net/cor/cor.h

   1 /*
   2  *      Connection oriented routing
   3  *      Copyright (C) 2007-2008 Michael Blizek
   4  *
   5  *      This program is free software; you can redistribute it and/or
   6  *      modify it under the terms of the GNU General Public License
   7  *      as published by the Free Software Foundation; either version 2
   8  *      of the License, or (at your option) any later version.
   9  *
  10  *      This program is distributed in the hope that it will be useful,
  11  *      but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  *      MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  *      GNU General Public License for more details.
  14  *
  15  *      You should have received a copy of the GNU General Public License
  16  *      along with this program; if not, write to the Free Software
  17  *      Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
  18  *      02110-1301, USA.
  19  */
  20
  21 #include <asm/atomic.h>
  22
  23 #include <linux/types.h>
  24 #include <linux/netdevice.h>
  25 #include <linux/skbuff.h>
  26 #include <linux/spinlock.h>
  27 #include <linux/workqueue.h>
  28 #include <linux/kref.h>
  29
  30 #include "settings.h"
  31
  32
  33 /* options */
  34 #define PIDOUT_NEWCONN 16
  35 #define PIDOUT_SENDDEF_THRES 8
  36 #define PIDOUT_SENDDEF_COUNT 16
  37
  38
  39
  40 #define ETH_P_COR 0x1022
  41 #define AF_COR 37
  42 #define PF_COR AF_COR
  43
  44 #define SOCKADDRTYPE_PORT 1
  45 struct cor_sockaddr {
  46         int type;
  47
  48         union {
  49                 __be64 port;
  50         } addr;
  51 };
  52
  53 #define MAX_CONN_CMD_LEN 4096
  54
  55
  56 #define PACKET_TYPE_ANNOUNCE 1
  57 #define PACKET_TYPE_DATA 2
  58
  59 /*
  60  * Kernel packet data - these commands are sent by the neighbor
  61  * The end nodes may cause these commands to be sent, but they see them beyond
  62  * the first hop.
  63  */
  64
  65 /* KP_PADDING[1] */
  66 #define KP_PADDING 1
  67
  68 /*
  69  * KP_PING[1] cookie[4]
  70  * KP_PONG[1] cookie[4] respdelay[4]
  71  *
  72  * This is needed to find out whether the other node is reachable. After a new
  73  * neighbor is seen, ping requests are sent and the neighbor is only reachable
  74  * after a few pongs are received. These requests are also used to find out
  75  * whether a neighber is gone.
  76  *
  77  * respdelay:
  78  * The receiver of a ping may delay the sending of the pong e.g. to create
  79  * bigger kernel packets. The respdelay is the time in microseconds the packet
  80  * was delayed.
  81  */
  82 #define KP_PING 2
  83 #define KP_PONG 3
  84
  85 /* KP_ACK[1] sent_conn_id[4] seqno[4]
  86  *
  87  * sent_conn_id means that this is *not* the conn_id we use if we sent something
  88  * through this conn, but the conn_id that the neighbor used to send us the
  89  * packet
  90  */
  91 #define KP_ACK 4
  92
  93 /*
  94  * KP_SPEED[1] conn_id[4] speedinfo[2]
  95  *
  96  * speedinfo[2] =
  97  *     buffer_state_value = speedinfo % 181
  98  *     speed_value = speedinfo / 181
  99  *
 100  *     buffer_state = 1024 * pow(2, buffer_state_value/3.0)
 101  *     speed = 1024 * pow(2, speed_value/12.0)
 102  *   see the .0 ...
 103  *
 104  * This has to be done either with floating points (which is no so nice) or
 105  * you can calculate:
 106  *     buffer_state =  pow(2, value/3) *
 107  *         1024 * pow(pow(2, 1.0/3), buffer_state_value%3)
 108  *   where 1024 * pow(pow(2, 1.0/4), value%3) can be just a table lookup
 109  *   (the "1024" should be part of the value in the table, because it increases
 110  *   the accuracy)
 111  *
 112  *   you can do the same with the speed
 113  *
 114  *
 115  * Some values have special meanings:
 116  *   if speedinfo is the highest possible value(65535), it means both values
 117  *     are inifinite
 118  *   if buffer_state_value if > 91, you have to subtract 90 and make the
 119  *     resulting buffer_state negative
 120  */
 121 #define KP_SPEED 5
 122
 123 /* NOTE on connection ids:
 124  * connection ids we send are used for the receive channel
 125  * connection ids we receive are used for the send channel
 126  */
 127
 128 /*
 129  * incoming connection
 130  * KP_CONNECT[1] conn_id[4]
 131  */
 132 #define KP_CONNECT 6
 133
 134 /*
 135  * incoming connection successful,
 136  * the first conn_id is the same as previously sent/received in KP_CONNECT
 137  * the second conn_id is generated by us and used for the other direction
 138  * KP_CONNECT_SUCCESS[1] conn_id[4] conn_id[4]
 139  */
 140 #define KP_CONNECT_SUCCESS 7
 141
 142 /* KP_CONN_DATA[1] conn_id[4] seqno[4] length[2] data[length] */
 143 #define KP_CONN_DATA 8
 144
 145 /*
 146  * { KP_RESET_CONN[1] conn_id[4] }
 147  * We send this, if there is an established connection we want to close.
 148  */
 149 #define KP_RESET_CONN 9
 150
 151
 152 /*
 153  * Connection data which in interpreted when connection has no target yet
 154  * These commands are sent by the end node.
 155  *
 156  * Format:
 157  * cmd[2] length[4] parameter[length]
 158  * unrecogniced commands are ignored
 159  * parameters which are longer than expected are ignored as well
 160  */
 161
 162 /* outgoing connection: CD_CONNECT_NB[2] length[4]
 163  * addrtypelen[2] addrlen[2] addrtype[addrtypelen] addr[addrlen] */
 164 #define CD_CONNECT_NB 1
 165
 166 /* connection to local open part: CD_CONNECT_PORT[2] length[4] port[8] */
 167 #define CD_CONNECT_PORT 2
 168
 169 /*
 170  * CD_LIST_NEIGH sends CDR_BINDATA if the command was successful. The response
 171  * format is:
 172  *
 173  * totalneighs[4] response_rows[4]
 174  * for every row:
 175  *   numaddr[2] (addrtypelen[2] addrlen[2] addrtype[addrtypelen] addr[addrlen]
 176  *       )[numaddr]
 177  *
 178  * Neighbors have to be sorted by uptime, new neighbors first. This is so that
 179  * the routing daemon can easily find out whether there are new neighbors. It
 180  * only needs to send a query with offset 0. If the totalneighs stays the same
 181  * while new were added, a connection to another neighbor was lost.
 182  */
 183
 184 /* list connected neighbors: CD_LIST_NEIGH[2] length[4] limit[4] offset[4] */
 185 #define CD_LIST_NEIGH 3
 186
 187 /*
 188  * Connection data response
 189  * Format is the same as with connection data
 190  */
 191
 192 /*
 193  * {CDR_EXECOK[2] || CDR_EXECFAILED[2]}
 194  * reasoncode[2] reasontextlength[2] reasontext[reasontextlength]
 195  * reasontextlength may be 0
 196  */
 197 #define CDR_EXECOK 32768
 198         #define CDR_EXECOK_OK 33024
 199
 200 #define CDR_EXECFAILED 32769
 201         #define CDR_EXECFAILED_UNKNOWN_COMMAND 33280
 202         #define CDR_EXECFAILED_PERMISSION_DENIED 33281
 203         #define CDR_EXECFAILED_TEMPORARILY_OUT_OF_RESSOURCES 33282
 204         #define CDR_EXECFAILED_CMD_TOO_SHORT 33283
 205         #define CDR_EXECFAILED_CMD_TOO_LONG 33284
 206         #define CDR_EXECFAILED_TARGETADDRTYPE_UNKNOWN 33285
 207         #define CDR_EXECFAILED_TARGETADDR_DOESNTEXIST 33286
 208         #define CDR_EXECFAILED_TARGETADDR_PORTCLOSED 33287
 209         #define CDR_EXECFAILED_LISTENERQUEUE_FULL 33288
 210
 211 /*
 212  * must be sent after CDR_EXEC{OK|FAILED}
 213  * CDR_EXEOK_BINDATA[2] bindatalen[4] bindata[bindatalen] */
 214 #define CDR_BINDATA 32770
 215
 216
 217 /* result codes for rcv.c/proc_packet */
 218 #define RC_DROP 0
 219 #define RC_FINISHED 1
 220
 221 #define RC_RCV1_ANNOUNCE 2
 222 #define RC_RCV1_KERNEL 3
 223 #define RC_RCV1_CONN 4
 224
 225 struct htab_entry{
 226         /* start of next element, *not* next htab_entry */
 227         void *next;
 228 };
 229
 230 struct htable{
 231         struct htab_entry **htable;
 232         __u32 htable_size;
 233         __u32 cell_size;
 234         __u32 num_elements;
 235
 236         int (*matches)(void *htentry, void *searcheditem);
 237         __u32 key_offset;
 238         __u32 entry_offset;
 239         __u32 kref_offset;
 240 };
 241
 242 struct ping_cookie{
 243         unsigned long time;
 244         __u32 cookie;
 245         __u8 pongs; /* count of pongs for pings sent after this one */
 246 };
 247
 248 #define NEIGHBOR_STATE_INITIAL 0
 249 #define NEIGHBOR_STATE_ACTIVE 1
 250 #define NEIGHBOR_STATE_STALLED 2
 251
 252 struct neighbor{
 253         struct list_head nb_list;
 254
 255         struct kref ref;
 256
 257         struct net_device *dev;
 258         char mac[MAX_ADDR_LEN];
 259
 260         char *addr;
 261         __u16 addrlen;
 262
 263         struct delayed_work cmsg_timer;
 264         struct mutex cmsg_lock;
 265         struct list_head control_msgs_out;
 266         __u64 timeout;
 267         __u32 length;
 268
 269
 270         unsigned long last_ping_time; /* protected by cmsg_lock */
 271         __u32 noping_cnt;/* protected by cmsg_lock */
 272
 273         struct mutex pingcookie_lock;
 274         __u32 ping_intransit;
 275         struct ping_cookie cookies[PING_COOKIES_PER_NEIGH];
 276         __u32 lastcookie;
 277         atomic_t latency; /* microsecs */
 278
 279         struct mutex state_lock;
 280         union {
 281                 __u64 last_state_change;/* initial state */
 282                 /**
 283                  * last_roundtrip:
 284                  * time of the last sent packet which has been acked or
 285                  * otherwise responded to (e.g. pong)
 286                  */
 287                 unsigned long last_roundtrip;/* active state */
 288         }state_time;
 289         __u8 state;
 290         __u16 ping_success;
 291
 292
 293         atomic_t kpacket_seqno;
 294         atomic_t ooo_packets;
 295
 296         /*
 297          * connecions which receive data from/send data to this node
 298          * used when terminating all connections of a neighbor
 299          */
 300         struct mutex conn_list_lock;
 301         struct list_head rcv_conn_list;
 302         struct list_head snd_conn_list;
 303
 304         /*
 305          * the timer has to be inited when adding the neighbor
 306          * init_timer(struct timer_list * timer);
 307          * add_timer(struct timer_list * timer);
 308          */
 309         spinlock_t retrans_lock;
 310         struct timer_list retrans_timer;
 311
 312         /*
 313          * next_retransmit are linked with
 314          * skb_procstate->funcstate.retransmit_queue
 315          * because the sk_buff next/prev fields are needed by the hashtable
 316          */
 317         struct sk_buff_head retrans_list;
 318
 319         struct conn *firstboundconn;
 320 };
 321
 322 struct cor_sched_data{
 323         spinlock_t lock;
 324         struct list_head conn_list;
 325         struct sk_buff_head requeue_queue;
 326 };
 327
 328 #define TYPE_BUF 0
 329 #define TYPE_SKB 1
 330
 331 struct data_buf_item{
 332         struct list_head buf_list;
 333
 334         union {
 335                 struct {
 336                         char *buf;
 337                         __u32 datalen;
 338
 339                 }buf;
 340
 341                 struct sk_buff *skb;
 342         }data;
 343
 344         __u8 type;
 345 };
 346
 347 struct data_buf{
 348         struct list_head items;
 349         struct data_buf_item *lastread;
 350         __u64 first_offset;
 351         __u64 read_offset;
 352
 353         __u32 totalsize;
 354         __u32 read_remaining;
 355
 356         __u16 last_read_offset;
 357
 358         __u16 last_buflen;
 359 };
 360
 361 struct connlistener;
 362
 363 struct bindnode{
 364         struct list_head lh;
 365         struct connlistener *owner;
 366         __be64 port;
 367 };
 368
 369 #define SOCKSTATE_LISTENER 1
 370 #define SOCKSTATE_CONN 2
 371
 372 struct sock_hdr {
 373         /* The first member of connlistener/conn (see sock.c) */
 374         __u8 sockstate;
 375 };
 376
 377 struct connlistener {
 378         /* The first member has to be the same as in conn (see sock.c) */
 379         __u8 sockstate;
 380         struct bindnode *bn;
 381         struct mutex lock;
 382         int queue_maxlen;
 383         int queue_len;
 384         struct list_head conn_queue;
 385         wait_queue_head_t wait;
 386
 387 };
 388
 389 /*
 390  * There are 2 conn objects per bi-directional connection. They refer to each
 391  * other with in the reversedir field. To distinguish them, the variables on
 392  * the stack are usually called rconn and sconn. rconn refers to the conn object
 393  * which has received a command. sconn is the other conn object. This means that
 394  * in send functions rconn means the connection we want to send the command to.
 395  */
 396
 397 struct conn{
 398         /* The first member has to be the same as in connlistener (see sock.c)*/
 399         __u8 sockstate;
 400
 401 #define SOURCE_NONE 0
 402 #define SOURCE_IN 1
 403 #define SOURCE_SOCK 2
 404
 405 #define TARGET_UNCONNECTED 0
 406 #define TARGET_OUT 1
 407 #define TARGET_SOCK 2
 408
 409         __u8    sourcetype:4,
 410                 targettype:4;
 411         __u8 isreset;
 412         __u8 qdisc_active;
 413         struct list_head queue_list;
 414
 415         struct kref ref;
 416
 417         struct mutex rcv_lock;
 418
 419         /* state */
 420         __u32 credits;
 421         /* credit rate */
 422         __s32 sender_crate;
 423         __s32 resp_crate;
 424
 425         union{
 426                 struct{
 427                         struct neighbor *nb;
 428                         /* list of all connections from this neighbor */
 429                         struct list_head nb_list;
 430
 431                         struct sk_buff_head reorder_queue;
 432
 433                         struct htab_entry htab_entry;
 434                         __u32 conn_id;
 435                         __u32 next_seqno;
 436                         __u32 ooo_packets;
 437                 }in;
 438
 439                 struct{
 440                         struct list_head cl_list;
 441                         wait_queue_head_t wait;
 442                         struct socket *sock;
 443                         int flags;
 444                 }sock;
 445         }source;
 446
 447         union{
 448                 struct{
 449                         __u32 paramlen;
 450                         __u32 cmdread;
 451                         __u16 cmd;
 452                         __u8 *cmdparams;
 453                 }unconnected;
 454
 455                 struct{
 456                         /* has to be first (because it is first in target
 457                          * kernel too)
 458                          */
 459                         struct neighbor *nb;
 460                         /* list of all connections to this neighbor */
 461                         struct list_head nb_list;
 462
 463                         __u32 conn_id;
 464                         __u32 seqno;
 465                 }out;
 466
 467                 struct{
 468                         wait_queue_head_t wait;
 469                 }sock;
 470         }target;
 471
 472         struct data_buf buf;
 473
 474         struct conn *reversedir;
 475 };
 476
 477 /* inside skb->cb */
 478 struct skb_procstate{
 479         struct conn *rconn;
 480
 481         union{
 482                 struct{
 483                         struct work_struct work;
 484                 }rcv;
 485
 486                 struct{
 487                         __u32 offset;
 488                 }announce;
 489
 490                 struct{
 491                         __u32 conn_id;
 492                         __u32 seqno;
 493                 }rcv2;
 494
 495                 struct{
 496                         struct htab_entry htab_entry;
 497                         struct kref ref;
 498                         unsigned long timeout;
 499                         __u32 conn_id;
 500                         __u32 seqno;
 501                         struct neighbor *nb;
 502                 }retransmit_queue;
 503         }funcstate;
 504 };
 505
 506
 507 /* common.c */
 508 extern char *htable_get(struct htable *ht, __u32 key, void *searcheditem);
 509
 510 extern int htable_delete(struct htable *ht, __u32 key, void *searcheditem,
 511                 void (*free) (struct kref *ref));
 512
 513 extern void htable_insert(struct htable *ht, char *newelement, __u32 key);
 514
 515 extern void htable_init(struct htable *ht, int (*matches)(void *htentry,
 516                 void *searcheditem), __u32 entry_offset,
 517                 __u32 kref_offset);
 518
 519 extern struct conn *get_conn(__u32 conn_id);
 520
 521 extern void free_conn(struct kref *ref);
 522
 523 extern int conn_init_out(struct conn *rconn, struct neighbor *nb);
 524
 525 extern void conn_init_sock_source(struct conn *conn);
 526 extern void conn_init_sock_target(struct conn *conn);
 527
 528 extern void close_port(struct connlistener *listener);
 529
 530 extern struct connlistener *open_port(__be64 port);
 531
 532 extern int connect_port(struct conn *rconn, __be64 port);
 533
 534 extern int connect_neigh(struct conn *rconn,
 535                 __u16 addrtypelen, __u8 *addrtype,
 536                 __u16 addrlen, __u8 *addr);
 537
 538 extern struct conn* alloc_conn(gfp_t allocflags);
 539
 540 extern void reset_conn(struct conn *conn);
 541
 542 /* neighbor.c */
 543 extern void neighbor_free(struct kref *ref);
 544
 545 extern struct neighbor *get_neigh_by_mac(struct sk_buff *skb);
 546
 547 extern struct neighbor *find_neigh(__u16 addrtypelen, __u8 *addrtype,
 548                 __u16 addrlen, __u8 *addr);
 549
 550 extern __u32 generate_neigh_list(char *buf, __u32 buflen, __u32 limit,
 551                 __u32 offset);
 552
 553 extern int get_neigh_state(struct neighbor *nb);
 554
 555 extern void ping_resp(struct neighbor *nb, __u32 cookie, __u32 respdelay);
 556
 557 extern __u32 add_ping_req(struct neighbor *nb);
 558
 559 extern int time_to_send_ping(struct neighbor *nb);
 560
 561 extern int force_ping(struct neighbor *nb);
 562
 563 extern void rcv_announce(struct sk_buff *skb);
 564
 565 extern int __init cor_neighbor_init(void);
 566
 567 /* rcv.c */
 568 extern void drain_ooo_queue(struct conn *rconn);
 569
 570 extern void conn_rcv_buildskb(char *data, __u32 datalen, __u32 conn_id,
 571                 __u32 seqno);
 572
 573 extern int __init cor_rcv_init(void);
 574
 575 /* kpacket_parse.c */
 576 extern void kernel_packet(struct neighbor *nb, struct sk_buff *skb, __u32 seqno);
 577
 578 /* kpacket_gen.c */
 579 extern void schedule_controlmsg_timerfunc(struct neighbor *nb);
 580
 581 struct control_msg_out;
 582
 583 extern struct control_msg_out *alloc_control_msg(void);
 584
 585 extern void free_control_msg(struct control_msg_out *cm);
 586
 587 extern void send_pong(struct control_msg_out *cm, struct neighbor *nb,
 588                 __u32 cookie);
 589
 590 extern void send_reset_conn(struct control_msg_out *cm, struct neighbor *nb,
 591                 __u32 conn_id);
 592
 593 extern void send_ack(struct control_msg_out *cm, struct neighbor *nb,
 594                 __u32 conn_id, __u32 seqno);
 595
 596 extern void send_connect_success(struct control_msg_out *cm,
 597                 struct neighbor *nb, __u32 rcvd_conn_id, __u32 gen_conn_id);
 598
 599 extern void send_connect_nb(struct control_msg_out *cm, struct neighbor *nb,
 600                 __u32 conn_id);
 601
 602 extern void send_conndata(struct control_msg_out *cm, struct neighbor *nb,
 603                 __u32 connid, __u32 seqno, char *data_orig, char *data,
 604                 __u32 datalen);
 605
 606 /* cpacket_parse.c */
 607 extern void parse(struct conn *rconn);
 608
 609 /* snd.c */
 610 extern void retransmit_timerfunc(unsigned long arg);
 611
 612 extern struct sk_buff *create_packet_conn(struct conn *target, int size,
 613                 gfp_t alloc_flags);
 614
 615 extern struct sk_buff *create_packet_kernel(struct neighbor *nb, int size,
 616                 gfp_t alloc_flags);
 617
 618 extern void send_conn_flushdata(struct conn *rconn, char *data, __u32 datalen);
 619
 620 extern void send_packet(struct sk_buff *skb, struct neighbor *nb,
 621                 int retransmit);
 622
 623 extern void ack_received(struct neighbor *nb, __u32 conn_id, __u32 seqno);
 624
 625 extern void flush_out(struct conn *rconn);
 626
 627 extern int __init cor_snd_init(void);
 628
 629 /* forward.c */
 630 extern void databuf_pull(struct data_buf *data, char *dst, int len);
 631
 632 extern size_t databuf_pulluser(struct conn *sconn, struct msghdr *msg);
 633
 634 extern void databuf_ack(struct data_buf *buf, __u64 pos);
 635
 636 extern void databuf_ackread(struct data_buf *buf);
 637
 638 extern int databuf_maypush(struct data_buf *buf);
 639
 640 extern void databuf_free(struct data_buf *data);
 641
 642 extern void databuf_init(struct data_buf *data);
 643
 644 extern int receive_userbuf(struct conn *rconn, struct msghdr *msg);
 645
 646 extern void receive_buf(struct conn *rconn, char *buf, int len);
 647
 648 extern int receive_skb(struct conn *rconn, struct sk_buff *skb);
 649
 650 extern void wake_sender(struct conn *rconn);
 651
 652 extern void forward_init(void);
 653
 654
 655
 656 static inline struct skb_procstate *skb_pstate(struct sk_buff *skb)
 657 {
 658         return (struct skb_procstate *) &(skb->cb[0]);
 659 }
 660
 661 static inline struct sk_buff *skb_from_pstate(struct skb_procstate *ps)
 662 {
 663         return (struct sk_buff *) (((char *)ps) - offsetof(struct sk_buff,cb));
 664 }
 665
 666
 667 static inline __u32 mss(struct neighbor *nb)
 668 {
 669         return nb->dev->mtu - LL_RESERVED_SPACE(nb->dev) - 9;
 670 }
 671
 672
 673 static inline void put_u64(char *dst, __u64 value, int convbo)
 674 {
 675         char *p_value = (char *) &value;
 676
 677         if (convbo)
 678                 value = cpu_to_be64(value);
 679
 680         dst[0] = p_value[0];
 681         dst[1] = p_value[1];
 682         dst[2] = p_value[2];
 683         dst[3] = p_value[3];
 684         dst[4] = p_value[4];
 685         dst[5] = p_value[5];
 686         dst[6] = p_value[6];
 687         dst[7] = p_value[7];
 688 }
 689
 690 static inline void put_u32(char *dst, __u32 value, int convbo)
 691 {
 692         char *p_value = (char *) &value;
 693
 694         if (convbo)
 695                 value = cpu_to_be32(value);
 696
 697         dst[0] = p_value[0];
 698         dst[1] = p_value[1];
 699         dst[2] = p_value[2];
 700         dst[3] = p_value[3];
 701 }
 702
 703 static inline void put_u16(char *dst, __u16 value, int convbo)
 704 {
 705         char *p_value = (char *) &value;
 706
 707         if (convbo)
 708                 value = cpu_to_be16(value);
 709
 710         dst[0] = p_value[0];
 711         dst[1] = p_value[1];
 712 }
 713
 714 static inline char *cor_pull_skb(struct sk_buff *skb, unsigned int len)
 715 {
 716         char *ptr = skb_pull(skb, len);
 717
 718         if(ptr == 0)
 719                 return 0;
 720
 721         return ptr - len;
 722 }