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