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);
45 rc = (struct dlm_rcom *) mb;
47 rc->rc_header.h_version = (DLM_HEADER_MAJOR | DLM_HEADER_MINOR);
48 rc->rc_header.h_lockspace = ls->ls_global_id;
49 rc->rc_header.h_nodeid = dlm_our_nodeid();
50 rc->rc_header.h_length = mb_len;
51 rc->rc_header.h_cmd = DLM_RCOM;
55 spin_lock(&ls->ls_recover_lock);
56 rc->rc_seq = ls->ls_recover_seq;
57 spin_unlock(&ls->ls_recover_lock);
64 static void send_rcom(struct dlm_ls *ls, struct dlm_mhandle *mh,
68 dlm_lowcomms_commit_buffer(mh);
71 static void set_rcom_status(struct dlm_ls *ls, struct rcom_status *rs,
74 rs->rs_flags = cpu_to_le32(flags);
77 /* When replying to a status request, a node also sends back its
78 configuration values. The requesting node then checks that the remote
79 node is configured the same way as itself. */
81 static void set_rcom_config(struct dlm_ls *ls, struct rcom_config *rf,
84 rf->rf_lvblen = cpu_to_le32(ls->ls_lvblen);
85 rf->rf_lsflags = cpu_to_le32(ls->ls_exflags);
87 rf->rf_our_slot = cpu_to_le16(ls->ls_slot);
88 rf->rf_num_slots = cpu_to_le16(num_slots);
89 rf->rf_generation = cpu_to_le32(ls->ls_generation);
92 static int check_rcom_config(struct dlm_ls *ls, struct dlm_rcom *rc, int nodeid)
94 struct rcom_config *rf = (struct rcom_config *) rc->rc_buf;
96 if ((rc->rc_header.h_version & 0xFFFF0000) != DLM_HEADER_MAJOR) {
97 log_error(ls, "version mismatch: %x nodeid %d: %x",
98 DLM_HEADER_MAJOR | DLM_HEADER_MINOR, nodeid,
99 rc->rc_header.h_version);
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));
114 static void allow_sync_reply(struct dlm_ls *ls, uint64_t *new_seq)
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);
122 static void disallow_sync_reply(struct dlm_ls *ls)
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);
131 * low nodeid gathers one slot value at a time from each node.
132 * it sets need_slots=0, and saves rf_our_slot returned from each
135 * other nodes gather all slot values at once from the low nodeid.
136 * they set need_slots=1, and ignore the rf_our_slot returned from each
137 * rcom_config. they use the rf_num_slots returned from the low
138 * node's rcom_config.
141 int dlm_rcom_status(struct dlm_ls *ls, int nodeid, uint32_t status_flags)
144 struct dlm_mhandle *mh;
147 ls->ls_recover_nodeid = nodeid;
149 if (nodeid == dlm_our_nodeid()) {
150 rc = ls->ls_recover_buf;
151 rc->rc_result = dlm_recover_status(ls);
156 error = create_rcom(ls, nodeid, DLM_RCOM_STATUS,
157 sizeof(struct rcom_status), &rc, &mh);
161 set_rcom_status(ls, (struct rcom_status *)rc->rc_buf, status_flags);
163 allow_sync_reply(ls, &rc->rc_id);
164 memset(ls->ls_recover_buf, 0, LOWCOMMS_MAX_TX_BUFFER_LEN);
166 send_rcom(ls, mh, rc);
168 error = dlm_wait_function(ls, &rcom_response);
169 disallow_sync_reply(ls);
170 if (error == -ETIMEDOUT)
175 rc = ls->ls_recover_buf;
177 if (rc->rc_result == -ESRCH) {
178 /* we pretend the remote lockspace exists with 0 status */
179 log_debug(ls, "remote node %d not ready", nodeid);
183 error = check_rcom_config(ls, rc, nodeid);
186 /* the caller looks at rc_result for the remote recovery status */
191 static void receive_rcom_status(struct dlm_ls *ls, struct dlm_rcom *rc_in)
194 struct dlm_mhandle *mh;
195 struct rcom_status *rs;
197 int nodeid = rc_in->rc_header.h_nodeid;
198 int len = sizeof(struct rcom_config);
202 if (!dlm_slots_version(&rc_in->rc_header)) {
203 status = dlm_recover_status(ls);
207 rs = (struct rcom_status *)rc_in->rc_buf;
209 if (!(le32_to_cpu(rs->rs_flags) & DLM_RSF_NEED_SLOTS)) {
210 status = dlm_recover_status(ls);
214 spin_lock(&ls->ls_recover_lock);
215 status = ls->ls_recover_status;
216 num_slots = ls->ls_num_slots;
217 spin_unlock(&ls->ls_recover_lock);
218 len += num_slots * sizeof(struct rcom_slot);
221 error = create_rcom(ls, nodeid, DLM_RCOM_STATUS_REPLY,
226 rc->rc_id = rc_in->rc_id;
227 rc->rc_seq_reply = rc_in->rc_seq;
228 rc->rc_result = status;
230 set_rcom_config(ls, (struct rcom_config *)rc->rc_buf, num_slots);
235 spin_lock(&ls->ls_recover_lock);
236 if (ls->ls_num_slots != num_slots) {
237 spin_unlock(&ls->ls_recover_lock);
238 log_debug(ls, "receive_rcom_status num_slots %d to %d",
239 num_slots, ls->ls_num_slots);
241 set_rcom_config(ls, (struct rcom_config *)rc->rc_buf, 0);
245 dlm_slots_copy_out(ls, rc);
246 spin_unlock(&ls->ls_recover_lock);
249 send_rcom(ls, mh, rc);
252 static void receive_sync_reply(struct dlm_ls *ls, struct dlm_rcom *rc_in)
254 spin_lock(&ls->ls_rcom_spin);
255 if (!test_bit(LSFL_RCOM_WAIT, &ls->ls_flags) ||
256 rc_in->rc_id != ls->ls_rcom_seq) {
257 log_debug(ls, "reject reply %d from %d seq %llx expect %llx",
258 rc_in->rc_type, rc_in->rc_header.h_nodeid,
259 (unsigned long long)rc_in->rc_id,
260 (unsigned long long)ls->ls_rcom_seq);
263 memcpy(ls->ls_recover_buf, rc_in, rc_in->rc_header.h_length);
264 set_bit(LSFL_RCOM_READY, &ls->ls_flags);
265 clear_bit(LSFL_RCOM_WAIT, &ls->ls_flags);
266 wake_up(&ls->ls_wait_general);
268 spin_unlock(&ls->ls_rcom_spin);
271 int dlm_rcom_names(struct dlm_ls *ls, int nodeid, char *last_name, int last_len)
274 struct dlm_mhandle *mh;
277 ls->ls_recover_nodeid = nodeid;
280 error = create_rcom(ls, nodeid, DLM_RCOM_NAMES, last_len, &rc, &mh);
283 memcpy(rc->rc_buf, last_name, last_len);
285 allow_sync_reply(ls, &rc->rc_id);
286 memset(ls->ls_recover_buf, 0, LOWCOMMS_MAX_TX_BUFFER_LEN);
288 send_rcom(ls, mh, rc);
290 error = dlm_wait_function(ls, &rcom_response);
291 disallow_sync_reply(ls);
292 if (error == -ETIMEDOUT)
298 static void receive_rcom_names(struct dlm_ls *ls, struct dlm_rcom *rc_in)
301 struct dlm_mhandle *mh;
302 int error, inlen, outlen, nodeid;
304 nodeid = rc_in->rc_header.h_nodeid;
305 inlen = rc_in->rc_header.h_length - sizeof(struct dlm_rcom);
306 outlen = LOWCOMMS_MAX_TX_BUFFER_LEN - sizeof(struct dlm_rcom);
308 error = create_rcom(ls, nodeid, DLM_RCOM_NAMES_REPLY, outlen, &rc, &mh);
311 rc->rc_id = rc_in->rc_id;
312 rc->rc_seq_reply = rc_in->rc_seq;
314 dlm_copy_master_names(ls, rc_in->rc_buf, inlen, rc->rc_buf, outlen,
316 send_rcom(ls, mh, rc);
319 int dlm_send_rcom_lookup(struct dlm_rsb *r, int dir_nodeid)
322 struct dlm_mhandle *mh;
323 struct dlm_ls *ls = r->res_ls;
326 error = create_rcom(ls, dir_nodeid, DLM_RCOM_LOOKUP, r->res_length,
330 memcpy(rc->rc_buf, r->res_name, r->res_length);
331 rc->rc_id = (unsigned long) r->res_id;
333 send_rcom(ls, mh, rc);
338 static void receive_rcom_lookup(struct dlm_ls *ls, struct dlm_rcom *rc_in)
341 struct dlm_mhandle *mh;
342 int error, ret_nodeid, nodeid = rc_in->rc_header.h_nodeid;
343 int len = rc_in->rc_header.h_length - sizeof(struct dlm_rcom);
345 error = create_rcom(ls, nodeid, DLM_RCOM_LOOKUP_REPLY, 0, &rc, &mh);
349 /* Old code would send this special id to trigger a debug dump. */
350 if (rc_in->rc_id == 0xFFFFFFFF) {
351 log_error(ls, "receive_rcom_lookup dump from %d", nodeid);
352 dlm_dump_rsb_name(ls, rc_in->rc_buf, len);
356 error = dlm_master_lookup(ls, nodeid, rc_in->rc_buf, len,
357 DLM_LU_RECOVER_MASTER, &ret_nodeid, NULL);
360 rc->rc_result = ret_nodeid;
361 rc->rc_id = rc_in->rc_id;
362 rc->rc_seq_reply = rc_in->rc_seq;
364 send_rcom(ls, mh, rc);
367 static void receive_rcom_lookup_reply(struct dlm_ls *ls, struct dlm_rcom *rc_in)
369 dlm_recover_master_reply(ls, rc_in);
372 static void pack_rcom_lock(struct dlm_rsb *r, struct dlm_lkb *lkb,
373 struct rcom_lock *rl)
375 memset(rl, 0, sizeof(*rl));
377 rl->rl_ownpid = cpu_to_le32(lkb->lkb_ownpid);
378 rl->rl_lkid = cpu_to_le32(lkb->lkb_id);
379 rl->rl_exflags = cpu_to_le32(lkb->lkb_exflags);
380 rl->rl_flags = cpu_to_le32(lkb->lkb_flags);
381 rl->rl_lvbseq = cpu_to_le32(lkb->lkb_lvbseq);
382 rl->rl_rqmode = lkb->lkb_rqmode;
383 rl->rl_grmode = lkb->lkb_grmode;
384 rl->rl_status = lkb->lkb_status;
385 rl->rl_wait_type = cpu_to_le16(lkb->lkb_wait_type);
388 rl->rl_asts |= DLM_CB_BAST;
390 rl->rl_asts |= DLM_CB_CAST;
392 rl->rl_namelen = cpu_to_le16(r->res_length);
393 memcpy(rl->rl_name, r->res_name, r->res_length);
395 /* FIXME: might we have an lvb without DLM_LKF_VALBLK set ?
396 If so, receive_rcom_lock_args() won't take this copy. */
399 memcpy(rl->rl_lvb, lkb->lkb_lvbptr, r->res_ls->ls_lvblen);
402 int dlm_send_rcom_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
404 struct dlm_ls *ls = r->res_ls;
406 struct dlm_mhandle *mh;
407 struct rcom_lock *rl;
408 int error, len = sizeof(struct rcom_lock);
411 len += ls->ls_lvblen;
413 error = create_rcom(ls, r->res_nodeid, DLM_RCOM_LOCK, len, &rc, &mh);
417 rl = (struct rcom_lock *) rc->rc_buf;
418 pack_rcom_lock(r, lkb, rl);
419 rc->rc_id = (unsigned long) r;
421 send_rcom(ls, mh, rc);
426 /* needs at least dlm_rcom + rcom_lock */
427 static void receive_rcom_lock(struct dlm_ls *ls, struct dlm_rcom *rc_in)
430 struct dlm_mhandle *mh;
431 int error, nodeid = rc_in->rc_header.h_nodeid;
433 dlm_recover_master_copy(ls, rc_in);
435 error = create_rcom(ls, nodeid, DLM_RCOM_LOCK_REPLY,
436 sizeof(struct rcom_lock), &rc, &mh);
440 /* We send back the same rcom_lock struct we received, but
441 dlm_recover_master_copy() has filled in rl_remid and rl_result */
443 memcpy(rc->rc_buf, rc_in->rc_buf, sizeof(struct rcom_lock));
444 rc->rc_id = rc_in->rc_id;
445 rc->rc_seq_reply = rc_in->rc_seq;
447 send_rcom(ls, mh, rc);
450 /* If the lockspace doesn't exist then still send a status message
451 back; it's possible that it just doesn't have its global_id yet. */
453 int dlm_send_ls_not_ready(int nodeid, struct dlm_rcom *rc_in)
456 struct rcom_config *rf;
457 struct dlm_mhandle *mh;
459 int mb_len = sizeof(struct dlm_rcom) + sizeof(struct rcom_config);
461 mh = dlm_lowcomms_get_buffer(nodeid, mb_len, GFP_NOFS, &mb);
465 rc = (struct dlm_rcom *) mb;
467 rc->rc_header.h_version = (DLM_HEADER_MAJOR | DLM_HEADER_MINOR);
468 rc->rc_header.h_lockspace = rc_in->rc_header.h_lockspace;
469 rc->rc_header.h_nodeid = dlm_our_nodeid();
470 rc->rc_header.h_length = mb_len;
471 rc->rc_header.h_cmd = DLM_RCOM;
473 rc->rc_type = DLM_RCOM_STATUS_REPLY;
474 rc->rc_id = rc_in->rc_id;
475 rc->rc_seq_reply = rc_in->rc_seq;
476 rc->rc_result = -ESRCH;
478 rf = (struct rcom_config *) rc->rc_buf;
479 rf->rf_lvblen = cpu_to_le32(~0U);
482 dlm_lowcomms_commit_buffer(mh);
488 * Ignore messages for stage Y before we set
489 * recover_status bit for stage X:
493 * dlm_recover_members()
496 * - ignore NAMES, NAMES_REPLY
497 * - ignore LOOKUP, LOOKUP_REPLY
498 * - ignore LOCK, LOCK_REPLY
500 * recover_status |= NODES
502 * dlm_recover_members_wait()
504 * dlm_recover_directory()
507 * - ignore LOOKUP, LOOKUP_REPLY
508 * - ignore LOCK, LOCK_REPLY
510 * recover_status |= DIR
512 * dlm_recover_directory_wait()
514 * dlm_recover_masters()
516 * - recv LOOKUP_REPLY
518 * dlm_recover_locks()
522 * recover_status |= LOCKS
524 * dlm_recover_locks_wait()
526 * recover_status |= DONE
529 /* Called by dlm_recv; corresponds to dlm_receive_message() but special
530 recovery-only comms are sent through here. */
532 void dlm_receive_rcom(struct dlm_ls *ls, struct dlm_rcom *rc, int nodeid)
534 int lock_size = sizeof(struct dlm_rcom) + sizeof(struct rcom_lock);
535 int stop, reply = 0, names = 0, lookup = 0, lock = 0;
539 switch (rc->rc_type) {
540 case DLM_RCOM_STATUS_REPLY:
546 case DLM_RCOM_NAMES_REPLY:
550 case DLM_RCOM_LOOKUP:
553 case DLM_RCOM_LOOKUP_REPLY:
560 case DLM_RCOM_LOCK_REPLY:
566 spin_lock(&ls->ls_recover_lock);
567 status = ls->ls_recover_status;
568 stop = test_bit(LSFL_RECOVER_STOP, &ls->ls_flags);
569 seq = ls->ls_recover_seq;
570 spin_unlock(&ls->ls_recover_lock);
572 if (stop && (rc->rc_type != DLM_RCOM_STATUS))
575 if (reply && (rc->rc_seq_reply != seq))
578 if (!(status & DLM_RS_NODES) && (names || lookup || lock))
581 if (!(status & DLM_RS_DIR) && (lookup || lock))
584 switch (rc->rc_type) {
585 case DLM_RCOM_STATUS:
586 receive_rcom_status(ls, rc);
590 receive_rcom_names(ls, rc);
593 case DLM_RCOM_LOOKUP:
594 receive_rcom_lookup(ls, rc);
598 if (rc->rc_header.h_length < lock_size)
600 receive_rcom_lock(ls, rc);
603 case DLM_RCOM_STATUS_REPLY:
604 receive_sync_reply(ls, rc);
607 case DLM_RCOM_NAMES_REPLY:
608 receive_sync_reply(ls, rc);
611 case DLM_RCOM_LOOKUP_REPLY:
612 receive_rcom_lookup_reply(ls, rc);
615 case DLM_RCOM_LOCK_REPLY:
616 if (rc->rc_header.h_length < lock_size)
618 dlm_recover_process_copy(ls, rc);
622 log_error(ls, "receive_rcom bad type %d", rc->rc_type);
627 log_limit(ls, "dlm_receive_rcom ignore msg %d "
628 "from %d %llu %llu recover seq %llu sts %x gen %u",
631 (unsigned long long)rc->rc_seq,
632 (unsigned long long)rc->rc_seq_reply,
633 (unsigned long long)seq,
634 status, ls->ls_generation);
637 log_error(ls, "recovery message %d from %d is too short",
638 rc->rc_type, nodeid);