Commit | Line | Data |
---|---|---|
bdbe6fbc AG |
1 | /* |
2 | * Copyright (c) 2006 Oracle. All rights reserved. | |
3 | * | |
4 | * This software is available to you under a choice of one of two | |
5 | * licenses. You may choose to be licensed under the terms of the GNU | |
6 | * General Public License (GPL) Version 2, available from the file | |
7 | * COPYING in the main directory of this source tree, or the | |
8 | * OpenIB.org BSD license below: | |
9 | * | |
10 | * Redistribution and use in source and binary forms, with or | |
11 | * without modification, are permitted provided that the following | |
12 | * conditions are met: | |
13 | * | |
14 | * - Redistributions of source code must retain the above | |
15 | * copyright notice, this list of conditions and the following | |
16 | * disclaimer. | |
17 | * | |
18 | * - Redistributions in binary form must reproduce the above | |
19 | * copyright notice, this list of conditions and the following | |
20 | * disclaimer in the documentation and/or other materials | |
21 | * provided with the distribution. | |
22 | * | |
23 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, | |
24 | * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF | |
25 | * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND | |
26 | * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS | |
27 | * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN | |
28 | * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN | |
29 | * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE | |
30 | * SOFTWARE. | |
31 | * | |
32 | */ | |
33 | #include <linux/kernel.h> | |
5a0e3ad6 | 34 | #include <linux/slab.h> |
bdbe6fbc AG |
35 | #include <net/sock.h> |
36 | #include <linux/in.h> | |
bc3b2d7f | 37 | #include <linux/export.h> |
5711f8b3 | 38 | #include <linux/time.h> |
39 | #include <linux/rds.h> | |
bdbe6fbc AG |
40 | |
41 | #include "rds.h" | |
bdbe6fbc AG |
42 | |
43 | void rds_inc_init(struct rds_incoming *inc, struct rds_connection *conn, | |
44 | __be32 saddr) | |
45 | { | |
3289025a SS |
46 | int i; |
47 | ||
bdbe6fbc AG |
48 | atomic_set(&inc->i_refcount, 1); |
49 | INIT_LIST_HEAD(&inc->i_item); | |
50 | inc->i_conn = conn; | |
51 | inc->i_saddr = saddr; | |
52 | inc->i_rdma_cookie = 0; | |
5711f8b3 | 53 | inc->i_rx_tstamp.tv_sec = 0; |
54 | inc->i_rx_tstamp.tv_usec = 0; | |
3289025a SS |
55 | |
56 | for (i = 0; i < RDS_RX_MAX_TRACES; i++) | |
57 | inc->i_rx_lat_trace[i] = 0; | |
bdbe6fbc | 58 | } |
616b757a | 59 | EXPORT_SYMBOL_GPL(rds_inc_init); |
bdbe6fbc | 60 | |
5e833e02 SV |
61 | void rds_inc_path_init(struct rds_incoming *inc, struct rds_conn_path *cp, |
62 | __be32 saddr) | |
63 | { | |
64 | atomic_set(&inc->i_refcount, 1); | |
65 | INIT_LIST_HEAD(&inc->i_item); | |
66 | inc->i_conn = cp->cp_conn; | |
67 | inc->i_conn_path = cp; | |
68 | inc->i_saddr = saddr; | |
69 | inc->i_rdma_cookie = 0; | |
70 | inc->i_rx_tstamp.tv_sec = 0; | |
71 | inc->i_rx_tstamp.tv_usec = 0; | |
72 | } | |
73 | EXPORT_SYMBOL_GPL(rds_inc_path_init); | |
74 | ||
ff51bf84 | 75 | static void rds_inc_addref(struct rds_incoming *inc) |
bdbe6fbc AG |
76 | { |
77 | rdsdebug("addref inc %p ref %d\n", inc, atomic_read(&inc->i_refcount)); | |
78 | atomic_inc(&inc->i_refcount); | |
79 | } | |
80 | ||
81 | void rds_inc_put(struct rds_incoming *inc) | |
82 | { | |
83 | rdsdebug("put inc %p ref %d\n", inc, atomic_read(&inc->i_refcount)); | |
84 | if (atomic_dec_and_test(&inc->i_refcount)) { | |
85 | BUG_ON(!list_empty(&inc->i_item)); | |
86 | ||
87 | inc->i_conn->c_trans->inc_free(inc); | |
88 | } | |
89 | } | |
616b757a | 90 | EXPORT_SYMBOL_GPL(rds_inc_put); |
bdbe6fbc AG |
91 | |
92 | static void rds_recv_rcvbuf_delta(struct rds_sock *rs, struct sock *sk, | |
93 | struct rds_cong_map *map, | |
94 | int delta, __be16 port) | |
95 | { | |
96 | int now_congested; | |
97 | ||
98 | if (delta == 0) | |
99 | return; | |
100 | ||
101 | rs->rs_rcv_bytes += delta; | |
192a798f VV |
102 | if (delta > 0) |
103 | rds_stats_add(s_recv_bytes_added_to_socket, delta); | |
104 | else | |
105 | rds_stats_add(s_recv_bytes_removed_from_socket, -delta); | |
bdbe6fbc AG |
106 | now_congested = rs->rs_rcv_bytes > rds_sk_rcvbuf(rs); |
107 | ||
108 | rdsdebug("rs %p (%pI4:%u) recv bytes %d buf %d " | |
109 | "now_cong %d delta %d\n", | |
110 | rs, &rs->rs_bound_addr, | |
111 | ntohs(rs->rs_bound_port), rs->rs_rcv_bytes, | |
112 | rds_sk_rcvbuf(rs), now_congested, delta); | |
113 | ||
114 | /* wasn't -> am congested */ | |
115 | if (!rs->rs_congested && now_congested) { | |
116 | rs->rs_congested = 1; | |
117 | rds_cong_set_bit(map, port); | |
118 | rds_cong_queue_updates(map); | |
119 | } | |
120 | /* was -> aren't congested */ | |
121 | /* Require more free space before reporting uncongested to prevent | |
122 | bouncing cong/uncong state too often */ | |
123 | else if (rs->rs_congested && (rs->rs_rcv_bytes < (rds_sk_rcvbuf(rs)/2))) { | |
124 | rs->rs_congested = 0; | |
125 | rds_cong_clear_bit(map, port); | |
126 | rds_cong_queue_updates(map); | |
127 | } | |
128 | ||
129 | /* do nothing if no change in cong state */ | |
130 | } | |
131 | ||
905dd418 SV |
132 | static void rds_conn_peer_gen_update(struct rds_connection *conn, |
133 | u32 peer_gen_num) | |
134 | { | |
135 | int i; | |
136 | struct rds_message *rm, *tmp; | |
137 | unsigned long flags; | |
138 | ||
139 | WARN_ON(conn->c_trans->t_type != RDS_TRANS_TCP); | |
140 | if (peer_gen_num != 0) { | |
141 | if (conn->c_peer_gen_num != 0 && | |
142 | peer_gen_num != conn->c_peer_gen_num) { | |
143 | for (i = 0; i < RDS_MPATH_WORKERS; i++) { | |
144 | struct rds_conn_path *cp; | |
145 | ||
146 | cp = &conn->c_path[i]; | |
147 | spin_lock_irqsave(&cp->cp_lock, flags); | |
148 | cp->cp_next_tx_seq = 1; | |
149 | cp->cp_next_rx_seq = 0; | |
150 | list_for_each_entry_safe(rm, tmp, | |
151 | &cp->cp_retrans, | |
152 | m_conn_item) { | |
153 | set_bit(RDS_MSG_FLUSH, &rm->m_flags); | |
154 | } | |
155 | spin_unlock_irqrestore(&cp->cp_lock, flags); | |
156 | } | |
157 | } | |
158 | conn->c_peer_gen_num = peer_gen_num; | |
159 | } | |
160 | } | |
161 | ||
bdbe6fbc AG |
162 | /* |
163 | * Process all extension headers that come with this message. | |
164 | */ | |
165 | static void rds_recv_incoming_exthdrs(struct rds_incoming *inc, struct rds_sock *rs) | |
166 | { | |
167 | struct rds_header *hdr = &inc->i_hdr; | |
168 | unsigned int pos = 0, type, len; | |
169 | union { | |
170 | struct rds_ext_header_version version; | |
171 | struct rds_ext_header_rdma rdma; | |
172 | struct rds_ext_header_rdma_dest rdma_dest; | |
173 | } buffer; | |
174 | ||
175 | while (1) { | |
176 | len = sizeof(buffer); | |
177 | type = rds_message_next_extension(hdr, &pos, &buffer, &len); | |
178 | if (type == RDS_EXTHDR_NONE) | |
179 | break; | |
180 | /* Process extension header here */ | |
181 | switch (type) { | |
182 | case RDS_EXTHDR_RDMA: | |
183 | rds_rdma_unuse(rs, be32_to_cpu(buffer.rdma.h_rdma_rkey), 0); | |
184 | break; | |
185 | ||
186 | case RDS_EXTHDR_RDMA_DEST: | |
187 | /* We ignore the size for now. We could stash it | |
188 | * somewhere and use it for error checking. */ | |
189 | inc->i_rdma_cookie = rds_rdma_make_cookie( | |
190 | be32_to_cpu(buffer.rdma_dest.h_rdma_rkey), | |
191 | be32_to_cpu(buffer.rdma_dest.h_rdma_offset)); | |
192 | ||
193 | break; | |
194 | } | |
195 | } | |
196 | } | |
197 | ||
5916e2c1 SV |
198 | static void rds_recv_hs_exthdrs(struct rds_header *hdr, |
199 | struct rds_connection *conn) | |
200 | { | |
201 | unsigned int pos = 0, type, len; | |
202 | union { | |
203 | struct rds_ext_header_version version; | |
204 | u16 rds_npaths; | |
905dd418 | 205 | u32 rds_gen_num; |
5916e2c1 | 206 | } buffer; |
905dd418 | 207 | u32 new_peer_gen_num = 0; |
5916e2c1 SV |
208 | |
209 | while (1) { | |
210 | len = sizeof(buffer); | |
211 | type = rds_message_next_extension(hdr, &pos, &buffer, &len); | |
212 | if (type == RDS_EXTHDR_NONE) | |
213 | break; | |
214 | /* Process extension header here */ | |
215 | switch (type) { | |
216 | case RDS_EXTHDR_NPATHS: | |
217 | conn->c_npaths = min_t(int, RDS_MPATH_WORKERS, | |
218 | buffer.rds_npaths); | |
219 | break; | |
905dd418 SV |
220 | case RDS_EXTHDR_GEN_NUM: |
221 | new_peer_gen_num = buffer.rds_gen_num; | |
222 | break; | |
5916e2c1 SV |
223 | default: |
224 | pr_warn_ratelimited("ignoring unknown exthdr type " | |
225 | "0x%x\n", type); | |
226 | } | |
227 | } | |
228 | /* if RDS_EXTHDR_NPATHS was not found, default to a single-path */ | |
229 | conn->c_npaths = max_t(int, conn->c_npaths, 1); | |
905dd418 | 230 | rds_conn_peer_gen_update(conn, new_peer_gen_num); |
5916e2c1 SV |
231 | } |
232 | ||
233 | /* rds_start_mprds() will synchronously start multiple paths when appropriate. | |
234 | * The scheme is based on the following rules: | |
235 | * | |
236 | * 1. rds_sendmsg on first connect attempt sends the probe ping, with the | |
237 | * sender's npaths (s_npaths) | |
238 | * 2. rcvr of probe-ping knows the mprds_paths = min(s_npaths, r_npaths). It | |
239 | * sends back a probe-pong with r_npaths. After that, if rcvr is the | |
240 | * smaller ip addr, it starts rds_conn_path_connect_if_down on all | |
241 | * mprds_paths. | |
242 | * 3. sender gets woken up, and can move to rds_conn_path_connect_if_down. | |
243 | * If it is the smaller ipaddr, rds_conn_path_connect_if_down can be | |
244 | * called after reception of the probe-pong on all mprds_paths. | |
245 | * Otherwise (sender of probe-ping is not the smaller ip addr): just call | |
246 | * rds_conn_path_connect_if_down on the hashed path. (see rule 4) | |
247 | * 4. when cp_index > 0, rds_connect_worker must only trigger | |
248 | * a connection if laddr < faddr. | |
249 | * 5. sender may end up queuing the packet on the cp. will get sent out later. | |
250 | * when connection is completed. | |
251 | */ | |
252 | static void rds_start_mprds(struct rds_connection *conn) | |
253 | { | |
254 | int i; | |
255 | struct rds_conn_path *cp; | |
256 | ||
257 | if (conn->c_npaths > 1 && conn->c_laddr < conn->c_faddr) { | |
258 | for (i = 1; i < conn->c_npaths; i++) { | |
259 | cp = &conn->c_path[i]; | |
260 | rds_conn_path_connect_if_down(cp); | |
261 | } | |
262 | } | |
263 | } | |
264 | ||
bdbe6fbc AG |
265 | /* |
266 | * The transport must make sure that this is serialized against other | |
267 | * rx and conn reset on this specific conn. | |
268 | * | |
269 | * We currently assert that only one fragmented message will be sent | |
270 | * down a connection at a time. This lets us reassemble in the conn | |
271 | * instead of per-flow which means that we don't have to go digging through | |
272 | * flows to tear down partial reassembly progress on conn failure and | |
273 | * we save flow lookup and locking for each frag arrival. It does mean | |
274 | * that small messages will wait behind large ones. Fragmenting at all | |
275 | * is only to reduce the memory consumption of pre-posted buffers. | |
276 | * | |
277 | * The caller passes in saddr and daddr instead of us getting it from the | |
278 | * conn. This lets loopback, who only has one conn for both directions, | |
279 | * tell us which roles the addrs in the conn are playing for this message. | |
280 | */ | |
281 | void rds_recv_incoming(struct rds_connection *conn, __be32 saddr, __be32 daddr, | |
6114eab5 | 282 | struct rds_incoming *inc, gfp_t gfp) |
bdbe6fbc AG |
283 | { |
284 | struct rds_sock *rs = NULL; | |
285 | struct sock *sk; | |
286 | unsigned long flags; | |
ef9e62c2 | 287 | struct rds_conn_path *cp; |
bdbe6fbc AG |
288 | |
289 | inc->i_conn = conn; | |
290 | inc->i_rx_jiffies = jiffies; | |
ef9e62c2 SV |
291 | if (conn->c_trans->t_mp_capable) |
292 | cp = inc->i_conn_path; | |
293 | else | |
294 | cp = &conn->c_path[0]; | |
bdbe6fbc AG |
295 | |
296 | rdsdebug("conn %p next %llu inc %p seq %llu len %u sport %u dport %u " | |
297 | "flags 0x%x rx_jiffies %lu\n", conn, | |
ef9e62c2 | 298 | (unsigned long long)cp->cp_next_rx_seq, |
bdbe6fbc AG |
299 | inc, |
300 | (unsigned long long)be64_to_cpu(inc->i_hdr.h_sequence), | |
301 | be32_to_cpu(inc->i_hdr.h_len), | |
302 | be16_to_cpu(inc->i_hdr.h_sport), | |
303 | be16_to_cpu(inc->i_hdr.h_dport), | |
304 | inc->i_hdr.h_flags, | |
305 | inc->i_rx_jiffies); | |
306 | ||
307 | /* | |
308 | * Sequence numbers should only increase. Messages get their | |
309 | * sequence number as they're queued in a sending conn. They | |
310 | * can be dropped, though, if the sending socket is closed before | |
311 | * they hit the wire. So sequence numbers can skip forward | |
312 | * under normal operation. They can also drop back in the conn | |
313 | * failover case as previously sent messages are resent down the | |
314 | * new instance of a conn. We drop those, otherwise we have | |
315 | * to assume that the next valid seq does not come after a | |
316 | * hole in the fragment stream. | |
317 | * | |
318 | * The headers don't give us a way to realize if fragments of | |
319 | * a message have been dropped. We assume that frags that arrive | |
320 | * to a flow are part of the current message on the flow that is | |
321 | * being reassembled. This means that senders can't drop messages | |
322 | * from the sending conn until all their frags are sent. | |
323 | * | |
324 | * XXX we could spend more on the wire to get more robust failure | |
325 | * detection, arguably worth it to avoid data corruption. | |
326 | */ | |
ef9e62c2 | 327 | if (be64_to_cpu(inc->i_hdr.h_sequence) < cp->cp_next_rx_seq && |
f64f9e71 | 328 | (inc->i_hdr.h_flags & RDS_FLAG_RETRANSMITTED)) { |
bdbe6fbc AG |
329 | rds_stats_inc(s_recv_drop_old_seq); |
330 | goto out; | |
331 | } | |
ef9e62c2 | 332 | cp->cp_next_rx_seq = be64_to_cpu(inc->i_hdr.h_sequence) + 1; |
bdbe6fbc AG |
333 | |
334 | if (rds_sysctl_ping_enable && inc->i_hdr.h_dport == 0) { | |
11bb62f7 SV |
335 | if (inc->i_hdr.h_sport == 0) { |
336 | rdsdebug("ignore ping with 0 sport from 0x%x\n", saddr); | |
337 | goto out; | |
338 | } | |
bdbe6fbc | 339 | rds_stats_inc(s_recv_ping); |
45997e9e | 340 | rds_send_pong(cp, inc->i_hdr.h_sport); |
5916e2c1 SV |
341 | /* if this is a handshake ping, start multipath if necessary */ |
342 | if (RDS_HS_PROBE(inc->i_hdr.h_sport, inc->i_hdr.h_dport)) { | |
343 | rds_recv_hs_exthdrs(&inc->i_hdr, cp->cp_conn); | |
344 | rds_start_mprds(cp->cp_conn); | |
345 | } | |
346 | goto out; | |
347 | } | |
348 | ||
349 | if (inc->i_hdr.h_dport == RDS_FLAG_PROBE_PORT && | |
350 | inc->i_hdr.h_sport == 0) { | |
351 | rds_recv_hs_exthdrs(&inc->i_hdr, cp->cp_conn); | |
352 | /* if this is a handshake pong, start multipath if necessary */ | |
353 | rds_start_mprds(cp->cp_conn); | |
354 | wake_up(&cp->cp_conn->c_hs_waitq); | |
bdbe6fbc AG |
355 | goto out; |
356 | } | |
357 | ||
358 | rs = rds_find_bound(daddr, inc->i_hdr.h_dport); | |
8690bfa1 | 359 | if (!rs) { |
bdbe6fbc AG |
360 | rds_stats_inc(s_recv_drop_no_sock); |
361 | goto out; | |
362 | } | |
363 | ||
364 | /* Process extension headers */ | |
365 | rds_recv_incoming_exthdrs(inc, rs); | |
366 | ||
367 | /* We can be racing with rds_release() which marks the socket dead. */ | |
368 | sk = rds_rs_to_sk(rs); | |
369 | ||
370 | /* serialize with rds_release -> sock_orphan */ | |
371 | write_lock_irqsave(&rs->rs_recv_lock, flags); | |
372 | if (!sock_flag(sk, SOCK_DEAD)) { | |
373 | rdsdebug("adding inc %p to rs %p's recv queue\n", inc, rs); | |
374 | rds_stats_inc(s_recv_queued); | |
375 | rds_recv_rcvbuf_delta(rs, sk, inc->i_conn->c_lcong, | |
376 | be32_to_cpu(inc->i_hdr.h_len), | |
377 | inc->i_hdr.h_dport); | |
5711f8b3 | 378 | if (sock_flag(sk, SOCK_RCVTSTAMP)) |
379 | do_gettimeofday(&inc->i_rx_tstamp); | |
bdbe6fbc | 380 | rds_inc_addref(inc); |
3289025a | 381 | inc->i_rx_lat_trace[RDS_MSG_RX_END] = local_clock(); |
bdbe6fbc AG |
382 | list_add_tail(&inc->i_item, &rs->rs_recv_queue); |
383 | __rds_wake_sk_sleep(sk); | |
384 | } else { | |
385 | rds_stats_inc(s_recv_drop_dead_sock); | |
386 | } | |
387 | write_unlock_irqrestore(&rs->rs_recv_lock, flags); | |
388 | ||
389 | out: | |
390 | if (rs) | |
391 | rds_sock_put(rs); | |
392 | } | |
616b757a | 393 | EXPORT_SYMBOL_GPL(rds_recv_incoming); |
bdbe6fbc AG |
394 | |
395 | /* | |
396 | * be very careful here. This is being called as the condition in | |
397 | * wait_event_*() needs to cope with being called many times. | |
398 | */ | |
399 | static int rds_next_incoming(struct rds_sock *rs, struct rds_incoming **inc) | |
400 | { | |
401 | unsigned long flags; | |
402 | ||
8690bfa1 | 403 | if (!*inc) { |
bdbe6fbc AG |
404 | read_lock_irqsave(&rs->rs_recv_lock, flags); |
405 | if (!list_empty(&rs->rs_recv_queue)) { | |
406 | *inc = list_entry(rs->rs_recv_queue.next, | |
407 | struct rds_incoming, | |
408 | i_item); | |
409 | rds_inc_addref(*inc); | |
410 | } | |
411 | read_unlock_irqrestore(&rs->rs_recv_lock, flags); | |
412 | } | |
413 | ||
414 | return *inc != NULL; | |
415 | } | |
416 | ||
417 | static int rds_still_queued(struct rds_sock *rs, struct rds_incoming *inc, | |
418 | int drop) | |
419 | { | |
420 | struct sock *sk = rds_rs_to_sk(rs); | |
421 | int ret = 0; | |
422 | unsigned long flags; | |
423 | ||
424 | write_lock_irqsave(&rs->rs_recv_lock, flags); | |
425 | if (!list_empty(&inc->i_item)) { | |
426 | ret = 1; | |
427 | if (drop) { | |
428 | /* XXX make sure this i_conn is reliable */ | |
429 | rds_recv_rcvbuf_delta(rs, sk, inc->i_conn->c_lcong, | |
430 | -be32_to_cpu(inc->i_hdr.h_len), | |
431 | inc->i_hdr.h_dport); | |
432 | list_del_init(&inc->i_item); | |
433 | rds_inc_put(inc); | |
434 | } | |
435 | } | |
436 | write_unlock_irqrestore(&rs->rs_recv_lock, flags); | |
437 | ||
438 | rdsdebug("inc %p rs %p still %d dropped %d\n", inc, rs, ret, drop); | |
439 | return ret; | |
440 | } | |
441 | ||
442 | /* | |
443 | * Pull errors off the error queue. | |
444 | * If msghdr is NULL, we will just purge the error queue. | |
445 | */ | |
446 | int rds_notify_queue_get(struct rds_sock *rs, struct msghdr *msghdr) | |
447 | { | |
448 | struct rds_notifier *notifier; | |
f037590f | 449 | struct rds_rdma_notify cmsg = { 0 }; /* fill holes with zero */ |
bdbe6fbc AG |
450 | unsigned int count = 0, max_messages = ~0U; |
451 | unsigned long flags; | |
452 | LIST_HEAD(copy); | |
453 | int err = 0; | |
454 | ||
455 | ||
456 | /* put_cmsg copies to user space and thus may sleep. We can't do this | |
457 | * with rs_lock held, so first grab as many notifications as we can stuff | |
458 | * in the user provided cmsg buffer. We don't try to copy more, to avoid | |
459 | * losing notifications - except when the buffer is so small that it wouldn't | |
460 | * even hold a single notification. Then we give him as much of this single | |
461 | * msg as we can squeeze in, and set MSG_CTRUNC. | |
462 | */ | |
463 | if (msghdr) { | |
464 | max_messages = msghdr->msg_controllen / CMSG_SPACE(sizeof(cmsg)); | |
465 | if (!max_messages) | |
466 | max_messages = 1; | |
467 | } | |
468 | ||
469 | spin_lock_irqsave(&rs->rs_lock, flags); | |
470 | while (!list_empty(&rs->rs_notify_queue) && count < max_messages) { | |
471 | notifier = list_entry(rs->rs_notify_queue.next, | |
472 | struct rds_notifier, n_list); | |
473 | list_move(¬ifier->n_list, ©); | |
474 | count++; | |
475 | } | |
476 | spin_unlock_irqrestore(&rs->rs_lock, flags); | |
477 | ||
478 | if (!count) | |
479 | return 0; | |
480 | ||
481 | while (!list_empty(©)) { | |
482 | notifier = list_entry(copy.next, struct rds_notifier, n_list); | |
483 | ||
484 | if (msghdr) { | |
485 | cmsg.user_token = notifier->n_user_token; | |
6200ed77 | 486 | cmsg.status = notifier->n_status; |
bdbe6fbc AG |
487 | |
488 | err = put_cmsg(msghdr, SOL_RDS, RDS_CMSG_RDMA_STATUS, | |
6200ed77 | 489 | sizeof(cmsg), &cmsg); |
bdbe6fbc AG |
490 | if (err) |
491 | break; | |
492 | } | |
493 | ||
494 | list_del_init(¬ifier->n_list); | |
495 | kfree(notifier); | |
496 | } | |
497 | ||
498 | /* If we bailed out because of an error in put_cmsg, | |
499 | * we may be left with one or more notifications that we | |
500 | * didn't process. Return them to the head of the list. */ | |
501 | if (!list_empty(©)) { | |
502 | spin_lock_irqsave(&rs->rs_lock, flags); | |
503 | list_splice(©, &rs->rs_notify_queue); | |
504 | spin_unlock_irqrestore(&rs->rs_lock, flags); | |
505 | } | |
506 | ||
507 | return err; | |
508 | } | |
509 | ||
510 | /* | |
511 | * Queue a congestion notification | |
512 | */ | |
513 | static int rds_notify_cong(struct rds_sock *rs, struct msghdr *msghdr) | |
514 | { | |
515 | uint64_t notify = rs->rs_cong_notify; | |
516 | unsigned long flags; | |
517 | int err; | |
518 | ||
519 | err = put_cmsg(msghdr, SOL_RDS, RDS_CMSG_CONG_UPDATE, | |
520 | sizeof(notify), ¬ify); | |
521 | if (err) | |
522 | return err; | |
523 | ||
524 | spin_lock_irqsave(&rs->rs_lock, flags); | |
525 | rs->rs_cong_notify &= ~notify; | |
526 | spin_unlock_irqrestore(&rs->rs_lock, flags); | |
527 | ||
528 | return 0; | |
529 | } | |
530 | ||
531 | /* | |
532 | * Receive any control messages. | |
533 | */ | |
5711f8b3 | 534 | static int rds_cmsg_recv(struct rds_incoming *inc, struct msghdr *msg, |
535 | struct rds_sock *rs) | |
bdbe6fbc AG |
536 | { |
537 | int ret = 0; | |
538 | ||
539 | if (inc->i_rdma_cookie) { | |
540 | ret = put_cmsg(msg, SOL_RDS, RDS_CMSG_RDMA_DEST, | |
541 | sizeof(inc->i_rdma_cookie), &inc->i_rdma_cookie); | |
542 | if (ret) | |
3289025a | 543 | goto out; |
bdbe6fbc AG |
544 | } |
545 | ||
5711f8b3 | 546 | if ((inc->i_rx_tstamp.tv_sec != 0) && |
547 | sock_flag(rds_rs_to_sk(rs), SOCK_RCVTSTAMP)) { | |
548 | ret = put_cmsg(msg, SOL_SOCKET, SCM_TIMESTAMP, | |
549 | sizeof(struct timeval), | |
550 | &inc->i_rx_tstamp); | |
551 | if (ret) | |
3289025a | 552 | goto out; |
5711f8b3 | 553 | } |
554 | ||
3289025a SS |
555 | if (rs->rs_rx_traces) { |
556 | struct rds_cmsg_rx_trace t; | |
557 | int i, j; | |
558 | ||
559 | inc->i_rx_lat_trace[RDS_MSG_RX_CMSG] = local_clock(); | |
560 | t.rx_traces = rs->rs_rx_traces; | |
561 | for (i = 0; i < rs->rs_rx_traces; i++) { | |
562 | j = rs->rs_rx_trace[i]; | |
563 | t.rx_trace_pos[i] = j; | |
564 | t.rx_trace[i] = inc->i_rx_lat_trace[j + 1] - | |
565 | inc->i_rx_lat_trace[j]; | |
566 | } | |
567 | ||
568 | ret = put_cmsg(msg, SOL_RDS, RDS_CMSG_RXPATH_LATENCY, | |
569 | sizeof(t), &t); | |
570 | if (ret) | |
571 | goto out; | |
572 | } | |
573 | ||
574 | out: | |
575 | return ret; | |
bdbe6fbc AG |
576 | } |
577 | ||
1b784140 YX |
578 | int rds_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, |
579 | int msg_flags) | |
bdbe6fbc AG |
580 | { |
581 | struct sock *sk = sock->sk; | |
582 | struct rds_sock *rs = rds_sk_to_rs(sk); | |
583 | long timeo; | |
584 | int ret = 0, nonblock = msg_flags & MSG_DONTWAIT; | |
342dfc30 | 585 | DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name); |
bdbe6fbc AG |
586 | struct rds_incoming *inc = NULL; |
587 | ||
588 | /* udp_recvmsg()->sock_recvtimeo() gets away without locking too.. */ | |
589 | timeo = sock_rcvtimeo(sk, nonblock); | |
590 | ||
591 | rdsdebug("size %zu flags 0x%x timeo %ld\n", size, msg_flags, timeo); | |
592 | ||
593 | if (msg_flags & MSG_OOB) | |
594 | goto out; | |
595 | ||
edacaeae | 596 | while (1) { |
c0371da6 | 597 | struct iov_iter save; |
edacaeae AG |
598 | /* If there are pending notifications, do those - and nothing else */ |
599 | if (!list_empty(&rs->rs_notify_queue)) { | |
600 | ret = rds_notify_queue_get(rs, msg); | |
601 | break; | |
602 | } | |
bdbe6fbc | 603 | |
edacaeae AG |
604 | if (rs->rs_cong_notify) { |
605 | ret = rds_notify_cong(rs, msg); | |
606 | break; | |
607 | } | |
bdbe6fbc | 608 | |
bdbe6fbc AG |
609 | if (!rds_next_incoming(rs, &inc)) { |
610 | if (nonblock) { | |
611 | ret = -EAGAIN; | |
612 | break; | |
613 | } | |
614 | ||
aa395145 | 615 | timeo = wait_event_interruptible_timeout(*sk_sleep(sk), |
f64f9e71 JP |
616 | (!list_empty(&rs->rs_notify_queue) || |
617 | rs->rs_cong_notify || | |
618 | rds_next_incoming(rs, &inc)), timeo); | |
bdbe6fbc AG |
619 | rdsdebug("recvmsg woke inc %p timeo %ld\n", inc, |
620 | timeo); | |
621 | if (timeo > 0 || timeo == MAX_SCHEDULE_TIMEOUT) | |
622 | continue; | |
623 | ||
624 | ret = timeo; | |
625 | if (ret == 0) | |
626 | ret = -ETIMEDOUT; | |
627 | break; | |
628 | } | |
629 | ||
630 | rdsdebug("copying inc %p from %pI4:%u to user\n", inc, | |
631 | &inc->i_conn->c_faddr, | |
632 | ntohs(inc->i_hdr.h_sport)); | |
c0371da6 AV |
633 | save = msg->msg_iter; |
634 | ret = inc->i_conn->c_trans->inc_copy_to_user(inc, &msg->msg_iter); | |
bdbe6fbc AG |
635 | if (ret < 0) |
636 | break; | |
637 | ||
638 | /* | |
639 | * if the message we just copied isn't at the head of the | |
640 | * recv queue then someone else raced us to return it, try | |
641 | * to get the next message. | |
642 | */ | |
643 | if (!rds_still_queued(rs, inc, !(msg_flags & MSG_PEEK))) { | |
644 | rds_inc_put(inc); | |
645 | inc = NULL; | |
646 | rds_stats_inc(s_recv_deliver_raced); | |
c0371da6 | 647 | msg->msg_iter = save; |
bdbe6fbc AG |
648 | continue; |
649 | } | |
650 | ||
651 | if (ret < be32_to_cpu(inc->i_hdr.h_len)) { | |
652 | if (msg_flags & MSG_TRUNC) | |
653 | ret = be32_to_cpu(inc->i_hdr.h_len); | |
654 | msg->msg_flags |= MSG_TRUNC; | |
655 | } | |
656 | ||
5711f8b3 | 657 | if (rds_cmsg_recv(inc, msg, rs)) { |
bdbe6fbc AG |
658 | ret = -EFAULT; |
659 | goto out; | |
660 | } | |
661 | ||
662 | rds_stats_inc(s_recv_delivered); | |
663 | ||
bdbe6fbc AG |
664 | if (sin) { |
665 | sin->sin_family = AF_INET; | |
666 | sin->sin_port = inc->i_hdr.h_sport; | |
667 | sin->sin_addr.s_addr = inc->i_saddr; | |
668 | memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); | |
06b6a1cf | 669 | msg->msg_namelen = sizeof(*sin); |
bdbe6fbc AG |
670 | } |
671 | break; | |
672 | } | |
673 | ||
674 | if (inc) | |
675 | rds_inc_put(inc); | |
676 | ||
677 | out: | |
678 | return ret; | |
679 | } | |
680 | ||
681 | /* | |
682 | * The socket is being shut down and we're asked to drop messages that were | |
683 | * queued for recvmsg. The caller has unbound the socket so the receive path | |
684 | * won't queue any more incoming fragments or messages on the socket. | |
685 | */ | |
686 | void rds_clear_recv_queue(struct rds_sock *rs) | |
687 | { | |
688 | struct sock *sk = rds_rs_to_sk(rs); | |
689 | struct rds_incoming *inc, *tmp; | |
690 | unsigned long flags; | |
691 | ||
692 | write_lock_irqsave(&rs->rs_recv_lock, flags); | |
693 | list_for_each_entry_safe(inc, tmp, &rs->rs_recv_queue, i_item) { | |
694 | rds_recv_rcvbuf_delta(rs, sk, inc->i_conn->c_lcong, | |
695 | -be32_to_cpu(inc->i_hdr.h_len), | |
696 | inc->i_hdr.h_dport); | |
697 | list_del_init(&inc->i_item); | |
698 | rds_inc_put(inc); | |
699 | } | |
700 | write_unlock_irqrestore(&rs->rs_recv_lock, flags); | |
701 | } | |
702 | ||
703 | /* | |
704 | * inc->i_saddr isn't used here because it is only set in the receive | |
705 | * path. | |
706 | */ | |
707 | void rds_inc_info_copy(struct rds_incoming *inc, | |
708 | struct rds_info_iterator *iter, | |
709 | __be32 saddr, __be32 daddr, int flip) | |
710 | { | |
711 | struct rds_info_message minfo; | |
712 | ||
713 | minfo.seq = be64_to_cpu(inc->i_hdr.h_sequence); | |
714 | minfo.len = be32_to_cpu(inc->i_hdr.h_len); | |
715 | ||
716 | if (flip) { | |
717 | minfo.laddr = daddr; | |
718 | minfo.faddr = saddr; | |
719 | minfo.lport = inc->i_hdr.h_dport; | |
720 | minfo.fport = inc->i_hdr.h_sport; | |
721 | } else { | |
722 | minfo.laddr = saddr; | |
723 | minfo.faddr = daddr; | |
724 | minfo.lport = inc->i_hdr.h_sport; | |
725 | minfo.fport = inc->i_hdr.h_dport; | |
726 | } | |
727 | ||
4116def2 KL |
728 | minfo.flags = 0; |
729 | ||
bdbe6fbc AG |
730 | rds_info_copy(iter, &minfo, sizeof(minfo)); |
731 | } |