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" |
813e6930 | 36 | #include "blk-rq-qos.h" |
3e252066 | 37 | |
838f13bf TH |
38 | /* |
39 | * blkcg_pol_mutex protects blkcg_policy[] and policy [de]activation. | |
40 | * blkcg_pol_register_mutex nests outside of it and synchronizes entire | |
41 | * policy [un]register operations including cgroup file additions / | |
42 | * removals. Putting cgroup file registration outside blkcg_pol_mutex | |
43 | * allows grabbing it from cgroup callbacks. | |
44 | */ | |
45 | static DEFINE_MUTEX(blkcg_pol_register_mutex); | |
bc0d6501 | 46 | static DEFINE_MUTEX(blkcg_pol_mutex); |
923adde1 | 47 | |
e48453c3 | 48 | struct blkcg blkcg_root; |
3c798398 | 49 | EXPORT_SYMBOL_GPL(blkcg_root); |
9d6a986c | 50 | |
496d5e75 | 51 | struct cgroup_subsys_state * const blkcg_root_css = &blkcg_root.css; |
9b0eb69b | 52 | EXPORT_SYMBOL_GPL(blkcg_root_css); |
496d5e75 | 53 | |
3c798398 | 54 | static struct blkcg_policy *blkcg_policy[BLKCG_MAX_POLS]; |
035d10b2 | 55 | |
7876f930 TH |
56 | static LIST_HEAD(all_blkcgs); /* protected by blkcg_pol_mutex */ |
57 | ||
07b0fdec | 58 | bool blkcg_debug_stats = false; |
d3f77dfd | 59 | static struct workqueue_struct *blkcg_punt_bio_wq; |
903d23f0 | 60 | |
a731763f YK |
61 | #define BLKG_DESTROY_BATCH_SIZE 64 |
62 | ||
3b8cc629 WL |
63 | /* |
64 | * Lockless lists for tracking IO stats update | |
65 | * | |
66 | * New IO stats are stored in the percpu iostat_cpu within blkcg_gq (blkg). | |
67 | * There are multiple blkg's (one for each block device) attached to each | |
68 | * blkcg. The rstat code keeps track of which cpu has IO stats updated, | |
69 | * but it doesn't know which blkg has the updated stats. If there are many | |
70 | * block devices in a system, the cost of iterating all the blkg's to flush | |
71 | * out the IO stats can be high. To reduce such overhead, a set of percpu | |
72 | * lockless lists (lhead) per blkcg are used to track the set of recently | |
73 | * updated iostat_cpu's since the last flush. An iostat_cpu will be put | |
74 | * onto the lockless list on the update side [blk_cgroup_bio_start()] if | |
75 | * not there yet and then removed when being flushed [blkcg_rstat_flush()]. | |
76 | * References to blkg are gotten and then put back in the process to | |
77 | * protect against blkg removal. | |
78 | * | |
79 | * Return: 0 if successful or -ENOMEM if allocation fails. | |
80 | */ | |
81 | static int init_blkcg_llists(struct blkcg *blkcg) | |
82 | { | |
83 | int cpu; | |
84 | ||
85 | blkcg->lhead = alloc_percpu_gfp(struct llist_head, GFP_KERNEL); | |
86 | if (!blkcg->lhead) | |
87 | return -ENOMEM; | |
88 | ||
89 | for_each_possible_cpu(cpu) | |
90 | init_llist_head(per_cpu_ptr(blkcg->lhead, cpu)); | |
91 | return 0; | |
92 | } | |
93 | ||
bc5fee91 CH |
94 | /** |
95 | * blkcg_css - find the current css | |
96 | * | |
97 | * Find the css associated with either the kthread or the current task. | |
98 | * This may return a dying css, so it is up to the caller to use tryget logic | |
99 | * to confirm it is alive and well. | |
100 | */ | |
101 | static struct cgroup_subsys_state *blkcg_css(void) | |
102 | { | |
103 | struct cgroup_subsys_state *css; | |
104 | ||
105 | css = kthread_blkcg(); | |
106 | if (css) | |
107 | return css; | |
108 | return task_css(current, io_cgrp_id); | |
109 | } | |
110 | ||
1231039d | 111 | static bool blkcg_policy_enabled(struct request_queue *q, |
3c798398 | 112 | const struct blkcg_policy *pol) |
a2b1693b | 113 | { |
1231039d | 114 | return pol && test_bit(pol->plid, q->blkcg_pols); |
a2b1693b TH |
115 | } |
116 | ||
dcb52201 | 117 | static void blkg_free_workfn(struct work_struct *work) |
0381411e | 118 | { |
dcb52201 CH |
119 | struct blkcg_gq *blkg = container_of(work, struct blkcg_gq, |
120 | free_work); | |
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 | */ | |
1231039d | 130 | mutex_lock(&blkg->disk->queue->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 CH |
136 | list_del_init(&blkg->q_node); |
137 | mutex_unlock(&blkg->disk->queue->blkcg_mutex); | |
f1c006f1 | 138 | |
84d7d462 | 139 | put_disk(blkg->disk); |
f7331648 | 140 | free_percpu(blkg->iostat_cpu); |
ef069b97 | 141 | percpu_ref_exit(&blkg->refcnt); |
549d3aa8 | 142 | kfree(blkg); |
0381411e TH |
143 | } |
144 | ||
dcb52201 CH |
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) | |
d578c770 | 152 | { |
dcb52201 CH |
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); | |
d578c770 ML |
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); | |
dcb52201 | 172 | blkg_free(blkg); |
7fcf2b03 DZ |
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; |
84d7d462 CH |
266 | |
267 | if (test_bit(GD_DEAD, &disk->state)) | |
0b6f93bd | 268 | goto out_free_iostat; |
84d7d462 CH |
269 | get_device(disk_to_dev(disk)); |
270 | blkg->disk = disk; | |
0a9a25ca | 271 | |
1231039d | 272 | INIT_LIST_HEAD(&blkg->q_node); |
d3f77dfd TH |
273 | spin_lock_init(&blkg->async_bio_lock); |
274 | bio_list_init(&blkg->async_bios); | |
275 | INIT_WORK(&blkg->async_bio_work, blkg_async_bio_workfn); | |
0381411e | 276 | blkg->blkcg = blkcg; |
0381411e | 277 | |
f7331648 | 278 | u64_stats_init(&blkg->iostat.sync); |
3b8cc629 | 279 | for_each_possible_cpu(cpu) { |
f7331648 | 280 | u64_stats_init(&per_cpu_ptr(blkg->iostat_cpu, cpu)->sync); |
3b8cc629 WL |
281 | per_cpu_ptr(blkg->iostat_cpu, cpu)->blkg = blkg; |
282 | } | |
f7331648 | 283 | |
8bd435b3 | 284 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
3c798398 | 285 | struct blkcg_policy *pol = blkcg_policy[i]; |
e8989fae | 286 | struct blkg_policy_data *pd; |
0381411e | 287 | |
1231039d | 288 | if (!blkcg_policy_enabled(disk->queue, pol)) |
e8989fae TH |
289 | continue; |
290 | ||
291 | /* alloc per-policy data and attach it to blkg */ | |
0a0b4f79 | 292 | pd = pol->pd_alloc_fn(disk, blkcg, gfp_mask); |
a051661c | 293 | if (!pd) |
0b6f93bd | 294 | goto out_free_pds; |
e8989fae TH |
295 | blkg->pd[i] = pd; |
296 | pd->blkg = blkg; | |
b276a876 | 297 | pd->plid = i; |
dfd6200a | 298 | pd->online = false; |
e8989fae TH |
299 | } |
300 | ||
0381411e | 301 | return blkg; |
a051661c | 302 | |
0b6f93bd CH |
303 | out_free_pds: |
304 | while (--i >= 0) | |
305 | if (blkg->pd[i]) | |
306 | blkcg_policy[i]->pd_free_fn(blkg->pd[i]); | |
84d7d462 | 307 | put_disk(blkg->disk); |
0b6f93bd CH |
308 | out_free_iostat: |
309 | free_percpu(blkg->iostat_cpu); | |
310 | out_exit_refcnt: | |
311 | percpu_ref_exit(&blkg->refcnt); | |
312 | out_free_blkg: | |
313 | kfree(blkg); | |
a051661c | 314 | return NULL; |
0381411e TH |
315 | } |
316 | ||
15974993 | 317 | /* |
d708f0d5 JA |
318 | * If @new_blkg is %NULL, this function tries to allocate a new one as |
319 | * necessary using %GFP_NOWAIT. @new_blkg is always consumed on return. | |
15974993 | 320 | */ |
99e60387 | 321 | static struct blkcg_gq *blkg_create(struct blkcg *blkcg, struct gendisk *disk, |
d708f0d5 | 322 | struct blkcg_gq *new_blkg) |
5624a4e4 | 323 | { |
d708f0d5 | 324 | struct blkcg_gq *blkg; |
f427d909 | 325 | int i, ret; |
5624a4e4 | 326 | |
99e60387 | 327 | lockdep_assert_held(&disk->queue->queue_lock); |
cd1604fa | 328 | |
0273ac34 | 329 | /* request_queue is dying, do not create/recreate a blkg */ |
99e60387 | 330 | if (blk_queue_dying(disk->queue)) { |
0273ac34 DZ |
331 | ret = -ENODEV; |
332 | goto err_free_blkg; | |
333 | } | |
334 | ||
7ee9c562 | 335 | /* blkg holds a reference to blkcg */ |
ec903c0c | 336 | if (!css_tryget_online(&blkcg->css)) { |
20386ce0 | 337 | ret = -ENODEV; |
93e6d5d8 | 338 | goto err_free_blkg; |
15974993 | 339 | } |
cd1604fa | 340 | |
d708f0d5 JA |
341 | /* allocate */ |
342 | if (!new_blkg) { | |
99e60387 | 343 | new_blkg = blkg_alloc(blkcg, disk, GFP_NOWAIT | __GFP_NOWARN); |
d708f0d5 JA |
344 | if (unlikely(!new_blkg)) { |
345 | ret = -ENOMEM; | |
8c911f3d | 346 | goto err_put_css; |
15974993 TH |
347 | } |
348 | } | |
d708f0d5 | 349 | blkg = new_blkg; |
cd1604fa | 350 | |
db613670 | 351 | /* link parent */ |
3c547865 | 352 | if (blkcg_parent(blkcg)) { |
9a9c261e | 353 | blkg->parent = blkg_lookup(blkcg_parent(blkcg), disk->queue); |
3c547865 | 354 | if (WARN_ON_ONCE(!blkg->parent)) { |
20386ce0 | 355 | ret = -ENODEV; |
8c911f3d | 356 | goto err_put_css; |
3c547865 TH |
357 | } |
358 | blkg_get(blkg->parent); | |
359 | } | |
360 | ||
db613670 TH |
361 | /* invoke per-policy init */ |
362 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
363 | struct blkcg_policy *pol = blkcg_policy[i]; | |
364 | ||
365 | if (blkg->pd[i] && pol->pd_init_fn) | |
a9520cd6 | 366 | pol->pd_init_fn(blkg->pd[i]); |
db613670 TH |
367 | } |
368 | ||
369 | /* insert */ | |
cd1604fa | 370 | spin_lock(&blkcg->lock); |
99e60387 | 371 | ret = radix_tree_insert(&blkcg->blkg_tree, disk->queue->id, blkg); |
a637120e TH |
372 | if (likely(!ret)) { |
373 | hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list); | |
1231039d | 374 | list_add(&blkg->q_node, &disk->queue->blkg_list); |
f427d909 TH |
375 | |
376 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
377 | struct blkcg_policy *pol = blkcg_policy[i]; | |
378 | ||
dfd6200a YK |
379 | if (blkg->pd[i]) { |
380 | if (pol->pd_online_fn) | |
381 | pol->pd_online_fn(blkg->pd[i]); | |
382 | blkg->pd[i]->online = true; | |
383 | } | |
f427d909 | 384 | } |
a637120e | 385 | } |
f427d909 | 386 | blkg->online = true; |
cd1604fa | 387 | spin_unlock(&blkcg->lock); |
496fb780 | 388 | |
ec13b1d6 | 389 | if (!ret) |
a637120e | 390 | return blkg; |
15974993 | 391 | |
3c547865 TH |
392 | /* @blkg failed fully initialized, use the usual release path */ |
393 | blkg_put(blkg); | |
394 | return ERR_PTR(ret); | |
395 | ||
d708f0d5 | 396 | err_put_css: |
496fb780 | 397 | css_put(&blkcg->css); |
93e6d5d8 | 398 | err_free_blkg: |
28e538a3 CH |
399 | if (new_blkg) |
400 | blkg_free(new_blkg); | |
93e6d5d8 | 401 | return ERR_PTR(ret); |
31e4c28d | 402 | } |
3c96cb32 | 403 | |
86cde6b6 | 404 | /** |
8c546287 | 405 | * blkg_lookup_create - lookup blkg, try to create one if not there |
86cde6b6 | 406 | * @blkcg: blkcg of interest |
99e60387 | 407 | * @disk: gendisk of interest |
86cde6b6 | 408 | * |
99e60387 | 409 | * Lookup blkg for the @blkcg - @disk pair. If it doesn't exist, try to |
3c547865 TH |
410 | * create one. blkg creation is performed recursively from blkcg_root such |
411 | * that all non-root blkg's have access to the parent blkg. This function | |
99e60387 | 412 | * should be called under RCU read lock and takes @disk->queue->queue_lock. |
86cde6b6 | 413 | * |
beea9da0 DZ |
414 | * Returns the blkg or the closest blkg if blkg_create() fails as it walks |
415 | * down from root. | |
86cde6b6 | 416 | */ |
8c546287 | 417 | static struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg, |
99e60387 | 418 | struct gendisk *disk) |
3c96cb32 | 419 | { |
99e60387 | 420 | struct request_queue *q = disk->queue; |
86cde6b6 | 421 | struct blkcg_gq *blkg; |
8c546287 | 422 | unsigned long flags; |
86cde6b6 TH |
423 | |
424 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
86cde6b6 | 425 | |
9a9c261e | 426 | blkg = blkg_lookup(blkcg, q); |
86cde6b6 TH |
427 | if (blkg) |
428 | return blkg; | |
429 | ||
8c546287 | 430 | spin_lock_irqsave(&q->queue_lock, flags); |
9a9c261e | 431 | blkg = blkg_lookup(blkcg, q); |
4a69f325 | 432 | if (blkg) { |
5765033c CH |
433 | if (blkcg != &blkcg_root && |
434 | blkg != rcu_dereference(blkcg->blkg_hint)) | |
435 | rcu_assign_pointer(blkcg->blkg_hint, blkg); | |
8c546287 | 436 | goto found; |
4a69f325 | 437 | } |
8c546287 | 438 | |
3c547865 TH |
439 | /* |
440 | * Create blkgs walking down from blkcg_root to @blkcg, so that all | |
beea9da0 DZ |
441 | * non-root blkgs have access to their parents. Returns the closest |
442 | * blkg to the intended blkg should blkg_create() fail. | |
3c547865 TH |
443 | */ |
444 | while (true) { | |
445 | struct blkcg *pos = blkcg; | |
446 | struct blkcg *parent = blkcg_parent(blkcg); | |
1231039d | 447 | struct blkcg_gq *ret_blkg = q->root_blkg; |
beea9da0 DZ |
448 | |
449 | while (parent) { | |
9a9c261e | 450 | blkg = blkg_lookup(parent, q); |
beea9da0 DZ |
451 | if (blkg) { |
452 | /* remember closest blkg */ | |
453 | ret_blkg = blkg; | |
454 | break; | |
455 | } | |
3c547865 TH |
456 | pos = parent; |
457 | parent = blkcg_parent(parent); | |
458 | } | |
459 | ||
99e60387 | 460 | blkg = blkg_create(pos, disk, NULL); |
8c546287 CH |
461 | if (IS_ERR(blkg)) { |
462 | blkg = ret_blkg; | |
463 | break; | |
464 | } | |
beea9da0 | 465 | if (pos == blkcg) |
8c546287 | 466 | break; |
b978962a DZ |
467 | } |
468 | ||
8c546287 CH |
469 | found: |
470 | spin_unlock_irqrestore(&q->queue_lock, flags); | |
b978962a DZ |
471 | return blkg; |
472 | } | |
473 | ||
3c798398 | 474 | static void blkg_destroy(struct blkcg_gq *blkg) |
03aa264a | 475 | { |
3c798398 | 476 | struct blkcg *blkcg = blkg->blkcg; |
6b065462 | 477 | int i; |
03aa264a | 478 | |
84d7d462 | 479 | lockdep_assert_held(&blkg->disk->queue->queue_lock); |
9f13ef67 | 480 | lockdep_assert_held(&blkcg->lock); |
03aa264a | 481 | |
f1c006f1 YK |
482 | /* |
483 | * blkg stays on the queue list until blkg_free_workfn(), see details in | |
484 | * blkg_free_workfn(), hence this function can be called from | |
485 | * blkcg_destroy_blkgs() first and again from blkg_destroy_all() before | |
486 | * blkg_free_workfn(). | |
487 | */ | |
488 | if (hlist_unhashed(&blkg->blkcg_node)) | |
489 | return; | |
a637120e | 490 | |
6b065462 DZF |
491 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
492 | struct blkcg_policy *pol = blkcg_policy[i]; | |
493 | ||
dfd6200a | 494 | if (blkg->pd[i] && blkg->pd[i]->online) { |
f37bf75c | 495 | blkg->pd[i]->online = false; |
dfd6200a YK |
496 | if (pol->pd_offline_fn) |
497 | pol->pd_offline_fn(blkg->pd[i]); | |
dfd6200a | 498 | } |
6b065462 DZF |
499 | } |
500 | ||
f427d909 TH |
501 | blkg->online = false; |
502 | ||
84d7d462 | 503 | radix_tree_delete(&blkcg->blkg_tree, blkg->disk->queue->id); |
9f13ef67 | 504 | hlist_del_init_rcu(&blkg->blkcg_node); |
03aa264a | 505 | |
a637120e TH |
506 | /* |
507 | * Both setting lookup hint to and clearing it from @blkg are done | |
508 | * under queue_lock. If it's not pointing to @blkg now, it never | |
509 | * will. Hint assignment itself can race safely. | |
510 | */ | |
ec6c676a | 511 | if (rcu_access_pointer(blkcg->blkg_hint) == blkg) |
a637120e TH |
512 | rcu_assign_pointer(blkcg->blkg_hint, NULL); |
513 | ||
03aa264a TH |
514 | /* |
515 | * Put the reference taken at the time of creation so that when all | |
516 | * queues are gone, group can be destroyed. | |
517 | */ | |
7fcf2b03 | 518 | percpu_ref_kill(&blkg->refcnt); |
03aa264a TH |
519 | } |
520 | ||
00ad6991 | 521 | static void blkg_destroy_all(struct gendisk *disk) |
72e06c25 | 522 | { |
00ad6991 | 523 | struct request_queue *q = disk->queue; |
3c798398 | 524 | struct blkcg_gq *blkg, *n; |
a731763f | 525 | int count = BLKG_DESTROY_BATCH_SIZE; |
72e06c25 | 526 | |
a731763f | 527 | restart: |
0d945c1f | 528 | spin_lock_irq(&q->queue_lock); |
1231039d | 529 | list_for_each_entry_safe(blkg, n, &q->blkg_list, q_node) { |
3c798398 | 530 | struct blkcg *blkcg = blkg->blkcg; |
72e06c25 | 531 | |
9f13ef67 TH |
532 | spin_lock(&blkcg->lock); |
533 | blkg_destroy(blkg); | |
534 | spin_unlock(&blkcg->lock); | |
a731763f YK |
535 | |
536 | /* | |
537 | * in order to avoid holding the spin lock for too long, release | |
538 | * it when a batch of blkgs are destroyed. | |
539 | */ | |
540 | if (!(--count)) { | |
541 | count = BLKG_DESTROY_BATCH_SIZE; | |
542 | spin_unlock_irq(&q->queue_lock); | |
543 | cond_resched(); | |
544 | goto restart; | |
545 | } | |
72e06c25 | 546 | } |
6fe810bd | 547 | |
1231039d | 548 | q->root_blkg = NULL; |
0d945c1f | 549 | spin_unlock_irq(&q->queue_lock); |
72e06c25 TH |
550 | } |
551 | ||
182446d0 TH |
552 | static int blkcg_reset_stats(struct cgroup_subsys_state *css, |
553 | struct cftype *cftype, u64 val) | |
303a3acb | 554 | { |
182446d0 | 555 | struct blkcg *blkcg = css_to_blkcg(css); |
3c798398 | 556 | struct blkcg_gq *blkg; |
f7331648 | 557 | int i, cpu; |
303a3acb | 558 | |
838f13bf | 559 | mutex_lock(&blkcg_pol_mutex); |
303a3acb | 560 | spin_lock_irq(&blkcg->lock); |
997a026c TH |
561 | |
562 | /* | |
563 | * Note that stat reset is racy - it doesn't synchronize against | |
564 | * stat updates. This is a debug feature which shouldn't exist | |
565 | * anyway. If you get hit by a race, retry. | |
566 | */ | |
b67bfe0d | 567 | hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) { |
f7331648 TH |
568 | for_each_possible_cpu(cpu) { |
569 | struct blkg_iostat_set *bis = | |
570 | per_cpu_ptr(blkg->iostat_cpu, cpu); | |
571 | memset(bis, 0, sizeof(*bis)); | |
572 | } | |
573 | memset(&blkg->iostat, 0, sizeof(blkg->iostat)); | |
77ea7338 | 574 | |
8bd435b3 | 575 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
3c798398 | 576 | struct blkcg_policy *pol = blkcg_policy[i]; |
549d3aa8 | 577 | |
a9520cd6 TH |
578 | if (blkg->pd[i] && pol->pd_reset_stats_fn) |
579 | pol->pd_reset_stats_fn(blkg->pd[i]); | |
bc0d6501 | 580 | } |
303a3acb | 581 | } |
f0bdc8cd | 582 | |
303a3acb | 583 | spin_unlock_irq(&blkcg->lock); |
bc0d6501 | 584 | mutex_unlock(&blkcg_pol_mutex); |
303a3acb DS |
585 | return 0; |
586 | } | |
587 | ||
dd165eb3 | 588 | const char *blkg_dev_name(struct blkcg_gq *blkg) |
303a3acb | 589 | { |
84d7d462 | 590 | return bdi_dev_name(blkg->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) { |
84d7d462 | 622 | spin_lock_irq(&blkg->disk->queue->queue_lock); |
1231039d | 623 | if (blkcg_policy_enabled(blkg->disk->queue, pol)) |
f95a04af | 624 | total += prfill(sf, blkg->pd[pol->plid], data); |
84d7d462 | 625 | spin_unlock_irq(&blkg->disk->queue->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 | /** |
22ae8ce8 | 655 | * blkcg_conf_open_bdev - parse and open bdev for per-blkg config update |
015d254c TH |
656 | * @inputp: input string pointer |
657 | * | |
658 | * Parse the device node prefix part, MAJ:MIN, of per-blkg config update | |
22ae8ce8 | 659 | * from @input and get and return the matching bdev. *@inputp is |
015d254c TH |
660 | * updated to point past the device node prefix. Returns an ERR_PTR() |
661 | * value on error. | |
662 | * | |
663 | * Use this function iff blkg_conf_prep() can't be used for some reason. | |
664 | */ | |
22ae8ce8 | 665 | struct block_device *blkcg_conf_open_bdev(char **inputp) |
015d254c TH |
666 | { |
667 | char *input = *inputp; | |
668 | unsigned int major, minor; | |
22ae8ce8 CH |
669 | struct block_device *bdev; |
670 | int key_len; | |
015d254c TH |
671 | |
672 | if (sscanf(input, "%u:%u%n", &major, &minor, &key_len) != 2) | |
673 | return ERR_PTR(-EINVAL); | |
674 | ||
675 | input += key_len; | |
676 | if (!isspace(*input)) | |
677 | return ERR_PTR(-EINVAL); | |
678 | input = skip_spaces(input); | |
679 | ||
22ae8ce8 CH |
680 | bdev = blkdev_get_no_open(MKDEV(major, minor)); |
681 | if (!bdev) | |
015d254c | 682 | return ERR_PTR(-ENODEV); |
22ae8ce8 CH |
683 | if (bdev_is_partition(bdev)) { |
684 | blkdev_put_no_open(bdev); | |
015d254c TH |
685 | return ERR_PTR(-ENODEV); |
686 | } | |
687 | ||
688 | *inputp = input; | |
22ae8ce8 | 689 | return bdev; |
015d254c TH |
690 | } |
691 | ||
3a8b31d3 TH |
692 | /** |
693 | * blkg_conf_prep - parse and prepare for per-blkg config update | |
694 | * @blkcg: target block cgroup | |
da8b0662 | 695 | * @pol: target policy |
3a8b31d3 TH |
696 | * @input: input string |
697 | * @ctx: blkg_conf_ctx to be filled | |
698 | * | |
699 | * Parse per-blkg config update from @input and initialize @ctx with the | |
36aa9e5f TH |
700 | * result. @ctx->blkg points to the blkg to be updated and @ctx->body the |
701 | * part of @input following MAJ:MIN. This function returns with RCU read | |
702 | * lock and queue lock held and must be paired with blkg_conf_finish(). | |
3a8b31d3 | 703 | */ |
3c798398 | 704 | int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol, |
36aa9e5f | 705 | char *input, struct blkg_conf_ctx *ctx) |
ed6cddef | 706 | __acquires(rcu) __acquires(&bdev->bd_queue->queue_lock) |
34d0f179 | 707 | { |
22ae8ce8 | 708 | struct block_device *bdev; |
99e60387 | 709 | struct gendisk *disk; |
457e490f | 710 | struct request_queue *q; |
3c798398 | 711 | struct blkcg_gq *blkg; |
015d254c | 712 | int ret; |
36aa9e5f | 713 | |
22ae8ce8 CH |
714 | bdev = blkcg_conf_open_bdev(&input); |
715 | if (IS_ERR(bdev)) | |
716 | return PTR_ERR(bdev); | |
99e60387 CH |
717 | disk = bdev->bd_disk; |
718 | q = disk->queue; | |
da8b0662 | 719 | |
0c9d338c YK |
720 | /* |
721 | * blkcg_deactivate_policy() requires queue to be frozen, we can grab | |
722 | * q_usage_counter to prevent concurrent with blkcg_deactivate_policy(). | |
723 | */ | |
724 | ret = blk_queue_enter(q, 0); | |
725 | if (ret) | |
15c30104 | 726 | goto fail; |
0c9d338c | 727 | |
457e490f | 728 | rcu_read_lock(); |
0d945c1f | 729 | spin_lock_irq(&q->queue_lock); |
e56da7e2 | 730 | |
1231039d | 731 | if (!blkcg_policy_enabled(q, pol)) { |
f753526e | 732 | ret = -EOPNOTSUPP; |
457e490f TE |
733 | goto fail_unlock; |
734 | } | |
735 | ||
9a9c261e | 736 | blkg = blkg_lookup(blkcg, q); |
5765033c | 737 | if (blkg) |
457e490f TE |
738 | goto success; |
739 | ||
740 | /* | |
741 | * Create blkgs walking down from blkcg_root to @blkcg, so that all | |
742 | * non-root blkgs have access to their parents. | |
743 | */ | |
744 | while (true) { | |
745 | struct blkcg *pos = blkcg; | |
746 | struct blkcg *parent; | |
747 | struct blkcg_gq *new_blkg; | |
748 | ||
749 | parent = blkcg_parent(blkcg); | |
9a9c261e | 750 | while (parent && !blkg_lookup(parent, q)) { |
457e490f TE |
751 | pos = parent; |
752 | parent = blkcg_parent(parent); | |
753 | } | |
754 | ||
755 | /* Drop locks to do new blkg allocation with GFP_KERNEL. */ | |
0d945c1f | 756 | spin_unlock_irq(&q->queue_lock); |
3a8b31d3 | 757 | rcu_read_unlock(); |
457e490f | 758 | |
99e60387 | 759 | new_blkg = blkg_alloc(pos, disk, GFP_KERNEL); |
457e490f TE |
760 | if (unlikely(!new_blkg)) { |
761 | ret = -ENOMEM; | |
15c30104 | 762 | goto fail_exit_queue; |
7702e8f4 | 763 | } |
3a8b31d3 | 764 | |
f255c19b GKB |
765 | if (radix_tree_preload(GFP_KERNEL)) { |
766 | blkg_free(new_blkg); | |
767 | ret = -ENOMEM; | |
15c30104 | 768 | goto fail_exit_queue; |
f255c19b GKB |
769 | } |
770 | ||
457e490f | 771 | rcu_read_lock(); |
0d945c1f | 772 | spin_lock_irq(&q->queue_lock); |
457e490f | 773 | |
1231039d | 774 | if (!blkcg_policy_enabled(q, pol)) { |
52abfcbd | 775 | blkg_free(new_blkg); |
f753526e | 776 | ret = -EOPNOTSUPP; |
f255c19b | 777 | goto fail_preloaded; |
457e490f TE |
778 | } |
779 | ||
9a9c261e | 780 | blkg = blkg_lookup(pos, q); |
457e490f TE |
781 | if (blkg) { |
782 | blkg_free(new_blkg); | |
783 | } else { | |
99e60387 | 784 | blkg = blkg_create(pos, disk, new_blkg); |
98d669b4 | 785 | if (IS_ERR(blkg)) { |
457e490f | 786 | ret = PTR_ERR(blkg); |
f255c19b | 787 | goto fail_preloaded; |
457e490f TE |
788 | } |
789 | } | |
790 | ||
f255c19b GKB |
791 | radix_tree_preload_end(); |
792 | ||
457e490f TE |
793 | if (pos == blkcg) |
794 | goto success; | |
795 | } | |
796 | success: | |
0c9d338c | 797 | blk_queue_exit(q); |
22ae8ce8 | 798 | ctx->bdev = bdev; |
3a8b31d3 | 799 | ctx->blkg = blkg; |
015d254c | 800 | ctx->body = input; |
726fa694 | 801 | return 0; |
457e490f | 802 | |
f255c19b GKB |
803 | fail_preloaded: |
804 | radix_tree_preload_end(); | |
457e490f | 805 | fail_unlock: |
0d945c1f | 806 | spin_unlock_irq(&q->queue_lock); |
457e490f | 807 | rcu_read_unlock(); |
15c30104 YK |
808 | fail_exit_queue: |
809 | blk_queue_exit(q); | |
457e490f | 810 | fail: |
22ae8ce8 | 811 | blkdev_put_no_open(bdev); |
457e490f TE |
812 | /* |
813 | * If queue was bypassing, we should retry. Do so after a | |
814 | * short msleep(). It isn't strictly necessary but queue | |
815 | * can be bypassing for some time and it's always nice to | |
816 | * avoid busy looping. | |
817 | */ | |
818 | if (ret == -EBUSY) { | |
819 | msleep(10); | |
820 | ret = restart_syscall(); | |
821 | } | |
822 | return ret; | |
34d0f179 | 823 | } |
89f3b6d6 | 824 | EXPORT_SYMBOL_GPL(blkg_conf_prep); |
34d0f179 | 825 | |
3a8b31d3 TH |
826 | /** |
827 | * blkg_conf_finish - finish up per-blkg config update | |
37754595 | 828 | * @ctx: blkg_conf_ctx initialized by blkg_conf_prep() |
3a8b31d3 TH |
829 | * |
830 | * Finish up after per-blkg config update. This function must be paired | |
831 | * with blkg_conf_prep(). | |
832 | */ | |
829fdb50 | 833 | void blkg_conf_finish(struct blkg_conf_ctx *ctx) |
ed6cddef | 834 | __releases(&ctx->bdev->bd_queue->queue_lock) __releases(rcu) |
34d0f179 | 835 | { |
ed6cddef | 836 | spin_unlock_irq(&bdev_get_queue(ctx->bdev)->queue_lock); |
3a8b31d3 | 837 | rcu_read_unlock(); |
22ae8ce8 | 838 | blkdev_put_no_open(ctx->bdev); |
34d0f179 | 839 | } |
89f3b6d6 | 840 | EXPORT_SYMBOL_GPL(blkg_conf_finish); |
34d0f179 | 841 | |
cd1fc4b9 BB |
842 | static void blkg_iostat_set(struct blkg_iostat *dst, struct blkg_iostat *src) |
843 | { | |
844 | int i; | |
845 | ||
846 | for (i = 0; i < BLKG_IOSTAT_NR; i++) { | |
847 | dst->bytes[i] = src->bytes[i]; | |
848 | dst->ios[i] = src->ios[i]; | |
849 | } | |
850 | } | |
851 | ||
852 | static void blkg_iostat_add(struct blkg_iostat *dst, struct blkg_iostat *src) | |
853 | { | |
854 | int i; | |
855 | ||
856 | for (i = 0; i < BLKG_IOSTAT_NR; i++) { | |
857 | dst->bytes[i] += src->bytes[i]; | |
858 | dst->ios[i] += src->ios[i]; | |
859 | } | |
860 | } | |
861 | ||
862 | static void blkg_iostat_sub(struct blkg_iostat *dst, struct blkg_iostat *src) | |
863 | { | |
864 | int i; | |
865 | ||
866 | for (i = 0; i < BLKG_IOSTAT_NR; i++) { | |
867 | dst->bytes[i] -= src->bytes[i]; | |
868 | dst->ios[i] -= src->ios[i]; | |
869 | } | |
870 | } | |
871 | ||
362b8c16 JY |
872 | static void blkcg_iostat_update(struct blkcg_gq *blkg, struct blkg_iostat *cur, |
873 | struct blkg_iostat *last) | |
874 | { | |
875 | struct blkg_iostat delta; | |
876 | unsigned long flags; | |
877 | ||
878 | /* propagate percpu delta to global */ | |
879 | flags = u64_stats_update_begin_irqsave(&blkg->iostat.sync); | |
880 | blkg_iostat_set(&delta, cur); | |
881 | blkg_iostat_sub(&delta, last); | |
882 | blkg_iostat_add(&blkg->iostat.cur, &delta); | |
883 | blkg_iostat_add(last, &delta); | |
884 | u64_stats_update_end_irqrestore(&blkg->iostat.sync, flags); | |
885 | } | |
886 | ||
cd1fc4b9 BB |
887 | static void blkcg_rstat_flush(struct cgroup_subsys_state *css, int cpu) |
888 | { | |
889 | struct blkcg *blkcg = css_to_blkcg(css); | |
3b8cc629 WL |
890 | struct llist_head *lhead = per_cpu_ptr(blkcg->lhead, cpu); |
891 | struct llist_node *lnode; | |
892 | struct blkg_iostat_set *bisc, *next_bisc; | |
cd1fc4b9 | 893 | |
dc26532a JW |
894 | /* Root-level stats are sourced from system-wide IO stats */ |
895 | if (!cgroup_parent(css->cgroup)) | |
896 | return; | |
897 | ||
cd1fc4b9 BB |
898 | rcu_read_lock(); |
899 | ||
3b8cc629 WL |
900 | lnode = llist_del_all(lhead); |
901 | if (!lnode) | |
902 | goto out; | |
903 | ||
904 | /* | |
905 | * Iterate only the iostat_cpu's queued in the lockless list. | |
906 | */ | |
907 | llist_for_each_entry_safe(bisc, next_bisc, lnode, lnode) { | |
908 | struct blkcg_gq *blkg = bisc->blkg; | |
cd1fc4b9 | 909 | struct blkcg_gq *parent = blkg->parent; |
362b8c16 | 910 | struct blkg_iostat cur; |
cd1fc4b9 BB |
911 | unsigned int seq; |
912 | ||
3b8cc629 WL |
913 | WRITE_ONCE(bisc->lqueued, false); |
914 | ||
cd1fc4b9 BB |
915 | /* fetch the current per-cpu values */ |
916 | do { | |
917 | seq = u64_stats_fetch_begin(&bisc->sync); | |
918 | blkg_iostat_set(&cur, &bisc->cur); | |
919 | } while (u64_stats_fetch_retry(&bisc->sync, seq)); | |
920 | ||
362b8c16 | 921 | blkcg_iostat_update(blkg, &cur, &bisc->last); |
cd1fc4b9 | 922 | |
dc26532a | 923 | /* propagate global delta to parent (unless that's root) */ |
362b8c16 JY |
924 | if (parent && parent->parent) |
925 | blkcg_iostat_update(parent, &blkg->iostat.cur, | |
926 | &blkg->iostat.last); | |
3b8cc629 | 927 | percpu_ref_put(&blkg->refcnt); |
cd1fc4b9 BB |
928 | } |
929 | ||
3b8cc629 | 930 | out: |
cd1fc4b9 BB |
931 | rcu_read_unlock(); |
932 | } | |
933 | ||
ef45fe47 | 934 | /* |
dc26532a JW |
935 | * We source root cgroup stats from the system-wide stats to avoid |
936 | * tracking the same information twice and incurring overhead when no | |
937 | * cgroups are defined. For that reason, cgroup_rstat_flush in | |
938 | * blkcg_print_stat does not actually fill out the iostat in the root | |
939 | * cgroup's blkcg_gq. | |
ef45fe47 BB |
940 | * |
941 | * However, we would like to re-use the printing code between the root and | |
942 | * non-root cgroups to the extent possible. For that reason, we simulate | |
943 | * flushing the root cgroup's stats by explicitly filling in the iostat | |
944 | * with disk level statistics. | |
945 | */ | |
946 | static void blkcg_fill_root_iostats(void) | |
947 | { | |
948 | struct class_dev_iter iter; | |
949 | struct device *dev; | |
950 | ||
951 | class_dev_iter_init(&iter, &block_class, NULL, &disk_type); | |
952 | while ((dev = class_dev_iter_next(&iter))) { | |
0d02129e | 953 | struct block_device *bdev = dev_to_bdev(dev); |
1231039d | 954 | struct blkcg_gq *blkg = bdev->bd_disk->queue->root_blkg; |
ef45fe47 BB |
955 | struct blkg_iostat tmp; |
956 | int cpu; | |
f122d103 | 957 | unsigned long flags; |
ef45fe47 BB |
958 | |
959 | memset(&tmp, 0, sizeof(tmp)); | |
960 | for_each_possible_cpu(cpu) { | |
961 | struct disk_stats *cpu_dkstats; | |
962 | ||
0d02129e | 963 | cpu_dkstats = per_cpu_ptr(bdev->bd_stats, cpu); |
ef45fe47 BB |
964 | tmp.ios[BLKG_IOSTAT_READ] += |
965 | cpu_dkstats->ios[STAT_READ]; | |
966 | tmp.ios[BLKG_IOSTAT_WRITE] += | |
967 | cpu_dkstats->ios[STAT_WRITE]; | |
968 | tmp.ios[BLKG_IOSTAT_DISCARD] += | |
969 | cpu_dkstats->ios[STAT_DISCARD]; | |
970 | // convert sectors to bytes | |
971 | tmp.bytes[BLKG_IOSTAT_READ] += | |
972 | cpu_dkstats->sectors[STAT_READ] << 9; | |
973 | tmp.bytes[BLKG_IOSTAT_WRITE] += | |
974 | cpu_dkstats->sectors[STAT_WRITE] << 9; | |
975 | tmp.bytes[BLKG_IOSTAT_DISCARD] += | |
976 | cpu_dkstats->sectors[STAT_DISCARD] << 9; | |
ef45fe47 | 977 | } |
f122d103 CZ |
978 | |
979 | flags = u64_stats_update_begin_irqsave(&blkg->iostat.sync); | |
980 | blkg_iostat_set(&blkg->iostat.cur, &tmp); | |
981 | u64_stats_update_end_irqrestore(&blkg->iostat.sync, flags); | |
ef45fe47 BB |
982 | } |
983 | } | |
984 | ||
49cb5168 | 985 | static void blkcg_print_one_stat(struct blkcg_gq *blkg, struct seq_file *s) |
2ee867dc | 986 | { |
49cb5168 CH |
987 | struct blkg_iostat_set *bis = &blkg->iostat; |
988 | u64 rbytes, wbytes, rios, wios, dbytes, dios; | |
49cb5168 CH |
989 | const char *dname; |
990 | unsigned seq; | |
49cb5168 | 991 | int i; |
ef45fe47 | 992 | |
49cb5168 CH |
993 | if (!blkg->online) |
994 | return; | |
2ee867dc | 995 | |
49cb5168 CH |
996 | dname = blkg_dev_name(blkg); |
997 | if (!dname) | |
998 | return; | |
2ee867dc | 999 | |
252c651a | 1000 | seq_printf(s, "%s ", dname); |
b0814361 | 1001 | |
49cb5168 CH |
1002 | do { |
1003 | seq = u64_stats_fetch_begin(&bis->sync); | |
b0814361 | 1004 | |
49cb5168 CH |
1005 | rbytes = bis->cur.bytes[BLKG_IOSTAT_READ]; |
1006 | wbytes = bis->cur.bytes[BLKG_IOSTAT_WRITE]; | |
1007 | dbytes = bis->cur.bytes[BLKG_IOSTAT_DISCARD]; | |
1008 | rios = bis->cur.ios[BLKG_IOSTAT_READ]; | |
1009 | wios = bis->cur.ios[BLKG_IOSTAT_WRITE]; | |
1010 | dios = bis->cur.ios[BLKG_IOSTAT_DISCARD]; | |
1011 | } while (u64_stats_fetch_retry(&bis->sync, seq)); | |
2ee867dc | 1012 | |
49cb5168 | 1013 | if (rbytes || wbytes || rios || wios) { |
252c651a | 1014 | seq_printf(s, "rbytes=%llu wbytes=%llu rios=%llu wios=%llu dbytes=%llu dios=%llu", |
49cb5168 CH |
1015 | rbytes, wbytes, rios, wios, |
1016 | dbytes, dios); | |
1017 | } | |
903d23f0 | 1018 | |
49cb5168 | 1019 | if (blkcg_debug_stats && atomic_read(&blkg->use_delay)) { |
252c651a | 1020 | seq_printf(s, " use_delay=%d delay_nsec=%llu", |
49cb5168 CH |
1021 | atomic_read(&blkg->use_delay), |
1022 | atomic64_read(&blkg->delay_nsec)); | |
1023 | } | |
2ee867dc | 1024 | |
49cb5168 CH |
1025 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
1026 | struct blkcg_policy *pol = blkcg_policy[i]; | |
2ee867dc | 1027 | |
49cb5168 CH |
1028 | if (!blkg->pd[i] || !pol->pd_stat_fn) |
1029 | continue; | |
903d23f0 | 1030 | |
3607849d | 1031 | pol->pd_stat_fn(blkg->pd[i], s); |
49cb5168 | 1032 | } |
d09d8df3 | 1033 | |
3607849d | 1034 | seq_puts(s, "\n"); |
49cb5168 | 1035 | } |
903d23f0 | 1036 | |
49cb5168 CH |
1037 | static int blkcg_print_stat(struct seq_file *sf, void *v) |
1038 | { | |
1039 | struct blkcg *blkcg = css_to_blkcg(seq_css(sf)); | |
1040 | struct blkcg_gq *blkg; | |
903d23f0 | 1041 | |
49cb5168 CH |
1042 | if (!seq_css(sf)->parent) |
1043 | blkcg_fill_root_iostats(); | |
1044 | else | |
1045 | cgroup_rstat_flush(blkcg->css.cgroup); | |
07b0fdec | 1046 | |
49cb5168 CH |
1047 | rcu_read_lock(); |
1048 | hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) { | |
84d7d462 | 1049 | spin_lock_irq(&blkg->disk->queue->queue_lock); |
49cb5168 | 1050 | blkcg_print_one_stat(blkg, sf); |
84d7d462 | 1051 | spin_unlock_irq(&blkg->disk->queue->queue_lock); |
2ee867dc | 1052 | } |
2ee867dc TH |
1053 | rcu_read_unlock(); |
1054 | return 0; | |
1055 | } | |
1056 | ||
e1f3b941 | 1057 | static struct cftype blkcg_files[] = { |
2ee867dc TH |
1058 | { |
1059 | .name = "stat", | |
1060 | .seq_show = blkcg_print_stat, | |
1061 | }, | |
1062 | { } /* terminate */ | |
1063 | }; | |
1064 | ||
e1f3b941 | 1065 | static struct cftype blkcg_legacy_files[] = { |
84c124da DS |
1066 | { |
1067 | .name = "reset_stats", | |
3c798398 | 1068 | .write_u64 = blkcg_reset_stats, |
22084190 | 1069 | }, |
4baf6e33 | 1070 | { } /* terminate */ |
31e4c28d VG |
1071 | }; |
1072 | ||
dec223c9 CH |
1073 | #ifdef CONFIG_CGROUP_WRITEBACK |
1074 | struct list_head *blkcg_get_cgwb_list(struct cgroup_subsys_state *css) | |
1075 | { | |
1076 | return &css_to_blkcg(css)->cgwb_list; | |
1077 | } | |
1078 | #endif | |
1079 | ||
59b57717 DZF |
1080 | /* |
1081 | * blkcg destruction is a three-stage process. | |
1082 | * | |
1083 | * 1. Destruction starts. The blkcg_css_offline() callback is invoked | |
1084 | * which offlines writeback. Here we tie the next stage of blkg destruction | |
1085 | * to the completion of writeback associated with the blkcg. This lets us | |
1086 | * avoid punting potentially large amounts of outstanding writeback to root | |
1087 | * while maintaining any ongoing policies. The next stage is triggered when | |
1088 | * the nr_cgwbs count goes to zero. | |
1089 | * | |
1090 | * 2. When the nr_cgwbs count goes to zero, blkcg_destroy_blkgs() is called | |
1091 | * and handles the destruction of blkgs. Here the css reference held by | |
1092 | * the blkg is put back eventually allowing blkcg_css_free() to be called. | |
1093 | * This work may occur in cgwb_release_workfn() on the cgwb_release | |
1094 | * workqueue. Any submitted ios that fail to get the blkg ref will be | |
1095 | * punted to the root_blkg. | |
1096 | * | |
1097 | * 3. Once the blkcg ref count goes to zero, blkcg_css_free() is called. | |
1098 | * This finally frees the blkcg. | |
1099 | */ | |
1100 | ||
59b57717 DZF |
1101 | /** |
1102 | * blkcg_destroy_blkgs - responsible for shooting down blkgs | |
1103 | * @blkcg: blkcg of interest | |
1104 | * | |
1105 | * blkgs should be removed while holding both q and blkcg locks. As blkcg lock | |
1106 | * is nested inside q lock, this function performs reverse double lock dancing. | |
1107 | * Destroying the blkgs releases the reference held on the blkcg's css allowing | |
1108 | * blkcg_css_free to eventually be called. | |
1109 | * | |
1110 | * This is the blkcg counterpart of ioc_release_fn(). | |
1111 | */ | |
397c9f46 | 1112 | static void blkcg_destroy_blkgs(struct blkcg *blkcg) |
59b57717 | 1113 | { |
6c635cae BW |
1114 | might_sleep(); |
1115 | ||
9f13ef67 | 1116 | spin_lock_irq(&blkcg->lock); |
7ee9c562 | 1117 | |
4c699480 JQ |
1118 | while (!hlist_empty(&blkcg->blkg_list)) { |
1119 | struct blkcg_gq *blkg = hlist_entry(blkcg->blkg_list.first, | |
6b065462 | 1120 | struct blkcg_gq, blkcg_node); |
84d7d462 | 1121 | struct request_queue *q = blkg->disk->queue; |
4c699480 | 1122 | |
6c635cae BW |
1123 | if (need_resched() || !spin_trylock(&q->queue_lock)) { |
1124 | /* | |
1125 | * Given that the system can accumulate a huge number | |
1126 | * of blkgs in pathological cases, check to see if we | |
1127 | * need to rescheduling to avoid softlockup. | |
1128 | */ | |
4c699480 | 1129 | spin_unlock_irq(&blkcg->lock); |
6c635cae | 1130 | cond_resched(); |
4c699480 | 1131 | spin_lock_irq(&blkcg->lock); |
6c635cae | 1132 | continue; |
4c699480 | 1133 | } |
6c635cae BW |
1134 | |
1135 | blkg_destroy(blkg); | |
1136 | spin_unlock(&q->queue_lock); | |
4c699480 | 1137 | } |
6b065462 | 1138 | |
4c699480 JQ |
1139 | spin_unlock_irq(&blkcg->lock); |
1140 | } | |
1141 | ||
397c9f46 CH |
1142 | /** |
1143 | * blkcg_pin_online - pin online state | |
1144 | * @blkcg_css: blkcg of interest | |
1145 | * | |
1146 | * While pinned, a blkcg is kept online. This is primarily used to | |
1147 | * impedance-match blkg and cgwb lifetimes so that blkg doesn't go offline | |
1148 | * while an associated cgwb is still active. | |
1149 | */ | |
1150 | void blkcg_pin_online(struct cgroup_subsys_state *blkcg_css) | |
1151 | { | |
1152 | refcount_inc(&css_to_blkcg(blkcg_css)->online_pin); | |
1153 | } | |
1154 | ||
1155 | /** | |
1156 | * blkcg_unpin_online - unpin online state | |
1157 | * @blkcg_css: blkcg of interest | |
1158 | * | |
1159 | * This is primarily used to impedance-match blkg and cgwb lifetimes so | |
1160 | * that blkg doesn't go offline while an associated cgwb is still active. | |
1161 | * When this count goes to zero, all active cgwbs have finished so the | |
1162 | * blkcg can continue destruction by calling blkcg_destroy_blkgs(). | |
1163 | */ | |
1164 | void blkcg_unpin_online(struct cgroup_subsys_state *blkcg_css) | |
1165 | { | |
1166 | struct blkcg *blkcg = css_to_blkcg(blkcg_css); | |
1167 | ||
1168 | do { | |
1169 | if (!refcount_dec_and_test(&blkcg->online_pin)) | |
1170 | break; | |
1171 | blkcg_destroy_blkgs(blkcg); | |
1172 | blkcg = blkcg_parent(blkcg); | |
1173 | } while (blkcg); | |
1174 | } | |
1175 | ||
1176 | /** | |
1177 | * blkcg_css_offline - cgroup css_offline callback | |
1178 | * @css: css of interest | |
1179 | * | |
1180 | * This function is called when @css is about to go away. Here the cgwbs are | |
1181 | * offlined first and only once writeback associated with the blkcg has | |
1182 | * finished do we start step 2 (see above). | |
1183 | */ | |
1184 | static void blkcg_css_offline(struct cgroup_subsys_state *css) | |
1185 | { | |
1186 | /* this prevents anyone from attaching or migrating to this blkcg */ | |
dec223c9 | 1187 | wb_blkcg_offline(css); |
397c9f46 CH |
1188 | |
1189 | /* put the base online pin allowing step 2 to be triggered */ | |
1190 | blkcg_unpin_online(css); | |
1191 | } | |
1192 | ||
eb95419b | 1193 | static void blkcg_css_free(struct cgroup_subsys_state *css) |
7ee9c562 | 1194 | { |
eb95419b | 1195 | struct blkcg *blkcg = css_to_blkcg(css); |
bc915e61 | 1196 | int i; |
7ee9c562 | 1197 | |
7876f930 | 1198 | mutex_lock(&blkcg_pol_mutex); |
e4a9bde9 | 1199 | |
7876f930 | 1200 | list_del(&blkcg->all_blkcgs_node); |
7876f930 | 1201 | |
bc915e61 | 1202 | for (i = 0; i < BLKCG_MAX_POLS; i++) |
e4a9bde9 TH |
1203 | if (blkcg->cpd[i]) |
1204 | blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]); | |
1205 | ||
1206 | mutex_unlock(&blkcg_pol_mutex); | |
1207 | ||
3b8cc629 | 1208 | free_percpu(blkcg->lhead); |
bc915e61 | 1209 | kfree(blkcg); |
31e4c28d VG |
1210 | } |
1211 | ||
eb95419b TH |
1212 | static struct cgroup_subsys_state * |
1213 | blkcg_css_alloc(struct cgroup_subsys_state *parent_css) | |
31e4c28d | 1214 | { |
3c798398 | 1215 | struct blkcg *blkcg; |
e48453c3 | 1216 | int i; |
31e4c28d | 1217 | |
7876f930 TH |
1218 | mutex_lock(&blkcg_pol_mutex); |
1219 | ||
eb95419b | 1220 | if (!parent_css) { |
3c798398 | 1221 | blkcg = &blkcg_root; |
bc915e61 TH |
1222 | } else { |
1223 | blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL); | |
b5a9adcb | 1224 | if (!blkcg) |
4c18c9e9 | 1225 | goto unlock; |
e48453c3 AA |
1226 | } |
1227 | ||
3b8cc629 WL |
1228 | if (init_blkcg_llists(blkcg)) |
1229 | goto free_blkcg; | |
1230 | ||
e48453c3 AA |
1231 | for (i = 0; i < BLKCG_MAX_POLS ; i++) { |
1232 | struct blkcg_policy *pol = blkcg_policy[i]; | |
1233 | struct blkcg_policy_data *cpd; | |
1234 | ||
1235 | /* | |
1236 | * If the policy hasn't been attached yet, wait for it | |
1237 | * to be attached before doing anything else. Otherwise, | |
1238 | * check if the policy requires any specific per-cgroup | |
1239 | * data: if it does, allocate and initialize it. | |
1240 | */ | |
e4a9bde9 | 1241 | if (!pol || !pol->cpd_alloc_fn) |
e48453c3 AA |
1242 | continue; |
1243 | ||
e4a9bde9 | 1244 | cpd = pol->cpd_alloc_fn(GFP_KERNEL); |
b5a9adcb | 1245 | if (!cpd) |
e48453c3 | 1246 | goto free_pd_blkcg; |
b5a9adcb | 1247 | |
81437648 TH |
1248 | blkcg->cpd[i] = cpd; |
1249 | cpd->blkcg = blkcg; | |
e48453c3 | 1250 | cpd->plid = i; |
e4a9bde9 TH |
1251 | if (pol->cpd_init_fn) |
1252 | pol->cpd_init_fn(cpd); | |
e48453c3 | 1253 | } |
31e4c28d | 1254 | |
31e4c28d | 1255 | spin_lock_init(&blkcg->lock); |
d866dbf6 | 1256 | refcount_set(&blkcg->online_pin, 1); |
e00f4f4d | 1257 | INIT_RADIX_TREE(&blkcg->blkg_tree, GFP_NOWAIT | __GFP_NOWARN); |
31e4c28d | 1258 | INIT_HLIST_HEAD(&blkcg->blkg_list); |
52ebea74 TH |
1259 | #ifdef CONFIG_CGROUP_WRITEBACK |
1260 | INIT_LIST_HEAD(&blkcg->cgwb_list); | |
1261 | #endif | |
7876f930 TH |
1262 | list_add_tail(&blkcg->all_blkcgs_node, &all_blkcgs); |
1263 | ||
1264 | mutex_unlock(&blkcg_pol_mutex); | |
31e4c28d | 1265 | return &blkcg->css; |
e48453c3 AA |
1266 | |
1267 | free_pd_blkcg: | |
1268 | for (i--; i >= 0; i--) | |
e4a9bde9 TH |
1269 | if (blkcg->cpd[i]) |
1270 | blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]); | |
3b8cc629 WL |
1271 | free_percpu(blkcg->lhead); |
1272 | free_blkcg: | |
4c18c9e9 | 1273 | if (blkcg != &blkcg_root) |
1274 | kfree(blkcg); | |
1275 | unlock: | |
7876f930 | 1276 | mutex_unlock(&blkcg_pol_mutex); |
b5a9adcb | 1277 | return ERR_PTR(-ENOMEM); |
31e4c28d VG |
1278 | } |
1279 | ||
4308a434 TH |
1280 | static int blkcg_css_online(struct cgroup_subsys_state *css) |
1281 | { | |
397c9f46 | 1282 | struct blkcg *parent = blkcg_parent(css_to_blkcg(css)); |
4308a434 TH |
1283 | |
1284 | /* | |
1285 | * blkcg_pin_online() is used to delay blkcg offline so that blkgs | |
1286 | * don't go offline while cgwbs are still active on them. Pin the | |
1287 | * parent so that offline always happens towards the root. | |
1288 | */ | |
1289 | if (parent) | |
d7dbd43f | 1290 | blkcg_pin_online(&parent->css); |
4308a434 TH |
1291 | return 0; |
1292 | } | |
1293 | ||
9823538f | 1294 | int blkcg_init_disk(struct gendisk *disk) |
5efd6113 | 1295 | { |
9823538f | 1296 | struct request_queue *q = disk->queue; |
d708f0d5 JA |
1297 | struct blkcg_gq *new_blkg, *blkg; |
1298 | bool preloaded; | |
ec13b1d6 TH |
1299 | int ret; |
1300 | ||
1231039d CH |
1301 | INIT_LIST_HEAD(&q->blkg_list); |
1302 | mutex_init(&q->blkcg_mutex); | |
472e4314 | 1303 | |
99e60387 | 1304 | new_blkg = blkg_alloc(&blkcg_root, disk, GFP_KERNEL); |
d708f0d5 JA |
1305 | if (!new_blkg) |
1306 | return -ENOMEM; | |
1307 | ||
1308 | preloaded = !radix_tree_preload(GFP_KERNEL); | |
1309 | ||
bea54883 | 1310 | /* Make sure the root blkg exists. */ |
77c570a1 | 1311 | /* spin_lock_irq can serve as RCU read-side critical section. */ |
0d945c1f | 1312 | spin_lock_irq(&q->queue_lock); |
99e60387 | 1313 | blkg = blkg_create(&blkcg_root, disk, new_blkg); |
901932a3 JB |
1314 | if (IS_ERR(blkg)) |
1315 | goto err_unlock; | |
1231039d | 1316 | q->root_blkg = blkg; |
0d945c1f | 1317 | spin_unlock_irq(&q->queue_lock); |
ec13b1d6 | 1318 | |
d708f0d5 JA |
1319 | if (preloaded) |
1320 | radix_tree_preload_end(); | |
1321 | ||
b0dde3f5 | 1322 | ret = blk_ioprio_init(disk); |
556910e3 BVA |
1323 | if (ret) |
1324 | goto err_destroy_all; | |
1325 | ||
e13793ba | 1326 | ret = blk_throtl_init(disk); |
19688d7f | 1327 | if (ret) |
33dc6279 | 1328 | goto err_ioprio_exit; |
19688d7f | 1329 | |
16fac1b5 | 1330 | ret = blk_iolatency_init(disk); |
33dc6279 CH |
1331 | if (ret) |
1332 | goto err_throtl_exit; | |
6f5ddde4 | 1333 | |
04be60b5 | 1334 | return 0; |
901932a3 | 1335 | |
33dc6279 | 1336 | err_throtl_exit: |
e13793ba | 1337 | blk_throtl_exit(disk); |
33dc6279 | 1338 | err_ioprio_exit: |
b0dde3f5 | 1339 | blk_ioprio_exit(disk); |
04be60b5 | 1340 | err_destroy_all: |
00ad6991 | 1341 | blkg_destroy_all(disk); |
04be60b5 | 1342 | return ret; |
901932a3 | 1343 | err_unlock: |
0d945c1f | 1344 | spin_unlock_irq(&q->queue_lock); |
901932a3 JB |
1345 | if (preloaded) |
1346 | radix_tree_preload_end(); | |
1347 | return PTR_ERR(blkg); | |
5efd6113 TH |
1348 | } |
1349 | ||
9823538f | 1350 | void blkcg_exit_disk(struct gendisk *disk) |
5efd6113 | 1351 | { |
00ad6991 | 1352 | blkg_destroy_all(disk); |
813e6930 | 1353 | rq_qos_exit(disk->queue); |
e13793ba | 1354 | blk_throtl_exit(disk); |
5efd6113 TH |
1355 | } |
1356 | ||
69d7fde5 TH |
1357 | static void blkcg_bind(struct cgroup_subsys_state *root_css) |
1358 | { | |
1359 | int i; | |
1360 | ||
1361 | mutex_lock(&blkcg_pol_mutex); | |
1362 | ||
1363 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
1364 | struct blkcg_policy *pol = blkcg_policy[i]; | |
1365 | struct blkcg *blkcg; | |
1366 | ||
1367 | if (!pol || !pol->cpd_bind_fn) | |
1368 | continue; | |
1369 | ||
1370 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) | |
1371 | if (blkcg->cpd[pol->plid]) | |
1372 | pol->cpd_bind_fn(blkcg->cpd[pol->plid]); | |
1373 | } | |
1374 | mutex_unlock(&blkcg_pol_mutex); | |
1375 | } | |
1376 | ||
d09d8df3 JB |
1377 | static void blkcg_exit(struct task_struct *tsk) |
1378 | { | |
f05837ed CH |
1379 | if (tsk->throttle_disk) |
1380 | put_disk(tsk->throttle_disk); | |
1381 | tsk->throttle_disk = NULL; | |
d09d8df3 JB |
1382 | } |
1383 | ||
c165b3e3 | 1384 | struct cgroup_subsys io_cgrp_subsys = { |
92fb9748 | 1385 | .css_alloc = blkcg_css_alloc, |
4308a434 | 1386 | .css_online = blkcg_css_online, |
92fb9748 TH |
1387 | .css_offline = blkcg_css_offline, |
1388 | .css_free = blkcg_css_free, | |
f7331648 | 1389 | .css_rstat_flush = blkcg_rstat_flush, |
69d7fde5 | 1390 | .bind = blkcg_bind, |
2ee867dc | 1391 | .dfl_cftypes = blkcg_files, |
880f50e2 | 1392 | .legacy_cftypes = blkcg_legacy_files, |
c165b3e3 | 1393 | .legacy_name = "blkio", |
d09d8df3 | 1394 | .exit = blkcg_exit, |
1ced953b TH |
1395 | #ifdef CONFIG_MEMCG |
1396 | /* | |
1397 | * This ensures that, if available, memcg is automatically enabled | |
1398 | * together on the default hierarchy so that the owner cgroup can | |
1399 | * be retrieved from writeback pages. | |
1400 | */ | |
1401 | .depends_on = 1 << memory_cgrp_id, | |
1402 | #endif | |
676f7c8f | 1403 | }; |
c165b3e3 | 1404 | EXPORT_SYMBOL_GPL(io_cgrp_subsys); |
676f7c8f | 1405 | |
a2b1693b | 1406 | /** |
40e4996e CH |
1407 | * blkcg_activate_policy - activate a blkcg policy on a gendisk |
1408 | * @disk: gendisk of interest | |
a2b1693b TH |
1409 | * @pol: blkcg policy to activate |
1410 | * | |
40e4996e | 1411 | * Activate @pol on @disk. Requires %GFP_KERNEL context. @disk goes through |
a2b1693b TH |
1412 | * bypass mode to populate its blkgs with policy_data for @pol. |
1413 | * | |
40e4996e | 1414 | * Activation happens with @disk bypassed, so nobody would be accessing blkgs |
a2b1693b TH |
1415 | * from IO path. Update of each blkg is protected by both queue and blkcg |
1416 | * locks so that holding either lock and testing blkcg_policy_enabled() is | |
1417 | * always enough for dereferencing policy data. | |
1418 | * | |
1419 | * The caller is responsible for synchronizing [de]activations and policy | |
1420 | * [un]registerations. Returns 0 on success, -errno on failure. | |
1421 | */ | |
40e4996e | 1422 | int blkcg_activate_policy(struct gendisk *disk, const struct blkcg_policy *pol) |
a2b1693b | 1423 | { |
40e4996e | 1424 | struct request_queue *q = disk->queue; |
4c55f4f9 | 1425 | struct blkg_policy_data *pd_prealloc = NULL; |
9d179b86 | 1426 | struct blkcg_gq *blkg, *pinned_blkg = NULL; |
4c55f4f9 | 1427 | int ret; |
a2b1693b | 1428 | |
1231039d | 1429 | if (blkcg_policy_enabled(q, pol)) |
a2b1693b TH |
1430 | return 0; |
1431 | ||
344e9ffc | 1432 | if (queue_is_mq(q)) |
bd166ef1 | 1433 | blk_mq_freeze_queue(q); |
9d179b86 | 1434 | retry: |
0d945c1f | 1435 | spin_lock_irq(&q->queue_lock); |
a2b1693b | 1436 | |
9d179b86 | 1437 | /* blkg_list is pushed at the head, reverse walk to allocate parents first */ |
1231039d | 1438 | list_for_each_entry_reverse(blkg, &q->blkg_list, q_node) { |
4c55f4f9 TH |
1439 | struct blkg_policy_data *pd; |
1440 | ||
1441 | if (blkg->pd[pol->plid]) | |
1442 | continue; | |
a2b1693b | 1443 | |
9d179b86 TH |
1444 | /* If prealloc matches, use it; otherwise try GFP_NOWAIT */ |
1445 | if (blkg == pinned_blkg) { | |
1446 | pd = pd_prealloc; | |
1447 | pd_prealloc = NULL; | |
1448 | } else { | |
0a0b4f79 CH |
1449 | pd = pol->pd_alloc_fn(disk, blkg->blkcg, |
1450 | GFP_NOWAIT | __GFP_NOWARN); | |
9d179b86 TH |
1451 | } |
1452 | ||
4c55f4f9 | 1453 | if (!pd) { |
9d179b86 TH |
1454 | /* |
1455 | * GFP_NOWAIT failed. Free the existing one and | |
1456 | * prealloc for @blkg w/ GFP_KERNEL. | |
1457 | */ | |
1458 | if (pinned_blkg) | |
1459 | blkg_put(pinned_blkg); | |
1460 | blkg_get(blkg); | |
1461 | pinned_blkg = blkg; | |
1462 | ||
0d945c1f | 1463 | spin_unlock_irq(&q->queue_lock); |
9d179b86 TH |
1464 | |
1465 | if (pd_prealloc) | |
1466 | pol->pd_free_fn(pd_prealloc); | |
0a0b4f79 CH |
1467 | pd_prealloc = pol->pd_alloc_fn(disk, blkg->blkcg, |
1468 | GFP_KERNEL); | |
9d179b86 TH |
1469 | if (pd_prealloc) |
1470 | goto retry; | |
1471 | else | |
1472 | goto enomem; | |
4c55f4f9 | 1473 | } |
a2b1693b TH |
1474 | |
1475 | blkg->pd[pol->plid] = pd; | |
1476 | pd->blkg = blkg; | |
b276a876 | 1477 | pd->plid = pol->plid; |
dfd6200a | 1478 | pd->online = false; |
a2b1693b TH |
1479 | } |
1480 | ||
9d179b86 TH |
1481 | /* all allocated, init in the same order */ |
1482 | if (pol->pd_init_fn) | |
1231039d | 1483 | list_for_each_entry_reverse(blkg, &q->blkg_list, q_node) |
9d179b86 TH |
1484 | pol->pd_init_fn(blkg->pd[pol->plid]); |
1485 | ||
1231039d | 1486 | list_for_each_entry_reverse(blkg, &q->blkg_list, q_node) { |
dfd6200a | 1487 | if (pol->pd_online_fn) |
e3ff8887 | 1488 | pol->pd_online_fn(blkg->pd[pol->plid]); |
dfd6200a YK |
1489 | blkg->pd[pol->plid]->online = true; |
1490 | } | |
e3ff8887 | 1491 | |
1231039d | 1492 | __set_bit(pol->plid, q->blkcg_pols); |
a2b1693b | 1493 | ret = 0; |
4c55f4f9 | 1494 | |
0d945c1f | 1495 | spin_unlock_irq(&q->queue_lock); |
9d179b86 | 1496 | out: |
344e9ffc | 1497 | if (queue_is_mq(q)) |
bd166ef1 | 1498 | blk_mq_unfreeze_queue(q); |
9d179b86 TH |
1499 | if (pinned_blkg) |
1500 | blkg_put(pinned_blkg); | |
001bea73 TH |
1501 | if (pd_prealloc) |
1502 | pol->pd_free_fn(pd_prealloc); | |
a2b1693b | 1503 | return ret; |
9d179b86 TH |
1504 | |
1505 | enomem: | |
1506 | /* alloc failed, nothing's initialized yet, free everything */ | |
1507 | spin_lock_irq(&q->queue_lock); | |
1231039d | 1508 | list_for_each_entry(blkg, &q->blkg_list, q_node) { |
858560b2 LJ |
1509 | struct blkcg *blkcg = blkg->blkcg; |
1510 | ||
1511 | spin_lock(&blkcg->lock); | |
9d179b86 TH |
1512 | if (blkg->pd[pol->plid]) { |
1513 | pol->pd_free_fn(blkg->pd[pol->plid]); | |
1514 | blkg->pd[pol->plid] = NULL; | |
1515 | } | |
858560b2 | 1516 | spin_unlock(&blkcg->lock); |
9d179b86 TH |
1517 | } |
1518 | spin_unlock_irq(&q->queue_lock); | |
1519 | ret = -ENOMEM; | |
1520 | goto out; | |
a2b1693b TH |
1521 | } |
1522 | EXPORT_SYMBOL_GPL(blkcg_activate_policy); | |
1523 | ||
1524 | /** | |
40e4996e CH |
1525 | * blkcg_deactivate_policy - deactivate a blkcg policy on a gendisk |
1526 | * @disk: gendisk of interest | |
a2b1693b TH |
1527 | * @pol: blkcg policy to deactivate |
1528 | * | |
40e4996e | 1529 | * Deactivate @pol on @disk. Follows the same synchronization rules as |
a2b1693b TH |
1530 | * blkcg_activate_policy(). |
1531 | */ | |
40e4996e | 1532 | void blkcg_deactivate_policy(struct gendisk *disk, |
3c798398 | 1533 | const struct blkcg_policy *pol) |
a2b1693b | 1534 | { |
40e4996e | 1535 | struct request_queue *q = disk->queue; |
3c798398 | 1536 | struct blkcg_gq *blkg; |
a2b1693b | 1537 | |
1231039d | 1538 | if (!blkcg_policy_enabled(q, pol)) |
a2b1693b TH |
1539 | return; |
1540 | ||
344e9ffc | 1541 | if (queue_is_mq(q)) |
bd166ef1 | 1542 | blk_mq_freeze_queue(q); |
bd166ef1 | 1543 | |
1231039d | 1544 | mutex_lock(&q->blkcg_mutex); |
0d945c1f | 1545 | spin_lock_irq(&q->queue_lock); |
a2b1693b | 1546 | |
1231039d | 1547 | __clear_bit(pol->plid, q->blkcg_pols); |
a2b1693b | 1548 | |
1231039d | 1549 | list_for_each_entry(blkg, &q->blkg_list, q_node) { |
858560b2 LJ |
1550 | struct blkcg *blkcg = blkg->blkcg; |
1551 | ||
1552 | spin_lock(&blkcg->lock); | |
001bea73 | 1553 | if (blkg->pd[pol->plid]) { |
dfd6200a | 1554 | if (blkg->pd[pol->plid]->online && pol->pd_offline_fn) |
a9520cd6 | 1555 | pol->pd_offline_fn(blkg->pd[pol->plid]); |
001bea73 TH |
1556 | pol->pd_free_fn(blkg->pd[pol->plid]); |
1557 | blkg->pd[pol->plid] = NULL; | |
1558 | } | |
858560b2 | 1559 | spin_unlock(&blkcg->lock); |
a2b1693b TH |
1560 | } |
1561 | ||
0d945c1f | 1562 | spin_unlock_irq(&q->queue_lock); |
1231039d | 1563 | mutex_unlock(&q->blkcg_mutex); |
bd166ef1 | 1564 | |
344e9ffc | 1565 | if (queue_is_mq(q)) |
bd166ef1 | 1566 | blk_mq_unfreeze_queue(q); |
a2b1693b TH |
1567 | } |
1568 | EXPORT_SYMBOL_GPL(blkcg_deactivate_policy); | |
1569 | ||
e55cf798 JY |
1570 | static void blkcg_free_all_cpd(struct blkcg_policy *pol) |
1571 | { | |
1572 | struct blkcg *blkcg; | |
1573 | ||
1574 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) { | |
1575 | if (blkcg->cpd[pol->plid]) { | |
1576 | pol->cpd_free_fn(blkcg->cpd[pol->plid]); | |
1577 | blkcg->cpd[pol->plid] = NULL; | |
1578 | } | |
1579 | } | |
1580 | } | |
1581 | ||
8bd435b3 | 1582 | /** |
3c798398 TH |
1583 | * blkcg_policy_register - register a blkcg policy |
1584 | * @pol: blkcg policy to register | |
8bd435b3 | 1585 | * |
3c798398 TH |
1586 | * Register @pol with blkcg core. Might sleep and @pol may be modified on |
1587 | * successful registration. Returns 0 on success and -errno on failure. | |
8bd435b3 | 1588 | */ |
d5bf0291 | 1589 | int blkcg_policy_register(struct blkcg_policy *pol) |
3e252066 | 1590 | { |
06b285bd | 1591 | struct blkcg *blkcg; |
8bd435b3 | 1592 | int i, ret; |
e8989fae | 1593 | |
838f13bf | 1594 | mutex_lock(&blkcg_pol_register_mutex); |
bc0d6501 TH |
1595 | mutex_lock(&blkcg_pol_mutex); |
1596 | ||
8bd435b3 TH |
1597 | /* find an empty slot */ |
1598 | ret = -ENOSPC; | |
1599 | for (i = 0; i < BLKCG_MAX_POLS; i++) | |
3c798398 | 1600 | if (!blkcg_policy[i]) |
8bd435b3 | 1601 | break; |
01c5f85a JA |
1602 | if (i >= BLKCG_MAX_POLS) { |
1603 | pr_warn("blkcg_policy_register: BLKCG_MAX_POLS too small\n"); | |
838f13bf | 1604 | goto err_unlock; |
01c5f85a | 1605 | } |
035d10b2 | 1606 | |
e8401073 | 1607 | /* Make sure cpd/pd_alloc_fn and cpd/pd_free_fn in pairs */ |
1608 | if ((!pol->cpd_alloc_fn ^ !pol->cpd_free_fn) || | |
1609 | (!pol->pd_alloc_fn ^ !pol->pd_free_fn)) | |
1610 | goto err_unlock; | |
1611 | ||
06b285bd | 1612 | /* register @pol */ |
3c798398 | 1613 | pol->plid = i; |
06b285bd TH |
1614 | blkcg_policy[pol->plid] = pol; |
1615 | ||
1616 | /* allocate and install cpd's */ | |
e4a9bde9 | 1617 | if (pol->cpd_alloc_fn) { |
06b285bd TH |
1618 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) { |
1619 | struct blkcg_policy_data *cpd; | |
1620 | ||
e4a9bde9 | 1621 | cpd = pol->cpd_alloc_fn(GFP_KERNEL); |
bbb427e3 | 1622 | if (!cpd) |
06b285bd | 1623 | goto err_free_cpds; |
06b285bd | 1624 | |
81437648 TH |
1625 | blkcg->cpd[pol->plid] = cpd; |
1626 | cpd->blkcg = blkcg; | |
06b285bd | 1627 | cpd->plid = pol->plid; |
86a5bba5 TH |
1628 | if (pol->cpd_init_fn) |
1629 | pol->cpd_init_fn(cpd); | |
06b285bd TH |
1630 | } |
1631 | } | |
1632 | ||
838f13bf | 1633 | mutex_unlock(&blkcg_pol_mutex); |
8bd435b3 | 1634 | |
8bd435b3 | 1635 | /* everything is in place, add intf files for the new policy */ |
2ee867dc TH |
1636 | if (pol->dfl_cftypes) |
1637 | WARN_ON(cgroup_add_dfl_cftypes(&io_cgrp_subsys, | |
1638 | pol->dfl_cftypes)); | |
880f50e2 | 1639 | if (pol->legacy_cftypes) |
c165b3e3 | 1640 | WARN_ON(cgroup_add_legacy_cftypes(&io_cgrp_subsys, |
880f50e2 | 1641 | pol->legacy_cftypes)); |
838f13bf TH |
1642 | mutex_unlock(&blkcg_pol_register_mutex); |
1643 | return 0; | |
1644 | ||
06b285bd | 1645 | err_free_cpds: |
e55cf798 JY |
1646 | if (pol->cpd_free_fn) |
1647 | blkcg_free_all_cpd(pol); | |
1648 | ||
06b285bd | 1649 | blkcg_policy[pol->plid] = NULL; |
838f13bf | 1650 | err_unlock: |
bc0d6501 | 1651 | mutex_unlock(&blkcg_pol_mutex); |
838f13bf | 1652 | mutex_unlock(&blkcg_pol_register_mutex); |
8bd435b3 | 1653 | return ret; |
3e252066 | 1654 | } |
3c798398 | 1655 | EXPORT_SYMBOL_GPL(blkcg_policy_register); |
3e252066 | 1656 | |
8bd435b3 | 1657 | /** |
3c798398 TH |
1658 | * blkcg_policy_unregister - unregister a blkcg policy |
1659 | * @pol: blkcg policy to unregister | |
8bd435b3 | 1660 | * |
3c798398 | 1661 | * Undo blkcg_policy_register(@pol). Might sleep. |
8bd435b3 | 1662 | */ |
3c798398 | 1663 | void blkcg_policy_unregister(struct blkcg_policy *pol) |
3e252066 | 1664 | { |
838f13bf | 1665 | mutex_lock(&blkcg_pol_register_mutex); |
bc0d6501 | 1666 | |
3c798398 | 1667 | if (WARN_ON(blkcg_policy[pol->plid] != pol)) |
8bd435b3 TH |
1668 | goto out_unlock; |
1669 | ||
1670 | /* kill the intf files first */ | |
2ee867dc TH |
1671 | if (pol->dfl_cftypes) |
1672 | cgroup_rm_cftypes(pol->dfl_cftypes); | |
880f50e2 TH |
1673 | if (pol->legacy_cftypes) |
1674 | cgroup_rm_cftypes(pol->legacy_cftypes); | |
44ea53de | 1675 | |
06b285bd | 1676 | /* remove cpds and unregister */ |
838f13bf | 1677 | mutex_lock(&blkcg_pol_mutex); |
06b285bd | 1678 | |
e55cf798 JY |
1679 | if (pol->cpd_free_fn) |
1680 | blkcg_free_all_cpd(pol); | |
1681 | ||
3c798398 | 1682 | blkcg_policy[pol->plid] = NULL; |
06b285bd | 1683 | |
bc0d6501 | 1684 | mutex_unlock(&blkcg_pol_mutex); |
838f13bf TH |
1685 | out_unlock: |
1686 | mutex_unlock(&blkcg_pol_register_mutex); | |
3e252066 | 1687 | } |
3c798398 | 1688 | EXPORT_SYMBOL_GPL(blkcg_policy_unregister); |
903d23f0 | 1689 | |
d3f77dfd TH |
1690 | bool __blkcg_punt_bio_submit(struct bio *bio) |
1691 | { | |
1692 | struct blkcg_gq *blkg = bio->bi_blkg; | |
1693 | ||
1694 | /* consume the flag first */ | |
1695 | bio->bi_opf &= ~REQ_CGROUP_PUNT; | |
1696 | ||
1697 | /* never bounce for the root cgroup */ | |
1698 | if (!blkg->parent) | |
1699 | return false; | |
1700 | ||
1701 | spin_lock_bh(&blkg->async_bio_lock); | |
1702 | bio_list_add(&blkg->async_bios, bio); | |
1703 | spin_unlock_bh(&blkg->async_bio_lock); | |
1704 | ||
1705 | queue_work(blkcg_punt_bio_wq, &blkg->async_bio_work); | |
1706 | return true; | |
1707 | } | |
1708 | ||
d09d8df3 JB |
1709 | /* |
1710 | * Scale the accumulated delay based on how long it has been since we updated | |
1711 | * the delay. We only call this when we are adding delay, in case it's been a | |
1712 | * while since we added delay, and when we are checking to see if we need to | |
1713 | * delay a task, to account for any delays that may have occurred. | |
1714 | */ | |
1715 | static void blkcg_scale_delay(struct blkcg_gq *blkg, u64 now) | |
1716 | { | |
1717 | u64 old = atomic64_read(&blkg->delay_start); | |
1718 | ||
54c52e10 TH |
1719 | /* negative use_delay means no scaling, see blkcg_set_delay() */ |
1720 | if (atomic_read(&blkg->use_delay) < 0) | |
1721 | return; | |
1722 | ||
d09d8df3 JB |
1723 | /* |
1724 | * We only want to scale down every second. The idea here is that we | |
1725 | * want to delay people for min(delay_nsec, NSEC_PER_SEC) in a certain | |
1726 | * time window. We only want to throttle tasks for recent delay that | |
1727 | * has occurred, in 1 second time windows since that's the maximum | |
1728 | * things can be throttled. We save the current delay window in | |
1729 | * blkg->last_delay so we know what amount is still left to be charged | |
1730 | * to the blkg from this point onward. blkg->last_use keeps track of | |
1731 | * the use_delay counter. The idea is if we're unthrottling the blkg we | |
1732 | * are ok with whatever is happening now, and we can take away more of | |
1733 | * the accumulated delay as we've already throttled enough that | |
1734 | * everybody is happy with their IO latencies. | |
1735 | */ | |
1736 | if (time_before64(old + NSEC_PER_SEC, now) && | |
96388f57 | 1737 | atomic64_try_cmpxchg(&blkg->delay_start, &old, now)) { |
d09d8df3 JB |
1738 | u64 cur = atomic64_read(&blkg->delay_nsec); |
1739 | u64 sub = min_t(u64, blkg->last_delay, now - old); | |
1740 | int cur_use = atomic_read(&blkg->use_delay); | |
1741 | ||
1742 | /* | |
1743 | * We've been unthrottled, subtract a larger chunk of our | |
1744 | * accumulated delay. | |
1745 | */ | |
1746 | if (cur_use < blkg->last_use) | |
1747 | sub = max_t(u64, sub, blkg->last_delay >> 1); | |
1748 | ||
1749 | /* | |
1750 | * This shouldn't happen, but handle it anyway. Our delay_nsec | |
1751 | * should only ever be growing except here where we subtract out | |
1752 | * min(last_delay, 1 second), but lord knows bugs happen and I'd | |
1753 | * rather not end up with negative numbers. | |
1754 | */ | |
1755 | if (unlikely(cur < sub)) { | |
1756 | atomic64_set(&blkg->delay_nsec, 0); | |
1757 | blkg->last_delay = 0; | |
1758 | } else { | |
1759 | atomic64_sub(sub, &blkg->delay_nsec); | |
1760 | blkg->last_delay = cur - sub; | |
1761 | } | |
1762 | blkg->last_use = cur_use; | |
1763 | } | |
1764 | } | |
1765 | ||
1766 | /* | |
1767 | * This is called when we want to actually walk up the hierarchy and check to | |
1768 | * see if we need to throttle, and then actually throttle if there is some | |
1769 | * accumulated delay. This should only be called upon return to user space so | |
1770 | * we're not holding some lock that would induce a priority inversion. | |
1771 | */ | |
1772 | static void blkcg_maybe_throttle_blkg(struct blkcg_gq *blkg, bool use_memdelay) | |
1773 | { | |
fd112c74 | 1774 | unsigned long pflags; |
5160a5a5 | 1775 | bool clamp; |
d09d8df3 JB |
1776 | u64 now = ktime_to_ns(ktime_get()); |
1777 | u64 exp; | |
1778 | u64 delay_nsec = 0; | |
1779 | int tok; | |
1780 | ||
1781 | while (blkg->parent) { | |
5160a5a5 TH |
1782 | int use_delay = atomic_read(&blkg->use_delay); |
1783 | ||
1784 | if (use_delay) { | |
1785 | u64 this_delay; | |
1786 | ||
d09d8df3 | 1787 | blkcg_scale_delay(blkg, now); |
5160a5a5 TH |
1788 | this_delay = atomic64_read(&blkg->delay_nsec); |
1789 | if (this_delay > delay_nsec) { | |
1790 | delay_nsec = this_delay; | |
1791 | clamp = use_delay > 0; | |
1792 | } | |
d09d8df3 JB |
1793 | } |
1794 | blkg = blkg->parent; | |
1795 | } | |
1796 | ||
1797 | if (!delay_nsec) | |
1798 | return; | |
1799 | ||
1800 | /* | |
1801 | * Let's not sleep for all eternity if we've amassed a huge delay. | |
1802 | * Swapping or metadata IO can accumulate 10's of seconds worth of | |
1803 | * delay, and we want userspace to be able to do _something_ so cap the | |
5160a5a5 TH |
1804 | * delays at 0.25s. If there's 10's of seconds worth of delay then the |
1805 | * tasks will be delayed for 0.25 second for every syscall. If | |
1806 | * blkcg_set_delay() was used as indicated by negative use_delay, the | |
1807 | * caller is responsible for regulating the range. | |
d09d8df3 | 1808 | */ |
5160a5a5 TH |
1809 | if (clamp) |
1810 | delay_nsec = min_t(u64, delay_nsec, 250 * NSEC_PER_MSEC); | |
d09d8df3 | 1811 | |
fd112c74 JB |
1812 | if (use_memdelay) |
1813 | psi_memstall_enter(&pflags); | |
d09d8df3 JB |
1814 | |
1815 | exp = ktime_add_ns(now, delay_nsec); | |
1816 | tok = io_schedule_prepare(); | |
1817 | do { | |
1818 | __set_current_state(TASK_KILLABLE); | |
1819 | if (!schedule_hrtimeout(&exp, HRTIMER_MODE_ABS)) | |
1820 | break; | |
1821 | } while (!fatal_signal_pending(current)); | |
1822 | io_schedule_finish(tok); | |
fd112c74 JB |
1823 | |
1824 | if (use_memdelay) | |
1825 | psi_memstall_leave(&pflags); | |
d09d8df3 JB |
1826 | } |
1827 | ||
1828 | /** | |
1829 | * blkcg_maybe_throttle_current - throttle the current task if it has been marked | |
1830 | * | |
1831 | * This is only called if we've been marked with set_notify_resume(). Obviously | |
1832 | * we can be set_notify_resume() for reasons other than blkcg throttling, so we | |
f05837ed | 1833 | * check to see if current->throttle_disk is set and if not this doesn't do |
d09d8df3 JB |
1834 | * anything. This should only ever be called by the resume code, it's not meant |
1835 | * to be called by people willy-nilly as it will actually do the work to | |
1836 | * throttle the task if it is setup for throttling. | |
1837 | */ | |
1838 | void blkcg_maybe_throttle_current(void) | |
1839 | { | |
f05837ed | 1840 | struct gendisk *disk = current->throttle_disk; |
d09d8df3 JB |
1841 | struct blkcg *blkcg; |
1842 | struct blkcg_gq *blkg; | |
1843 | bool use_memdelay = current->use_memdelay; | |
1844 | ||
f05837ed | 1845 | if (!disk) |
d09d8df3 JB |
1846 | return; |
1847 | ||
f05837ed | 1848 | current->throttle_disk = NULL; |
d09d8df3 JB |
1849 | current->use_memdelay = false; |
1850 | ||
1851 | rcu_read_lock(); | |
82778259 | 1852 | blkcg = css_to_blkcg(blkcg_css()); |
d09d8df3 JB |
1853 | if (!blkcg) |
1854 | goto out; | |
9a9c261e | 1855 | blkg = blkg_lookup(blkcg, disk->queue); |
d09d8df3 JB |
1856 | if (!blkg) |
1857 | goto out; | |
7754f669 | 1858 | if (!blkg_tryget(blkg)) |
d09d8df3 JB |
1859 | goto out; |
1860 | rcu_read_unlock(); | |
d09d8df3 JB |
1861 | |
1862 | blkcg_maybe_throttle_blkg(blkg, use_memdelay); | |
1863 | blkg_put(blkg); | |
f05837ed | 1864 | put_disk(disk); |
d09d8df3 JB |
1865 | return; |
1866 | out: | |
1867 | rcu_read_unlock(); | |
d09d8df3 | 1868 | } |
d09d8df3 JB |
1869 | |
1870 | /** | |
1871 | * blkcg_schedule_throttle - this task needs to check for throttling | |
1d6df9d3 | 1872 | * @disk: disk to throttle |
537d71b3 | 1873 | * @use_memdelay: do we charge this to memory delay for PSI |
d09d8df3 JB |
1874 | * |
1875 | * This is called by the IO controller when we know there's delay accumulated | |
1876 | * for the blkg for this task. We do not pass the blkg because there are places | |
1877 | * we call this that may not have that information, the swapping code for | |
de185b56 | 1878 | * instance will only have a block_device at that point. This set's the |
d09d8df3 JB |
1879 | * notify_resume for the task to check and see if it requires throttling before |
1880 | * returning to user space. | |
1881 | * | |
1882 | * We will only schedule once per syscall. You can call this over and over | |
1883 | * again and it will only do the check once upon return to user space, and only | |
1884 | * throttle once. If the task needs to be throttled again it'll need to be | |
1885 | * re-set at the next time we see the task. | |
1886 | */ | |
de185b56 | 1887 | void blkcg_schedule_throttle(struct gendisk *disk, bool use_memdelay) |
d09d8df3 JB |
1888 | { |
1889 | if (unlikely(current->flags & PF_KTHREAD)) | |
1890 | return; | |
1891 | ||
f05837ed CH |
1892 | if (current->throttle_disk != disk) { |
1893 | if (test_bit(GD_DEAD, &disk->state)) | |
49d1822b | 1894 | return; |
f05837ed | 1895 | get_device(disk_to_dev(disk)); |
49d1822b | 1896 | |
f05837ed CH |
1897 | if (current->throttle_disk) |
1898 | put_disk(current->throttle_disk); | |
1899 | current->throttle_disk = disk; | |
49d1822b | 1900 | } |
d09d8df3 | 1901 | |
d09d8df3 JB |
1902 | if (use_memdelay) |
1903 | current->use_memdelay = use_memdelay; | |
1904 | set_notify_resume(current); | |
1905 | } | |
d09d8df3 JB |
1906 | |
1907 | /** | |
1908 | * blkcg_add_delay - add delay to this blkg | |
537d71b3 BVA |
1909 | * @blkg: blkg of interest |
1910 | * @now: the current time in nanoseconds | |
1911 | * @delta: how many nanoseconds of delay to add | |
d09d8df3 JB |
1912 | * |
1913 | * Charge @delta to the blkg's current delay accumulation. This is used to | |
1914 | * throttle tasks if an IO controller thinks we need more throttling. | |
1915 | */ | |
1916 | void blkcg_add_delay(struct blkcg_gq *blkg, u64 now, u64 delta) | |
1917 | { | |
54c52e10 TH |
1918 | if (WARN_ON_ONCE(atomic_read(&blkg->use_delay) < 0)) |
1919 | return; | |
d09d8df3 JB |
1920 | blkcg_scale_delay(blkg, now); |
1921 | atomic64_add(delta, &blkg->delay_nsec); | |
1922 | } | |
d09d8df3 | 1923 | |
28fc591f CH |
1924 | /** |
1925 | * blkg_tryget_closest - try and get a blkg ref on the closet blkg | |
13c7863d CH |
1926 | * @bio: target bio |
1927 | * @css: target css | |
28fc591f | 1928 | * |
13c7863d CH |
1929 | * As the failure mode here is to walk up the blkg tree, this ensure that the |
1930 | * blkg->parent pointers are always valid. This returns the blkg that it ended | |
1931 | * up taking a reference on or %NULL if no reference was taken. | |
28fc591f | 1932 | */ |
13c7863d CH |
1933 | static inline struct blkcg_gq *blkg_tryget_closest(struct bio *bio, |
1934 | struct cgroup_subsys_state *css) | |
28fc591f | 1935 | { |
13c7863d | 1936 | struct blkcg_gq *blkg, *ret_blkg = NULL; |
28fc591f | 1937 | |
13c7863d | 1938 | rcu_read_lock(); |
99e60387 | 1939 | blkg = blkg_lookup_create(css_to_blkcg(css), bio->bi_bdev->bd_disk); |
28fc591f CH |
1940 | while (blkg) { |
1941 | if (blkg_tryget(blkg)) { | |
1942 | ret_blkg = blkg; | |
1943 | break; | |
1944 | } | |
1945 | blkg = blkg->parent; | |
1946 | } | |
13c7863d | 1947 | rcu_read_unlock(); |
28fc591f CH |
1948 | |
1949 | return ret_blkg; | |
1950 | } | |
1951 | ||
1952 | /** | |
1953 | * bio_associate_blkg_from_css - associate a bio with a specified css | |
1954 | * @bio: target bio | |
1955 | * @css: target css | |
1956 | * | |
1957 | * Associate @bio with the blkg found by combining the css's blkg and the | |
1958 | * request_queue of the @bio. An association failure is handled by walking up | |
1959 | * the blkg tree. Therefore, the blkg associated can be anything between @blkg | |
1231039d | 1960 | * and q->root_blkg. This situation only happens when a cgroup is dying and |
28fc591f CH |
1961 | * then the remaining bios will spill to the closest alive blkg. |
1962 | * | |
1963 | * A reference will be taken on the blkg and will be released when @bio is | |
1964 | * freed. | |
1965 | */ | |
1966 | void bio_associate_blkg_from_css(struct bio *bio, | |
1967 | struct cgroup_subsys_state *css) | |
1968 | { | |
28fc591f CH |
1969 | if (bio->bi_blkg) |
1970 | blkg_put(bio->bi_blkg); | |
1971 | ||
a5b97526 | 1972 | if (css && css->parent) { |
13c7863d | 1973 | bio->bi_blkg = blkg_tryget_closest(bio, css); |
a5b97526 | 1974 | } else { |
1231039d CH |
1975 | blkg_get(bdev_get_queue(bio->bi_bdev)->root_blkg); |
1976 | bio->bi_blkg = bdev_get_queue(bio->bi_bdev)->root_blkg; | |
a5b97526 | 1977 | } |
28fc591f CH |
1978 | } |
1979 | EXPORT_SYMBOL_GPL(bio_associate_blkg_from_css); | |
1980 | ||
1981 | /** | |
1982 | * bio_associate_blkg - associate a bio with a blkg | |
1983 | * @bio: target bio | |
1984 | * | |
1985 | * Associate @bio with the blkg found from the bio's css and request_queue. | |
1986 | * If one is not found, bio_lookup_blkg() creates the blkg. If a blkg is | |
1987 | * already associated, the css is reused and association redone as the | |
1988 | * request_queue may have changed. | |
1989 | */ | |
1990 | void bio_associate_blkg(struct bio *bio) | |
1991 | { | |
1992 | struct cgroup_subsys_state *css; | |
1993 | ||
1994 | rcu_read_lock(); | |
1995 | ||
1996 | if (bio->bi_blkg) | |
bbb1ebe7 | 1997 | css = bio_blkcg_css(bio); |
28fc591f CH |
1998 | else |
1999 | css = blkcg_css(); | |
2000 | ||
2001 | bio_associate_blkg_from_css(bio, css); | |
2002 | ||
2003 | rcu_read_unlock(); | |
2004 | } | |
2005 | EXPORT_SYMBOL_GPL(bio_associate_blkg); | |
2006 | ||
2007 | /** | |
2008 | * bio_clone_blkg_association - clone blkg association from src to dst bio | |
2009 | * @dst: destination bio | |
2010 | * @src: source bio | |
2011 | */ | |
2012 | void bio_clone_blkg_association(struct bio *dst, struct bio *src) | |
2013 | { | |
22b106e5 JK |
2014 | if (src->bi_blkg) |
2015 | bio_associate_blkg_from_css(dst, bio_blkcg_css(src)); | |
28fc591f CH |
2016 | } |
2017 | EXPORT_SYMBOL_GPL(bio_clone_blkg_association); | |
2018 | ||
db18a53e CH |
2019 | static int blk_cgroup_io_type(struct bio *bio) |
2020 | { | |
2021 | if (op_is_discard(bio->bi_opf)) | |
2022 | return BLKG_IOSTAT_DISCARD; | |
2023 | if (op_is_write(bio->bi_opf)) | |
2024 | return BLKG_IOSTAT_WRITE; | |
2025 | return BLKG_IOSTAT_READ; | |
2026 | } | |
2027 | ||
2028 | void blk_cgroup_bio_start(struct bio *bio) | |
2029 | { | |
3b8cc629 | 2030 | struct blkcg *blkcg = bio->bi_blkg->blkcg; |
db18a53e CH |
2031 | int rwd = blk_cgroup_io_type(bio), cpu; |
2032 | struct blkg_iostat_set *bis; | |
3c08b093 | 2033 | unsigned long flags; |
db18a53e CH |
2034 | |
2035 | cpu = get_cpu(); | |
2036 | bis = per_cpu_ptr(bio->bi_blkg->iostat_cpu, cpu); | |
3c08b093 | 2037 | flags = u64_stats_update_begin_irqsave(&bis->sync); |
db18a53e CH |
2038 | |
2039 | /* | |
2040 | * If the bio is flagged with BIO_CGROUP_ACCT it means this is a split | |
2041 | * bio and we would have already accounted for the size of the bio. | |
2042 | */ | |
2043 | if (!bio_flagged(bio, BIO_CGROUP_ACCT)) { | |
2044 | bio_set_flag(bio, BIO_CGROUP_ACCT); | |
0b8cc25d | 2045 | bis->cur.bytes[rwd] += bio->bi_iter.bi_size; |
db18a53e CH |
2046 | } |
2047 | bis->cur.ios[rwd]++; | |
2048 | ||
3b8cc629 WL |
2049 | /* |
2050 | * If the iostat_cpu isn't in a lockless list, put it into the | |
2051 | * list to indicate that a stat update is pending. | |
2052 | */ | |
2053 | if (!READ_ONCE(bis->lqueued)) { | |
2054 | struct llist_head *lhead = this_cpu_ptr(blkcg->lhead); | |
2055 | ||
2056 | llist_add(&bis->lnode, lhead); | |
2057 | WRITE_ONCE(bis->lqueued, true); | |
2058 | percpu_ref_get(&bis->blkg->refcnt); | |
2059 | } | |
2060 | ||
3c08b093 | 2061 | u64_stats_update_end_irqrestore(&bis->sync, flags); |
db18a53e | 2062 | if (cgroup_subsys_on_dfl(io_cgrp_subsys)) |
3b8cc629 | 2063 | cgroup_rstat_updated(blkcg->css.cgroup, cpu); |
db18a53e CH |
2064 | put_cpu(); |
2065 | } | |
2066 | ||
216889aa CH |
2067 | bool blk_cgroup_congested(void) |
2068 | { | |
2069 | struct cgroup_subsys_state *css; | |
2070 | bool ret = false; | |
2071 | ||
2072 | rcu_read_lock(); | |
d200ca14 | 2073 | for (css = blkcg_css(); css; css = css->parent) { |
216889aa CH |
2074 | if (atomic_read(&css->cgroup->congestion_count)) { |
2075 | ret = true; | |
2076 | break; | |
2077 | } | |
216889aa CH |
2078 | } |
2079 | rcu_read_unlock(); | |
2080 | return ret; | |
2081 | } | |
2082 | ||
d3f77dfd TH |
2083 | static int __init blkcg_init(void) |
2084 | { | |
2085 | blkcg_punt_bio_wq = alloc_workqueue("blkcg_punt_bio", | |
2086 | WQ_MEM_RECLAIM | WQ_FREEZABLE | | |
2087 | WQ_UNBOUND | WQ_SYSFS, 0); | |
2088 | if (!blkcg_punt_bio_wq) | |
2089 | return -ENOMEM; | |
2090 | return 0; | |
2091 | } | |
2092 | subsys_initcall(blkcg_init); | |
2093 | ||
903d23f0 JB |
2094 | module_param(blkcg_debug_stats, bool, 0644); |
2095 | MODULE_PARM_DESC(blkcg_debug_stats, "True if you want debug stats, false if not"); |