1 /* 2 * Copyright (c) 1998-2002 Luigi Rizzo, Universita` di Pisa 3 * Portions Copyright (c) 2000 Akamba Corp. 4 * All rights reserved 5 * 6 * Redistribution and use in source and binary forms, with or without 7 * modification, are permitted provided that the following conditions 8 * are met: 9 * 1. Redistributions of source code must retain the above copyright 10 * notice, this list of conditions and the following disclaimer. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 16 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 17 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 18 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 21 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 25 * SUCH DAMAGE. 26 * 27 * $FreeBSD: src/sys/netinet/ip_dummynet.h,v 1.10.2.9 2003/05/13 09:31:06 maxim Exp $ 28 * $DragonFly: src/sys/net/dummynet/ip_dummynet.h,v 1.18 2007/11/18 13:00:28 sephe Exp $ 29 */ 30 31 #ifndef _IP_DUMMYNET_H 32 #define _IP_DUMMYNET_H 33 34 /* 35 * We start with a heap, which is used in the scheduler to decide when to 36 * transmit packets etc. 37 * 38 * The key for the heap is used for two different values: 39 * 40 * 1. Timer ticks- max 10K/second, so 32 bits are enough; 41 * 42 * 2. Virtual times. These increase in steps of len/x, where len is the 43 * packet length, and x is either the weight of the flow, or the sum 44 * of all weights. 45 * If we limit to max 1000 flows and a max weight of 100, then x needs 46 * 17 bits. The packet size is 16 bits, so we can easily overflow if 47 * we do not allow errors. 48 * 49 * So we use a key "dn_key" which is 64 bits. 50 * 51 * MY_M is used as a shift count when doing fixed point arithmetic 52 * (a better name would be useful...). 53 */ 54 typedef uint64_t dn_key; /* sorting key */ 55 56 /* 57 * Number of left shift to obtain a larger precision 58 * 59 * XXX With this scaling, max 1000 flows, max weight 100, 1Gbit/s, the 60 * virtual time wraps every 15 days. 61 */ 62 #define MY_M 16 63 64 #ifdef _KERNEL 65 66 /* 67 * A heap entry is made of a key and a pointer to the actual object stored 68 * in the heap. 69 * 70 * The heap is an array of dn_heap_entry entries, dynamically allocated. 71 * Current size is "size", with "elements" actually in use. 72 * 73 * The heap normally supports only ordered insert and extract from the top. 74 * If we want to extract an object from the middle of the heap, we have to 75 * know where the object itself is located in the heap (or we need to scan 76 * the whole array). To this purpose, an object has a field (int) which 77 * contains the index of the object itself into the heap. When the object 78 * is moved, the field must also be updated. The offset of the index in the 79 * object is stored in the 'offset' field in the heap descriptor. The 80 * assumption is that this offset is non-zero if we want to support extract 81 * from the middle. 82 */ 83 struct dn_heap_entry { 84 dn_key key; /* sorting key. Topmost element is smallest one */ 85 void *object; /* object pointer */ 86 }; 87 88 struct dn_heap { 89 int size; 90 int elements; 91 int offset; /* XXX if > 0 this is the offset of direct ptr to obj */ 92 struct dn_heap_entry *p; /* really an array of "size" entries */ 93 }; 94 95 struct dn_flow_id { 96 uint16_t fid_type; /* ETHERTYPE_ */ 97 uint16_t pad; 98 union { 99 struct { 100 uint32_t dst_ip; 101 uint32_t src_ip; 102 uint16_t dst_port; 103 uint16_t src_port; 104 uint8_t proto; 105 uint8_t flags; 106 } inet; 107 } fid_u; 108 #define fid_dst_ip fid_u.inet.dst_ip 109 #define fid_src_ip fid_u.inet.src_ip 110 #define fid_dst_port fid_u.inet.dst_port 111 #define fid_src_port fid_u.inet.src_port 112 #define fid_proto fid_u.inet.proto 113 #define fid_flags fid_u.inet.flags 114 }; 115 116 typedef void (*ip_dn_unref_priv_t)(void *); 117 118 /* 119 * struct dn_pkt identifies a packet in the dummynet queue, but is also used 120 * to tag packets passed back to the various destinations (ip_input(), 121 * ip_output() and so on). 122 * 123 * It is a tag (PACKET_TAG_DUMMYNET) associated with the actual mbuf. 124 */ 125 struct dn_pkt { 126 struct mbuf *dn_m; 127 TAILQ_ENTRY(dn_pkt) dn_next; 128 129 void *dn_priv; 130 ip_dn_unref_priv_t dn_unref_priv; 131 132 uint32_t dn_flags; /* action when packet comes out. */ 133 #define DN_FLAGS_IS_PIPE 0x10 134 #define DN_FLAGS_DIR_MASK 0x0f 135 #define DN_TO_IP_OUT 1 136 #define DN_TO_IP_IN 2 137 #define DN_TO_ETH_DEMUX 4 138 #define DN_TO_ETH_OUT 5 139 #define DN_TO_MAX 6 140 141 dn_key output_time; /* when the pkt is due for delivery */ 142 struct ifnet *ifp; /* interface, for ip_output */ 143 struct sockaddr_in *dn_dst; 144 struct route ro; /* route, for ip_output. MUST COPY */ 145 int flags; /* flags, for ip_output (IPv6 ?) */ 146 147 u_short pipe_nr; /* pipe/flow_set number */ 148 u_short pad; 149 150 struct dn_flow_id id; /* flow id */ 151 int cpuid; /* target cpu, for IP_OUT/ETH_DEMUX/ETH_OUT */ 152 }; 153 TAILQ_HEAD(dn_pkt_queue, dn_pkt); 154 155 /* 156 * Overall structure of dummynet (with WF2Q+): 157 * 158 * In dummynet, packets are selected with the firewall rules, and passed to 159 * two different objects: PIPE or QUEUE. 160 * 161 * A QUEUE is just a queue with configurable size and queue management policy. 162 * It is also associated with a mask (to discriminate among different flows), 163 * a weight (used to give different shares of the bandwidth to different flows) 164 * and a "pipe", which essentially supplies the transmit clock for all queues 165 * associated with that pipe. 166 * 167 * A PIPE emulates a fixed-bandwidth link, whose bandwidth is configurable. 168 * The "clock" for a pipe comes from an internal timer. A pipe is also 169 * associated with one (or more, if masks are used) queue, where all packets 170 * for that pipe are stored. 171 * 172 * The bandwidth available on the pipe is shared by the queues associated with 173 * that pipe (only one in case the packet is sent to a PIPE) according to the 174 * WF2Q+ scheduling algorithm and the configured weights. 175 * 176 * In general, incoming packets are stored in the appropriate queue, which is 177 * then placed into one of a few heaps managed by a scheduler to decide when 178 * the packet should be extracted. The scheduler (a function called dummynet()) 179 * is run at every timer tick, and grabs queues from the head of the heaps when 180 * they are ready for processing. 181 * 182 * There are three data structures definining a pipe and associated queues: 183 * 184 * + dn_pipe, which contains the main configuration parameters related to 185 * delay and bandwidth; 186 * + dn_flow_set, which contains WF2Q+ configuration, flow masks, plr and 187 * RED configuration; 188 * + dn_flow_queue, which is the per-flow queue (containing the packets) 189 * 190 * Multiple dn_flow_set can be linked to the same pipe, and multiple 191 * dn_flow_queue can be linked to the same dn_flow_set. 192 * All data structures are linked in a linear list which is used for 193 * housekeeping purposes. 194 * 195 * During configuration, we create and initialize the dn_flow_set and dn_pipe 196 * structures (a dn_pipe also contains a dn_flow_set). 197 * 198 * At runtime: packets are sent to the appropriate dn_flow_set (either WFQ 199 * ones, or the one embedded in the dn_pipe for fixed-rate flows), which in 200 * turn dispatches them to the appropriate dn_flow_queue (created dynamically 201 * according to the masks). 202 * 203 * The transmit clock for fixed rate flows (ready_event()) selects the 204 * dn_flow_queue to be used to transmit the next packet. For WF2Q, 205 * wfq_ready_event() extract a pipe which in turn selects the right flow using 206 * a number of heaps defined into the pipe itself. 207 */ 208 209 /* 210 * Per flow queue. This contains the flow identifier, the queue of packets, 211 * counters, and parameters used to support both RED and WF2Q+. 212 * 213 * A dn_flow_queue is created and initialized whenever a packet for a new 214 * flow arrives. 215 */ 216 struct dn_flow_queue { 217 struct dn_flow_id id; 218 LIST_ENTRY(dn_flow_queue) q_link; 219 220 struct dn_pkt_queue queue; /* queue of packets */ 221 u_int len; 222 u_int len_bytes; 223 u_long numbytes; /* credit for transmission (dynamic queues) */ 224 225 uint64_t tot_pkts; /* statistics counters */ 226 uint64_t tot_bytes; 227 uint32_t drops; 228 229 int hash_slot; /* debugging/diagnostic */ 230 231 /* RED parameters */ 232 int avg; /* average queue length est. (scaled) */ 233 int count; /* arrivals since last RED drop */ 234 int random; /* random value (scaled) */ 235 uint32_t q_time; /* start of queue idle time */ 236 237 /* WF2Q+ support */ 238 struct dn_flow_set *fs; /* parent flow set */ 239 int heap_pos; /* position (index) of struct in heap */ 240 dn_key sched_time; /* current time when queue enters ready_heap */ 241 242 dn_key S, F; /* start time, finish time */ 243 /* 244 * Setting F < S means the timestamp is invalid. We only need 245 * to test this when the queue is empty. 246 */ 247 }; 248 LIST_HEAD(dn_flowqueue_head, dn_flow_queue); 249 250 /* 251 * flow_set descriptor. Contains the "template" parameters for the queue 252 * configuration, and pointers to the hash table of dn_flow_queue's. 253 * 254 * The hash table is an array of lists -- we identify the slot by hashing 255 * the flow-id, then scan the list looking for a match. 256 * The size of the hash table (buckets) is configurable on a per-queue basis. 257 * 258 * A dn_flow_set is created whenever a new queue or pipe is created (in the 259 * latter case, the structure is located inside the struct dn_pipe). 260 */ 261 struct dn_flow_set { 262 u_short fs_nr; /* flow_set number */ 263 u_short flags_fs; /* see 'Flow set flags' */ 264 265 LIST_ENTRY(dn_flow_set) fs_link; 266 267 struct dn_pipe *pipe; /* pointer to parent pipe */ 268 u_short parent_nr; /* parent pipe#, 0 if local to a pipe */ 269 270 int weight; /* WFQ queue weight */ 271 int qsize; /* queue size in slots or bytes */ 272 int plr; /* pkt loss rate (2^31-1 means 100%) */ 273 274 struct dn_flow_id flow_mask; 275 276 /* hash table of queues onto this flow_set */ 277 int rq_size; /* number of slots */ 278 int rq_elements; /* active elements */ 279 struct dn_flowqueue_head *rq;/* array of rq_size entries */ 280 281 uint32_t last_expired; /* do not expire too frequently */ 282 int backlogged; /* #active queues for this flowset */ 283 284 /* RED parameters */ 285 int w_q; /* queue weight (scaled) */ 286 int max_th; /* maximum threshold for queue (scaled) */ 287 int min_th; /* minimum threshold for queue (scaled) */ 288 int max_p; /* maximum value for p_b (scaled) */ 289 u_int c_1; /* max_p/(max_th-min_th) (scaled) */ 290 u_int c_2; /* max_p*min_th/(max_th-min_th) (scaled) */ 291 u_int c_3; /* for GRED, (1-max_p)/max_th (scaled) */ 292 u_int c_4; /* for GRED, 1 - 2*max_p (scaled) */ 293 u_int *w_q_lookup; /* lookup table for computing (1-w_q)^t */ 294 u_int lookup_depth; /* depth of lookup table */ 295 int lookup_step; /* granularity inside the lookup table */ 296 int lookup_weight; /* equal to (1-w_q)^t / (1-w_q)^(t+1) */ 297 int avg_pkt_size; /* medium packet size */ 298 int max_pkt_size; /* max packet size */ 299 }; 300 LIST_HEAD(dn_flowset_head, dn_flow_set); 301 302 /* 303 * Pipe descriptor. Contains global parameters, delay-line queue, and the 304 * flow_set used for fixed-rate queues. 305 * 306 * For WF2Q+ support it also has 3 heaps holding dn_flow_queue: 307 * + not_eligible_heap, for queues whose start time is higher than the 308 * virtual time. Sorted by start time. 309 * + scheduler_heap, for queues eligible for scheduling. Sorted by finish 310 * time. 311 * + idle_heap, all flows that are idle and can be removed. We do that on 312 * each tick so we do not slow down too much operations during forwarding. 313 */ 314 struct dn_pipe { /* a pipe */ 315 int pipe_nr; /* number */ 316 int bandwidth; /* really, bytes/tick. */ 317 int delay; /* really, ticks */ 318 319 struct dn_pkt_queue p_queue;/* packets in delay line */ 320 LIST_ENTRY(dn_pipe) p_link; 321 322 /* WF2Q+ */ 323 struct dn_heap scheduler_heap; /* top extract - key Finish time*/ 324 struct dn_heap not_eligible_heap; /* top extract- key Start time */ 325 struct dn_heap idle_heap; /* random extract - key Start=Finish time */ 326 327 dn_key V; /* virtual time */ 328 int sum; /* sum of weights of all active sessions */ 329 int numbytes; /* bits I can transmit (more or less). */ 330 331 dn_key sched_time; /* time pipe was scheduled in ready_heap */ 332 333 struct dn_flow_set fs; /* used with fixed-rate flows */ 334 }; 335 LIST_HEAD(dn_pipe_head, dn_pipe); 336 337 struct dn_sopt { 338 int dn_sopt_name; 339 void *dn_sopt_arg; 340 size_t dn_sopt_arglen; 341 }; 342 343 typedef int ip_dn_ctl_t(struct dn_sopt *); 344 typedef int ip_dn_io_t(struct mbuf *); 345 346 extern ip_dn_ctl_t *ip_dn_ctl_ptr; 347 extern ip_dn_io_t *ip_dn_io_ptr; 348 349 void ip_dn_queue(struct mbuf *); 350 void ip_dn_packet_free(struct dn_pkt *); 351 void ip_dn_packet_redispatch(struct dn_pkt *); 352 int ip_dn_sockopt(struct sockopt *); 353 354 #define DUMMYNET_LOADED (ip_dn_io_ptr != NULL) 355 356 #endif /* _KERNEL */ 357 358 struct dn_ioc_flowid { 359 uint16_t type; /* ETHERTYPE_ */ 360 uint16_t pad; 361 union { 362 struct { 363 uint32_t dst_ip; 364 uint32_t src_ip; 365 uint16_t dst_port; 366 uint16_t src_port; 367 uint8_t proto; 368 uint8_t flags; 369 } ip; 370 uint8_t pad[64]; 371 } u; 372 }; 373 374 struct dn_ioc_flowqueue { 375 u_int len; 376 u_int len_bytes; 377 378 uint64_t tot_pkts; 379 uint64_t tot_bytes; 380 uint32_t drops; 381 382 int hash_slot; /* debugging/diagnostic */ 383 dn_key S; /* virtual start time */ 384 dn_key F; /* virtual finish time */ 385 386 struct dn_ioc_flowid id; 387 uint8_t reserved[16]; 388 }; 389 390 struct dn_ioc_flowset { 391 u_short fs_type; /* DN_IS_{QUEUE,PIPE}, MUST be first */ 392 393 u_short fs_nr; /* flow_set number */ 394 u_short flags_fs; /* see 'Flow set flags' */ 395 u_short parent_nr; /* parent pipe#, 0 if local to a pipe */ 396 397 int weight; /* WFQ queue weight */ 398 int qsize; /* queue size in slots or bytes */ 399 int plr; /* pkt loss rate (2^31-1 means 100%) */ 400 401 /* Hash table information */ 402 int rq_size; /* number of slots */ 403 int rq_elements; /* active elements */ 404 405 /* RED parameters */ 406 int w_q; /* queue weight (scaled) */ 407 int max_th; /* maximum threshold for queue (scaled) */ 408 int min_th; /* minimum threshold for queue (scaled) */ 409 int max_p; /* maximum value for p_b (scaled) */ 410 int lookup_step; /* granularity inside the lookup table */ 411 int lookup_weight; /* equal to (1-w_q)^t / (1-w_q)^(t+1) */ 412 413 struct dn_ioc_flowid flow_mask; 414 uint8_t reserved[16]; 415 }; 416 417 struct dn_ioc_pipe { 418 struct dn_ioc_flowset fs; /* MUST be first */ 419 420 int pipe_nr; /* pipe number */ 421 int bandwidth; /* bit/second */ 422 int delay; /* milliseconds */ 423 424 dn_key V; /* virtual time */ 425 426 uint8_t reserved[16]; 427 }; 428 429 /* 430 * Flow set flags 431 */ 432 #define DN_HAVE_FLOW_MASK 0x0001 433 #define DN_IS_RED 0x0002 434 #define DN_IS_GENTLE_RED 0x0004 435 #define DN_QSIZE_IS_BYTES 0x0008 /* queue size is measured in bytes */ 436 #define DN_NOERROR 0x0010 /* do not report ENOBUFS on drops */ 437 #define DN_IS_PIPE 0x4000 438 #define DN_IS_QUEUE 0x8000 439 440 /* 441 * Macros for RED 442 */ 443 #define SCALE_RED 16 444 #define SCALE(x) ((x) << SCALE_RED) 445 #define SCALE_VAL(x) ((x) >> SCALE_RED) 446 #define SCALE_MUL(x, y) (((x) * (y)) >> SCALE_RED) 447 448 /* 449 * Maximum pipe number 450 */ 451 #define DN_PIPE_NR_MAX 65536 452 453 #endif /* !_IP_DUMMYNET_H */ 454