1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Monitoring code for network dropped packet alerts 4 * 5 * Copyright (C) 2009 Neil Horman <nhorman@tuxdriver.com> 6 */ 7 8 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 9 10 #include <linux/netdevice.h> 11 #include <linux/etherdevice.h> 12 #include <linux/string.h> 13 #include <linux/if_arp.h> 14 #include <linux/inetdevice.h> 15 #include <linux/inet.h> 16 #include <linux/interrupt.h> 17 #include <linux/netpoll.h> 18 #include <linux/sched.h> 19 #include <linux/delay.h> 20 #include <linux/types.h> 21 #include <linux/workqueue.h> 22 #include <linux/netlink.h> 23 #include <linux/net_dropmon.h> 24 #include <linux/bitfield.h> 25 #include <linux/percpu.h> 26 #include <linux/timer.h> 27 #include <linux/bitops.h> 28 #include <linux/slab.h> 29 #include <linux/module.h> 30 #include <net/genetlink.h> 31 #include <net/netevent.h> 32 #include <net/flow_offload.h> 33 #include <net/dropreason.h> 34 #include <net/devlink.h> 35 36 #include <trace/events/skb.h> 37 #include <trace/events/napi.h> 38 #include <trace/events/devlink.h> 39 40 #include <linux/unaligned.h> 41 42 #define TRACE_ON 1 43 #define TRACE_OFF 0 44 45 /* 46 * Globals, our netlink socket pointer 47 * and the work handle that will send up 48 * netlink alerts 49 */ 50 static int trace_state = TRACE_OFF; 51 static bool monitor_hw; 52 53 /* net_dm_mutex 54 * 55 * An overall lock guarding every operation coming from userspace. 56 */ 57 static DEFINE_MUTEX(net_dm_mutex); 58 59 struct net_dm_stats { 60 u64_stats_t dropped; 61 struct u64_stats_sync syncp; 62 }; 63 64 #define NET_DM_MAX_HW_TRAP_NAME_LEN 40 65 66 struct net_dm_hw_entry { 67 char trap_name[NET_DM_MAX_HW_TRAP_NAME_LEN]; 68 u32 count; 69 }; 70 71 struct net_dm_hw_entries { 72 u32 num_entries; 73 struct net_dm_hw_entry entries[]; 74 }; 75 76 struct per_cpu_dm_data { 77 raw_spinlock_t lock; /* Protects 'skb', 'hw_entries' and 78 * 'send_timer' 79 */ 80 union { 81 struct sk_buff *skb; 82 struct net_dm_hw_entries *hw_entries; 83 }; 84 struct sk_buff_head drop_queue; 85 struct work_struct dm_alert_work; 86 struct timer_list send_timer; 87 struct net_dm_stats stats; 88 }; 89 90 struct dm_hw_stat_delta { 91 unsigned long last_rx; 92 unsigned long last_drop_val; 93 struct rcu_head rcu; 94 }; 95 96 static struct genl_family net_drop_monitor_family; 97 98 static DEFINE_PER_CPU(struct per_cpu_dm_data, dm_cpu_data); 99 static DEFINE_PER_CPU(struct per_cpu_dm_data, dm_hw_cpu_data); 100 101 static int dm_hit_limit = 64; 102 static int dm_delay = 1; 103 static unsigned long dm_hw_check_delta = 2*HZ; 104 105 static enum net_dm_alert_mode net_dm_alert_mode = NET_DM_ALERT_MODE_SUMMARY; 106 static u32 net_dm_trunc_len; 107 static u32 net_dm_queue_len = 1000; 108 109 struct net_dm_alert_ops { 110 void (*kfree_skb_probe)(void *ignore, struct sk_buff *skb, 111 void *location, 112 enum skb_drop_reason reason, 113 const struct sock *rx_sk); 114 void (*napi_poll_probe)(void *ignore, struct napi_struct *napi, 115 int work, int budget); 116 void (*work_item_func)(struct work_struct *work); 117 void (*hw_work_item_func)(struct work_struct *work); 118 void (*hw_trap_probe)(void *ignore, const struct devlink *devlink, 119 struct sk_buff *skb, 120 const struct devlink_trap_metadata *metadata); 121 }; 122 123 struct net_dm_skb_cb { 124 union { 125 struct devlink_trap_metadata *hw_metadata; 126 void *pc; 127 }; 128 enum skb_drop_reason reason; 129 }; 130 131 #define NET_DM_SKB_CB(__skb) ((struct net_dm_skb_cb *)&((__skb)->cb[0])) 132 133 static struct sk_buff *reset_per_cpu_data(struct per_cpu_dm_data *data) 134 { 135 size_t al; 136 struct net_dm_alert_msg *msg; 137 struct nlattr *nla; 138 struct sk_buff *skb; 139 unsigned long flags; 140 void *msg_header; 141 142 al = sizeof(struct net_dm_alert_msg); 143 al += dm_hit_limit * sizeof(struct net_dm_drop_point); 144 al += sizeof(struct nlattr); 145 146 skb = genlmsg_new(al, GFP_KERNEL); 147 148 if (!skb) 149 goto err; 150 151 msg_header = genlmsg_put(skb, 0, 0, &net_drop_monitor_family, 152 0, NET_DM_CMD_ALERT); 153 if (!msg_header) { 154 nlmsg_free(skb); 155 skb = NULL; 156 goto err; 157 } 158 nla = nla_reserve(skb, NLA_UNSPEC, 159 sizeof(struct net_dm_alert_msg)); 160 if (!nla) { 161 nlmsg_free(skb); 162 skb = NULL; 163 goto err; 164 } 165 msg = nla_data(nla); 166 memset(msg, 0, al); 167 goto out; 168 169 err: 170 mod_timer(&data->send_timer, jiffies + HZ / 10); 171 out: 172 raw_spin_lock_irqsave(&data->lock, flags); 173 swap(data->skb, skb); 174 raw_spin_unlock_irqrestore(&data->lock, flags); 175 176 if (skb) { 177 struct nlmsghdr *nlh = (struct nlmsghdr *)skb->data; 178 struct genlmsghdr *gnlh = (struct genlmsghdr *)nlmsg_data(nlh); 179 180 genlmsg_end(skb, genlmsg_data(gnlh)); 181 } 182 183 return skb; 184 } 185 186 static const struct genl_multicast_group dropmon_mcgrps[] = { 187 { .name = "events", .flags = GENL_MCAST_CAP_SYS_ADMIN, }, 188 }; 189 190 static void send_dm_alert(struct work_struct *work) 191 { 192 struct sk_buff *skb; 193 struct per_cpu_dm_data *data; 194 195 data = container_of(work, struct per_cpu_dm_data, dm_alert_work); 196 197 skb = reset_per_cpu_data(data); 198 199 if (skb) 200 genlmsg_multicast(&net_drop_monitor_family, skb, 0, 201 0, GFP_KERNEL); 202 } 203 204 /* 205 * This is the timer function to delay the sending of an alert 206 * in the event that more drops will arrive during the 207 * hysteresis period. 208 */ 209 static void sched_send_work(struct timer_list *t) 210 { 211 struct per_cpu_dm_data *data = timer_container_of(data, t, send_timer); 212 213 schedule_work(&data->dm_alert_work); 214 } 215 216 static void trace_drop_common(struct sk_buff *skb, void *location) 217 { 218 struct net_dm_alert_msg *msg; 219 struct net_dm_drop_point *point; 220 struct nlmsghdr *nlh; 221 struct nlattr *nla; 222 int i; 223 struct sk_buff *dskb; 224 struct per_cpu_dm_data *data; 225 unsigned long flags; 226 227 local_irq_save(flags); 228 data = this_cpu_ptr(&dm_cpu_data); 229 raw_spin_lock(&data->lock); 230 dskb = data->skb; 231 232 if (!dskb) 233 goto out; 234 235 nlh = (struct nlmsghdr *)dskb->data; 236 nla = genlmsg_data(nlmsg_data(nlh)); 237 msg = nla_data(nla); 238 point = msg->points; 239 for (i = 0; i < msg->entries; i++) { 240 if (!memcmp(&location, &point->pc, sizeof(void *))) { 241 point->count++; 242 goto out; 243 } 244 point++; 245 } 246 if (msg->entries == dm_hit_limit) 247 goto out; 248 /* 249 * We need to create a new entry 250 */ 251 __nla_reserve_nohdr(dskb, sizeof(struct net_dm_drop_point)); 252 nla->nla_len += NLA_ALIGN(sizeof(struct net_dm_drop_point)); 253 memcpy(point->pc, &location, sizeof(void *)); 254 point->count = 1; 255 msg->entries++; 256 257 if (!timer_pending(&data->send_timer)) { 258 data->send_timer.expires = jiffies + dm_delay * HZ; 259 add_timer(&data->send_timer); 260 } 261 262 out: 263 raw_spin_unlock_irqrestore(&data->lock, flags); 264 } 265 266 static void trace_kfree_skb_hit(void *ignore, struct sk_buff *skb, 267 void *location, 268 enum skb_drop_reason reason, 269 const struct sock *rx_sk) 270 { 271 trace_drop_common(skb, location); 272 } 273 274 static void trace_napi_poll_hit(void *ignore, struct napi_struct *napi, 275 int work, int budget) 276 { 277 struct net_device *dev = napi->dev; 278 struct dm_hw_stat_delta *stat; 279 /* 280 * Don't check napi structures with no associated device 281 */ 282 if (!dev) 283 return; 284 285 rcu_read_lock(); 286 stat = rcu_dereference(dev->dm_private); 287 if (stat) { 288 /* 289 * only add a note to our monitor buffer if: 290 * 1) its after the last_rx delta 291 * 2) our rx_dropped count has gone up 292 */ 293 if (time_after(jiffies, stat->last_rx + dm_hw_check_delta) && 294 (dev->stats.rx_dropped != stat->last_drop_val)) { 295 trace_drop_common(NULL, NULL); 296 stat->last_drop_val = dev->stats.rx_dropped; 297 stat->last_rx = jiffies; 298 } 299 } 300 rcu_read_unlock(); 301 } 302 303 static struct net_dm_hw_entries * 304 net_dm_hw_reset_per_cpu_data(struct per_cpu_dm_data *hw_data) 305 { 306 struct net_dm_hw_entries *hw_entries; 307 unsigned long flags; 308 309 hw_entries = kzalloc_flex(*hw_entries, entries, dm_hit_limit); 310 if (!hw_entries) { 311 /* If the memory allocation failed, we try to perform another 312 * allocation in 1/10 second. Otherwise, the probe function 313 * will constantly bail out. 314 */ 315 mod_timer(&hw_data->send_timer, jiffies + HZ / 10); 316 } 317 318 raw_spin_lock_irqsave(&hw_data->lock, flags); 319 swap(hw_data->hw_entries, hw_entries); 320 raw_spin_unlock_irqrestore(&hw_data->lock, flags); 321 322 return hw_entries; 323 } 324 325 static int net_dm_hw_entry_put(struct sk_buff *msg, 326 const struct net_dm_hw_entry *hw_entry) 327 { 328 struct nlattr *attr; 329 330 attr = nla_nest_start(msg, NET_DM_ATTR_HW_ENTRY); 331 if (!attr) 332 return -EMSGSIZE; 333 334 if (nla_put_string(msg, NET_DM_ATTR_HW_TRAP_NAME, hw_entry->trap_name)) 335 goto nla_put_failure; 336 337 if (nla_put_u32(msg, NET_DM_ATTR_HW_TRAP_COUNT, hw_entry->count)) 338 goto nla_put_failure; 339 340 nla_nest_end(msg, attr); 341 342 return 0; 343 344 nla_put_failure: 345 nla_nest_cancel(msg, attr); 346 return -EMSGSIZE; 347 } 348 349 static int net_dm_hw_entries_put(struct sk_buff *msg, 350 const struct net_dm_hw_entries *hw_entries) 351 { 352 struct nlattr *attr; 353 int i; 354 355 attr = nla_nest_start(msg, NET_DM_ATTR_HW_ENTRIES); 356 if (!attr) 357 return -EMSGSIZE; 358 359 for (i = 0; i < hw_entries->num_entries; i++) { 360 int rc; 361 362 rc = net_dm_hw_entry_put(msg, &hw_entries->entries[i]); 363 if (rc) 364 goto nla_put_failure; 365 } 366 367 nla_nest_end(msg, attr); 368 369 return 0; 370 371 nla_put_failure: 372 nla_nest_cancel(msg, attr); 373 return -EMSGSIZE; 374 } 375 376 static int 377 net_dm_hw_summary_report_fill(struct sk_buff *msg, 378 const struct net_dm_hw_entries *hw_entries) 379 { 380 struct net_dm_alert_msg anc_hdr = { 0 }; 381 void *hdr; 382 int rc; 383 384 hdr = genlmsg_put(msg, 0, 0, &net_drop_monitor_family, 0, 385 NET_DM_CMD_ALERT); 386 if (!hdr) 387 return -EMSGSIZE; 388 389 /* We need to put the ancillary header in order not to break user 390 * space. 391 */ 392 if (nla_put(msg, NLA_UNSPEC, sizeof(anc_hdr), &anc_hdr)) 393 goto nla_put_failure; 394 395 rc = net_dm_hw_entries_put(msg, hw_entries); 396 if (rc) 397 goto nla_put_failure; 398 399 genlmsg_end(msg, hdr); 400 401 return 0; 402 403 nla_put_failure: 404 genlmsg_cancel(msg, hdr); 405 return -EMSGSIZE; 406 } 407 408 static void net_dm_hw_summary_work(struct work_struct *work) 409 { 410 struct net_dm_hw_entries *hw_entries; 411 struct per_cpu_dm_data *hw_data; 412 struct sk_buff *msg; 413 int rc; 414 415 hw_data = container_of(work, struct per_cpu_dm_data, dm_alert_work); 416 417 hw_entries = net_dm_hw_reset_per_cpu_data(hw_data); 418 if (!hw_entries) 419 return; 420 421 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 422 if (!msg) 423 goto out; 424 425 rc = net_dm_hw_summary_report_fill(msg, hw_entries); 426 if (rc) { 427 nlmsg_free(msg); 428 goto out; 429 } 430 431 genlmsg_multicast(&net_drop_monitor_family, msg, 0, 0, GFP_KERNEL); 432 433 out: 434 kfree(hw_entries); 435 } 436 437 static void 438 net_dm_hw_trap_summary_probe(void *ignore, const struct devlink *devlink, 439 struct sk_buff *skb, 440 const struct devlink_trap_metadata *metadata) 441 { 442 struct net_dm_hw_entries *hw_entries; 443 struct net_dm_hw_entry *hw_entry; 444 struct per_cpu_dm_data *hw_data; 445 unsigned long flags; 446 int i; 447 448 if (metadata->trap_type == DEVLINK_TRAP_TYPE_CONTROL) 449 return; 450 451 hw_data = this_cpu_ptr(&dm_hw_cpu_data); 452 raw_spin_lock_irqsave(&hw_data->lock, flags); 453 hw_entries = hw_data->hw_entries; 454 455 if (!hw_entries) 456 goto out; 457 458 for (i = 0; i < hw_entries->num_entries; i++) { 459 hw_entry = &hw_entries->entries[i]; 460 if (!strncmp(hw_entry->trap_name, metadata->trap_name, 461 NET_DM_MAX_HW_TRAP_NAME_LEN - 1)) { 462 hw_entry->count++; 463 goto out; 464 } 465 } 466 if (WARN_ON_ONCE(hw_entries->num_entries == dm_hit_limit)) 467 goto out; 468 469 hw_entry = &hw_entries->entries[hw_entries->num_entries]; 470 strscpy(hw_entry->trap_name, metadata->trap_name, 471 NET_DM_MAX_HW_TRAP_NAME_LEN - 1); 472 hw_entry->count = 1; 473 hw_entries->num_entries++; 474 475 if (!timer_pending(&hw_data->send_timer)) { 476 hw_data->send_timer.expires = jiffies + dm_delay * HZ; 477 add_timer(&hw_data->send_timer); 478 } 479 480 out: 481 raw_spin_unlock_irqrestore(&hw_data->lock, flags); 482 } 483 484 static const struct net_dm_alert_ops net_dm_alert_summary_ops = { 485 .kfree_skb_probe = trace_kfree_skb_hit, 486 .napi_poll_probe = trace_napi_poll_hit, 487 .work_item_func = send_dm_alert, 488 .hw_work_item_func = net_dm_hw_summary_work, 489 .hw_trap_probe = net_dm_hw_trap_summary_probe, 490 }; 491 492 static void net_dm_packet_trace_kfree_skb_hit(void *ignore, 493 struct sk_buff *skb, 494 void *location, 495 enum skb_drop_reason reason, 496 const struct sock *rx_sk) 497 { 498 ktime_t tstamp = ktime_get_real(); 499 struct per_cpu_dm_data *data; 500 struct net_dm_skb_cb *cb; 501 struct sk_buff *nskb; 502 unsigned long flags; 503 504 if (!skb_mac_header_was_set(skb)) 505 return; 506 507 nskb = skb_clone(skb, GFP_ATOMIC); 508 if (!nskb) 509 return; 510 511 cb = NET_DM_SKB_CB(nskb); 512 cb->reason = reason; 513 cb->pc = location; 514 /* Override the timestamp because we care about the time when the 515 * packet was dropped. 516 */ 517 nskb->tstamp = tstamp; 518 519 data = this_cpu_ptr(&dm_cpu_data); 520 521 spin_lock_irqsave(&data->drop_queue.lock, flags); 522 if (skb_queue_len(&data->drop_queue) < net_dm_queue_len) 523 __skb_queue_tail(&data->drop_queue, nskb); 524 else 525 goto unlock_free; 526 spin_unlock_irqrestore(&data->drop_queue.lock, flags); 527 528 schedule_work(&data->dm_alert_work); 529 530 return; 531 532 unlock_free: 533 u64_stats_update_begin(&data->stats.syncp); 534 u64_stats_inc(&data->stats.dropped); 535 u64_stats_update_end(&data->stats.syncp); 536 spin_unlock_irqrestore(&data->drop_queue.lock, flags); 537 consume_skb(nskb); 538 } 539 540 static void net_dm_packet_trace_napi_poll_hit(void *ignore, 541 struct napi_struct *napi, 542 int work, int budget) 543 { 544 } 545 546 static size_t net_dm_in_port_size(void) 547 { 548 /* NET_DM_ATTR_IN_PORT nest */ 549 return nla_total_size(0) + 550 /* NET_DM_ATTR_PORT_NETDEV_IFINDEX */ 551 nla_total_size(sizeof(u32)) + 552 /* NET_DM_ATTR_PORT_NETDEV_NAME */ 553 nla_total_size(IFNAMSIZ + 1); 554 } 555 556 #define NET_DM_MAX_SYMBOL_LEN 40 557 #define NET_DM_MAX_REASON_LEN 50 558 559 static size_t net_dm_packet_report_size(size_t payload_len) 560 { 561 size_t size; 562 563 size = nlmsg_msg_size(GENL_HDRLEN + net_drop_monitor_family.hdrsize); 564 565 return NLMSG_ALIGN(size) + 566 /* NET_DM_ATTR_ORIGIN */ 567 nla_total_size(sizeof(u16)) + 568 /* NET_DM_ATTR_PC */ 569 nla_total_size_64bit(sizeof(u64)) + 570 /* NET_DM_ATTR_SYMBOL */ 571 nla_total_size(NET_DM_MAX_SYMBOL_LEN + 1) + 572 /* NET_DM_ATTR_IN_PORT */ 573 net_dm_in_port_size() + 574 /* NET_DM_ATTR_TIMESTAMP */ 575 nla_total_size_64bit(sizeof(u64)) + 576 /* NET_DM_ATTR_ORIG_LEN */ 577 nla_total_size(sizeof(u32)) + 578 /* NET_DM_ATTR_PROTO */ 579 nla_total_size(sizeof(u16)) + 580 /* NET_DM_ATTR_REASON */ 581 nla_total_size(NET_DM_MAX_REASON_LEN + 1) + 582 /* NET_DM_ATTR_PAYLOAD */ 583 nla_total_size(payload_len); 584 } 585 586 static int net_dm_packet_report_in_port_put(struct sk_buff *msg, int ifindex, 587 const char *name) 588 { 589 struct nlattr *attr; 590 591 attr = nla_nest_start(msg, NET_DM_ATTR_IN_PORT); 592 if (!attr) 593 return -EMSGSIZE; 594 595 if (ifindex && 596 nla_put_u32(msg, NET_DM_ATTR_PORT_NETDEV_IFINDEX, ifindex)) 597 goto nla_put_failure; 598 599 if (name && nla_put_string(msg, NET_DM_ATTR_PORT_NETDEV_NAME, name)) 600 goto nla_put_failure; 601 602 nla_nest_end(msg, attr); 603 604 return 0; 605 606 nla_put_failure: 607 nla_nest_cancel(msg, attr); 608 return -EMSGSIZE; 609 } 610 611 static int net_dm_packet_report_fill(struct sk_buff *msg, struct sk_buff *skb, 612 size_t payload_len) 613 { 614 struct net_dm_skb_cb *cb = NET_DM_SKB_CB(skb); 615 const struct drop_reason_list *list = NULL; 616 unsigned int subsys, subsys_reason; 617 char buf[NET_DM_MAX_SYMBOL_LEN]; 618 struct nlattr *attr; 619 void *hdr; 620 int rc; 621 622 hdr = genlmsg_put(msg, 0, 0, &net_drop_monitor_family, 0, 623 NET_DM_CMD_PACKET_ALERT); 624 if (!hdr) 625 return -EMSGSIZE; 626 627 if (nla_put_u16(msg, NET_DM_ATTR_ORIGIN, NET_DM_ORIGIN_SW)) 628 goto nla_put_failure; 629 630 if (nla_put_u64_64bit(msg, NET_DM_ATTR_PC, (u64)(uintptr_t)cb->pc, 631 NET_DM_ATTR_PAD)) 632 goto nla_put_failure; 633 634 rcu_read_lock(); 635 subsys = u32_get_bits(cb->reason, SKB_DROP_REASON_SUBSYS_MASK); 636 if (subsys < SKB_DROP_REASON_SUBSYS_NUM) 637 list = rcu_dereference(drop_reasons_by_subsys[subsys]); 638 subsys_reason = cb->reason & ~SKB_DROP_REASON_SUBSYS_MASK; 639 if (!list || 640 subsys_reason >= list->n_reasons || 641 !list->reasons[subsys_reason] || 642 strlen(list->reasons[subsys_reason]) > NET_DM_MAX_REASON_LEN) { 643 list = rcu_dereference(drop_reasons_by_subsys[SKB_DROP_REASON_SUBSYS_CORE]); 644 subsys_reason = SKB_DROP_REASON_NOT_SPECIFIED; 645 } 646 if (nla_put_string(msg, NET_DM_ATTR_REASON, 647 list->reasons[subsys_reason])) { 648 rcu_read_unlock(); 649 goto nla_put_failure; 650 } 651 rcu_read_unlock(); 652 653 snprintf(buf, sizeof(buf), "%pS", cb->pc); 654 if (nla_put_string(msg, NET_DM_ATTR_SYMBOL, buf)) 655 goto nla_put_failure; 656 657 rc = net_dm_packet_report_in_port_put(msg, skb->skb_iif, NULL); 658 if (rc) 659 goto nla_put_failure; 660 661 if (nla_put_u64_64bit(msg, NET_DM_ATTR_TIMESTAMP, 662 ktime_to_ns(skb->tstamp), NET_DM_ATTR_PAD)) 663 goto nla_put_failure; 664 665 if (nla_put_u32(msg, NET_DM_ATTR_ORIG_LEN, skb->len)) 666 goto nla_put_failure; 667 668 if (!payload_len) 669 goto out; 670 671 if (nla_put_u16(msg, NET_DM_ATTR_PROTO, be16_to_cpu(skb->protocol))) 672 goto nla_put_failure; 673 674 attr = __nla_reserve(msg, NET_DM_ATTR_PAYLOAD, payload_len); 675 if (skb_copy_bits(skb, 0, nla_data(attr), payload_len)) 676 goto nla_put_failure; 677 678 out: 679 genlmsg_end(msg, hdr); 680 681 return 0; 682 683 nla_put_failure: 684 genlmsg_cancel(msg, hdr); 685 return -EMSGSIZE; 686 } 687 688 #define NET_DM_MAX_PACKET_SIZE (0xffff - NLA_HDRLEN - NLA_ALIGNTO) 689 690 static void net_dm_packet_report(struct sk_buff *skb) 691 { 692 struct sk_buff *msg; 693 size_t payload_len; 694 int rc; 695 696 /* Make sure we start copying the packet from the MAC header */ 697 if (skb->data > skb_mac_header(skb)) 698 skb_push(skb, skb->data - skb_mac_header(skb)); 699 else 700 skb_pull(skb, skb_mac_header(skb) - skb->data); 701 702 /* Ensure packet fits inside a single netlink attribute */ 703 payload_len = min_t(size_t, skb->len, NET_DM_MAX_PACKET_SIZE); 704 if (net_dm_trunc_len) 705 payload_len = min_t(size_t, net_dm_trunc_len, payload_len); 706 707 msg = nlmsg_new(net_dm_packet_report_size(payload_len), GFP_KERNEL); 708 if (!msg) 709 goto out; 710 711 rc = net_dm_packet_report_fill(msg, skb, payload_len); 712 if (rc) { 713 nlmsg_free(msg); 714 goto out; 715 } 716 717 genlmsg_multicast(&net_drop_monitor_family, msg, 0, 0, GFP_KERNEL); 718 719 out: 720 consume_skb(skb); 721 } 722 723 static void net_dm_packet_work(struct work_struct *work) 724 { 725 struct per_cpu_dm_data *data; 726 struct sk_buff_head list; 727 struct sk_buff *skb; 728 unsigned long flags; 729 730 data = container_of(work, struct per_cpu_dm_data, dm_alert_work); 731 732 __skb_queue_head_init(&list); 733 734 spin_lock_irqsave(&data->drop_queue.lock, flags); 735 skb_queue_splice_tail_init(&data->drop_queue, &list); 736 spin_unlock_irqrestore(&data->drop_queue.lock, flags); 737 738 while ((skb = __skb_dequeue(&list))) 739 net_dm_packet_report(skb); 740 } 741 742 static size_t 743 net_dm_flow_action_cookie_size(const struct devlink_trap_metadata *hw_metadata) 744 { 745 return hw_metadata->fa_cookie ? 746 nla_total_size(hw_metadata->fa_cookie->cookie_len) : 0; 747 } 748 749 static size_t 750 net_dm_hw_packet_report_size(size_t payload_len, 751 const struct devlink_trap_metadata *hw_metadata) 752 { 753 size_t size; 754 755 size = nlmsg_msg_size(GENL_HDRLEN + net_drop_monitor_family.hdrsize); 756 757 return NLMSG_ALIGN(size) + 758 /* NET_DM_ATTR_ORIGIN */ 759 nla_total_size(sizeof(u16)) + 760 /* NET_DM_ATTR_HW_TRAP_GROUP_NAME */ 761 nla_total_size(strlen(hw_metadata->trap_group_name) + 1) + 762 /* NET_DM_ATTR_HW_TRAP_NAME */ 763 nla_total_size(strlen(hw_metadata->trap_name) + 1) + 764 /* NET_DM_ATTR_IN_PORT */ 765 net_dm_in_port_size() + 766 /* NET_DM_ATTR_FLOW_ACTION_COOKIE */ 767 net_dm_flow_action_cookie_size(hw_metadata) + 768 /* NET_DM_ATTR_TIMESTAMP */ 769 nla_total_size_64bit(sizeof(u64)) + 770 /* NET_DM_ATTR_ORIG_LEN */ 771 nla_total_size(sizeof(u32)) + 772 /* NET_DM_ATTR_PROTO */ 773 nla_total_size(sizeof(u16)) + 774 /* NET_DM_ATTR_PAYLOAD */ 775 nla_total_size(payload_len); 776 } 777 778 static int net_dm_hw_packet_report_fill(struct sk_buff *msg, 779 struct sk_buff *skb, size_t payload_len) 780 { 781 struct devlink_trap_metadata *hw_metadata; 782 struct nlattr *attr; 783 void *hdr; 784 785 hw_metadata = NET_DM_SKB_CB(skb)->hw_metadata; 786 787 hdr = genlmsg_put(msg, 0, 0, &net_drop_monitor_family, 0, 788 NET_DM_CMD_PACKET_ALERT); 789 if (!hdr) 790 return -EMSGSIZE; 791 792 if (nla_put_u16(msg, NET_DM_ATTR_ORIGIN, NET_DM_ORIGIN_HW)) 793 goto nla_put_failure; 794 795 if (nla_put_string(msg, NET_DM_ATTR_HW_TRAP_GROUP_NAME, 796 hw_metadata->trap_group_name)) 797 goto nla_put_failure; 798 799 if (nla_put_string(msg, NET_DM_ATTR_HW_TRAP_NAME, 800 hw_metadata->trap_name)) 801 goto nla_put_failure; 802 803 if (hw_metadata->input_dev) { 804 struct net_device *dev = hw_metadata->input_dev; 805 int rc; 806 807 rc = net_dm_packet_report_in_port_put(msg, dev->ifindex, 808 dev->name); 809 if (rc) 810 goto nla_put_failure; 811 } 812 813 if (hw_metadata->fa_cookie && 814 nla_put(msg, NET_DM_ATTR_FLOW_ACTION_COOKIE, 815 hw_metadata->fa_cookie->cookie_len, 816 hw_metadata->fa_cookie->cookie)) 817 goto nla_put_failure; 818 819 if (nla_put_u64_64bit(msg, NET_DM_ATTR_TIMESTAMP, 820 ktime_to_ns(skb->tstamp), NET_DM_ATTR_PAD)) 821 goto nla_put_failure; 822 823 if (nla_put_u32(msg, NET_DM_ATTR_ORIG_LEN, skb->len)) 824 goto nla_put_failure; 825 826 if (!payload_len) 827 goto out; 828 829 if (nla_put_u16(msg, NET_DM_ATTR_PROTO, be16_to_cpu(skb->protocol))) 830 goto nla_put_failure; 831 832 attr = __nla_reserve(msg, NET_DM_ATTR_PAYLOAD, payload_len); 833 if (skb_copy_bits(skb, 0, nla_data(attr), payload_len)) 834 goto nla_put_failure; 835 836 out: 837 genlmsg_end(msg, hdr); 838 839 return 0; 840 841 nla_put_failure: 842 genlmsg_cancel(msg, hdr); 843 return -EMSGSIZE; 844 } 845 846 static struct devlink_trap_metadata * 847 net_dm_hw_metadata_copy(const struct devlink_trap_metadata *metadata) 848 { 849 const struct flow_action_cookie *fa_cookie; 850 struct devlink_trap_metadata *hw_metadata; 851 const char *trap_group_name; 852 const char *trap_name; 853 854 hw_metadata = kzalloc_obj(*hw_metadata, GFP_ATOMIC); 855 if (!hw_metadata) 856 return NULL; 857 858 trap_group_name = kstrdup(metadata->trap_group_name, GFP_ATOMIC); 859 if (!trap_group_name) 860 goto free_hw_metadata; 861 hw_metadata->trap_group_name = trap_group_name; 862 863 trap_name = kstrdup(metadata->trap_name, GFP_ATOMIC); 864 if (!trap_name) 865 goto free_trap_group; 866 hw_metadata->trap_name = trap_name; 867 868 if (metadata->fa_cookie) { 869 size_t cookie_size = sizeof(*fa_cookie) + 870 metadata->fa_cookie->cookie_len; 871 872 fa_cookie = kmemdup(metadata->fa_cookie, cookie_size, 873 GFP_ATOMIC); 874 if (!fa_cookie) 875 goto free_trap_name; 876 hw_metadata->fa_cookie = fa_cookie; 877 } 878 879 hw_metadata->input_dev = metadata->input_dev; 880 netdev_hold(hw_metadata->input_dev, &hw_metadata->dev_tracker, 881 GFP_ATOMIC); 882 883 return hw_metadata; 884 885 free_trap_name: 886 kfree(trap_name); 887 free_trap_group: 888 kfree(trap_group_name); 889 free_hw_metadata: 890 kfree(hw_metadata); 891 return NULL; 892 } 893 894 static void 895 net_dm_hw_metadata_free(struct devlink_trap_metadata *hw_metadata) 896 { 897 netdev_put(hw_metadata->input_dev, &hw_metadata->dev_tracker); 898 kfree(hw_metadata->fa_cookie); 899 kfree(hw_metadata->trap_name); 900 kfree(hw_metadata->trap_group_name); 901 kfree(hw_metadata); 902 } 903 904 static void net_dm_hw_packet_report(struct sk_buff *skb) 905 { 906 struct devlink_trap_metadata *hw_metadata; 907 struct sk_buff *msg; 908 size_t payload_len; 909 int rc; 910 911 if (skb->data > skb_mac_header(skb)) 912 skb_push(skb, skb->data - skb_mac_header(skb)); 913 else 914 skb_pull(skb, skb_mac_header(skb) - skb->data); 915 916 payload_len = min_t(size_t, skb->len, NET_DM_MAX_PACKET_SIZE); 917 if (net_dm_trunc_len) 918 payload_len = min_t(size_t, net_dm_trunc_len, payload_len); 919 920 hw_metadata = NET_DM_SKB_CB(skb)->hw_metadata; 921 msg = nlmsg_new(net_dm_hw_packet_report_size(payload_len, hw_metadata), 922 GFP_KERNEL); 923 if (!msg) 924 goto out; 925 926 rc = net_dm_hw_packet_report_fill(msg, skb, payload_len); 927 if (rc) { 928 nlmsg_free(msg); 929 goto out; 930 } 931 932 genlmsg_multicast(&net_drop_monitor_family, msg, 0, 0, GFP_KERNEL); 933 934 out: 935 net_dm_hw_metadata_free(NET_DM_SKB_CB(skb)->hw_metadata); 936 consume_skb(skb); 937 } 938 939 static void net_dm_hw_packet_work(struct work_struct *work) 940 { 941 struct per_cpu_dm_data *hw_data; 942 struct sk_buff_head list; 943 struct sk_buff *skb; 944 unsigned long flags; 945 946 hw_data = container_of(work, struct per_cpu_dm_data, dm_alert_work); 947 948 __skb_queue_head_init(&list); 949 950 spin_lock_irqsave(&hw_data->drop_queue.lock, flags); 951 skb_queue_splice_tail_init(&hw_data->drop_queue, &list); 952 spin_unlock_irqrestore(&hw_data->drop_queue.lock, flags); 953 954 while ((skb = __skb_dequeue(&list))) 955 net_dm_hw_packet_report(skb); 956 } 957 958 static void 959 net_dm_hw_trap_packet_probe(void *ignore, const struct devlink *devlink, 960 struct sk_buff *skb, 961 const struct devlink_trap_metadata *metadata) 962 { 963 struct devlink_trap_metadata *n_hw_metadata; 964 ktime_t tstamp = ktime_get_real(); 965 struct per_cpu_dm_data *hw_data; 966 struct sk_buff *nskb; 967 unsigned long flags; 968 969 if (metadata->trap_type == DEVLINK_TRAP_TYPE_CONTROL) 970 return; 971 972 if (!skb_mac_header_was_set(skb)) 973 return; 974 975 nskb = skb_clone(skb, GFP_ATOMIC); 976 if (!nskb) 977 return; 978 979 n_hw_metadata = net_dm_hw_metadata_copy(metadata); 980 if (!n_hw_metadata) 981 goto free; 982 983 NET_DM_SKB_CB(nskb)->hw_metadata = n_hw_metadata; 984 nskb->tstamp = tstamp; 985 986 hw_data = this_cpu_ptr(&dm_hw_cpu_data); 987 988 spin_lock_irqsave(&hw_data->drop_queue.lock, flags); 989 if (skb_queue_len(&hw_data->drop_queue) < net_dm_queue_len) 990 __skb_queue_tail(&hw_data->drop_queue, nskb); 991 else 992 goto unlock_free; 993 spin_unlock_irqrestore(&hw_data->drop_queue.lock, flags); 994 995 schedule_work(&hw_data->dm_alert_work); 996 997 return; 998 999 unlock_free: 1000 u64_stats_update_begin(&hw_data->stats.syncp); 1001 u64_stats_inc(&hw_data->stats.dropped); 1002 u64_stats_update_end(&hw_data->stats.syncp); 1003 spin_unlock_irqrestore(&hw_data->drop_queue.lock, flags); 1004 net_dm_hw_metadata_free(n_hw_metadata); 1005 free: 1006 consume_skb(nskb); 1007 } 1008 1009 static const struct net_dm_alert_ops net_dm_alert_packet_ops = { 1010 .kfree_skb_probe = net_dm_packet_trace_kfree_skb_hit, 1011 .napi_poll_probe = net_dm_packet_trace_napi_poll_hit, 1012 .work_item_func = net_dm_packet_work, 1013 .hw_work_item_func = net_dm_hw_packet_work, 1014 .hw_trap_probe = net_dm_hw_trap_packet_probe, 1015 }; 1016 1017 static const struct net_dm_alert_ops *net_dm_alert_ops_arr[] = { 1018 [NET_DM_ALERT_MODE_SUMMARY] = &net_dm_alert_summary_ops, 1019 [NET_DM_ALERT_MODE_PACKET] = &net_dm_alert_packet_ops, 1020 }; 1021 1022 #if IS_ENABLED(CONFIG_NET_DEVLINK) 1023 static int net_dm_hw_probe_register(const struct net_dm_alert_ops *ops) 1024 { 1025 return register_trace_devlink_trap_report(ops->hw_trap_probe, NULL); 1026 } 1027 1028 static void net_dm_hw_probe_unregister(const struct net_dm_alert_ops *ops) 1029 { 1030 unregister_trace_devlink_trap_report(ops->hw_trap_probe, NULL); 1031 tracepoint_synchronize_unregister(); 1032 } 1033 #else 1034 static int net_dm_hw_probe_register(const struct net_dm_alert_ops *ops) 1035 { 1036 return -EOPNOTSUPP; 1037 } 1038 1039 static void net_dm_hw_probe_unregister(const struct net_dm_alert_ops *ops) 1040 { 1041 } 1042 #endif 1043 1044 static int net_dm_hw_monitor_start(struct netlink_ext_ack *extack) 1045 { 1046 const struct net_dm_alert_ops *ops; 1047 int cpu, rc; 1048 1049 if (monitor_hw) { 1050 NL_SET_ERR_MSG_MOD(extack, "Hardware monitoring already enabled"); 1051 return -EAGAIN; 1052 } 1053 1054 ops = net_dm_alert_ops_arr[net_dm_alert_mode]; 1055 1056 if (!try_module_get(THIS_MODULE)) { 1057 NL_SET_ERR_MSG_MOD(extack, "Failed to take reference on module"); 1058 return -ENODEV; 1059 } 1060 1061 for_each_possible_cpu(cpu) { 1062 struct per_cpu_dm_data *hw_data = &per_cpu(dm_hw_cpu_data, cpu); 1063 struct net_dm_hw_entries *hw_entries; 1064 1065 INIT_WORK(&hw_data->dm_alert_work, ops->hw_work_item_func); 1066 timer_setup(&hw_data->send_timer, sched_send_work, 0); 1067 hw_entries = net_dm_hw_reset_per_cpu_data(hw_data); 1068 kfree(hw_entries); 1069 } 1070 1071 rc = net_dm_hw_probe_register(ops); 1072 if (rc) { 1073 NL_SET_ERR_MSG_MOD(extack, "Failed to connect probe to devlink_trap_probe() tracepoint"); 1074 goto err_module_put; 1075 } 1076 1077 monitor_hw = true; 1078 1079 return 0; 1080 1081 err_module_put: 1082 for_each_possible_cpu(cpu) { 1083 struct per_cpu_dm_data *hw_data = &per_cpu(dm_hw_cpu_data, cpu); 1084 struct sk_buff *skb; 1085 1086 timer_delete_sync(&hw_data->send_timer); 1087 cancel_work_sync(&hw_data->dm_alert_work); 1088 while ((skb = __skb_dequeue(&hw_data->drop_queue))) { 1089 struct devlink_trap_metadata *hw_metadata; 1090 1091 hw_metadata = NET_DM_SKB_CB(skb)->hw_metadata; 1092 net_dm_hw_metadata_free(hw_metadata); 1093 consume_skb(skb); 1094 } 1095 } 1096 module_put(THIS_MODULE); 1097 return rc; 1098 } 1099 1100 static void net_dm_hw_monitor_stop(struct netlink_ext_ack *extack) 1101 { 1102 const struct net_dm_alert_ops *ops; 1103 int cpu; 1104 1105 if (!monitor_hw) { 1106 NL_SET_ERR_MSG_MOD(extack, "Hardware monitoring already disabled"); 1107 return; 1108 } 1109 1110 ops = net_dm_alert_ops_arr[net_dm_alert_mode]; 1111 1112 monitor_hw = false; 1113 1114 net_dm_hw_probe_unregister(ops); 1115 1116 for_each_possible_cpu(cpu) { 1117 struct per_cpu_dm_data *hw_data = &per_cpu(dm_hw_cpu_data, cpu); 1118 struct sk_buff *skb; 1119 1120 timer_delete_sync(&hw_data->send_timer); 1121 cancel_work_sync(&hw_data->dm_alert_work); 1122 while ((skb = __skb_dequeue(&hw_data->drop_queue))) { 1123 struct devlink_trap_metadata *hw_metadata; 1124 1125 hw_metadata = NET_DM_SKB_CB(skb)->hw_metadata; 1126 net_dm_hw_metadata_free(hw_metadata); 1127 consume_skb(skb); 1128 } 1129 } 1130 1131 module_put(THIS_MODULE); 1132 } 1133 1134 static int net_dm_trace_on_set(struct netlink_ext_ack *extack) 1135 { 1136 const struct net_dm_alert_ops *ops; 1137 int cpu, rc; 1138 1139 ops = net_dm_alert_ops_arr[net_dm_alert_mode]; 1140 1141 if (!try_module_get(THIS_MODULE)) { 1142 NL_SET_ERR_MSG_MOD(extack, "Failed to take reference on module"); 1143 return -ENODEV; 1144 } 1145 1146 for_each_possible_cpu(cpu) { 1147 struct per_cpu_dm_data *data = &per_cpu(dm_cpu_data, cpu); 1148 struct sk_buff *skb; 1149 1150 INIT_WORK(&data->dm_alert_work, ops->work_item_func); 1151 timer_setup(&data->send_timer, sched_send_work, 0); 1152 /* Allocate a new per-CPU skb for the summary alert message and 1153 * free the old one which might contain stale data from 1154 * previous tracing. 1155 */ 1156 skb = reset_per_cpu_data(data); 1157 consume_skb(skb); 1158 } 1159 1160 rc = register_trace_kfree_skb(ops->kfree_skb_probe, NULL); 1161 if (rc) { 1162 NL_SET_ERR_MSG_MOD(extack, "Failed to connect probe to kfree_skb() tracepoint"); 1163 goto err_module_put; 1164 } 1165 1166 rc = register_trace_napi_poll(ops->napi_poll_probe, NULL); 1167 if (rc) { 1168 NL_SET_ERR_MSG_MOD(extack, "Failed to connect probe to napi_poll() tracepoint"); 1169 goto err_unregister_trace; 1170 } 1171 1172 return 0; 1173 1174 err_unregister_trace: 1175 unregister_trace_kfree_skb(ops->kfree_skb_probe, NULL); 1176 err_module_put: 1177 for_each_possible_cpu(cpu) { 1178 struct per_cpu_dm_data *data = &per_cpu(dm_cpu_data, cpu); 1179 struct sk_buff *skb; 1180 1181 timer_delete_sync(&data->send_timer); 1182 cancel_work_sync(&data->dm_alert_work); 1183 while ((skb = __skb_dequeue(&data->drop_queue))) 1184 consume_skb(skb); 1185 } 1186 module_put(THIS_MODULE); 1187 return rc; 1188 } 1189 1190 static void net_dm_trace_off_set(void) 1191 { 1192 const struct net_dm_alert_ops *ops; 1193 int cpu; 1194 1195 ops = net_dm_alert_ops_arr[net_dm_alert_mode]; 1196 1197 unregister_trace_napi_poll(ops->napi_poll_probe, NULL); 1198 unregister_trace_kfree_skb(ops->kfree_skb_probe, NULL); 1199 1200 tracepoint_synchronize_unregister(); 1201 1202 /* Make sure we do not send notifications to user space after request 1203 * to stop tracing returns. 1204 */ 1205 for_each_possible_cpu(cpu) { 1206 struct per_cpu_dm_data *data = &per_cpu(dm_cpu_data, cpu); 1207 struct sk_buff *skb; 1208 1209 timer_delete_sync(&data->send_timer); 1210 cancel_work_sync(&data->dm_alert_work); 1211 while ((skb = __skb_dequeue(&data->drop_queue))) 1212 consume_skb(skb); 1213 } 1214 1215 module_put(THIS_MODULE); 1216 } 1217 1218 static int set_all_monitor_traces(int state, struct netlink_ext_ack *extack) 1219 { 1220 int rc = 0; 1221 1222 if (state == trace_state) { 1223 NL_SET_ERR_MSG_MOD(extack, "Trace state already set to requested state"); 1224 return -EAGAIN; 1225 } 1226 1227 switch (state) { 1228 case TRACE_ON: 1229 rc = net_dm_trace_on_set(extack); 1230 break; 1231 case TRACE_OFF: 1232 net_dm_trace_off_set(); 1233 break; 1234 default: 1235 rc = 1; 1236 break; 1237 } 1238 1239 if (!rc) 1240 trace_state = state; 1241 else 1242 rc = -EINPROGRESS; 1243 1244 return rc; 1245 } 1246 1247 static bool net_dm_is_monitoring(void) 1248 { 1249 return trace_state == TRACE_ON || monitor_hw; 1250 } 1251 1252 static int net_dm_alert_mode_get_from_info(struct genl_info *info, 1253 enum net_dm_alert_mode *p_alert_mode) 1254 { 1255 u8 val; 1256 1257 val = nla_get_u8(info->attrs[NET_DM_ATTR_ALERT_MODE]); 1258 1259 switch (val) { 1260 case NET_DM_ALERT_MODE_SUMMARY: 1261 case NET_DM_ALERT_MODE_PACKET: 1262 *p_alert_mode = val; 1263 break; 1264 default: 1265 return -EINVAL; 1266 } 1267 1268 return 0; 1269 } 1270 1271 static int net_dm_alert_mode_set(struct genl_info *info) 1272 { 1273 struct netlink_ext_ack *extack = info->extack; 1274 enum net_dm_alert_mode alert_mode; 1275 int rc; 1276 1277 if (!info->attrs[NET_DM_ATTR_ALERT_MODE]) 1278 return 0; 1279 1280 rc = net_dm_alert_mode_get_from_info(info, &alert_mode); 1281 if (rc) { 1282 NL_SET_ERR_MSG_MOD(extack, "Invalid alert mode"); 1283 return -EINVAL; 1284 } 1285 1286 net_dm_alert_mode = alert_mode; 1287 1288 return 0; 1289 } 1290 1291 static void net_dm_trunc_len_set(struct genl_info *info) 1292 { 1293 if (!info->attrs[NET_DM_ATTR_TRUNC_LEN]) 1294 return; 1295 1296 net_dm_trunc_len = nla_get_u32(info->attrs[NET_DM_ATTR_TRUNC_LEN]); 1297 } 1298 1299 static void net_dm_queue_len_set(struct genl_info *info) 1300 { 1301 if (!info->attrs[NET_DM_ATTR_QUEUE_LEN]) 1302 return; 1303 1304 net_dm_queue_len = nla_get_u32(info->attrs[NET_DM_ATTR_QUEUE_LEN]); 1305 } 1306 1307 static int net_dm_cmd_config(struct sk_buff *skb, 1308 struct genl_info *info) 1309 { 1310 struct netlink_ext_ack *extack = info->extack; 1311 int rc; 1312 1313 if (net_dm_is_monitoring()) { 1314 NL_SET_ERR_MSG_MOD(extack, "Cannot configure drop monitor during monitoring"); 1315 return -EBUSY; 1316 } 1317 1318 rc = net_dm_alert_mode_set(info); 1319 if (rc) 1320 return rc; 1321 1322 net_dm_trunc_len_set(info); 1323 1324 net_dm_queue_len_set(info); 1325 1326 return 0; 1327 } 1328 1329 static int net_dm_monitor_start(bool set_sw, bool set_hw, 1330 struct netlink_ext_ack *extack) 1331 { 1332 bool sw_set = false; 1333 int rc; 1334 1335 if (set_sw) { 1336 rc = set_all_monitor_traces(TRACE_ON, extack); 1337 if (rc) 1338 return rc; 1339 sw_set = true; 1340 } 1341 1342 if (set_hw) { 1343 rc = net_dm_hw_monitor_start(extack); 1344 if (rc) 1345 goto err_monitor_hw; 1346 } 1347 1348 return 0; 1349 1350 err_monitor_hw: 1351 if (sw_set) 1352 set_all_monitor_traces(TRACE_OFF, extack); 1353 return rc; 1354 } 1355 1356 static void net_dm_monitor_stop(bool set_sw, bool set_hw, 1357 struct netlink_ext_ack *extack) 1358 { 1359 if (set_hw) 1360 net_dm_hw_monitor_stop(extack); 1361 if (set_sw) 1362 set_all_monitor_traces(TRACE_OFF, extack); 1363 } 1364 1365 static int net_dm_cmd_trace(struct sk_buff *skb, 1366 struct genl_info *info) 1367 { 1368 bool set_sw = !!info->attrs[NET_DM_ATTR_SW_DROPS]; 1369 bool set_hw = !!info->attrs[NET_DM_ATTR_HW_DROPS]; 1370 struct netlink_ext_ack *extack = info->extack; 1371 1372 /* To maintain backward compatibility, we start / stop monitoring of 1373 * software drops if no flag is specified. 1374 */ 1375 if (!set_sw && !set_hw) 1376 set_sw = true; 1377 1378 switch (info->genlhdr->cmd) { 1379 case NET_DM_CMD_START: 1380 return net_dm_monitor_start(set_sw, set_hw, extack); 1381 case NET_DM_CMD_STOP: 1382 net_dm_monitor_stop(set_sw, set_hw, extack); 1383 return 0; 1384 } 1385 1386 return -EOPNOTSUPP; 1387 } 1388 1389 static int net_dm_config_fill(struct sk_buff *msg, struct genl_info *info) 1390 { 1391 void *hdr; 1392 1393 hdr = genlmsg_put(msg, info->snd_portid, info->snd_seq, 1394 &net_drop_monitor_family, 0, NET_DM_CMD_CONFIG_NEW); 1395 if (!hdr) 1396 return -EMSGSIZE; 1397 1398 if (nla_put_u8(msg, NET_DM_ATTR_ALERT_MODE, net_dm_alert_mode)) 1399 goto nla_put_failure; 1400 1401 if (nla_put_u32(msg, NET_DM_ATTR_TRUNC_LEN, net_dm_trunc_len)) 1402 goto nla_put_failure; 1403 1404 if (nla_put_u32(msg, NET_DM_ATTR_QUEUE_LEN, net_dm_queue_len)) 1405 goto nla_put_failure; 1406 1407 genlmsg_end(msg, hdr); 1408 1409 return 0; 1410 1411 nla_put_failure: 1412 genlmsg_cancel(msg, hdr); 1413 return -EMSGSIZE; 1414 } 1415 1416 static int net_dm_cmd_config_get(struct sk_buff *skb, struct genl_info *info) 1417 { 1418 struct sk_buff *msg; 1419 int rc; 1420 1421 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 1422 if (!msg) 1423 return -ENOMEM; 1424 1425 rc = net_dm_config_fill(msg, info); 1426 if (rc) 1427 goto free_msg; 1428 1429 return genlmsg_reply(msg, info); 1430 1431 free_msg: 1432 nlmsg_free(msg); 1433 return rc; 1434 } 1435 1436 static void net_dm_stats_read(struct net_dm_stats *stats) 1437 { 1438 int cpu; 1439 1440 memset(stats, 0, sizeof(*stats)); 1441 for_each_possible_cpu(cpu) { 1442 struct per_cpu_dm_data *data = &per_cpu(dm_cpu_data, cpu); 1443 struct net_dm_stats *cpu_stats = &data->stats; 1444 unsigned int start; 1445 u64 dropped; 1446 1447 do { 1448 start = u64_stats_fetch_begin(&cpu_stats->syncp); 1449 dropped = u64_stats_read(&cpu_stats->dropped); 1450 } while (u64_stats_fetch_retry(&cpu_stats->syncp, start)); 1451 1452 u64_stats_add(&stats->dropped, dropped); 1453 } 1454 } 1455 1456 static int net_dm_stats_put(struct sk_buff *msg) 1457 { 1458 struct net_dm_stats stats; 1459 struct nlattr *attr; 1460 1461 net_dm_stats_read(&stats); 1462 1463 attr = nla_nest_start(msg, NET_DM_ATTR_STATS); 1464 if (!attr) 1465 return -EMSGSIZE; 1466 1467 if (nla_put_u64_64bit(msg, NET_DM_ATTR_STATS_DROPPED, 1468 u64_stats_read(&stats.dropped), NET_DM_ATTR_PAD)) 1469 goto nla_put_failure; 1470 1471 nla_nest_end(msg, attr); 1472 1473 return 0; 1474 1475 nla_put_failure: 1476 nla_nest_cancel(msg, attr); 1477 return -EMSGSIZE; 1478 } 1479 1480 static void net_dm_hw_stats_read(struct net_dm_stats *stats) 1481 { 1482 int cpu; 1483 1484 memset(stats, 0, sizeof(*stats)); 1485 for_each_possible_cpu(cpu) { 1486 struct per_cpu_dm_data *hw_data = &per_cpu(dm_hw_cpu_data, cpu); 1487 struct net_dm_stats *cpu_stats = &hw_data->stats; 1488 unsigned int start; 1489 u64 dropped; 1490 1491 do { 1492 start = u64_stats_fetch_begin(&cpu_stats->syncp); 1493 dropped = u64_stats_read(&cpu_stats->dropped); 1494 } while (u64_stats_fetch_retry(&cpu_stats->syncp, start)); 1495 1496 u64_stats_add(&stats->dropped, dropped); 1497 } 1498 } 1499 1500 static int net_dm_hw_stats_put(struct sk_buff *msg) 1501 { 1502 struct net_dm_stats stats; 1503 struct nlattr *attr; 1504 1505 net_dm_hw_stats_read(&stats); 1506 1507 attr = nla_nest_start(msg, NET_DM_ATTR_HW_STATS); 1508 if (!attr) 1509 return -EMSGSIZE; 1510 1511 if (nla_put_u64_64bit(msg, NET_DM_ATTR_STATS_DROPPED, 1512 u64_stats_read(&stats.dropped), NET_DM_ATTR_PAD)) 1513 goto nla_put_failure; 1514 1515 nla_nest_end(msg, attr); 1516 1517 return 0; 1518 1519 nla_put_failure: 1520 nla_nest_cancel(msg, attr); 1521 return -EMSGSIZE; 1522 } 1523 1524 static int net_dm_stats_fill(struct sk_buff *msg, struct genl_info *info) 1525 { 1526 void *hdr; 1527 int rc; 1528 1529 hdr = genlmsg_put(msg, info->snd_portid, info->snd_seq, 1530 &net_drop_monitor_family, 0, NET_DM_CMD_STATS_NEW); 1531 if (!hdr) 1532 return -EMSGSIZE; 1533 1534 rc = net_dm_stats_put(msg); 1535 if (rc) 1536 goto nla_put_failure; 1537 1538 rc = net_dm_hw_stats_put(msg); 1539 if (rc) 1540 goto nla_put_failure; 1541 1542 genlmsg_end(msg, hdr); 1543 1544 return 0; 1545 1546 nla_put_failure: 1547 genlmsg_cancel(msg, hdr); 1548 return -EMSGSIZE; 1549 } 1550 1551 static int net_dm_cmd_stats_get(struct sk_buff *skb, struct genl_info *info) 1552 { 1553 struct sk_buff *msg; 1554 int rc; 1555 1556 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL); 1557 if (!msg) 1558 return -ENOMEM; 1559 1560 rc = net_dm_stats_fill(msg, info); 1561 if (rc) 1562 goto free_msg; 1563 1564 return genlmsg_reply(msg, info); 1565 1566 free_msg: 1567 nlmsg_free(msg); 1568 return rc; 1569 } 1570 1571 static int dropmon_net_event(struct notifier_block *ev_block, 1572 unsigned long event, void *ptr) 1573 { 1574 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 1575 struct dm_hw_stat_delta *stat; 1576 1577 switch (event) { 1578 case NETDEV_REGISTER: 1579 if (WARN_ON_ONCE(rtnl_dereference(dev->dm_private))) 1580 break; 1581 stat = kzalloc_obj(*stat); 1582 if (!stat) 1583 break; 1584 1585 stat->last_rx = jiffies; 1586 rcu_assign_pointer(dev->dm_private, stat); 1587 1588 break; 1589 case NETDEV_UNREGISTER: 1590 stat = rtnl_dereference(dev->dm_private); 1591 if (stat) { 1592 rcu_assign_pointer(dev->dm_private, NULL); 1593 kfree_rcu(stat, rcu); 1594 } 1595 break; 1596 } 1597 return NOTIFY_DONE; 1598 } 1599 1600 static const struct nla_policy net_dm_nl_policy[NET_DM_ATTR_MAX + 1] = { 1601 [NET_DM_ATTR_UNSPEC] = { .strict_start_type = NET_DM_ATTR_UNSPEC + 1 }, 1602 [NET_DM_ATTR_ALERT_MODE] = { .type = NLA_U8 }, 1603 [NET_DM_ATTR_TRUNC_LEN] = { .type = NLA_U32 }, 1604 [NET_DM_ATTR_QUEUE_LEN] = { .type = NLA_U32 }, 1605 [NET_DM_ATTR_SW_DROPS] = {. type = NLA_FLAG }, 1606 [NET_DM_ATTR_HW_DROPS] = {. type = NLA_FLAG }, 1607 }; 1608 1609 static const struct genl_small_ops dropmon_ops[] = { 1610 { 1611 .cmd = NET_DM_CMD_CONFIG, 1612 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP, 1613 .doit = net_dm_cmd_config, 1614 .flags = GENL_ADMIN_PERM, 1615 }, 1616 { 1617 .cmd = NET_DM_CMD_START, 1618 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP, 1619 .doit = net_dm_cmd_trace, 1620 .flags = GENL_ADMIN_PERM, 1621 }, 1622 { 1623 .cmd = NET_DM_CMD_STOP, 1624 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP, 1625 .doit = net_dm_cmd_trace, 1626 .flags = GENL_ADMIN_PERM, 1627 }, 1628 { 1629 .cmd = NET_DM_CMD_CONFIG_GET, 1630 .doit = net_dm_cmd_config_get, 1631 }, 1632 { 1633 .cmd = NET_DM_CMD_STATS_GET, 1634 .doit = net_dm_cmd_stats_get, 1635 }, 1636 }; 1637 1638 static int net_dm_nl_pre_doit(const struct genl_split_ops *ops, 1639 struct sk_buff *skb, struct genl_info *info) 1640 { 1641 mutex_lock(&net_dm_mutex); 1642 1643 return 0; 1644 } 1645 1646 static void net_dm_nl_post_doit(const struct genl_split_ops *ops, 1647 struct sk_buff *skb, struct genl_info *info) 1648 { 1649 mutex_unlock(&net_dm_mutex); 1650 } 1651 1652 static struct genl_family net_drop_monitor_family __ro_after_init = { 1653 .hdrsize = 0, 1654 .name = "NET_DM", 1655 .version = 2, 1656 .maxattr = NET_DM_ATTR_MAX, 1657 .policy = net_dm_nl_policy, 1658 .pre_doit = net_dm_nl_pre_doit, 1659 .post_doit = net_dm_nl_post_doit, 1660 .module = THIS_MODULE, 1661 .small_ops = dropmon_ops, 1662 .n_small_ops = ARRAY_SIZE(dropmon_ops), 1663 .resv_start_op = NET_DM_CMD_STATS_GET + 1, 1664 .mcgrps = dropmon_mcgrps, 1665 .n_mcgrps = ARRAY_SIZE(dropmon_mcgrps), 1666 }; 1667 1668 static struct notifier_block dropmon_net_notifier = { 1669 .notifier_call = dropmon_net_event 1670 }; 1671 1672 static void __net_dm_cpu_data_init(struct per_cpu_dm_data *data) 1673 { 1674 raw_spin_lock_init(&data->lock); 1675 skb_queue_head_init(&data->drop_queue); 1676 u64_stats_init(&data->stats.syncp); 1677 } 1678 1679 static void __net_dm_cpu_data_fini(struct per_cpu_dm_data *data) 1680 { 1681 WARN_ON(!skb_queue_empty(&data->drop_queue)); 1682 } 1683 1684 static void net_dm_cpu_data_init(int cpu) 1685 { 1686 struct per_cpu_dm_data *data; 1687 1688 data = &per_cpu(dm_cpu_data, cpu); 1689 __net_dm_cpu_data_init(data); 1690 } 1691 1692 static void net_dm_cpu_data_fini(int cpu) 1693 { 1694 struct per_cpu_dm_data *data; 1695 1696 data = &per_cpu(dm_cpu_data, cpu); 1697 /* At this point, we should have exclusive access 1698 * to this struct and can free the skb inside it. 1699 */ 1700 consume_skb(data->skb); 1701 __net_dm_cpu_data_fini(data); 1702 } 1703 1704 static void net_dm_hw_cpu_data_init(int cpu) 1705 { 1706 struct per_cpu_dm_data *hw_data; 1707 1708 hw_data = &per_cpu(dm_hw_cpu_data, cpu); 1709 __net_dm_cpu_data_init(hw_data); 1710 } 1711 1712 static void net_dm_hw_cpu_data_fini(int cpu) 1713 { 1714 struct per_cpu_dm_data *hw_data; 1715 1716 hw_data = &per_cpu(dm_hw_cpu_data, cpu); 1717 kfree(hw_data->hw_entries); 1718 __net_dm_cpu_data_fini(hw_data); 1719 } 1720 1721 static int __init init_net_drop_monitor(void) 1722 { 1723 int cpu, rc; 1724 1725 pr_info("Initializing network drop monitor service\n"); 1726 1727 if (sizeof(void *) > 8) { 1728 pr_err("Unable to store program counters on this arch, Drop monitor failed\n"); 1729 return -ENOSPC; 1730 } 1731 1732 for_each_possible_cpu(cpu) { 1733 net_dm_cpu_data_init(cpu); 1734 net_dm_hw_cpu_data_init(cpu); 1735 } 1736 1737 rc = register_netdevice_notifier(&dropmon_net_notifier); 1738 if (rc < 0) { 1739 pr_crit("Failed to register netdevice notifier\n"); 1740 return rc; 1741 } 1742 1743 rc = genl_register_family(&net_drop_monitor_family); 1744 if (rc) { 1745 pr_err("Could not create drop monitor netlink family\n"); 1746 goto out_unreg; 1747 } 1748 WARN_ON(net_drop_monitor_family.mcgrp_offset != NET_DM_GRP_ALERT); 1749 1750 rc = 0; 1751 1752 goto out; 1753 1754 out_unreg: 1755 WARN_ON(unregister_netdevice_notifier(&dropmon_net_notifier)); 1756 out: 1757 return rc; 1758 } 1759 1760 static void exit_net_drop_monitor(void) 1761 { 1762 int cpu; 1763 1764 /* 1765 * Because of the module_get/put we do in the trace state change path 1766 * we are guaranteed not to have any current users when we get here 1767 */ 1768 BUG_ON(genl_unregister_family(&net_drop_monitor_family)); 1769 1770 BUG_ON(unregister_netdevice_notifier(&dropmon_net_notifier)); 1771 1772 for_each_possible_cpu(cpu) { 1773 net_dm_hw_cpu_data_fini(cpu); 1774 net_dm_cpu_data_fini(cpu); 1775 } 1776 } 1777 1778 module_init(init_net_drop_monitor); 1779 module_exit(exit_net_drop_monitor); 1780 1781 MODULE_LICENSE("GPL v2"); 1782 MODULE_AUTHOR("Neil Horman <nhorman@tuxdriver.com>"); 1783 MODULE_ALIAS_GENL_FAMILY("NET_DM"); 1784 MODULE_DESCRIPTION("Monitoring code for network dropped packet alerts"); 1785