Loading...
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
28static int rcom_response(struct dlm_ls *ls)
29{
30 return test_bit(LSFL_RCOM_READY, &ls->ls_flags);
31}
32
33static 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
68static 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
79static 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
85static 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
114static 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
122static 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
130int 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
171static 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
189static 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
208int 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
242static 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
263int 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
282static 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
303static 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
308static 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
338int 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 */
363static 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
389int 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
424static 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
452void 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 }
505out:
506 return;
507Eshort:
508 log_error(ls, "recovery message %x from %d is too short",
509 rc->rc_type, nodeid);
510}
511
1// SPDX-License-Identifier: GPL-2.0-only
2/******************************************************************************
3*******************************************************************************
4**
5** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
6** Copyright (C) 2005-2008 Red Hat, Inc. All rights reserved.
7**
8**
9*******************************************************************************
10******************************************************************************/
11
12#include "dlm_internal.h"
13#include "lockspace.h"
14#include "member.h"
15#include "lowcomms.h"
16#include "midcomms.h"
17#include "rcom.h"
18#include "recover.h"
19#include "dir.h"
20#include "config.h"
21#include "memory.h"
22#include "lock.h"
23#include "util.h"
24
25static int rcom_response(struct dlm_ls *ls)
26{
27 return test_bit(LSFL_RCOM_READY, &ls->ls_flags);
28}
29
30static void _create_rcom(struct dlm_ls *ls, int to_nodeid, int type, int len,
31 struct dlm_rcom **rc_ret, char *mb, int mb_len,
32 uint64_t seq)
33{
34 struct dlm_rcom *rc;
35
36 rc = (struct dlm_rcom *) mb;
37
38 rc->rc_header.h_version = cpu_to_le32(DLM_HEADER_MAJOR | DLM_HEADER_MINOR);
39 rc->rc_header.u.h_lockspace = cpu_to_le32(ls->ls_global_id);
40 rc->rc_header.h_nodeid = cpu_to_le32(dlm_our_nodeid());
41 rc->rc_header.h_length = cpu_to_le16(mb_len);
42 rc->rc_header.h_cmd = DLM_RCOM;
43
44 rc->rc_type = cpu_to_le32(type);
45 rc->rc_seq = cpu_to_le64(seq);
46
47 *rc_ret = rc;
48}
49
50static int create_rcom(struct dlm_ls *ls, int to_nodeid, int type, int len,
51 struct dlm_rcom **rc_ret, struct dlm_mhandle **mh_ret,
52 uint64_t seq)
53{
54 int mb_len = sizeof(struct dlm_rcom) + len;
55 struct dlm_mhandle *mh;
56 char *mb;
57
58 mh = dlm_midcomms_get_mhandle(to_nodeid, mb_len, GFP_NOFS, &mb);
59 if (!mh) {
60 log_print("%s to %d type %d len %d ENOBUFS",
61 __func__, to_nodeid, type, len);
62 return -ENOBUFS;
63 }
64
65 _create_rcom(ls, to_nodeid, type, len, rc_ret, mb, mb_len, seq);
66 *mh_ret = mh;
67 return 0;
68}
69
70static int create_rcom_stateless(struct dlm_ls *ls, int to_nodeid, int type,
71 int len, struct dlm_rcom **rc_ret,
72 struct dlm_msg **msg_ret, uint64_t seq)
73{
74 int mb_len = sizeof(struct dlm_rcom) + len;
75 struct dlm_msg *msg;
76 char *mb;
77
78 msg = dlm_lowcomms_new_msg(to_nodeid, mb_len, GFP_NOFS, &mb,
79 NULL, NULL);
80 if (!msg) {
81 log_print("create_rcom to %d type %d len %d ENOBUFS",
82 to_nodeid, type, len);
83 return -ENOBUFS;
84 }
85
86 _create_rcom(ls, to_nodeid, type, len, rc_ret, mb, mb_len, seq);
87 *msg_ret = msg;
88 return 0;
89}
90
91static void send_rcom(struct dlm_mhandle *mh, struct dlm_rcom *rc)
92{
93 dlm_midcomms_commit_mhandle(mh, NULL, 0);
94}
95
96static void send_rcom_stateless(struct dlm_msg *msg, struct dlm_rcom *rc)
97{
98 dlm_lowcomms_commit_msg(msg);
99 dlm_lowcomms_put_msg(msg);
100}
101
102static void set_rcom_status(struct dlm_ls *ls, struct rcom_status *rs,
103 uint32_t flags)
104{
105 rs->rs_flags = cpu_to_le32(flags);
106}
107
108/* When replying to a status request, a node also sends back its
109 configuration values. The requesting node then checks that the remote
110 node is configured the same way as itself. */
111
112static void set_rcom_config(struct dlm_ls *ls, struct rcom_config *rf,
113 uint32_t num_slots)
114{
115 rf->rf_lvblen = cpu_to_le32(ls->ls_lvblen);
116 rf->rf_lsflags = cpu_to_le32(ls->ls_exflags);
117
118 rf->rf_our_slot = cpu_to_le16(ls->ls_slot);
119 rf->rf_num_slots = cpu_to_le16(num_slots);
120 rf->rf_generation = cpu_to_le32(ls->ls_generation);
121}
122
123static int check_rcom_config(struct dlm_ls *ls, struct dlm_rcom *rc, int nodeid)
124{
125 struct rcom_config *rf = (struct rcom_config *) rc->rc_buf;
126
127 if ((le32_to_cpu(rc->rc_header.h_version) & 0xFFFF0000) != DLM_HEADER_MAJOR) {
128 log_error(ls, "version mismatch: %x nodeid %d: %x",
129 DLM_HEADER_MAJOR | DLM_HEADER_MINOR, nodeid,
130 le32_to_cpu(rc->rc_header.h_version));
131 return -EPROTO;
132 }
133
134 if (le32_to_cpu(rf->rf_lvblen) != ls->ls_lvblen ||
135 le32_to_cpu(rf->rf_lsflags) != ls->ls_exflags) {
136 log_error(ls, "config mismatch: %d,%x nodeid %d: %d,%x",
137 ls->ls_lvblen, ls->ls_exflags, nodeid,
138 le32_to_cpu(rf->rf_lvblen),
139 le32_to_cpu(rf->rf_lsflags));
140 return -EPROTO;
141 }
142 return 0;
143}
144
145static void allow_sync_reply(struct dlm_ls *ls, __le64 *new_seq)
146{
147 spin_lock(&ls->ls_rcom_spin);
148 *new_seq = cpu_to_le64(++ls->ls_rcom_seq);
149 set_bit(LSFL_RCOM_WAIT, &ls->ls_flags);
150 spin_unlock(&ls->ls_rcom_spin);
151}
152
153static void disallow_sync_reply(struct dlm_ls *ls)
154{
155 spin_lock(&ls->ls_rcom_spin);
156 clear_bit(LSFL_RCOM_WAIT, &ls->ls_flags);
157 clear_bit(LSFL_RCOM_READY, &ls->ls_flags);
158 spin_unlock(&ls->ls_rcom_spin);
159}
160
161/*
162 * low nodeid gathers one slot value at a time from each node.
163 * it sets need_slots=0, and saves rf_our_slot returned from each
164 * rcom_config.
165 *
166 * other nodes gather all slot values at once from the low nodeid.
167 * they set need_slots=1, and ignore the rf_our_slot returned from each
168 * rcom_config. they use the rf_num_slots returned from the low
169 * node's rcom_config.
170 */
171
172int dlm_rcom_status(struct dlm_ls *ls, int nodeid, uint32_t status_flags,
173 uint64_t seq)
174{
175 struct dlm_rcom *rc;
176 struct dlm_msg *msg;
177 int error = 0;
178
179 ls->ls_recover_nodeid = nodeid;
180
181 if (nodeid == dlm_our_nodeid()) {
182 rc = ls->ls_recover_buf;
183 rc->rc_result = cpu_to_le32(dlm_recover_status(ls));
184 goto out;
185 }
186
187retry:
188 error = create_rcom_stateless(ls, nodeid, DLM_RCOM_STATUS,
189 sizeof(struct rcom_status), &rc, &msg,
190 seq);
191 if (error)
192 goto out;
193
194 set_rcom_status(ls, (struct rcom_status *)rc->rc_buf, status_flags);
195
196 allow_sync_reply(ls, &rc->rc_id);
197 memset(ls->ls_recover_buf, 0, DLM_MAX_SOCKET_BUFSIZE);
198
199 send_rcom_stateless(msg, rc);
200
201 error = dlm_wait_function(ls, &rcom_response);
202 disallow_sync_reply(ls);
203 if (error == -ETIMEDOUT)
204 goto retry;
205 if (error)
206 goto out;
207
208 rc = ls->ls_recover_buf;
209
210 if (rc->rc_result == cpu_to_le32(-ESRCH)) {
211 /* we pretend the remote lockspace exists with 0 status */
212 log_debug(ls, "remote node %d not ready", nodeid);
213 rc->rc_result = 0;
214 error = 0;
215 } else {
216 error = check_rcom_config(ls, rc, nodeid);
217 }
218
219 /* the caller looks at rc_result for the remote recovery status */
220 out:
221 return error;
222}
223
224static void receive_rcom_status(struct dlm_ls *ls,
225 const struct dlm_rcom *rc_in,
226 uint64_t seq)
227{
228 struct dlm_rcom *rc;
229 struct rcom_status *rs;
230 uint32_t status;
231 int nodeid = le32_to_cpu(rc_in->rc_header.h_nodeid);
232 int len = sizeof(struct rcom_config);
233 struct dlm_msg *msg;
234 int num_slots = 0;
235 int error;
236
237 if (!dlm_slots_version(&rc_in->rc_header)) {
238 status = dlm_recover_status(ls);
239 goto do_create;
240 }
241
242 rs = (struct rcom_status *)rc_in->rc_buf;
243
244 if (!(le32_to_cpu(rs->rs_flags) & DLM_RSF_NEED_SLOTS)) {
245 status = dlm_recover_status(ls);
246 goto do_create;
247 }
248
249 spin_lock(&ls->ls_recover_lock);
250 status = ls->ls_recover_status;
251 num_slots = ls->ls_num_slots;
252 spin_unlock(&ls->ls_recover_lock);
253 len += num_slots * sizeof(struct rcom_slot);
254
255 do_create:
256 error = create_rcom_stateless(ls, nodeid, DLM_RCOM_STATUS_REPLY,
257 len, &rc, &msg, seq);
258 if (error)
259 return;
260
261 rc->rc_id = rc_in->rc_id;
262 rc->rc_seq_reply = rc_in->rc_seq;
263 rc->rc_result = cpu_to_le32(status);
264
265 set_rcom_config(ls, (struct rcom_config *)rc->rc_buf, num_slots);
266
267 if (!num_slots)
268 goto do_send;
269
270 spin_lock(&ls->ls_recover_lock);
271 if (ls->ls_num_slots != num_slots) {
272 spin_unlock(&ls->ls_recover_lock);
273 log_debug(ls, "receive_rcom_status num_slots %d to %d",
274 num_slots, ls->ls_num_slots);
275 rc->rc_result = 0;
276 set_rcom_config(ls, (struct rcom_config *)rc->rc_buf, 0);
277 goto do_send;
278 }
279
280 dlm_slots_copy_out(ls, rc);
281 spin_unlock(&ls->ls_recover_lock);
282
283 do_send:
284 send_rcom_stateless(msg, rc);
285}
286
287static void receive_sync_reply(struct dlm_ls *ls, const struct dlm_rcom *rc_in)
288{
289 spin_lock(&ls->ls_rcom_spin);
290 if (!test_bit(LSFL_RCOM_WAIT, &ls->ls_flags) ||
291 le64_to_cpu(rc_in->rc_id) != ls->ls_rcom_seq) {
292 log_debug(ls, "reject reply %d from %d seq %llx expect %llx",
293 le32_to_cpu(rc_in->rc_type),
294 le32_to_cpu(rc_in->rc_header.h_nodeid),
295 (unsigned long long)le64_to_cpu(rc_in->rc_id),
296 (unsigned long long)ls->ls_rcom_seq);
297 goto out;
298 }
299 memcpy(ls->ls_recover_buf, rc_in,
300 le16_to_cpu(rc_in->rc_header.h_length));
301 set_bit(LSFL_RCOM_READY, &ls->ls_flags);
302 clear_bit(LSFL_RCOM_WAIT, &ls->ls_flags);
303 wake_up(&ls->ls_wait_general);
304 out:
305 spin_unlock(&ls->ls_rcom_spin);
306}
307
308int dlm_rcom_names(struct dlm_ls *ls, int nodeid, char *last_name,
309 int last_len, uint64_t seq)
310{
311 struct dlm_mhandle *mh;
312 struct dlm_rcom *rc;
313 int error = 0;
314
315 ls->ls_recover_nodeid = nodeid;
316
317retry:
318 error = create_rcom(ls, nodeid, DLM_RCOM_NAMES, last_len,
319 &rc, &mh, seq);
320 if (error)
321 goto out;
322 memcpy(rc->rc_buf, last_name, last_len);
323
324 allow_sync_reply(ls, &rc->rc_id);
325 memset(ls->ls_recover_buf, 0, DLM_MAX_SOCKET_BUFSIZE);
326
327 send_rcom(mh, rc);
328
329 error = dlm_wait_function(ls, &rcom_response);
330 disallow_sync_reply(ls);
331 if (error == -ETIMEDOUT)
332 goto retry;
333 out:
334 return error;
335}
336
337static void receive_rcom_names(struct dlm_ls *ls, const struct dlm_rcom *rc_in,
338 uint64_t seq)
339{
340 struct dlm_mhandle *mh;
341 struct dlm_rcom *rc;
342 int error, inlen, outlen, nodeid;
343
344 nodeid = le32_to_cpu(rc_in->rc_header.h_nodeid);
345 inlen = le16_to_cpu(rc_in->rc_header.h_length) -
346 sizeof(struct dlm_rcom);
347 outlen = DLM_MAX_APP_BUFSIZE - sizeof(struct dlm_rcom);
348
349 error = create_rcom(ls, nodeid, DLM_RCOM_NAMES_REPLY, outlen,
350 &rc, &mh, seq);
351 if (error)
352 return;
353 rc->rc_id = rc_in->rc_id;
354 rc->rc_seq_reply = rc_in->rc_seq;
355
356 dlm_copy_master_names(ls, rc_in->rc_buf, inlen, rc->rc_buf, outlen,
357 nodeid);
358 send_rcom(mh, rc);
359}
360
361int dlm_send_rcom_lookup(struct dlm_rsb *r, int dir_nodeid, uint64_t seq)
362{
363 struct dlm_rcom *rc;
364 struct dlm_mhandle *mh;
365 struct dlm_ls *ls = r->res_ls;
366 int error;
367
368 error = create_rcom(ls, dir_nodeid, DLM_RCOM_LOOKUP, r->res_length,
369 &rc, &mh, seq);
370 if (error)
371 goto out;
372 memcpy(rc->rc_buf, r->res_name, r->res_length);
373 rc->rc_id = cpu_to_le64(r->res_id);
374
375 send_rcom(mh, rc);
376 out:
377 return error;
378}
379
380static void receive_rcom_lookup(struct dlm_ls *ls,
381 const struct dlm_rcom *rc_in, uint64_t seq)
382{
383 struct dlm_rcom *rc;
384 struct dlm_mhandle *mh;
385 int error, ret_nodeid, nodeid = le32_to_cpu(rc_in->rc_header.h_nodeid);
386 int len = le16_to_cpu(rc_in->rc_header.h_length) -
387 sizeof(struct dlm_rcom);
388
389 /* Old code would send this special id to trigger a debug dump. */
390 if (rc_in->rc_id == cpu_to_le64(0xFFFFFFFF)) {
391 log_error(ls, "receive_rcom_lookup dump from %d", nodeid);
392 dlm_dump_rsb_name(ls, rc_in->rc_buf, len);
393 return;
394 }
395
396 error = create_rcom(ls, nodeid, DLM_RCOM_LOOKUP_REPLY, 0, &rc, &mh,
397 seq);
398 if (error)
399 return;
400
401 error = dlm_master_lookup(ls, nodeid, rc_in->rc_buf, len,
402 DLM_LU_RECOVER_MASTER, &ret_nodeid, NULL);
403 if (error)
404 ret_nodeid = error;
405 rc->rc_result = cpu_to_le32(ret_nodeid);
406 rc->rc_id = rc_in->rc_id;
407 rc->rc_seq_reply = rc_in->rc_seq;
408
409 send_rcom(mh, rc);
410}
411
412static void receive_rcom_lookup_reply(struct dlm_ls *ls,
413 const struct dlm_rcom *rc_in)
414{
415 dlm_recover_master_reply(ls, rc_in);
416}
417
418static void pack_rcom_lock(struct dlm_rsb *r, struct dlm_lkb *lkb,
419 struct rcom_lock *rl)
420{
421 memset(rl, 0, sizeof(*rl));
422
423 rl->rl_ownpid = cpu_to_le32(lkb->lkb_ownpid);
424 rl->rl_lkid = cpu_to_le32(lkb->lkb_id);
425 rl->rl_exflags = cpu_to_le32(lkb->lkb_exflags);
426 rl->rl_flags = cpu_to_le32(dlm_dflags_val(lkb));
427 rl->rl_lvbseq = cpu_to_le32(lkb->lkb_lvbseq);
428 rl->rl_rqmode = lkb->lkb_rqmode;
429 rl->rl_grmode = lkb->lkb_grmode;
430 rl->rl_status = lkb->lkb_status;
431 rl->rl_wait_type = cpu_to_le16(lkb->lkb_wait_type);
432
433 if (lkb->lkb_bastfn)
434 rl->rl_asts |= DLM_CB_BAST;
435 if (lkb->lkb_astfn)
436 rl->rl_asts |= DLM_CB_CAST;
437
438 rl->rl_namelen = cpu_to_le16(r->res_length);
439 memcpy(rl->rl_name, r->res_name, r->res_length);
440
441 /* FIXME: might we have an lvb without DLM_LKF_VALBLK set ?
442 If so, receive_rcom_lock_args() won't take this copy. */
443
444 if (lkb->lkb_lvbptr)
445 memcpy(rl->rl_lvb, lkb->lkb_lvbptr, r->res_ls->ls_lvblen);
446}
447
448int dlm_send_rcom_lock(struct dlm_rsb *r, struct dlm_lkb *lkb, uint64_t seq)
449{
450 struct dlm_ls *ls = r->res_ls;
451 struct dlm_rcom *rc;
452 struct dlm_mhandle *mh;
453 struct rcom_lock *rl;
454 int error, len = sizeof(struct rcom_lock);
455
456 if (lkb->lkb_lvbptr)
457 len += ls->ls_lvblen;
458
459 error = create_rcom(ls, r->res_nodeid, DLM_RCOM_LOCK, len, &rc, &mh,
460 seq);
461 if (error)
462 goto out;
463
464 rl = (struct rcom_lock *) rc->rc_buf;
465 pack_rcom_lock(r, lkb, rl);
466 rc->rc_id = cpu_to_le64((uintptr_t)r);
467
468 send_rcom(mh, rc);
469 out:
470 return error;
471}
472
473/* needs at least dlm_rcom + rcom_lock */
474static void receive_rcom_lock(struct dlm_ls *ls, const struct dlm_rcom *rc_in,
475 uint64_t seq)
476{
477 __le32 rl_remid, rl_result;
478 struct rcom_lock *rl;
479 struct dlm_rcom *rc;
480 struct dlm_mhandle *mh;
481 int error, nodeid = le32_to_cpu(rc_in->rc_header.h_nodeid);
482
483 dlm_recover_master_copy(ls, rc_in, &rl_remid, &rl_result);
484
485 error = create_rcom(ls, nodeid, DLM_RCOM_LOCK_REPLY,
486 sizeof(struct rcom_lock), &rc, &mh, seq);
487 if (error)
488 return;
489
490 memcpy(rc->rc_buf, rc_in->rc_buf, sizeof(struct rcom_lock));
491 rl = (struct rcom_lock *)rc->rc_buf;
492 /* set rl_remid and rl_result from dlm_recover_master_copy() */
493 rl->rl_remid = rl_remid;
494 rl->rl_result = rl_result;
495
496 rc->rc_id = rc_in->rc_id;
497 rc->rc_seq_reply = rc_in->rc_seq;
498
499 send_rcom(mh, rc);
500}
501
502/* If the lockspace doesn't exist then still send a status message
503 back; it's possible that it just doesn't have its global_id yet. */
504
505int dlm_send_ls_not_ready(int nodeid, const struct dlm_rcom *rc_in)
506{
507 struct dlm_rcom *rc;
508 struct rcom_config *rf;
509 struct dlm_mhandle *mh;
510 char *mb;
511 int mb_len = sizeof(struct dlm_rcom) + sizeof(struct rcom_config);
512
513 mh = dlm_midcomms_get_mhandle(nodeid, mb_len, GFP_NOFS, &mb);
514 if (!mh)
515 return -ENOBUFS;
516
517 rc = (struct dlm_rcom *) mb;
518
519 rc->rc_header.h_version = cpu_to_le32(DLM_HEADER_MAJOR | DLM_HEADER_MINOR);
520 rc->rc_header.u.h_lockspace = rc_in->rc_header.u.h_lockspace;
521 rc->rc_header.h_nodeid = cpu_to_le32(dlm_our_nodeid());
522 rc->rc_header.h_length = cpu_to_le16(mb_len);
523 rc->rc_header.h_cmd = DLM_RCOM;
524
525 rc->rc_type = cpu_to_le32(DLM_RCOM_STATUS_REPLY);
526 rc->rc_id = rc_in->rc_id;
527 rc->rc_seq_reply = rc_in->rc_seq;
528 rc->rc_result = cpu_to_le32(-ESRCH);
529
530 rf = (struct rcom_config *) rc->rc_buf;
531 rf->rf_lvblen = cpu_to_le32(~0U);
532
533 dlm_midcomms_commit_mhandle(mh, NULL, 0);
534
535 return 0;
536}
537
538/*
539 * Ignore messages for stage Y before we set
540 * recover_status bit for stage X:
541 *
542 * recover_status = 0
543 *
544 * dlm_recover_members()
545 * - send nothing
546 * - recv nothing
547 * - ignore NAMES, NAMES_REPLY
548 * - ignore LOOKUP, LOOKUP_REPLY
549 * - ignore LOCK, LOCK_REPLY
550 *
551 * recover_status |= NODES
552 *
553 * dlm_recover_members_wait()
554 *
555 * dlm_recover_directory()
556 * - send NAMES
557 * - recv NAMES_REPLY
558 * - ignore LOOKUP, LOOKUP_REPLY
559 * - ignore LOCK, LOCK_REPLY
560 *
561 * recover_status |= DIR
562 *
563 * dlm_recover_directory_wait()
564 *
565 * dlm_recover_masters()
566 * - send LOOKUP
567 * - recv LOOKUP_REPLY
568 *
569 * dlm_recover_locks()
570 * - send LOCKS
571 * - recv LOCKS_REPLY
572 *
573 * recover_status |= LOCKS
574 *
575 * dlm_recover_locks_wait()
576 *
577 * recover_status |= DONE
578 */
579
580/* Called by dlm_recv; corresponds to dlm_receive_message() but special
581 recovery-only comms are sent through here. */
582
583void dlm_receive_rcom(struct dlm_ls *ls, const struct dlm_rcom *rc, int nodeid)
584{
585 int lock_size = sizeof(struct dlm_rcom) + sizeof(struct rcom_lock);
586 int stop, reply = 0, names = 0, lookup = 0, lock = 0;
587 uint32_t status;
588 uint64_t seq;
589
590 switch (rc->rc_type) {
591 case cpu_to_le32(DLM_RCOM_STATUS_REPLY):
592 reply = 1;
593 break;
594 case cpu_to_le32(DLM_RCOM_NAMES):
595 names = 1;
596 break;
597 case cpu_to_le32(DLM_RCOM_NAMES_REPLY):
598 names = 1;
599 reply = 1;
600 break;
601 case cpu_to_le32(DLM_RCOM_LOOKUP):
602 lookup = 1;
603 break;
604 case cpu_to_le32(DLM_RCOM_LOOKUP_REPLY):
605 lookup = 1;
606 reply = 1;
607 break;
608 case cpu_to_le32(DLM_RCOM_LOCK):
609 lock = 1;
610 break;
611 case cpu_to_le32(DLM_RCOM_LOCK_REPLY):
612 lock = 1;
613 reply = 1;
614 break;
615 }
616
617 spin_lock(&ls->ls_recover_lock);
618 status = ls->ls_recover_status;
619 stop = dlm_recovery_stopped(ls);
620 seq = ls->ls_recover_seq;
621 spin_unlock(&ls->ls_recover_lock);
622
623 if (stop && (rc->rc_type != cpu_to_le32(DLM_RCOM_STATUS)))
624 goto ignore;
625
626 if (reply && (le64_to_cpu(rc->rc_seq_reply) != seq))
627 goto ignore;
628
629 if (!(status & DLM_RS_NODES) && (names || lookup || lock))
630 goto ignore;
631
632 if (!(status & DLM_RS_DIR) && (lookup || lock))
633 goto ignore;
634
635 switch (rc->rc_type) {
636 case cpu_to_le32(DLM_RCOM_STATUS):
637 receive_rcom_status(ls, rc, seq);
638 break;
639
640 case cpu_to_le32(DLM_RCOM_NAMES):
641 receive_rcom_names(ls, rc, seq);
642 break;
643
644 case cpu_to_le32(DLM_RCOM_LOOKUP):
645 receive_rcom_lookup(ls, rc, seq);
646 break;
647
648 case cpu_to_le32(DLM_RCOM_LOCK):
649 if (le16_to_cpu(rc->rc_header.h_length) < lock_size)
650 goto Eshort;
651 receive_rcom_lock(ls, rc, seq);
652 break;
653
654 case cpu_to_le32(DLM_RCOM_STATUS_REPLY):
655 receive_sync_reply(ls, rc);
656 break;
657
658 case cpu_to_le32(DLM_RCOM_NAMES_REPLY):
659 receive_sync_reply(ls, rc);
660 break;
661
662 case cpu_to_le32(DLM_RCOM_LOOKUP_REPLY):
663 receive_rcom_lookup_reply(ls, rc);
664 break;
665
666 case cpu_to_le32(DLM_RCOM_LOCK_REPLY):
667 if (le16_to_cpu(rc->rc_header.h_length) < lock_size)
668 goto Eshort;
669 dlm_recover_process_copy(ls, rc, seq);
670 break;
671
672 default:
673 log_error(ls, "receive_rcom bad type %d",
674 le32_to_cpu(rc->rc_type));
675 }
676 return;
677
678ignore:
679 log_limit(ls, "dlm_receive_rcom ignore msg %d "
680 "from %d %llu %llu recover seq %llu sts %x gen %u",
681 le32_to_cpu(rc->rc_type),
682 nodeid,
683 (unsigned long long)le64_to_cpu(rc->rc_seq),
684 (unsigned long long)le64_to_cpu(rc->rc_seq_reply),
685 (unsigned long long)seq,
686 status, ls->ls_generation);
687 return;
688Eshort:
689 log_error(ls, "recovery message %d from %d is too short",
690 le32_to_cpu(rc->rc_type), nodeid);
691}
692