aboutsummaryrefslogtreecommitdiffstats
path: root/fs/ocfs2/dlm/dlmconvert.c
diff options
context:
space:
mode:
authorKurt Hackel <kurt.hackel@oracle.com>2005-12-15 17:31:23 -0500
committerJoel Becker <joel.becker@oracle.com>2006-01-03 14:45:47 -0500
commit6714d8e86bf443f6f7af50f9d432025649f091f5 (patch)
tree2c484bd1894a90cad7020869c7054f192d3bf34d /fs/ocfs2/dlm/dlmconvert.c
parent98211489d4147e41b11703e4245846d60b3acce4 (diff)
[PATCH] OCFS2: The Second Oracle Cluster Filesystem
A distributed lock manager built with the cluster file system use case in mind. The OCFS2 dlm exposes a VMS style API, though things have been simplified internally. The only lock levels implemented currently are NLMODE, PRMODE and EXMODE. Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com> Signed-off-by: Kurt Hackel <kurt.hackel@oracle.com>
Diffstat (limited to 'fs/ocfs2/dlm/dlmconvert.c')
-rw-r--r--fs/ocfs2/dlm/dlmconvert.c530
1 files changed, 530 insertions, 0 deletions
diff --git a/fs/ocfs2/dlm/dlmconvert.c b/fs/ocfs2/dlm/dlmconvert.c
new file mode 100644
index 000000000000..6001b22a997d
--- /dev/null
+++ b/fs/ocfs2/dlm/dlmconvert.c
@@ -0,0 +1,530 @@
1/* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
3 *
4 * dlmconvert.c
5 *
6 * underlying calls for lock conversion
7 *
8 * Copyright (C) 2004 Oracle. All rights reserved.
9 *
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public
12 * License as published by the Free Software Foundation; either
13 * version 2 of the License, or (at your option) any later version.
14 *
15 * This program is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * General Public License for more details.
19 *
20 * You should have received a copy of the GNU General Public
21 * License along with this program; if not, write to the
22 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
23 * Boston, MA 021110-1307, USA.
24 *
25 */
26
27
28#include <linux/module.h>
29#include <linux/fs.h>
30#include <linux/types.h>
31#include <linux/slab.h>
32#include <linux/highmem.h>
33#include <linux/utsname.h>
34#include <linux/init.h>
35#include <linux/sysctl.h>
36#include <linux/random.h>
37#include <linux/blkdev.h>
38#include <linux/socket.h>
39#include <linux/inet.h>
40#include <linux/spinlock.h>
41
42
43#include "cluster/heartbeat.h"
44#include "cluster/nodemanager.h"
45#include "cluster/tcp.h"
46
47#include "dlmapi.h"
48#include "dlmcommon.h"
49
50#include "dlmconvert.h"
51
52#define MLOG_MASK_PREFIX ML_DLM
53#include "cluster/masklog.h"
54
55/* NOTE: __dlmconvert_master is the only function in here that
56 * needs a spinlock held on entry (res->spinlock) and it is the
57 * only one that holds a lock on exit (res->spinlock).
58 * All other functions in here need no locks and drop all of
59 * the locks that they acquire. */
60static enum dlm_status __dlmconvert_master(struct dlm_ctxt *dlm,
61 struct dlm_lock_resource *res,
62 struct dlm_lock *lock, int flags,
63 int type, int *call_ast,
64 int *kick_thread);
65static enum dlm_status dlm_send_remote_convert_request(struct dlm_ctxt *dlm,
66 struct dlm_lock_resource *res,
67 struct dlm_lock *lock, int flags, int type);
68
69/*
70 * this is only called directly by dlmlock(), and only when the
71 * local node is the owner of the lockres
72 * locking:
73 * caller needs: none
74 * taken: takes and drops res->spinlock
75 * held on exit: none
76 * returns: see __dlmconvert_master
77 */
78enum dlm_status dlmconvert_master(struct dlm_ctxt *dlm,
79 struct dlm_lock_resource *res,
80 struct dlm_lock *lock, int flags, int type)
81{
82 int call_ast = 0, kick_thread = 0;
83 enum dlm_status status;
84
85 spin_lock(&res->spinlock);
86 /* we are not in a network handler, this is fine */
87 __dlm_wait_on_lockres(res);
88 __dlm_lockres_reserve_ast(res);
89 res->state |= DLM_LOCK_RES_IN_PROGRESS;
90
91 status = __dlmconvert_master(dlm, res, lock, flags, type,
92 &call_ast, &kick_thread);
93
94 res->state &= ~DLM_LOCK_RES_IN_PROGRESS;
95 spin_unlock(&res->spinlock);
96 wake_up(&res->wq);
97 if (status != DLM_NORMAL && status != DLM_NOTQUEUED)
98 dlm_error(status);
99
100 /* either queue the ast or release it */
101 if (call_ast)
102 dlm_queue_ast(dlm, lock);
103 else
104 dlm_lockres_release_ast(dlm, res);
105
106 if (kick_thread)
107 dlm_kick_thread(dlm, res);
108
109 return status;
110}
111
112/* performs lock conversion at the lockres master site
113 * locking:
114 * caller needs: res->spinlock
115 * taken: takes and drops lock->spinlock
116 * held on exit: res->spinlock
117 * returns: DLM_NORMAL, DLM_NOTQUEUED, DLM_DENIED
118 * call_ast: whether ast should be called for this lock
119 * kick_thread: whether dlm_kick_thread should be called
120 */
121static enum dlm_status __dlmconvert_master(struct dlm_ctxt *dlm,
122 struct dlm_lock_resource *res,
123 struct dlm_lock *lock, int flags,
124 int type, int *call_ast,
125 int *kick_thread)
126{
127 enum dlm_status status = DLM_NORMAL;
128 struct list_head *iter;
129 struct dlm_lock *tmplock=NULL;
130
131 assert_spin_locked(&res->spinlock);
132
133 mlog_entry("type=%d, convert_type=%d, new convert_type=%d\n",
134 lock->ml.type, lock->ml.convert_type, type);
135
136 spin_lock(&lock->spinlock);
137
138 /* already converting? */
139 if (lock->ml.convert_type != LKM_IVMODE) {
140 mlog(ML_ERROR, "attempted to convert a lock with a lock "
141 "conversion pending\n");
142 status = DLM_DENIED;
143 goto unlock_exit;
144 }
145
146 /* must be on grant queue to convert */
147 if (!dlm_lock_on_list(&res->granted, lock)) {
148 mlog(ML_ERROR, "attempted to convert a lock not on grant "
149 "queue\n");
150 status = DLM_DENIED;
151 goto unlock_exit;
152 }
153
154 if (flags & LKM_VALBLK) {
155 switch (lock->ml.type) {
156 case LKM_EXMODE:
157 /* EX + LKM_VALBLK + convert == set lvb */
158 mlog(0, "will set lvb: converting %s->%s\n",
159 dlm_lock_mode_name(lock->ml.type),
160 dlm_lock_mode_name(type));
161 lock->lksb->flags |= DLM_LKSB_PUT_LVB;
162 break;
163 case LKM_PRMODE:
164 case LKM_NLMODE:
165 /* refetch if new level is not NL */
166 if (type > LKM_NLMODE) {
167 mlog(0, "will fetch new value into "
168 "lvb: converting %s->%s\n",
169 dlm_lock_mode_name(lock->ml.type),
170 dlm_lock_mode_name(type));
171 lock->lksb->flags |= DLM_LKSB_GET_LVB;
172 } else {
173 mlog(0, "will NOT fetch new value "
174 "into lvb: converting %s->%s\n",
175 dlm_lock_mode_name(lock->ml.type),
176 dlm_lock_mode_name(type));
177 flags &= ~(LKM_VALBLK);
178 }
179 break;
180 }
181 }
182
183
184 /* in-place downconvert? */
185 if (type <= lock->ml.type)
186 goto grant;
187
188 /* upconvert from here on */
189 status = DLM_NORMAL;
190 list_for_each(iter, &res->granted) {
191 tmplock = list_entry(iter, struct dlm_lock, list);
192 if (tmplock == lock)
193 continue;
194 if (!dlm_lock_compatible(tmplock->ml.type, type))
195 goto switch_queues;
196 }
197
198 list_for_each(iter, &res->converting) {
199 tmplock = list_entry(iter, struct dlm_lock, list);
200 if (!dlm_lock_compatible(tmplock->ml.type, type))
201 goto switch_queues;
202 /* existing conversion requests take precedence */
203 if (!dlm_lock_compatible(tmplock->ml.convert_type, type))
204 goto switch_queues;
205 }
206
207 /* fall thru to grant */
208
209grant:
210 mlog(0, "res %.*s, granting %s lock\n", res->lockname.len,
211 res->lockname.name, dlm_lock_mode_name(type));
212 /* immediately grant the new lock type */
213 lock->lksb->status = DLM_NORMAL;
214 if (lock->ml.node == dlm->node_num)
215 mlog(0, "doing in-place convert for nonlocal lock\n");
216 lock->ml.type = type;
217 status = DLM_NORMAL;
218 *call_ast = 1;
219 goto unlock_exit;
220
221switch_queues:
222 if (flags & LKM_NOQUEUE) {
223 mlog(0, "failed to convert NOQUEUE lock %.*s from "
224 "%d to %d...\n", res->lockname.len, res->lockname.name,
225 lock->ml.type, type);
226 status = DLM_NOTQUEUED;
227 goto unlock_exit;
228 }
229 mlog(0, "res %.*s, queueing...\n", res->lockname.len,
230 res->lockname.name);
231
232 lock->ml.convert_type = type;
233 /* do not alter lock refcount. switching lists. */
234 list_del_init(&lock->list);
235 list_add_tail(&lock->list, &res->converting);
236
237unlock_exit:
238 spin_unlock(&lock->spinlock);
239 if (status == DLM_DENIED) {
240 __dlm_print_one_lock_resource(res);
241 }
242 if (status == DLM_NORMAL)
243 *kick_thread = 1;
244 return status;
245}
246
247void dlm_revert_pending_convert(struct dlm_lock_resource *res,
248 struct dlm_lock *lock)
249{
250 /* do not alter lock refcount. switching lists. */
251 list_del_init(&lock->list);
252 list_add_tail(&lock->list, &res->granted);
253 lock->ml.convert_type = LKM_IVMODE;
254 lock->lksb->flags &= ~(DLM_LKSB_GET_LVB|DLM_LKSB_PUT_LVB);
255}
256
257/* messages the master site to do lock conversion
258 * locking:
259 * caller needs: none
260 * taken: takes and drops res->spinlock, uses DLM_LOCK_RES_IN_PROGRESS
261 * held on exit: none
262 * returns: DLM_NORMAL, DLM_RECOVERING, status from remote node
263 */
264enum dlm_status dlmconvert_remote(struct dlm_ctxt *dlm,
265 struct dlm_lock_resource *res,
266 struct dlm_lock *lock, int flags, int type)
267{
268 enum dlm_status status;
269
270 mlog(0, "type=%d, convert_type=%d, busy=%d\n", lock->ml.type,
271 lock->ml.convert_type, res->state & DLM_LOCK_RES_IN_PROGRESS);
272
273 spin_lock(&res->spinlock);
274 if (res->state & DLM_LOCK_RES_RECOVERING) {
275 mlog(0, "bailing out early since res is RECOVERING "
276 "on secondary queue\n");
277 /* __dlm_print_one_lock_resource(res); */
278 status = DLM_RECOVERING;
279 goto bail;
280 }
281 /* will exit this call with spinlock held */
282 __dlm_wait_on_lockres(res);
283
284 if (lock->ml.convert_type != LKM_IVMODE) {
285 __dlm_print_one_lock_resource(res);
286 mlog(ML_ERROR, "converting a remote lock that is already "
287 "converting! (cookie=%"MLFu64", conv=%d)\n",
288 lock->ml.cookie, lock->ml.convert_type);
289 status = DLM_DENIED;
290 goto bail;
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_del_init(&lock->list);
296 list_add_tail(&lock->list, &res->converting);
297 lock->convert_pending = 1;
298 lock->ml.convert_type = type;
299
300 if (flags & LKM_VALBLK) {
301 if (lock->ml.type == LKM_EXMODE) {
302 flags |= LKM_PUT_LVB;
303 lock->lksb->flags |= DLM_LKSB_PUT_LVB;
304 } else {
305 if (lock->ml.convert_type == LKM_NLMODE)
306 flags &= ~LKM_VALBLK;
307 else {
308 flags |= LKM_GET_LVB;
309 lock->lksb->flags |= DLM_LKSB_GET_LVB;
310 }
311 }
312 }
313 spin_unlock(&res->spinlock);
314
315 /* no locks held here.
316 * need to wait for a reply as to whether it got queued or not. */
317 status = dlm_send_remote_convert_request(dlm, res, lock, flags, type);
318
319 spin_lock(&res->spinlock);
320 res->state &= ~DLM_LOCK_RES_IN_PROGRESS;
321 lock->convert_pending = 0;
322 /* if it failed, move it back to granted queue */
323 if (status != DLM_NORMAL) {
324 if (status != DLM_NOTQUEUED)
325 dlm_error(status);
326 dlm_revert_pending_convert(res, lock);
327 }
328bail:
329 spin_unlock(&res->spinlock);
330
331 /* TODO: should this be a wake_one? */
332 /* wake up any IN_PROGRESS waiters */
333 wake_up(&res->wq);
334
335 return status;
336}
337
338/* sends DLM_CONVERT_LOCK_MSG to master site
339 * locking:
340 * caller needs: none
341 * taken: none
342 * held on exit: none
343 * returns: DLM_NOLOCKMGR, status from remote node
344 */
345static enum dlm_status dlm_send_remote_convert_request(struct dlm_ctxt *dlm,
346 struct dlm_lock_resource *res,
347 struct dlm_lock *lock, int flags, int type)
348{
349 struct dlm_convert_lock convert;
350 int tmpret;
351 enum dlm_status ret;
352 int status = 0;
353 struct kvec vec[2];
354 size_t veclen = 1;
355
356 mlog_entry("%.*s\n", res->lockname.len, res->lockname.name);
357
358 memset(&convert, 0, sizeof(struct dlm_convert_lock));
359 convert.node_idx = dlm->node_num;
360 convert.requested_type = type;
361 convert.cookie = lock->ml.cookie;
362 convert.namelen = res->lockname.len;
363 convert.flags = cpu_to_be32(flags);
364 memcpy(convert.name, res->lockname.name, convert.namelen);
365
366 vec[0].iov_len = sizeof(struct dlm_convert_lock);
367 vec[0].iov_base = &convert;
368
369 if (flags & LKM_PUT_LVB) {
370 /* extra data to send if we are updating lvb */
371 vec[1].iov_len = DLM_LVB_LEN;
372 vec[1].iov_base = lock->lksb->lvb;
373 veclen++;
374 }
375
376 tmpret = o2net_send_message_vec(DLM_CONVERT_LOCK_MSG, dlm->key,
377 vec, veclen, res->owner, &status);
378 if (tmpret >= 0) {
379 // successfully sent and received
380 ret = status; // this is already a dlm_status
381 if (ret == DLM_RECOVERING) {
382 mlog(0, "node %u returned DLM_RECOVERING from convert "
383 "message!\n", res->owner);
384 } else if (ret == DLM_MIGRATING) {
385 mlog(0, "node %u returned DLM_MIGRATING from convert "
386 "message!\n", res->owner);
387 } else if (ret == DLM_FORWARD) {
388 mlog(0, "node %u returned DLM_FORWARD from convert "
389 "message!\n", res->owner);
390 } else if (ret != DLM_NORMAL && ret != DLM_NOTQUEUED)
391 dlm_error(ret);
392 } else {
393 mlog_errno(tmpret);
394 if (dlm_is_host_down(tmpret)) {
395 ret = DLM_RECOVERING;
396 mlog(0, "node %u died so returning DLM_RECOVERING "
397 "from convert message!\n", res->owner);
398 } else {
399 ret = dlm_err_to_dlm_status(tmpret);
400 }
401 }
402
403 return ret;
404}
405
406/* handler for DLM_CONVERT_LOCK_MSG on master site
407 * locking:
408 * caller needs: none
409 * taken: takes and drop res->spinlock
410 * held on exit: none
411 * returns: DLM_NORMAL, DLM_IVLOCKID, DLM_BADARGS,
412 * status from __dlmconvert_master
413 */
414int dlm_convert_lock_handler(struct o2net_msg *msg, u32 len, void *data)
415{
416 struct dlm_ctxt *dlm = data;
417 struct dlm_convert_lock *cnv = (struct dlm_convert_lock *)msg->buf;
418 struct dlm_lock_resource *res = NULL;
419 struct list_head *iter;
420 struct dlm_lock *lock = NULL;
421 struct dlm_lockstatus *lksb;
422 enum dlm_status status = DLM_NORMAL;
423 u32 flags;
424 int call_ast = 0, kick_thread = 0;
425
426 if (!dlm_grab(dlm)) {
427 dlm_error(DLM_REJECTED);
428 return DLM_REJECTED;
429 }
430
431 mlog_bug_on_msg(!dlm_domain_fully_joined(dlm),
432 "Domain %s not fully joined!\n", dlm->name);
433
434 if (cnv->namelen > DLM_LOCKID_NAME_MAX) {
435 status = DLM_IVBUFLEN;
436 dlm_error(status);
437 goto leave;
438 }
439
440 flags = be32_to_cpu(cnv->flags);
441
442 if ((flags & (LKM_PUT_LVB|LKM_GET_LVB)) ==
443 (LKM_PUT_LVB|LKM_GET_LVB)) {
444 mlog(ML_ERROR, "both PUT and GET lvb specified\n");
445 status = DLM_BADARGS;
446 goto leave;
447 }
448
449 mlog(0, "lvb: %s\n", flags & LKM_PUT_LVB ? "put lvb" :
450 (flags & LKM_GET_LVB ? "get lvb" : "none"));
451
452 status = DLM_IVLOCKID;
453 res = dlm_lookup_lockres(dlm, cnv->name, cnv->namelen);
454 if (!res) {
455 dlm_error(status);
456 goto leave;
457 }
458
459 spin_lock(&res->spinlock);
460 list_for_each(iter, &res->granted) {
461 lock = list_entry(iter, struct dlm_lock, list);
462 if (lock->ml.cookie == cnv->cookie &&
463 lock->ml.node == cnv->node_idx) {
464 dlm_lock_get(lock);
465 break;
466 }
467 lock = NULL;
468 }
469 spin_unlock(&res->spinlock);
470 if (!lock) {
471 status = DLM_IVLOCKID;
472 dlm_error(status);
473 goto leave;
474 }
475
476 /* found the lock */
477 lksb = lock->lksb;
478
479 /* see if caller needed to get/put lvb */
480 if (flags & LKM_PUT_LVB) {
481 BUG_ON(lksb->flags & (DLM_LKSB_PUT_LVB|DLM_LKSB_GET_LVB));
482 lksb->flags |= DLM_LKSB_PUT_LVB;
483 memcpy(&lksb->lvb[0], &cnv->lvb[0], DLM_LVB_LEN);
484 } else if (flags & LKM_GET_LVB) {
485 BUG_ON(lksb->flags & (DLM_LKSB_PUT_LVB|DLM_LKSB_GET_LVB));
486 lksb->flags |= DLM_LKSB_GET_LVB;
487 }
488
489 spin_lock(&res->spinlock);
490 status = __dlm_lockres_state_to_status(res);
491 if (status == DLM_NORMAL) {
492 __dlm_lockres_reserve_ast(res);
493 res->state |= DLM_LOCK_RES_IN_PROGRESS;
494 status = __dlmconvert_master(dlm, res, lock, flags,
495 cnv->requested_type,
496 &call_ast, &kick_thread);
497 res->state &= ~DLM_LOCK_RES_IN_PROGRESS;
498 }
499 spin_unlock(&res->spinlock);
500
501 if (status != DLM_NORMAL) {
502 if (status != DLM_NOTQUEUED)
503 dlm_error(status);
504 lksb->flags &= ~(DLM_LKSB_GET_LVB|DLM_LKSB_PUT_LVB);
505 }
506
507leave:
508 if (!lock)
509 mlog(ML_ERROR, "did not find lock to convert on grant queue! "
510 "cookie=%"MLFu64"\n",
511 cnv->cookie);
512 else
513 dlm_lock_put(lock);
514
515 /* either queue the ast or release it */
516 if (call_ast)
517 dlm_queue_ast(dlm, lock);
518 else
519 dlm_lockres_release_ast(dlm, res);
520
521 if (kick_thread)
522 dlm_kick_thread(dlm, res);
523
524 if (res)
525 dlm_lockres_put(res);
526
527 dlm_put(dlm);
528
529 return status;
530}