Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
3fcfab16 | 2 | |
c97ab271 | 3 | #include <linux/blkdev.h> |
3fcfab16 | 4 | #include <linux/wait.h> |
34f8fe50 | 5 | #include <linux/rbtree.h> |
03ba3782 | 6 | #include <linux/kthread.h> |
e41d12f5 CH |
7 | #include <linux/backing-dev.h> |
8 | #include <linux/blk-cgroup.h> | |
03ba3782 | 9 | #include <linux/freezer.h> |
3fcfab16 | 10 | #include <linux/fs.h> |
26160158 | 11 | #include <linux/pagemap.h> |
03ba3782 | 12 | #include <linux/mm.h> |
c1ca59a1 | 13 | #include <linux/sched/mm.h> |
3fcfab16 AM |
14 | #include <linux/sched.h> |
15 | #include <linux/module.h> | |
cf0ca9fe PZ |
16 | #include <linux/writeback.h> |
17 | #include <linux/device.h> | |
455b2864 | 18 | #include <trace/events/writeback.h> |
cf0ca9fe | 19 | |
f56753ac | 20 | struct backing_dev_info noop_backing_dev_info; |
a212b105 | 21 | EXPORT_SYMBOL_GPL(noop_backing_dev_info); |
5129a469 | 22 | |
cf0ca9fe | 23 | static struct class *bdi_class; |
eb7ae5e0 | 24 | static const char *bdi_unknown_name = "(unknown)"; |
cfc4ba53 JA |
25 | |
26 | /* | |
34f8fe50 TH |
27 | * bdi_lock protects bdi_tree and updates to bdi_list. bdi_list has RCU |
28 | * reader side locking. | |
cfc4ba53 | 29 | */ |
03ba3782 | 30 | DEFINE_SPINLOCK(bdi_lock); |
34f8fe50 TH |
31 | static u64 bdi_id_cursor; |
32 | static struct rb_root bdi_tree = RB_ROOT; | |
66f3b8e2 | 33 | LIST_HEAD(bdi_list); |
03ba3782 | 34 | |
839a8e86 TH |
35 | /* bdi_wq serves all asynchronous writeback tasks */ |
36 | struct workqueue_struct *bdi_wq; | |
37 | ||
6986c3e2 BW |
38 | #define K(x) ((x) << (PAGE_SHIFT - 10)) |
39 | ||
76f1418b MS |
40 | #ifdef CONFIG_DEBUG_FS |
41 | #include <linux/debugfs.h> | |
42 | #include <linux/seq_file.h> | |
43 | ||
44 | static struct dentry *bdi_debug_root; | |
45 | ||
46 | static void bdi_debug_init(void) | |
47 | { | |
48 | bdi_debug_root = debugfs_create_dir("bdi", NULL); | |
49 | } | |
50 | ||
51 | static int bdi_debug_stats_show(struct seq_file *m, void *v) | |
52 | { | |
53 | struct backing_dev_info *bdi = m->private; | |
c1955ce3 | 54 | struct bdi_writeback *wb = &bdi->wb; |
364aeb28 DR |
55 | unsigned long background_thresh; |
56 | unsigned long dirty_thresh; | |
0d960a38 | 57 | unsigned long wb_thresh; |
0ae45f63 | 58 | unsigned long nr_dirty, nr_io, nr_more_io, nr_dirty_time; |
f09b00d3 JA |
59 | struct inode *inode; |
60 | ||
0ae45f63 | 61 | nr_dirty = nr_io = nr_more_io = nr_dirty_time = 0; |
f758eeab | 62 | spin_lock(&wb->list_lock); |
c7f54084 | 63 | list_for_each_entry(inode, &wb->b_dirty, i_io_list) |
c1955ce3 | 64 | nr_dirty++; |
c7f54084 | 65 | list_for_each_entry(inode, &wb->b_io, i_io_list) |
c1955ce3 | 66 | nr_io++; |
c7f54084 | 67 | list_for_each_entry(inode, &wb->b_more_io, i_io_list) |
c1955ce3 | 68 | nr_more_io++; |
c7f54084 | 69 | list_for_each_entry(inode, &wb->b_dirty_time, i_io_list) |
0ae45f63 TT |
70 | if (inode->i_state & I_DIRTY_TIME) |
71 | nr_dirty_time++; | |
f758eeab | 72 | spin_unlock(&wb->list_lock); |
76f1418b | 73 | |
16c4042f | 74 | global_dirty_limits(&background_thresh, &dirty_thresh); |
0d960a38 | 75 | wb_thresh = wb_calc_thresh(wb, dirty_thresh); |
76f1418b | 76 | |
76f1418b | 77 | seq_printf(m, |
00821b00 WF |
78 | "BdiWriteback: %10lu kB\n" |
79 | "BdiReclaimable: %10lu kB\n" | |
80 | "BdiDirtyThresh: %10lu kB\n" | |
81 | "DirtyThresh: %10lu kB\n" | |
82 | "BackgroundThresh: %10lu kB\n" | |
c8e28ce0 | 83 | "BdiDirtied: %10lu kB\n" |
00821b00 WF |
84 | "BdiWritten: %10lu kB\n" |
85 | "BdiWriteBandwidth: %10lu kBps\n" | |
86 | "b_dirty: %10lu\n" | |
87 | "b_io: %10lu\n" | |
88 | "b_more_io: %10lu\n" | |
0ae45f63 | 89 | "b_dirty_time: %10lu\n" |
00821b00 WF |
90 | "bdi_list: %10u\n" |
91 | "state: %10lx\n", | |
93f78d88 TH |
92 | (unsigned long) K(wb_stat(wb, WB_WRITEBACK)), |
93 | (unsigned long) K(wb_stat(wb, WB_RECLAIMABLE)), | |
0d960a38 | 94 | K(wb_thresh), |
f7d2b1ec JK |
95 | K(dirty_thresh), |
96 | K(background_thresh), | |
93f78d88 TH |
97 | (unsigned long) K(wb_stat(wb, WB_DIRTIED)), |
98 | (unsigned long) K(wb_stat(wb, WB_WRITTEN)), | |
a88a341a | 99 | (unsigned long) K(wb->write_bandwidth), |
f7d2b1ec JK |
100 | nr_dirty, |
101 | nr_io, | |
102 | nr_more_io, | |
0ae45f63 | 103 | nr_dirty_time, |
4452226e | 104 | !list_empty(&bdi->bdi_list), bdi->wb.state); |
76f1418b MS |
105 | |
106 | return 0; | |
107 | } | |
5ad35093 | 108 | DEFINE_SHOW_ATTRIBUTE(bdi_debug_stats); |
76f1418b | 109 | |
2d146b92 | 110 | static void bdi_debug_register(struct backing_dev_info *bdi, const char *name) |
76f1418b MS |
111 | { |
112 | bdi->debug_dir = debugfs_create_dir(name, bdi_debug_root); | |
97f07697 | 113 | |
2d146b92 GKH |
114 | debugfs_create_file("stats", 0444, bdi->debug_dir, bdi, |
115 | &bdi_debug_stats_fops); | |
76f1418b MS |
116 | } |
117 | ||
118 | static void bdi_debug_unregister(struct backing_dev_info *bdi) | |
119 | { | |
2d146b92 | 120 | debugfs_remove_recursive(bdi->debug_dir); |
76f1418b MS |
121 | } |
122 | #else | |
123 | static inline void bdi_debug_init(void) | |
124 | { | |
125 | } | |
2d146b92 | 126 | static inline void bdi_debug_register(struct backing_dev_info *bdi, |
76f1418b MS |
127 | const char *name) |
128 | { | |
129 | } | |
130 | static inline void bdi_debug_unregister(struct backing_dev_info *bdi) | |
131 | { | |
132 | } | |
133 | #endif | |
134 | ||
cf0ca9fe PZ |
135 | static ssize_t read_ahead_kb_store(struct device *dev, |
136 | struct device_attribute *attr, | |
137 | const char *buf, size_t count) | |
138 | { | |
139 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
cf0ca9fe | 140 | unsigned long read_ahead_kb; |
7034ed13 | 141 | ssize_t ret; |
cf0ca9fe | 142 | |
7034ed13 NJ |
143 | ret = kstrtoul(buf, 10, &read_ahead_kb); |
144 | if (ret < 0) | |
145 | return ret; | |
146 | ||
147 | bdi->ra_pages = read_ahead_kb >> (PAGE_SHIFT - 10); | |
148 | ||
149 | return count; | |
cf0ca9fe PZ |
150 | } |
151 | ||
cf0ca9fe PZ |
152 | #define BDI_SHOW(name, expr) \ |
153 | static ssize_t name##_show(struct device *dev, \ | |
5e4c0d86 | 154 | struct device_attribute *attr, char *buf) \ |
cf0ca9fe PZ |
155 | { \ |
156 | struct backing_dev_info *bdi = dev_get_drvdata(dev); \ | |
157 | \ | |
5e4c0d86 | 158 | return sysfs_emit(buf, "%lld\n", (long long)expr); \ |
d9e1241e GKH |
159 | } \ |
160 | static DEVICE_ATTR_RW(name); | |
cf0ca9fe PZ |
161 | |
162 | BDI_SHOW(read_ahead_kb, K(bdi->ra_pages)) | |
163 | ||
189d3c4a PZ |
164 | static ssize_t min_ratio_store(struct device *dev, |
165 | struct device_attribute *attr, const char *buf, size_t count) | |
166 | { | |
167 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
189d3c4a | 168 | unsigned int ratio; |
7034ed13 NJ |
169 | ssize_t ret; |
170 | ||
171 | ret = kstrtouint(buf, 10, &ratio); | |
172 | if (ret < 0) | |
173 | return ret; | |
174 | ||
175 | ret = bdi_set_min_ratio(bdi, ratio); | |
176 | if (!ret) | |
177 | ret = count; | |
189d3c4a | 178 | |
189d3c4a PZ |
179 | return ret; |
180 | } | |
ae82291e | 181 | BDI_SHOW(min_ratio, bdi->min_ratio / BDI_RATIO_SCALE) |
189d3c4a | 182 | |
a42dde04 PZ |
183 | static ssize_t max_ratio_store(struct device *dev, |
184 | struct device_attribute *attr, const char *buf, size_t count) | |
185 | { | |
186 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
a42dde04 | 187 | unsigned int ratio; |
7034ed13 NJ |
188 | ssize_t ret; |
189 | ||
190 | ret = kstrtouint(buf, 10, &ratio); | |
191 | if (ret < 0) | |
192 | return ret; | |
193 | ||
194 | ret = bdi_set_max_ratio(bdi, ratio); | |
195 | if (!ret) | |
196 | ret = count; | |
a42dde04 | 197 | |
a42dde04 PZ |
198 | return ret; |
199 | } | |
ae82291e | 200 | BDI_SHOW(max_ratio, bdi->max_ratio / BDI_RATIO_SCALE) |
a42dde04 | 201 | |
9c84819b SR |
202 | static ssize_t min_bytes_show(struct device *dev, |
203 | struct device_attribute *attr, | |
204 | char *buf) | |
205 | { | |
206 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
207 | ||
208 | return sysfs_emit(buf, "%llu\n", bdi_get_min_bytes(bdi)); | |
209 | } | |
210 | ||
211 | static ssize_t min_bytes_store(struct device *dev, | |
212 | struct device_attribute *attr, const char *buf, size_t count) | |
213 | { | |
214 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
215 | u64 bytes; | |
216 | ssize_t ret; | |
217 | ||
218 | ret = kstrtoull(buf, 10, &bytes); | |
219 | if (ret < 0) | |
220 | return ret; | |
221 | ||
222 | ret = bdi_set_min_bytes(bdi, bytes); | |
223 | if (!ret) | |
224 | ret = count; | |
225 | ||
226 | return ret; | |
227 | } | |
228 | DEVICE_ATTR_RW(min_bytes); | |
229 | ||
c56e049a SR |
230 | static ssize_t max_bytes_show(struct device *dev, |
231 | struct device_attribute *attr, | |
232 | char *buf) | |
233 | { | |
234 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
235 | ||
236 | return sysfs_emit(buf, "%llu\n", bdi_get_max_bytes(bdi)); | |
237 | } | |
238 | ||
239 | static ssize_t max_bytes_store(struct device *dev, | |
240 | struct device_attribute *attr, const char *buf, size_t count) | |
241 | { | |
242 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
243 | u64 bytes; | |
244 | ssize_t ret; | |
245 | ||
246 | ret = kstrtoull(buf, 10, &bytes); | |
247 | if (ret < 0) | |
248 | return ret; | |
249 | ||
250 | ret = bdi_set_max_bytes(bdi, bytes); | |
251 | if (!ret) | |
252 | ret = count; | |
253 | ||
254 | return ret; | |
255 | } | |
256 | DEVICE_ATTR_RW(max_bytes); | |
257 | ||
7d311cda DW |
258 | static ssize_t stable_pages_required_show(struct device *dev, |
259 | struct device_attribute *attr, | |
5e4c0d86 | 260 | char *buf) |
7d311cda | 261 | { |
1cb039f3 CH |
262 | dev_warn_once(dev, |
263 | "the stable_pages_required attribute has been removed. Use the stable_writes queue attribute instead.\n"); | |
5e4c0d86 | 264 | return sysfs_emit(buf, "%d\n", 0); |
7d311cda | 265 | } |
d9e1241e GKH |
266 | static DEVICE_ATTR_RO(stable_pages_required); |
267 | ||
27bbe9d4 SR |
268 | static ssize_t strict_limit_store(struct device *dev, |
269 | struct device_attribute *attr, const char *buf, size_t count) | |
270 | { | |
271 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
272 | unsigned int strict_limit; | |
273 | ssize_t ret; | |
274 | ||
275 | ret = kstrtouint(buf, 10, &strict_limit); | |
276 | if (ret < 0) | |
277 | return ret; | |
278 | ||
279 | ret = bdi_set_strict_limit(bdi, strict_limit); | |
280 | if (!ret) | |
281 | ret = count; | |
282 | ||
283 | return ret; | |
284 | } | |
285 | ||
286 | static ssize_t strict_limit_show(struct device *dev, | |
287 | struct device_attribute *attr, char *buf) | |
288 | { | |
289 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
290 | ||
291 | return sysfs_emit(buf, "%d\n", | |
292 | !!(bdi->capabilities & BDI_CAP_STRICTLIMIT)); | |
293 | } | |
294 | static DEVICE_ATTR_RW(strict_limit); | |
295 | ||
d9e1241e GKH |
296 | static struct attribute *bdi_dev_attrs[] = { |
297 | &dev_attr_read_ahead_kb.attr, | |
298 | &dev_attr_min_ratio.attr, | |
299 | &dev_attr_max_ratio.attr, | |
9c84819b | 300 | &dev_attr_min_bytes.attr, |
c56e049a | 301 | &dev_attr_max_bytes.attr, |
d9e1241e | 302 | &dev_attr_stable_pages_required.attr, |
27bbe9d4 | 303 | &dev_attr_strict_limit.attr, |
d9e1241e | 304 | NULL, |
cf0ca9fe | 305 | }; |
d9e1241e | 306 | ATTRIBUTE_GROUPS(bdi_dev); |
cf0ca9fe PZ |
307 | |
308 | static __init int bdi_class_init(void) | |
309 | { | |
310 | bdi_class = class_create(THIS_MODULE, "bdi"); | |
14421453 AB |
311 | if (IS_ERR(bdi_class)) |
312 | return PTR_ERR(bdi_class); | |
313 | ||
d9e1241e | 314 | bdi_class->dev_groups = bdi_dev_groups; |
76f1418b | 315 | bdi_debug_init(); |
d03f6cdc | 316 | |
cf0ca9fe PZ |
317 | return 0; |
318 | } | |
76f1418b | 319 | postcore_initcall(bdi_class_init); |
cf0ca9fe | 320 | |
26160158 JA |
321 | static int __init default_bdi_init(void) |
322 | { | |
a2b90f11 MW |
323 | bdi_wq = alloc_workqueue("writeback", WQ_MEM_RECLAIM | WQ_UNBOUND | |
324 | WQ_SYSFS, 0); | |
839a8e86 TH |
325 | if (!bdi_wq) |
326 | return -ENOMEM; | |
4bca7e80 | 327 | return 0; |
26160158 JA |
328 | } |
329 | subsys_initcall(default_bdi_init); | |
330 | ||
6467716a | 331 | /* |
f0054bb1 | 332 | * This function is used when the first inode for this wb is marked dirty. It |
6467716a AB |
333 | * wakes-up the corresponding bdi thread which should then take care of the |
334 | * periodic background write-out of dirty inodes. Since the write-out would | |
335 | * starts only 'dirty_writeback_interval' centisecs from now anyway, we just | |
336 | * set up a timer which wakes the bdi thread up later. | |
337 | * | |
338 | * Note, we wouldn't bother setting up the timer, but this function is on the | |
339 | * fast-path (used by '__mark_inode_dirty()'), so we save few context switches | |
340 | * by delaying the wake-up. | |
6ca738d6 DB |
341 | * |
342 | * We have to be careful not to postpone flush work if it is scheduled for | |
343 | * earlier. Thus we use queue_delayed_work(). | |
6467716a | 344 | */ |
f0054bb1 | 345 | void wb_wakeup_delayed(struct bdi_writeback *wb) |
6467716a AB |
346 | { |
347 | unsigned long timeout; | |
348 | ||
349 | timeout = msecs_to_jiffies(dirty_writeback_interval * 10); | |
f87904c0 | 350 | spin_lock_irq(&wb->work_lock); |
f0054bb1 TH |
351 | if (test_bit(WB_registered, &wb->state)) |
352 | queue_delayed_work(bdi_wq, &wb->dwork, timeout); | |
f87904c0 | 353 | spin_unlock_irq(&wb->work_lock); |
03ba3782 JA |
354 | } |
355 | ||
45a2966f JK |
356 | static void wb_update_bandwidth_workfn(struct work_struct *work) |
357 | { | |
358 | struct bdi_writeback *wb = container_of(to_delayed_work(work), | |
359 | struct bdi_writeback, bw_dwork); | |
360 | ||
361 | wb_update_bandwidth(wb); | |
362 | } | |
363 | ||
cfc4ba53 | 364 | /* |
a88a341a | 365 | * Initial write bandwidth: 100 MB/s |
cfc4ba53 | 366 | */ |
a88a341a | 367 | #define INIT_BW (100 << (20 - PAGE_SHIFT)) |
cfc4ba53 | 368 | |
8395cd9f | 369 | static int wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi, |
8c911f3d | 370 | gfp_t gfp) |
cf0ca9fe | 371 | { |
93f78d88 | 372 | int i, err; |
cf0ca9fe | 373 | |
6467716a | 374 | memset(wb, 0, sizeof(*wb)); |
f1d0b063 | 375 | |
6467716a AB |
376 | wb->bdi = bdi; |
377 | wb->last_old_flush = jiffies; | |
378 | INIT_LIST_HEAD(&wb->b_dirty); | |
379 | INIT_LIST_HEAD(&wb->b_io); | |
380 | INIT_LIST_HEAD(&wb->b_more_io); | |
0ae45f63 | 381 | INIT_LIST_HEAD(&wb->b_dirty_time); |
f758eeab | 382 | spin_lock_init(&wb->list_lock); |
66f3b8e2 | 383 | |
633a2abb | 384 | atomic_set(&wb->writeback_inodes, 0); |
a88a341a TH |
385 | wb->bw_time_stamp = jiffies; |
386 | wb->balanced_dirty_ratelimit = INIT_BW; | |
387 | wb->dirty_ratelimit = INIT_BW; | |
388 | wb->write_bandwidth = INIT_BW; | |
389 | wb->avg_write_bandwidth = INIT_BW; | |
cf0ca9fe | 390 | |
f0054bb1 TH |
391 | spin_lock_init(&wb->work_lock); |
392 | INIT_LIST_HEAD(&wb->work_list); | |
393 | INIT_DELAYED_WORK(&wb->dwork, wb_workfn); | |
45a2966f | 394 | INIT_DELAYED_WORK(&wb->bw_dwork, wb_update_bandwidth_workfn); |
b57d74af | 395 | wb->dirty_sleep = jiffies; |
c284de61 | 396 | |
8395cd9f | 397 | err = fprop_local_init_percpu(&wb->completions, gfp); |
a88a341a | 398 | if (err) |
efee1713 | 399 | return err; |
c284de61 | 400 | |
93f78d88 | 401 | for (i = 0; i < NR_WB_STAT_ITEMS; i++) { |
8395cd9f | 402 | err = percpu_counter_init(&wb->stat[i], 0, gfp); |
a13f35e8 TH |
403 | if (err) |
404 | goto out_destroy_stat; | |
93f78d88 | 405 | } |
cf0ca9fe | 406 | |
93f78d88 | 407 | return 0; |
a13f35e8 TH |
408 | |
409 | out_destroy_stat: | |
078c6c3a | 410 | while (i--) |
a13f35e8 TH |
411 | percpu_counter_destroy(&wb->stat[i]); |
412 | fprop_local_destroy_percpu(&wb->completions); | |
a13f35e8 | 413 | return err; |
cf0ca9fe | 414 | } |
cf0ca9fe | 415 | |
e8cb72b3 JK |
416 | static void cgwb_remove_from_bdi_list(struct bdi_writeback *wb); |
417 | ||
03ba3782 JA |
418 | /* |
419 | * Remove bdi from the global list and shutdown any threads we have running | |
420 | */ | |
46100071 | 421 | static void wb_shutdown(struct bdi_writeback *wb) |
66f3b8e2 | 422 | { |
c4db59d3 | 423 | /* Make sure nobody queues further work */ |
f87904c0 | 424 | spin_lock_irq(&wb->work_lock); |
46100071 | 425 | if (!test_and_clear_bit(WB_registered, &wb->state)) { |
f87904c0 | 426 | spin_unlock_irq(&wb->work_lock); |
03ba3782 | 427 | return; |
c4db59d3 | 428 | } |
f87904c0 | 429 | spin_unlock_irq(&wb->work_lock); |
03ba3782 | 430 | |
e8cb72b3 | 431 | cgwb_remove_from_bdi_list(wb); |
03ba3782 | 432 | /* |
46100071 TH |
433 | * Drain work list and shutdown the delayed_work. !WB_registered |
434 | * tells wb_workfn() that @wb is dying and its work_list needs to | |
435 | * be drained no matter what. | |
03ba3782 | 436 | */ |
46100071 TH |
437 | mod_delayed_work(bdi_wq, &wb->dwork, 0); |
438 | flush_delayed_work(&wb->dwork); | |
439 | WARN_ON(!list_empty(&wb->work_list)); | |
45a2966f | 440 | flush_delayed_work(&wb->bw_dwork); |
46100071 TH |
441 | } |
442 | ||
f0054bb1 | 443 | static void wb_exit(struct bdi_writeback *wb) |
93f78d88 TH |
444 | { |
445 | int i; | |
446 | ||
447 | WARN_ON(delayed_work_pending(&wb->dwork)); | |
448 | ||
449 | for (i = 0; i < NR_WB_STAT_ITEMS; i++) | |
450 | percpu_counter_destroy(&wb->stat[i]); | |
6467716a | 451 | |
a88a341a TH |
452 | fprop_local_destroy_percpu(&wb->completions); |
453 | } | |
e98be2d5 | 454 | |
52ebea74 TH |
455 | #ifdef CONFIG_CGROUP_WRITEBACK |
456 | ||
457 | #include <linux/memcontrol.h> | |
458 | ||
459 | /* | |
c22d70a1 RG |
460 | * cgwb_lock protects bdi->cgwb_tree, blkcg->cgwb_list, offline_cgwbs and |
461 | * memcg->cgwb_list. bdi->cgwb_tree is also RCU protected. | |
52ebea74 TH |
462 | */ |
463 | static DEFINE_SPINLOCK(cgwb_lock); | |
f1834646 | 464 | static struct workqueue_struct *cgwb_release_wq; |
52ebea74 | 465 | |
c22d70a1 RG |
466 | static LIST_HEAD(offline_cgwbs); |
467 | static void cleanup_offline_cgwbs_workfn(struct work_struct *work); | |
468 | static DECLARE_WORK(cleanup_offline_cgwbs_work, cleanup_offline_cgwbs_workfn); | |
469 | ||
52ebea74 TH |
470 | static void cgwb_release_workfn(struct work_struct *work) |
471 | { | |
472 | struct bdi_writeback *wb = container_of(work, struct bdi_writeback, | |
473 | release_work); | |
efee1713 | 474 | struct backing_dev_info *bdi = wb->bdi; |
52ebea74 | 475 | |
3ee7e869 | 476 | mutex_lock(&wb->bdi->cgwb_release_mutex); |
52ebea74 TH |
477 | wb_shutdown(wb); |
478 | ||
479 | css_put(wb->memcg_css); | |
480 | css_put(wb->blkcg_css); | |
3ee7e869 | 481 | mutex_unlock(&wb->bdi->cgwb_release_mutex); |
52ebea74 | 482 | |
d866dbf6 | 483 | /* triggers blkg destruction if no online users left */ |
397c9f46 | 484 | blkcg_unpin_online(wb->blkcg_css); |
59b57717 | 485 | |
841710aa | 486 | fprop_local_destroy_percpu(&wb->memcg_completions); |
c22d70a1 RG |
487 | |
488 | spin_lock_irq(&cgwb_lock); | |
489 | list_del(&wb->offline_node); | |
490 | spin_unlock_irq(&cgwb_lock); | |
491 | ||
b43a9e76 | 492 | percpu_ref_exit(&wb->refcnt); |
52ebea74 | 493 | wb_exit(wb); |
efee1713 | 494 | bdi_put(bdi); |
f3b6a6df | 495 | WARN_ON_ONCE(!list_empty(&wb->b_attached)); |
52ebea74 | 496 | kfree_rcu(wb, rcu); |
52ebea74 TH |
497 | } |
498 | ||
499 | static void cgwb_release(struct percpu_ref *refcnt) | |
500 | { | |
501 | struct bdi_writeback *wb = container_of(refcnt, struct bdi_writeback, | |
502 | refcnt); | |
f1834646 | 503 | queue_work(cgwb_release_wq, &wb->release_work); |
52ebea74 TH |
504 | } |
505 | ||
506 | static void cgwb_kill(struct bdi_writeback *wb) | |
507 | { | |
508 | lockdep_assert_held(&cgwb_lock); | |
509 | ||
510 | WARN_ON(!radix_tree_delete(&wb->bdi->cgwb_tree, wb->memcg_css->id)); | |
511 | list_del(&wb->memcg_node); | |
512 | list_del(&wb->blkcg_node); | |
c22d70a1 | 513 | list_add(&wb->offline_node, &offline_cgwbs); |
52ebea74 TH |
514 | percpu_ref_kill(&wb->refcnt); |
515 | } | |
516 | ||
e8cb72b3 JK |
517 | static void cgwb_remove_from_bdi_list(struct bdi_writeback *wb) |
518 | { | |
519 | spin_lock_irq(&cgwb_lock); | |
520 | list_del_rcu(&wb->bdi_node); | |
521 | spin_unlock_irq(&cgwb_lock); | |
522 | } | |
523 | ||
52ebea74 TH |
524 | static int cgwb_create(struct backing_dev_info *bdi, |
525 | struct cgroup_subsys_state *memcg_css, gfp_t gfp) | |
526 | { | |
527 | struct mem_cgroup *memcg; | |
528 | struct cgroup_subsys_state *blkcg_css; | |
52ebea74 TH |
529 | struct list_head *memcg_cgwb_list, *blkcg_cgwb_list; |
530 | struct bdi_writeback *wb; | |
531 | unsigned long flags; | |
532 | int ret = 0; | |
533 | ||
534 | memcg = mem_cgroup_from_css(memcg_css); | |
c165b3e3 | 535 | blkcg_css = cgroup_get_e_css(memcg_css->cgroup, &io_cgrp_subsys); |
9ccc3617 | 536 | memcg_cgwb_list = &memcg->cgwb_list; |
dec223c9 | 537 | blkcg_cgwb_list = blkcg_get_cgwb_list(blkcg_css); |
52ebea74 TH |
538 | |
539 | /* look up again under lock and discard on blkcg mismatch */ | |
540 | spin_lock_irqsave(&cgwb_lock, flags); | |
541 | wb = radix_tree_lookup(&bdi->cgwb_tree, memcg_css->id); | |
542 | if (wb && wb->blkcg_css != blkcg_css) { | |
543 | cgwb_kill(wb); | |
544 | wb = NULL; | |
545 | } | |
546 | spin_unlock_irqrestore(&cgwb_lock, flags); | |
547 | if (wb) | |
548 | goto out_put; | |
549 | ||
550 | /* need to create a new one */ | |
551 | wb = kmalloc(sizeof(*wb), gfp); | |
0b045bd1 CJ |
552 | if (!wb) { |
553 | ret = -ENOMEM; | |
554 | goto out_put; | |
555 | } | |
52ebea74 | 556 | |
8c911f3d | 557 | ret = wb_init(wb, bdi, gfp); |
52ebea74 TH |
558 | if (ret) |
559 | goto err_free; | |
560 | ||
561 | ret = percpu_ref_init(&wb->refcnt, cgwb_release, 0, gfp); | |
562 | if (ret) | |
563 | goto err_wb_exit; | |
564 | ||
841710aa TH |
565 | ret = fprop_local_init_percpu(&wb->memcg_completions, gfp); |
566 | if (ret) | |
567 | goto err_ref_exit; | |
568 | ||
52ebea74 TH |
569 | wb->memcg_css = memcg_css; |
570 | wb->blkcg_css = blkcg_css; | |
f3b6a6df | 571 | INIT_LIST_HEAD(&wb->b_attached); |
52ebea74 TH |
572 | INIT_WORK(&wb->release_work, cgwb_release_workfn); |
573 | set_bit(WB_registered, &wb->state); | |
efee1713 | 574 | bdi_get(bdi); |
03ba3782 JA |
575 | |
576 | /* | |
52ebea74 TH |
577 | * The root wb determines the registered state of the whole bdi and |
578 | * memcg_cgwb_list and blkcg_cgwb_list's next pointers indicate | |
579 | * whether they're still online. Don't link @wb if any is dead. | |
580 | * See wb_memcg_offline() and wb_blkcg_offline(). | |
03ba3782 | 581 | */ |
52ebea74 TH |
582 | ret = -ENODEV; |
583 | spin_lock_irqsave(&cgwb_lock, flags); | |
584 | if (test_bit(WB_registered, &bdi->wb.state) && | |
585 | blkcg_cgwb_list->next && memcg_cgwb_list->next) { | |
586 | /* we might have raced another instance of this function */ | |
587 | ret = radix_tree_insert(&bdi->cgwb_tree, memcg_css->id, wb); | |
588 | if (!ret) { | |
b817525a | 589 | list_add_tail_rcu(&wb->bdi_node, &bdi->wb_list); |
52ebea74 TH |
590 | list_add(&wb->memcg_node, memcg_cgwb_list); |
591 | list_add(&wb->blkcg_node, blkcg_cgwb_list); | |
397c9f46 | 592 | blkcg_pin_online(blkcg_css); |
52ebea74 TH |
593 | css_get(memcg_css); |
594 | css_get(blkcg_css); | |
595 | } | |
596 | } | |
597 | spin_unlock_irqrestore(&cgwb_lock, flags); | |
598 | if (ret) { | |
599 | if (ret == -EEXIST) | |
600 | ret = 0; | |
a13f35e8 | 601 | goto err_fprop_exit; |
52ebea74 TH |
602 | } |
603 | goto out_put; | |
604 | ||
841710aa | 605 | err_fprop_exit: |
efee1713 | 606 | bdi_put(bdi); |
841710aa | 607 | fprop_local_destroy_percpu(&wb->memcg_completions); |
52ebea74 TH |
608 | err_ref_exit: |
609 | percpu_ref_exit(&wb->refcnt); | |
610 | err_wb_exit: | |
611 | wb_exit(wb); | |
612 | err_free: | |
613 | kfree(wb); | |
614 | out_put: | |
615 | css_put(blkcg_css); | |
616 | return ret; | |
66f3b8e2 JA |
617 | } |
618 | ||
52ebea74 | 619 | /** |
ed288dc0 | 620 | * wb_get_lookup - get wb for a given memcg |
52ebea74 TH |
621 | * @bdi: target bdi |
622 | * @memcg_css: cgroup_subsys_state of the target memcg (must have positive ref) | |
52ebea74 | 623 | * |
ed288dc0 TH |
624 | * Try to get the wb for @memcg_css on @bdi. The returned wb has its |
625 | * refcount incremented. | |
52ebea74 TH |
626 | * |
627 | * This function uses css_get() on @memcg_css and thus expects its refcnt | |
628 | * to be positive on invocation. IOW, rcu_read_lock() protection on | |
629 | * @memcg_css isn't enough. try_get it before calling this function. | |
630 | * | |
631 | * A wb is keyed by its associated memcg. As blkcg implicitly enables | |
632 | * memcg on the default hierarchy, memcg association is guaranteed to be | |
633 | * more specific (equal or descendant to the associated blkcg) and thus can | |
634 | * identify both the memcg and blkcg associations. | |
635 | * | |
636 | * Because the blkcg associated with a memcg may change as blkcg is enabled | |
637 | * and disabled closer to root in the hierarchy, each wb keeps track of | |
638 | * both the memcg and blkcg associated with it and verifies the blkcg on | |
639 | * each lookup. On mismatch, the existing wb is discarded and a new one is | |
640 | * created. | |
641 | */ | |
ed288dc0 TH |
642 | struct bdi_writeback *wb_get_lookup(struct backing_dev_info *bdi, |
643 | struct cgroup_subsys_state *memcg_css) | |
644 | { | |
645 | struct bdi_writeback *wb; | |
646 | ||
647 | if (!memcg_css->parent) | |
648 | return &bdi->wb; | |
649 | ||
650 | rcu_read_lock(); | |
651 | wb = radix_tree_lookup(&bdi->cgwb_tree, memcg_css->id); | |
652 | if (wb) { | |
653 | struct cgroup_subsys_state *blkcg_css; | |
654 | ||
655 | /* see whether the blkcg association has changed */ | |
656 | blkcg_css = cgroup_get_e_css(memcg_css->cgroup, &io_cgrp_subsys); | |
657 | if (unlikely(wb->blkcg_css != blkcg_css || !wb_tryget(wb))) | |
658 | wb = NULL; | |
659 | css_put(blkcg_css); | |
660 | } | |
661 | rcu_read_unlock(); | |
662 | ||
663 | return wb; | |
664 | } | |
665 | ||
666 | /** | |
667 | * wb_get_create - get wb for a given memcg, create if necessary | |
668 | * @bdi: target bdi | |
669 | * @memcg_css: cgroup_subsys_state of the target memcg (must have positive ref) | |
670 | * @gfp: allocation mask to use | |
671 | * | |
672 | * Try to get the wb for @memcg_css on @bdi. If it doesn't exist, try to | |
673 | * create one. See wb_get_lookup() for more details. | |
674 | */ | |
52ebea74 TH |
675 | struct bdi_writeback *wb_get_create(struct backing_dev_info *bdi, |
676 | struct cgroup_subsys_state *memcg_css, | |
677 | gfp_t gfp) | |
6467716a | 678 | { |
52ebea74 TH |
679 | struct bdi_writeback *wb; |
680 | ||
c1ca59a1 | 681 | might_alloc(gfp); |
52ebea74 TH |
682 | |
683 | if (!memcg_css->parent) | |
684 | return &bdi->wb; | |
685 | ||
686 | do { | |
ed288dc0 | 687 | wb = wb_get_lookup(bdi, memcg_css); |
52ebea74 TH |
688 | } while (!wb && !cgwb_create(bdi, memcg_css, gfp)); |
689 | ||
690 | return wb; | |
691 | } | |
6467716a | 692 | |
a13f35e8 | 693 | static int cgwb_bdi_init(struct backing_dev_info *bdi) |
52ebea74 | 694 | { |
a13f35e8 TH |
695 | int ret; |
696 | ||
52ebea74 | 697 | INIT_RADIX_TREE(&bdi->cgwb_tree, GFP_ATOMIC); |
3ee7e869 | 698 | mutex_init(&bdi->cgwb_release_mutex); |
7fc5854f | 699 | init_rwsem(&bdi->wb_switch_rwsem); |
a13f35e8 | 700 | |
8c911f3d | 701 | ret = wb_init(&bdi->wb, bdi, GFP_KERNEL); |
a13f35e8 | 702 | if (!ret) { |
7d828602 | 703 | bdi->wb.memcg_css = &root_mem_cgroup->css; |
a13f35e8 TH |
704 | bdi->wb.blkcg_css = blkcg_root_css; |
705 | } | |
706 | return ret; | |
6467716a AB |
707 | } |
708 | ||
b1c51afc | 709 | static void cgwb_bdi_unregister(struct backing_dev_info *bdi) |
52ebea74 TH |
710 | { |
711 | struct radix_tree_iter iter; | |
712 | void **slot; | |
5318ce7d | 713 | struct bdi_writeback *wb; |
52ebea74 TH |
714 | |
715 | WARN_ON(test_bit(WB_registered, &bdi->wb.state)); | |
716 | ||
717 | spin_lock_irq(&cgwb_lock); | |
718 | radix_tree_for_each_slot(slot, &bdi->cgwb_tree, &iter, 0) | |
719 | cgwb_kill(*slot); | |
3ee7e869 | 720 | spin_unlock_irq(&cgwb_lock); |
5318ce7d | 721 | |
3ee7e869 JK |
722 | mutex_lock(&bdi->cgwb_release_mutex); |
723 | spin_lock_irq(&cgwb_lock); | |
5318ce7d JK |
724 | while (!list_empty(&bdi->wb_list)) { |
725 | wb = list_first_entry(&bdi->wb_list, struct bdi_writeback, | |
726 | bdi_node); | |
727 | spin_unlock_irq(&cgwb_lock); | |
728 | wb_shutdown(wb); | |
729 | spin_lock_irq(&cgwb_lock); | |
730 | } | |
52ebea74 | 731 | spin_unlock_irq(&cgwb_lock); |
3ee7e869 | 732 | mutex_unlock(&bdi->cgwb_release_mutex); |
52ebea74 TH |
733 | } |
734 | ||
c22d70a1 RG |
735 | /* |
736 | * cleanup_offline_cgwbs_workfn - try to release dying cgwbs | |
737 | * | |
738 | * Try to release dying cgwbs by switching attached inodes to the nearest | |
739 | * living ancestor's writeback. Processed wbs are placed at the end | |
740 | * of the list to guarantee the forward progress. | |
741 | */ | |
742 | static void cleanup_offline_cgwbs_workfn(struct work_struct *work) | |
743 | { | |
744 | struct bdi_writeback *wb; | |
745 | LIST_HEAD(processed); | |
746 | ||
747 | spin_lock_irq(&cgwb_lock); | |
748 | ||
749 | while (!list_empty(&offline_cgwbs)) { | |
750 | wb = list_first_entry(&offline_cgwbs, struct bdi_writeback, | |
751 | offline_node); | |
752 | list_move(&wb->offline_node, &processed); | |
753 | ||
754 | /* | |
755 | * If wb is dirty, cleaning up the writeback by switching | |
756 | * attached inodes will result in an effective removal of any | |
757 | * bandwidth restrictions, which isn't the goal. Instead, | |
758 | * it can be postponed until the next time, when all io | |
759 | * will be likely completed. If in the meantime some inodes | |
760 | * will get re-dirtied, they should be eventually switched to | |
761 | * a new cgwb. | |
762 | */ | |
763 | if (wb_has_dirty_io(wb)) | |
764 | continue; | |
765 | ||
766 | if (!wb_tryget(wb)) | |
767 | continue; | |
768 | ||
769 | spin_unlock_irq(&cgwb_lock); | |
770 | while (cleanup_offline_cgwb(wb)) | |
771 | cond_resched(); | |
772 | spin_lock_irq(&cgwb_lock); | |
773 | ||
774 | wb_put(wb); | |
775 | } | |
776 | ||
777 | if (!list_empty(&processed)) | |
778 | list_splice_tail(&processed, &offline_cgwbs); | |
779 | ||
780 | spin_unlock_irq(&cgwb_lock); | |
781 | } | |
782 | ||
52ebea74 TH |
783 | /** |
784 | * wb_memcg_offline - kill all wb's associated with a memcg being offlined | |
785 | * @memcg: memcg being offlined | |
786 | * | |
787 | * Also prevents creation of any new wb's associated with @memcg. | |
e98be2d5 | 788 | */ |
52ebea74 TH |
789 | void wb_memcg_offline(struct mem_cgroup *memcg) |
790 | { | |
9ccc3617 | 791 | struct list_head *memcg_cgwb_list = &memcg->cgwb_list; |
52ebea74 TH |
792 | struct bdi_writeback *wb, *next; |
793 | ||
794 | spin_lock_irq(&cgwb_lock); | |
795 | list_for_each_entry_safe(wb, next, memcg_cgwb_list, memcg_node) | |
796 | cgwb_kill(wb); | |
797 | memcg_cgwb_list->next = NULL; /* prevent new wb's */ | |
798 | spin_unlock_irq(&cgwb_lock); | |
c22d70a1 RG |
799 | |
800 | queue_work(system_unbound_wq, &cleanup_offline_cgwbs_work); | |
52ebea74 TH |
801 | } |
802 | ||
803 | /** | |
804 | * wb_blkcg_offline - kill all wb's associated with a blkcg being offlined | |
dec223c9 | 805 | * @css: blkcg being offlined |
52ebea74 TH |
806 | * |
807 | * Also prevents creation of any new wb's associated with @blkcg. | |
808 | */ | |
dec223c9 | 809 | void wb_blkcg_offline(struct cgroup_subsys_state *css) |
52ebea74 | 810 | { |
52ebea74 | 811 | struct bdi_writeback *wb, *next; |
dec223c9 | 812 | struct list_head *list = blkcg_get_cgwb_list(css); |
52ebea74 TH |
813 | |
814 | spin_lock_irq(&cgwb_lock); | |
dec223c9 | 815 | list_for_each_entry_safe(wb, next, list, blkcg_node) |
52ebea74 | 816 | cgwb_kill(wb); |
dec223c9 | 817 | list->next = NULL; /* prevent new wb's */ |
52ebea74 TH |
818 | spin_unlock_irq(&cgwb_lock); |
819 | } | |
820 | ||
e8cb72b3 JK |
821 | static void cgwb_bdi_register(struct backing_dev_info *bdi) |
822 | { | |
823 | spin_lock_irq(&cgwb_lock); | |
824 | list_add_tail_rcu(&bdi->wb.bdi_node, &bdi->wb_list); | |
825 | spin_unlock_irq(&cgwb_lock); | |
826 | } | |
827 | ||
f1834646 TH |
828 | static int __init cgwb_init(void) |
829 | { | |
830 | /* | |
831 | * There can be many concurrent release work items overwhelming | |
832 | * system_wq. Put them in a separate wq and limit concurrency. | |
833 | * There's no point in executing many of these in parallel. | |
834 | */ | |
835 | cgwb_release_wq = alloc_workqueue("cgwb_release", 0, 1); | |
836 | if (!cgwb_release_wq) | |
837 | return -ENOMEM; | |
838 | ||
839 | return 0; | |
840 | } | |
841 | subsys_initcall(cgwb_init); | |
842 | ||
52ebea74 TH |
843 | #else /* CONFIG_CGROUP_WRITEBACK */ |
844 | ||
a13f35e8 TH |
845 | static int cgwb_bdi_init(struct backing_dev_info *bdi) |
846 | { | |
8c911f3d | 847 | return wb_init(&bdi->wb, bdi, GFP_KERNEL); |
a13f35e8 TH |
848 | } |
849 | ||
b1c51afc | 850 | static void cgwb_bdi_unregister(struct backing_dev_info *bdi) { } |
df23de55 | 851 | |
e8cb72b3 JK |
852 | static void cgwb_bdi_register(struct backing_dev_info *bdi) |
853 | { | |
854 | list_add_tail_rcu(&bdi->wb.bdi_node, &bdi->wb_list); | |
855 | } | |
856 | ||
857 | static void cgwb_remove_from_bdi_list(struct bdi_writeback *wb) | |
858 | { | |
859 | list_del_rcu(&wb->bdi_node); | |
860 | } | |
861 | ||
52ebea74 | 862 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
e98be2d5 | 863 | |
4bca7e80 | 864 | int bdi_init(struct backing_dev_info *bdi) |
b2e8fb6e | 865 | { |
cf0ca9fe PZ |
866 | bdi->dev = NULL; |
867 | ||
d03f6cdc | 868 | kref_init(&bdi->refcnt); |
189d3c4a | 869 | bdi->min_ratio = 0; |
ae82291e | 870 | bdi->max_ratio = 100 * BDI_RATIO_SCALE; |
eb608e3a | 871 | bdi->max_prop_frac = FPROP_FRAC_BASE; |
66f3b8e2 | 872 | INIT_LIST_HEAD(&bdi->bdi_list); |
b817525a | 873 | INIT_LIST_HEAD(&bdi->wb_list); |
cc395d7f | 874 | init_waitqueue_head(&bdi->wb_waitq); |
03ba3782 | 875 | |
3083da7b | 876 | return cgwb_bdi_init(bdi); |
b2e8fb6e | 877 | } |
e98be2d5 | 878 | |
aef33c2f | 879 | struct backing_dev_info *bdi_alloc(int node_id) |
d03f6cdc JK |
880 | { |
881 | struct backing_dev_info *bdi; | |
882 | ||
aef33c2f | 883 | bdi = kzalloc_node(sizeof(*bdi), GFP_KERNEL, node_id); |
d03f6cdc JK |
884 | if (!bdi) |
885 | return NULL; | |
886 | ||
887 | if (bdi_init(bdi)) { | |
888 | kfree(bdi); | |
889 | return NULL; | |
890 | } | |
f56753ac | 891 | bdi->capabilities = BDI_CAP_WRITEBACK | BDI_CAP_WRITEBACK_ACCT; |
55b2598e CH |
892 | bdi->ra_pages = VM_READAHEAD_PAGES; |
893 | bdi->io_pages = VM_READAHEAD_PAGES; | |
5ed964f8 | 894 | timer_setup(&bdi->laptop_mode_wb_timer, laptop_mode_timer_fn, 0); |
d03f6cdc JK |
895 | return bdi; |
896 | } | |
aef33c2f | 897 | EXPORT_SYMBOL(bdi_alloc); |
d03f6cdc | 898 | |
34f8fe50 TH |
899 | static struct rb_node **bdi_lookup_rb_node(u64 id, struct rb_node **parentp) |
900 | { | |
901 | struct rb_node **p = &bdi_tree.rb_node; | |
902 | struct rb_node *parent = NULL; | |
903 | struct backing_dev_info *bdi; | |
904 | ||
905 | lockdep_assert_held(&bdi_lock); | |
906 | ||
907 | while (*p) { | |
908 | parent = *p; | |
909 | bdi = rb_entry(parent, struct backing_dev_info, rb_node); | |
910 | ||
911 | if (bdi->id > id) | |
912 | p = &(*p)->rb_left; | |
913 | else if (bdi->id < id) | |
914 | p = &(*p)->rb_right; | |
915 | else | |
916 | break; | |
917 | } | |
918 | ||
919 | if (parentp) | |
920 | *parentp = parent; | |
921 | return p; | |
922 | } | |
923 | ||
924 | /** | |
925 | * bdi_get_by_id - lookup and get bdi from its id | |
926 | * @id: bdi id to lookup | |
927 | * | |
928 | * Find bdi matching @id and get it. Returns NULL if the matching bdi | |
929 | * doesn't exist or is already unregistered. | |
930 | */ | |
931 | struct backing_dev_info *bdi_get_by_id(u64 id) | |
932 | { | |
933 | struct backing_dev_info *bdi = NULL; | |
934 | struct rb_node **p; | |
935 | ||
936 | spin_lock_bh(&bdi_lock); | |
937 | p = bdi_lookup_rb_node(id, NULL); | |
938 | if (*p) { | |
939 | bdi = rb_entry(*p, struct backing_dev_info, rb_node); | |
940 | bdi_get(bdi); | |
941 | } | |
942 | spin_unlock_bh(&bdi_lock); | |
943 | ||
944 | return bdi; | |
945 | } | |
946 | ||
7c4cc300 | 947 | int bdi_register_va(struct backing_dev_info *bdi, const char *fmt, va_list args) |
46100071 | 948 | { |
46100071 | 949 | struct device *dev; |
34f8fe50 | 950 | struct rb_node *parent, **p; |
e98be2d5 | 951 | |
46100071 TH |
952 | if (bdi->dev) /* The driver needs to use separate queues per device */ |
953 | return 0; | |
e98be2d5 | 954 | |
6bd87eec CH |
955 | vsnprintf(bdi->dev_name, sizeof(bdi->dev_name), fmt, args); |
956 | dev = device_create(bdi_class, NULL, MKDEV(0, 0), bdi, bdi->dev_name); | |
46100071 TH |
957 | if (IS_ERR(dev)) |
958 | return PTR_ERR(dev); | |
04fbfdc1 | 959 | |
e8cb72b3 | 960 | cgwb_bdi_register(bdi); |
46100071 | 961 | bdi->dev = dev; |
b2e8fb6e | 962 | |
6d0e4827 | 963 | bdi_debug_register(bdi, dev_name(dev)); |
46100071 TH |
964 | set_bit(WB_registered, &bdi->wb.state); |
965 | ||
966 | spin_lock_bh(&bdi_lock); | |
34f8fe50 TH |
967 | |
968 | bdi->id = ++bdi_id_cursor; | |
969 | ||
970 | p = bdi_lookup_rb_node(bdi->id, &parent); | |
971 | rb_link_node(&bdi->rb_node, parent, p); | |
972 | rb_insert_color(&bdi->rb_node, &bdi_tree); | |
973 | ||
46100071 | 974 | list_add_tail_rcu(&bdi->bdi_list, &bdi_list); |
34f8fe50 | 975 | |
46100071 TH |
976 | spin_unlock_bh(&bdi_lock); |
977 | ||
978 | trace_writeback_bdi_register(bdi); | |
979 | return 0; | |
b2e8fb6e | 980 | } |
baf7a616 | 981 | |
7c4cc300 | 982 | int bdi_register(struct backing_dev_info *bdi, const char *fmt, ...) |
baf7a616 JK |
983 | { |
984 | va_list args; | |
985 | int ret; | |
986 | ||
987 | va_start(args, fmt); | |
7c4cc300 | 988 | ret = bdi_register_va(bdi, fmt, args); |
baf7a616 JK |
989 | va_end(args); |
990 | return ret; | |
991 | } | |
46100071 | 992 | EXPORT_SYMBOL(bdi_register); |
b2e8fb6e | 993 | |
3c5d202b | 994 | void bdi_set_owner(struct backing_dev_info *bdi, struct device *owner) |
df08c32c | 995 | { |
3c5d202b | 996 | WARN_ON_ONCE(bdi->owner); |
df08c32c DW |
997 | bdi->owner = owner; |
998 | get_device(owner); | |
df08c32c | 999 | } |
df08c32c | 1000 | |
46100071 TH |
1001 | /* |
1002 | * Remove bdi from bdi_list, and ensure that it is no longer visible | |
1003 | */ | |
1004 | static void bdi_remove_from_list(struct backing_dev_info *bdi) | |
1005 | { | |
1006 | spin_lock_bh(&bdi_lock); | |
34f8fe50 | 1007 | rb_erase(&bdi->rb_node, &bdi_tree); |
46100071 TH |
1008 | list_del_rcu(&bdi->bdi_list); |
1009 | spin_unlock_bh(&bdi_lock); | |
b2e8fb6e | 1010 | |
46100071 TH |
1011 | synchronize_rcu_expedited(); |
1012 | } | |
cf0ca9fe | 1013 | |
b02176f3 | 1014 | void bdi_unregister(struct backing_dev_info *bdi) |
b2e8fb6e | 1015 | { |
5ed964f8 CH |
1016 | del_timer_sync(&bdi->laptop_mode_wb_timer); |
1017 | ||
f0054bb1 TH |
1018 | /* make sure nobody finds us on the bdi_list anymore */ |
1019 | bdi_remove_from_list(bdi); | |
1020 | wb_shutdown(&bdi->wb); | |
b1c51afc | 1021 | cgwb_bdi_unregister(bdi); |
7a401a97 | 1022 | |
3c376dfa ML |
1023 | /* |
1024 | * If this BDI's min ratio has been set, use bdi_set_min_ratio() to | |
1025 | * update the global bdi_min_ratio. | |
1026 | */ | |
1027 | if (bdi->min_ratio) | |
1028 | bdi_set_min_ratio(bdi, 0); | |
1029 | ||
c4db59d3 CH |
1030 | if (bdi->dev) { |
1031 | bdi_debug_unregister(bdi); | |
1032 | device_unregister(bdi->dev); | |
1033 | bdi->dev = NULL; | |
1034 | } | |
df08c32c DW |
1035 | |
1036 | if (bdi->owner) { | |
1037 | put_device(bdi->owner); | |
1038 | bdi->owner = NULL; | |
1039 | } | |
b02176f3 | 1040 | } |
c6fd3ac0 | 1041 | EXPORT_SYMBOL(bdi_unregister); |
c4db59d3 | 1042 | |
d03f6cdc JK |
1043 | static void release_bdi(struct kref *ref) |
1044 | { | |
1045 | struct backing_dev_info *bdi = | |
1046 | container_of(ref, struct backing_dev_info, refcnt); | |
1047 | ||
702f2d1e | 1048 | WARN_ON_ONCE(test_bit(WB_registered, &bdi->wb.state)); |
2e82b84c JK |
1049 | WARN_ON_ONCE(bdi->dev); |
1050 | wb_exit(&bdi->wb); | |
d03f6cdc JK |
1051 | kfree(bdi); |
1052 | } | |
1053 | ||
1054 | void bdi_put(struct backing_dev_info *bdi) | |
1055 | { | |
1056 | kref_put(&bdi->refcnt, release_bdi); | |
1057 | } | |
62bf42ad | 1058 | EXPORT_SYMBOL(bdi_put); |
d03f6cdc | 1059 | |
ccdf7741 CH |
1060 | struct backing_dev_info *inode_to_bdi(struct inode *inode) |
1061 | { | |
1062 | struct super_block *sb; | |
1063 | ||
1064 | if (!inode) | |
1065 | return &noop_backing_dev_info; | |
1066 | ||
1067 | sb = inode->i_sb; | |
1068 | #ifdef CONFIG_BLOCK | |
1069 | if (sb_is_blkdev_sb(sb)) | |
1070 | return I_BDEV(inode)->bd_disk->bdi; | |
1071 | #endif | |
1072 | return sb->s_bdi; | |
1073 | } | |
1074 | EXPORT_SYMBOL(inode_to_bdi); | |
1075 | ||
eb7ae5e0 CH |
1076 | const char *bdi_dev_name(struct backing_dev_info *bdi) |
1077 | { | |
1078 | if (!bdi || !bdi->dev) | |
1079 | return bdi_unknown_name; | |
6bd87eec | 1080 | return bdi->dev_name; |
eb7ae5e0 CH |
1081 | } |
1082 | EXPORT_SYMBOL_GPL(bdi_dev_name); |