blob: c2a5eef41343c816f70e6dc16fc7a2fa9ea2d684 [file] [log] [blame]
Andy Grover39de8282009-02-24 15:30:19 +00001#ifndef _RDS_RDS_H
2#define _RDS_RDS_H
3
4#include <net/sock.h>
5#include <linux/scatterlist.h>
6#include <linux/highmem.h>
7#include <rdma/rdma_cm.h>
8#include <linux/mutex.h>
9#include <linux/rds.h>
10
11#include "info.h"
12
13/*
14 * RDS Network protocol version
15 */
16#define RDS_PROTOCOL_3_0 0x0300
17#define RDS_PROTOCOL_3_1 0x0301
18#define RDS_PROTOCOL_VERSION RDS_PROTOCOL_3_1
19#define RDS_PROTOCOL_MAJOR(v) ((v) >> 8)
20#define RDS_PROTOCOL_MINOR(v) ((v) & 255)
21#define RDS_PROTOCOL(maj, min) (((maj) << 8) | min)
22
23/*
24 * XXX randomly chosen, but at least seems to be unused:
25 * # 18464-18768 Unassigned
26 * We should do better. We want a reserved port to discourage unpriv'ed
27 * userspace from listening.
28 */
29#define RDS_PORT 18634
30
Andy Grover8cbd9602009-04-01 08:20:20 +000031#ifdef ATOMIC64_INIT
32#define KERNEL_HAS_ATOMIC64
33#endif
34
Andy Grover39de8282009-02-24 15:30:19 +000035#ifdef DEBUG
36#define rdsdebug(fmt, args...) pr_debug("%s(): " fmt, __func__ , ##args)
37#else
38/* sigh, pr_debug() causes unused variable warnings */
Joe Perchesb9075fa2011-10-31 17:11:33 -070039static inline __printf(1, 2)
40void rdsdebug(char *fmt, ...)
Andy Grover39de8282009-02-24 15:30:19 +000041{
42}
43#endif
44
45/* XXX is there one of these somewhere? */
46#define ceil(x, y) \
47 ({ unsigned long __x = (x), __y = (y); (__x + __y - 1) / __y; })
48
49#define RDS_FRAG_SHIFT 12
50#define RDS_FRAG_SIZE ((unsigned int)(1 << RDS_FRAG_SHIFT))
51
52#define RDS_CONG_MAP_BYTES (65536 / 8)
Andy Grover39de8282009-02-24 15:30:19 +000053#define RDS_CONG_MAP_PAGES (PAGE_ALIGN(RDS_CONG_MAP_BYTES) / PAGE_SIZE)
54#define RDS_CONG_MAP_PAGE_BITS (PAGE_SIZE * 8)
55
56struct rds_cong_map {
57 struct rb_node m_rb_node;
58 __be32 m_addr;
59 wait_queue_head_t m_waitq;
60 struct list_head m_conn_list;
61 unsigned long m_page_addrs[RDS_CONG_MAP_PAGES];
62};
63
64
65/*
66 * This is how we will track the connection state:
67 * A connection is always in one of the following
68 * states. Updates to the state are atomic and imply
69 * a memory barrier.
70 */
71enum {
72 RDS_CONN_DOWN = 0,
73 RDS_CONN_CONNECTING,
74 RDS_CONN_DISCONNECTING,
75 RDS_CONN_UP,
76 RDS_CONN_ERROR,
77};
78
79/* Bits for c_flags */
80#define RDS_LL_SEND_FULL 0
81#define RDS_RECONNECT_PENDING 1
Zach Brown0f4b1c72010-06-04 14:41:41 -070082#define RDS_IN_XMIT 2
Andy Grover39de8282009-02-24 15:30:19 +000083
84struct rds_connection {
85 struct hlist_node c_hash_node;
86 __be32 c_laddr;
87 __be32 c_faddr;
88 unsigned int c_loopback:1;
89 struct rds_connection *c_passive;
90
91 struct rds_cong_map *c_lcong;
92 struct rds_cong_map *c_fcong;
93
Andy Grover39de8282009-02-24 15:30:19 +000094 struct rds_message *c_xmit_rm;
95 unsigned long c_xmit_sg;
96 unsigned int c_xmit_hdr_off;
97 unsigned int c_xmit_data_off;
Andy Grover15133f62010-01-12 14:33:38 -080098 unsigned int c_xmit_atomic_sent;
Andy Grover5b2366b2010-02-03 19:36:44 -080099 unsigned int c_xmit_rdma_sent;
100 unsigned int c_xmit_data_sent;
Andy Grover39de8282009-02-24 15:30:19 +0000101
102 spinlock_t c_lock; /* protect msg queues */
103 u64 c_next_tx_seq;
104 struct list_head c_send_queue;
105 struct list_head c_retrans;
106
107 u64 c_next_rx_seq;
108
109 struct rds_transport *c_trans;
110 void *c_transport_data;
111
112 atomic_t c_state;
113 unsigned long c_flags;
114 unsigned long c_reconnect_jiffies;
115 struct delayed_work c_send_w;
116 struct delayed_work c_recv_w;
117 struct delayed_work c_conn_w;
118 struct work_struct c_down_w;
119 struct mutex c_cm_lock; /* protect conn state & cm */
Zach Brown0f4b1c72010-06-04 14:41:41 -0700120 wait_queue_head_t c_waitq;
Andy Grover39de8282009-02-24 15:30:19 +0000121
122 struct list_head c_map_item;
123 unsigned long c_map_queued;
Andy Grover39de8282009-02-24 15:30:19 +0000124
125 unsigned int c_unacked_packets;
126 unsigned int c_unacked_bytes;
127
128 /* Protocol version */
129 unsigned int c_version;
130};
131
132#define RDS_FLAG_CONG_BITMAP 0x01
133#define RDS_FLAG_ACK_REQUIRED 0x02
134#define RDS_FLAG_RETRANSMITTED 0x04
Steve Wise7b70d032009-04-09 14:09:39 +0000135#define RDS_MAX_ADV_CREDIT 255
Andy Grover39de8282009-02-24 15:30:19 +0000136
137/*
138 * Maximum space available for extension headers.
139 */
140#define RDS_HEADER_EXT_SPACE 16
141
142struct rds_header {
143 __be64 h_sequence;
144 __be64 h_ack;
145 __be32 h_len;
146 __be16 h_sport;
147 __be16 h_dport;
148 u8 h_flags;
149 u8 h_credit;
150 u8 h_padding[4];
151 __sum16 h_csum;
152
153 u8 h_exthdr[RDS_HEADER_EXT_SPACE];
154};
155
156/*
157 * Reserved - indicates end of extensions
158 */
159#define RDS_EXTHDR_NONE 0
160
161/*
162 * This extension header is included in the very
163 * first message that is sent on a new connection,
164 * and identifies the protocol level. This will help
165 * rolling updates if a future change requires breaking
166 * the protocol.
167 * NB: This is no longer true for IB, where we do a version
168 * negotiation during the connection setup phase (protocol
169 * version information is included in the RDMA CM private data).
170 */
171#define RDS_EXTHDR_VERSION 1
172struct rds_ext_header_version {
173 __be32 h_version;
174};
175
176/*
177 * This extension header is included in the RDS message
178 * chasing an RDMA operation.
179 */
180#define RDS_EXTHDR_RDMA 2
181struct rds_ext_header_rdma {
182 __be32 h_rdma_rkey;
183};
184
185/*
186 * This extension header tells the peer about the
187 * destination <R_Key,offset> of the requested RDMA
188 * operation.
189 */
190#define RDS_EXTHDR_RDMA_DEST 3
191struct rds_ext_header_rdma_dest {
192 __be32 h_rdma_rkey;
193 __be32 h_rdma_offset;
194};
195
196#define __RDS_EXTHDR_MAX 16 /* for now */
197
198struct rds_incoming {
199 atomic_t i_refcount;
200 struct list_head i_item;
201 struct rds_connection *i_conn;
202 struct rds_header i_hdr;
203 unsigned long i_rx_jiffies;
204 __be32 i_saddr;
205
206 rds_rdma_cookie_t i_rdma_cookie;
207};
208
Andy Grover21f79af2010-01-12 12:57:27 -0800209struct rds_mr {
210 struct rb_node r_rb_node;
211 atomic_t r_refcount;
212 u32 r_key;
213
214 /* A copy of the creation flags */
215 unsigned int r_use_once:1;
216 unsigned int r_invalidate:1;
217 unsigned int r_write:1;
218
219 /* This is for RDS_MR_DEAD.
220 * It would be nice & consistent to make this part of the above
221 * bit field here, but we need to use test_and_set_bit.
222 */
223 unsigned long r_state;
224 struct rds_sock *r_sock; /* back pointer to the socket that owns us */
225 struct rds_transport *r_trans;
226 void *r_trans_private;
227};
228
229/* Flags for mr->r_state */
230#define RDS_MR_DEAD 0
231
Andy Grover21f79af2010-01-12 12:57:27 -0800232static inline rds_rdma_cookie_t rds_rdma_make_cookie(u32 r_key, u32 offset)
233{
234 return r_key | (((u64) offset) << 32);
235}
236
237static inline u32 rds_rdma_cookie_key(rds_rdma_cookie_t cookie)
238{
239 return cookie;
240}
241
242static inline u32 rds_rdma_cookie_offset(rds_rdma_cookie_t cookie)
243{
244 return cookie >> 32;
245}
246
Andy Grover15133f62010-01-12 14:33:38 -0800247/* atomic operation types */
248#define RDS_ATOMIC_TYPE_CSWP 0
249#define RDS_ATOMIC_TYPE_FADD 1
250
Andy Grover39de8282009-02-24 15:30:19 +0000251/*
252 * m_sock_item and m_conn_item are on lists that are serialized under
253 * conn->c_lock. m_sock_item has additional meaning in that once it is empty
254 * the message will not be put back on the retransmit list after being sent.
255 * messages that are canceled while being sent rely on this.
256 *
257 * m_inc is used by loopback so that it can pass an incoming message straight
258 * back up into the rx path. It embeds a wire header which is also used by
259 * the send path, which is kind of awkward.
260 *
261 * m_sock_item indicates the message's presence on a socket's send or receive
262 * queue. m_rs will point to that socket.
263 *
264 * m_daddr is used by cancellation to prune messages to a given destination.
265 *
266 * The RDS_MSG_ON_SOCK and RDS_MSG_ON_CONN flags are used to avoid lock
267 * nesting. As paths iterate over messages on a sock, or conn, they must
268 * also lock the conn, or sock, to remove the message from those lists too.
269 * Testing the flag to determine if the message is still on the lists lets
270 * us avoid testing the list_head directly. That means each path can use
271 * the message's list_head to keep it on a local list while juggling locks
272 * without confusing the other path.
273 *
274 * m_ack_seq is an optional field set by transports who need a different
275 * sequence number range to invalidate. They can use this in a callback
276 * that they pass to rds_send_drop_acked() to see if each message has been
277 * acked. The HAS_ACK_SEQ flag can be used to detect messages which haven't
278 * had ack_seq set yet.
279 */
280#define RDS_MSG_ON_SOCK 1
281#define RDS_MSG_ON_CONN 2
282#define RDS_MSG_HAS_ACK_SEQ 3
283#define RDS_MSG_ACK_REQUIRED 4
284#define RDS_MSG_RETRANSMITTED 5
285#define RDS_MSG_MAPPED 6
286#define RDS_MSG_PAGEVEC 7
287
288struct rds_message {
289 atomic_t m_refcount;
290 struct list_head m_sock_item;
291 struct list_head m_conn_item;
292 struct rds_incoming m_inc;
293 u64 m_ack_seq;
294 __be32 m_daddr;
295 unsigned long m_flags;
296
297 /* Never access m_rs without holding m_rs_lock.
298 * Lock nesting is
299 * rm->m_rs_lock
300 * -> rs->rs_lock
301 */
302 spinlock_t m_rs_lock;
Chris Masonc83188d2010-04-21 13:09:28 -0700303 wait_queue_head_t m_flush_wait;
304
Andy Grover39de8282009-02-24 15:30:19 +0000305 struct rds_sock *m_rs;
Andy Grover7e3bd652010-03-01 16:04:59 -0800306
307 /* cookie to send to remote, in rds header */
Andy Grover39de8282009-02-24 15:30:19 +0000308 rds_rdma_cookie_t m_rdma_cookie;
Andy Grover7e3bd652010-03-01 16:04:59 -0800309
310 unsigned int m_used_sgs;
311 unsigned int m_total_sgs;
312
Andy Groverff3d7d32010-03-01 14:03:09 -0800313 void *m_final_op;
314
Andy Grovere7791372010-01-12 12:15:02 -0800315 struct {
Andy Grover15133f62010-01-12 14:33:38 -0800316 struct rm_atomic_op {
317 int op_type;
Andy Grover20c72bd2010-08-25 05:51:28 -0700318 union {
319 struct {
320 uint64_t compare;
321 uint64_t swap;
322 uint64_t compare_mask;
323 uint64_t swap_mask;
324 } op_m_cswp;
325 struct {
326 uint64_t add;
327 uint64_t nocarry_mask;
328 } op_m_fadd;
329 };
Andy Grover15133f62010-01-12 14:33:38 -0800330
331 u32 op_rkey;
332 u64 op_remote_addr;
333 unsigned int op_notify:1;
334 unsigned int op_recverr:1;
335 unsigned int op_mapped:1;
Andy Grover2c3a5f92010-03-01 16:10:40 -0800336 unsigned int op_silent:1;
Andy Grover15133f62010-01-12 14:33:38 -0800337 unsigned int op_active:1;
Andy Grover15133f62010-01-12 14:33:38 -0800338 struct scatterlist *op_sg;
Andy Groverf8b3aaf2010-03-01 14:11:53 -0800339 struct rds_notifier *op_notifier;
Andy Grover15133f62010-01-12 14:33:38 -0800340
341 struct rds_mr *op_rdma_mr;
342 } atomic;
343 struct rm_rdma_op {
Andy Groverf8b3aaf2010-03-01 14:11:53 -0800344 u32 op_rkey;
345 u64 op_remote_addr;
346 unsigned int op_write:1;
347 unsigned int op_fence:1;
348 unsigned int op_notify:1;
349 unsigned int op_recverr:1;
350 unsigned int op_mapped:1;
Andy Grover2c3a5f92010-03-01 16:10:40 -0800351 unsigned int op_silent:1;
Andy Groverf8b3aaf2010-03-01 14:11:53 -0800352 unsigned int op_active:1;
353 unsigned int op_bytes;
354 unsigned int op_nents;
355 unsigned int op_count;
356 struct scatterlist *op_sg;
357 struct rds_notifier *op_notifier;
358
359 struct rds_mr *op_rdma_mr;
Andy Grovere7791372010-01-12 12:15:02 -0800360 } rdma;
Andy Grover15133f62010-01-12 14:33:38 -0800361 struct rm_data_op {
Andy Grover241eef32010-01-19 21:25:26 -0800362 unsigned int op_active:1;
Andy Grover6c7cc6e2010-01-27 18:04:18 -0800363 unsigned int op_nents;
364 unsigned int op_count;
365 struct scatterlist *op_sg;
Andy Grovere7791372010-01-12 12:15:02 -0800366 } data;
367 };
Andy Grover39de8282009-02-24 15:30:19 +0000368};
369
370/*
371 * The RDS notifier is used (optionally) to tell the application about
372 * completed RDMA operations. Rather than keeping the whole rds message
373 * around on the queue, we allocate a small notifier that is put on the
374 * socket's notifier_list. Notifications are delivered to the application
375 * through control messages.
376 */
377struct rds_notifier {
378 struct list_head n_list;
379 uint64_t n_user_token;
380 int n_status;
381};
382
383/**
384 * struct rds_transport - transport specific behavioural hooks
385 *
386 * @xmit: .xmit is called by rds_send_xmit() to tell the transport to send
387 * part of a message. The caller serializes on the send_sem so this
388 * doesn't need to be reentrant for a given conn. The header must be
389 * sent before the data payload. .xmit must be prepared to send a
390 * message with no data payload. .xmit should return the number of
391 * bytes that were sent down the connection, including header bytes.
392 * Returning 0 tells the caller that it doesn't need to perform any
393 * additional work now. This is usually the case when the transport has
394 * filled the sending queue for its connection and will handle
395 * triggering the rds thread to continue the send when space becomes
396 * available. Returning -EAGAIN tells the caller to retry the send
397 * immediately. Returning -ENOMEM tells the caller to retry the send at
398 * some point in the future.
399 *
400 * @conn_shutdown: conn_shutdown stops traffic on the given connection. Once
401 * it returns the connection can not call rds_recv_incoming().
402 * This will only be called once after conn_connect returns
403 * non-zero success and will The caller serializes this with
404 * the send and connecting paths (xmit_* and conn_*). The
405 * transport is responsible for other serialization, including
406 * rds_recv_incoming(). This is called in process context but
407 * should try hard not to block.
Andy Grover39de8282009-02-24 15:30:19 +0000408 */
409
Andy Grover335776b2009-08-21 12:28:34 +0000410#define RDS_TRANS_IB 0
411#define RDS_TRANS_IWARP 1
412#define RDS_TRANS_TCP 2
413#define RDS_TRANS_COUNT 3
414
Andy Grover39de8282009-02-24 15:30:19 +0000415struct rds_transport {
416 char t_name[TRANSNAMSIZ];
417 struct list_head t_item;
418 struct module *t_owner;
419 unsigned int t_prefer_loopback:1;
Andy Grover335776b2009-08-21 12:28:34 +0000420 unsigned int t_type;
Andy Grover39de8282009-02-24 15:30:19 +0000421
422 int (*laddr_check)(__be32 addr);
423 int (*conn_alloc)(struct rds_connection *conn, gfp_t gfp);
424 void (*conn_free)(void *data);
425 int (*conn_connect)(struct rds_connection *conn);
426 void (*conn_shutdown)(struct rds_connection *conn);
427 void (*xmit_prepare)(struct rds_connection *conn);
428 void (*xmit_complete)(struct rds_connection *conn);
429 int (*xmit)(struct rds_connection *conn, struct rds_message *rm,
430 unsigned int hdr_off, unsigned int sg, unsigned int off);
Andy Groverf8b3aaf2010-03-01 14:11:53 -0800431 int (*xmit_rdma)(struct rds_connection *conn, struct rm_rdma_op *op);
Andy Groverff3d7d32010-03-01 14:03:09 -0800432 int (*xmit_atomic)(struct rds_connection *conn, struct rm_atomic_op *op);
Andy Grover39de8282009-02-24 15:30:19 +0000433 int (*recv)(struct rds_connection *conn);
Al Viroc310e722014-11-20 09:21:14 -0500434 int (*inc_copy_to_user)(struct rds_incoming *inc, struct iov_iter *to);
Andy Grover39de8282009-02-24 15:30:19 +0000435 void (*inc_free)(struct rds_incoming *inc);
436
437 int (*cm_handle_connect)(struct rdma_cm_id *cm_id,
438 struct rdma_cm_event *event);
439 int (*cm_initiate_connect)(struct rdma_cm_id *cm_id);
440 void (*cm_connect_complete)(struct rds_connection *conn,
441 struct rdma_cm_event *event);
442
443 unsigned int (*stats_info_copy)(struct rds_info_iterator *iter,
444 unsigned int avail);
445 void (*exit)(void);
446 void *(*get_mr)(struct scatterlist *sg, unsigned long nr_sg,
447 struct rds_sock *rs, u32 *key_ret);
448 void (*sync_mr)(void *trans_private, int direction);
449 void (*free_mr)(void *trans_private, int invalidate);
450 void (*flush_mrs)(void);
451};
452
453struct rds_sock {
454 struct sock rs_sk;
455
456 u64 rs_user_addr;
457 u64 rs_user_bytes;
458
459 /*
460 * bound_addr used for both incoming and outgoing, no INADDR_ANY
461 * support.
462 */
Chris Mason38a4e5e2010-05-11 15:09:45 -0700463 struct hlist_node rs_bound_node;
Andy Grover39de8282009-02-24 15:30:19 +0000464 __be32 rs_bound_addr;
465 __be32 rs_conn_addr;
466 __be16 rs_bound_port;
467 __be16 rs_conn_port;
Andy Grover39de8282009-02-24 15:30:19 +0000468 struct rds_transport *rs_transport;
469
470 /*
471 * rds_sendmsg caches the conn it used the last time around.
472 * This helps avoid costly lookups.
473 */
474 struct rds_connection *rs_conn;
475
476 /* flag indicating we were congested or not */
477 int rs_congested;
Andy Groverb98ba522010-03-11 13:50:04 +0000478 /* seen congestion (ENOBUFS) when sending? */
479 int rs_seen_congestion;
Andy Grover39de8282009-02-24 15:30:19 +0000480
481 /* rs_lock protects all these adjacent members before the newline */
482 spinlock_t rs_lock;
483 struct list_head rs_send_queue;
484 u32 rs_snd_bytes;
485 int rs_rcv_bytes;
486 struct list_head rs_notify_queue; /* currently used for failed RDMAs */
487
488 /* Congestion wake_up. If rs_cong_monitor is set, we use cong_mask
489 * to decide whether the application should be woken up.
490 * If not set, we use rs_cong_track to find out whether a cong map
491 * update arrived.
492 */
493 uint64_t rs_cong_mask;
494 uint64_t rs_cong_notify;
495 struct list_head rs_cong_list;
496 unsigned long rs_cong_track;
497
498 /*
499 * rs_recv_lock protects the receive queue, and is
500 * used to serialize with rds_release.
501 */
502 rwlock_t rs_recv_lock;
503 struct list_head rs_recv_queue;
504
505 /* just for stats reporting */
506 struct list_head rs_item;
507
508 /* these have their own lock */
509 spinlock_t rs_rdma_lock;
510 struct rb_root rs_rdma_keys;
511
512 /* Socket options - in case there will be more */
513 unsigned char rs_recverr,
514 rs_cong_monitor;
515};
516
517static inline struct rds_sock *rds_sk_to_rs(const struct sock *sk)
518{
519 return container_of(sk, struct rds_sock, rs_sk);
520}
521static inline struct sock *rds_rs_to_sk(struct rds_sock *rs)
522{
523 return &rs->rs_sk;
524}
525
526/*
527 * The stack assigns sk_sndbuf and sk_rcvbuf to twice the specified value
528 * to account for overhead. We don't account for overhead, we just apply
529 * the number of payload bytes to the specified value.
530 */
531static inline int rds_sk_sndbuf(struct rds_sock *rs)
532{
533 return rds_rs_to_sk(rs)->sk_sndbuf / 2;
534}
535static inline int rds_sk_rcvbuf(struct rds_sock *rs)
536{
537 return rds_rs_to_sk(rs)->sk_rcvbuf / 2;
538}
539
540struct rds_statistics {
541 uint64_t s_conn_reset;
542 uint64_t s_recv_drop_bad_checksum;
543 uint64_t s_recv_drop_old_seq;
544 uint64_t s_recv_drop_no_sock;
545 uint64_t s_recv_drop_dead_sock;
546 uint64_t s_recv_deliver_raced;
547 uint64_t s_recv_delivered;
548 uint64_t s_recv_queued;
549 uint64_t s_recv_immediate_retry;
550 uint64_t s_recv_delayed_retry;
551 uint64_t s_recv_ack_required;
552 uint64_t s_recv_rdma_bytes;
553 uint64_t s_recv_ping;
554 uint64_t s_send_queue_empty;
555 uint64_t s_send_queue_full;
Andy Grover049ee3f2010-03-23 17:39:07 -0700556 uint64_t s_send_lock_contention;
557 uint64_t s_send_lock_queue_raced;
Andy Grover39de8282009-02-24 15:30:19 +0000558 uint64_t s_send_immediate_retry;
559 uint64_t s_send_delayed_retry;
560 uint64_t s_send_drop_acked;
561 uint64_t s_send_ack_required;
562 uint64_t s_send_queued;
563 uint64_t s_send_rdma;
564 uint64_t s_send_rdma_bytes;
565 uint64_t s_send_pong;
566 uint64_t s_page_remainder_hit;
567 uint64_t s_page_remainder_miss;
568 uint64_t s_copy_to_user;
569 uint64_t s_copy_from_user;
570 uint64_t s_cong_update_queued;
571 uint64_t s_cong_update_received;
572 uint64_t s_cong_send_error;
573 uint64_t s_cong_send_blocked;
574};
575
576/* af_rds.c */
Zach Brown59f740a2010-08-03 13:52:47 -0700577char *rds_str_array(char **array, size_t elements, size_t index);
Andy Grover39de8282009-02-24 15:30:19 +0000578void rds_sock_addref(struct rds_sock *rs);
579void rds_sock_put(struct rds_sock *rs);
580void rds_wake_sk_sleep(struct rds_sock *rs);
581static inline void __rds_wake_sk_sleep(struct sock *sk)
582{
Eric Dumazetaa395142010-04-20 13:03:51 +0000583 wait_queue_head_t *waitq = sk_sleep(sk);
Andy Grover39de8282009-02-24 15:30:19 +0000584
585 if (!sock_flag(sk, SOCK_DEAD) && waitq)
586 wake_up(waitq);
587}
588extern wait_queue_head_t rds_poll_waitq;
589
590
591/* bind.c */
592int rds_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len);
593void rds_remove_bound(struct rds_sock *rs);
594struct rds_sock *rds_find_bound(__be32 addr, __be16 port);
595
596/* cong.c */
597int rds_cong_get_maps(struct rds_connection *conn);
598void rds_cong_add_conn(struct rds_connection *conn);
599void rds_cong_remove_conn(struct rds_connection *conn);
600void rds_cong_set_bit(struct rds_cong_map *map, __be16 port);
601void rds_cong_clear_bit(struct rds_cong_map *map, __be16 port);
602int rds_cong_wait(struct rds_cong_map *map, __be16 port, int nonblock, struct rds_sock *rs);
603void rds_cong_queue_updates(struct rds_cong_map *map);
604void rds_cong_map_updated(struct rds_cong_map *map, uint64_t);
605int rds_cong_updated_since(unsigned long *recent);
606void rds_cong_add_socket(struct rds_sock *);
607void rds_cong_remove_socket(struct rds_sock *);
608void rds_cong_exit(void);
609struct rds_message *rds_cong_update_alloc(struct rds_connection *conn);
610
611/* conn.c */
Zach Brownef87b7e2010-07-09 12:26:20 -0700612int rds_conn_init(void);
Andy Grover39de8282009-02-24 15:30:19 +0000613void rds_conn_exit(void);
614struct rds_connection *rds_conn_create(__be32 laddr, __be32 faddr,
615 struct rds_transport *trans, gfp_t gfp);
616struct rds_connection *rds_conn_create_outgoing(__be32 laddr, __be32 faddr,
617 struct rds_transport *trans, gfp_t gfp);
Andy Grover2dc39352010-06-11 13:49:13 -0700618void rds_conn_shutdown(struct rds_connection *conn);
Andy Grover39de8282009-02-24 15:30:19 +0000619void rds_conn_destroy(struct rds_connection *conn);
Andy Grover39de8282009-02-24 15:30:19 +0000620void rds_conn_drop(struct rds_connection *conn);
Zach Brownf3c68082010-05-24 13:14:36 -0700621void rds_conn_connect_if_down(struct rds_connection *conn);
Andy Grover39de8282009-02-24 15:30:19 +0000622void rds_for_each_conn_info(struct socket *sock, unsigned int len,
623 struct rds_info_iterator *iter,
624 struct rds_info_lengths *lens,
625 int (*visitor)(struct rds_connection *, void *),
626 size_t item_len);
Joe Perchesb9075fa2011-10-31 17:11:33 -0700627__printf(2, 3)
628void __rds_conn_error(struct rds_connection *conn, const char *, ...);
Andy Grover39de8282009-02-24 15:30:19 +0000629#define rds_conn_error(conn, fmt...) \
630 __rds_conn_error(conn, KERN_WARNING "RDS: " fmt)
631
632static inline int
633rds_conn_transition(struct rds_connection *conn, int old, int new)
634{
635 return atomic_cmpxchg(&conn->c_state, old, new) == old;
636}
637
638static inline int
639rds_conn_state(struct rds_connection *conn)
640{
641 return atomic_read(&conn->c_state);
642}
643
644static inline int
645rds_conn_up(struct rds_connection *conn)
646{
647 return atomic_read(&conn->c_state) == RDS_CONN_UP;
648}
649
650static inline int
651rds_conn_connecting(struct rds_connection *conn)
652{
653 return atomic_read(&conn->c_state) == RDS_CONN_CONNECTING;
654}
655
656/* message.c */
657struct rds_message *rds_message_alloc(unsigned int nents, gfp_t gfp);
Andy Groverfc445082010-01-12 12:56:06 -0800658struct scatterlist *rds_message_alloc_sgs(struct rds_message *rm, int nents);
Al Viro083735f2014-11-20 09:31:08 -0500659int rds_message_copy_from_user(struct rds_message *rm, struct iov_iter *from);
Andy Grover39de8282009-02-24 15:30:19 +0000660struct rds_message *rds_message_map_pages(unsigned long *page_addrs, unsigned int total_len);
661void rds_message_populate_header(struct rds_header *hdr, __be16 sport,
662 __be16 dport, u64 seq);
663int rds_message_add_extension(struct rds_header *hdr,
664 unsigned int type, const void *data, unsigned int len);
665int rds_message_next_extension(struct rds_header *hdr,
666 unsigned int *pos, void *buf, unsigned int *buflen);
Andy Grover39de8282009-02-24 15:30:19 +0000667int rds_message_add_rdma_dest_extension(struct rds_header *hdr, u32 r_key, u32 offset);
Al Viroc310e722014-11-20 09:21:14 -0500668int rds_message_inc_copy_to_user(struct rds_incoming *inc, struct iov_iter *to);
Andy Grover39de8282009-02-24 15:30:19 +0000669void rds_message_inc_free(struct rds_incoming *inc);
670void rds_message_addref(struct rds_message *rm);
671void rds_message_put(struct rds_message *rm);
672void rds_message_wait(struct rds_message *rm);
673void rds_message_unmapped(struct rds_message *rm);
674
675static inline void rds_message_make_checksum(struct rds_header *hdr)
676{
677 hdr->h_csum = 0;
678 hdr->h_csum = ip_fast_csum((void *) hdr, sizeof(*hdr) >> 2);
679}
680
681static inline int rds_message_verify_checksum(const struct rds_header *hdr)
682{
683 return !hdr->h_csum || ip_fast_csum((void *) hdr, sizeof(*hdr) >> 2) == 0;
684}
685
686
687/* page.c */
688int rds_page_remainder_alloc(struct scatterlist *scat, unsigned long bytes,
689 gfp_t gfp);
690int rds_page_copy_user(struct page *page, unsigned long offset,
691 void __user *ptr, unsigned long bytes,
692 int to_user);
693#define rds_page_copy_to_user(page, offset, ptr, bytes) \
694 rds_page_copy_user(page, offset, ptr, bytes, 1)
695#define rds_page_copy_from_user(page, offset, ptr, bytes) \
696 rds_page_copy_user(page, offset, ptr, bytes, 0)
697void rds_page_exit(void);
698
699/* recv.c */
700void rds_inc_init(struct rds_incoming *inc, struct rds_connection *conn,
701 __be32 saddr);
Andy Grover39de8282009-02-24 15:30:19 +0000702void rds_inc_put(struct rds_incoming *inc);
703void rds_recv_incoming(struct rds_connection *conn, __be32 saddr, __be32 daddr,
Cong Wang6114eab2011-11-25 23:14:40 +0800704 struct rds_incoming *inc, gfp_t gfp);
Andy Grover39de8282009-02-24 15:30:19 +0000705int rds_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
706 size_t size, int msg_flags);
707void rds_clear_recv_queue(struct rds_sock *rs);
708int rds_notify_queue_get(struct rds_sock *rs, struct msghdr *msg);
709void rds_inc_info_copy(struct rds_incoming *inc,
710 struct rds_info_iterator *iter,
711 __be32 saddr, __be32 daddr, int flip);
712
713/* send.c */
714int rds_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
715 size_t payload_len);
716void rds_send_reset(struct rds_connection *conn);
717int rds_send_xmit(struct rds_connection *conn);
718struct sockaddr_in;
719void rds_send_drop_to(struct rds_sock *rs, struct sockaddr_in *dest);
720typedef int (*is_acked_func)(struct rds_message *rm, uint64_t ack);
721void rds_send_drop_acked(struct rds_connection *conn, u64 ack,
722 is_acked_func is_acked);
Andy Grover39de8282009-02-24 15:30:19 +0000723int rds_send_pong(struct rds_connection *conn, __be16 dport);
724struct rds_message *rds_send_get_message(struct rds_connection *,
Andy Groverf8b3aaf2010-03-01 14:11:53 -0800725 struct rm_rdma_op *);
Andy Grover39de8282009-02-24 15:30:19 +0000726
727/* rdma.c */
728void rds_rdma_unuse(struct rds_sock *rs, u32 r_key, int force);
Andy Grover21f79af2010-01-12 12:57:27 -0800729int rds_get_mr(struct rds_sock *rs, char __user *optval, int optlen);
730int rds_get_mr_for_dest(struct rds_sock *rs, char __user *optval, int optlen);
731int rds_free_mr(struct rds_sock *rs, char __user *optval, int optlen);
732void rds_rdma_drop_keys(struct rds_sock *rs);
733int rds_rdma_extra_size(struct rds_rdma_args *args);
734int rds_cmsg_rdma_args(struct rds_sock *rs, struct rds_message *rm,
735 struct cmsghdr *cmsg);
736int rds_cmsg_rdma_dest(struct rds_sock *rs, struct rds_message *rm,
737 struct cmsghdr *cmsg);
738int rds_cmsg_rdma_args(struct rds_sock *rs, struct rds_message *rm,
739 struct cmsghdr *cmsg);
740int rds_cmsg_rdma_map(struct rds_sock *rs, struct rds_message *rm,
741 struct cmsghdr *cmsg);
Andy Groverf8b3aaf2010-03-01 14:11:53 -0800742void rds_rdma_free_op(struct rm_rdma_op *ro);
Andy Groverd0ab25a2010-01-27 16:15:48 -0800743void rds_atomic_free_op(struct rm_atomic_op *ao);
Andy Grover15133f62010-01-12 14:33:38 -0800744void rds_rdma_send_complete(struct rds_message *rm, int wc_status);
745void rds_atomic_send_complete(struct rds_message *rm, int wc_status);
746int rds_cmsg_atomic(struct rds_sock *rs, struct rds_message *rm,
747 struct cmsghdr *cmsg);
Andy Grover21f79af2010-01-12 12:57:27 -0800748
Joe Perchesc1b12032013-10-18 13:48:25 -0700749void __rds_put_mr_final(struct rds_mr *mr);
Andy Grover21f79af2010-01-12 12:57:27 -0800750static inline void rds_mr_put(struct rds_mr *mr)
751{
752 if (atomic_dec_and_test(&mr->r_refcount))
753 __rds_put_mr_final(mr);
754}
Andy Grover39de8282009-02-24 15:30:19 +0000755
756/* stats.c */
David Howells9b8de742009-04-21 23:00:24 +0100757DECLARE_PER_CPU_SHARED_ALIGNED(struct rds_statistics, rds_stats);
Andy Grover39de8282009-02-24 15:30:19 +0000758#define rds_stats_inc_which(which, member) do { \
759 per_cpu(which, get_cpu()).member++; \
760 put_cpu(); \
761} while (0)
762#define rds_stats_inc(member) rds_stats_inc_which(rds_stats, member)
763#define rds_stats_add_which(which, member, count) do { \
764 per_cpu(which, get_cpu()).member += count; \
765 put_cpu(); \
766} while (0)
767#define rds_stats_add(member, count) rds_stats_add_which(rds_stats, member, count)
Zach Brownef87b7e2010-07-09 12:26:20 -0700768int rds_stats_init(void);
Andy Grover39de8282009-02-24 15:30:19 +0000769void rds_stats_exit(void);
770void rds_stats_info_copy(struct rds_info_iterator *iter,
Jan Engelhardt36cbd3d2009-08-05 10:42:58 -0700771 uint64_t *values, const char *const *names,
772 size_t nr);
Andy Grover39de8282009-02-24 15:30:19 +0000773
774/* sysctl.c */
Zach Brownef87b7e2010-07-09 12:26:20 -0700775int rds_sysctl_init(void);
Andy Grover39de8282009-02-24 15:30:19 +0000776void rds_sysctl_exit(void);
777extern unsigned long rds_sysctl_sndbuf_min;
778extern unsigned long rds_sysctl_sndbuf_default;
779extern unsigned long rds_sysctl_sndbuf_max;
780extern unsigned long rds_sysctl_reconnect_min_jiffies;
781extern unsigned long rds_sysctl_reconnect_max_jiffies;
782extern unsigned int rds_sysctl_max_unacked_packets;
783extern unsigned int rds_sysctl_max_unacked_bytes;
784extern unsigned int rds_sysctl_ping_enable;
785extern unsigned long rds_sysctl_trace_flags;
786extern unsigned int rds_sysctl_trace_level;
787
788/* threads.c */
Zach Brownef87b7e2010-07-09 12:26:20 -0700789int rds_threads_init(void);
Andy Grover39de8282009-02-24 15:30:19 +0000790void rds_threads_exit(void);
791extern struct workqueue_struct *rds_wq;
Andy Grover2dc39352010-06-11 13:49:13 -0700792void rds_queue_reconnect(struct rds_connection *conn);
Andy Grover39de8282009-02-24 15:30:19 +0000793void rds_connect_worker(struct work_struct *);
794void rds_shutdown_worker(struct work_struct *);
795void rds_send_worker(struct work_struct *);
796void rds_recv_worker(struct work_struct *);
797void rds_connect_complete(struct rds_connection *conn);
798
799/* transport.c */
800int rds_trans_register(struct rds_transport *trans);
801void rds_trans_unregister(struct rds_transport *trans);
802struct rds_transport *rds_trans_get_preferred(__be32 addr);
Zach Brown5adb5bc2010-07-23 10:32:31 -0700803void rds_trans_put(struct rds_transport *trans);
Andy Grover39de8282009-02-24 15:30:19 +0000804unsigned int rds_trans_stats_info_copy(struct rds_info_iterator *iter,
805 unsigned int avail);
Zach Brownef87b7e2010-07-09 12:26:20 -0700806int rds_trans_init(void);
Andy Grover39de8282009-02-24 15:30:19 +0000807void rds_trans_exit(void);
808
809#endif