Commit | Line | Data |
---|---|---|
e7fd4179 DT |
1 | /****************************************************************************** |
2 | ******************************************************************************* | |
3 | ** | |
4 | ** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
5 | ** Copyright (C) 2005 Red Hat, Inc. All rights reserved. | |
6 | ** | |
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. | |
10 | ** | |
11 | ******************************************************************************* | |
12 | ******************************************************************************/ | |
13 | ||
14 | #include "dlm_internal.h" | |
15 | #include "lockspace.h" | |
16 | #include "member.h" | |
17 | #include "lowcomms.h" | |
18 | #include "midcomms.h" | |
19 | #include "rcom.h" | |
20 | #include "recover.h" | |
21 | #include "dir.h" | |
22 | #include "config.h" | |
23 | #include "memory.h" | |
24 | #include "lock.h" | |
25 | #include "util.h" | |
26 | ||
27 | ||
28 | static int rcom_response(struct dlm_ls *ls) | |
29 | { | |
30 | return test_bit(LSFL_RCOM_READY, &ls->ls_flags); | |
31 | } | |
32 | ||
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) | |
35 | { | |
36 | struct dlm_rcom *rc; | |
37 | struct dlm_mhandle *mh; | |
38 | char *mb; | |
39 | int mb_len = sizeof(struct dlm_rcom) + len; | |
40 | ||
41 | mh = dlm_lowcomms_get_buffer(to_nodeid, mb_len, GFP_KERNEL, &mb); | |
42 | if (!mh) { | |
43 | log_print("create_rcom to %d type %d len %d ENOBUFS", | |
44 | to_nodeid, type, len); | |
45 | return -ENOBUFS; | |
46 | } | |
47 | memset(mb, 0, mb_len); | |
48 | ||
49 | rc = (struct dlm_rcom *) mb; | |
50 | ||
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; | |
56 | ||
57 | rc->rc_type = type; | |
58 | ||
59 | *mh_ret = mh; | |
60 | *rc_ret = rc; | |
61 | return 0; | |
62 | } | |
63 | ||
64 | static void send_rcom(struct dlm_ls *ls, struct dlm_mhandle *mh, | |
65 | struct dlm_rcom *rc) | |
66 | { | |
67 | dlm_rcom_out(rc); | |
68 | dlm_lowcomms_commit_buffer(mh); | |
69 | } | |
70 | ||
71 | /* When replying to a status request, a node also sends back its | |
72 | configuration values. The requesting node then checks that the remote | |
73 | node is configured the same way as itself. */ | |
74 | ||
75 | static void make_config(struct dlm_ls *ls, struct rcom_config *rf) | |
76 | { | |
77 | rf->rf_lvblen = ls->ls_lvblen; | |
78 | rf->rf_lsflags = ls->ls_exflags; | |
79 | } | |
80 | ||
81 | static int check_config(struct dlm_ls *ls, struct rcom_config *rf, int nodeid) | |
82 | { | |
83 | if (rf->rf_lvblen != ls->ls_lvblen || | |
84 | rf->rf_lsflags != ls->ls_exflags) { | |
85 | log_error(ls, "config mismatch: %d,%x nodeid %d: %d,%x", | |
86 | ls->ls_lvblen, ls->ls_exflags, | |
87 | nodeid, rf->rf_lvblen, rf->rf_lsflags); | |
88 | return -EINVAL; | |
89 | } | |
90 | return 0; | |
91 | } | |
92 | ||
98f176fb DT |
93 | static void allow_sync_reply(struct dlm_ls *ls, uint64_t *new_seq) |
94 | { | |
95 | spin_lock(&ls->ls_rcom_spin); | |
96 | *new_seq = ++ls->ls_rcom_seq; | |
97 | set_bit(LSFL_RCOM_WAIT, &ls->ls_flags); | |
98 | spin_unlock(&ls->ls_rcom_spin); | |
99 | } | |
100 | ||
101 | static void disallow_sync_reply(struct dlm_ls *ls) | |
102 | { | |
103 | spin_lock(&ls->ls_rcom_spin); | |
104 | clear_bit(LSFL_RCOM_WAIT, &ls->ls_flags); | |
105 | clear_bit(LSFL_RCOM_READY, &ls->ls_flags); | |
106 | spin_unlock(&ls->ls_rcom_spin); | |
107 | } | |
108 | ||
e7fd4179 DT |
109 | int dlm_rcom_status(struct dlm_ls *ls, int nodeid) |
110 | { | |
111 | struct dlm_rcom *rc; | |
112 | struct dlm_mhandle *mh; | |
113 | int error = 0; | |
114 | ||
faa0f267 | 115 | ls->ls_recover_nodeid = nodeid; |
e7fd4179 DT |
116 | |
117 | if (nodeid == dlm_our_nodeid()) { | |
118 | rc = (struct dlm_rcom *) ls->ls_recover_buf; | |
119 | rc->rc_result = dlm_recover_status(ls); | |
120 | goto out; | |
121 | } | |
122 | ||
123 | error = create_rcom(ls, nodeid, DLM_RCOM_STATUS, 0, &rc, &mh); | |
124 | if (error) | |
125 | goto out; | |
98f176fb DT |
126 | |
127 | allow_sync_reply(ls, &rc->rc_id); | |
128 | memset(ls->ls_recover_buf, 0, dlm_config.buffer_size); | |
e7fd4179 DT |
129 | |
130 | send_rcom(ls, mh, rc); | |
131 | ||
132 | error = dlm_wait_function(ls, &rcom_response); | |
98f176fb | 133 | disallow_sync_reply(ls); |
e7fd4179 DT |
134 | if (error) |
135 | goto out; | |
136 | ||
137 | rc = (struct dlm_rcom *) ls->ls_recover_buf; | |
138 | ||
139 | if (rc->rc_result == -ESRCH) { | |
140 | /* we pretend the remote lockspace exists with 0 status */ | |
141 | log_debug(ls, "remote node %d not ready", nodeid); | |
142 | rc->rc_result = 0; | |
143 | } else | |
144 | error = check_config(ls, (struct rcom_config *) rc->rc_buf, | |
145 | nodeid); | |
146 | /* the caller looks at rc_result for the remote recovery status */ | |
147 | out: | |
148 | return error; | |
149 | } | |
150 | ||
151 | static void receive_rcom_status(struct dlm_ls *ls, struct dlm_rcom *rc_in) | |
152 | { | |
153 | struct dlm_rcom *rc; | |
154 | struct dlm_mhandle *mh; | |
155 | int error, nodeid = rc_in->rc_header.h_nodeid; | |
156 | ||
157 | error = create_rcom(ls, nodeid, DLM_RCOM_STATUS_REPLY, | |
158 | sizeof(struct rcom_config), &rc, &mh); | |
159 | if (error) | |
160 | return; | |
4a99c3d9 | 161 | rc->rc_id = rc_in->rc_id; |
e7fd4179 DT |
162 | rc->rc_result = dlm_recover_status(ls); |
163 | make_config(ls, (struct rcom_config *) rc->rc_buf); | |
164 | ||
165 | send_rcom(ls, mh, rc); | |
166 | } | |
167 | ||
4a99c3d9 | 168 | static void receive_sync_reply(struct dlm_ls *ls, struct dlm_rcom *rc_in) |
e7fd4179 | 169 | { |
98f176fb DT |
170 | spin_lock(&ls->ls_rcom_spin); |
171 | if (!test_bit(LSFL_RCOM_WAIT, &ls->ls_flags) || | |
172 | rc_in->rc_id != ls->ls_rcom_seq) { | |
173 | log_debug(ls, "reject reply %d from %d seq %llx expect %llx", | |
174 | rc_in->rc_type, rc_in->rc_header.h_nodeid, | |
175 | rc_in->rc_id, ls->ls_rcom_seq); | |
176 | goto out; | |
4a99c3d9 | 177 | } |
e7fd4179 DT |
178 | memcpy(ls->ls_recover_buf, rc_in, rc_in->rc_header.h_length); |
179 | set_bit(LSFL_RCOM_READY, &ls->ls_flags); | |
98f176fb | 180 | clear_bit(LSFL_RCOM_WAIT, &ls->ls_flags); |
e7fd4179 | 181 | wake_up(&ls->ls_wait_general); |
98f176fb DT |
182 | out: |
183 | spin_unlock(&ls->ls_rcom_spin); | |
e7fd4179 DT |
184 | } |
185 | ||
4a99c3d9 DT |
186 | static void receive_rcom_status_reply(struct dlm_ls *ls, struct dlm_rcom *rc_in) |
187 | { | |
188 | receive_sync_reply(ls, rc_in); | |
189 | } | |
190 | ||
e7fd4179 DT |
191 | int dlm_rcom_names(struct dlm_ls *ls, int nodeid, char *last_name, int last_len) |
192 | { | |
193 | struct dlm_rcom *rc; | |
194 | struct dlm_mhandle *mh; | |
195 | int error = 0, len = sizeof(struct dlm_rcom); | |
196 | ||
faa0f267 | 197 | ls->ls_recover_nodeid = nodeid; |
e7fd4179 DT |
198 | |
199 | if (nodeid == dlm_our_nodeid()) { | |
200 | dlm_copy_master_names(ls, last_name, last_len, | |
201 | ls->ls_recover_buf + len, | |
202 | dlm_config.buffer_size - len, nodeid); | |
203 | goto out; | |
204 | } | |
205 | ||
206 | error = create_rcom(ls, nodeid, DLM_RCOM_NAMES, last_len, &rc, &mh); | |
207 | if (error) | |
208 | goto out; | |
209 | memcpy(rc->rc_buf, last_name, last_len); | |
98f176fb DT |
210 | |
211 | allow_sync_reply(ls, &rc->rc_id); | |
212 | memset(ls->ls_recover_buf, 0, dlm_config.buffer_size); | |
e7fd4179 DT |
213 | |
214 | send_rcom(ls, mh, rc); | |
215 | ||
216 | error = dlm_wait_function(ls, &rcom_response); | |
98f176fb | 217 | disallow_sync_reply(ls); |
e7fd4179 DT |
218 | out: |
219 | return error; | |
220 | } | |
221 | ||
222 | static void receive_rcom_names(struct dlm_ls *ls, struct dlm_rcom *rc_in) | |
223 | { | |
224 | struct dlm_rcom *rc; | |
225 | struct dlm_mhandle *mh; | |
226 | int error, inlen, outlen; | |
227 | int nodeid = rc_in->rc_header.h_nodeid; | |
228 | uint32_t status = dlm_recover_status(ls); | |
229 | ||
230 | /* | |
231 | * We can't run dlm_dir_rebuild_send (which uses ls_nodes) while | |
232 | * dlm_recoverd is running ls_nodes_reconfig (which changes ls_nodes). | |
233 | * It could only happen in rare cases where we get a late NAMES | |
234 | * message from a previous instance of recovery. | |
235 | */ | |
236 | ||
237 | if (!(status & DLM_RS_NODES)) { | |
238 | log_debug(ls, "ignoring RCOM_NAMES from %u", nodeid); | |
239 | return; | |
240 | } | |
241 | ||
242 | nodeid = rc_in->rc_header.h_nodeid; | |
243 | inlen = rc_in->rc_header.h_length - sizeof(struct dlm_rcom); | |
244 | outlen = dlm_config.buffer_size - sizeof(struct dlm_rcom); | |
245 | ||
246 | error = create_rcom(ls, nodeid, DLM_RCOM_NAMES_REPLY, outlen, &rc, &mh); | |
247 | if (error) | |
248 | return; | |
4a99c3d9 | 249 | rc->rc_id = rc_in->rc_id; |
e7fd4179 DT |
250 | |
251 | dlm_copy_master_names(ls, rc_in->rc_buf, inlen, rc->rc_buf, outlen, | |
252 | nodeid); | |
253 | send_rcom(ls, mh, rc); | |
254 | } | |
255 | ||
256 | static void receive_rcom_names_reply(struct dlm_ls *ls, struct dlm_rcom *rc_in) | |
257 | { | |
4a99c3d9 | 258 | receive_sync_reply(ls, rc_in); |
e7fd4179 DT |
259 | } |
260 | ||
261 | int dlm_send_rcom_lookup(struct dlm_rsb *r, int dir_nodeid) | |
262 | { | |
263 | struct dlm_rcom *rc; | |
264 | struct dlm_mhandle *mh; | |
265 | struct dlm_ls *ls = r->res_ls; | |
266 | int error; | |
267 | ||
268 | error = create_rcom(ls, dir_nodeid, DLM_RCOM_LOOKUP, r->res_length, | |
269 | &rc, &mh); | |
270 | if (error) | |
271 | goto out; | |
272 | memcpy(rc->rc_buf, r->res_name, r->res_length); | |
273 | rc->rc_id = (unsigned long) r; | |
274 | ||
275 | send_rcom(ls, mh, rc); | |
276 | out: | |
277 | return error; | |
278 | } | |
279 | ||
280 | static void receive_rcom_lookup(struct dlm_ls *ls, struct dlm_rcom *rc_in) | |
281 | { | |
282 | struct dlm_rcom *rc; | |
283 | struct dlm_mhandle *mh; | |
284 | int error, ret_nodeid, nodeid = rc_in->rc_header.h_nodeid; | |
285 | int len = rc_in->rc_header.h_length - sizeof(struct dlm_rcom); | |
286 | ||
287 | error = create_rcom(ls, nodeid, DLM_RCOM_LOOKUP_REPLY, 0, &rc, &mh); | |
288 | if (error) | |
289 | return; | |
290 | ||
291 | error = dlm_dir_lookup(ls, nodeid, rc_in->rc_buf, len, &ret_nodeid); | |
292 | if (error) | |
293 | ret_nodeid = error; | |
294 | rc->rc_result = ret_nodeid; | |
295 | rc->rc_id = rc_in->rc_id; | |
296 | ||
297 | send_rcom(ls, mh, rc); | |
298 | } | |
299 | ||
300 | static void receive_rcom_lookup_reply(struct dlm_ls *ls, struct dlm_rcom *rc_in) | |
301 | { | |
302 | dlm_recover_master_reply(ls, rc_in); | |
303 | } | |
304 | ||
305 | static void pack_rcom_lock(struct dlm_rsb *r, struct dlm_lkb *lkb, | |
306 | struct rcom_lock *rl) | |
307 | { | |
308 | memset(rl, 0, sizeof(*rl)); | |
309 | ||
310 | rl->rl_ownpid = lkb->lkb_ownpid; | |
311 | rl->rl_lkid = lkb->lkb_id; | |
312 | rl->rl_exflags = lkb->lkb_exflags; | |
313 | rl->rl_flags = lkb->lkb_flags; | |
314 | rl->rl_lvbseq = lkb->lkb_lvbseq; | |
315 | rl->rl_rqmode = lkb->lkb_rqmode; | |
316 | rl->rl_grmode = lkb->lkb_grmode; | |
317 | rl->rl_status = lkb->lkb_status; | |
318 | rl->rl_wait_type = lkb->lkb_wait_type; | |
319 | ||
320 | if (lkb->lkb_bastaddr) | |
321 | rl->rl_asts |= AST_BAST; | |
322 | if (lkb->lkb_astaddr) | |
323 | rl->rl_asts |= AST_COMP; | |
324 | ||
e7fd4179 DT |
325 | rl->rl_namelen = r->res_length; |
326 | memcpy(rl->rl_name, r->res_name, r->res_length); | |
327 | ||
328 | /* FIXME: might we have an lvb without DLM_LKF_VALBLK set ? | |
329 | If so, receive_rcom_lock_args() won't take this copy. */ | |
330 | ||
331 | if (lkb->lkb_lvbptr) | |
332 | memcpy(rl->rl_lvb, lkb->lkb_lvbptr, r->res_ls->ls_lvblen); | |
333 | } | |
334 | ||
335 | int dlm_send_rcom_lock(struct dlm_rsb *r, struct dlm_lkb *lkb) | |
336 | { | |
337 | struct dlm_ls *ls = r->res_ls; | |
338 | struct dlm_rcom *rc; | |
339 | struct dlm_mhandle *mh; | |
340 | struct rcom_lock *rl; | |
341 | int error, len = sizeof(struct rcom_lock); | |
342 | ||
343 | if (lkb->lkb_lvbptr) | |
344 | len += ls->ls_lvblen; | |
345 | ||
346 | error = create_rcom(ls, r->res_nodeid, DLM_RCOM_LOCK, len, &rc, &mh); | |
347 | if (error) | |
348 | goto out; | |
349 | ||
350 | rl = (struct rcom_lock *) rc->rc_buf; | |
351 | pack_rcom_lock(r, lkb, rl); | |
352 | rc->rc_id = (unsigned long) r; | |
353 | ||
354 | send_rcom(ls, mh, rc); | |
355 | out: | |
356 | return error; | |
357 | } | |
358 | ||
359 | static void receive_rcom_lock(struct dlm_ls *ls, struct dlm_rcom *rc_in) | |
360 | { | |
361 | struct dlm_rcom *rc; | |
362 | struct dlm_mhandle *mh; | |
363 | int error, nodeid = rc_in->rc_header.h_nodeid; | |
364 | ||
365 | dlm_recover_master_copy(ls, rc_in); | |
366 | ||
367 | error = create_rcom(ls, nodeid, DLM_RCOM_LOCK_REPLY, | |
368 | sizeof(struct rcom_lock), &rc, &mh); | |
369 | if (error) | |
370 | return; | |
371 | ||
372 | /* We send back the same rcom_lock struct we received, but | |
373 | dlm_recover_master_copy() has filled in rl_remid and rl_result */ | |
374 | ||
375 | memcpy(rc->rc_buf, rc_in->rc_buf, sizeof(struct rcom_lock)); | |
376 | rc->rc_id = rc_in->rc_id; | |
377 | ||
378 | send_rcom(ls, mh, rc); | |
379 | } | |
380 | ||
381 | static void receive_rcom_lock_reply(struct dlm_ls *ls, struct dlm_rcom *rc_in) | |
382 | { | |
383 | uint32_t status = dlm_recover_status(ls); | |
384 | ||
385 | if (!(status & DLM_RS_DIR)) { | |
386 | log_debug(ls, "ignoring RCOM_LOCK_REPLY from %u", | |
387 | rc_in->rc_header.h_nodeid); | |
388 | return; | |
389 | } | |
390 | ||
391 | dlm_recover_process_copy(ls, rc_in); | |
392 | } | |
393 | ||
394 | static int send_ls_not_ready(int nodeid, struct dlm_rcom *rc_in) | |
395 | { | |
396 | struct dlm_rcom *rc; | |
1babdb45 | 397 | struct rcom_config *rf; |
e7fd4179 DT |
398 | struct dlm_mhandle *mh; |
399 | char *mb; | |
1babdb45 | 400 | int mb_len = sizeof(struct dlm_rcom) + sizeof(struct rcom_config); |
e7fd4179 DT |
401 | |
402 | mh = dlm_lowcomms_get_buffer(nodeid, mb_len, GFP_KERNEL, &mb); | |
403 | if (!mh) | |
404 | return -ENOBUFS; | |
405 | memset(mb, 0, mb_len); | |
406 | ||
407 | rc = (struct dlm_rcom *) mb; | |
408 | ||
409 | rc->rc_header.h_version = (DLM_HEADER_MAJOR | DLM_HEADER_MINOR); | |
410 | rc->rc_header.h_lockspace = rc_in->rc_header.h_lockspace; | |
411 | rc->rc_header.h_nodeid = dlm_our_nodeid(); | |
412 | rc->rc_header.h_length = mb_len; | |
413 | rc->rc_header.h_cmd = DLM_RCOM; | |
414 | ||
415 | rc->rc_type = DLM_RCOM_STATUS_REPLY; | |
f5888750 | 416 | rc->rc_id = rc_in->rc_id; |
e7fd4179 DT |
417 | rc->rc_result = -ESRCH; |
418 | ||
1babdb45 DT |
419 | rf = (struct rcom_config *) rc->rc_buf; |
420 | rf->rf_lvblen = -1; | |
421 | ||
e7fd4179 DT |
422 | dlm_rcom_out(rc); |
423 | dlm_lowcomms_commit_buffer(mh); | |
424 | ||
425 | return 0; | |
426 | } | |
427 | ||
428 | /* Called by dlm_recvd; corresponds to dlm_receive_message() but special | |
429 | recovery-only comms are sent through here. */ | |
430 | ||
431 | void dlm_receive_rcom(struct dlm_header *hd, int nodeid) | |
432 | { | |
433 | struct dlm_rcom *rc = (struct dlm_rcom *) hd; | |
434 | struct dlm_ls *ls; | |
435 | ||
436 | dlm_rcom_in(rc); | |
437 | ||
438 | /* If the lockspace doesn't exist then still send a status message | |
439 | back; it's possible that it just doesn't have its global_id yet. */ | |
440 | ||
441 | ls = dlm_find_lockspace_global(hd->h_lockspace); | |
442 | if (!ls) { | |
435618b7 DT |
443 | log_print("lockspace %x from %d type %x not found", |
444 | hd->h_lockspace, nodeid, rc->rc_type); | |
445 | if (rc->rc_type == DLM_RCOM_STATUS) | |
446 | send_ls_not_ready(nodeid, rc); | |
e7fd4179 DT |
447 | return; |
448 | } | |
449 | ||
450 | if (dlm_recovery_stopped(ls) && (rc->rc_type != DLM_RCOM_STATUS)) { | |
451 | log_error(ls, "ignoring recovery message %x from %d", | |
452 | rc->rc_type, nodeid); | |
453 | goto out; | |
454 | } | |
455 | ||
456 | if (nodeid != rc->rc_header.h_nodeid) { | |
457 | log_error(ls, "bad rcom nodeid %d from %d", | |
458 | rc->rc_header.h_nodeid, nodeid); | |
459 | goto out; | |
460 | } | |
461 | ||
462 | switch (rc->rc_type) { | |
463 | case DLM_RCOM_STATUS: | |
464 | receive_rcom_status(ls, rc); | |
465 | break; | |
466 | ||
467 | case DLM_RCOM_NAMES: | |
468 | receive_rcom_names(ls, rc); | |
469 | break; | |
470 | ||
471 | case DLM_RCOM_LOOKUP: | |
472 | receive_rcom_lookup(ls, rc); | |
473 | break; | |
474 | ||
475 | case DLM_RCOM_LOCK: | |
476 | receive_rcom_lock(ls, rc); | |
477 | break; | |
478 | ||
479 | case DLM_RCOM_STATUS_REPLY: | |
480 | receive_rcom_status_reply(ls, rc); | |
481 | break; | |
482 | ||
483 | case DLM_RCOM_NAMES_REPLY: | |
484 | receive_rcom_names_reply(ls, rc); | |
485 | break; | |
486 | ||
487 | case DLM_RCOM_LOOKUP_REPLY: | |
488 | receive_rcom_lookup_reply(ls, rc); | |
489 | break; | |
490 | ||
491 | case DLM_RCOM_LOCK_REPLY: | |
492 | receive_rcom_lock_reply(ls, rc); | |
493 | break; | |
494 | ||
495 | default: | |
496 | DLM_ASSERT(0, printk("rc_type=%x\n", rc->rc_type);); | |
497 | } | |
498 | out: | |
499 | dlm_put_lockspace(ls); | |
500 | } | |
501 |