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"
28 static int rcom_response(struct dlm_ls *ls)
30 return test_bit(LSFL_RCOM_READY, &ls->ls_flags);
33 static 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)
37 struct dlm_mhandle *mh;
39 int mb_len = sizeof(struct dlm_rcom) + len;
41 mh = dlm_lowcomms_get_buffer(to_nodeid, mb_len, ls->ls_allocation, &mb);
43 log_print("create_rcom to %d type %d len %d ENOBUFS",
44 to_nodeid, type, len);
47 memset(mb, 0, mb_len);
49 rc = (struct dlm_rcom *) mb;
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;
59 spin_lock(&ls->ls_recover_lock);
60 rc->rc_seq = ls->ls_recover_seq;
61 spin_unlock(&ls->ls_recover_lock);
68 static void send_rcom(struct dlm_ls *ls, struct dlm_mhandle *mh,
72 dlm_lowcomms_commit_buffer(mh);
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. */
79 static void make_config(struct dlm_ls *ls, struct rcom_config *rf)
81 rf->rf_lvblen = ls->ls_lvblen;
82 rf->rf_lsflags = ls->ls_exflags;
85 static int check_config(struct dlm_ls *ls, struct dlm_rcom *rc, int nodeid)
87 struct rcom_config *rf = (struct rcom_config *) rc->rc_buf;
89 if ((rc->rc_header.h_version & 0xFFFF0000) != DLM_HEADER_MAJOR) {
90 log_error(ls, "version mismatch: %x nodeid %d: %x",
91 DLM_HEADER_MAJOR | DLM_HEADER_MINOR, nodeid,
92 rc->rc_header.h_version);
96 if (rf->rf_lvblen != ls->ls_lvblen ||
97 rf->rf_lsflags != ls->ls_exflags) {
98 log_error(ls, "config mismatch: %d,%x nodeid %d: %d,%x",
99 ls->ls_lvblen, ls->ls_exflags,
100 nodeid, rf->rf_lvblen, rf->rf_lsflags);
106 static void allow_sync_reply(struct dlm_ls *ls, uint64_t *new_seq)
108 spin_lock(&ls->ls_rcom_spin);
109 *new_seq = ++ls->ls_rcom_seq;
110 set_bit(LSFL_RCOM_WAIT, &ls->ls_flags);
111 spin_unlock(&ls->ls_rcom_spin);
114 static void disallow_sync_reply(struct dlm_ls *ls)
116 spin_lock(&ls->ls_rcom_spin);
117 clear_bit(LSFL_RCOM_WAIT, &ls->ls_flags);
118 clear_bit(LSFL_RCOM_READY, &ls->ls_flags);
119 spin_unlock(&ls->ls_rcom_spin);
122 int dlm_rcom_status(struct dlm_ls *ls, int nodeid)
125 struct dlm_mhandle *mh;
128 ls->ls_recover_nodeid = nodeid;
130 if (nodeid == dlm_our_nodeid()) {
131 rc = (struct dlm_rcom *) ls->ls_recover_buf;
132 rc->rc_result = dlm_recover_status(ls);
136 error = create_rcom(ls, nodeid, DLM_RCOM_STATUS, 0, &rc, &mh);
140 allow_sync_reply(ls, &rc->rc_id);
141 memset(ls->ls_recover_buf, 0, dlm_config.ci_buffer_size);
143 send_rcom(ls, mh, rc);
145 error = dlm_wait_function(ls, &rcom_response);
146 disallow_sync_reply(ls);
150 rc = (struct dlm_rcom *) ls->ls_recover_buf;
152 if (rc->rc_result == -ESRCH) {
153 /* we pretend the remote lockspace exists with 0 status */
154 log_debug(ls, "remote node %d not ready", nodeid);
157 error = check_config(ls, rc, nodeid);
158 /* the caller looks at rc_result for the remote recovery status */
163 static void receive_rcom_status(struct dlm_ls *ls, struct dlm_rcom *rc_in)
166 struct dlm_mhandle *mh;
167 int error, nodeid = rc_in->rc_header.h_nodeid;
169 error = create_rcom(ls, nodeid, DLM_RCOM_STATUS_REPLY,
170 sizeof(struct rcom_config), &rc, &mh);
173 rc->rc_id = rc_in->rc_id;
174 rc->rc_seq_reply = rc_in->rc_seq;
175 rc->rc_result = dlm_recover_status(ls);
176 make_config(ls, (struct rcom_config *) rc->rc_buf);
178 send_rcom(ls, mh, rc);
181 static void receive_sync_reply(struct dlm_ls *ls, struct dlm_rcom *rc_in)
183 spin_lock(&ls->ls_rcom_spin);
184 if (!test_bit(LSFL_RCOM_WAIT, &ls->ls_flags) ||
185 rc_in->rc_id != ls->ls_rcom_seq) {
186 log_debug(ls, "reject reply %d from %d seq %llx expect %llx",
187 rc_in->rc_type, rc_in->rc_header.h_nodeid,
188 (unsigned long long)rc_in->rc_id,
189 (unsigned long long)ls->ls_rcom_seq);
192 memcpy(ls->ls_recover_buf, rc_in, rc_in->rc_header.h_length);
193 set_bit(LSFL_RCOM_READY, &ls->ls_flags);
194 clear_bit(LSFL_RCOM_WAIT, &ls->ls_flags);
195 wake_up(&ls->ls_wait_general);
197 spin_unlock(&ls->ls_rcom_spin);
200 int dlm_rcom_names(struct dlm_ls *ls, int nodeid, char *last_name, int last_len)
203 struct dlm_mhandle *mh;
204 int error = 0, len = sizeof(struct dlm_rcom);
206 ls->ls_recover_nodeid = nodeid;
208 if (nodeid == dlm_our_nodeid()) {
209 dlm_copy_master_names(ls, last_name, last_len,
210 ls->ls_recover_buf + len,
211 dlm_config.ci_buffer_size - len, nodeid);
215 error = create_rcom(ls, nodeid, DLM_RCOM_NAMES, last_len, &rc, &mh);
218 memcpy(rc->rc_buf, last_name, last_len);
220 allow_sync_reply(ls, &rc->rc_id);
221 memset(ls->ls_recover_buf, 0, dlm_config.ci_buffer_size);
223 send_rcom(ls, mh, rc);
225 error = dlm_wait_function(ls, &rcom_response);
226 disallow_sync_reply(ls);
231 static void receive_rcom_names(struct dlm_ls *ls, struct dlm_rcom *rc_in)
234 struct dlm_mhandle *mh;
235 int error, inlen, outlen, nodeid;
237 nodeid = rc_in->rc_header.h_nodeid;
238 inlen = rc_in->rc_header.h_length - sizeof(struct dlm_rcom);
239 outlen = dlm_config.ci_buffer_size - sizeof(struct dlm_rcom);
241 error = create_rcom(ls, nodeid, DLM_RCOM_NAMES_REPLY, outlen, &rc, &mh);
244 rc->rc_id = rc_in->rc_id;
245 rc->rc_seq_reply = rc_in->rc_seq;
247 dlm_copy_master_names(ls, rc_in->rc_buf, inlen, rc->rc_buf, outlen,
249 send_rcom(ls, mh, rc);
252 int dlm_send_rcom_lookup(struct dlm_rsb *r, int dir_nodeid)
255 struct dlm_mhandle *mh;
256 struct dlm_ls *ls = r->res_ls;
259 error = create_rcom(ls, dir_nodeid, DLM_RCOM_LOOKUP, r->res_length,
263 memcpy(rc->rc_buf, r->res_name, r->res_length);
264 rc->rc_id = (unsigned long) r;
266 send_rcom(ls, mh, rc);
271 static void receive_rcom_lookup(struct dlm_ls *ls, struct dlm_rcom *rc_in)
274 struct dlm_mhandle *mh;
275 int error, ret_nodeid, nodeid = rc_in->rc_header.h_nodeid;
276 int len = rc_in->rc_header.h_length - sizeof(struct dlm_rcom);
278 error = create_rcom(ls, nodeid, DLM_RCOM_LOOKUP_REPLY, 0, &rc, &mh);
282 error = dlm_dir_lookup(ls, nodeid, rc_in->rc_buf, len, &ret_nodeid);
285 rc->rc_result = ret_nodeid;
286 rc->rc_id = rc_in->rc_id;
287 rc->rc_seq_reply = rc_in->rc_seq;
289 send_rcom(ls, mh, rc);
292 static void receive_rcom_lookup_reply(struct dlm_ls *ls, struct dlm_rcom *rc_in)
294 dlm_recover_master_reply(ls, rc_in);
297 static void pack_rcom_lock(struct dlm_rsb *r, struct dlm_lkb *lkb,
298 struct rcom_lock *rl)
300 memset(rl, 0, sizeof(*rl));
302 rl->rl_ownpid = lkb->lkb_ownpid;
303 rl->rl_lkid = lkb->lkb_id;
304 rl->rl_exflags = lkb->lkb_exflags;
305 rl->rl_flags = lkb->lkb_flags;
306 rl->rl_lvbseq = lkb->lkb_lvbseq;
307 rl->rl_rqmode = lkb->lkb_rqmode;
308 rl->rl_grmode = lkb->lkb_grmode;
309 rl->rl_status = lkb->lkb_status;
310 rl->rl_wait_type = lkb->lkb_wait_type;
312 if (lkb->lkb_bastaddr)
313 rl->rl_asts |= AST_BAST;
314 if (lkb->lkb_astaddr)
315 rl->rl_asts |= AST_COMP;
317 rl->rl_namelen = r->res_length;
318 memcpy(rl->rl_name, r->res_name, r->res_length);
320 /* FIXME: might we have an lvb without DLM_LKF_VALBLK set ?
321 If so, receive_rcom_lock_args() won't take this copy. */
324 memcpy(rl->rl_lvb, lkb->lkb_lvbptr, r->res_ls->ls_lvblen);
327 int dlm_send_rcom_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
329 struct dlm_ls *ls = r->res_ls;
331 struct dlm_mhandle *mh;
332 struct rcom_lock *rl;
333 int error, len = sizeof(struct rcom_lock);
336 len += ls->ls_lvblen;
338 error = create_rcom(ls, r->res_nodeid, DLM_RCOM_LOCK, len, &rc, &mh);
342 rl = (struct rcom_lock *) rc->rc_buf;
343 pack_rcom_lock(r, lkb, rl);
344 rc->rc_id = (unsigned long) r;
346 send_rcom(ls, mh, rc);
351 static void receive_rcom_lock(struct dlm_ls *ls, struct dlm_rcom *rc_in)
354 struct dlm_mhandle *mh;
355 int error, nodeid = rc_in->rc_header.h_nodeid;
357 dlm_recover_master_copy(ls, rc_in);
359 error = create_rcom(ls, nodeid, DLM_RCOM_LOCK_REPLY,
360 sizeof(struct rcom_lock), &rc, &mh);
364 /* We send back the same rcom_lock struct we received, but
365 dlm_recover_master_copy() has filled in rl_remid and rl_result */
367 memcpy(rc->rc_buf, rc_in->rc_buf, sizeof(struct rcom_lock));
368 rc->rc_id = rc_in->rc_id;
369 rc->rc_seq_reply = rc_in->rc_seq;
371 send_rcom(ls, mh, rc);
374 /* If the lockspace doesn't exist then still send a status message
375 back; it's possible that it just doesn't have its global_id yet. */
377 int dlm_send_ls_not_ready(int nodeid, struct dlm_rcom *rc_in)
380 struct rcom_config *rf;
381 struct dlm_mhandle *mh;
383 int mb_len = sizeof(struct dlm_rcom) + sizeof(struct rcom_config);
385 mh = dlm_lowcomms_get_buffer(nodeid, mb_len, GFP_NOFS, &mb);
388 memset(mb, 0, mb_len);
390 rc = (struct dlm_rcom *) mb;
392 rc->rc_header.h_version = (DLM_HEADER_MAJOR | DLM_HEADER_MINOR);
393 rc->rc_header.h_lockspace = rc_in->rc_header.h_lockspace;
394 rc->rc_header.h_nodeid = dlm_our_nodeid();
395 rc->rc_header.h_length = mb_len;
396 rc->rc_header.h_cmd = DLM_RCOM;
398 rc->rc_type = DLM_RCOM_STATUS_REPLY;
399 rc->rc_id = rc_in->rc_id;
400 rc->rc_seq_reply = rc_in->rc_seq;
401 rc->rc_result = -ESRCH;
403 rf = (struct rcom_config *) rc->rc_buf;
407 dlm_lowcomms_commit_buffer(mh);
412 static int is_old_reply(struct dlm_ls *ls, struct dlm_rcom *rc)
417 switch (rc->rc_type) {
418 case DLM_RCOM_STATUS_REPLY:
419 case DLM_RCOM_NAMES_REPLY:
420 case DLM_RCOM_LOOKUP_REPLY:
421 case DLM_RCOM_LOCK_REPLY:
422 spin_lock(&ls->ls_recover_lock);
423 seq = ls->ls_recover_seq;
424 spin_unlock(&ls->ls_recover_lock);
425 if (rc->rc_seq_reply != seq) {
426 log_debug(ls, "ignoring old reply %x from %d "
427 "seq_reply %llx expect %llx",
428 rc->rc_type, rc->rc_header.h_nodeid,
429 (unsigned long long)rc->rc_seq_reply,
430 (unsigned long long)seq);
437 /* Called by dlm_recv; corresponds to dlm_receive_message() but special
438 recovery-only comms are sent through here. */
440 void dlm_receive_rcom(struct dlm_ls *ls, struct dlm_rcom *rc, int nodeid)
442 if (dlm_recovery_stopped(ls) && (rc->rc_type != DLM_RCOM_STATUS)) {
443 log_debug(ls, "ignoring recovery message %x from %d",
444 rc->rc_type, nodeid);
448 if (is_old_reply(ls, rc))
451 switch (rc->rc_type) {
452 case DLM_RCOM_STATUS:
453 receive_rcom_status(ls, rc);
457 receive_rcom_names(ls, rc);
460 case DLM_RCOM_LOOKUP:
461 receive_rcom_lookup(ls, rc);
465 receive_rcom_lock(ls, rc);
468 case DLM_RCOM_STATUS_REPLY:
469 receive_sync_reply(ls, rc);
472 case DLM_RCOM_NAMES_REPLY:
473 receive_sync_reply(ls, rc);
476 case DLM_RCOM_LOOKUP_REPLY:
477 receive_rcom_lookup_reply(ls, rc);
480 case DLM_RCOM_LOCK_REPLY:
481 dlm_recover_process_copy(ls, rc);
485 log_error(ls, "receive_rcom bad type %d", rc->rc_type);