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 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)
35 rc = (struct dlm_rcom *) mb;
37 rc->rc_header.h_version = (DLM_HEADER_MAJOR | DLM_HEADER_MINOR);
38 rc->rc_header.u.h_lockspace = ls->ls_global_id;
39 rc->rc_header.h_nodeid = dlm_our_nodeid();
40 rc->rc_header.h_length = mb_len;
41 rc->rc_header.h_cmd = DLM_RCOM;
45 spin_lock(&ls->ls_recover_lock);
46 rc->rc_seq = ls->ls_recover_seq;
47 spin_unlock(&ls->ls_recover_lock);
52 static int create_rcom(struct dlm_ls *ls, int to_nodeid, int type, int len,
53 struct dlm_rcom **rc_ret, struct dlm_mhandle **mh_ret)
55 int mb_len = sizeof(struct dlm_rcom) + len;
56 struct dlm_mhandle *mh;
59 mh = dlm_midcomms_get_mhandle(to_nodeid, mb_len, GFP_NOFS, &mb);
61 log_print("%s to %d type %d len %d ENOBUFS",
62 __func__, to_nodeid, type, len);
66 _create_rcom(ls, to_nodeid, type, len, rc_ret, mb, mb_len);
71 static int create_rcom_stateless(struct dlm_ls *ls, int to_nodeid, int type,
72 int len, struct dlm_rcom **rc_ret,
73 struct dlm_msg **msg_ret)
75 int mb_len = sizeof(struct dlm_rcom) + len;
79 msg = dlm_lowcomms_new_msg(to_nodeid, mb_len, GFP_NOFS, &mb,
82 log_print("create_rcom to %d type %d len %d ENOBUFS",
83 to_nodeid, type, len);
87 _create_rcom(ls, to_nodeid, type, len, rc_ret, mb, mb_len);
92 static void _send_rcom(struct dlm_ls *ls, struct dlm_rcom *rc)
97 static void send_rcom(struct dlm_ls *ls, struct dlm_mhandle *mh,
101 dlm_midcomms_commit_mhandle(mh);
104 static void send_rcom_stateless(struct dlm_ls *ls, struct dlm_msg *msg,
108 dlm_lowcomms_commit_msg(msg);
109 dlm_lowcomms_put_msg(msg);
112 static void set_rcom_status(struct dlm_ls *ls, struct rcom_status *rs,
115 rs->rs_flags = cpu_to_le32(flags);
118 /* When replying to a status request, a node also sends back its
119 configuration values. The requesting node then checks that the remote
120 node is configured the same way as itself. */
122 static void set_rcom_config(struct dlm_ls *ls, struct rcom_config *rf,
125 rf->rf_lvblen = cpu_to_le32(ls->ls_lvblen);
126 rf->rf_lsflags = cpu_to_le32(ls->ls_exflags);
128 rf->rf_our_slot = cpu_to_le16(ls->ls_slot);
129 rf->rf_num_slots = cpu_to_le16(num_slots);
130 rf->rf_generation = cpu_to_le32(ls->ls_generation);
133 static int check_rcom_config(struct dlm_ls *ls, struct dlm_rcom *rc, int nodeid)
135 struct rcom_config *rf = (struct rcom_config *) rc->rc_buf;
137 if ((rc->rc_header.h_version & 0xFFFF0000) != DLM_HEADER_MAJOR) {
138 log_error(ls, "version mismatch: %x nodeid %d: %x",
139 DLM_HEADER_MAJOR | DLM_HEADER_MINOR, nodeid,
140 rc->rc_header.h_version);
144 if (le32_to_cpu(rf->rf_lvblen) != ls->ls_lvblen ||
145 le32_to_cpu(rf->rf_lsflags) != ls->ls_exflags) {
146 log_error(ls, "config mismatch: %d,%x nodeid %d: %d,%x",
147 ls->ls_lvblen, ls->ls_exflags, nodeid,
148 le32_to_cpu(rf->rf_lvblen),
149 le32_to_cpu(rf->rf_lsflags));
155 static void allow_sync_reply(struct dlm_ls *ls, uint64_t *new_seq)
157 spin_lock(&ls->ls_rcom_spin);
158 *new_seq = ++ls->ls_rcom_seq;
159 set_bit(LSFL_RCOM_WAIT, &ls->ls_flags);
160 spin_unlock(&ls->ls_rcom_spin);
163 static void disallow_sync_reply(struct dlm_ls *ls)
165 spin_lock(&ls->ls_rcom_spin);
166 clear_bit(LSFL_RCOM_WAIT, &ls->ls_flags);
167 clear_bit(LSFL_RCOM_READY, &ls->ls_flags);
168 spin_unlock(&ls->ls_rcom_spin);
172 * low nodeid gathers one slot value at a time from each node.
173 * it sets need_slots=0, and saves rf_our_slot returned from each
176 * other nodes gather all slot values at once from the low nodeid.
177 * they set need_slots=1, and ignore the rf_our_slot returned from each
178 * rcom_config. they use the rf_num_slots returned from the low
179 * node's rcom_config.
182 int dlm_rcom_status(struct dlm_ls *ls, int nodeid, uint32_t status_flags)
188 ls->ls_recover_nodeid = nodeid;
190 if (nodeid == dlm_our_nodeid()) {
191 rc = ls->ls_recover_buf;
192 rc->rc_result = dlm_recover_status(ls);
197 error = create_rcom_stateless(ls, nodeid, DLM_RCOM_STATUS,
198 sizeof(struct rcom_status), &rc, &msg);
202 set_rcom_status(ls, (struct rcom_status *)rc->rc_buf, status_flags);
204 allow_sync_reply(ls, &rc->rc_id);
205 memset(ls->ls_recover_buf, 0, DLM_MAX_SOCKET_BUFSIZE);
207 send_rcom_stateless(ls, msg, rc);
209 error = dlm_wait_function(ls, &rcom_response);
210 disallow_sync_reply(ls);
211 if (error == -ETIMEDOUT)
216 rc = ls->ls_recover_buf;
218 if (rc->rc_result == -ESRCH) {
219 /* we pretend the remote lockspace exists with 0 status */
220 log_debug(ls, "remote node %d not ready", nodeid);
224 error = check_rcom_config(ls, rc, nodeid);
227 /* the caller looks at rc_result for the remote recovery status */
232 static void receive_rcom_status(struct dlm_ls *ls, struct dlm_rcom *rc_in)
235 struct rcom_status *rs;
237 int nodeid = rc_in->rc_header.h_nodeid;
238 int len = sizeof(struct rcom_config);
243 if (!dlm_slots_version(&rc_in->rc_header)) {
244 status = dlm_recover_status(ls);
248 rs = (struct rcom_status *)rc_in->rc_buf;
250 if (!(le32_to_cpu(rs->rs_flags) & DLM_RSF_NEED_SLOTS)) {
251 status = dlm_recover_status(ls);
255 spin_lock(&ls->ls_recover_lock);
256 status = ls->ls_recover_status;
257 num_slots = ls->ls_num_slots;
258 spin_unlock(&ls->ls_recover_lock);
259 len += num_slots * sizeof(struct rcom_slot);
262 error = create_rcom_stateless(ls, nodeid, DLM_RCOM_STATUS_REPLY,
267 rc->rc_id = rc_in->rc_id;
268 rc->rc_seq_reply = rc_in->rc_seq;
269 rc->rc_result = status;
271 set_rcom_config(ls, (struct rcom_config *)rc->rc_buf, num_slots);
276 spin_lock(&ls->ls_recover_lock);
277 if (ls->ls_num_slots != num_slots) {
278 spin_unlock(&ls->ls_recover_lock);
279 log_debug(ls, "receive_rcom_status num_slots %d to %d",
280 num_slots, ls->ls_num_slots);
282 set_rcom_config(ls, (struct rcom_config *)rc->rc_buf, 0);
286 dlm_slots_copy_out(ls, rc);
287 spin_unlock(&ls->ls_recover_lock);
290 send_rcom_stateless(ls, msg, rc);
293 static void receive_sync_reply(struct dlm_ls *ls, struct dlm_rcom *rc_in)
295 spin_lock(&ls->ls_rcom_spin);
296 if (!test_bit(LSFL_RCOM_WAIT, &ls->ls_flags) ||
297 rc_in->rc_id != ls->ls_rcom_seq) {
298 log_debug(ls, "reject reply %d from %d seq %llx expect %llx",
299 rc_in->rc_type, rc_in->rc_header.h_nodeid,
300 (unsigned long long)rc_in->rc_id,
301 (unsigned long long)ls->ls_rcom_seq);
304 memcpy(ls->ls_recover_buf, rc_in, rc_in->rc_header.h_length);
305 set_bit(LSFL_RCOM_READY, &ls->ls_flags);
306 clear_bit(LSFL_RCOM_WAIT, &ls->ls_flags);
307 wake_up(&ls->ls_wait_general);
309 spin_unlock(&ls->ls_rcom_spin);
312 int dlm_rcom_names(struct dlm_ls *ls, int nodeid, char *last_name, int last_len)
318 ls->ls_recover_nodeid = nodeid;
321 error = create_rcom_stateless(ls, nodeid, DLM_RCOM_NAMES, last_len,
325 memcpy(rc->rc_buf, last_name, last_len);
327 allow_sync_reply(ls, &rc->rc_id);
328 memset(ls->ls_recover_buf, 0, DLM_MAX_SOCKET_BUFSIZE);
330 send_rcom_stateless(ls, msg, rc);
332 error = dlm_wait_function(ls, &rcom_response);
333 disallow_sync_reply(ls);
334 if (error == -ETIMEDOUT)
340 static void receive_rcom_names(struct dlm_ls *ls, struct dlm_rcom *rc_in)
343 int error, inlen, outlen, nodeid;
346 nodeid = rc_in->rc_header.h_nodeid;
347 inlen = rc_in->rc_header.h_length - sizeof(struct dlm_rcom);
348 outlen = DLM_MAX_APP_BUFSIZE - sizeof(struct dlm_rcom);
350 error = create_rcom_stateless(ls, nodeid, DLM_RCOM_NAMES_REPLY, outlen,
354 rc->rc_id = rc_in->rc_id;
355 rc->rc_seq_reply = rc_in->rc_seq;
357 dlm_copy_master_names(ls, rc_in->rc_buf, inlen, rc->rc_buf, outlen,
359 send_rcom_stateless(ls, msg, rc);
362 int dlm_send_rcom_lookup(struct dlm_rsb *r, int dir_nodeid)
365 struct dlm_mhandle *mh;
366 struct dlm_ls *ls = r->res_ls;
369 error = create_rcom(ls, dir_nodeid, DLM_RCOM_LOOKUP, r->res_length,
373 memcpy(rc->rc_buf, r->res_name, r->res_length);
374 rc->rc_id = (unsigned long) r->res_id;
376 send_rcom(ls, mh, rc);
381 static void receive_rcom_lookup(struct dlm_ls *ls, struct dlm_rcom *rc_in)
384 struct dlm_mhandle *mh;
385 int error, ret_nodeid, nodeid = rc_in->rc_header.h_nodeid;
386 int len = rc_in->rc_header.h_length - sizeof(struct dlm_rcom);
388 /* Old code would send this special id to trigger a debug dump. */
389 if (rc_in->rc_id == 0xFFFFFFFF) {
390 log_error(ls, "receive_rcom_lookup dump from %d", nodeid);
391 dlm_dump_rsb_name(ls, rc_in->rc_buf, len);
395 error = create_rcom(ls, nodeid, DLM_RCOM_LOOKUP_REPLY, 0, &rc, &mh);
399 error = dlm_master_lookup(ls, nodeid, rc_in->rc_buf, len,
400 DLM_LU_RECOVER_MASTER, &ret_nodeid, NULL);
403 rc->rc_result = ret_nodeid;
404 rc->rc_id = rc_in->rc_id;
405 rc->rc_seq_reply = rc_in->rc_seq;
407 send_rcom(ls, mh, rc);
410 static void receive_rcom_lookup_reply(struct dlm_ls *ls, struct dlm_rcom *rc_in)
412 dlm_recover_master_reply(ls, rc_in);
415 static void pack_rcom_lock(struct dlm_rsb *r, struct dlm_lkb *lkb,
416 struct rcom_lock *rl)
418 memset(rl, 0, sizeof(*rl));
420 rl->rl_ownpid = cpu_to_le32(lkb->lkb_ownpid);
421 rl->rl_lkid = cpu_to_le32(lkb->lkb_id);
422 rl->rl_exflags = cpu_to_le32(lkb->lkb_exflags);
423 rl->rl_flags = cpu_to_le32(lkb->lkb_flags);
424 rl->rl_lvbseq = cpu_to_le32(lkb->lkb_lvbseq);
425 rl->rl_rqmode = lkb->lkb_rqmode;
426 rl->rl_grmode = lkb->lkb_grmode;
427 rl->rl_status = lkb->lkb_status;
428 rl->rl_wait_type = cpu_to_le16(lkb->lkb_wait_type);
431 rl->rl_asts |= DLM_CB_BAST;
433 rl->rl_asts |= DLM_CB_CAST;
435 rl->rl_namelen = cpu_to_le16(r->res_length);
436 memcpy(rl->rl_name, r->res_name, r->res_length);
438 /* FIXME: might we have an lvb without DLM_LKF_VALBLK set ?
439 If so, receive_rcom_lock_args() won't take this copy. */
442 memcpy(rl->rl_lvb, lkb->lkb_lvbptr, r->res_ls->ls_lvblen);
445 int dlm_send_rcom_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
447 struct dlm_ls *ls = r->res_ls;
449 struct dlm_mhandle *mh;
450 struct rcom_lock *rl;
451 int error, len = sizeof(struct rcom_lock);
454 len += ls->ls_lvblen;
456 error = create_rcom(ls, r->res_nodeid, DLM_RCOM_LOCK, len, &rc, &mh);
460 rl = (struct rcom_lock *) rc->rc_buf;
461 pack_rcom_lock(r, lkb, rl);
462 rc->rc_id = (unsigned long) r;
464 send_rcom(ls, mh, rc);
469 /* needs at least dlm_rcom + rcom_lock */
470 static void receive_rcom_lock(struct dlm_ls *ls, struct dlm_rcom *rc_in)
473 struct dlm_mhandle *mh;
474 int error, nodeid = rc_in->rc_header.h_nodeid;
476 dlm_recover_master_copy(ls, rc_in);
478 error = create_rcom(ls, nodeid, DLM_RCOM_LOCK_REPLY,
479 sizeof(struct rcom_lock), &rc, &mh);
483 /* We send back the same rcom_lock struct we received, but
484 dlm_recover_master_copy() has filled in rl_remid and rl_result */
486 memcpy(rc->rc_buf, rc_in->rc_buf, sizeof(struct rcom_lock));
487 rc->rc_id = rc_in->rc_id;
488 rc->rc_seq_reply = rc_in->rc_seq;
490 send_rcom(ls, mh, rc);
493 /* If the lockspace doesn't exist then still send a status message
494 back; it's possible that it just doesn't have its global_id yet. */
496 int dlm_send_ls_not_ready(int nodeid, struct dlm_rcom *rc_in)
499 struct rcom_config *rf;
500 struct dlm_mhandle *mh;
502 int mb_len = sizeof(struct dlm_rcom) + sizeof(struct rcom_config);
504 mh = dlm_midcomms_get_mhandle(nodeid, mb_len, GFP_NOFS, &mb);
508 rc = (struct dlm_rcom *) mb;
510 rc->rc_header.h_version = (DLM_HEADER_MAJOR | DLM_HEADER_MINOR);
511 rc->rc_header.u.h_lockspace = rc_in->rc_header.u.h_lockspace;
512 rc->rc_header.h_nodeid = dlm_our_nodeid();
513 rc->rc_header.h_length = mb_len;
514 rc->rc_header.h_cmd = DLM_RCOM;
516 rc->rc_type = DLM_RCOM_STATUS_REPLY;
517 rc->rc_id = rc_in->rc_id;
518 rc->rc_seq_reply = rc_in->rc_seq;
519 rc->rc_result = -ESRCH;
521 rf = (struct rcom_config *) rc->rc_buf;
522 rf->rf_lvblen = cpu_to_le32(~0U);
525 dlm_midcomms_commit_mhandle(mh);
531 * Ignore messages for stage Y before we set
532 * recover_status bit for stage X:
536 * dlm_recover_members()
539 * - ignore NAMES, NAMES_REPLY
540 * - ignore LOOKUP, LOOKUP_REPLY
541 * - ignore LOCK, LOCK_REPLY
543 * recover_status |= NODES
545 * dlm_recover_members_wait()
547 * dlm_recover_directory()
550 * - ignore LOOKUP, LOOKUP_REPLY
551 * - ignore LOCK, LOCK_REPLY
553 * recover_status |= DIR
555 * dlm_recover_directory_wait()
557 * dlm_recover_masters()
559 * - recv LOOKUP_REPLY
561 * dlm_recover_locks()
565 * recover_status |= LOCKS
567 * dlm_recover_locks_wait()
569 * recover_status |= DONE
572 /* Called by dlm_recv; corresponds to dlm_receive_message() but special
573 recovery-only comms are sent through here. */
575 void dlm_receive_rcom(struct dlm_ls *ls, struct dlm_rcom *rc, int nodeid)
577 int lock_size = sizeof(struct dlm_rcom) + sizeof(struct rcom_lock);
578 int stop, reply = 0, names = 0, lookup = 0, lock = 0;
582 switch (rc->rc_type) {
583 case DLM_RCOM_STATUS_REPLY:
589 case DLM_RCOM_NAMES_REPLY:
593 case DLM_RCOM_LOOKUP:
596 case DLM_RCOM_LOOKUP_REPLY:
603 case DLM_RCOM_LOCK_REPLY:
609 spin_lock(&ls->ls_recover_lock);
610 status = ls->ls_recover_status;
611 stop = test_bit(LSFL_RECOVER_STOP, &ls->ls_flags);
612 seq = ls->ls_recover_seq;
613 spin_unlock(&ls->ls_recover_lock);
615 if (stop && (rc->rc_type != DLM_RCOM_STATUS))
618 if (reply && (rc->rc_seq_reply != seq))
621 if (!(status & DLM_RS_NODES) && (names || lookup || lock))
624 if (!(status & DLM_RS_DIR) && (lookup || lock))
627 switch (rc->rc_type) {
628 case DLM_RCOM_STATUS:
629 receive_rcom_status(ls, rc);
633 receive_rcom_names(ls, rc);
636 case DLM_RCOM_LOOKUP:
637 receive_rcom_lookup(ls, rc);
641 if (rc->rc_header.h_length < lock_size)
643 receive_rcom_lock(ls, rc);
646 case DLM_RCOM_STATUS_REPLY:
647 receive_sync_reply(ls, rc);
650 case DLM_RCOM_NAMES_REPLY:
651 receive_sync_reply(ls, rc);
654 case DLM_RCOM_LOOKUP_REPLY:
655 receive_rcom_lookup_reply(ls, rc);
658 case DLM_RCOM_LOCK_REPLY:
659 if (rc->rc_header.h_length < lock_size)
661 dlm_recover_process_copy(ls, rc);
665 log_error(ls, "receive_rcom bad type %d", rc->rc_type);
670 log_limit(ls, "dlm_receive_rcom ignore msg %d "
671 "from %d %llu %llu recover seq %llu sts %x gen %u",
674 (unsigned long long)rc->rc_seq,
675 (unsigned long long)rc->rc_seq_reply,
676 (unsigned long long)seq,
677 status, ls->ls_generation);
680 log_error(ls, "recovery message %d from %d is too short",
681 rc->rc_type, nodeid);