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 * 141c742c59eSSunil Muthuswamy * format: <port>-facb-11e6-bd58-64006a7986d3. * 142ae0078fcSDexuan Cui **************************************************************************** 143ae0078fcSDexuan Cui * 144ae0078fcSDexuan Cui * When we write apps on the host to connect(), the GUID ServiceID is used. 145ae0078fcSDexuan Cui * When we write apps in Linux VM to connect(), we only need to specify the 146ae0078fcSDexuan Cui * port and the driver will form the GUID and use that to request the host. 147ae0078fcSDexuan Cui * 148ae0078fcSDexuan Cui */ 149ae0078fcSDexuan Cui 150ae0078fcSDexuan Cui /* 00000000-facb-11e6-bd58-64006a7986d3 */ 151ce103204SAndy Shevchenko static const guid_t srv_id_template = 152ce103204SAndy Shevchenko GUID_INIT(0x00000000, 0xfacb, 0x11e6, 0xbd, 0x58, 153ae0078fcSDexuan Cui 0x64, 0x00, 0x6a, 0x79, 0x86, 0xd3); 154ae0078fcSDexuan Cui 155c0cfa2d8SStefano Garzarella static bool hvs_check_transport(struct vsock_sock *vsk); 156c0cfa2d8SStefano Garzarella 157ce103204SAndy Shevchenko static bool is_valid_srv_id(const guid_t *id) 158ae0078fcSDexuan Cui { 159ce103204SAndy Shevchenko return !memcmp(&id->b[4], &srv_id_template.b[4], sizeof(guid_t) - 4); 160ae0078fcSDexuan Cui } 161ae0078fcSDexuan Cui 162ce103204SAndy Shevchenko static unsigned int get_port_by_srv_id(const guid_t *svr_id) 163ae0078fcSDexuan Cui { 164ae0078fcSDexuan Cui return *((unsigned int *)svr_id); 165ae0078fcSDexuan Cui } 166ae0078fcSDexuan Cui 167ce103204SAndy Shevchenko static void hvs_addr_init(struct sockaddr_vm *addr, const guid_t *svr_id) 168ae0078fcSDexuan Cui { 169ae0078fcSDexuan Cui unsigned int port = get_port_by_srv_id(svr_id); 170ae0078fcSDexuan Cui 171ae0078fcSDexuan Cui vsock_addr_init(addr, VMADDR_CID_ANY, port); 172ae0078fcSDexuan Cui } 173ae0078fcSDexuan Cui 174ae0078fcSDexuan Cui static void hvs_set_channel_pending_send_size(struct vmbus_channel *chan) 175ae0078fcSDexuan Cui { 176ae0078fcSDexuan Cui set_channel_pending_send_size(chan, 177ae0078fcSDexuan Cui HVS_PKT_LEN(HVS_SEND_BUF_SIZE)); 178ae0078fcSDexuan Cui 179ae0078fcSDexuan Cui virt_mb(); 180ae0078fcSDexuan Cui } 181ae0078fcSDexuan Cui 182ae0078fcSDexuan Cui static bool hvs_channel_readable(struct vmbus_channel *chan) 183ae0078fcSDexuan Cui { 184ae0078fcSDexuan Cui u32 readable = hv_get_bytes_to_read(&chan->inbound); 185ae0078fcSDexuan Cui 186ae0078fcSDexuan Cui /* 0-size payload means FIN */ 187ae0078fcSDexuan Cui return readable >= HVS_PKT_LEN(0); 188ae0078fcSDexuan Cui } 189ae0078fcSDexuan Cui 190ae0078fcSDexuan Cui static int hvs_channel_readable_payload(struct vmbus_channel *chan) 191ae0078fcSDexuan Cui { 192ae0078fcSDexuan Cui u32 readable = hv_get_bytes_to_read(&chan->inbound); 193ae0078fcSDexuan Cui 194ae0078fcSDexuan Cui if (readable > HVS_PKT_LEN(0)) { 195ae0078fcSDexuan Cui /* At least we have 1 byte to read. We don't need to return 196ae0078fcSDexuan Cui * the exact readable bytes: see vsock_stream_recvmsg() -> 197ae0078fcSDexuan Cui * vsock_stream_has_data(). 198ae0078fcSDexuan Cui */ 199ae0078fcSDexuan Cui return 1; 200ae0078fcSDexuan Cui } 201ae0078fcSDexuan Cui 202ae0078fcSDexuan Cui if (readable == HVS_PKT_LEN(0)) { 203ae0078fcSDexuan Cui /* 0-size payload means FIN */ 204ae0078fcSDexuan Cui return 0; 205ae0078fcSDexuan Cui } 206ae0078fcSDexuan Cui 207ae0078fcSDexuan Cui /* No payload or FIN */ 208ae0078fcSDexuan Cui return -1; 209ae0078fcSDexuan Cui } 210ae0078fcSDexuan Cui 211ae0078fcSDexuan Cui static size_t hvs_channel_writable_bytes(struct vmbus_channel *chan) 212ae0078fcSDexuan Cui { 213ae0078fcSDexuan Cui u32 writeable = hv_get_bytes_to_write(&chan->outbound); 214ae0078fcSDexuan Cui size_t ret; 215ae0078fcSDexuan Cui 216ae0078fcSDexuan Cui /* The ringbuffer mustn't be 100% full, and we should reserve a 217ae0078fcSDexuan Cui * zero-length-payload packet for the FIN: see hv_ringbuffer_write() 218ae0078fcSDexuan Cui * and hvs_shutdown(). 219ae0078fcSDexuan Cui */ 220ae0078fcSDexuan Cui if (writeable <= HVS_PKT_LEN(1) + HVS_PKT_LEN(0)) 221ae0078fcSDexuan Cui return 0; 222ae0078fcSDexuan Cui 223ae0078fcSDexuan Cui ret = writeable - HVS_PKT_LEN(1) - HVS_PKT_LEN(0); 224ae0078fcSDexuan Cui 225ae0078fcSDexuan Cui return round_down(ret, 8); 226ae0078fcSDexuan Cui } 227ae0078fcSDexuan Cui 228*c0e084e3SKees Cook static int __hvs_send_data(struct vmbus_channel *chan, 229*c0e084e3SKees Cook struct vmpipe_proto_header *hdr, 230*c0e084e3SKees Cook size_t to_write) 231*c0e084e3SKees Cook { 232*c0e084e3SKees Cook hdr->pkt_type = 1; 233*c0e084e3SKees Cook hdr->data_size = to_write; 234*c0e084e3SKees Cook return vmbus_sendpacket(chan, hdr, sizeof(*hdr) + to_write, 235*c0e084e3SKees Cook 0, VM_PKT_DATA_INBAND, 0); 236*c0e084e3SKees Cook } 237*c0e084e3SKees Cook 238ae0078fcSDexuan Cui static int hvs_send_data(struct vmbus_channel *chan, 239ae0078fcSDexuan Cui struct hvs_send_buf *send_buf, size_t to_write) 240ae0078fcSDexuan Cui { 241*c0e084e3SKees Cook return __hvs_send_data(chan, &send_buf->hdr, to_write); 242ae0078fcSDexuan Cui } 243ae0078fcSDexuan Cui 244ae0078fcSDexuan Cui static void hvs_channel_cb(void *ctx) 245ae0078fcSDexuan Cui { 246ae0078fcSDexuan Cui struct sock *sk = (struct sock *)ctx; 247ae0078fcSDexuan Cui struct vsock_sock *vsk = vsock_sk(sk); 248ae0078fcSDexuan Cui struct hvsock *hvs = vsk->trans; 249ae0078fcSDexuan Cui struct vmbus_channel *chan = hvs->chan; 250ae0078fcSDexuan Cui 251ae0078fcSDexuan Cui if (hvs_channel_readable(chan)) 252ae0078fcSDexuan Cui sk->sk_data_ready(sk); 253ae0078fcSDexuan Cui 254ae0078fcSDexuan Cui if (hv_get_bytes_to_write(&chan->outbound) > 0) 255ae0078fcSDexuan Cui sk->sk_write_space(sk); 256ae0078fcSDexuan Cui } 257ae0078fcSDexuan Cui 258a9eeb998SSunil Muthuswamy static void hvs_do_close_lock_held(struct vsock_sock *vsk, 259a9eeb998SSunil Muthuswamy bool cancel_timeout) 260a9eeb998SSunil Muthuswamy { 261a9eeb998SSunil Muthuswamy struct sock *sk = sk_vsock(vsk); 262a9eeb998SSunil Muthuswamy 263a9eeb998SSunil Muthuswamy sock_set_flag(sk, SOCK_DONE); 264a9eeb998SSunil Muthuswamy vsk->peer_shutdown = SHUTDOWN_MASK; 265a9eeb998SSunil Muthuswamy if (vsock_stream_has_data(vsk) <= 0) 266a9eeb998SSunil Muthuswamy sk->sk_state = TCP_CLOSING; 267a9eeb998SSunil Muthuswamy sk->sk_state_change(sk); 268a9eeb998SSunil Muthuswamy if (vsk->close_work_scheduled && 269a9eeb998SSunil Muthuswamy (!cancel_timeout || cancel_delayed_work(&vsk->close_work))) { 270a9eeb998SSunil Muthuswamy vsk->close_work_scheduled = false; 271a9eeb998SSunil Muthuswamy vsock_remove_sock(vsk); 272a9eeb998SSunil Muthuswamy 273a9eeb998SSunil Muthuswamy /* Release the reference taken while scheduling the timeout */ 274a9eeb998SSunil Muthuswamy sock_put(sk); 275a9eeb998SSunil Muthuswamy } 276a9eeb998SSunil Muthuswamy } 277a9eeb998SSunil Muthuswamy 278ae0078fcSDexuan Cui static void hvs_close_connection(struct vmbus_channel *chan) 279ae0078fcSDexuan Cui { 280ae0078fcSDexuan Cui struct sock *sk = get_per_channel_state(chan); 281ae0078fcSDexuan Cui 282b4562ca7SDexuan Cui lock_sock(sk); 283a9eeb998SSunil Muthuswamy hvs_do_close_lock_held(vsock_sk(sk), true); 284b4562ca7SDexuan Cui release_sock(sk); 285685703b4SDexuan Cui 286685703b4SDexuan Cui /* Release the refcnt for the channel that's opened in 287685703b4SDexuan Cui * hvs_open_connection(). 288685703b4SDexuan Cui */ 289685703b4SDexuan Cui sock_put(sk); 290ae0078fcSDexuan Cui } 291ae0078fcSDexuan Cui 292ae0078fcSDexuan Cui static void hvs_open_connection(struct vmbus_channel *chan) 293ae0078fcSDexuan Cui { 294ce103204SAndy Shevchenko guid_t *if_instance, *if_type; 295ae0078fcSDexuan Cui unsigned char conn_from_host; 296ae0078fcSDexuan Cui 297ae0078fcSDexuan Cui struct sockaddr_vm addr; 298ae0078fcSDexuan Cui struct sock *sk, *new = NULL; 299ac383f58SSunil Muthuswamy struct vsock_sock *vnew = NULL; 300ac383f58SSunil Muthuswamy struct hvsock *hvs = NULL; 301ac383f58SSunil Muthuswamy struct hvsock *hvs_new = NULL; 302ac383f58SSunil Muthuswamy int rcvbuf; 303ae0078fcSDexuan Cui int ret; 304ac383f58SSunil Muthuswamy int sndbuf; 305ae0078fcSDexuan Cui 306ae0078fcSDexuan Cui if_type = &chan->offermsg.offer.if_type; 307ae0078fcSDexuan Cui if_instance = &chan->offermsg.offer.if_instance; 308ae0078fcSDexuan Cui conn_from_host = chan->offermsg.offer.u.pipe.user_def[0]; 309c742c59eSSunil Muthuswamy if (!is_valid_srv_id(if_type)) 310ae0078fcSDexuan Cui return; 311ae0078fcSDexuan Cui 312ae0078fcSDexuan Cui hvs_addr_init(&addr, conn_from_host ? if_type : if_instance); 313ae0078fcSDexuan Cui sk = vsock_find_bound_socket(&addr); 314ae0078fcSDexuan Cui if (!sk) 315ae0078fcSDexuan Cui return; 316ae0078fcSDexuan Cui 317b4562ca7SDexuan Cui lock_sock(sk); 3183b4477d2SStefan Hajnoczi if ((conn_from_host && sk->sk_state != TCP_LISTEN) || 3193b4477d2SStefan Hajnoczi (!conn_from_host && sk->sk_state != TCP_SYN_SENT)) 320ae0078fcSDexuan Cui goto out; 321ae0078fcSDexuan Cui 322ae0078fcSDexuan Cui if (conn_from_host) { 323ae0078fcSDexuan Cui if (sk->sk_ack_backlog >= sk->sk_max_ack_backlog) 324ae0078fcSDexuan Cui goto out; 325ae0078fcSDexuan Cui 326b9ca2f5fSStefano Garzarella new = vsock_create_connected(sk); 327ae0078fcSDexuan Cui if (!new) 328ae0078fcSDexuan Cui goto out; 329ae0078fcSDexuan Cui 3303b4477d2SStefan Hajnoczi new->sk_state = TCP_SYN_SENT; 331ae0078fcSDexuan Cui vnew = vsock_sk(new); 332c0cfa2d8SStefano Garzarella 333c0cfa2d8SStefano Garzarella hvs_addr_init(&vnew->local_addr, if_type); 334c0cfa2d8SStefano Garzarella 335c742c59eSSunil Muthuswamy /* Remote peer is always the host */ 336c742c59eSSunil Muthuswamy vsock_addr_init(&vnew->remote_addr, 337c742c59eSSunil Muthuswamy VMADDR_CID_HOST, VMADDR_PORT_ANY); 338c742c59eSSunil Muthuswamy vnew->remote_addr.svm_port = get_port_by_srv_id(if_instance); 339c0cfa2d8SStefano Garzarella ret = vsock_assign_transport(vnew, vsock_sk(sk)); 340c0cfa2d8SStefano Garzarella /* Transport assigned (looking at remote_addr) must be the 341c0cfa2d8SStefano Garzarella * same where we received the request. 342c0cfa2d8SStefano Garzarella */ 343c0cfa2d8SStefano Garzarella if (ret || !hvs_check_transport(vnew)) { 344c0cfa2d8SStefano Garzarella sock_put(new); 345c0cfa2d8SStefano Garzarella goto out; 346c0cfa2d8SStefano Garzarella } 347ae0078fcSDexuan Cui hvs_new = vnew->trans; 348ae0078fcSDexuan Cui hvs_new->chan = chan; 349ae0078fcSDexuan Cui } else { 350ae0078fcSDexuan Cui hvs = vsock_sk(sk)->trans; 351ae0078fcSDexuan Cui hvs->chan = chan; 352ae0078fcSDexuan Cui } 353ae0078fcSDexuan Cui 354ae0078fcSDexuan Cui set_channel_read_mode(chan, HV_CALL_DIRECT); 355ac383f58SSunil Muthuswamy 356ac383f58SSunil Muthuswamy /* Use the socket buffer sizes as hints for the VMBUS ring size. For 357ac383f58SSunil Muthuswamy * server side sockets, 'sk' is the parent socket and thus, this will 358ac383f58SSunil Muthuswamy * allow the child sockets to inherit the size from the parent. Keep 359ac383f58SSunil Muthuswamy * the mins to the default value and align to page size as per VMBUS 360ac383f58SSunil Muthuswamy * requirements. 361ac383f58SSunil Muthuswamy * For the max, the socket core library will limit the socket buffer 362ac383f58SSunil Muthuswamy * size that can be set by the user, but, since currently, the hv_sock 363ac383f58SSunil Muthuswamy * VMBUS ring buffer is physically contiguous allocation, restrict it 364ac383f58SSunil Muthuswamy * further. 365ac383f58SSunil Muthuswamy * Older versions of hv_sock host side code cannot handle bigger VMBUS 366ac383f58SSunil Muthuswamy * ring buffer size. Use the version number to limit the change to newer 367ac383f58SSunil Muthuswamy * versions. 368ac383f58SSunil Muthuswamy */ 369ac383f58SSunil Muthuswamy if (vmbus_proto_version < VERSION_WIN10_V5) { 370ac383f58SSunil Muthuswamy sndbuf = RINGBUFFER_HVS_SND_SIZE; 371ac383f58SSunil Muthuswamy rcvbuf = RINGBUFFER_HVS_RCV_SIZE; 372ac383f58SSunil Muthuswamy } else { 373ac383f58SSunil Muthuswamy sndbuf = max_t(int, sk->sk_sndbuf, RINGBUFFER_HVS_SND_SIZE); 374ac383f58SSunil Muthuswamy sndbuf = min_t(int, sndbuf, RINGBUFFER_HVS_MAX_SIZE); 37577ffe333SHimadri Pandya sndbuf = ALIGN(sndbuf, HV_HYP_PAGE_SIZE); 376ac383f58SSunil Muthuswamy rcvbuf = max_t(int, sk->sk_rcvbuf, RINGBUFFER_HVS_RCV_SIZE); 377ac383f58SSunil Muthuswamy rcvbuf = min_t(int, rcvbuf, RINGBUFFER_HVS_MAX_SIZE); 37877ffe333SHimadri Pandya rcvbuf = ALIGN(rcvbuf, HV_HYP_PAGE_SIZE); 379ac383f58SSunil Muthuswamy } 380ac383f58SSunil Muthuswamy 381ac383f58SSunil Muthuswamy ret = vmbus_open(chan, sndbuf, rcvbuf, NULL, 0, hvs_channel_cb, 382ac383f58SSunil Muthuswamy conn_from_host ? new : sk); 383ae0078fcSDexuan Cui if (ret != 0) { 384ae0078fcSDexuan Cui if (conn_from_host) { 385ae0078fcSDexuan Cui hvs_new->chan = NULL; 386ae0078fcSDexuan Cui sock_put(new); 387ae0078fcSDexuan Cui } else { 388ae0078fcSDexuan Cui hvs->chan = NULL; 389ae0078fcSDexuan Cui } 390ae0078fcSDexuan Cui goto out; 391ae0078fcSDexuan Cui } 392ae0078fcSDexuan Cui 393ae0078fcSDexuan Cui set_per_channel_state(chan, conn_from_host ? new : sk); 394685703b4SDexuan Cui 395685703b4SDexuan Cui /* This reference will be dropped by hvs_close_connection(). */ 396685703b4SDexuan Cui sock_hold(conn_from_host ? new : sk); 397ae0078fcSDexuan Cui vmbus_set_chn_rescind_callback(chan, hvs_close_connection); 398ae0078fcSDexuan Cui 399cb359b60SSunil Muthuswamy /* Set the pending send size to max packet size to always get 400cb359b60SSunil Muthuswamy * notifications from the host when there is enough writable space. 401cb359b60SSunil Muthuswamy * The host is optimized to send notifications only when the pending 402cb359b60SSunil Muthuswamy * size boundary is crossed, and not always. 403cb359b60SSunil Muthuswamy */ 404cb359b60SSunil Muthuswamy hvs_set_channel_pending_send_size(chan); 405cb359b60SSunil Muthuswamy 406ae0078fcSDexuan Cui if (conn_from_host) { 4073b4477d2SStefan Hajnoczi new->sk_state = TCP_ESTABLISHED; 4087976a11bSEric Dumazet sk_acceptq_added(sk); 409ae0078fcSDexuan Cui 410ae0078fcSDexuan Cui hvs_new->vm_srv_id = *if_type; 411ae0078fcSDexuan Cui hvs_new->host_srv_id = *if_instance; 412ae0078fcSDexuan Cui 413ae0078fcSDexuan Cui vsock_insert_connected(vnew); 414ae0078fcSDexuan Cui 415ae0078fcSDexuan Cui vsock_enqueue_accept(sk, new); 416ae0078fcSDexuan Cui } else { 4173b4477d2SStefan Hajnoczi sk->sk_state = TCP_ESTABLISHED; 418ae0078fcSDexuan Cui sk->sk_socket->state = SS_CONNECTED; 419ae0078fcSDexuan Cui 420ae0078fcSDexuan Cui vsock_insert_connected(vsock_sk(sk)); 421ae0078fcSDexuan Cui } 422ae0078fcSDexuan Cui 423ae0078fcSDexuan Cui sk->sk_state_change(sk); 424ae0078fcSDexuan Cui 425ae0078fcSDexuan Cui out: 426ae0078fcSDexuan Cui /* Release refcnt obtained when we called vsock_find_bound_socket() */ 427ae0078fcSDexuan Cui sock_put(sk); 428b4562ca7SDexuan Cui 429b4562ca7SDexuan Cui release_sock(sk); 430ae0078fcSDexuan Cui } 431ae0078fcSDexuan Cui 432ae0078fcSDexuan Cui static u32 hvs_get_local_cid(void) 433ae0078fcSDexuan Cui { 434ae0078fcSDexuan Cui return VMADDR_CID_ANY; 435ae0078fcSDexuan Cui } 436ae0078fcSDexuan Cui 437ae0078fcSDexuan Cui static int hvs_sock_init(struct vsock_sock *vsk, struct vsock_sock *psk) 438ae0078fcSDexuan Cui { 439ae0078fcSDexuan Cui struct hvsock *hvs; 440ac383f58SSunil Muthuswamy struct sock *sk = sk_vsock(vsk); 441ae0078fcSDexuan Cui 442ae0078fcSDexuan Cui hvs = kzalloc(sizeof(*hvs), GFP_KERNEL); 443ae0078fcSDexuan Cui if (!hvs) 444ae0078fcSDexuan Cui return -ENOMEM; 445ae0078fcSDexuan Cui 446ae0078fcSDexuan Cui vsk->trans = hvs; 447ae0078fcSDexuan Cui hvs->vsk = vsk; 448ac383f58SSunil Muthuswamy sk->sk_sndbuf = RINGBUFFER_HVS_SND_SIZE; 449ac383f58SSunil Muthuswamy sk->sk_rcvbuf = RINGBUFFER_HVS_RCV_SIZE; 450ae0078fcSDexuan Cui return 0; 451ae0078fcSDexuan Cui } 452ae0078fcSDexuan Cui 453ae0078fcSDexuan Cui static int hvs_connect(struct vsock_sock *vsk) 454ae0078fcSDexuan Cui { 455ae0078fcSDexuan Cui union hvs_service_id vm, host; 456ae0078fcSDexuan Cui struct hvsock *h = vsk->trans; 457ae0078fcSDexuan Cui 458ae0078fcSDexuan Cui vm.srv_id = srv_id_template; 459ae0078fcSDexuan Cui vm.svm_port = vsk->local_addr.svm_port; 460ae0078fcSDexuan Cui h->vm_srv_id = vm.srv_id; 461ae0078fcSDexuan Cui 462ae0078fcSDexuan Cui host.srv_id = srv_id_template; 463ae0078fcSDexuan Cui host.svm_port = vsk->remote_addr.svm_port; 464ae0078fcSDexuan Cui h->host_srv_id = host.srv_id; 465ae0078fcSDexuan Cui 466ae0078fcSDexuan Cui return vmbus_send_tl_connect_request(&h->vm_srv_id, &h->host_srv_id); 467ae0078fcSDexuan Cui } 468ae0078fcSDexuan Cui 469a9eeb998SSunil Muthuswamy static void hvs_shutdown_lock_held(struct hvsock *hvs, int mode) 470a9eeb998SSunil Muthuswamy { 471a9eeb998SSunil Muthuswamy struct vmpipe_proto_header hdr; 472a9eeb998SSunil Muthuswamy 473a9eeb998SSunil Muthuswamy if (hvs->fin_sent || !hvs->chan) 474a9eeb998SSunil Muthuswamy return; 475a9eeb998SSunil Muthuswamy 476a9eeb998SSunil Muthuswamy /* It can't fail: see hvs_channel_writable_bytes(). */ 477*c0e084e3SKees Cook (void)__hvs_send_data(hvs->chan, &hdr, 0); 478a9eeb998SSunil Muthuswamy hvs->fin_sent = true; 479a9eeb998SSunil Muthuswamy } 480a9eeb998SSunil Muthuswamy 481ae0078fcSDexuan Cui static int hvs_shutdown(struct vsock_sock *vsk, int mode) 482ae0078fcSDexuan Cui { 483ae0078fcSDexuan Cui if (!(mode & SEND_SHUTDOWN)) 484ae0078fcSDexuan Cui return 0; 485ae0078fcSDexuan Cui 486a9eeb998SSunil Muthuswamy hvs_shutdown_lock_held(vsk->trans, mode); 487ae0078fcSDexuan Cui return 0; 488ae0078fcSDexuan Cui } 489ae0078fcSDexuan Cui 490a9eeb998SSunil Muthuswamy static void hvs_close_timeout(struct work_struct *work) 491a9eeb998SSunil Muthuswamy { 492a9eeb998SSunil Muthuswamy struct vsock_sock *vsk = 493a9eeb998SSunil Muthuswamy container_of(work, struct vsock_sock, close_work.work); 494a9eeb998SSunil Muthuswamy struct sock *sk = sk_vsock(vsk); 495a9eeb998SSunil Muthuswamy 496a9eeb998SSunil Muthuswamy sock_hold(sk); 497a9eeb998SSunil Muthuswamy lock_sock(sk); 498a9eeb998SSunil Muthuswamy if (!sock_flag(sk, SOCK_DONE)) 499a9eeb998SSunil Muthuswamy hvs_do_close_lock_held(vsk, false); 500a9eeb998SSunil Muthuswamy 501a9eeb998SSunil Muthuswamy vsk->close_work_scheduled = false; 502a9eeb998SSunil Muthuswamy release_sock(sk); 503a9eeb998SSunil Muthuswamy sock_put(sk); 504a9eeb998SSunil Muthuswamy } 505a9eeb998SSunil Muthuswamy 506a9eeb998SSunil Muthuswamy /* Returns true, if it is safe to remove socket; false otherwise */ 507a9eeb998SSunil Muthuswamy static bool hvs_close_lock_held(struct vsock_sock *vsk) 508a9eeb998SSunil Muthuswamy { 509a9eeb998SSunil Muthuswamy struct sock *sk = sk_vsock(vsk); 510a9eeb998SSunil Muthuswamy 511a9eeb998SSunil Muthuswamy if (!(sk->sk_state == TCP_ESTABLISHED || 512a9eeb998SSunil Muthuswamy sk->sk_state == TCP_CLOSING)) 513a9eeb998SSunil Muthuswamy return true; 514a9eeb998SSunil Muthuswamy 515a9eeb998SSunil Muthuswamy if ((sk->sk_shutdown & SHUTDOWN_MASK) != SHUTDOWN_MASK) 516a9eeb998SSunil Muthuswamy hvs_shutdown_lock_held(vsk->trans, SHUTDOWN_MASK); 517a9eeb998SSunil Muthuswamy 518a9eeb998SSunil Muthuswamy if (sock_flag(sk, SOCK_DONE)) 519a9eeb998SSunil Muthuswamy return true; 520a9eeb998SSunil Muthuswamy 521a9eeb998SSunil Muthuswamy /* This reference will be dropped by the delayed close routine */ 522a9eeb998SSunil Muthuswamy sock_hold(sk); 523a9eeb998SSunil Muthuswamy INIT_DELAYED_WORK(&vsk->close_work, hvs_close_timeout); 524a9eeb998SSunil Muthuswamy vsk->close_work_scheduled = true; 525a9eeb998SSunil Muthuswamy schedule_delayed_work(&vsk->close_work, HVS_CLOSE_TIMEOUT); 526a9eeb998SSunil Muthuswamy return false; 527a9eeb998SSunil Muthuswamy } 528a9eeb998SSunil Muthuswamy 529ae0078fcSDexuan Cui static void hvs_release(struct vsock_sock *vsk) 530ae0078fcSDexuan Cui { 531a9eeb998SSunil Muthuswamy bool remove_sock; 532ae0078fcSDexuan Cui 533a9eeb998SSunil Muthuswamy remove_sock = hvs_close_lock_held(vsk); 534a9eeb998SSunil Muthuswamy if (remove_sock) 535a9eeb998SSunil Muthuswamy vsock_remove_sock(vsk); 536ae0078fcSDexuan Cui } 537ae0078fcSDexuan Cui 538ae0078fcSDexuan Cui static void hvs_destruct(struct vsock_sock *vsk) 539ae0078fcSDexuan Cui { 540ae0078fcSDexuan Cui struct hvsock *hvs = vsk->trans; 541ae0078fcSDexuan Cui struct vmbus_channel *chan = hvs->chan; 542ae0078fcSDexuan Cui 543ae0078fcSDexuan Cui if (chan) 544ae0078fcSDexuan Cui vmbus_hvsock_device_unregister(chan); 545ae0078fcSDexuan Cui 546ae0078fcSDexuan Cui kfree(hvs); 547ae0078fcSDexuan Cui } 548ae0078fcSDexuan Cui 549ae0078fcSDexuan Cui static int hvs_dgram_bind(struct vsock_sock *vsk, struct sockaddr_vm *addr) 550ae0078fcSDexuan Cui { 551ae0078fcSDexuan Cui return -EOPNOTSUPP; 552ae0078fcSDexuan Cui } 553ae0078fcSDexuan Cui 554ae0078fcSDexuan Cui static int hvs_dgram_dequeue(struct vsock_sock *vsk, struct msghdr *msg, 555ae0078fcSDexuan Cui size_t len, int flags) 556ae0078fcSDexuan Cui { 557ae0078fcSDexuan Cui return -EOPNOTSUPP; 558ae0078fcSDexuan Cui } 559ae0078fcSDexuan Cui 560ae0078fcSDexuan Cui static int hvs_dgram_enqueue(struct vsock_sock *vsk, 561ae0078fcSDexuan Cui struct sockaddr_vm *remote, struct msghdr *msg, 562ae0078fcSDexuan Cui size_t dgram_len) 563ae0078fcSDexuan Cui { 564ae0078fcSDexuan Cui return -EOPNOTSUPP; 565ae0078fcSDexuan Cui } 566ae0078fcSDexuan Cui 567ae0078fcSDexuan Cui static bool hvs_dgram_allow(u32 cid, u32 port) 568ae0078fcSDexuan Cui { 569ae0078fcSDexuan Cui return false; 570ae0078fcSDexuan Cui } 571ae0078fcSDexuan Cui 572ae0078fcSDexuan Cui static int hvs_update_recv_data(struct hvsock *hvs) 573ae0078fcSDexuan Cui { 574ae0078fcSDexuan Cui struct hvs_recv_buf *recv_buf; 575ae0078fcSDexuan Cui u32 payload_len; 576ae0078fcSDexuan Cui 577ae0078fcSDexuan Cui recv_buf = (struct hvs_recv_buf *)(hvs->recv_desc + 1); 578ae0078fcSDexuan Cui payload_len = recv_buf->hdr.data_size; 579ae0078fcSDexuan Cui 580ae0078fcSDexuan Cui if (payload_len > HVS_MTU_SIZE) 581ae0078fcSDexuan Cui return -EIO; 582ae0078fcSDexuan Cui 583ae0078fcSDexuan Cui if (payload_len == 0) 584ae0078fcSDexuan Cui hvs->vsk->peer_shutdown |= SEND_SHUTDOWN; 585ae0078fcSDexuan Cui 586ae0078fcSDexuan Cui hvs->recv_data_len = payload_len; 587ae0078fcSDexuan Cui hvs->recv_data_off = 0; 588ae0078fcSDexuan Cui 589ae0078fcSDexuan Cui return 0; 590ae0078fcSDexuan Cui } 591ae0078fcSDexuan Cui 592ae0078fcSDexuan Cui static ssize_t hvs_stream_dequeue(struct vsock_sock *vsk, struct msghdr *msg, 593ae0078fcSDexuan Cui size_t len, int flags) 594ae0078fcSDexuan Cui { 595ae0078fcSDexuan Cui struct hvsock *hvs = vsk->trans; 596ae0078fcSDexuan Cui bool need_refill = !hvs->recv_desc; 597ae0078fcSDexuan Cui struct hvs_recv_buf *recv_buf; 598ae0078fcSDexuan Cui u32 to_read; 599ae0078fcSDexuan Cui int ret; 600ae0078fcSDexuan Cui 601ae0078fcSDexuan Cui if (flags & MSG_PEEK) 602ae0078fcSDexuan Cui return -EOPNOTSUPP; 603ae0078fcSDexuan Cui 604ae0078fcSDexuan Cui if (need_refill) { 605adae1e93SAndres Beltran hvs->recv_desc = hv_pkt_iter_first_raw(hvs->chan); 606ae0078fcSDexuan Cui ret = hvs_update_recv_data(hvs); 607ae0078fcSDexuan Cui if (ret) 608ae0078fcSDexuan Cui return ret; 609ae0078fcSDexuan Cui } 610ae0078fcSDexuan Cui 611ae0078fcSDexuan Cui recv_buf = (struct hvs_recv_buf *)(hvs->recv_desc + 1); 612ae0078fcSDexuan Cui to_read = min_t(u32, len, hvs->recv_data_len); 613ae0078fcSDexuan Cui ret = memcpy_to_msg(msg, recv_buf->data + hvs->recv_data_off, to_read); 614ae0078fcSDexuan Cui if (ret != 0) 615ae0078fcSDexuan Cui return ret; 616ae0078fcSDexuan Cui 617ae0078fcSDexuan Cui hvs->recv_data_len -= to_read; 618ae0078fcSDexuan Cui if (hvs->recv_data_len == 0) { 619adae1e93SAndres Beltran hvs->recv_desc = hv_pkt_iter_next_raw(hvs->chan, hvs->recv_desc); 620ae0078fcSDexuan Cui if (hvs->recv_desc) { 621ae0078fcSDexuan Cui ret = hvs_update_recv_data(hvs); 622ae0078fcSDexuan Cui if (ret) 623ae0078fcSDexuan Cui return ret; 624ae0078fcSDexuan Cui } 625ae0078fcSDexuan Cui } else { 626ae0078fcSDexuan Cui hvs->recv_data_off += to_read; 627ae0078fcSDexuan Cui } 628ae0078fcSDexuan Cui 629ae0078fcSDexuan Cui return to_read; 630ae0078fcSDexuan Cui } 631ae0078fcSDexuan Cui 632ae0078fcSDexuan Cui static ssize_t hvs_stream_enqueue(struct vsock_sock *vsk, struct msghdr *msg, 633ae0078fcSDexuan Cui size_t len) 634ae0078fcSDexuan Cui { 635ae0078fcSDexuan Cui struct hvsock *hvs = vsk->trans; 636ae0078fcSDexuan Cui struct vmbus_channel *chan = hvs->chan; 637ae0078fcSDexuan Cui struct hvs_send_buf *send_buf; 63814a1eaa8SSunil Muthuswamy ssize_t to_write, max_writable; 63914a1eaa8SSunil Muthuswamy ssize_t ret = 0; 64014a1eaa8SSunil Muthuswamy ssize_t bytes_written = 0; 641ae0078fcSDexuan Cui 64277ffe333SHimadri Pandya BUILD_BUG_ON(sizeof(*send_buf) != HV_HYP_PAGE_SIZE); 643ae0078fcSDexuan Cui 644ae0078fcSDexuan Cui send_buf = kmalloc(sizeof(*send_buf), GFP_KERNEL); 645ae0078fcSDexuan Cui if (!send_buf) 646ae0078fcSDexuan Cui return -ENOMEM; 647ae0078fcSDexuan Cui 64814a1eaa8SSunil Muthuswamy /* Reader(s) could be draining data from the channel as we write. 64914a1eaa8SSunil Muthuswamy * Maximize bandwidth, by iterating until the channel is found to be 65014a1eaa8SSunil Muthuswamy * full. 65114a1eaa8SSunil Muthuswamy */ 65214a1eaa8SSunil Muthuswamy while (len) { 653ae0078fcSDexuan Cui max_writable = hvs_channel_writable_bytes(chan); 65414a1eaa8SSunil Muthuswamy if (!max_writable) 65514a1eaa8SSunil Muthuswamy break; 656ae0078fcSDexuan Cui to_write = min_t(ssize_t, len, max_writable); 657ae0078fcSDexuan Cui to_write = min_t(ssize_t, to_write, HVS_SEND_BUF_SIZE); 65814a1eaa8SSunil Muthuswamy /* memcpy_from_msg is safe for loop as it advances the offsets 65914a1eaa8SSunil Muthuswamy * within the message iterator. 66014a1eaa8SSunil Muthuswamy */ 661ae0078fcSDexuan Cui ret = memcpy_from_msg(send_buf->data, msg, to_write); 662ae0078fcSDexuan Cui if (ret < 0) 663ae0078fcSDexuan Cui goto out; 664ae0078fcSDexuan Cui 665ae0078fcSDexuan Cui ret = hvs_send_data(hvs->chan, send_buf, to_write); 666ae0078fcSDexuan Cui if (ret < 0) 667ae0078fcSDexuan Cui goto out; 668ae0078fcSDexuan Cui 66914a1eaa8SSunil Muthuswamy bytes_written += to_write; 67014a1eaa8SSunil Muthuswamy len -= to_write; 67114a1eaa8SSunil Muthuswamy } 672ae0078fcSDexuan Cui out: 67314a1eaa8SSunil Muthuswamy /* If any data has been sent, return that */ 67414a1eaa8SSunil Muthuswamy if (bytes_written) 67514a1eaa8SSunil Muthuswamy ret = bytes_written; 676ae0078fcSDexuan Cui kfree(send_buf); 677ae0078fcSDexuan Cui return ret; 678ae0078fcSDexuan Cui } 679ae0078fcSDexuan Cui 680ae0078fcSDexuan Cui static s64 hvs_stream_has_data(struct vsock_sock *vsk) 681ae0078fcSDexuan Cui { 682ae0078fcSDexuan Cui struct hvsock *hvs = vsk->trans; 683ae0078fcSDexuan Cui s64 ret; 684ae0078fcSDexuan Cui 685ae0078fcSDexuan Cui if (hvs->recv_data_len > 0) 686ae0078fcSDexuan Cui return 1; 687ae0078fcSDexuan Cui 688ae0078fcSDexuan Cui switch (hvs_channel_readable_payload(hvs->chan)) { 689ae0078fcSDexuan Cui case 1: 690ae0078fcSDexuan Cui ret = 1; 691ae0078fcSDexuan Cui break; 692ae0078fcSDexuan Cui case 0: 693ae0078fcSDexuan Cui vsk->peer_shutdown |= SEND_SHUTDOWN; 694ae0078fcSDexuan Cui ret = 0; 695ae0078fcSDexuan Cui break; 696ae0078fcSDexuan Cui default: /* -1 */ 697ae0078fcSDexuan Cui ret = 0; 698ae0078fcSDexuan Cui break; 699ae0078fcSDexuan Cui } 700ae0078fcSDexuan Cui 701ae0078fcSDexuan Cui return ret; 702ae0078fcSDexuan Cui } 703ae0078fcSDexuan Cui 704ae0078fcSDexuan Cui static s64 hvs_stream_has_space(struct vsock_sock *vsk) 705ae0078fcSDexuan Cui { 706ae0078fcSDexuan Cui struct hvsock *hvs = vsk->trans; 707ae0078fcSDexuan Cui 708cb359b60SSunil Muthuswamy return hvs_channel_writable_bytes(hvs->chan); 709ae0078fcSDexuan Cui } 710ae0078fcSDexuan Cui 711ae0078fcSDexuan Cui static u64 hvs_stream_rcvhiwat(struct vsock_sock *vsk) 712ae0078fcSDexuan Cui { 713ae0078fcSDexuan Cui return HVS_MTU_SIZE + 1; 714ae0078fcSDexuan Cui } 715ae0078fcSDexuan Cui 716ae0078fcSDexuan Cui static bool hvs_stream_is_active(struct vsock_sock *vsk) 717ae0078fcSDexuan Cui { 718ae0078fcSDexuan Cui struct hvsock *hvs = vsk->trans; 719ae0078fcSDexuan Cui 720ae0078fcSDexuan Cui return hvs->chan != NULL; 721ae0078fcSDexuan Cui } 722ae0078fcSDexuan Cui 723ae0078fcSDexuan Cui static bool hvs_stream_allow(u32 cid, u32 port) 724ae0078fcSDexuan Cui { 725ae0078fcSDexuan Cui if (cid == VMADDR_CID_HOST) 726ae0078fcSDexuan Cui return true; 727ae0078fcSDexuan Cui 728ae0078fcSDexuan Cui return false; 729ae0078fcSDexuan Cui } 730ae0078fcSDexuan Cui 731ae0078fcSDexuan Cui static 732ae0078fcSDexuan Cui int hvs_notify_poll_in(struct vsock_sock *vsk, size_t target, bool *readable) 733ae0078fcSDexuan Cui { 734ae0078fcSDexuan Cui struct hvsock *hvs = vsk->trans; 735ae0078fcSDexuan Cui 736ae0078fcSDexuan Cui *readable = hvs_channel_readable(hvs->chan); 737ae0078fcSDexuan Cui return 0; 738ae0078fcSDexuan Cui } 739ae0078fcSDexuan Cui 740ae0078fcSDexuan Cui static 741ae0078fcSDexuan Cui int hvs_notify_poll_out(struct vsock_sock *vsk, size_t target, bool *writable) 742ae0078fcSDexuan Cui { 743ae0078fcSDexuan Cui *writable = hvs_stream_has_space(vsk) > 0; 744ae0078fcSDexuan Cui 745ae0078fcSDexuan Cui return 0; 746ae0078fcSDexuan Cui } 747ae0078fcSDexuan Cui 748ae0078fcSDexuan Cui static 749ae0078fcSDexuan Cui int hvs_notify_recv_init(struct vsock_sock *vsk, size_t target, 750ae0078fcSDexuan Cui struct vsock_transport_recv_notify_data *d) 751ae0078fcSDexuan Cui { 752ae0078fcSDexuan Cui return 0; 753ae0078fcSDexuan Cui } 754ae0078fcSDexuan Cui 755ae0078fcSDexuan Cui static 756ae0078fcSDexuan Cui int hvs_notify_recv_pre_block(struct vsock_sock *vsk, size_t target, 757ae0078fcSDexuan Cui struct vsock_transport_recv_notify_data *d) 758ae0078fcSDexuan Cui { 759ae0078fcSDexuan Cui return 0; 760ae0078fcSDexuan Cui } 761ae0078fcSDexuan Cui 762ae0078fcSDexuan Cui static 763ae0078fcSDexuan Cui int hvs_notify_recv_pre_dequeue(struct vsock_sock *vsk, size_t target, 764ae0078fcSDexuan Cui struct vsock_transport_recv_notify_data *d) 765ae0078fcSDexuan Cui { 766ae0078fcSDexuan Cui return 0; 767ae0078fcSDexuan Cui } 768ae0078fcSDexuan Cui 769ae0078fcSDexuan Cui static 770ae0078fcSDexuan Cui int hvs_notify_recv_post_dequeue(struct vsock_sock *vsk, size_t target, 771ae0078fcSDexuan Cui ssize_t copied, bool data_read, 772ae0078fcSDexuan Cui struct vsock_transport_recv_notify_data *d) 773ae0078fcSDexuan Cui { 774ae0078fcSDexuan Cui return 0; 775ae0078fcSDexuan Cui } 776ae0078fcSDexuan Cui 777ae0078fcSDexuan Cui static 778ae0078fcSDexuan Cui int hvs_notify_send_init(struct vsock_sock *vsk, 779ae0078fcSDexuan Cui struct vsock_transport_send_notify_data *d) 780ae0078fcSDexuan Cui { 781ae0078fcSDexuan Cui return 0; 782ae0078fcSDexuan Cui } 783ae0078fcSDexuan Cui 784ae0078fcSDexuan Cui static 785ae0078fcSDexuan Cui int hvs_notify_send_pre_block(struct vsock_sock *vsk, 786ae0078fcSDexuan Cui struct vsock_transport_send_notify_data *d) 787ae0078fcSDexuan Cui { 788ae0078fcSDexuan Cui return 0; 789ae0078fcSDexuan Cui } 790ae0078fcSDexuan Cui 791ae0078fcSDexuan Cui static 792ae0078fcSDexuan Cui int hvs_notify_send_pre_enqueue(struct vsock_sock *vsk, 793ae0078fcSDexuan Cui struct vsock_transport_send_notify_data *d) 794ae0078fcSDexuan Cui { 795ae0078fcSDexuan Cui return 0; 796ae0078fcSDexuan Cui } 797ae0078fcSDexuan Cui 798ae0078fcSDexuan Cui static 799ae0078fcSDexuan Cui int hvs_notify_send_post_enqueue(struct vsock_sock *vsk, ssize_t written, 800ae0078fcSDexuan Cui struct vsock_transport_send_notify_data *d) 801ae0078fcSDexuan Cui { 802ae0078fcSDexuan Cui return 0; 803ae0078fcSDexuan Cui } 804ae0078fcSDexuan Cui 805ae0078fcSDexuan Cui static struct vsock_transport hvs_transport = { 8066a2c0962SStefano Garzarella .module = THIS_MODULE, 8076a2c0962SStefano Garzarella 808ae0078fcSDexuan Cui .get_local_cid = hvs_get_local_cid, 809ae0078fcSDexuan Cui 810ae0078fcSDexuan Cui .init = hvs_sock_init, 811ae0078fcSDexuan Cui .destruct = hvs_destruct, 812ae0078fcSDexuan Cui .release = hvs_release, 813ae0078fcSDexuan Cui .connect = hvs_connect, 814ae0078fcSDexuan Cui .shutdown = hvs_shutdown, 815ae0078fcSDexuan Cui 816ae0078fcSDexuan Cui .dgram_bind = hvs_dgram_bind, 817ae0078fcSDexuan Cui .dgram_dequeue = hvs_dgram_dequeue, 818ae0078fcSDexuan Cui .dgram_enqueue = hvs_dgram_enqueue, 819ae0078fcSDexuan Cui .dgram_allow = hvs_dgram_allow, 820ae0078fcSDexuan Cui 821ae0078fcSDexuan Cui .stream_dequeue = hvs_stream_dequeue, 822ae0078fcSDexuan Cui .stream_enqueue = hvs_stream_enqueue, 823ae0078fcSDexuan Cui .stream_has_data = hvs_stream_has_data, 824ae0078fcSDexuan Cui .stream_has_space = hvs_stream_has_space, 825ae0078fcSDexuan Cui .stream_rcvhiwat = hvs_stream_rcvhiwat, 826ae0078fcSDexuan Cui .stream_is_active = hvs_stream_is_active, 827ae0078fcSDexuan Cui .stream_allow = hvs_stream_allow, 828ae0078fcSDexuan Cui 829ae0078fcSDexuan Cui .notify_poll_in = hvs_notify_poll_in, 830ae0078fcSDexuan Cui .notify_poll_out = hvs_notify_poll_out, 831ae0078fcSDexuan Cui .notify_recv_init = hvs_notify_recv_init, 832ae0078fcSDexuan Cui .notify_recv_pre_block = hvs_notify_recv_pre_block, 833ae0078fcSDexuan Cui .notify_recv_pre_dequeue = hvs_notify_recv_pre_dequeue, 834ae0078fcSDexuan Cui .notify_recv_post_dequeue = hvs_notify_recv_post_dequeue, 835ae0078fcSDexuan Cui .notify_send_init = hvs_notify_send_init, 836ae0078fcSDexuan Cui .notify_send_pre_block = hvs_notify_send_pre_block, 837ae0078fcSDexuan Cui .notify_send_pre_enqueue = hvs_notify_send_pre_enqueue, 838ae0078fcSDexuan Cui .notify_send_post_enqueue = hvs_notify_send_post_enqueue, 839ae0078fcSDexuan Cui 840ae0078fcSDexuan Cui }; 841ae0078fcSDexuan Cui 842c0cfa2d8SStefano Garzarella static bool hvs_check_transport(struct vsock_sock *vsk) 843c0cfa2d8SStefano Garzarella { 844c0cfa2d8SStefano Garzarella return vsk->transport == &hvs_transport; 845c0cfa2d8SStefano Garzarella } 846c0cfa2d8SStefano Garzarella 847ae0078fcSDexuan Cui static int hvs_probe(struct hv_device *hdev, 848ae0078fcSDexuan Cui const struct hv_vmbus_device_id *dev_id) 849ae0078fcSDexuan Cui { 850ae0078fcSDexuan Cui struct vmbus_channel *chan = hdev->channel; 851ae0078fcSDexuan Cui 852ae0078fcSDexuan Cui hvs_open_connection(chan); 853ae0078fcSDexuan Cui 854ae0078fcSDexuan Cui /* Always return success to suppress the unnecessary error message 855ae0078fcSDexuan Cui * in vmbus_probe(): on error the host will rescind the device in 856ae0078fcSDexuan Cui * 30 seconds and we can do cleanup at that time in 857ae0078fcSDexuan Cui * vmbus_onoffer_rescind(). 858ae0078fcSDexuan Cui */ 859ae0078fcSDexuan Cui return 0; 860ae0078fcSDexuan Cui } 861ae0078fcSDexuan Cui 862ae0078fcSDexuan Cui static int hvs_remove(struct hv_device *hdev) 863ae0078fcSDexuan Cui { 864ae0078fcSDexuan Cui struct vmbus_channel *chan = hdev->channel; 865ae0078fcSDexuan Cui 866ae0078fcSDexuan Cui vmbus_close(chan); 867ae0078fcSDexuan Cui 868ae0078fcSDexuan Cui return 0; 869ae0078fcSDexuan Cui } 870ae0078fcSDexuan Cui 8712194c2ebSDexuan Cui /* hv_sock connections can not persist across hibernation, and all the hv_sock 8722194c2ebSDexuan Cui * channels are forced to be rescinded before hibernation: see 8732194c2ebSDexuan Cui * vmbus_bus_suspend(). Here the dummy hvs_suspend() and hvs_resume() 8742194c2ebSDexuan Cui * are only needed because hibernation requires that every vmbus device's 8752194c2ebSDexuan Cui * driver should have a .suspend and .resume callback: see vmbus_suspend(). 8762194c2ebSDexuan Cui */ 8772194c2ebSDexuan Cui static int hvs_suspend(struct hv_device *hv_dev) 8782194c2ebSDexuan Cui { 8792194c2ebSDexuan Cui /* Dummy */ 8802194c2ebSDexuan Cui return 0; 8812194c2ebSDexuan Cui } 8822194c2ebSDexuan Cui 8832194c2ebSDexuan Cui static int hvs_resume(struct hv_device *dev) 8842194c2ebSDexuan Cui { 8852194c2ebSDexuan Cui /* Dummy */ 8862194c2ebSDexuan Cui return 0; 8872194c2ebSDexuan Cui } 8882194c2ebSDexuan Cui 889ae0078fcSDexuan Cui /* This isn't really used. See vmbus_match() and vmbus_probe() */ 890ae0078fcSDexuan Cui static const struct hv_vmbus_device_id id_table[] = { 891ae0078fcSDexuan Cui {}, 892ae0078fcSDexuan Cui }; 893ae0078fcSDexuan Cui 894ae0078fcSDexuan Cui static struct hv_driver hvs_drv = { 895ae0078fcSDexuan Cui .name = "hv_sock", 896ae0078fcSDexuan Cui .hvsock = true, 897ae0078fcSDexuan Cui .id_table = id_table, 898ae0078fcSDexuan Cui .probe = hvs_probe, 899ae0078fcSDexuan Cui .remove = hvs_remove, 9002194c2ebSDexuan Cui .suspend = hvs_suspend, 9012194c2ebSDexuan Cui .resume = hvs_resume, 902ae0078fcSDexuan Cui }; 903ae0078fcSDexuan Cui 904ae0078fcSDexuan Cui static int __init hvs_init(void) 905ae0078fcSDexuan Cui { 906ae0078fcSDexuan Cui int ret; 907ae0078fcSDexuan Cui 908ae0078fcSDexuan Cui if (vmbus_proto_version < VERSION_WIN10) 909ae0078fcSDexuan Cui return -ENODEV; 910ae0078fcSDexuan Cui 911ae0078fcSDexuan Cui ret = vmbus_driver_register(&hvs_drv); 912ae0078fcSDexuan Cui if (ret != 0) 913ae0078fcSDexuan Cui return ret; 914ae0078fcSDexuan Cui 915c0cfa2d8SStefano Garzarella ret = vsock_core_register(&hvs_transport, VSOCK_TRANSPORT_F_G2H); 916ae0078fcSDexuan Cui if (ret) { 917ae0078fcSDexuan Cui vmbus_driver_unregister(&hvs_drv); 918ae0078fcSDexuan Cui return ret; 919ae0078fcSDexuan Cui } 920ae0078fcSDexuan Cui 921ae0078fcSDexuan Cui return 0; 922ae0078fcSDexuan Cui } 923ae0078fcSDexuan Cui 924ae0078fcSDexuan Cui static void __exit hvs_exit(void) 925ae0078fcSDexuan Cui { 926c0cfa2d8SStefano Garzarella vsock_core_unregister(&hvs_transport); 927ae0078fcSDexuan Cui vmbus_driver_unregister(&hvs_drv); 928ae0078fcSDexuan Cui } 929ae0078fcSDexuan Cui 930ae0078fcSDexuan Cui module_init(hvs_init); 931ae0078fcSDexuan Cui module_exit(hvs_exit); 932ae0078fcSDexuan Cui 933ae0078fcSDexuan Cui MODULE_DESCRIPTION("Hyper-V Sockets"); 934ae0078fcSDexuan Cui MODULE_VERSION("1.0.0"); 935ae0078fcSDexuan Cui MODULE_LICENSE("GPL"); 936ae0078fcSDexuan Cui MODULE_ALIAS_NETPROTO(PF_VSOCK); 937