Linux Audio

Check our new training course

In-person Linux kernel drivers training

Jun 16-20, 2025
Register
Loading...
v6.8
   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 <asm/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 */
  50static int trace_state = TRACE_OFF;
  51static bool monitor_hw;
  52
  53/* net_dm_mutex
  54 *
  55 * An overall lock guarding every operation coming from userspace.
 
  56 */
  57static DEFINE_MUTEX(net_dm_mutex);
  58
  59struct 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
  66struct net_dm_hw_entry {
  67	char trap_name[NET_DM_MAX_HW_TRAP_NAME_LEN];
  68	u32 count;
  69};
  70
  71struct net_dm_hw_entries {
  72	u32 num_entries;
  73	struct net_dm_hw_entry entries[];
  74};
  75
  76struct per_cpu_dm_data {
  77	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
  90struct dm_hw_stat_delta {
 
  91	unsigned long last_rx;
  92	unsigned long last_drop_val;
  93	struct rcu_head rcu;
 
  94};
  95
  96static struct genl_family net_drop_monitor_family;
  97
  98static DEFINE_PER_CPU(struct per_cpu_dm_data, dm_cpu_data);
  99static DEFINE_PER_CPU(struct per_cpu_dm_data, dm_hw_cpu_data);
 100
 101static int dm_hit_limit = 64;
 102static int dm_delay = 1;
 103static unsigned long dm_hw_check_delta = 2*HZ;
 
 104
 105static enum net_dm_alert_mode net_dm_alert_mode = NET_DM_ALERT_MODE_SUMMARY;
 106static u32 net_dm_trunc_len;
 107static u32 net_dm_queue_len = 1000;
 108
 109struct 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	void (*napi_poll_probe)(void *ignore, struct napi_struct *napi,
 114				int work, int budget);
 115	void (*work_item_func)(struct work_struct *work);
 116	void (*hw_work_item_func)(struct work_struct *work);
 117	void (*hw_trap_probe)(void *ignore, const struct devlink *devlink,
 118			      struct sk_buff *skb,
 119			      const struct devlink_trap_metadata *metadata);
 120};
 121
 122struct net_dm_skb_cb {
 123	union {
 124		struct devlink_trap_metadata *hw_metadata;
 125		void *pc;
 126	};
 127	enum skb_drop_reason reason;
 128};
 129
 130#define NET_DM_SKB_CB(__skb) ((struct net_dm_skb_cb *)&((__skb)->cb[0]))
 131
 132static struct sk_buff *reset_per_cpu_data(struct per_cpu_dm_data *data)
 133{
 134	size_t al;
 135	struct net_dm_alert_msg *msg;
 136	struct nlattr *nla;
 137	struct sk_buff *skb;
 138	unsigned long flags;
 139	void *msg_header;
 140
 141	al = sizeof(struct net_dm_alert_msg);
 142	al += dm_hit_limit * sizeof(struct net_dm_drop_point);
 143	al += sizeof(struct nlattr);
 144
 145	skb = genlmsg_new(al, GFP_KERNEL);
 146
 147	if (!skb)
 148		goto err;
 149
 150	msg_header = genlmsg_put(skb, 0, 0, &net_drop_monitor_family,
 151				 0, NET_DM_CMD_ALERT);
 152	if (!msg_header) {
 153		nlmsg_free(skb);
 154		skb = NULL;
 155		goto err;
 156	}
 157	nla = nla_reserve(skb, NLA_UNSPEC,
 158			  sizeof(struct net_dm_alert_msg));
 159	if (!nla) {
 160		nlmsg_free(skb);
 161		skb = NULL;
 162		goto err;
 163	}
 164	msg = nla_data(nla);
 165	memset(msg, 0, al);
 166	goto out;
 167
 168err:
 169	mod_timer(&data->send_timer, jiffies + HZ / 10);
 170out:
 171	spin_lock_irqsave(&data->lock, flags);
 172	swap(data->skb, skb);
 173	spin_unlock_irqrestore(&data->lock, flags);
 174
 175	if (skb) {
 176		struct nlmsghdr *nlh = (struct nlmsghdr *)skb->data;
 177		struct genlmsghdr *gnlh = (struct genlmsghdr *)nlmsg_data(nlh);
 178
 179		genlmsg_end(skb, genlmsg_data(gnlh));
 180	}
 181
 182	return skb;
 183}
 184
 185static const struct genl_multicast_group dropmon_mcgrps[] = {
 186	{ .name = "events", .flags = GENL_MCAST_CAP_SYS_ADMIN, },
 187};
 188
 189static void send_dm_alert(struct work_struct *work)
 190{
 191	struct sk_buff *skb;
 192	struct per_cpu_dm_data *data;
 193
 194	data = container_of(work, struct per_cpu_dm_data, dm_alert_work);
 195
 196	skb = reset_per_cpu_data(data);
 197
 198	if (skb)
 199		genlmsg_multicast(&net_drop_monitor_family, skb, 0,
 200				  0, GFP_KERNEL);
 201}
 202
 203/*
 204 * This is the timer function to delay the sending of an alert
 205 * in the event that more drops will arrive during the
 206 * hysteresis period.
 207 */
 208static void sched_send_work(struct timer_list *t)
 209{
 210	struct per_cpu_dm_data *data = from_timer(data, t, send_timer);
 211
 212	schedule_work(&data->dm_alert_work);
 213}
 214
 215static void trace_drop_common(struct sk_buff *skb, void *location)
 216{
 217	struct net_dm_alert_msg *msg;
 218	struct net_dm_drop_point *point;
 219	struct nlmsghdr *nlh;
 220	struct nlattr *nla;
 221	int i;
 222	struct sk_buff *dskb;
 223	struct per_cpu_dm_data *data;
 224	unsigned long flags;
 225
 226	local_irq_save(flags);
 227	data = this_cpu_ptr(&dm_cpu_data);
 228	spin_lock(&data->lock);
 229	dskb = data->skb;
 230
 231	if (!dskb)
 232		goto out;
 233
 234	nlh = (struct nlmsghdr *)dskb->data;
 235	nla = genlmsg_data(nlmsg_data(nlh));
 236	msg = nla_data(nla);
 237	point = msg->points;
 238	for (i = 0; i < msg->entries; i++) {
 239		if (!memcmp(&location, &point->pc, sizeof(void *))) {
 240			point->count++;
 241			goto out;
 242		}
 243		point++;
 244	}
 245	if (msg->entries == dm_hit_limit)
 246		goto out;
 247	/*
 248	 * We need to create a new entry
 249	 */
 250	__nla_reserve_nohdr(dskb, sizeof(struct net_dm_drop_point));
 251	nla->nla_len += NLA_ALIGN(sizeof(struct net_dm_drop_point));
 252	memcpy(point->pc, &location, sizeof(void *));
 253	point->count = 1;
 254	msg->entries++;
 255
 256	if (!timer_pending(&data->send_timer)) {
 257		data->send_timer.expires = jiffies + dm_delay * HZ;
 258		add_timer(&data->send_timer);
 259	}
 260
 261out:
 262	spin_unlock_irqrestore(&data->lock, flags);
 263}
 264
 265static void trace_kfree_skb_hit(void *ignore, struct sk_buff *skb,
 266				void *location,
 267				enum skb_drop_reason reason)
 268{
 269	trace_drop_common(skb, location);
 270}
 271
 272static void trace_napi_poll_hit(void *ignore, struct napi_struct *napi,
 273				int work, int budget)
 274{
 275	struct net_device *dev = napi->dev;
 276	struct dm_hw_stat_delta *stat;
 277	/*
 278	 * Don't check napi structures with no associated device
 279	 */
 280	if (!dev)
 281		return;
 282
 283	rcu_read_lock();
 284	stat = rcu_dereference(dev->dm_private);
 285	if (stat) {
 286		/*
 287		 * only add a note to our monitor buffer if:
 288		 * 1) its after the last_rx delta
 289		 * 2) our rx_dropped count has gone up
 
 290		 */
 291		if (time_after(jiffies, stat->last_rx + dm_hw_check_delta) &&
 292		    (dev->stats.rx_dropped != stat->last_drop_val)) {
 
 293			trace_drop_common(NULL, NULL);
 294			stat->last_drop_val = dev->stats.rx_dropped;
 295			stat->last_rx = jiffies;
 
 296		}
 297	}
 298	rcu_read_unlock();
 299}
 300
 301static struct net_dm_hw_entries *
 302net_dm_hw_reset_per_cpu_data(struct per_cpu_dm_data *hw_data)
 303{
 304	struct net_dm_hw_entries *hw_entries;
 305	unsigned long flags;
 306
 307	hw_entries = kzalloc(struct_size(hw_entries, entries, dm_hit_limit),
 308			     GFP_KERNEL);
 309	if (!hw_entries) {
 310		/* If the memory allocation failed, we try to perform another
 311		 * allocation in 1/10 second. Otherwise, the probe function
 312		 * will constantly bail out.
 313		 */
 314		mod_timer(&hw_data->send_timer, jiffies + HZ / 10);
 315	}
 316
 317	spin_lock_irqsave(&hw_data->lock, flags);
 318	swap(hw_data->hw_entries, hw_entries);
 319	spin_unlock_irqrestore(&hw_data->lock, flags);
 320
 321	return hw_entries;
 322}
 323
 324static int net_dm_hw_entry_put(struct sk_buff *msg,
 325			       const struct net_dm_hw_entry *hw_entry)
 326{
 327	struct nlattr *attr;
 328
 329	attr = nla_nest_start(msg, NET_DM_ATTR_HW_ENTRY);
 330	if (!attr)
 331		return -EMSGSIZE;
 332
 333	if (nla_put_string(msg, NET_DM_ATTR_HW_TRAP_NAME, hw_entry->trap_name))
 334		goto nla_put_failure;
 335
 336	if (nla_put_u32(msg, NET_DM_ATTR_HW_TRAP_COUNT, hw_entry->count))
 337		goto nla_put_failure;
 338
 339	nla_nest_end(msg, attr);
 340
 341	return 0;
 342
 343nla_put_failure:
 344	nla_nest_cancel(msg, attr);
 345	return -EMSGSIZE;
 346}
 347
 348static int net_dm_hw_entries_put(struct sk_buff *msg,
 349				 const struct net_dm_hw_entries *hw_entries)
 350{
 351	struct nlattr *attr;
 352	int i;
 353
 354	attr = nla_nest_start(msg, NET_DM_ATTR_HW_ENTRIES);
 355	if (!attr)
 356		return -EMSGSIZE;
 357
 358	for (i = 0; i < hw_entries->num_entries; i++) {
 359		int rc;
 360
 361		rc = net_dm_hw_entry_put(msg, &hw_entries->entries[i]);
 362		if (rc)
 363			goto nla_put_failure;
 364	}
 365
 366	nla_nest_end(msg, attr);
 367
 368	return 0;
 369
 370nla_put_failure:
 371	nla_nest_cancel(msg, attr);
 372	return -EMSGSIZE;
 373}
 374
 375static int
 376net_dm_hw_summary_report_fill(struct sk_buff *msg,
 377			      const struct net_dm_hw_entries *hw_entries)
 378{
 379	struct net_dm_alert_msg anc_hdr = { 0 };
 380	void *hdr;
 381	int rc;
 382
 383	hdr = genlmsg_put(msg, 0, 0, &net_drop_monitor_family, 0,
 384			  NET_DM_CMD_ALERT);
 385	if (!hdr)
 386		return -EMSGSIZE;
 387
 388	/* We need to put the ancillary header in order not to break user
 389	 * space.
 390	 */
 391	if (nla_put(msg, NLA_UNSPEC, sizeof(anc_hdr), &anc_hdr))
 392		goto nla_put_failure;
 393
 394	rc = net_dm_hw_entries_put(msg, hw_entries);
 395	if (rc)
 396		goto nla_put_failure;
 397
 398	genlmsg_end(msg, hdr);
 399
 400	return 0;
 401
 402nla_put_failure:
 403	genlmsg_cancel(msg, hdr);
 404	return -EMSGSIZE;
 405}
 406
 407static void net_dm_hw_summary_work(struct work_struct *work)
 408{
 409	struct net_dm_hw_entries *hw_entries;
 410	struct per_cpu_dm_data *hw_data;
 411	struct sk_buff *msg;
 412	int rc;
 413
 414	hw_data = container_of(work, struct per_cpu_dm_data, dm_alert_work);
 415
 416	hw_entries = net_dm_hw_reset_per_cpu_data(hw_data);
 417	if (!hw_entries)
 418		return;
 419
 420	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
 421	if (!msg)
 422		goto out;
 423
 424	rc = net_dm_hw_summary_report_fill(msg, hw_entries);
 425	if (rc) {
 426		nlmsg_free(msg);
 427		goto out;
 428	}
 429
 430	genlmsg_multicast(&net_drop_monitor_family, msg, 0, 0, GFP_KERNEL);
 431
 432out:
 433	kfree(hw_entries);
 434}
 435
 436static void
 437net_dm_hw_trap_summary_probe(void *ignore, const struct devlink *devlink,
 438			     struct sk_buff *skb,
 439			     const struct devlink_trap_metadata *metadata)
 440{
 441	struct net_dm_hw_entries *hw_entries;
 442	struct net_dm_hw_entry *hw_entry;
 443	struct per_cpu_dm_data *hw_data;
 444	unsigned long flags;
 445	int i;
 446
 447	if (metadata->trap_type == DEVLINK_TRAP_TYPE_CONTROL)
 448		return;
 449
 450	hw_data = this_cpu_ptr(&dm_hw_cpu_data);
 451	spin_lock_irqsave(&hw_data->lock, flags);
 452	hw_entries = hw_data->hw_entries;
 453
 454	if (!hw_entries)
 455		goto out;
 456
 457	for (i = 0; i < hw_entries->num_entries; i++) {
 458		hw_entry = &hw_entries->entries[i];
 459		if (!strncmp(hw_entry->trap_name, metadata->trap_name,
 460			     NET_DM_MAX_HW_TRAP_NAME_LEN - 1)) {
 461			hw_entry->count++;
 462			goto out;
 463		}
 464	}
 465	if (WARN_ON_ONCE(hw_entries->num_entries == dm_hit_limit))
 466		goto out;
 467
 468	hw_entry = &hw_entries->entries[hw_entries->num_entries];
 469	strscpy(hw_entry->trap_name, metadata->trap_name,
 470		NET_DM_MAX_HW_TRAP_NAME_LEN - 1);
 471	hw_entry->count = 1;
 472	hw_entries->num_entries++;
 473
 474	if (!timer_pending(&hw_data->send_timer)) {
 475		hw_data->send_timer.expires = jiffies + dm_delay * HZ;
 476		add_timer(&hw_data->send_timer);
 477	}
 478
 479out:
 480	spin_unlock_irqrestore(&hw_data->lock, flags);
 481}
 482
 483static const struct net_dm_alert_ops net_dm_alert_summary_ops = {
 484	.kfree_skb_probe	= trace_kfree_skb_hit,
 485	.napi_poll_probe	= trace_napi_poll_hit,
 486	.work_item_func		= send_dm_alert,
 487	.hw_work_item_func	= net_dm_hw_summary_work,
 488	.hw_trap_probe		= net_dm_hw_trap_summary_probe,
 489};
 490
 491static void net_dm_packet_trace_kfree_skb_hit(void *ignore,
 492					      struct sk_buff *skb,
 493					      void *location,
 494					      enum skb_drop_reason reason)
 495{
 496	ktime_t tstamp = ktime_get_real();
 497	struct per_cpu_dm_data *data;
 498	struct net_dm_skb_cb *cb;
 499	struct sk_buff *nskb;
 500	unsigned long flags;
 501
 502	if (!skb_mac_header_was_set(skb))
 503		return;
 504
 505	nskb = skb_clone(skb, GFP_ATOMIC);
 506	if (!nskb)
 507		return;
 508
 509	cb = NET_DM_SKB_CB(nskb);
 510	cb->reason = reason;
 511	cb->pc = location;
 512	/* Override the timestamp because we care about the time when the
 513	 * packet was dropped.
 514	 */
 515	nskb->tstamp = tstamp;
 516
 517	data = this_cpu_ptr(&dm_cpu_data);
 518
 519	spin_lock_irqsave(&data->drop_queue.lock, flags);
 520	if (skb_queue_len(&data->drop_queue) < net_dm_queue_len)
 521		__skb_queue_tail(&data->drop_queue, nskb);
 522	else
 523		goto unlock_free;
 524	spin_unlock_irqrestore(&data->drop_queue.lock, flags);
 525
 526	schedule_work(&data->dm_alert_work);
 527
 528	return;
 529
 530unlock_free:
 531	spin_unlock_irqrestore(&data->drop_queue.lock, flags);
 532	u64_stats_update_begin(&data->stats.syncp);
 533	u64_stats_inc(&data->stats.dropped);
 534	u64_stats_update_end(&data->stats.syncp);
 535	consume_skb(nskb);
 536}
 537
 538static void net_dm_packet_trace_napi_poll_hit(void *ignore,
 539					      struct napi_struct *napi,
 540					      int work, int budget)
 541{
 542}
 543
 544static size_t net_dm_in_port_size(void)
 545{
 546	       /* NET_DM_ATTR_IN_PORT nest */
 547	return nla_total_size(0) +
 548	       /* NET_DM_ATTR_PORT_NETDEV_IFINDEX */
 549	       nla_total_size(sizeof(u32)) +
 550	       /* NET_DM_ATTR_PORT_NETDEV_NAME */
 551	       nla_total_size(IFNAMSIZ + 1);
 552}
 553
 554#define NET_DM_MAX_SYMBOL_LEN 40
 555#define NET_DM_MAX_REASON_LEN 50
 556
 557static size_t net_dm_packet_report_size(size_t payload_len)
 558{
 559	size_t size;
 560
 561	size = nlmsg_msg_size(GENL_HDRLEN + net_drop_monitor_family.hdrsize);
 562
 563	return NLMSG_ALIGN(size) +
 564	       /* NET_DM_ATTR_ORIGIN */
 565	       nla_total_size(sizeof(u16)) +
 566	       /* NET_DM_ATTR_PC */
 567	       nla_total_size(sizeof(u64)) +
 568	       /* NET_DM_ATTR_SYMBOL */
 569	       nla_total_size(NET_DM_MAX_SYMBOL_LEN + 1) +
 570	       /* NET_DM_ATTR_IN_PORT */
 571	       net_dm_in_port_size() +
 572	       /* NET_DM_ATTR_TIMESTAMP */
 573	       nla_total_size(sizeof(u64)) +
 574	       /* NET_DM_ATTR_ORIG_LEN */
 575	       nla_total_size(sizeof(u32)) +
 576	       /* NET_DM_ATTR_PROTO */
 577	       nla_total_size(sizeof(u16)) +
 578	       /* NET_DM_ATTR_REASON */
 579	       nla_total_size(NET_DM_MAX_REASON_LEN + 1) +
 580	       /* NET_DM_ATTR_PAYLOAD */
 581	       nla_total_size(payload_len);
 582}
 583
 584static int net_dm_packet_report_in_port_put(struct sk_buff *msg, int ifindex,
 585					    const char *name)
 586{
 587	struct nlattr *attr;
 588
 589	attr = nla_nest_start(msg, NET_DM_ATTR_IN_PORT);
 590	if (!attr)
 591		return -EMSGSIZE;
 592
 593	if (ifindex &&
 594	    nla_put_u32(msg, NET_DM_ATTR_PORT_NETDEV_IFINDEX, ifindex))
 595		goto nla_put_failure;
 596
 597	if (name && nla_put_string(msg, NET_DM_ATTR_PORT_NETDEV_NAME, name))
 598		goto nla_put_failure;
 599
 600	nla_nest_end(msg, attr);
 601
 602	return 0;
 603
 604nla_put_failure:
 605	nla_nest_cancel(msg, attr);
 606	return -EMSGSIZE;
 607}
 608
 609static int net_dm_packet_report_fill(struct sk_buff *msg, struct sk_buff *skb,
 610				     size_t payload_len)
 611{
 612	struct net_dm_skb_cb *cb = NET_DM_SKB_CB(skb);
 613	const struct drop_reason_list *list = NULL;
 614	unsigned int subsys, subsys_reason;
 615	char buf[NET_DM_MAX_SYMBOL_LEN];
 616	struct nlattr *attr;
 617	void *hdr;
 618	int rc;
 619
 620	hdr = genlmsg_put(msg, 0, 0, &net_drop_monitor_family, 0,
 621			  NET_DM_CMD_PACKET_ALERT);
 622	if (!hdr)
 623		return -EMSGSIZE;
 624
 625	if (nla_put_u16(msg, NET_DM_ATTR_ORIGIN, NET_DM_ORIGIN_SW))
 626		goto nla_put_failure;
 627
 628	if (nla_put_u64_64bit(msg, NET_DM_ATTR_PC, (u64)(uintptr_t)cb->pc,
 629			      NET_DM_ATTR_PAD))
 630		goto nla_put_failure;
 631
 632	rcu_read_lock();
 633	subsys = u32_get_bits(cb->reason, SKB_DROP_REASON_SUBSYS_MASK);
 634	if (subsys < SKB_DROP_REASON_SUBSYS_NUM)
 635		list = rcu_dereference(drop_reasons_by_subsys[subsys]);
 636	subsys_reason = cb->reason & ~SKB_DROP_REASON_SUBSYS_MASK;
 637	if (!list ||
 638	    subsys_reason >= list->n_reasons ||
 639	    !list->reasons[subsys_reason] ||
 640	    strlen(list->reasons[subsys_reason]) > NET_DM_MAX_REASON_LEN) {
 641		list = rcu_dereference(drop_reasons_by_subsys[SKB_DROP_REASON_SUBSYS_CORE]);
 642		subsys_reason = SKB_DROP_REASON_NOT_SPECIFIED;
 643	}
 644	if (nla_put_string(msg, NET_DM_ATTR_REASON,
 645			   list->reasons[subsys_reason])) {
 646		rcu_read_unlock();
 647		goto nla_put_failure;
 648	}
 649	rcu_read_unlock();
 650
 651	snprintf(buf, sizeof(buf), "%pS", cb->pc);
 652	if (nla_put_string(msg, NET_DM_ATTR_SYMBOL, buf))
 653		goto nla_put_failure;
 654
 655	rc = net_dm_packet_report_in_port_put(msg, skb->skb_iif, NULL);
 656	if (rc)
 657		goto nla_put_failure;
 658
 659	if (nla_put_u64_64bit(msg, NET_DM_ATTR_TIMESTAMP,
 660			      ktime_to_ns(skb->tstamp), NET_DM_ATTR_PAD))
 661		goto nla_put_failure;
 662
 663	if (nla_put_u32(msg, NET_DM_ATTR_ORIG_LEN, skb->len))
 664		goto nla_put_failure;
 665
 666	if (!payload_len)
 667		goto out;
 668
 669	if (nla_put_u16(msg, NET_DM_ATTR_PROTO, be16_to_cpu(skb->protocol)))
 670		goto nla_put_failure;
 671
 672	attr = skb_put(msg, nla_total_size(payload_len));
 673	attr->nla_type = NET_DM_ATTR_PAYLOAD;
 674	attr->nla_len = nla_attr_size(payload_len);
 675	if (skb_copy_bits(skb, 0, nla_data(attr), payload_len))
 676		goto nla_put_failure;
 677
 678out:
 679	genlmsg_end(msg, hdr);
 680
 681	return 0;
 682
 683nla_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
 690static 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
 719out:
 720	consume_skb(skb);
 721}
 722
 723static 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
 742static size_t
 743net_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
 749static size_t
 750net_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(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
 778static 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 = skb_put(msg, nla_total_size(payload_len));
 833	attr->nla_type = NET_DM_ATTR_PAYLOAD;
 834	attr->nla_len = nla_attr_size(payload_len);
 835	if (skb_copy_bits(skb, 0, nla_data(attr), payload_len))
 836		goto nla_put_failure;
 837
 838out:
 839	genlmsg_end(msg, hdr);
 840
 841	return 0;
 842
 843nla_put_failure:
 844	genlmsg_cancel(msg, hdr);
 845	return -EMSGSIZE;
 846}
 847
 848static struct devlink_trap_metadata *
 849net_dm_hw_metadata_copy(const struct devlink_trap_metadata *metadata)
 850{
 851	const struct flow_action_cookie *fa_cookie;
 852	struct devlink_trap_metadata *hw_metadata;
 853	const char *trap_group_name;
 854	const char *trap_name;
 855
 856	hw_metadata = kzalloc(sizeof(*hw_metadata), GFP_ATOMIC);
 857	if (!hw_metadata)
 858		return NULL;
 859
 860	trap_group_name = kstrdup(metadata->trap_group_name, GFP_ATOMIC);
 861	if (!trap_group_name)
 862		goto free_hw_metadata;
 863	hw_metadata->trap_group_name = trap_group_name;
 864
 865	trap_name = kstrdup(metadata->trap_name, GFP_ATOMIC);
 866	if (!trap_name)
 867		goto free_trap_group;
 868	hw_metadata->trap_name = trap_name;
 869
 870	if (metadata->fa_cookie) {
 871		size_t cookie_size = sizeof(*fa_cookie) +
 872				     metadata->fa_cookie->cookie_len;
 873
 874		fa_cookie = kmemdup(metadata->fa_cookie, cookie_size,
 875				    GFP_ATOMIC);
 876		if (!fa_cookie)
 877			goto free_trap_name;
 878		hw_metadata->fa_cookie = fa_cookie;
 879	}
 880
 881	hw_metadata->input_dev = metadata->input_dev;
 882	netdev_hold(hw_metadata->input_dev, &hw_metadata->dev_tracker,
 883		    GFP_ATOMIC);
 884
 885	return hw_metadata;
 886
 887free_trap_name:
 888	kfree(trap_name);
 889free_trap_group:
 890	kfree(trap_group_name);
 891free_hw_metadata:
 892	kfree(hw_metadata);
 893	return NULL;
 894}
 895
 896static void
 897net_dm_hw_metadata_free(struct devlink_trap_metadata *hw_metadata)
 898{
 899	netdev_put(hw_metadata->input_dev, &hw_metadata->dev_tracker);
 
 900	kfree(hw_metadata->fa_cookie);
 901	kfree(hw_metadata->trap_name);
 902	kfree(hw_metadata->trap_group_name);
 903	kfree(hw_metadata);
 904}
 905
 906static void net_dm_hw_packet_report(struct sk_buff *skb)
 907{
 908	struct devlink_trap_metadata *hw_metadata;
 909	struct sk_buff *msg;
 910	size_t payload_len;
 911	int rc;
 912
 913	if (skb->data > skb_mac_header(skb))
 914		skb_push(skb, skb->data - skb_mac_header(skb));
 915	else
 916		skb_pull(skb, skb_mac_header(skb) - skb->data);
 917
 918	payload_len = min_t(size_t, skb->len, NET_DM_MAX_PACKET_SIZE);
 919	if (net_dm_trunc_len)
 920		payload_len = min_t(size_t, net_dm_trunc_len, payload_len);
 921
 922	hw_metadata = NET_DM_SKB_CB(skb)->hw_metadata;
 923	msg = nlmsg_new(net_dm_hw_packet_report_size(payload_len, hw_metadata),
 924			GFP_KERNEL);
 925	if (!msg)
 926		goto out;
 927
 928	rc = net_dm_hw_packet_report_fill(msg, skb, payload_len);
 929	if (rc) {
 930		nlmsg_free(msg);
 931		goto out;
 932	}
 933
 934	genlmsg_multicast(&net_drop_monitor_family, msg, 0, 0, GFP_KERNEL);
 935
 936out:
 937	net_dm_hw_metadata_free(NET_DM_SKB_CB(skb)->hw_metadata);
 938	consume_skb(skb);
 939}
 940
 941static void net_dm_hw_packet_work(struct work_struct *work)
 942{
 943	struct per_cpu_dm_data *hw_data;
 944	struct sk_buff_head list;
 945	struct sk_buff *skb;
 946	unsigned long flags;
 947
 948	hw_data = container_of(work, struct per_cpu_dm_data, dm_alert_work);
 949
 950	__skb_queue_head_init(&list);
 951
 952	spin_lock_irqsave(&hw_data->drop_queue.lock, flags);
 953	skb_queue_splice_tail_init(&hw_data->drop_queue, &list);
 954	spin_unlock_irqrestore(&hw_data->drop_queue.lock, flags);
 955
 956	while ((skb = __skb_dequeue(&list)))
 957		net_dm_hw_packet_report(skb);
 958}
 959
 960static void
 961net_dm_hw_trap_packet_probe(void *ignore, const struct devlink *devlink,
 962			    struct sk_buff *skb,
 963			    const struct devlink_trap_metadata *metadata)
 964{
 965	struct devlink_trap_metadata *n_hw_metadata;
 966	ktime_t tstamp = ktime_get_real();
 967	struct per_cpu_dm_data *hw_data;
 968	struct sk_buff *nskb;
 969	unsigned long flags;
 970
 971	if (metadata->trap_type == DEVLINK_TRAP_TYPE_CONTROL)
 972		return;
 973
 974	if (!skb_mac_header_was_set(skb))
 975		return;
 976
 977	nskb = skb_clone(skb, GFP_ATOMIC);
 978	if (!nskb)
 979		return;
 980
 981	n_hw_metadata = net_dm_hw_metadata_copy(metadata);
 982	if (!n_hw_metadata)
 983		goto free;
 984
 985	NET_DM_SKB_CB(nskb)->hw_metadata = n_hw_metadata;
 986	nskb->tstamp = tstamp;
 987
 988	hw_data = this_cpu_ptr(&dm_hw_cpu_data);
 989
 990	spin_lock_irqsave(&hw_data->drop_queue.lock, flags);
 991	if (skb_queue_len(&hw_data->drop_queue) < net_dm_queue_len)
 992		__skb_queue_tail(&hw_data->drop_queue, nskb);
 993	else
 994		goto unlock_free;
 995	spin_unlock_irqrestore(&hw_data->drop_queue.lock, flags);
 996
 997	schedule_work(&hw_data->dm_alert_work);
 998
 999	return;
1000
1001unlock_free:
1002	spin_unlock_irqrestore(&hw_data->drop_queue.lock, flags);
1003	u64_stats_update_begin(&hw_data->stats.syncp);
1004	u64_stats_inc(&hw_data->stats.dropped);
1005	u64_stats_update_end(&hw_data->stats.syncp);
1006	net_dm_hw_metadata_free(n_hw_metadata);
1007free:
1008	consume_skb(nskb);
1009}
1010
1011static const struct net_dm_alert_ops net_dm_alert_packet_ops = {
1012	.kfree_skb_probe	= net_dm_packet_trace_kfree_skb_hit,
1013	.napi_poll_probe	= net_dm_packet_trace_napi_poll_hit,
1014	.work_item_func		= net_dm_packet_work,
1015	.hw_work_item_func	= net_dm_hw_packet_work,
1016	.hw_trap_probe		= net_dm_hw_trap_packet_probe,
1017};
1018
1019static const struct net_dm_alert_ops *net_dm_alert_ops_arr[] = {
1020	[NET_DM_ALERT_MODE_SUMMARY]	= &net_dm_alert_summary_ops,
1021	[NET_DM_ALERT_MODE_PACKET]	= &net_dm_alert_packet_ops,
1022};
1023
1024#if IS_ENABLED(CONFIG_NET_DEVLINK)
1025static int net_dm_hw_probe_register(const struct net_dm_alert_ops *ops)
1026{
1027	return register_trace_devlink_trap_report(ops->hw_trap_probe, NULL);
1028}
1029
1030static void net_dm_hw_probe_unregister(const struct net_dm_alert_ops *ops)
1031{
1032	unregister_trace_devlink_trap_report(ops->hw_trap_probe, NULL);
1033	tracepoint_synchronize_unregister();
1034}
1035#else
1036static int net_dm_hw_probe_register(const struct net_dm_alert_ops *ops)
1037{
1038	return -EOPNOTSUPP;
1039}
1040
1041static void net_dm_hw_probe_unregister(const struct net_dm_alert_ops *ops)
1042{
 
 
1043}
1044#endif
1045
1046static int net_dm_hw_monitor_start(struct netlink_ext_ack *extack)
1047{
1048	const struct net_dm_alert_ops *ops;
1049	int cpu, rc;
1050
1051	if (monitor_hw) {
1052		NL_SET_ERR_MSG_MOD(extack, "Hardware monitoring already enabled");
1053		return -EAGAIN;
1054	}
1055
1056	ops = net_dm_alert_ops_arr[net_dm_alert_mode];
1057
1058	if (!try_module_get(THIS_MODULE)) {
1059		NL_SET_ERR_MSG_MOD(extack, "Failed to take reference on module");
1060		return -ENODEV;
1061	}
1062
1063	for_each_possible_cpu(cpu) {
1064		struct per_cpu_dm_data *hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1065		struct net_dm_hw_entries *hw_entries;
1066
1067		INIT_WORK(&hw_data->dm_alert_work, ops->hw_work_item_func);
1068		timer_setup(&hw_data->send_timer, sched_send_work, 0);
1069		hw_entries = net_dm_hw_reset_per_cpu_data(hw_data);
1070		kfree(hw_entries);
1071	}
1072
1073	rc = net_dm_hw_probe_register(ops);
1074	if (rc) {
1075		NL_SET_ERR_MSG_MOD(extack, "Failed to connect probe to devlink_trap_probe() tracepoint");
1076		goto err_module_put;
1077	}
1078
1079	monitor_hw = true;
1080
1081	return 0;
1082
1083err_module_put:
1084	for_each_possible_cpu(cpu) {
1085		struct per_cpu_dm_data *hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1086		struct sk_buff *skb;
1087
1088		del_timer_sync(&hw_data->send_timer);
1089		cancel_work_sync(&hw_data->dm_alert_work);
1090		while ((skb = __skb_dequeue(&hw_data->drop_queue))) {
1091			struct devlink_trap_metadata *hw_metadata;
1092
1093			hw_metadata = NET_DM_SKB_CB(skb)->hw_metadata;
1094			net_dm_hw_metadata_free(hw_metadata);
1095			consume_skb(skb);
1096		}
1097	}
1098	module_put(THIS_MODULE);
1099	return rc;
1100}
1101
1102static void net_dm_hw_monitor_stop(struct netlink_ext_ack *extack)
1103{
1104	const struct net_dm_alert_ops *ops;
1105	int cpu;
1106
1107	if (!monitor_hw) {
1108		NL_SET_ERR_MSG_MOD(extack, "Hardware monitoring already disabled");
1109		return;
1110	}
1111
1112	ops = net_dm_alert_ops_arr[net_dm_alert_mode];
1113
1114	monitor_hw = false;
1115
1116	net_dm_hw_probe_unregister(ops);
 
 
 
1117
1118	for_each_possible_cpu(cpu) {
1119		struct per_cpu_dm_data *hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1120		struct sk_buff *skb;
1121
1122		del_timer_sync(&hw_data->send_timer);
1123		cancel_work_sync(&hw_data->dm_alert_work);
1124		while ((skb = __skb_dequeue(&hw_data->drop_queue))) {
1125			struct devlink_trap_metadata *hw_metadata;
1126
1127			hw_metadata = NET_DM_SKB_CB(skb)->hw_metadata;
1128			net_dm_hw_metadata_free(hw_metadata);
1129			consume_skb(skb);
1130		}
1131	}
1132
1133	module_put(THIS_MODULE);
1134}
1135
1136static int net_dm_trace_on_set(struct netlink_ext_ack *extack)
1137{
1138	const struct net_dm_alert_ops *ops;
1139	int cpu, rc;
1140
1141	ops = net_dm_alert_ops_arr[net_dm_alert_mode];
1142
1143	if (!try_module_get(THIS_MODULE)) {
1144		NL_SET_ERR_MSG_MOD(extack, "Failed to take reference on module");
1145		return -ENODEV;
1146	}
1147
1148	for_each_possible_cpu(cpu) {
1149		struct per_cpu_dm_data *data = &per_cpu(dm_cpu_data, cpu);
1150		struct sk_buff *skb;
1151
1152		INIT_WORK(&data->dm_alert_work, ops->work_item_func);
1153		timer_setup(&data->send_timer, sched_send_work, 0);
1154		/* Allocate a new per-CPU skb for the summary alert message and
1155		 * free the old one which might contain stale data from
1156		 * previous tracing.
1157		 */
1158		skb = reset_per_cpu_data(data);
1159		consume_skb(skb);
1160	}
1161
1162	rc = register_trace_kfree_skb(ops->kfree_skb_probe, NULL);
1163	if (rc) {
1164		NL_SET_ERR_MSG_MOD(extack, "Failed to connect probe to kfree_skb() tracepoint");
1165		goto err_module_put;
1166	}
1167
1168	rc = register_trace_napi_poll(ops->napi_poll_probe, NULL);
1169	if (rc) {
1170		NL_SET_ERR_MSG_MOD(extack, "Failed to connect probe to napi_poll() tracepoint");
1171		goto err_unregister_trace;
1172	}
1173
1174	return 0;
1175
1176err_unregister_trace:
1177	unregister_trace_kfree_skb(ops->kfree_skb_probe, NULL);
1178err_module_put:
1179	for_each_possible_cpu(cpu) {
1180		struct per_cpu_dm_data *data = &per_cpu(dm_cpu_data, cpu);
1181		struct sk_buff *skb;
1182
1183		del_timer_sync(&data->send_timer);
1184		cancel_work_sync(&data->dm_alert_work);
1185		while ((skb = __skb_dequeue(&data->drop_queue)))
1186			consume_skb(skb);
1187	}
1188	module_put(THIS_MODULE);
1189	return rc;
1190}
1191
1192static void net_dm_trace_off_set(void)
1193{
 
1194	const struct net_dm_alert_ops *ops;
1195	int cpu;
1196
1197	ops = net_dm_alert_ops_arr[net_dm_alert_mode];
1198
1199	unregister_trace_napi_poll(ops->napi_poll_probe, NULL);
1200	unregister_trace_kfree_skb(ops->kfree_skb_probe, NULL);
1201
1202	tracepoint_synchronize_unregister();
1203
1204	/* Make sure we do not send notifications to user space after request
1205	 * to stop tracing returns.
1206	 */
1207	for_each_possible_cpu(cpu) {
1208		struct per_cpu_dm_data *data = &per_cpu(dm_cpu_data, cpu);
1209		struct sk_buff *skb;
1210
1211		del_timer_sync(&data->send_timer);
1212		cancel_work_sync(&data->dm_alert_work);
1213		while ((skb = __skb_dequeue(&data->drop_queue)))
1214			consume_skb(skb);
1215	}
1216
 
 
 
 
 
 
 
1217	module_put(THIS_MODULE);
1218}
1219
1220static int set_all_monitor_traces(int state, struct netlink_ext_ack *extack)
1221{
1222	int rc = 0;
1223
1224	if (state == trace_state) {
1225		NL_SET_ERR_MSG_MOD(extack, "Trace state already set to requested state");
1226		return -EAGAIN;
1227	}
1228
1229	switch (state) {
1230	case TRACE_ON:
1231		rc = net_dm_trace_on_set(extack);
1232		break;
1233	case TRACE_OFF:
1234		net_dm_trace_off_set();
1235		break;
1236	default:
1237		rc = 1;
1238		break;
1239	}
1240
1241	if (!rc)
1242		trace_state = state;
1243	else
1244		rc = -EINPROGRESS;
1245
1246	return rc;
1247}
1248
1249static bool net_dm_is_monitoring(void)
1250{
1251	return trace_state == TRACE_ON || monitor_hw;
1252}
1253
1254static int net_dm_alert_mode_get_from_info(struct genl_info *info,
1255					   enum net_dm_alert_mode *p_alert_mode)
1256{
1257	u8 val;
1258
1259	val = nla_get_u8(info->attrs[NET_DM_ATTR_ALERT_MODE]);
1260
1261	switch (val) {
1262	case NET_DM_ALERT_MODE_SUMMARY:
1263	case NET_DM_ALERT_MODE_PACKET:
1264		*p_alert_mode = val;
1265		break;
1266	default:
1267		return -EINVAL;
1268	}
1269
1270	return 0;
1271}
1272
1273static int net_dm_alert_mode_set(struct genl_info *info)
1274{
1275	struct netlink_ext_ack *extack = info->extack;
1276	enum net_dm_alert_mode alert_mode;
1277	int rc;
1278
1279	if (!info->attrs[NET_DM_ATTR_ALERT_MODE])
1280		return 0;
1281
1282	rc = net_dm_alert_mode_get_from_info(info, &alert_mode);
1283	if (rc) {
1284		NL_SET_ERR_MSG_MOD(extack, "Invalid alert mode");
1285		return -EINVAL;
1286	}
1287
1288	net_dm_alert_mode = alert_mode;
1289
1290	return 0;
1291}
1292
1293static void net_dm_trunc_len_set(struct genl_info *info)
1294{
1295	if (!info->attrs[NET_DM_ATTR_TRUNC_LEN])
1296		return;
1297
1298	net_dm_trunc_len = nla_get_u32(info->attrs[NET_DM_ATTR_TRUNC_LEN]);
1299}
1300
1301static void net_dm_queue_len_set(struct genl_info *info)
1302{
1303	if (!info->attrs[NET_DM_ATTR_QUEUE_LEN])
1304		return;
1305
1306	net_dm_queue_len = nla_get_u32(info->attrs[NET_DM_ATTR_QUEUE_LEN]);
1307}
1308
1309static int net_dm_cmd_config(struct sk_buff *skb,
1310			struct genl_info *info)
1311{
1312	struct netlink_ext_ack *extack = info->extack;
1313	int rc;
1314
1315	if (net_dm_is_monitoring()) {
1316		NL_SET_ERR_MSG_MOD(extack, "Cannot configure drop monitor during monitoring");
1317		return -EBUSY;
1318	}
1319
1320	rc = net_dm_alert_mode_set(info);
1321	if (rc)
1322		return rc;
1323
1324	net_dm_trunc_len_set(info);
1325
1326	net_dm_queue_len_set(info);
1327
1328	return 0;
1329}
1330
1331static int net_dm_monitor_start(bool set_sw, bool set_hw,
1332				struct netlink_ext_ack *extack)
1333{
1334	bool sw_set = false;
1335	int rc;
1336
1337	if (set_sw) {
1338		rc = set_all_monitor_traces(TRACE_ON, extack);
1339		if (rc)
1340			return rc;
1341		sw_set = true;
1342	}
1343
1344	if (set_hw) {
1345		rc = net_dm_hw_monitor_start(extack);
1346		if (rc)
1347			goto err_monitor_hw;
1348	}
1349
1350	return 0;
1351
1352err_monitor_hw:
1353	if (sw_set)
1354		set_all_monitor_traces(TRACE_OFF, extack);
1355	return rc;
1356}
1357
1358static void net_dm_monitor_stop(bool set_sw, bool set_hw,
1359				struct netlink_ext_ack *extack)
1360{
1361	if (set_hw)
1362		net_dm_hw_monitor_stop(extack);
1363	if (set_sw)
1364		set_all_monitor_traces(TRACE_OFF, extack);
1365}
1366
1367static int net_dm_cmd_trace(struct sk_buff *skb,
1368			struct genl_info *info)
1369{
1370	bool set_sw = !!info->attrs[NET_DM_ATTR_SW_DROPS];
1371	bool set_hw = !!info->attrs[NET_DM_ATTR_HW_DROPS];
1372	struct netlink_ext_ack *extack = info->extack;
1373
1374	/* To maintain backward compatibility, we start / stop monitoring of
1375	 * software drops if no flag is specified.
1376	 */
1377	if (!set_sw && !set_hw)
1378		set_sw = true;
1379
1380	switch (info->genlhdr->cmd) {
1381	case NET_DM_CMD_START:
1382		return net_dm_monitor_start(set_sw, set_hw, extack);
1383	case NET_DM_CMD_STOP:
1384		net_dm_monitor_stop(set_sw, set_hw, extack);
1385		return 0;
1386	}
1387
1388	return -EOPNOTSUPP;
1389}
1390
1391static int net_dm_config_fill(struct sk_buff *msg, struct genl_info *info)
1392{
1393	void *hdr;
1394
1395	hdr = genlmsg_put(msg, info->snd_portid, info->snd_seq,
1396			  &net_drop_monitor_family, 0, NET_DM_CMD_CONFIG_NEW);
1397	if (!hdr)
1398		return -EMSGSIZE;
1399
1400	if (nla_put_u8(msg, NET_DM_ATTR_ALERT_MODE, net_dm_alert_mode))
1401		goto nla_put_failure;
1402
1403	if (nla_put_u32(msg, NET_DM_ATTR_TRUNC_LEN, net_dm_trunc_len))
1404		goto nla_put_failure;
1405
1406	if (nla_put_u32(msg, NET_DM_ATTR_QUEUE_LEN, net_dm_queue_len))
1407		goto nla_put_failure;
1408
1409	genlmsg_end(msg, hdr);
1410
1411	return 0;
1412
1413nla_put_failure:
1414	genlmsg_cancel(msg, hdr);
1415	return -EMSGSIZE;
1416}
1417
1418static int net_dm_cmd_config_get(struct sk_buff *skb, struct genl_info *info)
1419{
1420	struct sk_buff *msg;
1421	int rc;
1422
1423	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1424	if (!msg)
1425		return -ENOMEM;
1426
1427	rc = net_dm_config_fill(msg, info);
1428	if (rc)
1429		goto free_msg;
1430
1431	return genlmsg_reply(msg, info);
1432
1433free_msg:
1434	nlmsg_free(msg);
1435	return rc;
1436}
1437
1438static void net_dm_stats_read(struct net_dm_stats *stats)
1439{
1440	int cpu;
1441
1442	memset(stats, 0, sizeof(*stats));
1443	for_each_possible_cpu(cpu) {
1444		struct per_cpu_dm_data *data = &per_cpu(dm_cpu_data, cpu);
1445		struct net_dm_stats *cpu_stats = &data->stats;
1446		unsigned int start;
1447		u64 dropped;
1448
1449		do {
1450			start = u64_stats_fetch_begin(&cpu_stats->syncp);
1451			dropped = u64_stats_read(&cpu_stats->dropped);
1452		} while (u64_stats_fetch_retry(&cpu_stats->syncp, start));
1453
1454		u64_stats_add(&stats->dropped, dropped);
1455	}
1456}
1457
1458static int net_dm_stats_put(struct sk_buff *msg)
1459{
1460	struct net_dm_stats stats;
1461	struct nlattr *attr;
1462
1463	net_dm_stats_read(&stats);
1464
1465	attr = nla_nest_start(msg, NET_DM_ATTR_STATS);
1466	if (!attr)
1467		return -EMSGSIZE;
1468
1469	if (nla_put_u64_64bit(msg, NET_DM_ATTR_STATS_DROPPED,
1470			      u64_stats_read(&stats.dropped), NET_DM_ATTR_PAD))
1471		goto nla_put_failure;
1472
1473	nla_nest_end(msg, attr);
1474
1475	return 0;
1476
1477nla_put_failure:
1478	nla_nest_cancel(msg, attr);
1479	return -EMSGSIZE;
1480}
1481
1482static void net_dm_hw_stats_read(struct net_dm_stats *stats)
1483{
1484	int cpu;
1485
1486	memset(stats, 0, sizeof(*stats));
1487	for_each_possible_cpu(cpu) {
1488		struct per_cpu_dm_data *hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1489		struct net_dm_stats *cpu_stats = &hw_data->stats;
1490		unsigned int start;
1491		u64 dropped;
1492
1493		do {
1494			start = u64_stats_fetch_begin(&cpu_stats->syncp);
1495			dropped = u64_stats_read(&cpu_stats->dropped);
1496		} while (u64_stats_fetch_retry(&cpu_stats->syncp, start));
1497
1498		u64_stats_add(&stats->dropped, dropped);
1499	}
1500}
1501
1502static int net_dm_hw_stats_put(struct sk_buff *msg)
1503{
1504	struct net_dm_stats stats;
1505	struct nlattr *attr;
1506
1507	net_dm_hw_stats_read(&stats);
1508
1509	attr = nla_nest_start(msg, NET_DM_ATTR_HW_STATS);
1510	if (!attr)
1511		return -EMSGSIZE;
1512
1513	if (nla_put_u64_64bit(msg, NET_DM_ATTR_STATS_DROPPED,
1514			      u64_stats_read(&stats.dropped), NET_DM_ATTR_PAD))
1515		goto nla_put_failure;
1516
1517	nla_nest_end(msg, attr);
1518
1519	return 0;
1520
1521nla_put_failure:
1522	nla_nest_cancel(msg, attr);
1523	return -EMSGSIZE;
1524}
1525
1526static int net_dm_stats_fill(struct sk_buff *msg, struct genl_info *info)
1527{
1528	void *hdr;
1529	int rc;
1530
1531	hdr = genlmsg_put(msg, info->snd_portid, info->snd_seq,
1532			  &net_drop_monitor_family, 0, NET_DM_CMD_STATS_NEW);
1533	if (!hdr)
1534		return -EMSGSIZE;
1535
1536	rc = net_dm_stats_put(msg);
1537	if (rc)
1538		goto nla_put_failure;
1539
1540	rc = net_dm_hw_stats_put(msg);
1541	if (rc)
1542		goto nla_put_failure;
1543
1544	genlmsg_end(msg, hdr);
1545
1546	return 0;
1547
1548nla_put_failure:
1549	genlmsg_cancel(msg, hdr);
1550	return -EMSGSIZE;
1551}
1552
1553static int net_dm_cmd_stats_get(struct sk_buff *skb, struct genl_info *info)
1554{
1555	struct sk_buff *msg;
1556	int rc;
1557
1558	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1559	if (!msg)
1560		return -ENOMEM;
1561
1562	rc = net_dm_stats_fill(msg, info);
1563	if (rc)
1564		goto free_msg;
1565
1566	return genlmsg_reply(msg, info);
1567
1568free_msg:
1569	nlmsg_free(msg);
1570	return rc;
1571}
1572
1573static int dropmon_net_event(struct notifier_block *ev_block,
1574			     unsigned long event, void *ptr)
1575{
1576	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1577	struct dm_hw_stat_delta *stat;
 
1578
1579	switch (event) {
1580	case NETDEV_REGISTER:
1581		if (WARN_ON_ONCE(rtnl_dereference(dev->dm_private)))
1582			break;
1583		stat = kzalloc(sizeof(*stat), GFP_KERNEL);
1584		if (!stat)
1585			break;
1586
1587		stat->last_rx = jiffies;
1588		rcu_assign_pointer(dev->dm_private, stat);
1589
 
 
 
 
 
1590		break;
1591	case NETDEV_UNREGISTER:
1592		stat = rtnl_dereference(dev->dm_private);
1593		if (stat) {
1594			rcu_assign_pointer(dev->dm_private, NULL);
1595			kfree_rcu(stat, rcu);
 
 
 
 
 
 
1596		}
 
1597		break;
1598	}
 
1599	return NOTIFY_DONE;
1600}
1601
1602static const struct nla_policy net_dm_nl_policy[NET_DM_ATTR_MAX + 1] = {
1603	[NET_DM_ATTR_UNSPEC] = { .strict_start_type = NET_DM_ATTR_UNSPEC + 1 },
1604	[NET_DM_ATTR_ALERT_MODE] = { .type = NLA_U8 },
1605	[NET_DM_ATTR_TRUNC_LEN] = { .type = NLA_U32 },
1606	[NET_DM_ATTR_QUEUE_LEN] = { .type = NLA_U32 },
1607	[NET_DM_ATTR_SW_DROPS]	= {. type = NLA_FLAG },
1608	[NET_DM_ATTR_HW_DROPS]	= {. type = NLA_FLAG },
1609};
1610
1611static const struct genl_small_ops dropmon_ops[] = {
1612	{
1613		.cmd = NET_DM_CMD_CONFIG,
1614		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1615		.doit = net_dm_cmd_config,
1616		.flags = GENL_ADMIN_PERM,
1617	},
1618	{
1619		.cmd = NET_DM_CMD_START,
1620		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1621		.doit = net_dm_cmd_trace,
1622		.flags = GENL_ADMIN_PERM,
1623	},
1624	{
1625		.cmd = NET_DM_CMD_STOP,
1626		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1627		.doit = net_dm_cmd_trace,
1628		.flags = GENL_ADMIN_PERM,
1629	},
1630	{
1631		.cmd = NET_DM_CMD_CONFIG_GET,
1632		.doit = net_dm_cmd_config_get,
1633	},
1634	{
1635		.cmd = NET_DM_CMD_STATS_GET,
1636		.doit = net_dm_cmd_stats_get,
1637	},
1638};
1639
1640static int net_dm_nl_pre_doit(const struct genl_split_ops *ops,
1641			      struct sk_buff *skb, struct genl_info *info)
1642{
1643	mutex_lock(&net_dm_mutex);
1644
1645	return 0;
1646}
1647
1648static void net_dm_nl_post_doit(const struct genl_split_ops *ops,
1649				struct sk_buff *skb, struct genl_info *info)
1650{
1651	mutex_unlock(&net_dm_mutex);
1652}
1653
1654static struct genl_family net_drop_monitor_family __ro_after_init = {
1655	.hdrsize        = 0,
1656	.name           = "NET_DM",
1657	.version        = 2,
1658	.maxattr	= NET_DM_ATTR_MAX,
1659	.policy		= net_dm_nl_policy,
1660	.pre_doit	= net_dm_nl_pre_doit,
1661	.post_doit	= net_dm_nl_post_doit,
1662	.module		= THIS_MODULE,
1663	.small_ops	= dropmon_ops,
1664	.n_small_ops	= ARRAY_SIZE(dropmon_ops),
1665	.resv_start_op	= NET_DM_CMD_STATS_GET + 1,
1666	.mcgrps		= dropmon_mcgrps,
1667	.n_mcgrps	= ARRAY_SIZE(dropmon_mcgrps),
1668};
1669
1670static struct notifier_block dropmon_net_notifier = {
1671	.notifier_call = dropmon_net_event
1672};
1673
1674static void __net_dm_cpu_data_init(struct per_cpu_dm_data *data)
1675{
1676	spin_lock_init(&data->lock);
1677	skb_queue_head_init(&data->drop_queue);
1678	u64_stats_init(&data->stats.syncp);
1679}
1680
1681static void __net_dm_cpu_data_fini(struct per_cpu_dm_data *data)
1682{
1683	WARN_ON(!skb_queue_empty(&data->drop_queue));
1684}
1685
1686static void net_dm_cpu_data_init(int cpu)
1687{
1688	struct per_cpu_dm_data *data;
1689
1690	data = &per_cpu(dm_cpu_data, cpu);
1691	__net_dm_cpu_data_init(data);
1692}
1693
1694static void net_dm_cpu_data_fini(int cpu)
1695{
1696	struct per_cpu_dm_data *data;
1697
1698	data = &per_cpu(dm_cpu_data, cpu);
1699	/* At this point, we should have exclusive access
1700	 * to this struct and can free the skb inside it.
1701	 */
1702	consume_skb(data->skb);
1703	__net_dm_cpu_data_fini(data);
1704}
1705
1706static void net_dm_hw_cpu_data_init(int cpu)
1707{
1708	struct per_cpu_dm_data *hw_data;
1709
1710	hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1711	__net_dm_cpu_data_init(hw_data);
1712}
1713
1714static void net_dm_hw_cpu_data_fini(int cpu)
1715{
1716	struct per_cpu_dm_data *hw_data;
1717
1718	hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1719	kfree(hw_data->hw_entries);
1720	__net_dm_cpu_data_fini(hw_data);
1721}
1722
1723static int __init init_net_drop_monitor(void)
1724{
1725	int cpu, rc;
1726
1727	pr_info("Initializing network drop monitor service\n");
1728
1729	if (sizeof(void *) > 8) {
1730		pr_err("Unable to store program counters on this arch, Drop monitor failed\n");
1731		return -ENOSPC;
1732	}
1733
1734	rc = genl_register_family(&net_drop_monitor_family);
1735	if (rc) {
1736		pr_err("Could not create drop monitor netlink family\n");
1737		return rc;
1738	}
1739	WARN_ON(net_drop_monitor_family.mcgrp_offset != NET_DM_GRP_ALERT);
1740
1741	rc = register_netdevice_notifier(&dropmon_net_notifier);
1742	if (rc < 0) {
1743		pr_crit("Failed to register netdevice notifier\n");
1744		goto out_unreg;
1745	}
1746
1747	rc = 0;
1748
1749	for_each_possible_cpu(cpu) {
1750		net_dm_cpu_data_init(cpu);
1751		net_dm_hw_cpu_data_init(cpu);
1752	}
1753
1754	goto out;
1755
1756out_unreg:
1757	genl_unregister_family(&net_drop_monitor_family);
1758out:
1759	return rc;
1760}
1761
1762static void exit_net_drop_monitor(void)
1763{
1764	int cpu;
1765
1766	BUG_ON(unregister_netdevice_notifier(&dropmon_net_notifier));
1767
1768	/*
1769	 * Because of the module_get/put we do in the trace state change path
1770	 * we are guaranteed not to have any current users when we get here
1771	 */
1772
1773	for_each_possible_cpu(cpu) {
1774		net_dm_hw_cpu_data_fini(cpu);
1775		net_dm_cpu_data_fini(cpu);
1776	}
1777
1778	BUG_ON(genl_unregister_family(&net_drop_monitor_family));
1779}
1780
1781module_init(init_net_drop_monitor);
1782module_exit(exit_net_drop_monitor);
1783
1784MODULE_LICENSE("GPL v2");
1785MODULE_AUTHOR("Neil Horman <nhorman@tuxdriver.com>");
1786MODULE_ALIAS_GENL_FAMILY("NET_DM");
1787MODULE_DESCRIPTION("Monitoring code for network dropped packet alerts");
v5.9
   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/percpu.h>
  25#include <linux/timer.h>
  26#include <linux/bitops.h>
  27#include <linux/slab.h>
  28#include <linux/module.h>
  29#include <net/drop_monitor.h>
  30#include <net/genetlink.h>
  31#include <net/netevent.h>
  32#include <net/flow_offload.h>
 
 
  33
  34#include <trace/events/skb.h>
  35#include <trace/events/napi.h>
 
  36
  37#include <asm/unaligned.h>
  38
  39#define TRACE_ON 1
  40#define TRACE_OFF 0
  41
  42/*
  43 * Globals, our netlink socket pointer
  44 * and the work handle that will send up
  45 * netlink alerts
  46 */
  47static int trace_state = TRACE_OFF;
  48static bool monitor_hw;
  49
  50/* net_dm_mutex
  51 *
  52 * An overall lock guarding every operation coming from userspace.
  53 * It also guards the global 'hw_stats_list' list.
  54 */
  55static DEFINE_MUTEX(net_dm_mutex);
  56
  57struct net_dm_stats {
  58	u64 dropped;
  59	struct u64_stats_sync syncp;
  60};
  61
  62#define NET_DM_MAX_HW_TRAP_NAME_LEN 40
  63
  64struct net_dm_hw_entry {
  65	char trap_name[NET_DM_MAX_HW_TRAP_NAME_LEN];
  66	u32 count;
  67};
  68
  69struct net_dm_hw_entries {
  70	u32 num_entries;
  71	struct net_dm_hw_entry entries[];
  72};
  73
  74struct per_cpu_dm_data {
  75	spinlock_t		lock;	/* Protects 'skb', 'hw_entries' and
  76					 * 'send_timer'
  77					 */
  78	union {
  79		struct sk_buff			*skb;
  80		struct net_dm_hw_entries	*hw_entries;
  81	};
  82	struct sk_buff_head	drop_queue;
  83	struct work_struct	dm_alert_work;
  84	struct timer_list	send_timer;
  85	struct net_dm_stats	stats;
  86};
  87
  88struct dm_hw_stat_delta {
  89	struct net_device *dev;
  90	unsigned long last_rx;
  91	struct list_head list;
  92	struct rcu_head rcu;
  93	unsigned long last_drop_val;
  94};
  95
  96static struct genl_family net_drop_monitor_family;
  97
  98static DEFINE_PER_CPU(struct per_cpu_dm_data, dm_cpu_data);
  99static DEFINE_PER_CPU(struct per_cpu_dm_data, dm_hw_cpu_data);
 100
 101static int dm_hit_limit = 64;
 102static int dm_delay = 1;
 103static unsigned long dm_hw_check_delta = 2*HZ;
 104static LIST_HEAD(hw_stats_list);
 105
 106static enum net_dm_alert_mode net_dm_alert_mode = NET_DM_ALERT_MODE_SUMMARY;
 107static u32 net_dm_trunc_len;
 108static u32 net_dm_queue_len = 1000;
 109
 110struct net_dm_alert_ops {
 111	void (*kfree_skb_probe)(void *ignore, struct sk_buff *skb,
 112				void *location);
 
 113	void (*napi_poll_probe)(void *ignore, struct napi_struct *napi,
 114				int work, int budget);
 115	void (*work_item_func)(struct work_struct *work);
 116	void (*hw_work_item_func)(struct work_struct *work);
 117	void (*hw_probe)(struct sk_buff *skb,
 118			 const struct net_dm_hw_metadata *hw_metadata);
 
 119};
 120
 121struct net_dm_skb_cb {
 122	union {
 123		struct net_dm_hw_metadata *hw_metadata;
 124		void *pc;
 125	};
 
 126};
 127
 128#define NET_DM_SKB_CB(__skb) ((struct net_dm_skb_cb *)&((__skb)->cb[0]))
 129
 130static struct sk_buff *reset_per_cpu_data(struct per_cpu_dm_data *data)
 131{
 132	size_t al;
 133	struct net_dm_alert_msg *msg;
 134	struct nlattr *nla;
 135	struct sk_buff *skb;
 136	unsigned long flags;
 137	void *msg_header;
 138
 139	al = sizeof(struct net_dm_alert_msg);
 140	al += dm_hit_limit * sizeof(struct net_dm_drop_point);
 141	al += sizeof(struct nlattr);
 142
 143	skb = genlmsg_new(al, GFP_KERNEL);
 144
 145	if (!skb)
 146		goto err;
 147
 148	msg_header = genlmsg_put(skb, 0, 0, &net_drop_monitor_family,
 149				 0, NET_DM_CMD_ALERT);
 150	if (!msg_header) {
 151		nlmsg_free(skb);
 152		skb = NULL;
 153		goto err;
 154	}
 155	nla = nla_reserve(skb, NLA_UNSPEC,
 156			  sizeof(struct net_dm_alert_msg));
 157	if (!nla) {
 158		nlmsg_free(skb);
 159		skb = NULL;
 160		goto err;
 161	}
 162	msg = nla_data(nla);
 163	memset(msg, 0, al);
 164	goto out;
 165
 166err:
 167	mod_timer(&data->send_timer, jiffies + HZ / 10);
 168out:
 169	spin_lock_irqsave(&data->lock, flags);
 170	swap(data->skb, skb);
 171	spin_unlock_irqrestore(&data->lock, flags);
 172
 173	if (skb) {
 174		struct nlmsghdr *nlh = (struct nlmsghdr *)skb->data;
 175		struct genlmsghdr *gnlh = (struct genlmsghdr *)nlmsg_data(nlh);
 176
 177		genlmsg_end(skb, genlmsg_data(gnlh));
 178	}
 179
 180	return skb;
 181}
 182
 183static const struct genl_multicast_group dropmon_mcgrps[] = {
 184	{ .name = "events", },
 185};
 186
 187static void send_dm_alert(struct work_struct *work)
 188{
 189	struct sk_buff *skb;
 190	struct per_cpu_dm_data *data;
 191
 192	data = container_of(work, struct per_cpu_dm_data, dm_alert_work);
 193
 194	skb = reset_per_cpu_data(data);
 195
 196	if (skb)
 197		genlmsg_multicast(&net_drop_monitor_family, skb, 0,
 198				  0, GFP_KERNEL);
 199}
 200
 201/*
 202 * This is the timer function to delay the sending of an alert
 203 * in the event that more drops will arrive during the
 204 * hysteresis period.
 205 */
 206static void sched_send_work(struct timer_list *t)
 207{
 208	struct per_cpu_dm_data *data = from_timer(data, t, send_timer);
 209
 210	schedule_work(&data->dm_alert_work);
 211}
 212
 213static void trace_drop_common(struct sk_buff *skb, void *location)
 214{
 215	struct net_dm_alert_msg *msg;
 216	struct net_dm_drop_point *point;
 217	struct nlmsghdr *nlh;
 218	struct nlattr *nla;
 219	int i;
 220	struct sk_buff *dskb;
 221	struct per_cpu_dm_data *data;
 222	unsigned long flags;
 223
 224	local_irq_save(flags);
 225	data = this_cpu_ptr(&dm_cpu_data);
 226	spin_lock(&data->lock);
 227	dskb = data->skb;
 228
 229	if (!dskb)
 230		goto out;
 231
 232	nlh = (struct nlmsghdr *)dskb->data;
 233	nla = genlmsg_data(nlmsg_data(nlh));
 234	msg = nla_data(nla);
 235	point = msg->points;
 236	for (i = 0; i < msg->entries; i++) {
 237		if (!memcmp(&location, &point->pc, sizeof(void *))) {
 238			point->count++;
 239			goto out;
 240		}
 241		point++;
 242	}
 243	if (msg->entries == dm_hit_limit)
 244		goto out;
 245	/*
 246	 * We need to create a new entry
 247	 */
 248	__nla_reserve_nohdr(dskb, sizeof(struct net_dm_drop_point));
 249	nla->nla_len += NLA_ALIGN(sizeof(struct net_dm_drop_point));
 250	memcpy(point->pc, &location, sizeof(void *));
 251	point->count = 1;
 252	msg->entries++;
 253
 254	if (!timer_pending(&data->send_timer)) {
 255		data->send_timer.expires = jiffies + dm_delay * HZ;
 256		add_timer(&data->send_timer);
 257	}
 258
 259out:
 260	spin_unlock_irqrestore(&data->lock, flags);
 261}
 262
 263static void trace_kfree_skb_hit(void *ignore, struct sk_buff *skb, void *location)
 
 
 264{
 265	trace_drop_common(skb, location);
 266}
 267
 268static void trace_napi_poll_hit(void *ignore, struct napi_struct *napi,
 269				int work, int budget)
 270{
 271	struct dm_hw_stat_delta *new_stat;
 272
 273	/*
 274	 * Don't check napi structures with no associated device
 275	 */
 276	if (!napi->dev)
 277		return;
 278
 279	rcu_read_lock();
 280	list_for_each_entry_rcu(new_stat, &hw_stats_list, list) {
 
 281		/*
 282		 * only add a note to our monitor buffer if:
 283		 * 1) this is the dev we received on
 284		 * 2) its after the last_rx delta
 285		 * 3) our rx_dropped count has gone up
 286		 */
 287		if ((new_stat->dev == napi->dev)  &&
 288		    (time_after(jiffies, new_stat->last_rx + dm_hw_check_delta)) &&
 289		    (napi->dev->stats.rx_dropped != new_stat->last_drop_val)) {
 290			trace_drop_common(NULL, NULL);
 291			new_stat->last_drop_val = napi->dev->stats.rx_dropped;
 292			new_stat->last_rx = jiffies;
 293			break;
 294		}
 295	}
 296	rcu_read_unlock();
 297}
 298
 299static struct net_dm_hw_entries *
 300net_dm_hw_reset_per_cpu_data(struct per_cpu_dm_data *hw_data)
 301{
 302	struct net_dm_hw_entries *hw_entries;
 303	unsigned long flags;
 304
 305	hw_entries = kzalloc(struct_size(hw_entries, entries, dm_hit_limit),
 306			     GFP_KERNEL);
 307	if (!hw_entries) {
 308		/* If the memory allocation failed, we try to perform another
 309		 * allocation in 1/10 second. Otherwise, the probe function
 310		 * will constantly bail out.
 311		 */
 312		mod_timer(&hw_data->send_timer, jiffies + HZ / 10);
 313	}
 314
 315	spin_lock_irqsave(&hw_data->lock, flags);
 316	swap(hw_data->hw_entries, hw_entries);
 317	spin_unlock_irqrestore(&hw_data->lock, flags);
 318
 319	return hw_entries;
 320}
 321
 322static int net_dm_hw_entry_put(struct sk_buff *msg,
 323			       const struct net_dm_hw_entry *hw_entry)
 324{
 325	struct nlattr *attr;
 326
 327	attr = nla_nest_start(msg, NET_DM_ATTR_HW_ENTRY);
 328	if (!attr)
 329		return -EMSGSIZE;
 330
 331	if (nla_put_string(msg, NET_DM_ATTR_HW_TRAP_NAME, hw_entry->trap_name))
 332		goto nla_put_failure;
 333
 334	if (nla_put_u32(msg, NET_DM_ATTR_HW_TRAP_COUNT, hw_entry->count))
 335		goto nla_put_failure;
 336
 337	nla_nest_end(msg, attr);
 338
 339	return 0;
 340
 341nla_put_failure:
 342	nla_nest_cancel(msg, attr);
 343	return -EMSGSIZE;
 344}
 345
 346static int net_dm_hw_entries_put(struct sk_buff *msg,
 347				 const struct net_dm_hw_entries *hw_entries)
 348{
 349	struct nlattr *attr;
 350	int i;
 351
 352	attr = nla_nest_start(msg, NET_DM_ATTR_HW_ENTRIES);
 353	if (!attr)
 354		return -EMSGSIZE;
 355
 356	for (i = 0; i < hw_entries->num_entries; i++) {
 357		int rc;
 358
 359		rc = net_dm_hw_entry_put(msg, &hw_entries->entries[i]);
 360		if (rc)
 361			goto nla_put_failure;
 362	}
 363
 364	nla_nest_end(msg, attr);
 365
 366	return 0;
 367
 368nla_put_failure:
 369	nla_nest_cancel(msg, attr);
 370	return -EMSGSIZE;
 371}
 372
 373static int
 374net_dm_hw_summary_report_fill(struct sk_buff *msg,
 375			      const struct net_dm_hw_entries *hw_entries)
 376{
 377	struct net_dm_alert_msg anc_hdr = { 0 };
 378	void *hdr;
 379	int rc;
 380
 381	hdr = genlmsg_put(msg, 0, 0, &net_drop_monitor_family, 0,
 382			  NET_DM_CMD_ALERT);
 383	if (!hdr)
 384		return -EMSGSIZE;
 385
 386	/* We need to put the ancillary header in order not to break user
 387	 * space.
 388	 */
 389	if (nla_put(msg, NLA_UNSPEC, sizeof(anc_hdr), &anc_hdr))
 390		goto nla_put_failure;
 391
 392	rc = net_dm_hw_entries_put(msg, hw_entries);
 393	if (rc)
 394		goto nla_put_failure;
 395
 396	genlmsg_end(msg, hdr);
 397
 398	return 0;
 399
 400nla_put_failure:
 401	genlmsg_cancel(msg, hdr);
 402	return -EMSGSIZE;
 403}
 404
 405static void net_dm_hw_summary_work(struct work_struct *work)
 406{
 407	struct net_dm_hw_entries *hw_entries;
 408	struct per_cpu_dm_data *hw_data;
 409	struct sk_buff *msg;
 410	int rc;
 411
 412	hw_data = container_of(work, struct per_cpu_dm_data, dm_alert_work);
 413
 414	hw_entries = net_dm_hw_reset_per_cpu_data(hw_data);
 415	if (!hw_entries)
 416		return;
 417
 418	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
 419	if (!msg)
 420		goto out;
 421
 422	rc = net_dm_hw_summary_report_fill(msg, hw_entries);
 423	if (rc) {
 424		nlmsg_free(msg);
 425		goto out;
 426	}
 427
 428	genlmsg_multicast(&net_drop_monitor_family, msg, 0, 0, GFP_KERNEL);
 429
 430out:
 431	kfree(hw_entries);
 432}
 433
 434static void
 435net_dm_hw_summary_probe(struct sk_buff *skb,
 436			const struct net_dm_hw_metadata *hw_metadata)
 
 437{
 438	struct net_dm_hw_entries *hw_entries;
 439	struct net_dm_hw_entry *hw_entry;
 440	struct per_cpu_dm_data *hw_data;
 441	unsigned long flags;
 442	int i;
 443
 
 
 
 444	hw_data = this_cpu_ptr(&dm_hw_cpu_data);
 445	spin_lock_irqsave(&hw_data->lock, flags);
 446	hw_entries = hw_data->hw_entries;
 447
 448	if (!hw_entries)
 449		goto out;
 450
 451	for (i = 0; i < hw_entries->num_entries; i++) {
 452		hw_entry = &hw_entries->entries[i];
 453		if (!strncmp(hw_entry->trap_name, hw_metadata->trap_name,
 454			     NET_DM_MAX_HW_TRAP_NAME_LEN - 1)) {
 455			hw_entry->count++;
 456			goto out;
 457		}
 458	}
 459	if (WARN_ON_ONCE(hw_entries->num_entries == dm_hit_limit))
 460		goto out;
 461
 462	hw_entry = &hw_entries->entries[hw_entries->num_entries];
 463	strlcpy(hw_entry->trap_name, hw_metadata->trap_name,
 464		NET_DM_MAX_HW_TRAP_NAME_LEN - 1);
 465	hw_entry->count = 1;
 466	hw_entries->num_entries++;
 467
 468	if (!timer_pending(&hw_data->send_timer)) {
 469		hw_data->send_timer.expires = jiffies + dm_delay * HZ;
 470		add_timer(&hw_data->send_timer);
 471	}
 472
 473out:
 474	spin_unlock_irqrestore(&hw_data->lock, flags);
 475}
 476
 477static const struct net_dm_alert_ops net_dm_alert_summary_ops = {
 478	.kfree_skb_probe	= trace_kfree_skb_hit,
 479	.napi_poll_probe	= trace_napi_poll_hit,
 480	.work_item_func		= send_dm_alert,
 481	.hw_work_item_func	= net_dm_hw_summary_work,
 482	.hw_probe		= net_dm_hw_summary_probe,
 483};
 484
 485static void net_dm_packet_trace_kfree_skb_hit(void *ignore,
 486					      struct sk_buff *skb,
 487					      void *location)
 
 488{
 489	ktime_t tstamp = ktime_get_real();
 490	struct per_cpu_dm_data *data;
 
 491	struct sk_buff *nskb;
 492	unsigned long flags;
 493
 494	if (!skb_mac_header_was_set(skb))
 495		return;
 496
 497	nskb = skb_clone(skb, GFP_ATOMIC);
 498	if (!nskb)
 499		return;
 500
 501	NET_DM_SKB_CB(nskb)->pc = location;
 
 
 502	/* Override the timestamp because we care about the time when the
 503	 * packet was dropped.
 504	 */
 505	nskb->tstamp = tstamp;
 506
 507	data = this_cpu_ptr(&dm_cpu_data);
 508
 509	spin_lock_irqsave(&data->drop_queue.lock, flags);
 510	if (skb_queue_len(&data->drop_queue) < net_dm_queue_len)
 511		__skb_queue_tail(&data->drop_queue, nskb);
 512	else
 513		goto unlock_free;
 514	spin_unlock_irqrestore(&data->drop_queue.lock, flags);
 515
 516	schedule_work(&data->dm_alert_work);
 517
 518	return;
 519
 520unlock_free:
 521	spin_unlock_irqrestore(&data->drop_queue.lock, flags);
 522	u64_stats_update_begin(&data->stats.syncp);
 523	data->stats.dropped++;
 524	u64_stats_update_end(&data->stats.syncp);
 525	consume_skb(nskb);
 526}
 527
 528static void net_dm_packet_trace_napi_poll_hit(void *ignore,
 529					      struct napi_struct *napi,
 530					      int work, int budget)
 531{
 532}
 533
 534static size_t net_dm_in_port_size(void)
 535{
 536	       /* NET_DM_ATTR_IN_PORT nest */
 537	return nla_total_size(0) +
 538	       /* NET_DM_ATTR_PORT_NETDEV_IFINDEX */
 539	       nla_total_size(sizeof(u32)) +
 540	       /* NET_DM_ATTR_PORT_NETDEV_NAME */
 541	       nla_total_size(IFNAMSIZ + 1);
 542}
 543
 544#define NET_DM_MAX_SYMBOL_LEN 40
 
 545
 546static size_t net_dm_packet_report_size(size_t payload_len)
 547{
 548	size_t size;
 549
 550	size = nlmsg_msg_size(GENL_HDRLEN + net_drop_monitor_family.hdrsize);
 551
 552	return NLMSG_ALIGN(size) +
 553	       /* NET_DM_ATTR_ORIGIN */
 554	       nla_total_size(sizeof(u16)) +
 555	       /* NET_DM_ATTR_PC */
 556	       nla_total_size(sizeof(u64)) +
 557	       /* NET_DM_ATTR_SYMBOL */
 558	       nla_total_size(NET_DM_MAX_SYMBOL_LEN + 1) +
 559	       /* NET_DM_ATTR_IN_PORT */
 560	       net_dm_in_port_size() +
 561	       /* NET_DM_ATTR_TIMESTAMP */
 562	       nla_total_size(sizeof(u64)) +
 563	       /* NET_DM_ATTR_ORIG_LEN */
 564	       nla_total_size(sizeof(u32)) +
 565	       /* NET_DM_ATTR_PROTO */
 566	       nla_total_size(sizeof(u16)) +
 
 
 567	       /* NET_DM_ATTR_PAYLOAD */
 568	       nla_total_size(payload_len);
 569}
 570
 571static int net_dm_packet_report_in_port_put(struct sk_buff *msg, int ifindex,
 572					    const char *name)
 573{
 574	struct nlattr *attr;
 575
 576	attr = nla_nest_start(msg, NET_DM_ATTR_IN_PORT);
 577	if (!attr)
 578		return -EMSGSIZE;
 579
 580	if (ifindex &&
 581	    nla_put_u32(msg, NET_DM_ATTR_PORT_NETDEV_IFINDEX, ifindex))
 582		goto nla_put_failure;
 583
 584	if (name && nla_put_string(msg, NET_DM_ATTR_PORT_NETDEV_NAME, name))
 585		goto nla_put_failure;
 586
 587	nla_nest_end(msg, attr);
 588
 589	return 0;
 590
 591nla_put_failure:
 592	nla_nest_cancel(msg, attr);
 593	return -EMSGSIZE;
 594}
 595
 596static int net_dm_packet_report_fill(struct sk_buff *msg, struct sk_buff *skb,
 597				     size_t payload_len)
 598{
 599	u64 pc = (u64)(uintptr_t) NET_DM_SKB_CB(skb)->pc;
 
 
 600	char buf[NET_DM_MAX_SYMBOL_LEN];
 601	struct nlattr *attr;
 602	void *hdr;
 603	int rc;
 604
 605	hdr = genlmsg_put(msg, 0, 0, &net_drop_monitor_family, 0,
 606			  NET_DM_CMD_PACKET_ALERT);
 607	if (!hdr)
 608		return -EMSGSIZE;
 609
 610	if (nla_put_u16(msg, NET_DM_ATTR_ORIGIN, NET_DM_ORIGIN_SW))
 611		goto nla_put_failure;
 612
 613	if (nla_put_u64_64bit(msg, NET_DM_ATTR_PC, pc, NET_DM_ATTR_PAD))
 
 614		goto nla_put_failure;
 615
 616	snprintf(buf, sizeof(buf), "%pS", NET_DM_SKB_CB(skb)->pc);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 617	if (nla_put_string(msg, NET_DM_ATTR_SYMBOL, buf))
 618		goto nla_put_failure;
 619
 620	rc = net_dm_packet_report_in_port_put(msg, skb->skb_iif, NULL);
 621	if (rc)
 622		goto nla_put_failure;
 623
 624	if (nla_put_u64_64bit(msg, NET_DM_ATTR_TIMESTAMP,
 625			      ktime_to_ns(skb->tstamp), NET_DM_ATTR_PAD))
 626		goto nla_put_failure;
 627
 628	if (nla_put_u32(msg, NET_DM_ATTR_ORIG_LEN, skb->len))
 629		goto nla_put_failure;
 630
 631	if (!payload_len)
 632		goto out;
 633
 634	if (nla_put_u16(msg, NET_DM_ATTR_PROTO, be16_to_cpu(skb->protocol)))
 635		goto nla_put_failure;
 636
 637	attr = skb_put(msg, nla_total_size(payload_len));
 638	attr->nla_type = NET_DM_ATTR_PAYLOAD;
 639	attr->nla_len = nla_attr_size(payload_len);
 640	if (skb_copy_bits(skb, 0, nla_data(attr), payload_len))
 641		goto nla_put_failure;
 642
 643out:
 644	genlmsg_end(msg, hdr);
 645
 646	return 0;
 647
 648nla_put_failure:
 649	genlmsg_cancel(msg, hdr);
 650	return -EMSGSIZE;
 651}
 652
 653#define NET_DM_MAX_PACKET_SIZE (0xffff - NLA_HDRLEN - NLA_ALIGNTO)
 654
 655static void net_dm_packet_report(struct sk_buff *skb)
 656{
 657	struct sk_buff *msg;
 658	size_t payload_len;
 659	int rc;
 660
 661	/* Make sure we start copying the packet from the MAC header */
 662	if (skb->data > skb_mac_header(skb))
 663		skb_push(skb, skb->data - skb_mac_header(skb));
 664	else
 665		skb_pull(skb, skb_mac_header(skb) - skb->data);
 666
 667	/* Ensure packet fits inside a single netlink attribute */
 668	payload_len = min_t(size_t, skb->len, NET_DM_MAX_PACKET_SIZE);
 669	if (net_dm_trunc_len)
 670		payload_len = min_t(size_t, net_dm_trunc_len, payload_len);
 671
 672	msg = nlmsg_new(net_dm_packet_report_size(payload_len), GFP_KERNEL);
 673	if (!msg)
 674		goto out;
 675
 676	rc = net_dm_packet_report_fill(msg, skb, payload_len);
 677	if (rc) {
 678		nlmsg_free(msg);
 679		goto out;
 680	}
 681
 682	genlmsg_multicast(&net_drop_monitor_family, msg, 0, 0, GFP_KERNEL);
 683
 684out:
 685	consume_skb(skb);
 686}
 687
 688static void net_dm_packet_work(struct work_struct *work)
 689{
 690	struct per_cpu_dm_data *data;
 691	struct sk_buff_head list;
 692	struct sk_buff *skb;
 693	unsigned long flags;
 694
 695	data = container_of(work, struct per_cpu_dm_data, dm_alert_work);
 696
 697	__skb_queue_head_init(&list);
 698
 699	spin_lock_irqsave(&data->drop_queue.lock, flags);
 700	skb_queue_splice_tail_init(&data->drop_queue, &list);
 701	spin_unlock_irqrestore(&data->drop_queue.lock, flags);
 702
 703	while ((skb = __skb_dequeue(&list)))
 704		net_dm_packet_report(skb);
 705}
 706
 707static size_t
 708net_dm_flow_action_cookie_size(const struct net_dm_hw_metadata *hw_metadata)
 709{
 710	return hw_metadata->fa_cookie ?
 711	       nla_total_size(hw_metadata->fa_cookie->cookie_len) : 0;
 712}
 713
 714static size_t
 715net_dm_hw_packet_report_size(size_t payload_len,
 716			     const struct net_dm_hw_metadata *hw_metadata)
 717{
 718	size_t size;
 719
 720	size = nlmsg_msg_size(GENL_HDRLEN + net_drop_monitor_family.hdrsize);
 721
 722	return NLMSG_ALIGN(size) +
 723	       /* NET_DM_ATTR_ORIGIN */
 724	       nla_total_size(sizeof(u16)) +
 725	       /* NET_DM_ATTR_HW_TRAP_GROUP_NAME */
 726	       nla_total_size(strlen(hw_metadata->trap_group_name) + 1) +
 727	       /* NET_DM_ATTR_HW_TRAP_NAME */
 728	       nla_total_size(strlen(hw_metadata->trap_name) + 1) +
 729	       /* NET_DM_ATTR_IN_PORT */
 730	       net_dm_in_port_size() +
 731	       /* NET_DM_ATTR_FLOW_ACTION_COOKIE */
 732	       net_dm_flow_action_cookie_size(hw_metadata) +
 733	       /* NET_DM_ATTR_TIMESTAMP */
 734	       nla_total_size(sizeof(u64)) +
 735	       /* NET_DM_ATTR_ORIG_LEN */
 736	       nla_total_size(sizeof(u32)) +
 737	       /* NET_DM_ATTR_PROTO */
 738	       nla_total_size(sizeof(u16)) +
 739	       /* NET_DM_ATTR_PAYLOAD */
 740	       nla_total_size(payload_len);
 741}
 742
 743static int net_dm_hw_packet_report_fill(struct sk_buff *msg,
 744					struct sk_buff *skb, size_t payload_len)
 745{
 746	struct net_dm_hw_metadata *hw_metadata;
 747	struct nlattr *attr;
 748	void *hdr;
 749
 750	hw_metadata = NET_DM_SKB_CB(skb)->hw_metadata;
 751
 752	hdr = genlmsg_put(msg, 0, 0, &net_drop_monitor_family, 0,
 753			  NET_DM_CMD_PACKET_ALERT);
 754	if (!hdr)
 755		return -EMSGSIZE;
 756
 757	if (nla_put_u16(msg, NET_DM_ATTR_ORIGIN, NET_DM_ORIGIN_HW))
 758		goto nla_put_failure;
 759
 760	if (nla_put_string(msg, NET_DM_ATTR_HW_TRAP_GROUP_NAME,
 761			   hw_metadata->trap_group_name))
 762		goto nla_put_failure;
 763
 764	if (nla_put_string(msg, NET_DM_ATTR_HW_TRAP_NAME,
 765			   hw_metadata->trap_name))
 766		goto nla_put_failure;
 767
 768	if (hw_metadata->input_dev) {
 769		struct net_device *dev = hw_metadata->input_dev;
 770		int rc;
 771
 772		rc = net_dm_packet_report_in_port_put(msg, dev->ifindex,
 773						      dev->name);
 774		if (rc)
 775			goto nla_put_failure;
 776	}
 777
 778	if (hw_metadata->fa_cookie &&
 779	    nla_put(msg, NET_DM_ATTR_FLOW_ACTION_COOKIE,
 780		    hw_metadata->fa_cookie->cookie_len,
 781		    hw_metadata->fa_cookie->cookie))
 782		goto nla_put_failure;
 783
 784	if (nla_put_u64_64bit(msg, NET_DM_ATTR_TIMESTAMP,
 785			      ktime_to_ns(skb->tstamp), NET_DM_ATTR_PAD))
 786		goto nla_put_failure;
 787
 788	if (nla_put_u32(msg, NET_DM_ATTR_ORIG_LEN, skb->len))
 789		goto nla_put_failure;
 790
 791	if (!payload_len)
 792		goto out;
 793
 794	if (nla_put_u16(msg, NET_DM_ATTR_PROTO, be16_to_cpu(skb->protocol)))
 795		goto nla_put_failure;
 796
 797	attr = skb_put(msg, nla_total_size(payload_len));
 798	attr->nla_type = NET_DM_ATTR_PAYLOAD;
 799	attr->nla_len = nla_attr_size(payload_len);
 800	if (skb_copy_bits(skb, 0, nla_data(attr), payload_len))
 801		goto nla_put_failure;
 802
 803out:
 804	genlmsg_end(msg, hdr);
 805
 806	return 0;
 807
 808nla_put_failure:
 809	genlmsg_cancel(msg, hdr);
 810	return -EMSGSIZE;
 811}
 812
 813static struct net_dm_hw_metadata *
 814net_dm_hw_metadata_clone(const struct net_dm_hw_metadata *hw_metadata)
 815{
 816	const struct flow_action_cookie *fa_cookie;
 817	struct net_dm_hw_metadata *n_hw_metadata;
 818	const char *trap_group_name;
 819	const char *trap_name;
 820
 821	n_hw_metadata = kzalloc(sizeof(*hw_metadata), GFP_ATOMIC);
 822	if (!n_hw_metadata)
 823		return NULL;
 824
 825	trap_group_name = kstrdup(hw_metadata->trap_group_name, GFP_ATOMIC);
 826	if (!trap_group_name)
 827		goto free_hw_metadata;
 828	n_hw_metadata->trap_group_name = trap_group_name;
 829
 830	trap_name = kstrdup(hw_metadata->trap_name, GFP_ATOMIC);
 831	if (!trap_name)
 832		goto free_trap_group;
 833	n_hw_metadata->trap_name = trap_name;
 834
 835	if (hw_metadata->fa_cookie) {
 836		size_t cookie_size = sizeof(*fa_cookie) +
 837				     hw_metadata->fa_cookie->cookie_len;
 838
 839		fa_cookie = kmemdup(hw_metadata->fa_cookie, cookie_size,
 840				    GFP_ATOMIC);
 841		if (!fa_cookie)
 842			goto free_trap_name;
 843		n_hw_metadata->fa_cookie = fa_cookie;
 844	}
 845
 846	n_hw_metadata->input_dev = hw_metadata->input_dev;
 847	if (n_hw_metadata->input_dev)
 848		dev_hold(n_hw_metadata->input_dev);
 849
 850	return n_hw_metadata;
 851
 852free_trap_name:
 853	kfree(trap_name);
 854free_trap_group:
 855	kfree(trap_group_name);
 856free_hw_metadata:
 857	kfree(n_hw_metadata);
 858	return NULL;
 859}
 860
 861static void
 862net_dm_hw_metadata_free(const struct net_dm_hw_metadata *hw_metadata)
 863{
 864	if (hw_metadata->input_dev)
 865		dev_put(hw_metadata->input_dev);
 866	kfree(hw_metadata->fa_cookie);
 867	kfree(hw_metadata->trap_name);
 868	kfree(hw_metadata->trap_group_name);
 869	kfree(hw_metadata);
 870}
 871
 872static void net_dm_hw_packet_report(struct sk_buff *skb)
 873{
 874	struct net_dm_hw_metadata *hw_metadata;
 875	struct sk_buff *msg;
 876	size_t payload_len;
 877	int rc;
 878
 879	if (skb->data > skb_mac_header(skb))
 880		skb_push(skb, skb->data - skb_mac_header(skb));
 881	else
 882		skb_pull(skb, skb_mac_header(skb) - skb->data);
 883
 884	payload_len = min_t(size_t, skb->len, NET_DM_MAX_PACKET_SIZE);
 885	if (net_dm_trunc_len)
 886		payload_len = min_t(size_t, net_dm_trunc_len, payload_len);
 887
 888	hw_metadata = NET_DM_SKB_CB(skb)->hw_metadata;
 889	msg = nlmsg_new(net_dm_hw_packet_report_size(payload_len, hw_metadata),
 890			GFP_KERNEL);
 891	if (!msg)
 892		goto out;
 893
 894	rc = net_dm_hw_packet_report_fill(msg, skb, payload_len);
 895	if (rc) {
 896		nlmsg_free(msg);
 897		goto out;
 898	}
 899
 900	genlmsg_multicast(&net_drop_monitor_family, msg, 0, 0, GFP_KERNEL);
 901
 902out:
 903	net_dm_hw_metadata_free(NET_DM_SKB_CB(skb)->hw_metadata);
 904	consume_skb(skb);
 905}
 906
 907static void net_dm_hw_packet_work(struct work_struct *work)
 908{
 909	struct per_cpu_dm_data *hw_data;
 910	struct sk_buff_head list;
 911	struct sk_buff *skb;
 912	unsigned long flags;
 913
 914	hw_data = container_of(work, struct per_cpu_dm_data, dm_alert_work);
 915
 916	__skb_queue_head_init(&list);
 917
 918	spin_lock_irqsave(&hw_data->drop_queue.lock, flags);
 919	skb_queue_splice_tail_init(&hw_data->drop_queue, &list);
 920	spin_unlock_irqrestore(&hw_data->drop_queue.lock, flags);
 921
 922	while ((skb = __skb_dequeue(&list)))
 923		net_dm_hw_packet_report(skb);
 924}
 925
 926static void
 927net_dm_hw_packet_probe(struct sk_buff *skb,
 928		       const struct net_dm_hw_metadata *hw_metadata)
 
 929{
 930	struct net_dm_hw_metadata *n_hw_metadata;
 931	ktime_t tstamp = ktime_get_real();
 932	struct per_cpu_dm_data *hw_data;
 933	struct sk_buff *nskb;
 934	unsigned long flags;
 935
 
 
 
 936	if (!skb_mac_header_was_set(skb))
 937		return;
 938
 939	nskb = skb_clone(skb, GFP_ATOMIC);
 940	if (!nskb)
 941		return;
 942
 943	n_hw_metadata = net_dm_hw_metadata_clone(hw_metadata);
 944	if (!n_hw_metadata)
 945		goto free;
 946
 947	NET_DM_SKB_CB(nskb)->hw_metadata = n_hw_metadata;
 948	nskb->tstamp = tstamp;
 949
 950	hw_data = this_cpu_ptr(&dm_hw_cpu_data);
 951
 952	spin_lock_irqsave(&hw_data->drop_queue.lock, flags);
 953	if (skb_queue_len(&hw_data->drop_queue) < net_dm_queue_len)
 954		__skb_queue_tail(&hw_data->drop_queue, nskb);
 955	else
 956		goto unlock_free;
 957	spin_unlock_irqrestore(&hw_data->drop_queue.lock, flags);
 958
 959	schedule_work(&hw_data->dm_alert_work);
 960
 961	return;
 962
 963unlock_free:
 964	spin_unlock_irqrestore(&hw_data->drop_queue.lock, flags);
 965	u64_stats_update_begin(&hw_data->stats.syncp);
 966	hw_data->stats.dropped++;
 967	u64_stats_update_end(&hw_data->stats.syncp);
 968	net_dm_hw_metadata_free(n_hw_metadata);
 969free:
 970	consume_skb(nskb);
 971}
 972
 973static const struct net_dm_alert_ops net_dm_alert_packet_ops = {
 974	.kfree_skb_probe	= net_dm_packet_trace_kfree_skb_hit,
 975	.napi_poll_probe	= net_dm_packet_trace_napi_poll_hit,
 976	.work_item_func		= net_dm_packet_work,
 977	.hw_work_item_func	= net_dm_hw_packet_work,
 978	.hw_probe		= net_dm_hw_packet_probe,
 979};
 980
 981static const struct net_dm_alert_ops *net_dm_alert_ops_arr[] = {
 982	[NET_DM_ALERT_MODE_SUMMARY]	= &net_dm_alert_summary_ops,
 983	[NET_DM_ALERT_MODE_PACKET]	= &net_dm_alert_packet_ops,
 984};
 985
 986void net_dm_hw_report(struct sk_buff *skb,
 987		      const struct net_dm_hw_metadata *hw_metadata)
 988{
 989	rcu_read_lock();
 
 990
 991	if (!monitor_hw)
 992		goto out;
 
 
 
 
 
 
 
 
 993
 994	net_dm_alert_ops_arr[net_dm_alert_mode]->hw_probe(skb, hw_metadata);
 995
 996out:
 997	rcu_read_unlock();
 998}
 999EXPORT_SYMBOL_GPL(net_dm_hw_report);
1000
1001static int net_dm_hw_monitor_start(struct netlink_ext_ack *extack)
1002{
1003	const struct net_dm_alert_ops *ops;
1004	int cpu;
1005
1006	if (monitor_hw) {
1007		NL_SET_ERR_MSG_MOD(extack, "Hardware monitoring already enabled");
1008		return -EAGAIN;
1009	}
1010
1011	ops = net_dm_alert_ops_arr[net_dm_alert_mode];
1012
1013	if (!try_module_get(THIS_MODULE)) {
1014		NL_SET_ERR_MSG_MOD(extack, "Failed to take reference on module");
1015		return -ENODEV;
1016	}
1017
1018	for_each_possible_cpu(cpu) {
1019		struct per_cpu_dm_data *hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1020		struct net_dm_hw_entries *hw_entries;
1021
1022		INIT_WORK(&hw_data->dm_alert_work, ops->hw_work_item_func);
1023		timer_setup(&hw_data->send_timer, sched_send_work, 0);
1024		hw_entries = net_dm_hw_reset_per_cpu_data(hw_data);
1025		kfree(hw_entries);
1026	}
1027
 
 
 
 
 
 
1028	monitor_hw = true;
1029
1030	return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1031}
1032
1033static void net_dm_hw_monitor_stop(struct netlink_ext_ack *extack)
1034{
 
1035	int cpu;
1036
1037	if (!monitor_hw) {
1038		NL_SET_ERR_MSG_MOD(extack, "Hardware monitoring already disabled");
1039		return;
1040	}
1041
 
 
1042	monitor_hw = false;
1043
1044	/* After this call returns we are guaranteed that no CPU is processing
1045	 * any hardware drops.
1046	 */
1047	synchronize_rcu();
1048
1049	for_each_possible_cpu(cpu) {
1050		struct per_cpu_dm_data *hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1051		struct sk_buff *skb;
1052
1053		del_timer_sync(&hw_data->send_timer);
1054		cancel_work_sync(&hw_data->dm_alert_work);
1055		while ((skb = __skb_dequeue(&hw_data->drop_queue))) {
1056			struct net_dm_hw_metadata *hw_metadata;
1057
1058			hw_metadata = NET_DM_SKB_CB(skb)->hw_metadata;
1059			net_dm_hw_metadata_free(hw_metadata);
1060			consume_skb(skb);
1061		}
1062	}
1063
1064	module_put(THIS_MODULE);
1065}
1066
1067static int net_dm_trace_on_set(struct netlink_ext_ack *extack)
1068{
1069	const struct net_dm_alert_ops *ops;
1070	int cpu, rc;
1071
1072	ops = net_dm_alert_ops_arr[net_dm_alert_mode];
1073
1074	if (!try_module_get(THIS_MODULE)) {
1075		NL_SET_ERR_MSG_MOD(extack, "Failed to take reference on module");
1076		return -ENODEV;
1077	}
1078
1079	for_each_possible_cpu(cpu) {
1080		struct per_cpu_dm_data *data = &per_cpu(dm_cpu_data, cpu);
1081		struct sk_buff *skb;
1082
1083		INIT_WORK(&data->dm_alert_work, ops->work_item_func);
1084		timer_setup(&data->send_timer, sched_send_work, 0);
1085		/* Allocate a new per-CPU skb for the summary alert message and
1086		 * free the old one which might contain stale data from
1087		 * previous tracing.
1088		 */
1089		skb = reset_per_cpu_data(data);
1090		consume_skb(skb);
1091	}
1092
1093	rc = register_trace_kfree_skb(ops->kfree_skb_probe, NULL);
1094	if (rc) {
1095		NL_SET_ERR_MSG_MOD(extack, "Failed to connect probe to kfree_skb() tracepoint");
1096		goto err_module_put;
1097	}
1098
1099	rc = register_trace_napi_poll(ops->napi_poll_probe, NULL);
1100	if (rc) {
1101		NL_SET_ERR_MSG_MOD(extack, "Failed to connect probe to napi_poll() tracepoint");
1102		goto err_unregister_trace;
1103	}
1104
1105	return 0;
1106
1107err_unregister_trace:
1108	unregister_trace_kfree_skb(ops->kfree_skb_probe, NULL);
1109err_module_put:
 
 
 
 
 
 
 
 
 
1110	module_put(THIS_MODULE);
1111	return rc;
1112}
1113
1114static void net_dm_trace_off_set(void)
1115{
1116	struct dm_hw_stat_delta *new_stat, *temp;
1117	const struct net_dm_alert_ops *ops;
1118	int cpu;
1119
1120	ops = net_dm_alert_ops_arr[net_dm_alert_mode];
1121
1122	unregister_trace_napi_poll(ops->napi_poll_probe, NULL);
1123	unregister_trace_kfree_skb(ops->kfree_skb_probe, NULL);
1124
1125	tracepoint_synchronize_unregister();
1126
1127	/* Make sure we do not send notifications to user space after request
1128	 * to stop tracing returns.
1129	 */
1130	for_each_possible_cpu(cpu) {
1131		struct per_cpu_dm_data *data = &per_cpu(dm_cpu_data, cpu);
1132		struct sk_buff *skb;
1133
1134		del_timer_sync(&data->send_timer);
1135		cancel_work_sync(&data->dm_alert_work);
1136		while ((skb = __skb_dequeue(&data->drop_queue)))
1137			consume_skb(skb);
1138	}
1139
1140	list_for_each_entry_safe(new_stat, temp, &hw_stats_list, list) {
1141		if (new_stat->dev == NULL) {
1142			list_del_rcu(&new_stat->list);
1143			kfree_rcu(new_stat, rcu);
1144		}
1145	}
1146
1147	module_put(THIS_MODULE);
1148}
1149
1150static int set_all_monitor_traces(int state, struct netlink_ext_ack *extack)
1151{
1152	int rc = 0;
1153
1154	if (state == trace_state) {
1155		NL_SET_ERR_MSG_MOD(extack, "Trace state already set to requested state");
1156		return -EAGAIN;
1157	}
1158
1159	switch (state) {
1160	case TRACE_ON:
1161		rc = net_dm_trace_on_set(extack);
1162		break;
1163	case TRACE_OFF:
1164		net_dm_trace_off_set();
1165		break;
1166	default:
1167		rc = 1;
1168		break;
1169	}
1170
1171	if (!rc)
1172		trace_state = state;
1173	else
1174		rc = -EINPROGRESS;
1175
1176	return rc;
1177}
1178
1179static bool net_dm_is_monitoring(void)
1180{
1181	return trace_state == TRACE_ON || monitor_hw;
1182}
1183
1184static int net_dm_alert_mode_get_from_info(struct genl_info *info,
1185					   enum net_dm_alert_mode *p_alert_mode)
1186{
1187	u8 val;
1188
1189	val = nla_get_u8(info->attrs[NET_DM_ATTR_ALERT_MODE]);
1190
1191	switch (val) {
1192	case NET_DM_ALERT_MODE_SUMMARY:
1193	case NET_DM_ALERT_MODE_PACKET:
1194		*p_alert_mode = val;
1195		break;
1196	default:
1197		return -EINVAL;
1198	}
1199
1200	return 0;
1201}
1202
1203static int net_dm_alert_mode_set(struct genl_info *info)
1204{
1205	struct netlink_ext_ack *extack = info->extack;
1206	enum net_dm_alert_mode alert_mode;
1207	int rc;
1208
1209	if (!info->attrs[NET_DM_ATTR_ALERT_MODE])
1210		return 0;
1211
1212	rc = net_dm_alert_mode_get_from_info(info, &alert_mode);
1213	if (rc) {
1214		NL_SET_ERR_MSG_MOD(extack, "Invalid alert mode");
1215		return -EINVAL;
1216	}
1217
1218	net_dm_alert_mode = alert_mode;
1219
1220	return 0;
1221}
1222
1223static void net_dm_trunc_len_set(struct genl_info *info)
1224{
1225	if (!info->attrs[NET_DM_ATTR_TRUNC_LEN])
1226		return;
1227
1228	net_dm_trunc_len = nla_get_u32(info->attrs[NET_DM_ATTR_TRUNC_LEN]);
1229}
1230
1231static void net_dm_queue_len_set(struct genl_info *info)
1232{
1233	if (!info->attrs[NET_DM_ATTR_QUEUE_LEN])
1234		return;
1235
1236	net_dm_queue_len = nla_get_u32(info->attrs[NET_DM_ATTR_QUEUE_LEN]);
1237}
1238
1239static int net_dm_cmd_config(struct sk_buff *skb,
1240			struct genl_info *info)
1241{
1242	struct netlink_ext_ack *extack = info->extack;
1243	int rc;
1244
1245	if (net_dm_is_monitoring()) {
1246		NL_SET_ERR_MSG_MOD(extack, "Cannot configure drop monitor during monitoring");
1247		return -EBUSY;
1248	}
1249
1250	rc = net_dm_alert_mode_set(info);
1251	if (rc)
1252		return rc;
1253
1254	net_dm_trunc_len_set(info);
1255
1256	net_dm_queue_len_set(info);
1257
1258	return 0;
1259}
1260
1261static int net_dm_monitor_start(bool set_sw, bool set_hw,
1262				struct netlink_ext_ack *extack)
1263{
1264	bool sw_set = false;
1265	int rc;
1266
1267	if (set_sw) {
1268		rc = set_all_monitor_traces(TRACE_ON, extack);
1269		if (rc)
1270			return rc;
1271		sw_set = true;
1272	}
1273
1274	if (set_hw) {
1275		rc = net_dm_hw_monitor_start(extack);
1276		if (rc)
1277			goto err_monitor_hw;
1278	}
1279
1280	return 0;
1281
1282err_monitor_hw:
1283	if (sw_set)
1284		set_all_monitor_traces(TRACE_OFF, extack);
1285	return rc;
1286}
1287
1288static void net_dm_monitor_stop(bool set_sw, bool set_hw,
1289				struct netlink_ext_ack *extack)
1290{
1291	if (set_hw)
1292		net_dm_hw_monitor_stop(extack);
1293	if (set_sw)
1294		set_all_monitor_traces(TRACE_OFF, extack);
1295}
1296
1297static int net_dm_cmd_trace(struct sk_buff *skb,
1298			struct genl_info *info)
1299{
1300	bool set_sw = !!info->attrs[NET_DM_ATTR_SW_DROPS];
1301	bool set_hw = !!info->attrs[NET_DM_ATTR_HW_DROPS];
1302	struct netlink_ext_ack *extack = info->extack;
1303
1304	/* To maintain backward compatibility, we start / stop monitoring of
1305	 * software drops if no flag is specified.
1306	 */
1307	if (!set_sw && !set_hw)
1308		set_sw = true;
1309
1310	switch (info->genlhdr->cmd) {
1311	case NET_DM_CMD_START:
1312		return net_dm_monitor_start(set_sw, set_hw, extack);
1313	case NET_DM_CMD_STOP:
1314		net_dm_monitor_stop(set_sw, set_hw, extack);
1315		return 0;
1316	}
1317
1318	return -EOPNOTSUPP;
1319}
1320
1321static int net_dm_config_fill(struct sk_buff *msg, struct genl_info *info)
1322{
1323	void *hdr;
1324
1325	hdr = genlmsg_put(msg, info->snd_portid, info->snd_seq,
1326			  &net_drop_monitor_family, 0, NET_DM_CMD_CONFIG_NEW);
1327	if (!hdr)
1328		return -EMSGSIZE;
1329
1330	if (nla_put_u8(msg, NET_DM_ATTR_ALERT_MODE, net_dm_alert_mode))
1331		goto nla_put_failure;
1332
1333	if (nla_put_u32(msg, NET_DM_ATTR_TRUNC_LEN, net_dm_trunc_len))
1334		goto nla_put_failure;
1335
1336	if (nla_put_u32(msg, NET_DM_ATTR_QUEUE_LEN, net_dm_queue_len))
1337		goto nla_put_failure;
1338
1339	genlmsg_end(msg, hdr);
1340
1341	return 0;
1342
1343nla_put_failure:
1344	genlmsg_cancel(msg, hdr);
1345	return -EMSGSIZE;
1346}
1347
1348static int net_dm_cmd_config_get(struct sk_buff *skb, struct genl_info *info)
1349{
1350	struct sk_buff *msg;
1351	int rc;
1352
1353	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1354	if (!msg)
1355		return -ENOMEM;
1356
1357	rc = net_dm_config_fill(msg, info);
1358	if (rc)
1359		goto free_msg;
1360
1361	return genlmsg_reply(msg, info);
1362
1363free_msg:
1364	nlmsg_free(msg);
1365	return rc;
1366}
1367
1368static void net_dm_stats_read(struct net_dm_stats *stats)
1369{
1370	int cpu;
1371
1372	memset(stats, 0, sizeof(*stats));
1373	for_each_possible_cpu(cpu) {
1374		struct per_cpu_dm_data *data = &per_cpu(dm_cpu_data, cpu);
1375		struct net_dm_stats *cpu_stats = &data->stats;
1376		unsigned int start;
1377		u64 dropped;
1378
1379		do {
1380			start = u64_stats_fetch_begin_irq(&cpu_stats->syncp);
1381			dropped = cpu_stats->dropped;
1382		} while (u64_stats_fetch_retry_irq(&cpu_stats->syncp, start));
1383
1384		stats->dropped += dropped;
1385	}
1386}
1387
1388static int net_dm_stats_put(struct sk_buff *msg)
1389{
1390	struct net_dm_stats stats;
1391	struct nlattr *attr;
1392
1393	net_dm_stats_read(&stats);
1394
1395	attr = nla_nest_start(msg, NET_DM_ATTR_STATS);
1396	if (!attr)
1397		return -EMSGSIZE;
1398
1399	if (nla_put_u64_64bit(msg, NET_DM_ATTR_STATS_DROPPED,
1400			      stats.dropped, NET_DM_ATTR_PAD))
1401		goto nla_put_failure;
1402
1403	nla_nest_end(msg, attr);
1404
1405	return 0;
1406
1407nla_put_failure:
1408	nla_nest_cancel(msg, attr);
1409	return -EMSGSIZE;
1410}
1411
1412static void net_dm_hw_stats_read(struct net_dm_stats *stats)
1413{
1414	int cpu;
1415
1416	memset(stats, 0, sizeof(*stats));
1417	for_each_possible_cpu(cpu) {
1418		struct per_cpu_dm_data *hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1419		struct net_dm_stats *cpu_stats = &hw_data->stats;
1420		unsigned int start;
1421		u64 dropped;
1422
1423		do {
1424			start = u64_stats_fetch_begin_irq(&cpu_stats->syncp);
1425			dropped = cpu_stats->dropped;
1426		} while (u64_stats_fetch_retry_irq(&cpu_stats->syncp, start));
1427
1428		stats->dropped += dropped;
1429	}
1430}
1431
1432static int net_dm_hw_stats_put(struct sk_buff *msg)
1433{
1434	struct net_dm_stats stats;
1435	struct nlattr *attr;
1436
1437	net_dm_hw_stats_read(&stats);
1438
1439	attr = nla_nest_start(msg, NET_DM_ATTR_HW_STATS);
1440	if (!attr)
1441		return -EMSGSIZE;
1442
1443	if (nla_put_u64_64bit(msg, NET_DM_ATTR_STATS_DROPPED,
1444			      stats.dropped, NET_DM_ATTR_PAD))
1445		goto nla_put_failure;
1446
1447	nla_nest_end(msg, attr);
1448
1449	return 0;
1450
1451nla_put_failure:
1452	nla_nest_cancel(msg, attr);
1453	return -EMSGSIZE;
1454}
1455
1456static int net_dm_stats_fill(struct sk_buff *msg, struct genl_info *info)
1457{
1458	void *hdr;
1459	int rc;
1460
1461	hdr = genlmsg_put(msg, info->snd_portid, info->snd_seq,
1462			  &net_drop_monitor_family, 0, NET_DM_CMD_STATS_NEW);
1463	if (!hdr)
1464		return -EMSGSIZE;
1465
1466	rc = net_dm_stats_put(msg);
1467	if (rc)
1468		goto nla_put_failure;
1469
1470	rc = net_dm_hw_stats_put(msg);
1471	if (rc)
1472		goto nla_put_failure;
1473
1474	genlmsg_end(msg, hdr);
1475
1476	return 0;
1477
1478nla_put_failure:
1479	genlmsg_cancel(msg, hdr);
1480	return -EMSGSIZE;
1481}
1482
1483static int net_dm_cmd_stats_get(struct sk_buff *skb, struct genl_info *info)
1484{
1485	struct sk_buff *msg;
1486	int rc;
1487
1488	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1489	if (!msg)
1490		return -ENOMEM;
1491
1492	rc = net_dm_stats_fill(msg, info);
1493	if (rc)
1494		goto free_msg;
1495
1496	return genlmsg_reply(msg, info);
1497
1498free_msg:
1499	nlmsg_free(msg);
1500	return rc;
1501}
1502
1503static int dropmon_net_event(struct notifier_block *ev_block,
1504			     unsigned long event, void *ptr)
1505{
1506	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1507	struct dm_hw_stat_delta *new_stat = NULL;
1508	struct dm_hw_stat_delta *tmp;
1509
1510	switch (event) {
1511	case NETDEV_REGISTER:
1512		new_stat = kzalloc(sizeof(struct dm_hw_stat_delta), GFP_KERNEL);
 
 
 
 
1513
1514		if (!new_stat)
1515			goto out;
1516
1517		new_stat->dev = dev;
1518		new_stat->last_rx = jiffies;
1519		mutex_lock(&net_dm_mutex);
1520		list_add_rcu(&new_stat->list, &hw_stats_list);
1521		mutex_unlock(&net_dm_mutex);
1522		break;
1523	case NETDEV_UNREGISTER:
1524		mutex_lock(&net_dm_mutex);
1525		list_for_each_entry_safe(new_stat, tmp, &hw_stats_list, list) {
1526			if (new_stat->dev == dev) {
1527				new_stat->dev = NULL;
1528				if (trace_state == TRACE_OFF) {
1529					list_del_rcu(&new_stat->list);
1530					kfree_rcu(new_stat, rcu);
1531					break;
1532				}
1533			}
1534		}
1535		mutex_unlock(&net_dm_mutex);
1536		break;
1537	}
1538out:
1539	return NOTIFY_DONE;
1540}
1541
1542static const struct nla_policy net_dm_nl_policy[NET_DM_ATTR_MAX + 1] = {
1543	[NET_DM_ATTR_UNSPEC] = { .strict_start_type = NET_DM_ATTR_UNSPEC + 1 },
1544	[NET_DM_ATTR_ALERT_MODE] = { .type = NLA_U8 },
1545	[NET_DM_ATTR_TRUNC_LEN] = { .type = NLA_U32 },
1546	[NET_DM_ATTR_QUEUE_LEN] = { .type = NLA_U32 },
1547	[NET_DM_ATTR_SW_DROPS]	= {. type = NLA_FLAG },
1548	[NET_DM_ATTR_HW_DROPS]	= {. type = NLA_FLAG },
1549};
1550
1551static const struct genl_ops dropmon_ops[] = {
1552	{
1553		.cmd = NET_DM_CMD_CONFIG,
1554		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1555		.doit = net_dm_cmd_config,
1556		.flags = GENL_ADMIN_PERM,
1557	},
1558	{
1559		.cmd = NET_DM_CMD_START,
1560		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1561		.doit = net_dm_cmd_trace,
 
1562	},
1563	{
1564		.cmd = NET_DM_CMD_STOP,
1565		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1566		.doit = net_dm_cmd_trace,
 
1567	},
1568	{
1569		.cmd = NET_DM_CMD_CONFIG_GET,
1570		.doit = net_dm_cmd_config_get,
1571	},
1572	{
1573		.cmd = NET_DM_CMD_STATS_GET,
1574		.doit = net_dm_cmd_stats_get,
1575	},
1576};
1577
1578static int net_dm_nl_pre_doit(const struct genl_ops *ops,
1579			      struct sk_buff *skb, struct genl_info *info)
1580{
1581	mutex_lock(&net_dm_mutex);
1582
1583	return 0;
1584}
1585
1586static void net_dm_nl_post_doit(const struct genl_ops *ops,
1587				struct sk_buff *skb, struct genl_info *info)
1588{
1589	mutex_unlock(&net_dm_mutex);
1590}
1591
1592static struct genl_family net_drop_monitor_family __ro_after_init = {
1593	.hdrsize        = 0,
1594	.name           = "NET_DM",
1595	.version        = 2,
1596	.maxattr	= NET_DM_ATTR_MAX,
1597	.policy		= net_dm_nl_policy,
1598	.pre_doit	= net_dm_nl_pre_doit,
1599	.post_doit	= net_dm_nl_post_doit,
1600	.module		= THIS_MODULE,
1601	.ops		= dropmon_ops,
1602	.n_ops		= ARRAY_SIZE(dropmon_ops),
 
1603	.mcgrps		= dropmon_mcgrps,
1604	.n_mcgrps	= ARRAY_SIZE(dropmon_mcgrps),
1605};
1606
1607static struct notifier_block dropmon_net_notifier = {
1608	.notifier_call = dropmon_net_event
1609};
1610
1611static void __net_dm_cpu_data_init(struct per_cpu_dm_data *data)
1612{
1613	spin_lock_init(&data->lock);
1614	skb_queue_head_init(&data->drop_queue);
1615	u64_stats_init(&data->stats.syncp);
1616}
1617
1618static void __net_dm_cpu_data_fini(struct per_cpu_dm_data *data)
1619{
1620	WARN_ON(!skb_queue_empty(&data->drop_queue));
1621}
1622
1623static void net_dm_cpu_data_init(int cpu)
1624{
1625	struct per_cpu_dm_data *data;
1626
1627	data = &per_cpu(dm_cpu_data, cpu);
1628	__net_dm_cpu_data_init(data);
1629}
1630
1631static void net_dm_cpu_data_fini(int cpu)
1632{
1633	struct per_cpu_dm_data *data;
1634
1635	data = &per_cpu(dm_cpu_data, cpu);
1636	/* At this point, we should have exclusive access
1637	 * to this struct and can free the skb inside it.
1638	 */
1639	consume_skb(data->skb);
1640	__net_dm_cpu_data_fini(data);
1641}
1642
1643static void net_dm_hw_cpu_data_init(int cpu)
1644{
1645	struct per_cpu_dm_data *hw_data;
1646
1647	hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1648	__net_dm_cpu_data_init(hw_data);
1649}
1650
1651static void net_dm_hw_cpu_data_fini(int cpu)
1652{
1653	struct per_cpu_dm_data *hw_data;
1654
1655	hw_data = &per_cpu(dm_hw_cpu_data, cpu);
1656	kfree(hw_data->hw_entries);
1657	__net_dm_cpu_data_fini(hw_data);
1658}
1659
1660static int __init init_net_drop_monitor(void)
1661{
1662	int cpu, rc;
1663
1664	pr_info("Initializing network drop monitor service\n");
1665
1666	if (sizeof(void *) > 8) {
1667		pr_err("Unable to store program counters on this arch, Drop monitor failed\n");
1668		return -ENOSPC;
1669	}
1670
1671	rc = genl_register_family(&net_drop_monitor_family);
1672	if (rc) {
1673		pr_err("Could not create drop monitor netlink family\n");
1674		return rc;
1675	}
1676	WARN_ON(net_drop_monitor_family.mcgrp_offset != NET_DM_GRP_ALERT);
1677
1678	rc = register_netdevice_notifier(&dropmon_net_notifier);
1679	if (rc < 0) {
1680		pr_crit("Failed to register netdevice notifier\n");
1681		goto out_unreg;
1682	}
1683
1684	rc = 0;
1685
1686	for_each_possible_cpu(cpu) {
1687		net_dm_cpu_data_init(cpu);
1688		net_dm_hw_cpu_data_init(cpu);
1689	}
1690
1691	goto out;
1692
1693out_unreg:
1694	genl_unregister_family(&net_drop_monitor_family);
1695out:
1696	return rc;
1697}
1698
1699static void exit_net_drop_monitor(void)
1700{
1701	int cpu;
1702
1703	BUG_ON(unregister_netdevice_notifier(&dropmon_net_notifier));
1704
1705	/*
1706	 * Because of the module_get/put we do in the trace state change path
1707	 * we are guarnateed not to have any current users when we get here
1708	 */
1709
1710	for_each_possible_cpu(cpu) {
1711		net_dm_hw_cpu_data_fini(cpu);
1712		net_dm_cpu_data_fini(cpu);
1713	}
1714
1715	BUG_ON(genl_unregister_family(&net_drop_monitor_family));
1716}
1717
1718module_init(init_net_drop_monitor);
1719module_exit(exit_net_drop_monitor);
1720
1721MODULE_LICENSE("GPL v2");
1722MODULE_AUTHOR("Neil Horman <nhorman@tuxdriver.com>");
1723MODULE_ALIAS_GENL_FAMILY("NET_DM");
1724MODULE_DESCRIPTION("Monitoring code for network dropped packet alerts");