dequeue(struct vsock_sock *vsk, size_t target, 818ae0078fcSDexuan Cui struct vsock_transport_recv_notify_data *d) 819ae0078fcSDexuan Cui { 820ae0078fcSDexuan Cui return 0; 821ae0078fcSDexuan Cui } 822ae0078fcSDexuan Cui 823ae0078fcSDexuan Cui static 824ae0078fcSDexuan Cui int hvs_notify_recv_post_dequeue(struct vsock_sock *vsk, size_t target, 825ae0078fcSDexuan Cui ssize_t copied, bool data_read, 826ae0078fcSDexuan Cui struct vsock_transport_recv_notify_data *d) 827ae0078fcSDexuan Cui { 828ae0078fcSDexuan Cui return 0; 829ae0078fcSDexuan Cui } 830ae0078fcSDexuan Cui 831ae0078fcSDexuan Cui static 832ae0078fcSDexuan Cui int hvs_notify_send_init(struct vsock_sock *vsk, 833ae0078fcSDexuan Cui struct vsock_transport_send_notify_data *d) 834ae0078fcSDexuan Cui { 835ae0078fcSDexuan Cui return 0; 836ae0078fcSDexuan Cui } 837ae0078fcSDexuan Cui 838ae0078fcSDexuan Cui static 839ae0078fcSDexuan Cui int hvs_notify_send_pre_block(struct vsock_sock *vsk, 840ae0078fcSDexuan Cui struct vsock_transport_send_notify_data *d) 841ae0078fcSDexuan Cui { 842ae0078fcSDexuan Cui return 0; 843ae0078fcSDexuan Cui } 844ae0078fcSDexuan Cui 845ae0078fcSDexuan Cui static 846ae0078fcSDexuan Cui int hvs_notify_send_pre_enqueue(struct vsock_sock *vsk, 847ae0078fcSDexuan Cui struct vsock_transport_send_notify_data *d) 848ae0078fcSDexuan Cui { 849ae0078fcSDexuan Cui return 0; 850ae0078fcSDexuan Cui } 851ae0078fcSDexuan Cui 852ae0078fcSDexuan Cui static 853ae0078fcSDexuan Cui int hvs_notify_send_post_enqueue(struct vsock_sock *vsk, ssize_t written, 854ae0078fcSDexuan Cui struct vsock_transport_send_notify_data *d) 855ae0078fcSDexuan Cui { 856ae0078fcSDexuan Cui return 0; 857ae0078fcSDexuan Cui } 858ae0078fcSDexuan Cui 859ae0078fcSDexuan Cui static struct vsock_transport hvs_transport = { 860ae0078fcSDexuan Cui .get_local_cid = hvs_get_local_cid, 861ae0078fcSDexuan Cui 862ae0078fcSDexuan Cui .init = hvs_sock_init, 863ae0078fcSDexuan Cui .destruct = hvs_destruct, 864ae0078fcSDexuan Cui .release = hvs_release, 865ae0078fcSDexuan Cui .connect = hvs_connect, 866ae0078fcSDexuan Cui .shutdown = hvs_shutdown, 867ae0078fcSDexuan Cui 868ae0078fcSDexuan Cui .dgram_bind = hvs_dgram_bind, 869ae0078fcSDexuan Cui .dgram_dequeue = hvs_dgram_dequeue, 870ae0078fcSDexuan Cui .dgram_enqueue = hvs_dgram_enqueue, 871ae0078fcSDexuan Cui .dgram_allow = hvs_dgram_allow, 872ae0078fcSDexuan Cui 873ae0078fcSDexuan Cui .stream_dequeue = hvs_stream_dequeue, 874ae0078fcSDexuan Cui .stream_enqueue = hvs_stream_enqueue, 875ae0078fcSDexuan Cui .stream_has_data = hvs_stream_has_data, 876ae0078fcSDexuan Cui .stream_has_space = hvs_stream_has_space, 877ae0078fcSDexuan Cui .stream_rcvhiwat = hvs_stream_rcvhiwat, 878ae0078fcSDexuan Cui .stream_is_active = hvs_stream_is_active, 879ae0078fcSDexuan Cui .stream_allow = hvs_stream_allow, 880ae0078fcSDexuan Cui 881ae0078fcSDexuan Cui .notify_poll_in = hvs_notify_poll_in, 882ae0078fcSDexuan Cui .notify_poll_out = hvs_notify_poll_out, 883ae0078fcSDexuan Cui .notify_recv_init = hvs_notify_recv_init, 884ae0078fcSDexuan Cui .notify_recv_pre_block = hvs_notify_recv_pre_block, 885ae0078fcSDexuan Cui .notify_recv_pre_dequeue = hvs_notify_recv_pre_dequeue, 886ae0078fcSDexuan Cui .notify_recv_post_dequeue = hvs_notify_recv_post_dequeue, 887ae0078fcSDexuan Cui .notify_send_init = hvs_notify_send_init, 888ae0078fcSDexuan Cui .notify_send_pre_block = hvs_notify_send_pre_block, 889ae0078fcSDexuan Cui .notify_send_pre_enqueue = hvs_notify_send_pre_enqueue, 890ae0078fcSDexuan Cui .notify_send_post_enqueue = hvs_notify_send_post_enqueue, 891ae0078fcSDexuan Cui 892ae0078fcSDexuan Cui }; 893ae0078fcSDexuan Cui 894c0cfa2d8SStefano Garzarella static bool hvs_check_transport(struct vsock_sock *vsk) 895c0cfa2d8SStefano Garzarella { 896c0cfa2d8SStefano Garzarella return vsk->transport == &hvs_transport; 897c0cfa2d8SStefano Garzarella } 898c0cfa2d8SStefano Garzarella 899ae0078fcSDexuan Cui static int hvs_probe(struct hv_device *hdev, 900ae0078fcSDexuan Cui const struct hv_vmbus_device_id *dev_id) 901ae0078fcSDexuan Cui { 902ae0078fcSDexuan Cui struct vmbus_channel *chan = hdev->channel; 903ae0078fcSDexuan Cui 904ae0078fcSDexuan Cui hvs_open_connection(chan); 905ae0078fcSDexuan Cui 906ae0078fcSDexuan Cui /* Always return success to suppress the unnecessary error message 907ae0078fcSDexuan Cui * in vmbus_probe(): on error the host will rescind the device in 908ae0078fcSDexuan Cui * 30 seconds and we can do cleanup at that time in 909ae0078fcSDexuan Cui * vmbus_onoffer_rescind(). 910ae0078fcSDexuan Cui */ 911ae0078fcSDexuan Cui return 0; 912ae0078fcSDexuan Cui } 913ae0078fcSDexuan Cui 914ae0078fcSDexuan Cui static int hvs_remove(struct hv_device *hdev) 915ae0078fcSDexuan Cui { 916ae0078fcSDexuan Cui struct vmbus_channel *chan = hdev->channel; 917ae0078fcSDexuan Cui 918ae0078fcSDexuan Cui vmbus_close(chan); 919ae0078fcSDexuan Cui 920ae0078fcSDexuan Cui return 0; 921ae0078fcSDexuan Cui } 922ae0078fcSDexuan Cui 923ae0078fcSDexuan Cui /* This isn't really used. See vmbus_match() and vmbus_probe() */ 924ae0078fcSDexuan Cui static const struct hv_vmbus_device_id id_table[] = { 925ae0078fcSDexuan Cui {}, 926ae0078fcSDexuan Cui }; 927ae0078fcSDexuan Cui 928ae0078fcSDexuan Cui static struct hv_driver hvs_drv = { 929ae0078fcSDexuan Cui .name = "hv_sock", 930ae0078fcSDexuan Cui .hvsock = true, 931ae0078fcSDexuan Cui .id_table = id_table, 932ae0078fcSDexuan Cui .probe = hvs_probe, 933ae0078fcSDexuan Cui .remove = hvs_remove, 934ae0078fcSDexuan Cui }; 935ae0078fcSDexuan Cui 936ae0078fcSDexuan Cui static int __init hvs_init(void) 937ae0078fcSDexuan Cui { 938ae0078fcSDexuan Cui int ret; 939ae0078fcSDexuan Cui 940ae0078fcSDexuan Cui if (vmbus_proto_version < VERSION_WIN10) 941ae0078fcSDexuan Cui return -ENODEV; 942ae0078fcSDexuan Cui 943ae0078fcSDexuan Cui ret = vmbus_driver_register(&hvs_drv); 944ae0078fcSDexuan Cui if (ret != 0) 945ae0078fcSDexuan Cui return ret; 946ae0078fcSDexuan Cui 947c0cfa2d8SStefano Garzarella ret = vsock_core_register(&hvs_transport, VSOCK_TRANSPORT_F_G2H); 948ae0078fcSDexuan Cui if (ret) { 949ae0078fcSDexuan Cui vmbus_driver_unregister(&hvs_drv); 950ae0078fcSDexuan Cui return ret; 951ae0078fcSDexuan Cui } 952ae0078fcSDexuan Cui 953ae0078fcSDexuan Cui return 0; 954ae0078fcSDexuan Cui } 955ae0078fcSDexuan Cui 956ae0078fcSDexuan Cui static void __exit hvs_exit(void) 957ae0078fcSDexuan Cui { 958c0cfa2d8SStefano Garzarella vsock_core_unregister(&hvs_transport); 959ae0078fcSDexuan Cui vmbus_driver_unregister(&hvs_drv); 960ae0078fcSDexuan Cui } 961ae0078fcSDexuan Cui 962ae0078fcSDexuan Cui module_init(hvs_init); 963ae0078fcSDexuan Cui module_exit(hvs_exit); 964ae0078fcSDexuan Cui 965ae0078fcSDexuan Cui MODULE_DESCRIPTION("Hyper-V Sockets"); 966ae0078fcSDexuan Cui MODULE_VERSION("1.0.0"); 967ae0078fcSDexuan Cui MODULE_LICENSE("GPL"); 968ae0078fcSDexuan Cui MODULE_ALIAS_NETPROTO(PF_VSOCK); 969