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/******************************************************************************
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
29static int rcom_response(struct dlm_ls *ls)
30{
31 return test_bit(LSFL_RCOM_READY, &ls->ls_flags);
32}
33
34static 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
69static 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
76static 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
86static 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
97static 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
119static 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
127static 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
146int 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
193static 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
254static 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
273int 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
307static 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
328int 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
347static 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
368static 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
373static 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
403int 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 */
428static 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
454int 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/* Called by dlm_recv; corresponds to dlm_receive_message() but special
490 recovery-only comms are sent through here. */
491
492void dlm_receive_rcom(struct dlm_ls *ls, struct dlm_rcom *rc, int nodeid)
493{
494 int lock_size = sizeof(struct dlm_rcom) + sizeof(struct rcom_lock);
495 int stop, reply = 0, lock = 0;
496 uint32_t status;
497 uint64_t seq;
498
499 switch (rc->rc_type) {
500 case DLM_RCOM_LOCK:
501 lock = 1;
502 break;
503 case DLM_RCOM_LOCK_REPLY:
504 lock = 1;
505 reply = 1;
506 break;
507 case DLM_RCOM_STATUS_REPLY:
508 case DLM_RCOM_NAMES_REPLY:
509 case DLM_RCOM_LOOKUP_REPLY:
510 reply = 1;
511 };
512
513 spin_lock(&ls->ls_recover_lock);
514 status = ls->ls_recover_status;
515 stop = test_bit(LSFL_RECOVERY_STOP, &ls->ls_flags);
516 seq = ls->ls_recover_seq;
517 spin_unlock(&ls->ls_recover_lock);
518
519 if ((stop && (rc->rc_type != DLM_RCOM_STATUS)) ||
520 (reply && (rc->rc_seq_reply != seq)) ||
521 (lock && !(status & DLM_RS_DIR))) {
522 log_limit(ls, "dlm_receive_rcom ignore msg %d "
523 "from %d %llu %llu recover seq %llu sts %x gen %u",
524 rc->rc_type,
525 nodeid,
526 (unsigned long long)rc->rc_seq,
527 (unsigned long long)rc->rc_seq_reply,
528 (unsigned long long)seq,
529 status, ls->ls_generation);
530 goto out;
531 }
532
533 switch (rc->rc_type) {
534 case DLM_RCOM_STATUS:
535 receive_rcom_status(ls, rc);
536 break;
537
538 case DLM_RCOM_NAMES:
539 receive_rcom_names(ls, rc);
540 break;
541
542 case DLM_RCOM_LOOKUP:
543 receive_rcom_lookup(ls, rc);
544 break;
545
546 case DLM_RCOM_LOCK:
547 if (rc->rc_header.h_length < lock_size)
548 goto Eshort;
549 receive_rcom_lock(ls, rc);
550 break;
551
552 case DLM_RCOM_STATUS_REPLY:
553 receive_sync_reply(ls, rc);
554 break;
555
556 case DLM_RCOM_NAMES_REPLY:
557 receive_sync_reply(ls, rc);
558 break;
559
560 case DLM_RCOM_LOOKUP_REPLY:
561 receive_rcom_lookup_reply(ls, rc);
562 break;
563
564 case DLM_RCOM_LOCK_REPLY:
565 if (rc->rc_header.h_length < lock_size)
566 goto Eshort;
567 dlm_recover_process_copy(ls, rc);
568 break;
569
570 default:
571 log_error(ls, "receive_rcom bad type %d", rc->rc_type);
572 }
573out:
574 return;
575Eshort:
576 log_error(ls, "recovery message %x from %d is too short",
577 rc->rc_type, nodeid);
578}
579