aboutsummaryrefslogtreecommitdiffstats
path: root/fs/gfs2/locking/dlm
diff options
context:
space:
mode:
Diffstat (limited to 'fs/gfs2/locking/dlm')
-rw-r--r--fs/gfs2/locking/dlm/Makefile3
-rw-r--r--fs/gfs2/locking/dlm/lock.c524
-rw-r--r--fs/gfs2/locking/dlm/lock_dlm.h187
-rw-r--r--fs/gfs2/locking/dlm/main.c64
-rw-r--r--fs/gfs2/locking/dlm/mount.c255
-rw-r--r--fs/gfs2/locking/dlm/plock.c301
-rw-r--r--fs/gfs2/locking/dlm/sysfs.c226
-rw-r--r--fs/gfs2/locking/dlm/thread.c359
8 files changed, 1919 insertions, 0 deletions
diff --git a/fs/gfs2/locking/dlm/Makefile b/fs/gfs2/locking/dlm/Makefile
new file mode 100644
index 000000000000..89b93b6b45cf
--- /dev/null
+++ b/fs/gfs2/locking/dlm/Makefile
@@ -0,0 +1,3 @@
1obj-$(CONFIG_GFS2_FS_LOCKING_DLM) += lock_dlm.o
2lock_dlm-y := lock.o main.o mount.o sysfs.o thread.o plock.o
3
diff --git a/fs/gfs2/locking/dlm/lock.c b/fs/gfs2/locking/dlm/lock.c
new file mode 100644
index 000000000000..b167addf9fd1
--- /dev/null
+++ b/fs/gfs2/locking/dlm/lock.c
@@ -0,0 +1,524 @@
1/*
2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2005 Red Hat, Inc. All rights reserved.
4 *
5 * This copyrighted material is made available to anyone wishing to use,
6 * modify, copy, or redistribute it subject to the terms and conditions
7 * of the GNU General Public License version 2.
8 */
9
10#include "lock_dlm.h"
11
12static char junk_lvb[GDLM_LVB_SIZE];
13
14static void queue_complete(struct gdlm_lock *lp)
15{
16 struct gdlm_ls *ls = lp->ls;
17
18 clear_bit(LFL_ACTIVE, &lp->flags);
19
20 spin_lock(&ls->async_lock);
21 list_add_tail(&lp->clist, &ls->complete);
22 spin_unlock(&ls->async_lock);
23 wake_up(&ls->thread_wait);
24}
25
26static inline void gdlm_ast(void *astarg)
27{
28 queue_complete(astarg);
29}
30
31static inline void gdlm_bast(void *astarg, int mode)
32{
33 struct gdlm_lock *lp = astarg;
34 struct gdlm_ls *ls = lp->ls;
35
36 if (!mode) {
37 printk(KERN_INFO "lock_dlm: bast mode zero %x,%llx\n",
38 lp->lockname.ln_type,
39 (unsigned long long)lp->lockname.ln_number);
40 return;
41 }
42
43 spin_lock(&ls->async_lock);
44 if (!lp->bast_mode) {
45 list_add_tail(&lp->blist, &ls->blocking);
46 lp->bast_mode = mode;
47 } else if (lp->bast_mode < mode)
48 lp->bast_mode = mode;
49 spin_unlock(&ls->async_lock);
50 wake_up(&ls->thread_wait);
51}
52
53void gdlm_queue_delayed(struct gdlm_lock *lp)
54{
55 struct gdlm_ls *ls = lp->ls;
56
57 spin_lock(&ls->async_lock);
58 list_add_tail(&lp->delay_list, &ls->delayed);
59 spin_unlock(&ls->async_lock);
60}
61
62/* convert gfs lock-state to dlm lock-mode */
63
64static s16 make_mode(s16 lmstate)
65{
66 switch (lmstate) {
67 case LM_ST_UNLOCKED:
68 return DLM_LOCK_NL;
69 case LM_ST_EXCLUSIVE:
70 return DLM_LOCK_EX;
71 case LM_ST_DEFERRED:
72 return DLM_LOCK_CW;
73 case LM_ST_SHARED:
74 return DLM_LOCK_PR;
75 }
76 gdlm_assert(0, "unknown LM state %d", lmstate);
77 return -1;
78}
79
80/* convert dlm lock-mode to gfs lock-state */
81
82s16 gdlm_make_lmstate(s16 dlmmode)
83{
84 switch (dlmmode) {
85 case DLM_LOCK_IV:
86 case DLM_LOCK_NL:
87 return LM_ST_UNLOCKED;
88 case DLM_LOCK_EX:
89 return LM_ST_EXCLUSIVE;
90 case DLM_LOCK_CW:
91 return LM_ST_DEFERRED;
92 case DLM_LOCK_PR:
93 return LM_ST_SHARED;
94 }
95 gdlm_assert(0, "unknown DLM mode %d", dlmmode);
96 return -1;
97}
98
99/* verify agreement with GFS on the current lock state, NB: DLM_LOCK_NL and
100 DLM_LOCK_IV are both considered LM_ST_UNLOCKED by GFS. */
101
102static void check_cur_state(struct gdlm_lock *lp, unsigned int cur_state)
103{
104 s16 cur = make_mode(cur_state);
105 if (lp->cur != DLM_LOCK_IV)
106 gdlm_assert(lp->cur == cur, "%d, %d", lp->cur, cur);
107}
108
109static inline unsigned int make_flags(struct gdlm_lock *lp,
110 unsigned int gfs_flags,
111 s16 cur, s16 req)
112{
113 unsigned int lkf = 0;
114
115 if (gfs_flags & LM_FLAG_TRY)
116 lkf |= DLM_LKF_NOQUEUE;
117
118 if (gfs_flags & LM_FLAG_TRY_1CB) {
119 lkf |= DLM_LKF_NOQUEUE;
120 lkf |= DLM_LKF_NOQUEUEBAST;
121 }
122
123 if (gfs_flags & LM_FLAG_PRIORITY) {
124 lkf |= DLM_LKF_NOORDER;
125 lkf |= DLM_LKF_HEADQUE;
126 }
127
128 if (gfs_flags & LM_FLAG_ANY) {
129 if (req == DLM_LOCK_PR)
130 lkf |= DLM_LKF_ALTCW;
131 else if (req == DLM_LOCK_CW)
132 lkf |= DLM_LKF_ALTPR;
133 }
134
135 if (lp->lksb.sb_lkid != 0) {
136 lkf |= DLM_LKF_CONVERT;
137
138 /* Conversion deadlock avoidance by DLM */
139
140 if (!test_bit(LFL_FORCE_PROMOTE, &lp->flags) &&
141 !(lkf & DLM_LKF_NOQUEUE) &&
142 cur > DLM_LOCK_NL && req > DLM_LOCK_NL && cur != req)
143 lkf |= DLM_LKF_CONVDEADLK;
144 }
145
146 if (lp->lvb)
147 lkf |= DLM_LKF_VALBLK;
148
149 return lkf;
150}
151
152/* make_strname - convert GFS lock numbers to a string */
153
154static inline void make_strname(struct lm_lockname *lockname,
155 struct gdlm_strname *str)
156{
157 sprintf(str->name, "%8x%16llx", lockname->ln_type,
158 (unsigned long long)lockname->ln_number);
159 str->namelen = GDLM_STRNAME_BYTES;
160}
161
162static int gdlm_create_lp(struct gdlm_ls *ls, struct lm_lockname *name,
163 struct gdlm_lock **lpp)
164{
165 struct gdlm_lock *lp;
166
167 lp = kzalloc(sizeof(struct gdlm_lock), GFP_KERNEL);
168 if (!lp)
169 return -ENOMEM;
170
171 lp->lockname = *name;
172 lp->ls = ls;
173 lp->cur = DLM_LOCK_IV;
174 lp->lvb = NULL;
175 lp->hold_null = NULL;
176 init_completion(&lp->ast_wait);
177 INIT_LIST_HEAD(&lp->clist);
178 INIT_LIST_HEAD(&lp->blist);
179 INIT_LIST_HEAD(&lp->delay_list);
180
181 spin_lock(&ls->async_lock);
182 list_add(&lp->all_list, &ls->all_locks);
183 ls->all_locks_count++;
184 spin_unlock(&ls->async_lock);
185
186 *lpp = lp;
187 return 0;
188}
189
190void gdlm_delete_lp(struct gdlm_lock *lp)
191{
192 struct gdlm_ls *ls = lp->ls;
193
194 spin_lock(&ls->async_lock);
195 if (!list_empty(&lp->clist))
196 list_del_init(&lp->clist);
197 if (!list_empty(&lp->blist))
198 list_del_init(&lp->blist);
199 if (!list_empty(&lp->delay_list))
200 list_del_init(&lp->delay_list);
201 gdlm_assert(!list_empty(&lp->all_list), "%x,%llx", lp->lockname.ln_type,
202 (unsigned long long)lp->lockname.ln_number);
203 list_del_init(&lp->all_list);
204 ls->all_locks_count--;
205 spin_unlock(&ls->async_lock);
206
207 kfree(lp);
208}
209
210int gdlm_get_lock(void *lockspace, struct lm_lockname *name,
211 void **lockp)
212{
213 struct gdlm_lock *lp;
214 int error;
215
216 error = gdlm_create_lp(lockspace, name, &lp);
217
218 *lockp = lp;
219 return error;
220}
221
222void gdlm_put_lock(void *lock)
223{
224 gdlm_delete_lp(lock);
225}
226
227unsigned int gdlm_do_lock(struct gdlm_lock *lp)
228{
229 struct gdlm_ls *ls = lp->ls;
230 struct gdlm_strname str;
231 int error, bast = 1;
232
233 /*
234 * When recovery is in progress, delay lock requests for submission
235 * once recovery is done. Requests for recovery (NOEXP) and unlocks
236 * can pass.
237 */
238
239 if (test_bit(DFL_BLOCK_LOCKS, &ls->flags) &&
240 !test_bit(LFL_NOBLOCK, &lp->flags) && lp->req != DLM_LOCK_NL) {
241 gdlm_queue_delayed(lp);
242 return LM_OUT_ASYNC;
243 }
244
245 /*
246 * Submit the actual lock request.
247 */
248
249 if (test_bit(LFL_NOBAST, &lp->flags))
250 bast = 0;
251
252 make_strname(&lp->lockname, &str);
253
254 set_bit(LFL_ACTIVE, &lp->flags);
255
256 log_debug("lk %x,%llx id %x %d,%d %x", lp->lockname.ln_type,
257 (unsigned long long)lp->lockname.ln_number, lp->lksb.sb_lkid,
258 lp->cur, lp->req, lp->lkf);
259
260 error = dlm_lock(ls->dlm_lockspace, lp->req, &lp->lksb, lp->lkf,
261 str.name, str.namelen, 0, gdlm_ast, lp,
262 bast ? gdlm_bast : NULL);
263
264 if ((error == -EAGAIN) && (lp->lkf & DLM_LKF_NOQUEUE)) {
265 lp->lksb.sb_status = -EAGAIN;
266 queue_complete(lp);
267 error = 0;
268 }
269
270 if (error) {
271 log_debug("%s: gdlm_lock %x,%llx err=%d cur=%d req=%d lkf=%x "
272 "flags=%lx", ls->fsname, lp->lockname.ln_type,
273 (unsigned long long)lp->lockname.ln_number, error,
274 lp->cur, lp->req, lp->lkf, lp->flags);
275 return LM_OUT_ERROR;
276 }
277 return LM_OUT_ASYNC;
278}
279
280static unsigned int gdlm_do_unlock(struct gdlm_lock *lp)
281{
282 struct gdlm_ls *ls = lp->ls;
283 unsigned int lkf = 0;
284 int error;
285
286 set_bit(LFL_DLM_UNLOCK, &lp->flags);
287 set_bit(LFL_ACTIVE, &lp->flags);
288
289 if (lp->lvb)
290 lkf = DLM_LKF_VALBLK;
291
292 log_debug("un %x,%llx %x %d %x", lp->lockname.ln_type,
293 (unsigned long long)lp->lockname.ln_number,
294 lp->lksb.sb_lkid, lp->cur, lkf);
295
296 error = dlm_unlock(ls->dlm_lockspace, lp->lksb.sb_lkid, lkf, NULL, lp);
297
298 if (error) {
299 log_debug("%s: gdlm_unlock %x,%llx err=%d cur=%d req=%d lkf=%x "
300 "flags=%lx", ls->fsname, lp->lockname.ln_type,
301 (unsigned long long)lp->lockname.ln_number, error,
302 lp->cur, lp->req, lp->lkf, lp->flags);
303 return LM_OUT_ERROR;
304 }
305 return LM_OUT_ASYNC;
306}
307
308unsigned int gdlm_lock(void *lock, unsigned int cur_state,
309 unsigned int req_state, unsigned int flags)
310{
311 struct gdlm_lock *lp = lock;
312
313 clear_bit(LFL_DLM_CANCEL, &lp->flags);
314 if (flags & LM_FLAG_NOEXP)
315 set_bit(LFL_NOBLOCK, &lp->flags);
316
317 check_cur_state(lp, cur_state);
318 lp->req = make_mode(req_state);
319 lp->lkf = make_flags(lp, flags, lp->cur, lp->req);
320
321 return gdlm_do_lock(lp);
322}
323
324unsigned int gdlm_unlock(void *lock, unsigned int cur_state)
325{
326 struct gdlm_lock *lp = lock;
327
328 clear_bit(LFL_DLM_CANCEL, &lp->flags);
329 if (lp->cur == DLM_LOCK_IV)
330 return 0;
331 return gdlm_do_unlock(lp);
332}
333
334void gdlm_cancel(void *lock)
335{
336 struct gdlm_lock *lp = lock;
337 struct gdlm_ls *ls = lp->ls;
338 int error, delay_list = 0;
339
340 if (test_bit(LFL_DLM_CANCEL, &lp->flags))
341 return;
342
343 log_info("gdlm_cancel %x,%llx flags %lx", lp->lockname.ln_type,
344 (unsigned long long)lp->lockname.ln_number, lp->flags);
345
346 spin_lock(&ls->async_lock);
347 if (!list_empty(&lp->delay_list)) {
348 list_del_init(&lp->delay_list);
349 delay_list = 1;
350 }
351 spin_unlock(&ls->async_lock);
352
353 if (delay_list) {
354 set_bit(LFL_CANCEL, &lp->flags);
355 set_bit(LFL_ACTIVE, &lp->flags);
356 queue_complete(lp);
357 return;
358 }
359
360 if (!test_bit(LFL_ACTIVE, &lp->flags) ||
361 test_bit(LFL_DLM_UNLOCK, &lp->flags)) {
362 log_info("gdlm_cancel skip %x,%llx flags %lx",
363 lp->lockname.ln_type,
364 (unsigned long long)lp->lockname.ln_number, lp->flags);
365 return;
366 }
367
368 /* the lock is blocked in the dlm */
369
370 set_bit(LFL_DLM_CANCEL, &lp->flags);
371 set_bit(LFL_ACTIVE, &lp->flags);
372
373 error = dlm_unlock(ls->dlm_lockspace, lp->lksb.sb_lkid, DLM_LKF_CANCEL,
374 NULL, lp);
375
376 log_info("gdlm_cancel rv %d %x,%llx flags %lx", error,
377 lp->lockname.ln_type,
378 (unsigned long long)lp->lockname.ln_number, lp->flags);
379
380 if (error == -EBUSY)
381 clear_bit(LFL_DLM_CANCEL, &lp->flags);
382}
383
384static int gdlm_add_lvb(struct gdlm_lock *lp)
385{
386 char *lvb;
387
388 lvb = kzalloc(GDLM_LVB_SIZE, GFP_KERNEL);
389 if (!lvb)
390 return -ENOMEM;
391
392 lp->lksb.sb_lvbptr = lvb;
393 lp->lvb = lvb;
394 return 0;
395}
396
397static void gdlm_del_lvb(struct gdlm_lock *lp)
398{
399 kfree(lp->lvb);
400 lp->lvb = NULL;
401 lp->lksb.sb_lvbptr = NULL;
402}
403
404/* This can do a synchronous dlm request (requiring a lock_dlm thread to get
405 the completion) because gfs won't call hold_lvb() during a callback (from
406 the context of a lock_dlm thread). */
407
408static int hold_null_lock(struct gdlm_lock *lp)
409{
410 struct gdlm_lock *lpn = NULL;
411 int error;
412
413 if (lp->hold_null) {
414 printk(KERN_INFO "lock_dlm: lvb already held\n");
415 return 0;
416 }
417
418 error = gdlm_create_lp(lp->ls, &lp->lockname, &lpn);
419 if (error)
420 goto out;
421
422 lpn->lksb.sb_lvbptr = junk_lvb;
423 lpn->lvb = junk_lvb;
424
425 lpn->req = DLM_LOCK_NL;
426 lpn->lkf = DLM_LKF_VALBLK | DLM_LKF_EXPEDITE;
427 set_bit(LFL_NOBAST, &lpn->flags);
428 set_bit(LFL_INLOCK, &lpn->flags);
429
430 init_completion(&lpn->ast_wait);
431 gdlm_do_lock(lpn);
432 wait_for_completion(&lpn->ast_wait);
433 error = lpn->lksb.sb_status;
434 if (error) {
435 printk(KERN_INFO "lock_dlm: hold_null_lock dlm error %d\n",
436 error);
437 gdlm_delete_lp(lpn);
438 lpn = NULL;
439 }
440out:
441 lp->hold_null = lpn;
442 return error;
443}
444
445/* This cannot do a synchronous dlm request (requiring a lock_dlm thread to get
446 the completion) because gfs may call unhold_lvb() during a callback (from
447 the context of a lock_dlm thread) which could cause a deadlock since the
448 other lock_dlm thread could be engaged in recovery. */
449
450static void unhold_null_lock(struct gdlm_lock *lp)
451{
452 struct gdlm_lock *lpn = lp->hold_null;
453
454 gdlm_assert(lpn, "%x,%llx", lp->lockname.ln_type,
455 (unsigned long long)lp->lockname.ln_number);
456 lpn->lksb.sb_lvbptr = NULL;
457 lpn->lvb = NULL;
458 set_bit(LFL_UNLOCK_DELETE, &lpn->flags);
459 gdlm_do_unlock(lpn);
460 lp->hold_null = NULL;
461}
462
463/* Acquire a NL lock because gfs requires the value block to remain
464 intact on the resource while the lvb is "held" even if it's holding no locks
465 on the resource. */
466
467int gdlm_hold_lvb(void *lock, char **lvbp)
468{
469 struct gdlm_lock *lp = lock;
470 int error;
471
472 error = gdlm_add_lvb(lp);
473 if (error)
474 return error;
475
476 *lvbp = lp->lvb;
477
478 error = hold_null_lock(lp);
479 if (error)
480 gdlm_del_lvb(lp);
481
482 return error;
483}
484
485void gdlm_unhold_lvb(void *lock, char *lvb)
486{
487 struct gdlm_lock *lp = lock;
488
489 unhold_null_lock(lp);
490 gdlm_del_lvb(lp);
491}
492
493void gdlm_submit_delayed(struct gdlm_ls *ls)
494{
495 struct gdlm_lock *lp, *safe;
496
497 spin_lock(&ls->async_lock);
498 list_for_each_entry_safe(lp, safe, &ls->delayed, delay_list) {
499 list_del_init(&lp->delay_list);
500 list_add_tail(&lp->delay_list, &ls->submit);
501 }
502 spin_unlock(&ls->async_lock);
503 wake_up(&ls->thread_wait);
504}
505
506int gdlm_release_all_locks(struct gdlm_ls *ls)
507{
508 struct gdlm_lock *lp, *safe;
509 int count = 0;
510
511 spin_lock(&ls->async_lock);
512 list_for_each_entry_safe(lp, safe, &ls->all_locks, all_list) {
513 list_del_init(&lp->all_list);
514
515 if (lp->lvb && lp->lvb != junk_lvb)
516 kfree(lp->lvb);
517 kfree(lp);
518 count++;
519 }
520 spin_unlock(&ls->async_lock);
521
522 return count;
523}
524
diff --git a/fs/gfs2/locking/dlm/lock_dlm.h b/fs/gfs2/locking/dlm/lock_dlm.h
new file mode 100644
index 000000000000..33af707a4d3f
--- /dev/null
+++ b/fs/gfs2/locking/dlm/lock_dlm.h
@@ -0,0 +1,187 @@
1/*
2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2005 Red Hat, Inc. All rights reserved.
4 *
5 * This copyrighted material is made available to anyone wishing to use,
6 * modify, copy, or redistribute it subject to the terms and conditions
7 * of the GNU General Public License version 2.
8 */
9
10#ifndef LOCK_DLM_DOT_H
11#define LOCK_DLM_DOT_H
12
13#include <linux/module.h>
14#include <linux/slab.h>
15#include <linux/spinlock.h>
16#include <linux/module.h>
17#include <linux/types.h>
18#include <linux/string.h>
19#include <linux/list.h>
20#include <linux/socket.h>
21#include <linux/delay.h>
22#include <linux/kthread.h>
23#include <linux/kobject.h>
24#include <linux/fcntl.h>
25#include <linux/wait.h>
26#include <net/sock.h>
27
28#include <linux/dlm.h>
29#include <linux/lm_interface.h>
30
31/*
32 * Internally, we prefix things with gdlm_ and GDLM_ (for gfs-dlm) since a
33 * prefix of lock_dlm_ gets awkward. Externally, GFS refers to this module
34 * as "lock_dlm".
35 */
36
37#define GDLM_STRNAME_BYTES 24
38#define GDLM_LVB_SIZE 32
39#define GDLM_DROP_COUNT 50000
40#define GDLM_DROP_PERIOD 60
41#define GDLM_NAME_LEN 128
42
43/* GFS uses 12 bytes to identify a resource (32 bit type + 64 bit number).
44 We sprintf these numbers into a 24 byte string of hex values to make them
45 human-readable (to make debugging simpler.) */
46
47struct gdlm_strname {
48 unsigned char name[GDLM_STRNAME_BYTES];
49 unsigned short namelen;
50};
51
52enum {
53 DFL_BLOCK_LOCKS = 0,
54 DFL_SPECTATOR = 1,
55 DFL_WITHDRAW = 2,
56};
57
58struct gdlm_ls {
59 u32 id;
60 int jid;
61 int first;
62 int first_done;
63 unsigned long flags;
64 struct kobject kobj;
65 char clustername[GDLM_NAME_LEN];
66 char fsname[GDLM_NAME_LEN];
67 int fsflags;
68 dlm_lockspace_t *dlm_lockspace;
69 lm_callback_t fscb;
70 struct gfs2_sbd *sdp;
71 int recover_jid;
72 int recover_jid_done;
73 int recover_jid_status;
74 spinlock_t async_lock;
75 struct list_head complete;
76 struct list_head blocking;
77 struct list_head delayed;
78 struct list_head submit;
79 struct list_head all_locks;
80 u32 all_locks_count;
81 wait_queue_head_t wait_control;
82 struct task_struct *thread1;
83 struct task_struct *thread2;
84 wait_queue_head_t thread_wait;
85 unsigned long drop_time;
86 int drop_locks_count;
87 int drop_locks_period;
88};
89
90enum {
91 LFL_NOBLOCK = 0,
92 LFL_NOCACHE = 1,
93 LFL_DLM_UNLOCK = 2,
94 LFL_DLM_CANCEL = 3,
95 LFL_SYNC_LVB = 4,
96 LFL_FORCE_PROMOTE = 5,
97 LFL_REREQUEST = 6,
98 LFL_ACTIVE = 7,
99 LFL_INLOCK = 8,
100 LFL_CANCEL = 9,
101 LFL_NOBAST = 10,
102 LFL_HEADQUE = 11,
103 LFL_UNLOCK_DELETE = 12,
104};
105
106struct gdlm_lock {
107 struct gdlm_ls *ls;
108 struct lm_lockname lockname;
109 char *lvb;
110 struct dlm_lksb lksb;
111
112 s16 cur;
113 s16 req;
114 s16 prev_req;
115 u32 lkf; /* dlm flags DLM_LKF_ */
116 unsigned long flags; /* lock_dlm flags LFL_ */
117
118 int bast_mode; /* protected by async_lock */
119 struct completion ast_wait;
120
121 struct list_head clist; /* complete */
122 struct list_head blist; /* blocking */
123 struct list_head delay_list; /* delayed */
124 struct list_head all_list; /* all locks for the fs */
125 struct gdlm_lock *hold_null; /* NL lock for hold_lvb */
126};
127
128#define gdlm_assert(assertion, fmt, args...) \
129do { \
130 if (unlikely(!(assertion))) { \
131 printk(KERN_EMERG "lock_dlm: fatal assertion failed \"%s\"\n" \
132 "lock_dlm: " fmt "\n", \
133 #assertion, ##args); \
134 BUG(); \
135 } \
136} while (0)
137
138#define log_print(lev, fmt, arg...) printk(lev "lock_dlm: " fmt "\n" , ## arg)
139#define log_info(fmt, arg...) log_print(KERN_INFO , fmt , ## arg)
140#define log_error(fmt, arg...) log_print(KERN_ERR , fmt , ## arg)
141#ifdef LOCK_DLM_LOG_DEBUG
142#define log_debug(fmt, arg...) log_print(KERN_DEBUG , fmt , ## arg)
143#else
144#define log_debug(fmt, arg...)
145#endif
146
147/* sysfs.c */
148
149int gdlm_sysfs_init(void);
150void gdlm_sysfs_exit(void);
151int gdlm_kobject_setup(struct gdlm_ls *, struct kobject *);
152void gdlm_kobject_release(struct gdlm_ls *);
153
154/* thread.c */
155
156int gdlm_init_threads(struct gdlm_ls *);
157void gdlm_release_threads(struct gdlm_ls *);
158
159/* lock.c */
160
161s16 gdlm_make_lmstate(s16);
162void gdlm_queue_delayed(struct gdlm_lock *);
163void gdlm_submit_delayed(struct gdlm_ls *);
164int gdlm_release_all_locks(struct gdlm_ls *);
165void gdlm_delete_lp(struct gdlm_lock *);
166unsigned int gdlm_do_lock(struct gdlm_lock *);
167
168int gdlm_get_lock(void *, struct lm_lockname *, void **);
169void gdlm_put_lock(void *);
170unsigned int gdlm_lock(void *, unsigned int, unsigned int, unsigned int);
171unsigned int gdlm_unlock(void *, unsigned int);
172void gdlm_cancel(void *);
173int gdlm_hold_lvb(void *, char **);
174void gdlm_unhold_lvb(void *, char *);
175
176/* plock.c */
177
178int gdlm_plock_init(void);
179void gdlm_plock_exit(void);
180int gdlm_plock(void *, struct lm_lockname *, struct file *, int,
181 struct file_lock *);
182int gdlm_plock_get(void *, struct lm_lockname *, struct file *,
183 struct file_lock *);
184int gdlm_punlock(void *, struct lm_lockname *, struct file *,
185 struct file_lock *);
186#endif
187
diff --git a/fs/gfs2/locking/dlm/main.c b/fs/gfs2/locking/dlm/main.c
new file mode 100644
index 000000000000..2194b1d5b5ec
--- /dev/null
+++ b/fs/gfs2/locking/dlm/main.c
@@ -0,0 +1,64 @@
1/*
2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2005 Red Hat, Inc. All rights reserved.
4 *
5 * This copyrighted material is made available to anyone wishing to use,
6 * modify, copy, or redistribute it subject to the terms and conditions
7 * of the GNU General Public License version 2.
8 */
9
10#include <linux/init.h>
11
12#include "lock_dlm.h"
13
14extern int gdlm_drop_count;
15extern int gdlm_drop_period;
16
17extern struct lm_lockops gdlm_ops;
18
19static int __init init_lock_dlm(void)
20{
21 int error;
22
23 error = gfs2_register_lockproto(&gdlm_ops);
24 if (error) {
25 printk(KERN_WARNING "lock_dlm: can't register protocol: %d\n",
26 error);
27 return error;
28 }
29
30 error = gdlm_sysfs_init();
31 if (error) {
32 gfs2_unregister_lockproto(&gdlm_ops);
33 return error;
34 }
35
36 error = gdlm_plock_init();
37 if (error) {
38 gdlm_sysfs_exit();
39 gfs2_unregister_lockproto(&gdlm_ops);
40 return error;
41 }
42
43 gdlm_drop_count = GDLM_DROP_COUNT;
44 gdlm_drop_period = GDLM_DROP_PERIOD;
45
46 printk(KERN_INFO
47 "Lock_DLM (built %s %s) installed\n", __DATE__, __TIME__);
48 return 0;
49}
50
51static void __exit exit_lock_dlm(void)
52{
53 gdlm_plock_exit();
54 gdlm_sysfs_exit();
55 gfs2_unregister_lockproto(&gdlm_ops);
56}
57
58module_init(init_lock_dlm);
59module_exit(exit_lock_dlm);
60
61MODULE_DESCRIPTION("GFS DLM Locking Module");
62MODULE_AUTHOR("Red Hat, Inc.");
63MODULE_LICENSE("GPL");
64
diff --git a/fs/gfs2/locking/dlm/mount.c b/fs/gfs2/locking/dlm/mount.c
new file mode 100644
index 000000000000..1f94dd35a943
--- /dev/null
+++ b/fs/gfs2/locking/dlm/mount.c
@@ -0,0 +1,255 @@
1/*
2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2005 Red Hat, Inc. All rights reserved.
4 *
5 * This copyrighted material is made available to anyone wishing to use,
6 * modify, copy, or redistribute it subject to the terms and conditions
7 * of the GNU General Public License version 2.
8 */
9
10#include "lock_dlm.h"
11
12int gdlm_drop_count;
13int gdlm_drop_period;
14const struct lm_lockops gdlm_ops;
15
16
17static struct gdlm_ls *init_gdlm(lm_callback_t cb, struct gfs2_sbd *sdp,
18 int flags, char *table_name)
19{
20 struct gdlm_ls *ls;
21 char buf[256], *p;
22
23 ls = kzalloc(sizeof(struct gdlm_ls), GFP_KERNEL);
24 if (!ls)
25 return NULL;
26
27 ls->drop_locks_count = gdlm_drop_count;
28 ls->drop_locks_period = gdlm_drop_period;
29 ls->fscb = cb;
30 ls->sdp = sdp;
31 ls->fsflags = flags;
32 spin_lock_init(&ls->async_lock);
33 INIT_LIST_HEAD(&ls->complete);
34 INIT_LIST_HEAD(&ls->blocking);
35 INIT_LIST_HEAD(&ls->delayed);
36 INIT_LIST_HEAD(&ls->submit);
37 INIT_LIST_HEAD(&ls->all_locks);
38 init_waitqueue_head(&ls->thread_wait);
39 init_waitqueue_head(&ls->wait_control);
40 ls->thread1 = NULL;
41 ls->thread2 = NULL;
42 ls->drop_time = jiffies;
43 ls->jid = -1;
44
45 strncpy(buf, table_name, 256);
46 buf[255] = '\0';
47
48 p = strstr(buf, ":");
49 if (!p) {
50 log_info("invalid table_name \"%s\"", table_name);
51 kfree(ls);
52 return NULL;
53 }
54 *p = '\0';
55 p++;
56
57 strncpy(ls->clustername, buf, GDLM_NAME_LEN);
58 strncpy(ls->fsname, p, GDLM_NAME_LEN);
59
60 return ls;
61}
62
63static int make_args(struct gdlm_ls *ls, char *data_arg, int *nodir)
64{
65 char data[256];
66 char *options, *x, *y;
67 int error = 0;
68
69 memset(data, 0, 256);
70 strncpy(data, data_arg, 255);
71
72 for (options = data; (x = strsep(&options, ":")); ) {
73 if (!*x)
74 continue;
75
76 y = strchr(x, '=');
77 if (y)
78 *y++ = 0;
79
80 if (!strcmp(x, "jid")) {
81 if (!y) {
82 log_error("need argument to jid");
83 error = -EINVAL;
84 break;
85 }
86 sscanf(y, "%u", &ls->jid);
87
88 } else if (!strcmp(x, "first")) {
89 if (!y) {
90 log_error("need argument to first");
91 error = -EINVAL;
92 break;
93 }
94 sscanf(y, "%u", &ls->first);
95
96 } else if (!strcmp(x, "id")) {
97 if (!y) {
98 log_error("need argument to id");
99 error = -EINVAL;
100 break;
101 }
102 sscanf(y, "%u", &ls->id);
103
104 } else if (!strcmp(x, "nodir")) {
105 if (!y) {
106 log_error("need argument to nodir");
107 error = -EINVAL;
108 break;
109 }
110 sscanf(y, "%u", nodir);
111
112 } else {
113 log_error("unkonwn option: %s", x);
114 error = -EINVAL;
115 break;
116 }
117 }
118
119 return error;
120}
121
122static int gdlm_mount(char *table_name, char *host_data,
123 lm_callback_t cb, void *cb_data,
124 unsigned int min_lvb_size, int flags,
125 struct lm_lockstruct *lockstruct,
126 struct kobject *fskobj)
127{
128 struct gdlm_ls *ls;
129 int error = -ENOMEM, nodir = 0;
130
131 if (min_lvb_size > GDLM_LVB_SIZE)
132 goto out;
133
134 ls = init_gdlm(cb, cb_data, flags, table_name);
135 if (!ls)
136 goto out;
137
138 error = make_args(ls, host_data, &nodir);
139 if (error)
140 goto out;
141
142 error = gdlm_init_threads(ls);
143 if (error)
144 goto out_free;
145
146 error = gdlm_kobject_setup(ls, fskobj);
147 if (error)
148 goto out_thread;
149
150 error = dlm_new_lockspace(ls->fsname, strlen(ls->fsname),
151 &ls->dlm_lockspace,
152 nodir ? DLM_LSFL_NODIR : 0,
153 GDLM_LVB_SIZE);
154 if (error) {
155 log_error("dlm_new_lockspace error %d", error);
156 goto out_kobj;
157 }
158
159 lockstruct->ls_jid = ls->jid;
160 lockstruct->ls_first = ls->first;
161 lockstruct->ls_lockspace = ls;
162 lockstruct->ls_ops = &gdlm_ops;
163 lockstruct->ls_flags = 0;
164 lockstruct->ls_lvb_size = GDLM_LVB_SIZE;
165 return 0;
166
167out_kobj:
168 gdlm_kobject_release(ls);
169out_thread:
170 gdlm_release_threads(ls);
171out_free:
172 kfree(ls);
173out:
174 return error;
175}
176
177static void gdlm_unmount(void *lockspace)
178{
179 struct gdlm_ls *ls = lockspace;
180 int rv;
181
182 log_debug("unmount flags %lx", ls->flags);
183
184 /* FIXME: serialize unmount and withdraw in case they
185 happen at once. Also, if unmount follows withdraw,
186 wait for withdraw to finish. */
187
188 if (test_bit(DFL_WITHDRAW, &ls->flags))
189 goto out;
190
191 gdlm_kobject_release(ls);
192 dlm_release_lockspace(ls->dlm_lockspace, 2);
193 gdlm_release_threads(ls);
194 rv = gdlm_release_all_locks(ls);
195 if (rv)
196 log_info("gdlm_unmount: %d stray locks freed", rv);
197out:
198 kfree(ls);
199}
200
201static void gdlm_recovery_done(void *lockspace, unsigned int jid,
202 unsigned int message)
203{
204 struct gdlm_ls *ls = lockspace;
205 ls->recover_jid_done = jid;
206 ls->recover_jid_status = message;
207 kobject_uevent(&ls->kobj, KOBJ_CHANGE);
208}
209
210static void gdlm_others_may_mount(void *lockspace)
211{
212 struct gdlm_ls *ls = lockspace;
213 ls->first_done = 1;
214 kobject_uevent(&ls->kobj, KOBJ_CHANGE);
215}
216
217/* Userspace gets the offline uevent, blocks new gfs locks on
218 other mounters, and lets us know (sets WITHDRAW flag). Then,
219 userspace leaves the mount group while we leave the lockspace. */
220
221static void gdlm_withdraw(void *lockspace)
222{
223 struct gdlm_ls *ls = lockspace;
224
225 kobject_uevent(&ls->kobj, KOBJ_OFFLINE);
226
227 wait_event_interruptible(ls->wait_control,
228 test_bit(DFL_WITHDRAW, &ls->flags));
229
230 dlm_release_lockspace(ls->dlm_lockspace, 2);
231 gdlm_release_threads(ls);
232 gdlm_release_all_locks(ls);
233 gdlm_kobject_release(ls);
234}
235
236const struct lm_lockops gdlm_ops = {
237 .lm_proto_name = "lock_dlm",
238 .lm_mount = gdlm_mount,
239 .lm_others_may_mount = gdlm_others_may_mount,
240 .lm_unmount = gdlm_unmount,
241 .lm_withdraw = gdlm_withdraw,
242 .lm_get_lock = gdlm_get_lock,
243 .lm_put_lock = gdlm_put_lock,
244 .lm_lock = gdlm_lock,
245 .lm_unlock = gdlm_unlock,
246 .lm_plock = gdlm_plock,
247 .lm_punlock = gdlm_punlock,
248 .lm_plock_get = gdlm_plock_get,
249 .lm_cancel = gdlm_cancel,
250 .lm_hold_lvb = gdlm_hold_lvb,
251 .lm_unhold_lvb = gdlm_unhold_lvb,
252 .lm_recovery_done = gdlm_recovery_done,
253 .lm_owner = THIS_MODULE,
254};
255
diff --git a/fs/gfs2/locking/dlm/plock.c b/fs/gfs2/locking/dlm/plock.c
new file mode 100644
index 000000000000..7365aec9511b
--- /dev/null
+++ b/fs/gfs2/locking/dlm/plock.c
@@ -0,0 +1,301 @@
1/*
2 * Copyright (C) 2005 Red Hat, Inc. All rights reserved.
3 *
4 * This copyrighted material is made available to anyone wishing to use,
5 * modify, copy, or redistribute it subject to the terms and conditions
6 * of the GNU General Public License version 2.
7 */
8
9#include <linux/miscdevice.h>
10#include <linux/lock_dlm_plock.h>
11
12#include "lock_dlm.h"
13
14
15static spinlock_t ops_lock;
16static struct list_head send_list;
17static struct list_head recv_list;
18static wait_queue_head_t send_wq;
19static wait_queue_head_t recv_wq;
20
21struct plock_op {
22 struct list_head list;
23 int done;
24 struct gdlm_plock_info info;
25};
26
27static inline void set_version(struct gdlm_plock_info *info)
28{
29 info->version[0] = GDLM_PLOCK_VERSION_MAJOR;
30 info->version[1] = GDLM_PLOCK_VERSION_MINOR;
31 info->version[2] = GDLM_PLOCK_VERSION_PATCH;
32}
33
34static int check_version(struct gdlm_plock_info *info)
35{
36 if ((GDLM_PLOCK_VERSION_MAJOR != info->version[0]) ||
37 (GDLM_PLOCK_VERSION_MINOR < info->version[1])) {
38 log_error("plock device version mismatch: "
39 "kernel (%u.%u.%u), user (%u.%u.%u)",
40 GDLM_PLOCK_VERSION_MAJOR,
41 GDLM_PLOCK_VERSION_MINOR,
42 GDLM_PLOCK_VERSION_PATCH,
43 info->version[0],
44 info->version[1],
45 info->version[2]);
46 return -EINVAL;
47 }
48 return 0;
49}
50
51static void send_op(struct plock_op *op)
52{
53 set_version(&op->info);
54 INIT_LIST_HEAD(&op->list);
55 spin_lock(&ops_lock);
56 list_add_tail(&op->list, &send_list);
57 spin_unlock(&ops_lock);
58 wake_up(&send_wq);
59}
60
61int gdlm_plock(void *lockspace, struct lm_lockname *name,
62 struct file *file, int cmd, struct file_lock *fl)
63{
64 struct gdlm_ls *ls = lockspace;
65 struct plock_op *op;
66 int rv;
67
68 op = kzalloc(sizeof(*op), GFP_KERNEL);
69 if (!op)
70 return -ENOMEM;
71
72 op->info.optype = GDLM_PLOCK_OP_LOCK;
73 op->info.pid = fl->fl_pid;
74 op->info.ex = (fl->fl_type == F_WRLCK);
75 op->info.wait = IS_SETLKW(cmd);
76 op->info.fsid = ls->id;
77 op->info.number = name->ln_number;
78 op->info.start = fl->fl_start;
79 op->info.end = fl->fl_end;
80 op->info.owner = (__u64)(long) fl->fl_owner;
81
82 send_op(op);
83 wait_event(recv_wq, (op->done != 0));
84
85 spin_lock(&ops_lock);
86 if (!list_empty(&op->list)) {
87 printk(KERN_INFO "plock op on list\n");
88 list_del(&op->list);
89 }
90 spin_unlock(&ops_lock);
91
92 rv = op->info.rv;
93
94 if (!rv) {
95 if (posix_lock_file_wait(file, fl) < 0)
96 log_error("gdlm_plock: vfs lock error %x,%llx",
97 name->ln_type,
98 (unsigned long long)name->ln_number);
99 }
100
101 kfree(op);
102 return rv;
103}
104
105int gdlm_punlock(void *lockspace, struct lm_lockname *name,
106 struct file *file, struct file_lock *fl)
107{
108 struct gdlm_ls *ls = lockspace;
109 struct plock_op *op;
110 int rv;
111
112 op = kzalloc(sizeof(*op), GFP_KERNEL);
113 if (!op)
114 return -ENOMEM;
115
116 if (posix_lock_file_wait(file, fl) < 0)
117 log_error("gdlm_punlock: vfs unlock error %x,%llx",
118 name->ln_type, (unsigned long long)name->ln_number);
119
120 op->info.optype = GDLM_PLOCK_OP_UNLOCK;
121 op->info.pid = fl->fl_pid;
122 op->info.fsid = ls->id;
123 op->info.number = name->ln_number;
124 op->info.start = fl->fl_start;
125 op->info.end = fl->fl_end;
126 op->info.owner = (__u64)(long) fl->fl_owner;
127
128 send_op(op);
129 wait_event(recv_wq, (op->done != 0));
130
131 spin_lock(&ops_lock);
132 if (!list_empty(&op->list)) {
133 printk(KERN_INFO "punlock op on list\n");
134 list_del(&op->list);
135 }
136 spin_unlock(&ops_lock);
137
138 rv = op->info.rv;
139
140 kfree(op);
141 return rv;
142}
143
144int gdlm_plock_get(void *lockspace, struct lm_lockname *name,
145 struct file *file, struct file_lock *fl)
146{
147 struct gdlm_ls *ls = lockspace;
148 struct plock_op *op;
149 int rv;
150
151 op = kzalloc(sizeof(*op), GFP_KERNEL);
152 if (!op)
153 return -ENOMEM;
154
155 op->info.optype = GDLM_PLOCK_OP_GET;
156 op->info.pid = fl->fl_pid;
157 op->info.ex = (fl->fl_type == F_WRLCK);
158 op->info.fsid = ls->id;
159 op->info.number = name->ln_number;
160 op->info.start = fl->fl_start;
161 op->info.end = fl->fl_end;
162
163 send_op(op);
164 wait_event(recv_wq, (op->done != 0));
165
166 spin_lock(&ops_lock);
167 if (!list_empty(&op->list)) {
168 printk(KERN_INFO "plock_get op on list\n");
169 list_del(&op->list);
170 }
171 spin_unlock(&ops_lock);
172
173 rv = op->info.rv;
174
175 if (rv == 0)
176 fl->fl_type = F_UNLCK;
177 else if (rv > 0) {
178 fl->fl_type = (op->info.ex) ? F_WRLCK : F_RDLCK;
179 fl->fl_pid = op->info.pid;
180 fl->fl_start = op->info.start;
181 fl->fl_end = op->info.end;
182 }
183
184 kfree(op);
185 return rv;
186}
187
188/* a read copies out one plock request from the send list */
189static ssize_t dev_read(struct file *file, char __user *u, size_t count,
190 loff_t *ppos)
191{
192 struct gdlm_plock_info info;
193 struct plock_op *op = NULL;
194
195 if (count < sizeof(info))
196 return -EINVAL;
197
198 spin_lock(&ops_lock);
199 if (!list_empty(&send_list)) {
200 op = list_entry(send_list.next, struct plock_op, list);
201 list_move(&op->list, &recv_list);
202 memcpy(&info, &op->info, sizeof(info));
203 }
204 spin_unlock(&ops_lock);
205
206 if (!op)
207 return -EAGAIN;
208
209 if (copy_to_user(u, &info, sizeof(info)))
210 return -EFAULT;
211 return sizeof(info);
212}
213
214/* a write copies in one plock result that should match a plock_op
215 on the recv list */
216static ssize_t dev_write(struct file *file, const char __user *u, size_t count,
217 loff_t *ppos)
218{
219 struct gdlm_plock_info info;
220 struct plock_op *op;
221 int found = 0;
222
223 if (count != sizeof(info))
224 return -EINVAL;
225
226 if (copy_from_user(&info, u, sizeof(info)))
227 return -EFAULT;
228
229 if (check_version(&info))
230 return -EINVAL;
231
232 spin_lock(&ops_lock);
233 list_for_each_entry(op, &recv_list, list) {
234 if (op->info.fsid == info.fsid && op->info.number == info.number &&
235 op->info.owner == info.owner) {
236 list_del_init(&op->list);
237 found = 1;
238 op->done = 1;
239 memcpy(&op->info, &info, sizeof(info));
240 break;
241 }
242 }
243 spin_unlock(&ops_lock);
244
245 if (found)
246 wake_up(&recv_wq);
247 else
248 printk(KERN_INFO "gdlm dev_write no op %x %llx\n", info.fsid,
249 (unsigned long long)info.number);
250 return count;
251}
252
253static unsigned int dev_poll(struct file *file, poll_table *wait)
254{
255 poll_wait(file, &send_wq, wait);
256
257 spin_lock(&ops_lock);
258 if (!list_empty(&send_list)) {
259 spin_unlock(&ops_lock);
260 return POLLIN | POLLRDNORM;
261 }
262 spin_unlock(&ops_lock);
263 return 0;
264}
265
266static struct file_operations dev_fops = {
267 .read = dev_read,
268 .write = dev_write,
269 .poll = dev_poll,
270 .owner = THIS_MODULE
271};
272
273static struct miscdevice plock_dev_misc = {
274 .minor = MISC_DYNAMIC_MINOR,
275 .name = GDLM_PLOCK_MISC_NAME,
276 .fops = &dev_fops
277};
278
279int gdlm_plock_init(void)
280{
281 int rv;
282
283 spin_lock_init(&ops_lock);
284 INIT_LIST_HEAD(&send_list);
285 INIT_LIST_HEAD(&recv_list);
286 init_waitqueue_head(&send_wq);
287 init_waitqueue_head(&recv_wq);
288
289 rv = misc_register(&plock_dev_misc);
290 if (rv)
291 printk(KERN_INFO "gdlm_plock_init: misc_register failed %d",
292 rv);
293 return rv;
294}
295
296void gdlm_plock_exit(void)
297{
298 if (misc_deregister(&plock_dev_misc) < 0)
299 printk(KERN_INFO "gdlm_plock_exit: misc_deregister failed");
300}
301
diff --git a/fs/gfs2/locking/dlm/sysfs.c b/fs/gfs2/locking/dlm/sysfs.c
new file mode 100644
index 000000000000..29ae06f94944
--- /dev/null
+++ b/fs/gfs2/locking/dlm/sysfs.c
@@ -0,0 +1,226 @@
1/*
2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2005 Red Hat, Inc. All rights reserved.
4 *
5 * This copyrighted material is made available to anyone wishing to use,
6 * modify, copy, or redistribute it subject to the terms and conditions
7 * of the GNU General Public License version 2.
8 */
9
10#include <linux/ctype.h>
11#include <linux/stat.h>
12
13#include "lock_dlm.h"
14
15extern struct lm_lockops gdlm_ops;
16
17static ssize_t proto_name_show(struct gdlm_ls *ls, char *buf)
18{
19 return sprintf(buf, "%s\n", gdlm_ops.lm_proto_name);
20}
21
22static ssize_t block_show(struct gdlm_ls *ls, char *buf)
23{
24 ssize_t ret;
25 int val = 0;
26
27 if (test_bit(DFL_BLOCK_LOCKS, &ls->flags))
28 val = 1;
29 ret = sprintf(buf, "%d\n", val);
30 return ret;
31}
32
33static ssize_t block_store(struct gdlm_ls *ls, const char *buf, size_t len)
34{
35 ssize_t ret = len;
36 int val;
37
38 val = simple_strtol(buf, NULL, 0);
39
40 if (val == 1)
41 set_bit(DFL_BLOCK_LOCKS, &ls->flags);
42 else if (val == 0) {
43 clear_bit(DFL_BLOCK_LOCKS, &ls->flags);
44 gdlm_submit_delayed(ls);
45 } else {
46 ret = -EINVAL;
47 }
48 return ret;
49}
50
51static ssize_t withdraw_show(struct gdlm_ls *ls, char *buf)
52{
53 ssize_t ret;
54 int val = 0;
55
56 if (test_bit(DFL_WITHDRAW, &ls->flags))
57 val = 1;
58 ret = sprintf(buf, "%d\n", val);
59 return ret;
60}
61
62static ssize_t withdraw_store(struct gdlm_ls *ls, const char *buf, size_t len)
63{
64 ssize_t ret = len;
65 int val;
66
67 val = simple_strtol(buf, NULL, 0);
68
69 if (val == 1)
70 set_bit(DFL_WITHDRAW, &ls->flags);
71 else
72 ret = -EINVAL;
73 wake_up(&ls->wait_control);
74 return ret;
75}
76
77static ssize_t id_show(struct gdlm_ls *ls, char *buf)
78{
79 return sprintf(buf, "%u\n", ls->id);
80}
81
82static ssize_t jid_show(struct gdlm_ls *ls, char *buf)
83{
84 return sprintf(buf, "%d\n", ls->jid);
85}
86
87static ssize_t first_show(struct gdlm_ls *ls, char *buf)
88{
89 return sprintf(buf, "%d\n", ls->first);
90}
91
92static ssize_t first_done_show(struct gdlm_ls *ls, char *buf)
93{
94 return sprintf(buf, "%d\n", ls->first_done);
95}
96
97static ssize_t recover_show(struct gdlm_ls *ls, char *buf)
98{
99 return sprintf(buf, "%d\n", ls->recover_jid);
100}
101
102static ssize_t recover_store(struct gdlm_ls *ls, const char *buf, size_t len)
103{
104 ls->recover_jid = simple_strtol(buf, NULL, 0);
105 ls->fscb(ls->sdp, LM_CB_NEED_RECOVERY, &ls->recover_jid);
106 return len;
107}
108
109static ssize_t recover_done_show(struct gdlm_ls *ls, char *buf)
110{
111 return sprintf(buf, "%d\n", ls->recover_jid_done);
112}
113
114static ssize_t recover_status_show(struct gdlm_ls *ls, char *buf)
115{
116 return sprintf(buf, "%d\n", ls->recover_jid_status);
117}
118
119struct gdlm_attr {
120 struct attribute attr;
121 ssize_t (*show)(struct gdlm_ls *, char *);
122 ssize_t (*store)(struct gdlm_ls *, const char *, size_t);
123};
124
125#define GDLM_ATTR(_name,_mode,_show,_store) \
126static struct gdlm_attr gdlm_attr_##_name = __ATTR(_name,_mode,_show,_store)
127
128GDLM_ATTR(proto_name, 0444, proto_name_show, NULL);
129GDLM_ATTR(block, 0644, block_show, block_store);
130GDLM_ATTR(withdraw, 0644, withdraw_show, withdraw_store);
131GDLM_ATTR(id, 0444, id_show, NULL);
132GDLM_ATTR(jid, 0444, jid_show, NULL);
133GDLM_ATTR(first, 0444, first_show, NULL);
134GDLM_ATTR(first_done, 0444, first_done_show, NULL);
135GDLM_ATTR(recover, 0644, recover_show, recover_store);
136GDLM_ATTR(recover_done, 0444, recover_done_show, NULL);
137GDLM_ATTR(recover_status, 0444, recover_status_show, NULL);
138
139static struct attribute *gdlm_attrs[] = {
140 &gdlm_attr_proto_name.attr,
141 &gdlm_attr_block.attr,
142 &gdlm_attr_withdraw.attr,
143 &gdlm_attr_id.attr,
144 &gdlm_attr_jid.attr,
145 &gdlm_attr_first.attr,
146 &gdlm_attr_first_done.attr,
147 &gdlm_attr_recover.attr,
148 &gdlm_attr_recover_done.attr,
149 &gdlm_attr_recover_status.attr,
150 NULL,
151};
152
153static ssize_t gdlm_attr_show(struct kobject *kobj, struct attribute *attr,
154 char *buf)
155{
156 struct gdlm_ls *ls = container_of(kobj, struct gdlm_ls, kobj);
157 struct gdlm_attr *a = container_of(attr, struct gdlm_attr, attr);
158 return a->show ? a->show(ls, buf) : 0;
159}
160
161static ssize_t gdlm_attr_store(struct kobject *kobj, struct attribute *attr,
162 const char *buf, size_t len)
163{
164 struct gdlm_ls *ls = container_of(kobj, struct gdlm_ls, kobj);
165 struct gdlm_attr *a = container_of(attr, struct gdlm_attr, attr);
166 return a->store ? a->store(ls, buf, len) : len;
167}
168
169static struct sysfs_ops gdlm_attr_ops = {
170 .show = gdlm_attr_show,
171 .store = gdlm_attr_store,
172};
173
174static struct kobj_type gdlm_ktype = {
175 .default_attrs = gdlm_attrs,
176 .sysfs_ops = &gdlm_attr_ops,
177};
178
179static struct kset gdlm_kset = {
180 .subsys = &kernel_subsys,
181 .kobj = {.name = "lock_dlm",},
182 .ktype = &gdlm_ktype,
183};
184
185int gdlm_kobject_setup(struct gdlm_ls *ls, struct kobject *fskobj)
186{
187 int error;
188
189 error = kobject_set_name(&ls->kobj, "%s", "lock_module");
190 if (error) {
191 log_error("can't set kobj name %d", error);
192 return error;
193 }
194
195 ls->kobj.kset = &gdlm_kset;
196 ls->kobj.ktype = &gdlm_ktype;
197 ls->kobj.parent = fskobj;
198
199 error = kobject_register(&ls->kobj);
200 if (error)
201 log_error("can't register kobj %d", error);
202
203 return error;
204}
205
206void gdlm_kobject_release(struct gdlm_ls *ls)
207{
208 kobject_unregister(&ls->kobj);
209}
210
211int gdlm_sysfs_init(void)
212{
213 int error;
214
215 error = kset_register(&gdlm_kset);
216 if (error)
217 printk("lock_dlm: cannot register kset %d\n", error);
218
219 return error;
220}
221
222void gdlm_sysfs_exit(void)
223{
224 kset_unregister(&gdlm_kset);
225}
226
diff --git a/fs/gfs2/locking/dlm/thread.c b/fs/gfs2/locking/dlm/thread.c
new file mode 100644
index 000000000000..9cf1f168eaf8
--- /dev/null
+++ b/fs/gfs2/locking/dlm/thread.c
@@ -0,0 +1,359 @@
1/*
2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2005 Red Hat, Inc. All rights reserved.
4 *
5 * This copyrighted material is made available to anyone wishing to use,
6 * modify, copy, or redistribute it subject to the terms and conditions
7 * of the GNU General Public License version 2.
8 */
9
10#include "lock_dlm.h"
11
12/* A lock placed on this queue is re-submitted to DLM as soon as the lock_dlm
13 thread gets to it. */
14
15static void queue_submit(struct gdlm_lock *lp)
16{
17 struct gdlm_ls *ls = lp->ls;
18
19 spin_lock(&ls->async_lock);
20 list_add_tail(&lp->delay_list, &ls->submit);
21 spin_unlock(&ls->async_lock);
22 wake_up(&ls->thread_wait);
23}
24
25static void process_blocking(struct gdlm_lock *lp, int bast_mode)
26{
27 struct gdlm_ls *ls = lp->ls;
28 unsigned int cb = 0;
29
30 switch (gdlm_make_lmstate(bast_mode)) {
31 case LM_ST_EXCLUSIVE:
32 cb = LM_CB_NEED_E;
33 break;
34 case LM_ST_DEFERRED:
35 cb = LM_CB_NEED_D;
36 break;
37 case LM_ST_SHARED:
38 cb = LM_CB_NEED_S;
39 break;
40 default:
41 gdlm_assert(0, "unknown bast mode %u", lp->bast_mode);
42 }
43
44 ls->fscb(ls->sdp, cb, &lp->lockname);
45}
46
47static void process_complete(struct gdlm_lock *lp)
48{
49 struct gdlm_ls *ls = lp->ls;
50 struct lm_async_cb acb;
51 s16 prev_mode = lp->cur;
52
53 memset(&acb, 0, sizeof(acb));
54
55 if (lp->lksb.sb_status == -DLM_ECANCEL) {
56 log_info("complete dlm cancel %x,%llx flags %lx",
57 lp->lockname.ln_type,
58 (unsigned long long)lp->lockname.ln_number,
59 lp->flags);
60
61 lp->req = lp->cur;
62 acb.lc_ret |= LM_OUT_CANCELED;
63 if (lp->cur == DLM_LOCK_IV)
64 lp->lksb.sb_lkid = 0;
65 goto out;
66 }
67
68 if (test_and_clear_bit(LFL_DLM_UNLOCK, &lp->flags)) {
69 if (lp->lksb.sb_status != -DLM_EUNLOCK) {
70 log_info("unlock sb_status %d %x,%llx flags %lx",
71 lp->lksb.sb_status, lp->lockname.ln_type,
72 (unsigned long long)lp->lockname.ln_number,
73 lp->flags);
74 return;
75 }
76
77 lp->cur = DLM_LOCK_IV;
78 lp->req = DLM_LOCK_IV;
79 lp->lksb.sb_lkid = 0;
80
81 if (test_and_clear_bit(LFL_UNLOCK_DELETE, &lp->flags)) {
82 gdlm_delete_lp(lp);
83 return;
84 }
85 goto out;
86 }
87
88 if (lp->lksb.sb_flags & DLM_SBF_VALNOTVALID)
89 memset(lp->lksb.sb_lvbptr, 0, GDLM_LVB_SIZE);
90
91 if (lp->lksb.sb_flags & DLM_SBF_ALTMODE) {
92 if (lp->req == DLM_LOCK_PR)
93 lp->req = DLM_LOCK_CW;
94 else if (lp->req == DLM_LOCK_CW)
95 lp->req = DLM_LOCK_PR;
96 }
97
98 /*
99 * A canceled lock request. The lock was just taken off the delayed
100 * list and was never even submitted to dlm.
101 */
102
103 if (test_and_clear_bit(LFL_CANCEL, &lp->flags)) {
104 log_info("complete internal cancel %x,%llx",
105 lp->lockname.ln_type,
106 (unsigned long long)lp->lockname.ln_number);
107 lp->req = lp->cur;
108 acb.lc_ret |= LM_OUT_CANCELED;
109 goto out;
110 }
111
112 /*
113 * An error occured.
114 */
115
116 if (lp->lksb.sb_status) {
117 /* a "normal" error */
118 if ((lp->lksb.sb_status == -EAGAIN) &&
119 (lp->lkf & DLM_LKF_NOQUEUE)) {
120 lp->req = lp->cur;
121 if (lp->cur == DLM_LOCK_IV)
122 lp->lksb.sb_lkid = 0;
123 goto out;
124 }
125
126 /* this could only happen with cancels I think */
127 log_info("ast sb_status %d %x,%llx flags %lx",
128 lp->lksb.sb_status, lp->lockname.ln_type,
129 (unsigned long long)lp->lockname.ln_number,
130 lp->flags);
131 return;
132 }
133
134 /*
135 * This is an AST for an EX->EX conversion for sync_lvb from GFS.
136 */
137
138 if (test_and_clear_bit(LFL_SYNC_LVB, &lp->flags)) {
139 complete(&lp->ast_wait);
140 return;
141 }
142
143 /*
144 * A lock has been demoted to NL because it initially completed during
145 * BLOCK_LOCKS. Now it must be requested in the originally requested
146 * mode.
147 */
148
149 if (test_and_clear_bit(LFL_REREQUEST, &lp->flags)) {
150 gdlm_assert(lp->req == DLM_LOCK_NL, "%x,%llx",
151 lp->lockname.ln_type,
152 (unsigned long long)lp->lockname.ln_number);
153 gdlm_assert(lp->prev_req > DLM_LOCK_NL, "%x,%llx",
154 lp->lockname.ln_type,
155 (unsigned long long)lp->lockname.ln_number);
156
157 lp->cur = DLM_LOCK_NL;
158 lp->req = lp->prev_req;
159 lp->prev_req = DLM_LOCK_IV;
160 lp->lkf &= ~DLM_LKF_CONVDEADLK;
161
162 set_bit(LFL_NOCACHE, &lp->flags);
163
164 if (test_bit(DFL_BLOCK_LOCKS, &ls->flags) &&
165 !test_bit(LFL_NOBLOCK, &lp->flags))
166 gdlm_queue_delayed(lp);
167 else
168 queue_submit(lp);
169 return;
170 }
171
172 /*
173 * A request is granted during dlm recovery. It may be granted
174 * because the locks of a failed node were cleared. In that case,
175 * there may be inconsistent data beneath this lock and we must wait
176 * for recovery to complete to use it. When gfs recovery is done this
177 * granted lock will be converted to NL and then reacquired in this
178 * granted state.
179 */
180
181 if (test_bit(DFL_BLOCK_LOCKS, &ls->flags) &&
182 !test_bit(LFL_NOBLOCK, &lp->flags) &&
183 lp->req != DLM_LOCK_NL) {
184
185 lp->cur = lp->req;
186 lp->prev_req = lp->req;
187 lp->req = DLM_LOCK_NL;
188 lp->lkf |= DLM_LKF_CONVERT;
189 lp->lkf &= ~DLM_LKF_CONVDEADLK;
190
191 log_debug("rereq %x,%llx id %x %d,%d",
192 lp->lockname.ln_type,
193 (unsigned long long)lp->lockname.ln_number,
194 lp->lksb.sb_lkid, lp->cur, lp->req);
195
196 set_bit(LFL_REREQUEST, &lp->flags);
197 queue_submit(lp);
198 return;
199 }
200
201 /*
202 * DLM demoted the lock to NL before it was granted so GFS must be
203 * told it cannot cache data for this lock.
204 */
205
206 if (lp->lksb.sb_flags & DLM_SBF_DEMOTED)
207 set_bit(LFL_NOCACHE, &lp->flags);
208
209out:
210 /*
211 * This is an internal lock_dlm lock
212 */
213
214 if (test_bit(LFL_INLOCK, &lp->flags)) {
215 clear_bit(LFL_NOBLOCK, &lp->flags);
216 lp->cur = lp->req;
217 complete(&lp->ast_wait);
218 return;
219 }
220
221 /*
222 * Normal completion of a lock request. Tell GFS it now has the lock.
223 */
224
225 clear_bit(LFL_NOBLOCK, &lp->flags);
226 lp->cur = lp->req;
227
228 acb.lc_name = lp->lockname;
229 acb.lc_ret |= gdlm_make_lmstate(lp->cur);
230
231 if (!test_and_clear_bit(LFL_NOCACHE, &lp->flags) &&
232 (lp->cur > DLM_LOCK_NL) && (prev_mode > DLM_LOCK_NL))
233 acb.lc_ret |= LM_OUT_CACHEABLE;
234
235 ls->fscb(ls->sdp, LM_CB_ASYNC, &acb);
236}
237
238static inline int no_work(struct gdlm_ls *ls, int blocking)
239{
240 int ret;
241
242 spin_lock(&ls->async_lock);
243 ret = list_empty(&ls->complete) && list_empty(&ls->submit);
244 if (ret && blocking)
245 ret = list_empty(&ls->blocking);
246 spin_unlock(&ls->async_lock);
247
248 return ret;
249}
250
251static inline int check_drop(struct gdlm_ls *ls)
252{
253 if (!ls->drop_locks_count)
254 return 0;
255
256 if (time_after(jiffies, ls->drop_time + ls->drop_locks_period * HZ)) {
257 ls->drop_time = jiffies;
258 if (ls->all_locks_count >= ls->drop_locks_count)
259 return 1;
260 }
261 return 0;
262}
263
264static int gdlm_thread(void *data)
265{
266 struct gdlm_ls *ls = (struct gdlm_ls *) data;
267 struct gdlm_lock *lp = NULL;
268 int blist = 0;
269 uint8_t complete, blocking, submit, drop;
270 DECLARE_WAITQUEUE(wait, current);
271
272 /* Only thread1 is allowed to do blocking callbacks since gfs
273 may wait for a completion callback within a blocking cb. */
274
275 if (current == ls->thread1)
276 blist = 1;
277
278 while (!kthread_should_stop()) {
279 set_current_state(TASK_INTERRUPTIBLE);
280 add_wait_queue(&ls->thread_wait, &wait);
281 if (no_work(ls, blist))
282 schedule();
283 remove_wait_queue(&ls->thread_wait, &wait);
284 set_current_state(TASK_RUNNING);
285
286 complete = blocking = submit = drop = 0;
287
288 spin_lock(&ls->async_lock);
289
290 if (blist && !list_empty(&ls->blocking)) {
291 lp = list_entry(ls->blocking.next, struct gdlm_lock,
292 blist);
293 list_del_init(&lp->blist);
294 blocking = lp->bast_mode;
295 lp->bast_mode = 0;
296 } else if (!list_empty(&ls->complete)) {
297 lp = list_entry(ls->complete.next, struct gdlm_lock,
298 clist);
299 list_del_init(&lp->clist);
300 complete = 1;
301 } else if (!list_empty(&ls->submit)) {
302 lp = list_entry(ls->submit.next, struct gdlm_lock,
303 delay_list);
304 list_del_init(&lp->delay_list);
305 submit = 1;
306 }
307
308 drop = check_drop(ls);
309 spin_unlock(&ls->async_lock);
310
311 if (complete)
312 process_complete(lp);
313
314 else if (blocking)
315 process_blocking(lp, blocking);
316
317 else if (submit)
318 gdlm_do_lock(lp);
319
320 if (drop)
321 ls->fscb(ls->sdp, LM_CB_DROPLOCKS, NULL);
322
323 schedule();
324 }
325
326 return 0;
327}
328
329int gdlm_init_threads(struct gdlm_ls *ls)
330{
331 struct task_struct *p;
332 int error;
333
334 p = kthread_run(gdlm_thread, ls, "lock_dlm1");
335 error = IS_ERR(p);
336 if (error) {
337 log_error("can't start lock_dlm1 thread %d", error);
338 return error;
339 }
340 ls->thread1 = p;
341
342 p = kthread_run(gdlm_thread, ls, "lock_dlm2");
343 error = IS_ERR(p);
344 if (error) {
345 log_error("can't start lock_dlm2 thread %d", error);
346 kthread_stop(ls->thread1);
347 return error;
348 }
349 ls->thread2 = p;
350
351 return 0;
352}
353
354void gdlm_release_threads(struct gdlm_ls *ls)
355{
356 kthread_stop(ls->thread1);
357 kthread_stop(ls->thread2);
358}
359