linux/net/batman-adv/tp_meter.c
<<
>>
Prefs
   1// SPDX-License-Identifier: GPL-2.0
   2/* Copyright (C) 2012-2019  B.A.T.M.A.N. contributors:
   3 *
   4 * Edo Monticelli, Antonio Quartulli
   5 */
   6
   7#include "tp_meter.h"
   8#include "main.h"
   9
  10#include <linux/atomic.h>
  11#include <linux/build_bug.h>
  12#include <linux/byteorder/generic.h>
  13#include <linux/cache.h>
  14#include <linux/compiler.h>
  15#include <linux/err.h>
  16#include <linux/etherdevice.h>
  17#include <linux/gfp.h>
  18#include <linux/if_ether.h>
  19#include <linux/init.h>
  20#include <linux/jiffies.h>
  21#include <linux/kernel.h>
  22#include <linux/kref.h>
  23#include <linux/kthread.h>
  24#include <linux/limits.h>
  25#include <linux/list.h>
  26#include <linux/netdevice.h>
  27#include <linux/param.h>
  28#include <linux/printk.h>
  29#include <linux/random.h>
  30#include <linux/rculist.h>
  31#include <linux/rcupdate.h>
  32#include <linux/sched.h>
  33#include <linux/skbuff.h>
  34#include <linux/slab.h>
  35#include <linux/spinlock.h>
  36#include <linux/stddef.h>
  37#include <linux/string.h>
  38#include <linux/timer.h>
  39#include <linux/wait.h>
  40#include <linux/workqueue.h>
  41#include <uapi/linux/batadv_packet.h>
  42#include <uapi/linux/batman_adv.h>
  43
  44#include "hard-interface.h"
  45#include "log.h"
  46#include "netlink.h"
  47#include "originator.h"
  48#include "send.h"
  49
  50/**
  51 * BATADV_TP_DEF_TEST_LENGTH - Default test length if not specified by the user
  52 *  in milliseconds
  53 */
  54#define BATADV_TP_DEF_TEST_LENGTH 10000
  55
  56/**
  57 * BATADV_TP_AWND - Advertised window by the receiver (in bytes)
  58 */
  59#define BATADV_TP_AWND 0x20000000
  60
  61/**
  62 * BATADV_TP_RECV_TIMEOUT - Receiver activity timeout. If the receiver does not
  63 *  get anything for such amount of milliseconds, the connection is killed
  64 */
  65#define BATADV_TP_RECV_TIMEOUT 1000
  66
  67/**
  68 * BATADV_TP_MAX_RTO - Maximum sender timeout. If the sender RTO gets beyond
  69 * such amound of milliseconds, the receiver is considered unreachable and the
  70 * connection is killed
  71 */
  72#define BATADV_TP_MAX_RTO 30000
  73
  74/**
  75 * BATADV_TP_FIRST_SEQ - First seqno of each session. The number is rather high
  76 *  in order to immediately trigger a wrap around (test purposes)
  77 */
  78#define BATADV_TP_FIRST_SEQ ((u32)-1 - 2000)
  79
  80/**
  81 * BATADV_TP_PLEN - length of the payload (data after the batadv_unicast header)
  82 *  to simulate
  83 */
  84#define BATADV_TP_PLEN (BATADV_TP_PACKET_LEN - ETH_HLEN - \
  85                        sizeof(struct batadv_unicast_packet))
  86
  87static u8 batadv_tp_prerandom[4096] __read_mostly;
  88
  89/**
  90 * batadv_tp_session_cookie() - generate session cookie based on session ids
  91 * @session: TP session identifier
  92 * @icmp_uid: icmp pseudo uid of the tp session
  93 *
  94 * Return: 32 bit tp_meter session cookie
  95 */
  96static u32 batadv_tp_session_cookie(const u8 session[2], u8 icmp_uid)
  97{
  98        u32 cookie;
  99
 100        cookie = icmp_uid << 16;
 101        cookie |= session[0] << 8;
 102        cookie |= session[1];
 103
 104        return cookie;
 105}
 106
 107/**
 108 * batadv_tp_cwnd() - compute the new cwnd size
 109 * @base: base cwnd size value
 110 * @increment: the value to add to base to get the new size
 111 * @min: minumim cwnd value (usually MSS)
 112 *
 113 * Return the new cwnd size and ensures it does not exceed the Advertised
 114 * Receiver Window size. It is wrap around safe.
 115 * For details refer to Section 3.1 of RFC5681
 116 *
 117 * Return: new congestion window size in bytes
 118 */
 119static u32 batadv_tp_cwnd(u32 base, u32 increment, u32 min)
 120{
 121        u32 new_size = base + increment;
 122
 123        /* check for wrap-around */
 124        if (new_size < base)
 125                new_size = (u32)ULONG_MAX;
 126
 127        new_size = min_t(u32, new_size, BATADV_TP_AWND);
 128
 129        return max_t(u32, new_size, min);
 130}
 131
 132/**
 133 * batadv_tp_updated_cwnd() - update the Congestion Windows
 134 * @tp_vars: the private data of the current TP meter session
 135 * @mss: maximum segment size of transmission
 136 *
 137 * 1) if the session is in Slow Start, the CWND has to be increased by 1
 138 * MSS every unique received ACK
 139 * 2) if the session is in Congestion Avoidance, the CWND has to be
 140 * increased by MSS * MSS / CWND for every unique received ACK
 141 */
 142static void batadv_tp_update_cwnd(struct batadv_tp_vars *tp_vars, u32 mss)
 143{
 144        spin_lock_bh(&tp_vars->cwnd_lock);
 145
 146        /* slow start... */
 147        if (tp_vars->cwnd <= tp_vars->ss_threshold) {
 148                tp_vars->dec_cwnd = 0;
 149                tp_vars->cwnd = batadv_tp_cwnd(tp_vars->cwnd, mss, mss);
 150                spin_unlock_bh(&tp_vars->cwnd_lock);
 151                return;
 152        }
 153
 154        /* increment CWND at least of 1 (section 3.1 of RFC5681) */
 155        tp_vars->dec_cwnd += max_t(u32, 1U << 3,
 156                                   ((mss * mss) << 6) / (tp_vars->cwnd << 3));
 157        if (tp_vars->dec_cwnd < (mss << 3)) {
 158                spin_unlock_bh(&tp_vars->cwnd_lock);
 159                return;
 160        }
 161
 162        tp_vars->cwnd = batadv_tp_cwnd(tp_vars->cwnd, mss, mss);
 163        tp_vars->dec_cwnd = 0;
 164
 165        spin_unlock_bh(&tp_vars->cwnd_lock);
 166}
 167
 168/**
 169 * batadv_tp_update_rto() - calculate new retransmission timeout
 170 * @tp_vars: the private data of the current TP meter session
 171 * @new_rtt: new roundtrip time in msec
 172 */
 173static void batadv_tp_update_rto(struct batadv_tp_vars *tp_vars,
 174                                 u32 new_rtt)
 175{
 176        long m = new_rtt;
 177
 178        /* RTT update
 179         * Details in Section 2.2 and 2.3 of RFC6298
 180         *
 181         * It's tricky to understand. Don't lose hair please.
 182         * Inspired by tcp_rtt_estimator() tcp_input.c
 183         */
 184        if (tp_vars->srtt != 0) {
 185                m -= (tp_vars->srtt >> 3); /* m is now error in rtt est */
 186                tp_vars->srtt += m; /* rtt = 7/8 srtt + 1/8 new */
 187                if (m < 0)
 188                        m = -m;
 189
 190                m -= (tp_vars->rttvar >> 2);
 191                tp_vars->rttvar += m; /* mdev ~= 3/4 rttvar + 1/4 new */
 192        } else {
 193                /* first measure getting in */
 194                tp_vars->srtt = m << 3; /* take the measured time to be srtt */
 195                tp_vars->rttvar = m << 1; /* new_rtt / 2 */
 196        }
 197
 198        /* rto = srtt + 4 * rttvar.
 199         * rttvar is scaled by 4, therefore doesn't need to be multiplied
 200         */
 201        tp_vars->rto = (tp_vars->srtt >> 3) + tp_vars->rttvar;
 202}
 203
 204/**
 205 * batadv_tp_batctl_notify() - send client status result to client
 206 * @reason: reason for tp meter session stop
 207 * @dst: destination of tp_meter session
 208 * @bat_priv: the bat priv with all the soft interface information
 209 * @start_time: start of transmission in jiffies
 210 * @total_sent: bytes acked to the receiver
 211 * @cookie: cookie of tp_meter session
 212 */
 213static void batadv_tp_batctl_notify(enum batadv_tp_meter_reason reason,
 214                                    const u8 *dst, struct batadv_priv *bat_priv,
 215                                    unsigned long start_time, u64 total_sent,
 216                                    u32 cookie)
 217{
 218        u32 test_time;
 219        u8 result;
 220        u32 total_bytes;
 221
 222        if (!batadv_tp_is_error(reason)) {
 223                result = BATADV_TP_REASON_COMPLETE;
 224                test_time = jiffies_to_msecs(jiffies - start_time);
 225                total_bytes = total_sent;
 226        } else {
 227                result = reason;
 228                test_time = 0;
 229                total_bytes = 0;
 230        }
 231
 232        batadv_netlink_tpmeter_notify(bat_priv, dst, result, test_time,
 233                                      total_bytes, cookie);
 234}
 235
 236/**
 237 * batadv_tp_batctl_error_notify() - send client error result to client
 238 * @reason: reason for tp meter session stop
 239 * @dst: destination of tp_meter session
 240 * @bat_priv: the bat priv with all the soft interface information
 241 * @cookie: cookie of tp_meter session
 242 */
 243static void batadv_tp_batctl_error_notify(enum batadv_tp_meter_reason reason,
 244                                          const u8 *dst,
 245                                          struct batadv_priv *bat_priv,
 246                                          u32 cookie)
 247{
 248        batadv_tp_batctl_notify(reason, dst, bat_priv, 0, 0, cookie);
 249}
 250
 251/**
 252 * batadv_tp_list_find() - find a tp_vars object in the global list
 253 * @bat_priv: the bat priv with all the soft interface information
 254 * @dst: the other endpoint MAC address to look for
 255 *
 256 * Look for a tp_vars object matching dst as end_point and return it after
 257 * having incremented the refcounter. Return NULL is not found
 258 *
 259 * Return: matching tp_vars or NULL when no tp_vars with @dst was found
 260 */
 261static struct batadv_tp_vars *batadv_tp_list_find(struct batadv_priv *bat_priv,
 262                                                  const u8 *dst)
 263{
 264        struct batadv_tp_vars *pos, *tp_vars = NULL;
 265
 266        rcu_read_lock();
 267        hlist_for_each_entry_rcu(pos, &bat_priv->tp_list, list) {
 268                if (!batadv_compare_eth(pos->other_end, dst))
 269                        continue;
 270
 271                /* most of the time this function is invoked during the normal
 272                 * process..it makes sens to pay more when the session is
 273                 * finished and to speed the process up during the measurement
 274                 */
 275                if (unlikely(!kref_get_unless_zero(&pos->refcount)))
 276                        continue;
 277
 278                tp_vars = pos;
 279                break;
 280        }
 281        rcu_read_unlock();
 282
 283        return tp_vars;
 284}
 285
 286/**
 287 * batadv_tp_list_find_session() - find tp_vars session object in the global
 288 *  list
 289 * @bat_priv: the bat priv with all the soft interface information
 290 * @dst: the other endpoint MAC address to look for
 291 * @session: session identifier
 292 *
 293 * Look for a tp_vars object matching dst as end_point, session as tp meter
 294 * session and return it after having incremented the refcounter. Return NULL
 295 * is not found
 296 *
 297 * Return: matching tp_vars or NULL when no tp_vars was found
 298 */
 299static struct batadv_tp_vars *
 300batadv_tp_list_find_session(struct batadv_priv *bat_priv, const u8 *dst,
 301                            const u8 *session)
 302{
 303        struct batadv_tp_vars *pos, *tp_vars = NULL;
 304
 305        rcu_read_lock();
 306        hlist_for_each_entry_rcu(pos, &bat_priv->tp_list, list) {
 307                if (!batadv_compare_eth(pos->other_end, dst))
 308                        continue;
 309
 310                if (memcmp(pos->session, session, sizeof(pos->session)) != 0)
 311                        continue;
 312
 313                /* most of the time this function is invoked during the normal
 314                 * process..it makes sense to pay more when the session is
 315                 * finished and to speed the process up during the measurement
 316                 */
 317                if (unlikely(!kref_get_unless_zero(&pos->refcount)))
 318                        continue;
 319
 320                tp_vars = pos;
 321                break;
 322        }
 323        rcu_read_unlock();
 324
 325        return tp_vars;
 326}
 327
 328/**
 329 * batadv_tp_vars_release() - release batadv_tp_vars from lists and queue for
 330 *  free after rcu grace period
 331 * @ref: kref pointer of the batadv_tp_vars
 332 */
 333static void batadv_tp_vars_release(struct kref *ref)
 334{
 335        struct batadv_tp_vars *tp_vars;
 336        struct batadv_tp_unacked *un, *safe;
 337
 338        tp_vars = container_of(ref, struct batadv_tp_vars, refcount);
 339
 340        /* lock should not be needed because this object is now out of any
 341         * context!
 342         */
 343        spin_lock_bh(&tp_vars->unacked_lock);
 344        list_for_each_entry_safe(un, safe, &tp_vars->unacked_list, list) {
 345                list_del(&un->list);
 346                kfree(un);
 347        }
 348        spin_unlock_bh(&tp_vars->unacked_lock);
 349
 350        kfree_rcu(tp_vars, rcu);
 351}
 352
 353/**
 354 * batadv_tp_vars_put() - decrement the batadv_tp_vars refcounter and possibly
 355 *  release it
 356 * @tp_vars: the private data of the current TP meter session to be free'd
 357 */
 358static void batadv_tp_vars_put(struct batadv_tp_vars *tp_vars)
 359{
 360        kref_put(&tp_vars->refcount, batadv_tp_vars_release);
 361}
 362
 363/**
 364 * batadv_tp_sender_cleanup() - cleanup sender data and drop and timer
 365 * @bat_priv: the bat priv with all the soft interface information
 366 * @tp_vars: the private data of the current TP meter session to cleanup
 367 */
 368static void batadv_tp_sender_cleanup(struct batadv_priv *bat_priv,
 369                                     struct batadv_tp_vars *tp_vars)
 370{
 371        cancel_delayed_work(&tp_vars->finish_work);
 372
 373        spin_lock_bh(&tp_vars->bat_priv->tp_list_lock);
 374        hlist_del_rcu(&tp_vars->list);
 375        spin_unlock_bh(&tp_vars->bat_priv->tp_list_lock);
 376
 377        /* drop list reference */
 378        batadv_tp_vars_put(tp_vars);
 379
 380        atomic_dec(&tp_vars->bat_priv->tp_num);
 381
 382        /* kill the timer and remove its reference */
 383        del_timer_sync(&tp_vars->timer);
 384        /* the worker might have rearmed itself therefore we kill it again. Note
 385         * that if the worker should run again before invoking the following
 386         * del_timer(), it would not re-arm itself once again because the status
 387         * is OFF now
 388         */
 389        del_timer(&tp_vars->timer);
 390        batadv_tp_vars_put(tp_vars);
 391}
 392
 393/**
 394 * batadv_tp_sender_end() - print info about ended session and inform client
 395 * @bat_priv: the bat priv with all the soft interface information
 396 * @tp_vars: the private data of the current TP meter session
 397 */
 398static void batadv_tp_sender_end(struct batadv_priv *bat_priv,
 399                                 struct batadv_tp_vars *tp_vars)
 400{
 401        u32 session_cookie;
 402
 403        batadv_dbg(BATADV_DBG_TP_METER, bat_priv,
 404                   "Test towards %pM finished..shutting down (reason=%d)\n",
 405                   tp_vars->other_end, tp_vars->reason);
 406
 407        batadv_dbg(BATADV_DBG_TP_METER, bat_priv,
 408                   "Last timing stats: SRTT=%ums RTTVAR=%ums RTO=%ums\n",
 409                   tp_vars->srtt >> 3, tp_vars->rttvar >> 2, tp_vars->rto);
 410
 411        batadv_dbg(BATADV_DBG_TP_METER, bat_priv,
 412                   "Final values: cwnd=%u ss_threshold=%u\n",
 413                   tp_vars->cwnd, tp_vars->ss_threshold);
 414
 415        session_cookie = batadv_tp_session_cookie(tp_vars->session,
 416                                                  tp_vars->icmp_uid);
 417
 418        batadv_tp_batctl_notify(tp_vars->reason,
 419                                tp_vars->other_end,
 420                                bat_priv,
 421                                tp_vars->start_time,
 422                                atomic64_read(&tp_vars->tot_sent),
 423                                session_cookie);
 424}
 425
 426/**
 427 * batadv_tp_sender_shutdown() - let sender thread/timer stop gracefully
 428 * @tp_vars: the private data of the current TP meter session
 429 * @reason: reason for tp meter session stop
 430 */
 431static void batadv_tp_sender_shutdown(struct batadv_tp_vars *tp_vars,
 432                                      enum batadv_tp_meter_reason reason)
 433{
 434        if (!atomic_dec_and_test(&tp_vars->sending))
 435                return;
 436
 437        tp_vars->reason = reason;
 438}
 439
 440/**
 441 * batadv_tp_sender_finish() - stop sender session after test_length was reached
 442 * @work: delayed work reference of the related tp_vars
 443 */
 444static void batadv_tp_sender_finish(struct work_struct *work)
 445{
 446        struct delayed_work *delayed_work;
 447        struct batadv_tp_vars *tp_vars;
 448
 449        delayed_work = to_delayed_work(work);
 450        tp_vars = container_of(delayed_work, struct batadv_tp_vars,
 451                               finish_work);
 452
 453        batadv_tp_sender_shutdown(tp_vars, BATADV_TP_REASON_COMPLETE);
 454}
 455
 456/**
 457 * batadv_tp_reset_sender_timer() - reschedule the sender timer
 458 * @tp_vars: the private TP meter data for this session
 459 *
 460 * Reschedule the timer using tp_vars->rto as delay
 461 */
 462static void batadv_tp_reset_sender_timer(struct batadv_tp_vars *tp_vars)
 463{
 464        /* most of the time this function is invoked while normal packet
 465         * reception...
 466         */
 467        if (unlikely(atomic_read(&tp_vars->sending) == 0))
 468                /* timer ref will be dropped in batadv_tp_sender_cleanup */
 469                return;
 470
 471        mod_timer(&tp_vars->timer, jiffies + msecs_to_jiffies(tp_vars->rto));
 472}
 473
 474/**
 475 * batadv_tp_sender_timeout() - timer that fires in case of packet loss
 476 * @t: address to timer_list inside tp_vars
 477 *
 478 * If fired it means that there was packet loss.
 479 * Switch to Slow Start, set the ss_threshold to half of the current cwnd and
 480 * reset the cwnd to 3*MSS
 481 */
 482static void batadv_tp_sender_timeout(struct timer_list *t)
 483{
 484        struct batadv_tp_vars *tp_vars = from_timer(tp_vars, t, timer);
 485        struct batadv_priv *bat_priv = tp_vars->bat_priv;
 486
 487        if (atomic_read(&tp_vars->sending) == 0)
 488                return;
 489
 490        /* if the user waited long enough...shutdown the test */
 491        if (unlikely(tp_vars->rto >= BATADV_TP_MAX_RTO)) {
 492                batadv_tp_sender_shutdown(tp_vars,
 493                                          BATADV_TP_REASON_DST_UNREACHABLE);
 494                return;
 495        }
 496
 497        /* RTO exponential backoff
 498         * Details in Section 5.5 of RFC6298
 499         */
 500        tp_vars->rto <<= 1;
 501
 502        spin_lock_bh(&tp_vars->cwnd_lock);
 503
 504        tp_vars->ss_threshold = tp_vars->cwnd >> 1;
 505        if (tp_vars->ss_threshold < BATADV_TP_PLEN * 2)
 506                tp_vars->ss_threshold = BATADV_TP_PLEN * 2;
 507
 508        batadv_dbg(BATADV_DBG_TP_METER, bat_priv,
 509                   "Meter: RTO fired during test towards %pM! cwnd=%u new ss_thr=%u, resetting last_sent to %u\n",
 510                   tp_vars->other_end, tp_vars->cwnd, tp_vars->ss_threshold,
 511                   atomic_read(&tp_vars->last_acked));
 512
 513        tp_vars->cwnd = BATADV_TP_PLEN * 3;
 514
 515        spin_unlock_bh(&tp_vars->cwnd_lock);
 516
 517        /* resend the non-ACKed packets.. */
 518        tp_vars->last_sent = atomic_read(&tp_vars->last_acked);
 519        wake_up(&tp_vars->more_bytes);
 520
 521        batadv_tp_reset_sender_timer(tp_vars);
 522}
 523
 524/**
 525 * batadv_tp_fill_prerandom() - Fill buffer with prefetched random bytes
 526 * @tp_vars: the private TP meter data for this session
 527 * @buf: Buffer to fill with bytes
 528 * @nbytes: amount of pseudorandom bytes
 529 */
 530static void batadv_tp_fill_prerandom(struct batadv_tp_vars *tp_vars,
 531                                     u8 *buf, size_t nbytes)
 532{
 533        u32 local_offset;
 534        size_t bytes_inbuf;
 535        size_t to_copy;
 536        size_t pos = 0;
 537
 538        spin_lock_bh(&tp_vars->prerandom_lock);
 539        local_offset = tp_vars->prerandom_offset;
 540        tp_vars->prerandom_offset += nbytes;
 541        tp_vars->prerandom_offset %= sizeof(batadv_tp_prerandom);
 542        spin_unlock_bh(&tp_vars->prerandom_lock);
 543
 544        while (nbytes) {
 545                local_offset %= sizeof(batadv_tp_prerandom);
 546                bytes_inbuf = sizeof(batadv_tp_prerandom) - local_offset;
 547                to_copy = min(nbytes, bytes_inbuf);
 548
 549                memcpy(&buf[pos], &batadv_tp_prerandom[local_offset], to_copy);
 550                pos += to_copy;
 551                nbytes -= to_copy;
 552                local_offset = 0;
 553        }
 554}
 555
 556/**
 557 * batadv_tp_send_msg() - send a single message
 558 * @tp_vars: the private TP meter data for this session
 559 * @src: source mac address
 560 * @orig_node: the originator of the destination
 561 * @seqno: sequence number of this packet
 562 * @len: length of the entire packet
 563 * @session: session identifier
 564 * @uid: local ICMP "socket" index
 565 * @timestamp: timestamp in jiffies which is replied in ack
 566 *
 567 * Create and send a single TP Meter message.
 568 *
 569 * Return: 0 on success, BATADV_TP_REASON_DST_UNREACHABLE if the destination is
 570 * not reachable, BATADV_TP_REASON_MEMORY_ERROR if the packet couldn't be
 571 * allocated
 572 */
 573static int batadv_tp_send_msg(struct batadv_tp_vars *tp_vars, const u8 *src,
 574                              struct batadv_orig_node *orig_node,
 575                              u32 seqno, size_t len, const u8 *session,
 576                              int uid, u32 timestamp)
 577{
 578        struct batadv_icmp_tp_packet *icmp;
 579        struct sk_buff *skb;
 580        int r;
 581        u8 *data;
 582        size_t data_len;
 583
 584        skb = netdev_alloc_skb_ip_align(NULL, len + ETH_HLEN);
 585        if (unlikely(!skb))
 586                return BATADV_TP_REASON_MEMORY_ERROR;
 587
 588        skb_reserve(skb, ETH_HLEN);
 589        icmp = skb_put(skb, sizeof(*icmp));
 590
 591        /* fill the icmp header */
 592        ether_addr_copy(icmp->dst, orig_node->orig);
 593        ether_addr_copy(icmp->orig, src);
 594        icmp->version = BATADV_COMPAT_VERSION;
 595        icmp->packet_type = BATADV_ICMP;
 596        icmp->ttl = BATADV_TTL;
 597        icmp->msg_type = BATADV_TP;
 598        icmp->uid = uid;
 599
 600        icmp->subtype = BATADV_TP_MSG;
 601        memcpy(icmp->session, session, sizeof(icmp->session));
 602        icmp->seqno = htonl(seqno);
 603        icmp->timestamp = htonl(timestamp);
 604
 605        data_len = len - sizeof(*icmp);
 606        data = skb_put(skb, data_len);
 607        batadv_tp_fill_prerandom(tp_vars, data, data_len);
 608
 609        r = batadv_send_skb_to_orig(skb, orig_node, NULL);
 610        if (r == NET_XMIT_SUCCESS)
 611                return 0;
 612
 613        return BATADV_TP_REASON_CANT_SEND;
 614}
 615
 616/**
 617 * batadv_tp_recv_ack() - ACK receiving function
 618 * @bat_priv: the bat priv with all the soft interface information
 619 * @skb: the buffer containing the received packet
 620 *
 621 * Process a received TP ACK packet
 622 */
 623static void batadv_tp_recv_ack(struct batadv_priv *bat_priv,
 624                               const struct sk_buff *skb)
 625{
 626        struct batadv_hard_iface *primary_if = NULL;
 627        struct batadv_orig_node *orig_node = NULL;
 628        const struct batadv_icmp_tp_packet *icmp;
 629        struct batadv_tp_vars *tp_vars;
 630        size_t packet_len, mss;
 631        u32 rtt, recv_ack, cwnd;
 632        unsigned char *dev_addr;
 633
 634        packet_len = BATADV_TP_PLEN;
 635        mss = BATADV_TP_PLEN;
 636        packet_len += sizeof(struct batadv_unicast_packet);
 637
 638        icmp = (struct batadv_icmp_tp_packet *)skb->data;
 639
 640        /* find the tp_vars */
 641        tp_vars = batadv_tp_list_find_session(bat_priv, icmp->orig,
 642                                              icmp->session);
 643        if (unlikely(!tp_vars))
 644                return;
 645
 646        if (unlikely(atomic_read(&tp_vars->sending) == 0))
 647                goto out;
 648
 649        /* old ACK? silently drop it.. */
 650        if (batadv_seq_before(ntohl(icmp->seqno),
 651                              (u32)atomic_read(&tp_vars->last_acked)))
 652                goto out;
 653
 654        primary_if = batadv_primary_if_get_selected(bat_priv);
 655        if (unlikely(!primary_if))
 656                goto out;
 657
 658        orig_node = batadv_orig_hash_find(bat_priv, icmp->orig);
 659        if (unlikely(!orig_node))
 660                goto out;
 661
 662        /* update RTO with the new sampled RTT, if any */
 663        rtt = jiffies_to_msecs(jiffies) - ntohl(icmp->timestamp);
 664        if (icmp->timestamp && rtt)
 665                batadv_tp_update_rto(tp_vars, rtt);
 666
 667        /* ACK for new data... reset the timer */
 668        batadv_tp_reset_sender_timer(tp_vars);
 669
 670        recv_ack = ntohl(icmp->seqno);
 671
 672        /* check if this ACK is a duplicate */
 673        if (atomic_read(&tp_vars->last_acked) == recv_ack) {
 674                atomic_inc(&tp_vars->dup_acks);
 675                if (atomic_read(&tp_vars->dup_acks) != 3)
 676                        goto out;
 677
 678                if (recv_ack >= tp_vars->recover)
 679                        goto out;
 680
 681                /* if this is the third duplicate ACK do Fast Retransmit */
 682                batadv_tp_send_msg(tp_vars, primary_if->net_dev->dev_addr,
 683                                   orig_node, recv_ack, packet_len,
 684                                   icmp->session, icmp->uid,
 685                                   jiffies_to_msecs(jiffies));
 686
 687                spin_lock_bh(&tp_vars->cwnd_lock);
 688
 689                /* Fast Recovery */
 690                tp_vars->fast_recovery = true;
 691                /* Set recover to the last outstanding seqno when Fast Recovery
 692                 * is entered. RFC6582, Section 3.2, step 1
 693                 */
 694                tp_vars->recover = tp_vars->last_sent;
 695                tp_vars->ss_threshold = tp_vars->cwnd >> 1;
 696                batadv_dbg(BATADV_DBG_TP_METER, bat_priv,
 697                           "Meter: Fast Recovery, (cur cwnd=%u) ss_thr=%u last_sent=%u recv_ack=%u\n",
 698                           tp_vars->cwnd, tp_vars->ss_threshold,
 699                           tp_vars->last_sent, recv_ack);
 700                tp_vars->cwnd = batadv_tp_cwnd(tp_vars->ss_threshold, 3 * mss,
 701                                               mss);
 702                tp_vars->dec_cwnd = 0;
 703                tp_vars->last_sent = recv_ack;
 704
 705                spin_unlock_bh(&tp_vars->cwnd_lock);
 706        } else {
 707                /* count the acked data */
 708                atomic64_add(recv_ack - atomic_read(&tp_vars->last_acked),
 709                             &tp_vars->tot_sent);
 710                /* reset the duplicate ACKs counter */
 711                atomic_set(&tp_vars->dup_acks, 0);
 712
 713                if (tp_vars->fast_recovery) {
 714                        /* partial ACK */
 715                        if (batadv_seq_before(recv_ack, tp_vars->recover)) {
 716                                /* this is another hole in the window. React
 717                                 * immediately as specified by NewReno (see
 718                                 * Section 3.2 of RFC6582 for details)
 719                                 */
 720                                dev_addr = primary_if->net_dev->dev_addr;
 721                                batadv_tp_send_msg(tp_vars, dev_addr,
 722                                                   orig_node, recv_ack,
 723                                                   packet_len, icmp->session,
 724                                                   icmp->uid,
 725                                                   jiffies_to_msecs(jiffies));
 726                                tp_vars->cwnd = batadv_tp_cwnd(tp_vars->cwnd,
 727                                                               mss, mss);
 728                        } else {
 729                                tp_vars->fast_recovery = false;
 730                                /* set cwnd to the value of ss_threshold at the
 731                                 * moment that Fast Recovery was entered.
 732                                 * RFC6582, Section 3.2, step 3
 733                                 */
 734                                cwnd = batadv_tp_cwnd(tp_vars->ss_threshold, 0,
 735                                                      mss);
 736                                tp_vars->cwnd = cwnd;
 737                        }
 738                        goto move_twnd;
 739                }
 740
 741                if (recv_ack - atomic_read(&tp_vars->last_acked) >= mss)
 742                        batadv_tp_update_cwnd(tp_vars, mss);
 743move_twnd:
 744                /* move the Transmit Window */
 745                atomic_set(&tp_vars->last_acked, recv_ack);
 746        }
 747
 748        wake_up(&tp_vars->more_bytes);
 749out:
 750        if (likely(primary_if))
 751                batadv_hardif_put(primary_if);
 752        if (likely(orig_node))
 753                batadv_orig_node_put(orig_node);
 754        if (likely(tp_vars))
 755                batadv_tp_vars_put(tp_vars);
 756}
 757
 758/**
 759 * batadv_tp_avail() - check if congestion window is not full
 760 * @tp_vars: the private data of the current TP meter session
 761 * @payload_len: size of the payload of a single message
 762 *
 763 * Return: true when congestion window is not full, false otherwise
 764 */
 765static bool batadv_tp_avail(struct batadv_tp_vars *tp_vars,
 766                            size_t payload_len)
 767{
 768        u32 win_left, win_limit;
 769
 770        win_limit = atomic_read(&tp_vars->last_acked) + tp_vars->cwnd;
 771        win_left = win_limit - tp_vars->last_sent;
 772
 773        return win_left >= payload_len;
 774}
 775
 776/**
 777 * batadv_tp_wait_available() - wait until congestion window becomes free or
 778 *  timeout is reached
 779 * @tp_vars: the private data of the current TP meter session
 780 * @plen: size of the payload of a single message
 781 *
 782 * Return: 0 if the condition evaluated to false after the timeout elapsed,
 783 *  1 if the condition evaluated to true after the timeout elapsed, the
 784 *  remaining jiffies (at least 1) if the condition evaluated to true before
 785 *  the timeout elapsed, or -ERESTARTSYS if it was interrupted by a signal.
 786 */
 787static int batadv_tp_wait_available(struct batadv_tp_vars *tp_vars, size_t plen)
 788{
 789        int ret;
 790
 791        ret = wait_event_interruptible_timeout(tp_vars->more_bytes,
 792                                               batadv_tp_avail(tp_vars, plen),
 793                                               HZ / 10);
 794
 795        return ret;
 796}
 797
 798/**
 799 * batadv_tp_send() - main sending thread of a tp meter session
 800 * @arg: address of the related tp_vars
 801 *
 802 * Return: nothing, this function never returns
 803 */
 804static int batadv_tp_send(void *arg)
 805{
 806        struct batadv_tp_vars *tp_vars = arg;
 807        struct batadv_priv *bat_priv = tp_vars->bat_priv;
 808        struct batadv_hard_iface *primary_if = NULL;
 809        struct batadv_orig_node *orig_node = NULL;
 810        size_t payload_len, packet_len;
 811        int err = 0;
 812
 813        if (unlikely(tp_vars->role != BATADV_TP_SENDER)) {
 814                err = BATADV_TP_REASON_DST_UNREACHABLE;
 815                tp_vars->reason = err;
 816                goto out;
 817        }
 818
 819        orig_node = batadv_orig_hash_find(bat_priv, tp_vars->other_end);
 820        if (unlikely(!orig_node)) {
 821                err = BATADV_TP_REASON_DST_UNREACHABLE;
 822                tp_vars->reason = err;
 823                goto out;
 824        }
 825
 826        primary_if = batadv_primary_if_get_selected(bat_priv);
 827        if (unlikely(!primary_if)) {
 828                err = BATADV_TP_REASON_DST_UNREACHABLE;
 829                tp_vars->reason = err;
 830                goto out;
 831        }
 832
 833        /* assume that all the hard_interfaces have a correctly
 834         * configured MTU, so use the soft_iface MTU as MSS.
 835         * This might not be true and in that case the fragmentation
 836         * should be used.
 837         * Now, try to send the packet as it is
 838         */
 839        payload_len = BATADV_TP_PLEN;
 840        BUILD_BUG_ON(sizeof(struct batadv_icmp_tp_packet) > BATADV_TP_PLEN);
 841
 842        batadv_tp_reset_sender_timer(tp_vars);
 843
 844        /* queue the worker in charge of terminating the test */
 845        queue_delayed_work(batadv_event_workqueue, &tp_vars->finish_work,
 846                           msecs_to_jiffies(tp_vars->test_length));
 847
 848        while (atomic_read(&tp_vars->sending) != 0) {
 849                if (unlikely(!batadv_tp_avail(tp_vars, payload_len))) {
 850                        batadv_tp_wait_available(tp_vars, payload_len);
 851                        continue;
 852                }
 853
 854                /* to emulate normal unicast traffic, add to the payload len
 855                 * the size of the unicast header
 856                 */
 857                packet_len = payload_len + sizeof(struct batadv_unicast_packet);
 858
 859                err = batadv_tp_send_msg(tp_vars, primary_if->net_dev->dev_addr,
 860                                         orig_node, tp_vars->last_sent,
 861                                         packet_len,
 862                                         tp_vars->session, tp_vars->icmp_uid,
 863                                         jiffies_to_msecs(jiffies));
 864
 865                /* something went wrong during the preparation/transmission */
 866                if (unlikely(err && err != BATADV_TP_REASON_CANT_SEND)) {
 867                        batadv_dbg(BATADV_DBG_TP_METER, bat_priv,
 868                                   "Meter: %s() cannot send packets (%d)\n",
 869                                   __func__, err);
 870                        /* ensure nobody else tries to stop the thread now */
 871                        if (atomic_dec_and_test(&tp_vars->sending))
 872                                tp_vars->reason = err;
 873                        break;
 874                }
 875
 876                /* right-shift the TWND */
 877                if (!err)
 878                        tp_vars->last_sent += payload_len;
 879
 880                cond_resched();
 881        }
 882
 883out:
 884        if (likely(primary_if))
 885                batadv_hardif_put(primary_if);
 886        if (likely(orig_node))
 887                batadv_orig_node_put(orig_node);
 888
 889        batadv_tp_sender_end(bat_priv, tp_vars);
 890        batadv_tp_sender_cleanup(bat_priv, tp_vars);
 891
 892        batadv_tp_vars_put(tp_vars);
 893
 894        do_exit(0);
 895}
 896
 897/**
 898 * batadv_tp_start_kthread() - start new thread which manages the tp meter
 899 *  sender
 900 * @tp_vars: the private data of the current TP meter session
 901 */
 902static void batadv_tp_start_kthread(struct batadv_tp_vars *tp_vars)
 903{
 904        struct task_struct *kthread;
 905        struct batadv_priv *bat_priv = tp_vars->bat_priv;
 906        u32 session_cookie;
 907
 908        kref_get(&tp_vars->refcount);
 909        kthread = kthread_create(batadv_tp_send, tp_vars, "kbatadv_tp_meter");
 910        if (IS_ERR(kthread)) {
 911                session_cookie = batadv_tp_session_cookie(tp_vars->session,
 912                                                          tp_vars->icmp_uid);
 913                pr_err("batadv: cannot create tp meter kthread\n");
 914                batadv_tp_batctl_error_notify(BATADV_TP_REASON_MEMORY_ERROR,
 915                                              tp_vars->other_end,
 916                                              bat_priv, session_cookie);
 917
 918                /* drop reserved reference for kthread */
 919                batadv_tp_vars_put(tp_vars);
 920
 921                /* cleanup of failed tp meter variables */
 922                batadv_tp_sender_cleanup(bat_priv, tp_vars);
 923                return;
 924        }
 925
 926        wake_up_process(kthread);
 927}
 928
 929/**
 930 * batadv_tp_start() - start a new tp meter session
 931 * @bat_priv: the bat priv with all the soft interface information
 932 * @dst: the receiver MAC address
 933 * @test_length: test length in milliseconds
 934 * @cookie: session cookie
 935 */
 936void batadv_tp_start(struct batadv_priv *bat_priv, const u8 *dst,
 937                     u32 test_length, u32 *cookie)
 938{
 939        struct batadv_tp_vars *tp_vars;
 940        u8 session_id[2];
 941        u8 icmp_uid;
 942        u32 session_cookie;
 943
 944        get_random_bytes(session_id, sizeof(session_id));
 945        get_random_bytes(&icmp_uid, 1);
 946        session_cookie = batadv_tp_session_cookie(session_id, icmp_uid);
 947        *cookie = session_cookie;
 948
 949        /* look for an already existing test towards this node */
 950        spin_lock_bh(&bat_priv->tp_list_lock);
 951        tp_vars = batadv_tp_list_find(bat_priv, dst);
 952        if (tp_vars) {
 953                spin_unlock_bh(&bat_priv->tp_list_lock);
 954                batadv_tp_vars_put(tp_vars);
 955                batadv_dbg(BATADV_DBG_TP_METER, bat_priv,
 956                           "Meter: test to or from the same node already ongoing, aborting\n");
 957                batadv_tp_batctl_error_notify(BATADV_TP_REASON_ALREADY_ONGOING,
 958                                              dst, bat_priv, session_cookie);
 959                return;
 960        }
 961
 962        if (!atomic_add_unless(&bat_priv->tp_num, 1, BATADV_TP_MAX_NUM)) {
 963                spin_unlock_bh(&bat_priv->tp_list_lock);
 964                batadv_dbg(BATADV_DBG_TP_METER, bat_priv,
 965                           "Meter: too many ongoing sessions, aborting (SEND)\n");
 966                batadv_tp_batctl_error_notify(BATADV_TP_REASON_TOO_MANY, dst,
 967                                              bat_priv, session_cookie);
 968                return;
 969        }
 970
 971        tp_vars = kmalloc(sizeof(*tp_vars), GFP_ATOMIC);
 972        if (!tp_vars) {
 973                spin_unlock_bh(&bat_priv->tp_list_lock);
 974                batadv_dbg(BATADV_DBG_TP_METER, bat_priv,
 975                           "Meter: %s cannot allocate list elements\n",
 976                           __func__);
 977                batadv_tp_batctl_error_notify(BATADV_TP_REASON_MEMORY_ERROR,
 978                                              dst, bat_priv, session_cookie);
 979                return;
 980        }
 981
 982        /* initialize tp_vars */
 983        ether_addr_copy(tp_vars->other_end, dst);
 984        kref_init(&tp_vars->refcount);
 985        tp_vars->role = BATADV_TP_SENDER;
 986        atomic_set(&tp_vars->sending, 1);
 987        memcpy(tp_vars->session, session_id, sizeof(session_id));
 988        tp_vars->icmp_uid = icmp_uid;
 989
 990        tp_vars->last_sent = BATADV_TP_FIRST_SEQ;
 991        atomic_set(&tp_vars->last_acked, BATADV_TP_FIRST_SEQ);
 992        tp_vars->fast_recovery = false;
 993        tp_vars->recover = BATADV_TP_FIRST_SEQ;
 994
 995        /* initialise the CWND to 3*MSS (Section 3.1 in RFC5681).
 996         * For batman-adv the MSS is the size of the payload received by the
 997         * soft_interface, hence its MTU
 998         */
 999        tp_vars->cwnd = BATADV_TP_PLEN * 3;
1000        /* at the beginning initialise the SS threshold to the biggest possible
1001         * window size, hence the AWND size
1002         */
1003        tp_vars->ss_threshold = BATADV_TP_AWND;
1004
1005        /* RTO initial value is 3 seconds.
1006         * Details in Section 2.1 of RFC6298
1007         */
1008        tp_vars->rto = 1000;
1009        tp_vars->srtt = 0;
1010        tp_vars->rttvar = 0;
1011
1012        atomic64_set(&tp_vars->tot_sent, 0);
1013
1014        kref_get(&tp_vars->refcount);
1015        timer_setup(&tp_vars->timer, batadv_tp_sender_timeout, 0);
1016
1017        tp_vars->bat_priv = bat_priv;
1018        tp_vars->start_time = jiffies;
1019
1020        init_waitqueue_head(&tp_vars->more_bytes);
1021
1022        spin_lock_init(&tp_vars->unacked_lock);
1023        INIT_LIST_HEAD(&tp_vars->unacked_list);
1024
1025        spin_lock_init(&tp_vars->cwnd_lock);
1026
1027        tp_vars->prerandom_offset = 0;
1028        spin_lock_init(&tp_vars->prerandom_lock);
1029
1030        kref_get(&tp_vars->refcount);
1031        hlist_add_head_rcu(&tp_vars->list, &bat_priv->tp_list);
1032        spin_unlock_bh(&bat_priv->tp_list_lock);
1033
1034        tp_vars->test_length = test_length;
1035        if (!tp_vars->test_length)
1036                tp_vars->test_length = BATADV_TP_DEF_TEST_LENGTH;
1037
1038        batadv_dbg(BATADV_DBG_TP_METER, bat_priv,
1039                   "Meter: starting throughput meter towards %pM (length=%ums)\n",
1040                   dst, test_length);
1041
1042        /* init work item for finished tp tests */
1043        INIT_DELAYED_WORK(&tp_vars->finish_work, batadv_tp_sender_finish);
1044
1045        /* start tp kthread. This way the write() call issued from userspace can
1046         * happily return and avoid to block
1047         */
1048        batadv_tp_start_kthread(tp_vars);
1049
1050        /* don't return reference to new tp_vars */
1051        batadv_tp_vars_put(tp_vars);
1052}
1053
1054/**
1055 * batadv_tp_stop() - stop currently running tp meter session
1056 * @bat_priv: the bat priv with all the soft interface information
1057 * @dst: the receiver MAC address
1058 * @return_value: reason for tp meter session stop
1059 */
1060void batadv_tp_stop(struct batadv_priv *bat_priv, const u8 *dst,
1061                    u8 return_value)
1062{
1063        struct batadv_orig_node *orig_node;
1064        struct batadv_tp_vars *tp_vars;
1065
1066        batadv_dbg(BATADV_DBG_TP_METER, bat_priv,
1067                   "Meter: stopping test towards %pM\n", dst);
1068
1069        orig_node = batadv_orig_hash_find(bat_priv, dst);
1070        if (!orig_node)
1071                return;
1072
1073        tp_vars = batadv_tp_list_find(bat_priv, orig_node->orig);
1074        if (!tp_vars) {
1075                batadv_dbg(BATADV_DBG_TP_METER, bat_priv,
1076                           "Meter: trying to interrupt an already over connection\n");
1077                goto out;
1078        }
1079
1080        batadv_tp_sender_shutdown(tp_vars, return_value);
1081        batadv_tp_vars_put(tp_vars);
1082out:
1083        batadv_orig_node_put(orig_node);
1084}
1085
1086/**
1087 * batadv_tp_reset_receiver_timer() - reset the receiver shutdown timer
1088 * @tp_vars: the private data of the current TP meter session
1089 *
1090 * start the receiver shutdown timer or reset it if already started
1091 */
1092static void batadv_tp_reset_receiver_timer(struct batadv_tp_vars *tp_vars)
1093{
1094        mod_timer(&tp_vars->timer,
1095                  jiffies + msecs_to_jiffies(BATADV_TP_RECV_TIMEOUT));
1096}
1097
1098/**
1099 * batadv_tp_receiver_shutdown() - stop a tp meter receiver when timeout is
1100 *  reached without received ack
1101 * @t: address to timer_list inside tp_vars
1102 */
1103static void batadv_tp_receiver_shutdown(struct timer_list *t)
1104{
1105        struct batadv_tp_vars *tp_vars = from_timer(tp_vars, t, timer);
1106        struct batadv_tp_unacked *un, *safe;
1107        struct batadv_priv *bat_priv;
1108
1109        bat_priv = tp_vars->bat_priv;
1110
1111        /* if there is recent activity rearm the timer */
1112        if (!batadv_has_timed_out(tp_vars->last_recv_time,
1113                                  BATADV_TP_RECV_TIMEOUT)) {
1114                /* reset the receiver shutdown timer */
1115                batadv_tp_reset_receiver_timer(tp_vars);
1116                return;
1117        }
1118
1119        batadv_dbg(BATADV_DBG_TP_METER, bat_priv,
1120                   "Shutting down for inactivity (more than %dms) from %pM\n",
1121                   BATADV_TP_RECV_TIMEOUT, tp_vars->other_end);
1122
1123        spin_lock_bh(&tp_vars->bat_priv->tp_list_lock);
1124        hlist_del_rcu(&tp_vars->list);
1125        spin_unlock_bh(&tp_vars->bat_priv->tp_list_lock);
1126
1127        /* drop list reference */
1128        batadv_tp_vars_put(tp_vars);
1129
1130        atomic_dec(&bat_priv->tp_num);
1131
1132        spin_lock_bh(&tp_vars->unacked_lock);
1133        list_for_each_entry_safe(un, safe, &tp_vars->unacked_list, list) {
1134                list_del(&un->list);
1135                kfree(un);
1136        }
1137        spin_unlock_bh(&tp_vars->unacked_lock);
1138
1139        /* drop reference of timer */
1140        batadv_tp_vars_put(tp_vars);
1141}
1142
1143/**
1144 * batadv_tp_send_ack() - send an ACK packet
1145 * @bat_priv: the bat priv with all the soft interface information
1146 * @dst: the mac address of the destination originator
1147 * @seq: the sequence number to ACK
1148 * @timestamp: the timestamp to echo back in the ACK
1149 * @session: session identifier
1150 * @socket_index: local ICMP socket identifier
1151 *
1152 * Return: 0 on success, a positive integer representing the reason of the
1153 * failure otherwise
1154 */
1155static int batadv_tp_send_ack(struct batadv_priv *bat_priv, const u8 *dst,
1156                              u32 seq, __be32 timestamp, const u8 *session,
1157                              int socket_index)
1158{
1159        struct batadv_hard_iface *primary_if = NULL;
1160        struct batadv_orig_node *orig_node;
1161        struct batadv_icmp_tp_packet *icmp;
1162        struct sk_buff *skb;
1163        int r, ret;
1164
1165        orig_node = batadv_orig_hash_find(bat_priv, dst);
1166        if (unlikely(!orig_node)) {
1167                ret = BATADV_TP_REASON_DST_UNREACHABLE;
1168                goto out;
1169        }
1170
1171        primary_if = batadv_primary_if_get_selected(bat_priv);
1172        if (unlikely(!primary_if)) {
1173                ret = BATADV_TP_REASON_DST_UNREACHABLE;
1174                goto out;
1175        }
1176
1177        skb = netdev_alloc_skb_ip_align(NULL, sizeof(*icmp) + ETH_HLEN);
1178        if (unlikely(!skb)) {
1179                ret = BATADV_TP_REASON_MEMORY_ERROR;
1180                goto out;
1181        }
1182
1183        skb_reserve(skb, ETH_HLEN);
1184        icmp = skb_put(skb, sizeof(*icmp));
1185        icmp->packet_type = BATADV_ICMP;
1186        icmp->version = BATADV_COMPAT_VERSION;
1187        icmp->ttl = BATADV_TTL;
1188        icmp->msg_type = BATADV_TP;
1189        ether_addr_copy(icmp->dst, orig_node->orig);
1190        ether_addr_copy(icmp->orig, primary_if->net_dev->dev_addr);
1191        icmp->uid = socket_index;
1192
1193        icmp->subtype = BATADV_TP_ACK;
1194        memcpy(icmp->session, session, sizeof(icmp->session));
1195        icmp->seqno = htonl(seq);
1196        icmp->timestamp = timestamp;
1197
1198        /* send the ack */
1199        r = batadv_send_skb_to_orig(skb, orig_node, NULL);
1200        if (unlikely(r < 0) || r == NET_XMIT_DROP) {
1201                ret = BATADV_TP_REASON_DST_UNREACHABLE;
1202                goto out;
1203        }
1204        ret = 0;
1205
1206out:
1207        if (likely(orig_node))
1208                batadv_orig_node_put(orig_node);
1209        if (likely(primary_if))
1210                batadv_hardif_put(primary_if);
1211
1212        return ret;
1213}
1214
1215/**
1216 * batadv_tp_handle_out_of_order() - store an out of order packet
1217 * @tp_vars: the private data of the current TP meter session
1218 * @skb: the buffer containing the received packet
1219 *
1220 * Store the out of order packet in the unacked list for late processing. This
1221 * packets are kept in this list so that they can be ACKed at once as soon as
1222 * all the previous packets have been received
1223 *
1224 * Return: true if the packed has been successfully processed, false otherwise
1225 */
1226static bool batadv_tp_handle_out_of_order(struct batadv_tp_vars *tp_vars,
1227                                          const struct sk_buff *skb)
1228{
1229        const struct batadv_icmp_tp_packet *icmp;
1230        struct batadv_tp_unacked *un, *new;
1231        u32 payload_len;
1232        bool added = false;
1233
1234        new = kmalloc(sizeof(*new), GFP_ATOMIC);
1235        if (unlikely(!new))
1236                return false;
1237
1238        icmp = (struct batadv_icmp_tp_packet *)skb->data;
1239
1240        new->seqno = ntohl(icmp->seqno);
1241        payload_len = skb->len - sizeof(struct batadv_unicast_packet);
1242        new->len = payload_len;
1243
1244        spin_lock_bh(&tp_vars->unacked_lock);
1245        /* if the list is empty immediately attach this new object */
1246        if (list_empty(&tp_vars->unacked_list)) {
1247                list_add(&new->list, &tp_vars->unacked_list);
1248                goto out;
1249        }
1250
1251        /* otherwise loop over the list and either drop the packet because this
1252         * is a duplicate or store it at the right position.
1253         *
1254         * The iteration is done in the reverse way because it is likely that
1255         * the last received packet (the one being processed now) has a bigger
1256         * seqno than all the others already stored.
1257         */
1258        list_for_each_entry_reverse(un, &tp_vars->unacked_list, list) {
1259                /* check for duplicates */
1260                if (new->seqno == un->seqno) {
1261                        if (new->len > un->len)
1262                                un->len = new->len;
1263                        kfree(new);
1264                        added = true;
1265                        break;
1266                }
1267
1268                /* look for the right position */
1269                if (batadv_seq_before(new->seqno, un->seqno))
1270                        continue;
1271
1272                /* as soon as an entry having a bigger seqno is found, the new
1273                 * one is attached _after_ it. In this way the list is kept in
1274                 * ascending order
1275                 */
1276                list_add_tail(&new->list, &un->list);
1277                added = true;
1278                break;
1279        }
1280
1281        /* received packet with smallest seqno out of order; add it to front */
1282        if (!added)
1283                list_add(&new->list, &tp_vars->unacked_list);
1284
1285out:
1286        spin_unlock_bh(&tp_vars->unacked_lock);
1287
1288        return true;
1289}
1290
1291/**
1292 * batadv_tp_ack_unordered() - update number received bytes in current stream
1293 *  without gaps
1294 * @tp_vars: the private data of the current TP meter session
1295 */
1296static void batadv_tp_ack_unordered(struct batadv_tp_vars *tp_vars)
1297{
1298        struct batadv_tp_unacked *un, *safe;
1299        u32 to_ack;
1300
1301        /* go through the unacked packet list and possibly ACK them as
1302         * well
1303         */
1304        spin_lock_bh(&tp_vars->unacked_lock);
1305        list_for_each_entry_safe(un, safe, &tp_vars->unacked_list, list) {
1306                /* the list is ordered, therefore it is possible to stop as soon
1307                 * there is a gap between the last acked seqno and the seqno of
1308                 * the packet under inspection
1309                 */
1310                if (batadv_seq_before(tp_vars->last_recv, un->seqno))
1311                        break;
1312
1313                to_ack = un->seqno + un->len - tp_vars->last_recv;
1314
1315                if (batadv_seq_before(tp_vars->last_recv, un->seqno + un->len))
1316                        tp_vars->last_recv += to_ack;
1317
1318                list_del(&un->list);
1319                kfree(un);
1320        }
1321        spin_unlock_bh(&tp_vars->unacked_lock);
1322}
1323
1324/**
1325 * batadv_tp_init_recv() - return matching or create new receiver tp_vars
1326 * @bat_priv: the bat priv with all the soft interface information
1327 * @icmp: received icmp tp msg
1328 *
1329 * Return: corresponding tp_vars or NULL on errors
1330 */
1331static struct batadv_tp_vars *
1332batadv_tp_init_recv(struct batadv_priv *bat_priv,
1333                    const struct batadv_icmp_tp_packet *icmp)
1334{
1335        struct batadv_tp_vars *tp_vars;
1336
1337        spin_lock_bh(&bat_priv->tp_list_lock);
1338        tp_vars = batadv_tp_list_find_session(bat_priv, icmp->orig,
1339                                              icmp->session);
1340        if (tp_vars)
1341                goto out_unlock;
1342
1343        if (!atomic_add_unless(&bat_priv->tp_num, 1, BATADV_TP_MAX_NUM)) {
1344                batadv_dbg(BATADV_DBG_TP_METER, bat_priv,
1345                           "Meter: too many ongoing sessions, aborting (RECV)\n");
1346                goto out_unlock;
1347        }
1348
1349        tp_vars = kmalloc(sizeof(*tp_vars), GFP_ATOMIC);
1350        if (!tp_vars)
1351                goto out_unlock;
1352
1353        ether_addr_copy(tp_vars->other_end, icmp->orig);
1354        tp_vars->role = BATADV_TP_RECEIVER;
1355        memcpy(tp_vars->session, icmp->session, sizeof(tp_vars->session));
1356        tp_vars->last_recv = BATADV_TP_FIRST_SEQ;
1357        tp_vars->bat_priv = bat_priv;
1358        kref_init(&tp_vars->refcount);
1359
1360        spin_lock_init(&tp_vars->unacked_lock);
1361        INIT_LIST_HEAD(&tp_vars->unacked_list);
1362
1363        kref_get(&tp_vars->refcount);
1364        hlist_add_head_rcu(&tp_vars->list, &bat_priv->tp_list);
1365
1366        kref_get(&tp_vars->refcount);
1367        timer_setup(&tp_vars->timer, batadv_tp_receiver_shutdown, 0);
1368
1369        batadv_tp_reset_receiver_timer(tp_vars);
1370
1371out_unlock:
1372        spin_unlock_bh(&bat_priv->tp_list_lock);
1373
1374        return tp_vars;
1375}
1376
1377/**
1378 * batadv_tp_recv_msg() - process a single data message
1379 * @bat_priv: the bat priv with all the soft interface information
1380 * @skb: the buffer containing the received packet
1381 *
1382 * Process a received TP MSG packet
1383 */
1384static void batadv_tp_recv_msg(struct batadv_priv *bat_priv,
1385                               const struct sk_buff *skb)
1386{
1387        const struct batadv_icmp_tp_packet *icmp;
1388        struct batadv_tp_vars *tp_vars;
1389        size_t packet_size;
1390        u32 seqno;
1391
1392        icmp = (struct batadv_icmp_tp_packet *)skb->data;
1393
1394        seqno = ntohl(icmp->seqno);
1395        /* check if this is the first seqno. This means that if the
1396         * first packet is lost, the tp meter does not work anymore!
1397         */
1398        if (seqno == BATADV_TP_FIRST_SEQ) {
1399                tp_vars = batadv_tp_init_recv(bat_priv, icmp);
1400                if (!tp_vars) {
1401                        batadv_dbg(BATADV_DBG_TP_METER, bat_priv,
1402                                   "Meter: seqno != BATADV_TP_FIRST_SEQ cannot initiate connection\n");
1403                        goto out;
1404                }
1405        } else {
1406                tp_vars = batadv_tp_list_find_session(bat_priv, icmp->orig,
1407                                                      icmp->session);
1408                if (!tp_vars) {
1409                        batadv_dbg(BATADV_DBG_TP_METER, bat_priv,
1410                                   "Unexpected packet from %pM!\n",
1411                                   icmp->orig);
1412                        goto out;
1413                }
1414        }
1415
1416        if (unlikely(tp_vars->role != BATADV_TP_RECEIVER)) {
1417                batadv_dbg(BATADV_DBG_TP_METER, bat_priv,
1418                           "Meter: dropping packet: not expected (role=%u)\n",
1419                           tp_vars->role);
1420                goto out;
1421        }
1422
1423        tp_vars->last_recv_time = jiffies;
1424
1425        /* if the packet is a duplicate, it may be the case that an ACK has been
1426         * lost. Resend the ACK
1427         */
1428        if (batadv_seq_before(seqno, tp_vars->last_recv))
1429                goto send_ack;
1430
1431        /* if the packet is out of order enqueue it */
1432        if (ntohl(icmp->seqno) != tp_vars->last_recv) {
1433                /* exit immediately (and do not send any ACK) if the packet has
1434                 * not been enqueued correctly
1435                 */
1436                if (!batadv_tp_handle_out_of_order(tp_vars, skb))
1437                        goto out;
1438
1439                /* send a duplicate ACK */
1440                goto send_ack;
1441        }
1442
1443        /* if everything was fine count the ACKed bytes */
1444        packet_size = skb->len - sizeof(struct batadv_unicast_packet);
1445        tp_vars->last_recv += packet_size;
1446
1447        /* check if this ordered message filled a gap.... */
1448        batadv_tp_ack_unordered(tp_vars);
1449
1450send_ack:
1451        /* send the ACK. If the received packet was out of order, the ACK that
1452         * is going to be sent is a duplicate (the sender will count them and
1453         * possibly enter Fast Retransmit as soon as it has reached 3)
1454         */
1455        batadv_tp_send_ack(bat_priv, icmp->orig, tp_vars->last_recv,
1456                           icmp->timestamp, icmp->session, icmp->uid);
1457out:
1458        if (likely(tp_vars))
1459                batadv_tp_vars_put(tp_vars);
1460}
1461
1462/**
1463 * batadv_tp_meter_recv() - main TP Meter receiving function
1464 * @bat_priv: the bat priv with all the soft interface information
1465 * @skb: the buffer containing the received packet
1466 */
1467void batadv_tp_meter_recv(struct batadv_priv *bat_priv, struct sk_buff *skb)
1468{
1469        struct batadv_icmp_tp_packet *icmp;
1470
1471        icmp = (struct batadv_icmp_tp_packet *)skb->data;
1472
1473        switch (icmp->subtype) {
1474        case BATADV_TP_MSG:
1475                batadv_tp_recv_msg(bat_priv, skb);
1476                break;
1477        case BATADV_TP_ACK:
1478                batadv_tp_recv_ack(bat_priv, skb);
1479                break;
1480        default:
1481                batadv_dbg(BATADV_DBG_TP_METER, bat_priv,
1482                           "Received unknown TP Metric packet type %u\n",
1483                           icmp->subtype);
1484        }
1485        consume_skb(skb);
1486}
1487
1488/**
1489 * batadv_tp_meter_init() - initialize global tp_meter structures
1490 */
1491void __init batadv_tp_meter_init(void)
1492{
1493        get_random_bytes(batadv_tp_prerandom, sizeof(batadv_tp_prerandom));
1494}
1495