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 PC |
108 | #define CF_READ_PENDING 1 |
109 | #define CF_WRITE_PENDING 2 | |
110 | #define CF_CONNECT_PENDING 3 | |
6ed7257b PC |
111 | #define CF_INIT_PENDING 4 |
112 | #define CF_IS_OTHERCON 5 | |
063c4c99 | 113 | #define CF_CLOSE 6 |
b36930dd | 114 | #define CF_APP_LIMITED 7 |
ac33d071 | 115 | struct list_head writequeue; /* List of outgoing writequeue_entries */ |
fdda387f PC |
116 | spinlock_t writequeue_lock; |
117 | int (*rx_action) (struct connection *); /* What to do when active */ | |
6ed7257b | 118 | void (*connect_action) (struct connection *); /* What to do to connect */ |
fdda387f PC |
119 | struct page *rx_page; |
120 | struct cbuf cb; | |
121 | int retries; | |
fdda387f | 122 | #define MAX_CONNECT_RETRIES 3 |
5e9ccc37 | 123 | struct hlist_node list; |
fdda387f | 124 | struct connection *othercon; |
1d6e8131 PC |
125 | struct work_struct rwork; /* Receive workqueue */ |
126 | struct work_struct swork; /* Send workqueue */ | |
b3a5bbfd | 127 | void (*orig_error_report)(struct sock *sk); |
fdda387f PC |
128 | }; |
129 | #define sock2con(x) ((struct connection *)(x)->sk_user_data) | |
130 | ||
131 | /* An entry waiting to be sent */ | |
132 | struct writequeue_entry { | |
133 | struct list_head list; | |
134 | struct page *page; | |
135 | int offset; | |
136 | int len; | |
137 | int end; | |
138 | int users; | |
139 | struct connection *con; | |
140 | }; | |
141 | ||
36b71a8b DT |
142 | struct dlm_node_addr { |
143 | struct list_head list; | |
144 | int nodeid; | |
145 | int addr_count; | |
98e1b60e | 146 | int curr_addr_index; |
36b71a8b DT |
147 | struct sockaddr_storage *addr[DLM_MAX_ADDR_COUNT]; |
148 | }; | |
149 | ||
150 | static LIST_HEAD(dlm_node_addrs); | |
151 | static DEFINE_SPINLOCK(dlm_node_addrs_spin); | |
152 | ||
6ed7257b PC |
153 | static struct sockaddr_storage *dlm_local_addr[DLM_MAX_ADDR_COUNT]; |
154 | static int dlm_local_count; | |
513ef596 | 155 | static int dlm_allow_conn; |
fdda387f | 156 | |
1d6e8131 PC |
157 | /* Work queues */ |
158 | static struct workqueue_struct *recv_workqueue; | |
159 | static struct workqueue_struct *send_workqueue; | |
fdda387f | 160 | |
5e9ccc37 | 161 | static struct hlist_head connection_hash[CONN_HASH_SIZE]; |
7a936ce7 | 162 | static DEFINE_MUTEX(connections_lock); |
c80e7c83 | 163 | static struct kmem_cache *con_cache; |
fdda387f | 164 | |
1d6e8131 PC |
165 | static void process_recv_sockets(struct work_struct *work); |
166 | static void process_send_sockets(struct work_struct *work); | |
fdda387f | 167 | |
5e9ccc37 CC |
168 | |
169 | /* This is deliberately very simple because most clusters have simple | |
170 | sequential nodeids, so we should be able to go straight to a connection | |
171 | struct in the array */ | |
172 | static inline int nodeid_hash(int nodeid) | |
173 | { | |
174 | return nodeid & (CONN_HASH_SIZE-1); | |
175 | } | |
176 | ||
177 | static struct connection *__find_con(int nodeid) | |
178 | { | |
179 | int r; | |
5e9ccc37 CC |
180 | struct connection *con; |
181 | ||
182 | r = nodeid_hash(nodeid); | |
183 | ||
b67bfe0d | 184 | hlist_for_each_entry(con, &connection_hash[r], list) { |
5e9ccc37 CC |
185 | if (con->nodeid == nodeid) |
186 | return con; | |
187 | } | |
188 | return NULL; | |
189 | } | |
190 | ||
6ed7257b PC |
191 | /* |
192 | * If 'allocation' is zero then we don't attempt to create a new | |
193 | * connection structure for this node. | |
194 | */ | |
195 | static struct connection *__nodeid2con(int nodeid, gfp_t alloc) | |
fdda387f PC |
196 | { |
197 | struct connection *con = NULL; | |
6ed7257b | 198 | int r; |
fdda387f | 199 | |
5e9ccc37 | 200 | con = __find_con(nodeid); |
6ed7257b PC |
201 | if (con || !alloc) |
202 | return con; | |
fdda387f | 203 | |
6ed7257b PC |
204 | con = kmem_cache_zalloc(con_cache, alloc); |
205 | if (!con) | |
206 | return NULL; | |
fdda387f | 207 | |
5e9ccc37 CC |
208 | r = nodeid_hash(nodeid); |
209 | hlist_add_head(&con->list, &connection_hash[r]); | |
fdda387f | 210 | |
6ed7257b PC |
211 | con->nodeid = nodeid; |
212 | mutex_init(&con->sock_mutex); | |
213 | INIT_LIST_HEAD(&con->writequeue); | |
214 | spin_lock_init(&con->writequeue_lock); | |
215 | INIT_WORK(&con->swork, process_send_sockets); | |
216 | INIT_WORK(&con->rwork, process_recv_sockets); | |
fdda387f | 217 | |
6ed7257b PC |
218 | /* Setup action pointers for child sockets */ |
219 | if (con->nodeid) { | |
5e9ccc37 | 220 | struct connection *zerocon = __find_con(0); |
fdda387f | 221 | |
6ed7257b PC |
222 | con->connect_action = zerocon->connect_action; |
223 | if (!con->rx_action) | |
224 | con->rx_action = zerocon->rx_action; | |
fdda387f PC |
225 | } |
226 | ||
6ed7257b PC |
227 | return con; |
228 | } | |
229 | ||
5e9ccc37 CC |
230 | /* Loop round all connections */ |
231 | static void foreach_conn(void (*conn_func)(struct connection *c)) | |
232 | { | |
233 | int i; | |
b67bfe0d | 234 | struct hlist_node *n; |
5e9ccc37 CC |
235 | struct connection *con; |
236 | ||
237 | for (i = 0; i < CONN_HASH_SIZE; i++) { | |
b67bfe0d | 238 | hlist_for_each_entry_safe(con, n, &connection_hash[i], list) |
5e9ccc37 | 239 | conn_func(con); |
5e9ccc37 CC |
240 | } |
241 | } | |
242 | ||
6ed7257b PC |
243 | static struct connection *nodeid2con(int nodeid, gfp_t allocation) |
244 | { | |
245 | struct connection *con; | |
246 | ||
7a936ce7 | 247 | mutex_lock(&connections_lock); |
6ed7257b | 248 | con = __nodeid2con(nodeid, allocation); |
7a936ce7 | 249 | mutex_unlock(&connections_lock); |
6ed7257b | 250 | |
fdda387f PC |
251 | return con; |
252 | } | |
253 | ||
36b71a8b DT |
254 | static struct dlm_node_addr *find_node_addr(int nodeid) |
255 | { | |
256 | struct dlm_node_addr *na; | |
257 | ||
258 | list_for_each_entry(na, &dlm_node_addrs, list) { | |
259 | if (na->nodeid == nodeid) | |
260 | return na; | |
261 | } | |
262 | return NULL; | |
263 | } | |
264 | ||
265 | static int addr_compare(struct sockaddr_storage *x, struct sockaddr_storage *y) | |
6ed7257b | 266 | { |
36b71a8b DT |
267 | switch (x->ss_family) { |
268 | case AF_INET: { | |
269 | struct sockaddr_in *sinx = (struct sockaddr_in *)x; | |
270 | struct sockaddr_in *siny = (struct sockaddr_in *)y; | |
271 | if (sinx->sin_addr.s_addr != siny->sin_addr.s_addr) | |
272 | return 0; | |
273 | if (sinx->sin_port != siny->sin_port) | |
274 | return 0; | |
275 | break; | |
276 | } | |
277 | case AF_INET6: { | |
278 | struct sockaddr_in6 *sinx = (struct sockaddr_in6 *)x; | |
279 | struct sockaddr_in6 *siny = (struct sockaddr_in6 *)y; | |
280 | if (!ipv6_addr_equal(&sinx->sin6_addr, &siny->sin6_addr)) | |
281 | return 0; | |
282 | if (sinx->sin6_port != siny->sin6_port) | |
283 | return 0; | |
284 | break; | |
285 | } | |
286 | default: | |
287 | return 0; | |
288 | } | |
289 | return 1; | |
290 | } | |
291 | ||
292 | static int nodeid_to_addr(int nodeid, struct sockaddr_storage *sas_out, | |
98e1b60e | 293 | struct sockaddr *sa_out, bool try_new_addr) |
36b71a8b DT |
294 | { |
295 | struct sockaddr_storage sas; | |
296 | struct dlm_node_addr *na; | |
6ed7257b PC |
297 | |
298 | if (!dlm_local_count) | |
299 | return -1; | |
300 | ||
36b71a8b DT |
301 | spin_lock(&dlm_node_addrs_spin); |
302 | na = find_node_addr(nodeid); | |
98e1b60e | 303 | if (na && na->addr_count) { |
ee44b4bc MRL |
304 | memcpy(&sas, na->addr[na->curr_addr_index], |
305 | sizeof(struct sockaddr_storage)); | |
306 | ||
98e1b60e MC |
307 | if (try_new_addr) { |
308 | na->curr_addr_index++; | |
309 | if (na->curr_addr_index == na->addr_count) | |
310 | na->curr_addr_index = 0; | |
311 | } | |
98e1b60e | 312 | } |
36b71a8b DT |
313 | spin_unlock(&dlm_node_addrs_spin); |
314 | ||
315 | if (!na) | |
316 | return -EEXIST; | |
317 | ||
318 | if (!na->addr_count) | |
319 | return -ENOENT; | |
320 | ||
321 | if (sas_out) | |
322 | memcpy(sas_out, &sas, sizeof(struct sockaddr_storage)); | |
323 | ||
324 | if (!sa_out) | |
325 | return 0; | |
6ed7257b PC |
326 | |
327 | if (dlm_local_addr[0]->ss_family == AF_INET) { | |
36b71a8b DT |
328 | struct sockaddr_in *in4 = (struct sockaddr_in *) &sas; |
329 | struct sockaddr_in *ret4 = (struct sockaddr_in *) sa_out; | |
6ed7257b PC |
330 | ret4->sin_addr.s_addr = in4->sin_addr.s_addr; |
331 | } else { | |
36b71a8b DT |
332 | struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) &sas; |
333 | struct sockaddr_in6 *ret6 = (struct sockaddr_in6 *) sa_out; | |
4e3fd7a0 | 334 | ret6->sin6_addr = in6->sin6_addr; |
6ed7257b PC |
335 | } |
336 | ||
337 | return 0; | |
338 | } | |
339 | ||
36b71a8b DT |
340 | static int addr_to_nodeid(struct sockaddr_storage *addr, int *nodeid) |
341 | { | |
342 | struct dlm_node_addr *na; | |
343 | int rv = -EEXIST; | |
98e1b60e | 344 | int addr_i; |
36b71a8b DT |
345 | |
346 | spin_lock(&dlm_node_addrs_spin); | |
347 | list_for_each_entry(na, &dlm_node_addrs, list) { | |
348 | if (!na->addr_count) | |
349 | continue; | |
350 | ||
98e1b60e MC |
351 | for (addr_i = 0; addr_i < na->addr_count; addr_i++) { |
352 | if (addr_compare(na->addr[addr_i], addr)) { | |
353 | *nodeid = na->nodeid; | |
354 | rv = 0; | |
355 | goto unlock; | |
356 | } | |
357 | } | |
36b71a8b | 358 | } |
98e1b60e | 359 | unlock: |
36b71a8b DT |
360 | spin_unlock(&dlm_node_addrs_spin); |
361 | return rv; | |
362 | } | |
363 | ||
364 | int dlm_lowcomms_addr(int nodeid, struct sockaddr_storage *addr, int len) | |
365 | { | |
366 | struct sockaddr_storage *new_addr; | |
367 | struct dlm_node_addr *new_node, *na; | |
368 | ||
369 | new_node = kzalloc(sizeof(struct dlm_node_addr), GFP_NOFS); | |
370 | if (!new_node) | |
371 | return -ENOMEM; | |
372 | ||
373 | new_addr = kzalloc(sizeof(struct sockaddr_storage), GFP_NOFS); | |
374 | if (!new_addr) { | |
375 | kfree(new_node); | |
376 | return -ENOMEM; | |
377 | } | |
378 | ||
379 | memcpy(new_addr, addr, len); | |
380 | ||
381 | spin_lock(&dlm_node_addrs_spin); | |
382 | na = find_node_addr(nodeid); | |
383 | if (!na) { | |
384 | new_node->nodeid = nodeid; | |
385 | new_node->addr[0] = new_addr; | |
386 | new_node->addr_count = 1; | |
387 | list_add(&new_node->list, &dlm_node_addrs); | |
388 | spin_unlock(&dlm_node_addrs_spin); | |
389 | return 0; | |
390 | } | |
391 | ||
392 | if (na->addr_count >= DLM_MAX_ADDR_COUNT) { | |
393 | spin_unlock(&dlm_node_addrs_spin); | |
394 | kfree(new_addr); | |
395 | kfree(new_node); | |
396 | return -ENOSPC; | |
397 | } | |
398 | ||
399 | na->addr[na->addr_count++] = new_addr; | |
400 | spin_unlock(&dlm_node_addrs_spin); | |
401 | kfree(new_node); | |
402 | return 0; | |
403 | } | |
404 | ||
fdda387f | 405 | /* Data available on socket or listen socket received a connect */ |
676d2369 | 406 | static void lowcomms_data_ready(struct sock *sk) |
fdda387f PC |
407 | { |
408 | struct connection *con = sock2con(sk); | |
afb853fb | 409 | if (con && !test_and_set_bit(CF_READ_PENDING, &con->flags)) |
1d6e8131 | 410 | queue_work(recv_workqueue, &con->rwork); |
fdda387f PC |
411 | } |
412 | ||
413 | static void lowcomms_write_space(struct sock *sk) | |
414 | { | |
415 | struct connection *con = sock2con(sk); | |
416 | ||
b36930dd DM |
417 | if (!con) |
418 | return; | |
419 | ||
420 | clear_bit(SOCK_NOSPACE, &con->sock->flags); | |
421 | ||
422 | if (test_and_clear_bit(CF_APP_LIMITED, &con->flags)) { | |
423 | con->sock->sk->sk_write_pending--; | |
9cd3e072 | 424 | clear_bit(SOCKWQ_ASYNC_NOSPACE, &con->sock->flags); |
b36930dd DM |
425 | } |
426 | ||
427 | if (!test_and_set_bit(CF_WRITE_PENDING, &con->flags)) | |
1d6e8131 | 428 | queue_work(send_workqueue, &con->swork); |
fdda387f PC |
429 | } |
430 | ||
431 | static inline void lowcomms_connect_sock(struct connection *con) | |
432 | { | |
063c4c99 LMB |
433 | if (test_bit(CF_CLOSE, &con->flags)) |
434 | return; | |
1d6e8131 PC |
435 | if (!test_and_set_bit(CF_CONNECT_PENDING, &con->flags)) |
436 | queue_work(send_workqueue, &con->swork); | |
fdda387f PC |
437 | } |
438 | ||
439 | static void lowcomms_state_change(struct sock *sk) | |
440 | { | |
ee44b4bc MRL |
441 | /* SCTP layer is not calling sk_data_ready when the connection |
442 | * is done, so we catch the signal through here. Also, it | |
443 | * doesn't switch socket state when entering shutdown, so we | |
444 | * skip the write in that case. | |
445 | */ | |
446 | if (sk->sk_shutdown) { | |
447 | if (sk->sk_shutdown == RCV_SHUTDOWN) | |
448 | lowcomms_data_ready(sk); | |
449 | } else if (sk->sk_state == TCP_ESTABLISHED) { | |
fdda387f | 450 | lowcomms_write_space(sk); |
ee44b4bc | 451 | } |
fdda387f PC |
452 | } |
453 | ||
391fbdc5 CC |
454 | int dlm_lowcomms_connect_node(int nodeid) |
455 | { | |
456 | struct connection *con; | |
457 | ||
458 | if (nodeid == dlm_our_nodeid()) | |
459 | return 0; | |
460 | ||
461 | con = nodeid2con(nodeid, GFP_NOFS); | |
462 | if (!con) | |
463 | return -ENOMEM; | |
464 | lowcomms_connect_sock(con); | |
465 | return 0; | |
466 | } | |
467 | ||
b3a5bbfd BP |
468 | static void lowcomms_error_report(struct sock *sk) |
469 | { | |
470 | struct connection *con = sock2con(sk); | |
471 | struct sockaddr_storage saddr; | |
1a31833d | 472 | int buflen; |
b3a5bbfd | 473 | |
1a31833d BP |
474 | if (con->sock == NULL || |
475 | kernel_getpeername(con->sock, (struct sockaddr *)&saddr, &buflen)) { | |
b3a5bbfd BP |
476 | printk_ratelimited(KERN_ERR "dlm: node %d: socket error " |
477 | "sending to node %d, port %d, " | |
478 | "sk_err=%d/%d\n", dlm_our_nodeid(), | |
479 | con->nodeid, dlm_config.ci_tcp_port, | |
480 | sk->sk_err, sk->sk_err_soft); | |
481 | return; | |
482 | } else if (saddr.ss_family == AF_INET) { | |
483 | struct sockaddr_in *sin4 = (struct sockaddr_in *)&saddr; | |
484 | ||
485 | printk_ratelimited(KERN_ERR "dlm: node %d: socket error " | |
486 | "sending to node %d at %pI4, port %d, " | |
487 | "sk_err=%d/%d\n", dlm_our_nodeid(), | |
488 | con->nodeid, &sin4->sin_addr.s_addr, | |
489 | dlm_config.ci_tcp_port, sk->sk_err, | |
490 | sk->sk_err_soft); | |
491 | } else { | |
492 | struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&saddr; | |
493 | ||
494 | printk_ratelimited(KERN_ERR "dlm: node %d: socket error " | |
495 | "sending to node %d at %u.%u.%u.%u, " | |
496 | "port %d, sk_err=%d/%d\n", dlm_our_nodeid(), | |
497 | con->nodeid, sin6->sin6_addr.s6_addr32[0], | |
498 | sin6->sin6_addr.s6_addr32[1], | |
499 | sin6->sin6_addr.s6_addr32[2], | |
500 | sin6->sin6_addr.s6_addr32[3], | |
501 | dlm_config.ci_tcp_port, sk->sk_err, | |
502 | sk->sk_err_soft); | |
503 | } | |
504 | con->orig_error_report(sk); | |
505 | } | |
506 | ||
fdda387f | 507 | /* Make a socket active */ |
4dd40f0c | 508 | static void add_sock(struct socket *sock, struct connection *con) |
fdda387f PC |
509 | { |
510 | con->sock = sock; | |
511 | ||
512 | /* Install a data_ready callback */ | |
513 | con->sock->sk->sk_data_ready = lowcomms_data_ready; | |
514 | con->sock->sk->sk_write_space = lowcomms_write_space; | |
515 | con->sock->sk->sk_state_change = lowcomms_state_change; | |
6ed7257b | 516 | con->sock->sk->sk_user_data = con; |
d6d7b702 | 517 | con->sock->sk->sk_allocation = GFP_NOFS; |
b3a5bbfd BP |
518 | con->orig_error_report = con->sock->sk->sk_error_report; |
519 | con->sock->sk->sk_error_report = lowcomms_error_report; | |
fdda387f PC |
520 | } |
521 | ||
6ed7257b | 522 | /* Add the port number to an IPv6 or 4 sockaddr and return the address |
fdda387f PC |
523 | length */ |
524 | static void make_sockaddr(struct sockaddr_storage *saddr, uint16_t port, | |
525 | int *addr_len) | |
526 | { | |
6ed7257b | 527 | saddr->ss_family = dlm_local_addr[0]->ss_family; |
ac33d071 | 528 | if (saddr->ss_family == AF_INET) { |
fdda387f PC |
529 | struct sockaddr_in *in4_addr = (struct sockaddr_in *)saddr; |
530 | in4_addr->sin_port = cpu_to_be16(port); | |
531 | *addr_len = sizeof(struct sockaddr_in); | |
6ed7257b | 532 | memset(&in4_addr->sin_zero, 0, sizeof(in4_addr->sin_zero)); |
ac33d071 | 533 | } else { |
fdda387f PC |
534 | struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)saddr; |
535 | in6_addr->sin6_port = cpu_to_be16(port); | |
536 | *addr_len = sizeof(struct sockaddr_in6); | |
537 | } | |
01c8cab2 | 538 | memset((char *)saddr + *addr_len, 0, sizeof(struct sockaddr_storage) - *addr_len); |
fdda387f PC |
539 | } |
540 | ||
541 | /* Close a remote connection and tidy up */ | |
0d737a8c MRL |
542 | static void close_connection(struct connection *con, bool and_other, |
543 | bool tx, bool rx) | |
fdda387f | 544 | { |
0d737a8c MRL |
545 | clear_bit(CF_CONNECT_PENDING, &con->flags); |
546 | clear_bit(CF_WRITE_PENDING, &con->flags); | |
547 | if (tx && cancel_work_sync(&con->swork)) | |
548 | log_print("canceled swork for node %d", con->nodeid); | |
549 | if (rx && cancel_work_sync(&con->rwork)) | |
550 | log_print("canceled rwork for node %d", con->nodeid); | |
fdda387f | 551 | |
0d737a8c | 552 | mutex_lock(&con->sock_mutex); |
fdda387f PC |
553 | if (con->sock) { |
554 | sock_release(con->sock); | |
555 | con->sock = NULL; | |
556 | } | |
557 | if (con->othercon && and_other) { | |
ac33d071 | 558 | /* Will only re-enter once. */ |
0d737a8c | 559 | close_connection(con->othercon, false, true, true); |
fdda387f PC |
560 | } |
561 | if (con->rx_page) { | |
562 | __free_page(con->rx_page); | |
563 | con->rx_page = NULL; | |
564 | } | |
9e5f2825 | 565 | |
61d96be0 PC |
566 | con->retries = 0; |
567 | mutex_unlock(&con->sock_mutex); | |
fdda387f PC |
568 | } |
569 | ||
570 | /* Data received from remote end */ | |
571 | static int receive_from_sock(struct connection *con) | |
572 | { | |
573 | int ret = 0; | |
58addbff AV |
574 | struct msghdr msg = {}; |
575 | struct kvec iov[2]; | |
fdda387f PC |
576 | unsigned len; |
577 | int r; | |
578 | int call_again_soon = 0; | |
58addbff | 579 | int nvec; |
fdda387f | 580 | |
f1f1c1cc | 581 | mutex_lock(&con->sock_mutex); |
fdda387f | 582 | |
a34fbc63 PC |
583 | if (con->sock == NULL) { |
584 | ret = -EAGAIN; | |
585 | goto out_close; | |
586 | } | |
acee4e52 MRL |
587 | if (con->nodeid == 0) { |
588 | ret = -EINVAL; | |
589 | goto out_close; | |
590 | } | |
a34fbc63 | 591 | |
fdda387f PC |
592 | if (con->rx_page == NULL) { |
593 | /* | |
594 | * This doesn't need to be atomic, but I think it should | |
595 | * improve performance if it is. | |
596 | */ | |
597 | con->rx_page = alloc_page(GFP_ATOMIC); | |
598 | if (con->rx_page == NULL) | |
599 | goto out_resched; | |
ac33d071 | 600 | cbuf_init(&con->cb, PAGE_CACHE_SIZE); |
fdda387f PC |
601 | } |
602 | ||
fdda387f PC |
603 | /* |
604 | * iov[0] is the bit of the circular buffer between the current end | |
605 | * point (cb.base + cb.len) and the end of the buffer. | |
606 | */ | |
ac33d071 PC |
607 | iov[0].iov_len = con->cb.base - cbuf_data(&con->cb); |
608 | iov[0].iov_base = page_address(con->rx_page) + cbuf_data(&con->cb); | |
89adc934 | 609 | iov[1].iov_len = 0; |
58addbff | 610 | nvec = 1; |
fdda387f PC |
611 | |
612 | /* | |
613 | * iov[1] is the bit of the circular buffer between the start of the | |
614 | * buffer and the start of the currently used section (cb.base) | |
615 | */ | |
ac33d071 PC |
616 | if (cbuf_data(&con->cb) >= con->cb.base) { |
617 | iov[0].iov_len = PAGE_CACHE_SIZE - cbuf_data(&con->cb); | |
fdda387f PC |
618 | iov[1].iov_len = con->cb.base; |
619 | iov[1].iov_base = page_address(con->rx_page); | |
58addbff | 620 | nvec = 2; |
fdda387f PC |
621 | } |
622 | len = iov[0].iov_len + iov[1].iov_len; | |
623 | ||
58addbff | 624 | r = ret = kernel_recvmsg(con->sock, &msg, iov, nvec, len, |
fdda387f | 625 | MSG_DONTWAIT | MSG_NOSIGNAL); |
fdda387f PC |
626 | if (ret <= 0) |
627 | goto out_close; | |
ee44b4bc MRL |
628 | else if (ret == len) |
629 | call_again_soon = 1; | |
bd44e2b0 | 630 | |
ac33d071 | 631 | cbuf_add(&con->cb, ret); |
fdda387f PC |
632 | ret = dlm_process_incoming_buffer(con->nodeid, |
633 | page_address(con->rx_page), | |
634 | con->cb.base, con->cb.len, | |
635 | PAGE_CACHE_SIZE); | |
636 | if (ret == -EBADMSG) { | |
ee44b4bc MRL |
637 | log_print("lowcomms: addr=%p, base=%u, len=%u, read=%d", |
638 | page_address(con->rx_page), con->cb.base, | |
639 | con->cb.len, r); | |
fdda387f PC |
640 | } |
641 | if (ret < 0) | |
642 | goto out_close; | |
ac33d071 | 643 | cbuf_eat(&con->cb, ret); |
fdda387f | 644 | |
ac33d071 | 645 | if (cbuf_empty(&con->cb) && !call_again_soon) { |
fdda387f PC |
646 | __free_page(con->rx_page); |
647 | con->rx_page = NULL; | |
648 | } | |
649 | ||
fdda387f PC |
650 | if (call_again_soon) |
651 | goto out_resched; | |
f1f1c1cc | 652 | mutex_unlock(&con->sock_mutex); |
ac33d071 | 653 | return 0; |
fdda387f | 654 | |
ac33d071 | 655 | out_resched: |
1d6e8131 PC |
656 | if (!test_and_set_bit(CF_READ_PENDING, &con->flags)) |
657 | queue_work(recv_workqueue, &con->rwork); | |
f1f1c1cc | 658 | mutex_unlock(&con->sock_mutex); |
bd44e2b0 | 659 | return -EAGAIN; |
fdda387f | 660 | |
ac33d071 | 661 | out_close: |
f1f1c1cc | 662 | mutex_unlock(&con->sock_mutex); |
9e5f2825 | 663 | if (ret != -EAGAIN) { |
0d737a8c | 664 | close_connection(con, false, true, false); |
fdda387f PC |
665 | /* Reconnect when there is something to send */ |
666 | } | |
a34fbc63 PC |
667 | /* Don't return success if we really got EOF */ |
668 | if (ret == 0) | |
669 | ret = -EAGAIN; | |
fdda387f | 670 | |
fdda387f PC |
671 | return ret; |
672 | } | |
673 | ||
674 | /* Listening socket is busy, accept a connection */ | |
6ed7257b | 675 | static int tcp_accept_from_sock(struct connection *con) |
fdda387f PC |
676 | { |
677 | int result; | |
678 | struct sockaddr_storage peeraddr; | |
679 | struct socket *newsock; | |
680 | int len; | |
681 | int nodeid; | |
682 | struct connection *newcon; | |
bd44e2b0 | 683 | struct connection *addcon; |
fdda387f | 684 | |
513ef596 DT |
685 | mutex_lock(&connections_lock); |
686 | if (!dlm_allow_conn) { | |
687 | mutex_unlock(&connections_lock); | |
688 | return -1; | |
689 | } | |
690 | mutex_unlock(&connections_lock); | |
691 | ||
fdda387f | 692 | memset(&peeraddr, 0, sizeof(peeraddr)); |
eeb1bd5c EB |
693 | result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family, |
694 | SOCK_STREAM, IPPROTO_TCP, &newsock); | |
fdda387f PC |
695 | if (result < 0) |
696 | return -ENOMEM; | |
697 | ||
f1f1c1cc | 698 | mutex_lock_nested(&con->sock_mutex, 0); |
fdda387f PC |
699 | |
700 | result = -ENOTCONN; | |
701 | if (con->sock == NULL) | |
702 | goto accept_err; | |
703 | ||
704 | newsock->type = con->sock->type; | |
705 | newsock->ops = con->sock->ops; | |
706 | ||
707 | result = con->sock->ops->accept(con->sock, newsock, O_NONBLOCK); | |
708 | if (result < 0) | |
709 | goto accept_err; | |
710 | ||
711 | /* Get the connected socket's peer */ | |
712 | memset(&peeraddr, 0, sizeof(peeraddr)); | |
713 | if (newsock->ops->getname(newsock, (struct sockaddr *)&peeraddr, | |
714 | &len, 2)) { | |
715 | result = -ECONNABORTED; | |
716 | goto accept_err; | |
717 | } | |
718 | ||
719 | /* Get the new node's NODEID */ | |
720 | make_sockaddr(&peeraddr, 0, &len); | |
36b71a8b | 721 | if (addr_to_nodeid(&peeraddr, &nodeid)) { |
bcaadf5c | 722 | unsigned char *b=(unsigned char *)&peeraddr; |
617e82e1 | 723 | log_print("connect from non cluster node"); |
bcaadf5c MY |
724 | print_hex_dump_bytes("ss: ", DUMP_PREFIX_NONE, |
725 | b, sizeof(struct sockaddr_storage)); | |
fdda387f | 726 | sock_release(newsock); |
f1f1c1cc | 727 | mutex_unlock(&con->sock_mutex); |
fdda387f PC |
728 | return -1; |
729 | } | |
730 | ||
731 | log_print("got connection from %d", nodeid); | |
732 | ||
733 | /* Check to see if we already have a connection to this node. This | |
734 | * could happen if the two nodes initiate a connection at roughly | |
735 | * the same time and the connections cross on the wire. | |
fdda387f PC |
736 | * In this case we store the incoming one in "othercon" |
737 | */ | |
748285cc | 738 | newcon = nodeid2con(nodeid, GFP_NOFS); |
fdda387f PC |
739 | if (!newcon) { |
740 | result = -ENOMEM; | |
741 | goto accept_err; | |
742 | } | |
f1f1c1cc | 743 | mutex_lock_nested(&newcon->sock_mutex, 1); |
fdda387f | 744 | if (newcon->sock) { |
ac33d071 | 745 | struct connection *othercon = newcon->othercon; |
fdda387f PC |
746 | |
747 | if (!othercon) { | |
748285cc | 748 | othercon = kmem_cache_zalloc(con_cache, GFP_NOFS); |
fdda387f | 749 | if (!othercon) { |
617e82e1 | 750 | log_print("failed to allocate incoming socket"); |
f1f1c1cc | 751 | mutex_unlock(&newcon->sock_mutex); |
fdda387f PC |
752 | result = -ENOMEM; |
753 | goto accept_err; | |
754 | } | |
fdda387f PC |
755 | othercon->nodeid = nodeid; |
756 | othercon->rx_action = receive_from_sock; | |
f1f1c1cc | 757 | mutex_init(&othercon->sock_mutex); |
1d6e8131 PC |
758 | INIT_WORK(&othercon->swork, process_send_sockets); |
759 | INIT_WORK(&othercon->rwork, process_recv_sockets); | |
fdda387f | 760 | set_bit(CF_IS_OTHERCON, &othercon->flags); |
61d96be0 PC |
761 | } |
762 | if (!othercon->sock) { | |
fdda387f | 763 | newcon->othercon = othercon; |
97d84836 PC |
764 | othercon->sock = newsock; |
765 | newsock->sk->sk_user_data = othercon; | |
766 | add_sock(newsock, othercon); | |
767 | addcon = othercon; | |
768 | } | |
769 | else { | |
770 | printk("Extra connection from node %d attempted\n", nodeid); | |
771 | result = -EAGAIN; | |
f4fadb23 | 772 | mutex_unlock(&newcon->sock_mutex); |
97d84836 | 773 | goto accept_err; |
fdda387f | 774 | } |
fdda387f PC |
775 | } |
776 | else { | |
777 | newsock->sk->sk_user_data = newcon; | |
778 | newcon->rx_action = receive_from_sock; | |
779 | add_sock(newsock, newcon); | |
bd44e2b0 | 780 | addcon = newcon; |
fdda387f PC |
781 | } |
782 | ||
f1f1c1cc | 783 | mutex_unlock(&newcon->sock_mutex); |
fdda387f PC |
784 | |
785 | /* | |
786 | * Add it to the active queue in case we got data | |
25985edc | 787 | * between processing the accept adding the socket |
fdda387f PC |
788 | * to the read_sockets list |
789 | */ | |
bd44e2b0 PC |
790 | if (!test_and_set_bit(CF_READ_PENDING, &addcon->flags)) |
791 | queue_work(recv_workqueue, &addcon->rwork); | |
f1f1c1cc | 792 | mutex_unlock(&con->sock_mutex); |
fdda387f PC |
793 | |
794 | return 0; | |
795 | ||
ac33d071 | 796 | accept_err: |
f1f1c1cc | 797 | mutex_unlock(&con->sock_mutex); |
fdda387f PC |
798 | sock_release(newsock); |
799 | ||
800 | if (result != -EAGAIN) | |
617e82e1 | 801 | log_print("error accepting connection from node: %d", result); |
fdda387f PC |
802 | return result; |
803 | } | |
804 | ||
18df8a87 | 805 | static int sctp_accept_from_sock(struct connection *con) |
ee44b4bc MRL |
806 | { |
807 | /* Check that the new node is in the lockspace */ | |
808 | struct sctp_prim prim; | |
809 | int nodeid; | |
810 | int prim_len, ret; | |
811 | int addr_len; | |
812 | struct connection *newcon; | |
813 | struct connection *addcon; | |
814 | struct socket *newsock; | |
815 | ||
816 | mutex_lock(&connections_lock); | |
817 | if (!dlm_allow_conn) { | |
818 | mutex_unlock(&connections_lock); | |
819 | return -1; | |
820 | } | |
821 | mutex_unlock(&connections_lock); | |
822 | ||
823 | mutex_lock_nested(&con->sock_mutex, 0); | |
824 | ||
825 | ret = kernel_accept(con->sock, &newsock, O_NONBLOCK); | |
826 | if (ret < 0) | |
827 | goto accept_err; | |
828 | ||
829 | memset(&prim, 0, sizeof(struct sctp_prim)); | |
830 | prim_len = sizeof(struct sctp_prim); | |
831 | ||
832 | ret = kernel_getsockopt(newsock, IPPROTO_SCTP, SCTP_PRIMARY_ADDR, | |
833 | (char *)&prim, &prim_len); | |
834 | if (ret < 0) { | |
835 | log_print("getsockopt/sctp_primary_addr failed: %d", ret); | |
836 | goto accept_err; | |
837 | } | |
838 | ||
839 | make_sockaddr(&prim.ssp_addr, 0, &addr_len); | |
840 | if (addr_to_nodeid(&prim.ssp_addr, &nodeid)) { | |
841 | unsigned char *b = (unsigned char *)&prim.ssp_addr; | |
842 | ||
843 | log_print("reject connect from unknown addr"); | |
844 | print_hex_dump_bytes("ss: ", DUMP_PREFIX_NONE, | |
845 | b, sizeof(struct sockaddr_storage)); | |
846 | goto accept_err; | |
847 | } | |
848 | ||
849 | newcon = nodeid2con(nodeid, GFP_NOFS); | |
850 | if (!newcon) { | |
851 | ret = -ENOMEM; | |
852 | goto accept_err; | |
853 | } | |
854 | ||
855 | mutex_lock_nested(&newcon->sock_mutex, 1); | |
856 | ||
857 | if (newcon->sock) { | |
858 | struct connection *othercon = newcon->othercon; | |
859 | ||
860 | if (!othercon) { | |
861 | othercon = kmem_cache_zalloc(con_cache, GFP_NOFS); | |
862 | if (!othercon) { | |
863 | log_print("failed to allocate incoming socket"); | |
864 | mutex_unlock(&newcon->sock_mutex); | |
865 | ret = -ENOMEM; | |
866 | goto accept_err; | |
867 | } | |
868 | othercon->nodeid = nodeid; | |
869 | othercon->rx_action = receive_from_sock; | |
870 | mutex_init(&othercon->sock_mutex); | |
871 | INIT_WORK(&othercon->swork, process_send_sockets); | |
872 | INIT_WORK(&othercon->rwork, process_recv_sockets); | |
873 | set_bit(CF_IS_OTHERCON, &othercon->flags); | |
874 | } | |
875 | if (!othercon->sock) { | |
876 | newcon->othercon = othercon; | |
877 | othercon->sock = newsock; | |
878 | newsock->sk->sk_user_data = othercon; | |
879 | add_sock(newsock, othercon); | |
880 | addcon = othercon; | |
881 | } else { | |
882 | printk("Extra connection from node %d attempted\n", nodeid); | |
883 | ret = -EAGAIN; | |
884 | mutex_unlock(&newcon->sock_mutex); | |
885 | goto accept_err; | |
886 | } | |
887 | } else { | |
888 | newsock->sk->sk_user_data = newcon; | |
889 | newcon->rx_action = receive_from_sock; | |
890 | add_sock(newsock, newcon); | |
891 | addcon = newcon; | |
892 | } | |
893 | ||
894 | log_print("connected to %d", nodeid); | |
895 | ||
896 | mutex_unlock(&newcon->sock_mutex); | |
897 | ||
898 | /* | |
899 | * Add it to the active queue in case we got data | |
900 | * between processing the accept adding the socket | |
901 | * to the read_sockets list | |
902 | */ | |
903 | if (!test_and_set_bit(CF_READ_PENDING, &addcon->flags)) | |
904 | queue_work(recv_workqueue, &addcon->rwork); | |
905 | mutex_unlock(&con->sock_mutex); | |
906 | ||
907 | return 0; | |
908 | ||
909 | accept_err: | |
910 | mutex_unlock(&con->sock_mutex); | |
911 | if (newsock) | |
912 | sock_release(newsock); | |
913 | if (ret != -EAGAIN) | |
914 | log_print("error accepting connection from node: %d", ret); | |
915 | ||
916 | return ret; | |
917 | } | |
918 | ||
6ed7257b PC |
919 | static void free_entry(struct writequeue_entry *e) |
920 | { | |
921 | __free_page(e->page); | |
922 | kfree(e); | |
923 | } | |
924 | ||
5d689871 MC |
925 | /* |
926 | * writequeue_entry_complete - try to delete and free write queue entry | |
927 | * @e: write queue entry to try to delete | |
928 | * @completed: bytes completed | |
929 | * | |
930 | * writequeue_lock must be held. | |
931 | */ | |
932 | static void writequeue_entry_complete(struct writequeue_entry *e, int completed) | |
933 | { | |
934 | e->offset += completed; | |
935 | e->len -= completed; | |
936 | ||
937 | if (e->len == 0 && e->users == 0) { | |
938 | list_del(&e->list); | |
939 | free_entry(e); | |
940 | } | |
941 | } | |
942 | ||
ee44b4bc MRL |
943 | /* |
944 | * sctp_bind_addrs - bind a SCTP socket to all our addresses | |
945 | */ | |
946 | static int sctp_bind_addrs(struct connection *con, uint16_t port) | |
947 | { | |
948 | struct sockaddr_storage localaddr; | |
949 | int i, addr_len, result = 0; | |
950 | ||
951 | for (i = 0; i < dlm_local_count; i++) { | |
952 | memcpy(&localaddr, dlm_local_addr[i], sizeof(localaddr)); | |
953 | make_sockaddr(&localaddr, port, &addr_len); | |
954 | ||
955 | if (!i) | |
956 | result = kernel_bind(con->sock, | |
957 | (struct sockaddr *)&localaddr, | |
958 | addr_len); | |
959 | else | |
960 | result = kernel_setsockopt(con->sock, SOL_SCTP, | |
961 | SCTP_SOCKOPT_BINDX_ADD, | |
962 | (char *)&localaddr, addr_len); | |
963 | ||
964 | if (result < 0) { | |
965 | log_print("Can't bind to %d addr number %d, %d.\n", | |
966 | port, i + 1, result); | |
967 | break; | |
968 | } | |
969 | } | |
970 | return result; | |
971 | } | |
972 | ||
6ed7257b PC |
973 | /* Initiate an SCTP association. |
974 | This is a special case of send_to_sock() in that we don't yet have a | |
975 | peeled-off socket for this association, so we use the listening socket | |
976 | and add the primary IP address of the remote node. | |
977 | */ | |
ee44b4bc | 978 | static void sctp_connect_to_sock(struct connection *con) |
6ed7257b | 979 | { |
ee44b4bc MRL |
980 | struct sockaddr_storage daddr; |
981 | int one = 1; | |
982 | int result; | |
983 | int addr_len; | |
984 | struct socket *sock; | |
985 | ||
986 | if (con->nodeid == 0) { | |
987 | log_print("attempt to connect sock 0 foiled"); | |
988 | return; | |
989 | } | |
6ed7257b | 990 | |
5d689871 | 991 | mutex_lock(&con->sock_mutex); |
6ed7257b | 992 | |
ee44b4bc MRL |
993 | /* Some odd races can cause double-connects, ignore them */ |
994 | if (con->retries++ > MAX_CONNECT_RETRIES) | |
995 | goto out; | |
996 | ||
997 | if (con->sock) { | |
998 | log_print("node %d already connected.", con->nodeid); | |
999 | goto out; | |
1000 | } | |
1001 | ||
1002 | memset(&daddr, 0, sizeof(daddr)); | |
1003 | result = nodeid_to_addr(con->nodeid, &daddr, NULL, true); | |
1004 | if (result < 0) { | |
6ed7257b | 1005 | log_print("no address for nodeid %d", con->nodeid); |
ee44b4bc | 1006 | goto out; |
6ed7257b | 1007 | } |
6ed7257b | 1008 | |
ee44b4bc MRL |
1009 | /* Create a socket to communicate with */ |
1010 | result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family, | |
1011 | SOCK_STREAM, IPPROTO_SCTP, &sock); | |
1012 | if (result < 0) | |
1013 | goto socket_err; | |
6ed7257b | 1014 | |
ee44b4bc MRL |
1015 | sock->sk->sk_user_data = con; |
1016 | con->rx_action = receive_from_sock; | |
1017 | con->connect_action = sctp_connect_to_sock; | |
1018 | add_sock(sock, con); | |
6ed7257b | 1019 | |
ee44b4bc MRL |
1020 | /* Bind to all addresses. */ |
1021 | if (sctp_bind_addrs(con, 0)) | |
1022 | goto bind_err; | |
6ed7257b | 1023 | |
ee44b4bc | 1024 | make_sockaddr(&daddr, dlm_config.ci_tcp_port, &addr_len); |
6ed7257b | 1025 | |
ee44b4bc | 1026 | log_print("connecting to %d", con->nodeid); |
6ed7257b | 1027 | |
ee44b4bc MRL |
1028 | /* Turn off Nagle's algorithm */ |
1029 | kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY, (char *)&one, | |
1030 | sizeof(one)); | |
6ed7257b | 1031 | |
ee44b4bc MRL |
1032 | result = sock->ops->connect(sock, (struct sockaddr *)&daddr, addr_len, |
1033 | O_NONBLOCK); | |
1034 | if (result == -EINPROGRESS) | |
1035 | result = 0; | |
1036 | if (result == 0) | |
1037 | goto out; | |
98e1b60e | 1038 | |
6ed7257b | 1039 | |
ee44b4bc MRL |
1040 | bind_err: |
1041 | con->sock = NULL; | |
1042 | sock_release(sock); | |
6ed7257b | 1043 | |
ee44b4bc MRL |
1044 | socket_err: |
1045 | /* | |
1046 | * Some errors are fatal and this list might need adjusting. For other | |
1047 | * errors we try again until the max number of retries is reached. | |
1048 | */ | |
1049 | if (result != -EHOSTUNREACH && | |
1050 | result != -ENETUNREACH && | |
1051 | result != -ENETDOWN && | |
1052 | result != -EINVAL && | |
1053 | result != -EPROTONOSUPPORT) { | |
1054 | log_print("connect %d try %d error %d", con->nodeid, | |
1055 | con->retries, result); | |
1056 | mutex_unlock(&con->sock_mutex); | |
1057 | msleep(1000); | |
6ed7257b | 1058 | clear_bit(CF_CONNECT_PENDING, &con->flags); |
ee44b4bc MRL |
1059 | lowcomms_connect_sock(con); |
1060 | return; | |
6ed7257b | 1061 | } |
5d689871 | 1062 | |
ee44b4bc | 1063 | out: |
5d689871 | 1064 | mutex_unlock(&con->sock_mutex); |
00dcffae | 1065 | set_bit(CF_WRITE_PENDING, &con->flags); |
6ed7257b PC |
1066 | } |
1067 | ||
fdda387f | 1068 | /* Connect a new socket to its peer */ |
6ed7257b | 1069 | static void tcp_connect_to_sock(struct connection *con) |
fdda387f | 1070 | { |
6bd8feda | 1071 | struct sockaddr_storage saddr, src_addr; |
fdda387f | 1072 | int addr_len; |
a89d63a1 | 1073 | struct socket *sock = NULL; |
cb2d45da | 1074 | int one = 1; |
36b71a8b | 1075 | int result; |
fdda387f PC |
1076 | |
1077 | if (con->nodeid == 0) { | |
1078 | log_print("attempt to connect sock 0 foiled"); | |
ac33d071 | 1079 | return; |
fdda387f PC |
1080 | } |
1081 | ||
f1f1c1cc | 1082 | mutex_lock(&con->sock_mutex); |
fdda387f PC |
1083 | if (con->retries++ > MAX_CONNECT_RETRIES) |
1084 | goto out; | |
1085 | ||
1086 | /* Some odd races can cause double-connects, ignore them */ | |
36b71a8b | 1087 | if (con->sock) |
fdda387f | 1088 | goto out; |
fdda387f PC |
1089 | |
1090 | /* Create a socket to communicate with */ | |
eeb1bd5c EB |
1091 | result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family, |
1092 | SOCK_STREAM, IPPROTO_TCP, &sock); | |
fdda387f PC |
1093 | if (result < 0) |
1094 | goto out_err; | |
1095 | ||
1096 | memset(&saddr, 0, sizeof(saddr)); | |
98e1b60e | 1097 | result = nodeid_to_addr(con->nodeid, &saddr, NULL, false); |
36b71a8b DT |
1098 | if (result < 0) { |
1099 | log_print("no address for nodeid %d", con->nodeid); | |
ac33d071 | 1100 | goto out_err; |
36b71a8b | 1101 | } |
fdda387f PC |
1102 | |
1103 | sock->sk->sk_user_data = con; | |
1104 | con->rx_action = receive_from_sock; | |
6ed7257b PC |
1105 | con->connect_action = tcp_connect_to_sock; |
1106 | add_sock(sock, con); | |
fdda387f | 1107 | |
6bd8feda LH |
1108 | /* Bind to our cluster-known address connecting to avoid |
1109 | routing problems */ | |
1110 | memcpy(&src_addr, dlm_local_addr[0], sizeof(src_addr)); | |
1111 | make_sockaddr(&src_addr, 0, &addr_len); | |
1112 | result = sock->ops->bind(sock, (struct sockaddr *) &src_addr, | |
1113 | addr_len); | |
1114 | if (result < 0) { | |
1115 | log_print("could not bind for connect: %d", result); | |
1116 | /* This *may* not indicate a critical error */ | |
1117 | } | |
1118 | ||
68c817a1 | 1119 | make_sockaddr(&saddr, dlm_config.ci_tcp_port, &addr_len); |
fdda387f | 1120 | |
fdda387f | 1121 | log_print("connecting to %d", con->nodeid); |
cb2d45da DT |
1122 | |
1123 | /* Turn off Nagle's algorithm */ | |
1124 | kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY, (char *)&one, | |
1125 | sizeof(one)); | |
1126 | ||
36b71a8b | 1127 | result = sock->ops->connect(sock, (struct sockaddr *)&saddr, addr_len, |
ac33d071 | 1128 | O_NONBLOCK); |
fdda387f PC |
1129 | if (result == -EINPROGRESS) |
1130 | result = 0; | |
ac33d071 PC |
1131 | if (result == 0) |
1132 | goto out; | |
fdda387f | 1133 | |
ac33d071 | 1134 | out_err: |
fdda387f PC |
1135 | if (con->sock) { |
1136 | sock_release(con->sock); | |
1137 | con->sock = NULL; | |
a89d63a1 CD |
1138 | } else if (sock) { |
1139 | sock_release(sock); | |
fdda387f PC |
1140 | } |
1141 | /* | |
1142 | * Some errors are fatal and this list might need adjusting. For other | |
1143 | * errors we try again until the max number of retries is reached. | |
1144 | */ | |
36b71a8b DT |
1145 | if (result != -EHOSTUNREACH && |
1146 | result != -ENETUNREACH && | |
1147 | result != -ENETDOWN && | |
1148 | result != -EINVAL && | |
1149 | result != -EPROTONOSUPPORT) { | |
1150 | log_print("connect %d try %d error %d", con->nodeid, | |
1151 | con->retries, result); | |
1152 | mutex_unlock(&con->sock_mutex); | |
1153 | msleep(1000); | |
356344c4 | 1154 | clear_bit(CF_CONNECT_PENDING, &con->flags); |
fdda387f | 1155 | lowcomms_connect_sock(con); |
36b71a8b | 1156 | return; |
fdda387f | 1157 | } |
ac33d071 | 1158 | out: |
f1f1c1cc | 1159 | mutex_unlock(&con->sock_mutex); |
00dcffae | 1160 | set_bit(CF_WRITE_PENDING, &con->flags); |
ac33d071 | 1161 | return; |
fdda387f PC |
1162 | } |
1163 | ||
6ed7257b PC |
1164 | static struct socket *tcp_create_listen_sock(struct connection *con, |
1165 | struct sockaddr_storage *saddr) | |
fdda387f | 1166 | { |
ac33d071 | 1167 | struct socket *sock = NULL; |
fdda387f PC |
1168 | int result = 0; |
1169 | int one = 1; | |
1170 | int addr_len; | |
1171 | ||
6ed7257b | 1172 | if (dlm_local_addr[0]->ss_family == AF_INET) |
fdda387f PC |
1173 | addr_len = sizeof(struct sockaddr_in); |
1174 | else | |
1175 | addr_len = sizeof(struct sockaddr_in6); | |
1176 | ||
1177 | /* Create a socket to communicate with */ | |
eeb1bd5c EB |
1178 | result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family, |
1179 | SOCK_STREAM, IPPROTO_TCP, &sock); | |
fdda387f | 1180 | if (result < 0) { |
617e82e1 | 1181 | log_print("Can't create listening comms socket"); |
fdda387f PC |
1182 | goto create_out; |
1183 | } | |
1184 | ||
cb2d45da DT |
1185 | /* Turn off Nagle's algorithm */ |
1186 | kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY, (char *)&one, | |
1187 | sizeof(one)); | |
1188 | ||
6ed7257b PC |
1189 | result = kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEADDR, |
1190 | (char *)&one, sizeof(one)); | |
1191 | ||
fdda387f | 1192 | if (result < 0) { |
617e82e1 | 1193 | log_print("Failed to set SO_REUSEADDR on socket: %d", result); |
fdda387f | 1194 | } |
6ed7257b PC |
1195 | con->rx_action = tcp_accept_from_sock; |
1196 | con->connect_action = tcp_connect_to_sock; | |
fdda387f PC |
1197 | |
1198 | /* Bind to our port */ | |
68c817a1 | 1199 | make_sockaddr(saddr, dlm_config.ci_tcp_port, &addr_len); |
fdda387f PC |
1200 | result = sock->ops->bind(sock, (struct sockaddr *) saddr, addr_len); |
1201 | if (result < 0) { | |
617e82e1 | 1202 | log_print("Can't bind to port %d", dlm_config.ci_tcp_port); |
fdda387f PC |
1203 | sock_release(sock); |
1204 | sock = NULL; | |
1205 | con->sock = NULL; | |
1206 | goto create_out; | |
1207 | } | |
6ed7257b | 1208 | result = kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE, |
ac33d071 | 1209 | (char *)&one, sizeof(one)); |
fdda387f | 1210 | if (result < 0) { |
617e82e1 | 1211 | log_print("Set keepalive failed: %d", result); |
fdda387f PC |
1212 | } |
1213 | ||
1214 | result = sock->ops->listen(sock, 5); | |
1215 | if (result < 0) { | |
617e82e1 | 1216 | log_print("Can't listen on port %d", dlm_config.ci_tcp_port); |
fdda387f PC |
1217 | sock_release(sock); |
1218 | sock = NULL; | |
1219 | goto create_out; | |
1220 | } | |
1221 | ||
ac33d071 | 1222 | create_out: |
fdda387f PC |
1223 | return sock; |
1224 | } | |
1225 | ||
6ed7257b PC |
1226 | /* Get local addresses */ |
1227 | static void init_local(void) | |
1228 | { | |
1229 | struct sockaddr_storage sas, *addr; | |
1230 | int i; | |
1231 | ||
30d3a237 | 1232 | dlm_local_count = 0; |
1b189b88 | 1233 | for (i = 0; i < DLM_MAX_ADDR_COUNT; i++) { |
6ed7257b PC |
1234 | if (dlm_our_addr(&sas, i)) |
1235 | break; | |
1236 | ||
573c24c4 | 1237 | addr = kmalloc(sizeof(*addr), GFP_NOFS); |
6ed7257b PC |
1238 | if (!addr) |
1239 | break; | |
1240 | memcpy(addr, &sas, sizeof(*addr)); | |
1241 | dlm_local_addr[dlm_local_count++] = addr; | |
1242 | } | |
1243 | } | |
1244 | ||
6ed7257b PC |
1245 | /* Initialise SCTP socket and bind to all interfaces */ |
1246 | static int sctp_listen_for_all(void) | |
1247 | { | |
1248 | struct socket *sock = NULL; | |
ee44b4bc | 1249 | int result = -EINVAL; |
573c24c4 | 1250 | struct connection *con = nodeid2con(0, GFP_NOFS); |
6ed7257b | 1251 | int bufsize = NEEDED_RMEM; |
86e92ad2 | 1252 | int one = 1; |
6ed7257b PC |
1253 | |
1254 | if (!con) | |
1255 | return -ENOMEM; | |
1256 | ||
1257 | log_print("Using SCTP for communications"); | |
1258 | ||
eeb1bd5c | 1259 | result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family, |
ee44b4bc | 1260 | SOCK_STREAM, IPPROTO_SCTP, &sock); |
6ed7257b PC |
1261 | if (result < 0) { |
1262 | log_print("Can't create comms socket, check SCTP is loaded"); | |
1263 | goto out; | |
1264 | } | |
1265 | ||
df61c952 | 1266 | result = kernel_setsockopt(sock, SOL_SOCKET, SO_RCVBUFFORCE, |
6ed7257b PC |
1267 | (char *)&bufsize, sizeof(bufsize)); |
1268 | if (result) | |
617e82e1 | 1269 | log_print("Error increasing buffer space on socket %d", result); |
6ed7257b | 1270 | |
86e92ad2 MC |
1271 | result = kernel_setsockopt(sock, SOL_SCTP, SCTP_NODELAY, (char *)&one, |
1272 | sizeof(one)); | |
1273 | if (result < 0) | |
1274 | log_print("Could not set SCTP NODELAY error %d\n", result); | |
1275 | ||
6ed7257b PC |
1276 | /* Init con struct */ |
1277 | sock->sk->sk_user_data = con; | |
1278 | con->sock = sock; | |
1279 | con->sock->sk->sk_data_ready = lowcomms_data_ready; | |
ee44b4bc MRL |
1280 | con->rx_action = sctp_accept_from_sock; |
1281 | con->connect_action = sctp_connect_to_sock; | |
6ed7257b | 1282 | |
ee44b4bc MRL |
1283 | /* Bind to all addresses. */ |
1284 | if (sctp_bind_addrs(con, dlm_config.ci_tcp_port)) | |
1285 | goto create_delsock; | |
6ed7257b PC |
1286 | |
1287 | result = sock->ops->listen(sock, 5); | |
1288 | if (result < 0) { | |
1289 | log_print("Can't set socket listening"); | |
1290 | goto create_delsock; | |
1291 | } | |
1292 | ||
1293 | return 0; | |
1294 | ||
1295 | create_delsock: | |
1296 | sock_release(sock); | |
1297 | con->sock = NULL; | |
1298 | out: | |
1299 | return result; | |
1300 | } | |
1301 | ||
1302 | static int tcp_listen_for_all(void) | |
fdda387f PC |
1303 | { |
1304 | struct socket *sock = NULL; | |
573c24c4 | 1305 | struct connection *con = nodeid2con(0, GFP_NOFS); |
fdda387f PC |
1306 | int result = -EINVAL; |
1307 | ||
6ed7257b PC |
1308 | if (!con) |
1309 | return -ENOMEM; | |
1310 | ||
fdda387f | 1311 | /* We don't support multi-homed hosts */ |
6ed7257b | 1312 | if (dlm_local_addr[1] != NULL) { |
617e82e1 DT |
1313 | log_print("TCP protocol can't handle multi-homed hosts, " |
1314 | "try SCTP"); | |
6ed7257b PC |
1315 | return -EINVAL; |
1316 | } | |
1317 | ||
1318 | log_print("Using TCP for communications"); | |
1319 | ||
6ed7257b | 1320 | sock = tcp_create_listen_sock(con, dlm_local_addr[0]); |
fdda387f PC |
1321 | if (sock) { |
1322 | add_sock(sock, con); | |
1323 | result = 0; | |
1324 | } | |
1325 | else { | |
1326 | result = -EADDRINUSE; | |
1327 | } | |
1328 | ||
1329 | return result; | |
1330 | } | |
1331 | ||
1332 | ||
1333 | ||
1334 | static struct writequeue_entry *new_writequeue_entry(struct connection *con, | |
1335 | gfp_t allocation) | |
1336 | { | |
1337 | struct writequeue_entry *entry; | |
1338 | ||
1339 | entry = kmalloc(sizeof(struct writequeue_entry), allocation); | |
1340 | if (!entry) | |
1341 | return NULL; | |
1342 | ||
1343 | entry->page = alloc_page(allocation); | |
1344 | if (!entry->page) { | |
1345 | kfree(entry); | |
1346 | return NULL; | |
1347 | } | |
1348 | ||
1349 | entry->offset = 0; | |
1350 | entry->len = 0; | |
1351 | entry->end = 0; | |
1352 | entry->users = 0; | |
1353 | entry->con = con; | |
1354 | ||
1355 | return entry; | |
1356 | } | |
1357 | ||
617e82e1 | 1358 | void *dlm_lowcomms_get_buffer(int nodeid, int len, gfp_t allocation, char **ppc) |
fdda387f PC |
1359 | { |
1360 | struct connection *con; | |
1361 | struct writequeue_entry *e; | |
1362 | int offset = 0; | |
fdda387f | 1363 | |
fdda387f PC |
1364 | con = nodeid2con(nodeid, allocation); |
1365 | if (!con) | |
1366 | return NULL; | |
1367 | ||
4edde74e | 1368 | spin_lock(&con->writequeue_lock); |
fdda387f | 1369 | e = list_entry(con->writequeue.prev, struct writequeue_entry, list); |
ac33d071 | 1370 | if ((&e->list == &con->writequeue) || |
fdda387f PC |
1371 | (PAGE_CACHE_SIZE - e->end < len)) { |
1372 | e = NULL; | |
1373 | } else { | |
1374 | offset = e->end; | |
1375 | e->end += len; | |
eeee2b5f | 1376 | e->users++; |
fdda387f PC |
1377 | } |
1378 | spin_unlock(&con->writequeue_lock); | |
1379 | ||
1380 | if (e) { | |
ac33d071 | 1381 | got_one: |
fdda387f PC |
1382 | *ppc = page_address(e->page) + offset; |
1383 | return e; | |
1384 | } | |
1385 | ||
1386 | e = new_writequeue_entry(con, allocation); | |
1387 | if (e) { | |
1388 | spin_lock(&con->writequeue_lock); | |
1389 | offset = e->end; | |
1390 | e->end += len; | |
eeee2b5f | 1391 | e->users++; |
fdda387f PC |
1392 | list_add_tail(&e->list, &con->writequeue); |
1393 | spin_unlock(&con->writequeue_lock); | |
1394 | goto got_one; | |
1395 | } | |
1396 | return NULL; | |
1397 | } | |
1398 | ||
1399 | void dlm_lowcomms_commit_buffer(void *mh) | |
1400 | { | |
1401 | struct writequeue_entry *e = (struct writequeue_entry *)mh; | |
1402 | struct connection *con = e->con; | |
1403 | int users; | |
1404 | ||
4edde74e | 1405 | spin_lock(&con->writequeue_lock); |
fdda387f PC |
1406 | users = --e->users; |
1407 | if (users) | |
1408 | goto out; | |
1409 | e->len = e->end - e->offset; | |
fdda387f PC |
1410 | spin_unlock(&con->writequeue_lock); |
1411 | ||
1d6e8131 PC |
1412 | if (!test_and_set_bit(CF_WRITE_PENDING, &con->flags)) { |
1413 | queue_work(send_workqueue, &con->swork); | |
fdda387f PC |
1414 | } |
1415 | return; | |
1416 | ||
ac33d071 | 1417 | out: |
fdda387f PC |
1418 | spin_unlock(&con->writequeue_lock); |
1419 | return; | |
1420 | } | |
1421 | ||
fdda387f | 1422 | /* Send a message */ |
ac33d071 | 1423 | static void send_to_sock(struct connection *con) |
fdda387f PC |
1424 | { |
1425 | int ret = 0; | |
fdda387f PC |
1426 | const int msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL; |
1427 | struct writequeue_entry *e; | |
1428 | int len, offset; | |
f92c8dd7 | 1429 | int count = 0; |
fdda387f | 1430 | |
f1f1c1cc | 1431 | mutex_lock(&con->sock_mutex); |
fdda387f PC |
1432 | if (con->sock == NULL) |
1433 | goto out_connect; | |
1434 | ||
fdda387f PC |
1435 | spin_lock(&con->writequeue_lock); |
1436 | for (;;) { | |
1437 | e = list_entry(con->writequeue.next, struct writequeue_entry, | |
1438 | list); | |
1439 | if ((struct list_head *) e == &con->writequeue) | |
1440 | break; | |
1441 | ||
1442 | len = e->len; | |
1443 | offset = e->offset; | |
1444 | BUG_ON(len == 0 && e->users == 0); | |
1445 | spin_unlock(&con->writequeue_lock); | |
1446 | ||
1447 | ret = 0; | |
1448 | if (len) { | |
1329e3f2 PB |
1449 | ret = kernel_sendpage(con->sock, e->page, offset, len, |
1450 | msg_flags); | |
d66f8277 | 1451 | if (ret == -EAGAIN || ret == 0) { |
b36930dd | 1452 | if (ret == -EAGAIN && |
9cd3e072 | 1453 | test_bit(SOCKWQ_ASYNC_NOSPACE, &con->sock->flags) && |
b36930dd DM |
1454 | !test_and_set_bit(CF_APP_LIMITED, &con->flags)) { |
1455 | /* Notify TCP that we're limited by the | |
1456 | * application window size. | |
1457 | */ | |
1458 | set_bit(SOCK_NOSPACE, &con->sock->flags); | |
1459 | con->sock->sk->sk_write_pending++; | |
1460 | } | |
d66f8277 | 1461 | cond_resched(); |
fdda387f | 1462 | goto out; |
9c5bef58 | 1463 | } else if (ret < 0) |
fdda387f | 1464 | goto send_error; |
d66f8277 | 1465 | } |
f92c8dd7 BP |
1466 | |
1467 | /* Don't starve people filling buffers */ | |
1468 | if (++count >= MAX_SEND_MSG_COUNT) { | |
ac33d071 | 1469 | cond_resched(); |
f92c8dd7 BP |
1470 | count = 0; |
1471 | } | |
fdda387f PC |
1472 | |
1473 | spin_lock(&con->writequeue_lock); | |
5d689871 | 1474 | writequeue_entry_complete(e, ret); |
fdda387f PC |
1475 | } |
1476 | spin_unlock(&con->writequeue_lock); | |
ac33d071 | 1477 | out: |
f1f1c1cc | 1478 | mutex_unlock(&con->sock_mutex); |
ac33d071 | 1479 | return; |
fdda387f | 1480 | |
ac33d071 | 1481 | send_error: |
f1f1c1cc | 1482 | mutex_unlock(&con->sock_mutex); |
0d737a8c | 1483 | close_connection(con, false, false, true); |
fdda387f | 1484 | lowcomms_connect_sock(con); |
ac33d071 | 1485 | return; |
fdda387f | 1486 | |
ac33d071 | 1487 | out_connect: |
f1f1c1cc | 1488 | mutex_unlock(&con->sock_mutex); |
ee44b4bc | 1489 | lowcomms_connect_sock(con); |
fdda387f PC |
1490 | } |
1491 | ||
1492 | static void clean_one_writequeue(struct connection *con) | |
1493 | { | |
5e9ccc37 | 1494 | struct writequeue_entry *e, *safe; |
fdda387f PC |
1495 | |
1496 | spin_lock(&con->writequeue_lock); | |
5e9ccc37 | 1497 | list_for_each_entry_safe(e, safe, &con->writequeue, list) { |
fdda387f PC |
1498 | list_del(&e->list); |
1499 | free_entry(e); | |
1500 | } | |
1501 | spin_unlock(&con->writequeue_lock); | |
1502 | } | |
1503 | ||
1504 | /* Called from recovery when it knows that a node has | |
1505 | left the cluster */ | |
1506 | int dlm_lowcomms_close(int nodeid) | |
1507 | { | |
1508 | struct connection *con; | |
36b71a8b | 1509 | struct dlm_node_addr *na; |
fdda387f | 1510 | |
fdda387f PC |
1511 | log_print("closing connection to node %d", nodeid); |
1512 | con = nodeid2con(nodeid, 0); | |
1513 | if (con) { | |
063c4c99 | 1514 | set_bit(CF_CLOSE, &con->flags); |
0d737a8c | 1515 | close_connection(con, true, true, true); |
fdda387f | 1516 | clean_one_writequeue(con); |
fdda387f | 1517 | } |
36b71a8b DT |
1518 | |
1519 | spin_lock(&dlm_node_addrs_spin); | |
1520 | na = find_node_addr(nodeid); | |
1521 | if (na) { | |
1522 | list_del(&na->list); | |
1523 | while (na->addr_count--) | |
1524 | kfree(na->addr[na->addr_count]); | |
1525 | kfree(na); | |
1526 | } | |
1527 | spin_unlock(&dlm_node_addrs_spin); | |
1528 | ||
fdda387f | 1529 | return 0; |
fdda387f PC |
1530 | } |
1531 | ||
6ed7257b | 1532 | /* Receive workqueue function */ |
1d6e8131 | 1533 | static void process_recv_sockets(struct work_struct *work) |
fdda387f | 1534 | { |
1d6e8131 PC |
1535 | struct connection *con = container_of(work, struct connection, rwork); |
1536 | int err; | |
fdda387f | 1537 | |
1d6e8131 PC |
1538 | clear_bit(CF_READ_PENDING, &con->flags); |
1539 | do { | |
1540 | err = con->rx_action(con); | |
1541 | } while (!err); | |
fdda387f PC |
1542 | } |
1543 | ||
6ed7257b | 1544 | /* Send workqueue function */ |
1d6e8131 | 1545 | static void process_send_sockets(struct work_struct *work) |
fdda387f | 1546 | { |
1d6e8131 | 1547 | struct connection *con = container_of(work, struct connection, swork); |
fdda387f | 1548 | |
00dcffae | 1549 | if (test_and_clear_bit(CF_CONNECT_PENDING, &con->flags)) |
6ed7257b | 1550 | con->connect_action(con); |
063c4c99 LMB |
1551 | if (test_and_clear_bit(CF_WRITE_PENDING, &con->flags)) |
1552 | send_to_sock(con); | |
fdda387f PC |
1553 | } |
1554 | ||
1555 | ||
1556 | /* Discard all entries on the write queues */ | |
1557 | static void clean_writequeues(void) | |
1558 | { | |
5e9ccc37 | 1559 | foreach_conn(clean_one_writequeue); |
fdda387f PC |
1560 | } |
1561 | ||
1d6e8131 | 1562 | static void work_stop(void) |
fdda387f | 1563 | { |
1d6e8131 PC |
1564 | destroy_workqueue(recv_workqueue); |
1565 | destroy_workqueue(send_workqueue); | |
fdda387f PC |
1566 | } |
1567 | ||
1d6e8131 | 1568 | static int work_start(void) |
fdda387f | 1569 | { |
e43f055a DT |
1570 | recv_workqueue = alloc_workqueue("dlm_recv", |
1571 | WQ_UNBOUND | WQ_MEM_RECLAIM, 1); | |
b9d41052 NK |
1572 | if (!recv_workqueue) { |
1573 | log_print("can't start dlm_recv"); | |
1574 | return -ENOMEM; | |
fdda387f | 1575 | } |
fdda387f | 1576 | |
e43f055a DT |
1577 | send_workqueue = alloc_workqueue("dlm_send", |
1578 | WQ_UNBOUND | WQ_MEM_RECLAIM, 1); | |
b9d41052 NK |
1579 | if (!send_workqueue) { |
1580 | log_print("can't start dlm_send"); | |
1d6e8131 | 1581 | destroy_workqueue(recv_workqueue); |
b9d41052 | 1582 | return -ENOMEM; |
fdda387f | 1583 | } |
fdda387f PC |
1584 | |
1585 | return 0; | |
1586 | } | |
1587 | ||
5e9ccc37 | 1588 | static void stop_conn(struct connection *con) |
fdda387f | 1589 | { |
5e9ccc37 | 1590 | con->flags |= 0x0F; |
391fbdc5 | 1591 | if (con->sock && con->sock->sk) |
5e9ccc37 CC |
1592 | con->sock->sk->sk_user_data = NULL; |
1593 | } | |
fdda387f | 1594 | |
5e9ccc37 CC |
1595 | static void free_conn(struct connection *con) |
1596 | { | |
0d737a8c | 1597 | close_connection(con, true, true, true); |
5e9ccc37 CC |
1598 | if (con->othercon) |
1599 | kmem_cache_free(con_cache, con->othercon); | |
1600 | hlist_del(&con->list); | |
1601 | kmem_cache_free(con_cache, con); | |
1602 | } | |
1603 | ||
1604 | void dlm_lowcomms_stop(void) | |
1605 | { | |
ac33d071 | 1606 | /* Set all the flags to prevent any |
fdda387f PC |
1607 | socket activity. |
1608 | */ | |
7a936ce7 | 1609 | mutex_lock(&connections_lock); |
513ef596 | 1610 | dlm_allow_conn = 0; |
5e9ccc37 | 1611 | foreach_conn(stop_conn); |
7a936ce7 | 1612 | mutex_unlock(&connections_lock); |
ac33d071 | 1613 | |
1d6e8131 | 1614 | work_stop(); |
6ed7257b | 1615 | |
7a936ce7 | 1616 | mutex_lock(&connections_lock); |
fdda387f PC |
1617 | clean_writequeues(); |
1618 | ||
5e9ccc37 CC |
1619 | foreach_conn(free_conn); |
1620 | ||
7a936ce7 | 1621 | mutex_unlock(&connections_lock); |
fdda387f PC |
1622 | kmem_cache_destroy(con_cache); |
1623 | } | |
1624 | ||
fdda387f PC |
1625 | int dlm_lowcomms_start(void) |
1626 | { | |
6ed7257b PC |
1627 | int error = -EINVAL; |
1628 | struct connection *con; | |
5e9ccc37 CC |
1629 | int i; |
1630 | ||
1631 | for (i = 0; i < CONN_HASH_SIZE; i++) | |
1632 | INIT_HLIST_HEAD(&connection_hash[i]); | |
fdda387f | 1633 | |
6ed7257b PC |
1634 | init_local(); |
1635 | if (!dlm_local_count) { | |
617e82e1 | 1636 | error = -ENOTCONN; |
fdda387f | 1637 | log_print("no local IP address has been set"); |
513ef596 | 1638 | goto fail; |
fdda387f PC |
1639 | } |
1640 | ||
6ed7257b | 1641 | error = -ENOMEM; |
fdda387f | 1642 | con_cache = kmem_cache_create("dlm_conn", sizeof(struct connection), |
ac33d071 | 1643 | __alignof__(struct connection), 0, |
20c2df83 | 1644 | NULL); |
fdda387f | 1645 | if (!con_cache) |
513ef596 DT |
1646 | goto fail; |
1647 | ||
1648 | error = work_start(); | |
1649 | if (error) | |
1650 | goto fail_destroy; | |
1651 | ||
1652 | dlm_allow_conn = 1; | |
fdda387f | 1653 | |
fdda387f | 1654 | /* Start listening */ |
6ed7257b PC |
1655 | if (dlm_config.ci_protocol == 0) |
1656 | error = tcp_listen_for_all(); | |
1657 | else | |
1658 | error = sctp_listen_for_all(); | |
fdda387f PC |
1659 | if (error) |
1660 | goto fail_unlisten; | |
1661 | ||
fdda387f PC |
1662 | return 0; |
1663 | ||
ac33d071 | 1664 | fail_unlisten: |
513ef596 | 1665 | dlm_allow_conn = 0; |
6ed7257b PC |
1666 | con = nodeid2con(0,0); |
1667 | if (con) { | |
0d737a8c | 1668 | close_connection(con, false, true, true); |
6ed7257b PC |
1669 | kmem_cache_free(con_cache, con); |
1670 | } | |
513ef596 | 1671 | fail_destroy: |
fdda387f | 1672 | kmem_cache_destroy(con_cache); |
513ef596 | 1673 | fail: |
fdda387f PC |
1674 | return error; |
1675 | } | |
36b71a8b DT |
1676 | |
1677 | void dlm_lowcomms_exit(void) | |
1678 | { | |
1679 | struct dlm_node_addr *na, *safe; | |
1680 | ||
1681 | spin_lock(&dlm_node_addrs_spin); | |
1682 | list_for_each_entry_safe(na, safe, &dlm_node_addrs, list) { | |
1683 | list_del(&na->list); | |
1684 | while (na->addr_count--) | |
1685 | kfree(na->addr[na->addr_count]); | |
1686 | kfree(na); | |
1687 | } | |
1688 | spin_unlock(&dlm_node_addrs_spin); | |
1689 | } |