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); | |
25720c2d PC |
331 | kmem_cache_free(con_cache, con->othercon); |
332 | con->othercon = NULL; | |
fdda387f PC |
333 | } |
334 | if (con->rx_page) { | |
335 | __free_page(con->rx_page); | |
336 | con->rx_page = NULL; | |
337 | } | |
338 | con->retries = 0; | |
f1f1c1cc | 339 | mutex_unlock(&con->sock_mutex); |
fdda387f PC |
340 | } |
341 | ||
6ed7257b PC |
342 | /* We only send shutdown messages to nodes that are not part of the cluster */ |
343 | static void sctp_send_shutdown(sctp_assoc_t associd) | |
344 | { | |
345 | static char outcmsg[CMSG_SPACE(sizeof(struct sctp_sndrcvinfo))]; | |
346 | struct msghdr outmessage; | |
347 | struct cmsghdr *cmsg; | |
348 | struct sctp_sndrcvinfo *sinfo; | |
349 | int ret; | |
350 | struct connection *con; | |
351 | ||
352 | con = nodeid2con(0,0); | |
353 | BUG_ON(con == NULL); | |
354 | ||
355 | outmessage.msg_name = NULL; | |
356 | outmessage.msg_namelen = 0; | |
357 | outmessage.msg_control = outcmsg; | |
358 | outmessage.msg_controllen = sizeof(outcmsg); | |
359 | outmessage.msg_flags = MSG_EOR; | |
360 | ||
361 | cmsg = CMSG_FIRSTHDR(&outmessage); | |
362 | cmsg->cmsg_level = IPPROTO_SCTP; | |
363 | cmsg->cmsg_type = SCTP_SNDRCV; | |
364 | cmsg->cmsg_len = CMSG_LEN(sizeof(struct sctp_sndrcvinfo)); | |
365 | outmessage.msg_controllen = cmsg->cmsg_len; | |
366 | sinfo = CMSG_DATA(cmsg); | |
367 | memset(sinfo, 0x00, sizeof(struct sctp_sndrcvinfo)); | |
368 | ||
369 | sinfo->sinfo_flags |= MSG_EOF; | |
370 | sinfo->sinfo_assoc_id = associd; | |
371 | ||
372 | ret = kernel_sendmsg(con->sock, &outmessage, NULL, 0, 0); | |
373 | ||
374 | if (ret != 0) | |
375 | log_print("send EOF to node failed: %d", ret); | |
376 | } | |
377 | ||
378 | /* INIT failed but we don't know which node... | |
379 | restart INIT on all pending nodes */ | |
380 | static void sctp_init_failed(void) | |
381 | { | |
382 | int i; | |
383 | struct connection *con; | |
384 | ||
385 | down(&connections_lock); | |
386 | for (i=1; i<=max_nodeid; i++) { | |
387 | con = __nodeid2con(i, 0); | |
388 | if (!con) | |
389 | continue; | |
390 | con->sctp_assoc = 0; | |
391 | if (test_and_clear_bit(CF_CONNECT_PENDING, &con->flags)) { | |
392 | if (!test_and_set_bit(CF_WRITE_PENDING, &con->flags)) { | |
393 | queue_work(send_workqueue, &con->swork); | |
394 | } | |
395 | } | |
396 | } | |
397 | up(&connections_lock); | |
398 | } | |
399 | ||
400 | /* Something happened to an association */ | |
617e82e1 DT |
401 | static void process_sctp_notification(struct connection *con, |
402 | struct msghdr *msg, char *buf) | |
6ed7257b PC |
403 | { |
404 | union sctp_notification *sn = (union sctp_notification *)buf; | |
405 | ||
406 | if (sn->sn_header.sn_type == SCTP_ASSOC_CHANGE) { | |
407 | switch (sn->sn_assoc_change.sac_state) { | |
408 | ||
409 | case SCTP_COMM_UP: | |
410 | case SCTP_RESTART: | |
411 | { | |
412 | /* Check that the new node is in the lockspace */ | |
413 | struct sctp_prim prim; | |
414 | int nodeid; | |
415 | int prim_len, ret; | |
416 | int addr_len; | |
417 | struct connection *new_con; | |
418 | struct file *file; | |
419 | sctp_peeloff_arg_t parg; | |
420 | int parglen = sizeof(parg); | |
421 | ||
422 | /* | |
423 | * We get this before any data for an association. | |
424 | * We verify that the node is in the cluster and | |
425 | * then peel off a socket for it. | |
426 | */ | |
427 | if ((int)sn->sn_assoc_change.sac_assoc_id <= 0) { | |
428 | log_print("COMM_UP for invalid assoc ID %d", | |
617e82e1 | 429 | (int)sn->sn_assoc_change.sac_assoc_id); |
6ed7257b PC |
430 | sctp_init_failed(); |
431 | return; | |
432 | } | |
433 | memset(&prim, 0, sizeof(struct sctp_prim)); | |
434 | prim_len = sizeof(struct sctp_prim); | |
435 | prim.ssp_assoc_id = sn->sn_assoc_change.sac_assoc_id; | |
436 | ||
437 | ret = kernel_getsockopt(con->sock, | |
438 | IPPROTO_SCTP, | |
439 | SCTP_PRIMARY_ADDR, | |
440 | (char*)&prim, | |
441 | &prim_len); | |
442 | if (ret < 0) { | |
443 | log_print("getsockopt/sctp_primary_addr on " | |
444 | "new assoc %d failed : %d", | |
445 | (int)sn->sn_assoc_change.sac_assoc_id, | |
446 | ret); | |
447 | ||
448 | /* Retry INIT later */ | |
449 | new_con = assoc2con(sn->sn_assoc_change.sac_assoc_id); | |
450 | if (new_con) | |
451 | clear_bit(CF_CONNECT_PENDING, &con->flags); | |
452 | return; | |
453 | } | |
454 | make_sockaddr(&prim.ssp_addr, 0, &addr_len); | |
455 | if (dlm_addr_to_nodeid(&prim.ssp_addr, &nodeid)) { | |
456 | int i; | |
457 | unsigned char *b=(unsigned char *)&prim.ssp_addr; | |
458 | log_print("reject connect from unknown addr"); | |
459 | for (i=0; i<sizeof(struct sockaddr_storage);i++) | |
460 | printk("%02x ", b[i]); | |
461 | printk("\n"); | |
462 | sctp_send_shutdown(prim.ssp_assoc_id); | |
463 | return; | |
464 | } | |
465 | ||
466 | new_con = nodeid2con(nodeid, GFP_KERNEL); | |
467 | if (!new_con) | |
468 | return; | |
469 | ||
470 | /* Peel off a new sock */ | |
471 | parg.associd = sn->sn_assoc_change.sac_assoc_id; | |
617e82e1 DT |
472 | ret = kernel_getsockopt(con->sock, IPPROTO_SCTP, |
473 | SCTP_SOCKOPT_PEELOFF, | |
6ed7257b | 474 | (void *)&parg, &parglen); |
30d3a237 | 475 | if (ret) { |
617e82e1 DT |
476 | log_print("Can't peel off a socket for " |
477 | "connection %d to node %d: err=%d\n", | |
6ed7257b | 478 | parg.associd, nodeid, ret); |
6ed7257b | 479 | } |
6ed7257b PC |
480 | file = fget(parg.sd); |
481 | new_con->sock = SOCKET_I(file->f_dentry->d_inode); | |
482 | add_sock(new_con->sock, new_con); | |
483 | fput(file); | |
484 | put_unused_fd(parg.sd); | |
485 | ||
486 | log_print("got new/restarted association %d nodeid %d", | |
617e82e1 | 487 | (int)sn->sn_assoc_change.sac_assoc_id, nodeid); |
6ed7257b PC |
488 | |
489 | /* Send any pending writes */ | |
490 | clear_bit(CF_CONNECT_PENDING, &new_con->flags); | |
491 | clear_bit(CF_INIT_PENDING, &con->flags); | |
492 | if (!test_and_set_bit(CF_WRITE_PENDING, &new_con->flags)) { | |
493 | queue_work(send_workqueue, &new_con->swork); | |
494 | } | |
495 | if (!test_and_set_bit(CF_READ_PENDING, &new_con->flags)) | |
496 | queue_work(recv_workqueue, &new_con->rwork); | |
497 | } | |
498 | break; | |
499 | ||
500 | case SCTP_COMM_LOST: | |
501 | case SCTP_SHUTDOWN_COMP: | |
502 | { | |
503 | con = assoc2con(sn->sn_assoc_change.sac_assoc_id); | |
504 | if (con) { | |
505 | con->sctp_assoc = 0; | |
506 | } | |
507 | } | |
508 | break; | |
509 | ||
510 | /* We don't know which INIT failed, so clear the PENDING flags | |
511 | * on them all. if assoc_id is zero then it will then try | |
512 | * again */ | |
513 | ||
514 | case SCTP_CANT_STR_ASSOC: | |
515 | { | |
516 | log_print("Can't start SCTP association - retrying"); | |
517 | sctp_init_failed(); | |
518 | } | |
519 | break; | |
520 | ||
521 | default: | |
522 | log_print("unexpected SCTP assoc change id=%d state=%d", | |
523 | (int)sn->sn_assoc_change.sac_assoc_id, | |
524 | sn->sn_assoc_change.sac_state); | |
525 | } | |
526 | } | |
527 | } | |
528 | ||
fdda387f PC |
529 | /* Data received from remote end */ |
530 | static int receive_from_sock(struct connection *con) | |
531 | { | |
532 | int ret = 0; | |
58addbff AV |
533 | struct msghdr msg = {}; |
534 | struct kvec iov[2]; | |
fdda387f PC |
535 | unsigned len; |
536 | int r; | |
537 | int call_again_soon = 0; | |
58addbff | 538 | int nvec; |
6ed7257b | 539 | char incmsg[CMSG_SPACE(sizeof(struct sctp_sndrcvinfo))]; |
fdda387f | 540 | |
f1f1c1cc | 541 | mutex_lock(&con->sock_mutex); |
fdda387f | 542 | |
a34fbc63 PC |
543 | if (con->sock == NULL) { |
544 | ret = -EAGAIN; | |
545 | goto out_close; | |
546 | } | |
547 | ||
fdda387f PC |
548 | if (con->rx_page == NULL) { |
549 | /* | |
550 | * This doesn't need to be atomic, but I think it should | |
551 | * improve performance if it is. | |
552 | */ | |
553 | con->rx_page = alloc_page(GFP_ATOMIC); | |
554 | if (con->rx_page == NULL) | |
555 | goto out_resched; | |
ac33d071 | 556 | cbuf_init(&con->cb, PAGE_CACHE_SIZE); |
fdda387f PC |
557 | } |
558 | ||
6ed7257b PC |
559 | /* Only SCTP needs these really */ |
560 | memset(&incmsg, 0, sizeof(incmsg)); | |
561 | msg.msg_control = incmsg; | |
562 | msg.msg_controllen = sizeof(incmsg); | |
563 | ||
fdda387f PC |
564 | /* |
565 | * iov[0] is the bit of the circular buffer between the current end | |
566 | * point (cb.base + cb.len) and the end of the buffer. | |
567 | */ | |
ac33d071 PC |
568 | iov[0].iov_len = con->cb.base - cbuf_data(&con->cb); |
569 | iov[0].iov_base = page_address(con->rx_page) + cbuf_data(&con->cb); | |
89adc934 | 570 | iov[1].iov_len = 0; |
58addbff | 571 | nvec = 1; |
fdda387f PC |
572 | |
573 | /* | |
574 | * iov[1] is the bit of the circular buffer between the start of the | |
575 | * buffer and the start of the currently used section (cb.base) | |
576 | */ | |
ac33d071 PC |
577 | if (cbuf_data(&con->cb) >= con->cb.base) { |
578 | iov[0].iov_len = PAGE_CACHE_SIZE - cbuf_data(&con->cb); | |
fdda387f PC |
579 | iov[1].iov_len = con->cb.base; |
580 | iov[1].iov_base = page_address(con->rx_page); | |
58addbff | 581 | nvec = 2; |
fdda387f PC |
582 | } |
583 | len = iov[0].iov_len + iov[1].iov_len; | |
584 | ||
58addbff | 585 | r = ret = kernel_recvmsg(con->sock, &msg, iov, nvec, len, |
fdda387f | 586 | MSG_DONTWAIT | MSG_NOSIGNAL); |
fdda387f PC |
587 | if (ret <= 0) |
588 | goto out_close; | |
bd44e2b0 | 589 | |
6ed7257b PC |
590 | /* Process SCTP notifications */ |
591 | if (msg.msg_flags & MSG_NOTIFICATION) { | |
6ed7257b PC |
592 | msg.msg_control = incmsg; |
593 | msg.msg_controllen = sizeof(incmsg); | |
594 | ||
595 | process_sctp_notification(con, &msg, | |
617e82e1 | 596 | page_address(con->rx_page) + con->cb.base); |
6ed7257b PC |
597 | mutex_unlock(&con->sock_mutex); |
598 | return 0; | |
599 | } | |
600 | BUG_ON(con->nodeid == 0); | |
601 | ||
fdda387f PC |
602 | if (ret == len) |
603 | call_again_soon = 1; | |
ac33d071 | 604 | cbuf_add(&con->cb, ret); |
fdda387f PC |
605 | ret = dlm_process_incoming_buffer(con->nodeid, |
606 | page_address(con->rx_page), | |
607 | con->cb.base, con->cb.len, | |
608 | PAGE_CACHE_SIZE); | |
609 | if (ret == -EBADMSG) { | |
617e82e1 DT |
610 | log_print("lowcomms: addr=%p, base=%u, len=%u, " |
611 | "iov_len=%u, iov_base[0]=%p, read=%d", | |
612 | page_address(con->rx_page), con->cb.base, con->cb.len, | |
613 | len, iov[0].iov_base, r); | |
fdda387f PC |
614 | } |
615 | if (ret < 0) | |
616 | goto out_close; | |
ac33d071 | 617 | cbuf_eat(&con->cb, ret); |
fdda387f | 618 | |
ac33d071 | 619 | if (cbuf_empty(&con->cb) && !call_again_soon) { |
fdda387f PC |
620 | __free_page(con->rx_page); |
621 | con->rx_page = NULL; | |
622 | } | |
623 | ||
fdda387f PC |
624 | if (call_again_soon) |
625 | goto out_resched; | |
f1f1c1cc | 626 | mutex_unlock(&con->sock_mutex); |
ac33d071 | 627 | return 0; |
fdda387f | 628 | |
ac33d071 | 629 | out_resched: |
1d6e8131 PC |
630 | if (!test_and_set_bit(CF_READ_PENDING, &con->flags)) |
631 | queue_work(recv_workqueue, &con->rwork); | |
f1f1c1cc | 632 | mutex_unlock(&con->sock_mutex); |
bd44e2b0 | 633 | return -EAGAIN; |
fdda387f | 634 | |
ac33d071 | 635 | out_close: |
f1f1c1cc | 636 | mutex_unlock(&con->sock_mutex); |
fdda387f | 637 | if (ret != -EAGAIN && !test_bit(CF_IS_OTHERCON, &con->flags)) { |
ac33d071 | 638 | close_connection(con, false); |
fdda387f PC |
639 | /* Reconnect when there is something to send */ |
640 | } | |
a34fbc63 PC |
641 | /* Don't return success if we really got EOF */ |
642 | if (ret == 0) | |
643 | ret = -EAGAIN; | |
fdda387f | 644 | |
fdda387f PC |
645 | return ret; |
646 | } | |
647 | ||
648 | /* Listening socket is busy, accept a connection */ | |
6ed7257b | 649 | static int tcp_accept_from_sock(struct connection *con) |
fdda387f PC |
650 | { |
651 | int result; | |
652 | struct sockaddr_storage peeraddr; | |
653 | struct socket *newsock; | |
654 | int len; | |
655 | int nodeid; | |
656 | struct connection *newcon; | |
bd44e2b0 | 657 | struct connection *addcon; |
fdda387f PC |
658 | |
659 | memset(&peeraddr, 0, sizeof(peeraddr)); | |
6ed7257b | 660 | result = sock_create_kern(dlm_local_addr[0]->ss_family, SOCK_STREAM, |
ac33d071 | 661 | IPPROTO_TCP, &newsock); |
fdda387f PC |
662 | if (result < 0) |
663 | return -ENOMEM; | |
664 | ||
f1f1c1cc | 665 | mutex_lock_nested(&con->sock_mutex, 0); |
fdda387f PC |
666 | |
667 | result = -ENOTCONN; | |
668 | if (con->sock == NULL) | |
669 | goto accept_err; | |
670 | ||
671 | newsock->type = con->sock->type; | |
672 | newsock->ops = con->sock->ops; | |
673 | ||
674 | result = con->sock->ops->accept(con->sock, newsock, O_NONBLOCK); | |
675 | if (result < 0) | |
676 | goto accept_err; | |
677 | ||
678 | /* Get the connected socket's peer */ | |
679 | memset(&peeraddr, 0, sizeof(peeraddr)); | |
680 | if (newsock->ops->getname(newsock, (struct sockaddr *)&peeraddr, | |
681 | &len, 2)) { | |
682 | result = -ECONNABORTED; | |
683 | goto accept_err; | |
684 | } | |
685 | ||
686 | /* Get the new node's NODEID */ | |
687 | make_sockaddr(&peeraddr, 0, &len); | |
688 | if (dlm_addr_to_nodeid(&peeraddr, &nodeid)) { | |
617e82e1 | 689 | log_print("connect from non cluster node"); |
fdda387f | 690 | sock_release(newsock); |
f1f1c1cc | 691 | mutex_unlock(&con->sock_mutex); |
fdda387f PC |
692 | return -1; |
693 | } | |
694 | ||
695 | log_print("got connection from %d", nodeid); | |
696 | ||
697 | /* Check to see if we already have a connection to this node. This | |
698 | * could happen if the two nodes initiate a connection at roughly | |
699 | * the same time and the connections cross on the wire. | |
fdda387f PC |
700 | * In this case we store the incoming one in "othercon" |
701 | */ | |
702 | newcon = nodeid2con(nodeid, GFP_KERNEL); | |
703 | if (!newcon) { | |
704 | result = -ENOMEM; | |
705 | goto accept_err; | |
706 | } | |
f1f1c1cc | 707 | mutex_lock_nested(&newcon->sock_mutex, 1); |
fdda387f | 708 | if (newcon->sock) { |
ac33d071 | 709 | struct connection *othercon = newcon->othercon; |
fdda387f PC |
710 | |
711 | if (!othercon) { | |
ac33d071 | 712 | othercon = kmem_cache_zalloc(con_cache, GFP_KERNEL); |
fdda387f | 713 | if (!othercon) { |
617e82e1 | 714 | log_print("failed to allocate incoming socket"); |
f1f1c1cc | 715 | mutex_unlock(&newcon->sock_mutex); |
fdda387f PC |
716 | result = -ENOMEM; |
717 | goto accept_err; | |
718 | } | |
fdda387f PC |
719 | othercon->nodeid = nodeid; |
720 | othercon->rx_action = receive_from_sock; | |
f1f1c1cc | 721 | mutex_init(&othercon->sock_mutex); |
1d6e8131 PC |
722 | INIT_WORK(&othercon->swork, process_send_sockets); |
723 | INIT_WORK(&othercon->rwork, process_recv_sockets); | |
fdda387f PC |
724 | set_bit(CF_IS_OTHERCON, &othercon->flags); |
725 | newcon->othercon = othercon; | |
97d84836 PC |
726 | othercon->sock = newsock; |
727 | newsock->sk->sk_user_data = othercon; | |
728 | add_sock(newsock, othercon); | |
729 | addcon = othercon; | |
730 | } | |
731 | else { | |
732 | printk("Extra connection from node %d attempted\n", nodeid); | |
733 | result = -EAGAIN; | |
f4fadb23 | 734 | mutex_unlock(&newcon->sock_mutex); |
97d84836 | 735 | goto accept_err; |
fdda387f | 736 | } |
fdda387f PC |
737 | } |
738 | else { | |
739 | newsock->sk->sk_user_data = newcon; | |
740 | newcon->rx_action = receive_from_sock; | |
741 | add_sock(newsock, newcon); | |
bd44e2b0 | 742 | addcon = newcon; |
fdda387f PC |
743 | } |
744 | ||
f1f1c1cc | 745 | mutex_unlock(&newcon->sock_mutex); |
fdda387f PC |
746 | |
747 | /* | |
748 | * Add it to the active queue in case we got data | |
749 | * beween processing the accept adding the socket | |
750 | * to the read_sockets list | |
751 | */ | |
bd44e2b0 PC |
752 | if (!test_and_set_bit(CF_READ_PENDING, &addcon->flags)) |
753 | queue_work(recv_workqueue, &addcon->rwork); | |
f1f1c1cc | 754 | mutex_unlock(&con->sock_mutex); |
fdda387f PC |
755 | |
756 | return 0; | |
757 | ||
ac33d071 | 758 | accept_err: |
f1f1c1cc | 759 | mutex_unlock(&con->sock_mutex); |
fdda387f PC |
760 | sock_release(newsock); |
761 | ||
762 | if (result != -EAGAIN) | |
617e82e1 | 763 | log_print("error accepting connection from node: %d", result); |
fdda387f PC |
764 | return result; |
765 | } | |
766 | ||
6ed7257b PC |
767 | static void free_entry(struct writequeue_entry *e) |
768 | { | |
769 | __free_page(e->page); | |
770 | kfree(e); | |
771 | } | |
772 | ||
773 | /* Initiate an SCTP association. | |
774 | This is a special case of send_to_sock() in that we don't yet have a | |
775 | peeled-off socket for this association, so we use the listening socket | |
776 | and add the primary IP address of the remote node. | |
777 | */ | |
778 | static void sctp_init_assoc(struct connection *con) | |
779 | { | |
780 | struct sockaddr_storage rem_addr; | |
781 | char outcmsg[CMSG_SPACE(sizeof(struct sctp_sndrcvinfo))]; | |
782 | struct msghdr outmessage; | |
783 | struct cmsghdr *cmsg; | |
784 | struct sctp_sndrcvinfo *sinfo; | |
785 | struct connection *base_con; | |
786 | struct writequeue_entry *e; | |
787 | int len, offset; | |
788 | int ret; | |
789 | int addrlen; | |
790 | struct kvec iov[1]; | |
791 | ||
792 | if (test_and_set_bit(CF_INIT_PENDING, &con->flags)) | |
793 | return; | |
794 | ||
795 | if (con->retries++ > MAX_CONNECT_RETRIES) | |
796 | return; | |
797 | ||
798 | log_print("Initiating association with node %d", con->nodeid); | |
799 | ||
800 | if (nodeid_to_addr(con->nodeid, (struct sockaddr *)&rem_addr)) { | |
801 | log_print("no address for nodeid %d", con->nodeid); | |
802 | return; | |
803 | } | |
804 | base_con = nodeid2con(0, 0); | |
805 | BUG_ON(base_con == NULL); | |
806 | ||
807 | make_sockaddr(&rem_addr, dlm_config.ci_tcp_port, &addrlen); | |
808 | ||
809 | outmessage.msg_name = &rem_addr; | |
810 | outmessage.msg_namelen = addrlen; | |
811 | outmessage.msg_control = outcmsg; | |
812 | outmessage.msg_controllen = sizeof(outcmsg); | |
813 | outmessage.msg_flags = MSG_EOR; | |
814 | ||
815 | spin_lock(&con->writequeue_lock); | |
816 | e = list_entry(con->writequeue.next, struct writequeue_entry, | |
817 | list); | |
818 | ||
819 | BUG_ON((struct list_head *) e == &con->writequeue); | |
820 | ||
821 | len = e->len; | |
822 | offset = e->offset; | |
823 | spin_unlock(&con->writequeue_lock); | |
824 | kmap(e->page); | |
825 | ||
826 | /* Send the first block off the write queue */ | |
827 | iov[0].iov_base = page_address(e->page)+offset; | |
828 | iov[0].iov_len = len; | |
829 | ||
830 | cmsg = CMSG_FIRSTHDR(&outmessage); | |
831 | cmsg->cmsg_level = IPPROTO_SCTP; | |
832 | cmsg->cmsg_type = SCTP_SNDRCV; | |
833 | cmsg->cmsg_len = CMSG_LEN(sizeof(struct sctp_sndrcvinfo)); | |
834 | sinfo = CMSG_DATA(cmsg); | |
835 | memset(sinfo, 0x00, sizeof(struct sctp_sndrcvinfo)); | |
836 | sinfo->sinfo_ppid = cpu_to_le32(dlm_our_nodeid()); | |
837 | outmessage.msg_controllen = cmsg->cmsg_len; | |
838 | ||
839 | ret = kernel_sendmsg(base_con->sock, &outmessage, iov, 1, len); | |
840 | if (ret < 0) { | |
617e82e1 DT |
841 | log_print("Send first packet to node %d failed: %d", |
842 | con->nodeid, ret); | |
6ed7257b PC |
843 | |
844 | /* Try again later */ | |
845 | clear_bit(CF_CONNECT_PENDING, &con->flags); | |
846 | clear_bit(CF_INIT_PENDING, &con->flags); | |
847 | } | |
848 | else { | |
849 | spin_lock(&con->writequeue_lock); | |
850 | e->offset += ret; | |
851 | e->len -= ret; | |
852 | ||
853 | if (e->len == 0 && e->users == 0) { | |
854 | list_del(&e->list); | |
855 | kunmap(e->page); | |
856 | free_entry(e); | |
857 | } | |
858 | spin_unlock(&con->writequeue_lock); | |
859 | } | |
860 | } | |
861 | ||
fdda387f | 862 | /* Connect a new socket to its peer */ |
6ed7257b | 863 | static void tcp_connect_to_sock(struct connection *con) |
fdda387f PC |
864 | { |
865 | int result = -EHOSTUNREACH; | |
866 | struct sockaddr_storage saddr; | |
867 | int addr_len; | |
868 | struct socket *sock; | |
869 | ||
870 | if (con->nodeid == 0) { | |
871 | log_print("attempt to connect sock 0 foiled"); | |
ac33d071 | 872 | return; |
fdda387f PC |
873 | } |
874 | ||
f1f1c1cc | 875 | mutex_lock(&con->sock_mutex); |
fdda387f PC |
876 | if (con->retries++ > MAX_CONNECT_RETRIES) |
877 | goto out; | |
878 | ||
879 | /* Some odd races can cause double-connects, ignore them */ | |
880 | if (con->sock) { | |
881 | result = 0; | |
882 | goto out; | |
883 | } | |
884 | ||
885 | /* Create a socket to communicate with */ | |
6ed7257b | 886 | result = sock_create_kern(dlm_local_addr[0]->ss_family, SOCK_STREAM, |
ac33d071 | 887 | IPPROTO_TCP, &sock); |
fdda387f PC |
888 | if (result < 0) |
889 | goto out_err; | |
890 | ||
891 | memset(&saddr, 0, sizeof(saddr)); | |
892 | if (dlm_nodeid_to_addr(con->nodeid, &saddr)) | |
ac33d071 | 893 | goto out_err; |
fdda387f PC |
894 | |
895 | sock->sk->sk_user_data = con; | |
896 | con->rx_action = receive_from_sock; | |
6ed7257b PC |
897 | con->connect_action = tcp_connect_to_sock; |
898 | add_sock(sock, con); | |
fdda387f | 899 | |
68c817a1 | 900 | make_sockaddr(&saddr, dlm_config.ci_tcp_port, &addr_len); |
fdda387f | 901 | |
fdda387f PC |
902 | log_print("connecting to %d", con->nodeid); |
903 | result = | |
904 | sock->ops->connect(sock, (struct sockaddr *)&saddr, addr_len, | |
ac33d071 | 905 | O_NONBLOCK); |
fdda387f PC |
906 | if (result == -EINPROGRESS) |
907 | result = 0; | |
ac33d071 PC |
908 | if (result == 0) |
909 | goto out; | |
fdda387f | 910 | |
ac33d071 | 911 | out_err: |
fdda387f PC |
912 | if (con->sock) { |
913 | sock_release(con->sock); | |
914 | con->sock = NULL; | |
915 | } | |
916 | /* | |
917 | * Some errors are fatal and this list might need adjusting. For other | |
918 | * errors we try again until the max number of retries is reached. | |
919 | */ | |
920 | if (result != -EHOSTUNREACH && result != -ENETUNREACH && | |
921 | result != -ENETDOWN && result != EINVAL | |
922 | && result != -EPROTONOSUPPORT) { | |
923 | lowcomms_connect_sock(con); | |
924 | result = 0; | |
925 | } | |
ac33d071 | 926 | out: |
f1f1c1cc | 927 | mutex_unlock(&con->sock_mutex); |
ac33d071 | 928 | return; |
fdda387f PC |
929 | } |
930 | ||
6ed7257b PC |
931 | static struct socket *tcp_create_listen_sock(struct connection *con, |
932 | struct sockaddr_storage *saddr) | |
fdda387f | 933 | { |
ac33d071 | 934 | struct socket *sock = NULL; |
fdda387f PC |
935 | int result = 0; |
936 | int one = 1; | |
937 | int addr_len; | |
938 | ||
6ed7257b | 939 | if (dlm_local_addr[0]->ss_family == AF_INET) |
fdda387f PC |
940 | addr_len = sizeof(struct sockaddr_in); |
941 | else | |
942 | addr_len = sizeof(struct sockaddr_in6); | |
943 | ||
944 | /* Create a socket to communicate with */ | |
617e82e1 DT |
945 | result = sock_create_kern(dlm_local_addr[0]->ss_family, SOCK_STREAM, |
946 | IPPROTO_TCP, &sock); | |
fdda387f | 947 | if (result < 0) { |
617e82e1 | 948 | log_print("Can't create listening comms socket"); |
fdda387f PC |
949 | goto create_out; |
950 | } | |
951 | ||
6ed7257b PC |
952 | result = kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEADDR, |
953 | (char *)&one, sizeof(one)); | |
954 | ||
fdda387f | 955 | if (result < 0) { |
617e82e1 | 956 | log_print("Failed to set SO_REUSEADDR on socket: %d", result); |
fdda387f PC |
957 | } |
958 | sock->sk->sk_user_data = con; | |
6ed7257b PC |
959 | con->rx_action = tcp_accept_from_sock; |
960 | con->connect_action = tcp_connect_to_sock; | |
fdda387f PC |
961 | con->sock = sock; |
962 | ||
963 | /* Bind to our port */ | |
68c817a1 | 964 | make_sockaddr(saddr, dlm_config.ci_tcp_port, &addr_len); |
fdda387f PC |
965 | result = sock->ops->bind(sock, (struct sockaddr *) saddr, addr_len); |
966 | if (result < 0) { | |
617e82e1 | 967 | log_print("Can't bind to port %d", dlm_config.ci_tcp_port); |
fdda387f PC |
968 | sock_release(sock); |
969 | sock = NULL; | |
970 | con->sock = NULL; | |
971 | goto create_out; | |
972 | } | |
6ed7257b | 973 | result = kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE, |
ac33d071 | 974 | (char *)&one, sizeof(one)); |
fdda387f | 975 | if (result < 0) { |
617e82e1 | 976 | log_print("Set keepalive failed: %d", result); |
fdda387f PC |
977 | } |
978 | ||
979 | result = sock->ops->listen(sock, 5); | |
980 | if (result < 0) { | |
617e82e1 | 981 | log_print("Can't listen on port %d", dlm_config.ci_tcp_port); |
fdda387f PC |
982 | sock_release(sock); |
983 | sock = NULL; | |
984 | goto create_out; | |
985 | } | |
986 | ||
ac33d071 | 987 | create_out: |
fdda387f PC |
988 | return sock; |
989 | } | |
990 | ||
6ed7257b PC |
991 | /* Get local addresses */ |
992 | static void init_local(void) | |
993 | { | |
994 | struct sockaddr_storage sas, *addr; | |
995 | int i; | |
996 | ||
30d3a237 | 997 | dlm_local_count = 0; |
6ed7257b PC |
998 | for (i = 0; i < DLM_MAX_ADDR_COUNT - 1; i++) { |
999 | if (dlm_our_addr(&sas, i)) | |
1000 | break; | |
1001 | ||
1002 | addr = kmalloc(sizeof(*addr), GFP_KERNEL); | |
1003 | if (!addr) | |
1004 | break; | |
1005 | memcpy(addr, &sas, sizeof(*addr)); | |
1006 | dlm_local_addr[dlm_local_count++] = addr; | |
1007 | } | |
1008 | } | |
1009 | ||
617e82e1 DT |
1010 | /* Bind to an IP address. SCTP allows multiple address so it can do |
1011 | multi-homing */ | |
1012 | static int add_sctp_bind_addr(struct connection *sctp_con, | |
1013 | struct sockaddr_storage *addr, | |
1014 | int addr_len, int num) | |
6ed7257b PC |
1015 | { |
1016 | int result = 0; | |
1017 | ||
1018 | if (num == 1) | |
1019 | result = kernel_bind(sctp_con->sock, | |
1020 | (struct sockaddr *) addr, | |
1021 | addr_len); | |
1022 | else | |
1023 | result = kernel_setsockopt(sctp_con->sock, SOL_SCTP, | |
1024 | SCTP_SOCKOPT_BINDX_ADD, | |
1025 | (char *)addr, addr_len); | |
1026 | ||
1027 | if (result < 0) | |
1028 | log_print("Can't bind to port %d addr number %d", | |
1029 | dlm_config.ci_tcp_port, num); | |
1030 | ||
1031 | return result; | |
1032 | } | |
fdda387f | 1033 | |
6ed7257b PC |
1034 | /* Initialise SCTP socket and bind to all interfaces */ |
1035 | static int sctp_listen_for_all(void) | |
1036 | { | |
1037 | struct socket *sock = NULL; | |
1038 | struct sockaddr_storage localaddr; | |
1039 | struct sctp_event_subscribe subscribe; | |
1040 | int result = -EINVAL, num = 1, i, addr_len; | |
1041 | struct connection *con = nodeid2con(0, GFP_KERNEL); | |
1042 | int bufsize = NEEDED_RMEM; | |
1043 | ||
1044 | if (!con) | |
1045 | return -ENOMEM; | |
1046 | ||
1047 | log_print("Using SCTP for communications"); | |
1048 | ||
1049 | result = sock_create_kern(dlm_local_addr[0]->ss_family, SOCK_SEQPACKET, | |
1050 | IPPROTO_SCTP, &sock); | |
1051 | if (result < 0) { | |
1052 | log_print("Can't create comms socket, check SCTP is loaded"); | |
1053 | goto out; | |
1054 | } | |
1055 | ||
1056 | /* Listen for events */ | |
1057 | memset(&subscribe, 0, sizeof(subscribe)); | |
1058 | subscribe.sctp_data_io_event = 1; | |
1059 | subscribe.sctp_association_event = 1; | |
1060 | subscribe.sctp_send_failure_event = 1; | |
1061 | subscribe.sctp_shutdown_event = 1; | |
1062 | subscribe.sctp_partial_delivery_event = 1; | |
1063 | ||
1064 | result = kernel_setsockopt(sock, SOL_SOCKET, SO_RCVBUF, | |
1065 | (char *)&bufsize, sizeof(bufsize)); | |
1066 | if (result) | |
617e82e1 | 1067 | log_print("Error increasing buffer space on socket %d", result); |
6ed7257b PC |
1068 | |
1069 | result = kernel_setsockopt(sock, SOL_SCTP, SCTP_EVENTS, | |
617e82e1 | 1070 | (char *)&subscribe, sizeof(subscribe)); |
6ed7257b PC |
1071 | if (result < 0) { |
1072 | log_print("Failed to set SCTP_EVENTS on socket: result=%d", | |
1073 | result); | |
1074 | goto create_delsock; | |
1075 | } | |
1076 | ||
1077 | /* Init con struct */ | |
1078 | sock->sk->sk_user_data = con; | |
1079 | con->sock = sock; | |
1080 | con->sock->sk->sk_data_ready = lowcomms_data_ready; | |
1081 | con->rx_action = receive_from_sock; | |
1082 | con->connect_action = sctp_init_assoc; | |
1083 | ||
1084 | /* Bind to all interfaces. */ | |
1085 | for (i = 0; i < dlm_local_count; i++) { | |
1086 | memcpy(&localaddr, dlm_local_addr[i], sizeof(localaddr)); | |
1087 | make_sockaddr(&localaddr, dlm_config.ci_tcp_port, &addr_len); | |
1088 | ||
1089 | result = add_sctp_bind_addr(con, &localaddr, addr_len, num); | |
1090 | if (result) | |
1091 | goto create_delsock; | |
1092 | ++num; | |
1093 | } | |
1094 | ||
1095 | result = sock->ops->listen(sock, 5); | |
1096 | if (result < 0) { | |
1097 | log_print("Can't set socket listening"); | |
1098 | goto create_delsock; | |
1099 | } | |
1100 | ||
1101 | return 0; | |
1102 | ||
1103 | create_delsock: | |
1104 | sock_release(sock); | |
1105 | con->sock = NULL; | |
1106 | out: | |
1107 | return result; | |
1108 | } | |
1109 | ||
1110 | static int tcp_listen_for_all(void) | |
fdda387f PC |
1111 | { |
1112 | struct socket *sock = NULL; | |
1113 | struct connection *con = nodeid2con(0, GFP_KERNEL); | |
1114 | int result = -EINVAL; | |
1115 | ||
6ed7257b PC |
1116 | if (!con) |
1117 | return -ENOMEM; | |
1118 | ||
fdda387f | 1119 | /* We don't support multi-homed hosts */ |
6ed7257b | 1120 | if (dlm_local_addr[1] != NULL) { |
617e82e1 DT |
1121 | log_print("TCP protocol can't handle multi-homed hosts, " |
1122 | "try SCTP"); | |
6ed7257b PC |
1123 | return -EINVAL; |
1124 | } | |
1125 | ||
1126 | log_print("Using TCP for communications"); | |
1127 | ||
fdda387f PC |
1128 | set_bit(CF_IS_OTHERCON, &con->flags); |
1129 | ||
6ed7257b | 1130 | sock = tcp_create_listen_sock(con, dlm_local_addr[0]); |
fdda387f PC |
1131 | if (sock) { |
1132 | add_sock(sock, con); | |
1133 | result = 0; | |
1134 | } | |
1135 | else { | |
1136 | result = -EADDRINUSE; | |
1137 | } | |
1138 | ||
1139 | return result; | |
1140 | } | |
1141 | ||
1142 | ||
1143 | ||
1144 | static struct writequeue_entry *new_writequeue_entry(struct connection *con, | |
1145 | gfp_t allocation) | |
1146 | { | |
1147 | struct writequeue_entry *entry; | |
1148 | ||
1149 | entry = kmalloc(sizeof(struct writequeue_entry), allocation); | |
1150 | if (!entry) | |
1151 | return NULL; | |
1152 | ||
1153 | entry->page = alloc_page(allocation); | |
1154 | if (!entry->page) { | |
1155 | kfree(entry); | |
1156 | return NULL; | |
1157 | } | |
1158 | ||
1159 | entry->offset = 0; | |
1160 | entry->len = 0; | |
1161 | entry->end = 0; | |
1162 | entry->users = 0; | |
1163 | entry->con = con; | |
1164 | ||
1165 | return entry; | |
1166 | } | |
1167 | ||
617e82e1 | 1168 | void *dlm_lowcomms_get_buffer(int nodeid, int len, gfp_t allocation, char **ppc) |
fdda387f PC |
1169 | { |
1170 | struct connection *con; | |
1171 | struct writequeue_entry *e; | |
1172 | int offset = 0; | |
1173 | int users = 0; | |
1174 | ||
fdda387f PC |
1175 | con = nodeid2con(nodeid, allocation); |
1176 | if (!con) | |
1177 | return NULL; | |
1178 | ||
4edde74e | 1179 | spin_lock(&con->writequeue_lock); |
fdda387f | 1180 | e = list_entry(con->writequeue.prev, struct writequeue_entry, list); |
ac33d071 | 1181 | if ((&e->list == &con->writequeue) || |
fdda387f PC |
1182 | (PAGE_CACHE_SIZE - e->end < len)) { |
1183 | e = NULL; | |
1184 | } else { | |
1185 | offset = e->end; | |
1186 | e->end += len; | |
1187 | users = e->users++; | |
1188 | } | |
1189 | spin_unlock(&con->writequeue_lock); | |
1190 | ||
1191 | if (e) { | |
ac33d071 | 1192 | got_one: |
fdda387f PC |
1193 | if (users == 0) |
1194 | kmap(e->page); | |
1195 | *ppc = page_address(e->page) + offset; | |
1196 | return e; | |
1197 | } | |
1198 | ||
1199 | e = new_writequeue_entry(con, allocation); | |
1200 | if (e) { | |
1201 | spin_lock(&con->writequeue_lock); | |
1202 | offset = e->end; | |
1203 | e->end += len; | |
1204 | users = e->users++; | |
1205 | list_add_tail(&e->list, &con->writequeue); | |
1206 | spin_unlock(&con->writequeue_lock); | |
1207 | goto got_one; | |
1208 | } | |
1209 | return NULL; | |
1210 | } | |
1211 | ||
1212 | void dlm_lowcomms_commit_buffer(void *mh) | |
1213 | { | |
1214 | struct writequeue_entry *e = (struct writequeue_entry *)mh; | |
1215 | struct connection *con = e->con; | |
1216 | int users; | |
1217 | ||
4edde74e | 1218 | spin_lock(&con->writequeue_lock); |
fdda387f PC |
1219 | users = --e->users; |
1220 | if (users) | |
1221 | goto out; | |
1222 | e->len = e->end - e->offset; | |
1223 | kunmap(e->page); | |
1224 | spin_unlock(&con->writequeue_lock); | |
1225 | ||
1d6e8131 PC |
1226 | if (!test_and_set_bit(CF_WRITE_PENDING, &con->flags)) { |
1227 | queue_work(send_workqueue, &con->swork); | |
fdda387f PC |
1228 | } |
1229 | return; | |
1230 | ||
ac33d071 | 1231 | out: |
fdda387f PC |
1232 | spin_unlock(&con->writequeue_lock); |
1233 | return; | |
1234 | } | |
1235 | ||
fdda387f | 1236 | /* Send a message */ |
ac33d071 | 1237 | static void send_to_sock(struct connection *con) |
fdda387f PC |
1238 | { |
1239 | int ret = 0; | |
1240 | ssize_t(*sendpage) (struct socket *, struct page *, int, size_t, int); | |
1241 | const int msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL; | |
1242 | struct writequeue_entry *e; | |
1243 | int len, offset; | |
1244 | ||
f1f1c1cc | 1245 | mutex_lock(&con->sock_mutex); |
fdda387f PC |
1246 | if (con->sock == NULL) |
1247 | goto out_connect; | |
1248 | ||
1249 | sendpage = con->sock->ops->sendpage; | |
1250 | ||
1251 | spin_lock(&con->writequeue_lock); | |
1252 | for (;;) { | |
1253 | e = list_entry(con->writequeue.next, struct writequeue_entry, | |
1254 | list); | |
1255 | if ((struct list_head *) e == &con->writequeue) | |
1256 | break; | |
1257 | ||
1258 | len = e->len; | |
1259 | offset = e->offset; | |
1260 | BUG_ON(len == 0 && e->users == 0); | |
1261 | spin_unlock(&con->writequeue_lock); | |
1d6e8131 | 1262 | kmap(e->page); |
fdda387f PC |
1263 | |
1264 | ret = 0; | |
1265 | if (len) { | |
1266 | ret = sendpage(con->sock, e->page, offset, len, | |
1267 | msg_flags); | |
1268 | if (ret == -EAGAIN || ret == 0) | |
1269 | goto out; | |
1270 | if (ret <= 0) | |
1271 | goto send_error; | |
617e82e1 | 1272 | } else { |
fdda387f | 1273 | /* Don't starve people filling buffers */ |
ac33d071 | 1274 | cond_resched(); |
fdda387f PC |
1275 | } |
1276 | ||
1277 | spin_lock(&con->writequeue_lock); | |
1278 | e->offset += ret; | |
1279 | e->len -= ret; | |
1280 | ||
1281 | if (e->len == 0 && e->users == 0) { | |
1282 | list_del(&e->list); | |
ac33d071 | 1283 | kunmap(e->page); |
fdda387f PC |
1284 | free_entry(e); |
1285 | continue; | |
1286 | } | |
1287 | } | |
1288 | spin_unlock(&con->writequeue_lock); | |
ac33d071 | 1289 | out: |
f1f1c1cc | 1290 | mutex_unlock(&con->sock_mutex); |
ac33d071 | 1291 | return; |
fdda387f | 1292 | |
ac33d071 | 1293 | send_error: |
f1f1c1cc | 1294 | mutex_unlock(&con->sock_mutex); |
ac33d071 | 1295 | close_connection(con, false); |
fdda387f | 1296 | lowcomms_connect_sock(con); |
ac33d071 | 1297 | return; |
fdda387f | 1298 | |
ac33d071 | 1299 | out_connect: |
f1f1c1cc | 1300 | mutex_unlock(&con->sock_mutex); |
6ed7257b PC |
1301 | if (!test_bit(CF_INIT_PENDING, &con->flags)) |
1302 | lowcomms_connect_sock(con); | |
ac33d071 | 1303 | return; |
fdda387f PC |
1304 | } |
1305 | ||
1306 | static void clean_one_writequeue(struct connection *con) | |
1307 | { | |
1308 | struct list_head *list; | |
1309 | struct list_head *temp; | |
1310 | ||
1311 | spin_lock(&con->writequeue_lock); | |
1312 | list_for_each_safe(list, temp, &con->writequeue) { | |
1313 | struct writequeue_entry *e = | |
1314 | list_entry(list, struct writequeue_entry, list); | |
1315 | list_del(&e->list); | |
1316 | free_entry(e); | |
1317 | } | |
1318 | spin_unlock(&con->writequeue_lock); | |
1319 | } | |
1320 | ||
1321 | /* Called from recovery when it knows that a node has | |
1322 | left the cluster */ | |
1323 | int dlm_lowcomms_close(int nodeid) | |
1324 | { | |
1325 | struct connection *con; | |
1326 | ||
fdda387f PC |
1327 | log_print("closing connection to node %d", nodeid); |
1328 | con = nodeid2con(nodeid, 0); | |
1329 | if (con) { | |
1330 | clean_one_writequeue(con); | |
ac33d071 | 1331 | close_connection(con, true); |
fdda387f PC |
1332 | } |
1333 | return 0; | |
fdda387f PC |
1334 | } |
1335 | ||
6ed7257b | 1336 | /* Receive workqueue function */ |
1d6e8131 | 1337 | static void process_recv_sockets(struct work_struct *work) |
fdda387f | 1338 | { |
1d6e8131 PC |
1339 | struct connection *con = container_of(work, struct connection, rwork); |
1340 | int err; | |
fdda387f | 1341 | |
1d6e8131 PC |
1342 | clear_bit(CF_READ_PENDING, &con->flags); |
1343 | do { | |
1344 | err = con->rx_action(con); | |
1345 | } while (!err); | |
fdda387f PC |
1346 | } |
1347 | ||
6ed7257b | 1348 | /* Send workqueue function */ |
1d6e8131 | 1349 | static void process_send_sockets(struct work_struct *work) |
fdda387f | 1350 | { |
1d6e8131 | 1351 | struct connection *con = container_of(work, struct connection, swork); |
fdda387f | 1352 | |
1d6e8131 | 1353 | if (test_and_clear_bit(CF_CONNECT_PENDING, &con->flags)) { |
6ed7257b | 1354 | con->connect_action(con); |
fdda387f | 1355 | } |
bd44e2b0 PC |
1356 | clear_bit(CF_WRITE_PENDING, &con->flags); |
1357 | send_to_sock(con); | |
fdda387f PC |
1358 | } |
1359 | ||
1360 | ||
1361 | /* Discard all entries on the write queues */ | |
1362 | static void clean_writequeues(void) | |
1363 | { | |
1364 | int nodeid; | |
1365 | ||
30d3a237 PC |
1366 | for (nodeid = 1; nodeid <= max_nodeid; nodeid++) { |
1367 | struct connection *con = __nodeid2con(nodeid, 0); | |
fdda387f PC |
1368 | |
1369 | if (con) | |
1370 | clean_one_writequeue(con); | |
1371 | } | |
1372 | } | |
1373 | ||
1d6e8131 | 1374 | static void work_stop(void) |
fdda387f | 1375 | { |
1d6e8131 PC |
1376 | destroy_workqueue(recv_workqueue); |
1377 | destroy_workqueue(send_workqueue); | |
fdda387f PC |
1378 | } |
1379 | ||
1d6e8131 | 1380 | static int work_start(void) |
fdda387f | 1381 | { |
fdda387f | 1382 | int error; |
1d6e8131 PC |
1383 | recv_workqueue = create_workqueue("dlm_recv"); |
1384 | error = IS_ERR(recv_workqueue); | |
ac33d071 | 1385 | if (error) { |
1d6e8131 | 1386 | log_print("can't start dlm_recv %d", error); |
fdda387f PC |
1387 | return error; |
1388 | } | |
fdda387f | 1389 | |
1d6e8131 PC |
1390 | send_workqueue = create_singlethread_workqueue("dlm_send"); |
1391 | error = IS_ERR(send_workqueue); | |
ac33d071 | 1392 | if (error) { |
1d6e8131 PC |
1393 | log_print("can't start dlm_send %d", error); |
1394 | destroy_workqueue(recv_workqueue); | |
fdda387f PC |
1395 | return error; |
1396 | } | |
fdda387f PC |
1397 | |
1398 | return 0; | |
1399 | } | |
1400 | ||
fdda387f PC |
1401 | void dlm_lowcomms_stop(void) |
1402 | { | |
1403 | int i; | |
6ed7257b | 1404 | struct connection *con; |
fdda387f | 1405 | |
ac33d071 | 1406 | /* Set all the flags to prevent any |
fdda387f PC |
1407 | socket activity. |
1408 | */ | |
6ed7257b | 1409 | down(&connections_lock); |
30d3a237 | 1410 | for (i = 0; i <= max_nodeid; i++) { |
6ed7257b | 1411 | con = __nodeid2con(i, 0); |
afb853fb | 1412 | if (con) { |
6ed7257b | 1413 | con->flags |= 0xFF; |
afb853fb PC |
1414 | if (con->sock) |
1415 | con->sock->sk->sk_user_data = NULL; | |
1416 | } | |
fdda387f | 1417 | } |
6ed7257b | 1418 | up(&connections_lock); |
ac33d071 | 1419 | |
1d6e8131 | 1420 | work_stop(); |
6ed7257b PC |
1421 | |
1422 | down(&connections_lock); | |
fdda387f PC |
1423 | clean_writequeues(); |
1424 | ||
30d3a237 | 1425 | for (i = 0; i <= max_nodeid; i++) { |
2439fe50 | 1426 | con = __nodeid2con(i, 0); |
6ed7257b PC |
1427 | if (con) { |
1428 | close_connection(con, true); | |
1429 | if (con->othercon) | |
1430 | kmem_cache_free(con_cache, con->othercon); | |
1431 | kmem_cache_free(con_cache, con); | |
fdda387f PC |
1432 | } |
1433 | } | |
30d3a237 | 1434 | max_nodeid = 0; |
6ed7257b | 1435 | up(&connections_lock); |
fdda387f | 1436 | kmem_cache_destroy(con_cache); |
30d3a237 | 1437 | idr_init(&connections_idr); |
fdda387f PC |
1438 | } |
1439 | ||
fdda387f PC |
1440 | int dlm_lowcomms_start(void) |
1441 | { | |
6ed7257b PC |
1442 | int error = -EINVAL; |
1443 | struct connection *con; | |
fdda387f | 1444 | |
6ed7257b PC |
1445 | init_local(); |
1446 | if (!dlm_local_count) { | |
617e82e1 | 1447 | error = -ENOTCONN; |
fdda387f | 1448 | log_print("no local IP address has been set"); |
6ed7257b | 1449 | goto out; |
fdda387f PC |
1450 | } |
1451 | ||
6ed7257b | 1452 | error = -ENOMEM; |
fdda387f | 1453 | con_cache = kmem_cache_create("dlm_conn", sizeof(struct connection), |
ac33d071 | 1454 | __alignof__(struct connection), 0, |
20c2df83 | 1455 | NULL); |
fdda387f | 1456 | if (!con_cache) |
6ed7257b | 1457 | goto out; |
fdda387f | 1458 | |
6ed7257b PC |
1459 | /* Set some sysctl minima */ |
1460 | if (sysctl_rmem_max < NEEDED_RMEM) | |
1461 | sysctl_rmem_max = NEEDED_RMEM; | |
fdda387f PC |
1462 | |
1463 | /* Start listening */ | |
6ed7257b PC |
1464 | if (dlm_config.ci_protocol == 0) |
1465 | error = tcp_listen_for_all(); | |
1466 | else | |
1467 | error = sctp_listen_for_all(); | |
fdda387f PC |
1468 | if (error) |
1469 | goto fail_unlisten; | |
1470 | ||
1d6e8131 | 1471 | error = work_start(); |
fdda387f PC |
1472 | if (error) |
1473 | goto fail_unlisten; | |
1474 | ||
fdda387f PC |
1475 | return 0; |
1476 | ||
ac33d071 | 1477 | fail_unlisten: |
6ed7257b PC |
1478 | con = nodeid2con(0,0); |
1479 | if (con) { | |
1480 | close_connection(con, false); | |
1481 | kmem_cache_free(con_cache, con); | |
1482 | } | |
fdda387f PC |
1483 | kmem_cache_destroy(con_cache); |
1484 | ||
ac33d071 | 1485 | out: |
fdda387f PC |
1486 | return error; |
1487 | } |