1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
6 * Code which implements the kernel side of a minimal userspace
7 * interface to our DLM.
9 * Many of the functions here are pared down versions of dlmglue.c
12 * Copyright (C) 2003, 2004 Oracle. All rights reserved.
14 * This program is free software; you can redistribute it and/or
15 * modify it under the terms of the GNU General Public
16 * License as published by the Free Software Foundation; either
17 * version 2 of the License, or (at your option) any later version.
19 * This program is distributed in the hope that it will be useful,
20 * but WITHOUT ANY WARRANTY; without even the implied warranty of
21 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
22 * General Public License for more details.
24 * You should have received a copy of the GNU General Public
25 * License along with this program; if not, write to the
26 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
27 * Boston, MA 021110-1307, USA.
30 #include <linux/signal.h>
32 #include <linux/module.h>
34 #include <linux/types.h>
35 #include <linux/crc32.h>
38 #include "cluster/nodemanager.h"
39 #include "cluster/heartbeat.h"
40 #include "cluster/tcp.h"
46 #define MLOG_MASK_PREFIX ML_DLMFS
47 #include "cluster/masklog.h"
49 static inline int user_check_wait_flag(struct user_lock_res
*lockres
,
54 spin_lock(&lockres
->l_lock
);
55 ret
= lockres
->l_flags
& flag
;
56 spin_unlock(&lockres
->l_lock
);
61 static inline void user_wait_on_busy_lock(struct user_lock_res
*lockres
)
64 wait_event(lockres
->l_event
,
65 !user_check_wait_flag(lockres
, USER_LOCK_BUSY
));
68 static inline void user_wait_on_blocked_lock(struct user_lock_res
*lockres
)
71 wait_event(lockres
->l_event
,
72 !user_check_wait_flag(lockres
, USER_LOCK_BLOCKED
));
75 /* I heart container_of... */
76 static inline struct dlm_ctxt
*
77 dlm_ctxt_from_user_lockres(struct user_lock_res
*lockres
)
79 struct dlmfs_inode_private
*ip
;
81 ip
= container_of(lockres
,
82 struct dlmfs_inode_private
,
88 user_dlm_inode_from_user_lockres(struct user_lock_res
*lockres
)
90 struct dlmfs_inode_private
*ip
;
92 ip
= container_of(lockres
,
93 struct dlmfs_inode_private
,
95 return &ip
->ip_vfs_inode
;
98 static inline void user_recover_from_dlm_error(struct user_lock_res
*lockres
)
100 spin_lock(&lockres
->l_lock
);
101 lockres
->l_flags
&= ~USER_LOCK_BUSY
;
102 spin_unlock(&lockres
->l_lock
);
105 #define user_log_dlm_error(_func, _stat, _lockres) do { \
106 mlog(ML_ERROR, "Dlm error \"%s\" while calling %s on " \
107 "resource %s: %s\n", dlm_errname(_stat), _func, \
108 _lockres->l_name, dlm_errmsg(_stat)); \
111 /* WARNING: This function lives in a world where the only three lock
112 * levels are EX, PR, and NL. It *will* have to be adjusted when more
113 * lock types are added. */
114 static inline int user_highest_compat_lock_level(int level
)
116 int new_level
= LKM_EXMODE
;
118 if (level
== LKM_EXMODE
)
119 new_level
= LKM_NLMODE
;
120 else if (level
== LKM_PRMODE
)
121 new_level
= LKM_PRMODE
;
125 static void user_ast(void *opaque
)
127 struct user_lock_res
*lockres
= opaque
;
128 struct dlm_lockstatus
*lksb
;
130 mlog(0, "AST fired for lockres %s\n", lockres
->l_name
);
132 spin_lock(&lockres
->l_lock
);
134 lksb
= &(lockres
->l_lksb
);
135 if (lksb
->status
!= DLM_NORMAL
) {
136 mlog(ML_ERROR
, "lksb status value of %u on lockres %s\n",
137 lksb
->status
, lockres
->l_name
);
138 spin_unlock(&lockres
->l_lock
);
142 /* we're downconverting. */
143 if (lockres
->l_requested
< lockres
->l_level
) {
144 if (lockres
->l_requested
<=
145 user_highest_compat_lock_level(lockres
->l_blocking
)) {
146 lockres
->l_blocking
= LKM_NLMODE
;
147 lockres
->l_flags
&= ~USER_LOCK_BLOCKED
;
151 lockres
->l_level
= lockres
->l_requested
;
152 lockres
->l_requested
= LKM_IVMODE
;
153 lockres
->l_flags
|= USER_LOCK_ATTACHED
;
154 lockres
->l_flags
&= ~USER_LOCK_BUSY
;
156 spin_unlock(&lockres
->l_lock
);
158 wake_up(&lockres
->l_event
);
161 static inline void user_dlm_grab_inode_ref(struct user_lock_res
*lockres
)
164 inode
= user_dlm_inode_from_user_lockres(lockres
);
169 static void user_dlm_unblock_lock(void *opaque
);
171 static void __user_dlm_queue_lockres(struct user_lock_res
*lockres
)
173 if (!(lockres
->l_flags
& USER_LOCK_QUEUED
)) {
174 user_dlm_grab_inode_ref(lockres
);
176 INIT_WORK(&lockres
->l_work
, user_dlm_unblock_lock
,
179 queue_work(user_dlm_worker
, &lockres
->l_work
);
180 lockres
->l_flags
|= USER_LOCK_QUEUED
;
184 static void __user_dlm_cond_queue_lockres(struct user_lock_res
*lockres
)
188 if (!(lockres
->l_flags
& USER_LOCK_BLOCKED
))
191 switch (lockres
->l_blocking
) {
193 if (!lockres
->l_ex_holders
&& !lockres
->l_ro_holders
)
197 if (!lockres
->l_ex_holders
)
205 __user_dlm_queue_lockres(lockres
);
208 static void user_bast(void *opaque
, int level
)
210 struct user_lock_res
*lockres
= opaque
;
212 mlog(0, "Blocking AST fired for lockres %s. Blocking level %d\n",
213 lockres
->l_name
, level
);
215 spin_lock(&lockres
->l_lock
);
216 lockres
->l_flags
|= USER_LOCK_BLOCKED
;
217 if (level
> lockres
->l_blocking
)
218 lockres
->l_blocking
= level
;
220 __user_dlm_queue_lockres(lockres
);
221 spin_unlock(&lockres
->l_lock
);
223 wake_up(&lockres
->l_event
);
226 static void user_unlock_ast(void *opaque
, enum dlm_status status
)
228 struct user_lock_res
*lockres
= opaque
;
230 mlog(0, "UNLOCK AST called on lock %s\n", lockres
->l_name
);
232 if (status
!= DLM_NORMAL
)
233 mlog(ML_ERROR
, "Dlm returns status %d\n", status
);
235 spin_lock(&lockres
->l_lock
);
236 if (lockres
->l_flags
& USER_LOCK_IN_TEARDOWN
)
237 lockres
->l_level
= LKM_IVMODE
;
239 lockres
->l_requested
= LKM_IVMODE
; /* cancel an
242 lockres
->l_flags
&= ~USER_LOCK_IN_CANCEL
;
243 /* we want the unblock thread to look at it again
245 __user_dlm_queue_lockres(lockres
);
248 lockres
->l_flags
&= ~USER_LOCK_BUSY
;
249 spin_unlock(&lockres
->l_lock
);
251 wake_up(&lockres
->l_event
);
254 static inline void user_dlm_drop_inode_ref(struct user_lock_res
*lockres
)
257 inode
= user_dlm_inode_from_user_lockres(lockres
);
261 static void user_dlm_unblock_lock(void *opaque
)
263 int new_level
, status
;
264 struct user_lock_res
*lockres
= (struct user_lock_res
*) opaque
;
265 struct dlm_ctxt
*dlm
= dlm_ctxt_from_user_lockres(lockres
);
267 mlog(0, "processing lockres %s\n", lockres
->l_name
);
269 spin_lock(&lockres
->l_lock
);
271 BUG_ON(!(lockres
->l_flags
& USER_LOCK_BLOCKED
));
272 BUG_ON(!(lockres
->l_flags
& USER_LOCK_QUEUED
));
274 /* notice that we don't clear USER_LOCK_BLOCKED here. That's
275 * for user_ast to do. */
276 lockres
->l_flags
&= ~USER_LOCK_QUEUED
;
278 if (lockres
->l_flags
& USER_LOCK_IN_TEARDOWN
) {
279 mlog(0, "lock is in teardown so we do nothing\n");
280 spin_unlock(&lockres
->l_lock
);
284 if (lockres
->l_flags
& USER_LOCK_BUSY
) {
285 mlog(0, "BUSY flag detected...\n");
286 if (lockres
->l_flags
& USER_LOCK_IN_CANCEL
) {
287 spin_unlock(&lockres
->l_lock
);
291 lockres
->l_flags
|= USER_LOCK_IN_CANCEL
;
292 spin_unlock(&lockres
->l_lock
);
294 status
= dlmunlock(dlm
,
299 if (status
== DLM_CANCELGRANT
) {
300 /* If we got this, then the ast was fired
301 * before we could cancel. We cleanup our
302 * state, and restart the function. */
303 spin_lock(&lockres
->l_lock
);
304 lockres
->l_flags
&= ~USER_LOCK_IN_CANCEL
;
305 spin_unlock(&lockres
->l_lock
);
306 } else if (status
!= DLM_NORMAL
)
307 user_log_dlm_error("dlmunlock", status
, lockres
);
311 /* If there are still incompat holders, we can exit safely
312 * without worrying about re-queueing this lock as that will
313 * happen on the last call to user_cluster_unlock. */
314 if ((lockres
->l_blocking
== LKM_EXMODE
)
315 && (lockres
->l_ex_holders
|| lockres
->l_ro_holders
)) {
316 spin_unlock(&lockres
->l_lock
);
317 mlog(0, "can't downconvert for ex: ro = %u, ex = %u\n",
318 lockres
->l_ro_holders
, lockres
->l_ex_holders
);
322 if ((lockres
->l_blocking
== LKM_PRMODE
)
323 && lockres
->l_ex_holders
) {
324 spin_unlock(&lockres
->l_lock
);
325 mlog(0, "can't downconvert for pr: ex = %u\n",
326 lockres
->l_ex_holders
);
330 /* yay, we can downconvert now. */
331 new_level
= user_highest_compat_lock_level(lockres
->l_blocking
);
332 lockres
->l_requested
= new_level
;
333 lockres
->l_flags
|= USER_LOCK_BUSY
;
334 mlog(0, "Downconvert lock from %d to %d\n",
335 lockres
->l_level
, new_level
);
336 spin_unlock(&lockres
->l_lock
);
338 /* need lock downconvert request now... */
339 status
= dlmlock(dlm
,
342 LKM_CONVERT
|LKM_VALBLK
,
347 if (status
!= DLM_NORMAL
) {
348 user_log_dlm_error("dlmlock", status
, lockres
);
349 user_recover_from_dlm_error(lockres
);
353 user_dlm_drop_inode_ref(lockres
);
356 static inline void user_dlm_inc_holders(struct user_lock_res
*lockres
,
361 lockres
->l_ex_holders
++;
364 lockres
->l_ro_holders
++;
371 /* predict what lock level we'll be dropping down to on behalf
372 * of another node, and return true if the currently wanted
373 * level will be compatible with it. */
375 user_may_continue_on_blocked_lock(struct user_lock_res
*lockres
,
378 BUG_ON(!(lockres
->l_flags
& USER_LOCK_BLOCKED
));
380 return wanted
<= user_highest_compat_lock_level(lockres
->l_blocking
);
383 int user_dlm_cluster_lock(struct user_lock_res
*lockres
,
387 int status
, local_flags
;
388 struct dlm_ctxt
*dlm
= dlm_ctxt_from_user_lockres(lockres
);
390 if (level
!= LKM_EXMODE
&&
391 level
!= LKM_PRMODE
) {
392 mlog(ML_ERROR
, "lockres %s: invalid request!\n",
398 mlog(0, "lockres %s: asking for %s lock, passed flags = 0x%x\n",
400 (level
== LKM_EXMODE
) ? "LKM_EXMODE" : "LKM_PRMODE",
404 if (signal_pending(current
)) {
405 status
= -ERESTARTSYS
;
409 spin_lock(&lockres
->l_lock
);
411 /* We only compare against the currently granted level
412 * here. If the lock is blocked waiting on a downconvert,
413 * we'll get caught below. */
414 if ((lockres
->l_flags
& USER_LOCK_BUSY
) &&
415 (level
> lockres
->l_level
)) {
416 /* is someone sitting in dlm_lock? If so, wait on
418 spin_unlock(&lockres
->l_lock
);
420 user_wait_on_busy_lock(lockres
);
424 if ((lockres
->l_flags
& USER_LOCK_BLOCKED
) &&
425 (!user_may_continue_on_blocked_lock(lockres
, level
))) {
426 /* is the lock is currently blocked on behalf of
428 spin_unlock(&lockres
->l_lock
);
430 user_wait_on_blocked_lock(lockres
);
434 if (level
> lockres
->l_level
) {
435 local_flags
= lkm_flags
| LKM_VALBLK
;
436 if (lockres
->l_level
!= LKM_IVMODE
)
437 local_flags
|= LKM_CONVERT
;
439 lockres
->l_requested
= level
;
440 lockres
->l_flags
|= USER_LOCK_BUSY
;
441 spin_unlock(&lockres
->l_lock
);
443 BUG_ON(level
== LKM_IVMODE
);
444 BUG_ON(level
== LKM_NLMODE
);
446 mlog(0, "lock %s, get lock from %d to level = %d\n",
447 lockres
->l_name
, lockres
->l_level
, level
);
449 /* call dlm_lock to upgrade lock now */
450 status
= dlmlock(dlm
,
458 if (status
!= DLM_NORMAL
) {
459 if ((lkm_flags
& LKM_NOQUEUE
) &&
460 (status
== DLM_NOTQUEUED
))
463 user_log_dlm_error("dlmlock", status
, lockres
);
466 user_recover_from_dlm_error(lockres
);
470 mlog(0, "lock %s, successfull return from dlmlock\n",
473 user_wait_on_busy_lock(lockres
);
477 user_dlm_inc_holders(lockres
, level
);
478 spin_unlock(&lockres
->l_lock
);
480 mlog(0, "lockres %s: Got %s lock!\n", lockres
->l_name
,
481 (level
== LKM_EXMODE
) ? "LKM_EXMODE" : "LKM_PRMODE");
488 static inline void user_dlm_dec_holders(struct user_lock_res
*lockres
,
493 BUG_ON(!lockres
->l_ex_holders
);
494 lockres
->l_ex_holders
--;
497 BUG_ON(!lockres
->l_ro_holders
);
498 lockres
->l_ro_holders
--;
505 void user_dlm_cluster_unlock(struct user_lock_res
*lockres
,
508 if (level
!= LKM_EXMODE
&&
509 level
!= LKM_PRMODE
) {
510 mlog(ML_ERROR
, "lockres %s: invalid request!\n", lockres
->l_name
);
514 mlog(0, "lockres %s: dropping %s lock\n", lockres
->l_name
,
515 (level
== LKM_EXMODE
) ? "LKM_EXMODE" : "LKM_PRMODE");
517 spin_lock(&lockres
->l_lock
);
518 user_dlm_dec_holders(lockres
, level
);
519 __user_dlm_cond_queue_lockres(lockres
);
520 spin_unlock(&lockres
->l_lock
);
523 void user_dlm_write_lvb(struct inode
*inode
,
527 struct user_lock_res
*lockres
= &DLMFS_I(inode
)->ip_lockres
;
528 char *lvb
= lockres
->l_lksb
.lvb
;
530 BUG_ON(len
> DLM_LVB_LEN
);
532 spin_lock(&lockres
->l_lock
);
534 BUG_ON(lockres
->l_level
< LKM_EXMODE
);
535 memcpy(lvb
, val
, len
);
537 spin_unlock(&lockres
->l_lock
);
540 void user_dlm_read_lvb(struct inode
*inode
,
544 struct user_lock_res
*lockres
= &DLMFS_I(inode
)->ip_lockres
;
545 char *lvb
= lockres
->l_lksb
.lvb
;
547 BUG_ON(len
> DLM_LVB_LEN
);
549 spin_lock(&lockres
->l_lock
);
551 BUG_ON(lockres
->l_level
< LKM_PRMODE
);
552 memcpy(val
, lvb
, len
);
554 spin_unlock(&lockres
->l_lock
);
557 void user_dlm_lock_res_init(struct user_lock_res
*lockres
,
558 struct dentry
*dentry
)
560 memset(lockres
, 0, sizeof(*lockres
));
562 spin_lock_init(&lockres
->l_lock
);
563 init_waitqueue_head(&lockres
->l_event
);
564 lockres
->l_level
= LKM_IVMODE
;
565 lockres
->l_requested
= LKM_IVMODE
;
566 lockres
->l_blocking
= LKM_IVMODE
;
568 /* should have been checked before getting here. */
569 BUG_ON(dentry
->d_name
.len
>= USER_DLM_LOCK_ID_MAX_LEN
);
571 memcpy(lockres
->l_name
,
576 int user_dlm_destroy_lock(struct user_lock_res
*lockres
)
579 struct dlm_ctxt
*dlm
= dlm_ctxt_from_user_lockres(lockres
);
581 mlog(0, "asked to destroy %s\n", lockres
->l_name
);
583 spin_lock(&lockres
->l_lock
);
584 while (lockres
->l_flags
& USER_LOCK_BUSY
) {
585 spin_unlock(&lockres
->l_lock
);
587 mlog(0, "lock %s is busy\n", lockres
->l_name
);
589 user_wait_on_busy_lock(lockres
);
591 spin_lock(&lockres
->l_lock
);
594 if (lockres
->l_ro_holders
|| lockres
->l_ex_holders
) {
595 spin_unlock(&lockres
->l_lock
);
596 mlog(0, "lock %s has holders\n", lockres
->l_name
);
601 if (!(lockres
->l_flags
& USER_LOCK_ATTACHED
)) {
602 spin_unlock(&lockres
->l_lock
);
603 mlog(0, "lock %s is not attached\n", lockres
->l_name
);
607 lockres
->l_flags
&= ~USER_LOCK_ATTACHED
;
608 lockres
->l_flags
|= USER_LOCK_BUSY
;
609 lockres
->l_flags
|= USER_LOCK_IN_TEARDOWN
;
610 spin_unlock(&lockres
->l_lock
);
612 mlog(0, "unlocking lockres %s\n", lockres
->l_name
);
613 status
= dlmunlock(dlm
,
618 if (status
!= DLM_NORMAL
) {
619 user_log_dlm_error("dlmunlock", status
, lockres
);
624 user_wait_on_busy_lock(lockres
);
631 struct dlm_ctxt
*user_dlm_register_context(struct qstr
*name
)
633 struct dlm_ctxt
*dlm
;
637 domain
= kmalloc(name
->len
+ 1, GFP_KERNEL
);
640 return ERR_PTR(-ENOMEM
);
643 dlm_key
= crc32_le(0, name
->name
, name
->len
);
645 snprintf(domain
, name
->len
+ 1, "%.*s", name
->len
, name
->name
);
647 dlm
= dlm_register_domain(domain
, dlm_key
);
649 mlog_errno(PTR_ERR(dlm
));
655 void user_dlm_unregister_context(struct dlm_ctxt
*dlm
)
657 dlm_unregister_domain(dlm
);