]>
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 PC |
53 | #include <linux/sctp.h> |
54 | #include <net/sctp/user.h> | |
44ad532b | 55 | #include <net/ipv6.h> |
fdda387f PC |
56 | |
57 | #include "dlm_internal.h" | |
58 | #include "lowcomms.h" | |
59 | #include "midcomms.h" | |
60 | #include "config.h" | |
61 | ||
6ed7257b | 62 | #define NEEDED_RMEM (4*1024*1024) |
5e9ccc37 | 63 | #define CONN_HASH_SIZE 32 |
6ed7257b | 64 | |
fdda387f | 65 | struct cbuf { |
ac33d071 PC |
66 | unsigned int base; |
67 | unsigned int len; | |
68 | unsigned int mask; | |
fdda387f PC |
69 | }; |
70 | ||
ac33d071 PC |
71 | static void cbuf_add(struct cbuf *cb, int n) |
72 | { | |
73 | cb->len += n; | |
74 | } | |
fdda387f | 75 | |
ac33d071 PC |
76 | static int cbuf_data(struct cbuf *cb) |
77 | { | |
78 | return ((cb->base + cb->len) & cb->mask); | |
79 | } | |
80 | ||
81 | static void cbuf_init(struct cbuf *cb, int size) | |
82 | { | |
83 | cb->base = cb->len = 0; | |
84 | cb->mask = size-1; | |
85 | } | |
86 | ||
87 | static void cbuf_eat(struct cbuf *cb, int n) | |
88 | { | |
89 | cb->len -= n; | |
90 | cb->base += n; | |
91 | cb->base &= cb->mask; | |
92 | } | |
93 | ||
94 | static bool cbuf_empty(struct cbuf *cb) | |
95 | { | |
96 | return cb->len == 0; | |
97 | } | |
fdda387f | 98 | |
fdda387f PC |
99 | struct connection { |
100 | struct socket *sock; /* NULL if not connected */ | |
101 | uint32_t nodeid; /* So we know who we are in the list */ | |
f1f1c1cc | 102 | struct mutex sock_mutex; |
6ed7257b | 103 | unsigned long flags; |
fdda387f PC |
104 | #define CF_READ_PENDING 1 |
105 | #define CF_WRITE_PENDING 2 | |
106 | #define CF_CONNECT_PENDING 3 | |
6ed7257b PC |
107 | #define CF_INIT_PENDING 4 |
108 | #define CF_IS_OTHERCON 5 | |
ac33d071 | 109 | struct list_head writequeue; /* List of outgoing writequeue_entries */ |
fdda387f PC |
110 | spinlock_t writequeue_lock; |
111 | int (*rx_action) (struct connection *); /* What to do when active */ | |
6ed7257b | 112 | void (*connect_action) (struct connection *); /* What to do to connect */ |
fdda387f PC |
113 | struct page *rx_page; |
114 | struct cbuf cb; | |
115 | int retries; | |
fdda387f | 116 | #define MAX_CONNECT_RETRIES 3 |
6ed7257b | 117 | int sctp_assoc; |
5e9ccc37 | 118 | struct hlist_node list; |
fdda387f | 119 | struct connection *othercon; |
1d6e8131 PC |
120 | struct work_struct rwork; /* Receive workqueue */ |
121 | struct work_struct swork; /* Send workqueue */ | |
fdda387f PC |
122 | }; |
123 | #define sock2con(x) ((struct connection *)(x)->sk_user_data) | |
124 | ||
125 | /* An entry waiting to be sent */ | |
126 | struct writequeue_entry { | |
127 | struct list_head list; | |
128 | struct page *page; | |
129 | int offset; | |
130 | int len; | |
131 | int end; | |
132 | int users; | |
133 | struct connection *con; | |
134 | }; | |
135 | ||
6ed7257b PC |
136 | static struct sockaddr_storage *dlm_local_addr[DLM_MAX_ADDR_COUNT]; |
137 | static int dlm_local_count; | |
fdda387f | 138 | |
1d6e8131 PC |
139 | /* Work queues */ |
140 | static struct workqueue_struct *recv_workqueue; | |
141 | static struct workqueue_struct *send_workqueue; | |
fdda387f | 142 | |
5e9ccc37 | 143 | static struct hlist_head connection_hash[CONN_HASH_SIZE]; |
7a936ce7 | 144 | static DEFINE_MUTEX(connections_lock); |
c80e7c83 | 145 | static struct kmem_cache *con_cache; |
fdda387f | 146 | |
1d6e8131 PC |
147 | static void process_recv_sockets(struct work_struct *work); |
148 | static void process_send_sockets(struct work_struct *work); | |
fdda387f | 149 | |
5e9ccc37 CC |
150 | |
151 | /* This is deliberately very simple because most clusters have simple | |
152 | sequential nodeids, so we should be able to go straight to a connection | |
153 | struct in the array */ | |
154 | static inline int nodeid_hash(int nodeid) | |
155 | { | |
156 | return nodeid & (CONN_HASH_SIZE-1); | |
157 | } | |
158 | ||
159 | static struct connection *__find_con(int nodeid) | |
160 | { | |
161 | int r; | |
162 | struct hlist_node *h; | |
163 | struct connection *con; | |
164 | ||
165 | r = nodeid_hash(nodeid); | |
166 | ||
167 | hlist_for_each_entry(con, h, &connection_hash[r], list) { | |
168 | if (con->nodeid == nodeid) | |
169 | return con; | |
170 | } | |
171 | return NULL; | |
172 | } | |
173 | ||
6ed7257b PC |
174 | /* |
175 | * If 'allocation' is zero then we don't attempt to create a new | |
176 | * connection structure for this node. | |
177 | */ | |
178 | static struct connection *__nodeid2con(int nodeid, gfp_t alloc) | |
fdda387f PC |
179 | { |
180 | struct connection *con = NULL; | |
6ed7257b | 181 | int r; |
fdda387f | 182 | |
5e9ccc37 | 183 | con = __find_con(nodeid); |
6ed7257b PC |
184 | if (con || !alloc) |
185 | return con; | |
fdda387f | 186 | |
6ed7257b PC |
187 | con = kmem_cache_zalloc(con_cache, alloc); |
188 | if (!con) | |
189 | return NULL; | |
fdda387f | 190 | |
5e9ccc37 CC |
191 | r = nodeid_hash(nodeid); |
192 | hlist_add_head(&con->list, &connection_hash[r]); | |
fdda387f | 193 | |
6ed7257b PC |
194 | con->nodeid = nodeid; |
195 | mutex_init(&con->sock_mutex); | |
196 | INIT_LIST_HEAD(&con->writequeue); | |
197 | spin_lock_init(&con->writequeue_lock); | |
198 | INIT_WORK(&con->swork, process_send_sockets); | |
199 | INIT_WORK(&con->rwork, process_recv_sockets); | |
fdda387f | 200 | |
6ed7257b PC |
201 | /* Setup action pointers for child sockets */ |
202 | if (con->nodeid) { | |
5e9ccc37 | 203 | struct connection *zerocon = __find_con(0); |
fdda387f | 204 | |
6ed7257b PC |
205 | con->connect_action = zerocon->connect_action; |
206 | if (!con->rx_action) | |
207 | con->rx_action = zerocon->rx_action; | |
fdda387f PC |
208 | } |
209 | ||
6ed7257b PC |
210 | return con; |
211 | } | |
212 | ||
5e9ccc37 CC |
213 | /* Loop round all connections */ |
214 | static void foreach_conn(void (*conn_func)(struct connection *c)) | |
215 | { | |
216 | int i; | |
217 | struct hlist_node *h, *n; | |
218 | struct connection *con; | |
219 | ||
220 | for (i = 0; i < CONN_HASH_SIZE; i++) { | |
221 | hlist_for_each_entry_safe(con, h, n, &connection_hash[i], list){ | |
222 | conn_func(con); | |
223 | } | |
224 | } | |
225 | } | |
226 | ||
6ed7257b PC |
227 | static struct connection *nodeid2con(int nodeid, gfp_t allocation) |
228 | { | |
229 | struct connection *con; | |
230 | ||
7a936ce7 | 231 | mutex_lock(&connections_lock); |
6ed7257b | 232 | con = __nodeid2con(nodeid, allocation); |
7a936ce7 | 233 | mutex_unlock(&connections_lock); |
6ed7257b | 234 | |
fdda387f PC |
235 | return con; |
236 | } | |
237 | ||
6ed7257b PC |
238 | /* This is a bit drastic, but only called when things go wrong */ |
239 | static struct connection *assoc2con(int assoc_id) | |
240 | { | |
241 | int i; | |
5e9ccc37 | 242 | struct hlist_node *h; |
6ed7257b PC |
243 | struct connection *con; |
244 | ||
7a936ce7 | 245 | mutex_lock(&connections_lock); |
5e9ccc37 CC |
246 | |
247 | for (i = 0 ; i < CONN_HASH_SIZE; i++) { | |
248 | hlist_for_each_entry(con, h, &connection_hash[i], list) { | |
249 | if (con && con->sctp_assoc == assoc_id) { | |
250 | mutex_unlock(&connections_lock); | |
251 | return con; | |
252 | } | |
6ed7257b PC |
253 | } |
254 | } | |
7a936ce7 | 255 | mutex_unlock(&connections_lock); |
6ed7257b PC |
256 | return NULL; |
257 | } | |
258 | ||
259 | static int nodeid_to_addr(int nodeid, struct sockaddr *retaddr) | |
260 | { | |
261 | struct sockaddr_storage addr; | |
262 | int error; | |
263 | ||
264 | if (!dlm_local_count) | |
265 | return -1; | |
266 | ||
267 | error = dlm_nodeid_to_addr(nodeid, &addr); | |
268 | if (error) | |
269 | return error; | |
270 | ||
271 | if (dlm_local_addr[0]->ss_family == AF_INET) { | |
272 | struct sockaddr_in *in4 = (struct sockaddr_in *) &addr; | |
273 | struct sockaddr_in *ret4 = (struct sockaddr_in *) retaddr; | |
274 | ret4->sin_addr.s_addr = in4->sin_addr.s_addr; | |
275 | } else { | |
276 | struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) &addr; | |
277 | struct sockaddr_in6 *ret6 = (struct sockaddr_in6 *) retaddr; | |
44ad532b | 278 | ipv6_addr_copy(&ret6->sin6_addr, &in6->sin6_addr); |
6ed7257b PC |
279 | } |
280 | ||
281 | return 0; | |
282 | } | |
283 | ||
fdda387f PC |
284 | /* Data available on socket or listen socket received a connect */ |
285 | static void lowcomms_data_ready(struct sock *sk, int count_unused) | |
286 | { | |
287 | struct connection *con = sock2con(sk); | |
afb853fb | 288 | if (con && !test_and_set_bit(CF_READ_PENDING, &con->flags)) |
1d6e8131 | 289 | queue_work(recv_workqueue, &con->rwork); |
fdda387f PC |
290 | } |
291 | ||
292 | static void lowcomms_write_space(struct sock *sk) | |
293 | { | |
294 | struct connection *con = sock2con(sk); | |
295 | ||
afb853fb | 296 | if (con && !test_and_set_bit(CF_WRITE_PENDING, &con->flags)) |
1d6e8131 | 297 | queue_work(send_workqueue, &con->swork); |
fdda387f PC |
298 | } |
299 | ||
300 | static inline void lowcomms_connect_sock(struct connection *con) | |
301 | { | |
1d6e8131 PC |
302 | if (!test_and_set_bit(CF_CONNECT_PENDING, &con->flags)) |
303 | queue_work(send_workqueue, &con->swork); | |
fdda387f PC |
304 | } |
305 | ||
306 | static void lowcomms_state_change(struct sock *sk) | |
307 | { | |
ac33d071 | 308 | if (sk->sk_state == TCP_ESTABLISHED) |
fdda387f | 309 | lowcomms_write_space(sk); |
fdda387f PC |
310 | } |
311 | ||
391fbdc5 CC |
312 | int dlm_lowcomms_connect_node(int nodeid) |
313 | { | |
314 | struct connection *con; | |
315 | ||
316 | if (nodeid == dlm_our_nodeid()) | |
317 | return 0; | |
318 | ||
319 | con = nodeid2con(nodeid, GFP_NOFS); | |
320 | if (!con) | |
321 | return -ENOMEM; | |
322 | lowcomms_connect_sock(con); | |
323 | return 0; | |
324 | } | |
325 | ||
fdda387f PC |
326 | /* Make a socket active */ |
327 | static int add_sock(struct socket *sock, struct connection *con) | |
328 | { | |
329 | con->sock = sock; | |
330 | ||
331 | /* Install a data_ready callback */ | |
332 | con->sock->sk->sk_data_ready = lowcomms_data_ready; | |
333 | con->sock->sk->sk_write_space = lowcomms_write_space; | |
334 | con->sock->sk->sk_state_change = lowcomms_state_change; | |
6ed7257b | 335 | con->sock->sk->sk_user_data = con; |
d6d7b702 | 336 | con->sock->sk->sk_allocation = GFP_NOFS; |
fdda387f PC |
337 | return 0; |
338 | } | |
339 | ||
6ed7257b | 340 | /* Add the port number to an IPv6 or 4 sockaddr and return the address |
fdda387f PC |
341 | length */ |
342 | static void make_sockaddr(struct sockaddr_storage *saddr, uint16_t port, | |
343 | int *addr_len) | |
344 | { | |
6ed7257b | 345 | saddr->ss_family = dlm_local_addr[0]->ss_family; |
ac33d071 | 346 | if (saddr->ss_family == AF_INET) { |
fdda387f PC |
347 | struct sockaddr_in *in4_addr = (struct sockaddr_in *)saddr; |
348 | in4_addr->sin_port = cpu_to_be16(port); | |
349 | *addr_len = sizeof(struct sockaddr_in); | |
6ed7257b | 350 | memset(&in4_addr->sin_zero, 0, sizeof(in4_addr->sin_zero)); |
ac33d071 | 351 | } else { |
fdda387f PC |
352 | struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)saddr; |
353 | in6_addr->sin6_port = cpu_to_be16(port); | |
354 | *addr_len = sizeof(struct sockaddr_in6); | |
355 | } | |
01c8cab2 | 356 | memset((char *)saddr + *addr_len, 0, sizeof(struct sockaddr_storage) - *addr_len); |
fdda387f PC |
357 | } |
358 | ||
359 | /* Close a remote connection and tidy up */ | |
ac33d071 | 360 | static void close_connection(struct connection *con, bool and_other) |
fdda387f | 361 | { |
f1f1c1cc | 362 | mutex_lock(&con->sock_mutex); |
fdda387f PC |
363 | |
364 | if (con->sock) { | |
365 | sock_release(con->sock); | |
366 | con->sock = NULL; | |
367 | } | |
368 | if (con->othercon && and_other) { | |
ac33d071 PC |
369 | /* Will only re-enter once. */ |
370 | close_connection(con->othercon, false); | |
fdda387f PC |
371 | } |
372 | if (con->rx_page) { | |
373 | __free_page(con->rx_page); | |
374 | con->rx_page = NULL; | |
375 | } | |
9e5f2825 | 376 | |
61d96be0 PC |
377 | con->retries = 0; |
378 | mutex_unlock(&con->sock_mutex); | |
fdda387f PC |
379 | } |
380 | ||
6ed7257b PC |
381 | /* We only send shutdown messages to nodes that are not part of the cluster */ |
382 | static void sctp_send_shutdown(sctp_assoc_t associd) | |
383 | { | |
384 | static char outcmsg[CMSG_SPACE(sizeof(struct sctp_sndrcvinfo))]; | |
385 | struct msghdr outmessage; | |
386 | struct cmsghdr *cmsg; | |
387 | struct sctp_sndrcvinfo *sinfo; | |
388 | int ret; | |
389 | struct connection *con; | |
390 | ||
391 | con = nodeid2con(0,0); | |
392 | BUG_ON(con == NULL); | |
393 | ||
394 | outmessage.msg_name = NULL; | |
395 | outmessage.msg_namelen = 0; | |
396 | outmessage.msg_control = outcmsg; | |
397 | outmessage.msg_controllen = sizeof(outcmsg); | |
398 | outmessage.msg_flags = MSG_EOR; | |
399 | ||
400 | cmsg = CMSG_FIRSTHDR(&outmessage); | |
401 | cmsg->cmsg_level = IPPROTO_SCTP; | |
402 | cmsg->cmsg_type = SCTP_SNDRCV; | |
403 | cmsg->cmsg_len = CMSG_LEN(sizeof(struct sctp_sndrcvinfo)); | |
404 | outmessage.msg_controllen = cmsg->cmsg_len; | |
405 | sinfo = CMSG_DATA(cmsg); | |
406 | memset(sinfo, 0x00, sizeof(struct sctp_sndrcvinfo)); | |
407 | ||
408 | sinfo->sinfo_flags |= MSG_EOF; | |
409 | sinfo->sinfo_assoc_id = associd; | |
410 | ||
411 | ret = kernel_sendmsg(con->sock, &outmessage, NULL, 0, 0); | |
412 | ||
413 | if (ret != 0) | |
414 | log_print("send EOF to node failed: %d", ret); | |
415 | } | |
416 | ||
5e9ccc37 CC |
417 | static void sctp_init_failed_foreach(struct connection *con) |
418 | { | |
419 | con->sctp_assoc = 0; | |
420 | if (test_and_clear_bit(CF_CONNECT_PENDING, &con->flags)) { | |
421 | if (!test_and_set_bit(CF_WRITE_PENDING, &con->flags)) | |
422 | queue_work(send_workqueue, &con->swork); | |
423 | } | |
424 | } | |
425 | ||
6ed7257b PC |
426 | /* INIT failed but we don't know which node... |
427 | restart INIT on all pending nodes */ | |
428 | static void sctp_init_failed(void) | |
429 | { | |
7a936ce7 | 430 | mutex_lock(&connections_lock); |
5e9ccc37 CC |
431 | |
432 | foreach_conn(sctp_init_failed_foreach); | |
433 | ||
7a936ce7 | 434 | mutex_unlock(&connections_lock); |
6ed7257b PC |
435 | } |
436 | ||
437 | /* Something happened to an association */ | |
617e82e1 DT |
438 | static void process_sctp_notification(struct connection *con, |
439 | struct msghdr *msg, char *buf) | |
6ed7257b PC |
440 | { |
441 | union sctp_notification *sn = (union sctp_notification *)buf; | |
442 | ||
443 | if (sn->sn_header.sn_type == SCTP_ASSOC_CHANGE) { | |
444 | switch (sn->sn_assoc_change.sac_state) { | |
445 | ||
446 | case SCTP_COMM_UP: | |
447 | case SCTP_RESTART: | |
448 | { | |
449 | /* Check that the new node is in the lockspace */ | |
450 | struct sctp_prim prim; | |
451 | int nodeid; | |
452 | int prim_len, ret; | |
453 | int addr_len; | |
454 | struct connection *new_con; | |
455 | struct file *file; | |
456 | sctp_peeloff_arg_t parg; | |
457 | int parglen = sizeof(parg); | |
458 | ||
459 | /* | |
460 | * We get this before any data for an association. | |
461 | * We verify that the node is in the cluster and | |
462 | * then peel off a socket for it. | |
463 | */ | |
464 | if ((int)sn->sn_assoc_change.sac_assoc_id <= 0) { | |
465 | log_print("COMM_UP for invalid assoc ID %d", | |
617e82e1 | 466 | (int)sn->sn_assoc_change.sac_assoc_id); |
6ed7257b PC |
467 | sctp_init_failed(); |
468 | return; | |
469 | } | |
470 | memset(&prim, 0, sizeof(struct sctp_prim)); | |
471 | prim_len = sizeof(struct sctp_prim); | |
472 | prim.ssp_assoc_id = sn->sn_assoc_change.sac_assoc_id; | |
473 | ||
474 | ret = kernel_getsockopt(con->sock, | |
475 | IPPROTO_SCTP, | |
476 | SCTP_PRIMARY_ADDR, | |
477 | (char*)&prim, | |
478 | &prim_len); | |
479 | if (ret < 0) { | |
480 | log_print("getsockopt/sctp_primary_addr on " | |
481 | "new assoc %d failed : %d", | |
482 | (int)sn->sn_assoc_change.sac_assoc_id, | |
483 | ret); | |
484 | ||
485 | /* Retry INIT later */ | |
486 | new_con = assoc2con(sn->sn_assoc_change.sac_assoc_id); | |
487 | if (new_con) | |
488 | clear_bit(CF_CONNECT_PENDING, &con->flags); | |
489 | return; | |
490 | } | |
491 | make_sockaddr(&prim.ssp_addr, 0, &addr_len); | |
492 | if (dlm_addr_to_nodeid(&prim.ssp_addr, &nodeid)) { | |
493 | int i; | |
494 | unsigned char *b=(unsigned char *)&prim.ssp_addr; | |
495 | log_print("reject connect from unknown addr"); | |
496 | for (i=0; i<sizeof(struct sockaddr_storage);i++) | |
497 | printk("%02x ", b[i]); | |
498 | printk("\n"); | |
499 | sctp_send_shutdown(prim.ssp_assoc_id); | |
500 | return; | |
501 | } | |
502 | ||
748285cc | 503 | new_con = nodeid2con(nodeid, GFP_NOFS); |
6ed7257b PC |
504 | if (!new_con) |
505 | return; | |
506 | ||
507 | /* Peel off a new sock */ | |
508 | parg.associd = sn->sn_assoc_change.sac_assoc_id; | |
617e82e1 DT |
509 | ret = kernel_getsockopt(con->sock, IPPROTO_SCTP, |
510 | SCTP_SOCKOPT_PEELOFF, | |
6ed7257b | 511 | (void *)&parg, &parglen); |
30d3a237 | 512 | if (ret) { |
617e82e1 DT |
513 | log_print("Can't peel off a socket for " |
514 | "connection %d to node %d: err=%d\n", | |
6ed7257b | 515 | parg.associd, nodeid, ret); |
6ed7257b | 516 | } |
6ed7257b PC |
517 | file = fget(parg.sd); |
518 | new_con->sock = SOCKET_I(file->f_dentry->d_inode); | |
519 | add_sock(new_con->sock, new_con); | |
520 | fput(file); | |
521 | put_unused_fd(parg.sd); | |
522 | ||
523 | log_print("got new/restarted association %d nodeid %d", | |
617e82e1 | 524 | (int)sn->sn_assoc_change.sac_assoc_id, nodeid); |
6ed7257b PC |
525 | |
526 | /* Send any pending writes */ | |
527 | clear_bit(CF_CONNECT_PENDING, &new_con->flags); | |
528 | clear_bit(CF_INIT_PENDING, &con->flags); | |
529 | if (!test_and_set_bit(CF_WRITE_PENDING, &new_con->flags)) { | |
530 | queue_work(send_workqueue, &new_con->swork); | |
531 | } | |
532 | if (!test_and_set_bit(CF_READ_PENDING, &new_con->flags)) | |
533 | queue_work(recv_workqueue, &new_con->rwork); | |
534 | } | |
535 | break; | |
536 | ||
537 | case SCTP_COMM_LOST: | |
538 | case SCTP_SHUTDOWN_COMP: | |
539 | { | |
540 | con = assoc2con(sn->sn_assoc_change.sac_assoc_id); | |
541 | if (con) { | |
542 | con->sctp_assoc = 0; | |
543 | } | |
544 | } | |
545 | break; | |
546 | ||
547 | /* We don't know which INIT failed, so clear the PENDING flags | |
548 | * on them all. if assoc_id is zero then it will then try | |
549 | * again */ | |
550 | ||
551 | case SCTP_CANT_STR_ASSOC: | |
552 | { | |
553 | log_print("Can't start SCTP association - retrying"); | |
554 | sctp_init_failed(); | |
555 | } | |
556 | break; | |
557 | ||
558 | default: | |
559 | log_print("unexpected SCTP assoc change id=%d state=%d", | |
560 | (int)sn->sn_assoc_change.sac_assoc_id, | |
561 | sn->sn_assoc_change.sac_state); | |
562 | } | |
563 | } | |
564 | } | |
565 | ||
fdda387f PC |
566 | /* Data received from remote end */ |
567 | static int receive_from_sock(struct connection *con) | |
568 | { | |
569 | int ret = 0; | |
58addbff AV |
570 | struct msghdr msg = {}; |
571 | struct kvec iov[2]; | |
fdda387f PC |
572 | unsigned len; |
573 | int r; | |
574 | int call_again_soon = 0; | |
58addbff | 575 | int nvec; |
6ed7257b | 576 | char incmsg[CMSG_SPACE(sizeof(struct sctp_sndrcvinfo))]; |
fdda387f | 577 | |
f1f1c1cc | 578 | mutex_lock(&con->sock_mutex); |
fdda387f | 579 | |
a34fbc63 PC |
580 | if (con->sock == NULL) { |
581 | ret = -EAGAIN; | |
582 | goto out_close; | |
583 | } | |
584 | ||
fdda387f PC |
585 | if (con->rx_page == NULL) { |
586 | /* | |
587 | * This doesn't need to be atomic, but I think it should | |
588 | * improve performance if it is. | |
589 | */ | |
590 | con->rx_page = alloc_page(GFP_ATOMIC); | |
591 | if (con->rx_page == NULL) | |
592 | goto out_resched; | |
ac33d071 | 593 | cbuf_init(&con->cb, PAGE_CACHE_SIZE); |
fdda387f PC |
594 | } |
595 | ||
6ed7257b PC |
596 | /* Only SCTP needs these really */ |
597 | memset(&incmsg, 0, sizeof(incmsg)); | |
598 | msg.msg_control = incmsg; | |
599 | msg.msg_controllen = sizeof(incmsg); | |
600 | ||
fdda387f PC |
601 | /* |
602 | * iov[0] is the bit of the circular buffer between the current end | |
603 | * point (cb.base + cb.len) and the end of the buffer. | |
604 | */ | |
ac33d071 PC |
605 | iov[0].iov_len = con->cb.base - cbuf_data(&con->cb); |
606 | iov[0].iov_base = page_address(con->rx_page) + cbuf_data(&con->cb); | |
89adc934 | 607 | iov[1].iov_len = 0; |
58addbff | 608 | nvec = 1; |
fdda387f PC |
609 | |
610 | /* | |
611 | * iov[1] is the bit of the circular buffer between the start of the | |
612 | * buffer and the start of the currently used section (cb.base) | |
613 | */ | |
ac33d071 PC |
614 | if (cbuf_data(&con->cb) >= con->cb.base) { |
615 | iov[0].iov_len = PAGE_CACHE_SIZE - cbuf_data(&con->cb); | |
fdda387f PC |
616 | iov[1].iov_len = con->cb.base; |
617 | iov[1].iov_base = page_address(con->rx_page); | |
58addbff | 618 | nvec = 2; |
fdda387f PC |
619 | } |
620 | len = iov[0].iov_len + iov[1].iov_len; | |
621 | ||
58addbff | 622 | r = ret = kernel_recvmsg(con->sock, &msg, iov, nvec, len, |
fdda387f | 623 | MSG_DONTWAIT | MSG_NOSIGNAL); |
fdda387f PC |
624 | if (ret <= 0) |
625 | goto out_close; | |
bd44e2b0 | 626 | |
6ed7257b PC |
627 | /* Process SCTP notifications */ |
628 | if (msg.msg_flags & MSG_NOTIFICATION) { | |
6ed7257b PC |
629 | msg.msg_control = incmsg; |
630 | msg.msg_controllen = sizeof(incmsg); | |
631 | ||
632 | process_sctp_notification(con, &msg, | |
617e82e1 | 633 | page_address(con->rx_page) + con->cb.base); |
6ed7257b PC |
634 | mutex_unlock(&con->sock_mutex); |
635 | return 0; | |
636 | } | |
637 | BUG_ON(con->nodeid == 0); | |
638 | ||
fdda387f PC |
639 | if (ret == len) |
640 | call_again_soon = 1; | |
ac33d071 | 641 | cbuf_add(&con->cb, ret); |
fdda387f PC |
642 | ret = dlm_process_incoming_buffer(con->nodeid, |
643 | page_address(con->rx_page), | |
644 | con->cb.base, con->cb.len, | |
645 | PAGE_CACHE_SIZE); | |
646 | if (ret == -EBADMSG) { | |
617e82e1 DT |
647 | log_print("lowcomms: addr=%p, base=%u, len=%u, " |
648 | "iov_len=%u, iov_base[0]=%p, read=%d", | |
649 | page_address(con->rx_page), con->cb.base, con->cb.len, | |
650 | len, iov[0].iov_base, r); | |
fdda387f PC |
651 | } |
652 | if (ret < 0) | |
653 | goto out_close; | |
ac33d071 | 654 | cbuf_eat(&con->cb, ret); |
fdda387f | 655 | |
ac33d071 | 656 | if (cbuf_empty(&con->cb) && !call_again_soon) { |
fdda387f PC |
657 | __free_page(con->rx_page); |
658 | con->rx_page = NULL; | |
659 | } | |
660 | ||
fdda387f PC |
661 | if (call_again_soon) |
662 | goto out_resched; | |
f1f1c1cc | 663 | mutex_unlock(&con->sock_mutex); |
ac33d071 | 664 | return 0; |
fdda387f | 665 | |
ac33d071 | 666 | out_resched: |
1d6e8131 PC |
667 | if (!test_and_set_bit(CF_READ_PENDING, &con->flags)) |
668 | queue_work(recv_workqueue, &con->rwork); | |
f1f1c1cc | 669 | mutex_unlock(&con->sock_mutex); |
bd44e2b0 | 670 | return -EAGAIN; |
fdda387f | 671 | |
ac33d071 | 672 | out_close: |
f1f1c1cc | 673 | mutex_unlock(&con->sock_mutex); |
9e5f2825 | 674 | if (ret != -EAGAIN) { |
ac33d071 | 675 | close_connection(con, false); |
fdda387f PC |
676 | /* Reconnect when there is something to send */ |
677 | } | |
a34fbc63 PC |
678 | /* Don't return success if we really got EOF */ |
679 | if (ret == 0) | |
680 | ret = -EAGAIN; | |
fdda387f | 681 | |
fdda387f PC |
682 | return ret; |
683 | } | |
684 | ||
685 | /* Listening socket is busy, accept a connection */ | |
6ed7257b | 686 | static int tcp_accept_from_sock(struct connection *con) |
fdda387f PC |
687 | { |
688 | int result; | |
689 | struct sockaddr_storage peeraddr; | |
690 | struct socket *newsock; | |
691 | int len; | |
692 | int nodeid; | |
693 | struct connection *newcon; | |
bd44e2b0 | 694 | struct connection *addcon; |
fdda387f PC |
695 | |
696 | memset(&peeraddr, 0, sizeof(peeraddr)); | |
6ed7257b | 697 | result = sock_create_kern(dlm_local_addr[0]->ss_family, SOCK_STREAM, |
ac33d071 | 698 | IPPROTO_TCP, &newsock); |
fdda387f PC |
699 | if (result < 0) |
700 | return -ENOMEM; | |
701 | ||
f1f1c1cc | 702 | mutex_lock_nested(&con->sock_mutex, 0); |
fdda387f PC |
703 | |
704 | result = -ENOTCONN; | |
705 | if (con->sock == NULL) | |
706 | goto accept_err; | |
707 | ||
708 | newsock->type = con->sock->type; | |
709 | newsock->ops = con->sock->ops; | |
710 | ||
711 | result = con->sock->ops->accept(con->sock, newsock, O_NONBLOCK); | |
712 | if (result < 0) | |
713 | goto accept_err; | |
714 | ||
715 | /* Get the connected socket's peer */ | |
716 | memset(&peeraddr, 0, sizeof(peeraddr)); | |
717 | if (newsock->ops->getname(newsock, (struct sockaddr *)&peeraddr, | |
718 | &len, 2)) { | |
719 | result = -ECONNABORTED; | |
720 | goto accept_err; | |
721 | } | |
722 | ||
723 | /* Get the new node's NODEID */ | |
724 | make_sockaddr(&peeraddr, 0, &len); | |
725 | if (dlm_addr_to_nodeid(&peeraddr, &nodeid)) { | |
617e82e1 | 726 | log_print("connect from non cluster node"); |
fdda387f | 727 | sock_release(newsock); |
f1f1c1cc | 728 | mutex_unlock(&con->sock_mutex); |
fdda387f PC |
729 | return -1; |
730 | } | |
731 | ||
732 | log_print("got connection from %d", nodeid); | |
733 | ||
734 | /* Check to see if we already have a connection to this node. This | |
735 | * could happen if the two nodes initiate a connection at roughly | |
736 | * the same time and the connections cross on the wire. | |
fdda387f PC |
737 | * In this case we store the incoming one in "othercon" |
738 | */ | |
748285cc | 739 | newcon = nodeid2con(nodeid, GFP_NOFS); |
fdda387f PC |
740 | if (!newcon) { |
741 | result = -ENOMEM; | |
742 | goto accept_err; | |
743 | } | |
f1f1c1cc | 744 | mutex_lock_nested(&newcon->sock_mutex, 1); |
fdda387f | 745 | if (newcon->sock) { |
ac33d071 | 746 | struct connection *othercon = newcon->othercon; |
fdda387f PC |
747 | |
748 | if (!othercon) { | |
748285cc | 749 | othercon = kmem_cache_zalloc(con_cache, GFP_NOFS); |
fdda387f | 750 | if (!othercon) { |
617e82e1 | 751 | log_print("failed to allocate incoming socket"); |
f1f1c1cc | 752 | mutex_unlock(&newcon->sock_mutex); |
fdda387f PC |
753 | result = -ENOMEM; |
754 | goto accept_err; | |
755 | } | |
fdda387f PC |
756 | othercon->nodeid = nodeid; |
757 | othercon->rx_action = receive_from_sock; | |
f1f1c1cc | 758 | mutex_init(&othercon->sock_mutex); |
1d6e8131 PC |
759 | INIT_WORK(&othercon->swork, process_send_sockets); |
760 | INIT_WORK(&othercon->rwork, process_recv_sockets); | |
fdda387f | 761 | set_bit(CF_IS_OTHERCON, &othercon->flags); |
61d96be0 PC |
762 | } |
763 | if (!othercon->sock) { | |
fdda387f | 764 | newcon->othercon = othercon; |
97d84836 PC |
765 | othercon->sock = newsock; |
766 | newsock->sk->sk_user_data = othercon; | |
767 | add_sock(newsock, othercon); | |
768 | addcon = othercon; | |
769 | } | |
770 | else { | |
771 | printk("Extra connection from node %d attempted\n", nodeid); | |
772 | result = -EAGAIN; | |
f4fadb23 | 773 | mutex_unlock(&newcon->sock_mutex); |
97d84836 | 774 | goto accept_err; |
fdda387f | 775 | } |
fdda387f PC |
776 | } |
777 | else { | |
778 | newsock->sk->sk_user_data = newcon; | |
779 | newcon->rx_action = receive_from_sock; | |
780 | add_sock(newsock, newcon); | |
bd44e2b0 | 781 | addcon = newcon; |
fdda387f PC |
782 | } |
783 | ||
f1f1c1cc | 784 | mutex_unlock(&newcon->sock_mutex); |
fdda387f PC |
785 | |
786 | /* | |
787 | * Add it to the active queue in case we got data | |
788 | * beween processing the accept adding the socket | |
789 | * to the read_sockets list | |
790 | */ | |
bd44e2b0 PC |
791 | if (!test_and_set_bit(CF_READ_PENDING, &addcon->flags)) |
792 | queue_work(recv_workqueue, &addcon->rwork); | |
f1f1c1cc | 793 | mutex_unlock(&con->sock_mutex); |
fdda387f PC |
794 | |
795 | return 0; | |
796 | ||
ac33d071 | 797 | accept_err: |
f1f1c1cc | 798 | mutex_unlock(&con->sock_mutex); |
fdda387f PC |
799 | sock_release(newsock); |
800 | ||
801 | if (result != -EAGAIN) | |
617e82e1 | 802 | log_print("error accepting connection from node: %d", result); |
fdda387f PC |
803 | return result; |
804 | } | |
805 | ||
6ed7257b PC |
806 | static void free_entry(struct writequeue_entry *e) |
807 | { | |
808 | __free_page(e->page); | |
809 | kfree(e); | |
810 | } | |
811 | ||
812 | /* Initiate an SCTP association. | |
813 | This is a special case of send_to_sock() in that we don't yet have a | |
814 | peeled-off socket for this association, so we use the listening socket | |
815 | and add the primary IP address of the remote node. | |
816 | */ | |
817 | static void sctp_init_assoc(struct connection *con) | |
818 | { | |
819 | struct sockaddr_storage rem_addr; | |
820 | char outcmsg[CMSG_SPACE(sizeof(struct sctp_sndrcvinfo))]; | |
821 | struct msghdr outmessage; | |
822 | struct cmsghdr *cmsg; | |
823 | struct sctp_sndrcvinfo *sinfo; | |
824 | struct connection *base_con; | |
825 | struct writequeue_entry *e; | |
826 | int len, offset; | |
827 | int ret; | |
828 | int addrlen; | |
829 | struct kvec iov[1]; | |
830 | ||
831 | if (test_and_set_bit(CF_INIT_PENDING, &con->flags)) | |
832 | return; | |
833 | ||
834 | if (con->retries++ > MAX_CONNECT_RETRIES) | |
835 | return; | |
836 | ||
837 | log_print("Initiating association with node %d", con->nodeid); | |
838 | ||
839 | if (nodeid_to_addr(con->nodeid, (struct sockaddr *)&rem_addr)) { | |
840 | log_print("no address for nodeid %d", con->nodeid); | |
841 | return; | |
842 | } | |
843 | base_con = nodeid2con(0, 0); | |
844 | BUG_ON(base_con == NULL); | |
845 | ||
846 | make_sockaddr(&rem_addr, dlm_config.ci_tcp_port, &addrlen); | |
847 | ||
848 | outmessage.msg_name = &rem_addr; | |
849 | outmessage.msg_namelen = addrlen; | |
850 | outmessage.msg_control = outcmsg; | |
851 | outmessage.msg_controllen = sizeof(outcmsg); | |
852 | outmessage.msg_flags = MSG_EOR; | |
853 | ||
854 | spin_lock(&con->writequeue_lock); | |
855 | e = list_entry(con->writequeue.next, struct writequeue_entry, | |
856 | list); | |
857 | ||
858 | BUG_ON((struct list_head *) e == &con->writequeue); | |
859 | ||
860 | len = e->len; | |
861 | offset = e->offset; | |
862 | spin_unlock(&con->writequeue_lock); | |
6ed7257b PC |
863 | |
864 | /* Send the first block off the write queue */ | |
865 | iov[0].iov_base = page_address(e->page)+offset; | |
866 | iov[0].iov_len = len; | |
867 | ||
868 | cmsg = CMSG_FIRSTHDR(&outmessage); | |
869 | cmsg->cmsg_level = IPPROTO_SCTP; | |
870 | cmsg->cmsg_type = SCTP_SNDRCV; | |
871 | cmsg->cmsg_len = CMSG_LEN(sizeof(struct sctp_sndrcvinfo)); | |
872 | sinfo = CMSG_DATA(cmsg); | |
873 | memset(sinfo, 0x00, sizeof(struct sctp_sndrcvinfo)); | |
874 | sinfo->sinfo_ppid = cpu_to_le32(dlm_our_nodeid()); | |
875 | outmessage.msg_controllen = cmsg->cmsg_len; | |
876 | ||
877 | ret = kernel_sendmsg(base_con->sock, &outmessage, iov, 1, len); | |
878 | if (ret < 0) { | |
617e82e1 DT |
879 | log_print("Send first packet to node %d failed: %d", |
880 | con->nodeid, ret); | |
6ed7257b PC |
881 | |
882 | /* Try again later */ | |
883 | clear_bit(CF_CONNECT_PENDING, &con->flags); | |
884 | clear_bit(CF_INIT_PENDING, &con->flags); | |
885 | } | |
886 | else { | |
887 | spin_lock(&con->writequeue_lock); | |
888 | e->offset += ret; | |
889 | e->len -= ret; | |
890 | ||
891 | if (e->len == 0 && e->users == 0) { | |
892 | list_del(&e->list); | |
6ed7257b PC |
893 | free_entry(e); |
894 | } | |
895 | spin_unlock(&con->writequeue_lock); | |
896 | } | |
897 | } | |
898 | ||
fdda387f | 899 | /* Connect a new socket to its peer */ |
6ed7257b | 900 | static void tcp_connect_to_sock(struct connection *con) |
fdda387f PC |
901 | { |
902 | int result = -EHOSTUNREACH; | |
6bd8feda | 903 | struct sockaddr_storage saddr, src_addr; |
fdda387f PC |
904 | int addr_len; |
905 | struct socket *sock; | |
906 | ||
907 | if (con->nodeid == 0) { | |
908 | log_print("attempt to connect sock 0 foiled"); | |
ac33d071 | 909 | return; |
fdda387f PC |
910 | } |
911 | ||
f1f1c1cc | 912 | mutex_lock(&con->sock_mutex); |
fdda387f PC |
913 | if (con->retries++ > MAX_CONNECT_RETRIES) |
914 | goto out; | |
915 | ||
916 | /* Some odd races can cause double-connects, ignore them */ | |
917 | if (con->sock) { | |
918 | result = 0; | |
919 | goto out; | |
920 | } | |
921 | ||
922 | /* Create a socket to communicate with */ | |
6ed7257b | 923 | result = sock_create_kern(dlm_local_addr[0]->ss_family, SOCK_STREAM, |
ac33d071 | 924 | IPPROTO_TCP, &sock); |
fdda387f PC |
925 | if (result < 0) |
926 | goto out_err; | |
927 | ||
928 | memset(&saddr, 0, sizeof(saddr)); | |
311f6fc7 MY |
929 | if (dlm_nodeid_to_addr(con->nodeid, &saddr)) { |
930 | sock_release(sock); | |
ac33d071 | 931 | goto out_err; |
311f6fc7 | 932 | } |
fdda387f PC |
933 | |
934 | sock->sk->sk_user_data = con; | |
935 | con->rx_action = receive_from_sock; | |
6ed7257b PC |
936 | con->connect_action = tcp_connect_to_sock; |
937 | add_sock(sock, con); | |
fdda387f | 938 | |
6bd8feda LH |
939 | /* Bind to our cluster-known address connecting to avoid |
940 | routing problems */ | |
941 | memcpy(&src_addr, dlm_local_addr[0], sizeof(src_addr)); | |
942 | make_sockaddr(&src_addr, 0, &addr_len); | |
943 | result = sock->ops->bind(sock, (struct sockaddr *) &src_addr, | |
944 | addr_len); | |
945 | if (result < 0) { | |
946 | log_print("could not bind for connect: %d", result); | |
947 | /* This *may* not indicate a critical error */ | |
948 | } | |
949 | ||
68c817a1 | 950 | make_sockaddr(&saddr, dlm_config.ci_tcp_port, &addr_len); |
fdda387f | 951 | |
fdda387f PC |
952 | log_print("connecting to %d", con->nodeid); |
953 | result = | |
954 | sock->ops->connect(sock, (struct sockaddr *)&saddr, addr_len, | |
ac33d071 | 955 | O_NONBLOCK); |
fdda387f PC |
956 | if (result == -EINPROGRESS) |
957 | result = 0; | |
ac33d071 PC |
958 | if (result == 0) |
959 | goto out; | |
fdda387f | 960 | |
ac33d071 | 961 | out_err: |
fdda387f PC |
962 | if (con->sock) { |
963 | sock_release(con->sock); | |
964 | con->sock = NULL; | |
965 | } | |
966 | /* | |
967 | * Some errors are fatal and this list might need adjusting. For other | |
968 | * errors we try again until the max number of retries is reached. | |
969 | */ | |
970 | if (result != -EHOSTUNREACH && result != -ENETUNREACH && | |
0035a4b1 | 971 | result != -ENETDOWN && result != -EINVAL |
fdda387f PC |
972 | && result != -EPROTONOSUPPORT) { |
973 | lowcomms_connect_sock(con); | |
974 | result = 0; | |
975 | } | |
ac33d071 | 976 | out: |
f1f1c1cc | 977 | mutex_unlock(&con->sock_mutex); |
ac33d071 | 978 | return; |
fdda387f PC |
979 | } |
980 | ||
6ed7257b PC |
981 | static struct socket *tcp_create_listen_sock(struct connection *con, |
982 | struct sockaddr_storage *saddr) | |
fdda387f | 983 | { |
ac33d071 | 984 | struct socket *sock = NULL; |
fdda387f PC |
985 | int result = 0; |
986 | int one = 1; | |
987 | int addr_len; | |
988 | ||
6ed7257b | 989 | if (dlm_local_addr[0]->ss_family == AF_INET) |
fdda387f PC |
990 | addr_len = sizeof(struct sockaddr_in); |
991 | else | |
992 | addr_len = sizeof(struct sockaddr_in6); | |
993 | ||
994 | /* Create a socket to communicate with */ | |
617e82e1 DT |
995 | result = sock_create_kern(dlm_local_addr[0]->ss_family, SOCK_STREAM, |
996 | IPPROTO_TCP, &sock); | |
fdda387f | 997 | if (result < 0) { |
617e82e1 | 998 | log_print("Can't create listening comms socket"); |
fdda387f PC |
999 | goto create_out; |
1000 | } | |
1001 | ||
6ed7257b PC |
1002 | result = kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEADDR, |
1003 | (char *)&one, sizeof(one)); | |
1004 | ||
fdda387f | 1005 | if (result < 0) { |
617e82e1 | 1006 | log_print("Failed to set SO_REUSEADDR on socket: %d", result); |
fdda387f PC |
1007 | } |
1008 | sock->sk->sk_user_data = con; | |
6ed7257b PC |
1009 | con->rx_action = tcp_accept_from_sock; |
1010 | con->connect_action = tcp_connect_to_sock; | |
fdda387f PC |
1011 | con->sock = sock; |
1012 | ||
1013 | /* Bind to our port */ | |
68c817a1 | 1014 | make_sockaddr(saddr, dlm_config.ci_tcp_port, &addr_len); |
fdda387f PC |
1015 | result = sock->ops->bind(sock, (struct sockaddr *) saddr, addr_len); |
1016 | if (result < 0) { | |
617e82e1 | 1017 | log_print("Can't bind to port %d", dlm_config.ci_tcp_port); |
fdda387f PC |
1018 | sock_release(sock); |
1019 | sock = NULL; | |
1020 | con->sock = NULL; | |
1021 | goto create_out; | |
1022 | } | |
6ed7257b | 1023 | result = kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE, |
ac33d071 | 1024 | (char *)&one, sizeof(one)); |
fdda387f | 1025 | if (result < 0) { |
617e82e1 | 1026 | log_print("Set keepalive failed: %d", result); |
fdda387f PC |
1027 | } |
1028 | ||
1029 | result = sock->ops->listen(sock, 5); | |
1030 | if (result < 0) { | |
617e82e1 | 1031 | log_print("Can't listen on port %d", dlm_config.ci_tcp_port); |
fdda387f PC |
1032 | sock_release(sock); |
1033 | sock = NULL; | |
1034 | goto create_out; | |
1035 | } | |
1036 | ||
ac33d071 | 1037 | create_out: |
fdda387f PC |
1038 | return sock; |
1039 | } | |
1040 | ||
6ed7257b PC |
1041 | /* Get local addresses */ |
1042 | static void init_local(void) | |
1043 | { | |
1044 | struct sockaddr_storage sas, *addr; | |
1045 | int i; | |
1046 | ||
30d3a237 | 1047 | dlm_local_count = 0; |
6ed7257b PC |
1048 | for (i = 0; i < DLM_MAX_ADDR_COUNT - 1; i++) { |
1049 | if (dlm_our_addr(&sas, i)) | |
1050 | break; | |
1051 | ||
1052 | addr = kmalloc(sizeof(*addr), GFP_KERNEL); | |
1053 | if (!addr) | |
1054 | break; | |
1055 | memcpy(addr, &sas, sizeof(*addr)); | |
1056 | dlm_local_addr[dlm_local_count++] = addr; | |
1057 | } | |
1058 | } | |
1059 | ||
617e82e1 DT |
1060 | /* Bind to an IP address. SCTP allows multiple address so it can do |
1061 | multi-homing */ | |
1062 | static int add_sctp_bind_addr(struct connection *sctp_con, | |
1063 | struct sockaddr_storage *addr, | |
1064 | int addr_len, int num) | |
6ed7257b PC |
1065 | { |
1066 | int result = 0; | |
1067 | ||
1068 | if (num == 1) | |
1069 | result = kernel_bind(sctp_con->sock, | |
1070 | (struct sockaddr *) addr, | |
1071 | addr_len); | |
1072 | else | |
1073 | result = kernel_setsockopt(sctp_con->sock, SOL_SCTP, | |
1074 | SCTP_SOCKOPT_BINDX_ADD, | |
1075 | (char *)addr, addr_len); | |
1076 | ||
1077 | if (result < 0) | |
1078 | log_print("Can't bind to port %d addr number %d", | |
1079 | dlm_config.ci_tcp_port, num); | |
1080 | ||
1081 | return result; | |
1082 | } | |
fdda387f | 1083 | |
6ed7257b PC |
1084 | /* Initialise SCTP socket and bind to all interfaces */ |
1085 | static int sctp_listen_for_all(void) | |
1086 | { | |
1087 | struct socket *sock = NULL; | |
1088 | struct sockaddr_storage localaddr; | |
1089 | struct sctp_event_subscribe subscribe; | |
1090 | int result = -EINVAL, num = 1, i, addr_len; | |
1091 | struct connection *con = nodeid2con(0, GFP_KERNEL); | |
1092 | int bufsize = NEEDED_RMEM; | |
1093 | ||
1094 | if (!con) | |
1095 | return -ENOMEM; | |
1096 | ||
1097 | log_print("Using SCTP for communications"); | |
1098 | ||
1099 | result = sock_create_kern(dlm_local_addr[0]->ss_family, SOCK_SEQPACKET, | |
1100 | IPPROTO_SCTP, &sock); | |
1101 | if (result < 0) { | |
1102 | log_print("Can't create comms socket, check SCTP is loaded"); | |
1103 | goto out; | |
1104 | } | |
1105 | ||
1106 | /* Listen for events */ | |
1107 | memset(&subscribe, 0, sizeof(subscribe)); | |
1108 | subscribe.sctp_data_io_event = 1; | |
1109 | subscribe.sctp_association_event = 1; | |
1110 | subscribe.sctp_send_failure_event = 1; | |
1111 | subscribe.sctp_shutdown_event = 1; | |
1112 | subscribe.sctp_partial_delivery_event = 1; | |
1113 | ||
df61c952 | 1114 | result = kernel_setsockopt(sock, SOL_SOCKET, SO_RCVBUFFORCE, |
6ed7257b PC |
1115 | (char *)&bufsize, sizeof(bufsize)); |
1116 | if (result) | |
617e82e1 | 1117 | log_print("Error increasing buffer space on socket %d", result); |
6ed7257b PC |
1118 | |
1119 | result = kernel_setsockopt(sock, SOL_SCTP, SCTP_EVENTS, | |
617e82e1 | 1120 | (char *)&subscribe, sizeof(subscribe)); |
6ed7257b PC |
1121 | if (result < 0) { |
1122 | log_print("Failed to set SCTP_EVENTS on socket: result=%d", | |
1123 | result); | |
1124 | goto create_delsock; | |
1125 | } | |
1126 | ||
1127 | /* Init con struct */ | |
1128 | sock->sk->sk_user_data = con; | |
1129 | con->sock = sock; | |
1130 | con->sock->sk->sk_data_ready = lowcomms_data_ready; | |
1131 | con->rx_action = receive_from_sock; | |
1132 | con->connect_action = sctp_init_assoc; | |
1133 | ||
1134 | /* Bind to all interfaces. */ | |
1135 | for (i = 0; i < dlm_local_count; i++) { | |
1136 | memcpy(&localaddr, dlm_local_addr[i], sizeof(localaddr)); | |
1137 | make_sockaddr(&localaddr, dlm_config.ci_tcp_port, &addr_len); | |
1138 | ||
1139 | result = add_sctp_bind_addr(con, &localaddr, addr_len, num); | |
1140 | if (result) | |
1141 | goto create_delsock; | |
1142 | ++num; | |
1143 | } | |
1144 | ||
1145 | result = sock->ops->listen(sock, 5); | |
1146 | if (result < 0) { | |
1147 | log_print("Can't set socket listening"); | |
1148 | goto create_delsock; | |
1149 | } | |
1150 | ||
1151 | return 0; | |
1152 | ||
1153 | create_delsock: | |
1154 | sock_release(sock); | |
1155 | con->sock = NULL; | |
1156 | out: | |
1157 | return result; | |
1158 | } | |
1159 | ||
1160 | static int tcp_listen_for_all(void) | |
fdda387f PC |
1161 | { |
1162 | struct socket *sock = NULL; | |
1163 | struct connection *con = nodeid2con(0, GFP_KERNEL); | |
1164 | int result = -EINVAL; | |
1165 | ||
6ed7257b PC |
1166 | if (!con) |
1167 | return -ENOMEM; | |
1168 | ||
fdda387f | 1169 | /* We don't support multi-homed hosts */ |
6ed7257b | 1170 | if (dlm_local_addr[1] != NULL) { |
617e82e1 DT |
1171 | log_print("TCP protocol can't handle multi-homed hosts, " |
1172 | "try SCTP"); | |
6ed7257b PC |
1173 | return -EINVAL; |
1174 | } | |
1175 | ||
1176 | log_print("Using TCP for communications"); | |
1177 | ||
6ed7257b | 1178 | sock = tcp_create_listen_sock(con, dlm_local_addr[0]); |
fdda387f PC |
1179 | if (sock) { |
1180 | add_sock(sock, con); | |
1181 | result = 0; | |
1182 | } | |
1183 | else { | |
1184 | result = -EADDRINUSE; | |
1185 | } | |
1186 | ||
1187 | return result; | |
1188 | } | |
1189 | ||
1190 | ||
1191 | ||
1192 | static struct writequeue_entry *new_writequeue_entry(struct connection *con, | |
1193 | gfp_t allocation) | |
1194 | { | |
1195 | struct writequeue_entry *entry; | |
1196 | ||
1197 | entry = kmalloc(sizeof(struct writequeue_entry), allocation); | |
1198 | if (!entry) | |
1199 | return NULL; | |
1200 | ||
1201 | entry->page = alloc_page(allocation); | |
1202 | if (!entry->page) { | |
1203 | kfree(entry); | |
1204 | return NULL; | |
1205 | } | |
1206 | ||
1207 | entry->offset = 0; | |
1208 | entry->len = 0; | |
1209 | entry->end = 0; | |
1210 | entry->users = 0; | |
1211 | entry->con = con; | |
1212 | ||
1213 | return entry; | |
1214 | } | |
1215 | ||
617e82e1 | 1216 | void *dlm_lowcomms_get_buffer(int nodeid, int len, gfp_t allocation, char **ppc) |
fdda387f PC |
1217 | { |
1218 | struct connection *con; | |
1219 | struct writequeue_entry *e; | |
1220 | int offset = 0; | |
1221 | int users = 0; | |
1222 | ||
fdda387f PC |
1223 | con = nodeid2con(nodeid, allocation); |
1224 | if (!con) | |
1225 | return NULL; | |
1226 | ||
4edde74e | 1227 | spin_lock(&con->writequeue_lock); |
fdda387f | 1228 | e = list_entry(con->writequeue.prev, struct writequeue_entry, list); |
ac33d071 | 1229 | if ((&e->list == &con->writequeue) || |
fdda387f PC |
1230 | (PAGE_CACHE_SIZE - e->end < len)) { |
1231 | e = NULL; | |
1232 | } else { | |
1233 | offset = e->end; | |
1234 | e->end += len; | |
1235 | users = e->users++; | |
1236 | } | |
1237 | spin_unlock(&con->writequeue_lock); | |
1238 | ||
1239 | if (e) { | |
ac33d071 | 1240 | got_one: |
fdda387f PC |
1241 | *ppc = page_address(e->page) + offset; |
1242 | return e; | |
1243 | } | |
1244 | ||
1245 | e = new_writequeue_entry(con, allocation); | |
1246 | if (e) { | |
1247 | spin_lock(&con->writequeue_lock); | |
1248 | offset = e->end; | |
1249 | e->end += len; | |
1250 | users = e->users++; | |
1251 | list_add_tail(&e->list, &con->writequeue); | |
1252 | spin_unlock(&con->writequeue_lock); | |
1253 | goto got_one; | |
1254 | } | |
1255 | return NULL; | |
1256 | } | |
1257 | ||
1258 | void dlm_lowcomms_commit_buffer(void *mh) | |
1259 | { | |
1260 | struct writequeue_entry *e = (struct writequeue_entry *)mh; | |
1261 | struct connection *con = e->con; | |
1262 | int users; | |
1263 | ||
4edde74e | 1264 | spin_lock(&con->writequeue_lock); |
fdda387f PC |
1265 | users = --e->users; |
1266 | if (users) | |
1267 | goto out; | |
1268 | e->len = e->end - e->offset; | |
fdda387f PC |
1269 | spin_unlock(&con->writequeue_lock); |
1270 | ||
1d6e8131 PC |
1271 | if (!test_and_set_bit(CF_WRITE_PENDING, &con->flags)) { |
1272 | queue_work(send_workqueue, &con->swork); | |
fdda387f PC |
1273 | } |
1274 | return; | |
1275 | ||
ac33d071 | 1276 | out: |
fdda387f PC |
1277 | spin_unlock(&con->writequeue_lock); |
1278 | return; | |
1279 | } | |
1280 | ||
fdda387f | 1281 | /* Send a message */ |
ac33d071 | 1282 | static void send_to_sock(struct connection *con) |
fdda387f PC |
1283 | { |
1284 | int ret = 0; | |
1285 | ssize_t(*sendpage) (struct socket *, struct page *, int, size_t, int); | |
1286 | const int msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL; | |
1287 | struct writequeue_entry *e; | |
1288 | int len, offset; | |
1289 | ||
f1f1c1cc | 1290 | mutex_lock(&con->sock_mutex); |
fdda387f PC |
1291 | if (con->sock == NULL) |
1292 | goto out_connect; | |
1293 | ||
1294 | sendpage = con->sock->ops->sendpage; | |
1295 | ||
1296 | spin_lock(&con->writequeue_lock); | |
1297 | for (;;) { | |
1298 | e = list_entry(con->writequeue.next, struct writequeue_entry, | |
1299 | list); | |
1300 | if ((struct list_head *) e == &con->writequeue) | |
1301 | break; | |
1302 | ||
1303 | len = e->len; | |
1304 | offset = e->offset; | |
1305 | BUG_ON(len == 0 && e->users == 0); | |
1306 | spin_unlock(&con->writequeue_lock); | |
1307 | ||
1308 | ret = 0; | |
1309 | if (len) { | |
1310 | ret = sendpage(con->sock, e->page, offset, len, | |
1311 | msg_flags); | |
d66f8277 PC |
1312 | if (ret == -EAGAIN || ret == 0) { |
1313 | cond_resched(); | |
fdda387f | 1314 | goto out; |
d66f8277 | 1315 | } |
fdda387f PC |
1316 | if (ret <= 0) |
1317 | goto send_error; | |
d66f8277 | 1318 | } |
fdda387f | 1319 | /* Don't starve people filling buffers */ |
ac33d071 | 1320 | cond_resched(); |
fdda387f PC |
1321 | |
1322 | spin_lock(&con->writequeue_lock); | |
1323 | e->offset += ret; | |
1324 | e->len -= ret; | |
1325 | ||
1326 | if (e->len == 0 && e->users == 0) { | |
1327 | list_del(&e->list); | |
1328 | free_entry(e); | |
1329 | continue; | |
1330 | } | |
1331 | } | |
1332 | spin_unlock(&con->writequeue_lock); | |
ac33d071 | 1333 | out: |
f1f1c1cc | 1334 | mutex_unlock(&con->sock_mutex); |
ac33d071 | 1335 | return; |
fdda387f | 1336 | |
ac33d071 | 1337 | send_error: |
f1f1c1cc | 1338 | mutex_unlock(&con->sock_mutex); |
ac33d071 | 1339 | close_connection(con, false); |
fdda387f | 1340 | lowcomms_connect_sock(con); |
ac33d071 | 1341 | return; |
fdda387f | 1342 | |
ac33d071 | 1343 | out_connect: |
f1f1c1cc | 1344 | mutex_unlock(&con->sock_mutex); |
6ed7257b PC |
1345 | if (!test_bit(CF_INIT_PENDING, &con->flags)) |
1346 | lowcomms_connect_sock(con); | |
ac33d071 | 1347 | return; |
fdda387f PC |
1348 | } |
1349 | ||
1350 | static void clean_one_writequeue(struct connection *con) | |
1351 | { | |
5e9ccc37 | 1352 | struct writequeue_entry *e, *safe; |
fdda387f PC |
1353 | |
1354 | spin_lock(&con->writequeue_lock); | |
5e9ccc37 | 1355 | list_for_each_entry_safe(e, safe, &con->writequeue, list) { |
fdda387f PC |
1356 | list_del(&e->list); |
1357 | free_entry(e); | |
1358 | } | |
1359 | spin_unlock(&con->writequeue_lock); | |
1360 | } | |
1361 | ||
1362 | /* Called from recovery when it knows that a node has | |
1363 | left the cluster */ | |
1364 | int dlm_lowcomms_close(int nodeid) | |
1365 | { | |
1366 | struct connection *con; | |
1367 | ||
fdda387f PC |
1368 | log_print("closing connection to node %d", nodeid); |
1369 | con = nodeid2con(nodeid, 0); | |
1370 | if (con) { | |
1371 | clean_one_writequeue(con); | |
ac33d071 | 1372 | close_connection(con, true); |
fdda387f PC |
1373 | } |
1374 | return 0; | |
fdda387f PC |
1375 | } |
1376 | ||
6ed7257b | 1377 | /* Receive workqueue function */ |
1d6e8131 | 1378 | static void process_recv_sockets(struct work_struct *work) |
fdda387f | 1379 | { |
1d6e8131 PC |
1380 | struct connection *con = container_of(work, struct connection, rwork); |
1381 | int err; | |
fdda387f | 1382 | |
1d6e8131 PC |
1383 | clear_bit(CF_READ_PENDING, &con->flags); |
1384 | do { | |
1385 | err = con->rx_action(con); | |
1386 | } while (!err); | |
fdda387f PC |
1387 | } |
1388 | ||
6ed7257b | 1389 | /* Send workqueue function */ |
1d6e8131 | 1390 | static void process_send_sockets(struct work_struct *work) |
fdda387f | 1391 | { |
1d6e8131 | 1392 | struct connection *con = container_of(work, struct connection, swork); |
fdda387f | 1393 | |
1d6e8131 | 1394 | if (test_and_clear_bit(CF_CONNECT_PENDING, &con->flags)) { |
6ed7257b | 1395 | con->connect_action(con); |
fdda387f | 1396 | } |
bd44e2b0 PC |
1397 | clear_bit(CF_WRITE_PENDING, &con->flags); |
1398 | send_to_sock(con); | |
fdda387f PC |
1399 | } |
1400 | ||
1401 | ||
1402 | /* Discard all entries on the write queues */ | |
1403 | static void clean_writequeues(void) | |
1404 | { | |
5e9ccc37 | 1405 | foreach_conn(clean_one_writequeue); |
fdda387f PC |
1406 | } |
1407 | ||
1d6e8131 | 1408 | static void work_stop(void) |
fdda387f | 1409 | { |
1d6e8131 PC |
1410 | destroy_workqueue(recv_workqueue); |
1411 | destroy_workqueue(send_workqueue); | |
fdda387f PC |
1412 | } |
1413 | ||
1d6e8131 | 1414 | static int work_start(void) |
fdda387f | 1415 | { |
fdda387f | 1416 | int error; |
1d6e8131 PC |
1417 | recv_workqueue = create_workqueue("dlm_recv"); |
1418 | error = IS_ERR(recv_workqueue); | |
ac33d071 | 1419 | if (error) { |
1d6e8131 | 1420 | log_print("can't start dlm_recv %d", error); |
fdda387f PC |
1421 | return error; |
1422 | } | |
fdda387f | 1423 | |
1d6e8131 PC |
1424 | send_workqueue = create_singlethread_workqueue("dlm_send"); |
1425 | error = IS_ERR(send_workqueue); | |
ac33d071 | 1426 | if (error) { |
1d6e8131 PC |
1427 | log_print("can't start dlm_send %d", error); |
1428 | destroy_workqueue(recv_workqueue); | |
fdda387f PC |
1429 | return error; |
1430 | } | |
fdda387f PC |
1431 | |
1432 | return 0; | |
1433 | } | |
1434 | ||
5e9ccc37 | 1435 | static void stop_conn(struct connection *con) |
fdda387f | 1436 | { |
5e9ccc37 | 1437 | con->flags |= 0x0F; |
391fbdc5 | 1438 | if (con->sock && con->sock->sk) |
5e9ccc37 CC |
1439 | con->sock->sk->sk_user_data = NULL; |
1440 | } | |
fdda387f | 1441 | |
5e9ccc37 CC |
1442 | static void free_conn(struct connection *con) |
1443 | { | |
1444 | close_connection(con, true); | |
1445 | if (con->othercon) | |
1446 | kmem_cache_free(con_cache, con->othercon); | |
1447 | hlist_del(&con->list); | |
1448 | kmem_cache_free(con_cache, con); | |
1449 | } | |
1450 | ||
1451 | void dlm_lowcomms_stop(void) | |
1452 | { | |
ac33d071 | 1453 | /* Set all the flags to prevent any |
fdda387f PC |
1454 | socket activity. |
1455 | */ | |
7a936ce7 | 1456 | mutex_lock(&connections_lock); |
5e9ccc37 | 1457 | foreach_conn(stop_conn); |
7a936ce7 | 1458 | mutex_unlock(&connections_lock); |
ac33d071 | 1459 | |
1d6e8131 | 1460 | work_stop(); |
6ed7257b | 1461 | |
7a936ce7 | 1462 | mutex_lock(&connections_lock); |
fdda387f PC |
1463 | clean_writequeues(); |
1464 | ||
5e9ccc37 CC |
1465 | foreach_conn(free_conn); |
1466 | ||
7a936ce7 | 1467 | mutex_unlock(&connections_lock); |
fdda387f PC |
1468 | kmem_cache_destroy(con_cache); |
1469 | } | |
1470 | ||
fdda387f PC |
1471 | int dlm_lowcomms_start(void) |
1472 | { | |
6ed7257b PC |
1473 | int error = -EINVAL; |
1474 | struct connection *con; | |
5e9ccc37 CC |
1475 | int i; |
1476 | ||
1477 | for (i = 0; i < CONN_HASH_SIZE; i++) | |
1478 | INIT_HLIST_HEAD(&connection_hash[i]); | |
fdda387f | 1479 | |
6ed7257b PC |
1480 | init_local(); |
1481 | if (!dlm_local_count) { | |
617e82e1 | 1482 | error = -ENOTCONN; |
fdda387f | 1483 | log_print("no local IP address has been set"); |
6ed7257b | 1484 | goto out; |
fdda387f PC |
1485 | } |
1486 | ||
6ed7257b | 1487 | error = -ENOMEM; |
fdda387f | 1488 | con_cache = kmem_cache_create("dlm_conn", sizeof(struct connection), |
ac33d071 | 1489 | __alignof__(struct connection), 0, |
20c2df83 | 1490 | NULL); |
fdda387f | 1491 | if (!con_cache) |
6ed7257b | 1492 | goto out; |
fdda387f | 1493 | |
fdda387f | 1494 | /* Start listening */ |
6ed7257b PC |
1495 | if (dlm_config.ci_protocol == 0) |
1496 | error = tcp_listen_for_all(); | |
1497 | else | |
1498 | error = sctp_listen_for_all(); | |
fdda387f PC |
1499 | if (error) |
1500 | goto fail_unlisten; | |
1501 | ||
1d6e8131 | 1502 | error = work_start(); |
fdda387f PC |
1503 | if (error) |
1504 | goto fail_unlisten; | |
1505 | ||
fdda387f PC |
1506 | return 0; |
1507 | ||
ac33d071 | 1508 | fail_unlisten: |
6ed7257b PC |
1509 | con = nodeid2con(0,0); |
1510 | if (con) { | |
1511 | close_connection(con, false); | |
1512 | kmem_cache_free(con_cache, con); | |
1513 | } | |
fdda387f PC |
1514 | kmem_cache_destroy(con_cache); |
1515 | ||
ac33d071 | 1516 | out: |
fdda387f PC |
1517 | return error; |
1518 | } |