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 | |
838f13bf TH |
37 | /* |
38 | * blkcg_pol_mutex protects blkcg_policy[] and policy [de]activation. | |
39 | * blkcg_pol_register_mutex nests outside of it and synchronizes entire | |
40 | * policy [un]register operations including cgroup file additions / | |
41 | * removals. Putting cgroup file registration outside blkcg_pol_mutex | |
42 | * allows grabbing it from cgroup callbacks. | |
43 | */ | |
44 | static DEFINE_MUTEX(blkcg_pol_register_mutex); | |
bc0d6501 | 45 | static DEFINE_MUTEX(blkcg_pol_mutex); |
923adde1 | 46 | |
e48453c3 | 47 | struct blkcg blkcg_root; |
3c798398 | 48 | EXPORT_SYMBOL_GPL(blkcg_root); |
9d6a986c | 49 | |
496d5e75 | 50 | struct cgroup_subsys_state * const blkcg_root_css = &blkcg_root.css; |
9b0eb69b | 51 | EXPORT_SYMBOL_GPL(blkcg_root_css); |
496d5e75 | 52 | |
3c798398 | 53 | static struct blkcg_policy *blkcg_policy[BLKCG_MAX_POLS]; |
035d10b2 | 54 | |
7876f930 TH |
55 | static LIST_HEAD(all_blkcgs); /* protected by blkcg_pol_mutex */ |
56 | ||
07b0fdec | 57 | bool blkcg_debug_stats = false; |
d3f77dfd | 58 | static struct workqueue_struct *blkcg_punt_bio_wq; |
903d23f0 | 59 | |
a731763f YK |
60 | #define BLKG_DESTROY_BATCH_SIZE 64 |
61 | ||
3b8cc629 WL |
62 | /* |
63 | * Lockless lists for tracking IO stats update | |
64 | * | |
65 | * New IO stats are stored in the percpu iostat_cpu within blkcg_gq (blkg). | |
66 | * There are multiple blkg's (one for each block device) attached to each | |
67 | * blkcg. The rstat code keeps track of which cpu has IO stats updated, | |
68 | * but it doesn't know which blkg has the updated stats. If there are many | |
69 | * block devices in a system, the cost of iterating all the blkg's to flush | |
70 | * out the IO stats can be high. To reduce such overhead, a set of percpu | |
71 | * lockless lists (lhead) per blkcg are used to track the set of recently | |
72 | * updated iostat_cpu's since the last flush. An iostat_cpu will be put | |
73 | * onto the lockless list on the update side [blk_cgroup_bio_start()] if | |
74 | * not there yet and then removed when being flushed [blkcg_rstat_flush()]. | |
75 | * References to blkg are gotten and then put back in the process to | |
76 | * protect against blkg removal. | |
77 | * | |
78 | * Return: 0 if successful or -ENOMEM if allocation fails. | |
79 | */ | |
80 | static int init_blkcg_llists(struct blkcg *blkcg) | |
81 | { | |
82 | int cpu; | |
83 | ||
84 | blkcg->lhead = alloc_percpu_gfp(struct llist_head, GFP_KERNEL); | |
85 | if (!blkcg->lhead) | |
86 | return -ENOMEM; | |
87 | ||
88 | for_each_possible_cpu(cpu) | |
89 | init_llist_head(per_cpu_ptr(blkcg->lhead, cpu)); | |
90 | return 0; | |
91 | } | |
92 | ||
bc5fee91 CH |
93 | /** |
94 | * blkcg_css - find the current css | |
95 | * | |
96 | * Find the css associated with either the kthread or the current task. | |
97 | * This may return a dying css, so it is up to the caller to use tryget logic | |
98 | * to confirm it is alive and well. | |
99 | */ | |
100 | static struct cgroup_subsys_state *blkcg_css(void) | |
101 | { | |
102 | struct cgroup_subsys_state *css; | |
103 | ||
104 | css = kthread_blkcg(); | |
105 | if (css) | |
106 | return css; | |
107 | return task_css(current, io_cgrp_id); | |
108 | } | |
109 | ||
a2b1693b | 110 | static bool blkcg_policy_enabled(struct request_queue *q, |
3c798398 | 111 | const struct blkcg_policy *pol) |
a2b1693b TH |
112 | { |
113 | return pol && test_bit(pol->plid, q->blkcg_pols); | |
114 | } | |
115 | ||
d578c770 | 116 | static void blkg_free_workfn(struct work_struct *work) |
0381411e | 117 | { |
d578c770 ML |
118 | struct blkcg_gq *blkg = container_of(work, struct blkcg_gq, |
119 | free_work); | |
a06377c5 | 120 | struct request_queue *q = blkg->q; |
e8989fae | 121 | int i; |
549d3aa8 | 122 | |
f1c006f1 YK |
123 | /* |
124 | * pd_free_fn() can also be called from blkcg_deactivate_policy(), | |
125 | * in order to make sure pd_free_fn() is called in order, the deletion | |
1231039d | 126 | * of the list blkg->q_node is delayed to here from blkg_destroy(), and |
f1c006f1 YK |
127 | * blkcg_mutex is used to synchronize blkg_free_workfn() and |
128 | * blkcg_deactivate_policy(). | |
129 | */ | |
a06377c5 | 130 | mutex_lock(&q->blkcg_mutex); |
db613670 | 131 | for (i = 0; i < BLKCG_MAX_POLS; i++) |
001bea73 TH |
132 | if (blkg->pd[i]) |
133 | blkcg_policy[i]->pd_free_fn(blkg->pd[i]); | |
c7241bab YK |
134 | if (blkg->parent) |
135 | blkg_put(blkg->parent); | |
1231039d | 136 | list_del_init(&blkg->q_node); |
a06377c5 | 137 | mutex_unlock(&q->blkcg_mutex); |
e8989fae | 138 | |
a06377c5 | 139 | blk_put_queue(q); |
f7331648 | 140 | free_percpu(blkg->iostat_cpu); |
ef069b97 | 141 | percpu_ref_exit(&blkg->refcnt); |
549d3aa8 | 142 | kfree(blkg); |
0381411e TH |
143 | } |
144 | ||
d578c770 ML |
145 | /** |
146 | * blkg_free - free a blkg | |
147 | * @blkg: blkg to free | |
148 | * | |
149 | * Free @blkg which may be partially allocated. | |
150 | */ | |
151 | static void blkg_free(struct blkcg_gq *blkg) | |
152 | { | |
153 | if (!blkg) | |
154 | return; | |
155 | ||
156 | /* | |
157 | * Both ->pd_free_fn() and request queue's release handler may | |
158 | * sleep, so free us by scheduling one work func | |
159 | */ | |
160 | INIT_WORK(&blkg->free_work, blkg_free_workfn); | |
161 | schedule_work(&blkg->free_work); | |
162 | } | |
163 | ||
7fcf2b03 DZ |
164 | static void __blkg_release(struct rcu_head *rcu) |
165 | { | |
166 | struct blkcg_gq *blkg = container_of(rcu, struct blkcg_gq, rcu_head); | |
167 | ||
d3f77dfd TH |
168 | WARN_ON(!bio_list_empty(&blkg->async_bios)); |
169 | ||
7fcf2b03 DZ |
170 | /* release the blkcg and parent blkg refs this blkg has been holding */ |
171 | css_put(&blkg->blkcg->css); | |
7fcf2b03 DZ |
172 | blkg_free(blkg); |
173 | } | |
174 | ||
175 | /* | |
176 | * A group is RCU protected, but having an rcu lock does not mean that one | |
177 | * can access all the fields of blkg and assume these are valid. For | |
178 | * example, don't try to follow throtl_data and request queue links. | |
179 | * | |
180 | * Having a reference to blkg under an rcu allows accesses to only values | |
181 | * local to groups like group stats and group rate limits. | |
182 | */ | |
183 | static void blkg_release(struct percpu_ref *ref) | |
184 | { | |
185 | struct blkcg_gq *blkg = container_of(ref, struct blkcg_gq, refcnt); | |
186 | ||
187 | call_rcu(&blkg->rcu_head, __blkg_release); | |
188 | } | |
189 | ||
d3f77dfd TH |
190 | static void blkg_async_bio_workfn(struct work_struct *work) |
191 | { | |
192 | struct blkcg_gq *blkg = container_of(work, struct blkcg_gq, | |
193 | async_bio_work); | |
194 | struct bio_list bios = BIO_EMPTY_LIST; | |
195 | struct bio *bio; | |
192f1c6b XT |
196 | struct blk_plug plug; |
197 | bool need_plug = false; | |
d3f77dfd TH |
198 | |
199 | /* as long as there are pending bios, @blkg can't go away */ | |
200 | spin_lock_bh(&blkg->async_bio_lock); | |
201 | bio_list_merge(&bios, &blkg->async_bios); | |
202 | bio_list_init(&blkg->async_bios); | |
203 | spin_unlock_bh(&blkg->async_bio_lock); | |
204 | ||
192f1c6b XT |
205 | /* start plug only when bio_list contains at least 2 bios */ |
206 | if (bios.head && bios.head->bi_next) { | |
207 | need_plug = true; | |
208 | blk_start_plug(&plug); | |
209 | } | |
d3f77dfd TH |
210 | while ((bio = bio_list_pop(&bios))) |
211 | submit_bio(bio); | |
192f1c6b XT |
212 | if (need_plug) |
213 | blk_finish_plug(&plug); | |
d3f77dfd TH |
214 | } |
215 | ||
bbb1ebe7 CH |
216 | /** |
217 | * bio_blkcg_css - return the blkcg CSS associated with a bio | |
218 | * @bio: target bio | |
219 | * | |
220 | * This returns the CSS for the blkcg associated with a bio, or %NULL if not | |
221 | * associated. Callers are expected to either handle %NULL or know association | |
222 | * has been done prior to calling this. | |
223 | */ | |
224 | struct cgroup_subsys_state *bio_blkcg_css(struct bio *bio) | |
225 | { | |
226 | if (!bio || !bio->bi_blkg) | |
227 | return NULL; | |
228 | return &bio->bi_blkg->blkcg->css; | |
229 | } | |
230 | EXPORT_SYMBOL_GPL(bio_blkcg_css); | |
231 | ||
397c9f46 CH |
232 | /** |
233 | * blkcg_parent - get the parent of a blkcg | |
234 | * @blkcg: blkcg of interest | |
235 | * | |
236 | * Return the parent blkcg of @blkcg. Can be called anytime. | |
237 | */ | |
238 | static inline struct blkcg *blkcg_parent(struct blkcg *blkcg) | |
239 | { | |
240 | return css_to_blkcg(blkcg->css.parent); | |
241 | } | |
242 | ||
0381411e TH |
243 | /** |
244 | * blkg_alloc - allocate a blkg | |
245 | * @blkcg: block cgroup the new blkg is associated with | |
99e60387 | 246 | * @disk: gendisk the new blkg is associated with |
15974993 | 247 | * @gfp_mask: allocation mask to use |
0381411e | 248 | * |
e8989fae | 249 | * Allocate a new blkg assocating @blkcg and @q. |
0381411e | 250 | */ |
99e60387 | 251 | static struct blkcg_gq *blkg_alloc(struct blkcg *blkcg, struct gendisk *disk, |
15974993 | 252 | gfp_t gfp_mask) |
0381411e | 253 | { |
3c798398 | 254 | struct blkcg_gq *blkg; |
f7331648 | 255 | int i, cpu; |
0381411e TH |
256 | |
257 | /* alloc and init base part */ | |
99e60387 | 258 | blkg = kzalloc_node(sizeof(*blkg), gfp_mask, disk->queue->node); |
0381411e TH |
259 | if (!blkg) |
260 | return NULL; | |
ef069b97 | 261 | if (percpu_ref_init(&blkg->refcnt, blkg_release, 0, gfp_mask)) |
0b6f93bd | 262 | goto out_free_blkg; |
f7331648 TH |
263 | blkg->iostat_cpu = alloc_percpu_gfp(struct blkg_iostat_set, gfp_mask); |
264 | if (!blkg->iostat_cpu) | |
0b6f93bd | 265 | goto out_exit_refcnt; |
99e60387 | 266 | if (!blk_get_queue(disk->queue)) |
0b6f93bd | 267 | goto out_free_iostat; |
0a9a25ca | 268 | |
99e60387 | 269 | blkg->q = disk->queue; |
e8989fae | 270 | INIT_LIST_HEAD(&blkg->q_node); |
d3f77dfd TH |
271 | spin_lock_init(&blkg->async_bio_lock); |
272 | bio_list_init(&blkg->async_bios); | |
273 | INIT_WORK(&blkg->async_bio_work, blkg_async_bio_workfn); | |
0381411e | 274 | blkg->blkcg = blkcg; |
0381411e | 275 | |
f7331648 | 276 | u64_stats_init(&blkg->iostat.sync); |
3b8cc629 | 277 | for_each_possible_cpu(cpu) { |
f7331648 | 278 | u64_stats_init(&per_cpu_ptr(blkg->iostat_cpu, cpu)->sync); |
3b8cc629 WL |
279 | per_cpu_ptr(blkg->iostat_cpu, cpu)->blkg = blkg; |
280 | } | |
f7331648 | 281 | |
8bd435b3 | 282 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
3c798398 | 283 | struct blkcg_policy *pol = blkcg_policy[i]; |
e8989fae | 284 | struct blkg_policy_data *pd; |
0381411e | 285 | |
99e60387 | 286 | if (!blkcg_policy_enabled(disk->queue, pol)) |
e8989fae TH |
287 | continue; |
288 | ||
289 | /* alloc per-policy data and attach it to blkg */ | |
0a0b4f79 | 290 | pd = pol->pd_alloc_fn(disk, blkcg, gfp_mask); |
a051661c | 291 | if (!pd) |
0b6f93bd | 292 | goto out_free_pds; |
e8989fae TH |
293 | blkg->pd[i] = pd; |
294 | pd->blkg = blkg; | |
b276a876 | 295 | pd->plid = i; |
dfd6200a | 296 | pd->online = false; |
e8989fae TH |
297 | } |
298 | ||
0381411e | 299 | return blkg; |
a051661c | 300 | |
0b6f93bd CH |
301 | out_free_pds: |
302 | while (--i >= 0) | |
303 | if (blkg->pd[i]) | |
304 | blkcg_policy[i]->pd_free_fn(blkg->pd[i]); | |
a06377c5 | 305 | blk_put_queue(disk->queue); |
0b6f93bd CH |
306 | out_free_iostat: |
307 | free_percpu(blkg->iostat_cpu); | |
308 | out_exit_refcnt: | |
309 | percpu_ref_exit(&blkg->refcnt); | |
310 | out_free_blkg: | |
311 | kfree(blkg); | |
a051661c | 312 | return NULL; |
0381411e TH |
313 | } |
314 | ||
15974993 | 315 | /* |
d708f0d5 JA |
316 | * If @new_blkg is %NULL, this function tries to allocate a new one as |
317 | * necessary using %GFP_NOWAIT. @new_blkg is always consumed on return. | |
15974993 | 318 | */ |
99e60387 | 319 | static struct blkcg_gq *blkg_create(struct blkcg *blkcg, struct gendisk *disk, |
d708f0d5 | 320 | struct blkcg_gq *new_blkg) |
5624a4e4 | 321 | { |
d708f0d5 | 322 | struct blkcg_gq *blkg; |
f427d909 | 323 | int i, ret; |
5624a4e4 | 324 | |
99e60387 | 325 | lockdep_assert_held(&disk->queue->queue_lock); |
cd1604fa | 326 | |
0273ac34 | 327 | /* request_queue is dying, do not create/recreate a blkg */ |
99e60387 | 328 | if (blk_queue_dying(disk->queue)) { |
0273ac34 DZ |
329 | ret = -ENODEV; |
330 | goto err_free_blkg; | |
331 | } | |
332 | ||
7ee9c562 | 333 | /* blkg holds a reference to blkcg */ |
ec903c0c | 334 | if (!css_tryget_online(&blkcg->css)) { |
20386ce0 | 335 | ret = -ENODEV; |
93e6d5d8 | 336 | goto err_free_blkg; |
15974993 | 337 | } |
cd1604fa | 338 | |
d708f0d5 JA |
339 | /* allocate */ |
340 | if (!new_blkg) { | |
99e60387 | 341 | new_blkg = blkg_alloc(blkcg, disk, GFP_NOWAIT | __GFP_NOWARN); |
d708f0d5 JA |
342 | if (unlikely(!new_blkg)) { |
343 | ret = -ENOMEM; | |
8c911f3d | 344 | goto err_put_css; |
15974993 TH |
345 | } |
346 | } | |
d708f0d5 | 347 | blkg = new_blkg; |
cd1604fa | 348 | |
db613670 | 349 | /* link parent */ |
3c547865 | 350 | if (blkcg_parent(blkcg)) { |
99e60387 | 351 | blkg->parent = blkg_lookup(blkcg_parent(blkcg), disk->queue); |
3c547865 | 352 | if (WARN_ON_ONCE(!blkg->parent)) { |
20386ce0 | 353 | ret = -ENODEV; |
8c911f3d | 354 | goto err_put_css; |
3c547865 TH |
355 | } |
356 | blkg_get(blkg->parent); | |
357 | } | |
358 | ||
db613670 TH |
359 | /* invoke per-policy init */ |
360 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
361 | struct blkcg_policy *pol = blkcg_policy[i]; | |
362 | ||
363 | if (blkg->pd[i] && pol->pd_init_fn) | |
a9520cd6 | 364 | pol->pd_init_fn(blkg->pd[i]); |
db613670 TH |
365 | } |
366 | ||
367 | /* insert */ | |
cd1604fa | 368 | spin_lock(&blkcg->lock); |
99e60387 | 369 | ret = radix_tree_insert(&blkcg->blkg_tree, disk->queue->id, blkg); |
a637120e TH |
370 | if (likely(!ret)) { |
371 | hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list); | |
99e60387 | 372 | list_add(&blkg->q_node, &disk->queue->blkg_list); |
f427d909 TH |
373 | |
374 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
375 | struct blkcg_policy *pol = blkcg_policy[i]; | |
376 | ||
dfd6200a YK |
377 | if (blkg->pd[i]) { |
378 | if (pol->pd_online_fn) | |
379 | pol->pd_online_fn(blkg->pd[i]); | |
380 | blkg->pd[i]->online = true; | |
381 | } | |
f427d909 | 382 | } |
a637120e | 383 | } |
f427d909 | 384 | blkg->online = true; |
cd1604fa | 385 | spin_unlock(&blkcg->lock); |
496fb780 | 386 | |
ec13b1d6 | 387 | if (!ret) |
a637120e | 388 | return blkg; |
15974993 | 389 | |
3c547865 TH |
390 | /* @blkg failed fully initialized, use the usual release path */ |
391 | blkg_put(blkg); | |
392 | return ERR_PTR(ret); | |
393 | ||
d708f0d5 | 394 | err_put_css: |
496fb780 | 395 | css_put(&blkcg->css); |
93e6d5d8 | 396 | err_free_blkg: |
28e538a3 CH |
397 | if (new_blkg) |
398 | blkg_free(new_blkg); | |
93e6d5d8 | 399 | return ERR_PTR(ret); |
31e4c28d | 400 | } |
3c96cb32 | 401 | |
86cde6b6 | 402 | /** |
8c546287 | 403 | * blkg_lookup_create - lookup blkg, try to create one if not there |
86cde6b6 | 404 | * @blkcg: blkcg of interest |
99e60387 | 405 | * @disk: gendisk of interest |
86cde6b6 | 406 | * |
99e60387 | 407 | * Lookup blkg for the @blkcg - @disk pair. If it doesn't exist, try to |
3c547865 TH |
408 | * create one. blkg creation is performed recursively from blkcg_root such |
409 | * that all non-root blkg's have access to the parent blkg. This function | |
99e60387 | 410 | * should be called under RCU read lock and takes @disk->queue->queue_lock. |
86cde6b6 | 411 | * |
beea9da0 DZ |
412 | * Returns the blkg or the closest blkg if blkg_create() fails as it walks |
413 | * down from root. | |
86cde6b6 | 414 | */ |
8c546287 | 415 | static struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg, |
99e60387 | 416 | struct gendisk *disk) |
3c96cb32 | 417 | { |
99e60387 | 418 | struct request_queue *q = disk->queue; |
86cde6b6 | 419 | struct blkcg_gq *blkg; |
8c546287 | 420 | unsigned long flags; |
86cde6b6 TH |
421 | |
422 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
86cde6b6 | 423 | |
8c546287 | 424 | blkg = blkg_lookup(blkcg, q); |
86cde6b6 TH |
425 | if (blkg) |
426 | return blkg; | |
427 | ||
8c546287 | 428 | spin_lock_irqsave(&q->queue_lock, flags); |
4a69f325 CH |
429 | blkg = blkg_lookup(blkcg, q); |
430 | if (blkg) { | |
5765033c CH |
431 | if (blkcg != &blkcg_root && |
432 | blkg != rcu_dereference(blkcg->blkg_hint)) | |
433 | rcu_assign_pointer(blkcg->blkg_hint, blkg); | |
8c546287 | 434 | goto found; |
4a69f325 | 435 | } |
8c546287 | 436 | |
3c547865 TH |
437 | /* |
438 | * Create blkgs walking down from blkcg_root to @blkcg, so that all | |
beea9da0 DZ |
439 | * non-root blkgs have access to their parents. Returns the closest |
440 | * blkg to the intended blkg should blkg_create() fail. | |
3c547865 TH |
441 | */ |
442 | while (true) { | |
443 | struct blkcg *pos = blkcg; | |
444 | struct blkcg *parent = blkcg_parent(blkcg); | |
beea9da0 DZ |
445 | struct blkcg_gq *ret_blkg = q->root_blkg; |
446 | ||
447 | while (parent) { | |
79fcc5be | 448 | blkg = blkg_lookup(parent, q); |
beea9da0 DZ |
449 | if (blkg) { |
450 | /* remember closest blkg */ | |
451 | ret_blkg = blkg; | |
452 | break; | |
453 | } | |
3c547865 TH |
454 | pos = parent; |
455 | parent = blkcg_parent(parent); | |
456 | } | |
457 | ||
99e60387 | 458 | blkg = blkg_create(pos, disk, NULL); |
8c546287 CH |
459 | if (IS_ERR(blkg)) { |
460 | blkg = ret_blkg; | |
461 | break; | |
462 | } | |
beea9da0 | 463 | if (pos == blkcg) |
8c546287 | 464 | break; |
b978962a DZ |
465 | } |
466 | ||
8c546287 CH |
467 | found: |
468 | spin_unlock_irqrestore(&q->queue_lock, flags); | |
b978962a DZ |
469 | return blkg; |
470 | } | |
471 | ||
3c798398 | 472 | static void blkg_destroy(struct blkcg_gq *blkg) |
03aa264a | 473 | { |
3c798398 | 474 | struct blkcg *blkcg = blkg->blkcg; |
6b065462 | 475 | int i; |
03aa264a | 476 | |
0d945c1f | 477 | lockdep_assert_held(&blkg->q->queue_lock); |
9f13ef67 | 478 | lockdep_assert_held(&blkcg->lock); |
03aa264a | 479 | |
f1c006f1 YK |
480 | /* |
481 | * blkg stays on the queue list until blkg_free_workfn(), see details in | |
482 | * blkg_free_workfn(), hence this function can be called from | |
483 | * blkcg_destroy_blkgs() first and again from blkg_destroy_all() before | |
484 | * blkg_free_workfn(). | |
485 | */ | |
486 | if (hlist_unhashed(&blkg->blkcg_node)) | |
487 | return; | |
a637120e | 488 | |
6b065462 DZF |
489 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
490 | struct blkcg_policy *pol = blkcg_policy[i]; | |
491 | ||
dfd6200a | 492 | if (blkg->pd[i] && blkg->pd[i]->online) { |
f37bf75c | 493 | blkg->pd[i]->online = false; |
dfd6200a YK |
494 | if (pol->pd_offline_fn) |
495 | pol->pd_offline_fn(blkg->pd[i]); | |
dfd6200a | 496 | } |
6b065462 DZF |
497 | } |
498 | ||
f427d909 TH |
499 | blkg->online = false; |
500 | ||
a637120e | 501 | radix_tree_delete(&blkcg->blkg_tree, blkg->q->id); |
9f13ef67 | 502 | hlist_del_init_rcu(&blkg->blkcg_node); |
03aa264a | 503 | |
a637120e TH |
504 | /* |
505 | * Both setting lookup hint to and clearing it from @blkg are done | |
506 | * under queue_lock. If it's not pointing to @blkg now, it never | |
507 | * will. Hint assignment itself can race safely. | |
508 | */ | |
ec6c676a | 509 | if (rcu_access_pointer(blkcg->blkg_hint) == blkg) |
a637120e TH |
510 | rcu_assign_pointer(blkcg->blkg_hint, NULL); |
511 | ||
03aa264a TH |
512 | /* |
513 | * Put the reference taken at the time of creation so that when all | |
514 | * queues are gone, group can be destroyed. | |
515 | */ | |
7fcf2b03 | 516 | percpu_ref_kill(&blkg->refcnt); |
03aa264a TH |
517 | } |
518 | ||
00ad6991 | 519 | static void blkg_destroy_all(struct gendisk *disk) |
72e06c25 | 520 | { |
00ad6991 | 521 | struct request_queue *q = disk->queue; |
3c798398 | 522 | struct blkcg_gq *blkg, *n; |
a731763f | 523 | int count = BLKG_DESTROY_BATCH_SIZE; |
72e06c25 | 524 | |
a731763f | 525 | restart: |
0d945c1f | 526 | spin_lock_irq(&q->queue_lock); |
9f13ef67 | 527 | list_for_each_entry_safe(blkg, n, &q->blkg_list, q_node) { |
3c798398 | 528 | struct blkcg *blkcg = blkg->blkcg; |
72e06c25 | 529 | |
9f13ef67 TH |
530 | spin_lock(&blkcg->lock); |
531 | blkg_destroy(blkg); | |
532 | spin_unlock(&blkcg->lock); | |
a731763f YK |
533 | |
534 | /* | |
535 | * in order to avoid holding the spin lock for too long, release | |
536 | * it when a batch of blkgs are destroyed. | |
537 | */ | |
538 | if (!(--count)) { | |
539 | count = BLKG_DESTROY_BATCH_SIZE; | |
540 | spin_unlock_irq(&q->queue_lock); | |
541 | cond_resched(); | |
542 | goto restart; | |
543 | } | |
72e06c25 | 544 | } |
6fe810bd TH |
545 | |
546 | q->root_blkg = NULL; | |
0d945c1f | 547 | spin_unlock_irq(&q->queue_lock); |
72e06c25 TH |
548 | } |
549 | ||
182446d0 TH |
550 | static int blkcg_reset_stats(struct cgroup_subsys_state *css, |
551 | struct cftype *cftype, u64 val) | |
303a3acb | 552 | { |
182446d0 | 553 | struct blkcg *blkcg = css_to_blkcg(css); |
3c798398 | 554 | struct blkcg_gq *blkg; |
f7331648 | 555 | int i, cpu; |
303a3acb | 556 | |
838f13bf | 557 | mutex_lock(&blkcg_pol_mutex); |
303a3acb | 558 | spin_lock_irq(&blkcg->lock); |
997a026c TH |
559 | |
560 | /* | |
561 | * Note that stat reset is racy - it doesn't synchronize against | |
562 | * stat updates. This is a debug feature which shouldn't exist | |
563 | * anyway. If you get hit by a race, retry. | |
564 | */ | |
b67bfe0d | 565 | hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) { |
f7331648 TH |
566 | for_each_possible_cpu(cpu) { |
567 | struct blkg_iostat_set *bis = | |
568 | per_cpu_ptr(blkg->iostat_cpu, cpu); | |
569 | memset(bis, 0, sizeof(*bis)); | |
570 | } | |
571 | memset(&blkg->iostat, 0, sizeof(blkg->iostat)); | |
77ea7338 | 572 | |
8bd435b3 | 573 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
3c798398 | 574 | struct blkcg_policy *pol = blkcg_policy[i]; |
549d3aa8 | 575 | |
a9520cd6 TH |
576 | if (blkg->pd[i] && pol->pd_reset_stats_fn) |
577 | pol->pd_reset_stats_fn(blkg->pd[i]); | |
bc0d6501 | 578 | } |
303a3acb | 579 | } |
f0bdc8cd | 580 | |
303a3acb | 581 | spin_unlock_irq(&blkcg->lock); |
bc0d6501 | 582 | mutex_unlock(&blkcg_pol_mutex); |
303a3acb DS |
583 | return 0; |
584 | } | |
585 | ||
dd165eb3 | 586 | const char *blkg_dev_name(struct blkcg_gq *blkg) |
303a3acb | 587 | { |
a06377c5 | 588 | if (!blkg->q->disk) |
edb0872f | 589 | return NULL; |
d152c682 | 590 | return bdi_dev_name(blkg->q->disk->bdi); |
303a3acb DS |
591 | } |
592 | ||
d3d32e69 TH |
593 | /** |
594 | * blkcg_print_blkgs - helper for printing per-blkg data | |
595 | * @sf: seq_file to print to | |
596 | * @blkcg: blkcg of interest | |
597 | * @prfill: fill function to print out a blkg | |
598 | * @pol: policy in question | |
599 | * @data: data to be passed to @prfill | |
600 | * @show_total: to print out sum of prfill return values or not | |
601 | * | |
602 | * This function invokes @prfill on each blkg of @blkcg if pd for the | |
603 | * policy specified by @pol exists. @prfill is invoked with @sf, the | |
810ecfa7 TH |
604 | * policy data and @data and the matching queue lock held. If @show_total |
605 | * is %true, the sum of the return values from @prfill is printed with | |
606 | * "Total" label at the end. | |
d3d32e69 TH |
607 | * |
608 | * This is to be used to construct print functions for | |
609 | * cftype->read_seq_string method. | |
610 | */ | |
3c798398 | 611 | void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg, |
f95a04af TH |
612 | u64 (*prfill)(struct seq_file *, |
613 | struct blkg_policy_data *, int), | |
3c798398 | 614 | const struct blkcg_policy *pol, int data, |
ec399347 | 615 | bool show_total) |
5624a4e4 | 616 | { |
3c798398 | 617 | struct blkcg_gq *blkg; |
d3d32e69 | 618 | u64 total = 0; |
5624a4e4 | 619 | |
810ecfa7 | 620 | rcu_read_lock(); |
ee89f812 | 621 | hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) { |
0d945c1f | 622 | spin_lock_irq(&blkg->q->queue_lock); |
a2b1693b | 623 | if (blkcg_policy_enabled(blkg->q, pol)) |
f95a04af | 624 | total += prfill(sf, blkg->pd[pol->plid], data); |
0d945c1f | 625 | spin_unlock_irq(&blkg->q->queue_lock); |
810ecfa7 TH |
626 | } |
627 | rcu_read_unlock(); | |
d3d32e69 TH |
628 | |
629 | if (show_total) | |
630 | seq_printf(sf, "Total %llu\n", (unsigned long long)total); | |
631 | } | |
829fdb50 | 632 | EXPORT_SYMBOL_GPL(blkcg_print_blkgs); |
d3d32e69 TH |
633 | |
634 | /** | |
635 | * __blkg_prfill_u64 - prfill helper for a single u64 value | |
636 | * @sf: seq_file to print to | |
f95a04af | 637 | * @pd: policy private data of interest |
d3d32e69 TH |
638 | * @v: value to print |
639 | * | |
37754595 | 640 | * Print @v to @sf for the device associated with @pd. |
d3d32e69 | 641 | */ |
f95a04af | 642 | u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v) |
d3d32e69 | 643 | { |
f95a04af | 644 | const char *dname = blkg_dev_name(pd->blkg); |
d3d32e69 TH |
645 | |
646 | if (!dname) | |
647 | return 0; | |
648 | ||
649 | seq_printf(sf, "%s %llu\n", dname, (unsigned long long)v); | |
650 | return v; | |
651 | } | |
829fdb50 | 652 | EXPORT_SYMBOL_GPL(__blkg_prfill_u64); |
d3d32e69 | 653 | |
015d254c | 654 | /** |
faffaab2 TH |
655 | * blkg_conf_init - initialize a blkg_conf_ctx |
656 | * @ctx: blkg_conf_ctx to initialize | |
657 | * @input: input string | |
015d254c | 658 | * |
faffaab2 TH |
659 | * Initialize @ctx which can be used to parse blkg config input string @input. |
660 | * Once initialized, @ctx can be used with blkg_conf_open_bdev() and | |
661 | * blkg_conf_prep(), and must be cleaned up with blkg_conf_exit(). | |
662 | */ | |
663 | void blkg_conf_init(struct blkg_conf_ctx *ctx, char *input) | |
664 | { | |
665 | *ctx = (struct blkg_conf_ctx){ .input = input }; | |
666 | } | |
667 | EXPORT_SYMBOL_GPL(blkg_conf_init); | |
668 | ||
669 | /** | |
670 | * blkg_conf_open_bdev - parse and open bdev for per-blkg config update | |
671 | * @ctx: blkg_conf_ctx initialized with blkg_conf_init() | |
015d254c | 672 | * |
faffaab2 TH |
673 | * Parse the device node prefix part, MAJ:MIN, of per-blkg config update from |
674 | * @ctx->input and get and store the matching bdev in @ctx->bdev. @ctx->body is | |
675 | * set to point past the device node prefix. | |
676 | * | |
677 | * This function may be called multiple times on @ctx and the extra calls become | |
678 | * NOOPs. blkg_conf_prep() implicitly calls this function. Use this function | |
679 | * explicitly if bdev access is needed without resolving the blkcg / policy part | |
680 | * of @ctx->input. Returns -errno on error. | |
015d254c | 681 | */ |
faffaab2 | 682 | int blkg_conf_open_bdev(struct blkg_conf_ctx *ctx) |
015d254c | 683 | { |
faffaab2 | 684 | char *input = ctx->input; |
015d254c | 685 | unsigned int major, minor; |
22ae8ce8 CH |
686 | struct block_device *bdev; |
687 | int key_len; | |
015d254c | 688 | |
faffaab2 TH |
689 | if (ctx->bdev) |
690 | return 0; | |
691 | ||
015d254c | 692 | if (sscanf(input, "%u:%u%n", &major, &minor, &key_len) != 2) |
faffaab2 | 693 | return -EINVAL; |
015d254c TH |
694 | |
695 | input += key_len; | |
696 | if (!isspace(*input)) | |
faffaab2 | 697 | return -EINVAL; |
015d254c TH |
698 | input = skip_spaces(input); |
699 | ||
22ae8ce8 CH |
700 | bdev = blkdev_get_no_open(MKDEV(major, minor)); |
701 | if (!bdev) | |
faffaab2 | 702 | return -ENODEV; |
22ae8ce8 CH |
703 | if (bdev_is_partition(bdev)) { |
704 | blkdev_put_no_open(bdev); | |
faffaab2 | 705 | return -ENODEV; |
015d254c TH |
706 | } |
707 | ||
faffaab2 TH |
708 | ctx->body = input; |
709 | ctx->bdev = bdev; | |
710 | return 0; | |
015d254c TH |
711 | } |
712 | ||
3a8b31d3 TH |
713 | /** |
714 | * blkg_conf_prep - parse and prepare for per-blkg config update | |
715 | * @blkcg: target block cgroup | |
da8b0662 | 716 | * @pol: target policy |
faffaab2 TH |
717 | * @ctx: blkg_conf_ctx initialized with blkg_conf_init() |
718 | * | |
719 | * Parse per-blkg config update from @ctx->input and initialize @ctx | |
720 | * accordingly. On success, @ctx->body points to the part of @ctx->input | |
721 | * following MAJ:MIN, @ctx->bdev points to the target block device and | |
722 | * @ctx->blkg to the blkg being configured. | |
3a8b31d3 | 723 | * |
faffaab2 TH |
724 | * blkg_conf_open_bdev() may be called on @ctx beforehand. On success, this |
725 | * function returns with queue lock held and must be followed by | |
726 | * blkg_conf_exit(). | |
3a8b31d3 | 727 | */ |
3c798398 | 728 | int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol, |
faffaab2 | 729 | struct blkg_conf_ctx *ctx) |
83462a6c | 730 | __acquires(&bdev->bd_queue->queue_lock) |
34d0f179 | 731 | { |
99e60387 | 732 | struct gendisk *disk; |
457e490f | 733 | struct request_queue *q; |
3c798398 | 734 | struct blkcg_gq *blkg; |
015d254c | 735 | int ret; |
36aa9e5f | 736 | |
faffaab2 TH |
737 | ret = blkg_conf_open_bdev(ctx); |
738 | if (ret) | |
739 | return ret; | |
740 | ||
741 | disk = ctx->bdev->bd_disk; | |
99e60387 | 742 | q = disk->queue; |
da8b0662 | 743 | |
0c9d338c YK |
744 | /* |
745 | * blkcg_deactivate_policy() requires queue to be frozen, we can grab | |
746 | * q_usage_counter to prevent concurrent with blkcg_deactivate_policy(). | |
747 | */ | |
748 | ret = blk_queue_enter(q, 0); | |
749 | if (ret) | |
15c30104 | 750 | goto fail; |
0c9d338c | 751 | |
0d945c1f | 752 | spin_lock_irq(&q->queue_lock); |
e56da7e2 | 753 | |
f753526e CH |
754 | if (!blkcg_policy_enabled(q, pol)) { |
755 | ret = -EOPNOTSUPP; | |
457e490f TE |
756 | goto fail_unlock; |
757 | } | |
758 | ||
f753526e | 759 | blkg = blkg_lookup(blkcg, q); |
5765033c | 760 | if (blkg) |
457e490f TE |
761 | goto success; |
762 | ||
763 | /* | |
764 | * Create blkgs walking down from blkcg_root to @blkcg, so that all | |
765 | * non-root blkgs have access to their parents. | |
766 | */ | |
767 | while (true) { | |
768 | struct blkcg *pos = blkcg; | |
769 | struct blkcg *parent; | |
770 | struct blkcg_gq *new_blkg; | |
771 | ||
772 | parent = blkcg_parent(blkcg); | |
79fcc5be | 773 | while (parent && !blkg_lookup(parent, q)) { |
457e490f TE |
774 | pos = parent; |
775 | parent = blkcg_parent(parent); | |
776 | } | |
777 | ||
778 | /* Drop locks to do new blkg allocation with GFP_KERNEL. */ | |
0d945c1f | 779 | spin_unlock_irq(&q->queue_lock); |
457e490f | 780 | |
99e60387 | 781 | new_blkg = blkg_alloc(pos, disk, GFP_KERNEL); |
457e490f TE |
782 | if (unlikely(!new_blkg)) { |
783 | ret = -ENOMEM; | |
15c30104 | 784 | goto fail_exit_queue; |
7702e8f4 | 785 | } |
3a8b31d3 | 786 | |
f255c19b GKB |
787 | if (radix_tree_preload(GFP_KERNEL)) { |
788 | blkg_free(new_blkg); | |
789 | ret = -ENOMEM; | |
15c30104 | 790 | goto fail_exit_queue; |
f255c19b GKB |
791 | } |
792 | ||
0d945c1f | 793 | spin_lock_irq(&q->queue_lock); |
457e490f | 794 | |
f753526e | 795 | if (!blkcg_policy_enabled(q, pol)) { |
52abfcbd | 796 | blkg_free(new_blkg); |
f753526e | 797 | ret = -EOPNOTSUPP; |
f255c19b | 798 | goto fail_preloaded; |
457e490f TE |
799 | } |
800 | ||
f753526e | 801 | blkg = blkg_lookup(pos, q); |
457e490f TE |
802 | if (blkg) { |
803 | blkg_free(new_blkg); | |
804 | } else { | |
99e60387 | 805 | blkg = blkg_create(pos, disk, new_blkg); |
98d669b4 | 806 | if (IS_ERR(blkg)) { |
457e490f | 807 | ret = PTR_ERR(blkg); |
f255c19b | 808 | goto fail_preloaded; |
457e490f TE |
809 | } |
810 | } | |
811 | ||
f255c19b GKB |
812 | radix_tree_preload_end(); |
813 | ||
457e490f TE |
814 | if (pos == blkcg) |
815 | goto success; | |
816 | } | |
817 | success: | |
0c9d338c | 818 | blk_queue_exit(q); |
3a8b31d3 | 819 | ctx->blkg = blkg; |
726fa694 | 820 | return 0; |
457e490f | 821 | |
f255c19b GKB |
822 | fail_preloaded: |
823 | radix_tree_preload_end(); | |
457e490f | 824 | fail_unlock: |
0d945c1f | 825 | spin_unlock_irq(&q->queue_lock); |
15c30104 YK |
826 | fail_exit_queue: |
827 | blk_queue_exit(q); | |
457e490f | 828 | fail: |
457e490f TE |
829 | /* |
830 | * If queue was bypassing, we should retry. Do so after a | |
831 | * short msleep(). It isn't strictly necessary but queue | |
832 | * can be bypassing for some time and it's always nice to | |
833 | * avoid busy looping. | |
834 | */ | |
835 | if (ret == -EBUSY) { | |
836 | msleep(10); | |
837 | ret = restart_syscall(); | |
838 | } | |
839 | return ret; | |
34d0f179 | 840 | } |
89f3b6d6 | 841 | EXPORT_SYMBOL_GPL(blkg_conf_prep); |
34d0f179 | 842 | |
3a8b31d3 | 843 | /** |
faffaab2 TH |
844 | * blkg_conf_exit - clean up per-blkg config update |
845 | * @ctx: blkg_conf_ctx initialized with blkg_conf_init() | |
3a8b31d3 | 846 | * |
faffaab2 TH |
847 | * Clean up after per-blkg config update. This function must be called on all |
848 | * blkg_conf_ctx's initialized with blkg_conf_init(). | |
3a8b31d3 | 849 | */ |
faffaab2 | 850 | void blkg_conf_exit(struct blkg_conf_ctx *ctx) |
83462a6c | 851 | __releases(&ctx->bdev->bd_queue->queue_lock) |
34d0f179 | 852 | { |
faffaab2 TH |
853 | if (ctx->blkg) { |
854 | spin_unlock_irq(&bdev_get_queue(ctx->bdev)->queue_lock); | |
855 | ctx->blkg = NULL; | |
856 | } | |
857 | ||
858 | if (ctx->bdev) { | |
859 | blkdev_put_no_open(ctx->bdev); | |
860 | ctx->body = NULL; | |
861 | ctx->bdev = NULL; | |
862 | } | |
34d0f179 | 863 | } |
faffaab2 | 864 | EXPORT_SYMBOL_GPL(blkg_conf_exit); |
34d0f179 | 865 | |
cd1fc4b9 BB |
866 | static void blkg_iostat_set(struct blkg_iostat *dst, struct blkg_iostat *src) |
867 | { | |
868 | int i; | |
869 | ||
870 | for (i = 0; i < BLKG_IOSTAT_NR; i++) { | |
871 | dst->bytes[i] = src->bytes[i]; | |
872 | dst->ios[i] = src->ios[i]; | |
873 | } | |
874 | } | |
875 | ||
876 | static void blkg_iostat_add(struct blkg_iostat *dst, struct blkg_iostat *src) | |
877 | { | |
878 | int i; | |
879 | ||
880 | for (i = 0; i < BLKG_IOSTAT_NR; i++) { | |
881 | dst->bytes[i] += src->bytes[i]; | |
882 | dst->ios[i] += src->ios[i]; | |
883 | } | |
884 | } | |
885 | ||
886 | static void blkg_iostat_sub(struct blkg_iostat *dst, struct blkg_iostat *src) | |
887 | { | |
888 | int i; | |
889 | ||
890 | for (i = 0; i < BLKG_IOSTAT_NR; i++) { | |
891 | dst->bytes[i] -= src->bytes[i]; | |
892 | dst->ios[i] -= src->ios[i]; | |
893 | } | |
894 | } | |
895 | ||
362b8c16 JY |
896 | static void blkcg_iostat_update(struct blkcg_gq *blkg, struct blkg_iostat *cur, |
897 | struct blkg_iostat *last) | |
898 | { | |
899 | struct blkg_iostat delta; | |
900 | unsigned long flags; | |
901 | ||
902 | /* propagate percpu delta to global */ | |
903 | flags = u64_stats_update_begin_irqsave(&blkg->iostat.sync); | |
904 | blkg_iostat_set(&delta, cur); | |
905 | blkg_iostat_sub(&delta, last); | |
906 | blkg_iostat_add(&blkg->iostat.cur, &delta); | |
907 | blkg_iostat_add(last, &delta); | |
908 | u64_stats_update_end_irqrestore(&blkg->iostat.sync, flags); | |
909 | } | |
910 | ||
cd1fc4b9 BB |
911 | static void blkcg_rstat_flush(struct cgroup_subsys_state *css, int cpu) |
912 | { | |
913 | struct blkcg *blkcg = css_to_blkcg(css); | |
3b8cc629 WL |
914 | struct llist_head *lhead = per_cpu_ptr(blkcg->lhead, cpu); |
915 | struct llist_node *lnode; | |
916 | struct blkg_iostat_set *bisc, *next_bisc; | |
cd1fc4b9 | 917 | |
dc26532a JW |
918 | /* Root-level stats are sourced from system-wide IO stats */ |
919 | if (!cgroup_parent(css->cgroup)) | |
920 | return; | |
921 | ||
cd1fc4b9 BB |
922 | rcu_read_lock(); |
923 | ||
3b8cc629 WL |
924 | lnode = llist_del_all(lhead); |
925 | if (!lnode) | |
926 | goto out; | |
927 | ||
928 | /* | |
929 | * Iterate only the iostat_cpu's queued in the lockless list. | |
930 | */ | |
931 | llist_for_each_entry_safe(bisc, next_bisc, lnode, lnode) { | |
932 | struct blkcg_gq *blkg = bisc->blkg; | |
cd1fc4b9 | 933 | struct blkcg_gq *parent = blkg->parent; |
362b8c16 | 934 | struct blkg_iostat cur; |
cd1fc4b9 BB |
935 | unsigned int seq; |
936 | ||
3b8cc629 WL |
937 | WRITE_ONCE(bisc->lqueued, false); |
938 | ||
cd1fc4b9 BB |
939 | /* fetch the current per-cpu values */ |
940 | do { | |
941 | seq = u64_stats_fetch_begin(&bisc->sync); | |
942 | blkg_iostat_set(&cur, &bisc->cur); | |
943 | } while (u64_stats_fetch_retry(&bisc->sync, seq)); | |
944 | ||
362b8c16 | 945 | blkcg_iostat_update(blkg, &cur, &bisc->last); |
cd1fc4b9 | 946 | |
dc26532a | 947 | /* propagate global delta to parent (unless that's root) */ |
362b8c16 JY |
948 | if (parent && parent->parent) |
949 | blkcg_iostat_update(parent, &blkg->iostat.cur, | |
950 | &blkg->iostat.last); | |
3b8cc629 | 951 | percpu_ref_put(&blkg->refcnt); |
cd1fc4b9 BB |
952 | } |
953 | ||
3b8cc629 | 954 | out: |
cd1fc4b9 BB |
955 | rcu_read_unlock(); |
956 | } | |
957 | ||
ef45fe47 | 958 | /* |
dc26532a JW |
959 | * We source root cgroup stats from the system-wide stats to avoid |
960 | * tracking the same information twice and incurring overhead when no | |
961 | * cgroups are defined. For that reason, cgroup_rstat_flush in | |
962 | * blkcg_print_stat does not actually fill out the iostat in the root | |
963 | * cgroup's blkcg_gq. | |
ef45fe47 BB |
964 | * |
965 | * However, we would like to re-use the printing code between the root and | |
966 | * non-root cgroups to the extent possible. For that reason, we simulate | |
967 | * flushing the root cgroup's stats by explicitly filling in the iostat | |
968 | * with disk level statistics. | |
969 | */ | |
970 | static void blkcg_fill_root_iostats(void) | |
971 | { | |
972 | struct class_dev_iter iter; | |
973 | struct device *dev; | |
974 | ||
975 | class_dev_iter_init(&iter, &block_class, NULL, &disk_type); | |
976 | while ((dev = class_dev_iter_next(&iter))) { | |
0d02129e | 977 | struct block_device *bdev = dev_to_bdev(dev); |
928f6f00 | 978 | struct blkcg_gq *blkg = bdev->bd_disk->queue->root_blkg; |
ef45fe47 BB |
979 | struct blkg_iostat tmp; |
980 | int cpu; | |
f122d103 | 981 | unsigned long flags; |
ef45fe47 BB |
982 | |
983 | memset(&tmp, 0, sizeof(tmp)); | |
984 | for_each_possible_cpu(cpu) { | |
985 | struct disk_stats *cpu_dkstats; | |
986 | ||
0d02129e | 987 | cpu_dkstats = per_cpu_ptr(bdev->bd_stats, cpu); |
ef45fe47 BB |
988 | tmp.ios[BLKG_IOSTAT_READ] += |
989 | cpu_dkstats->ios[STAT_READ]; | |
990 | tmp.ios[BLKG_IOSTAT_WRITE] += | |
991 | cpu_dkstats->ios[STAT_WRITE]; | |
992 | tmp.ios[BLKG_IOSTAT_DISCARD] += | |
993 | cpu_dkstats->ios[STAT_DISCARD]; | |
994 | // convert sectors to bytes | |
995 | tmp.bytes[BLKG_IOSTAT_READ] += | |
996 | cpu_dkstats->sectors[STAT_READ] << 9; | |
997 | tmp.bytes[BLKG_IOSTAT_WRITE] += | |
998 | cpu_dkstats->sectors[STAT_WRITE] << 9; | |
999 | tmp.bytes[BLKG_IOSTAT_DISCARD] += | |
1000 | cpu_dkstats->sectors[STAT_DISCARD] << 9; | |
ef45fe47 | 1001 | } |
f122d103 CZ |
1002 | |
1003 | flags = u64_stats_update_begin_irqsave(&blkg->iostat.sync); | |
1004 | blkg_iostat_set(&blkg->iostat.cur, &tmp); | |
1005 | u64_stats_update_end_irqrestore(&blkg->iostat.sync, flags); | |
ef45fe47 BB |
1006 | } |
1007 | } | |
1008 | ||
49cb5168 | 1009 | static void blkcg_print_one_stat(struct blkcg_gq *blkg, struct seq_file *s) |
2ee867dc | 1010 | { |
49cb5168 CH |
1011 | struct blkg_iostat_set *bis = &blkg->iostat; |
1012 | u64 rbytes, wbytes, rios, wios, dbytes, dios; | |
49cb5168 CH |
1013 | const char *dname; |
1014 | unsigned seq; | |
49cb5168 | 1015 | int i; |
ef45fe47 | 1016 | |
49cb5168 CH |
1017 | if (!blkg->online) |
1018 | return; | |
2ee867dc | 1019 | |
49cb5168 CH |
1020 | dname = blkg_dev_name(blkg); |
1021 | if (!dname) | |
1022 | return; | |
2ee867dc | 1023 | |
252c651a | 1024 | seq_printf(s, "%s ", dname); |
b0814361 | 1025 | |
49cb5168 CH |
1026 | do { |
1027 | seq = u64_stats_fetch_begin(&bis->sync); | |
b0814361 | 1028 | |
49cb5168 CH |
1029 | rbytes = bis->cur.bytes[BLKG_IOSTAT_READ]; |
1030 | wbytes = bis->cur.bytes[BLKG_IOSTAT_WRITE]; | |
1031 | dbytes = bis->cur.bytes[BLKG_IOSTAT_DISCARD]; | |
1032 | rios = bis->cur.ios[BLKG_IOSTAT_READ]; | |
1033 | wios = bis->cur.ios[BLKG_IOSTAT_WRITE]; | |
1034 | dios = bis->cur.ios[BLKG_IOSTAT_DISCARD]; | |
1035 | } while (u64_stats_fetch_retry(&bis->sync, seq)); | |
2ee867dc | 1036 | |
49cb5168 | 1037 | if (rbytes || wbytes || rios || wios) { |
252c651a | 1038 | seq_printf(s, "rbytes=%llu wbytes=%llu rios=%llu wios=%llu dbytes=%llu dios=%llu", |
49cb5168 CH |
1039 | rbytes, wbytes, rios, wios, |
1040 | dbytes, dios); | |
1041 | } | |
903d23f0 | 1042 | |
49cb5168 | 1043 | if (blkcg_debug_stats && atomic_read(&blkg->use_delay)) { |
252c651a | 1044 | seq_printf(s, " use_delay=%d delay_nsec=%llu", |
49cb5168 CH |
1045 | atomic_read(&blkg->use_delay), |
1046 | atomic64_read(&blkg->delay_nsec)); | |
1047 | } | |
2ee867dc | 1048 | |
49cb5168 CH |
1049 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
1050 | struct blkcg_policy *pol = blkcg_policy[i]; | |
2ee867dc | 1051 | |
49cb5168 CH |
1052 | if (!blkg->pd[i] || !pol->pd_stat_fn) |
1053 | continue; | |
903d23f0 | 1054 | |
3607849d | 1055 | pol->pd_stat_fn(blkg->pd[i], s); |
49cb5168 | 1056 | } |
d09d8df3 | 1057 | |
3607849d | 1058 | seq_puts(s, "\n"); |
49cb5168 | 1059 | } |
903d23f0 | 1060 | |
49cb5168 CH |
1061 | static int blkcg_print_stat(struct seq_file *sf, void *v) |
1062 | { | |
1063 | struct blkcg *blkcg = css_to_blkcg(seq_css(sf)); | |
1064 | struct blkcg_gq *blkg; | |
903d23f0 | 1065 | |
49cb5168 CH |
1066 | if (!seq_css(sf)->parent) |
1067 | blkcg_fill_root_iostats(); | |
1068 | else | |
1069 | cgroup_rstat_flush(blkcg->css.cgroup); | |
07b0fdec | 1070 | |
49cb5168 CH |
1071 | rcu_read_lock(); |
1072 | hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) { | |
1073 | spin_lock_irq(&blkg->q->queue_lock); | |
1074 | blkcg_print_one_stat(blkg, sf); | |
b0814361 | 1075 | spin_unlock_irq(&blkg->q->queue_lock); |
2ee867dc | 1076 | } |
2ee867dc TH |
1077 | rcu_read_unlock(); |
1078 | return 0; | |
1079 | } | |
1080 | ||
e1f3b941 | 1081 | static struct cftype blkcg_files[] = { |
2ee867dc TH |
1082 | { |
1083 | .name = "stat", | |
1084 | .seq_show = blkcg_print_stat, | |
1085 | }, | |
1086 | { } /* terminate */ | |
1087 | }; | |
1088 | ||
e1f3b941 | 1089 | static struct cftype blkcg_legacy_files[] = { |
84c124da DS |
1090 | { |
1091 | .name = "reset_stats", | |
3c798398 | 1092 | .write_u64 = blkcg_reset_stats, |
22084190 | 1093 | }, |
4baf6e33 | 1094 | { } /* terminate */ |
31e4c28d VG |
1095 | }; |
1096 | ||
dec223c9 CH |
1097 | #ifdef CONFIG_CGROUP_WRITEBACK |
1098 | struct list_head *blkcg_get_cgwb_list(struct cgroup_subsys_state *css) | |
1099 | { | |
1100 | return &css_to_blkcg(css)->cgwb_list; | |
1101 | } | |
1102 | #endif | |
1103 | ||
59b57717 DZF |
1104 | /* |
1105 | * blkcg destruction is a three-stage process. | |
1106 | * | |
1107 | * 1. Destruction starts. The blkcg_css_offline() callback is invoked | |
1108 | * which offlines writeback. Here we tie the next stage of blkg destruction | |
1109 | * to the completion of writeback associated with the blkcg. This lets us | |
1110 | * avoid punting potentially large amounts of outstanding writeback to root | |
1111 | * while maintaining any ongoing policies. The next stage is triggered when | |
1112 | * the nr_cgwbs count goes to zero. | |
1113 | * | |
1114 | * 2. When the nr_cgwbs count goes to zero, blkcg_destroy_blkgs() is called | |
1115 | * and handles the destruction of blkgs. Here the css reference held by | |
1116 | * the blkg is put back eventually allowing blkcg_css_free() to be called. | |
1117 | * This work may occur in cgwb_release_workfn() on the cgwb_release | |
1118 | * workqueue. Any submitted ios that fail to get the blkg ref will be | |
1119 | * punted to the root_blkg. | |
1120 | * | |
1121 | * 3. Once the blkcg ref count goes to zero, blkcg_css_free() is called. | |
1122 | * This finally frees the blkcg. | |
1123 | */ | |
1124 | ||
59b57717 DZF |
1125 | /** |
1126 | * blkcg_destroy_blkgs - responsible for shooting down blkgs | |
1127 | * @blkcg: blkcg of interest | |
1128 | * | |
1129 | * blkgs should be removed while holding both q and blkcg locks. As blkcg lock | |
1130 | * is nested inside q lock, this function performs reverse double lock dancing. | |
1131 | * Destroying the blkgs releases the reference held on the blkcg's css allowing | |
1132 | * blkcg_css_free to eventually be called. | |
1133 | * | |
1134 | * This is the blkcg counterpart of ioc_release_fn(). | |
1135 | */ | |
397c9f46 | 1136 | static void blkcg_destroy_blkgs(struct blkcg *blkcg) |
59b57717 | 1137 | { |
6c635cae BW |
1138 | might_sleep(); |
1139 | ||
9f13ef67 | 1140 | spin_lock_irq(&blkcg->lock); |
7ee9c562 | 1141 | |
4c699480 JQ |
1142 | while (!hlist_empty(&blkcg->blkg_list)) { |
1143 | struct blkcg_gq *blkg = hlist_entry(blkcg->blkg_list.first, | |
6b065462 | 1144 | struct blkcg_gq, blkcg_node); |
4c699480 JQ |
1145 | struct request_queue *q = blkg->q; |
1146 | ||
6c635cae BW |
1147 | if (need_resched() || !spin_trylock(&q->queue_lock)) { |
1148 | /* | |
1149 | * Given that the system can accumulate a huge number | |
1150 | * of blkgs in pathological cases, check to see if we | |
1151 | * need to rescheduling to avoid softlockup. | |
1152 | */ | |
4c699480 | 1153 | spin_unlock_irq(&blkcg->lock); |
6c635cae | 1154 | cond_resched(); |
4c699480 | 1155 | spin_lock_irq(&blkcg->lock); |
6c635cae | 1156 | continue; |
4c699480 | 1157 | } |
6c635cae BW |
1158 | |
1159 | blkg_destroy(blkg); | |
1160 | spin_unlock(&q->queue_lock); | |
4c699480 | 1161 | } |
6b065462 | 1162 | |
4c699480 JQ |
1163 | spin_unlock_irq(&blkcg->lock); |
1164 | } | |
1165 | ||
397c9f46 CH |
1166 | /** |
1167 | * blkcg_pin_online - pin online state | |
1168 | * @blkcg_css: blkcg of interest | |
1169 | * | |
1170 | * While pinned, a blkcg is kept online. This is primarily used to | |
1171 | * impedance-match blkg and cgwb lifetimes so that blkg doesn't go offline | |
1172 | * while an associated cgwb is still active. | |
1173 | */ | |
1174 | void blkcg_pin_online(struct cgroup_subsys_state *blkcg_css) | |
1175 | { | |
1176 | refcount_inc(&css_to_blkcg(blkcg_css)->online_pin); | |
1177 | } | |
1178 | ||
1179 | /** | |
1180 | * blkcg_unpin_online - unpin online state | |
1181 | * @blkcg_css: blkcg of interest | |
1182 | * | |
1183 | * This is primarily used to impedance-match blkg and cgwb lifetimes so | |
1184 | * that blkg doesn't go offline while an associated cgwb is still active. | |
1185 | * When this count goes to zero, all active cgwbs have finished so the | |
1186 | * blkcg can continue destruction by calling blkcg_destroy_blkgs(). | |
1187 | */ | |
1188 | void blkcg_unpin_online(struct cgroup_subsys_state *blkcg_css) | |
1189 | { | |
1190 | struct blkcg *blkcg = css_to_blkcg(blkcg_css); | |
1191 | ||
1192 | do { | |
1193 | if (!refcount_dec_and_test(&blkcg->online_pin)) | |
1194 | break; | |
1195 | blkcg_destroy_blkgs(blkcg); | |
1196 | blkcg = blkcg_parent(blkcg); | |
1197 | } while (blkcg); | |
1198 | } | |
1199 | ||
1200 | /** | |
1201 | * blkcg_css_offline - cgroup css_offline callback | |
1202 | * @css: css of interest | |
1203 | * | |
1204 | * This function is called when @css is about to go away. Here the cgwbs are | |
1205 | * offlined first and only once writeback associated with the blkcg has | |
1206 | * finished do we start step 2 (see above). | |
1207 | */ | |
1208 | static void blkcg_css_offline(struct cgroup_subsys_state *css) | |
1209 | { | |
1210 | /* this prevents anyone from attaching or migrating to this blkcg */ | |
dec223c9 | 1211 | wb_blkcg_offline(css); |
397c9f46 CH |
1212 | |
1213 | /* put the base online pin allowing step 2 to be triggered */ | |
1214 | blkcg_unpin_online(css); | |
1215 | } | |
1216 | ||
eb95419b | 1217 | static void blkcg_css_free(struct cgroup_subsys_state *css) |
7ee9c562 | 1218 | { |
eb95419b | 1219 | struct blkcg *blkcg = css_to_blkcg(css); |
bc915e61 | 1220 | int i; |
7ee9c562 | 1221 | |
7876f930 | 1222 | mutex_lock(&blkcg_pol_mutex); |
e4a9bde9 | 1223 | |
7876f930 | 1224 | list_del(&blkcg->all_blkcgs_node); |
7876f930 | 1225 | |
bc915e61 | 1226 | for (i = 0; i < BLKCG_MAX_POLS; i++) |
e4a9bde9 TH |
1227 | if (blkcg->cpd[i]) |
1228 | blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]); | |
1229 | ||
1230 | mutex_unlock(&blkcg_pol_mutex); | |
1231 | ||
3b8cc629 | 1232 | free_percpu(blkcg->lhead); |
bc915e61 | 1233 | kfree(blkcg); |
31e4c28d VG |
1234 | } |
1235 | ||
eb95419b TH |
1236 | static struct cgroup_subsys_state * |
1237 | blkcg_css_alloc(struct cgroup_subsys_state *parent_css) | |
31e4c28d | 1238 | { |
3c798398 | 1239 | struct blkcg *blkcg; |
e48453c3 | 1240 | int i; |
31e4c28d | 1241 | |
7876f930 TH |
1242 | mutex_lock(&blkcg_pol_mutex); |
1243 | ||
eb95419b | 1244 | if (!parent_css) { |
3c798398 | 1245 | blkcg = &blkcg_root; |
bc915e61 TH |
1246 | } else { |
1247 | blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL); | |
b5a9adcb | 1248 | if (!blkcg) |
4c18c9e9 | 1249 | goto unlock; |
e48453c3 AA |
1250 | } |
1251 | ||
3b8cc629 WL |
1252 | if (init_blkcg_llists(blkcg)) |
1253 | goto free_blkcg; | |
1254 | ||
e48453c3 AA |
1255 | for (i = 0; i < BLKCG_MAX_POLS ; i++) { |
1256 | struct blkcg_policy *pol = blkcg_policy[i]; | |
1257 | struct blkcg_policy_data *cpd; | |
1258 | ||
1259 | /* | |
1260 | * If the policy hasn't been attached yet, wait for it | |
1261 | * to be attached before doing anything else. Otherwise, | |
1262 | * check if the policy requires any specific per-cgroup | |
1263 | * data: if it does, allocate and initialize it. | |
1264 | */ | |
e4a9bde9 | 1265 | if (!pol || !pol->cpd_alloc_fn) |
e48453c3 AA |
1266 | continue; |
1267 | ||
e4a9bde9 | 1268 | cpd = pol->cpd_alloc_fn(GFP_KERNEL); |
b5a9adcb | 1269 | if (!cpd) |
e48453c3 | 1270 | goto free_pd_blkcg; |
b5a9adcb | 1271 | |
81437648 TH |
1272 | blkcg->cpd[i] = cpd; |
1273 | cpd->blkcg = blkcg; | |
e48453c3 | 1274 | cpd->plid = i; |
e48453c3 | 1275 | } |
31e4c28d | 1276 | |
31e4c28d | 1277 | spin_lock_init(&blkcg->lock); |
d866dbf6 | 1278 | refcount_set(&blkcg->online_pin, 1); |
e00f4f4d | 1279 | INIT_RADIX_TREE(&blkcg->blkg_tree, GFP_NOWAIT | __GFP_NOWARN); |
31e4c28d | 1280 | INIT_HLIST_HEAD(&blkcg->blkg_list); |
52ebea74 TH |
1281 | #ifdef CONFIG_CGROUP_WRITEBACK |
1282 | INIT_LIST_HEAD(&blkcg->cgwb_list); | |
1283 | #endif | |
7876f930 TH |
1284 | list_add_tail(&blkcg->all_blkcgs_node, &all_blkcgs); |
1285 | ||
1286 | mutex_unlock(&blkcg_pol_mutex); | |
31e4c28d | 1287 | return &blkcg->css; |
e48453c3 AA |
1288 | |
1289 | free_pd_blkcg: | |
1290 | for (i--; i >= 0; i--) | |
e4a9bde9 TH |
1291 | if (blkcg->cpd[i]) |
1292 | blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]); | |
3b8cc629 WL |
1293 | free_percpu(blkcg->lhead); |
1294 | free_blkcg: | |
4c18c9e9 | 1295 | if (blkcg != &blkcg_root) |
1296 | kfree(blkcg); | |
1297 | unlock: | |
7876f930 | 1298 | mutex_unlock(&blkcg_pol_mutex); |
b5a9adcb | 1299 | return ERR_PTR(-ENOMEM); |
31e4c28d VG |
1300 | } |
1301 | ||
4308a434 TH |
1302 | static int blkcg_css_online(struct cgroup_subsys_state *css) |
1303 | { | |
397c9f46 | 1304 | struct blkcg *parent = blkcg_parent(css_to_blkcg(css)); |
4308a434 TH |
1305 | |
1306 | /* | |
1307 | * blkcg_pin_online() is used to delay blkcg offline so that blkgs | |
1308 | * don't go offline while cgwbs are still active on them. Pin the | |
1309 | * parent so that offline always happens towards the root. | |
1310 | */ | |
1311 | if (parent) | |
d7dbd43f | 1312 | blkcg_pin_online(&parent->css); |
4308a434 TH |
1313 | return 0; |
1314 | } | |
1315 | ||
9823538f | 1316 | int blkcg_init_disk(struct gendisk *disk) |
5efd6113 | 1317 | { |
9823538f | 1318 | struct request_queue *q = disk->queue; |
d708f0d5 JA |
1319 | struct blkcg_gq *new_blkg, *blkg; |
1320 | bool preloaded; | |
ec13b1d6 TH |
1321 | int ret; |
1322 | ||
472e4314 | 1323 | INIT_LIST_HEAD(&q->blkg_list); |
1231039d | 1324 | mutex_init(&q->blkcg_mutex); |
472e4314 | 1325 | |
99e60387 | 1326 | new_blkg = blkg_alloc(&blkcg_root, disk, GFP_KERNEL); |
d708f0d5 JA |
1327 | if (!new_blkg) |
1328 | return -ENOMEM; | |
1329 | ||
1330 | preloaded = !radix_tree_preload(GFP_KERNEL); | |
1331 | ||
bea54883 | 1332 | /* Make sure the root blkg exists. */ |
77c570a1 | 1333 | /* spin_lock_irq can serve as RCU read-side critical section. */ |
0d945c1f | 1334 | spin_lock_irq(&q->queue_lock); |
99e60387 | 1335 | blkg = blkg_create(&blkcg_root, disk, new_blkg); |
901932a3 JB |
1336 | if (IS_ERR(blkg)) |
1337 | goto err_unlock; | |
1338 | q->root_blkg = blkg; | |
0d945c1f | 1339 | spin_unlock_irq(&q->queue_lock); |
ec13b1d6 | 1340 | |
d708f0d5 JA |
1341 | if (preloaded) |
1342 | radix_tree_preload_end(); | |
1343 | ||
b0dde3f5 | 1344 | ret = blk_ioprio_init(disk); |
556910e3 BVA |
1345 | if (ret) |
1346 | goto err_destroy_all; | |
1347 | ||
e13793ba | 1348 | ret = blk_throtl_init(disk); |
19688d7f | 1349 | if (ret) |
33dc6279 | 1350 | goto err_ioprio_exit; |
19688d7f | 1351 | |
04be60b5 | 1352 | return 0; |
901932a3 | 1353 | |
33dc6279 | 1354 | err_ioprio_exit: |
b0dde3f5 | 1355 | blk_ioprio_exit(disk); |
04be60b5 | 1356 | err_destroy_all: |
00ad6991 | 1357 | blkg_destroy_all(disk); |
04be60b5 | 1358 | return ret; |
901932a3 | 1359 | err_unlock: |
0d945c1f | 1360 | spin_unlock_irq(&q->queue_lock); |
901932a3 JB |
1361 | if (preloaded) |
1362 | radix_tree_preload_end(); | |
1363 | return PTR_ERR(blkg); | |
5efd6113 TH |
1364 | } |
1365 | ||
9823538f | 1366 | void blkcg_exit_disk(struct gendisk *disk) |
5efd6113 | 1367 | { |
00ad6991 | 1368 | blkg_destroy_all(disk); |
e13793ba | 1369 | blk_throtl_exit(disk); |
5efd6113 TH |
1370 | } |
1371 | ||
d09d8df3 JB |
1372 | static void blkcg_exit(struct task_struct *tsk) |
1373 | { | |
f05837ed CH |
1374 | if (tsk->throttle_disk) |
1375 | put_disk(tsk->throttle_disk); | |
1376 | tsk->throttle_disk = NULL; | |
d09d8df3 JB |
1377 | } |
1378 | ||
c165b3e3 | 1379 | struct cgroup_subsys io_cgrp_subsys = { |
92fb9748 | 1380 | .css_alloc = blkcg_css_alloc, |
4308a434 | 1381 | .css_online = blkcg_css_online, |
92fb9748 TH |
1382 | .css_offline = blkcg_css_offline, |
1383 | .css_free = blkcg_css_free, | |
f7331648 | 1384 | .css_rstat_flush = blkcg_rstat_flush, |
2ee867dc | 1385 | .dfl_cftypes = blkcg_files, |
880f50e2 | 1386 | .legacy_cftypes = blkcg_legacy_files, |
c165b3e3 | 1387 | .legacy_name = "blkio", |
d09d8df3 | 1388 | .exit = blkcg_exit, |
1ced953b TH |
1389 | #ifdef CONFIG_MEMCG |
1390 | /* | |
1391 | * This ensures that, if available, memcg is automatically enabled | |
1392 | * together on the default hierarchy so that the owner cgroup can | |
1393 | * be retrieved from writeback pages. | |
1394 | */ | |
1395 | .depends_on = 1 << memory_cgrp_id, | |
1396 | #endif | |
676f7c8f | 1397 | }; |
c165b3e3 | 1398 | EXPORT_SYMBOL_GPL(io_cgrp_subsys); |
676f7c8f | 1399 | |
a2b1693b | 1400 | /** |
40e4996e CH |
1401 | * blkcg_activate_policy - activate a blkcg policy on a gendisk |
1402 | * @disk: gendisk of interest | |
a2b1693b TH |
1403 | * @pol: blkcg policy to activate |
1404 | * | |
40e4996e | 1405 | * Activate @pol on @disk. Requires %GFP_KERNEL context. @disk goes through |
a2b1693b TH |
1406 | * bypass mode to populate its blkgs with policy_data for @pol. |
1407 | * | |
40e4996e | 1408 | * Activation happens with @disk bypassed, so nobody would be accessing blkgs |
a2b1693b TH |
1409 | * from IO path. Update of each blkg is protected by both queue and blkcg |
1410 | * locks so that holding either lock and testing blkcg_policy_enabled() is | |
1411 | * always enough for dereferencing policy data. | |
1412 | * | |
1413 | * The caller is responsible for synchronizing [de]activations and policy | |
1414 | * [un]registerations. Returns 0 on success, -errno on failure. | |
1415 | */ | |
40e4996e | 1416 | int blkcg_activate_policy(struct gendisk *disk, const struct blkcg_policy *pol) |
a2b1693b | 1417 | { |
40e4996e | 1418 | struct request_queue *q = disk->queue; |
4c55f4f9 | 1419 | struct blkg_policy_data *pd_prealloc = NULL; |
9d179b86 | 1420 | struct blkcg_gq *blkg, *pinned_blkg = NULL; |
4c55f4f9 | 1421 | int ret; |
a2b1693b TH |
1422 | |
1423 | if (blkcg_policy_enabled(q, pol)) | |
1424 | return 0; | |
1425 | ||
344e9ffc | 1426 | if (queue_is_mq(q)) |
bd166ef1 | 1427 | blk_mq_freeze_queue(q); |
9d179b86 | 1428 | retry: |
0d945c1f | 1429 | spin_lock_irq(&q->queue_lock); |
a2b1693b | 1430 | |
9d179b86 | 1431 | /* blkg_list is pushed at the head, reverse walk to allocate parents first */ |
71c81407 | 1432 | list_for_each_entry_reverse(blkg, &q->blkg_list, q_node) { |
4c55f4f9 TH |
1433 | struct blkg_policy_data *pd; |
1434 | ||
1435 | if (blkg->pd[pol->plid]) | |
1436 | continue; | |
a2b1693b | 1437 | |
9d179b86 TH |
1438 | /* If prealloc matches, use it; otherwise try GFP_NOWAIT */ |
1439 | if (blkg == pinned_blkg) { | |
1440 | pd = pd_prealloc; | |
1441 | pd_prealloc = NULL; | |
1442 | } else { | |
0a0b4f79 CH |
1443 | pd = pol->pd_alloc_fn(disk, blkg->blkcg, |
1444 | GFP_NOWAIT | __GFP_NOWARN); | |
9d179b86 TH |
1445 | } |
1446 | ||
4c55f4f9 | 1447 | if (!pd) { |
9d179b86 TH |
1448 | /* |
1449 | * GFP_NOWAIT failed. Free the existing one and | |
1450 | * prealloc for @blkg w/ GFP_KERNEL. | |
1451 | */ | |
1452 | if (pinned_blkg) | |
1453 | blkg_put(pinned_blkg); | |
1454 | blkg_get(blkg); | |
1455 | pinned_blkg = blkg; | |
1456 | ||
0d945c1f | 1457 | spin_unlock_irq(&q->queue_lock); |
9d179b86 TH |
1458 | |
1459 | if (pd_prealloc) | |
1460 | pol->pd_free_fn(pd_prealloc); | |
0a0b4f79 CH |
1461 | pd_prealloc = pol->pd_alloc_fn(disk, blkg->blkcg, |
1462 | GFP_KERNEL); | |
9d179b86 TH |
1463 | if (pd_prealloc) |
1464 | goto retry; | |
1465 | else | |
1466 | goto enomem; | |
4c55f4f9 | 1467 | } |
a2b1693b TH |
1468 | |
1469 | blkg->pd[pol->plid] = pd; | |
1470 | pd->blkg = blkg; | |
b276a876 | 1471 | pd->plid = pol->plid; |
dfd6200a | 1472 | pd->online = false; |
a2b1693b TH |
1473 | } |
1474 | ||
9d179b86 TH |
1475 | /* all allocated, init in the same order */ |
1476 | if (pol->pd_init_fn) | |
1477 | list_for_each_entry_reverse(blkg, &q->blkg_list, q_node) | |
1478 | pol->pd_init_fn(blkg->pd[pol->plid]); | |
1479 | ||
1231039d | 1480 | list_for_each_entry_reverse(blkg, &q->blkg_list, q_node) { |
dfd6200a | 1481 | if (pol->pd_online_fn) |
e3ff8887 | 1482 | pol->pd_online_fn(blkg->pd[pol->plid]); |
dfd6200a YK |
1483 | blkg->pd[pol->plid]->online = true; |
1484 | } | |
e3ff8887 | 1485 | |
a2b1693b TH |
1486 | __set_bit(pol->plid, q->blkcg_pols); |
1487 | ret = 0; | |
4c55f4f9 | 1488 | |
0d945c1f | 1489 | spin_unlock_irq(&q->queue_lock); |
9d179b86 | 1490 | out: |
344e9ffc | 1491 | if (queue_is_mq(q)) |
bd166ef1 | 1492 | blk_mq_unfreeze_queue(q); |
9d179b86 TH |
1493 | if (pinned_blkg) |
1494 | blkg_put(pinned_blkg); | |
001bea73 TH |
1495 | if (pd_prealloc) |
1496 | pol->pd_free_fn(pd_prealloc); | |
a2b1693b | 1497 | return ret; |
9d179b86 TH |
1498 | |
1499 | enomem: | |
1500 | /* alloc failed, nothing's initialized yet, free everything */ | |
1501 | spin_lock_irq(&q->queue_lock); | |
1502 | list_for_each_entry(blkg, &q->blkg_list, q_node) { | |
858560b2 LJ |
1503 | struct blkcg *blkcg = blkg->blkcg; |
1504 | ||
1505 | spin_lock(&blkcg->lock); | |
9d179b86 TH |
1506 | if (blkg->pd[pol->plid]) { |
1507 | pol->pd_free_fn(blkg->pd[pol->plid]); | |
1508 | blkg->pd[pol->plid] = NULL; | |
1509 | } | |
858560b2 | 1510 | spin_unlock(&blkcg->lock); |
9d179b86 TH |
1511 | } |
1512 | spin_unlock_irq(&q->queue_lock); | |
1513 | ret = -ENOMEM; | |
1514 | goto out; | |
a2b1693b TH |
1515 | } |
1516 | EXPORT_SYMBOL_GPL(blkcg_activate_policy); | |
1517 | ||
1518 | /** | |
40e4996e CH |
1519 | * blkcg_deactivate_policy - deactivate a blkcg policy on a gendisk |
1520 | * @disk: gendisk of interest | |
a2b1693b TH |
1521 | * @pol: blkcg policy to deactivate |
1522 | * | |
40e4996e | 1523 | * Deactivate @pol on @disk. Follows the same synchronization rules as |
a2b1693b TH |
1524 | * blkcg_activate_policy(). |
1525 | */ | |
40e4996e | 1526 | void blkcg_deactivate_policy(struct gendisk *disk, |
3c798398 | 1527 | const struct blkcg_policy *pol) |
a2b1693b | 1528 | { |
40e4996e | 1529 | struct request_queue *q = disk->queue; |
3c798398 | 1530 | struct blkcg_gq *blkg; |
a2b1693b TH |
1531 | |
1532 | if (!blkcg_policy_enabled(q, pol)) | |
1533 | return; | |
1534 | ||
344e9ffc | 1535 | if (queue_is_mq(q)) |
bd166ef1 | 1536 | blk_mq_freeze_queue(q); |
bd166ef1 | 1537 | |
1231039d | 1538 | mutex_lock(&q->blkcg_mutex); |
0d945c1f | 1539 | spin_lock_irq(&q->queue_lock); |
a2b1693b TH |
1540 | |
1541 | __clear_bit(pol->plid, q->blkcg_pols); | |
1542 | ||
1543 | list_for_each_entry(blkg, &q->blkg_list, q_node) { | |
858560b2 LJ |
1544 | struct blkcg *blkcg = blkg->blkcg; |
1545 | ||
1546 | spin_lock(&blkcg->lock); | |
001bea73 | 1547 | if (blkg->pd[pol->plid]) { |
dfd6200a | 1548 | if (blkg->pd[pol->plid]->online && pol->pd_offline_fn) |
a9520cd6 | 1549 | pol->pd_offline_fn(blkg->pd[pol->plid]); |
001bea73 TH |
1550 | pol->pd_free_fn(blkg->pd[pol->plid]); |
1551 | blkg->pd[pol->plid] = NULL; | |
1552 | } | |
858560b2 | 1553 | spin_unlock(&blkcg->lock); |
a2b1693b TH |
1554 | } |
1555 | ||
0d945c1f | 1556 | spin_unlock_irq(&q->queue_lock); |
1231039d | 1557 | mutex_unlock(&q->blkcg_mutex); |
bd166ef1 | 1558 | |
344e9ffc | 1559 | if (queue_is_mq(q)) |
bd166ef1 | 1560 | blk_mq_unfreeze_queue(q); |
a2b1693b TH |
1561 | } |
1562 | EXPORT_SYMBOL_GPL(blkcg_deactivate_policy); | |
1563 | ||
e55cf798 JY |
1564 | static void blkcg_free_all_cpd(struct blkcg_policy *pol) |
1565 | { | |
1566 | struct blkcg *blkcg; | |
1567 | ||
1568 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) { | |
1569 | if (blkcg->cpd[pol->plid]) { | |
1570 | pol->cpd_free_fn(blkcg->cpd[pol->plid]); | |
1571 | blkcg->cpd[pol->plid] = NULL; | |
1572 | } | |
1573 | } | |
1574 | } | |
1575 | ||
8bd435b3 | 1576 | /** |
3c798398 TH |
1577 | * blkcg_policy_register - register a blkcg policy |
1578 | * @pol: blkcg policy to register | |
8bd435b3 | 1579 | * |
3c798398 TH |
1580 | * Register @pol with blkcg core. Might sleep and @pol may be modified on |
1581 | * successful registration. Returns 0 on success and -errno on failure. | |
8bd435b3 | 1582 | */ |
d5bf0291 | 1583 | int blkcg_policy_register(struct blkcg_policy *pol) |
3e252066 | 1584 | { |
06b285bd | 1585 | struct blkcg *blkcg; |
8bd435b3 | 1586 | int i, ret; |
e8989fae | 1587 | |
838f13bf | 1588 | mutex_lock(&blkcg_pol_register_mutex); |
bc0d6501 TH |
1589 | mutex_lock(&blkcg_pol_mutex); |
1590 | ||
8bd435b3 TH |
1591 | /* find an empty slot */ |
1592 | ret = -ENOSPC; | |
1593 | for (i = 0; i < BLKCG_MAX_POLS; i++) | |
3c798398 | 1594 | if (!blkcg_policy[i]) |
8bd435b3 | 1595 | break; |
01c5f85a JA |
1596 | if (i >= BLKCG_MAX_POLS) { |
1597 | pr_warn("blkcg_policy_register: BLKCG_MAX_POLS too small\n"); | |
838f13bf | 1598 | goto err_unlock; |
01c5f85a | 1599 | } |
035d10b2 | 1600 | |
e8401073 | 1601 | /* Make sure cpd/pd_alloc_fn and cpd/pd_free_fn in pairs */ |
1602 | if ((!pol->cpd_alloc_fn ^ !pol->cpd_free_fn) || | |
1603 | (!pol->pd_alloc_fn ^ !pol->pd_free_fn)) | |
1604 | goto err_unlock; | |
1605 | ||
06b285bd | 1606 | /* register @pol */ |
3c798398 | 1607 | pol->plid = i; |
06b285bd TH |
1608 | blkcg_policy[pol->plid] = pol; |
1609 | ||
1610 | /* allocate and install cpd's */ | |
e4a9bde9 | 1611 | if (pol->cpd_alloc_fn) { |
06b285bd TH |
1612 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) { |
1613 | struct blkcg_policy_data *cpd; | |
1614 | ||
e4a9bde9 | 1615 | cpd = pol->cpd_alloc_fn(GFP_KERNEL); |
bbb427e3 | 1616 | if (!cpd) |
06b285bd | 1617 | goto err_free_cpds; |
06b285bd | 1618 | |
81437648 TH |
1619 | blkcg->cpd[pol->plid] = cpd; |
1620 | cpd->blkcg = blkcg; | |
06b285bd | 1621 | cpd->plid = pol->plid; |
06b285bd TH |
1622 | } |
1623 | } | |
1624 | ||
838f13bf | 1625 | mutex_unlock(&blkcg_pol_mutex); |
8bd435b3 | 1626 | |
8bd435b3 | 1627 | /* everything is in place, add intf files for the new policy */ |
2ee867dc TH |
1628 | if (pol->dfl_cftypes) |
1629 | WARN_ON(cgroup_add_dfl_cftypes(&io_cgrp_subsys, | |
1630 | pol->dfl_cftypes)); | |
880f50e2 | 1631 | if (pol->legacy_cftypes) |
c165b3e3 | 1632 | WARN_ON(cgroup_add_legacy_cftypes(&io_cgrp_subsys, |
880f50e2 | 1633 | pol->legacy_cftypes)); |
838f13bf TH |
1634 | mutex_unlock(&blkcg_pol_register_mutex); |
1635 | return 0; | |
1636 | ||
06b285bd | 1637 | err_free_cpds: |
e55cf798 JY |
1638 | if (pol->cpd_free_fn) |
1639 | blkcg_free_all_cpd(pol); | |
1640 | ||
06b285bd | 1641 | blkcg_policy[pol->plid] = NULL; |
838f13bf | 1642 | err_unlock: |
bc0d6501 | 1643 | mutex_unlock(&blkcg_pol_mutex); |
838f13bf | 1644 | mutex_unlock(&blkcg_pol_register_mutex); |
8bd435b3 | 1645 | return ret; |
3e252066 | 1646 | } |
3c798398 | 1647 | EXPORT_SYMBOL_GPL(blkcg_policy_register); |
3e252066 | 1648 | |
8bd435b3 | 1649 | /** |
3c798398 TH |
1650 | * blkcg_policy_unregister - unregister a blkcg policy |
1651 | * @pol: blkcg policy to unregister | |
8bd435b3 | 1652 | * |
3c798398 | 1653 | * Undo blkcg_policy_register(@pol). Might sleep. |
8bd435b3 | 1654 | */ |
3c798398 | 1655 | void blkcg_policy_unregister(struct blkcg_policy *pol) |
3e252066 | 1656 | { |
838f13bf | 1657 | mutex_lock(&blkcg_pol_register_mutex); |
bc0d6501 | 1658 | |
3c798398 | 1659 | if (WARN_ON(blkcg_policy[pol->plid] != pol)) |
8bd435b3 TH |
1660 | goto out_unlock; |
1661 | ||
1662 | /* kill the intf files first */ | |
2ee867dc TH |
1663 | if (pol->dfl_cftypes) |
1664 | cgroup_rm_cftypes(pol->dfl_cftypes); | |
880f50e2 TH |
1665 | if (pol->legacy_cftypes) |
1666 | cgroup_rm_cftypes(pol->legacy_cftypes); | |
44ea53de | 1667 | |
06b285bd | 1668 | /* remove cpds and unregister */ |
838f13bf | 1669 | mutex_lock(&blkcg_pol_mutex); |
06b285bd | 1670 | |
e55cf798 JY |
1671 | if (pol->cpd_free_fn) |
1672 | blkcg_free_all_cpd(pol); | |
1673 | ||
3c798398 | 1674 | blkcg_policy[pol->plid] = NULL; |
06b285bd | 1675 | |
bc0d6501 | 1676 | mutex_unlock(&blkcg_pol_mutex); |
838f13bf TH |
1677 | out_unlock: |
1678 | mutex_unlock(&blkcg_pol_register_mutex); | |
3e252066 | 1679 | } |
3c798398 | 1680 | EXPORT_SYMBOL_GPL(blkcg_policy_unregister); |
903d23f0 | 1681 | |
d3f77dfd TH |
1682 | bool __blkcg_punt_bio_submit(struct bio *bio) |
1683 | { | |
1684 | struct blkcg_gq *blkg = bio->bi_blkg; | |
1685 | ||
1686 | /* consume the flag first */ | |
1687 | bio->bi_opf &= ~REQ_CGROUP_PUNT; | |
1688 | ||
1689 | /* never bounce for the root cgroup */ | |
1690 | if (!blkg->parent) | |
1691 | return false; | |
1692 | ||
1693 | spin_lock_bh(&blkg->async_bio_lock); | |
1694 | bio_list_add(&blkg->async_bios, bio); | |
1695 | spin_unlock_bh(&blkg->async_bio_lock); | |
1696 | ||
1697 | queue_work(blkcg_punt_bio_wq, &blkg->async_bio_work); | |
1698 | return true; | |
1699 | } | |
1700 | ||
d09d8df3 JB |
1701 | /* |
1702 | * Scale the accumulated delay based on how long it has been since we updated | |
1703 | * the delay. We only call this when we are adding delay, in case it's been a | |
1704 | * while since we added delay, and when we are checking to see if we need to | |
1705 | * delay a task, to account for any delays that may have occurred. | |
1706 | */ | |
1707 | static void blkcg_scale_delay(struct blkcg_gq *blkg, u64 now) | |
1708 | { | |
1709 | u64 old = atomic64_read(&blkg->delay_start); | |
1710 | ||
54c52e10 TH |
1711 | /* negative use_delay means no scaling, see blkcg_set_delay() */ |
1712 | if (atomic_read(&blkg->use_delay) < 0) | |
1713 | return; | |
1714 | ||
d09d8df3 JB |
1715 | /* |
1716 | * We only want to scale down every second. The idea here is that we | |
1717 | * want to delay people for min(delay_nsec, NSEC_PER_SEC) in a certain | |
1718 | * time window. We only want to throttle tasks for recent delay that | |
1719 | * has occurred, in 1 second time windows since that's the maximum | |
1720 | * things can be throttled. We save the current delay window in | |
1721 | * blkg->last_delay so we know what amount is still left to be charged | |
1722 | * to the blkg from this point onward. blkg->last_use keeps track of | |
1723 | * the use_delay counter. The idea is if we're unthrottling the blkg we | |
1724 | * are ok with whatever is happening now, and we can take away more of | |
1725 | * the accumulated delay as we've already throttled enough that | |
1726 | * everybody is happy with their IO latencies. | |
1727 | */ | |
1728 | if (time_before64(old + NSEC_PER_SEC, now) && | |
96388f57 | 1729 | atomic64_try_cmpxchg(&blkg->delay_start, &old, now)) { |
d09d8df3 JB |
1730 | u64 cur = atomic64_read(&blkg->delay_nsec); |
1731 | u64 sub = min_t(u64, blkg->last_delay, now - old); | |
1732 | int cur_use = atomic_read(&blkg->use_delay); | |
1733 | ||
1734 | /* | |
1735 | * We've been unthrottled, subtract a larger chunk of our | |
1736 | * accumulated delay. | |
1737 | */ | |
1738 | if (cur_use < blkg->last_use) | |
1739 | sub = max_t(u64, sub, blkg->last_delay >> 1); | |
1740 | ||
1741 | /* | |
1742 | * This shouldn't happen, but handle it anyway. Our delay_nsec | |
1743 | * should only ever be growing except here where we subtract out | |
1744 | * min(last_delay, 1 second), but lord knows bugs happen and I'd | |
1745 | * rather not end up with negative numbers. | |
1746 | */ | |
1747 | if (unlikely(cur < sub)) { | |
1748 | atomic64_set(&blkg->delay_nsec, 0); | |
1749 | blkg->last_delay = 0; | |
1750 | } else { | |
1751 | atomic64_sub(sub, &blkg->delay_nsec); | |
1752 | blkg->last_delay = cur - sub; | |
1753 | } | |
1754 | blkg->last_use = cur_use; | |
1755 | } | |
1756 | } | |
1757 | ||
1758 | /* | |
1759 | * This is called when we want to actually walk up the hierarchy and check to | |
1760 | * see if we need to throttle, and then actually throttle if there is some | |
1761 | * accumulated delay. This should only be called upon return to user space so | |
1762 | * we're not holding some lock that would induce a priority inversion. | |
1763 | */ | |
1764 | static void blkcg_maybe_throttle_blkg(struct blkcg_gq *blkg, bool use_memdelay) | |
1765 | { | |
fd112c74 | 1766 | unsigned long pflags; |
5160a5a5 | 1767 | bool clamp; |
d09d8df3 JB |
1768 | u64 now = ktime_to_ns(ktime_get()); |
1769 | u64 exp; | |
1770 | u64 delay_nsec = 0; | |
1771 | int tok; | |
1772 | ||
1773 | while (blkg->parent) { | |
5160a5a5 TH |
1774 | int use_delay = atomic_read(&blkg->use_delay); |
1775 | ||
1776 | if (use_delay) { | |
1777 | u64 this_delay; | |
1778 | ||
d09d8df3 | 1779 | blkcg_scale_delay(blkg, now); |
5160a5a5 TH |
1780 | this_delay = atomic64_read(&blkg->delay_nsec); |
1781 | if (this_delay > delay_nsec) { | |
1782 | delay_nsec = this_delay; | |
1783 | clamp = use_delay > 0; | |
1784 | } | |
d09d8df3 JB |
1785 | } |
1786 | blkg = blkg->parent; | |
1787 | } | |
1788 | ||
1789 | if (!delay_nsec) | |
1790 | return; | |
1791 | ||
1792 | /* | |
1793 | * Let's not sleep for all eternity if we've amassed a huge delay. | |
1794 | * Swapping or metadata IO can accumulate 10's of seconds worth of | |
1795 | * delay, and we want userspace to be able to do _something_ so cap the | |
5160a5a5 TH |
1796 | * delays at 0.25s. If there's 10's of seconds worth of delay then the |
1797 | * tasks will be delayed for 0.25 second for every syscall. If | |
1798 | * blkcg_set_delay() was used as indicated by negative use_delay, the | |
1799 | * caller is responsible for regulating the range. | |
d09d8df3 | 1800 | */ |
5160a5a5 TH |
1801 | if (clamp) |
1802 | delay_nsec = min_t(u64, delay_nsec, 250 * NSEC_PER_MSEC); | |
d09d8df3 | 1803 | |
fd112c74 JB |
1804 | if (use_memdelay) |
1805 | psi_memstall_enter(&pflags); | |
d09d8df3 JB |
1806 | |
1807 | exp = ktime_add_ns(now, delay_nsec); | |
1808 | tok = io_schedule_prepare(); | |
1809 | do { | |
1810 | __set_current_state(TASK_KILLABLE); | |
1811 | if (!schedule_hrtimeout(&exp, HRTIMER_MODE_ABS)) | |
1812 | break; | |
1813 | } while (!fatal_signal_pending(current)); | |
1814 | io_schedule_finish(tok); | |
fd112c74 JB |
1815 | |
1816 | if (use_memdelay) | |
1817 | psi_memstall_leave(&pflags); | |
d09d8df3 JB |
1818 | } |
1819 | ||
1820 | /** | |
1821 | * blkcg_maybe_throttle_current - throttle the current task if it has been marked | |
1822 | * | |
1823 | * This is only called if we've been marked with set_notify_resume(). Obviously | |
1824 | * we can be set_notify_resume() for reasons other than blkcg throttling, so we | |
f05837ed | 1825 | * check to see if current->throttle_disk is set and if not this doesn't do |
d09d8df3 JB |
1826 | * anything. This should only ever be called by the resume code, it's not meant |
1827 | * to be called by people willy-nilly as it will actually do the work to | |
1828 | * throttle the task if it is setup for throttling. | |
1829 | */ | |
1830 | void blkcg_maybe_throttle_current(void) | |
1831 | { | |
f05837ed | 1832 | struct gendisk *disk = current->throttle_disk; |
d09d8df3 JB |
1833 | struct blkcg *blkcg; |
1834 | struct blkcg_gq *blkg; | |
1835 | bool use_memdelay = current->use_memdelay; | |
1836 | ||
f05837ed | 1837 | if (!disk) |
d09d8df3 JB |
1838 | return; |
1839 | ||
f05837ed | 1840 | current->throttle_disk = NULL; |
d09d8df3 JB |
1841 | current->use_memdelay = false; |
1842 | ||
1843 | rcu_read_lock(); | |
82778259 | 1844 | blkcg = css_to_blkcg(blkcg_css()); |
d09d8df3 JB |
1845 | if (!blkcg) |
1846 | goto out; | |
9a9c261e | 1847 | blkg = blkg_lookup(blkcg, disk->queue); |
d09d8df3 JB |
1848 | if (!blkg) |
1849 | goto out; | |
7754f669 | 1850 | if (!blkg_tryget(blkg)) |
d09d8df3 JB |
1851 | goto out; |
1852 | rcu_read_unlock(); | |
d09d8df3 JB |
1853 | |
1854 | blkcg_maybe_throttle_blkg(blkg, use_memdelay); | |
1855 | blkg_put(blkg); | |
f05837ed | 1856 | put_disk(disk); |
d09d8df3 JB |
1857 | return; |
1858 | out: | |
1859 | rcu_read_unlock(); | |
d09d8df3 | 1860 | } |
d09d8df3 JB |
1861 | |
1862 | /** | |
1863 | * blkcg_schedule_throttle - this task needs to check for throttling | |
1d6df9d3 | 1864 | * @disk: disk to throttle |
537d71b3 | 1865 | * @use_memdelay: do we charge this to memory delay for PSI |
d09d8df3 JB |
1866 | * |
1867 | * This is called by the IO controller when we know there's delay accumulated | |
1868 | * for the blkg for this task. We do not pass the blkg because there are places | |
1869 | * we call this that may not have that information, the swapping code for | |
de185b56 | 1870 | * instance will only have a block_device at that point. This set's the |
d09d8df3 JB |
1871 | * notify_resume for the task to check and see if it requires throttling before |
1872 | * returning to user space. | |
1873 | * | |
1874 | * We will only schedule once per syscall. You can call this over and over | |
1875 | * again and it will only do the check once upon return to user space, and only | |
1876 | * throttle once. If the task needs to be throttled again it'll need to be | |
1877 | * re-set at the next time we see the task. | |
1878 | */ | |
de185b56 | 1879 | void blkcg_schedule_throttle(struct gendisk *disk, bool use_memdelay) |
d09d8df3 JB |
1880 | { |
1881 | if (unlikely(current->flags & PF_KTHREAD)) | |
1882 | return; | |
1883 | ||
f05837ed CH |
1884 | if (current->throttle_disk != disk) { |
1885 | if (test_bit(GD_DEAD, &disk->state)) | |
49d1822b | 1886 | return; |
f05837ed | 1887 | get_device(disk_to_dev(disk)); |
49d1822b | 1888 | |
f05837ed CH |
1889 | if (current->throttle_disk) |
1890 | put_disk(current->throttle_disk); | |
1891 | current->throttle_disk = disk; | |
49d1822b | 1892 | } |
d09d8df3 | 1893 | |
d09d8df3 JB |
1894 | if (use_memdelay) |
1895 | current->use_memdelay = use_memdelay; | |
1896 | set_notify_resume(current); | |
1897 | } | |
d09d8df3 JB |
1898 | |
1899 | /** | |
1900 | * blkcg_add_delay - add delay to this blkg | |
537d71b3 BVA |
1901 | * @blkg: blkg of interest |
1902 | * @now: the current time in nanoseconds | |
1903 | * @delta: how many nanoseconds of delay to add | |
d09d8df3 JB |
1904 | * |
1905 | * Charge @delta to the blkg's current delay accumulation. This is used to | |
1906 | * throttle tasks if an IO controller thinks we need more throttling. | |
1907 | */ | |
1908 | void blkcg_add_delay(struct blkcg_gq *blkg, u64 now, u64 delta) | |
1909 | { | |
54c52e10 TH |
1910 | if (WARN_ON_ONCE(atomic_read(&blkg->use_delay) < 0)) |
1911 | return; | |
d09d8df3 JB |
1912 | blkcg_scale_delay(blkg, now); |
1913 | atomic64_add(delta, &blkg->delay_nsec); | |
1914 | } | |
d09d8df3 | 1915 | |
28fc591f CH |
1916 | /** |
1917 | * blkg_tryget_closest - try and get a blkg ref on the closet blkg | |
13c7863d CH |
1918 | * @bio: target bio |
1919 | * @css: target css | |
28fc591f | 1920 | * |
13c7863d CH |
1921 | * As the failure mode here is to walk up the blkg tree, this ensure that the |
1922 | * blkg->parent pointers are always valid. This returns the blkg that it ended | |
1923 | * up taking a reference on or %NULL if no reference was taken. | |
28fc591f | 1924 | */ |
13c7863d CH |
1925 | static inline struct blkcg_gq *blkg_tryget_closest(struct bio *bio, |
1926 | struct cgroup_subsys_state *css) | |
28fc591f | 1927 | { |
13c7863d | 1928 | struct blkcg_gq *blkg, *ret_blkg = NULL; |
28fc591f | 1929 | |
13c7863d | 1930 | rcu_read_lock(); |
99e60387 | 1931 | blkg = blkg_lookup_create(css_to_blkcg(css), bio->bi_bdev->bd_disk); |
28fc591f CH |
1932 | while (blkg) { |
1933 | if (blkg_tryget(blkg)) { | |
1934 | ret_blkg = blkg; | |
1935 | break; | |
1936 | } | |
1937 | blkg = blkg->parent; | |
1938 | } | |
13c7863d | 1939 | rcu_read_unlock(); |
28fc591f CH |
1940 | |
1941 | return ret_blkg; | |
1942 | } | |
1943 | ||
1944 | /** | |
1945 | * bio_associate_blkg_from_css - associate a bio with a specified css | |
1946 | * @bio: target bio | |
1947 | * @css: target css | |
1948 | * | |
1949 | * Associate @bio with the blkg found by combining the css's blkg and the | |
1950 | * request_queue of the @bio. An association failure is handled by walking up | |
1951 | * the blkg tree. Therefore, the blkg associated can be anything between @blkg | |
1952 | * and q->root_blkg. This situation only happens when a cgroup is dying and | |
1953 | * then the remaining bios will spill to the closest alive blkg. | |
1954 | * | |
1955 | * A reference will be taken on the blkg and will be released when @bio is | |
1956 | * freed. | |
1957 | */ | |
1958 | void bio_associate_blkg_from_css(struct bio *bio, | |
1959 | struct cgroup_subsys_state *css) | |
1960 | { | |
28fc591f CH |
1961 | if (bio->bi_blkg) |
1962 | blkg_put(bio->bi_blkg); | |
1963 | ||
a5b97526 | 1964 | if (css && css->parent) { |
13c7863d | 1965 | bio->bi_blkg = blkg_tryget_closest(bio, css); |
a5b97526 | 1966 | } else { |
ed6cddef PB |
1967 | blkg_get(bdev_get_queue(bio->bi_bdev)->root_blkg); |
1968 | bio->bi_blkg = bdev_get_queue(bio->bi_bdev)->root_blkg; | |
a5b97526 | 1969 | } |
28fc591f CH |
1970 | } |
1971 | EXPORT_SYMBOL_GPL(bio_associate_blkg_from_css); | |
1972 | ||
1973 | /** | |
1974 | * bio_associate_blkg - associate a bio with a blkg | |
1975 | * @bio: target bio | |
1976 | * | |
1977 | * Associate @bio with the blkg found from the bio's css and request_queue. | |
1978 | * If one is not found, bio_lookup_blkg() creates the blkg. If a blkg is | |
1979 | * already associated, the css is reused and association redone as the | |
1980 | * request_queue may have changed. | |
1981 | */ | |
1982 | void bio_associate_blkg(struct bio *bio) | |
1983 | { | |
1984 | struct cgroup_subsys_state *css; | |
1985 | ||
1986 | rcu_read_lock(); | |
1987 | ||
1988 | if (bio->bi_blkg) | |
bbb1ebe7 | 1989 | css = bio_blkcg_css(bio); |
28fc591f CH |
1990 | else |
1991 | css = blkcg_css(); | |
1992 | ||
1993 | bio_associate_blkg_from_css(bio, css); | |
1994 | ||
1995 | rcu_read_unlock(); | |
1996 | } | |
1997 | EXPORT_SYMBOL_GPL(bio_associate_blkg); | |
1998 | ||
1999 | /** | |
2000 | * bio_clone_blkg_association - clone blkg association from src to dst bio | |
2001 | * @dst: destination bio | |
2002 | * @src: source bio | |
2003 | */ | |
2004 | void bio_clone_blkg_association(struct bio *dst, struct bio *src) | |
2005 | { | |
22b106e5 JK |
2006 | if (src->bi_blkg) |
2007 | bio_associate_blkg_from_css(dst, bio_blkcg_css(src)); | |
28fc591f CH |
2008 | } |
2009 | EXPORT_SYMBOL_GPL(bio_clone_blkg_association); | |
2010 | ||
db18a53e CH |
2011 | static int blk_cgroup_io_type(struct bio *bio) |
2012 | { | |
2013 | if (op_is_discard(bio->bi_opf)) | |
2014 | return BLKG_IOSTAT_DISCARD; | |
2015 | if (op_is_write(bio->bi_opf)) | |
2016 | return BLKG_IOSTAT_WRITE; | |
2017 | return BLKG_IOSTAT_READ; | |
2018 | } | |
2019 | ||
2020 | void blk_cgroup_bio_start(struct bio *bio) | |
2021 | { | |
3b8cc629 | 2022 | struct blkcg *blkcg = bio->bi_blkg->blkcg; |
db18a53e CH |
2023 | int rwd = blk_cgroup_io_type(bio), cpu; |
2024 | struct blkg_iostat_set *bis; | |
3c08b093 | 2025 | unsigned long flags; |
db18a53e | 2026 | |
0416f3be ML |
2027 | /* Root-level stats are sourced from system-wide IO stats */ |
2028 | if (!cgroup_parent(blkcg->css.cgroup)) | |
2029 | return; | |
2030 | ||
db18a53e CH |
2031 | cpu = get_cpu(); |
2032 | bis = per_cpu_ptr(bio->bi_blkg->iostat_cpu, cpu); | |
3c08b093 | 2033 | flags = u64_stats_update_begin_irqsave(&bis->sync); |
db18a53e CH |
2034 | |
2035 | /* | |
2036 | * If the bio is flagged with BIO_CGROUP_ACCT it means this is a split | |
2037 | * bio and we would have already accounted for the size of the bio. | |
2038 | */ | |
2039 | if (!bio_flagged(bio, BIO_CGROUP_ACCT)) { | |
2040 | bio_set_flag(bio, BIO_CGROUP_ACCT); | |
0b8cc25d | 2041 | bis->cur.bytes[rwd] += bio->bi_iter.bi_size; |
db18a53e CH |
2042 | } |
2043 | bis->cur.ios[rwd]++; | |
2044 | ||
3b8cc629 WL |
2045 | /* |
2046 | * If the iostat_cpu isn't in a lockless list, put it into the | |
2047 | * list to indicate that a stat update is pending. | |
2048 | */ | |
2049 | if (!READ_ONCE(bis->lqueued)) { | |
2050 | struct llist_head *lhead = this_cpu_ptr(blkcg->lhead); | |
2051 | ||
2052 | llist_add(&bis->lnode, lhead); | |
2053 | WRITE_ONCE(bis->lqueued, true); | |
2054 | percpu_ref_get(&bis->blkg->refcnt); | |
2055 | } | |
2056 | ||
3c08b093 | 2057 | u64_stats_update_end_irqrestore(&bis->sync, flags); |
db18a53e | 2058 | if (cgroup_subsys_on_dfl(io_cgrp_subsys)) |
3b8cc629 | 2059 | cgroup_rstat_updated(blkcg->css.cgroup, cpu); |
db18a53e CH |
2060 | put_cpu(); |
2061 | } | |
2062 | ||
216889aa CH |
2063 | bool blk_cgroup_congested(void) |
2064 | { | |
2065 | struct cgroup_subsys_state *css; | |
2066 | bool ret = false; | |
2067 | ||
2068 | rcu_read_lock(); | |
d200ca14 | 2069 | for (css = blkcg_css(); css; css = css->parent) { |
216889aa CH |
2070 | if (atomic_read(&css->cgroup->congestion_count)) { |
2071 | ret = true; | |
2072 | break; | |
2073 | } | |
216889aa CH |
2074 | } |
2075 | rcu_read_unlock(); | |
2076 | return ret; | |
2077 | } | |
2078 | ||
d3f77dfd TH |
2079 | static int __init blkcg_init(void) |
2080 | { | |
2081 | blkcg_punt_bio_wq = alloc_workqueue("blkcg_punt_bio", | |
2082 | WQ_MEM_RECLAIM | WQ_FREEZABLE | | |
2083 | WQ_UNBOUND | WQ_SYSFS, 0); | |
2084 | if (!blkcg_punt_bio_wq) | |
2085 | return -ENOMEM; | |
2086 | return 0; | |
2087 | } | |
2088 | subsys_initcall(blkcg_init); | |
2089 | ||
903d23f0 JB |
2090 | module_param(blkcg_debug_stats, bool, 0644); |
2091 | MODULE_PARM_DESC(blkcg_debug_stats, "True if you want debug stats, false if not"); |