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 <exec-env/rte_kni_common.h> 20 21 #include "compat.h" 22 #include "kni_dev.h" 23 24 MODULE_LICENSE("Dual BSD/GPL"); 25 MODULE_AUTHOR("Intel Corporation"); 26 MODULE_DESCRIPTION("Kernel Module for managing kni devices"); 27 28 #define KNI_RX_LOOP_NUM 1000 29 30 #define KNI_MAX_DEVICES 32 31 32 extern const struct pci_device_id ixgbe_pci_tbl[]; 33 extern const struct pci_device_id igb_pci_tbl[]; 34 35 /* loopback mode */ 36 static char *lo_mode; 37 38 /* Kernel thread mode */ 39 static char *kthread_mode; 40 static uint32_t multiple_kthread_on; 41 42 #define KNI_DEV_IN_USE_BIT_NUM 0 /* Bit number for device in use */ 43 44 static int kni_net_id; 45 46 struct kni_net { 47 unsigned long device_in_use; /* device in use flag */ 48 struct mutex kni_kthread_lock; 49 struct task_struct *kni_kthread; 50 struct rw_semaphore kni_list_lock; 51 struct list_head kni_list_head; 52 }; 53 54 static int __net_init 55 kni_init_net(struct net *net) 56 { 57 #ifdef HAVE_SIMPLIFIED_PERNET_OPERATIONS 58 struct kni_net *knet = net_generic(net, kni_net_id); 59 60 memset(knet, 0, sizeof(*knet)); 61 #else 62 struct kni_net *knet; 63 int ret; 64 65 knet = kzalloc(sizeof(struct kni_net), GFP_KERNEL); 66 if (!knet) { 67 ret = -ENOMEM; 68 return ret; 69 } 70 #endif 71 72 /* Clear the bit of device in use */ 73 clear_bit(KNI_DEV_IN_USE_BIT_NUM, &knet->device_in_use); 74 75 mutex_init(&knet->kni_kthread_lock); 76 77 init_rwsem(&knet->kni_list_lock); 78 INIT_LIST_HEAD(&knet->kni_list_head); 79 80 #ifdef HAVE_SIMPLIFIED_PERNET_OPERATIONS 81 return 0; 82 #else 83 ret = net_assign_generic(net, kni_net_id, knet); 84 if (ret < 0) 85 kfree(knet); 86 87 return ret; 88 #endif 89 } 90 91 static void __net_exit 92 kni_exit_net(struct net *net) 93 { 94 struct kni_net *knet __maybe_unused; 95 96 knet = net_generic(net, kni_net_id); 97 mutex_destroy(&knet->kni_kthread_lock); 98 99 #ifndef HAVE_SIMPLIFIED_PERNET_OPERATIONS 100 kfree(knet); 101 #endif 102 } 103 104 static struct pernet_operations kni_net_ops = { 105 .init = kni_init_net, 106 .exit = kni_exit_net, 107 #ifdef HAVE_SIMPLIFIED_PERNET_OPERATIONS 108 .id = &kni_net_id, 109 .size = sizeof(struct kni_net), 110 #endif 111 }; 112 113 static int 114 kni_thread_single(void *data) 115 { 116 struct kni_net *knet = data; 117 int j; 118 struct kni_dev *dev; 119 120 while (!kthread_should_stop()) { 121 down_read(&knet->kni_list_lock); 122 for (j = 0; j < KNI_RX_LOOP_NUM; j++) { 123 list_for_each_entry(dev, &knet->kni_list_head, list) { 124 kni_net_rx(dev); 125 kni_net_poll_resp(dev); 126 } 127 } 128 up_read(&knet->kni_list_lock); 129 #ifdef RTE_KNI_PREEMPT_DEFAULT 130 /* reschedule out for a while */ 131 schedule_timeout_interruptible( 132 usecs_to_jiffies(KNI_KTHREAD_RESCHEDULE_INTERVAL)); 133 #endif 134 } 135 136 return 0; 137 } 138 139 static int 140 kni_thread_multiple(void *param) 141 { 142 int j; 143 struct kni_dev *dev = param; 144 145 while (!kthread_should_stop()) { 146 for (j = 0; j < KNI_RX_LOOP_NUM; j++) { 147 kni_net_rx(dev); 148 kni_net_poll_resp(dev); 149 } 150 #ifdef RTE_KNI_PREEMPT_DEFAULT 151 schedule_timeout_interruptible( 152 usecs_to_jiffies(KNI_KTHREAD_RESCHEDULE_INTERVAL)); 153 #endif 154 } 155 156 return 0; 157 } 158 159 static int 160 kni_open(struct inode *inode, struct file *file) 161 { 162 struct net *net = current->nsproxy->net_ns; 163 struct kni_net *knet = net_generic(net, kni_net_id); 164 165 /* kni device can be opened by one user only per netns */ 166 if (test_and_set_bit(KNI_DEV_IN_USE_BIT_NUM, &knet->device_in_use)) 167 return -EBUSY; 168 169 file->private_data = get_net(net); 170 pr_debug("/dev/kni opened\n"); 171 172 return 0; 173 } 174 175 static int 176 kni_dev_remove(struct kni_dev *dev) 177 { 178 if (!dev) 179 return -ENODEV; 180 181 #ifdef RTE_KNI_KMOD_ETHTOOL 182 if (dev->pci_dev) { 183 if (pci_match_id(ixgbe_pci_tbl, dev->pci_dev)) 184 ixgbe_kni_remove(dev->pci_dev); 185 else if (pci_match_id(igb_pci_tbl, dev->pci_dev)) 186 igb_kni_remove(dev->pci_dev); 187 } 188 #endif 189 190 if (dev->net_dev) { 191 unregister_netdev(dev->net_dev); 192 free_netdev(dev->net_dev); 193 } 194 195 return 0; 196 } 197 198 static int 199 kni_release(struct inode *inode, struct file *file) 200 { 201 struct net *net = file->private_data; 202 struct kni_net *knet = net_generic(net, kni_net_id); 203 struct kni_dev *dev, *n; 204 205 /* Stop kernel thread for single mode */ 206 if (multiple_kthread_on == 0) { 207 mutex_lock(&knet->kni_kthread_lock); 208 /* Stop kernel thread */ 209 if (knet->kni_kthread != NULL) { 210 kthread_stop(knet->kni_kthread); 211 knet->kni_kthread = NULL; 212 } 213 mutex_unlock(&knet->kni_kthread_lock); 214 } 215 216 down_write(&knet->kni_list_lock); 217 list_for_each_entry_safe(dev, n, &knet->kni_list_head, list) { 218 /* Stop kernel thread for multiple mode */ 219 if (multiple_kthread_on && dev->pthread != NULL) { 220 kthread_stop(dev->pthread); 221 dev->pthread = NULL; 222 } 223 224 kni_dev_remove(dev); 225 list_del(&dev->list); 226 } 227 up_write(&knet->kni_list_lock); 228 229 /* Clear the bit of device in use */ 230 clear_bit(KNI_DEV_IN_USE_BIT_NUM, &knet->device_in_use); 231 232 put_net(net); 233 pr_debug("/dev/kni closed\n"); 234 235 return 0; 236 } 237 238 static int 239 kni_check_param(struct kni_dev *kni, struct rte_kni_device_info *dev) 240 { 241 if (!kni || !dev) 242 return -1; 243 244 /* Check if network name has been used */ 245 if (!strncmp(kni->name, dev->name, RTE_KNI_NAMESIZE)) { 246 pr_err("KNI name %s duplicated\n", dev->name); 247 return -1; 248 } 249 250 return 0; 251 } 252 253 static int 254 kni_run_thread(struct kni_net *knet, struct kni_dev *kni, uint8_t force_bind) 255 { 256 /** 257 * Create a new kernel thread for multiple mode, set its core affinity, 258 * and finally wake it up. 259 */ 260 if (multiple_kthread_on) { 261 kni->pthread = kthread_create(kni_thread_multiple, 262 (void *)kni, "kni_%s", kni->name); 263 if (IS_ERR(kni->pthread)) { 264 kni_dev_remove(kni); 265 return -ECANCELED; 266 } 267 268 if (force_bind) 269 kthread_bind(kni->pthread, kni->core_id); 270 wake_up_process(kni->pthread); 271 } else { 272 mutex_lock(&knet->kni_kthread_lock); 273 274 if (knet->kni_kthread == NULL) { 275 knet->kni_kthread = kthread_create(kni_thread_single, 276 (void *)knet, "kni_single"); 277 if (IS_ERR(knet->kni_kthread)) { 278 mutex_unlock(&knet->kni_kthread_lock); 279 kni_dev_remove(kni); 280 return -ECANCELED; 281 } 282 283 if (force_bind) 284 kthread_bind(knet->kni_kthread, kni->core_id); 285 wake_up_process(knet->kni_kthread); 286 } 287 288 mutex_unlock(&knet->kni_kthread_lock); 289 } 290 291 return 0; 292 } 293 294 static int 295 kni_ioctl_create(struct net *net, uint32_t ioctl_num, 296 unsigned long ioctl_param) 297 { 298 struct kni_net *knet = net_generic(net, kni_net_id); 299 int ret; 300 struct rte_kni_device_info dev_info; 301 struct net_device *net_dev = NULL; 302 struct kni_dev *kni, *dev, *n; 303 #ifdef RTE_KNI_KMOD_ETHTOOL 304 struct pci_dev *found_pci = NULL; 305 struct net_device *lad_dev = NULL; 306 struct pci_dev *pci = NULL; 307 #endif 308 309 pr_info("Creating kni...\n"); 310 /* Check the buffer size, to avoid warning */ 311 if (_IOC_SIZE(ioctl_num) > sizeof(dev_info)) 312 return -EINVAL; 313 314 /* Copy kni info from user space */ 315 ret = copy_from_user(&dev_info, (void *)ioctl_param, sizeof(dev_info)); 316 if (ret) { 317 pr_err("copy_from_user in kni_ioctl_create"); 318 return -EIO; 319 } 320 321 /* Check if name is zero-ended */ 322 if (strnlen(dev_info.name, sizeof(dev_info.name)) == sizeof(dev_info.name)) { 323 pr_err("kni.name not zero-terminated"); 324 return -EINVAL; 325 } 326 327 /** 328 * Check if the cpu core id is valid for binding. 329 */ 330 if (dev_info.force_bind && !cpu_online(dev_info.core_id)) { 331 pr_err("cpu %u is not online\n", dev_info.core_id); 332 return -EINVAL; 333 } 334 335 /* Check if it has been created */ 336 down_read(&knet->kni_list_lock); 337 list_for_each_entry_safe(dev, n, &knet->kni_list_head, list) { 338 if (kni_check_param(dev, &dev_info) < 0) { 339 up_read(&knet->kni_list_lock); 340 return -EINVAL; 341 } 342 } 343 up_read(&knet->kni_list_lock); 344 345 net_dev = alloc_netdev(sizeof(struct kni_dev), dev_info.name, 346 #ifdef NET_NAME_USER 347 NET_NAME_USER, 348 #endif 349 kni_net_init); 350 if (net_dev == NULL) { 351 pr_err("error allocating device \"%s\"\n", dev_info.name); 352 return -EBUSY; 353 } 354 355 dev_net_set(net_dev, net); 356 357 kni = netdev_priv(net_dev); 358 359 kni->net_dev = net_dev; 360 kni->group_id = dev_info.group_id; 361 kni->core_id = dev_info.core_id; 362 strncpy(kni->name, dev_info.name, RTE_KNI_NAMESIZE); 363 364 /* Translate user space info into kernel space info */ 365 kni->tx_q = phys_to_virt(dev_info.tx_phys); 366 kni->rx_q = phys_to_virt(dev_info.rx_phys); 367 kni->alloc_q = phys_to_virt(dev_info.alloc_phys); 368 kni->free_q = phys_to_virt(dev_info.free_phys); 369 370 kni->req_q = phys_to_virt(dev_info.req_phys); 371 kni->resp_q = phys_to_virt(dev_info.resp_phys); 372 kni->sync_va = dev_info.sync_va; 373 kni->sync_kva = phys_to_virt(dev_info.sync_phys); 374 375 kni->mbuf_size = dev_info.mbuf_size; 376 377 pr_debug("tx_phys: 0x%016llx, tx_q addr: 0x%p\n", 378 (unsigned long long) dev_info.tx_phys, kni->tx_q); 379 pr_debug("rx_phys: 0x%016llx, rx_q addr: 0x%p\n", 380 (unsigned long long) dev_info.rx_phys, kni->rx_q); 381 pr_debug("alloc_phys: 0x%016llx, alloc_q addr: 0x%p\n", 382 (unsigned long long) dev_info.alloc_phys, kni->alloc_q); 383 pr_debug("free_phys: 0x%016llx, free_q addr: 0x%p\n", 384 (unsigned long long) dev_info.free_phys, kni->free_q); 385 pr_debug("req_phys: 0x%016llx, req_q addr: 0x%p\n", 386 (unsigned long long) dev_info.req_phys, kni->req_q); 387 pr_debug("resp_phys: 0x%016llx, resp_q addr: 0x%p\n", 388 (unsigned long long) dev_info.resp_phys, kni->resp_q); 389 pr_debug("mbuf_size: %u\n", kni->mbuf_size); 390 391 pr_debug("PCI: %02x:%02x.%02x %04x:%04x\n", 392 dev_info.bus, 393 dev_info.devid, 394 dev_info.function, 395 dev_info.vendor_id, 396 dev_info.device_id); 397 #ifdef RTE_KNI_KMOD_ETHTOOL 398 pci = pci_get_device(dev_info.vendor_id, dev_info.device_id, NULL); 399 400 /* Support Ethtool */ 401 while (pci) { 402 pr_debug("pci_bus: %02x:%02x:%02x\n", 403 pci->bus->number, 404 PCI_SLOT(pci->devfn), 405 PCI_FUNC(pci->devfn)); 406 407 if ((pci->bus->number == dev_info.bus) && 408 (PCI_SLOT(pci->devfn) == dev_info.devid) && 409 (PCI_FUNC(pci->devfn) == dev_info.function)) { 410 found_pci = pci; 411 412 if (pci_match_id(ixgbe_pci_tbl, found_pci)) 413 ret = ixgbe_kni_probe(found_pci, &lad_dev); 414 else if (pci_match_id(igb_pci_tbl, found_pci)) 415 ret = igb_kni_probe(found_pci, &lad_dev); 416 else 417 ret = -1; 418 419 pr_debug("PCI found: pci=0x%p, lad_dev=0x%p\n", 420 pci, lad_dev); 421 if (ret == 0) { 422 kni->lad_dev = lad_dev; 423 kni_set_ethtool_ops(kni->net_dev); 424 } else { 425 pr_err("Device not supported by ethtool"); 426 kni->lad_dev = NULL; 427 } 428 429 kni->pci_dev = found_pci; 430 kni->device_id = dev_info.device_id; 431 break; 432 } 433 pci = pci_get_device(dev_info.vendor_id, 434 dev_info.device_id, pci); 435 } 436 if (pci) 437 pci_dev_put(pci); 438 #endif 439 440 if (kni->lad_dev) 441 ether_addr_copy(net_dev->dev_addr, kni->lad_dev->dev_addr); 442 else { 443 /* if user has provided a valid mac address */ 444 if (is_valid_ether_addr((unsigned char *)(dev_info.mac_addr))) 445 memcpy(net_dev->dev_addr, dev_info.mac_addr, ETH_ALEN); 446 else 447 /* 448 * Generate random mac address. eth_random_addr() is the 449 * newer version of generating mac address in kernel. 450 */ 451 random_ether_addr(net_dev->dev_addr); 452 } 453 454 if (dev_info.mtu) 455 net_dev->mtu = dev_info.mtu; 456 457 ret = register_netdev(net_dev); 458 if (ret) { 459 pr_err("error %i registering device \"%s\"\n", 460 ret, dev_info.name); 461 kni->net_dev = NULL; 462 kni_dev_remove(kni); 463 free_netdev(net_dev); 464 return -ENODEV; 465 } 466 467 ret = kni_run_thread(knet, kni, dev_info.force_bind); 468 if (ret != 0) 469 return ret; 470 471 down_write(&knet->kni_list_lock); 472 list_add(&kni->list, &knet->kni_list_head); 473 up_write(&knet->kni_list_lock); 474 475 return 0; 476 } 477 478 static int 479 kni_ioctl_release(struct net *net, uint32_t ioctl_num, 480 unsigned long ioctl_param) 481 { 482 struct kni_net *knet = net_generic(net, kni_net_id); 483 int ret = -EINVAL; 484 struct kni_dev *dev, *n; 485 struct rte_kni_device_info dev_info; 486 487 if (_IOC_SIZE(ioctl_num) > sizeof(dev_info)) 488 return -EINVAL; 489 490 ret = copy_from_user(&dev_info, (void *)ioctl_param, sizeof(dev_info)); 491 if (ret) { 492 pr_err("copy_from_user in kni_ioctl_release"); 493 return -EIO; 494 } 495 496 /* Release the network device according to its name */ 497 if (strlen(dev_info.name) == 0) 498 return ret; 499 500 down_write(&knet->kni_list_lock); 501 list_for_each_entry_safe(dev, n, &knet->kni_list_head, list) { 502 if (strncmp(dev->name, dev_info.name, RTE_KNI_NAMESIZE) != 0) 503 continue; 504 505 if (multiple_kthread_on && dev->pthread != NULL) { 506 kthread_stop(dev->pthread); 507 dev->pthread = NULL; 508 } 509 510 kni_dev_remove(dev); 511 list_del(&dev->list); 512 ret = 0; 513 break; 514 } 515 up_write(&knet->kni_list_lock); 516 pr_info("%s release kni named %s\n", 517 (ret == 0 ? "Successfully" : "Unsuccessfully"), dev_info.name); 518 519 return ret; 520 } 521 522 static int 523 kni_ioctl(struct inode *inode, uint32_t ioctl_num, unsigned long ioctl_param) 524 { 525 int ret = -EINVAL; 526 struct net *net = current->nsproxy->net_ns; 527 528 pr_debug("IOCTL num=0x%0x param=0x%0lx\n", ioctl_num, ioctl_param); 529 530 /* 531 * Switch according to the ioctl called 532 */ 533 switch (_IOC_NR(ioctl_num)) { 534 case _IOC_NR(RTE_KNI_IOCTL_TEST): 535 /* For test only, not used */ 536 break; 537 case _IOC_NR(RTE_KNI_IOCTL_CREATE): 538 ret = kni_ioctl_create(net, ioctl_num, ioctl_param); 539 break; 540 case _IOC_NR(RTE_KNI_IOCTL_RELEASE): 541 ret = kni_ioctl_release(net, ioctl_num, ioctl_param); 542 break; 543 default: 544 pr_debug("IOCTL default\n"); 545 break; 546 } 547 548 return ret; 549 } 550 551 static int 552 kni_compat_ioctl(struct inode *inode, uint32_t ioctl_num, 553 unsigned long ioctl_param) 554 { 555 /* 32 bits app on 64 bits OS to be supported later */ 556 pr_debug("Not implemented.\n"); 557 558 return -EINVAL; 559 } 560 561 static const struct file_operations kni_fops = { 562 .owner = THIS_MODULE, 563 .open = kni_open, 564 .release = kni_release, 565 .unlocked_ioctl = (void *)kni_ioctl, 566 .compat_ioctl = (void *)kni_compat_ioctl, 567 }; 568 569 static struct miscdevice kni_misc = { 570 .minor = MISC_DYNAMIC_MINOR, 571 .name = KNI_DEVICE, 572 .fops = &kni_fops, 573 }; 574 575 static int __init 576 kni_parse_kthread_mode(void) 577 { 578 if (!kthread_mode) 579 return 0; 580 581 if (strcmp(kthread_mode, "single") == 0) 582 return 0; 583 else if (strcmp(kthread_mode, "multiple") == 0) 584 multiple_kthread_on = 1; 585 else 586 return -1; 587 588 return 0; 589 } 590 591 static int __init 592 kni_init(void) 593 { 594 int rc; 595 596 if (kni_parse_kthread_mode() < 0) { 597 pr_err("Invalid parameter for kthread_mode\n"); 598 return -EINVAL; 599 } 600 601 if (multiple_kthread_on == 0) 602 pr_debug("Single kernel thread for all KNI devices\n"); 603 else 604 pr_debug("Multiple kernel thread mode enabled\n"); 605 606 #ifdef HAVE_SIMPLIFIED_PERNET_OPERATIONS 607 rc = register_pernet_subsys(&kni_net_ops); 608 #else 609 rc = register_pernet_gen_subsys(&kni_net_id, &kni_net_ops); 610 #endif 611 if (rc) 612 return -EPERM; 613 614 rc = misc_register(&kni_misc); 615 if (rc != 0) { 616 pr_err("Misc registration failed\n"); 617 goto out; 618 } 619 620 /* Configure the lo mode according to the input parameter */ 621 kni_net_config_lo_mode(lo_mode); 622 623 return 0; 624 625 out: 626 #ifdef HAVE_SIMPLIFIED_PERNET_OPERATIONS 627 unregister_pernet_subsys(&kni_net_ops); 628 #else 629 unregister_pernet_gen_subsys(kni_net_id, &kni_net_ops); 630 #endif 631 return rc; 632 } 633 634 static void __exit 635 kni_exit(void) 636 { 637 misc_deregister(&kni_misc); 638 #ifdef HAVE_SIMPLIFIED_PERNET_OPERATIONS 639 unregister_pernet_subsys(&kni_net_ops); 640 #else 641 unregister_pernet_gen_subsys(kni_net_id, &kni_net_ops); 642 #endif 643 } 644 645 module_init(kni_init); 646 module_exit(kni_exit); 647 648 module_param(lo_mode, charp, S_IRUGO | S_IWUSR); 649 MODULE_PARM_DESC(lo_mode, 650 "KNI loopback mode (default=lo_mode_none):\n" 651 " lo_mode_none Kernel loopback disabled\n" 652 " lo_mode_fifo Enable kernel loopback with fifo\n" 653 " lo_mode_fifo_skb Enable kernel loopback with fifo and skb buffer\n" 654 "\n" 655 ); 656 657 module_param(kthread_mode, charp, S_IRUGO); 658 MODULE_PARM_DESC(kthread_mode, 659 "Kernel thread mode (default=single):\n" 660 " single Single kernel thread mode enabled.\n" 661 " multiple Multiple kernel thread mode enabled.\n" 662 "\n" 663 ); 664