xref: /openbmc/linux/fs/dlm/rcom.c (revision 63dc02bd)
1 /******************************************************************************
2 *******************************************************************************
3 **
4 **  Copyright (C) Sistina Software, Inc.  1997-2003  All rights reserved.
5 **  Copyright (C) 2005-2008 Red Hat, Inc.  All rights reserved.
6 **
7 **  This copyrighted material is made available to anyone wishing to use,
8 **  modify, copy, or redistribute it subject to the terms and conditions
9 **  of the GNU General Public License v.2.
10 **
11 *******************************************************************************
12 ******************************************************************************/
13 
14 #include "dlm_internal.h"
15 #include "lockspace.h"
16 #include "member.h"
17 #include "lowcomms.h"
18 #include "midcomms.h"
19 #include "rcom.h"
20 #include "recover.h"
21 #include "dir.h"
22 #include "config.h"
23 #include "memory.h"
24 #include "lock.h"
25 #include "util.h"
26 #include "member.h"
27 
28 
29 static int rcom_response(struct dlm_ls *ls)
30 {
31 	return test_bit(LSFL_RCOM_READY, &ls->ls_flags);
32 }
33 
34 static int create_rcom(struct dlm_ls *ls, int to_nodeid, int type, int len,
35 		       struct dlm_rcom **rc_ret, struct dlm_mhandle **mh_ret)
36 {
37 	struct dlm_rcom *rc;
38 	struct dlm_mhandle *mh;
39 	char *mb;
40 	int mb_len = sizeof(struct dlm_rcom) + len;
41 
42 	mh = dlm_lowcomms_get_buffer(to_nodeid, mb_len, GFP_NOFS, &mb);
43 	if (!mh) {
44 		log_print("create_rcom to %d type %d len %d ENOBUFS",
45 			  to_nodeid, type, len);
46 		return -ENOBUFS;
47 	}
48 	memset(mb, 0, mb_len);
49 
50 	rc = (struct dlm_rcom *) mb;
51 
52 	rc->rc_header.h_version = (DLM_HEADER_MAJOR | DLM_HEADER_MINOR);
53 	rc->rc_header.h_lockspace = ls->ls_global_id;
54 	rc->rc_header.h_nodeid = dlm_our_nodeid();
55 	rc->rc_header.h_length = mb_len;
56 	rc->rc_header.h_cmd = DLM_RCOM;
57 
58 	rc->rc_type = type;
59 
60 	spin_lock(&ls->ls_recover_lock);
61 	rc->rc_seq = ls->ls_recover_seq;
62 	spin_unlock(&ls->ls_recover_lock);
63 
64 	*mh_ret = mh;
65 	*rc_ret = rc;
66 	return 0;
67 }
68 
69 static void send_rcom(struct dlm_ls *ls, struct dlm_mhandle *mh,
70 		      struct dlm_rcom *rc)
71 {
72 	dlm_rcom_out(rc);
73 	dlm_lowcomms_commit_buffer(mh);
74 }
75 
76 static void set_rcom_status(struct dlm_ls *ls, struct rcom_status *rs,
77 			    uint32_t flags)
78 {
79 	rs->rs_flags = cpu_to_le32(flags);
80 }
81 
82 /* When replying to a status request, a node also sends back its
83    configuration values.  The requesting node then checks that the remote
84    node is configured the same way as itself. */
85 
86 static void set_rcom_config(struct dlm_ls *ls, struct rcom_config *rf,
87 			    uint32_t num_slots)
88 {
89 	rf->rf_lvblen = cpu_to_le32(ls->ls_lvblen);
90 	rf->rf_lsflags = cpu_to_le32(ls->ls_exflags);
91 
92 	rf->rf_our_slot = cpu_to_le16(ls->ls_slot);
93 	rf->rf_num_slots = cpu_to_le16(num_slots);
94 	rf->rf_generation =  cpu_to_le32(ls->ls_generation);
95 }
96 
97 static int check_rcom_config(struct dlm_ls *ls, struct dlm_rcom *rc, int nodeid)
98 {
99 	struct rcom_config *rf = (struct rcom_config *) rc->rc_buf;
100 
101 	if ((rc->rc_header.h_version & 0xFFFF0000) != DLM_HEADER_MAJOR) {
102 		log_error(ls, "version mismatch: %x nodeid %d: %x",
103 			  DLM_HEADER_MAJOR | DLM_HEADER_MINOR, nodeid,
104 			  rc->rc_header.h_version);
105 		return -EPROTO;
106 	}
107 
108 	if (le32_to_cpu(rf->rf_lvblen) != ls->ls_lvblen ||
109 	    le32_to_cpu(rf->rf_lsflags) != ls->ls_exflags) {
110 		log_error(ls, "config mismatch: %d,%x nodeid %d: %d,%x",
111 			  ls->ls_lvblen, ls->ls_exflags, nodeid,
112 			  le32_to_cpu(rf->rf_lvblen),
113 			  le32_to_cpu(rf->rf_lsflags));
114 		return -EPROTO;
115 	}
116 	return 0;
117 }
118 
119 static void allow_sync_reply(struct dlm_ls *ls, uint64_t *new_seq)
120 {
121 	spin_lock(&ls->ls_rcom_spin);
122 	*new_seq = ++ls->ls_rcom_seq;
123 	set_bit(LSFL_RCOM_WAIT, &ls->ls_flags);
124 	spin_unlock(&ls->ls_rcom_spin);
125 }
126 
127 static void disallow_sync_reply(struct dlm_ls *ls)
128 {
129 	spin_lock(&ls->ls_rcom_spin);
130 	clear_bit(LSFL_RCOM_WAIT, &ls->ls_flags);
131 	clear_bit(LSFL_RCOM_READY, &ls->ls_flags);
132 	spin_unlock(&ls->ls_rcom_spin);
133 }
134 
135 /*
136  * low nodeid gathers one slot value at a time from each node.
137  * it sets need_slots=0, and saves rf_our_slot returned from each
138  * rcom_config.
139  *
140  * other nodes gather all slot values at once from the low nodeid.
141  * they set need_slots=1, and ignore the rf_our_slot returned from each
142  * rcom_config.  they use the rf_num_slots returned from the low
143  * node's rcom_config.
144  */
145 
146 int dlm_rcom_status(struct dlm_ls *ls, int nodeid, uint32_t status_flags)
147 {
148 	struct dlm_rcom *rc;
149 	struct dlm_mhandle *mh;
150 	int error = 0;
151 
152 	ls->ls_recover_nodeid = nodeid;
153 
154 	if (nodeid == dlm_our_nodeid()) {
155 		rc = ls->ls_recover_buf;
156 		rc->rc_result = dlm_recover_status(ls);
157 		goto out;
158 	}
159 
160 	error = create_rcom(ls, nodeid, DLM_RCOM_STATUS,
161 			    sizeof(struct rcom_status), &rc, &mh);
162 	if (error)
163 		goto out;
164 
165 	set_rcom_status(ls, (struct rcom_status *)rc->rc_buf, status_flags);
166 
167 	allow_sync_reply(ls, &rc->rc_id);
168 	memset(ls->ls_recover_buf, 0, dlm_config.ci_buffer_size);
169 
170 	send_rcom(ls, mh, rc);
171 
172 	error = dlm_wait_function(ls, &rcom_response);
173 	disallow_sync_reply(ls);
174 	if (error)
175 		goto out;
176 
177 	rc = ls->ls_recover_buf;
178 
179 	if (rc->rc_result == -ESRCH) {
180 		/* we pretend the remote lockspace exists with 0 status */
181 		log_debug(ls, "remote node %d not ready", nodeid);
182 		rc->rc_result = 0;
183 		error = 0;
184 	} else {
185 		error = check_rcom_config(ls, rc, nodeid);
186 	}
187 
188 	/* the caller looks at rc_result for the remote recovery status */
189  out:
190 	return error;
191 }
192 
193 static void receive_rcom_status(struct dlm_ls *ls, struct dlm_rcom *rc_in)
194 {
195 	struct dlm_rcom *rc;
196 	struct dlm_mhandle *mh;
197 	struct rcom_status *rs;
198 	uint32_t status;
199 	int nodeid = rc_in->rc_header.h_nodeid;
200 	int len = sizeof(struct rcom_config);
201 	int num_slots = 0;
202 	int error;
203 
204 	if (!dlm_slots_version(&rc_in->rc_header)) {
205 		status = dlm_recover_status(ls);
206 		goto do_create;
207 	}
208 
209 	rs = (struct rcom_status *)rc_in->rc_buf;
210 
211 	if (!(rs->rs_flags & DLM_RSF_NEED_SLOTS)) {
212 		status = dlm_recover_status(ls);
213 		goto do_create;
214 	}
215 
216 	spin_lock(&ls->ls_recover_lock);
217 	status = ls->ls_recover_status;
218 	num_slots = ls->ls_num_slots;
219 	spin_unlock(&ls->ls_recover_lock);
220 	len += num_slots * sizeof(struct rcom_slot);
221 
222  do_create:
223 	error = create_rcom(ls, nodeid, DLM_RCOM_STATUS_REPLY,
224 			    len, &rc, &mh);
225 	if (error)
226 		return;
227 
228 	rc->rc_id = rc_in->rc_id;
229 	rc->rc_seq_reply = rc_in->rc_seq;
230 	rc->rc_result = status;
231 
232 	set_rcom_config(ls, (struct rcom_config *)rc->rc_buf, num_slots);
233 
234 	if (!num_slots)
235 		goto do_send;
236 
237 	spin_lock(&ls->ls_recover_lock);
238 	if (ls->ls_num_slots != num_slots) {
239 		spin_unlock(&ls->ls_recover_lock);
240 		log_debug(ls, "receive_rcom_status num_slots %d to %d",
241 			  num_slots, ls->ls_num_slots);
242 		rc->rc_result = 0;
243 		set_rcom_config(ls, (struct rcom_config *)rc->rc_buf, 0);
244 		goto do_send;
245 	}
246 
247 	dlm_slots_copy_out(ls, rc);
248 	spin_unlock(&ls->ls_recover_lock);
249 
250  do_send:
251 	send_rcom(ls, mh, rc);
252 }
253 
254 static void receive_sync_reply(struct dlm_ls *ls, struct dlm_rcom *rc_in)
255 {
256 	spin_lock(&ls->ls_rcom_spin);
257 	if (!test_bit(LSFL_RCOM_WAIT, &ls->ls_flags) ||
258 	    rc_in->rc_id != ls->ls_rcom_seq) {
259 		log_debug(ls, "reject reply %d from %d seq %llx expect %llx",
260 			  rc_in->rc_type, rc_in->rc_header.h_nodeid,
261 			  (unsigned long long)rc_in->rc_id,
262 			  (unsigned long long)ls->ls_rcom_seq);
263 		goto out;
264 	}
265 	memcpy(ls->ls_recover_buf, rc_in, rc_in->rc_header.h_length);
266 	set_bit(LSFL_RCOM_READY, &ls->ls_flags);
267 	clear_bit(LSFL_RCOM_WAIT, &ls->ls_flags);
268 	wake_up(&ls->ls_wait_general);
269  out:
270 	spin_unlock(&ls->ls_rcom_spin);
271 }
272 
273 int dlm_rcom_names(struct dlm_ls *ls, int nodeid, char *last_name, int last_len)
274 {
275 	struct dlm_rcom *rc;
276 	struct dlm_mhandle *mh;
277 	int error = 0;
278 	int max_size = dlm_config.ci_buffer_size - sizeof(struct dlm_rcom);
279 
280 	ls->ls_recover_nodeid = nodeid;
281 
282 	if (nodeid == dlm_our_nodeid()) {
283 		ls->ls_recover_buf->rc_header.h_length =
284 			dlm_config.ci_buffer_size;
285 		dlm_copy_master_names(ls, last_name, last_len,
286 		                      ls->ls_recover_buf->rc_buf,
287 		                      max_size, nodeid);
288 		goto out;
289 	}
290 
291 	error = create_rcom(ls, nodeid, DLM_RCOM_NAMES, last_len, &rc, &mh);
292 	if (error)
293 		goto out;
294 	memcpy(rc->rc_buf, last_name, last_len);
295 
296 	allow_sync_reply(ls, &rc->rc_id);
297 	memset(ls->ls_recover_buf, 0, dlm_config.ci_buffer_size);
298 
299 	send_rcom(ls, mh, rc);
300 
301 	error = dlm_wait_function(ls, &rcom_response);
302 	disallow_sync_reply(ls);
303  out:
304 	return error;
305 }
306 
307 static void receive_rcom_names(struct dlm_ls *ls, struct dlm_rcom *rc_in)
308 {
309 	struct dlm_rcom *rc;
310 	struct dlm_mhandle *mh;
311 	int error, inlen, outlen, nodeid;
312 
313 	nodeid = rc_in->rc_header.h_nodeid;
314 	inlen = rc_in->rc_header.h_length - sizeof(struct dlm_rcom);
315 	outlen = dlm_config.ci_buffer_size - sizeof(struct dlm_rcom);
316 
317 	error = create_rcom(ls, nodeid, DLM_RCOM_NAMES_REPLY, outlen, &rc, &mh);
318 	if (error)
319 		return;
320 	rc->rc_id = rc_in->rc_id;
321 	rc->rc_seq_reply = rc_in->rc_seq;
322 
323 	dlm_copy_master_names(ls, rc_in->rc_buf, inlen, rc->rc_buf, outlen,
324 			      nodeid);
325 	send_rcom(ls, mh, rc);
326 }
327 
328 int dlm_send_rcom_lookup(struct dlm_rsb *r, int dir_nodeid)
329 {
330 	struct dlm_rcom *rc;
331 	struct dlm_mhandle *mh;
332 	struct dlm_ls *ls = r->res_ls;
333 	int error;
334 
335 	error = create_rcom(ls, dir_nodeid, DLM_RCOM_LOOKUP, r->res_length,
336 			    &rc, &mh);
337 	if (error)
338 		goto out;
339 	memcpy(rc->rc_buf, r->res_name, r->res_length);
340 	rc->rc_id = (unsigned long) r;
341 
342 	send_rcom(ls, mh, rc);
343  out:
344 	return error;
345 }
346 
347 static void receive_rcom_lookup(struct dlm_ls *ls, struct dlm_rcom *rc_in)
348 {
349 	struct dlm_rcom *rc;
350 	struct dlm_mhandle *mh;
351 	int error, ret_nodeid, nodeid = rc_in->rc_header.h_nodeid;
352 	int len = rc_in->rc_header.h_length - sizeof(struct dlm_rcom);
353 
354 	error = create_rcom(ls, nodeid, DLM_RCOM_LOOKUP_REPLY, 0, &rc, &mh);
355 	if (error)
356 		return;
357 
358 	error = dlm_dir_lookup(ls, nodeid, rc_in->rc_buf, len, &ret_nodeid);
359 	if (error)
360 		ret_nodeid = error;
361 	rc->rc_result = ret_nodeid;
362 	rc->rc_id = rc_in->rc_id;
363 	rc->rc_seq_reply = rc_in->rc_seq;
364 
365 	send_rcom(ls, mh, rc);
366 }
367 
368 static void receive_rcom_lookup_reply(struct dlm_ls *ls, struct dlm_rcom *rc_in)
369 {
370 	dlm_recover_master_reply(ls, rc_in);
371 }
372 
373 static void pack_rcom_lock(struct dlm_rsb *r, struct dlm_lkb *lkb,
374 			   struct rcom_lock *rl)
375 {
376 	memset(rl, 0, sizeof(*rl));
377 
378 	rl->rl_ownpid = cpu_to_le32(lkb->lkb_ownpid);
379 	rl->rl_lkid = cpu_to_le32(lkb->lkb_id);
380 	rl->rl_exflags = cpu_to_le32(lkb->lkb_exflags);
381 	rl->rl_flags = cpu_to_le32(lkb->lkb_flags);
382 	rl->rl_lvbseq = cpu_to_le32(lkb->lkb_lvbseq);
383 	rl->rl_rqmode = lkb->lkb_rqmode;
384 	rl->rl_grmode = lkb->lkb_grmode;
385 	rl->rl_status = lkb->lkb_status;
386 	rl->rl_wait_type = cpu_to_le16(lkb->lkb_wait_type);
387 
388 	if (lkb->lkb_bastfn)
389 		rl->rl_asts |= DLM_CB_BAST;
390 	if (lkb->lkb_astfn)
391 		rl->rl_asts |= DLM_CB_CAST;
392 
393 	rl->rl_namelen = cpu_to_le16(r->res_length);
394 	memcpy(rl->rl_name, r->res_name, r->res_length);
395 
396 	/* FIXME: might we have an lvb without DLM_LKF_VALBLK set ?
397 	   If so, receive_rcom_lock_args() won't take this copy. */
398 
399 	if (lkb->lkb_lvbptr)
400 		memcpy(rl->rl_lvb, lkb->lkb_lvbptr, r->res_ls->ls_lvblen);
401 }
402 
403 int dlm_send_rcom_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
404 {
405 	struct dlm_ls *ls = r->res_ls;
406 	struct dlm_rcom *rc;
407 	struct dlm_mhandle *mh;
408 	struct rcom_lock *rl;
409 	int error, len = sizeof(struct rcom_lock);
410 
411 	if (lkb->lkb_lvbptr)
412 		len += ls->ls_lvblen;
413 
414 	error = create_rcom(ls, r->res_nodeid, DLM_RCOM_LOCK, len, &rc, &mh);
415 	if (error)
416 		goto out;
417 
418 	rl = (struct rcom_lock *) rc->rc_buf;
419 	pack_rcom_lock(r, lkb, rl);
420 	rc->rc_id = (unsigned long) r;
421 
422 	send_rcom(ls, mh, rc);
423  out:
424 	return error;
425 }
426 
427 /* needs at least dlm_rcom + rcom_lock */
428 static void receive_rcom_lock(struct dlm_ls *ls, struct dlm_rcom *rc_in)
429 {
430 	struct dlm_rcom *rc;
431 	struct dlm_mhandle *mh;
432 	int error, nodeid = rc_in->rc_header.h_nodeid;
433 
434 	dlm_recover_master_copy(ls, rc_in);
435 
436 	error = create_rcom(ls, nodeid, DLM_RCOM_LOCK_REPLY,
437 			    sizeof(struct rcom_lock), &rc, &mh);
438 	if (error)
439 		return;
440 
441 	/* We send back the same rcom_lock struct we received, but
442 	   dlm_recover_master_copy() has filled in rl_remid and rl_result */
443 
444 	memcpy(rc->rc_buf, rc_in->rc_buf, sizeof(struct rcom_lock));
445 	rc->rc_id = rc_in->rc_id;
446 	rc->rc_seq_reply = rc_in->rc_seq;
447 
448 	send_rcom(ls, mh, rc);
449 }
450 
451 /* If the lockspace doesn't exist then still send a status message
452    back; it's possible that it just doesn't have its global_id yet. */
453 
454 int dlm_send_ls_not_ready(int nodeid, struct dlm_rcom *rc_in)
455 {
456 	struct dlm_rcom *rc;
457 	struct rcom_config *rf;
458 	struct dlm_mhandle *mh;
459 	char *mb;
460 	int mb_len = sizeof(struct dlm_rcom) + sizeof(struct rcom_config);
461 
462 	mh = dlm_lowcomms_get_buffer(nodeid, mb_len, GFP_NOFS, &mb);
463 	if (!mh)
464 		return -ENOBUFS;
465 	memset(mb, 0, mb_len);
466 
467 	rc = (struct dlm_rcom *) mb;
468 
469 	rc->rc_header.h_version = (DLM_HEADER_MAJOR | DLM_HEADER_MINOR);
470 	rc->rc_header.h_lockspace = rc_in->rc_header.h_lockspace;
471 	rc->rc_header.h_nodeid = dlm_our_nodeid();
472 	rc->rc_header.h_length = mb_len;
473 	rc->rc_header.h_cmd = DLM_RCOM;
474 
475 	rc->rc_type = DLM_RCOM_STATUS_REPLY;
476 	rc->rc_id = rc_in->rc_id;
477 	rc->rc_seq_reply = rc_in->rc_seq;
478 	rc->rc_result = -ESRCH;
479 
480 	rf = (struct rcom_config *) rc->rc_buf;
481 	rf->rf_lvblen = cpu_to_le32(~0U);
482 
483 	dlm_rcom_out(rc);
484 	dlm_lowcomms_commit_buffer(mh);
485 
486 	return 0;
487 }
488 
489 static int is_old_reply(struct dlm_ls *ls, struct dlm_rcom *rc)
490 {
491 	uint64_t seq;
492 	int rv = 0;
493 
494 	switch (rc->rc_type) {
495 	case DLM_RCOM_STATUS_REPLY:
496 	case DLM_RCOM_NAMES_REPLY:
497 	case DLM_RCOM_LOOKUP_REPLY:
498 	case DLM_RCOM_LOCK_REPLY:
499 		spin_lock(&ls->ls_recover_lock);
500 		seq = ls->ls_recover_seq;
501 		spin_unlock(&ls->ls_recover_lock);
502 		if (rc->rc_seq_reply != seq) {
503 			log_debug(ls, "ignoring old reply %x from %d "
504 				      "seq_reply %llx expect %llx",
505 				      rc->rc_type, rc->rc_header.h_nodeid,
506 				      (unsigned long long)rc->rc_seq_reply,
507 				      (unsigned long long)seq);
508 			rv = 1;
509 		}
510 	}
511 	return rv;
512 }
513 
514 /* Called by dlm_recv; corresponds to dlm_receive_message() but special
515    recovery-only comms are sent through here. */
516 
517 void dlm_receive_rcom(struct dlm_ls *ls, struct dlm_rcom *rc, int nodeid)
518 {
519 	int lock_size = sizeof(struct dlm_rcom) + sizeof(struct rcom_lock);
520 
521 	if (dlm_recovery_stopped(ls) && (rc->rc_type != DLM_RCOM_STATUS)) {
522 		log_debug(ls, "ignoring recovery message %x from %d",
523 			  rc->rc_type, nodeid);
524 		goto out;
525 	}
526 
527 	if (is_old_reply(ls, rc))
528 		goto out;
529 
530 	switch (rc->rc_type) {
531 	case DLM_RCOM_STATUS:
532 		receive_rcom_status(ls, rc);
533 		break;
534 
535 	case DLM_RCOM_NAMES:
536 		receive_rcom_names(ls, rc);
537 		break;
538 
539 	case DLM_RCOM_LOOKUP:
540 		receive_rcom_lookup(ls, rc);
541 		break;
542 
543 	case DLM_RCOM_LOCK:
544 		if (rc->rc_header.h_length < lock_size)
545 			goto Eshort;
546 		receive_rcom_lock(ls, rc);
547 		break;
548 
549 	case DLM_RCOM_STATUS_REPLY:
550 		receive_sync_reply(ls, rc);
551 		break;
552 
553 	case DLM_RCOM_NAMES_REPLY:
554 		receive_sync_reply(ls, rc);
555 		break;
556 
557 	case DLM_RCOM_LOOKUP_REPLY:
558 		receive_rcom_lookup_reply(ls, rc);
559 		break;
560 
561 	case DLM_RCOM_LOCK_REPLY:
562 		if (rc->rc_header.h_length < lock_size)
563 			goto Eshort;
564 		dlm_recover_process_copy(ls, rc);
565 		break;
566 
567 	default:
568 		log_error(ls, "receive_rcom bad type %d", rc->rc_type);
569 	}
570 out:
571 	return;
572 Eshort:
573 	log_error(ls, "recovery message %x from %d is too short",
574 			  rc->rc_type, nodeid);
575 }
576 
577