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 | |
838f13bf TH |
35 | /* |
36 | * blkcg_pol_mutex protects blkcg_policy[] and policy [de]activation. | |
37 | * blkcg_pol_register_mutex nests outside of it and synchronizes entire | |
38 | * policy [un]register operations including cgroup file additions / | |
39 | * removals. Putting cgroup file registration outside blkcg_pol_mutex | |
40 | * allows grabbing it from cgroup callbacks. | |
41 | */ | |
42 | static DEFINE_MUTEX(blkcg_pol_register_mutex); | |
bc0d6501 | 43 | static DEFINE_MUTEX(blkcg_pol_mutex); |
923adde1 | 44 | |
e48453c3 | 45 | struct blkcg blkcg_root; |
3c798398 | 46 | EXPORT_SYMBOL_GPL(blkcg_root); |
9d6a986c | 47 | |
496d5e75 | 48 | struct cgroup_subsys_state * const blkcg_root_css = &blkcg_root.css; |
9b0eb69b | 49 | EXPORT_SYMBOL_GPL(blkcg_root_css); |
496d5e75 | 50 | |
3c798398 | 51 | static struct blkcg_policy *blkcg_policy[BLKCG_MAX_POLS]; |
035d10b2 | 52 | |
7876f930 TH |
53 | static LIST_HEAD(all_blkcgs); /* protected by blkcg_pol_mutex */ |
54 | ||
07b0fdec | 55 | bool blkcg_debug_stats = false; |
d3f77dfd | 56 | static struct workqueue_struct *blkcg_punt_bio_wq; |
903d23f0 | 57 | |
a2b1693b | 58 | static bool blkcg_policy_enabled(struct request_queue *q, |
3c798398 | 59 | const struct blkcg_policy *pol) |
a2b1693b TH |
60 | { |
61 | return pol && test_bit(pol->plid, q->blkcg_pols); | |
62 | } | |
63 | ||
0381411e TH |
64 | /** |
65 | * blkg_free - free a blkg | |
66 | * @blkg: blkg to free | |
67 | * | |
68 | * Free @blkg which may be partially allocated. | |
69 | */ | |
3c798398 | 70 | static void blkg_free(struct blkcg_gq *blkg) |
0381411e | 71 | { |
e8989fae | 72 | int i; |
549d3aa8 TH |
73 | |
74 | if (!blkg) | |
75 | return; | |
76 | ||
db613670 | 77 | for (i = 0; i < BLKCG_MAX_POLS; i++) |
001bea73 TH |
78 | if (blkg->pd[i]) |
79 | blkcg_policy[i]->pd_free_fn(blkg->pd[i]); | |
e8989fae | 80 | |
f7331648 | 81 | free_percpu(blkg->iostat_cpu); |
ef069b97 | 82 | percpu_ref_exit(&blkg->refcnt); |
549d3aa8 | 83 | kfree(blkg); |
0381411e TH |
84 | } |
85 | ||
7fcf2b03 DZ |
86 | static void __blkg_release(struct rcu_head *rcu) |
87 | { | |
88 | struct blkcg_gq *blkg = container_of(rcu, struct blkcg_gq, rcu_head); | |
89 | ||
d3f77dfd TH |
90 | WARN_ON(!bio_list_empty(&blkg->async_bios)); |
91 | ||
7fcf2b03 DZ |
92 | /* release the blkcg and parent blkg refs this blkg has been holding */ |
93 | css_put(&blkg->blkcg->css); | |
94 | if (blkg->parent) | |
95 | blkg_put(blkg->parent); | |
7fcf2b03 DZ |
96 | blkg_free(blkg); |
97 | } | |
98 | ||
99 | /* | |
100 | * A group is RCU protected, but having an rcu lock does not mean that one | |
101 | * can access all the fields of blkg and assume these are valid. For | |
102 | * example, don't try to follow throtl_data and request queue links. | |
103 | * | |
104 | * Having a reference to blkg under an rcu allows accesses to only values | |
105 | * local to groups like group stats and group rate limits. | |
106 | */ | |
107 | static void blkg_release(struct percpu_ref *ref) | |
108 | { | |
109 | struct blkcg_gq *blkg = container_of(ref, struct blkcg_gq, refcnt); | |
110 | ||
111 | call_rcu(&blkg->rcu_head, __blkg_release); | |
112 | } | |
113 | ||
d3f77dfd TH |
114 | static void blkg_async_bio_workfn(struct work_struct *work) |
115 | { | |
116 | struct blkcg_gq *blkg = container_of(work, struct blkcg_gq, | |
117 | async_bio_work); | |
118 | struct bio_list bios = BIO_EMPTY_LIST; | |
119 | struct bio *bio; | |
192f1c6b XT |
120 | struct blk_plug plug; |
121 | bool need_plug = false; | |
d3f77dfd TH |
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 | ||
192f1c6b XT |
129 | /* start plug only when bio_list contains at least 2 bios */ |
130 | if (bios.head && bios.head->bi_next) { | |
131 | need_plug = true; | |
132 | blk_start_plug(&plug); | |
133 | } | |
d3f77dfd TH |
134 | while ((bio = bio_list_pop(&bios))) |
135 | submit_bio(bio); | |
192f1c6b XT |
136 | if (need_plug) |
137 | blk_finish_plug(&plug); | |
d3f77dfd TH |
138 | } |
139 | ||
0381411e TH |
140 | /** |
141 | * blkg_alloc - allocate a blkg | |
142 | * @blkcg: block cgroup the new blkg is associated with | |
143 | * @q: request_queue the new blkg is associated with | |
15974993 | 144 | * @gfp_mask: allocation mask to use |
0381411e | 145 | * |
e8989fae | 146 | * Allocate a new blkg assocating @blkcg and @q. |
0381411e | 147 | */ |
15974993 TH |
148 | static struct blkcg_gq *blkg_alloc(struct blkcg *blkcg, struct request_queue *q, |
149 | gfp_t gfp_mask) | |
0381411e | 150 | { |
3c798398 | 151 | struct blkcg_gq *blkg; |
f7331648 | 152 | int i, cpu; |
0381411e TH |
153 | |
154 | /* alloc and init base part */ | |
15974993 | 155 | blkg = kzalloc_node(sizeof(*blkg), gfp_mask, q->node); |
0381411e TH |
156 | if (!blkg) |
157 | return NULL; | |
158 | ||
ef069b97 TH |
159 | if (percpu_ref_init(&blkg->refcnt, blkg_release, 0, gfp_mask)) |
160 | goto err_free; | |
161 | ||
f7331648 TH |
162 | blkg->iostat_cpu = alloc_percpu_gfp(struct blkg_iostat_set, gfp_mask); |
163 | if (!blkg->iostat_cpu) | |
77ea7338 TH |
164 | goto err_free; |
165 | ||
c875f4d0 | 166 | blkg->q = q; |
e8989fae | 167 | INIT_LIST_HEAD(&blkg->q_node); |
d3f77dfd TH |
168 | spin_lock_init(&blkg->async_bio_lock); |
169 | bio_list_init(&blkg->async_bios); | |
170 | INIT_WORK(&blkg->async_bio_work, blkg_async_bio_workfn); | |
0381411e | 171 | blkg->blkcg = blkcg; |
0381411e | 172 | |
f7331648 TH |
173 | u64_stats_init(&blkg->iostat.sync); |
174 | for_each_possible_cpu(cpu) | |
175 | u64_stats_init(&per_cpu_ptr(blkg->iostat_cpu, cpu)->sync); | |
176 | ||
8bd435b3 | 177 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
3c798398 | 178 | struct blkcg_policy *pol = blkcg_policy[i]; |
e8989fae | 179 | struct blkg_policy_data *pd; |
0381411e | 180 | |
a2b1693b | 181 | if (!blkcg_policy_enabled(q, pol)) |
e8989fae TH |
182 | continue; |
183 | ||
184 | /* alloc per-policy data and attach it to blkg */ | |
cf09a8ee | 185 | pd = pol->pd_alloc_fn(gfp_mask, q, blkcg); |
a051661c TH |
186 | if (!pd) |
187 | goto err_free; | |
549d3aa8 | 188 | |
e8989fae TH |
189 | blkg->pd[i] = pd; |
190 | pd->blkg = blkg; | |
b276a876 | 191 | pd->plid = i; |
e8989fae TH |
192 | } |
193 | ||
0381411e | 194 | return blkg; |
a051661c TH |
195 | |
196 | err_free: | |
197 | blkg_free(blkg); | |
198 | return NULL; | |
0381411e TH |
199 | } |
200 | ||
24f29046 TH |
201 | struct blkcg_gq *blkg_lookup_slowpath(struct blkcg *blkcg, |
202 | struct request_queue *q, bool update_hint) | |
80fd9979 | 203 | { |
3c798398 | 204 | struct blkcg_gq *blkg; |
80fd9979 | 205 | |
a637120e | 206 | /* |
86cde6b6 TH |
207 | * Hint didn't match. Look up from the radix tree. Note that the |
208 | * hint can only be updated under queue_lock as otherwise @blkg | |
209 | * could have already been removed from blkg_tree. The caller is | |
210 | * responsible for grabbing queue_lock if @update_hint. | |
a637120e TH |
211 | */ |
212 | blkg = radix_tree_lookup(&blkcg->blkg_tree, q->id); | |
86cde6b6 TH |
213 | if (blkg && blkg->q == q) { |
214 | if (update_hint) { | |
0d945c1f | 215 | lockdep_assert_held(&q->queue_lock); |
86cde6b6 TH |
216 | rcu_assign_pointer(blkcg->blkg_hint, blkg); |
217 | } | |
a637120e | 218 | return blkg; |
86cde6b6 | 219 | } |
a637120e | 220 | |
80fd9979 TH |
221 | return NULL; |
222 | } | |
ae118896 | 223 | EXPORT_SYMBOL_GPL(blkg_lookup_slowpath); |
80fd9979 | 224 | |
15974993 | 225 | /* |
d708f0d5 JA |
226 | * If @new_blkg is %NULL, this function tries to allocate a new one as |
227 | * necessary using %GFP_NOWAIT. @new_blkg is always consumed on return. | |
15974993 | 228 | */ |
86cde6b6 | 229 | static struct blkcg_gq *blkg_create(struct blkcg *blkcg, |
d708f0d5 JA |
230 | struct request_queue *q, |
231 | struct blkcg_gq *new_blkg) | |
5624a4e4 | 232 | { |
d708f0d5 | 233 | struct blkcg_gq *blkg; |
f427d909 | 234 | int i, ret; |
5624a4e4 | 235 | |
cd1604fa | 236 | WARN_ON_ONCE(!rcu_read_lock_held()); |
0d945c1f | 237 | lockdep_assert_held(&q->queue_lock); |
cd1604fa | 238 | |
0273ac34 DZ |
239 | /* request_queue is dying, do not create/recreate a blkg */ |
240 | if (blk_queue_dying(q)) { | |
241 | ret = -ENODEV; | |
242 | goto err_free_blkg; | |
243 | } | |
244 | ||
7ee9c562 | 245 | /* blkg holds a reference to blkcg */ |
ec903c0c | 246 | if (!css_tryget_online(&blkcg->css)) { |
20386ce0 | 247 | ret = -ENODEV; |
93e6d5d8 | 248 | goto err_free_blkg; |
15974993 | 249 | } |
cd1604fa | 250 | |
d708f0d5 JA |
251 | /* allocate */ |
252 | if (!new_blkg) { | |
253 | new_blkg = blkg_alloc(blkcg, q, GFP_NOWAIT | __GFP_NOWARN); | |
254 | if (unlikely(!new_blkg)) { | |
255 | ret = -ENOMEM; | |
8c911f3d | 256 | goto err_put_css; |
15974993 TH |
257 | } |
258 | } | |
d708f0d5 | 259 | blkg = new_blkg; |
cd1604fa | 260 | |
db613670 | 261 | /* link parent */ |
3c547865 TH |
262 | if (blkcg_parent(blkcg)) { |
263 | blkg->parent = __blkg_lookup(blkcg_parent(blkcg), q, false); | |
264 | if (WARN_ON_ONCE(!blkg->parent)) { | |
20386ce0 | 265 | ret = -ENODEV; |
8c911f3d | 266 | goto err_put_css; |
3c547865 TH |
267 | } |
268 | blkg_get(blkg->parent); | |
269 | } | |
270 | ||
db613670 TH |
271 | /* invoke per-policy init */ |
272 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
273 | struct blkcg_policy *pol = blkcg_policy[i]; | |
274 | ||
275 | if (blkg->pd[i] && pol->pd_init_fn) | |
a9520cd6 | 276 | pol->pd_init_fn(blkg->pd[i]); |
db613670 TH |
277 | } |
278 | ||
279 | /* insert */ | |
cd1604fa | 280 | spin_lock(&blkcg->lock); |
a637120e TH |
281 | ret = radix_tree_insert(&blkcg->blkg_tree, q->id, blkg); |
282 | if (likely(!ret)) { | |
283 | hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list); | |
284 | list_add(&blkg->q_node, &q->blkg_list); | |
f427d909 TH |
285 | |
286 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
287 | struct blkcg_policy *pol = blkcg_policy[i]; | |
288 | ||
289 | if (blkg->pd[i] && pol->pd_online_fn) | |
a9520cd6 | 290 | pol->pd_online_fn(blkg->pd[i]); |
f427d909 | 291 | } |
a637120e | 292 | } |
f427d909 | 293 | blkg->online = true; |
cd1604fa | 294 | spin_unlock(&blkcg->lock); |
496fb780 | 295 | |
ec13b1d6 | 296 | if (!ret) |
a637120e | 297 | return blkg; |
15974993 | 298 | |
3c547865 TH |
299 | /* @blkg failed fully initialized, use the usual release path */ |
300 | blkg_put(blkg); | |
301 | return ERR_PTR(ret); | |
302 | ||
d708f0d5 | 303 | err_put_css: |
496fb780 | 304 | css_put(&blkcg->css); |
93e6d5d8 | 305 | err_free_blkg: |
d708f0d5 | 306 | blkg_free(new_blkg); |
93e6d5d8 | 307 | return ERR_PTR(ret); |
31e4c28d | 308 | } |
3c96cb32 | 309 | |
86cde6b6 | 310 | /** |
8c546287 | 311 | * blkg_lookup_create - lookup blkg, try to create one if not there |
86cde6b6 TH |
312 | * @blkcg: blkcg of interest |
313 | * @q: request_queue of interest | |
314 | * | |
315 | * Lookup blkg for the @blkcg - @q pair. If it doesn't exist, try to | |
3c547865 TH |
316 | * create one. blkg creation is performed recursively from blkcg_root such |
317 | * that all non-root blkg's have access to the parent blkg. This function | |
8c546287 | 318 | * should be called under RCU read lock and takes @q->queue_lock. |
86cde6b6 | 319 | * |
beea9da0 DZ |
320 | * Returns the blkg or the closest blkg if blkg_create() fails as it walks |
321 | * down from root. | |
86cde6b6 | 322 | */ |
8c546287 CH |
323 | static struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg, |
324 | struct request_queue *q) | |
3c96cb32 | 325 | { |
86cde6b6 | 326 | struct blkcg_gq *blkg; |
8c546287 | 327 | unsigned long flags; |
86cde6b6 TH |
328 | |
329 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
86cde6b6 | 330 | |
8c546287 | 331 | blkg = blkg_lookup(blkcg, q); |
86cde6b6 TH |
332 | if (blkg) |
333 | return blkg; | |
334 | ||
8c546287 CH |
335 | spin_lock_irqsave(&q->queue_lock, flags); |
336 | blkg = __blkg_lookup(blkcg, q, true); | |
337 | if (blkg) | |
338 | goto found; | |
339 | ||
3c547865 TH |
340 | /* |
341 | * Create blkgs walking down from blkcg_root to @blkcg, so that all | |
beea9da0 DZ |
342 | * non-root blkgs have access to their parents. Returns the closest |
343 | * blkg to the intended blkg should blkg_create() fail. | |
3c547865 TH |
344 | */ |
345 | while (true) { | |
346 | struct blkcg *pos = blkcg; | |
347 | struct blkcg *parent = blkcg_parent(blkcg); | |
beea9da0 DZ |
348 | struct blkcg_gq *ret_blkg = q->root_blkg; |
349 | ||
350 | while (parent) { | |
351 | blkg = __blkg_lookup(parent, q, false); | |
352 | if (blkg) { | |
353 | /* remember closest blkg */ | |
354 | ret_blkg = blkg; | |
355 | break; | |
356 | } | |
3c547865 TH |
357 | pos = parent; |
358 | parent = blkcg_parent(parent); | |
359 | } | |
360 | ||
d708f0d5 | 361 | blkg = blkg_create(pos, q, NULL); |
8c546287 CH |
362 | if (IS_ERR(blkg)) { |
363 | blkg = ret_blkg; | |
364 | break; | |
365 | } | |
beea9da0 | 366 | if (pos == blkcg) |
8c546287 | 367 | break; |
b978962a DZ |
368 | } |
369 | ||
8c546287 CH |
370 | found: |
371 | spin_unlock_irqrestore(&q->queue_lock, flags); | |
b978962a DZ |
372 | return blkg; |
373 | } | |
374 | ||
3c798398 | 375 | static void blkg_destroy(struct blkcg_gq *blkg) |
03aa264a | 376 | { |
3c798398 | 377 | struct blkcg *blkcg = blkg->blkcg; |
6b065462 | 378 | int i; |
03aa264a | 379 | |
0d945c1f | 380 | lockdep_assert_held(&blkg->q->queue_lock); |
9f13ef67 | 381 | lockdep_assert_held(&blkcg->lock); |
03aa264a TH |
382 | |
383 | /* Something wrong if we are trying to remove same group twice */ | |
e8989fae | 384 | WARN_ON_ONCE(list_empty(&blkg->q_node)); |
9f13ef67 | 385 | WARN_ON_ONCE(hlist_unhashed(&blkg->blkcg_node)); |
a637120e | 386 | |
6b065462 DZF |
387 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
388 | struct blkcg_policy *pol = blkcg_policy[i]; | |
389 | ||
390 | if (blkg->pd[i] && pol->pd_offline_fn) | |
391 | pol->pd_offline_fn(blkg->pd[i]); | |
392 | } | |
393 | ||
f427d909 TH |
394 | blkg->online = false; |
395 | ||
a637120e | 396 | radix_tree_delete(&blkcg->blkg_tree, blkg->q->id); |
e8989fae | 397 | list_del_init(&blkg->q_node); |
9f13ef67 | 398 | hlist_del_init_rcu(&blkg->blkcg_node); |
03aa264a | 399 | |
a637120e TH |
400 | /* |
401 | * Both setting lookup hint to and clearing it from @blkg are done | |
402 | * under queue_lock. If it's not pointing to @blkg now, it never | |
403 | * will. Hint assignment itself can race safely. | |
404 | */ | |
ec6c676a | 405 | if (rcu_access_pointer(blkcg->blkg_hint) == blkg) |
a637120e TH |
406 | rcu_assign_pointer(blkcg->blkg_hint, NULL); |
407 | ||
03aa264a TH |
408 | /* |
409 | * Put the reference taken at the time of creation so that when all | |
410 | * queues are gone, group can be destroyed. | |
411 | */ | |
7fcf2b03 | 412 | percpu_ref_kill(&blkg->refcnt); |
03aa264a TH |
413 | } |
414 | ||
9f13ef67 TH |
415 | /** |
416 | * blkg_destroy_all - destroy all blkgs associated with a request_queue | |
417 | * @q: request_queue of interest | |
9f13ef67 | 418 | * |
3c96cb32 | 419 | * Destroy all blkgs associated with @q. |
9f13ef67 | 420 | */ |
3c96cb32 | 421 | static void blkg_destroy_all(struct request_queue *q) |
72e06c25 | 422 | { |
3c798398 | 423 | struct blkcg_gq *blkg, *n; |
72e06c25 | 424 | |
0d945c1f | 425 | spin_lock_irq(&q->queue_lock); |
9f13ef67 | 426 | list_for_each_entry_safe(blkg, n, &q->blkg_list, q_node) { |
3c798398 | 427 | struct blkcg *blkcg = blkg->blkcg; |
72e06c25 | 428 | |
9f13ef67 TH |
429 | spin_lock(&blkcg->lock); |
430 | blkg_destroy(blkg); | |
431 | spin_unlock(&blkcg->lock); | |
72e06c25 | 432 | } |
6fe810bd TH |
433 | |
434 | q->root_blkg = NULL; | |
0d945c1f | 435 | spin_unlock_irq(&q->queue_lock); |
72e06c25 TH |
436 | } |
437 | ||
182446d0 TH |
438 | static int blkcg_reset_stats(struct cgroup_subsys_state *css, |
439 | struct cftype *cftype, u64 val) | |
303a3acb | 440 | { |
182446d0 | 441 | struct blkcg *blkcg = css_to_blkcg(css); |
3c798398 | 442 | struct blkcg_gq *blkg; |
f7331648 | 443 | int i, cpu; |
303a3acb | 444 | |
838f13bf | 445 | mutex_lock(&blkcg_pol_mutex); |
303a3acb | 446 | spin_lock_irq(&blkcg->lock); |
997a026c TH |
447 | |
448 | /* | |
449 | * Note that stat reset is racy - it doesn't synchronize against | |
450 | * stat updates. This is a debug feature which shouldn't exist | |
451 | * anyway. If you get hit by a race, retry. | |
452 | */ | |
b67bfe0d | 453 | hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) { |
f7331648 TH |
454 | for_each_possible_cpu(cpu) { |
455 | struct blkg_iostat_set *bis = | |
456 | per_cpu_ptr(blkg->iostat_cpu, cpu); | |
457 | memset(bis, 0, sizeof(*bis)); | |
458 | } | |
459 | memset(&blkg->iostat, 0, sizeof(blkg->iostat)); | |
77ea7338 | 460 | |
8bd435b3 | 461 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
3c798398 | 462 | struct blkcg_policy *pol = blkcg_policy[i]; |
549d3aa8 | 463 | |
a9520cd6 TH |
464 | if (blkg->pd[i] && pol->pd_reset_stats_fn) |
465 | pol->pd_reset_stats_fn(blkg->pd[i]); | |
bc0d6501 | 466 | } |
303a3acb | 467 | } |
f0bdc8cd | 468 | |
303a3acb | 469 | spin_unlock_irq(&blkcg->lock); |
bc0d6501 | 470 | mutex_unlock(&blkcg_pol_mutex); |
303a3acb DS |
471 | return 0; |
472 | } | |
473 | ||
dd165eb3 | 474 | const char *blkg_dev_name(struct blkcg_gq *blkg) |
303a3acb | 475 | { |
d3d32e69 | 476 | /* some drivers (floppy) instantiate a queue w/o disk registered */ |
dc3b17cc | 477 | if (blkg->q->backing_dev_info->dev) |
d51cfc53 | 478 | return bdi_dev_name(blkg->q->backing_dev_info); |
d3d32e69 | 479 | return NULL; |
303a3acb DS |
480 | } |
481 | ||
d3d32e69 TH |
482 | /** |
483 | * blkcg_print_blkgs - helper for printing per-blkg data | |
484 | * @sf: seq_file to print to | |
485 | * @blkcg: blkcg of interest | |
486 | * @prfill: fill function to print out a blkg | |
487 | * @pol: policy in question | |
488 | * @data: data to be passed to @prfill | |
489 | * @show_total: to print out sum of prfill return values or not | |
490 | * | |
491 | * This function invokes @prfill on each blkg of @blkcg if pd for the | |
492 | * policy specified by @pol exists. @prfill is invoked with @sf, the | |
810ecfa7 TH |
493 | * policy data and @data and the matching queue lock held. If @show_total |
494 | * is %true, the sum of the return values from @prfill is printed with | |
495 | * "Total" label at the end. | |
d3d32e69 TH |
496 | * |
497 | * This is to be used to construct print functions for | |
498 | * cftype->read_seq_string method. | |
499 | */ | |
3c798398 | 500 | void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg, |
f95a04af TH |
501 | u64 (*prfill)(struct seq_file *, |
502 | struct blkg_policy_data *, int), | |
3c798398 | 503 | const struct blkcg_policy *pol, int data, |
ec399347 | 504 | bool show_total) |
5624a4e4 | 505 | { |
3c798398 | 506 | struct blkcg_gq *blkg; |
d3d32e69 | 507 | u64 total = 0; |
5624a4e4 | 508 | |
810ecfa7 | 509 | rcu_read_lock(); |
ee89f812 | 510 | hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) { |
0d945c1f | 511 | spin_lock_irq(&blkg->q->queue_lock); |
a2b1693b | 512 | if (blkcg_policy_enabled(blkg->q, pol)) |
f95a04af | 513 | total += prfill(sf, blkg->pd[pol->plid], data); |
0d945c1f | 514 | spin_unlock_irq(&blkg->q->queue_lock); |
810ecfa7 TH |
515 | } |
516 | rcu_read_unlock(); | |
d3d32e69 TH |
517 | |
518 | if (show_total) | |
519 | seq_printf(sf, "Total %llu\n", (unsigned long long)total); | |
520 | } | |
829fdb50 | 521 | EXPORT_SYMBOL_GPL(blkcg_print_blkgs); |
d3d32e69 TH |
522 | |
523 | /** | |
524 | * __blkg_prfill_u64 - prfill helper for a single u64 value | |
525 | * @sf: seq_file to print to | |
f95a04af | 526 | * @pd: policy private data of interest |
d3d32e69 TH |
527 | * @v: value to print |
528 | * | |
f95a04af | 529 | * Print @v to @sf for the device assocaited with @pd. |
d3d32e69 | 530 | */ |
f95a04af | 531 | u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v) |
d3d32e69 | 532 | { |
f95a04af | 533 | const char *dname = blkg_dev_name(pd->blkg); |
d3d32e69 TH |
534 | |
535 | if (!dname) | |
536 | return 0; | |
537 | ||
538 | seq_printf(sf, "%s %llu\n", dname, (unsigned long long)v); | |
539 | return v; | |
540 | } | |
829fdb50 | 541 | EXPORT_SYMBOL_GPL(__blkg_prfill_u64); |
d3d32e69 | 542 | |
457e490f TE |
543 | /* Performs queue bypass and policy enabled checks then looks up blkg. */ |
544 | static struct blkcg_gq *blkg_lookup_check(struct blkcg *blkcg, | |
545 | const struct blkcg_policy *pol, | |
546 | struct request_queue *q) | |
547 | { | |
548 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
0d945c1f | 549 | lockdep_assert_held(&q->queue_lock); |
457e490f TE |
550 | |
551 | if (!blkcg_policy_enabled(q, pol)) | |
552 | return ERR_PTR(-EOPNOTSUPP); | |
457e490f TE |
553 | return __blkg_lookup(blkcg, q, true /* update_hint */); |
554 | } | |
555 | ||
015d254c | 556 | /** |
22ae8ce8 | 557 | * blkcg_conf_open_bdev - parse and open bdev for per-blkg config update |
015d254c TH |
558 | * @inputp: input string pointer |
559 | * | |
560 | * Parse the device node prefix part, MAJ:MIN, of per-blkg config update | |
22ae8ce8 | 561 | * from @input and get and return the matching bdev. *@inputp is |
015d254c TH |
562 | * updated to point past the device node prefix. Returns an ERR_PTR() |
563 | * value on error. | |
564 | * | |
565 | * Use this function iff blkg_conf_prep() can't be used for some reason. | |
566 | */ | |
22ae8ce8 | 567 | struct block_device *blkcg_conf_open_bdev(char **inputp) |
015d254c TH |
568 | { |
569 | char *input = *inputp; | |
570 | unsigned int major, minor; | |
22ae8ce8 CH |
571 | struct block_device *bdev; |
572 | int key_len; | |
015d254c TH |
573 | |
574 | if (sscanf(input, "%u:%u%n", &major, &minor, &key_len) != 2) | |
575 | return ERR_PTR(-EINVAL); | |
576 | ||
577 | input += key_len; | |
578 | if (!isspace(*input)) | |
579 | return ERR_PTR(-EINVAL); | |
580 | input = skip_spaces(input); | |
581 | ||
22ae8ce8 CH |
582 | bdev = blkdev_get_no_open(MKDEV(major, minor)); |
583 | if (!bdev) | |
015d254c | 584 | return ERR_PTR(-ENODEV); |
22ae8ce8 CH |
585 | if (bdev_is_partition(bdev)) { |
586 | blkdev_put_no_open(bdev); | |
015d254c TH |
587 | return ERR_PTR(-ENODEV); |
588 | } | |
589 | ||
590 | *inputp = input; | |
22ae8ce8 | 591 | return bdev; |
015d254c TH |
592 | } |
593 | ||
3a8b31d3 TH |
594 | /** |
595 | * blkg_conf_prep - parse and prepare for per-blkg config update | |
596 | * @blkcg: target block cgroup | |
da8b0662 | 597 | * @pol: target policy |
3a8b31d3 TH |
598 | * @input: input string |
599 | * @ctx: blkg_conf_ctx to be filled | |
600 | * | |
601 | * Parse per-blkg config update from @input and initialize @ctx with the | |
36aa9e5f TH |
602 | * result. @ctx->blkg points to the blkg to be updated and @ctx->body the |
603 | * part of @input following MAJ:MIN. This function returns with RCU read | |
604 | * lock and queue lock held and must be paired with blkg_conf_finish(). | |
3a8b31d3 | 605 | */ |
3c798398 | 606 | int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol, |
36aa9e5f | 607 | char *input, struct blkg_conf_ctx *ctx) |
22ae8ce8 | 608 | __acquires(rcu) __acquires(&bdev->bd_disk->queue->queue_lock) |
34d0f179 | 609 | { |
22ae8ce8 | 610 | struct block_device *bdev; |
457e490f | 611 | struct request_queue *q; |
3c798398 | 612 | struct blkcg_gq *blkg; |
015d254c | 613 | int ret; |
36aa9e5f | 614 | |
22ae8ce8 CH |
615 | bdev = blkcg_conf_open_bdev(&input); |
616 | if (IS_ERR(bdev)) | |
617 | return PTR_ERR(bdev); | |
e56da7e2 | 618 | |
22ae8ce8 | 619 | q = bdev->bd_disk->queue; |
da8b0662 | 620 | |
457e490f | 621 | rcu_read_lock(); |
0d945c1f | 622 | spin_lock_irq(&q->queue_lock); |
e56da7e2 | 623 | |
457e490f | 624 | blkg = blkg_lookup_check(blkcg, pol, q); |
4bfd482e TH |
625 | if (IS_ERR(blkg)) { |
626 | ret = PTR_ERR(blkg); | |
457e490f TE |
627 | goto fail_unlock; |
628 | } | |
629 | ||
630 | if (blkg) | |
631 | goto success; | |
632 | ||
633 | /* | |
634 | * Create blkgs walking down from blkcg_root to @blkcg, so that all | |
635 | * non-root blkgs have access to their parents. | |
636 | */ | |
637 | while (true) { | |
638 | struct blkcg *pos = blkcg; | |
639 | struct blkcg *parent; | |
640 | struct blkcg_gq *new_blkg; | |
641 | ||
642 | parent = blkcg_parent(blkcg); | |
643 | while (parent && !__blkg_lookup(parent, q, false)) { | |
644 | pos = parent; | |
645 | parent = blkcg_parent(parent); | |
646 | } | |
647 | ||
648 | /* Drop locks to do new blkg allocation with GFP_KERNEL. */ | |
0d945c1f | 649 | spin_unlock_irq(&q->queue_lock); |
3a8b31d3 | 650 | rcu_read_unlock(); |
457e490f TE |
651 | |
652 | new_blkg = blkg_alloc(pos, q, GFP_KERNEL); | |
653 | if (unlikely(!new_blkg)) { | |
654 | ret = -ENOMEM; | |
655 | goto fail; | |
7702e8f4 | 656 | } |
3a8b31d3 | 657 | |
f255c19b GKB |
658 | if (radix_tree_preload(GFP_KERNEL)) { |
659 | blkg_free(new_blkg); | |
660 | ret = -ENOMEM; | |
661 | goto fail; | |
662 | } | |
663 | ||
457e490f | 664 | rcu_read_lock(); |
0d945c1f | 665 | spin_lock_irq(&q->queue_lock); |
457e490f TE |
666 | |
667 | blkg = blkg_lookup_check(pos, pol, q); | |
668 | if (IS_ERR(blkg)) { | |
669 | ret = PTR_ERR(blkg); | |
52abfcbd | 670 | blkg_free(new_blkg); |
f255c19b | 671 | goto fail_preloaded; |
457e490f TE |
672 | } |
673 | ||
674 | if (blkg) { | |
675 | blkg_free(new_blkg); | |
676 | } else { | |
677 | blkg = blkg_create(pos, q, new_blkg); | |
98d669b4 | 678 | if (IS_ERR(blkg)) { |
457e490f | 679 | ret = PTR_ERR(blkg); |
f255c19b | 680 | goto fail_preloaded; |
457e490f TE |
681 | } |
682 | } | |
683 | ||
f255c19b GKB |
684 | radix_tree_preload_end(); |
685 | ||
457e490f TE |
686 | if (pos == blkcg) |
687 | goto success; | |
688 | } | |
689 | success: | |
22ae8ce8 | 690 | ctx->bdev = bdev; |
3a8b31d3 | 691 | ctx->blkg = blkg; |
015d254c | 692 | ctx->body = input; |
726fa694 | 693 | return 0; |
457e490f | 694 | |
f255c19b GKB |
695 | fail_preloaded: |
696 | radix_tree_preload_end(); | |
457e490f | 697 | fail_unlock: |
0d945c1f | 698 | spin_unlock_irq(&q->queue_lock); |
457e490f TE |
699 | rcu_read_unlock(); |
700 | fail: | |
22ae8ce8 | 701 | blkdev_put_no_open(bdev); |
457e490f TE |
702 | /* |
703 | * If queue was bypassing, we should retry. Do so after a | |
704 | * short msleep(). It isn't strictly necessary but queue | |
705 | * can be bypassing for some time and it's always nice to | |
706 | * avoid busy looping. | |
707 | */ | |
708 | if (ret == -EBUSY) { | |
709 | msleep(10); | |
710 | ret = restart_syscall(); | |
711 | } | |
712 | return ret; | |
34d0f179 | 713 | } |
89f3b6d6 | 714 | EXPORT_SYMBOL_GPL(blkg_conf_prep); |
34d0f179 | 715 | |
3a8b31d3 TH |
716 | /** |
717 | * blkg_conf_finish - finish up per-blkg config update | |
718 | * @ctx: blkg_conf_ctx intiailized by blkg_conf_prep() | |
719 | * | |
720 | * Finish up after per-blkg config update. This function must be paired | |
721 | * with blkg_conf_prep(). | |
722 | */ | |
829fdb50 | 723 | void blkg_conf_finish(struct blkg_conf_ctx *ctx) |
22ae8ce8 | 724 | __releases(&ctx->bdev->bd_disk->queue->queue_lock) __releases(rcu) |
34d0f179 | 725 | { |
22ae8ce8 | 726 | spin_unlock_irq(&ctx->bdev->bd_disk->queue->queue_lock); |
3a8b31d3 | 727 | rcu_read_unlock(); |
22ae8ce8 | 728 | blkdev_put_no_open(ctx->bdev); |
34d0f179 | 729 | } |
89f3b6d6 | 730 | EXPORT_SYMBOL_GPL(blkg_conf_finish); |
34d0f179 | 731 | |
cd1fc4b9 BB |
732 | static void blkg_iostat_set(struct blkg_iostat *dst, struct blkg_iostat *src) |
733 | { | |
734 | int i; | |
735 | ||
736 | for (i = 0; i < BLKG_IOSTAT_NR; i++) { | |
737 | dst->bytes[i] = src->bytes[i]; | |
738 | dst->ios[i] = src->ios[i]; | |
739 | } | |
740 | } | |
741 | ||
742 | static void blkg_iostat_add(struct blkg_iostat *dst, struct blkg_iostat *src) | |
743 | { | |
744 | int i; | |
745 | ||
746 | for (i = 0; i < BLKG_IOSTAT_NR; i++) { | |
747 | dst->bytes[i] += src->bytes[i]; | |
748 | dst->ios[i] += src->ios[i]; | |
749 | } | |
750 | } | |
751 | ||
752 | static void blkg_iostat_sub(struct blkg_iostat *dst, struct blkg_iostat *src) | |
753 | { | |
754 | int i; | |
755 | ||
756 | for (i = 0; i < BLKG_IOSTAT_NR; i++) { | |
757 | dst->bytes[i] -= src->bytes[i]; | |
758 | dst->ios[i] -= src->ios[i]; | |
759 | } | |
760 | } | |
761 | ||
762 | static void blkcg_rstat_flush(struct cgroup_subsys_state *css, int cpu) | |
763 | { | |
764 | struct blkcg *blkcg = css_to_blkcg(css); | |
765 | struct blkcg_gq *blkg; | |
766 | ||
dc26532a JW |
767 | /* Root-level stats are sourced from system-wide IO stats */ |
768 | if (!cgroup_parent(css->cgroup)) | |
769 | return; | |
770 | ||
cd1fc4b9 BB |
771 | rcu_read_lock(); |
772 | ||
773 | hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) { | |
774 | struct blkcg_gq *parent = blkg->parent; | |
775 | struct blkg_iostat_set *bisc = per_cpu_ptr(blkg->iostat_cpu, cpu); | |
776 | struct blkg_iostat cur, delta; | |
777 | unsigned int seq; | |
778 | ||
779 | /* fetch the current per-cpu values */ | |
780 | do { | |
781 | seq = u64_stats_fetch_begin(&bisc->sync); | |
782 | blkg_iostat_set(&cur, &bisc->cur); | |
783 | } while (u64_stats_fetch_retry(&bisc->sync, seq)); | |
784 | ||
785 | /* propagate percpu delta to global */ | |
786 | u64_stats_update_begin(&blkg->iostat.sync); | |
787 | blkg_iostat_set(&delta, &cur); | |
788 | blkg_iostat_sub(&delta, &bisc->last); | |
789 | blkg_iostat_add(&blkg->iostat.cur, &delta); | |
790 | blkg_iostat_add(&bisc->last, &delta); | |
791 | u64_stats_update_end(&blkg->iostat.sync); | |
792 | ||
dc26532a JW |
793 | /* propagate global delta to parent (unless that's root) */ |
794 | if (parent && parent->parent) { | |
cd1fc4b9 BB |
795 | u64_stats_update_begin(&parent->iostat.sync); |
796 | blkg_iostat_set(&delta, &blkg->iostat.cur); | |
797 | blkg_iostat_sub(&delta, &blkg->iostat.last); | |
798 | blkg_iostat_add(&parent->iostat.cur, &delta); | |
799 | blkg_iostat_add(&blkg->iostat.last, &delta); | |
800 | u64_stats_update_end(&parent->iostat.sync); | |
801 | } | |
802 | } | |
803 | ||
804 | rcu_read_unlock(); | |
805 | } | |
806 | ||
ef45fe47 | 807 | /* |
dc26532a JW |
808 | * We source root cgroup stats from the system-wide stats to avoid |
809 | * tracking the same information twice and incurring overhead when no | |
810 | * cgroups are defined. For that reason, cgroup_rstat_flush in | |
811 | * blkcg_print_stat does not actually fill out the iostat in the root | |
812 | * cgroup's blkcg_gq. | |
ef45fe47 BB |
813 | * |
814 | * However, we would like to re-use the printing code between the root and | |
815 | * non-root cgroups to the extent possible. For that reason, we simulate | |
816 | * flushing the root cgroup's stats by explicitly filling in the iostat | |
817 | * with disk level statistics. | |
818 | */ | |
819 | static void blkcg_fill_root_iostats(void) | |
820 | { | |
821 | struct class_dev_iter iter; | |
822 | struct device *dev; | |
823 | ||
824 | class_dev_iter_init(&iter, &block_class, NULL, &disk_type); | |
825 | while ((dev = class_dev_iter_next(&iter))) { | |
0d02129e CH |
826 | struct block_device *bdev = dev_to_bdev(dev); |
827 | struct blkcg_gq *blkg = | |
828 | blk_queue_root_blkg(bdev->bd_disk->queue); | |
ef45fe47 BB |
829 | struct blkg_iostat tmp; |
830 | int cpu; | |
831 | ||
832 | memset(&tmp, 0, sizeof(tmp)); | |
833 | for_each_possible_cpu(cpu) { | |
834 | struct disk_stats *cpu_dkstats; | |
835 | ||
0d02129e | 836 | cpu_dkstats = per_cpu_ptr(bdev->bd_stats, cpu); |
ef45fe47 BB |
837 | tmp.ios[BLKG_IOSTAT_READ] += |
838 | cpu_dkstats->ios[STAT_READ]; | |
839 | tmp.ios[BLKG_IOSTAT_WRITE] += | |
840 | cpu_dkstats->ios[STAT_WRITE]; | |
841 | tmp.ios[BLKG_IOSTAT_DISCARD] += | |
842 | cpu_dkstats->ios[STAT_DISCARD]; | |
843 | // convert sectors to bytes | |
844 | tmp.bytes[BLKG_IOSTAT_READ] += | |
845 | cpu_dkstats->sectors[STAT_READ] << 9; | |
846 | tmp.bytes[BLKG_IOSTAT_WRITE] += | |
847 | cpu_dkstats->sectors[STAT_WRITE] << 9; | |
848 | tmp.bytes[BLKG_IOSTAT_DISCARD] += | |
849 | cpu_dkstats->sectors[STAT_DISCARD] << 9; | |
850 | ||
851 | u64_stats_update_begin(&blkg->iostat.sync); | |
852 | blkg_iostat_set(&blkg->iostat.cur, &tmp); | |
853 | u64_stats_update_end(&blkg->iostat.sync); | |
854 | } | |
855 | } | |
856 | } | |
857 | ||
2ee867dc TH |
858 | static int blkcg_print_stat(struct seq_file *sf, void *v) |
859 | { | |
860 | struct blkcg *blkcg = css_to_blkcg(seq_css(sf)); | |
861 | struct blkcg_gq *blkg; | |
862 | ||
ef45fe47 BB |
863 | if (!seq_css(sf)->parent) |
864 | blkcg_fill_root_iostats(); | |
865 | else | |
866 | cgroup_rstat_flush(blkcg->css.cgroup); | |
867 | ||
2ee867dc TH |
868 | rcu_read_lock(); |
869 | ||
870 | hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) { | |
f7331648 | 871 | struct blkg_iostat_set *bis = &blkg->iostat; |
2ee867dc | 872 | const char *dname; |
903d23f0 | 873 | char *buf; |
636620b6 | 874 | u64 rbytes, wbytes, rios, wios, dbytes, dios; |
903d23f0 JB |
875 | size_t size = seq_get_buf(sf, &buf), off = 0; |
876 | int i; | |
877 | bool has_stats = false; | |
f7331648 | 878 | unsigned seq; |
2ee867dc | 879 | |
b0814361 TH |
880 | spin_lock_irq(&blkg->q->queue_lock); |
881 | ||
882 | if (!blkg->online) | |
883 | goto skip; | |
884 | ||
2ee867dc TH |
885 | dname = blkg_dev_name(blkg); |
886 | if (!dname) | |
b0814361 | 887 | goto skip; |
2ee867dc | 888 | |
903d23f0 JB |
889 | /* |
890 | * Hooray string manipulation, count is the size written NOT | |
891 | * INCLUDING THE \0, so size is now count+1 less than what we | |
892 | * had before, but we want to start writing the next bit from | |
893 | * the \0 so we only add count to buf. | |
894 | */ | |
895 | off += scnprintf(buf+off, size-off, "%s ", dname); | |
896 | ||
f7331648 TH |
897 | do { |
898 | seq = u64_stats_fetch_begin(&bis->sync); | |
2ee867dc | 899 | |
f7331648 TH |
900 | rbytes = bis->cur.bytes[BLKG_IOSTAT_READ]; |
901 | wbytes = bis->cur.bytes[BLKG_IOSTAT_WRITE]; | |
902 | dbytes = bis->cur.bytes[BLKG_IOSTAT_DISCARD]; | |
903 | rios = bis->cur.ios[BLKG_IOSTAT_READ]; | |
904 | wios = bis->cur.ios[BLKG_IOSTAT_WRITE]; | |
905 | dios = bis->cur.ios[BLKG_IOSTAT_DISCARD]; | |
906 | } while (u64_stats_fetch_retry(&bis->sync, seq)); | |
2ee867dc | 907 | |
903d23f0 JB |
908 | if (rbytes || wbytes || rios || wios) { |
909 | has_stats = true; | |
910 | off += scnprintf(buf+off, size-off, | |
636620b6 TH |
911 | "rbytes=%llu wbytes=%llu rios=%llu wios=%llu dbytes=%llu dios=%llu", |
912 | rbytes, wbytes, rios, wios, | |
913 | dbytes, dios); | |
903d23f0 JB |
914 | } |
915 | ||
07b0fdec | 916 | if (blkcg_debug_stats && atomic_read(&blkg->use_delay)) { |
d09d8df3 JB |
917 | has_stats = true; |
918 | off += scnprintf(buf+off, size-off, | |
919 | " use_delay=%d delay_nsec=%llu", | |
920 | atomic_read(&blkg->use_delay), | |
921 | (unsigned long long)atomic64_read(&blkg->delay_nsec)); | |
922 | } | |
923 | ||
903d23f0 JB |
924 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
925 | struct blkcg_policy *pol = blkcg_policy[i]; | |
926 | size_t written; | |
927 | ||
928 | if (!blkg->pd[i] || !pol->pd_stat_fn) | |
929 | continue; | |
930 | ||
931 | written = pol->pd_stat_fn(blkg->pd[i], buf+off, size-off); | |
932 | if (written) | |
933 | has_stats = true; | |
934 | off += written; | |
935 | } | |
07b0fdec | 936 | |
903d23f0 | 937 | if (has_stats) { |
f539da82 TH |
938 | if (off < size - 1) { |
939 | off += scnprintf(buf+off, size-off, "\n"); | |
940 | seq_commit(sf, off); | |
941 | } else { | |
942 | seq_commit(sf, -1); | |
943 | } | |
903d23f0 | 944 | } |
b0814361 TH |
945 | skip: |
946 | spin_unlock_irq(&blkg->q->queue_lock); | |
2ee867dc TH |
947 | } |
948 | ||
949 | rcu_read_unlock(); | |
950 | return 0; | |
951 | } | |
952 | ||
e1f3b941 | 953 | static struct cftype blkcg_files[] = { |
2ee867dc TH |
954 | { |
955 | .name = "stat", | |
956 | .seq_show = blkcg_print_stat, | |
957 | }, | |
958 | { } /* terminate */ | |
959 | }; | |
960 | ||
e1f3b941 | 961 | static struct cftype blkcg_legacy_files[] = { |
84c124da DS |
962 | { |
963 | .name = "reset_stats", | |
3c798398 | 964 | .write_u64 = blkcg_reset_stats, |
22084190 | 965 | }, |
4baf6e33 | 966 | { } /* terminate */ |
31e4c28d VG |
967 | }; |
968 | ||
59b57717 DZF |
969 | /* |
970 | * blkcg destruction is a three-stage process. | |
971 | * | |
972 | * 1. Destruction starts. The blkcg_css_offline() callback is invoked | |
973 | * which offlines writeback. Here we tie the next stage of blkg destruction | |
974 | * to the completion of writeback associated with the blkcg. This lets us | |
975 | * avoid punting potentially large amounts of outstanding writeback to root | |
976 | * while maintaining any ongoing policies. The next stage is triggered when | |
977 | * the nr_cgwbs count goes to zero. | |
978 | * | |
979 | * 2. When the nr_cgwbs count goes to zero, blkcg_destroy_blkgs() is called | |
980 | * and handles the destruction of blkgs. Here the css reference held by | |
981 | * the blkg is put back eventually allowing blkcg_css_free() to be called. | |
982 | * This work may occur in cgwb_release_workfn() on the cgwb_release | |
983 | * workqueue. Any submitted ios that fail to get the blkg ref will be | |
984 | * punted to the root_blkg. | |
985 | * | |
986 | * 3. Once the blkcg ref count goes to zero, blkcg_css_free() is called. | |
987 | * This finally frees the blkcg. | |
988 | */ | |
989 | ||
9f13ef67 | 990 | /** |
92fb9748 | 991 | * blkcg_css_offline - cgroup css_offline callback |
eb95419b | 992 | * @css: css of interest |
9f13ef67 | 993 | * |
59b57717 DZF |
994 | * This function is called when @css is about to go away. Here the cgwbs are |
995 | * offlined first and only once writeback associated with the blkcg has | |
996 | * finished do we start step 2 (see above). | |
9f13ef67 | 997 | */ |
eb95419b | 998 | static void blkcg_css_offline(struct cgroup_subsys_state *css) |
31e4c28d | 999 | { |
eb95419b | 1000 | struct blkcg *blkcg = css_to_blkcg(css); |
b1c35769 | 1001 | |
59b57717 DZF |
1002 | /* this prevents anyone from attaching or migrating to this blkcg */ |
1003 | wb_blkcg_offline(blkcg); | |
1004 | ||
d866dbf6 TH |
1005 | /* put the base online pin allowing step 2 to be triggered */ |
1006 | blkcg_unpin_online(blkcg); | |
59b57717 DZF |
1007 | } |
1008 | ||
1009 | /** | |
1010 | * blkcg_destroy_blkgs - responsible for shooting down blkgs | |
1011 | * @blkcg: blkcg of interest | |
1012 | * | |
1013 | * blkgs should be removed while holding both q and blkcg locks. As blkcg lock | |
1014 | * is nested inside q lock, this function performs reverse double lock dancing. | |
1015 | * Destroying the blkgs releases the reference held on the blkcg's css allowing | |
1016 | * blkcg_css_free to eventually be called. | |
1017 | * | |
1018 | * This is the blkcg counterpart of ioc_release_fn(). | |
1019 | */ | |
1020 | void blkcg_destroy_blkgs(struct blkcg *blkcg) | |
1021 | { | |
6c635cae BW |
1022 | might_sleep(); |
1023 | ||
9f13ef67 | 1024 | spin_lock_irq(&blkcg->lock); |
7ee9c562 | 1025 | |
4c699480 JQ |
1026 | while (!hlist_empty(&blkcg->blkg_list)) { |
1027 | struct blkcg_gq *blkg = hlist_entry(blkcg->blkg_list.first, | |
6b065462 | 1028 | struct blkcg_gq, blkcg_node); |
4c699480 JQ |
1029 | struct request_queue *q = blkg->q; |
1030 | ||
6c635cae BW |
1031 | if (need_resched() || !spin_trylock(&q->queue_lock)) { |
1032 | /* | |
1033 | * Given that the system can accumulate a huge number | |
1034 | * of blkgs in pathological cases, check to see if we | |
1035 | * need to rescheduling to avoid softlockup. | |
1036 | */ | |
4c699480 | 1037 | spin_unlock_irq(&blkcg->lock); |
6c635cae | 1038 | cond_resched(); |
4c699480 | 1039 | spin_lock_irq(&blkcg->lock); |
6c635cae | 1040 | continue; |
4c699480 | 1041 | } |
6c635cae BW |
1042 | |
1043 | blkg_destroy(blkg); | |
1044 | spin_unlock(&q->queue_lock); | |
4c699480 | 1045 | } |
6b065462 | 1046 | |
4c699480 JQ |
1047 | spin_unlock_irq(&blkcg->lock); |
1048 | } | |
1049 | ||
eb95419b | 1050 | static void blkcg_css_free(struct cgroup_subsys_state *css) |
7ee9c562 | 1051 | { |
eb95419b | 1052 | struct blkcg *blkcg = css_to_blkcg(css); |
bc915e61 | 1053 | int i; |
7ee9c562 | 1054 | |
7876f930 | 1055 | mutex_lock(&blkcg_pol_mutex); |
e4a9bde9 | 1056 | |
7876f930 | 1057 | list_del(&blkcg->all_blkcgs_node); |
7876f930 | 1058 | |
bc915e61 | 1059 | for (i = 0; i < BLKCG_MAX_POLS; i++) |
e4a9bde9 TH |
1060 | if (blkcg->cpd[i]) |
1061 | blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]); | |
1062 | ||
1063 | mutex_unlock(&blkcg_pol_mutex); | |
1064 | ||
bc915e61 | 1065 | kfree(blkcg); |
31e4c28d VG |
1066 | } |
1067 | ||
eb95419b TH |
1068 | static struct cgroup_subsys_state * |
1069 | blkcg_css_alloc(struct cgroup_subsys_state *parent_css) | |
31e4c28d | 1070 | { |
3c798398 | 1071 | struct blkcg *blkcg; |
e48453c3 AA |
1072 | struct cgroup_subsys_state *ret; |
1073 | int i; | |
31e4c28d | 1074 | |
7876f930 TH |
1075 | mutex_lock(&blkcg_pol_mutex); |
1076 | ||
eb95419b | 1077 | if (!parent_css) { |
3c798398 | 1078 | blkcg = &blkcg_root; |
bc915e61 TH |
1079 | } else { |
1080 | blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL); | |
1081 | if (!blkcg) { | |
1082 | ret = ERR_PTR(-ENOMEM); | |
4c18c9e9 | 1083 | goto unlock; |
bc915e61 | 1084 | } |
e48453c3 AA |
1085 | } |
1086 | ||
1087 | for (i = 0; i < BLKCG_MAX_POLS ; i++) { | |
1088 | struct blkcg_policy *pol = blkcg_policy[i]; | |
1089 | struct blkcg_policy_data *cpd; | |
1090 | ||
1091 | /* | |
1092 | * If the policy hasn't been attached yet, wait for it | |
1093 | * to be attached before doing anything else. Otherwise, | |
1094 | * check if the policy requires any specific per-cgroup | |
1095 | * data: if it does, allocate and initialize it. | |
1096 | */ | |
e4a9bde9 | 1097 | if (!pol || !pol->cpd_alloc_fn) |
e48453c3 AA |
1098 | continue; |
1099 | ||
e4a9bde9 | 1100 | cpd = pol->cpd_alloc_fn(GFP_KERNEL); |
e48453c3 AA |
1101 | if (!cpd) { |
1102 | ret = ERR_PTR(-ENOMEM); | |
1103 | goto free_pd_blkcg; | |
1104 | } | |
81437648 TH |
1105 | blkcg->cpd[i] = cpd; |
1106 | cpd->blkcg = blkcg; | |
e48453c3 | 1107 | cpd->plid = i; |
e4a9bde9 TH |
1108 | if (pol->cpd_init_fn) |
1109 | pol->cpd_init_fn(cpd); | |
e48453c3 | 1110 | } |
31e4c28d | 1111 | |
31e4c28d | 1112 | spin_lock_init(&blkcg->lock); |
d866dbf6 | 1113 | refcount_set(&blkcg->online_pin, 1); |
e00f4f4d | 1114 | INIT_RADIX_TREE(&blkcg->blkg_tree, GFP_NOWAIT | __GFP_NOWARN); |
31e4c28d | 1115 | INIT_HLIST_HEAD(&blkcg->blkg_list); |
52ebea74 TH |
1116 | #ifdef CONFIG_CGROUP_WRITEBACK |
1117 | INIT_LIST_HEAD(&blkcg->cgwb_list); | |
1118 | #endif | |
7876f930 TH |
1119 | list_add_tail(&blkcg->all_blkcgs_node, &all_blkcgs); |
1120 | ||
1121 | mutex_unlock(&blkcg_pol_mutex); | |
31e4c28d | 1122 | return &blkcg->css; |
e48453c3 AA |
1123 | |
1124 | free_pd_blkcg: | |
1125 | for (i--; i >= 0; i--) | |
e4a9bde9 TH |
1126 | if (blkcg->cpd[i]) |
1127 | blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]); | |
4c18c9e9 | 1128 | |
1129 | if (blkcg != &blkcg_root) | |
1130 | kfree(blkcg); | |
1131 | unlock: | |
7876f930 | 1132 | mutex_unlock(&blkcg_pol_mutex); |
e48453c3 | 1133 | return ret; |
31e4c28d VG |
1134 | } |
1135 | ||
4308a434 TH |
1136 | static int blkcg_css_online(struct cgroup_subsys_state *css) |
1137 | { | |
1138 | struct blkcg *blkcg = css_to_blkcg(css); | |
1139 | struct blkcg *parent = blkcg_parent(blkcg); | |
1140 | ||
1141 | /* | |
1142 | * blkcg_pin_online() is used to delay blkcg offline so that blkgs | |
1143 | * don't go offline while cgwbs are still active on them. Pin the | |
1144 | * parent so that offline always happens towards the root. | |
1145 | */ | |
1146 | if (parent) | |
1147 | blkcg_pin_online(parent); | |
1148 | return 0; | |
1149 | } | |
1150 | ||
5efd6113 TH |
1151 | /** |
1152 | * blkcg_init_queue - initialize blkcg part of request queue | |
1153 | * @q: request_queue to initialize | |
1154 | * | |
c62b37d9 | 1155 | * Called from blk_alloc_queue(). Responsible for initializing blkcg |
5efd6113 TH |
1156 | * part of new request_queue @q. |
1157 | * | |
1158 | * RETURNS: | |
1159 | * 0 on success, -errno on failure. | |
1160 | */ | |
1161 | int blkcg_init_queue(struct request_queue *q) | |
1162 | { | |
d708f0d5 JA |
1163 | struct blkcg_gq *new_blkg, *blkg; |
1164 | bool preloaded; | |
ec13b1d6 TH |
1165 | int ret; |
1166 | ||
d708f0d5 JA |
1167 | new_blkg = blkg_alloc(&blkcg_root, q, GFP_KERNEL); |
1168 | if (!new_blkg) | |
1169 | return -ENOMEM; | |
1170 | ||
1171 | preloaded = !radix_tree_preload(GFP_KERNEL); | |
1172 | ||
bea54883 | 1173 | /* Make sure the root blkg exists. */ |
ec13b1d6 | 1174 | rcu_read_lock(); |
0d945c1f | 1175 | spin_lock_irq(&q->queue_lock); |
d708f0d5 | 1176 | blkg = blkg_create(&blkcg_root, q, new_blkg); |
901932a3 JB |
1177 | if (IS_ERR(blkg)) |
1178 | goto err_unlock; | |
1179 | q->root_blkg = blkg; | |
0d945c1f | 1180 | spin_unlock_irq(&q->queue_lock); |
ec13b1d6 TH |
1181 | rcu_read_unlock(); |
1182 | ||
d708f0d5 JA |
1183 | if (preloaded) |
1184 | radix_tree_preload_end(); | |
1185 | ||
27029b4b | 1186 | ret = blk_throtl_init(q); |
04be60b5 CH |
1187 | if (ret) |
1188 | goto err_destroy_all; | |
d7067512 | 1189 | |
27029b4b YY |
1190 | ret = blk_iolatency_init(q); |
1191 | if (ret) { | |
1192 | blk_throtl_exit(q); | |
04be60b5 | 1193 | goto err_destroy_all; |
27029b4b | 1194 | } |
04be60b5 | 1195 | return 0; |
901932a3 | 1196 | |
04be60b5 | 1197 | err_destroy_all: |
04be60b5 | 1198 | blkg_destroy_all(q); |
04be60b5 | 1199 | return ret; |
901932a3 | 1200 | err_unlock: |
0d945c1f | 1201 | spin_unlock_irq(&q->queue_lock); |
901932a3 JB |
1202 | rcu_read_unlock(); |
1203 | if (preloaded) | |
1204 | radix_tree_preload_end(); | |
1205 | return PTR_ERR(blkg); | |
5efd6113 TH |
1206 | } |
1207 | ||
5efd6113 TH |
1208 | /** |
1209 | * blkcg_exit_queue - exit and release blkcg part of request_queue | |
1210 | * @q: request_queue being released | |
1211 | * | |
7585d508 | 1212 | * Called from blk_exit_queue(). Responsible for exiting blkcg part. |
5efd6113 TH |
1213 | */ |
1214 | void blkcg_exit_queue(struct request_queue *q) | |
1215 | { | |
3c96cb32 | 1216 | blkg_destroy_all(q); |
5efd6113 TH |
1217 | blk_throtl_exit(q); |
1218 | } | |
1219 | ||
31e4c28d VG |
1220 | /* |
1221 | * We cannot support shared io contexts, as we have no mean to support | |
1222 | * two tasks with the same ioc in two different groups without major rework | |
1223 | * of the main cic data structures. For now we allow a task to change | |
1224 | * its cgroup only if it's the only owner of its ioc. | |
1225 | */ | |
1f7dd3e5 | 1226 | static int blkcg_can_attach(struct cgroup_taskset *tset) |
31e4c28d | 1227 | { |
bb9d97b6 | 1228 | struct task_struct *task; |
1f7dd3e5 | 1229 | struct cgroup_subsys_state *dst_css; |
31e4c28d VG |
1230 | struct io_context *ioc; |
1231 | int ret = 0; | |
1232 | ||
1233 | /* task_lock() is needed to avoid races with exit_io_context() */ | |
1f7dd3e5 | 1234 | cgroup_taskset_for_each(task, dst_css, tset) { |
bb9d97b6 TH |
1235 | task_lock(task); |
1236 | ioc = task->io_context; | |
1237 | if (ioc && atomic_read(&ioc->nr_tasks) > 1) | |
1238 | ret = -EINVAL; | |
1239 | task_unlock(task); | |
1240 | if (ret) | |
1241 | break; | |
1242 | } | |
31e4c28d VG |
1243 | return ret; |
1244 | } | |
1245 | ||
69d7fde5 TH |
1246 | static void blkcg_bind(struct cgroup_subsys_state *root_css) |
1247 | { | |
1248 | int i; | |
1249 | ||
1250 | mutex_lock(&blkcg_pol_mutex); | |
1251 | ||
1252 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
1253 | struct blkcg_policy *pol = blkcg_policy[i]; | |
1254 | struct blkcg *blkcg; | |
1255 | ||
1256 | if (!pol || !pol->cpd_bind_fn) | |
1257 | continue; | |
1258 | ||
1259 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) | |
1260 | if (blkcg->cpd[pol->plid]) | |
1261 | pol->cpd_bind_fn(blkcg->cpd[pol->plid]); | |
1262 | } | |
1263 | mutex_unlock(&blkcg_pol_mutex); | |
1264 | } | |
1265 | ||
d09d8df3 JB |
1266 | static void blkcg_exit(struct task_struct *tsk) |
1267 | { | |
1268 | if (tsk->throttle_queue) | |
1269 | blk_put_queue(tsk->throttle_queue); | |
1270 | tsk->throttle_queue = NULL; | |
1271 | } | |
1272 | ||
c165b3e3 | 1273 | struct cgroup_subsys io_cgrp_subsys = { |
92fb9748 | 1274 | .css_alloc = blkcg_css_alloc, |
4308a434 | 1275 | .css_online = blkcg_css_online, |
92fb9748 TH |
1276 | .css_offline = blkcg_css_offline, |
1277 | .css_free = blkcg_css_free, | |
3c798398 | 1278 | .can_attach = blkcg_can_attach, |
f7331648 | 1279 | .css_rstat_flush = blkcg_rstat_flush, |
69d7fde5 | 1280 | .bind = blkcg_bind, |
2ee867dc | 1281 | .dfl_cftypes = blkcg_files, |
880f50e2 | 1282 | .legacy_cftypes = blkcg_legacy_files, |
c165b3e3 | 1283 | .legacy_name = "blkio", |
d09d8df3 | 1284 | .exit = blkcg_exit, |
1ced953b TH |
1285 | #ifdef CONFIG_MEMCG |
1286 | /* | |
1287 | * This ensures that, if available, memcg is automatically enabled | |
1288 | * together on the default hierarchy so that the owner cgroup can | |
1289 | * be retrieved from writeback pages. | |
1290 | */ | |
1291 | .depends_on = 1 << memory_cgrp_id, | |
1292 | #endif | |
676f7c8f | 1293 | }; |
c165b3e3 | 1294 | EXPORT_SYMBOL_GPL(io_cgrp_subsys); |
676f7c8f | 1295 | |
a2b1693b TH |
1296 | /** |
1297 | * blkcg_activate_policy - activate a blkcg policy on a request_queue | |
1298 | * @q: request_queue of interest | |
1299 | * @pol: blkcg policy to activate | |
1300 | * | |
1301 | * Activate @pol on @q. Requires %GFP_KERNEL context. @q goes through | |
1302 | * bypass mode to populate its blkgs with policy_data for @pol. | |
1303 | * | |
1304 | * Activation happens with @q bypassed, so nobody would be accessing blkgs | |
1305 | * from IO path. Update of each blkg is protected by both queue and blkcg | |
1306 | * locks so that holding either lock and testing blkcg_policy_enabled() is | |
1307 | * always enough for dereferencing policy data. | |
1308 | * | |
1309 | * The caller is responsible for synchronizing [de]activations and policy | |
1310 | * [un]registerations. Returns 0 on success, -errno on failure. | |
1311 | */ | |
1312 | int blkcg_activate_policy(struct request_queue *q, | |
3c798398 | 1313 | const struct blkcg_policy *pol) |
a2b1693b | 1314 | { |
4c55f4f9 | 1315 | struct blkg_policy_data *pd_prealloc = NULL; |
9d179b86 | 1316 | struct blkcg_gq *blkg, *pinned_blkg = NULL; |
4c55f4f9 | 1317 | int ret; |
a2b1693b TH |
1318 | |
1319 | if (blkcg_policy_enabled(q, pol)) | |
1320 | return 0; | |
1321 | ||
344e9ffc | 1322 | if (queue_is_mq(q)) |
bd166ef1 | 1323 | blk_mq_freeze_queue(q); |
9d179b86 | 1324 | retry: |
0d945c1f | 1325 | spin_lock_irq(&q->queue_lock); |
a2b1693b | 1326 | |
9d179b86 | 1327 | /* blkg_list is pushed at the head, reverse walk to allocate parents first */ |
71c81407 | 1328 | list_for_each_entry_reverse(blkg, &q->blkg_list, q_node) { |
4c55f4f9 TH |
1329 | struct blkg_policy_data *pd; |
1330 | ||
1331 | if (blkg->pd[pol->plid]) | |
1332 | continue; | |
a2b1693b | 1333 | |
9d179b86 TH |
1334 | /* If prealloc matches, use it; otherwise try GFP_NOWAIT */ |
1335 | if (blkg == pinned_blkg) { | |
1336 | pd = pd_prealloc; | |
1337 | pd_prealloc = NULL; | |
1338 | } else { | |
1339 | pd = pol->pd_alloc_fn(GFP_NOWAIT | __GFP_NOWARN, q, | |
1340 | blkg->blkcg); | |
1341 | } | |
1342 | ||
4c55f4f9 | 1343 | if (!pd) { |
9d179b86 TH |
1344 | /* |
1345 | * GFP_NOWAIT failed. Free the existing one and | |
1346 | * prealloc for @blkg w/ GFP_KERNEL. | |
1347 | */ | |
1348 | if (pinned_blkg) | |
1349 | blkg_put(pinned_blkg); | |
1350 | blkg_get(blkg); | |
1351 | pinned_blkg = blkg; | |
1352 | ||
0d945c1f | 1353 | spin_unlock_irq(&q->queue_lock); |
9d179b86 TH |
1354 | |
1355 | if (pd_prealloc) | |
1356 | pol->pd_free_fn(pd_prealloc); | |
1357 | pd_prealloc = pol->pd_alloc_fn(GFP_KERNEL, q, | |
1358 | blkg->blkcg); | |
1359 | if (pd_prealloc) | |
1360 | goto retry; | |
1361 | else | |
1362 | goto enomem; | |
4c55f4f9 | 1363 | } |
a2b1693b TH |
1364 | |
1365 | blkg->pd[pol->plid] = pd; | |
1366 | pd->blkg = blkg; | |
b276a876 | 1367 | pd->plid = pol->plid; |
a2b1693b TH |
1368 | } |
1369 | ||
9d179b86 TH |
1370 | /* all allocated, init in the same order */ |
1371 | if (pol->pd_init_fn) | |
1372 | list_for_each_entry_reverse(blkg, &q->blkg_list, q_node) | |
1373 | pol->pd_init_fn(blkg->pd[pol->plid]); | |
1374 | ||
a2b1693b TH |
1375 | __set_bit(pol->plid, q->blkcg_pols); |
1376 | ret = 0; | |
4c55f4f9 | 1377 | |
0d945c1f | 1378 | spin_unlock_irq(&q->queue_lock); |
9d179b86 | 1379 | out: |
344e9ffc | 1380 | if (queue_is_mq(q)) |
bd166ef1 | 1381 | blk_mq_unfreeze_queue(q); |
9d179b86 TH |
1382 | if (pinned_blkg) |
1383 | blkg_put(pinned_blkg); | |
001bea73 TH |
1384 | if (pd_prealloc) |
1385 | pol->pd_free_fn(pd_prealloc); | |
a2b1693b | 1386 | return ret; |
9d179b86 TH |
1387 | |
1388 | enomem: | |
1389 | /* alloc failed, nothing's initialized yet, free everything */ | |
1390 | spin_lock_irq(&q->queue_lock); | |
1391 | list_for_each_entry(blkg, &q->blkg_list, q_node) { | |
1392 | if (blkg->pd[pol->plid]) { | |
1393 | pol->pd_free_fn(blkg->pd[pol->plid]); | |
1394 | blkg->pd[pol->plid] = NULL; | |
1395 | } | |
1396 | } | |
1397 | spin_unlock_irq(&q->queue_lock); | |
1398 | ret = -ENOMEM; | |
1399 | goto out; | |
a2b1693b TH |
1400 | } |
1401 | EXPORT_SYMBOL_GPL(blkcg_activate_policy); | |
1402 | ||
1403 | /** | |
1404 | * blkcg_deactivate_policy - deactivate a blkcg policy on a request_queue | |
1405 | * @q: request_queue of interest | |
1406 | * @pol: blkcg policy to deactivate | |
1407 | * | |
1408 | * Deactivate @pol on @q. Follows the same synchronization rules as | |
1409 | * blkcg_activate_policy(). | |
1410 | */ | |
1411 | void blkcg_deactivate_policy(struct request_queue *q, | |
3c798398 | 1412 | const struct blkcg_policy *pol) |
a2b1693b | 1413 | { |
3c798398 | 1414 | struct blkcg_gq *blkg; |
a2b1693b TH |
1415 | |
1416 | if (!blkcg_policy_enabled(q, pol)) | |
1417 | return; | |
1418 | ||
344e9ffc | 1419 | if (queue_is_mq(q)) |
bd166ef1 | 1420 | blk_mq_freeze_queue(q); |
bd166ef1 | 1421 | |
0d945c1f | 1422 | spin_lock_irq(&q->queue_lock); |
a2b1693b TH |
1423 | |
1424 | __clear_bit(pol->plid, q->blkcg_pols); | |
1425 | ||
1426 | list_for_each_entry(blkg, &q->blkg_list, q_node) { | |
001bea73 | 1427 | if (blkg->pd[pol->plid]) { |
6b065462 | 1428 | if (pol->pd_offline_fn) |
a9520cd6 | 1429 | pol->pd_offline_fn(blkg->pd[pol->plid]); |
001bea73 TH |
1430 | pol->pd_free_fn(blkg->pd[pol->plid]); |
1431 | blkg->pd[pol->plid] = NULL; | |
1432 | } | |
a2b1693b TH |
1433 | } |
1434 | ||
0d945c1f | 1435 | spin_unlock_irq(&q->queue_lock); |
bd166ef1 | 1436 | |
344e9ffc | 1437 | if (queue_is_mq(q)) |
bd166ef1 | 1438 | blk_mq_unfreeze_queue(q); |
a2b1693b TH |
1439 | } |
1440 | EXPORT_SYMBOL_GPL(blkcg_deactivate_policy); | |
1441 | ||
8bd435b3 | 1442 | /** |
3c798398 TH |
1443 | * blkcg_policy_register - register a blkcg policy |
1444 | * @pol: blkcg policy to register | |
8bd435b3 | 1445 | * |
3c798398 TH |
1446 | * Register @pol with blkcg core. Might sleep and @pol may be modified on |
1447 | * successful registration. Returns 0 on success and -errno on failure. | |
8bd435b3 | 1448 | */ |
d5bf0291 | 1449 | int blkcg_policy_register(struct blkcg_policy *pol) |
3e252066 | 1450 | { |
06b285bd | 1451 | struct blkcg *blkcg; |
8bd435b3 | 1452 | int i, ret; |
e8989fae | 1453 | |
838f13bf | 1454 | mutex_lock(&blkcg_pol_register_mutex); |
bc0d6501 TH |
1455 | mutex_lock(&blkcg_pol_mutex); |
1456 | ||
8bd435b3 TH |
1457 | /* find an empty slot */ |
1458 | ret = -ENOSPC; | |
1459 | for (i = 0; i < BLKCG_MAX_POLS; i++) | |
3c798398 | 1460 | if (!blkcg_policy[i]) |
8bd435b3 | 1461 | break; |
01c5f85a JA |
1462 | if (i >= BLKCG_MAX_POLS) { |
1463 | pr_warn("blkcg_policy_register: BLKCG_MAX_POLS too small\n"); | |
838f13bf | 1464 | goto err_unlock; |
01c5f85a | 1465 | } |
035d10b2 | 1466 | |
e8401073 | 1467 | /* Make sure cpd/pd_alloc_fn and cpd/pd_free_fn in pairs */ |
1468 | if ((!pol->cpd_alloc_fn ^ !pol->cpd_free_fn) || | |
1469 | (!pol->pd_alloc_fn ^ !pol->pd_free_fn)) | |
1470 | goto err_unlock; | |
1471 | ||
06b285bd | 1472 | /* register @pol */ |
3c798398 | 1473 | pol->plid = i; |
06b285bd TH |
1474 | blkcg_policy[pol->plid] = pol; |
1475 | ||
1476 | /* allocate and install cpd's */ | |
e4a9bde9 | 1477 | if (pol->cpd_alloc_fn) { |
06b285bd TH |
1478 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) { |
1479 | struct blkcg_policy_data *cpd; | |
1480 | ||
e4a9bde9 | 1481 | cpd = pol->cpd_alloc_fn(GFP_KERNEL); |
bbb427e3 | 1482 | if (!cpd) |
06b285bd | 1483 | goto err_free_cpds; |
06b285bd | 1484 | |
81437648 TH |
1485 | blkcg->cpd[pol->plid] = cpd; |
1486 | cpd->blkcg = blkcg; | |
06b285bd | 1487 | cpd->plid = pol->plid; |
86a5bba5 TH |
1488 | if (pol->cpd_init_fn) |
1489 | pol->cpd_init_fn(cpd); | |
06b285bd TH |
1490 | } |
1491 | } | |
1492 | ||
838f13bf | 1493 | mutex_unlock(&blkcg_pol_mutex); |
8bd435b3 | 1494 | |
8bd435b3 | 1495 | /* everything is in place, add intf files for the new policy */ |
2ee867dc TH |
1496 | if (pol->dfl_cftypes) |
1497 | WARN_ON(cgroup_add_dfl_cftypes(&io_cgrp_subsys, | |
1498 | pol->dfl_cftypes)); | |
880f50e2 | 1499 | if (pol->legacy_cftypes) |
c165b3e3 | 1500 | WARN_ON(cgroup_add_legacy_cftypes(&io_cgrp_subsys, |
880f50e2 | 1501 | pol->legacy_cftypes)); |
838f13bf TH |
1502 | mutex_unlock(&blkcg_pol_register_mutex); |
1503 | return 0; | |
1504 | ||
06b285bd | 1505 | err_free_cpds: |
58a9edce | 1506 | if (pol->cpd_free_fn) { |
06b285bd | 1507 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) { |
e4a9bde9 TH |
1508 | if (blkcg->cpd[pol->plid]) { |
1509 | pol->cpd_free_fn(blkcg->cpd[pol->plid]); | |
1510 | blkcg->cpd[pol->plid] = NULL; | |
1511 | } | |
06b285bd TH |
1512 | } |
1513 | } | |
1514 | blkcg_policy[pol->plid] = NULL; | |
838f13bf | 1515 | err_unlock: |
bc0d6501 | 1516 | mutex_unlock(&blkcg_pol_mutex); |
838f13bf | 1517 | mutex_unlock(&blkcg_pol_register_mutex); |
8bd435b3 | 1518 | return ret; |
3e252066 | 1519 | } |
3c798398 | 1520 | EXPORT_SYMBOL_GPL(blkcg_policy_register); |
3e252066 | 1521 | |
8bd435b3 | 1522 | /** |
3c798398 TH |
1523 | * blkcg_policy_unregister - unregister a blkcg policy |
1524 | * @pol: blkcg policy to unregister | |
8bd435b3 | 1525 | * |
3c798398 | 1526 | * Undo blkcg_policy_register(@pol). Might sleep. |
8bd435b3 | 1527 | */ |
3c798398 | 1528 | void blkcg_policy_unregister(struct blkcg_policy *pol) |
3e252066 | 1529 | { |
06b285bd TH |
1530 | struct blkcg *blkcg; |
1531 | ||
838f13bf | 1532 | mutex_lock(&blkcg_pol_register_mutex); |
bc0d6501 | 1533 | |
3c798398 | 1534 | if (WARN_ON(blkcg_policy[pol->plid] != pol)) |
8bd435b3 TH |
1535 | goto out_unlock; |
1536 | ||
1537 | /* kill the intf files first */ | |
2ee867dc TH |
1538 | if (pol->dfl_cftypes) |
1539 | cgroup_rm_cftypes(pol->dfl_cftypes); | |
880f50e2 TH |
1540 | if (pol->legacy_cftypes) |
1541 | cgroup_rm_cftypes(pol->legacy_cftypes); | |
44ea53de | 1542 | |
06b285bd | 1543 | /* remove cpds and unregister */ |
838f13bf | 1544 | mutex_lock(&blkcg_pol_mutex); |
06b285bd | 1545 | |
58a9edce | 1546 | if (pol->cpd_free_fn) { |
06b285bd | 1547 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) { |
e4a9bde9 TH |
1548 | if (blkcg->cpd[pol->plid]) { |
1549 | pol->cpd_free_fn(blkcg->cpd[pol->plid]); | |
1550 | blkcg->cpd[pol->plid] = NULL; | |
1551 | } | |
06b285bd TH |
1552 | } |
1553 | } | |
3c798398 | 1554 | blkcg_policy[pol->plid] = NULL; |
06b285bd | 1555 | |
bc0d6501 | 1556 | mutex_unlock(&blkcg_pol_mutex); |
838f13bf TH |
1557 | out_unlock: |
1558 | mutex_unlock(&blkcg_pol_register_mutex); | |
3e252066 | 1559 | } |
3c798398 | 1560 | EXPORT_SYMBOL_GPL(blkcg_policy_unregister); |
903d23f0 | 1561 | |
d3f77dfd TH |
1562 | bool __blkcg_punt_bio_submit(struct bio *bio) |
1563 | { | |
1564 | struct blkcg_gq *blkg = bio->bi_blkg; | |
1565 | ||
1566 | /* consume the flag first */ | |
1567 | bio->bi_opf &= ~REQ_CGROUP_PUNT; | |
1568 | ||
1569 | /* never bounce for the root cgroup */ | |
1570 | if (!blkg->parent) | |
1571 | return false; | |
1572 | ||
1573 | spin_lock_bh(&blkg->async_bio_lock); | |
1574 | bio_list_add(&blkg->async_bios, bio); | |
1575 | spin_unlock_bh(&blkg->async_bio_lock); | |
1576 | ||
1577 | queue_work(blkcg_punt_bio_wq, &blkg->async_bio_work); | |
1578 | return true; | |
1579 | } | |
1580 | ||
d09d8df3 JB |
1581 | /* |
1582 | * Scale the accumulated delay based on how long it has been since we updated | |
1583 | * the delay. We only call this when we are adding delay, in case it's been a | |
1584 | * while since we added delay, and when we are checking to see if we need to | |
1585 | * delay a task, to account for any delays that may have occurred. | |
1586 | */ | |
1587 | static void blkcg_scale_delay(struct blkcg_gq *blkg, u64 now) | |
1588 | { | |
1589 | u64 old = atomic64_read(&blkg->delay_start); | |
1590 | ||
54c52e10 TH |
1591 | /* negative use_delay means no scaling, see blkcg_set_delay() */ |
1592 | if (atomic_read(&blkg->use_delay) < 0) | |
1593 | return; | |
1594 | ||
d09d8df3 JB |
1595 | /* |
1596 | * We only want to scale down every second. The idea here is that we | |
1597 | * want to delay people for min(delay_nsec, NSEC_PER_SEC) in a certain | |
1598 | * time window. We only want to throttle tasks for recent delay that | |
1599 | * has occurred, in 1 second time windows since that's the maximum | |
1600 | * things can be throttled. We save the current delay window in | |
1601 | * blkg->last_delay so we know what amount is still left to be charged | |
1602 | * to the blkg from this point onward. blkg->last_use keeps track of | |
1603 | * the use_delay counter. The idea is if we're unthrottling the blkg we | |
1604 | * are ok with whatever is happening now, and we can take away more of | |
1605 | * the accumulated delay as we've already throttled enough that | |
1606 | * everybody is happy with their IO latencies. | |
1607 | */ | |
1608 | if (time_before64(old + NSEC_PER_SEC, now) && | |
1609 | atomic64_cmpxchg(&blkg->delay_start, old, now) == old) { | |
1610 | u64 cur = atomic64_read(&blkg->delay_nsec); | |
1611 | u64 sub = min_t(u64, blkg->last_delay, now - old); | |
1612 | int cur_use = atomic_read(&blkg->use_delay); | |
1613 | ||
1614 | /* | |
1615 | * We've been unthrottled, subtract a larger chunk of our | |
1616 | * accumulated delay. | |
1617 | */ | |
1618 | if (cur_use < blkg->last_use) | |
1619 | sub = max_t(u64, sub, blkg->last_delay >> 1); | |
1620 | ||
1621 | /* | |
1622 | * This shouldn't happen, but handle it anyway. Our delay_nsec | |
1623 | * should only ever be growing except here where we subtract out | |
1624 | * min(last_delay, 1 second), but lord knows bugs happen and I'd | |
1625 | * rather not end up with negative numbers. | |
1626 | */ | |
1627 | if (unlikely(cur < sub)) { | |
1628 | atomic64_set(&blkg->delay_nsec, 0); | |
1629 | blkg->last_delay = 0; | |
1630 | } else { | |
1631 | atomic64_sub(sub, &blkg->delay_nsec); | |
1632 | blkg->last_delay = cur - sub; | |
1633 | } | |
1634 | blkg->last_use = cur_use; | |
1635 | } | |
1636 | } | |
1637 | ||
1638 | /* | |
1639 | * This is called when we want to actually walk up the hierarchy and check to | |
1640 | * see if we need to throttle, and then actually throttle if there is some | |
1641 | * accumulated delay. This should only be called upon return to user space so | |
1642 | * we're not holding some lock that would induce a priority inversion. | |
1643 | */ | |
1644 | static void blkcg_maybe_throttle_blkg(struct blkcg_gq *blkg, bool use_memdelay) | |
1645 | { | |
fd112c74 | 1646 | unsigned long pflags; |
5160a5a5 | 1647 | bool clamp; |
d09d8df3 JB |
1648 | u64 now = ktime_to_ns(ktime_get()); |
1649 | u64 exp; | |
1650 | u64 delay_nsec = 0; | |
1651 | int tok; | |
1652 | ||
1653 | while (blkg->parent) { | |
5160a5a5 TH |
1654 | int use_delay = atomic_read(&blkg->use_delay); |
1655 | ||
1656 | if (use_delay) { | |
1657 | u64 this_delay; | |
1658 | ||
d09d8df3 | 1659 | blkcg_scale_delay(blkg, now); |
5160a5a5 TH |
1660 | this_delay = atomic64_read(&blkg->delay_nsec); |
1661 | if (this_delay > delay_nsec) { | |
1662 | delay_nsec = this_delay; | |
1663 | clamp = use_delay > 0; | |
1664 | } | |
d09d8df3 JB |
1665 | } |
1666 | blkg = blkg->parent; | |
1667 | } | |
1668 | ||
1669 | if (!delay_nsec) | |
1670 | return; | |
1671 | ||
1672 | /* | |
1673 | * Let's not sleep for all eternity if we've amassed a huge delay. | |
1674 | * Swapping or metadata IO can accumulate 10's of seconds worth of | |
1675 | * delay, and we want userspace to be able to do _something_ so cap the | |
5160a5a5 TH |
1676 | * delays at 0.25s. If there's 10's of seconds worth of delay then the |
1677 | * tasks will be delayed for 0.25 second for every syscall. If | |
1678 | * blkcg_set_delay() was used as indicated by negative use_delay, the | |
1679 | * caller is responsible for regulating the range. | |
d09d8df3 | 1680 | */ |
5160a5a5 TH |
1681 | if (clamp) |
1682 | delay_nsec = min_t(u64, delay_nsec, 250 * NSEC_PER_MSEC); | |
d09d8df3 | 1683 | |
fd112c74 JB |
1684 | if (use_memdelay) |
1685 | psi_memstall_enter(&pflags); | |
d09d8df3 JB |
1686 | |
1687 | exp = ktime_add_ns(now, delay_nsec); | |
1688 | tok = io_schedule_prepare(); | |
1689 | do { | |
1690 | __set_current_state(TASK_KILLABLE); | |
1691 | if (!schedule_hrtimeout(&exp, HRTIMER_MODE_ABS)) | |
1692 | break; | |
1693 | } while (!fatal_signal_pending(current)); | |
1694 | io_schedule_finish(tok); | |
fd112c74 JB |
1695 | |
1696 | if (use_memdelay) | |
1697 | psi_memstall_leave(&pflags); | |
d09d8df3 JB |
1698 | } |
1699 | ||
1700 | /** | |
1701 | * blkcg_maybe_throttle_current - throttle the current task if it has been marked | |
1702 | * | |
1703 | * This is only called if we've been marked with set_notify_resume(). Obviously | |
1704 | * we can be set_notify_resume() for reasons other than blkcg throttling, so we | |
1705 | * check to see if current->throttle_queue is set and if not this doesn't do | |
1706 | * anything. This should only ever be called by the resume code, it's not meant | |
1707 | * to be called by people willy-nilly as it will actually do the work to | |
1708 | * throttle the task if it is setup for throttling. | |
1709 | */ | |
1710 | void blkcg_maybe_throttle_current(void) | |
1711 | { | |
1712 | struct request_queue *q = current->throttle_queue; | |
1713 | struct cgroup_subsys_state *css; | |
1714 | struct blkcg *blkcg; | |
1715 | struct blkcg_gq *blkg; | |
1716 | bool use_memdelay = current->use_memdelay; | |
1717 | ||
1718 | if (!q) | |
1719 | return; | |
1720 | ||
1721 | current->throttle_queue = NULL; | |
1722 | current->use_memdelay = false; | |
1723 | ||
1724 | rcu_read_lock(); | |
1725 | css = kthread_blkcg(); | |
1726 | if (css) | |
1727 | blkcg = css_to_blkcg(css); | |
1728 | else | |
1729 | blkcg = css_to_blkcg(task_css(current, io_cgrp_id)); | |
1730 | ||
1731 | if (!blkcg) | |
1732 | goto out; | |
1733 | blkg = blkg_lookup(blkcg, q); | |
1734 | if (!blkg) | |
1735 | goto out; | |
7754f669 | 1736 | if (!blkg_tryget(blkg)) |
d09d8df3 JB |
1737 | goto out; |
1738 | rcu_read_unlock(); | |
d09d8df3 JB |
1739 | |
1740 | blkcg_maybe_throttle_blkg(blkg, use_memdelay); | |
1741 | blkg_put(blkg); | |
cc7ecc25 | 1742 | blk_put_queue(q); |
d09d8df3 JB |
1743 | return; |
1744 | out: | |
1745 | rcu_read_unlock(); | |
1746 | blk_put_queue(q); | |
1747 | } | |
d09d8df3 JB |
1748 | |
1749 | /** | |
1750 | * blkcg_schedule_throttle - this task needs to check for throttling | |
537d71b3 BVA |
1751 | * @q: the request queue IO was submitted on |
1752 | * @use_memdelay: do we charge this to memory delay for PSI | |
d09d8df3 JB |
1753 | * |
1754 | * This is called by the IO controller when we know there's delay accumulated | |
1755 | * for the blkg for this task. We do not pass the blkg because there are places | |
1756 | * we call this that may not have that information, the swapping code for | |
1757 | * instance will only have a request_queue at that point. This set's the | |
1758 | * notify_resume for the task to check and see if it requires throttling before | |
1759 | * returning to user space. | |
1760 | * | |
1761 | * We will only schedule once per syscall. You can call this over and over | |
1762 | * again and it will only do the check once upon return to user space, and only | |
1763 | * throttle once. If the task needs to be throttled again it'll need to be | |
1764 | * re-set at the next time we see the task. | |
1765 | */ | |
1766 | void blkcg_schedule_throttle(struct request_queue *q, bool use_memdelay) | |
1767 | { | |
1768 | if (unlikely(current->flags & PF_KTHREAD)) | |
1769 | return; | |
1770 | ||
49d1822b CX |
1771 | if (current->throttle_queue != q) { |
1772 | if (!blk_get_queue(q)) | |
1773 | return; | |
1774 | ||
1775 | if (current->throttle_queue) | |
1776 | blk_put_queue(current->throttle_queue); | |
1777 | current->throttle_queue = q; | |
1778 | } | |
d09d8df3 | 1779 | |
d09d8df3 JB |
1780 | if (use_memdelay) |
1781 | current->use_memdelay = use_memdelay; | |
1782 | set_notify_resume(current); | |
1783 | } | |
d09d8df3 JB |
1784 | |
1785 | /** | |
1786 | * blkcg_add_delay - add delay to this blkg | |
537d71b3 BVA |
1787 | * @blkg: blkg of interest |
1788 | * @now: the current time in nanoseconds | |
1789 | * @delta: how many nanoseconds of delay to add | |
d09d8df3 JB |
1790 | * |
1791 | * Charge @delta to the blkg's current delay accumulation. This is used to | |
1792 | * throttle tasks if an IO controller thinks we need more throttling. | |
1793 | */ | |
1794 | void blkcg_add_delay(struct blkcg_gq *blkg, u64 now, u64 delta) | |
1795 | { | |
54c52e10 TH |
1796 | if (WARN_ON_ONCE(atomic_read(&blkg->use_delay) < 0)) |
1797 | return; | |
d09d8df3 JB |
1798 | blkcg_scale_delay(blkg, now); |
1799 | atomic64_add(delta, &blkg->delay_nsec); | |
1800 | } | |
d09d8df3 | 1801 | |
28fc591f CH |
1802 | /** |
1803 | * blkg_tryget_closest - try and get a blkg ref on the closet blkg | |
13c7863d CH |
1804 | * @bio: target bio |
1805 | * @css: target css | |
28fc591f | 1806 | * |
13c7863d CH |
1807 | * As the failure mode here is to walk up the blkg tree, this ensure that the |
1808 | * blkg->parent pointers are always valid. This returns the blkg that it ended | |
1809 | * up taking a reference on or %NULL if no reference was taken. | |
28fc591f | 1810 | */ |
13c7863d CH |
1811 | static inline struct blkcg_gq *blkg_tryget_closest(struct bio *bio, |
1812 | struct cgroup_subsys_state *css) | |
28fc591f | 1813 | { |
13c7863d | 1814 | struct blkcg_gq *blkg, *ret_blkg = NULL; |
28fc591f | 1815 | |
13c7863d | 1816 | rcu_read_lock(); |
309dca30 CH |
1817 | blkg = blkg_lookup_create(css_to_blkcg(css), |
1818 | bio->bi_bdev->bd_disk->queue); | |
28fc591f CH |
1819 | while (blkg) { |
1820 | if (blkg_tryget(blkg)) { | |
1821 | ret_blkg = blkg; | |
1822 | break; | |
1823 | } | |
1824 | blkg = blkg->parent; | |
1825 | } | |
13c7863d | 1826 | rcu_read_unlock(); |
28fc591f CH |
1827 | |
1828 | return ret_blkg; | |
1829 | } | |
1830 | ||
1831 | /** | |
1832 | * bio_associate_blkg_from_css - associate a bio with a specified css | |
1833 | * @bio: target bio | |
1834 | * @css: target css | |
1835 | * | |
1836 | * Associate @bio with the blkg found by combining the css's blkg and the | |
1837 | * request_queue of the @bio. An association failure is handled by walking up | |
1838 | * the blkg tree. Therefore, the blkg associated can be anything between @blkg | |
1839 | * and q->root_blkg. This situation only happens when a cgroup is dying and | |
1840 | * then the remaining bios will spill to the closest alive blkg. | |
1841 | * | |
1842 | * A reference will be taken on the blkg and will be released when @bio is | |
1843 | * freed. | |
1844 | */ | |
1845 | void bio_associate_blkg_from_css(struct bio *bio, | |
1846 | struct cgroup_subsys_state *css) | |
1847 | { | |
28fc591f CH |
1848 | if (bio->bi_blkg) |
1849 | blkg_put(bio->bi_blkg); | |
1850 | ||
a5b97526 | 1851 | if (css && css->parent) { |
13c7863d | 1852 | bio->bi_blkg = blkg_tryget_closest(bio, css); |
a5b97526 | 1853 | } else { |
309dca30 CH |
1854 | blkg_get(bio->bi_bdev->bd_disk->queue->root_blkg); |
1855 | bio->bi_blkg = bio->bi_bdev->bd_disk->queue->root_blkg; | |
a5b97526 | 1856 | } |
28fc591f CH |
1857 | } |
1858 | EXPORT_SYMBOL_GPL(bio_associate_blkg_from_css); | |
1859 | ||
1860 | /** | |
1861 | * bio_associate_blkg - associate a bio with a blkg | |
1862 | * @bio: target bio | |
1863 | * | |
1864 | * Associate @bio with the blkg found from the bio's css and request_queue. | |
1865 | * If one is not found, bio_lookup_blkg() creates the blkg. If a blkg is | |
1866 | * already associated, the css is reused and association redone as the | |
1867 | * request_queue may have changed. | |
1868 | */ | |
1869 | void bio_associate_blkg(struct bio *bio) | |
1870 | { | |
1871 | struct cgroup_subsys_state *css; | |
1872 | ||
1873 | rcu_read_lock(); | |
1874 | ||
1875 | if (bio->bi_blkg) | |
1876 | css = &bio_blkcg(bio)->css; | |
1877 | else | |
1878 | css = blkcg_css(); | |
1879 | ||
1880 | bio_associate_blkg_from_css(bio, css); | |
1881 | ||
1882 | rcu_read_unlock(); | |
1883 | } | |
1884 | EXPORT_SYMBOL_GPL(bio_associate_blkg); | |
1885 | ||
1886 | /** | |
1887 | * bio_clone_blkg_association - clone blkg association from src to dst bio | |
1888 | * @dst: destination bio | |
1889 | * @src: source bio | |
1890 | */ | |
1891 | void bio_clone_blkg_association(struct bio *dst, struct bio *src) | |
1892 | { | |
1893 | if (src->bi_blkg) { | |
1894 | if (dst->bi_blkg) | |
1895 | blkg_put(dst->bi_blkg); | |
1896 | blkg_get(src->bi_blkg); | |
1897 | dst->bi_blkg = src->bi_blkg; | |
1898 | } | |
1899 | } | |
1900 | EXPORT_SYMBOL_GPL(bio_clone_blkg_association); | |
1901 | ||
db18a53e CH |
1902 | static int blk_cgroup_io_type(struct bio *bio) |
1903 | { | |
1904 | if (op_is_discard(bio->bi_opf)) | |
1905 | return BLKG_IOSTAT_DISCARD; | |
1906 | if (op_is_write(bio->bi_opf)) | |
1907 | return BLKG_IOSTAT_WRITE; | |
1908 | return BLKG_IOSTAT_READ; | |
1909 | } | |
1910 | ||
1911 | void blk_cgroup_bio_start(struct bio *bio) | |
1912 | { | |
1913 | int rwd = blk_cgroup_io_type(bio), cpu; | |
1914 | struct blkg_iostat_set *bis; | |
1915 | ||
1916 | cpu = get_cpu(); | |
1917 | bis = per_cpu_ptr(bio->bi_blkg->iostat_cpu, cpu); | |
1918 | u64_stats_update_begin(&bis->sync); | |
1919 | ||
1920 | /* | |
1921 | * If the bio is flagged with BIO_CGROUP_ACCT it means this is a split | |
1922 | * bio and we would have already accounted for the size of the bio. | |
1923 | */ | |
1924 | if (!bio_flagged(bio, BIO_CGROUP_ACCT)) { | |
1925 | bio_set_flag(bio, BIO_CGROUP_ACCT); | |
0b8cc25d | 1926 | bis->cur.bytes[rwd] += bio->bi_iter.bi_size; |
db18a53e CH |
1927 | } |
1928 | bis->cur.ios[rwd]++; | |
1929 | ||
1930 | u64_stats_update_end(&bis->sync); | |
1931 | if (cgroup_subsys_on_dfl(io_cgrp_subsys)) | |
1932 | cgroup_rstat_updated(bio->bi_blkg->blkcg->css.cgroup, cpu); | |
1933 | put_cpu(); | |
1934 | } | |
1935 | ||
d3f77dfd TH |
1936 | static int __init blkcg_init(void) |
1937 | { | |
1938 | blkcg_punt_bio_wq = alloc_workqueue("blkcg_punt_bio", | |
1939 | WQ_MEM_RECLAIM | WQ_FREEZABLE | | |
1940 | WQ_UNBOUND | WQ_SYSFS, 0); | |
1941 | if (!blkcg_punt_bio_wq) | |
1942 | return -ENOMEM; | |
1943 | return 0; | |
1944 | } | |
1945 | subsys_initcall(blkcg_init); | |
1946 | ||
903d23f0 JB |
1947 | module_param(blkcg_debug_stats, bool, 0644); |
1948 | MODULE_PARM_DESC(blkcg_debug_stats, "True if you want debug stats, false if not"); |