1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Copyright(c) 2010-2014 Intel Corporation. 4 */ 5 6 #include <linux/version.h> 7 #include <linux/module.h> 8 #include <linux/miscdevice.h> 9 #include <linux/netdevice.h> 10 #include <linux/etherdevice.h> 11 #include <linux/pci.h> 12 #include <linux/kthread.h> 13 #include <linux/rwsem.h> 14 #include <linux/mutex.h> 15 #include <linux/nsproxy.h> 16 #include <net/net_namespace.h> 17 #include <net/netns/generic.h> 18 19 #include <rte_kni_common.h> 20 21 #include "compat.h" 22 #include "kni_dev.h" 23 24 MODULE_VERSION(KNI_VERSION); 25 MODULE_LICENSE("Dual BSD/GPL"); 26 MODULE_AUTHOR("Intel Corporation"); 27 MODULE_DESCRIPTION("Kernel Module for managing kni devices"); 28 29 #define KNI_RX_LOOP_NUM 1000 30 31 #define KNI_MAX_DEVICES 32 32 33 /* loopback mode */ 34 static char *lo_mode; 35 36 /* Kernel thread mode */ 37 static char *kthread_mode; 38 static uint32_t multiple_kthread_on; 39 40 /* Default carrier state for created KNI network interfaces */ 41 static char *carrier; 42 uint32_t dflt_carrier; 43 44 #define KNI_DEV_IN_USE_BIT_NUM 0 /* Bit number for device in use */ 45 46 static int kni_net_id; 47 48 struct kni_net { 49 unsigned long device_in_use; /* device in use flag */ 50 struct mutex kni_kthread_lock; 51 struct task_struct *kni_kthread; 52 struct rw_semaphore kni_list_lock; 53 struct list_head kni_list_head; 54 }; 55 56 static int __net_init 57 kni_init_net(struct net *net) 58 { 59 #ifdef HAVE_SIMPLIFIED_PERNET_OPERATIONS 60 struct kni_net *knet = net_generic(net, kni_net_id); 61 62 memset(knet, 0, sizeof(*knet)); 63 #else 64 struct kni_net *knet; 65 int ret; 66 67 knet = kzalloc(sizeof(struct kni_net), GFP_KERNEL); 68 if (!knet) { 69 ret = -ENOMEM; 70 return ret; 71 } 72 #endif 73 74 /* Clear the bit of device in use */ 75 clear_bit(KNI_DEV_IN_USE_BIT_NUM, &knet->device_in_use); 76 77 mutex_init(&knet->kni_kthread_lock); 78 79 init_rwsem(&knet->kni_list_lock); 80 INIT_LIST_HEAD(&knet->kni_list_head); 81 82 #ifdef HAVE_SIMPLIFIED_PERNET_OPERATIONS 83 return 0; 84 #else 85 ret = net_assign_generic(net, kni_net_id, knet); 86 if (ret < 0) 87 kfree(knet); 88 89 return ret; 90 #endif 91 } 92 93 static void __net_exit 94 kni_exit_net(struct net *net) 95 { 96 struct kni_net *knet __maybe_unused; 97 98 knet = net_generic(net, kni_net_id); 99 mutex_destroy(&knet->kni_kthread_lock); 100 101 #ifndef HAVE_SIMPLIFIED_PERNET_OPERATIONS 102 kfree(knet); 103 #endif 104 } 105 106 static struct pernet_operations kni_net_ops = { 107 .init = kni_init_net, 108 .exit = kni_exit_net, 109 #ifdef HAVE_SIMPLIFIED_PERNET_OPERATIONS 110 .id = &kni_net_id, 111 .size = sizeof(struct kni_net), 112 #endif 113 }; 114 115 static int 116 kni_thread_single(void *data) 117 { 118 struct kni_net *knet = data; 119 int j; 120 struct kni_dev *dev; 121 122 while (!kthread_should_stop()) { 123 down_read(&knet->kni_list_lock); 124 for (j = 0; j < KNI_RX_LOOP_NUM; j++) { 125 list_for_each_entry(dev, &knet->kni_list_head, list) { 126 kni_net_rx(dev); 127 kni_net_poll_resp(dev); 128 } 129 } 130 up_read(&knet->kni_list_lock); 131 #ifdef RTE_KNI_PREEMPT_DEFAULT 132 /* reschedule out for a while */ 133 schedule_timeout_interruptible( 134 usecs_to_jiffies(KNI_KTHREAD_RESCHEDULE_INTERVAL)); 135 #endif 136 } 137 138 return 0; 139 } 140 141 static int 142 kni_thread_multiple(void *param) 143 { 144 int j; 145 struct kni_dev *dev = param; 146 147 while (!kthread_should_stop()) { 148 for (j = 0; j < KNI_RX_LOOP_NUM; j++) { 149 kni_net_rx(dev); 150 kni_net_poll_resp(dev); 151 } 152 #ifdef RTE_KNI_PREEMPT_DEFAULT 153 schedule_timeout_interruptible( 154 usecs_to_jiffies(KNI_KTHREAD_RESCHEDULE_INTERVAL)); 155 #endif 156 } 157 158 return 0; 159 } 160 161 static int 162 kni_open(struct inode *inode, struct file *file) 163 { 164 struct net *net = current->nsproxy->net_ns; 165 struct kni_net *knet = net_generic(net, kni_net_id); 166 167 /* kni device can be opened by one user only per netns */ 168 if (test_and_set_bit(KNI_DEV_IN_USE_BIT_NUM, &knet->device_in_use)) 169 return -EBUSY; 170 171 file->private_data = get_net(net); 172 pr_debug("/dev/kni opened\n"); 173 174 return 0; 175 } 176 177 static int 178 kni_dev_remove(struct kni_dev *dev) 179 { 180 if (!dev) 181 return -ENODEV; 182 183 if (dev->net_dev) { 184 unregister_netdev(dev->net_dev); 185 free_netdev(dev->net_dev); 186 } 187 188 kni_net_release_fifo_phy(dev); 189 190 return 0; 191 } 192 193 static int 194 kni_release(struct inode *inode, struct file *file) 195 { 196 struct net *net = file->private_data; 197 struct kni_net *knet = net_generic(net, kni_net_id); 198 struct kni_dev *dev, *n; 199 200 /* Stop kernel thread for single mode */ 201 if (multiple_kthread_on == 0) { 202 mutex_lock(&knet->kni_kthread_lock); 203 /* Stop kernel thread */ 204 if (knet->kni_kthread != NULL) { 205 kthread_stop(knet->kni_kthread); 206 knet->kni_kthread = NULL; 207 } 208 mutex_unlock(&knet->kni_kthread_lock); 209 } 210 211 down_write(&knet->kni_list_lock); 212 list_for_each_entry_safe(dev, n, &knet->kni_list_head, list) { 213 /* Stop kernel thread for multiple mode */ 214 if (multiple_kthread_on && dev->pthread != NULL) { 215 kthread_stop(dev->pthread); 216 dev->pthread = NULL; 217 } 218 219 kni_dev_remove(dev); 220 list_del(&dev->list); 221 } 222 up_write(&knet->kni_list_lock); 223 224 /* Clear the bit of device in use */ 225 clear_bit(KNI_DEV_IN_USE_BIT_NUM, &knet->device_in_use); 226 227 put_net(net); 228 pr_debug("/dev/kni closed\n"); 229 230 return 0; 231 } 232 233 static int 234 kni_check_param(struct kni_dev *kni, struct rte_kni_device_info *dev) 235 { 236 if (!kni || !dev) 237 return -1; 238 239 /* Check if network name has been used */ 240 if (!strncmp(kni->name, dev->name, RTE_KNI_NAMESIZE)) { 241 pr_err("KNI name %s duplicated\n", dev->name); 242 return -1; 243 } 244 245 return 0; 246 } 247 248 static int 249 kni_run_thread(struct kni_net *knet, struct kni_dev *kni, uint8_t force_bind) 250 { 251 /** 252 * Create a new kernel thread for multiple mode, set its core affinity, 253 * and finally wake it up. 254 */ 255 if (multiple_kthread_on) { 256 kni->pthread = kthread_create(kni_thread_multiple, 257 (void *)kni, "kni_%s", kni->name); 258 if (IS_ERR(kni->pthread)) { 259 kni_dev_remove(kni); 260 return -ECANCELED; 261 } 262 263 if (force_bind) 264 kthread_bind(kni->pthread, kni->core_id); 265 wake_up_process(kni->pthread); 266 } else { 267 mutex_lock(&knet->kni_kthread_lock); 268 269 if (knet->kni_kthread == NULL) { 270 knet->kni_kthread = kthread_create(kni_thread_single, 271 (void *)knet, "kni_single"); 272 if (IS_ERR(knet->kni_kthread)) { 273 mutex_unlock(&knet->kni_kthread_lock); 274 kni_dev_remove(kni); 275 return -ECANCELED; 276 } 277 278 if (force_bind) 279 kthread_bind(knet->kni_kthread, kni->core_id); 280 wake_up_process(knet->kni_kthread); 281 } 282 283 mutex_unlock(&knet->kni_kthread_lock); 284 } 285 286 return 0; 287 } 288 289 static int 290 kni_ioctl_create(struct net *net, uint32_t ioctl_num, 291 unsigned long ioctl_param) 292 { 293 struct kni_net *knet = net_generic(net, kni_net_id); 294 int ret; 295 struct rte_kni_device_info dev_info; 296 struct net_device *net_dev = NULL; 297 struct kni_dev *kni, *dev, *n; 298 299 pr_info("Creating kni...\n"); 300 /* Check the buffer size, to avoid warning */ 301 if (_IOC_SIZE(ioctl_num) > sizeof(dev_info)) 302 return -EINVAL; 303 304 /* Copy kni info from user space */ 305 if (copy_from_user(&dev_info, (void *)ioctl_param, sizeof(dev_info))) 306 return -EFAULT; 307 308 /* Check if name is zero-ended */ 309 if (strnlen(dev_info.name, sizeof(dev_info.name)) == sizeof(dev_info.name)) { 310 pr_err("kni.name not zero-terminated"); 311 return -EINVAL; 312 } 313 314 /** 315 * Check if the cpu core id is valid for binding. 316 */ 317 if (dev_info.force_bind && !cpu_online(dev_info.core_id)) { 318 pr_err("cpu %u is not online\n", dev_info.core_id); 319 return -EINVAL; 320 } 321 322 /* Check if it has been created */ 323 down_read(&knet->kni_list_lock); 324 list_for_each_entry_safe(dev, n, &knet->kni_list_head, list) { 325 if (kni_check_param(dev, &dev_info) < 0) { 326 up_read(&knet->kni_list_lock); 327 return -EINVAL; 328 } 329 } 330 up_read(&knet->kni_list_lock); 331 332 net_dev = alloc_netdev(sizeof(struct kni_dev), dev_info.name, 333 #ifdef NET_NAME_USER 334 NET_NAME_USER, 335 #endif 336 kni_net_init); 337 if (net_dev == NULL) { 338 pr_err("error allocating device \"%s\"\n", dev_info.name); 339 return -EBUSY; 340 } 341 342 dev_net_set(net_dev, net); 343 344 kni = netdev_priv(net_dev); 345 346 kni->net_dev = net_dev; 347 kni->core_id = dev_info.core_id; 348 strncpy(kni->name, dev_info.name, RTE_KNI_NAMESIZE); 349 350 /* Translate user space info into kernel space info */ 351 kni->tx_q = phys_to_virt(dev_info.tx_phys); 352 kni->rx_q = phys_to_virt(dev_info.rx_phys); 353 kni->alloc_q = phys_to_virt(dev_info.alloc_phys); 354 kni->free_q = phys_to_virt(dev_info.free_phys); 355 356 kni->req_q = phys_to_virt(dev_info.req_phys); 357 kni->resp_q = phys_to_virt(dev_info.resp_phys); 358 kni->sync_va = dev_info.sync_va; 359 kni->sync_kva = phys_to_virt(dev_info.sync_phys); 360 361 kni->mbuf_size = dev_info.mbuf_size; 362 363 pr_debug("tx_phys: 0x%016llx, tx_q addr: 0x%p\n", 364 (unsigned long long) dev_info.tx_phys, kni->tx_q); 365 pr_debug("rx_phys: 0x%016llx, rx_q addr: 0x%p\n", 366 (unsigned long long) dev_info.rx_phys, kni->rx_q); 367 pr_debug("alloc_phys: 0x%016llx, alloc_q addr: 0x%p\n", 368 (unsigned long long) dev_info.alloc_phys, kni->alloc_q); 369 pr_debug("free_phys: 0x%016llx, free_q addr: 0x%p\n", 370 (unsigned long long) dev_info.free_phys, kni->free_q); 371 pr_debug("req_phys: 0x%016llx, req_q addr: 0x%p\n", 372 (unsigned long long) dev_info.req_phys, kni->req_q); 373 pr_debug("resp_phys: 0x%016llx, resp_q addr: 0x%p\n", 374 (unsigned long long) dev_info.resp_phys, kni->resp_q); 375 pr_debug("mbuf_size: %u\n", kni->mbuf_size); 376 377 /* if user has provided a valid mac address */ 378 if (is_valid_ether_addr(dev_info.mac_addr)) 379 memcpy(net_dev->dev_addr, dev_info.mac_addr, ETH_ALEN); 380 else 381 /* 382 * Generate random mac address. eth_random_addr() is the 383 * newer version of generating mac address in kernel. 384 */ 385 random_ether_addr(net_dev->dev_addr); 386 387 if (dev_info.mtu) 388 net_dev->mtu = dev_info.mtu; 389 #ifdef HAVE_MAX_MTU_PARAM 390 net_dev->max_mtu = net_dev->mtu; 391 #endif 392 393 ret = register_netdev(net_dev); 394 if (ret) { 395 pr_err("error %i registering device \"%s\"\n", 396 ret, dev_info.name); 397 kni->net_dev = NULL; 398 kni_dev_remove(kni); 399 free_netdev(net_dev); 400 return -ENODEV; 401 } 402 403 netif_carrier_off(net_dev); 404 405 ret = kni_run_thread(knet, kni, dev_info.force_bind); 406 if (ret != 0) 407 return ret; 408 409 down_write(&knet->kni_list_lock); 410 list_add(&kni->list, &knet->kni_list_head); 411 up_write(&knet->kni_list_lock); 412 413 return 0; 414 } 415 416 static int 417 kni_ioctl_release(struct net *net, uint32_t ioctl_num, 418 unsigned long ioctl_param) 419 { 420 struct kni_net *knet = net_generic(net, kni_net_id); 421 int ret = -EINVAL; 422 struct kni_dev *dev, *n; 423 struct rte_kni_device_info dev_info; 424 425 if (_IOC_SIZE(ioctl_num) > sizeof(dev_info)) 426 return -EINVAL; 427 428 if (copy_from_user(&dev_info, (void *)ioctl_param, sizeof(dev_info))) 429 return -EFAULT; 430 431 /* Release the network device according to its name */ 432 if (strlen(dev_info.name) == 0) 433 return -EINVAL; 434 435 down_write(&knet->kni_list_lock); 436 list_for_each_entry_safe(dev, n, &knet->kni_list_head, list) { 437 if (strncmp(dev->name, dev_info.name, RTE_KNI_NAMESIZE) != 0) 438 continue; 439 440 if (multiple_kthread_on && dev->pthread != NULL) { 441 kthread_stop(dev->pthread); 442 dev->pthread = NULL; 443 } 444 445 kni_dev_remove(dev); 446 list_del(&dev->list); 447 ret = 0; 448 break; 449 } 450 up_write(&knet->kni_list_lock); 451 pr_info("%s release kni named %s\n", 452 (ret == 0 ? "Successfully" : "Unsuccessfully"), dev_info.name); 453 454 return ret; 455 } 456 457 static int 458 kni_ioctl(struct inode *inode, uint32_t ioctl_num, unsigned long ioctl_param) 459 { 460 int ret = -EINVAL; 461 struct net *net = current->nsproxy->net_ns; 462 463 pr_debug("IOCTL num=0x%0x param=0x%0lx\n", ioctl_num, ioctl_param); 464 465 /* 466 * Switch according to the ioctl called 467 */ 468 switch (_IOC_NR(ioctl_num)) { 469 case _IOC_NR(RTE_KNI_IOCTL_TEST): 470 /* For test only, not used */ 471 break; 472 case _IOC_NR(RTE_KNI_IOCTL_CREATE): 473 ret = kni_ioctl_create(net, ioctl_num, ioctl_param); 474 break; 475 case _IOC_NR(RTE_KNI_IOCTL_RELEASE): 476 ret = kni_ioctl_release(net, ioctl_num, ioctl_param); 477 break; 478 default: 479 pr_debug("IOCTL default\n"); 480 break; 481 } 482 483 return ret; 484 } 485 486 static int 487 kni_compat_ioctl(struct inode *inode, uint32_t ioctl_num, 488 unsigned long ioctl_param) 489 { 490 /* 32 bits app on 64 bits OS to be supported later */ 491 pr_debug("Not implemented.\n"); 492 493 return -EINVAL; 494 } 495 496 static const struct file_operations kni_fops = { 497 .owner = THIS_MODULE, 498 .open = kni_open, 499 .release = kni_release, 500 .unlocked_ioctl = (void *)kni_ioctl, 501 .compat_ioctl = (void *)kni_compat_ioctl, 502 }; 503 504 static struct miscdevice kni_misc = { 505 .minor = MISC_DYNAMIC_MINOR, 506 .name = KNI_DEVICE, 507 .fops = &kni_fops, 508 }; 509 510 static int __init 511 kni_parse_kthread_mode(void) 512 { 513 if (!kthread_mode) 514 return 0; 515 516 if (strcmp(kthread_mode, "single") == 0) 517 return 0; 518 else if (strcmp(kthread_mode, "multiple") == 0) 519 multiple_kthread_on = 1; 520 else 521 return -1; 522 523 return 0; 524 } 525 526 static int __init 527 kni_parse_carrier_state(void) 528 { 529 if (!carrier) { 530 dflt_carrier = 0; 531 return 0; 532 } 533 534 if (strcmp(carrier, "off") == 0) 535 dflt_carrier = 0; 536 else if (strcmp(carrier, "on") == 0) 537 dflt_carrier = 1; 538 else 539 return -1; 540 541 return 0; 542 } 543 544 static int __init 545 kni_init(void) 546 { 547 int rc; 548 549 if (kni_parse_kthread_mode() < 0) { 550 pr_err("Invalid parameter for kthread_mode\n"); 551 return -EINVAL; 552 } 553 554 if (multiple_kthread_on == 0) 555 pr_debug("Single kernel thread for all KNI devices\n"); 556 else 557 pr_debug("Multiple kernel thread mode enabled\n"); 558 559 if (kni_parse_carrier_state() < 0) { 560 pr_err("Invalid parameter for carrier\n"); 561 return -EINVAL; 562 } 563 564 if (dflt_carrier == 0) 565 pr_debug("Default carrier state set to off.\n"); 566 else 567 pr_debug("Default carrier state set to on.\n"); 568 569 #ifdef HAVE_SIMPLIFIED_PERNET_OPERATIONS 570 rc = register_pernet_subsys(&kni_net_ops); 571 #else 572 rc = register_pernet_gen_subsys(&kni_net_id, &kni_net_ops); 573 #endif 574 if (rc) 575 return -EPERM; 576 577 rc = misc_register(&kni_misc); 578 if (rc != 0) { 579 pr_err("Misc registration failed\n"); 580 goto out; 581 } 582 583 /* Configure the lo mode according to the input parameter */ 584 kni_net_config_lo_mode(lo_mode); 585 586 return 0; 587 588 out: 589 #ifdef HAVE_SIMPLIFIED_PERNET_OPERATIONS 590 unregister_pernet_subsys(&kni_net_ops); 591 #else 592 unregister_pernet_gen_subsys(kni_net_id, &kni_net_ops); 593 #endif 594 return rc; 595 } 596 597 static void __exit 598 kni_exit(void) 599 { 600 misc_deregister(&kni_misc); 601 #ifdef HAVE_SIMPLIFIED_PERNET_OPERATIONS 602 unregister_pernet_subsys(&kni_net_ops); 603 #else 604 unregister_pernet_gen_subsys(kni_net_id, &kni_net_ops); 605 #endif 606 } 607 608 module_init(kni_init); 609 module_exit(kni_exit); 610 611 module_param(lo_mode, charp, 0644); 612 MODULE_PARM_DESC(lo_mode, 613 "KNI loopback mode (default=lo_mode_none):\n" 614 "\t\tlo_mode_none Kernel loopback disabled\n" 615 "\t\tlo_mode_fifo Enable kernel loopback with fifo\n" 616 "\t\tlo_mode_fifo_skb Enable kernel loopback with fifo and skb buffer\n" 617 "\t\t" 618 ); 619 620 module_param(kthread_mode, charp, 0644); 621 MODULE_PARM_DESC(kthread_mode, 622 "Kernel thread mode (default=single):\n" 623 "\t\tsingle Single kernel thread mode enabled.\n" 624 "\t\tmultiple Multiple kernel thread mode enabled.\n" 625 "\t\t" 626 ); 627 628 module_param(carrier, charp, 0644); 629 MODULE_PARM_DESC(carrier, 630 "Default carrier state for KNI interface (default=off):\n" 631 "\t\toff Interfaces will be created with carrier state set to off.\n" 632 "\t\ton Interfaces will be created with carrier state set to on.\n" 633 "\t\t" 634 ); 635