Commit | Line | Data |
---|---|---|
fdda387f PC |
1 | /****************************************************************************** |
2 | ******************************************************************************* | |
3 | ** | |
4 | ** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
a34fbc63 | 5 | ** Copyright (C) 2004-2007 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 | |
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 | * | |
6ed7257b PC |
39 | * lowcomms will choose to use wither TCP or SCTP as its transport layer |
40 | * depending on the configuration variable 'protocol'. This should be set | |
41 | * to 0 (default) for TCP or 1 for SCTP. It shouldbe configured using a | |
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 PC |
51 | #include <linux/idr.h> |
52 | #include <linux/file.h> | |
53 | #include <linux/sctp.h> | |
54 | #include <net/sctp/user.h> | |
fdda387f PC |
55 | |
56 | #include "dlm_internal.h" | |
57 | #include "lowcomms.h" | |
58 | #include "midcomms.h" | |
59 | #include "config.h" | |
60 | ||
6ed7257b PC |
61 | #define NEEDED_RMEM (4*1024*1024) |
62 | ||
fdda387f | 63 | struct cbuf { |
ac33d071 PC |
64 | unsigned int base; |
65 | unsigned int len; | |
66 | unsigned int mask; | |
fdda387f PC |
67 | }; |
68 | ||
ac33d071 PC |
69 | static void cbuf_add(struct cbuf *cb, int n) |
70 | { | |
71 | cb->len += n; | |
72 | } | |
fdda387f | 73 | |
ac33d071 PC |
74 | static int cbuf_data(struct cbuf *cb) |
75 | { | |
76 | return ((cb->base + cb->len) & cb->mask); | |
77 | } | |
78 | ||
79 | static void cbuf_init(struct cbuf *cb, int size) | |
80 | { | |
81 | cb->base = cb->len = 0; | |
82 | cb->mask = size-1; | |
83 | } | |
84 | ||
85 | static void cbuf_eat(struct cbuf *cb, int n) | |
86 | { | |
87 | cb->len -= n; | |
88 | cb->base += n; | |
89 | cb->base &= cb->mask; | |
90 | } | |
91 | ||
92 | static bool cbuf_empty(struct cbuf *cb) | |
93 | { | |
94 | return cb->len == 0; | |
95 | } | |
fdda387f | 96 | |
fdda387f PC |
97 | struct connection { |
98 | struct socket *sock; /* NULL if not connected */ | |
99 | uint32_t nodeid; /* So we know who we are in the list */ | |
f1f1c1cc | 100 | struct mutex sock_mutex; |
6ed7257b | 101 | unsigned long flags; |
fdda387f PC |
102 | #define CF_READ_PENDING 1 |
103 | #define CF_WRITE_PENDING 2 | |
104 | #define CF_CONNECT_PENDING 3 | |
6ed7257b PC |
105 | #define CF_INIT_PENDING 4 |
106 | #define CF_IS_OTHERCON 5 | |
ac33d071 | 107 | struct list_head writequeue; /* List of outgoing writequeue_entries */ |
fdda387f PC |
108 | spinlock_t writequeue_lock; |
109 | int (*rx_action) (struct connection *); /* What to do when active */ | |
6ed7257b | 110 | void (*connect_action) (struct connection *); /* What to do to connect */ |
fdda387f PC |
111 | struct page *rx_page; |
112 | struct cbuf cb; | |
113 | int retries; | |
fdda387f | 114 | #define MAX_CONNECT_RETRIES 3 |
6ed7257b | 115 | int sctp_assoc; |
fdda387f | 116 | struct connection *othercon; |
1d6e8131 PC |
117 | struct work_struct rwork; /* Receive workqueue */ |
118 | struct work_struct swork; /* Send workqueue */ | |
fdda387f PC |
119 | }; |
120 | #define sock2con(x) ((struct connection *)(x)->sk_user_data) | |
121 | ||
122 | /* An entry waiting to be sent */ | |
123 | struct writequeue_entry { | |
124 | struct list_head list; | |
125 | struct page *page; | |
126 | int offset; | |
127 | int len; | |
128 | int end; | |
129 | int users; | |
130 | struct connection *con; | |
131 | }; | |
132 | ||
6ed7257b PC |
133 | static struct sockaddr_storage *dlm_local_addr[DLM_MAX_ADDR_COUNT]; |
134 | static int dlm_local_count; | |
fdda387f | 135 | |
1d6e8131 PC |
136 | /* Work queues */ |
137 | static struct workqueue_struct *recv_workqueue; | |
138 | static struct workqueue_struct *send_workqueue; | |
fdda387f | 139 | |
6ed7257b | 140 | static DEFINE_IDR(connections_idr); |
ac33d071 | 141 | static DECLARE_MUTEX(connections_lock); |
6ed7257b | 142 | static int max_nodeid; |
c80e7c83 | 143 | static struct kmem_cache *con_cache; |
fdda387f | 144 | |
1d6e8131 PC |
145 | static void process_recv_sockets(struct work_struct *work); |
146 | static void process_send_sockets(struct work_struct *work); | |
fdda387f | 147 | |
6ed7257b PC |
148 | /* |
149 | * If 'allocation' is zero then we don't attempt to create a new | |
150 | * connection structure for this node. | |
151 | */ | |
152 | static struct connection *__nodeid2con(int nodeid, gfp_t alloc) | |
fdda387f PC |
153 | { |
154 | struct connection *con = NULL; | |
6ed7257b PC |
155 | int r; |
156 | int n; | |
fdda387f | 157 | |
6ed7257b PC |
158 | con = idr_find(&connections_idr, nodeid); |
159 | if (con || !alloc) | |
160 | return con; | |
fdda387f | 161 | |
6ed7257b PC |
162 | r = idr_pre_get(&connections_idr, alloc); |
163 | if (!r) | |
164 | return NULL; | |
165 | ||
166 | con = kmem_cache_zalloc(con_cache, alloc); | |
167 | if (!con) | |
168 | return NULL; | |
fdda387f | 169 | |
6ed7257b PC |
170 | r = idr_get_new_above(&connections_idr, con, nodeid, &n); |
171 | if (r) { | |
172 | kmem_cache_free(con_cache, con); | |
173 | return NULL; | |
174 | } | |
fdda387f | 175 | |
6ed7257b PC |
176 | if (n != nodeid) { |
177 | idr_remove(&connections_idr, n); | |
178 | kmem_cache_free(con_cache, con); | |
179 | return NULL; | |
fdda387f PC |
180 | } |
181 | ||
6ed7257b PC |
182 | con->nodeid = nodeid; |
183 | mutex_init(&con->sock_mutex); | |
184 | INIT_LIST_HEAD(&con->writequeue); | |
185 | spin_lock_init(&con->writequeue_lock); | |
186 | INIT_WORK(&con->swork, process_send_sockets); | |
187 | INIT_WORK(&con->rwork, process_recv_sockets); | |
fdda387f | 188 | |
6ed7257b PC |
189 | /* Setup action pointers for child sockets */ |
190 | if (con->nodeid) { | |
191 | struct connection *zerocon = idr_find(&connections_idr, 0); | |
fdda387f | 192 | |
6ed7257b PC |
193 | con->connect_action = zerocon->connect_action; |
194 | if (!con->rx_action) | |
195 | con->rx_action = zerocon->rx_action; | |
fdda387f PC |
196 | } |
197 | ||
6ed7257b PC |
198 | if (nodeid > max_nodeid) |
199 | max_nodeid = nodeid; | |
200 | ||
201 | return con; | |
202 | } | |
203 | ||
204 | static struct connection *nodeid2con(int nodeid, gfp_t allocation) | |
205 | { | |
206 | struct connection *con; | |
207 | ||
208 | down(&connections_lock); | |
209 | con = __nodeid2con(nodeid, allocation); | |
fdda387f | 210 | up(&connections_lock); |
6ed7257b | 211 | |
fdda387f PC |
212 | return con; |
213 | } | |
214 | ||
6ed7257b PC |
215 | /* This is a bit drastic, but only called when things go wrong */ |
216 | static struct connection *assoc2con(int assoc_id) | |
217 | { | |
218 | int i; | |
219 | struct connection *con; | |
220 | ||
221 | down(&connections_lock); | |
30d3a237 | 222 | for (i=0; i<=max_nodeid; i++) { |
6ed7257b PC |
223 | con = __nodeid2con(i, 0); |
224 | if (con && con->sctp_assoc == assoc_id) { | |
225 | up(&connections_lock); | |
226 | return con; | |
227 | } | |
228 | } | |
229 | up(&connections_lock); | |
230 | return NULL; | |
231 | } | |
232 | ||
233 | static int nodeid_to_addr(int nodeid, struct sockaddr *retaddr) | |
234 | { | |
235 | struct sockaddr_storage addr; | |
236 | int error; | |
237 | ||
238 | if (!dlm_local_count) | |
239 | return -1; | |
240 | ||
241 | error = dlm_nodeid_to_addr(nodeid, &addr); | |
242 | if (error) | |
243 | return error; | |
244 | ||
245 | if (dlm_local_addr[0]->ss_family == AF_INET) { | |
246 | struct sockaddr_in *in4 = (struct sockaddr_in *) &addr; | |
247 | struct sockaddr_in *ret4 = (struct sockaddr_in *) retaddr; | |
248 | ret4->sin_addr.s_addr = in4->sin_addr.s_addr; | |
249 | } else { | |
250 | struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) &addr; | |
251 | struct sockaddr_in6 *ret6 = (struct sockaddr_in6 *) retaddr; | |
252 | memcpy(&ret6->sin6_addr, &in6->sin6_addr, | |
253 | sizeof(in6->sin6_addr)); | |
254 | } | |
255 | ||
256 | return 0; | |
257 | } | |
258 | ||
fdda387f PC |
259 | /* Data available on socket or listen socket received a connect */ |
260 | static void lowcomms_data_ready(struct sock *sk, int count_unused) | |
261 | { | |
262 | struct connection *con = sock2con(sk); | |
afb853fb | 263 | if (con && !test_and_set_bit(CF_READ_PENDING, &con->flags)) |
1d6e8131 | 264 | queue_work(recv_workqueue, &con->rwork); |
fdda387f PC |
265 | } |
266 | ||
267 | static void lowcomms_write_space(struct sock *sk) | |
268 | { | |
269 | struct connection *con = sock2con(sk); | |
270 | ||
afb853fb | 271 | if (con && !test_and_set_bit(CF_WRITE_PENDING, &con->flags)) |
1d6e8131 | 272 | queue_work(send_workqueue, &con->swork); |
fdda387f PC |
273 | } |
274 | ||
275 | static inline void lowcomms_connect_sock(struct connection *con) | |
276 | { | |
1d6e8131 PC |
277 | if (!test_and_set_bit(CF_CONNECT_PENDING, &con->flags)) |
278 | queue_work(send_workqueue, &con->swork); | |
fdda387f PC |
279 | } |
280 | ||
281 | static void lowcomms_state_change(struct sock *sk) | |
282 | { | |
ac33d071 | 283 | if (sk->sk_state == TCP_ESTABLISHED) |
fdda387f | 284 | lowcomms_write_space(sk); |
fdda387f PC |
285 | } |
286 | ||
287 | /* Make a socket active */ | |
288 | static int add_sock(struct socket *sock, struct connection *con) | |
289 | { | |
290 | con->sock = sock; | |
291 | ||
292 | /* Install a data_ready callback */ | |
293 | con->sock->sk->sk_data_ready = lowcomms_data_ready; | |
294 | con->sock->sk->sk_write_space = lowcomms_write_space; | |
295 | con->sock->sk->sk_state_change = lowcomms_state_change; | |
6ed7257b | 296 | con->sock->sk->sk_user_data = con; |
fdda387f PC |
297 | return 0; |
298 | } | |
299 | ||
6ed7257b | 300 | /* Add the port number to an IPv6 or 4 sockaddr and return the address |
fdda387f PC |
301 | length */ |
302 | static void make_sockaddr(struct sockaddr_storage *saddr, uint16_t port, | |
303 | int *addr_len) | |
304 | { | |
6ed7257b | 305 | saddr->ss_family = dlm_local_addr[0]->ss_family; |
ac33d071 | 306 | if (saddr->ss_family == AF_INET) { |
fdda387f PC |
307 | struct sockaddr_in *in4_addr = (struct sockaddr_in *)saddr; |
308 | in4_addr->sin_port = cpu_to_be16(port); | |
309 | *addr_len = sizeof(struct sockaddr_in); | |
6ed7257b | 310 | memset(&in4_addr->sin_zero, 0, sizeof(in4_addr->sin_zero)); |
ac33d071 | 311 | } else { |
fdda387f PC |
312 | struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)saddr; |
313 | in6_addr->sin6_port = cpu_to_be16(port); | |
314 | *addr_len = sizeof(struct sockaddr_in6); | |
315 | } | |
01c8cab2 | 316 | memset((char *)saddr + *addr_len, 0, sizeof(struct sockaddr_storage) - *addr_len); |
fdda387f PC |
317 | } |
318 | ||
319 | /* Close a remote connection and tidy up */ | |
ac33d071 | 320 | static void close_connection(struct connection *con, bool and_other) |
fdda387f | 321 | { |
f1f1c1cc | 322 | mutex_lock(&con->sock_mutex); |
fdda387f PC |
323 | |
324 | if (con->sock) { | |
325 | sock_release(con->sock); | |
326 | con->sock = NULL; | |
327 | } | |
328 | if (con->othercon && and_other) { | |
ac33d071 PC |
329 | /* Will only re-enter once. */ |
330 | close_connection(con->othercon, false); | |
fdda387f PC |
331 | } |
332 | if (con->rx_page) { | |
333 | __free_page(con->rx_page); | |
334 | con->rx_page = NULL; | |
335 | } | |
9e5f2825 | 336 | |
61d96be0 PC |
337 | con->retries = 0; |
338 | mutex_unlock(&con->sock_mutex); | |
fdda387f PC |
339 | } |
340 | ||
6ed7257b PC |
341 | /* We only send shutdown messages to nodes that are not part of the cluster */ |
342 | static void sctp_send_shutdown(sctp_assoc_t associd) | |
343 | { | |
344 | static char outcmsg[CMSG_SPACE(sizeof(struct sctp_sndrcvinfo))]; | |
345 | struct msghdr outmessage; | |
346 | struct cmsghdr *cmsg; | |
347 | struct sctp_sndrcvinfo *sinfo; | |
348 | int ret; | |
349 | struct connection *con; | |
350 | ||
351 | con = nodeid2con(0,0); | |
352 | BUG_ON(con == NULL); | |
353 | ||
354 | outmessage.msg_name = NULL; | |
355 | outmessage.msg_namelen = 0; | |
356 | outmessage.msg_control = outcmsg; | |
357 | outmessage.msg_controllen = sizeof(outcmsg); | |
358 | outmessage.msg_flags = MSG_EOR; | |
359 | ||
360 | cmsg = CMSG_FIRSTHDR(&outmessage); | |
361 | cmsg->cmsg_level = IPPROTO_SCTP; | |
362 | cmsg->cmsg_type = SCTP_SNDRCV; | |
363 | cmsg->cmsg_len = CMSG_LEN(sizeof(struct sctp_sndrcvinfo)); | |
364 | outmessage.msg_controllen = cmsg->cmsg_len; | |
365 | sinfo = CMSG_DATA(cmsg); | |
366 | memset(sinfo, 0x00, sizeof(struct sctp_sndrcvinfo)); | |
367 | ||
368 | sinfo->sinfo_flags |= MSG_EOF; | |
369 | sinfo->sinfo_assoc_id = associd; | |
370 | ||
371 | ret = kernel_sendmsg(con->sock, &outmessage, NULL, 0, 0); | |
372 | ||
373 | if (ret != 0) | |
374 | log_print("send EOF to node failed: %d", ret); | |
375 | } | |
376 | ||
377 | /* INIT failed but we don't know which node... | |
378 | restart INIT on all pending nodes */ | |
379 | static void sctp_init_failed(void) | |
380 | { | |
381 | int i; | |
382 | struct connection *con; | |
383 | ||
384 | down(&connections_lock); | |
385 | for (i=1; i<=max_nodeid; i++) { | |
386 | con = __nodeid2con(i, 0); | |
387 | if (!con) | |
388 | continue; | |
389 | con->sctp_assoc = 0; | |
390 | if (test_and_clear_bit(CF_CONNECT_PENDING, &con->flags)) { | |
391 | if (!test_and_set_bit(CF_WRITE_PENDING, &con->flags)) { | |
392 | queue_work(send_workqueue, &con->swork); | |
393 | } | |
394 | } | |
395 | } | |
396 | up(&connections_lock); | |
397 | } | |
398 | ||
399 | /* Something happened to an association */ | |
617e82e1 DT |
400 | static void process_sctp_notification(struct connection *con, |
401 | struct msghdr *msg, char *buf) | |
6ed7257b PC |
402 | { |
403 | union sctp_notification *sn = (union sctp_notification *)buf; | |
404 | ||
405 | if (sn->sn_header.sn_type == SCTP_ASSOC_CHANGE) { | |
406 | switch (sn->sn_assoc_change.sac_state) { | |
407 | ||
408 | case SCTP_COMM_UP: | |
409 | case SCTP_RESTART: | |
410 | { | |
411 | /* Check that the new node is in the lockspace */ | |
412 | struct sctp_prim prim; | |
413 | int nodeid; | |
414 | int prim_len, ret; | |
415 | int addr_len; | |
416 | struct connection *new_con; | |
417 | struct file *file; | |
418 | sctp_peeloff_arg_t parg; | |
419 | int parglen = sizeof(parg); | |
420 | ||
421 | /* | |
422 | * We get this before any data for an association. | |
423 | * We verify that the node is in the cluster and | |
424 | * then peel off a socket for it. | |
425 | */ | |
426 | if ((int)sn->sn_assoc_change.sac_assoc_id <= 0) { | |
427 | log_print("COMM_UP for invalid assoc ID %d", | |
617e82e1 | 428 | (int)sn->sn_assoc_change.sac_assoc_id); |
6ed7257b PC |
429 | sctp_init_failed(); |
430 | return; | |
431 | } | |
432 | memset(&prim, 0, sizeof(struct sctp_prim)); | |
433 | prim_len = sizeof(struct sctp_prim); | |
434 | prim.ssp_assoc_id = sn->sn_assoc_change.sac_assoc_id; | |
435 | ||
436 | ret = kernel_getsockopt(con->sock, | |
437 | IPPROTO_SCTP, | |
438 | SCTP_PRIMARY_ADDR, | |
439 | (char*)&prim, | |
440 | &prim_len); | |
441 | if (ret < 0) { | |
442 | log_print("getsockopt/sctp_primary_addr on " | |
443 | "new assoc %d failed : %d", | |
444 | (int)sn->sn_assoc_change.sac_assoc_id, | |
445 | ret); | |
446 | ||
447 | /* Retry INIT later */ | |
448 | new_con = assoc2con(sn->sn_assoc_change.sac_assoc_id); | |
449 | if (new_con) | |
450 | clear_bit(CF_CONNECT_PENDING, &con->flags); | |
451 | return; | |
452 | } | |
453 | make_sockaddr(&prim.ssp_addr, 0, &addr_len); | |
454 | if (dlm_addr_to_nodeid(&prim.ssp_addr, &nodeid)) { | |
455 | int i; | |
456 | unsigned char *b=(unsigned char *)&prim.ssp_addr; | |
457 | log_print("reject connect from unknown addr"); | |
458 | for (i=0; i<sizeof(struct sockaddr_storage);i++) | |
459 | printk("%02x ", b[i]); | |
460 | printk("\n"); | |
461 | sctp_send_shutdown(prim.ssp_assoc_id); | |
462 | return; | |
463 | } | |
464 | ||
465 | new_con = nodeid2con(nodeid, GFP_KERNEL); | |
466 | if (!new_con) | |
467 | return; | |
468 | ||
469 | /* Peel off a new sock */ | |
470 | parg.associd = sn->sn_assoc_change.sac_assoc_id; | |
617e82e1 DT |
471 | ret = kernel_getsockopt(con->sock, IPPROTO_SCTP, |
472 | SCTP_SOCKOPT_PEELOFF, | |
6ed7257b | 473 | (void *)&parg, &parglen); |
30d3a237 | 474 | if (ret) { |
617e82e1 DT |
475 | log_print("Can't peel off a socket for " |
476 | "connection %d to node %d: err=%d\n", | |
6ed7257b | 477 | parg.associd, nodeid, ret); |
6ed7257b | 478 | } |
6ed7257b PC |
479 | file = fget(parg.sd); |
480 | new_con->sock = SOCKET_I(file->f_dentry->d_inode); | |
481 | add_sock(new_con->sock, new_con); | |
482 | fput(file); | |
483 | put_unused_fd(parg.sd); | |
484 | ||
485 | log_print("got new/restarted association %d nodeid %d", | |
617e82e1 | 486 | (int)sn->sn_assoc_change.sac_assoc_id, nodeid); |
6ed7257b PC |
487 | |
488 | /* Send any pending writes */ | |
489 | clear_bit(CF_CONNECT_PENDING, &new_con->flags); | |
490 | clear_bit(CF_INIT_PENDING, &con->flags); | |
491 | if (!test_and_set_bit(CF_WRITE_PENDING, &new_con->flags)) { | |
492 | queue_work(send_workqueue, &new_con->swork); | |
493 | } | |
494 | if (!test_and_set_bit(CF_READ_PENDING, &new_con->flags)) | |
495 | queue_work(recv_workqueue, &new_con->rwork); | |
496 | } | |
497 | break; | |
498 | ||
499 | case SCTP_COMM_LOST: | |
500 | case SCTP_SHUTDOWN_COMP: | |
501 | { | |
502 | con = assoc2con(sn->sn_assoc_change.sac_assoc_id); | |
503 | if (con) { | |
504 | con->sctp_assoc = 0; | |
505 | } | |
506 | } | |
507 | break; | |
508 | ||
509 | /* We don't know which INIT failed, so clear the PENDING flags | |
510 | * on them all. if assoc_id is zero then it will then try | |
511 | * again */ | |
512 | ||
513 | case SCTP_CANT_STR_ASSOC: | |
514 | { | |
515 | log_print("Can't start SCTP association - retrying"); | |
516 | sctp_init_failed(); | |
517 | } | |
518 | break; | |
519 | ||
520 | default: | |
521 | log_print("unexpected SCTP assoc change id=%d state=%d", | |
522 | (int)sn->sn_assoc_change.sac_assoc_id, | |
523 | sn->sn_assoc_change.sac_state); | |
524 | } | |
525 | } | |
526 | } | |
527 | ||
fdda387f PC |
528 | /* Data received from remote end */ |
529 | static int receive_from_sock(struct connection *con) | |
530 | { | |
531 | int ret = 0; | |
58addbff AV |
532 | struct msghdr msg = {}; |
533 | struct kvec iov[2]; | |
fdda387f PC |
534 | unsigned len; |
535 | int r; | |
536 | int call_again_soon = 0; | |
58addbff | 537 | int nvec; |
6ed7257b | 538 | char incmsg[CMSG_SPACE(sizeof(struct sctp_sndrcvinfo))]; |
fdda387f | 539 | |
f1f1c1cc | 540 | mutex_lock(&con->sock_mutex); |
fdda387f | 541 | |
a34fbc63 PC |
542 | if (con->sock == NULL) { |
543 | ret = -EAGAIN; | |
544 | goto out_close; | |
545 | } | |
546 | ||
fdda387f PC |
547 | if (con->rx_page == NULL) { |
548 | /* | |
549 | * This doesn't need to be atomic, but I think it should | |
550 | * improve performance if it is. | |
551 | */ | |
552 | con->rx_page = alloc_page(GFP_ATOMIC); | |
553 | if (con->rx_page == NULL) | |
554 | goto out_resched; | |
ac33d071 | 555 | cbuf_init(&con->cb, PAGE_CACHE_SIZE); |
fdda387f PC |
556 | } |
557 | ||
6ed7257b PC |
558 | /* Only SCTP needs these really */ |
559 | memset(&incmsg, 0, sizeof(incmsg)); | |
560 | msg.msg_control = incmsg; | |
561 | msg.msg_controllen = sizeof(incmsg); | |
562 | ||
fdda387f PC |
563 | /* |
564 | * iov[0] is the bit of the circular buffer between the current end | |
565 | * point (cb.base + cb.len) and the end of the buffer. | |
566 | */ | |
ac33d071 PC |
567 | iov[0].iov_len = con->cb.base - cbuf_data(&con->cb); |
568 | iov[0].iov_base = page_address(con->rx_page) + cbuf_data(&con->cb); | |
89adc934 | 569 | iov[1].iov_len = 0; |
58addbff | 570 | nvec = 1; |
fdda387f PC |
571 | |
572 | /* | |
573 | * iov[1] is the bit of the circular buffer between the start of the | |
574 | * buffer and the start of the currently used section (cb.base) | |
575 | */ | |
ac33d071 PC |
576 | if (cbuf_data(&con->cb) >= con->cb.base) { |
577 | iov[0].iov_len = PAGE_CACHE_SIZE - cbuf_data(&con->cb); | |
fdda387f PC |
578 | iov[1].iov_len = con->cb.base; |
579 | iov[1].iov_base = page_address(con->rx_page); | |
58addbff | 580 | nvec = 2; |
fdda387f PC |
581 | } |
582 | len = iov[0].iov_len + iov[1].iov_len; | |
583 | ||
58addbff | 584 | r = ret = kernel_recvmsg(con->sock, &msg, iov, nvec, len, |
fdda387f | 585 | MSG_DONTWAIT | MSG_NOSIGNAL); |
fdda387f PC |
586 | if (ret <= 0) |
587 | goto out_close; | |
bd44e2b0 | 588 | |
6ed7257b PC |
589 | /* Process SCTP notifications */ |
590 | if (msg.msg_flags & MSG_NOTIFICATION) { | |
6ed7257b PC |
591 | msg.msg_control = incmsg; |
592 | msg.msg_controllen = sizeof(incmsg); | |
593 | ||
594 | process_sctp_notification(con, &msg, | |
617e82e1 | 595 | page_address(con->rx_page) + con->cb.base); |
6ed7257b PC |
596 | mutex_unlock(&con->sock_mutex); |
597 | return 0; | |
598 | } | |
599 | BUG_ON(con->nodeid == 0); | |
600 | ||
fdda387f PC |
601 | if (ret == len) |
602 | call_again_soon = 1; | |
ac33d071 | 603 | cbuf_add(&con->cb, ret); |
fdda387f PC |
604 | ret = dlm_process_incoming_buffer(con->nodeid, |
605 | page_address(con->rx_page), | |
606 | con->cb.base, con->cb.len, | |
607 | PAGE_CACHE_SIZE); | |
608 | if (ret == -EBADMSG) { | |
617e82e1 DT |
609 | log_print("lowcomms: addr=%p, base=%u, len=%u, " |
610 | "iov_len=%u, iov_base[0]=%p, read=%d", | |
611 | page_address(con->rx_page), con->cb.base, con->cb.len, | |
612 | len, iov[0].iov_base, r); | |
fdda387f PC |
613 | } |
614 | if (ret < 0) | |
615 | goto out_close; | |
ac33d071 | 616 | cbuf_eat(&con->cb, ret); |
fdda387f | 617 | |
ac33d071 | 618 | if (cbuf_empty(&con->cb) && !call_again_soon) { |
fdda387f PC |
619 | __free_page(con->rx_page); |
620 | con->rx_page = NULL; | |
621 | } | |
622 | ||
fdda387f PC |
623 | if (call_again_soon) |
624 | goto out_resched; | |
f1f1c1cc | 625 | mutex_unlock(&con->sock_mutex); |
ac33d071 | 626 | return 0; |
fdda387f | 627 | |
ac33d071 | 628 | out_resched: |
1d6e8131 PC |
629 | if (!test_and_set_bit(CF_READ_PENDING, &con->flags)) |
630 | queue_work(recv_workqueue, &con->rwork); | |
f1f1c1cc | 631 | mutex_unlock(&con->sock_mutex); |
bd44e2b0 | 632 | return -EAGAIN; |
fdda387f | 633 | |
ac33d071 | 634 | out_close: |
f1f1c1cc | 635 | mutex_unlock(&con->sock_mutex); |
9e5f2825 | 636 | if (ret != -EAGAIN) { |
ac33d071 | 637 | close_connection(con, false); |
fdda387f PC |
638 | /* Reconnect when there is something to send */ |
639 | } | |
a34fbc63 PC |
640 | /* Don't return success if we really got EOF */ |
641 | if (ret == 0) | |
642 | ret = -EAGAIN; | |
fdda387f | 643 | |
fdda387f PC |
644 | return ret; |
645 | } | |
646 | ||
647 | /* Listening socket is busy, accept a connection */ | |
6ed7257b | 648 | static int tcp_accept_from_sock(struct connection *con) |
fdda387f PC |
649 | { |
650 | int result; | |
651 | struct sockaddr_storage peeraddr; | |
652 | struct socket *newsock; | |
653 | int len; | |
654 | int nodeid; | |
655 | struct connection *newcon; | |
bd44e2b0 | 656 | struct connection *addcon; |
fdda387f PC |
657 | |
658 | memset(&peeraddr, 0, sizeof(peeraddr)); | |
6ed7257b | 659 | result = sock_create_kern(dlm_local_addr[0]->ss_family, SOCK_STREAM, |
ac33d071 | 660 | IPPROTO_TCP, &newsock); |
fdda387f PC |
661 | if (result < 0) |
662 | return -ENOMEM; | |
663 | ||
f1f1c1cc | 664 | mutex_lock_nested(&con->sock_mutex, 0); |
fdda387f PC |
665 | |
666 | result = -ENOTCONN; | |
667 | if (con->sock == NULL) | |
668 | goto accept_err; | |
669 | ||
670 | newsock->type = con->sock->type; | |
671 | newsock->ops = con->sock->ops; | |
672 | ||
673 | result = con->sock->ops->accept(con->sock, newsock, O_NONBLOCK); | |
674 | if (result < 0) | |
675 | goto accept_err; | |
676 | ||
677 | /* Get the connected socket's peer */ | |
678 | memset(&peeraddr, 0, sizeof(peeraddr)); | |
679 | if (newsock->ops->getname(newsock, (struct sockaddr *)&peeraddr, | |
680 | &len, 2)) { | |
681 | result = -ECONNABORTED; | |
682 | goto accept_err; | |
683 | } | |
684 | ||
685 | /* Get the new node's NODEID */ | |
686 | make_sockaddr(&peeraddr, 0, &len); | |
687 | if (dlm_addr_to_nodeid(&peeraddr, &nodeid)) { | |
617e82e1 | 688 | log_print("connect from non cluster node"); |
fdda387f | 689 | sock_release(newsock); |
f1f1c1cc | 690 | mutex_unlock(&con->sock_mutex); |
fdda387f PC |
691 | return -1; |
692 | } | |
693 | ||
694 | log_print("got connection from %d", nodeid); | |
695 | ||
696 | /* Check to see if we already have a connection to this node. This | |
697 | * could happen if the two nodes initiate a connection at roughly | |
698 | * the same time and the connections cross on the wire. | |
fdda387f PC |
699 | * In this case we store the incoming one in "othercon" |
700 | */ | |
701 | newcon = nodeid2con(nodeid, GFP_KERNEL); | |
702 | if (!newcon) { | |
703 | result = -ENOMEM; | |
704 | goto accept_err; | |
705 | } | |
f1f1c1cc | 706 | mutex_lock_nested(&newcon->sock_mutex, 1); |
fdda387f | 707 | if (newcon->sock) { |
ac33d071 | 708 | struct connection *othercon = newcon->othercon; |
fdda387f PC |
709 | |
710 | if (!othercon) { | |
ac33d071 | 711 | othercon = kmem_cache_zalloc(con_cache, GFP_KERNEL); |
fdda387f | 712 | if (!othercon) { |
617e82e1 | 713 | log_print("failed to allocate incoming socket"); |
f1f1c1cc | 714 | mutex_unlock(&newcon->sock_mutex); |
fdda387f PC |
715 | result = -ENOMEM; |
716 | goto accept_err; | |
717 | } | |
fdda387f PC |
718 | othercon->nodeid = nodeid; |
719 | othercon->rx_action = receive_from_sock; | |
f1f1c1cc | 720 | mutex_init(&othercon->sock_mutex); |
1d6e8131 PC |
721 | INIT_WORK(&othercon->swork, process_send_sockets); |
722 | INIT_WORK(&othercon->rwork, process_recv_sockets); | |
fdda387f | 723 | set_bit(CF_IS_OTHERCON, &othercon->flags); |
61d96be0 PC |
724 | } |
725 | if (!othercon->sock) { | |
fdda387f | 726 | newcon->othercon = othercon; |
97d84836 PC |
727 | othercon->sock = newsock; |
728 | newsock->sk->sk_user_data = othercon; | |
729 | add_sock(newsock, othercon); | |
730 | addcon = othercon; | |
731 | } | |
732 | else { | |
733 | printk("Extra connection from node %d attempted\n", nodeid); | |
734 | result = -EAGAIN; | |
f4fadb23 | 735 | mutex_unlock(&newcon->sock_mutex); |
97d84836 | 736 | goto accept_err; |
fdda387f | 737 | } |
fdda387f PC |
738 | } |
739 | else { | |
740 | newsock->sk->sk_user_data = newcon; | |
741 | newcon->rx_action = receive_from_sock; | |
742 | add_sock(newsock, newcon); | |
bd44e2b0 | 743 | addcon = newcon; |
fdda387f PC |
744 | } |
745 | ||
f1f1c1cc | 746 | mutex_unlock(&newcon->sock_mutex); |
fdda387f PC |
747 | |
748 | /* | |
749 | * Add it to the active queue in case we got data | |
750 | * beween processing the accept adding the socket | |
751 | * to the read_sockets list | |
752 | */ | |
bd44e2b0 PC |
753 | if (!test_and_set_bit(CF_READ_PENDING, &addcon->flags)) |
754 | queue_work(recv_workqueue, &addcon->rwork); | |
f1f1c1cc | 755 | mutex_unlock(&con->sock_mutex); |
fdda387f PC |
756 | |
757 | return 0; | |
758 | ||
ac33d071 | 759 | accept_err: |
f1f1c1cc | 760 | mutex_unlock(&con->sock_mutex); |
fdda387f PC |
761 | sock_release(newsock); |
762 | ||
763 | if (result != -EAGAIN) | |
617e82e1 | 764 | log_print("error accepting connection from node: %d", result); |
fdda387f PC |
765 | return result; |
766 | } | |
767 | ||
6ed7257b PC |
768 | static void free_entry(struct writequeue_entry *e) |
769 | { | |
770 | __free_page(e->page); | |
771 | kfree(e); | |
772 | } | |
773 | ||
774 | /* Initiate an SCTP association. | |
775 | This is a special case of send_to_sock() in that we don't yet have a | |
776 | peeled-off socket for this association, so we use the listening socket | |
777 | and add the primary IP address of the remote node. | |
778 | */ | |
779 | static void sctp_init_assoc(struct connection *con) | |
780 | { | |
781 | struct sockaddr_storage rem_addr; | |
782 | char outcmsg[CMSG_SPACE(sizeof(struct sctp_sndrcvinfo))]; | |
783 | struct msghdr outmessage; | |
784 | struct cmsghdr *cmsg; | |
785 | struct sctp_sndrcvinfo *sinfo; | |
786 | struct connection *base_con; | |
787 | struct writequeue_entry *e; | |
788 | int len, offset; | |
789 | int ret; | |
790 | int addrlen; | |
791 | struct kvec iov[1]; | |
792 | ||
793 | if (test_and_set_bit(CF_INIT_PENDING, &con->flags)) | |
794 | return; | |
795 | ||
796 | if (con->retries++ > MAX_CONNECT_RETRIES) | |
797 | return; | |
798 | ||
799 | log_print("Initiating association with node %d", con->nodeid); | |
800 | ||
801 | if (nodeid_to_addr(con->nodeid, (struct sockaddr *)&rem_addr)) { | |
802 | log_print("no address for nodeid %d", con->nodeid); | |
803 | return; | |
804 | } | |
805 | base_con = nodeid2con(0, 0); | |
806 | BUG_ON(base_con == NULL); | |
807 | ||
808 | make_sockaddr(&rem_addr, dlm_config.ci_tcp_port, &addrlen); | |
809 | ||
810 | outmessage.msg_name = &rem_addr; | |
811 | outmessage.msg_namelen = addrlen; | |
812 | outmessage.msg_control = outcmsg; | |
813 | outmessage.msg_controllen = sizeof(outcmsg); | |
814 | outmessage.msg_flags = MSG_EOR; | |
815 | ||
816 | spin_lock(&con->writequeue_lock); | |
817 | e = list_entry(con->writequeue.next, struct writequeue_entry, | |
818 | list); | |
819 | ||
820 | BUG_ON((struct list_head *) e == &con->writequeue); | |
821 | ||
822 | len = e->len; | |
823 | offset = e->offset; | |
824 | spin_unlock(&con->writequeue_lock); | |
825 | kmap(e->page); | |
826 | ||
827 | /* Send the first block off the write queue */ | |
828 | iov[0].iov_base = page_address(e->page)+offset; | |
829 | iov[0].iov_len = len; | |
830 | ||
831 | cmsg = CMSG_FIRSTHDR(&outmessage); | |
832 | cmsg->cmsg_level = IPPROTO_SCTP; | |
833 | cmsg->cmsg_type = SCTP_SNDRCV; | |
834 | cmsg->cmsg_len = CMSG_LEN(sizeof(struct sctp_sndrcvinfo)); | |
835 | sinfo = CMSG_DATA(cmsg); | |
836 | memset(sinfo, 0x00, sizeof(struct sctp_sndrcvinfo)); | |
837 | sinfo->sinfo_ppid = cpu_to_le32(dlm_our_nodeid()); | |
838 | outmessage.msg_controllen = cmsg->cmsg_len; | |
839 | ||
840 | ret = kernel_sendmsg(base_con->sock, &outmessage, iov, 1, len); | |
841 | if (ret < 0) { | |
617e82e1 DT |
842 | log_print("Send first packet to node %d failed: %d", |
843 | con->nodeid, ret); | |
6ed7257b PC |
844 | |
845 | /* Try again later */ | |
846 | clear_bit(CF_CONNECT_PENDING, &con->flags); | |
847 | clear_bit(CF_INIT_PENDING, &con->flags); | |
848 | } | |
849 | else { | |
850 | spin_lock(&con->writequeue_lock); | |
851 | e->offset += ret; | |
852 | e->len -= ret; | |
853 | ||
854 | if (e->len == 0 && e->users == 0) { | |
855 | list_del(&e->list); | |
856 | kunmap(e->page); | |
857 | free_entry(e); | |
858 | } | |
859 | spin_unlock(&con->writequeue_lock); | |
860 | } | |
861 | } | |
862 | ||
fdda387f | 863 | /* Connect a new socket to its peer */ |
6ed7257b | 864 | static void tcp_connect_to_sock(struct connection *con) |
fdda387f PC |
865 | { |
866 | int result = -EHOSTUNREACH; | |
6bd8feda | 867 | struct sockaddr_storage saddr, src_addr; |
fdda387f PC |
868 | int addr_len; |
869 | struct socket *sock; | |
870 | ||
871 | if (con->nodeid == 0) { | |
872 | log_print("attempt to connect sock 0 foiled"); | |
ac33d071 | 873 | return; |
fdda387f PC |
874 | } |
875 | ||
f1f1c1cc | 876 | mutex_lock(&con->sock_mutex); |
fdda387f PC |
877 | if (con->retries++ > MAX_CONNECT_RETRIES) |
878 | goto out; | |
879 | ||
880 | /* Some odd races can cause double-connects, ignore them */ | |
881 | if (con->sock) { | |
882 | result = 0; | |
883 | goto out; | |
884 | } | |
885 | ||
886 | /* Create a socket to communicate with */ | |
6ed7257b | 887 | result = sock_create_kern(dlm_local_addr[0]->ss_family, SOCK_STREAM, |
ac33d071 | 888 | IPPROTO_TCP, &sock); |
fdda387f PC |
889 | if (result < 0) |
890 | goto out_err; | |
891 | ||
892 | memset(&saddr, 0, sizeof(saddr)); | |
893 | if (dlm_nodeid_to_addr(con->nodeid, &saddr)) | |
ac33d071 | 894 | goto out_err; |
fdda387f PC |
895 | |
896 | sock->sk->sk_user_data = con; | |
897 | con->rx_action = receive_from_sock; | |
6ed7257b PC |
898 | con->connect_action = tcp_connect_to_sock; |
899 | add_sock(sock, con); | |
fdda387f | 900 | |
6bd8feda LH |
901 | /* Bind to our cluster-known address connecting to avoid |
902 | routing problems */ | |
903 | memcpy(&src_addr, dlm_local_addr[0], sizeof(src_addr)); | |
904 | make_sockaddr(&src_addr, 0, &addr_len); | |
905 | result = sock->ops->bind(sock, (struct sockaddr *) &src_addr, | |
906 | addr_len); | |
907 | if (result < 0) { | |
908 | log_print("could not bind for connect: %d", result); | |
909 | /* This *may* not indicate a critical error */ | |
910 | } | |
911 | ||
68c817a1 | 912 | make_sockaddr(&saddr, dlm_config.ci_tcp_port, &addr_len); |
fdda387f | 913 | |
fdda387f PC |
914 | log_print("connecting to %d", con->nodeid); |
915 | result = | |
916 | sock->ops->connect(sock, (struct sockaddr *)&saddr, addr_len, | |
ac33d071 | 917 | O_NONBLOCK); |
fdda387f PC |
918 | if (result == -EINPROGRESS) |
919 | result = 0; | |
ac33d071 PC |
920 | if (result == 0) |
921 | goto out; | |
fdda387f | 922 | |
ac33d071 | 923 | out_err: |
fdda387f PC |
924 | if (con->sock) { |
925 | sock_release(con->sock); | |
926 | con->sock = NULL; | |
927 | } | |
928 | /* | |
929 | * Some errors are fatal and this list might need adjusting. For other | |
930 | * errors we try again until the max number of retries is reached. | |
931 | */ | |
932 | if (result != -EHOSTUNREACH && result != -ENETUNREACH && | |
933 | result != -ENETDOWN && result != EINVAL | |
934 | && result != -EPROTONOSUPPORT) { | |
935 | lowcomms_connect_sock(con); | |
936 | result = 0; | |
937 | } | |
ac33d071 | 938 | out: |
f1f1c1cc | 939 | mutex_unlock(&con->sock_mutex); |
ac33d071 | 940 | return; |
fdda387f PC |
941 | } |
942 | ||
6ed7257b PC |
943 | static struct socket *tcp_create_listen_sock(struct connection *con, |
944 | struct sockaddr_storage *saddr) | |
fdda387f | 945 | { |
ac33d071 | 946 | struct socket *sock = NULL; |
fdda387f PC |
947 | int result = 0; |
948 | int one = 1; | |
949 | int addr_len; | |
950 | ||
6ed7257b | 951 | if (dlm_local_addr[0]->ss_family == AF_INET) |
fdda387f PC |
952 | addr_len = sizeof(struct sockaddr_in); |
953 | else | |
954 | addr_len = sizeof(struct sockaddr_in6); | |
955 | ||
956 | /* Create a socket to communicate with */ | |
617e82e1 DT |
957 | result = sock_create_kern(dlm_local_addr[0]->ss_family, SOCK_STREAM, |
958 | IPPROTO_TCP, &sock); | |
fdda387f | 959 | if (result < 0) { |
617e82e1 | 960 | log_print("Can't create listening comms socket"); |
fdda387f PC |
961 | goto create_out; |
962 | } | |
963 | ||
6ed7257b PC |
964 | result = kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEADDR, |
965 | (char *)&one, sizeof(one)); | |
966 | ||
fdda387f | 967 | if (result < 0) { |
617e82e1 | 968 | log_print("Failed to set SO_REUSEADDR on socket: %d", result); |
fdda387f PC |
969 | } |
970 | sock->sk->sk_user_data = con; | |
6ed7257b PC |
971 | con->rx_action = tcp_accept_from_sock; |
972 | con->connect_action = tcp_connect_to_sock; | |
fdda387f PC |
973 | con->sock = sock; |
974 | ||
975 | /* Bind to our port */ | |
68c817a1 | 976 | make_sockaddr(saddr, dlm_config.ci_tcp_port, &addr_len); |
fdda387f PC |
977 | result = sock->ops->bind(sock, (struct sockaddr *) saddr, addr_len); |
978 | if (result < 0) { | |
617e82e1 | 979 | log_print("Can't bind to port %d", dlm_config.ci_tcp_port); |
fdda387f PC |
980 | sock_release(sock); |
981 | sock = NULL; | |
982 | con->sock = NULL; | |
983 | goto create_out; | |
984 | } | |
6ed7257b | 985 | result = kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE, |
ac33d071 | 986 | (char *)&one, sizeof(one)); |
fdda387f | 987 | if (result < 0) { |
617e82e1 | 988 | log_print("Set keepalive failed: %d", result); |
fdda387f PC |
989 | } |
990 | ||
991 | result = sock->ops->listen(sock, 5); | |
992 | if (result < 0) { | |
617e82e1 | 993 | log_print("Can't listen on port %d", dlm_config.ci_tcp_port); |
fdda387f PC |
994 | sock_release(sock); |
995 | sock = NULL; | |
996 | goto create_out; | |
997 | } | |
998 | ||
ac33d071 | 999 | create_out: |
fdda387f PC |
1000 | return sock; |
1001 | } | |
1002 | ||
6ed7257b PC |
1003 | /* Get local addresses */ |
1004 | static void init_local(void) | |
1005 | { | |
1006 | struct sockaddr_storage sas, *addr; | |
1007 | int i; | |
1008 | ||
30d3a237 | 1009 | dlm_local_count = 0; |
6ed7257b PC |
1010 | for (i = 0; i < DLM_MAX_ADDR_COUNT - 1; i++) { |
1011 | if (dlm_our_addr(&sas, i)) | |
1012 | break; | |
1013 | ||
1014 | addr = kmalloc(sizeof(*addr), GFP_KERNEL); | |
1015 | if (!addr) | |
1016 | break; | |
1017 | memcpy(addr, &sas, sizeof(*addr)); | |
1018 | dlm_local_addr[dlm_local_count++] = addr; | |
1019 | } | |
1020 | } | |
1021 | ||
617e82e1 DT |
1022 | /* Bind to an IP address. SCTP allows multiple address so it can do |
1023 | multi-homing */ | |
1024 | static int add_sctp_bind_addr(struct connection *sctp_con, | |
1025 | struct sockaddr_storage *addr, | |
1026 | int addr_len, int num) | |
6ed7257b PC |
1027 | { |
1028 | int result = 0; | |
1029 | ||
1030 | if (num == 1) | |
1031 | result = kernel_bind(sctp_con->sock, | |
1032 | (struct sockaddr *) addr, | |
1033 | addr_len); | |
1034 | else | |
1035 | result = kernel_setsockopt(sctp_con->sock, SOL_SCTP, | |
1036 | SCTP_SOCKOPT_BINDX_ADD, | |
1037 | (char *)addr, addr_len); | |
1038 | ||
1039 | if (result < 0) | |
1040 | log_print("Can't bind to port %d addr number %d", | |
1041 | dlm_config.ci_tcp_port, num); | |
1042 | ||
1043 | return result; | |
1044 | } | |
fdda387f | 1045 | |
6ed7257b PC |
1046 | /* Initialise SCTP socket and bind to all interfaces */ |
1047 | static int sctp_listen_for_all(void) | |
1048 | { | |
1049 | struct socket *sock = NULL; | |
1050 | struct sockaddr_storage localaddr; | |
1051 | struct sctp_event_subscribe subscribe; | |
1052 | int result = -EINVAL, num = 1, i, addr_len; | |
1053 | struct connection *con = nodeid2con(0, GFP_KERNEL); | |
1054 | int bufsize = NEEDED_RMEM; | |
1055 | ||
1056 | if (!con) | |
1057 | return -ENOMEM; | |
1058 | ||
1059 | log_print("Using SCTP for communications"); | |
1060 | ||
1061 | result = sock_create_kern(dlm_local_addr[0]->ss_family, SOCK_SEQPACKET, | |
1062 | IPPROTO_SCTP, &sock); | |
1063 | if (result < 0) { | |
1064 | log_print("Can't create comms socket, check SCTP is loaded"); | |
1065 | goto out; | |
1066 | } | |
1067 | ||
1068 | /* Listen for events */ | |
1069 | memset(&subscribe, 0, sizeof(subscribe)); | |
1070 | subscribe.sctp_data_io_event = 1; | |
1071 | subscribe.sctp_association_event = 1; | |
1072 | subscribe.sctp_send_failure_event = 1; | |
1073 | subscribe.sctp_shutdown_event = 1; | |
1074 | subscribe.sctp_partial_delivery_event = 1; | |
1075 | ||
df61c952 | 1076 | result = kernel_setsockopt(sock, SOL_SOCKET, SO_RCVBUFFORCE, |
6ed7257b PC |
1077 | (char *)&bufsize, sizeof(bufsize)); |
1078 | if (result) | |
617e82e1 | 1079 | log_print("Error increasing buffer space on socket %d", result); |
6ed7257b PC |
1080 | |
1081 | result = kernel_setsockopt(sock, SOL_SCTP, SCTP_EVENTS, | |
617e82e1 | 1082 | (char *)&subscribe, sizeof(subscribe)); |
6ed7257b PC |
1083 | if (result < 0) { |
1084 | log_print("Failed to set SCTP_EVENTS on socket: result=%d", | |
1085 | result); | |
1086 | goto create_delsock; | |
1087 | } | |
1088 | ||
1089 | /* Init con struct */ | |
1090 | sock->sk->sk_user_data = con; | |
1091 | con->sock = sock; | |
1092 | con->sock->sk->sk_data_ready = lowcomms_data_ready; | |
1093 | con->rx_action = receive_from_sock; | |
1094 | con->connect_action = sctp_init_assoc; | |
1095 | ||
1096 | /* Bind to all interfaces. */ | |
1097 | for (i = 0; i < dlm_local_count; i++) { | |
1098 | memcpy(&localaddr, dlm_local_addr[i], sizeof(localaddr)); | |
1099 | make_sockaddr(&localaddr, dlm_config.ci_tcp_port, &addr_len); | |
1100 | ||
1101 | result = add_sctp_bind_addr(con, &localaddr, addr_len, num); | |
1102 | if (result) | |
1103 | goto create_delsock; | |
1104 | ++num; | |
1105 | } | |
1106 | ||
1107 | result = sock->ops->listen(sock, 5); | |
1108 | if (result < 0) { | |
1109 | log_print("Can't set socket listening"); | |
1110 | goto create_delsock; | |
1111 | } | |
1112 | ||
1113 | return 0; | |
1114 | ||
1115 | create_delsock: | |
1116 | sock_release(sock); | |
1117 | con->sock = NULL; | |
1118 | out: | |
1119 | return result; | |
1120 | } | |
1121 | ||
1122 | static int tcp_listen_for_all(void) | |
fdda387f PC |
1123 | { |
1124 | struct socket *sock = NULL; | |
1125 | struct connection *con = nodeid2con(0, GFP_KERNEL); | |
1126 | int result = -EINVAL; | |
1127 | ||
6ed7257b PC |
1128 | if (!con) |
1129 | return -ENOMEM; | |
1130 | ||
fdda387f | 1131 | /* We don't support multi-homed hosts */ |
6ed7257b | 1132 | if (dlm_local_addr[1] != NULL) { |
617e82e1 DT |
1133 | log_print("TCP protocol can't handle multi-homed hosts, " |
1134 | "try SCTP"); | |
6ed7257b PC |
1135 | return -EINVAL; |
1136 | } | |
1137 | ||
1138 | log_print("Using TCP for communications"); | |
1139 | ||
6ed7257b | 1140 | sock = tcp_create_listen_sock(con, dlm_local_addr[0]); |
fdda387f PC |
1141 | if (sock) { |
1142 | add_sock(sock, con); | |
1143 | result = 0; | |
1144 | } | |
1145 | else { | |
1146 | result = -EADDRINUSE; | |
1147 | } | |
1148 | ||
1149 | return result; | |
1150 | } | |
1151 | ||
1152 | ||
1153 | ||
1154 | static struct writequeue_entry *new_writequeue_entry(struct connection *con, | |
1155 | gfp_t allocation) | |
1156 | { | |
1157 | struct writequeue_entry *entry; | |
1158 | ||
1159 | entry = kmalloc(sizeof(struct writequeue_entry), allocation); | |
1160 | if (!entry) | |
1161 | return NULL; | |
1162 | ||
1163 | entry->page = alloc_page(allocation); | |
1164 | if (!entry->page) { | |
1165 | kfree(entry); | |
1166 | return NULL; | |
1167 | } | |
1168 | ||
1169 | entry->offset = 0; | |
1170 | entry->len = 0; | |
1171 | entry->end = 0; | |
1172 | entry->users = 0; | |
1173 | entry->con = con; | |
1174 | ||
1175 | return entry; | |
1176 | } | |
1177 | ||
617e82e1 | 1178 | void *dlm_lowcomms_get_buffer(int nodeid, int len, gfp_t allocation, char **ppc) |
fdda387f PC |
1179 | { |
1180 | struct connection *con; | |
1181 | struct writequeue_entry *e; | |
1182 | int offset = 0; | |
1183 | int users = 0; | |
1184 | ||
fdda387f PC |
1185 | con = nodeid2con(nodeid, allocation); |
1186 | if (!con) | |
1187 | return NULL; | |
1188 | ||
4edde74e | 1189 | spin_lock(&con->writequeue_lock); |
fdda387f | 1190 | e = list_entry(con->writequeue.prev, struct writequeue_entry, list); |
ac33d071 | 1191 | if ((&e->list == &con->writequeue) || |
fdda387f PC |
1192 | (PAGE_CACHE_SIZE - e->end < len)) { |
1193 | e = NULL; | |
1194 | } else { | |
1195 | offset = e->end; | |
1196 | e->end += len; | |
1197 | users = e->users++; | |
1198 | } | |
1199 | spin_unlock(&con->writequeue_lock); | |
1200 | ||
1201 | if (e) { | |
ac33d071 | 1202 | got_one: |
fdda387f PC |
1203 | if (users == 0) |
1204 | kmap(e->page); | |
1205 | *ppc = page_address(e->page) + offset; | |
1206 | return e; | |
1207 | } | |
1208 | ||
1209 | e = new_writequeue_entry(con, allocation); | |
1210 | if (e) { | |
1211 | spin_lock(&con->writequeue_lock); | |
1212 | offset = e->end; | |
1213 | e->end += len; | |
1214 | users = e->users++; | |
1215 | list_add_tail(&e->list, &con->writequeue); | |
1216 | spin_unlock(&con->writequeue_lock); | |
1217 | goto got_one; | |
1218 | } | |
1219 | return NULL; | |
1220 | } | |
1221 | ||
1222 | void dlm_lowcomms_commit_buffer(void *mh) | |
1223 | { | |
1224 | struct writequeue_entry *e = (struct writequeue_entry *)mh; | |
1225 | struct connection *con = e->con; | |
1226 | int users; | |
1227 | ||
4edde74e | 1228 | spin_lock(&con->writequeue_lock); |
fdda387f PC |
1229 | users = --e->users; |
1230 | if (users) | |
1231 | goto out; | |
1232 | e->len = e->end - e->offset; | |
1233 | kunmap(e->page); | |
1234 | spin_unlock(&con->writequeue_lock); | |
1235 | ||
1d6e8131 PC |
1236 | if (!test_and_set_bit(CF_WRITE_PENDING, &con->flags)) { |
1237 | queue_work(send_workqueue, &con->swork); | |
fdda387f PC |
1238 | } |
1239 | return; | |
1240 | ||
ac33d071 | 1241 | out: |
fdda387f PC |
1242 | spin_unlock(&con->writequeue_lock); |
1243 | return; | |
1244 | } | |
1245 | ||
fdda387f | 1246 | /* Send a message */ |
ac33d071 | 1247 | static void send_to_sock(struct connection *con) |
fdda387f PC |
1248 | { |
1249 | int ret = 0; | |
1250 | ssize_t(*sendpage) (struct socket *, struct page *, int, size_t, int); | |
1251 | const int msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL; | |
1252 | struct writequeue_entry *e; | |
1253 | int len, offset; | |
1254 | ||
f1f1c1cc | 1255 | mutex_lock(&con->sock_mutex); |
fdda387f PC |
1256 | if (con->sock == NULL) |
1257 | goto out_connect; | |
1258 | ||
1259 | sendpage = con->sock->ops->sendpage; | |
1260 | ||
1261 | spin_lock(&con->writequeue_lock); | |
1262 | for (;;) { | |
1263 | e = list_entry(con->writequeue.next, struct writequeue_entry, | |
1264 | list); | |
1265 | if ((struct list_head *) e == &con->writequeue) | |
1266 | break; | |
1267 | ||
1268 | len = e->len; | |
1269 | offset = e->offset; | |
1270 | BUG_ON(len == 0 && e->users == 0); | |
1271 | spin_unlock(&con->writequeue_lock); | |
1d6e8131 | 1272 | kmap(e->page); |
fdda387f PC |
1273 | |
1274 | ret = 0; | |
1275 | if (len) { | |
1276 | ret = sendpage(con->sock, e->page, offset, len, | |
1277 | msg_flags); | |
d66f8277 PC |
1278 | if (ret == -EAGAIN || ret == 0) { |
1279 | cond_resched(); | |
fdda387f | 1280 | goto out; |
d66f8277 | 1281 | } |
fdda387f PC |
1282 | if (ret <= 0) |
1283 | goto send_error; | |
d66f8277 | 1284 | } |
fdda387f | 1285 | /* Don't starve people filling buffers */ |
ac33d071 | 1286 | cond_resched(); |
fdda387f PC |
1287 | |
1288 | spin_lock(&con->writequeue_lock); | |
1289 | e->offset += ret; | |
1290 | e->len -= ret; | |
1291 | ||
1292 | if (e->len == 0 && e->users == 0) { | |
1293 | list_del(&e->list); | |
ac33d071 | 1294 | kunmap(e->page); |
fdda387f PC |
1295 | free_entry(e); |
1296 | continue; | |
1297 | } | |
1298 | } | |
1299 | spin_unlock(&con->writequeue_lock); | |
ac33d071 | 1300 | out: |
f1f1c1cc | 1301 | mutex_unlock(&con->sock_mutex); |
ac33d071 | 1302 | return; |
fdda387f | 1303 | |
ac33d071 | 1304 | send_error: |
f1f1c1cc | 1305 | mutex_unlock(&con->sock_mutex); |
ac33d071 | 1306 | close_connection(con, false); |
fdda387f | 1307 | lowcomms_connect_sock(con); |
ac33d071 | 1308 | return; |
fdda387f | 1309 | |
ac33d071 | 1310 | out_connect: |
f1f1c1cc | 1311 | mutex_unlock(&con->sock_mutex); |
6ed7257b PC |
1312 | if (!test_bit(CF_INIT_PENDING, &con->flags)) |
1313 | lowcomms_connect_sock(con); | |
ac33d071 | 1314 | return; |
fdda387f PC |
1315 | } |
1316 | ||
1317 | static void clean_one_writequeue(struct connection *con) | |
1318 | { | |
1319 | struct list_head *list; | |
1320 | struct list_head *temp; | |
1321 | ||
1322 | spin_lock(&con->writequeue_lock); | |
1323 | list_for_each_safe(list, temp, &con->writequeue) { | |
1324 | struct writequeue_entry *e = | |
1325 | list_entry(list, struct writequeue_entry, list); | |
1326 | list_del(&e->list); | |
1327 | free_entry(e); | |
1328 | } | |
1329 | spin_unlock(&con->writequeue_lock); | |
1330 | } | |
1331 | ||
1332 | /* Called from recovery when it knows that a node has | |
1333 | left the cluster */ | |
1334 | int dlm_lowcomms_close(int nodeid) | |
1335 | { | |
1336 | struct connection *con; | |
1337 | ||
fdda387f PC |
1338 | log_print("closing connection to node %d", nodeid); |
1339 | con = nodeid2con(nodeid, 0); | |
1340 | if (con) { | |
1341 | clean_one_writequeue(con); | |
ac33d071 | 1342 | close_connection(con, true); |
fdda387f PC |
1343 | } |
1344 | return 0; | |
fdda387f PC |
1345 | } |
1346 | ||
6ed7257b | 1347 | /* Receive workqueue function */ |
1d6e8131 | 1348 | static void process_recv_sockets(struct work_struct *work) |
fdda387f | 1349 | { |
1d6e8131 PC |
1350 | struct connection *con = container_of(work, struct connection, rwork); |
1351 | int err; | |
fdda387f | 1352 | |
1d6e8131 PC |
1353 | clear_bit(CF_READ_PENDING, &con->flags); |
1354 | do { | |
1355 | err = con->rx_action(con); | |
1356 | } while (!err); | |
fdda387f PC |
1357 | } |
1358 | ||
6ed7257b | 1359 | /* Send workqueue function */ |
1d6e8131 | 1360 | static void process_send_sockets(struct work_struct *work) |
fdda387f | 1361 | { |
1d6e8131 | 1362 | struct connection *con = container_of(work, struct connection, swork); |
fdda387f | 1363 | |
1d6e8131 | 1364 | if (test_and_clear_bit(CF_CONNECT_PENDING, &con->flags)) { |
6ed7257b | 1365 | con->connect_action(con); |
fdda387f | 1366 | } |
bd44e2b0 PC |
1367 | clear_bit(CF_WRITE_PENDING, &con->flags); |
1368 | send_to_sock(con); | |
fdda387f PC |
1369 | } |
1370 | ||
1371 | ||
1372 | /* Discard all entries on the write queues */ | |
1373 | static void clean_writequeues(void) | |
1374 | { | |
1375 | int nodeid; | |
1376 | ||
30d3a237 PC |
1377 | for (nodeid = 1; nodeid <= max_nodeid; nodeid++) { |
1378 | struct connection *con = __nodeid2con(nodeid, 0); | |
fdda387f PC |
1379 | |
1380 | if (con) | |
1381 | clean_one_writequeue(con); | |
1382 | } | |
1383 | } | |
1384 | ||
1d6e8131 | 1385 | static void work_stop(void) |
fdda387f | 1386 | { |
1d6e8131 PC |
1387 | destroy_workqueue(recv_workqueue); |
1388 | destroy_workqueue(send_workqueue); | |
fdda387f PC |
1389 | } |
1390 | ||
1d6e8131 | 1391 | static int work_start(void) |
fdda387f | 1392 | { |
fdda387f | 1393 | int error; |
1d6e8131 PC |
1394 | recv_workqueue = create_workqueue("dlm_recv"); |
1395 | error = IS_ERR(recv_workqueue); | |
ac33d071 | 1396 | if (error) { |
1d6e8131 | 1397 | log_print("can't start dlm_recv %d", error); |
fdda387f PC |
1398 | return error; |
1399 | } | |
fdda387f | 1400 | |
1d6e8131 PC |
1401 | send_workqueue = create_singlethread_workqueue("dlm_send"); |
1402 | error = IS_ERR(send_workqueue); | |
ac33d071 | 1403 | if (error) { |
1d6e8131 PC |
1404 | log_print("can't start dlm_send %d", error); |
1405 | destroy_workqueue(recv_workqueue); | |
fdda387f PC |
1406 | return error; |
1407 | } | |
fdda387f PC |
1408 | |
1409 | return 0; | |
1410 | } | |
1411 | ||
fdda387f PC |
1412 | void dlm_lowcomms_stop(void) |
1413 | { | |
1414 | int i; | |
6ed7257b | 1415 | struct connection *con; |
fdda387f | 1416 | |
ac33d071 | 1417 | /* Set all the flags to prevent any |
fdda387f PC |
1418 | socket activity. |
1419 | */ | |
6ed7257b | 1420 | down(&connections_lock); |
30d3a237 | 1421 | for (i = 0; i <= max_nodeid; i++) { |
6ed7257b | 1422 | con = __nodeid2con(i, 0); |
afb853fb | 1423 | if (con) { |
9e5f2825 | 1424 | con->flags |= 0x0F; |
afb853fb PC |
1425 | if (con->sock) |
1426 | con->sock->sk->sk_user_data = NULL; | |
1427 | } | |
fdda387f | 1428 | } |
6ed7257b | 1429 | up(&connections_lock); |
ac33d071 | 1430 | |
1d6e8131 | 1431 | work_stop(); |
6ed7257b PC |
1432 | |
1433 | down(&connections_lock); | |
fdda387f PC |
1434 | clean_writequeues(); |
1435 | ||
30d3a237 | 1436 | for (i = 0; i <= max_nodeid; i++) { |
2439fe50 | 1437 | con = __nodeid2con(i, 0); |
6ed7257b PC |
1438 | if (con) { |
1439 | close_connection(con, true); | |
39bd4177 PC |
1440 | if (con->othercon) |
1441 | kmem_cache_free(con_cache, con->othercon); | |
6ed7257b | 1442 | kmem_cache_free(con_cache, con); |
fdda387f PC |
1443 | } |
1444 | } | |
30d3a237 | 1445 | max_nodeid = 0; |
6ed7257b | 1446 | up(&connections_lock); |
fdda387f | 1447 | kmem_cache_destroy(con_cache); |
30d3a237 | 1448 | idr_init(&connections_idr); |
fdda387f PC |
1449 | } |
1450 | ||
fdda387f PC |
1451 | int dlm_lowcomms_start(void) |
1452 | { | |
6ed7257b PC |
1453 | int error = -EINVAL; |
1454 | struct connection *con; | |
fdda387f | 1455 | |
6ed7257b PC |
1456 | init_local(); |
1457 | if (!dlm_local_count) { | |
617e82e1 | 1458 | error = -ENOTCONN; |
fdda387f | 1459 | log_print("no local IP address has been set"); |
6ed7257b | 1460 | goto out; |
fdda387f PC |
1461 | } |
1462 | ||
6ed7257b | 1463 | error = -ENOMEM; |
fdda387f | 1464 | con_cache = kmem_cache_create("dlm_conn", sizeof(struct connection), |
ac33d071 | 1465 | __alignof__(struct connection), 0, |
20c2df83 | 1466 | NULL); |
fdda387f | 1467 | if (!con_cache) |
6ed7257b | 1468 | goto out; |
fdda387f | 1469 | |
fdda387f | 1470 | /* Start listening */ |
6ed7257b PC |
1471 | if (dlm_config.ci_protocol == 0) |
1472 | error = tcp_listen_for_all(); | |
1473 | else | |
1474 | error = sctp_listen_for_all(); | |
fdda387f PC |
1475 | if (error) |
1476 | goto fail_unlisten; | |
1477 | ||
1d6e8131 | 1478 | error = work_start(); |
fdda387f PC |
1479 | if (error) |
1480 | goto fail_unlisten; | |
1481 | ||
fdda387f PC |
1482 | return 0; |
1483 | ||
ac33d071 | 1484 | fail_unlisten: |
6ed7257b PC |
1485 | con = nodeid2con(0,0); |
1486 | if (con) { | |
1487 | close_connection(con, false); | |
1488 | kmem_cache_free(con_cache, con); | |
1489 | } | |
fdda387f PC |
1490 | kmem_cache_destroy(con_cache); |
1491 | ||
ac33d071 | 1492 | out: |
fdda387f PC |
1493 | return error; |
1494 | } |