]>
Commit | Line | Data |
---|---|---|
8b712842 CM |
1 | /* |
2 | * Copyright (C) 2007 Oracle. All rights reserved. | |
08a9ff32 | 3 | * Copyright (C) 2014 Fujitsu. All rights reserved. |
8b712842 CM |
4 | * |
5 | * This program is free software; you can redistribute it and/or | |
6 | * modify it under the terms of the GNU General Public | |
7 | * License v2 as published by the Free Software Foundation. | |
8 | * | |
9 | * This program is distributed in the hope that it will be useful, | |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
12 | * General Public License for more details. | |
13 | * | |
14 | * You should have received a copy of the GNU General Public | |
15 | * License along with this program; if not, write to the | |
16 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
17 | * Boston, MA 021110-1307, USA. | |
18 | */ | |
19 | ||
20 | #include <linux/kthread.h> | |
5a0e3ad6 | 21 | #include <linux/slab.h> |
8b712842 CM |
22 | #include <linux/list.h> |
23 | #include <linux/spinlock.h> | |
b51912c9 | 24 | #include <linux/freezer.h> |
8b712842 | 25 | #include "async-thread.h" |
52483bc2 | 26 | #include "ctree.h" |
8b712842 | 27 | |
a046e9c8 QW |
28 | #define WORK_DONE_BIT 0 |
29 | #define WORK_ORDER_DONE_BIT 1 | |
30 | #define WORK_HIGH_PRIO_BIT 2 | |
4a69a410 | 31 | |
0bd9289c QW |
32 | #define NO_THRESHOLD (-1) |
33 | #define DFT_THRESHOLD (32) | |
34 | ||
d458b054 | 35 | struct __btrfs_workqueue { |
08a9ff32 | 36 | struct workqueue_struct *normal_wq; |
cb001095 JM |
37 | |
38 | /* File system this workqueue services */ | |
39 | struct btrfs_fs_info *fs_info; | |
40 | ||
08a9ff32 QW |
41 | /* List head pointing to ordered work list */ |
42 | struct list_head ordered_list; | |
43 | ||
44 | /* Spinlock for ordered_list */ | |
45 | spinlock_t list_lock; | |
0bd9289c QW |
46 | |
47 | /* Thresholding related variants */ | |
48 | atomic_t pending; | |
c6dd6ea5 QW |
49 | |
50 | /* Up limit of concurrency workers */ | |
51 | int limit_active; | |
52 | ||
53 | /* Current number of concurrency workers */ | |
54 | int current_active; | |
55 | ||
56 | /* Threshold to change current_active */ | |
0bd9289c QW |
57 | int thresh; |
58 | unsigned int count; | |
59 | spinlock_t thres_lock; | |
08a9ff32 QW |
60 | }; |
61 | ||
d458b054 QW |
62 | struct btrfs_workqueue { |
63 | struct __btrfs_workqueue *normal; | |
64 | struct __btrfs_workqueue *high; | |
1ca08976 QW |
65 | }; |
66 | ||
9e0af237 LB |
67 | static void normal_work_helper(struct btrfs_work *work); |
68 | ||
69 | #define BTRFS_WORK_HELPER(name) \ | |
70 | void btrfs_##name(struct work_struct *arg) \ | |
71 | { \ | |
72 | struct btrfs_work *work = container_of(arg, struct btrfs_work, \ | |
73 | normal_work); \ | |
74 | normal_work_helper(work); \ | |
75 | } | |
76 | ||
cb001095 JM |
77 | struct btrfs_fs_info * |
78 | btrfs_workqueue_owner(struct __btrfs_workqueue *wq) | |
79 | { | |
80 | return wq->fs_info; | |
81 | } | |
82 | ||
83 | struct btrfs_fs_info * | |
84 | btrfs_work_owner(struct btrfs_work *work) | |
85 | { | |
86 | return work->wq->fs_info; | |
87 | } | |
88 | ||
9e0af237 LB |
89 | BTRFS_WORK_HELPER(worker_helper); |
90 | BTRFS_WORK_HELPER(delalloc_helper); | |
91 | BTRFS_WORK_HELPER(flush_delalloc_helper); | |
92 | BTRFS_WORK_HELPER(cache_helper); | |
93 | BTRFS_WORK_HELPER(submit_helper); | |
94 | BTRFS_WORK_HELPER(fixup_helper); | |
95 | BTRFS_WORK_HELPER(endio_helper); | |
96 | BTRFS_WORK_HELPER(endio_meta_helper); | |
97 | BTRFS_WORK_HELPER(endio_meta_write_helper); | |
98 | BTRFS_WORK_HELPER(endio_raid56_helper); | |
8b110e39 | 99 | BTRFS_WORK_HELPER(endio_repair_helper); |
9e0af237 LB |
100 | BTRFS_WORK_HELPER(rmw_helper); |
101 | BTRFS_WORK_HELPER(endio_write_helper); | |
102 | BTRFS_WORK_HELPER(freespace_write_helper); | |
103 | BTRFS_WORK_HELPER(delayed_meta_helper); | |
104 | BTRFS_WORK_HELPER(readahead_helper); | |
105 | BTRFS_WORK_HELPER(qgroup_rescan_helper); | |
106 | BTRFS_WORK_HELPER(extent_refs_helper); | |
107 | BTRFS_WORK_HELPER(scrub_helper); | |
108 | BTRFS_WORK_HELPER(scrubwrc_helper); | |
109 | BTRFS_WORK_HELPER(scrubnc_helper); | |
20b2e302 | 110 | BTRFS_WORK_HELPER(scrubparity_helper); |
9e0af237 LB |
111 | |
112 | static struct __btrfs_workqueue * | |
cb001095 JM |
113 | __btrfs_alloc_workqueue(struct btrfs_fs_info *fs_info, const char *name, |
114 | unsigned int flags, int limit_active, int thresh) | |
1ca08976 | 115 | { |
61dd5ae6 | 116 | struct __btrfs_workqueue *ret = kzalloc(sizeof(*ret), GFP_KERNEL); |
1ca08976 | 117 | |
5d99a998 | 118 | if (!ret) |
1ca08976 QW |
119 | return NULL; |
120 | ||
cb001095 | 121 | ret->fs_info = fs_info; |
c6dd6ea5 | 122 | ret->limit_active = limit_active; |
0bd9289c QW |
123 | atomic_set(&ret->pending, 0); |
124 | if (thresh == 0) | |
125 | thresh = DFT_THRESHOLD; | |
126 | /* For low threshold, disabling threshold is a better choice */ | |
127 | if (thresh < DFT_THRESHOLD) { | |
c6dd6ea5 | 128 | ret->current_active = limit_active; |
0bd9289c QW |
129 | ret->thresh = NO_THRESHOLD; |
130 | } else { | |
c6dd6ea5 QW |
131 | /* |
132 | * For threshold-able wq, let its concurrency grow on demand. | |
133 | * Use minimal max_active at alloc time to reduce resource | |
134 | * usage. | |
135 | */ | |
136 | ret->current_active = 1; | |
0bd9289c QW |
137 | ret->thresh = thresh; |
138 | } | |
139 | ||
1ca08976 QW |
140 | if (flags & WQ_HIGHPRI) |
141 | ret->normal_wq = alloc_workqueue("%s-%s-high", flags, | |
c6dd6ea5 QW |
142 | ret->current_active, "btrfs", |
143 | name); | |
1ca08976 QW |
144 | else |
145 | ret->normal_wq = alloc_workqueue("%s-%s", flags, | |
c6dd6ea5 | 146 | ret->current_active, "btrfs", |
0bd9289c | 147 | name); |
5d99a998 | 148 | if (!ret->normal_wq) { |
1ca08976 QW |
149 | kfree(ret); |
150 | return NULL; | |
151 | } | |
152 | ||
153 | INIT_LIST_HEAD(&ret->ordered_list); | |
154 | spin_lock_init(&ret->list_lock); | |
0bd9289c | 155 | spin_lock_init(&ret->thres_lock); |
c3a46891 | 156 | trace_btrfs_workqueue_alloc(ret, name, flags & WQ_HIGHPRI); |
1ca08976 QW |
157 | return ret; |
158 | } | |
159 | ||
160 | static inline void | |
d458b054 | 161 | __btrfs_destroy_workqueue(struct __btrfs_workqueue *wq); |
1ca08976 | 162 | |
cb001095 JM |
163 | struct btrfs_workqueue *btrfs_alloc_workqueue(struct btrfs_fs_info *fs_info, |
164 | const char *name, | |
6f011058 | 165 | unsigned int flags, |
c6dd6ea5 | 166 | int limit_active, |
d458b054 | 167 | int thresh) |
08a9ff32 | 168 | { |
61dd5ae6 | 169 | struct btrfs_workqueue *ret = kzalloc(sizeof(*ret), GFP_KERNEL); |
08a9ff32 | 170 | |
5d99a998 | 171 | if (!ret) |
08a9ff32 QW |
172 | return NULL; |
173 | ||
cb001095 JM |
174 | ret->normal = __btrfs_alloc_workqueue(fs_info, name, |
175 | flags & ~WQ_HIGHPRI, | |
c6dd6ea5 | 176 | limit_active, thresh); |
5d99a998 | 177 | if (!ret->normal) { |
08a9ff32 QW |
178 | kfree(ret); |
179 | return NULL; | |
180 | } | |
181 | ||
1ca08976 | 182 | if (flags & WQ_HIGHPRI) { |
cb001095 JM |
183 | ret->high = __btrfs_alloc_workqueue(fs_info, name, flags, |
184 | limit_active, thresh); | |
5d99a998 | 185 | if (!ret->high) { |
1ca08976 QW |
186 | __btrfs_destroy_workqueue(ret->normal); |
187 | kfree(ret); | |
188 | return NULL; | |
189 | } | |
190 | } | |
08a9ff32 QW |
191 | return ret; |
192 | } | |
193 | ||
0bd9289c QW |
194 | /* |
195 | * Hook for threshold which will be called in btrfs_queue_work. | |
196 | * This hook WILL be called in IRQ handler context, | |
197 | * so workqueue_set_max_active MUST NOT be called in this hook | |
198 | */ | |
d458b054 | 199 | static inline void thresh_queue_hook(struct __btrfs_workqueue *wq) |
0bd9289c QW |
200 | { |
201 | if (wq->thresh == NO_THRESHOLD) | |
202 | return; | |
203 | atomic_inc(&wq->pending); | |
204 | } | |
205 | ||
206 | /* | |
207 | * Hook for threshold which will be called before executing the work, | |
208 | * This hook is called in kthread content. | |
209 | * So workqueue_set_max_active is called here. | |
210 | */ | |
d458b054 | 211 | static inline void thresh_exec_hook(struct __btrfs_workqueue *wq) |
0bd9289c | 212 | { |
c6dd6ea5 | 213 | int new_current_active; |
0bd9289c QW |
214 | long pending; |
215 | int need_change = 0; | |
216 | ||
217 | if (wq->thresh == NO_THRESHOLD) | |
218 | return; | |
219 | ||
220 | atomic_dec(&wq->pending); | |
221 | spin_lock(&wq->thres_lock); | |
222 | /* | |
223 | * Use wq->count to limit the calling frequency of | |
224 | * workqueue_set_max_active. | |
225 | */ | |
226 | wq->count++; | |
227 | wq->count %= (wq->thresh / 4); | |
228 | if (!wq->count) | |
229 | goto out; | |
c6dd6ea5 | 230 | new_current_active = wq->current_active; |
0bd9289c QW |
231 | |
232 | /* | |
233 | * pending may be changed later, but it's OK since we really | |
234 | * don't need it so accurate to calculate new_max_active. | |
235 | */ | |
236 | pending = atomic_read(&wq->pending); | |
237 | if (pending > wq->thresh) | |
c6dd6ea5 | 238 | new_current_active++; |
0bd9289c | 239 | if (pending < wq->thresh / 2) |
c6dd6ea5 QW |
240 | new_current_active--; |
241 | new_current_active = clamp_val(new_current_active, 1, wq->limit_active); | |
242 | if (new_current_active != wq->current_active) { | |
0bd9289c | 243 | need_change = 1; |
c6dd6ea5 | 244 | wq->current_active = new_current_active; |
0bd9289c QW |
245 | } |
246 | out: | |
247 | spin_unlock(&wq->thres_lock); | |
248 | ||
249 | if (need_change) { | |
c6dd6ea5 | 250 | workqueue_set_max_active(wq->normal_wq, wq->current_active); |
0bd9289c QW |
251 | } |
252 | } | |
253 | ||
d458b054 | 254 | static void run_ordered_work(struct __btrfs_workqueue *wq) |
08a9ff32 QW |
255 | { |
256 | struct list_head *list = &wq->ordered_list; | |
d458b054 | 257 | struct btrfs_work *work; |
08a9ff32 QW |
258 | spinlock_t *lock = &wq->list_lock; |
259 | unsigned long flags; | |
260 | ||
261 | while (1) { | |
262 | spin_lock_irqsave(lock, flags); | |
263 | if (list_empty(list)) | |
264 | break; | |
d458b054 | 265 | work = list_entry(list->next, struct btrfs_work, |
08a9ff32 QW |
266 | ordered_list); |
267 | if (!test_bit(WORK_DONE_BIT, &work->flags)) | |
268 | break; | |
269 | ||
270 | /* | |
271 | * we are going to call the ordered done function, but | |
272 | * we leave the work item on the list as a barrier so | |
273 | * that later work items that are done don't have their | |
274 | * functions called before this one returns | |
275 | */ | |
276 | if (test_and_set_bit(WORK_ORDER_DONE_BIT, &work->flags)) | |
277 | break; | |
52483bc2 | 278 | trace_btrfs_ordered_sched(work); |
08a9ff32 QW |
279 | spin_unlock_irqrestore(lock, flags); |
280 | work->ordered_func(work); | |
281 | ||
282 | /* now take the lock again and drop our item from the list */ | |
283 | spin_lock_irqsave(lock, flags); | |
284 | list_del(&work->ordered_list); | |
285 | spin_unlock_irqrestore(lock, flags); | |
286 | ||
287 | /* | |
288 | * we don't want to call the ordered free functions | |
289 | * with the lock held though | |
290 | */ | |
291 | work->ordered_free(work); | |
52483bc2 | 292 | trace_btrfs_all_work_done(work); |
08a9ff32 QW |
293 | } |
294 | spin_unlock_irqrestore(lock, flags); | |
295 | } | |
296 | ||
9e0af237 | 297 | static void normal_work_helper(struct btrfs_work *work) |
08a9ff32 | 298 | { |
d458b054 | 299 | struct __btrfs_workqueue *wq; |
08a9ff32 QW |
300 | int need_order = 0; |
301 | ||
08a9ff32 QW |
302 | /* |
303 | * We should not touch things inside work in the following cases: | |
304 | * 1) after work->func() if it has no ordered_free | |
305 | * Since the struct is freed in work->func(). | |
306 | * 2) after setting WORK_DONE_BIT | |
307 | * The work may be freed in other threads almost instantly. | |
308 | * So we save the needed things here. | |
309 | */ | |
310 | if (work->ordered_func) | |
311 | need_order = 1; | |
312 | wq = work->wq; | |
313 | ||
52483bc2 | 314 | trace_btrfs_work_sched(work); |
0bd9289c | 315 | thresh_exec_hook(wq); |
08a9ff32 QW |
316 | work->func(work); |
317 | if (need_order) { | |
318 | set_bit(WORK_DONE_BIT, &work->flags); | |
319 | run_ordered_work(wq); | |
320 | } | |
52483bc2 QW |
321 | if (!need_order) |
322 | trace_btrfs_all_work_done(work); | |
08a9ff32 QW |
323 | } |
324 | ||
9e0af237 | 325 | void btrfs_init_work(struct btrfs_work *work, btrfs_work_func_t uniq_func, |
6db8914f QW |
326 | btrfs_func_t func, |
327 | btrfs_func_t ordered_func, | |
328 | btrfs_func_t ordered_free) | |
08a9ff32 QW |
329 | { |
330 | work->func = func; | |
331 | work->ordered_func = ordered_func; | |
332 | work->ordered_free = ordered_free; | |
9e0af237 | 333 | INIT_WORK(&work->normal_work, uniq_func); |
08a9ff32 QW |
334 | INIT_LIST_HEAD(&work->ordered_list); |
335 | work->flags = 0; | |
336 | } | |
337 | ||
d458b054 QW |
338 | static inline void __btrfs_queue_work(struct __btrfs_workqueue *wq, |
339 | struct btrfs_work *work) | |
08a9ff32 QW |
340 | { |
341 | unsigned long flags; | |
342 | ||
343 | work->wq = wq; | |
0bd9289c | 344 | thresh_queue_hook(wq); |
08a9ff32 QW |
345 | if (work->ordered_func) { |
346 | spin_lock_irqsave(&wq->list_lock, flags); | |
347 | list_add_tail(&work->ordered_list, &wq->ordered_list); | |
348 | spin_unlock_irqrestore(&wq->list_lock, flags); | |
349 | } | |
52483bc2 | 350 | trace_btrfs_work_queued(work); |
0a95b851 | 351 | queue_work(wq->normal_wq, &work->normal_work); |
08a9ff32 QW |
352 | } |
353 | ||
d458b054 QW |
354 | void btrfs_queue_work(struct btrfs_workqueue *wq, |
355 | struct btrfs_work *work) | |
1ca08976 | 356 | { |
d458b054 | 357 | struct __btrfs_workqueue *dest_wq; |
1ca08976 QW |
358 | |
359 | if (test_bit(WORK_HIGH_PRIO_BIT, &work->flags) && wq->high) | |
360 | dest_wq = wq->high; | |
361 | else | |
362 | dest_wq = wq->normal; | |
363 | __btrfs_queue_work(dest_wq, work); | |
364 | } | |
365 | ||
366 | static inline void | |
d458b054 | 367 | __btrfs_destroy_workqueue(struct __btrfs_workqueue *wq) |
08a9ff32 QW |
368 | { |
369 | destroy_workqueue(wq->normal_wq); | |
c3a46891 | 370 | trace_btrfs_workqueue_destroy(wq); |
08a9ff32 QW |
371 | kfree(wq); |
372 | } | |
373 | ||
d458b054 | 374 | void btrfs_destroy_workqueue(struct btrfs_workqueue *wq) |
1ca08976 QW |
375 | { |
376 | if (!wq) | |
377 | return; | |
378 | if (wq->high) | |
379 | __btrfs_destroy_workqueue(wq->high); | |
380 | __btrfs_destroy_workqueue(wq->normal); | |
ef66af10 | 381 | kfree(wq); |
1ca08976 QW |
382 | } |
383 | ||
c6dd6ea5 | 384 | void btrfs_workqueue_set_max(struct btrfs_workqueue *wq, int limit_active) |
08a9ff32 | 385 | { |
800ee224 ST |
386 | if (!wq) |
387 | return; | |
c6dd6ea5 | 388 | wq->normal->limit_active = limit_active; |
1ca08976 | 389 | if (wq->high) |
c6dd6ea5 | 390 | wq->high->limit_active = limit_active; |
1ca08976 QW |
391 | } |
392 | ||
d458b054 | 393 | void btrfs_set_work_high_priority(struct btrfs_work *work) |
1ca08976 QW |
394 | { |
395 | set_bit(WORK_HIGH_PRIO_BIT, &work->flags); | |
08a9ff32 | 396 | } |