1 // SPDX-License-Identifier: GPL-2.0 OR BSD-3-Clause 2 3 /* COMMON Applications Kept Enhanced (CAKE) discipline 4 * 5 * Copyright (C) 2014-2018 Jonathan Morton <chromatix99@gmail.com> 6 * Copyright (C) 2015-2018 Toke Høiland-Jørgensen <toke@toke.dk> 7 * Copyright (C) 2014-2018 Dave Täht <dave.taht@gmail.com> 8 * Copyright (C) 2015-2018 Sebastian Moeller <moeller0@gmx.de> 9 * (C) 2015-2018 Kevin Darbyshire-Bryant <kevin@darbyshire-bryant.me.uk> 10 * Copyright (C) 2017-2018 Ryan Mounce <ryan@mounce.com.au> 11 * 12 * The CAKE Principles: 13 * (or, how to have your cake and eat it too) 14 * 15 * This is a combination of several shaping, AQM and FQ techniques into one 16 * easy-to-use package: 17 * 18 * - An overall bandwidth shaper, to move the bottleneck away from dumb CPE 19 * equipment and bloated MACs. This operates in deficit mode (as in sch_fq), 20 * eliminating the need for any sort of burst parameter (eg. token bucket 21 * depth). Burst support is limited to that necessary to overcome scheduling 22 * latency. 23 * 24 * - A Diffserv-aware priority queue, giving more priority to certain classes, 25 * up to a specified fraction of bandwidth. Above that bandwidth threshold, 26 * the priority is reduced to avoid starving other tins. 27 * 28 * - Each priority tin has a separate Flow Queue system, to isolate traffic 29 * flows from each other. This prevents a burst on one flow from increasing 30 * the delay to another. Flows are distributed to queues using a 31 * set-associative hash function. 32 * 33 * - Each queue is actively managed by Cobalt, which is a combination of the 34 * Codel and Blue AQM algorithms. This serves flows fairly, and signals 35 * congestion early via ECN (if available) and/or packet drops, to keep 36 * latency low. The codel parameters are auto-tuned based on the bandwidth 37 * setting, as is necessary at low bandwidths. 38 * 39 * The configuration parameters are kept deliberately simple for ease of use. 40 * Everything has sane defaults. Complete generality of configuration is *not* 41 * a goal. 42 * 43 * The priority queue operates according to a weighted DRR scheme, combined with 44 * a bandwidth tracker which reuses the shaper logic to detect which side of the 45 * bandwidth sharing threshold the tin is operating. This determines whether a 46 * priority-based weight (high) or a bandwidth-based weight (low) is used for 47 * that tin in the current pass. 48 * 49 * This qdisc was inspired by Eric Dumazet's fq_codel code, which he kindly 50 * granted us permission to leverage. 51 */ 52 53 #include <linux/module.h> 54 #include <linux/types.h> 55 #include <linux/kernel.h> 56 #include <linux/jiffies.h> 57 #include <linux/string.h> 58 #include <linux/in.h> 59 #include <linux/errno.h> 60 #include <linux/init.h> 61 #include <linux/skbuff.h> 62 #include <linux/jhash.h> 63 #include <linux/slab.h> 64 #include <linux/vmalloc.h> 65 #include <linux/reciprocal_div.h> 66 #include <net/netlink.h> 67 #include <linux/if_vlan.h> 68 #include <net/pkt_sched.h> 69 #include <net/pkt_cls.h> 70 #include <net/tcp.h> 71 #include <net/flow_dissector.h> 72 73 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 74 #include <net/netfilter/nf_conntrack_core.h> 75 #endif 76 77 #define CAKE_SET_WAYS (8) 78 #define CAKE_MAX_TINS (8) 79 #define CAKE_QUEUES (1024) 80 #define CAKE_FLOW_MASK 63 81 #define CAKE_FLOW_NAT_FLAG 64 82 83 /* struct cobalt_params - contains codel and blue parameters 84 * @interval: codel initial drop rate 85 * @target: maximum persistent sojourn time & blue update rate 86 * @mtu_time: serialisation delay of maximum-size packet 87 * @p_inc: increment of blue drop probability (0.32 fxp) 88 * @p_dec: decrement of blue drop probability (0.32 fxp) 89 */ 90 struct cobalt_params { 91 u64 interval; 92 u64 target; 93 u64 mtu_time; 94 u32 p_inc; 95 u32 p_dec; 96 }; 97 98 /* struct cobalt_vars - contains codel and blue variables 99 * @count: codel dropping frequency 100 * @rec_inv_sqrt: reciprocal value of sqrt(count) >> 1 101 * @drop_next: time to drop next packet, or when we dropped last 102 * @blue_timer: Blue time to next drop 103 * @p_drop: BLUE drop probability (0.32 fxp) 104 * @dropping: set if in dropping state 105 * @ecn_marked: set if marked 106 */ 107 struct cobalt_vars { 108 u32 count; 109 u32 rec_inv_sqrt; 110 ktime_t drop_next; 111 ktime_t blue_timer; 112 u32 p_drop; 113 bool dropping; 114 bool ecn_marked; 115 }; 116 117 enum { 118 CAKE_SET_NONE = 0, 119 CAKE_SET_SPARSE, 120 CAKE_SET_SPARSE_WAIT, /* counted in SPARSE, actually in BULK */ 121 CAKE_SET_BULK, 122 CAKE_SET_DECAYING 123 }; 124 125 struct cake_flow { 126 /* this stuff is all needed per-flow at dequeue time */ 127 struct sk_buff *head; 128 struct sk_buff *tail; 129 struct list_head flowchain; 130 s32 deficit; 131 u32 dropped; 132 struct cobalt_vars cvars; 133 u16 srchost; /* index into cake_host table */ 134 u16 dsthost; 135 u8 set; 136 }; /* please try to keep this structure <= 64 bytes */ 137 138 struct cake_host { 139 u32 srchost_tag; 140 u32 dsthost_tag; 141 u16 srchost_bulk_flow_count; 142 u16 dsthost_bulk_flow_count; 143 }; 144 145 struct cake_heap_entry { 146 u16 t:3, b:10; 147 }; 148 149 struct cake_tin_data { 150 struct cake_flow flows[CAKE_QUEUES]; 151 u32 backlogs[CAKE_QUEUES]; 152 u32 tags[CAKE_QUEUES]; /* for set association */ 153 u16 overflow_idx[CAKE_QUEUES]; 154 struct cake_host hosts[CAKE_QUEUES]; /* for triple isolation */ 155 u16 flow_quantum; 156 157 struct cobalt_params cparams; 158 u32 drop_overlimit; 159 u16 bulk_flow_count; 160 u16 sparse_flow_count; 161 u16 decaying_flow_count; 162 u16 unresponsive_flow_count; 163 164 u32 max_skblen; 165 166 struct list_head new_flows; 167 struct list_head old_flows; 168 struct list_head decaying_flows; 169 170 /* time_next = time_this + ((len * rate_ns) >> rate_shft) */ 171 ktime_t time_next_packet; 172 u64 tin_rate_ns; 173 u64 tin_rate_bps; 174 u16 tin_rate_shft; 175 176 u16 tin_quantum; 177 s32 tin_deficit; 178 u32 tin_backlog; 179 u32 tin_dropped; 180 u32 tin_ecn_mark; 181 182 u32 packets; 183 u64 bytes; 184 185 u32 ack_drops; 186 187 /* moving averages */ 188 u64 avge_delay; 189 u64 peak_delay; 190 u64 base_delay; 191 192 /* hash function stats */ 193 u32 way_directs; 194 u32 way_hits; 195 u32 way_misses; 196 u32 way_collisions; 197 }; /* number of tins is small, so size of this struct doesn't matter much */ 198 199 struct cake_sched_data { 200 struct tcf_proto __rcu *filter_list; /* optional external classifier */ 201 struct tcf_block *block; 202 struct cake_tin_data *tins; 203 204 struct cake_heap_entry overflow_heap[CAKE_QUEUES * CAKE_MAX_TINS]; 205 u16 overflow_timeout; 206 207 u16 tin_cnt; 208 u8 tin_mode; 209 u8 flow_mode; 210 u8 ack_filter; 211 u8 atm_mode; 212 213 u32 fwmark_mask; 214 u16 fwmark_shft; 215 216 /* time_next = time_this + ((len * rate_ns) >> rate_shft) */ 217 u16 rate_shft; 218 ktime_t time_next_packet; 219 ktime_t failsafe_next_packet; 220 u64 rate_ns; 221 u64 rate_bps; 222 u16 rate_flags; 223 s16 rate_overhead; 224 u16 rate_mpu; 225 u64 interval; 226 u64 target; 227 228 /* resource tracking */ 229 u32 buffer_used; 230 u32 buffer_max_used; 231 u32 buffer_limit; 232 u32 buffer_config_limit; 233 234 /* indices for dequeue */ 235 u16 cur_tin; 236 u16 cur_flow; 237 238 struct qdisc_watchdog watchdog; 239 const u8 *tin_index; 240 const u8 *tin_order; 241 242 /* bandwidth capacity estimate */ 243 ktime_t last_packet_time; 244 ktime_t avg_window_begin; 245 u64 avg_packet_interval; 246 u64 avg_window_bytes; 247 u64 avg_peak_bandwidth; 248 ktime_t last_reconfig_time; 249 250 /* packet length stats */ 251 u32 avg_netoff; 252 u16 max_netlen; 253 u16 max_adjlen; 254 u16 min_netlen; 255 u16 min_adjlen; 256 }; 257 258 enum { 259 CAKE_FLAG_OVERHEAD = BIT(0), 260 CAKE_FLAG_AUTORATE_INGRESS = BIT(1), 261 CAKE_FLAG_INGRESS = BIT(2), 262 CAKE_FLAG_WASH = BIT(3), 263 CAKE_FLAG_SPLIT_GSO = BIT(4) 264 }; 265 266 /* COBALT operates the Codel and BLUE algorithms in parallel, in order to 267 * obtain the best features of each. Codel is excellent on flows which 268 * respond to congestion signals in a TCP-like way. BLUE is more effective on 269 * unresponsive flows. 270 */ 271 272 struct cobalt_skb_cb { 273 ktime_t enqueue_time; 274 u32 adjusted_len; 275 }; 276 277 static u64 us_to_ns(u64 us) 278 { 279 return us * NSEC_PER_USEC; 280 } 281 282 static struct cobalt_skb_cb *get_cobalt_cb(const struct sk_buff *skb) 283 { 284 qdisc_cb_private_validate(skb, sizeof(struct cobalt_skb_cb)); 285 return (struct cobalt_skb_cb *)qdisc_skb_cb(skb)->data; 286 } 287 288 static ktime_t cobalt_get_enqueue_time(const struct sk_buff *skb) 289 { 290 return get_cobalt_cb(skb)->enqueue_time; 291 } 292 293 static void cobalt_set_enqueue_time(struct sk_buff *skb, 294 ktime_t now) 295 { 296 get_cobalt_cb(skb)->enqueue_time = now; 297 } 298 299 static u16 quantum_div[CAKE_QUEUES + 1] = {0}; 300 301 /* Diffserv lookup tables */ 302 303 static const u8 precedence[] = { 304 0, 0, 0, 0, 0, 0, 0, 0, 305 1, 1, 1, 1, 1, 1, 1, 1, 306 2, 2, 2, 2, 2, 2, 2, 2, 307 3, 3, 3, 3, 3, 3, 3, 3, 308 4, 4, 4, 4, 4, 4, 4, 4, 309 5, 5, 5, 5, 5, 5, 5, 5, 310 6, 6, 6, 6, 6, 6, 6, 6, 311 7, 7, 7, 7, 7, 7, 7, 7, 312 }; 313 314 static const u8 diffserv8[] = { 315 2, 0, 1, 2, 4, 2, 2, 2, 316 1, 2, 1, 2, 1, 2, 1, 2, 317 5, 2, 4, 2, 4, 2, 4, 2, 318 3, 2, 3, 2, 3, 2, 3, 2, 319 6, 2, 3, 2, 3, 2, 3, 2, 320 6, 2, 2, 2, 6, 2, 6, 2, 321 7, 2, 2, 2, 2, 2, 2, 2, 322 7, 2, 2, 2, 2, 2, 2, 2, 323 }; 324 325 static const u8 diffserv4[] = { 326 0, 1, 0, 0, 2, 0, 0, 0, 327 1, 0, 0, 0, 0, 0, 0, 0, 328 2, 0, 2, 0, 2, 0, 2, 0, 329 2, 0, 2, 0, 2, 0, 2, 0, 330 3, 0, 2, 0, 2, 0, 2, 0, 331 3, 0, 0, 0, 3, 0, 3, 0, 332 3, 0, 0, 0, 0, 0, 0, 0, 333 3, 0, 0, 0, 0, 0, 0, 0, 334 }; 335 336 static const u8 diffserv3[] = { 337 0, 1, 0, 0, 2, 0, 0, 0, 338 1, 0, 0, 0, 0, 0, 0, 0, 339 0, 0, 0, 0, 0, 0, 0, 0, 340 0, 0, 0, 0, 0, 0, 0, 0, 341 0, 0, 0, 0, 0, 0, 0, 0, 342 0, 0, 0, 0, 2, 0, 2, 0, 343 2, 0, 0, 0, 0, 0, 0, 0, 344 2, 0, 0, 0, 0, 0, 0, 0, 345 }; 346 347 static const u8 besteffort[] = { 348 0, 0, 0, 0, 0, 0, 0, 0, 349 0, 0, 0, 0, 0, 0, 0, 0, 350 0, 0, 0, 0, 0, 0, 0, 0, 351 0, 0, 0, 0, 0, 0, 0, 0, 352 0, 0, 0, 0, 0, 0, 0, 0, 353 0, 0, 0, 0, 0, 0, 0, 0, 354 0, 0, 0, 0, 0, 0, 0, 0, 355 0, 0, 0, 0, 0, 0, 0, 0, 356 }; 357 358 /* tin priority order for stats dumping */ 359 360 static const u8 normal_order[] = {0, 1, 2, 3, 4, 5, 6, 7}; 361 static const u8 bulk_order[] = {1, 0, 2, 3}; 362 363 #define REC_INV_SQRT_CACHE (16) 364 static u32 cobalt_rec_inv_sqrt_cache[REC_INV_SQRT_CACHE] = {0}; 365 366 /* http://en.wikipedia.org/wiki/Methods_of_computing_square_roots 367 * new_invsqrt = (invsqrt / 2) * (3 - count * invsqrt^2) 368 * 369 * Here, invsqrt is a fixed point number (< 1.0), 32bit mantissa, aka Q0.32 370 */ 371 372 static void cobalt_newton_step(struct cobalt_vars *vars) 373 { 374 u32 invsqrt, invsqrt2; 375 u64 val; 376 377 invsqrt = vars->rec_inv_sqrt; 378 invsqrt2 = ((u64)invsqrt * invsqrt) >> 32; 379 val = (3LL << 32) - ((u64)vars->count * invsqrt2); 380 381 val >>= 2; /* avoid overflow in following multiply */ 382 val = (val * invsqrt) >> (32 - 2 + 1); 383 384 vars->rec_inv_sqrt = val; 385 } 386 387 static void cobalt_invsqrt(struct cobalt_vars *vars) 388 { 389 if (vars->count < REC_INV_SQRT_CACHE) 390 vars->rec_inv_sqrt = cobalt_rec_inv_sqrt_cache[vars->count]; 391 else 392 cobalt_newton_step(vars); 393 } 394 395 /* There is a big difference in timing between the accurate values placed in 396 * the cache and the approximations given by a single Newton step for small 397 * count values, particularly when stepping from count 1 to 2 or vice versa. 398 * Above 16, a single Newton step gives sufficient accuracy in either 399 * direction, given the precision stored. 400 * 401 * The magnitude of the error when stepping up to count 2 is such as to give 402 * the value that *should* have been produced at count 4. 403 */ 404 405 static void cobalt_cache_init(void) 406 { 407 struct cobalt_vars v; 408 409 memset(&v, 0, sizeof(v)); 410 v.rec_inv_sqrt = ~0U; 411 cobalt_rec_inv_sqrt_cache[0] = v.rec_inv_sqrt; 412 413 for (v.count = 1; v.count < REC_INV_SQRT_CACHE; v.count++) { 414 cobalt_newton_step(&v); 415 cobalt_newton_step(&v); 416 cobalt_newton_step(&v); 417 cobalt_newton_step(&v); 418 419 cobalt_rec_inv_sqrt_cache[v.count] = v.rec_inv_sqrt; 420 } 421 } 422 423 static void cobalt_vars_init(struct cobalt_vars *vars) 424 { 425 memset(vars, 0, sizeof(*vars)); 426 427 if (!cobalt_rec_inv_sqrt_cache[0]) { 428 cobalt_cache_init(); 429 cobalt_rec_inv_sqrt_cache[0] = ~0; 430 } 431 } 432 433 /* CoDel control_law is t + interval/sqrt(count) 434 * We maintain in rec_inv_sqrt the reciprocal value of sqrt(count) to avoid 435 * both sqrt() and divide operation. 436 */ 437 static ktime_t cobalt_control(ktime_t t, 438 u64 interval, 439 u32 rec_inv_sqrt) 440 { 441 return ktime_add_ns(t, reciprocal_scale(interval, 442 rec_inv_sqrt)); 443 } 444 445 /* Call this when a packet had to be dropped due to queue overflow. Returns 446 * true if the BLUE state was quiescent before but active after this call. 447 */ 448 static bool cobalt_queue_full(struct cobalt_vars *vars, 449 struct cobalt_params *p, 450 ktime_t now) 451 { 452 bool up = false; 453 454 if (ktime_to_ns(ktime_sub(now, vars->blue_timer)) > p->target) { 455 up = !vars->p_drop; 456 vars->p_drop += p->p_inc; 457 if (vars->p_drop < p->p_inc) 458 vars->p_drop = ~0; 459 vars->blue_timer = now; 460 } 461 vars->dropping = true; 462 vars->drop_next = now; 463 if (!vars->count) 464 vars->count = 1; 465 466 return up; 467 } 468 469 /* Call this when the queue was serviced but turned out to be empty. Returns 470 * true if the BLUE state was active before but quiescent after this call. 471 */ 472 static bool cobalt_queue_empty(struct cobalt_vars *vars, 473 struct cobalt_params *p, 474 ktime_t now) 475 { 476 bool down = false; 477 478 if (vars->p_drop && 479 ktime_to_ns(ktime_sub(now, vars->blue_timer)) > p->target) { 480 if (vars->p_drop < p->p_dec) 481 vars->p_drop = 0; 482 else 483 vars->p_drop -= p->p_dec; 484 vars->blue_timer = now; 485 down = !vars->p_drop; 486 } 487 vars->dropping = false; 488 489 if (vars->count && ktime_to_ns(ktime_sub(now, vars->drop_next)) >= 0) { 490 vars->count--; 491 cobalt_invsqrt(vars); 492 vars->drop_next = cobalt_control(vars->drop_next, 493 p->interval, 494 vars->rec_inv_sqrt); 495 } 496 497 return down; 498 } 499 500 /* Call this with a freshly dequeued packet for possible congestion marking. 501 * Returns true as an instruction to drop the packet, false for delivery. 502 */ 503 static bool cobalt_should_drop(struct cobalt_vars *vars, 504 struct cobalt_params *p, 505 ktime_t now, 506 struct sk_buff *skb, 507 u32 bulk_flows) 508 { 509 bool next_due, over_target, drop = false; 510 ktime_t schedule; 511 u64 sojourn; 512 513 /* The 'schedule' variable records, in its sign, whether 'now' is before or 514 * after 'drop_next'. This allows 'drop_next' to be updated before the next 515 * scheduling decision is actually branched, without destroying that 516 * information. Similarly, the first 'schedule' value calculated is preserved 517 * in the boolean 'next_due'. 518 * 519 * As for 'drop_next', we take advantage of the fact that 'interval' is both 520 * the delay between first exceeding 'target' and the first signalling event, 521 * *and* the scaling factor for the signalling frequency. It's therefore very 522 * natural to use a single mechanism for both purposes, and eliminates a 523 * significant amount of reference Codel's spaghetti code. To help with this, 524 * both the '0' and '1' entries in the invsqrt cache are 0xFFFFFFFF, as close 525 * as possible to 1.0 in fixed-point. 526 */ 527 528 sojourn = ktime_to_ns(ktime_sub(now, cobalt_get_enqueue_time(skb))); 529 schedule = ktime_sub(now, vars->drop_next); 530 over_target = sojourn > p->target && 531 sojourn > p->mtu_time * bulk_flows * 2 && 532 sojourn > p->mtu_time * 4; 533 next_due = vars->count && ktime_to_ns(schedule) >= 0; 534 535 vars->ecn_marked = false; 536 537 if (over_target) { 538 if (!vars->dropping) { 539 vars->dropping = true; 540 vars->drop_next = cobalt_control(now, 541 p->interval, 542 vars->rec_inv_sqrt); 543 } 544 if (!vars->count) 545 vars->count = 1; 546 } else if (vars->dropping) { 547 vars->dropping = false; 548 } 549 550 if (next_due && vars->dropping) { 551 /* Use ECN mark if possible, otherwise drop */ 552 drop = !(vars->ecn_marked = INET_ECN_set_ce(skb)); 553 554 vars->count++; 555 if (!vars->count) 556 vars->count--; 557 cobalt_invsqrt(vars); 558 vars->drop_next = cobalt_control(vars->drop_next, 559 p->interval, 560 vars->rec_inv_sqrt); 561 schedule = ktime_sub(now, vars->drop_next); 562 } else { 563 while (next_due) { 564 vars->count--; 565 cobalt_invsqrt(vars); 566 vars->drop_next = cobalt_control(vars->drop_next, 567 p->interval, 568 vars->rec_inv_sqrt); 569 schedule = ktime_sub(now, vars->drop_next); 570 next_due = vars->count && ktime_to_ns(schedule) >= 0; 571 } 572 } 573 574 /* Simple BLUE implementation. Lack of ECN is deliberate. */ 575 if (vars->p_drop) 576 drop |= (prandom_u32() < vars->p_drop); 577 578 /* Overload the drop_next field as an activity timeout */ 579 if (!vars->count) 580 vars->drop_next = ktime_add_ns(now, p->interval); 581 else if (ktime_to_ns(schedule) > 0 && !drop) 582 vars->drop_next = now; 583 584 return drop; 585 } 586 587 static bool cake_update_flowkeys(struct flow_keys *keys, 588 const struct sk_buff *skb) 589 { 590 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 591 struct nf_conntrack_tuple tuple = {}; 592 bool rev = !skb->_nfct, upd = false; 593 __be32 ip; 594 595 if (skb_protocol(skb, true) != htons(ETH_P_IP)) 596 return false; 597 598 if (!nf_ct_get_tuple_skb(&tuple, skb)) 599 return false; 600 601 ip = rev ? tuple.dst.u3.ip : tuple.src.u3.ip; 602 if (ip != keys->addrs.v4addrs.src) { 603 keys->addrs.v4addrs.src = ip; 604 upd = true; 605 } 606 ip = rev ? tuple.src.u3.ip : tuple.dst.u3.ip; 607 if (ip != keys->addrs.v4addrs.dst) { 608 keys->addrs.v4addrs.dst = ip; 609 upd = true; 610 } 611 612 if (keys->ports.ports) { 613 __be16 port; 614 615 port = rev ? tuple.dst.u.all : tuple.src.u.all; 616 if (port != keys->ports.src) { 617 keys->ports.src = port; 618 upd = true; 619 } 620 port = rev ? tuple.src.u.all : tuple.dst.u.all; 621 if (port != keys->ports.dst) { 622 port = keys->ports.dst; 623 upd = true; 624 } 625 } 626 return upd; 627 #else 628 return false; 629 #endif 630 } 631 632 /* Cake has several subtle multiple bit settings. In these cases you 633 * would be matching triple isolate mode as well. 634 */ 635 636 static bool cake_dsrc(int flow_mode) 637 { 638 return (flow_mode & CAKE_FLOW_DUAL_SRC) == CAKE_FLOW_DUAL_SRC; 639 } 640 641 static bool cake_ddst(int flow_mode) 642 { 643 return (flow_mode & CAKE_FLOW_DUAL_DST) == CAKE_FLOW_DUAL_DST; 644 } 645 646 static u32 cake_hash(struct cake_tin_data *q, const struct sk_buff *skb, 647 int flow_mode, u16 flow_override, u16 host_override) 648 { 649 bool hash_flows = (!flow_override && !!(flow_mode & CAKE_FLOW_FLOWS)); 650 bool hash_hosts = (!host_override && !!(flow_mode & CAKE_FLOW_HOSTS)); 651 bool nat_enabled = !!(flow_mode & CAKE_FLOW_NAT_FLAG); 652 u32 flow_hash = 0, srchost_hash = 0, dsthost_hash = 0; 653 u16 reduced_hash, srchost_idx, dsthost_idx; 654 struct flow_keys keys, host_keys; 655 bool use_skbhash = skb->l4_hash; 656 657 if (unlikely(flow_mode == CAKE_FLOW_NONE)) 658 return 0; 659 660 /* If both overrides are set, or we can use the SKB hash and nat mode is 661 * disabled, we can skip packet dissection entirely. If nat mode is 662 * enabled there's another check below after doing the conntrack lookup. 663 */ 664 if ((!hash_flows || (use_skbhash && !nat_enabled)) && !hash_hosts) 665 goto skip_hash; 666 667 skb_flow_dissect_flow_keys(skb, &keys, 668 FLOW_DISSECTOR_F_STOP_AT_FLOW_LABEL); 669 670 /* Don't use the SKB hash if we change the lookup keys from conntrack */ 671 if (nat_enabled && cake_update_flowkeys(&keys, skb)) 672 use_skbhash = false; 673 674 /* If we can still use the SKB hash and don't need the host hash, we can 675 * skip the rest of the hashing procedure 676 */ 677 if (use_skbhash && !hash_hosts) 678 goto skip_hash; 679 680 /* flow_hash_from_keys() sorts the addresses by value, so we have 681 * to preserve their order in a separate data structure to treat 682 * src and dst host addresses as independently selectable. 683 */ 684 host_keys = keys; 685 host_keys.ports.ports = 0; 686 host_keys.basic.ip_proto = 0; 687 host_keys.keyid.keyid = 0; 688 host_keys.tags.flow_label = 0; 689 690 switch (host_keys.control.addr_type) { 691 case FLOW_DISSECTOR_KEY_IPV4_ADDRS: 692 host_keys.addrs.v4addrs.src = 0; 693 dsthost_hash = flow_hash_from_keys(&host_keys); 694 host_keys.addrs.v4addrs.src = keys.addrs.v4addrs.src; 695 host_keys.addrs.v4addrs.dst = 0; 696 srchost_hash = flow_hash_from_keys(&host_keys); 697 break; 698 699 case FLOW_DISSECTOR_KEY_IPV6_ADDRS: 700 memset(&host_keys.addrs.v6addrs.src, 0, 701 sizeof(host_keys.addrs.v6addrs.src)); 702 dsthost_hash = flow_hash_from_keys(&host_keys); 703 host_keys.addrs.v6addrs.src = keys.addrs.v6addrs.src; 704 memset(&host_keys.addrs.v6addrs.dst, 0, 705 sizeof(host_keys.addrs.v6addrs.dst)); 706 srchost_hash = flow_hash_from_keys(&host_keys); 707 break; 708 709 default: 710 dsthost_hash = 0; 711 srchost_hash = 0; 712 } 713 714 /* This *must* be after the above switch, since as a 715 * side-effect it sorts the src and dst addresses. 716 */ 717 if (hash_flows && !use_skbhash) 718 flow_hash = flow_hash_from_keys(&keys); 719 720 skip_hash: 721 if (flow_override) 722 flow_hash = flow_override - 1; 723 else if (use_skbhash) 724 flow_hash = skb->hash; 725 if (host_override) { 726 dsthost_hash = host_override - 1; 727 srchost_hash = host_override - 1; 728 } 729 730 if (!(flow_mode & CAKE_FLOW_FLOWS)) { 731 if (flow_mode & CAKE_FLOW_SRC_IP) 732 flow_hash ^= srchost_hash; 733 734 if (flow_mode & CAKE_FLOW_DST_IP) 735 flow_hash ^= dsthost_hash; 736 } 737 738 reduced_hash = flow_hash % CAKE_QUEUES; 739 740 /* set-associative hashing */ 741 /* fast path if no hash collision (direct lookup succeeds) */ 742 if (likely(q->tags[reduced_hash] == flow_hash && 743 q->flows[reduced_hash].set)) { 744 q->way_directs++; 745 } else { 746 u32 inner_hash = reduced_hash % CAKE_SET_WAYS; 747 u32 outer_hash = reduced_hash - inner_hash; 748 bool allocate_src = false; 749 bool allocate_dst = false; 750 u32 i, k; 751 752 /* check if any active queue in the set is reserved for 753 * this flow. 754 */ 755 for (i = 0, k = inner_hash; i < CAKE_SET_WAYS; 756 i++, k = (k + 1) % CAKE_SET_WAYS) { 757 if (q->tags[outer_hash + k] == flow_hash) { 758 if (i) 759 q->way_hits++; 760 761 if (!q->flows[outer_hash + k].set) { 762 /* need to increment host refcnts */ 763 allocate_src = cake_dsrc(flow_mode); 764 allocate_dst = cake_ddst(flow_mode); 765 } 766 767 goto found; 768 } 769 } 770 771 /* no queue is reserved for this flow, look for an 772 * empty one. 773 */ 774 for (i = 0; i < CAKE_SET_WAYS; 775 i++, k = (k + 1) % CAKE_SET_WAYS) { 776 if (!q->flows[outer_hash + k].set) { 777 q->way_misses++; 778 allocate_src = cake_dsrc(flow_mode); 779 allocate_dst = cake_ddst(flow_mode); 780 goto found; 781 } 782 } 783 784 /* With no empty queues, default to the original 785 * queue, accept the collision, update the host tags. 786 */ 787 q->way_collisions++; 788 if (q->flows[outer_hash + k].set == CAKE_SET_BULK) { 789 q->hosts[q->flows[reduced_hash].srchost].srchost_bulk_flow_count--; 790 q->hosts[q->flows[reduced_hash].dsthost].dsthost_bulk_flow_count--; 791 } 792 allocate_src = cake_dsrc(flow_mode); 793 allocate_dst = cake_ddst(flow_mode); 794 found: 795 /* reserve queue for future packets in same flow */ 796 reduced_hash = outer_hash + k; 797 q->tags[reduced_hash] = flow_hash; 798 799 if (allocate_src) { 800 srchost_idx = srchost_hash % CAKE_QUEUES; 801 inner_hash = srchost_idx % CAKE_SET_WAYS; 802 outer_hash = srchost_idx - inner_hash; 803 for (i = 0, k = inner_hash; i < CAKE_SET_WAYS; 804 i++, k = (k + 1) % CAKE_SET_WAYS) { 805 if (q->hosts[outer_hash + k].srchost_tag == 806 srchost_hash) 807 goto found_src; 808 } 809 for (i = 0; i < CAKE_SET_WAYS; 810 i++, k = (k + 1) % CAKE_SET_WAYS) { 811 if (!q->hosts[outer_hash + k].srchost_bulk_flow_count) 812 break; 813 } 814 q->hosts[outer_hash + k].srchost_tag = srchost_hash; 815 found_src: 816 srchost_idx = outer_hash + k; 817 if (q->flows[reduced_hash].set == CAKE_SET_BULK) 818 q->hosts[srchost_idx].srchost_bulk_flow_count++; 819 q->flows[reduced_hash].srchost = srchost_idx; 820 } 821 822 if (allocate_dst) { 823 dsthost_idx = dsthost_hash % CAKE_QUEUES; 824 inner_hash = dsthost_idx % CAKE_SET_WAYS; 825 outer_hash = dsthost_idx - inner_hash; 826 for (i = 0, k = inner_hash; i < CAKE_SET_WAYS; 827 i++, k = (k + 1) % CAKE_SET_WAYS) { 828 if (q->hosts[outer_hash + k].dsthost_tag == 829 dsthost_hash) 830 goto found_dst; 831 } 832 for (i = 0; i < CAKE_SET_WAYS; 833 i++, k = (k + 1) % CAKE_SET_WAYS) { 834 if (!q->hosts[outer_hash + k].dsthost_bulk_flow_count) 835 break; 836 } 837 q->hosts[outer_hash + k].dsthost_tag = dsthost_hash; 838 found_dst: 839 dsthost_idx = outer_hash + k; 840 if (q->flows[reduced_hash].set == CAKE_SET_BULK) 841 q->hosts[dsthost_idx].dsthost_bulk_flow_count++; 842 q->flows[reduced_hash].dsthost = dsthost_idx; 843 } 844 } 845 846 return reduced_hash; 847 } 848 849 /* helper functions : might be changed when/if skb use a standard list_head */ 850 /* remove one skb from head of slot queue */ 851 852 static struct sk_buff *dequeue_head(struct cake_flow *flow) 853 { 854 struct sk_buff *skb = flow->head; 855 856 if (skb) { 857 flow->head = skb->next; 858 skb_mark_not_on_list(skb); 859 } 860 861 return skb; 862 } 863 864 /* add skb to flow queue (tail add) */ 865 866 static void flow_queue_add(struct cake_flow *flow, struct sk_buff *skb) 867 { 868 if (!flow->head) 869 flow->head = skb; 870 else 871 flow->tail->next = skb; 872 flow->tail = skb; 873 skb->next = NULL; 874 } 875 876 static struct iphdr *cake_get_iphdr(const struct sk_buff *skb, 877 struct ipv6hdr *buf) 878 { 879 unsigned int offset = skb_network_offset(skb); 880 struct iphdr *iph; 881 882 iph = skb_header_pointer(skb, offset, sizeof(struct iphdr), buf); 883 884 if (!iph) 885 return NULL; 886 887 if (iph->version == 4 && iph->protocol == IPPROTO_IPV6) 888 return skb_header_pointer(skb, offset + iph->ihl * 4, 889 sizeof(struct ipv6hdr), buf); 890 891 else if (iph->version == 4) 892 return iph; 893 894 else if (iph->version == 6) 895 return skb_header_pointer(skb, offset, sizeof(struct ipv6hdr), 896 buf); 897 898 return NULL; 899 } 900 901 static struct tcphdr *cake_get_tcphdr(const struct sk_buff *skb, 902 void *buf, unsigned int bufsize) 903 { 904 unsigned int offset = skb_network_offset(skb); 905 const struct ipv6hdr *ipv6h; 906 const struct tcphdr *tcph; 907 const struct iphdr *iph; 908 struct ipv6hdr _ipv6h; 909 struct tcphdr _tcph; 910 911 ipv6h = skb_header_pointer(skb, offset, sizeof(_ipv6h), &_ipv6h); 912 913 if (!ipv6h) 914 return NULL; 915 916 if (ipv6h->version == 4) { 917 iph = (struct iphdr *)ipv6h; 918 offset += iph->ihl * 4; 919 920 /* special-case 6in4 tunnelling, as that is a common way to get 921 * v6 connectivity in the home 922 */ 923 if (iph->protocol == IPPROTO_IPV6) { 924 ipv6h = skb_header_pointer(skb, offset, 925 sizeof(_ipv6h), &_ipv6h); 926 927 if (!ipv6h || ipv6h->nexthdr != IPPROTO_TCP) 928 return NULL; 929 930 offset += sizeof(struct ipv6hdr); 931 932 } else if (iph->protocol != IPPROTO_TCP) { 933 return NULL; 934 } 935 936 } else if (ipv6h->version == 6) { 937 if (ipv6h->nexthdr != IPPROTO_TCP) 938 return NULL; 939 940 offset += sizeof(struct ipv6hdr); 941 } else { 942 return NULL; 943 } 944 945 tcph = skb_header_pointer(skb, offset, sizeof(_tcph), &_tcph); 946 if (!tcph || tcph->doff < 5) 947 return NULL; 948 949 return skb_header_pointer(skb, offset, 950 min(__tcp_hdrlen(tcph), bufsize), buf); 951 } 952 953 static const void *cake_get_tcpopt(const struct tcphdr *tcph, 954 int code, int *oplen) 955 { 956 /* inspired by tcp_parse_options in tcp_input.c */ 957 int length = __tcp_hdrlen(tcph) - sizeof(struct tcphdr); 958 const u8 *ptr = (const u8 *)(tcph + 1); 959 960 while (length > 0) { 961 int opcode = *ptr++; 962 int opsize; 963 964 if (opcode == TCPOPT_EOL) 965 break; 966 if (opcode == TCPOPT_NOP) { 967 length--; 968 continue; 969 } 970 if (length < 2) 971 break; 972 opsize = *ptr++; 973 if (opsize < 2 || opsize > length) 974 break; 975 976 if (opcode == code) { 977 *oplen = opsize; 978 return ptr; 979 } 980 981 ptr += opsize - 2; 982 length -= opsize; 983 } 984 985 return NULL; 986 } 987 988 /* Compare two SACK sequences. A sequence is considered greater if it SACKs more 989 * bytes than the other. In the case where both sequences ACKs bytes that the 990 * other doesn't, A is considered greater. DSACKs in A also makes A be 991 * considered greater. 992 * 993 * @return -1, 0 or 1 as normal compare functions 994 */ 995 static int cake_tcph_sack_compare(const struct tcphdr *tcph_a, 996 const struct tcphdr *tcph_b) 997 { 998 const struct tcp_sack_block_wire *sack_a, *sack_b; 999 u32 ack_seq_a = ntohl(tcph_a->ack_seq); 1000 u32 bytes_a = 0, bytes_b = 0; 1001 int oplen_a, oplen_b; 1002 bool first = true; 1003 1004 sack_a = cake_get_tcpopt(tcph_a, TCPOPT_SACK, &oplen_a); 1005 sack_b = cake_get_tcpopt(tcph_b, TCPOPT_SACK, &oplen_b); 1006 1007 /* pointers point to option contents */ 1008 oplen_a -= TCPOLEN_SACK_BASE; 1009 oplen_b -= TCPOLEN_SACK_BASE; 1010 1011 if (sack_a && oplen_a >= sizeof(*sack_a) && 1012 (!sack_b || oplen_b < sizeof(*sack_b))) 1013 return -1; 1014 else if (sack_b && oplen_b >= sizeof(*sack_b) && 1015 (!sack_a || oplen_a < sizeof(*sack_a))) 1016 return 1; 1017 else if ((!sack_a || oplen_a < sizeof(*sack_a)) && 1018 (!sack_b || oplen_b < sizeof(*sack_b))) 1019 return 0; 1020 1021 while (oplen_a >= sizeof(*sack_a)) { 1022 const struct tcp_sack_block_wire *sack_tmp = sack_b; 1023 u32 start_a = get_unaligned_be32(&sack_a->start_seq); 1024 u32 end_a = get_unaligned_be32(&sack_a->end_seq); 1025 int oplen_tmp = oplen_b; 1026 bool found = false; 1027 1028 /* DSACK; always considered greater to prevent dropping */ 1029 if (before(start_a, ack_seq_a)) 1030 return -1; 1031 1032 bytes_a += end_a - start_a; 1033 1034 while (oplen_tmp >= sizeof(*sack_tmp)) { 1035 u32 start_b = get_unaligned_be32(&sack_tmp->start_seq); 1036 u32 end_b = get_unaligned_be32(&sack_tmp->end_seq); 1037 1038 /* first time through we count the total size */ 1039 if (first) 1040 bytes_b += end_b - start_b; 1041 1042 if (!after(start_b, start_a) && !before(end_b, end_a)) { 1043 found = true; 1044 if (!first) 1045 break; 1046 } 1047 oplen_tmp -= sizeof(*sack_tmp); 1048 sack_tmp++; 1049 } 1050 1051 if (!found) 1052 return -1; 1053 1054 oplen_a -= sizeof(*sack_a); 1055 sack_a++; 1056 first = false; 1057 } 1058 1059 /* If we made it this far, all ranges SACKed by A are covered by B, so 1060 * either the SACKs are equal, or B SACKs more bytes. 1061 */ 1062 return bytes_b > bytes_a ? 1 : 0; 1063 } 1064 1065 static void cake_tcph_get_tstamp(const struct tcphdr *tcph, 1066 u32 *tsval, u32 *tsecr) 1067 { 1068 const u8 *ptr; 1069 int opsize; 1070 1071 ptr = cake_get_tcpopt(tcph, TCPOPT_TIMESTAMP, &opsize); 1072 1073 if (ptr && opsize == TCPOLEN_TIMESTAMP) { 1074 *tsval = get_unaligned_be32(ptr); 1075 *tsecr = get_unaligned_be32(ptr + 4); 1076 } 1077 } 1078 1079 static bool cake_tcph_may_drop(const struct tcphdr *tcph, 1080 u32 tstamp_new, u32 tsecr_new) 1081 { 1082 /* inspired by tcp_parse_options in tcp_input.c */ 1083 int length = __tcp_hdrlen(tcph) - sizeof(struct tcphdr); 1084 const u8 *ptr = (const u8 *)(tcph + 1); 1085 u32 tstamp, tsecr; 1086 1087 /* 3 reserved flags must be unset to avoid future breakage 1088 * ACK must be set 1089 * ECE/CWR are handled separately 1090 * All other flags URG/PSH/RST/SYN/FIN must be unset 1091 * 0x0FFF0000 = all TCP flags (confirm ACK=1, others zero) 1092 * 0x00C00000 = CWR/ECE (handled separately) 1093 * 0x0F3F0000 = 0x0FFF0000 & ~0x00C00000 1094 */ 1095 if (((tcp_flag_word(tcph) & 1096 cpu_to_be32(0x0F3F0000)) != TCP_FLAG_ACK)) 1097 return false; 1098 1099 while (length > 0) { 1100 int opcode = *ptr++; 1101 int opsize; 1102 1103 if (opcode == TCPOPT_EOL) 1104 break; 1105 if (opcode == TCPOPT_NOP) { 1106 length--; 1107 continue; 1108 } 1109 if (length < 2) 1110 break; 1111 opsize = *ptr++; 1112 if (opsize < 2 || opsize > length) 1113 break; 1114 1115 switch (opcode) { 1116 case TCPOPT_MD5SIG: /* doesn't influence state */ 1117 break; 1118 1119 case TCPOPT_SACK: /* stricter checking performed later */ 1120 if (opsize % 8 != 2) 1121 return false; 1122 break; 1123 1124 case TCPOPT_TIMESTAMP: 1125 /* only drop timestamps lower than new */ 1126 if (opsize != TCPOLEN_TIMESTAMP) 1127 return false; 1128 tstamp = get_unaligned_be32(ptr); 1129 tsecr = get_unaligned_be32(ptr + 4); 1130 if (after(tstamp, tstamp_new) || 1131 after(tsecr, tsecr_new)) 1132 return false; 1133 break; 1134 1135 case TCPOPT_MSS: /* these should only be set on SYN */ 1136 case TCPOPT_WINDOW: 1137 case TCPOPT_SACK_PERM: 1138 case TCPOPT_FASTOPEN: 1139 case TCPOPT_EXP: 1140 default: /* don't drop if any unknown options are present */ 1141 return false; 1142 } 1143 1144 ptr += opsize - 2; 1145 length -= opsize; 1146 } 1147 1148 return true; 1149 } 1150 1151 static struct sk_buff *cake_ack_filter(struct cake_sched_data *q, 1152 struct cake_flow *flow) 1153 { 1154 bool aggressive = q->ack_filter == CAKE_ACK_AGGRESSIVE; 1155 struct sk_buff *elig_ack = NULL, *elig_ack_prev = NULL; 1156 struct sk_buff *skb_check, *skb_prev = NULL; 1157 const struct ipv6hdr *ipv6h, *ipv6h_check; 1158 unsigned char _tcph[64], _tcph_check[64]; 1159 const struct tcphdr *tcph, *tcph_check; 1160 const struct iphdr *iph, *iph_check; 1161 struct ipv6hdr _iph, _iph_check; 1162 const struct sk_buff *skb; 1163 int seglen, num_found = 0; 1164 u32 tstamp = 0, tsecr = 0; 1165 __be32 elig_flags = 0; 1166 int sack_comp; 1167 1168 /* no other possible ACKs to filter */ 1169 if (flow->head == flow->tail) 1170 return NULL; 1171 1172 skb = flow->tail; 1173 tcph = cake_get_tcphdr(skb, _tcph, sizeof(_tcph)); 1174 iph = cake_get_iphdr(skb, &_iph); 1175 if (!tcph) 1176 return NULL; 1177 1178 cake_tcph_get_tstamp(tcph, &tstamp, &tsecr); 1179 1180 /* the 'triggering' packet need only have the ACK flag set. 1181 * also check that SYN is not set, as there won't be any previous ACKs. 1182 */ 1183 if ((tcp_flag_word(tcph) & 1184 (TCP_FLAG_ACK | TCP_FLAG_SYN)) != TCP_FLAG_ACK) 1185 return NULL; 1186 1187 /* the 'triggering' ACK is at the tail of the queue, we have already 1188 * returned if it is the only packet in the flow. loop through the rest 1189 * of the queue looking for pure ACKs with the same 5-tuple as the 1190 * triggering one. 1191 */ 1192 for (skb_check = flow->head; 1193 skb_check && skb_check != skb; 1194 skb_prev = skb_check, skb_check = skb_check->next) { 1195 iph_check = cake_get_iphdr(skb_check, &_iph_check); 1196 tcph_check = cake_get_tcphdr(skb_check, &_tcph_check, 1197 sizeof(_tcph_check)); 1198 1199 /* only TCP packets with matching 5-tuple are eligible, and only 1200 * drop safe headers 1201 */ 1202 if (!tcph_check || iph->version != iph_check->version || 1203 tcph_check->source != tcph->source || 1204 tcph_check->dest != tcph->dest) 1205 continue; 1206 1207 if (iph_check->version == 4) { 1208 if (iph_check->saddr != iph->saddr || 1209 iph_check->daddr != iph->daddr) 1210 continue; 1211 1212 seglen = ntohs(iph_check->tot_len) - 1213 (4 * iph_check->ihl); 1214 } else if (iph_check->version == 6) { 1215 ipv6h = (struct ipv6hdr *)iph; 1216 ipv6h_check = (struct ipv6hdr *)iph_check; 1217 1218 if (ipv6_addr_cmp(&ipv6h_check->saddr, &ipv6h->saddr) || 1219 ipv6_addr_cmp(&ipv6h_check->daddr, &ipv6h->daddr)) 1220 continue; 1221 1222 seglen = ntohs(ipv6h_check->payload_len); 1223 } else { 1224 WARN_ON(1); /* shouldn't happen */ 1225 continue; 1226 } 1227 1228 /* If the ECE/CWR flags changed from the previous eligible 1229 * packet in the same flow, we should no longer be dropping that 1230 * previous packet as this would lose information. 1231 */ 1232 if (elig_ack && (tcp_flag_word(tcph_check) & 1233 (TCP_FLAG_ECE | TCP_FLAG_CWR)) != elig_flags) { 1234 elig_ack = NULL; 1235 elig_ack_prev = NULL; 1236 num_found--; 1237 } 1238 1239 /* Check TCP options and flags, don't drop ACKs with segment 1240 * data, and don't drop ACKs with a higher cumulative ACK 1241 * counter than the triggering packet. Check ACK seqno here to 1242 * avoid parsing SACK options of packets we are going to exclude 1243 * anyway. 1244 */ 1245 if (!cake_tcph_may_drop(tcph_check, tstamp, tsecr) || 1246 (seglen - __tcp_hdrlen(tcph_check)) != 0 || 1247 after(ntohl(tcph_check->ack_seq), ntohl(tcph->ack_seq))) 1248 continue; 1249 1250 /* Check SACK options. The triggering packet must SACK more data 1251 * than the ACK under consideration, or SACK the same range but 1252 * have a larger cumulative ACK counter. The latter is a 1253 * pathological case, but is contained in the following check 1254 * anyway, just to be safe. 1255 */ 1256 sack_comp = cake_tcph_sack_compare(tcph_check, tcph); 1257 1258 if (sack_comp < 0 || 1259 (ntohl(tcph_check->ack_seq) == ntohl(tcph->ack_seq) && 1260 sack_comp == 0)) 1261 continue; 1262 1263 /* At this point we have found an eligible pure ACK to drop; if 1264 * we are in aggressive mode, we are done. Otherwise, keep 1265 * searching unless this is the second eligible ACK we 1266 * found. 1267 * 1268 * Since we want to drop ACK closest to the head of the queue, 1269 * save the first eligible ACK we find, even if we need to loop 1270 * again. 1271 */ 1272 if (!elig_ack) { 1273 elig_ack = skb_check; 1274 elig_ack_prev = skb_prev; 1275 elig_flags = (tcp_flag_word(tcph_check) 1276 & (TCP_FLAG_ECE | TCP_FLAG_CWR)); 1277 } 1278 1279 if (num_found++ > 0) 1280 goto found; 1281 } 1282 1283 /* We made it through the queue without finding two eligible ACKs . If 1284 * we found a single eligible ACK we can drop it in aggressive mode if 1285 * we can guarantee that this does not interfere with ECN flag 1286 * information. We ensure this by dropping it only if the enqueued 1287 * packet is consecutive with the eligible ACK, and their flags match. 1288 */ 1289 if (elig_ack && aggressive && elig_ack->next == skb && 1290 (elig_flags == (tcp_flag_word(tcph) & 1291 (TCP_FLAG_ECE | TCP_FLAG_CWR)))) 1292 goto found; 1293 1294 return NULL; 1295 1296 found: 1297 if (elig_ack_prev) 1298 elig_ack_prev->next = elig_ack->next; 1299 else 1300 flow->head = elig_ack->next; 1301 1302 skb_mark_not_on_list(elig_ack); 1303 1304 return elig_ack; 1305 } 1306 1307 static u64 cake_ewma(u64 avg, u64 sample, u32 shift) 1308 { 1309 avg -= avg >> shift; 1310 avg += sample >> shift; 1311 return avg; 1312 } 1313 1314 static u32 cake_calc_overhead(struct cake_sched_data *q, u32 len, u32 off) 1315 { 1316 if (q->rate_flags & CAKE_FLAG_OVERHEAD) 1317 len -= off; 1318 1319 if (q->max_netlen < len) 1320 q->max_netlen = len; 1321 if (q->min_netlen > len) 1322 q->min_netlen = len; 1323 1324 len += q->rate_overhead; 1325 1326 if (len < q->rate_mpu) 1327 len = q->rate_mpu; 1328 1329 if (q->atm_mode == CAKE_ATM_ATM) { 1330 len += 47; 1331 len /= 48; 1332 len *= 53; 1333 } else if (q->atm_mode == CAKE_ATM_PTM) { 1334 /* Add one byte per 64 bytes or part thereof. 1335 * This is conservative and easier to calculate than the 1336 * precise value. 1337 */ 1338 len += (len + 63) / 64; 1339 } 1340 1341 if (q->max_adjlen < len) 1342 q->max_adjlen = len; 1343 if (q->min_adjlen > len) 1344 q->min_adjlen = len; 1345 1346 return len; 1347 } 1348 1349 static u32 cake_overhead(struct cake_sched_data *q, const struct sk_buff *skb) 1350 { 1351 const struct skb_shared_info *shinfo = skb_shinfo(skb); 1352 unsigned int hdr_len, last_len = 0; 1353 u32 off = skb_network_offset(skb); 1354 u32 len = qdisc_pkt_len(skb); 1355 u16 segs = 1; 1356 1357 q->avg_netoff = cake_ewma(q->avg_netoff, off << 16, 8); 1358 1359 if (!shinfo->gso_size) 1360 return cake_calc_overhead(q, len, off); 1361 1362 /* borrowed from qdisc_pkt_len_init() */ 1363 hdr_len = skb_transport_header(skb) - skb_mac_header(skb); 1364 1365 /* + transport layer */ 1366 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | 1367 SKB_GSO_TCPV6))) { 1368 const struct tcphdr *th; 1369 struct tcphdr _tcphdr; 1370 1371 th = skb_header_pointer(skb, skb_transport_offset(skb), 1372 sizeof(_tcphdr), &_tcphdr); 1373 if (likely(th)) 1374 hdr_len += __tcp_hdrlen(th); 1375 } else { 1376 struct udphdr _udphdr; 1377 1378 if (skb_header_pointer(skb, skb_transport_offset(skb), 1379 sizeof(_udphdr), &_udphdr)) 1380 hdr_len += sizeof(struct udphdr); 1381 } 1382 1383 if (unlikely(shinfo->gso_type & SKB_GSO_DODGY)) 1384 segs = DIV_ROUND_UP(skb->len - hdr_len, 1385 shinfo->gso_size); 1386 else 1387 segs = shinfo->gso_segs; 1388 1389 len = shinfo->gso_size + hdr_len; 1390 last_len = skb->len - shinfo->gso_size * (segs - 1); 1391 1392 return (cake_calc_overhead(q, len, off) * (segs - 1) + 1393 cake_calc_overhead(q, last_len, off)); 1394 } 1395 1396 static void cake_heap_swap(struct cake_sched_data *q, u16 i, u16 j) 1397 { 1398 struct cake_heap_entry ii = q->overflow_heap[i]; 1399 struct cake_heap_entry jj = q->overflow_heap[j]; 1400 1401 q->overflow_heap[i] = jj; 1402 q->overflow_heap[j] = ii; 1403 1404 q->tins[ii.t].overflow_idx[ii.b] = j; 1405 q->tins[jj.t].overflow_idx[jj.b] = i; 1406 } 1407 1408 static u32 cake_heap_get_backlog(const struct cake_sched_data *q, u16 i) 1409 { 1410 struct cake_heap_entry ii = q->overflow_heap[i]; 1411 1412 return q->tins[ii.t].backlogs[ii.b]; 1413 } 1414 1415 static void cake_heapify(struct cake_sched_data *q, u16 i) 1416 { 1417 static const u32 a = CAKE_MAX_TINS * CAKE_QUEUES; 1418 u32 mb = cake_heap_get_backlog(q, i); 1419 u32 m = i; 1420 1421 while (m < a) { 1422 u32 l = m + m + 1; 1423 u32 r = l + 1; 1424 1425 if (l < a) { 1426 u32 lb = cake_heap_get_backlog(q, l); 1427 1428 if (lb > mb) { 1429 m = l; 1430 mb = lb; 1431 } 1432 } 1433 1434 if (r < a) { 1435 u32 rb = cake_heap_get_backlog(q, r); 1436 1437 if (rb > mb) { 1438 m = r; 1439 mb = rb; 1440 } 1441 } 1442 1443 if (m != i) { 1444 cake_heap_swap(q, i, m); 1445 i = m; 1446 } else { 1447 break; 1448 } 1449 } 1450 } 1451 1452 static void cake_heapify_up(struct cake_sched_data *q, u16 i) 1453 { 1454 while (i > 0 && i < CAKE_MAX_TINS * CAKE_QUEUES) { 1455 u16 p = (i - 1) >> 1; 1456 u32 ib = cake_heap_get_backlog(q, i); 1457 u32 pb = cake_heap_get_backlog(q, p); 1458 1459 if (ib > pb) { 1460 cake_heap_swap(q, i, p); 1461 i = p; 1462 } else { 1463 break; 1464 } 1465 } 1466 } 1467 1468 static int cake_advance_shaper(struct cake_sched_data *q, 1469 struct cake_tin_data *b, 1470 struct sk_buff *skb, 1471 ktime_t now, bool drop) 1472 { 1473 u32 len = get_cobalt_cb(skb)->adjusted_len; 1474 1475 /* charge packet bandwidth to this tin 1476 * and to the global shaper. 1477 */ 1478 if (q->rate_ns) { 1479 u64 tin_dur = (len * b->tin_rate_ns) >> b->tin_rate_shft; 1480 u64 global_dur = (len * q->rate_ns) >> q->rate_shft; 1481 u64 failsafe_dur = global_dur + (global_dur >> 1); 1482 1483 if (ktime_before(b->time_next_packet, now)) 1484 b->time_next_packet = ktime_add_ns(b->time_next_packet, 1485 tin_dur); 1486 1487 else if (ktime_before(b->time_next_packet, 1488 ktime_add_ns(now, tin_dur))) 1489 b->time_next_packet = ktime_add_ns(now, tin_dur); 1490 1491 q->time_next_packet = ktime_add_ns(q->time_next_packet, 1492 global_dur); 1493 if (!drop) 1494 q->failsafe_next_packet = \ 1495 ktime_add_ns(q->failsafe_next_packet, 1496 failsafe_dur); 1497 } 1498 return len; 1499 } 1500 1501 static unsigned int cake_drop(struct Qdisc *sch, struct sk_buff **to_free) 1502 { 1503 struct cake_sched_data *q = qdisc_priv(sch); 1504 ktime_t now = ktime_get(); 1505 u32 idx = 0, tin = 0, len; 1506 struct cake_heap_entry qq; 1507 struct cake_tin_data *b; 1508 struct cake_flow *flow; 1509 struct sk_buff *skb; 1510 1511 if (!q->overflow_timeout) { 1512 int i; 1513 /* Build fresh max-heap */ 1514 for (i = CAKE_MAX_TINS * CAKE_QUEUES / 2; i >= 0; i--) 1515 cake_heapify(q, i); 1516 } 1517 q->overflow_timeout = 65535; 1518 1519 /* select longest queue for pruning */ 1520 qq = q->overflow_heap[0]; 1521 tin = qq.t; 1522 idx = qq.b; 1523 1524 b = &q->tins[tin]; 1525 flow = &b->flows[idx]; 1526 skb = dequeue_head(flow); 1527 if (unlikely(!skb)) { 1528 /* heap has gone wrong, rebuild it next time */ 1529 q->overflow_timeout = 0; 1530 return idx + (tin << 16); 1531 } 1532 1533 if (cobalt_queue_full(&flow->cvars, &b->cparams, now)) 1534 b->unresponsive_flow_count++; 1535 1536 len = qdisc_pkt_len(skb); 1537 q->buffer_used -= skb->truesize; 1538 b->backlogs[idx] -= len; 1539 b->tin_backlog -= len; 1540 sch->qstats.backlog -= len; 1541 qdisc_tree_reduce_backlog(sch, 1, len); 1542 1543 flow->dropped++; 1544 b->tin_dropped++; 1545 sch->qstats.drops++; 1546 1547 if (q->rate_flags & CAKE_FLAG_INGRESS) 1548 cake_advance_shaper(q, b, skb, now, true); 1549 1550 __qdisc_drop(skb, to_free); 1551 sch->q.qlen--; 1552 1553 cake_heapify(q, 0); 1554 1555 return idx + (tin << 16); 1556 } 1557 1558 static u8 cake_handle_diffserv(struct sk_buff *skb, bool wash) 1559 { 1560 const int offset = skb_network_offset(skb); 1561 u16 *buf, buf_; 1562 u8 dscp; 1563 1564 switch (skb_protocol(skb, true)) { 1565 case htons(ETH_P_IP): 1566 buf = skb_header_pointer(skb, offset, sizeof(buf_), &buf_); 1567 if (unlikely(!buf)) 1568 return 0; 1569 1570 /* ToS is in the second byte of iphdr */ 1571 dscp = ipv4_get_dsfield((struct iphdr *)buf) >> 2; 1572 1573 if (wash && dscp) { 1574 const int wlen = offset + sizeof(struct iphdr); 1575 1576 if (!pskb_may_pull(skb, wlen) || 1577 skb_try_make_writable(skb, wlen)) 1578 return 0; 1579 1580 ipv4_change_dsfield(ip_hdr(skb), INET_ECN_MASK, 0); 1581 } 1582 1583 return dscp; 1584 1585 case htons(ETH_P_IPV6): 1586 buf = skb_header_pointer(skb, offset, sizeof(buf_), &buf_); 1587 if (unlikely(!buf)) 1588 return 0; 1589 1590 /* Traffic class is in the first and second bytes of ipv6hdr */ 1591 dscp = ipv6_get_dsfield((struct ipv6hdr *)buf) >> 2; 1592 1593 if (wash && dscp) { 1594 const int wlen = offset + sizeof(struct ipv6hdr); 1595 1596 if (!pskb_may_pull(skb, wlen) || 1597 skb_try_make_writable(skb, wlen)) 1598 return 0; 1599 1600 ipv6_change_dsfield(ipv6_hdr(skb), INET_ECN_MASK, 0); 1601 } 1602 1603 return dscp; 1604 1605 case htons(ETH_P_ARP): 1606 return 0x38; /* CS7 - Net Control */ 1607 1608 default: 1609 /* If there is no Diffserv field, treat as best-effort */ 1610 return 0; 1611 } 1612 } 1613 1614 static struct cake_tin_data *cake_select_tin(struct Qdisc *sch, 1615 struct sk_buff *skb) 1616 { 1617 struct cake_sched_data *q = qdisc_priv(sch); 1618 u32 tin, mark; 1619 bool wash; 1620 u8 dscp; 1621 1622 /* Tin selection: Default to diffserv-based selection, allow overriding 1623 * using firewall marks or skb->priority. Call DSCP parsing early if 1624 * wash is enabled, otherwise defer to below to skip unneeded parsing. 1625 */ 1626 mark = (skb->mark & q->fwmark_mask) >> q->fwmark_shft; 1627 wash = !!(q->rate_flags & CAKE_FLAG_WASH); 1628 if (wash) 1629 dscp = cake_handle_diffserv(skb, wash); 1630 1631 if (q->tin_mode == CAKE_DIFFSERV_BESTEFFORT) 1632 tin = 0; 1633 1634 else if (mark && mark <= q->tin_cnt) 1635 tin = q->tin_order[mark - 1]; 1636 1637 else if (TC_H_MAJ(skb->priority) == sch->handle && 1638 TC_H_MIN(skb->priority) > 0 && 1639 TC_H_MIN(skb->priority) <= q->tin_cnt) 1640 tin = q->tin_order[TC_H_MIN(skb->priority) - 1]; 1641 1642 else { 1643 if (!wash) 1644 dscp = cake_handle_diffserv(skb, wash); 1645 tin = q->tin_index[dscp]; 1646 1647 if (unlikely(tin >= q->tin_cnt)) 1648 tin = 0; 1649 } 1650 1651 return &q->tins[tin]; 1652 } 1653 1654 static u32 cake_classify(struct Qdisc *sch, struct cake_tin_data **t, 1655 struct sk_buff *skb, int flow_mode, int *qerr) 1656 { 1657 struct cake_sched_data *q = qdisc_priv(sch); 1658 struct tcf_proto *filter; 1659 struct tcf_result res; 1660 u16 flow = 0, host = 0; 1661 int result; 1662 1663 filter = rcu_dereference_bh(q->filter_list); 1664 if (!filter) 1665 goto hash; 1666 1667 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS; 1668 result = tcf_classify(skb, NULL, filter, &res, false); 1669 1670 if (result >= 0) { 1671 #ifdef CONFIG_NET_CLS_ACT 1672 switch (result) { 1673 case TC_ACT_STOLEN: 1674 case TC_ACT_QUEUED: 1675 case TC_ACT_TRAP: 1676 *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN; 1677 fallthrough; 1678 case TC_ACT_SHOT: 1679 return 0; 1680 } 1681 #endif 1682 if (TC_H_MIN(res.classid) <= CAKE_QUEUES) 1683 flow = TC_H_MIN(res.classid); 1684 if (TC_H_MAJ(res.classid) <= (CAKE_QUEUES << 16)) 1685 host = TC_H_MAJ(res.classid) >> 16; 1686 } 1687 hash: 1688 *t = cake_select_tin(sch, skb); 1689 return cake_hash(*t, skb, flow_mode, flow, host) + 1; 1690 } 1691 1692 static void cake_reconfigure(struct Qdisc *sch); 1693 1694 static s32 cake_enqueue(struct sk_buff *skb, struct Qdisc *sch, 1695 struct sk_buff **to_free) 1696 { 1697 struct cake_sched_data *q = qdisc_priv(sch); 1698 int len = qdisc_pkt_len(skb); 1699 int ret; 1700 struct sk_buff *ack = NULL; 1701 ktime_t now = ktime_get(); 1702 struct cake_tin_data *b; 1703 struct cake_flow *flow; 1704 u32 idx; 1705 1706 /* choose flow to insert into */ 1707 idx = cake_classify(sch, &b, skb, q->flow_mode, &ret); 1708 if (idx == 0) { 1709 if (ret & __NET_XMIT_BYPASS) 1710 qdisc_qstats_drop(sch); 1711 __qdisc_drop(skb, to_free); 1712 return ret; 1713 } 1714 idx--; 1715 flow = &b->flows[idx]; 1716 1717 /* ensure shaper state isn't stale */ 1718 if (!b->tin_backlog) { 1719 if (ktime_before(b->time_next_packet, now)) 1720 b->time_next_packet = now; 1721 1722 if (!sch->q.qlen) { 1723 if (ktime_before(q->time_next_packet, now)) { 1724 q->failsafe_next_packet = now; 1725 q->time_next_packet = now; 1726 } else if (ktime_after(q->time_next_packet, now) && 1727 ktime_after(q->failsafe_next_packet, now)) { 1728 u64 next = \ 1729 min(ktime_to_ns(q->time_next_packet), 1730 ktime_to_ns( 1731 q->failsafe_next_packet)); 1732 sch->qstats.overlimits++; 1733 qdisc_watchdog_schedule_ns(&q->watchdog, next); 1734 } 1735 } 1736 } 1737 1738 if (unlikely(len > b->max_skblen)) 1739 b->max_skblen = len; 1740 1741 if (skb_is_gso(skb) && q->rate_flags & CAKE_FLAG_SPLIT_GSO) { 1742 struct sk_buff *segs, *nskb; 1743 netdev_features_t features = netif_skb_features(skb); 1744 unsigned int slen = 0, numsegs = 0; 1745 1746 segs = skb_gso_segment(skb, features & ~NETIF_F_GSO_MASK); 1747 if (IS_ERR_OR_NULL(segs)) 1748 return qdisc_drop(skb, sch, to_free); 1749 1750 skb_list_walk_safe(segs, segs, nskb) { 1751 skb_mark_not_on_list(segs); 1752 qdisc_skb_cb(segs)->pkt_len = segs->len; 1753 cobalt_set_enqueue_time(segs, now); 1754 get_cobalt_cb(segs)->adjusted_len = cake_overhead(q, 1755 segs); 1756 flow_queue_add(flow, segs); 1757 1758 sch->q.qlen++; 1759 numsegs++; 1760 slen += segs->len; 1761 q->buffer_used += segs->truesize; 1762 b->packets++; 1763 } 1764 1765 /* stats */ 1766 b->bytes += slen; 1767 b->backlogs[idx] += slen; 1768 b->tin_backlog += slen; 1769 sch->qstats.backlog += slen; 1770 q->avg_window_bytes += slen; 1771 1772 qdisc_tree_reduce_backlog(sch, 1-numsegs, len-slen); 1773 consume_skb(skb); 1774 } else { 1775 /* not splitting */ 1776 cobalt_set_enqueue_time(skb, now); 1777 get_cobalt_cb(skb)->adjusted_len = cake_overhead(q, skb); 1778 flow_queue_add(flow, skb); 1779 1780 if (q->ack_filter) 1781 ack = cake_ack_filter(q, flow); 1782 1783 if (ack) { 1784 b->ack_drops++; 1785 sch->qstats.drops++; 1786 b->bytes += qdisc_pkt_len(ack); 1787 len -= qdisc_pkt_len(ack); 1788 q->buffer_used += skb->truesize - ack->truesize; 1789 if (q->rate_flags & CAKE_FLAG_INGRESS) 1790 cake_advance_shaper(q, b, ack, now, true); 1791 1792 qdisc_tree_reduce_backlog(sch, 1, qdisc_pkt_len(ack)); 1793 consume_skb(ack); 1794 } else { 1795 sch->q.qlen++; 1796 q->buffer_used += skb->truesize; 1797 } 1798 1799 /* stats */ 1800 b->packets++; 1801 b->bytes += len; 1802 b->backlogs[idx] += len; 1803 b->tin_backlog += len; 1804 sch->qstats.backlog += len; 1805 q->avg_window_bytes += len; 1806 } 1807 1808 if (q->overflow_timeout) 1809 cake_heapify_up(q, b->overflow_idx[idx]); 1810 1811 /* incoming bandwidth capacity estimate */ 1812 if (q->rate_flags & CAKE_FLAG_AUTORATE_INGRESS) { 1813 u64 packet_interval = \ 1814 ktime_to_ns(ktime_sub(now, q->last_packet_time)); 1815 1816 if (packet_interval > NSEC_PER_SEC) 1817 packet_interval = NSEC_PER_SEC; 1818 1819 /* filter out short-term bursts, eg. wifi aggregation */ 1820 q->avg_packet_interval = \ 1821 cake_ewma(q->avg_packet_interval, 1822 packet_interval, 1823 (packet_interval > q->avg_packet_interval ? 1824 2 : 8)); 1825 1826 q->last_packet_time = now; 1827 1828 if (packet_interval > q->avg_packet_interval) { 1829 u64 window_interval = \ 1830 ktime_to_ns(ktime_sub(now, 1831 q->avg_window_begin)); 1832 u64 b = q->avg_window_bytes * (u64)NSEC_PER_SEC; 1833 1834 b = div64_u64(b, window_interval); 1835 q->avg_peak_bandwidth = 1836 cake_ewma(q->avg_peak_bandwidth, b, 1837 b > q->avg_peak_bandwidth ? 2 : 8); 1838 q->avg_window_bytes = 0; 1839 q->avg_window_begin = now; 1840 1841 if (ktime_after(now, 1842 ktime_add_ms(q->last_reconfig_time, 1843 250))) { 1844 q->rate_bps = (q->avg_peak_bandwidth * 15) >> 4; 1845 cake_reconfigure(sch); 1846 } 1847 } 1848 } else { 1849 q->avg_window_bytes = 0; 1850 q->last_packet_time = now; 1851 } 1852 1853 /* flowchain */ 1854 if (!flow->set || flow->set == CAKE_SET_DECAYING) { 1855 struct cake_host *srchost = &b->hosts[flow->srchost]; 1856 struct cake_host *dsthost = &b->hosts[flow->dsthost]; 1857 u16 host_load = 1; 1858 1859 if (!flow->set) { 1860 list_add_tail(&flow->flowchain, &b->new_flows); 1861 } else { 1862 b->decaying_flow_count--; 1863 list_move_tail(&flow->flowchain, &b->new_flows); 1864 } 1865 flow->set = CAKE_SET_SPARSE; 1866 b->sparse_flow_count++; 1867 1868 if (cake_dsrc(q->flow_mode)) 1869 host_load = max(host_load, srchost->srchost_bulk_flow_count); 1870 1871 if (cake_ddst(q->flow_mode)) 1872 host_load = max(host_load, dsthost->dsthost_bulk_flow_count); 1873 1874 flow->deficit = (b->flow_quantum * 1875 quantum_div[host_load]) >> 16; 1876 } else if (flow->set == CAKE_SET_SPARSE_WAIT) { 1877 struct cake_host *srchost = &b->hosts[flow->srchost]; 1878 struct cake_host *dsthost = &b->hosts[flow->dsthost]; 1879 1880 /* this flow was empty, accounted as a sparse flow, but actually 1881 * in the bulk rotation. 1882 */ 1883 flow->set = CAKE_SET_BULK; 1884 b->sparse_flow_count--; 1885 b->bulk_flow_count++; 1886 1887 if (cake_dsrc(q->flow_mode)) 1888 srchost->srchost_bulk_flow_count++; 1889 1890 if (cake_ddst(q->flow_mode)) 1891 dsthost->dsthost_bulk_flow_count++; 1892 1893 } 1894 1895 if (q->buffer_used > q->buffer_max_used) 1896 q->buffer_max_used = q->buffer_used; 1897 1898 if (q->buffer_used > q->buffer_limit) { 1899 u32 dropped = 0; 1900 1901 while (q->buffer_used > q->buffer_limit) { 1902 dropped++; 1903 cake_drop(sch, to_free); 1904 } 1905 b->drop_overlimit += dropped; 1906 } 1907 return NET_XMIT_SUCCESS; 1908 } 1909 1910 static struct sk_buff *cake_dequeue_one(struct Qdisc *sch) 1911 { 1912 struct cake_sched_data *q = qdisc_priv(sch); 1913 struct cake_tin_data *b = &q->tins[q->cur_tin]; 1914 struct cake_flow *flow = &b->flows[q->cur_flow]; 1915 struct sk_buff *skb = NULL; 1916 u32 len; 1917 1918 if (flow->head) { 1919 skb = dequeue_head(flow); 1920 len = qdisc_pkt_len(skb); 1921 b->backlogs[q->cur_flow] -= len; 1922 b->tin_backlog -= len; 1923 sch->qstats.backlog -= len; 1924 q->buffer_used -= skb->truesize; 1925 sch->q.qlen--; 1926 1927 if (q->overflow_timeout) 1928 cake_heapify(q, b->overflow_idx[q->cur_flow]); 1929 } 1930 return skb; 1931 } 1932 1933 /* Discard leftover packets from a tin no longer in use. */ 1934 static void cake_clear_tin(struct Qdisc *sch, u16 tin) 1935 { 1936 struct cake_sched_data *q = qdisc_priv(sch); 1937 struct sk_buff *skb; 1938 1939 q->cur_tin = tin; 1940 for (q->cur_flow = 0; q->cur_flow < CAKE_QUEUES; q->cur_flow++) 1941 while (!!(skb = cake_dequeue_one(sch))) 1942 kfree_skb(skb); 1943 } 1944 1945 static struct sk_buff *cake_dequeue(struct Qdisc *sch) 1946 { 1947 struct cake_sched_data *q = qdisc_priv(sch); 1948 struct cake_tin_data *b = &q->tins[q->cur_tin]; 1949 struct cake_host *srchost, *dsthost; 1950 ktime_t now = ktime_get(); 1951 struct cake_flow *flow; 1952 struct list_head *head; 1953 bool first_flow = true; 1954 struct sk_buff *skb; 1955 u16 host_load; 1956 u64 delay; 1957 u32 len; 1958 1959 begin: 1960 if (!sch->q.qlen) 1961 return NULL; 1962 1963 /* global hard shaper */ 1964 if (ktime_after(q->time_next_packet, now) && 1965 ktime_after(q->failsafe_next_packet, now)) { 1966 u64 next = min(ktime_to_ns(q->time_next_packet), 1967 ktime_to_ns(q->failsafe_next_packet)); 1968 1969 sch->qstats.overlimits++; 1970 qdisc_watchdog_schedule_ns(&q->watchdog, next); 1971 return NULL; 1972 } 1973 1974 /* Choose a class to work on. */ 1975 if (!q->rate_ns) { 1976 /* In unlimited mode, can't rely on shaper timings, just balance 1977 * with DRR 1978 */ 1979 bool wrapped = false, empty = true; 1980 1981 while (b->tin_deficit < 0 || 1982 !(b->sparse_flow_count + b->bulk_flow_count)) { 1983 if (b->tin_deficit <= 0) 1984 b->tin_deficit += b->tin_quantum; 1985 if (b->sparse_flow_count + b->bulk_flow_count) 1986 empty = false; 1987 1988 q->cur_tin++; 1989 b++; 1990 if (q->cur_tin >= q->tin_cnt) { 1991 q->cur_tin = 0; 1992 b = q->tins; 1993 1994 if (wrapped) { 1995 /* It's possible for q->qlen to be 1996 * nonzero when we actually have no 1997 * packets anywhere. 1998 */ 1999 if (empty) 2000 return NULL; 2001 } else { 2002 wrapped = true; 2003 } 2004 } 2005 } 2006 } else { 2007 /* In shaped mode, choose: 2008 * - Highest-priority tin with queue and meeting schedule, or 2009 * - The earliest-scheduled tin with queue. 2010 */ 2011 ktime_t best_time = KTIME_MAX; 2012 int tin, best_tin = 0; 2013 2014 for (tin = 0; tin < q->tin_cnt; tin++) { 2015 b = q->tins + tin; 2016 if ((b->sparse_flow_count + b->bulk_flow_count) > 0) { 2017 ktime_t time_to_pkt = \ 2018 ktime_sub(b->time_next_packet, now); 2019 2020 if (ktime_to_ns(time_to_pkt) <= 0 || 2021 ktime_compare(time_to_pkt, 2022 best_time) <= 0) { 2023 best_time = time_to_pkt; 2024 best_tin = tin; 2025 } 2026 } 2027 } 2028 2029 q->cur_tin = best_tin; 2030 b = q->tins + best_tin; 2031 2032 /* No point in going further if no packets to deliver. */ 2033 if (unlikely(!(b->sparse_flow_count + b->bulk_flow_count))) 2034 return NULL; 2035 } 2036 2037 retry: 2038 /* service this class */ 2039 head = &b->decaying_flows; 2040 if (!first_flow || list_empty(head)) { 2041 head = &b->new_flows; 2042 if (list_empty(head)) { 2043 head = &b->old_flows; 2044 if (unlikely(list_empty(head))) { 2045 head = &b->decaying_flows; 2046 if (unlikely(list_empty(head))) 2047 goto begin; 2048 } 2049 } 2050 } 2051 flow = list_first_entry(head, struct cake_flow, flowchain); 2052 q->cur_flow = flow - b->flows; 2053 first_flow = false; 2054 2055 /* triple isolation (modified DRR++) */ 2056 srchost = &b->hosts[flow->srchost]; 2057 dsthost = &b->hosts[flow->dsthost]; 2058 host_load = 1; 2059 2060 /* flow isolation (DRR++) */ 2061 if (flow->deficit <= 0) { 2062 /* Keep all flows with deficits out of the sparse and decaying 2063 * rotations. No non-empty flow can go into the decaying 2064 * rotation, so they can't get deficits 2065 */ 2066 if (flow->set == CAKE_SET_SPARSE) { 2067 if (flow->head) { 2068 b->sparse_flow_count--; 2069 b->bulk_flow_count++; 2070 2071 if (cake_dsrc(q->flow_mode)) 2072 srchost->srchost_bulk_flow_count++; 2073 2074 if (cake_ddst(q->flow_mode)) 2075 dsthost->dsthost_bulk_flow_count++; 2076 2077 flow->set = CAKE_SET_BULK; 2078 } else { 2079 /* we've moved it to the bulk rotation for 2080 * correct deficit accounting but we still want 2081 * to count it as a sparse flow, not a bulk one. 2082 */ 2083 flow->set = CAKE_SET_SPARSE_WAIT; 2084 } 2085 } 2086 2087 if (cake_dsrc(q->flow_mode)) 2088 host_load = max(host_load, srchost->srchost_bulk_flow_count); 2089 2090 if (cake_ddst(q->flow_mode)) 2091 host_load = max(host_load, dsthost->dsthost_bulk_flow_count); 2092 2093 WARN_ON(host_load > CAKE_QUEUES); 2094 2095 /* The shifted prandom_u32() is a way to apply dithering to 2096 * avoid accumulating roundoff errors 2097 */ 2098 flow->deficit += (b->flow_quantum * quantum_div[host_load] + 2099 (prandom_u32() >> 16)) >> 16; 2100 list_move_tail(&flow->flowchain, &b->old_flows); 2101 2102 goto retry; 2103 } 2104 2105 /* Retrieve a packet via the AQM */ 2106 while (1) { 2107 skb = cake_dequeue_one(sch); 2108 if (!skb) { 2109 /* this queue was actually empty */ 2110 if (cobalt_queue_empty(&flow->cvars, &b->cparams, now)) 2111 b->unresponsive_flow_count--; 2112 2113 if (flow->cvars.p_drop || flow->cvars.count || 2114 ktime_before(now, flow->cvars.drop_next)) { 2115 /* keep in the flowchain until the state has 2116 * decayed to rest 2117 */ 2118 list_move_tail(&flow->flowchain, 2119 &b->decaying_flows); 2120 if (flow->set == CAKE_SET_BULK) { 2121 b->bulk_flow_count--; 2122 2123 if (cake_dsrc(q->flow_mode)) 2124 srchost->srchost_bulk_flow_count--; 2125 2126 if (cake_ddst(q->flow_mode)) 2127 dsthost->dsthost_bulk_flow_count--; 2128 2129 b->decaying_flow_count++; 2130 } else if (flow->set == CAKE_SET_SPARSE || 2131 flow->set == CAKE_SET_SPARSE_WAIT) { 2132 b->sparse_flow_count--; 2133 b->decaying_flow_count++; 2134 } 2135 flow->set = CAKE_SET_DECAYING; 2136 } else { 2137 /* remove empty queue from the flowchain */ 2138 list_del_init(&flow->flowchain); 2139 if (flow->set == CAKE_SET_SPARSE || 2140 flow->set == CAKE_SET_SPARSE_WAIT) 2141 b->sparse_flow_count--; 2142 else if (flow->set == CAKE_SET_BULK) { 2143 b->bulk_flow_count--; 2144 2145 if (cake_dsrc(q->flow_mode)) 2146 srchost->srchost_bulk_flow_count--; 2147 2148 if (cake_ddst(q->flow_mode)) 2149 dsthost->dsthost_bulk_flow_count--; 2150 2151 } else 2152 b->decaying_flow_count--; 2153 2154 flow->set = CAKE_SET_NONE; 2155 } 2156 goto begin; 2157 } 2158 2159 /* Last packet in queue may be marked, shouldn't be dropped */ 2160 if (!cobalt_should_drop(&flow->cvars, &b->cparams, now, skb, 2161 (b->bulk_flow_count * 2162 !!(q->rate_flags & 2163 CAKE_FLAG_INGRESS))) || 2164 !flow->head) 2165 break; 2166 2167 /* drop this packet, get another one */ 2168 if (q->rate_flags & CAKE_FLAG_INGRESS) { 2169 len = cake_advance_shaper(q, b, skb, 2170 now, true); 2171 flow->deficit -= len; 2172 b->tin_deficit -= len; 2173 } 2174 flow->dropped++; 2175 b->tin_dropped++; 2176 qdisc_tree_reduce_backlog(sch, 1, qdisc_pkt_len(skb)); 2177 qdisc_qstats_drop(sch); 2178 kfree_skb(skb); 2179 if (q->rate_flags & CAKE_FLAG_INGRESS) 2180 goto retry; 2181 } 2182 2183 b->tin_ecn_mark += !!flow->cvars.ecn_marked; 2184 qdisc_bstats_update(sch, skb); 2185 2186 /* collect delay stats */ 2187 delay = ktime_to_ns(ktime_sub(now, cobalt_get_enqueue_time(skb))); 2188 b->avge_delay = cake_ewma(b->avge_delay, delay, 8); 2189 b->peak_delay = cake_ewma(b->peak_delay, delay, 2190 delay > b->peak_delay ? 2 : 8); 2191 b->base_delay = cake_ewma(b->base_delay, delay, 2192 delay < b->base_delay ? 2 : 8); 2193 2194 len = cake_advance_shaper(q, b, skb, now, false); 2195 flow->deficit -= len; 2196 b->tin_deficit -= len; 2197 2198 if (ktime_after(q->time_next_packet, now) && sch->q.qlen) { 2199 u64 next = min(ktime_to_ns(q->time_next_packet), 2200 ktime_to_ns(q->failsafe_next_packet)); 2201 2202 qdisc_watchdog_schedule_ns(&q->watchdog, next); 2203 } else if (!sch->q.qlen) { 2204 int i; 2205 2206 for (i = 0; i < q->tin_cnt; i++) { 2207 if (q->tins[i].decaying_flow_count) { 2208 ktime_t next = \ 2209 ktime_add_ns(now, 2210 q->tins[i].cparams.target); 2211 2212 qdisc_watchdog_schedule_ns(&q->watchdog, 2213 ktime_to_ns(next)); 2214 break; 2215 } 2216 } 2217 } 2218 2219 if (q->overflow_timeout) 2220 q->overflow_timeout--; 2221 2222 return skb; 2223 } 2224 2225 static void cake_reset(struct Qdisc *sch) 2226 { 2227 u32 c; 2228 2229 for (c = 0; c < CAKE_MAX_TINS; c++) 2230 cake_clear_tin(sch, c); 2231 } 2232 2233 static const struct nla_policy cake_policy[TCA_CAKE_MAX + 1] = { 2234 [TCA_CAKE_BASE_RATE64] = { .type = NLA_U64 }, 2235 [TCA_CAKE_DIFFSERV_MODE] = { .type = NLA_U32 }, 2236 [TCA_CAKE_ATM] = { .type = NLA_U32 }, 2237 [TCA_CAKE_FLOW_MODE] = { .type = NLA_U32 }, 2238 [TCA_CAKE_OVERHEAD] = { .type = NLA_S32 }, 2239 [TCA_CAKE_RTT] = { .type = NLA_U32 }, 2240 [TCA_CAKE_TARGET] = { .type = NLA_U32 }, 2241 [TCA_CAKE_AUTORATE] = { .type = NLA_U32 }, 2242 [TCA_CAKE_MEMORY] = { .type = NLA_U32 }, 2243 [TCA_CAKE_NAT] = { .type = NLA_U32 }, 2244 [TCA_CAKE_RAW] = { .type = NLA_U32 }, 2245 [TCA_CAKE_WASH] = { .type = NLA_U32 }, 2246 [TCA_CAKE_MPU] = { .type = NLA_U32 }, 2247 [TCA_CAKE_INGRESS] = { .type = NLA_U32 }, 2248 [TCA_CAKE_ACK_FILTER] = { .type = NLA_U32 }, 2249 [TCA_CAKE_SPLIT_GSO] = { .type = NLA_U32 }, 2250 [TCA_CAKE_FWMARK] = { .type = NLA_U32 }, 2251 }; 2252 2253 static void cake_set_rate(struct cake_tin_data *b, u64 rate, u32 mtu, 2254 u64 target_ns, u64 rtt_est_ns) 2255 { 2256 /* convert byte-rate into time-per-byte 2257 * so it will always unwedge in reasonable time. 2258 */ 2259 static const u64 MIN_RATE = 64; 2260 u32 byte_target = mtu; 2261 u64 byte_target_ns; 2262 u8 rate_shft = 0; 2263 u64 rate_ns = 0; 2264 2265 b->flow_quantum = 1514; 2266 if (rate) { 2267 b->flow_quantum = max(min(rate >> 12, 1514ULL), 300ULL); 2268 rate_shft = 34; 2269 rate_ns = ((u64)NSEC_PER_SEC) << rate_shft; 2270 rate_ns = div64_u64(rate_ns, max(MIN_RATE, rate)); 2271 while (!!(rate_ns >> 34)) { 2272 rate_ns >>= 1; 2273 rate_shft--; 2274 } 2275 } /* else unlimited, ie. zero delay */ 2276 2277 b->tin_rate_bps = rate; 2278 b->tin_rate_ns = rate_ns; 2279 b->tin_rate_shft = rate_shft; 2280 2281 byte_target_ns = (byte_target * rate_ns) >> rate_shft; 2282 2283 b->cparams.target = max((byte_target_ns * 3) / 2, target_ns); 2284 b->cparams.interval = max(rtt_est_ns + 2285 b->cparams.target - target_ns, 2286 b->cparams.target * 2); 2287 b->cparams.mtu_time = byte_target_ns; 2288 b->cparams.p_inc = 1 << 24; /* 1/256 */ 2289 b->cparams.p_dec = 1 << 20; /* 1/4096 */ 2290 } 2291 2292 static int cake_config_besteffort(struct Qdisc *sch) 2293 { 2294 struct cake_sched_data *q = qdisc_priv(sch); 2295 struct cake_tin_data *b = &q->tins[0]; 2296 u32 mtu = psched_mtu(qdisc_dev(sch)); 2297 u64 rate = q->rate_bps; 2298 2299 q->tin_cnt = 1; 2300 2301 q->tin_index = besteffort; 2302 q->tin_order = normal_order; 2303 2304 cake_set_rate(b, rate, mtu, 2305 us_to_ns(q->target), us_to_ns(q->interval)); 2306 b->tin_quantum = 65535; 2307 2308 return 0; 2309 } 2310 2311 static int cake_config_precedence(struct Qdisc *sch) 2312 { 2313 /* convert high-level (user visible) parameters into internal format */ 2314 struct cake_sched_data *q = qdisc_priv(sch); 2315 u32 mtu = psched_mtu(qdisc_dev(sch)); 2316 u64 rate = q->rate_bps; 2317 u32 quantum = 256; 2318 u32 i; 2319 2320 q->tin_cnt = 8; 2321 q->tin_index = precedence; 2322 q->tin_order = normal_order; 2323 2324 for (i = 0; i < q->tin_cnt; i++) { 2325 struct cake_tin_data *b = &q->tins[i]; 2326 2327 cake_set_rate(b, rate, mtu, us_to_ns(q->target), 2328 us_to_ns(q->interval)); 2329 2330 b->tin_quantum = max_t(u16, 1U, quantum); 2331 2332 /* calculate next class's parameters */ 2333 rate *= 7; 2334 rate >>= 3; 2335 2336 quantum *= 7; 2337 quantum >>= 3; 2338 } 2339 2340 return 0; 2341 } 2342 2343 /* List of known Diffserv codepoints: 2344 * 2345 * Least Effort (CS1, LE) 2346 * Best Effort (CS0) 2347 * Max Reliability & LLT "Lo" (TOS1) 2348 * Max Throughput (TOS2) 2349 * Min Delay (TOS4) 2350 * LLT "La" (TOS5) 2351 * Assured Forwarding 1 (AF1x) - x3 2352 * Assured Forwarding 2 (AF2x) - x3 2353 * Assured Forwarding 3 (AF3x) - x3 2354 * Assured Forwarding 4 (AF4x) - x3 2355 * Precedence Class 2 (CS2) 2356 * Precedence Class 3 (CS3) 2357 * Precedence Class 4 (CS4) 2358 * Precedence Class 5 (CS5) 2359 * Precedence Class 6 (CS6) 2360 * Precedence Class 7 (CS7) 2361 * Voice Admit (VA) 2362 * Expedited Forwarding (EF) 2363 2364 * Total 25 codepoints. 2365 */ 2366 2367 /* List of traffic classes in RFC 4594, updated by RFC 8622: 2368 * (roughly descending order of contended priority) 2369 * (roughly ascending order of uncontended throughput) 2370 * 2371 * Network Control (CS6,CS7) - routing traffic 2372 * Telephony (EF,VA) - aka. VoIP streams 2373 * Signalling (CS5) - VoIP setup 2374 * Multimedia Conferencing (AF4x) - aka. video calls 2375 * Realtime Interactive (CS4) - eg. games 2376 * Multimedia Streaming (AF3x) - eg. YouTube, NetFlix, Twitch 2377 * Broadcast Video (CS3) 2378 * Low Latency Data (AF2x,TOS4) - eg. database 2379 * Ops, Admin, Management (CS2,TOS1) - eg. ssh 2380 * Standard Service (CS0 & unrecognised codepoints) 2381 * High Throughput Data (AF1x,TOS2) - eg. web traffic 2382 * Low Priority Data (CS1,LE) - eg. BitTorrent 2383 2384 * Total 12 traffic classes. 2385 */ 2386 2387 static int cake_config_diffserv8(struct Qdisc *sch) 2388 { 2389 /* Pruned list of traffic classes for typical applications: 2390 * 2391 * Network Control (CS6, CS7) 2392 * Minimum Latency (EF, VA, CS5, CS4) 2393 * Interactive Shell (CS2, TOS1) 2394 * Low Latency Transactions (AF2x, TOS4) 2395 * Video Streaming (AF4x, AF3x, CS3) 2396 * Bog Standard (CS0 etc.) 2397 * High Throughput (AF1x, TOS2) 2398 * Background Traffic (CS1, LE) 2399 * 2400 * Total 8 traffic classes. 2401 */ 2402 2403 struct cake_sched_data *q = qdisc_priv(sch); 2404 u32 mtu = psched_mtu(qdisc_dev(sch)); 2405 u64 rate = q->rate_bps; 2406 u32 quantum = 256; 2407 u32 i; 2408 2409 q->tin_cnt = 8; 2410 2411 /* codepoint to class mapping */ 2412 q->tin_index = diffserv8; 2413 q->tin_order = normal_order; 2414 2415 /* class characteristics */ 2416 for (i = 0; i < q->tin_cnt; i++) { 2417 struct cake_tin_data *b = &q->tins[i]; 2418 2419 cake_set_rate(b, rate, mtu, us_to_ns(q->target), 2420 us_to_ns(q->interval)); 2421 2422 b->tin_quantum = max_t(u16, 1U, quantum); 2423 2424 /* calculate next class's parameters */ 2425 rate *= 7; 2426 rate >>= 3; 2427 2428 quantum *= 7; 2429 quantum >>= 3; 2430 } 2431 2432 return 0; 2433 } 2434 2435 static int cake_config_diffserv4(struct Qdisc *sch) 2436 { 2437 /* Further pruned list of traffic classes for four-class system: 2438 * 2439 * Latency Sensitive (CS7, CS6, EF, VA, CS5, CS4) 2440 * Streaming Media (AF4x, AF3x, CS3, AF2x, TOS4, CS2, TOS1) 2441 * Best Effort (CS0, AF1x, TOS2, and those not specified) 2442 * Background Traffic (CS1, LE) 2443 * 2444 * Total 4 traffic classes. 2445 */ 2446 2447 struct cake_sched_data *q = qdisc_priv(sch); 2448 u32 mtu = psched_mtu(qdisc_dev(sch)); 2449 u64 rate = q->rate_bps; 2450 u32 quantum = 1024; 2451 2452 q->tin_cnt = 4; 2453 2454 /* codepoint to class mapping */ 2455 q->tin_index = diffserv4; 2456 q->tin_order = bulk_order; 2457 2458 /* class characteristics */ 2459 cake_set_rate(&q->tins[0], rate, mtu, 2460 us_to_ns(q->target), us_to_ns(q->interval)); 2461 cake_set_rate(&q->tins[1], rate >> 4, mtu, 2462 us_to_ns(q->target), us_to_ns(q->interval)); 2463 cake_set_rate(&q->tins[2], rate >> 1, mtu, 2464 us_to_ns(q->target), us_to_ns(q->interval)); 2465 cake_set_rate(&q->tins[3], rate >> 2, mtu, 2466 us_to_ns(q->target), us_to_ns(q->interval)); 2467 2468 /* bandwidth-sharing weights */ 2469 q->tins[0].tin_quantum = quantum; 2470 q->tins[1].tin_quantum = quantum >> 4; 2471 q->tins[2].tin_quantum = quantum >> 1; 2472 q->tins[3].tin_quantum = quantum >> 2; 2473 2474 return 0; 2475 } 2476 2477 static int cake_config_diffserv3(struct Qdisc *sch) 2478 { 2479 /* Simplified Diffserv structure with 3 tins. 2480 * Low Priority (CS1, LE) 2481 * Best Effort 2482 * Latency Sensitive (TOS4, VA, EF, CS6, CS7) 2483 */ 2484 struct cake_sched_data *q = qdisc_priv(sch); 2485 u32 mtu = psched_mtu(qdisc_dev(sch)); 2486 u64 rate = q->rate_bps; 2487 u32 quantum = 1024; 2488 2489 q->tin_cnt = 3; 2490 2491 /* codepoint to class mapping */ 2492 q->tin_index = diffserv3; 2493 q->tin_order = bulk_order; 2494 2495 /* class characteristics */ 2496 cake_set_rate(&q->tins[0], rate, mtu, 2497 us_to_ns(q->target), us_to_ns(q->interval)); 2498 cake_set_rate(&q->tins[1], rate >> 4, mtu, 2499 us_to_ns(q->target), us_to_ns(q->interval)); 2500 cake_set_rate(&q->tins[2], rate >> 2, mtu, 2501 us_to_ns(q->target), us_to_ns(q->interval)); 2502 2503 /* bandwidth-sharing weights */ 2504 q->tins[0].tin_quantum = quantum; 2505 q->tins[1].tin_quantum = quantum >> 4; 2506 q->tins[2].tin_quantum = quantum >> 2; 2507 2508 return 0; 2509 } 2510 2511 static void cake_reconfigure(struct Qdisc *sch) 2512 { 2513 struct cake_sched_data *q = qdisc_priv(sch); 2514 int c, ft; 2515 2516 switch (q->tin_mode) { 2517 case CAKE_DIFFSERV_BESTEFFORT: 2518 ft = cake_config_besteffort(sch); 2519 break; 2520 2521 case CAKE_DIFFSERV_PRECEDENCE: 2522 ft = cake_config_precedence(sch); 2523 break; 2524 2525 case CAKE_DIFFSERV_DIFFSERV8: 2526 ft = cake_config_diffserv8(sch); 2527 break; 2528 2529 case CAKE_DIFFSERV_DIFFSERV4: 2530 ft = cake_config_diffserv4(sch); 2531 break; 2532 2533 case CAKE_DIFFSERV_DIFFSERV3: 2534 default: 2535 ft = cake_config_diffserv3(sch); 2536 break; 2537 } 2538 2539 for (c = q->tin_cnt; c < CAKE_MAX_TINS; c++) { 2540 cake_clear_tin(sch, c); 2541 q->tins[c].cparams.mtu_time = q->tins[ft].cparams.mtu_time; 2542 } 2543 2544 q->rate_ns = q->tins[ft].tin_rate_ns; 2545 q->rate_shft = q->tins[ft].tin_rate_shft; 2546 2547 if (q->buffer_config_limit) { 2548 q->buffer_limit = q->buffer_config_limit; 2549 } else if (q->rate_bps) { 2550 u64 t = q->rate_bps * q->interval; 2551 2552 do_div(t, USEC_PER_SEC / 4); 2553 q->buffer_limit = max_t(u32, t, 4U << 20); 2554 } else { 2555 q->buffer_limit = ~0; 2556 } 2557 2558 sch->flags &= ~TCQ_F_CAN_BYPASS; 2559 2560 q->buffer_limit = min(q->buffer_limit, 2561 max(sch->limit * psched_mtu(qdisc_dev(sch)), 2562 q->buffer_config_limit)); 2563 } 2564 2565 static int cake_change(struct Qdisc *sch, struct nlattr *opt, 2566 struct netlink_ext_ack *extack) 2567 { 2568 struct cake_sched_data *q = qdisc_priv(sch); 2569 struct nlattr *tb[TCA_CAKE_MAX + 1]; 2570 int err; 2571 2572 if (!opt) 2573 return -EINVAL; 2574 2575 err = nla_parse_nested_deprecated(tb, TCA_CAKE_MAX, opt, cake_policy, 2576 extack); 2577 if (err < 0) 2578 return err; 2579 2580 if (tb[TCA_CAKE_NAT]) { 2581 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 2582 q->flow_mode &= ~CAKE_FLOW_NAT_FLAG; 2583 q->flow_mode |= CAKE_FLOW_NAT_FLAG * 2584 !!nla_get_u32(tb[TCA_CAKE_NAT]); 2585 #else 2586 NL_SET_ERR_MSG_ATTR(extack, tb[TCA_CAKE_NAT], 2587 "No conntrack support in kernel"); 2588 return -EOPNOTSUPP; 2589 #endif 2590 } 2591 2592 if (tb[TCA_CAKE_BASE_RATE64]) 2593 q->rate_bps = nla_get_u64(tb[TCA_CAKE_BASE_RATE64]); 2594 2595 if (tb[TCA_CAKE_DIFFSERV_MODE]) 2596 q->tin_mode = nla_get_u32(tb[TCA_CAKE_DIFFSERV_MODE]); 2597 2598 if (tb[TCA_CAKE_WASH]) { 2599 if (!!nla_get_u32(tb[TCA_CAKE_WASH])) 2600 q->rate_flags |= CAKE_FLAG_WASH; 2601 else 2602 q->rate_flags &= ~CAKE_FLAG_WASH; 2603 } 2604 2605 if (tb[TCA_CAKE_FLOW_MODE]) 2606 q->flow_mode = ((q->flow_mode & CAKE_FLOW_NAT_FLAG) | 2607 (nla_get_u32(tb[TCA_CAKE_FLOW_MODE]) & 2608 CAKE_FLOW_MASK)); 2609 2610 if (tb[TCA_CAKE_ATM]) 2611 q->atm_mode = nla_get_u32(tb[TCA_CAKE_ATM]); 2612 2613 if (tb[TCA_CAKE_OVERHEAD]) { 2614 q->rate_overhead = nla_get_s32(tb[TCA_CAKE_OVERHEAD]); 2615 q->rate_flags |= CAKE_FLAG_OVERHEAD; 2616 2617 q->max_netlen = 0; 2618 q->max_adjlen = 0; 2619 q->min_netlen = ~0; 2620 q->min_adjlen = ~0; 2621 } 2622 2623 if (tb[TCA_CAKE_RAW]) { 2624 q->rate_flags &= ~CAKE_FLAG_OVERHEAD; 2625 2626 q->max_netlen = 0; 2627 q->max_adjlen = 0; 2628 q->min_netlen = ~0; 2629 q->min_adjlen = ~0; 2630 } 2631 2632 if (tb[TCA_CAKE_MPU]) 2633 q->rate_mpu = nla_get_u32(tb[TCA_CAKE_MPU]); 2634 2635 if (tb[TCA_CAKE_RTT]) { 2636 q->interval = nla_get_u32(tb[TCA_CAKE_RTT]); 2637 2638 if (!q->interval) 2639 q->interval = 1; 2640 } 2641 2642 if (tb[TCA_CAKE_TARGET]) { 2643 q->target = nla_get_u32(tb[TCA_CAKE_TARGET]); 2644 2645 if (!q->target) 2646 q->target = 1; 2647 } 2648 2649 if (tb[TCA_CAKE_AUTORATE]) { 2650 if (!!nla_get_u32(tb[TCA_CAKE_AUTORATE])) 2651 q->rate_flags |= CAKE_FLAG_AUTORATE_INGRESS; 2652 else 2653 q->rate_flags &= ~CAKE_FLAG_AUTORATE_INGRESS; 2654 } 2655 2656 if (tb[TCA_CAKE_INGRESS]) { 2657 if (!!nla_get_u32(tb[TCA_CAKE_INGRESS])) 2658 q->rate_flags |= CAKE_FLAG_INGRESS; 2659 else 2660 q->rate_flags &= ~CAKE_FLAG_INGRESS; 2661 } 2662 2663 if (tb[TCA_CAKE_ACK_FILTER]) 2664 q->ack_filter = nla_get_u32(tb[TCA_CAKE_ACK_FILTER]); 2665 2666 if (tb[TCA_CAKE_MEMORY]) 2667 q->buffer_config_limit = nla_get_u32(tb[TCA_CAKE_MEMORY]); 2668 2669 if (tb[TCA_CAKE_SPLIT_GSO]) { 2670 if (!!nla_get_u32(tb[TCA_CAKE_SPLIT_GSO])) 2671 q->rate_flags |= CAKE_FLAG_SPLIT_GSO; 2672 else 2673 q->rate_flags &= ~CAKE_FLAG_SPLIT_GSO; 2674 } 2675 2676 if (tb[TCA_CAKE_FWMARK]) { 2677 q->fwmark_mask = nla_get_u32(tb[TCA_CAKE_FWMARK]); 2678 q->fwmark_shft = q->fwmark_mask ? __ffs(q->fwmark_mask) : 0; 2679 } 2680 2681 if (q->tins) { 2682 sch_tree_lock(sch); 2683 cake_reconfigure(sch); 2684 sch_tree_unlock(sch); 2685 } 2686 2687 return 0; 2688 } 2689 2690 static void cake_destroy(struct Qdisc *sch) 2691 { 2692 struct cake_sched_data *q = qdisc_priv(sch); 2693 2694 qdisc_watchdog_cancel(&q->watchdog); 2695 tcf_block_put(q->block); 2696 kvfree(q->tins); 2697 } 2698 2699 static int cake_init(struct Qdisc *sch, struct nlattr *opt, 2700 struct netlink_ext_ack *extack) 2701 { 2702 struct cake_sched_data *q = qdisc_priv(sch); 2703 int i, j, err; 2704 2705 sch->limit = 10240; 2706 q->tin_mode = CAKE_DIFFSERV_DIFFSERV3; 2707 q->flow_mode = CAKE_FLOW_TRIPLE; 2708 2709 q->rate_bps = 0; /* unlimited by default */ 2710 2711 q->interval = 100000; /* 100ms default */ 2712 q->target = 5000; /* 5ms: codel RFC argues 2713 * for 5 to 10% of interval 2714 */ 2715 q->rate_flags |= CAKE_FLAG_SPLIT_GSO; 2716 q->cur_tin = 0; 2717 q->cur_flow = 0; 2718 2719 qdisc_watchdog_init(&q->watchdog, sch); 2720 2721 if (opt) { 2722 err = cake_change(sch, opt, extack); 2723 2724 if (err) 2725 return err; 2726 } 2727 2728 err = tcf_block_get(&q->block, &q->filter_list, sch, extack); 2729 if (err) 2730 return err; 2731 2732 quantum_div[0] = ~0; 2733 for (i = 1; i <= CAKE_QUEUES; i++) 2734 quantum_div[i] = 65535 / i; 2735 2736 q->tins = kvcalloc(CAKE_MAX_TINS, sizeof(struct cake_tin_data), 2737 GFP_KERNEL); 2738 if (!q->tins) 2739 goto nomem; 2740 2741 for (i = 0; i < CAKE_MAX_TINS; i++) { 2742 struct cake_tin_data *b = q->tins + i; 2743 2744 INIT_LIST_HEAD(&b->new_flows); 2745 INIT_LIST_HEAD(&b->old_flows); 2746 INIT_LIST_HEAD(&b->decaying_flows); 2747 b->sparse_flow_count = 0; 2748 b->bulk_flow_count = 0; 2749 b->decaying_flow_count = 0; 2750 2751 for (j = 0; j < CAKE_QUEUES; j++) { 2752 struct cake_flow *flow = b->flows + j; 2753 u32 k = j * CAKE_MAX_TINS + i; 2754 2755 INIT_LIST_HEAD(&flow->flowchain); 2756 cobalt_vars_init(&flow->cvars); 2757 2758 q->overflow_heap[k].t = i; 2759 q->overflow_heap[k].b = j; 2760 b->overflow_idx[j] = k; 2761 } 2762 } 2763 2764 cake_reconfigure(sch); 2765 q->avg_peak_bandwidth = q->rate_bps; 2766 q->min_netlen = ~0; 2767 q->min_adjlen = ~0; 2768 return 0; 2769 2770 nomem: 2771 cake_destroy(sch); 2772 return -ENOMEM; 2773 } 2774 2775 static int cake_dump(struct Qdisc *sch, struct sk_buff *skb) 2776 { 2777 struct cake_sched_data *q = qdisc_priv(sch); 2778 struct nlattr *opts; 2779 2780 opts = nla_nest_start_noflag(skb, TCA_OPTIONS); 2781 if (!opts) 2782 goto nla_put_failure; 2783 2784 if (nla_put_u64_64bit(skb, TCA_CAKE_BASE_RATE64, q->rate_bps, 2785 TCA_CAKE_PAD)) 2786 goto nla_put_failure; 2787 2788 if (nla_put_u32(skb, TCA_CAKE_FLOW_MODE, 2789 q->flow_mode & CAKE_FLOW_MASK)) 2790 goto nla_put_failure; 2791 2792 if (nla_put_u32(skb, TCA_CAKE_RTT, q->interval)) 2793 goto nla_put_failure; 2794 2795 if (nla_put_u32(skb, TCA_CAKE_TARGET, q->target)) 2796 goto nla_put_failure; 2797 2798 if (nla_put_u32(skb, TCA_CAKE_MEMORY, q->buffer_config_limit)) 2799 goto nla_put_failure; 2800 2801 if (nla_put_u32(skb, TCA_CAKE_AUTORATE, 2802 !!(q->rate_flags & CAKE_FLAG_AUTORATE_INGRESS))) 2803 goto nla_put_failure; 2804 2805 if (nla_put_u32(skb, TCA_CAKE_INGRESS, 2806 !!(q->rate_flags & CAKE_FLAG_INGRESS))) 2807 goto nla_put_failure; 2808 2809 if (nla_put_u32(skb, TCA_CAKE_ACK_FILTER, q->ack_filter)) 2810 goto nla_put_failure; 2811 2812 if (nla_put_u32(skb, TCA_CAKE_NAT, 2813 !!(q->flow_mode & CAKE_FLOW_NAT_FLAG))) 2814 goto nla_put_failure; 2815 2816 if (nla_put_u32(skb, TCA_CAKE_DIFFSERV_MODE, q->tin_mode)) 2817 goto nla_put_failure; 2818 2819 if (nla_put_u32(skb, TCA_CAKE_WASH, 2820 !!(q->rate_flags & CAKE_FLAG_WASH))) 2821 goto nla_put_failure; 2822 2823 if (nla_put_u32(skb, TCA_CAKE_OVERHEAD, q->rate_overhead)) 2824 goto nla_put_failure; 2825 2826 if (!(q->rate_flags & CAKE_FLAG_OVERHEAD)) 2827 if (nla_put_u32(skb, TCA_CAKE_RAW, 0)) 2828 goto nla_put_failure; 2829 2830 if (nla_put_u32(skb, TCA_CAKE_ATM, q->atm_mode)) 2831 goto nla_put_failure; 2832 2833 if (nla_put_u32(skb, TCA_CAKE_MPU, q->rate_mpu)) 2834 goto nla_put_failure; 2835 2836 if (nla_put_u32(skb, TCA_CAKE_SPLIT_GSO, 2837 !!(q->rate_flags & CAKE_FLAG_SPLIT_GSO))) 2838 goto nla_put_failure; 2839 2840 if (nla_put_u32(skb, TCA_CAKE_FWMARK, q->fwmark_mask)) 2841 goto nla_put_failure; 2842 2843 return nla_nest_end(skb, opts); 2844 2845 nla_put_failure: 2846 return -1; 2847 } 2848 2849 static int cake_dump_stats(struct Qdisc *sch, struct gnet_dump *d) 2850 { 2851 struct nlattr *stats = nla_nest_start_noflag(d->skb, TCA_STATS_APP); 2852 struct cake_sched_data *q = qdisc_priv(sch); 2853 struct nlattr *tstats, *ts; 2854 int i; 2855 2856 if (!stats) 2857 return -1; 2858 2859 #define PUT_STAT_U32(attr, data) do { \ 2860 if (nla_put_u32(d->skb, TCA_CAKE_STATS_ ## attr, data)) \ 2861 goto nla_put_failure; \ 2862 } while (0) 2863 #define PUT_STAT_U64(attr, data) do { \ 2864 if (nla_put_u64_64bit(d->skb, TCA_CAKE_STATS_ ## attr, \ 2865 data, TCA_CAKE_STATS_PAD)) \ 2866 goto nla_put_failure; \ 2867 } while (0) 2868 2869 PUT_STAT_U64(CAPACITY_ESTIMATE64, q->avg_peak_bandwidth); 2870 PUT_STAT_U32(MEMORY_LIMIT, q->buffer_limit); 2871 PUT_STAT_U32(MEMORY_USED, q->buffer_max_used); 2872 PUT_STAT_U32(AVG_NETOFF, ((q->avg_netoff + 0x8000) >> 16)); 2873 PUT_STAT_U32(MAX_NETLEN, q->max_netlen); 2874 PUT_STAT_U32(MAX_ADJLEN, q->max_adjlen); 2875 PUT_STAT_U32(MIN_NETLEN, q->min_netlen); 2876 PUT_STAT_U32(MIN_ADJLEN, q->min_adjlen); 2877 2878 #undef PUT_STAT_U32 2879 #undef PUT_STAT_U64 2880 2881 tstats = nla_nest_start_noflag(d->skb, TCA_CAKE_STATS_TIN_STATS); 2882 if (!tstats) 2883 goto nla_put_failure; 2884 2885 #define PUT_TSTAT_U32(attr, data) do { \ 2886 if (nla_put_u32(d->skb, TCA_CAKE_TIN_STATS_ ## attr, data)) \ 2887 goto nla_put_failure; \ 2888 } while (0) 2889 #define PUT_TSTAT_U64(attr, data) do { \ 2890 if (nla_put_u64_64bit(d->skb, TCA_CAKE_TIN_STATS_ ## attr, \ 2891 data, TCA_CAKE_TIN_STATS_PAD)) \ 2892 goto nla_put_failure; \ 2893 } while (0) 2894 2895 for (i = 0; i < q->tin_cnt; i++) { 2896 struct cake_tin_data *b = &q->tins[q->tin_order[i]]; 2897 2898 ts = nla_nest_start_noflag(d->skb, i + 1); 2899 if (!ts) 2900 goto nla_put_failure; 2901 2902 PUT_TSTAT_U64(THRESHOLD_RATE64, b->tin_rate_bps); 2903 PUT_TSTAT_U64(SENT_BYTES64, b->bytes); 2904 PUT_TSTAT_U32(BACKLOG_BYTES, b->tin_backlog); 2905 2906 PUT_TSTAT_U32(TARGET_US, 2907 ktime_to_us(ns_to_ktime(b->cparams.target))); 2908 PUT_TSTAT_U32(INTERVAL_US, 2909 ktime_to_us(ns_to_ktime(b->cparams.interval))); 2910 2911 PUT_TSTAT_U32(SENT_PACKETS, b->packets); 2912 PUT_TSTAT_U32(DROPPED_PACKETS, b->tin_dropped); 2913 PUT_TSTAT_U32(ECN_MARKED_PACKETS, b->tin_ecn_mark); 2914 PUT_TSTAT_U32(ACKS_DROPPED_PACKETS, b->ack_drops); 2915 2916 PUT_TSTAT_U32(PEAK_DELAY_US, 2917 ktime_to_us(ns_to_ktime(b->peak_delay))); 2918 PUT_TSTAT_U32(AVG_DELAY_US, 2919 ktime_to_us(ns_to_ktime(b->avge_delay))); 2920 PUT_TSTAT_U32(BASE_DELAY_US, 2921 ktime_to_us(ns_to_ktime(b->base_delay))); 2922 2923 PUT_TSTAT_U32(WAY_INDIRECT_HITS, b->way_hits); 2924 PUT_TSTAT_U32(WAY_MISSES, b->way_misses); 2925 PUT_TSTAT_U32(WAY_COLLISIONS, b->way_collisions); 2926 2927 PUT_TSTAT_U32(SPARSE_FLOWS, b->sparse_flow_count + 2928 b->decaying_flow_count); 2929 PUT_TSTAT_U32(BULK_FLOWS, b->bulk_flow_count); 2930 PUT_TSTAT_U32(UNRESPONSIVE_FLOWS, b->unresponsive_flow_count); 2931 PUT_TSTAT_U32(MAX_SKBLEN, b->max_skblen); 2932 2933 PUT_TSTAT_U32(FLOW_QUANTUM, b->flow_quantum); 2934 nla_nest_end(d->skb, ts); 2935 } 2936 2937 #undef PUT_TSTAT_U32 2938 #undef PUT_TSTAT_U64 2939 2940 nla_nest_end(d->skb, tstats); 2941 return nla_nest_end(d->skb, stats); 2942 2943 nla_put_failure: 2944 nla_nest_cancel(d->skb, stats); 2945 return -1; 2946 } 2947 2948 static struct Qdisc *cake_leaf(struct Qdisc *sch, unsigned long arg) 2949 { 2950 return NULL; 2951 } 2952 2953 static unsigned long cake_find(struct Qdisc *sch, u32 classid) 2954 { 2955 return 0; 2956 } 2957 2958 static unsigned long cake_bind(struct Qdisc *sch, unsigned long parent, 2959 u32 classid) 2960 { 2961 return 0; 2962 } 2963 2964 static void cake_unbind(struct Qdisc *q, unsigned long cl) 2965 { 2966 } 2967 2968 static struct tcf_block *cake_tcf_block(struct Qdisc *sch, unsigned long cl, 2969 struct netlink_ext_ack *extack) 2970 { 2971 struct cake_sched_data *q = qdisc_priv(sch); 2972 2973 if (cl) 2974 return NULL; 2975 return q->block; 2976 } 2977 2978 static int cake_dump_class(struct Qdisc *sch, unsigned long cl, 2979 struct sk_buff *skb, struct tcmsg *tcm) 2980 { 2981 tcm->tcm_handle |= TC_H_MIN(cl); 2982 return 0; 2983 } 2984 2985 static int cake_dump_class_stats(struct Qdisc *sch, unsigned long cl, 2986 struct gnet_dump *d) 2987 { 2988 struct cake_sched_data *q = qdisc_priv(sch); 2989 const struct cake_flow *flow = NULL; 2990 struct gnet_stats_queue qs = { 0 }; 2991 struct nlattr *stats; 2992 u32 idx = cl - 1; 2993 2994 if (idx < CAKE_QUEUES * q->tin_cnt) { 2995 const struct cake_tin_data *b = \ 2996 &q->tins[q->tin_order[idx / CAKE_QUEUES]]; 2997 const struct sk_buff *skb; 2998 2999 flow = &b->flows[idx % CAKE_QUEUES]; 3000 3001 if (flow->head) { 3002 sch_tree_lock(sch); 3003 skb = flow->head; 3004 while (skb) { 3005 qs.qlen++; 3006 skb = skb->next; 3007 } 3008 sch_tree_unlock(sch); 3009 } 3010 qs.backlog = b->backlogs[idx % CAKE_QUEUES]; 3011 qs.drops = flow->dropped; 3012 } 3013 if (gnet_stats_copy_queue(d, NULL, &qs, qs.qlen) < 0) 3014 return -1; 3015 if (flow) { 3016 ktime_t now = ktime_get(); 3017 3018 stats = nla_nest_start_noflag(d->skb, TCA_STATS_APP); 3019 if (!stats) 3020 return -1; 3021 3022 #define PUT_STAT_U32(attr, data) do { \ 3023 if (nla_put_u32(d->skb, TCA_CAKE_STATS_ ## attr, data)) \ 3024 goto nla_put_failure; \ 3025 } while (0) 3026 #define PUT_STAT_S32(attr, data) do { \ 3027 if (nla_put_s32(d->skb, TCA_CAKE_STATS_ ## attr, data)) \ 3028 goto nla_put_failure; \ 3029 } while (0) 3030 3031 PUT_STAT_S32(DEFICIT, flow->deficit); 3032 PUT_STAT_U32(DROPPING, flow->cvars.dropping); 3033 PUT_STAT_U32(COBALT_COUNT, flow->cvars.count); 3034 PUT_STAT_U32(P_DROP, flow->cvars.p_drop); 3035 if (flow->cvars.p_drop) { 3036 PUT_STAT_S32(BLUE_TIMER_US, 3037 ktime_to_us( 3038 ktime_sub(now, 3039 flow->cvars.blue_timer))); 3040 } 3041 if (flow->cvars.dropping) { 3042 PUT_STAT_S32(DROP_NEXT_US, 3043 ktime_to_us( 3044 ktime_sub(now, 3045 flow->cvars.drop_next))); 3046 } 3047 3048 if (nla_nest_end(d->skb, stats) < 0) 3049 return -1; 3050 } 3051 3052 return 0; 3053 3054 nla_put_failure: 3055 nla_nest_cancel(d->skb, stats); 3056 return -1; 3057 } 3058 3059 static void cake_walk(struct Qdisc *sch, struct qdisc_walker *arg) 3060 { 3061 struct cake_sched_data *q = qdisc_priv(sch); 3062 unsigned int i, j; 3063 3064 if (arg->stop) 3065 return; 3066 3067 for (i = 0; i < q->tin_cnt; i++) { 3068 struct cake_tin_data *b = &q->tins[q->tin_order[i]]; 3069 3070 for (j = 0; j < CAKE_QUEUES; j++) { 3071 if (list_empty(&b->flows[j].flowchain) || 3072 arg->count < arg->skip) { 3073 arg->count++; 3074 continue; 3075 } 3076 if (arg->fn(sch, i * CAKE_QUEUES + j + 1, arg) < 0) { 3077 arg->stop = 1; 3078 break; 3079 } 3080 arg->count++; 3081 } 3082 } 3083 } 3084 3085 static const struct Qdisc_class_ops cake_class_ops = { 3086 .leaf = cake_leaf, 3087 .find = cake_find, 3088 .tcf_block = cake_tcf_block, 3089 .bind_tcf = cake_bind, 3090 .unbind_tcf = cake_unbind, 3091 .dump = cake_dump_class, 3092 .dump_stats = cake_dump_class_stats, 3093 .walk = cake_walk, 3094 }; 3095 3096 static struct Qdisc_ops cake_qdisc_ops __read_mostly = { 3097 .cl_ops = &cake_class_ops, 3098 .id = "cake", 3099 .priv_size = sizeof(struct cake_sched_data), 3100 .enqueue = cake_enqueue, 3101 .dequeue = cake_dequeue, 3102 .peek = qdisc_peek_dequeued, 3103 .init = cake_init, 3104 .reset = cake_reset, 3105 .destroy = cake_destroy, 3106 .change = cake_change, 3107 .dump = cake_dump, 3108 .dump_stats = cake_dump_stats, 3109 .owner = THIS_MODULE, 3110 }; 3111 3112 static int __init cake_module_init(void) 3113 { 3114 return register_qdisc(&cake_qdisc_ops); 3115 } 3116 3117 static void __exit cake_module_exit(void) 3118 { 3119 unregister_qdisc(&cake_qdisc_ops); 3120 } 3121 3122 module_init(cake_module_init) 3123 module_exit(cake_module_exit) 3124 MODULE_AUTHOR("Jonathan Morton"); 3125 MODULE_LICENSE("Dual BSD/GPL"); 3126 MODULE_DESCRIPTION("The CAKE shaper."); 3127