]>
Commit | Line | Data |
---|---|---|
2522fe45 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
fdda387f PC |
2 | /****************************************************************************** |
3 | ******************************************************************************* | |
4 | ** | |
5 | ** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
5e9ccc37 | 6 | ** Copyright (C) 2004-2009 Red Hat, Inc. All rights reserved. |
fdda387f | 7 | ** |
fdda387f PC |
8 | ** |
9 | ******************************************************************************* | |
10 | ******************************************************************************/ | |
11 | ||
12 | /* | |
13 | * lowcomms.c | |
14 | * | |
15 | * This is the "low-level" comms layer. | |
16 | * | |
17 | * It is responsible for sending/receiving messages | |
18 | * from other nodes in the cluster. | |
19 | * | |
20 | * Cluster nodes are referred to by their nodeids. nodeids are | |
21 | * simply 32 bit numbers to the locking module - if they need to | |
2cf12c0b | 22 | * be expanded for the cluster infrastructure then that is its |
fdda387f PC |
23 | * responsibility. It is this layer's |
24 | * responsibility to resolve these into IP address or | |
25 | * whatever it needs for inter-node communication. | |
26 | * | |
27 | * The comms level is two kernel threads that deal mainly with | |
28 | * the receiving of messages from other nodes and passing them | |
29 | * up to the mid-level comms layer (which understands the | |
30 | * message format) for execution by the locking core, and | |
31 | * a send thread which does all the setting up of connections | |
32 | * to remote nodes and the sending of data. Threads are not allowed | |
33 | * to send their own data because it may cause them to wait in times | |
34 | * of high load. Also, this way, the sending thread can collect together | |
35 | * messages bound for one node and send them in one block. | |
36 | * | |
2cf12c0b | 37 | * lowcomms will choose to use either TCP or SCTP as its transport layer |
6ed7257b | 38 | * depending on the configuration variable 'protocol'. This should be set |
2cf12c0b | 39 | * to 0 (default) for TCP or 1 for SCTP. It should be configured using a |
6ed7257b PC |
40 | * cluster-wide mechanism as it must be the same on all nodes of the cluster |
41 | * for the DLM to function. | |
fdda387f PC |
42 | * |
43 | */ | |
44 | ||
fdda387f PC |
45 | #include <asm/ioctls.h> |
46 | #include <net/sock.h> | |
47 | #include <net/tcp.h> | |
48 | #include <linux/pagemap.h> | |
6ed7257b | 49 | #include <linux/file.h> |
7a936ce7 | 50 | #include <linux/mutex.h> |
6ed7257b | 51 | #include <linux/sctp.h> |
5a0e3ad6 | 52 | #include <linux/slab.h> |
2f2d76cc | 53 | #include <net/sctp/sctp.h> |
44ad532b | 54 | #include <net/ipv6.h> |
fdda387f PC |
55 | |
56 | #include "dlm_internal.h" | |
57 | #include "lowcomms.h" | |
58 | #include "midcomms.h" | |
59 | #include "config.h" | |
60 | ||
6ed7257b | 61 | #define NEEDED_RMEM (4*1024*1024) |
5e9ccc37 | 62 | #define CONN_HASH_SIZE 32 |
6ed7257b | 63 | |
f92c8dd7 BP |
64 | /* Number of messages to send before rescheduling */ |
65 | #define MAX_SEND_MSG_COUNT 25 | |
055923bf | 66 | #define DLM_SHUTDOWN_WAIT_TIMEOUT msecs_to_jiffies(10000) |
f92c8dd7 | 67 | |
fdda387f PC |
68 | struct connection { |
69 | struct socket *sock; /* NULL if not connected */ | |
70 | uint32_t nodeid; /* So we know who we are in the list */ | |
f1f1c1cc | 71 | struct mutex sock_mutex; |
6ed7257b | 72 | unsigned long flags; |
fdda387f | 73 | #define CF_READ_PENDING 1 |
8a4abb08 | 74 | #define CF_WRITE_PENDING 2 |
6ed7257b PC |
75 | #define CF_INIT_PENDING 4 |
76 | #define CF_IS_OTHERCON 5 | |
063c4c99 | 77 | #define CF_CLOSE 6 |
b36930dd | 78 | #define CF_APP_LIMITED 7 |
b2a66629 | 79 | #define CF_CLOSING 8 |
055923bf | 80 | #define CF_SHUTDOWN 9 |
ac33d071 | 81 | struct list_head writequeue; /* List of outgoing writequeue_entries */ |
fdda387f PC |
82 | spinlock_t writequeue_lock; |
83 | int (*rx_action) (struct connection *); /* What to do when active */ | |
6ed7257b | 84 | void (*connect_action) (struct connection *); /* What to do to connect */ |
055923bf | 85 | void (*shutdown_action)(struct connection *con); /* What to do to shutdown */ |
fdda387f | 86 | int retries; |
fdda387f | 87 | #define MAX_CONNECT_RETRIES 3 |
5e9ccc37 | 88 | struct hlist_node list; |
fdda387f | 89 | struct connection *othercon; |
1d6e8131 PC |
90 | struct work_struct rwork; /* Receive workqueue */ |
91 | struct work_struct swork; /* Send workqueue */ | |
055923bf | 92 | wait_queue_head_t shutdown_wait; /* wait for graceful shutdown */ |
4798cbbf AA |
93 | unsigned char *rx_buf; |
94 | int rx_buflen; | |
95 | int rx_leftover; | |
a47666eb | 96 | struct rcu_head rcu; |
fdda387f PC |
97 | }; |
98 | #define sock2con(x) ((struct connection *)(x)->sk_user_data) | |
99 | ||
100 | /* An entry waiting to be sent */ | |
101 | struct writequeue_entry { | |
102 | struct list_head list; | |
103 | struct page *page; | |
104 | int offset; | |
105 | int len; | |
106 | int end; | |
107 | int users; | |
108 | struct connection *con; | |
109 | }; | |
110 | ||
36b71a8b DT |
111 | struct dlm_node_addr { |
112 | struct list_head list; | |
113 | int nodeid; | |
114 | int addr_count; | |
98e1b60e | 115 | int curr_addr_index; |
36b71a8b DT |
116 | struct sockaddr_storage *addr[DLM_MAX_ADDR_COUNT]; |
117 | }; | |
118 | ||
cc661fc9 BP |
119 | static struct listen_sock_callbacks { |
120 | void (*sk_error_report)(struct sock *); | |
121 | void (*sk_data_ready)(struct sock *); | |
122 | void (*sk_state_change)(struct sock *); | |
123 | void (*sk_write_space)(struct sock *); | |
124 | } listen_sock; | |
125 | ||
36b71a8b DT |
126 | static LIST_HEAD(dlm_node_addrs); |
127 | static DEFINE_SPINLOCK(dlm_node_addrs_spin); | |
128 | ||
6ed7257b PC |
129 | static struct sockaddr_storage *dlm_local_addr[DLM_MAX_ADDR_COUNT]; |
130 | static int dlm_local_count; | |
513ef596 | 131 | static int dlm_allow_conn; |
fdda387f | 132 | |
1d6e8131 PC |
133 | /* Work queues */ |
134 | static struct workqueue_struct *recv_workqueue; | |
135 | static struct workqueue_struct *send_workqueue; | |
fdda387f | 136 | |
5e9ccc37 | 137 | static struct hlist_head connection_hash[CONN_HASH_SIZE]; |
a47666eb AA |
138 | static DEFINE_SPINLOCK(connections_lock); |
139 | DEFINE_STATIC_SRCU(connections_srcu); | |
fdda387f | 140 | |
1d6e8131 PC |
141 | static void process_recv_sockets(struct work_struct *work); |
142 | static void process_send_sockets(struct work_struct *work); | |
fdda387f | 143 | |
5e9ccc37 CC |
144 | |
145 | /* This is deliberately very simple because most clusters have simple | |
146 | sequential nodeids, so we should be able to go straight to a connection | |
147 | struct in the array */ | |
148 | static inline int nodeid_hash(int nodeid) | |
149 | { | |
150 | return nodeid & (CONN_HASH_SIZE-1); | |
151 | } | |
152 | ||
153 | static struct connection *__find_con(int nodeid) | |
154 | { | |
a47666eb | 155 | int r, idx; |
5e9ccc37 CC |
156 | struct connection *con; |
157 | ||
158 | r = nodeid_hash(nodeid); | |
159 | ||
a47666eb AA |
160 | idx = srcu_read_lock(&connections_srcu); |
161 | hlist_for_each_entry_rcu(con, &connection_hash[r], list) { | |
162 | if (con->nodeid == nodeid) { | |
163 | srcu_read_unlock(&connections_srcu, idx); | |
5e9ccc37 | 164 | return con; |
a47666eb | 165 | } |
5e9ccc37 | 166 | } |
a47666eb AA |
167 | srcu_read_unlock(&connections_srcu, idx); |
168 | ||
5e9ccc37 CC |
169 | return NULL; |
170 | } | |
171 | ||
6ed7257b PC |
172 | /* |
173 | * If 'allocation' is zero then we don't attempt to create a new | |
174 | * connection structure for this node. | |
175 | */ | |
a47666eb | 176 | static struct connection *nodeid2con(int nodeid, gfp_t alloc) |
fdda387f PC |
177 | { |
178 | struct connection *con = NULL; | |
6ed7257b | 179 | int r; |
fdda387f | 180 | |
5e9ccc37 | 181 | con = __find_con(nodeid); |
6ed7257b PC |
182 | if (con || !alloc) |
183 | return con; | |
fdda387f | 184 | |
a47666eb | 185 | con = kzalloc(sizeof(*con), alloc); |
6ed7257b PC |
186 | if (!con) |
187 | return NULL; | |
fdda387f | 188 | |
4798cbbf AA |
189 | con->rx_buflen = dlm_config.ci_buffer_size; |
190 | con->rx_buf = kmalloc(con->rx_buflen, GFP_NOFS); | |
191 | if (!con->rx_buf) { | |
192 | kfree(con); | |
193 | return NULL; | |
194 | } | |
195 | ||
6ed7257b PC |
196 | con->nodeid = nodeid; |
197 | mutex_init(&con->sock_mutex); | |
198 | INIT_LIST_HEAD(&con->writequeue); | |
199 | spin_lock_init(&con->writequeue_lock); | |
200 | INIT_WORK(&con->swork, process_send_sockets); | |
201 | INIT_WORK(&con->rwork, process_recv_sockets); | |
055923bf | 202 | init_waitqueue_head(&con->shutdown_wait); |
fdda387f | 203 | |
6ed7257b PC |
204 | /* Setup action pointers for child sockets */ |
205 | if (con->nodeid) { | |
5e9ccc37 | 206 | struct connection *zerocon = __find_con(0); |
fdda387f | 207 | |
6ed7257b PC |
208 | con->connect_action = zerocon->connect_action; |
209 | if (!con->rx_action) | |
210 | con->rx_action = zerocon->rx_action; | |
fdda387f PC |
211 | } |
212 | ||
a47666eb AA |
213 | r = nodeid_hash(nodeid); |
214 | ||
215 | spin_lock(&connections_lock); | |
216 | hlist_add_head_rcu(&con->list, &connection_hash[r]); | |
217 | spin_unlock(&connections_lock); | |
218 | ||
6ed7257b PC |
219 | return con; |
220 | } | |
221 | ||
5e9ccc37 CC |
222 | /* Loop round all connections */ |
223 | static void foreach_conn(void (*conn_func)(struct connection *c)) | |
224 | { | |
a47666eb | 225 | int i, idx; |
5e9ccc37 CC |
226 | struct connection *con; |
227 | ||
a47666eb | 228 | idx = srcu_read_lock(&connections_srcu); |
5e9ccc37 | 229 | for (i = 0; i < CONN_HASH_SIZE; i++) { |
a47666eb | 230 | hlist_for_each_entry_rcu(con, &connection_hash[i], list) |
5e9ccc37 | 231 | conn_func(con); |
5e9ccc37 | 232 | } |
a47666eb | 233 | srcu_read_unlock(&connections_srcu, idx); |
fdda387f PC |
234 | } |
235 | ||
36b71a8b DT |
236 | static struct dlm_node_addr *find_node_addr(int nodeid) |
237 | { | |
238 | struct dlm_node_addr *na; | |
239 | ||
240 | list_for_each_entry(na, &dlm_node_addrs, list) { | |
241 | if (na->nodeid == nodeid) | |
242 | return na; | |
243 | } | |
244 | return NULL; | |
245 | } | |
246 | ||
247 | static int addr_compare(struct sockaddr_storage *x, struct sockaddr_storage *y) | |
6ed7257b | 248 | { |
36b71a8b DT |
249 | switch (x->ss_family) { |
250 | case AF_INET: { | |
251 | struct sockaddr_in *sinx = (struct sockaddr_in *)x; | |
252 | struct sockaddr_in *siny = (struct sockaddr_in *)y; | |
253 | if (sinx->sin_addr.s_addr != siny->sin_addr.s_addr) | |
254 | return 0; | |
255 | if (sinx->sin_port != siny->sin_port) | |
256 | return 0; | |
257 | break; | |
258 | } | |
259 | case AF_INET6: { | |
260 | struct sockaddr_in6 *sinx = (struct sockaddr_in6 *)x; | |
261 | struct sockaddr_in6 *siny = (struct sockaddr_in6 *)y; | |
262 | if (!ipv6_addr_equal(&sinx->sin6_addr, &siny->sin6_addr)) | |
263 | return 0; | |
264 | if (sinx->sin6_port != siny->sin6_port) | |
265 | return 0; | |
266 | break; | |
267 | } | |
268 | default: | |
269 | return 0; | |
270 | } | |
271 | return 1; | |
272 | } | |
273 | ||
274 | static int nodeid_to_addr(int nodeid, struct sockaddr_storage *sas_out, | |
98e1b60e | 275 | struct sockaddr *sa_out, bool try_new_addr) |
36b71a8b DT |
276 | { |
277 | struct sockaddr_storage sas; | |
278 | struct dlm_node_addr *na; | |
6ed7257b PC |
279 | |
280 | if (!dlm_local_count) | |
281 | return -1; | |
282 | ||
36b71a8b DT |
283 | spin_lock(&dlm_node_addrs_spin); |
284 | na = find_node_addr(nodeid); | |
98e1b60e | 285 | if (na && na->addr_count) { |
ee44b4bc MRL |
286 | memcpy(&sas, na->addr[na->curr_addr_index], |
287 | sizeof(struct sockaddr_storage)); | |
288 | ||
98e1b60e MC |
289 | if (try_new_addr) { |
290 | na->curr_addr_index++; | |
291 | if (na->curr_addr_index == na->addr_count) | |
292 | na->curr_addr_index = 0; | |
293 | } | |
98e1b60e | 294 | } |
36b71a8b DT |
295 | spin_unlock(&dlm_node_addrs_spin); |
296 | ||
297 | if (!na) | |
298 | return -EEXIST; | |
299 | ||
300 | if (!na->addr_count) | |
301 | return -ENOENT; | |
302 | ||
303 | if (sas_out) | |
304 | memcpy(sas_out, &sas, sizeof(struct sockaddr_storage)); | |
305 | ||
306 | if (!sa_out) | |
307 | return 0; | |
6ed7257b PC |
308 | |
309 | if (dlm_local_addr[0]->ss_family == AF_INET) { | |
36b71a8b DT |
310 | struct sockaddr_in *in4 = (struct sockaddr_in *) &sas; |
311 | struct sockaddr_in *ret4 = (struct sockaddr_in *) sa_out; | |
6ed7257b PC |
312 | ret4->sin_addr.s_addr = in4->sin_addr.s_addr; |
313 | } else { | |
36b71a8b DT |
314 | struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) &sas; |
315 | struct sockaddr_in6 *ret6 = (struct sockaddr_in6 *) sa_out; | |
4e3fd7a0 | 316 | ret6->sin6_addr = in6->sin6_addr; |
6ed7257b PC |
317 | } |
318 | ||
319 | return 0; | |
320 | } | |
321 | ||
36b71a8b DT |
322 | static int addr_to_nodeid(struct sockaddr_storage *addr, int *nodeid) |
323 | { | |
324 | struct dlm_node_addr *na; | |
325 | int rv = -EEXIST; | |
98e1b60e | 326 | int addr_i; |
36b71a8b DT |
327 | |
328 | spin_lock(&dlm_node_addrs_spin); | |
329 | list_for_each_entry(na, &dlm_node_addrs, list) { | |
330 | if (!na->addr_count) | |
331 | continue; | |
332 | ||
98e1b60e MC |
333 | for (addr_i = 0; addr_i < na->addr_count; addr_i++) { |
334 | if (addr_compare(na->addr[addr_i], addr)) { | |
335 | *nodeid = na->nodeid; | |
336 | rv = 0; | |
337 | goto unlock; | |
338 | } | |
339 | } | |
36b71a8b | 340 | } |
98e1b60e | 341 | unlock: |
36b71a8b DT |
342 | spin_unlock(&dlm_node_addrs_spin); |
343 | return rv; | |
344 | } | |
345 | ||
346 | int dlm_lowcomms_addr(int nodeid, struct sockaddr_storage *addr, int len) | |
347 | { | |
348 | struct sockaddr_storage *new_addr; | |
349 | struct dlm_node_addr *new_node, *na; | |
350 | ||
351 | new_node = kzalloc(sizeof(struct dlm_node_addr), GFP_NOFS); | |
352 | if (!new_node) | |
353 | return -ENOMEM; | |
354 | ||
355 | new_addr = kzalloc(sizeof(struct sockaddr_storage), GFP_NOFS); | |
356 | if (!new_addr) { | |
357 | kfree(new_node); | |
358 | return -ENOMEM; | |
359 | } | |
360 | ||
361 | memcpy(new_addr, addr, len); | |
362 | ||
363 | spin_lock(&dlm_node_addrs_spin); | |
364 | na = find_node_addr(nodeid); | |
365 | if (!na) { | |
366 | new_node->nodeid = nodeid; | |
367 | new_node->addr[0] = new_addr; | |
368 | new_node->addr_count = 1; | |
369 | list_add(&new_node->list, &dlm_node_addrs); | |
370 | spin_unlock(&dlm_node_addrs_spin); | |
371 | return 0; | |
372 | } | |
373 | ||
374 | if (na->addr_count >= DLM_MAX_ADDR_COUNT) { | |
375 | spin_unlock(&dlm_node_addrs_spin); | |
376 | kfree(new_addr); | |
377 | kfree(new_node); | |
378 | return -ENOSPC; | |
379 | } | |
380 | ||
381 | na->addr[na->addr_count++] = new_addr; | |
382 | spin_unlock(&dlm_node_addrs_spin); | |
383 | kfree(new_node); | |
384 | return 0; | |
385 | } | |
386 | ||
fdda387f | 387 | /* Data available on socket or listen socket received a connect */ |
676d2369 | 388 | static void lowcomms_data_ready(struct sock *sk) |
fdda387f | 389 | { |
93eaadeb | 390 | struct connection *con; |
391 | ||
392 | read_lock_bh(&sk->sk_callback_lock); | |
393 | con = sock2con(sk); | |
afb853fb | 394 | if (con && !test_and_set_bit(CF_READ_PENDING, &con->flags)) |
1d6e8131 | 395 | queue_work(recv_workqueue, &con->rwork); |
93eaadeb | 396 | read_unlock_bh(&sk->sk_callback_lock); |
fdda387f PC |
397 | } |
398 | ||
399 | static void lowcomms_write_space(struct sock *sk) | |
400 | { | |
93eaadeb | 401 | struct connection *con; |
fdda387f | 402 | |
93eaadeb | 403 | read_lock_bh(&sk->sk_callback_lock); |
404 | con = sock2con(sk); | |
b36930dd | 405 | if (!con) |
93eaadeb | 406 | goto out; |
b36930dd DM |
407 | |
408 | clear_bit(SOCK_NOSPACE, &con->sock->flags); | |
409 | ||
410 | if (test_and_clear_bit(CF_APP_LIMITED, &con->flags)) { | |
411 | con->sock->sk->sk_write_pending--; | |
9cd3e072 | 412 | clear_bit(SOCKWQ_ASYNC_NOSPACE, &con->sock->flags); |
b36930dd DM |
413 | } |
414 | ||
01da24d3 | 415 | queue_work(send_workqueue, &con->swork); |
93eaadeb | 416 | out: |
417 | read_unlock_bh(&sk->sk_callback_lock); | |
fdda387f PC |
418 | } |
419 | ||
420 | static inline void lowcomms_connect_sock(struct connection *con) | |
421 | { | |
063c4c99 LMB |
422 | if (test_bit(CF_CLOSE, &con->flags)) |
423 | return; | |
61d9102b BP |
424 | queue_work(send_workqueue, &con->swork); |
425 | cond_resched(); | |
fdda387f PC |
426 | } |
427 | ||
428 | static void lowcomms_state_change(struct sock *sk) | |
429 | { | |
ee44b4bc MRL |
430 | /* SCTP layer is not calling sk_data_ready when the connection |
431 | * is done, so we catch the signal through here. Also, it | |
432 | * doesn't switch socket state when entering shutdown, so we | |
433 | * skip the write in that case. | |
434 | */ | |
435 | if (sk->sk_shutdown) { | |
436 | if (sk->sk_shutdown == RCV_SHUTDOWN) | |
437 | lowcomms_data_ready(sk); | |
438 | } else if (sk->sk_state == TCP_ESTABLISHED) { | |
fdda387f | 439 | lowcomms_write_space(sk); |
ee44b4bc | 440 | } |
fdda387f PC |
441 | } |
442 | ||
391fbdc5 CC |
443 | int dlm_lowcomms_connect_node(int nodeid) |
444 | { | |
445 | struct connection *con; | |
446 | ||
447 | if (nodeid == dlm_our_nodeid()) | |
448 | return 0; | |
449 | ||
450 | con = nodeid2con(nodeid, GFP_NOFS); | |
451 | if (!con) | |
452 | return -ENOMEM; | |
453 | lowcomms_connect_sock(con); | |
454 | return 0; | |
455 | } | |
456 | ||
b3a5bbfd BP |
457 | static void lowcomms_error_report(struct sock *sk) |
458 | { | |
b81171cb | 459 | struct connection *con; |
b3a5bbfd | 460 | struct sockaddr_storage saddr; |
b81171cb | 461 | void (*orig_report)(struct sock *) = NULL; |
b3a5bbfd | 462 | |
b81171cb BP |
463 | read_lock_bh(&sk->sk_callback_lock); |
464 | con = sock2con(sk); | |
465 | if (con == NULL) | |
466 | goto out; | |
467 | ||
cc661fc9 | 468 | orig_report = listen_sock.sk_error_report; |
1a31833d | 469 | if (con->sock == NULL || |
9b2c45d4 | 470 | kernel_getpeername(con->sock, (struct sockaddr *)&saddr) < 0) { |
b3a5bbfd BP |
471 | printk_ratelimited(KERN_ERR "dlm: node %d: socket error " |
472 | "sending to node %d, port %d, " | |
473 | "sk_err=%d/%d\n", dlm_our_nodeid(), | |
474 | con->nodeid, dlm_config.ci_tcp_port, | |
475 | sk->sk_err, sk->sk_err_soft); | |
b3a5bbfd BP |
476 | } else if (saddr.ss_family == AF_INET) { |
477 | struct sockaddr_in *sin4 = (struct sockaddr_in *)&saddr; | |
478 | ||
479 | printk_ratelimited(KERN_ERR "dlm: node %d: socket error " | |
480 | "sending to node %d at %pI4, port %d, " | |
481 | "sk_err=%d/%d\n", dlm_our_nodeid(), | |
482 | con->nodeid, &sin4->sin_addr.s_addr, | |
483 | dlm_config.ci_tcp_port, sk->sk_err, | |
484 | sk->sk_err_soft); | |
485 | } else { | |
486 | struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&saddr; | |
487 | ||
488 | printk_ratelimited(KERN_ERR "dlm: node %d: socket error " | |
489 | "sending to node %d at %u.%u.%u.%u, " | |
490 | "port %d, sk_err=%d/%d\n", dlm_our_nodeid(), | |
491 | con->nodeid, sin6->sin6_addr.s6_addr32[0], | |
492 | sin6->sin6_addr.s6_addr32[1], | |
493 | sin6->sin6_addr.s6_addr32[2], | |
494 | sin6->sin6_addr.s6_addr32[3], | |
495 | dlm_config.ci_tcp_port, sk->sk_err, | |
496 | sk->sk_err_soft); | |
497 | } | |
b81171cb BP |
498 | out: |
499 | read_unlock_bh(&sk->sk_callback_lock); | |
500 | if (orig_report) | |
501 | orig_report(sk); | |
502 | } | |
503 | ||
504 | /* Note: sk_callback_lock must be locked before calling this function. */ | |
cc661fc9 | 505 | static void save_listen_callbacks(struct socket *sock) |
b81171cb | 506 | { |
cc661fc9 BP |
507 | struct sock *sk = sock->sk; |
508 | ||
509 | listen_sock.sk_data_ready = sk->sk_data_ready; | |
510 | listen_sock.sk_state_change = sk->sk_state_change; | |
511 | listen_sock.sk_write_space = sk->sk_write_space; | |
512 | listen_sock.sk_error_report = sk->sk_error_report; | |
b81171cb BP |
513 | } |
514 | ||
cc661fc9 | 515 | static void restore_callbacks(struct socket *sock) |
b81171cb | 516 | { |
cc661fc9 BP |
517 | struct sock *sk = sock->sk; |
518 | ||
b81171cb | 519 | write_lock_bh(&sk->sk_callback_lock); |
b81171cb | 520 | sk->sk_user_data = NULL; |
cc661fc9 BP |
521 | sk->sk_data_ready = listen_sock.sk_data_ready; |
522 | sk->sk_state_change = listen_sock.sk_state_change; | |
523 | sk->sk_write_space = listen_sock.sk_write_space; | |
524 | sk->sk_error_report = listen_sock.sk_error_report; | |
b81171cb | 525 | write_unlock_bh(&sk->sk_callback_lock); |
b3a5bbfd BP |
526 | } |
527 | ||
fdda387f | 528 | /* Make a socket active */ |
988419a9 | 529 | static void add_sock(struct socket *sock, struct connection *con) |
fdda387f | 530 | { |
b81171cb BP |
531 | struct sock *sk = sock->sk; |
532 | ||
533 | write_lock_bh(&sk->sk_callback_lock); | |
fdda387f PC |
534 | con->sock = sock; |
535 | ||
b81171cb | 536 | sk->sk_user_data = con; |
fdda387f | 537 | /* Install a data_ready callback */ |
b81171cb BP |
538 | sk->sk_data_ready = lowcomms_data_ready; |
539 | sk->sk_write_space = lowcomms_write_space; | |
540 | sk->sk_state_change = lowcomms_state_change; | |
541 | sk->sk_allocation = GFP_NOFS; | |
542 | sk->sk_error_report = lowcomms_error_report; | |
543 | write_unlock_bh(&sk->sk_callback_lock); | |
fdda387f PC |
544 | } |
545 | ||
6ed7257b | 546 | /* Add the port number to an IPv6 or 4 sockaddr and return the address |
fdda387f PC |
547 | length */ |
548 | static void make_sockaddr(struct sockaddr_storage *saddr, uint16_t port, | |
549 | int *addr_len) | |
550 | { | |
6ed7257b | 551 | saddr->ss_family = dlm_local_addr[0]->ss_family; |
ac33d071 | 552 | if (saddr->ss_family == AF_INET) { |
fdda387f PC |
553 | struct sockaddr_in *in4_addr = (struct sockaddr_in *)saddr; |
554 | in4_addr->sin_port = cpu_to_be16(port); | |
555 | *addr_len = sizeof(struct sockaddr_in); | |
6ed7257b | 556 | memset(&in4_addr->sin_zero, 0, sizeof(in4_addr->sin_zero)); |
ac33d071 | 557 | } else { |
fdda387f PC |
558 | struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)saddr; |
559 | in6_addr->sin6_port = cpu_to_be16(port); | |
560 | *addr_len = sizeof(struct sockaddr_in6); | |
561 | } | |
01c8cab2 | 562 | memset((char *)saddr + *addr_len, 0, sizeof(struct sockaddr_storage) - *addr_len); |
fdda387f PC |
563 | } |
564 | ||
565 | /* Close a remote connection and tidy up */ | |
0d737a8c MRL |
566 | static void close_connection(struct connection *con, bool and_other, |
567 | bool tx, bool rx) | |
fdda387f | 568 | { |
b2a66629 | 569 | bool closing = test_and_set_bit(CF_CLOSING, &con->flags); |
570 | ||
0aa18464 | 571 | if (tx && !closing && cancel_work_sync(&con->swork)) { |
0d737a8c | 572 | log_print("canceled swork for node %d", con->nodeid); |
0aa18464 | 573 | clear_bit(CF_WRITE_PENDING, &con->flags); |
574 | } | |
575 | if (rx && !closing && cancel_work_sync(&con->rwork)) { | |
0d737a8c | 576 | log_print("canceled rwork for node %d", con->nodeid); |
0aa18464 | 577 | clear_bit(CF_READ_PENDING, &con->flags); |
578 | } | |
fdda387f | 579 | |
0d737a8c | 580 | mutex_lock(&con->sock_mutex); |
fdda387f | 581 | if (con->sock) { |
cc661fc9 | 582 | restore_callbacks(con->sock); |
fdda387f PC |
583 | sock_release(con->sock); |
584 | con->sock = NULL; | |
585 | } | |
586 | if (con->othercon && and_other) { | |
ac33d071 | 587 | /* Will only re-enter once. */ |
0d737a8c | 588 | close_connection(con->othercon, false, true, true); |
fdda387f | 589 | } |
9e5f2825 | 590 | |
4798cbbf | 591 | con->rx_leftover = 0; |
61d96be0 PC |
592 | con->retries = 0; |
593 | mutex_unlock(&con->sock_mutex); | |
b2a66629 | 594 | clear_bit(CF_CLOSING, &con->flags); |
fdda387f PC |
595 | } |
596 | ||
055923bf AA |
597 | static void shutdown_connection(struct connection *con) |
598 | { | |
599 | int ret; | |
600 | ||
601 | if (cancel_work_sync(&con->swork)) { | |
602 | log_print("canceled swork for node %d", con->nodeid); | |
603 | clear_bit(CF_WRITE_PENDING, &con->flags); | |
604 | } | |
605 | ||
606 | mutex_lock(&con->sock_mutex); | |
607 | /* nothing to shutdown */ | |
608 | if (!con->sock) { | |
609 | mutex_unlock(&con->sock_mutex); | |
610 | return; | |
611 | } | |
612 | ||
613 | set_bit(CF_SHUTDOWN, &con->flags); | |
614 | ret = kernel_sock_shutdown(con->sock, SHUT_WR); | |
615 | mutex_unlock(&con->sock_mutex); | |
616 | if (ret) { | |
617 | log_print("Connection %p failed to shutdown: %d will force close", | |
618 | con, ret); | |
619 | goto force_close; | |
620 | } else { | |
621 | ret = wait_event_timeout(con->shutdown_wait, | |
622 | !test_bit(CF_SHUTDOWN, &con->flags), | |
623 | DLM_SHUTDOWN_WAIT_TIMEOUT); | |
624 | if (ret == 0) { | |
625 | log_print("Connection %p shutdown timed out, will force close", | |
626 | con); | |
627 | goto force_close; | |
628 | } | |
629 | } | |
630 | ||
631 | return; | |
632 | ||
633 | force_close: | |
634 | clear_bit(CF_SHUTDOWN, &con->flags); | |
635 | close_connection(con, false, true, true); | |
636 | } | |
637 | ||
638 | static void dlm_tcp_shutdown(struct connection *con) | |
639 | { | |
640 | if (con->othercon) | |
641 | shutdown_connection(con->othercon); | |
642 | shutdown_connection(con); | |
643 | } | |
644 | ||
4798cbbf AA |
645 | static int con_realloc_receive_buf(struct connection *con, int newlen) |
646 | { | |
647 | unsigned char *newbuf; | |
648 | ||
649 | newbuf = kmalloc(newlen, GFP_NOFS); | |
650 | if (!newbuf) | |
651 | return -ENOMEM; | |
652 | ||
653 | /* copy any leftover from last receive */ | |
654 | if (con->rx_leftover) | |
655 | memmove(newbuf, con->rx_buf, con->rx_leftover); | |
656 | ||
657 | /* swap to new buffer space */ | |
658 | kfree(con->rx_buf); | |
659 | con->rx_buflen = newlen; | |
660 | con->rx_buf = newbuf; | |
661 | ||
662 | return 0; | |
663 | } | |
664 | ||
fdda387f PC |
665 | /* Data received from remote end */ |
666 | static int receive_from_sock(struct connection *con) | |
667 | { | |
fdda387f | 668 | int call_again_soon = 0; |
4798cbbf AA |
669 | struct msghdr msg; |
670 | struct kvec iov; | |
671 | int ret, buflen; | |
fdda387f | 672 | |
f1f1c1cc | 673 | mutex_lock(&con->sock_mutex); |
fdda387f | 674 | |
a34fbc63 PC |
675 | if (con->sock == NULL) { |
676 | ret = -EAGAIN; | |
677 | goto out_close; | |
678 | } | |
4798cbbf | 679 | |
acee4e52 MRL |
680 | if (con->nodeid == 0) { |
681 | ret = -EINVAL; | |
682 | goto out_close; | |
683 | } | |
a34fbc63 | 684 | |
4798cbbf AA |
685 | /* realloc if we get new buffer size to read out */ |
686 | buflen = dlm_config.ci_buffer_size; | |
687 | if (con->rx_buflen != buflen && con->rx_leftover <= buflen) { | |
688 | ret = con_realloc_receive_buf(con, buflen); | |
689 | if (ret < 0) | |
fdda387f | 690 | goto out_resched; |
fdda387f PC |
691 | } |
692 | ||
4798cbbf AA |
693 | /* calculate new buffer parameter regarding last receive and |
694 | * possible leftover bytes | |
fdda387f | 695 | */ |
4798cbbf AA |
696 | iov.iov_base = con->rx_buf + con->rx_leftover; |
697 | iov.iov_len = con->rx_buflen - con->rx_leftover; | |
fdda387f | 698 | |
4798cbbf AA |
699 | memset(&msg, 0, sizeof(msg)); |
700 | msg.msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL; | |
701 | ret = kernel_recvmsg(con->sock, &msg, &iov, 1, iov.iov_len, | |
702 | msg.msg_flags); | |
fdda387f PC |
703 | if (ret <= 0) |
704 | goto out_close; | |
4798cbbf | 705 | else if (ret == iov.iov_len) |
ee44b4bc | 706 | call_again_soon = 1; |
bd44e2b0 | 707 | |
4798cbbf AA |
708 | /* new buflen according readed bytes and leftover from last receive */ |
709 | buflen = ret + con->rx_leftover; | |
710 | ret = dlm_process_incoming_buffer(con->nodeid, con->rx_buf, buflen); | |
711 | if (ret < 0) | |
712 | goto out_close; | |
fdda387f | 713 | |
4798cbbf AA |
714 | /* calculate leftover bytes from process and put it into begin of |
715 | * the receive buffer, so next receive we have the full message | |
716 | * at the start address of the receive buffer. | |
717 | */ | |
718 | con->rx_leftover = buflen - ret; | |
719 | if (con->rx_leftover) { | |
720 | memmove(con->rx_buf, con->rx_buf + ret, | |
721 | con->rx_leftover); | |
722 | call_again_soon = true; | |
fdda387f PC |
723 | } |
724 | ||
fdda387f PC |
725 | if (call_again_soon) |
726 | goto out_resched; | |
4798cbbf | 727 | |
f1f1c1cc | 728 | mutex_unlock(&con->sock_mutex); |
ac33d071 | 729 | return 0; |
fdda387f | 730 | |
ac33d071 | 731 | out_resched: |
1d6e8131 PC |
732 | if (!test_and_set_bit(CF_READ_PENDING, &con->flags)) |
733 | queue_work(recv_workqueue, &con->rwork); | |
f1f1c1cc | 734 | mutex_unlock(&con->sock_mutex); |
bd44e2b0 | 735 | return -EAGAIN; |
fdda387f | 736 | |
ac33d071 | 737 | out_close: |
f1f1c1cc | 738 | mutex_unlock(&con->sock_mutex); |
9e5f2825 | 739 | if (ret != -EAGAIN) { |
fdda387f | 740 | /* Reconnect when there is something to send */ |
055923bf AA |
741 | close_connection(con, false, true, false); |
742 | if (ret == 0) { | |
743 | log_print("connection %p got EOF from %d", | |
744 | con, con->nodeid); | |
745 | /* handling for tcp shutdown */ | |
746 | clear_bit(CF_SHUTDOWN, &con->flags); | |
747 | wake_up(&con->shutdown_wait); | |
748 | /* signal to breaking receive worker */ | |
749 | ret = -1; | |
750 | } | |
fdda387f | 751 | } |
fdda387f PC |
752 | return ret; |
753 | } | |
754 | ||
755 | /* Listening socket is busy, accept a connection */ | |
0774dc76 | 756 | static int accept_from_sock(struct connection *con) |
fdda387f PC |
757 | { |
758 | int result; | |
759 | struct sockaddr_storage peeraddr; | |
760 | struct socket *newsock; | |
761 | int len; | |
762 | int nodeid; | |
763 | struct connection *newcon; | |
bd44e2b0 | 764 | struct connection *addcon; |
3f78cd7d | 765 | unsigned int mark; |
fdda387f | 766 | |
513ef596 | 767 | if (!dlm_allow_conn) { |
513ef596 DT |
768 | return -1; |
769 | } | |
513ef596 | 770 | |
f1f1c1cc | 771 | mutex_lock_nested(&con->sock_mutex, 0); |
fdda387f | 772 | |
3421fb15 | 773 | if (!con->sock) { |
774 | mutex_unlock(&con->sock_mutex); | |
775 | return -ENOTCONN; | |
776 | } | |
fdda387f | 777 | |
3421fb15 | 778 | result = kernel_accept(con->sock, &newsock, O_NONBLOCK); |
fdda387f PC |
779 | if (result < 0) |
780 | goto accept_err; | |
781 | ||
782 | /* Get the connected socket's peer */ | |
783 | memset(&peeraddr, 0, sizeof(peeraddr)); | |
9b2c45d4 DV |
784 | len = newsock->ops->getname(newsock, (struct sockaddr *)&peeraddr, 2); |
785 | if (len < 0) { | |
fdda387f PC |
786 | result = -ECONNABORTED; |
787 | goto accept_err; | |
788 | } | |
789 | ||
790 | /* Get the new node's NODEID */ | |
791 | make_sockaddr(&peeraddr, 0, &len); | |
36b71a8b | 792 | if (addr_to_nodeid(&peeraddr, &nodeid)) { |
bcaadf5c | 793 | unsigned char *b=(unsigned char *)&peeraddr; |
617e82e1 | 794 | log_print("connect from non cluster node"); |
bcaadf5c MY |
795 | print_hex_dump_bytes("ss: ", DUMP_PREFIX_NONE, |
796 | b, sizeof(struct sockaddr_storage)); | |
fdda387f | 797 | sock_release(newsock); |
f1f1c1cc | 798 | mutex_unlock(&con->sock_mutex); |
fdda387f PC |
799 | return -1; |
800 | } | |
801 | ||
3f78cd7d AA |
802 | dlm_comm_mark(nodeid, &mark); |
803 | sock_set_mark(newsock->sk, mark); | |
804 | ||
fdda387f PC |
805 | log_print("got connection from %d", nodeid); |
806 | ||
807 | /* Check to see if we already have a connection to this node. This | |
808 | * could happen if the two nodes initiate a connection at roughly | |
809 | * the same time and the connections cross on the wire. | |
fdda387f PC |
810 | * In this case we store the incoming one in "othercon" |
811 | */ | |
748285cc | 812 | newcon = nodeid2con(nodeid, GFP_NOFS); |
fdda387f PC |
813 | if (!newcon) { |
814 | result = -ENOMEM; | |
815 | goto accept_err; | |
816 | } | |
f1f1c1cc | 817 | mutex_lock_nested(&newcon->sock_mutex, 1); |
fdda387f | 818 | if (newcon->sock) { |
ac33d071 | 819 | struct connection *othercon = newcon->othercon; |
fdda387f PC |
820 | |
821 | if (!othercon) { | |
a47666eb | 822 | othercon = kzalloc(sizeof(*othercon), GFP_NOFS); |
fdda387f | 823 | if (!othercon) { |
617e82e1 | 824 | log_print("failed to allocate incoming socket"); |
f1f1c1cc | 825 | mutex_unlock(&newcon->sock_mutex); |
fdda387f PC |
826 | result = -ENOMEM; |
827 | goto accept_err; | |
828 | } | |
4798cbbf AA |
829 | |
830 | othercon->rx_buflen = dlm_config.ci_buffer_size; | |
831 | othercon->rx_buf = kmalloc(othercon->rx_buflen, GFP_NOFS); | |
832 | if (!othercon->rx_buf) { | |
833 | mutex_unlock(&newcon->sock_mutex); | |
834 | kfree(othercon); | |
835 | log_print("failed to allocate incoming socket receive buffer"); | |
836 | result = -ENOMEM; | |
837 | goto accept_err; | |
838 | } | |
839 | ||
fdda387f PC |
840 | othercon->nodeid = nodeid; |
841 | othercon->rx_action = receive_from_sock; | |
f1f1c1cc | 842 | mutex_init(&othercon->sock_mutex); |
26b41099 | 843 | INIT_LIST_HEAD(&othercon->writequeue); |
844 | spin_lock_init(&othercon->writequeue_lock); | |
1d6e8131 PC |
845 | INIT_WORK(&othercon->swork, process_send_sockets); |
846 | INIT_WORK(&othercon->rwork, process_recv_sockets); | |
055923bf | 847 | init_waitqueue_head(&othercon->shutdown_wait); |
fdda387f | 848 | set_bit(CF_IS_OTHERCON, &othercon->flags); |
ba3ab3ca AA |
849 | } else { |
850 | /* close other sock con if we have something new */ | |
851 | close_connection(othercon, false, true, false); | |
61d96be0 | 852 | } |
ba3ab3ca | 853 | |
c7355827 | 854 | mutex_lock_nested(&othercon->sock_mutex, 2); |
ba3ab3ca AA |
855 | newcon->othercon = othercon; |
856 | add_sock(newsock, othercon); | |
857 | addcon = othercon; | |
858 | mutex_unlock(&othercon->sock_mutex); | |
fdda387f PC |
859 | } |
860 | else { | |
fdda387f | 861 | newcon->rx_action = receive_from_sock; |
3735b4b9 BP |
862 | /* accept copies the sk after we've saved the callbacks, so we |
863 | don't want to save them a second time or comm errors will | |
864 | result in calling sk_error_report recursively. */ | |
988419a9 | 865 | add_sock(newsock, newcon); |
bd44e2b0 | 866 | addcon = newcon; |
fdda387f PC |
867 | } |
868 | ||
f1f1c1cc | 869 | mutex_unlock(&newcon->sock_mutex); |
fdda387f PC |
870 | |
871 | /* | |
872 | * Add it to the active queue in case we got data | |
25985edc | 873 | * between processing the accept adding the socket |
fdda387f PC |
874 | * to the read_sockets list |
875 | */ | |
bd44e2b0 PC |
876 | if (!test_and_set_bit(CF_READ_PENDING, &addcon->flags)) |
877 | queue_work(recv_workqueue, &addcon->rwork); | |
f1f1c1cc | 878 | mutex_unlock(&con->sock_mutex); |
fdda387f PC |
879 | |
880 | return 0; | |
881 | ||
ac33d071 | 882 | accept_err: |
f1f1c1cc | 883 | mutex_unlock(&con->sock_mutex); |
3421fb15 | 884 | if (newsock) |
885 | sock_release(newsock); | |
fdda387f PC |
886 | |
887 | if (result != -EAGAIN) | |
617e82e1 | 888 | log_print("error accepting connection from node: %d", result); |
fdda387f PC |
889 | return result; |
890 | } | |
891 | ||
6ed7257b PC |
892 | static void free_entry(struct writequeue_entry *e) |
893 | { | |
894 | __free_page(e->page); | |
895 | kfree(e); | |
896 | } | |
897 | ||
5d689871 MC |
898 | /* |
899 | * writequeue_entry_complete - try to delete and free write queue entry | |
900 | * @e: write queue entry to try to delete | |
901 | * @completed: bytes completed | |
902 | * | |
903 | * writequeue_lock must be held. | |
904 | */ | |
905 | static void writequeue_entry_complete(struct writequeue_entry *e, int completed) | |
906 | { | |
907 | e->offset += completed; | |
908 | e->len -= completed; | |
909 | ||
910 | if (e->len == 0 && e->users == 0) { | |
911 | list_del(&e->list); | |
912 | free_entry(e); | |
913 | } | |
914 | } | |
915 | ||
ee44b4bc MRL |
916 | /* |
917 | * sctp_bind_addrs - bind a SCTP socket to all our addresses | |
918 | */ | |
919 | static int sctp_bind_addrs(struct connection *con, uint16_t port) | |
920 | { | |
921 | struct sockaddr_storage localaddr; | |
c0425a42 | 922 | struct sockaddr *addr = (struct sockaddr *)&localaddr; |
ee44b4bc MRL |
923 | int i, addr_len, result = 0; |
924 | ||
925 | for (i = 0; i < dlm_local_count; i++) { | |
926 | memcpy(&localaddr, dlm_local_addr[i], sizeof(localaddr)); | |
927 | make_sockaddr(&localaddr, port, &addr_len); | |
928 | ||
929 | if (!i) | |
c0425a42 | 930 | result = kernel_bind(con->sock, addr, addr_len); |
ee44b4bc | 931 | else |
c0425a42 | 932 | result = sock_bind_add(con->sock->sk, addr, addr_len); |
ee44b4bc MRL |
933 | |
934 | if (result < 0) { | |
935 | log_print("Can't bind to %d addr number %d, %d.\n", | |
936 | port, i + 1, result); | |
937 | break; | |
938 | } | |
939 | } | |
940 | return result; | |
941 | } | |
942 | ||
6ed7257b PC |
943 | /* Initiate an SCTP association. |
944 | This is a special case of send_to_sock() in that we don't yet have a | |
945 | peeled-off socket for this association, so we use the listening socket | |
946 | and add the primary IP address of the remote node. | |
947 | */ | |
ee44b4bc | 948 | static void sctp_connect_to_sock(struct connection *con) |
6ed7257b | 949 | { |
ee44b4bc | 950 | struct sockaddr_storage daddr; |
ee44b4bc MRL |
951 | int result; |
952 | int addr_len; | |
953 | struct socket *sock; | |
9c9f168f | 954 | unsigned int mark; |
ee44b4bc MRL |
955 | |
956 | if (con->nodeid == 0) { | |
957 | log_print("attempt to connect sock 0 foiled"); | |
958 | return; | |
959 | } | |
6ed7257b | 960 | |
3f78cd7d | 961 | dlm_comm_mark(con->nodeid, &mark); |
0461e0db | 962 | |
5d689871 | 963 | mutex_lock(&con->sock_mutex); |
6ed7257b | 964 | |
ee44b4bc MRL |
965 | /* Some odd races can cause double-connects, ignore them */ |
966 | if (con->retries++ > MAX_CONNECT_RETRIES) | |
967 | goto out; | |
968 | ||
969 | if (con->sock) { | |
970 | log_print("node %d already connected.", con->nodeid); | |
971 | goto out; | |
972 | } | |
973 | ||
974 | memset(&daddr, 0, sizeof(daddr)); | |
975 | result = nodeid_to_addr(con->nodeid, &daddr, NULL, true); | |
976 | if (result < 0) { | |
6ed7257b | 977 | log_print("no address for nodeid %d", con->nodeid); |
ee44b4bc | 978 | goto out; |
6ed7257b | 979 | } |
6ed7257b | 980 | |
ee44b4bc MRL |
981 | /* Create a socket to communicate with */ |
982 | result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family, | |
983 | SOCK_STREAM, IPPROTO_SCTP, &sock); | |
984 | if (result < 0) | |
985 | goto socket_err; | |
6ed7257b | 986 | |
9c9f168f AA |
987 | sock_set_mark(sock->sk, mark); |
988 | ||
ee44b4bc MRL |
989 | con->rx_action = receive_from_sock; |
990 | con->connect_action = sctp_connect_to_sock; | |
988419a9 | 991 | add_sock(sock, con); |
6ed7257b | 992 | |
ee44b4bc MRL |
993 | /* Bind to all addresses. */ |
994 | if (sctp_bind_addrs(con, 0)) | |
995 | goto bind_err; | |
6ed7257b | 996 | |
ee44b4bc | 997 | make_sockaddr(&daddr, dlm_config.ci_tcp_port, &addr_len); |
6ed7257b | 998 | |
ee44b4bc | 999 | log_print("connecting to %d", con->nodeid); |
6ed7257b | 1000 | |
ee44b4bc | 1001 | /* Turn off Nagle's algorithm */ |
40ef92c6 | 1002 | sctp_sock_set_nodelay(sock->sk); |
6ed7257b | 1003 | |
f706d830 GH |
1004 | /* |
1005 | * Make sock->ops->connect() function return in specified time, | |
1006 | * since O_NONBLOCK argument in connect() function does not work here, | |
1007 | * then, we should restore the default value of this attribute. | |
1008 | */ | |
76ee0785 | 1009 | sock_set_sndtimeo(sock->sk, 5); |
ee44b4bc | 1010 | result = sock->ops->connect(sock, (struct sockaddr *)&daddr, addr_len, |
da3627c3 | 1011 | 0); |
76ee0785 | 1012 | sock_set_sndtimeo(sock->sk, 0); |
f706d830 | 1013 | |
ee44b4bc MRL |
1014 | if (result == -EINPROGRESS) |
1015 | result = 0; | |
1016 | if (result == 0) | |
1017 | goto out; | |
98e1b60e | 1018 | |
ee44b4bc MRL |
1019 | bind_err: |
1020 | con->sock = NULL; | |
1021 | sock_release(sock); | |
6ed7257b | 1022 | |
ee44b4bc MRL |
1023 | socket_err: |
1024 | /* | |
1025 | * Some errors are fatal and this list might need adjusting. For other | |
1026 | * errors we try again until the max number of retries is reached. | |
1027 | */ | |
1028 | if (result != -EHOSTUNREACH && | |
1029 | result != -ENETUNREACH && | |
1030 | result != -ENETDOWN && | |
1031 | result != -EINVAL && | |
1032 | result != -EPROTONOSUPPORT) { | |
1033 | log_print("connect %d try %d error %d", con->nodeid, | |
1034 | con->retries, result); | |
1035 | mutex_unlock(&con->sock_mutex); | |
1036 | msleep(1000); | |
ee44b4bc MRL |
1037 | lowcomms_connect_sock(con); |
1038 | return; | |
6ed7257b | 1039 | } |
5d689871 | 1040 | |
ee44b4bc | 1041 | out: |
5d689871 | 1042 | mutex_unlock(&con->sock_mutex); |
6ed7257b PC |
1043 | } |
1044 | ||
fdda387f | 1045 | /* Connect a new socket to its peer */ |
6ed7257b | 1046 | static void tcp_connect_to_sock(struct connection *con) |
fdda387f | 1047 | { |
6bd8feda | 1048 | struct sockaddr_storage saddr, src_addr; |
fdda387f | 1049 | int addr_len; |
a89d63a1 | 1050 | struct socket *sock = NULL; |
9c9f168f | 1051 | unsigned int mark; |
36b71a8b | 1052 | int result; |
fdda387f PC |
1053 | |
1054 | if (con->nodeid == 0) { | |
1055 | log_print("attempt to connect sock 0 foiled"); | |
ac33d071 | 1056 | return; |
fdda387f PC |
1057 | } |
1058 | ||
3f78cd7d | 1059 | dlm_comm_mark(con->nodeid, &mark); |
0461e0db | 1060 | |
f1f1c1cc | 1061 | mutex_lock(&con->sock_mutex); |
fdda387f PC |
1062 | if (con->retries++ > MAX_CONNECT_RETRIES) |
1063 | goto out; | |
1064 | ||
1065 | /* Some odd races can cause double-connects, ignore them */ | |
36b71a8b | 1066 | if (con->sock) |
fdda387f | 1067 | goto out; |
fdda387f PC |
1068 | |
1069 | /* Create a socket to communicate with */ | |
eeb1bd5c EB |
1070 | result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family, |
1071 | SOCK_STREAM, IPPROTO_TCP, &sock); | |
fdda387f PC |
1072 | if (result < 0) |
1073 | goto out_err; | |
1074 | ||
9c9f168f AA |
1075 | sock_set_mark(sock->sk, mark); |
1076 | ||
fdda387f | 1077 | memset(&saddr, 0, sizeof(saddr)); |
98e1b60e | 1078 | result = nodeid_to_addr(con->nodeid, &saddr, NULL, false); |
36b71a8b DT |
1079 | if (result < 0) { |
1080 | log_print("no address for nodeid %d", con->nodeid); | |
ac33d071 | 1081 | goto out_err; |
36b71a8b | 1082 | } |
fdda387f | 1083 | |
fdda387f | 1084 | con->rx_action = receive_from_sock; |
6ed7257b | 1085 | con->connect_action = tcp_connect_to_sock; |
055923bf | 1086 | con->shutdown_action = dlm_tcp_shutdown; |
988419a9 | 1087 | add_sock(sock, con); |
fdda387f | 1088 | |
6bd8feda LH |
1089 | /* Bind to our cluster-known address connecting to avoid |
1090 | routing problems */ | |
1091 | memcpy(&src_addr, dlm_local_addr[0], sizeof(src_addr)); | |
1092 | make_sockaddr(&src_addr, 0, &addr_len); | |
1093 | result = sock->ops->bind(sock, (struct sockaddr *) &src_addr, | |
1094 | addr_len); | |
1095 | if (result < 0) { | |
1096 | log_print("could not bind for connect: %d", result); | |
1097 | /* This *may* not indicate a critical error */ | |
1098 | } | |
1099 | ||
68c817a1 | 1100 | make_sockaddr(&saddr, dlm_config.ci_tcp_port, &addr_len); |
fdda387f | 1101 | |
fdda387f | 1102 | log_print("connecting to %d", con->nodeid); |
cb2d45da DT |
1103 | |
1104 | /* Turn off Nagle's algorithm */ | |
12abc5ee | 1105 | tcp_sock_set_nodelay(sock->sk); |
cb2d45da | 1106 | |
36b71a8b | 1107 | result = sock->ops->connect(sock, (struct sockaddr *)&saddr, addr_len, |
ac33d071 | 1108 | O_NONBLOCK); |
fdda387f PC |
1109 | if (result == -EINPROGRESS) |
1110 | result = 0; | |
ac33d071 PC |
1111 | if (result == 0) |
1112 | goto out; | |
fdda387f | 1113 | |
ac33d071 | 1114 | out_err: |
fdda387f PC |
1115 | if (con->sock) { |
1116 | sock_release(con->sock); | |
1117 | con->sock = NULL; | |
a89d63a1 CD |
1118 | } else if (sock) { |
1119 | sock_release(sock); | |
fdda387f PC |
1120 | } |
1121 | /* | |
1122 | * Some errors are fatal and this list might need adjusting. For other | |
1123 | * errors we try again until the max number of retries is reached. | |
1124 | */ | |
36b71a8b DT |
1125 | if (result != -EHOSTUNREACH && |
1126 | result != -ENETUNREACH && | |
1127 | result != -ENETDOWN && | |
1128 | result != -EINVAL && | |
1129 | result != -EPROTONOSUPPORT) { | |
1130 | log_print("connect %d try %d error %d", con->nodeid, | |
1131 | con->retries, result); | |
1132 | mutex_unlock(&con->sock_mutex); | |
1133 | msleep(1000); | |
fdda387f | 1134 | lowcomms_connect_sock(con); |
36b71a8b | 1135 | return; |
fdda387f | 1136 | } |
ac33d071 | 1137 | out: |
f1f1c1cc | 1138 | mutex_unlock(&con->sock_mutex); |
ac33d071 | 1139 | return; |
fdda387f PC |
1140 | } |
1141 | ||
6ed7257b PC |
1142 | static struct socket *tcp_create_listen_sock(struct connection *con, |
1143 | struct sockaddr_storage *saddr) | |
fdda387f | 1144 | { |
ac33d071 | 1145 | struct socket *sock = NULL; |
fdda387f | 1146 | int result = 0; |
fdda387f PC |
1147 | int addr_len; |
1148 | ||
6ed7257b | 1149 | if (dlm_local_addr[0]->ss_family == AF_INET) |
fdda387f PC |
1150 | addr_len = sizeof(struct sockaddr_in); |
1151 | else | |
1152 | addr_len = sizeof(struct sockaddr_in6); | |
1153 | ||
1154 | /* Create a socket to communicate with */ | |
eeb1bd5c EB |
1155 | result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family, |
1156 | SOCK_STREAM, IPPROTO_TCP, &sock); | |
fdda387f | 1157 | if (result < 0) { |
617e82e1 | 1158 | log_print("Can't create listening comms socket"); |
fdda387f PC |
1159 | goto create_out; |
1160 | } | |
1161 | ||
a5b7ab63 AA |
1162 | sock_set_mark(sock->sk, dlm_config.ci_mark); |
1163 | ||
cb2d45da | 1164 | /* Turn off Nagle's algorithm */ |
12abc5ee | 1165 | tcp_sock_set_nodelay(sock->sk); |
cb2d45da | 1166 | |
b58f0e8f | 1167 | sock_set_reuseaddr(sock->sk); |
6ed7257b | 1168 | |
93eaadeb | 1169 | write_lock_bh(&sock->sk->sk_callback_lock); |
b81171cb | 1170 | sock->sk->sk_user_data = con; |
cc661fc9 | 1171 | save_listen_callbacks(sock); |
0774dc76 | 1172 | con->rx_action = accept_from_sock; |
6ed7257b | 1173 | con->connect_action = tcp_connect_to_sock; |
93eaadeb | 1174 | write_unlock_bh(&sock->sk->sk_callback_lock); |
fdda387f PC |
1175 | |
1176 | /* Bind to our port */ | |
68c817a1 | 1177 | make_sockaddr(saddr, dlm_config.ci_tcp_port, &addr_len); |
fdda387f PC |
1178 | result = sock->ops->bind(sock, (struct sockaddr *) saddr, addr_len); |
1179 | if (result < 0) { | |
617e82e1 | 1180 | log_print("Can't bind to port %d", dlm_config.ci_tcp_port); |
fdda387f PC |
1181 | sock_release(sock); |
1182 | sock = NULL; | |
1183 | con->sock = NULL; | |
1184 | goto create_out; | |
1185 | } | |
ce3d9544 | 1186 | sock_set_keepalive(sock->sk); |
fdda387f PC |
1187 | |
1188 | result = sock->ops->listen(sock, 5); | |
1189 | if (result < 0) { | |
617e82e1 | 1190 | log_print("Can't listen on port %d", dlm_config.ci_tcp_port); |
fdda387f PC |
1191 | sock_release(sock); |
1192 | sock = NULL; | |
1193 | goto create_out; | |
1194 | } | |
1195 | ||
ac33d071 | 1196 | create_out: |
fdda387f PC |
1197 | return sock; |
1198 | } | |
1199 | ||
6ed7257b PC |
1200 | /* Get local addresses */ |
1201 | static void init_local(void) | |
1202 | { | |
1203 | struct sockaddr_storage sas, *addr; | |
1204 | int i; | |
1205 | ||
30d3a237 | 1206 | dlm_local_count = 0; |
1b189b88 | 1207 | for (i = 0; i < DLM_MAX_ADDR_COUNT; i++) { |
6ed7257b PC |
1208 | if (dlm_our_addr(&sas, i)) |
1209 | break; | |
1210 | ||
5c93f56f | 1211 | addr = kmemdup(&sas, sizeof(*addr), GFP_NOFS); |
6ed7257b PC |
1212 | if (!addr) |
1213 | break; | |
6ed7257b PC |
1214 | dlm_local_addr[dlm_local_count++] = addr; |
1215 | } | |
1216 | } | |
1217 | ||
043697f0 AA |
1218 | static void deinit_local(void) |
1219 | { | |
1220 | int i; | |
1221 | ||
1222 | for (i = 0; i < dlm_local_count; i++) | |
1223 | kfree(dlm_local_addr[i]); | |
1224 | } | |
1225 | ||
6ed7257b PC |
1226 | /* Initialise SCTP socket and bind to all interfaces */ |
1227 | static int sctp_listen_for_all(void) | |
1228 | { | |
1229 | struct socket *sock = NULL; | |
ee44b4bc | 1230 | int result = -EINVAL; |
573c24c4 | 1231 | struct connection *con = nodeid2con(0, GFP_NOFS); |
6ed7257b PC |
1232 | |
1233 | if (!con) | |
1234 | return -ENOMEM; | |
1235 | ||
1236 | log_print("Using SCTP for communications"); | |
1237 | ||
eeb1bd5c | 1238 | result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family, |
ee44b4bc | 1239 | SOCK_STREAM, IPPROTO_SCTP, &sock); |
6ed7257b PC |
1240 | if (result < 0) { |
1241 | log_print("Can't create comms socket, check SCTP is loaded"); | |
1242 | goto out; | |
1243 | } | |
1244 | ||
26cfabf9 | 1245 | sock_set_rcvbuf(sock->sk, NEEDED_RMEM); |
a5b7ab63 | 1246 | sock_set_mark(sock->sk, dlm_config.ci_mark); |
40ef92c6 | 1247 | sctp_sock_set_nodelay(sock->sk); |
86e92ad2 | 1248 | |
b81171cb | 1249 | write_lock_bh(&sock->sk->sk_callback_lock); |
6ed7257b PC |
1250 | /* Init con struct */ |
1251 | sock->sk->sk_user_data = con; | |
cc661fc9 | 1252 | save_listen_callbacks(sock); |
6ed7257b PC |
1253 | con->sock = sock; |
1254 | con->sock->sk->sk_data_ready = lowcomms_data_ready; | |
0774dc76 | 1255 | con->rx_action = accept_from_sock; |
ee44b4bc | 1256 | con->connect_action = sctp_connect_to_sock; |
6ed7257b | 1257 | |
b81171cb BP |
1258 | write_unlock_bh(&sock->sk->sk_callback_lock); |
1259 | ||
ee44b4bc MRL |
1260 | /* Bind to all addresses. */ |
1261 | if (sctp_bind_addrs(con, dlm_config.ci_tcp_port)) | |
1262 | goto create_delsock; | |
6ed7257b PC |
1263 | |
1264 | result = sock->ops->listen(sock, 5); | |
1265 | if (result < 0) { | |
1266 | log_print("Can't set socket listening"); | |
1267 | goto create_delsock; | |
1268 | } | |
1269 | ||
1270 | return 0; | |
1271 | ||
1272 | create_delsock: | |
1273 | sock_release(sock); | |
1274 | con->sock = NULL; | |
1275 | out: | |
1276 | return result; | |
1277 | } | |
1278 | ||
1279 | static int tcp_listen_for_all(void) | |
fdda387f PC |
1280 | { |
1281 | struct socket *sock = NULL; | |
573c24c4 | 1282 | struct connection *con = nodeid2con(0, GFP_NOFS); |
fdda387f PC |
1283 | int result = -EINVAL; |
1284 | ||
6ed7257b PC |
1285 | if (!con) |
1286 | return -ENOMEM; | |
1287 | ||
fdda387f | 1288 | /* We don't support multi-homed hosts */ |
6ed7257b | 1289 | if (dlm_local_addr[1] != NULL) { |
617e82e1 DT |
1290 | log_print("TCP protocol can't handle multi-homed hosts, " |
1291 | "try SCTP"); | |
6ed7257b PC |
1292 | return -EINVAL; |
1293 | } | |
1294 | ||
1295 | log_print("Using TCP for communications"); | |
1296 | ||
6ed7257b | 1297 | sock = tcp_create_listen_sock(con, dlm_local_addr[0]); |
fdda387f | 1298 | if (sock) { |
988419a9 | 1299 | add_sock(sock, con); |
fdda387f PC |
1300 | result = 0; |
1301 | } | |
1302 | else { | |
1303 | result = -EADDRINUSE; | |
1304 | } | |
1305 | ||
1306 | return result; | |
1307 | } | |
1308 | ||
1309 | ||
1310 | ||
1311 | static struct writequeue_entry *new_writequeue_entry(struct connection *con, | |
1312 | gfp_t allocation) | |
1313 | { | |
1314 | struct writequeue_entry *entry; | |
1315 | ||
1316 | entry = kmalloc(sizeof(struct writequeue_entry), allocation); | |
1317 | if (!entry) | |
1318 | return NULL; | |
1319 | ||
1320 | entry->page = alloc_page(allocation); | |
1321 | if (!entry->page) { | |
1322 | kfree(entry); | |
1323 | return NULL; | |
1324 | } | |
1325 | ||
1326 | entry->offset = 0; | |
1327 | entry->len = 0; | |
1328 | entry->end = 0; | |
1329 | entry->users = 0; | |
1330 | entry->con = con; | |
1331 | ||
1332 | return entry; | |
1333 | } | |
1334 | ||
617e82e1 | 1335 | void *dlm_lowcomms_get_buffer(int nodeid, int len, gfp_t allocation, char **ppc) |
fdda387f PC |
1336 | { |
1337 | struct connection *con; | |
1338 | struct writequeue_entry *e; | |
1339 | int offset = 0; | |
fdda387f | 1340 | |
fdda387f PC |
1341 | con = nodeid2con(nodeid, allocation); |
1342 | if (!con) | |
1343 | return NULL; | |
1344 | ||
4edde74e | 1345 | spin_lock(&con->writequeue_lock); |
fdda387f | 1346 | e = list_entry(con->writequeue.prev, struct writequeue_entry, list); |
ac33d071 | 1347 | if ((&e->list == &con->writequeue) || |
09cbfeaf | 1348 | (PAGE_SIZE - e->end < len)) { |
fdda387f PC |
1349 | e = NULL; |
1350 | } else { | |
1351 | offset = e->end; | |
1352 | e->end += len; | |
eeee2b5f | 1353 | e->users++; |
fdda387f PC |
1354 | } |
1355 | spin_unlock(&con->writequeue_lock); | |
1356 | ||
1357 | if (e) { | |
ac33d071 | 1358 | got_one: |
fdda387f PC |
1359 | *ppc = page_address(e->page) + offset; |
1360 | return e; | |
1361 | } | |
1362 | ||
1363 | e = new_writequeue_entry(con, allocation); | |
1364 | if (e) { | |
1365 | spin_lock(&con->writequeue_lock); | |
1366 | offset = e->end; | |
1367 | e->end += len; | |
eeee2b5f | 1368 | e->users++; |
fdda387f PC |
1369 | list_add_tail(&e->list, &con->writequeue); |
1370 | spin_unlock(&con->writequeue_lock); | |
1371 | goto got_one; | |
1372 | } | |
1373 | return NULL; | |
1374 | } | |
1375 | ||
1376 | void dlm_lowcomms_commit_buffer(void *mh) | |
1377 | { | |
1378 | struct writequeue_entry *e = (struct writequeue_entry *)mh; | |
1379 | struct connection *con = e->con; | |
1380 | int users; | |
1381 | ||
4edde74e | 1382 | spin_lock(&con->writequeue_lock); |
fdda387f PC |
1383 | users = --e->users; |
1384 | if (users) | |
1385 | goto out; | |
1386 | e->len = e->end - e->offset; | |
fdda387f PC |
1387 | spin_unlock(&con->writequeue_lock); |
1388 | ||
01da24d3 | 1389 | queue_work(send_workqueue, &con->swork); |
fdda387f PC |
1390 | return; |
1391 | ||
ac33d071 | 1392 | out: |
fdda387f PC |
1393 | spin_unlock(&con->writequeue_lock); |
1394 | return; | |
1395 | } | |
1396 | ||
fdda387f | 1397 | /* Send a message */ |
ac33d071 | 1398 | static void send_to_sock(struct connection *con) |
fdda387f PC |
1399 | { |
1400 | int ret = 0; | |
fdda387f PC |
1401 | const int msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL; |
1402 | struct writequeue_entry *e; | |
1403 | int len, offset; | |
f92c8dd7 | 1404 | int count = 0; |
fdda387f | 1405 | |
f1f1c1cc | 1406 | mutex_lock(&con->sock_mutex); |
fdda387f PC |
1407 | if (con->sock == NULL) |
1408 | goto out_connect; | |
1409 | ||
fdda387f PC |
1410 | spin_lock(&con->writequeue_lock); |
1411 | for (;;) { | |
1412 | e = list_entry(con->writequeue.next, struct writequeue_entry, | |
1413 | list); | |
1414 | if ((struct list_head *) e == &con->writequeue) | |
1415 | break; | |
1416 | ||
1417 | len = e->len; | |
1418 | offset = e->offset; | |
1419 | BUG_ON(len == 0 && e->users == 0); | |
1420 | spin_unlock(&con->writequeue_lock); | |
1421 | ||
1422 | ret = 0; | |
1423 | if (len) { | |
1329e3f2 PB |
1424 | ret = kernel_sendpage(con->sock, e->page, offset, len, |
1425 | msg_flags); | |
d66f8277 | 1426 | if (ret == -EAGAIN || ret == 0) { |
b36930dd | 1427 | if (ret == -EAGAIN && |
9cd3e072 | 1428 | test_bit(SOCKWQ_ASYNC_NOSPACE, &con->sock->flags) && |
b36930dd DM |
1429 | !test_and_set_bit(CF_APP_LIMITED, &con->flags)) { |
1430 | /* Notify TCP that we're limited by the | |
1431 | * application window size. | |
1432 | */ | |
1433 | set_bit(SOCK_NOSPACE, &con->sock->flags); | |
1434 | con->sock->sk->sk_write_pending++; | |
1435 | } | |
d66f8277 | 1436 | cond_resched(); |
fdda387f | 1437 | goto out; |
9c5bef58 | 1438 | } else if (ret < 0) |
fdda387f | 1439 | goto send_error; |
d66f8277 | 1440 | } |
f92c8dd7 BP |
1441 | |
1442 | /* Don't starve people filling buffers */ | |
1443 | if (++count >= MAX_SEND_MSG_COUNT) { | |
ac33d071 | 1444 | cond_resched(); |
f92c8dd7 BP |
1445 | count = 0; |
1446 | } | |
fdda387f PC |
1447 | |
1448 | spin_lock(&con->writequeue_lock); | |
5d689871 | 1449 | writequeue_entry_complete(e, ret); |
fdda387f PC |
1450 | } |
1451 | spin_unlock(&con->writequeue_lock); | |
ac33d071 | 1452 | out: |
f1f1c1cc | 1453 | mutex_unlock(&con->sock_mutex); |
ac33d071 | 1454 | return; |
fdda387f | 1455 | |
ac33d071 | 1456 | send_error: |
f1f1c1cc | 1457 | mutex_unlock(&con->sock_mutex); |
ba3ab3ca | 1458 | close_connection(con, false, false, true); |
01da24d3 BP |
1459 | /* Requeue the send work. When the work daemon runs again, it will try |
1460 | a new connection, then call this function again. */ | |
1461 | queue_work(send_workqueue, &con->swork); | |
ac33d071 | 1462 | return; |
fdda387f | 1463 | |
ac33d071 | 1464 | out_connect: |
f1f1c1cc | 1465 | mutex_unlock(&con->sock_mutex); |
01da24d3 BP |
1466 | queue_work(send_workqueue, &con->swork); |
1467 | cond_resched(); | |
fdda387f PC |
1468 | } |
1469 | ||
1470 | static void clean_one_writequeue(struct connection *con) | |
1471 | { | |
5e9ccc37 | 1472 | struct writequeue_entry *e, *safe; |
fdda387f PC |
1473 | |
1474 | spin_lock(&con->writequeue_lock); | |
5e9ccc37 | 1475 | list_for_each_entry_safe(e, safe, &con->writequeue, list) { |
fdda387f PC |
1476 | list_del(&e->list); |
1477 | free_entry(e); | |
1478 | } | |
1479 | spin_unlock(&con->writequeue_lock); | |
1480 | } | |
1481 | ||
1482 | /* Called from recovery when it knows that a node has | |
1483 | left the cluster */ | |
1484 | int dlm_lowcomms_close(int nodeid) | |
1485 | { | |
1486 | struct connection *con; | |
36b71a8b | 1487 | struct dlm_node_addr *na; |
fdda387f | 1488 | |
fdda387f PC |
1489 | log_print("closing connection to node %d", nodeid); |
1490 | con = nodeid2con(nodeid, 0); | |
1491 | if (con) { | |
063c4c99 | 1492 | set_bit(CF_CLOSE, &con->flags); |
0d737a8c | 1493 | close_connection(con, true, true, true); |
fdda387f | 1494 | clean_one_writequeue(con); |
fdda387f | 1495 | } |
36b71a8b DT |
1496 | |
1497 | spin_lock(&dlm_node_addrs_spin); | |
1498 | na = find_node_addr(nodeid); | |
1499 | if (na) { | |
1500 | list_del(&na->list); | |
1501 | while (na->addr_count--) | |
1502 | kfree(na->addr[na->addr_count]); | |
1503 | kfree(na); | |
1504 | } | |
1505 | spin_unlock(&dlm_node_addrs_spin); | |
1506 | ||
fdda387f | 1507 | return 0; |
fdda387f PC |
1508 | } |
1509 | ||
6ed7257b | 1510 | /* Receive workqueue function */ |
1d6e8131 | 1511 | static void process_recv_sockets(struct work_struct *work) |
fdda387f | 1512 | { |
1d6e8131 PC |
1513 | struct connection *con = container_of(work, struct connection, rwork); |
1514 | int err; | |
fdda387f | 1515 | |
1d6e8131 PC |
1516 | clear_bit(CF_READ_PENDING, &con->flags); |
1517 | do { | |
1518 | err = con->rx_action(con); | |
1519 | } while (!err); | |
fdda387f PC |
1520 | } |
1521 | ||
6ed7257b | 1522 | /* Send workqueue function */ |
1d6e8131 | 1523 | static void process_send_sockets(struct work_struct *work) |
fdda387f | 1524 | { |
1d6e8131 | 1525 | struct connection *con = container_of(work, struct connection, swork); |
fdda387f | 1526 | |
8a4abb08 | 1527 | clear_bit(CF_WRITE_PENDING, &con->flags); |
61d9102b | 1528 | if (con->sock == NULL) /* not mutex protected so check it inside too */ |
6ed7257b | 1529 | con->connect_action(con); |
01da24d3 | 1530 | if (!list_empty(&con->writequeue)) |
063c4c99 | 1531 | send_to_sock(con); |
fdda387f PC |
1532 | } |
1533 | ||
1d6e8131 | 1534 | static void work_stop(void) |
fdda387f | 1535 | { |
b355516f DW |
1536 | if (recv_workqueue) |
1537 | destroy_workqueue(recv_workqueue); | |
1538 | if (send_workqueue) | |
1539 | destroy_workqueue(send_workqueue); | |
fdda387f PC |
1540 | } |
1541 | ||
1d6e8131 | 1542 | static int work_start(void) |
fdda387f | 1543 | { |
e43f055a DT |
1544 | recv_workqueue = alloc_workqueue("dlm_recv", |
1545 | WQ_UNBOUND | WQ_MEM_RECLAIM, 1); | |
b9d41052 NK |
1546 | if (!recv_workqueue) { |
1547 | log_print("can't start dlm_recv"); | |
1548 | return -ENOMEM; | |
fdda387f | 1549 | } |
fdda387f | 1550 | |
e43f055a DT |
1551 | send_workqueue = alloc_workqueue("dlm_send", |
1552 | WQ_UNBOUND | WQ_MEM_RECLAIM, 1); | |
b9d41052 NK |
1553 | if (!send_workqueue) { |
1554 | log_print("can't start dlm_send"); | |
1d6e8131 | 1555 | destroy_workqueue(recv_workqueue); |
b9d41052 | 1556 | return -ENOMEM; |
fdda387f | 1557 | } |
fdda387f PC |
1558 | |
1559 | return 0; | |
1560 | } | |
1561 | ||
f0fb83cb | 1562 | static void _stop_conn(struct connection *con, bool and_other) |
fdda387f | 1563 | { |
f0fb83cb | 1564 | mutex_lock(&con->sock_mutex); |
173a31fe | 1565 | set_bit(CF_CLOSE, &con->flags); |
f0fb83cb | 1566 | set_bit(CF_READ_PENDING, &con->flags); |
8a4abb08 | 1567 | set_bit(CF_WRITE_PENDING, &con->flags); |
93eaadeb | 1568 | if (con->sock && con->sock->sk) { |
1569 | write_lock_bh(&con->sock->sk->sk_callback_lock); | |
5e9ccc37 | 1570 | con->sock->sk->sk_user_data = NULL; |
93eaadeb | 1571 | write_unlock_bh(&con->sock->sk->sk_callback_lock); |
1572 | } | |
f0fb83cb | 1573 | if (con->othercon && and_other) |
1574 | _stop_conn(con->othercon, false); | |
1575 | mutex_unlock(&con->sock_mutex); | |
1576 | } | |
1577 | ||
1578 | static void stop_conn(struct connection *con) | |
1579 | { | |
1580 | _stop_conn(con, true); | |
5e9ccc37 | 1581 | } |
fdda387f | 1582 | |
055923bf AA |
1583 | static void shutdown_conn(struct connection *con) |
1584 | { | |
1585 | if (con->shutdown_action) | |
1586 | con->shutdown_action(con); | |
1587 | } | |
1588 | ||
4798cbbf AA |
1589 | static void connection_release(struct rcu_head *rcu) |
1590 | { | |
1591 | struct connection *con = container_of(rcu, struct connection, rcu); | |
1592 | ||
1593 | kfree(con->rx_buf); | |
1594 | kfree(con); | |
1595 | } | |
1596 | ||
5e9ccc37 CC |
1597 | static void free_conn(struct connection *con) |
1598 | { | |
0d737a8c | 1599 | close_connection(con, true, true, true); |
a47666eb AA |
1600 | spin_lock(&connections_lock); |
1601 | hlist_del_rcu(&con->list); | |
1602 | spin_unlock(&connections_lock); | |
948c47e9 AA |
1603 | if (con->othercon) { |
1604 | clean_one_writequeue(con->othercon); | |
4798cbbf | 1605 | call_rcu(&con->othercon->rcu, connection_release); |
948c47e9 | 1606 | } |
0de98432 | 1607 | clean_one_writequeue(con); |
4798cbbf | 1608 | call_rcu(&con->rcu, connection_release); |
5e9ccc37 CC |
1609 | } |
1610 | ||
f0fb83cb | 1611 | static void work_flush(void) |
1612 | { | |
a47666eb | 1613 | int ok, idx; |
f0fb83cb | 1614 | int i; |
f0fb83cb | 1615 | struct connection *con; |
1616 | ||
f0fb83cb | 1617 | do { |
1618 | ok = 1; | |
1619 | foreach_conn(stop_conn); | |
b355516f DW |
1620 | if (recv_workqueue) |
1621 | flush_workqueue(recv_workqueue); | |
1622 | if (send_workqueue) | |
1623 | flush_workqueue(send_workqueue); | |
a47666eb | 1624 | idx = srcu_read_lock(&connections_srcu); |
f0fb83cb | 1625 | for (i = 0; i < CONN_HASH_SIZE && ok; i++) { |
a47666eb AA |
1626 | hlist_for_each_entry_rcu(con, &connection_hash[i], |
1627 | list) { | |
f0fb83cb | 1628 | ok &= test_bit(CF_READ_PENDING, &con->flags); |
8a4abb08 | 1629 | ok &= test_bit(CF_WRITE_PENDING, &con->flags); |
1630 | if (con->othercon) { | |
f0fb83cb | 1631 | ok &= test_bit(CF_READ_PENDING, |
1632 | &con->othercon->flags); | |
8a4abb08 | 1633 | ok &= test_bit(CF_WRITE_PENDING, |
1634 | &con->othercon->flags); | |
1635 | } | |
f0fb83cb | 1636 | } |
1637 | } | |
a47666eb | 1638 | srcu_read_unlock(&connections_srcu, idx); |
f0fb83cb | 1639 | } while (!ok); |
1640 | } | |
1641 | ||
5e9ccc37 CC |
1642 | void dlm_lowcomms_stop(void) |
1643 | { | |
ac33d071 | 1644 | /* Set all the flags to prevent any |
fdda387f PC |
1645 | socket activity. |
1646 | */ | |
513ef596 | 1647 | dlm_allow_conn = 0; |
aa7ab1e2 AA |
1648 | |
1649 | if (recv_workqueue) | |
1650 | flush_workqueue(recv_workqueue); | |
1651 | if (send_workqueue) | |
1652 | flush_workqueue(send_workqueue); | |
1653 | ||
055923bf | 1654 | foreach_conn(shutdown_conn); |
f0fb83cb | 1655 | work_flush(); |
3a8db798 | 1656 | foreach_conn(free_conn); |
1d6e8131 | 1657 | work_stop(); |
043697f0 | 1658 | deinit_local(); |
fdda387f PC |
1659 | } |
1660 | ||
fdda387f PC |
1661 | int dlm_lowcomms_start(void) |
1662 | { | |
6ed7257b PC |
1663 | int error = -EINVAL; |
1664 | struct connection *con; | |
5e9ccc37 CC |
1665 | int i; |
1666 | ||
1667 | for (i = 0; i < CONN_HASH_SIZE; i++) | |
1668 | INIT_HLIST_HEAD(&connection_hash[i]); | |
fdda387f | 1669 | |
6ed7257b PC |
1670 | init_local(); |
1671 | if (!dlm_local_count) { | |
617e82e1 | 1672 | error = -ENOTCONN; |
fdda387f | 1673 | log_print("no local IP address has been set"); |
513ef596 | 1674 | goto fail; |
fdda387f PC |
1675 | } |
1676 | ||
513ef596 DT |
1677 | error = work_start(); |
1678 | if (error) | |
a47666eb | 1679 | goto fail; |
513ef596 DT |
1680 | |
1681 | dlm_allow_conn = 1; | |
fdda387f | 1682 | |
fdda387f | 1683 | /* Start listening */ |
6ed7257b PC |
1684 | if (dlm_config.ci_protocol == 0) |
1685 | error = tcp_listen_for_all(); | |
1686 | else | |
1687 | error = sctp_listen_for_all(); | |
fdda387f PC |
1688 | if (error) |
1689 | goto fail_unlisten; | |
1690 | ||
fdda387f PC |
1691 | return 0; |
1692 | ||
ac33d071 | 1693 | fail_unlisten: |
513ef596 | 1694 | dlm_allow_conn = 0; |
6ed7257b | 1695 | con = nodeid2con(0,0); |
7ae0451e AA |
1696 | if (con) |
1697 | free_conn(con); | |
513ef596 | 1698 | fail: |
fdda387f PC |
1699 | return error; |
1700 | } | |
36b71a8b DT |
1701 | |
1702 | void dlm_lowcomms_exit(void) | |
1703 | { | |
1704 | struct dlm_node_addr *na, *safe; | |
1705 | ||
1706 | spin_lock(&dlm_node_addrs_spin); | |
1707 | list_for_each_entry_safe(na, safe, &dlm_node_addrs, list) { | |
1708 | list_del(&na->list); | |
1709 | while (na->addr_count--) | |
1710 | kfree(na->addr[na->addr_count]); | |
1711 | kfree(na); | |
1712 | } | |
1713 | spin_unlock(&dlm_node_addrs_spin); | |
1714 | } |