1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * dlmconvert.c
4 *
5 * underlying calls for lock conversion
6 *
7 * Copyright (C) 2004 Oracle. All rights reserved.
8 */
9
10
11 #include <linux/module.h>
12 #include <linux/fs.h>
13 #include <linux/types.h>
14 #include <linux/highmem.h>
15 #include <linux/init.h>
16 #include <linux/sysctl.h>
17 #include <linux/random.h>
18 #include <linux/blkdev.h>
19 #include <linux/socket.h>
20 #include <linux/inet.h>
21 #include <linux/spinlock.h>
22
23
24 #include "../cluster/heartbeat.h"
25 #include "../cluster/nodemanager.h"
26 #include "../cluster/tcp.h"
27
28 #include "dlmapi.h"
29 #include "dlmcommon.h"
30
31 #include "dlmconvert.h"
32
33 #define MLOG_MASK_PREFIX ML_DLM
34 #include "../cluster/masklog.h"
35
36 /* NOTE: __dlmconvert_master is the only function in here that
37 * needs a spinlock held on entry (res->spinlock) and it is the
38 * only one that holds a lock on exit (res->spinlock).
39 * All other functions in here need no locks and drop all of
40 * the locks that they acquire. */
41 static enum dlm_status __dlmconvert_master(struct dlm_ctxt *dlm,
42 struct dlm_lock_resource *res,
43 struct dlm_lock *lock, int flags,
44 int type, int *call_ast,
45 int *kick_thread);
46 static enum dlm_status dlm_send_remote_convert_request(struct dlm_ctxt *dlm,
47 struct dlm_lock_resource *res,
48 struct dlm_lock *lock, int flags, int type);
49
50 /*
51 * this is only called directly by dlmlock(), and only when the
52 * local node is the owner of the lockres
53 * locking:
54 * caller needs: none
55 * taken: takes and drops res->spinlock
56 * held on exit: none
57 * returns: see __dlmconvert_master
58 */
dlmconvert_master(struct dlm_ctxt * dlm,struct dlm_lock_resource * res,struct dlm_lock * lock,int flags,int type)59 enum dlm_status dlmconvert_master(struct dlm_ctxt *dlm,
60 struct dlm_lock_resource *res,
61 struct dlm_lock *lock, int flags, int type)
62 {
63 int call_ast = 0, kick_thread = 0;
64 enum dlm_status status;
65
66 spin_lock(&res->spinlock);
67 /* we are not in a network handler, this is fine */
68 __dlm_wait_on_lockres(res);
69 __dlm_lockres_reserve_ast(res);
70 res->state |= DLM_LOCK_RES_IN_PROGRESS;
71
72 status = __dlmconvert_master(dlm, res, lock, flags, type,
73 &call_ast, &kick_thread);
74
75 res->state &= ~DLM_LOCK_RES_IN_PROGRESS;
76 spin_unlock(&res->spinlock);
77 wake_up(&res->wq);
78 if (status != DLM_NORMAL && status != DLM_NOTQUEUED)
79 dlm_error(status);
80
81 /* either queue the ast or release it */
82 if (call_ast)
83 dlm_queue_ast(dlm, lock);
84 else
85 dlm_lockres_release_ast(dlm, res);
86
87 if (kick_thread)
88 dlm_kick_thread(dlm, res);
89
90 return status;
91 }
92
93 /* performs lock conversion at the lockres master site
94 * locking:
95 * caller needs: res->spinlock
96 * taken: takes and drops lock->spinlock
97 * held on exit: res->spinlock
98 * returns: DLM_NORMAL, DLM_NOTQUEUED, DLM_DENIED
99 * call_ast: whether ast should be called for this lock
100 * kick_thread: whether dlm_kick_thread should be called
101 */
__dlmconvert_master(struct dlm_ctxt * dlm,struct dlm_lock_resource * res,struct dlm_lock * lock,int flags,int type,int * call_ast,int * kick_thread)102 static enum dlm_status __dlmconvert_master(struct dlm_ctxt *dlm,
103 struct dlm_lock_resource *res,
104 struct dlm_lock *lock, int flags,
105 int type, int *call_ast,
106 int *kick_thread)
107 {
108 enum dlm_status status = DLM_NORMAL;
109 struct dlm_lock *tmplock=NULL;
110
111 assert_spin_locked(&res->spinlock);
112
113 mlog(0, "type=%d, convert_type=%d, new convert_type=%d\n",
114 lock->ml.type, lock->ml.convert_type, type);
115
116 spin_lock(&lock->spinlock);
117
118 /* already converting? */
119 if (lock->ml.convert_type != LKM_IVMODE) {
120 mlog(ML_ERROR, "attempted to convert a lock with a lock "
121 "conversion pending\n");
122 status = DLM_DENIED;
123 goto unlock_exit;
124 }
125
126 /* must be on grant queue to convert */
127 if (!dlm_lock_on_list(&res->granted, lock)) {
128 mlog(ML_ERROR, "attempted to convert a lock not on grant "
129 "queue\n");
130 status = DLM_DENIED;
131 goto unlock_exit;
132 }
133
134 if (flags & LKM_VALBLK) {
135 switch (lock->ml.type) {
136 case LKM_EXMODE:
137 /* EX + LKM_VALBLK + convert == set lvb */
138 mlog(0, "will set lvb: converting %s->%s\n",
139 dlm_lock_mode_name(lock->ml.type),
140 dlm_lock_mode_name(type));
141 lock->lksb->flags |= DLM_LKSB_PUT_LVB;
142 break;
143 case LKM_PRMODE:
144 case LKM_NLMODE:
145 /* refetch if new level is not NL */
146 if (type > LKM_NLMODE) {
147 mlog(0, "will fetch new value into "
148 "lvb: converting %s->%s\n",
149 dlm_lock_mode_name(lock->ml.type),
150 dlm_lock_mode_name(type));
151 lock->lksb->flags |= DLM_LKSB_GET_LVB;
152 } else {
153 mlog(0, "will NOT fetch new value "
154 "into lvb: converting %s->%s\n",
155 dlm_lock_mode_name(lock->ml.type),
156 dlm_lock_mode_name(type));
157 flags &= ~(LKM_VALBLK);
158 }
159 break;
160 }
161 }
162
163
164 /* in-place downconvert? */
165 if (type <= lock->ml.type)
166 goto grant;
167
168 /* upconvert from here on */
169 status = DLM_NORMAL;
170 list_for_each_entry(tmplock, &res->granted, list) {
171 if (tmplock == lock)
172 continue;
173 if (!dlm_lock_compatible(tmplock->ml.type, type))
174 goto switch_queues;
175 }
176
177 list_for_each_entry(tmplock, &res->converting, list) {
178 if (!dlm_lock_compatible(tmplock->ml.type, type))
179 goto switch_queues;
180 /* existing conversion requests take precedence */
181 if (!dlm_lock_compatible(tmplock->ml.convert_type, type))
182 goto switch_queues;
183 }
184
185 /* fall thru to grant */
186
187 grant:
188 mlog(0, "res %.*s, granting %s lock\n", res->lockname.len,
189 res->lockname.name, dlm_lock_mode_name(type));
190 /* immediately grant the new lock type */
191 lock->lksb->status = DLM_NORMAL;
192 if (lock->ml.node == dlm->node_num)
193 mlog(0, "doing in-place convert for nonlocal lock\n");
194 lock->ml.type = type;
195 if (lock->lksb->flags & DLM_LKSB_PUT_LVB)
196 memcpy(res->lvb, lock->lksb->lvb, DLM_LVB_LEN);
197
198 /*
199 * Move the lock to the tail because it may be the only lock which has
200 * an invalid lvb.
201 */
202 list_move_tail(&lock->list, &res->granted);
203
204 status = DLM_NORMAL;
205 *call_ast = 1;
206 goto unlock_exit;
207
208 switch_queues:
209 if (flags & LKM_NOQUEUE) {
210 mlog(0, "failed to convert NOQUEUE lock %.*s from "
211 "%d to %d...\n", res->lockname.len, res->lockname.name,
212 lock->ml.type, type);
213 status = DLM_NOTQUEUED;
214 goto unlock_exit;
215 }
216 mlog(0, "res %.*s, queueing...\n", res->lockname.len,
217 res->lockname.name);
218
219 lock->ml.convert_type = type;
220 /* do not alter lock refcount. switching lists. */
221 list_move_tail(&lock->list, &res->converting);
222
223 unlock_exit:
224 spin_unlock(&lock->spinlock);
225 if (status == DLM_DENIED) {
226 __dlm_print_one_lock_resource(res);
227 }
228 if (status == DLM_NORMAL)
229 *kick_thread = 1;
230 return status;
231 }
232
dlm_revert_pending_convert(struct dlm_lock_resource * res,struct dlm_lock * lock)233 void dlm_revert_pending_convert(struct dlm_lock_resource *res,
234 struct dlm_lock *lock)
235 {
236 /* do not alter lock refcount. switching lists. */
237 list_move_tail(&lock->list, &res->granted);
238 lock->ml.convert_type = LKM_IVMODE;
239 lock->lksb->flags &= ~(DLM_LKSB_GET_LVB|DLM_LKSB_PUT_LVB);
240 }
241
242 /* messages the master site to do lock conversion
243 * locking:
244 * caller needs: none
245 * taken: takes and drops res->spinlock, uses DLM_LOCK_RES_IN_PROGRESS
246 * held on exit: none
247 * returns: DLM_NORMAL, DLM_RECOVERING, status from remote node
248 */
dlmconvert_remote(struct dlm_ctxt * dlm,struct dlm_lock_resource * res,struct dlm_lock * lock,int flags,int type)249 enum dlm_status dlmconvert_remote(struct dlm_ctxt *dlm,
250 struct dlm_lock_resource *res,
251 struct dlm_lock *lock, int flags, int type)
252 {
253 enum dlm_status status;
254
255 mlog(0, "type=%d, convert_type=%d, busy=%d\n", lock->ml.type,
256 lock->ml.convert_type, res->state & DLM_LOCK_RES_IN_PROGRESS);
257
258 spin_lock(&res->spinlock);
259 if (res->state & DLM_LOCK_RES_RECOVERING) {
260 mlog(0, "bailing out early since res is RECOVERING "
261 "on secondary queue\n");
262 /* __dlm_print_one_lock_resource(res); */
263 status = DLM_RECOVERING;
264 goto bail;
265 }
266 /* will exit this call with spinlock held */
267 __dlm_wait_on_lockres(res);
268
269 if (lock->ml.convert_type != LKM_IVMODE) {
270 __dlm_print_one_lock_resource(res);
271 mlog(ML_ERROR, "converting a remote lock that is already "
272 "converting! (cookie=%u:%llu, conv=%d)\n",
273 dlm_get_lock_cookie_node(be64_to_cpu(lock->ml.cookie)),
274 dlm_get_lock_cookie_seq(be64_to_cpu(lock->ml.cookie)),
275 lock->ml.convert_type);
276 status = DLM_DENIED;
277 goto bail;
278 }
279
280 if (lock->ml.type == type && lock->ml.convert_type == LKM_IVMODE) {
281 mlog(0, "last convert request returned DLM_RECOVERING, but "
282 "owner has already queued and sent ast to me. res %.*s, "
283 "(cookie=%u:%llu, type=%d, conv=%d)\n",
284 res->lockname.len, res->lockname.name,
285 dlm_get_lock_cookie_node(be64_to_cpu(lock->ml.cookie)),
286 dlm_get_lock_cookie_seq(be64_to_cpu(lock->ml.cookie)),
287 lock->ml.type, lock->ml.convert_type);
288 status = DLM_NORMAL;
289 goto bail;
290 }
291
292 res->state |= DLM_LOCK_RES_IN_PROGRESS;
293 /* move lock to local convert queue */
294 /* do not alter lock refcount. switching lists. */
295 list_move_tail(&lock->list, &res->converting);
296 lock->convert_pending = 1;
297 lock->ml.convert_type = type;
298
299 if (flags & LKM_VALBLK) {
300 if (lock->ml.type == LKM_EXMODE) {
301 flags |= LKM_PUT_LVB;
302 lock->lksb->flags |= DLM_LKSB_PUT_LVB;
303 } else {
304 if (lock->ml.convert_type == LKM_NLMODE)
305 flags &= ~LKM_VALBLK;
306 else {
307 flags |= LKM_GET_LVB;
308 lock->lksb->flags |= DLM_LKSB_GET_LVB;
309 }
310 }
311 }
312 spin_unlock(&res->spinlock);
313
314 /* no locks held here.
315 * need to wait for a reply as to whether it got queued or not. */
316 status = dlm_send_remote_convert_request(dlm, res, lock, flags, type);
317
318 spin_lock(&res->spinlock);
319 res->state &= ~DLM_LOCK_RES_IN_PROGRESS;
320 /* if it failed, move it back to granted queue.
321 * if master returns DLM_NORMAL and then down before sending ast,
322 * it may have already been moved to granted queue, reset to
323 * DLM_RECOVERING and retry convert */
324 if (status != DLM_NORMAL) {
325 if (status != DLM_NOTQUEUED)
326 dlm_error(status);
327 dlm_revert_pending_convert(res, lock);
328 } else if (!lock->convert_pending) {
329 mlog(0, "%s: res %.*s, owner died and lock has been moved back "
330 "to granted list, retry convert.\n",
331 dlm->name, res->lockname.len, res->lockname.name);
332 status = DLM_RECOVERING;
333 }
334
335 lock->convert_pending = 0;
336 bail:
337 spin_unlock(&res->spinlock);
338
339 /* TODO: should this be a wake_one? */
340 /* wake up any IN_PROGRESS waiters */
341 wake_up(&res->wq);
342
343 return status;
344 }
345
346 /* sends DLM_CONVERT_LOCK_MSG to master site
347 * locking:
348 * caller needs: none
349 * taken: none
350 * held on exit: none
351 * returns: DLM_NOLOCKMGR, status from remote node
352 */
dlm_send_remote_convert_request(struct dlm_ctxt * dlm,struct dlm_lock_resource * res,struct dlm_lock * lock,int flags,int type)353 static enum dlm_status dlm_send_remote_convert_request(struct dlm_ctxt *dlm,
354 struct dlm_lock_resource *res,
355 struct dlm_lock *lock, int flags, int type)
356 {
357 struct dlm_convert_lock convert;
358 int tmpret;
359 enum dlm_status ret;
360 int status = 0;
361 struct kvec vec[2];
362 size_t veclen = 1;
363
364 mlog(0, "%.*s\n", res->lockname.len, res->lockname.name);
365
366 memset(&convert, 0, sizeof(struct dlm_convert_lock));
367 convert.node_idx = dlm->node_num;
368 convert.requested_type = type;
369 convert.cookie = lock->ml.cookie;
370 convert.namelen = res->lockname.len;
371 convert.flags = cpu_to_be32(flags);
372 memcpy(convert.name, res->lockname.name, convert.namelen);
373
374 vec[0].iov_len = sizeof(struct dlm_convert_lock);
375 vec[0].iov_base = &convert;
376
377 if (flags & LKM_PUT_LVB) {
378 /* extra data to send if we are updating lvb */
379 vec[1].iov_len = DLM_LVB_LEN;
380 vec[1].iov_base = lock->lksb->lvb;
381 veclen++;
382 }
383
384 tmpret = o2net_send_message_vec(DLM_CONVERT_LOCK_MSG, dlm->key,
385 vec, veclen, res->owner, &status);
386 if (tmpret >= 0) {
387 // successfully sent and received
388 ret = status; // this is already a dlm_status
389 if (ret == DLM_RECOVERING) {
390 mlog(0, "node %u returned DLM_RECOVERING from convert "
391 "message!\n", res->owner);
392 } else if (ret == DLM_MIGRATING) {
393 mlog(0, "node %u returned DLM_MIGRATING from convert "
394 "message!\n", res->owner);
395 } else if (ret == DLM_FORWARD) {
396 mlog(0, "node %u returned DLM_FORWARD from convert "
397 "message!\n", res->owner);
398 } else if (ret != DLM_NORMAL && ret != DLM_NOTQUEUED)
399 dlm_error(ret);
400 } else {
401 mlog(ML_ERROR, "Error %d when sending message %u (key 0x%x) to "
402 "node %u\n", tmpret, DLM_CONVERT_LOCK_MSG, dlm->key,
403 res->owner);
404 if (dlm_is_host_down(tmpret)) {
405 /* instead of logging the same network error over
406 * and over, sleep here and wait for the heartbeat
407 * to notice the node is dead. times out after 5s. */
408 dlm_wait_for_node_death(dlm, res->owner,
409 DLM_NODE_DEATH_WAIT_MAX);
410 ret = DLM_RECOVERING;
411 mlog(0, "node %u died so returning DLM_RECOVERING "
412 "from convert message!\n", res->owner);
413 } else {
414 ret = dlm_err_to_dlm_status(tmpret);
415 }
416 }
417
418 return ret;
419 }
420
421 /* handler for DLM_CONVERT_LOCK_MSG on master site
422 * locking:
423 * caller needs: none
424 * taken: takes and drop res->spinlock
425 * held on exit: none
426 * returns: DLM_NORMAL, DLM_IVLOCKID, DLM_BADARGS,
427 * status from __dlmconvert_master
428 */
dlm_convert_lock_handler(struct o2net_msg * msg,u32 len,void * data,void ** ret_data)429 int dlm_convert_lock_handler(struct o2net_msg *msg, u32 len, void *data,
430 void **ret_data)
431 {
432 struct dlm_ctxt *dlm = data;
433 struct dlm_convert_lock *cnv = (struct dlm_convert_lock *)msg->buf;
434 struct dlm_lock_resource *res = NULL;
435 struct dlm_lock *lock = NULL;
436 struct dlm_lock *tmp_lock;
437 struct dlm_lockstatus *lksb;
438 enum dlm_status status = DLM_NORMAL;
439 u32 flags;
440 int call_ast = 0, kick_thread = 0, ast_reserved = 0, wake = 0;
441
442 if (!dlm_grab(dlm)) {
443 dlm_error(DLM_REJECTED);
444 return DLM_REJECTED;
445 }
446
447 mlog_bug_on_msg(!dlm_domain_fully_joined(dlm),
448 "Domain %s not fully joined!\n", dlm->name);
449
450 if (cnv->namelen > DLM_LOCKID_NAME_MAX) {
451 status = DLM_IVBUFLEN;
452 dlm_error(status);
453 goto leave;
454 }
455
456 flags = be32_to_cpu(cnv->flags);
457
458 if ((flags & (LKM_PUT_LVB|LKM_GET_LVB)) ==
459 (LKM_PUT_LVB|LKM_GET_LVB)) {
460 mlog(ML_ERROR, "both PUT and GET lvb specified\n");
461 status = DLM_BADARGS;
462 goto leave;
463 }
464
465 mlog(0, "lvb: %s\n", flags & LKM_PUT_LVB ? "put lvb" :
466 (flags & LKM_GET_LVB ? "get lvb" : "none"));
467
468 status = DLM_IVLOCKID;
469 res = dlm_lookup_lockres(dlm, cnv->name, cnv->namelen);
470 if (!res) {
471 dlm_error(status);
472 goto leave;
473 }
474
475 spin_lock(&res->spinlock);
476 status = __dlm_lockres_state_to_status(res);
477 if (status != DLM_NORMAL) {
478 spin_unlock(&res->spinlock);
479 dlm_error(status);
480 goto leave;
481 }
482 list_for_each_entry(tmp_lock, &res->granted, list) {
483 if (tmp_lock->ml.cookie == cnv->cookie &&
484 tmp_lock->ml.node == cnv->node_idx) {
485 lock = tmp_lock;
486 dlm_lock_get(lock);
487 break;
488 }
489 }
490 spin_unlock(&res->spinlock);
491 if (!lock) {
492 status = DLM_IVLOCKID;
493 mlog(ML_ERROR, "did not find lock to convert on grant queue! "
494 "cookie=%u:%llu\n",
495 dlm_get_lock_cookie_node(be64_to_cpu(cnv->cookie)),
496 dlm_get_lock_cookie_seq(be64_to_cpu(cnv->cookie)));
497 dlm_print_one_lock_resource(res);
498 goto leave;
499 }
500
501 /* found the lock */
502 lksb = lock->lksb;
503
504 /* see if caller needed to get/put lvb */
505 if (flags & LKM_PUT_LVB) {
506 BUG_ON(lksb->flags & (DLM_LKSB_PUT_LVB|DLM_LKSB_GET_LVB));
507 lksb->flags |= DLM_LKSB_PUT_LVB;
508 memcpy(&lksb->lvb[0], &cnv->lvb[0], DLM_LVB_LEN);
509 } else if (flags & LKM_GET_LVB) {
510 BUG_ON(lksb->flags & (DLM_LKSB_PUT_LVB|DLM_LKSB_GET_LVB));
511 lksb->flags |= DLM_LKSB_GET_LVB;
512 }
513
514 spin_lock(&res->spinlock);
515 status = __dlm_lockres_state_to_status(res);
516 if (status == DLM_NORMAL) {
517 __dlm_lockres_reserve_ast(res);
518 ast_reserved = 1;
519 res->state |= DLM_LOCK_RES_IN_PROGRESS;
520 status = __dlmconvert_master(dlm, res, lock, flags,
521 cnv->requested_type,
522 &call_ast, &kick_thread);
523 res->state &= ~DLM_LOCK_RES_IN_PROGRESS;
524 wake = 1;
525 }
526 spin_unlock(&res->spinlock);
527 if (wake)
528 wake_up(&res->wq);
529
530 if (status != DLM_NORMAL) {
531 if (status != DLM_NOTQUEUED)
532 dlm_error(status);
533 lksb->flags &= ~(DLM_LKSB_GET_LVB|DLM_LKSB_PUT_LVB);
534 }
535
536 leave:
537 if (lock)
538 dlm_lock_put(lock);
539
540 /* either queue the ast or release it, if reserved */
541 if (call_ast)
542 dlm_queue_ast(dlm, lock);
543 else if (ast_reserved)
544 dlm_lockres_release_ast(dlm, res);
545
546 if (kick_thread)
547 dlm_kick_thread(dlm, res);
548
549 if (res)
550 dlm_lockres_put(res);
551
552 dlm_put(dlm);
553
554 return status;
555 }
556