aboutsummaryrefslogtreecommitdiffstats
path: root/fs/gfs2/locking/dlm/lock.c
diff options
context:
space:
mode:
Diffstat (limited to 'fs/gfs2/locking/dlm/lock.c')
-rw-r--r--fs/gfs2/locking/dlm/lock.c368
1 files changed, 274 insertions, 94 deletions
diff --git a/fs/gfs2/locking/dlm/lock.c b/fs/gfs2/locking/dlm/lock.c
index cf7ea8abec87..2482c9047505 100644
--- a/fs/gfs2/locking/dlm/lock.c
+++ b/fs/gfs2/locking/dlm/lock.c
@@ -11,46 +11,60 @@
11 11
12static char junk_lvb[GDLM_LVB_SIZE]; 12static char junk_lvb[GDLM_LVB_SIZE];
13 13
14static void queue_complete(struct gdlm_lock *lp) 14
15/* convert dlm lock-mode to gfs lock-state */
16
17static s16 gdlm_make_lmstate(s16 dlmmode)
15{ 18{
16 struct gdlm_ls *ls = lp->ls; 19 switch (dlmmode) {
20 case DLM_LOCK_IV:
21 case DLM_LOCK_NL:
22 return LM_ST_UNLOCKED;
23 case DLM_LOCK_EX:
24 return LM_ST_EXCLUSIVE;
25 case DLM_LOCK_CW:
26 return LM_ST_DEFERRED;
27 case DLM_LOCK_PR:
28 return LM_ST_SHARED;
29 }
30 gdlm_assert(0, "unknown DLM mode %d", dlmmode);
31 return -1;
32}
17 33
18 clear_bit(LFL_ACTIVE, &lp->flags); 34/* A lock placed on this queue is re-submitted to DLM as soon as the lock_dlm
35 thread gets to it. */
36
37static void queue_submit(struct gdlm_lock *lp)
38{
39 struct gdlm_ls *ls = lp->ls;
19 40
20 spin_lock(&ls->async_lock); 41 spin_lock(&ls->async_lock);
21 list_add_tail(&lp->clist, &ls->complete); 42 list_add_tail(&lp->delay_list, &ls->submit);
22 spin_unlock(&ls->async_lock); 43 spin_unlock(&ls->async_lock);
23 wake_up(&ls->thread_wait); 44 wake_up(&ls->thread_wait);
24} 45}
25 46
26static inline void gdlm_ast(void *astarg) 47static void wake_up_ast(struct gdlm_lock *lp)
27{ 48{
28 queue_complete(astarg); 49 clear_bit(LFL_AST_WAIT, &lp->flags);
50 smp_mb__after_clear_bit();
51 wake_up_bit(&lp->flags, LFL_AST_WAIT);
29} 52}
30 53
31static inline void gdlm_bast(void *astarg, int mode) 54static void gdlm_delete_lp(struct gdlm_lock *lp)
32{ 55{
33 struct gdlm_lock *lp = astarg;
34 struct gdlm_ls *ls = lp->ls; 56 struct gdlm_ls *ls = lp->ls;
35 57
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); 58 spin_lock(&ls->async_lock);
44 if (!lp->bast_mode) { 59 if (!list_empty(&lp->delay_list))
45 list_add_tail(&lp->blist, &ls->blocking); 60 list_del_init(&lp->delay_list);
46 lp->bast_mode = mode; 61 ls->all_locks_count--;
47 } else if (lp->bast_mode < mode)
48 lp->bast_mode = mode;
49 spin_unlock(&ls->async_lock); 62 spin_unlock(&ls->async_lock);
50 wake_up(&ls->thread_wait); 63
64 kfree(lp);
51} 65}
52 66
53void gdlm_queue_delayed(struct gdlm_lock *lp) 67static void gdlm_queue_delayed(struct gdlm_lock *lp)
54{ 68{
55 struct gdlm_ls *ls = lp->ls; 69 struct gdlm_ls *ls = lp->ls;
56 70
@@ -59,6 +73,236 @@ void gdlm_queue_delayed(struct gdlm_lock *lp)
59 spin_unlock(&ls->async_lock); 73 spin_unlock(&ls->async_lock);
60} 74}
61 75
76static void process_complete(struct gdlm_lock *lp)
77{
78 struct gdlm_ls *ls = lp->ls;
79 struct lm_async_cb acb;
80
81 memset(&acb, 0, sizeof(acb));
82
83 if (lp->lksb.sb_status == -DLM_ECANCEL) {
84 log_info("complete dlm cancel %x,%llx flags %lx",
85 lp->lockname.ln_type,
86 (unsigned long long)lp->lockname.ln_number,
87 lp->flags);
88
89 lp->req = lp->cur;
90 acb.lc_ret |= LM_OUT_CANCELED;
91 if (lp->cur == DLM_LOCK_IV)
92 lp->lksb.sb_lkid = 0;
93 goto out;
94 }
95
96 if (test_and_clear_bit(LFL_DLM_UNLOCK, &lp->flags)) {
97 if (lp->lksb.sb_status != -DLM_EUNLOCK) {
98 log_info("unlock sb_status %d %x,%llx flags %lx",
99 lp->lksb.sb_status, lp->lockname.ln_type,
100 (unsigned long long)lp->lockname.ln_number,
101 lp->flags);
102 return;
103 }
104
105 lp->cur = DLM_LOCK_IV;
106 lp->req = DLM_LOCK_IV;
107 lp->lksb.sb_lkid = 0;
108
109 if (test_and_clear_bit(LFL_UNLOCK_DELETE, &lp->flags)) {
110 gdlm_delete_lp(lp);
111 return;
112 }
113 goto out;
114 }
115
116 if (lp->lksb.sb_flags & DLM_SBF_VALNOTVALID)
117 memset(lp->lksb.sb_lvbptr, 0, GDLM_LVB_SIZE);
118
119 if (lp->lksb.sb_flags & DLM_SBF_ALTMODE) {
120 if (lp->req == DLM_LOCK_PR)
121 lp->req = DLM_LOCK_CW;
122 else if (lp->req == DLM_LOCK_CW)
123 lp->req = DLM_LOCK_PR;
124 }
125
126 /*
127 * A canceled lock request. The lock was just taken off the delayed
128 * list and was never even submitted to dlm.
129 */
130
131 if (test_and_clear_bit(LFL_CANCEL, &lp->flags)) {
132 log_info("complete internal cancel %x,%llx",
133 lp->lockname.ln_type,
134 (unsigned long long)lp->lockname.ln_number);
135 lp->req = lp->cur;
136 acb.lc_ret |= LM_OUT_CANCELED;
137 goto out;
138 }
139
140 /*
141 * An error occured.
142 */
143
144 if (lp->lksb.sb_status) {
145 /* a "normal" error */
146 if ((lp->lksb.sb_status == -EAGAIN) &&
147 (lp->lkf & DLM_LKF_NOQUEUE)) {
148 lp->req = lp->cur;
149 if (lp->cur == DLM_LOCK_IV)
150 lp->lksb.sb_lkid = 0;
151 goto out;
152 }
153
154 /* this could only happen with cancels I think */
155 log_info("ast sb_status %d %x,%llx flags %lx",
156 lp->lksb.sb_status, lp->lockname.ln_type,
157 (unsigned long long)lp->lockname.ln_number,
158 lp->flags);
159 return;
160 }
161
162 /*
163 * This is an AST for an EX->EX conversion for sync_lvb from GFS.
164 */
165
166 if (test_and_clear_bit(LFL_SYNC_LVB, &lp->flags)) {
167 wake_up_ast(lp);
168 return;
169 }
170
171 /*
172 * A lock has been demoted to NL because it initially completed during
173 * BLOCK_LOCKS. Now it must be requested in the originally requested
174 * mode.
175 */
176
177 if (test_and_clear_bit(LFL_REREQUEST, &lp->flags)) {
178 gdlm_assert(lp->req == DLM_LOCK_NL, "%x,%llx",
179 lp->lockname.ln_type,
180 (unsigned long long)lp->lockname.ln_number);
181 gdlm_assert(lp->prev_req > DLM_LOCK_NL, "%x,%llx",
182 lp->lockname.ln_type,
183 (unsigned long long)lp->lockname.ln_number);
184
185 lp->cur = DLM_LOCK_NL;
186 lp->req = lp->prev_req;
187 lp->prev_req = DLM_LOCK_IV;
188 lp->lkf &= ~DLM_LKF_CONVDEADLK;
189
190 set_bit(LFL_NOCACHE, &lp->flags);
191
192 if (test_bit(DFL_BLOCK_LOCKS, &ls->flags) &&
193 !test_bit(LFL_NOBLOCK, &lp->flags))
194 gdlm_queue_delayed(lp);
195 else
196 queue_submit(lp);
197 return;
198 }
199
200 /*
201 * A request is granted during dlm recovery. It may be granted
202 * because the locks of a failed node were cleared. In that case,
203 * there may be inconsistent data beneath this lock and we must wait
204 * for recovery to complete to use it. When gfs recovery is done this
205 * granted lock will be converted to NL and then reacquired in this
206 * granted state.
207 */
208
209 if (test_bit(DFL_BLOCK_LOCKS, &ls->flags) &&
210 !test_bit(LFL_NOBLOCK, &lp->flags) &&
211 lp->req != DLM_LOCK_NL) {
212
213 lp->cur = lp->req;
214 lp->prev_req = lp->req;
215 lp->req = DLM_LOCK_NL;
216 lp->lkf |= DLM_LKF_CONVERT;
217 lp->lkf &= ~DLM_LKF_CONVDEADLK;
218
219 log_debug("rereq %x,%llx id %x %d,%d",
220 lp->lockname.ln_type,
221 (unsigned long long)lp->lockname.ln_number,
222 lp->lksb.sb_lkid, lp->cur, lp->req);
223
224 set_bit(LFL_REREQUEST, &lp->flags);
225 queue_submit(lp);
226 return;
227 }
228
229 /*
230 * DLM demoted the lock to NL before it was granted so GFS must be
231 * told it cannot cache data for this lock.
232 */
233
234 if (lp->lksb.sb_flags & DLM_SBF_DEMOTED)
235 set_bit(LFL_NOCACHE, &lp->flags);
236
237out:
238 /*
239 * This is an internal lock_dlm lock
240 */
241
242 if (test_bit(LFL_INLOCK, &lp->flags)) {
243 clear_bit(LFL_NOBLOCK, &lp->flags);
244 lp->cur = lp->req;
245 wake_up_ast(lp);
246 return;
247 }
248
249 /*
250 * Normal completion of a lock request. Tell GFS it now has the lock.
251 */
252
253 clear_bit(LFL_NOBLOCK, &lp->flags);
254 lp->cur = lp->req;
255
256 acb.lc_name = lp->lockname;
257 acb.lc_ret |= gdlm_make_lmstate(lp->cur);
258
259 ls->fscb(ls->sdp, LM_CB_ASYNC, &acb);
260}
261
262static void gdlm_ast(void *astarg)
263{
264 struct gdlm_lock *lp = astarg;
265 clear_bit(LFL_ACTIVE, &lp->flags);
266 process_complete(lp);
267}
268
269static void process_blocking(struct gdlm_lock *lp, int bast_mode)
270{
271 struct gdlm_ls *ls = lp->ls;
272 unsigned int cb = 0;
273
274 switch (gdlm_make_lmstate(bast_mode)) {
275 case LM_ST_EXCLUSIVE:
276 cb = LM_CB_NEED_E;
277 break;
278 case LM_ST_DEFERRED:
279 cb = LM_CB_NEED_D;
280 break;
281 case LM_ST_SHARED:
282 cb = LM_CB_NEED_S;
283 break;
284 default:
285 gdlm_assert(0, "unknown bast mode %u", bast_mode);
286 }
287
288 ls->fscb(ls->sdp, cb, &lp->lockname);
289}
290
291
292static void gdlm_bast(void *astarg, int mode)
293{
294 struct gdlm_lock *lp = astarg;
295
296 if (!mode) {
297 printk(KERN_INFO "lock_dlm: bast mode zero %x,%llx\n",
298 lp->lockname.ln_type,
299 (unsigned long long)lp->lockname.ln_number);
300 return;
301 }
302
303 process_blocking(lp, mode);
304}
305
62/* convert gfs lock-state to dlm lock-mode */ 306/* convert gfs lock-state to dlm lock-mode */
63 307
64static s16 make_mode(s16 lmstate) 308static s16 make_mode(s16 lmstate)
@@ -77,24 +321,6 @@ static s16 make_mode(s16 lmstate)
77 return -1; 321 return -1;
78} 322}
79 323
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}
98 324
99/* verify agreement with GFS on the current lock state, NB: DLM_LOCK_NL and 325/* 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. */ 326 DLM_LOCK_IV are both considered LM_ST_UNLOCKED by GFS. */
@@ -134,14 +360,6 @@ static inline unsigned int make_flags(struct gdlm_lock *lp,
134 360
135 if (lp->lksb.sb_lkid != 0) { 361 if (lp->lksb.sb_lkid != 0) {
136 lkf |= DLM_LKF_CONVERT; 362 lkf |= DLM_LKF_CONVERT;
137
138 /* Conversion deadlock avoidance by DLM */
139
140 if (!(lp->ls->fsflags & LM_MFLAG_CONV_NODROP) &&
141 !test_bit(LFL_FORCE_PROMOTE, &lp->flags) &&
142 !(lkf & DLM_LKF_NOQUEUE) &&
143 cur > DLM_LOCK_NL && req > DLM_LOCK_NL && cur != req)
144 lkf |= DLM_LKF_CONVDEADLK;
145 } 363 }
146 364
147 if (lp->lvb) 365 if (lp->lvb)
@@ -173,14 +391,9 @@ static int gdlm_create_lp(struct gdlm_ls *ls, struct lm_lockname *name,
173 make_strname(name, &lp->strname); 391 make_strname(name, &lp->strname);
174 lp->ls = ls; 392 lp->ls = ls;
175 lp->cur = DLM_LOCK_IV; 393 lp->cur = DLM_LOCK_IV;
176 lp->lvb = NULL;
177 lp->hold_null = NULL;
178 INIT_LIST_HEAD(&lp->clist);
179 INIT_LIST_HEAD(&lp->blist);
180 INIT_LIST_HEAD(&lp->delay_list); 394 INIT_LIST_HEAD(&lp->delay_list);
181 395
182 spin_lock(&ls->async_lock); 396 spin_lock(&ls->async_lock);
183 list_add(&lp->all_list, &ls->all_locks);
184 ls->all_locks_count++; 397 ls->all_locks_count++;
185 spin_unlock(&ls->async_lock); 398 spin_unlock(&ls->async_lock);
186 399
@@ -188,26 +401,6 @@ static int gdlm_create_lp(struct gdlm_ls *ls, struct lm_lockname *name,
188 return 0; 401 return 0;
189} 402}
190 403
191void gdlm_delete_lp(struct gdlm_lock *lp)
192{
193 struct gdlm_ls *ls = lp->ls;
194
195 spin_lock(&ls->async_lock);
196 if (!list_empty(&lp->clist))
197 list_del_init(&lp->clist);
198 if (!list_empty(&lp->blist))
199 list_del_init(&lp->blist);
200 if (!list_empty(&lp->delay_list))
201 list_del_init(&lp->delay_list);
202 gdlm_assert(!list_empty(&lp->all_list), "%x,%llx", lp->lockname.ln_type,
203 (unsigned long long)lp->lockname.ln_number);
204 list_del_init(&lp->all_list);
205 ls->all_locks_count--;
206 spin_unlock(&ls->async_lock);
207
208 kfree(lp);
209}
210
211int gdlm_get_lock(void *lockspace, struct lm_lockname *name, 404int gdlm_get_lock(void *lockspace, struct lm_lockname *name,
212 void **lockp) 405 void **lockp)
213{ 406{
@@ -261,7 +454,7 @@ unsigned int gdlm_do_lock(struct gdlm_lock *lp)
261 454
262 if ((error == -EAGAIN) && (lp->lkf & DLM_LKF_NOQUEUE)) { 455 if ((error == -EAGAIN) && (lp->lkf & DLM_LKF_NOQUEUE)) {
263 lp->lksb.sb_status = -EAGAIN; 456 lp->lksb.sb_status = -EAGAIN;
264 queue_complete(lp); 457 gdlm_ast(lp);
265 error = 0; 458 error = 0;
266 } 459 }
267 460
@@ -308,6 +501,12 @@ unsigned int gdlm_lock(void *lock, unsigned int cur_state,
308{ 501{
309 struct gdlm_lock *lp = lock; 502 struct gdlm_lock *lp = lock;
310 503
504 if (req_state == LM_ST_UNLOCKED)
505 return gdlm_unlock(lock, cur_state);
506
507 if (req_state == LM_ST_UNLOCKED)
508 return gdlm_unlock(lock, cur_state);
509
311 clear_bit(LFL_DLM_CANCEL, &lp->flags); 510 clear_bit(LFL_DLM_CANCEL, &lp->flags);
312 if (flags & LM_FLAG_NOEXP) 511 if (flags & LM_FLAG_NOEXP)
313 set_bit(LFL_NOBLOCK, &lp->flags); 512 set_bit(LFL_NOBLOCK, &lp->flags);
@@ -351,7 +550,7 @@ void gdlm_cancel(void *lock)
351 if (delay_list) { 550 if (delay_list) {
352 set_bit(LFL_CANCEL, &lp->flags); 551 set_bit(LFL_CANCEL, &lp->flags);
353 set_bit(LFL_ACTIVE, &lp->flags); 552 set_bit(LFL_ACTIVE, &lp->flags);
354 queue_complete(lp); 553 gdlm_ast(lp);
355 return; 554 return;
356 } 555 }
357 556
@@ -507,22 +706,3 @@ void gdlm_submit_delayed(struct gdlm_ls *ls)
507 wake_up(&ls->thread_wait); 706 wake_up(&ls->thread_wait);
508} 707}
509 708
510int gdlm_release_all_locks(struct gdlm_ls *ls)
511{
512 struct gdlm_lock *lp, *safe;
513 int count = 0;
514
515 spin_lock(&ls->async_lock);
516 list_for_each_entry_safe(lp, safe, &ls->all_locks, all_list) {
517 list_del_init(&lp->all_list);
518
519 if (lp->lvb && lp->lvb != junk_lvb)
520 kfree(lp->lvb);
521 kfree(lp);
522 count++;
523 }
524 spin_unlock(&ls->async_lock);
525
526 return count;
527}
528