Blame view
net/rds/send.c
30 KB
5c1155904 RDS: send.c |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 |
/* * Copyright (c) 2006 Oracle. All rights reserved. * * This software is available to you under a choice of one of two * licenses. You may choose to be licensed under the terms of the GNU * General Public License (GPL) Version 2, available from the file * COPYING in the main directory of this source tree, or the * OpenIB.org BSD license below: * * Redistribution and use in source and binary forms, with or * without modification, are permitted provided that the following * conditions are met: * * - Redistributions of source code must retain the above * copyright notice, this list of conditions and the following * disclaimer. * * - Redistributions in binary form must reproduce the above * copyright notice, this list of conditions and the following * disclaimer in the documentation and/or other materials * provided with the distribution. * * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE * SOFTWARE. * */ #include <linux/kernel.h> |
d9b938421 net: add modulepa... |
34 |
#include <linux/moduleparam.h> |
5a0e3ad6a include cleanup: ... |
35 |
#include <linux/gfp.h> |
5c1155904 RDS: send.c |
36 37 38 |
#include <net/sock.h> #include <linux/in.h> #include <linux/list.h> |
cb0a60564 net/rds: use prin... |
39 |
#include <linux/ratelimit.h> |
bc3b2d7fb net: Add export.h... |
40 |
#include <linux/export.h> |
5c1155904 RDS: send.c |
41 42 |
#include "rds.h" |
5c1155904 RDS: send.c |
43 44 45 46 47 48 49 50 51 52 53 54 55 56 |
/* When transmitting messages in rds_send_xmit, we need to emerge from * time to time and briefly release the CPU. Otherwise the softlock watchdog * will kick our shin. * Also, it seems fairer to not let one busy connection stall all the * others. * * send_batch_count is the number of times we'll loop in send_xmit. Setting * it to 0 will restore the old behavior (where we looped until we had * drained the queue). */ static int send_batch_count = 64; module_param(send_batch_count, int, 0444); MODULE_PARM_DESC(send_batch_count, " batch factor when working the send queue"); |
ff51bf841 rds: make local f... |
57 |
static void rds_send_remove_from_sock(struct list_head *messages, int status); |
5c1155904 RDS: send.c |
58 |
/* |
0f4b1c7e8 rds: fix rds_send... |
59 60 |
* Reset the send state. Callers must ensure that this doesn't race with * rds_send_xmit(). |
5c1155904 RDS: send.c |
61 62 63 64 65 66 67 |
*/ void rds_send_reset(struct rds_connection *conn) { struct rds_message *rm, *tmp; unsigned long flags; if (conn->c_xmit_rm) { |
7e3f2952e rds: don't let RD... |
68 69 |
rm = conn->c_xmit_rm; conn->c_xmit_rm = NULL; |
5c1155904 RDS: send.c |
70 71 72 73 |
/* Tell the user the RDMA op is no longer mapped by the * transport. This isn't entirely true (it's flushed out * independently) but as the connection is down, there's * no ongoing RDMA to/from that memory */ |
7e3f2952e rds: don't let RD... |
74 |
rds_message_unmapped(rm); |
7e3f2952e rds: don't let RD... |
75 |
rds_message_put(rm); |
5c1155904 RDS: send.c |
76 |
} |
7e3f2952e rds: don't let RD... |
77 |
|
5c1155904 RDS: send.c |
78 79 80 |
conn->c_xmit_sg = 0; conn->c_xmit_hdr_off = 0; conn->c_xmit_data_off = 0; |
15133f6e6 RDS: Implement at... |
81 |
conn->c_xmit_atomic_sent = 0; |
5b2366bd2 RDS: Rewrite rds_... |
82 83 |
conn->c_xmit_rdma_sent = 0; conn->c_xmit_data_sent = 0; |
5c1155904 RDS: send.c |
84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 |
conn->c_map_queued = 0; conn->c_unacked_packets = rds_sysctl_max_unacked_packets; conn->c_unacked_bytes = rds_sysctl_max_unacked_bytes; /* Mark messages as retransmissions, and move them to the send q */ spin_lock_irqsave(&conn->c_lock, flags); list_for_each_entry_safe(rm, tmp, &conn->c_retrans, m_conn_item) { set_bit(RDS_MSG_ACK_REQUIRED, &rm->m_flags); set_bit(RDS_MSG_RETRANSMITTED, &rm->m_flags); } list_splice_init(&conn->c_retrans, &conn->c_send_queue); spin_unlock_irqrestore(&conn->c_lock, flags); } |
0f4b1c7e8 rds: fix rds_send... |
99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 |
static int acquire_in_xmit(struct rds_connection *conn) { return test_and_set_bit(RDS_IN_XMIT, &conn->c_flags) == 0; } static void release_in_xmit(struct rds_connection *conn) { clear_bit(RDS_IN_XMIT, &conn->c_flags); smp_mb__after_clear_bit(); /* * We don't use wait_on_bit()/wake_up_bit() because our waking is in a * hot path and finding waiters is very rare. We don't want to walk * the system-wide hashed waitqueue buckets in the fast path only to * almost never find waiters. */ if (waitqueue_active(&conn->c_waitq)) wake_up_all(&conn->c_waitq); } |
5c1155904 RDS: send.c |
117 |
/* |
25985edce Fix common misspe... |
118 |
* We're making the conscious trade-off here to only send one message |
5c1155904 RDS: send.c |
119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 |
* down the connection at a time. * Pro: * - tx queueing is a simple fifo list * - reassembly is optional and easily done by transports per conn * - no per flow rx lookup at all, straight to the socket * - less per-frag memory and wire overhead * Con: * - queued acks can be delayed behind large messages * Depends: * - small message latency is higher behind queued large messages * - large message latency isn't starved by intervening small sends */ int rds_send_xmit(struct rds_connection *conn) { struct rds_message *rm; unsigned long flags; unsigned int tmp; |
5c1155904 RDS: send.c |
136 137 |
struct scatterlist *sg; int ret = 0; |
5c1155904 RDS: send.c |
138 |
LIST_HEAD(to_be_dropped); |
fcc5450c6 RDS: Remove send_... |
139 |
restart: |
049ee3f50 RDS: Change send ... |
140 |
|
5c1155904 RDS: send.c |
141 142 143 144 145 146 |
/* * sendmsg calls here after having queued its message on the send * queue. We only have one task feeding the connection at a time. If * another thread is already feeding the queue then we back off. This * avoids blocking the caller and trading per-connection data between * caches per message. |
5c1155904 RDS: send.c |
147 |
*/ |
0f4b1c7e8 rds: fix rds_send... |
148 |
if (!acquire_in_xmit(conn)) { |
049ee3f50 RDS: Change send ... |
149 |
rds_stats_inc(s_send_lock_contention); |
5c1155904 RDS: send.c |
150 151 152 |
ret = -ENOMEM; goto out; } |
0f4b1c7e8 rds: fix rds_send... |
153 154 155 156 157 158 159 160 161 162 |
/* * rds_conn_shutdown() sets the conn state and then tests RDS_IN_XMIT, * we do the opposite to avoid races. */ if (!rds_conn_up(conn)) { release_in_xmit(conn); ret = 0; goto out; } |
5c1155904 RDS: send.c |
163 164 165 166 167 168 |
if (conn->c_trans->xmit_prepare) conn->c_trans->xmit_prepare(conn); /* * spin trying to push headers and data down the connection until |
5b2366bd2 RDS: Rewrite rds_... |
169 |
* the connection doesn't make forward progress. |
5c1155904 RDS: send.c |
170 |
*/ |
fcc5450c6 RDS: Remove send_... |
171 |
while (1) { |
5c1155904 RDS: send.c |
172 |
|
5c1155904 RDS: send.c |
173 |
rm = conn->c_xmit_rm; |
5c1155904 RDS: send.c |
174 |
|
5b2366bd2 RDS: Rewrite rds_... |
175 176 177 |
/* * If between sending messages, we can send a pending congestion * map update. |
5c1155904 RDS: send.c |
178 |
*/ |
8690bfa17 RDS: cleanup: rem... |
179 |
if (!rm && test_and_clear_bit(0, &conn->c_map_queued)) { |
77dd550e5 RDS: Stop support... |
180 181 182 183 |
rm = rds_cong_update_alloc(conn); if (IS_ERR(rm)) { ret = PTR_ERR(rm); break; |
5b2366bd2 RDS: Rewrite rds_... |
184 |
} |
77dd550e5 RDS: Stop support... |
185 186 187 |
rm->data.op_active = 1; conn->c_xmit_rm = rm; |
5c1155904 RDS: send.c |
188 189 190 |
} /* |
5b2366bd2 RDS: Rewrite rds_... |
191 |
* If not already working on one, grab the next message. |
5c1155904 RDS: send.c |
192 193 194 195 196 |
* * c_xmit_rm holds a ref while we're sending this message down * the connction. We can use this ref while holding the * send_sem.. rds_send_reset() is serialized with it. */ |
8690bfa17 RDS: cleanup: rem... |
197 |
if (!rm) { |
5c1155904 RDS: send.c |
198 |
unsigned int len; |
0f4b1c7e8 rds: fix rds_send... |
199 |
spin_lock_irqsave(&conn->c_lock, flags); |
5c1155904 RDS: send.c |
200 201 202 203 204 205 206 207 208 209 210 211 212 |
if (!list_empty(&conn->c_send_queue)) { rm = list_entry(conn->c_send_queue.next, struct rds_message, m_conn_item); rds_message_addref(rm); /* * Move the message from the send queue to the retransmit * list right away. */ list_move_tail(&rm->m_conn_item, &conn->c_retrans); } |
0f4b1c7e8 rds: fix rds_send... |
213 |
spin_unlock_irqrestore(&conn->c_lock, flags); |
5c1155904 RDS: send.c |
214 |
|
fcc5450c6 RDS: Remove send_... |
215 |
if (!rm) |
5c1155904 RDS: send.c |
216 |
break; |
5c1155904 RDS: send.c |
217 218 219 220 221 222 223 224 |
/* Unfortunately, the way Infiniband deals with * RDMA to a bad MR key is by moving the entire * queue pair to error state. We cold possibly * recover from that, but right now we drop the * connection. * Therefore, we never retransmit messages with RDMA ops. */ |
f8b3aaf2b RDS: Remove struc... |
225 |
if (rm->rdma.op_active && |
f64f9e719 net: Move && and ... |
226 |
test_bit(RDS_MSG_RETRANSMITTED, &rm->m_flags)) { |
0f4b1c7e8 rds: fix rds_send... |
227 |
spin_lock_irqsave(&conn->c_lock, flags); |
5c1155904 RDS: send.c |
228 229 |
if (test_and_clear_bit(RDS_MSG_ON_CONN, &rm->m_flags)) list_move(&rm->m_conn_item, &to_be_dropped); |
0f4b1c7e8 rds: fix rds_send... |
230 |
spin_unlock_irqrestore(&conn->c_lock, flags); |
5c1155904 RDS: send.c |
231 232 233 234 235 |
continue; } /* Require an ACK every once in a while */ len = ntohl(rm->m_inc.i_hdr.h_len); |
f64f9e719 net: Move && and ... |
236 237 |
if (conn->c_unacked_packets == 0 || conn->c_unacked_bytes < len) { |
5c1155904 RDS: send.c |
238 239 240 241 242 243 244 245 246 247 248 249 |
__set_bit(RDS_MSG_ACK_REQUIRED, &rm->m_flags); conn->c_unacked_packets = rds_sysctl_max_unacked_packets; conn->c_unacked_bytes = rds_sysctl_max_unacked_bytes; rds_stats_inc(s_send_ack_required); } else { conn->c_unacked_bytes -= len; conn->c_unacked_packets--; } conn->c_xmit_rm = rm; } |
2c3a5f9ab RDS: Add flag for... |
250 251 |
/* The transport either sends the whole rdma or none of it */ if (rm->rdma.op_active && !conn->c_xmit_rdma_sent) { |
ff3d7d361 RDS: Perform unma... |
252 |
rm->m_final_op = &rm->rdma; |
2c3a5f9ab RDS: Add flag for... |
253 |
ret = conn->c_trans->xmit_rdma(conn, &rm->rdma); |
1cc2228c5 rds: Fix referenc... |
254 |
if (ret) |
15133f6e6 RDS: Implement at... |
255 |
break; |
2c3a5f9ab RDS: Add flag for... |
256 |
conn->c_xmit_rdma_sent = 1; |
15133f6e6 RDS: Implement at... |
257 258 259 260 |
/* The transport owns the mapped memory for now. * You can't unmap it while it's on the send queue */ set_bit(RDS_MSG_MAPPED, &rm->m_flags); } |
2c3a5f9ab RDS: Add flag for... |
261 |
if (rm->atomic.op_active && !conn->c_xmit_atomic_sent) { |
ff3d7d361 RDS: Perform unma... |
262 263 |
rm->m_final_op = &rm->atomic; ret = conn->c_trans->xmit_atomic(conn, &rm->atomic); |
1cc2228c5 rds: Fix referenc... |
264 |
if (ret) |
5c1155904 RDS: send.c |
265 |
break; |
2c3a5f9ab RDS: Add flag for... |
266 |
conn->c_xmit_atomic_sent = 1; |
ff3d7d361 RDS: Perform unma... |
267 |
|
5c1155904 RDS: send.c |
268 269 270 271 |
/* The transport owns the mapped memory for now. * You can't unmap it while it's on the send queue */ set_bit(RDS_MSG_MAPPED, &rm->m_flags); } |
2c3a5f9ab RDS: Add flag for... |
272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 |
/* * A number of cases require an RDS header to be sent * even if there is no data. * We permit 0-byte sends; rds-ping depends on this. * However, if there are exclusively attached silent ops, * we skip the hdr/data send, to enable silent operation. */ if (rm->data.op_nents == 0) { int ops_present; int all_ops_are_silent = 1; ops_present = (rm->atomic.op_active || rm->rdma.op_active); if (rm->atomic.op_active && !rm->atomic.op_silent) all_ops_are_silent = 0; if (rm->rdma.op_active && !rm->rdma.op_silent) all_ops_are_silent = 0; if (ops_present && all_ops_are_silent && !rm->m_rdma_cookie) rm->data.op_active = 0; } |
5b2366bd2 RDS: Rewrite rds_... |
293 |
if (rm->data.op_active && !conn->c_xmit_data_sent) { |
ff3d7d361 RDS: Perform unma... |
294 |
rm->m_final_op = &rm->data; |
5c1155904 RDS: send.c |
295 296 297 298 299 300 301 302 303 304 305 306 307 308 |
ret = conn->c_trans->xmit(conn, rm, conn->c_xmit_hdr_off, conn->c_xmit_sg, conn->c_xmit_data_off); if (ret <= 0) break; if (conn->c_xmit_hdr_off < sizeof(struct rds_header)) { tmp = min_t(int, ret, sizeof(struct rds_header) - conn->c_xmit_hdr_off); conn->c_xmit_hdr_off += tmp; ret -= tmp; } |
6c7cc6e46 RDS: Rename data ... |
309 |
sg = &rm->data.op_sg[conn->c_xmit_sg]; |
5c1155904 RDS: send.c |
310 311 312 313 314 315 316 317 318 319 |
while (ret) { tmp = min_t(int, ret, sg->length - conn->c_xmit_data_off); conn->c_xmit_data_off += tmp; ret -= tmp; if (conn->c_xmit_data_off == sg->length) { conn->c_xmit_data_off = 0; sg++; conn->c_xmit_sg++; BUG_ON(ret != 0 && |
6c7cc6e46 RDS: Rename data ... |
320 |
conn->c_xmit_sg == rm->data.op_nents); |
5c1155904 RDS: send.c |
321 322 |
} } |
5b2366bd2 RDS: Rewrite rds_... |
323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 |
if (conn->c_xmit_hdr_off == sizeof(struct rds_header) && (conn->c_xmit_sg == rm->data.op_nents)) conn->c_xmit_data_sent = 1; } /* * A rm will only take multiple times through this loop * if there is a data op. Thus, if the data is sent (or there was * none), then we're done with the rm. */ if (!rm->data.op_active || conn->c_xmit_data_sent) { conn->c_xmit_rm = NULL; conn->c_xmit_sg = 0; conn->c_xmit_hdr_off = 0; conn->c_xmit_data_off = 0; conn->c_xmit_rdma_sent = 0; conn->c_xmit_atomic_sent = 0; conn->c_xmit_data_sent = 0; rds_message_put(rm); |
5c1155904 RDS: send.c |
344 345 |
} } |
5c1155904 RDS: send.c |
346 347 |
if (conn->c_trans->xmit_complete) conn->c_trans->xmit_complete(conn); |
0f4b1c7e8 rds: fix rds_send... |
348 |
release_in_xmit(conn); |
5c1155904 RDS: send.c |
349 |
|
2ad8099b5 RDS: rds_send_xmi... |
350 351 352 353 354 355 356 |
/* Nuke any messages we decided not to retransmit. */ if (!list_empty(&to_be_dropped)) { /* irqs on here, so we can put(), unlike above */ list_for_each_entry(rm, &to_be_dropped, m_conn_item) rds_message_put(rm); rds_send_remove_from_sock(&to_be_dropped, RDS_RDMA_DROPPED); } |
fcc5450c6 RDS: Remove send_... |
357 |
/* |
0f4b1c7e8 rds: fix rds_send... |
358 359 360 361 362 |
* Other senders can queue a message after we last test the send queue * but before we clear RDS_IN_XMIT. In that case they'd back off and * not try and send their newly queued message. We need to check the * send queue after having cleared RDS_IN_XMIT so that their message * doesn't get stuck on the send queue. |
fcc5450c6 RDS: Remove send_... |
363 364 365 366 367 368 |
* * If the transport cannot continue (i.e ret != 0), then it must * call us when more room is available, such as from the tx * completion handler. */ if (ret == 0) { |
9e29db0e3 RDS: Use a genera... |
369 |
smp_mb(); |
5c1155904 RDS: send.c |
370 |
if (!list_empty(&conn->c_send_queue)) { |
049ee3f50 RDS: Change send ... |
371 |
rds_stats_inc(s_send_lock_queue_raced); |
0f4b1c7e8 rds: fix rds_send... |
372 |
goto restart; |
5c1155904 RDS: send.c |
373 |
} |
5c1155904 RDS: send.c |
374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 |
} out: return ret; } static void rds_send_sndbuf_remove(struct rds_sock *rs, struct rds_message *rm) { u32 len = be32_to_cpu(rm->m_inc.i_hdr.h_len); assert_spin_locked(&rs->rs_lock); BUG_ON(rs->rs_snd_bytes < len); rs->rs_snd_bytes -= len; if (rs->rs_snd_bytes == 0) rds_stats_inc(s_send_queue_empty); } static inline int rds_send_is_acked(struct rds_message *rm, u64 ack, is_acked_func is_acked) { if (is_acked) return is_acked(rm, ack); return be64_to_cpu(rm->m_inc.i_hdr.h_sequence) <= ack; } /* |
5c1155904 RDS: send.c |
401 402 403 404 405 406 407 408 |
* This is pretty similar to what happens below in the ACK * handling code - except that we call here as soon as we get * the IB send completion on the RDMA op and the accompanying * message. */ void rds_rdma_send_complete(struct rds_message *rm, int status) { struct rds_sock *rs = NULL; |
f8b3aaf2b RDS: Remove struc... |
409 |
struct rm_rdma_op *ro; |
5c1155904 RDS: send.c |
410 |
struct rds_notifier *notifier; |
9de0864cf RDS: Fix locking ... |
411 |
unsigned long flags; |
5c1155904 RDS: send.c |
412 |
|
9de0864cf RDS: Fix locking ... |
413 |
spin_lock_irqsave(&rm->m_rs_lock, flags); |
5c1155904 RDS: send.c |
414 |
|
f8b3aaf2b RDS: Remove struc... |
415 |
ro = &rm->rdma; |
f64f9e719 net: Move && and ... |
416 |
if (test_bit(RDS_MSG_ON_SOCK, &rm->m_flags) && |
f8b3aaf2b RDS: Remove struc... |
417 418 |
ro->op_active && ro->op_notify && ro->op_notifier) { notifier = ro->op_notifier; |
5c1155904 RDS: send.c |
419 420 421 422 423 424 425 |
rs = rm->m_rs; sock_hold(rds_rs_to_sk(rs)); notifier->n_status = status; spin_lock(&rs->rs_lock); list_add_tail(¬ifier->n_list, &rs->rs_notify_queue); spin_unlock(&rs->rs_lock); |
f8b3aaf2b RDS: Remove struc... |
426 |
ro->op_notifier = NULL; |
5c1155904 RDS: send.c |
427 |
} |
9de0864cf RDS: Fix locking ... |
428 |
spin_unlock_irqrestore(&rm->m_rs_lock, flags); |
5c1155904 RDS: send.c |
429 430 431 432 433 434 |
if (rs) { rds_wake_sk_sleep(rs); sock_put(rds_rs_to_sk(rs)); } } |
616b757ae RDS: Export symbo... |
435 |
EXPORT_SYMBOL_GPL(rds_rdma_send_complete); |
5c1155904 RDS: send.c |
436 437 |
/* |
15133f6e6 RDS: Implement at... |
438 439 440 441 442 443 444 |
* Just like above, except looks at atomic op */ void rds_atomic_send_complete(struct rds_message *rm, int status) { struct rds_sock *rs = NULL; struct rm_atomic_op *ao; struct rds_notifier *notifier; |
cf4b7389e RDS: Fix locking ... |
445 |
unsigned long flags; |
15133f6e6 RDS: Implement at... |
446 |
|
cf4b7389e RDS: Fix locking ... |
447 |
spin_lock_irqsave(&rm->m_rs_lock, flags); |
15133f6e6 RDS: Implement at... |
448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 |
ao = &rm->atomic; if (test_bit(RDS_MSG_ON_SOCK, &rm->m_flags) && ao->op_active && ao->op_notify && ao->op_notifier) { notifier = ao->op_notifier; rs = rm->m_rs; sock_hold(rds_rs_to_sk(rs)); notifier->n_status = status; spin_lock(&rs->rs_lock); list_add_tail(¬ifier->n_list, &rs->rs_notify_queue); spin_unlock(&rs->rs_lock); ao->op_notifier = NULL; } |
cf4b7389e RDS: Fix locking ... |
463 |
spin_unlock_irqrestore(&rm->m_rs_lock, flags); |
15133f6e6 RDS: Implement at... |
464 465 466 467 468 469 470 471 472 |
if (rs) { rds_wake_sk_sleep(rs); sock_put(rds_rs_to_sk(rs)); } } EXPORT_SYMBOL_GPL(rds_atomic_send_complete); /* |
5c1155904 RDS: send.c |
473 474 475 476 477 |
* This is the same as rds_rdma_send_complete except we * don't do any locking - we have all the ingredients (message, * socket, socket lock) and can just move the notifier. */ static inline void |
940786eb0 RDS: queue failur... |
478 |
__rds_send_complete(struct rds_sock *rs, struct rds_message *rm, int status) |
5c1155904 RDS: send.c |
479 |
{ |
f8b3aaf2b RDS: Remove struc... |
480 |
struct rm_rdma_op *ro; |
940786eb0 RDS: queue failur... |
481 |
struct rm_atomic_op *ao; |
5c1155904 RDS: send.c |
482 |
|
f8b3aaf2b RDS: Remove struc... |
483 484 485 486 487 |
ro = &rm->rdma; if (ro->op_active && ro->op_notify && ro->op_notifier) { ro->op_notifier->n_status = status; list_add_tail(&ro->op_notifier->n_list, &rs->rs_notify_queue); ro->op_notifier = NULL; |
5c1155904 RDS: send.c |
488 |
} |
940786eb0 RDS: queue failur... |
489 490 491 492 493 494 |
ao = &rm->atomic; if (ao->op_active && ao->op_notify && ao->op_notifier) { ao->op_notifier->n_status = status; list_add_tail(&ao->op_notifier->n_list, &rs->rs_notify_queue); ao->op_notifier = NULL; } |
5c1155904 RDS: send.c |
495 496 497 498 499 500 501 502 503 |
/* No need to wake the app - caller does this */ } /* * This is called from the IB send completion when we detect * a RDMA operation that failed with remote access error. * So speed is not an issue here. */ struct rds_message *rds_send_get_message(struct rds_connection *conn, |
f8b3aaf2b RDS: Remove struc... |
504 |
struct rm_rdma_op *op) |
5c1155904 RDS: send.c |
505 506 507 508 509 510 511 |
{ struct rds_message *rm, *tmp, *found = NULL; unsigned long flags; spin_lock_irqsave(&conn->c_lock, flags); list_for_each_entry_safe(rm, tmp, &conn->c_retrans, m_conn_item) { |
f8b3aaf2b RDS: Remove struc... |
512 |
if (&rm->rdma == op) { |
5c1155904 RDS: send.c |
513 514 515 516 517 518 519 |
atomic_inc(&rm->m_refcount); found = rm; goto out; } } list_for_each_entry_safe(rm, tmp, &conn->c_send_queue, m_conn_item) { |
f8b3aaf2b RDS: Remove struc... |
520 |
if (&rm->rdma == op) { |
5c1155904 RDS: send.c |
521 522 523 524 525 526 527 528 529 530 531 |
atomic_inc(&rm->m_refcount); found = rm; break; } } out: spin_unlock_irqrestore(&conn->c_lock, flags); return found; } |
616b757ae RDS: Export symbo... |
532 |
EXPORT_SYMBOL_GPL(rds_send_get_message); |
5c1155904 RDS: send.c |
533 534 535 536 537 538 539 540 541 |
/* * This removes messages from the socket's list if they're on it. The list * argument must be private to the caller, we must be able to modify it * without locks. The messages must have a reference held for their * position on the list. This function will drop that reference after * removing the messages from the 'messages' list regardless of if it found * the messages on the socket list or not. */ |
ff51bf841 rds: make local f... |
542 |
static void rds_send_remove_from_sock(struct list_head *messages, int status) |
5c1155904 RDS: send.c |
543 |
{ |
561c7df63 RDS: Do not call ... |
544 |
unsigned long flags; |
5c1155904 RDS: send.c |
545 546 |
struct rds_sock *rs = NULL; struct rds_message *rm; |
5c1155904 RDS: send.c |
547 |
while (!list_empty(messages)) { |
561c7df63 RDS: Do not call ... |
548 |
int was_on_sock = 0; |
5c1155904 RDS: send.c |
549 550 551 552 553 554 555 556 557 558 559 560 561 562 |
rm = list_entry(messages->next, struct rds_message, m_conn_item); list_del_init(&rm->m_conn_item); /* * If we see this flag cleared then we're *sure* that someone * else beat us to removing it from the sock. If we race * with their flag update we'll get the lock and then really * see that the flag has been cleared. * * The message spinlock makes sure nobody clears rm->m_rs * while we're messing with it. It does not prevent the * message from being removed from the socket, though. */ |
561c7df63 RDS: Do not call ... |
563 |
spin_lock_irqsave(&rm->m_rs_lock, flags); |
5c1155904 RDS: send.c |
564 565 566 567 568 |
if (!test_bit(RDS_MSG_ON_SOCK, &rm->m_flags)) goto unlock_and_drop; if (rs != rm->m_rs) { if (rs) { |
5c1155904 RDS: send.c |
569 570 571 572 |
rds_wake_sk_sleep(rs); sock_put(rds_rs_to_sk(rs)); } rs = rm->m_rs; |
5c1155904 RDS: send.c |
573 574 |
sock_hold(rds_rs_to_sk(rs)); } |
048c15e64 RDS: Fix send loc... |
575 |
spin_lock(&rs->rs_lock); |
5c1155904 RDS: send.c |
576 577 |
if (test_and_clear_bit(RDS_MSG_ON_SOCK, &rm->m_flags)) { |
f8b3aaf2b RDS: Remove struc... |
578 |
struct rm_rdma_op *ro = &rm->rdma; |
5c1155904 RDS: send.c |
579 580 581 582 |
struct rds_notifier *notifier; list_del_init(&rm->m_sock_item); rds_send_sndbuf_remove(rs, rm); |
f8b3aaf2b RDS: Remove struc... |
583 584 585 |
if (ro->op_active && ro->op_notifier && (ro->op_notify || (ro->op_recverr && status))) { notifier = ro->op_notifier; |
5c1155904 RDS: send.c |
586 587 588 589 |
list_add_tail(¬ifier->n_list, &rs->rs_notify_queue); if (!notifier->n_status) notifier->n_status = status; |
f8b3aaf2b RDS: Remove struc... |
590 |
rm->rdma.op_notifier = NULL; |
5c1155904 RDS: send.c |
591 |
} |
561c7df63 RDS: Do not call ... |
592 |
was_on_sock = 1; |
5c1155904 RDS: send.c |
593 594 |
rm->m_rs = NULL; } |
048c15e64 RDS: Fix send loc... |
595 |
spin_unlock(&rs->rs_lock); |
5c1155904 RDS: send.c |
596 597 |
unlock_and_drop: |
561c7df63 RDS: Do not call ... |
598 |
spin_unlock_irqrestore(&rm->m_rs_lock, flags); |
5c1155904 RDS: send.c |
599 |
rds_message_put(rm); |
561c7df63 RDS: Do not call ... |
600 601 |
if (was_on_sock) rds_message_put(rm); |
5c1155904 RDS: send.c |
602 603 604 |
} if (rs) { |
5c1155904 RDS: send.c |
605 606 607 |
rds_wake_sk_sleep(rs); sock_put(rds_rs_to_sk(rs)); } |
5c1155904 RDS: send.c |
608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 |
} /* * Transports call here when they've determined that the receiver queued * messages up to, and including, the given sequence number. Messages are * moved to the retrans queue when rds_send_xmit picks them off the send * queue. This means that in the TCP case, the message may not have been * assigned the m_ack_seq yet - but that's fine as long as tcp_is_acked * checks the RDS_MSG_HAS_ACK_SEQ bit. * * XXX It's not clear to me how this is safely serialized with socket * destruction. Maybe it should bail if it sees SOCK_DEAD. */ void rds_send_drop_acked(struct rds_connection *conn, u64 ack, is_acked_func is_acked) { struct rds_message *rm, *tmp; unsigned long flags; LIST_HEAD(list); spin_lock_irqsave(&conn->c_lock, flags); list_for_each_entry_safe(rm, tmp, &conn->c_retrans, m_conn_item) { if (!rds_send_is_acked(rm, ack, is_acked)) break; list_move(&rm->m_conn_item, &list); clear_bit(RDS_MSG_ON_CONN, &rm->m_flags); } /* order flag updates with spin locks */ if (!list_empty(&list)) smp_mb__after_clear_bit(); spin_unlock_irqrestore(&conn->c_lock, flags); /* now remove the messages from the sock list as needed */ rds_send_remove_from_sock(&list, RDS_RDMA_SUCCESS); } |
616b757ae RDS: Export symbo... |
647 |
EXPORT_SYMBOL_GPL(rds_send_drop_acked); |
5c1155904 RDS: send.c |
648 649 650 651 652 |
void rds_send_drop_to(struct rds_sock *rs, struct sockaddr_in *dest) { struct rds_message *rm, *tmp; struct rds_connection *conn; |
7c82eaf00 RDS: Rewrite rds_... |
653 |
unsigned long flags; |
5c1155904 RDS: send.c |
654 |
LIST_HEAD(list); |
5c1155904 RDS: send.c |
655 656 657 658 659 660 661 662 |
/* get all the messages we're dropping under the rs lock */ spin_lock_irqsave(&rs->rs_lock, flags); list_for_each_entry_safe(rm, tmp, &rs->rs_send_queue, m_sock_item) { if (dest && (dest->sin_addr.s_addr != rm->m_daddr || dest->sin_port != rm->m_inc.i_hdr.h_dport)) continue; |
5c1155904 RDS: send.c |
663 664 665 |
list_move(&rm->m_sock_item, &list); rds_send_sndbuf_remove(rs, rm); clear_bit(RDS_MSG_ON_SOCK, &rm->m_flags); |
5c1155904 RDS: send.c |
666 667 668 |
} /* order flag updates with the rs lock */ |
7c82eaf00 RDS: Rewrite rds_... |
669 |
smp_mb__after_clear_bit(); |
5c1155904 RDS: send.c |
670 671 |
spin_unlock_irqrestore(&rs->rs_lock, flags); |
7c82eaf00 RDS: Rewrite rds_... |
672 673 |
if (list_empty(&list)) return; |
5c1155904 RDS: send.c |
674 |
|
7c82eaf00 RDS: Rewrite rds_... |
675 |
/* Remove the messages from the conn */ |
5c1155904 RDS: send.c |
676 |
list_for_each_entry(rm, &list, m_sock_item) { |
7c82eaf00 RDS: Rewrite rds_... |
677 678 |
conn = rm->m_inc.i_conn; |
5c1155904 RDS: send.c |
679 |
|
9de0864cf RDS: Fix locking ... |
680 |
spin_lock_irqsave(&conn->c_lock, flags); |
5c1155904 RDS: send.c |
681 |
/* |
7c82eaf00 RDS: Rewrite rds_... |
682 683 684 |
* Maybe someone else beat us to removing rm from the conn. * If we race with their flag update we'll get the lock and * then really see that the flag has been cleared. |
5c1155904 RDS: send.c |
685 |
*/ |
7c82eaf00 RDS: Rewrite rds_... |
686 687 |
if (!test_and_clear_bit(RDS_MSG_ON_CONN, &rm->m_flags)) { spin_unlock_irqrestore(&conn->c_lock, flags); |
5c1155904 RDS: send.c |
688 |
continue; |
5c1155904 RDS: send.c |
689 |
} |
9de0864cf RDS: Fix locking ... |
690 691 |
list_del_init(&rm->m_conn_item); spin_unlock_irqrestore(&conn->c_lock, flags); |
5c1155904 RDS: send.c |
692 |
|
7c82eaf00 RDS: Rewrite rds_... |
693 694 695 696 |
/* * Couldn't grab m_rs_lock in top loop (lock ordering), * but we can now. */ |
9de0864cf RDS: Fix locking ... |
697 |
spin_lock_irqsave(&rm->m_rs_lock, flags); |
5c1155904 RDS: send.c |
698 |
|
7c82eaf00 RDS: Rewrite rds_... |
699 |
spin_lock(&rs->rs_lock); |
940786eb0 RDS: queue failur... |
700 |
__rds_send_complete(rs, rm, RDS_RDMA_CANCELED); |
7c82eaf00 RDS: Rewrite rds_... |
701 702 703 |
spin_unlock(&rs->rs_lock); rm->m_rs = NULL; |
9de0864cf RDS: Fix locking ... |
704 |
spin_unlock_irqrestore(&rm->m_rs_lock, flags); |
7c82eaf00 RDS: Rewrite rds_... |
705 |
|
7c82eaf00 RDS: Rewrite rds_... |
706 |
rds_message_put(rm); |
7c82eaf00 RDS: Rewrite rds_... |
707 |
} |
5c1155904 RDS: send.c |
708 |
|
7c82eaf00 RDS: Rewrite rds_... |
709 |
rds_wake_sk_sleep(rs); |
550a8002e RDS: Fix locking ... |
710 |
|
5c1155904 RDS: send.c |
711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 |
while (!list_empty(&list)) { rm = list_entry(list.next, struct rds_message, m_sock_item); list_del_init(&rm->m_sock_item); rds_message_wait(rm); rds_message_put(rm); } } /* * we only want this to fire once so we use the callers 'queued'. It's * possible that another thread can race with us and remove the * message from the flow with RDS_CANCEL_SENT_TO. */ static int rds_send_queue_rm(struct rds_sock *rs, struct rds_connection *conn, struct rds_message *rm, __be16 sport, __be16 dport, int *queued) { unsigned long flags; u32 len; if (*queued) goto out; len = be32_to_cpu(rm->m_inc.i_hdr.h_len); /* this is the only place which holds both the socket's rs_lock * and the connection's c_lock */ spin_lock_irqsave(&rs->rs_lock, flags); /* * If there is a little space in sndbuf, we don't queue anything, * and userspace gets -EAGAIN. But poll() indicates there's send * room. This can lead to bad behavior (spinning) if snd_bytes isn't * freed up by incoming acks. So we check the *old* value of * rs_snd_bytes here to allow the last msg to exceed the buffer, * and poll() now knows no more data can be sent. */ if (rs->rs_snd_bytes < rds_sk_sndbuf(rs)) { rs->rs_snd_bytes += len; /* let recv side know we are close to send space exhaustion. * This is probably not the optimal way to do it, as this * means we set the flag on *all* messages as soon as our * throughput hits a certain threshold. */ if (rs->rs_snd_bytes >= rds_sk_sndbuf(rs) / 2) __set_bit(RDS_MSG_ACK_REQUIRED, &rm->m_flags); list_add_tail(&rm->m_sock_item, &rs->rs_send_queue); set_bit(RDS_MSG_ON_SOCK, &rm->m_flags); rds_message_addref(rm); rm->m_rs = rs; /* The code ordering is a little weird, but we're trying to minimize the time we hold c_lock */ rds_message_populate_header(&rm->m_inc.i_hdr, sport, dport, 0); rm->m_inc.i_conn = conn; rds_message_addref(rm); spin_lock(&conn->c_lock); rm->m_inc.i_hdr.h_sequence = cpu_to_be64(conn->c_next_tx_seq++); list_add_tail(&rm->m_conn_item, &conn->c_send_queue); set_bit(RDS_MSG_ON_CONN, &rm->m_flags); spin_unlock(&conn->c_lock); rdsdebug("queued msg %p len %d, rs %p bytes %d seq %llu ", rm, len, rs, rs->rs_snd_bytes, (unsigned long long)be64_to_cpu(rm->m_inc.i_hdr.h_sequence)); *queued = 1; } spin_unlock_irqrestore(&rs->rs_lock, flags); out: return *queued; } |
fc445084f RDS: Explicitly a... |
789 790 791 792 793 794 |
/* * rds_message is getting to be quite complicated, and we'd like to allocate * it all in one go. This figures out how big it needs to be up front. */ static int rds_rm_size(struct msghdr *msg, int data_len) { |
ff87e97a9 RDS: make m_rdma_... |
795 |
struct cmsghdr *cmsg; |
fc445084f RDS: Explicitly a... |
796 |
int size = 0; |
aa0a4ef4a RDS: Make sure cm... |
797 |
int cmsg_groups = 0; |
ff87e97a9 RDS: make m_rdma_... |
798 799 800 801 802 803 804 805 806 807 808 |
int retval; for (cmsg = CMSG_FIRSTHDR(msg); cmsg; cmsg = CMSG_NXTHDR(msg, cmsg)) { if (!CMSG_OK(msg, cmsg)) return -EINVAL; if (cmsg->cmsg_level != SOL_RDS) continue; switch (cmsg->cmsg_type) { case RDS_CMSG_RDMA_ARGS: |
aa0a4ef4a RDS: Make sure cm... |
809 |
cmsg_groups |= 1; |
ff87e97a9 RDS: make m_rdma_... |
810 811 812 813 |
retval = rds_rdma_extra_size(CMSG_DATA(cmsg)); if (retval < 0) return retval; size += retval; |
aa0a4ef4a RDS: Make sure cm... |
814 |
|
ff87e97a9 RDS: make m_rdma_... |
815 816 817 818 |
break; case RDS_CMSG_RDMA_DEST: case RDS_CMSG_RDMA_MAP: |
aa0a4ef4a RDS: Make sure cm... |
819 |
cmsg_groups |= 2; |
ff87e97a9 RDS: make m_rdma_... |
820 821 |
/* these are valid but do no add any size */ break; |
15133f6e6 RDS: Implement at... |
822 823 |
case RDS_CMSG_ATOMIC_CSWP: case RDS_CMSG_ATOMIC_FADD: |
20c72bd5f RDS: Implement ma... |
824 825 |
case RDS_CMSG_MASKED_ATOMIC_CSWP: case RDS_CMSG_MASKED_ATOMIC_FADD: |
aa0a4ef4a RDS: Make sure cm... |
826 |
cmsg_groups |= 1; |
15133f6e6 RDS: Implement at... |
827 828 |
size += sizeof(struct scatterlist); break; |
ff87e97a9 RDS: make m_rdma_... |
829 830 831 832 833 |
default: return -EINVAL; } } |
fc445084f RDS: Explicitly a... |
834 |
|
ff87e97a9 RDS: make m_rdma_... |
835 |
size += ceil(data_len, PAGE_SIZE) * sizeof(struct scatterlist); |
fc445084f RDS: Explicitly a... |
836 |
|
aa0a4ef4a RDS: Make sure cm... |
837 838 839 |
/* Ensure (DEST, MAP) are never used with (ARGS, ATOMIC) */ if (cmsg_groups == 3) return -EINVAL; |
fc445084f RDS: Explicitly a... |
840 841 |
return size; } |
5c1155904 RDS: send.c |
842 843 844 845 846 847 848 849 850 851 852 853 854 855 |
static int rds_cmsg_send(struct rds_sock *rs, struct rds_message *rm, struct msghdr *msg, int *allocated_mr) { struct cmsghdr *cmsg; int ret = 0; for (cmsg = CMSG_FIRSTHDR(msg); cmsg; cmsg = CMSG_NXTHDR(msg, cmsg)) { if (!CMSG_OK(msg, cmsg)) return -EINVAL; if (cmsg->cmsg_level != SOL_RDS) continue; /* As a side effect, RDMA_DEST and RDMA_MAP will set |
15133f6e6 RDS: Implement at... |
856 |
* rm->rdma.m_rdma_cookie and rm->rdma.m_rdma_mr. |
5c1155904 RDS: send.c |
857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 |
*/ switch (cmsg->cmsg_type) { case RDS_CMSG_RDMA_ARGS: ret = rds_cmsg_rdma_args(rs, rm, cmsg); break; case RDS_CMSG_RDMA_DEST: ret = rds_cmsg_rdma_dest(rs, rm, cmsg); break; case RDS_CMSG_RDMA_MAP: ret = rds_cmsg_rdma_map(rs, rm, cmsg); if (!ret) *allocated_mr = 1; break; |
15133f6e6 RDS: Implement at... |
872 873 |
case RDS_CMSG_ATOMIC_CSWP: case RDS_CMSG_ATOMIC_FADD: |
20c72bd5f RDS: Implement ma... |
874 875 |
case RDS_CMSG_MASKED_ATOMIC_CSWP: case RDS_CMSG_MASKED_ATOMIC_FADD: |
15133f6e6 RDS: Implement at... |
876 877 |
ret = rds_cmsg_atomic(rs, rm, cmsg); break; |
5c1155904 RDS: send.c |
878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 |
default: return -EINVAL; } if (ret) break; } return ret; } int rds_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg, size_t payload_len) { struct sock *sk = sock->sk; struct rds_sock *rs = rds_sk_to_rs(sk); struct sockaddr_in *usin = (struct sockaddr_in *)msg->msg_name; __be32 daddr; __be16 dport; struct rds_message *rm = NULL; struct rds_connection *conn; int ret = 0; int queued = 0, allocated_mr = 0; int nonblock = msg->msg_flags & MSG_DONTWAIT; |
1123fd734 RDS: sendmsg() sh... |
903 |
long timeo = sock_sndtimeo(sk, nonblock); |
5c1155904 RDS: send.c |
904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 |
/* Mirror Linux UDP mirror of BSD error message compatibility */ /* XXX: Perhaps MSG_MORE someday */ if (msg->msg_flags & ~(MSG_DONTWAIT | MSG_CMSG_COMPAT)) { printk(KERN_INFO "msg_flags 0x%08X ", msg->msg_flags); ret = -EOPNOTSUPP; goto out; } if (msg->msg_namelen) { /* XXX fail non-unicast destination IPs? */ if (msg->msg_namelen < sizeof(*usin) || usin->sin_family != AF_INET) { ret = -EINVAL; goto out; } daddr = usin->sin_addr.s_addr; dport = usin->sin_port; } else { /* We only care about consistency with ->connect() */ lock_sock(sk); daddr = rs->rs_conn_addr; dport = rs->rs_conn_port; release_sock(sk); } /* racing with another thread binding seems ok here */ if (daddr == 0 || rs->rs_bound_addr == 0) { ret = -ENOTCONN; /* XXX not a great errno */ goto out; } |
fc445084f RDS: Explicitly a... |
935 936 937 938 939 940 941 942 |
/* size of rm including all sgs */ ret = rds_rm_size(msg, payload_len); if (ret < 0) goto out; rm = rds_message_alloc(ret, GFP_KERNEL); if (!rm) { ret = -ENOMEM; |
5c1155904 RDS: send.c |
943 944 |
goto out; } |
372cd7ded RDS: Do not set o... |
945 946 947 |
/* Attach data to the rm */ if (payload_len) { rm->data.op_sg = rds_message_alloc_sgs(rm, ceil(payload_len, PAGE_SIZE)); |
d139ff090 RDS: Let rds_mess... |
948 949 950 951 |
if (!rm->data.op_sg) { ret = -ENOMEM; goto out; } |
372cd7ded RDS: Do not set o... |
952 953 954 955 956 |
ret = rds_message_copy_from_user(rm, msg->msg_iov, payload_len); if (ret) goto out; } rm->data.op_active = 1; |
fc445084f RDS: Explicitly a... |
957 |
|
5c1155904 RDS: send.c |
958 |
rm->m_daddr = daddr; |
5c1155904 RDS: send.c |
959 960 961 962 963 964 965 966 967 968 969 970 971 972 |
/* rds_conn_create has a spinlock that runs with IRQ off. * Caching the conn in the socket helps a lot. */ if (rs->rs_conn && rs->rs_conn->c_faddr == daddr) conn = rs->rs_conn; else { conn = rds_conn_create_outgoing(rs->rs_bound_addr, daddr, rs->rs_transport, sock->sk->sk_allocation); if (IS_ERR(conn)) { ret = PTR_ERR(conn); goto out; } rs->rs_conn = conn; } |
49f696914 RDS: Establish co... |
973 974 975 976 |
/* Parse any control messages the user may have included. */ ret = rds_cmsg_send(rs, rm, msg, &allocated_mr); if (ret) goto out; |
2c3a5f9ab RDS: Add flag for... |
977 |
if (rm->rdma.op_active && !conn->c_trans->xmit_rdma) { |
cb0a60564 net/rds: use prin... |
978 979 |
printk_ratelimited(KERN_NOTICE "rdma_op %p conn xmit_rdma %p ", |
f8b3aaf2b RDS: Remove struc... |
980 |
&rm->rdma, conn->c_trans->xmit_rdma); |
15133f6e6 RDS: Implement at... |
981 982 983 984 985 |
ret = -EOPNOTSUPP; goto out; } if (rm->atomic.op_active && !conn->c_trans->xmit_atomic) { |
cb0a60564 net/rds: use prin... |
986 987 |
printk_ratelimited(KERN_NOTICE "atomic_op %p conn xmit_atomic %p ", |
15133f6e6 RDS: Implement at... |
988 |
&rm->atomic, conn->c_trans->xmit_atomic); |
5c1155904 RDS: send.c |
989 990 991 |
ret = -EOPNOTSUPP; goto out; } |
f3c6808d3 RDS: introduce rd... |
992 |
rds_conn_connect_if_down(conn); |
5c1155904 RDS: send.c |
993 994 |
ret = rds_cong_wait(conn->c_fcong, dport, nonblock, rs); |
b98ba52f9 RDS: only put soc... |
995 996 |
if (ret) { rs->rs_seen_congestion = 1; |
5c1155904 RDS: send.c |
997 |
goto out; |
b98ba52f9 RDS: only put soc... |
998 |
} |
5c1155904 RDS: send.c |
999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 |
while (!rds_send_queue_rm(rs, conn, rm, rs->rs_bound_port, dport, &queued)) { rds_stats_inc(s_send_queue_full); /* XXX make sure this is reasonable */ if (payload_len > rds_sk_sndbuf(rs)) { ret = -EMSGSIZE; goto out; } if (nonblock) { ret = -EAGAIN; goto out; } |
aa3951451 net: sk_sleep() h... |
1012 |
timeo = wait_event_interruptible_timeout(*sk_sleep(sk), |
5c1155904 RDS: send.c |
1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 |
rds_send_queue_rm(rs, conn, rm, rs->rs_bound_port, dport, &queued), timeo); rdsdebug("sendmsg woke queued %d timeo %ld ", queued, timeo); if (timeo > 0 || timeo == MAX_SCHEDULE_TIMEOUT) continue; ret = timeo; if (ret == 0) ret = -ETIMEDOUT; goto out; } /* * By now we've committed to the send. We reuse rds_send_worker() * to retry sends in the rds thread if the transport asks us to. */ rds_stats_inc(s_send_queued); if (!test_bit(RDS_LL_SEND_FULL, &conn->c_flags)) |
a7d3a2814 RDS: Call rds_sen... |
1036 |
rds_send_xmit(conn); |
5c1155904 RDS: send.c |
1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 |
rds_message_put(rm); return payload_len; out: /* If the user included a RDMA_MAP cmsg, we allocated a MR on the fly. * If the sendmsg goes through, we keep the MR. If it fails with EAGAIN * or in any other way, we need to destroy the MR again */ if (allocated_mr) rds_rdma_unuse(rs, rds_rdma_cookie_key(rm->m_rdma_cookie), 1); if (rm) rds_message_put(rm); return ret; } /* * Reply to a ping packet. */ int rds_send_pong(struct rds_connection *conn, __be16 dport) { struct rds_message *rm; unsigned long flags; int ret = 0; rm = rds_message_alloc(0, GFP_ATOMIC); |
8690bfa17 RDS: cleanup: rem... |
1064 |
if (!rm) { |
5c1155904 RDS: send.c |
1065 1066 1067 1068 1069 |
ret = -ENOMEM; goto out; } rm->m_daddr = conn->c_faddr; |
acfcd4d4e RDS: Get pong wor... |
1070 |
rm->data.op_active = 1; |
5c1155904 RDS: send.c |
1071 |
|
f3c6808d3 RDS: introduce rd... |
1072 |
rds_conn_connect_if_down(conn); |
5c1155904 RDS: send.c |
1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 |
ret = rds_cong_wait(conn->c_fcong, dport, 1, NULL); if (ret) goto out; spin_lock_irqsave(&conn->c_lock, flags); list_add_tail(&rm->m_conn_item, &conn->c_send_queue); set_bit(RDS_MSG_ON_CONN, &rm->m_flags); rds_message_addref(rm); rm->m_inc.i_conn = conn; rds_message_populate_header(&rm->m_inc.i_hdr, 0, dport, conn->c_next_tx_seq); conn->c_next_tx_seq++; spin_unlock_irqrestore(&conn->c_lock, flags); rds_stats_inc(s_send_queued); rds_stats_inc(s_send_pong); |
acfcd4d4e RDS: Get pong wor... |
1091 1092 |
if (!test_bit(RDS_LL_SEND_FULL, &conn->c_flags)) rds_send_xmit(conn); |
5c1155904 RDS: send.c |
1093 1094 1095 1096 1097 1098 1099 1100 |
rds_message_put(rm); return 0; out: if (rm) rds_message_put(rm); return ret; } |