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