1 /******************************************************************************
2 *******************************************************************************
4 ** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
5 ** Copyright (C) 2005-2008 Red Hat, Inc. All rights reserved.
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.
11 *******************************************************************************
12 ******************************************************************************/
14 #include "dlm_internal.h"
15 #include "lockspace.h"
27 static int rcom_response(struct dlm_ls *ls)
29 return test_bit(LSFL_RCOM_READY, &ls->ls_flags);
32 static int create_rcom(struct dlm_ls *ls, int to_nodeid, int type, int len,
33 struct dlm_rcom **rc_ret, struct dlm_mhandle **mh_ret)
36 struct dlm_mhandle *mh;
38 int mb_len = sizeof(struct dlm_rcom) + len;
40 mh = dlm_lowcomms_get_buffer(to_nodeid, mb_len, GFP_NOFS, &mb);
42 log_print("create_rcom to %d type %d len %d ENOBUFS",
43 to_nodeid, type, len);
46 memset(mb, 0, mb_len);
48 rc = (struct dlm_rcom *) mb;
50 rc->rc_header.h_version = (DLM_HEADER_MAJOR | DLM_HEADER_MINOR);
51 rc->rc_header.h_lockspace = ls->ls_global_id;
52 rc->rc_header.h_nodeid = dlm_our_nodeid();
53 rc->rc_header.h_length = mb_len;
54 rc->rc_header.h_cmd = DLM_RCOM;
58 spin_lock(&ls->ls_recover_lock);
59 rc->rc_seq = ls->ls_recover_seq;
60 spin_unlock(&ls->ls_recover_lock);
67 static void send_rcom(struct dlm_ls *ls, struct dlm_mhandle *mh,
71 dlm_lowcomms_commit_buffer(mh);
74 static void set_rcom_status(struct dlm_ls *ls, struct rcom_status *rs,
77 rs->rs_flags = cpu_to_le32(flags);
80 /* When replying to a status request, a node also sends back its
81 configuration values. The requesting node then checks that the remote
82 node is configured the same way as itself. */
84 static void set_rcom_config(struct dlm_ls *ls, struct rcom_config *rf,
87 rf->rf_lvblen = cpu_to_le32(ls->ls_lvblen);
88 rf->rf_lsflags = cpu_to_le32(ls->ls_exflags);
90 rf->rf_our_slot = cpu_to_le16(ls->ls_slot);
91 rf->rf_num_slots = cpu_to_le16(num_slots);
92 rf->rf_generation = cpu_to_le32(ls->ls_generation);
95 static int check_rcom_config(struct dlm_ls *ls, struct dlm_rcom *rc, int nodeid)
97 struct rcom_config *rf = (struct rcom_config *) rc->rc_buf;
99 if ((rc->rc_header.h_version & 0xFFFF0000) != DLM_HEADER_MAJOR) {
100 log_error(ls, "version mismatch: %x nodeid %d: %x",
101 DLM_HEADER_MAJOR | DLM_HEADER_MINOR, nodeid,
102 rc->rc_header.h_version);
106 if (le32_to_cpu(rf->rf_lvblen) != ls->ls_lvblen ||
107 le32_to_cpu(rf->rf_lsflags) != ls->ls_exflags) {
108 log_error(ls, "config mismatch: %d,%x nodeid %d: %d,%x",
109 ls->ls_lvblen, ls->ls_exflags, nodeid,
110 le32_to_cpu(rf->rf_lvblen),
111 le32_to_cpu(rf->rf_lsflags));
117 static void allow_sync_reply(struct dlm_ls *ls, uint64_t *new_seq)
119 spin_lock(&ls->ls_rcom_spin);
120 *new_seq = ++ls->ls_rcom_seq;
121 set_bit(LSFL_RCOM_WAIT, &ls->ls_flags);
122 spin_unlock(&ls->ls_rcom_spin);
125 static void disallow_sync_reply(struct dlm_ls *ls)
127 spin_lock(&ls->ls_rcom_spin);
128 clear_bit(LSFL_RCOM_WAIT, &ls->ls_flags);
129 clear_bit(LSFL_RCOM_READY, &ls->ls_flags);
130 spin_unlock(&ls->ls_rcom_spin);
134 * low nodeid gathers one slot value at a time from each node.
135 * it sets need_slots=0, and saves rf_our_slot returned from each
138 * other nodes gather all slot values at once from the low nodeid.
139 * they set need_slots=1, and ignore the rf_our_slot returned from each
140 * rcom_config. they use the rf_num_slots returned from the low
141 * node's rcom_config.
144 int dlm_rcom_status(struct dlm_ls *ls, int nodeid, uint32_t status_flags)
147 struct dlm_mhandle *mh;
150 ls->ls_recover_nodeid = nodeid;
152 if (nodeid == dlm_our_nodeid()) {
153 rc = ls->ls_recover_buf;
154 rc->rc_result = dlm_recover_status(ls);
159 error = create_rcom(ls, nodeid, DLM_RCOM_STATUS,
160 sizeof(struct rcom_status), &rc, &mh);
164 set_rcom_status(ls, (struct rcom_status *)rc->rc_buf, status_flags);
166 allow_sync_reply(ls, &rc->rc_id);
167 memset(ls->ls_recover_buf, 0, dlm_config.ci_buffer_size);
169 send_rcom(ls, mh, rc);
171 error = dlm_wait_function(ls, &rcom_response);
172 disallow_sync_reply(ls);
173 if (error == -ETIMEDOUT)
178 rc = ls->ls_recover_buf;
180 if (rc->rc_result == -ESRCH) {
181 /* we pretend the remote lockspace exists with 0 status */
182 log_debug(ls, "remote node %d not ready", nodeid);
186 error = check_rcom_config(ls, rc, nodeid);
189 /* the caller looks at rc_result for the remote recovery status */
194 static void receive_rcom_status(struct dlm_ls *ls, struct dlm_rcom *rc_in)
197 struct dlm_mhandle *mh;
198 struct rcom_status *rs;
200 int nodeid = rc_in->rc_header.h_nodeid;
201 int len = sizeof(struct rcom_config);
205 if (!dlm_slots_version(&rc_in->rc_header)) {
206 status = dlm_recover_status(ls);
210 rs = (struct rcom_status *)rc_in->rc_buf;
212 if (!(le32_to_cpu(rs->rs_flags) & DLM_RSF_NEED_SLOTS)) {
213 status = dlm_recover_status(ls);
217 spin_lock(&ls->ls_recover_lock);
218 status = ls->ls_recover_status;
219 num_slots = ls->ls_num_slots;
220 spin_unlock(&ls->ls_recover_lock);
221 len += num_slots * sizeof(struct rcom_slot);
224 error = create_rcom(ls, nodeid, DLM_RCOM_STATUS_REPLY,
229 rc->rc_id = rc_in->rc_id;
230 rc->rc_seq_reply = rc_in->rc_seq;
231 rc->rc_result = status;
233 set_rcom_config(ls, (struct rcom_config *)rc->rc_buf, num_slots);
238 spin_lock(&ls->ls_recover_lock);
239 if (ls->ls_num_slots != num_slots) {
240 spin_unlock(&ls->ls_recover_lock);
241 log_debug(ls, "receive_rcom_status num_slots %d to %d",
242 num_slots, ls->ls_num_slots);
244 set_rcom_config(ls, (struct rcom_config *)rc->rc_buf, 0);
248 dlm_slots_copy_out(ls, rc);
249 spin_unlock(&ls->ls_recover_lock);
252 send_rcom(ls, mh, rc);
255 static void receive_sync_reply(struct dlm_ls *ls, struct dlm_rcom *rc_in)
257 spin_lock(&ls->ls_rcom_spin);
258 if (!test_bit(LSFL_RCOM_WAIT, &ls->ls_flags) ||
259 rc_in->rc_id != ls->ls_rcom_seq) {
260 log_debug(ls, "reject reply %d from %d seq %llx expect %llx",
261 rc_in->rc_type, rc_in->rc_header.h_nodeid,
262 (unsigned long long)rc_in->rc_id,
263 (unsigned long long)ls->ls_rcom_seq);
266 memcpy(ls->ls_recover_buf, rc_in, rc_in->rc_header.h_length);
267 set_bit(LSFL_RCOM_READY, &ls->ls_flags);
268 clear_bit(LSFL_RCOM_WAIT, &ls->ls_flags);
269 wake_up(&ls->ls_wait_general);
271 spin_unlock(&ls->ls_rcom_spin);
274 int dlm_rcom_names(struct dlm_ls *ls, int nodeid, char *last_name, int last_len)
277 struct dlm_mhandle *mh;
280 ls->ls_recover_nodeid = nodeid;
283 error = create_rcom(ls, nodeid, DLM_RCOM_NAMES, last_len, &rc, &mh);
286 memcpy(rc->rc_buf, last_name, last_len);
288 allow_sync_reply(ls, &rc->rc_id);
289 memset(ls->ls_recover_buf, 0, dlm_config.ci_buffer_size);
291 send_rcom(ls, mh, rc);
293 error = dlm_wait_function(ls, &rcom_response);
294 disallow_sync_reply(ls);
295 if (error == -ETIMEDOUT)
301 static void receive_rcom_names(struct dlm_ls *ls, struct dlm_rcom *rc_in)
304 struct dlm_mhandle *mh;
305 int error, inlen, outlen, nodeid;
307 nodeid = rc_in->rc_header.h_nodeid;
308 inlen = rc_in->rc_header.h_length - sizeof(struct dlm_rcom);
309 outlen = dlm_config.ci_buffer_size - sizeof(struct dlm_rcom);
311 error = create_rcom(ls, nodeid, DLM_RCOM_NAMES_REPLY, outlen, &rc, &mh);
314 rc->rc_id = rc_in->rc_id;
315 rc->rc_seq_reply = rc_in->rc_seq;
317 dlm_copy_master_names(ls, rc_in->rc_buf, inlen, rc->rc_buf, outlen,
319 send_rcom(ls, mh, rc);
322 int dlm_send_rcom_lookup(struct dlm_rsb *r, int dir_nodeid)
325 struct dlm_mhandle *mh;
326 struct dlm_ls *ls = r->res_ls;
329 error = create_rcom(ls, dir_nodeid, DLM_RCOM_LOOKUP, r->res_length,
333 memcpy(rc->rc_buf, r->res_name, r->res_length);
334 rc->rc_id = (unsigned long) r->res_id;
336 send_rcom(ls, mh, rc);
341 static void receive_rcom_lookup(struct dlm_ls *ls, struct dlm_rcom *rc_in)
344 struct dlm_mhandle *mh;
345 int error, ret_nodeid, nodeid = rc_in->rc_header.h_nodeid;
346 int len = rc_in->rc_header.h_length - sizeof(struct dlm_rcom);
348 error = create_rcom(ls, nodeid, DLM_RCOM_LOOKUP_REPLY, 0, &rc, &mh);
352 /* Old code would send this special id to trigger a debug dump. */
353 if (rc_in->rc_id == 0xFFFFFFFF) {
354 log_error(ls, "receive_rcom_lookup dump from %d", nodeid);
355 dlm_dump_rsb_name(ls, rc_in->rc_buf, len);
359 error = dlm_master_lookup(ls, nodeid, rc_in->rc_buf, len,
360 DLM_LU_RECOVER_MASTER, &ret_nodeid, NULL);
363 rc->rc_result = ret_nodeid;
364 rc->rc_id = rc_in->rc_id;
365 rc->rc_seq_reply = rc_in->rc_seq;
367 send_rcom(ls, mh, rc);
370 static void receive_rcom_lookup_reply(struct dlm_ls *ls, struct dlm_rcom *rc_in)
372 dlm_recover_master_reply(ls, rc_in);
375 static void pack_rcom_lock(struct dlm_rsb *r, struct dlm_lkb *lkb,
376 struct rcom_lock *rl)
378 memset(rl, 0, sizeof(*rl));
380 rl->rl_ownpid = cpu_to_le32(lkb->lkb_ownpid);
381 rl->rl_lkid = cpu_to_le32(lkb->lkb_id);
382 rl->rl_exflags = cpu_to_le32(lkb->lkb_exflags);
383 rl->rl_flags = cpu_to_le32(lkb->lkb_flags);
384 rl->rl_lvbseq = cpu_to_le32(lkb->lkb_lvbseq);
385 rl->rl_rqmode = lkb->lkb_rqmode;
386 rl->rl_grmode = lkb->lkb_grmode;
387 rl->rl_status = lkb->lkb_status;
388 rl->rl_wait_type = cpu_to_le16(lkb->lkb_wait_type);
391 rl->rl_asts |= DLM_CB_BAST;
393 rl->rl_asts |= DLM_CB_CAST;
395 rl->rl_namelen = cpu_to_le16(r->res_length);
396 memcpy(rl->rl_name, r->res_name, r->res_length);
398 /* FIXME: might we have an lvb without DLM_LKF_VALBLK set ?
399 If so, receive_rcom_lock_args() won't take this copy. */
402 memcpy(rl->rl_lvb, lkb->lkb_lvbptr, r->res_ls->ls_lvblen);
405 int dlm_send_rcom_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
407 struct dlm_ls *ls = r->res_ls;
409 struct dlm_mhandle *mh;
410 struct rcom_lock *rl;
411 int error, len = sizeof(struct rcom_lock);
414 len += ls->ls_lvblen;
416 error = create_rcom(ls, r->res_nodeid, DLM_RCOM_LOCK, len, &rc, &mh);
420 rl = (struct rcom_lock *) rc->rc_buf;
421 pack_rcom_lock(r, lkb, rl);
422 rc->rc_id = (unsigned long) r;
424 send_rcom(ls, mh, rc);
429 /* needs at least dlm_rcom + rcom_lock */
430 static void receive_rcom_lock(struct dlm_ls *ls, struct dlm_rcom *rc_in)
433 struct dlm_mhandle *mh;
434 int error, nodeid = rc_in->rc_header.h_nodeid;
436 dlm_recover_master_copy(ls, rc_in);
438 error = create_rcom(ls, nodeid, DLM_RCOM_LOCK_REPLY,
439 sizeof(struct rcom_lock), &rc, &mh);
443 /* We send back the same rcom_lock struct we received, but
444 dlm_recover_master_copy() has filled in rl_remid and rl_result */
446 memcpy(rc->rc_buf, rc_in->rc_buf, sizeof(struct rcom_lock));
447 rc->rc_id = rc_in->rc_id;
448 rc->rc_seq_reply = rc_in->rc_seq;
450 send_rcom(ls, mh, rc);
453 /* If the lockspace doesn't exist then still send a status message
454 back; it's possible that it just doesn't have its global_id yet. */
456 int dlm_send_ls_not_ready(int nodeid, struct dlm_rcom *rc_in)
459 struct rcom_config *rf;
460 struct dlm_mhandle *mh;
462 int mb_len = sizeof(struct dlm_rcom) + sizeof(struct rcom_config);
464 mh = dlm_lowcomms_get_buffer(nodeid, mb_len, GFP_NOFS, &mb);
467 memset(mb, 0, mb_len);
469 rc = (struct dlm_rcom *) mb;
471 rc->rc_header.h_version = (DLM_HEADER_MAJOR | DLM_HEADER_MINOR);
472 rc->rc_header.h_lockspace = rc_in->rc_header.h_lockspace;
473 rc->rc_header.h_nodeid = dlm_our_nodeid();
474 rc->rc_header.h_length = mb_len;
475 rc->rc_header.h_cmd = DLM_RCOM;
477 rc->rc_type = DLM_RCOM_STATUS_REPLY;
478 rc->rc_id = rc_in->rc_id;
479 rc->rc_seq_reply = rc_in->rc_seq;
480 rc->rc_result = -ESRCH;
482 rf = (struct rcom_config *) rc->rc_buf;
483 rf->rf_lvblen = cpu_to_le32(~0U);
486 dlm_lowcomms_commit_buffer(mh);
492 * Ignore messages for stage Y before we set
493 * recover_status bit for stage X:
497 * dlm_recover_members()
500 * - ignore NAMES, NAMES_REPLY
501 * - ignore LOOKUP, LOOKUP_REPLY
502 * - ignore LOCK, LOCK_REPLY
504 * recover_status |= NODES
506 * dlm_recover_members_wait()
508 * dlm_recover_directory()
511 * - ignore LOOKUP, LOOKUP_REPLY
512 * - ignore LOCK, LOCK_REPLY
514 * recover_status |= DIR
516 * dlm_recover_directory_wait()
518 * dlm_recover_masters()
520 * - recv LOOKUP_REPLY
522 * dlm_recover_locks()
526 * recover_status |= LOCKS
528 * dlm_recover_locks_wait()
530 * recover_status |= DONE
533 /* Called by dlm_recv; corresponds to dlm_receive_message() but special
534 recovery-only comms are sent through here. */
536 void dlm_receive_rcom(struct dlm_ls *ls, struct dlm_rcom *rc, int nodeid)
538 int lock_size = sizeof(struct dlm_rcom) + sizeof(struct rcom_lock);
539 int stop, reply = 0, names = 0, lookup = 0, lock = 0;
543 switch (rc->rc_type) {
544 case DLM_RCOM_STATUS_REPLY:
550 case DLM_RCOM_NAMES_REPLY:
554 case DLM_RCOM_LOOKUP:
557 case DLM_RCOM_LOOKUP_REPLY:
564 case DLM_RCOM_LOCK_REPLY:
570 spin_lock(&ls->ls_recover_lock);
571 status = ls->ls_recover_status;
572 stop = test_bit(LSFL_RECOVER_STOP, &ls->ls_flags);
573 seq = ls->ls_recover_seq;
574 spin_unlock(&ls->ls_recover_lock);
576 if (stop && (rc->rc_type != DLM_RCOM_STATUS))
579 if (reply && (rc->rc_seq_reply != seq))
582 if (!(status & DLM_RS_NODES) && (names || lookup || lock))
585 if (!(status & DLM_RS_DIR) && (lookup || lock))
588 switch (rc->rc_type) {
589 case DLM_RCOM_STATUS:
590 receive_rcom_status(ls, rc);
594 receive_rcom_names(ls, rc);
597 case DLM_RCOM_LOOKUP:
598 receive_rcom_lookup(ls, rc);
602 if (rc->rc_header.h_length < lock_size)
604 receive_rcom_lock(ls, rc);
607 case DLM_RCOM_STATUS_REPLY:
608 receive_sync_reply(ls, rc);
611 case DLM_RCOM_NAMES_REPLY:
612 receive_sync_reply(ls, rc);
615 case DLM_RCOM_LOOKUP_REPLY:
616 receive_rcom_lookup_reply(ls, rc);
619 case DLM_RCOM_LOCK_REPLY:
620 if (rc->rc_header.h_length < lock_size)
622 dlm_recover_process_copy(ls, rc);
626 log_error(ls, "receive_rcom bad type %d", rc->rc_type);
631 log_limit(ls, "dlm_receive_rcom ignore msg %d "
632 "from %d %llu %llu recover seq %llu sts %x gen %u",
635 (unsigned long long)rc->rc_seq,
636 (unsigned long long)rc->rc_seq_reply,
637 (unsigned long long)seq,
638 status, ls->ls_generation);
641 log_error(ls, "recovery message %d from %d is too short",
642 rc->rc_type, nodeid);