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 | ||
a2b1693b | 111 | static bool blkcg_policy_enabled(struct request_queue *q, |
3c798398 | 112 | const struct blkcg_policy *pol) |
a2b1693b TH |
113 | { |
114 | return pol && test_bit(pol->plid, q->blkcg_pols); | |
115 | } | |
116 | ||
d578c770 | 117 | static void blkg_free_workfn(struct work_struct *work) |
0381411e | 118 | { |
d578c770 ML |
119 | struct blkcg_gq *blkg = container_of(work, struct blkcg_gq, |
120 | free_work); | |
a06377c5 | 121 | struct request_queue *q = blkg->q; |
e8989fae | 122 | int i; |
549d3aa8 | 123 | |
f1c006f1 YK |
124 | /* |
125 | * pd_free_fn() can also be called from blkcg_deactivate_policy(), | |
126 | * in order to make sure pd_free_fn() is called in order, the deletion | |
1231039d | 127 | * of the list blkg->q_node is delayed to here from blkg_destroy(), and |
f1c006f1 YK |
128 | * blkcg_mutex is used to synchronize blkg_free_workfn() and |
129 | * blkcg_deactivate_policy(). | |
130 | */ | |
a06377c5 | 131 | mutex_lock(&q->blkcg_mutex); |
db613670 | 132 | for (i = 0; i < BLKCG_MAX_POLS; i++) |
001bea73 TH |
133 | if (blkg->pd[i]) |
134 | blkcg_policy[i]->pd_free_fn(blkg->pd[i]); | |
c7241bab YK |
135 | if (blkg->parent) |
136 | blkg_put(blkg->parent); | |
1231039d | 137 | list_del_init(&blkg->q_node); |
a06377c5 | 138 | mutex_unlock(&q->blkcg_mutex); |
e8989fae | 139 | |
a06377c5 | 140 | blk_put_queue(q); |
f7331648 | 141 | free_percpu(blkg->iostat_cpu); |
ef069b97 | 142 | percpu_ref_exit(&blkg->refcnt); |
549d3aa8 | 143 | kfree(blkg); |
0381411e TH |
144 | } |
145 | ||
d578c770 ML |
146 | /** |
147 | * blkg_free - free a blkg | |
148 | * @blkg: blkg to free | |
149 | * | |
150 | * Free @blkg which may be partially allocated. | |
151 | */ | |
152 | static void blkg_free(struct blkcg_gq *blkg) | |
153 | { | |
154 | if (!blkg) | |
155 | return; | |
156 | ||
157 | /* | |
158 | * Both ->pd_free_fn() and request queue's release handler may | |
159 | * sleep, so free us by scheduling one work func | |
160 | */ | |
161 | INIT_WORK(&blkg->free_work, blkg_free_workfn); | |
162 | schedule_work(&blkg->free_work); | |
163 | } | |
164 | ||
7fcf2b03 DZ |
165 | static void __blkg_release(struct rcu_head *rcu) |
166 | { | |
167 | struct blkcg_gq *blkg = container_of(rcu, struct blkcg_gq, rcu_head); | |
168 | ||
d3f77dfd TH |
169 | WARN_ON(!bio_list_empty(&blkg->async_bios)); |
170 | ||
7fcf2b03 DZ |
171 | /* release the blkcg and parent blkg refs this blkg has been holding */ |
172 | css_put(&blkg->blkcg->css); | |
7fcf2b03 DZ |
173 | blkg_free(blkg); |
174 | } | |
175 | ||
176 | /* | |
177 | * A group is RCU protected, but having an rcu lock does not mean that one | |
178 | * can access all the fields of blkg and assume these are valid. For | |
179 | * example, don't try to follow throtl_data and request queue links. | |
180 | * | |
181 | * Having a reference to blkg under an rcu allows accesses to only values | |
182 | * local to groups like group stats and group rate limits. | |
183 | */ | |
184 | static void blkg_release(struct percpu_ref *ref) | |
185 | { | |
186 | struct blkcg_gq *blkg = container_of(ref, struct blkcg_gq, refcnt); | |
187 | ||
188 | call_rcu(&blkg->rcu_head, __blkg_release); | |
189 | } | |
190 | ||
d3f77dfd TH |
191 | static void blkg_async_bio_workfn(struct work_struct *work) |
192 | { | |
193 | struct blkcg_gq *blkg = container_of(work, struct blkcg_gq, | |
194 | async_bio_work); | |
195 | struct bio_list bios = BIO_EMPTY_LIST; | |
196 | struct bio *bio; | |
192f1c6b XT |
197 | struct blk_plug plug; |
198 | bool need_plug = false; | |
d3f77dfd TH |
199 | |
200 | /* as long as there are pending bios, @blkg can't go away */ | |
201 | spin_lock_bh(&blkg->async_bio_lock); | |
202 | bio_list_merge(&bios, &blkg->async_bios); | |
203 | bio_list_init(&blkg->async_bios); | |
204 | spin_unlock_bh(&blkg->async_bio_lock); | |
205 | ||
192f1c6b XT |
206 | /* start plug only when bio_list contains at least 2 bios */ |
207 | if (bios.head && bios.head->bi_next) { | |
208 | need_plug = true; | |
209 | blk_start_plug(&plug); | |
210 | } | |
d3f77dfd TH |
211 | while ((bio = bio_list_pop(&bios))) |
212 | submit_bio(bio); | |
192f1c6b XT |
213 | if (need_plug) |
214 | blk_finish_plug(&plug); | |
d3f77dfd TH |
215 | } |
216 | ||
bbb1ebe7 CH |
217 | /** |
218 | * bio_blkcg_css - return the blkcg CSS associated with a bio | |
219 | * @bio: target bio | |
220 | * | |
221 | * This returns the CSS for the blkcg associated with a bio, or %NULL if not | |
222 | * associated. Callers are expected to either handle %NULL or know association | |
223 | * has been done prior to calling this. | |
224 | */ | |
225 | struct cgroup_subsys_state *bio_blkcg_css(struct bio *bio) | |
226 | { | |
227 | if (!bio || !bio->bi_blkg) | |
228 | return NULL; | |
229 | return &bio->bi_blkg->blkcg->css; | |
230 | } | |
231 | EXPORT_SYMBOL_GPL(bio_blkcg_css); | |
232 | ||
397c9f46 CH |
233 | /** |
234 | * blkcg_parent - get the parent of a blkcg | |
235 | * @blkcg: blkcg of interest | |
236 | * | |
237 | * Return the parent blkcg of @blkcg. Can be called anytime. | |
238 | */ | |
239 | static inline struct blkcg *blkcg_parent(struct blkcg *blkcg) | |
240 | { | |
241 | return css_to_blkcg(blkcg->css.parent); | |
242 | } | |
243 | ||
0381411e TH |
244 | /** |
245 | * blkg_alloc - allocate a blkg | |
246 | * @blkcg: block cgroup the new blkg is associated with | |
99e60387 | 247 | * @disk: gendisk the new blkg is associated with |
15974993 | 248 | * @gfp_mask: allocation mask to use |
0381411e | 249 | * |
e8989fae | 250 | * Allocate a new blkg assocating @blkcg and @q. |
0381411e | 251 | */ |
99e60387 | 252 | static struct blkcg_gq *blkg_alloc(struct blkcg *blkcg, struct gendisk *disk, |
15974993 | 253 | gfp_t gfp_mask) |
0381411e | 254 | { |
3c798398 | 255 | struct blkcg_gq *blkg; |
f7331648 | 256 | int i, cpu; |
0381411e TH |
257 | |
258 | /* alloc and init base part */ | |
99e60387 | 259 | blkg = kzalloc_node(sizeof(*blkg), gfp_mask, disk->queue->node); |
0381411e TH |
260 | if (!blkg) |
261 | return NULL; | |
ef069b97 | 262 | if (percpu_ref_init(&blkg->refcnt, blkg_release, 0, gfp_mask)) |
0b6f93bd | 263 | goto out_free_blkg; |
f7331648 TH |
264 | blkg->iostat_cpu = alloc_percpu_gfp(struct blkg_iostat_set, gfp_mask); |
265 | if (!blkg->iostat_cpu) | |
0b6f93bd | 266 | goto out_exit_refcnt; |
99e60387 | 267 | if (!blk_get_queue(disk->queue)) |
0b6f93bd | 268 | goto out_free_iostat; |
0a9a25ca | 269 | |
99e60387 | 270 | blkg->q = disk->queue; |
e8989fae | 271 | INIT_LIST_HEAD(&blkg->q_node); |
d3f77dfd TH |
272 | spin_lock_init(&blkg->async_bio_lock); |
273 | bio_list_init(&blkg->async_bios); | |
274 | INIT_WORK(&blkg->async_bio_work, blkg_async_bio_workfn); | |
0381411e | 275 | blkg->blkcg = blkcg; |
0381411e | 276 | |
f7331648 | 277 | u64_stats_init(&blkg->iostat.sync); |
3b8cc629 | 278 | for_each_possible_cpu(cpu) { |
f7331648 | 279 | u64_stats_init(&per_cpu_ptr(blkg->iostat_cpu, cpu)->sync); |
3b8cc629 WL |
280 | per_cpu_ptr(blkg->iostat_cpu, cpu)->blkg = blkg; |
281 | } | |
f7331648 | 282 | |
8bd435b3 | 283 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
3c798398 | 284 | struct blkcg_policy *pol = blkcg_policy[i]; |
e8989fae | 285 | struct blkg_policy_data *pd; |
0381411e | 286 | |
99e60387 | 287 | if (!blkcg_policy_enabled(disk->queue, pol)) |
e8989fae TH |
288 | continue; |
289 | ||
290 | /* alloc per-policy data and attach it to blkg */ | |
0a0b4f79 | 291 | pd = pol->pd_alloc_fn(disk, blkcg, gfp_mask); |
a051661c | 292 | if (!pd) |
0b6f93bd | 293 | goto out_free_pds; |
e8989fae TH |
294 | blkg->pd[i] = pd; |
295 | pd->blkg = blkg; | |
b276a876 | 296 | pd->plid = i; |
dfd6200a | 297 | pd->online = false; |
e8989fae TH |
298 | } |
299 | ||
0381411e | 300 | return blkg; |
a051661c | 301 | |
0b6f93bd CH |
302 | out_free_pds: |
303 | while (--i >= 0) | |
304 | if (blkg->pd[i]) | |
305 | blkcg_policy[i]->pd_free_fn(blkg->pd[i]); | |
a06377c5 | 306 | blk_put_queue(disk->queue); |
0b6f93bd CH |
307 | out_free_iostat: |
308 | free_percpu(blkg->iostat_cpu); | |
309 | out_exit_refcnt: | |
310 | percpu_ref_exit(&blkg->refcnt); | |
311 | out_free_blkg: | |
312 | kfree(blkg); | |
a051661c | 313 | return NULL; |
0381411e TH |
314 | } |
315 | ||
15974993 | 316 | /* |
d708f0d5 JA |
317 | * If @new_blkg is %NULL, this function tries to allocate a new one as |
318 | * necessary using %GFP_NOWAIT. @new_blkg is always consumed on return. | |
15974993 | 319 | */ |
99e60387 | 320 | static struct blkcg_gq *blkg_create(struct blkcg *blkcg, struct gendisk *disk, |
d708f0d5 | 321 | struct blkcg_gq *new_blkg) |
5624a4e4 | 322 | { |
d708f0d5 | 323 | struct blkcg_gq *blkg; |
f427d909 | 324 | int i, ret; |
5624a4e4 | 325 | |
99e60387 | 326 | lockdep_assert_held(&disk->queue->queue_lock); |
cd1604fa | 327 | |
0273ac34 | 328 | /* request_queue is dying, do not create/recreate a blkg */ |
99e60387 | 329 | if (blk_queue_dying(disk->queue)) { |
0273ac34 DZ |
330 | ret = -ENODEV; |
331 | goto err_free_blkg; | |
332 | } | |
333 | ||
7ee9c562 | 334 | /* blkg holds a reference to blkcg */ |
ec903c0c | 335 | if (!css_tryget_online(&blkcg->css)) { |
20386ce0 | 336 | ret = -ENODEV; |
93e6d5d8 | 337 | goto err_free_blkg; |
15974993 | 338 | } |
cd1604fa | 339 | |
d708f0d5 JA |
340 | /* allocate */ |
341 | if (!new_blkg) { | |
99e60387 | 342 | new_blkg = blkg_alloc(blkcg, disk, GFP_NOWAIT | __GFP_NOWARN); |
d708f0d5 JA |
343 | if (unlikely(!new_blkg)) { |
344 | ret = -ENOMEM; | |
8c911f3d | 345 | goto err_put_css; |
15974993 TH |
346 | } |
347 | } | |
d708f0d5 | 348 | blkg = new_blkg; |
cd1604fa | 349 | |
db613670 | 350 | /* link parent */ |
3c547865 | 351 | if (blkcg_parent(blkcg)) { |
99e60387 | 352 | blkg->parent = blkg_lookup(blkcg_parent(blkcg), disk->queue); |
3c547865 | 353 | if (WARN_ON_ONCE(!blkg->parent)) { |
20386ce0 | 354 | ret = -ENODEV; |
8c911f3d | 355 | goto err_put_css; |
3c547865 TH |
356 | } |
357 | blkg_get(blkg->parent); | |
358 | } | |
359 | ||
db613670 TH |
360 | /* invoke per-policy init */ |
361 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
362 | struct blkcg_policy *pol = blkcg_policy[i]; | |
363 | ||
364 | if (blkg->pd[i] && pol->pd_init_fn) | |
a9520cd6 | 365 | pol->pd_init_fn(blkg->pd[i]); |
db613670 TH |
366 | } |
367 | ||
368 | /* insert */ | |
cd1604fa | 369 | spin_lock(&blkcg->lock); |
99e60387 | 370 | ret = radix_tree_insert(&blkcg->blkg_tree, disk->queue->id, blkg); |
a637120e TH |
371 | if (likely(!ret)) { |
372 | hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list); | |
99e60387 | 373 | list_add(&blkg->q_node, &disk->queue->blkg_list); |
f427d909 TH |
374 | |
375 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
376 | struct blkcg_policy *pol = blkcg_policy[i]; | |
377 | ||
dfd6200a YK |
378 | if (blkg->pd[i]) { |
379 | if (pol->pd_online_fn) | |
380 | pol->pd_online_fn(blkg->pd[i]); | |
381 | blkg->pd[i]->online = true; | |
382 | } | |
f427d909 | 383 | } |
a637120e | 384 | } |
f427d909 | 385 | blkg->online = true; |
cd1604fa | 386 | spin_unlock(&blkcg->lock); |
496fb780 | 387 | |
ec13b1d6 | 388 | if (!ret) |
a637120e | 389 | return blkg; |
15974993 | 390 | |
3c547865 TH |
391 | /* @blkg failed fully initialized, use the usual release path */ |
392 | blkg_put(blkg); | |
393 | return ERR_PTR(ret); | |
394 | ||
d708f0d5 | 395 | err_put_css: |
496fb780 | 396 | css_put(&blkcg->css); |
93e6d5d8 | 397 | err_free_blkg: |
28e538a3 CH |
398 | if (new_blkg) |
399 | blkg_free(new_blkg); | |
93e6d5d8 | 400 | return ERR_PTR(ret); |
31e4c28d | 401 | } |
3c96cb32 | 402 | |
86cde6b6 | 403 | /** |
8c546287 | 404 | * blkg_lookup_create - lookup blkg, try to create one if not there |
86cde6b6 | 405 | * @blkcg: blkcg of interest |
99e60387 | 406 | * @disk: gendisk of interest |
86cde6b6 | 407 | * |
99e60387 | 408 | * Lookup blkg for the @blkcg - @disk pair. If it doesn't exist, try to |
3c547865 TH |
409 | * create one. blkg creation is performed recursively from blkcg_root such |
410 | * that all non-root blkg's have access to the parent blkg. This function | |
99e60387 | 411 | * should be called under RCU read lock and takes @disk->queue->queue_lock. |
86cde6b6 | 412 | * |
beea9da0 DZ |
413 | * Returns the blkg or the closest blkg if blkg_create() fails as it walks |
414 | * down from root. | |
86cde6b6 | 415 | */ |
8c546287 | 416 | static struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg, |
99e60387 | 417 | struct gendisk *disk) |
3c96cb32 | 418 | { |
99e60387 | 419 | struct request_queue *q = disk->queue; |
86cde6b6 | 420 | struct blkcg_gq *blkg; |
8c546287 | 421 | unsigned long flags; |
86cde6b6 TH |
422 | |
423 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
86cde6b6 | 424 | |
8c546287 | 425 | blkg = blkg_lookup(blkcg, q); |
86cde6b6 TH |
426 | if (blkg) |
427 | return blkg; | |
428 | ||
8c546287 | 429 | spin_lock_irqsave(&q->queue_lock, flags); |
4a69f325 CH |
430 | blkg = blkg_lookup(blkcg, q); |
431 | if (blkg) { | |
5765033c CH |
432 | if (blkcg != &blkcg_root && |
433 | blkg != rcu_dereference(blkcg->blkg_hint)) | |
434 | rcu_assign_pointer(blkcg->blkg_hint, blkg); | |
8c546287 | 435 | goto found; |
4a69f325 | 436 | } |
8c546287 | 437 | |
3c547865 TH |
438 | /* |
439 | * Create blkgs walking down from blkcg_root to @blkcg, so that all | |
beea9da0 DZ |
440 | * non-root blkgs have access to their parents. Returns the closest |
441 | * blkg to the intended blkg should blkg_create() fail. | |
3c547865 TH |
442 | */ |
443 | while (true) { | |
444 | struct blkcg *pos = blkcg; | |
445 | struct blkcg *parent = blkcg_parent(blkcg); | |
beea9da0 DZ |
446 | struct blkcg_gq *ret_blkg = q->root_blkg; |
447 | ||
448 | while (parent) { | |
79fcc5be | 449 | blkg = blkg_lookup(parent, q); |
beea9da0 DZ |
450 | if (blkg) { |
451 | /* remember closest blkg */ | |
452 | ret_blkg = blkg; | |
453 | break; | |
454 | } | |
3c547865 TH |
455 | pos = parent; |
456 | parent = blkcg_parent(parent); | |
457 | } | |
458 | ||
99e60387 | 459 | blkg = blkg_create(pos, disk, NULL); |
8c546287 CH |
460 | if (IS_ERR(blkg)) { |
461 | blkg = ret_blkg; | |
462 | break; | |
463 | } | |
beea9da0 | 464 | if (pos == blkcg) |
8c546287 | 465 | break; |
b978962a DZ |
466 | } |
467 | ||
8c546287 CH |
468 | found: |
469 | spin_unlock_irqrestore(&q->queue_lock, flags); | |
b978962a DZ |
470 | return blkg; |
471 | } | |
472 | ||
3c798398 | 473 | static void blkg_destroy(struct blkcg_gq *blkg) |
03aa264a | 474 | { |
3c798398 | 475 | struct blkcg *blkcg = blkg->blkcg; |
6b065462 | 476 | int i; |
03aa264a | 477 | |
0d945c1f | 478 | lockdep_assert_held(&blkg->q->queue_lock); |
9f13ef67 | 479 | lockdep_assert_held(&blkcg->lock); |
03aa264a | 480 | |
f1c006f1 YK |
481 | /* |
482 | * blkg stays on the queue list until blkg_free_workfn(), see details in | |
483 | * blkg_free_workfn(), hence this function can be called from | |
484 | * blkcg_destroy_blkgs() first and again from blkg_destroy_all() before | |
485 | * blkg_free_workfn(). | |
486 | */ | |
487 | if (hlist_unhashed(&blkg->blkcg_node)) | |
488 | return; | |
a637120e | 489 | |
6b065462 DZF |
490 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
491 | struct blkcg_policy *pol = blkcg_policy[i]; | |
492 | ||
dfd6200a | 493 | if (blkg->pd[i] && blkg->pd[i]->online) { |
f37bf75c | 494 | blkg->pd[i]->online = false; |
dfd6200a YK |
495 | if (pol->pd_offline_fn) |
496 | pol->pd_offline_fn(blkg->pd[i]); | |
dfd6200a | 497 | } |
6b065462 DZF |
498 | } |
499 | ||
f427d909 TH |
500 | blkg->online = false; |
501 | ||
a637120e | 502 | radix_tree_delete(&blkcg->blkg_tree, blkg->q->id); |
9f13ef67 | 503 | hlist_del_init_rcu(&blkg->blkcg_node); |
03aa264a | 504 | |
a637120e TH |
505 | /* |
506 | * Both setting lookup hint to and clearing it from @blkg are done | |
507 | * under queue_lock. If it's not pointing to @blkg now, it never | |
508 | * will. Hint assignment itself can race safely. | |
509 | */ | |
ec6c676a | 510 | if (rcu_access_pointer(blkcg->blkg_hint) == blkg) |
a637120e TH |
511 | rcu_assign_pointer(blkcg->blkg_hint, NULL); |
512 | ||
03aa264a TH |
513 | /* |
514 | * Put the reference taken at the time of creation so that when all | |
515 | * queues are gone, group can be destroyed. | |
516 | */ | |
7fcf2b03 | 517 | percpu_ref_kill(&blkg->refcnt); |
03aa264a TH |
518 | } |
519 | ||
00ad6991 | 520 | static void blkg_destroy_all(struct gendisk *disk) |
72e06c25 | 521 | { |
00ad6991 | 522 | struct request_queue *q = disk->queue; |
3c798398 | 523 | struct blkcg_gq *blkg, *n; |
a731763f | 524 | int count = BLKG_DESTROY_BATCH_SIZE; |
72e06c25 | 525 | |
a731763f | 526 | restart: |
0d945c1f | 527 | spin_lock_irq(&q->queue_lock); |
9f13ef67 | 528 | list_for_each_entry_safe(blkg, n, &q->blkg_list, q_node) { |
3c798398 | 529 | struct blkcg *blkcg = blkg->blkcg; |
72e06c25 | 530 | |
9f13ef67 TH |
531 | spin_lock(&blkcg->lock); |
532 | blkg_destroy(blkg); | |
533 | spin_unlock(&blkcg->lock); | |
a731763f YK |
534 | |
535 | /* | |
536 | * in order to avoid holding the spin lock for too long, release | |
537 | * it when a batch of blkgs are destroyed. | |
538 | */ | |
539 | if (!(--count)) { | |
540 | count = BLKG_DESTROY_BATCH_SIZE; | |
541 | spin_unlock_irq(&q->queue_lock); | |
542 | cond_resched(); | |
543 | goto restart; | |
544 | } | |
72e06c25 | 545 | } |
6fe810bd TH |
546 | |
547 | q->root_blkg = NULL; | |
0d945c1f | 548 | spin_unlock_irq(&q->queue_lock); |
72e06c25 TH |
549 | } |
550 | ||
182446d0 TH |
551 | static int blkcg_reset_stats(struct cgroup_subsys_state *css, |
552 | struct cftype *cftype, u64 val) | |
303a3acb | 553 | { |
182446d0 | 554 | struct blkcg *blkcg = css_to_blkcg(css); |
3c798398 | 555 | struct blkcg_gq *blkg; |
f7331648 | 556 | int i, cpu; |
303a3acb | 557 | |
838f13bf | 558 | mutex_lock(&blkcg_pol_mutex); |
303a3acb | 559 | spin_lock_irq(&blkcg->lock); |
997a026c TH |
560 | |
561 | /* | |
562 | * Note that stat reset is racy - it doesn't synchronize against | |
563 | * stat updates. This is a debug feature which shouldn't exist | |
564 | * anyway. If you get hit by a race, retry. | |
565 | */ | |
b67bfe0d | 566 | hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) { |
f7331648 TH |
567 | for_each_possible_cpu(cpu) { |
568 | struct blkg_iostat_set *bis = | |
569 | per_cpu_ptr(blkg->iostat_cpu, cpu); | |
570 | memset(bis, 0, sizeof(*bis)); | |
571 | } | |
572 | memset(&blkg->iostat, 0, sizeof(blkg->iostat)); | |
77ea7338 | 573 | |
8bd435b3 | 574 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
3c798398 | 575 | struct blkcg_policy *pol = blkcg_policy[i]; |
549d3aa8 | 576 | |
a9520cd6 TH |
577 | if (blkg->pd[i] && pol->pd_reset_stats_fn) |
578 | pol->pd_reset_stats_fn(blkg->pd[i]); | |
bc0d6501 | 579 | } |
303a3acb | 580 | } |
f0bdc8cd | 581 | |
303a3acb | 582 | spin_unlock_irq(&blkcg->lock); |
bc0d6501 | 583 | mutex_unlock(&blkcg_pol_mutex); |
303a3acb DS |
584 | return 0; |
585 | } | |
586 | ||
dd165eb3 | 587 | const char *blkg_dev_name(struct blkcg_gq *blkg) |
303a3acb | 588 | { |
a06377c5 | 589 | if (!blkg->q->disk) |
edb0872f | 590 | return NULL; |
d152c682 | 591 | return bdi_dev_name(blkg->q->disk->bdi); |
303a3acb DS |
592 | } |
593 | ||
d3d32e69 TH |
594 | /** |
595 | * blkcg_print_blkgs - helper for printing per-blkg data | |
596 | * @sf: seq_file to print to | |
597 | * @blkcg: blkcg of interest | |
598 | * @prfill: fill function to print out a blkg | |
599 | * @pol: policy in question | |
600 | * @data: data to be passed to @prfill | |
601 | * @show_total: to print out sum of prfill return values or not | |
602 | * | |
603 | * This function invokes @prfill on each blkg of @blkcg if pd for the | |
604 | * policy specified by @pol exists. @prfill is invoked with @sf, the | |
810ecfa7 TH |
605 | * policy data and @data and the matching queue lock held. If @show_total |
606 | * is %true, the sum of the return values from @prfill is printed with | |
607 | * "Total" label at the end. | |
d3d32e69 TH |
608 | * |
609 | * This is to be used to construct print functions for | |
610 | * cftype->read_seq_string method. | |
611 | */ | |
3c798398 | 612 | void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg, |
f95a04af TH |
613 | u64 (*prfill)(struct seq_file *, |
614 | struct blkg_policy_data *, int), | |
3c798398 | 615 | const struct blkcg_policy *pol, int data, |
ec399347 | 616 | bool show_total) |
5624a4e4 | 617 | { |
3c798398 | 618 | struct blkcg_gq *blkg; |
d3d32e69 | 619 | u64 total = 0; |
5624a4e4 | 620 | |
810ecfa7 | 621 | rcu_read_lock(); |
ee89f812 | 622 | hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) { |
0d945c1f | 623 | spin_lock_irq(&blkg->q->queue_lock); |
a2b1693b | 624 | if (blkcg_policy_enabled(blkg->q, pol)) |
f95a04af | 625 | total += prfill(sf, blkg->pd[pol->plid], data); |
0d945c1f | 626 | spin_unlock_irq(&blkg->q->queue_lock); |
810ecfa7 TH |
627 | } |
628 | rcu_read_unlock(); | |
d3d32e69 TH |
629 | |
630 | if (show_total) | |
631 | seq_printf(sf, "Total %llu\n", (unsigned long long)total); | |
632 | } | |
829fdb50 | 633 | EXPORT_SYMBOL_GPL(blkcg_print_blkgs); |
d3d32e69 TH |
634 | |
635 | /** | |
636 | * __blkg_prfill_u64 - prfill helper for a single u64 value | |
637 | * @sf: seq_file to print to | |
f95a04af | 638 | * @pd: policy private data of interest |
d3d32e69 TH |
639 | * @v: value to print |
640 | * | |
37754595 | 641 | * Print @v to @sf for the device associated with @pd. |
d3d32e69 | 642 | */ |
f95a04af | 643 | u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v) |
d3d32e69 | 644 | { |
f95a04af | 645 | const char *dname = blkg_dev_name(pd->blkg); |
d3d32e69 TH |
646 | |
647 | if (!dname) | |
648 | return 0; | |
649 | ||
650 | seq_printf(sf, "%s %llu\n", dname, (unsigned long long)v); | |
651 | return v; | |
652 | } | |
829fdb50 | 653 | EXPORT_SYMBOL_GPL(__blkg_prfill_u64); |
d3d32e69 | 654 | |
015d254c | 655 | /** |
22ae8ce8 | 656 | * blkcg_conf_open_bdev - parse and open bdev for per-blkg config update |
015d254c TH |
657 | * @inputp: input string pointer |
658 | * | |
659 | * Parse the device node prefix part, MAJ:MIN, of per-blkg config update | |
22ae8ce8 | 660 | * from @input and get and return the matching bdev. *@inputp is |
015d254c TH |
661 | * updated to point past the device node prefix. Returns an ERR_PTR() |
662 | * value on error. | |
663 | * | |
664 | * Use this function iff blkg_conf_prep() can't be used for some reason. | |
665 | */ | |
22ae8ce8 | 666 | struct block_device *blkcg_conf_open_bdev(char **inputp) |
015d254c TH |
667 | { |
668 | char *input = *inputp; | |
669 | unsigned int major, minor; | |
22ae8ce8 CH |
670 | struct block_device *bdev; |
671 | int key_len; | |
015d254c TH |
672 | |
673 | if (sscanf(input, "%u:%u%n", &major, &minor, &key_len) != 2) | |
674 | return ERR_PTR(-EINVAL); | |
675 | ||
676 | input += key_len; | |
677 | if (!isspace(*input)) | |
678 | return ERR_PTR(-EINVAL); | |
679 | input = skip_spaces(input); | |
680 | ||
22ae8ce8 CH |
681 | bdev = blkdev_get_no_open(MKDEV(major, minor)); |
682 | if (!bdev) | |
015d254c | 683 | return ERR_PTR(-ENODEV); |
22ae8ce8 CH |
684 | if (bdev_is_partition(bdev)) { |
685 | blkdev_put_no_open(bdev); | |
015d254c TH |
686 | return ERR_PTR(-ENODEV); |
687 | } | |
688 | ||
689 | *inputp = input; | |
22ae8ce8 | 690 | return bdev; |
015d254c TH |
691 | } |
692 | ||
3a8b31d3 TH |
693 | /** |
694 | * blkg_conf_prep - parse and prepare for per-blkg config update | |
695 | * @blkcg: target block cgroup | |
da8b0662 | 696 | * @pol: target policy |
3a8b31d3 TH |
697 | * @input: input string |
698 | * @ctx: blkg_conf_ctx to be filled | |
699 | * | |
700 | * Parse per-blkg config update from @input and initialize @ctx with the | |
36aa9e5f | 701 | * result. @ctx->blkg points to the blkg to be updated and @ctx->body the |
83462a6c TH |
702 | * part of @input following MAJ:MIN. This function returns with queue lock |
703 | * held and must be paired with blkg_conf_finish(). | |
3a8b31d3 | 704 | */ |
3c798398 | 705 | int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol, |
36aa9e5f | 706 | char *input, struct blkg_conf_ctx *ctx) |
83462a6c | 707 | __acquires(&bdev->bd_queue->queue_lock) |
34d0f179 | 708 | { |
22ae8ce8 | 709 | struct block_device *bdev; |
99e60387 | 710 | struct gendisk *disk; |
457e490f | 711 | struct request_queue *q; |
3c798398 | 712 | struct blkcg_gq *blkg; |
015d254c | 713 | int ret; |
36aa9e5f | 714 | |
22ae8ce8 CH |
715 | bdev = blkcg_conf_open_bdev(&input); |
716 | if (IS_ERR(bdev)) | |
717 | return PTR_ERR(bdev); | |
99e60387 CH |
718 | disk = bdev->bd_disk; |
719 | q = disk->queue; | |
da8b0662 | 720 | |
0c9d338c YK |
721 | /* |
722 | * blkcg_deactivate_policy() requires queue to be frozen, we can grab | |
723 | * q_usage_counter to prevent concurrent with blkcg_deactivate_policy(). | |
724 | */ | |
725 | ret = blk_queue_enter(q, 0); | |
726 | if (ret) | |
15c30104 | 727 | goto fail; |
0c9d338c | 728 | |
0d945c1f | 729 | spin_lock_irq(&q->queue_lock); |
e56da7e2 | 730 | |
f753526e CH |
731 | if (!blkcg_policy_enabled(q, pol)) { |
732 | ret = -EOPNOTSUPP; | |
457e490f TE |
733 | goto fail_unlock; |
734 | } | |
735 | ||
f753526e | 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); | |
79fcc5be | 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); |
457e490f | 757 | |
99e60387 | 758 | new_blkg = blkg_alloc(pos, disk, GFP_KERNEL); |
457e490f TE |
759 | if (unlikely(!new_blkg)) { |
760 | ret = -ENOMEM; | |
15c30104 | 761 | goto fail_exit_queue; |
7702e8f4 | 762 | } |
3a8b31d3 | 763 | |
f255c19b GKB |
764 | if (radix_tree_preload(GFP_KERNEL)) { |
765 | blkg_free(new_blkg); | |
766 | ret = -ENOMEM; | |
15c30104 | 767 | goto fail_exit_queue; |
f255c19b GKB |
768 | } |
769 | ||
0d945c1f | 770 | spin_lock_irq(&q->queue_lock); |
457e490f | 771 | |
f753526e | 772 | if (!blkcg_policy_enabled(q, pol)) { |
52abfcbd | 773 | blkg_free(new_blkg); |
f753526e | 774 | ret = -EOPNOTSUPP; |
f255c19b | 775 | goto fail_preloaded; |
457e490f TE |
776 | } |
777 | ||
f753526e | 778 | blkg = blkg_lookup(pos, q); |
457e490f TE |
779 | if (blkg) { |
780 | blkg_free(new_blkg); | |
781 | } else { | |
99e60387 | 782 | blkg = blkg_create(pos, disk, new_blkg); |
98d669b4 | 783 | if (IS_ERR(blkg)) { |
457e490f | 784 | ret = PTR_ERR(blkg); |
f255c19b | 785 | goto fail_preloaded; |
457e490f TE |
786 | } |
787 | } | |
788 | ||
f255c19b GKB |
789 | radix_tree_preload_end(); |
790 | ||
457e490f TE |
791 | if (pos == blkcg) |
792 | goto success; | |
793 | } | |
794 | success: | |
0c9d338c | 795 | blk_queue_exit(q); |
22ae8ce8 | 796 | ctx->bdev = bdev; |
3a8b31d3 | 797 | ctx->blkg = blkg; |
015d254c | 798 | ctx->body = input; |
726fa694 | 799 | return 0; |
457e490f | 800 | |
f255c19b GKB |
801 | fail_preloaded: |
802 | radix_tree_preload_end(); | |
457e490f | 803 | fail_unlock: |
0d945c1f | 804 | spin_unlock_irq(&q->queue_lock); |
15c30104 YK |
805 | fail_exit_queue: |
806 | blk_queue_exit(q); | |
457e490f | 807 | fail: |
22ae8ce8 | 808 | blkdev_put_no_open(bdev); |
457e490f TE |
809 | /* |
810 | * If queue was bypassing, we should retry. Do so after a | |
811 | * short msleep(). It isn't strictly necessary but queue | |
812 | * can be bypassing for some time and it's always nice to | |
813 | * avoid busy looping. | |
814 | */ | |
815 | if (ret == -EBUSY) { | |
816 | msleep(10); | |
817 | ret = restart_syscall(); | |
818 | } | |
819 | return ret; | |
34d0f179 | 820 | } |
89f3b6d6 | 821 | EXPORT_SYMBOL_GPL(blkg_conf_prep); |
34d0f179 | 822 | |
3a8b31d3 TH |
823 | /** |
824 | * blkg_conf_finish - finish up per-blkg config update | |
37754595 | 825 | * @ctx: blkg_conf_ctx initialized by blkg_conf_prep() |
3a8b31d3 TH |
826 | * |
827 | * Finish up after per-blkg config update. This function must be paired | |
828 | * with blkg_conf_prep(). | |
829 | */ | |
829fdb50 | 830 | void blkg_conf_finish(struct blkg_conf_ctx *ctx) |
83462a6c | 831 | __releases(&ctx->bdev->bd_queue->queue_lock) |
34d0f179 | 832 | { |
ed6cddef | 833 | spin_unlock_irq(&bdev_get_queue(ctx->bdev)->queue_lock); |
22ae8ce8 | 834 | blkdev_put_no_open(ctx->bdev); |
34d0f179 | 835 | } |
89f3b6d6 | 836 | EXPORT_SYMBOL_GPL(blkg_conf_finish); |
34d0f179 | 837 | |
cd1fc4b9 BB |
838 | static void blkg_iostat_set(struct blkg_iostat *dst, struct blkg_iostat *src) |
839 | { | |
840 | int i; | |
841 | ||
842 | for (i = 0; i < BLKG_IOSTAT_NR; i++) { | |
843 | dst->bytes[i] = src->bytes[i]; | |
844 | dst->ios[i] = src->ios[i]; | |
845 | } | |
846 | } | |
847 | ||
848 | static void blkg_iostat_add(struct blkg_iostat *dst, struct blkg_iostat *src) | |
849 | { | |
850 | int i; | |
851 | ||
852 | for (i = 0; i < BLKG_IOSTAT_NR; i++) { | |
853 | dst->bytes[i] += src->bytes[i]; | |
854 | dst->ios[i] += src->ios[i]; | |
855 | } | |
856 | } | |
857 | ||
858 | static void blkg_iostat_sub(struct blkg_iostat *dst, struct blkg_iostat *src) | |
859 | { | |
860 | int i; | |
861 | ||
862 | for (i = 0; i < BLKG_IOSTAT_NR; i++) { | |
863 | dst->bytes[i] -= src->bytes[i]; | |
864 | dst->ios[i] -= src->ios[i]; | |
865 | } | |
866 | } | |
867 | ||
362b8c16 JY |
868 | static void blkcg_iostat_update(struct blkcg_gq *blkg, struct blkg_iostat *cur, |
869 | struct blkg_iostat *last) | |
870 | { | |
871 | struct blkg_iostat delta; | |
872 | unsigned long flags; | |
873 | ||
874 | /* propagate percpu delta to global */ | |
875 | flags = u64_stats_update_begin_irqsave(&blkg->iostat.sync); | |
876 | blkg_iostat_set(&delta, cur); | |
877 | blkg_iostat_sub(&delta, last); | |
878 | blkg_iostat_add(&blkg->iostat.cur, &delta); | |
879 | blkg_iostat_add(last, &delta); | |
880 | u64_stats_update_end_irqrestore(&blkg->iostat.sync, flags); | |
881 | } | |
882 | ||
cd1fc4b9 BB |
883 | static void blkcg_rstat_flush(struct cgroup_subsys_state *css, int cpu) |
884 | { | |
885 | struct blkcg *blkcg = css_to_blkcg(css); | |
3b8cc629 WL |
886 | struct llist_head *lhead = per_cpu_ptr(blkcg->lhead, cpu); |
887 | struct llist_node *lnode; | |
888 | struct blkg_iostat_set *bisc, *next_bisc; | |
cd1fc4b9 | 889 | |
dc26532a JW |
890 | /* Root-level stats are sourced from system-wide IO stats */ |
891 | if (!cgroup_parent(css->cgroup)) | |
892 | return; | |
893 | ||
cd1fc4b9 BB |
894 | rcu_read_lock(); |
895 | ||
3b8cc629 WL |
896 | lnode = llist_del_all(lhead); |
897 | if (!lnode) | |
898 | goto out; | |
899 | ||
900 | /* | |
901 | * Iterate only the iostat_cpu's queued in the lockless list. | |
902 | */ | |
903 | llist_for_each_entry_safe(bisc, next_bisc, lnode, lnode) { | |
904 | struct blkcg_gq *blkg = bisc->blkg; | |
cd1fc4b9 | 905 | struct blkcg_gq *parent = blkg->parent; |
362b8c16 | 906 | struct blkg_iostat cur; |
cd1fc4b9 BB |
907 | unsigned int seq; |
908 | ||
3b8cc629 WL |
909 | WRITE_ONCE(bisc->lqueued, false); |
910 | ||
cd1fc4b9 BB |
911 | /* fetch the current per-cpu values */ |
912 | do { | |
913 | seq = u64_stats_fetch_begin(&bisc->sync); | |
914 | blkg_iostat_set(&cur, &bisc->cur); | |
915 | } while (u64_stats_fetch_retry(&bisc->sync, seq)); | |
916 | ||
362b8c16 | 917 | blkcg_iostat_update(blkg, &cur, &bisc->last); |
cd1fc4b9 | 918 | |
dc26532a | 919 | /* propagate global delta to parent (unless that's root) */ |
362b8c16 JY |
920 | if (parent && parent->parent) |
921 | blkcg_iostat_update(parent, &blkg->iostat.cur, | |
922 | &blkg->iostat.last); | |
3b8cc629 | 923 | percpu_ref_put(&blkg->refcnt); |
cd1fc4b9 BB |
924 | } |
925 | ||
3b8cc629 | 926 | out: |
cd1fc4b9 BB |
927 | rcu_read_unlock(); |
928 | } | |
929 | ||
ef45fe47 | 930 | /* |
dc26532a JW |
931 | * We source root cgroup stats from the system-wide stats to avoid |
932 | * tracking the same information twice and incurring overhead when no | |
933 | * cgroups are defined. For that reason, cgroup_rstat_flush in | |
934 | * blkcg_print_stat does not actually fill out the iostat in the root | |
935 | * cgroup's blkcg_gq. | |
ef45fe47 BB |
936 | * |
937 | * However, we would like to re-use the printing code between the root and | |
938 | * non-root cgroups to the extent possible. For that reason, we simulate | |
939 | * flushing the root cgroup's stats by explicitly filling in the iostat | |
940 | * with disk level statistics. | |
941 | */ | |
942 | static void blkcg_fill_root_iostats(void) | |
943 | { | |
944 | struct class_dev_iter iter; | |
945 | struct device *dev; | |
946 | ||
947 | class_dev_iter_init(&iter, &block_class, NULL, &disk_type); | |
948 | while ((dev = class_dev_iter_next(&iter))) { | |
0d02129e | 949 | struct block_device *bdev = dev_to_bdev(dev); |
928f6f00 | 950 | struct blkcg_gq *blkg = bdev->bd_disk->queue->root_blkg; |
ef45fe47 BB |
951 | struct blkg_iostat tmp; |
952 | int cpu; | |
f122d103 | 953 | unsigned long flags; |
ef45fe47 BB |
954 | |
955 | memset(&tmp, 0, sizeof(tmp)); | |
956 | for_each_possible_cpu(cpu) { | |
957 | struct disk_stats *cpu_dkstats; | |
958 | ||
0d02129e | 959 | cpu_dkstats = per_cpu_ptr(bdev->bd_stats, cpu); |
ef45fe47 BB |
960 | tmp.ios[BLKG_IOSTAT_READ] += |
961 | cpu_dkstats->ios[STAT_READ]; | |
962 | tmp.ios[BLKG_IOSTAT_WRITE] += | |
963 | cpu_dkstats->ios[STAT_WRITE]; | |
964 | tmp.ios[BLKG_IOSTAT_DISCARD] += | |
965 | cpu_dkstats->ios[STAT_DISCARD]; | |
966 | // convert sectors to bytes | |
967 | tmp.bytes[BLKG_IOSTAT_READ] += | |
968 | cpu_dkstats->sectors[STAT_READ] << 9; | |
969 | tmp.bytes[BLKG_IOSTAT_WRITE] += | |
970 | cpu_dkstats->sectors[STAT_WRITE] << 9; | |
971 | tmp.bytes[BLKG_IOSTAT_DISCARD] += | |
972 | cpu_dkstats->sectors[STAT_DISCARD] << 9; | |
ef45fe47 | 973 | } |
f122d103 CZ |
974 | |
975 | flags = u64_stats_update_begin_irqsave(&blkg->iostat.sync); | |
976 | blkg_iostat_set(&blkg->iostat.cur, &tmp); | |
977 | u64_stats_update_end_irqrestore(&blkg->iostat.sync, flags); | |
ef45fe47 BB |
978 | } |
979 | } | |
980 | ||
49cb5168 | 981 | static void blkcg_print_one_stat(struct blkcg_gq *blkg, struct seq_file *s) |
2ee867dc | 982 | { |
49cb5168 CH |
983 | struct blkg_iostat_set *bis = &blkg->iostat; |
984 | u64 rbytes, wbytes, rios, wios, dbytes, dios; | |
49cb5168 CH |
985 | const char *dname; |
986 | unsigned seq; | |
49cb5168 | 987 | int i; |
ef45fe47 | 988 | |
49cb5168 CH |
989 | if (!blkg->online) |
990 | return; | |
2ee867dc | 991 | |
49cb5168 CH |
992 | dname = blkg_dev_name(blkg); |
993 | if (!dname) | |
994 | return; | |
2ee867dc | 995 | |
252c651a | 996 | seq_printf(s, "%s ", dname); |
b0814361 | 997 | |
49cb5168 CH |
998 | do { |
999 | seq = u64_stats_fetch_begin(&bis->sync); | |
b0814361 | 1000 | |
49cb5168 CH |
1001 | rbytes = bis->cur.bytes[BLKG_IOSTAT_READ]; |
1002 | wbytes = bis->cur.bytes[BLKG_IOSTAT_WRITE]; | |
1003 | dbytes = bis->cur.bytes[BLKG_IOSTAT_DISCARD]; | |
1004 | rios = bis->cur.ios[BLKG_IOSTAT_READ]; | |
1005 | wios = bis->cur.ios[BLKG_IOSTAT_WRITE]; | |
1006 | dios = bis->cur.ios[BLKG_IOSTAT_DISCARD]; | |
1007 | } while (u64_stats_fetch_retry(&bis->sync, seq)); | |
2ee867dc | 1008 | |
49cb5168 | 1009 | if (rbytes || wbytes || rios || wios) { |
252c651a | 1010 | seq_printf(s, "rbytes=%llu wbytes=%llu rios=%llu wios=%llu dbytes=%llu dios=%llu", |
49cb5168 CH |
1011 | rbytes, wbytes, rios, wios, |
1012 | dbytes, dios); | |
1013 | } | |
903d23f0 | 1014 | |
49cb5168 | 1015 | if (blkcg_debug_stats && atomic_read(&blkg->use_delay)) { |
252c651a | 1016 | seq_printf(s, " use_delay=%d delay_nsec=%llu", |
49cb5168 CH |
1017 | atomic_read(&blkg->use_delay), |
1018 | atomic64_read(&blkg->delay_nsec)); | |
1019 | } | |
2ee867dc | 1020 | |
49cb5168 CH |
1021 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
1022 | struct blkcg_policy *pol = blkcg_policy[i]; | |
2ee867dc | 1023 | |
49cb5168 CH |
1024 | if (!blkg->pd[i] || !pol->pd_stat_fn) |
1025 | continue; | |
903d23f0 | 1026 | |
3607849d | 1027 | pol->pd_stat_fn(blkg->pd[i], s); |
49cb5168 | 1028 | } |
d09d8df3 | 1029 | |
3607849d | 1030 | seq_puts(s, "\n"); |
49cb5168 | 1031 | } |
903d23f0 | 1032 | |
49cb5168 CH |
1033 | static int blkcg_print_stat(struct seq_file *sf, void *v) |
1034 | { | |
1035 | struct blkcg *blkcg = css_to_blkcg(seq_css(sf)); | |
1036 | struct blkcg_gq *blkg; | |
903d23f0 | 1037 | |
49cb5168 CH |
1038 | if (!seq_css(sf)->parent) |
1039 | blkcg_fill_root_iostats(); | |
1040 | else | |
1041 | cgroup_rstat_flush(blkcg->css.cgroup); | |
07b0fdec | 1042 | |
49cb5168 CH |
1043 | rcu_read_lock(); |
1044 | hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) { | |
1045 | spin_lock_irq(&blkg->q->queue_lock); | |
1046 | blkcg_print_one_stat(blkg, sf); | |
b0814361 | 1047 | spin_unlock_irq(&blkg->q->queue_lock); |
2ee867dc | 1048 | } |
2ee867dc TH |
1049 | rcu_read_unlock(); |
1050 | return 0; | |
1051 | } | |
1052 | ||
e1f3b941 | 1053 | static struct cftype blkcg_files[] = { |
2ee867dc TH |
1054 | { |
1055 | .name = "stat", | |
1056 | .seq_show = blkcg_print_stat, | |
1057 | }, | |
1058 | { } /* terminate */ | |
1059 | }; | |
1060 | ||
e1f3b941 | 1061 | static struct cftype blkcg_legacy_files[] = { |
84c124da DS |
1062 | { |
1063 | .name = "reset_stats", | |
3c798398 | 1064 | .write_u64 = blkcg_reset_stats, |
22084190 | 1065 | }, |
4baf6e33 | 1066 | { } /* terminate */ |
31e4c28d VG |
1067 | }; |
1068 | ||
dec223c9 CH |
1069 | #ifdef CONFIG_CGROUP_WRITEBACK |
1070 | struct list_head *blkcg_get_cgwb_list(struct cgroup_subsys_state *css) | |
1071 | { | |
1072 | return &css_to_blkcg(css)->cgwb_list; | |
1073 | } | |
1074 | #endif | |
1075 | ||
59b57717 DZF |
1076 | /* |
1077 | * blkcg destruction is a three-stage process. | |
1078 | * | |
1079 | * 1. Destruction starts. The blkcg_css_offline() callback is invoked | |
1080 | * which offlines writeback. Here we tie the next stage of blkg destruction | |
1081 | * to the completion of writeback associated with the blkcg. This lets us | |
1082 | * avoid punting potentially large amounts of outstanding writeback to root | |
1083 | * while maintaining any ongoing policies. The next stage is triggered when | |
1084 | * the nr_cgwbs count goes to zero. | |
1085 | * | |
1086 | * 2. When the nr_cgwbs count goes to zero, blkcg_destroy_blkgs() is called | |
1087 | * and handles the destruction of blkgs. Here the css reference held by | |
1088 | * the blkg is put back eventually allowing blkcg_css_free() to be called. | |
1089 | * This work may occur in cgwb_release_workfn() on the cgwb_release | |
1090 | * workqueue. Any submitted ios that fail to get the blkg ref will be | |
1091 | * punted to the root_blkg. | |
1092 | * | |
1093 | * 3. Once the blkcg ref count goes to zero, blkcg_css_free() is called. | |
1094 | * This finally frees the blkcg. | |
1095 | */ | |
1096 | ||
59b57717 DZF |
1097 | /** |
1098 | * blkcg_destroy_blkgs - responsible for shooting down blkgs | |
1099 | * @blkcg: blkcg of interest | |
1100 | * | |
1101 | * blkgs should be removed while holding both q and blkcg locks. As blkcg lock | |
1102 | * is nested inside q lock, this function performs reverse double lock dancing. | |
1103 | * Destroying the blkgs releases the reference held on the blkcg's css allowing | |
1104 | * blkcg_css_free to eventually be called. | |
1105 | * | |
1106 | * This is the blkcg counterpart of ioc_release_fn(). | |
1107 | */ | |
397c9f46 | 1108 | static void blkcg_destroy_blkgs(struct blkcg *blkcg) |
59b57717 | 1109 | { |
6c635cae BW |
1110 | might_sleep(); |
1111 | ||
9f13ef67 | 1112 | spin_lock_irq(&blkcg->lock); |
7ee9c562 | 1113 | |
4c699480 JQ |
1114 | while (!hlist_empty(&blkcg->blkg_list)) { |
1115 | struct blkcg_gq *blkg = hlist_entry(blkcg->blkg_list.first, | |
6b065462 | 1116 | struct blkcg_gq, blkcg_node); |
4c699480 JQ |
1117 | struct request_queue *q = blkg->q; |
1118 | ||
6c635cae BW |
1119 | if (need_resched() || !spin_trylock(&q->queue_lock)) { |
1120 | /* | |
1121 | * Given that the system can accumulate a huge number | |
1122 | * of blkgs in pathological cases, check to see if we | |
1123 | * need to rescheduling to avoid softlockup. | |
1124 | */ | |
4c699480 | 1125 | spin_unlock_irq(&blkcg->lock); |
6c635cae | 1126 | cond_resched(); |
4c699480 | 1127 | spin_lock_irq(&blkcg->lock); |
6c635cae | 1128 | continue; |
4c699480 | 1129 | } |
6c635cae BW |
1130 | |
1131 | blkg_destroy(blkg); | |
1132 | spin_unlock(&q->queue_lock); | |
4c699480 | 1133 | } |
6b065462 | 1134 | |
4c699480 JQ |
1135 | spin_unlock_irq(&blkcg->lock); |
1136 | } | |
1137 | ||
397c9f46 CH |
1138 | /** |
1139 | * blkcg_pin_online - pin online state | |
1140 | * @blkcg_css: blkcg of interest | |
1141 | * | |
1142 | * While pinned, a blkcg is kept online. This is primarily used to | |
1143 | * impedance-match blkg and cgwb lifetimes so that blkg doesn't go offline | |
1144 | * while an associated cgwb is still active. | |
1145 | */ | |
1146 | void blkcg_pin_online(struct cgroup_subsys_state *blkcg_css) | |
1147 | { | |
1148 | refcount_inc(&css_to_blkcg(blkcg_css)->online_pin); | |
1149 | } | |
1150 | ||
1151 | /** | |
1152 | * blkcg_unpin_online - unpin online state | |
1153 | * @blkcg_css: blkcg of interest | |
1154 | * | |
1155 | * This is primarily used to impedance-match blkg and cgwb lifetimes so | |
1156 | * that blkg doesn't go offline while an associated cgwb is still active. | |
1157 | * When this count goes to zero, all active cgwbs have finished so the | |
1158 | * blkcg can continue destruction by calling blkcg_destroy_blkgs(). | |
1159 | */ | |
1160 | void blkcg_unpin_online(struct cgroup_subsys_state *blkcg_css) | |
1161 | { | |
1162 | struct blkcg *blkcg = css_to_blkcg(blkcg_css); | |
1163 | ||
1164 | do { | |
1165 | if (!refcount_dec_and_test(&blkcg->online_pin)) | |
1166 | break; | |
1167 | blkcg_destroy_blkgs(blkcg); | |
1168 | blkcg = blkcg_parent(blkcg); | |
1169 | } while (blkcg); | |
1170 | } | |
1171 | ||
1172 | /** | |
1173 | * blkcg_css_offline - cgroup css_offline callback | |
1174 | * @css: css of interest | |
1175 | * | |
1176 | * This function is called when @css is about to go away. Here the cgwbs are | |
1177 | * offlined first and only once writeback associated with the blkcg has | |
1178 | * finished do we start step 2 (see above). | |
1179 | */ | |
1180 | static void blkcg_css_offline(struct cgroup_subsys_state *css) | |
1181 | { | |
1182 | /* this prevents anyone from attaching or migrating to this blkcg */ | |
dec223c9 | 1183 | wb_blkcg_offline(css); |
397c9f46 CH |
1184 | |
1185 | /* put the base online pin allowing step 2 to be triggered */ | |
1186 | blkcg_unpin_online(css); | |
1187 | } | |
1188 | ||
eb95419b | 1189 | static void blkcg_css_free(struct cgroup_subsys_state *css) |
7ee9c562 | 1190 | { |
eb95419b | 1191 | struct blkcg *blkcg = css_to_blkcg(css); |
bc915e61 | 1192 | int i; |
7ee9c562 | 1193 | |
7876f930 | 1194 | mutex_lock(&blkcg_pol_mutex); |
e4a9bde9 | 1195 | |
7876f930 | 1196 | list_del(&blkcg->all_blkcgs_node); |
7876f930 | 1197 | |
bc915e61 | 1198 | for (i = 0; i < BLKCG_MAX_POLS; i++) |
e4a9bde9 TH |
1199 | if (blkcg->cpd[i]) |
1200 | blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]); | |
1201 | ||
1202 | mutex_unlock(&blkcg_pol_mutex); | |
1203 | ||
3b8cc629 | 1204 | free_percpu(blkcg->lhead); |
bc915e61 | 1205 | kfree(blkcg); |
31e4c28d VG |
1206 | } |
1207 | ||
eb95419b TH |
1208 | static struct cgroup_subsys_state * |
1209 | blkcg_css_alloc(struct cgroup_subsys_state *parent_css) | |
31e4c28d | 1210 | { |
3c798398 | 1211 | struct blkcg *blkcg; |
e48453c3 | 1212 | int i; |
31e4c28d | 1213 | |
7876f930 TH |
1214 | mutex_lock(&blkcg_pol_mutex); |
1215 | ||
eb95419b | 1216 | if (!parent_css) { |
3c798398 | 1217 | blkcg = &blkcg_root; |
bc915e61 TH |
1218 | } else { |
1219 | blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL); | |
b5a9adcb | 1220 | if (!blkcg) |
4c18c9e9 | 1221 | goto unlock; |
e48453c3 AA |
1222 | } |
1223 | ||
3b8cc629 WL |
1224 | if (init_blkcg_llists(blkcg)) |
1225 | goto free_blkcg; | |
1226 | ||
e48453c3 AA |
1227 | for (i = 0; i < BLKCG_MAX_POLS ; i++) { |
1228 | struct blkcg_policy *pol = blkcg_policy[i]; | |
1229 | struct blkcg_policy_data *cpd; | |
1230 | ||
1231 | /* | |
1232 | * If the policy hasn't been attached yet, wait for it | |
1233 | * to be attached before doing anything else. Otherwise, | |
1234 | * check if the policy requires any specific per-cgroup | |
1235 | * data: if it does, allocate and initialize it. | |
1236 | */ | |
e4a9bde9 | 1237 | if (!pol || !pol->cpd_alloc_fn) |
e48453c3 AA |
1238 | continue; |
1239 | ||
e4a9bde9 | 1240 | cpd = pol->cpd_alloc_fn(GFP_KERNEL); |
b5a9adcb | 1241 | if (!cpd) |
e48453c3 | 1242 | goto free_pd_blkcg; |
b5a9adcb | 1243 | |
81437648 TH |
1244 | blkcg->cpd[i] = cpd; |
1245 | cpd->blkcg = blkcg; | |
e48453c3 | 1246 | cpd->plid = i; |
e48453c3 | 1247 | } |
31e4c28d | 1248 | |
31e4c28d | 1249 | spin_lock_init(&blkcg->lock); |
d866dbf6 | 1250 | refcount_set(&blkcg->online_pin, 1); |
e00f4f4d | 1251 | INIT_RADIX_TREE(&blkcg->blkg_tree, GFP_NOWAIT | __GFP_NOWARN); |
31e4c28d | 1252 | INIT_HLIST_HEAD(&blkcg->blkg_list); |
52ebea74 TH |
1253 | #ifdef CONFIG_CGROUP_WRITEBACK |
1254 | INIT_LIST_HEAD(&blkcg->cgwb_list); | |
1255 | #endif | |
7876f930 TH |
1256 | list_add_tail(&blkcg->all_blkcgs_node, &all_blkcgs); |
1257 | ||
1258 | mutex_unlock(&blkcg_pol_mutex); | |
31e4c28d | 1259 | return &blkcg->css; |
e48453c3 AA |
1260 | |
1261 | free_pd_blkcg: | |
1262 | for (i--; i >= 0; i--) | |
e4a9bde9 TH |
1263 | if (blkcg->cpd[i]) |
1264 | blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]); | |
3b8cc629 WL |
1265 | free_percpu(blkcg->lhead); |
1266 | free_blkcg: | |
4c18c9e9 | 1267 | if (blkcg != &blkcg_root) |
1268 | kfree(blkcg); | |
1269 | unlock: | |
7876f930 | 1270 | mutex_unlock(&blkcg_pol_mutex); |
b5a9adcb | 1271 | return ERR_PTR(-ENOMEM); |
31e4c28d VG |
1272 | } |
1273 | ||
4308a434 TH |
1274 | static int blkcg_css_online(struct cgroup_subsys_state *css) |
1275 | { | |
397c9f46 | 1276 | struct blkcg *parent = blkcg_parent(css_to_blkcg(css)); |
4308a434 TH |
1277 | |
1278 | /* | |
1279 | * blkcg_pin_online() is used to delay blkcg offline so that blkgs | |
1280 | * don't go offline while cgwbs are still active on them. Pin the | |
1281 | * parent so that offline always happens towards the root. | |
1282 | */ | |
1283 | if (parent) | |
d7dbd43f | 1284 | blkcg_pin_online(&parent->css); |
4308a434 TH |
1285 | return 0; |
1286 | } | |
1287 | ||
9823538f | 1288 | int blkcg_init_disk(struct gendisk *disk) |
5efd6113 | 1289 | { |
9823538f | 1290 | struct request_queue *q = disk->queue; |
d708f0d5 JA |
1291 | struct blkcg_gq *new_blkg, *blkg; |
1292 | bool preloaded; | |
ec13b1d6 TH |
1293 | int ret; |
1294 | ||
472e4314 | 1295 | INIT_LIST_HEAD(&q->blkg_list); |
1231039d | 1296 | mutex_init(&q->blkcg_mutex); |
472e4314 | 1297 | |
99e60387 | 1298 | new_blkg = blkg_alloc(&blkcg_root, disk, GFP_KERNEL); |
d708f0d5 JA |
1299 | if (!new_blkg) |
1300 | return -ENOMEM; | |
1301 | ||
1302 | preloaded = !radix_tree_preload(GFP_KERNEL); | |
1303 | ||
bea54883 | 1304 | /* Make sure the root blkg exists. */ |
77c570a1 | 1305 | /* spin_lock_irq can serve as RCU read-side critical section. */ |
0d945c1f | 1306 | spin_lock_irq(&q->queue_lock); |
99e60387 | 1307 | blkg = blkg_create(&blkcg_root, disk, new_blkg); |
901932a3 JB |
1308 | if (IS_ERR(blkg)) |
1309 | goto err_unlock; | |
1310 | q->root_blkg = blkg; | |
0d945c1f | 1311 | spin_unlock_irq(&q->queue_lock); |
ec13b1d6 | 1312 | |
d708f0d5 JA |
1313 | if (preloaded) |
1314 | radix_tree_preload_end(); | |
1315 | ||
b0dde3f5 | 1316 | ret = blk_ioprio_init(disk); |
556910e3 BVA |
1317 | if (ret) |
1318 | goto err_destroy_all; | |
1319 | ||
e13793ba | 1320 | ret = blk_throtl_init(disk); |
19688d7f | 1321 | if (ret) |
33dc6279 | 1322 | goto err_ioprio_exit; |
19688d7f | 1323 | |
16fac1b5 | 1324 | ret = blk_iolatency_init(disk); |
33dc6279 CH |
1325 | if (ret) |
1326 | goto err_throtl_exit; | |
6f5ddde4 | 1327 | |
04be60b5 | 1328 | return 0; |
901932a3 | 1329 | |
33dc6279 | 1330 | err_throtl_exit: |
e13793ba | 1331 | blk_throtl_exit(disk); |
33dc6279 | 1332 | err_ioprio_exit: |
b0dde3f5 | 1333 | blk_ioprio_exit(disk); |
04be60b5 | 1334 | err_destroy_all: |
00ad6991 | 1335 | blkg_destroy_all(disk); |
04be60b5 | 1336 | return ret; |
901932a3 | 1337 | err_unlock: |
0d945c1f | 1338 | spin_unlock_irq(&q->queue_lock); |
901932a3 JB |
1339 | if (preloaded) |
1340 | radix_tree_preload_end(); | |
1341 | return PTR_ERR(blkg); | |
5efd6113 TH |
1342 | } |
1343 | ||
9823538f | 1344 | void blkcg_exit_disk(struct gendisk *disk) |
5efd6113 | 1345 | { |
00ad6991 | 1346 | blkg_destroy_all(disk); |
813e6930 | 1347 | rq_qos_exit(disk->queue); |
e13793ba | 1348 | blk_throtl_exit(disk); |
5efd6113 TH |
1349 | } |
1350 | ||
d09d8df3 JB |
1351 | static void blkcg_exit(struct task_struct *tsk) |
1352 | { | |
f05837ed CH |
1353 | if (tsk->throttle_disk) |
1354 | put_disk(tsk->throttle_disk); | |
1355 | tsk->throttle_disk = NULL; | |
d09d8df3 JB |
1356 | } |
1357 | ||
c165b3e3 | 1358 | struct cgroup_subsys io_cgrp_subsys = { |
92fb9748 | 1359 | .css_alloc = blkcg_css_alloc, |
4308a434 | 1360 | .css_online = blkcg_css_online, |
92fb9748 TH |
1361 | .css_offline = blkcg_css_offline, |
1362 | .css_free = blkcg_css_free, | |
f7331648 | 1363 | .css_rstat_flush = blkcg_rstat_flush, |
2ee867dc | 1364 | .dfl_cftypes = blkcg_files, |
880f50e2 | 1365 | .legacy_cftypes = blkcg_legacy_files, |
c165b3e3 | 1366 | .legacy_name = "blkio", |
d09d8df3 | 1367 | .exit = blkcg_exit, |
1ced953b TH |
1368 | #ifdef CONFIG_MEMCG |
1369 | /* | |
1370 | * This ensures that, if available, memcg is automatically enabled | |
1371 | * together on the default hierarchy so that the owner cgroup can | |
1372 | * be retrieved from writeback pages. | |
1373 | */ | |
1374 | .depends_on = 1 << memory_cgrp_id, | |
1375 | #endif | |
676f7c8f | 1376 | }; |
c165b3e3 | 1377 | EXPORT_SYMBOL_GPL(io_cgrp_subsys); |
676f7c8f | 1378 | |
a2b1693b | 1379 | /** |
40e4996e CH |
1380 | * blkcg_activate_policy - activate a blkcg policy on a gendisk |
1381 | * @disk: gendisk of interest | |
a2b1693b TH |
1382 | * @pol: blkcg policy to activate |
1383 | * | |
40e4996e | 1384 | * Activate @pol on @disk. Requires %GFP_KERNEL context. @disk goes through |
a2b1693b TH |
1385 | * bypass mode to populate its blkgs with policy_data for @pol. |
1386 | * | |
40e4996e | 1387 | * Activation happens with @disk bypassed, so nobody would be accessing blkgs |
a2b1693b TH |
1388 | * from IO path. Update of each blkg is protected by both queue and blkcg |
1389 | * locks so that holding either lock and testing blkcg_policy_enabled() is | |
1390 | * always enough for dereferencing policy data. | |
1391 | * | |
1392 | * The caller is responsible for synchronizing [de]activations and policy | |
1393 | * [un]registerations. Returns 0 on success, -errno on failure. | |
1394 | */ | |
40e4996e | 1395 | int blkcg_activate_policy(struct gendisk *disk, const struct blkcg_policy *pol) |
a2b1693b | 1396 | { |
40e4996e | 1397 | struct request_queue *q = disk->queue; |
4c55f4f9 | 1398 | struct blkg_policy_data *pd_prealloc = NULL; |
9d179b86 | 1399 | struct blkcg_gq *blkg, *pinned_blkg = NULL; |
4c55f4f9 | 1400 | int ret; |
a2b1693b TH |
1401 | |
1402 | if (blkcg_policy_enabled(q, pol)) | |
1403 | return 0; | |
1404 | ||
344e9ffc | 1405 | if (queue_is_mq(q)) |
bd166ef1 | 1406 | blk_mq_freeze_queue(q); |
9d179b86 | 1407 | retry: |
0d945c1f | 1408 | spin_lock_irq(&q->queue_lock); |
a2b1693b | 1409 | |
9d179b86 | 1410 | /* blkg_list is pushed at the head, reverse walk to allocate parents first */ |
71c81407 | 1411 | list_for_each_entry_reverse(blkg, &q->blkg_list, q_node) { |
4c55f4f9 TH |
1412 | struct blkg_policy_data *pd; |
1413 | ||
1414 | if (blkg->pd[pol->plid]) | |
1415 | continue; | |
a2b1693b | 1416 | |
9d179b86 TH |
1417 | /* If prealloc matches, use it; otherwise try GFP_NOWAIT */ |
1418 | if (blkg == pinned_blkg) { | |
1419 | pd = pd_prealloc; | |
1420 | pd_prealloc = NULL; | |
1421 | } else { | |
0a0b4f79 CH |
1422 | pd = pol->pd_alloc_fn(disk, blkg->blkcg, |
1423 | GFP_NOWAIT | __GFP_NOWARN); | |
9d179b86 TH |
1424 | } |
1425 | ||
4c55f4f9 | 1426 | if (!pd) { |
9d179b86 TH |
1427 | /* |
1428 | * GFP_NOWAIT failed. Free the existing one and | |
1429 | * prealloc for @blkg w/ GFP_KERNEL. | |
1430 | */ | |
1431 | if (pinned_blkg) | |
1432 | blkg_put(pinned_blkg); | |
1433 | blkg_get(blkg); | |
1434 | pinned_blkg = blkg; | |
1435 | ||
0d945c1f | 1436 | spin_unlock_irq(&q->queue_lock); |
9d179b86 TH |
1437 | |
1438 | if (pd_prealloc) | |
1439 | pol->pd_free_fn(pd_prealloc); | |
0a0b4f79 CH |
1440 | pd_prealloc = pol->pd_alloc_fn(disk, blkg->blkcg, |
1441 | GFP_KERNEL); | |
9d179b86 TH |
1442 | if (pd_prealloc) |
1443 | goto retry; | |
1444 | else | |
1445 | goto enomem; | |
4c55f4f9 | 1446 | } |
a2b1693b TH |
1447 | |
1448 | blkg->pd[pol->plid] = pd; | |
1449 | pd->blkg = blkg; | |
b276a876 | 1450 | pd->plid = pol->plid; |
dfd6200a | 1451 | pd->online = false; |
a2b1693b TH |
1452 | } |
1453 | ||
9d179b86 TH |
1454 | /* all allocated, init in the same order */ |
1455 | if (pol->pd_init_fn) | |
1456 | list_for_each_entry_reverse(blkg, &q->blkg_list, q_node) | |
1457 | pol->pd_init_fn(blkg->pd[pol->plid]); | |
1458 | ||
1231039d | 1459 | list_for_each_entry_reverse(blkg, &q->blkg_list, q_node) { |
dfd6200a | 1460 | if (pol->pd_online_fn) |
e3ff8887 | 1461 | pol->pd_online_fn(blkg->pd[pol->plid]); |
dfd6200a YK |
1462 | blkg->pd[pol->plid]->online = true; |
1463 | } | |
e3ff8887 | 1464 | |
a2b1693b TH |
1465 | __set_bit(pol->plid, q->blkcg_pols); |
1466 | ret = 0; | |
4c55f4f9 | 1467 | |
0d945c1f | 1468 | spin_unlock_irq(&q->queue_lock); |
9d179b86 | 1469 | out: |
344e9ffc | 1470 | if (queue_is_mq(q)) |
bd166ef1 | 1471 | blk_mq_unfreeze_queue(q); |
9d179b86 TH |
1472 | if (pinned_blkg) |
1473 | blkg_put(pinned_blkg); | |
001bea73 TH |
1474 | if (pd_prealloc) |
1475 | pol->pd_free_fn(pd_prealloc); | |
a2b1693b | 1476 | return ret; |
9d179b86 TH |
1477 | |
1478 | enomem: | |
1479 | /* alloc failed, nothing's initialized yet, free everything */ | |
1480 | spin_lock_irq(&q->queue_lock); | |
1481 | list_for_each_entry(blkg, &q->blkg_list, q_node) { | |
858560b2 LJ |
1482 | struct blkcg *blkcg = blkg->blkcg; |
1483 | ||
1484 | spin_lock(&blkcg->lock); | |
9d179b86 TH |
1485 | if (blkg->pd[pol->plid]) { |
1486 | pol->pd_free_fn(blkg->pd[pol->plid]); | |
1487 | blkg->pd[pol->plid] = NULL; | |
1488 | } | |
858560b2 | 1489 | spin_unlock(&blkcg->lock); |
9d179b86 TH |
1490 | } |
1491 | spin_unlock_irq(&q->queue_lock); | |
1492 | ret = -ENOMEM; | |
1493 | goto out; | |
a2b1693b TH |
1494 | } |
1495 | EXPORT_SYMBOL_GPL(blkcg_activate_policy); | |
1496 | ||
1497 | /** | |
40e4996e CH |
1498 | * blkcg_deactivate_policy - deactivate a blkcg policy on a gendisk |
1499 | * @disk: gendisk of interest | |
a2b1693b TH |
1500 | * @pol: blkcg policy to deactivate |
1501 | * | |
40e4996e | 1502 | * Deactivate @pol on @disk. Follows the same synchronization rules as |
a2b1693b TH |
1503 | * blkcg_activate_policy(). |
1504 | */ | |
40e4996e | 1505 | void blkcg_deactivate_policy(struct gendisk *disk, |
3c798398 | 1506 | const struct blkcg_policy *pol) |
a2b1693b | 1507 | { |
40e4996e | 1508 | struct request_queue *q = disk->queue; |
3c798398 | 1509 | struct blkcg_gq *blkg; |
a2b1693b TH |
1510 | |
1511 | if (!blkcg_policy_enabled(q, pol)) | |
1512 | return; | |
1513 | ||
344e9ffc | 1514 | if (queue_is_mq(q)) |
bd166ef1 | 1515 | blk_mq_freeze_queue(q); |
bd166ef1 | 1516 | |
1231039d | 1517 | mutex_lock(&q->blkcg_mutex); |
0d945c1f | 1518 | spin_lock_irq(&q->queue_lock); |
a2b1693b TH |
1519 | |
1520 | __clear_bit(pol->plid, q->blkcg_pols); | |
1521 | ||
1522 | list_for_each_entry(blkg, &q->blkg_list, q_node) { | |
858560b2 LJ |
1523 | struct blkcg *blkcg = blkg->blkcg; |
1524 | ||
1525 | spin_lock(&blkcg->lock); | |
001bea73 | 1526 | if (blkg->pd[pol->plid]) { |
dfd6200a | 1527 | if (blkg->pd[pol->plid]->online && pol->pd_offline_fn) |
a9520cd6 | 1528 | pol->pd_offline_fn(blkg->pd[pol->plid]); |
001bea73 TH |
1529 | pol->pd_free_fn(blkg->pd[pol->plid]); |
1530 | blkg->pd[pol->plid] = NULL; | |
1531 | } | |
858560b2 | 1532 | spin_unlock(&blkcg->lock); |
a2b1693b TH |
1533 | } |
1534 | ||
0d945c1f | 1535 | spin_unlock_irq(&q->queue_lock); |
1231039d | 1536 | mutex_unlock(&q->blkcg_mutex); |
bd166ef1 | 1537 | |
344e9ffc | 1538 | if (queue_is_mq(q)) |
bd166ef1 | 1539 | blk_mq_unfreeze_queue(q); |
a2b1693b TH |
1540 | } |
1541 | EXPORT_SYMBOL_GPL(blkcg_deactivate_policy); | |
1542 | ||
e55cf798 JY |
1543 | static void blkcg_free_all_cpd(struct blkcg_policy *pol) |
1544 | { | |
1545 | struct blkcg *blkcg; | |
1546 | ||
1547 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) { | |
1548 | if (blkcg->cpd[pol->plid]) { | |
1549 | pol->cpd_free_fn(blkcg->cpd[pol->plid]); | |
1550 | blkcg->cpd[pol->plid] = NULL; | |
1551 | } | |
1552 | } | |
1553 | } | |
1554 | ||
8bd435b3 | 1555 | /** |
3c798398 TH |
1556 | * blkcg_policy_register - register a blkcg policy |
1557 | * @pol: blkcg policy to register | |
8bd435b3 | 1558 | * |
3c798398 TH |
1559 | * Register @pol with blkcg core. Might sleep and @pol may be modified on |
1560 | * successful registration. Returns 0 on success and -errno on failure. | |
8bd435b3 | 1561 | */ |
d5bf0291 | 1562 | int blkcg_policy_register(struct blkcg_policy *pol) |
3e252066 | 1563 | { |
06b285bd | 1564 | struct blkcg *blkcg; |
8bd435b3 | 1565 | int i, ret; |
e8989fae | 1566 | |
838f13bf | 1567 | mutex_lock(&blkcg_pol_register_mutex); |
bc0d6501 TH |
1568 | mutex_lock(&blkcg_pol_mutex); |
1569 | ||
8bd435b3 TH |
1570 | /* find an empty slot */ |
1571 | ret = -ENOSPC; | |
1572 | for (i = 0; i < BLKCG_MAX_POLS; i++) | |
3c798398 | 1573 | if (!blkcg_policy[i]) |
8bd435b3 | 1574 | break; |
01c5f85a JA |
1575 | if (i >= BLKCG_MAX_POLS) { |
1576 | pr_warn("blkcg_policy_register: BLKCG_MAX_POLS too small\n"); | |
838f13bf | 1577 | goto err_unlock; |
01c5f85a | 1578 | } |
035d10b2 | 1579 | |
e8401073 | 1580 | /* Make sure cpd/pd_alloc_fn and cpd/pd_free_fn in pairs */ |
1581 | if ((!pol->cpd_alloc_fn ^ !pol->cpd_free_fn) || | |
1582 | (!pol->pd_alloc_fn ^ !pol->pd_free_fn)) | |
1583 | goto err_unlock; | |
1584 | ||
06b285bd | 1585 | /* register @pol */ |
3c798398 | 1586 | pol->plid = i; |
06b285bd TH |
1587 | blkcg_policy[pol->plid] = pol; |
1588 | ||
1589 | /* allocate and install cpd's */ | |
e4a9bde9 | 1590 | if (pol->cpd_alloc_fn) { |
06b285bd TH |
1591 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) { |
1592 | struct blkcg_policy_data *cpd; | |
1593 | ||
e4a9bde9 | 1594 | cpd = pol->cpd_alloc_fn(GFP_KERNEL); |
bbb427e3 | 1595 | if (!cpd) |
06b285bd | 1596 | goto err_free_cpds; |
06b285bd | 1597 | |
81437648 TH |
1598 | blkcg->cpd[pol->plid] = cpd; |
1599 | cpd->blkcg = blkcg; | |
06b285bd | 1600 | cpd->plid = pol->plid; |
06b285bd TH |
1601 | } |
1602 | } | |
1603 | ||
838f13bf | 1604 | mutex_unlock(&blkcg_pol_mutex); |
8bd435b3 | 1605 | |
8bd435b3 | 1606 | /* everything is in place, add intf files for the new policy */ |
2ee867dc TH |
1607 | if (pol->dfl_cftypes) |
1608 | WARN_ON(cgroup_add_dfl_cftypes(&io_cgrp_subsys, | |
1609 | pol->dfl_cftypes)); | |
880f50e2 | 1610 | if (pol->legacy_cftypes) |
c165b3e3 | 1611 | WARN_ON(cgroup_add_legacy_cftypes(&io_cgrp_subsys, |
880f50e2 | 1612 | pol->legacy_cftypes)); |
838f13bf TH |
1613 | mutex_unlock(&blkcg_pol_register_mutex); |
1614 | return 0; | |
1615 | ||
06b285bd | 1616 | err_free_cpds: |
e55cf798 JY |
1617 | if (pol->cpd_free_fn) |
1618 | blkcg_free_all_cpd(pol); | |
1619 | ||
06b285bd | 1620 | blkcg_policy[pol->plid] = NULL; |
838f13bf | 1621 | err_unlock: |
bc0d6501 | 1622 | mutex_unlock(&blkcg_pol_mutex); |
838f13bf | 1623 | mutex_unlock(&blkcg_pol_register_mutex); |
8bd435b3 | 1624 | return ret; |
3e252066 | 1625 | } |
3c798398 | 1626 | EXPORT_SYMBOL_GPL(blkcg_policy_register); |
3e252066 | 1627 | |
8bd435b3 | 1628 | /** |
3c798398 TH |
1629 | * blkcg_policy_unregister - unregister a blkcg policy |
1630 | * @pol: blkcg policy to unregister | |
8bd435b3 | 1631 | * |
3c798398 | 1632 | * Undo blkcg_policy_register(@pol). Might sleep. |
8bd435b3 | 1633 | */ |
3c798398 | 1634 | void blkcg_policy_unregister(struct blkcg_policy *pol) |
3e252066 | 1635 | { |
838f13bf | 1636 | mutex_lock(&blkcg_pol_register_mutex); |
bc0d6501 | 1637 | |
3c798398 | 1638 | if (WARN_ON(blkcg_policy[pol->plid] != pol)) |
8bd435b3 TH |
1639 | goto out_unlock; |
1640 | ||
1641 | /* kill the intf files first */ | |
2ee867dc TH |
1642 | if (pol->dfl_cftypes) |
1643 | cgroup_rm_cftypes(pol->dfl_cftypes); | |
880f50e2 TH |
1644 | if (pol->legacy_cftypes) |
1645 | cgroup_rm_cftypes(pol->legacy_cftypes); | |
44ea53de | 1646 | |
06b285bd | 1647 | /* remove cpds and unregister */ |
838f13bf | 1648 | mutex_lock(&blkcg_pol_mutex); |
06b285bd | 1649 | |
e55cf798 JY |
1650 | if (pol->cpd_free_fn) |
1651 | blkcg_free_all_cpd(pol); | |
1652 | ||
3c798398 | 1653 | blkcg_policy[pol->plid] = NULL; |
06b285bd | 1654 | |
bc0d6501 | 1655 | mutex_unlock(&blkcg_pol_mutex); |
838f13bf TH |
1656 | out_unlock: |
1657 | mutex_unlock(&blkcg_pol_register_mutex); | |
3e252066 | 1658 | } |
3c798398 | 1659 | EXPORT_SYMBOL_GPL(blkcg_policy_unregister); |
903d23f0 | 1660 | |
d3f77dfd TH |
1661 | bool __blkcg_punt_bio_submit(struct bio *bio) |
1662 | { | |
1663 | struct blkcg_gq *blkg = bio->bi_blkg; | |
1664 | ||
1665 | /* consume the flag first */ | |
1666 | bio->bi_opf &= ~REQ_CGROUP_PUNT; | |
1667 | ||
1668 | /* never bounce for the root cgroup */ | |
1669 | if (!blkg->parent) | |
1670 | return false; | |
1671 | ||
1672 | spin_lock_bh(&blkg->async_bio_lock); | |
1673 | bio_list_add(&blkg->async_bios, bio); | |
1674 | spin_unlock_bh(&blkg->async_bio_lock); | |
1675 | ||
1676 | queue_work(blkcg_punt_bio_wq, &blkg->async_bio_work); | |
1677 | return true; | |
1678 | } | |
1679 | ||
d09d8df3 JB |
1680 | /* |
1681 | * Scale the accumulated delay based on how long it has been since we updated | |
1682 | * the delay. We only call this when we are adding delay, in case it's been a | |
1683 | * while since we added delay, and when we are checking to see if we need to | |
1684 | * delay a task, to account for any delays that may have occurred. | |
1685 | */ | |
1686 | static void blkcg_scale_delay(struct blkcg_gq *blkg, u64 now) | |
1687 | { | |
1688 | u64 old = atomic64_read(&blkg->delay_start); | |
1689 | ||
54c52e10 TH |
1690 | /* negative use_delay means no scaling, see blkcg_set_delay() */ |
1691 | if (atomic_read(&blkg->use_delay) < 0) | |
1692 | return; | |
1693 | ||
d09d8df3 JB |
1694 | /* |
1695 | * We only want to scale down every second. The idea here is that we | |
1696 | * want to delay people for min(delay_nsec, NSEC_PER_SEC) in a certain | |
1697 | * time window. We only want to throttle tasks for recent delay that | |
1698 | * has occurred, in 1 second time windows since that's the maximum | |
1699 | * things can be throttled. We save the current delay window in | |
1700 | * blkg->last_delay so we know what amount is still left to be charged | |
1701 | * to the blkg from this point onward. blkg->last_use keeps track of | |
1702 | * the use_delay counter. The idea is if we're unthrottling the blkg we | |
1703 | * are ok with whatever is happening now, and we can take away more of | |
1704 | * the accumulated delay as we've already throttled enough that | |
1705 | * everybody is happy with their IO latencies. | |
1706 | */ | |
1707 | if (time_before64(old + NSEC_PER_SEC, now) && | |
96388f57 | 1708 | atomic64_try_cmpxchg(&blkg->delay_start, &old, now)) { |
d09d8df3 JB |
1709 | u64 cur = atomic64_read(&blkg->delay_nsec); |
1710 | u64 sub = min_t(u64, blkg->last_delay, now - old); | |
1711 | int cur_use = atomic_read(&blkg->use_delay); | |
1712 | ||
1713 | /* | |
1714 | * We've been unthrottled, subtract a larger chunk of our | |
1715 | * accumulated delay. | |
1716 | */ | |
1717 | if (cur_use < blkg->last_use) | |
1718 | sub = max_t(u64, sub, blkg->last_delay >> 1); | |
1719 | ||
1720 | /* | |
1721 | * This shouldn't happen, but handle it anyway. Our delay_nsec | |
1722 | * should only ever be growing except here where we subtract out | |
1723 | * min(last_delay, 1 second), but lord knows bugs happen and I'd | |
1724 | * rather not end up with negative numbers. | |
1725 | */ | |
1726 | if (unlikely(cur < sub)) { | |
1727 | atomic64_set(&blkg->delay_nsec, 0); | |
1728 | blkg->last_delay = 0; | |
1729 | } else { | |
1730 | atomic64_sub(sub, &blkg->delay_nsec); | |
1731 | blkg->last_delay = cur - sub; | |
1732 | } | |
1733 | blkg->last_use = cur_use; | |
1734 | } | |
1735 | } | |
1736 | ||
1737 | /* | |
1738 | * This is called when we want to actually walk up the hierarchy and check to | |
1739 | * see if we need to throttle, and then actually throttle if there is some | |
1740 | * accumulated delay. This should only be called upon return to user space so | |
1741 | * we're not holding some lock that would induce a priority inversion. | |
1742 | */ | |
1743 | static void blkcg_maybe_throttle_blkg(struct blkcg_gq *blkg, bool use_memdelay) | |
1744 | { | |
fd112c74 | 1745 | unsigned long pflags; |
5160a5a5 | 1746 | bool clamp; |
d09d8df3 JB |
1747 | u64 now = ktime_to_ns(ktime_get()); |
1748 | u64 exp; | |
1749 | u64 delay_nsec = 0; | |
1750 | int tok; | |
1751 | ||
1752 | while (blkg->parent) { | |
5160a5a5 TH |
1753 | int use_delay = atomic_read(&blkg->use_delay); |
1754 | ||
1755 | if (use_delay) { | |
1756 | u64 this_delay; | |
1757 | ||
d09d8df3 | 1758 | blkcg_scale_delay(blkg, now); |
5160a5a5 TH |
1759 | this_delay = atomic64_read(&blkg->delay_nsec); |
1760 | if (this_delay > delay_nsec) { | |
1761 | delay_nsec = this_delay; | |
1762 | clamp = use_delay > 0; | |
1763 | } | |
d09d8df3 JB |
1764 | } |
1765 | blkg = blkg->parent; | |
1766 | } | |
1767 | ||
1768 | if (!delay_nsec) | |
1769 | return; | |
1770 | ||
1771 | /* | |
1772 | * Let's not sleep for all eternity if we've amassed a huge delay. | |
1773 | * Swapping or metadata IO can accumulate 10's of seconds worth of | |
1774 | * delay, and we want userspace to be able to do _something_ so cap the | |
5160a5a5 TH |
1775 | * delays at 0.25s. If there's 10's of seconds worth of delay then the |
1776 | * tasks will be delayed for 0.25 second for every syscall. If | |
1777 | * blkcg_set_delay() was used as indicated by negative use_delay, the | |
1778 | * caller is responsible for regulating the range. | |
d09d8df3 | 1779 | */ |
5160a5a5 TH |
1780 | if (clamp) |
1781 | delay_nsec = min_t(u64, delay_nsec, 250 * NSEC_PER_MSEC); | |
d09d8df3 | 1782 | |
fd112c74 JB |
1783 | if (use_memdelay) |
1784 | psi_memstall_enter(&pflags); | |
d09d8df3 JB |
1785 | |
1786 | exp = ktime_add_ns(now, delay_nsec); | |
1787 | tok = io_schedule_prepare(); | |
1788 | do { | |
1789 | __set_current_state(TASK_KILLABLE); | |
1790 | if (!schedule_hrtimeout(&exp, HRTIMER_MODE_ABS)) | |
1791 | break; | |
1792 | } while (!fatal_signal_pending(current)); | |
1793 | io_schedule_finish(tok); | |
fd112c74 JB |
1794 | |
1795 | if (use_memdelay) | |
1796 | psi_memstall_leave(&pflags); | |
d09d8df3 JB |
1797 | } |
1798 | ||
1799 | /** | |
1800 | * blkcg_maybe_throttle_current - throttle the current task if it has been marked | |
1801 | * | |
1802 | * This is only called if we've been marked with set_notify_resume(). Obviously | |
1803 | * we can be set_notify_resume() for reasons other than blkcg throttling, so we | |
f05837ed | 1804 | * check to see if current->throttle_disk is set and if not this doesn't do |
d09d8df3 JB |
1805 | * anything. This should only ever be called by the resume code, it's not meant |
1806 | * to be called by people willy-nilly as it will actually do the work to | |
1807 | * throttle the task if it is setup for throttling. | |
1808 | */ | |
1809 | void blkcg_maybe_throttle_current(void) | |
1810 | { | |
f05837ed | 1811 | struct gendisk *disk = current->throttle_disk; |
d09d8df3 JB |
1812 | struct blkcg *blkcg; |
1813 | struct blkcg_gq *blkg; | |
1814 | bool use_memdelay = current->use_memdelay; | |
1815 | ||
f05837ed | 1816 | if (!disk) |
d09d8df3 JB |
1817 | return; |
1818 | ||
f05837ed | 1819 | current->throttle_disk = NULL; |
d09d8df3 JB |
1820 | current->use_memdelay = false; |
1821 | ||
1822 | rcu_read_lock(); | |
82778259 | 1823 | blkcg = css_to_blkcg(blkcg_css()); |
d09d8df3 JB |
1824 | if (!blkcg) |
1825 | goto out; | |
9a9c261e | 1826 | blkg = blkg_lookup(blkcg, disk->queue); |
d09d8df3 JB |
1827 | if (!blkg) |
1828 | goto out; | |
7754f669 | 1829 | if (!blkg_tryget(blkg)) |
d09d8df3 JB |
1830 | goto out; |
1831 | rcu_read_unlock(); | |
d09d8df3 JB |
1832 | |
1833 | blkcg_maybe_throttle_blkg(blkg, use_memdelay); | |
1834 | blkg_put(blkg); | |
f05837ed | 1835 | put_disk(disk); |
d09d8df3 JB |
1836 | return; |
1837 | out: | |
1838 | rcu_read_unlock(); | |
d09d8df3 | 1839 | } |
d09d8df3 JB |
1840 | |
1841 | /** | |
1842 | * blkcg_schedule_throttle - this task needs to check for throttling | |
1d6df9d3 | 1843 | * @disk: disk to throttle |
537d71b3 | 1844 | * @use_memdelay: do we charge this to memory delay for PSI |
d09d8df3 JB |
1845 | * |
1846 | * This is called by the IO controller when we know there's delay accumulated | |
1847 | * for the blkg for this task. We do not pass the blkg because there are places | |
1848 | * we call this that may not have that information, the swapping code for | |
de185b56 | 1849 | * instance will only have a block_device at that point. This set's the |
d09d8df3 JB |
1850 | * notify_resume for the task to check and see if it requires throttling before |
1851 | * returning to user space. | |
1852 | * | |
1853 | * We will only schedule once per syscall. You can call this over and over | |
1854 | * again and it will only do the check once upon return to user space, and only | |
1855 | * throttle once. If the task needs to be throttled again it'll need to be | |
1856 | * re-set at the next time we see the task. | |
1857 | */ | |
de185b56 | 1858 | void blkcg_schedule_throttle(struct gendisk *disk, bool use_memdelay) |
d09d8df3 JB |
1859 | { |
1860 | if (unlikely(current->flags & PF_KTHREAD)) | |
1861 | return; | |
1862 | ||
f05837ed CH |
1863 | if (current->throttle_disk != disk) { |
1864 | if (test_bit(GD_DEAD, &disk->state)) | |
49d1822b | 1865 | return; |
f05837ed | 1866 | get_device(disk_to_dev(disk)); |
49d1822b | 1867 | |
f05837ed CH |
1868 | if (current->throttle_disk) |
1869 | put_disk(current->throttle_disk); | |
1870 | current->throttle_disk = disk; | |
49d1822b | 1871 | } |
d09d8df3 | 1872 | |
d09d8df3 JB |
1873 | if (use_memdelay) |
1874 | current->use_memdelay = use_memdelay; | |
1875 | set_notify_resume(current); | |
1876 | } | |
d09d8df3 JB |
1877 | |
1878 | /** | |
1879 | * blkcg_add_delay - add delay to this blkg | |
537d71b3 BVA |
1880 | * @blkg: blkg of interest |
1881 | * @now: the current time in nanoseconds | |
1882 | * @delta: how many nanoseconds of delay to add | |
d09d8df3 JB |
1883 | * |
1884 | * Charge @delta to the blkg's current delay accumulation. This is used to | |
1885 | * throttle tasks if an IO controller thinks we need more throttling. | |
1886 | */ | |
1887 | void blkcg_add_delay(struct blkcg_gq *blkg, u64 now, u64 delta) | |
1888 | { | |
54c52e10 TH |
1889 | if (WARN_ON_ONCE(atomic_read(&blkg->use_delay) < 0)) |
1890 | return; | |
d09d8df3 JB |
1891 | blkcg_scale_delay(blkg, now); |
1892 | atomic64_add(delta, &blkg->delay_nsec); | |
1893 | } | |
d09d8df3 | 1894 | |
28fc591f CH |
1895 | /** |
1896 | * blkg_tryget_closest - try and get a blkg ref on the closet blkg | |
13c7863d CH |
1897 | * @bio: target bio |
1898 | * @css: target css | |
28fc591f | 1899 | * |
13c7863d CH |
1900 | * As the failure mode here is to walk up the blkg tree, this ensure that the |
1901 | * blkg->parent pointers are always valid. This returns the blkg that it ended | |
1902 | * up taking a reference on or %NULL if no reference was taken. | |
28fc591f | 1903 | */ |
13c7863d CH |
1904 | static inline struct blkcg_gq *blkg_tryget_closest(struct bio *bio, |
1905 | struct cgroup_subsys_state *css) | |
28fc591f | 1906 | { |
13c7863d | 1907 | struct blkcg_gq *blkg, *ret_blkg = NULL; |
28fc591f | 1908 | |
13c7863d | 1909 | rcu_read_lock(); |
99e60387 | 1910 | blkg = blkg_lookup_create(css_to_blkcg(css), bio->bi_bdev->bd_disk); |
28fc591f CH |
1911 | while (blkg) { |
1912 | if (blkg_tryget(blkg)) { | |
1913 | ret_blkg = blkg; | |
1914 | break; | |
1915 | } | |
1916 | blkg = blkg->parent; | |
1917 | } | |
13c7863d | 1918 | rcu_read_unlock(); |
28fc591f CH |
1919 | |
1920 | return ret_blkg; | |
1921 | } | |
1922 | ||
1923 | /** | |
1924 | * bio_associate_blkg_from_css - associate a bio with a specified css | |
1925 | * @bio: target bio | |
1926 | * @css: target css | |
1927 | * | |
1928 | * Associate @bio with the blkg found by combining the css's blkg and the | |
1929 | * request_queue of the @bio. An association failure is handled by walking up | |
1930 | * the blkg tree. Therefore, the blkg associated can be anything between @blkg | |
1931 | * and q->root_blkg. This situation only happens when a cgroup is dying and | |
1932 | * then the remaining bios will spill to the closest alive blkg. | |
1933 | * | |
1934 | * A reference will be taken on the blkg and will be released when @bio is | |
1935 | * freed. | |
1936 | */ | |
1937 | void bio_associate_blkg_from_css(struct bio *bio, | |
1938 | struct cgroup_subsys_state *css) | |
1939 | { | |
28fc591f CH |
1940 | if (bio->bi_blkg) |
1941 | blkg_put(bio->bi_blkg); | |
1942 | ||
a5b97526 | 1943 | if (css && css->parent) { |
13c7863d | 1944 | bio->bi_blkg = blkg_tryget_closest(bio, css); |
a5b97526 | 1945 | } else { |
ed6cddef PB |
1946 | blkg_get(bdev_get_queue(bio->bi_bdev)->root_blkg); |
1947 | bio->bi_blkg = bdev_get_queue(bio->bi_bdev)->root_blkg; | |
a5b97526 | 1948 | } |
28fc591f CH |
1949 | } |
1950 | EXPORT_SYMBOL_GPL(bio_associate_blkg_from_css); | |
1951 | ||
1952 | /** | |
1953 | * bio_associate_blkg - associate a bio with a blkg | |
1954 | * @bio: target bio | |
1955 | * | |
1956 | * Associate @bio with the blkg found from the bio's css and request_queue. | |
1957 | * If one is not found, bio_lookup_blkg() creates the blkg. If a blkg is | |
1958 | * already associated, the css is reused and association redone as the | |
1959 | * request_queue may have changed. | |
1960 | */ | |
1961 | void bio_associate_blkg(struct bio *bio) | |
1962 | { | |
1963 | struct cgroup_subsys_state *css; | |
1964 | ||
1965 | rcu_read_lock(); | |
1966 | ||
1967 | if (bio->bi_blkg) | |
bbb1ebe7 | 1968 | css = bio_blkcg_css(bio); |
28fc591f CH |
1969 | else |
1970 | css = blkcg_css(); | |
1971 | ||
1972 | bio_associate_blkg_from_css(bio, css); | |
1973 | ||
1974 | rcu_read_unlock(); | |
1975 | } | |
1976 | EXPORT_SYMBOL_GPL(bio_associate_blkg); | |
1977 | ||
1978 | /** | |
1979 | * bio_clone_blkg_association - clone blkg association from src to dst bio | |
1980 | * @dst: destination bio | |
1981 | * @src: source bio | |
1982 | */ | |
1983 | void bio_clone_blkg_association(struct bio *dst, struct bio *src) | |
1984 | { | |
22b106e5 JK |
1985 | if (src->bi_blkg) |
1986 | bio_associate_blkg_from_css(dst, bio_blkcg_css(src)); | |
28fc591f CH |
1987 | } |
1988 | EXPORT_SYMBOL_GPL(bio_clone_blkg_association); | |
1989 | ||
db18a53e CH |
1990 | static int blk_cgroup_io_type(struct bio *bio) |
1991 | { | |
1992 | if (op_is_discard(bio->bi_opf)) | |
1993 | return BLKG_IOSTAT_DISCARD; | |
1994 | if (op_is_write(bio->bi_opf)) | |
1995 | return BLKG_IOSTAT_WRITE; | |
1996 | return BLKG_IOSTAT_READ; | |
1997 | } | |
1998 | ||
1999 | void blk_cgroup_bio_start(struct bio *bio) | |
2000 | { | |
3b8cc629 | 2001 | struct blkcg *blkcg = bio->bi_blkg->blkcg; |
db18a53e CH |
2002 | int rwd = blk_cgroup_io_type(bio), cpu; |
2003 | struct blkg_iostat_set *bis; | |
3c08b093 | 2004 | unsigned long flags; |
db18a53e | 2005 | |
0416f3be ML |
2006 | /* Root-level stats are sourced from system-wide IO stats */ |
2007 | if (!cgroup_parent(blkcg->css.cgroup)) | |
2008 | return; | |
2009 | ||
db18a53e CH |
2010 | cpu = get_cpu(); |
2011 | bis = per_cpu_ptr(bio->bi_blkg->iostat_cpu, cpu); | |
3c08b093 | 2012 | flags = u64_stats_update_begin_irqsave(&bis->sync); |
db18a53e CH |
2013 | |
2014 | /* | |
2015 | * If the bio is flagged with BIO_CGROUP_ACCT it means this is a split | |
2016 | * bio and we would have already accounted for the size of the bio. | |
2017 | */ | |
2018 | if (!bio_flagged(bio, BIO_CGROUP_ACCT)) { | |
2019 | bio_set_flag(bio, BIO_CGROUP_ACCT); | |
0b8cc25d | 2020 | bis->cur.bytes[rwd] += bio->bi_iter.bi_size; |
db18a53e CH |
2021 | } |
2022 | bis->cur.ios[rwd]++; | |
2023 | ||
3b8cc629 WL |
2024 | /* |
2025 | * If the iostat_cpu isn't in a lockless list, put it into the | |
2026 | * list to indicate that a stat update is pending. | |
2027 | */ | |
2028 | if (!READ_ONCE(bis->lqueued)) { | |
2029 | struct llist_head *lhead = this_cpu_ptr(blkcg->lhead); | |
2030 | ||
2031 | llist_add(&bis->lnode, lhead); | |
2032 | WRITE_ONCE(bis->lqueued, true); | |
2033 | percpu_ref_get(&bis->blkg->refcnt); | |
2034 | } | |
2035 | ||
3c08b093 | 2036 | u64_stats_update_end_irqrestore(&bis->sync, flags); |
db18a53e | 2037 | if (cgroup_subsys_on_dfl(io_cgrp_subsys)) |
3b8cc629 | 2038 | cgroup_rstat_updated(blkcg->css.cgroup, cpu); |
db18a53e CH |
2039 | put_cpu(); |
2040 | } | |
2041 | ||
216889aa CH |
2042 | bool blk_cgroup_congested(void) |
2043 | { | |
2044 | struct cgroup_subsys_state *css; | |
2045 | bool ret = false; | |
2046 | ||
2047 | rcu_read_lock(); | |
d200ca14 | 2048 | for (css = blkcg_css(); css; css = css->parent) { |
216889aa CH |
2049 | if (atomic_read(&css->cgroup->congestion_count)) { |
2050 | ret = true; | |
2051 | break; | |
2052 | } | |
216889aa CH |
2053 | } |
2054 | rcu_read_unlock(); | |
2055 | return ret; | |
2056 | } | |
2057 | ||
d3f77dfd TH |
2058 | static int __init blkcg_init(void) |
2059 | { | |
2060 | blkcg_punt_bio_wq = alloc_workqueue("blkcg_punt_bio", | |
2061 | WQ_MEM_RECLAIM | WQ_FREEZABLE | | |
2062 | WQ_UNBOUND | WQ_SYSFS, 0); | |
2063 | if (!blkcg_punt_bio_wq) | |
2064 | return -ENOMEM; | |
2065 | return 0; | |
2066 | } | |
2067 | subsys_initcall(blkcg_init); | |
2068 | ||
903d23f0 JB |
2069 | module_param(blkcg_debug_stats, bool, 0644); |
2070 | MODULE_PARM_DESC(blkcg_debug_stats, "True if you want debug stats, false if not"); |