]>
Commit | Line | Data |
---|---|---|
fdda387f PC |
1 | /****************************************************************************** |
2 | ******************************************************************************* | |
3 | ** | |
4 | ** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
5e9ccc37 | 5 | ** Copyright (C) 2004-2009 Red Hat, Inc. All rights reserved. |
fdda387f PC |
6 | ** |
7 | ** This copyrighted material is made available to anyone wishing to use, | |
8 | ** modify, copy, or redistribute it subject to the terms and conditions | |
9 | ** of the GNU General Public License v.2. | |
10 | ** | |
11 | ******************************************************************************* | |
12 | ******************************************************************************/ | |
13 | ||
14 | /* | |
15 | * lowcomms.c | |
16 | * | |
17 | * This is the "low-level" comms layer. | |
18 | * | |
19 | * It is responsible for sending/receiving messages | |
20 | * from other nodes in the cluster. | |
21 | * | |
22 | * Cluster nodes are referred to by their nodeids. nodeids are | |
23 | * simply 32 bit numbers to the locking module - if they need to | |
2cf12c0b | 24 | * be expanded for the cluster infrastructure then that is its |
fdda387f PC |
25 | * responsibility. It is this layer's |
26 | * responsibility to resolve these into IP address or | |
27 | * whatever it needs for inter-node communication. | |
28 | * | |
29 | * The comms level is two kernel threads that deal mainly with | |
30 | * the receiving of messages from other nodes and passing them | |
31 | * up to the mid-level comms layer (which understands the | |
32 | * message format) for execution by the locking core, and | |
33 | * a send thread which does all the setting up of connections | |
34 | * to remote nodes and the sending of data. Threads are not allowed | |
35 | * to send their own data because it may cause them to wait in times | |
36 | * of high load. Also, this way, the sending thread can collect together | |
37 | * messages bound for one node and send them in one block. | |
38 | * | |
2cf12c0b | 39 | * lowcomms will choose to use either TCP or SCTP as its transport layer |
6ed7257b | 40 | * depending on the configuration variable 'protocol'. This should be set |
2cf12c0b | 41 | * to 0 (default) for TCP or 1 for SCTP. It should be configured using a |
6ed7257b PC |
42 | * cluster-wide mechanism as it must be the same on all nodes of the cluster |
43 | * for the DLM to function. | |
fdda387f PC |
44 | * |
45 | */ | |
46 | ||
fdda387f PC |
47 | #include <asm/ioctls.h> |
48 | #include <net/sock.h> | |
49 | #include <net/tcp.h> | |
50 | #include <linux/pagemap.h> | |
6ed7257b | 51 | #include <linux/file.h> |
7a936ce7 | 52 | #include <linux/mutex.h> |
6ed7257b | 53 | #include <linux/sctp.h> |
5a0e3ad6 | 54 | #include <linux/slab.h> |
2f2d76cc | 55 | #include <net/sctp/sctp.h> |
44ad532b | 56 | #include <net/ipv6.h> |
fdda387f PC |
57 | |
58 | #include "dlm_internal.h" | |
59 | #include "lowcomms.h" | |
60 | #include "midcomms.h" | |
61 | #include "config.h" | |
62 | ||
6ed7257b | 63 | #define NEEDED_RMEM (4*1024*1024) |
5e9ccc37 | 64 | #define CONN_HASH_SIZE 32 |
6ed7257b | 65 | |
f92c8dd7 BP |
66 | /* Number of messages to send before rescheduling */ |
67 | #define MAX_SEND_MSG_COUNT 25 | |
68 | ||
fdda387f | 69 | struct cbuf { |
ac33d071 PC |
70 | unsigned int base; |
71 | unsigned int len; | |
72 | unsigned int mask; | |
fdda387f PC |
73 | }; |
74 | ||
ac33d071 PC |
75 | static void cbuf_add(struct cbuf *cb, int n) |
76 | { | |
77 | cb->len += n; | |
78 | } | |
fdda387f | 79 | |
ac33d071 PC |
80 | static int cbuf_data(struct cbuf *cb) |
81 | { | |
82 | return ((cb->base + cb->len) & cb->mask); | |
83 | } | |
84 | ||
85 | static void cbuf_init(struct cbuf *cb, int size) | |
86 | { | |
87 | cb->base = cb->len = 0; | |
88 | cb->mask = size-1; | |
89 | } | |
90 | ||
91 | static void cbuf_eat(struct cbuf *cb, int n) | |
92 | { | |
93 | cb->len -= n; | |
94 | cb->base += n; | |
95 | cb->base &= cb->mask; | |
96 | } | |
97 | ||
98 | static bool cbuf_empty(struct cbuf *cb) | |
99 | { | |
100 | return cb->len == 0; | |
101 | } | |
fdda387f | 102 | |
fdda387f PC |
103 | struct connection { |
104 | struct socket *sock; /* NULL if not connected */ | |
105 | uint32_t nodeid; /* So we know who we are in the list */ | |
f1f1c1cc | 106 | struct mutex sock_mutex; |
6ed7257b | 107 | unsigned long flags; |
fdda387f | 108 | #define CF_READ_PENDING 1 |
6ed7257b PC |
109 | #define CF_INIT_PENDING 4 |
110 | #define CF_IS_OTHERCON 5 | |
063c4c99 | 111 | #define CF_CLOSE 6 |
b36930dd | 112 | #define CF_APP_LIMITED 7 |
b2a66629 | 113 | #define CF_CLOSING 8 |
ac33d071 | 114 | struct list_head writequeue; /* List of outgoing writequeue_entries */ |
fdda387f PC |
115 | spinlock_t writequeue_lock; |
116 | int (*rx_action) (struct connection *); /* What to do when active */ | |
6ed7257b | 117 | void (*connect_action) (struct connection *); /* What to do to connect */ |
fdda387f PC |
118 | struct page *rx_page; |
119 | struct cbuf cb; | |
120 | int retries; | |
fdda387f | 121 | #define MAX_CONNECT_RETRIES 3 |
5e9ccc37 | 122 | struct hlist_node list; |
fdda387f | 123 | struct connection *othercon; |
1d6e8131 PC |
124 | struct work_struct rwork; /* Receive workqueue */ |
125 | struct work_struct swork; /* Send workqueue */ | |
fdda387f PC |
126 | }; |
127 | #define sock2con(x) ((struct connection *)(x)->sk_user_data) | |
128 | ||
129 | /* An entry waiting to be sent */ | |
130 | struct writequeue_entry { | |
131 | struct list_head list; | |
132 | struct page *page; | |
133 | int offset; | |
134 | int len; | |
135 | int end; | |
136 | int users; | |
137 | struct connection *con; | |
138 | }; | |
139 | ||
36b71a8b DT |
140 | struct dlm_node_addr { |
141 | struct list_head list; | |
142 | int nodeid; | |
143 | int addr_count; | |
98e1b60e | 144 | int curr_addr_index; |
36b71a8b DT |
145 | struct sockaddr_storage *addr[DLM_MAX_ADDR_COUNT]; |
146 | }; | |
147 | ||
cc661fc9 BP |
148 | static struct listen_sock_callbacks { |
149 | void (*sk_error_report)(struct sock *); | |
150 | void (*sk_data_ready)(struct sock *); | |
151 | void (*sk_state_change)(struct sock *); | |
152 | void (*sk_write_space)(struct sock *); | |
153 | } listen_sock; | |
154 | ||
36b71a8b DT |
155 | static LIST_HEAD(dlm_node_addrs); |
156 | static DEFINE_SPINLOCK(dlm_node_addrs_spin); | |
157 | ||
6ed7257b PC |
158 | static struct sockaddr_storage *dlm_local_addr[DLM_MAX_ADDR_COUNT]; |
159 | static int dlm_local_count; | |
513ef596 | 160 | static int dlm_allow_conn; |
fdda387f | 161 | |
1d6e8131 PC |
162 | /* Work queues */ |
163 | static struct workqueue_struct *recv_workqueue; | |
164 | static struct workqueue_struct *send_workqueue; | |
fdda387f | 165 | |
5e9ccc37 | 166 | static struct hlist_head connection_hash[CONN_HASH_SIZE]; |
7a936ce7 | 167 | static DEFINE_MUTEX(connections_lock); |
c80e7c83 | 168 | static struct kmem_cache *con_cache; |
fdda387f | 169 | |
1d6e8131 PC |
170 | static void process_recv_sockets(struct work_struct *work); |
171 | static void process_send_sockets(struct work_struct *work); | |
fdda387f | 172 | |
5e9ccc37 CC |
173 | |
174 | /* This is deliberately very simple because most clusters have simple | |
175 | sequential nodeids, so we should be able to go straight to a connection | |
176 | struct in the array */ | |
177 | static inline int nodeid_hash(int nodeid) | |
178 | { | |
179 | return nodeid & (CONN_HASH_SIZE-1); | |
180 | } | |
181 | ||
182 | static struct connection *__find_con(int nodeid) | |
183 | { | |
184 | int r; | |
5e9ccc37 CC |
185 | struct connection *con; |
186 | ||
187 | r = nodeid_hash(nodeid); | |
188 | ||
b67bfe0d | 189 | hlist_for_each_entry(con, &connection_hash[r], list) { |
5e9ccc37 CC |
190 | if (con->nodeid == nodeid) |
191 | return con; | |
192 | } | |
193 | return NULL; | |
194 | } | |
195 | ||
6ed7257b PC |
196 | /* |
197 | * If 'allocation' is zero then we don't attempt to create a new | |
198 | * connection structure for this node. | |
199 | */ | |
200 | static struct connection *__nodeid2con(int nodeid, gfp_t alloc) | |
fdda387f PC |
201 | { |
202 | struct connection *con = NULL; | |
6ed7257b | 203 | int r; |
fdda387f | 204 | |
5e9ccc37 | 205 | con = __find_con(nodeid); |
6ed7257b PC |
206 | if (con || !alloc) |
207 | return con; | |
fdda387f | 208 | |
6ed7257b PC |
209 | con = kmem_cache_zalloc(con_cache, alloc); |
210 | if (!con) | |
211 | return NULL; | |
fdda387f | 212 | |
5e9ccc37 CC |
213 | r = nodeid_hash(nodeid); |
214 | hlist_add_head(&con->list, &connection_hash[r]); | |
fdda387f | 215 | |
6ed7257b PC |
216 | con->nodeid = nodeid; |
217 | mutex_init(&con->sock_mutex); | |
218 | INIT_LIST_HEAD(&con->writequeue); | |
219 | spin_lock_init(&con->writequeue_lock); | |
220 | INIT_WORK(&con->swork, process_send_sockets); | |
221 | INIT_WORK(&con->rwork, process_recv_sockets); | |
fdda387f | 222 | |
6ed7257b PC |
223 | /* Setup action pointers for child sockets */ |
224 | if (con->nodeid) { | |
5e9ccc37 | 225 | struct connection *zerocon = __find_con(0); |
fdda387f | 226 | |
6ed7257b PC |
227 | con->connect_action = zerocon->connect_action; |
228 | if (!con->rx_action) | |
229 | con->rx_action = zerocon->rx_action; | |
fdda387f PC |
230 | } |
231 | ||
6ed7257b PC |
232 | return con; |
233 | } | |
234 | ||
5e9ccc37 CC |
235 | /* Loop round all connections */ |
236 | static void foreach_conn(void (*conn_func)(struct connection *c)) | |
237 | { | |
238 | int i; | |
b67bfe0d | 239 | struct hlist_node *n; |
5e9ccc37 CC |
240 | struct connection *con; |
241 | ||
242 | for (i = 0; i < CONN_HASH_SIZE; i++) { | |
b67bfe0d | 243 | hlist_for_each_entry_safe(con, n, &connection_hash[i], list) |
5e9ccc37 | 244 | conn_func(con); |
5e9ccc37 CC |
245 | } |
246 | } | |
247 | ||
6ed7257b PC |
248 | static struct connection *nodeid2con(int nodeid, gfp_t allocation) |
249 | { | |
250 | struct connection *con; | |
251 | ||
7a936ce7 | 252 | mutex_lock(&connections_lock); |
6ed7257b | 253 | con = __nodeid2con(nodeid, allocation); |
7a936ce7 | 254 | mutex_unlock(&connections_lock); |
6ed7257b | 255 | |
fdda387f PC |
256 | return con; |
257 | } | |
258 | ||
36b71a8b DT |
259 | static struct dlm_node_addr *find_node_addr(int nodeid) |
260 | { | |
261 | struct dlm_node_addr *na; | |
262 | ||
263 | list_for_each_entry(na, &dlm_node_addrs, list) { | |
264 | if (na->nodeid == nodeid) | |
265 | return na; | |
266 | } | |
267 | return NULL; | |
268 | } | |
269 | ||
270 | static int addr_compare(struct sockaddr_storage *x, struct sockaddr_storage *y) | |
6ed7257b | 271 | { |
36b71a8b DT |
272 | switch (x->ss_family) { |
273 | case AF_INET: { | |
274 | struct sockaddr_in *sinx = (struct sockaddr_in *)x; | |
275 | struct sockaddr_in *siny = (struct sockaddr_in *)y; | |
276 | if (sinx->sin_addr.s_addr != siny->sin_addr.s_addr) | |
277 | return 0; | |
278 | if (sinx->sin_port != siny->sin_port) | |
279 | return 0; | |
280 | break; | |
281 | } | |
282 | case AF_INET6: { | |
283 | struct sockaddr_in6 *sinx = (struct sockaddr_in6 *)x; | |
284 | struct sockaddr_in6 *siny = (struct sockaddr_in6 *)y; | |
285 | if (!ipv6_addr_equal(&sinx->sin6_addr, &siny->sin6_addr)) | |
286 | return 0; | |
287 | if (sinx->sin6_port != siny->sin6_port) | |
288 | return 0; | |
289 | break; | |
290 | } | |
291 | default: | |
292 | return 0; | |
293 | } | |
294 | return 1; | |
295 | } | |
296 | ||
297 | static int nodeid_to_addr(int nodeid, struct sockaddr_storage *sas_out, | |
98e1b60e | 298 | struct sockaddr *sa_out, bool try_new_addr) |
36b71a8b DT |
299 | { |
300 | struct sockaddr_storage sas; | |
301 | struct dlm_node_addr *na; | |
6ed7257b PC |
302 | |
303 | if (!dlm_local_count) | |
304 | return -1; | |
305 | ||
36b71a8b DT |
306 | spin_lock(&dlm_node_addrs_spin); |
307 | na = find_node_addr(nodeid); | |
98e1b60e | 308 | if (na && na->addr_count) { |
ee44b4bc MRL |
309 | memcpy(&sas, na->addr[na->curr_addr_index], |
310 | sizeof(struct sockaddr_storage)); | |
311 | ||
98e1b60e MC |
312 | if (try_new_addr) { |
313 | na->curr_addr_index++; | |
314 | if (na->curr_addr_index == na->addr_count) | |
315 | na->curr_addr_index = 0; | |
316 | } | |
98e1b60e | 317 | } |
36b71a8b DT |
318 | spin_unlock(&dlm_node_addrs_spin); |
319 | ||
320 | if (!na) | |
321 | return -EEXIST; | |
322 | ||
323 | if (!na->addr_count) | |
324 | return -ENOENT; | |
325 | ||
326 | if (sas_out) | |
327 | memcpy(sas_out, &sas, sizeof(struct sockaddr_storage)); | |
328 | ||
329 | if (!sa_out) | |
330 | return 0; | |
6ed7257b PC |
331 | |
332 | if (dlm_local_addr[0]->ss_family == AF_INET) { | |
36b71a8b DT |
333 | struct sockaddr_in *in4 = (struct sockaddr_in *) &sas; |
334 | struct sockaddr_in *ret4 = (struct sockaddr_in *) sa_out; | |
6ed7257b PC |
335 | ret4->sin_addr.s_addr = in4->sin_addr.s_addr; |
336 | } else { | |
36b71a8b DT |
337 | struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) &sas; |
338 | struct sockaddr_in6 *ret6 = (struct sockaddr_in6 *) sa_out; | |
4e3fd7a0 | 339 | ret6->sin6_addr = in6->sin6_addr; |
6ed7257b PC |
340 | } |
341 | ||
342 | return 0; | |
343 | } | |
344 | ||
36b71a8b DT |
345 | static int addr_to_nodeid(struct sockaddr_storage *addr, int *nodeid) |
346 | { | |
347 | struct dlm_node_addr *na; | |
348 | int rv = -EEXIST; | |
98e1b60e | 349 | int addr_i; |
36b71a8b DT |
350 | |
351 | spin_lock(&dlm_node_addrs_spin); | |
352 | list_for_each_entry(na, &dlm_node_addrs, list) { | |
353 | if (!na->addr_count) | |
354 | continue; | |
355 | ||
98e1b60e MC |
356 | for (addr_i = 0; addr_i < na->addr_count; addr_i++) { |
357 | if (addr_compare(na->addr[addr_i], addr)) { | |
358 | *nodeid = na->nodeid; | |
359 | rv = 0; | |
360 | goto unlock; | |
361 | } | |
362 | } | |
36b71a8b | 363 | } |
98e1b60e | 364 | unlock: |
36b71a8b DT |
365 | spin_unlock(&dlm_node_addrs_spin); |
366 | return rv; | |
367 | } | |
368 | ||
369 | int dlm_lowcomms_addr(int nodeid, struct sockaddr_storage *addr, int len) | |
370 | { | |
371 | struct sockaddr_storage *new_addr; | |
372 | struct dlm_node_addr *new_node, *na; | |
373 | ||
374 | new_node = kzalloc(sizeof(struct dlm_node_addr), GFP_NOFS); | |
375 | if (!new_node) | |
376 | return -ENOMEM; | |
377 | ||
378 | new_addr = kzalloc(sizeof(struct sockaddr_storage), GFP_NOFS); | |
379 | if (!new_addr) { | |
380 | kfree(new_node); | |
381 | return -ENOMEM; | |
382 | } | |
383 | ||
384 | memcpy(new_addr, addr, len); | |
385 | ||
386 | spin_lock(&dlm_node_addrs_spin); | |
387 | na = find_node_addr(nodeid); | |
388 | if (!na) { | |
389 | new_node->nodeid = nodeid; | |
390 | new_node->addr[0] = new_addr; | |
391 | new_node->addr_count = 1; | |
392 | list_add(&new_node->list, &dlm_node_addrs); | |
393 | spin_unlock(&dlm_node_addrs_spin); | |
394 | return 0; | |
395 | } | |
396 | ||
397 | if (na->addr_count >= DLM_MAX_ADDR_COUNT) { | |
398 | spin_unlock(&dlm_node_addrs_spin); | |
399 | kfree(new_addr); | |
400 | kfree(new_node); | |
401 | return -ENOSPC; | |
402 | } | |
403 | ||
404 | na->addr[na->addr_count++] = new_addr; | |
405 | spin_unlock(&dlm_node_addrs_spin); | |
406 | kfree(new_node); | |
407 | return 0; | |
408 | } | |
409 | ||
fdda387f | 410 | /* Data available on socket or listen socket received a connect */ |
676d2369 | 411 | static void lowcomms_data_ready(struct sock *sk) |
fdda387f PC |
412 | { |
413 | struct connection *con = sock2con(sk); | |
afb853fb | 414 | if (con && !test_and_set_bit(CF_READ_PENDING, &con->flags)) |
1d6e8131 | 415 | queue_work(recv_workqueue, &con->rwork); |
fdda387f PC |
416 | } |
417 | ||
418 | static void lowcomms_write_space(struct sock *sk) | |
419 | { | |
420 | struct connection *con = sock2con(sk); | |
421 | ||
b36930dd DM |
422 | if (!con) |
423 | return; | |
424 | ||
425 | clear_bit(SOCK_NOSPACE, &con->sock->flags); | |
426 | ||
427 | if (test_and_clear_bit(CF_APP_LIMITED, &con->flags)) { | |
428 | con->sock->sk->sk_write_pending--; | |
9cd3e072 | 429 | clear_bit(SOCKWQ_ASYNC_NOSPACE, &con->sock->flags); |
b36930dd DM |
430 | } |
431 | ||
01da24d3 | 432 | queue_work(send_workqueue, &con->swork); |
fdda387f PC |
433 | } |
434 | ||
435 | static inline void lowcomms_connect_sock(struct connection *con) | |
436 | { | |
063c4c99 LMB |
437 | if (test_bit(CF_CLOSE, &con->flags)) |
438 | return; | |
61d9102b BP |
439 | queue_work(send_workqueue, &con->swork); |
440 | cond_resched(); | |
fdda387f PC |
441 | } |
442 | ||
443 | static void lowcomms_state_change(struct sock *sk) | |
444 | { | |
ee44b4bc MRL |
445 | /* SCTP layer is not calling sk_data_ready when the connection |
446 | * is done, so we catch the signal through here. Also, it | |
447 | * doesn't switch socket state when entering shutdown, so we | |
448 | * skip the write in that case. | |
449 | */ | |
450 | if (sk->sk_shutdown) { | |
451 | if (sk->sk_shutdown == RCV_SHUTDOWN) | |
452 | lowcomms_data_ready(sk); | |
453 | } else if (sk->sk_state == TCP_ESTABLISHED) { | |
fdda387f | 454 | lowcomms_write_space(sk); |
ee44b4bc | 455 | } |
fdda387f PC |
456 | } |
457 | ||
391fbdc5 CC |
458 | int dlm_lowcomms_connect_node(int nodeid) |
459 | { | |
460 | struct connection *con; | |
461 | ||
462 | if (nodeid == dlm_our_nodeid()) | |
463 | return 0; | |
464 | ||
465 | con = nodeid2con(nodeid, GFP_NOFS); | |
466 | if (!con) | |
467 | return -ENOMEM; | |
468 | lowcomms_connect_sock(con); | |
469 | return 0; | |
470 | } | |
471 | ||
b3a5bbfd BP |
472 | static void lowcomms_error_report(struct sock *sk) |
473 | { | |
b81171cb | 474 | struct connection *con; |
b3a5bbfd | 475 | struct sockaddr_storage saddr; |
1a31833d | 476 | int buflen; |
b81171cb | 477 | void (*orig_report)(struct sock *) = NULL; |
b3a5bbfd | 478 | |
b81171cb BP |
479 | read_lock_bh(&sk->sk_callback_lock); |
480 | con = sock2con(sk); | |
481 | if (con == NULL) | |
482 | goto out; | |
483 | ||
cc661fc9 | 484 | orig_report = listen_sock.sk_error_report; |
1a31833d BP |
485 | if (con->sock == NULL || |
486 | kernel_getpeername(con->sock, (struct sockaddr *)&saddr, &buflen)) { | |
b3a5bbfd BP |
487 | printk_ratelimited(KERN_ERR "dlm: node %d: socket error " |
488 | "sending to node %d, port %d, " | |
489 | "sk_err=%d/%d\n", dlm_our_nodeid(), | |
490 | con->nodeid, dlm_config.ci_tcp_port, | |
491 | sk->sk_err, sk->sk_err_soft); | |
b3a5bbfd BP |
492 | } else if (saddr.ss_family == AF_INET) { |
493 | struct sockaddr_in *sin4 = (struct sockaddr_in *)&saddr; | |
494 | ||
495 | printk_ratelimited(KERN_ERR "dlm: node %d: socket error " | |
496 | "sending to node %d at %pI4, port %d, " | |
497 | "sk_err=%d/%d\n", dlm_our_nodeid(), | |
498 | con->nodeid, &sin4->sin_addr.s_addr, | |
499 | dlm_config.ci_tcp_port, sk->sk_err, | |
500 | sk->sk_err_soft); | |
501 | } else { | |
502 | struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&saddr; | |
503 | ||
504 | printk_ratelimited(KERN_ERR "dlm: node %d: socket error " | |
505 | "sending to node %d at %u.%u.%u.%u, " | |
506 | "port %d, sk_err=%d/%d\n", dlm_our_nodeid(), | |
507 | con->nodeid, sin6->sin6_addr.s6_addr32[0], | |
508 | sin6->sin6_addr.s6_addr32[1], | |
509 | sin6->sin6_addr.s6_addr32[2], | |
510 | sin6->sin6_addr.s6_addr32[3], | |
511 | dlm_config.ci_tcp_port, sk->sk_err, | |
512 | sk->sk_err_soft); | |
513 | } | |
b81171cb BP |
514 | out: |
515 | read_unlock_bh(&sk->sk_callback_lock); | |
516 | if (orig_report) | |
517 | orig_report(sk); | |
518 | } | |
519 | ||
520 | /* Note: sk_callback_lock must be locked before calling this function. */ | |
cc661fc9 | 521 | static void save_listen_callbacks(struct socket *sock) |
b81171cb | 522 | { |
cc661fc9 BP |
523 | struct sock *sk = sock->sk; |
524 | ||
525 | listen_sock.sk_data_ready = sk->sk_data_ready; | |
526 | listen_sock.sk_state_change = sk->sk_state_change; | |
527 | listen_sock.sk_write_space = sk->sk_write_space; | |
528 | listen_sock.sk_error_report = sk->sk_error_report; | |
b81171cb BP |
529 | } |
530 | ||
cc661fc9 | 531 | static void restore_callbacks(struct socket *sock) |
b81171cb | 532 | { |
cc661fc9 BP |
533 | struct sock *sk = sock->sk; |
534 | ||
b81171cb | 535 | write_lock_bh(&sk->sk_callback_lock); |
b81171cb | 536 | sk->sk_user_data = NULL; |
cc661fc9 BP |
537 | sk->sk_data_ready = listen_sock.sk_data_ready; |
538 | sk->sk_state_change = listen_sock.sk_state_change; | |
539 | sk->sk_write_space = listen_sock.sk_write_space; | |
540 | sk->sk_error_report = listen_sock.sk_error_report; | |
b81171cb | 541 | write_unlock_bh(&sk->sk_callback_lock); |
b3a5bbfd BP |
542 | } |
543 | ||
fdda387f | 544 | /* Make a socket active */ |
988419a9 | 545 | static void add_sock(struct socket *sock, struct connection *con) |
fdda387f | 546 | { |
b81171cb BP |
547 | struct sock *sk = sock->sk; |
548 | ||
549 | write_lock_bh(&sk->sk_callback_lock); | |
fdda387f PC |
550 | con->sock = sock; |
551 | ||
b81171cb | 552 | sk->sk_user_data = con; |
fdda387f | 553 | /* Install a data_ready callback */ |
b81171cb BP |
554 | sk->sk_data_ready = lowcomms_data_ready; |
555 | sk->sk_write_space = lowcomms_write_space; | |
556 | sk->sk_state_change = lowcomms_state_change; | |
557 | sk->sk_allocation = GFP_NOFS; | |
558 | sk->sk_error_report = lowcomms_error_report; | |
559 | write_unlock_bh(&sk->sk_callback_lock); | |
fdda387f PC |
560 | } |
561 | ||
6ed7257b | 562 | /* Add the port number to an IPv6 or 4 sockaddr and return the address |
fdda387f PC |
563 | length */ |
564 | static void make_sockaddr(struct sockaddr_storage *saddr, uint16_t port, | |
565 | int *addr_len) | |
566 | { | |
6ed7257b | 567 | saddr->ss_family = dlm_local_addr[0]->ss_family; |
ac33d071 | 568 | if (saddr->ss_family == AF_INET) { |
fdda387f PC |
569 | struct sockaddr_in *in4_addr = (struct sockaddr_in *)saddr; |
570 | in4_addr->sin_port = cpu_to_be16(port); | |
571 | *addr_len = sizeof(struct sockaddr_in); | |
6ed7257b | 572 | memset(&in4_addr->sin_zero, 0, sizeof(in4_addr->sin_zero)); |
ac33d071 | 573 | } else { |
fdda387f PC |
574 | struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)saddr; |
575 | in6_addr->sin6_port = cpu_to_be16(port); | |
576 | *addr_len = sizeof(struct sockaddr_in6); | |
577 | } | |
01c8cab2 | 578 | memset((char *)saddr + *addr_len, 0, sizeof(struct sockaddr_storage) - *addr_len); |
fdda387f PC |
579 | } |
580 | ||
581 | /* Close a remote connection and tidy up */ | |
0d737a8c MRL |
582 | static void close_connection(struct connection *con, bool and_other, |
583 | bool tx, bool rx) | |
fdda387f | 584 | { |
b2a66629 | 585 | bool closing = test_and_set_bit(CF_CLOSING, &con->flags); |
586 | ||
587 | if (tx && !closing && cancel_work_sync(&con->swork)) | |
0d737a8c | 588 | log_print("canceled swork for node %d", con->nodeid); |
b2a66629 | 589 | if (rx && !closing && cancel_work_sync(&con->rwork)) |
0d737a8c | 590 | log_print("canceled rwork for node %d", con->nodeid); |
fdda387f | 591 | |
0d737a8c | 592 | mutex_lock(&con->sock_mutex); |
fdda387f | 593 | if (con->sock) { |
cc661fc9 | 594 | restore_callbacks(con->sock); |
fdda387f PC |
595 | sock_release(con->sock); |
596 | con->sock = NULL; | |
597 | } | |
598 | if (con->othercon && and_other) { | |
ac33d071 | 599 | /* Will only re-enter once. */ |
0d737a8c | 600 | close_connection(con->othercon, false, true, true); |
fdda387f PC |
601 | } |
602 | if (con->rx_page) { | |
603 | __free_page(con->rx_page); | |
604 | con->rx_page = NULL; | |
605 | } | |
9e5f2825 | 606 | |
61d96be0 PC |
607 | con->retries = 0; |
608 | mutex_unlock(&con->sock_mutex); | |
b2a66629 | 609 | clear_bit(CF_CLOSING, &con->flags); |
fdda387f PC |
610 | } |
611 | ||
612 | /* Data received from remote end */ | |
613 | static int receive_from_sock(struct connection *con) | |
614 | { | |
615 | int ret = 0; | |
58addbff AV |
616 | struct msghdr msg = {}; |
617 | struct kvec iov[2]; | |
fdda387f PC |
618 | unsigned len; |
619 | int r; | |
620 | int call_again_soon = 0; | |
58addbff | 621 | int nvec; |
fdda387f | 622 | |
f1f1c1cc | 623 | mutex_lock(&con->sock_mutex); |
fdda387f | 624 | |
a34fbc63 PC |
625 | if (con->sock == NULL) { |
626 | ret = -EAGAIN; | |
627 | goto out_close; | |
628 | } | |
acee4e52 MRL |
629 | if (con->nodeid == 0) { |
630 | ret = -EINVAL; | |
631 | goto out_close; | |
632 | } | |
a34fbc63 | 633 | |
fdda387f PC |
634 | if (con->rx_page == NULL) { |
635 | /* | |
636 | * This doesn't need to be atomic, but I think it should | |
637 | * improve performance if it is. | |
638 | */ | |
639 | con->rx_page = alloc_page(GFP_ATOMIC); | |
640 | if (con->rx_page == NULL) | |
641 | goto out_resched; | |
09cbfeaf | 642 | cbuf_init(&con->cb, PAGE_SIZE); |
fdda387f PC |
643 | } |
644 | ||
fdda387f PC |
645 | /* |
646 | * iov[0] is the bit of the circular buffer between the current end | |
647 | * point (cb.base + cb.len) and the end of the buffer. | |
648 | */ | |
ac33d071 PC |
649 | iov[0].iov_len = con->cb.base - cbuf_data(&con->cb); |
650 | iov[0].iov_base = page_address(con->rx_page) + cbuf_data(&con->cb); | |
89adc934 | 651 | iov[1].iov_len = 0; |
58addbff | 652 | nvec = 1; |
fdda387f PC |
653 | |
654 | /* | |
655 | * iov[1] is the bit of the circular buffer between the start of the | |
656 | * buffer and the start of the currently used section (cb.base) | |
657 | */ | |
ac33d071 | 658 | if (cbuf_data(&con->cb) >= con->cb.base) { |
09cbfeaf | 659 | iov[0].iov_len = PAGE_SIZE - cbuf_data(&con->cb); |
fdda387f PC |
660 | iov[1].iov_len = con->cb.base; |
661 | iov[1].iov_base = page_address(con->rx_page); | |
58addbff | 662 | nvec = 2; |
fdda387f PC |
663 | } |
664 | len = iov[0].iov_len + iov[1].iov_len; | |
665 | ||
58addbff | 666 | r = ret = kernel_recvmsg(con->sock, &msg, iov, nvec, len, |
fdda387f | 667 | MSG_DONTWAIT | MSG_NOSIGNAL); |
fdda387f PC |
668 | if (ret <= 0) |
669 | goto out_close; | |
ee44b4bc MRL |
670 | else if (ret == len) |
671 | call_again_soon = 1; | |
bd44e2b0 | 672 | |
ac33d071 | 673 | cbuf_add(&con->cb, ret); |
fdda387f PC |
674 | ret = dlm_process_incoming_buffer(con->nodeid, |
675 | page_address(con->rx_page), | |
676 | con->cb.base, con->cb.len, | |
09cbfeaf | 677 | PAGE_SIZE); |
fdda387f | 678 | if (ret == -EBADMSG) { |
ee44b4bc MRL |
679 | log_print("lowcomms: addr=%p, base=%u, len=%u, read=%d", |
680 | page_address(con->rx_page), con->cb.base, | |
681 | con->cb.len, r); | |
fdda387f PC |
682 | } |
683 | if (ret < 0) | |
684 | goto out_close; | |
ac33d071 | 685 | cbuf_eat(&con->cb, ret); |
fdda387f | 686 | |
ac33d071 | 687 | if (cbuf_empty(&con->cb) && !call_again_soon) { |
fdda387f PC |
688 | __free_page(con->rx_page); |
689 | con->rx_page = NULL; | |
690 | } | |
691 | ||
fdda387f PC |
692 | if (call_again_soon) |
693 | goto out_resched; | |
f1f1c1cc | 694 | mutex_unlock(&con->sock_mutex); |
ac33d071 | 695 | return 0; |
fdda387f | 696 | |
ac33d071 | 697 | out_resched: |
1d6e8131 PC |
698 | if (!test_and_set_bit(CF_READ_PENDING, &con->flags)) |
699 | queue_work(recv_workqueue, &con->rwork); | |
f1f1c1cc | 700 | mutex_unlock(&con->sock_mutex); |
bd44e2b0 | 701 | return -EAGAIN; |
fdda387f | 702 | |
ac33d071 | 703 | out_close: |
f1f1c1cc | 704 | mutex_unlock(&con->sock_mutex); |
9e5f2825 | 705 | if (ret != -EAGAIN) { |
c553e173 | 706 | close_connection(con, true, true, false); |
fdda387f PC |
707 | /* Reconnect when there is something to send */ |
708 | } | |
a34fbc63 PC |
709 | /* Don't return success if we really got EOF */ |
710 | if (ret == 0) | |
711 | ret = -EAGAIN; | |
fdda387f | 712 | |
fdda387f PC |
713 | return ret; |
714 | } | |
715 | ||
716 | /* Listening socket is busy, accept a connection */ | |
6ed7257b | 717 | static int tcp_accept_from_sock(struct connection *con) |
fdda387f PC |
718 | { |
719 | int result; | |
720 | struct sockaddr_storage peeraddr; | |
721 | struct socket *newsock; | |
722 | int len; | |
723 | int nodeid; | |
724 | struct connection *newcon; | |
bd44e2b0 | 725 | struct connection *addcon; |
fdda387f | 726 | |
513ef596 DT |
727 | mutex_lock(&connections_lock); |
728 | if (!dlm_allow_conn) { | |
729 | mutex_unlock(&connections_lock); | |
730 | return -1; | |
731 | } | |
732 | mutex_unlock(&connections_lock); | |
733 | ||
fdda387f | 734 | memset(&peeraddr, 0, sizeof(peeraddr)); |
1c242853 | 735 | result = sock_create_lite(dlm_local_addr[0]->ss_family, |
eeb1bd5c | 736 | SOCK_STREAM, IPPROTO_TCP, &newsock); |
fdda387f PC |
737 | if (result < 0) |
738 | return -ENOMEM; | |
739 | ||
f1f1c1cc | 740 | mutex_lock_nested(&con->sock_mutex, 0); |
fdda387f PC |
741 | |
742 | result = -ENOTCONN; | |
743 | if (con->sock == NULL) | |
744 | goto accept_err; | |
745 | ||
746 | newsock->type = con->sock->type; | |
747 | newsock->ops = con->sock->ops; | |
748 | ||
cdfbabfb | 749 | result = con->sock->ops->accept(con->sock, newsock, O_NONBLOCK, true); |
fdda387f PC |
750 | if (result < 0) |
751 | goto accept_err; | |
752 | ||
753 | /* Get the connected socket's peer */ | |
754 | memset(&peeraddr, 0, sizeof(peeraddr)); | |
755 | if (newsock->ops->getname(newsock, (struct sockaddr *)&peeraddr, | |
756 | &len, 2)) { | |
757 | result = -ECONNABORTED; | |
758 | goto accept_err; | |
759 | } | |
760 | ||
761 | /* Get the new node's NODEID */ | |
762 | make_sockaddr(&peeraddr, 0, &len); | |
36b71a8b | 763 | if (addr_to_nodeid(&peeraddr, &nodeid)) { |
bcaadf5c | 764 | unsigned char *b=(unsigned char *)&peeraddr; |
617e82e1 | 765 | log_print("connect from non cluster node"); |
bcaadf5c MY |
766 | print_hex_dump_bytes("ss: ", DUMP_PREFIX_NONE, |
767 | b, sizeof(struct sockaddr_storage)); | |
fdda387f | 768 | sock_release(newsock); |
f1f1c1cc | 769 | mutex_unlock(&con->sock_mutex); |
fdda387f PC |
770 | return -1; |
771 | } | |
772 | ||
773 | log_print("got connection from %d", nodeid); | |
774 | ||
775 | /* Check to see if we already have a connection to this node. This | |
776 | * could happen if the two nodes initiate a connection at roughly | |
777 | * the same time and the connections cross on the wire. | |
fdda387f PC |
778 | * In this case we store the incoming one in "othercon" |
779 | */ | |
748285cc | 780 | newcon = nodeid2con(nodeid, GFP_NOFS); |
fdda387f PC |
781 | if (!newcon) { |
782 | result = -ENOMEM; | |
783 | goto accept_err; | |
784 | } | |
f1f1c1cc | 785 | mutex_lock_nested(&newcon->sock_mutex, 1); |
fdda387f | 786 | if (newcon->sock) { |
ac33d071 | 787 | struct connection *othercon = newcon->othercon; |
fdda387f PC |
788 | |
789 | if (!othercon) { | |
748285cc | 790 | othercon = kmem_cache_zalloc(con_cache, GFP_NOFS); |
fdda387f | 791 | if (!othercon) { |
617e82e1 | 792 | log_print("failed to allocate incoming socket"); |
f1f1c1cc | 793 | mutex_unlock(&newcon->sock_mutex); |
fdda387f PC |
794 | result = -ENOMEM; |
795 | goto accept_err; | |
796 | } | |
fdda387f PC |
797 | othercon->nodeid = nodeid; |
798 | othercon->rx_action = receive_from_sock; | |
f1f1c1cc | 799 | mutex_init(&othercon->sock_mutex); |
1d6e8131 PC |
800 | INIT_WORK(&othercon->swork, process_send_sockets); |
801 | INIT_WORK(&othercon->rwork, process_recv_sockets); | |
fdda387f | 802 | set_bit(CF_IS_OTHERCON, &othercon->flags); |
61d96be0 | 803 | } |
c7355827 | 804 | mutex_lock_nested(&othercon->sock_mutex, 2); |
61d96be0 | 805 | if (!othercon->sock) { |
fdda387f | 806 | newcon->othercon = othercon; |
97d84836 PC |
807 | othercon->sock = newsock; |
808 | newsock->sk->sk_user_data = othercon; | |
988419a9 | 809 | add_sock(newsock, othercon); |
97d84836 | 810 | addcon = othercon; |
c7355827 | 811 | mutex_unlock(&othercon->sock_mutex); |
97d84836 PC |
812 | } |
813 | else { | |
814 | printk("Extra connection from node %d attempted\n", nodeid); | |
815 | result = -EAGAIN; | |
c7355827 | 816 | mutex_unlock(&othercon->sock_mutex); |
f4fadb23 | 817 | mutex_unlock(&newcon->sock_mutex); |
97d84836 | 818 | goto accept_err; |
fdda387f | 819 | } |
fdda387f PC |
820 | } |
821 | else { | |
822 | newsock->sk->sk_user_data = newcon; | |
823 | newcon->rx_action = receive_from_sock; | |
3735b4b9 BP |
824 | /* accept copies the sk after we've saved the callbacks, so we |
825 | don't want to save them a second time or comm errors will | |
826 | result in calling sk_error_report recursively. */ | |
988419a9 | 827 | add_sock(newsock, newcon); |
bd44e2b0 | 828 | addcon = newcon; |
fdda387f PC |
829 | } |
830 | ||
f1f1c1cc | 831 | mutex_unlock(&newcon->sock_mutex); |
fdda387f PC |
832 | |
833 | /* | |
834 | * Add it to the active queue in case we got data | |
25985edc | 835 | * between processing the accept adding the socket |
fdda387f PC |
836 | * to the read_sockets list |
837 | */ | |
bd44e2b0 PC |
838 | if (!test_and_set_bit(CF_READ_PENDING, &addcon->flags)) |
839 | queue_work(recv_workqueue, &addcon->rwork); | |
f1f1c1cc | 840 | mutex_unlock(&con->sock_mutex); |
fdda387f PC |
841 | |
842 | return 0; | |
843 | ||
ac33d071 | 844 | accept_err: |
f1f1c1cc | 845 | mutex_unlock(&con->sock_mutex); |
fdda387f PC |
846 | sock_release(newsock); |
847 | ||
848 | if (result != -EAGAIN) | |
617e82e1 | 849 | log_print("error accepting connection from node: %d", result); |
fdda387f PC |
850 | return result; |
851 | } | |
852 | ||
18df8a87 | 853 | static int sctp_accept_from_sock(struct connection *con) |
ee44b4bc MRL |
854 | { |
855 | /* Check that the new node is in the lockspace */ | |
856 | struct sctp_prim prim; | |
857 | int nodeid; | |
858 | int prim_len, ret; | |
859 | int addr_len; | |
860 | struct connection *newcon; | |
861 | struct connection *addcon; | |
862 | struct socket *newsock; | |
863 | ||
864 | mutex_lock(&connections_lock); | |
865 | if (!dlm_allow_conn) { | |
866 | mutex_unlock(&connections_lock); | |
867 | return -1; | |
868 | } | |
869 | mutex_unlock(&connections_lock); | |
870 | ||
871 | mutex_lock_nested(&con->sock_mutex, 0); | |
872 | ||
873 | ret = kernel_accept(con->sock, &newsock, O_NONBLOCK); | |
874 | if (ret < 0) | |
875 | goto accept_err; | |
876 | ||
877 | memset(&prim, 0, sizeof(struct sctp_prim)); | |
878 | prim_len = sizeof(struct sctp_prim); | |
879 | ||
880 | ret = kernel_getsockopt(newsock, IPPROTO_SCTP, SCTP_PRIMARY_ADDR, | |
881 | (char *)&prim, &prim_len); | |
882 | if (ret < 0) { | |
883 | log_print("getsockopt/sctp_primary_addr failed: %d", ret); | |
884 | goto accept_err; | |
885 | } | |
886 | ||
887 | make_sockaddr(&prim.ssp_addr, 0, &addr_len); | |
26c1ec2f WY |
888 | ret = addr_to_nodeid(&prim.ssp_addr, &nodeid); |
889 | if (ret) { | |
ee44b4bc MRL |
890 | unsigned char *b = (unsigned char *)&prim.ssp_addr; |
891 | ||
892 | log_print("reject connect from unknown addr"); | |
893 | print_hex_dump_bytes("ss: ", DUMP_PREFIX_NONE, | |
894 | b, sizeof(struct sockaddr_storage)); | |
895 | goto accept_err; | |
896 | } | |
897 | ||
898 | newcon = nodeid2con(nodeid, GFP_NOFS); | |
899 | if (!newcon) { | |
900 | ret = -ENOMEM; | |
901 | goto accept_err; | |
902 | } | |
903 | ||
904 | mutex_lock_nested(&newcon->sock_mutex, 1); | |
905 | ||
906 | if (newcon->sock) { | |
907 | struct connection *othercon = newcon->othercon; | |
908 | ||
909 | if (!othercon) { | |
910 | othercon = kmem_cache_zalloc(con_cache, GFP_NOFS); | |
911 | if (!othercon) { | |
912 | log_print("failed to allocate incoming socket"); | |
913 | mutex_unlock(&newcon->sock_mutex); | |
914 | ret = -ENOMEM; | |
915 | goto accept_err; | |
916 | } | |
917 | othercon->nodeid = nodeid; | |
918 | othercon->rx_action = receive_from_sock; | |
919 | mutex_init(&othercon->sock_mutex); | |
920 | INIT_WORK(&othercon->swork, process_send_sockets); | |
921 | INIT_WORK(&othercon->rwork, process_recv_sockets); | |
922 | set_bit(CF_IS_OTHERCON, &othercon->flags); | |
923 | } | |
c7355827 | 924 | mutex_lock_nested(&othercon->sock_mutex, 2); |
ee44b4bc MRL |
925 | if (!othercon->sock) { |
926 | newcon->othercon = othercon; | |
927 | othercon->sock = newsock; | |
928 | newsock->sk->sk_user_data = othercon; | |
988419a9 | 929 | add_sock(newsock, othercon); |
ee44b4bc | 930 | addcon = othercon; |
c7355827 | 931 | mutex_unlock(&othercon->sock_mutex); |
ee44b4bc MRL |
932 | } else { |
933 | printk("Extra connection from node %d attempted\n", nodeid); | |
934 | ret = -EAGAIN; | |
c7355827 | 935 | mutex_unlock(&othercon->sock_mutex); |
ee44b4bc MRL |
936 | mutex_unlock(&newcon->sock_mutex); |
937 | goto accept_err; | |
938 | } | |
939 | } else { | |
940 | newsock->sk->sk_user_data = newcon; | |
941 | newcon->rx_action = receive_from_sock; | |
988419a9 | 942 | add_sock(newsock, newcon); |
ee44b4bc MRL |
943 | addcon = newcon; |
944 | } | |
945 | ||
946 | log_print("connected to %d", nodeid); | |
947 | ||
948 | mutex_unlock(&newcon->sock_mutex); | |
949 | ||
950 | /* | |
951 | * Add it to the active queue in case we got data | |
952 | * between processing the accept adding the socket | |
953 | * to the read_sockets list | |
954 | */ | |
955 | if (!test_and_set_bit(CF_READ_PENDING, &addcon->flags)) | |
956 | queue_work(recv_workqueue, &addcon->rwork); | |
957 | mutex_unlock(&con->sock_mutex); | |
958 | ||
959 | return 0; | |
960 | ||
961 | accept_err: | |
962 | mutex_unlock(&con->sock_mutex); | |
963 | if (newsock) | |
964 | sock_release(newsock); | |
965 | if (ret != -EAGAIN) | |
966 | log_print("error accepting connection from node: %d", ret); | |
967 | ||
968 | return ret; | |
969 | } | |
970 | ||
6ed7257b PC |
971 | static void free_entry(struct writequeue_entry *e) |
972 | { | |
973 | __free_page(e->page); | |
974 | kfree(e); | |
975 | } | |
976 | ||
5d689871 MC |
977 | /* |
978 | * writequeue_entry_complete - try to delete and free write queue entry | |
979 | * @e: write queue entry to try to delete | |
980 | * @completed: bytes completed | |
981 | * | |
982 | * writequeue_lock must be held. | |
983 | */ | |
984 | static void writequeue_entry_complete(struct writequeue_entry *e, int completed) | |
985 | { | |
986 | e->offset += completed; | |
987 | e->len -= completed; | |
988 | ||
989 | if (e->len == 0 && e->users == 0) { | |
990 | list_del(&e->list); | |
991 | free_entry(e); | |
992 | } | |
993 | } | |
994 | ||
ee44b4bc MRL |
995 | /* |
996 | * sctp_bind_addrs - bind a SCTP socket to all our addresses | |
997 | */ | |
998 | static int sctp_bind_addrs(struct connection *con, uint16_t port) | |
999 | { | |
1000 | struct sockaddr_storage localaddr; | |
1001 | int i, addr_len, result = 0; | |
1002 | ||
1003 | for (i = 0; i < dlm_local_count; i++) { | |
1004 | memcpy(&localaddr, dlm_local_addr[i], sizeof(localaddr)); | |
1005 | make_sockaddr(&localaddr, port, &addr_len); | |
1006 | ||
1007 | if (!i) | |
1008 | result = kernel_bind(con->sock, | |
1009 | (struct sockaddr *)&localaddr, | |
1010 | addr_len); | |
1011 | else | |
1012 | result = kernel_setsockopt(con->sock, SOL_SCTP, | |
1013 | SCTP_SOCKOPT_BINDX_ADD, | |
1014 | (char *)&localaddr, addr_len); | |
1015 | ||
1016 | if (result < 0) { | |
1017 | log_print("Can't bind to %d addr number %d, %d.\n", | |
1018 | port, i + 1, result); | |
1019 | break; | |
1020 | } | |
1021 | } | |
1022 | return result; | |
1023 | } | |
1024 | ||
6ed7257b PC |
1025 | /* Initiate an SCTP association. |
1026 | This is a special case of send_to_sock() in that we don't yet have a | |
1027 | peeled-off socket for this association, so we use the listening socket | |
1028 | and add the primary IP address of the remote node. | |
1029 | */ | |
ee44b4bc | 1030 | static void sctp_connect_to_sock(struct connection *con) |
6ed7257b | 1031 | { |
ee44b4bc MRL |
1032 | struct sockaddr_storage daddr; |
1033 | int one = 1; | |
1034 | int result; | |
1035 | int addr_len; | |
1036 | struct socket *sock; | |
1037 | ||
1038 | if (con->nodeid == 0) { | |
1039 | log_print("attempt to connect sock 0 foiled"); | |
1040 | return; | |
1041 | } | |
6ed7257b | 1042 | |
5d689871 | 1043 | mutex_lock(&con->sock_mutex); |
6ed7257b | 1044 | |
ee44b4bc MRL |
1045 | /* Some odd races can cause double-connects, ignore them */ |
1046 | if (con->retries++ > MAX_CONNECT_RETRIES) | |
1047 | goto out; | |
1048 | ||
1049 | if (con->sock) { | |
1050 | log_print("node %d already connected.", con->nodeid); | |
1051 | goto out; | |
1052 | } | |
1053 | ||
1054 | memset(&daddr, 0, sizeof(daddr)); | |
1055 | result = nodeid_to_addr(con->nodeid, &daddr, NULL, true); | |
1056 | if (result < 0) { | |
6ed7257b | 1057 | log_print("no address for nodeid %d", con->nodeid); |
ee44b4bc | 1058 | goto out; |
6ed7257b | 1059 | } |
6ed7257b | 1060 | |
ee44b4bc MRL |
1061 | /* Create a socket to communicate with */ |
1062 | result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family, | |
1063 | SOCK_STREAM, IPPROTO_SCTP, &sock); | |
1064 | if (result < 0) | |
1065 | goto socket_err; | |
6ed7257b | 1066 | |
ee44b4bc MRL |
1067 | sock->sk->sk_user_data = con; |
1068 | con->rx_action = receive_from_sock; | |
1069 | con->connect_action = sctp_connect_to_sock; | |
988419a9 | 1070 | add_sock(sock, con); |
6ed7257b | 1071 | |
ee44b4bc MRL |
1072 | /* Bind to all addresses. */ |
1073 | if (sctp_bind_addrs(con, 0)) | |
1074 | goto bind_err; | |
6ed7257b | 1075 | |
ee44b4bc | 1076 | make_sockaddr(&daddr, dlm_config.ci_tcp_port, &addr_len); |
6ed7257b | 1077 | |
ee44b4bc | 1078 | log_print("connecting to %d", con->nodeid); |
6ed7257b | 1079 | |
ee44b4bc MRL |
1080 | /* Turn off Nagle's algorithm */ |
1081 | kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY, (char *)&one, | |
1082 | sizeof(one)); | |
6ed7257b | 1083 | |
ee44b4bc MRL |
1084 | result = sock->ops->connect(sock, (struct sockaddr *)&daddr, addr_len, |
1085 | O_NONBLOCK); | |
1086 | if (result == -EINPROGRESS) | |
1087 | result = 0; | |
1088 | if (result == 0) | |
1089 | goto out; | |
98e1b60e | 1090 | |
ee44b4bc MRL |
1091 | bind_err: |
1092 | con->sock = NULL; | |
1093 | sock_release(sock); | |
6ed7257b | 1094 | |
ee44b4bc MRL |
1095 | socket_err: |
1096 | /* | |
1097 | * Some errors are fatal and this list might need adjusting. For other | |
1098 | * errors we try again until the max number of retries is reached. | |
1099 | */ | |
1100 | if (result != -EHOSTUNREACH && | |
1101 | result != -ENETUNREACH && | |
1102 | result != -ENETDOWN && | |
1103 | result != -EINVAL && | |
1104 | result != -EPROTONOSUPPORT) { | |
1105 | log_print("connect %d try %d error %d", con->nodeid, | |
1106 | con->retries, result); | |
1107 | mutex_unlock(&con->sock_mutex); | |
1108 | msleep(1000); | |
ee44b4bc MRL |
1109 | lowcomms_connect_sock(con); |
1110 | return; | |
6ed7257b | 1111 | } |
5d689871 | 1112 | |
ee44b4bc | 1113 | out: |
5d689871 | 1114 | mutex_unlock(&con->sock_mutex); |
6ed7257b PC |
1115 | } |
1116 | ||
fdda387f | 1117 | /* Connect a new socket to its peer */ |
6ed7257b | 1118 | static void tcp_connect_to_sock(struct connection *con) |
fdda387f | 1119 | { |
6bd8feda | 1120 | struct sockaddr_storage saddr, src_addr; |
fdda387f | 1121 | int addr_len; |
a89d63a1 | 1122 | struct socket *sock = NULL; |
cb2d45da | 1123 | int one = 1; |
36b71a8b | 1124 | int result; |
fdda387f PC |
1125 | |
1126 | if (con->nodeid == 0) { | |
1127 | log_print("attempt to connect sock 0 foiled"); | |
ac33d071 | 1128 | return; |
fdda387f PC |
1129 | } |
1130 | ||
f1f1c1cc | 1131 | mutex_lock(&con->sock_mutex); |
fdda387f PC |
1132 | if (con->retries++ > MAX_CONNECT_RETRIES) |
1133 | goto out; | |
1134 | ||
1135 | /* Some odd races can cause double-connects, ignore them */ | |
36b71a8b | 1136 | if (con->sock) |
fdda387f | 1137 | goto out; |
fdda387f PC |
1138 | |
1139 | /* Create a socket to communicate with */ | |
eeb1bd5c EB |
1140 | result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family, |
1141 | SOCK_STREAM, IPPROTO_TCP, &sock); | |
fdda387f PC |
1142 | if (result < 0) |
1143 | goto out_err; | |
1144 | ||
1145 | memset(&saddr, 0, sizeof(saddr)); | |
98e1b60e | 1146 | result = nodeid_to_addr(con->nodeid, &saddr, NULL, false); |
36b71a8b DT |
1147 | if (result < 0) { |
1148 | log_print("no address for nodeid %d", con->nodeid); | |
ac33d071 | 1149 | goto out_err; |
36b71a8b | 1150 | } |
fdda387f PC |
1151 | |
1152 | sock->sk->sk_user_data = con; | |
1153 | con->rx_action = receive_from_sock; | |
6ed7257b | 1154 | con->connect_action = tcp_connect_to_sock; |
988419a9 | 1155 | add_sock(sock, con); |
fdda387f | 1156 | |
6bd8feda LH |
1157 | /* Bind to our cluster-known address connecting to avoid |
1158 | routing problems */ | |
1159 | memcpy(&src_addr, dlm_local_addr[0], sizeof(src_addr)); | |
1160 | make_sockaddr(&src_addr, 0, &addr_len); | |
1161 | result = sock->ops->bind(sock, (struct sockaddr *) &src_addr, | |
1162 | addr_len); | |
1163 | if (result < 0) { | |
1164 | log_print("could not bind for connect: %d", result); | |
1165 | /* This *may* not indicate a critical error */ | |
1166 | } | |
1167 | ||
68c817a1 | 1168 | make_sockaddr(&saddr, dlm_config.ci_tcp_port, &addr_len); |
fdda387f | 1169 | |
fdda387f | 1170 | log_print("connecting to %d", con->nodeid); |
cb2d45da DT |
1171 | |
1172 | /* Turn off Nagle's algorithm */ | |
1173 | kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY, (char *)&one, | |
1174 | sizeof(one)); | |
1175 | ||
36b71a8b | 1176 | result = sock->ops->connect(sock, (struct sockaddr *)&saddr, addr_len, |
ac33d071 | 1177 | O_NONBLOCK); |
fdda387f PC |
1178 | if (result == -EINPROGRESS) |
1179 | result = 0; | |
ac33d071 PC |
1180 | if (result == 0) |
1181 | goto out; | |
fdda387f | 1182 | |
ac33d071 | 1183 | out_err: |
fdda387f PC |
1184 | if (con->sock) { |
1185 | sock_release(con->sock); | |
1186 | con->sock = NULL; | |
a89d63a1 CD |
1187 | } else if (sock) { |
1188 | sock_release(sock); | |
fdda387f PC |
1189 | } |
1190 | /* | |
1191 | * Some errors are fatal and this list might need adjusting. For other | |
1192 | * errors we try again until the max number of retries is reached. | |
1193 | */ | |
36b71a8b DT |
1194 | if (result != -EHOSTUNREACH && |
1195 | result != -ENETUNREACH && | |
1196 | result != -ENETDOWN && | |
1197 | result != -EINVAL && | |
1198 | result != -EPROTONOSUPPORT) { | |
1199 | log_print("connect %d try %d error %d", con->nodeid, | |
1200 | con->retries, result); | |
1201 | mutex_unlock(&con->sock_mutex); | |
1202 | msleep(1000); | |
fdda387f | 1203 | lowcomms_connect_sock(con); |
36b71a8b | 1204 | return; |
fdda387f | 1205 | } |
ac33d071 | 1206 | out: |
f1f1c1cc | 1207 | mutex_unlock(&con->sock_mutex); |
ac33d071 | 1208 | return; |
fdda387f PC |
1209 | } |
1210 | ||
6ed7257b PC |
1211 | static struct socket *tcp_create_listen_sock(struct connection *con, |
1212 | struct sockaddr_storage *saddr) | |
fdda387f | 1213 | { |
ac33d071 | 1214 | struct socket *sock = NULL; |
fdda387f PC |
1215 | int result = 0; |
1216 | int one = 1; | |
1217 | int addr_len; | |
1218 | ||
6ed7257b | 1219 | if (dlm_local_addr[0]->ss_family == AF_INET) |
fdda387f PC |
1220 | addr_len = sizeof(struct sockaddr_in); |
1221 | else | |
1222 | addr_len = sizeof(struct sockaddr_in6); | |
1223 | ||
1224 | /* Create a socket to communicate with */ | |
eeb1bd5c EB |
1225 | result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family, |
1226 | SOCK_STREAM, IPPROTO_TCP, &sock); | |
fdda387f | 1227 | if (result < 0) { |
617e82e1 | 1228 | log_print("Can't create listening comms socket"); |
fdda387f PC |
1229 | goto create_out; |
1230 | } | |
1231 | ||
cb2d45da DT |
1232 | /* Turn off Nagle's algorithm */ |
1233 | kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY, (char *)&one, | |
1234 | sizeof(one)); | |
1235 | ||
6ed7257b PC |
1236 | result = kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEADDR, |
1237 | (char *)&one, sizeof(one)); | |
1238 | ||
fdda387f | 1239 | if (result < 0) { |
617e82e1 | 1240 | log_print("Failed to set SO_REUSEADDR on socket: %d", result); |
fdda387f | 1241 | } |
b81171cb | 1242 | sock->sk->sk_user_data = con; |
cc661fc9 | 1243 | save_listen_callbacks(sock); |
6ed7257b PC |
1244 | con->rx_action = tcp_accept_from_sock; |
1245 | con->connect_action = tcp_connect_to_sock; | |
fdda387f PC |
1246 | |
1247 | /* Bind to our port */ | |
68c817a1 | 1248 | make_sockaddr(saddr, dlm_config.ci_tcp_port, &addr_len); |
fdda387f PC |
1249 | result = sock->ops->bind(sock, (struct sockaddr *) saddr, addr_len); |
1250 | if (result < 0) { | |
617e82e1 | 1251 | log_print("Can't bind to port %d", dlm_config.ci_tcp_port); |
fdda387f PC |
1252 | sock_release(sock); |
1253 | sock = NULL; | |
1254 | con->sock = NULL; | |
1255 | goto create_out; | |
1256 | } | |
6ed7257b | 1257 | result = kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE, |
ac33d071 | 1258 | (char *)&one, sizeof(one)); |
fdda387f | 1259 | if (result < 0) { |
617e82e1 | 1260 | log_print("Set keepalive failed: %d", result); |
fdda387f PC |
1261 | } |
1262 | ||
1263 | result = sock->ops->listen(sock, 5); | |
1264 | if (result < 0) { | |
617e82e1 | 1265 | log_print("Can't listen on port %d", dlm_config.ci_tcp_port); |
fdda387f PC |
1266 | sock_release(sock); |
1267 | sock = NULL; | |
1268 | goto create_out; | |
1269 | } | |
1270 | ||
ac33d071 | 1271 | create_out: |
fdda387f PC |
1272 | return sock; |
1273 | } | |
1274 | ||
6ed7257b PC |
1275 | /* Get local addresses */ |
1276 | static void init_local(void) | |
1277 | { | |
1278 | struct sockaddr_storage sas, *addr; | |
1279 | int i; | |
1280 | ||
30d3a237 | 1281 | dlm_local_count = 0; |
1b189b88 | 1282 | for (i = 0; i < DLM_MAX_ADDR_COUNT; i++) { |
6ed7257b PC |
1283 | if (dlm_our_addr(&sas, i)) |
1284 | break; | |
1285 | ||
5c93f56f | 1286 | addr = kmemdup(&sas, sizeof(*addr), GFP_NOFS); |
6ed7257b PC |
1287 | if (!addr) |
1288 | break; | |
6ed7257b PC |
1289 | dlm_local_addr[dlm_local_count++] = addr; |
1290 | } | |
1291 | } | |
1292 | ||
6ed7257b PC |
1293 | /* Initialise SCTP socket and bind to all interfaces */ |
1294 | static int sctp_listen_for_all(void) | |
1295 | { | |
1296 | struct socket *sock = NULL; | |
ee44b4bc | 1297 | int result = -EINVAL; |
573c24c4 | 1298 | struct connection *con = nodeid2con(0, GFP_NOFS); |
6ed7257b | 1299 | int bufsize = NEEDED_RMEM; |
86e92ad2 | 1300 | int one = 1; |
6ed7257b PC |
1301 | |
1302 | if (!con) | |
1303 | return -ENOMEM; | |
1304 | ||
1305 | log_print("Using SCTP for communications"); | |
1306 | ||
eeb1bd5c | 1307 | result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family, |
ee44b4bc | 1308 | SOCK_STREAM, IPPROTO_SCTP, &sock); |
6ed7257b PC |
1309 | if (result < 0) { |
1310 | log_print("Can't create comms socket, check SCTP is loaded"); | |
1311 | goto out; | |
1312 | } | |
1313 | ||
df61c952 | 1314 | result = kernel_setsockopt(sock, SOL_SOCKET, SO_RCVBUFFORCE, |
6ed7257b PC |
1315 | (char *)&bufsize, sizeof(bufsize)); |
1316 | if (result) | |
617e82e1 | 1317 | log_print("Error increasing buffer space on socket %d", result); |
6ed7257b | 1318 | |
86e92ad2 MC |
1319 | result = kernel_setsockopt(sock, SOL_SCTP, SCTP_NODELAY, (char *)&one, |
1320 | sizeof(one)); | |
1321 | if (result < 0) | |
1322 | log_print("Could not set SCTP NODELAY error %d\n", result); | |
1323 | ||
b81171cb | 1324 | write_lock_bh(&sock->sk->sk_callback_lock); |
6ed7257b PC |
1325 | /* Init con struct */ |
1326 | sock->sk->sk_user_data = con; | |
cc661fc9 | 1327 | save_listen_callbacks(sock); |
6ed7257b PC |
1328 | con->sock = sock; |
1329 | con->sock->sk->sk_data_ready = lowcomms_data_ready; | |
ee44b4bc MRL |
1330 | con->rx_action = sctp_accept_from_sock; |
1331 | con->connect_action = sctp_connect_to_sock; | |
6ed7257b | 1332 | |
b81171cb BP |
1333 | write_unlock_bh(&sock->sk->sk_callback_lock); |
1334 | ||
ee44b4bc MRL |
1335 | /* Bind to all addresses. */ |
1336 | if (sctp_bind_addrs(con, dlm_config.ci_tcp_port)) | |
1337 | goto create_delsock; | |
6ed7257b PC |
1338 | |
1339 | result = sock->ops->listen(sock, 5); | |
1340 | if (result < 0) { | |
1341 | log_print("Can't set socket listening"); | |
1342 | goto create_delsock; | |
1343 | } | |
1344 | ||
1345 | return 0; | |
1346 | ||
1347 | create_delsock: | |
1348 | sock_release(sock); | |
1349 | con->sock = NULL; | |
1350 | out: | |
1351 | return result; | |
1352 | } | |
1353 | ||
1354 | static int tcp_listen_for_all(void) | |
fdda387f PC |
1355 | { |
1356 | struct socket *sock = NULL; | |
573c24c4 | 1357 | struct connection *con = nodeid2con(0, GFP_NOFS); |
fdda387f PC |
1358 | int result = -EINVAL; |
1359 | ||
6ed7257b PC |
1360 | if (!con) |
1361 | return -ENOMEM; | |
1362 | ||
fdda387f | 1363 | /* We don't support multi-homed hosts */ |
6ed7257b | 1364 | if (dlm_local_addr[1] != NULL) { |
617e82e1 DT |
1365 | log_print("TCP protocol can't handle multi-homed hosts, " |
1366 | "try SCTP"); | |
6ed7257b PC |
1367 | return -EINVAL; |
1368 | } | |
1369 | ||
1370 | log_print("Using TCP for communications"); | |
1371 | ||
6ed7257b | 1372 | sock = tcp_create_listen_sock(con, dlm_local_addr[0]); |
fdda387f | 1373 | if (sock) { |
988419a9 | 1374 | add_sock(sock, con); |
fdda387f PC |
1375 | result = 0; |
1376 | } | |
1377 | else { | |
1378 | result = -EADDRINUSE; | |
1379 | } | |
1380 | ||
1381 | return result; | |
1382 | } | |
1383 | ||
1384 | ||
1385 | ||
1386 | static struct writequeue_entry *new_writequeue_entry(struct connection *con, | |
1387 | gfp_t allocation) | |
1388 | { | |
1389 | struct writequeue_entry *entry; | |
1390 | ||
1391 | entry = kmalloc(sizeof(struct writequeue_entry), allocation); | |
1392 | if (!entry) | |
1393 | return NULL; | |
1394 | ||
1395 | entry->page = alloc_page(allocation); | |
1396 | if (!entry->page) { | |
1397 | kfree(entry); | |
1398 | return NULL; | |
1399 | } | |
1400 | ||
1401 | entry->offset = 0; | |
1402 | entry->len = 0; | |
1403 | entry->end = 0; | |
1404 | entry->users = 0; | |
1405 | entry->con = con; | |
1406 | ||
1407 | return entry; | |
1408 | } | |
1409 | ||
617e82e1 | 1410 | void *dlm_lowcomms_get_buffer(int nodeid, int len, gfp_t allocation, char **ppc) |
fdda387f PC |
1411 | { |
1412 | struct connection *con; | |
1413 | struct writequeue_entry *e; | |
1414 | int offset = 0; | |
fdda387f | 1415 | |
fdda387f PC |
1416 | con = nodeid2con(nodeid, allocation); |
1417 | if (!con) | |
1418 | return NULL; | |
1419 | ||
4edde74e | 1420 | spin_lock(&con->writequeue_lock); |
fdda387f | 1421 | e = list_entry(con->writequeue.prev, struct writequeue_entry, list); |
ac33d071 | 1422 | if ((&e->list == &con->writequeue) || |
09cbfeaf | 1423 | (PAGE_SIZE - e->end < len)) { |
fdda387f PC |
1424 | e = NULL; |
1425 | } else { | |
1426 | offset = e->end; | |
1427 | e->end += len; | |
eeee2b5f | 1428 | e->users++; |
fdda387f PC |
1429 | } |
1430 | spin_unlock(&con->writequeue_lock); | |
1431 | ||
1432 | if (e) { | |
ac33d071 | 1433 | got_one: |
fdda387f PC |
1434 | *ppc = page_address(e->page) + offset; |
1435 | return e; | |
1436 | } | |
1437 | ||
1438 | e = new_writequeue_entry(con, allocation); | |
1439 | if (e) { | |
1440 | spin_lock(&con->writequeue_lock); | |
1441 | offset = e->end; | |
1442 | e->end += len; | |
eeee2b5f | 1443 | e->users++; |
fdda387f PC |
1444 | list_add_tail(&e->list, &con->writequeue); |
1445 | spin_unlock(&con->writequeue_lock); | |
1446 | goto got_one; | |
1447 | } | |
1448 | return NULL; | |
1449 | } | |
1450 | ||
1451 | void dlm_lowcomms_commit_buffer(void *mh) | |
1452 | { | |
1453 | struct writequeue_entry *e = (struct writequeue_entry *)mh; | |
1454 | struct connection *con = e->con; | |
1455 | int users; | |
1456 | ||
4edde74e | 1457 | spin_lock(&con->writequeue_lock); |
fdda387f PC |
1458 | users = --e->users; |
1459 | if (users) | |
1460 | goto out; | |
1461 | e->len = e->end - e->offset; | |
fdda387f PC |
1462 | spin_unlock(&con->writequeue_lock); |
1463 | ||
01da24d3 | 1464 | queue_work(send_workqueue, &con->swork); |
fdda387f PC |
1465 | return; |
1466 | ||
ac33d071 | 1467 | out: |
fdda387f PC |
1468 | spin_unlock(&con->writequeue_lock); |
1469 | return; | |
1470 | } | |
1471 | ||
fdda387f | 1472 | /* Send a message */ |
ac33d071 | 1473 | static void send_to_sock(struct connection *con) |
fdda387f PC |
1474 | { |
1475 | int ret = 0; | |
fdda387f PC |
1476 | const int msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL; |
1477 | struct writequeue_entry *e; | |
1478 | int len, offset; | |
f92c8dd7 | 1479 | int count = 0; |
fdda387f | 1480 | |
f1f1c1cc | 1481 | mutex_lock(&con->sock_mutex); |
fdda387f PC |
1482 | if (con->sock == NULL) |
1483 | goto out_connect; | |
1484 | ||
fdda387f PC |
1485 | spin_lock(&con->writequeue_lock); |
1486 | for (;;) { | |
1487 | e = list_entry(con->writequeue.next, struct writequeue_entry, | |
1488 | list); | |
1489 | if ((struct list_head *) e == &con->writequeue) | |
1490 | break; | |
1491 | ||
1492 | len = e->len; | |
1493 | offset = e->offset; | |
1494 | BUG_ON(len == 0 && e->users == 0); | |
1495 | spin_unlock(&con->writequeue_lock); | |
1496 | ||
1497 | ret = 0; | |
1498 | if (len) { | |
1329e3f2 PB |
1499 | ret = kernel_sendpage(con->sock, e->page, offset, len, |
1500 | msg_flags); | |
d66f8277 | 1501 | if (ret == -EAGAIN || ret == 0) { |
b36930dd | 1502 | if (ret == -EAGAIN && |
9cd3e072 | 1503 | test_bit(SOCKWQ_ASYNC_NOSPACE, &con->sock->flags) && |
b36930dd DM |
1504 | !test_and_set_bit(CF_APP_LIMITED, &con->flags)) { |
1505 | /* Notify TCP that we're limited by the | |
1506 | * application window size. | |
1507 | */ | |
1508 | set_bit(SOCK_NOSPACE, &con->sock->flags); | |
1509 | con->sock->sk->sk_write_pending++; | |
1510 | } | |
d66f8277 | 1511 | cond_resched(); |
fdda387f | 1512 | goto out; |
9c5bef58 | 1513 | } else if (ret < 0) |
fdda387f | 1514 | goto send_error; |
d66f8277 | 1515 | } |
f92c8dd7 BP |
1516 | |
1517 | /* Don't starve people filling buffers */ | |
1518 | if (++count >= MAX_SEND_MSG_COUNT) { | |
ac33d071 | 1519 | cond_resched(); |
f92c8dd7 BP |
1520 | count = 0; |
1521 | } | |
fdda387f PC |
1522 | |
1523 | spin_lock(&con->writequeue_lock); | |
5d689871 | 1524 | writequeue_entry_complete(e, ret); |
fdda387f PC |
1525 | } |
1526 | spin_unlock(&con->writequeue_lock); | |
ac33d071 | 1527 | out: |
f1f1c1cc | 1528 | mutex_unlock(&con->sock_mutex); |
ac33d071 | 1529 | return; |
fdda387f | 1530 | |
ac33d071 | 1531 | send_error: |
f1f1c1cc | 1532 | mutex_unlock(&con->sock_mutex); |
c553e173 | 1533 | close_connection(con, true, false, true); |
01da24d3 BP |
1534 | /* Requeue the send work. When the work daemon runs again, it will try |
1535 | a new connection, then call this function again. */ | |
1536 | queue_work(send_workqueue, &con->swork); | |
ac33d071 | 1537 | return; |
fdda387f | 1538 | |
ac33d071 | 1539 | out_connect: |
f1f1c1cc | 1540 | mutex_unlock(&con->sock_mutex); |
01da24d3 BP |
1541 | queue_work(send_workqueue, &con->swork); |
1542 | cond_resched(); | |
fdda387f PC |
1543 | } |
1544 | ||
1545 | static void clean_one_writequeue(struct connection *con) | |
1546 | { | |
5e9ccc37 | 1547 | struct writequeue_entry *e, *safe; |
fdda387f PC |
1548 | |
1549 | spin_lock(&con->writequeue_lock); | |
5e9ccc37 | 1550 | list_for_each_entry_safe(e, safe, &con->writequeue, list) { |
fdda387f PC |
1551 | list_del(&e->list); |
1552 | free_entry(e); | |
1553 | } | |
1554 | spin_unlock(&con->writequeue_lock); | |
1555 | } | |
1556 | ||
1557 | /* Called from recovery when it knows that a node has | |
1558 | left the cluster */ | |
1559 | int dlm_lowcomms_close(int nodeid) | |
1560 | { | |
1561 | struct connection *con; | |
36b71a8b | 1562 | struct dlm_node_addr *na; |
fdda387f | 1563 | |
fdda387f PC |
1564 | log_print("closing connection to node %d", nodeid); |
1565 | con = nodeid2con(nodeid, 0); | |
1566 | if (con) { | |
063c4c99 | 1567 | set_bit(CF_CLOSE, &con->flags); |
0d737a8c | 1568 | close_connection(con, true, true, true); |
fdda387f | 1569 | clean_one_writequeue(con); |
fdda387f | 1570 | } |
36b71a8b DT |
1571 | |
1572 | spin_lock(&dlm_node_addrs_spin); | |
1573 | na = find_node_addr(nodeid); | |
1574 | if (na) { | |
1575 | list_del(&na->list); | |
1576 | while (na->addr_count--) | |
1577 | kfree(na->addr[na->addr_count]); | |
1578 | kfree(na); | |
1579 | } | |
1580 | spin_unlock(&dlm_node_addrs_spin); | |
1581 | ||
fdda387f | 1582 | return 0; |
fdda387f PC |
1583 | } |
1584 | ||
6ed7257b | 1585 | /* Receive workqueue function */ |
1d6e8131 | 1586 | static void process_recv_sockets(struct work_struct *work) |
fdda387f | 1587 | { |
1d6e8131 PC |
1588 | struct connection *con = container_of(work, struct connection, rwork); |
1589 | int err; | |
fdda387f | 1590 | |
1d6e8131 PC |
1591 | clear_bit(CF_READ_PENDING, &con->flags); |
1592 | do { | |
1593 | err = con->rx_action(con); | |
1594 | } while (!err); | |
fdda387f PC |
1595 | } |
1596 | ||
6ed7257b | 1597 | /* Send workqueue function */ |
1d6e8131 | 1598 | static void process_send_sockets(struct work_struct *work) |
fdda387f | 1599 | { |
1d6e8131 | 1600 | struct connection *con = container_of(work, struct connection, swork); |
fdda387f | 1601 | |
61d9102b | 1602 | if (con->sock == NULL) /* not mutex protected so check it inside too */ |
6ed7257b | 1603 | con->connect_action(con); |
01da24d3 | 1604 | if (!list_empty(&con->writequeue)) |
063c4c99 | 1605 | send_to_sock(con); |
fdda387f PC |
1606 | } |
1607 | ||
1608 | ||
1609 | /* Discard all entries on the write queues */ | |
1610 | static void clean_writequeues(void) | |
1611 | { | |
5e9ccc37 | 1612 | foreach_conn(clean_one_writequeue); |
fdda387f PC |
1613 | } |
1614 | ||
1d6e8131 | 1615 | static void work_stop(void) |
fdda387f | 1616 | { |
1d6e8131 PC |
1617 | destroy_workqueue(recv_workqueue); |
1618 | destroy_workqueue(send_workqueue); | |
fdda387f PC |
1619 | } |
1620 | ||
1d6e8131 | 1621 | static int work_start(void) |
fdda387f | 1622 | { |
e43f055a DT |
1623 | recv_workqueue = alloc_workqueue("dlm_recv", |
1624 | WQ_UNBOUND | WQ_MEM_RECLAIM, 1); | |
b9d41052 NK |
1625 | if (!recv_workqueue) { |
1626 | log_print("can't start dlm_recv"); | |
1627 | return -ENOMEM; | |
fdda387f | 1628 | } |
fdda387f | 1629 | |
e43f055a DT |
1630 | send_workqueue = alloc_workqueue("dlm_send", |
1631 | WQ_UNBOUND | WQ_MEM_RECLAIM, 1); | |
b9d41052 NK |
1632 | if (!send_workqueue) { |
1633 | log_print("can't start dlm_send"); | |
1d6e8131 | 1634 | destroy_workqueue(recv_workqueue); |
b9d41052 | 1635 | return -ENOMEM; |
fdda387f | 1636 | } |
fdda387f PC |
1637 | |
1638 | return 0; | |
1639 | } | |
1640 | ||
f0fb83cb | 1641 | static void _stop_conn(struct connection *con, bool and_other) |
fdda387f | 1642 | { |
f0fb83cb | 1643 | mutex_lock(&con->sock_mutex); |
1644 | set_bit(CF_READ_PENDING, &con->flags); | |
391fbdc5 | 1645 | if (con->sock && con->sock->sk) |
5e9ccc37 | 1646 | con->sock->sk->sk_user_data = NULL; |
f0fb83cb | 1647 | if (con->othercon && and_other) |
1648 | _stop_conn(con->othercon, false); | |
1649 | mutex_unlock(&con->sock_mutex); | |
1650 | } | |
1651 | ||
1652 | static void stop_conn(struct connection *con) | |
1653 | { | |
1654 | _stop_conn(con, true); | |
5e9ccc37 | 1655 | } |
fdda387f | 1656 | |
5e9ccc37 CC |
1657 | static void free_conn(struct connection *con) |
1658 | { | |
0d737a8c | 1659 | close_connection(con, true, true, true); |
5e9ccc37 CC |
1660 | if (con->othercon) |
1661 | kmem_cache_free(con_cache, con->othercon); | |
1662 | hlist_del(&con->list); | |
1663 | kmem_cache_free(con_cache, con); | |
1664 | } | |
1665 | ||
f0fb83cb | 1666 | static void work_flush(void) |
1667 | { | |
1668 | int ok; | |
1669 | int i; | |
1670 | struct hlist_node *n; | |
1671 | struct connection *con; | |
1672 | ||
1673 | flush_workqueue(recv_workqueue); | |
1674 | flush_workqueue(send_workqueue); | |
1675 | do { | |
1676 | ok = 1; | |
1677 | foreach_conn(stop_conn); | |
1678 | flush_workqueue(recv_workqueue); | |
1679 | flush_workqueue(send_workqueue); | |
1680 | for (i = 0; i < CONN_HASH_SIZE && ok; i++) { | |
1681 | hlist_for_each_entry_safe(con, n, | |
1682 | &connection_hash[i], list) { | |
1683 | ok &= test_bit(CF_READ_PENDING, &con->flags); | |
1684 | if (con->othercon) | |
1685 | ok &= test_bit(CF_READ_PENDING, | |
1686 | &con->othercon->flags); | |
1687 | } | |
1688 | } | |
1689 | } while (!ok); | |
1690 | } | |
1691 | ||
5e9ccc37 CC |
1692 | void dlm_lowcomms_stop(void) |
1693 | { | |
ac33d071 | 1694 | /* Set all the flags to prevent any |
fdda387f PC |
1695 | socket activity. |
1696 | */ | |
7a936ce7 | 1697 | mutex_lock(&connections_lock); |
513ef596 | 1698 | dlm_allow_conn = 0; |
f0fb83cb | 1699 | mutex_unlock(&connections_lock); |
1700 | work_flush(); | |
3a8db798 MRL |
1701 | clean_writequeues(); |
1702 | foreach_conn(free_conn); | |
1d6e8131 | 1703 | work_stop(); |
6ed7257b | 1704 | |
fdda387f PC |
1705 | kmem_cache_destroy(con_cache); |
1706 | } | |
1707 | ||
fdda387f PC |
1708 | int dlm_lowcomms_start(void) |
1709 | { | |
6ed7257b PC |
1710 | int error = -EINVAL; |
1711 | struct connection *con; | |
5e9ccc37 CC |
1712 | int i; |
1713 | ||
1714 | for (i = 0; i < CONN_HASH_SIZE; i++) | |
1715 | INIT_HLIST_HEAD(&connection_hash[i]); | |
fdda387f | 1716 | |
6ed7257b PC |
1717 | init_local(); |
1718 | if (!dlm_local_count) { | |
617e82e1 | 1719 | error = -ENOTCONN; |
fdda387f | 1720 | log_print("no local IP address has been set"); |
513ef596 | 1721 | goto fail; |
fdda387f PC |
1722 | } |
1723 | ||
6ed7257b | 1724 | error = -ENOMEM; |
fdda387f | 1725 | con_cache = kmem_cache_create("dlm_conn", sizeof(struct connection), |
ac33d071 | 1726 | __alignof__(struct connection), 0, |
20c2df83 | 1727 | NULL); |
fdda387f | 1728 | if (!con_cache) |
513ef596 DT |
1729 | goto fail; |
1730 | ||
1731 | error = work_start(); | |
1732 | if (error) | |
1733 | goto fail_destroy; | |
1734 | ||
1735 | dlm_allow_conn = 1; | |
fdda387f | 1736 | |
fdda387f | 1737 | /* Start listening */ |
6ed7257b PC |
1738 | if (dlm_config.ci_protocol == 0) |
1739 | error = tcp_listen_for_all(); | |
1740 | else | |
1741 | error = sctp_listen_for_all(); | |
fdda387f PC |
1742 | if (error) |
1743 | goto fail_unlisten; | |
1744 | ||
fdda387f PC |
1745 | return 0; |
1746 | ||
ac33d071 | 1747 | fail_unlisten: |
513ef596 | 1748 | dlm_allow_conn = 0; |
6ed7257b PC |
1749 | con = nodeid2con(0,0); |
1750 | if (con) { | |
0d737a8c | 1751 | close_connection(con, false, true, true); |
6ed7257b PC |
1752 | kmem_cache_free(con_cache, con); |
1753 | } | |
513ef596 | 1754 | fail_destroy: |
fdda387f | 1755 | kmem_cache_destroy(con_cache); |
513ef596 | 1756 | fail: |
fdda387f PC |
1757 | return error; |
1758 | } | |
36b71a8b DT |
1759 | |
1760 | void dlm_lowcomms_exit(void) | |
1761 | { | |
1762 | struct dlm_node_addr *na, *safe; | |
1763 | ||
1764 | spin_lock(&dlm_node_addrs_spin); | |
1765 | list_for_each_entry_safe(na, safe, &dlm_node_addrs, list) { | |
1766 | list_del(&na->list); | |
1767 | while (na->addr_count--) | |
1768 | kfree(na->addr[na->addr_count]); | |
1769 | kfree(na); | |
1770 | } | |
1771 | spin_unlock(&dlm_node_addrs_spin); | |
1772 | } |