Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
3fcfab16 AM |
2 | |
3 | #include <linux/wait.h> | |
34f8fe50 | 4 | #include <linux/rbtree.h> |
3fcfab16 | 5 | #include <linux/backing-dev.h> |
03ba3782 JA |
6 | #include <linux/kthread.h> |
7 | #include <linux/freezer.h> | |
3fcfab16 | 8 | #include <linux/fs.h> |
26160158 | 9 | #include <linux/pagemap.h> |
03ba3782 | 10 | #include <linux/mm.h> |
c1ca59a1 | 11 | #include <linux/sched/mm.h> |
3fcfab16 AM |
12 | #include <linux/sched.h> |
13 | #include <linux/module.h> | |
cf0ca9fe PZ |
14 | #include <linux/writeback.h> |
15 | #include <linux/device.h> | |
455b2864 | 16 | #include <trace/events/writeback.h> |
cf0ca9fe | 17 | |
f56753ac | 18 | struct backing_dev_info noop_backing_dev_info; |
a212b105 | 19 | EXPORT_SYMBOL_GPL(noop_backing_dev_info); |
5129a469 | 20 | |
cf0ca9fe | 21 | static struct class *bdi_class; |
eb7ae5e0 | 22 | static const char *bdi_unknown_name = "(unknown)"; |
cfc4ba53 JA |
23 | |
24 | /* | |
34f8fe50 TH |
25 | * bdi_lock protects bdi_tree and updates to bdi_list. bdi_list has RCU |
26 | * reader side locking. | |
cfc4ba53 | 27 | */ |
03ba3782 | 28 | DEFINE_SPINLOCK(bdi_lock); |
34f8fe50 TH |
29 | static u64 bdi_id_cursor; |
30 | static struct rb_root bdi_tree = RB_ROOT; | |
66f3b8e2 | 31 | LIST_HEAD(bdi_list); |
03ba3782 | 32 | |
839a8e86 TH |
33 | /* bdi_wq serves all asynchronous writeback tasks */ |
34 | struct workqueue_struct *bdi_wq; | |
35 | ||
6986c3e2 BW |
36 | #define K(x) ((x) << (PAGE_SHIFT - 10)) |
37 | ||
76f1418b MS |
38 | #ifdef CONFIG_DEBUG_FS |
39 | #include <linux/debugfs.h> | |
40 | #include <linux/seq_file.h> | |
41 | ||
42 | static struct dentry *bdi_debug_root; | |
43 | ||
44 | static void bdi_debug_init(void) | |
45 | { | |
46 | bdi_debug_root = debugfs_create_dir("bdi", NULL); | |
47 | } | |
48 | ||
49 | static int bdi_debug_stats_show(struct seq_file *m, void *v) | |
50 | { | |
51 | struct backing_dev_info *bdi = m->private; | |
c1955ce3 | 52 | struct bdi_writeback *wb = &bdi->wb; |
364aeb28 DR |
53 | unsigned long background_thresh; |
54 | unsigned long dirty_thresh; | |
0d960a38 | 55 | unsigned long wb_thresh; |
0ae45f63 | 56 | unsigned long nr_dirty, nr_io, nr_more_io, nr_dirty_time; |
f09b00d3 JA |
57 | struct inode *inode; |
58 | ||
0ae45f63 | 59 | nr_dirty = nr_io = nr_more_io = nr_dirty_time = 0; |
f758eeab | 60 | spin_lock(&wb->list_lock); |
c7f54084 | 61 | list_for_each_entry(inode, &wb->b_dirty, i_io_list) |
c1955ce3 | 62 | nr_dirty++; |
c7f54084 | 63 | list_for_each_entry(inode, &wb->b_io, i_io_list) |
c1955ce3 | 64 | nr_io++; |
c7f54084 | 65 | list_for_each_entry(inode, &wb->b_more_io, i_io_list) |
c1955ce3 | 66 | nr_more_io++; |
c7f54084 | 67 | list_for_each_entry(inode, &wb->b_dirty_time, i_io_list) |
0ae45f63 TT |
68 | if (inode->i_state & I_DIRTY_TIME) |
69 | nr_dirty_time++; | |
f758eeab | 70 | spin_unlock(&wb->list_lock); |
76f1418b | 71 | |
16c4042f | 72 | global_dirty_limits(&background_thresh, &dirty_thresh); |
0d960a38 | 73 | wb_thresh = wb_calc_thresh(wb, dirty_thresh); |
76f1418b | 74 | |
76f1418b | 75 | seq_printf(m, |
00821b00 WF |
76 | "BdiWriteback: %10lu kB\n" |
77 | "BdiReclaimable: %10lu kB\n" | |
78 | "BdiDirtyThresh: %10lu kB\n" | |
79 | "DirtyThresh: %10lu kB\n" | |
80 | "BackgroundThresh: %10lu kB\n" | |
c8e28ce0 | 81 | "BdiDirtied: %10lu kB\n" |
00821b00 WF |
82 | "BdiWritten: %10lu kB\n" |
83 | "BdiWriteBandwidth: %10lu kBps\n" | |
84 | "b_dirty: %10lu\n" | |
85 | "b_io: %10lu\n" | |
86 | "b_more_io: %10lu\n" | |
0ae45f63 | 87 | "b_dirty_time: %10lu\n" |
00821b00 WF |
88 | "bdi_list: %10u\n" |
89 | "state: %10lx\n", | |
93f78d88 TH |
90 | (unsigned long) K(wb_stat(wb, WB_WRITEBACK)), |
91 | (unsigned long) K(wb_stat(wb, WB_RECLAIMABLE)), | |
0d960a38 | 92 | K(wb_thresh), |
f7d2b1ec JK |
93 | K(dirty_thresh), |
94 | K(background_thresh), | |
93f78d88 TH |
95 | (unsigned long) K(wb_stat(wb, WB_DIRTIED)), |
96 | (unsigned long) K(wb_stat(wb, WB_WRITTEN)), | |
a88a341a | 97 | (unsigned long) K(wb->write_bandwidth), |
f7d2b1ec JK |
98 | nr_dirty, |
99 | nr_io, | |
100 | nr_more_io, | |
0ae45f63 | 101 | nr_dirty_time, |
4452226e | 102 | !list_empty(&bdi->bdi_list), bdi->wb.state); |
76f1418b MS |
103 | |
104 | return 0; | |
105 | } | |
5ad35093 | 106 | DEFINE_SHOW_ATTRIBUTE(bdi_debug_stats); |
76f1418b | 107 | |
2d146b92 | 108 | static void bdi_debug_register(struct backing_dev_info *bdi, const char *name) |
76f1418b MS |
109 | { |
110 | bdi->debug_dir = debugfs_create_dir(name, bdi_debug_root); | |
97f07697 | 111 | |
2d146b92 GKH |
112 | debugfs_create_file("stats", 0444, bdi->debug_dir, bdi, |
113 | &bdi_debug_stats_fops); | |
76f1418b MS |
114 | } |
115 | ||
116 | static void bdi_debug_unregister(struct backing_dev_info *bdi) | |
117 | { | |
2d146b92 | 118 | debugfs_remove_recursive(bdi->debug_dir); |
76f1418b MS |
119 | } |
120 | #else | |
121 | static inline void bdi_debug_init(void) | |
122 | { | |
123 | } | |
2d146b92 | 124 | static inline void bdi_debug_register(struct backing_dev_info *bdi, |
76f1418b MS |
125 | const char *name) |
126 | { | |
127 | } | |
128 | static inline void bdi_debug_unregister(struct backing_dev_info *bdi) | |
129 | { | |
130 | } | |
131 | #endif | |
132 | ||
cf0ca9fe PZ |
133 | static ssize_t read_ahead_kb_store(struct device *dev, |
134 | struct device_attribute *attr, | |
135 | const char *buf, size_t count) | |
136 | { | |
137 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
cf0ca9fe | 138 | unsigned long read_ahead_kb; |
7034ed13 | 139 | ssize_t ret; |
cf0ca9fe | 140 | |
7034ed13 NJ |
141 | ret = kstrtoul(buf, 10, &read_ahead_kb); |
142 | if (ret < 0) | |
143 | return ret; | |
144 | ||
145 | bdi->ra_pages = read_ahead_kb >> (PAGE_SHIFT - 10); | |
146 | ||
147 | return count; | |
cf0ca9fe PZ |
148 | } |
149 | ||
cf0ca9fe PZ |
150 | #define BDI_SHOW(name, expr) \ |
151 | static ssize_t name##_show(struct device *dev, \ | |
5e4c0d86 | 152 | struct device_attribute *attr, char *buf) \ |
cf0ca9fe PZ |
153 | { \ |
154 | struct backing_dev_info *bdi = dev_get_drvdata(dev); \ | |
155 | \ | |
5e4c0d86 | 156 | return sysfs_emit(buf, "%lld\n", (long long)expr); \ |
d9e1241e GKH |
157 | } \ |
158 | static DEVICE_ATTR_RW(name); | |
cf0ca9fe PZ |
159 | |
160 | BDI_SHOW(read_ahead_kb, K(bdi->ra_pages)) | |
161 | ||
189d3c4a PZ |
162 | static ssize_t min_ratio_store(struct device *dev, |
163 | struct device_attribute *attr, const char *buf, size_t count) | |
164 | { | |
165 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
189d3c4a | 166 | unsigned int ratio; |
7034ed13 NJ |
167 | ssize_t ret; |
168 | ||
169 | ret = kstrtouint(buf, 10, &ratio); | |
170 | if (ret < 0) | |
171 | return ret; | |
172 | ||
173 | ret = bdi_set_min_ratio(bdi, ratio); | |
174 | if (!ret) | |
175 | ret = count; | |
189d3c4a | 176 | |
189d3c4a PZ |
177 | return ret; |
178 | } | |
179 | BDI_SHOW(min_ratio, bdi->min_ratio) | |
180 | ||
a42dde04 PZ |
181 | static ssize_t max_ratio_store(struct device *dev, |
182 | struct device_attribute *attr, const char *buf, size_t count) | |
183 | { | |
184 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
a42dde04 | 185 | unsigned int ratio; |
7034ed13 NJ |
186 | ssize_t ret; |
187 | ||
188 | ret = kstrtouint(buf, 10, &ratio); | |
189 | if (ret < 0) | |
190 | return ret; | |
191 | ||
192 | ret = bdi_set_max_ratio(bdi, ratio); | |
193 | if (!ret) | |
194 | ret = count; | |
a42dde04 | 195 | |
a42dde04 PZ |
196 | return ret; |
197 | } | |
198 | BDI_SHOW(max_ratio, bdi->max_ratio) | |
199 | ||
7d311cda DW |
200 | static ssize_t stable_pages_required_show(struct device *dev, |
201 | struct device_attribute *attr, | |
5e4c0d86 | 202 | char *buf) |
7d311cda | 203 | { |
1cb039f3 CH |
204 | dev_warn_once(dev, |
205 | "the stable_pages_required attribute has been removed. Use the stable_writes queue attribute instead.\n"); | |
5e4c0d86 | 206 | return sysfs_emit(buf, "%d\n", 0); |
7d311cda | 207 | } |
d9e1241e GKH |
208 | static DEVICE_ATTR_RO(stable_pages_required); |
209 | ||
210 | static struct attribute *bdi_dev_attrs[] = { | |
211 | &dev_attr_read_ahead_kb.attr, | |
212 | &dev_attr_min_ratio.attr, | |
213 | &dev_attr_max_ratio.attr, | |
214 | &dev_attr_stable_pages_required.attr, | |
215 | NULL, | |
cf0ca9fe | 216 | }; |
d9e1241e | 217 | ATTRIBUTE_GROUPS(bdi_dev); |
cf0ca9fe PZ |
218 | |
219 | static __init int bdi_class_init(void) | |
220 | { | |
221 | bdi_class = class_create(THIS_MODULE, "bdi"); | |
14421453 AB |
222 | if (IS_ERR(bdi_class)) |
223 | return PTR_ERR(bdi_class); | |
224 | ||
d9e1241e | 225 | bdi_class->dev_groups = bdi_dev_groups; |
76f1418b | 226 | bdi_debug_init(); |
d03f6cdc | 227 | |
cf0ca9fe PZ |
228 | return 0; |
229 | } | |
76f1418b | 230 | postcore_initcall(bdi_class_init); |
cf0ca9fe | 231 | |
2e82b84c JK |
232 | static int bdi_init(struct backing_dev_info *bdi); |
233 | ||
26160158 JA |
234 | static int __init default_bdi_init(void) |
235 | { | |
236 | int err; | |
237 | ||
a2b90f11 MW |
238 | bdi_wq = alloc_workqueue("writeback", WQ_MEM_RECLAIM | WQ_UNBOUND | |
239 | WQ_SYSFS, 0); | |
839a8e86 TH |
240 | if (!bdi_wq) |
241 | return -ENOMEM; | |
242 | ||
976e48f8 | 243 | err = bdi_init(&noop_backing_dev_info); |
26160158 JA |
244 | |
245 | return err; | |
246 | } | |
247 | subsys_initcall(default_bdi_init); | |
248 | ||
6467716a | 249 | /* |
f0054bb1 | 250 | * This function is used when the first inode for this wb is marked dirty. It |
6467716a AB |
251 | * wakes-up the corresponding bdi thread which should then take care of the |
252 | * periodic background write-out of dirty inodes. Since the write-out would | |
253 | * starts only 'dirty_writeback_interval' centisecs from now anyway, we just | |
254 | * set up a timer which wakes the bdi thread up later. | |
255 | * | |
256 | * Note, we wouldn't bother setting up the timer, but this function is on the | |
257 | * fast-path (used by '__mark_inode_dirty()'), so we save few context switches | |
258 | * by delaying the wake-up. | |
6ca738d6 DB |
259 | * |
260 | * We have to be careful not to postpone flush work if it is scheduled for | |
261 | * earlier. Thus we use queue_delayed_work(). | |
6467716a | 262 | */ |
f0054bb1 | 263 | void wb_wakeup_delayed(struct bdi_writeback *wb) |
6467716a AB |
264 | { |
265 | unsigned long timeout; | |
266 | ||
267 | timeout = msecs_to_jiffies(dirty_writeback_interval * 10); | |
f0054bb1 TH |
268 | spin_lock_bh(&wb->work_lock); |
269 | if (test_bit(WB_registered, &wb->state)) | |
270 | queue_delayed_work(bdi_wq, &wb->dwork, timeout); | |
271 | spin_unlock_bh(&wb->work_lock); | |
03ba3782 JA |
272 | } |
273 | ||
45a2966f JK |
274 | static void wb_update_bandwidth_workfn(struct work_struct *work) |
275 | { | |
276 | struct bdi_writeback *wb = container_of(to_delayed_work(work), | |
277 | struct bdi_writeback, bw_dwork); | |
278 | ||
279 | wb_update_bandwidth(wb); | |
280 | } | |
281 | ||
cfc4ba53 | 282 | /* |
a88a341a | 283 | * Initial write bandwidth: 100 MB/s |
cfc4ba53 | 284 | */ |
a88a341a | 285 | #define INIT_BW (100 << (20 - PAGE_SHIFT)) |
cfc4ba53 | 286 | |
8395cd9f | 287 | static int wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi, |
8c911f3d | 288 | gfp_t gfp) |
cf0ca9fe | 289 | { |
93f78d88 | 290 | int i, err; |
cf0ca9fe | 291 | |
6467716a | 292 | memset(wb, 0, sizeof(*wb)); |
f1d0b063 | 293 | |
810df54a JK |
294 | if (wb != &bdi->wb) |
295 | bdi_get(bdi); | |
6467716a AB |
296 | wb->bdi = bdi; |
297 | wb->last_old_flush = jiffies; | |
298 | INIT_LIST_HEAD(&wb->b_dirty); | |
299 | INIT_LIST_HEAD(&wb->b_io); | |
300 | INIT_LIST_HEAD(&wb->b_more_io); | |
0ae45f63 | 301 | INIT_LIST_HEAD(&wb->b_dirty_time); |
f758eeab | 302 | spin_lock_init(&wb->list_lock); |
66f3b8e2 | 303 | |
633a2abb | 304 | atomic_set(&wb->writeback_inodes, 0); |
a88a341a TH |
305 | wb->bw_time_stamp = jiffies; |
306 | wb->balanced_dirty_ratelimit = INIT_BW; | |
307 | wb->dirty_ratelimit = INIT_BW; | |
308 | wb->write_bandwidth = INIT_BW; | |
309 | wb->avg_write_bandwidth = INIT_BW; | |
cf0ca9fe | 310 | |
f0054bb1 TH |
311 | spin_lock_init(&wb->work_lock); |
312 | INIT_LIST_HEAD(&wb->work_list); | |
313 | INIT_DELAYED_WORK(&wb->dwork, wb_workfn); | |
45a2966f | 314 | INIT_DELAYED_WORK(&wb->bw_dwork, wb_update_bandwidth_workfn); |
b57d74af | 315 | wb->dirty_sleep = jiffies; |
c284de61 | 316 | |
8395cd9f | 317 | err = fprop_local_init_percpu(&wb->completions, gfp); |
a88a341a | 318 | if (err) |
8c911f3d | 319 | goto out_put_bdi; |
c284de61 | 320 | |
93f78d88 | 321 | for (i = 0; i < NR_WB_STAT_ITEMS; i++) { |
8395cd9f | 322 | err = percpu_counter_init(&wb->stat[i], 0, gfp); |
a13f35e8 TH |
323 | if (err) |
324 | goto out_destroy_stat; | |
93f78d88 | 325 | } |
cf0ca9fe | 326 | |
93f78d88 | 327 | return 0; |
a13f35e8 TH |
328 | |
329 | out_destroy_stat: | |
078c6c3a | 330 | while (i--) |
a13f35e8 TH |
331 | percpu_counter_destroy(&wb->stat[i]); |
332 | fprop_local_destroy_percpu(&wb->completions); | |
810df54a JK |
333 | out_put_bdi: |
334 | if (wb != &bdi->wb) | |
335 | bdi_put(bdi); | |
a13f35e8 | 336 | return err; |
cf0ca9fe | 337 | } |
cf0ca9fe | 338 | |
e8cb72b3 JK |
339 | static void cgwb_remove_from_bdi_list(struct bdi_writeback *wb); |
340 | ||
03ba3782 JA |
341 | /* |
342 | * Remove bdi from the global list and shutdown any threads we have running | |
343 | */ | |
46100071 | 344 | static void wb_shutdown(struct bdi_writeback *wb) |
66f3b8e2 | 345 | { |
c4db59d3 | 346 | /* Make sure nobody queues further work */ |
46100071 TH |
347 | spin_lock_bh(&wb->work_lock); |
348 | if (!test_and_clear_bit(WB_registered, &wb->state)) { | |
349 | spin_unlock_bh(&wb->work_lock); | |
03ba3782 | 350 | return; |
c4db59d3 | 351 | } |
46100071 | 352 | spin_unlock_bh(&wb->work_lock); |
03ba3782 | 353 | |
e8cb72b3 | 354 | cgwb_remove_from_bdi_list(wb); |
03ba3782 | 355 | /* |
46100071 TH |
356 | * Drain work list and shutdown the delayed_work. !WB_registered |
357 | * tells wb_workfn() that @wb is dying and its work_list needs to | |
358 | * be drained no matter what. | |
03ba3782 | 359 | */ |
46100071 TH |
360 | mod_delayed_work(bdi_wq, &wb->dwork, 0); |
361 | flush_delayed_work(&wb->dwork); | |
362 | WARN_ON(!list_empty(&wb->work_list)); | |
45a2966f | 363 | flush_delayed_work(&wb->bw_dwork); |
46100071 TH |
364 | } |
365 | ||
f0054bb1 | 366 | static void wb_exit(struct bdi_writeback *wb) |
93f78d88 TH |
367 | { |
368 | int i; | |
369 | ||
370 | WARN_ON(delayed_work_pending(&wb->dwork)); | |
371 | ||
372 | for (i = 0; i < NR_WB_STAT_ITEMS; i++) | |
373 | percpu_counter_destroy(&wb->stat[i]); | |
6467716a | 374 | |
a88a341a | 375 | fprop_local_destroy_percpu(&wb->completions); |
810df54a JK |
376 | if (wb != &wb->bdi->wb) |
377 | bdi_put(wb->bdi); | |
a88a341a | 378 | } |
e98be2d5 | 379 | |
52ebea74 TH |
380 | #ifdef CONFIG_CGROUP_WRITEBACK |
381 | ||
382 | #include <linux/memcontrol.h> | |
383 | ||
384 | /* | |
c22d70a1 RG |
385 | * cgwb_lock protects bdi->cgwb_tree, blkcg->cgwb_list, offline_cgwbs and |
386 | * memcg->cgwb_list. bdi->cgwb_tree is also RCU protected. | |
52ebea74 TH |
387 | */ |
388 | static DEFINE_SPINLOCK(cgwb_lock); | |
f1834646 | 389 | static struct workqueue_struct *cgwb_release_wq; |
52ebea74 | 390 | |
c22d70a1 RG |
391 | static LIST_HEAD(offline_cgwbs); |
392 | static void cleanup_offline_cgwbs_workfn(struct work_struct *work); | |
393 | static DECLARE_WORK(cleanup_offline_cgwbs_work, cleanup_offline_cgwbs_workfn); | |
394 | ||
52ebea74 TH |
395 | static void cgwb_release_workfn(struct work_struct *work) |
396 | { | |
397 | struct bdi_writeback *wb = container_of(work, struct bdi_writeback, | |
398 | release_work); | |
59b57717 | 399 | struct blkcg *blkcg = css_to_blkcg(wb->blkcg_css); |
52ebea74 | 400 | |
3ee7e869 | 401 | mutex_lock(&wb->bdi->cgwb_release_mutex); |
52ebea74 TH |
402 | wb_shutdown(wb); |
403 | ||
404 | css_put(wb->memcg_css); | |
405 | css_put(wb->blkcg_css); | |
3ee7e869 | 406 | mutex_unlock(&wb->bdi->cgwb_release_mutex); |
52ebea74 | 407 | |
d866dbf6 TH |
408 | /* triggers blkg destruction if no online users left */ |
409 | blkcg_unpin_online(blkcg); | |
59b57717 | 410 | |
841710aa | 411 | fprop_local_destroy_percpu(&wb->memcg_completions); |
c22d70a1 RG |
412 | |
413 | spin_lock_irq(&cgwb_lock); | |
414 | list_del(&wb->offline_node); | |
415 | spin_unlock_irq(&cgwb_lock); | |
416 | ||
b43a9e76 | 417 | percpu_ref_exit(&wb->refcnt); |
52ebea74 | 418 | wb_exit(wb); |
f3b6a6df | 419 | WARN_ON_ONCE(!list_empty(&wb->b_attached)); |
52ebea74 | 420 | kfree_rcu(wb, rcu); |
52ebea74 TH |
421 | } |
422 | ||
423 | static void cgwb_release(struct percpu_ref *refcnt) | |
424 | { | |
425 | struct bdi_writeback *wb = container_of(refcnt, struct bdi_writeback, | |
426 | refcnt); | |
f1834646 | 427 | queue_work(cgwb_release_wq, &wb->release_work); |
52ebea74 TH |
428 | } |
429 | ||
430 | static void cgwb_kill(struct bdi_writeback *wb) | |
431 | { | |
432 | lockdep_assert_held(&cgwb_lock); | |
433 | ||
434 | WARN_ON(!radix_tree_delete(&wb->bdi->cgwb_tree, wb->memcg_css->id)); | |
435 | list_del(&wb->memcg_node); | |
436 | list_del(&wb->blkcg_node); | |
c22d70a1 | 437 | list_add(&wb->offline_node, &offline_cgwbs); |
52ebea74 TH |
438 | percpu_ref_kill(&wb->refcnt); |
439 | } | |
440 | ||
e8cb72b3 JK |
441 | static void cgwb_remove_from_bdi_list(struct bdi_writeback *wb) |
442 | { | |
443 | spin_lock_irq(&cgwb_lock); | |
444 | list_del_rcu(&wb->bdi_node); | |
445 | spin_unlock_irq(&cgwb_lock); | |
446 | } | |
447 | ||
52ebea74 TH |
448 | static int cgwb_create(struct backing_dev_info *bdi, |
449 | struct cgroup_subsys_state *memcg_css, gfp_t gfp) | |
450 | { | |
451 | struct mem_cgroup *memcg; | |
452 | struct cgroup_subsys_state *blkcg_css; | |
453 | struct blkcg *blkcg; | |
454 | struct list_head *memcg_cgwb_list, *blkcg_cgwb_list; | |
455 | struct bdi_writeback *wb; | |
456 | unsigned long flags; | |
457 | int ret = 0; | |
458 | ||
459 | memcg = mem_cgroup_from_css(memcg_css); | |
c165b3e3 | 460 | blkcg_css = cgroup_get_e_css(memcg_css->cgroup, &io_cgrp_subsys); |
52ebea74 | 461 | blkcg = css_to_blkcg(blkcg_css); |
9ccc3617 | 462 | memcg_cgwb_list = &memcg->cgwb_list; |
52ebea74 TH |
463 | blkcg_cgwb_list = &blkcg->cgwb_list; |
464 | ||
465 | /* look up again under lock and discard on blkcg mismatch */ | |
466 | spin_lock_irqsave(&cgwb_lock, flags); | |
467 | wb = radix_tree_lookup(&bdi->cgwb_tree, memcg_css->id); | |
468 | if (wb && wb->blkcg_css != blkcg_css) { | |
469 | cgwb_kill(wb); | |
470 | wb = NULL; | |
471 | } | |
472 | spin_unlock_irqrestore(&cgwb_lock, flags); | |
473 | if (wb) | |
474 | goto out_put; | |
475 | ||
476 | /* need to create a new one */ | |
477 | wb = kmalloc(sizeof(*wb), gfp); | |
0b045bd1 CJ |
478 | if (!wb) { |
479 | ret = -ENOMEM; | |
480 | goto out_put; | |
481 | } | |
52ebea74 | 482 | |
8c911f3d | 483 | ret = wb_init(wb, bdi, gfp); |
52ebea74 TH |
484 | if (ret) |
485 | goto err_free; | |
486 | ||
487 | ret = percpu_ref_init(&wb->refcnt, cgwb_release, 0, gfp); | |
488 | if (ret) | |
489 | goto err_wb_exit; | |
490 | ||
841710aa TH |
491 | ret = fprop_local_init_percpu(&wb->memcg_completions, gfp); |
492 | if (ret) | |
493 | goto err_ref_exit; | |
494 | ||
52ebea74 TH |
495 | wb->memcg_css = memcg_css; |
496 | wb->blkcg_css = blkcg_css; | |
f3b6a6df | 497 | INIT_LIST_HEAD(&wb->b_attached); |
52ebea74 TH |
498 | INIT_WORK(&wb->release_work, cgwb_release_workfn); |
499 | set_bit(WB_registered, &wb->state); | |
03ba3782 JA |
500 | |
501 | /* | |
52ebea74 TH |
502 | * The root wb determines the registered state of the whole bdi and |
503 | * memcg_cgwb_list and blkcg_cgwb_list's next pointers indicate | |
504 | * whether they're still online. Don't link @wb if any is dead. | |
505 | * See wb_memcg_offline() and wb_blkcg_offline(). | |
03ba3782 | 506 | */ |
52ebea74 TH |
507 | ret = -ENODEV; |
508 | spin_lock_irqsave(&cgwb_lock, flags); | |
509 | if (test_bit(WB_registered, &bdi->wb.state) && | |
510 | blkcg_cgwb_list->next && memcg_cgwb_list->next) { | |
511 | /* we might have raced another instance of this function */ | |
512 | ret = radix_tree_insert(&bdi->cgwb_tree, memcg_css->id, wb); | |
513 | if (!ret) { | |
b817525a | 514 | list_add_tail_rcu(&wb->bdi_node, &bdi->wb_list); |
52ebea74 TH |
515 | list_add(&wb->memcg_node, memcg_cgwb_list); |
516 | list_add(&wb->blkcg_node, blkcg_cgwb_list); | |
d866dbf6 | 517 | blkcg_pin_online(blkcg); |
52ebea74 TH |
518 | css_get(memcg_css); |
519 | css_get(blkcg_css); | |
520 | } | |
521 | } | |
522 | spin_unlock_irqrestore(&cgwb_lock, flags); | |
523 | if (ret) { | |
524 | if (ret == -EEXIST) | |
525 | ret = 0; | |
a13f35e8 | 526 | goto err_fprop_exit; |
52ebea74 TH |
527 | } |
528 | goto out_put; | |
529 | ||
841710aa TH |
530 | err_fprop_exit: |
531 | fprop_local_destroy_percpu(&wb->memcg_completions); | |
52ebea74 TH |
532 | err_ref_exit: |
533 | percpu_ref_exit(&wb->refcnt); | |
534 | err_wb_exit: | |
535 | wb_exit(wb); | |
536 | err_free: | |
537 | kfree(wb); | |
538 | out_put: | |
539 | css_put(blkcg_css); | |
540 | return ret; | |
66f3b8e2 JA |
541 | } |
542 | ||
52ebea74 | 543 | /** |
ed288dc0 | 544 | * wb_get_lookup - get wb for a given memcg |
52ebea74 TH |
545 | * @bdi: target bdi |
546 | * @memcg_css: cgroup_subsys_state of the target memcg (must have positive ref) | |
52ebea74 | 547 | * |
ed288dc0 TH |
548 | * Try to get the wb for @memcg_css on @bdi. The returned wb has its |
549 | * refcount incremented. | |
52ebea74 TH |
550 | * |
551 | * This function uses css_get() on @memcg_css and thus expects its refcnt | |
552 | * to be positive on invocation. IOW, rcu_read_lock() protection on | |
553 | * @memcg_css isn't enough. try_get it before calling this function. | |
554 | * | |
555 | * A wb is keyed by its associated memcg. As blkcg implicitly enables | |
556 | * memcg on the default hierarchy, memcg association is guaranteed to be | |
557 | * more specific (equal or descendant to the associated blkcg) and thus can | |
558 | * identify both the memcg and blkcg associations. | |
559 | * | |
560 | * Because the blkcg associated with a memcg may change as blkcg is enabled | |
561 | * and disabled closer to root in the hierarchy, each wb keeps track of | |
562 | * both the memcg and blkcg associated with it and verifies the blkcg on | |
563 | * each lookup. On mismatch, the existing wb is discarded and a new one is | |
564 | * created. | |
565 | */ | |
ed288dc0 TH |
566 | struct bdi_writeback *wb_get_lookup(struct backing_dev_info *bdi, |
567 | struct cgroup_subsys_state *memcg_css) | |
568 | { | |
569 | struct bdi_writeback *wb; | |
570 | ||
571 | if (!memcg_css->parent) | |
572 | return &bdi->wb; | |
573 | ||
574 | rcu_read_lock(); | |
575 | wb = radix_tree_lookup(&bdi->cgwb_tree, memcg_css->id); | |
576 | if (wb) { | |
577 | struct cgroup_subsys_state *blkcg_css; | |
578 | ||
579 | /* see whether the blkcg association has changed */ | |
580 | blkcg_css = cgroup_get_e_css(memcg_css->cgroup, &io_cgrp_subsys); | |
581 | if (unlikely(wb->blkcg_css != blkcg_css || !wb_tryget(wb))) | |
582 | wb = NULL; | |
583 | css_put(blkcg_css); | |
584 | } | |
585 | rcu_read_unlock(); | |
586 | ||
587 | return wb; | |
588 | } | |
589 | ||
590 | /** | |
591 | * wb_get_create - get wb for a given memcg, create if necessary | |
592 | * @bdi: target bdi | |
593 | * @memcg_css: cgroup_subsys_state of the target memcg (must have positive ref) | |
594 | * @gfp: allocation mask to use | |
595 | * | |
596 | * Try to get the wb for @memcg_css on @bdi. If it doesn't exist, try to | |
597 | * create one. See wb_get_lookup() for more details. | |
598 | */ | |
52ebea74 TH |
599 | struct bdi_writeback *wb_get_create(struct backing_dev_info *bdi, |
600 | struct cgroup_subsys_state *memcg_css, | |
601 | gfp_t gfp) | |
6467716a | 602 | { |
52ebea74 TH |
603 | struct bdi_writeback *wb; |
604 | ||
c1ca59a1 | 605 | might_alloc(gfp); |
52ebea74 TH |
606 | |
607 | if (!memcg_css->parent) | |
608 | return &bdi->wb; | |
609 | ||
610 | do { | |
ed288dc0 | 611 | wb = wb_get_lookup(bdi, memcg_css); |
52ebea74 TH |
612 | } while (!wb && !cgwb_create(bdi, memcg_css, gfp)); |
613 | ||
614 | return wb; | |
615 | } | |
6467716a | 616 | |
a13f35e8 | 617 | static int cgwb_bdi_init(struct backing_dev_info *bdi) |
52ebea74 | 618 | { |
a13f35e8 TH |
619 | int ret; |
620 | ||
52ebea74 | 621 | INIT_RADIX_TREE(&bdi->cgwb_tree, GFP_ATOMIC); |
3ee7e869 | 622 | mutex_init(&bdi->cgwb_release_mutex); |
7fc5854f | 623 | init_rwsem(&bdi->wb_switch_rwsem); |
a13f35e8 | 624 | |
8c911f3d | 625 | ret = wb_init(&bdi->wb, bdi, GFP_KERNEL); |
a13f35e8 | 626 | if (!ret) { |
7d828602 | 627 | bdi->wb.memcg_css = &root_mem_cgroup->css; |
a13f35e8 TH |
628 | bdi->wb.blkcg_css = blkcg_root_css; |
629 | } | |
630 | return ret; | |
6467716a AB |
631 | } |
632 | ||
b1c51afc | 633 | static void cgwb_bdi_unregister(struct backing_dev_info *bdi) |
52ebea74 TH |
634 | { |
635 | struct radix_tree_iter iter; | |
636 | void **slot; | |
5318ce7d | 637 | struct bdi_writeback *wb; |
52ebea74 TH |
638 | |
639 | WARN_ON(test_bit(WB_registered, &bdi->wb.state)); | |
640 | ||
641 | spin_lock_irq(&cgwb_lock); | |
642 | radix_tree_for_each_slot(slot, &bdi->cgwb_tree, &iter, 0) | |
643 | cgwb_kill(*slot); | |
3ee7e869 | 644 | spin_unlock_irq(&cgwb_lock); |
5318ce7d | 645 | |
3ee7e869 JK |
646 | mutex_lock(&bdi->cgwb_release_mutex); |
647 | spin_lock_irq(&cgwb_lock); | |
5318ce7d JK |
648 | while (!list_empty(&bdi->wb_list)) { |
649 | wb = list_first_entry(&bdi->wb_list, struct bdi_writeback, | |
650 | bdi_node); | |
651 | spin_unlock_irq(&cgwb_lock); | |
652 | wb_shutdown(wb); | |
653 | spin_lock_irq(&cgwb_lock); | |
654 | } | |
52ebea74 | 655 | spin_unlock_irq(&cgwb_lock); |
3ee7e869 | 656 | mutex_unlock(&bdi->cgwb_release_mutex); |
52ebea74 TH |
657 | } |
658 | ||
c22d70a1 RG |
659 | /* |
660 | * cleanup_offline_cgwbs_workfn - try to release dying cgwbs | |
661 | * | |
662 | * Try to release dying cgwbs by switching attached inodes to the nearest | |
663 | * living ancestor's writeback. Processed wbs are placed at the end | |
664 | * of the list to guarantee the forward progress. | |
665 | */ | |
666 | static void cleanup_offline_cgwbs_workfn(struct work_struct *work) | |
667 | { | |
668 | struct bdi_writeback *wb; | |
669 | LIST_HEAD(processed); | |
670 | ||
671 | spin_lock_irq(&cgwb_lock); | |
672 | ||
673 | while (!list_empty(&offline_cgwbs)) { | |
674 | wb = list_first_entry(&offline_cgwbs, struct bdi_writeback, | |
675 | offline_node); | |
676 | list_move(&wb->offline_node, &processed); | |
677 | ||
678 | /* | |
679 | * If wb is dirty, cleaning up the writeback by switching | |
680 | * attached inodes will result in an effective removal of any | |
681 | * bandwidth restrictions, which isn't the goal. Instead, | |
682 | * it can be postponed until the next time, when all io | |
683 | * will be likely completed. If in the meantime some inodes | |
684 | * will get re-dirtied, they should be eventually switched to | |
685 | * a new cgwb. | |
686 | */ | |
687 | if (wb_has_dirty_io(wb)) | |
688 | continue; | |
689 | ||
690 | if (!wb_tryget(wb)) | |
691 | continue; | |
692 | ||
693 | spin_unlock_irq(&cgwb_lock); | |
694 | while (cleanup_offline_cgwb(wb)) | |
695 | cond_resched(); | |
696 | spin_lock_irq(&cgwb_lock); | |
697 | ||
698 | wb_put(wb); | |
699 | } | |
700 | ||
701 | if (!list_empty(&processed)) | |
702 | list_splice_tail(&processed, &offline_cgwbs); | |
703 | ||
704 | spin_unlock_irq(&cgwb_lock); | |
705 | } | |
706 | ||
52ebea74 TH |
707 | /** |
708 | * wb_memcg_offline - kill all wb's associated with a memcg being offlined | |
709 | * @memcg: memcg being offlined | |
710 | * | |
711 | * Also prevents creation of any new wb's associated with @memcg. | |
e98be2d5 | 712 | */ |
52ebea74 TH |
713 | void wb_memcg_offline(struct mem_cgroup *memcg) |
714 | { | |
9ccc3617 | 715 | struct list_head *memcg_cgwb_list = &memcg->cgwb_list; |
52ebea74 TH |
716 | struct bdi_writeback *wb, *next; |
717 | ||
718 | spin_lock_irq(&cgwb_lock); | |
719 | list_for_each_entry_safe(wb, next, memcg_cgwb_list, memcg_node) | |
720 | cgwb_kill(wb); | |
721 | memcg_cgwb_list->next = NULL; /* prevent new wb's */ | |
722 | spin_unlock_irq(&cgwb_lock); | |
c22d70a1 RG |
723 | |
724 | queue_work(system_unbound_wq, &cleanup_offline_cgwbs_work); | |
52ebea74 TH |
725 | } |
726 | ||
727 | /** | |
728 | * wb_blkcg_offline - kill all wb's associated with a blkcg being offlined | |
729 | * @blkcg: blkcg being offlined | |
730 | * | |
731 | * Also prevents creation of any new wb's associated with @blkcg. | |
732 | */ | |
733 | void wb_blkcg_offline(struct blkcg *blkcg) | |
734 | { | |
52ebea74 TH |
735 | struct bdi_writeback *wb, *next; |
736 | ||
737 | spin_lock_irq(&cgwb_lock); | |
738 | list_for_each_entry_safe(wb, next, &blkcg->cgwb_list, blkcg_node) | |
739 | cgwb_kill(wb); | |
740 | blkcg->cgwb_list.next = NULL; /* prevent new wb's */ | |
741 | spin_unlock_irq(&cgwb_lock); | |
742 | } | |
743 | ||
e8cb72b3 JK |
744 | static void cgwb_bdi_register(struct backing_dev_info *bdi) |
745 | { | |
746 | spin_lock_irq(&cgwb_lock); | |
747 | list_add_tail_rcu(&bdi->wb.bdi_node, &bdi->wb_list); | |
748 | spin_unlock_irq(&cgwb_lock); | |
749 | } | |
750 | ||
f1834646 TH |
751 | static int __init cgwb_init(void) |
752 | { | |
753 | /* | |
754 | * There can be many concurrent release work items overwhelming | |
755 | * system_wq. Put them in a separate wq and limit concurrency. | |
756 | * There's no point in executing many of these in parallel. | |
757 | */ | |
758 | cgwb_release_wq = alloc_workqueue("cgwb_release", 0, 1); | |
759 | if (!cgwb_release_wq) | |
760 | return -ENOMEM; | |
761 | ||
762 | return 0; | |
763 | } | |
764 | subsys_initcall(cgwb_init); | |
765 | ||
52ebea74 TH |
766 | #else /* CONFIG_CGROUP_WRITEBACK */ |
767 | ||
a13f35e8 TH |
768 | static int cgwb_bdi_init(struct backing_dev_info *bdi) |
769 | { | |
8c911f3d | 770 | return wb_init(&bdi->wb, bdi, GFP_KERNEL); |
a13f35e8 TH |
771 | } |
772 | ||
b1c51afc | 773 | static void cgwb_bdi_unregister(struct backing_dev_info *bdi) { } |
df23de55 | 774 | |
e8cb72b3 JK |
775 | static void cgwb_bdi_register(struct backing_dev_info *bdi) |
776 | { | |
777 | list_add_tail_rcu(&bdi->wb.bdi_node, &bdi->wb_list); | |
778 | } | |
779 | ||
780 | static void cgwb_remove_from_bdi_list(struct bdi_writeback *wb) | |
781 | { | |
782 | list_del_rcu(&wb->bdi_node); | |
783 | } | |
784 | ||
52ebea74 | 785 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
e98be2d5 | 786 | |
2e82b84c | 787 | static int bdi_init(struct backing_dev_info *bdi) |
b2e8fb6e | 788 | { |
b817525a TH |
789 | int ret; |
790 | ||
cf0ca9fe PZ |
791 | bdi->dev = NULL; |
792 | ||
d03f6cdc | 793 | kref_init(&bdi->refcnt); |
189d3c4a | 794 | bdi->min_ratio = 0; |
a42dde04 | 795 | bdi->max_ratio = 100; |
eb608e3a | 796 | bdi->max_prop_frac = FPROP_FRAC_BASE; |
66f3b8e2 | 797 | INIT_LIST_HEAD(&bdi->bdi_list); |
b817525a | 798 | INIT_LIST_HEAD(&bdi->wb_list); |
cc395d7f | 799 | init_waitqueue_head(&bdi->wb_waitq); |
03ba3782 | 800 | |
b817525a TH |
801 | ret = cgwb_bdi_init(bdi); |
802 | ||
b817525a | 803 | return ret; |
b2e8fb6e | 804 | } |
e98be2d5 | 805 | |
aef33c2f | 806 | struct backing_dev_info *bdi_alloc(int node_id) |
d03f6cdc JK |
807 | { |
808 | struct backing_dev_info *bdi; | |
809 | ||
aef33c2f | 810 | bdi = kzalloc_node(sizeof(*bdi), GFP_KERNEL, node_id); |
d03f6cdc JK |
811 | if (!bdi) |
812 | return NULL; | |
813 | ||
814 | if (bdi_init(bdi)) { | |
815 | kfree(bdi); | |
816 | return NULL; | |
817 | } | |
f56753ac | 818 | bdi->capabilities = BDI_CAP_WRITEBACK | BDI_CAP_WRITEBACK_ACCT; |
55b2598e CH |
819 | bdi->ra_pages = VM_READAHEAD_PAGES; |
820 | bdi->io_pages = VM_READAHEAD_PAGES; | |
5ed964f8 | 821 | timer_setup(&bdi->laptop_mode_wb_timer, laptop_mode_timer_fn, 0); |
d03f6cdc JK |
822 | return bdi; |
823 | } | |
aef33c2f | 824 | EXPORT_SYMBOL(bdi_alloc); |
d03f6cdc | 825 | |
34f8fe50 TH |
826 | static struct rb_node **bdi_lookup_rb_node(u64 id, struct rb_node **parentp) |
827 | { | |
828 | struct rb_node **p = &bdi_tree.rb_node; | |
829 | struct rb_node *parent = NULL; | |
830 | struct backing_dev_info *bdi; | |
831 | ||
832 | lockdep_assert_held(&bdi_lock); | |
833 | ||
834 | while (*p) { | |
835 | parent = *p; | |
836 | bdi = rb_entry(parent, struct backing_dev_info, rb_node); | |
837 | ||
838 | if (bdi->id > id) | |
839 | p = &(*p)->rb_left; | |
840 | else if (bdi->id < id) | |
841 | p = &(*p)->rb_right; | |
842 | else | |
843 | break; | |
844 | } | |
845 | ||
846 | if (parentp) | |
847 | *parentp = parent; | |
848 | return p; | |
849 | } | |
850 | ||
851 | /** | |
852 | * bdi_get_by_id - lookup and get bdi from its id | |
853 | * @id: bdi id to lookup | |
854 | * | |
855 | * Find bdi matching @id and get it. Returns NULL if the matching bdi | |
856 | * doesn't exist or is already unregistered. | |
857 | */ | |
858 | struct backing_dev_info *bdi_get_by_id(u64 id) | |
859 | { | |
860 | struct backing_dev_info *bdi = NULL; | |
861 | struct rb_node **p; | |
862 | ||
863 | spin_lock_bh(&bdi_lock); | |
864 | p = bdi_lookup_rb_node(id, NULL); | |
865 | if (*p) { | |
866 | bdi = rb_entry(*p, struct backing_dev_info, rb_node); | |
867 | bdi_get(bdi); | |
868 | } | |
869 | spin_unlock_bh(&bdi_lock); | |
870 | ||
871 | return bdi; | |
872 | } | |
873 | ||
7c4cc300 | 874 | int bdi_register_va(struct backing_dev_info *bdi, const char *fmt, va_list args) |
46100071 | 875 | { |
46100071 | 876 | struct device *dev; |
34f8fe50 | 877 | struct rb_node *parent, **p; |
e98be2d5 | 878 | |
46100071 TH |
879 | if (bdi->dev) /* The driver needs to use separate queues per device */ |
880 | return 0; | |
e98be2d5 | 881 | |
6bd87eec CH |
882 | vsnprintf(bdi->dev_name, sizeof(bdi->dev_name), fmt, args); |
883 | dev = device_create(bdi_class, NULL, MKDEV(0, 0), bdi, bdi->dev_name); | |
46100071 TH |
884 | if (IS_ERR(dev)) |
885 | return PTR_ERR(dev); | |
04fbfdc1 | 886 | |
e8cb72b3 | 887 | cgwb_bdi_register(bdi); |
46100071 | 888 | bdi->dev = dev; |
b2e8fb6e | 889 | |
6d0e4827 | 890 | bdi_debug_register(bdi, dev_name(dev)); |
46100071 TH |
891 | set_bit(WB_registered, &bdi->wb.state); |
892 | ||
893 | spin_lock_bh(&bdi_lock); | |
34f8fe50 TH |
894 | |
895 | bdi->id = ++bdi_id_cursor; | |
896 | ||
897 | p = bdi_lookup_rb_node(bdi->id, &parent); | |
898 | rb_link_node(&bdi->rb_node, parent, p); | |
899 | rb_insert_color(&bdi->rb_node, &bdi_tree); | |
900 | ||
46100071 | 901 | list_add_tail_rcu(&bdi->bdi_list, &bdi_list); |
34f8fe50 | 902 | |
46100071 TH |
903 | spin_unlock_bh(&bdi_lock); |
904 | ||
905 | trace_writeback_bdi_register(bdi); | |
906 | return 0; | |
b2e8fb6e | 907 | } |
baf7a616 | 908 | |
7c4cc300 | 909 | int bdi_register(struct backing_dev_info *bdi, const char *fmt, ...) |
baf7a616 JK |
910 | { |
911 | va_list args; | |
912 | int ret; | |
913 | ||
914 | va_start(args, fmt); | |
7c4cc300 | 915 | ret = bdi_register_va(bdi, fmt, args); |
baf7a616 JK |
916 | va_end(args); |
917 | return ret; | |
918 | } | |
46100071 | 919 | EXPORT_SYMBOL(bdi_register); |
b2e8fb6e | 920 | |
3c5d202b | 921 | void bdi_set_owner(struct backing_dev_info *bdi, struct device *owner) |
df08c32c | 922 | { |
3c5d202b | 923 | WARN_ON_ONCE(bdi->owner); |
df08c32c DW |
924 | bdi->owner = owner; |
925 | get_device(owner); | |
df08c32c | 926 | } |
df08c32c | 927 | |
46100071 TH |
928 | /* |
929 | * Remove bdi from bdi_list, and ensure that it is no longer visible | |
930 | */ | |
931 | static void bdi_remove_from_list(struct backing_dev_info *bdi) | |
932 | { | |
933 | spin_lock_bh(&bdi_lock); | |
34f8fe50 | 934 | rb_erase(&bdi->rb_node, &bdi_tree); |
46100071 TH |
935 | list_del_rcu(&bdi->bdi_list); |
936 | spin_unlock_bh(&bdi_lock); | |
b2e8fb6e | 937 | |
46100071 TH |
938 | synchronize_rcu_expedited(); |
939 | } | |
cf0ca9fe | 940 | |
b02176f3 | 941 | void bdi_unregister(struct backing_dev_info *bdi) |
b2e8fb6e | 942 | { |
5ed964f8 CH |
943 | del_timer_sync(&bdi->laptop_mode_wb_timer); |
944 | ||
f0054bb1 TH |
945 | /* make sure nobody finds us on the bdi_list anymore */ |
946 | bdi_remove_from_list(bdi); | |
947 | wb_shutdown(&bdi->wb); | |
b1c51afc | 948 | cgwb_bdi_unregister(bdi); |
7a401a97 | 949 | |
c4db59d3 CH |
950 | if (bdi->dev) { |
951 | bdi_debug_unregister(bdi); | |
952 | device_unregister(bdi->dev); | |
953 | bdi->dev = NULL; | |
954 | } | |
df08c32c DW |
955 | |
956 | if (bdi->owner) { | |
957 | put_device(bdi->owner); | |
958 | bdi->owner = NULL; | |
959 | } | |
b02176f3 | 960 | } |
c4db59d3 | 961 | |
d03f6cdc JK |
962 | static void release_bdi(struct kref *ref) |
963 | { | |
964 | struct backing_dev_info *bdi = | |
965 | container_of(ref, struct backing_dev_info, refcnt); | |
966 | ||
5af110b2 JK |
967 | if (test_bit(WB_registered, &bdi->wb.state)) |
968 | bdi_unregister(bdi); | |
2e82b84c JK |
969 | WARN_ON_ONCE(bdi->dev); |
970 | wb_exit(&bdi->wb); | |
d03f6cdc JK |
971 | kfree(bdi); |
972 | } | |
973 | ||
974 | void bdi_put(struct backing_dev_info *bdi) | |
975 | { | |
976 | kref_put(&bdi->refcnt, release_bdi); | |
977 | } | |
62bf42ad | 978 | EXPORT_SYMBOL(bdi_put); |
d03f6cdc | 979 | |
eb7ae5e0 CH |
980 | const char *bdi_dev_name(struct backing_dev_info *bdi) |
981 | { | |
982 | if (!bdi || !bdi->dev) | |
983 | return bdi_unknown_name; | |
6bd87eec | 984 | return bdi->dev_name; |
eb7ae5e0 CH |
985 | } |
986 | EXPORT_SYMBOL_GPL(bdi_dev_name); | |
987 | ||
3fcfab16 AM |
988 | static wait_queue_head_t congestion_wqh[2] = { |
989 | __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]), | |
990 | __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1]) | |
991 | }; | |
ec8a6f26 | 992 | static atomic_t nr_wb_congested[2]; |
3fcfab16 | 993 | |
492d76b2 | 994 | void clear_bdi_congested(struct backing_dev_info *bdi, int sync) |
3fcfab16 | 995 | { |
1faa16d2 | 996 | wait_queue_head_t *wqh = &congestion_wqh[sync]; |
c877ef8a | 997 | enum wb_congested_state bit; |
3fcfab16 | 998 | |
4452226e | 999 | bit = sync ? WB_sync_congested : WB_async_congested; |
8c911f3d | 1000 | if (test_and_clear_bit(bit, &bdi->wb.congested)) |
ec8a6f26 | 1001 | atomic_dec(&nr_wb_congested[sync]); |
4e857c58 | 1002 | smp_mb__after_atomic(); |
3fcfab16 AM |
1003 | if (waitqueue_active(wqh)) |
1004 | wake_up(wqh); | |
1005 | } | |
492d76b2 | 1006 | EXPORT_SYMBOL(clear_bdi_congested); |
3fcfab16 | 1007 | |
492d76b2 | 1008 | void set_bdi_congested(struct backing_dev_info *bdi, int sync) |
3fcfab16 | 1009 | { |
c877ef8a | 1010 | enum wb_congested_state bit; |
3fcfab16 | 1011 | |
4452226e | 1012 | bit = sync ? WB_sync_congested : WB_async_congested; |
8c911f3d | 1013 | if (!test_and_set_bit(bit, &bdi->wb.congested)) |
ec8a6f26 | 1014 | atomic_inc(&nr_wb_congested[sync]); |
3fcfab16 | 1015 | } |
492d76b2 | 1016 | EXPORT_SYMBOL(set_bdi_congested); |
3fcfab16 AM |
1017 | |
1018 | /** | |
1019 | * congestion_wait - wait for a backing_dev to become uncongested | |
8aa7e847 | 1020 | * @sync: SYNC or ASYNC IO |
3fcfab16 AM |
1021 | * @timeout: timeout in jiffies |
1022 | * | |
1023 | * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit | |
1024 | * write congestion. If no backing_devs are congested then just wait for the | |
1025 | * next write to be completed. | |
1026 | */ | |
8aa7e847 | 1027 | long congestion_wait(int sync, long timeout) |
3fcfab16 AM |
1028 | { |
1029 | long ret; | |
52bb9198 | 1030 | unsigned long start = jiffies; |
3fcfab16 | 1031 | DEFINE_WAIT(wait); |
8aa7e847 | 1032 | wait_queue_head_t *wqh = &congestion_wqh[sync]; |
3fcfab16 AM |
1033 | |
1034 | prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); | |
1035 | ret = io_schedule_timeout(timeout); | |
1036 | finish_wait(wqh, &wait); | |
52bb9198 MG |
1037 | |
1038 | trace_writeback_congestion_wait(jiffies_to_usecs(timeout), | |
1039 | jiffies_to_usecs(jiffies - start)); | |
1040 | ||
3fcfab16 AM |
1041 | return ret; |
1042 | } | |
1043 | EXPORT_SYMBOL(congestion_wait); | |
04fbfdc1 | 1044 | |
0e093d99 | 1045 | /** |
599d0c95 | 1046 | * wait_iff_congested - Conditionally wait for a backing_dev to become uncongested or a pgdat to complete writes |
0e093d99 MG |
1047 | * @sync: SYNC or ASYNC IO |
1048 | * @timeout: timeout in jiffies | |
1049 | * | |
e3c1ac58 AR |
1050 | * In the event of a congested backing_dev (any backing_dev) this waits |
1051 | * for up to @timeout jiffies for either a BDI to exit congestion of the | |
1052 | * given @sync queue or a write to complete. | |
0e093d99 MG |
1053 | * |
1054 | * The return value is 0 if the sleep is for the full timeout. Otherwise, | |
1055 | * it is the number of jiffies that were still remaining when the function | |
1056 | * returned. return_value == timeout implies the function did not sleep. | |
1057 | */ | |
e3c1ac58 | 1058 | long wait_iff_congested(int sync, long timeout) |
0e093d99 MG |
1059 | { |
1060 | long ret; | |
1061 | unsigned long start = jiffies; | |
1062 | DEFINE_WAIT(wait); | |
1063 | wait_queue_head_t *wqh = &congestion_wqh[sync]; | |
1064 | ||
1065 | /* | |
e3c1ac58 | 1066 | * If there is no congestion, yield if necessary instead |
0e093d99 MG |
1067 | * of sleeping on the congestion queue |
1068 | */ | |
e3c1ac58 | 1069 | if (atomic_read(&nr_wb_congested[sync]) == 0) { |
ede37713 | 1070 | cond_resched(); |
599d0c95 | 1071 | |
0e093d99 MG |
1072 | /* In case we scheduled, work out time remaining */ |
1073 | ret = timeout - (jiffies - start); | |
1074 | if (ret < 0) | |
1075 | ret = 0; | |
1076 | ||
1077 | goto out; | |
1078 | } | |
1079 | ||
1080 | /* Sleep until uncongested or a write happens */ | |
1081 | prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); | |
1082 | ret = io_schedule_timeout(timeout); | |
1083 | finish_wait(wqh, &wait); | |
1084 | ||
1085 | out: | |
1086 | trace_writeback_wait_iff_congested(jiffies_to_usecs(timeout), | |
1087 | jiffies_to_usecs(jiffies - start)); | |
1088 | ||
1089 | return ret; | |
1090 | } | |
1091 | EXPORT_SYMBOL(wait_iff_congested); |