Commit | Line | Data |
---|---|---|
869d81df DT |
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 | |
e9fc2aa0 | 7 | * of the GNU General Public License version 2. |
869d81df | 8 | */ |
29b7998d DT |
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 | ||
15 | static 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 | ||
29b7998d DT |
25 | static void process_blocking(struct gdlm_lock *lp, int bast_mode) |
26 | { | |
27 | struct gdlm_ls *ls = lp->ls; | |
869d81df | 28 | unsigned int cb = 0; |
29b7998d DT |
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: | |
869d81df | 41 | gdlm_assert(0, "unknown bast mode %u", lp->bast_mode); |
29b7998d DT |
42 | } |
43 | ||
1c089c32 | 44 | ls->fscb(ls->sdp, cb, &lp->lockname); |
29b7998d DT |
45 | } |
46 | ||
41d7db0a SW |
47 | static void wake_up_ast(struct gdlm_lock *lp) |
48 | { | |
49 | clear_bit(LFL_AST_WAIT, &lp->flags); | |
50 | smp_mb__after_clear_bit(); | |
51 | wake_up_bit(&lp->flags, LFL_AST_WAIT); | |
52 | } | |
53 | ||
29b7998d DT |
54 | static void process_complete(struct gdlm_lock *lp) |
55 | { | |
56 | struct gdlm_ls *ls = lp->ls; | |
57 | struct lm_async_cb acb; | |
cd915493 | 58 | s16 prev_mode = lp->cur; |
29b7998d DT |
59 | |
60 | memset(&acb, 0, sizeof(acb)); | |
61 | ||
62 | if (lp->lksb.sb_status == -DLM_ECANCEL) { | |
869d81df | 63 | log_info("complete dlm cancel %x,%llx flags %lx", |
907b9bce | 64 | lp->lockname.ln_type, |
0239c4ae | 65 | (unsigned long long)lp->lockname.ln_number, |
869d81df | 66 | lp->flags); |
29b7998d DT |
67 | |
68 | lp->req = lp->cur; | |
69 | acb.lc_ret |= LM_OUT_CANCELED; | |
70 | if (lp->cur == DLM_LOCK_IV) | |
71 | lp->lksb.sb_lkid = 0; | |
72 | goto out; | |
73 | } | |
74 | ||
75 | if (test_and_clear_bit(LFL_DLM_UNLOCK, &lp->flags)) { | |
76 | if (lp->lksb.sb_status != -DLM_EUNLOCK) { | |
869d81df DT |
77 | log_info("unlock sb_status %d %x,%llx flags %lx", |
78 | lp->lksb.sb_status, lp->lockname.ln_type, | |
0239c4ae DW |
79 | (unsigned long long)lp->lockname.ln_number, |
80 | lp->flags); | |
29b7998d DT |
81 | return; |
82 | } | |
83 | ||
84 | lp->cur = DLM_LOCK_IV; | |
85 | lp->req = DLM_LOCK_IV; | |
86 | lp->lksb.sb_lkid = 0; | |
87 | ||
88 | if (test_and_clear_bit(LFL_UNLOCK_DELETE, &lp->flags)) { | |
89 | gdlm_delete_lp(lp); | |
90 | return; | |
91 | } | |
92 | goto out; | |
93 | } | |
94 | ||
95 | if (lp->lksb.sb_flags & DLM_SBF_VALNOTVALID) | |
96 | memset(lp->lksb.sb_lvbptr, 0, GDLM_LVB_SIZE); | |
97 | ||
98 | if (lp->lksb.sb_flags & DLM_SBF_ALTMODE) { | |
99 | if (lp->req == DLM_LOCK_PR) | |
100 | lp->req = DLM_LOCK_CW; | |
101 | else if (lp->req == DLM_LOCK_CW) | |
102 | lp->req = DLM_LOCK_PR; | |
103 | } | |
104 | ||
105 | /* | |
106 | * A canceled lock request. The lock was just taken off the delayed | |
107 | * list and was never even submitted to dlm. | |
108 | */ | |
109 | ||
110 | if (test_and_clear_bit(LFL_CANCEL, &lp->flags)) { | |
869d81df | 111 | log_info("complete internal cancel %x,%llx", |
907b9bce | 112 | lp->lockname.ln_type, |
0239c4ae | 113 | (unsigned long long)lp->lockname.ln_number); |
29b7998d DT |
114 | lp->req = lp->cur; |
115 | acb.lc_ret |= LM_OUT_CANCELED; | |
116 | goto out; | |
117 | } | |
118 | ||
119 | /* | |
120 | * An error occured. | |
121 | */ | |
122 | ||
123 | if (lp->lksb.sb_status) { | |
124 | /* a "normal" error */ | |
125 | if ((lp->lksb.sb_status == -EAGAIN) && | |
126 | (lp->lkf & DLM_LKF_NOQUEUE)) { | |
127 | lp->req = lp->cur; | |
128 | if (lp->cur == DLM_LOCK_IV) | |
129 | lp->lksb.sb_lkid = 0; | |
130 | goto out; | |
131 | } | |
132 | ||
133 | /* this could only happen with cancels I think */ | |
869d81df DT |
134 | log_info("ast sb_status %d %x,%llx flags %lx", |
135 | lp->lksb.sb_status, lp->lockname.ln_type, | |
0239c4ae DW |
136 | (unsigned long long)lp->lockname.ln_number, |
137 | lp->flags); | |
29b7998d DT |
138 | return; |
139 | } | |
140 | ||
141 | /* | |
142 | * This is an AST for an EX->EX conversion for sync_lvb from GFS. | |
143 | */ | |
144 | ||
145 | if (test_and_clear_bit(LFL_SYNC_LVB, &lp->flags)) { | |
41d7db0a | 146 | wake_up_ast(lp); |
29b7998d DT |
147 | return; |
148 | } | |
149 | ||
150 | /* | |
151 | * A lock has been demoted to NL because it initially completed during | |
152 | * BLOCK_LOCKS. Now it must be requested in the originally requested | |
153 | * mode. | |
154 | */ | |
155 | ||
156 | if (test_and_clear_bit(LFL_REREQUEST, &lp->flags)) { | |
869d81df | 157 | gdlm_assert(lp->req == DLM_LOCK_NL, "%x,%llx", |
0239c4ae DW |
158 | lp->lockname.ln_type, |
159 | (unsigned long long)lp->lockname.ln_number); | |
869d81df | 160 | gdlm_assert(lp->prev_req > DLM_LOCK_NL, "%x,%llx", |
0239c4ae DW |
161 | lp->lockname.ln_type, |
162 | (unsigned long long)lp->lockname.ln_number); | |
29b7998d DT |
163 | |
164 | lp->cur = DLM_LOCK_NL; | |
165 | lp->req = lp->prev_req; | |
166 | lp->prev_req = DLM_LOCK_IV; | |
167 | lp->lkf &= ~DLM_LKF_CONVDEADLK; | |
168 | ||
169 | set_bit(LFL_NOCACHE, &lp->flags); | |
170 | ||
171 | if (test_bit(DFL_BLOCK_LOCKS, &ls->flags) && | |
172 | !test_bit(LFL_NOBLOCK, &lp->flags)) | |
173 | gdlm_queue_delayed(lp); | |
174 | else | |
175 | queue_submit(lp); | |
176 | return; | |
177 | } | |
178 | ||
179 | /* | |
180 | * A request is granted during dlm recovery. It may be granted | |
181 | * because the locks of a failed node were cleared. In that case, | |
182 | * there may be inconsistent data beneath this lock and we must wait | |
183 | * for recovery to complete to use it. When gfs recovery is done this | |
184 | * granted lock will be converted to NL and then reacquired in this | |
185 | * granted state. | |
186 | */ | |
187 | ||
188 | if (test_bit(DFL_BLOCK_LOCKS, &ls->flags) && | |
189 | !test_bit(LFL_NOBLOCK, &lp->flags) && | |
190 | lp->req != DLM_LOCK_NL) { | |
191 | ||
192 | lp->cur = lp->req; | |
193 | lp->prev_req = lp->req; | |
194 | lp->req = DLM_LOCK_NL; | |
195 | lp->lkf |= DLM_LKF_CONVERT; | |
196 | lp->lkf &= ~DLM_LKF_CONVDEADLK; | |
197 | ||
869d81df | 198 | log_debug("rereq %x,%llx id %x %d,%d", |
0239c4ae DW |
199 | lp->lockname.ln_type, |
200 | (unsigned long long)lp->lockname.ln_number, | |
29b7998d DT |
201 | lp->lksb.sb_lkid, lp->cur, lp->req); |
202 | ||
203 | set_bit(LFL_REREQUEST, &lp->flags); | |
204 | queue_submit(lp); | |
205 | return; | |
206 | } | |
207 | ||
208 | /* | |
209 | * DLM demoted the lock to NL before it was granted so GFS must be | |
210 | * told it cannot cache data for this lock. | |
211 | */ | |
212 | ||
213 | if (lp->lksb.sb_flags & DLM_SBF_DEMOTED) | |
214 | set_bit(LFL_NOCACHE, &lp->flags); | |
215 | ||
a91ea69f | 216 | out: |
29b7998d DT |
217 | /* |
218 | * This is an internal lock_dlm lock | |
219 | */ | |
220 | ||
221 | if (test_bit(LFL_INLOCK, &lp->flags)) { | |
222 | clear_bit(LFL_NOBLOCK, &lp->flags); | |
223 | lp->cur = lp->req; | |
41d7db0a | 224 | wake_up_ast(lp); |
29b7998d DT |
225 | return; |
226 | } | |
227 | ||
228 | /* | |
229 | * Normal completion of a lock request. Tell GFS it now has the lock. | |
230 | */ | |
231 | ||
232 | clear_bit(LFL_NOBLOCK, &lp->flags); | |
233 | lp->cur = lp->req; | |
234 | ||
235 | acb.lc_name = lp->lockname; | |
236 | acb.lc_ret |= gdlm_make_lmstate(lp->cur); | |
237 | ||
238 | if (!test_and_clear_bit(LFL_NOCACHE, &lp->flags) && | |
239 | (lp->cur > DLM_LOCK_NL) && (prev_mode > DLM_LOCK_NL)) | |
240 | acb.lc_ret |= LM_OUT_CACHEABLE; | |
241 | ||
1c089c32 | 242 | ls->fscb(ls->sdp, LM_CB_ASYNC, &acb); |
29b7998d DT |
243 | } |
244 | ||
245 | static inline int no_work(struct gdlm_ls *ls, int blocking) | |
246 | { | |
247 | int ret; | |
248 | ||
249 | spin_lock(&ls->async_lock); | |
250 | ret = list_empty(&ls->complete) && list_empty(&ls->submit); | |
251 | if (ret && blocking) | |
252 | ret = list_empty(&ls->blocking); | |
253 | spin_unlock(&ls->async_lock); | |
254 | ||
255 | return ret; | |
256 | } | |
257 | ||
258 | static inline int check_drop(struct gdlm_ls *ls) | |
259 | { | |
260 | if (!ls->drop_locks_count) | |
261 | return 0; | |
262 | ||
263 | if (time_after(jiffies, ls->drop_time + ls->drop_locks_period * HZ)) { | |
264 | ls->drop_time = jiffies; | |
265 | if (ls->all_locks_count >= ls->drop_locks_count) | |
266 | return 1; | |
267 | } | |
268 | return 0; | |
269 | } | |
270 | ||
55c0c4ac | 271 | static int gdlm_thread(void *data, int blist) |
29b7998d DT |
272 | { |
273 | struct gdlm_ls *ls = (struct gdlm_ls *) data; | |
274 | struct gdlm_lock *lp = NULL; | |
29b7998d DT |
275 | uint8_t complete, blocking, submit, drop; |
276 | DECLARE_WAITQUEUE(wait, current); | |
277 | ||
278 | /* Only thread1 is allowed to do blocking callbacks since gfs | |
279 | may wait for a completion callback within a blocking cb. */ | |
280 | ||
29b7998d DT |
281 | while (!kthread_should_stop()) { |
282 | set_current_state(TASK_INTERRUPTIBLE); | |
283 | add_wait_queue(&ls->thread_wait, &wait); | |
284 | if (no_work(ls, blist)) | |
285 | schedule(); | |
286 | remove_wait_queue(&ls->thread_wait, &wait); | |
287 | set_current_state(TASK_RUNNING); | |
288 | ||
289 | complete = blocking = submit = drop = 0; | |
290 | ||
291 | spin_lock(&ls->async_lock); | |
292 | ||
293 | if (blist && !list_empty(&ls->blocking)) { | |
294 | lp = list_entry(ls->blocking.next, struct gdlm_lock, | |
295 | blist); | |
296 | list_del_init(&lp->blist); | |
297 | blocking = lp->bast_mode; | |
298 | lp->bast_mode = 0; | |
299 | } else if (!list_empty(&ls->complete)) { | |
300 | lp = list_entry(ls->complete.next, struct gdlm_lock, | |
301 | clist); | |
302 | list_del_init(&lp->clist); | |
303 | complete = 1; | |
304 | } else if (!list_empty(&ls->submit)) { | |
305 | lp = list_entry(ls->submit.next, struct gdlm_lock, | |
306 | delay_list); | |
307 | list_del_init(&lp->delay_list); | |
308 | submit = 1; | |
309 | } | |
310 | ||
311 | drop = check_drop(ls); | |
312 | spin_unlock(&ls->async_lock); | |
313 | ||
314 | if (complete) | |
315 | process_complete(lp); | |
316 | ||
317 | else if (blocking) | |
318 | process_blocking(lp, blocking); | |
319 | ||
320 | else if (submit) | |
8d3b35a4 | 321 | gdlm_do_lock(lp); |
29b7998d DT |
322 | |
323 | if (drop) | |
1c089c32 | 324 | ls->fscb(ls->sdp, LM_CB_DROPLOCKS, NULL); |
29b7998d DT |
325 | |
326 | schedule(); | |
327 | } | |
328 | ||
329 | return 0; | |
330 | } | |
331 | ||
55c0c4ac BP |
332 | static int gdlm_thread1(void *data) |
333 | { | |
334 | return gdlm_thread(data, 1); | |
335 | } | |
336 | ||
337 | static int gdlm_thread2(void *data) | |
338 | { | |
339 | return gdlm_thread(data, 0); | |
340 | } | |
341 | ||
29b7998d DT |
342 | int gdlm_init_threads(struct gdlm_ls *ls) |
343 | { | |
344 | struct task_struct *p; | |
345 | int error; | |
346 | ||
55c0c4ac | 347 | p = kthread_run(gdlm_thread1, ls, "lock_dlm1"); |
29b7998d DT |
348 | error = IS_ERR(p); |
349 | if (error) { | |
869d81df | 350 | log_error("can't start lock_dlm1 thread %d", error); |
29b7998d DT |
351 | return error; |
352 | } | |
353 | ls->thread1 = p; | |
354 | ||
55c0c4ac | 355 | p = kthread_run(gdlm_thread2, ls, "lock_dlm2"); |
29b7998d DT |
356 | error = IS_ERR(p); |
357 | if (error) { | |
869d81df | 358 | log_error("can't start lock_dlm2 thread %d", error); |
29b7998d DT |
359 | kthread_stop(ls->thread1); |
360 | return error; | |
361 | } | |
362 | ls->thread2 = p; | |
363 | ||
364 | return 0; | |
365 | } | |
366 | ||
367 | void gdlm_release_threads(struct gdlm_ls *ls) | |
368 | { | |
369 | kthread_stop(ls->thread1); | |
370 | kthread_stop(ls->thread2); | |
371 | } | |
372 |