1 // SPDX-License-Identifier: GPL-2.0-only
2 /******************************************************************************
3 *******************************************************************************
5 ** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
6 ** Copyright (C) 2005-2008 Red Hat, Inc. All rights reserved.
9 *******************************************************************************
10 ******************************************************************************/
12 #include "dlm_internal.h"
13 #include "lockspace.h"
25 static int rcom_response(struct dlm_ls
*ls
)
27 return test_bit(LSFL_RCOM_READY
, &ls
->ls_flags
);
30 static int create_rcom(struct dlm_ls
*ls
, int to_nodeid
, int type
, int len
,
31 struct dlm_rcom
**rc_ret
, struct dlm_mhandle
**mh_ret
)
34 struct dlm_mhandle
*mh
;
36 int mb_len
= sizeof(struct dlm_rcom
) + len
;
38 mh
= dlm_lowcomms_get_buffer(to_nodeid
, mb_len
, GFP_NOFS
, &mb
);
40 log_print("create_rcom to %d type %d len %d ENOBUFS",
41 to_nodeid
, type
, len
);
44 memset(mb
, 0, mb_len
);
46 rc
= (struct dlm_rcom
*) mb
;
48 rc
->rc_header
.h_version
= (DLM_HEADER_MAJOR
| DLM_HEADER_MINOR
);
49 rc
->rc_header
.h_lockspace
= ls
->ls_global_id
;
50 rc
->rc_header
.h_nodeid
= dlm_our_nodeid();
51 rc
->rc_header
.h_length
= mb_len
;
52 rc
->rc_header
.h_cmd
= DLM_RCOM
;
56 spin_lock(&ls
->ls_recover_lock
);
57 rc
->rc_seq
= ls
->ls_recover_seq
;
58 spin_unlock(&ls
->ls_recover_lock
);
65 static void send_rcom(struct dlm_ls
*ls
, struct dlm_mhandle
*mh
,
69 dlm_lowcomms_commit_buffer(mh
);
72 static void set_rcom_status(struct dlm_ls
*ls
, struct rcom_status
*rs
,
75 rs
->rs_flags
= cpu_to_le32(flags
);
78 /* When replying to a status request, a node also sends back its
79 configuration values. The requesting node then checks that the remote
80 node is configured the same way as itself. */
82 static void set_rcom_config(struct dlm_ls
*ls
, struct rcom_config
*rf
,
85 rf
->rf_lvblen
= cpu_to_le32(ls
->ls_lvblen
);
86 rf
->rf_lsflags
= cpu_to_le32(ls
->ls_exflags
);
88 rf
->rf_our_slot
= cpu_to_le16(ls
->ls_slot
);
89 rf
->rf_num_slots
= cpu_to_le16(num_slots
);
90 rf
->rf_generation
= cpu_to_le32(ls
->ls_generation
);
93 static int check_rcom_config(struct dlm_ls
*ls
, struct dlm_rcom
*rc
, int nodeid
)
95 struct rcom_config
*rf
= (struct rcom_config
*) rc
->rc_buf
;
97 if ((rc
->rc_header
.h_version
& 0xFFFF0000) != DLM_HEADER_MAJOR
) {
98 log_error(ls
, "version mismatch: %x nodeid %d: %x",
99 DLM_HEADER_MAJOR
| DLM_HEADER_MINOR
, nodeid
,
100 rc
->rc_header
.h_version
);
104 if (le32_to_cpu(rf
->rf_lvblen
) != ls
->ls_lvblen
||
105 le32_to_cpu(rf
->rf_lsflags
) != ls
->ls_exflags
) {
106 log_error(ls
, "config mismatch: %d,%x nodeid %d: %d,%x",
107 ls
->ls_lvblen
, ls
->ls_exflags
, nodeid
,
108 le32_to_cpu(rf
->rf_lvblen
),
109 le32_to_cpu(rf
->rf_lsflags
));
115 static void allow_sync_reply(struct dlm_ls
*ls
, uint64_t *new_seq
)
117 spin_lock(&ls
->ls_rcom_spin
);
118 *new_seq
= ++ls
->ls_rcom_seq
;
119 set_bit(LSFL_RCOM_WAIT
, &ls
->ls_flags
);
120 spin_unlock(&ls
->ls_rcom_spin
);
123 static void disallow_sync_reply(struct dlm_ls
*ls
)
125 spin_lock(&ls
->ls_rcom_spin
);
126 clear_bit(LSFL_RCOM_WAIT
, &ls
->ls_flags
);
127 clear_bit(LSFL_RCOM_READY
, &ls
->ls_flags
);
128 spin_unlock(&ls
->ls_rcom_spin
);
132 * low nodeid gathers one slot value at a time from each node.
133 * it sets need_slots=0, and saves rf_our_slot returned from each
136 * other nodes gather all slot values at once from the low nodeid.
137 * they set need_slots=1, and ignore the rf_our_slot returned from each
138 * rcom_config. they use the rf_num_slots returned from the low
139 * node's rcom_config.
142 int dlm_rcom_status(struct dlm_ls
*ls
, int nodeid
, uint32_t status_flags
)
145 struct dlm_mhandle
*mh
;
148 ls
->ls_recover_nodeid
= nodeid
;
150 if (nodeid
== dlm_our_nodeid()) {
151 rc
= ls
->ls_recover_buf
;
152 rc
->rc_result
= dlm_recover_status(ls
);
157 error
= create_rcom(ls
, nodeid
, DLM_RCOM_STATUS
,
158 sizeof(struct rcom_status
), &rc
, &mh
);
162 set_rcom_status(ls
, (struct rcom_status
*)rc
->rc_buf
, status_flags
);
164 allow_sync_reply(ls
, &rc
->rc_id
);
165 memset(ls
->ls_recover_buf
, 0, dlm_config
.ci_buffer_size
);
167 send_rcom(ls
, mh
, rc
);
169 error
= dlm_wait_function(ls
, &rcom_response
);
170 disallow_sync_reply(ls
);
171 if (error
== -ETIMEDOUT
)
176 rc
= ls
->ls_recover_buf
;
178 if (rc
->rc_result
== -ESRCH
) {
179 /* we pretend the remote lockspace exists with 0 status */
180 log_debug(ls
, "remote node %d not ready", nodeid
);
184 error
= check_rcom_config(ls
, rc
, nodeid
);
187 /* the caller looks at rc_result for the remote recovery status */
192 static void receive_rcom_status(struct dlm_ls
*ls
, struct dlm_rcom
*rc_in
)
195 struct dlm_mhandle
*mh
;
196 struct rcom_status
*rs
;
198 int nodeid
= rc_in
->rc_header
.h_nodeid
;
199 int len
= sizeof(struct rcom_config
);
203 if (!dlm_slots_version(&rc_in
->rc_header
)) {
204 status
= dlm_recover_status(ls
);
208 rs
= (struct rcom_status
*)rc_in
->rc_buf
;
210 if (!(le32_to_cpu(rs
->rs_flags
) & DLM_RSF_NEED_SLOTS
)) {
211 status
= dlm_recover_status(ls
);
215 spin_lock(&ls
->ls_recover_lock
);
216 status
= ls
->ls_recover_status
;
217 num_slots
= ls
->ls_num_slots
;
218 spin_unlock(&ls
->ls_recover_lock
);
219 len
+= num_slots
* sizeof(struct rcom_slot
);
222 error
= create_rcom(ls
, nodeid
, DLM_RCOM_STATUS_REPLY
,
227 rc
->rc_id
= rc_in
->rc_id
;
228 rc
->rc_seq_reply
= rc_in
->rc_seq
;
229 rc
->rc_result
= status
;
231 set_rcom_config(ls
, (struct rcom_config
*)rc
->rc_buf
, num_slots
);
236 spin_lock(&ls
->ls_recover_lock
);
237 if (ls
->ls_num_slots
!= num_slots
) {
238 spin_unlock(&ls
->ls_recover_lock
);
239 log_debug(ls
, "receive_rcom_status num_slots %d to %d",
240 num_slots
, ls
->ls_num_slots
);
242 set_rcom_config(ls
, (struct rcom_config
*)rc
->rc_buf
, 0);
246 dlm_slots_copy_out(ls
, rc
);
247 spin_unlock(&ls
->ls_recover_lock
);
250 send_rcom(ls
, mh
, rc
);
253 static void receive_sync_reply(struct dlm_ls
*ls
, struct dlm_rcom
*rc_in
)
255 spin_lock(&ls
->ls_rcom_spin
);
256 if (!test_bit(LSFL_RCOM_WAIT
, &ls
->ls_flags
) ||
257 rc_in
->rc_id
!= ls
->ls_rcom_seq
) {
258 log_debug(ls
, "reject reply %d from %d seq %llx expect %llx",
259 rc_in
->rc_type
, rc_in
->rc_header
.h_nodeid
,
260 (unsigned long long)rc_in
->rc_id
,
261 (unsigned long long)ls
->ls_rcom_seq
);
264 memcpy(ls
->ls_recover_buf
, rc_in
, rc_in
->rc_header
.h_length
);
265 set_bit(LSFL_RCOM_READY
, &ls
->ls_flags
);
266 clear_bit(LSFL_RCOM_WAIT
, &ls
->ls_flags
);
267 wake_up(&ls
->ls_wait_general
);
269 spin_unlock(&ls
->ls_rcom_spin
);
272 int dlm_rcom_names(struct dlm_ls
*ls
, int nodeid
, char *last_name
, int last_len
)
275 struct dlm_mhandle
*mh
;
278 ls
->ls_recover_nodeid
= nodeid
;
281 error
= create_rcom(ls
, nodeid
, DLM_RCOM_NAMES
, last_len
, &rc
, &mh
);
284 memcpy(rc
->rc_buf
, last_name
, last_len
);
286 allow_sync_reply(ls
, &rc
->rc_id
);
287 memset(ls
->ls_recover_buf
, 0, dlm_config
.ci_buffer_size
);
289 send_rcom(ls
, mh
, rc
);
291 error
= dlm_wait_function(ls
, &rcom_response
);
292 disallow_sync_reply(ls
);
293 if (error
== -ETIMEDOUT
)
299 static void receive_rcom_names(struct dlm_ls
*ls
, struct dlm_rcom
*rc_in
)
302 struct dlm_mhandle
*mh
;
303 int error
, inlen
, outlen
, nodeid
;
305 nodeid
= rc_in
->rc_header
.h_nodeid
;
306 inlen
= rc_in
->rc_header
.h_length
- sizeof(struct dlm_rcom
);
307 outlen
= dlm_config
.ci_buffer_size
- sizeof(struct dlm_rcom
);
309 error
= create_rcom(ls
, nodeid
, DLM_RCOM_NAMES_REPLY
, outlen
, &rc
, &mh
);
312 rc
->rc_id
= rc_in
->rc_id
;
313 rc
->rc_seq_reply
= rc_in
->rc_seq
;
315 dlm_copy_master_names(ls
, rc_in
->rc_buf
, inlen
, rc
->rc_buf
, outlen
,
317 send_rcom(ls
, mh
, rc
);
320 int dlm_send_rcom_lookup(struct dlm_rsb
*r
, int dir_nodeid
)
323 struct dlm_mhandle
*mh
;
324 struct dlm_ls
*ls
= r
->res_ls
;
327 error
= create_rcom(ls
, dir_nodeid
, DLM_RCOM_LOOKUP
, r
->res_length
,
331 memcpy(rc
->rc_buf
, r
->res_name
, r
->res_length
);
332 rc
->rc_id
= (unsigned long) r
->res_id
;
334 send_rcom(ls
, mh
, rc
);
339 static void receive_rcom_lookup(struct dlm_ls
*ls
, struct dlm_rcom
*rc_in
)
342 struct dlm_mhandle
*mh
;
343 int error
, ret_nodeid
, nodeid
= rc_in
->rc_header
.h_nodeid
;
344 int len
= rc_in
->rc_header
.h_length
- sizeof(struct dlm_rcom
);
346 error
= create_rcom(ls
, nodeid
, DLM_RCOM_LOOKUP_REPLY
, 0, &rc
, &mh
);
350 /* Old code would send this special id to trigger a debug dump. */
351 if (rc_in
->rc_id
== 0xFFFFFFFF) {
352 log_error(ls
, "receive_rcom_lookup dump from %d", nodeid
);
353 dlm_dump_rsb_name(ls
, rc_in
->rc_buf
, len
);
357 error
= dlm_master_lookup(ls
, nodeid
, rc_in
->rc_buf
, len
,
358 DLM_LU_RECOVER_MASTER
, &ret_nodeid
, NULL
);
361 rc
->rc_result
= ret_nodeid
;
362 rc
->rc_id
= rc_in
->rc_id
;
363 rc
->rc_seq_reply
= rc_in
->rc_seq
;
365 send_rcom(ls
, mh
, rc
);
368 static void receive_rcom_lookup_reply(struct dlm_ls
*ls
, struct dlm_rcom
*rc_in
)
370 dlm_recover_master_reply(ls
, rc_in
);
373 static void pack_rcom_lock(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
,
374 struct rcom_lock
*rl
)
376 memset(rl
, 0, sizeof(*rl
));
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
);
389 rl
->rl_asts
|= DLM_CB_BAST
;
391 rl
->rl_asts
|= DLM_CB_CAST
;
393 rl
->rl_namelen
= cpu_to_le16(r
->res_length
);
394 memcpy(rl
->rl_name
, r
->res_name
, r
->res_length
);
396 /* FIXME: might we have an lvb without DLM_LKF_VALBLK set ?
397 If so, receive_rcom_lock_args() won't take this copy. */
400 memcpy(rl
->rl_lvb
, lkb
->lkb_lvbptr
, r
->res_ls
->ls_lvblen
);
403 int dlm_send_rcom_lock(struct dlm_rsb
*r
, struct dlm_lkb
*lkb
)
405 struct dlm_ls
*ls
= r
->res_ls
;
407 struct dlm_mhandle
*mh
;
408 struct rcom_lock
*rl
;
409 int error
, len
= sizeof(struct rcom_lock
);
412 len
+= ls
->ls_lvblen
;
414 error
= create_rcom(ls
, r
->res_nodeid
, DLM_RCOM_LOCK
, len
, &rc
, &mh
);
418 rl
= (struct rcom_lock
*) rc
->rc_buf
;
419 pack_rcom_lock(r
, lkb
, rl
);
420 rc
->rc_id
= (unsigned long) r
;
422 send_rcom(ls
, mh
, rc
);
427 /* needs at least dlm_rcom + rcom_lock */
428 static void receive_rcom_lock(struct dlm_ls
*ls
, struct dlm_rcom
*rc_in
)
431 struct dlm_mhandle
*mh
;
432 int error
, nodeid
= rc_in
->rc_header
.h_nodeid
;
434 dlm_recover_master_copy(ls
, rc_in
);
436 error
= create_rcom(ls
, nodeid
, DLM_RCOM_LOCK_REPLY
,
437 sizeof(struct rcom_lock
), &rc
, &mh
);
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 */
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
;
448 send_rcom(ls
, mh
, rc
);
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. */
454 int dlm_send_ls_not_ready(int nodeid
, struct dlm_rcom
*rc_in
)
457 struct rcom_config
*rf
;
458 struct dlm_mhandle
*mh
;
460 int mb_len
= sizeof(struct dlm_rcom
) + sizeof(struct rcom_config
);
462 mh
= dlm_lowcomms_get_buffer(nodeid
, mb_len
, GFP_NOFS
, &mb
);
465 memset(mb
, 0, mb_len
);
467 rc
= (struct dlm_rcom
*) mb
;
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
;
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
;
480 rf
= (struct rcom_config
*) rc
->rc_buf
;
481 rf
->rf_lvblen
= cpu_to_le32(~0U);
484 dlm_lowcomms_commit_buffer(mh
);
490 * Ignore messages for stage Y before we set
491 * recover_status bit for stage X:
495 * dlm_recover_members()
498 * - ignore NAMES, NAMES_REPLY
499 * - ignore LOOKUP, LOOKUP_REPLY
500 * - ignore LOCK, LOCK_REPLY
502 * recover_status |= NODES
504 * dlm_recover_members_wait()
506 * dlm_recover_directory()
509 * - ignore LOOKUP, LOOKUP_REPLY
510 * - ignore LOCK, LOCK_REPLY
512 * recover_status |= DIR
514 * dlm_recover_directory_wait()
516 * dlm_recover_masters()
518 * - recv LOOKUP_REPLY
520 * dlm_recover_locks()
524 * recover_status |= LOCKS
526 * dlm_recover_locks_wait()
528 * recover_status |= DONE
531 /* Called by dlm_recv; corresponds to dlm_receive_message() but special
532 recovery-only comms are sent through here. */
534 void dlm_receive_rcom(struct dlm_ls
*ls
, struct dlm_rcom
*rc
, int nodeid
)
536 int lock_size
= sizeof(struct dlm_rcom
) + sizeof(struct rcom_lock
);
537 int stop
, reply
= 0, names
= 0, lookup
= 0, lock
= 0;
541 switch (rc
->rc_type
) {
542 case DLM_RCOM_STATUS_REPLY
:
548 case DLM_RCOM_NAMES_REPLY
:
552 case DLM_RCOM_LOOKUP
:
555 case DLM_RCOM_LOOKUP_REPLY
:
562 case DLM_RCOM_LOCK_REPLY
:
568 spin_lock(&ls
->ls_recover_lock
);
569 status
= ls
->ls_recover_status
;
570 stop
= test_bit(LSFL_RECOVER_STOP
, &ls
->ls_flags
);
571 seq
= ls
->ls_recover_seq
;
572 spin_unlock(&ls
->ls_recover_lock
);
574 if (stop
&& (rc
->rc_type
!= DLM_RCOM_STATUS
))
577 if (reply
&& (rc
->rc_seq_reply
!= seq
))
580 if (!(status
& DLM_RS_NODES
) && (names
|| lookup
|| lock
))
583 if (!(status
& DLM_RS_DIR
) && (lookup
|| lock
))
586 switch (rc
->rc_type
) {
587 case DLM_RCOM_STATUS
:
588 receive_rcom_status(ls
, rc
);
592 receive_rcom_names(ls
, rc
);
595 case DLM_RCOM_LOOKUP
:
596 receive_rcom_lookup(ls
, rc
);
600 if (rc
->rc_header
.h_length
< lock_size
)
602 receive_rcom_lock(ls
, rc
);
605 case DLM_RCOM_STATUS_REPLY
:
606 receive_sync_reply(ls
, rc
);
609 case DLM_RCOM_NAMES_REPLY
:
610 receive_sync_reply(ls
, rc
);
613 case DLM_RCOM_LOOKUP_REPLY
:
614 receive_rcom_lookup_reply(ls
, rc
);
617 case DLM_RCOM_LOCK_REPLY
:
618 if (rc
->rc_header
.h_length
< lock_size
)
620 dlm_recover_process_copy(ls
, rc
);
624 log_error(ls
, "receive_rcom bad type %d", rc
->rc_type
);
629 log_limit(ls
, "dlm_receive_rcom ignore msg %d "
630 "from %d %llu %llu recover seq %llu sts %x gen %u",
633 (unsigned long long)rc
->rc_seq
,
634 (unsigned long long)rc
->rc_seq_reply
,
635 (unsigned long long)seq
,
636 status
, ls
->ls_generation
);
639 log_error(ls
, "recovery message %d from %d is too short",
640 rc
->rc_type
, nodeid
);