rdsv3.h revision 12863:5783b6fe7ed5
1/* 2 * This file contains definitions imported from the OFED rds header rds.h. 3 * Oracle elects to have and use the contents of rds.h under and 4 * governed by the OpenIB.org BSD license. 5 */ 6 7/* 8 * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved. 9 */ 10 11#ifndef _RDSV3_RDSV3_H 12#define _RDSV3_RDSV3_H 13 14/* 15 * The name of this file is rds.h in ofed. 16 */ 17 18#ifdef __cplusplus 19extern "C" { 20#endif 21 22#include <sys/sunndi.h> 23#include <netinet/in.h> 24#include <sys/synch.h> 25#include <sys/stropts.h> 26#include <sys/socket.h> 27#include <sys/socketvar.h> 28#include <inet/ip.h> 29#include <sys/avl.h> 30#include <sys/param.h> 31#include <sys/time.h> 32#include <sys/rds.h> 33 34#include <sys/ib/ibtl/ibti.h> 35#include <sys/ib/clients/of/rdma/ib_verbs.h> 36#include <sys/ib/clients/of/rdma/ib_addr.h> 37#include <sys/ib/clients/of/rdma/rdma_cm.h> 38#include <sys/ib/clients/rdsv3/rdsv3_impl.h> 39#include <sys/ib/clients/rdsv3/info.h> 40 41#define NIPQUAD(addr) \ 42 (unsigned char)((ntohl(addr) >> 24) & 0xFF), \ 43 (unsigned char)((ntohl(addr) >> 16) & 0xFF), \ 44 (unsigned char)((ntohl(addr) >> 8) & 0xFF), \ 45 (unsigned char)(ntohl(addr) & 0xFF) 46 47/* 48 * RDS Network protocol version 49 */ 50#define RDS_PROTOCOL_3_0 0x0300 51#define RDS_PROTOCOL_3_1 0x0301 52#define RDS_PROTOCOL_VERSION RDS_PROTOCOL_3_1 53#define RDS_PROTOCOL_MAJOR(v) ((v) >> 8) 54#define RDS_PROTOCOL_MINOR(v) ((v) & 255) 55#define RDS_PROTOCOL(maj, min) (((maj) << 8) | min) 56 57/* 58 * XXX randomly chosen, but at least seems to be unused: 59 * # 18464-18768 Unassigned 60 * We should do better. We want a reserved port to discourage unpriv'ed 61 * userspace from listening. 62 * 63 * port 18633 was the version that had ack frames on the wire. 64 */ 65#define RDSV3_PORT 18634 66 67#define RDSV3_REAPER_WAIT_SECS (5*60) 68#define RDSV3_REAPER_WAIT_JIFFIES SEC_TO_TICK(RDSV3_REAPER_WAIT_SECS) 69 70/* 71 * This is the sad making. Some kernels have a bug in the per_cpu() api which 72 * makes DEFINE_PER_CPU trigger an oops on insmod because the per-cpu section 73 * in the module is not cacheline-aligned. As much as we'd like to tell users 74 * with older kernels to stuff it, that's not reasonable. We'll roll our own 75 * until this doesn't have to build against older kernels. 76 */ 77#define RDSV3_DEFINE_PER_CPU(type, var) type var[NR_CPUS] 78#define RDSV3_DECLARE_PER_CPU(type, var) extern type var[NR_CPUS] 79#define rdsv3_per_cpu(var, cpu) var[cpu] 80 81static inline ulong_t 82ceil(ulong_t x, ulong_t y) 83{ 84 return ((x + y - 1) / y); 85} 86 87#define RDSV3_FRAG_SHIFT 12 88#define RDSV3_FRAG_SIZE ((unsigned int)(1 << RDSV3_FRAG_SHIFT)) 89 90#define RDSV3_CONG_MAP_BYTES (65536 / 8) 91#define RDSV3_CONG_MAP_LONGS (RDSV3_CONG_MAP_BYTES / sizeof (unsigned long)) 92#define RDSV3_CONG_MAP_PAGES (RDSV3_CONG_MAP_BYTES / PAGE_SIZE) 93#define RDSV3_CONG_MAP_PAGE_BITS (PAGE_SIZE * 8) 94 95struct rdsv3_cong_map { 96 struct avl_node m_rb_node; 97 uint32_be_t m_addr; 98 rdsv3_wait_queue_t m_waitq; 99 struct list m_conn_list; 100 unsigned long m_page_addrs[RDSV3_CONG_MAP_PAGES]; 101}; 102 103/* 104 * This is how we will track the connection state: 105 * A connection is always in one of the following 106 * states. Updates to the state are atomic and imply 107 * a memory barrier. 108 */ 109enum { 110 RDSV3_CONN_DOWN = 0, 111 RDSV3_CONN_CONNECTING, 112 RDSV3_CONN_DISCONNECTING, 113 RDSV3_CONN_UP, 114 RDSV3_CONN_ERROR, 115}; 116 117/* Bits for c_flags */ 118#define RDSV3_LL_SEND_FULL 0 119#define RDSV3_RECONNECT_PENDING 1 120 121struct rdsv3_connection { 122 struct avl_node c_hash_node; 123 uint32_be_t c_laddr; 124 uint32_be_t c_faddr; 125 unsigned int c_loopback:1; 126 struct rdsv3_connection *c_passive; 127 128 struct rdsv3_cong_map *c_lcong; 129 struct rdsv3_cong_map *c_fcong; 130 131 struct mutex c_send_lock; /* protect send ring */ 132 atomic_t c_send_generation; 133 atomic_t c_senders; 134 135 struct rdsv3_message *c_xmit_rm; 136 unsigned long c_xmit_sg; 137 unsigned int c_xmit_hdr_off; 138 unsigned int c_xmit_data_off; 139 unsigned int c_xmit_rdma_sent; 140 141 kmutex_t c_lock; /* protect msg queues */ 142 uint64_t c_next_tx_seq; 143 struct list c_send_queue; 144 struct list c_retrans; 145 146 uint64_t c_next_rx_seq; 147 148 struct rdsv3_transport *c_trans; 149 void *c_transport_data; 150 151 atomic_t c_state; 152 unsigned long c_flags; 153 unsigned long c_reconnect_jiffies; 154 clock_t c_last_connect_jiffies; 155 156 struct rdsv3_delayed_work_s c_send_w; 157 struct rdsv3_delayed_work_s c_recv_w; 158 struct rdsv3_delayed_work_s c_conn_w; 159 struct rdsv3_delayed_work_s c_reap_w; 160 struct rdsv3_work_s c_down_w; 161 struct mutex c_cm_lock; /* protect conn state & cm */ 162 163 struct list_node c_map_item; 164 unsigned long c_map_queued; 165 unsigned long c_map_offset; 166 unsigned long c_map_bytes; 167 168 unsigned int c_unacked_packets; 169 unsigned int c_unacked_bytes; 170 171 /* Protocol version */ 172 unsigned int c_version; 173}; 174 175#define RDSV3_FLAG_CONG_BITMAP 0x01 176#define RDSV3_FLAG_ACK_REQUIRED 0x02 177#define RDSV3_FLAG_RETRANSMITTED 0x04 178#define RDSV3_MAX_ADV_CREDIT 127 179 180/* 181 * Maximum space available for extension headers. 182 */ 183#define RDSV3_HEADER_EXT_SPACE 16 184 185struct rdsv3_header { 186 uint64_be_t h_sequence; 187 uint64_be_t h_ack; 188 uint32_be_t h_len; 189 uint16_be_t h_sport; 190 uint16_be_t h_dport; 191 uint8_t h_flags; 192 uint8_t h_credit; 193 uint8_t h_padding[4]; 194 uint16_be_t h_csum; 195 196 uint8_t h_exthdr[RDSV3_HEADER_EXT_SPACE]; 197}; 198 199/* Reserved - indicates end of extensions */ 200#define RDSV3_EXTHDR_NONE 0 201 202/* 203 * This extension header is included in the very 204 * first message that is sent on a new connection, 205 * and identifies the protocol level. This will help 206 * rolling updates if a future change requires breaking 207 * the protocol. 208 */ 209#define RDSV3_EXTHDR_VERSION 1 210struct rdsv3_ext_header_version { 211 uint32_be_t h_version; 212}; 213 214/* 215 * This extension header is included in the RDS message 216 * chasing an RDMA operation. 217 */ 218#define RDSV3_EXTHDR_RDMA 2 219struct rdsv3_ext_header_rdma { 220 uint32_be_t h_rdma_rkey; 221}; 222 223/* 224 * This extension header tells the peer about the 225 * destination <R_Key,offset> of the requested RDMA 226 * operation. 227 */ 228#define RDSV3_EXTHDR_RDMA_DEST 3 229struct rdsv3_ext_header_rdma_dest { 230 uint32_be_t h_rdma_rkey; 231 uint32_be_t h_rdma_offset; 232}; 233 234#define __RDSV3_EXTHDR_MAX 16 /* for now */ 235 236struct rdsv3_incoming { 237 atomic_t i_refcount; 238 struct list_node i_item; 239 struct rdsv3_connection *i_conn; 240 struct rdsv3_header i_hdr; 241 unsigned long i_rx_jiffies; 242 uint32_be_t i_saddr; 243 244 rds_rdma_cookie_t i_rdma_cookie; 245}; 246 247/* 248 * m_sock_item and m_conn_item are on lists that are serialized under 249 * conn->c_lock. m_sock_item has additional meaning in that once it is empty 250 * the message will not be put back on the retransmit list after being sent. 251 * messages that are canceled while being sent rely on this. 252 * 253 * m_inc is used by loopback so that it can pass an incoming message straight 254 * back up into the rx path. It embeds a wire header which is also used by 255 * the send path, which is kind of awkward. 256 * 257 * m_sock_item indicates the message's presence on a socket's send or receive 258 * queue. m_rs will point to that socket. 259 * 260 * m_daddr is used by cancellation to prune messages to a given destination. 261 * 262 * The RDS_MSG_ON_SOCK and RDS_MSG_ON_CONN flags are used to avoid lock 263 * nesting. As paths iterate over messages on a sock, or conn, they must 264 * also lock the conn, or sock, to remove the message from those lists too. 265 * Testing the flag to determine if the message is still on the lists lets 266 * us avoid testing the list_head directly. That means each path can use 267 * the message's list_head to keep it on a local list while juggling locks 268 * without confusing the other path. 269 * 270 * m_ack_seq is an optional field set by transports who need a different 271 * sequence number range to invalidate. They can use this in a callback 272 * that they pass to rdsv3_send_drop_acked() to see if each message has been 273 * acked. The HAS_ACK_SEQ flag can be used to detect messages which haven't 274 * had ack_seq set yet. 275 */ 276#define RDSV3_MSG_ON_SOCK 1 277#define RDSV3_MSG_ON_CONN 2 278#define RDSV3_MSG_HAS_ACK_SEQ 3 279#define RDSV3_MSG_ACK_REQUIRED 4 280#define RDSV3_MSG_RETRANSMITTED 5 281#define RDSV3_MSG_MAPPED 6 282#define RDSV3_MSG_PAGEVEC 7 283 284struct rdsv3_message { 285 atomic_t m_refcount; 286 struct list_node m_sock_item; 287 struct list_node m_conn_item; 288 struct rdsv3_incoming m_inc; 289 uint64_t m_ack_seq; 290 uint32_be_t m_daddr; 291 unsigned long m_flags; 292 293 /* 294 * Never access m_rs without holding m_rs_lock. 295 * Lock nesting is 296 * rm->m_rs_lock 297 * -> rs->rs_lock 298 */ 299 kmutex_t m_rs_lock; 300 rdsv3_wait_queue_t m_flush_wait; 301 302 struct rdsv3_sock *m_rs; 303 struct rdsv3_rdma_op *m_rdma_op; 304 rds_rdma_cookie_t m_rdma_cookie; 305 struct rdsv3_mr *m_rdma_mr; 306 unsigned int m_nents; 307 unsigned int m_count; 308 struct rdsv3_scatterlist m_sg[1]; 309}; 310 311/* 312 * The RDS notifier is used (optionally) to tell the application about 313 * completed RDMA operations. Rather than keeping the whole rds message 314 * around on the queue, we allocate a small notifier that is put on the 315 * socket's notifier_list. Notifications are delivered to the application 316 * through control messages. 317 */ 318struct rdsv3_notifier { 319 list_node_t n_list; 320 uint64_t n_user_token; 321 int n_status; 322}; 323 324/* 325 * struct rdsv3_transport - transport specific behavioural hooks 326 * 327 * @xmit: .xmit is called by rdsv3_send_xmit() to tell the transport to send 328 * part of a message. The caller serializes on the send_sem so this 329 * doesn't need to be reentrant for a given conn. The header must be 330 * sent before the data payload. .xmit must be prepared to send a 331 * message with no data payload. .xmit should return the number of 332 * bytes that were sent down the connection, including header bytes. 333 * Returning 0 tells the caller that it doesn't need to perform any 334 * additional work now. This is usually the case when the transport has 335 * filled the sending queue for its connection and will handle 336 * triggering the rds thread to continue the send when space becomes 337 * available. Returning -EAGAIN tells the caller to retry the send 338 * immediately. Returning -ENOMEM tells the caller to retry the send at 339 * some point in the future. 340 * 341 * @conn_shutdown: conn_shutdown stops traffic on the given connection. Once 342 * it returns the connection can not call rdsv3_recv_incoming(). 343 * This will only be called once after conn_connect returns 344 * non-zero success and will The caller serializes this with 345 * the send and connecting paths (xmit_* and conn_*). The 346 * transport is responsible for other serialization, including 347 * rdsv3_recv_incoming(). This is called in process context but 348 * should try hard not to block. 349 * 350 * @xmit_cong_map: This asks the transport to send the local bitmap down the 351 * given connection. XXX get a better story about the bitmap 352 * flag and header. 353 */ 354 355#define RDS_TRANS_IB 0 356#define RDS_TRANS_IWARP 1 357#define RDS_TRANS_TCP 2 358#define RDS_TRANS_COUNT 3 359 360struct rdsv3_transport { 361 char t_name[TRANSNAMSIZ]; 362 struct list_node t_item; 363 unsigned int t_type; 364 unsigned int t_prefer_loopback:1; 365 366 int (*laddr_check)(uint32_be_t addr); 367 int (*conn_alloc)(struct rdsv3_connection *conn, int gfp); 368 void (*conn_free)(void *data); 369 int (*conn_connect)(struct rdsv3_connection *conn); 370 void (*conn_shutdown)(struct rdsv3_connection *conn); 371 void (*xmit_prepare)(struct rdsv3_connection *conn); 372 void (*xmit_complete)(struct rdsv3_connection *conn); 373 int (*xmit)(struct rdsv3_connection *conn, struct rdsv3_message *rm, 374 unsigned int hdr_off, unsigned int sg, unsigned int off); 375 int (*xmit_cong_map)(struct rdsv3_connection *conn, 376 struct rdsv3_cong_map *map, unsigned long offset); 377 int (*xmit_rdma)(struct rdsv3_connection *conn, 378 struct rdsv3_rdma_op *op); 379 int (*recv)(struct rdsv3_connection *conn); 380 int (*inc_copy_to_user)(struct rdsv3_incoming *inc, uio_t *uio, 381 size_t size); 382 void (*inc_free)(struct rdsv3_incoming *inc); 383 384 int (*cm_handle_connect)(struct rdma_cm_id *cm_id, 385 struct rdma_cm_event *event); 386 int (*cm_initiate_connect)(struct rdma_cm_id *cm_id); 387 void (*cm_connect_complete)(struct rdsv3_connection *conn, 388 struct rdma_cm_event *event); 389 390 unsigned int (*stats_info_copy)(struct rdsv3_info_iterator *iter, 391 unsigned int avail); 392 void (*exit)(void); 393 void *(*get_mr)(struct rds_iovec *sg, unsigned long nr_sg, 394 struct rdsv3_sock *rs, uint32_t *key_ret); 395 void (*sync_mr)(void *trans_private, int direction); 396 void (*free_mr)(void *trans_private, int invalidate); 397 void (*flush_mrs)(void); 398}; 399 400struct rdsv3_sock { 401 struct rsock *rs_sk; 402 uint64_t rs_user_addr; 403 uint64_t rs_user_bytes; 404 405 /* 406 * bound_addr used for both incoming and outgoing, no INADDR_ANY 407 * support. 408 */ 409 struct avl_node rs_bound_node; 410 uint32_be_t rs_bound_addr; 411 uint32_be_t rs_conn_addr; 412 uint16_be_t rs_bound_port; 413 uint16_be_t rs_conn_port; 414 415 /* 416 * This is only used to communicate the transport between bind and 417 * initiating connections. All other trans use is referenced through 418 * the connection. 419 */ 420 struct rdsv3_transport *rs_transport; 421 422 /* 423 * rdsv3_sendmsg caches the conn it used the last time around. 424 * This helps avoid costly lookups. 425 */ 426 struct rdsv3_connection *rs_conn; 427 kmutex_t rs_conn_lock; 428 429 /* flag indicating we were congested or not */ 430 int rs_congested; 431 /* seen congestion (ENOBUFS) when sending? */ 432 int rs_seen_congestion; 433 kmutex_t rs_congested_lock; 434 kcondvar_t rs_congested_cv; 435 436 /* rs_lock protects all these adjacent members before the newline */ 437 kmutex_t rs_lock; 438 struct list rs_send_queue; 439 uint32_t rs_snd_bytes; 440 int rs_rcv_bytes; 441 /* currently used for failed RDMAs */ 442 struct list rs_notify_queue; 443 444 /* 445 * Congestion wake_up. If rs_cong_monitor is set, we use cong_mask 446 * to decide whether the application should be woken up. 447 * If not set, we use rs_cong_track to find out whether a cong map 448 * update arrived. 449 */ 450 uint64_t rs_cong_mask; 451 uint64_t rs_cong_notify; 452 struct list_node rs_cong_list; 453 unsigned long rs_cong_track; 454 455 /* 456 * rs_recv_lock protects the receive queue, and is 457 * used to serialize with rdsv3_release. 458 */ 459 krwlock_t rs_recv_lock; 460 struct list rs_recv_queue; 461 462 /* just for stats reporting */ 463 struct list_node rs_item; 464 465 /* these have their own lock */ 466 kmutex_t rs_rdma_lock; 467 struct avl_tree rs_rdma_keys; 468 469 /* Socket options - in case there will be more */ 470 unsigned char rs_recverr, 471 rs_cong_monitor; 472 473 cred_t *rs_cred; 474 zoneid_t rs_zoneid; 475}; 476 477static inline struct rdsv3_sock * 478rdsv3_sk_to_rs(const struct rsock *sk) 479{ 480 return ((struct rdsv3_sock *)sk->sk_protinfo); 481} 482 483static inline struct rsock * 484rdsv3_rs_to_sk(const struct rdsv3_sock *rs) 485{ 486 return ((struct rsock *)rs->rs_sk); 487} 488 489/* 490 * The stack assigns sk_sndbuf and sk_rcvbuf to twice the specified value 491 * to account for overhead. We don't account for overhead, we just apply 492 * the number of payload bytes to the specified value. 493 */ 494static inline int 495rdsv3_sk_sndbuf(struct rdsv3_sock *rs) 496{ 497 /* XXX */ 498 return (rdsv3_rs_to_sk(rs)->sk_sndbuf); 499} 500 501static inline int 502rdsv3_sk_rcvbuf(struct rdsv3_sock *rs) 503{ 504 /* XXX */ 505 return (rdsv3_rs_to_sk(rs)->sk_rcvbuf); 506} 507 508struct rdsv3_statistics { 509 uint64_t s_conn_reset; 510 uint64_t s_recv_drop_bad_checksum; 511 uint64_t s_recv_drop_old_seq; 512 uint64_t s_recv_drop_no_sock; 513 uint64_t s_recv_drop_dead_sock; 514 uint64_t s_recv_deliver_raced; 515 uint64_t s_recv_delivered; 516 uint64_t s_recv_queued; 517 uint64_t s_recv_immediate_retry; 518 uint64_t s_recv_delayed_retry; 519 uint64_t s_recv_ack_required; 520 uint64_t s_recv_rdma_bytes; 521 uint64_t s_recv_ping; 522 uint64_t s_send_queue_empty; 523 uint64_t s_send_queue_full; 524 uint64_t s_send_sem_contention; 525 uint64_t s_send_sem_queue_raced; 526 uint64_t s_send_immediate_retry; 527 uint64_t s_send_delayed_retry; 528 uint64_t s_send_drop_acked; 529 uint64_t s_send_ack_required; 530 uint64_t s_send_queued; 531 uint64_t s_send_rdma; 532 uint64_t s_send_rdma_bytes; 533 uint64_t s_send_pong; 534 uint64_t s_page_remainder_hit; 535 uint64_t s_page_remainder_miss; 536 uint64_t s_copy_to_user; 537 uint64_t s_copy_from_user; 538 uint64_t s_cong_update_queued; 539 uint64_t s_cong_update_received; 540 uint64_t s_cong_send_error; 541 uint64_t s_cong_send_blocked; 542}; 543 544/* af_rds.c */ 545void rdsv3_sock_addref(struct rdsv3_sock *rs); 546void rdsv3_sock_put(struct rdsv3_sock *rs); 547void rdsv3_wake_sk_sleep(struct rdsv3_sock *rs); 548void __rdsv3_wake_sk_sleep(struct rsock *sk); 549 550/* bind.c */ 551int rdsv3_bind(sock_lower_handle_t proto_handle, struct sockaddr *sa, 552 socklen_t len, cred_t *cr); 553void rdsv3_remove_bound(struct rdsv3_sock *rs); 554struct rdsv3_sock *rdsv3_find_bound(uint32_be_t addr, uint16_be_t port); 555 556/* conn.c */ 557int rdsv3_conn_init(void); 558void rdsv3_conn_exit(void); 559struct rdsv3_connection *rdsv3_conn_create(uint32_be_t laddr, uint32_be_t faddr, 560 struct rdsv3_transport *trans, int gfp); 561struct rdsv3_connection *rdsv3_conn_create_outgoing(uint32_be_t laddr, 562 uint32_be_t faddr, 563 struct rdsv3_transport *trans, int gfp); 564void rdsv3_conn_shutdown(struct rdsv3_connection *conn); 565void rdsv3_conn_destroy(struct rdsv3_connection *conn); 566void rdsv3_conn_reset(struct rdsv3_connection *conn); 567void rdsv3_conn_drop(struct rdsv3_connection *conn); 568void rdsv3_for_each_conn_info(struct rsock *sock, unsigned int len, 569 struct rdsv3_info_iterator *iter, 570 struct rdsv3_info_lengths *lens, 571 int (*visitor)(struct rdsv3_connection *, void *), 572 size_t item_len); 573 574static inline int 575rdsv3_conn_transition(struct rdsv3_connection *conn, int old, int new) 576{ 577 return (atomic_cmpxchg(&conn->c_state, old, new) == old); 578} 579 580static inline int 581rdsv3_conn_state(struct rdsv3_connection *conn) 582{ 583 return (atomic_get(&conn->c_state)); 584} 585 586static inline int 587rdsv3_conn_up(struct rdsv3_connection *conn) 588{ 589 return (atomic_get(&conn->c_state) == RDSV3_CONN_UP); 590} 591 592static inline int 593rdsv3_conn_connecting(struct rdsv3_connection *conn) 594{ 595 return (atomic_get(&conn->c_state) == RDSV3_CONN_CONNECTING); 596} 597 598/* recv.c */ 599void rdsv3_inc_init(struct rdsv3_incoming *inc, struct rdsv3_connection *conn, 600 uint32_be_t saddr); 601void rdsv3_inc_addref(struct rdsv3_incoming *inc); 602void rdsv3_inc_put(struct rdsv3_incoming *inc); 603void rdsv3_recv_incoming(struct rdsv3_connection *conn, uint32_be_t saddr, 604 uint32_be_t daddr, 605 struct rdsv3_incoming *inc, int gfp); 606int rdsv3_recvmsg(struct rdsv3_sock *rs, uio_t *uio, 607 struct msghdr *msg, size_t size, int msg_flags); 608void rdsv3_clear_recv_queue(struct rdsv3_sock *rs); 609int rdsv3_notify_queue_get(struct rdsv3_sock *rs, struct msghdr *msg); 610void rdsv3_inc_info_copy(struct rdsv3_incoming *inc, 611 struct rdsv3_info_iterator *iter, 612 uint32_be_t saddr, uint32_be_t daddr, int flip); 613 614/* page.c */ 615int rdsv3_page_remainder_alloc(struct rdsv3_scatterlist *scat, 616 unsigned long bytes, int gfp); 617 618/* send.c */ 619int rdsv3_sendmsg(struct rdsv3_sock *rs, uio_t *uio, struct nmsghdr *msg, 620 size_t payload_len); 621void rdsv3_send_reset(struct rdsv3_connection *conn); 622int rdsv3_send_xmit(struct rdsv3_connection *conn); 623struct sockaddr_in; 624void rdsv3_send_drop_to(struct rdsv3_sock *rs, struct sockaddr_in *dest); 625typedef int (*is_acked_func)(struct rdsv3_message *rm, uint64_t ack); 626void rdsv3_send_drop_acked(struct rdsv3_connection *conn, uint64_t ack, 627 is_acked_func is_acked); 628int rdsv3_send_acked_before(struct rdsv3_connection *conn, uint64_t seq); 629void rdsv3_send_remove_from_sock(struct list *messages, int status); 630int rdsv3_send_pong(struct rdsv3_connection *conn, uint16_be_t dport); 631struct rdsv3_message *rdsv3_send_get_message(struct rdsv3_connection *, 632 struct rdsv3_rdma_op *); 633 634/* rdma.c */ 635void rdsv3_rdma_unuse(struct rdsv3_sock *rs, uint32_t r_key, int force); 636 637/* cong.c */ 638void rdsv3_cong_init(void); 639int rdsv3_cong_get_maps(struct rdsv3_connection *conn); 640void rdsv3_cong_add_conn(struct rdsv3_connection *conn); 641void rdsv3_cong_remove_conn(struct rdsv3_connection *conn); 642void rdsv3_cong_set_bit(struct rdsv3_cong_map *map, uint16_be_t port); 643void rdsv3_cong_clear_bit(struct rdsv3_cong_map *map, uint16_be_t port); 644int rdsv3_cong_wait(struct rdsv3_cong_map *map, uint16_be_t port, int nonblock, 645 struct rdsv3_sock *rs); 646void rdsv3_cong_queue_updates(struct rdsv3_cong_map *map); 647void rdsv3_cong_map_updated(struct rdsv3_cong_map *map, uint64_t); 648int rdsv3_cong_updated_since(unsigned long *recent); 649void rdsv3_cong_add_socket(struct rdsv3_sock *); 650void rdsv3_cong_remove_socket(struct rdsv3_sock *); 651void rdsv3_cong_exit(void); 652struct rdsv3_message *rdsv3_cong_update_alloc(struct rdsv3_connection *conn); 653 654/* stats.c */ 655RDSV3_DECLARE_PER_CPU(struct rdsv3_statistics, rdsv3_stats); 656#define rdsv3_stats_inc_which(which, member) do { \ 657 rdsv3_per_cpu(which, get_cpu()).member++; \ 658 put_cpu(); \ 659} while (0) 660#define rdsv3_stats_inc(member) rdsv3_stats_inc_which(rdsv3_stats, member) 661#define rdsv3_stats_add_which(which, member, count) do { \ 662 rdsv3_per_cpu(which, get_cpu()).member += count; \ 663 put_cpu(); \ 664} while (0) 665#define rdsv3_stats_add(member, count) \ 666 rdsv3_stats_add_which(rdsv3_stats, member, count) 667int rdsv3_stats_init(void); 668void rdsv3_stats_exit(void); 669void rdsv3_stats_info_copy(struct rdsv3_info_iterator *iter, 670 uint64_t *values, char **names, size_t nr); 671 672 673/* sysctl.c */ 674int rdsv3_sysctl_init(void); 675void rdsv3_sysctl_exit(void); 676extern unsigned long rdsv3_sysctl_sndbuf_min; 677extern unsigned long rdsv3_sysctl_sndbuf_default; 678extern unsigned long rdsv3_sysctl_sndbuf_max; 679extern unsigned long rdsv3_sysctl_reconnect_min_jiffies; 680extern unsigned long rdsv3_sysctl_reconnect_max_jiffies; 681extern unsigned int rdsv3_sysctl_max_unacked_packets; 682extern unsigned int rdsv3_sysctl_max_unacked_bytes; 683extern unsigned int rdsv3_sysctl_ping_enable; 684extern unsigned long rdsv3_sysctl_trace_flags; 685extern unsigned int rdsv3_sysctl_trace_level; 686 687/* threads.c */ 688int rdsv3_threads_init(); 689void rdsv3_threads_exit(void); 690extern struct rdsv3_workqueue_struct_s *rdsv3_wq; 691void rdsv3_queue_reconnect(struct rdsv3_connection *conn); 692void rdsv3_connect_worker(struct rdsv3_work_s *); 693void rdsv3_shutdown_worker(struct rdsv3_work_s *); 694void rdsv3_send_worker(struct rdsv3_work_s *); 695void rdsv3_recv_worker(struct rdsv3_work_s *); 696void rdsv3_reaper_worker(struct rdsv3_work_s *); 697void rdsv3_connect_complete(struct rdsv3_connection *conn); 698 699/* transport.c */ 700int rdsv3_trans_register(struct rdsv3_transport *trans); 701void rdsv3_trans_unregister(struct rdsv3_transport *trans); 702struct rdsv3_transport *rdsv3_trans_get_preferred(uint32_be_t addr); 703unsigned int rdsv3_trans_stats_info_copy(struct rdsv3_info_iterator *iter, 704 unsigned int avail); 705void rdsv3_trans_exit(void); 706 707/* message.c */ 708struct rdsv3_message *rdsv3_message_alloc(unsigned int nents, int gfp); 709struct rdsv3_message *rdsv3_message_copy_from_user(struct uio *uiop, 710 size_t total_len); 711struct rdsv3_message *rdsv3_message_map_pages(unsigned long *page_addrs, 712 unsigned int total_len); 713void rdsv3_message_populate_header(struct rdsv3_header *hdr, uint16_be_t sport, 714 uint16_be_t dport, uint64_t seq); 715int rdsv3_message_add_extension(struct rdsv3_header *hdr, 716 unsigned int type, const void *data, unsigned int len); 717int rdsv3_message_next_extension(struct rdsv3_header *hdr, 718 unsigned int *pos, void *buf, unsigned int *buflen); 719int rdsv3_message_add_version_extension(struct rdsv3_header *hdr, 720 unsigned int version); 721int rdsv3_message_get_version_extension(struct rdsv3_header *hdr, 722 unsigned int *version); 723int rdsv3_message_add_rdma_dest_extension(struct rdsv3_header *hdr, 724 uint32_t r_key, uint32_t offset); 725int rdsv3_message_inc_copy_to_user(struct rdsv3_incoming *inc, 726 uio_t *uio, size_t size); 727void rdsv3_message_inc_free(struct rdsv3_incoming *inc); 728void rdsv3_message_addref(struct rdsv3_message *rm); 729void rdsv3_message_put(struct rdsv3_message *rm); 730void rdsv3_message_wait(struct rdsv3_message *rm); 731void rdsv3_message_unmapped(struct rdsv3_message *rm); 732 733static inline void 734rdsv3_message_make_checksum(struct rdsv3_header *hdr) 735{ 736 hdr->h_csum = 0; 737 hdr->h_csum = 738 rdsv3_ip_fast_csum((void *)hdr, sizeof (*hdr) >> 2); 739} 740 741static inline int 742rdsv3_message_verify_checksum(const struct rdsv3_header *hdr) 743{ 744 return (!hdr->h_csum || 745 rdsv3_ip_fast_csum((void *)hdr, sizeof (*hdr) >> 2) == 0); 746} 747 748/* rdsv3_sc.c */ 749extern boolean_t rdsv3_if_lookup_by_name(char *if_name); 750extern int rdsv3_sc_path_lookup(ipaddr_t *localip, ipaddr_t *remip); 751extern ipaddr_t rdsv3_scaddr_to_ibaddr(ipaddr_t addr); 752 753#ifdef __cplusplus 754} 755#endif 756 757#endif /* _RDSV3_RDSV3_H */ 758