Commit | Line | Data |
---|---|---|
3dcf60bc | 1 | // SPDX-License-Identifier: GPL-2.0 |
31e4c28d VG |
2 | /* |
3 | * Common Block IO controller cgroup interface | |
4 | * | |
5 | * Based on ideas and code from CFQ, CFS and BFQ: | |
6 | * Copyright (C) 2003 Jens Axboe <axboe@kernel.dk> | |
7 | * | |
8 | * Copyright (C) 2008 Fabio Checconi <fabio@gandalf.sssup.it> | |
9 | * Paolo Valente <paolo.valente@unimore.it> | |
10 | * | |
11 | * Copyright (C) 2009 Vivek Goyal <vgoyal@redhat.com> | |
12 | * Nauman Rafique <nauman@google.com> | |
e48453c3 AA |
13 | * |
14 | * For policy-specific per-blkcg data: | |
15 | * Copyright (C) 2015 Paolo Valente <paolo.valente@unimore.it> | |
16 | * Arianna Avanzini <avanzini.arianna@gmail.com> | |
31e4c28d VG |
17 | */ |
18 | #include <linux/ioprio.h> | |
22084190 | 19 | #include <linux/kdev_t.h> |
9d6a986c | 20 | #include <linux/module.h> |
174cd4b1 | 21 | #include <linux/sched/signal.h> |
accee785 | 22 | #include <linux/err.h> |
9195291e | 23 | #include <linux/blkdev.h> |
52ebea74 | 24 | #include <linux/backing-dev.h> |
5a0e3ad6 | 25 | #include <linux/slab.h> |
72e06c25 | 26 | #include <linux/delay.h> |
9a9e8a26 | 27 | #include <linux/atomic.h> |
36aa9e5f | 28 | #include <linux/ctype.h> |
03248add | 29 | #include <linux/resume_user_mode.h> |
fd112c74 | 30 | #include <linux/psi.h> |
82d981d4 | 31 | #include <linux/part_stat.h> |
5efd6113 | 32 | #include "blk.h" |
672fdcf0 | 33 | #include "blk-cgroup.h" |
556910e3 | 34 | #include "blk-ioprio.h" |
a7b36ee6 | 35 | #include "blk-throttle.h" |
3e252066 | 36 | |
20cb1c2f ML |
37 | static void __blkcg_rstat_flush(struct blkcg *blkcg, int cpu); |
38 | ||
838f13bf TH |
39 | /* |
40 | * blkcg_pol_mutex protects blkcg_policy[] and policy [de]activation. | |
41 | * blkcg_pol_register_mutex nests outside of it and synchronizes entire | |
42 | * policy [un]register operations including cgroup file additions / | |
43 | * removals. Putting cgroup file registration outside blkcg_pol_mutex | |
44 | * allows grabbing it from cgroup callbacks. | |
45 | */ | |
46 | static DEFINE_MUTEX(blkcg_pol_register_mutex); | |
bc0d6501 | 47 | static DEFINE_MUTEX(blkcg_pol_mutex); |
923adde1 | 48 | |
e48453c3 | 49 | struct blkcg blkcg_root; |
3c798398 | 50 | EXPORT_SYMBOL_GPL(blkcg_root); |
9d6a986c | 51 | |
496d5e75 | 52 | struct cgroup_subsys_state * const blkcg_root_css = &blkcg_root.css; |
9b0eb69b | 53 | EXPORT_SYMBOL_GPL(blkcg_root_css); |
496d5e75 | 54 | |
3c798398 | 55 | static struct blkcg_policy *blkcg_policy[BLKCG_MAX_POLS]; |
035d10b2 | 56 | |
7876f930 TH |
57 | static LIST_HEAD(all_blkcgs); /* protected by blkcg_pol_mutex */ |
58 | ||
07b0fdec | 59 | bool blkcg_debug_stats = false; |
903d23f0 | 60 | |
20cb1c2f ML |
61 | static DEFINE_RAW_SPINLOCK(blkg_stat_lock); |
62 | ||
a731763f YK |
63 | #define BLKG_DESTROY_BATCH_SIZE 64 |
64 | ||
3b8cc629 WL |
65 | /* |
66 | * Lockless lists for tracking IO stats update | |
67 | * | |
68 | * New IO stats are stored in the percpu iostat_cpu within blkcg_gq (blkg). | |
69 | * There are multiple blkg's (one for each block device) attached to each | |
70 | * blkcg. The rstat code keeps track of which cpu has IO stats updated, | |
71 | * but it doesn't know which blkg has the updated stats. If there are many | |
72 | * block devices in a system, the cost of iterating all the blkg's to flush | |
73 | * out the IO stats can be high. To reduce such overhead, a set of percpu | |
74 | * lockless lists (lhead) per blkcg are used to track the set of recently | |
75 | * updated iostat_cpu's since the last flush. An iostat_cpu will be put | |
76 | * onto the lockless list on the update side [blk_cgroup_bio_start()] if | |
77 | * not there yet and then removed when being flushed [blkcg_rstat_flush()]. | |
78 | * References to blkg are gotten and then put back in the process to | |
79 | * protect against blkg removal. | |
80 | * | |
81 | * Return: 0 if successful or -ENOMEM if allocation fails. | |
82 | */ | |
83 | static int init_blkcg_llists(struct blkcg *blkcg) | |
84 | { | |
85 | int cpu; | |
86 | ||
87 | blkcg->lhead = alloc_percpu_gfp(struct llist_head, GFP_KERNEL); | |
88 | if (!blkcg->lhead) | |
89 | return -ENOMEM; | |
90 | ||
91 | for_each_possible_cpu(cpu) | |
92 | init_llist_head(per_cpu_ptr(blkcg->lhead, cpu)); | |
93 | return 0; | |
94 | } | |
95 | ||
bc5fee91 CH |
96 | /** |
97 | * blkcg_css - find the current css | |
98 | * | |
99 | * Find the css associated with either the kthread or the current task. | |
100 | * This may return a dying css, so it is up to the caller to use tryget logic | |
101 | * to confirm it is alive and well. | |
102 | */ | |
103 | static struct cgroup_subsys_state *blkcg_css(void) | |
104 | { | |
105 | struct cgroup_subsys_state *css; | |
106 | ||
107 | css = kthread_blkcg(); | |
108 | if (css) | |
109 | return css; | |
110 | return task_css(current, io_cgrp_id); | |
111 | } | |
112 | ||
a2b1693b | 113 | static bool blkcg_policy_enabled(struct request_queue *q, |
3c798398 | 114 | const struct blkcg_policy *pol) |
a2b1693b TH |
115 | { |
116 | return pol && test_bit(pol->plid, q->blkcg_pols); | |
117 | } | |
118 | ||
d578c770 | 119 | static void blkg_free_workfn(struct work_struct *work) |
0381411e | 120 | { |
d578c770 ML |
121 | struct blkcg_gq *blkg = container_of(work, struct blkcg_gq, |
122 | free_work); | |
a06377c5 | 123 | struct request_queue *q = blkg->q; |
e8989fae | 124 | int i; |
549d3aa8 | 125 | |
f1c006f1 YK |
126 | /* |
127 | * pd_free_fn() can also be called from blkcg_deactivate_policy(), | |
128 | * in order to make sure pd_free_fn() is called in order, the deletion | |
1231039d | 129 | * of the list blkg->q_node is delayed to here from blkg_destroy(), and |
f1c006f1 YK |
130 | * blkcg_mutex is used to synchronize blkg_free_workfn() and |
131 | * blkcg_deactivate_policy(). | |
132 | */ | |
a06377c5 | 133 | mutex_lock(&q->blkcg_mutex); |
db613670 | 134 | for (i = 0; i < BLKCG_MAX_POLS; i++) |
001bea73 TH |
135 | if (blkg->pd[i]) |
136 | blkcg_policy[i]->pd_free_fn(blkg->pd[i]); | |
c7241bab YK |
137 | if (blkg->parent) |
138 | blkg_put(blkg->parent); | |
c164c7bc | 139 | spin_lock_irq(&q->queue_lock); |
1231039d | 140 | list_del_init(&blkg->q_node); |
c164c7bc | 141 | spin_unlock_irq(&q->queue_lock); |
a06377c5 | 142 | mutex_unlock(&q->blkcg_mutex); |
e8989fae | 143 | |
a06377c5 | 144 | blk_put_queue(q); |
f7331648 | 145 | free_percpu(blkg->iostat_cpu); |
ef069b97 | 146 | percpu_ref_exit(&blkg->refcnt); |
549d3aa8 | 147 | kfree(blkg); |
0381411e TH |
148 | } |
149 | ||
d578c770 ML |
150 | /** |
151 | * blkg_free - free a blkg | |
152 | * @blkg: blkg to free | |
153 | * | |
154 | * Free @blkg which may be partially allocated. | |
155 | */ | |
156 | static void blkg_free(struct blkcg_gq *blkg) | |
157 | { | |
158 | if (!blkg) | |
159 | return; | |
160 | ||
161 | /* | |
162 | * Both ->pd_free_fn() and request queue's release handler may | |
163 | * sleep, so free us by scheduling one work func | |
164 | */ | |
165 | INIT_WORK(&blkg->free_work, blkg_free_workfn); | |
166 | schedule_work(&blkg->free_work); | |
167 | } | |
168 | ||
7fcf2b03 DZ |
169 | static void __blkg_release(struct rcu_head *rcu) |
170 | { | |
171 | struct blkcg_gq *blkg = container_of(rcu, struct blkcg_gq, rcu_head); | |
20cb1c2f ML |
172 | struct blkcg *blkcg = blkg->blkcg; |
173 | int cpu; | |
7fcf2b03 | 174 | |
2c275afe | 175 | #ifdef CONFIG_BLK_CGROUP_PUNT_BIO |
d3f77dfd | 176 | WARN_ON(!bio_list_empty(&blkg->async_bios)); |
2c275afe | 177 | #endif |
20cb1c2f ML |
178 | /* |
179 | * Flush all the non-empty percpu lockless lists before releasing | |
180 | * us, given these stat belongs to us. | |
181 | * | |
182 | * blkg_stat_lock is for serializing blkg stat update | |
183 | */ | |
184 | for_each_possible_cpu(cpu) | |
185 | __blkcg_rstat_flush(blkcg, cpu); | |
d3f77dfd | 186 | |
7fcf2b03 DZ |
187 | /* release the blkcg and parent blkg refs this blkg has been holding */ |
188 | css_put(&blkg->blkcg->css); | |
7fcf2b03 DZ |
189 | blkg_free(blkg); |
190 | } | |
191 | ||
192 | /* | |
193 | * A group is RCU protected, but having an rcu lock does not mean that one | |
194 | * can access all the fields of blkg and assume these are valid. For | |
195 | * example, don't try to follow throtl_data and request queue links. | |
196 | * | |
197 | * Having a reference to blkg under an rcu allows accesses to only values | |
198 | * local to groups like group stats and group rate limits. | |
199 | */ | |
200 | static void blkg_release(struct percpu_ref *ref) | |
201 | { | |
202 | struct blkcg_gq *blkg = container_of(ref, struct blkcg_gq, refcnt); | |
203 | ||
204 | call_rcu(&blkg->rcu_head, __blkg_release); | |
205 | } | |
206 | ||
2c275afe CH |
207 | #ifdef CONFIG_BLK_CGROUP_PUNT_BIO |
208 | static struct workqueue_struct *blkcg_punt_bio_wq; | |
209 | ||
d3f77dfd TH |
210 | static void blkg_async_bio_workfn(struct work_struct *work) |
211 | { | |
212 | struct blkcg_gq *blkg = container_of(work, struct blkcg_gq, | |
213 | async_bio_work); | |
214 | struct bio_list bios = BIO_EMPTY_LIST; | |
215 | struct bio *bio; | |
192f1c6b XT |
216 | struct blk_plug plug; |
217 | bool need_plug = false; | |
d3f77dfd TH |
218 | |
219 | /* as long as there are pending bios, @blkg can't go away */ | |
12be09fe | 220 | spin_lock(&blkg->async_bio_lock); |
7a324d83 | 221 | bio_list_merge_init(&bios, &blkg->async_bios); |
12be09fe | 222 | spin_unlock(&blkg->async_bio_lock); |
d3f77dfd | 223 | |
192f1c6b XT |
224 | /* start plug only when bio_list contains at least 2 bios */ |
225 | if (bios.head && bios.head->bi_next) { | |
226 | need_plug = true; | |
227 | blk_start_plug(&plug); | |
228 | } | |
d3f77dfd TH |
229 | while ((bio = bio_list_pop(&bios))) |
230 | submit_bio(bio); | |
192f1c6b XT |
231 | if (need_plug) |
232 | blk_finish_plug(&plug); | |
d3f77dfd TH |
233 | } |
234 | ||
2c275afe CH |
235 | /* |
236 | * When a shared kthread issues a bio for a cgroup, doing so synchronously can | |
237 | * lead to priority inversions as the kthread can be trapped waiting for that | |
238 | * cgroup. Use this helper instead of submit_bio to punt the actual issuing to | |
239 | * a dedicated per-blkcg work item to avoid such priority inversions. | |
240 | */ | |
241 | void blkcg_punt_bio_submit(struct bio *bio) | |
242 | { | |
243 | struct blkcg_gq *blkg = bio->bi_blkg; | |
244 | ||
245 | if (blkg->parent) { | |
246 | spin_lock(&blkg->async_bio_lock); | |
247 | bio_list_add(&blkg->async_bios, bio); | |
248 | spin_unlock(&blkg->async_bio_lock); | |
249 | queue_work(blkcg_punt_bio_wq, &blkg->async_bio_work); | |
250 | } else { | |
251 | /* never bounce for the root cgroup */ | |
252 | submit_bio(bio); | |
253 | } | |
254 | } | |
255 | EXPORT_SYMBOL_GPL(blkcg_punt_bio_submit); | |
256 | ||
257 | static int __init blkcg_punt_bio_init(void) | |
258 | { | |
259 | blkcg_punt_bio_wq = alloc_workqueue("blkcg_punt_bio", | |
260 | WQ_MEM_RECLAIM | WQ_FREEZABLE | | |
261 | WQ_UNBOUND | WQ_SYSFS, 0); | |
262 | if (!blkcg_punt_bio_wq) | |
263 | return -ENOMEM; | |
264 | return 0; | |
265 | } | |
266 | subsys_initcall(blkcg_punt_bio_init); | |
267 | #endif /* CONFIG_BLK_CGROUP_PUNT_BIO */ | |
268 | ||
bbb1ebe7 CH |
269 | /** |
270 | * bio_blkcg_css - return the blkcg CSS associated with a bio | |
271 | * @bio: target bio | |
272 | * | |
273 | * This returns the CSS for the blkcg associated with a bio, or %NULL if not | |
274 | * associated. Callers are expected to either handle %NULL or know association | |
275 | * has been done prior to calling this. | |
276 | */ | |
277 | struct cgroup_subsys_state *bio_blkcg_css(struct bio *bio) | |
278 | { | |
279 | if (!bio || !bio->bi_blkg) | |
280 | return NULL; | |
281 | return &bio->bi_blkg->blkcg->css; | |
282 | } | |
283 | EXPORT_SYMBOL_GPL(bio_blkcg_css); | |
284 | ||
397c9f46 CH |
285 | /** |
286 | * blkcg_parent - get the parent of a blkcg | |
287 | * @blkcg: blkcg of interest | |
288 | * | |
289 | * Return the parent blkcg of @blkcg. Can be called anytime. | |
290 | */ | |
291 | static inline struct blkcg *blkcg_parent(struct blkcg *blkcg) | |
292 | { | |
293 | return css_to_blkcg(blkcg->css.parent); | |
294 | } | |
295 | ||
0381411e TH |
296 | /** |
297 | * blkg_alloc - allocate a blkg | |
298 | * @blkcg: block cgroup the new blkg is associated with | |
99e60387 | 299 | * @disk: gendisk the new blkg is associated with |
15974993 | 300 | * @gfp_mask: allocation mask to use |
0381411e | 301 | * |
521277d1 | 302 | * Allocate a new blkg associating @blkcg and @disk. |
0381411e | 303 | */ |
99e60387 | 304 | static struct blkcg_gq *blkg_alloc(struct blkcg *blkcg, struct gendisk *disk, |
15974993 | 305 | gfp_t gfp_mask) |
0381411e | 306 | { |
3c798398 | 307 | struct blkcg_gq *blkg; |
f7331648 | 308 | int i, cpu; |
0381411e TH |
309 | |
310 | /* alloc and init base part */ | |
99e60387 | 311 | blkg = kzalloc_node(sizeof(*blkg), gfp_mask, disk->queue->node); |
0381411e TH |
312 | if (!blkg) |
313 | return NULL; | |
ef069b97 | 314 | if (percpu_ref_init(&blkg->refcnt, blkg_release, 0, gfp_mask)) |
0b6f93bd | 315 | goto out_free_blkg; |
f7331648 TH |
316 | blkg->iostat_cpu = alloc_percpu_gfp(struct blkg_iostat_set, gfp_mask); |
317 | if (!blkg->iostat_cpu) | |
0b6f93bd | 318 | goto out_exit_refcnt; |
99e60387 | 319 | if (!blk_get_queue(disk->queue)) |
0b6f93bd | 320 | goto out_free_iostat; |
0a9a25ca | 321 | |
99e60387 | 322 | blkg->q = disk->queue; |
e8989fae | 323 | INIT_LIST_HEAD(&blkg->q_node); |
2c275afe CH |
324 | blkg->blkcg = blkcg; |
325 | #ifdef CONFIG_BLK_CGROUP_PUNT_BIO | |
d3f77dfd TH |
326 | spin_lock_init(&blkg->async_bio_lock); |
327 | bio_list_init(&blkg->async_bios); | |
328 | INIT_WORK(&blkg->async_bio_work, blkg_async_bio_workfn); | |
2c275afe | 329 | #endif |
0381411e | 330 | |
f7331648 | 331 | u64_stats_init(&blkg->iostat.sync); |
3b8cc629 | 332 | for_each_possible_cpu(cpu) { |
f7331648 | 333 | u64_stats_init(&per_cpu_ptr(blkg->iostat_cpu, cpu)->sync); |
3b8cc629 WL |
334 | per_cpu_ptr(blkg->iostat_cpu, cpu)->blkg = blkg; |
335 | } | |
f7331648 | 336 | |
8bd435b3 | 337 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
3c798398 | 338 | struct blkcg_policy *pol = blkcg_policy[i]; |
e8989fae | 339 | struct blkg_policy_data *pd; |
0381411e | 340 | |
99e60387 | 341 | if (!blkcg_policy_enabled(disk->queue, pol)) |
e8989fae TH |
342 | continue; |
343 | ||
344 | /* alloc per-policy data and attach it to blkg */ | |
0a0b4f79 | 345 | pd = pol->pd_alloc_fn(disk, blkcg, gfp_mask); |
a051661c | 346 | if (!pd) |
0b6f93bd | 347 | goto out_free_pds; |
e8989fae TH |
348 | blkg->pd[i] = pd; |
349 | pd->blkg = blkg; | |
b276a876 | 350 | pd->plid = i; |
dfd6200a | 351 | pd->online = false; |
e8989fae TH |
352 | } |
353 | ||
0381411e | 354 | return blkg; |
a051661c | 355 | |
0b6f93bd CH |
356 | out_free_pds: |
357 | while (--i >= 0) | |
358 | if (blkg->pd[i]) | |
359 | blkcg_policy[i]->pd_free_fn(blkg->pd[i]); | |
a06377c5 | 360 | blk_put_queue(disk->queue); |
0b6f93bd CH |
361 | out_free_iostat: |
362 | free_percpu(blkg->iostat_cpu); | |
363 | out_exit_refcnt: | |
364 | percpu_ref_exit(&blkg->refcnt); | |
365 | out_free_blkg: | |
366 | kfree(blkg); | |
a051661c | 367 | return NULL; |
0381411e TH |
368 | } |
369 | ||
15974993 | 370 | /* |
d708f0d5 JA |
371 | * If @new_blkg is %NULL, this function tries to allocate a new one as |
372 | * necessary using %GFP_NOWAIT. @new_blkg is always consumed on return. | |
15974993 | 373 | */ |
99e60387 | 374 | static struct blkcg_gq *blkg_create(struct blkcg *blkcg, struct gendisk *disk, |
d708f0d5 | 375 | struct blkcg_gq *new_blkg) |
5624a4e4 | 376 | { |
d708f0d5 | 377 | struct blkcg_gq *blkg; |
f427d909 | 378 | int i, ret; |
5624a4e4 | 379 | |
99e60387 | 380 | lockdep_assert_held(&disk->queue->queue_lock); |
cd1604fa | 381 | |
0273ac34 | 382 | /* request_queue is dying, do not create/recreate a blkg */ |
99e60387 | 383 | if (blk_queue_dying(disk->queue)) { |
0273ac34 DZ |
384 | ret = -ENODEV; |
385 | goto err_free_blkg; | |
386 | } | |
387 | ||
7ee9c562 | 388 | /* blkg holds a reference to blkcg */ |
ec903c0c | 389 | if (!css_tryget_online(&blkcg->css)) { |
20386ce0 | 390 | ret = -ENODEV; |
93e6d5d8 | 391 | goto err_free_blkg; |
15974993 | 392 | } |
cd1604fa | 393 | |
d708f0d5 JA |
394 | /* allocate */ |
395 | if (!new_blkg) { | |
99e60387 | 396 | new_blkg = blkg_alloc(blkcg, disk, GFP_NOWAIT | __GFP_NOWARN); |
d708f0d5 JA |
397 | if (unlikely(!new_blkg)) { |
398 | ret = -ENOMEM; | |
8c911f3d | 399 | goto err_put_css; |
15974993 TH |
400 | } |
401 | } | |
d708f0d5 | 402 | blkg = new_blkg; |
cd1604fa | 403 | |
db613670 | 404 | /* link parent */ |
3c547865 | 405 | if (blkcg_parent(blkcg)) { |
99e60387 | 406 | blkg->parent = blkg_lookup(blkcg_parent(blkcg), disk->queue); |
3c547865 | 407 | if (WARN_ON_ONCE(!blkg->parent)) { |
20386ce0 | 408 | ret = -ENODEV; |
8c911f3d | 409 | goto err_put_css; |
3c547865 TH |
410 | } |
411 | blkg_get(blkg->parent); | |
412 | } | |
413 | ||
db613670 TH |
414 | /* invoke per-policy init */ |
415 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
416 | struct blkcg_policy *pol = blkcg_policy[i]; | |
417 | ||
418 | if (blkg->pd[i] && pol->pd_init_fn) | |
a9520cd6 | 419 | pol->pd_init_fn(blkg->pd[i]); |
db613670 TH |
420 | } |
421 | ||
422 | /* insert */ | |
cd1604fa | 423 | spin_lock(&blkcg->lock); |
99e60387 | 424 | ret = radix_tree_insert(&blkcg->blkg_tree, disk->queue->id, blkg); |
a637120e TH |
425 | if (likely(!ret)) { |
426 | hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list); | |
99e60387 | 427 | list_add(&blkg->q_node, &disk->queue->blkg_list); |
f427d909 TH |
428 | |
429 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
430 | struct blkcg_policy *pol = blkcg_policy[i]; | |
431 | ||
dfd6200a YK |
432 | if (blkg->pd[i]) { |
433 | if (pol->pd_online_fn) | |
434 | pol->pd_online_fn(blkg->pd[i]); | |
435 | blkg->pd[i]->online = true; | |
436 | } | |
f427d909 | 437 | } |
a637120e | 438 | } |
f427d909 | 439 | blkg->online = true; |
cd1604fa | 440 | spin_unlock(&blkcg->lock); |
496fb780 | 441 | |
ec13b1d6 | 442 | if (!ret) |
a637120e | 443 | return blkg; |
15974993 | 444 | |
3c547865 TH |
445 | /* @blkg failed fully initialized, use the usual release path */ |
446 | blkg_put(blkg); | |
447 | return ERR_PTR(ret); | |
448 | ||
d708f0d5 | 449 | err_put_css: |
496fb780 | 450 | css_put(&blkcg->css); |
93e6d5d8 | 451 | err_free_blkg: |
28e538a3 CH |
452 | if (new_blkg) |
453 | blkg_free(new_blkg); | |
93e6d5d8 | 454 | return ERR_PTR(ret); |
31e4c28d | 455 | } |
3c96cb32 | 456 | |
86cde6b6 | 457 | /** |
8c546287 | 458 | * blkg_lookup_create - lookup blkg, try to create one if not there |
86cde6b6 | 459 | * @blkcg: blkcg of interest |
99e60387 | 460 | * @disk: gendisk of interest |
86cde6b6 | 461 | * |
99e60387 | 462 | * Lookup blkg for the @blkcg - @disk pair. If it doesn't exist, try to |
3c547865 TH |
463 | * create one. blkg creation is performed recursively from blkcg_root such |
464 | * that all non-root blkg's have access to the parent blkg. This function | |
99e60387 | 465 | * should be called under RCU read lock and takes @disk->queue->queue_lock. |
86cde6b6 | 466 | * |
beea9da0 DZ |
467 | * Returns the blkg or the closest blkg if blkg_create() fails as it walks |
468 | * down from root. | |
86cde6b6 | 469 | */ |
8c546287 | 470 | static struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg, |
99e60387 | 471 | struct gendisk *disk) |
3c96cb32 | 472 | { |
99e60387 | 473 | struct request_queue *q = disk->queue; |
86cde6b6 | 474 | struct blkcg_gq *blkg; |
8c546287 | 475 | unsigned long flags; |
86cde6b6 TH |
476 | |
477 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
86cde6b6 | 478 | |
8c546287 | 479 | blkg = blkg_lookup(blkcg, q); |
86cde6b6 TH |
480 | if (blkg) |
481 | return blkg; | |
482 | ||
8c546287 | 483 | spin_lock_irqsave(&q->queue_lock, flags); |
4a69f325 CH |
484 | blkg = blkg_lookup(blkcg, q); |
485 | if (blkg) { | |
5765033c CH |
486 | if (blkcg != &blkcg_root && |
487 | blkg != rcu_dereference(blkcg->blkg_hint)) | |
488 | rcu_assign_pointer(blkcg->blkg_hint, blkg); | |
8c546287 | 489 | goto found; |
4a69f325 | 490 | } |
8c546287 | 491 | |
3c547865 TH |
492 | /* |
493 | * Create blkgs walking down from blkcg_root to @blkcg, so that all | |
beea9da0 DZ |
494 | * non-root blkgs have access to their parents. Returns the closest |
495 | * blkg to the intended blkg should blkg_create() fail. | |
3c547865 TH |
496 | */ |
497 | while (true) { | |
498 | struct blkcg *pos = blkcg; | |
499 | struct blkcg *parent = blkcg_parent(blkcg); | |
beea9da0 DZ |
500 | struct blkcg_gq *ret_blkg = q->root_blkg; |
501 | ||
502 | while (parent) { | |
79fcc5be | 503 | blkg = blkg_lookup(parent, q); |
beea9da0 DZ |
504 | if (blkg) { |
505 | /* remember closest blkg */ | |
506 | ret_blkg = blkg; | |
507 | break; | |
508 | } | |
3c547865 TH |
509 | pos = parent; |
510 | parent = blkcg_parent(parent); | |
511 | } | |
512 | ||
99e60387 | 513 | blkg = blkg_create(pos, disk, NULL); |
8c546287 CH |
514 | if (IS_ERR(blkg)) { |
515 | blkg = ret_blkg; | |
516 | break; | |
517 | } | |
beea9da0 | 518 | if (pos == blkcg) |
8c546287 | 519 | break; |
b978962a DZ |
520 | } |
521 | ||
8c546287 CH |
522 | found: |
523 | spin_unlock_irqrestore(&q->queue_lock, flags); | |
b978962a DZ |
524 | return blkg; |
525 | } | |
526 | ||
3c798398 | 527 | static void blkg_destroy(struct blkcg_gq *blkg) |
03aa264a | 528 | { |
3c798398 | 529 | struct blkcg *blkcg = blkg->blkcg; |
6b065462 | 530 | int i; |
03aa264a | 531 | |
0d945c1f | 532 | lockdep_assert_held(&blkg->q->queue_lock); |
9f13ef67 | 533 | lockdep_assert_held(&blkcg->lock); |
03aa264a | 534 | |
f1c006f1 YK |
535 | /* |
536 | * blkg stays on the queue list until blkg_free_workfn(), see details in | |
537 | * blkg_free_workfn(), hence this function can be called from | |
538 | * blkcg_destroy_blkgs() first and again from blkg_destroy_all() before | |
539 | * blkg_free_workfn(). | |
540 | */ | |
541 | if (hlist_unhashed(&blkg->blkcg_node)) | |
542 | return; | |
a637120e | 543 | |
6b065462 DZF |
544 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
545 | struct blkcg_policy *pol = blkcg_policy[i]; | |
546 | ||
dfd6200a | 547 | if (blkg->pd[i] && blkg->pd[i]->online) { |
f37bf75c | 548 | blkg->pd[i]->online = false; |
dfd6200a YK |
549 | if (pol->pd_offline_fn) |
550 | pol->pd_offline_fn(blkg->pd[i]); | |
dfd6200a | 551 | } |
6b065462 DZF |
552 | } |
553 | ||
f427d909 TH |
554 | blkg->online = false; |
555 | ||
a637120e | 556 | radix_tree_delete(&blkcg->blkg_tree, blkg->q->id); |
9f13ef67 | 557 | hlist_del_init_rcu(&blkg->blkcg_node); |
03aa264a | 558 | |
a637120e TH |
559 | /* |
560 | * Both setting lookup hint to and clearing it from @blkg are done | |
561 | * under queue_lock. If it's not pointing to @blkg now, it never | |
562 | * will. Hint assignment itself can race safely. | |
563 | */ | |
ec6c676a | 564 | if (rcu_access_pointer(blkcg->blkg_hint) == blkg) |
a637120e TH |
565 | rcu_assign_pointer(blkcg->blkg_hint, NULL); |
566 | ||
03aa264a TH |
567 | /* |
568 | * Put the reference taken at the time of creation so that when all | |
569 | * queues are gone, group can be destroyed. | |
570 | */ | |
7fcf2b03 | 571 | percpu_ref_kill(&blkg->refcnt); |
03aa264a TH |
572 | } |
573 | ||
00ad6991 | 574 | static void blkg_destroy_all(struct gendisk *disk) |
72e06c25 | 575 | { |
00ad6991 | 576 | struct request_queue *q = disk->queue; |
fab4c16c | 577 | struct blkcg_gq *blkg; |
a731763f | 578 | int count = BLKG_DESTROY_BATCH_SIZE; |
e63a5730 | 579 | int i; |
72e06c25 | 580 | |
a731763f | 581 | restart: |
0d945c1f | 582 | spin_lock_irq(&q->queue_lock); |
fab4c16c | 583 | list_for_each_entry(blkg, &q->blkg_list, q_node) { |
3c798398 | 584 | struct blkcg *blkcg = blkg->blkcg; |
72e06c25 | 585 | |
8176080d TS |
586 | if (hlist_unhashed(&blkg->blkcg_node)) |
587 | continue; | |
588 | ||
9f13ef67 TH |
589 | spin_lock(&blkcg->lock); |
590 | blkg_destroy(blkg); | |
591 | spin_unlock(&blkcg->lock); | |
a731763f YK |
592 | |
593 | /* | |
594 | * in order to avoid holding the spin lock for too long, release | |
595 | * it when a batch of blkgs are destroyed. | |
596 | */ | |
597 | if (!(--count)) { | |
598 | count = BLKG_DESTROY_BATCH_SIZE; | |
599 | spin_unlock_irq(&q->queue_lock); | |
600 | cond_resched(); | |
601 | goto restart; | |
602 | } | |
72e06c25 | 603 | } |
6fe810bd | 604 | |
e63a5730 ML |
605 | /* |
606 | * Mark policy deactivated since policy offline has been done, and | |
607 | * the free is scheduled, so future blkcg_deactivate_policy() can | |
608 | * be bypassed | |
609 | */ | |
610 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
611 | struct blkcg_policy *pol = blkcg_policy[i]; | |
612 | ||
613 | if (pol) | |
614 | __clear_bit(pol->plid, q->blkcg_pols); | |
615 | } | |
616 | ||
6fe810bd | 617 | q->root_blkg = NULL; |
0d945c1f | 618 | spin_unlock_irq(&q->queue_lock); |
72e06c25 TH |
619 | } |
620 | ||
6da66806 ML |
621 | static void blkg_iostat_set(struct blkg_iostat *dst, struct blkg_iostat *src) |
622 | { | |
623 | int i; | |
624 | ||
625 | for (i = 0; i < BLKG_IOSTAT_NR; i++) { | |
626 | dst->bytes[i] = src->bytes[i]; | |
627 | dst->ios[i] = src->ios[i]; | |
628 | } | |
629 | } | |
630 | ||
631 | static void __blkg_clear_stat(struct blkg_iostat_set *bis) | |
632 | { | |
633 | struct blkg_iostat cur = {0}; | |
634 | unsigned long flags; | |
635 | ||
636 | flags = u64_stats_update_begin_irqsave(&bis->sync); | |
637 | blkg_iostat_set(&bis->cur, &cur); | |
638 | blkg_iostat_set(&bis->last, &cur); | |
639 | u64_stats_update_end_irqrestore(&bis->sync, flags); | |
640 | } | |
641 | ||
642 | static void blkg_clear_stat(struct blkcg_gq *blkg) | |
643 | { | |
644 | int cpu; | |
645 | ||
646 | for_each_possible_cpu(cpu) { | |
647 | struct blkg_iostat_set *s = per_cpu_ptr(blkg->iostat_cpu, cpu); | |
648 | ||
649 | __blkg_clear_stat(s); | |
650 | } | |
651 | __blkg_clear_stat(&blkg->iostat); | |
652 | } | |
653 | ||
182446d0 TH |
654 | static int blkcg_reset_stats(struct cgroup_subsys_state *css, |
655 | struct cftype *cftype, u64 val) | |
303a3acb | 656 | { |
182446d0 | 657 | struct blkcg *blkcg = css_to_blkcg(css); |
3c798398 | 658 | struct blkcg_gq *blkg; |
6da66806 | 659 | int i; |
303a3acb | 660 | |
838f13bf | 661 | mutex_lock(&blkcg_pol_mutex); |
303a3acb | 662 | spin_lock_irq(&blkcg->lock); |
997a026c TH |
663 | |
664 | /* | |
665 | * Note that stat reset is racy - it doesn't synchronize against | |
666 | * stat updates. This is a debug feature which shouldn't exist | |
667 | * anyway. If you get hit by a race, retry. | |
668 | */ | |
b67bfe0d | 669 | hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) { |
6da66806 | 670 | blkg_clear_stat(blkg); |
8bd435b3 | 671 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
3c798398 | 672 | struct blkcg_policy *pol = blkcg_policy[i]; |
549d3aa8 | 673 | |
a9520cd6 TH |
674 | if (blkg->pd[i] && pol->pd_reset_stats_fn) |
675 | pol->pd_reset_stats_fn(blkg->pd[i]); | |
bc0d6501 | 676 | } |
303a3acb | 677 | } |
f0bdc8cd | 678 | |
303a3acb | 679 | spin_unlock_irq(&blkcg->lock); |
bc0d6501 | 680 | mutex_unlock(&blkcg_pol_mutex); |
303a3acb DS |
681 | return 0; |
682 | } | |
683 | ||
dd165eb3 | 684 | const char *blkg_dev_name(struct blkcg_gq *blkg) |
303a3acb | 685 | { |
a06377c5 | 686 | if (!blkg->q->disk) |
edb0872f | 687 | return NULL; |
d152c682 | 688 | return bdi_dev_name(blkg->q->disk->bdi); |
303a3acb DS |
689 | } |
690 | ||
d3d32e69 TH |
691 | /** |
692 | * blkcg_print_blkgs - helper for printing per-blkg data | |
693 | * @sf: seq_file to print to | |
694 | * @blkcg: blkcg of interest | |
695 | * @prfill: fill function to print out a blkg | |
696 | * @pol: policy in question | |
697 | * @data: data to be passed to @prfill | |
698 | * @show_total: to print out sum of prfill return values or not | |
699 | * | |
700 | * This function invokes @prfill on each blkg of @blkcg if pd for the | |
701 | * policy specified by @pol exists. @prfill is invoked with @sf, the | |
810ecfa7 TH |
702 | * policy data and @data and the matching queue lock held. If @show_total |
703 | * is %true, the sum of the return values from @prfill is printed with | |
704 | * "Total" label at the end. | |
d3d32e69 TH |
705 | * |
706 | * This is to be used to construct print functions for | |
707 | * cftype->read_seq_string method. | |
708 | */ | |
3c798398 | 709 | void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg, |
f95a04af TH |
710 | u64 (*prfill)(struct seq_file *, |
711 | struct blkg_policy_data *, int), | |
3c798398 | 712 | const struct blkcg_policy *pol, int data, |
ec399347 | 713 | bool show_total) |
5624a4e4 | 714 | { |
3c798398 | 715 | struct blkcg_gq *blkg; |
d3d32e69 | 716 | u64 total = 0; |
5624a4e4 | 717 | |
810ecfa7 | 718 | rcu_read_lock(); |
ee89f812 | 719 | hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) { |
0d945c1f | 720 | spin_lock_irq(&blkg->q->queue_lock); |
a2b1693b | 721 | if (blkcg_policy_enabled(blkg->q, pol)) |
f95a04af | 722 | total += prfill(sf, blkg->pd[pol->plid], data); |
0d945c1f | 723 | spin_unlock_irq(&blkg->q->queue_lock); |
810ecfa7 TH |
724 | } |
725 | rcu_read_unlock(); | |
d3d32e69 TH |
726 | |
727 | if (show_total) | |
728 | seq_printf(sf, "Total %llu\n", (unsigned long long)total); | |
729 | } | |
829fdb50 | 730 | EXPORT_SYMBOL_GPL(blkcg_print_blkgs); |
d3d32e69 TH |
731 | |
732 | /** | |
733 | * __blkg_prfill_u64 - prfill helper for a single u64 value | |
734 | * @sf: seq_file to print to | |
f95a04af | 735 | * @pd: policy private data of interest |
d3d32e69 TH |
736 | * @v: value to print |
737 | * | |
37754595 | 738 | * Print @v to @sf for the device associated with @pd. |
d3d32e69 | 739 | */ |
f95a04af | 740 | u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v) |
d3d32e69 | 741 | { |
f95a04af | 742 | const char *dname = blkg_dev_name(pd->blkg); |
d3d32e69 TH |
743 | |
744 | if (!dname) | |
745 | return 0; | |
746 | ||
747 | seq_printf(sf, "%s %llu\n", dname, (unsigned long long)v); | |
748 | return v; | |
749 | } | |
829fdb50 | 750 | EXPORT_SYMBOL_GPL(__blkg_prfill_u64); |
d3d32e69 | 751 | |
015d254c | 752 | /** |
faffaab2 TH |
753 | * blkg_conf_init - initialize a blkg_conf_ctx |
754 | * @ctx: blkg_conf_ctx to initialize | |
755 | * @input: input string | |
015d254c | 756 | * |
faffaab2 TH |
757 | * Initialize @ctx which can be used to parse blkg config input string @input. |
758 | * Once initialized, @ctx can be used with blkg_conf_open_bdev() and | |
759 | * blkg_conf_prep(), and must be cleaned up with blkg_conf_exit(). | |
760 | */ | |
761 | void blkg_conf_init(struct blkg_conf_ctx *ctx, char *input) | |
762 | { | |
763 | *ctx = (struct blkg_conf_ctx){ .input = input }; | |
764 | } | |
765 | EXPORT_SYMBOL_GPL(blkg_conf_init); | |
766 | ||
767 | /** | |
768 | * blkg_conf_open_bdev - parse and open bdev for per-blkg config update | |
769 | * @ctx: blkg_conf_ctx initialized with blkg_conf_init() | |
015d254c | 770 | * |
faffaab2 TH |
771 | * Parse the device node prefix part, MAJ:MIN, of per-blkg config update from |
772 | * @ctx->input and get and store the matching bdev in @ctx->bdev. @ctx->body is | |
773 | * set to point past the device node prefix. | |
774 | * | |
775 | * This function may be called multiple times on @ctx and the extra calls become | |
776 | * NOOPs. blkg_conf_prep() implicitly calls this function. Use this function | |
777 | * explicitly if bdev access is needed without resolving the blkcg / policy part | |
778 | * of @ctx->input. Returns -errno on error. | |
015d254c | 779 | */ |
faffaab2 | 780 | int blkg_conf_open_bdev(struct blkg_conf_ctx *ctx) |
015d254c | 781 | { |
faffaab2 | 782 | char *input = ctx->input; |
015d254c | 783 | unsigned int major, minor; |
22ae8ce8 CH |
784 | struct block_device *bdev; |
785 | int key_len; | |
015d254c | 786 | |
faffaab2 TH |
787 | if (ctx->bdev) |
788 | return 0; | |
789 | ||
015d254c | 790 | if (sscanf(input, "%u:%u%n", &major, &minor, &key_len) != 2) |
faffaab2 | 791 | return -EINVAL; |
015d254c TH |
792 | |
793 | input += key_len; | |
794 | if (!isspace(*input)) | |
faffaab2 | 795 | return -EINVAL; |
015d254c TH |
796 | input = skip_spaces(input); |
797 | ||
22ae8ce8 CH |
798 | bdev = blkdev_get_no_open(MKDEV(major, minor)); |
799 | if (!bdev) | |
faffaab2 | 800 | return -ENODEV; |
22ae8ce8 CH |
801 | if (bdev_is_partition(bdev)) { |
802 | blkdev_put_no_open(bdev); | |
faffaab2 | 803 | return -ENODEV; |
015d254c TH |
804 | } |
805 | ||
a13bd91b YK |
806 | mutex_lock(&bdev->bd_queue->rq_qos_mutex); |
807 | if (!disk_live(bdev->bd_disk)) { | |
808 | blkdev_put_no_open(bdev); | |
809 | mutex_unlock(&bdev->bd_queue->rq_qos_mutex); | |
810 | return -ENODEV; | |
811 | } | |
812 | ||
faffaab2 TH |
813 | ctx->body = input; |
814 | ctx->bdev = bdev; | |
815 | return 0; | |
015d254c TH |
816 | } |
817 | ||
3a8b31d3 TH |
818 | /** |
819 | * blkg_conf_prep - parse and prepare for per-blkg config update | |
820 | * @blkcg: target block cgroup | |
da8b0662 | 821 | * @pol: target policy |
faffaab2 TH |
822 | * @ctx: blkg_conf_ctx initialized with blkg_conf_init() |
823 | * | |
824 | * Parse per-blkg config update from @ctx->input and initialize @ctx | |
825 | * accordingly. On success, @ctx->body points to the part of @ctx->input | |
826 | * following MAJ:MIN, @ctx->bdev points to the target block device and | |
827 | * @ctx->blkg to the blkg being configured. | |
3a8b31d3 | 828 | * |
faffaab2 TH |
829 | * blkg_conf_open_bdev() may be called on @ctx beforehand. On success, this |
830 | * function returns with queue lock held and must be followed by | |
831 | * blkg_conf_exit(). | |
3a8b31d3 | 832 | */ |
3c798398 | 833 | int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol, |
faffaab2 | 834 | struct blkg_conf_ctx *ctx) |
83462a6c | 835 | __acquires(&bdev->bd_queue->queue_lock) |
34d0f179 | 836 | { |
99e60387 | 837 | struct gendisk *disk; |
457e490f | 838 | struct request_queue *q; |
3c798398 | 839 | struct blkcg_gq *blkg; |
015d254c | 840 | int ret; |
36aa9e5f | 841 | |
faffaab2 TH |
842 | ret = blkg_conf_open_bdev(ctx); |
843 | if (ret) | |
844 | return ret; | |
845 | ||
846 | disk = ctx->bdev->bd_disk; | |
99e60387 | 847 | q = disk->queue; |
da8b0662 | 848 | |
0c9d338c YK |
849 | /* |
850 | * blkcg_deactivate_policy() requires queue to be frozen, we can grab | |
851 | * q_usage_counter to prevent concurrent with blkcg_deactivate_policy(). | |
852 | */ | |
853 | ret = blk_queue_enter(q, 0); | |
854 | if (ret) | |
15c30104 | 855 | goto fail; |
0c9d338c | 856 | |
0d945c1f | 857 | spin_lock_irq(&q->queue_lock); |
e56da7e2 | 858 | |
f753526e CH |
859 | if (!blkcg_policy_enabled(q, pol)) { |
860 | ret = -EOPNOTSUPP; | |
457e490f TE |
861 | goto fail_unlock; |
862 | } | |
863 | ||
f753526e | 864 | blkg = blkg_lookup(blkcg, q); |
5765033c | 865 | if (blkg) |
457e490f TE |
866 | goto success; |
867 | ||
868 | /* | |
869 | * Create blkgs walking down from blkcg_root to @blkcg, so that all | |
870 | * non-root blkgs have access to their parents. | |
871 | */ | |
872 | while (true) { | |
873 | struct blkcg *pos = blkcg; | |
874 | struct blkcg *parent; | |
875 | struct blkcg_gq *new_blkg; | |
876 | ||
877 | parent = blkcg_parent(blkcg); | |
79fcc5be | 878 | while (parent && !blkg_lookup(parent, q)) { |
457e490f TE |
879 | pos = parent; |
880 | parent = blkcg_parent(parent); | |
881 | } | |
882 | ||
883 | /* Drop locks to do new blkg allocation with GFP_KERNEL. */ | |
0d945c1f | 884 | spin_unlock_irq(&q->queue_lock); |
457e490f | 885 | |
99e60387 | 886 | new_blkg = blkg_alloc(pos, disk, GFP_KERNEL); |
457e490f TE |
887 | if (unlikely(!new_blkg)) { |
888 | ret = -ENOMEM; | |
15c30104 | 889 | goto fail_exit_queue; |
7702e8f4 | 890 | } |
3a8b31d3 | 891 | |
f255c19b GKB |
892 | if (radix_tree_preload(GFP_KERNEL)) { |
893 | blkg_free(new_blkg); | |
894 | ret = -ENOMEM; | |
15c30104 | 895 | goto fail_exit_queue; |
f255c19b GKB |
896 | } |
897 | ||
0d945c1f | 898 | spin_lock_irq(&q->queue_lock); |
457e490f | 899 | |
f753526e | 900 | if (!blkcg_policy_enabled(q, pol)) { |
52abfcbd | 901 | blkg_free(new_blkg); |
f753526e | 902 | ret = -EOPNOTSUPP; |
f255c19b | 903 | goto fail_preloaded; |
457e490f TE |
904 | } |
905 | ||
f753526e | 906 | blkg = blkg_lookup(pos, q); |
457e490f TE |
907 | if (blkg) { |
908 | blkg_free(new_blkg); | |
909 | } else { | |
99e60387 | 910 | blkg = blkg_create(pos, disk, new_blkg); |
98d669b4 | 911 | if (IS_ERR(blkg)) { |
457e490f | 912 | ret = PTR_ERR(blkg); |
f255c19b | 913 | goto fail_preloaded; |
457e490f TE |
914 | } |
915 | } | |
916 | ||
f255c19b GKB |
917 | radix_tree_preload_end(); |
918 | ||
457e490f TE |
919 | if (pos == blkcg) |
920 | goto success; | |
921 | } | |
922 | success: | |
0c9d338c | 923 | blk_queue_exit(q); |
3a8b31d3 | 924 | ctx->blkg = blkg; |
726fa694 | 925 | return 0; |
457e490f | 926 | |
f255c19b GKB |
927 | fail_preloaded: |
928 | radix_tree_preload_end(); | |
457e490f | 929 | fail_unlock: |
0d945c1f | 930 | spin_unlock_irq(&q->queue_lock); |
15c30104 YK |
931 | fail_exit_queue: |
932 | blk_queue_exit(q); | |
457e490f | 933 | fail: |
457e490f TE |
934 | /* |
935 | * If queue was bypassing, we should retry. Do so after a | |
936 | * short msleep(). It isn't strictly necessary but queue | |
937 | * can be bypassing for some time and it's always nice to | |
938 | * avoid busy looping. | |
939 | */ | |
940 | if (ret == -EBUSY) { | |
941 | msleep(10); | |
942 | ret = restart_syscall(); | |
943 | } | |
944 | return ret; | |
34d0f179 | 945 | } |
89f3b6d6 | 946 | EXPORT_SYMBOL_GPL(blkg_conf_prep); |
34d0f179 | 947 | |
3a8b31d3 | 948 | /** |
faffaab2 TH |
949 | * blkg_conf_exit - clean up per-blkg config update |
950 | * @ctx: blkg_conf_ctx initialized with blkg_conf_init() | |
3a8b31d3 | 951 | * |
faffaab2 TH |
952 | * Clean up after per-blkg config update. This function must be called on all |
953 | * blkg_conf_ctx's initialized with blkg_conf_init(). | |
3a8b31d3 | 954 | */ |
faffaab2 | 955 | void blkg_conf_exit(struct blkg_conf_ctx *ctx) |
83462a6c | 956 | __releases(&ctx->bdev->bd_queue->queue_lock) |
a13bd91b | 957 | __releases(&ctx->bdev->bd_queue->rq_qos_mutex) |
34d0f179 | 958 | { |
faffaab2 TH |
959 | if (ctx->blkg) { |
960 | spin_unlock_irq(&bdev_get_queue(ctx->bdev)->queue_lock); | |
961 | ctx->blkg = NULL; | |
962 | } | |
963 | ||
964 | if (ctx->bdev) { | |
a13bd91b | 965 | mutex_unlock(&ctx->bdev->bd_queue->rq_qos_mutex); |
faffaab2 TH |
966 | blkdev_put_no_open(ctx->bdev); |
967 | ctx->body = NULL; | |
968 | ctx->bdev = NULL; | |
969 | } | |
34d0f179 | 970 | } |
faffaab2 | 971 | EXPORT_SYMBOL_GPL(blkg_conf_exit); |
34d0f179 | 972 | |
cd1fc4b9 BB |
973 | static void blkg_iostat_add(struct blkg_iostat *dst, struct blkg_iostat *src) |
974 | { | |
975 | int i; | |
976 | ||
977 | for (i = 0; i < BLKG_IOSTAT_NR; i++) { | |
978 | dst->bytes[i] += src->bytes[i]; | |
979 | dst->ios[i] += src->ios[i]; | |
980 | } | |
981 | } | |
982 | ||
983 | static void blkg_iostat_sub(struct blkg_iostat *dst, struct blkg_iostat *src) | |
984 | { | |
985 | int i; | |
986 | ||
987 | for (i = 0; i < BLKG_IOSTAT_NR; i++) { | |
988 | dst->bytes[i] -= src->bytes[i]; | |
989 | dst->ios[i] -= src->ios[i]; | |
990 | } | |
991 | } | |
992 | ||
362b8c16 JY |
993 | static void blkcg_iostat_update(struct blkcg_gq *blkg, struct blkg_iostat *cur, |
994 | struct blkg_iostat *last) | |
995 | { | |
996 | struct blkg_iostat delta; | |
997 | unsigned long flags; | |
998 | ||
999 | /* propagate percpu delta to global */ | |
1000 | flags = u64_stats_update_begin_irqsave(&blkg->iostat.sync); | |
1001 | blkg_iostat_set(&delta, cur); | |
1002 | blkg_iostat_sub(&delta, last); | |
1003 | blkg_iostat_add(&blkg->iostat.cur, &delta); | |
1004 | blkg_iostat_add(last, &delta); | |
1005 | u64_stats_update_end_irqrestore(&blkg->iostat.sync, flags); | |
1006 | } | |
1007 | ||
20cb1c2f | 1008 | static void __blkcg_rstat_flush(struct blkcg *blkcg, int cpu) |
cd1fc4b9 | 1009 | { |
3b8cc629 WL |
1010 | struct llist_head *lhead = per_cpu_ptr(blkcg->lhead, cpu); |
1011 | struct llist_node *lnode; | |
1012 | struct blkg_iostat_set *bisc, *next_bisc; | |
9c39b7a9 | 1013 | unsigned long flags; |
cd1fc4b9 BB |
1014 | |
1015 | rcu_read_lock(); | |
1016 | ||
3b8cc629 WL |
1017 | lnode = llist_del_all(lhead); |
1018 | if (!lnode) | |
1019 | goto out; | |
1020 | ||
20cb1c2f ML |
1021 | /* |
1022 | * For covering concurrent parent blkg update from blkg_release(). | |
1023 | * | |
1024 | * When flushing from cgroup, cgroup_rstat_lock is always held, so | |
1025 | * this lock won't cause contention most of time. | |
1026 | */ | |
9c39b7a9 | 1027 | raw_spin_lock_irqsave(&blkg_stat_lock, flags); |
20cb1c2f | 1028 | |
3b8cc629 WL |
1029 | /* |
1030 | * Iterate only the iostat_cpu's queued in the lockless list. | |
1031 | */ | |
1032 | llist_for_each_entry_safe(bisc, next_bisc, lnode, lnode) { | |
1033 | struct blkcg_gq *blkg = bisc->blkg; | |
cd1fc4b9 | 1034 | struct blkcg_gq *parent = blkg->parent; |
362b8c16 | 1035 | struct blkg_iostat cur; |
cd1fc4b9 BB |
1036 | unsigned int seq; |
1037 | ||
d0aac236 ML |
1038 | /* |
1039 | * Order assignment of `next_bisc` from `bisc->lnode.next` in | |
1040 | * llist_for_each_entry_safe and clearing `bisc->lqueued` for | |
1041 | * avoiding to assign `next_bisc` with new next pointer added | |
1042 | * in blk_cgroup_bio_start() in case of re-ordering. | |
1043 | * | |
1044 | * The pair barrier is implied in llist_add() in blk_cgroup_bio_start(). | |
1045 | */ | |
1046 | smp_mb(); | |
1047 | ||
3b8cc629 WL |
1048 | WRITE_ONCE(bisc->lqueued, false); |
1049 | ||
cd1fc4b9 BB |
1050 | /* fetch the current per-cpu values */ |
1051 | do { | |
1052 | seq = u64_stats_fetch_begin(&bisc->sync); | |
1053 | blkg_iostat_set(&cur, &bisc->cur); | |
1054 | } while (u64_stats_fetch_retry(&bisc->sync, seq)); | |
1055 | ||
362b8c16 | 1056 | blkcg_iostat_update(blkg, &cur, &bisc->last); |
cd1fc4b9 | 1057 | |
dc26532a | 1058 | /* propagate global delta to parent (unless that's root) */ |
362b8c16 JY |
1059 | if (parent && parent->parent) |
1060 | blkcg_iostat_update(parent, &blkg->iostat.cur, | |
1061 | &blkg->iostat.last); | |
cd1fc4b9 | 1062 | } |
9c39b7a9 | 1063 | raw_spin_unlock_irqrestore(&blkg_stat_lock, flags); |
3b8cc629 | 1064 | out: |
cd1fc4b9 BB |
1065 | rcu_read_unlock(); |
1066 | } | |
1067 | ||
20cb1c2f ML |
1068 | static void blkcg_rstat_flush(struct cgroup_subsys_state *css, int cpu) |
1069 | { | |
1070 | /* Root-level stats are sourced from system-wide IO stats */ | |
1071 | if (cgroup_parent(css->cgroup)) | |
1072 | __blkcg_rstat_flush(css_to_blkcg(css), cpu); | |
1073 | } | |
1074 | ||
ef45fe47 | 1075 | /* |
dc26532a JW |
1076 | * We source root cgroup stats from the system-wide stats to avoid |
1077 | * tracking the same information twice and incurring overhead when no | |
1078 | * cgroups are defined. For that reason, cgroup_rstat_flush in | |
1079 | * blkcg_print_stat does not actually fill out the iostat in the root | |
1080 | * cgroup's blkcg_gq. | |
ef45fe47 BB |
1081 | * |
1082 | * However, we would like to re-use the printing code between the root and | |
1083 | * non-root cgroups to the extent possible. For that reason, we simulate | |
1084 | * flushing the root cgroup's stats by explicitly filling in the iostat | |
1085 | * with disk level statistics. | |
1086 | */ | |
1087 | static void blkcg_fill_root_iostats(void) | |
1088 | { | |
1089 | struct class_dev_iter iter; | |
1090 | struct device *dev; | |
1091 | ||
1092 | class_dev_iter_init(&iter, &block_class, NULL, &disk_type); | |
1093 | while ((dev = class_dev_iter_next(&iter))) { | |
0d02129e | 1094 | struct block_device *bdev = dev_to_bdev(dev); |
928f6f00 | 1095 | struct blkcg_gq *blkg = bdev->bd_disk->queue->root_blkg; |
ef45fe47 BB |
1096 | struct blkg_iostat tmp; |
1097 | int cpu; | |
f122d103 | 1098 | unsigned long flags; |
ef45fe47 BB |
1099 | |
1100 | memset(&tmp, 0, sizeof(tmp)); | |
1101 | for_each_possible_cpu(cpu) { | |
1102 | struct disk_stats *cpu_dkstats; | |
1103 | ||
0d02129e | 1104 | cpu_dkstats = per_cpu_ptr(bdev->bd_stats, cpu); |
ef45fe47 BB |
1105 | tmp.ios[BLKG_IOSTAT_READ] += |
1106 | cpu_dkstats->ios[STAT_READ]; | |
1107 | tmp.ios[BLKG_IOSTAT_WRITE] += | |
1108 | cpu_dkstats->ios[STAT_WRITE]; | |
1109 | tmp.ios[BLKG_IOSTAT_DISCARD] += | |
1110 | cpu_dkstats->ios[STAT_DISCARD]; | |
1111 | // convert sectors to bytes | |
1112 | tmp.bytes[BLKG_IOSTAT_READ] += | |
1113 | cpu_dkstats->sectors[STAT_READ] << 9; | |
1114 | tmp.bytes[BLKG_IOSTAT_WRITE] += | |
1115 | cpu_dkstats->sectors[STAT_WRITE] << 9; | |
1116 | tmp.bytes[BLKG_IOSTAT_DISCARD] += | |
1117 | cpu_dkstats->sectors[STAT_DISCARD] << 9; | |
ef45fe47 | 1118 | } |
f122d103 CZ |
1119 | |
1120 | flags = u64_stats_update_begin_irqsave(&blkg->iostat.sync); | |
1121 | blkg_iostat_set(&blkg->iostat.cur, &tmp); | |
1122 | u64_stats_update_end_irqrestore(&blkg->iostat.sync, flags); | |
ef45fe47 BB |
1123 | } |
1124 | } | |
1125 | ||
49cb5168 | 1126 | static void blkcg_print_one_stat(struct blkcg_gq *blkg, struct seq_file *s) |
2ee867dc | 1127 | { |
49cb5168 CH |
1128 | struct blkg_iostat_set *bis = &blkg->iostat; |
1129 | u64 rbytes, wbytes, rios, wios, dbytes, dios; | |
49cb5168 CH |
1130 | const char *dname; |
1131 | unsigned seq; | |
49cb5168 | 1132 | int i; |
ef45fe47 | 1133 | |
49cb5168 CH |
1134 | if (!blkg->online) |
1135 | return; | |
2ee867dc | 1136 | |
49cb5168 CH |
1137 | dname = blkg_dev_name(blkg); |
1138 | if (!dname) | |
1139 | return; | |
2ee867dc | 1140 | |
252c651a | 1141 | seq_printf(s, "%s ", dname); |
b0814361 | 1142 | |
49cb5168 CH |
1143 | do { |
1144 | seq = u64_stats_fetch_begin(&bis->sync); | |
b0814361 | 1145 | |
49cb5168 CH |
1146 | rbytes = bis->cur.bytes[BLKG_IOSTAT_READ]; |
1147 | wbytes = bis->cur.bytes[BLKG_IOSTAT_WRITE]; | |
1148 | dbytes = bis->cur.bytes[BLKG_IOSTAT_DISCARD]; | |
1149 | rios = bis->cur.ios[BLKG_IOSTAT_READ]; | |
1150 | wios = bis->cur.ios[BLKG_IOSTAT_WRITE]; | |
1151 | dios = bis->cur.ios[BLKG_IOSTAT_DISCARD]; | |
1152 | } while (u64_stats_fetch_retry(&bis->sync, seq)); | |
2ee867dc | 1153 | |
49cb5168 | 1154 | if (rbytes || wbytes || rios || wios) { |
252c651a | 1155 | seq_printf(s, "rbytes=%llu wbytes=%llu rios=%llu wios=%llu dbytes=%llu dios=%llu", |
49cb5168 CH |
1156 | rbytes, wbytes, rios, wios, |
1157 | dbytes, dios); | |
1158 | } | |
903d23f0 | 1159 | |
49cb5168 | 1160 | if (blkcg_debug_stats && atomic_read(&blkg->use_delay)) { |
252c651a | 1161 | seq_printf(s, " use_delay=%d delay_nsec=%llu", |
49cb5168 CH |
1162 | atomic_read(&blkg->use_delay), |
1163 | atomic64_read(&blkg->delay_nsec)); | |
1164 | } | |
2ee867dc | 1165 | |
49cb5168 CH |
1166 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
1167 | struct blkcg_policy *pol = blkcg_policy[i]; | |
2ee867dc | 1168 | |
49cb5168 CH |
1169 | if (!blkg->pd[i] || !pol->pd_stat_fn) |
1170 | continue; | |
903d23f0 | 1171 | |
3607849d | 1172 | pol->pd_stat_fn(blkg->pd[i], s); |
49cb5168 | 1173 | } |
d09d8df3 | 1174 | |
3607849d | 1175 | seq_puts(s, "\n"); |
49cb5168 | 1176 | } |
903d23f0 | 1177 | |
49cb5168 CH |
1178 | static int blkcg_print_stat(struct seq_file *sf, void *v) |
1179 | { | |
1180 | struct blkcg *blkcg = css_to_blkcg(seq_css(sf)); | |
1181 | struct blkcg_gq *blkg; | |
903d23f0 | 1182 | |
49cb5168 CH |
1183 | if (!seq_css(sf)->parent) |
1184 | blkcg_fill_root_iostats(); | |
1185 | else | |
1186 | cgroup_rstat_flush(blkcg->css.cgroup); | |
07b0fdec | 1187 | |
49cb5168 CH |
1188 | rcu_read_lock(); |
1189 | hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) { | |
1190 | spin_lock_irq(&blkg->q->queue_lock); | |
1191 | blkcg_print_one_stat(blkg, sf); | |
b0814361 | 1192 | spin_unlock_irq(&blkg->q->queue_lock); |
2ee867dc | 1193 | } |
2ee867dc TH |
1194 | rcu_read_unlock(); |
1195 | return 0; | |
1196 | } | |
1197 | ||
e1f3b941 | 1198 | static struct cftype blkcg_files[] = { |
2ee867dc TH |
1199 | { |
1200 | .name = "stat", | |
1201 | .seq_show = blkcg_print_stat, | |
1202 | }, | |
1203 | { } /* terminate */ | |
1204 | }; | |
1205 | ||
e1f3b941 | 1206 | static struct cftype blkcg_legacy_files[] = { |
84c124da DS |
1207 | { |
1208 | .name = "reset_stats", | |
3c798398 | 1209 | .write_u64 = blkcg_reset_stats, |
22084190 | 1210 | }, |
4baf6e33 | 1211 | { } /* terminate */ |
31e4c28d VG |
1212 | }; |
1213 | ||
dec223c9 CH |
1214 | #ifdef CONFIG_CGROUP_WRITEBACK |
1215 | struct list_head *blkcg_get_cgwb_list(struct cgroup_subsys_state *css) | |
1216 | { | |
1217 | return &css_to_blkcg(css)->cgwb_list; | |
1218 | } | |
1219 | #endif | |
1220 | ||
59b57717 DZF |
1221 | /* |
1222 | * blkcg destruction is a three-stage process. | |
1223 | * | |
1224 | * 1. Destruction starts. The blkcg_css_offline() callback is invoked | |
1225 | * which offlines writeback. Here we tie the next stage of blkg destruction | |
1226 | * to the completion of writeback associated with the blkcg. This lets us | |
1227 | * avoid punting potentially large amounts of outstanding writeback to root | |
1228 | * while maintaining any ongoing policies. The next stage is triggered when | |
1229 | * the nr_cgwbs count goes to zero. | |
1230 | * | |
1231 | * 2. When the nr_cgwbs count goes to zero, blkcg_destroy_blkgs() is called | |
1232 | * and handles the destruction of blkgs. Here the css reference held by | |
1233 | * the blkg is put back eventually allowing blkcg_css_free() to be called. | |
1234 | * This work may occur in cgwb_release_workfn() on the cgwb_release | |
1235 | * workqueue. Any submitted ios that fail to get the blkg ref will be | |
1236 | * punted to the root_blkg. | |
1237 | * | |
1238 | * 3. Once the blkcg ref count goes to zero, blkcg_css_free() is called. | |
1239 | * This finally frees the blkcg. | |
1240 | */ | |
1241 | ||
59b57717 DZF |
1242 | /** |
1243 | * blkcg_destroy_blkgs - responsible for shooting down blkgs | |
1244 | * @blkcg: blkcg of interest | |
1245 | * | |
1246 | * blkgs should be removed while holding both q and blkcg locks. As blkcg lock | |
1247 | * is nested inside q lock, this function performs reverse double lock dancing. | |
1248 | * Destroying the blkgs releases the reference held on the blkcg's css allowing | |
1249 | * blkcg_css_free to eventually be called. | |
1250 | * | |
1251 | * This is the blkcg counterpart of ioc_release_fn(). | |
1252 | */ | |
397c9f46 | 1253 | static void blkcg_destroy_blkgs(struct blkcg *blkcg) |
59b57717 | 1254 | { |
6c635cae BW |
1255 | might_sleep(); |
1256 | ||
9f13ef67 | 1257 | spin_lock_irq(&blkcg->lock); |
7ee9c562 | 1258 | |
4c699480 JQ |
1259 | while (!hlist_empty(&blkcg->blkg_list)) { |
1260 | struct blkcg_gq *blkg = hlist_entry(blkcg->blkg_list.first, | |
6b065462 | 1261 | struct blkcg_gq, blkcg_node); |
4c699480 JQ |
1262 | struct request_queue *q = blkg->q; |
1263 | ||
6c635cae BW |
1264 | if (need_resched() || !spin_trylock(&q->queue_lock)) { |
1265 | /* | |
1266 | * Given that the system can accumulate a huge number | |
1267 | * of blkgs in pathological cases, check to see if we | |
1268 | * need to rescheduling to avoid softlockup. | |
1269 | */ | |
4c699480 | 1270 | spin_unlock_irq(&blkcg->lock); |
6c635cae | 1271 | cond_resched(); |
4c699480 | 1272 | spin_lock_irq(&blkcg->lock); |
6c635cae | 1273 | continue; |
4c699480 | 1274 | } |
6c635cae BW |
1275 | |
1276 | blkg_destroy(blkg); | |
1277 | spin_unlock(&q->queue_lock); | |
4c699480 | 1278 | } |
6b065462 | 1279 | |
4c699480 JQ |
1280 | spin_unlock_irq(&blkcg->lock); |
1281 | } | |
1282 | ||
397c9f46 CH |
1283 | /** |
1284 | * blkcg_pin_online - pin online state | |
1285 | * @blkcg_css: blkcg of interest | |
1286 | * | |
1287 | * While pinned, a blkcg is kept online. This is primarily used to | |
1288 | * impedance-match blkg and cgwb lifetimes so that blkg doesn't go offline | |
1289 | * while an associated cgwb is still active. | |
1290 | */ | |
1291 | void blkcg_pin_online(struct cgroup_subsys_state *blkcg_css) | |
1292 | { | |
1293 | refcount_inc(&css_to_blkcg(blkcg_css)->online_pin); | |
1294 | } | |
1295 | ||
1296 | /** | |
1297 | * blkcg_unpin_online - unpin online state | |
1298 | * @blkcg_css: blkcg of interest | |
1299 | * | |
1300 | * This is primarily used to impedance-match blkg and cgwb lifetimes so | |
1301 | * that blkg doesn't go offline while an associated cgwb is still active. | |
1302 | * When this count goes to zero, all active cgwbs have finished so the | |
1303 | * blkcg can continue destruction by calling blkcg_destroy_blkgs(). | |
1304 | */ | |
1305 | void blkcg_unpin_online(struct cgroup_subsys_state *blkcg_css) | |
1306 | { | |
1307 | struct blkcg *blkcg = css_to_blkcg(blkcg_css); | |
1308 | ||
1309 | do { | |
1310 | if (!refcount_dec_and_test(&blkcg->online_pin)) | |
1311 | break; | |
1312 | blkcg_destroy_blkgs(blkcg); | |
1313 | blkcg = blkcg_parent(blkcg); | |
1314 | } while (blkcg); | |
1315 | } | |
1316 | ||
1317 | /** | |
1318 | * blkcg_css_offline - cgroup css_offline callback | |
1319 | * @css: css of interest | |
1320 | * | |
1321 | * This function is called when @css is about to go away. Here the cgwbs are | |
1322 | * offlined first and only once writeback associated with the blkcg has | |
1323 | * finished do we start step 2 (see above). | |
1324 | */ | |
1325 | static void blkcg_css_offline(struct cgroup_subsys_state *css) | |
1326 | { | |
1327 | /* this prevents anyone from attaching or migrating to this blkcg */ | |
dec223c9 | 1328 | wb_blkcg_offline(css); |
397c9f46 CH |
1329 | |
1330 | /* put the base online pin allowing step 2 to be triggered */ | |
1331 | blkcg_unpin_online(css); | |
1332 | } | |
1333 | ||
eb95419b | 1334 | static void blkcg_css_free(struct cgroup_subsys_state *css) |
7ee9c562 | 1335 | { |
eb95419b | 1336 | struct blkcg *blkcg = css_to_blkcg(css); |
bc915e61 | 1337 | int i; |
7ee9c562 | 1338 | |
7876f930 | 1339 | mutex_lock(&blkcg_pol_mutex); |
e4a9bde9 | 1340 | |
7876f930 | 1341 | list_del(&blkcg->all_blkcgs_node); |
7876f930 | 1342 | |
bc915e61 | 1343 | for (i = 0; i < BLKCG_MAX_POLS; i++) |
e4a9bde9 TH |
1344 | if (blkcg->cpd[i]) |
1345 | blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]); | |
1346 | ||
1347 | mutex_unlock(&blkcg_pol_mutex); | |
1348 | ||
3b8cc629 | 1349 | free_percpu(blkcg->lhead); |
bc915e61 | 1350 | kfree(blkcg); |
31e4c28d VG |
1351 | } |
1352 | ||
eb95419b TH |
1353 | static struct cgroup_subsys_state * |
1354 | blkcg_css_alloc(struct cgroup_subsys_state *parent_css) | |
31e4c28d | 1355 | { |
3c798398 | 1356 | struct blkcg *blkcg; |
e48453c3 | 1357 | int i; |
31e4c28d | 1358 | |
7876f930 TH |
1359 | mutex_lock(&blkcg_pol_mutex); |
1360 | ||
eb95419b | 1361 | if (!parent_css) { |
3c798398 | 1362 | blkcg = &blkcg_root; |
bc915e61 TH |
1363 | } else { |
1364 | blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL); | |
b5a9adcb | 1365 | if (!blkcg) |
4c18c9e9 | 1366 | goto unlock; |
e48453c3 AA |
1367 | } |
1368 | ||
3b8cc629 WL |
1369 | if (init_blkcg_llists(blkcg)) |
1370 | goto free_blkcg; | |
1371 | ||
e48453c3 AA |
1372 | for (i = 0; i < BLKCG_MAX_POLS ; i++) { |
1373 | struct blkcg_policy *pol = blkcg_policy[i]; | |
1374 | struct blkcg_policy_data *cpd; | |
1375 | ||
1376 | /* | |
1377 | * If the policy hasn't been attached yet, wait for it | |
1378 | * to be attached before doing anything else. Otherwise, | |
1379 | * check if the policy requires any specific per-cgroup | |
1380 | * data: if it does, allocate and initialize it. | |
1381 | */ | |
e4a9bde9 | 1382 | if (!pol || !pol->cpd_alloc_fn) |
e48453c3 AA |
1383 | continue; |
1384 | ||
e4a9bde9 | 1385 | cpd = pol->cpd_alloc_fn(GFP_KERNEL); |
b5a9adcb | 1386 | if (!cpd) |
e48453c3 | 1387 | goto free_pd_blkcg; |
b5a9adcb | 1388 | |
81437648 TH |
1389 | blkcg->cpd[i] = cpd; |
1390 | cpd->blkcg = blkcg; | |
e48453c3 | 1391 | cpd->plid = i; |
e48453c3 | 1392 | } |
31e4c28d | 1393 | |
31e4c28d | 1394 | spin_lock_init(&blkcg->lock); |
d866dbf6 | 1395 | refcount_set(&blkcg->online_pin, 1); |
e00f4f4d | 1396 | INIT_RADIX_TREE(&blkcg->blkg_tree, GFP_NOWAIT | __GFP_NOWARN); |
31e4c28d | 1397 | INIT_HLIST_HEAD(&blkcg->blkg_list); |
52ebea74 TH |
1398 | #ifdef CONFIG_CGROUP_WRITEBACK |
1399 | INIT_LIST_HEAD(&blkcg->cgwb_list); | |
1400 | #endif | |
7876f930 TH |
1401 | list_add_tail(&blkcg->all_blkcgs_node, &all_blkcgs); |
1402 | ||
1403 | mutex_unlock(&blkcg_pol_mutex); | |
31e4c28d | 1404 | return &blkcg->css; |
e48453c3 AA |
1405 | |
1406 | free_pd_blkcg: | |
1407 | for (i--; i >= 0; i--) | |
e4a9bde9 TH |
1408 | if (blkcg->cpd[i]) |
1409 | blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]); | |
3b8cc629 WL |
1410 | free_percpu(blkcg->lhead); |
1411 | free_blkcg: | |
4c18c9e9 | 1412 | if (blkcg != &blkcg_root) |
1413 | kfree(blkcg); | |
1414 | unlock: | |
7876f930 | 1415 | mutex_unlock(&blkcg_pol_mutex); |
b5a9adcb | 1416 | return ERR_PTR(-ENOMEM); |
31e4c28d VG |
1417 | } |
1418 | ||
4308a434 TH |
1419 | static int blkcg_css_online(struct cgroup_subsys_state *css) |
1420 | { | |
397c9f46 | 1421 | struct blkcg *parent = blkcg_parent(css_to_blkcg(css)); |
4308a434 TH |
1422 | |
1423 | /* | |
1424 | * blkcg_pin_online() is used to delay blkcg offline so that blkgs | |
1425 | * don't go offline while cgwbs are still active on them. Pin the | |
1426 | * parent so that offline always happens towards the root. | |
1427 | */ | |
1428 | if (parent) | |
d7dbd43f | 1429 | blkcg_pin_online(&parent->css); |
4308a434 TH |
1430 | return 0; |
1431 | } | |
1432 | ||
8b8ace08 ML |
1433 | void blkg_init_queue(struct request_queue *q) |
1434 | { | |
1435 | INIT_LIST_HEAD(&q->blkg_list); | |
1436 | mutex_init(&q->blkcg_mutex); | |
1437 | } | |
1438 | ||
9823538f | 1439 | int blkcg_init_disk(struct gendisk *disk) |
5efd6113 | 1440 | { |
9823538f | 1441 | struct request_queue *q = disk->queue; |
d708f0d5 JA |
1442 | struct blkcg_gq *new_blkg, *blkg; |
1443 | bool preloaded; | |
ec13b1d6 TH |
1444 | int ret; |
1445 | ||
99e60387 | 1446 | new_blkg = blkg_alloc(&blkcg_root, disk, GFP_KERNEL); |
d708f0d5 JA |
1447 | if (!new_blkg) |
1448 | return -ENOMEM; | |
1449 | ||
1450 | preloaded = !radix_tree_preload(GFP_KERNEL); | |
1451 | ||
bea54883 | 1452 | /* Make sure the root blkg exists. */ |
77c570a1 | 1453 | /* spin_lock_irq can serve as RCU read-side critical section. */ |
0d945c1f | 1454 | spin_lock_irq(&q->queue_lock); |
99e60387 | 1455 | blkg = blkg_create(&blkcg_root, disk, new_blkg); |
901932a3 JB |
1456 | if (IS_ERR(blkg)) |
1457 | goto err_unlock; | |
1458 | q->root_blkg = blkg; | |
0d945c1f | 1459 | spin_unlock_irq(&q->queue_lock); |
ec13b1d6 | 1460 | |
d708f0d5 JA |
1461 | if (preloaded) |
1462 | radix_tree_preload_end(); | |
1463 | ||
b0dde3f5 | 1464 | ret = blk_ioprio_init(disk); |
556910e3 BVA |
1465 | if (ret) |
1466 | goto err_destroy_all; | |
1467 | ||
04be60b5 | 1468 | return 0; |
901932a3 | 1469 | |
04be60b5 | 1470 | err_destroy_all: |
00ad6991 | 1471 | blkg_destroy_all(disk); |
04be60b5 | 1472 | return ret; |
901932a3 | 1473 | err_unlock: |
0d945c1f | 1474 | spin_unlock_irq(&q->queue_lock); |
901932a3 JB |
1475 | if (preloaded) |
1476 | radix_tree_preload_end(); | |
1477 | return PTR_ERR(blkg); | |
5efd6113 TH |
1478 | } |
1479 | ||
9823538f | 1480 | void blkcg_exit_disk(struct gendisk *disk) |
5efd6113 | 1481 | { |
00ad6991 | 1482 | blkg_destroy_all(disk); |
e13793ba | 1483 | blk_throtl_exit(disk); |
5efd6113 TH |
1484 | } |
1485 | ||
d09d8df3 JB |
1486 | static void blkcg_exit(struct task_struct *tsk) |
1487 | { | |
f05837ed CH |
1488 | if (tsk->throttle_disk) |
1489 | put_disk(tsk->throttle_disk); | |
1490 | tsk->throttle_disk = NULL; | |
d09d8df3 JB |
1491 | } |
1492 | ||
c165b3e3 | 1493 | struct cgroup_subsys io_cgrp_subsys = { |
92fb9748 | 1494 | .css_alloc = blkcg_css_alloc, |
4308a434 | 1495 | .css_online = blkcg_css_online, |
92fb9748 TH |
1496 | .css_offline = blkcg_css_offline, |
1497 | .css_free = blkcg_css_free, | |
f7331648 | 1498 | .css_rstat_flush = blkcg_rstat_flush, |
2ee867dc | 1499 | .dfl_cftypes = blkcg_files, |
880f50e2 | 1500 | .legacy_cftypes = blkcg_legacy_files, |
c165b3e3 | 1501 | .legacy_name = "blkio", |
d09d8df3 | 1502 | .exit = blkcg_exit, |
1ced953b TH |
1503 | #ifdef CONFIG_MEMCG |
1504 | /* | |
1505 | * This ensures that, if available, memcg is automatically enabled | |
1506 | * together on the default hierarchy so that the owner cgroup can | |
1507 | * be retrieved from writeback pages. | |
1508 | */ | |
1509 | .depends_on = 1 << memory_cgrp_id, | |
1510 | #endif | |
676f7c8f | 1511 | }; |
c165b3e3 | 1512 | EXPORT_SYMBOL_GPL(io_cgrp_subsys); |
676f7c8f | 1513 | |
a2b1693b | 1514 | /** |
40e4996e CH |
1515 | * blkcg_activate_policy - activate a blkcg policy on a gendisk |
1516 | * @disk: gendisk of interest | |
a2b1693b TH |
1517 | * @pol: blkcg policy to activate |
1518 | * | |
40e4996e | 1519 | * Activate @pol on @disk. Requires %GFP_KERNEL context. @disk goes through |
a2b1693b TH |
1520 | * bypass mode to populate its blkgs with policy_data for @pol. |
1521 | * | |
40e4996e | 1522 | * Activation happens with @disk bypassed, so nobody would be accessing blkgs |
a2b1693b TH |
1523 | * from IO path. Update of each blkg is protected by both queue and blkcg |
1524 | * locks so that holding either lock and testing blkcg_policy_enabled() is | |
1525 | * always enough for dereferencing policy data. | |
1526 | * | |
1527 | * The caller is responsible for synchronizing [de]activations and policy | |
1528 | * [un]registerations. Returns 0 on success, -errno on failure. | |
1529 | */ | |
40e4996e | 1530 | int blkcg_activate_policy(struct gendisk *disk, const struct blkcg_policy *pol) |
a2b1693b | 1531 | { |
40e4996e | 1532 | struct request_queue *q = disk->queue; |
4c55f4f9 | 1533 | struct blkg_policy_data *pd_prealloc = NULL; |
9d179b86 | 1534 | struct blkcg_gq *blkg, *pinned_blkg = NULL; |
4c55f4f9 | 1535 | int ret; |
a2b1693b TH |
1536 | |
1537 | if (blkcg_policy_enabled(q, pol)) | |
1538 | return 0; | |
1539 | ||
344e9ffc | 1540 | if (queue_is_mq(q)) |
bd166ef1 | 1541 | blk_mq_freeze_queue(q); |
9d179b86 | 1542 | retry: |
0d945c1f | 1543 | spin_lock_irq(&q->queue_lock); |
a2b1693b | 1544 | |
ec14a87e | 1545 | /* blkg_list is pushed at the head, reverse walk to initialize parents first */ |
71c81407 | 1546 | list_for_each_entry_reverse(blkg, &q->blkg_list, q_node) { |
4c55f4f9 TH |
1547 | struct blkg_policy_data *pd; |
1548 | ||
1549 | if (blkg->pd[pol->plid]) | |
1550 | continue; | |
a2b1693b | 1551 | |
9d179b86 TH |
1552 | /* If prealloc matches, use it; otherwise try GFP_NOWAIT */ |
1553 | if (blkg == pinned_blkg) { | |
1554 | pd = pd_prealloc; | |
1555 | pd_prealloc = NULL; | |
1556 | } else { | |
0a0b4f79 CH |
1557 | pd = pol->pd_alloc_fn(disk, blkg->blkcg, |
1558 | GFP_NOWAIT | __GFP_NOWARN); | |
9d179b86 TH |
1559 | } |
1560 | ||
4c55f4f9 | 1561 | if (!pd) { |
9d179b86 TH |
1562 | /* |
1563 | * GFP_NOWAIT failed. Free the existing one and | |
1564 | * prealloc for @blkg w/ GFP_KERNEL. | |
1565 | */ | |
1566 | if (pinned_blkg) | |
1567 | blkg_put(pinned_blkg); | |
1568 | blkg_get(blkg); | |
1569 | pinned_blkg = blkg; | |
1570 | ||
0d945c1f | 1571 | spin_unlock_irq(&q->queue_lock); |
9d179b86 TH |
1572 | |
1573 | if (pd_prealloc) | |
1574 | pol->pd_free_fn(pd_prealloc); | |
0a0b4f79 CH |
1575 | pd_prealloc = pol->pd_alloc_fn(disk, blkg->blkcg, |
1576 | GFP_KERNEL); | |
9d179b86 TH |
1577 | if (pd_prealloc) |
1578 | goto retry; | |
1579 | else | |
1580 | goto enomem; | |
4c55f4f9 | 1581 | } |
a2b1693b | 1582 | |
ec14a87e TH |
1583 | spin_lock(&blkg->blkcg->lock); |
1584 | ||
a2b1693b | 1585 | pd->blkg = blkg; |
b276a876 | 1586 | pd->plid = pol->plid; |
ec14a87e | 1587 | blkg->pd[pol->plid] = pd; |
a2b1693b | 1588 | |
ec14a87e TH |
1589 | if (pol->pd_init_fn) |
1590 | pol->pd_init_fn(pd); | |
9d179b86 | 1591 | |
dfd6200a | 1592 | if (pol->pd_online_fn) |
ec14a87e TH |
1593 | pol->pd_online_fn(pd); |
1594 | pd->online = true; | |
1595 | ||
1596 | spin_unlock(&blkg->blkcg->lock); | |
dfd6200a | 1597 | } |
e3ff8887 | 1598 | |
a2b1693b TH |
1599 | __set_bit(pol->plid, q->blkcg_pols); |
1600 | ret = 0; | |
4c55f4f9 | 1601 | |
0d945c1f | 1602 | spin_unlock_irq(&q->queue_lock); |
9d179b86 | 1603 | out: |
344e9ffc | 1604 | if (queue_is_mq(q)) |
bd166ef1 | 1605 | blk_mq_unfreeze_queue(q); |
9d179b86 TH |
1606 | if (pinned_blkg) |
1607 | blkg_put(pinned_blkg); | |
001bea73 TH |
1608 | if (pd_prealloc) |
1609 | pol->pd_free_fn(pd_prealloc); | |
a2b1693b | 1610 | return ret; |
9d179b86 TH |
1611 | |
1612 | enomem: | |
ec14a87e | 1613 | /* alloc failed, take down everything */ |
9d179b86 TH |
1614 | spin_lock_irq(&q->queue_lock); |
1615 | list_for_each_entry(blkg, &q->blkg_list, q_node) { | |
858560b2 | 1616 | struct blkcg *blkcg = blkg->blkcg; |
ec14a87e | 1617 | struct blkg_policy_data *pd; |
858560b2 LJ |
1618 | |
1619 | spin_lock(&blkcg->lock); | |
ec14a87e TH |
1620 | pd = blkg->pd[pol->plid]; |
1621 | if (pd) { | |
1622 | if (pd->online && pol->pd_offline_fn) | |
1623 | pol->pd_offline_fn(pd); | |
1624 | pd->online = false; | |
1625 | pol->pd_free_fn(pd); | |
9d179b86 TH |
1626 | blkg->pd[pol->plid] = NULL; |
1627 | } | |
858560b2 | 1628 | spin_unlock(&blkcg->lock); |
9d179b86 TH |
1629 | } |
1630 | spin_unlock_irq(&q->queue_lock); | |
1631 | ret = -ENOMEM; | |
1632 | goto out; | |
a2b1693b TH |
1633 | } |
1634 | EXPORT_SYMBOL_GPL(blkcg_activate_policy); | |
1635 | ||
1636 | /** | |
40e4996e CH |
1637 | * blkcg_deactivate_policy - deactivate a blkcg policy on a gendisk |
1638 | * @disk: gendisk of interest | |
a2b1693b TH |
1639 | * @pol: blkcg policy to deactivate |
1640 | * | |
40e4996e | 1641 | * Deactivate @pol on @disk. Follows the same synchronization rules as |
a2b1693b TH |
1642 | * blkcg_activate_policy(). |
1643 | */ | |
40e4996e | 1644 | void blkcg_deactivate_policy(struct gendisk *disk, |
3c798398 | 1645 | const struct blkcg_policy *pol) |
a2b1693b | 1646 | { |
40e4996e | 1647 | struct request_queue *q = disk->queue; |
3c798398 | 1648 | struct blkcg_gq *blkg; |
a2b1693b TH |
1649 | |
1650 | if (!blkcg_policy_enabled(q, pol)) | |
1651 | return; | |
1652 | ||
344e9ffc | 1653 | if (queue_is_mq(q)) |
bd166ef1 | 1654 | blk_mq_freeze_queue(q); |
bd166ef1 | 1655 | |
1231039d | 1656 | mutex_lock(&q->blkcg_mutex); |
0d945c1f | 1657 | spin_lock_irq(&q->queue_lock); |
a2b1693b TH |
1658 | |
1659 | __clear_bit(pol->plid, q->blkcg_pols); | |
1660 | ||
1661 | list_for_each_entry(blkg, &q->blkg_list, q_node) { | |
858560b2 LJ |
1662 | struct blkcg *blkcg = blkg->blkcg; |
1663 | ||
1664 | spin_lock(&blkcg->lock); | |
001bea73 | 1665 | if (blkg->pd[pol->plid]) { |
dfd6200a | 1666 | if (blkg->pd[pol->plid]->online && pol->pd_offline_fn) |
a9520cd6 | 1667 | pol->pd_offline_fn(blkg->pd[pol->plid]); |
001bea73 TH |
1668 | pol->pd_free_fn(blkg->pd[pol->plid]); |
1669 | blkg->pd[pol->plid] = NULL; | |
1670 | } | |
858560b2 | 1671 | spin_unlock(&blkcg->lock); |
a2b1693b TH |
1672 | } |
1673 | ||
0d945c1f | 1674 | spin_unlock_irq(&q->queue_lock); |
1231039d | 1675 | mutex_unlock(&q->blkcg_mutex); |
bd166ef1 | 1676 | |
344e9ffc | 1677 | if (queue_is_mq(q)) |
bd166ef1 | 1678 | blk_mq_unfreeze_queue(q); |
a2b1693b TH |
1679 | } |
1680 | EXPORT_SYMBOL_GPL(blkcg_deactivate_policy); | |
1681 | ||
e55cf798 JY |
1682 | static void blkcg_free_all_cpd(struct blkcg_policy *pol) |
1683 | { | |
1684 | struct blkcg *blkcg; | |
1685 | ||
1686 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) { | |
1687 | if (blkcg->cpd[pol->plid]) { | |
1688 | pol->cpd_free_fn(blkcg->cpd[pol->plid]); | |
1689 | blkcg->cpd[pol->plid] = NULL; | |
1690 | } | |
1691 | } | |
1692 | } | |
1693 | ||
8bd435b3 | 1694 | /** |
3c798398 TH |
1695 | * blkcg_policy_register - register a blkcg policy |
1696 | * @pol: blkcg policy to register | |
8bd435b3 | 1697 | * |
3c798398 TH |
1698 | * Register @pol with blkcg core. Might sleep and @pol may be modified on |
1699 | * successful registration. Returns 0 on success and -errno on failure. | |
8bd435b3 | 1700 | */ |
d5bf0291 | 1701 | int blkcg_policy_register(struct blkcg_policy *pol) |
3e252066 | 1702 | { |
06b285bd | 1703 | struct blkcg *blkcg; |
8bd435b3 | 1704 | int i, ret; |
e8989fae | 1705 | |
838f13bf | 1706 | mutex_lock(&blkcg_pol_register_mutex); |
bc0d6501 TH |
1707 | mutex_lock(&blkcg_pol_mutex); |
1708 | ||
8bd435b3 TH |
1709 | /* find an empty slot */ |
1710 | ret = -ENOSPC; | |
1711 | for (i = 0; i < BLKCG_MAX_POLS; i++) | |
3c798398 | 1712 | if (!blkcg_policy[i]) |
8bd435b3 | 1713 | break; |
01c5f85a JA |
1714 | if (i >= BLKCG_MAX_POLS) { |
1715 | pr_warn("blkcg_policy_register: BLKCG_MAX_POLS too small\n"); | |
838f13bf | 1716 | goto err_unlock; |
01c5f85a | 1717 | } |
035d10b2 | 1718 | |
e8401073 | 1719 | /* Make sure cpd/pd_alloc_fn and cpd/pd_free_fn in pairs */ |
1720 | if ((!pol->cpd_alloc_fn ^ !pol->cpd_free_fn) || | |
1721 | (!pol->pd_alloc_fn ^ !pol->pd_free_fn)) | |
1722 | goto err_unlock; | |
1723 | ||
06b285bd | 1724 | /* register @pol */ |
3c798398 | 1725 | pol->plid = i; |
06b285bd TH |
1726 | blkcg_policy[pol->plid] = pol; |
1727 | ||
1728 | /* allocate and install cpd's */ | |
e4a9bde9 | 1729 | if (pol->cpd_alloc_fn) { |
06b285bd TH |
1730 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) { |
1731 | struct blkcg_policy_data *cpd; | |
1732 | ||
e4a9bde9 | 1733 | cpd = pol->cpd_alloc_fn(GFP_KERNEL); |
bbb427e3 | 1734 | if (!cpd) |
06b285bd | 1735 | goto err_free_cpds; |
06b285bd | 1736 | |
81437648 TH |
1737 | blkcg->cpd[pol->plid] = cpd; |
1738 | cpd->blkcg = blkcg; | |
06b285bd | 1739 | cpd->plid = pol->plid; |
06b285bd TH |
1740 | } |
1741 | } | |
1742 | ||
838f13bf | 1743 | mutex_unlock(&blkcg_pol_mutex); |
8bd435b3 | 1744 | |
8bd435b3 | 1745 | /* everything is in place, add intf files for the new policy */ |
2ee867dc TH |
1746 | if (pol->dfl_cftypes) |
1747 | WARN_ON(cgroup_add_dfl_cftypes(&io_cgrp_subsys, | |
1748 | pol->dfl_cftypes)); | |
880f50e2 | 1749 | if (pol->legacy_cftypes) |
c165b3e3 | 1750 | WARN_ON(cgroup_add_legacy_cftypes(&io_cgrp_subsys, |
880f50e2 | 1751 | pol->legacy_cftypes)); |
838f13bf TH |
1752 | mutex_unlock(&blkcg_pol_register_mutex); |
1753 | return 0; | |
1754 | ||
06b285bd | 1755 | err_free_cpds: |
e55cf798 JY |
1756 | if (pol->cpd_free_fn) |
1757 | blkcg_free_all_cpd(pol); | |
1758 | ||
06b285bd | 1759 | blkcg_policy[pol->plid] = NULL; |
838f13bf | 1760 | err_unlock: |
bc0d6501 | 1761 | mutex_unlock(&blkcg_pol_mutex); |
838f13bf | 1762 | mutex_unlock(&blkcg_pol_register_mutex); |
8bd435b3 | 1763 | return ret; |
3e252066 | 1764 | } |
3c798398 | 1765 | EXPORT_SYMBOL_GPL(blkcg_policy_register); |
3e252066 | 1766 | |
8bd435b3 | 1767 | /** |
3c798398 TH |
1768 | * blkcg_policy_unregister - unregister a blkcg policy |
1769 | * @pol: blkcg policy to unregister | |
8bd435b3 | 1770 | * |
3c798398 | 1771 | * Undo blkcg_policy_register(@pol). Might sleep. |
8bd435b3 | 1772 | */ |
3c798398 | 1773 | void blkcg_policy_unregister(struct blkcg_policy *pol) |
3e252066 | 1774 | { |
838f13bf | 1775 | mutex_lock(&blkcg_pol_register_mutex); |
bc0d6501 | 1776 | |
3c798398 | 1777 | if (WARN_ON(blkcg_policy[pol->plid] != pol)) |
8bd435b3 TH |
1778 | goto out_unlock; |
1779 | ||
1780 | /* kill the intf files first */ | |
2ee867dc TH |
1781 | if (pol->dfl_cftypes) |
1782 | cgroup_rm_cftypes(pol->dfl_cftypes); | |
880f50e2 TH |
1783 | if (pol->legacy_cftypes) |
1784 | cgroup_rm_cftypes(pol->legacy_cftypes); | |
44ea53de | 1785 | |
06b285bd | 1786 | /* remove cpds and unregister */ |
838f13bf | 1787 | mutex_lock(&blkcg_pol_mutex); |
06b285bd | 1788 | |
e55cf798 JY |
1789 | if (pol->cpd_free_fn) |
1790 | blkcg_free_all_cpd(pol); | |
1791 | ||
3c798398 | 1792 | blkcg_policy[pol->plid] = NULL; |
06b285bd | 1793 | |
bc0d6501 | 1794 | mutex_unlock(&blkcg_pol_mutex); |
838f13bf TH |
1795 | out_unlock: |
1796 | mutex_unlock(&blkcg_pol_register_mutex); | |
3e252066 | 1797 | } |
3c798398 | 1798 | EXPORT_SYMBOL_GPL(blkcg_policy_unregister); |
903d23f0 | 1799 | |
d09d8df3 JB |
1800 | /* |
1801 | * Scale the accumulated delay based on how long it has been since we updated | |
1802 | * the delay. We only call this when we are adding delay, in case it's been a | |
1803 | * while since we added delay, and when we are checking to see if we need to | |
1804 | * delay a task, to account for any delays that may have occurred. | |
1805 | */ | |
1806 | static void blkcg_scale_delay(struct blkcg_gq *blkg, u64 now) | |
1807 | { | |
1808 | u64 old = atomic64_read(&blkg->delay_start); | |
1809 | ||
54c52e10 TH |
1810 | /* negative use_delay means no scaling, see blkcg_set_delay() */ |
1811 | if (atomic_read(&blkg->use_delay) < 0) | |
1812 | return; | |
1813 | ||
d09d8df3 JB |
1814 | /* |
1815 | * We only want to scale down every second. The idea here is that we | |
1816 | * want to delay people for min(delay_nsec, NSEC_PER_SEC) in a certain | |
1817 | * time window. We only want to throttle tasks for recent delay that | |
1818 | * has occurred, in 1 second time windows since that's the maximum | |
1819 | * things can be throttled. We save the current delay window in | |
1820 | * blkg->last_delay so we know what amount is still left to be charged | |
1821 | * to the blkg from this point onward. blkg->last_use keeps track of | |
1822 | * the use_delay counter. The idea is if we're unthrottling the blkg we | |
1823 | * are ok with whatever is happening now, and we can take away more of | |
1824 | * the accumulated delay as we've already throttled enough that | |
1825 | * everybody is happy with their IO latencies. | |
1826 | */ | |
1827 | if (time_before64(old + NSEC_PER_SEC, now) && | |
96388f57 | 1828 | atomic64_try_cmpxchg(&blkg->delay_start, &old, now)) { |
d09d8df3 JB |
1829 | u64 cur = atomic64_read(&blkg->delay_nsec); |
1830 | u64 sub = min_t(u64, blkg->last_delay, now - old); | |
1831 | int cur_use = atomic_read(&blkg->use_delay); | |
1832 | ||
1833 | /* | |
1834 | * We've been unthrottled, subtract a larger chunk of our | |
1835 | * accumulated delay. | |
1836 | */ | |
1837 | if (cur_use < blkg->last_use) | |
1838 | sub = max_t(u64, sub, blkg->last_delay >> 1); | |
1839 | ||
1840 | /* | |
1841 | * This shouldn't happen, but handle it anyway. Our delay_nsec | |
1842 | * should only ever be growing except here where we subtract out | |
1843 | * min(last_delay, 1 second), but lord knows bugs happen and I'd | |
1844 | * rather not end up with negative numbers. | |
1845 | */ | |
1846 | if (unlikely(cur < sub)) { | |
1847 | atomic64_set(&blkg->delay_nsec, 0); | |
1848 | blkg->last_delay = 0; | |
1849 | } else { | |
1850 | atomic64_sub(sub, &blkg->delay_nsec); | |
1851 | blkg->last_delay = cur - sub; | |
1852 | } | |
1853 | blkg->last_use = cur_use; | |
1854 | } | |
1855 | } | |
1856 | ||
1857 | /* | |
1858 | * This is called when we want to actually walk up the hierarchy and check to | |
1859 | * see if we need to throttle, and then actually throttle if there is some | |
1860 | * accumulated delay. This should only be called upon return to user space so | |
1861 | * we're not holding some lock that would induce a priority inversion. | |
1862 | */ | |
1863 | static void blkcg_maybe_throttle_blkg(struct blkcg_gq *blkg, bool use_memdelay) | |
1864 | { | |
fd112c74 | 1865 | unsigned long pflags; |
5160a5a5 | 1866 | bool clamp; |
08420cf7 | 1867 | u64 now = blk_time_get_ns(); |
d09d8df3 JB |
1868 | u64 exp; |
1869 | u64 delay_nsec = 0; | |
1870 | int tok; | |
1871 | ||
1872 | while (blkg->parent) { | |
5160a5a5 TH |
1873 | int use_delay = atomic_read(&blkg->use_delay); |
1874 | ||
1875 | if (use_delay) { | |
1876 | u64 this_delay; | |
1877 | ||
d09d8df3 | 1878 | blkcg_scale_delay(blkg, now); |
5160a5a5 TH |
1879 | this_delay = atomic64_read(&blkg->delay_nsec); |
1880 | if (this_delay > delay_nsec) { | |
1881 | delay_nsec = this_delay; | |
1882 | clamp = use_delay > 0; | |
1883 | } | |
d09d8df3 JB |
1884 | } |
1885 | blkg = blkg->parent; | |
1886 | } | |
1887 | ||
1888 | if (!delay_nsec) | |
1889 | return; | |
1890 | ||
1891 | /* | |
1892 | * Let's not sleep for all eternity if we've amassed a huge delay. | |
1893 | * Swapping or metadata IO can accumulate 10's of seconds worth of | |
1894 | * delay, and we want userspace to be able to do _something_ so cap the | |
5160a5a5 TH |
1895 | * delays at 0.25s. If there's 10's of seconds worth of delay then the |
1896 | * tasks will be delayed for 0.25 second for every syscall. If | |
1897 | * blkcg_set_delay() was used as indicated by negative use_delay, the | |
1898 | * caller is responsible for regulating the range. | |
d09d8df3 | 1899 | */ |
5160a5a5 TH |
1900 | if (clamp) |
1901 | delay_nsec = min_t(u64, delay_nsec, 250 * NSEC_PER_MSEC); | |
d09d8df3 | 1902 | |
fd112c74 JB |
1903 | if (use_memdelay) |
1904 | psi_memstall_enter(&pflags); | |
d09d8df3 JB |
1905 | |
1906 | exp = ktime_add_ns(now, delay_nsec); | |
1907 | tok = io_schedule_prepare(); | |
1908 | do { | |
1909 | __set_current_state(TASK_KILLABLE); | |
1910 | if (!schedule_hrtimeout(&exp, HRTIMER_MODE_ABS)) | |
1911 | break; | |
1912 | } while (!fatal_signal_pending(current)); | |
1913 | io_schedule_finish(tok); | |
fd112c74 JB |
1914 | |
1915 | if (use_memdelay) | |
1916 | psi_memstall_leave(&pflags); | |
d09d8df3 JB |
1917 | } |
1918 | ||
1919 | /** | |
1920 | * blkcg_maybe_throttle_current - throttle the current task if it has been marked | |
1921 | * | |
1922 | * This is only called if we've been marked with set_notify_resume(). Obviously | |
1923 | * we can be set_notify_resume() for reasons other than blkcg throttling, so we | |
f05837ed | 1924 | * check to see if current->throttle_disk is set and if not this doesn't do |
d09d8df3 JB |
1925 | * anything. This should only ever be called by the resume code, it's not meant |
1926 | * to be called by people willy-nilly as it will actually do the work to | |
1927 | * throttle the task if it is setup for throttling. | |
1928 | */ | |
1929 | void blkcg_maybe_throttle_current(void) | |
1930 | { | |
f05837ed | 1931 | struct gendisk *disk = current->throttle_disk; |
d09d8df3 JB |
1932 | struct blkcg *blkcg; |
1933 | struct blkcg_gq *blkg; | |
1934 | bool use_memdelay = current->use_memdelay; | |
1935 | ||
f05837ed | 1936 | if (!disk) |
d09d8df3 JB |
1937 | return; |
1938 | ||
f05837ed | 1939 | current->throttle_disk = NULL; |
d09d8df3 JB |
1940 | current->use_memdelay = false; |
1941 | ||
1942 | rcu_read_lock(); | |
82778259 | 1943 | blkcg = css_to_blkcg(blkcg_css()); |
d09d8df3 JB |
1944 | if (!blkcg) |
1945 | goto out; | |
9a9c261e | 1946 | blkg = blkg_lookup(blkcg, disk->queue); |
d09d8df3 JB |
1947 | if (!blkg) |
1948 | goto out; | |
7754f669 | 1949 | if (!blkg_tryget(blkg)) |
d09d8df3 JB |
1950 | goto out; |
1951 | rcu_read_unlock(); | |
d09d8df3 JB |
1952 | |
1953 | blkcg_maybe_throttle_blkg(blkg, use_memdelay); | |
1954 | blkg_put(blkg); | |
f05837ed | 1955 | put_disk(disk); |
d09d8df3 JB |
1956 | return; |
1957 | out: | |
1958 | rcu_read_unlock(); | |
d09d8df3 | 1959 | } |
d09d8df3 JB |
1960 | |
1961 | /** | |
1962 | * blkcg_schedule_throttle - this task needs to check for throttling | |
1d6df9d3 | 1963 | * @disk: disk to throttle |
537d71b3 | 1964 | * @use_memdelay: do we charge this to memory delay for PSI |
d09d8df3 JB |
1965 | * |
1966 | * This is called by the IO controller when we know there's delay accumulated | |
1967 | * for the blkg for this task. We do not pass the blkg because there are places | |
1968 | * we call this that may not have that information, the swapping code for | |
de185b56 | 1969 | * instance will only have a block_device at that point. This set's the |
d09d8df3 JB |
1970 | * notify_resume for the task to check and see if it requires throttling before |
1971 | * returning to user space. | |
1972 | * | |
1973 | * We will only schedule once per syscall. You can call this over and over | |
1974 | * again and it will only do the check once upon return to user space, and only | |
1975 | * throttle once. If the task needs to be throttled again it'll need to be | |
1976 | * re-set at the next time we see the task. | |
1977 | */ | |
de185b56 | 1978 | void blkcg_schedule_throttle(struct gendisk *disk, bool use_memdelay) |
d09d8df3 JB |
1979 | { |
1980 | if (unlikely(current->flags & PF_KTHREAD)) | |
1981 | return; | |
1982 | ||
f05837ed CH |
1983 | if (current->throttle_disk != disk) { |
1984 | if (test_bit(GD_DEAD, &disk->state)) | |
49d1822b | 1985 | return; |
f05837ed | 1986 | get_device(disk_to_dev(disk)); |
49d1822b | 1987 | |
f05837ed CH |
1988 | if (current->throttle_disk) |
1989 | put_disk(current->throttle_disk); | |
1990 | current->throttle_disk = disk; | |
49d1822b | 1991 | } |
d09d8df3 | 1992 | |
d09d8df3 JB |
1993 | if (use_memdelay) |
1994 | current->use_memdelay = use_memdelay; | |
1995 | set_notify_resume(current); | |
1996 | } | |
d09d8df3 JB |
1997 | |
1998 | /** | |
1999 | * blkcg_add_delay - add delay to this blkg | |
537d71b3 BVA |
2000 | * @blkg: blkg of interest |
2001 | * @now: the current time in nanoseconds | |
2002 | * @delta: how many nanoseconds of delay to add | |
d09d8df3 JB |
2003 | * |
2004 | * Charge @delta to the blkg's current delay accumulation. This is used to | |
2005 | * throttle tasks if an IO controller thinks we need more throttling. | |
2006 | */ | |
2007 | void blkcg_add_delay(struct blkcg_gq *blkg, u64 now, u64 delta) | |
2008 | { | |
54c52e10 TH |
2009 | if (WARN_ON_ONCE(atomic_read(&blkg->use_delay) < 0)) |
2010 | return; | |
d09d8df3 JB |
2011 | blkcg_scale_delay(blkg, now); |
2012 | atomic64_add(delta, &blkg->delay_nsec); | |
2013 | } | |
d09d8df3 | 2014 | |
28fc591f CH |
2015 | /** |
2016 | * blkg_tryget_closest - try and get a blkg ref on the closet blkg | |
13c7863d CH |
2017 | * @bio: target bio |
2018 | * @css: target css | |
28fc591f | 2019 | * |
13c7863d CH |
2020 | * As the failure mode here is to walk up the blkg tree, this ensure that the |
2021 | * blkg->parent pointers are always valid. This returns the blkg that it ended | |
2022 | * up taking a reference on or %NULL if no reference was taken. | |
28fc591f | 2023 | */ |
13c7863d CH |
2024 | static inline struct blkcg_gq *blkg_tryget_closest(struct bio *bio, |
2025 | struct cgroup_subsys_state *css) | |
28fc591f | 2026 | { |
13c7863d | 2027 | struct blkcg_gq *blkg, *ret_blkg = NULL; |
28fc591f | 2028 | |
13c7863d | 2029 | rcu_read_lock(); |
99e60387 | 2030 | blkg = blkg_lookup_create(css_to_blkcg(css), bio->bi_bdev->bd_disk); |
28fc591f CH |
2031 | while (blkg) { |
2032 | if (blkg_tryget(blkg)) { | |
2033 | ret_blkg = blkg; | |
2034 | break; | |
2035 | } | |
2036 | blkg = blkg->parent; | |
2037 | } | |
13c7863d | 2038 | rcu_read_unlock(); |
28fc591f CH |
2039 | |
2040 | return ret_blkg; | |
2041 | } | |
2042 | ||
2043 | /** | |
2044 | * bio_associate_blkg_from_css - associate a bio with a specified css | |
2045 | * @bio: target bio | |
2046 | * @css: target css | |
2047 | * | |
2048 | * Associate @bio with the blkg found by combining the css's blkg and the | |
2049 | * request_queue of the @bio. An association failure is handled by walking up | |
2050 | * the blkg tree. Therefore, the blkg associated can be anything between @blkg | |
2051 | * and q->root_blkg. This situation only happens when a cgroup is dying and | |
2052 | * then the remaining bios will spill to the closest alive blkg. | |
2053 | * | |
2054 | * A reference will be taken on the blkg and will be released when @bio is | |
2055 | * freed. | |
2056 | */ | |
2057 | void bio_associate_blkg_from_css(struct bio *bio, | |
2058 | struct cgroup_subsys_state *css) | |
2059 | { | |
28fc591f CH |
2060 | if (bio->bi_blkg) |
2061 | blkg_put(bio->bi_blkg); | |
2062 | ||
a5b97526 | 2063 | if (css && css->parent) { |
13c7863d | 2064 | bio->bi_blkg = blkg_tryget_closest(bio, css); |
a5b97526 | 2065 | } else { |
ed6cddef PB |
2066 | blkg_get(bdev_get_queue(bio->bi_bdev)->root_blkg); |
2067 | bio->bi_blkg = bdev_get_queue(bio->bi_bdev)->root_blkg; | |
a5b97526 | 2068 | } |
28fc591f CH |
2069 | } |
2070 | EXPORT_SYMBOL_GPL(bio_associate_blkg_from_css); | |
2071 | ||
2072 | /** | |
2073 | * bio_associate_blkg - associate a bio with a blkg | |
2074 | * @bio: target bio | |
2075 | * | |
2076 | * Associate @bio with the blkg found from the bio's css and request_queue. | |
2077 | * If one is not found, bio_lookup_blkg() creates the blkg. If a blkg is | |
2078 | * already associated, the css is reused and association redone as the | |
2079 | * request_queue may have changed. | |
2080 | */ | |
2081 | void bio_associate_blkg(struct bio *bio) | |
2082 | { | |
2083 | struct cgroup_subsys_state *css; | |
2084 | ||
6c9b9708 KK |
2085 | if (blk_op_is_passthrough(bio->bi_opf)) |
2086 | return; | |
2087 | ||
28fc591f CH |
2088 | rcu_read_lock(); |
2089 | ||
2090 | if (bio->bi_blkg) | |
bbb1ebe7 | 2091 | css = bio_blkcg_css(bio); |
28fc591f CH |
2092 | else |
2093 | css = blkcg_css(); | |
2094 | ||
2095 | bio_associate_blkg_from_css(bio, css); | |
2096 | ||
2097 | rcu_read_unlock(); | |
2098 | } | |
2099 | EXPORT_SYMBOL_GPL(bio_associate_blkg); | |
2100 | ||
2101 | /** | |
2102 | * bio_clone_blkg_association - clone blkg association from src to dst bio | |
2103 | * @dst: destination bio | |
2104 | * @src: source bio | |
2105 | */ | |
2106 | void bio_clone_blkg_association(struct bio *dst, struct bio *src) | |
2107 | { | |
22b106e5 JK |
2108 | if (src->bi_blkg) |
2109 | bio_associate_blkg_from_css(dst, bio_blkcg_css(src)); | |
28fc591f CH |
2110 | } |
2111 | EXPORT_SYMBOL_GPL(bio_clone_blkg_association); | |
2112 | ||
db18a53e CH |
2113 | static int blk_cgroup_io_type(struct bio *bio) |
2114 | { | |
2115 | if (op_is_discard(bio->bi_opf)) | |
2116 | return BLKG_IOSTAT_DISCARD; | |
2117 | if (op_is_write(bio->bi_opf)) | |
2118 | return BLKG_IOSTAT_WRITE; | |
2119 | return BLKG_IOSTAT_READ; | |
2120 | } | |
2121 | ||
2122 | void blk_cgroup_bio_start(struct bio *bio) | |
2123 | { | |
3b8cc629 | 2124 | struct blkcg *blkcg = bio->bi_blkg->blkcg; |
db18a53e CH |
2125 | int rwd = blk_cgroup_io_type(bio), cpu; |
2126 | struct blkg_iostat_set *bis; | |
3c08b093 | 2127 | unsigned long flags; |
db18a53e | 2128 | |
ad7c3b41 JH |
2129 | if (!cgroup_subsys_on_dfl(io_cgrp_subsys)) |
2130 | return; | |
2131 | ||
0416f3be ML |
2132 | /* Root-level stats are sourced from system-wide IO stats */ |
2133 | if (!cgroup_parent(blkcg->css.cgroup)) | |
2134 | return; | |
2135 | ||
db18a53e CH |
2136 | cpu = get_cpu(); |
2137 | bis = per_cpu_ptr(bio->bi_blkg->iostat_cpu, cpu); | |
3c08b093 | 2138 | flags = u64_stats_update_begin_irqsave(&bis->sync); |
db18a53e CH |
2139 | |
2140 | /* | |
2141 | * If the bio is flagged with BIO_CGROUP_ACCT it means this is a split | |
2142 | * bio and we would have already accounted for the size of the bio. | |
2143 | */ | |
2144 | if (!bio_flagged(bio, BIO_CGROUP_ACCT)) { | |
2145 | bio_set_flag(bio, BIO_CGROUP_ACCT); | |
0b8cc25d | 2146 | bis->cur.bytes[rwd] += bio->bi_iter.bi_size; |
db18a53e CH |
2147 | } |
2148 | bis->cur.ios[rwd]++; | |
2149 | ||
3b8cc629 WL |
2150 | /* |
2151 | * If the iostat_cpu isn't in a lockless list, put it into the | |
2152 | * list to indicate that a stat update is pending. | |
2153 | */ | |
2154 | if (!READ_ONCE(bis->lqueued)) { | |
2155 | struct llist_head *lhead = this_cpu_ptr(blkcg->lhead); | |
2156 | ||
2157 | llist_add(&bis->lnode, lhead); | |
2158 | WRITE_ONCE(bis->lqueued, true); | |
3b8cc629 WL |
2159 | } |
2160 | ||
3c08b093 | 2161 | u64_stats_update_end_irqrestore(&bis->sync, flags); |
ad7c3b41 | 2162 | cgroup_rstat_updated(blkcg->css.cgroup, cpu); |
db18a53e CH |
2163 | put_cpu(); |
2164 | } | |
2165 | ||
216889aa CH |
2166 | bool blk_cgroup_congested(void) |
2167 | { | |
2168 | struct cgroup_subsys_state *css; | |
2169 | bool ret = false; | |
2170 | ||
2171 | rcu_read_lock(); | |
d200ca14 | 2172 | for (css = blkcg_css(); css; css = css->parent) { |
216889aa CH |
2173 | if (atomic_read(&css->cgroup->congestion_count)) { |
2174 | ret = true; | |
2175 | break; | |
2176 | } | |
216889aa CH |
2177 | } |
2178 | rcu_read_unlock(); | |
2179 | return ret; | |
2180 | } | |
2181 | ||
903d23f0 JB |
2182 | module_param(blkcg_debug_stats, bool, 0644); |
2183 | MODULE_PARM_DESC(blkcg_debug_stats, "True if you want debug stats, false if not"); |