Commit | Line | Data |
---|---|---|
70041088 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> |
70041088 AG |
35 | #include <net/tcp.h> |
36 | ||
0cb43965 | 37 | #include "rds_single_path.h" |
70041088 AG |
38 | #include "rds.h" |
39 | #include "tcp.h" | |
40 | ||
41 | static struct kmem_cache *rds_tcp_incoming_slab; | |
42 | ||
809fa148 | 43 | static void rds_tcp_inc_purge(struct rds_incoming *inc) |
70041088 AG |
44 | { |
45 | struct rds_tcp_incoming *tinc; | |
46 | tinc = container_of(inc, struct rds_tcp_incoming, ti_inc); | |
47 | rdsdebug("purging tinc %p inc %p\n", tinc, inc); | |
48 | skb_queue_purge(&tinc->ti_skb_list); | |
49 | } | |
50 | ||
51 | void rds_tcp_inc_free(struct rds_incoming *inc) | |
52 | { | |
53 | struct rds_tcp_incoming *tinc; | |
54 | tinc = container_of(inc, struct rds_tcp_incoming, ti_inc); | |
55 | rds_tcp_inc_purge(inc); | |
56 | rdsdebug("freeing tinc %p inc %p\n", tinc, inc); | |
57 | kmem_cache_free(rds_tcp_incoming_slab, tinc); | |
58 | } | |
59 | ||
60 | /* | |
61 | * this is pretty lame, but, whatever. | |
62 | */ | |
c310e72c | 63 | int rds_tcp_inc_copy_to_user(struct rds_incoming *inc, struct iov_iter *to) |
70041088 AG |
64 | { |
65 | struct rds_tcp_incoming *tinc; | |
70041088 | 66 | struct sk_buff *skb; |
70041088 AG |
67 | int ret = 0; |
68 | ||
c310e72c | 69 | if (!iov_iter_count(to)) |
70041088 AG |
70 | goto out; |
71 | ||
72 | tinc = container_of(inc, struct rds_tcp_incoming, ti_inc); | |
70041088 AG |
73 | |
74 | skb_queue_walk(&tinc->ti_skb_list, skb) { | |
c310e72c AV |
75 | unsigned long to_copy, skb_off; |
76 | for (skb_off = 0; skb_off < skb->len; skb_off += to_copy) { | |
77 | to_copy = iov_iter_count(to); | |
70041088 AG |
78 | to_copy = min(to_copy, skb->len - skb_off); |
79 | ||
c310e72c AV |
80 | if (skb_copy_datagram_iter(skb, skb_off, to, to_copy)) |
81 | return -EFAULT; | |
70041088 | 82 | |
b075cfdb | 83 | rds_stats_add(s_copy_to_user, to_copy); |
70041088 | 84 | ret += to_copy; |
c310e72c AV |
85 | |
86 | if (!iov_iter_count(to)) | |
70041088 AG |
87 | goto out; |
88 | } | |
89 | } | |
90 | out: | |
91 | return ret; | |
92 | } | |
93 | ||
94 | /* | |
95 | * We have a series of skbs that have fragmented pieces of the congestion | |
96 | * bitmap. They must add up to the exact size of the congestion bitmap. We | |
97 | * use the skb helpers to copy those into the pages that make up the in-memory | |
98 | * congestion bitmap for the remote address of this connection. We then tell | |
99 | * the congestion core that the bitmap has been changed so that it can wake up | |
100 | * sleepers. | |
101 | * | |
102 | * This is racing with sending paths which are using test_bit to see if the | |
103 | * bitmap indicates that their recipient is congested. | |
104 | */ | |
105 | ||
106 | static void rds_tcp_cong_recv(struct rds_connection *conn, | |
107 | struct rds_tcp_incoming *tinc) | |
108 | { | |
109 | struct sk_buff *skb; | |
110 | unsigned int to_copy, skb_off; | |
111 | unsigned int map_off; | |
112 | unsigned int map_page; | |
113 | struct rds_cong_map *map; | |
114 | int ret; | |
115 | ||
116 | /* catch completely corrupt packets */ | |
117 | if (be32_to_cpu(tinc->ti_inc.i_hdr.h_len) != RDS_CONG_MAP_BYTES) | |
118 | return; | |
119 | ||
120 | map_page = 0; | |
121 | map_off = 0; | |
122 | map = conn->c_fcong; | |
123 | ||
124 | skb_queue_walk(&tinc->ti_skb_list, skb) { | |
125 | skb_off = 0; | |
126 | while (skb_off < skb->len) { | |
127 | to_copy = min_t(unsigned int, PAGE_SIZE - map_off, | |
128 | skb->len - skb_off); | |
129 | ||
130 | BUG_ON(map_page >= RDS_CONG_MAP_PAGES); | |
131 | ||
132 | /* only returns 0 or -error */ | |
133 | ret = skb_copy_bits(skb, skb_off, | |
134 | (void *)map->m_page_addrs[map_page] + map_off, | |
135 | to_copy); | |
136 | BUG_ON(ret != 0); | |
137 | ||
138 | skb_off += to_copy; | |
139 | map_off += to_copy; | |
140 | if (map_off == PAGE_SIZE) { | |
141 | map_off = 0; | |
142 | map_page++; | |
143 | } | |
144 | } | |
145 | } | |
146 | ||
147 | rds_cong_map_updated(map, ~(u64) 0); | |
148 | } | |
149 | ||
150 | struct rds_tcp_desc_arg { | |
151 | struct rds_connection *conn; | |
152 | gfp_t gfp; | |
70041088 AG |
153 | }; |
154 | ||
155 | static int rds_tcp_data_recv(read_descriptor_t *desc, struct sk_buff *skb, | |
156 | unsigned int offset, size_t len) | |
157 | { | |
158 | struct rds_tcp_desc_arg *arg = desc->arg.data; | |
159 | struct rds_connection *conn = arg->conn; | |
160 | struct rds_tcp_connection *tc = conn->c_transport_data; | |
161 | struct rds_tcp_incoming *tinc = tc->t_tinc; | |
162 | struct sk_buff *clone; | |
163 | size_t left = len, to_copy; | |
164 | ||
165 | rdsdebug("tcp data tc %p skb %p offset %u len %zu\n", tc, skb, offset, | |
166 | len); | |
167 | ||
168 | /* | |
169 | * tcp_read_sock() interprets partial progress as an indication to stop | |
170 | * processing. | |
171 | */ | |
172 | while (left) { | |
8690bfa1 | 173 | if (!tinc) { |
70041088 | 174 | tinc = kmem_cache_alloc(rds_tcp_incoming_slab, |
5c3da57d | 175 | arg->gfp); |
8690bfa1 | 176 | if (!tinc) { |
70041088 AG |
177 | desc->error = -ENOMEM; |
178 | goto out; | |
179 | } | |
180 | tc->t_tinc = tinc; | |
181 | rdsdebug("alloced tinc %p\n", tinc); | |
182 | rds_inc_init(&tinc->ti_inc, conn, conn->c_faddr); | |
183 | /* | |
184 | * XXX * we might be able to use the __ variants when | |
185 | * we've already serialized at a higher level. | |
186 | */ | |
187 | skb_queue_head_init(&tinc->ti_skb_list); | |
188 | } | |
189 | ||
190 | if (left && tc->t_tinc_hdr_rem) { | |
191 | to_copy = min(tc->t_tinc_hdr_rem, left); | |
192 | rdsdebug("copying %zu header from skb %p\n", to_copy, | |
193 | skb); | |
194 | skb_copy_bits(skb, offset, | |
195 | (char *)&tinc->ti_inc.i_hdr + | |
196 | sizeof(struct rds_header) - | |
197 | tc->t_tinc_hdr_rem, | |
198 | to_copy); | |
199 | tc->t_tinc_hdr_rem -= to_copy; | |
200 | left -= to_copy; | |
201 | offset += to_copy; | |
202 | ||
203 | if (tc->t_tinc_hdr_rem == 0) { | |
204 | /* could be 0 for a 0 len message */ | |
205 | tc->t_tinc_data_rem = | |
206 | be32_to_cpu(tinc->ti_inc.i_hdr.h_len); | |
207 | } | |
208 | } | |
209 | ||
210 | if (left && tc->t_tinc_data_rem) { | |
947d2756 SV |
211 | to_copy = min(tc->t_tinc_data_rem, left); |
212 | ||
213 | clone = pskb_extract(skb, offset, to_copy, arg->gfp); | |
8690bfa1 | 214 | if (!clone) { |
70041088 AG |
215 | desc->error = -ENOMEM; |
216 | goto out; | |
217 | } | |
218 | ||
70041088 AG |
219 | skb_queue_tail(&tinc->ti_skb_list, clone); |
220 | ||
221 | rdsdebug("skb %p data %p len %d off %u to_copy %zu -> " | |
222 | "clone %p data %p len %d\n", | |
223 | skb, skb->data, skb->len, offset, to_copy, | |
224 | clone, clone->data, clone->len); | |
225 | ||
226 | tc->t_tinc_data_rem -= to_copy; | |
227 | left -= to_copy; | |
228 | offset += to_copy; | |
229 | } | |
230 | ||
231 | if (tc->t_tinc_hdr_rem == 0 && tc->t_tinc_data_rem == 0) { | |
232 | if (tinc->ti_inc.i_hdr.h_flags == RDS_FLAG_CONG_BITMAP) | |
233 | rds_tcp_cong_recv(conn, tinc); | |
234 | else | |
235 | rds_recv_incoming(conn, conn->c_faddr, | |
236 | conn->c_laddr, &tinc->ti_inc, | |
6114eab5 | 237 | arg->gfp); |
70041088 AG |
238 | |
239 | tc->t_tinc_hdr_rem = sizeof(struct rds_header); | |
240 | tc->t_tinc_data_rem = 0; | |
241 | tc->t_tinc = NULL; | |
242 | rds_inc_put(&tinc->ti_inc); | |
243 | tinc = NULL; | |
244 | } | |
245 | } | |
246 | out: | |
247 | rdsdebug("returning len %zu left %zu skb len %d rx queue depth %d\n", | |
248 | len, left, skb->len, | |
249 | skb_queue_len(&tc->t_sock->sk->sk_receive_queue)); | |
250 | return len - left; | |
251 | } | |
252 | ||
253 | /* the caller has to hold the sock lock */ | |
6114eab5 | 254 | static int rds_tcp_read_sock(struct rds_connection *conn, gfp_t gfp) |
70041088 AG |
255 | { |
256 | struct rds_tcp_connection *tc = conn->c_transport_data; | |
257 | struct socket *sock = tc->t_sock; | |
258 | read_descriptor_t desc; | |
259 | struct rds_tcp_desc_arg arg; | |
260 | ||
261 | /* It's like glib in the kernel! */ | |
262 | arg.conn = conn; | |
263 | arg.gfp = gfp; | |
70041088 AG |
264 | desc.arg.data = &arg; |
265 | desc.error = 0; | |
266 | desc.count = 1; /* give more than one skb per call */ | |
267 | ||
268 | tcp_read_sock(sock->sk, &desc, rds_tcp_data_recv); | |
269 | rdsdebug("tcp_read_sock for tc %p gfp 0x%x returned %d\n", tc, gfp, | |
270 | desc.error); | |
271 | ||
272 | return desc.error; | |
273 | } | |
274 | ||
275 | /* | |
276 | * We hold the sock lock to serialize our rds_tcp_recv->tcp_read_sock from | |
277 | * data_ready. | |
278 | * | |
279 | * if we fail to allocate we're in trouble.. blindly wait some time before | |
280 | * trying again to see if the VM can free up something for us. | |
281 | */ | |
282 | int rds_tcp_recv(struct rds_connection *conn) | |
283 | { | |
284 | struct rds_tcp_connection *tc = conn->c_transport_data; | |
285 | struct socket *sock = tc->t_sock; | |
286 | int ret = 0; | |
287 | ||
288 | rdsdebug("recv worker conn %p tc %p sock %p\n", conn, tc, sock); | |
289 | ||
290 | lock_sock(sock->sk); | |
6114eab5 | 291 | ret = rds_tcp_read_sock(conn, GFP_KERNEL); |
70041088 AG |
292 | release_sock(sock->sk); |
293 | ||
294 | return ret; | |
295 | } | |
296 | ||
676d2369 | 297 | void rds_tcp_data_ready(struct sock *sk) |
70041088 | 298 | { |
676d2369 | 299 | void (*ready)(struct sock *sk); |
70041088 AG |
300 | struct rds_connection *conn; |
301 | struct rds_tcp_connection *tc; | |
302 | ||
676d2369 | 303 | rdsdebug("data ready sk %p\n", sk); |
70041088 | 304 | |
38036629 | 305 | read_lock_bh(&sk->sk_callback_lock); |
70041088 | 306 | conn = sk->sk_user_data; |
8690bfa1 | 307 | if (!conn) { /* check for teardown race */ |
70041088 AG |
308 | ready = sk->sk_data_ready; |
309 | goto out; | |
310 | } | |
311 | ||
312 | tc = conn->c_transport_data; | |
313 | ready = tc->t_orig_data_ready; | |
314 | rds_tcp_stats_inc(s_tcp_data_ready_calls); | |
315 | ||
6114eab5 | 316 | if (rds_tcp_read_sock(conn, GFP_ATOMIC) == -ENOMEM) |
70041088 AG |
317 | queue_delayed_work(rds_wq, &conn->c_recv_w, 0); |
318 | out: | |
38036629 | 319 | read_unlock_bh(&sk->sk_callback_lock); |
676d2369 | 320 | ready(sk); |
70041088 AG |
321 | } |
322 | ||
ef87b7ea | 323 | int rds_tcp_recv_init(void) |
70041088 AG |
324 | { |
325 | rds_tcp_incoming_slab = kmem_cache_create("rds_tcp_incoming", | |
326 | sizeof(struct rds_tcp_incoming), | |
327 | 0, 0, NULL); | |
8690bfa1 | 328 | if (!rds_tcp_incoming_slab) |
70041088 AG |
329 | return -ENOMEM; |
330 | return 0; | |
331 | } | |
332 | ||
333 | void rds_tcp_recv_exit(void) | |
334 | { | |
335 | kmem_cache_destroy(rds_tcp_incoming_slab); | |
336 | } |