Commit | Line | Data |
---|---|---|
fdda387f PC |
1 | /****************************************************************************** |
2 | ******************************************************************************* | |
3 | ** | |
4 | ** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
5 | ** Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved. | |
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 | |
24 | * be expanded for the cluster infrastructure then that is it's | |
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 | * | |
39 | * I don't see any problem with the recv thread executing the locking | |
40 | * code on behalf of remote processes as the locking code is | |
41 | * short, efficient and never waits. | |
42 | * | |
43 | */ | |
44 | ||
45 | ||
46 | #include <asm/ioctls.h> | |
47 | #include <net/sock.h> | |
48 | #include <net/tcp.h> | |
49 | #include <linux/pagemap.h> | |
50 | ||
51 | #include "dlm_internal.h" | |
52 | #include "lowcomms.h" | |
53 | #include "midcomms.h" | |
54 | #include "config.h" | |
55 | ||
56 | struct cbuf { | |
ac33d071 PC |
57 | unsigned int base; |
58 | unsigned int len; | |
59 | unsigned int mask; | |
fdda387f PC |
60 | }; |
61 | ||
fdda387f | 62 | #define NODE_INCREMENT 32 |
ac33d071 PC |
63 | static void cbuf_add(struct cbuf *cb, int n) |
64 | { | |
65 | cb->len += n; | |
66 | } | |
fdda387f | 67 | |
ac33d071 PC |
68 | static int cbuf_data(struct cbuf *cb) |
69 | { | |
70 | return ((cb->base + cb->len) & cb->mask); | |
71 | } | |
72 | ||
73 | static void cbuf_init(struct cbuf *cb, int size) | |
74 | { | |
75 | cb->base = cb->len = 0; | |
76 | cb->mask = size-1; | |
77 | } | |
78 | ||
79 | static void cbuf_eat(struct cbuf *cb, int n) | |
80 | { | |
81 | cb->len -= n; | |
82 | cb->base += n; | |
83 | cb->base &= cb->mask; | |
84 | } | |
85 | ||
86 | static bool cbuf_empty(struct cbuf *cb) | |
87 | { | |
88 | return cb->len == 0; | |
89 | } | |
fdda387f PC |
90 | |
91 | /* Maximum number of incoming messages to process before | |
ac33d071 | 92 | doing a cond_resched() |
fdda387f PC |
93 | */ |
94 | #define MAX_RX_MSG_COUNT 25 | |
95 | ||
96 | struct connection { | |
97 | struct socket *sock; /* NULL if not connected */ | |
98 | uint32_t nodeid; /* So we know who we are in the list */ | |
f1f1c1cc | 99 | struct mutex sock_mutex; |
fdda387f PC |
100 | unsigned long flags; /* bit 1,2 = We are on the read/write lists */ |
101 | #define CF_READ_PENDING 1 | |
102 | #define CF_WRITE_PENDING 2 | |
103 | #define CF_CONNECT_PENDING 3 | |
104 | #define CF_IS_OTHERCON 4 | |
ac33d071 PC |
105 | struct list_head writequeue; /* List of outgoing writequeue_entries */ |
106 | struct list_head listenlist; /* List of allocated listening sockets */ | |
fdda387f PC |
107 | spinlock_t writequeue_lock; |
108 | int (*rx_action) (struct connection *); /* What to do when active */ | |
109 | struct page *rx_page; | |
110 | struct cbuf cb; | |
111 | int retries; | |
112 | atomic_t waiting_requests; | |
113 | #define MAX_CONNECT_RETRIES 3 | |
114 | struct connection *othercon; | |
1d6e8131 PC |
115 | struct work_struct rwork; /* Receive workqueue */ |
116 | struct work_struct swork; /* Send workqueue */ | |
fdda387f PC |
117 | }; |
118 | #define sock2con(x) ((struct connection *)(x)->sk_user_data) | |
119 | ||
120 | /* An entry waiting to be sent */ | |
121 | struct writequeue_entry { | |
122 | struct list_head list; | |
123 | struct page *page; | |
124 | int offset; | |
125 | int len; | |
126 | int end; | |
127 | int users; | |
128 | struct connection *con; | |
129 | }; | |
130 | ||
131 | static struct sockaddr_storage dlm_local_addr; | |
132 | ||
1d6e8131 PC |
133 | /* Work queues */ |
134 | static struct workqueue_struct *recv_workqueue; | |
135 | static struct workqueue_struct *send_workqueue; | |
fdda387f PC |
136 | |
137 | /* An array of pointers to connections, indexed by NODEID */ | |
138 | static struct connection **connections; | |
ac33d071 | 139 | static DECLARE_MUTEX(connections_lock); |
c80e7c83 | 140 | static struct kmem_cache *con_cache; |
fdda387f | 141 | static int conn_array_size; |
fdda387f | 142 | |
1d6e8131 PC |
143 | static void process_recv_sockets(struct work_struct *work); |
144 | static void process_send_sockets(struct work_struct *work); | |
fdda387f PC |
145 | |
146 | static struct connection *nodeid2con(int nodeid, gfp_t allocation) | |
147 | { | |
148 | struct connection *con = NULL; | |
149 | ||
150 | down(&connections_lock); | |
151 | if (nodeid >= conn_array_size) { | |
152 | int new_size = nodeid + NODE_INCREMENT; | |
153 | struct connection **new_conns; | |
154 | ||
ac33d071 | 155 | new_conns = kzalloc(sizeof(struct connection *) * |
fdda387f PC |
156 | new_size, allocation); |
157 | if (!new_conns) | |
158 | goto finish; | |
159 | ||
fdda387f PC |
160 | memcpy(new_conns, connections, sizeof(struct connection *) * conn_array_size); |
161 | conn_array_size = new_size; | |
162 | kfree(connections); | |
163 | connections = new_conns; | |
164 | ||
165 | } | |
166 | ||
167 | con = connections[nodeid]; | |
168 | if (con == NULL && allocation) { | |
ac33d071 | 169 | con = kmem_cache_zalloc(con_cache, allocation); |
fdda387f PC |
170 | if (!con) |
171 | goto finish; | |
172 | ||
fdda387f | 173 | con->nodeid = nodeid; |
f1f1c1cc | 174 | mutex_init(&con->sock_mutex); |
fdda387f PC |
175 | INIT_LIST_HEAD(&con->writequeue); |
176 | spin_lock_init(&con->writequeue_lock); | |
1d6e8131 PC |
177 | INIT_WORK(&con->swork, process_send_sockets); |
178 | INIT_WORK(&con->rwork, process_recv_sockets); | |
fdda387f PC |
179 | |
180 | connections[nodeid] = con; | |
181 | } | |
182 | ||
ac33d071 | 183 | finish: |
fdda387f PC |
184 | up(&connections_lock); |
185 | return con; | |
186 | } | |
187 | ||
188 | /* Data available on socket or listen socket received a connect */ | |
189 | static void lowcomms_data_ready(struct sock *sk, int count_unused) | |
190 | { | |
191 | struct connection *con = sock2con(sk); | |
192 | ||
1d6e8131 PC |
193 | if (!test_and_set_bit(CF_READ_PENDING, &con->flags)) |
194 | queue_work(recv_workqueue, &con->rwork); | |
fdda387f PC |
195 | } |
196 | ||
197 | static void lowcomms_write_space(struct sock *sk) | |
198 | { | |
199 | struct connection *con = sock2con(sk); | |
200 | ||
1d6e8131 PC |
201 | if (!test_and_set_bit(CF_WRITE_PENDING, &con->flags)) |
202 | queue_work(send_workqueue, &con->swork); | |
fdda387f PC |
203 | } |
204 | ||
205 | static inline void lowcomms_connect_sock(struct connection *con) | |
206 | { | |
1d6e8131 PC |
207 | if (!test_and_set_bit(CF_CONNECT_PENDING, &con->flags)) |
208 | queue_work(send_workqueue, &con->swork); | |
fdda387f PC |
209 | } |
210 | ||
211 | static void lowcomms_state_change(struct sock *sk) | |
212 | { | |
ac33d071 | 213 | if (sk->sk_state == TCP_ESTABLISHED) |
fdda387f | 214 | lowcomms_write_space(sk); |
fdda387f PC |
215 | } |
216 | ||
217 | /* Make a socket active */ | |
218 | static int add_sock(struct socket *sock, struct connection *con) | |
219 | { | |
220 | con->sock = sock; | |
221 | ||
222 | /* Install a data_ready callback */ | |
223 | con->sock->sk->sk_data_ready = lowcomms_data_ready; | |
224 | con->sock->sk->sk_write_space = lowcomms_write_space; | |
225 | con->sock->sk->sk_state_change = lowcomms_state_change; | |
226 | ||
227 | return 0; | |
228 | } | |
229 | ||
230 | /* Add the port number to an IP6 or 4 sockaddr and return the address | |
231 | length */ | |
232 | static void make_sockaddr(struct sockaddr_storage *saddr, uint16_t port, | |
233 | int *addr_len) | |
234 | { | |
ac33d071 PC |
235 | saddr->ss_family = dlm_local_addr.ss_family; |
236 | if (saddr->ss_family == AF_INET) { | |
fdda387f PC |
237 | struct sockaddr_in *in4_addr = (struct sockaddr_in *)saddr; |
238 | in4_addr->sin_port = cpu_to_be16(port); | |
239 | *addr_len = sizeof(struct sockaddr_in); | |
ac33d071 | 240 | } else { |
fdda387f PC |
241 | struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)saddr; |
242 | in6_addr->sin6_port = cpu_to_be16(port); | |
243 | *addr_len = sizeof(struct sockaddr_in6); | |
244 | } | |
245 | } | |
246 | ||
247 | /* Close a remote connection and tidy up */ | |
ac33d071 | 248 | static void close_connection(struct connection *con, bool and_other) |
fdda387f | 249 | { |
f1f1c1cc | 250 | mutex_lock(&con->sock_mutex); |
fdda387f PC |
251 | |
252 | if (con->sock) { | |
253 | sock_release(con->sock); | |
254 | con->sock = NULL; | |
255 | } | |
256 | if (con->othercon && and_other) { | |
ac33d071 PC |
257 | /* Will only re-enter once. */ |
258 | close_connection(con->othercon, false); | |
fdda387f PC |
259 | } |
260 | if (con->rx_page) { | |
261 | __free_page(con->rx_page); | |
262 | con->rx_page = NULL; | |
263 | } | |
264 | con->retries = 0; | |
f1f1c1cc | 265 | mutex_unlock(&con->sock_mutex); |
fdda387f PC |
266 | } |
267 | ||
268 | /* Data received from remote end */ | |
269 | static int receive_from_sock(struct connection *con) | |
270 | { | |
271 | int ret = 0; | |
272 | struct msghdr msg; | |
273 | struct iovec iov[2]; | |
274 | mm_segment_t fs; | |
275 | unsigned len; | |
276 | int r; | |
277 | int call_again_soon = 0; | |
278 | ||
f1f1c1cc | 279 | mutex_lock(&con->sock_mutex); |
fdda387f PC |
280 | |
281 | if (con->sock == NULL) | |
282 | goto out; | |
283 | if (con->rx_page == NULL) { | |
284 | /* | |
285 | * This doesn't need to be atomic, but I think it should | |
286 | * improve performance if it is. | |
287 | */ | |
288 | con->rx_page = alloc_page(GFP_ATOMIC); | |
289 | if (con->rx_page == NULL) | |
290 | goto out_resched; | |
ac33d071 | 291 | cbuf_init(&con->cb, PAGE_CACHE_SIZE); |
fdda387f PC |
292 | } |
293 | ||
294 | msg.msg_control = NULL; | |
295 | msg.msg_controllen = 0; | |
296 | msg.msg_iovlen = 1; | |
297 | msg.msg_iov = iov; | |
298 | msg.msg_name = NULL; | |
299 | msg.msg_namelen = 0; | |
300 | msg.msg_flags = 0; | |
301 | ||
302 | /* | |
303 | * iov[0] is the bit of the circular buffer between the current end | |
304 | * point (cb.base + cb.len) and the end of the buffer. | |
305 | */ | |
ac33d071 PC |
306 | iov[0].iov_len = con->cb.base - cbuf_data(&con->cb); |
307 | iov[0].iov_base = page_address(con->rx_page) + cbuf_data(&con->cb); | |
fdda387f PC |
308 | iov[1].iov_len = 0; |
309 | ||
310 | /* | |
311 | * iov[1] is the bit of the circular buffer between the start of the | |
312 | * buffer and the start of the currently used section (cb.base) | |
313 | */ | |
ac33d071 PC |
314 | if (cbuf_data(&con->cb) >= con->cb.base) { |
315 | iov[0].iov_len = PAGE_CACHE_SIZE - cbuf_data(&con->cb); | |
fdda387f PC |
316 | iov[1].iov_len = con->cb.base; |
317 | iov[1].iov_base = page_address(con->rx_page); | |
318 | msg.msg_iovlen = 2; | |
319 | } | |
320 | len = iov[0].iov_len + iov[1].iov_len; | |
321 | ||
322 | fs = get_fs(); | |
323 | set_fs(get_ds()); | |
324 | r = ret = sock_recvmsg(con->sock, &msg, len, | |
325 | MSG_DONTWAIT | MSG_NOSIGNAL); | |
326 | set_fs(fs); | |
327 | ||
328 | if (ret <= 0) | |
329 | goto out_close; | |
bd44e2b0 PC |
330 | if (ret == -EAGAIN) |
331 | goto out_resched; | |
332 | ||
fdda387f PC |
333 | if (ret == len) |
334 | call_again_soon = 1; | |
ac33d071 | 335 | cbuf_add(&con->cb, ret); |
fdda387f PC |
336 | ret = dlm_process_incoming_buffer(con->nodeid, |
337 | page_address(con->rx_page), | |
338 | con->cb.base, con->cb.len, | |
339 | PAGE_CACHE_SIZE); | |
340 | if (ret == -EBADMSG) { | |
341 | printk(KERN_INFO "dlm: lowcomms: addr=%p, base=%u, len=%u, " | |
342 | "iov_len=%u, iov_base[0]=%p, read=%d\n", | |
343 | page_address(con->rx_page), con->cb.base, con->cb.len, | |
344 | len, iov[0].iov_base, r); | |
345 | } | |
346 | if (ret < 0) | |
347 | goto out_close; | |
ac33d071 | 348 | cbuf_eat(&con->cb, ret); |
fdda387f | 349 | |
ac33d071 | 350 | if (cbuf_empty(&con->cb) && !call_again_soon) { |
fdda387f PC |
351 | __free_page(con->rx_page); |
352 | con->rx_page = NULL; | |
353 | } | |
354 | ||
ac33d071 | 355 | out: |
fdda387f PC |
356 | if (call_again_soon) |
357 | goto out_resched; | |
f1f1c1cc | 358 | mutex_unlock(&con->sock_mutex); |
ac33d071 | 359 | return 0; |
fdda387f | 360 | |
ac33d071 | 361 | out_resched: |
1d6e8131 PC |
362 | if (!test_and_set_bit(CF_READ_PENDING, &con->flags)) |
363 | queue_work(recv_workqueue, &con->rwork); | |
f1f1c1cc | 364 | mutex_unlock(&con->sock_mutex); |
bd44e2b0 | 365 | return -EAGAIN; |
fdda387f | 366 | |
ac33d071 | 367 | out_close: |
f1f1c1cc | 368 | mutex_unlock(&con->sock_mutex); |
fdda387f | 369 | if (ret != -EAGAIN && !test_bit(CF_IS_OTHERCON, &con->flags)) { |
ac33d071 | 370 | close_connection(con, false); |
fdda387f PC |
371 | /* Reconnect when there is something to send */ |
372 | } | |
373 | ||
fdda387f PC |
374 | return ret; |
375 | } | |
376 | ||
377 | /* Listening socket is busy, accept a connection */ | |
378 | static int accept_from_sock(struct connection *con) | |
379 | { | |
380 | int result; | |
381 | struct sockaddr_storage peeraddr; | |
382 | struct socket *newsock; | |
383 | int len; | |
384 | int nodeid; | |
385 | struct connection *newcon; | |
bd44e2b0 | 386 | struct connection *addcon; |
fdda387f PC |
387 | |
388 | memset(&peeraddr, 0, sizeof(peeraddr)); | |
ac33d071 PC |
389 | result = sock_create_kern(dlm_local_addr.ss_family, SOCK_STREAM, |
390 | IPPROTO_TCP, &newsock); | |
fdda387f PC |
391 | if (result < 0) |
392 | return -ENOMEM; | |
393 | ||
f1f1c1cc | 394 | mutex_lock_nested(&con->sock_mutex, 0); |
fdda387f PC |
395 | |
396 | result = -ENOTCONN; | |
397 | if (con->sock == NULL) | |
398 | goto accept_err; | |
399 | ||
400 | newsock->type = con->sock->type; | |
401 | newsock->ops = con->sock->ops; | |
402 | ||
403 | result = con->sock->ops->accept(con->sock, newsock, O_NONBLOCK); | |
404 | if (result < 0) | |
405 | goto accept_err; | |
406 | ||
407 | /* Get the connected socket's peer */ | |
408 | memset(&peeraddr, 0, sizeof(peeraddr)); | |
409 | if (newsock->ops->getname(newsock, (struct sockaddr *)&peeraddr, | |
410 | &len, 2)) { | |
411 | result = -ECONNABORTED; | |
412 | goto accept_err; | |
413 | } | |
414 | ||
415 | /* Get the new node's NODEID */ | |
416 | make_sockaddr(&peeraddr, 0, &len); | |
417 | if (dlm_addr_to_nodeid(&peeraddr, &nodeid)) { | |
ac33d071 | 418 | printk("dlm: connect from non cluster node\n"); |
fdda387f | 419 | sock_release(newsock); |
f1f1c1cc | 420 | mutex_unlock(&con->sock_mutex); |
fdda387f PC |
421 | return -1; |
422 | } | |
423 | ||
424 | log_print("got connection from %d", nodeid); | |
425 | ||
426 | /* Check to see if we already have a connection to this node. This | |
427 | * could happen if the two nodes initiate a connection at roughly | |
428 | * the same time and the connections cross on the wire. | |
429 | * TEMPORARY FIX: | |
430 | * In this case we store the incoming one in "othercon" | |
431 | */ | |
432 | newcon = nodeid2con(nodeid, GFP_KERNEL); | |
433 | if (!newcon) { | |
434 | result = -ENOMEM; | |
435 | goto accept_err; | |
436 | } | |
f1f1c1cc | 437 | mutex_lock_nested(&newcon->sock_mutex, 1); |
fdda387f | 438 | if (newcon->sock) { |
ac33d071 | 439 | struct connection *othercon = newcon->othercon; |
fdda387f PC |
440 | |
441 | if (!othercon) { | |
ac33d071 | 442 | othercon = kmem_cache_zalloc(con_cache, GFP_KERNEL); |
fdda387f PC |
443 | if (!othercon) { |
444 | printk("dlm: failed to allocate incoming socket\n"); | |
f1f1c1cc | 445 | mutex_unlock(&newcon->sock_mutex); |
fdda387f PC |
446 | result = -ENOMEM; |
447 | goto accept_err; | |
448 | } | |
fdda387f PC |
449 | othercon->nodeid = nodeid; |
450 | othercon->rx_action = receive_from_sock; | |
f1f1c1cc | 451 | mutex_init(&othercon->sock_mutex); |
1d6e8131 PC |
452 | INIT_WORK(&othercon->swork, process_send_sockets); |
453 | INIT_WORK(&othercon->rwork, process_recv_sockets); | |
fdda387f PC |
454 | set_bit(CF_IS_OTHERCON, &othercon->flags); |
455 | newcon->othercon = othercon; | |
456 | } | |
457 | othercon->sock = newsock; | |
458 | newsock->sk->sk_user_data = othercon; | |
459 | add_sock(newsock, othercon); | |
bd44e2b0 | 460 | addcon = othercon; |
fdda387f PC |
461 | } |
462 | else { | |
463 | newsock->sk->sk_user_data = newcon; | |
464 | newcon->rx_action = receive_from_sock; | |
465 | add_sock(newsock, newcon); | |
bd44e2b0 | 466 | addcon = newcon; |
fdda387f PC |
467 | } |
468 | ||
f1f1c1cc | 469 | mutex_unlock(&newcon->sock_mutex); |
fdda387f PC |
470 | |
471 | /* | |
472 | * Add it to the active queue in case we got data | |
473 | * beween processing the accept adding the socket | |
474 | * to the read_sockets list | |
475 | */ | |
bd44e2b0 PC |
476 | if (!test_and_set_bit(CF_READ_PENDING, &addcon->flags)) |
477 | queue_work(recv_workqueue, &addcon->rwork); | |
f1f1c1cc | 478 | mutex_unlock(&con->sock_mutex); |
fdda387f PC |
479 | |
480 | return 0; | |
481 | ||
ac33d071 | 482 | accept_err: |
f1f1c1cc | 483 | mutex_unlock(&con->sock_mutex); |
fdda387f PC |
484 | sock_release(newsock); |
485 | ||
486 | if (result != -EAGAIN) | |
487 | printk("dlm: error accepting connection from node: %d\n", result); | |
488 | return result; | |
489 | } | |
490 | ||
491 | /* Connect a new socket to its peer */ | |
ac33d071 | 492 | static void connect_to_sock(struct connection *con) |
fdda387f PC |
493 | { |
494 | int result = -EHOSTUNREACH; | |
495 | struct sockaddr_storage saddr; | |
496 | int addr_len; | |
497 | struct socket *sock; | |
498 | ||
499 | if (con->nodeid == 0) { | |
500 | log_print("attempt to connect sock 0 foiled"); | |
ac33d071 | 501 | return; |
fdda387f PC |
502 | } |
503 | ||
f1f1c1cc | 504 | mutex_lock(&con->sock_mutex); |
fdda387f PC |
505 | if (con->retries++ > MAX_CONNECT_RETRIES) |
506 | goto out; | |
507 | ||
508 | /* Some odd races can cause double-connects, ignore them */ | |
509 | if (con->sock) { | |
510 | result = 0; | |
511 | goto out; | |
512 | } | |
513 | ||
514 | /* Create a socket to communicate with */ | |
ac33d071 PC |
515 | result = sock_create_kern(dlm_local_addr.ss_family, SOCK_STREAM, |
516 | IPPROTO_TCP, &sock); | |
fdda387f PC |
517 | if (result < 0) |
518 | goto out_err; | |
519 | ||
520 | memset(&saddr, 0, sizeof(saddr)); | |
521 | if (dlm_nodeid_to_addr(con->nodeid, &saddr)) | |
ac33d071 | 522 | goto out_err; |
fdda387f PC |
523 | |
524 | sock->sk->sk_user_data = con; | |
525 | con->rx_action = receive_from_sock; | |
526 | ||
68c817a1 | 527 | make_sockaddr(&saddr, dlm_config.ci_tcp_port, &addr_len); |
fdda387f PC |
528 | |
529 | add_sock(sock, con); | |
530 | ||
531 | log_print("connecting to %d", con->nodeid); | |
532 | result = | |
533 | sock->ops->connect(sock, (struct sockaddr *)&saddr, addr_len, | |
ac33d071 | 534 | O_NONBLOCK); |
fdda387f PC |
535 | if (result == -EINPROGRESS) |
536 | result = 0; | |
ac33d071 PC |
537 | if (result == 0) |
538 | goto out; | |
fdda387f | 539 | |
ac33d071 | 540 | out_err: |
fdda387f PC |
541 | if (con->sock) { |
542 | sock_release(con->sock); | |
543 | con->sock = NULL; | |
544 | } | |
545 | /* | |
546 | * Some errors are fatal and this list might need adjusting. For other | |
547 | * errors we try again until the max number of retries is reached. | |
548 | */ | |
549 | if (result != -EHOSTUNREACH && result != -ENETUNREACH && | |
550 | result != -ENETDOWN && result != EINVAL | |
551 | && result != -EPROTONOSUPPORT) { | |
552 | lowcomms_connect_sock(con); | |
553 | result = 0; | |
554 | } | |
ac33d071 | 555 | out: |
f1f1c1cc | 556 | mutex_unlock(&con->sock_mutex); |
ac33d071 | 557 | return; |
fdda387f PC |
558 | } |
559 | ||
ac33d071 PC |
560 | static struct socket *create_listen_sock(struct connection *con, |
561 | struct sockaddr_storage *saddr) | |
fdda387f | 562 | { |
ac33d071 | 563 | struct socket *sock = NULL; |
fdda387f PC |
564 | mm_segment_t fs; |
565 | int result = 0; | |
566 | int one = 1; | |
567 | int addr_len; | |
568 | ||
569 | if (dlm_local_addr.ss_family == AF_INET) | |
570 | addr_len = sizeof(struct sockaddr_in); | |
571 | else | |
572 | addr_len = sizeof(struct sockaddr_in6); | |
573 | ||
574 | /* Create a socket to communicate with */ | |
575 | result = sock_create_kern(dlm_local_addr.ss_family, SOCK_STREAM, IPPROTO_TCP, &sock); | |
576 | if (result < 0) { | |
577 | printk("dlm: Can't create listening comms socket\n"); | |
578 | goto create_out; | |
579 | } | |
580 | ||
581 | fs = get_fs(); | |
582 | set_fs(get_ds()); | |
ac33d071 PC |
583 | result = sock_setsockopt(sock, SOL_SOCKET, SO_REUSEADDR, |
584 | (char *)&one, sizeof(one)); | |
fdda387f PC |
585 | set_fs(fs); |
586 | if (result < 0) { | |
ac33d071 PC |
587 | printk("dlm: Failed to set SO_REUSEADDR on socket: result=%d\n", |
588 | result); | |
fdda387f PC |
589 | } |
590 | sock->sk->sk_user_data = con; | |
591 | con->rx_action = accept_from_sock; | |
592 | con->sock = sock; | |
593 | ||
594 | /* Bind to our port */ | |
68c817a1 | 595 | make_sockaddr(saddr, dlm_config.ci_tcp_port, &addr_len); |
fdda387f PC |
596 | result = sock->ops->bind(sock, (struct sockaddr *) saddr, addr_len); |
597 | if (result < 0) { | |
68c817a1 | 598 | printk("dlm: Can't bind to port %d\n", dlm_config.ci_tcp_port); |
fdda387f PC |
599 | sock_release(sock); |
600 | sock = NULL; | |
601 | con->sock = NULL; | |
602 | goto create_out; | |
603 | } | |
604 | ||
605 | fs = get_fs(); | |
606 | set_fs(get_ds()); | |
607 | ||
ac33d071 PC |
608 | result = sock_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE, |
609 | (char *)&one, sizeof(one)); | |
fdda387f PC |
610 | set_fs(fs); |
611 | if (result < 0) { | |
612 | printk("dlm: Set keepalive failed: %d\n", result); | |
613 | } | |
614 | ||
615 | result = sock->ops->listen(sock, 5); | |
616 | if (result < 0) { | |
bd44e2b0 | 617 | printk("dlm: Can't listen on port %d\n", dlm_config.ci_tcp_port); |
fdda387f PC |
618 | sock_release(sock); |
619 | sock = NULL; | |
620 | goto create_out; | |
621 | } | |
622 | ||
ac33d071 | 623 | create_out: |
fdda387f PC |
624 | return sock; |
625 | } | |
626 | ||
627 | ||
628 | /* Listen on all interfaces */ | |
629 | static int listen_for_all(void) | |
630 | { | |
631 | struct socket *sock = NULL; | |
632 | struct connection *con = nodeid2con(0, GFP_KERNEL); | |
633 | int result = -EINVAL; | |
634 | ||
635 | /* We don't support multi-homed hosts */ | |
fdda387f PC |
636 | set_bit(CF_IS_OTHERCON, &con->flags); |
637 | ||
638 | sock = create_listen_sock(con, &dlm_local_addr); | |
639 | if (sock) { | |
640 | add_sock(sock, con); | |
641 | result = 0; | |
642 | } | |
643 | else { | |
644 | result = -EADDRINUSE; | |
645 | } | |
646 | ||
647 | return result; | |
648 | } | |
649 | ||
650 | ||
651 | ||
652 | static struct writequeue_entry *new_writequeue_entry(struct connection *con, | |
653 | gfp_t allocation) | |
654 | { | |
655 | struct writequeue_entry *entry; | |
656 | ||
657 | entry = kmalloc(sizeof(struct writequeue_entry), allocation); | |
658 | if (!entry) | |
659 | return NULL; | |
660 | ||
661 | entry->page = alloc_page(allocation); | |
662 | if (!entry->page) { | |
663 | kfree(entry); | |
664 | return NULL; | |
665 | } | |
666 | ||
667 | entry->offset = 0; | |
668 | entry->len = 0; | |
669 | entry->end = 0; | |
670 | entry->users = 0; | |
671 | entry->con = con; | |
672 | ||
673 | return entry; | |
674 | } | |
675 | ||
676 | void *dlm_lowcomms_get_buffer(int nodeid, int len, | |
677 | gfp_t allocation, char **ppc) | |
678 | { | |
679 | struct connection *con; | |
680 | struct writequeue_entry *e; | |
681 | int offset = 0; | |
682 | int users = 0; | |
683 | ||
fdda387f PC |
684 | con = nodeid2con(nodeid, allocation); |
685 | if (!con) | |
686 | return NULL; | |
687 | ||
4edde74e | 688 | spin_lock(&con->writequeue_lock); |
fdda387f | 689 | e = list_entry(con->writequeue.prev, struct writequeue_entry, list); |
ac33d071 | 690 | if ((&e->list == &con->writequeue) || |
fdda387f PC |
691 | (PAGE_CACHE_SIZE - e->end < len)) { |
692 | e = NULL; | |
693 | } else { | |
694 | offset = e->end; | |
695 | e->end += len; | |
696 | users = e->users++; | |
697 | } | |
698 | spin_unlock(&con->writequeue_lock); | |
699 | ||
700 | if (e) { | |
ac33d071 | 701 | got_one: |
fdda387f PC |
702 | if (users == 0) |
703 | kmap(e->page); | |
704 | *ppc = page_address(e->page) + offset; | |
705 | return e; | |
706 | } | |
707 | ||
708 | e = new_writequeue_entry(con, allocation); | |
709 | if (e) { | |
710 | spin_lock(&con->writequeue_lock); | |
711 | offset = e->end; | |
712 | e->end += len; | |
713 | users = e->users++; | |
714 | list_add_tail(&e->list, &con->writequeue); | |
715 | spin_unlock(&con->writequeue_lock); | |
716 | goto got_one; | |
717 | } | |
718 | return NULL; | |
719 | } | |
720 | ||
721 | void dlm_lowcomms_commit_buffer(void *mh) | |
722 | { | |
723 | struct writequeue_entry *e = (struct writequeue_entry *)mh; | |
724 | struct connection *con = e->con; | |
725 | int users; | |
726 | ||
4edde74e | 727 | spin_lock(&con->writequeue_lock); |
fdda387f PC |
728 | users = --e->users; |
729 | if (users) | |
730 | goto out; | |
731 | e->len = e->end - e->offset; | |
732 | kunmap(e->page); | |
733 | spin_unlock(&con->writequeue_lock); | |
734 | ||
1d6e8131 PC |
735 | if (!test_and_set_bit(CF_WRITE_PENDING, &con->flags)) { |
736 | queue_work(send_workqueue, &con->swork); | |
fdda387f PC |
737 | } |
738 | return; | |
739 | ||
ac33d071 | 740 | out: |
fdda387f PC |
741 | spin_unlock(&con->writequeue_lock); |
742 | return; | |
743 | } | |
744 | ||
745 | static void free_entry(struct writequeue_entry *e) | |
746 | { | |
747 | __free_page(e->page); | |
748 | kfree(e); | |
749 | } | |
750 | ||
751 | /* Send a message */ | |
ac33d071 | 752 | static void send_to_sock(struct connection *con) |
fdda387f PC |
753 | { |
754 | int ret = 0; | |
755 | ssize_t(*sendpage) (struct socket *, struct page *, int, size_t, int); | |
756 | const int msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL; | |
757 | struct writequeue_entry *e; | |
758 | int len, offset; | |
759 | ||
f1f1c1cc | 760 | mutex_lock(&con->sock_mutex); |
fdda387f PC |
761 | if (con->sock == NULL) |
762 | goto out_connect; | |
763 | ||
764 | sendpage = con->sock->ops->sendpage; | |
765 | ||
766 | spin_lock(&con->writequeue_lock); | |
767 | for (;;) { | |
768 | e = list_entry(con->writequeue.next, struct writequeue_entry, | |
769 | list); | |
770 | if ((struct list_head *) e == &con->writequeue) | |
771 | break; | |
772 | ||
773 | len = e->len; | |
774 | offset = e->offset; | |
775 | BUG_ON(len == 0 && e->users == 0); | |
776 | spin_unlock(&con->writequeue_lock); | |
1d6e8131 | 777 | kmap(e->page); |
fdda387f PC |
778 | |
779 | ret = 0; | |
780 | if (len) { | |
781 | ret = sendpage(con->sock, e->page, offset, len, | |
782 | msg_flags); | |
783 | if (ret == -EAGAIN || ret == 0) | |
784 | goto out; | |
785 | if (ret <= 0) | |
786 | goto send_error; | |
787 | } | |
788 | else { | |
789 | /* Don't starve people filling buffers */ | |
ac33d071 | 790 | cond_resched(); |
fdda387f PC |
791 | } |
792 | ||
793 | spin_lock(&con->writequeue_lock); | |
794 | e->offset += ret; | |
795 | e->len -= ret; | |
796 | ||
797 | if (e->len == 0 && e->users == 0) { | |
798 | list_del(&e->list); | |
ac33d071 | 799 | kunmap(e->page); |
fdda387f PC |
800 | free_entry(e); |
801 | continue; | |
802 | } | |
803 | } | |
804 | spin_unlock(&con->writequeue_lock); | |
ac33d071 | 805 | out: |
f1f1c1cc | 806 | mutex_unlock(&con->sock_mutex); |
ac33d071 | 807 | return; |
fdda387f | 808 | |
ac33d071 | 809 | send_error: |
f1f1c1cc | 810 | mutex_unlock(&con->sock_mutex); |
ac33d071 | 811 | close_connection(con, false); |
fdda387f | 812 | lowcomms_connect_sock(con); |
ac33d071 | 813 | return; |
fdda387f | 814 | |
ac33d071 | 815 | out_connect: |
f1f1c1cc | 816 | mutex_unlock(&con->sock_mutex); |
bd44e2b0 | 817 | connect_to_sock(con); |
ac33d071 | 818 | return; |
fdda387f PC |
819 | } |
820 | ||
821 | static void clean_one_writequeue(struct connection *con) | |
822 | { | |
823 | struct list_head *list; | |
824 | struct list_head *temp; | |
825 | ||
826 | spin_lock(&con->writequeue_lock); | |
827 | list_for_each_safe(list, temp, &con->writequeue) { | |
828 | struct writequeue_entry *e = | |
829 | list_entry(list, struct writequeue_entry, list); | |
830 | list_del(&e->list); | |
831 | free_entry(e); | |
832 | } | |
833 | spin_unlock(&con->writequeue_lock); | |
834 | } | |
835 | ||
836 | /* Called from recovery when it knows that a node has | |
837 | left the cluster */ | |
838 | int dlm_lowcomms_close(int nodeid) | |
839 | { | |
840 | struct connection *con; | |
841 | ||
842 | if (!connections) | |
843 | goto out; | |
844 | ||
845 | log_print("closing connection to node %d", nodeid); | |
846 | con = nodeid2con(nodeid, 0); | |
847 | if (con) { | |
848 | clean_one_writequeue(con); | |
ac33d071 | 849 | close_connection(con, true); |
fdda387f PC |
850 | atomic_set(&con->waiting_requests, 0); |
851 | } | |
852 | return 0; | |
853 | ||
ac33d071 | 854 | out: |
fdda387f PC |
855 | return -1; |
856 | } | |
857 | ||
fdda387f | 858 | /* Look for activity on active sockets */ |
1d6e8131 | 859 | static void process_recv_sockets(struct work_struct *work) |
fdda387f | 860 | { |
1d6e8131 PC |
861 | struct connection *con = container_of(work, struct connection, rwork); |
862 | int err; | |
fdda387f | 863 | |
1d6e8131 PC |
864 | clear_bit(CF_READ_PENDING, &con->flags); |
865 | do { | |
866 | err = con->rx_action(con); | |
867 | } while (!err); | |
fdda387f PC |
868 | } |
869 | ||
fdda387f | 870 | |
1d6e8131 | 871 | static void process_send_sockets(struct work_struct *work) |
fdda387f | 872 | { |
1d6e8131 | 873 | struct connection *con = container_of(work, struct connection, swork); |
fdda387f | 874 | |
1d6e8131 | 875 | if (test_and_clear_bit(CF_CONNECT_PENDING, &con->flags)) { |
ac33d071 | 876 | connect_to_sock(con); |
fdda387f | 877 | } |
1d6e8131 | 878 | |
bd44e2b0 PC |
879 | clear_bit(CF_WRITE_PENDING, &con->flags); |
880 | send_to_sock(con); | |
fdda387f PC |
881 | } |
882 | ||
883 | ||
884 | /* Discard all entries on the write queues */ | |
885 | static void clean_writequeues(void) | |
886 | { | |
887 | int nodeid; | |
888 | ||
889 | for (nodeid = 1; nodeid < conn_array_size; nodeid++) { | |
890 | struct connection *con = nodeid2con(nodeid, 0); | |
891 | ||
892 | if (con) | |
893 | clean_one_writequeue(con); | |
894 | } | |
895 | } | |
896 | ||
1d6e8131 | 897 | static void work_stop(void) |
fdda387f | 898 | { |
1d6e8131 PC |
899 | destroy_workqueue(recv_workqueue); |
900 | destroy_workqueue(send_workqueue); | |
fdda387f PC |
901 | } |
902 | ||
1d6e8131 | 903 | static int work_start(void) |
fdda387f | 904 | { |
fdda387f | 905 | int error; |
1d6e8131 PC |
906 | recv_workqueue = create_workqueue("dlm_recv"); |
907 | error = IS_ERR(recv_workqueue); | |
ac33d071 | 908 | if (error) { |
1d6e8131 | 909 | log_print("can't start dlm_recv %d", error); |
fdda387f PC |
910 | return error; |
911 | } | |
fdda387f | 912 | |
1d6e8131 PC |
913 | send_workqueue = create_singlethread_workqueue("dlm_send"); |
914 | error = IS_ERR(send_workqueue); | |
ac33d071 | 915 | if (error) { |
1d6e8131 PC |
916 | log_print("can't start dlm_send %d", error); |
917 | destroy_workqueue(recv_workqueue); | |
fdda387f PC |
918 | return error; |
919 | } | |
fdda387f PC |
920 | |
921 | return 0; | |
922 | } | |
923 | ||
fdda387f PC |
924 | void dlm_lowcomms_stop(void) |
925 | { | |
926 | int i; | |
927 | ||
ac33d071 | 928 | /* Set all the flags to prevent any |
fdda387f PC |
929 | socket activity. |
930 | */ | |
931 | for (i = 0; i < conn_array_size; i++) { | |
932 | if (connections[i]) | |
ac33d071 | 933 | connections[i]->flags |= 0xFF; |
fdda387f | 934 | } |
ac33d071 | 935 | |
1d6e8131 | 936 | work_stop(); |
fdda387f PC |
937 | clean_writequeues(); |
938 | ||
939 | for (i = 0; i < conn_array_size; i++) { | |
940 | if (connections[i]) { | |
ac33d071 | 941 | close_connection(connections[i], true); |
fdda387f PC |
942 | if (connections[i]->othercon) |
943 | kmem_cache_free(con_cache, connections[i]->othercon); | |
944 | kmem_cache_free(con_cache, connections[i]); | |
945 | } | |
946 | } | |
947 | ||
948 | kfree(connections); | |
949 | connections = NULL; | |
950 | ||
951 | kmem_cache_destroy(con_cache); | |
952 | } | |
953 | ||
954 | /* This is quite likely to sleep... */ | |
955 | int dlm_lowcomms_start(void) | |
956 | { | |
957 | int error = 0; | |
958 | ||
fdda387f | 959 | error = -ENOMEM; |
ac33d071 | 960 | connections = kzalloc(sizeof(struct connection *) * |
fdda387f PC |
961 | NODE_INCREMENT, GFP_KERNEL); |
962 | if (!connections) | |
963 | goto out; | |
964 | ||
fdda387f PC |
965 | conn_array_size = NODE_INCREMENT; |
966 | ||
967 | if (dlm_our_addr(&dlm_local_addr, 0)) { | |
968 | log_print("no local IP address has been set"); | |
969 | goto fail_free_conn; | |
970 | } | |
971 | if (!dlm_our_addr(&dlm_local_addr, 1)) { | |
972 | log_print("This dlm comms module does not support multi-homed clustering"); | |
973 | goto fail_free_conn; | |
974 | } | |
975 | ||
976 | con_cache = kmem_cache_create("dlm_conn", sizeof(struct connection), | |
ac33d071 PC |
977 | __alignof__(struct connection), 0, |
978 | NULL, NULL); | |
fdda387f PC |
979 | if (!con_cache) |
980 | goto fail_free_conn; | |
981 | ||
982 | ||
983 | /* Start listening */ | |
984 | error = listen_for_all(); | |
985 | if (error) | |
986 | goto fail_unlisten; | |
987 | ||
1d6e8131 | 988 | error = work_start(); |
fdda387f PC |
989 | if (error) |
990 | goto fail_unlisten; | |
991 | ||
fdda387f PC |
992 | return 0; |
993 | ||
ac33d071 PC |
994 | fail_unlisten: |
995 | close_connection(connections[0], false); | |
fdda387f PC |
996 | kmem_cache_free(con_cache, connections[0]); |
997 | kmem_cache_destroy(con_cache); | |
998 | ||
ac33d071 | 999 | fail_free_conn: |
fdda387f PC |
1000 | kfree(connections); |
1001 | ||
ac33d071 | 1002 | out: |
fdda387f PC |
1003 | return error; |
1004 | } | |
1005 | ||
fdda387f PC |
1006 | /* |
1007 | * Overrides for Emacs so that we follow Linus's tabbing style. | |
1008 | * Emacs will notice this stuff at the end of the file and automatically | |
1009 | * adjust the settings for this buffer only. This must remain at the end | |
1010 | * of the file. | |
1011 | * --------------------------------------------------------------------------- | |
1012 | * Local variables: | |
1013 | * c-file-style: "linux" | |
1014 | * End: | |
1015 | */ |