blob: c00dafffbb5acf3d844716b49d5fe434e710d006 [file] [log] [blame]
Andy Grover70041082009-08-21 12:28:31 +00001/*
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>
34#include <net/tcp.h>
35
36#include "rds.h"
37#include "tcp.h"
38
39static struct kmem_cache *rds_tcp_incoming_slab;
40
41void rds_tcp_inc_purge(struct rds_incoming *inc)
42{
43 struct rds_tcp_incoming *tinc;
44 tinc = container_of(inc, struct rds_tcp_incoming, ti_inc);
45 rdsdebug("purging tinc %p inc %p\n", tinc, inc);
46 skb_queue_purge(&tinc->ti_skb_list);
47}
48
49void rds_tcp_inc_free(struct rds_incoming *inc)
50{
51 struct rds_tcp_incoming *tinc;
52 tinc = container_of(inc, struct rds_tcp_incoming, ti_inc);
53 rds_tcp_inc_purge(inc);
54 rdsdebug("freeing tinc %p inc %p\n", tinc, inc);
55 kmem_cache_free(rds_tcp_incoming_slab, tinc);
56}
57
58/*
59 * this is pretty lame, but, whatever.
60 */
61int rds_tcp_inc_copy_to_user(struct rds_incoming *inc, struct iovec *first_iov,
62 size_t size)
63{
64 struct rds_tcp_incoming *tinc;
65 struct iovec *iov, tmp;
66 struct sk_buff *skb;
67 unsigned long to_copy, skb_off;
68 int ret = 0;
69
70 if (size == 0)
71 goto out;
72
73 tinc = container_of(inc, struct rds_tcp_incoming, ti_inc);
74 iov = first_iov;
75 tmp = *iov;
76
77 skb_queue_walk(&tinc->ti_skb_list, skb) {
78 skb_off = 0;
79 while (skb_off < skb->len) {
80 while (tmp.iov_len == 0) {
81 iov++;
82 tmp = *iov;
83 }
84
85 to_copy = min(tmp.iov_len, size);
86 to_copy = min(to_copy, skb->len - skb_off);
87
88 rdsdebug("ret %d size %zu skb %p skb_off %lu "
89 "skblen %d iov_base %p iov_len %zu cpy %lu\n",
90 ret, size, skb, skb_off, skb->len,
91 tmp.iov_base, tmp.iov_len, to_copy);
92
93 /* modifies tmp as it copies */
94 if (skb_copy_datagram_iovec(skb, skb_off, &tmp,
95 to_copy)) {
96 ret = -EFAULT;
97 goto out;
98 }
99
100 size -= to_copy;
101 ret += to_copy;
102 skb_off += to_copy;
103 if (size == 0)
104 goto out;
105 }
106 }
107out:
108 return ret;
109}
110
111/*
112 * We have a series of skbs that have fragmented pieces of the congestion
113 * bitmap. They must add up to the exact size of the congestion bitmap. We
114 * use the skb helpers to copy those into the pages that make up the in-memory
115 * congestion bitmap for the remote address of this connection. We then tell
116 * the congestion core that the bitmap has been changed so that it can wake up
117 * sleepers.
118 *
119 * This is racing with sending paths which are using test_bit to see if the
120 * bitmap indicates that their recipient is congested.
121 */
122
123static void rds_tcp_cong_recv(struct rds_connection *conn,
124 struct rds_tcp_incoming *tinc)
125{
126 struct sk_buff *skb;
127 unsigned int to_copy, skb_off;
128 unsigned int map_off;
129 unsigned int map_page;
130 struct rds_cong_map *map;
131 int ret;
132
133 /* catch completely corrupt packets */
134 if (be32_to_cpu(tinc->ti_inc.i_hdr.h_len) != RDS_CONG_MAP_BYTES)
135 return;
136
137 map_page = 0;
138 map_off = 0;
139 map = conn->c_fcong;
140
141 skb_queue_walk(&tinc->ti_skb_list, skb) {
142 skb_off = 0;
143 while (skb_off < skb->len) {
144 to_copy = min_t(unsigned int, PAGE_SIZE - map_off,
145 skb->len - skb_off);
146
147 BUG_ON(map_page >= RDS_CONG_MAP_PAGES);
148
149 /* only returns 0 or -error */
150 ret = skb_copy_bits(skb, skb_off,
151 (void *)map->m_page_addrs[map_page] + map_off,
152 to_copy);
153 BUG_ON(ret != 0);
154
155 skb_off += to_copy;
156 map_off += to_copy;
157 if (map_off == PAGE_SIZE) {
158 map_off = 0;
159 map_page++;
160 }
161 }
162 }
163
164 rds_cong_map_updated(map, ~(u64) 0);
165}
166
167struct rds_tcp_desc_arg {
168 struct rds_connection *conn;
169 gfp_t gfp;
170 enum km_type km;
171};
172
173static int rds_tcp_data_recv(read_descriptor_t *desc, struct sk_buff *skb,
174 unsigned int offset, size_t len)
175{
176 struct rds_tcp_desc_arg *arg = desc->arg.data;
177 struct rds_connection *conn = arg->conn;
178 struct rds_tcp_connection *tc = conn->c_transport_data;
179 struct rds_tcp_incoming *tinc = tc->t_tinc;
180 struct sk_buff *clone;
181 size_t left = len, to_copy;
182
183 rdsdebug("tcp data tc %p skb %p offset %u len %zu\n", tc, skb, offset,
184 len);
185
186 /*
187 * tcp_read_sock() interprets partial progress as an indication to stop
188 * processing.
189 */
190 while (left) {
191 if (tinc == NULL) {
192 tinc = kmem_cache_alloc(rds_tcp_incoming_slab,
193 arg->gfp);
194 if (tinc == NULL) {
195 desc->error = -ENOMEM;
196 goto out;
197 }
198 tc->t_tinc = tinc;
199 rdsdebug("alloced tinc %p\n", tinc);
200 rds_inc_init(&tinc->ti_inc, conn, conn->c_faddr);
201 /*
202 * XXX * we might be able to use the __ variants when
203 * we've already serialized at a higher level.
204 */
205 skb_queue_head_init(&tinc->ti_skb_list);
206 }
207
208 if (left && tc->t_tinc_hdr_rem) {
209 to_copy = min(tc->t_tinc_hdr_rem, left);
210 rdsdebug("copying %zu header from skb %p\n", to_copy,
211 skb);
212 skb_copy_bits(skb, offset,
213 (char *)&tinc->ti_inc.i_hdr +
214 sizeof(struct rds_header) -
215 tc->t_tinc_hdr_rem,
216 to_copy);
217 tc->t_tinc_hdr_rem -= to_copy;
218 left -= to_copy;
219 offset += to_copy;
220
221 if (tc->t_tinc_hdr_rem == 0) {
222 /* could be 0 for a 0 len message */
223 tc->t_tinc_data_rem =
224 be32_to_cpu(tinc->ti_inc.i_hdr.h_len);
225 }
226 }
227
228 if (left && tc->t_tinc_data_rem) {
229 clone = skb_clone(skb, arg->gfp);
230 if (clone == NULL) {
231 desc->error = -ENOMEM;
232 goto out;
233 }
234
235 to_copy = min(tc->t_tinc_data_rem, left);
236 pskb_pull(clone, offset);
237 pskb_trim(clone, to_copy);
238 skb_queue_tail(&tinc->ti_skb_list, clone);
239
240 rdsdebug("skb %p data %p len %d off %u to_copy %zu -> "
241 "clone %p data %p len %d\n",
242 skb, skb->data, skb->len, offset, to_copy,
243 clone, clone->data, clone->len);
244
245 tc->t_tinc_data_rem -= to_copy;
246 left -= to_copy;
247 offset += to_copy;
248 }
249
250 if (tc->t_tinc_hdr_rem == 0 && tc->t_tinc_data_rem == 0) {
251 if (tinc->ti_inc.i_hdr.h_flags == RDS_FLAG_CONG_BITMAP)
252 rds_tcp_cong_recv(conn, tinc);
253 else
254 rds_recv_incoming(conn, conn->c_faddr,
255 conn->c_laddr, &tinc->ti_inc,
256 arg->gfp, arg->km);
257
258 tc->t_tinc_hdr_rem = sizeof(struct rds_header);
259 tc->t_tinc_data_rem = 0;
260 tc->t_tinc = NULL;
261 rds_inc_put(&tinc->ti_inc);
262 tinc = NULL;
263 }
264 }
265out:
266 rdsdebug("returning len %zu left %zu skb len %d rx queue depth %d\n",
267 len, left, skb->len,
268 skb_queue_len(&tc->t_sock->sk->sk_receive_queue));
269 return len - left;
270}
271
272/* the caller has to hold the sock lock */
273int rds_tcp_read_sock(struct rds_connection *conn, gfp_t gfp, enum km_type km)
274{
275 struct rds_tcp_connection *tc = conn->c_transport_data;
276 struct socket *sock = tc->t_sock;
277 read_descriptor_t desc;
278 struct rds_tcp_desc_arg arg;
279
280 /* It's like glib in the kernel! */
281 arg.conn = conn;
282 arg.gfp = gfp;
283 arg.km = km;
284 desc.arg.data = &arg;
285 desc.error = 0;
286 desc.count = 1; /* give more than one skb per call */
287
288 tcp_read_sock(sock->sk, &desc, rds_tcp_data_recv);
289 rdsdebug("tcp_read_sock for tc %p gfp 0x%x returned %d\n", tc, gfp,
290 desc.error);
291
292 return desc.error;
293}
294
295/*
296 * We hold the sock lock to serialize our rds_tcp_recv->tcp_read_sock from
297 * data_ready.
298 *
299 * if we fail to allocate we're in trouble.. blindly wait some time before
300 * trying again to see if the VM can free up something for us.
301 */
302int rds_tcp_recv(struct rds_connection *conn)
303{
304 struct rds_tcp_connection *tc = conn->c_transport_data;
305 struct socket *sock = tc->t_sock;
306 int ret = 0;
307
308 rdsdebug("recv worker conn %p tc %p sock %p\n", conn, tc, sock);
309
310 lock_sock(sock->sk);
311 ret = rds_tcp_read_sock(conn, GFP_KERNEL, KM_USER0);
312 release_sock(sock->sk);
313
314 return ret;
315}
316
317void rds_tcp_data_ready(struct sock *sk, int bytes)
318{
319 void (*ready)(struct sock *sk, int bytes);
320 struct rds_connection *conn;
321 struct rds_tcp_connection *tc;
322
323 rdsdebug("data ready sk %p bytes %d\n", sk, bytes);
324
325 read_lock(&sk->sk_callback_lock);
326 conn = sk->sk_user_data;
327 if (conn == NULL) { /* check for teardown race */
328 ready = sk->sk_data_ready;
329 goto out;
330 }
331
332 tc = conn->c_transport_data;
333 ready = tc->t_orig_data_ready;
334 rds_tcp_stats_inc(s_tcp_data_ready_calls);
335
336 if (rds_tcp_read_sock(conn, GFP_ATOMIC, KM_SOFTIRQ0) == -ENOMEM)
337 queue_delayed_work(rds_wq, &conn->c_recv_w, 0);
338out:
339 read_unlock(&sk->sk_callback_lock);
340 ready(sk, bytes);
341}
342
343int __init rds_tcp_recv_init(void)
344{
345 rds_tcp_incoming_slab = kmem_cache_create("rds_tcp_incoming",
346 sizeof(struct rds_tcp_incoming),
347 0, 0, NULL);
348 if (rds_tcp_incoming_slab == NULL)
349 return -ENOMEM;
350 return 0;
351}
352
353void rds_tcp_recv_exit(void)
354{
355 kmem_cache_destroy(rds_tcp_incoming_slab);
356}