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); |
d3f77dfd TH |
221 | bio_list_merge(&bios, &blkg->async_bios); |
222 | bio_list_init(&blkg->async_bios); | |
12be09fe | 223 | spin_unlock(&blkg->async_bio_lock); |
d3f77dfd | 224 | |
192f1c6b XT |
225 | /* start plug only when bio_list contains at least 2 bios */ |
226 | if (bios.head && bios.head->bi_next) { | |
227 | need_plug = true; | |
228 | blk_start_plug(&plug); | |
229 | } | |
d3f77dfd TH |
230 | while ((bio = bio_list_pop(&bios))) |
231 | submit_bio(bio); | |
192f1c6b XT |
232 | if (need_plug) |
233 | blk_finish_plug(&plug); | |
d3f77dfd TH |
234 | } |
235 | ||
2c275afe CH |
236 | /* |
237 | * When a shared kthread issues a bio for a cgroup, doing so synchronously can | |
238 | * lead to priority inversions as the kthread can be trapped waiting for that | |
239 | * cgroup. Use this helper instead of submit_bio to punt the actual issuing to | |
240 | * a dedicated per-blkcg work item to avoid such priority inversions. | |
241 | */ | |
242 | void blkcg_punt_bio_submit(struct bio *bio) | |
243 | { | |
244 | struct blkcg_gq *blkg = bio->bi_blkg; | |
245 | ||
246 | if (blkg->parent) { | |
247 | spin_lock(&blkg->async_bio_lock); | |
248 | bio_list_add(&blkg->async_bios, bio); | |
249 | spin_unlock(&blkg->async_bio_lock); | |
250 | queue_work(blkcg_punt_bio_wq, &blkg->async_bio_work); | |
251 | } else { | |
252 | /* never bounce for the root cgroup */ | |
253 | submit_bio(bio); | |
254 | } | |
255 | } | |
256 | EXPORT_SYMBOL_GPL(blkcg_punt_bio_submit); | |
257 | ||
258 | static int __init blkcg_punt_bio_init(void) | |
259 | { | |
260 | blkcg_punt_bio_wq = alloc_workqueue("blkcg_punt_bio", | |
261 | WQ_MEM_RECLAIM | WQ_FREEZABLE | | |
262 | WQ_UNBOUND | WQ_SYSFS, 0); | |
263 | if (!blkcg_punt_bio_wq) | |
264 | return -ENOMEM; | |
265 | return 0; | |
266 | } | |
267 | subsys_initcall(blkcg_punt_bio_init); | |
268 | #endif /* CONFIG_BLK_CGROUP_PUNT_BIO */ | |
269 | ||
bbb1ebe7 CH |
270 | /** |
271 | * bio_blkcg_css - return the blkcg CSS associated with a bio | |
272 | * @bio: target bio | |
273 | * | |
274 | * This returns the CSS for the blkcg associated with a bio, or %NULL if not | |
275 | * associated. Callers are expected to either handle %NULL or know association | |
276 | * has been done prior to calling this. | |
277 | */ | |
278 | struct cgroup_subsys_state *bio_blkcg_css(struct bio *bio) | |
279 | { | |
280 | if (!bio || !bio->bi_blkg) | |
281 | return NULL; | |
282 | return &bio->bi_blkg->blkcg->css; | |
283 | } | |
284 | EXPORT_SYMBOL_GPL(bio_blkcg_css); | |
285 | ||
397c9f46 CH |
286 | /** |
287 | * blkcg_parent - get the parent of a blkcg | |
288 | * @blkcg: blkcg of interest | |
289 | * | |
290 | * Return the parent blkcg of @blkcg. Can be called anytime. | |
291 | */ | |
292 | static inline struct blkcg *blkcg_parent(struct blkcg *blkcg) | |
293 | { | |
294 | return css_to_blkcg(blkcg->css.parent); | |
295 | } | |
296 | ||
0381411e TH |
297 | /** |
298 | * blkg_alloc - allocate a blkg | |
299 | * @blkcg: block cgroup the new blkg is associated with | |
99e60387 | 300 | * @disk: gendisk the new blkg is associated with |
15974993 | 301 | * @gfp_mask: allocation mask to use |
0381411e | 302 | * |
e8989fae | 303 | * Allocate a new blkg assocating @blkcg and @q. |
0381411e | 304 | */ |
99e60387 | 305 | static struct blkcg_gq *blkg_alloc(struct blkcg *blkcg, struct gendisk *disk, |
15974993 | 306 | gfp_t gfp_mask) |
0381411e | 307 | { |
3c798398 | 308 | struct blkcg_gq *blkg; |
f7331648 | 309 | int i, cpu; |
0381411e TH |
310 | |
311 | /* alloc and init base part */ | |
99e60387 | 312 | blkg = kzalloc_node(sizeof(*blkg), gfp_mask, disk->queue->node); |
0381411e TH |
313 | if (!blkg) |
314 | return NULL; | |
ef069b97 | 315 | if (percpu_ref_init(&blkg->refcnt, blkg_release, 0, gfp_mask)) |
0b6f93bd | 316 | goto out_free_blkg; |
f7331648 TH |
317 | blkg->iostat_cpu = alloc_percpu_gfp(struct blkg_iostat_set, gfp_mask); |
318 | if (!blkg->iostat_cpu) | |
0b6f93bd | 319 | goto out_exit_refcnt; |
99e60387 | 320 | if (!blk_get_queue(disk->queue)) |
0b6f93bd | 321 | goto out_free_iostat; |
0a9a25ca | 322 | |
99e60387 | 323 | blkg->q = disk->queue; |
e8989fae | 324 | INIT_LIST_HEAD(&blkg->q_node); |
2c275afe CH |
325 | blkg->blkcg = blkcg; |
326 | #ifdef CONFIG_BLK_CGROUP_PUNT_BIO | |
d3f77dfd TH |
327 | spin_lock_init(&blkg->async_bio_lock); |
328 | bio_list_init(&blkg->async_bios); | |
329 | INIT_WORK(&blkg->async_bio_work, blkg_async_bio_workfn); | |
2c275afe | 330 | #endif |
0381411e | 331 | |
f7331648 | 332 | u64_stats_init(&blkg->iostat.sync); |
3b8cc629 | 333 | for_each_possible_cpu(cpu) { |
f7331648 | 334 | u64_stats_init(&per_cpu_ptr(blkg->iostat_cpu, cpu)->sync); |
3b8cc629 WL |
335 | per_cpu_ptr(blkg->iostat_cpu, cpu)->blkg = blkg; |
336 | } | |
f7331648 | 337 | |
8bd435b3 | 338 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
3c798398 | 339 | struct blkcg_policy *pol = blkcg_policy[i]; |
e8989fae | 340 | struct blkg_policy_data *pd; |
0381411e | 341 | |
99e60387 | 342 | if (!blkcg_policy_enabled(disk->queue, pol)) |
e8989fae TH |
343 | continue; |
344 | ||
345 | /* alloc per-policy data and attach it to blkg */ | |
0a0b4f79 | 346 | pd = pol->pd_alloc_fn(disk, blkcg, gfp_mask); |
a051661c | 347 | if (!pd) |
0b6f93bd | 348 | goto out_free_pds; |
e8989fae TH |
349 | blkg->pd[i] = pd; |
350 | pd->blkg = blkg; | |
b276a876 | 351 | pd->plid = i; |
dfd6200a | 352 | pd->online = false; |
e8989fae TH |
353 | } |
354 | ||
0381411e | 355 | return blkg; |
a051661c | 356 | |
0b6f93bd CH |
357 | out_free_pds: |
358 | while (--i >= 0) | |
359 | if (blkg->pd[i]) | |
360 | blkcg_policy[i]->pd_free_fn(blkg->pd[i]); | |
a06377c5 | 361 | blk_put_queue(disk->queue); |
0b6f93bd CH |
362 | out_free_iostat: |
363 | free_percpu(blkg->iostat_cpu); | |
364 | out_exit_refcnt: | |
365 | percpu_ref_exit(&blkg->refcnt); | |
366 | out_free_blkg: | |
367 | kfree(blkg); | |
a051661c | 368 | return NULL; |
0381411e TH |
369 | } |
370 | ||
15974993 | 371 | /* |
d708f0d5 JA |
372 | * If @new_blkg is %NULL, this function tries to allocate a new one as |
373 | * necessary using %GFP_NOWAIT. @new_blkg is always consumed on return. | |
15974993 | 374 | */ |
99e60387 | 375 | static struct blkcg_gq *blkg_create(struct blkcg *blkcg, struct gendisk *disk, |
d708f0d5 | 376 | struct blkcg_gq *new_blkg) |
5624a4e4 | 377 | { |
d708f0d5 | 378 | struct blkcg_gq *blkg; |
f427d909 | 379 | int i, ret; |
5624a4e4 | 380 | |
99e60387 | 381 | lockdep_assert_held(&disk->queue->queue_lock); |
cd1604fa | 382 | |
0273ac34 | 383 | /* request_queue is dying, do not create/recreate a blkg */ |
99e60387 | 384 | if (blk_queue_dying(disk->queue)) { |
0273ac34 DZ |
385 | ret = -ENODEV; |
386 | goto err_free_blkg; | |
387 | } | |
388 | ||
7ee9c562 | 389 | /* blkg holds a reference to blkcg */ |
ec903c0c | 390 | if (!css_tryget_online(&blkcg->css)) { |
20386ce0 | 391 | ret = -ENODEV; |
93e6d5d8 | 392 | goto err_free_blkg; |
15974993 | 393 | } |
cd1604fa | 394 | |
d708f0d5 JA |
395 | /* allocate */ |
396 | if (!new_blkg) { | |
99e60387 | 397 | new_blkg = blkg_alloc(blkcg, disk, GFP_NOWAIT | __GFP_NOWARN); |
d708f0d5 JA |
398 | if (unlikely(!new_blkg)) { |
399 | ret = -ENOMEM; | |
8c911f3d | 400 | goto err_put_css; |
15974993 TH |
401 | } |
402 | } | |
d708f0d5 | 403 | blkg = new_blkg; |
cd1604fa | 404 | |
db613670 | 405 | /* link parent */ |
3c547865 | 406 | if (blkcg_parent(blkcg)) { |
99e60387 | 407 | blkg->parent = blkg_lookup(blkcg_parent(blkcg), disk->queue); |
3c547865 | 408 | if (WARN_ON_ONCE(!blkg->parent)) { |
20386ce0 | 409 | ret = -ENODEV; |
8c911f3d | 410 | goto err_put_css; |
3c547865 TH |
411 | } |
412 | blkg_get(blkg->parent); | |
413 | } | |
414 | ||
db613670 TH |
415 | /* invoke per-policy init */ |
416 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
417 | struct blkcg_policy *pol = blkcg_policy[i]; | |
418 | ||
419 | if (blkg->pd[i] && pol->pd_init_fn) | |
a9520cd6 | 420 | pol->pd_init_fn(blkg->pd[i]); |
db613670 TH |
421 | } |
422 | ||
423 | /* insert */ | |
cd1604fa | 424 | spin_lock(&blkcg->lock); |
99e60387 | 425 | ret = radix_tree_insert(&blkcg->blkg_tree, disk->queue->id, blkg); |
a637120e TH |
426 | if (likely(!ret)) { |
427 | hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list); | |
99e60387 | 428 | list_add(&blkg->q_node, &disk->queue->blkg_list); |
f427d909 TH |
429 | |
430 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
431 | struct blkcg_policy *pol = blkcg_policy[i]; | |
432 | ||
dfd6200a YK |
433 | if (blkg->pd[i]) { |
434 | if (pol->pd_online_fn) | |
435 | pol->pd_online_fn(blkg->pd[i]); | |
436 | blkg->pd[i]->online = true; | |
437 | } | |
f427d909 | 438 | } |
a637120e | 439 | } |
f427d909 | 440 | blkg->online = true; |
cd1604fa | 441 | spin_unlock(&blkcg->lock); |
496fb780 | 442 | |
ec13b1d6 | 443 | if (!ret) |
a637120e | 444 | return blkg; |
15974993 | 445 | |
3c547865 TH |
446 | /* @blkg failed fully initialized, use the usual release path */ |
447 | blkg_put(blkg); | |
448 | return ERR_PTR(ret); | |
449 | ||
d708f0d5 | 450 | err_put_css: |
496fb780 | 451 | css_put(&blkcg->css); |
93e6d5d8 | 452 | err_free_blkg: |
28e538a3 CH |
453 | if (new_blkg) |
454 | blkg_free(new_blkg); | |
93e6d5d8 | 455 | return ERR_PTR(ret); |
31e4c28d | 456 | } |
3c96cb32 | 457 | |
86cde6b6 | 458 | /** |
8c546287 | 459 | * blkg_lookup_create - lookup blkg, try to create one if not there |
86cde6b6 | 460 | * @blkcg: blkcg of interest |
99e60387 | 461 | * @disk: gendisk of interest |
86cde6b6 | 462 | * |
99e60387 | 463 | * Lookup blkg for the @blkcg - @disk pair. If it doesn't exist, try to |
3c547865 TH |
464 | * create one. blkg creation is performed recursively from blkcg_root such |
465 | * that all non-root blkg's have access to the parent blkg. This function | |
99e60387 | 466 | * should be called under RCU read lock and takes @disk->queue->queue_lock. |
86cde6b6 | 467 | * |
beea9da0 DZ |
468 | * Returns the blkg or the closest blkg if blkg_create() fails as it walks |
469 | * down from root. | |
86cde6b6 | 470 | */ |
8c546287 | 471 | static struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg, |
99e60387 | 472 | struct gendisk *disk) |
3c96cb32 | 473 | { |
99e60387 | 474 | struct request_queue *q = disk->queue; |
86cde6b6 | 475 | struct blkcg_gq *blkg; |
8c546287 | 476 | unsigned long flags; |
86cde6b6 TH |
477 | |
478 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
86cde6b6 | 479 | |
8c546287 | 480 | blkg = blkg_lookup(blkcg, q); |
86cde6b6 TH |
481 | if (blkg) |
482 | return blkg; | |
483 | ||
8c546287 | 484 | spin_lock_irqsave(&q->queue_lock, flags); |
4a69f325 CH |
485 | blkg = blkg_lookup(blkcg, q); |
486 | if (blkg) { | |
5765033c CH |
487 | if (blkcg != &blkcg_root && |
488 | blkg != rcu_dereference(blkcg->blkg_hint)) | |
489 | rcu_assign_pointer(blkcg->blkg_hint, blkg); | |
8c546287 | 490 | goto found; |
4a69f325 | 491 | } |
8c546287 | 492 | |
3c547865 TH |
493 | /* |
494 | * Create blkgs walking down from blkcg_root to @blkcg, so that all | |
beea9da0 DZ |
495 | * non-root blkgs have access to their parents. Returns the closest |
496 | * blkg to the intended blkg should blkg_create() fail. | |
3c547865 TH |
497 | */ |
498 | while (true) { | |
499 | struct blkcg *pos = blkcg; | |
500 | struct blkcg *parent = blkcg_parent(blkcg); | |
beea9da0 DZ |
501 | struct blkcg_gq *ret_blkg = q->root_blkg; |
502 | ||
503 | while (parent) { | |
79fcc5be | 504 | blkg = blkg_lookup(parent, q); |
beea9da0 DZ |
505 | if (blkg) { |
506 | /* remember closest blkg */ | |
507 | ret_blkg = blkg; | |
508 | break; | |
509 | } | |
3c547865 TH |
510 | pos = parent; |
511 | parent = blkcg_parent(parent); | |
512 | } | |
513 | ||
99e60387 | 514 | blkg = blkg_create(pos, disk, NULL); |
8c546287 CH |
515 | if (IS_ERR(blkg)) { |
516 | blkg = ret_blkg; | |
517 | break; | |
518 | } | |
beea9da0 | 519 | if (pos == blkcg) |
8c546287 | 520 | break; |
b978962a DZ |
521 | } |
522 | ||
8c546287 CH |
523 | found: |
524 | spin_unlock_irqrestore(&q->queue_lock, flags); | |
b978962a DZ |
525 | return blkg; |
526 | } | |
527 | ||
3c798398 | 528 | static void blkg_destroy(struct blkcg_gq *blkg) |
03aa264a | 529 | { |
3c798398 | 530 | struct blkcg *blkcg = blkg->blkcg; |
6b065462 | 531 | int i; |
03aa264a | 532 | |
0d945c1f | 533 | lockdep_assert_held(&blkg->q->queue_lock); |
9f13ef67 | 534 | lockdep_assert_held(&blkcg->lock); |
03aa264a | 535 | |
f1c006f1 YK |
536 | /* |
537 | * blkg stays on the queue list until blkg_free_workfn(), see details in | |
538 | * blkg_free_workfn(), hence this function can be called from | |
539 | * blkcg_destroy_blkgs() first and again from blkg_destroy_all() before | |
540 | * blkg_free_workfn(). | |
541 | */ | |
542 | if (hlist_unhashed(&blkg->blkcg_node)) | |
543 | return; | |
a637120e | 544 | |
6b065462 DZF |
545 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
546 | struct blkcg_policy *pol = blkcg_policy[i]; | |
547 | ||
dfd6200a | 548 | if (blkg->pd[i] && blkg->pd[i]->online) { |
f37bf75c | 549 | blkg->pd[i]->online = false; |
dfd6200a YK |
550 | if (pol->pd_offline_fn) |
551 | pol->pd_offline_fn(blkg->pd[i]); | |
dfd6200a | 552 | } |
6b065462 DZF |
553 | } |
554 | ||
f427d909 TH |
555 | blkg->online = false; |
556 | ||
a637120e | 557 | radix_tree_delete(&blkcg->blkg_tree, blkg->q->id); |
9f13ef67 | 558 | hlist_del_init_rcu(&blkg->blkcg_node); |
03aa264a | 559 | |
a637120e TH |
560 | /* |
561 | * Both setting lookup hint to and clearing it from @blkg are done | |
562 | * under queue_lock. If it's not pointing to @blkg now, it never | |
563 | * will. Hint assignment itself can race safely. | |
564 | */ | |
ec6c676a | 565 | if (rcu_access_pointer(blkcg->blkg_hint) == blkg) |
a637120e TH |
566 | rcu_assign_pointer(blkcg->blkg_hint, NULL); |
567 | ||
03aa264a TH |
568 | /* |
569 | * Put the reference taken at the time of creation so that when all | |
570 | * queues are gone, group can be destroyed. | |
571 | */ | |
7fcf2b03 | 572 | percpu_ref_kill(&blkg->refcnt); |
03aa264a TH |
573 | } |
574 | ||
00ad6991 | 575 | static void blkg_destroy_all(struct gendisk *disk) |
72e06c25 | 576 | { |
00ad6991 | 577 | struct request_queue *q = disk->queue; |
3c798398 | 578 | struct blkcg_gq *blkg, *n; |
a731763f | 579 | int count = BLKG_DESTROY_BATCH_SIZE; |
72e06c25 | 580 | |
a731763f | 581 | restart: |
0d945c1f | 582 | spin_lock_irq(&q->queue_lock); |
9f13ef67 | 583 | list_for_each_entry_safe(blkg, n, &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 TH |
604 | |
605 | q->root_blkg = NULL; | |
0d945c1f | 606 | spin_unlock_irq(&q->queue_lock); |
72e06c25 TH |
607 | } |
608 | ||
182446d0 TH |
609 | static int blkcg_reset_stats(struct cgroup_subsys_state *css, |
610 | struct cftype *cftype, u64 val) | |
303a3acb | 611 | { |
182446d0 | 612 | struct blkcg *blkcg = css_to_blkcg(css); |
3c798398 | 613 | struct blkcg_gq *blkg; |
f7331648 | 614 | int i, cpu; |
303a3acb | 615 | |
838f13bf | 616 | mutex_lock(&blkcg_pol_mutex); |
303a3acb | 617 | spin_lock_irq(&blkcg->lock); |
997a026c TH |
618 | |
619 | /* | |
620 | * Note that stat reset is racy - it doesn't synchronize against | |
621 | * stat updates. This is a debug feature which shouldn't exist | |
622 | * anyway. If you get hit by a race, retry. | |
623 | */ | |
b67bfe0d | 624 | hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) { |
f7331648 TH |
625 | for_each_possible_cpu(cpu) { |
626 | struct blkg_iostat_set *bis = | |
627 | per_cpu_ptr(blkg->iostat_cpu, cpu); | |
628 | memset(bis, 0, sizeof(*bis)); | |
3d2af77e WL |
629 | |
630 | /* Re-initialize the cleared blkg_iostat_set */ | |
631 | u64_stats_init(&bis->sync); | |
632 | bis->blkg = blkg; | |
f7331648 TH |
633 | } |
634 | memset(&blkg->iostat, 0, sizeof(blkg->iostat)); | |
3d2af77e | 635 | u64_stats_init(&blkg->iostat.sync); |
77ea7338 | 636 | |
8bd435b3 | 637 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
3c798398 | 638 | struct blkcg_policy *pol = blkcg_policy[i]; |
549d3aa8 | 639 | |
a9520cd6 TH |
640 | if (blkg->pd[i] && pol->pd_reset_stats_fn) |
641 | pol->pd_reset_stats_fn(blkg->pd[i]); | |
bc0d6501 | 642 | } |
303a3acb | 643 | } |
f0bdc8cd | 644 | |
303a3acb | 645 | spin_unlock_irq(&blkcg->lock); |
bc0d6501 | 646 | mutex_unlock(&blkcg_pol_mutex); |
303a3acb DS |
647 | return 0; |
648 | } | |
649 | ||
dd165eb3 | 650 | const char *blkg_dev_name(struct blkcg_gq *blkg) |
303a3acb | 651 | { |
a06377c5 | 652 | if (!blkg->q->disk) |
edb0872f | 653 | return NULL; |
d152c682 | 654 | return bdi_dev_name(blkg->q->disk->bdi); |
303a3acb DS |
655 | } |
656 | ||
d3d32e69 TH |
657 | /** |
658 | * blkcg_print_blkgs - helper for printing per-blkg data | |
659 | * @sf: seq_file to print to | |
660 | * @blkcg: blkcg of interest | |
661 | * @prfill: fill function to print out a blkg | |
662 | * @pol: policy in question | |
663 | * @data: data to be passed to @prfill | |
664 | * @show_total: to print out sum of prfill return values or not | |
665 | * | |
666 | * This function invokes @prfill on each blkg of @blkcg if pd for the | |
667 | * policy specified by @pol exists. @prfill is invoked with @sf, the | |
810ecfa7 TH |
668 | * policy data and @data and the matching queue lock held. If @show_total |
669 | * is %true, the sum of the return values from @prfill is printed with | |
670 | * "Total" label at the end. | |
d3d32e69 TH |
671 | * |
672 | * This is to be used to construct print functions for | |
673 | * cftype->read_seq_string method. | |
674 | */ | |
3c798398 | 675 | void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg, |
f95a04af TH |
676 | u64 (*prfill)(struct seq_file *, |
677 | struct blkg_policy_data *, int), | |
3c798398 | 678 | const struct blkcg_policy *pol, int data, |
ec399347 | 679 | bool show_total) |
5624a4e4 | 680 | { |
3c798398 | 681 | struct blkcg_gq *blkg; |
d3d32e69 | 682 | u64 total = 0; |
5624a4e4 | 683 | |
810ecfa7 | 684 | rcu_read_lock(); |
ee89f812 | 685 | hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) { |
0d945c1f | 686 | spin_lock_irq(&blkg->q->queue_lock); |
a2b1693b | 687 | if (blkcg_policy_enabled(blkg->q, pol)) |
f95a04af | 688 | total += prfill(sf, blkg->pd[pol->plid], data); |
0d945c1f | 689 | spin_unlock_irq(&blkg->q->queue_lock); |
810ecfa7 TH |
690 | } |
691 | rcu_read_unlock(); | |
d3d32e69 TH |
692 | |
693 | if (show_total) | |
694 | seq_printf(sf, "Total %llu\n", (unsigned long long)total); | |
695 | } | |
829fdb50 | 696 | EXPORT_SYMBOL_GPL(blkcg_print_blkgs); |
d3d32e69 TH |
697 | |
698 | /** | |
699 | * __blkg_prfill_u64 - prfill helper for a single u64 value | |
700 | * @sf: seq_file to print to | |
f95a04af | 701 | * @pd: policy private data of interest |
d3d32e69 TH |
702 | * @v: value to print |
703 | * | |
37754595 | 704 | * Print @v to @sf for the device associated with @pd. |
d3d32e69 | 705 | */ |
f95a04af | 706 | u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v) |
d3d32e69 | 707 | { |
f95a04af | 708 | const char *dname = blkg_dev_name(pd->blkg); |
d3d32e69 TH |
709 | |
710 | if (!dname) | |
711 | return 0; | |
712 | ||
713 | seq_printf(sf, "%s %llu\n", dname, (unsigned long long)v); | |
714 | return v; | |
715 | } | |
829fdb50 | 716 | EXPORT_SYMBOL_GPL(__blkg_prfill_u64); |
d3d32e69 | 717 | |
015d254c | 718 | /** |
faffaab2 TH |
719 | * blkg_conf_init - initialize a blkg_conf_ctx |
720 | * @ctx: blkg_conf_ctx to initialize | |
721 | * @input: input string | |
015d254c | 722 | * |
faffaab2 TH |
723 | * Initialize @ctx which can be used to parse blkg config input string @input. |
724 | * Once initialized, @ctx can be used with blkg_conf_open_bdev() and | |
725 | * blkg_conf_prep(), and must be cleaned up with blkg_conf_exit(). | |
726 | */ | |
727 | void blkg_conf_init(struct blkg_conf_ctx *ctx, char *input) | |
728 | { | |
729 | *ctx = (struct blkg_conf_ctx){ .input = input }; | |
730 | } | |
731 | EXPORT_SYMBOL_GPL(blkg_conf_init); | |
732 | ||
733 | /** | |
734 | * blkg_conf_open_bdev - parse and open bdev for per-blkg config update | |
735 | * @ctx: blkg_conf_ctx initialized with blkg_conf_init() | |
015d254c | 736 | * |
faffaab2 TH |
737 | * Parse the device node prefix part, MAJ:MIN, of per-blkg config update from |
738 | * @ctx->input and get and store the matching bdev in @ctx->bdev. @ctx->body is | |
739 | * set to point past the device node prefix. | |
740 | * | |
741 | * This function may be called multiple times on @ctx and the extra calls become | |
742 | * NOOPs. blkg_conf_prep() implicitly calls this function. Use this function | |
743 | * explicitly if bdev access is needed without resolving the blkcg / policy part | |
744 | * of @ctx->input. Returns -errno on error. | |
015d254c | 745 | */ |
faffaab2 | 746 | int blkg_conf_open_bdev(struct blkg_conf_ctx *ctx) |
015d254c | 747 | { |
faffaab2 | 748 | char *input = ctx->input; |
015d254c | 749 | unsigned int major, minor; |
22ae8ce8 CH |
750 | struct block_device *bdev; |
751 | int key_len; | |
015d254c | 752 | |
faffaab2 TH |
753 | if (ctx->bdev) |
754 | return 0; | |
755 | ||
015d254c | 756 | if (sscanf(input, "%u:%u%n", &major, &minor, &key_len) != 2) |
faffaab2 | 757 | return -EINVAL; |
015d254c TH |
758 | |
759 | input += key_len; | |
760 | if (!isspace(*input)) | |
faffaab2 | 761 | return -EINVAL; |
015d254c TH |
762 | input = skip_spaces(input); |
763 | ||
22ae8ce8 CH |
764 | bdev = blkdev_get_no_open(MKDEV(major, minor)); |
765 | if (!bdev) | |
faffaab2 | 766 | return -ENODEV; |
22ae8ce8 CH |
767 | if (bdev_is_partition(bdev)) { |
768 | blkdev_put_no_open(bdev); | |
faffaab2 | 769 | return -ENODEV; |
015d254c TH |
770 | } |
771 | ||
a13bd91b YK |
772 | mutex_lock(&bdev->bd_queue->rq_qos_mutex); |
773 | if (!disk_live(bdev->bd_disk)) { | |
774 | blkdev_put_no_open(bdev); | |
775 | mutex_unlock(&bdev->bd_queue->rq_qos_mutex); | |
776 | return -ENODEV; | |
777 | } | |
778 | ||
faffaab2 TH |
779 | ctx->body = input; |
780 | ctx->bdev = bdev; | |
781 | return 0; | |
015d254c TH |
782 | } |
783 | ||
3a8b31d3 TH |
784 | /** |
785 | * blkg_conf_prep - parse and prepare for per-blkg config update | |
786 | * @blkcg: target block cgroup | |
da8b0662 | 787 | * @pol: target policy |
faffaab2 TH |
788 | * @ctx: blkg_conf_ctx initialized with blkg_conf_init() |
789 | * | |
790 | * Parse per-blkg config update from @ctx->input and initialize @ctx | |
791 | * accordingly. On success, @ctx->body points to the part of @ctx->input | |
792 | * following MAJ:MIN, @ctx->bdev points to the target block device and | |
793 | * @ctx->blkg to the blkg being configured. | |
3a8b31d3 | 794 | * |
faffaab2 TH |
795 | * blkg_conf_open_bdev() may be called on @ctx beforehand. On success, this |
796 | * function returns with queue lock held and must be followed by | |
797 | * blkg_conf_exit(). | |
3a8b31d3 | 798 | */ |
3c798398 | 799 | int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol, |
faffaab2 | 800 | struct blkg_conf_ctx *ctx) |
83462a6c | 801 | __acquires(&bdev->bd_queue->queue_lock) |
34d0f179 | 802 | { |
99e60387 | 803 | struct gendisk *disk; |
457e490f | 804 | struct request_queue *q; |
3c798398 | 805 | struct blkcg_gq *blkg; |
015d254c | 806 | int ret; |
36aa9e5f | 807 | |
faffaab2 TH |
808 | ret = blkg_conf_open_bdev(ctx); |
809 | if (ret) | |
810 | return ret; | |
811 | ||
812 | disk = ctx->bdev->bd_disk; | |
99e60387 | 813 | q = disk->queue; |
da8b0662 | 814 | |
0c9d338c YK |
815 | /* |
816 | * blkcg_deactivate_policy() requires queue to be frozen, we can grab | |
817 | * q_usage_counter to prevent concurrent with blkcg_deactivate_policy(). | |
818 | */ | |
819 | ret = blk_queue_enter(q, 0); | |
820 | if (ret) | |
15c30104 | 821 | goto fail; |
0c9d338c | 822 | |
0d945c1f | 823 | spin_lock_irq(&q->queue_lock); |
e56da7e2 | 824 | |
f753526e CH |
825 | if (!blkcg_policy_enabled(q, pol)) { |
826 | ret = -EOPNOTSUPP; | |
457e490f TE |
827 | goto fail_unlock; |
828 | } | |
829 | ||
f753526e | 830 | blkg = blkg_lookup(blkcg, q); |
5765033c | 831 | if (blkg) |
457e490f TE |
832 | goto success; |
833 | ||
834 | /* | |
835 | * Create blkgs walking down from blkcg_root to @blkcg, so that all | |
836 | * non-root blkgs have access to their parents. | |
837 | */ | |
838 | while (true) { | |
839 | struct blkcg *pos = blkcg; | |
840 | struct blkcg *parent; | |
841 | struct blkcg_gq *new_blkg; | |
842 | ||
843 | parent = blkcg_parent(blkcg); | |
79fcc5be | 844 | while (parent && !blkg_lookup(parent, q)) { |
457e490f TE |
845 | pos = parent; |
846 | parent = blkcg_parent(parent); | |
847 | } | |
848 | ||
849 | /* Drop locks to do new blkg allocation with GFP_KERNEL. */ | |
0d945c1f | 850 | spin_unlock_irq(&q->queue_lock); |
457e490f | 851 | |
99e60387 | 852 | new_blkg = blkg_alloc(pos, disk, GFP_KERNEL); |
457e490f TE |
853 | if (unlikely(!new_blkg)) { |
854 | ret = -ENOMEM; | |
15c30104 | 855 | goto fail_exit_queue; |
7702e8f4 | 856 | } |
3a8b31d3 | 857 | |
f255c19b GKB |
858 | if (radix_tree_preload(GFP_KERNEL)) { |
859 | blkg_free(new_blkg); | |
860 | ret = -ENOMEM; | |
15c30104 | 861 | goto fail_exit_queue; |
f255c19b GKB |
862 | } |
863 | ||
0d945c1f | 864 | spin_lock_irq(&q->queue_lock); |
457e490f | 865 | |
f753526e | 866 | if (!blkcg_policy_enabled(q, pol)) { |
52abfcbd | 867 | blkg_free(new_blkg); |
f753526e | 868 | ret = -EOPNOTSUPP; |
f255c19b | 869 | goto fail_preloaded; |
457e490f TE |
870 | } |
871 | ||
f753526e | 872 | blkg = blkg_lookup(pos, q); |
457e490f TE |
873 | if (blkg) { |
874 | blkg_free(new_blkg); | |
875 | } else { | |
99e60387 | 876 | blkg = blkg_create(pos, disk, new_blkg); |
98d669b4 | 877 | if (IS_ERR(blkg)) { |
457e490f | 878 | ret = PTR_ERR(blkg); |
f255c19b | 879 | goto fail_preloaded; |
457e490f TE |
880 | } |
881 | } | |
882 | ||
f255c19b GKB |
883 | radix_tree_preload_end(); |
884 | ||
457e490f TE |
885 | if (pos == blkcg) |
886 | goto success; | |
887 | } | |
888 | success: | |
0c9d338c | 889 | blk_queue_exit(q); |
3a8b31d3 | 890 | ctx->blkg = blkg; |
726fa694 | 891 | return 0; |
457e490f | 892 | |
f255c19b GKB |
893 | fail_preloaded: |
894 | radix_tree_preload_end(); | |
457e490f | 895 | fail_unlock: |
0d945c1f | 896 | spin_unlock_irq(&q->queue_lock); |
15c30104 YK |
897 | fail_exit_queue: |
898 | blk_queue_exit(q); | |
457e490f | 899 | fail: |
457e490f TE |
900 | /* |
901 | * If queue was bypassing, we should retry. Do so after a | |
902 | * short msleep(). It isn't strictly necessary but queue | |
903 | * can be bypassing for some time and it's always nice to | |
904 | * avoid busy looping. | |
905 | */ | |
906 | if (ret == -EBUSY) { | |
907 | msleep(10); | |
908 | ret = restart_syscall(); | |
909 | } | |
910 | return ret; | |
34d0f179 | 911 | } |
89f3b6d6 | 912 | EXPORT_SYMBOL_GPL(blkg_conf_prep); |
34d0f179 | 913 | |
3a8b31d3 | 914 | /** |
faffaab2 TH |
915 | * blkg_conf_exit - clean up per-blkg config update |
916 | * @ctx: blkg_conf_ctx initialized with blkg_conf_init() | |
3a8b31d3 | 917 | * |
faffaab2 TH |
918 | * Clean up after per-blkg config update. This function must be called on all |
919 | * blkg_conf_ctx's initialized with blkg_conf_init(). | |
3a8b31d3 | 920 | */ |
faffaab2 | 921 | void blkg_conf_exit(struct blkg_conf_ctx *ctx) |
83462a6c | 922 | __releases(&ctx->bdev->bd_queue->queue_lock) |
a13bd91b | 923 | __releases(&ctx->bdev->bd_queue->rq_qos_mutex) |
34d0f179 | 924 | { |
faffaab2 TH |
925 | if (ctx->blkg) { |
926 | spin_unlock_irq(&bdev_get_queue(ctx->bdev)->queue_lock); | |
927 | ctx->blkg = NULL; | |
928 | } | |
929 | ||
930 | if (ctx->bdev) { | |
a13bd91b | 931 | mutex_unlock(&ctx->bdev->bd_queue->rq_qos_mutex); |
faffaab2 TH |
932 | blkdev_put_no_open(ctx->bdev); |
933 | ctx->body = NULL; | |
934 | ctx->bdev = NULL; | |
935 | } | |
34d0f179 | 936 | } |
faffaab2 | 937 | EXPORT_SYMBOL_GPL(blkg_conf_exit); |
34d0f179 | 938 | |
cd1fc4b9 BB |
939 | static void blkg_iostat_set(struct blkg_iostat *dst, struct blkg_iostat *src) |
940 | { | |
941 | int i; | |
942 | ||
943 | for (i = 0; i < BLKG_IOSTAT_NR; i++) { | |
944 | dst->bytes[i] = src->bytes[i]; | |
945 | dst->ios[i] = src->ios[i]; | |
946 | } | |
947 | } | |
948 | ||
949 | static void blkg_iostat_add(struct blkg_iostat *dst, struct blkg_iostat *src) | |
950 | { | |
951 | int i; | |
952 | ||
953 | for (i = 0; i < BLKG_IOSTAT_NR; i++) { | |
954 | dst->bytes[i] += src->bytes[i]; | |
955 | dst->ios[i] += src->ios[i]; | |
956 | } | |
957 | } | |
958 | ||
959 | static void blkg_iostat_sub(struct blkg_iostat *dst, struct blkg_iostat *src) | |
960 | { | |
961 | int i; | |
962 | ||
963 | for (i = 0; i < BLKG_IOSTAT_NR; i++) { | |
964 | dst->bytes[i] -= src->bytes[i]; | |
965 | dst->ios[i] -= src->ios[i]; | |
966 | } | |
967 | } | |
968 | ||
362b8c16 JY |
969 | static void blkcg_iostat_update(struct blkcg_gq *blkg, struct blkg_iostat *cur, |
970 | struct blkg_iostat *last) | |
971 | { | |
972 | struct blkg_iostat delta; | |
973 | unsigned long flags; | |
974 | ||
975 | /* propagate percpu delta to global */ | |
976 | flags = u64_stats_update_begin_irqsave(&blkg->iostat.sync); | |
977 | blkg_iostat_set(&delta, cur); | |
978 | blkg_iostat_sub(&delta, last); | |
979 | blkg_iostat_add(&blkg->iostat.cur, &delta); | |
980 | blkg_iostat_add(last, &delta); | |
981 | u64_stats_update_end_irqrestore(&blkg->iostat.sync, flags); | |
982 | } | |
983 | ||
20cb1c2f | 984 | static void __blkcg_rstat_flush(struct blkcg *blkcg, int cpu) |
cd1fc4b9 | 985 | { |
3b8cc629 WL |
986 | struct llist_head *lhead = per_cpu_ptr(blkcg->lhead, cpu); |
987 | struct llist_node *lnode; | |
988 | struct blkg_iostat_set *bisc, *next_bisc; | |
9c39b7a9 | 989 | unsigned long flags; |
cd1fc4b9 BB |
990 | |
991 | rcu_read_lock(); | |
992 | ||
3b8cc629 WL |
993 | lnode = llist_del_all(lhead); |
994 | if (!lnode) | |
995 | goto out; | |
996 | ||
20cb1c2f ML |
997 | /* |
998 | * For covering concurrent parent blkg update from blkg_release(). | |
999 | * | |
1000 | * When flushing from cgroup, cgroup_rstat_lock is always held, so | |
1001 | * this lock won't cause contention most of time. | |
1002 | */ | |
9c39b7a9 | 1003 | raw_spin_lock_irqsave(&blkg_stat_lock, flags); |
20cb1c2f | 1004 | |
3b8cc629 WL |
1005 | /* |
1006 | * Iterate only the iostat_cpu's queued in the lockless list. | |
1007 | */ | |
1008 | llist_for_each_entry_safe(bisc, next_bisc, lnode, lnode) { | |
1009 | struct blkcg_gq *blkg = bisc->blkg; | |
cd1fc4b9 | 1010 | struct blkcg_gq *parent = blkg->parent; |
362b8c16 | 1011 | struct blkg_iostat cur; |
cd1fc4b9 BB |
1012 | unsigned int seq; |
1013 | ||
3b8cc629 WL |
1014 | WRITE_ONCE(bisc->lqueued, false); |
1015 | ||
cd1fc4b9 BB |
1016 | /* fetch the current per-cpu values */ |
1017 | do { | |
1018 | seq = u64_stats_fetch_begin(&bisc->sync); | |
1019 | blkg_iostat_set(&cur, &bisc->cur); | |
1020 | } while (u64_stats_fetch_retry(&bisc->sync, seq)); | |
1021 | ||
362b8c16 | 1022 | blkcg_iostat_update(blkg, &cur, &bisc->last); |
cd1fc4b9 | 1023 | |
dc26532a | 1024 | /* propagate global delta to parent (unless that's root) */ |
362b8c16 JY |
1025 | if (parent && parent->parent) |
1026 | blkcg_iostat_update(parent, &blkg->iostat.cur, | |
1027 | &blkg->iostat.last); | |
cd1fc4b9 | 1028 | } |
9c39b7a9 | 1029 | raw_spin_unlock_irqrestore(&blkg_stat_lock, flags); |
3b8cc629 | 1030 | out: |
cd1fc4b9 BB |
1031 | rcu_read_unlock(); |
1032 | } | |
1033 | ||
20cb1c2f ML |
1034 | static void blkcg_rstat_flush(struct cgroup_subsys_state *css, int cpu) |
1035 | { | |
1036 | /* Root-level stats are sourced from system-wide IO stats */ | |
1037 | if (cgroup_parent(css->cgroup)) | |
1038 | __blkcg_rstat_flush(css_to_blkcg(css), cpu); | |
1039 | } | |
1040 | ||
ef45fe47 | 1041 | /* |
dc26532a JW |
1042 | * We source root cgroup stats from the system-wide stats to avoid |
1043 | * tracking the same information twice and incurring overhead when no | |
1044 | * cgroups are defined. For that reason, cgroup_rstat_flush in | |
1045 | * blkcg_print_stat does not actually fill out the iostat in the root | |
1046 | * cgroup's blkcg_gq. | |
ef45fe47 BB |
1047 | * |
1048 | * However, we would like to re-use the printing code between the root and | |
1049 | * non-root cgroups to the extent possible. For that reason, we simulate | |
1050 | * flushing the root cgroup's stats by explicitly filling in the iostat | |
1051 | * with disk level statistics. | |
1052 | */ | |
1053 | static void blkcg_fill_root_iostats(void) | |
1054 | { | |
1055 | struct class_dev_iter iter; | |
1056 | struct device *dev; | |
1057 | ||
1058 | class_dev_iter_init(&iter, &block_class, NULL, &disk_type); | |
1059 | while ((dev = class_dev_iter_next(&iter))) { | |
0d02129e | 1060 | struct block_device *bdev = dev_to_bdev(dev); |
928f6f00 | 1061 | struct blkcg_gq *blkg = bdev->bd_disk->queue->root_blkg; |
ef45fe47 BB |
1062 | struct blkg_iostat tmp; |
1063 | int cpu; | |
f122d103 | 1064 | unsigned long flags; |
ef45fe47 BB |
1065 | |
1066 | memset(&tmp, 0, sizeof(tmp)); | |
1067 | for_each_possible_cpu(cpu) { | |
1068 | struct disk_stats *cpu_dkstats; | |
1069 | ||
0d02129e | 1070 | cpu_dkstats = per_cpu_ptr(bdev->bd_stats, cpu); |
ef45fe47 BB |
1071 | tmp.ios[BLKG_IOSTAT_READ] += |
1072 | cpu_dkstats->ios[STAT_READ]; | |
1073 | tmp.ios[BLKG_IOSTAT_WRITE] += | |
1074 | cpu_dkstats->ios[STAT_WRITE]; | |
1075 | tmp.ios[BLKG_IOSTAT_DISCARD] += | |
1076 | cpu_dkstats->ios[STAT_DISCARD]; | |
1077 | // convert sectors to bytes | |
1078 | tmp.bytes[BLKG_IOSTAT_READ] += | |
1079 | cpu_dkstats->sectors[STAT_READ] << 9; | |
1080 | tmp.bytes[BLKG_IOSTAT_WRITE] += | |
1081 | cpu_dkstats->sectors[STAT_WRITE] << 9; | |
1082 | tmp.bytes[BLKG_IOSTAT_DISCARD] += | |
1083 | cpu_dkstats->sectors[STAT_DISCARD] << 9; | |
ef45fe47 | 1084 | } |
f122d103 CZ |
1085 | |
1086 | flags = u64_stats_update_begin_irqsave(&blkg->iostat.sync); | |
1087 | blkg_iostat_set(&blkg->iostat.cur, &tmp); | |
1088 | u64_stats_update_end_irqrestore(&blkg->iostat.sync, flags); | |
ef45fe47 BB |
1089 | } |
1090 | } | |
1091 | ||
49cb5168 | 1092 | static void blkcg_print_one_stat(struct blkcg_gq *blkg, struct seq_file *s) |
2ee867dc | 1093 | { |
49cb5168 CH |
1094 | struct blkg_iostat_set *bis = &blkg->iostat; |
1095 | u64 rbytes, wbytes, rios, wios, dbytes, dios; | |
49cb5168 CH |
1096 | const char *dname; |
1097 | unsigned seq; | |
49cb5168 | 1098 | int i; |
ef45fe47 | 1099 | |
49cb5168 CH |
1100 | if (!blkg->online) |
1101 | return; | |
2ee867dc | 1102 | |
49cb5168 CH |
1103 | dname = blkg_dev_name(blkg); |
1104 | if (!dname) | |
1105 | return; | |
2ee867dc | 1106 | |
252c651a | 1107 | seq_printf(s, "%s ", dname); |
b0814361 | 1108 | |
49cb5168 CH |
1109 | do { |
1110 | seq = u64_stats_fetch_begin(&bis->sync); | |
b0814361 | 1111 | |
49cb5168 CH |
1112 | rbytes = bis->cur.bytes[BLKG_IOSTAT_READ]; |
1113 | wbytes = bis->cur.bytes[BLKG_IOSTAT_WRITE]; | |
1114 | dbytes = bis->cur.bytes[BLKG_IOSTAT_DISCARD]; | |
1115 | rios = bis->cur.ios[BLKG_IOSTAT_READ]; | |
1116 | wios = bis->cur.ios[BLKG_IOSTAT_WRITE]; | |
1117 | dios = bis->cur.ios[BLKG_IOSTAT_DISCARD]; | |
1118 | } while (u64_stats_fetch_retry(&bis->sync, seq)); | |
2ee867dc | 1119 | |
49cb5168 | 1120 | if (rbytes || wbytes || rios || wios) { |
252c651a | 1121 | seq_printf(s, "rbytes=%llu wbytes=%llu rios=%llu wios=%llu dbytes=%llu dios=%llu", |
49cb5168 CH |
1122 | rbytes, wbytes, rios, wios, |
1123 | dbytes, dios); | |
1124 | } | |
903d23f0 | 1125 | |
49cb5168 | 1126 | if (blkcg_debug_stats && atomic_read(&blkg->use_delay)) { |
252c651a | 1127 | seq_printf(s, " use_delay=%d delay_nsec=%llu", |
49cb5168 CH |
1128 | atomic_read(&blkg->use_delay), |
1129 | atomic64_read(&blkg->delay_nsec)); | |
1130 | } | |
2ee867dc | 1131 | |
49cb5168 CH |
1132 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
1133 | struct blkcg_policy *pol = blkcg_policy[i]; | |
2ee867dc | 1134 | |
49cb5168 CH |
1135 | if (!blkg->pd[i] || !pol->pd_stat_fn) |
1136 | continue; | |
903d23f0 | 1137 | |
3607849d | 1138 | pol->pd_stat_fn(blkg->pd[i], s); |
49cb5168 | 1139 | } |
d09d8df3 | 1140 | |
3607849d | 1141 | seq_puts(s, "\n"); |
49cb5168 | 1142 | } |
903d23f0 | 1143 | |
49cb5168 CH |
1144 | static int blkcg_print_stat(struct seq_file *sf, void *v) |
1145 | { | |
1146 | struct blkcg *blkcg = css_to_blkcg(seq_css(sf)); | |
1147 | struct blkcg_gq *blkg; | |
903d23f0 | 1148 | |
49cb5168 CH |
1149 | if (!seq_css(sf)->parent) |
1150 | blkcg_fill_root_iostats(); | |
1151 | else | |
1152 | cgroup_rstat_flush(blkcg->css.cgroup); | |
07b0fdec | 1153 | |
49cb5168 CH |
1154 | rcu_read_lock(); |
1155 | hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) { | |
1156 | spin_lock_irq(&blkg->q->queue_lock); | |
1157 | blkcg_print_one_stat(blkg, sf); | |
b0814361 | 1158 | spin_unlock_irq(&blkg->q->queue_lock); |
2ee867dc | 1159 | } |
2ee867dc TH |
1160 | rcu_read_unlock(); |
1161 | return 0; | |
1162 | } | |
1163 | ||
e1f3b941 | 1164 | static struct cftype blkcg_files[] = { |
2ee867dc TH |
1165 | { |
1166 | .name = "stat", | |
1167 | .seq_show = blkcg_print_stat, | |
1168 | }, | |
1169 | { } /* terminate */ | |
1170 | }; | |
1171 | ||
e1f3b941 | 1172 | static struct cftype blkcg_legacy_files[] = { |
84c124da DS |
1173 | { |
1174 | .name = "reset_stats", | |
3c798398 | 1175 | .write_u64 = blkcg_reset_stats, |
22084190 | 1176 | }, |
4baf6e33 | 1177 | { } /* terminate */ |
31e4c28d VG |
1178 | }; |
1179 | ||
dec223c9 CH |
1180 | #ifdef CONFIG_CGROUP_WRITEBACK |
1181 | struct list_head *blkcg_get_cgwb_list(struct cgroup_subsys_state *css) | |
1182 | { | |
1183 | return &css_to_blkcg(css)->cgwb_list; | |
1184 | } | |
1185 | #endif | |
1186 | ||
59b57717 DZF |
1187 | /* |
1188 | * blkcg destruction is a three-stage process. | |
1189 | * | |
1190 | * 1. Destruction starts. The blkcg_css_offline() callback is invoked | |
1191 | * which offlines writeback. Here we tie the next stage of blkg destruction | |
1192 | * to the completion of writeback associated with the blkcg. This lets us | |
1193 | * avoid punting potentially large amounts of outstanding writeback to root | |
1194 | * while maintaining any ongoing policies. The next stage is triggered when | |
1195 | * the nr_cgwbs count goes to zero. | |
1196 | * | |
1197 | * 2. When the nr_cgwbs count goes to zero, blkcg_destroy_blkgs() is called | |
1198 | * and handles the destruction of blkgs. Here the css reference held by | |
1199 | * the blkg is put back eventually allowing blkcg_css_free() to be called. | |
1200 | * This work may occur in cgwb_release_workfn() on the cgwb_release | |
1201 | * workqueue. Any submitted ios that fail to get the blkg ref will be | |
1202 | * punted to the root_blkg. | |
1203 | * | |
1204 | * 3. Once the blkcg ref count goes to zero, blkcg_css_free() is called. | |
1205 | * This finally frees the blkcg. | |
1206 | */ | |
1207 | ||
59b57717 DZF |
1208 | /** |
1209 | * blkcg_destroy_blkgs - responsible for shooting down blkgs | |
1210 | * @blkcg: blkcg of interest | |
1211 | * | |
1212 | * blkgs should be removed while holding both q and blkcg locks. As blkcg lock | |
1213 | * is nested inside q lock, this function performs reverse double lock dancing. | |
1214 | * Destroying the blkgs releases the reference held on the blkcg's css allowing | |
1215 | * blkcg_css_free to eventually be called. | |
1216 | * | |
1217 | * This is the blkcg counterpart of ioc_release_fn(). | |
1218 | */ | |
397c9f46 | 1219 | static void blkcg_destroy_blkgs(struct blkcg *blkcg) |
59b57717 | 1220 | { |
6c635cae BW |
1221 | might_sleep(); |
1222 | ||
9f13ef67 | 1223 | spin_lock_irq(&blkcg->lock); |
7ee9c562 | 1224 | |
4c699480 JQ |
1225 | while (!hlist_empty(&blkcg->blkg_list)) { |
1226 | struct blkcg_gq *blkg = hlist_entry(blkcg->blkg_list.first, | |
6b065462 | 1227 | struct blkcg_gq, blkcg_node); |
4c699480 JQ |
1228 | struct request_queue *q = blkg->q; |
1229 | ||
6c635cae BW |
1230 | if (need_resched() || !spin_trylock(&q->queue_lock)) { |
1231 | /* | |
1232 | * Given that the system can accumulate a huge number | |
1233 | * of blkgs in pathological cases, check to see if we | |
1234 | * need to rescheduling to avoid softlockup. | |
1235 | */ | |
4c699480 | 1236 | spin_unlock_irq(&blkcg->lock); |
6c635cae | 1237 | cond_resched(); |
4c699480 | 1238 | spin_lock_irq(&blkcg->lock); |
6c635cae | 1239 | continue; |
4c699480 | 1240 | } |
6c635cae BW |
1241 | |
1242 | blkg_destroy(blkg); | |
1243 | spin_unlock(&q->queue_lock); | |
4c699480 | 1244 | } |
6b065462 | 1245 | |
4c699480 JQ |
1246 | spin_unlock_irq(&blkcg->lock); |
1247 | } | |
1248 | ||
397c9f46 CH |
1249 | /** |
1250 | * blkcg_pin_online - pin online state | |
1251 | * @blkcg_css: blkcg of interest | |
1252 | * | |
1253 | * While pinned, a blkcg is kept online. This is primarily used to | |
1254 | * impedance-match blkg and cgwb lifetimes so that blkg doesn't go offline | |
1255 | * while an associated cgwb is still active. | |
1256 | */ | |
1257 | void blkcg_pin_online(struct cgroup_subsys_state *blkcg_css) | |
1258 | { | |
1259 | refcount_inc(&css_to_blkcg(blkcg_css)->online_pin); | |
1260 | } | |
1261 | ||
1262 | /** | |
1263 | * blkcg_unpin_online - unpin online state | |
1264 | * @blkcg_css: blkcg of interest | |
1265 | * | |
1266 | * This is primarily used to impedance-match blkg and cgwb lifetimes so | |
1267 | * that blkg doesn't go offline while an associated cgwb is still active. | |
1268 | * When this count goes to zero, all active cgwbs have finished so the | |
1269 | * blkcg can continue destruction by calling blkcg_destroy_blkgs(). | |
1270 | */ | |
1271 | void blkcg_unpin_online(struct cgroup_subsys_state *blkcg_css) | |
1272 | { | |
1273 | struct blkcg *blkcg = css_to_blkcg(blkcg_css); | |
1274 | ||
1275 | do { | |
1276 | if (!refcount_dec_and_test(&blkcg->online_pin)) | |
1277 | break; | |
1278 | blkcg_destroy_blkgs(blkcg); | |
1279 | blkcg = blkcg_parent(blkcg); | |
1280 | } while (blkcg); | |
1281 | } | |
1282 | ||
1283 | /** | |
1284 | * blkcg_css_offline - cgroup css_offline callback | |
1285 | * @css: css of interest | |
1286 | * | |
1287 | * This function is called when @css is about to go away. Here the cgwbs are | |
1288 | * offlined first and only once writeback associated with the blkcg has | |
1289 | * finished do we start step 2 (see above). | |
1290 | */ | |
1291 | static void blkcg_css_offline(struct cgroup_subsys_state *css) | |
1292 | { | |
1293 | /* this prevents anyone from attaching or migrating to this blkcg */ | |
dec223c9 | 1294 | wb_blkcg_offline(css); |
397c9f46 CH |
1295 | |
1296 | /* put the base online pin allowing step 2 to be triggered */ | |
1297 | blkcg_unpin_online(css); | |
1298 | } | |
1299 | ||
eb95419b | 1300 | static void blkcg_css_free(struct cgroup_subsys_state *css) |
7ee9c562 | 1301 | { |
eb95419b | 1302 | struct blkcg *blkcg = css_to_blkcg(css); |
bc915e61 | 1303 | int i; |
7ee9c562 | 1304 | |
7876f930 | 1305 | mutex_lock(&blkcg_pol_mutex); |
e4a9bde9 | 1306 | |
7876f930 | 1307 | list_del(&blkcg->all_blkcgs_node); |
7876f930 | 1308 | |
bc915e61 | 1309 | for (i = 0; i < BLKCG_MAX_POLS; i++) |
e4a9bde9 TH |
1310 | if (blkcg->cpd[i]) |
1311 | blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]); | |
1312 | ||
1313 | mutex_unlock(&blkcg_pol_mutex); | |
1314 | ||
3b8cc629 | 1315 | free_percpu(blkcg->lhead); |
bc915e61 | 1316 | kfree(blkcg); |
31e4c28d VG |
1317 | } |
1318 | ||
eb95419b TH |
1319 | static struct cgroup_subsys_state * |
1320 | blkcg_css_alloc(struct cgroup_subsys_state *parent_css) | |
31e4c28d | 1321 | { |
3c798398 | 1322 | struct blkcg *blkcg; |
e48453c3 | 1323 | int i; |
31e4c28d | 1324 | |
7876f930 TH |
1325 | mutex_lock(&blkcg_pol_mutex); |
1326 | ||
eb95419b | 1327 | if (!parent_css) { |
3c798398 | 1328 | blkcg = &blkcg_root; |
bc915e61 TH |
1329 | } else { |
1330 | blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL); | |
b5a9adcb | 1331 | if (!blkcg) |
4c18c9e9 | 1332 | goto unlock; |
e48453c3 AA |
1333 | } |
1334 | ||
3b8cc629 WL |
1335 | if (init_blkcg_llists(blkcg)) |
1336 | goto free_blkcg; | |
1337 | ||
e48453c3 AA |
1338 | for (i = 0; i < BLKCG_MAX_POLS ; i++) { |
1339 | struct blkcg_policy *pol = blkcg_policy[i]; | |
1340 | struct blkcg_policy_data *cpd; | |
1341 | ||
1342 | /* | |
1343 | * If the policy hasn't been attached yet, wait for it | |
1344 | * to be attached before doing anything else. Otherwise, | |
1345 | * check if the policy requires any specific per-cgroup | |
1346 | * data: if it does, allocate and initialize it. | |
1347 | */ | |
e4a9bde9 | 1348 | if (!pol || !pol->cpd_alloc_fn) |
e48453c3 AA |
1349 | continue; |
1350 | ||
e4a9bde9 | 1351 | cpd = pol->cpd_alloc_fn(GFP_KERNEL); |
b5a9adcb | 1352 | if (!cpd) |
e48453c3 | 1353 | goto free_pd_blkcg; |
b5a9adcb | 1354 | |
81437648 TH |
1355 | blkcg->cpd[i] = cpd; |
1356 | cpd->blkcg = blkcg; | |
e48453c3 | 1357 | cpd->plid = i; |
e48453c3 | 1358 | } |
31e4c28d | 1359 | |
31e4c28d | 1360 | spin_lock_init(&blkcg->lock); |
d866dbf6 | 1361 | refcount_set(&blkcg->online_pin, 1); |
e00f4f4d | 1362 | INIT_RADIX_TREE(&blkcg->blkg_tree, GFP_NOWAIT | __GFP_NOWARN); |
31e4c28d | 1363 | INIT_HLIST_HEAD(&blkcg->blkg_list); |
52ebea74 TH |
1364 | #ifdef CONFIG_CGROUP_WRITEBACK |
1365 | INIT_LIST_HEAD(&blkcg->cgwb_list); | |
1366 | #endif | |
7876f930 TH |
1367 | list_add_tail(&blkcg->all_blkcgs_node, &all_blkcgs); |
1368 | ||
1369 | mutex_unlock(&blkcg_pol_mutex); | |
31e4c28d | 1370 | return &blkcg->css; |
e48453c3 AA |
1371 | |
1372 | free_pd_blkcg: | |
1373 | for (i--; i >= 0; i--) | |
e4a9bde9 TH |
1374 | if (blkcg->cpd[i]) |
1375 | blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]); | |
3b8cc629 WL |
1376 | free_percpu(blkcg->lhead); |
1377 | free_blkcg: | |
4c18c9e9 | 1378 | if (blkcg != &blkcg_root) |
1379 | kfree(blkcg); | |
1380 | unlock: | |
7876f930 | 1381 | mutex_unlock(&blkcg_pol_mutex); |
b5a9adcb | 1382 | return ERR_PTR(-ENOMEM); |
31e4c28d VG |
1383 | } |
1384 | ||
4308a434 TH |
1385 | static int blkcg_css_online(struct cgroup_subsys_state *css) |
1386 | { | |
397c9f46 | 1387 | struct blkcg *parent = blkcg_parent(css_to_blkcg(css)); |
4308a434 TH |
1388 | |
1389 | /* | |
1390 | * blkcg_pin_online() is used to delay blkcg offline so that blkgs | |
1391 | * don't go offline while cgwbs are still active on them. Pin the | |
1392 | * parent so that offline always happens towards the root. | |
1393 | */ | |
1394 | if (parent) | |
d7dbd43f | 1395 | blkcg_pin_online(&parent->css); |
4308a434 TH |
1396 | return 0; |
1397 | } | |
1398 | ||
9823538f | 1399 | int blkcg_init_disk(struct gendisk *disk) |
5efd6113 | 1400 | { |
9823538f | 1401 | struct request_queue *q = disk->queue; |
d708f0d5 JA |
1402 | struct blkcg_gq *new_blkg, *blkg; |
1403 | bool preloaded; | |
ec13b1d6 TH |
1404 | int ret; |
1405 | ||
472e4314 | 1406 | INIT_LIST_HEAD(&q->blkg_list); |
1231039d | 1407 | mutex_init(&q->blkcg_mutex); |
472e4314 | 1408 | |
99e60387 | 1409 | new_blkg = blkg_alloc(&blkcg_root, disk, GFP_KERNEL); |
d708f0d5 JA |
1410 | if (!new_blkg) |
1411 | return -ENOMEM; | |
1412 | ||
1413 | preloaded = !radix_tree_preload(GFP_KERNEL); | |
1414 | ||
bea54883 | 1415 | /* Make sure the root blkg exists. */ |
77c570a1 | 1416 | /* spin_lock_irq can serve as RCU read-side critical section. */ |
0d945c1f | 1417 | spin_lock_irq(&q->queue_lock); |
99e60387 | 1418 | blkg = blkg_create(&blkcg_root, disk, new_blkg); |
901932a3 JB |
1419 | if (IS_ERR(blkg)) |
1420 | goto err_unlock; | |
1421 | q->root_blkg = blkg; | |
0d945c1f | 1422 | spin_unlock_irq(&q->queue_lock); |
ec13b1d6 | 1423 | |
d708f0d5 JA |
1424 | if (preloaded) |
1425 | radix_tree_preload_end(); | |
1426 | ||
b0dde3f5 | 1427 | ret = blk_ioprio_init(disk); |
556910e3 BVA |
1428 | if (ret) |
1429 | goto err_destroy_all; | |
1430 | ||
e13793ba | 1431 | ret = blk_throtl_init(disk); |
19688d7f | 1432 | if (ret) |
33dc6279 | 1433 | goto err_ioprio_exit; |
19688d7f | 1434 | |
04be60b5 | 1435 | return 0; |
901932a3 | 1436 | |
33dc6279 | 1437 | err_ioprio_exit: |
b0dde3f5 | 1438 | blk_ioprio_exit(disk); |
04be60b5 | 1439 | err_destroy_all: |
00ad6991 | 1440 | blkg_destroy_all(disk); |
04be60b5 | 1441 | return ret; |
901932a3 | 1442 | err_unlock: |
0d945c1f | 1443 | spin_unlock_irq(&q->queue_lock); |
901932a3 JB |
1444 | if (preloaded) |
1445 | radix_tree_preload_end(); | |
1446 | return PTR_ERR(blkg); | |
5efd6113 TH |
1447 | } |
1448 | ||
9823538f | 1449 | void blkcg_exit_disk(struct gendisk *disk) |
5efd6113 | 1450 | { |
00ad6991 | 1451 | blkg_destroy_all(disk); |
e13793ba | 1452 | blk_throtl_exit(disk); |
5efd6113 TH |
1453 | } |
1454 | ||
d09d8df3 JB |
1455 | static void blkcg_exit(struct task_struct *tsk) |
1456 | { | |
f05837ed CH |
1457 | if (tsk->throttle_disk) |
1458 | put_disk(tsk->throttle_disk); | |
1459 | tsk->throttle_disk = NULL; | |
d09d8df3 JB |
1460 | } |
1461 | ||
c165b3e3 | 1462 | struct cgroup_subsys io_cgrp_subsys = { |
92fb9748 | 1463 | .css_alloc = blkcg_css_alloc, |
4308a434 | 1464 | .css_online = blkcg_css_online, |
92fb9748 TH |
1465 | .css_offline = blkcg_css_offline, |
1466 | .css_free = blkcg_css_free, | |
f7331648 | 1467 | .css_rstat_flush = blkcg_rstat_flush, |
2ee867dc | 1468 | .dfl_cftypes = blkcg_files, |
880f50e2 | 1469 | .legacy_cftypes = blkcg_legacy_files, |
c165b3e3 | 1470 | .legacy_name = "blkio", |
d09d8df3 | 1471 | .exit = blkcg_exit, |
1ced953b TH |
1472 | #ifdef CONFIG_MEMCG |
1473 | /* | |
1474 | * This ensures that, if available, memcg is automatically enabled | |
1475 | * together on the default hierarchy so that the owner cgroup can | |
1476 | * be retrieved from writeback pages. | |
1477 | */ | |
1478 | .depends_on = 1 << memory_cgrp_id, | |
1479 | #endif | |
676f7c8f | 1480 | }; |
c165b3e3 | 1481 | EXPORT_SYMBOL_GPL(io_cgrp_subsys); |
676f7c8f | 1482 | |
a2b1693b | 1483 | /** |
40e4996e CH |
1484 | * blkcg_activate_policy - activate a blkcg policy on a gendisk |
1485 | * @disk: gendisk of interest | |
a2b1693b TH |
1486 | * @pol: blkcg policy to activate |
1487 | * | |
40e4996e | 1488 | * Activate @pol on @disk. Requires %GFP_KERNEL context. @disk goes through |
a2b1693b TH |
1489 | * bypass mode to populate its blkgs with policy_data for @pol. |
1490 | * | |
40e4996e | 1491 | * Activation happens with @disk bypassed, so nobody would be accessing blkgs |
a2b1693b TH |
1492 | * from IO path. Update of each blkg is protected by both queue and blkcg |
1493 | * locks so that holding either lock and testing blkcg_policy_enabled() is | |
1494 | * always enough for dereferencing policy data. | |
1495 | * | |
1496 | * The caller is responsible for synchronizing [de]activations and policy | |
1497 | * [un]registerations. Returns 0 on success, -errno on failure. | |
1498 | */ | |
40e4996e | 1499 | int blkcg_activate_policy(struct gendisk *disk, const struct blkcg_policy *pol) |
a2b1693b | 1500 | { |
40e4996e | 1501 | struct request_queue *q = disk->queue; |
4c55f4f9 | 1502 | struct blkg_policy_data *pd_prealloc = NULL; |
9d179b86 | 1503 | struct blkcg_gq *blkg, *pinned_blkg = NULL; |
4c55f4f9 | 1504 | int ret; |
a2b1693b TH |
1505 | |
1506 | if (blkcg_policy_enabled(q, pol)) | |
1507 | return 0; | |
1508 | ||
344e9ffc | 1509 | if (queue_is_mq(q)) |
bd166ef1 | 1510 | blk_mq_freeze_queue(q); |
9d179b86 | 1511 | retry: |
0d945c1f | 1512 | spin_lock_irq(&q->queue_lock); |
a2b1693b | 1513 | |
ec14a87e | 1514 | /* blkg_list is pushed at the head, reverse walk to initialize parents first */ |
71c81407 | 1515 | list_for_each_entry_reverse(blkg, &q->blkg_list, q_node) { |
4c55f4f9 TH |
1516 | struct blkg_policy_data *pd; |
1517 | ||
1518 | if (blkg->pd[pol->plid]) | |
1519 | continue; | |
a2b1693b | 1520 | |
9d179b86 TH |
1521 | /* If prealloc matches, use it; otherwise try GFP_NOWAIT */ |
1522 | if (blkg == pinned_blkg) { | |
1523 | pd = pd_prealloc; | |
1524 | pd_prealloc = NULL; | |
1525 | } else { | |
0a0b4f79 CH |
1526 | pd = pol->pd_alloc_fn(disk, blkg->blkcg, |
1527 | GFP_NOWAIT | __GFP_NOWARN); | |
9d179b86 TH |
1528 | } |
1529 | ||
4c55f4f9 | 1530 | if (!pd) { |
9d179b86 TH |
1531 | /* |
1532 | * GFP_NOWAIT failed. Free the existing one and | |
1533 | * prealloc for @blkg w/ GFP_KERNEL. | |
1534 | */ | |
1535 | if (pinned_blkg) | |
1536 | blkg_put(pinned_blkg); | |
1537 | blkg_get(blkg); | |
1538 | pinned_blkg = blkg; | |
1539 | ||
0d945c1f | 1540 | spin_unlock_irq(&q->queue_lock); |
9d179b86 TH |
1541 | |
1542 | if (pd_prealloc) | |
1543 | pol->pd_free_fn(pd_prealloc); | |
0a0b4f79 CH |
1544 | pd_prealloc = pol->pd_alloc_fn(disk, blkg->blkcg, |
1545 | GFP_KERNEL); | |
9d179b86 TH |
1546 | if (pd_prealloc) |
1547 | goto retry; | |
1548 | else | |
1549 | goto enomem; | |
4c55f4f9 | 1550 | } |
a2b1693b | 1551 | |
ec14a87e TH |
1552 | spin_lock(&blkg->blkcg->lock); |
1553 | ||
a2b1693b | 1554 | pd->blkg = blkg; |
b276a876 | 1555 | pd->plid = pol->plid; |
ec14a87e | 1556 | blkg->pd[pol->plid] = pd; |
a2b1693b | 1557 | |
ec14a87e TH |
1558 | if (pol->pd_init_fn) |
1559 | pol->pd_init_fn(pd); | |
9d179b86 | 1560 | |
dfd6200a | 1561 | if (pol->pd_online_fn) |
ec14a87e TH |
1562 | pol->pd_online_fn(pd); |
1563 | pd->online = true; | |
1564 | ||
1565 | spin_unlock(&blkg->blkcg->lock); | |
dfd6200a | 1566 | } |
e3ff8887 | 1567 | |
a2b1693b TH |
1568 | __set_bit(pol->plid, q->blkcg_pols); |
1569 | ret = 0; | |
4c55f4f9 | 1570 | |
0d945c1f | 1571 | spin_unlock_irq(&q->queue_lock); |
9d179b86 | 1572 | out: |
344e9ffc | 1573 | if (queue_is_mq(q)) |
bd166ef1 | 1574 | blk_mq_unfreeze_queue(q); |
9d179b86 TH |
1575 | if (pinned_blkg) |
1576 | blkg_put(pinned_blkg); | |
001bea73 TH |
1577 | if (pd_prealloc) |
1578 | pol->pd_free_fn(pd_prealloc); | |
a2b1693b | 1579 | return ret; |
9d179b86 TH |
1580 | |
1581 | enomem: | |
ec14a87e | 1582 | /* alloc failed, take down everything */ |
9d179b86 TH |
1583 | spin_lock_irq(&q->queue_lock); |
1584 | list_for_each_entry(blkg, &q->blkg_list, q_node) { | |
858560b2 | 1585 | struct blkcg *blkcg = blkg->blkcg; |
ec14a87e | 1586 | struct blkg_policy_data *pd; |
858560b2 LJ |
1587 | |
1588 | spin_lock(&blkcg->lock); | |
ec14a87e TH |
1589 | pd = blkg->pd[pol->plid]; |
1590 | if (pd) { | |
1591 | if (pd->online && pol->pd_offline_fn) | |
1592 | pol->pd_offline_fn(pd); | |
1593 | pd->online = false; | |
1594 | pol->pd_free_fn(pd); | |
9d179b86 TH |
1595 | blkg->pd[pol->plid] = NULL; |
1596 | } | |
858560b2 | 1597 | spin_unlock(&blkcg->lock); |
9d179b86 TH |
1598 | } |
1599 | spin_unlock_irq(&q->queue_lock); | |
1600 | ret = -ENOMEM; | |
1601 | goto out; | |
a2b1693b TH |
1602 | } |
1603 | EXPORT_SYMBOL_GPL(blkcg_activate_policy); | |
1604 | ||
1605 | /** | |
40e4996e CH |
1606 | * blkcg_deactivate_policy - deactivate a blkcg policy on a gendisk |
1607 | * @disk: gendisk of interest | |
a2b1693b TH |
1608 | * @pol: blkcg policy to deactivate |
1609 | * | |
40e4996e | 1610 | * Deactivate @pol on @disk. Follows the same synchronization rules as |
a2b1693b TH |
1611 | * blkcg_activate_policy(). |
1612 | */ | |
40e4996e | 1613 | void blkcg_deactivate_policy(struct gendisk *disk, |
3c798398 | 1614 | const struct blkcg_policy *pol) |
a2b1693b | 1615 | { |
40e4996e | 1616 | struct request_queue *q = disk->queue; |
3c798398 | 1617 | struct blkcg_gq *blkg; |
a2b1693b TH |
1618 | |
1619 | if (!blkcg_policy_enabled(q, pol)) | |
1620 | return; | |
1621 | ||
344e9ffc | 1622 | if (queue_is_mq(q)) |
bd166ef1 | 1623 | blk_mq_freeze_queue(q); |
bd166ef1 | 1624 | |
1231039d | 1625 | mutex_lock(&q->blkcg_mutex); |
0d945c1f | 1626 | spin_lock_irq(&q->queue_lock); |
a2b1693b TH |
1627 | |
1628 | __clear_bit(pol->plid, q->blkcg_pols); | |
1629 | ||
1630 | list_for_each_entry(blkg, &q->blkg_list, q_node) { | |
858560b2 LJ |
1631 | struct blkcg *blkcg = blkg->blkcg; |
1632 | ||
1633 | spin_lock(&blkcg->lock); | |
001bea73 | 1634 | if (blkg->pd[pol->plid]) { |
dfd6200a | 1635 | if (blkg->pd[pol->plid]->online && pol->pd_offline_fn) |
a9520cd6 | 1636 | pol->pd_offline_fn(blkg->pd[pol->plid]); |
001bea73 TH |
1637 | pol->pd_free_fn(blkg->pd[pol->plid]); |
1638 | blkg->pd[pol->plid] = NULL; | |
1639 | } | |
858560b2 | 1640 | spin_unlock(&blkcg->lock); |
a2b1693b TH |
1641 | } |
1642 | ||
0d945c1f | 1643 | spin_unlock_irq(&q->queue_lock); |
1231039d | 1644 | mutex_unlock(&q->blkcg_mutex); |
bd166ef1 | 1645 | |
344e9ffc | 1646 | if (queue_is_mq(q)) |
bd166ef1 | 1647 | blk_mq_unfreeze_queue(q); |
a2b1693b TH |
1648 | } |
1649 | EXPORT_SYMBOL_GPL(blkcg_deactivate_policy); | |
1650 | ||
e55cf798 JY |
1651 | static void blkcg_free_all_cpd(struct blkcg_policy *pol) |
1652 | { | |
1653 | struct blkcg *blkcg; | |
1654 | ||
1655 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) { | |
1656 | if (blkcg->cpd[pol->plid]) { | |
1657 | pol->cpd_free_fn(blkcg->cpd[pol->plid]); | |
1658 | blkcg->cpd[pol->plid] = NULL; | |
1659 | } | |
1660 | } | |
1661 | } | |
1662 | ||
8bd435b3 | 1663 | /** |
3c798398 TH |
1664 | * blkcg_policy_register - register a blkcg policy |
1665 | * @pol: blkcg policy to register | |
8bd435b3 | 1666 | * |
3c798398 TH |
1667 | * Register @pol with blkcg core. Might sleep and @pol may be modified on |
1668 | * successful registration. Returns 0 on success and -errno on failure. | |
8bd435b3 | 1669 | */ |
d5bf0291 | 1670 | int blkcg_policy_register(struct blkcg_policy *pol) |
3e252066 | 1671 | { |
06b285bd | 1672 | struct blkcg *blkcg; |
8bd435b3 | 1673 | int i, ret; |
e8989fae | 1674 | |
838f13bf | 1675 | mutex_lock(&blkcg_pol_register_mutex); |
bc0d6501 TH |
1676 | mutex_lock(&blkcg_pol_mutex); |
1677 | ||
8bd435b3 TH |
1678 | /* find an empty slot */ |
1679 | ret = -ENOSPC; | |
1680 | for (i = 0; i < BLKCG_MAX_POLS; i++) | |
3c798398 | 1681 | if (!blkcg_policy[i]) |
8bd435b3 | 1682 | break; |
01c5f85a JA |
1683 | if (i >= BLKCG_MAX_POLS) { |
1684 | pr_warn("blkcg_policy_register: BLKCG_MAX_POLS too small\n"); | |
838f13bf | 1685 | goto err_unlock; |
01c5f85a | 1686 | } |
035d10b2 | 1687 | |
e8401073 | 1688 | /* Make sure cpd/pd_alloc_fn and cpd/pd_free_fn in pairs */ |
1689 | if ((!pol->cpd_alloc_fn ^ !pol->cpd_free_fn) || | |
1690 | (!pol->pd_alloc_fn ^ !pol->pd_free_fn)) | |
1691 | goto err_unlock; | |
1692 | ||
06b285bd | 1693 | /* register @pol */ |
3c798398 | 1694 | pol->plid = i; |
06b285bd TH |
1695 | blkcg_policy[pol->plid] = pol; |
1696 | ||
1697 | /* allocate and install cpd's */ | |
e4a9bde9 | 1698 | if (pol->cpd_alloc_fn) { |
06b285bd TH |
1699 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) { |
1700 | struct blkcg_policy_data *cpd; | |
1701 | ||
e4a9bde9 | 1702 | cpd = pol->cpd_alloc_fn(GFP_KERNEL); |
bbb427e3 | 1703 | if (!cpd) |
06b285bd | 1704 | goto err_free_cpds; |
06b285bd | 1705 | |
81437648 TH |
1706 | blkcg->cpd[pol->plid] = cpd; |
1707 | cpd->blkcg = blkcg; | |
06b285bd | 1708 | cpd->plid = pol->plid; |
06b285bd TH |
1709 | } |
1710 | } | |
1711 | ||
838f13bf | 1712 | mutex_unlock(&blkcg_pol_mutex); |
8bd435b3 | 1713 | |
8bd435b3 | 1714 | /* everything is in place, add intf files for the new policy */ |
2ee867dc TH |
1715 | if (pol->dfl_cftypes) |
1716 | WARN_ON(cgroup_add_dfl_cftypes(&io_cgrp_subsys, | |
1717 | pol->dfl_cftypes)); | |
880f50e2 | 1718 | if (pol->legacy_cftypes) |
c165b3e3 | 1719 | WARN_ON(cgroup_add_legacy_cftypes(&io_cgrp_subsys, |
880f50e2 | 1720 | pol->legacy_cftypes)); |
838f13bf TH |
1721 | mutex_unlock(&blkcg_pol_register_mutex); |
1722 | return 0; | |
1723 | ||
06b285bd | 1724 | err_free_cpds: |
e55cf798 JY |
1725 | if (pol->cpd_free_fn) |
1726 | blkcg_free_all_cpd(pol); | |
1727 | ||
06b285bd | 1728 | blkcg_policy[pol->plid] = NULL; |
838f13bf | 1729 | err_unlock: |
bc0d6501 | 1730 | mutex_unlock(&blkcg_pol_mutex); |
838f13bf | 1731 | mutex_unlock(&blkcg_pol_register_mutex); |
8bd435b3 | 1732 | return ret; |
3e252066 | 1733 | } |
3c798398 | 1734 | EXPORT_SYMBOL_GPL(blkcg_policy_register); |
3e252066 | 1735 | |
8bd435b3 | 1736 | /** |
3c798398 TH |
1737 | * blkcg_policy_unregister - unregister a blkcg policy |
1738 | * @pol: blkcg policy to unregister | |
8bd435b3 | 1739 | * |
3c798398 | 1740 | * Undo blkcg_policy_register(@pol). Might sleep. |
8bd435b3 | 1741 | */ |
3c798398 | 1742 | void blkcg_policy_unregister(struct blkcg_policy *pol) |
3e252066 | 1743 | { |
838f13bf | 1744 | mutex_lock(&blkcg_pol_register_mutex); |
bc0d6501 | 1745 | |
3c798398 | 1746 | if (WARN_ON(blkcg_policy[pol->plid] != pol)) |
8bd435b3 TH |
1747 | goto out_unlock; |
1748 | ||
1749 | /* kill the intf files first */ | |
2ee867dc TH |
1750 | if (pol->dfl_cftypes) |
1751 | cgroup_rm_cftypes(pol->dfl_cftypes); | |
880f50e2 TH |
1752 | if (pol->legacy_cftypes) |
1753 | cgroup_rm_cftypes(pol->legacy_cftypes); | |
44ea53de | 1754 | |
06b285bd | 1755 | /* remove cpds and unregister */ |
838f13bf | 1756 | mutex_lock(&blkcg_pol_mutex); |
06b285bd | 1757 | |
e55cf798 JY |
1758 | if (pol->cpd_free_fn) |
1759 | blkcg_free_all_cpd(pol); | |
1760 | ||
3c798398 | 1761 | blkcg_policy[pol->plid] = NULL; |
06b285bd | 1762 | |
bc0d6501 | 1763 | mutex_unlock(&blkcg_pol_mutex); |
838f13bf TH |
1764 | out_unlock: |
1765 | mutex_unlock(&blkcg_pol_register_mutex); | |
3e252066 | 1766 | } |
3c798398 | 1767 | EXPORT_SYMBOL_GPL(blkcg_policy_unregister); |
903d23f0 | 1768 | |
d09d8df3 JB |
1769 | /* |
1770 | * Scale the accumulated delay based on how long it has been since we updated | |
1771 | * the delay. We only call this when we are adding delay, in case it's been a | |
1772 | * while since we added delay, and when we are checking to see if we need to | |
1773 | * delay a task, to account for any delays that may have occurred. | |
1774 | */ | |
1775 | static void blkcg_scale_delay(struct blkcg_gq *blkg, u64 now) | |
1776 | { | |
1777 | u64 old = atomic64_read(&blkg->delay_start); | |
1778 | ||
54c52e10 TH |
1779 | /* negative use_delay means no scaling, see blkcg_set_delay() */ |
1780 | if (atomic_read(&blkg->use_delay) < 0) | |
1781 | return; | |
1782 | ||
d09d8df3 JB |
1783 | /* |
1784 | * We only want to scale down every second. The idea here is that we | |
1785 | * want to delay people for min(delay_nsec, NSEC_PER_SEC) in a certain | |
1786 | * time window. We only want to throttle tasks for recent delay that | |
1787 | * has occurred, in 1 second time windows since that's the maximum | |
1788 | * things can be throttled. We save the current delay window in | |
1789 | * blkg->last_delay so we know what amount is still left to be charged | |
1790 | * to the blkg from this point onward. blkg->last_use keeps track of | |
1791 | * the use_delay counter. The idea is if we're unthrottling the blkg we | |
1792 | * are ok with whatever is happening now, and we can take away more of | |
1793 | * the accumulated delay as we've already throttled enough that | |
1794 | * everybody is happy with their IO latencies. | |
1795 | */ | |
1796 | if (time_before64(old + NSEC_PER_SEC, now) && | |
96388f57 | 1797 | atomic64_try_cmpxchg(&blkg->delay_start, &old, now)) { |
d09d8df3 JB |
1798 | u64 cur = atomic64_read(&blkg->delay_nsec); |
1799 | u64 sub = min_t(u64, blkg->last_delay, now - old); | |
1800 | int cur_use = atomic_read(&blkg->use_delay); | |
1801 | ||
1802 | /* | |
1803 | * We've been unthrottled, subtract a larger chunk of our | |
1804 | * accumulated delay. | |
1805 | */ | |
1806 | if (cur_use < blkg->last_use) | |
1807 | sub = max_t(u64, sub, blkg->last_delay >> 1); | |
1808 | ||
1809 | /* | |
1810 | * This shouldn't happen, but handle it anyway. Our delay_nsec | |
1811 | * should only ever be growing except here where we subtract out | |
1812 | * min(last_delay, 1 second), but lord knows bugs happen and I'd | |
1813 | * rather not end up with negative numbers. | |
1814 | */ | |
1815 | if (unlikely(cur < sub)) { | |
1816 | atomic64_set(&blkg->delay_nsec, 0); | |
1817 | blkg->last_delay = 0; | |
1818 | } else { | |
1819 | atomic64_sub(sub, &blkg->delay_nsec); | |
1820 | blkg->last_delay = cur - sub; | |
1821 | } | |
1822 | blkg->last_use = cur_use; | |
1823 | } | |
1824 | } | |
1825 | ||
1826 | /* | |
1827 | * This is called when we want to actually walk up the hierarchy and check to | |
1828 | * see if we need to throttle, and then actually throttle if there is some | |
1829 | * accumulated delay. This should only be called upon return to user space so | |
1830 | * we're not holding some lock that would induce a priority inversion. | |
1831 | */ | |
1832 | static void blkcg_maybe_throttle_blkg(struct blkcg_gq *blkg, bool use_memdelay) | |
1833 | { | |
fd112c74 | 1834 | unsigned long pflags; |
5160a5a5 | 1835 | bool clamp; |
d09d8df3 JB |
1836 | u64 now = ktime_to_ns(ktime_get()); |
1837 | u64 exp; | |
1838 | u64 delay_nsec = 0; | |
1839 | int tok; | |
1840 | ||
1841 | while (blkg->parent) { | |
5160a5a5 TH |
1842 | int use_delay = atomic_read(&blkg->use_delay); |
1843 | ||
1844 | if (use_delay) { | |
1845 | u64 this_delay; | |
1846 | ||
d09d8df3 | 1847 | blkcg_scale_delay(blkg, now); |
5160a5a5 TH |
1848 | this_delay = atomic64_read(&blkg->delay_nsec); |
1849 | if (this_delay > delay_nsec) { | |
1850 | delay_nsec = this_delay; | |
1851 | clamp = use_delay > 0; | |
1852 | } | |
d09d8df3 JB |
1853 | } |
1854 | blkg = blkg->parent; | |
1855 | } | |
1856 | ||
1857 | if (!delay_nsec) | |
1858 | return; | |
1859 | ||
1860 | /* | |
1861 | * Let's not sleep for all eternity if we've amassed a huge delay. | |
1862 | * Swapping or metadata IO can accumulate 10's of seconds worth of | |
1863 | * delay, and we want userspace to be able to do _something_ so cap the | |
5160a5a5 TH |
1864 | * delays at 0.25s. If there's 10's of seconds worth of delay then the |
1865 | * tasks will be delayed for 0.25 second for every syscall. If | |
1866 | * blkcg_set_delay() was used as indicated by negative use_delay, the | |
1867 | * caller is responsible for regulating the range. | |
d09d8df3 | 1868 | */ |
5160a5a5 TH |
1869 | if (clamp) |
1870 | delay_nsec = min_t(u64, delay_nsec, 250 * NSEC_PER_MSEC); | |
d09d8df3 | 1871 | |
fd112c74 JB |
1872 | if (use_memdelay) |
1873 | psi_memstall_enter(&pflags); | |
d09d8df3 JB |
1874 | |
1875 | exp = ktime_add_ns(now, delay_nsec); | |
1876 | tok = io_schedule_prepare(); | |
1877 | do { | |
1878 | __set_current_state(TASK_KILLABLE); | |
1879 | if (!schedule_hrtimeout(&exp, HRTIMER_MODE_ABS)) | |
1880 | break; | |
1881 | } while (!fatal_signal_pending(current)); | |
1882 | io_schedule_finish(tok); | |
fd112c74 JB |
1883 | |
1884 | if (use_memdelay) | |
1885 | psi_memstall_leave(&pflags); | |
d09d8df3 JB |
1886 | } |
1887 | ||
1888 | /** | |
1889 | * blkcg_maybe_throttle_current - throttle the current task if it has been marked | |
1890 | * | |
1891 | * This is only called if we've been marked with set_notify_resume(). Obviously | |
1892 | * we can be set_notify_resume() for reasons other than blkcg throttling, so we | |
f05837ed | 1893 | * check to see if current->throttle_disk is set and if not this doesn't do |
d09d8df3 JB |
1894 | * anything. This should only ever be called by the resume code, it's not meant |
1895 | * to be called by people willy-nilly as it will actually do the work to | |
1896 | * throttle the task if it is setup for throttling. | |
1897 | */ | |
1898 | void blkcg_maybe_throttle_current(void) | |
1899 | { | |
f05837ed | 1900 | struct gendisk *disk = current->throttle_disk; |
d09d8df3 JB |
1901 | struct blkcg *blkcg; |
1902 | struct blkcg_gq *blkg; | |
1903 | bool use_memdelay = current->use_memdelay; | |
1904 | ||
f05837ed | 1905 | if (!disk) |
d09d8df3 JB |
1906 | return; |
1907 | ||
f05837ed | 1908 | current->throttle_disk = NULL; |
d09d8df3 JB |
1909 | current->use_memdelay = false; |
1910 | ||
1911 | rcu_read_lock(); | |
82778259 | 1912 | blkcg = css_to_blkcg(blkcg_css()); |
d09d8df3 JB |
1913 | if (!blkcg) |
1914 | goto out; | |
9a9c261e | 1915 | blkg = blkg_lookup(blkcg, disk->queue); |
d09d8df3 JB |
1916 | if (!blkg) |
1917 | goto out; | |
7754f669 | 1918 | if (!blkg_tryget(blkg)) |
d09d8df3 JB |
1919 | goto out; |
1920 | rcu_read_unlock(); | |
d09d8df3 JB |
1921 | |
1922 | blkcg_maybe_throttle_blkg(blkg, use_memdelay); | |
1923 | blkg_put(blkg); | |
f05837ed | 1924 | put_disk(disk); |
d09d8df3 JB |
1925 | return; |
1926 | out: | |
1927 | rcu_read_unlock(); | |
d09d8df3 | 1928 | } |
d09d8df3 JB |
1929 | |
1930 | /** | |
1931 | * blkcg_schedule_throttle - this task needs to check for throttling | |
1d6df9d3 | 1932 | * @disk: disk to throttle |
537d71b3 | 1933 | * @use_memdelay: do we charge this to memory delay for PSI |
d09d8df3 JB |
1934 | * |
1935 | * This is called by the IO controller when we know there's delay accumulated | |
1936 | * for the blkg for this task. We do not pass the blkg because there are places | |
1937 | * we call this that may not have that information, the swapping code for | |
de185b56 | 1938 | * instance will only have a block_device at that point. This set's the |
d09d8df3 JB |
1939 | * notify_resume for the task to check and see if it requires throttling before |
1940 | * returning to user space. | |
1941 | * | |
1942 | * We will only schedule once per syscall. You can call this over and over | |
1943 | * again and it will only do the check once upon return to user space, and only | |
1944 | * throttle once. If the task needs to be throttled again it'll need to be | |
1945 | * re-set at the next time we see the task. | |
1946 | */ | |
de185b56 | 1947 | void blkcg_schedule_throttle(struct gendisk *disk, bool use_memdelay) |
d09d8df3 JB |
1948 | { |
1949 | if (unlikely(current->flags & PF_KTHREAD)) | |
1950 | return; | |
1951 | ||
f05837ed CH |
1952 | if (current->throttle_disk != disk) { |
1953 | if (test_bit(GD_DEAD, &disk->state)) | |
49d1822b | 1954 | return; |
f05837ed | 1955 | get_device(disk_to_dev(disk)); |
49d1822b | 1956 | |
f05837ed CH |
1957 | if (current->throttle_disk) |
1958 | put_disk(current->throttle_disk); | |
1959 | current->throttle_disk = disk; | |
49d1822b | 1960 | } |
d09d8df3 | 1961 | |
d09d8df3 JB |
1962 | if (use_memdelay) |
1963 | current->use_memdelay = use_memdelay; | |
1964 | set_notify_resume(current); | |
1965 | } | |
d09d8df3 JB |
1966 | |
1967 | /** | |
1968 | * blkcg_add_delay - add delay to this blkg | |
537d71b3 BVA |
1969 | * @blkg: blkg of interest |
1970 | * @now: the current time in nanoseconds | |
1971 | * @delta: how many nanoseconds of delay to add | |
d09d8df3 JB |
1972 | * |
1973 | * Charge @delta to the blkg's current delay accumulation. This is used to | |
1974 | * throttle tasks if an IO controller thinks we need more throttling. | |
1975 | */ | |
1976 | void blkcg_add_delay(struct blkcg_gq *blkg, u64 now, u64 delta) | |
1977 | { | |
54c52e10 TH |
1978 | if (WARN_ON_ONCE(atomic_read(&blkg->use_delay) < 0)) |
1979 | return; | |
d09d8df3 JB |
1980 | blkcg_scale_delay(blkg, now); |
1981 | atomic64_add(delta, &blkg->delay_nsec); | |
1982 | } | |
d09d8df3 | 1983 | |
28fc591f CH |
1984 | /** |
1985 | * blkg_tryget_closest - try and get a blkg ref on the closet blkg | |
13c7863d CH |
1986 | * @bio: target bio |
1987 | * @css: target css | |
28fc591f | 1988 | * |
13c7863d CH |
1989 | * As the failure mode here is to walk up the blkg tree, this ensure that the |
1990 | * blkg->parent pointers are always valid. This returns the blkg that it ended | |
1991 | * up taking a reference on or %NULL if no reference was taken. | |
28fc591f | 1992 | */ |
13c7863d CH |
1993 | static inline struct blkcg_gq *blkg_tryget_closest(struct bio *bio, |
1994 | struct cgroup_subsys_state *css) | |
28fc591f | 1995 | { |
13c7863d | 1996 | struct blkcg_gq *blkg, *ret_blkg = NULL; |
28fc591f | 1997 | |
13c7863d | 1998 | rcu_read_lock(); |
99e60387 | 1999 | blkg = blkg_lookup_create(css_to_blkcg(css), bio->bi_bdev->bd_disk); |
28fc591f CH |
2000 | while (blkg) { |
2001 | if (blkg_tryget(blkg)) { | |
2002 | ret_blkg = blkg; | |
2003 | break; | |
2004 | } | |
2005 | blkg = blkg->parent; | |
2006 | } | |
13c7863d | 2007 | rcu_read_unlock(); |
28fc591f CH |
2008 | |
2009 | return ret_blkg; | |
2010 | } | |
2011 | ||
2012 | /** | |
2013 | * bio_associate_blkg_from_css - associate a bio with a specified css | |
2014 | * @bio: target bio | |
2015 | * @css: target css | |
2016 | * | |
2017 | * Associate @bio with the blkg found by combining the css's blkg and the | |
2018 | * request_queue of the @bio. An association failure is handled by walking up | |
2019 | * the blkg tree. Therefore, the blkg associated can be anything between @blkg | |
2020 | * and q->root_blkg. This situation only happens when a cgroup is dying and | |
2021 | * then the remaining bios will spill to the closest alive blkg. | |
2022 | * | |
2023 | * A reference will be taken on the blkg and will be released when @bio is | |
2024 | * freed. | |
2025 | */ | |
2026 | void bio_associate_blkg_from_css(struct bio *bio, | |
2027 | struct cgroup_subsys_state *css) | |
2028 | { | |
28fc591f CH |
2029 | if (bio->bi_blkg) |
2030 | blkg_put(bio->bi_blkg); | |
2031 | ||
a5b97526 | 2032 | if (css && css->parent) { |
13c7863d | 2033 | bio->bi_blkg = blkg_tryget_closest(bio, css); |
a5b97526 | 2034 | } else { |
ed6cddef PB |
2035 | blkg_get(bdev_get_queue(bio->bi_bdev)->root_blkg); |
2036 | bio->bi_blkg = bdev_get_queue(bio->bi_bdev)->root_blkg; | |
a5b97526 | 2037 | } |
28fc591f CH |
2038 | } |
2039 | EXPORT_SYMBOL_GPL(bio_associate_blkg_from_css); | |
2040 | ||
2041 | /** | |
2042 | * bio_associate_blkg - associate a bio with a blkg | |
2043 | * @bio: target bio | |
2044 | * | |
2045 | * Associate @bio with the blkg found from the bio's css and request_queue. | |
2046 | * If one is not found, bio_lookup_blkg() creates the blkg. If a blkg is | |
2047 | * already associated, the css is reused and association redone as the | |
2048 | * request_queue may have changed. | |
2049 | */ | |
2050 | void bio_associate_blkg(struct bio *bio) | |
2051 | { | |
2052 | struct cgroup_subsys_state *css; | |
2053 | ||
2054 | rcu_read_lock(); | |
2055 | ||
2056 | if (bio->bi_blkg) | |
bbb1ebe7 | 2057 | css = bio_blkcg_css(bio); |
28fc591f CH |
2058 | else |
2059 | css = blkcg_css(); | |
2060 | ||
2061 | bio_associate_blkg_from_css(bio, css); | |
2062 | ||
2063 | rcu_read_unlock(); | |
2064 | } | |
2065 | EXPORT_SYMBOL_GPL(bio_associate_blkg); | |
2066 | ||
2067 | /** | |
2068 | * bio_clone_blkg_association - clone blkg association from src to dst bio | |
2069 | * @dst: destination bio | |
2070 | * @src: source bio | |
2071 | */ | |
2072 | void bio_clone_blkg_association(struct bio *dst, struct bio *src) | |
2073 | { | |
22b106e5 JK |
2074 | if (src->bi_blkg) |
2075 | bio_associate_blkg_from_css(dst, bio_blkcg_css(src)); | |
28fc591f CH |
2076 | } |
2077 | EXPORT_SYMBOL_GPL(bio_clone_blkg_association); | |
2078 | ||
db18a53e CH |
2079 | static int blk_cgroup_io_type(struct bio *bio) |
2080 | { | |
2081 | if (op_is_discard(bio->bi_opf)) | |
2082 | return BLKG_IOSTAT_DISCARD; | |
2083 | if (op_is_write(bio->bi_opf)) | |
2084 | return BLKG_IOSTAT_WRITE; | |
2085 | return BLKG_IOSTAT_READ; | |
2086 | } | |
2087 | ||
2088 | void blk_cgroup_bio_start(struct bio *bio) | |
2089 | { | |
3b8cc629 | 2090 | struct blkcg *blkcg = bio->bi_blkg->blkcg; |
db18a53e CH |
2091 | int rwd = blk_cgroup_io_type(bio), cpu; |
2092 | struct blkg_iostat_set *bis; | |
3c08b093 | 2093 | unsigned long flags; |
db18a53e | 2094 | |
ad7c3b41 JH |
2095 | if (!cgroup_subsys_on_dfl(io_cgrp_subsys)) |
2096 | return; | |
2097 | ||
0416f3be ML |
2098 | /* Root-level stats are sourced from system-wide IO stats */ |
2099 | if (!cgroup_parent(blkcg->css.cgroup)) | |
2100 | return; | |
2101 | ||
db18a53e CH |
2102 | cpu = get_cpu(); |
2103 | bis = per_cpu_ptr(bio->bi_blkg->iostat_cpu, cpu); | |
3c08b093 | 2104 | flags = u64_stats_update_begin_irqsave(&bis->sync); |
db18a53e CH |
2105 | |
2106 | /* | |
2107 | * If the bio is flagged with BIO_CGROUP_ACCT it means this is a split | |
2108 | * bio and we would have already accounted for the size of the bio. | |
2109 | */ | |
2110 | if (!bio_flagged(bio, BIO_CGROUP_ACCT)) { | |
2111 | bio_set_flag(bio, BIO_CGROUP_ACCT); | |
0b8cc25d | 2112 | bis->cur.bytes[rwd] += bio->bi_iter.bi_size; |
db18a53e CH |
2113 | } |
2114 | bis->cur.ios[rwd]++; | |
2115 | ||
3b8cc629 WL |
2116 | /* |
2117 | * If the iostat_cpu isn't in a lockless list, put it into the | |
2118 | * list to indicate that a stat update is pending. | |
2119 | */ | |
2120 | if (!READ_ONCE(bis->lqueued)) { | |
2121 | struct llist_head *lhead = this_cpu_ptr(blkcg->lhead); | |
2122 | ||
2123 | llist_add(&bis->lnode, lhead); | |
2124 | WRITE_ONCE(bis->lqueued, true); | |
3b8cc629 WL |
2125 | } |
2126 | ||
3c08b093 | 2127 | u64_stats_update_end_irqrestore(&bis->sync, flags); |
ad7c3b41 | 2128 | cgroup_rstat_updated(blkcg->css.cgroup, cpu); |
db18a53e CH |
2129 | put_cpu(); |
2130 | } | |
2131 | ||
216889aa CH |
2132 | bool blk_cgroup_congested(void) |
2133 | { | |
2134 | struct cgroup_subsys_state *css; | |
2135 | bool ret = false; | |
2136 | ||
2137 | rcu_read_lock(); | |
d200ca14 | 2138 | for (css = blkcg_css(); css; css = css->parent) { |
216889aa CH |
2139 | if (atomic_read(&css->cgroup->congestion_count)) { |
2140 | ret = true; | |
2141 | break; | |
2142 | } | |
216889aa CH |
2143 | } |
2144 | rcu_read_unlock(); | |
2145 | return ret; | |
2146 | } | |
2147 | ||
903d23f0 JB |
2148 | module_param(blkcg_debug_stats, bool, 0644); |
2149 | MODULE_PARM_DESC(blkcg_debug_stats, "True if you want debug stats, false if not"); |