Commit | Line | Data |
---|---|---|
3dcf60bc | 1 | // SPDX-License-Identifier: GPL-2.0 |
31e4c28d VG |
2 | /* |
3 | * Common Block IO controller cgroup interface | |
4 | * | |
5 | * Based on ideas and code from CFQ, CFS and BFQ: | |
6 | * Copyright (C) 2003 Jens Axboe <axboe@kernel.dk> | |
7 | * | |
8 | * Copyright (C) 2008 Fabio Checconi <fabio@gandalf.sssup.it> | |
9 | * Paolo Valente <paolo.valente@unimore.it> | |
10 | * | |
11 | * Copyright (C) 2009 Vivek Goyal <vgoyal@redhat.com> | |
12 | * Nauman Rafique <nauman@google.com> | |
e48453c3 AA |
13 | * |
14 | * For policy-specific per-blkcg data: | |
15 | * Copyright (C) 2015 Paolo Valente <paolo.valente@unimore.it> | |
16 | * Arianna Avanzini <avanzini.arianna@gmail.com> | |
31e4c28d VG |
17 | */ |
18 | #include <linux/ioprio.h> | |
22084190 | 19 | #include <linux/kdev_t.h> |
9d6a986c | 20 | #include <linux/module.h> |
174cd4b1 | 21 | #include <linux/sched/signal.h> |
accee785 | 22 | #include <linux/err.h> |
9195291e | 23 | #include <linux/blkdev.h> |
52ebea74 | 24 | #include <linux/backing-dev.h> |
5a0e3ad6 | 25 | #include <linux/slab.h> |
72e06c25 | 26 | #include <linux/delay.h> |
9a9e8a26 | 27 | #include <linux/atomic.h> |
36aa9e5f | 28 | #include <linux/ctype.h> |
03248add | 29 | #include <linux/resume_user_mode.h> |
fd112c74 | 30 | #include <linux/psi.h> |
82d981d4 | 31 | #include <linux/part_stat.h> |
5efd6113 | 32 | #include "blk.h" |
672fdcf0 | 33 | #include "blk-cgroup.h" |
556910e3 | 34 | #include "blk-ioprio.h" |
a7b36ee6 | 35 | #include "blk-throttle.h" |
3e252066 | 36 | |
838f13bf TH |
37 | /* |
38 | * blkcg_pol_mutex protects blkcg_policy[] and policy [de]activation. | |
39 | * blkcg_pol_register_mutex nests outside of it and synchronizes entire | |
40 | * policy [un]register operations including cgroup file additions / | |
41 | * removals. Putting cgroup file registration outside blkcg_pol_mutex | |
42 | * allows grabbing it from cgroup callbacks. | |
43 | */ | |
44 | static DEFINE_MUTEX(blkcg_pol_register_mutex); | |
bc0d6501 | 45 | static DEFINE_MUTEX(blkcg_pol_mutex); |
923adde1 | 46 | |
e48453c3 | 47 | struct blkcg blkcg_root; |
3c798398 | 48 | EXPORT_SYMBOL_GPL(blkcg_root); |
9d6a986c | 49 | |
496d5e75 | 50 | struct cgroup_subsys_state * const blkcg_root_css = &blkcg_root.css; |
9b0eb69b | 51 | EXPORT_SYMBOL_GPL(blkcg_root_css); |
496d5e75 | 52 | |
3c798398 | 53 | static struct blkcg_policy *blkcg_policy[BLKCG_MAX_POLS]; |
035d10b2 | 54 | |
7876f930 TH |
55 | static LIST_HEAD(all_blkcgs); /* protected by blkcg_pol_mutex */ |
56 | ||
07b0fdec | 57 | bool blkcg_debug_stats = false; |
d3f77dfd | 58 | static struct workqueue_struct *blkcg_punt_bio_wq; |
903d23f0 | 59 | |
a731763f YK |
60 | #define BLKG_DESTROY_BATCH_SIZE 64 |
61 | ||
bc5fee91 CH |
62 | /** |
63 | * blkcg_css - find the current css | |
64 | * | |
65 | * Find the css associated with either the kthread or the current task. | |
66 | * This may return a dying css, so it is up to the caller to use tryget logic | |
67 | * to confirm it is alive and well. | |
68 | */ | |
69 | static struct cgroup_subsys_state *blkcg_css(void) | |
70 | { | |
71 | struct cgroup_subsys_state *css; | |
72 | ||
73 | css = kthread_blkcg(); | |
74 | if (css) | |
75 | return css; | |
76 | return task_css(current, io_cgrp_id); | |
77 | } | |
78 | ||
a2b1693b | 79 | static bool blkcg_policy_enabled(struct request_queue *q, |
3c798398 | 80 | const struct blkcg_policy *pol) |
a2b1693b TH |
81 | { |
82 | return pol && test_bit(pol->plid, q->blkcg_pols); | |
83 | } | |
84 | ||
d578c770 | 85 | static void blkg_free_workfn(struct work_struct *work) |
0381411e | 86 | { |
d578c770 ML |
87 | struct blkcg_gq *blkg = container_of(work, struct blkcg_gq, |
88 | free_work); | |
e8989fae | 89 | int i; |
549d3aa8 | 90 | |
db613670 | 91 | for (i = 0; i < BLKCG_MAX_POLS; i++) |
001bea73 TH |
92 | if (blkg->pd[i]) |
93 | blkcg_policy[i]->pd_free_fn(blkg->pd[i]); | |
e8989fae | 94 | |
0a9a25ca ML |
95 | if (blkg->q) |
96 | blk_put_queue(blkg->q); | |
f7331648 | 97 | free_percpu(blkg->iostat_cpu); |
ef069b97 | 98 | percpu_ref_exit(&blkg->refcnt); |
549d3aa8 | 99 | kfree(blkg); |
0381411e TH |
100 | } |
101 | ||
d578c770 ML |
102 | /** |
103 | * blkg_free - free a blkg | |
104 | * @blkg: blkg to free | |
105 | * | |
106 | * Free @blkg which may be partially allocated. | |
107 | */ | |
108 | static void blkg_free(struct blkcg_gq *blkg) | |
109 | { | |
110 | if (!blkg) | |
111 | return; | |
112 | ||
113 | /* | |
114 | * Both ->pd_free_fn() and request queue's release handler may | |
115 | * sleep, so free us by scheduling one work func | |
116 | */ | |
117 | INIT_WORK(&blkg->free_work, blkg_free_workfn); | |
118 | schedule_work(&blkg->free_work); | |
119 | } | |
120 | ||
7fcf2b03 DZ |
121 | static void __blkg_release(struct rcu_head *rcu) |
122 | { | |
123 | struct blkcg_gq *blkg = container_of(rcu, struct blkcg_gq, rcu_head); | |
124 | ||
d3f77dfd TH |
125 | WARN_ON(!bio_list_empty(&blkg->async_bios)); |
126 | ||
7fcf2b03 DZ |
127 | /* release the blkcg and parent blkg refs this blkg has been holding */ |
128 | css_put(&blkg->blkcg->css); | |
129 | if (blkg->parent) | |
130 | blkg_put(blkg->parent); | |
7fcf2b03 DZ |
131 | blkg_free(blkg); |
132 | } | |
133 | ||
134 | /* | |
135 | * A group is RCU protected, but having an rcu lock does not mean that one | |
136 | * can access all the fields of blkg and assume these are valid. For | |
137 | * example, don't try to follow throtl_data and request queue links. | |
138 | * | |
139 | * Having a reference to blkg under an rcu allows accesses to only values | |
140 | * local to groups like group stats and group rate limits. | |
141 | */ | |
142 | static void blkg_release(struct percpu_ref *ref) | |
143 | { | |
144 | struct blkcg_gq *blkg = container_of(ref, struct blkcg_gq, refcnt); | |
145 | ||
146 | call_rcu(&blkg->rcu_head, __blkg_release); | |
147 | } | |
148 | ||
d3f77dfd TH |
149 | static void blkg_async_bio_workfn(struct work_struct *work) |
150 | { | |
151 | struct blkcg_gq *blkg = container_of(work, struct blkcg_gq, | |
152 | async_bio_work); | |
153 | struct bio_list bios = BIO_EMPTY_LIST; | |
154 | struct bio *bio; | |
192f1c6b XT |
155 | struct blk_plug plug; |
156 | bool need_plug = false; | |
d3f77dfd TH |
157 | |
158 | /* as long as there are pending bios, @blkg can't go away */ | |
159 | spin_lock_bh(&blkg->async_bio_lock); | |
160 | bio_list_merge(&bios, &blkg->async_bios); | |
161 | bio_list_init(&blkg->async_bios); | |
162 | spin_unlock_bh(&blkg->async_bio_lock); | |
163 | ||
192f1c6b XT |
164 | /* start plug only when bio_list contains at least 2 bios */ |
165 | if (bios.head && bios.head->bi_next) { | |
166 | need_plug = true; | |
167 | blk_start_plug(&plug); | |
168 | } | |
d3f77dfd TH |
169 | while ((bio = bio_list_pop(&bios))) |
170 | submit_bio(bio); | |
192f1c6b XT |
171 | if (need_plug) |
172 | blk_finish_plug(&plug); | |
d3f77dfd TH |
173 | } |
174 | ||
bbb1ebe7 CH |
175 | /** |
176 | * bio_blkcg_css - return the blkcg CSS associated with a bio | |
177 | * @bio: target bio | |
178 | * | |
179 | * This returns the CSS for the blkcg associated with a bio, or %NULL if not | |
180 | * associated. Callers are expected to either handle %NULL or know association | |
181 | * has been done prior to calling this. | |
182 | */ | |
183 | struct cgroup_subsys_state *bio_blkcg_css(struct bio *bio) | |
184 | { | |
185 | if (!bio || !bio->bi_blkg) | |
186 | return NULL; | |
187 | return &bio->bi_blkg->blkcg->css; | |
188 | } | |
189 | EXPORT_SYMBOL_GPL(bio_blkcg_css); | |
190 | ||
397c9f46 CH |
191 | /** |
192 | * blkcg_parent - get the parent of a blkcg | |
193 | * @blkcg: blkcg of interest | |
194 | * | |
195 | * Return the parent blkcg of @blkcg. Can be called anytime. | |
196 | */ | |
197 | static inline struct blkcg *blkcg_parent(struct blkcg *blkcg) | |
198 | { | |
199 | return css_to_blkcg(blkcg->css.parent); | |
200 | } | |
201 | ||
0381411e TH |
202 | /** |
203 | * blkg_alloc - allocate a blkg | |
204 | * @blkcg: block cgroup the new blkg is associated with | |
205 | * @q: request_queue the new blkg is associated with | |
15974993 | 206 | * @gfp_mask: allocation mask to use |
0381411e | 207 | * |
e8989fae | 208 | * Allocate a new blkg assocating @blkcg and @q. |
0381411e | 209 | */ |
15974993 TH |
210 | static struct blkcg_gq *blkg_alloc(struct blkcg *blkcg, struct request_queue *q, |
211 | gfp_t gfp_mask) | |
0381411e | 212 | { |
3c798398 | 213 | struct blkcg_gq *blkg; |
f7331648 | 214 | int i, cpu; |
0381411e TH |
215 | |
216 | /* alloc and init base part */ | |
15974993 | 217 | blkg = kzalloc_node(sizeof(*blkg), gfp_mask, q->node); |
0381411e TH |
218 | if (!blkg) |
219 | return NULL; | |
220 | ||
ef069b97 TH |
221 | if (percpu_ref_init(&blkg->refcnt, blkg_release, 0, gfp_mask)) |
222 | goto err_free; | |
223 | ||
f7331648 TH |
224 | blkg->iostat_cpu = alloc_percpu_gfp(struct blkg_iostat_set, gfp_mask); |
225 | if (!blkg->iostat_cpu) | |
77ea7338 TH |
226 | goto err_free; |
227 | ||
0a9a25ca ML |
228 | if (!blk_get_queue(q)) |
229 | goto err_free; | |
230 | ||
c875f4d0 | 231 | blkg->q = q; |
e8989fae | 232 | INIT_LIST_HEAD(&blkg->q_node); |
d3f77dfd TH |
233 | spin_lock_init(&blkg->async_bio_lock); |
234 | bio_list_init(&blkg->async_bios); | |
235 | INIT_WORK(&blkg->async_bio_work, blkg_async_bio_workfn); | |
0381411e | 236 | blkg->blkcg = blkcg; |
0381411e | 237 | |
f7331648 TH |
238 | u64_stats_init(&blkg->iostat.sync); |
239 | for_each_possible_cpu(cpu) | |
240 | u64_stats_init(&per_cpu_ptr(blkg->iostat_cpu, cpu)->sync); | |
241 | ||
8bd435b3 | 242 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
3c798398 | 243 | struct blkcg_policy *pol = blkcg_policy[i]; |
e8989fae | 244 | struct blkg_policy_data *pd; |
0381411e | 245 | |
a2b1693b | 246 | if (!blkcg_policy_enabled(q, pol)) |
e8989fae TH |
247 | continue; |
248 | ||
249 | /* alloc per-policy data and attach it to blkg */ | |
cf09a8ee | 250 | pd = pol->pd_alloc_fn(gfp_mask, q, blkcg); |
a051661c TH |
251 | if (!pd) |
252 | goto err_free; | |
549d3aa8 | 253 | |
e8989fae TH |
254 | blkg->pd[i] = pd; |
255 | pd->blkg = blkg; | |
b276a876 | 256 | pd->plid = i; |
e8989fae TH |
257 | } |
258 | ||
0381411e | 259 | return blkg; |
a051661c TH |
260 | |
261 | err_free: | |
262 | blkg_free(blkg); | |
263 | return NULL; | |
0381411e TH |
264 | } |
265 | ||
24f29046 TH |
266 | struct blkcg_gq *blkg_lookup_slowpath(struct blkcg *blkcg, |
267 | struct request_queue *q, bool update_hint) | |
80fd9979 | 268 | { |
3c798398 | 269 | struct blkcg_gq *blkg; |
80fd9979 | 270 | |
a637120e | 271 | /* |
86cde6b6 TH |
272 | * Hint didn't match. Look up from the radix tree. Note that the |
273 | * hint can only be updated under queue_lock as otherwise @blkg | |
274 | * could have already been removed from blkg_tree. The caller is | |
275 | * responsible for grabbing queue_lock if @update_hint. | |
a637120e TH |
276 | */ |
277 | blkg = radix_tree_lookup(&blkcg->blkg_tree, q->id); | |
86cde6b6 TH |
278 | if (blkg && blkg->q == q) { |
279 | if (update_hint) { | |
0d945c1f | 280 | lockdep_assert_held(&q->queue_lock); |
86cde6b6 TH |
281 | rcu_assign_pointer(blkcg->blkg_hint, blkg); |
282 | } | |
a637120e | 283 | return blkg; |
86cde6b6 | 284 | } |
a637120e | 285 | |
80fd9979 TH |
286 | return NULL; |
287 | } | |
ae118896 | 288 | EXPORT_SYMBOL_GPL(blkg_lookup_slowpath); |
80fd9979 | 289 | |
15974993 | 290 | /* |
d708f0d5 JA |
291 | * If @new_blkg is %NULL, this function tries to allocate a new one as |
292 | * necessary using %GFP_NOWAIT. @new_blkg is always consumed on return. | |
15974993 | 293 | */ |
86cde6b6 | 294 | static struct blkcg_gq *blkg_create(struct blkcg *blkcg, |
d708f0d5 JA |
295 | struct request_queue *q, |
296 | struct blkcg_gq *new_blkg) | |
5624a4e4 | 297 | { |
d708f0d5 | 298 | struct blkcg_gq *blkg; |
f427d909 | 299 | int i, ret; |
5624a4e4 | 300 | |
0d945c1f | 301 | lockdep_assert_held(&q->queue_lock); |
cd1604fa | 302 | |
0273ac34 DZ |
303 | /* request_queue is dying, do not create/recreate a blkg */ |
304 | if (blk_queue_dying(q)) { | |
305 | ret = -ENODEV; | |
306 | goto err_free_blkg; | |
307 | } | |
308 | ||
7ee9c562 | 309 | /* blkg holds a reference to blkcg */ |
ec903c0c | 310 | if (!css_tryget_online(&blkcg->css)) { |
20386ce0 | 311 | ret = -ENODEV; |
93e6d5d8 | 312 | goto err_free_blkg; |
15974993 | 313 | } |
cd1604fa | 314 | |
d708f0d5 JA |
315 | /* allocate */ |
316 | if (!new_blkg) { | |
317 | new_blkg = blkg_alloc(blkcg, q, GFP_NOWAIT | __GFP_NOWARN); | |
318 | if (unlikely(!new_blkg)) { | |
319 | ret = -ENOMEM; | |
8c911f3d | 320 | goto err_put_css; |
15974993 TH |
321 | } |
322 | } | |
d708f0d5 | 323 | blkg = new_blkg; |
cd1604fa | 324 | |
db613670 | 325 | /* link parent */ |
3c547865 TH |
326 | if (blkcg_parent(blkcg)) { |
327 | blkg->parent = __blkg_lookup(blkcg_parent(blkcg), q, false); | |
328 | if (WARN_ON_ONCE(!blkg->parent)) { | |
20386ce0 | 329 | ret = -ENODEV; |
8c911f3d | 330 | goto err_put_css; |
3c547865 TH |
331 | } |
332 | blkg_get(blkg->parent); | |
333 | } | |
334 | ||
db613670 TH |
335 | /* invoke per-policy init */ |
336 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
337 | struct blkcg_policy *pol = blkcg_policy[i]; | |
338 | ||
339 | if (blkg->pd[i] && pol->pd_init_fn) | |
a9520cd6 | 340 | pol->pd_init_fn(blkg->pd[i]); |
db613670 TH |
341 | } |
342 | ||
343 | /* insert */ | |
cd1604fa | 344 | spin_lock(&blkcg->lock); |
a637120e TH |
345 | ret = radix_tree_insert(&blkcg->blkg_tree, q->id, blkg); |
346 | if (likely(!ret)) { | |
347 | hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list); | |
348 | list_add(&blkg->q_node, &q->blkg_list); | |
f427d909 TH |
349 | |
350 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
351 | struct blkcg_policy *pol = blkcg_policy[i]; | |
352 | ||
353 | if (blkg->pd[i] && pol->pd_online_fn) | |
a9520cd6 | 354 | pol->pd_online_fn(blkg->pd[i]); |
f427d909 | 355 | } |
a637120e | 356 | } |
f427d909 | 357 | blkg->online = true; |
cd1604fa | 358 | spin_unlock(&blkcg->lock); |
496fb780 | 359 | |
ec13b1d6 | 360 | if (!ret) |
a637120e | 361 | return blkg; |
15974993 | 362 | |
3c547865 TH |
363 | /* @blkg failed fully initialized, use the usual release path */ |
364 | blkg_put(blkg); | |
365 | return ERR_PTR(ret); | |
366 | ||
d708f0d5 | 367 | err_put_css: |
496fb780 | 368 | css_put(&blkcg->css); |
93e6d5d8 | 369 | err_free_blkg: |
d708f0d5 | 370 | blkg_free(new_blkg); |
93e6d5d8 | 371 | return ERR_PTR(ret); |
31e4c28d | 372 | } |
3c96cb32 | 373 | |
86cde6b6 | 374 | /** |
8c546287 | 375 | * blkg_lookup_create - lookup blkg, try to create one if not there |
86cde6b6 TH |
376 | * @blkcg: blkcg of interest |
377 | * @q: request_queue of interest | |
378 | * | |
379 | * Lookup blkg for the @blkcg - @q pair. If it doesn't exist, try to | |
3c547865 TH |
380 | * create one. blkg creation is performed recursively from blkcg_root such |
381 | * that all non-root blkg's have access to the parent blkg. This function | |
8c546287 | 382 | * should be called under RCU read lock and takes @q->queue_lock. |
86cde6b6 | 383 | * |
beea9da0 DZ |
384 | * Returns the blkg or the closest blkg if blkg_create() fails as it walks |
385 | * down from root. | |
86cde6b6 | 386 | */ |
8c546287 CH |
387 | static struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg, |
388 | struct request_queue *q) | |
3c96cb32 | 389 | { |
86cde6b6 | 390 | struct blkcg_gq *blkg; |
8c546287 | 391 | unsigned long flags; |
86cde6b6 TH |
392 | |
393 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
86cde6b6 | 394 | |
8c546287 | 395 | blkg = blkg_lookup(blkcg, q); |
86cde6b6 TH |
396 | if (blkg) |
397 | return blkg; | |
398 | ||
8c546287 CH |
399 | spin_lock_irqsave(&q->queue_lock, flags); |
400 | blkg = __blkg_lookup(blkcg, q, true); | |
401 | if (blkg) | |
402 | goto found; | |
403 | ||
3c547865 TH |
404 | /* |
405 | * Create blkgs walking down from blkcg_root to @blkcg, so that all | |
beea9da0 DZ |
406 | * non-root blkgs have access to their parents. Returns the closest |
407 | * blkg to the intended blkg should blkg_create() fail. | |
3c547865 TH |
408 | */ |
409 | while (true) { | |
410 | struct blkcg *pos = blkcg; | |
411 | struct blkcg *parent = blkcg_parent(blkcg); | |
beea9da0 DZ |
412 | struct blkcg_gq *ret_blkg = q->root_blkg; |
413 | ||
414 | while (parent) { | |
415 | blkg = __blkg_lookup(parent, q, false); | |
416 | if (blkg) { | |
417 | /* remember closest blkg */ | |
418 | ret_blkg = blkg; | |
419 | break; | |
420 | } | |
3c547865 TH |
421 | pos = parent; |
422 | parent = blkcg_parent(parent); | |
423 | } | |
424 | ||
d708f0d5 | 425 | blkg = blkg_create(pos, q, NULL); |
8c546287 CH |
426 | if (IS_ERR(blkg)) { |
427 | blkg = ret_blkg; | |
428 | break; | |
429 | } | |
beea9da0 | 430 | if (pos == blkcg) |
8c546287 | 431 | break; |
b978962a DZ |
432 | } |
433 | ||
8c546287 CH |
434 | found: |
435 | spin_unlock_irqrestore(&q->queue_lock, flags); | |
b978962a DZ |
436 | return blkg; |
437 | } | |
438 | ||
3c798398 | 439 | static void blkg_destroy(struct blkcg_gq *blkg) |
03aa264a | 440 | { |
3c798398 | 441 | struct blkcg *blkcg = blkg->blkcg; |
6b065462 | 442 | int i; |
03aa264a | 443 | |
0d945c1f | 444 | lockdep_assert_held(&blkg->q->queue_lock); |
9f13ef67 | 445 | lockdep_assert_held(&blkcg->lock); |
03aa264a TH |
446 | |
447 | /* Something wrong if we are trying to remove same group twice */ | |
e8989fae | 448 | WARN_ON_ONCE(list_empty(&blkg->q_node)); |
9f13ef67 | 449 | WARN_ON_ONCE(hlist_unhashed(&blkg->blkcg_node)); |
a637120e | 450 | |
6b065462 DZF |
451 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
452 | struct blkcg_policy *pol = blkcg_policy[i]; | |
453 | ||
454 | if (blkg->pd[i] && pol->pd_offline_fn) | |
455 | pol->pd_offline_fn(blkg->pd[i]); | |
456 | } | |
457 | ||
f427d909 TH |
458 | blkg->online = false; |
459 | ||
a637120e | 460 | radix_tree_delete(&blkcg->blkg_tree, blkg->q->id); |
e8989fae | 461 | list_del_init(&blkg->q_node); |
9f13ef67 | 462 | hlist_del_init_rcu(&blkg->blkcg_node); |
03aa264a | 463 | |
a637120e TH |
464 | /* |
465 | * Both setting lookup hint to and clearing it from @blkg are done | |
466 | * under queue_lock. If it's not pointing to @blkg now, it never | |
467 | * will. Hint assignment itself can race safely. | |
468 | */ | |
ec6c676a | 469 | if (rcu_access_pointer(blkcg->blkg_hint) == blkg) |
a637120e TH |
470 | rcu_assign_pointer(blkcg->blkg_hint, NULL); |
471 | ||
03aa264a TH |
472 | /* |
473 | * Put the reference taken at the time of creation so that when all | |
474 | * queues are gone, group can be destroyed. | |
475 | */ | |
7fcf2b03 | 476 | percpu_ref_kill(&blkg->refcnt); |
03aa264a TH |
477 | } |
478 | ||
9f13ef67 TH |
479 | /** |
480 | * blkg_destroy_all - destroy all blkgs associated with a request_queue | |
481 | * @q: request_queue of interest | |
9f13ef67 | 482 | * |
3c96cb32 | 483 | * Destroy all blkgs associated with @q. |
9f13ef67 | 484 | */ |
3c96cb32 | 485 | static void blkg_destroy_all(struct request_queue *q) |
72e06c25 | 486 | { |
3c798398 | 487 | struct blkcg_gq *blkg, *n; |
a731763f | 488 | int count = BLKG_DESTROY_BATCH_SIZE; |
72e06c25 | 489 | |
a731763f | 490 | restart: |
0d945c1f | 491 | spin_lock_irq(&q->queue_lock); |
9f13ef67 | 492 | list_for_each_entry_safe(blkg, n, &q->blkg_list, q_node) { |
3c798398 | 493 | struct blkcg *blkcg = blkg->blkcg; |
72e06c25 | 494 | |
9f13ef67 TH |
495 | spin_lock(&blkcg->lock); |
496 | blkg_destroy(blkg); | |
497 | spin_unlock(&blkcg->lock); | |
a731763f YK |
498 | |
499 | /* | |
500 | * in order to avoid holding the spin lock for too long, release | |
501 | * it when a batch of blkgs are destroyed. | |
502 | */ | |
503 | if (!(--count)) { | |
504 | count = BLKG_DESTROY_BATCH_SIZE; | |
505 | spin_unlock_irq(&q->queue_lock); | |
506 | cond_resched(); | |
507 | goto restart; | |
508 | } | |
72e06c25 | 509 | } |
6fe810bd TH |
510 | |
511 | q->root_blkg = NULL; | |
0d945c1f | 512 | spin_unlock_irq(&q->queue_lock); |
72e06c25 TH |
513 | } |
514 | ||
182446d0 TH |
515 | static int blkcg_reset_stats(struct cgroup_subsys_state *css, |
516 | struct cftype *cftype, u64 val) | |
303a3acb | 517 | { |
182446d0 | 518 | struct blkcg *blkcg = css_to_blkcg(css); |
3c798398 | 519 | struct blkcg_gq *blkg; |
f7331648 | 520 | int i, cpu; |
303a3acb | 521 | |
838f13bf | 522 | mutex_lock(&blkcg_pol_mutex); |
303a3acb | 523 | spin_lock_irq(&blkcg->lock); |
997a026c TH |
524 | |
525 | /* | |
526 | * Note that stat reset is racy - it doesn't synchronize against | |
527 | * stat updates. This is a debug feature which shouldn't exist | |
528 | * anyway. If you get hit by a race, retry. | |
529 | */ | |
b67bfe0d | 530 | hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) { |
f7331648 TH |
531 | for_each_possible_cpu(cpu) { |
532 | struct blkg_iostat_set *bis = | |
533 | per_cpu_ptr(blkg->iostat_cpu, cpu); | |
534 | memset(bis, 0, sizeof(*bis)); | |
535 | } | |
536 | memset(&blkg->iostat, 0, sizeof(blkg->iostat)); | |
77ea7338 | 537 | |
8bd435b3 | 538 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
3c798398 | 539 | struct blkcg_policy *pol = blkcg_policy[i]; |
549d3aa8 | 540 | |
a9520cd6 TH |
541 | if (blkg->pd[i] && pol->pd_reset_stats_fn) |
542 | pol->pd_reset_stats_fn(blkg->pd[i]); | |
bc0d6501 | 543 | } |
303a3acb | 544 | } |
f0bdc8cd | 545 | |
303a3acb | 546 | spin_unlock_irq(&blkcg->lock); |
bc0d6501 | 547 | mutex_unlock(&blkcg_pol_mutex); |
303a3acb DS |
548 | return 0; |
549 | } | |
550 | ||
dd165eb3 | 551 | const char *blkg_dev_name(struct blkcg_gq *blkg) |
303a3acb | 552 | { |
d152c682 | 553 | if (!blkg->q->disk || !blkg->q->disk->bdi->dev) |
edb0872f | 554 | return NULL; |
d152c682 | 555 | return bdi_dev_name(blkg->q->disk->bdi); |
303a3acb DS |
556 | } |
557 | ||
d3d32e69 TH |
558 | /** |
559 | * blkcg_print_blkgs - helper for printing per-blkg data | |
560 | * @sf: seq_file to print to | |
561 | * @blkcg: blkcg of interest | |
562 | * @prfill: fill function to print out a blkg | |
563 | * @pol: policy in question | |
564 | * @data: data to be passed to @prfill | |
565 | * @show_total: to print out sum of prfill return values or not | |
566 | * | |
567 | * This function invokes @prfill on each blkg of @blkcg if pd for the | |
568 | * policy specified by @pol exists. @prfill is invoked with @sf, the | |
810ecfa7 TH |
569 | * policy data and @data and the matching queue lock held. If @show_total |
570 | * is %true, the sum of the return values from @prfill is printed with | |
571 | * "Total" label at the end. | |
d3d32e69 TH |
572 | * |
573 | * This is to be used to construct print functions for | |
574 | * cftype->read_seq_string method. | |
575 | */ | |
3c798398 | 576 | void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg, |
f95a04af TH |
577 | u64 (*prfill)(struct seq_file *, |
578 | struct blkg_policy_data *, int), | |
3c798398 | 579 | const struct blkcg_policy *pol, int data, |
ec399347 | 580 | bool show_total) |
5624a4e4 | 581 | { |
3c798398 | 582 | struct blkcg_gq *blkg; |
d3d32e69 | 583 | u64 total = 0; |
5624a4e4 | 584 | |
810ecfa7 | 585 | rcu_read_lock(); |
ee89f812 | 586 | hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) { |
0d945c1f | 587 | spin_lock_irq(&blkg->q->queue_lock); |
a2b1693b | 588 | if (blkcg_policy_enabled(blkg->q, pol)) |
f95a04af | 589 | total += prfill(sf, blkg->pd[pol->plid], data); |
0d945c1f | 590 | spin_unlock_irq(&blkg->q->queue_lock); |
810ecfa7 TH |
591 | } |
592 | rcu_read_unlock(); | |
d3d32e69 TH |
593 | |
594 | if (show_total) | |
595 | seq_printf(sf, "Total %llu\n", (unsigned long long)total); | |
596 | } | |
829fdb50 | 597 | EXPORT_SYMBOL_GPL(blkcg_print_blkgs); |
d3d32e69 TH |
598 | |
599 | /** | |
600 | * __blkg_prfill_u64 - prfill helper for a single u64 value | |
601 | * @sf: seq_file to print to | |
f95a04af | 602 | * @pd: policy private data of interest |
d3d32e69 TH |
603 | * @v: value to print |
604 | * | |
f95a04af | 605 | * Print @v to @sf for the device assocaited with @pd. |
d3d32e69 | 606 | */ |
f95a04af | 607 | u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v) |
d3d32e69 | 608 | { |
f95a04af | 609 | const char *dname = blkg_dev_name(pd->blkg); |
d3d32e69 TH |
610 | |
611 | if (!dname) | |
612 | return 0; | |
613 | ||
614 | seq_printf(sf, "%s %llu\n", dname, (unsigned long long)v); | |
615 | return v; | |
616 | } | |
829fdb50 | 617 | EXPORT_SYMBOL_GPL(__blkg_prfill_u64); |
d3d32e69 | 618 | |
457e490f TE |
619 | /* Performs queue bypass and policy enabled checks then looks up blkg. */ |
620 | static struct blkcg_gq *blkg_lookup_check(struct blkcg *blkcg, | |
621 | const struct blkcg_policy *pol, | |
622 | struct request_queue *q) | |
623 | { | |
624 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
0d945c1f | 625 | lockdep_assert_held(&q->queue_lock); |
457e490f TE |
626 | |
627 | if (!blkcg_policy_enabled(q, pol)) | |
628 | return ERR_PTR(-EOPNOTSUPP); | |
457e490f TE |
629 | return __blkg_lookup(blkcg, q, true /* update_hint */); |
630 | } | |
631 | ||
015d254c | 632 | /** |
22ae8ce8 | 633 | * blkcg_conf_open_bdev - parse and open bdev for per-blkg config update |
015d254c TH |
634 | * @inputp: input string pointer |
635 | * | |
636 | * Parse the device node prefix part, MAJ:MIN, of per-blkg config update | |
22ae8ce8 | 637 | * from @input and get and return the matching bdev. *@inputp is |
015d254c TH |
638 | * updated to point past the device node prefix. Returns an ERR_PTR() |
639 | * value on error. | |
640 | * | |
641 | * Use this function iff blkg_conf_prep() can't be used for some reason. | |
642 | */ | |
22ae8ce8 | 643 | struct block_device *blkcg_conf_open_bdev(char **inputp) |
015d254c TH |
644 | { |
645 | char *input = *inputp; | |
646 | unsigned int major, minor; | |
22ae8ce8 CH |
647 | struct block_device *bdev; |
648 | int key_len; | |
015d254c TH |
649 | |
650 | if (sscanf(input, "%u:%u%n", &major, &minor, &key_len) != 2) | |
651 | return ERR_PTR(-EINVAL); | |
652 | ||
653 | input += key_len; | |
654 | if (!isspace(*input)) | |
655 | return ERR_PTR(-EINVAL); | |
656 | input = skip_spaces(input); | |
657 | ||
22ae8ce8 CH |
658 | bdev = blkdev_get_no_open(MKDEV(major, minor)); |
659 | if (!bdev) | |
015d254c | 660 | return ERR_PTR(-ENODEV); |
22ae8ce8 CH |
661 | if (bdev_is_partition(bdev)) { |
662 | blkdev_put_no_open(bdev); | |
015d254c TH |
663 | return ERR_PTR(-ENODEV); |
664 | } | |
665 | ||
666 | *inputp = input; | |
22ae8ce8 | 667 | return bdev; |
015d254c TH |
668 | } |
669 | ||
3a8b31d3 TH |
670 | /** |
671 | * blkg_conf_prep - parse and prepare for per-blkg config update | |
672 | * @blkcg: target block cgroup | |
da8b0662 | 673 | * @pol: target policy |
3a8b31d3 TH |
674 | * @input: input string |
675 | * @ctx: blkg_conf_ctx to be filled | |
676 | * | |
677 | * Parse per-blkg config update from @input and initialize @ctx with the | |
36aa9e5f TH |
678 | * result. @ctx->blkg points to the blkg to be updated and @ctx->body the |
679 | * part of @input following MAJ:MIN. This function returns with RCU read | |
680 | * lock and queue lock held and must be paired with blkg_conf_finish(). | |
3a8b31d3 | 681 | */ |
3c798398 | 682 | int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol, |
36aa9e5f | 683 | char *input, struct blkg_conf_ctx *ctx) |
ed6cddef | 684 | __acquires(rcu) __acquires(&bdev->bd_queue->queue_lock) |
34d0f179 | 685 | { |
22ae8ce8 | 686 | struct block_device *bdev; |
457e490f | 687 | struct request_queue *q; |
3c798398 | 688 | struct blkcg_gq *blkg; |
015d254c | 689 | int ret; |
36aa9e5f | 690 | |
22ae8ce8 CH |
691 | bdev = blkcg_conf_open_bdev(&input); |
692 | if (IS_ERR(bdev)) | |
693 | return PTR_ERR(bdev); | |
e56da7e2 | 694 | |
ed6cddef | 695 | q = bdev_get_queue(bdev); |
da8b0662 | 696 | |
0c9d338c YK |
697 | /* |
698 | * blkcg_deactivate_policy() requires queue to be frozen, we can grab | |
699 | * q_usage_counter to prevent concurrent with blkcg_deactivate_policy(). | |
700 | */ | |
701 | ret = blk_queue_enter(q, 0); | |
702 | if (ret) | |
15c30104 | 703 | goto fail; |
0c9d338c | 704 | |
457e490f | 705 | rcu_read_lock(); |
0d945c1f | 706 | spin_lock_irq(&q->queue_lock); |
e56da7e2 | 707 | |
457e490f | 708 | blkg = blkg_lookup_check(blkcg, pol, q); |
4bfd482e TH |
709 | if (IS_ERR(blkg)) { |
710 | ret = PTR_ERR(blkg); | |
457e490f TE |
711 | goto fail_unlock; |
712 | } | |
713 | ||
714 | if (blkg) | |
715 | goto success; | |
716 | ||
717 | /* | |
718 | * Create blkgs walking down from blkcg_root to @blkcg, so that all | |
719 | * non-root blkgs have access to their parents. | |
720 | */ | |
721 | while (true) { | |
722 | struct blkcg *pos = blkcg; | |
723 | struct blkcg *parent; | |
724 | struct blkcg_gq *new_blkg; | |
725 | ||
726 | parent = blkcg_parent(blkcg); | |
727 | while (parent && !__blkg_lookup(parent, q, false)) { | |
728 | pos = parent; | |
729 | parent = blkcg_parent(parent); | |
730 | } | |
731 | ||
732 | /* Drop locks to do new blkg allocation with GFP_KERNEL. */ | |
0d945c1f | 733 | spin_unlock_irq(&q->queue_lock); |
3a8b31d3 | 734 | rcu_read_unlock(); |
457e490f TE |
735 | |
736 | new_blkg = blkg_alloc(pos, q, GFP_KERNEL); | |
737 | if (unlikely(!new_blkg)) { | |
738 | ret = -ENOMEM; | |
15c30104 | 739 | goto fail_exit_queue; |
7702e8f4 | 740 | } |
3a8b31d3 | 741 | |
f255c19b GKB |
742 | if (radix_tree_preload(GFP_KERNEL)) { |
743 | blkg_free(new_blkg); | |
744 | ret = -ENOMEM; | |
15c30104 | 745 | goto fail_exit_queue; |
f255c19b GKB |
746 | } |
747 | ||
457e490f | 748 | rcu_read_lock(); |
0d945c1f | 749 | spin_lock_irq(&q->queue_lock); |
457e490f TE |
750 | |
751 | blkg = blkg_lookup_check(pos, pol, q); | |
752 | if (IS_ERR(blkg)) { | |
753 | ret = PTR_ERR(blkg); | |
52abfcbd | 754 | blkg_free(new_blkg); |
f255c19b | 755 | goto fail_preloaded; |
457e490f TE |
756 | } |
757 | ||
758 | if (blkg) { | |
759 | blkg_free(new_blkg); | |
760 | } else { | |
761 | blkg = blkg_create(pos, q, new_blkg); | |
98d669b4 | 762 | if (IS_ERR(blkg)) { |
457e490f | 763 | ret = PTR_ERR(blkg); |
f255c19b | 764 | goto fail_preloaded; |
457e490f TE |
765 | } |
766 | } | |
767 | ||
f255c19b GKB |
768 | radix_tree_preload_end(); |
769 | ||
457e490f TE |
770 | if (pos == blkcg) |
771 | goto success; | |
772 | } | |
773 | success: | |
0c9d338c | 774 | blk_queue_exit(q); |
22ae8ce8 | 775 | ctx->bdev = bdev; |
3a8b31d3 | 776 | ctx->blkg = blkg; |
015d254c | 777 | ctx->body = input; |
726fa694 | 778 | return 0; |
457e490f | 779 | |
f255c19b GKB |
780 | fail_preloaded: |
781 | radix_tree_preload_end(); | |
457e490f | 782 | fail_unlock: |
0d945c1f | 783 | spin_unlock_irq(&q->queue_lock); |
457e490f | 784 | rcu_read_unlock(); |
15c30104 YK |
785 | fail_exit_queue: |
786 | blk_queue_exit(q); | |
457e490f | 787 | fail: |
22ae8ce8 | 788 | blkdev_put_no_open(bdev); |
457e490f TE |
789 | /* |
790 | * If queue was bypassing, we should retry. Do so after a | |
791 | * short msleep(). It isn't strictly necessary but queue | |
792 | * can be bypassing for some time and it's always nice to | |
793 | * avoid busy looping. | |
794 | */ | |
795 | if (ret == -EBUSY) { | |
796 | msleep(10); | |
797 | ret = restart_syscall(); | |
798 | } | |
799 | return ret; | |
34d0f179 | 800 | } |
89f3b6d6 | 801 | EXPORT_SYMBOL_GPL(blkg_conf_prep); |
34d0f179 | 802 | |
3a8b31d3 TH |
803 | /** |
804 | * blkg_conf_finish - finish up per-blkg config update | |
805 | * @ctx: blkg_conf_ctx intiailized by blkg_conf_prep() | |
806 | * | |
807 | * Finish up after per-blkg config update. This function must be paired | |
808 | * with blkg_conf_prep(). | |
809 | */ | |
829fdb50 | 810 | void blkg_conf_finish(struct blkg_conf_ctx *ctx) |
ed6cddef | 811 | __releases(&ctx->bdev->bd_queue->queue_lock) __releases(rcu) |
34d0f179 | 812 | { |
ed6cddef | 813 | spin_unlock_irq(&bdev_get_queue(ctx->bdev)->queue_lock); |
3a8b31d3 | 814 | rcu_read_unlock(); |
22ae8ce8 | 815 | blkdev_put_no_open(ctx->bdev); |
34d0f179 | 816 | } |
89f3b6d6 | 817 | EXPORT_SYMBOL_GPL(blkg_conf_finish); |
34d0f179 | 818 | |
cd1fc4b9 BB |
819 | static void blkg_iostat_set(struct blkg_iostat *dst, struct blkg_iostat *src) |
820 | { | |
821 | int i; | |
822 | ||
823 | for (i = 0; i < BLKG_IOSTAT_NR; i++) { | |
824 | dst->bytes[i] = src->bytes[i]; | |
825 | dst->ios[i] = src->ios[i]; | |
826 | } | |
827 | } | |
828 | ||
829 | static void blkg_iostat_add(struct blkg_iostat *dst, struct blkg_iostat *src) | |
830 | { | |
831 | int i; | |
832 | ||
833 | for (i = 0; i < BLKG_IOSTAT_NR; i++) { | |
834 | dst->bytes[i] += src->bytes[i]; | |
835 | dst->ios[i] += src->ios[i]; | |
836 | } | |
837 | } | |
838 | ||
839 | static void blkg_iostat_sub(struct blkg_iostat *dst, struct blkg_iostat *src) | |
840 | { | |
841 | int i; | |
842 | ||
843 | for (i = 0; i < BLKG_IOSTAT_NR; i++) { | |
844 | dst->bytes[i] -= src->bytes[i]; | |
845 | dst->ios[i] -= src->ios[i]; | |
846 | } | |
847 | } | |
848 | ||
362b8c16 JY |
849 | static void blkcg_iostat_update(struct blkcg_gq *blkg, struct blkg_iostat *cur, |
850 | struct blkg_iostat *last) | |
851 | { | |
852 | struct blkg_iostat delta; | |
853 | unsigned long flags; | |
854 | ||
855 | /* propagate percpu delta to global */ | |
856 | flags = u64_stats_update_begin_irqsave(&blkg->iostat.sync); | |
857 | blkg_iostat_set(&delta, cur); | |
858 | blkg_iostat_sub(&delta, last); | |
859 | blkg_iostat_add(&blkg->iostat.cur, &delta); | |
860 | blkg_iostat_add(last, &delta); | |
861 | u64_stats_update_end_irqrestore(&blkg->iostat.sync, flags); | |
862 | } | |
863 | ||
cd1fc4b9 BB |
864 | static void blkcg_rstat_flush(struct cgroup_subsys_state *css, int cpu) |
865 | { | |
866 | struct blkcg *blkcg = css_to_blkcg(css); | |
867 | struct blkcg_gq *blkg; | |
868 | ||
dc26532a JW |
869 | /* Root-level stats are sourced from system-wide IO stats */ |
870 | if (!cgroup_parent(css->cgroup)) | |
871 | return; | |
872 | ||
cd1fc4b9 BB |
873 | rcu_read_lock(); |
874 | ||
875 | hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) { | |
876 | struct blkcg_gq *parent = blkg->parent; | |
877 | struct blkg_iostat_set *bisc = per_cpu_ptr(blkg->iostat_cpu, cpu); | |
362b8c16 | 878 | struct blkg_iostat cur; |
cd1fc4b9 BB |
879 | unsigned int seq; |
880 | ||
881 | /* fetch the current per-cpu values */ | |
882 | do { | |
883 | seq = u64_stats_fetch_begin(&bisc->sync); | |
884 | blkg_iostat_set(&cur, &bisc->cur); | |
885 | } while (u64_stats_fetch_retry(&bisc->sync, seq)); | |
886 | ||
362b8c16 | 887 | blkcg_iostat_update(blkg, &cur, &bisc->last); |
cd1fc4b9 | 888 | |
dc26532a | 889 | /* propagate global delta to parent (unless that's root) */ |
362b8c16 JY |
890 | if (parent && parent->parent) |
891 | blkcg_iostat_update(parent, &blkg->iostat.cur, | |
892 | &blkg->iostat.last); | |
cd1fc4b9 BB |
893 | } |
894 | ||
895 | rcu_read_unlock(); | |
896 | } | |
897 | ||
ef45fe47 | 898 | /* |
dc26532a JW |
899 | * We source root cgroup stats from the system-wide stats to avoid |
900 | * tracking the same information twice and incurring overhead when no | |
901 | * cgroups are defined. For that reason, cgroup_rstat_flush in | |
902 | * blkcg_print_stat does not actually fill out the iostat in the root | |
903 | * cgroup's blkcg_gq. | |
ef45fe47 BB |
904 | * |
905 | * However, we would like to re-use the printing code between the root and | |
906 | * non-root cgroups to the extent possible. For that reason, we simulate | |
907 | * flushing the root cgroup's stats by explicitly filling in the iostat | |
908 | * with disk level statistics. | |
909 | */ | |
910 | static void blkcg_fill_root_iostats(void) | |
911 | { | |
912 | struct class_dev_iter iter; | |
913 | struct device *dev; | |
914 | ||
915 | class_dev_iter_init(&iter, &block_class, NULL, &disk_type); | |
916 | while ((dev = class_dev_iter_next(&iter))) { | |
0d02129e CH |
917 | struct block_device *bdev = dev_to_bdev(dev); |
918 | struct blkcg_gq *blkg = | |
ed6cddef | 919 | blk_queue_root_blkg(bdev_get_queue(bdev)); |
ef45fe47 BB |
920 | struct blkg_iostat tmp; |
921 | int cpu; | |
f122d103 | 922 | unsigned long flags; |
ef45fe47 BB |
923 | |
924 | memset(&tmp, 0, sizeof(tmp)); | |
925 | for_each_possible_cpu(cpu) { | |
926 | struct disk_stats *cpu_dkstats; | |
927 | ||
0d02129e | 928 | cpu_dkstats = per_cpu_ptr(bdev->bd_stats, cpu); |
ef45fe47 BB |
929 | tmp.ios[BLKG_IOSTAT_READ] += |
930 | cpu_dkstats->ios[STAT_READ]; | |
931 | tmp.ios[BLKG_IOSTAT_WRITE] += | |
932 | cpu_dkstats->ios[STAT_WRITE]; | |
933 | tmp.ios[BLKG_IOSTAT_DISCARD] += | |
934 | cpu_dkstats->ios[STAT_DISCARD]; | |
935 | // convert sectors to bytes | |
936 | tmp.bytes[BLKG_IOSTAT_READ] += | |
937 | cpu_dkstats->sectors[STAT_READ] << 9; | |
938 | tmp.bytes[BLKG_IOSTAT_WRITE] += | |
939 | cpu_dkstats->sectors[STAT_WRITE] << 9; | |
940 | tmp.bytes[BLKG_IOSTAT_DISCARD] += | |
941 | cpu_dkstats->sectors[STAT_DISCARD] << 9; | |
ef45fe47 | 942 | } |
f122d103 CZ |
943 | |
944 | flags = u64_stats_update_begin_irqsave(&blkg->iostat.sync); | |
945 | blkg_iostat_set(&blkg->iostat.cur, &tmp); | |
946 | u64_stats_update_end_irqrestore(&blkg->iostat.sync, flags); | |
ef45fe47 BB |
947 | } |
948 | } | |
949 | ||
49cb5168 | 950 | static void blkcg_print_one_stat(struct blkcg_gq *blkg, struct seq_file *s) |
2ee867dc | 951 | { |
49cb5168 CH |
952 | struct blkg_iostat_set *bis = &blkg->iostat; |
953 | u64 rbytes, wbytes, rios, wios, dbytes, dios; | |
49cb5168 CH |
954 | const char *dname; |
955 | unsigned seq; | |
49cb5168 | 956 | int i; |
ef45fe47 | 957 | |
49cb5168 CH |
958 | if (!blkg->online) |
959 | return; | |
2ee867dc | 960 | |
49cb5168 CH |
961 | dname = blkg_dev_name(blkg); |
962 | if (!dname) | |
963 | return; | |
2ee867dc | 964 | |
252c651a | 965 | seq_printf(s, "%s ", dname); |
b0814361 | 966 | |
49cb5168 CH |
967 | do { |
968 | seq = u64_stats_fetch_begin(&bis->sync); | |
b0814361 | 969 | |
49cb5168 CH |
970 | rbytes = bis->cur.bytes[BLKG_IOSTAT_READ]; |
971 | wbytes = bis->cur.bytes[BLKG_IOSTAT_WRITE]; | |
972 | dbytes = bis->cur.bytes[BLKG_IOSTAT_DISCARD]; | |
973 | rios = bis->cur.ios[BLKG_IOSTAT_READ]; | |
974 | wios = bis->cur.ios[BLKG_IOSTAT_WRITE]; | |
975 | dios = bis->cur.ios[BLKG_IOSTAT_DISCARD]; | |
976 | } while (u64_stats_fetch_retry(&bis->sync, seq)); | |
2ee867dc | 977 | |
49cb5168 | 978 | if (rbytes || wbytes || rios || wios) { |
252c651a | 979 | seq_printf(s, "rbytes=%llu wbytes=%llu rios=%llu wios=%llu dbytes=%llu dios=%llu", |
49cb5168 CH |
980 | rbytes, wbytes, rios, wios, |
981 | dbytes, dios); | |
982 | } | |
903d23f0 | 983 | |
49cb5168 | 984 | if (blkcg_debug_stats && atomic_read(&blkg->use_delay)) { |
252c651a | 985 | seq_printf(s, " use_delay=%d delay_nsec=%llu", |
49cb5168 CH |
986 | atomic_read(&blkg->use_delay), |
987 | atomic64_read(&blkg->delay_nsec)); | |
988 | } | |
2ee867dc | 989 | |
49cb5168 CH |
990 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
991 | struct blkcg_policy *pol = blkcg_policy[i]; | |
2ee867dc | 992 | |
49cb5168 CH |
993 | if (!blkg->pd[i] || !pol->pd_stat_fn) |
994 | continue; | |
903d23f0 | 995 | |
3607849d | 996 | pol->pd_stat_fn(blkg->pd[i], s); |
49cb5168 | 997 | } |
d09d8df3 | 998 | |
3607849d | 999 | seq_puts(s, "\n"); |
49cb5168 | 1000 | } |
903d23f0 | 1001 | |
49cb5168 CH |
1002 | static int blkcg_print_stat(struct seq_file *sf, void *v) |
1003 | { | |
1004 | struct blkcg *blkcg = css_to_blkcg(seq_css(sf)); | |
1005 | struct blkcg_gq *blkg; | |
903d23f0 | 1006 | |
49cb5168 CH |
1007 | if (!seq_css(sf)->parent) |
1008 | blkcg_fill_root_iostats(); | |
1009 | else | |
1010 | cgroup_rstat_flush(blkcg->css.cgroup); | |
07b0fdec | 1011 | |
49cb5168 CH |
1012 | rcu_read_lock(); |
1013 | hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) { | |
1014 | spin_lock_irq(&blkg->q->queue_lock); | |
1015 | blkcg_print_one_stat(blkg, sf); | |
b0814361 | 1016 | spin_unlock_irq(&blkg->q->queue_lock); |
2ee867dc | 1017 | } |
2ee867dc TH |
1018 | rcu_read_unlock(); |
1019 | return 0; | |
1020 | } | |
1021 | ||
e1f3b941 | 1022 | static struct cftype blkcg_files[] = { |
2ee867dc TH |
1023 | { |
1024 | .name = "stat", | |
1025 | .seq_show = blkcg_print_stat, | |
1026 | }, | |
1027 | { } /* terminate */ | |
1028 | }; | |
1029 | ||
e1f3b941 | 1030 | static struct cftype blkcg_legacy_files[] = { |
84c124da DS |
1031 | { |
1032 | .name = "reset_stats", | |
3c798398 | 1033 | .write_u64 = blkcg_reset_stats, |
22084190 | 1034 | }, |
4baf6e33 | 1035 | { } /* terminate */ |
31e4c28d VG |
1036 | }; |
1037 | ||
dec223c9 CH |
1038 | #ifdef CONFIG_CGROUP_WRITEBACK |
1039 | struct list_head *blkcg_get_cgwb_list(struct cgroup_subsys_state *css) | |
1040 | { | |
1041 | return &css_to_blkcg(css)->cgwb_list; | |
1042 | } | |
1043 | #endif | |
1044 | ||
59b57717 DZF |
1045 | /* |
1046 | * blkcg destruction is a three-stage process. | |
1047 | * | |
1048 | * 1. Destruction starts. The blkcg_css_offline() callback is invoked | |
1049 | * which offlines writeback. Here we tie the next stage of blkg destruction | |
1050 | * to the completion of writeback associated with the blkcg. This lets us | |
1051 | * avoid punting potentially large amounts of outstanding writeback to root | |
1052 | * while maintaining any ongoing policies. The next stage is triggered when | |
1053 | * the nr_cgwbs count goes to zero. | |
1054 | * | |
1055 | * 2. When the nr_cgwbs count goes to zero, blkcg_destroy_blkgs() is called | |
1056 | * and handles the destruction of blkgs. Here the css reference held by | |
1057 | * the blkg is put back eventually allowing blkcg_css_free() to be called. | |
1058 | * This work may occur in cgwb_release_workfn() on the cgwb_release | |
1059 | * workqueue. Any submitted ios that fail to get the blkg ref will be | |
1060 | * punted to the root_blkg. | |
1061 | * | |
1062 | * 3. Once the blkcg ref count goes to zero, blkcg_css_free() is called. | |
1063 | * This finally frees the blkcg. | |
1064 | */ | |
1065 | ||
59b57717 DZF |
1066 | /** |
1067 | * blkcg_destroy_blkgs - responsible for shooting down blkgs | |
1068 | * @blkcg: blkcg of interest | |
1069 | * | |
1070 | * blkgs should be removed while holding both q and blkcg locks. As blkcg lock | |
1071 | * is nested inside q lock, this function performs reverse double lock dancing. | |
1072 | * Destroying the blkgs releases the reference held on the blkcg's css allowing | |
1073 | * blkcg_css_free to eventually be called. | |
1074 | * | |
1075 | * This is the blkcg counterpart of ioc_release_fn(). | |
1076 | */ | |
397c9f46 | 1077 | static void blkcg_destroy_blkgs(struct blkcg *blkcg) |
59b57717 | 1078 | { |
6c635cae BW |
1079 | might_sleep(); |
1080 | ||
9f13ef67 | 1081 | spin_lock_irq(&blkcg->lock); |
7ee9c562 | 1082 | |
4c699480 JQ |
1083 | while (!hlist_empty(&blkcg->blkg_list)) { |
1084 | struct blkcg_gq *blkg = hlist_entry(blkcg->blkg_list.first, | |
6b065462 | 1085 | struct blkcg_gq, blkcg_node); |
4c699480 JQ |
1086 | struct request_queue *q = blkg->q; |
1087 | ||
6c635cae BW |
1088 | if (need_resched() || !spin_trylock(&q->queue_lock)) { |
1089 | /* | |
1090 | * Given that the system can accumulate a huge number | |
1091 | * of blkgs in pathological cases, check to see if we | |
1092 | * need to rescheduling to avoid softlockup. | |
1093 | */ | |
4c699480 | 1094 | spin_unlock_irq(&blkcg->lock); |
6c635cae | 1095 | cond_resched(); |
4c699480 | 1096 | spin_lock_irq(&blkcg->lock); |
6c635cae | 1097 | continue; |
4c699480 | 1098 | } |
6c635cae BW |
1099 | |
1100 | blkg_destroy(blkg); | |
1101 | spin_unlock(&q->queue_lock); | |
4c699480 | 1102 | } |
6b065462 | 1103 | |
4c699480 JQ |
1104 | spin_unlock_irq(&blkcg->lock); |
1105 | } | |
1106 | ||
397c9f46 CH |
1107 | /** |
1108 | * blkcg_pin_online - pin online state | |
1109 | * @blkcg_css: blkcg of interest | |
1110 | * | |
1111 | * While pinned, a blkcg is kept online. This is primarily used to | |
1112 | * impedance-match blkg and cgwb lifetimes so that blkg doesn't go offline | |
1113 | * while an associated cgwb is still active. | |
1114 | */ | |
1115 | void blkcg_pin_online(struct cgroup_subsys_state *blkcg_css) | |
1116 | { | |
1117 | refcount_inc(&css_to_blkcg(blkcg_css)->online_pin); | |
1118 | } | |
1119 | ||
1120 | /** | |
1121 | * blkcg_unpin_online - unpin online state | |
1122 | * @blkcg_css: blkcg of interest | |
1123 | * | |
1124 | * This is primarily used to impedance-match blkg and cgwb lifetimes so | |
1125 | * that blkg doesn't go offline while an associated cgwb is still active. | |
1126 | * When this count goes to zero, all active cgwbs have finished so the | |
1127 | * blkcg can continue destruction by calling blkcg_destroy_blkgs(). | |
1128 | */ | |
1129 | void blkcg_unpin_online(struct cgroup_subsys_state *blkcg_css) | |
1130 | { | |
1131 | struct blkcg *blkcg = css_to_blkcg(blkcg_css); | |
1132 | ||
1133 | do { | |
1134 | if (!refcount_dec_and_test(&blkcg->online_pin)) | |
1135 | break; | |
1136 | blkcg_destroy_blkgs(blkcg); | |
1137 | blkcg = blkcg_parent(blkcg); | |
1138 | } while (blkcg); | |
1139 | } | |
1140 | ||
1141 | /** | |
1142 | * blkcg_css_offline - cgroup css_offline callback | |
1143 | * @css: css of interest | |
1144 | * | |
1145 | * This function is called when @css is about to go away. Here the cgwbs are | |
1146 | * offlined first and only once writeback associated with the blkcg has | |
1147 | * finished do we start step 2 (see above). | |
1148 | */ | |
1149 | static void blkcg_css_offline(struct cgroup_subsys_state *css) | |
1150 | { | |
1151 | /* this prevents anyone from attaching or migrating to this blkcg */ | |
dec223c9 | 1152 | wb_blkcg_offline(css); |
397c9f46 CH |
1153 | |
1154 | /* put the base online pin allowing step 2 to be triggered */ | |
1155 | blkcg_unpin_online(css); | |
1156 | } | |
1157 | ||
eb95419b | 1158 | static void blkcg_css_free(struct cgroup_subsys_state *css) |
7ee9c562 | 1159 | { |
eb95419b | 1160 | struct blkcg *blkcg = css_to_blkcg(css); |
bc915e61 | 1161 | int i; |
7ee9c562 | 1162 | |
7876f930 | 1163 | mutex_lock(&blkcg_pol_mutex); |
e4a9bde9 | 1164 | |
7876f930 | 1165 | list_del(&blkcg->all_blkcgs_node); |
7876f930 | 1166 | |
bc915e61 | 1167 | for (i = 0; i < BLKCG_MAX_POLS; i++) |
e4a9bde9 TH |
1168 | if (blkcg->cpd[i]) |
1169 | blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]); | |
1170 | ||
1171 | mutex_unlock(&blkcg_pol_mutex); | |
1172 | ||
bc915e61 | 1173 | kfree(blkcg); |
31e4c28d VG |
1174 | } |
1175 | ||
eb95419b TH |
1176 | static struct cgroup_subsys_state * |
1177 | blkcg_css_alloc(struct cgroup_subsys_state *parent_css) | |
31e4c28d | 1178 | { |
3c798398 | 1179 | struct blkcg *blkcg; |
e48453c3 AA |
1180 | struct cgroup_subsys_state *ret; |
1181 | int i; | |
31e4c28d | 1182 | |
7876f930 TH |
1183 | mutex_lock(&blkcg_pol_mutex); |
1184 | ||
eb95419b | 1185 | if (!parent_css) { |
3c798398 | 1186 | blkcg = &blkcg_root; |
bc915e61 TH |
1187 | } else { |
1188 | blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL); | |
1189 | if (!blkcg) { | |
1190 | ret = ERR_PTR(-ENOMEM); | |
4c18c9e9 | 1191 | goto unlock; |
bc915e61 | 1192 | } |
e48453c3 AA |
1193 | } |
1194 | ||
1195 | for (i = 0; i < BLKCG_MAX_POLS ; i++) { | |
1196 | struct blkcg_policy *pol = blkcg_policy[i]; | |
1197 | struct blkcg_policy_data *cpd; | |
1198 | ||
1199 | /* | |
1200 | * If the policy hasn't been attached yet, wait for it | |
1201 | * to be attached before doing anything else. Otherwise, | |
1202 | * check if the policy requires any specific per-cgroup | |
1203 | * data: if it does, allocate and initialize it. | |
1204 | */ | |
e4a9bde9 | 1205 | if (!pol || !pol->cpd_alloc_fn) |
e48453c3 AA |
1206 | continue; |
1207 | ||
e4a9bde9 | 1208 | cpd = pol->cpd_alloc_fn(GFP_KERNEL); |
e48453c3 AA |
1209 | if (!cpd) { |
1210 | ret = ERR_PTR(-ENOMEM); | |
1211 | goto free_pd_blkcg; | |
1212 | } | |
81437648 TH |
1213 | blkcg->cpd[i] = cpd; |
1214 | cpd->blkcg = blkcg; | |
e48453c3 | 1215 | cpd->plid = i; |
e4a9bde9 TH |
1216 | if (pol->cpd_init_fn) |
1217 | pol->cpd_init_fn(cpd); | |
e48453c3 | 1218 | } |
31e4c28d | 1219 | |
31e4c28d | 1220 | spin_lock_init(&blkcg->lock); |
d866dbf6 | 1221 | refcount_set(&blkcg->online_pin, 1); |
e00f4f4d | 1222 | INIT_RADIX_TREE(&blkcg->blkg_tree, GFP_NOWAIT | __GFP_NOWARN); |
31e4c28d | 1223 | INIT_HLIST_HEAD(&blkcg->blkg_list); |
52ebea74 TH |
1224 | #ifdef CONFIG_CGROUP_WRITEBACK |
1225 | INIT_LIST_HEAD(&blkcg->cgwb_list); | |
1226 | #endif | |
7876f930 TH |
1227 | list_add_tail(&blkcg->all_blkcgs_node, &all_blkcgs); |
1228 | ||
1229 | mutex_unlock(&blkcg_pol_mutex); | |
31e4c28d | 1230 | return &blkcg->css; |
e48453c3 AA |
1231 | |
1232 | free_pd_blkcg: | |
1233 | for (i--; i >= 0; i--) | |
e4a9bde9 TH |
1234 | if (blkcg->cpd[i]) |
1235 | blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]); | |
4c18c9e9 | 1236 | |
1237 | if (blkcg != &blkcg_root) | |
1238 | kfree(blkcg); | |
1239 | unlock: | |
7876f930 | 1240 | mutex_unlock(&blkcg_pol_mutex); |
e48453c3 | 1241 | return ret; |
31e4c28d VG |
1242 | } |
1243 | ||
4308a434 TH |
1244 | static int blkcg_css_online(struct cgroup_subsys_state *css) |
1245 | { | |
397c9f46 | 1246 | struct blkcg *parent = blkcg_parent(css_to_blkcg(css)); |
4308a434 TH |
1247 | |
1248 | /* | |
1249 | * blkcg_pin_online() is used to delay blkcg offline so that blkgs | |
1250 | * don't go offline while cgwbs are still active on them. Pin the | |
1251 | * parent so that offline always happens towards the root. | |
1252 | */ | |
1253 | if (parent) | |
397c9f46 | 1254 | blkcg_pin_online(css); |
4308a434 TH |
1255 | return 0; |
1256 | } | |
1257 | ||
5efd6113 TH |
1258 | /** |
1259 | * blkcg_init_queue - initialize blkcg part of request queue | |
1260 | * @q: request_queue to initialize | |
1261 | * | |
c62b37d9 | 1262 | * Called from blk_alloc_queue(). Responsible for initializing blkcg |
5efd6113 TH |
1263 | * part of new request_queue @q. |
1264 | * | |
1265 | * RETURNS: | |
1266 | * 0 on success, -errno on failure. | |
1267 | */ | |
1268 | int blkcg_init_queue(struct request_queue *q) | |
1269 | { | |
d708f0d5 JA |
1270 | struct blkcg_gq *new_blkg, *blkg; |
1271 | bool preloaded; | |
ec13b1d6 TH |
1272 | int ret; |
1273 | ||
472e4314 ML |
1274 | INIT_LIST_HEAD(&q->blkg_list); |
1275 | ||
d708f0d5 JA |
1276 | new_blkg = blkg_alloc(&blkcg_root, q, GFP_KERNEL); |
1277 | if (!new_blkg) | |
1278 | return -ENOMEM; | |
1279 | ||
1280 | preloaded = !radix_tree_preload(GFP_KERNEL); | |
1281 | ||
bea54883 | 1282 | /* Make sure the root blkg exists. */ |
77c570a1 | 1283 | /* spin_lock_irq can serve as RCU read-side critical section. */ |
0d945c1f | 1284 | spin_lock_irq(&q->queue_lock); |
d708f0d5 | 1285 | blkg = blkg_create(&blkcg_root, q, new_blkg); |
901932a3 JB |
1286 | if (IS_ERR(blkg)) |
1287 | goto err_unlock; | |
1288 | q->root_blkg = blkg; | |
0d945c1f | 1289 | spin_unlock_irq(&q->queue_lock); |
ec13b1d6 | 1290 | |
d708f0d5 JA |
1291 | if (preloaded) |
1292 | radix_tree_preload_end(); | |
1293 | ||
556910e3 BVA |
1294 | ret = blk_ioprio_init(q); |
1295 | if (ret) | |
1296 | goto err_destroy_all; | |
1297 | ||
19688d7f BVA |
1298 | ret = blk_throtl_init(q); |
1299 | if (ret) | |
04be60b5 | 1300 | goto err_destroy_all; |
19688d7f | 1301 | |
6f5ddde4 YX |
1302 | ret = blk_iolatency_init(q); |
1303 | if (ret) { | |
1304 | blk_throtl_exit(q); | |
82b74cac | 1305 | blk_ioprio_exit(q); |
6f5ddde4 YX |
1306 | goto err_destroy_all; |
1307 | } | |
1308 | ||
04be60b5 | 1309 | return 0; |
901932a3 | 1310 | |
04be60b5 | 1311 | err_destroy_all: |
04be60b5 | 1312 | blkg_destroy_all(q); |
04be60b5 | 1313 | return ret; |
901932a3 | 1314 | err_unlock: |
0d945c1f | 1315 | spin_unlock_irq(&q->queue_lock); |
901932a3 JB |
1316 | if (preloaded) |
1317 | radix_tree_preload_end(); | |
1318 | return PTR_ERR(blkg); | |
5efd6113 TH |
1319 | } |
1320 | ||
5efd6113 TH |
1321 | /** |
1322 | * blkcg_exit_queue - exit and release blkcg part of request_queue | |
1323 | * @q: request_queue being released | |
1324 | * | |
7585d508 | 1325 | * Called from blk_exit_queue(). Responsible for exiting blkcg part. |
5efd6113 TH |
1326 | */ |
1327 | void blkcg_exit_queue(struct request_queue *q) | |
1328 | { | |
3c96cb32 | 1329 | blkg_destroy_all(q); |
5efd6113 TH |
1330 | blk_throtl_exit(q); |
1331 | } | |
1332 | ||
69d7fde5 TH |
1333 | static void blkcg_bind(struct cgroup_subsys_state *root_css) |
1334 | { | |
1335 | int i; | |
1336 | ||
1337 | mutex_lock(&blkcg_pol_mutex); | |
1338 | ||
1339 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
1340 | struct blkcg_policy *pol = blkcg_policy[i]; | |
1341 | struct blkcg *blkcg; | |
1342 | ||
1343 | if (!pol || !pol->cpd_bind_fn) | |
1344 | continue; | |
1345 | ||
1346 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) | |
1347 | if (blkcg->cpd[pol->plid]) | |
1348 | pol->cpd_bind_fn(blkcg->cpd[pol->plid]); | |
1349 | } | |
1350 | mutex_unlock(&blkcg_pol_mutex); | |
1351 | } | |
1352 | ||
d09d8df3 JB |
1353 | static void blkcg_exit(struct task_struct *tsk) |
1354 | { | |
1355 | if (tsk->throttle_queue) | |
1356 | blk_put_queue(tsk->throttle_queue); | |
1357 | tsk->throttle_queue = NULL; | |
1358 | } | |
1359 | ||
c165b3e3 | 1360 | struct cgroup_subsys io_cgrp_subsys = { |
92fb9748 | 1361 | .css_alloc = blkcg_css_alloc, |
4308a434 | 1362 | .css_online = blkcg_css_online, |
92fb9748 TH |
1363 | .css_offline = blkcg_css_offline, |
1364 | .css_free = blkcg_css_free, | |
f7331648 | 1365 | .css_rstat_flush = blkcg_rstat_flush, |
69d7fde5 | 1366 | .bind = blkcg_bind, |
2ee867dc | 1367 | .dfl_cftypes = blkcg_files, |
880f50e2 | 1368 | .legacy_cftypes = blkcg_legacy_files, |
c165b3e3 | 1369 | .legacy_name = "blkio", |
d09d8df3 | 1370 | .exit = blkcg_exit, |
1ced953b TH |
1371 | #ifdef CONFIG_MEMCG |
1372 | /* | |
1373 | * This ensures that, if available, memcg is automatically enabled | |
1374 | * together on the default hierarchy so that the owner cgroup can | |
1375 | * be retrieved from writeback pages. | |
1376 | */ | |
1377 | .depends_on = 1 << memory_cgrp_id, | |
1378 | #endif | |
676f7c8f | 1379 | }; |
c165b3e3 | 1380 | EXPORT_SYMBOL_GPL(io_cgrp_subsys); |
676f7c8f | 1381 | |
a2b1693b TH |
1382 | /** |
1383 | * blkcg_activate_policy - activate a blkcg policy on a request_queue | |
1384 | * @q: request_queue of interest | |
1385 | * @pol: blkcg policy to activate | |
1386 | * | |
1387 | * Activate @pol on @q. Requires %GFP_KERNEL context. @q goes through | |
1388 | * bypass mode to populate its blkgs with policy_data for @pol. | |
1389 | * | |
1390 | * Activation happens with @q bypassed, so nobody would be accessing blkgs | |
1391 | * from IO path. Update of each blkg is protected by both queue and blkcg | |
1392 | * locks so that holding either lock and testing blkcg_policy_enabled() is | |
1393 | * always enough for dereferencing policy data. | |
1394 | * | |
1395 | * The caller is responsible for synchronizing [de]activations and policy | |
1396 | * [un]registerations. Returns 0 on success, -errno on failure. | |
1397 | */ | |
1398 | int blkcg_activate_policy(struct request_queue *q, | |
3c798398 | 1399 | const struct blkcg_policy *pol) |
a2b1693b | 1400 | { |
4c55f4f9 | 1401 | struct blkg_policy_data *pd_prealloc = NULL; |
9d179b86 | 1402 | struct blkcg_gq *blkg, *pinned_blkg = NULL; |
4c55f4f9 | 1403 | int ret; |
a2b1693b TH |
1404 | |
1405 | if (blkcg_policy_enabled(q, pol)) | |
1406 | return 0; | |
1407 | ||
344e9ffc | 1408 | if (queue_is_mq(q)) |
bd166ef1 | 1409 | blk_mq_freeze_queue(q); |
9d179b86 | 1410 | retry: |
0d945c1f | 1411 | spin_lock_irq(&q->queue_lock); |
a2b1693b | 1412 | |
9d179b86 | 1413 | /* blkg_list is pushed at the head, reverse walk to allocate parents first */ |
71c81407 | 1414 | list_for_each_entry_reverse(blkg, &q->blkg_list, q_node) { |
4c55f4f9 TH |
1415 | struct blkg_policy_data *pd; |
1416 | ||
1417 | if (blkg->pd[pol->plid]) | |
1418 | continue; | |
a2b1693b | 1419 | |
9d179b86 TH |
1420 | /* If prealloc matches, use it; otherwise try GFP_NOWAIT */ |
1421 | if (blkg == pinned_blkg) { | |
1422 | pd = pd_prealloc; | |
1423 | pd_prealloc = NULL; | |
1424 | } else { | |
1425 | pd = pol->pd_alloc_fn(GFP_NOWAIT | __GFP_NOWARN, q, | |
1426 | blkg->blkcg); | |
1427 | } | |
1428 | ||
4c55f4f9 | 1429 | if (!pd) { |
9d179b86 TH |
1430 | /* |
1431 | * GFP_NOWAIT failed. Free the existing one and | |
1432 | * prealloc for @blkg w/ GFP_KERNEL. | |
1433 | */ | |
1434 | if (pinned_blkg) | |
1435 | blkg_put(pinned_blkg); | |
1436 | blkg_get(blkg); | |
1437 | pinned_blkg = blkg; | |
1438 | ||
0d945c1f | 1439 | spin_unlock_irq(&q->queue_lock); |
9d179b86 TH |
1440 | |
1441 | if (pd_prealloc) | |
1442 | pol->pd_free_fn(pd_prealloc); | |
1443 | pd_prealloc = pol->pd_alloc_fn(GFP_KERNEL, q, | |
1444 | blkg->blkcg); | |
1445 | if (pd_prealloc) | |
1446 | goto retry; | |
1447 | else | |
1448 | goto enomem; | |
4c55f4f9 | 1449 | } |
a2b1693b TH |
1450 | |
1451 | blkg->pd[pol->plid] = pd; | |
1452 | pd->blkg = blkg; | |
b276a876 | 1453 | pd->plid = pol->plid; |
a2b1693b TH |
1454 | } |
1455 | ||
9d179b86 TH |
1456 | /* all allocated, init in the same order */ |
1457 | if (pol->pd_init_fn) | |
1458 | list_for_each_entry_reverse(blkg, &q->blkg_list, q_node) | |
1459 | pol->pd_init_fn(blkg->pd[pol->plid]); | |
1460 | ||
a2b1693b TH |
1461 | __set_bit(pol->plid, q->blkcg_pols); |
1462 | ret = 0; | |
4c55f4f9 | 1463 | |
0d945c1f | 1464 | spin_unlock_irq(&q->queue_lock); |
9d179b86 | 1465 | out: |
344e9ffc | 1466 | if (queue_is_mq(q)) |
bd166ef1 | 1467 | blk_mq_unfreeze_queue(q); |
9d179b86 TH |
1468 | if (pinned_blkg) |
1469 | blkg_put(pinned_blkg); | |
001bea73 TH |
1470 | if (pd_prealloc) |
1471 | pol->pd_free_fn(pd_prealloc); | |
a2b1693b | 1472 | return ret; |
9d179b86 TH |
1473 | |
1474 | enomem: | |
1475 | /* alloc failed, nothing's initialized yet, free everything */ | |
1476 | spin_lock_irq(&q->queue_lock); | |
1477 | list_for_each_entry(blkg, &q->blkg_list, q_node) { | |
858560b2 LJ |
1478 | struct blkcg *blkcg = blkg->blkcg; |
1479 | ||
1480 | spin_lock(&blkcg->lock); | |
9d179b86 TH |
1481 | if (blkg->pd[pol->plid]) { |
1482 | pol->pd_free_fn(blkg->pd[pol->plid]); | |
1483 | blkg->pd[pol->plid] = NULL; | |
1484 | } | |
858560b2 | 1485 | spin_unlock(&blkcg->lock); |
9d179b86 TH |
1486 | } |
1487 | spin_unlock_irq(&q->queue_lock); | |
1488 | ret = -ENOMEM; | |
1489 | goto out; | |
a2b1693b TH |
1490 | } |
1491 | EXPORT_SYMBOL_GPL(blkcg_activate_policy); | |
1492 | ||
1493 | /** | |
1494 | * blkcg_deactivate_policy - deactivate a blkcg policy on a request_queue | |
1495 | * @q: request_queue of interest | |
1496 | * @pol: blkcg policy to deactivate | |
1497 | * | |
1498 | * Deactivate @pol on @q. Follows the same synchronization rules as | |
1499 | * blkcg_activate_policy(). | |
1500 | */ | |
1501 | void blkcg_deactivate_policy(struct request_queue *q, | |
3c798398 | 1502 | const struct blkcg_policy *pol) |
a2b1693b | 1503 | { |
3c798398 | 1504 | struct blkcg_gq *blkg; |
a2b1693b TH |
1505 | |
1506 | if (!blkcg_policy_enabled(q, pol)) | |
1507 | return; | |
1508 | ||
344e9ffc | 1509 | if (queue_is_mq(q)) |
bd166ef1 | 1510 | blk_mq_freeze_queue(q); |
bd166ef1 | 1511 | |
0d945c1f | 1512 | spin_lock_irq(&q->queue_lock); |
a2b1693b TH |
1513 | |
1514 | __clear_bit(pol->plid, q->blkcg_pols); | |
1515 | ||
1516 | list_for_each_entry(blkg, &q->blkg_list, q_node) { | |
858560b2 LJ |
1517 | struct blkcg *blkcg = blkg->blkcg; |
1518 | ||
1519 | spin_lock(&blkcg->lock); | |
001bea73 | 1520 | if (blkg->pd[pol->plid]) { |
6b065462 | 1521 | if (pol->pd_offline_fn) |
a9520cd6 | 1522 | pol->pd_offline_fn(blkg->pd[pol->plid]); |
001bea73 TH |
1523 | pol->pd_free_fn(blkg->pd[pol->plid]); |
1524 | blkg->pd[pol->plid] = NULL; | |
1525 | } | |
858560b2 | 1526 | spin_unlock(&blkcg->lock); |
a2b1693b TH |
1527 | } |
1528 | ||
0d945c1f | 1529 | spin_unlock_irq(&q->queue_lock); |
bd166ef1 | 1530 | |
344e9ffc | 1531 | if (queue_is_mq(q)) |
bd166ef1 | 1532 | blk_mq_unfreeze_queue(q); |
a2b1693b TH |
1533 | } |
1534 | EXPORT_SYMBOL_GPL(blkcg_deactivate_policy); | |
1535 | ||
e55cf798 JY |
1536 | static void blkcg_free_all_cpd(struct blkcg_policy *pol) |
1537 | { | |
1538 | struct blkcg *blkcg; | |
1539 | ||
1540 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) { | |
1541 | if (blkcg->cpd[pol->plid]) { | |
1542 | pol->cpd_free_fn(blkcg->cpd[pol->plid]); | |
1543 | blkcg->cpd[pol->plid] = NULL; | |
1544 | } | |
1545 | } | |
1546 | } | |
1547 | ||
8bd435b3 | 1548 | /** |
3c798398 TH |
1549 | * blkcg_policy_register - register a blkcg policy |
1550 | * @pol: blkcg policy to register | |
8bd435b3 | 1551 | * |
3c798398 TH |
1552 | * Register @pol with blkcg core. Might sleep and @pol may be modified on |
1553 | * successful registration. Returns 0 on success and -errno on failure. | |
8bd435b3 | 1554 | */ |
d5bf0291 | 1555 | int blkcg_policy_register(struct blkcg_policy *pol) |
3e252066 | 1556 | { |
06b285bd | 1557 | struct blkcg *blkcg; |
8bd435b3 | 1558 | int i, ret; |
e8989fae | 1559 | |
838f13bf | 1560 | mutex_lock(&blkcg_pol_register_mutex); |
bc0d6501 TH |
1561 | mutex_lock(&blkcg_pol_mutex); |
1562 | ||
8bd435b3 TH |
1563 | /* find an empty slot */ |
1564 | ret = -ENOSPC; | |
1565 | for (i = 0; i < BLKCG_MAX_POLS; i++) | |
3c798398 | 1566 | if (!blkcg_policy[i]) |
8bd435b3 | 1567 | break; |
01c5f85a JA |
1568 | if (i >= BLKCG_MAX_POLS) { |
1569 | pr_warn("blkcg_policy_register: BLKCG_MAX_POLS too small\n"); | |
838f13bf | 1570 | goto err_unlock; |
01c5f85a | 1571 | } |
035d10b2 | 1572 | |
e8401073 | 1573 | /* Make sure cpd/pd_alloc_fn and cpd/pd_free_fn in pairs */ |
1574 | if ((!pol->cpd_alloc_fn ^ !pol->cpd_free_fn) || | |
1575 | (!pol->pd_alloc_fn ^ !pol->pd_free_fn)) | |
1576 | goto err_unlock; | |
1577 | ||
06b285bd | 1578 | /* register @pol */ |
3c798398 | 1579 | pol->plid = i; |
06b285bd TH |
1580 | blkcg_policy[pol->plid] = pol; |
1581 | ||
1582 | /* allocate and install cpd's */ | |
e4a9bde9 | 1583 | if (pol->cpd_alloc_fn) { |
06b285bd TH |
1584 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) { |
1585 | struct blkcg_policy_data *cpd; | |
1586 | ||
e4a9bde9 | 1587 | cpd = pol->cpd_alloc_fn(GFP_KERNEL); |
bbb427e3 | 1588 | if (!cpd) |
06b285bd | 1589 | goto err_free_cpds; |
06b285bd | 1590 | |
81437648 TH |
1591 | blkcg->cpd[pol->plid] = cpd; |
1592 | cpd->blkcg = blkcg; | |
06b285bd | 1593 | cpd->plid = pol->plid; |
86a5bba5 TH |
1594 | if (pol->cpd_init_fn) |
1595 | pol->cpd_init_fn(cpd); | |
06b285bd TH |
1596 | } |
1597 | } | |
1598 | ||
838f13bf | 1599 | mutex_unlock(&blkcg_pol_mutex); |
8bd435b3 | 1600 | |
8bd435b3 | 1601 | /* everything is in place, add intf files for the new policy */ |
2ee867dc TH |
1602 | if (pol->dfl_cftypes) |
1603 | WARN_ON(cgroup_add_dfl_cftypes(&io_cgrp_subsys, | |
1604 | pol->dfl_cftypes)); | |
880f50e2 | 1605 | if (pol->legacy_cftypes) |
c165b3e3 | 1606 | WARN_ON(cgroup_add_legacy_cftypes(&io_cgrp_subsys, |
880f50e2 | 1607 | pol->legacy_cftypes)); |
838f13bf TH |
1608 | mutex_unlock(&blkcg_pol_register_mutex); |
1609 | return 0; | |
1610 | ||
06b285bd | 1611 | err_free_cpds: |
e55cf798 JY |
1612 | if (pol->cpd_free_fn) |
1613 | blkcg_free_all_cpd(pol); | |
1614 | ||
06b285bd | 1615 | blkcg_policy[pol->plid] = NULL; |
838f13bf | 1616 | err_unlock: |
bc0d6501 | 1617 | mutex_unlock(&blkcg_pol_mutex); |
838f13bf | 1618 | mutex_unlock(&blkcg_pol_register_mutex); |
8bd435b3 | 1619 | return ret; |
3e252066 | 1620 | } |
3c798398 | 1621 | EXPORT_SYMBOL_GPL(blkcg_policy_register); |
3e252066 | 1622 | |
8bd435b3 | 1623 | /** |
3c798398 TH |
1624 | * blkcg_policy_unregister - unregister a blkcg policy |
1625 | * @pol: blkcg policy to unregister | |
8bd435b3 | 1626 | * |
3c798398 | 1627 | * Undo blkcg_policy_register(@pol). Might sleep. |
8bd435b3 | 1628 | */ |
3c798398 | 1629 | void blkcg_policy_unregister(struct blkcg_policy *pol) |
3e252066 | 1630 | { |
838f13bf | 1631 | mutex_lock(&blkcg_pol_register_mutex); |
bc0d6501 | 1632 | |
3c798398 | 1633 | if (WARN_ON(blkcg_policy[pol->plid] != pol)) |
8bd435b3 TH |
1634 | goto out_unlock; |
1635 | ||
1636 | /* kill the intf files first */ | |
2ee867dc TH |
1637 | if (pol->dfl_cftypes) |
1638 | cgroup_rm_cftypes(pol->dfl_cftypes); | |
880f50e2 TH |
1639 | if (pol->legacy_cftypes) |
1640 | cgroup_rm_cftypes(pol->legacy_cftypes); | |
44ea53de | 1641 | |
06b285bd | 1642 | /* remove cpds and unregister */ |
838f13bf | 1643 | mutex_lock(&blkcg_pol_mutex); |
06b285bd | 1644 | |
e55cf798 JY |
1645 | if (pol->cpd_free_fn) |
1646 | blkcg_free_all_cpd(pol); | |
1647 | ||
3c798398 | 1648 | blkcg_policy[pol->plid] = NULL; |
06b285bd | 1649 | |
bc0d6501 | 1650 | mutex_unlock(&blkcg_pol_mutex); |
838f13bf TH |
1651 | out_unlock: |
1652 | mutex_unlock(&blkcg_pol_register_mutex); | |
3e252066 | 1653 | } |
3c798398 | 1654 | EXPORT_SYMBOL_GPL(blkcg_policy_unregister); |
903d23f0 | 1655 | |
d3f77dfd TH |
1656 | bool __blkcg_punt_bio_submit(struct bio *bio) |
1657 | { | |
1658 | struct blkcg_gq *blkg = bio->bi_blkg; | |
1659 | ||
1660 | /* consume the flag first */ | |
1661 | bio->bi_opf &= ~REQ_CGROUP_PUNT; | |
1662 | ||
1663 | /* never bounce for the root cgroup */ | |
1664 | if (!blkg->parent) | |
1665 | return false; | |
1666 | ||
1667 | spin_lock_bh(&blkg->async_bio_lock); | |
1668 | bio_list_add(&blkg->async_bios, bio); | |
1669 | spin_unlock_bh(&blkg->async_bio_lock); | |
1670 | ||
1671 | queue_work(blkcg_punt_bio_wq, &blkg->async_bio_work); | |
1672 | return true; | |
1673 | } | |
1674 | ||
d09d8df3 JB |
1675 | /* |
1676 | * Scale the accumulated delay based on how long it has been since we updated | |
1677 | * the delay. We only call this when we are adding delay, in case it's been a | |
1678 | * while since we added delay, and when we are checking to see if we need to | |
1679 | * delay a task, to account for any delays that may have occurred. | |
1680 | */ | |
1681 | static void blkcg_scale_delay(struct blkcg_gq *blkg, u64 now) | |
1682 | { | |
1683 | u64 old = atomic64_read(&blkg->delay_start); | |
1684 | ||
54c52e10 TH |
1685 | /* negative use_delay means no scaling, see blkcg_set_delay() */ |
1686 | if (atomic_read(&blkg->use_delay) < 0) | |
1687 | return; | |
1688 | ||
d09d8df3 JB |
1689 | /* |
1690 | * We only want to scale down every second. The idea here is that we | |
1691 | * want to delay people for min(delay_nsec, NSEC_PER_SEC) in a certain | |
1692 | * time window. We only want to throttle tasks for recent delay that | |
1693 | * has occurred, in 1 second time windows since that's the maximum | |
1694 | * things can be throttled. We save the current delay window in | |
1695 | * blkg->last_delay so we know what amount is still left to be charged | |
1696 | * to the blkg from this point onward. blkg->last_use keeps track of | |
1697 | * the use_delay counter. The idea is if we're unthrottling the blkg we | |
1698 | * are ok with whatever is happening now, and we can take away more of | |
1699 | * the accumulated delay as we've already throttled enough that | |
1700 | * everybody is happy with their IO latencies. | |
1701 | */ | |
1702 | if (time_before64(old + NSEC_PER_SEC, now) && | |
96388f57 | 1703 | atomic64_try_cmpxchg(&blkg->delay_start, &old, now)) { |
d09d8df3 JB |
1704 | u64 cur = atomic64_read(&blkg->delay_nsec); |
1705 | u64 sub = min_t(u64, blkg->last_delay, now - old); | |
1706 | int cur_use = atomic_read(&blkg->use_delay); | |
1707 | ||
1708 | /* | |
1709 | * We've been unthrottled, subtract a larger chunk of our | |
1710 | * accumulated delay. | |
1711 | */ | |
1712 | if (cur_use < blkg->last_use) | |
1713 | sub = max_t(u64, sub, blkg->last_delay >> 1); | |
1714 | ||
1715 | /* | |
1716 | * This shouldn't happen, but handle it anyway. Our delay_nsec | |
1717 | * should only ever be growing except here where we subtract out | |
1718 | * min(last_delay, 1 second), but lord knows bugs happen and I'd | |
1719 | * rather not end up with negative numbers. | |
1720 | */ | |
1721 | if (unlikely(cur < sub)) { | |
1722 | atomic64_set(&blkg->delay_nsec, 0); | |
1723 | blkg->last_delay = 0; | |
1724 | } else { | |
1725 | atomic64_sub(sub, &blkg->delay_nsec); | |
1726 | blkg->last_delay = cur - sub; | |
1727 | } | |
1728 | blkg->last_use = cur_use; | |
1729 | } | |
1730 | } | |
1731 | ||
1732 | /* | |
1733 | * This is called when we want to actually walk up the hierarchy and check to | |
1734 | * see if we need to throttle, and then actually throttle if there is some | |
1735 | * accumulated delay. This should only be called upon return to user space so | |
1736 | * we're not holding some lock that would induce a priority inversion. | |
1737 | */ | |
1738 | static void blkcg_maybe_throttle_blkg(struct blkcg_gq *blkg, bool use_memdelay) | |
1739 | { | |
fd112c74 | 1740 | unsigned long pflags; |
5160a5a5 | 1741 | bool clamp; |
d09d8df3 JB |
1742 | u64 now = ktime_to_ns(ktime_get()); |
1743 | u64 exp; | |
1744 | u64 delay_nsec = 0; | |
1745 | int tok; | |
1746 | ||
1747 | while (blkg->parent) { | |
5160a5a5 TH |
1748 | int use_delay = atomic_read(&blkg->use_delay); |
1749 | ||
1750 | if (use_delay) { | |
1751 | u64 this_delay; | |
1752 | ||
d09d8df3 | 1753 | blkcg_scale_delay(blkg, now); |
5160a5a5 TH |
1754 | this_delay = atomic64_read(&blkg->delay_nsec); |
1755 | if (this_delay > delay_nsec) { | |
1756 | delay_nsec = this_delay; | |
1757 | clamp = use_delay > 0; | |
1758 | } | |
d09d8df3 JB |
1759 | } |
1760 | blkg = blkg->parent; | |
1761 | } | |
1762 | ||
1763 | if (!delay_nsec) | |
1764 | return; | |
1765 | ||
1766 | /* | |
1767 | * Let's not sleep for all eternity if we've amassed a huge delay. | |
1768 | * Swapping or metadata IO can accumulate 10's of seconds worth of | |
1769 | * delay, and we want userspace to be able to do _something_ so cap the | |
5160a5a5 TH |
1770 | * delays at 0.25s. If there's 10's of seconds worth of delay then the |
1771 | * tasks will be delayed for 0.25 second for every syscall. If | |
1772 | * blkcg_set_delay() was used as indicated by negative use_delay, the | |
1773 | * caller is responsible for regulating the range. | |
d09d8df3 | 1774 | */ |
5160a5a5 TH |
1775 | if (clamp) |
1776 | delay_nsec = min_t(u64, delay_nsec, 250 * NSEC_PER_MSEC); | |
d09d8df3 | 1777 | |
fd112c74 JB |
1778 | if (use_memdelay) |
1779 | psi_memstall_enter(&pflags); | |
d09d8df3 JB |
1780 | |
1781 | exp = ktime_add_ns(now, delay_nsec); | |
1782 | tok = io_schedule_prepare(); | |
1783 | do { | |
1784 | __set_current_state(TASK_KILLABLE); | |
1785 | if (!schedule_hrtimeout(&exp, HRTIMER_MODE_ABS)) | |
1786 | break; | |
1787 | } while (!fatal_signal_pending(current)); | |
1788 | io_schedule_finish(tok); | |
fd112c74 JB |
1789 | |
1790 | if (use_memdelay) | |
1791 | psi_memstall_leave(&pflags); | |
d09d8df3 JB |
1792 | } |
1793 | ||
1794 | /** | |
1795 | * blkcg_maybe_throttle_current - throttle the current task if it has been marked | |
1796 | * | |
1797 | * This is only called if we've been marked with set_notify_resume(). Obviously | |
1798 | * we can be set_notify_resume() for reasons other than blkcg throttling, so we | |
1799 | * check to see if current->throttle_queue is set and if not this doesn't do | |
1800 | * anything. This should only ever be called by the resume code, it's not meant | |
1801 | * to be called by people willy-nilly as it will actually do the work to | |
1802 | * throttle the task if it is setup for throttling. | |
1803 | */ | |
1804 | void blkcg_maybe_throttle_current(void) | |
1805 | { | |
1806 | struct request_queue *q = current->throttle_queue; | |
d09d8df3 JB |
1807 | struct blkcg *blkcg; |
1808 | struct blkcg_gq *blkg; | |
1809 | bool use_memdelay = current->use_memdelay; | |
1810 | ||
1811 | if (!q) | |
1812 | return; | |
1813 | ||
1814 | current->throttle_queue = NULL; | |
1815 | current->use_memdelay = false; | |
1816 | ||
1817 | rcu_read_lock(); | |
82778259 | 1818 | blkcg = css_to_blkcg(blkcg_css()); |
d09d8df3 JB |
1819 | if (!blkcg) |
1820 | goto out; | |
1821 | blkg = blkg_lookup(blkcg, q); | |
1822 | if (!blkg) | |
1823 | goto out; | |
7754f669 | 1824 | if (!blkg_tryget(blkg)) |
d09d8df3 JB |
1825 | goto out; |
1826 | rcu_read_unlock(); | |
d09d8df3 JB |
1827 | |
1828 | blkcg_maybe_throttle_blkg(blkg, use_memdelay); | |
1829 | blkg_put(blkg); | |
cc7ecc25 | 1830 | blk_put_queue(q); |
d09d8df3 JB |
1831 | return; |
1832 | out: | |
1833 | rcu_read_unlock(); | |
1834 | blk_put_queue(q); | |
1835 | } | |
d09d8df3 JB |
1836 | |
1837 | /** | |
1838 | * blkcg_schedule_throttle - this task needs to check for throttling | |
537d71b3 BVA |
1839 | * @q: the request queue IO was submitted on |
1840 | * @use_memdelay: do we charge this to memory delay for PSI | |
d09d8df3 JB |
1841 | * |
1842 | * This is called by the IO controller when we know there's delay accumulated | |
1843 | * for the blkg for this task. We do not pass the blkg because there are places | |
1844 | * we call this that may not have that information, the swapping code for | |
1845 | * instance will only have a request_queue at that point. This set's the | |
1846 | * notify_resume for the task to check and see if it requires throttling before | |
1847 | * returning to user space. | |
1848 | * | |
1849 | * We will only schedule once per syscall. You can call this over and over | |
1850 | * again and it will only do the check once upon return to user space, and only | |
1851 | * throttle once. If the task needs to be throttled again it'll need to be | |
1852 | * re-set at the next time we see the task. | |
1853 | */ | |
1854 | void blkcg_schedule_throttle(struct request_queue *q, bool use_memdelay) | |
1855 | { | |
1856 | if (unlikely(current->flags & PF_KTHREAD)) | |
1857 | return; | |
1858 | ||
49d1822b CX |
1859 | if (current->throttle_queue != q) { |
1860 | if (!blk_get_queue(q)) | |
1861 | return; | |
1862 | ||
1863 | if (current->throttle_queue) | |
1864 | blk_put_queue(current->throttle_queue); | |
1865 | current->throttle_queue = q; | |
1866 | } | |
d09d8df3 | 1867 | |
d09d8df3 JB |
1868 | if (use_memdelay) |
1869 | current->use_memdelay = use_memdelay; | |
1870 | set_notify_resume(current); | |
1871 | } | |
d09d8df3 JB |
1872 | |
1873 | /** | |
1874 | * blkcg_add_delay - add delay to this blkg | |
537d71b3 BVA |
1875 | * @blkg: blkg of interest |
1876 | * @now: the current time in nanoseconds | |
1877 | * @delta: how many nanoseconds of delay to add | |
d09d8df3 JB |
1878 | * |
1879 | * Charge @delta to the blkg's current delay accumulation. This is used to | |
1880 | * throttle tasks if an IO controller thinks we need more throttling. | |
1881 | */ | |
1882 | void blkcg_add_delay(struct blkcg_gq *blkg, u64 now, u64 delta) | |
1883 | { | |
54c52e10 TH |
1884 | if (WARN_ON_ONCE(atomic_read(&blkg->use_delay) < 0)) |
1885 | return; | |
d09d8df3 JB |
1886 | blkcg_scale_delay(blkg, now); |
1887 | atomic64_add(delta, &blkg->delay_nsec); | |
1888 | } | |
d09d8df3 | 1889 | |
28fc591f CH |
1890 | /** |
1891 | * blkg_tryget_closest - try and get a blkg ref on the closet blkg | |
13c7863d CH |
1892 | * @bio: target bio |
1893 | * @css: target css | |
28fc591f | 1894 | * |
13c7863d CH |
1895 | * As the failure mode here is to walk up the blkg tree, this ensure that the |
1896 | * blkg->parent pointers are always valid. This returns the blkg that it ended | |
1897 | * up taking a reference on or %NULL if no reference was taken. | |
28fc591f | 1898 | */ |
13c7863d CH |
1899 | static inline struct blkcg_gq *blkg_tryget_closest(struct bio *bio, |
1900 | struct cgroup_subsys_state *css) | |
28fc591f | 1901 | { |
13c7863d | 1902 | struct blkcg_gq *blkg, *ret_blkg = NULL; |
28fc591f | 1903 | |
13c7863d | 1904 | rcu_read_lock(); |
309dca30 | 1905 | blkg = blkg_lookup_create(css_to_blkcg(css), |
ed6cddef | 1906 | bdev_get_queue(bio->bi_bdev)); |
28fc591f CH |
1907 | while (blkg) { |
1908 | if (blkg_tryget(blkg)) { | |
1909 | ret_blkg = blkg; | |
1910 | break; | |
1911 | } | |
1912 | blkg = blkg->parent; | |
1913 | } | |
13c7863d | 1914 | rcu_read_unlock(); |
28fc591f CH |
1915 | |
1916 | return ret_blkg; | |
1917 | } | |
1918 | ||
1919 | /** | |
1920 | * bio_associate_blkg_from_css - associate a bio with a specified css | |
1921 | * @bio: target bio | |
1922 | * @css: target css | |
1923 | * | |
1924 | * Associate @bio with the blkg found by combining the css's blkg and the | |
1925 | * request_queue of the @bio. An association failure is handled by walking up | |
1926 | * the blkg tree. Therefore, the blkg associated can be anything between @blkg | |
1927 | * and q->root_blkg. This situation only happens when a cgroup is dying and | |
1928 | * then the remaining bios will spill to the closest alive blkg. | |
1929 | * | |
1930 | * A reference will be taken on the blkg and will be released when @bio is | |
1931 | * freed. | |
1932 | */ | |
1933 | void bio_associate_blkg_from_css(struct bio *bio, | |
1934 | struct cgroup_subsys_state *css) | |
1935 | { | |
28fc591f CH |
1936 | if (bio->bi_blkg) |
1937 | blkg_put(bio->bi_blkg); | |
1938 | ||
a5b97526 | 1939 | if (css && css->parent) { |
13c7863d | 1940 | bio->bi_blkg = blkg_tryget_closest(bio, css); |
a5b97526 | 1941 | } else { |
ed6cddef PB |
1942 | blkg_get(bdev_get_queue(bio->bi_bdev)->root_blkg); |
1943 | bio->bi_blkg = bdev_get_queue(bio->bi_bdev)->root_blkg; | |
a5b97526 | 1944 | } |
28fc591f CH |
1945 | } |
1946 | EXPORT_SYMBOL_GPL(bio_associate_blkg_from_css); | |
1947 | ||
1948 | /** | |
1949 | * bio_associate_blkg - associate a bio with a blkg | |
1950 | * @bio: target bio | |
1951 | * | |
1952 | * Associate @bio with the blkg found from the bio's css and request_queue. | |
1953 | * If one is not found, bio_lookup_blkg() creates the blkg. If a blkg is | |
1954 | * already associated, the css is reused and association redone as the | |
1955 | * request_queue may have changed. | |
1956 | */ | |
1957 | void bio_associate_blkg(struct bio *bio) | |
1958 | { | |
1959 | struct cgroup_subsys_state *css; | |
1960 | ||
1961 | rcu_read_lock(); | |
1962 | ||
1963 | if (bio->bi_blkg) | |
bbb1ebe7 | 1964 | css = bio_blkcg_css(bio); |
28fc591f CH |
1965 | else |
1966 | css = blkcg_css(); | |
1967 | ||
1968 | bio_associate_blkg_from_css(bio, css); | |
1969 | ||
1970 | rcu_read_unlock(); | |
1971 | } | |
1972 | EXPORT_SYMBOL_GPL(bio_associate_blkg); | |
1973 | ||
1974 | /** | |
1975 | * bio_clone_blkg_association - clone blkg association from src to dst bio | |
1976 | * @dst: destination bio | |
1977 | * @src: source bio | |
1978 | */ | |
1979 | void bio_clone_blkg_association(struct bio *dst, struct bio *src) | |
1980 | { | |
22b106e5 JK |
1981 | if (src->bi_blkg) |
1982 | bio_associate_blkg_from_css(dst, bio_blkcg_css(src)); | |
28fc591f CH |
1983 | } |
1984 | EXPORT_SYMBOL_GPL(bio_clone_blkg_association); | |
1985 | ||
db18a53e CH |
1986 | static int blk_cgroup_io_type(struct bio *bio) |
1987 | { | |
1988 | if (op_is_discard(bio->bi_opf)) | |
1989 | return BLKG_IOSTAT_DISCARD; | |
1990 | if (op_is_write(bio->bi_opf)) | |
1991 | return BLKG_IOSTAT_WRITE; | |
1992 | return BLKG_IOSTAT_READ; | |
1993 | } | |
1994 | ||
1995 | void blk_cgroup_bio_start(struct bio *bio) | |
1996 | { | |
1997 | int rwd = blk_cgroup_io_type(bio), cpu; | |
1998 | struct blkg_iostat_set *bis; | |
3c08b093 | 1999 | unsigned long flags; |
db18a53e CH |
2000 | |
2001 | cpu = get_cpu(); | |
2002 | bis = per_cpu_ptr(bio->bi_blkg->iostat_cpu, cpu); | |
3c08b093 | 2003 | flags = u64_stats_update_begin_irqsave(&bis->sync); |
db18a53e CH |
2004 | |
2005 | /* | |
2006 | * If the bio is flagged with BIO_CGROUP_ACCT it means this is a split | |
2007 | * bio and we would have already accounted for the size of the bio. | |
2008 | */ | |
2009 | if (!bio_flagged(bio, BIO_CGROUP_ACCT)) { | |
2010 | bio_set_flag(bio, BIO_CGROUP_ACCT); | |
0b8cc25d | 2011 | bis->cur.bytes[rwd] += bio->bi_iter.bi_size; |
db18a53e CH |
2012 | } |
2013 | bis->cur.ios[rwd]++; | |
2014 | ||
3c08b093 | 2015 | u64_stats_update_end_irqrestore(&bis->sync, flags); |
db18a53e CH |
2016 | if (cgroup_subsys_on_dfl(io_cgrp_subsys)) |
2017 | cgroup_rstat_updated(bio->bi_blkg->blkcg->css.cgroup, cpu); | |
2018 | put_cpu(); | |
2019 | } | |
2020 | ||
216889aa CH |
2021 | bool blk_cgroup_congested(void) |
2022 | { | |
2023 | struct cgroup_subsys_state *css; | |
2024 | bool ret = false; | |
2025 | ||
2026 | rcu_read_lock(); | |
d200ca14 | 2027 | for (css = blkcg_css(); css; css = css->parent) { |
216889aa CH |
2028 | if (atomic_read(&css->cgroup->congestion_count)) { |
2029 | ret = true; | |
2030 | break; | |
2031 | } | |
216889aa CH |
2032 | } |
2033 | rcu_read_unlock(); | |
2034 | return ret; | |
2035 | } | |
2036 | ||
d3f77dfd TH |
2037 | static int __init blkcg_init(void) |
2038 | { | |
2039 | blkcg_punt_bio_wq = alloc_workqueue("blkcg_punt_bio", | |
2040 | WQ_MEM_RECLAIM | WQ_FREEZABLE | | |
2041 | WQ_UNBOUND | WQ_SYSFS, 0); | |
2042 | if (!blkcg_punt_bio_wq) | |
2043 | return -ENOMEM; | |
2044 | return 0; | |
2045 | } | |
2046 | subsys_initcall(blkcg_init); | |
2047 | ||
903d23f0 JB |
2048 | module_param(blkcg_debug_stats, bool, 0644); |
2049 | MODULE_PARM_DESC(blkcg_debug_stats, "True if you want debug stats, false if not"); |