1 /* 1 /*
2 * Copyright (c) 2006, 2017 Oracle and/or its 2 * Copyright (c) 2006, 2017 Oracle and/or its affiliates. All rights reserved.
3 * 3 *
4 * This software is available to you under a c 4 * This software is available to you under a choice of one of two
5 * licenses. You may choose to be licensed un 5 * licenses. You may choose to be licensed under the terms of the GNU
6 * General Public License (GPL) Version 2, ava 6 * General Public License (GPL) Version 2, available from the file
7 * COPYING in the main directory of this sourc 7 * COPYING in the main directory of this source tree, or the
8 * OpenIB.org BSD license below: 8 * OpenIB.org BSD license below:
9 * 9 *
10 * Redistribution and use in source and bi 10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted pro 11 * without modification, are permitted provided that the following
12 * conditions are met: 12 * conditions are met:
13 * 13 *
14 * - Redistributions of source code must 14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of condi 15 * copyright notice, this list of conditions and the following
16 * disclaimer. 16 * disclaimer.
17 * 17 *
18 * - Redistributions in binary form must 18 * - Redistributions in binary form must reproduce the above
19 * copyright notice, this list of condi 19 * copyright notice, this list of conditions and the following
20 * disclaimer in the documentation and/ 20 * disclaimer in the documentation and/or other materials
21 * provided with the distribution. 21 * provided with the distribution.
22 * 22 *
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT W 23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMIT 24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR P 25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTH 26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER L 27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARIS 28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29 * CONNECTION WITH THE SOFTWARE OR THE USE OR 29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
30 * SOFTWARE. 30 * SOFTWARE.
31 * 31 *
32 */ 32 */
33 #include <linux/kernel.h> 33 #include <linux/kernel.h>
34 #include <linux/slab.h> 34 #include <linux/slab.h>
35 #include <net/tcp.h> 35 #include <net/tcp.h>
36 #include <trace/events/sock.h> <<
37 36
38 #include "rds.h" 37 #include "rds.h"
39 #include "tcp.h" 38 #include "tcp.h"
40 39
41 static struct kmem_cache *rds_tcp_incoming_sla 40 static struct kmem_cache *rds_tcp_incoming_slab;
42 41
43 static void rds_tcp_inc_purge(struct rds_incom 42 static void rds_tcp_inc_purge(struct rds_incoming *inc)
44 { 43 {
45 struct rds_tcp_incoming *tinc; 44 struct rds_tcp_incoming *tinc;
46 tinc = container_of(inc, struct rds_tc 45 tinc = container_of(inc, struct rds_tcp_incoming, ti_inc);
47 rdsdebug("purging tinc %p inc %p\n", t 46 rdsdebug("purging tinc %p inc %p\n", tinc, inc);
48 skb_queue_purge(&tinc->ti_skb_list); 47 skb_queue_purge(&tinc->ti_skb_list);
49 } 48 }
50 49
51 void rds_tcp_inc_free(struct rds_incoming *inc 50 void rds_tcp_inc_free(struct rds_incoming *inc)
52 { 51 {
53 struct rds_tcp_incoming *tinc; 52 struct rds_tcp_incoming *tinc;
54 tinc = container_of(inc, struct rds_tc 53 tinc = container_of(inc, struct rds_tcp_incoming, ti_inc);
55 rds_tcp_inc_purge(inc); 54 rds_tcp_inc_purge(inc);
56 rdsdebug("freeing tinc %p inc %p\n", t 55 rdsdebug("freeing tinc %p inc %p\n", tinc, inc);
57 kmem_cache_free(rds_tcp_incoming_slab, 56 kmem_cache_free(rds_tcp_incoming_slab, tinc);
58 } 57 }
59 58
60 /* 59 /*
61 * this is pretty lame, but, whatever. 60 * this is pretty lame, but, whatever.
62 */ 61 */
63 int rds_tcp_inc_copy_to_user(struct rds_incomi 62 int rds_tcp_inc_copy_to_user(struct rds_incoming *inc, struct iov_iter *to)
64 { 63 {
65 struct rds_tcp_incoming *tinc; 64 struct rds_tcp_incoming *tinc;
66 struct sk_buff *skb; 65 struct sk_buff *skb;
67 int ret = 0; 66 int ret = 0;
68 67
69 if (!iov_iter_count(to)) 68 if (!iov_iter_count(to))
70 goto out; 69 goto out;
71 70
72 tinc = container_of(inc, struct rds_tc 71 tinc = container_of(inc, struct rds_tcp_incoming, ti_inc);
73 72
74 skb_queue_walk(&tinc->ti_skb_list, skb 73 skb_queue_walk(&tinc->ti_skb_list, skb) {
75 unsigned long to_copy, skb_off 74 unsigned long to_copy, skb_off;
76 for (skb_off = 0; skb_off < sk 75 for (skb_off = 0; skb_off < skb->len; skb_off += to_copy) {
77 to_copy = iov_iter_cou 76 to_copy = iov_iter_count(to);
78 to_copy = min(to_copy, 77 to_copy = min(to_copy, skb->len - skb_off);
79 78
80 if (skb_copy_datagram_ 79 if (skb_copy_datagram_iter(skb, skb_off, to, to_copy))
81 return -EFAULT 80 return -EFAULT;
82 81
83 rds_stats_add(s_copy_t 82 rds_stats_add(s_copy_to_user, to_copy);
84 ret += to_copy; 83 ret += to_copy;
85 84
86 if (!iov_iter_count(to 85 if (!iov_iter_count(to))
87 goto out; 86 goto out;
88 } 87 }
89 } 88 }
90 out: 89 out:
91 return ret; 90 return ret;
92 } 91 }
93 92
94 /* 93 /*
95 * We have a series of skbs that have fragment 94 * We have a series of skbs that have fragmented pieces of the congestion
96 * bitmap. They must add up to the exact size 95 * bitmap. They must add up to the exact size of the congestion bitmap. We
97 * use the skb helpers to copy those into the 96 * use the skb helpers to copy those into the pages that make up the in-memory
98 * congestion bitmap for the remote address of 97 * congestion bitmap for the remote address of this connection. We then tell
99 * the congestion core that the bitmap has bee 98 * the congestion core that the bitmap has been changed so that it can wake up
100 * sleepers. 99 * sleepers.
101 * 100 *
102 * This is racing with sending paths which are 101 * This is racing with sending paths which are using test_bit to see if the
103 * bitmap indicates that their recipient is co 102 * bitmap indicates that their recipient is congested.
104 */ 103 */
105 104
106 static void rds_tcp_cong_recv(struct rds_conne 105 static void rds_tcp_cong_recv(struct rds_connection *conn,
107 struct rds_tcp_i 106 struct rds_tcp_incoming *tinc)
108 { 107 {
109 struct sk_buff *skb; 108 struct sk_buff *skb;
110 unsigned int to_copy, skb_off; 109 unsigned int to_copy, skb_off;
111 unsigned int map_off; 110 unsigned int map_off;
112 unsigned int map_page; 111 unsigned int map_page;
113 struct rds_cong_map *map; 112 struct rds_cong_map *map;
114 int ret; 113 int ret;
115 114
116 /* catch completely corrupt packets */ 115 /* catch completely corrupt packets */
117 if (be32_to_cpu(tinc->ti_inc.i_hdr.h_l 116 if (be32_to_cpu(tinc->ti_inc.i_hdr.h_len) != RDS_CONG_MAP_BYTES)
118 return; 117 return;
119 118
120 map_page = 0; 119 map_page = 0;
121 map_off = 0; 120 map_off = 0;
122 map = conn->c_fcong; 121 map = conn->c_fcong;
123 122
124 skb_queue_walk(&tinc->ti_skb_list, skb 123 skb_queue_walk(&tinc->ti_skb_list, skb) {
125 skb_off = 0; 124 skb_off = 0;
126 while (skb_off < skb->len) { 125 while (skb_off < skb->len) {
127 to_copy = min_t(unsign 126 to_copy = min_t(unsigned int, PAGE_SIZE - map_off,
128 skb->l 127 skb->len - skb_off);
129 128
130 BUG_ON(map_page >= RDS 129 BUG_ON(map_page >= RDS_CONG_MAP_PAGES);
131 130
132 /* only returns 0 or - 131 /* only returns 0 or -error */
133 ret = skb_copy_bits(sk 132 ret = skb_copy_bits(skb, skb_off,
134 (void *)map->m 133 (void *)map->m_page_addrs[map_page] + map_off,
135 to_copy); 134 to_copy);
136 BUG_ON(ret != 0); 135 BUG_ON(ret != 0);
137 136
138 skb_off += to_copy; 137 skb_off += to_copy;
139 map_off += to_copy; 138 map_off += to_copy;
140 if (map_off == PAGE_SI 139 if (map_off == PAGE_SIZE) {
141 map_off = 0; 140 map_off = 0;
142 map_page++; 141 map_page++;
143 } 142 }
144 } 143 }
145 } 144 }
146 145
147 rds_cong_map_updated(map, ~(u64) 0); 146 rds_cong_map_updated(map, ~(u64) 0);
148 } 147 }
149 148
150 struct rds_tcp_desc_arg { 149 struct rds_tcp_desc_arg {
151 struct rds_conn_path *conn_path; 150 struct rds_conn_path *conn_path;
152 gfp_t gfp; 151 gfp_t gfp;
153 }; 152 };
154 153
155 static int rds_tcp_data_recv(read_descriptor_t 154 static int rds_tcp_data_recv(read_descriptor_t *desc, struct sk_buff *skb,
156 unsigned int offs 155 unsigned int offset, size_t len)
157 { 156 {
158 struct rds_tcp_desc_arg *arg = desc->a 157 struct rds_tcp_desc_arg *arg = desc->arg.data;
159 struct rds_conn_path *cp = arg->conn_p 158 struct rds_conn_path *cp = arg->conn_path;
160 struct rds_tcp_connection *tc = cp->cp 159 struct rds_tcp_connection *tc = cp->cp_transport_data;
161 struct rds_tcp_incoming *tinc = tc->t_ 160 struct rds_tcp_incoming *tinc = tc->t_tinc;
162 struct sk_buff *clone; 161 struct sk_buff *clone;
163 size_t left = len, to_copy; 162 size_t left = len, to_copy;
164 163
165 rdsdebug("tcp data tc %p skb %p offset 164 rdsdebug("tcp data tc %p skb %p offset %u len %zu\n", tc, skb, offset,
166 len); 165 len);
167 166
168 /* 167 /*
169 * tcp_read_sock() interprets partial 168 * tcp_read_sock() interprets partial progress as an indication to stop
170 * processing. 169 * processing.
171 */ 170 */
172 while (left) { 171 while (left) {
173 if (!tinc) { 172 if (!tinc) {
174 tinc = kmem_cache_allo 173 tinc = kmem_cache_alloc(rds_tcp_incoming_slab,
175 174 arg->gfp);
176 if (!tinc) { 175 if (!tinc) {
177 desc->error = 176 desc->error = -ENOMEM;
178 goto out; 177 goto out;
179 } 178 }
180 tc->t_tinc = tinc; 179 tc->t_tinc = tinc;
181 rdsdebug("allocated ti !! 180 rdsdebug("alloced tinc %p\n", tinc);
182 rds_inc_path_init(&tin 181 rds_inc_path_init(&tinc->ti_inc, cp,
183 &cp- 182 &cp->cp_conn->c_faddr);
184 tinc->ti_inc.i_rx_lat_ 183 tinc->ti_inc.i_rx_lat_trace[RDS_MSG_RX_HDR] =
185 local_ 184 local_clock();
186 185
187 /* 186 /*
188 * XXX * we might be a 187 * XXX * we might be able to use the __ variants when
189 * we've already seria 188 * we've already serialized at a higher level.
190 */ 189 */
191 skb_queue_head_init(&t 190 skb_queue_head_init(&tinc->ti_skb_list);
192 } 191 }
193 192
194 if (left && tc->t_tinc_hdr_rem 193 if (left && tc->t_tinc_hdr_rem) {
195 to_copy = min(tc->t_ti 194 to_copy = min(tc->t_tinc_hdr_rem, left);
196 rdsdebug("copying %zu 195 rdsdebug("copying %zu header from skb %p\n", to_copy,
197 skb); 196 skb);
198 skb_copy_bits(skb, off 197 skb_copy_bits(skb, offset,
199 (char *) 198 (char *)&tinc->ti_inc.i_hdr +
200 199 sizeof(struct rds_header) -
201 200 tc->t_tinc_hdr_rem,
202 to_copy) 201 to_copy);
203 tc->t_tinc_hdr_rem -= 202 tc->t_tinc_hdr_rem -= to_copy;
204 left -= to_copy; 203 left -= to_copy;
205 offset += to_copy; 204 offset += to_copy;
206 205
207 if (tc->t_tinc_hdr_rem 206 if (tc->t_tinc_hdr_rem == 0) {
208 /* could be 0 207 /* could be 0 for a 0 len message */
209 tc->t_tinc_dat 208 tc->t_tinc_data_rem =
210 be32_t 209 be32_to_cpu(tinc->ti_inc.i_hdr.h_len);
211 tinc->ti_inc.i 210 tinc->ti_inc.i_rx_lat_trace[RDS_MSG_RX_START] =
212 local_ 211 local_clock();
213 } 212 }
214 } 213 }
215 214
216 if (left && tc->t_tinc_data_re 215 if (left && tc->t_tinc_data_rem) {
217 to_copy = min(tc->t_ti 216 to_copy = min(tc->t_tinc_data_rem, left);
218 217
219 clone = pskb_extract(s 218 clone = pskb_extract(skb, offset, to_copy, arg->gfp);
220 if (!clone) { 219 if (!clone) {
221 desc->error = 220 desc->error = -ENOMEM;
222 goto out; 221 goto out;
223 } 222 }
224 223
225 skb_queue_tail(&tinc-> 224 skb_queue_tail(&tinc->ti_skb_list, clone);
226 225
227 rdsdebug("skb %p data 226 rdsdebug("skb %p data %p len %d off %u to_copy %zu -> "
228 "clone %p dat 227 "clone %p data %p len %d\n",
229 skb, skb->dat 228 skb, skb->data, skb->len, offset, to_copy,
230 clone, clone- 229 clone, clone->data, clone->len);
231 230
232 tc->t_tinc_data_rem -= 231 tc->t_tinc_data_rem -= to_copy;
233 left -= to_copy; 232 left -= to_copy;
234 offset += to_copy; 233 offset += to_copy;
235 } 234 }
236 235
237 if (tc->t_tinc_hdr_rem == 0 && 236 if (tc->t_tinc_hdr_rem == 0 && tc->t_tinc_data_rem == 0) {
238 struct rds_connection 237 struct rds_connection *conn = cp->cp_conn;
239 238
240 if (tinc->ti_inc.i_hdr 239 if (tinc->ti_inc.i_hdr.h_flags == RDS_FLAG_CONG_BITMAP)
241 rds_tcp_cong_r 240 rds_tcp_cong_recv(conn, tinc);
242 else 241 else
243 rds_recv_incom 242 rds_recv_incoming(conn, &conn->c_faddr,
244 243 &conn->c_laddr,
245 244 &tinc->ti_inc,
246 245 arg->gfp);
247 246
248 tc->t_tinc_hdr_rem = s 247 tc->t_tinc_hdr_rem = sizeof(struct rds_header);
249 tc->t_tinc_data_rem = 248 tc->t_tinc_data_rem = 0;
250 tc->t_tinc = NULL; 249 tc->t_tinc = NULL;
251 rds_inc_put(&tinc->ti_ 250 rds_inc_put(&tinc->ti_inc);
252 tinc = NULL; 251 tinc = NULL;
253 } 252 }
254 } 253 }
255 out: 254 out:
256 rdsdebug("returning len %zu left %zu s 255 rdsdebug("returning len %zu left %zu skb len %d rx queue depth %d\n",
257 len, left, skb->len, 256 len, left, skb->len,
258 skb_queue_len(&tc->t_sock->sk 257 skb_queue_len(&tc->t_sock->sk->sk_receive_queue));
259 return len - left; 258 return len - left;
260 } 259 }
261 260
262 /* the caller has to hold the sock lock */ 261 /* the caller has to hold the sock lock */
263 static int rds_tcp_read_sock(struct rds_conn_p 262 static int rds_tcp_read_sock(struct rds_conn_path *cp, gfp_t gfp)
264 { 263 {
265 struct rds_tcp_connection *tc = cp->cp 264 struct rds_tcp_connection *tc = cp->cp_transport_data;
266 struct socket *sock = tc->t_sock; 265 struct socket *sock = tc->t_sock;
267 read_descriptor_t desc; 266 read_descriptor_t desc;
268 struct rds_tcp_desc_arg arg; 267 struct rds_tcp_desc_arg arg;
269 268
270 /* It's like glib in the kernel! */ 269 /* It's like glib in the kernel! */
271 arg.conn_path = cp; 270 arg.conn_path = cp;
272 arg.gfp = gfp; 271 arg.gfp = gfp;
273 desc.arg.data = &arg; 272 desc.arg.data = &arg;
274 desc.error = 0; 273 desc.error = 0;
275 desc.count = 1; /* give more than one 274 desc.count = 1; /* give more than one skb per call */
276 275
277 tcp_read_sock(sock->sk, &desc, rds_tcp 276 tcp_read_sock(sock->sk, &desc, rds_tcp_data_recv);
278 rdsdebug("tcp_read_sock for tc %p gfp 277 rdsdebug("tcp_read_sock for tc %p gfp 0x%x returned %d\n", tc, gfp,
279 desc.error); 278 desc.error);
280 279
281 return desc.error; 280 return desc.error;
282 } 281 }
283 282
284 /* 283 /*
285 * We hold the sock lock to serialize our rds_ 284 * We hold the sock lock to serialize our rds_tcp_recv->tcp_read_sock from
286 * data_ready. 285 * data_ready.
287 * 286 *
288 * if we fail to allocate we're in trouble.. b 287 * if we fail to allocate we're in trouble.. blindly wait some time before
289 * trying again to see if the VM can free up s 288 * trying again to see if the VM can free up something for us.
290 */ 289 */
291 int rds_tcp_recv_path(struct rds_conn_path *cp 290 int rds_tcp_recv_path(struct rds_conn_path *cp)
292 { 291 {
293 struct rds_tcp_connection *tc = cp->cp 292 struct rds_tcp_connection *tc = cp->cp_transport_data;
294 struct socket *sock = tc->t_sock; 293 struct socket *sock = tc->t_sock;
295 int ret = 0; 294 int ret = 0;
296 295
297 rdsdebug("recv worker path [%d] tc %p 296 rdsdebug("recv worker path [%d] tc %p sock %p\n",
298 cp->cp_index, tc, sock); 297 cp->cp_index, tc, sock);
299 298
300 lock_sock(sock->sk); 299 lock_sock(sock->sk);
301 ret = rds_tcp_read_sock(cp, GFP_KERNEL 300 ret = rds_tcp_read_sock(cp, GFP_KERNEL);
302 release_sock(sock->sk); 301 release_sock(sock->sk);
303 302
304 return ret; 303 return ret;
305 } 304 }
306 305
307 void rds_tcp_data_ready(struct sock *sk) 306 void rds_tcp_data_ready(struct sock *sk)
308 { 307 {
309 void (*ready)(struct sock *sk); 308 void (*ready)(struct sock *sk);
310 struct rds_conn_path *cp; 309 struct rds_conn_path *cp;
311 struct rds_tcp_connection *tc; 310 struct rds_tcp_connection *tc;
312 311
313 trace_sk_data_ready(sk); <<
314 rdsdebug("data ready sk %p\n", sk); 312 rdsdebug("data ready sk %p\n", sk);
315 313
316 read_lock_bh(&sk->sk_callback_lock); 314 read_lock_bh(&sk->sk_callback_lock);
317 cp = sk->sk_user_data; 315 cp = sk->sk_user_data;
318 if (!cp) { /* check for teardown race 316 if (!cp) { /* check for teardown race */
319 ready = sk->sk_data_ready; 317 ready = sk->sk_data_ready;
320 goto out; 318 goto out;
321 } 319 }
322 320
323 tc = cp->cp_transport_data; 321 tc = cp->cp_transport_data;
324 ready = tc->t_orig_data_ready; 322 ready = tc->t_orig_data_ready;
325 rds_tcp_stats_inc(s_tcp_data_ready_cal 323 rds_tcp_stats_inc(s_tcp_data_ready_calls);
326 324
327 if (rds_tcp_read_sock(cp, GFP_ATOMIC) 325 if (rds_tcp_read_sock(cp, GFP_ATOMIC) == -ENOMEM) {
328 rcu_read_lock(); 326 rcu_read_lock();
329 if (!rds_destroy_pending(cp->c 327 if (!rds_destroy_pending(cp->cp_conn))
330 queue_delayed_work(rds 328 queue_delayed_work(rds_wq, &cp->cp_recv_w, 0);
331 rcu_read_unlock(); 329 rcu_read_unlock();
332 } 330 }
333 out: 331 out:
334 read_unlock_bh(&sk->sk_callback_lock); 332 read_unlock_bh(&sk->sk_callback_lock);
335 ready(sk); 333 ready(sk);
336 } 334 }
337 335
338 int rds_tcp_recv_init(void) 336 int rds_tcp_recv_init(void)
339 { 337 {
340 rds_tcp_incoming_slab = KMEM_CACHE(rds !! 338 rds_tcp_incoming_slab = kmem_cache_create("rds_tcp_incoming",
>> 339 sizeof(struct rds_tcp_incoming),
>> 340 0, 0, NULL);
341 if (!rds_tcp_incoming_slab) 341 if (!rds_tcp_incoming_slab)
342 return -ENOMEM; 342 return -ENOMEM;
343 return 0; 343 return 0;
344 } 344 }
345 345
346 void rds_tcp_recv_exit(void) 346 void rds_tcp_recv_exit(void)
347 { 347 {
348 kmem_cache_destroy(rds_tcp_incoming_sl 348 kmem_cache_destroy(rds_tcp_incoming_slab);
349 } 349 }
350 350
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.