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> |
03ba3782 | 5 | #include <linux/kthread.h> |
e41d12f5 CH |
6 | #include <linux/backing-dev.h> |
7 | #include <linux/blk-cgroup.h> | |
03ba3782 | 8 | #include <linux/freezer.h> |
3fcfab16 | 9 | #include <linux/fs.h> |
26160158 | 10 | #include <linux/pagemap.h> |
03ba3782 | 11 | #include <linux/mm.h> |
c1ca59a1 | 12 | #include <linux/sched/mm.h> |
3fcfab16 AM |
13 | #include <linux/sched.h> |
14 | #include <linux/module.h> | |
cf0ca9fe PZ |
15 | #include <linux/writeback.h> |
16 | #include <linux/device.h> | |
455b2864 | 17 | #include <trace/events/writeback.h> |
cf0ca9fe | 18 | |
f56753ac | 19 | struct backing_dev_info noop_backing_dev_info; |
a212b105 | 20 | EXPORT_SYMBOL_GPL(noop_backing_dev_info); |
5129a469 | 21 | |
cf0ca9fe | 22 | static struct class *bdi_class; |
eb7ae5e0 | 23 | static const char *bdi_unknown_name = "(unknown)"; |
cfc4ba53 JA |
24 | |
25 | /* | |
34f8fe50 TH |
26 | * bdi_lock protects bdi_tree and updates to bdi_list. bdi_list has RCU |
27 | * reader side locking. | |
cfc4ba53 | 28 | */ |
03ba3782 | 29 | DEFINE_SPINLOCK(bdi_lock); |
34f8fe50 TH |
30 | static u64 bdi_id_cursor; |
31 | static struct rb_root bdi_tree = RB_ROOT; | |
66f3b8e2 | 32 | LIST_HEAD(bdi_list); |
03ba3782 | 33 | |
839a8e86 TH |
34 | /* bdi_wq serves all asynchronous writeback tasks */ |
35 | struct workqueue_struct *bdi_wq; | |
36 | ||
6986c3e2 BW |
37 | #define K(x) ((x) << (PAGE_SHIFT - 10)) |
38 | ||
76f1418b MS |
39 | #ifdef CONFIG_DEBUG_FS |
40 | #include <linux/debugfs.h> | |
41 | #include <linux/seq_file.h> | |
42 | ||
43 | static struct dentry *bdi_debug_root; | |
44 | ||
45 | static void bdi_debug_init(void) | |
46 | { | |
47 | bdi_debug_root = debugfs_create_dir("bdi", NULL); | |
48 | } | |
49 | ||
50 | static int bdi_debug_stats_show(struct seq_file *m, void *v) | |
51 | { | |
52 | struct backing_dev_info *bdi = m->private; | |
c1955ce3 | 53 | struct bdi_writeback *wb = &bdi->wb; |
364aeb28 DR |
54 | unsigned long background_thresh; |
55 | unsigned long dirty_thresh; | |
0d960a38 | 56 | unsigned long wb_thresh; |
0ae45f63 | 57 | unsigned long nr_dirty, nr_io, nr_more_io, nr_dirty_time; |
f09b00d3 JA |
58 | struct inode *inode; |
59 | ||
0ae45f63 | 60 | nr_dirty = nr_io = nr_more_io = nr_dirty_time = 0; |
f758eeab | 61 | spin_lock(&wb->list_lock); |
c7f54084 | 62 | list_for_each_entry(inode, &wb->b_dirty, i_io_list) |
c1955ce3 | 63 | nr_dirty++; |
c7f54084 | 64 | list_for_each_entry(inode, &wb->b_io, i_io_list) |
c1955ce3 | 65 | nr_io++; |
c7f54084 | 66 | list_for_each_entry(inode, &wb->b_more_io, i_io_list) |
c1955ce3 | 67 | nr_more_io++; |
c7f54084 | 68 | list_for_each_entry(inode, &wb->b_dirty_time, i_io_list) |
0ae45f63 TT |
69 | if (inode->i_state & I_DIRTY_TIME) |
70 | nr_dirty_time++; | |
f758eeab | 71 | spin_unlock(&wb->list_lock); |
76f1418b | 72 | |
16c4042f | 73 | global_dirty_limits(&background_thresh, &dirty_thresh); |
0d960a38 | 74 | wb_thresh = wb_calc_thresh(wb, dirty_thresh); |
76f1418b | 75 | |
76f1418b | 76 | seq_printf(m, |
00821b00 WF |
77 | "BdiWriteback: %10lu kB\n" |
78 | "BdiReclaimable: %10lu kB\n" | |
79 | "BdiDirtyThresh: %10lu kB\n" | |
80 | "DirtyThresh: %10lu kB\n" | |
81 | "BackgroundThresh: %10lu kB\n" | |
c8e28ce0 | 82 | "BdiDirtied: %10lu kB\n" |
00821b00 WF |
83 | "BdiWritten: %10lu kB\n" |
84 | "BdiWriteBandwidth: %10lu kBps\n" | |
85 | "b_dirty: %10lu\n" | |
86 | "b_io: %10lu\n" | |
87 | "b_more_io: %10lu\n" | |
0ae45f63 | 88 | "b_dirty_time: %10lu\n" |
00821b00 WF |
89 | "bdi_list: %10u\n" |
90 | "state: %10lx\n", | |
93f78d88 TH |
91 | (unsigned long) K(wb_stat(wb, WB_WRITEBACK)), |
92 | (unsigned long) K(wb_stat(wb, WB_RECLAIMABLE)), | |
0d960a38 | 93 | K(wb_thresh), |
f7d2b1ec JK |
94 | K(dirty_thresh), |
95 | K(background_thresh), | |
93f78d88 TH |
96 | (unsigned long) K(wb_stat(wb, WB_DIRTIED)), |
97 | (unsigned long) K(wb_stat(wb, WB_WRITTEN)), | |
a88a341a | 98 | (unsigned long) K(wb->write_bandwidth), |
f7d2b1ec JK |
99 | nr_dirty, |
100 | nr_io, | |
101 | nr_more_io, | |
0ae45f63 | 102 | nr_dirty_time, |
4452226e | 103 | !list_empty(&bdi->bdi_list), bdi->wb.state); |
76f1418b MS |
104 | |
105 | return 0; | |
106 | } | |
5ad35093 | 107 | DEFINE_SHOW_ATTRIBUTE(bdi_debug_stats); |
76f1418b | 108 | |
2d146b92 | 109 | static void bdi_debug_register(struct backing_dev_info *bdi, const char *name) |
76f1418b MS |
110 | { |
111 | bdi->debug_dir = debugfs_create_dir(name, bdi_debug_root); | |
97f07697 | 112 | |
2d146b92 GKH |
113 | debugfs_create_file("stats", 0444, bdi->debug_dir, bdi, |
114 | &bdi_debug_stats_fops); | |
76f1418b MS |
115 | } |
116 | ||
117 | static void bdi_debug_unregister(struct backing_dev_info *bdi) | |
118 | { | |
2d146b92 | 119 | debugfs_remove_recursive(bdi->debug_dir); |
76f1418b MS |
120 | } |
121 | #else | |
122 | static inline void bdi_debug_init(void) | |
123 | { | |
124 | } | |
2d146b92 | 125 | static inline void bdi_debug_register(struct backing_dev_info *bdi, |
76f1418b MS |
126 | const char *name) |
127 | { | |
128 | } | |
129 | static inline void bdi_debug_unregister(struct backing_dev_info *bdi) | |
130 | { | |
131 | } | |
132 | #endif | |
133 | ||
cf0ca9fe PZ |
134 | static ssize_t read_ahead_kb_store(struct device *dev, |
135 | struct device_attribute *attr, | |
136 | const char *buf, size_t count) | |
137 | { | |
138 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
cf0ca9fe | 139 | unsigned long read_ahead_kb; |
7034ed13 | 140 | ssize_t ret; |
cf0ca9fe | 141 | |
7034ed13 NJ |
142 | ret = kstrtoul(buf, 10, &read_ahead_kb); |
143 | if (ret < 0) | |
144 | return ret; | |
145 | ||
146 | bdi->ra_pages = read_ahead_kb >> (PAGE_SHIFT - 10); | |
147 | ||
148 | return count; | |
cf0ca9fe PZ |
149 | } |
150 | ||
cf0ca9fe PZ |
151 | #define BDI_SHOW(name, expr) \ |
152 | static ssize_t name##_show(struct device *dev, \ | |
5e4c0d86 | 153 | struct device_attribute *attr, char *buf) \ |
cf0ca9fe PZ |
154 | { \ |
155 | struct backing_dev_info *bdi = dev_get_drvdata(dev); \ | |
156 | \ | |
5e4c0d86 | 157 | return sysfs_emit(buf, "%lld\n", (long long)expr); \ |
d9e1241e GKH |
158 | } \ |
159 | static DEVICE_ATTR_RW(name); | |
cf0ca9fe PZ |
160 | |
161 | BDI_SHOW(read_ahead_kb, K(bdi->ra_pages)) | |
162 | ||
189d3c4a PZ |
163 | static ssize_t min_ratio_store(struct device *dev, |
164 | struct device_attribute *attr, const char *buf, size_t count) | |
165 | { | |
166 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
189d3c4a | 167 | unsigned int ratio; |
7034ed13 NJ |
168 | ssize_t ret; |
169 | ||
170 | ret = kstrtouint(buf, 10, &ratio); | |
171 | if (ret < 0) | |
172 | return ret; | |
173 | ||
174 | ret = bdi_set_min_ratio(bdi, ratio); | |
175 | if (!ret) | |
176 | ret = count; | |
189d3c4a | 177 | |
189d3c4a PZ |
178 | return ret; |
179 | } | |
180 | BDI_SHOW(min_ratio, bdi->min_ratio) | |
181 | ||
a42dde04 PZ |
182 | static ssize_t max_ratio_store(struct device *dev, |
183 | struct device_attribute *attr, const char *buf, size_t count) | |
184 | { | |
185 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
a42dde04 | 186 | unsigned int ratio; |
7034ed13 NJ |
187 | ssize_t ret; |
188 | ||
189 | ret = kstrtouint(buf, 10, &ratio); | |
190 | if (ret < 0) | |
191 | return ret; | |
192 | ||
193 | ret = bdi_set_max_ratio(bdi, ratio); | |
194 | if (!ret) | |
195 | ret = count; | |
a42dde04 | 196 | |
a42dde04 PZ |
197 | return ret; |
198 | } | |
199 | BDI_SHOW(max_ratio, bdi->max_ratio) | |
200 | ||
7d311cda DW |
201 | static ssize_t stable_pages_required_show(struct device *dev, |
202 | struct device_attribute *attr, | |
5e4c0d86 | 203 | char *buf) |
7d311cda | 204 | { |
1cb039f3 CH |
205 | dev_warn_once(dev, |
206 | "the stable_pages_required attribute has been removed. Use the stable_writes queue attribute instead.\n"); | |
5e4c0d86 | 207 | return sysfs_emit(buf, "%d\n", 0); |
7d311cda | 208 | } |
d9e1241e GKH |
209 | static DEVICE_ATTR_RO(stable_pages_required); |
210 | ||
211 | static struct attribute *bdi_dev_attrs[] = { | |
212 | &dev_attr_read_ahead_kb.attr, | |
213 | &dev_attr_min_ratio.attr, | |
214 | &dev_attr_max_ratio.attr, | |
215 | &dev_attr_stable_pages_required.attr, | |
216 | NULL, | |
cf0ca9fe | 217 | }; |
d9e1241e | 218 | ATTRIBUTE_GROUPS(bdi_dev); |
cf0ca9fe PZ |
219 | |
220 | static __init int bdi_class_init(void) | |
221 | { | |
222 | bdi_class = class_create(THIS_MODULE, "bdi"); | |
14421453 AB |
223 | if (IS_ERR(bdi_class)) |
224 | return PTR_ERR(bdi_class); | |
225 | ||
d9e1241e | 226 | bdi_class->dev_groups = bdi_dev_groups; |
76f1418b | 227 | bdi_debug_init(); |
d03f6cdc | 228 | |
cf0ca9fe PZ |
229 | return 0; |
230 | } | |
76f1418b | 231 | postcore_initcall(bdi_class_init); |
cf0ca9fe | 232 | |
2e82b84c JK |
233 | static int bdi_init(struct backing_dev_info *bdi); |
234 | ||
26160158 JA |
235 | static int __init default_bdi_init(void) |
236 | { | |
237 | int err; | |
238 | ||
a2b90f11 MW |
239 | bdi_wq = alloc_workqueue("writeback", WQ_MEM_RECLAIM | WQ_UNBOUND | |
240 | WQ_SYSFS, 0); | |
839a8e86 TH |
241 | if (!bdi_wq) |
242 | return -ENOMEM; | |
243 | ||
976e48f8 | 244 | err = bdi_init(&noop_backing_dev_info); |
26160158 JA |
245 | |
246 | return err; | |
247 | } | |
248 | subsys_initcall(default_bdi_init); | |
249 | ||
6467716a | 250 | /* |
f0054bb1 | 251 | * This function is used when the first inode for this wb is marked dirty. It |
6467716a AB |
252 | * wakes-up the corresponding bdi thread which should then take care of the |
253 | * periodic background write-out of dirty inodes. Since the write-out would | |
254 | * starts only 'dirty_writeback_interval' centisecs from now anyway, we just | |
255 | * set up a timer which wakes the bdi thread up later. | |
256 | * | |
257 | * Note, we wouldn't bother setting up the timer, but this function is on the | |
258 | * fast-path (used by '__mark_inode_dirty()'), so we save few context switches | |
259 | * by delaying the wake-up. | |
6ca738d6 DB |
260 | * |
261 | * We have to be careful not to postpone flush work if it is scheduled for | |
262 | * earlier. Thus we use queue_delayed_work(). | |
6467716a | 263 | */ |
f0054bb1 | 264 | void wb_wakeup_delayed(struct bdi_writeback *wb) |
6467716a AB |
265 | { |
266 | unsigned long timeout; | |
267 | ||
268 | timeout = msecs_to_jiffies(dirty_writeback_interval * 10); | |
f0054bb1 TH |
269 | spin_lock_bh(&wb->work_lock); |
270 | if (test_bit(WB_registered, &wb->state)) | |
271 | queue_delayed_work(bdi_wq, &wb->dwork, timeout); | |
272 | spin_unlock_bh(&wb->work_lock); | |
03ba3782 JA |
273 | } |
274 | ||
45a2966f JK |
275 | static void wb_update_bandwidth_workfn(struct work_struct *work) |
276 | { | |
277 | struct bdi_writeback *wb = container_of(to_delayed_work(work), | |
278 | struct bdi_writeback, bw_dwork); | |
279 | ||
280 | wb_update_bandwidth(wb); | |
281 | } | |
282 | ||
cfc4ba53 | 283 | /* |
a88a341a | 284 | * Initial write bandwidth: 100 MB/s |
cfc4ba53 | 285 | */ |
a88a341a | 286 | #define INIT_BW (100 << (20 - PAGE_SHIFT)) |
cfc4ba53 | 287 | |
8395cd9f | 288 | static int wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi, |
8c911f3d | 289 | gfp_t gfp) |
cf0ca9fe | 290 | { |
93f78d88 | 291 | int i, err; |
cf0ca9fe | 292 | |
6467716a | 293 | memset(wb, 0, sizeof(*wb)); |
f1d0b063 | 294 | |
6467716a AB |
295 | wb->bdi = bdi; |
296 | wb->last_old_flush = jiffies; | |
297 | INIT_LIST_HEAD(&wb->b_dirty); | |
298 | INIT_LIST_HEAD(&wb->b_io); | |
299 | INIT_LIST_HEAD(&wb->b_more_io); | |
0ae45f63 | 300 | INIT_LIST_HEAD(&wb->b_dirty_time); |
f758eeab | 301 | spin_lock_init(&wb->list_lock); |
66f3b8e2 | 302 | |
633a2abb | 303 | atomic_set(&wb->writeback_inodes, 0); |
a88a341a TH |
304 | wb->bw_time_stamp = jiffies; |
305 | wb->balanced_dirty_ratelimit = INIT_BW; | |
306 | wb->dirty_ratelimit = INIT_BW; | |
307 | wb->write_bandwidth = INIT_BW; | |
308 | wb->avg_write_bandwidth = INIT_BW; | |
cf0ca9fe | 309 | |
f0054bb1 TH |
310 | spin_lock_init(&wb->work_lock); |
311 | INIT_LIST_HEAD(&wb->work_list); | |
312 | INIT_DELAYED_WORK(&wb->dwork, wb_workfn); | |
45a2966f | 313 | INIT_DELAYED_WORK(&wb->bw_dwork, wb_update_bandwidth_workfn); |
b57d74af | 314 | wb->dirty_sleep = jiffies; |
c284de61 | 315 | |
8395cd9f | 316 | err = fprop_local_init_percpu(&wb->completions, gfp); |
a88a341a | 317 | if (err) |
efee1713 | 318 | return err; |
c284de61 | 319 | |
93f78d88 | 320 | for (i = 0; i < NR_WB_STAT_ITEMS; i++) { |
8395cd9f | 321 | err = percpu_counter_init(&wb->stat[i], 0, gfp); |
a13f35e8 TH |
322 | if (err) |
323 | goto out_destroy_stat; | |
93f78d88 | 324 | } |
cf0ca9fe | 325 | |
93f78d88 | 326 | return 0; |
a13f35e8 TH |
327 | |
328 | out_destroy_stat: | |
078c6c3a | 329 | while (i--) |
a13f35e8 TH |
330 | percpu_counter_destroy(&wb->stat[i]); |
331 | fprop_local_destroy_percpu(&wb->completions); | |
a13f35e8 | 332 | return err; |
cf0ca9fe | 333 | } |
cf0ca9fe | 334 | |
e8cb72b3 JK |
335 | static void cgwb_remove_from_bdi_list(struct bdi_writeback *wb); |
336 | ||
03ba3782 JA |
337 | /* |
338 | * Remove bdi from the global list and shutdown any threads we have running | |
339 | */ | |
46100071 | 340 | static void wb_shutdown(struct bdi_writeback *wb) |
66f3b8e2 | 341 | { |
c4db59d3 | 342 | /* Make sure nobody queues further work */ |
46100071 TH |
343 | spin_lock_bh(&wb->work_lock); |
344 | if (!test_and_clear_bit(WB_registered, &wb->state)) { | |
345 | spin_unlock_bh(&wb->work_lock); | |
03ba3782 | 346 | return; |
c4db59d3 | 347 | } |
46100071 | 348 | spin_unlock_bh(&wb->work_lock); |
03ba3782 | 349 | |
e8cb72b3 | 350 | cgwb_remove_from_bdi_list(wb); |
03ba3782 | 351 | /* |
46100071 TH |
352 | * Drain work list and shutdown the delayed_work. !WB_registered |
353 | * tells wb_workfn() that @wb is dying and its work_list needs to | |
354 | * be drained no matter what. | |
03ba3782 | 355 | */ |
46100071 TH |
356 | mod_delayed_work(bdi_wq, &wb->dwork, 0); |
357 | flush_delayed_work(&wb->dwork); | |
358 | WARN_ON(!list_empty(&wb->work_list)); | |
45a2966f | 359 | flush_delayed_work(&wb->bw_dwork); |
46100071 TH |
360 | } |
361 | ||
f0054bb1 | 362 | static void wb_exit(struct bdi_writeback *wb) |
93f78d88 TH |
363 | { |
364 | int i; | |
365 | ||
366 | WARN_ON(delayed_work_pending(&wb->dwork)); | |
367 | ||
368 | for (i = 0; i < NR_WB_STAT_ITEMS; i++) | |
369 | percpu_counter_destroy(&wb->stat[i]); | |
6467716a | 370 | |
a88a341a TH |
371 | fprop_local_destroy_percpu(&wb->completions); |
372 | } | |
e98be2d5 | 373 | |
52ebea74 TH |
374 | #ifdef CONFIG_CGROUP_WRITEBACK |
375 | ||
376 | #include <linux/memcontrol.h> | |
377 | ||
378 | /* | |
c22d70a1 RG |
379 | * cgwb_lock protects bdi->cgwb_tree, blkcg->cgwb_list, offline_cgwbs and |
380 | * memcg->cgwb_list. bdi->cgwb_tree is also RCU protected. | |
52ebea74 TH |
381 | */ |
382 | static DEFINE_SPINLOCK(cgwb_lock); | |
f1834646 | 383 | static struct workqueue_struct *cgwb_release_wq; |
52ebea74 | 384 | |
c22d70a1 RG |
385 | static LIST_HEAD(offline_cgwbs); |
386 | static void cleanup_offline_cgwbs_workfn(struct work_struct *work); | |
387 | static DECLARE_WORK(cleanup_offline_cgwbs_work, cleanup_offline_cgwbs_workfn); | |
388 | ||
52ebea74 TH |
389 | static void cgwb_release_workfn(struct work_struct *work) |
390 | { | |
391 | struct bdi_writeback *wb = container_of(work, struct bdi_writeback, | |
392 | release_work); | |
59b57717 | 393 | struct blkcg *blkcg = css_to_blkcg(wb->blkcg_css); |
efee1713 | 394 | struct backing_dev_info *bdi = wb->bdi; |
52ebea74 | 395 | |
3ee7e869 | 396 | mutex_lock(&wb->bdi->cgwb_release_mutex); |
52ebea74 TH |
397 | wb_shutdown(wb); |
398 | ||
399 | css_put(wb->memcg_css); | |
400 | css_put(wb->blkcg_css); | |
3ee7e869 | 401 | mutex_unlock(&wb->bdi->cgwb_release_mutex); |
52ebea74 | 402 | |
d866dbf6 TH |
403 | /* triggers blkg destruction if no online users left */ |
404 | blkcg_unpin_online(blkcg); | |
59b57717 | 405 | |
841710aa | 406 | fprop_local_destroy_percpu(&wb->memcg_completions); |
c22d70a1 RG |
407 | |
408 | spin_lock_irq(&cgwb_lock); | |
409 | list_del(&wb->offline_node); | |
410 | spin_unlock_irq(&cgwb_lock); | |
411 | ||
b43a9e76 | 412 | percpu_ref_exit(&wb->refcnt); |
52ebea74 | 413 | wb_exit(wb); |
efee1713 | 414 | bdi_put(bdi); |
f3b6a6df | 415 | WARN_ON_ONCE(!list_empty(&wb->b_attached)); |
52ebea74 | 416 | kfree_rcu(wb, rcu); |
52ebea74 TH |
417 | } |
418 | ||
419 | static void cgwb_release(struct percpu_ref *refcnt) | |
420 | { | |
421 | struct bdi_writeback *wb = container_of(refcnt, struct bdi_writeback, | |
422 | refcnt); | |
f1834646 | 423 | queue_work(cgwb_release_wq, &wb->release_work); |
52ebea74 TH |
424 | } |
425 | ||
426 | static void cgwb_kill(struct bdi_writeback *wb) | |
427 | { | |
428 | lockdep_assert_held(&cgwb_lock); | |
429 | ||
430 | WARN_ON(!radix_tree_delete(&wb->bdi->cgwb_tree, wb->memcg_css->id)); | |
431 | list_del(&wb->memcg_node); | |
432 | list_del(&wb->blkcg_node); | |
c22d70a1 | 433 | list_add(&wb->offline_node, &offline_cgwbs); |
52ebea74 TH |
434 | percpu_ref_kill(&wb->refcnt); |
435 | } | |
436 | ||
e8cb72b3 JK |
437 | static void cgwb_remove_from_bdi_list(struct bdi_writeback *wb) |
438 | { | |
439 | spin_lock_irq(&cgwb_lock); | |
440 | list_del_rcu(&wb->bdi_node); | |
441 | spin_unlock_irq(&cgwb_lock); | |
442 | } | |
443 | ||
52ebea74 TH |
444 | static int cgwb_create(struct backing_dev_info *bdi, |
445 | struct cgroup_subsys_state *memcg_css, gfp_t gfp) | |
446 | { | |
447 | struct mem_cgroup *memcg; | |
448 | struct cgroup_subsys_state *blkcg_css; | |
449 | struct blkcg *blkcg; | |
450 | struct list_head *memcg_cgwb_list, *blkcg_cgwb_list; | |
451 | struct bdi_writeback *wb; | |
452 | unsigned long flags; | |
453 | int ret = 0; | |
454 | ||
455 | memcg = mem_cgroup_from_css(memcg_css); | |
c165b3e3 | 456 | blkcg_css = cgroup_get_e_css(memcg_css->cgroup, &io_cgrp_subsys); |
52ebea74 | 457 | blkcg = css_to_blkcg(blkcg_css); |
9ccc3617 | 458 | memcg_cgwb_list = &memcg->cgwb_list; |
52ebea74 TH |
459 | blkcg_cgwb_list = &blkcg->cgwb_list; |
460 | ||
461 | /* look up again under lock and discard on blkcg mismatch */ | |
462 | spin_lock_irqsave(&cgwb_lock, flags); | |
463 | wb = radix_tree_lookup(&bdi->cgwb_tree, memcg_css->id); | |
464 | if (wb && wb->blkcg_css != blkcg_css) { | |
465 | cgwb_kill(wb); | |
466 | wb = NULL; | |
467 | } | |
468 | spin_unlock_irqrestore(&cgwb_lock, flags); | |
469 | if (wb) | |
470 | goto out_put; | |
471 | ||
472 | /* need to create a new one */ | |
473 | wb = kmalloc(sizeof(*wb), gfp); | |
0b045bd1 CJ |
474 | if (!wb) { |
475 | ret = -ENOMEM; | |
476 | goto out_put; | |
477 | } | |
52ebea74 | 478 | |
8c911f3d | 479 | ret = wb_init(wb, bdi, gfp); |
52ebea74 TH |
480 | if (ret) |
481 | goto err_free; | |
482 | ||
483 | ret = percpu_ref_init(&wb->refcnt, cgwb_release, 0, gfp); | |
484 | if (ret) | |
485 | goto err_wb_exit; | |
486 | ||
841710aa TH |
487 | ret = fprop_local_init_percpu(&wb->memcg_completions, gfp); |
488 | if (ret) | |
489 | goto err_ref_exit; | |
490 | ||
52ebea74 TH |
491 | wb->memcg_css = memcg_css; |
492 | wb->blkcg_css = blkcg_css; | |
f3b6a6df | 493 | INIT_LIST_HEAD(&wb->b_attached); |
52ebea74 TH |
494 | INIT_WORK(&wb->release_work, cgwb_release_workfn); |
495 | set_bit(WB_registered, &wb->state); | |
efee1713 | 496 | bdi_get(bdi); |
03ba3782 JA |
497 | |
498 | /* | |
52ebea74 TH |
499 | * The root wb determines the registered state of the whole bdi and |
500 | * memcg_cgwb_list and blkcg_cgwb_list's next pointers indicate | |
501 | * whether they're still online. Don't link @wb if any is dead. | |
502 | * See wb_memcg_offline() and wb_blkcg_offline(). | |
03ba3782 | 503 | */ |
52ebea74 TH |
504 | ret = -ENODEV; |
505 | spin_lock_irqsave(&cgwb_lock, flags); | |
506 | if (test_bit(WB_registered, &bdi->wb.state) && | |
507 | blkcg_cgwb_list->next && memcg_cgwb_list->next) { | |
508 | /* we might have raced another instance of this function */ | |
509 | ret = radix_tree_insert(&bdi->cgwb_tree, memcg_css->id, wb); | |
510 | if (!ret) { | |
b817525a | 511 | list_add_tail_rcu(&wb->bdi_node, &bdi->wb_list); |
52ebea74 TH |
512 | list_add(&wb->memcg_node, memcg_cgwb_list); |
513 | list_add(&wb->blkcg_node, blkcg_cgwb_list); | |
d866dbf6 | 514 | blkcg_pin_online(blkcg); |
52ebea74 TH |
515 | css_get(memcg_css); |
516 | css_get(blkcg_css); | |
517 | } | |
518 | } | |
519 | spin_unlock_irqrestore(&cgwb_lock, flags); | |
520 | if (ret) { | |
521 | if (ret == -EEXIST) | |
522 | ret = 0; | |
a13f35e8 | 523 | goto err_fprop_exit; |
52ebea74 TH |
524 | } |
525 | goto out_put; | |
526 | ||
841710aa | 527 | err_fprop_exit: |
efee1713 | 528 | bdi_put(bdi); |
841710aa | 529 | fprop_local_destroy_percpu(&wb->memcg_completions); |
52ebea74 TH |
530 | err_ref_exit: |
531 | percpu_ref_exit(&wb->refcnt); | |
532 | err_wb_exit: | |
533 | wb_exit(wb); | |
534 | err_free: | |
535 | kfree(wb); | |
536 | out_put: | |
537 | css_put(blkcg_css); | |
538 | return ret; | |
66f3b8e2 JA |
539 | } |
540 | ||
52ebea74 | 541 | /** |
ed288dc0 | 542 | * wb_get_lookup - get wb for a given memcg |
52ebea74 TH |
543 | * @bdi: target bdi |
544 | * @memcg_css: cgroup_subsys_state of the target memcg (must have positive ref) | |
52ebea74 | 545 | * |
ed288dc0 TH |
546 | * Try to get the wb for @memcg_css on @bdi. The returned wb has its |
547 | * refcount incremented. | |
52ebea74 TH |
548 | * |
549 | * This function uses css_get() on @memcg_css and thus expects its refcnt | |
550 | * to be positive on invocation. IOW, rcu_read_lock() protection on | |
551 | * @memcg_css isn't enough. try_get it before calling this function. | |
552 | * | |
553 | * A wb is keyed by its associated memcg. As blkcg implicitly enables | |
554 | * memcg on the default hierarchy, memcg association is guaranteed to be | |
555 | * more specific (equal or descendant to the associated blkcg) and thus can | |
556 | * identify both the memcg and blkcg associations. | |
557 | * | |
558 | * Because the blkcg associated with a memcg may change as blkcg is enabled | |
559 | * and disabled closer to root in the hierarchy, each wb keeps track of | |
560 | * both the memcg and blkcg associated with it and verifies the blkcg on | |
561 | * each lookup. On mismatch, the existing wb is discarded and a new one is | |
562 | * created. | |
563 | */ | |
ed288dc0 TH |
564 | struct bdi_writeback *wb_get_lookup(struct backing_dev_info *bdi, |
565 | struct cgroup_subsys_state *memcg_css) | |
566 | { | |
567 | struct bdi_writeback *wb; | |
568 | ||
569 | if (!memcg_css->parent) | |
570 | return &bdi->wb; | |
571 | ||
572 | rcu_read_lock(); | |
573 | wb = radix_tree_lookup(&bdi->cgwb_tree, memcg_css->id); | |
574 | if (wb) { | |
575 | struct cgroup_subsys_state *blkcg_css; | |
576 | ||
577 | /* see whether the blkcg association has changed */ | |
578 | blkcg_css = cgroup_get_e_css(memcg_css->cgroup, &io_cgrp_subsys); | |
579 | if (unlikely(wb->blkcg_css != blkcg_css || !wb_tryget(wb))) | |
580 | wb = NULL; | |
581 | css_put(blkcg_css); | |
582 | } | |
583 | rcu_read_unlock(); | |
584 | ||
585 | return wb; | |
586 | } | |
587 | ||
588 | /** | |
589 | * wb_get_create - get wb for a given memcg, create if necessary | |
590 | * @bdi: target bdi | |
591 | * @memcg_css: cgroup_subsys_state of the target memcg (must have positive ref) | |
592 | * @gfp: allocation mask to use | |
593 | * | |
594 | * Try to get the wb for @memcg_css on @bdi. If it doesn't exist, try to | |
595 | * create one. See wb_get_lookup() for more details. | |
596 | */ | |
52ebea74 TH |
597 | struct bdi_writeback *wb_get_create(struct backing_dev_info *bdi, |
598 | struct cgroup_subsys_state *memcg_css, | |
599 | gfp_t gfp) | |
6467716a | 600 | { |
52ebea74 TH |
601 | struct bdi_writeback *wb; |
602 | ||
c1ca59a1 | 603 | might_alloc(gfp); |
52ebea74 TH |
604 | |
605 | if (!memcg_css->parent) | |
606 | return &bdi->wb; | |
607 | ||
608 | do { | |
ed288dc0 | 609 | wb = wb_get_lookup(bdi, memcg_css); |
52ebea74 TH |
610 | } while (!wb && !cgwb_create(bdi, memcg_css, gfp)); |
611 | ||
612 | return wb; | |
613 | } | |
6467716a | 614 | |
a13f35e8 | 615 | static int cgwb_bdi_init(struct backing_dev_info *bdi) |
52ebea74 | 616 | { |
a13f35e8 TH |
617 | int ret; |
618 | ||
52ebea74 | 619 | INIT_RADIX_TREE(&bdi->cgwb_tree, GFP_ATOMIC); |
3ee7e869 | 620 | mutex_init(&bdi->cgwb_release_mutex); |
7fc5854f | 621 | init_rwsem(&bdi->wb_switch_rwsem); |
a13f35e8 | 622 | |
8c911f3d | 623 | ret = wb_init(&bdi->wb, bdi, GFP_KERNEL); |
a13f35e8 | 624 | if (!ret) { |
7d828602 | 625 | bdi->wb.memcg_css = &root_mem_cgroup->css; |
a13f35e8 TH |
626 | bdi->wb.blkcg_css = blkcg_root_css; |
627 | } | |
628 | return ret; | |
6467716a AB |
629 | } |
630 | ||
b1c51afc | 631 | static void cgwb_bdi_unregister(struct backing_dev_info *bdi) |
52ebea74 TH |
632 | { |
633 | struct radix_tree_iter iter; | |
634 | void **slot; | |
5318ce7d | 635 | struct bdi_writeback *wb; |
52ebea74 TH |
636 | |
637 | WARN_ON(test_bit(WB_registered, &bdi->wb.state)); | |
638 | ||
639 | spin_lock_irq(&cgwb_lock); | |
640 | radix_tree_for_each_slot(slot, &bdi->cgwb_tree, &iter, 0) | |
641 | cgwb_kill(*slot); | |
3ee7e869 | 642 | spin_unlock_irq(&cgwb_lock); |
5318ce7d | 643 | |
3ee7e869 JK |
644 | mutex_lock(&bdi->cgwb_release_mutex); |
645 | spin_lock_irq(&cgwb_lock); | |
5318ce7d JK |
646 | while (!list_empty(&bdi->wb_list)) { |
647 | wb = list_first_entry(&bdi->wb_list, struct bdi_writeback, | |
648 | bdi_node); | |
649 | spin_unlock_irq(&cgwb_lock); | |
650 | wb_shutdown(wb); | |
651 | spin_lock_irq(&cgwb_lock); | |
652 | } | |
52ebea74 | 653 | spin_unlock_irq(&cgwb_lock); |
3ee7e869 | 654 | mutex_unlock(&bdi->cgwb_release_mutex); |
52ebea74 TH |
655 | } |
656 | ||
c22d70a1 RG |
657 | /* |
658 | * cleanup_offline_cgwbs_workfn - try to release dying cgwbs | |
659 | * | |
660 | * Try to release dying cgwbs by switching attached inodes to the nearest | |
661 | * living ancestor's writeback. Processed wbs are placed at the end | |
662 | * of the list to guarantee the forward progress. | |
663 | */ | |
664 | static void cleanup_offline_cgwbs_workfn(struct work_struct *work) | |
665 | { | |
666 | struct bdi_writeback *wb; | |
667 | LIST_HEAD(processed); | |
668 | ||
669 | spin_lock_irq(&cgwb_lock); | |
670 | ||
671 | while (!list_empty(&offline_cgwbs)) { | |
672 | wb = list_first_entry(&offline_cgwbs, struct bdi_writeback, | |
673 | offline_node); | |
674 | list_move(&wb->offline_node, &processed); | |
675 | ||
676 | /* | |
677 | * If wb is dirty, cleaning up the writeback by switching | |
678 | * attached inodes will result in an effective removal of any | |
679 | * bandwidth restrictions, which isn't the goal. Instead, | |
680 | * it can be postponed until the next time, when all io | |
681 | * will be likely completed. If in the meantime some inodes | |
682 | * will get re-dirtied, they should be eventually switched to | |
683 | * a new cgwb. | |
684 | */ | |
685 | if (wb_has_dirty_io(wb)) | |
686 | continue; | |
687 | ||
688 | if (!wb_tryget(wb)) | |
689 | continue; | |
690 | ||
691 | spin_unlock_irq(&cgwb_lock); | |
692 | while (cleanup_offline_cgwb(wb)) | |
693 | cond_resched(); | |
694 | spin_lock_irq(&cgwb_lock); | |
695 | ||
696 | wb_put(wb); | |
697 | } | |
698 | ||
699 | if (!list_empty(&processed)) | |
700 | list_splice_tail(&processed, &offline_cgwbs); | |
701 | ||
702 | spin_unlock_irq(&cgwb_lock); | |
703 | } | |
704 | ||
52ebea74 TH |
705 | /** |
706 | * wb_memcg_offline - kill all wb's associated with a memcg being offlined | |
707 | * @memcg: memcg being offlined | |
708 | * | |
709 | * Also prevents creation of any new wb's associated with @memcg. | |
e98be2d5 | 710 | */ |
52ebea74 TH |
711 | void wb_memcg_offline(struct mem_cgroup *memcg) |
712 | { | |
9ccc3617 | 713 | struct list_head *memcg_cgwb_list = &memcg->cgwb_list; |
52ebea74 TH |
714 | struct bdi_writeback *wb, *next; |
715 | ||
716 | spin_lock_irq(&cgwb_lock); | |
717 | list_for_each_entry_safe(wb, next, memcg_cgwb_list, memcg_node) | |
718 | cgwb_kill(wb); | |
719 | memcg_cgwb_list->next = NULL; /* prevent new wb's */ | |
720 | spin_unlock_irq(&cgwb_lock); | |
c22d70a1 RG |
721 | |
722 | queue_work(system_unbound_wq, &cleanup_offline_cgwbs_work); | |
52ebea74 TH |
723 | } |
724 | ||
725 | /** | |
726 | * wb_blkcg_offline - kill all wb's associated with a blkcg being offlined | |
727 | * @blkcg: blkcg being offlined | |
728 | * | |
729 | * Also prevents creation of any new wb's associated with @blkcg. | |
730 | */ | |
731 | void wb_blkcg_offline(struct blkcg *blkcg) | |
732 | { | |
52ebea74 TH |
733 | struct bdi_writeback *wb, *next; |
734 | ||
735 | spin_lock_irq(&cgwb_lock); | |
736 | list_for_each_entry_safe(wb, next, &blkcg->cgwb_list, blkcg_node) | |
737 | cgwb_kill(wb); | |
738 | blkcg->cgwb_list.next = NULL; /* prevent new wb's */ | |
739 | spin_unlock_irq(&cgwb_lock); | |
740 | } | |
741 | ||
e8cb72b3 JK |
742 | static void cgwb_bdi_register(struct backing_dev_info *bdi) |
743 | { | |
744 | spin_lock_irq(&cgwb_lock); | |
745 | list_add_tail_rcu(&bdi->wb.bdi_node, &bdi->wb_list); | |
746 | spin_unlock_irq(&cgwb_lock); | |
747 | } | |
748 | ||
f1834646 TH |
749 | static int __init cgwb_init(void) |
750 | { | |
751 | /* | |
752 | * There can be many concurrent release work items overwhelming | |
753 | * system_wq. Put them in a separate wq and limit concurrency. | |
754 | * There's no point in executing many of these in parallel. | |
755 | */ | |
756 | cgwb_release_wq = alloc_workqueue("cgwb_release", 0, 1); | |
757 | if (!cgwb_release_wq) | |
758 | return -ENOMEM; | |
759 | ||
760 | return 0; | |
761 | } | |
762 | subsys_initcall(cgwb_init); | |
763 | ||
52ebea74 TH |
764 | #else /* CONFIG_CGROUP_WRITEBACK */ |
765 | ||
a13f35e8 TH |
766 | static int cgwb_bdi_init(struct backing_dev_info *bdi) |
767 | { | |
8c911f3d | 768 | return wb_init(&bdi->wb, bdi, GFP_KERNEL); |
a13f35e8 TH |
769 | } |
770 | ||
b1c51afc | 771 | static void cgwb_bdi_unregister(struct backing_dev_info *bdi) { } |
df23de55 | 772 | |
e8cb72b3 JK |
773 | static void cgwb_bdi_register(struct backing_dev_info *bdi) |
774 | { | |
775 | list_add_tail_rcu(&bdi->wb.bdi_node, &bdi->wb_list); | |
776 | } | |
777 | ||
778 | static void cgwb_remove_from_bdi_list(struct bdi_writeback *wb) | |
779 | { | |
780 | list_del_rcu(&wb->bdi_node); | |
781 | } | |
782 | ||
52ebea74 | 783 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
e98be2d5 | 784 | |
2e82b84c | 785 | static int bdi_init(struct backing_dev_info *bdi) |
b2e8fb6e | 786 | { |
b817525a TH |
787 | int ret; |
788 | ||
cf0ca9fe PZ |
789 | bdi->dev = NULL; |
790 | ||
d03f6cdc | 791 | kref_init(&bdi->refcnt); |
189d3c4a | 792 | bdi->min_ratio = 0; |
a42dde04 | 793 | bdi->max_ratio = 100; |
eb608e3a | 794 | bdi->max_prop_frac = FPROP_FRAC_BASE; |
66f3b8e2 | 795 | INIT_LIST_HEAD(&bdi->bdi_list); |
b817525a | 796 | INIT_LIST_HEAD(&bdi->wb_list); |
cc395d7f | 797 | init_waitqueue_head(&bdi->wb_waitq); |
03ba3782 | 798 | |
b817525a TH |
799 | ret = cgwb_bdi_init(bdi); |
800 | ||
b817525a | 801 | return ret; |
b2e8fb6e | 802 | } |
e98be2d5 | 803 | |
aef33c2f | 804 | struct backing_dev_info *bdi_alloc(int node_id) |
d03f6cdc JK |
805 | { |
806 | struct backing_dev_info *bdi; | |
807 | ||
aef33c2f | 808 | bdi = kzalloc_node(sizeof(*bdi), GFP_KERNEL, node_id); |
d03f6cdc JK |
809 | if (!bdi) |
810 | return NULL; | |
811 | ||
812 | if (bdi_init(bdi)) { | |
813 | kfree(bdi); | |
814 | return NULL; | |
815 | } | |
f56753ac | 816 | bdi->capabilities = BDI_CAP_WRITEBACK | BDI_CAP_WRITEBACK_ACCT; |
55b2598e CH |
817 | bdi->ra_pages = VM_READAHEAD_PAGES; |
818 | bdi->io_pages = VM_READAHEAD_PAGES; | |
5ed964f8 | 819 | timer_setup(&bdi->laptop_mode_wb_timer, laptop_mode_timer_fn, 0); |
d03f6cdc JK |
820 | return bdi; |
821 | } | |
aef33c2f | 822 | EXPORT_SYMBOL(bdi_alloc); |
d03f6cdc | 823 | |
34f8fe50 TH |
824 | static struct rb_node **bdi_lookup_rb_node(u64 id, struct rb_node **parentp) |
825 | { | |
826 | struct rb_node **p = &bdi_tree.rb_node; | |
827 | struct rb_node *parent = NULL; | |
828 | struct backing_dev_info *bdi; | |
829 | ||
830 | lockdep_assert_held(&bdi_lock); | |
831 | ||
832 | while (*p) { | |
833 | parent = *p; | |
834 | bdi = rb_entry(parent, struct backing_dev_info, rb_node); | |
835 | ||
836 | if (bdi->id > id) | |
837 | p = &(*p)->rb_left; | |
838 | else if (bdi->id < id) | |
839 | p = &(*p)->rb_right; | |
840 | else | |
841 | break; | |
842 | } | |
843 | ||
844 | if (parentp) | |
845 | *parentp = parent; | |
846 | return p; | |
847 | } | |
848 | ||
849 | /** | |
850 | * bdi_get_by_id - lookup and get bdi from its id | |
851 | * @id: bdi id to lookup | |
852 | * | |
853 | * Find bdi matching @id and get it. Returns NULL if the matching bdi | |
854 | * doesn't exist or is already unregistered. | |
855 | */ | |
856 | struct backing_dev_info *bdi_get_by_id(u64 id) | |
857 | { | |
858 | struct backing_dev_info *bdi = NULL; | |
859 | struct rb_node **p; | |
860 | ||
861 | spin_lock_bh(&bdi_lock); | |
862 | p = bdi_lookup_rb_node(id, NULL); | |
863 | if (*p) { | |
864 | bdi = rb_entry(*p, struct backing_dev_info, rb_node); | |
865 | bdi_get(bdi); | |
866 | } | |
867 | spin_unlock_bh(&bdi_lock); | |
868 | ||
869 | return bdi; | |
870 | } | |
871 | ||
7c4cc300 | 872 | int bdi_register_va(struct backing_dev_info *bdi, const char *fmt, va_list args) |
46100071 | 873 | { |
46100071 | 874 | struct device *dev; |
34f8fe50 | 875 | struct rb_node *parent, **p; |
e98be2d5 | 876 | |
46100071 TH |
877 | if (bdi->dev) /* The driver needs to use separate queues per device */ |
878 | return 0; | |
e98be2d5 | 879 | |
6bd87eec CH |
880 | vsnprintf(bdi->dev_name, sizeof(bdi->dev_name), fmt, args); |
881 | dev = device_create(bdi_class, NULL, MKDEV(0, 0), bdi, bdi->dev_name); | |
46100071 TH |
882 | if (IS_ERR(dev)) |
883 | return PTR_ERR(dev); | |
04fbfdc1 | 884 | |
e8cb72b3 | 885 | cgwb_bdi_register(bdi); |
46100071 | 886 | bdi->dev = dev; |
b2e8fb6e | 887 | |
6d0e4827 | 888 | bdi_debug_register(bdi, dev_name(dev)); |
46100071 TH |
889 | set_bit(WB_registered, &bdi->wb.state); |
890 | ||
891 | spin_lock_bh(&bdi_lock); | |
34f8fe50 TH |
892 | |
893 | bdi->id = ++bdi_id_cursor; | |
894 | ||
895 | p = bdi_lookup_rb_node(bdi->id, &parent); | |
896 | rb_link_node(&bdi->rb_node, parent, p); | |
897 | rb_insert_color(&bdi->rb_node, &bdi_tree); | |
898 | ||
46100071 | 899 | list_add_tail_rcu(&bdi->bdi_list, &bdi_list); |
34f8fe50 | 900 | |
46100071 TH |
901 | spin_unlock_bh(&bdi_lock); |
902 | ||
903 | trace_writeback_bdi_register(bdi); | |
904 | return 0; | |
b2e8fb6e | 905 | } |
baf7a616 | 906 | |
7c4cc300 | 907 | int bdi_register(struct backing_dev_info *bdi, const char *fmt, ...) |
baf7a616 JK |
908 | { |
909 | va_list args; | |
910 | int ret; | |
911 | ||
912 | va_start(args, fmt); | |
7c4cc300 | 913 | ret = bdi_register_va(bdi, fmt, args); |
baf7a616 JK |
914 | va_end(args); |
915 | return ret; | |
916 | } | |
46100071 | 917 | EXPORT_SYMBOL(bdi_register); |
b2e8fb6e | 918 | |
3c5d202b | 919 | void bdi_set_owner(struct backing_dev_info *bdi, struct device *owner) |
df08c32c | 920 | { |
3c5d202b | 921 | WARN_ON_ONCE(bdi->owner); |
df08c32c DW |
922 | bdi->owner = owner; |
923 | get_device(owner); | |
df08c32c | 924 | } |
df08c32c | 925 | |
46100071 TH |
926 | /* |
927 | * Remove bdi from bdi_list, and ensure that it is no longer visible | |
928 | */ | |
929 | static void bdi_remove_from_list(struct backing_dev_info *bdi) | |
930 | { | |
931 | spin_lock_bh(&bdi_lock); | |
34f8fe50 | 932 | rb_erase(&bdi->rb_node, &bdi_tree); |
46100071 TH |
933 | list_del_rcu(&bdi->bdi_list); |
934 | spin_unlock_bh(&bdi_lock); | |
b2e8fb6e | 935 | |
46100071 TH |
936 | synchronize_rcu_expedited(); |
937 | } | |
cf0ca9fe | 938 | |
b02176f3 | 939 | void bdi_unregister(struct backing_dev_info *bdi) |
b2e8fb6e | 940 | { |
5ed964f8 CH |
941 | del_timer_sync(&bdi->laptop_mode_wb_timer); |
942 | ||
f0054bb1 TH |
943 | /* make sure nobody finds us on the bdi_list anymore */ |
944 | bdi_remove_from_list(bdi); | |
945 | wb_shutdown(&bdi->wb); | |
b1c51afc | 946 | cgwb_bdi_unregister(bdi); |
7a401a97 | 947 | |
c4db59d3 CH |
948 | if (bdi->dev) { |
949 | bdi_debug_unregister(bdi); | |
950 | device_unregister(bdi->dev); | |
951 | bdi->dev = NULL; | |
952 | } | |
df08c32c DW |
953 | |
954 | if (bdi->owner) { | |
955 | put_device(bdi->owner); | |
956 | bdi->owner = NULL; | |
957 | } | |
b02176f3 | 958 | } |
c6fd3ac0 | 959 | EXPORT_SYMBOL(bdi_unregister); |
c4db59d3 | 960 | |
d03f6cdc JK |
961 | static void release_bdi(struct kref *ref) |
962 | { | |
963 | struct backing_dev_info *bdi = | |
964 | container_of(ref, struct backing_dev_info, refcnt); | |
965 | ||
702f2d1e | 966 | WARN_ON_ONCE(test_bit(WB_registered, &bdi->wb.state)); |
2e82b84c JK |
967 | WARN_ON_ONCE(bdi->dev); |
968 | wb_exit(&bdi->wb); | |
d03f6cdc JK |
969 | kfree(bdi); |
970 | } | |
971 | ||
972 | void bdi_put(struct backing_dev_info *bdi) | |
973 | { | |
974 | kref_put(&bdi->refcnt, release_bdi); | |
975 | } | |
62bf42ad | 976 | EXPORT_SYMBOL(bdi_put); |
d03f6cdc | 977 | |
ccdf7741 CH |
978 | struct backing_dev_info *inode_to_bdi(struct inode *inode) |
979 | { | |
980 | struct super_block *sb; | |
981 | ||
982 | if (!inode) | |
983 | return &noop_backing_dev_info; | |
984 | ||
985 | sb = inode->i_sb; | |
986 | #ifdef CONFIG_BLOCK | |
987 | if (sb_is_blkdev_sb(sb)) | |
988 | return I_BDEV(inode)->bd_disk->bdi; | |
989 | #endif | |
990 | return sb->s_bdi; | |
991 | } | |
992 | EXPORT_SYMBOL(inode_to_bdi); | |
993 | ||
eb7ae5e0 CH |
994 | const char *bdi_dev_name(struct backing_dev_info *bdi) |
995 | { | |
996 | if (!bdi || !bdi->dev) | |
997 | return bdi_unknown_name; | |
6bd87eec | 998 | return bdi->dev_name; |
eb7ae5e0 CH |
999 | } |
1000 | EXPORT_SYMBOL_GPL(bdi_dev_name); | |
1001 | ||
3fcfab16 AM |
1002 | static wait_queue_head_t congestion_wqh[2] = { |
1003 | __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]), | |
1004 | __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1]) | |
1005 | }; | |
ec8a6f26 | 1006 | static atomic_t nr_wb_congested[2]; |
3fcfab16 | 1007 | |
492d76b2 | 1008 | void clear_bdi_congested(struct backing_dev_info *bdi, int sync) |
3fcfab16 | 1009 | { |
1faa16d2 | 1010 | wait_queue_head_t *wqh = &congestion_wqh[sync]; |
c877ef8a | 1011 | enum wb_congested_state bit; |
3fcfab16 | 1012 | |
4452226e | 1013 | bit = sync ? WB_sync_congested : WB_async_congested; |
8c911f3d | 1014 | if (test_and_clear_bit(bit, &bdi->wb.congested)) |
ec8a6f26 | 1015 | atomic_dec(&nr_wb_congested[sync]); |
4e857c58 | 1016 | smp_mb__after_atomic(); |
3fcfab16 AM |
1017 | if (waitqueue_active(wqh)) |
1018 | wake_up(wqh); | |
1019 | } | |
492d76b2 | 1020 | EXPORT_SYMBOL(clear_bdi_congested); |
3fcfab16 | 1021 | |
492d76b2 | 1022 | void set_bdi_congested(struct backing_dev_info *bdi, int sync) |
3fcfab16 | 1023 | { |
c877ef8a | 1024 | enum wb_congested_state bit; |
3fcfab16 | 1025 | |
4452226e | 1026 | bit = sync ? WB_sync_congested : WB_async_congested; |
8c911f3d | 1027 | if (!test_and_set_bit(bit, &bdi->wb.congested)) |
ec8a6f26 | 1028 | atomic_inc(&nr_wb_congested[sync]); |
3fcfab16 | 1029 | } |
492d76b2 | 1030 | EXPORT_SYMBOL(set_bdi_congested); |
3fcfab16 AM |
1031 | |
1032 | /** | |
1033 | * congestion_wait - wait for a backing_dev to become uncongested | |
8aa7e847 | 1034 | * @sync: SYNC or ASYNC IO |
3fcfab16 AM |
1035 | * @timeout: timeout in jiffies |
1036 | * | |
1037 | * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit | |
1038 | * write congestion. If no backing_devs are congested then just wait for the | |
1039 | * next write to be completed. | |
1040 | */ | |
8aa7e847 | 1041 | long congestion_wait(int sync, long timeout) |
3fcfab16 AM |
1042 | { |
1043 | long ret; | |
52bb9198 | 1044 | unsigned long start = jiffies; |
3fcfab16 | 1045 | DEFINE_WAIT(wait); |
8aa7e847 | 1046 | wait_queue_head_t *wqh = &congestion_wqh[sync]; |
3fcfab16 AM |
1047 | |
1048 | prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); | |
1049 | ret = io_schedule_timeout(timeout); | |
1050 | finish_wait(wqh, &wait); | |
52bb9198 MG |
1051 | |
1052 | trace_writeback_congestion_wait(jiffies_to_usecs(timeout), | |
1053 | jiffies_to_usecs(jiffies - start)); | |
1054 | ||
3fcfab16 AM |
1055 | return ret; |
1056 | } | |
1057 | EXPORT_SYMBOL(congestion_wait); |