1 | // SPDX-License-Identifier: GPL-2.0-only |
2 | /****************************************************************************** |
3 | ******************************************************************************* |
4 | ** |
5 | ** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. |
6 | ** Copyright (C) 2005-2008 Red Hat, Inc. All rights reserved. |
7 | ** |
8 | ** |
9 | ******************************************************************************* |
10 | ******************************************************************************/ |
11 | |
12 | #include "dlm_internal.h" |
13 | #include "lockspace.h" |
14 | #include "member.h" |
15 | #include "lowcomms.h" |
16 | #include "midcomms.h" |
17 | #include "rcom.h" |
18 | #include "recover.h" |
19 | #include "dir.h" |
20 | #include "config.h" |
21 | #include "memory.h" |
22 | #include "lock.h" |
23 | #include "util.h" |
24 | |
25 | static int rcom_response(struct dlm_ls *ls) |
26 | { |
27 | return test_bit(LSFL_RCOM_READY, &ls->ls_flags); |
28 | } |
29 | |
30 | static void _create_rcom(struct dlm_ls *ls, int to_nodeid, int type, int len, |
31 | struct dlm_rcom **rc_ret, char *mb, int mb_len, |
32 | uint64_t seq) |
33 | { |
34 | struct dlm_rcom *rc; |
35 | |
36 | rc = (struct dlm_rcom *) mb; |
37 | |
38 | rc->rc_header.h_version = cpu_to_le32(DLM_HEADER_MAJOR | DLM_HEADER_MINOR); |
39 | rc->rc_header.u.h_lockspace = cpu_to_le32(ls->ls_global_id); |
40 | rc->rc_header.h_nodeid = cpu_to_le32(dlm_our_nodeid()); |
41 | rc->rc_header.h_length = cpu_to_le16(mb_len); |
42 | rc->rc_header.h_cmd = DLM_RCOM; |
43 | |
44 | rc->rc_type = cpu_to_le32(type); |
45 | rc->rc_seq = cpu_to_le64(seq); |
46 | |
47 | *rc_ret = rc; |
48 | } |
49 | |
50 | static int create_rcom(struct dlm_ls *ls, int to_nodeid, int type, int len, |
51 | struct dlm_rcom **rc_ret, struct dlm_mhandle **mh_ret, |
52 | uint64_t seq) |
53 | { |
54 | int mb_len = sizeof(struct dlm_rcom) + len; |
55 | struct dlm_mhandle *mh; |
56 | char *mb; |
57 | |
58 | mh = dlm_midcomms_get_mhandle(nodeid: to_nodeid, len: mb_len, GFP_NOFS, ppc: &mb); |
59 | if (!mh) { |
60 | log_print("%s to %d type %d len %d ENOBUFS" , |
61 | __func__, to_nodeid, type, len); |
62 | return -ENOBUFS; |
63 | } |
64 | |
65 | _create_rcom(ls, to_nodeid, type, len, rc_ret, mb, mb_len, seq); |
66 | *mh_ret = mh; |
67 | return 0; |
68 | } |
69 | |
70 | static int create_rcom_stateless(struct dlm_ls *ls, int to_nodeid, int type, |
71 | int len, struct dlm_rcom **rc_ret, |
72 | struct dlm_msg **msg_ret, uint64_t seq) |
73 | { |
74 | int mb_len = sizeof(struct dlm_rcom) + len; |
75 | struct dlm_msg *msg; |
76 | char *mb; |
77 | |
78 | msg = dlm_lowcomms_new_msg(nodeid: to_nodeid, len: mb_len, GFP_NOFS, ppc: &mb, |
79 | NULL, NULL); |
80 | if (!msg) { |
81 | log_print("create_rcom to %d type %d len %d ENOBUFS" , |
82 | to_nodeid, type, len); |
83 | return -ENOBUFS; |
84 | } |
85 | |
86 | _create_rcom(ls, to_nodeid, type, len, rc_ret, mb, mb_len, seq); |
87 | *msg_ret = msg; |
88 | return 0; |
89 | } |
90 | |
91 | static void send_rcom(struct dlm_mhandle *mh, struct dlm_rcom *rc) |
92 | { |
93 | dlm_midcomms_commit_mhandle(mh, NULL, namelen: 0); |
94 | } |
95 | |
96 | static void send_rcom_stateless(struct dlm_msg *msg, struct dlm_rcom *rc) |
97 | { |
98 | dlm_lowcomms_commit_msg(msg); |
99 | dlm_lowcomms_put_msg(msg); |
100 | } |
101 | |
102 | static void set_rcom_status(struct dlm_ls *ls, struct rcom_status *rs, |
103 | uint32_t flags) |
104 | { |
105 | rs->rs_flags = cpu_to_le32(flags); |
106 | } |
107 | |
108 | /* When replying to a status request, a node also sends back its |
109 | configuration values. The requesting node then checks that the remote |
110 | node is configured the same way as itself. */ |
111 | |
112 | static void set_rcom_config(struct dlm_ls *ls, struct rcom_config *rf, |
113 | uint32_t num_slots) |
114 | { |
115 | rf->rf_lvblen = cpu_to_le32(ls->ls_lvblen); |
116 | rf->rf_lsflags = cpu_to_le32(ls->ls_exflags); |
117 | |
118 | rf->rf_our_slot = cpu_to_le16(ls->ls_slot); |
119 | rf->rf_num_slots = cpu_to_le16(num_slots); |
120 | rf->rf_generation = cpu_to_le32(ls->ls_generation); |
121 | } |
122 | |
123 | static int check_rcom_config(struct dlm_ls *ls, struct dlm_rcom *rc, int nodeid) |
124 | { |
125 | struct rcom_config *rf = (struct rcom_config *) rc->rc_buf; |
126 | |
127 | if ((le32_to_cpu(rc->rc_header.h_version) & 0xFFFF0000) != DLM_HEADER_MAJOR) { |
128 | log_error(ls, "version mismatch: %x nodeid %d: %x" , |
129 | DLM_HEADER_MAJOR | DLM_HEADER_MINOR, nodeid, |
130 | le32_to_cpu(rc->rc_header.h_version)); |
131 | return -EPROTO; |
132 | } |
133 | |
134 | if (le32_to_cpu(rf->rf_lvblen) != ls->ls_lvblen || |
135 | le32_to_cpu(rf->rf_lsflags) != ls->ls_exflags) { |
136 | log_error(ls, "config mismatch: %d,%x nodeid %d: %d,%x" , |
137 | ls->ls_lvblen, ls->ls_exflags, nodeid, |
138 | le32_to_cpu(rf->rf_lvblen), |
139 | le32_to_cpu(rf->rf_lsflags)); |
140 | return -EPROTO; |
141 | } |
142 | return 0; |
143 | } |
144 | |
145 | static void allow_sync_reply(struct dlm_ls *ls, __le64 *new_seq) |
146 | { |
147 | spin_lock(lock: &ls->ls_rcom_spin); |
148 | *new_seq = cpu_to_le64(++ls->ls_rcom_seq); |
149 | set_bit(LSFL_RCOM_WAIT, addr: &ls->ls_flags); |
150 | spin_unlock(lock: &ls->ls_rcom_spin); |
151 | } |
152 | |
153 | static void disallow_sync_reply(struct dlm_ls *ls) |
154 | { |
155 | spin_lock(lock: &ls->ls_rcom_spin); |
156 | clear_bit(LSFL_RCOM_WAIT, addr: &ls->ls_flags); |
157 | clear_bit(LSFL_RCOM_READY, addr: &ls->ls_flags); |
158 | spin_unlock(lock: &ls->ls_rcom_spin); |
159 | } |
160 | |
161 | /* |
162 | * low nodeid gathers one slot value at a time from each node. |
163 | * it sets need_slots=0, and saves rf_our_slot returned from each |
164 | * rcom_config. |
165 | * |
166 | * other nodes gather all slot values at once from the low nodeid. |
167 | * they set need_slots=1, and ignore the rf_our_slot returned from each |
168 | * rcom_config. they use the rf_num_slots returned from the low |
169 | * node's rcom_config. |
170 | */ |
171 | |
172 | int dlm_rcom_status(struct dlm_ls *ls, int nodeid, uint32_t status_flags, |
173 | uint64_t seq) |
174 | { |
175 | struct dlm_rcom *rc; |
176 | struct dlm_msg *msg; |
177 | int error = 0; |
178 | |
179 | ls->ls_recover_nodeid = nodeid; |
180 | |
181 | if (nodeid == dlm_our_nodeid()) { |
182 | rc = ls->ls_recover_buf; |
183 | rc->rc_result = cpu_to_le32(dlm_recover_status(ls)); |
184 | goto out; |
185 | } |
186 | |
187 | retry: |
188 | error = create_rcom_stateless(ls, to_nodeid: nodeid, DLM_RCOM_STATUS, |
189 | len: sizeof(struct rcom_status), rc_ret: &rc, msg_ret: &msg, |
190 | seq); |
191 | if (error) |
192 | goto out; |
193 | |
194 | set_rcom_status(ls, rs: (struct rcom_status *)rc->rc_buf, flags: status_flags); |
195 | |
196 | allow_sync_reply(ls, new_seq: &rc->rc_id); |
197 | memset(ls->ls_recover_buf, 0, DLM_MAX_SOCKET_BUFSIZE); |
198 | |
199 | send_rcom_stateless(msg, rc); |
200 | |
201 | error = dlm_wait_function(ls, testfn: &rcom_response); |
202 | disallow_sync_reply(ls); |
203 | if (error == -ETIMEDOUT) |
204 | goto retry; |
205 | if (error) |
206 | goto out; |
207 | |
208 | rc = ls->ls_recover_buf; |
209 | |
210 | if (rc->rc_result == cpu_to_le32(-ESRCH)) { |
211 | /* we pretend the remote lockspace exists with 0 status */ |
212 | log_debug(ls, "remote node %d not ready" , nodeid); |
213 | rc->rc_result = 0; |
214 | error = 0; |
215 | } else { |
216 | error = check_rcom_config(ls, rc, nodeid); |
217 | } |
218 | |
219 | /* the caller looks at rc_result for the remote recovery status */ |
220 | out: |
221 | return error; |
222 | } |
223 | |
224 | static void receive_rcom_status(struct dlm_ls *ls, |
225 | const struct dlm_rcom *rc_in, |
226 | uint64_t seq) |
227 | { |
228 | struct dlm_rcom *rc; |
229 | struct rcom_status *rs; |
230 | uint32_t status; |
231 | int nodeid = le32_to_cpu(rc_in->rc_header.h_nodeid); |
232 | int len = sizeof(struct rcom_config); |
233 | struct dlm_msg *msg; |
234 | int num_slots = 0; |
235 | int error; |
236 | |
237 | if (!dlm_slots_version(h: &rc_in->rc_header)) { |
238 | status = dlm_recover_status(ls); |
239 | goto do_create; |
240 | } |
241 | |
242 | rs = (struct rcom_status *)rc_in->rc_buf; |
243 | |
244 | if (!(le32_to_cpu(rs->rs_flags) & DLM_RSF_NEED_SLOTS)) { |
245 | status = dlm_recover_status(ls); |
246 | goto do_create; |
247 | } |
248 | |
249 | spin_lock(lock: &ls->ls_recover_lock); |
250 | status = ls->ls_recover_status; |
251 | num_slots = ls->ls_num_slots; |
252 | spin_unlock(lock: &ls->ls_recover_lock); |
253 | len += num_slots * sizeof(struct rcom_slot); |
254 | |
255 | do_create: |
256 | error = create_rcom_stateless(ls, to_nodeid: nodeid, DLM_RCOM_STATUS_REPLY, |
257 | len, rc_ret: &rc, msg_ret: &msg, seq); |
258 | if (error) |
259 | return; |
260 | |
261 | rc->rc_id = rc_in->rc_id; |
262 | rc->rc_seq_reply = rc_in->rc_seq; |
263 | rc->rc_result = cpu_to_le32(status); |
264 | |
265 | set_rcom_config(ls, rf: (struct rcom_config *)rc->rc_buf, num_slots); |
266 | |
267 | if (!num_slots) |
268 | goto do_send; |
269 | |
270 | spin_lock(lock: &ls->ls_recover_lock); |
271 | if (ls->ls_num_slots != num_slots) { |
272 | spin_unlock(lock: &ls->ls_recover_lock); |
273 | log_debug(ls, "receive_rcom_status num_slots %d to %d" , |
274 | num_slots, ls->ls_num_slots); |
275 | rc->rc_result = 0; |
276 | set_rcom_config(ls, rf: (struct rcom_config *)rc->rc_buf, num_slots: 0); |
277 | goto do_send; |
278 | } |
279 | |
280 | dlm_slots_copy_out(ls, rc); |
281 | spin_unlock(lock: &ls->ls_recover_lock); |
282 | |
283 | do_send: |
284 | send_rcom_stateless(msg, rc); |
285 | } |
286 | |
287 | static void receive_sync_reply(struct dlm_ls *ls, const struct dlm_rcom *rc_in) |
288 | { |
289 | spin_lock(lock: &ls->ls_rcom_spin); |
290 | if (!test_bit(LSFL_RCOM_WAIT, &ls->ls_flags) || |
291 | le64_to_cpu(rc_in->rc_id) != ls->ls_rcom_seq) { |
292 | log_debug(ls, "reject reply %d from %d seq %llx expect %llx" , |
293 | le32_to_cpu(rc_in->rc_type), |
294 | le32_to_cpu(rc_in->rc_header.h_nodeid), |
295 | (unsigned long long)le64_to_cpu(rc_in->rc_id), |
296 | (unsigned long long)ls->ls_rcom_seq); |
297 | goto out; |
298 | } |
299 | memcpy(ls->ls_recover_buf, rc_in, |
300 | le16_to_cpu(rc_in->rc_header.h_length)); |
301 | set_bit(LSFL_RCOM_READY, addr: &ls->ls_flags); |
302 | clear_bit(LSFL_RCOM_WAIT, addr: &ls->ls_flags); |
303 | wake_up(&ls->ls_wait_general); |
304 | out: |
305 | spin_unlock(lock: &ls->ls_rcom_spin); |
306 | } |
307 | |
308 | int dlm_rcom_names(struct dlm_ls *ls, int nodeid, char *last_name, |
309 | int last_len, uint64_t seq) |
310 | { |
311 | struct dlm_mhandle *mh; |
312 | struct dlm_rcom *rc; |
313 | int error = 0; |
314 | |
315 | ls->ls_recover_nodeid = nodeid; |
316 | |
317 | retry: |
318 | error = create_rcom(ls, to_nodeid: nodeid, DLM_RCOM_NAMES, len: last_len, |
319 | rc_ret: &rc, mh_ret: &mh, seq); |
320 | if (error) |
321 | goto out; |
322 | memcpy(rc->rc_buf, last_name, last_len); |
323 | |
324 | allow_sync_reply(ls, new_seq: &rc->rc_id); |
325 | memset(ls->ls_recover_buf, 0, DLM_MAX_SOCKET_BUFSIZE); |
326 | |
327 | send_rcom(mh, rc); |
328 | |
329 | error = dlm_wait_function(ls, testfn: &rcom_response); |
330 | disallow_sync_reply(ls); |
331 | if (error == -ETIMEDOUT) |
332 | goto retry; |
333 | out: |
334 | return error; |
335 | } |
336 | |
337 | static void receive_rcom_names(struct dlm_ls *ls, const struct dlm_rcom *rc_in, |
338 | uint64_t seq) |
339 | { |
340 | struct dlm_mhandle *mh; |
341 | struct dlm_rcom *rc; |
342 | int error, inlen, outlen, nodeid; |
343 | |
344 | nodeid = le32_to_cpu(rc_in->rc_header.h_nodeid); |
345 | inlen = le16_to_cpu(rc_in->rc_header.h_length) - |
346 | sizeof(struct dlm_rcom); |
347 | outlen = DLM_MAX_APP_BUFSIZE - sizeof(struct dlm_rcom); |
348 | |
349 | error = create_rcom(ls, to_nodeid: nodeid, DLM_RCOM_NAMES_REPLY, len: outlen, |
350 | rc_ret: &rc, mh_ret: &mh, seq); |
351 | if (error) |
352 | return; |
353 | rc->rc_id = rc_in->rc_id; |
354 | rc->rc_seq_reply = rc_in->rc_seq; |
355 | |
356 | dlm_copy_master_names(ls, inbuf: rc_in->rc_buf, inlen, outbuf: rc->rc_buf, outlen, |
357 | nodeid); |
358 | send_rcom(mh, rc); |
359 | } |
360 | |
361 | int dlm_send_rcom_lookup(struct dlm_rsb *r, int dir_nodeid, uint64_t seq) |
362 | { |
363 | struct dlm_rcom *rc; |
364 | struct dlm_mhandle *mh; |
365 | struct dlm_ls *ls = r->res_ls; |
366 | int error; |
367 | |
368 | error = create_rcom(ls, to_nodeid: dir_nodeid, DLM_RCOM_LOOKUP, len: r->res_length, |
369 | rc_ret: &rc, mh_ret: &mh, seq); |
370 | if (error) |
371 | goto out; |
372 | memcpy(rc->rc_buf, r->res_name, r->res_length); |
373 | rc->rc_id = cpu_to_le64(r->res_id); |
374 | |
375 | send_rcom(mh, rc); |
376 | out: |
377 | return error; |
378 | } |
379 | |
380 | static void receive_rcom_lookup(struct dlm_ls *ls, |
381 | const struct dlm_rcom *rc_in, uint64_t seq) |
382 | { |
383 | struct dlm_rcom *rc; |
384 | struct dlm_mhandle *mh; |
385 | int error, ret_nodeid, nodeid = le32_to_cpu(rc_in->rc_header.h_nodeid); |
386 | int len = le16_to_cpu(rc_in->rc_header.h_length) - |
387 | sizeof(struct dlm_rcom); |
388 | |
389 | /* Old code would send this special id to trigger a debug dump. */ |
390 | if (rc_in->rc_id == cpu_to_le64(0xFFFFFFFF)) { |
391 | log_error(ls, "receive_rcom_lookup dump from %d" , nodeid); |
392 | dlm_dump_rsb_name(ls, name: rc_in->rc_buf, len); |
393 | return; |
394 | } |
395 | |
396 | error = create_rcom(ls, to_nodeid: nodeid, DLM_RCOM_LOOKUP_REPLY, len: 0, rc_ret: &rc, mh_ret: &mh, |
397 | seq); |
398 | if (error) |
399 | return; |
400 | |
401 | error = dlm_master_lookup(ls, from_nodeid: nodeid, name: rc_in->rc_buf, len, |
402 | DLM_LU_RECOVER_MASTER, r_nodeid: &ret_nodeid, NULL); |
403 | if (error) |
404 | ret_nodeid = error; |
405 | rc->rc_result = cpu_to_le32(ret_nodeid); |
406 | rc->rc_id = rc_in->rc_id; |
407 | rc->rc_seq_reply = rc_in->rc_seq; |
408 | |
409 | send_rcom(mh, rc); |
410 | } |
411 | |
412 | static void receive_rcom_lookup_reply(struct dlm_ls *ls, |
413 | const struct dlm_rcom *rc_in) |
414 | { |
415 | dlm_recover_master_reply(ls, rc: rc_in); |
416 | } |
417 | |
418 | static void pack_rcom_lock(struct dlm_rsb *r, struct dlm_lkb *lkb, |
419 | struct rcom_lock *rl) |
420 | { |
421 | memset(rl, 0, sizeof(*rl)); |
422 | |
423 | rl->rl_ownpid = cpu_to_le32(lkb->lkb_ownpid); |
424 | rl->rl_lkid = cpu_to_le32(lkb->lkb_id); |
425 | rl->rl_exflags = cpu_to_le32(lkb->lkb_exflags); |
426 | rl->rl_flags = cpu_to_le32(dlm_dflags_val(lkb)); |
427 | rl->rl_lvbseq = cpu_to_le32(lkb->lkb_lvbseq); |
428 | rl->rl_rqmode = lkb->lkb_rqmode; |
429 | rl->rl_grmode = lkb->lkb_grmode; |
430 | rl->rl_status = lkb->lkb_status; |
431 | rl->rl_wait_type = cpu_to_le16(lkb->lkb_wait_type); |
432 | |
433 | if (lkb->lkb_bastfn) |
434 | rl->rl_asts |= DLM_CB_BAST; |
435 | if (lkb->lkb_astfn) |
436 | rl->rl_asts |= DLM_CB_CAST; |
437 | |
438 | rl->rl_namelen = cpu_to_le16(r->res_length); |
439 | memcpy(rl->rl_name, r->res_name, r->res_length); |
440 | |
441 | /* FIXME: might we have an lvb without DLM_LKF_VALBLK set ? |
442 | If so, receive_rcom_lock_args() won't take this copy. */ |
443 | |
444 | if (lkb->lkb_lvbptr) |
445 | memcpy(rl->rl_lvb, lkb->lkb_lvbptr, r->res_ls->ls_lvblen); |
446 | } |
447 | |
448 | int dlm_send_rcom_lock(struct dlm_rsb *r, struct dlm_lkb *lkb, uint64_t seq) |
449 | { |
450 | struct dlm_ls *ls = r->res_ls; |
451 | struct dlm_rcom *rc; |
452 | struct dlm_mhandle *mh; |
453 | struct rcom_lock *rl; |
454 | int error, len = sizeof(struct rcom_lock); |
455 | |
456 | if (lkb->lkb_lvbptr) |
457 | len += ls->ls_lvblen; |
458 | |
459 | error = create_rcom(ls, to_nodeid: r->res_nodeid, DLM_RCOM_LOCK, len, rc_ret: &rc, mh_ret: &mh, |
460 | seq); |
461 | if (error) |
462 | goto out; |
463 | |
464 | rl = (struct rcom_lock *) rc->rc_buf; |
465 | pack_rcom_lock(r, lkb, rl); |
466 | rc->rc_id = cpu_to_le64((uintptr_t)r); |
467 | |
468 | send_rcom(mh, rc); |
469 | out: |
470 | return error; |
471 | } |
472 | |
473 | /* needs at least dlm_rcom + rcom_lock */ |
474 | static void receive_rcom_lock(struct dlm_ls *ls, const struct dlm_rcom *rc_in, |
475 | uint64_t seq) |
476 | { |
477 | __le32 rl_remid, rl_result; |
478 | struct rcom_lock *rl; |
479 | struct dlm_rcom *rc; |
480 | struct dlm_mhandle *mh; |
481 | int error, nodeid = le32_to_cpu(rc_in->rc_header.h_nodeid); |
482 | |
483 | dlm_recover_master_copy(ls, rc: rc_in, rl_remid: &rl_remid, rl_result: &rl_result); |
484 | |
485 | error = create_rcom(ls, to_nodeid: nodeid, DLM_RCOM_LOCK_REPLY, |
486 | len: sizeof(struct rcom_lock), rc_ret: &rc, mh_ret: &mh, seq); |
487 | if (error) |
488 | return; |
489 | |
490 | memcpy(rc->rc_buf, rc_in->rc_buf, sizeof(struct rcom_lock)); |
491 | rl = (struct rcom_lock *)rc->rc_buf; |
492 | /* set rl_remid and rl_result from dlm_recover_master_copy() */ |
493 | rl->rl_remid = rl_remid; |
494 | rl->rl_result = rl_result; |
495 | |
496 | rc->rc_id = rc_in->rc_id; |
497 | rc->rc_seq_reply = rc_in->rc_seq; |
498 | |
499 | send_rcom(mh, rc); |
500 | } |
501 | |
502 | /* If the lockspace doesn't exist then still send a status message |
503 | back; it's possible that it just doesn't have its global_id yet. */ |
504 | |
505 | int dlm_send_ls_not_ready(int nodeid, const struct dlm_rcom *rc_in) |
506 | { |
507 | struct dlm_rcom *rc; |
508 | struct rcom_config *rf; |
509 | struct dlm_mhandle *mh; |
510 | char *mb; |
511 | int mb_len = sizeof(struct dlm_rcom) + sizeof(struct rcom_config); |
512 | |
513 | mh = dlm_midcomms_get_mhandle(nodeid, len: mb_len, GFP_NOFS, ppc: &mb); |
514 | if (!mh) |
515 | return -ENOBUFS; |
516 | |
517 | rc = (struct dlm_rcom *) mb; |
518 | |
519 | rc->rc_header.h_version = cpu_to_le32(DLM_HEADER_MAJOR | DLM_HEADER_MINOR); |
520 | rc->rc_header.u.h_lockspace = rc_in->rc_header.u.h_lockspace; |
521 | rc->rc_header.h_nodeid = cpu_to_le32(dlm_our_nodeid()); |
522 | rc->rc_header.h_length = cpu_to_le16(mb_len); |
523 | rc->rc_header.h_cmd = DLM_RCOM; |
524 | |
525 | rc->rc_type = cpu_to_le32(DLM_RCOM_STATUS_REPLY); |
526 | rc->rc_id = rc_in->rc_id; |
527 | rc->rc_seq_reply = rc_in->rc_seq; |
528 | rc->rc_result = cpu_to_le32(-ESRCH); |
529 | |
530 | rf = (struct rcom_config *) rc->rc_buf; |
531 | rf->rf_lvblen = cpu_to_le32(~0U); |
532 | |
533 | dlm_midcomms_commit_mhandle(mh, NULL, namelen: 0); |
534 | |
535 | return 0; |
536 | } |
537 | |
538 | /* |
539 | * Ignore messages for stage Y before we set |
540 | * recover_status bit for stage X: |
541 | * |
542 | * recover_status = 0 |
543 | * |
544 | * dlm_recover_members() |
545 | * - send nothing |
546 | * - recv nothing |
547 | * - ignore NAMES, NAMES_REPLY |
548 | * - ignore LOOKUP, LOOKUP_REPLY |
549 | * - ignore LOCK, LOCK_REPLY |
550 | * |
551 | * recover_status |= NODES |
552 | * |
553 | * dlm_recover_members_wait() |
554 | * |
555 | * dlm_recover_directory() |
556 | * - send NAMES |
557 | * - recv NAMES_REPLY |
558 | * - ignore LOOKUP, LOOKUP_REPLY |
559 | * - ignore LOCK, LOCK_REPLY |
560 | * |
561 | * recover_status |= DIR |
562 | * |
563 | * dlm_recover_directory_wait() |
564 | * |
565 | * dlm_recover_masters() |
566 | * - send LOOKUP |
567 | * - recv LOOKUP_REPLY |
568 | * |
569 | * dlm_recover_locks() |
570 | * - send LOCKS |
571 | * - recv LOCKS_REPLY |
572 | * |
573 | * recover_status |= LOCKS |
574 | * |
575 | * dlm_recover_locks_wait() |
576 | * |
577 | * recover_status |= DONE |
578 | */ |
579 | |
580 | /* Called by dlm_recv; corresponds to dlm_receive_message() but special |
581 | recovery-only comms are sent through here. */ |
582 | |
583 | void dlm_receive_rcom(struct dlm_ls *ls, const struct dlm_rcom *rc, int nodeid) |
584 | { |
585 | int lock_size = sizeof(struct dlm_rcom) + sizeof(struct rcom_lock); |
586 | int stop, reply = 0, names = 0, lookup = 0, lock = 0; |
587 | uint32_t status; |
588 | uint64_t seq; |
589 | |
590 | switch (rc->rc_type) { |
591 | case cpu_to_le32(DLM_RCOM_STATUS_REPLY): |
592 | reply = 1; |
593 | break; |
594 | case cpu_to_le32(DLM_RCOM_NAMES): |
595 | names = 1; |
596 | break; |
597 | case cpu_to_le32(DLM_RCOM_NAMES_REPLY): |
598 | names = 1; |
599 | reply = 1; |
600 | break; |
601 | case cpu_to_le32(DLM_RCOM_LOOKUP): |
602 | lookup = 1; |
603 | break; |
604 | case cpu_to_le32(DLM_RCOM_LOOKUP_REPLY): |
605 | lookup = 1; |
606 | reply = 1; |
607 | break; |
608 | case cpu_to_le32(DLM_RCOM_LOCK): |
609 | lock = 1; |
610 | break; |
611 | case cpu_to_le32(DLM_RCOM_LOCK_REPLY): |
612 | lock = 1; |
613 | reply = 1; |
614 | break; |
615 | } |
616 | |
617 | spin_lock(lock: &ls->ls_recover_lock); |
618 | status = ls->ls_recover_status; |
619 | stop = dlm_recovery_stopped(ls); |
620 | seq = ls->ls_recover_seq; |
621 | spin_unlock(lock: &ls->ls_recover_lock); |
622 | |
623 | if (stop && (rc->rc_type != cpu_to_le32(DLM_RCOM_STATUS))) |
624 | goto ignore; |
625 | |
626 | if (reply && (le64_to_cpu(rc->rc_seq_reply) != seq)) |
627 | goto ignore; |
628 | |
629 | if (!(status & DLM_RS_NODES) && (names || lookup || lock)) |
630 | goto ignore; |
631 | |
632 | if (!(status & DLM_RS_DIR) && (lookup || lock)) |
633 | goto ignore; |
634 | |
635 | switch (rc->rc_type) { |
636 | case cpu_to_le32(DLM_RCOM_STATUS): |
637 | receive_rcom_status(ls, rc_in: rc, seq); |
638 | break; |
639 | |
640 | case cpu_to_le32(DLM_RCOM_NAMES): |
641 | receive_rcom_names(ls, rc_in: rc, seq); |
642 | break; |
643 | |
644 | case cpu_to_le32(DLM_RCOM_LOOKUP): |
645 | receive_rcom_lookup(ls, rc_in: rc, seq); |
646 | break; |
647 | |
648 | case cpu_to_le32(DLM_RCOM_LOCK): |
649 | if (le16_to_cpu(rc->rc_header.h_length) < lock_size) |
650 | goto Eshort; |
651 | receive_rcom_lock(ls, rc_in: rc, seq); |
652 | break; |
653 | |
654 | case cpu_to_le32(DLM_RCOM_STATUS_REPLY): |
655 | receive_sync_reply(ls, rc_in: rc); |
656 | break; |
657 | |
658 | case cpu_to_le32(DLM_RCOM_NAMES_REPLY): |
659 | receive_sync_reply(ls, rc_in: rc); |
660 | break; |
661 | |
662 | case cpu_to_le32(DLM_RCOM_LOOKUP_REPLY): |
663 | receive_rcom_lookup_reply(ls, rc_in: rc); |
664 | break; |
665 | |
666 | case cpu_to_le32(DLM_RCOM_LOCK_REPLY): |
667 | if (le16_to_cpu(rc->rc_header.h_length) < lock_size) |
668 | goto Eshort; |
669 | dlm_recover_process_copy(ls, rc, seq); |
670 | break; |
671 | |
672 | default: |
673 | log_error(ls, "receive_rcom bad type %d" , |
674 | le32_to_cpu(rc->rc_type)); |
675 | } |
676 | return; |
677 | |
678 | ignore: |
679 | log_limit(ls, "dlm_receive_rcom ignore msg %d " |
680 | "from %d %llu %llu recover seq %llu sts %x gen %u" , |
681 | le32_to_cpu(rc->rc_type), |
682 | nodeid, |
683 | (unsigned long long)le64_to_cpu(rc->rc_seq), |
684 | (unsigned long long)le64_to_cpu(rc->rc_seq_reply), |
685 | (unsigned long long)seq, |
686 | status, ls->ls_generation); |
687 | return; |
688 | Eshort: |
689 | log_error(ls, "recovery message %d from %d is too short" , |
690 | le32_to_cpu(rc->rc_type), nodeid); |
691 | } |
692 | |
693 | |