]>
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> |
34d0f179 | 26 | #include <linux/genhd.h> |
72e06c25 | 27 | #include <linux/delay.h> |
9a9e8a26 | 28 | #include <linux/atomic.h> |
36aa9e5f | 29 | #include <linux/ctype.h> |
eea8f41c | 30 | #include <linux/blk-cgroup.h> |
d09d8df3 | 31 | #include <linux/tracehook.h> |
fd112c74 | 32 | #include <linux/psi.h> |
5efd6113 | 33 | #include "blk.h" |
3e252066 | 34 | |
84c124da DS |
35 | #define MAX_KEY_LEN 100 |
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 | |
a2b1693b | 60 | static bool blkcg_policy_enabled(struct request_queue *q, |
3c798398 | 61 | const struct blkcg_policy *pol) |
a2b1693b TH |
62 | { |
63 | return pol && test_bit(pol->plid, q->blkcg_pols); | |
64 | } | |
65 | ||
0381411e TH |
66 | /** |
67 | * blkg_free - free a blkg | |
68 | * @blkg: blkg to free | |
69 | * | |
70 | * Free @blkg which may be partially allocated. | |
71 | */ | |
3c798398 | 72 | static void blkg_free(struct blkcg_gq *blkg) |
0381411e | 73 | { |
e8989fae | 74 | int i; |
549d3aa8 TH |
75 | |
76 | if (!blkg) | |
77 | return; | |
78 | ||
db613670 | 79 | for (i = 0; i < BLKCG_MAX_POLS; i++) |
001bea73 TH |
80 | if (blkg->pd[i]) |
81 | blkcg_policy[i]->pd_free_fn(blkg->pd[i]); | |
e8989fae | 82 | |
f7331648 | 83 | free_percpu(blkg->iostat_cpu); |
ef069b97 | 84 | percpu_ref_exit(&blkg->refcnt); |
549d3aa8 | 85 | kfree(blkg); |
0381411e TH |
86 | } |
87 | ||
7fcf2b03 DZ |
88 | static void __blkg_release(struct rcu_head *rcu) |
89 | { | |
90 | struct blkcg_gq *blkg = container_of(rcu, struct blkcg_gq, rcu_head); | |
91 | ||
d3f77dfd TH |
92 | WARN_ON(!bio_list_empty(&blkg->async_bios)); |
93 | ||
7fcf2b03 DZ |
94 | /* release the blkcg and parent blkg refs this blkg has been holding */ |
95 | css_put(&blkg->blkcg->css); | |
96 | if (blkg->parent) | |
97 | blkg_put(blkg->parent); | |
7fcf2b03 DZ |
98 | blkg_free(blkg); |
99 | } | |
100 | ||
101 | /* | |
102 | * A group is RCU protected, but having an rcu lock does not mean that one | |
103 | * can access all the fields of blkg and assume these are valid. For | |
104 | * example, don't try to follow throtl_data and request queue links. | |
105 | * | |
106 | * Having a reference to blkg under an rcu allows accesses to only values | |
107 | * local to groups like group stats and group rate limits. | |
108 | */ | |
109 | static void blkg_release(struct percpu_ref *ref) | |
110 | { | |
111 | struct blkcg_gq *blkg = container_of(ref, struct blkcg_gq, refcnt); | |
112 | ||
113 | call_rcu(&blkg->rcu_head, __blkg_release); | |
114 | } | |
115 | ||
d3f77dfd TH |
116 | static void blkg_async_bio_workfn(struct work_struct *work) |
117 | { | |
118 | struct blkcg_gq *blkg = container_of(work, struct blkcg_gq, | |
119 | async_bio_work); | |
120 | struct bio_list bios = BIO_EMPTY_LIST; | |
121 | struct bio *bio; | |
122 | ||
123 | /* as long as there are pending bios, @blkg can't go away */ | |
124 | spin_lock_bh(&blkg->async_bio_lock); | |
125 | bio_list_merge(&bios, &blkg->async_bios); | |
126 | bio_list_init(&blkg->async_bios); | |
127 | spin_unlock_bh(&blkg->async_bio_lock); | |
128 | ||
129 | while ((bio = bio_list_pop(&bios))) | |
130 | submit_bio(bio); | |
131 | } | |
132 | ||
0381411e TH |
133 | /** |
134 | * blkg_alloc - allocate a blkg | |
135 | * @blkcg: block cgroup the new blkg is associated with | |
136 | * @q: request_queue the new blkg is associated with | |
15974993 | 137 | * @gfp_mask: allocation mask to use |
0381411e | 138 | * |
e8989fae | 139 | * Allocate a new blkg assocating @blkcg and @q. |
0381411e | 140 | */ |
15974993 TH |
141 | static struct blkcg_gq *blkg_alloc(struct blkcg *blkcg, struct request_queue *q, |
142 | gfp_t gfp_mask) | |
0381411e | 143 | { |
3c798398 | 144 | struct blkcg_gq *blkg; |
f7331648 | 145 | int i, cpu; |
0381411e TH |
146 | |
147 | /* alloc and init base part */ | |
15974993 | 148 | blkg = kzalloc_node(sizeof(*blkg), gfp_mask, q->node); |
0381411e TH |
149 | if (!blkg) |
150 | return NULL; | |
151 | ||
ef069b97 TH |
152 | if (percpu_ref_init(&blkg->refcnt, blkg_release, 0, gfp_mask)) |
153 | goto err_free; | |
154 | ||
f7331648 TH |
155 | blkg->iostat_cpu = alloc_percpu_gfp(struct blkg_iostat_set, gfp_mask); |
156 | if (!blkg->iostat_cpu) | |
77ea7338 TH |
157 | goto err_free; |
158 | ||
c875f4d0 | 159 | blkg->q = q; |
e8989fae | 160 | INIT_LIST_HEAD(&blkg->q_node); |
d3f77dfd TH |
161 | spin_lock_init(&blkg->async_bio_lock); |
162 | bio_list_init(&blkg->async_bios); | |
163 | INIT_WORK(&blkg->async_bio_work, blkg_async_bio_workfn); | |
0381411e | 164 | blkg->blkcg = blkcg; |
0381411e | 165 | |
f7331648 TH |
166 | u64_stats_init(&blkg->iostat.sync); |
167 | for_each_possible_cpu(cpu) | |
168 | u64_stats_init(&per_cpu_ptr(blkg->iostat_cpu, cpu)->sync); | |
169 | ||
8bd435b3 | 170 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
3c798398 | 171 | struct blkcg_policy *pol = blkcg_policy[i]; |
e8989fae | 172 | struct blkg_policy_data *pd; |
0381411e | 173 | |
a2b1693b | 174 | if (!blkcg_policy_enabled(q, pol)) |
e8989fae TH |
175 | continue; |
176 | ||
177 | /* alloc per-policy data and attach it to blkg */ | |
cf09a8ee | 178 | pd = pol->pd_alloc_fn(gfp_mask, q, blkcg); |
a051661c TH |
179 | if (!pd) |
180 | goto err_free; | |
549d3aa8 | 181 | |
e8989fae TH |
182 | blkg->pd[i] = pd; |
183 | pd->blkg = blkg; | |
b276a876 | 184 | pd->plid = i; |
e8989fae TH |
185 | } |
186 | ||
0381411e | 187 | return blkg; |
a051661c TH |
188 | |
189 | err_free: | |
190 | blkg_free(blkg); | |
191 | return NULL; | |
0381411e TH |
192 | } |
193 | ||
24f29046 TH |
194 | struct blkcg_gq *blkg_lookup_slowpath(struct blkcg *blkcg, |
195 | struct request_queue *q, bool update_hint) | |
80fd9979 | 196 | { |
3c798398 | 197 | struct blkcg_gq *blkg; |
80fd9979 | 198 | |
a637120e | 199 | /* |
86cde6b6 TH |
200 | * Hint didn't match. Look up from the radix tree. Note that the |
201 | * hint can only be updated under queue_lock as otherwise @blkg | |
202 | * could have already been removed from blkg_tree. The caller is | |
203 | * responsible for grabbing queue_lock if @update_hint. | |
a637120e TH |
204 | */ |
205 | blkg = radix_tree_lookup(&blkcg->blkg_tree, q->id); | |
86cde6b6 TH |
206 | if (blkg && blkg->q == q) { |
207 | if (update_hint) { | |
0d945c1f | 208 | lockdep_assert_held(&q->queue_lock); |
86cde6b6 TH |
209 | rcu_assign_pointer(blkcg->blkg_hint, blkg); |
210 | } | |
a637120e | 211 | return blkg; |
86cde6b6 | 212 | } |
a637120e | 213 | |
80fd9979 TH |
214 | return NULL; |
215 | } | |
ae118896 | 216 | EXPORT_SYMBOL_GPL(blkg_lookup_slowpath); |
80fd9979 | 217 | |
15974993 | 218 | /* |
d708f0d5 JA |
219 | * If @new_blkg is %NULL, this function tries to allocate a new one as |
220 | * necessary using %GFP_NOWAIT. @new_blkg is always consumed on return. | |
15974993 | 221 | */ |
86cde6b6 | 222 | static struct blkcg_gq *blkg_create(struct blkcg *blkcg, |
d708f0d5 JA |
223 | struct request_queue *q, |
224 | struct blkcg_gq *new_blkg) | |
5624a4e4 | 225 | { |
d708f0d5 | 226 | struct blkcg_gq *blkg; |
f427d909 | 227 | int i, ret; |
5624a4e4 | 228 | |
cd1604fa | 229 | WARN_ON_ONCE(!rcu_read_lock_held()); |
0d945c1f | 230 | lockdep_assert_held(&q->queue_lock); |
cd1604fa | 231 | |
0273ac34 DZ |
232 | /* request_queue is dying, do not create/recreate a blkg */ |
233 | if (blk_queue_dying(q)) { | |
234 | ret = -ENODEV; | |
235 | goto err_free_blkg; | |
236 | } | |
237 | ||
7ee9c562 | 238 | /* blkg holds a reference to blkcg */ |
ec903c0c | 239 | if (!css_tryget_online(&blkcg->css)) { |
20386ce0 | 240 | ret = -ENODEV; |
93e6d5d8 | 241 | goto err_free_blkg; |
15974993 | 242 | } |
cd1604fa | 243 | |
d708f0d5 JA |
244 | /* allocate */ |
245 | if (!new_blkg) { | |
246 | new_blkg = blkg_alloc(blkcg, q, GFP_NOWAIT | __GFP_NOWARN); | |
247 | if (unlikely(!new_blkg)) { | |
248 | ret = -ENOMEM; | |
8c911f3d | 249 | goto err_put_css; |
15974993 TH |
250 | } |
251 | } | |
d708f0d5 | 252 | blkg = new_blkg; |
cd1604fa | 253 | |
db613670 | 254 | /* link parent */ |
3c547865 TH |
255 | if (blkcg_parent(blkcg)) { |
256 | blkg->parent = __blkg_lookup(blkcg_parent(blkcg), q, false); | |
257 | if (WARN_ON_ONCE(!blkg->parent)) { | |
20386ce0 | 258 | ret = -ENODEV; |
8c911f3d | 259 | goto err_put_css; |
3c547865 TH |
260 | } |
261 | blkg_get(blkg->parent); | |
262 | } | |
263 | ||
db613670 TH |
264 | /* invoke per-policy init */ |
265 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
266 | struct blkcg_policy *pol = blkcg_policy[i]; | |
267 | ||
268 | if (blkg->pd[i] && pol->pd_init_fn) | |
a9520cd6 | 269 | pol->pd_init_fn(blkg->pd[i]); |
db613670 TH |
270 | } |
271 | ||
272 | /* insert */ | |
cd1604fa | 273 | spin_lock(&blkcg->lock); |
a637120e TH |
274 | ret = radix_tree_insert(&blkcg->blkg_tree, q->id, blkg); |
275 | if (likely(!ret)) { | |
276 | hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list); | |
277 | list_add(&blkg->q_node, &q->blkg_list); | |
f427d909 TH |
278 | |
279 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
280 | struct blkcg_policy *pol = blkcg_policy[i]; | |
281 | ||
282 | if (blkg->pd[i] && pol->pd_online_fn) | |
a9520cd6 | 283 | pol->pd_online_fn(blkg->pd[i]); |
f427d909 | 284 | } |
a637120e | 285 | } |
f427d909 | 286 | blkg->online = true; |
cd1604fa | 287 | spin_unlock(&blkcg->lock); |
496fb780 | 288 | |
ec13b1d6 | 289 | if (!ret) |
a637120e | 290 | return blkg; |
15974993 | 291 | |
3c547865 TH |
292 | /* @blkg failed fully initialized, use the usual release path */ |
293 | blkg_put(blkg); | |
294 | return ERR_PTR(ret); | |
295 | ||
d708f0d5 | 296 | err_put_css: |
496fb780 | 297 | css_put(&blkcg->css); |
93e6d5d8 | 298 | err_free_blkg: |
d708f0d5 | 299 | blkg_free(new_blkg); |
93e6d5d8 | 300 | return ERR_PTR(ret); |
31e4c28d | 301 | } |
3c96cb32 | 302 | |
86cde6b6 | 303 | /** |
8c546287 | 304 | * blkg_lookup_create - lookup blkg, try to create one if not there |
86cde6b6 TH |
305 | * @blkcg: blkcg of interest |
306 | * @q: request_queue of interest | |
307 | * | |
308 | * Lookup blkg for the @blkcg - @q pair. If it doesn't exist, try to | |
3c547865 TH |
309 | * create one. blkg creation is performed recursively from blkcg_root such |
310 | * that all non-root blkg's have access to the parent blkg. This function | |
8c546287 | 311 | * should be called under RCU read lock and takes @q->queue_lock. |
86cde6b6 | 312 | * |
beea9da0 DZ |
313 | * Returns the blkg or the closest blkg if blkg_create() fails as it walks |
314 | * down from root. | |
86cde6b6 | 315 | */ |
8c546287 CH |
316 | static struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg, |
317 | struct request_queue *q) | |
3c96cb32 | 318 | { |
86cde6b6 | 319 | struct blkcg_gq *blkg; |
8c546287 | 320 | unsigned long flags; |
86cde6b6 TH |
321 | |
322 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
86cde6b6 | 323 | |
8c546287 | 324 | blkg = blkg_lookup(blkcg, q); |
86cde6b6 TH |
325 | if (blkg) |
326 | return blkg; | |
327 | ||
8c546287 CH |
328 | spin_lock_irqsave(&q->queue_lock, flags); |
329 | blkg = __blkg_lookup(blkcg, q, true); | |
330 | if (blkg) | |
331 | goto found; | |
332 | ||
3c547865 TH |
333 | /* |
334 | * Create blkgs walking down from blkcg_root to @blkcg, so that all | |
beea9da0 DZ |
335 | * non-root blkgs have access to their parents. Returns the closest |
336 | * blkg to the intended blkg should blkg_create() fail. | |
3c547865 TH |
337 | */ |
338 | while (true) { | |
339 | struct blkcg *pos = blkcg; | |
340 | struct blkcg *parent = blkcg_parent(blkcg); | |
beea9da0 DZ |
341 | struct blkcg_gq *ret_blkg = q->root_blkg; |
342 | ||
343 | while (parent) { | |
344 | blkg = __blkg_lookup(parent, q, false); | |
345 | if (blkg) { | |
346 | /* remember closest blkg */ | |
347 | ret_blkg = blkg; | |
348 | break; | |
349 | } | |
3c547865 TH |
350 | pos = parent; |
351 | parent = blkcg_parent(parent); | |
352 | } | |
353 | ||
d708f0d5 | 354 | blkg = blkg_create(pos, q, NULL); |
8c546287 CH |
355 | if (IS_ERR(blkg)) { |
356 | blkg = ret_blkg; | |
357 | break; | |
358 | } | |
beea9da0 | 359 | if (pos == blkcg) |
8c546287 | 360 | break; |
b978962a DZ |
361 | } |
362 | ||
8c546287 CH |
363 | found: |
364 | spin_unlock_irqrestore(&q->queue_lock, flags); | |
b978962a DZ |
365 | return blkg; |
366 | } | |
367 | ||
3c798398 | 368 | static void blkg_destroy(struct blkcg_gq *blkg) |
03aa264a | 369 | { |
3c798398 | 370 | struct blkcg *blkcg = blkg->blkcg; |
6b065462 | 371 | int i; |
03aa264a | 372 | |
0d945c1f | 373 | lockdep_assert_held(&blkg->q->queue_lock); |
9f13ef67 | 374 | lockdep_assert_held(&blkcg->lock); |
03aa264a TH |
375 | |
376 | /* Something wrong if we are trying to remove same group twice */ | |
e8989fae | 377 | WARN_ON_ONCE(list_empty(&blkg->q_node)); |
9f13ef67 | 378 | WARN_ON_ONCE(hlist_unhashed(&blkg->blkcg_node)); |
a637120e | 379 | |
6b065462 DZF |
380 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
381 | struct blkcg_policy *pol = blkcg_policy[i]; | |
382 | ||
383 | if (blkg->pd[i] && pol->pd_offline_fn) | |
384 | pol->pd_offline_fn(blkg->pd[i]); | |
385 | } | |
386 | ||
f427d909 TH |
387 | blkg->online = false; |
388 | ||
a637120e | 389 | radix_tree_delete(&blkcg->blkg_tree, blkg->q->id); |
e8989fae | 390 | list_del_init(&blkg->q_node); |
9f13ef67 | 391 | hlist_del_init_rcu(&blkg->blkcg_node); |
03aa264a | 392 | |
a637120e TH |
393 | /* |
394 | * Both setting lookup hint to and clearing it from @blkg are done | |
395 | * under queue_lock. If it's not pointing to @blkg now, it never | |
396 | * will. Hint assignment itself can race safely. | |
397 | */ | |
ec6c676a | 398 | if (rcu_access_pointer(blkcg->blkg_hint) == blkg) |
a637120e TH |
399 | rcu_assign_pointer(blkcg->blkg_hint, NULL); |
400 | ||
03aa264a TH |
401 | /* |
402 | * Put the reference taken at the time of creation so that when all | |
403 | * queues are gone, group can be destroyed. | |
404 | */ | |
7fcf2b03 | 405 | percpu_ref_kill(&blkg->refcnt); |
03aa264a TH |
406 | } |
407 | ||
9f13ef67 TH |
408 | /** |
409 | * blkg_destroy_all - destroy all blkgs associated with a request_queue | |
410 | * @q: request_queue of interest | |
9f13ef67 | 411 | * |
3c96cb32 | 412 | * Destroy all blkgs associated with @q. |
9f13ef67 | 413 | */ |
3c96cb32 | 414 | static void blkg_destroy_all(struct request_queue *q) |
72e06c25 | 415 | { |
3c798398 | 416 | struct blkcg_gq *blkg, *n; |
72e06c25 | 417 | |
0d945c1f | 418 | spin_lock_irq(&q->queue_lock); |
9f13ef67 | 419 | list_for_each_entry_safe(blkg, n, &q->blkg_list, q_node) { |
3c798398 | 420 | struct blkcg *blkcg = blkg->blkcg; |
72e06c25 | 421 | |
9f13ef67 TH |
422 | spin_lock(&blkcg->lock); |
423 | blkg_destroy(blkg); | |
424 | spin_unlock(&blkcg->lock); | |
72e06c25 | 425 | } |
6fe810bd TH |
426 | |
427 | q->root_blkg = NULL; | |
0d945c1f | 428 | spin_unlock_irq(&q->queue_lock); |
72e06c25 TH |
429 | } |
430 | ||
182446d0 TH |
431 | static int blkcg_reset_stats(struct cgroup_subsys_state *css, |
432 | struct cftype *cftype, u64 val) | |
303a3acb | 433 | { |
182446d0 | 434 | struct blkcg *blkcg = css_to_blkcg(css); |
3c798398 | 435 | struct blkcg_gq *blkg; |
f7331648 | 436 | int i, cpu; |
303a3acb | 437 | |
838f13bf | 438 | mutex_lock(&blkcg_pol_mutex); |
303a3acb | 439 | spin_lock_irq(&blkcg->lock); |
997a026c TH |
440 | |
441 | /* | |
442 | * Note that stat reset is racy - it doesn't synchronize against | |
443 | * stat updates. This is a debug feature which shouldn't exist | |
444 | * anyway. If you get hit by a race, retry. | |
445 | */ | |
b67bfe0d | 446 | hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) { |
f7331648 TH |
447 | for_each_possible_cpu(cpu) { |
448 | struct blkg_iostat_set *bis = | |
449 | per_cpu_ptr(blkg->iostat_cpu, cpu); | |
450 | memset(bis, 0, sizeof(*bis)); | |
451 | } | |
452 | memset(&blkg->iostat, 0, sizeof(blkg->iostat)); | |
77ea7338 | 453 | |
8bd435b3 | 454 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
3c798398 | 455 | struct blkcg_policy *pol = blkcg_policy[i]; |
549d3aa8 | 456 | |
a9520cd6 TH |
457 | if (blkg->pd[i] && pol->pd_reset_stats_fn) |
458 | pol->pd_reset_stats_fn(blkg->pd[i]); | |
bc0d6501 | 459 | } |
303a3acb | 460 | } |
f0bdc8cd | 461 | |
303a3acb | 462 | spin_unlock_irq(&blkcg->lock); |
bc0d6501 | 463 | mutex_unlock(&blkcg_pol_mutex); |
303a3acb DS |
464 | return 0; |
465 | } | |
466 | ||
dd165eb3 | 467 | const char *blkg_dev_name(struct blkcg_gq *blkg) |
303a3acb | 468 | { |
d3d32e69 | 469 | /* some drivers (floppy) instantiate a queue w/o disk registered */ |
dc3b17cc | 470 | if (blkg->q->backing_dev_info->dev) |
d51cfc53 | 471 | return bdi_dev_name(blkg->q->backing_dev_info); |
d3d32e69 | 472 | return NULL; |
303a3acb DS |
473 | } |
474 | ||
d3d32e69 TH |
475 | /** |
476 | * blkcg_print_blkgs - helper for printing per-blkg data | |
477 | * @sf: seq_file to print to | |
478 | * @blkcg: blkcg of interest | |
479 | * @prfill: fill function to print out a blkg | |
480 | * @pol: policy in question | |
481 | * @data: data to be passed to @prfill | |
482 | * @show_total: to print out sum of prfill return values or not | |
483 | * | |
484 | * This function invokes @prfill on each blkg of @blkcg if pd for the | |
485 | * policy specified by @pol exists. @prfill is invoked with @sf, the | |
810ecfa7 TH |
486 | * policy data and @data and the matching queue lock held. If @show_total |
487 | * is %true, the sum of the return values from @prfill is printed with | |
488 | * "Total" label at the end. | |
d3d32e69 TH |
489 | * |
490 | * This is to be used to construct print functions for | |
491 | * cftype->read_seq_string method. | |
492 | */ | |
3c798398 | 493 | void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg, |
f95a04af TH |
494 | u64 (*prfill)(struct seq_file *, |
495 | struct blkg_policy_data *, int), | |
3c798398 | 496 | const struct blkcg_policy *pol, int data, |
ec399347 | 497 | bool show_total) |
5624a4e4 | 498 | { |
3c798398 | 499 | struct blkcg_gq *blkg; |
d3d32e69 | 500 | u64 total = 0; |
5624a4e4 | 501 | |
810ecfa7 | 502 | rcu_read_lock(); |
ee89f812 | 503 | hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) { |
0d945c1f | 504 | spin_lock_irq(&blkg->q->queue_lock); |
a2b1693b | 505 | if (blkcg_policy_enabled(blkg->q, pol)) |
f95a04af | 506 | total += prfill(sf, blkg->pd[pol->plid], data); |
0d945c1f | 507 | spin_unlock_irq(&blkg->q->queue_lock); |
810ecfa7 TH |
508 | } |
509 | rcu_read_unlock(); | |
d3d32e69 TH |
510 | |
511 | if (show_total) | |
512 | seq_printf(sf, "Total %llu\n", (unsigned long long)total); | |
513 | } | |
829fdb50 | 514 | EXPORT_SYMBOL_GPL(blkcg_print_blkgs); |
d3d32e69 TH |
515 | |
516 | /** | |
517 | * __blkg_prfill_u64 - prfill helper for a single u64 value | |
518 | * @sf: seq_file to print to | |
f95a04af | 519 | * @pd: policy private data of interest |
d3d32e69 TH |
520 | * @v: value to print |
521 | * | |
f95a04af | 522 | * Print @v to @sf for the device assocaited with @pd. |
d3d32e69 | 523 | */ |
f95a04af | 524 | u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v) |
d3d32e69 | 525 | { |
f95a04af | 526 | const char *dname = blkg_dev_name(pd->blkg); |
d3d32e69 TH |
527 | |
528 | if (!dname) | |
529 | return 0; | |
530 | ||
531 | seq_printf(sf, "%s %llu\n", dname, (unsigned long long)v); | |
532 | return v; | |
533 | } | |
829fdb50 | 534 | EXPORT_SYMBOL_GPL(__blkg_prfill_u64); |
d3d32e69 | 535 | |
457e490f TE |
536 | /* Performs queue bypass and policy enabled checks then looks up blkg. */ |
537 | static struct blkcg_gq *blkg_lookup_check(struct blkcg *blkcg, | |
538 | const struct blkcg_policy *pol, | |
539 | struct request_queue *q) | |
540 | { | |
541 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
0d945c1f | 542 | lockdep_assert_held(&q->queue_lock); |
457e490f TE |
543 | |
544 | if (!blkcg_policy_enabled(q, pol)) | |
545 | return ERR_PTR(-EOPNOTSUPP); | |
457e490f TE |
546 | return __blkg_lookup(blkcg, q, true /* update_hint */); |
547 | } | |
548 | ||
015d254c TH |
549 | /** |
550 | * blkg_conf_prep - parse and prepare for per-blkg config update | |
551 | * @inputp: input string pointer | |
552 | * | |
553 | * Parse the device node prefix part, MAJ:MIN, of per-blkg config update | |
554 | * from @input and get and return the matching gendisk. *@inputp is | |
555 | * updated to point past the device node prefix. Returns an ERR_PTR() | |
556 | * value on error. | |
557 | * | |
558 | * Use this function iff blkg_conf_prep() can't be used for some reason. | |
559 | */ | |
560 | struct gendisk *blkcg_conf_get_disk(char **inputp) | |
561 | { | |
562 | char *input = *inputp; | |
563 | unsigned int major, minor; | |
564 | struct gendisk *disk; | |
565 | int key_len, part; | |
566 | ||
567 | if (sscanf(input, "%u:%u%n", &major, &minor, &key_len) != 2) | |
568 | return ERR_PTR(-EINVAL); | |
569 | ||
570 | input += key_len; | |
571 | if (!isspace(*input)) | |
572 | return ERR_PTR(-EINVAL); | |
573 | input = skip_spaces(input); | |
574 | ||
575 | disk = get_gendisk(MKDEV(major, minor), &part); | |
576 | if (!disk) | |
577 | return ERR_PTR(-ENODEV); | |
578 | if (part) { | |
579 | put_disk_and_module(disk); | |
580 | return ERR_PTR(-ENODEV); | |
581 | } | |
582 | ||
583 | *inputp = input; | |
584 | return disk; | |
585 | } | |
586 | ||
3a8b31d3 TH |
587 | /** |
588 | * blkg_conf_prep - parse and prepare for per-blkg config update | |
589 | * @blkcg: target block cgroup | |
da8b0662 | 590 | * @pol: target policy |
3a8b31d3 TH |
591 | * @input: input string |
592 | * @ctx: blkg_conf_ctx to be filled | |
593 | * | |
594 | * Parse per-blkg config update from @input and initialize @ctx with the | |
36aa9e5f TH |
595 | * result. @ctx->blkg points to the blkg to be updated and @ctx->body the |
596 | * part of @input following MAJ:MIN. This function returns with RCU read | |
597 | * lock and queue lock held and must be paired with blkg_conf_finish(). | |
3a8b31d3 | 598 | */ |
3c798398 | 599 | int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol, |
36aa9e5f | 600 | char *input, struct blkg_conf_ctx *ctx) |
0d945c1f | 601 | __acquires(rcu) __acquires(&disk->queue->queue_lock) |
34d0f179 | 602 | { |
3a8b31d3 | 603 | struct gendisk *disk; |
457e490f | 604 | struct request_queue *q; |
3c798398 | 605 | struct blkcg_gq *blkg; |
015d254c | 606 | int ret; |
36aa9e5f | 607 | |
015d254c TH |
608 | disk = blkcg_conf_get_disk(&input); |
609 | if (IS_ERR(disk)) | |
610 | return PTR_ERR(disk); | |
e56da7e2 | 611 | |
457e490f | 612 | q = disk->queue; |
da8b0662 | 613 | |
457e490f | 614 | rcu_read_lock(); |
0d945c1f | 615 | spin_lock_irq(&q->queue_lock); |
e56da7e2 | 616 | |
457e490f | 617 | blkg = blkg_lookup_check(blkcg, pol, q); |
4bfd482e TH |
618 | if (IS_ERR(blkg)) { |
619 | ret = PTR_ERR(blkg); | |
457e490f TE |
620 | goto fail_unlock; |
621 | } | |
622 | ||
623 | if (blkg) | |
624 | goto success; | |
625 | ||
626 | /* | |
627 | * Create blkgs walking down from blkcg_root to @blkcg, so that all | |
628 | * non-root blkgs have access to their parents. | |
629 | */ | |
630 | while (true) { | |
631 | struct blkcg *pos = blkcg; | |
632 | struct blkcg *parent; | |
633 | struct blkcg_gq *new_blkg; | |
634 | ||
635 | parent = blkcg_parent(blkcg); | |
636 | while (parent && !__blkg_lookup(parent, q, false)) { | |
637 | pos = parent; | |
638 | parent = blkcg_parent(parent); | |
639 | } | |
640 | ||
641 | /* Drop locks to do new blkg allocation with GFP_KERNEL. */ | |
0d945c1f | 642 | spin_unlock_irq(&q->queue_lock); |
3a8b31d3 | 643 | rcu_read_unlock(); |
457e490f TE |
644 | |
645 | new_blkg = blkg_alloc(pos, q, GFP_KERNEL); | |
646 | if (unlikely(!new_blkg)) { | |
647 | ret = -ENOMEM; | |
648 | goto fail; | |
7702e8f4 | 649 | } |
3a8b31d3 | 650 | |
457e490f | 651 | rcu_read_lock(); |
0d945c1f | 652 | spin_lock_irq(&q->queue_lock); |
457e490f TE |
653 | |
654 | blkg = blkg_lookup_check(pos, pol, q); | |
655 | if (IS_ERR(blkg)) { | |
656 | ret = PTR_ERR(blkg); | |
657 | goto fail_unlock; | |
658 | } | |
659 | ||
660 | if (blkg) { | |
661 | blkg_free(new_blkg); | |
662 | } else { | |
663 | blkg = blkg_create(pos, q, new_blkg); | |
98d669b4 | 664 | if (IS_ERR(blkg)) { |
457e490f TE |
665 | ret = PTR_ERR(blkg); |
666 | goto fail_unlock; | |
667 | } | |
668 | } | |
669 | ||
670 | if (pos == blkcg) | |
671 | goto success; | |
672 | } | |
673 | success: | |
3a8b31d3 TH |
674 | ctx->disk = disk; |
675 | ctx->blkg = blkg; | |
015d254c | 676 | ctx->body = input; |
726fa694 | 677 | return 0; |
457e490f TE |
678 | |
679 | fail_unlock: | |
0d945c1f | 680 | spin_unlock_irq(&q->queue_lock); |
457e490f TE |
681 | rcu_read_unlock(); |
682 | fail: | |
9df6c299 | 683 | put_disk_and_module(disk); |
457e490f TE |
684 | /* |
685 | * If queue was bypassing, we should retry. Do so after a | |
686 | * short msleep(). It isn't strictly necessary but queue | |
687 | * can be bypassing for some time and it's always nice to | |
688 | * avoid busy looping. | |
689 | */ | |
690 | if (ret == -EBUSY) { | |
691 | msleep(10); | |
692 | ret = restart_syscall(); | |
693 | } | |
694 | return ret; | |
34d0f179 | 695 | } |
89f3b6d6 | 696 | EXPORT_SYMBOL_GPL(blkg_conf_prep); |
34d0f179 | 697 | |
3a8b31d3 TH |
698 | /** |
699 | * blkg_conf_finish - finish up per-blkg config update | |
700 | * @ctx: blkg_conf_ctx intiailized by blkg_conf_prep() | |
701 | * | |
702 | * Finish up after per-blkg config update. This function must be paired | |
703 | * with blkg_conf_prep(). | |
704 | */ | |
829fdb50 | 705 | void blkg_conf_finish(struct blkg_conf_ctx *ctx) |
0d945c1f | 706 | __releases(&ctx->disk->queue->queue_lock) __releases(rcu) |
34d0f179 | 707 | { |
0d945c1f | 708 | spin_unlock_irq(&ctx->disk->queue->queue_lock); |
3a8b31d3 | 709 | rcu_read_unlock(); |
9df6c299 | 710 | put_disk_and_module(ctx->disk); |
34d0f179 | 711 | } |
89f3b6d6 | 712 | EXPORT_SYMBOL_GPL(blkg_conf_finish); |
34d0f179 | 713 | |
cd1fc4b9 BB |
714 | static void blkg_iostat_set(struct blkg_iostat *dst, struct blkg_iostat *src) |
715 | { | |
716 | int i; | |
717 | ||
718 | for (i = 0; i < BLKG_IOSTAT_NR; i++) { | |
719 | dst->bytes[i] = src->bytes[i]; | |
720 | dst->ios[i] = src->ios[i]; | |
721 | } | |
722 | } | |
723 | ||
724 | static void blkg_iostat_add(struct blkg_iostat *dst, struct blkg_iostat *src) | |
725 | { | |
726 | int i; | |
727 | ||
728 | for (i = 0; i < BLKG_IOSTAT_NR; i++) { | |
729 | dst->bytes[i] += src->bytes[i]; | |
730 | dst->ios[i] += src->ios[i]; | |
731 | } | |
732 | } | |
733 | ||
734 | static void blkg_iostat_sub(struct blkg_iostat *dst, struct blkg_iostat *src) | |
735 | { | |
736 | int i; | |
737 | ||
738 | for (i = 0; i < BLKG_IOSTAT_NR; i++) { | |
739 | dst->bytes[i] -= src->bytes[i]; | |
740 | dst->ios[i] -= src->ios[i]; | |
741 | } | |
742 | } | |
743 | ||
744 | static void blkcg_rstat_flush(struct cgroup_subsys_state *css, int cpu) | |
745 | { | |
746 | struct blkcg *blkcg = css_to_blkcg(css); | |
747 | struct blkcg_gq *blkg; | |
748 | ||
749 | rcu_read_lock(); | |
750 | ||
751 | hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) { | |
752 | struct blkcg_gq *parent = blkg->parent; | |
753 | struct blkg_iostat_set *bisc = per_cpu_ptr(blkg->iostat_cpu, cpu); | |
754 | struct blkg_iostat cur, delta; | |
755 | unsigned int seq; | |
756 | ||
757 | /* fetch the current per-cpu values */ | |
758 | do { | |
759 | seq = u64_stats_fetch_begin(&bisc->sync); | |
760 | blkg_iostat_set(&cur, &bisc->cur); | |
761 | } while (u64_stats_fetch_retry(&bisc->sync, seq)); | |
762 | ||
763 | /* propagate percpu delta to global */ | |
764 | u64_stats_update_begin(&blkg->iostat.sync); | |
765 | blkg_iostat_set(&delta, &cur); | |
766 | blkg_iostat_sub(&delta, &bisc->last); | |
767 | blkg_iostat_add(&blkg->iostat.cur, &delta); | |
768 | blkg_iostat_add(&bisc->last, &delta); | |
769 | u64_stats_update_end(&blkg->iostat.sync); | |
770 | ||
771 | /* propagate global delta to parent */ | |
772 | if (parent) { | |
773 | u64_stats_update_begin(&parent->iostat.sync); | |
774 | blkg_iostat_set(&delta, &blkg->iostat.cur); | |
775 | blkg_iostat_sub(&delta, &blkg->iostat.last); | |
776 | blkg_iostat_add(&parent->iostat.cur, &delta); | |
777 | blkg_iostat_add(&blkg->iostat.last, &delta); | |
778 | u64_stats_update_end(&parent->iostat.sync); | |
779 | } | |
780 | } | |
781 | ||
782 | rcu_read_unlock(); | |
783 | } | |
784 | ||
ef45fe47 BB |
785 | /* |
786 | * The rstat algorithms intentionally don't handle the root cgroup to avoid | |
787 | * incurring overhead when no cgroups are defined. For that reason, | |
788 | * cgroup_rstat_flush in blkcg_print_stat does not actually fill out the | |
789 | * iostat in the root cgroup's blkcg_gq. | |
790 | * | |
791 | * However, we would like to re-use the printing code between the root and | |
792 | * non-root cgroups to the extent possible. For that reason, we simulate | |
793 | * flushing the root cgroup's stats by explicitly filling in the iostat | |
794 | * with disk level statistics. | |
795 | */ | |
796 | static void blkcg_fill_root_iostats(void) | |
797 | { | |
798 | struct class_dev_iter iter; | |
799 | struct device *dev; | |
800 | ||
801 | class_dev_iter_init(&iter, &block_class, NULL, &disk_type); | |
802 | while ((dev = class_dev_iter_next(&iter))) { | |
803 | struct gendisk *disk = dev_to_disk(dev); | |
804 | struct hd_struct *part = disk_get_part(disk, 0); | |
805 | struct blkcg_gq *blkg = blk_queue_root_blkg(disk->queue); | |
806 | struct blkg_iostat tmp; | |
807 | int cpu; | |
808 | ||
809 | memset(&tmp, 0, sizeof(tmp)); | |
810 | for_each_possible_cpu(cpu) { | |
811 | struct disk_stats *cpu_dkstats; | |
812 | ||
813 | cpu_dkstats = per_cpu_ptr(part->dkstats, cpu); | |
814 | tmp.ios[BLKG_IOSTAT_READ] += | |
815 | cpu_dkstats->ios[STAT_READ]; | |
816 | tmp.ios[BLKG_IOSTAT_WRITE] += | |
817 | cpu_dkstats->ios[STAT_WRITE]; | |
818 | tmp.ios[BLKG_IOSTAT_DISCARD] += | |
819 | cpu_dkstats->ios[STAT_DISCARD]; | |
820 | // convert sectors to bytes | |
821 | tmp.bytes[BLKG_IOSTAT_READ] += | |
822 | cpu_dkstats->sectors[STAT_READ] << 9; | |
823 | tmp.bytes[BLKG_IOSTAT_WRITE] += | |
824 | cpu_dkstats->sectors[STAT_WRITE] << 9; | |
825 | tmp.bytes[BLKG_IOSTAT_DISCARD] += | |
826 | cpu_dkstats->sectors[STAT_DISCARD] << 9; | |
827 | ||
828 | u64_stats_update_begin(&blkg->iostat.sync); | |
829 | blkg_iostat_set(&blkg->iostat.cur, &tmp); | |
830 | u64_stats_update_end(&blkg->iostat.sync); | |
831 | } | |
832 | } | |
833 | } | |
834 | ||
2ee867dc TH |
835 | static int blkcg_print_stat(struct seq_file *sf, void *v) |
836 | { | |
837 | struct blkcg *blkcg = css_to_blkcg(seq_css(sf)); | |
838 | struct blkcg_gq *blkg; | |
839 | ||
ef45fe47 BB |
840 | if (!seq_css(sf)->parent) |
841 | blkcg_fill_root_iostats(); | |
842 | else | |
843 | cgroup_rstat_flush(blkcg->css.cgroup); | |
844 | ||
2ee867dc TH |
845 | rcu_read_lock(); |
846 | ||
847 | hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) { | |
f7331648 | 848 | struct blkg_iostat_set *bis = &blkg->iostat; |
2ee867dc | 849 | const char *dname; |
903d23f0 | 850 | char *buf; |
636620b6 | 851 | u64 rbytes, wbytes, rios, wios, dbytes, dios; |
903d23f0 JB |
852 | size_t size = seq_get_buf(sf, &buf), off = 0; |
853 | int i; | |
854 | bool has_stats = false; | |
f7331648 | 855 | unsigned seq; |
2ee867dc | 856 | |
b0814361 TH |
857 | spin_lock_irq(&blkg->q->queue_lock); |
858 | ||
859 | if (!blkg->online) | |
860 | goto skip; | |
861 | ||
2ee867dc TH |
862 | dname = blkg_dev_name(blkg); |
863 | if (!dname) | |
b0814361 | 864 | goto skip; |
2ee867dc | 865 | |
903d23f0 JB |
866 | /* |
867 | * Hooray string manipulation, count is the size written NOT | |
868 | * INCLUDING THE \0, so size is now count+1 less than what we | |
869 | * had before, but we want to start writing the next bit from | |
870 | * the \0 so we only add count to buf. | |
871 | */ | |
872 | off += scnprintf(buf+off, size-off, "%s ", dname); | |
873 | ||
f7331648 TH |
874 | do { |
875 | seq = u64_stats_fetch_begin(&bis->sync); | |
2ee867dc | 876 | |
f7331648 TH |
877 | rbytes = bis->cur.bytes[BLKG_IOSTAT_READ]; |
878 | wbytes = bis->cur.bytes[BLKG_IOSTAT_WRITE]; | |
879 | dbytes = bis->cur.bytes[BLKG_IOSTAT_DISCARD]; | |
880 | rios = bis->cur.ios[BLKG_IOSTAT_READ]; | |
881 | wios = bis->cur.ios[BLKG_IOSTAT_WRITE]; | |
882 | dios = bis->cur.ios[BLKG_IOSTAT_DISCARD]; | |
883 | } while (u64_stats_fetch_retry(&bis->sync, seq)); | |
2ee867dc | 884 | |
903d23f0 JB |
885 | if (rbytes || wbytes || rios || wios) { |
886 | has_stats = true; | |
887 | off += scnprintf(buf+off, size-off, | |
636620b6 TH |
888 | "rbytes=%llu wbytes=%llu rios=%llu wios=%llu dbytes=%llu dios=%llu", |
889 | rbytes, wbytes, rios, wios, | |
890 | dbytes, dios); | |
903d23f0 JB |
891 | } |
892 | ||
07b0fdec | 893 | if (blkcg_debug_stats && atomic_read(&blkg->use_delay)) { |
d09d8df3 JB |
894 | has_stats = true; |
895 | off += scnprintf(buf+off, size-off, | |
896 | " use_delay=%d delay_nsec=%llu", | |
897 | atomic_read(&blkg->use_delay), | |
898 | (unsigned long long)atomic64_read(&blkg->delay_nsec)); | |
899 | } | |
900 | ||
903d23f0 JB |
901 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
902 | struct blkcg_policy *pol = blkcg_policy[i]; | |
903 | size_t written; | |
904 | ||
905 | if (!blkg->pd[i] || !pol->pd_stat_fn) | |
906 | continue; | |
907 | ||
908 | written = pol->pd_stat_fn(blkg->pd[i], buf+off, size-off); | |
909 | if (written) | |
910 | has_stats = true; | |
911 | off += written; | |
912 | } | |
07b0fdec | 913 | |
903d23f0 | 914 | if (has_stats) { |
f539da82 TH |
915 | if (off < size - 1) { |
916 | off += scnprintf(buf+off, size-off, "\n"); | |
917 | seq_commit(sf, off); | |
918 | } else { | |
919 | seq_commit(sf, -1); | |
920 | } | |
903d23f0 | 921 | } |
b0814361 TH |
922 | skip: |
923 | spin_unlock_irq(&blkg->q->queue_lock); | |
2ee867dc TH |
924 | } |
925 | ||
926 | rcu_read_unlock(); | |
927 | return 0; | |
928 | } | |
929 | ||
e1f3b941 | 930 | static struct cftype blkcg_files[] = { |
2ee867dc TH |
931 | { |
932 | .name = "stat", | |
933 | .seq_show = blkcg_print_stat, | |
934 | }, | |
935 | { } /* terminate */ | |
936 | }; | |
937 | ||
e1f3b941 | 938 | static struct cftype blkcg_legacy_files[] = { |
84c124da DS |
939 | { |
940 | .name = "reset_stats", | |
3c798398 | 941 | .write_u64 = blkcg_reset_stats, |
22084190 | 942 | }, |
4baf6e33 | 943 | { } /* terminate */ |
31e4c28d VG |
944 | }; |
945 | ||
59b57717 DZF |
946 | /* |
947 | * blkcg destruction is a three-stage process. | |
948 | * | |
949 | * 1. Destruction starts. The blkcg_css_offline() callback is invoked | |
950 | * which offlines writeback. Here we tie the next stage of blkg destruction | |
951 | * to the completion of writeback associated with the blkcg. This lets us | |
952 | * avoid punting potentially large amounts of outstanding writeback to root | |
953 | * while maintaining any ongoing policies. The next stage is triggered when | |
954 | * the nr_cgwbs count goes to zero. | |
955 | * | |
956 | * 2. When the nr_cgwbs count goes to zero, blkcg_destroy_blkgs() is called | |
957 | * and handles the destruction of blkgs. Here the css reference held by | |
958 | * the blkg is put back eventually allowing blkcg_css_free() to be called. | |
959 | * This work may occur in cgwb_release_workfn() on the cgwb_release | |
960 | * workqueue. Any submitted ios that fail to get the blkg ref will be | |
961 | * punted to the root_blkg. | |
962 | * | |
963 | * 3. Once the blkcg ref count goes to zero, blkcg_css_free() is called. | |
964 | * This finally frees the blkcg. | |
965 | */ | |
966 | ||
9f13ef67 | 967 | /** |
92fb9748 | 968 | * blkcg_css_offline - cgroup css_offline callback |
eb95419b | 969 | * @css: css of interest |
9f13ef67 | 970 | * |
59b57717 DZF |
971 | * This function is called when @css is about to go away. Here the cgwbs are |
972 | * offlined first and only once writeback associated with the blkcg has | |
973 | * finished do we start step 2 (see above). | |
9f13ef67 | 974 | */ |
eb95419b | 975 | static void blkcg_css_offline(struct cgroup_subsys_state *css) |
31e4c28d | 976 | { |
eb95419b | 977 | struct blkcg *blkcg = css_to_blkcg(css); |
b1c35769 | 978 | |
59b57717 DZF |
979 | /* this prevents anyone from attaching or migrating to this blkcg */ |
980 | wb_blkcg_offline(blkcg); | |
981 | ||
d866dbf6 TH |
982 | /* put the base online pin allowing step 2 to be triggered */ |
983 | blkcg_unpin_online(blkcg); | |
59b57717 DZF |
984 | } |
985 | ||
986 | /** | |
987 | * blkcg_destroy_blkgs - responsible for shooting down blkgs | |
988 | * @blkcg: blkcg of interest | |
989 | * | |
990 | * blkgs should be removed while holding both q and blkcg locks. As blkcg lock | |
991 | * is nested inside q lock, this function performs reverse double lock dancing. | |
992 | * Destroying the blkgs releases the reference held on the blkcg's css allowing | |
993 | * blkcg_css_free to eventually be called. | |
994 | * | |
995 | * This is the blkcg counterpart of ioc_release_fn(). | |
996 | */ | |
997 | void blkcg_destroy_blkgs(struct blkcg *blkcg) | |
998 | { | |
9f13ef67 | 999 | spin_lock_irq(&blkcg->lock); |
7ee9c562 | 1000 | |
4c699480 JQ |
1001 | while (!hlist_empty(&blkcg->blkg_list)) { |
1002 | struct blkcg_gq *blkg = hlist_entry(blkcg->blkg_list.first, | |
6b065462 | 1003 | struct blkcg_gq, blkcg_node); |
4c699480 JQ |
1004 | struct request_queue *q = blkg->q; |
1005 | ||
0d945c1f | 1006 | if (spin_trylock(&q->queue_lock)) { |
4c699480 | 1007 | blkg_destroy(blkg); |
0d945c1f | 1008 | spin_unlock(&q->queue_lock); |
4c699480 JQ |
1009 | } else { |
1010 | spin_unlock_irq(&blkcg->lock); | |
1011 | cpu_relax(); | |
1012 | spin_lock_irq(&blkcg->lock); | |
1013 | } | |
1014 | } | |
6b065462 | 1015 | |
4c699480 JQ |
1016 | spin_unlock_irq(&blkcg->lock); |
1017 | } | |
1018 | ||
eb95419b | 1019 | static void blkcg_css_free(struct cgroup_subsys_state *css) |
7ee9c562 | 1020 | { |
eb95419b | 1021 | struct blkcg *blkcg = css_to_blkcg(css); |
bc915e61 | 1022 | int i; |
7ee9c562 | 1023 | |
7876f930 | 1024 | mutex_lock(&blkcg_pol_mutex); |
e4a9bde9 | 1025 | |
7876f930 | 1026 | list_del(&blkcg->all_blkcgs_node); |
7876f930 | 1027 | |
bc915e61 | 1028 | for (i = 0; i < BLKCG_MAX_POLS; i++) |
e4a9bde9 TH |
1029 | if (blkcg->cpd[i]) |
1030 | blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]); | |
1031 | ||
1032 | mutex_unlock(&blkcg_pol_mutex); | |
1033 | ||
bc915e61 | 1034 | kfree(blkcg); |
31e4c28d VG |
1035 | } |
1036 | ||
eb95419b TH |
1037 | static struct cgroup_subsys_state * |
1038 | blkcg_css_alloc(struct cgroup_subsys_state *parent_css) | |
31e4c28d | 1039 | { |
3c798398 | 1040 | struct blkcg *blkcg; |
e48453c3 AA |
1041 | struct cgroup_subsys_state *ret; |
1042 | int i; | |
31e4c28d | 1043 | |
7876f930 TH |
1044 | mutex_lock(&blkcg_pol_mutex); |
1045 | ||
eb95419b | 1046 | if (!parent_css) { |
3c798398 | 1047 | blkcg = &blkcg_root; |
bc915e61 TH |
1048 | } else { |
1049 | blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL); | |
1050 | if (!blkcg) { | |
1051 | ret = ERR_PTR(-ENOMEM); | |
4c18c9e9 | 1052 | goto unlock; |
bc915e61 | 1053 | } |
e48453c3 AA |
1054 | } |
1055 | ||
1056 | for (i = 0; i < BLKCG_MAX_POLS ; i++) { | |
1057 | struct blkcg_policy *pol = blkcg_policy[i]; | |
1058 | struct blkcg_policy_data *cpd; | |
1059 | ||
1060 | /* | |
1061 | * If the policy hasn't been attached yet, wait for it | |
1062 | * to be attached before doing anything else. Otherwise, | |
1063 | * check if the policy requires any specific per-cgroup | |
1064 | * data: if it does, allocate and initialize it. | |
1065 | */ | |
e4a9bde9 | 1066 | if (!pol || !pol->cpd_alloc_fn) |
e48453c3 AA |
1067 | continue; |
1068 | ||
e4a9bde9 | 1069 | cpd = pol->cpd_alloc_fn(GFP_KERNEL); |
e48453c3 AA |
1070 | if (!cpd) { |
1071 | ret = ERR_PTR(-ENOMEM); | |
1072 | goto free_pd_blkcg; | |
1073 | } | |
81437648 TH |
1074 | blkcg->cpd[i] = cpd; |
1075 | cpd->blkcg = blkcg; | |
e48453c3 | 1076 | cpd->plid = i; |
e4a9bde9 TH |
1077 | if (pol->cpd_init_fn) |
1078 | pol->cpd_init_fn(cpd); | |
e48453c3 | 1079 | } |
31e4c28d | 1080 | |
31e4c28d | 1081 | spin_lock_init(&blkcg->lock); |
d866dbf6 | 1082 | refcount_set(&blkcg->online_pin, 1); |
e00f4f4d | 1083 | INIT_RADIX_TREE(&blkcg->blkg_tree, GFP_NOWAIT | __GFP_NOWARN); |
31e4c28d | 1084 | INIT_HLIST_HEAD(&blkcg->blkg_list); |
52ebea74 TH |
1085 | #ifdef CONFIG_CGROUP_WRITEBACK |
1086 | INIT_LIST_HEAD(&blkcg->cgwb_list); | |
1087 | #endif | |
7876f930 TH |
1088 | list_add_tail(&blkcg->all_blkcgs_node, &all_blkcgs); |
1089 | ||
1090 | mutex_unlock(&blkcg_pol_mutex); | |
31e4c28d | 1091 | return &blkcg->css; |
e48453c3 AA |
1092 | |
1093 | free_pd_blkcg: | |
1094 | for (i--; i >= 0; i--) | |
e4a9bde9 TH |
1095 | if (blkcg->cpd[i]) |
1096 | blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]); | |
4c18c9e9 | 1097 | |
1098 | if (blkcg != &blkcg_root) | |
1099 | kfree(blkcg); | |
1100 | unlock: | |
7876f930 | 1101 | mutex_unlock(&blkcg_pol_mutex); |
e48453c3 | 1102 | return ret; |
31e4c28d VG |
1103 | } |
1104 | ||
4308a434 TH |
1105 | static int blkcg_css_online(struct cgroup_subsys_state *css) |
1106 | { | |
1107 | struct blkcg *blkcg = css_to_blkcg(css); | |
1108 | struct blkcg *parent = blkcg_parent(blkcg); | |
1109 | ||
1110 | /* | |
1111 | * blkcg_pin_online() is used to delay blkcg offline so that blkgs | |
1112 | * don't go offline while cgwbs are still active on them. Pin the | |
1113 | * parent so that offline always happens towards the root. | |
1114 | */ | |
1115 | if (parent) | |
1116 | blkcg_pin_online(parent); | |
1117 | return 0; | |
1118 | } | |
1119 | ||
5efd6113 TH |
1120 | /** |
1121 | * blkcg_init_queue - initialize blkcg part of request queue | |
1122 | * @q: request_queue to initialize | |
1123 | * | |
c62b37d9 | 1124 | * Called from blk_alloc_queue(). Responsible for initializing blkcg |
5efd6113 TH |
1125 | * part of new request_queue @q. |
1126 | * | |
1127 | * RETURNS: | |
1128 | * 0 on success, -errno on failure. | |
1129 | */ | |
1130 | int blkcg_init_queue(struct request_queue *q) | |
1131 | { | |
d708f0d5 JA |
1132 | struct blkcg_gq *new_blkg, *blkg; |
1133 | bool preloaded; | |
ec13b1d6 TH |
1134 | int ret; |
1135 | ||
d708f0d5 JA |
1136 | new_blkg = blkg_alloc(&blkcg_root, q, GFP_KERNEL); |
1137 | if (!new_blkg) | |
1138 | return -ENOMEM; | |
1139 | ||
1140 | preloaded = !radix_tree_preload(GFP_KERNEL); | |
1141 | ||
bea54883 | 1142 | /* Make sure the root blkg exists. */ |
ec13b1d6 | 1143 | rcu_read_lock(); |
0d945c1f | 1144 | spin_lock_irq(&q->queue_lock); |
d708f0d5 | 1145 | blkg = blkg_create(&blkcg_root, q, new_blkg); |
901932a3 JB |
1146 | if (IS_ERR(blkg)) |
1147 | goto err_unlock; | |
1148 | q->root_blkg = blkg; | |
0d945c1f | 1149 | spin_unlock_irq(&q->queue_lock); |
ec13b1d6 TH |
1150 | rcu_read_unlock(); |
1151 | ||
d708f0d5 JA |
1152 | if (preloaded) |
1153 | radix_tree_preload_end(); | |
1154 | ||
d7067512 | 1155 | ret = blk_iolatency_init(q); |
04be60b5 CH |
1156 | if (ret) |
1157 | goto err_destroy_all; | |
d7067512 | 1158 | |
ec13b1d6 | 1159 | ret = blk_throtl_init(q); |
04be60b5 CH |
1160 | if (ret) |
1161 | goto err_destroy_all; | |
1162 | return 0; | |
901932a3 | 1163 | |
04be60b5 | 1164 | err_destroy_all: |
04be60b5 | 1165 | blkg_destroy_all(q); |
04be60b5 | 1166 | return ret; |
901932a3 | 1167 | err_unlock: |
0d945c1f | 1168 | spin_unlock_irq(&q->queue_lock); |
901932a3 JB |
1169 | rcu_read_unlock(); |
1170 | if (preloaded) | |
1171 | radix_tree_preload_end(); | |
1172 | return PTR_ERR(blkg); | |
5efd6113 TH |
1173 | } |
1174 | ||
5efd6113 TH |
1175 | /** |
1176 | * blkcg_exit_queue - exit and release blkcg part of request_queue | |
1177 | * @q: request_queue being released | |
1178 | * | |
7585d508 | 1179 | * Called from blk_exit_queue(). Responsible for exiting blkcg part. |
5efd6113 TH |
1180 | */ |
1181 | void blkcg_exit_queue(struct request_queue *q) | |
1182 | { | |
3c96cb32 | 1183 | blkg_destroy_all(q); |
5efd6113 TH |
1184 | blk_throtl_exit(q); |
1185 | } | |
1186 | ||
31e4c28d VG |
1187 | /* |
1188 | * We cannot support shared io contexts, as we have no mean to support | |
1189 | * two tasks with the same ioc in two different groups without major rework | |
1190 | * of the main cic data structures. For now we allow a task to change | |
1191 | * its cgroup only if it's the only owner of its ioc. | |
1192 | */ | |
1f7dd3e5 | 1193 | static int blkcg_can_attach(struct cgroup_taskset *tset) |
31e4c28d | 1194 | { |
bb9d97b6 | 1195 | struct task_struct *task; |
1f7dd3e5 | 1196 | struct cgroup_subsys_state *dst_css; |
31e4c28d VG |
1197 | struct io_context *ioc; |
1198 | int ret = 0; | |
1199 | ||
1200 | /* task_lock() is needed to avoid races with exit_io_context() */ | |
1f7dd3e5 | 1201 | cgroup_taskset_for_each(task, dst_css, tset) { |
bb9d97b6 TH |
1202 | task_lock(task); |
1203 | ioc = task->io_context; | |
1204 | if (ioc && atomic_read(&ioc->nr_tasks) > 1) | |
1205 | ret = -EINVAL; | |
1206 | task_unlock(task); | |
1207 | if (ret) | |
1208 | break; | |
1209 | } | |
31e4c28d VG |
1210 | return ret; |
1211 | } | |
1212 | ||
69d7fde5 TH |
1213 | static void blkcg_bind(struct cgroup_subsys_state *root_css) |
1214 | { | |
1215 | int i; | |
1216 | ||
1217 | mutex_lock(&blkcg_pol_mutex); | |
1218 | ||
1219 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
1220 | struct blkcg_policy *pol = blkcg_policy[i]; | |
1221 | struct blkcg *blkcg; | |
1222 | ||
1223 | if (!pol || !pol->cpd_bind_fn) | |
1224 | continue; | |
1225 | ||
1226 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) | |
1227 | if (blkcg->cpd[pol->plid]) | |
1228 | pol->cpd_bind_fn(blkcg->cpd[pol->plid]); | |
1229 | } | |
1230 | mutex_unlock(&blkcg_pol_mutex); | |
1231 | } | |
1232 | ||
d09d8df3 JB |
1233 | static void blkcg_exit(struct task_struct *tsk) |
1234 | { | |
1235 | if (tsk->throttle_queue) | |
1236 | blk_put_queue(tsk->throttle_queue); | |
1237 | tsk->throttle_queue = NULL; | |
1238 | } | |
1239 | ||
c165b3e3 | 1240 | struct cgroup_subsys io_cgrp_subsys = { |
92fb9748 | 1241 | .css_alloc = blkcg_css_alloc, |
4308a434 | 1242 | .css_online = blkcg_css_online, |
92fb9748 TH |
1243 | .css_offline = blkcg_css_offline, |
1244 | .css_free = blkcg_css_free, | |
3c798398 | 1245 | .can_attach = blkcg_can_attach, |
f7331648 | 1246 | .css_rstat_flush = blkcg_rstat_flush, |
69d7fde5 | 1247 | .bind = blkcg_bind, |
2ee867dc | 1248 | .dfl_cftypes = blkcg_files, |
880f50e2 | 1249 | .legacy_cftypes = blkcg_legacy_files, |
c165b3e3 | 1250 | .legacy_name = "blkio", |
d09d8df3 | 1251 | .exit = blkcg_exit, |
1ced953b TH |
1252 | #ifdef CONFIG_MEMCG |
1253 | /* | |
1254 | * This ensures that, if available, memcg is automatically enabled | |
1255 | * together on the default hierarchy so that the owner cgroup can | |
1256 | * be retrieved from writeback pages. | |
1257 | */ | |
1258 | .depends_on = 1 << memory_cgrp_id, | |
1259 | #endif | |
676f7c8f | 1260 | }; |
c165b3e3 | 1261 | EXPORT_SYMBOL_GPL(io_cgrp_subsys); |
676f7c8f | 1262 | |
a2b1693b TH |
1263 | /** |
1264 | * blkcg_activate_policy - activate a blkcg policy on a request_queue | |
1265 | * @q: request_queue of interest | |
1266 | * @pol: blkcg policy to activate | |
1267 | * | |
1268 | * Activate @pol on @q. Requires %GFP_KERNEL context. @q goes through | |
1269 | * bypass mode to populate its blkgs with policy_data for @pol. | |
1270 | * | |
1271 | * Activation happens with @q bypassed, so nobody would be accessing blkgs | |
1272 | * from IO path. Update of each blkg is protected by both queue and blkcg | |
1273 | * locks so that holding either lock and testing blkcg_policy_enabled() is | |
1274 | * always enough for dereferencing policy data. | |
1275 | * | |
1276 | * The caller is responsible for synchronizing [de]activations and policy | |
1277 | * [un]registerations. Returns 0 on success, -errno on failure. | |
1278 | */ | |
1279 | int blkcg_activate_policy(struct request_queue *q, | |
3c798398 | 1280 | const struct blkcg_policy *pol) |
a2b1693b | 1281 | { |
4c55f4f9 | 1282 | struct blkg_policy_data *pd_prealloc = NULL; |
9d179b86 | 1283 | struct blkcg_gq *blkg, *pinned_blkg = NULL; |
4c55f4f9 | 1284 | int ret; |
a2b1693b TH |
1285 | |
1286 | if (blkcg_policy_enabled(q, pol)) | |
1287 | return 0; | |
1288 | ||
344e9ffc | 1289 | if (queue_is_mq(q)) |
bd166ef1 | 1290 | blk_mq_freeze_queue(q); |
9d179b86 | 1291 | retry: |
0d945c1f | 1292 | spin_lock_irq(&q->queue_lock); |
a2b1693b | 1293 | |
9d179b86 | 1294 | /* blkg_list is pushed at the head, reverse walk to allocate parents first */ |
71c81407 | 1295 | list_for_each_entry_reverse(blkg, &q->blkg_list, q_node) { |
4c55f4f9 TH |
1296 | struct blkg_policy_data *pd; |
1297 | ||
1298 | if (blkg->pd[pol->plid]) | |
1299 | continue; | |
a2b1693b | 1300 | |
9d179b86 TH |
1301 | /* If prealloc matches, use it; otherwise try GFP_NOWAIT */ |
1302 | if (blkg == pinned_blkg) { | |
1303 | pd = pd_prealloc; | |
1304 | pd_prealloc = NULL; | |
1305 | } else { | |
1306 | pd = pol->pd_alloc_fn(GFP_NOWAIT | __GFP_NOWARN, q, | |
1307 | blkg->blkcg); | |
1308 | } | |
1309 | ||
4c55f4f9 | 1310 | if (!pd) { |
9d179b86 TH |
1311 | /* |
1312 | * GFP_NOWAIT failed. Free the existing one and | |
1313 | * prealloc for @blkg w/ GFP_KERNEL. | |
1314 | */ | |
1315 | if (pinned_blkg) | |
1316 | blkg_put(pinned_blkg); | |
1317 | blkg_get(blkg); | |
1318 | pinned_blkg = blkg; | |
1319 | ||
0d945c1f | 1320 | spin_unlock_irq(&q->queue_lock); |
9d179b86 TH |
1321 | |
1322 | if (pd_prealloc) | |
1323 | pol->pd_free_fn(pd_prealloc); | |
1324 | pd_prealloc = pol->pd_alloc_fn(GFP_KERNEL, q, | |
1325 | blkg->blkcg); | |
1326 | if (pd_prealloc) | |
1327 | goto retry; | |
1328 | else | |
1329 | goto enomem; | |
4c55f4f9 | 1330 | } |
a2b1693b TH |
1331 | |
1332 | blkg->pd[pol->plid] = pd; | |
1333 | pd->blkg = blkg; | |
b276a876 | 1334 | pd->plid = pol->plid; |
a2b1693b TH |
1335 | } |
1336 | ||
9d179b86 TH |
1337 | /* all allocated, init in the same order */ |
1338 | if (pol->pd_init_fn) | |
1339 | list_for_each_entry_reverse(blkg, &q->blkg_list, q_node) | |
1340 | pol->pd_init_fn(blkg->pd[pol->plid]); | |
1341 | ||
a2b1693b TH |
1342 | __set_bit(pol->plid, q->blkcg_pols); |
1343 | ret = 0; | |
4c55f4f9 | 1344 | |
0d945c1f | 1345 | spin_unlock_irq(&q->queue_lock); |
9d179b86 | 1346 | out: |
344e9ffc | 1347 | if (queue_is_mq(q)) |
bd166ef1 | 1348 | blk_mq_unfreeze_queue(q); |
9d179b86 TH |
1349 | if (pinned_blkg) |
1350 | blkg_put(pinned_blkg); | |
001bea73 TH |
1351 | if (pd_prealloc) |
1352 | pol->pd_free_fn(pd_prealloc); | |
a2b1693b | 1353 | return ret; |
9d179b86 TH |
1354 | |
1355 | enomem: | |
1356 | /* alloc failed, nothing's initialized yet, free everything */ | |
1357 | spin_lock_irq(&q->queue_lock); | |
1358 | list_for_each_entry(blkg, &q->blkg_list, q_node) { | |
1359 | if (blkg->pd[pol->plid]) { | |
1360 | pol->pd_free_fn(blkg->pd[pol->plid]); | |
1361 | blkg->pd[pol->plid] = NULL; | |
1362 | } | |
1363 | } | |
1364 | spin_unlock_irq(&q->queue_lock); | |
1365 | ret = -ENOMEM; | |
1366 | goto out; | |
a2b1693b TH |
1367 | } |
1368 | EXPORT_SYMBOL_GPL(blkcg_activate_policy); | |
1369 | ||
1370 | /** | |
1371 | * blkcg_deactivate_policy - deactivate a blkcg policy on a request_queue | |
1372 | * @q: request_queue of interest | |
1373 | * @pol: blkcg policy to deactivate | |
1374 | * | |
1375 | * Deactivate @pol on @q. Follows the same synchronization rules as | |
1376 | * blkcg_activate_policy(). | |
1377 | */ | |
1378 | void blkcg_deactivate_policy(struct request_queue *q, | |
3c798398 | 1379 | const struct blkcg_policy *pol) |
a2b1693b | 1380 | { |
3c798398 | 1381 | struct blkcg_gq *blkg; |
a2b1693b TH |
1382 | |
1383 | if (!blkcg_policy_enabled(q, pol)) | |
1384 | return; | |
1385 | ||
344e9ffc | 1386 | if (queue_is_mq(q)) |
bd166ef1 | 1387 | blk_mq_freeze_queue(q); |
bd166ef1 | 1388 | |
0d945c1f | 1389 | spin_lock_irq(&q->queue_lock); |
a2b1693b TH |
1390 | |
1391 | __clear_bit(pol->plid, q->blkcg_pols); | |
1392 | ||
1393 | list_for_each_entry(blkg, &q->blkg_list, q_node) { | |
001bea73 | 1394 | if (blkg->pd[pol->plid]) { |
6b065462 | 1395 | if (pol->pd_offline_fn) |
a9520cd6 | 1396 | pol->pd_offline_fn(blkg->pd[pol->plid]); |
001bea73 TH |
1397 | pol->pd_free_fn(blkg->pd[pol->plid]); |
1398 | blkg->pd[pol->plid] = NULL; | |
1399 | } | |
a2b1693b TH |
1400 | } |
1401 | ||
0d945c1f | 1402 | spin_unlock_irq(&q->queue_lock); |
bd166ef1 | 1403 | |
344e9ffc | 1404 | if (queue_is_mq(q)) |
bd166ef1 | 1405 | blk_mq_unfreeze_queue(q); |
a2b1693b TH |
1406 | } |
1407 | EXPORT_SYMBOL_GPL(blkcg_deactivate_policy); | |
1408 | ||
8bd435b3 | 1409 | /** |
3c798398 TH |
1410 | * blkcg_policy_register - register a blkcg policy |
1411 | * @pol: blkcg policy to register | |
8bd435b3 | 1412 | * |
3c798398 TH |
1413 | * Register @pol with blkcg core. Might sleep and @pol may be modified on |
1414 | * successful registration. Returns 0 on success and -errno on failure. | |
8bd435b3 | 1415 | */ |
d5bf0291 | 1416 | int blkcg_policy_register(struct blkcg_policy *pol) |
3e252066 | 1417 | { |
06b285bd | 1418 | struct blkcg *blkcg; |
8bd435b3 | 1419 | int i, ret; |
e8989fae | 1420 | |
838f13bf | 1421 | mutex_lock(&blkcg_pol_register_mutex); |
bc0d6501 TH |
1422 | mutex_lock(&blkcg_pol_mutex); |
1423 | ||
8bd435b3 TH |
1424 | /* find an empty slot */ |
1425 | ret = -ENOSPC; | |
1426 | for (i = 0; i < BLKCG_MAX_POLS; i++) | |
3c798398 | 1427 | if (!blkcg_policy[i]) |
8bd435b3 | 1428 | break; |
01c5f85a JA |
1429 | if (i >= BLKCG_MAX_POLS) { |
1430 | pr_warn("blkcg_policy_register: BLKCG_MAX_POLS too small\n"); | |
838f13bf | 1431 | goto err_unlock; |
01c5f85a | 1432 | } |
035d10b2 | 1433 | |
e8401073 | 1434 | /* Make sure cpd/pd_alloc_fn and cpd/pd_free_fn in pairs */ |
1435 | if ((!pol->cpd_alloc_fn ^ !pol->cpd_free_fn) || | |
1436 | (!pol->pd_alloc_fn ^ !pol->pd_free_fn)) | |
1437 | goto err_unlock; | |
1438 | ||
06b285bd | 1439 | /* register @pol */ |
3c798398 | 1440 | pol->plid = i; |
06b285bd TH |
1441 | blkcg_policy[pol->plid] = pol; |
1442 | ||
1443 | /* allocate and install cpd's */ | |
e4a9bde9 | 1444 | if (pol->cpd_alloc_fn) { |
06b285bd TH |
1445 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) { |
1446 | struct blkcg_policy_data *cpd; | |
1447 | ||
e4a9bde9 | 1448 | cpd = pol->cpd_alloc_fn(GFP_KERNEL); |
bbb427e3 | 1449 | if (!cpd) |
06b285bd | 1450 | goto err_free_cpds; |
06b285bd | 1451 | |
81437648 TH |
1452 | blkcg->cpd[pol->plid] = cpd; |
1453 | cpd->blkcg = blkcg; | |
06b285bd | 1454 | cpd->plid = pol->plid; |
86a5bba5 TH |
1455 | if (pol->cpd_init_fn) |
1456 | pol->cpd_init_fn(cpd); | |
06b285bd TH |
1457 | } |
1458 | } | |
1459 | ||
838f13bf | 1460 | mutex_unlock(&blkcg_pol_mutex); |
8bd435b3 | 1461 | |
8bd435b3 | 1462 | /* everything is in place, add intf files for the new policy */ |
2ee867dc TH |
1463 | if (pol->dfl_cftypes) |
1464 | WARN_ON(cgroup_add_dfl_cftypes(&io_cgrp_subsys, | |
1465 | pol->dfl_cftypes)); | |
880f50e2 | 1466 | if (pol->legacy_cftypes) |
c165b3e3 | 1467 | WARN_ON(cgroup_add_legacy_cftypes(&io_cgrp_subsys, |
880f50e2 | 1468 | pol->legacy_cftypes)); |
838f13bf TH |
1469 | mutex_unlock(&blkcg_pol_register_mutex); |
1470 | return 0; | |
1471 | ||
06b285bd | 1472 | err_free_cpds: |
58a9edce | 1473 | if (pol->cpd_free_fn) { |
06b285bd | 1474 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) { |
e4a9bde9 TH |
1475 | if (blkcg->cpd[pol->plid]) { |
1476 | pol->cpd_free_fn(blkcg->cpd[pol->plid]); | |
1477 | blkcg->cpd[pol->plid] = NULL; | |
1478 | } | |
06b285bd TH |
1479 | } |
1480 | } | |
1481 | blkcg_policy[pol->plid] = NULL; | |
838f13bf | 1482 | err_unlock: |
bc0d6501 | 1483 | mutex_unlock(&blkcg_pol_mutex); |
838f13bf | 1484 | mutex_unlock(&blkcg_pol_register_mutex); |
8bd435b3 | 1485 | return ret; |
3e252066 | 1486 | } |
3c798398 | 1487 | EXPORT_SYMBOL_GPL(blkcg_policy_register); |
3e252066 | 1488 | |
8bd435b3 | 1489 | /** |
3c798398 TH |
1490 | * blkcg_policy_unregister - unregister a blkcg policy |
1491 | * @pol: blkcg policy to unregister | |
8bd435b3 | 1492 | * |
3c798398 | 1493 | * Undo blkcg_policy_register(@pol). Might sleep. |
8bd435b3 | 1494 | */ |
3c798398 | 1495 | void blkcg_policy_unregister(struct blkcg_policy *pol) |
3e252066 | 1496 | { |
06b285bd TH |
1497 | struct blkcg *blkcg; |
1498 | ||
838f13bf | 1499 | mutex_lock(&blkcg_pol_register_mutex); |
bc0d6501 | 1500 | |
3c798398 | 1501 | if (WARN_ON(blkcg_policy[pol->plid] != pol)) |
8bd435b3 TH |
1502 | goto out_unlock; |
1503 | ||
1504 | /* kill the intf files first */ | |
2ee867dc TH |
1505 | if (pol->dfl_cftypes) |
1506 | cgroup_rm_cftypes(pol->dfl_cftypes); | |
880f50e2 TH |
1507 | if (pol->legacy_cftypes) |
1508 | cgroup_rm_cftypes(pol->legacy_cftypes); | |
44ea53de | 1509 | |
06b285bd | 1510 | /* remove cpds and unregister */ |
838f13bf | 1511 | mutex_lock(&blkcg_pol_mutex); |
06b285bd | 1512 | |
58a9edce | 1513 | if (pol->cpd_free_fn) { |
06b285bd | 1514 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) { |
e4a9bde9 TH |
1515 | if (blkcg->cpd[pol->plid]) { |
1516 | pol->cpd_free_fn(blkcg->cpd[pol->plid]); | |
1517 | blkcg->cpd[pol->plid] = NULL; | |
1518 | } | |
06b285bd TH |
1519 | } |
1520 | } | |
3c798398 | 1521 | blkcg_policy[pol->plid] = NULL; |
06b285bd | 1522 | |
bc0d6501 | 1523 | mutex_unlock(&blkcg_pol_mutex); |
838f13bf TH |
1524 | out_unlock: |
1525 | mutex_unlock(&blkcg_pol_register_mutex); | |
3e252066 | 1526 | } |
3c798398 | 1527 | EXPORT_SYMBOL_GPL(blkcg_policy_unregister); |
903d23f0 | 1528 | |
d3f77dfd TH |
1529 | bool __blkcg_punt_bio_submit(struct bio *bio) |
1530 | { | |
1531 | struct blkcg_gq *blkg = bio->bi_blkg; | |
1532 | ||
1533 | /* consume the flag first */ | |
1534 | bio->bi_opf &= ~REQ_CGROUP_PUNT; | |
1535 | ||
1536 | /* never bounce for the root cgroup */ | |
1537 | if (!blkg->parent) | |
1538 | return false; | |
1539 | ||
1540 | spin_lock_bh(&blkg->async_bio_lock); | |
1541 | bio_list_add(&blkg->async_bios, bio); | |
1542 | spin_unlock_bh(&blkg->async_bio_lock); | |
1543 | ||
1544 | queue_work(blkcg_punt_bio_wq, &blkg->async_bio_work); | |
1545 | return true; | |
1546 | } | |
1547 | ||
d09d8df3 JB |
1548 | /* |
1549 | * Scale the accumulated delay based on how long it has been since we updated | |
1550 | * the delay. We only call this when we are adding delay, in case it's been a | |
1551 | * while since we added delay, and when we are checking to see if we need to | |
1552 | * delay a task, to account for any delays that may have occurred. | |
1553 | */ | |
1554 | static void blkcg_scale_delay(struct blkcg_gq *blkg, u64 now) | |
1555 | { | |
1556 | u64 old = atomic64_read(&blkg->delay_start); | |
1557 | ||
54c52e10 TH |
1558 | /* negative use_delay means no scaling, see blkcg_set_delay() */ |
1559 | if (atomic_read(&blkg->use_delay) < 0) | |
1560 | return; | |
1561 | ||
d09d8df3 JB |
1562 | /* |
1563 | * We only want to scale down every second. The idea here is that we | |
1564 | * want to delay people for min(delay_nsec, NSEC_PER_SEC) in a certain | |
1565 | * time window. We only want to throttle tasks for recent delay that | |
1566 | * has occurred, in 1 second time windows since that's the maximum | |
1567 | * things can be throttled. We save the current delay window in | |
1568 | * blkg->last_delay so we know what amount is still left to be charged | |
1569 | * to the blkg from this point onward. blkg->last_use keeps track of | |
1570 | * the use_delay counter. The idea is if we're unthrottling the blkg we | |
1571 | * are ok with whatever is happening now, and we can take away more of | |
1572 | * the accumulated delay as we've already throttled enough that | |
1573 | * everybody is happy with their IO latencies. | |
1574 | */ | |
1575 | if (time_before64(old + NSEC_PER_SEC, now) && | |
1576 | atomic64_cmpxchg(&blkg->delay_start, old, now) == old) { | |
1577 | u64 cur = atomic64_read(&blkg->delay_nsec); | |
1578 | u64 sub = min_t(u64, blkg->last_delay, now - old); | |
1579 | int cur_use = atomic_read(&blkg->use_delay); | |
1580 | ||
1581 | /* | |
1582 | * We've been unthrottled, subtract a larger chunk of our | |
1583 | * accumulated delay. | |
1584 | */ | |
1585 | if (cur_use < blkg->last_use) | |
1586 | sub = max_t(u64, sub, blkg->last_delay >> 1); | |
1587 | ||
1588 | /* | |
1589 | * This shouldn't happen, but handle it anyway. Our delay_nsec | |
1590 | * should only ever be growing except here where we subtract out | |
1591 | * min(last_delay, 1 second), but lord knows bugs happen and I'd | |
1592 | * rather not end up with negative numbers. | |
1593 | */ | |
1594 | if (unlikely(cur < sub)) { | |
1595 | atomic64_set(&blkg->delay_nsec, 0); | |
1596 | blkg->last_delay = 0; | |
1597 | } else { | |
1598 | atomic64_sub(sub, &blkg->delay_nsec); | |
1599 | blkg->last_delay = cur - sub; | |
1600 | } | |
1601 | blkg->last_use = cur_use; | |
1602 | } | |
1603 | } | |
1604 | ||
1605 | /* | |
1606 | * This is called when we want to actually walk up the hierarchy and check to | |
1607 | * see if we need to throttle, and then actually throttle if there is some | |
1608 | * accumulated delay. This should only be called upon return to user space so | |
1609 | * we're not holding some lock that would induce a priority inversion. | |
1610 | */ | |
1611 | static void blkcg_maybe_throttle_blkg(struct blkcg_gq *blkg, bool use_memdelay) | |
1612 | { | |
fd112c74 | 1613 | unsigned long pflags; |
d09d8df3 JB |
1614 | u64 now = ktime_to_ns(ktime_get()); |
1615 | u64 exp; | |
1616 | u64 delay_nsec = 0; | |
1617 | int tok; | |
1618 | ||
1619 | while (blkg->parent) { | |
1620 | if (atomic_read(&blkg->use_delay)) { | |
1621 | blkcg_scale_delay(blkg, now); | |
1622 | delay_nsec = max_t(u64, delay_nsec, | |
1623 | atomic64_read(&blkg->delay_nsec)); | |
1624 | } | |
1625 | blkg = blkg->parent; | |
1626 | } | |
1627 | ||
1628 | if (!delay_nsec) | |
1629 | return; | |
1630 | ||
1631 | /* | |
1632 | * Let's not sleep for all eternity if we've amassed a huge delay. | |
1633 | * Swapping or metadata IO can accumulate 10's of seconds worth of | |
1634 | * delay, and we want userspace to be able to do _something_ so cap the | |
1635 | * delays at 1 second. If there's 10's of seconds worth of delay then | |
1636 | * the tasks will be delayed for 1 second for every syscall. | |
1637 | */ | |
1638 | delay_nsec = min_t(u64, delay_nsec, 250 * NSEC_PER_MSEC); | |
1639 | ||
fd112c74 JB |
1640 | if (use_memdelay) |
1641 | psi_memstall_enter(&pflags); | |
d09d8df3 JB |
1642 | |
1643 | exp = ktime_add_ns(now, delay_nsec); | |
1644 | tok = io_schedule_prepare(); | |
1645 | do { | |
1646 | __set_current_state(TASK_KILLABLE); | |
1647 | if (!schedule_hrtimeout(&exp, HRTIMER_MODE_ABS)) | |
1648 | break; | |
1649 | } while (!fatal_signal_pending(current)); | |
1650 | io_schedule_finish(tok); | |
fd112c74 JB |
1651 | |
1652 | if (use_memdelay) | |
1653 | psi_memstall_leave(&pflags); | |
d09d8df3 JB |
1654 | } |
1655 | ||
1656 | /** | |
1657 | * blkcg_maybe_throttle_current - throttle the current task if it has been marked | |
1658 | * | |
1659 | * This is only called if we've been marked with set_notify_resume(). Obviously | |
1660 | * we can be set_notify_resume() for reasons other than blkcg throttling, so we | |
1661 | * check to see if current->throttle_queue is set and if not this doesn't do | |
1662 | * anything. This should only ever be called by the resume code, it's not meant | |
1663 | * to be called by people willy-nilly as it will actually do the work to | |
1664 | * throttle the task if it is setup for throttling. | |
1665 | */ | |
1666 | void blkcg_maybe_throttle_current(void) | |
1667 | { | |
1668 | struct request_queue *q = current->throttle_queue; | |
1669 | struct cgroup_subsys_state *css; | |
1670 | struct blkcg *blkcg; | |
1671 | struct blkcg_gq *blkg; | |
1672 | bool use_memdelay = current->use_memdelay; | |
1673 | ||
1674 | if (!q) | |
1675 | return; | |
1676 | ||
1677 | current->throttle_queue = NULL; | |
1678 | current->use_memdelay = false; | |
1679 | ||
1680 | rcu_read_lock(); | |
1681 | css = kthread_blkcg(); | |
1682 | if (css) | |
1683 | blkcg = css_to_blkcg(css); | |
1684 | else | |
1685 | blkcg = css_to_blkcg(task_css(current, io_cgrp_id)); | |
1686 | ||
1687 | if (!blkcg) | |
1688 | goto out; | |
1689 | blkg = blkg_lookup(blkcg, q); | |
1690 | if (!blkg) | |
1691 | goto out; | |
7754f669 | 1692 | if (!blkg_tryget(blkg)) |
d09d8df3 JB |
1693 | goto out; |
1694 | rcu_read_unlock(); | |
d09d8df3 JB |
1695 | |
1696 | blkcg_maybe_throttle_blkg(blkg, use_memdelay); | |
1697 | blkg_put(blkg); | |
cc7ecc25 | 1698 | blk_put_queue(q); |
d09d8df3 JB |
1699 | return; |
1700 | out: | |
1701 | rcu_read_unlock(); | |
1702 | blk_put_queue(q); | |
1703 | } | |
d09d8df3 JB |
1704 | |
1705 | /** | |
1706 | * blkcg_schedule_throttle - this task needs to check for throttling | |
537d71b3 BVA |
1707 | * @q: the request queue IO was submitted on |
1708 | * @use_memdelay: do we charge this to memory delay for PSI | |
d09d8df3 JB |
1709 | * |
1710 | * This is called by the IO controller when we know there's delay accumulated | |
1711 | * for the blkg for this task. We do not pass the blkg because there are places | |
1712 | * we call this that may not have that information, the swapping code for | |
1713 | * instance will only have a request_queue at that point. This set's the | |
1714 | * notify_resume for the task to check and see if it requires throttling before | |
1715 | * returning to user space. | |
1716 | * | |
1717 | * We will only schedule once per syscall. You can call this over and over | |
1718 | * again and it will only do the check once upon return to user space, and only | |
1719 | * throttle once. If the task needs to be throttled again it'll need to be | |
1720 | * re-set at the next time we see the task. | |
1721 | */ | |
1722 | void blkcg_schedule_throttle(struct request_queue *q, bool use_memdelay) | |
1723 | { | |
1724 | if (unlikely(current->flags & PF_KTHREAD)) | |
1725 | return; | |
1726 | ||
1727 | if (!blk_get_queue(q)) | |
1728 | return; | |
1729 | ||
1730 | if (current->throttle_queue) | |
1731 | blk_put_queue(current->throttle_queue); | |
1732 | current->throttle_queue = q; | |
1733 | if (use_memdelay) | |
1734 | current->use_memdelay = use_memdelay; | |
1735 | set_notify_resume(current); | |
1736 | } | |
d09d8df3 JB |
1737 | |
1738 | /** | |
1739 | * blkcg_add_delay - add delay to this blkg | |
537d71b3 BVA |
1740 | * @blkg: blkg of interest |
1741 | * @now: the current time in nanoseconds | |
1742 | * @delta: how many nanoseconds of delay to add | |
d09d8df3 JB |
1743 | * |
1744 | * Charge @delta to the blkg's current delay accumulation. This is used to | |
1745 | * throttle tasks if an IO controller thinks we need more throttling. | |
1746 | */ | |
1747 | void blkcg_add_delay(struct blkcg_gq *blkg, u64 now, u64 delta) | |
1748 | { | |
54c52e10 TH |
1749 | if (WARN_ON_ONCE(atomic_read(&blkg->use_delay) < 0)) |
1750 | return; | |
d09d8df3 JB |
1751 | blkcg_scale_delay(blkg, now); |
1752 | atomic64_add(delta, &blkg->delay_nsec); | |
1753 | } | |
d09d8df3 | 1754 | |
28fc591f CH |
1755 | /** |
1756 | * blkg_tryget_closest - try and get a blkg ref on the closet blkg | |
13c7863d CH |
1757 | * @bio: target bio |
1758 | * @css: target css | |
28fc591f | 1759 | * |
13c7863d CH |
1760 | * As the failure mode here is to walk up the blkg tree, this ensure that the |
1761 | * blkg->parent pointers are always valid. This returns the blkg that it ended | |
1762 | * up taking a reference on or %NULL if no reference was taken. | |
28fc591f | 1763 | */ |
13c7863d CH |
1764 | static inline struct blkcg_gq *blkg_tryget_closest(struct bio *bio, |
1765 | struct cgroup_subsys_state *css) | |
28fc591f | 1766 | { |
13c7863d | 1767 | struct blkcg_gq *blkg, *ret_blkg = NULL; |
28fc591f | 1768 | |
13c7863d CH |
1769 | rcu_read_lock(); |
1770 | blkg = blkg_lookup_create(css_to_blkcg(css), bio->bi_disk->queue); | |
28fc591f CH |
1771 | while (blkg) { |
1772 | if (blkg_tryget(blkg)) { | |
1773 | ret_blkg = blkg; | |
1774 | break; | |
1775 | } | |
1776 | blkg = blkg->parent; | |
1777 | } | |
13c7863d | 1778 | rcu_read_unlock(); |
28fc591f CH |
1779 | |
1780 | return ret_blkg; | |
1781 | } | |
1782 | ||
1783 | /** | |
1784 | * bio_associate_blkg_from_css - associate a bio with a specified css | |
1785 | * @bio: target bio | |
1786 | * @css: target css | |
1787 | * | |
1788 | * Associate @bio with the blkg found by combining the css's blkg and the | |
1789 | * request_queue of the @bio. An association failure is handled by walking up | |
1790 | * the blkg tree. Therefore, the blkg associated can be anything between @blkg | |
1791 | * and q->root_blkg. This situation only happens when a cgroup is dying and | |
1792 | * then the remaining bios will spill to the closest alive blkg. | |
1793 | * | |
1794 | * A reference will be taken on the blkg and will be released when @bio is | |
1795 | * freed. | |
1796 | */ | |
1797 | void bio_associate_blkg_from_css(struct bio *bio, | |
1798 | struct cgroup_subsys_state *css) | |
1799 | { | |
28fc591f CH |
1800 | if (bio->bi_blkg) |
1801 | blkg_put(bio->bi_blkg); | |
1802 | ||
a5b97526 | 1803 | if (css && css->parent) { |
13c7863d | 1804 | bio->bi_blkg = blkg_tryget_closest(bio, css); |
a5b97526 | 1805 | } else { |
13c7863d CH |
1806 | blkg_get(bio->bi_disk->queue->root_blkg); |
1807 | bio->bi_blkg = bio->bi_disk->queue->root_blkg; | |
a5b97526 | 1808 | } |
28fc591f CH |
1809 | } |
1810 | EXPORT_SYMBOL_GPL(bio_associate_blkg_from_css); | |
1811 | ||
1812 | /** | |
1813 | * bio_associate_blkg - associate a bio with a blkg | |
1814 | * @bio: target bio | |
1815 | * | |
1816 | * Associate @bio with the blkg found from the bio's css and request_queue. | |
1817 | * If one is not found, bio_lookup_blkg() creates the blkg. If a blkg is | |
1818 | * already associated, the css is reused and association redone as the | |
1819 | * request_queue may have changed. | |
1820 | */ | |
1821 | void bio_associate_blkg(struct bio *bio) | |
1822 | { | |
1823 | struct cgroup_subsys_state *css; | |
1824 | ||
1825 | rcu_read_lock(); | |
1826 | ||
1827 | if (bio->bi_blkg) | |
1828 | css = &bio_blkcg(bio)->css; | |
1829 | else | |
1830 | css = blkcg_css(); | |
1831 | ||
1832 | bio_associate_blkg_from_css(bio, css); | |
1833 | ||
1834 | rcu_read_unlock(); | |
1835 | } | |
1836 | EXPORT_SYMBOL_GPL(bio_associate_blkg); | |
1837 | ||
1838 | /** | |
1839 | * bio_clone_blkg_association - clone blkg association from src to dst bio | |
1840 | * @dst: destination bio | |
1841 | * @src: source bio | |
1842 | */ | |
1843 | void bio_clone_blkg_association(struct bio *dst, struct bio *src) | |
1844 | { | |
1845 | if (src->bi_blkg) { | |
1846 | if (dst->bi_blkg) | |
1847 | blkg_put(dst->bi_blkg); | |
1848 | blkg_get(src->bi_blkg); | |
1849 | dst->bi_blkg = src->bi_blkg; | |
1850 | } | |
1851 | } | |
1852 | EXPORT_SYMBOL_GPL(bio_clone_blkg_association); | |
1853 | ||
db18a53e CH |
1854 | static int blk_cgroup_io_type(struct bio *bio) |
1855 | { | |
1856 | if (op_is_discard(bio->bi_opf)) | |
1857 | return BLKG_IOSTAT_DISCARD; | |
1858 | if (op_is_write(bio->bi_opf)) | |
1859 | return BLKG_IOSTAT_WRITE; | |
1860 | return BLKG_IOSTAT_READ; | |
1861 | } | |
1862 | ||
1863 | void blk_cgroup_bio_start(struct bio *bio) | |
1864 | { | |
1865 | int rwd = blk_cgroup_io_type(bio), cpu; | |
1866 | struct blkg_iostat_set *bis; | |
1867 | ||
1868 | cpu = get_cpu(); | |
1869 | bis = per_cpu_ptr(bio->bi_blkg->iostat_cpu, cpu); | |
1870 | u64_stats_update_begin(&bis->sync); | |
1871 | ||
1872 | /* | |
1873 | * If the bio is flagged with BIO_CGROUP_ACCT it means this is a split | |
1874 | * bio and we would have already accounted for the size of the bio. | |
1875 | */ | |
1876 | if (!bio_flagged(bio, BIO_CGROUP_ACCT)) { | |
1877 | bio_set_flag(bio, BIO_CGROUP_ACCT); | |
0b8cc25d | 1878 | bis->cur.bytes[rwd] += bio->bi_iter.bi_size; |
db18a53e CH |
1879 | } |
1880 | bis->cur.ios[rwd]++; | |
1881 | ||
1882 | u64_stats_update_end(&bis->sync); | |
1883 | if (cgroup_subsys_on_dfl(io_cgrp_subsys)) | |
1884 | cgroup_rstat_updated(bio->bi_blkg->blkcg->css.cgroup, cpu); | |
1885 | put_cpu(); | |
1886 | } | |
1887 | ||
d3f77dfd TH |
1888 | static int __init blkcg_init(void) |
1889 | { | |
1890 | blkcg_punt_bio_wq = alloc_workqueue("blkcg_punt_bio", | |
1891 | WQ_MEM_RECLAIM | WQ_FREEZABLE | | |
1892 | WQ_UNBOUND | WQ_SYSFS, 0); | |
1893 | if (!blkcg_punt_bio_wq) | |
1894 | return -ENOMEM; | |
1895 | return 0; | |
1896 | } | |
1897 | subsys_initcall(blkcg_init); | |
1898 | ||
903d23f0 JB |
1899 | module_param(blkcg_debug_stats, bool, 0644); |
1900 | MODULE_PARM_DESC(blkcg_debug_stats, "True if you want debug stats, false if not"); |