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> 36 #include <trace/events/sock.h>
37 37
38 #include "rds.h" 38 #include "rds.h"
39 #include "tcp.h" 39 #include "tcp.h"
40 40
41 static struct kmem_cache *rds_tcp_incoming_sla 41 static struct kmem_cache *rds_tcp_incoming_slab;
42 42
43 static void rds_tcp_inc_purge(struct rds_incom 43 static void rds_tcp_inc_purge(struct rds_incoming *inc)
44 { 44 {
45 struct rds_tcp_incoming *tinc; 45 struct rds_tcp_incoming *tinc;
46 tinc = container_of(inc, struct rds_tc 46 tinc = container_of(inc, struct rds_tcp_incoming, ti_inc);
47 rdsdebug("purging tinc %p inc %p\n", t 47 rdsdebug("purging tinc %p inc %p\n", tinc, inc);
48 skb_queue_purge(&tinc->ti_skb_list); 48 skb_queue_purge(&tinc->ti_skb_list);
49 } 49 }
50 50
51 void rds_tcp_inc_free(struct rds_incoming *inc 51 void rds_tcp_inc_free(struct rds_incoming *inc)
52 { 52 {
53 struct rds_tcp_incoming *tinc; 53 struct rds_tcp_incoming *tinc;
54 tinc = container_of(inc, struct rds_tc 54 tinc = container_of(inc, struct rds_tcp_incoming, ti_inc);
55 rds_tcp_inc_purge(inc); 55 rds_tcp_inc_purge(inc);
56 rdsdebug("freeing tinc %p inc %p\n", t 56 rdsdebug("freeing tinc %p inc %p\n", tinc, inc);
57 kmem_cache_free(rds_tcp_incoming_slab, 57 kmem_cache_free(rds_tcp_incoming_slab, tinc);
58 } 58 }
59 59
60 /* 60 /*
61 * this is pretty lame, but, whatever. 61 * this is pretty lame, but, whatever.
62 */ 62 */
63 int rds_tcp_inc_copy_to_user(struct rds_incomi 63 int rds_tcp_inc_copy_to_user(struct rds_incoming *inc, struct iov_iter *to)
64 { 64 {
65 struct rds_tcp_incoming *tinc; 65 struct rds_tcp_incoming *tinc;
66 struct sk_buff *skb; 66 struct sk_buff *skb;
67 int ret = 0; 67 int ret = 0;
68 68
69 if (!iov_iter_count(to)) 69 if (!iov_iter_count(to))
70 goto out; 70 goto out;
71 71
72 tinc = container_of(inc, struct rds_tc 72 tinc = container_of(inc, struct rds_tcp_incoming, ti_inc);
73 73
74 skb_queue_walk(&tinc->ti_skb_list, skb 74 skb_queue_walk(&tinc->ti_skb_list, skb) {
75 unsigned long to_copy, skb_off 75 unsigned long to_copy, skb_off;
76 for (skb_off = 0; skb_off < sk 76 for (skb_off = 0; skb_off < skb->len; skb_off += to_copy) {
77 to_copy = iov_iter_cou 77 to_copy = iov_iter_count(to);
78 to_copy = min(to_copy, 78 to_copy = min(to_copy, skb->len - skb_off);
79 79
80 if (skb_copy_datagram_ 80 if (skb_copy_datagram_iter(skb, skb_off, to, to_copy))
81 return -EFAULT 81 return -EFAULT;
82 82
83 rds_stats_add(s_copy_t 83 rds_stats_add(s_copy_to_user, to_copy);
84 ret += to_copy; 84 ret += to_copy;
85 85
86 if (!iov_iter_count(to 86 if (!iov_iter_count(to))
87 goto out; 87 goto out;
88 } 88 }
89 } 89 }
90 out: 90 out:
91 return ret; 91 return ret;
92 } 92 }
93 93
94 /* 94 /*
95 * We have a series of skbs that have fragment 95 * We have a series of skbs that have fragmented pieces of the congestion
96 * bitmap. They must add up to the exact size 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 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 98 * congestion bitmap for the remote address of this connection. We then tell
99 * the congestion core that the bitmap has bee 99 * the congestion core that the bitmap has been changed so that it can wake up
100 * sleepers. 100 * sleepers.
101 * 101 *
102 * This is racing with sending paths which are 102 * This is racing with sending paths which are using test_bit to see if the
103 * bitmap indicates that their recipient is co 103 * bitmap indicates that their recipient is congested.
104 */ 104 */
105 105
106 static void rds_tcp_cong_recv(struct rds_conne 106 static void rds_tcp_cong_recv(struct rds_connection *conn,
107 struct rds_tcp_i 107 struct rds_tcp_incoming *tinc)
108 { 108 {
109 struct sk_buff *skb; 109 struct sk_buff *skb;
110 unsigned int to_copy, skb_off; 110 unsigned int to_copy, skb_off;
111 unsigned int map_off; 111 unsigned int map_off;
112 unsigned int map_page; 112 unsigned int map_page;
113 struct rds_cong_map *map; 113 struct rds_cong_map *map;
114 int ret; 114 int ret;
115 115
116 /* catch completely corrupt packets */ 116 /* catch completely corrupt packets */
117 if (be32_to_cpu(tinc->ti_inc.i_hdr.h_l 117 if (be32_to_cpu(tinc->ti_inc.i_hdr.h_len) != RDS_CONG_MAP_BYTES)
118 return; 118 return;
119 119
120 map_page = 0; 120 map_page = 0;
121 map_off = 0; 121 map_off = 0;
122 map = conn->c_fcong; 122 map = conn->c_fcong;
123 123
124 skb_queue_walk(&tinc->ti_skb_list, skb 124 skb_queue_walk(&tinc->ti_skb_list, skb) {
125 skb_off = 0; 125 skb_off = 0;
126 while (skb_off < skb->len) { 126 while (skb_off < skb->len) {
127 to_copy = min_t(unsign 127 to_copy = min_t(unsigned int, PAGE_SIZE - map_off,
128 skb->l 128 skb->len - skb_off);
129 129
130 BUG_ON(map_page >= RDS 130 BUG_ON(map_page >= RDS_CONG_MAP_PAGES);
131 131
132 /* only returns 0 or - 132 /* only returns 0 or -error */
133 ret = skb_copy_bits(sk 133 ret = skb_copy_bits(skb, skb_off,
134 (void *)map->m 134 (void *)map->m_page_addrs[map_page] + map_off,
135 to_copy); 135 to_copy);
136 BUG_ON(ret != 0); 136 BUG_ON(ret != 0);
137 137
138 skb_off += to_copy; 138 skb_off += to_copy;
139 map_off += to_copy; 139 map_off += to_copy;
140 if (map_off == PAGE_SI 140 if (map_off == PAGE_SIZE) {
141 map_off = 0; 141 map_off = 0;
142 map_page++; 142 map_page++;
143 } 143 }
144 } 144 }
145 } 145 }
146 146
147 rds_cong_map_updated(map, ~(u64) 0); 147 rds_cong_map_updated(map, ~(u64) 0);
148 } 148 }
149 149
150 struct rds_tcp_desc_arg { 150 struct rds_tcp_desc_arg {
151 struct rds_conn_path *conn_path; 151 struct rds_conn_path *conn_path;
152 gfp_t gfp; 152 gfp_t gfp;
153 }; 153 };
154 154
155 static int rds_tcp_data_recv(read_descriptor_t 155 static int rds_tcp_data_recv(read_descriptor_t *desc, struct sk_buff *skb,
156 unsigned int offs 156 unsigned int offset, size_t len)
157 { 157 {
158 struct rds_tcp_desc_arg *arg = desc->a 158 struct rds_tcp_desc_arg *arg = desc->arg.data;
159 struct rds_conn_path *cp = arg->conn_p 159 struct rds_conn_path *cp = arg->conn_path;
160 struct rds_tcp_connection *tc = cp->cp 160 struct rds_tcp_connection *tc = cp->cp_transport_data;
161 struct rds_tcp_incoming *tinc = tc->t_ 161 struct rds_tcp_incoming *tinc = tc->t_tinc;
162 struct sk_buff *clone; 162 struct sk_buff *clone;
163 size_t left = len, to_copy; 163 size_t left = len, to_copy;
164 164
165 rdsdebug("tcp data tc %p skb %p offset 165 rdsdebug("tcp data tc %p skb %p offset %u len %zu\n", tc, skb, offset,
166 len); 166 len);
167 167
168 /* 168 /*
169 * tcp_read_sock() interprets partial 169 * tcp_read_sock() interprets partial progress as an indication to stop
170 * processing. 170 * processing.
171 */ 171 */
172 while (left) { 172 while (left) {
173 if (!tinc) { 173 if (!tinc) {
174 tinc = kmem_cache_allo 174 tinc = kmem_cache_alloc(rds_tcp_incoming_slab,
175 175 arg->gfp);
176 if (!tinc) { 176 if (!tinc) {
177 desc->error = 177 desc->error = -ENOMEM;
178 goto out; 178 goto out;
179 } 179 }
180 tc->t_tinc = tinc; 180 tc->t_tinc = tinc;
181 rdsdebug("allocated ti 181 rdsdebug("allocated tinc %p\n", tinc);
182 rds_inc_path_init(&tin 182 rds_inc_path_init(&tinc->ti_inc, cp,
183 &cp- 183 &cp->cp_conn->c_faddr);
184 tinc->ti_inc.i_rx_lat_ 184 tinc->ti_inc.i_rx_lat_trace[RDS_MSG_RX_HDR] =
185 local_ 185 local_clock();
186 186
187 /* 187 /*
188 * XXX * we might be a 188 * XXX * we might be able to use the __ variants when
189 * we've already seria 189 * we've already serialized at a higher level.
190 */ 190 */
191 skb_queue_head_init(&t 191 skb_queue_head_init(&tinc->ti_skb_list);
192 } 192 }
193 193
194 if (left && tc->t_tinc_hdr_rem 194 if (left && tc->t_tinc_hdr_rem) {
195 to_copy = min(tc->t_ti 195 to_copy = min(tc->t_tinc_hdr_rem, left);
196 rdsdebug("copying %zu 196 rdsdebug("copying %zu header from skb %p\n", to_copy,
197 skb); 197 skb);
198 skb_copy_bits(skb, off 198 skb_copy_bits(skb, offset,
199 (char *) 199 (char *)&tinc->ti_inc.i_hdr +
200 200 sizeof(struct rds_header) -
201 201 tc->t_tinc_hdr_rem,
202 to_copy) 202 to_copy);
203 tc->t_tinc_hdr_rem -= 203 tc->t_tinc_hdr_rem -= to_copy;
204 left -= to_copy; 204 left -= to_copy;
205 offset += to_copy; 205 offset += to_copy;
206 206
207 if (tc->t_tinc_hdr_rem 207 if (tc->t_tinc_hdr_rem == 0) {
208 /* could be 0 208 /* could be 0 for a 0 len message */
209 tc->t_tinc_dat 209 tc->t_tinc_data_rem =
210 be32_t 210 be32_to_cpu(tinc->ti_inc.i_hdr.h_len);
211 tinc->ti_inc.i 211 tinc->ti_inc.i_rx_lat_trace[RDS_MSG_RX_START] =
212 local_ 212 local_clock();
213 } 213 }
214 } 214 }
215 215
216 if (left && tc->t_tinc_data_re 216 if (left && tc->t_tinc_data_rem) {
217 to_copy = min(tc->t_ti 217 to_copy = min(tc->t_tinc_data_rem, left);
218 218
219 clone = pskb_extract(s 219 clone = pskb_extract(skb, offset, to_copy, arg->gfp);
220 if (!clone) { 220 if (!clone) {
221 desc->error = 221 desc->error = -ENOMEM;
222 goto out; 222 goto out;
223 } 223 }
224 224
225 skb_queue_tail(&tinc-> 225 skb_queue_tail(&tinc->ti_skb_list, clone);
226 226
227 rdsdebug("skb %p data 227 rdsdebug("skb %p data %p len %d off %u to_copy %zu -> "
228 "clone %p dat 228 "clone %p data %p len %d\n",
229 skb, skb->dat 229 skb, skb->data, skb->len, offset, to_copy,
230 clone, clone- 230 clone, clone->data, clone->len);
231 231
232 tc->t_tinc_data_rem -= 232 tc->t_tinc_data_rem -= to_copy;
233 left -= to_copy; 233 left -= to_copy;
234 offset += to_copy; 234 offset += to_copy;
235 } 235 }
236 236
237 if (tc->t_tinc_hdr_rem == 0 && 237 if (tc->t_tinc_hdr_rem == 0 && tc->t_tinc_data_rem == 0) {
238 struct rds_connection 238 struct rds_connection *conn = cp->cp_conn;
239 239
240 if (tinc->ti_inc.i_hdr 240 if (tinc->ti_inc.i_hdr.h_flags == RDS_FLAG_CONG_BITMAP)
241 rds_tcp_cong_r 241 rds_tcp_cong_recv(conn, tinc);
242 else 242 else
243 rds_recv_incom 243 rds_recv_incoming(conn, &conn->c_faddr,
244 244 &conn->c_laddr,
245 245 &tinc->ti_inc,
246 246 arg->gfp);
247 247
248 tc->t_tinc_hdr_rem = s 248 tc->t_tinc_hdr_rem = sizeof(struct rds_header);
249 tc->t_tinc_data_rem = 249 tc->t_tinc_data_rem = 0;
250 tc->t_tinc = NULL; 250 tc->t_tinc = NULL;
251 rds_inc_put(&tinc->ti_ 251 rds_inc_put(&tinc->ti_inc);
252 tinc = NULL; 252 tinc = NULL;
253 } 253 }
254 } 254 }
255 out: 255 out:
256 rdsdebug("returning len %zu left %zu s 256 rdsdebug("returning len %zu left %zu skb len %d rx queue depth %d\n",
257 len, left, skb->len, 257 len, left, skb->len,
258 skb_queue_len(&tc->t_sock->sk 258 skb_queue_len(&tc->t_sock->sk->sk_receive_queue));
259 return len - left; 259 return len - left;
260 } 260 }
261 261
262 /* the caller has to hold the sock lock */ 262 /* the caller has to hold the sock lock */
263 static int rds_tcp_read_sock(struct rds_conn_p 263 static int rds_tcp_read_sock(struct rds_conn_path *cp, gfp_t gfp)
264 { 264 {
265 struct rds_tcp_connection *tc = cp->cp 265 struct rds_tcp_connection *tc = cp->cp_transport_data;
266 struct socket *sock = tc->t_sock; 266 struct socket *sock = tc->t_sock;
267 read_descriptor_t desc; 267 read_descriptor_t desc;
268 struct rds_tcp_desc_arg arg; 268 struct rds_tcp_desc_arg arg;
269 269
270 /* It's like glib in the kernel! */ 270 /* It's like glib in the kernel! */
271 arg.conn_path = cp; 271 arg.conn_path = cp;
272 arg.gfp = gfp; 272 arg.gfp = gfp;
273 desc.arg.data = &arg; 273 desc.arg.data = &arg;
274 desc.error = 0; 274 desc.error = 0;
275 desc.count = 1; /* give more than one 275 desc.count = 1; /* give more than one skb per call */
276 276
277 tcp_read_sock(sock->sk, &desc, rds_tcp 277 tcp_read_sock(sock->sk, &desc, rds_tcp_data_recv);
278 rdsdebug("tcp_read_sock for tc %p gfp 278 rdsdebug("tcp_read_sock for tc %p gfp 0x%x returned %d\n", tc, gfp,
279 desc.error); 279 desc.error);
280 280
281 return desc.error; 281 return desc.error;
282 } 282 }
283 283
284 /* 284 /*
285 * We hold the sock lock to serialize our rds_ 285 * We hold the sock lock to serialize our rds_tcp_recv->tcp_read_sock from
286 * data_ready. 286 * data_ready.
287 * 287 *
288 * if we fail to allocate we're in trouble.. b 288 * 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 289 * trying again to see if the VM can free up something for us.
290 */ 290 */
291 int rds_tcp_recv_path(struct rds_conn_path *cp 291 int rds_tcp_recv_path(struct rds_conn_path *cp)
292 { 292 {
293 struct rds_tcp_connection *tc = cp->cp 293 struct rds_tcp_connection *tc = cp->cp_transport_data;
294 struct socket *sock = tc->t_sock; 294 struct socket *sock = tc->t_sock;
295 int ret = 0; 295 int ret = 0;
296 296
297 rdsdebug("recv worker path [%d] tc %p 297 rdsdebug("recv worker path [%d] tc %p sock %p\n",
298 cp->cp_index, tc, sock); 298 cp->cp_index, tc, sock);
299 299
300 lock_sock(sock->sk); 300 lock_sock(sock->sk);
301 ret = rds_tcp_read_sock(cp, GFP_KERNEL 301 ret = rds_tcp_read_sock(cp, GFP_KERNEL);
302 release_sock(sock->sk); 302 release_sock(sock->sk);
303 303
304 return ret; 304 return ret;
305 } 305 }
306 306
307 void rds_tcp_data_ready(struct sock *sk) 307 void rds_tcp_data_ready(struct sock *sk)
308 { 308 {
309 void (*ready)(struct sock *sk); 309 void (*ready)(struct sock *sk);
310 struct rds_conn_path *cp; 310 struct rds_conn_path *cp;
311 struct rds_tcp_connection *tc; 311 struct rds_tcp_connection *tc;
312 312
313 trace_sk_data_ready(sk); 313 trace_sk_data_ready(sk);
314 rdsdebug("data ready sk %p\n", sk); 314 rdsdebug("data ready sk %p\n", sk);
315 315
316 read_lock_bh(&sk->sk_callback_lock); 316 read_lock_bh(&sk->sk_callback_lock);
317 cp = sk->sk_user_data; 317 cp = sk->sk_user_data;
318 if (!cp) { /* check for teardown race 318 if (!cp) { /* check for teardown race */
319 ready = sk->sk_data_ready; 319 ready = sk->sk_data_ready;
320 goto out; 320 goto out;
321 } 321 }
322 322
323 tc = cp->cp_transport_data; 323 tc = cp->cp_transport_data;
324 ready = tc->t_orig_data_ready; 324 ready = tc->t_orig_data_ready;
325 rds_tcp_stats_inc(s_tcp_data_ready_cal 325 rds_tcp_stats_inc(s_tcp_data_ready_calls);
326 326
327 if (rds_tcp_read_sock(cp, GFP_ATOMIC) 327 if (rds_tcp_read_sock(cp, GFP_ATOMIC) == -ENOMEM) {
328 rcu_read_lock(); 328 rcu_read_lock();
329 if (!rds_destroy_pending(cp->c 329 if (!rds_destroy_pending(cp->cp_conn))
330 queue_delayed_work(rds 330 queue_delayed_work(rds_wq, &cp->cp_recv_w, 0);
331 rcu_read_unlock(); 331 rcu_read_unlock();
332 } 332 }
333 out: 333 out:
334 read_unlock_bh(&sk->sk_callback_lock); 334 read_unlock_bh(&sk->sk_callback_lock);
335 ready(sk); 335 ready(sk);
336 } 336 }
337 337
338 int rds_tcp_recv_init(void) 338 int rds_tcp_recv_init(void)
339 { 339 {
340 rds_tcp_incoming_slab = KMEM_CACHE(rds !! 340 rds_tcp_incoming_slab = kmem_cache_create("rds_tcp_incoming",
>> 341 sizeof(struct rds_tcp_incoming),
>> 342 0, 0, NULL);
341 if (!rds_tcp_incoming_slab) 343 if (!rds_tcp_incoming_slab)
342 return -ENOMEM; 344 return -ENOMEM;
343 return 0; 345 return 0;
344 } 346 }
345 347
346 void rds_tcp_recv_exit(void) 348 void rds_tcp_recv_exit(void)
347 { 349 {
348 kmem_cache_destroy(rds_tcp_incoming_sl 350 kmem_cache_destroy(rds_tcp_incoming_slab);
349 } 351 }
350 352
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.