1 /* 1 /*
2 * Copyright (c) 2006, 2018 Oracle and/or its !! 2 * Copyright (c) 2006 Oracle. 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 <linux/in.h> 35 #include <linux/in.h>
36 #include <linux/module.h> 36 #include <linux/module.h>
37 #include <net/tcp.h> 37 #include <net/tcp.h>
38 #include <net/net_namespace.h> 38 #include <net/net_namespace.h>
39 #include <net/netns/generic.h> 39 #include <net/netns/generic.h>
40 #include <net/addrconf.h> <<
41 40
42 #include "rds.h" 41 #include "rds.h"
43 #include "tcp.h" 42 #include "tcp.h"
44 43
45 /* only for info exporting */ 44 /* only for info exporting */
46 static DEFINE_SPINLOCK(rds_tcp_tc_list_lock); 45 static DEFINE_SPINLOCK(rds_tcp_tc_list_lock);
47 static LIST_HEAD(rds_tcp_tc_list); 46 static LIST_HEAD(rds_tcp_tc_list);
48 <<
49 /* rds_tcp_tc_count counts only IPv4 connectio <<
50 * rds6_tcp_tc_count counts both IPv4 and IPv6 <<
51 */ <<
52 static unsigned int rds_tcp_tc_count; 47 static unsigned int rds_tcp_tc_count;
53 #if IS_ENABLED(CONFIG_IPV6) <<
54 static unsigned int rds6_tcp_tc_count; <<
55 #endif <<
56 48
57 /* Track rds_tcp_connection structs so they ca 49 /* Track rds_tcp_connection structs so they can be cleaned up */
58 static DEFINE_SPINLOCK(rds_tcp_conn_lock); 50 static DEFINE_SPINLOCK(rds_tcp_conn_lock);
59 static LIST_HEAD(rds_tcp_conn_list); 51 static LIST_HEAD(rds_tcp_conn_list);
60 static atomic_t rds_tcp_unloading = ATOMIC_INI <<
61 52
62 static struct kmem_cache *rds_tcp_conn_slab; 53 static struct kmem_cache *rds_tcp_conn_slab;
63 54
64 static int rds_tcp_skbuf_handler(const struct !! 55 static int rds_tcp_skbuf_handler(struct ctl_table *ctl, int write,
65 void *buffer, !! 56 void __user *buffer, size_t *lenp,
>> 57 loff_t *fpos);
66 58
67 static int rds_tcp_min_sndbuf = SOCK_MIN_SNDBU 59 static int rds_tcp_min_sndbuf = SOCK_MIN_SNDBUF;
68 static int rds_tcp_min_rcvbuf = SOCK_MIN_RCVBU 60 static int rds_tcp_min_rcvbuf = SOCK_MIN_RCVBUF;
69 61
70 static struct ctl_table rds_tcp_sysctl_table[] 62 static struct ctl_table rds_tcp_sysctl_table[] = {
71 #define RDS_TCP_SNDBUF 0 63 #define RDS_TCP_SNDBUF 0
72 { 64 {
73 .procname = "rds_tcp_snd 65 .procname = "rds_tcp_sndbuf",
74 /* data is per-net pointer */ 66 /* data is per-net pointer */
75 .maxlen = sizeof(int), 67 .maxlen = sizeof(int),
76 .mode = 0644, 68 .mode = 0644,
77 .proc_handler = rds_tcp_skbu 69 .proc_handler = rds_tcp_skbuf_handler,
78 .extra1 = &rds_tcp_min 70 .extra1 = &rds_tcp_min_sndbuf,
79 }, 71 },
80 #define RDS_TCP_RCVBUF 1 72 #define RDS_TCP_RCVBUF 1
81 { 73 {
82 .procname = "rds_tcp_rcv 74 .procname = "rds_tcp_rcvbuf",
83 /* data is per-net pointer */ 75 /* data is per-net pointer */
84 .maxlen = sizeof(int), 76 .maxlen = sizeof(int),
85 .mode = 0644, 77 .mode = 0644,
86 .proc_handler = rds_tcp_skbu 78 .proc_handler = rds_tcp_skbuf_handler,
87 .extra1 = &rds_tcp_min 79 .extra1 = &rds_tcp_min_rcvbuf,
88 }, 80 },
>> 81 { }
89 }; 82 };
90 83
91 u32 rds_tcp_write_seq(struct rds_tcp_connectio !! 84 /* doing it this way avoids calling tcp_sk() */
>> 85 void rds_tcp_nonagle(struct socket *sock)
>> 86 {
>> 87 mm_segment_t oldfs = get_fs();
>> 88 int val = 1;
>> 89
>> 90 set_fs(KERNEL_DS);
>> 91 sock->ops->setsockopt(sock, SOL_TCP, TCP_NODELAY, (char __user *)&val,
>> 92 sizeof(val));
>> 93 set_fs(oldfs);
>> 94 }
>> 95
>> 96 u32 rds_tcp_snd_nxt(struct rds_tcp_connection *tc)
92 { 97 {
93 /* seq# of the last byte of data in tc !! 98 return tcp_sk(tc->t_sock->sk)->snd_nxt;
94 return tcp_sk(tc->t_sock->sk)->write_s <<
95 } 99 }
96 100
97 u32 rds_tcp_snd_una(struct rds_tcp_connection 101 u32 rds_tcp_snd_una(struct rds_tcp_connection *tc)
98 { 102 {
99 return tcp_sk(tc->t_sock->sk)->snd_una 103 return tcp_sk(tc->t_sock->sk)->snd_una;
100 } 104 }
101 105
102 void rds_tcp_restore_callbacks(struct socket * 106 void rds_tcp_restore_callbacks(struct socket *sock,
103 struct rds_tcp_ 107 struct rds_tcp_connection *tc)
104 { 108 {
105 rdsdebug("restoring sock %p callbacks 109 rdsdebug("restoring sock %p callbacks from tc %p\n", sock, tc);
106 write_lock_bh(&sock->sk->sk_callback_l 110 write_lock_bh(&sock->sk->sk_callback_lock);
107 111
108 /* done under the callback_lock to ser 112 /* done under the callback_lock to serialize with write_space */
109 spin_lock(&rds_tcp_tc_list_lock); 113 spin_lock(&rds_tcp_tc_list_lock);
110 list_del_init(&tc->t_list_item); 114 list_del_init(&tc->t_list_item);
111 #if IS_ENABLED(CONFIG_IPV6) !! 115 rds_tcp_tc_count--;
112 rds6_tcp_tc_count--; <<
113 #endif <<
114 if (!tc->t_cpath->cp_conn->c_isv6) <<
115 rds_tcp_tc_count--; <<
116 spin_unlock(&rds_tcp_tc_list_lock); 116 spin_unlock(&rds_tcp_tc_list_lock);
117 117
118 tc->t_sock = NULL; 118 tc->t_sock = NULL;
119 119
120 sock->sk->sk_write_space = tc->t_orig_ 120 sock->sk->sk_write_space = tc->t_orig_write_space;
121 sock->sk->sk_data_ready = tc->t_orig_d 121 sock->sk->sk_data_ready = tc->t_orig_data_ready;
122 sock->sk->sk_state_change = tc->t_orig 122 sock->sk->sk_state_change = tc->t_orig_state_change;
123 sock->sk->sk_user_data = NULL; 123 sock->sk->sk_user_data = NULL;
124 124
125 write_unlock_bh(&sock->sk->sk_callback 125 write_unlock_bh(&sock->sk->sk_callback_lock);
126 } 126 }
127 127
128 /* 128 /*
129 * rds_tcp_reset_callbacks() switches the to t 129 * rds_tcp_reset_callbacks() switches the to the new sock and
130 * returns the existing tc->t_sock. 130 * returns the existing tc->t_sock.
131 * 131 *
132 * The only functions that set tc->t_sock are 132 * The only functions that set tc->t_sock are rds_tcp_set_callbacks
133 * and rds_tcp_reset_callbacks. Send and rece 133 * and rds_tcp_reset_callbacks. Send and receive trust that
134 * it is set. The absence of RDS_CONN_UP bit 134 * it is set. The absence of RDS_CONN_UP bit protects those paths
135 * from being called while it isn't set. 135 * from being called while it isn't set.
136 */ 136 */
137 void rds_tcp_reset_callbacks(struct socket *so 137 void rds_tcp_reset_callbacks(struct socket *sock,
138 struct rds_conn_p 138 struct rds_conn_path *cp)
139 { 139 {
140 struct rds_tcp_connection *tc = cp->cp 140 struct rds_tcp_connection *tc = cp->cp_transport_data;
141 struct socket *osock = tc->t_sock; 141 struct socket *osock = tc->t_sock;
142 142
143 if (!osock) 143 if (!osock)
144 goto newsock; 144 goto newsock;
145 145
146 /* Need to resolve a duelling SYN betw 146 /* Need to resolve a duelling SYN between peers.
147 * We have an outstanding SYN to this 147 * We have an outstanding SYN to this peer, which may
148 * potentially have transitioned to th 148 * potentially have transitioned to the RDS_CONN_UP state,
149 * so we must quiesce any send threads 149 * so we must quiesce any send threads before resetting
150 * cp_transport_data. We quiesce these 150 * cp_transport_data. We quiesce these threads by setting
151 * cp_state to something other than RD 151 * cp_state to something other than RDS_CONN_UP, and then
152 * waiting for any existing threads in 152 * waiting for any existing threads in rds_send_xmit to
153 * complete release_in_xmit(). (Subseq 153 * complete release_in_xmit(). (Subsequent threads entering
154 * rds_send_xmit() will bail on !rds_c 154 * rds_send_xmit() will bail on !rds_conn_up().
155 * 155 *
156 * However an incoming syn-ack at this 156 * However an incoming syn-ack at this point would end up
157 * marking the conn as RDS_CONN_UP, an 157 * marking the conn as RDS_CONN_UP, and would again permit
158 * rds_send_xmi() threads through, so 158 * rds_send_xmi() threads through, so ideally we would
159 * synchronize on RDS_CONN_UP after lo 159 * synchronize on RDS_CONN_UP after lock_sock(), but cannot
160 * do that: waiting on !RDS_IN_XMIT af 160 * do that: waiting on !RDS_IN_XMIT after lock_sock() may
161 * end up deadlocking with tcp_sendmsg 161 * end up deadlocking with tcp_sendmsg(), and the RDS_IN_XMIT
162 * would not get set. As a result, we 162 * would not get set. As a result, we set c_state to
163 * RDS_CONN_RESETTTING, to ensure that 163 * RDS_CONN_RESETTTING, to ensure that rds_tcp_state_change
164 * cannot mark rds_conn_path_up() in t 164 * cannot mark rds_conn_path_up() in the window before lock_sock()
165 */ 165 */
166 atomic_set(&cp->cp_state, RDS_CONN_RES 166 atomic_set(&cp->cp_state, RDS_CONN_RESETTING);
167 wait_event(cp->cp_waitq, !test_bit(RDS 167 wait_event(cp->cp_waitq, !test_bit(RDS_IN_XMIT, &cp->cp_flags));
>> 168 lock_sock(osock->sk);
168 /* reset receive side state for rds_tc 169 /* reset receive side state for rds_tcp_data_recv() for osock */
169 cancel_delayed_work_sync(&cp->cp_send_ 170 cancel_delayed_work_sync(&cp->cp_send_w);
170 cancel_delayed_work_sync(&cp->cp_recv_ 171 cancel_delayed_work_sync(&cp->cp_recv_w);
171 lock_sock(osock->sk); <<
172 if (tc->t_tinc) { 172 if (tc->t_tinc) {
173 rds_inc_put(&tc->t_tinc->ti_in 173 rds_inc_put(&tc->t_tinc->ti_inc);
174 tc->t_tinc = NULL; 174 tc->t_tinc = NULL;
175 } 175 }
176 tc->t_tinc_hdr_rem = sizeof(struct rds 176 tc->t_tinc_hdr_rem = sizeof(struct rds_header);
177 tc->t_tinc_data_rem = 0; 177 tc->t_tinc_data_rem = 0;
178 rds_tcp_restore_callbacks(osock, tc); 178 rds_tcp_restore_callbacks(osock, tc);
179 release_sock(osock->sk); 179 release_sock(osock->sk);
180 sock_release(osock); 180 sock_release(osock);
181 newsock: 181 newsock:
182 rds_send_path_reset(cp); 182 rds_send_path_reset(cp);
183 lock_sock(sock->sk); 183 lock_sock(sock->sk);
184 rds_tcp_set_callbacks(sock, cp); 184 rds_tcp_set_callbacks(sock, cp);
185 release_sock(sock->sk); 185 release_sock(sock->sk);
186 } 186 }
187 187
188 /* Add tc to rds_tcp_tc_list and set tc->t_soc 188 /* Add tc to rds_tcp_tc_list and set tc->t_sock. See comments
189 * above rds_tcp_reset_callbacks for notes abo 189 * above rds_tcp_reset_callbacks for notes about synchronization
190 * with data path 190 * with data path
191 */ 191 */
192 void rds_tcp_set_callbacks(struct socket *sock 192 void rds_tcp_set_callbacks(struct socket *sock, struct rds_conn_path *cp)
193 { 193 {
194 struct rds_tcp_connection *tc = cp->cp 194 struct rds_tcp_connection *tc = cp->cp_transport_data;
195 195
196 rdsdebug("setting sock %p callbacks to 196 rdsdebug("setting sock %p callbacks to tc %p\n", sock, tc);
197 write_lock_bh(&sock->sk->sk_callback_l 197 write_lock_bh(&sock->sk->sk_callback_lock);
198 198
199 /* done under the callback_lock to ser 199 /* done under the callback_lock to serialize with write_space */
200 spin_lock(&rds_tcp_tc_list_lock); 200 spin_lock(&rds_tcp_tc_list_lock);
201 list_add_tail(&tc->t_list_item, &rds_t 201 list_add_tail(&tc->t_list_item, &rds_tcp_tc_list);
202 #if IS_ENABLED(CONFIG_IPV6) !! 202 rds_tcp_tc_count++;
203 rds6_tcp_tc_count++; <<
204 #endif <<
205 if (!tc->t_cpath->cp_conn->c_isv6) <<
206 rds_tcp_tc_count++; <<
207 spin_unlock(&rds_tcp_tc_list_lock); 203 spin_unlock(&rds_tcp_tc_list_lock);
208 204
209 /* accepted sockets need our listen da 205 /* accepted sockets need our listen data ready undone */
210 if (sock->sk->sk_data_ready == rds_tcp 206 if (sock->sk->sk_data_ready == rds_tcp_listen_data_ready)
211 sock->sk->sk_data_ready = sock 207 sock->sk->sk_data_ready = sock->sk->sk_user_data;
212 208
213 tc->t_sock = sock; 209 tc->t_sock = sock;
214 tc->t_cpath = cp; 210 tc->t_cpath = cp;
215 tc->t_orig_data_ready = sock->sk->sk_d 211 tc->t_orig_data_ready = sock->sk->sk_data_ready;
216 tc->t_orig_write_space = sock->sk->sk_ 212 tc->t_orig_write_space = sock->sk->sk_write_space;
217 tc->t_orig_state_change = sock->sk->sk 213 tc->t_orig_state_change = sock->sk->sk_state_change;
218 214
219 sock->sk->sk_user_data = cp; 215 sock->sk->sk_user_data = cp;
220 sock->sk->sk_data_ready = rds_tcp_data 216 sock->sk->sk_data_ready = rds_tcp_data_ready;
221 sock->sk->sk_write_space = rds_tcp_wri 217 sock->sk->sk_write_space = rds_tcp_write_space;
222 sock->sk->sk_state_change = rds_tcp_st 218 sock->sk->sk_state_change = rds_tcp_state_change;
223 219
224 write_unlock_bh(&sock->sk->sk_callback 220 write_unlock_bh(&sock->sk->sk_callback_lock);
225 } 221 }
226 222
227 /* Handle RDS_INFO_TCP_SOCKETS socket option. <<
228 * connections for backward compatibility. <<
229 */ <<
230 static void rds_tcp_tc_info(struct socket *rds 223 static void rds_tcp_tc_info(struct socket *rds_sock, unsigned int len,
231 struct rds_info_it 224 struct rds_info_iterator *iter,
232 struct rds_info_le 225 struct rds_info_lengths *lens)
233 { 226 {
234 struct rds_info_tcp_socket tsinfo; 227 struct rds_info_tcp_socket tsinfo;
235 struct rds_tcp_connection *tc; 228 struct rds_tcp_connection *tc;
236 unsigned long flags; 229 unsigned long flags;
>> 230 struct sockaddr_in sin;
>> 231 int sinlen;
>> 232 struct socket *sock;
237 233
238 spin_lock_irqsave(&rds_tcp_tc_list_loc 234 spin_lock_irqsave(&rds_tcp_tc_list_lock, flags);
239 235
240 if (len / sizeof(tsinfo) < rds_tcp_tc_ 236 if (len / sizeof(tsinfo) < rds_tcp_tc_count)
241 goto out; 237 goto out;
242 238
243 list_for_each_entry(tc, &rds_tcp_tc_li 239 list_for_each_entry(tc, &rds_tcp_tc_list, t_list_item) {
244 struct inet_sock *inet = inet_ <<
245 <<
246 if (tc->t_cpath->cp_conn->c_is <<
247 continue; <<
248 240
249 tsinfo.local_addr = inet->inet !! 241 sock = tc->t_sock;
250 tsinfo.local_port = inet->inet !! 242 if (sock) {
251 tsinfo.peer_addr = inet->inet_ !! 243 sock->ops->getname(sock, (struct sockaddr *)&sin,
252 tsinfo.peer_port = inet->inet_ !! 244 &sinlen, 0);
>> 245 tsinfo.local_addr = sin.sin_addr.s_addr;
>> 246 tsinfo.local_port = sin.sin_port;
>> 247 sock->ops->getname(sock, (struct sockaddr *)&sin,
>> 248 &sinlen, 1);
>> 249 tsinfo.peer_addr = sin.sin_addr.s_addr;
>> 250 tsinfo.peer_port = sin.sin_port;
>> 251 }
253 252
254 tsinfo.hdr_rem = tc->t_tinc_hd 253 tsinfo.hdr_rem = tc->t_tinc_hdr_rem;
255 tsinfo.data_rem = tc->t_tinc_d 254 tsinfo.data_rem = tc->t_tinc_data_rem;
256 tsinfo.last_sent_nxt = tc->t_l 255 tsinfo.last_sent_nxt = tc->t_last_sent_nxt;
257 tsinfo.last_expected_una = tc- 256 tsinfo.last_expected_una = tc->t_last_expected_una;
258 tsinfo.last_seen_una = tc->t_l 257 tsinfo.last_seen_una = tc->t_last_seen_una;
259 tsinfo.tos = tc->t_cpath->cp_c <<
260 258
261 rds_info_copy(iter, &tsinfo, s 259 rds_info_copy(iter, &tsinfo, sizeof(tsinfo));
262 } 260 }
263 261
264 out: 262 out:
265 lens->nr = rds_tcp_tc_count; 263 lens->nr = rds_tcp_tc_count;
266 lens->each = sizeof(tsinfo); 264 lens->each = sizeof(tsinfo);
267 265
268 spin_unlock_irqrestore(&rds_tcp_tc_lis 266 spin_unlock_irqrestore(&rds_tcp_tc_list_lock, flags);
269 } 267 }
270 268
271 #if IS_ENABLED(CONFIG_IPV6) !! 269 static int rds_tcp_laddr_check(struct net *net, __be32 addr)
272 /* Handle RDS6_INFO_TCP_SOCKETS socket option. <<
273 * IPv6 connections. IPv4 connection address i <<
274 * address. <<
275 */ <<
276 static void rds6_tcp_tc_info(struct socket *so <<
277 struct rds_info_i <<
278 struct rds_info_l <<
279 { <<
280 struct rds6_info_tcp_socket tsinfo6; <<
281 struct rds_tcp_connection *tc; <<
282 unsigned long flags; <<
283 <<
284 spin_lock_irqsave(&rds_tcp_tc_list_loc <<
285 <<
286 if (len / sizeof(tsinfo6) < rds6_tcp_t <<
287 goto out; <<
288 <<
289 list_for_each_entry(tc, &rds_tcp_tc_li <<
290 struct sock *sk = tc->t_sock-> <<
291 struct inet_sock *inet = inet_ <<
292 <<
293 tsinfo6.local_addr = sk->sk_v6 <<
294 tsinfo6.local_port = inet->ine <<
295 tsinfo6.peer_addr = sk->sk_v6_ <<
296 tsinfo6.peer_port = inet->inet <<
297 <<
298 tsinfo6.hdr_rem = tc->t_tinc_h <<
299 tsinfo6.data_rem = tc->t_tinc_ <<
300 tsinfo6.last_sent_nxt = tc->t_ <<
301 tsinfo6.last_expected_una = tc <<
302 tsinfo6.last_seen_una = tc->t_ <<
303 <<
304 rds_info_copy(iter, &tsinfo6, <<
305 } <<
306 <<
307 out: <<
308 lens->nr = rds6_tcp_tc_count; <<
309 lens->each = sizeof(tsinfo6); <<
310 <<
311 spin_unlock_irqrestore(&rds_tcp_tc_lis <<
312 } <<
313 #endif <<
314 <<
315 int rds_tcp_laddr_check(struct net *net, const <<
316 __u32 scope_id) <<
317 { 270 {
318 struct net_device *dev = NULL; !! 271 if (inet_addr_type(net, addr) == RTN_LOCAL)
319 #if IS_ENABLED(CONFIG_IPV6) <<
320 int ret; <<
321 #endif <<
322 <<
323 if (ipv6_addr_v4mapped(addr)) { <<
324 if (inet_addr_type(net, addr-> <<
325 return 0; <<
326 return -EADDRNOTAVAIL; <<
327 } <<
328 <<
329 /* If the scope_id is specified, check <<
330 * hosted on the specified interface. <<
331 */ <<
332 if (scope_id != 0) { <<
333 rcu_read_lock(); <<
334 dev = dev_get_by_index_rcu(net <<
335 /* scope_id is not valid... */ <<
336 if (!dev) { <<
337 rcu_read_unlock(); <<
338 return -EADDRNOTAVAIL; <<
339 } <<
340 rcu_read_unlock(); <<
341 } <<
342 #if IS_ENABLED(CONFIG_IPV6) <<
343 ret = ipv6_chk_addr(net, addr, dev, 0) <<
344 if (ret) <<
345 return 0; 272 return 0;
346 #endif <<
347 return -EADDRNOTAVAIL; 273 return -EADDRNOTAVAIL;
348 } 274 }
349 275
350 static void rds_tcp_conn_free(void *arg) <<
351 { <<
352 struct rds_tcp_connection *tc = arg; <<
353 unsigned long flags; <<
354 <<
355 rdsdebug("freeing tc %p\n", tc); <<
356 <<
357 spin_lock_irqsave(&rds_tcp_conn_lock, <<
358 if (!tc->t_tcp_node_detached) <<
359 list_del(&tc->t_tcp_node); <<
360 spin_unlock_irqrestore(&rds_tcp_conn_l <<
361 <<
362 kmem_cache_free(rds_tcp_conn_slab, tc) <<
363 } <<
364 <<
365 static int rds_tcp_conn_alloc(struct rds_conne 276 static int rds_tcp_conn_alloc(struct rds_connection *conn, gfp_t gfp)
366 { 277 {
367 struct rds_tcp_connection *tc; 278 struct rds_tcp_connection *tc;
368 int i, j; !! 279 int i;
369 int ret = 0; <<
370 280
371 for (i = 0; i < RDS_MPATH_WORKERS; i++ 281 for (i = 0; i < RDS_MPATH_WORKERS; i++) {
372 tc = kmem_cache_alloc(rds_tcp_ 282 tc = kmem_cache_alloc(rds_tcp_conn_slab, gfp);
373 if (!tc) { !! 283 if (!tc)
374 ret = -ENOMEM; !! 284 return -ENOMEM;
375 goto fail; !! 285
376 } <<
377 mutex_init(&tc->t_conn_path_lo 286 mutex_init(&tc->t_conn_path_lock);
378 tc->t_sock = NULL; 287 tc->t_sock = NULL;
379 tc->t_tinc = NULL; 288 tc->t_tinc = NULL;
380 tc->t_tinc_hdr_rem = sizeof(st 289 tc->t_tinc_hdr_rem = sizeof(struct rds_header);
381 tc->t_tinc_data_rem = 0; 290 tc->t_tinc_data_rem = 0;
382 291
383 conn->c_path[i].cp_transport_d 292 conn->c_path[i].cp_transport_data = tc;
384 tc->t_cpath = &conn->c_path[i] 293 tc->t_cpath = &conn->c_path[i];
385 tc->t_tcp_node_detached = true <<
386 294
>> 295 spin_lock_irq(&rds_tcp_conn_lock);
>> 296 list_add_tail(&tc->t_tcp_node, &rds_tcp_conn_list);
>> 297 spin_unlock_irq(&rds_tcp_conn_lock);
387 rdsdebug("rds_conn_path [%d] t 298 rdsdebug("rds_conn_path [%d] tc %p\n", i,
388 conn->c_path[i].cp_tr 299 conn->c_path[i].cp_transport_data);
389 } 300 }
390 spin_lock_irq(&rds_tcp_conn_lock); !! 301
391 for (i = 0; i < RDS_MPATH_WORKERS; i++ !! 302 return 0;
392 tc = conn->c_path[i].cp_transp !! 303 }
393 tc->t_tcp_node_detached = fals !! 304
394 list_add_tail(&tc->t_tcp_node, !! 305 static void rds_tcp_conn_free(void *arg)
395 } !! 306 {
396 spin_unlock_irq(&rds_tcp_conn_lock); !! 307 struct rds_tcp_connection *tc = arg;
397 fail: !! 308 unsigned long flags;
398 if (ret) { !! 309 rdsdebug("freeing tc %p\n", tc);
399 for (j = 0; j < i; j++) !! 310
400 rds_tcp_conn_free(conn !! 311 spin_lock_irqsave(&rds_tcp_conn_lock, flags);
401 } !! 312 list_del(&tc->t_tcp_node);
402 return ret; !! 313 spin_unlock_irqrestore(&rds_tcp_conn_lock, flags);
>> 314
>> 315 kmem_cache_free(rds_tcp_conn_slab, tc);
403 } 316 }
404 317
405 static bool list_has_conn(struct list_head *li 318 static bool list_has_conn(struct list_head *list, struct rds_connection *conn)
406 { 319 {
407 struct rds_tcp_connection *tc, *_tc; 320 struct rds_tcp_connection *tc, *_tc;
408 321
409 list_for_each_entry_safe(tc, _tc, list 322 list_for_each_entry_safe(tc, _tc, list, t_tcp_node) {
410 if (tc->t_cpath->cp_conn == co 323 if (tc->t_cpath->cp_conn == conn)
411 return true; 324 return true;
412 } 325 }
413 return false; 326 return false;
414 } 327 }
415 328
416 static void rds_tcp_set_unloading(void) <<
417 { <<
418 atomic_set(&rds_tcp_unloading, 1); <<
419 } <<
420 <<
421 static bool rds_tcp_is_unloading(struct rds_co <<
422 { <<
423 return atomic_read(&rds_tcp_unloading) <<
424 } <<
425 <<
426 static void rds_tcp_destroy_conns(void) 329 static void rds_tcp_destroy_conns(void)
427 { 330 {
428 struct rds_tcp_connection *tc, *_tc; 331 struct rds_tcp_connection *tc, *_tc;
429 LIST_HEAD(tmp_list); 332 LIST_HEAD(tmp_list);
430 333
431 /* avoid calling conn_destroy with irq 334 /* avoid calling conn_destroy with irqs off */
432 spin_lock_irq(&rds_tcp_conn_lock); 335 spin_lock_irq(&rds_tcp_conn_lock);
433 list_for_each_entry_safe(tc, _tc, &rds 336 list_for_each_entry_safe(tc, _tc, &rds_tcp_conn_list, t_tcp_node) {
434 if (!list_has_conn(&tmp_list, 337 if (!list_has_conn(&tmp_list, tc->t_cpath->cp_conn))
435 list_move_tail(&tc->t_ 338 list_move_tail(&tc->t_tcp_node, &tmp_list);
436 } 339 }
437 spin_unlock_irq(&rds_tcp_conn_lock); 340 spin_unlock_irq(&rds_tcp_conn_lock);
438 341
439 list_for_each_entry_safe(tc, _tc, &tmp 342 list_for_each_entry_safe(tc, _tc, &tmp_list, t_tcp_node)
440 rds_conn_destroy(tc->t_cpath-> 343 rds_conn_destroy(tc->t_cpath->cp_conn);
441 } 344 }
442 345
443 static void rds_tcp_exit(void); 346 static void rds_tcp_exit(void);
444 347
445 static u8 rds_tcp_get_tos_map(u8 tos) <<
446 { <<
447 /* all user tos mapped to default 0 fo <<
448 return 0; <<
449 } <<
450 <<
451 struct rds_transport rds_tcp_transport = { 348 struct rds_transport rds_tcp_transport = {
452 .laddr_check = rds_tcp_ladd 349 .laddr_check = rds_tcp_laddr_check,
453 .xmit_path_prepare = rds_tcp_xmit 350 .xmit_path_prepare = rds_tcp_xmit_path_prepare,
454 .xmit_path_complete = rds_tcp_xmit 351 .xmit_path_complete = rds_tcp_xmit_path_complete,
455 .xmit = rds_tcp_xmit 352 .xmit = rds_tcp_xmit,
456 .recv_path = rds_tcp_recv 353 .recv_path = rds_tcp_recv_path,
457 .conn_alloc = rds_tcp_conn 354 .conn_alloc = rds_tcp_conn_alloc,
458 .conn_free = rds_tcp_conn 355 .conn_free = rds_tcp_conn_free,
459 .conn_path_connect = rds_tcp_conn 356 .conn_path_connect = rds_tcp_conn_path_connect,
460 .conn_path_shutdown = rds_tcp_conn 357 .conn_path_shutdown = rds_tcp_conn_path_shutdown,
461 .inc_copy_to_user = rds_tcp_inc_ 358 .inc_copy_to_user = rds_tcp_inc_copy_to_user,
462 .inc_free = rds_tcp_inc_ 359 .inc_free = rds_tcp_inc_free,
463 .stats_info_copy = rds_tcp_stat 360 .stats_info_copy = rds_tcp_stats_info_copy,
464 .exit = rds_tcp_exit 361 .exit = rds_tcp_exit,
465 .get_tos_map = rds_tcp_get_ <<
466 .t_owner = THIS_MODULE, 362 .t_owner = THIS_MODULE,
467 .t_name = "tcp", 363 .t_name = "tcp",
468 .t_type = RDS_TRANS_TC 364 .t_type = RDS_TRANS_TCP,
469 .t_prefer_loopback = 1, 365 .t_prefer_loopback = 1,
470 .t_mp_capable = 1, 366 .t_mp_capable = 1,
471 .t_unloading = rds_tcp_is_u <<
472 }; 367 };
473 368
474 static unsigned int rds_tcp_netid; 369 static unsigned int rds_tcp_netid;
475 370
476 /* per-network namespace private data for this 371 /* per-network namespace private data for this module */
477 struct rds_tcp_net { 372 struct rds_tcp_net {
478 struct socket *rds_tcp_listen_sock; 373 struct socket *rds_tcp_listen_sock;
479 struct work_struct rds_tcp_accept_w; 374 struct work_struct rds_tcp_accept_w;
480 struct ctl_table_header *rds_tcp_sysct 375 struct ctl_table_header *rds_tcp_sysctl;
481 struct ctl_table *ctl_table; 376 struct ctl_table *ctl_table;
482 int sndbuf_size; 377 int sndbuf_size;
483 int rcvbuf_size; 378 int rcvbuf_size;
484 }; 379 };
485 380
486 /* All module specific customizations to the R 381 /* All module specific customizations to the RDS-TCP socket should be done in
487 * rds_tcp_tune() and applied after socket cre 382 * rds_tcp_tune() and applied after socket creation.
488 */ 383 */
489 bool rds_tcp_tune(struct socket *sock) !! 384 void rds_tcp_tune(struct socket *sock)
490 { 385 {
491 struct sock *sk = sock->sk; 386 struct sock *sk = sock->sk;
492 struct net *net = sock_net(sk); 387 struct net *net = sock_net(sk);
493 struct rds_tcp_net *rtn; !! 388 struct rds_tcp_net *rtn = net_generic(net, rds_tcp_netid);
494 389
495 tcp_sock_set_nodelay(sock->sk); !! 390 rds_tcp_nonagle(sock);
496 lock_sock(sk); 391 lock_sock(sk);
497 /* TCP timer functions might access ne <<
498 * a process which created this net na <<
499 */ <<
500 if (!sk->sk_net_refcnt) { <<
501 if (!maybe_get_net(net)) { <<
502 release_sock(sk); <<
503 return false; <<
504 } <<
505 /* Update ns_tracker to curren <<
506 __netns_tracker_free(net, &sk- <<
507 <<
508 sk->sk_net_refcnt = 1; <<
509 netns_tracker_alloc(net, &sk-> <<
510 sock_inuse_add(net, 1); <<
511 } <<
512 rtn = net_generic(net, rds_tcp_netid); <<
513 if (rtn->sndbuf_size > 0) { 392 if (rtn->sndbuf_size > 0) {
514 sk->sk_sndbuf = rtn->sndbuf_si 393 sk->sk_sndbuf = rtn->sndbuf_size;
515 sk->sk_userlocks |= SOCK_SNDBU 394 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
516 } 395 }
517 if (rtn->rcvbuf_size > 0) { 396 if (rtn->rcvbuf_size > 0) {
518 sk->sk_rcvbuf = rtn->rcvbuf_si !! 397 sk->sk_sndbuf = rtn->rcvbuf_size;
519 sk->sk_userlocks |= SOCK_RCVBU 398 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
520 } 399 }
521 release_sock(sk); 400 release_sock(sk);
522 return true; <<
523 } 401 }
524 402
525 static void rds_tcp_accept_worker(struct work_ 403 static void rds_tcp_accept_worker(struct work_struct *work)
526 { 404 {
527 struct rds_tcp_net *rtn = container_of 405 struct rds_tcp_net *rtn = container_of(work,
528 406 struct rds_tcp_net,
529 407 rds_tcp_accept_w);
530 408
531 while (rds_tcp_accept_one(rtn->rds_tcp 409 while (rds_tcp_accept_one(rtn->rds_tcp_listen_sock) == 0)
532 cond_resched(); 410 cond_resched();
533 } 411 }
534 412
535 void rds_tcp_accept_work(struct sock *sk) 413 void rds_tcp_accept_work(struct sock *sk)
536 { 414 {
537 struct net *net = sock_net(sk); 415 struct net *net = sock_net(sk);
538 struct rds_tcp_net *rtn = net_generic( 416 struct rds_tcp_net *rtn = net_generic(net, rds_tcp_netid);
539 417
540 queue_work(rds_wq, &rtn->rds_tcp_accep 418 queue_work(rds_wq, &rtn->rds_tcp_accept_w);
541 } 419 }
542 420
543 static __net_init int rds_tcp_init_net(struct 421 static __net_init int rds_tcp_init_net(struct net *net)
544 { 422 {
545 struct rds_tcp_net *rtn = net_generic( 423 struct rds_tcp_net *rtn = net_generic(net, rds_tcp_netid);
546 struct ctl_table *tbl; 424 struct ctl_table *tbl;
547 int err = 0; 425 int err = 0;
548 426
549 memset(rtn, 0, sizeof(*rtn)); 427 memset(rtn, 0, sizeof(*rtn));
550 428
551 /* {snd, rcv}buf_size default to 0, wh 429 /* {snd, rcv}buf_size default to 0, which implies we let the
552 * stack pick the value, and permit au 430 * stack pick the value, and permit auto-tuning of buffer size.
553 */ 431 */
554 if (net == &init_net) { 432 if (net == &init_net) {
555 tbl = rds_tcp_sysctl_table; 433 tbl = rds_tcp_sysctl_table;
556 } else { 434 } else {
557 tbl = kmemdup(rds_tcp_sysctl_t 435 tbl = kmemdup(rds_tcp_sysctl_table,
558 sizeof(rds_tcp_s 436 sizeof(rds_tcp_sysctl_table), GFP_KERNEL);
559 if (!tbl) { 437 if (!tbl) {
560 pr_warn("could not set !! 438 pr_warn("could not set allocate syctl table\n");
561 return -ENOMEM; 439 return -ENOMEM;
562 } 440 }
563 rtn->ctl_table = tbl; 441 rtn->ctl_table = tbl;
564 } 442 }
565 tbl[RDS_TCP_SNDBUF].data = &rtn->sndbu 443 tbl[RDS_TCP_SNDBUF].data = &rtn->sndbuf_size;
566 tbl[RDS_TCP_RCVBUF].data = &rtn->rcvbu 444 tbl[RDS_TCP_RCVBUF].data = &rtn->rcvbuf_size;
567 rtn->rds_tcp_sysctl = register_net_sys !! 445 rtn->rds_tcp_sysctl = register_net_sysctl(net, "net/rds/tcp", tbl);
568 <<
569 if (!rtn->rds_tcp_sysctl) { 446 if (!rtn->rds_tcp_sysctl) {
570 pr_warn("could not register sy 447 pr_warn("could not register sysctl\n");
571 err = -ENOMEM; 448 err = -ENOMEM;
572 goto fail; 449 goto fail;
573 } 450 }
574 !! 451 rtn->rds_tcp_listen_sock = rds_tcp_listen_init(net);
575 #if IS_ENABLED(CONFIG_IPV6) <<
576 rtn->rds_tcp_listen_sock = rds_tcp_lis <<
577 #else <<
578 rtn->rds_tcp_listen_sock = rds_tcp_lis <<
579 #endif <<
580 if (!rtn->rds_tcp_listen_sock) { 452 if (!rtn->rds_tcp_listen_sock) {
581 pr_warn("could not set up IPv6 !! 453 pr_warn("could not set up listen sock\n");
582 !! 454 unregister_net_sysctl_table(rtn->rds_tcp_sysctl);
583 #if IS_ENABLED(CONFIG_IPV6) !! 455 rtn->rds_tcp_sysctl = NULL;
584 /* Try IPv4 as some systems di !! 456 err = -EAFNOSUPPORT;
585 rtn->rds_tcp_listen_sock = rds !! 457 goto fail;
586 if (!rtn->rds_tcp_listen_sock) <<
587 #endif <<
588 unregister_net_sysctl_ <<
589 rtn->rds_tcp_sysctl = <<
590 err = -EAFNOSUPPORT; <<
591 goto fail; <<
592 #if IS_ENABLED(CONFIG_IPV6) <<
593 } <<
594 #endif <<
595 } 458 }
596 INIT_WORK(&rtn->rds_tcp_accept_w, rds_ 459 INIT_WORK(&rtn->rds_tcp_accept_w, rds_tcp_accept_worker);
597 return 0; 460 return 0;
598 461
599 fail: 462 fail:
600 if (net != &init_net) 463 if (net != &init_net)
601 kfree(tbl); 464 kfree(tbl);
602 return err; 465 return err;
603 } 466 }
604 467
>> 468 static void __net_exit rds_tcp_exit_net(struct net *net)
>> 469 {
>> 470 struct rds_tcp_net *rtn = net_generic(net, rds_tcp_netid);
>> 471
>> 472 if (rtn->rds_tcp_sysctl)
>> 473 unregister_net_sysctl_table(rtn->rds_tcp_sysctl);
>> 474
>> 475 if (net != &init_net && rtn->ctl_table)
>> 476 kfree(rtn->ctl_table);
>> 477
>> 478 /* If rds_tcp_exit_net() is called as a result of netns deletion,
>> 479 * the rds_tcp_kill_sock() device notifier would already have cleaned
>> 480 * up the listen socket, thus there is no work to do in this function.
>> 481 *
>> 482 * If rds_tcp_exit_net() is called as a result of module unload,
>> 483 * i.e., due to rds_tcp_exit() -> unregister_pernet_subsys(), then
>> 484 * we do need to clean up the listen socket here.
>> 485 */
>> 486 if (rtn->rds_tcp_listen_sock) {
>> 487 struct socket *lsock = rtn->rds_tcp_listen_sock;
>> 488
>> 489 rtn->rds_tcp_listen_sock = NULL;
>> 490 rds_tcp_listen_stop(lsock, &rtn->rds_tcp_accept_w);
>> 491 }
>> 492 }
>> 493
>> 494 static struct pernet_operations rds_tcp_net_ops = {
>> 495 .init = rds_tcp_init_net,
>> 496 .exit = rds_tcp_exit_net,
>> 497 .id = &rds_tcp_netid,
>> 498 .size = sizeof(struct rds_tcp_net),
>> 499 };
>> 500
>> 501 /* explicitly send a RST on each socket, thereby releasing any socket refcnts
>> 502 * that may otherwise hold up netns deletion.
>> 503 */
>> 504 static void rds_tcp_conn_paths_destroy(struct rds_connection *conn)
>> 505 {
>> 506 struct rds_conn_path *cp;
>> 507 struct rds_tcp_connection *tc;
>> 508 int i;
>> 509 struct sock *sk;
>> 510
>> 511 for (i = 0; i < RDS_MPATH_WORKERS; i++) {
>> 512 cp = &conn->c_path[i];
>> 513 tc = cp->cp_transport_data;
>> 514 if (!tc->t_sock)
>> 515 continue;
>> 516 sk = tc->t_sock->sk;
>> 517 sk->sk_prot->disconnect(sk, 0);
>> 518 tcp_done(sk);
>> 519 }
>> 520 }
>> 521
605 static void rds_tcp_kill_sock(struct net *net) 522 static void rds_tcp_kill_sock(struct net *net)
606 { 523 {
607 struct rds_tcp_connection *tc, *_tc; 524 struct rds_tcp_connection *tc, *_tc;
608 LIST_HEAD(tmp_list); 525 LIST_HEAD(tmp_list);
609 struct rds_tcp_net *rtn = net_generic( 526 struct rds_tcp_net *rtn = net_generic(net, rds_tcp_netid);
610 struct socket *lsock = rtn->rds_tcp_li 527 struct socket *lsock = rtn->rds_tcp_listen_sock;
611 528
612 rtn->rds_tcp_listen_sock = NULL; 529 rtn->rds_tcp_listen_sock = NULL;
613 rds_tcp_listen_stop(lsock, &rtn->rds_t 530 rds_tcp_listen_stop(lsock, &rtn->rds_tcp_accept_w);
614 spin_lock_irq(&rds_tcp_conn_lock); 531 spin_lock_irq(&rds_tcp_conn_lock);
615 list_for_each_entry_safe(tc, _tc, &rds 532 list_for_each_entry_safe(tc, _tc, &rds_tcp_conn_list, t_tcp_node) {
616 struct net *c_net = read_pnet( !! 533 struct net *c_net = tc->t_cpath->cp_conn->c_net;
617 534
618 if (net != c_net) !! 535 if (net != c_net || !tc->t_sock)
619 continue; 536 continue;
620 if (!list_has_conn(&tmp_list, !! 537 if (!list_has_conn(&tmp_list, tc->t_cpath->cp_conn))
621 list_move_tail(&tc->t_ 538 list_move_tail(&tc->t_tcp_node, &tmp_list);
622 } else { <<
623 list_del(&tc->t_tcp_no <<
624 tc->t_tcp_node_detache <<
625 } <<
626 } 539 }
627 spin_unlock_irq(&rds_tcp_conn_lock); 540 spin_unlock_irq(&rds_tcp_conn_lock);
628 list_for_each_entry_safe(tc, _tc, &tmp !! 541 list_for_each_entry_safe(tc, _tc, &tmp_list, t_tcp_node) {
>> 542 rds_tcp_conn_paths_destroy(tc->t_cpath->cp_conn);
629 rds_conn_destroy(tc->t_cpath-> 543 rds_conn_destroy(tc->t_cpath->cp_conn);
>> 544 }
630 } 545 }
631 546
632 static void __net_exit rds_tcp_exit_net(struct <<
633 { <<
634 struct rds_tcp_net *rtn = net_generic( <<
635 <<
636 rds_tcp_kill_sock(net); <<
637 <<
638 if (rtn->rds_tcp_sysctl) <<
639 unregister_net_sysctl_table(rt <<
640 <<
641 if (net != &init_net) <<
642 kfree(rtn->ctl_table); <<
643 } <<
644 <<
645 static struct pernet_operations rds_tcp_net_op <<
646 .init = rds_tcp_init_net, <<
647 .exit = rds_tcp_exit_net, <<
648 .id = &rds_tcp_netid, <<
649 .size = sizeof(struct rds_tcp_net), <<
650 }; <<
651 <<
652 void *rds_tcp_listen_sock_def_readable(struct 547 void *rds_tcp_listen_sock_def_readable(struct net *net)
653 { 548 {
654 struct rds_tcp_net *rtn = net_generic( 549 struct rds_tcp_net *rtn = net_generic(net, rds_tcp_netid);
655 struct socket *lsock = rtn->rds_tcp_li 550 struct socket *lsock = rtn->rds_tcp_listen_sock;
656 551
657 if (!lsock) 552 if (!lsock)
658 return NULL; 553 return NULL;
659 554
660 return lsock->sk->sk_user_data; 555 return lsock->sk->sk_user_data;
661 } 556 }
662 557
>> 558 static int rds_tcp_dev_event(struct notifier_block *this,
>> 559 unsigned long event, void *ptr)
>> 560 {
>> 561 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
>> 562
>> 563 /* rds-tcp registers as a pernet subys, so the ->exit will only
>> 564 * get invoked after network acitivity has quiesced. We need to
>> 565 * clean up all sockets to quiesce network activity, and use
>> 566 * the unregistration of the per-net loopback device as a trigger
>> 567 * to start that cleanup.
>> 568 */
>> 569 if (event == NETDEV_UNREGISTER_FINAL &&
>> 570 dev->ifindex == LOOPBACK_IFINDEX)
>> 571 rds_tcp_kill_sock(dev_net(dev));
>> 572
>> 573 return NOTIFY_DONE;
>> 574 }
>> 575
>> 576 static struct notifier_block rds_tcp_dev_notifier = {
>> 577 .notifier_call = rds_tcp_dev_event,
>> 578 .priority = -10, /* must be called after other network notifiers */
>> 579 };
>> 580
663 /* when sysctl is used to modify some kernel s 581 /* when sysctl is used to modify some kernel socket parameters,this
664 * function resets the RDS connections in tha 582 * function resets the RDS connections in that netns so that we can
665 * restart with new parameters. The assumptio 583 * restart with new parameters. The assumption is that such reset
666 * events are few and far-between. 584 * events are few and far-between.
667 */ 585 */
668 static void rds_tcp_sysctl_reset(struct net *n 586 static void rds_tcp_sysctl_reset(struct net *net)
669 { 587 {
670 struct rds_tcp_connection *tc, *_tc; 588 struct rds_tcp_connection *tc, *_tc;
671 589
672 spin_lock_irq(&rds_tcp_conn_lock); 590 spin_lock_irq(&rds_tcp_conn_lock);
673 list_for_each_entry_safe(tc, _tc, &rds 591 list_for_each_entry_safe(tc, _tc, &rds_tcp_conn_list, t_tcp_node) {
674 struct net *c_net = read_pnet( !! 592 struct net *c_net = tc->t_cpath->cp_conn->c_net;
675 593
676 if (net != c_net || !tc->t_soc 594 if (net != c_net || !tc->t_sock)
677 continue; 595 continue;
678 596
679 /* reconnect with new paramete 597 /* reconnect with new parameters */
680 rds_conn_path_drop(tc->t_cpath !! 598 rds_conn_path_drop(tc->t_cpath);
681 } 599 }
682 spin_unlock_irq(&rds_tcp_conn_lock); 600 spin_unlock_irq(&rds_tcp_conn_lock);
683 } 601 }
684 602
685 static int rds_tcp_skbuf_handler(const struct !! 603 static int rds_tcp_skbuf_handler(struct ctl_table *ctl, int write,
686 void *buffer, !! 604 void __user *buffer, size_t *lenp,
>> 605 loff_t *fpos)
687 { 606 {
688 struct net *net = current->nsproxy->ne 607 struct net *net = current->nsproxy->net_ns;
689 int err; 608 int err;
690 609
691 err = proc_dointvec_minmax(ctl, write, 610 err = proc_dointvec_minmax(ctl, write, buffer, lenp, fpos);
692 if (err < 0) { 611 if (err < 0) {
693 pr_warn("Invalid input. Must b 612 pr_warn("Invalid input. Must be >= %d\n",
694 *(int *)(ctl->extra1)) 613 *(int *)(ctl->extra1));
695 return err; 614 return err;
696 } 615 }
697 if (write) 616 if (write)
698 rds_tcp_sysctl_reset(net); 617 rds_tcp_sysctl_reset(net);
699 return 0; 618 return 0;
700 } 619 }
701 620
702 static void rds_tcp_exit(void) 621 static void rds_tcp_exit(void)
703 { 622 {
704 rds_tcp_set_unloading(); <<
705 synchronize_rcu(); <<
706 rds_info_deregister_func(RDS_INFO_TCP_ 623 rds_info_deregister_func(RDS_INFO_TCP_SOCKETS, rds_tcp_tc_info);
707 #if IS_ENABLED(CONFIG_IPV6) !! 624 unregister_pernet_subsys(&rds_tcp_net_ops);
708 rds_info_deregister_func(RDS6_INFO_TCP !! 625 if (unregister_netdevice_notifier(&rds_tcp_dev_notifier))
709 #endif !! 626 pr_warn("could not unregister rds_tcp_dev_notifier\n");
710 unregister_pernet_device(&rds_tcp_net_ <<
711 rds_tcp_destroy_conns(); 627 rds_tcp_destroy_conns();
712 rds_trans_unregister(&rds_tcp_transpor 628 rds_trans_unregister(&rds_tcp_transport);
713 rds_tcp_recv_exit(); 629 rds_tcp_recv_exit();
714 kmem_cache_destroy(rds_tcp_conn_slab); 630 kmem_cache_destroy(rds_tcp_conn_slab);
715 } 631 }
716 module_exit(rds_tcp_exit); 632 module_exit(rds_tcp_exit);
717 633
718 static int __init rds_tcp_init(void) !! 634 static int rds_tcp_init(void)
719 { 635 {
720 int ret; 636 int ret;
721 637
722 rds_tcp_conn_slab = KMEM_CACHE(rds_tcp !! 638 rds_tcp_conn_slab = kmem_cache_create("rds_tcp_connection",
>> 639 sizeof(struct rds_tcp_connection),
>> 640 0, 0, NULL);
723 if (!rds_tcp_conn_slab) { 641 if (!rds_tcp_conn_slab) {
724 ret = -ENOMEM; 642 ret = -ENOMEM;
725 goto out; 643 goto out;
726 } 644 }
727 645
728 ret = rds_tcp_recv_init(); 646 ret = rds_tcp_recv_init();
729 if (ret) 647 if (ret)
730 goto out_slab; 648 goto out_slab;
731 649
732 ret = register_pernet_device(&rds_tcp_ !! 650 ret = register_pernet_subsys(&rds_tcp_net_ops);
733 if (ret) 651 if (ret)
734 goto out_recv; 652 goto out_recv;
735 653
>> 654 ret = register_netdevice_notifier(&rds_tcp_dev_notifier);
>> 655 if (ret) {
>> 656 pr_warn("could not register rds_tcp_dev_notifier\n");
>> 657 goto out_pernet;
>> 658 }
>> 659
736 rds_trans_register(&rds_tcp_transport) 660 rds_trans_register(&rds_tcp_transport);
737 661
738 rds_info_register_func(RDS_INFO_TCP_SO 662 rds_info_register_func(RDS_INFO_TCP_SOCKETS, rds_tcp_tc_info);
739 #if IS_ENABLED(CONFIG_IPV6) <<
740 rds_info_register_func(RDS6_INFO_TCP_S <<
741 #endif <<
742 663
743 goto out; 664 goto out;
>> 665
>> 666 out_pernet:
>> 667 unregister_pernet_subsys(&rds_tcp_net_ops);
744 out_recv: 668 out_recv:
745 rds_tcp_recv_exit(); 669 rds_tcp_recv_exit();
746 out_slab: 670 out_slab:
747 kmem_cache_destroy(rds_tcp_conn_slab); 671 kmem_cache_destroy(rds_tcp_conn_slab);
748 out: 672 out:
749 return ret; 673 return ret;
750 } 674 }
751 module_init(rds_tcp_init); 675 module_init(rds_tcp_init);
752 676
753 MODULE_AUTHOR("Oracle Corporation <rds-devel@o 677 MODULE_AUTHOR("Oracle Corporation <rds-devel@oss.oracle.com>");
754 MODULE_DESCRIPTION("RDS: TCP transport"); 678 MODULE_DESCRIPTION("RDS: TCP transport");
755 MODULE_LICENSE("Dual BSD/GPL"); 679 MODULE_LICENSE("Dual BSD/GPL");
>> 680
756 681
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.