Vladislav Yaroshchuk | 81ad296 | 2022-03-17 20:28:34 +0300 | [diff] [blame] | 1 | /* |
| 2 | * vmnet-common.m - network client wrapper for Apple vmnet.framework |
| 3 | * |
| 4 | * Copyright(c) 2022 Vladislav Yaroshchuk <vladislav.yaroshchuk@jetbrains.com> |
| 5 | * Copyright(c) 2021 Phillip Tennen <phillip@axleos.com> |
| 6 | * |
| 7 | * This work is licensed under the terms of the GNU GPL, version 2 or later. |
| 8 | * See the COPYING file in the top-level directory. |
| 9 | * |
| 10 | */ |
| 11 | |
| 12 | #include "qemu/osdep.h" |
Vladislav Yaroshchuk | 73f99db | 2022-03-17 20:28:35 +0300 | [diff] [blame] | 13 | #include "qemu/main-loop.h" |
| 14 | #include "qemu/log.h" |
Vladislav Yaroshchuk | 81ad296 | 2022-03-17 20:28:34 +0300 | [diff] [blame] | 15 | #include "qapi/qapi-types-net.h" |
| 16 | #include "vmnet_int.h" |
| 17 | #include "clients.h" |
| 18 | #include "qemu/error-report.h" |
| 19 | #include "qapi/error.h" |
Joelle van Dyne | 993f71e | 2023-01-01 17:08:21 -0800 | [diff] [blame] | 20 | #include "sysemu/runstate.h" |
Vladislav Yaroshchuk | 81ad296 | 2022-03-17 20:28:34 +0300 | [diff] [blame] | 21 | |
| 22 | #include <vmnet/vmnet.h> |
Vladislav Yaroshchuk | 73f99db | 2022-03-17 20:28:35 +0300 | [diff] [blame] | 23 | #include <dispatch/dispatch.h> |
| 24 | |
| 25 | |
| 26 | static void vmnet_send_completed(NetClientState *nc, ssize_t len); |
| 27 | |
| 28 | |
| 29 | const char *vmnet_status_map_str(vmnet_return_t status) |
| 30 | { |
| 31 | switch (status) { |
| 32 | case VMNET_SUCCESS: |
| 33 | return "success"; |
| 34 | case VMNET_FAILURE: |
| 35 | return "general failure (possibly not enough privileges)"; |
| 36 | case VMNET_MEM_FAILURE: |
| 37 | return "memory allocation failure"; |
| 38 | case VMNET_INVALID_ARGUMENT: |
| 39 | return "invalid argument specified"; |
| 40 | case VMNET_SETUP_INCOMPLETE: |
| 41 | return "interface setup is not complete"; |
| 42 | case VMNET_INVALID_ACCESS: |
| 43 | return "invalid access, permission denied"; |
| 44 | case VMNET_PACKET_TOO_BIG: |
| 45 | return "packet size is larger than MTU"; |
| 46 | case VMNET_BUFFER_EXHAUSTED: |
| 47 | return "buffers exhausted in kernel"; |
| 48 | case VMNET_TOO_MANY_PACKETS: |
| 49 | return "packet count exceeds limit"; |
| 50 | #if defined(MAC_OS_VERSION_11_0) && \ |
| 51 | MAC_OS_X_VERSION_MIN_REQUIRED >= MAC_OS_VERSION_11_0 |
| 52 | case VMNET_SHARING_SERVICE_BUSY: |
| 53 | return "conflict, sharing service is in use"; |
| 54 | #endif |
| 55 | default: |
| 56 | return "unknown vmnet error"; |
| 57 | } |
| 58 | } |
| 59 | |
| 60 | |
| 61 | /** |
| 62 | * Write packets from QEMU to vmnet interface. |
| 63 | * |
| 64 | * vmnet.framework supports iov, but writing more than |
| 65 | * one iov into vmnet interface fails with |
| 66 | * 'VMNET_INVALID_ARGUMENT'. Collecting provided iovs into |
| 67 | * one and passing it to vmnet works fine. That's the |
| 68 | * reason why receive_iov() left unimplemented. But it still |
| 69 | * works with good performance having .receive() only. |
| 70 | */ |
| 71 | ssize_t vmnet_receive_common(NetClientState *nc, |
| 72 | const uint8_t *buf, |
| 73 | size_t size) |
| 74 | { |
| 75 | VmnetState *s = DO_UPCAST(VmnetState, nc, nc); |
| 76 | struct vmpktdesc packet; |
| 77 | struct iovec iov; |
| 78 | int pkt_cnt; |
| 79 | vmnet_return_t if_status; |
| 80 | |
| 81 | if (size > s->max_packet_size) { |
| 82 | warn_report("vmnet: packet is too big, %zu > %" PRIu64, |
| 83 | packet.vm_pkt_size, |
| 84 | s->max_packet_size); |
| 85 | return -1; |
| 86 | } |
| 87 | |
| 88 | iov.iov_base = (char *) buf; |
| 89 | iov.iov_len = size; |
| 90 | |
| 91 | packet.vm_pkt_iovcnt = 1; |
| 92 | packet.vm_flags = 0; |
| 93 | packet.vm_pkt_size = size; |
| 94 | packet.vm_pkt_iov = &iov; |
| 95 | pkt_cnt = 1; |
| 96 | |
| 97 | if_status = vmnet_write(s->vmnet_if, &packet, &pkt_cnt); |
| 98 | if (if_status != VMNET_SUCCESS) { |
| 99 | error_report("vmnet: write error: %s\n", |
| 100 | vmnet_status_map_str(if_status)); |
| 101 | return -1; |
| 102 | } |
| 103 | |
| 104 | if (pkt_cnt) { |
| 105 | return size; |
| 106 | } |
| 107 | return 0; |
| 108 | } |
| 109 | |
| 110 | |
| 111 | /** |
| 112 | * Read packets from vmnet interface and write them |
| 113 | * to temporary buffers in VmnetState. |
| 114 | * |
| 115 | * Returns read packets number (may be 0) on success, |
| 116 | * -1 on error |
| 117 | */ |
| 118 | static int vmnet_read_packets(VmnetState *s) |
| 119 | { |
| 120 | assert(s->packets_send_current_pos == s->packets_send_end_pos); |
| 121 | |
| 122 | struct vmpktdesc *packets = s->packets_buf; |
| 123 | vmnet_return_t status; |
| 124 | int i; |
| 125 | |
| 126 | /* Read as many packets as present */ |
| 127 | s->packets_send_current_pos = 0; |
| 128 | s->packets_send_end_pos = VMNET_PACKETS_LIMIT; |
| 129 | for (i = 0; i < s->packets_send_end_pos; ++i) { |
| 130 | packets[i].vm_pkt_size = s->max_packet_size; |
| 131 | packets[i].vm_pkt_iovcnt = 1; |
| 132 | packets[i].vm_flags = 0; |
| 133 | } |
| 134 | |
| 135 | status = vmnet_read(s->vmnet_if, packets, &s->packets_send_end_pos); |
| 136 | if (status != VMNET_SUCCESS) { |
| 137 | error_printf("vmnet: read failed: %s\n", |
| 138 | vmnet_status_map_str(status)); |
| 139 | s->packets_send_current_pos = 0; |
| 140 | s->packets_send_end_pos = 0; |
| 141 | return -1; |
| 142 | } |
| 143 | return s->packets_send_end_pos; |
| 144 | } |
| 145 | |
| 146 | |
| 147 | /** |
| 148 | * Write packets from temporary buffers in VmnetState |
| 149 | * to QEMU. |
| 150 | */ |
| 151 | static void vmnet_write_packets_to_qemu(VmnetState *s) |
| 152 | { |
| 153 | while (s->packets_send_current_pos < s->packets_send_end_pos) { |
| 154 | ssize_t size = qemu_send_packet_async(&s->nc, |
| 155 | s->iov_buf[s->packets_send_current_pos].iov_base, |
| 156 | s->packets_buf[s->packets_send_current_pos].vm_pkt_size, |
| 157 | vmnet_send_completed); |
| 158 | |
| 159 | if (size == 0) { |
| 160 | /* QEMU is not ready to consume more packets - |
| 161 | * stop and wait for completion callback call */ |
| 162 | return; |
| 163 | } |
| 164 | ++s->packets_send_current_pos; |
| 165 | } |
| 166 | } |
| 167 | |
| 168 | |
| 169 | /** |
| 170 | * Bottom half callback that transfers packets from vmnet interface |
| 171 | * to QEMU. |
| 172 | * |
| 173 | * The process of transferring packets is three-staged: |
| 174 | * 1. Handle vmnet event; |
| 175 | * 2. Read packets from vmnet interface into temporary buffer; |
| 176 | * 3. Write packets from temporary buffer to QEMU. |
| 177 | * |
| 178 | * QEMU may suspend this process on the last stage, returning 0 from |
| 179 | * qemu_send_packet_async function. If this happens, we should |
| 180 | * respectfully wait until it is ready to consume more packets, |
| 181 | * write left ones in temporary buffer and only after this |
| 182 | * continue reading more packets from vmnet interface. |
| 183 | * |
| 184 | * Packets to be transferred are stored into packets_buf, |
| 185 | * in the window [packets_send_current_pos..packets_send_end_pos) |
| 186 | * including current_pos, excluding end_pos. |
| 187 | * |
| 188 | * Thus, if QEMU is not ready, buffer is not read and |
| 189 | * packets_send_current_pos < packets_send_end_pos. |
| 190 | */ |
| 191 | static void vmnet_send_bh(void *opaque) |
| 192 | { |
| 193 | NetClientState *nc = (NetClientState *) opaque; |
| 194 | VmnetState *s = DO_UPCAST(VmnetState, nc, nc); |
| 195 | |
| 196 | /* |
| 197 | * Do nothing if QEMU is not ready - wait |
| 198 | * for completion callback invocation |
| 199 | */ |
| 200 | if (s->packets_send_current_pos < s->packets_send_end_pos) { |
| 201 | return; |
| 202 | } |
| 203 | |
| 204 | /* Read packets from vmnet interface */ |
| 205 | if (vmnet_read_packets(s) > 0) { |
| 206 | /* Send them to QEMU */ |
| 207 | vmnet_write_packets_to_qemu(s); |
| 208 | } |
| 209 | } |
| 210 | |
| 211 | |
| 212 | /** |
| 213 | * Completion callback to be invoked by QEMU when it becomes |
| 214 | * ready to consume more packets. |
| 215 | */ |
| 216 | static void vmnet_send_completed(NetClientState *nc, ssize_t len) |
| 217 | { |
| 218 | VmnetState *s = DO_UPCAST(VmnetState, nc, nc); |
| 219 | |
| 220 | /* Callback is invoked eq queued packet is sent */ |
| 221 | ++s->packets_send_current_pos; |
| 222 | |
| 223 | /* Complete sending packets left in VmnetState buffers */ |
| 224 | vmnet_write_packets_to_qemu(s); |
| 225 | |
| 226 | /* And read new ones from vmnet if VmnetState buffer is ready */ |
| 227 | if (s->packets_send_current_pos < s->packets_send_end_pos) { |
| 228 | qemu_bh_schedule(s->send_bh); |
| 229 | } |
| 230 | } |
| 231 | |
| 232 | |
| 233 | static void vmnet_bufs_init(VmnetState *s) |
| 234 | { |
| 235 | struct vmpktdesc *packets = s->packets_buf; |
| 236 | struct iovec *iov = s->iov_buf; |
| 237 | int i; |
| 238 | |
| 239 | for (i = 0; i < VMNET_PACKETS_LIMIT; ++i) { |
| 240 | iov[i].iov_len = s->max_packet_size; |
| 241 | iov[i].iov_base = g_malloc0(iov[i].iov_len); |
| 242 | packets[i].vm_pkt_iov = iov + i; |
| 243 | } |
| 244 | } |
| 245 | |
Joelle van Dyne | 993f71e | 2023-01-01 17:08:21 -0800 | [diff] [blame] | 246 | /** |
| 247 | * Called on state change to un-register/re-register handlers |
| 248 | */ |
| 249 | static void vmnet_vm_state_change_cb(void *opaque, bool running, RunState state) |
| 250 | { |
| 251 | VmnetState *s = opaque; |
| 252 | |
| 253 | if (running) { |
| 254 | vmnet_interface_set_event_callback( |
| 255 | s->vmnet_if, |
| 256 | VMNET_INTERFACE_PACKETS_AVAILABLE, |
| 257 | s->if_queue, |
| 258 | ^(interface_event_t event_id, xpc_object_t event) { |
| 259 | assert(event_id == VMNET_INTERFACE_PACKETS_AVAILABLE); |
| 260 | /* |
| 261 | * This function is being called from a non qemu thread, so |
| 262 | * we only schedule a BH, and do the rest of the io completion |
| 263 | * handling from vmnet_send_bh() which runs in a qemu context. |
| 264 | */ |
| 265 | qemu_bh_schedule(s->send_bh); |
| 266 | }); |
| 267 | } else { |
| 268 | vmnet_interface_set_event_callback( |
| 269 | s->vmnet_if, |
| 270 | VMNET_INTERFACE_PACKETS_AVAILABLE, |
| 271 | NULL, |
| 272 | NULL); |
| 273 | } |
| 274 | } |
Vladislav Yaroshchuk | 73f99db | 2022-03-17 20:28:35 +0300 | [diff] [blame] | 275 | |
| 276 | int vmnet_if_create(NetClientState *nc, |
| 277 | xpc_object_t if_desc, |
| 278 | Error **errp) |
| 279 | { |
| 280 | VmnetState *s = DO_UPCAST(VmnetState, nc, nc); |
| 281 | dispatch_semaphore_t if_created_sem = dispatch_semaphore_create(0); |
| 282 | __block vmnet_return_t if_status; |
| 283 | |
| 284 | s->if_queue = dispatch_queue_create( |
| 285 | "org.qemu.vmnet.if_queue", |
| 286 | DISPATCH_QUEUE_SERIAL |
| 287 | ); |
| 288 | |
| 289 | xpc_dictionary_set_bool( |
| 290 | if_desc, |
| 291 | vmnet_allocate_mac_address_key, |
| 292 | false |
| 293 | ); |
| 294 | |
| 295 | #ifdef DEBUG |
| 296 | qemu_log("vmnet.start.interface_desc:\n"); |
| 297 | xpc_dictionary_apply(if_desc, |
| 298 | ^bool(const char *k, xpc_object_t v) { |
| 299 | char *desc = xpc_copy_description(v); |
| 300 | qemu_log(" %s=%s\n", k, desc); |
| 301 | free(desc); |
| 302 | return true; |
| 303 | }); |
| 304 | #endif /* DEBUG */ |
| 305 | |
| 306 | s->vmnet_if = vmnet_start_interface( |
| 307 | if_desc, |
| 308 | s->if_queue, |
| 309 | ^(vmnet_return_t status, xpc_object_t interface_param) { |
| 310 | if_status = status; |
| 311 | if (status != VMNET_SUCCESS || !interface_param) { |
| 312 | dispatch_semaphore_signal(if_created_sem); |
| 313 | return; |
| 314 | } |
| 315 | |
| 316 | #ifdef DEBUG |
| 317 | qemu_log("vmnet.start.interface_param:\n"); |
| 318 | xpc_dictionary_apply(interface_param, |
| 319 | ^bool(const char *k, xpc_object_t v) { |
| 320 | char *desc = xpc_copy_description(v); |
| 321 | qemu_log(" %s=%s\n", k, desc); |
| 322 | free(desc); |
| 323 | return true; |
| 324 | }); |
| 325 | #endif /* DEBUG */ |
| 326 | |
| 327 | s->mtu = xpc_dictionary_get_uint64( |
| 328 | interface_param, |
| 329 | vmnet_mtu_key); |
| 330 | s->max_packet_size = xpc_dictionary_get_uint64( |
| 331 | interface_param, |
| 332 | vmnet_max_packet_size_key); |
| 333 | |
| 334 | dispatch_semaphore_signal(if_created_sem); |
| 335 | }); |
| 336 | |
| 337 | if (s->vmnet_if == NULL) { |
| 338 | dispatch_release(s->if_queue); |
| 339 | dispatch_release(if_created_sem); |
| 340 | error_setg(errp, |
| 341 | "unable to create interface with requested params"); |
| 342 | return -1; |
| 343 | } |
| 344 | |
| 345 | dispatch_semaphore_wait(if_created_sem, DISPATCH_TIME_FOREVER); |
| 346 | dispatch_release(if_created_sem); |
| 347 | |
| 348 | if (if_status != VMNET_SUCCESS) { |
| 349 | dispatch_release(s->if_queue); |
| 350 | error_setg(errp, |
| 351 | "cannot create vmnet interface: %s", |
| 352 | vmnet_status_map_str(if_status)); |
| 353 | return -1; |
| 354 | } |
| 355 | |
| 356 | s->send_bh = aio_bh_new(qemu_get_aio_context(), vmnet_send_bh, nc); |
| 357 | vmnet_bufs_init(s); |
| 358 | |
| 359 | s->packets_send_current_pos = 0; |
| 360 | s->packets_send_end_pos = 0; |
| 361 | |
Joelle van Dyne | 993f71e | 2023-01-01 17:08:21 -0800 | [diff] [blame] | 362 | vmnet_vm_state_change_cb(s, 1, RUN_STATE_RUNNING); |
| 363 | |
| 364 | s->change = qemu_add_vm_change_state_handler(vmnet_vm_state_change_cb, s); |
Vladislav Yaroshchuk | 73f99db | 2022-03-17 20:28:35 +0300 | [diff] [blame] | 365 | |
| 366 | return 0; |
| 367 | } |
| 368 | |
| 369 | |
| 370 | void vmnet_cleanup_common(NetClientState *nc) |
| 371 | { |
| 372 | VmnetState *s = DO_UPCAST(VmnetState, nc, nc); |
| 373 | dispatch_semaphore_t if_stopped_sem; |
| 374 | |
| 375 | if (s->vmnet_if == NULL) { |
| 376 | return; |
| 377 | } |
| 378 | |
Joelle van Dyne | 993f71e | 2023-01-01 17:08:21 -0800 | [diff] [blame] | 379 | vmnet_vm_state_change_cb(s, 0, RUN_STATE_SHUTDOWN); |
| 380 | qemu_del_vm_change_state_handler(s->change); |
Vladislav Yaroshchuk | 73f99db | 2022-03-17 20:28:35 +0300 | [diff] [blame] | 381 | if_stopped_sem = dispatch_semaphore_create(0); |
| 382 | vmnet_stop_interface( |
| 383 | s->vmnet_if, |
| 384 | s->if_queue, |
| 385 | ^(vmnet_return_t status) { |
| 386 | assert(status == VMNET_SUCCESS); |
| 387 | dispatch_semaphore_signal(if_stopped_sem); |
| 388 | }); |
| 389 | dispatch_semaphore_wait(if_stopped_sem, DISPATCH_TIME_FOREVER); |
| 390 | |
| 391 | qemu_purge_queued_packets(nc); |
| 392 | |
| 393 | qemu_bh_delete(s->send_bh); |
| 394 | dispatch_release(if_stopped_sem); |
| 395 | dispatch_release(s->if_queue); |
| 396 | |
| 397 | for (int i = 0; i < VMNET_PACKETS_LIMIT; ++i) { |
| 398 | g_free(s->iov_buf[i].iov_base); |
| 399 | } |
| 400 | } |