1 /* 2 * Copyright (C) 2017 Netronome Systems, Inc. 3 * 4 * This software is licensed under the GNU General License Version 2, 5 * June 1991 as shown in the file COPYING in the top-level directory of this 6 * source tree. 7 * 8 * THE COPYRIGHT HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS" 9 * WITHOUT WARRANTY OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, 10 * BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 11 * FOR A PARTICULAR PURPOSE. THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE 12 * OF THE PROGRAM IS WITH YOU. SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME 13 * THE COST OF ALL NECESSARY SERVICING, REPAIR OR CORRECTION. 14 */ 15 16 #include <linux/debugfs.h> 17 #include <linux/etherdevice.h> 18 #include <linux/ethtool_netlink.h> 19 #include <linux/kernel.h> 20 #include <linux/module.h> 21 #include <linux/netdevice.h> 22 #include <linux/slab.h> 23 #include <net/netdev_queues.h> 24 #include <net/netdev_rx_queue.h> 25 #include <net/page_pool/helpers.h> 26 #include <net/netlink.h> 27 #include <net/net_shaper.h> 28 #include <net/netdev_lock.h> 29 #include <net/pkt_cls.h> 30 #include <net/rtnetlink.h> 31 #include <net/udp_tunnel.h> 32 #include <net/busy_poll.h> 33 34 #include "netdevsim.h" 35 36 MODULE_IMPORT_NS("NETDEV_INTERNAL"); 37 38 #define NSIM_RING_SIZE 256 39 40 static void nsim_start_peer_tx_queue(struct net_device *dev, struct nsim_rq *rq) 41 { 42 struct netdevsim *ns = netdev_priv(dev); 43 struct net_device *peer_dev; 44 struct netdevsim *peer_ns; 45 struct netdev_queue *txq; 46 u16 idx; 47 48 idx = rq->napi.index; 49 rcu_read_lock(); 50 peer_ns = rcu_dereference(ns->peer); 51 if (!peer_ns) 52 goto out; 53 54 /* TX device */ 55 peer_dev = peer_ns->netdev; 56 if (dev->real_num_tx_queues != peer_dev->num_rx_queues) 57 goto out; 58 59 txq = netdev_get_tx_queue(peer_dev, idx); 60 if (!netif_tx_queue_stopped(txq)) 61 goto out; 62 63 netif_tx_wake_queue(txq); 64 out: 65 rcu_read_unlock(); 66 } 67 68 static void nsim_stop_tx_queue(struct net_device *tx_dev, 69 struct net_device *rx_dev, 70 struct nsim_rq *rq, 71 u16 idx) 72 { 73 /* If different queues size, do not stop, since it is not 74 * easy to find which TX queue is mapped here 75 */ 76 if (rx_dev->real_num_tx_queues != tx_dev->num_rx_queues) 77 return; 78 79 /* rq is the queue on the receive side */ 80 netif_subqueue_try_stop(tx_dev, idx, 81 NSIM_RING_SIZE - skb_queue_len(&rq->skb_queue), 82 NSIM_RING_SIZE / 2); 83 } 84 85 static int nsim_napi_rx(struct net_device *tx_dev, struct net_device *rx_dev, 86 struct nsim_rq *rq, struct sk_buff *skb) 87 { 88 if (skb_queue_len(&rq->skb_queue) > NSIM_RING_SIZE) { 89 dev_kfree_skb_any(skb); 90 return NET_RX_DROP; 91 } 92 93 skb_queue_tail(&rq->skb_queue, skb); 94 95 /* Stop the peer TX queue avoiding dropping packets later */ 96 if (skb_queue_len(&rq->skb_queue) >= NSIM_RING_SIZE) 97 nsim_stop_tx_queue(tx_dev, rx_dev, rq, 98 skb_get_queue_mapping(skb)); 99 100 return NET_RX_SUCCESS; 101 } 102 103 static int nsim_forward_skb(struct net_device *tx_dev, 104 struct net_device *rx_dev, 105 struct sk_buff *skb, 106 struct nsim_rq *rq, 107 struct skb_ext *psp_ext) 108 { 109 int ret; 110 111 ret = __dev_forward_skb(rx_dev, skb); 112 if (ret) { 113 if (psp_ext) 114 __skb_ext_put(psp_ext); 115 return ret; 116 } 117 118 nsim_psp_handle_ext(skb, psp_ext); 119 120 return nsim_napi_rx(tx_dev, rx_dev, rq, skb); 121 } 122 123 static netdev_tx_t nsim_start_xmit(struct sk_buff *skb, struct net_device *dev) 124 { 125 struct netdevsim *ns = netdev_priv(dev); 126 struct skb_ext *psp_ext = NULL; 127 struct net_device *peer_dev; 128 unsigned int len = skb->len; 129 struct netdevsim *peer_ns; 130 struct netdev_config *cfg; 131 struct nsim_rq *rq; 132 int rxq; 133 int dr; 134 135 rcu_read_lock(); 136 if (!nsim_ipsec_tx(ns, skb)) 137 goto out_drop_any; 138 139 /* Check if loopback mode is enabled */ 140 if (dev->features & NETIF_F_LOOPBACK) { 141 peer_ns = ns; 142 peer_dev = dev; 143 } else { 144 peer_ns = rcu_dereference(ns->peer); 145 if (!peer_ns) 146 goto out_drop_any; 147 peer_dev = peer_ns->netdev; 148 } 149 150 dr = nsim_do_psp(skb, ns, peer_ns, &psp_ext); 151 if (dr) 152 goto out_drop_free; 153 154 rxq = skb_get_queue_mapping(skb); 155 if (rxq >= peer_dev->num_rx_queues) 156 rxq = rxq % peer_dev->num_rx_queues; 157 rq = peer_ns->rq[rxq]; 158 159 cfg = peer_dev->cfg; 160 if (skb_is_nonlinear(skb) && 161 (cfg->hds_config != ETHTOOL_TCP_DATA_SPLIT_ENABLED || 162 (cfg->hds_config == ETHTOOL_TCP_DATA_SPLIT_ENABLED && 163 cfg->hds_thresh > len))) 164 skb_linearize(skb); 165 166 skb_tx_timestamp(skb); 167 if (unlikely(nsim_forward_skb(dev, peer_dev, 168 skb, rq, psp_ext) == NET_RX_DROP)) 169 goto out_drop_cnt; 170 171 if (!hrtimer_active(&rq->napi_timer)) 172 hrtimer_start(&rq->napi_timer, us_to_ktime(5), HRTIMER_MODE_REL); 173 174 rcu_read_unlock(); 175 dev_dstats_tx_add(dev, len); 176 return NETDEV_TX_OK; 177 178 out_drop_any: 179 dr = SKB_DROP_REASON_NOT_SPECIFIED; 180 out_drop_free: 181 kfree_skb_reason(skb, dr); 182 out_drop_cnt: 183 rcu_read_unlock(); 184 dev_dstats_tx_dropped(dev); 185 return NETDEV_TX_OK; 186 } 187 188 static int nsim_set_rx_mode(struct net_device *dev, 189 struct netdev_hw_addr_list *uc, 190 struct netdev_hw_addr_list *mc) 191 { 192 return 0; 193 } 194 195 static int nsim_change_mtu(struct net_device *dev, int new_mtu) 196 { 197 struct netdevsim *ns = netdev_priv(dev); 198 199 if (ns->xdp.prog && !ns->xdp.prog->aux->xdp_has_frags && 200 new_mtu > NSIM_XDP_MAX_MTU) 201 return -EBUSY; 202 203 WRITE_ONCE(dev->mtu, new_mtu); 204 205 return 0; 206 } 207 208 static int nsim_set_vf_mac(struct net_device *dev, int vf, u8 *mac) 209 { 210 struct netdevsim *ns = netdev_priv(dev); 211 struct nsim_dev *nsim_dev = ns->nsim_dev; 212 213 /* Only refuse multicast addresses, zero address can mean unset/any. */ 214 if (vf >= nsim_dev_get_vfs(nsim_dev) || is_multicast_ether_addr(mac)) 215 return -EINVAL; 216 memcpy(nsim_dev->vfconfigs[vf].vf_mac, mac, ETH_ALEN); 217 218 return 0; 219 } 220 221 static int nsim_set_vf_vlan(struct net_device *dev, int vf, 222 u16 vlan, u8 qos, __be16 vlan_proto) 223 { 224 struct netdevsim *ns = netdev_priv(dev); 225 struct nsim_dev *nsim_dev = ns->nsim_dev; 226 227 if (vf >= nsim_dev_get_vfs(nsim_dev) || vlan > 4095 || qos > 7) 228 return -EINVAL; 229 230 nsim_dev->vfconfigs[vf].vlan = vlan; 231 nsim_dev->vfconfigs[vf].qos = qos; 232 nsim_dev->vfconfigs[vf].vlan_proto = vlan_proto; 233 234 return 0; 235 } 236 237 static int nsim_set_vf_rate(struct net_device *dev, int vf, int min, int max) 238 { 239 struct netdevsim *ns = netdev_priv(dev); 240 struct nsim_dev *nsim_dev = ns->nsim_dev; 241 242 if (nsim_esw_mode_is_switchdev(ns->nsim_dev)) { 243 pr_err("Not supported in switchdev mode. Please use devlink API.\n"); 244 return -EOPNOTSUPP; 245 } 246 247 if (vf >= nsim_dev_get_vfs(nsim_dev)) 248 return -EINVAL; 249 250 nsim_dev->vfconfigs[vf].min_tx_rate = min; 251 nsim_dev->vfconfigs[vf].max_tx_rate = max; 252 253 return 0; 254 } 255 256 static int nsim_set_vf_spoofchk(struct net_device *dev, int vf, bool val) 257 { 258 struct netdevsim *ns = netdev_priv(dev); 259 struct nsim_dev *nsim_dev = ns->nsim_dev; 260 261 if (vf >= nsim_dev_get_vfs(nsim_dev)) 262 return -EINVAL; 263 nsim_dev->vfconfigs[vf].spoofchk_enabled = val; 264 265 return 0; 266 } 267 268 static int nsim_set_vf_rss_query_en(struct net_device *dev, int vf, bool val) 269 { 270 struct netdevsim *ns = netdev_priv(dev); 271 struct nsim_dev *nsim_dev = ns->nsim_dev; 272 273 if (vf >= nsim_dev_get_vfs(nsim_dev)) 274 return -EINVAL; 275 nsim_dev->vfconfigs[vf].rss_query_enabled = val; 276 277 return 0; 278 } 279 280 static int nsim_set_vf_trust(struct net_device *dev, int vf, bool val) 281 { 282 struct netdevsim *ns = netdev_priv(dev); 283 struct nsim_dev *nsim_dev = ns->nsim_dev; 284 285 if (vf >= nsim_dev_get_vfs(nsim_dev)) 286 return -EINVAL; 287 nsim_dev->vfconfigs[vf].trusted = val; 288 289 return 0; 290 } 291 292 static int 293 nsim_get_vf_config(struct net_device *dev, int vf, struct ifla_vf_info *ivi) 294 { 295 struct netdevsim *ns = netdev_priv(dev); 296 struct nsim_dev *nsim_dev = ns->nsim_dev; 297 298 if (vf >= nsim_dev_get_vfs(nsim_dev)) 299 return -EINVAL; 300 301 ivi->vf = vf; 302 ivi->linkstate = nsim_dev->vfconfigs[vf].link_state; 303 ivi->min_tx_rate = nsim_dev->vfconfigs[vf].min_tx_rate; 304 ivi->max_tx_rate = nsim_dev->vfconfigs[vf].max_tx_rate; 305 ivi->vlan = nsim_dev->vfconfigs[vf].vlan; 306 ivi->vlan_proto = nsim_dev->vfconfigs[vf].vlan_proto; 307 ivi->qos = nsim_dev->vfconfigs[vf].qos; 308 memcpy(&ivi->mac, nsim_dev->vfconfigs[vf].vf_mac, ETH_ALEN); 309 ivi->spoofchk = nsim_dev->vfconfigs[vf].spoofchk_enabled; 310 ivi->trusted = nsim_dev->vfconfigs[vf].trusted; 311 ivi->rss_query_en = nsim_dev->vfconfigs[vf].rss_query_enabled; 312 313 return 0; 314 } 315 316 static int nsim_set_vf_link_state(struct net_device *dev, int vf, int state) 317 { 318 struct netdevsim *ns = netdev_priv(dev); 319 struct nsim_dev *nsim_dev = ns->nsim_dev; 320 321 if (vf >= nsim_dev_get_vfs(nsim_dev)) 322 return -EINVAL; 323 324 switch (state) { 325 case IFLA_VF_LINK_STATE_AUTO: 326 case IFLA_VF_LINK_STATE_ENABLE: 327 case IFLA_VF_LINK_STATE_DISABLE: 328 break; 329 default: 330 return -EINVAL; 331 } 332 333 nsim_dev->vfconfigs[vf].link_state = state; 334 335 return 0; 336 } 337 338 static int 339 nsim_set_features(struct net_device *dev, netdev_features_t features) 340 { 341 struct netdevsim *ns = netdev_priv(dev); 342 343 if ((dev->features & NETIF_F_HW_TC) > (features & NETIF_F_HW_TC)) 344 return nsim_bpf_disable_tc(ns); 345 346 return 0; 347 } 348 349 static int nsim_get_iflink(const struct net_device *dev) 350 { 351 struct netdevsim *nsim, *peer; 352 int iflink; 353 354 nsim = netdev_priv(dev); 355 356 rcu_read_lock(); 357 peer = rcu_dereference(nsim->peer); 358 iflink = peer ? READ_ONCE(peer->netdev->ifindex) : 359 READ_ONCE(dev->ifindex); 360 rcu_read_unlock(); 361 362 return iflink; 363 } 364 365 static int nsim_rcv(struct nsim_rq *rq, int budget) 366 { 367 struct net_device *dev = rq->napi.dev; 368 struct bpf_prog *xdp_prog; 369 struct netdevsim *ns; 370 struct sk_buff *skb; 371 unsigned int skblen; 372 int i, ret; 373 374 ns = netdev_priv(dev); 375 xdp_prog = READ_ONCE(ns->xdp.prog); 376 377 for (i = 0; i < budget; i++) { 378 if (skb_queue_empty(&rq->skb_queue)) 379 break; 380 381 skb = skb_dequeue(&rq->skb_queue); 382 383 if (xdp_prog) { 384 /* skb might be freed directly by XDP, save the len */ 385 skblen = skb->len; 386 387 if (skb->ip_summed == CHECKSUM_PARTIAL) 388 skb_checksum_help(skb); 389 ret = do_xdp_generic(xdp_prog, &skb); 390 if (ret != XDP_PASS) { 391 dev_dstats_rx_add(dev, skblen); 392 continue; 393 } 394 } 395 396 /* skb might be discard at netif_receive_skb, save the len */ 397 dev_dstats_rx_add(dev, skb->len); 398 napi_gro_receive(&rq->napi, skb); 399 } 400 401 nsim_start_peer_tx_queue(dev, rq); 402 return i; 403 } 404 405 static int nsim_poll(struct napi_struct *napi, int budget) 406 { 407 struct nsim_rq *rq = container_of(napi, struct nsim_rq, napi); 408 int done; 409 410 done = nsim_rcv(rq, budget); 411 if (done < budget) 412 napi_complete_done(napi, done); 413 414 return done; 415 } 416 417 static int nsim_create_page_pool(struct page_pool **p, struct napi_struct *napi) 418 { 419 struct page_pool_params params = { 420 .order = 0, 421 .pool_size = NSIM_RING_SIZE, 422 .nid = NUMA_NO_NODE, 423 .dev = &napi->dev->dev, 424 .napi = napi, 425 .dma_dir = DMA_BIDIRECTIONAL, 426 .netdev = napi->dev, 427 }; 428 struct page_pool *pool; 429 430 pool = page_pool_create(¶ms); 431 if (IS_ERR(pool)) 432 return PTR_ERR(pool); 433 434 *p = pool; 435 return 0; 436 } 437 438 static int nsim_init_napi(struct netdevsim *ns) 439 { 440 struct net_device *dev = ns->netdev; 441 struct nsim_rq *rq; 442 int err, i; 443 444 for (i = 0; i < dev->num_rx_queues; i++) { 445 rq = ns->rq[i]; 446 447 netif_napi_add_config_locked(dev, &rq->napi, nsim_poll, i); 448 } 449 450 for (i = 0; i < dev->num_rx_queues; i++) { 451 rq = ns->rq[i]; 452 453 err = nsim_create_page_pool(&rq->page_pool, &rq->napi); 454 if (err) 455 goto err_pp_destroy; 456 } 457 458 return 0; 459 460 err_pp_destroy: 461 while (i--) { 462 page_pool_destroy(ns->rq[i]->page_pool); 463 ns->rq[i]->page_pool = NULL; 464 } 465 466 for (i = 0; i < dev->num_rx_queues; i++) 467 __netif_napi_del_locked(&ns->rq[i]->napi); 468 469 return err; 470 } 471 472 static enum hrtimer_restart nsim_napi_schedule(struct hrtimer *timer) 473 { 474 struct nsim_rq *rq; 475 476 rq = container_of(timer, struct nsim_rq, napi_timer); 477 napi_schedule(&rq->napi); 478 479 return HRTIMER_NORESTART; 480 } 481 482 static void nsim_rq_timer_init(struct nsim_rq *rq) 483 { 484 hrtimer_setup(&rq->napi_timer, nsim_napi_schedule, CLOCK_MONOTONIC, 485 HRTIMER_MODE_REL); 486 } 487 488 static void nsim_enable_napi(struct netdevsim *ns) 489 { 490 struct net_device *dev = ns->netdev; 491 int i; 492 493 for (i = 0; i < dev->num_rx_queues; i++) { 494 struct nsim_rq *rq = ns->rq[i]; 495 496 netif_queue_set_napi(dev, i, NETDEV_QUEUE_TYPE_RX, &rq->napi); 497 napi_enable_locked(&rq->napi); 498 } 499 } 500 501 static int nsim_open(struct net_device *dev) 502 { 503 struct netdevsim *ns = netdev_priv(dev); 504 struct netdevsim *peer; 505 int err; 506 507 netdev_assert_locked(dev); 508 509 err = nsim_init_napi(ns); 510 if (err) 511 return err; 512 513 nsim_enable_napi(ns); 514 515 peer = rtnl_dereference(ns->peer); 516 if (peer && netif_running(peer->netdev)) { 517 netif_carrier_on(dev); 518 netif_carrier_on(peer->netdev); 519 } 520 521 return 0; 522 } 523 524 static void nsim_del_napi(struct netdevsim *ns) 525 { 526 struct net_device *dev = ns->netdev; 527 int i; 528 529 for (i = 0; i < dev->num_rx_queues; i++) { 530 struct nsim_rq *rq = ns->rq[i]; 531 532 napi_disable_locked(&rq->napi); 533 __netif_napi_del_locked(&rq->napi); 534 } 535 synchronize_net(); 536 537 for (i = 0; i < dev->num_rx_queues; i++) { 538 page_pool_destroy(ns->rq[i]->page_pool); 539 ns->rq[i]->page_pool = NULL; 540 } 541 } 542 543 static int nsim_stop(struct net_device *dev) 544 { 545 struct netdevsim *ns = netdev_priv(dev); 546 struct netdevsim *peer; 547 548 netdev_assert_locked(dev); 549 550 netif_carrier_off(dev); 551 peer = rtnl_dereference(ns->peer); 552 if (peer) 553 netif_carrier_off(peer->netdev); 554 555 nsim_del_napi(ns); 556 557 return 0; 558 } 559 560 static int nsim_vlan_rx_add_vid(struct net_device *dev, __be16 proto, u16 vid) 561 { 562 struct netdevsim *ns = netdev_priv(dev); 563 564 if (vid >= VLAN_N_VID) 565 return -EINVAL; 566 567 if (proto == htons(ETH_P_8021Q)) 568 WARN_ON_ONCE(test_and_set_bit(vid, ns->vlan.ctag)); 569 else if (proto == htons(ETH_P_8021AD)) 570 WARN_ON_ONCE(test_and_set_bit(vid, ns->vlan.stag)); 571 572 return 0; 573 } 574 575 static int nsim_vlan_rx_kill_vid(struct net_device *dev, __be16 proto, u16 vid) 576 { 577 struct netdevsim *ns = netdev_priv(dev); 578 579 if (vid >= VLAN_N_VID) 580 return -EINVAL; 581 582 if (proto == htons(ETH_P_8021Q)) 583 WARN_ON_ONCE(!test_and_clear_bit(vid, ns->vlan.ctag)); 584 else if (proto == htons(ETH_P_8021AD)) 585 WARN_ON_ONCE(!test_and_clear_bit(vid, ns->vlan.stag)); 586 587 return 0; 588 } 589 590 static int nsim_shaper_set(struct net_shaper_binding *binding, 591 const struct net_shaper *shaper, 592 struct netlink_ext_ack *extack) 593 { 594 return 0; 595 } 596 597 static int nsim_shaper_del(struct net_shaper_binding *binding, 598 const struct net_shaper_handle *handle, 599 struct netlink_ext_ack *extack) 600 { 601 return 0; 602 } 603 604 static int nsim_shaper_group(struct net_shaper_binding *binding, 605 int leaves_count, 606 const struct net_shaper *leaves, 607 const struct net_shaper *root, 608 struct netlink_ext_ack *extack) 609 { 610 return 0; 611 } 612 613 static void nsim_shaper_cap(struct net_shaper_binding *binding, 614 enum net_shaper_scope scope, 615 unsigned long *flags) 616 { 617 *flags = ULONG_MAX; 618 } 619 620 static const struct net_shaper_ops nsim_shaper_ops = { 621 .set = nsim_shaper_set, 622 .delete = nsim_shaper_del, 623 .group = nsim_shaper_group, 624 .capabilities = nsim_shaper_cap, 625 }; 626 627 static const struct net_device_ops nsim_netdev_ops = { 628 .ndo_start_xmit = nsim_start_xmit, 629 .ndo_set_rx_mode_async = nsim_set_rx_mode, 630 .ndo_set_mac_address = eth_mac_addr, 631 .ndo_validate_addr = eth_validate_addr, 632 .ndo_change_mtu = nsim_change_mtu, 633 .ndo_set_vf_mac = nsim_set_vf_mac, 634 .ndo_set_vf_vlan = nsim_set_vf_vlan, 635 .ndo_set_vf_rate = nsim_set_vf_rate, 636 .ndo_set_vf_spoofchk = nsim_set_vf_spoofchk, 637 .ndo_set_vf_trust = nsim_set_vf_trust, 638 .ndo_get_vf_config = nsim_get_vf_config, 639 .ndo_set_vf_link_state = nsim_set_vf_link_state, 640 .ndo_set_vf_rss_query_en = nsim_set_vf_rss_query_en, 641 .ndo_setup_tc = nsim_setup_tc, 642 .ndo_set_features = nsim_set_features, 643 .ndo_get_iflink = nsim_get_iflink, 644 .ndo_bpf = nsim_bpf, 645 .ndo_open = nsim_open, 646 .ndo_stop = nsim_stop, 647 .ndo_vlan_rx_add_vid = nsim_vlan_rx_add_vid, 648 .ndo_vlan_rx_kill_vid = nsim_vlan_rx_kill_vid, 649 .net_shaper_ops = &nsim_shaper_ops, 650 }; 651 652 static const struct net_device_ops nsim_vf_netdev_ops = { 653 .ndo_start_xmit = nsim_start_xmit, 654 .ndo_set_rx_mode_async = nsim_set_rx_mode, 655 .ndo_set_mac_address = eth_mac_addr, 656 .ndo_validate_addr = eth_validate_addr, 657 .ndo_change_mtu = nsim_change_mtu, 658 .ndo_setup_tc = nsim_setup_tc, 659 .ndo_set_features = nsim_set_features, 660 .ndo_vlan_rx_add_vid = nsim_vlan_rx_add_vid, 661 .ndo_vlan_rx_kill_vid = nsim_vlan_rx_kill_vid, 662 }; 663 664 /* We don't have true per-queue stats, yet, so do some random fakery here. 665 * Only report stuff for queue 0. 666 */ 667 static void nsim_get_queue_stats_rx(struct net_device *dev, int idx, 668 struct netdev_queue_stats_rx *stats) 669 { 670 struct rtnl_link_stats64 rtstats = {}; 671 672 if (!idx) 673 dev_get_stats(dev, &rtstats); 674 675 stats->packets = rtstats.rx_packets - !!rtstats.rx_packets; 676 stats->bytes = rtstats.rx_bytes; 677 } 678 679 static void nsim_get_queue_stats_tx(struct net_device *dev, int idx, 680 struct netdev_queue_stats_tx *stats) 681 { 682 struct rtnl_link_stats64 rtstats = {}; 683 684 if (!idx) 685 dev_get_stats(dev, &rtstats); 686 687 stats->packets = rtstats.tx_packets - !!rtstats.tx_packets; 688 stats->bytes = rtstats.tx_bytes; 689 } 690 691 static void nsim_get_base_stats(struct net_device *dev, 692 struct netdev_queue_stats_rx *rx, 693 struct netdev_queue_stats_tx *tx) 694 { 695 struct rtnl_link_stats64 rtstats = {}; 696 697 dev_get_stats(dev, &rtstats); 698 699 rx->packets = !!rtstats.rx_packets; 700 rx->bytes = 0; 701 tx->packets = !!rtstats.tx_packets; 702 tx->bytes = 0; 703 } 704 705 static const struct netdev_stat_ops nsim_stat_ops = { 706 .get_queue_stats_tx = nsim_get_queue_stats_tx, 707 .get_queue_stats_rx = nsim_get_queue_stats_rx, 708 .get_base_stats = nsim_get_base_stats, 709 }; 710 711 static struct nsim_rq *nsim_queue_alloc(void) 712 { 713 struct nsim_rq *rq; 714 715 rq = kzalloc_obj(*rq, GFP_KERNEL_ACCOUNT); 716 if (!rq) 717 return NULL; 718 719 skb_queue_head_init(&rq->skb_queue); 720 nsim_rq_timer_init(rq); 721 return rq; 722 } 723 724 static void nsim_queue_free(struct net_device *dev, struct nsim_rq *rq) 725 { 726 hrtimer_cancel(&rq->napi_timer); 727 728 if (rq->skb_queue.qlen) { 729 local_bh_disable(); 730 dev_dstats_rx_dropped_add(dev, rq->skb_queue.qlen); 731 local_bh_enable(); 732 } 733 734 skb_queue_purge_reason(&rq->skb_queue, SKB_DROP_REASON_QUEUE_PURGE); 735 kfree(rq); 736 } 737 738 /* Queue reset mode is controlled by ns->rq_reset_mode. 739 * - normal - new NAPI new pool (old NAPI enabled when new added) 740 * - mode 1 - allocate new pool (NAPI is only disabled / enabled) 741 * - mode 2 - new NAPI new pool (old NAPI removed before new added) 742 * - mode 3 - new NAPI new pool (old NAPI disabled when new added) 743 */ 744 struct nsim_queue_mem { 745 struct nsim_rq *rq; 746 struct page_pool *pp; 747 }; 748 749 static int 750 nsim_queue_mem_alloc(struct net_device *dev, 751 struct netdev_queue_config *qcfg, 752 void *per_queue_mem, int idx) 753 { 754 struct nsim_queue_mem *qmem = per_queue_mem; 755 struct netdevsim *ns = netdev_priv(dev); 756 int err; 757 758 if (ns->rq_reset_mode > 3) 759 return -EINVAL; 760 761 if (ns->rq_reset_mode == 1) { 762 if (!netif_running(ns->netdev)) 763 return -ENETDOWN; 764 return nsim_create_page_pool(&qmem->pp, &ns->rq[idx]->napi); 765 } 766 767 qmem->rq = nsim_queue_alloc(); 768 if (!qmem->rq) 769 return -ENOMEM; 770 771 err = nsim_create_page_pool(&qmem->rq->page_pool, &qmem->rq->napi); 772 if (err) 773 goto err_free; 774 775 if (!ns->rq_reset_mode) 776 netif_napi_add_config_locked(dev, &qmem->rq->napi, nsim_poll, 777 idx); 778 779 return 0; 780 781 err_free: 782 nsim_queue_free(dev, qmem->rq); 783 return err; 784 } 785 786 static void nsim_queue_mem_free(struct net_device *dev, void *per_queue_mem) 787 { 788 struct nsim_queue_mem *qmem = per_queue_mem; 789 struct netdevsim *ns = netdev_priv(dev); 790 791 page_pool_destroy(qmem->pp); 792 if (qmem->rq) { 793 if (!ns->rq_reset_mode) 794 netif_napi_del_locked(&qmem->rq->napi); 795 page_pool_destroy(qmem->rq->page_pool); 796 nsim_queue_free(dev, qmem->rq); 797 } 798 } 799 800 static int 801 nsim_queue_start(struct net_device *dev, struct netdev_queue_config *qcfg, 802 void *per_queue_mem, int idx) 803 { 804 struct nsim_queue_mem *qmem = per_queue_mem; 805 struct netdevsim *ns = netdev_priv(dev); 806 807 netdev_assert_locked(dev); 808 809 if (ns->rq_reset_mode == 1) { 810 ns->rq[idx]->page_pool = qmem->pp; 811 napi_enable_locked(&ns->rq[idx]->napi); 812 return 0; 813 } 814 815 /* netif_napi_add()/_del() should normally be called from alloc/free, 816 * here we want to test various call orders. 817 */ 818 if (ns->rq_reset_mode == 2) { 819 netif_napi_del_locked(&ns->rq[idx]->napi); 820 netif_napi_add_config_locked(dev, &qmem->rq->napi, nsim_poll, 821 idx); 822 } else if (ns->rq_reset_mode == 3) { 823 netif_napi_add_config_locked(dev, &qmem->rq->napi, nsim_poll, 824 idx); 825 netif_napi_del_locked(&ns->rq[idx]->napi); 826 } 827 828 ns->rq[idx] = qmem->rq; 829 napi_enable_locked(&ns->rq[idx]->napi); 830 831 return 0; 832 } 833 834 static int nsim_queue_stop(struct net_device *dev, void *per_queue_mem, int idx) 835 { 836 struct nsim_queue_mem *qmem = per_queue_mem; 837 struct netdevsim *ns = netdev_priv(dev); 838 839 netdev_assert_locked(dev); 840 841 napi_disable_locked(&ns->rq[idx]->napi); 842 843 if (ns->rq_reset_mode == 1) { 844 qmem->pp = ns->rq[idx]->page_pool; 845 page_pool_disable_direct_recycling(qmem->pp); 846 } else { 847 qmem->rq = ns->rq[idx]; 848 } 849 850 return 0; 851 } 852 853 static const struct netdev_queue_mgmt_ops nsim_queue_mgmt_ops = { 854 .ndo_queue_mem_size = sizeof(struct nsim_queue_mem), 855 .ndo_queue_mem_alloc = nsim_queue_mem_alloc, 856 .ndo_queue_mem_free = nsim_queue_mem_free, 857 .ndo_queue_start = nsim_queue_start, 858 .ndo_queue_stop = nsim_queue_stop, 859 }; 860 861 static ssize_t 862 nsim_qreset_write(struct file *file, const char __user *data, 863 size_t count, loff_t *ppos) 864 { 865 struct netdevsim *ns = file->private_data; 866 unsigned int queue, mode; 867 char buf[32]; 868 ssize_t ret; 869 870 if (count >= sizeof(buf)) 871 return -EINVAL; 872 if (copy_from_user(buf, data, count)) 873 return -EFAULT; 874 buf[count] = '\0'; 875 876 ret = sscanf(buf, "%u %u", &queue, &mode); 877 if (ret != 2) 878 return -EINVAL; 879 880 netdev_lock(ns->netdev); 881 if (queue >= ns->netdev->real_num_rx_queues) { 882 ret = -EINVAL; 883 goto exit_unlock; 884 } 885 886 ns->rq_reset_mode = mode; 887 ret = netdev_rx_queue_restart(ns->netdev, queue); 888 ns->rq_reset_mode = 0; 889 if (ret) 890 goto exit_unlock; 891 892 ret = count; 893 exit_unlock: 894 netdev_unlock(ns->netdev); 895 return ret; 896 } 897 898 static const struct file_operations nsim_qreset_fops = { 899 .open = simple_open, 900 .write = nsim_qreset_write, 901 .owner = THIS_MODULE, 902 }; 903 904 static ssize_t 905 nsim_pp_hold_read(struct file *file, char __user *data, 906 size_t count, loff_t *ppos) 907 { 908 struct netdevsim *ns = file->private_data; 909 char buf[3] = "n\n"; 910 911 if (ns->page) 912 buf[0] = 'y'; 913 914 return simple_read_from_buffer(data, count, ppos, buf, 2); 915 } 916 917 static ssize_t 918 nsim_pp_hold_write(struct file *file, const char __user *data, 919 size_t count, loff_t *ppos) 920 { 921 struct netdevsim *ns = file->private_data; 922 ssize_t ret; 923 bool val; 924 925 ret = kstrtobool_from_user(data, count, &val); 926 if (ret) 927 return ret; 928 929 rtnl_lock(); 930 ret = count; 931 if (val == !!ns->page) 932 goto exit; 933 934 if (!netif_running(ns->netdev) && val) { 935 ret = -ENETDOWN; 936 } else if (val) { 937 ns->page = page_pool_dev_alloc_pages(ns->rq[0]->page_pool); 938 if (!ns->page) 939 ret = -ENOMEM; 940 } else { 941 page_pool_put_full_page(pp_page_to_nmdesc(ns->page)->pp, 942 ns->page, false); 943 ns->page = NULL; 944 } 945 946 exit: 947 rtnl_unlock(); 948 return ret; 949 } 950 951 static const struct file_operations nsim_pp_hold_fops = { 952 .open = simple_open, 953 .read = nsim_pp_hold_read, 954 .write = nsim_pp_hold_write, 955 .llseek = generic_file_llseek, 956 .owner = THIS_MODULE, 957 }; 958 959 static int nsim_vlan_show(struct seq_file *s, void *data) 960 { 961 struct netdevsim *ns = s->private; 962 int vid; 963 964 for_each_set_bit(vid, ns->vlan.ctag, VLAN_N_VID) 965 seq_printf(s, "ctag %d\n", vid); 966 for_each_set_bit(vid, ns->vlan.stag, VLAN_N_VID) 967 seq_printf(s, "stag %d\n", vid); 968 969 return 0; 970 } 971 DEFINE_SHOW_ATTRIBUTE(nsim_vlan); 972 973 static void nsim_setup(struct net_device *dev) 974 { 975 ether_setup(dev); 976 eth_hw_addr_random(dev); 977 978 dev->flags &= ~IFF_MULTICAST; 979 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE; 980 dev->features |= NETIF_F_HIGHDMA | 981 NETIF_F_SG | 982 NETIF_F_FRAGLIST | 983 NETIF_F_HW_CSUM | 984 NETIF_F_LRO | 985 NETIF_F_TSO | 986 NETIF_F_HW_VLAN_CTAG_FILTER | 987 NETIF_F_HW_VLAN_STAG_FILTER; 988 dev->hw_features |= NETIF_F_HW_TC | 989 NETIF_F_SG | 990 NETIF_F_FRAGLIST | 991 NETIF_F_HW_CSUM | 992 NETIF_F_LRO | 993 NETIF_F_TSO | 994 NETIF_F_LOOPBACK | 995 NETIF_F_HW_VLAN_CTAG_FILTER | 996 NETIF_F_HW_VLAN_STAG_FILTER; 997 dev->pcpu_stat_type = NETDEV_PCPU_STAT_DSTATS; 998 dev->max_mtu = ETH_MAX_MTU; 999 dev->xdp_features = NETDEV_XDP_ACT_BASIC | NETDEV_XDP_ACT_HW_OFFLOAD; 1000 } 1001 1002 static int nsim_queue_init(struct netdevsim *ns) 1003 { 1004 struct net_device *dev = ns->netdev; 1005 int i; 1006 1007 ns->rq = kzalloc_objs(*ns->rq, dev->num_rx_queues, GFP_KERNEL_ACCOUNT); 1008 if (!ns->rq) 1009 return -ENOMEM; 1010 1011 for (i = 0; i < dev->num_rx_queues; i++) { 1012 ns->rq[i] = nsim_queue_alloc(); 1013 if (!ns->rq[i]) 1014 goto err_free_prev; 1015 } 1016 1017 return 0; 1018 1019 err_free_prev: 1020 while (i--) 1021 kfree(ns->rq[i]); 1022 kfree(ns->rq); 1023 return -ENOMEM; 1024 } 1025 1026 static void nsim_queue_uninit(struct netdevsim *ns) 1027 { 1028 struct net_device *dev = ns->netdev; 1029 int i; 1030 1031 for (i = 0; i < dev->num_rx_queues; i++) 1032 nsim_queue_free(dev, ns->rq[i]); 1033 1034 kfree(ns->rq); 1035 ns->rq = NULL; 1036 } 1037 1038 static int nsim_init_netdevsim(struct netdevsim *ns) 1039 { 1040 struct netdevsim *peer; 1041 struct mock_phc *phc; 1042 int err; 1043 1044 phc = mock_phc_create(&ns->nsim_bus_dev->dev); 1045 if (IS_ERR(phc)) 1046 return PTR_ERR(phc); 1047 1048 ns->phc = phc; 1049 ns->netdev->netdev_ops = &nsim_netdev_ops; 1050 ns->netdev->stat_ops = &nsim_stat_ops; 1051 ns->netdev->queue_mgmt_ops = &nsim_queue_mgmt_ops; 1052 netdev_lockdep_set_classes(ns->netdev); 1053 1054 err = nsim_udp_tunnels_info_create(ns->nsim_dev, ns->netdev); 1055 if (err) 1056 goto err_phc_destroy; 1057 1058 rtnl_lock(); 1059 err = nsim_queue_init(ns); 1060 if (err) 1061 goto err_utn_destroy; 1062 1063 err = nsim_bpf_init(ns); 1064 if (err) 1065 goto err_rq_destroy; 1066 1067 nsim_macsec_init(ns); 1068 nsim_ipsec_init(ns); 1069 1070 err = register_netdevice(ns->netdev); 1071 if (err) 1072 goto err_ipsec_teardown; 1073 rtnl_unlock(); 1074 1075 err = nsim_psp_init(ns); 1076 if (err) 1077 goto err_unregister_netdev; 1078 1079 if (IS_ENABLED(CONFIG_DEBUG_NET)) { 1080 ns->nb.notifier_call = netdev_debug_event; 1081 if (register_netdevice_notifier_dev_net(ns->netdev, &ns->nb, 1082 &ns->nn)) 1083 ns->nb.notifier_call = NULL; 1084 } 1085 1086 return 0; 1087 1088 err_unregister_netdev: 1089 rtnl_lock(); 1090 peer = rtnl_dereference(ns->peer); 1091 if (peer) 1092 RCU_INIT_POINTER(peer->peer, NULL); 1093 RCU_INIT_POINTER(ns->peer, NULL); 1094 unregister_netdevice(ns->netdev); 1095 err_ipsec_teardown: 1096 nsim_ipsec_teardown(ns); 1097 nsim_macsec_teardown(ns); 1098 nsim_bpf_uninit(ns); 1099 err_rq_destroy: 1100 nsim_queue_uninit(ns); 1101 err_utn_destroy: 1102 rtnl_unlock(); 1103 nsim_udp_tunnels_info_destroy(ns->netdev); 1104 err_phc_destroy: 1105 mock_phc_destroy(ns->phc); 1106 return err; 1107 } 1108 1109 static int nsim_init_netdevsim_vf(struct netdevsim *ns) 1110 { 1111 int err; 1112 1113 ns->netdev->netdev_ops = &nsim_vf_netdev_ops; 1114 rtnl_lock(); 1115 err = register_netdevice(ns->netdev); 1116 rtnl_unlock(); 1117 return err; 1118 } 1119 1120 static void nsim_exit_netdevsim(struct netdevsim *ns) 1121 { 1122 nsim_udp_tunnels_info_destroy(ns->netdev); 1123 mock_phc_destroy(ns->phc); 1124 } 1125 1126 struct netdevsim *nsim_create(struct nsim_dev *nsim_dev, 1127 struct nsim_dev_port *nsim_dev_port, 1128 u8 perm_addr[ETH_ALEN]) 1129 { 1130 struct net_device *dev; 1131 struct netdevsim *ns; 1132 int err; 1133 1134 dev = alloc_netdev_mq(sizeof(*ns), "eth%d", NET_NAME_UNKNOWN, nsim_setup, 1135 nsim_dev->nsim_bus_dev->num_queues); 1136 if (!dev) 1137 return ERR_PTR(-ENOMEM); 1138 1139 if (perm_addr) 1140 memcpy(dev->perm_addr, perm_addr, ETH_ALEN); 1141 1142 dev_net_set(dev, nsim_dev_net(nsim_dev)); 1143 ns = netdev_priv(dev); 1144 ns->netdev = dev; 1145 ns->nsim_dev = nsim_dev; 1146 ns->nsim_dev_port = nsim_dev_port; 1147 ns->nsim_bus_dev = nsim_dev->nsim_bus_dev; 1148 SET_NETDEV_DEV(dev, &ns->nsim_bus_dev->dev); 1149 SET_NETDEV_DEVLINK_PORT(dev, &nsim_dev_port->devlink_port); 1150 nsim_ethtool_init(ns); 1151 if (nsim_dev_port_is_pf(nsim_dev_port)) 1152 err = nsim_init_netdevsim(ns); 1153 else 1154 err = nsim_init_netdevsim_vf(ns); 1155 if (err) 1156 goto err_free_netdev; 1157 1158 ns->pp_dfs = debugfs_create_file("pp_hold", 0600, nsim_dev_port->ddir, 1159 ns, &nsim_pp_hold_fops); 1160 ns->qr_dfs = debugfs_create_file("queue_reset", 0200, 1161 nsim_dev_port->ddir, ns, 1162 &nsim_qreset_fops); 1163 ns->vlan_dfs = debugfs_create_file("vlan", 0400, nsim_dev_port->ddir, 1164 ns, &nsim_vlan_fops); 1165 return ns; 1166 1167 err_free_netdev: 1168 free_netdev(dev); 1169 return ERR_PTR(err); 1170 } 1171 1172 void nsim_destroy(struct netdevsim *ns) 1173 { 1174 struct net_device *dev = ns->netdev; 1175 struct netdevsim *peer; 1176 u16 vid; 1177 1178 debugfs_remove(ns->vlan_dfs); 1179 debugfs_remove(ns->qr_dfs); 1180 debugfs_remove(ns->pp_dfs); 1181 1182 if (ns->nb.notifier_call) 1183 unregister_netdevice_notifier_dev_net(ns->netdev, &ns->nb, 1184 &ns->nn); 1185 1186 if (nsim_dev_port_is_pf(ns->nsim_dev_port)) 1187 nsim_psp_uninit(ns); 1188 1189 rtnl_lock(); 1190 peer = rtnl_dereference(ns->peer); 1191 if (peer) 1192 RCU_INIT_POINTER(peer->peer, NULL); 1193 RCU_INIT_POINTER(ns->peer, NULL); 1194 unregister_netdevice(dev); 1195 if (nsim_dev_port_is_pf(ns->nsim_dev_port)) { 1196 nsim_macsec_teardown(ns); 1197 nsim_ipsec_teardown(ns); 1198 nsim_bpf_uninit(ns); 1199 nsim_queue_uninit(ns); 1200 } 1201 rtnl_unlock(); 1202 if (nsim_dev_port_is_pf(ns->nsim_dev_port)) 1203 nsim_exit_netdevsim(ns); 1204 1205 for_each_set_bit(vid, ns->vlan.ctag, VLAN_N_VID) 1206 WARN_ON_ONCE(1); 1207 for_each_set_bit(vid, ns->vlan.stag, VLAN_N_VID) 1208 WARN_ON_ONCE(1); 1209 1210 /* Put this intentionally late to exercise the orphaning path */ 1211 if (ns->page) { 1212 page_pool_put_full_page(pp_page_to_nmdesc(ns->page)->pp, 1213 ns->page, false); 1214 ns->page = NULL; 1215 } 1216 1217 free_netdev(dev); 1218 } 1219 1220 bool netdev_is_nsim(struct net_device *dev) 1221 { 1222 return dev->netdev_ops == &nsim_netdev_ops; 1223 } 1224 1225 static int nsim_validate(struct nlattr *tb[], struct nlattr *data[], 1226 struct netlink_ext_ack *extack) 1227 { 1228 NL_SET_ERR_MSG_MOD(extack, 1229 "Please use: echo \"[ID] [PORT_COUNT] [NUM_QUEUES]\" > /sys/bus/netdevsim/new_device"); 1230 return -EOPNOTSUPP; 1231 } 1232 1233 static struct rtnl_link_ops nsim_link_ops __read_mostly = { 1234 .kind = DRV_NAME, 1235 .validate = nsim_validate, 1236 }; 1237 1238 static int __init nsim_module_init(void) 1239 { 1240 int err; 1241 1242 err = nsim_dev_init(); 1243 if (err) 1244 return err; 1245 1246 err = nsim_bus_init(); 1247 if (err) 1248 goto err_dev_exit; 1249 1250 err = rtnl_link_register(&nsim_link_ops); 1251 if (err) 1252 goto err_bus_exit; 1253 1254 return 0; 1255 1256 err_bus_exit: 1257 nsim_bus_exit(); 1258 err_dev_exit: 1259 nsim_dev_exit(); 1260 return err; 1261 } 1262 1263 static void __exit nsim_module_exit(void) 1264 { 1265 rtnl_link_unregister(&nsim_link_ops); 1266 nsim_bus_exit(); 1267 nsim_dev_exit(); 1268 } 1269 1270 module_init(nsim_module_init); 1271 module_exit(nsim_module_exit); 1272 MODULE_LICENSE("GPL"); 1273 MODULE_DESCRIPTION("Simulated networking device for testing"); 1274 MODULE_ALIAS_RTNL_LINK(DRV_NAME); 1275