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