ac5c616c969643addc81c8920f4da2cf439356e9
[linux-3.10.git] / fs / dlm / rcom.c
1 /******************************************************************************
2 *******************************************************************************
3 **
4 **  Copyright (C) Sistina Software, Inc.  1997-2003  All rights reserved.
5 **  Copyright (C) 2005-2008 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 #include "member.h"
27
28
29 static int rcom_response(struct dlm_ls *ls)
30 {
31         return test_bit(LSFL_RCOM_READY, &ls->ls_flags);
32 }
33
34 static int create_rcom(struct dlm_ls *ls, int to_nodeid, int type, int len,
35                        struct dlm_rcom **rc_ret, struct dlm_mhandle **mh_ret)
36 {
37         struct dlm_rcom *rc;
38         struct dlm_mhandle *mh;
39         char *mb;
40         int mb_len = sizeof(struct dlm_rcom) + len;
41
42         mh = dlm_lowcomms_get_buffer(to_nodeid, mb_len, GFP_NOFS, &mb);
43         if (!mh) {
44                 log_print("create_rcom to %d type %d len %d ENOBUFS",
45                           to_nodeid, type, len);
46                 return -ENOBUFS;
47         }
48         memset(mb, 0, mb_len);
49
50         rc = (struct dlm_rcom *) mb;
51
52         rc->rc_header.h_version = (DLM_HEADER_MAJOR | DLM_HEADER_MINOR);
53         rc->rc_header.h_lockspace = ls->ls_global_id;
54         rc->rc_header.h_nodeid = dlm_our_nodeid();
55         rc->rc_header.h_length = mb_len;
56         rc->rc_header.h_cmd = DLM_RCOM;
57
58         rc->rc_type = type;
59
60         spin_lock(&ls->ls_recover_lock);
61         rc->rc_seq = ls->ls_recover_seq;
62         spin_unlock(&ls->ls_recover_lock);
63
64         *mh_ret = mh;
65         *rc_ret = rc;
66         return 0;
67 }
68
69 static void send_rcom(struct dlm_ls *ls, struct dlm_mhandle *mh,
70                       struct dlm_rcom *rc)
71 {
72         dlm_rcom_out(rc);
73         dlm_lowcomms_commit_buffer(mh);
74 }
75
76 static void set_rcom_status(struct dlm_ls *ls, struct rcom_status *rs,
77                             uint32_t flags)
78 {
79         rs->rs_flags = cpu_to_le32(flags);
80 }
81
82 /* When replying to a status request, a node also sends back its
83    configuration values.  The requesting node then checks that the remote
84    node is configured the same way as itself. */
85
86 static void set_rcom_config(struct dlm_ls *ls, struct rcom_config *rf,
87                             uint32_t num_slots)
88 {
89         rf->rf_lvblen = cpu_to_le32(ls->ls_lvblen);
90         rf->rf_lsflags = cpu_to_le32(ls->ls_exflags);
91
92         rf->rf_our_slot = cpu_to_le16(ls->ls_slot);
93         rf->rf_num_slots = cpu_to_le16(num_slots);
94         rf->rf_generation =  cpu_to_le32(ls->ls_generation);
95 }
96
97 static int check_rcom_config(struct dlm_ls *ls, struct dlm_rcom *rc, int nodeid)
98 {
99         struct rcom_config *rf = (struct rcom_config *) rc->rc_buf;
100
101         if ((rc->rc_header.h_version & 0xFFFF0000) != DLM_HEADER_MAJOR) {
102                 log_error(ls, "version mismatch: %x nodeid %d: %x",
103                           DLM_HEADER_MAJOR | DLM_HEADER_MINOR, nodeid,
104                           rc->rc_header.h_version);
105                 return -EPROTO;
106         }
107
108         if (le32_to_cpu(rf->rf_lvblen) != ls->ls_lvblen ||
109             le32_to_cpu(rf->rf_lsflags) != ls->ls_exflags) {
110                 log_error(ls, "config mismatch: %d,%x nodeid %d: %d,%x",
111                           ls->ls_lvblen, ls->ls_exflags, nodeid,
112                           le32_to_cpu(rf->rf_lvblen),
113                           le32_to_cpu(rf->rf_lsflags));
114                 return -EPROTO;
115         }
116         return 0;
117 }
118
119 static void allow_sync_reply(struct dlm_ls *ls, uint64_t *new_seq)
120 {
121         spin_lock(&ls->ls_rcom_spin);
122         *new_seq = ++ls->ls_rcom_seq;
123         set_bit(LSFL_RCOM_WAIT, &ls->ls_flags);
124         spin_unlock(&ls->ls_rcom_spin);
125 }
126
127 static void disallow_sync_reply(struct dlm_ls *ls)
128 {
129         spin_lock(&ls->ls_rcom_spin);
130         clear_bit(LSFL_RCOM_WAIT, &ls->ls_flags);
131         clear_bit(LSFL_RCOM_READY, &ls->ls_flags);
132         spin_unlock(&ls->ls_rcom_spin);
133 }
134
135 /*
136  * low nodeid gathers one slot value at a time from each node.
137  * it sets need_slots=0, and saves rf_our_slot returned from each
138  * rcom_config.
139  *
140  * other nodes gather all slot values at once from the low nodeid.
141  * they set need_slots=1, and ignore the rf_our_slot returned from each
142  * rcom_config.  they use the rf_num_slots returned from the low
143  * node's rcom_config.
144  */
145
146 int dlm_rcom_status(struct dlm_ls *ls, int nodeid, uint32_t status_flags)
147 {
148         struct dlm_rcom *rc;
149         struct dlm_mhandle *mh;
150         int error = 0;
151
152         ls->ls_recover_nodeid = nodeid;
153
154         if (nodeid == dlm_our_nodeid()) {
155                 rc = ls->ls_recover_buf;
156                 rc->rc_result = dlm_recover_status(ls);
157                 goto out;
158         }
159
160         error = create_rcom(ls, nodeid, DLM_RCOM_STATUS,
161                             sizeof(struct rcom_status), &rc, &mh);
162         if (error)
163                 goto out;
164
165         set_rcom_status(ls, (struct rcom_status *)rc->rc_buf, status_flags);
166
167         allow_sync_reply(ls, &rc->rc_id);
168         memset(ls->ls_recover_buf, 0, dlm_config.ci_buffer_size);
169
170         send_rcom(ls, mh, rc);
171
172         error = dlm_wait_function(ls, &rcom_response);
173         disallow_sync_reply(ls);
174         if (error)
175                 goto out;
176
177         rc = ls->ls_recover_buf;
178
179         if (rc->rc_result == -ESRCH) {
180                 /* we pretend the remote lockspace exists with 0 status */
181                 log_debug(ls, "remote node %d not ready", nodeid);
182                 rc->rc_result = 0;
183                 error = 0;
184         } else {
185                 error = check_rcom_config(ls, rc, nodeid);
186         }
187
188         /* the caller looks at rc_result for the remote recovery status */
189  out:
190         return error;
191 }
192
193 static void receive_rcom_status(struct dlm_ls *ls, struct dlm_rcom *rc_in)
194 {
195         struct dlm_rcom *rc;
196         struct dlm_mhandle *mh;
197         struct rcom_status *rs;
198         uint32_t status;
199         int nodeid = rc_in->rc_header.h_nodeid;
200         int len = sizeof(struct rcom_config);
201         int num_slots = 0;
202         int error;
203
204         if (!dlm_slots_version(&rc_in->rc_header)) {
205                 status = dlm_recover_status(ls);
206                 goto do_create;
207         }
208
209         rs = (struct rcom_status *)rc_in->rc_buf;
210
211         if (!(rs->rs_flags & DLM_RSF_NEED_SLOTS)) {
212                 status = dlm_recover_status(ls);
213                 goto do_create;
214         }
215
216         spin_lock(&ls->ls_recover_lock);
217         status = ls->ls_recover_status;
218         num_slots = ls->ls_num_slots;
219         spin_unlock(&ls->ls_recover_lock);
220         len += num_slots * sizeof(struct rcom_slot);
221
222  do_create:
223         error = create_rcom(ls, nodeid, DLM_RCOM_STATUS_REPLY,
224                             len, &rc, &mh);
225         if (error)
226                 return;
227
228         rc->rc_id = rc_in->rc_id;
229         rc->rc_seq_reply = rc_in->rc_seq;
230         rc->rc_result = status;
231
232         set_rcom_config(ls, (struct rcom_config *)rc->rc_buf, num_slots);
233
234         if (!num_slots)
235                 goto do_send;
236
237         spin_lock(&ls->ls_recover_lock);
238         if (ls->ls_num_slots != num_slots) {
239                 spin_unlock(&ls->ls_recover_lock);
240                 log_debug(ls, "receive_rcom_status num_slots %d to %d",
241                           num_slots, ls->ls_num_slots);
242                 rc->rc_result = 0;
243                 set_rcom_config(ls, (struct rcom_config *)rc->rc_buf, 0);
244                 goto do_send;
245         }
246
247         dlm_slots_copy_out(ls, rc);
248         spin_unlock(&ls->ls_recover_lock);
249
250  do_send:
251         send_rcom(ls, mh, rc);
252 }
253
254 static void receive_sync_reply(struct dlm_ls *ls, struct dlm_rcom *rc_in)
255 {
256         spin_lock(&ls->ls_rcom_spin);
257         if (!test_bit(LSFL_RCOM_WAIT, &ls->ls_flags) ||
258             rc_in->rc_id != ls->ls_rcom_seq) {
259                 log_debug(ls, "reject reply %d from %d seq %llx expect %llx",
260                           rc_in->rc_type, rc_in->rc_header.h_nodeid,
261                           (unsigned long long)rc_in->rc_id,
262                           (unsigned long long)ls->ls_rcom_seq);
263                 goto out;
264         }
265         memcpy(ls->ls_recover_buf, rc_in, rc_in->rc_header.h_length);
266         set_bit(LSFL_RCOM_READY, &ls->ls_flags);
267         clear_bit(LSFL_RCOM_WAIT, &ls->ls_flags);
268         wake_up(&ls->ls_wait_general);
269  out:
270         spin_unlock(&ls->ls_rcom_spin);
271 }
272
273 int dlm_rcom_names(struct dlm_ls *ls, int nodeid, char *last_name, int last_len)
274 {
275         struct dlm_rcom *rc;
276         struct dlm_mhandle *mh;
277         int error = 0;
278         int max_size = dlm_config.ci_buffer_size - sizeof(struct dlm_rcom);
279
280         ls->ls_recover_nodeid = nodeid;
281
282         if (nodeid == dlm_our_nodeid()) {
283                 ls->ls_recover_buf->rc_header.h_length =
284                         dlm_config.ci_buffer_size;
285                 dlm_copy_master_names(ls, last_name, last_len,
286                                       ls->ls_recover_buf->rc_buf,
287                                       max_size, nodeid);
288                 goto out;
289         }
290
291         error = create_rcom(ls, nodeid, DLM_RCOM_NAMES, last_len, &rc, &mh);
292         if (error)
293                 goto out;
294         memcpy(rc->rc_buf, last_name, last_len);
295
296         allow_sync_reply(ls, &rc->rc_id);
297         memset(ls->ls_recover_buf, 0, dlm_config.ci_buffer_size);
298
299         send_rcom(ls, mh, rc);
300
301         error = dlm_wait_function(ls, &rcom_response);
302         disallow_sync_reply(ls);
303  out:
304         return error;
305 }
306
307 static void receive_rcom_names(struct dlm_ls *ls, struct dlm_rcom *rc_in)
308 {
309         struct dlm_rcom *rc;
310         struct dlm_mhandle *mh;
311         int error, inlen, outlen, nodeid;
312
313         nodeid = rc_in->rc_header.h_nodeid;
314         inlen = rc_in->rc_header.h_length - sizeof(struct dlm_rcom);
315         outlen = dlm_config.ci_buffer_size - sizeof(struct dlm_rcom);
316
317         error = create_rcom(ls, nodeid, DLM_RCOM_NAMES_REPLY, outlen, &rc, &mh);
318         if (error)
319                 return;
320         rc->rc_id = rc_in->rc_id;
321         rc->rc_seq_reply = rc_in->rc_seq;
322
323         dlm_copy_master_names(ls, rc_in->rc_buf, inlen, rc->rc_buf, outlen,
324                               nodeid);
325         send_rcom(ls, mh, rc);
326 }
327
328 int dlm_send_rcom_lookup(struct dlm_rsb *r, int dir_nodeid)
329 {
330         struct dlm_rcom *rc;
331         struct dlm_mhandle *mh;
332         struct dlm_ls *ls = r->res_ls;
333         int error;
334
335         error = create_rcom(ls, dir_nodeid, DLM_RCOM_LOOKUP, r->res_length,
336                             &rc, &mh);
337         if (error)
338                 goto out;
339         memcpy(rc->rc_buf, r->res_name, r->res_length);
340         rc->rc_id = (unsigned long) r;
341
342         send_rcom(ls, mh, rc);
343  out:
344         return error;
345 }
346
347 static void receive_rcom_lookup(struct dlm_ls *ls, struct dlm_rcom *rc_in)
348 {
349         struct dlm_rcom *rc;
350         struct dlm_mhandle *mh;
351         int error, ret_nodeid, nodeid = rc_in->rc_header.h_nodeid;
352         int len = rc_in->rc_header.h_length - sizeof(struct dlm_rcom);
353
354         error = create_rcom(ls, nodeid, DLM_RCOM_LOOKUP_REPLY, 0, &rc, &mh);
355         if (error)
356                 return;
357
358         error = dlm_dir_lookup(ls, nodeid, rc_in->rc_buf, len, &ret_nodeid);
359         if (error)
360                 ret_nodeid = error;
361         rc->rc_result = ret_nodeid;
362         rc->rc_id = rc_in->rc_id;
363         rc->rc_seq_reply = rc_in->rc_seq;
364
365         send_rcom(ls, mh, rc);
366 }
367
368 static void receive_rcom_lookup_reply(struct dlm_ls *ls, struct dlm_rcom *rc_in)
369 {
370         dlm_recover_master_reply(ls, rc_in);
371 }
372
373 static void pack_rcom_lock(struct dlm_rsb *r, struct dlm_lkb *lkb,
374                            struct rcom_lock *rl)
375 {
376         memset(rl, 0, sizeof(*rl));
377
378         rl->rl_ownpid = cpu_to_le32(lkb->lkb_ownpid);
379         rl->rl_lkid = cpu_to_le32(lkb->lkb_id);
380         rl->rl_exflags = cpu_to_le32(lkb->lkb_exflags);
381         rl->rl_flags = cpu_to_le32(lkb->lkb_flags);
382         rl->rl_lvbseq = cpu_to_le32(lkb->lkb_lvbseq);
383         rl->rl_rqmode = lkb->lkb_rqmode;
384         rl->rl_grmode = lkb->lkb_grmode;
385         rl->rl_status = lkb->lkb_status;
386         rl->rl_wait_type = cpu_to_le16(lkb->lkb_wait_type);
387
388         if (lkb->lkb_bastfn)
389                 rl->rl_asts |= DLM_CB_BAST;
390         if (lkb->lkb_astfn)
391                 rl->rl_asts |= DLM_CB_CAST;
392
393         rl->rl_namelen = cpu_to_le16(r->res_length);
394         memcpy(rl->rl_name, r->res_name, r->res_length);
395
396         /* FIXME: might we have an lvb without DLM_LKF_VALBLK set ?
397            If so, receive_rcom_lock_args() won't take this copy. */
398
399         if (lkb->lkb_lvbptr)
400                 memcpy(rl->rl_lvb, lkb->lkb_lvbptr, r->res_ls->ls_lvblen);
401 }
402
403 int dlm_send_rcom_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
404 {
405         struct dlm_ls *ls = r->res_ls;
406         struct dlm_rcom *rc;
407         struct dlm_mhandle *mh;
408         struct rcom_lock *rl;
409         int error, len = sizeof(struct rcom_lock);
410
411         if (lkb->lkb_lvbptr)
412                 len += ls->ls_lvblen;
413
414         error = create_rcom(ls, r->res_nodeid, DLM_RCOM_LOCK, len, &rc, &mh);
415         if (error)
416                 goto out;
417
418         rl = (struct rcom_lock *) rc->rc_buf;
419         pack_rcom_lock(r, lkb, rl);
420         rc->rc_id = (unsigned long) r;
421
422         send_rcom(ls, mh, rc);
423  out:
424         return error;
425 }
426
427 /* needs at least dlm_rcom + rcom_lock */
428 static void receive_rcom_lock(struct dlm_ls *ls, struct dlm_rcom *rc_in)
429 {
430         struct dlm_rcom *rc;
431         struct dlm_mhandle *mh;
432         int error, nodeid = rc_in->rc_header.h_nodeid;
433
434         dlm_recover_master_copy(ls, rc_in);
435
436         error = create_rcom(ls, nodeid, DLM_RCOM_LOCK_REPLY,
437                             sizeof(struct rcom_lock), &rc, &mh);
438         if (error)
439                 return;
440
441         /* We send back the same rcom_lock struct we received, but
442            dlm_recover_master_copy() has filled in rl_remid and rl_result */
443
444         memcpy(rc->rc_buf, rc_in->rc_buf, sizeof(struct rcom_lock));
445         rc->rc_id = rc_in->rc_id;
446         rc->rc_seq_reply = rc_in->rc_seq;
447
448         send_rcom(ls, mh, rc);
449 }
450
451 /* If the lockspace doesn't exist then still send a status message
452    back; it's possible that it just doesn't have its global_id yet. */
453
454 int dlm_send_ls_not_ready(int nodeid, struct dlm_rcom *rc_in)
455 {
456         struct dlm_rcom *rc;
457         struct rcom_config *rf;
458         struct dlm_mhandle *mh;
459         char *mb;
460         int mb_len = sizeof(struct dlm_rcom) + sizeof(struct rcom_config);
461
462         mh = dlm_lowcomms_get_buffer(nodeid, mb_len, GFP_NOFS, &mb);
463         if (!mh)
464                 return -ENOBUFS;
465         memset(mb, 0, mb_len);
466
467         rc = (struct dlm_rcom *) mb;
468
469         rc->rc_header.h_version = (DLM_HEADER_MAJOR | DLM_HEADER_MINOR);
470         rc->rc_header.h_lockspace = rc_in->rc_header.h_lockspace;
471         rc->rc_header.h_nodeid = dlm_our_nodeid();
472         rc->rc_header.h_length = mb_len;
473         rc->rc_header.h_cmd = DLM_RCOM;
474
475         rc->rc_type = DLM_RCOM_STATUS_REPLY;
476         rc->rc_id = rc_in->rc_id;
477         rc->rc_seq_reply = rc_in->rc_seq;
478         rc->rc_result = -ESRCH;
479
480         rf = (struct rcom_config *) rc->rc_buf;
481         rf->rf_lvblen = cpu_to_le32(~0U);
482
483         dlm_rcom_out(rc);
484         dlm_lowcomms_commit_buffer(mh);
485
486         return 0;
487 }
488
489 static int is_old_reply(struct dlm_ls *ls, struct dlm_rcom *rc)
490 {
491         uint64_t seq;
492         int rv = 0;
493
494         switch (rc->rc_type) {
495         case DLM_RCOM_STATUS_REPLY:
496         case DLM_RCOM_NAMES_REPLY:
497         case DLM_RCOM_LOOKUP_REPLY:
498         case DLM_RCOM_LOCK_REPLY:
499                 spin_lock(&ls->ls_recover_lock);
500                 seq = ls->ls_recover_seq;
501                 spin_unlock(&ls->ls_recover_lock);
502                 if (rc->rc_seq_reply != seq) {
503                         log_debug(ls, "ignoring old reply %x from %d "
504                                       "seq_reply %llx expect %llx",
505                                       rc->rc_type, rc->rc_header.h_nodeid,
506                                       (unsigned long long)rc->rc_seq_reply,
507                                       (unsigned long long)seq);
508                         rv = 1;
509                 }
510         }
511         return rv;
512 }
513
514 /* Called by dlm_recv; corresponds to dlm_receive_message() but special
515    recovery-only comms are sent through here. */
516
517 void dlm_receive_rcom(struct dlm_ls *ls, struct dlm_rcom *rc, int nodeid)
518 {
519         int lock_size = sizeof(struct dlm_rcom) + sizeof(struct rcom_lock);
520
521         if (dlm_recovery_stopped(ls) && (rc->rc_type != DLM_RCOM_STATUS)) {
522                 log_debug(ls, "ignoring recovery message %x from %d",
523                           rc->rc_type, nodeid);
524                 goto out;
525         }
526
527         if (is_old_reply(ls, rc))
528                 goto out;
529
530         switch (rc->rc_type) {
531         case DLM_RCOM_STATUS:
532                 receive_rcom_status(ls, rc);
533                 break;
534
535         case DLM_RCOM_NAMES:
536                 receive_rcom_names(ls, rc);
537                 break;
538
539         case DLM_RCOM_LOOKUP:
540                 receive_rcom_lookup(ls, rc);
541                 break;
542
543         case DLM_RCOM_LOCK:
544                 if (rc->rc_header.h_length < lock_size)
545                         goto Eshort;
546                 receive_rcom_lock(ls, rc);
547                 break;
548
549         case DLM_RCOM_STATUS_REPLY:
550                 receive_sync_reply(ls, rc);
551                 break;
552
553         case DLM_RCOM_NAMES_REPLY:
554                 receive_sync_reply(ls, rc);
555                 break;
556
557         case DLM_RCOM_LOOKUP_REPLY:
558                 receive_rcom_lookup_reply(ls, rc);
559                 break;
560
561         case DLM_RCOM_LOCK_REPLY:
562                 if (rc->rc_header.h_length < lock_size)
563                         goto Eshort;
564                 dlm_recover_process_copy(ls, rc);
565                 break;
566
567         default:
568                 log_error(ls, "receive_rcom bad type %d", rc->rc_type);
569         }
570 out:
571         return;
572 Eshort:
573         log_error(ls, "recovery message %x from %d is too short",
574                           rc->rc_type, nodeid);
575 }
576