blob: 3635e42b066962c14219baae56828b7479f4e0a7 [file] [log] [blame]
Thomas Gleixner2522fe42019-05-28 09:57:20 -07001// SPDX-License-Identifier: GPL-2.0-only
David Teiglande7fd4172006-01-18 09:30:29 +00002/******************************************************************************
3*******************************************************************************
4**
5** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
Alexander Aring489d8e52021-05-21 15:08:46 -04006** Copyright (C) 2004-2021 Red Hat, Inc. All rights reserved.
David Teiglande7fd4172006-01-18 09:30:29 +00007**
David Teiglande7fd4172006-01-18 09:30:29 +00008**
9*******************************************************************************
10******************************************************************************/
11
12/*
13 * midcomms.c
14 *
Alexander Aring489d8e52021-05-21 15:08:46 -040015 * This is the appallingly named "mid-level" comms layer. It takes care about
16 * deliver an on application layer "reliable" communication above the used
17 * lowcomms transport layer.
David Teiglande7fd4172006-01-18 09:30:29 +000018 *
Alexander Aring489d8e52021-05-21 15:08:46 -040019 * How it works:
David Teiglande7fd4172006-01-18 09:30:29 +000020 *
Alexander Aring489d8e52021-05-21 15:08:46 -040021 * Each nodes keeps track of all send DLM messages in send_queue with a sequence
22 * number. The receive will send an DLM_ACK message back for every DLM message
23 * received at the other side. If a reconnect happens in lowcomms we will send
24 * all unacknowledged dlm messages again. The receiving side might drop any already
25 * received message by comparing sequence numbers.
26 *
27 * How version detection works:
28 *
29 * Due the fact that dlm has pre-configured node addresses on every side
30 * it is in it's nature that every side connects at starts to transmit
31 * dlm messages which ends in a race. However DLM_RCOM_NAMES, DLM_RCOM_STATUS
32 * and their replies are the first messages which are exchanges. Due backwards
33 * compatibility these messages are not covered by the midcomms re-transmission
34 * layer. These messages have their own re-transmission handling in the dlm
35 * application layer. The version field of every node will be set on these RCOM
36 * messages as soon as they arrived and the node isn't yet part of the nodes
37 * hash. There exists also logic to detect version mismatched if something weird
38 * going on or the first messages isn't an expected one.
39 *
40 * Termination:
41 *
42 * The midcomms layer does a 4 way handshake for termination on DLM protocol
43 * like TCP supports it with half-closed socket support. SCTP doesn't support
44 * half-closed socket, so we do it on DLM layer. Also socket shutdown() can be
45 * interrupted by .e.g. tcp reset itself. Additional there exists the othercon
46 * paradigm in lowcomms which cannot be easily without breaking backwards
47 * compatibility. A node cannot send anything to another node when a DLM_FIN
48 * message was send. There exists additional logic to print a warning if
49 * DLM wants to do it. There exists a state handling like RFC 793 but reduced
50 * to termination only. The event "member removal event" describes the cluster
51 * manager removed the node from internal lists, at this point DLM does not
52 * send any message to the other node. There exists two cases:
53 *
54 * 1. The cluster member was removed and we received a FIN
55 * OR
56 * 2. We received a FIN but the member was not removed yet
57 *
58 * One of these cases will do the CLOSE_WAIT to LAST_ACK change.
59 *
60 *
61 * +---------+
62 * | CLOSED |
63 * +---------+
64 * | add member/receive RCOM version
65 * | detection msg
66 * V
67 * +---------+
68 * | ESTAB |
69 * +---------+
70 * CLOSE | | rcv FIN
71 * ------- | | -------
72 * +---------+ snd FIN / \ snd ACK +---------+
73 * | FIN |<----------------- ------------------>| CLOSE |
74 * | WAIT-1 |------------------ | WAIT |
75 * +---------+ rcv FIN \ +---------+
76 * | rcv ACK of FIN ------- | CLOSE | member
77 * | -------------- snd ACK | ------- | removal
78 * V x V snd FIN V event
79 * +---------+ +---------+ +---------+
80 * |FINWAIT-2| | CLOSING | | LAST-ACK|
81 * +---------+ +---------+ +---------+
82 * | rcv ACK of FIN | rcv ACK of FIN |
83 * | rcv FIN -------------- | -------------- |
84 * | ------- x V x V
85 * \ snd ACK +---------+ +---------+
86 * ------------------------>| CLOSED | | CLOSED |
87 * +---------+ +---------+
88 *
89 * NOTE: any state can interrupted by midcomms_close() and state will be
90 * switched to CLOSED in case of fencing. There exists also some timeout
91 * handling when we receive the version detection RCOM messages which is
92 * made by observation.
93 *
94 * Future improvements:
95 *
96 * There exists some known issues/improvements of the dlm handling. Some
97 * of them should be done in a next major dlm version bump which makes
98 * it incompatible with previous versions.
99 *
100 * Unaligned memory access:
101 *
102 * There exists cases when the dlm message buffer length is not aligned
103 * to 8 byte. However seems nobody detected any problem with it. This
104 * can be fixed in the next major version bump of dlm.
105 *
106 * Version detection:
107 *
108 * The version detection and how it's done is related to backwards
109 * compatibility. There exists better ways to make a better handling.
110 * However this should be changed in the next major version bump of dlm.
111 *
Alexander Aring489d8e52021-05-21 15:08:46 -0400112 * Tail Size checking:
113 *
114 * There exists a message tail payload in e.g. DLM_MSG however we don't
115 * check it against the message length yet regarding to the receive buffer
116 * length. That need to be validated.
117 *
118 * Fencing bad nodes:
119 *
120 * At timeout places or weird sequence number behaviours we should send
121 * a fencing request to the cluster manager.
David Teiglande7fd4172006-01-18 09:30:29 +0000122 */
123
Alexander Aring489d8e52021-05-21 15:08:46 -0400124/* Debug switch to enable a 5 seconds sleep waiting of a termination.
125 * This can be useful to test fencing while termination is running.
126 * This requires a setup with only gfs2 as dlm user, so that the
127 * last umount will terminate the connection.
128 *
129 * However it became useful to test, while the 5 seconds block in umount
130 * just press the reset button. In a lot of dropping the termination
131 * process can could take several seconds.
132 */
133#define DLM_DEBUG_FENCE_TERMINATION 0
134
135#include <net/tcp.h>
136
David Teiglande7fd4172006-01-18 09:30:29 +0000137#include "dlm_internal.h"
138#include "lowcomms.h"
139#include "config.h"
Alexander Aring6c547f22021-11-30 14:47:18 -0500140#include "memory.h"
David Teiglande7fd4172006-01-18 09:30:29 +0000141#include "lock.h"
Alexander Aring489d8e52021-05-21 15:08:46 -0400142#include "util.h"
David Teiglande7fd4172006-01-18 09:30:29 +0000143#include "midcomms.h"
144
Alexander Aring489d8e52021-05-21 15:08:46 -0400145/* init value for sequence numbers for testing purpose only e.g. overflows */
146#define DLM_SEQ_INIT 0
147/* 3 minutes wait to sync ending of dlm */
148#define DLM_SHUTDOWN_TIMEOUT msecs_to_jiffies(3 * 60 * 1000)
149#define DLM_VERSION_NOT_SET 0
150
151struct midcomms_node {
152 int nodeid;
153 uint32_t version;
154 uint32_t seq_send;
155 uint32_t seq_next;
156 /* These queues are unbound because we cannot drop any message in dlm.
157 * We could send a fence signal for a specific node to the cluster
158 * manager if queues hits some maximum value, however this handling
159 * not supported yet.
160 */
161 struct list_head send_queue;
162 spinlock_t send_queue_lock;
163 atomic_t send_queue_cnt;
164#define DLM_NODE_FLAG_CLOSE 1
165#define DLM_NODE_FLAG_STOP_TX 2
166#define DLM_NODE_FLAG_STOP_RX 3
Alexander Aringb97f8522021-08-18 16:27:13 -0400167#define DLM_NODE_ULP_DELIVERED 4
Alexander Aring489d8e52021-05-21 15:08:46 -0400168 unsigned long flags;
169 wait_queue_head_t shutdown_wait;
170
171 /* dlm tcp termination state */
172#define DLM_CLOSED 1
173#define DLM_ESTABLISHED 2
174#define DLM_FIN_WAIT1 3
175#define DLM_FIN_WAIT2 4
176#define DLM_CLOSE_WAIT 5
177#define DLM_LAST_ACK 6
178#define DLM_CLOSING 7
179 int state;
180 spinlock_t state_lock;
181
182 /* counts how many lockspaces are using this node
183 * this refcount is necessary to determine if the
184 * node wants to disconnect.
185 */
186 int users;
187
Alexander Aring5b2f9812021-05-21 15:08:47 -0400188 /* not protected by srcu, node_hash lifetime */
189 void *debugfs;
190
Alexander Aring489d8e52021-05-21 15:08:46 -0400191 struct hlist_node hlist;
192 struct rcu_head rcu;
193};
194
195struct dlm_mhandle {
196 const struct dlm_header *inner_hd;
197 struct midcomms_node *node;
198 struct dlm_opts *opts;
199 struct dlm_msg *msg;
200 bool committed;
201 uint32_t seq;
202
203 void (*ack_rcv)(struct midcomms_node *node);
204
205 /* get_mhandle/commit srcu idx exchange */
206 int idx;
207
208 struct list_head list;
209 struct rcu_head rcu;
210};
211
212static struct hlist_head node_hash[CONN_HASH_SIZE];
213static DEFINE_SPINLOCK(nodes_lock);
214DEFINE_STATIC_SRCU(nodes_srcu);
215
216/* This mutex prevents that midcomms_close() is running while
217 * stop() or remove(). As I experienced invalid memory access
218 * behaviours when DLM_DEBUG_FENCE_TERMINATION is enabled and
219 * resetting machines. I will end in some double deletion in nodes
220 * datastructure.
221 */
222static DEFINE_MUTEX(close_lock);
223
Alexander Aring6c547f22021-11-30 14:47:18 -0500224struct kmem_cache *dlm_midcomms_cache_create(void)
225{
226 return kmem_cache_create("dlm_mhandle", sizeof(struct dlm_mhandle),
227 0, 0, NULL);
228}
229
Alexander Aring489d8e52021-05-21 15:08:46 -0400230static inline const char *dlm_state_str(int state)
Alexander Aringa070a912021-05-21 15:08:41 -0400231{
Alexander Aring489d8e52021-05-21 15:08:46 -0400232 switch (state) {
233 case DLM_CLOSED:
234 return "CLOSED";
235 case DLM_ESTABLISHED:
236 return "ESTABLISHED";
237 case DLM_FIN_WAIT1:
238 return "FIN_WAIT1";
239 case DLM_FIN_WAIT2:
240 return "FIN_WAIT2";
241 case DLM_CLOSE_WAIT:
242 return "CLOSE_WAIT";
243 case DLM_LAST_ACK:
244 return "LAST_ACK";
245 case DLM_CLOSING:
246 return "CLOSING";
247 default:
248 return "UNKNOWN";
249 }
Alexander Aringa070a912021-05-21 15:08:41 -0400250}
251
Alexander Aring5b2f9812021-05-21 15:08:47 -0400252const char *dlm_midcomms_state(struct midcomms_node *node)
253{
254 return dlm_state_str(node->state);
255}
256
257unsigned long dlm_midcomms_flags(struct midcomms_node *node)
258{
259 return node->flags;
260}
261
262int dlm_midcomms_send_queue_cnt(struct midcomms_node *node)
263{
264 return atomic_read(&node->send_queue_cnt);
265}
266
267uint32_t dlm_midcomms_version(struct midcomms_node *node)
268{
269 return node->version;
270}
271
Alexander Aring489d8e52021-05-21 15:08:46 -0400272static struct midcomms_node *__find_node(int nodeid, int r)
Alexander Aringa070a912021-05-21 15:08:41 -0400273{
Alexander Aring489d8e52021-05-21 15:08:46 -0400274 struct midcomms_node *node;
275
276 hlist_for_each_entry_rcu(node, &node_hash[r], hlist) {
277 if (node->nodeid == nodeid)
278 return node;
279 }
280
281 return NULL;
Alexander Aringa070a912021-05-21 15:08:41 -0400282}
283
Alexander Aring489d8e52021-05-21 15:08:46 -0400284static void dlm_mhandle_release(struct rcu_head *rcu)
Alexander Aringa070a912021-05-21 15:08:41 -0400285{
Alexander Aring489d8e52021-05-21 15:08:46 -0400286 struct dlm_mhandle *mh = container_of(rcu, struct dlm_mhandle, rcu);
287
288 dlm_lowcomms_put_msg(mh->msg);
Alexander Aring6c547f22021-11-30 14:47:18 -0500289 dlm_free_mhandle(mh);
Alexander Aringa070a912021-05-21 15:08:41 -0400290}
291
Alexander Aringf5fe8d52021-06-11 12:55:40 -0400292static void dlm_mhandle_delete(struct midcomms_node *node,
293 struct dlm_mhandle *mh)
294{
295 list_del_rcu(&mh->list);
296 atomic_dec(&node->send_queue_cnt);
297 call_rcu(&mh->rcu, dlm_mhandle_release);
298}
299
Alexander Aring489d8e52021-05-21 15:08:46 -0400300static void dlm_send_queue_flush(struct midcomms_node *node)
Alexander Aringa070a912021-05-21 15:08:41 -0400301{
Alexander Aring489d8e52021-05-21 15:08:46 -0400302 struct dlm_mhandle *mh;
303
304 pr_debug("flush midcomms send queue of node %d\n", node->nodeid);
305
306 rcu_read_lock();
Alexander Aringf5fe8d52021-06-11 12:55:40 -0400307 spin_lock(&node->send_queue_lock);
Alexander Aring489d8e52021-05-21 15:08:46 -0400308 list_for_each_entry_rcu(mh, &node->send_queue, list) {
Alexander Aringf5fe8d52021-06-11 12:55:40 -0400309 dlm_mhandle_delete(node, mh);
Alexander Aring489d8e52021-05-21 15:08:46 -0400310 }
Alexander Aringf5fe8d52021-06-11 12:55:40 -0400311 spin_unlock(&node->send_queue_lock);
Alexander Aring489d8e52021-05-21 15:08:46 -0400312 rcu_read_unlock();
Alexander Aringa070a912021-05-21 15:08:41 -0400313}
314
Alexander Aring489d8e52021-05-21 15:08:46 -0400315static void midcomms_node_reset(struct midcomms_node *node)
Alexander Aringa070a912021-05-21 15:08:41 -0400316{
Alexander Aring489d8e52021-05-21 15:08:46 -0400317 pr_debug("reset node %d\n", node->nodeid);
318
319 node->seq_next = DLM_SEQ_INIT;
320 node->seq_send = DLM_SEQ_INIT;
321 node->version = DLM_VERSION_NOT_SET;
322 node->flags = 0;
323
324 dlm_send_queue_flush(node);
325 node->state = DLM_CLOSED;
326 wake_up(&node->shutdown_wait);
327}
328
329static struct midcomms_node *nodeid2node(int nodeid, gfp_t alloc)
330{
331 struct midcomms_node *node, *tmp;
332 int r = nodeid_hash(nodeid);
333
334 node = __find_node(nodeid, r);
335 if (node || !alloc)
336 return node;
337
338 node = kmalloc(sizeof(*node), alloc);
339 if (!node)
340 return NULL;
341
342 node->nodeid = nodeid;
343 spin_lock_init(&node->state_lock);
344 spin_lock_init(&node->send_queue_lock);
345 atomic_set(&node->send_queue_cnt, 0);
346 INIT_LIST_HEAD(&node->send_queue);
347 init_waitqueue_head(&node->shutdown_wait);
348 node->users = 0;
349 midcomms_node_reset(node);
350
351 spin_lock(&nodes_lock);
352 /* check again if there was somebody else
353 * earlier here to add the node
354 */
355 tmp = __find_node(nodeid, r);
356 if (tmp) {
357 spin_unlock(&nodes_lock);
358 kfree(node);
359 return tmp;
360 }
361
362 hlist_add_head_rcu(&node->hlist, &node_hash[r]);
363 spin_unlock(&nodes_lock);
Alexander Aring5b2f9812021-05-21 15:08:47 -0400364
365 node->debugfs = dlm_create_debug_comms_file(nodeid, node);
Alexander Aring489d8e52021-05-21 15:08:46 -0400366 return node;
367}
368
369static int dlm_send_ack(int nodeid, uint32_t seq)
370{
371 int mb_len = sizeof(struct dlm_header);
372 struct dlm_header *m_header;
373 struct dlm_msg *msg;
374 char *ppc;
375
376 msg = dlm_lowcomms_new_msg(nodeid, mb_len, GFP_NOFS, &ppc,
377 NULL, NULL);
378 if (!msg)
379 return -ENOMEM;
380
381 m_header = (struct dlm_header *)ppc;
382
383 m_header->h_version = (DLM_HEADER_MAJOR | DLM_HEADER_MINOR);
384 m_header->h_nodeid = dlm_our_nodeid();
385 m_header->h_length = mb_len;
386 m_header->h_cmd = DLM_ACK;
387 m_header->u.h_seq = seq;
388
389 header_out(m_header);
390 dlm_lowcomms_commit_msg(msg);
391 dlm_lowcomms_put_msg(msg);
392
393 return 0;
394}
395
396static int dlm_send_fin(struct midcomms_node *node,
397 void (*ack_rcv)(struct midcomms_node *node))
398{
399 int mb_len = sizeof(struct dlm_header);
400 struct dlm_header *m_header;
401 struct dlm_mhandle *mh;
402 char *ppc;
403
404 mh = dlm_midcomms_get_mhandle(node->nodeid, mb_len, GFP_NOFS, &ppc);
405 if (!mh)
406 return -ENOMEM;
407
408 mh->ack_rcv = ack_rcv;
409
410 m_header = (struct dlm_header *)ppc;
411
412 m_header->h_version = (DLM_HEADER_MAJOR | DLM_HEADER_MINOR);
413 m_header->h_nodeid = dlm_our_nodeid();
414 m_header->h_length = mb_len;
415 m_header->h_cmd = DLM_FIN;
416
417 header_out(m_header);
418
419 pr_debug("sending fin msg to node %d\n", node->nodeid);
420 dlm_midcomms_commit_mhandle(mh);
421 set_bit(DLM_NODE_FLAG_STOP_TX, &node->flags);
422
423 return 0;
424}
425
426static void dlm_receive_ack(struct midcomms_node *node, uint32_t seq)
427{
428 struct dlm_mhandle *mh;
429
430 rcu_read_lock();
431 list_for_each_entry_rcu(mh, &node->send_queue, list) {
432 if (before(mh->seq, seq)) {
Alexander Aring489d8e52021-05-21 15:08:46 -0400433 if (mh->ack_rcv)
434 mh->ack_rcv(node);
Alexander Aring489d8e52021-05-21 15:08:46 -0400435 } else {
436 /* send queue should be ordered */
437 break;
438 }
439 }
Alexander Aringf5fe8d52021-06-11 12:55:40 -0400440
441 spin_lock(&node->send_queue_lock);
442 list_for_each_entry_rcu(mh, &node->send_queue, list) {
443 if (before(mh->seq, seq)) {
444 dlm_mhandle_delete(node, mh);
445 } else {
446 /* send queue should be ordered */
447 break;
448 }
449 }
450 spin_unlock(&node->send_queue_lock);
Alexander Aring489d8e52021-05-21 15:08:46 -0400451 rcu_read_unlock();
452}
453
454static void dlm_pas_fin_ack_rcv(struct midcomms_node *node)
455{
456 spin_lock(&node->state_lock);
457 pr_debug("receive passive fin ack from node %d with state %s\n",
458 node->nodeid, dlm_state_str(node->state));
459
460 switch (node->state) {
461 case DLM_LAST_ACK:
462 /* DLM_CLOSED */
463 midcomms_node_reset(node);
464 break;
465 case DLM_CLOSED:
466 /* not valid but somehow we got what we want */
467 wake_up(&node->shutdown_wait);
468 break;
469 default:
470 spin_unlock(&node->state_lock);
471 log_print("%s: unexpected state: %d\n",
472 __func__, node->state);
473 WARN_ON(1);
474 return;
475 }
476 spin_unlock(&node->state_lock);
477}
478
479static void dlm_midcomms_receive_buffer(union dlm_packet *p,
480 struct midcomms_node *node,
481 uint32_t seq)
482{
483 if (seq == node->seq_next) {
484 node->seq_next++;
Alexander Aring489d8e52021-05-21 15:08:46 -0400485
486 switch (p->header.h_cmd) {
487 case DLM_FIN:
Alexander Aringb97f8522021-08-18 16:27:13 -0400488 /* send ack before fin */
489 dlm_send_ack(node->nodeid, node->seq_next);
490
Alexander Aring489d8e52021-05-21 15:08:46 -0400491 spin_lock(&node->state_lock);
492 pr_debug("receive fin msg from node %d with state %s\n",
493 node->nodeid, dlm_state_str(node->state));
494
495 switch (node->state) {
496 case DLM_ESTABLISHED:
497 node->state = DLM_CLOSE_WAIT;
498 pr_debug("switch node %d to state %s\n",
499 node->nodeid, dlm_state_str(node->state));
500 /* passive shutdown DLM_LAST_ACK case 1
501 * additional we check if the node is used by
502 * cluster manager events at all.
503 */
504 if (node->users == 0) {
505 node->state = DLM_LAST_ACK;
506 pr_debug("switch node %d to state %s case 1\n",
507 node->nodeid, dlm_state_str(node->state));
508 spin_unlock(&node->state_lock);
509 goto send_fin;
510 }
511 break;
512 case DLM_FIN_WAIT1:
513 node->state = DLM_CLOSING;
514 pr_debug("switch node %d to state %s\n",
515 node->nodeid, dlm_state_str(node->state));
516 break;
517 case DLM_FIN_WAIT2:
518 midcomms_node_reset(node);
519 pr_debug("switch node %d to state %s\n",
520 node->nodeid, dlm_state_str(node->state));
521 wake_up(&node->shutdown_wait);
522 break;
523 case DLM_LAST_ACK:
524 /* probably remove_member caught it, do nothing */
525 break;
526 default:
527 spin_unlock(&node->state_lock);
528 log_print("%s: unexpected state: %d\n",
529 __func__, node->state);
530 WARN_ON(1);
531 return;
532 }
533 spin_unlock(&node->state_lock);
534
535 set_bit(DLM_NODE_FLAG_STOP_RX, &node->flags);
536 break;
537 default:
538 WARN_ON(test_bit(DLM_NODE_FLAG_STOP_RX, &node->flags));
539 dlm_receive_buffer(p, node->nodeid);
Alexander Aringb97f8522021-08-18 16:27:13 -0400540 set_bit(DLM_NODE_ULP_DELIVERED, &node->flags);
Alexander Aring489d8e52021-05-21 15:08:46 -0400541 break;
542 }
543 } else {
544 /* retry to ack message which we already have by sending back
545 * current node->seq_next number as ack.
546 */
547 if (seq < node->seq_next)
548 dlm_send_ack(node->nodeid, node->seq_next);
549
550 log_print_ratelimited("ignore dlm msg because seq mismatch, seq: %u, expected: %u, nodeid: %d",
551 seq, node->seq_next, node->nodeid);
552 }
553
554 return;
555
556send_fin:
557 set_bit(DLM_NODE_FLAG_STOP_RX, &node->flags);
558 dlm_send_fin(node, dlm_pas_fin_ack_rcv);
559}
560
561static struct midcomms_node *
562dlm_midcomms_recv_node_lookup(int nodeid, const union dlm_packet *p,
563 uint16_t msglen, int (*cb)(struct midcomms_node *node))
564{
565 struct midcomms_node *node = NULL;
566 gfp_t allocation = 0;
567 int ret;
568
569 switch (p->header.h_cmd) {
570 case DLM_RCOM:
571 if (msglen < sizeof(struct dlm_rcom)) {
572 log_print("rcom msg too small: %u, will skip this message from node %d",
573 msglen, nodeid);
574 return NULL;
575 }
576
577 switch (le32_to_cpu(p->rcom.rc_type)) {
578 case DLM_RCOM_NAMES:
579 fallthrough;
580 case DLM_RCOM_NAMES_REPLY:
581 fallthrough;
582 case DLM_RCOM_STATUS:
583 fallthrough;
584 case DLM_RCOM_STATUS_REPLY:
585 node = nodeid2node(nodeid, 0);
586 if (node) {
587 spin_lock(&node->state_lock);
588 if (node->state != DLM_ESTABLISHED)
589 pr_debug("receive begin RCOM msg from node %d with state %s\n",
590 node->nodeid, dlm_state_str(node->state));
591
592 switch (node->state) {
593 case DLM_CLOSED:
594 node->state = DLM_ESTABLISHED;
595 pr_debug("switch node %d to state %s\n",
596 node->nodeid, dlm_state_str(node->state));
597 break;
598 case DLM_ESTABLISHED:
599 break;
600 default:
601 /* some invalid state passive shutdown
602 * was failed, we try to reset and
603 * hope it will go on.
604 */
Colin Ian King7d3848c2021-05-26 09:53:39 +0100605 log_print("reset node %d because shutdown stuck",
Alexander Aring489d8e52021-05-21 15:08:46 -0400606 node->nodeid);
607
608 midcomms_node_reset(node);
609 node->state = DLM_ESTABLISHED;
610 break;
611 }
612 spin_unlock(&node->state_lock);
613 }
614
615 allocation = GFP_NOFS;
616 break;
617 default:
618 break;
619 }
620
621 break;
622 default:
623 break;
624 }
625
626 node = nodeid2node(nodeid, allocation);
627 if (!node) {
Alexander Aring957adb62021-06-11 12:55:41 -0400628 switch (p->header.h_cmd) {
629 case DLM_OPTS:
630 if (msglen < sizeof(struct dlm_opts)) {
631 log_print("opts msg too small: %u, will skip this message from node %d",
632 msglen, nodeid);
633 return NULL;
634 }
635
636 log_print_ratelimited("received dlm opts message nextcmd %d from node %d in an invalid sequence",
637 p->opts.o_nextcmd, nodeid);
638 break;
639 default:
640 log_print_ratelimited("received dlm message cmd %d from node %d in an invalid sequence",
641 p->header.h_cmd, nodeid);
642 break;
643 }
644
Alexander Aring489d8e52021-05-21 15:08:46 -0400645 return NULL;
646 }
647
648 ret = cb(node);
649 if (ret < 0)
650 return NULL;
651
652 return node;
653}
654
655static int dlm_midcomms_version_check_3_2(struct midcomms_node *node)
656{
657 switch (node->version) {
658 case DLM_VERSION_NOT_SET:
659 node->version = DLM_VERSION_3_2;
660 log_print("version 0x%08x for node %d detected", DLM_VERSION_3_2,
661 node->nodeid);
662 break;
663 case DLM_VERSION_3_2:
664 break;
665 default:
666 log_print_ratelimited("version mismatch detected, assumed 0x%08x but node %d has 0x%08x",
667 DLM_VERSION_3_2, node->nodeid, node->version);
668 return -1;
669 }
670
671 return 0;
672}
673
674static int dlm_opts_check_msglen(union dlm_packet *p, uint16_t msglen, int nodeid)
675{
676 int len = msglen;
677
678 /* we only trust outer header msglen because
679 * it's checked against receive buffer length.
680 */
681 if (len < sizeof(struct dlm_opts))
682 return -1;
683 len -= sizeof(struct dlm_opts);
684
685 if (len < le16_to_cpu(p->opts.o_optlen))
686 return -1;
687 len -= le16_to_cpu(p->opts.o_optlen);
688
689 switch (p->opts.o_nextcmd) {
690 case DLM_FIN:
691 if (len < sizeof(struct dlm_header)) {
692 log_print("fin too small: %d, will skip this message from node %d",
693 len, nodeid);
694 return -1;
695 }
696
697 break;
698 case DLM_MSG:
699 if (len < sizeof(struct dlm_message)) {
700 log_print("msg too small: %d, will skip this message from node %d",
701 msglen, nodeid);
702 return -1;
703 }
704
705 break;
706 case DLM_RCOM:
707 if (len < sizeof(struct dlm_rcom)) {
708 log_print("rcom msg too small: %d, will skip this message from node %d",
709 len, nodeid);
710 return -1;
711 }
712
713 break;
714 default:
715 log_print("unsupported o_nextcmd received: %u, will skip this message from node %d",
716 p->opts.o_nextcmd, nodeid);
717 return -1;
718 }
719
720 return 0;
721}
722
723static void dlm_midcomms_receive_buffer_3_2(union dlm_packet *p, int nodeid)
724{
725 uint16_t msglen = le16_to_cpu(p->header.h_length);
726 struct midcomms_node *node;
727 uint32_t seq;
728 int ret, idx;
729
730 idx = srcu_read_lock(&nodes_srcu);
731 node = dlm_midcomms_recv_node_lookup(nodeid, p, msglen,
732 dlm_midcomms_version_check_3_2);
733 if (!node)
734 goto out;
735
736 switch (p->header.h_cmd) {
737 case DLM_RCOM:
738 /* these rcom message we use to determine version.
739 * they have their own retransmission handling and
740 * are the first messages of dlm.
741 *
742 * length already checked.
743 */
744 switch (le32_to_cpu(p->rcom.rc_type)) {
745 case DLM_RCOM_NAMES:
746 fallthrough;
747 case DLM_RCOM_NAMES_REPLY:
748 fallthrough;
749 case DLM_RCOM_STATUS:
750 fallthrough;
751 case DLM_RCOM_STATUS_REPLY:
752 break;
753 default:
754 log_print("unsupported rcom type received: %u, will skip this message from node %d",
755 le32_to_cpu(p->rcom.rc_type), nodeid);
756 goto out;
757 }
758
759 WARN_ON(test_bit(DLM_NODE_FLAG_STOP_RX, &node->flags));
760 dlm_receive_buffer(p, nodeid);
761 break;
762 case DLM_OPTS:
763 seq = le32_to_cpu(p->header.u.h_seq);
764
765 ret = dlm_opts_check_msglen(p, msglen, nodeid);
766 if (ret < 0) {
767 log_print("opts msg too small: %u, will skip this message from node %d",
768 msglen, nodeid);
769 goto out;
770 }
771
772 p = (union dlm_packet *)((unsigned char *)p->opts.o_opts +
773 le16_to_cpu(p->opts.o_optlen));
774
775 /* recheck inner msglen just if it's not garbage */
776 msglen = le16_to_cpu(p->header.h_length);
777 switch (p->header.h_cmd) {
778 case DLM_RCOM:
779 if (msglen < sizeof(struct dlm_rcom)) {
780 log_print("inner rcom msg too small: %u, will skip this message from node %d",
781 msglen, nodeid);
782 goto out;
783 }
784
785 break;
786 case DLM_MSG:
787 if (msglen < sizeof(struct dlm_message)) {
788 log_print("inner msg too small: %u, will skip this message from node %d",
789 msglen, nodeid);
790 goto out;
791 }
792
793 break;
794 case DLM_FIN:
795 if (msglen < sizeof(struct dlm_header)) {
796 log_print("inner fin too small: %u, will skip this message from node %d",
797 msglen, nodeid);
798 goto out;
799 }
800
801 break;
802 default:
803 log_print("unsupported inner h_cmd received: %u, will skip this message from node %d",
804 msglen, nodeid);
805 goto out;
806 }
807
808 dlm_midcomms_receive_buffer(p, node, seq);
809 break;
810 case DLM_ACK:
811 seq = le32_to_cpu(p->header.u.h_seq);
812 dlm_receive_ack(node, seq);
813 break;
814 default:
815 log_print("unsupported h_cmd received: %u, will skip this message from node %d",
816 p->header.h_cmd, nodeid);
817 break;
818 }
819
820out:
821 srcu_read_unlock(&nodes_srcu, idx);
822}
823
824static int dlm_midcomms_version_check_3_1(struct midcomms_node *node)
825{
826 switch (node->version) {
827 case DLM_VERSION_NOT_SET:
828 node->version = DLM_VERSION_3_1;
829 log_print("version 0x%08x for node %d detected", DLM_VERSION_3_1,
830 node->nodeid);
831 break;
832 case DLM_VERSION_3_1:
833 break;
834 default:
835 log_print_ratelimited("version mismatch detected, assumed 0x%08x but node %d has 0x%08x",
836 DLM_VERSION_3_1, node->nodeid, node->version);
837 return -1;
838 }
839
840 return 0;
841}
842
843static void dlm_midcomms_receive_buffer_3_1(union dlm_packet *p, int nodeid)
844{
845 uint16_t msglen = le16_to_cpu(p->header.h_length);
846 struct midcomms_node *node;
847 int idx;
848
849 idx = srcu_read_lock(&nodes_srcu);
850 node = dlm_midcomms_recv_node_lookup(nodeid, p, msglen,
851 dlm_midcomms_version_check_3_1);
852 if (!node) {
853 srcu_read_unlock(&nodes_srcu, idx);
854 return;
855 }
856 srcu_read_unlock(&nodes_srcu, idx);
857
858 switch (p->header.h_cmd) {
859 case DLM_RCOM:
860 /* length already checked */
861 break;
862 case DLM_MSG:
863 if (msglen < sizeof(struct dlm_message)) {
864 log_print("msg too small: %u, will skip this message from node %d",
865 msglen, nodeid);
866 return;
867 }
868
869 break;
870 default:
871 log_print("unsupported h_cmd received: %u, will skip this message from node %d",
872 p->header.h_cmd, nodeid);
873 return;
874 }
875
876 dlm_receive_buffer(p, nodeid);
Alexander Aringa070a912021-05-21 15:08:41 -0400877}
878
David Teiglande7fd4172006-01-18 09:30:29 +0000879/*
880 * Called from the low-level comms layer to process a buffer of
881 * commands.
David Teiglande7fd4172006-01-18 09:30:29 +0000882 */
883
Alexander Aring4798cbb2020-09-24 10:31:26 -0400884int dlm_process_incoming_buffer(int nodeid, unsigned char *buf, int len)
David Teiglande7fd4172006-01-18 09:30:29 +0000885{
Alexander Aring4798cbb2020-09-24 10:31:26 -0400886 const unsigned char *ptr = buf;
887 const struct dlm_header *hd;
David Teiglande7fd4172006-01-18 09:30:29 +0000888 uint16_t msglen;
Alexander Aring4798cbb2020-09-24 10:31:26 -0400889 int ret = 0;
David Teiglande7fd4172006-01-18 09:30:29 +0000890
Alexander Aring4798cbb2020-09-24 10:31:26 -0400891 while (len >= sizeof(struct dlm_header)) {
892 hd = (struct dlm_header *)ptr;
David Teiglande7fd4172006-01-18 09:30:29 +0000893
Alexander Aringd10a0b882021-06-02 09:45:20 -0400894 /* no message should be more than DLM_MAX_SOCKET_BUFSIZE or
Alexander Aringdf9e06b2021-03-01 17:05:18 -0500895 * less than dlm_header size.
896 *
897 * Some messages does not have a 8 byte length boundary yet
898 * which can occur in a unaligned memory access of some dlm
899 * messages. However this problem need to be fixed at the
900 * sending side, for now it seems nobody run into architecture
901 * related issues yet but it slows down some processing.
902 * Fixing this issue should be scheduled in future by doing
903 * the next major version bump.
Alexander Aring4798cbb2020-09-24 10:31:26 -0400904 */
Alexander Aringdf9e06b2021-03-01 17:05:18 -0500905 msglen = le16_to_cpu(hd->h_length);
Alexander Aringd10a0b882021-06-02 09:45:20 -0400906 if (msglen > DLM_MAX_SOCKET_BUFSIZE ||
Alexander Aring710176e2021-03-01 17:05:17 -0500907 msglen < sizeof(struct dlm_header)) {
908 log_print("received invalid length header: %u from node %d, will abort message parsing",
909 msglen, nodeid);
Alexander Aring4798cbb2020-09-24 10:31:26 -0400910 return -EBADMSG;
Al Viroeef7d732008-01-25 00:58:46 -0500911 }
David Teiglande7fd4172006-01-18 09:30:29 +0000912
Alexander Aring4798cbb2020-09-24 10:31:26 -0400913 /* caller will take care that leftover
914 * will be parsed next call with more data
915 */
David Teiglande7fd4172006-01-18 09:30:29 +0000916 if (msglen > len)
917 break;
918
Alexander Aring658bd572021-11-02 15:17:11 -0400919 switch (hd->h_version) {
920 case cpu_to_le32(DLM_VERSION_3_1):
Alexander Aring489d8e52021-05-21 15:08:46 -0400921 dlm_midcomms_receive_buffer_3_1((union dlm_packet *)ptr, nodeid);
Alexander Aring4798cbb2020-09-24 10:31:26 -0400922 break;
Alexander Aring658bd572021-11-02 15:17:11 -0400923 case cpu_to_le32(DLM_VERSION_3_2):
Alexander Aring489d8e52021-05-21 15:08:46 -0400924 dlm_midcomms_receive_buffer_3_2((union dlm_packet *)ptr, nodeid);
Alexander Aring4798cbb2020-09-24 10:31:26 -0400925 break;
926 default:
Alexander Aring489d8e52021-05-21 15:08:46 -0400927 log_print("received invalid version header: %u from node %d, will skip this message",
928 le32_to_cpu(hd->h_version), nodeid);
929 break;
David Teiglande7fd4172006-01-18 09:30:29 +0000930 }
931
David Teiglande7fd4172006-01-18 09:30:29 +0000932 ret += msglen;
David Teiglande7fd4172006-01-18 09:30:29 +0000933 len -= msglen;
Alexander Aring4798cbb2020-09-24 10:31:26 -0400934 ptr += msglen;
David Teiglande7fd4172006-01-18 09:30:29 +0000935 }
936
Alexander Aring4798cbb2020-09-24 10:31:26 -0400937 return ret;
David Teiglande7fd4172006-01-18 09:30:29 +0000938}
Alexander Aring489d8e52021-05-21 15:08:46 -0400939
Alexander Aringb97f8522021-08-18 16:27:13 -0400940void dlm_midcomms_receive_done(int nodeid)
941{
942 struct midcomms_node *node;
943 int idx;
944
945 idx = srcu_read_lock(&nodes_srcu);
946 node = nodeid2node(nodeid, 0);
947 if (!node) {
948 srcu_read_unlock(&nodes_srcu, idx);
949 return;
950 }
951
952 /* old protocol, we do nothing */
953 switch (node->version) {
954 case DLM_VERSION_3_2:
955 break;
956 default:
957 srcu_read_unlock(&nodes_srcu, idx);
958 return;
959 }
960
961 /* do nothing if we didn't delivered stateful to ulp */
962 if (!test_and_clear_bit(DLM_NODE_ULP_DELIVERED,
963 &node->flags)) {
964 srcu_read_unlock(&nodes_srcu, idx);
965 return;
966 }
967
968 spin_lock(&node->state_lock);
969 /* we only ack if state is ESTABLISHED */
970 switch (node->state) {
971 case DLM_ESTABLISHED:
972 spin_unlock(&node->state_lock);
973 dlm_send_ack(node->nodeid, node->seq_next);
974 break;
975 default:
976 spin_unlock(&node->state_lock);
977 /* do nothing FIN has it's own ack send */
978 break;
Zhang Mingyuc8b9f342021-11-05 01:43:20 +0000979 }
Alexander Aringb97f8522021-08-18 16:27:13 -0400980 srcu_read_unlock(&nodes_srcu, idx);
981}
982
Alexander Aring489d8e52021-05-21 15:08:46 -0400983void dlm_midcomms_unack_msg_resend(int nodeid)
984{
985 struct midcomms_node *node;
986 struct dlm_mhandle *mh;
987 int idx, ret;
988
989 idx = srcu_read_lock(&nodes_srcu);
990 node = nodeid2node(nodeid, 0);
991 if (!node) {
992 srcu_read_unlock(&nodes_srcu, idx);
993 return;
994 }
995
996 /* old protocol, we don't support to retransmit on failure */
997 switch (node->version) {
998 case DLM_VERSION_3_2:
999 break;
1000 default:
1001 srcu_read_unlock(&nodes_srcu, idx);
1002 return;
1003 }
1004
1005 rcu_read_lock();
1006 list_for_each_entry_rcu(mh, &node->send_queue, list) {
1007 if (!mh->committed)
1008 continue;
1009
1010 ret = dlm_lowcomms_resend_msg(mh->msg);
1011 if (!ret)
1012 log_print_ratelimited("retransmit dlm msg, seq %u, nodeid %d",
1013 mh->seq, node->nodeid);
1014 }
1015 rcu_read_unlock();
1016 srcu_read_unlock(&nodes_srcu, idx);
1017}
1018
1019static void dlm_fill_opts_header(struct dlm_opts *opts, uint16_t inner_len,
1020 uint32_t seq)
1021{
1022 opts->o_header.h_cmd = DLM_OPTS;
1023 opts->o_header.h_version = (DLM_HEADER_MAJOR | DLM_HEADER_MINOR);
1024 opts->o_header.h_nodeid = dlm_our_nodeid();
1025 opts->o_header.h_length = DLM_MIDCOMMS_OPT_LEN + inner_len;
1026 opts->o_header.u.h_seq = seq;
1027 header_out(&opts->o_header);
1028}
1029
Alexander Aring5c16feb2021-11-02 15:17:19 -04001030static void midcomms_new_msg_cb(void *data)
Alexander Aring489d8e52021-05-21 15:08:46 -04001031{
Alexander Aring5c16feb2021-11-02 15:17:19 -04001032 struct dlm_mhandle *mh = data;
1033
Alexander Aring489d8e52021-05-21 15:08:46 -04001034 atomic_inc(&mh->node->send_queue_cnt);
1035
1036 spin_lock(&mh->node->send_queue_lock);
1037 list_add_tail_rcu(&mh->list, &mh->node->send_queue);
1038 spin_unlock(&mh->node->send_queue_lock);
1039
1040 mh->seq = mh->node->seq_send++;
1041}
1042
1043static struct dlm_msg *dlm_midcomms_get_msg_3_2(struct dlm_mhandle *mh, int nodeid,
1044 int len, gfp_t allocation, char **ppc)
1045{
1046 struct dlm_opts *opts;
1047 struct dlm_msg *msg;
1048
1049 msg = dlm_lowcomms_new_msg(nodeid, len + DLM_MIDCOMMS_OPT_LEN,
1050 allocation, ppc, midcomms_new_msg_cb, mh);
1051 if (!msg)
1052 return NULL;
1053
1054 opts = (struct dlm_opts *)*ppc;
1055 mh->opts = opts;
1056
1057 /* add possible options here */
1058 dlm_fill_opts_header(opts, len, mh->seq);
1059
1060 *ppc += sizeof(*opts);
1061 mh->inner_hd = (const struct dlm_header *)*ppc;
1062 return msg;
1063}
1064
1065struct dlm_mhandle *dlm_midcomms_get_mhandle(int nodeid, int len,
1066 gfp_t allocation, char **ppc)
1067{
1068 struct midcomms_node *node;
1069 struct dlm_mhandle *mh;
1070 struct dlm_msg *msg;
1071 int idx;
1072
1073 idx = srcu_read_lock(&nodes_srcu);
1074 node = nodeid2node(nodeid, 0);
1075 if (!node) {
1076 WARN_ON_ONCE(1);
1077 goto err;
1078 }
1079
1080 /* this is a bug, however we going on and hope it will be resolved */
1081 WARN_ON(test_bit(DLM_NODE_FLAG_STOP_TX, &node->flags));
1082
Alexander Aring6c547f22021-11-30 14:47:18 -05001083 mh = dlm_allocate_mhandle();
Alexander Aring489d8e52021-05-21 15:08:46 -04001084 if (!mh)
1085 goto err;
1086
Alexander Aring6c547f22021-11-30 14:47:18 -05001087 mh->committed = false;
1088 mh->ack_rcv = NULL;
Alexander Aring489d8e52021-05-21 15:08:46 -04001089 mh->idx = idx;
1090 mh->node = node;
1091
1092 switch (node->version) {
1093 case DLM_VERSION_3_1:
1094 msg = dlm_lowcomms_new_msg(nodeid, len, allocation, ppc,
1095 NULL, NULL);
1096 if (!msg) {
Alexander Aring6c547f22021-11-30 14:47:18 -05001097 dlm_free_mhandle(mh);
Alexander Aring489d8e52021-05-21 15:08:46 -04001098 goto err;
1099 }
1100
1101 break;
1102 case DLM_VERSION_3_2:
1103 msg = dlm_midcomms_get_msg_3_2(mh, nodeid, len, allocation,
1104 ppc);
1105 if (!msg) {
Alexander Aring6c547f22021-11-30 14:47:18 -05001106 dlm_free_mhandle(mh);
Alexander Aring489d8e52021-05-21 15:08:46 -04001107 goto err;
1108 }
1109
1110 break;
1111 default:
Alexander Aring6c547f22021-11-30 14:47:18 -05001112 dlm_free_mhandle(mh);
Alexander Aring489d8e52021-05-21 15:08:46 -04001113 WARN_ON(1);
1114 goto err;
1115 }
1116
1117 mh->msg = msg;
1118
1119 /* keep in mind that is a must to call
1120 * dlm_midcomms_commit_msg() which releases
1121 * nodes_srcu using mh->idx which is assumed
1122 * here that the application will call it.
1123 */
1124 return mh;
1125
1126err:
1127 srcu_read_unlock(&nodes_srcu, idx);
1128 return NULL;
1129}
1130
1131static void dlm_midcomms_commit_msg_3_2(struct dlm_mhandle *mh)
1132{
1133 /* nexthdr chain for fast lookup */
1134 mh->opts->o_nextcmd = mh->inner_hd->h_cmd;
1135 mh->committed = true;
1136 dlm_lowcomms_commit_msg(mh->msg);
1137}
1138
1139void dlm_midcomms_commit_mhandle(struct dlm_mhandle *mh)
1140{
1141 switch (mh->node->version) {
1142 case DLM_VERSION_3_1:
1143 srcu_read_unlock(&nodes_srcu, mh->idx);
1144
1145 dlm_lowcomms_commit_msg(mh->msg);
1146 dlm_lowcomms_put_msg(mh->msg);
1147 /* mh is not part of rcu list in this case */
Alexander Aring6c547f22021-11-30 14:47:18 -05001148 dlm_free_mhandle(mh);
Alexander Aring489d8e52021-05-21 15:08:46 -04001149 break;
1150 case DLM_VERSION_3_2:
1151 dlm_midcomms_commit_msg_3_2(mh);
1152 srcu_read_unlock(&nodes_srcu, mh->idx);
1153 break;
1154 default:
1155 srcu_read_unlock(&nodes_srcu, mh->idx);
1156 WARN_ON(1);
1157 break;
1158 }
1159}
1160
1161int dlm_midcomms_start(void)
1162{
1163 int i;
1164
1165 for (i = 0; i < CONN_HASH_SIZE; i++)
1166 INIT_HLIST_HEAD(&node_hash[i]);
1167
1168 return dlm_lowcomms_start();
1169}
1170
1171static void dlm_act_fin_ack_rcv(struct midcomms_node *node)
1172{
1173 spin_lock(&node->state_lock);
1174 pr_debug("receive active fin ack from node %d with state %s\n",
1175 node->nodeid, dlm_state_str(node->state));
1176
1177 switch (node->state) {
1178 case DLM_FIN_WAIT1:
1179 node->state = DLM_FIN_WAIT2;
1180 pr_debug("switch node %d to state %s\n",
1181 node->nodeid, dlm_state_str(node->state));
1182 break;
1183 case DLM_CLOSING:
1184 midcomms_node_reset(node);
1185 pr_debug("switch node %d to state %s\n",
1186 node->nodeid, dlm_state_str(node->state));
1187 wake_up(&node->shutdown_wait);
1188 break;
1189 case DLM_CLOSED:
1190 /* not valid but somehow we got what we want */
1191 wake_up(&node->shutdown_wait);
1192 break;
1193 default:
1194 spin_unlock(&node->state_lock);
1195 log_print("%s: unexpected state: %d\n",
1196 __func__, node->state);
1197 WARN_ON(1);
1198 return;
1199 }
1200 spin_unlock(&node->state_lock);
1201}
1202
1203void dlm_midcomms_add_member(int nodeid)
1204{
1205 struct midcomms_node *node;
1206 int idx;
1207
1208 if (nodeid == dlm_our_nodeid())
1209 return;
1210
1211 idx = srcu_read_lock(&nodes_srcu);
1212 node = nodeid2node(nodeid, GFP_NOFS);
1213 if (!node) {
1214 srcu_read_unlock(&nodes_srcu, idx);
1215 return;
1216 }
1217
1218 spin_lock(&node->state_lock);
1219 if (!node->users) {
1220 pr_debug("receive add member from node %d with state %s\n",
1221 node->nodeid, dlm_state_str(node->state));
1222 switch (node->state) {
1223 case DLM_ESTABLISHED:
1224 break;
1225 case DLM_CLOSED:
1226 node->state = DLM_ESTABLISHED;
1227 pr_debug("switch node %d to state %s\n",
1228 node->nodeid, dlm_state_str(node->state));
1229 break;
1230 default:
1231 /* some invalid state passive shutdown
1232 * was failed, we try to reset and
1233 * hope it will go on.
1234 */
Colin Ian King7d3848c2021-05-26 09:53:39 +01001235 log_print("reset node %d because shutdown stuck",
Alexander Aring489d8e52021-05-21 15:08:46 -04001236 node->nodeid);
1237
1238 midcomms_node_reset(node);
1239 node->state = DLM_ESTABLISHED;
1240 break;
1241 }
1242 }
1243
1244 node->users++;
Alexander Aring1aafd9c2021-11-02 15:17:09 -04001245 pr_debug("node %d users inc count %d\n", nodeid, node->users);
Alexander Aring489d8e52021-05-21 15:08:46 -04001246 spin_unlock(&node->state_lock);
1247
1248 srcu_read_unlock(&nodes_srcu, idx);
1249}
1250
1251void dlm_midcomms_remove_member(int nodeid)
1252{
1253 struct midcomms_node *node;
1254 int idx;
1255
1256 if (nodeid == dlm_our_nodeid())
1257 return;
1258
1259 idx = srcu_read_lock(&nodes_srcu);
1260 node = nodeid2node(nodeid, 0);
1261 if (!node) {
1262 srcu_read_unlock(&nodes_srcu, idx);
1263 return;
1264 }
1265
1266 spin_lock(&node->state_lock);
1267 node->users--;
Alexander Aring1aafd9c2021-11-02 15:17:09 -04001268 pr_debug("node %d users dec count %d\n", nodeid, node->users);
Alexander Aring489d8e52021-05-21 15:08:46 -04001269
1270 /* hitting users count to zero means the
1271 * other side is running dlm_midcomms_stop()
1272 * we meet us to have a clean disconnect.
1273 */
1274 if (node->users == 0) {
1275 pr_debug("receive remove member from node %d with state %s\n",
1276 node->nodeid, dlm_state_str(node->state));
1277 switch (node->state) {
1278 case DLM_ESTABLISHED:
1279 break;
1280 case DLM_CLOSE_WAIT:
1281 /* passive shutdown DLM_LAST_ACK case 2 */
1282 node->state = DLM_LAST_ACK;
1283 spin_unlock(&node->state_lock);
1284
1285 pr_debug("switch node %d to state %s case 2\n",
1286 node->nodeid, dlm_state_str(node->state));
1287 goto send_fin;
1288 case DLM_LAST_ACK:
1289 /* probably receive fin caught it, do nothing */
1290 break;
1291 case DLM_CLOSED:
1292 /* already gone, do nothing */
1293 break;
1294 default:
1295 log_print("%s: unexpected state: %d\n",
1296 __func__, node->state);
1297 break;
1298 }
1299 }
1300 spin_unlock(&node->state_lock);
1301
1302 srcu_read_unlock(&nodes_srcu, idx);
1303 return;
1304
1305send_fin:
1306 set_bit(DLM_NODE_FLAG_STOP_RX, &node->flags);
1307 dlm_send_fin(node, dlm_pas_fin_ack_rcv);
1308 srcu_read_unlock(&nodes_srcu, idx);
1309}
1310
1311static void midcomms_node_release(struct rcu_head *rcu)
1312{
1313 struct midcomms_node *node = container_of(rcu, struct midcomms_node, rcu);
1314
1315 WARN_ON(atomic_read(&node->send_queue_cnt));
1316 kfree(node);
1317}
1318
1319static void midcomms_shutdown(struct midcomms_node *node)
1320{
1321 int ret;
1322
1323 /* old protocol, we don't wait for pending operations */
1324 switch (node->version) {
1325 case DLM_VERSION_3_2:
1326 break;
1327 default:
1328 return;
1329 }
1330
1331 spin_lock(&node->state_lock);
1332 pr_debug("receive active shutdown for node %d with state %s\n",
1333 node->nodeid, dlm_state_str(node->state));
1334 switch (node->state) {
1335 case DLM_ESTABLISHED:
1336 node->state = DLM_FIN_WAIT1;
1337 pr_debug("switch node %d to state %s case 2\n",
1338 node->nodeid, dlm_state_str(node->state));
1339 break;
1340 case DLM_CLOSED:
1341 /* we have what we want */
1342 spin_unlock(&node->state_lock);
1343 return;
1344 default:
1345 /* busy to enter DLM_FIN_WAIT1, wait until passive
1346 * done in shutdown_wait to enter DLM_CLOSED.
1347 */
1348 break;
1349 }
1350 spin_unlock(&node->state_lock);
1351
1352 if (node->state == DLM_FIN_WAIT1) {
1353 dlm_send_fin(node, dlm_act_fin_ack_rcv);
1354
1355 if (DLM_DEBUG_FENCE_TERMINATION)
1356 msleep(5000);
1357 }
1358
1359 /* wait for other side dlm + fin */
1360 ret = wait_event_timeout(node->shutdown_wait,
1361 node->state == DLM_CLOSED ||
1362 test_bit(DLM_NODE_FLAG_CLOSE, &node->flags),
1363 DLM_SHUTDOWN_TIMEOUT);
1364 if (!ret || test_bit(DLM_NODE_FLAG_CLOSE, &node->flags)) {
1365 pr_debug("active shutdown timed out for node %d with state %s\n",
1366 node->nodeid, dlm_state_str(node->state));
1367 midcomms_node_reset(node);
1368 return;
1369 }
1370
1371 pr_debug("active shutdown done for node %d with state %s\n",
1372 node->nodeid, dlm_state_str(node->state));
1373}
1374
1375void dlm_midcomms_shutdown(void)
1376{
1377 struct midcomms_node *node;
1378 int i, idx;
1379
1380 mutex_lock(&close_lock);
1381 idx = srcu_read_lock(&nodes_srcu);
1382 for (i = 0; i < CONN_HASH_SIZE; i++) {
1383 hlist_for_each_entry_rcu(node, &node_hash[i], hlist) {
1384 midcomms_shutdown(node);
1385
Alexander Aring5b2f9812021-05-21 15:08:47 -04001386 dlm_delete_debug_comms_file(node->debugfs);
1387
Alexander Aring489d8e52021-05-21 15:08:46 -04001388 spin_lock(&nodes_lock);
1389 hlist_del_rcu(&node->hlist);
1390 spin_unlock(&nodes_lock);
1391
1392 call_srcu(&nodes_srcu, &node->rcu, midcomms_node_release);
1393 }
1394 }
1395 srcu_read_unlock(&nodes_srcu, idx);
1396 mutex_unlock(&close_lock);
1397
1398 dlm_lowcomms_shutdown();
1399}
1400
1401int dlm_midcomms_close(int nodeid)
1402{
1403 struct midcomms_node *node;
1404 int idx, ret;
1405
1406 if (nodeid == dlm_our_nodeid())
1407 return 0;
1408
1409 idx = srcu_read_lock(&nodes_srcu);
1410 /* Abort pending close/remove operation */
1411 node = nodeid2node(nodeid, 0);
1412 if (node) {
1413 /* let shutdown waiters leave */
1414 set_bit(DLM_NODE_FLAG_CLOSE, &node->flags);
1415 wake_up(&node->shutdown_wait);
1416 }
1417 srcu_read_unlock(&nodes_srcu, idx);
1418
1419 synchronize_srcu(&nodes_srcu);
1420
1421 idx = srcu_read_lock(&nodes_srcu);
1422 mutex_lock(&close_lock);
1423 node = nodeid2node(nodeid, 0);
1424 if (!node) {
1425 mutex_unlock(&close_lock);
1426 srcu_read_unlock(&nodes_srcu, idx);
1427 return dlm_lowcomms_close(nodeid);
1428 }
1429
1430 ret = dlm_lowcomms_close(nodeid);
1431 spin_lock(&node->state_lock);
1432 midcomms_node_reset(node);
1433 spin_unlock(&node->state_lock);
1434 srcu_read_unlock(&nodes_srcu, idx);
1435 mutex_unlock(&close_lock);
1436
1437 return ret;
1438}
Alexander Aring9af5b8f2021-11-02 15:17:20 -04001439
1440/* debug functionality to send raw dlm msg from user space */
1441struct dlm_rawmsg_data {
1442 struct midcomms_node *node;
1443 void *buf;
1444};
1445
1446static void midcomms_new_rawmsg_cb(void *data)
1447{
1448 struct dlm_rawmsg_data *rd = data;
1449 struct dlm_header *h = rd->buf;
1450
1451 switch (h->h_version) {
1452 case cpu_to_le32(DLM_VERSION_3_1):
1453 break;
1454 default:
1455 switch (h->h_cmd) {
1456 case DLM_OPTS:
1457 if (!h->u.h_seq)
1458 h->u.h_seq = rd->node->seq_send++;
1459 break;
1460 default:
1461 break;
1462 }
1463 break;
1464 }
1465}
1466
1467int dlm_midcomms_rawmsg_send(struct midcomms_node *node, void *buf,
1468 int buflen)
1469{
1470 struct dlm_rawmsg_data rd;
1471 struct dlm_msg *msg;
1472 char *msgbuf;
1473
1474 rd.node = node;
1475 rd.buf = buf;
1476
1477 msg = dlm_lowcomms_new_msg(node->nodeid, buflen, GFP_NOFS,
1478 &msgbuf, midcomms_new_rawmsg_cb, &rd);
1479 if (!msg)
1480 return -ENOMEM;
1481
1482 memcpy(msgbuf, buf, buflen);
1483 dlm_lowcomms_commit_msg(msg);
1484 return 0;
1485}
1486