Commit | Line | Data |
---|---|---|
3dcf60bc | 1 | // SPDX-License-Identifier: GPL-2.0 |
31e4c28d VG |
2 | /* |
3 | * Common Block IO controller cgroup interface | |
4 | * | |
5 | * Based on ideas and code from CFQ, CFS and BFQ: | |
6 | * Copyright (C) 2003 Jens Axboe <axboe@kernel.dk> | |
7 | * | |
8 | * Copyright (C) 2008 Fabio Checconi <fabio@gandalf.sssup.it> | |
9 | * Paolo Valente <paolo.valente@unimore.it> | |
10 | * | |
11 | * Copyright (C) 2009 Vivek Goyal <vgoyal@redhat.com> | |
12 | * Nauman Rafique <nauman@google.com> | |
e48453c3 AA |
13 | * |
14 | * For policy-specific per-blkcg data: | |
15 | * Copyright (C) 2015 Paolo Valente <paolo.valente@unimore.it> | |
16 | * Arianna Avanzini <avanzini.arianna@gmail.com> | |
31e4c28d VG |
17 | */ |
18 | #include <linux/ioprio.h> | |
22084190 | 19 | #include <linux/kdev_t.h> |
9d6a986c | 20 | #include <linux/module.h> |
174cd4b1 | 21 | #include <linux/sched/signal.h> |
accee785 | 22 | #include <linux/err.h> |
9195291e | 23 | #include <linux/blkdev.h> |
52ebea74 | 24 | #include <linux/backing-dev.h> |
5a0e3ad6 | 25 | #include <linux/slab.h> |
72e06c25 | 26 | #include <linux/delay.h> |
9a9e8a26 | 27 | #include <linux/atomic.h> |
36aa9e5f | 28 | #include <linux/ctype.h> |
03248add | 29 | #include <linux/resume_user_mode.h> |
fd112c74 | 30 | #include <linux/psi.h> |
82d981d4 | 31 | #include <linux/part_stat.h> |
5efd6113 | 32 | #include "blk.h" |
672fdcf0 | 33 | #include "blk-cgroup.h" |
556910e3 | 34 | #include "blk-ioprio.h" |
a7b36ee6 | 35 | #include "blk-throttle.h" |
3e252066 | 36 | |
838f13bf TH |
37 | /* |
38 | * blkcg_pol_mutex protects blkcg_policy[] and policy [de]activation. | |
39 | * blkcg_pol_register_mutex nests outside of it and synchronizes entire | |
40 | * policy [un]register operations including cgroup file additions / | |
41 | * removals. Putting cgroup file registration outside blkcg_pol_mutex | |
42 | * allows grabbing it from cgroup callbacks. | |
43 | */ | |
44 | static DEFINE_MUTEX(blkcg_pol_register_mutex); | |
bc0d6501 | 45 | static DEFINE_MUTEX(blkcg_pol_mutex); |
923adde1 | 46 | |
e48453c3 | 47 | struct blkcg blkcg_root; |
3c798398 | 48 | EXPORT_SYMBOL_GPL(blkcg_root); |
9d6a986c | 49 | |
496d5e75 | 50 | struct cgroup_subsys_state * const blkcg_root_css = &blkcg_root.css; |
9b0eb69b | 51 | EXPORT_SYMBOL_GPL(blkcg_root_css); |
496d5e75 | 52 | |
3c798398 | 53 | static struct blkcg_policy *blkcg_policy[BLKCG_MAX_POLS]; |
035d10b2 | 54 | |
7876f930 TH |
55 | static LIST_HEAD(all_blkcgs); /* protected by blkcg_pol_mutex */ |
56 | ||
07b0fdec | 57 | bool blkcg_debug_stats = false; |
d3f77dfd | 58 | static struct workqueue_struct *blkcg_punt_bio_wq; |
903d23f0 | 59 | |
a731763f YK |
60 | #define BLKG_DESTROY_BATCH_SIZE 64 |
61 | ||
a2b1693b | 62 | static bool blkcg_policy_enabled(struct request_queue *q, |
3c798398 | 63 | const struct blkcg_policy *pol) |
a2b1693b TH |
64 | { |
65 | return pol && test_bit(pol->plid, q->blkcg_pols); | |
66 | } | |
67 | ||
d578c770 | 68 | static void blkg_free_workfn(struct work_struct *work) |
0381411e | 69 | { |
d578c770 ML |
70 | struct blkcg_gq *blkg = container_of(work, struct blkcg_gq, |
71 | free_work); | |
e8989fae | 72 | int i; |
549d3aa8 | 73 | |
db613670 | 74 | for (i = 0; i < BLKCG_MAX_POLS; i++) |
001bea73 TH |
75 | if (blkg->pd[i]) |
76 | blkcg_policy[i]->pd_free_fn(blkg->pd[i]); | |
e8989fae | 77 | |
0a9a25ca ML |
78 | if (blkg->q) |
79 | blk_put_queue(blkg->q); | |
f7331648 | 80 | free_percpu(blkg->iostat_cpu); |
ef069b97 | 81 | percpu_ref_exit(&blkg->refcnt); |
549d3aa8 | 82 | kfree(blkg); |
0381411e TH |
83 | } |
84 | ||
d578c770 ML |
85 | /** |
86 | * blkg_free - free a blkg | |
87 | * @blkg: blkg to free | |
88 | * | |
89 | * Free @blkg which may be partially allocated. | |
90 | */ | |
91 | static void blkg_free(struct blkcg_gq *blkg) | |
92 | { | |
93 | if (!blkg) | |
94 | return; | |
95 | ||
96 | /* | |
97 | * Both ->pd_free_fn() and request queue's release handler may | |
98 | * sleep, so free us by scheduling one work func | |
99 | */ | |
100 | INIT_WORK(&blkg->free_work, blkg_free_workfn); | |
101 | schedule_work(&blkg->free_work); | |
102 | } | |
103 | ||
7fcf2b03 DZ |
104 | static void __blkg_release(struct rcu_head *rcu) |
105 | { | |
106 | struct blkcg_gq *blkg = container_of(rcu, struct blkcg_gq, rcu_head); | |
107 | ||
d3f77dfd TH |
108 | WARN_ON(!bio_list_empty(&blkg->async_bios)); |
109 | ||
7fcf2b03 DZ |
110 | /* release the blkcg and parent blkg refs this blkg has been holding */ |
111 | css_put(&blkg->blkcg->css); | |
112 | if (blkg->parent) | |
113 | blkg_put(blkg->parent); | |
7fcf2b03 DZ |
114 | blkg_free(blkg); |
115 | } | |
116 | ||
117 | /* | |
118 | * A group is RCU protected, but having an rcu lock does not mean that one | |
119 | * can access all the fields of blkg and assume these are valid. For | |
120 | * example, don't try to follow throtl_data and request queue links. | |
121 | * | |
122 | * Having a reference to blkg under an rcu allows accesses to only values | |
123 | * local to groups like group stats and group rate limits. | |
124 | */ | |
125 | static void blkg_release(struct percpu_ref *ref) | |
126 | { | |
127 | struct blkcg_gq *blkg = container_of(ref, struct blkcg_gq, refcnt); | |
128 | ||
129 | call_rcu(&blkg->rcu_head, __blkg_release); | |
130 | } | |
131 | ||
d3f77dfd TH |
132 | static void blkg_async_bio_workfn(struct work_struct *work) |
133 | { | |
134 | struct blkcg_gq *blkg = container_of(work, struct blkcg_gq, | |
135 | async_bio_work); | |
136 | struct bio_list bios = BIO_EMPTY_LIST; | |
137 | struct bio *bio; | |
192f1c6b XT |
138 | struct blk_plug plug; |
139 | bool need_plug = false; | |
d3f77dfd TH |
140 | |
141 | /* as long as there are pending bios, @blkg can't go away */ | |
142 | spin_lock_bh(&blkg->async_bio_lock); | |
143 | bio_list_merge(&bios, &blkg->async_bios); | |
144 | bio_list_init(&blkg->async_bios); | |
145 | spin_unlock_bh(&blkg->async_bio_lock); | |
146 | ||
192f1c6b XT |
147 | /* start plug only when bio_list contains at least 2 bios */ |
148 | if (bios.head && bios.head->bi_next) { | |
149 | need_plug = true; | |
150 | blk_start_plug(&plug); | |
151 | } | |
d3f77dfd TH |
152 | while ((bio = bio_list_pop(&bios))) |
153 | submit_bio(bio); | |
192f1c6b XT |
154 | if (need_plug) |
155 | blk_finish_plug(&plug); | |
d3f77dfd TH |
156 | } |
157 | ||
0381411e TH |
158 | /** |
159 | * blkg_alloc - allocate a blkg | |
160 | * @blkcg: block cgroup the new blkg is associated with | |
161 | * @q: request_queue the new blkg is associated with | |
15974993 | 162 | * @gfp_mask: allocation mask to use |
0381411e | 163 | * |
e8989fae | 164 | * Allocate a new blkg assocating @blkcg and @q. |
0381411e | 165 | */ |
15974993 TH |
166 | static struct blkcg_gq *blkg_alloc(struct blkcg *blkcg, struct request_queue *q, |
167 | gfp_t gfp_mask) | |
0381411e | 168 | { |
3c798398 | 169 | struct blkcg_gq *blkg; |
f7331648 | 170 | int i, cpu; |
0381411e TH |
171 | |
172 | /* alloc and init base part */ | |
15974993 | 173 | blkg = kzalloc_node(sizeof(*blkg), gfp_mask, q->node); |
0381411e TH |
174 | if (!blkg) |
175 | return NULL; | |
176 | ||
ef069b97 TH |
177 | if (percpu_ref_init(&blkg->refcnt, blkg_release, 0, gfp_mask)) |
178 | goto err_free; | |
179 | ||
f7331648 TH |
180 | blkg->iostat_cpu = alloc_percpu_gfp(struct blkg_iostat_set, gfp_mask); |
181 | if (!blkg->iostat_cpu) | |
77ea7338 TH |
182 | goto err_free; |
183 | ||
0a9a25ca ML |
184 | if (!blk_get_queue(q)) |
185 | goto err_free; | |
186 | ||
c875f4d0 | 187 | blkg->q = q; |
e8989fae | 188 | INIT_LIST_HEAD(&blkg->q_node); |
d3f77dfd TH |
189 | spin_lock_init(&blkg->async_bio_lock); |
190 | bio_list_init(&blkg->async_bios); | |
191 | INIT_WORK(&blkg->async_bio_work, blkg_async_bio_workfn); | |
0381411e | 192 | blkg->blkcg = blkcg; |
0381411e | 193 | |
f7331648 TH |
194 | u64_stats_init(&blkg->iostat.sync); |
195 | for_each_possible_cpu(cpu) | |
196 | u64_stats_init(&per_cpu_ptr(blkg->iostat_cpu, cpu)->sync); | |
197 | ||
8bd435b3 | 198 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
3c798398 | 199 | struct blkcg_policy *pol = blkcg_policy[i]; |
e8989fae | 200 | struct blkg_policy_data *pd; |
0381411e | 201 | |
a2b1693b | 202 | if (!blkcg_policy_enabled(q, pol)) |
e8989fae TH |
203 | continue; |
204 | ||
205 | /* alloc per-policy data and attach it to blkg */ | |
cf09a8ee | 206 | pd = pol->pd_alloc_fn(gfp_mask, q, blkcg); |
a051661c TH |
207 | if (!pd) |
208 | goto err_free; | |
549d3aa8 | 209 | |
e8989fae TH |
210 | blkg->pd[i] = pd; |
211 | pd->blkg = blkg; | |
b276a876 | 212 | pd->plid = i; |
e8989fae TH |
213 | } |
214 | ||
0381411e | 215 | return blkg; |
a051661c TH |
216 | |
217 | err_free: | |
218 | blkg_free(blkg); | |
219 | return NULL; | |
0381411e TH |
220 | } |
221 | ||
24f29046 TH |
222 | struct blkcg_gq *blkg_lookup_slowpath(struct blkcg *blkcg, |
223 | struct request_queue *q, bool update_hint) | |
80fd9979 | 224 | { |
3c798398 | 225 | struct blkcg_gq *blkg; |
80fd9979 | 226 | |
a637120e | 227 | /* |
86cde6b6 TH |
228 | * Hint didn't match. Look up from the radix tree. Note that the |
229 | * hint can only be updated under queue_lock as otherwise @blkg | |
230 | * could have already been removed from blkg_tree. The caller is | |
231 | * responsible for grabbing queue_lock if @update_hint. | |
a637120e TH |
232 | */ |
233 | blkg = radix_tree_lookup(&blkcg->blkg_tree, q->id); | |
86cde6b6 TH |
234 | if (blkg && blkg->q == q) { |
235 | if (update_hint) { | |
0d945c1f | 236 | lockdep_assert_held(&q->queue_lock); |
86cde6b6 TH |
237 | rcu_assign_pointer(blkcg->blkg_hint, blkg); |
238 | } | |
a637120e | 239 | return blkg; |
86cde6b6 | 240 | } |
a637120e | 241 | |
80fd9979 TH |
242 | return NULL; |
243 | } | |
ae118896 | 244 | EXPORT_SYMBOL_GPL(blkg_lookup_slowpath); |
80fd9979 | 245 | |
15974993 | 246 | /* |
d708f0d5 JA |
247 | * If @new_blkg is %NULL, this function tries to allocate a new one as |
248 | * necessary using %GFP_NOWAIT. @new_blkg is always consumed on return. | |
15974993 | 249 | */ |
86cde6b6 | 250 | static struct blkcg_gq *blkg_create(struct blkcg *blkcg, |
d708f0d5 JA |
251 | struct request_queue *q, |
252 | struct blkcg_gq *new_blkg) | |
5624a4e4 | 253 | { |
d708f0d5 | 254 | struct blkcg_gq *blkg; |
f427d909 | 255 | int i, ret; |
5624a4e4 | 256 | |
cd1604fa | 257 | WARN_ON_ONCE(!rcu_read_lock_held()); |
0d945c1f | 258 | lockdep_assert_held(&q->queue_lock); |
cd1604fa | 259 | |
0273ac34 DZ |
260 | /* request_queue is dying, do not create/recreate a blkg */ |
261 | if (blk_queue_dying(q)) { | |
262 | ret = -ENODEV; | |
263 | goto err_free_blkg; | |
264 | } | |
265 | ||
7ee9c562 | 266 | /* blkg holds a reference to blkcg */ |
ec903c0c | 267 | if (!css_tryget_online(&blkcg->css)) { |
20386ce0 | 268 | ret = -ENODEV; |
93e6d5d8 | 269 | goto err_free_blkg; |
15974993 | 270 | } |
cd1604fa | 271 | |
d708f0d5 JA |
272 | /* allocate */ |
273 | if (!new_blkg) { | |
274 | new_blkg = blkg_alloc(blkcg, q, GFP_NOWAIT | __GFP_NOWARN); | |
275 | if (unlikely(!new_blkg)) { | |
276 | ret = -ENOMEM; | |
8c911f3d | 277 | goto err_put_css; |
15974993 TH |
278 | } |
279 | } | |
d708f0d5 | 280 | blkg = new_blkg; |
cd1604fa | 281 | |
db613670 | 282 | /* link parent */ |
3c547865 TH |
283 | if (blkcg_parent(blkcg)) { |
284 | blkg->parent = __blkg_lookup(blkcg_parent(blkcg), q, false); | |
285 | if (WARN_ON_ONCE(!blkg->parent)) { | |
20386ce0 | 286 | ret = -ENODEV; |
8c911f3d | 287 | goto err_put_css; |
3c547865 TH |
288 | } |
289 | blkg_get(blkg->parent); | |
290 | } | |
291 | ||
db613670 TH |
292 | /* invoke per-policy init */ |
293 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
294 | struct blkcg_policy *pol = blkcg_policy[i]; | |
295 | ||
296 | if (blkg->pd[i] && pol->pd_init_fn) | |
a9520cd6 | 297 | pol->pd_init_fn(blkg->pd[i]); |
db613670 TH |
298 | } |
299 | ||
300 | /* insert */ | |
cd1604fa | 301 | spin_lock(&blkcg->lock); |
a637120e TH |
302 | ret = radix_tree_insert(&blkcg->blkg_tree, q->id, blkg); |
303 | if (likely(!ret)) { | |
304 | hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list); | |
305 | list_add(&blkg->q_node, &q->blkg_list); | |
f427d909 TH |
306 | |
307 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
308 | struct blkcg_policy *pol = blkcg_policy[i]; | |
309 | ||
310 | if (blkg->pd[i] && pol->pd_online_fn) | |
a9520cd6 | 311 | pol->pd_online_fn(blkg->pd[i]); |
f427d909 | 312 | } |
a637120e | 313 | } |
f427d909 | 314 | blkg->online = true; |
cd1604fa | 315 | spin_unlock(&blkcg->lock); |
496fb780 | 316 | |
ec13b1d6 | 317 | if (!ret) |
a637120e | 318 | return blkg; |
15974993 | 319 | |
3c547865 TH |
320 | /* @blkg failed fully initialized, use the usual release path */ |
321 | blkg_put(blkg); | |
322 | return ERR_PTR(ret); | |
323 | ||
d708f0d5 | 324 | err_put_css: |
496fb780 | 325 | css_put(&blkcg->css); |
93e6d5d8 | 326 | err_free_blkg: |
d708f0d5 | 327 | blkg_free(new_blkg); |
93e6d5d8 | 328 | return ERR_PTR(ret); |
31e4c28d | 329 | } |
3c96cb32 | 330 | |
86cde6b6 | 331 | /** |
8c546287 | 332 | * blkg_lookup_create - lookup blkg, try to create one if not there |
86cde6b6 TH |
333 | * @blkcg: blkcg of interest |
334 | * @q: request_queue of interest | |
335 | * | |
336 | * Lookup blkg for the @blkcg - @q pair. If it doesn't exist, try to | |
3c547865 TH |
337 | * create one. blkg creation is performed recursively from blkcg_root such |
338 | * that all non-root blkg's have access to the parent blkg. This function | |
8c546287 | 339 | * should be called under RCU read lock and takes @q->queue_lock. |
86cde6b6 | 340 | * |
beea9da0 DZ |
341 | * Returns the blkg or the closest blkg if blkg_create() fails as it walks |
342 | * down from root. | |
86cde6b6 | 343 | */ |
8c546287 CH |
344 | static struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg, |
345 | struct request_queue *q) | |
3c96cb32 | 346 | { |
86cde6b6 | 347 | struct blkcg_gq *blkg; |
8c546287 | 348 | unsigned long flags; |
86cde6b6 TH |
349 | |
350 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
86cde6b6 | 351 | |
8c546287 | 352 | blkg = blkg_lookup(blkcg, q); |
86cde6b6 TH |
353 | if (blkg) |
354 | return blkg; | |
355 | ||
8c546287 CH |
356 | spin_lock_irqsave(&q->queue_lock, flags); |
357 | blkg = __blkg_lookup(blkcg, q, true); | |
358 | if (blkg) | |
359 | goto found; | |
360 | ||
3c547865 TH |
361 | /* |
362 | * Create blkgs walking down from blkcg_root to @blkcg, so that all | |
beea9da0 DZ |
363 | * non-root blkgs have access to their parents. Returns the closest |
364 | * blkg to the intended blkg should blkg_create() fail. | |
3c547865 TH |
365 | */ |
366 | while (true) { | |
367 | struct blkcg *pos = blkcg; | |
368 | struct blkcg *parent = blkcg_parent(blkcg); | |
beea9da0 DZ |
369 | struct blkcg_gq *ret_blkg = q->root_blkg; |
370 | ||
371 | while (parent) { | |
372 | blkg = __blkg_lookup(parent, q, false); | |
373 | if (blkg) { | |
374 | /* remember closest blkg */ | |
375 | ret_blkg = blkg; | |
376 | break; | |
377 | } | |
3c547865 TH |
378 | pos = parent; |
379 | parent = blkcg_parent(parent); | |
380 | } | |
381 | ||
d708f0d5 | 382 | blkg = blkg_create(pos, q, NULL); |
8c546287 CH |
383 | if (IS_ERR(blkg)) { |
384 | blkg = ret_blkg; | |
385 | break; | |
386 | } | |
beea9da0 | 387 | if (pos == blkcg) |
8c546287 | 388 | break; |
b978962a DZ |
389 | } |
390 | ||
8c546287 CH |
391 | found: |
392 | spin_unlock_irqrestore(&q->queue_lock, flags); | |
b978962a DZ |
393 | return blkg; |
394 | } | |
395 | ||
3c798398 | 396 | static void blkg_destroy(struct blkcg_gq *blkg) |
03aa264a | 397 | { |
3c798398 | 398 | struct blkcg *blkcg = blkg->blkcg; |
6b065462 | 399 | int i; |
03aa264a | 400 | |
0d945c1f | 401 | lockdep_assert_held(&blkg->q->queue_lock); |
9f13ef67 | 402 | lockdep_assert_held(&blkcg->lock); |
03aa264a TH |
403 | |
404 | /* Something wrong if we are trying to remove same group twice */ | |
e8989fae | 405 | WARN_ON_ONCE(list_empty(&blkg->q_node)); |
9f13ef67 | 406 | WARN_ON_ONCE(hlist_unhashed(&blkg->blkcg_node)); |
a637120e | 407 | |
6b065462 DZF |
408 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
409 | struct blkcg_policy *pol = blkcg_policy[i]; | |
410 | ||
411 | if (blkg->pd[i] && pol->pd_offline_fn) | |
412 | pol->pd_offline_fn(blkg->pd[i]); | |
413 | } | |
414 | ||
f427d909 TH |
415 | blkg->online = false; |
416 | ||
a637120e | 417 | radix_tree_delete(&blkcg->blkg_tree, blkg->q->id); |
e8989fae | 418 | list_del_init(&blkg->q_node); |
9f13ef67 | 419 | hlist_del_init_rcu(&blkg->blkcg_node); |
03aa264a | 420 | |
a637120e TH |
421 | /* |
422 | * Both setting lookup hint to and clearing it from @blkg are done | |
423 | * under queue_lock. If it's not pointing to @blkg now, it never | |
424 | * will. Hint assignment itself can race safely. | |
425 | */ | |
ec6c676a | 426 | if (rcu_access_pointer(blkcg->blkg_hint) == blkg) |
a637120e TH |
427 | rcu_assign_pointer(blkcg->blkg_hint, NULL); |
428 | ||
03aa264a TH |
429 | /* |
430 | * Put the reference taken at the time of creation so that when all | |
431 | * queues are gone, group can be destroyed. | |
432 | */ | |
7fcf2b03 | 433 | percpu_ref_kill(&blkg->refcnt); |
03aa264a TH |
434 | } |
435 | ||
9f13ef67 TH |
436 | /** |
437 | * blkg_destroy_all - destroy all blkgs associated with a request_queue | |
438 | * @q: request_queue of interest | |
9f13ef67 | 439 | * |
3c96cb32 | 440 | * Destroy all blkgs associated with @q. |
9f13ef67 | 441 | */ |
3c96cb32 | 442 | static void blkg_destroy_all(struct request_queue *q) |
72e06c25 | 443 | { |
3c798398 | 444 | struct blkcg_gq *blkg, *n; |
a731763f | 445 | int count = BLKG_DESTROY_BATCH_SIZE; |
72e06c25 | 446 | |
a731763f | 447 | restart: |
0d945c1f | 448 | spin_lock_irq(&q->queue_lock); |
9f13ef67 | 449 | list_for_each_entry_safe(blkg, n, &q->blkg_list, q_node) { |
3c798398 | 450 | struct blkcg *blkcg = blkg->blkcg; |
72e06c25 | 451 | |
9f13ef67 TH |
452 | spin_lock(&blkcg->lock); |
453 | blkg_destroy(blkg); | |
454 | spin_unlock(&blkcg->lock); | |
a731763f YK |
455 | |
456 | /* | |
457 | * in order to avoid holding the spin lock for too long, release | |
458 | * it when a batch of blkgs are destroyed. | |
459 | */ | |
460 | if (!(--count)) { | |
461 | count = BLKG_DESTROY_BATCH_SIZE; | |
462 | spin_unlock_irq(&q->queue_lock); | |
463 | cond_resched(); | |
464 | goto restart; | |
465 | } | |
72e06c25 | 466 | } |
6fe810bd TH |
467 | |
468 | q->root_blkg = NULL; | |
0d945c1f | 469 | spin_unlock_irq(&q->queue_lock); |
72e06c25 TH |
470 | } |
471 | ||
182446d0 TH |
472 | static int blkcg_reset_stats(struct cgroup_subsys_state *css, |
473 | struct cftype *cftype, u64 val) | |
303a3acb | 474 | { |
182446d0 | 475 | struct blkcg *blkcg = css_to_blkcg(css); |
3c798398 | 476 | struct blkcg_gq *blkg; |
f7331648 | 477 | int i, cpu; |
303a3acb | 478 | |
838f13bf | 479 | mutex_lock(&blkcg_pol_mutex); |
303a3acb | 480 | spin_lock_irq(&blkcg->lock); |
997a026c TH |
481 | |
482 | /* | |
483 | * Note that stat reset is racy - it doesn't synchronize against | |
484 | * stat updates. This is a debug feature which shouldn't exist | |
485 | * anyway. If you get hit by a race, retry. | |
486 | */ | |
b67bfe0d | 487 | hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) { |
f7331648 TH |
488 | for_each_possible_cpu(cpu) { |
489 | struct blkg_iostat_set *bis = | |
490 | per_cpu_ptr(blkg->iostat_cpu, cpu); | |
491 | memset(bis, 0, sizeof(*bis)); | |
492 | } | |
493 | memset(&blkg->iostat, 0, sizeof(blkg->iostat)); | |
77ea7338 | 494 | |
8bd435b3 | 495 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
3c798398 | 496 | struct blkcg_policy *pol = blkcg_policy[i]; |
549d3aa8 | 497 | |
a9520cd6 TH |
498 | if (blkg->pd[i] && pol->pd_reset_stats_fn) |
499 | pol->pd_reset_stats_fn(blkg->pd[i]); | |
bc0d6501 | 500 | } |
303a3acb | 501 | } |
f0bdc8cd | 502 | |
303a3acb | 503 | spin_unlock_irq(&blkcg->lock); |
bc0d6501 | 504 | mutex_unlock(&blkcg_pol_mutex); |
303a3acb DS |
505 | return 0; |
506 | } | |
507 | ||
dd165eb3 | 508 | const char *blkg_dev_name(struct blkcg_gq *blkg) |
303a3acb | 509 | { |
d152c682 | 510 | if (!blkg->q->disk || !blkg->q->disk->bdi->dev) |
edb0872f | 511 | return NULL; |
d152c682 | 512 | return bdi_dev_name(blkg->q->disk->bdi); |
303a3acb DS |
513 | } |
514 | ||
d3d32e69 TH |
515 | /** |
516 | * blkcg_print_blkgs - helper for printing per-blkg data | |
517 | * @sf: seq_file to print to | |
518 | * @blkcg: blkcg of interest | |
519 | * @prfill: fill function to print out a blkg | |
520 | * @pol: policy in question | |
521 | * @data: data to be passed to @prfill | |
522 | * @show_total: to print out sum of prfill return values or not | |
523 | * | |
524 | * This function invokes @prfill on each blkg of @blkcg if pd for the | |
525 | * policy specified by @pol exists. @prfill is invoked with @sf, the | |
810ecfa7 TH |
526 | * policy data and @data and the matching queue lock held. If @show_total |
527 | * is %true, the sum of the return values from @prfill is printed with | |
528 | * "Total" label at the end. | |
d3d32e69 TH |
529 | * |
530 | * This is to be used to construct print functions for | |
531 | * cftype->read_seq_string method. | |
532 | */ | |
3c798398 | 533 | void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg, |
f95a04af TH |
534 | u64 (*prfill)(struct seq_file *, |
535 | struct blkg_policy_data *, int), | |
3c798398 | 536 | const struct blkcg_policy *pol, int data, |
ec399347 | 537 | bool show_total) |
5624a4e4 | 538 | { |
3c798398 | 539 | struct blkcg_gq *blkg; |
d3d32e69 | 540 | u64 total = 0; |
5624a4e4 | 541 | |
810ecfa7 | 542 | rcu_read_lock(); |
ee89f812 | 543 | hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) { |
0d945c1f | 544 | spin_lock_irq(&blkg->q->queue_lock); |
a2b1693b | 545 | if (blkcg_policy_enabled(blkg->q, pol)) |
f95a04af | 546 | total += prfill(sf, blkg->pd[pol->plid], data); |
0d945c1f | 547 | spin_unlock_irq(&blkg->q->queue_lock); |
810ecfa7 TH |
548 | } |
549 | rcu_read_unlock(); | |
d3d32e69 TH |
550 | |
551 | if (show_total) | |
552 | seq_printf(sf, "Total %llu\n", (unsigned long long)total); | |
553 | } | |
829fdb50 | 554 | EXPORT_SYMBOL_GPL(blkcg_print_blkgs); |
d3d32e69 TH |
555 | |
556 | /** | |
557 | * __blkg_prfill_u64 - prfill helper for a single u64 value | |
558 | * @sf: seq_file to print to | |
f95a04af | 559 | * @pd: policy private data of interest |
d3d32e69 TH |
560 | * @v: value to print |
561 | * | |
f95a04af | 562 | * Print @v to @sf for the device assocaited with @pd. |
d3d32e69 | 563 | */ |
f95a04af | 564 | u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v) |
d3d32e69 | 565 | { |
f95a04af | 566 | const char *dname = blkg_dev_name(pd->blkg); |
d3d32e69 TH |
567 | |
568 | if (!dname) | |
569 | return 0; | |
570 | ||
571 | seq_printf(sf, "%s %llu\n", dname, (unsigned long long)v); | |
572 | return v; | |
573 | } | |
829fdb50 | 574 | EXPORT_SYMBOL_GPL(__blkg_prfill_u64); |
d3d32e69 | 575 | |
457e490f TE |
576 | /* Performs queue bypass and policy enabled checks then looks up blkg. */ |
577 | static struct blkcg_gq *blkg_lookup_check(struct blkcg *blkcg, | |
578 | const struct blkcg_policy *pol, | |
579 | struct request_queue *q) | |
580 | { | |
581 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
0d945c1f | 582 | lockdep_assert_held(&q->queue_lock); |
457e490f TE |
583 | |
584 | if (!blkcg_policy_enabled(q, pol)) | |
585 | return ERR_PTR(-EOPNOTSUPP); | |
457e490f TE |
586 | return __blkg_lookup(blkcg, q, true /* update_hint */); |
587 | } | |
588 | ||
015d254c | 589 | /** |
22ae8ce8 | 590 | * blkcg_conf_open_bdev - parse and open bdev for per-blkg config update |
015d254c TH |
591 | * @inputp: input string pointer |
592 | * | |
593 | * Parse the device node prefix part, MAJ:MIN, of per-blkg config update | |
22ae8ce8 | 594 | * from @input and get and return the matching bdev. *@inputp is |
015d254c TH |
595 | * updated to point past the device node prefix. Returns an ERR_PTR() |
596 | * value on error. | |
597 | * | |
598 | * Use this function iff blkg_conf_prep() can't be used for some reason. | |
599 | */ | |
22ae8ce8 | 600 | struct block_device *blkcg_conf_open_bdev(char **inputp) |
015d254c TH |
601 | { |
602 | char *input = *inputp; | |
603 | unsigned int major, minor; | |
22ae8ce8 CH |
604 | struct block_device *bdev; |
605 | int key_len; | |
015d254c TH |
606 | |
607 | if (sscanf(input, "%u:%u%n", &major, &minor, &key_len) != 2) | |
608 | return ERR_PTR(-EINVAL); | |
609 | ||
610 | input += key_len; | |
611 | if (!isspace(*input)) | |
612 | return ERR_PTR(-EINVAL); | |
613 | input = skip_spaces(input); | |
614 | ||
22ae8ce8 CH |
615 | bdev = blkdev_get_no_open(MKDEV(major, minor)); |
616 | if (!bdev) | |
015d254c | 617 | return ERR_PTR(-ENODEV); |
22ae8ce8 CH |
618 | if (bdev_is_partition(bdev)) { |
619 | blkdev_put_no_open(bdev); | |
015d254c TH |
620 | return ERR_PTR(-ENODEV); |
621 | } | |
622 | ||
623 | *inputp = input; | |
22ae8ce8 | 624 | return bdev; |
015d254c TH |
625 | } |
626 | ||
3a8b31d3 TH |
627 | /** |
628 | * blkg_conf_prep - parse and prepare for per-blkg config update | |
629 | * @blkcg: target block cgroup | |
da8b0662 | 630 | * @pol: target policy |
3a8b31d3 TH |
631 | * @input: input string |
632 | * @ctx: blkg_conf_ctx to be filled | |
633 | * | |
634 | * Parse per-blkg config update from @input and initialize @ctx with the | |
36aa9e5f TH |
635 | * result. @ctx->blkg points to the blkg to be updated and @ctx->body the |
636 | * part of @input following MAJ:MIN. This function returns with RCU read | |
637 | * lock and queue lock held and must be paired with blkg_conf_finish(). | |
3a8b31d3 | 638 | */ |
3c798398 | 639 | int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol, |
36aa9e5f | 640 | char *input, struct blkg_conf_ctx *ctx) |
ed6cddef | 641 | __acquires(rcu) __acquires(&bdev->bd_queue->queue_lock) |
34d0f179 | 642 | { |
22ae8ce8 | 643 | struct block_device *bdev; |
457e490f | 644 | struct request_queue *q; |
3c798398 | 645 | struct blkcg_gq *blkg; |
015d254c | 646 | int ret; |
36aa9e5f | 647 | |
22ae8ce8 CH |
648 | bdev = blkcg_conf_open_bdev(&input); |
649 | if (IS_ERR(bdev)) | |
650 | return PTR_ERR(bdev); | |
e56da7e2 | 651 | |
ed6cddef | 652 | q = bdev_get_queue(bdev); |
da8b0662 | 653 | |
0c9d338c YK |
654 | /* |
655 | * blkcg_deactivate_policy() requires queue to be frozen, we can grab | |
656 | * q_usage_counter to prevent concurrent with blkcg_deactivate_policy(). | |
657 | */ | |
658 | ret = blk_queue_enter(q, 0); | |
659 | if (ret) | |
15c30104 | 660 | goto fail; |
0c9d338c | 661 | |
457e490f | 662 | rcu_read_lock(); |
0d945c1f | 663 | spin_lock_irq(&q->queue_lock); |
e56da7e2 | 664 | |
457e490f | 665 | blkg = blkg_lookup_check(blkcg, pol, q); |
4bfd482e TH |
666 | if (IS_ERR(blkg)) { |
667 | ret = PTR_ERR(blkg); | |
457e490f TE |
668 | goto fail_unlock; |
669 | } | |
670 | ||
671 | if (blkg) | |
672 | goto success; | |
673 | ||
674 | /* | |
675 | * Create blkgs walking down from blkcg_root to @blkcg, so that all | |
676 | * non-root blkgs have access to their parents. | |
677 | */ | |
678 | while (true) { | |
679 | struct blkcg *pos = blkcg; | |
680 | struct blkcg *parent; | |
681 | struct blkcg_gq *new_blkg; | |
682 | ||
683 | parent = blkcg_parent(blkcg); | |
684 | while (parent && !__blkg_lookup(parent, q, false)) { | |
685 | pos = parent; | |
686 | parent = blkcg_parent(parent); | |
687 | } | |
688 | ||
689 | /* Drop locks to do new blkg allocation with GFP_KERNEL. */ | |
0d945c1f | 690 | spin_unlock_irq(&q->queue_lock); |
3a8b31d3 | 691 | rcu_read_unlock(); |
457e490f TE |
692 | |
693 | new_blkg = blkg_alloc(pos, q, GFP_KERNEL); | |
694 | if (unlikely(!new_blkg)) { | |
695 | ret = -ENOMEM; | |
15c30104 | 696 | goto fail_exit_queue; |
7702e8f4 | 697 | } |
3a8b31d3 | 698 | |
f255c19b GKB |
699 | if (radix_tree_preload(GFP_KERNEL)) { |
700 | blkg_free(new_blkg); | |
701 | ret = -ENOMEM; | |
15c30104 | 702 | goto fail_exit_queue; |
f255c19b GKB |
703 | } |
704 | ||
457e490f | 705 | rcu_read_lock(); |
0d945c1f | 706 | spin_lock_irq(&q->queue_lock); |
457e490f TE |
707 | |
708 | blkg = blkg_lookup_check(pos, pol, q); | |
709 | if (IS_ERR(blkg)) { | |
710 | ret = PTR_ERR(blkg); | |
52abfcbd | 711 | blkg_free(new_blkg); |
f255c19b | 712 | goto fail_preloaded; |
457e490f TE |
713 | } |
714 | ||
715 | if (blkg) { | |
716 | blkg_free(new_blkg); | |
717 | } else { | |
718 | blkg = blkg_create(pos, q, new_blkg); | |
98d669b4 | 719 | if (IS_ERR(blkg)) { |
457e490f | 720 | ret = PTR_ERR(blkg); |
f255c19b | 721 | goto fail_preloaded; |
457e490f TE |
722 | } |
723 | } | |
724 | ||
f255c19b GKB |
725 | radix_tree_preload_end(); |
726 | ||
457e490f TE |
727 | if (pos == blkcg) |
728 | goto success; | |
729 | } | |
730 | success: | |
0c9d338c | 731 | blk_queue_exit(q); |
22ae8ce8 | 732 | ctx->bdev = bdev; |
3a8b31d3 | 733 | ctx->blkg = blkg; |
015d254c | 734 | ctx->body = input; |
726fa694 | 735 | return 0; |
457e490f | 736 | |
f255c19b GKB |
737 | fail_preloaded: |
738 | radix_tree_preload_end(); | |
457e490f | 739 | fail_unlock: |
0d945c1f | 740 | spin_unlock_irq(&q->queue_lock); |
457e490f | 741 | rcu_read_unlock(); |
15c30104 YK |
742 | fail_exit_queue: |
743 | blk_queue_exit(q); | |
457e490f | 744 | fail: |
22ae8ce8 | 745 | blkdev_put_no_open(bdev); |
457e490f TE |
746 | /* |
747 | * If queue was bypassing, we should retry. Do so after a | |
748 | * short msleep(). It isn't strictly necessary but queue | |
749 | * can be bypassing for some time and it's always nice to | |
750 | * avoid busy looping. | |
751 | */ | |
752 | if (ret == -EBUSY) { | |
753 | msleep(10); | |
754 | ret = restart_syscall(); | |
755 | } | |
756 | return ret; | |
34d0f179 | 757 | } |
89f3b6d6 | 758 | EXPORT_SYMBOL_GPL(blkg_conf_prep); |
34d0f179 | 759 | |
3a8b31d3 TH |
760 | /** |
761 | * blkg_conf_finish - finish up per-blkg config update | |
762 | * @ctx: blkg_conf_ctx intiailized by blkg_conf_prep() | |
763 | * | |
764 | * Finish up after per-blkg config update. This function must be paired | |
765 | * with blkg_conf_prep(). | |
766 | */ | |
829fdb50 | 767 | void blkg_conf_finish(struct blkg_conf_ctx *ctx) |
ed6cddef | 768 | __releases(&ctx->bdev->bd_queue->queue_lock) __releases(rcu) |
34d0f179 | 769 | { |
ed6cddef | 770 | spin_unlock_irq(&bdev_get_queue(ctx->bdev)->queue_lock); |
3a8b31d3 | 771 | rcu_read_unlock(); |
22ae8ce8 | 772 | blkdev_put_no_open(ctx->bdev); |
34d0f179 | 773 | } |
89f3b6d6 | 774 | EXPORT_SYMBOL_GPL(blkg_conf_finish); |
34d0f179 | 775 | |
cd1fc4b9 BB |
776 | static void blkg_iostat_set(struct blkg_iostat *dst, struct blkg_iostat *src) |
777 | { | |
778 | int i; | |
779 | ||
780 | for (i = 0; i < BLKG_IOSTAT_NR; i++) { | |
781 | dst->bytes[i] = src->bytes[i]; | |
782 | dst->ios[i] = src->ios[i]; | |
783 | } | |
784 | } | |
785 | ||
786 | static void blkg_iostat_add(struct blkg_iostat *dst, struct blkg_iostat *src) | |
787 | { | |
788 | int i; | |
789 | ||
790 | for (i = 0; i < BLKG_IOSTAT_NR; i++) { | |
791 | dst->bytes[i] += src->bytes[i]; | |
792 | dst->ios[i] += src->ios[i]; | |
793 | } | |
794 | } | |
795 | ||
796 | static void blkg_iostat_sub(struct blkg_iostat *dst, struct blkg_iostat *src) | |
797 | { | |
798 | int i; | |
799 | ||
800 | for (i = 0; i < BLKG_IOSTAT_NR; i++) { | |
801 | dst->bytes[i] -= src->bytes[i]; | |
802 | dst->ios[i] -= src->ios[i]; | |
803 | } | |
804 | } | |
805 | ||
806 | static void blkcg_rstat_flush(struct cgroup_subsys_state *css, int cpu) | |
807 | { | |
808 | struct blkcg *blkcg = css_to_blkcg(css); | |
809 | struct blkcg_gq *blkg; | |
810 | ||
dc26532a JW |
811 | /* Root-level stats are sourced from system-wide IO stats */ |
812 | if (!cgroup_parent(css->cgroup)) | |
813 | return; | |
814 | ||
cd1fc4b9 BB |
815 | rcu_read_lock(); |
816 | ||
817 | hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) { | |
818 | struct blkcg_gq *parent = blkg->parent; | |
819 | struct blkg_iostat_set *bisc = per_cpu_ptr(blkg->iostat_cpu, cpu); | |
820 | struct blkg_iostat cur, delta; | |
c3df5fb5 | 821 | unsigned long flags; |
cd1fc4b9 BB |
822 | unsigned int seq; |
823 | ||
824 | /* fetch the current per-cpu values */ | |
825 | do { | |
826 | seq = u64_stats_fetch_begin(&bisc->sync); | |
827 | blkg_iostat_set(&cur, &bisc->cur); | |
828 | } while (u64_stats_fetch_retry(&bisc->sync, seq)); | |
829 | ||
830 | /* propagate percpu delta to global */ | |
c3df5fb5 | 831 | flags = u64_stats_update_begin_irqsave(&blkg->iostat.sync); |
cd1fc4b9 BB |
832 | blkg_iostat_set(&delta, &cur); |
833 | blkg_iostat_sub(&delta, &bisc->last); | |
834 | blkg_iostat_add(&blkg->iostat.cur, &delta); | |
835 | blkg_iostat_add(&bisc->last, &delta); | |
c3df5fb5 | 836 | u64_stats_update_end_irqrestore(&blkg->iostat.sync, flags); |
cd1fc4b9 | 837 | |
dc26532a JW |
838 | /* propagate global delta to parent (unless that's root) */ |
839 | if (parent && parent->parent) { | |
c3df5fb5 | 840 | flags = u64_stats_update_begin_irqsave(&parent->iostat.sync); |
cd1fc4b9 BB |
841 | blkg_iostat_set(&delta, &blkg->iostat.cur); |
842 | blkg_iostat_sub(&delta, &blkg->iostat.last); | |
843 | blkg_iostat_add(&parent->iostat.cur, &delta); | |
844 | blkg_iostat_add(&blkg->iostat.last, &delta); | |
c3df5fb5 | 845 | u64_stats_update_end_irqrestore(&parent->iostat.sync, flags); |
cd1fc4b9 BB |
846 | } |
847 | } | |
848 | ||
849 | rcu_read_unlock(); | |
850 | } | |
851 | ||
ef45fe47 | 852 | /* |
dc26532a JW |
853 | * We source root cgroup stats from the system-wide stats to avoid |
854 | * tracking the same information twice and incurring overhead when no | |
855 | * cgroups are defined. For that reason, cgroup_rstat_flush in | |
856 | * blkcg_print_stat does not actually fill out the iostat in the root | |
857 | * cgroup's blkcg_gq. | |
ef45fe47 BB |
858 | * |
859 | * However, we would like to re-use the printing code between the root and | |
860 | * non-root cgroups to the extent possible. For that reason, we simulate | |
861 | * flushing the root cgroup's stats by explicitly filling in the iostat | |
862 | * with disk level statistics. | |
863 | */ | |
864 | static void blkcg_fill_root_iostats(void) | |
865 | { | |
866 | struct class_dev_iter iter; | |
867 | struct device *dev; | |
868 | ||
869 | class_dev_iter_init(&iter, &block_class, NULL, &disk_type); | |
870 | while ((dev = class_dev_iter_next(&iter))) { | |
0d02129e CH |
871 | struct block_device *bdev = dev_to_bdev(dev); |
872 | struct blkcg_gq *blkg = | |
ed6cddef | 873 | blk_queue_root_blkg(bdev_get_queue(bdev)); |
ef45fe47 BB |
874 | struct blkg_iostat tmp; |
875 | int cpu; | |
f122d103 | 876 | unsigned long flags; |
ef45fe47 BB |
877 | |
878 | memset(&tmp, 0, sizeof(tmp)); | |
879 | for_each_possible_cpu(cpu) { | |
880 | struct disk_stats *cpu_dkstats; | |
881 | ||
0d02129e | 882 | cpu_dkstats = per_cpu_ptr(bdev->bd_stats, cpu); |
ef45fe47 BB |
883 | tmp.ios[BLKG_IOSTAT_READ] += |
884 | cpu_dkstats->ios[STAT_READ]; | |
885 | tmp.ios[BLKG_IOSTAT_WRITE] += | |
886 | cpu_dkstats->ios[STAT_WRITE]; | |
887 | tmp.ios[BLKG_IOSTAT_DISCARD] += | |
888 | cpu_dkstats->ios[STAT_DISCARD]; | |
889 | // convert sectors to bytes | |
890 | tmp.bytes[BLKG_IOSTAT_READ] += | |
891 | cpu_dkstats->sectors[STAT_READ] << 9; | |
892 | tmp.bytes[BLKG_IOSTAT_WRITE] += | |
893 | cpu_dkstats->sectors[STAT_WRITE] << 9; | |
894 | tmp.bytes[BLKG_IOSTAT_DISCARD] += | |
895 | cpu_dkstats->sectors[STAT_DISCARD] << 9; | |
ef45fe47 | 896 | } |
f122d103 CZ |
897 | |
898 | flags = u64_stats_update_begin_irqsave(&blkg->iostat.sync); | |
899 | blkg_iostat_set(&blkg->iostat.cur, &tmp); | |
900 | u64_stats_update_end_irqrestore(&blkg->iostat.sync, flags); | |
ef45fe47 BB |
901 | } |
902 | } | |
903 | ||
49cb5168 | 904 | static void blkcg_print_one_stat(struct blkcg_gq *blkg, struct seq_file *s) |
2ee867dc | 905 | { |
49cb5168 CH |
906 | struct blkg_iostat_set *bis = &blkg->iostat; |
907 | u64 rbytes, wbytes, rios, wios, dbytes, dios; | |
908 | bool has_stats = false; | |
909 | const char *dname; | |
910 | unsigned seq; | |
49cb5168 | 911 | int i; |
ef45fe47 | 912 | |
49cb5168 CH |
913 | if (!blkg->online) |
914 | return; | |
2ee867dc | 915 | |
49cb5168 CH |
916 | dname = blkg_dev_name(blkg); |
917 | if (!dname) | |
918 | return; | |
2ee867dc | 919 | |
252c651a | 920 | seq_printf(s, "%s ", dname); |
b0814361 | 921 | |
49cb5168 CH |
922 | do { |
923 | seq = u64_stats_fetch_begin(&bis->sync); | |
b0814361 | 924 | |
49cb5168 CH |
925 | rbytes = bis->cur.bytes[BLKG_IOSTAT_READ]; |
926 | wbytes = bis->cur.bytes[BLKG_IOSTAT_WRITE]; | |
927 | dbytes = bis->cur.bytes[BLKG_IOSTAT_DISCARD]; | |
928 | rios = bis->cur.ios[BLKG_IOSTAT_READ]; | |
929 | wios = bis->cur.ios[BLKG_IOSTAT_WRITE]; | |
930 | dios = bis->cur.ios[BLKG_IOSTAT_DISCARD]; | |
931 | } while (u64_stats_fetch_retry(&bis->sync, seq)); | |
2ee867dc | 932 | |
49cb5168 CH |
933 | if (rbytes || wbytes || rios || wios) { |
934 | has_stats = true; | |
252c651a | 935 | seq_printf(s, "rbytes=%llu wbytes=%llu rios=%llu wios=%llu dbytes=%llu dios=%llu", |
49cb5168 CH |
936 | rbytes, wbytes, rios, wios, |
937 | dbytes, dios); | |
938 | } | |
903d23f0 | 939 | |
49cb5168 CH |
940 | if (blkcg_debug_stats && atomic_read(&blkg->use_delay)) { |
941 | has_stats = true; | |
252c651a | 942 | seq_printf(s, " use_delay=%d delay_nsec=%llu", |
49cb5168 CH |
943 | atomic_read(&blkg->use_delay), |
944 | atomic64_read(&blkg->delay_nsec)); | |
945 | } | |
2ee867dc | 946 | |
49cb5168 CH |
947 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
948 | struct blkcg_policy *pol = blkcg_policy[i]; | |
2ee867dc | 949 | |
49cb5168 CH |
950 | if (!blkg->pd[i] || !pol->pd_stat_fn) |
951 | continue; | |
903d23f0 | 952 | |
252c651a | 953 | if (pol->pd_stat_fn(blkg->pd[i], s)) |
d09d8df3 | 954 | has_stats = true; |
49cb5168 | 955 | } |
d09d8df3 | 956 | |
252c651a CH |
957 | if (has_stats) |
958 | seq_printf(s, "\n"); | |
49cb5168 | 959 | } |
903d23f0 | 960 | |
49cb5168 CH |
961 | static int blkcg_print_stat(struct seq_file *sf, void *v) |
962 | { | |
963 | struct blkcg *blkcg = css_to_blkcg(seq_css(sf)); | |
964 | struct blkcg_gq *blkg; | |
903d23f0 | 965 | |
49cb5168 CH |
966 | if (!seq_css(sf)->parent) |
967 | blkcg_fill_root_iostats(); | |
968 | else | |
969 | cgroup_rstat_flush(blkcg->css.cgroup); | |
07b0fdec | 970 | |
49cb5168 CH |
971 | rcu_read_lock(); |
972 | hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) { | |
973 | spin_lock_irq(&blkg->q->queue_lock); | |
974 | blkcg_print_one_stat(blkg, sf); | |
b0814361 | 975 | spin_unlock_irq(&blkg->q->queue_lock); |
2ee867dc | 976 | } |
2ee867dc TH |
977 | rcu_read_unlock(); |
978 | return 0; | |
979 | } | |
980 | ||
e1f3b941 | 981 | static struct cftype blkcg_files[] = { |
2ee867dc TH |
982 | { |
983 | .name = "stat", | |
984 | .seq_show = blkcg_print_stat, | |
985 | }, | |
986 | { } /* terminate */ | |
987 | }; | |
988 | ||
e1f3b941 | 989 | static struct cftype blkcg_legacy_files[] = { |
84c124da DS |
990 | { |
991 | .name = "reset_stats", | |
3c798398 | 992 | .write_u64 = blkcg_reset_stats, |
22084190 | 993 | }, |
4baf6e33 | 994 | { } /* terminate */ |
31e4c28d VG |
995 | }; |
996 | ||
59b57717 DZF |
997 | /* |
998 | * blkcg destruction is a three-stage process. | |
999 | * | |
1000 | * 1. Destruction starts. The blkcg_css_offline() callback is invoked | |
1001 | * which offlines writeback. Here we tie the next stage of blkg destruction | |
1002 | * to the completion of writeback associated with the blkcg. This lets us | |
1003 | * avoid punting potentially large amounts of outstanding writeback to root | |
1004 | * while maintaining any ongoing policies. The next stage is triggered when | |
1005 | * the nr_cgwbs count goes to zero. | |
1006 | * | |
1007 | * 2. When the nr_cgwbs count goes to zero, blkcg_destroy_blkgs() is called | |
1008 | * and handles the destruction of blkgs. Here the css reference held by | |
1009 | * the blkg is put back eventually allowing blkcg_css_free() to be called. | |
1010 | * This work may occur in cgwb_release_workfn() on the cgwb_release | |
1011 | * workqueue. Any submitted ios that fail to get the blkg ref will be | |
1012 | * punted to the root_blkg. | |
1013 | * | |
1014 | * 3. Once the blkcg ref count goes to zero, blkcg_css_free() is called. | |
1015 | * This finally frees the blkcg. | |
1016 | */ | |
1017 | ||
9f13ef67 | 1018 | /** |
92fb9748 | 1019 | * blkcg_css_offline - cgroup css_offline callback |
eb95419b | 1020 | * @css: css of interest |
9f13ef67 | 1021 | * |
59b57717 DZF |
1022 | * This function is called when @css is about to go away. Here the cgwbs are |
1023 | * offlined first and only once writeback associated with the blkcg has | |
1024 | * finished do we start step 2 (see above). | |
9f13ef67 | 1025 | */ |
eb95419b | 1026 | static void blkcg_css_offline(struct cgroup_subsys_state *css) |
31e4c28d | 1027 | { |
eb95419b | 1028 | struct blkcg *blkcg = css_to_blkcg(css); |
b1c35769 | 1029 | |
59b57717 DZF |
1030 | /* this prevents anyone from attaching or migrating to this blkcg */ |
1031 | wb_blkcg_offline(blkcg); | |
1032 | ||
d866dbf6 TH |
1033 | /* put the base online pin allowing step 2 to be triggered */ |
1034 | blkcg_unpin_online(blkcg); | |
59b57717 DZF |
1035 | } |
1036 | ||
1037 | /** | |
1038 | * blkcg_destroy_blkgs - responsible for shooting down blkgs | |
1039 | * @blkcg: blkcg of interest | |
1040 | * | |
1041 | * blkgs should be removed while holding both q and blkcg locks. As blkcg lock | |
1042 | * is nested inside q lock, this function performs reverse double lock dancing. | |
1043 | * Destroying the blkgs releases the reference held on the blkcg's css allowing | |
1044 | * blkcg_css_free to eventually be called. | |
1045 | * | |
1046 | * This is the blkcg counterpart of ioc_release_fn(). | |
1047 | */ | |
1048 | void blkcg_destroy_blkgs(struct blkcg *blkcg) | |
1049 | { | |
6c635cae BW |
1050 | might_sleep(); |
1051 | ||
9f13ef67 | 1052 | spin_lock_irq(&blkcg->lock); |
7ee9c562 | 1053 | |
4c699480 JQ |
1054 | while (!hlist_empty(&blkcg->blkg_list)) { |
1055 | struct blkcg_gq *blkg = hlist_entry(blkcg->blkg_list.first, | |
6b065462 | 1056 | struct blkcg_gq, blkcg_node); |
4c699480 JQ |
1057 | struct request_queue *q = blkg->q; |
1058 | ||
6c635cae BW |
1059 | if (need_resched() || !spin_trylock(&q->queue_lock)) { |
1060 | /* | |
1061 | * Given that the system can accumulate a huge number | |
1062 | * of blkgs in pathological cases, check to see if we | |
1063 | * need to rescheduling to avoid softlockup. | |
1064 | */ | |
4c699480 | 1065 | spin_unlock_irq(&blkcg->lock); |
6c635cae | 1066 | cond_resched(); |
4c699480 | 1067 | spin_lock_irq(&blkcg->lock); |
6c635cae | 1068 | continue; |
4c699480 | 1069 | } |
6c635cae BW |
1070 | |
1071 | blkg_destroy(blkg); | |
1072 | spin_unlock(&q->queue_lock); | |
4c699480 | 1073 | } |
6b065462 | 1074 | |
4c699480 JQ |
1075 | spin_unlock_irq(&blkcg->lock); |
1076 | } | |
1077 | ||
eb95419b | 1078 | static void blkcg_css_free(struct cgroup_subsys_state *css) |
7ee9c562 | 1079 | { |
eb95419b | 1080 | struct blkcg *blkcg = css_to_blkcg(css); |
bc915e61 | 1081 | int i; |
7ee9c562 | 1082 | |
7876f930 | 1083 | mutex_lock(&blkcg_pol_mutex); |
e4a9bde9 | 1084 | |
7876f930 | 1085 | list_del(&blkcg->all_blkcgs_node); |
7876f930 | 1086 | |
bc915e61 | 1087 | for (i = 0; i < BLKCG_MAX_POLS; i++) |
e4a9bde9 TH |
1088 | if (blkcg->cpd[i]) |
1089 | blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]); | |
1090 | ||
1091 | mutex_unlock(&blkcg_pol_mutex); | |
1092 | ||
bc915e61 | 1093 | kfree(blkcg); |
31e4c28d VG |
1094 | } |
1095 | ||
eb95419b TH |
1096 | static struct cgroup_subsys_state * |
1097 | blkcg_css_alloc(struct cgroup_subsys_state *parent_css) | |
31e4c28d | 1098 | { |
3c798398 | 1099 | struct blkcg *blkcg; |
e48453c3 AA |
1100 | struct cgroup_subsys_state *ret; |
1101 | int i; | |
31e4c28d | 1102 | |
7876f930 TH |
1103 | mutex_lock(&blkcg_pol_mutex); |
1104 | ||
eb95419b | 1105 | if (!parent_css) { |
3c798398 | 1106 | blkcg = &blkcg_root; |
bc915e61 TH |
1107 | } else { |
1108 | blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL); | |
1109 | if (!blkcg) { | |
1110 | ret = ERR_PTR(-ENOMEM); | |
4c18c9e9 | 1111 | goto unlock; |
bc915e61 | 1112 | } |
e48453c3 AA |
1113 | } |
1114 | ||
1115 | for (i = 0; i < BLKCG_MAX_POLS ; i++) { | |
1116 | struct blkcg_policy *pol = blkcg_policy[i]; | |
1117 | struct blkcg_policy_data *cpd; | |
1118 | ||
1119 | /* | |
1120 | * If the policy hasn't been attached yet, wait for it | |
1121 | * to be attached before doing anything else. Otherwise, | |
1122 | * check if the policy requires any specific per-cgroup | |
1123 | * data: if it does, allocate and initialize it. | |
1124 | */ | |
e4a9bde9 | 1125 | if (!pol || !pol->cpd_alloc_fn) |
e48453c3 AA |
1126 | continue; |
1127 | ||
e4a9bde9 | 1128 | cpd = pol->cpd_alloc_fn(GFP_KERNEL); |
e48453c3 AA |
1129 | if (!cpd) { |
1130 | ret = ERR_PTR(-ENOMEM); | |
1131 | goto free_pd_blkcg; | |
1132 | } | |
81437648 TH |
1133 | blkcg->cpd[i] = cpd; |
1134 | cpd->blkcg = blkcg; | |
e48453c3 | 1135 | cpd->plid = i; |
e4a9bde9 TH |
1136 | if (pol->cpd_init_fn) |
1137 | pol->cpd_init_fn(cpd); | |
e48453c3 | 1138 | } |
31e4c28d | 1139 | |
31e4c28d | 1140 | spin_lock_init(&blkcg->lock); |
d866dbf6 | 1141 | refcount_set(&blkcg->online_pin, 1); |
e00f4f4d | 1142 | INIT_RADIX_TREE(&blkcg->blkg_tree, GFP_NOWAIT | __GFP_NOWARN); |
31e4c28d | 1143 | INIT_HLIST_HEAD(&blkcg->blkg_list); |
52ebea74 TH |
1144 | #ifdef CONFIG_CGROUP_WRITEBACK |
1145 | INIT_LIST_HEAD(&blkcg->cgwb_list); | |
1146 | #endif | |
7876f930 TH |
1147 | list_add_tail(&blkcg->all_blkcgs_node, &all_blkcgs); |
1148 | ||
1149 | mutex_unlock(&blkcg_pol_mutex); | |
31e4c28d | 1150 | return &blkcg->css; |
e48453c3 AA |
1151 | |
1152 | free_pd_blkcg: | |
1153 | for (i--; i >= 0; i--) | |
e4a9bde9 TH |
1154 | if (blkcg->cpd[i]) |
1155 | blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]); | |
4c18c9e9 | 1156 | |
1157 | if (blkcg != &blkcg_root) | |
1158 | kfree(blkcg); | |
1159 | unlock: | |
7876f930 | 1160 | mutex_unlock(&blkcg_pol_mutex); |
e48453c3 | 1161 | return ret; |
31e4c28d VG |
1162 | } |
1163 | ||
4308a434 TH |
1164 | static int blkcg_css_online(struct cgroup_subsys_state *css) |
1165 | { | |
1166 | struct blkcg *blkcg = css_to_blkcg(css); | |
1167 | struct blkcg *parent = blkcg_parent(blkcg); | |
1168 | ||
1169 | /* | |
1170 | * blkcg_pin_online() is used to delay blkcg offline so that blkgs | |
1171 | * don't go offline while cgwbs are still active on them. Pin the | |
1172 | * parent so that offline always happens towards the root. | |
1173 | */ | |
1174 | if (parent) | |
1175 | blkcg_pin_online(parent); | |
1176 | return 0; | |
1177 | } | |
1178 | ||
5efd6113 TH |
1179 | /** |
1180 | * blkcg_init_queue - initialize blkcg part of request queue | |
1181 | * @q: request_queue to initialize | |
1182 | * | |
c62b37d9 | 1183 | * Called from blk_alloc_queue(). Responsible for initializing blkcg |
5efd6113 TH |
1184 | * part of new request_queue @q. |
1185 | * | |
1186 | * RETURNS: | |
1187 | * 0 on success, -errno on failure. | |
1188 | */ | |
1189 | int blkcg_init_queue(struct request_queue *q) | |
1190 | { | |
d708f0d5 JA |
1191 | struct blkcg_gq *new_blkg, *blkg; |
1192 | bool preloaded; | |
ec13b1d6 TH |
1193 | int ret; |
1194 | ||
472e4314 ML |
1195 | INIT_LIST_HEAD(&q->blkg_list); |
1196 | ||
d708f0d5 JA |
1197 | new_blkg = blkg_alloc(&blkcg_root, q, GFP_KERNEL); |
1198 | if (!new_blkg) | |
1199 | return -ENOMEM; | |
1200 | ||
1201 | preloaded = !radix_tree_preload(GFP_KERNEL); | |
1202 | ||
bea54883 | 1203 | /* Make sure the root blkg exists. */ |
ec13b1d6 | 1204 | rcu_read_lock(); |
0d945c1f | 1205 | spin_lock_irq(&q->queue_lock); |
d708f0d5 | 1206 | blkg = blkg_create(&blkcg_root, q, new_blkg); |
901932a3 JB |
1207 | if (IS_ERR(blkg)) |
1208 | goto err_unlock; | |
1209 | q->root_blkg = blkg; | |
0d945c1f | 1210 | spin_unlock_irq(&q->queue_lock); |
ec13b1d6 TH |
1211 | rcu_read_unlock(); |
1212 | ||
d708f0d5 JA |
1213 | if (preloaded) |
1214 | radix_tree_preload_end(); | |
1215 | ||
556910e3 BVA |
1216 | ret = blk_ioprio_init(q); |
1217 | if (ret) | |
1218 | goto err_destroy_all; | |
1219 | ||
19688d7f BVA |
1220 | ret = blk_throtl_init(q); |
1221 | if (ret) | |
04be60b5 | 1222 | goto err_destroy_all; |
19688d7f | 1223 | |
6f5ddde4 YX |
1224 | ret = blk_iolatency_init(q); |
1225 | if (ret) { | |
1226 | blk_throtl_exit(q); | |
1227 | goto err_destroy_all; | |
1228 | } | |
1229 | ||
04be60b5 | 1230 | return 0; |
901932a3 | 1231 | |
04be60b5 | 1232 | err_destroy_all: |
04be60b5 | 1233 | blkg_destroy_all(q); |
04be60b5 | 1234 | return ret; |
901932a3 | 1235 | err_unlock: |
0d945c1f | 1236 | spin_unlock_irq(&q->queue_lock); |
901932a3 JB |
1237 | rcu_read_unlock(); |
1238 | if (preloaded) | |
1239 | radix_tree_preload_end(); | |
1240 | return PTR_ERR(blkg); | |
5efd6113 TH |
1241 | } |
1242 | ||
5efd6113 TH |
1243 | /** |
1244 | * blkcg_exit_queue - exit and release blkcg part of request_queue | |
1245 | * @q: request_queue being released | |
1246 | * | |
7585d508 | 1247 | * Called from blk_exit_queue(). Responsible for exiting blkcg part. |
5efd6113 TH |
1248 | */ |
1249 | void blkcg_exit_queue(struct request_queue *q) | |
1250 | { | |
3c96cb32 | 1251 | blkg_destroy_all(q); |
5efd6113 TH |
1252 | blk_throtl_exit(q); |
1253 | } | |
1254 | ||
69d7fde5 TH |
1255 | static void blkcg_bind(struct cgroup_subsys_state *root_css) |
1256 | { | |
1257 | int i; | |
1258 | ||
1259 | mutex_lock(&blkcg_pol_mutex); | |
1260 | ||
1261 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
1262 | struct blkcg_policy *pol = blkcg_policy[i]; | |
1263 | struct blkcg *blkcg; | |
1264 | ||
1265 | if (!pol || !pol->cpd_bind_fn) | |
1266 | continue; | |
1267 | ||
1268 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) | |
1269 | if (blkcg->cpd[pol->plid]) | |
1270 | pol->cpd_bind_fn(blkcg->cpd[pol->plid]); | |
1271 | } | |
1272 | mutex_unlock(&blkcg_pol_mutex); | |
1273 | } | |
1274 | ||
d09d8df3 JB |
1275 | static void blkcg_exit(struct task_struct *tsk) |
1276 | { | |
1277 | if (tsk->throttle_queue) | |
1278 | blk_put_queue(tsk->throttle_queue); | |
1279 | tsk->throttle_queue = NULL; | |
1280 | } | |
1281 | ||
c165b3e3 | 1282 | struct cgroup_subsys io_cgrp_subsys = { |
92fb9748 | 1283 | .css_alloc = blkcg_css_alloc, |
4308a434 | 1284 | .css_online = blkcg_css_online, |
92fb9748 TH |
1285 | .css_offline = blkcg_css_offline, |
1286 | .css_free = blkcg_css_free, | |
f7331648 | 1287 | .css_rstat_flush = blkcg_rstat_flush, |
69d7fde5 | 1288 | .bind = blkcg_bind, |
2ee867dc | 1289 | .dfl_cftypes = blkcg_files, |
880f50e2 | 1290 | .legacy_cftypes = blkcg_legacy_files, |
c165b3e3 | 1291 | .legacy_name = "blkio", |
d09d8df3 | 1292 | .exit = blkcg_exit, |
1ced953b TH |
1293 | #ifdef CONFIG_MEMCG |
1294 | /* | |
1295 | * This ensures that, if available, memcg is automatically enabled | |
1296 | * together on the default hierarchy so that the owner cgroup can | |
1297 | * be retrieved from writeback pages. | |
1298 | */ | |
1299 | .depends_on = 1 << memory_cgrp_id, | |
1300 | #endif | |
676f7c8f | 1301 | }; |
c165b3e3 | 1302 | EXPORT_SYMBOL_GPL(io_cgrp_subsys); |
676f7c8f | 1303 | |
a2b1693b TH |
1304 | /** |
1305 | * blkcg_activate_policy - activate a blkcg policy on a request_queue | |
1306 | * @q: request_queue of interest | |
1307 | * @pol: blkcg policy to activate | |
1308 | * | |
1309 | * Activate @pol on @q. Requires %GFP_KERNEL context. @q goes through | |
1310 | * bypass mode to populate its blkgs with policy_data for @pol. | |
1311 | * | |
1312 | * Activation happens with @q bypassed, so nobody would be accessing blkgs | |
1313 | * from IO path. Update of each blkg is protected by both queue and blkcg | |
1314 | * locks so that holding either lock and testing blkcg_policy_enabled() is | |
1315 | * always enough for dereferencing policy data. | |
1316 | * | |
1317 | * The caller is responsible for synchronizing [de]activations and policy | |
1318 | * [un]registerations. Returns 0 on success, -errno on failure. | |
1319 | */ | |
1320 | int blkcg_activate_policy(struct request_queue *q, | |
3c798398 | 1321 | const struct blkcg_policy *pol) |
a2b1693b | 1322 | { |
4c55f4f9 | 1323 | struct blkg_policy_data *pd_prealloc = NULL; |
9d179b86 | 1324 | struct blkcg_gq *blkg, *pinned_blkg = NULL; |
4c55f4f9 | 1325 | int ret; |
a2b1693b TH |
1326 | |
1327 | if (blkcg_policy_enabled(q, pol)) | |
1328 | return 0; | |
1329 | ||
344e9ffc | 1330 | if (queue_is_mq(q)) |
bd166ef1 | 1331 | blk_mq_freeze_queue(q); |
9d179b86 | 1332 | retry: |
0d945c1f | 1333 | spin_lock_irq(&q->queue_lock); |
a2b1693b | 1334 | |
9d179b86 | 1335 | /* blkg_list is pushed at the head, reverse walk to allocate parents first */ |
71c81407 | 1336 | list_for_each_entry_reverse(blkg, &q->blkg_list, q_node) { |
4c55f4f9 TH |
1337 | struct blkg_policy_data *pd; |
1338 | ||
1339 | if (blkg->pd[pol->plid]) | |
1340 | continue; | |
a2b1693b | 1341 | |
9d179b86 TH |
1342 | /* If prealloc matches, use it; otherwise try GFP_NOWAIT */ |
1343 | if (blkg == pinned_blkg) { | |
1344 | pd = pd_prealloc; | |
1345 | pd_prealloc = NULL; | |
1346 | } else { | |
1347 | pd = pol->pd_alloc_fn(GFP_NOWAIT | __GFP_NOWARN, q, | |
1348 | blkg->blkcg); | |
1349 | } | |
1350 | ||
4c55f4f9 | 1351 | if (!pd) { |
9d179b86 TH |
1352 | /* |
1353 | * GFP_NOWAIT failed. Free the existing one and | |
1354 | * prealloc for @blkg w/ GFP_KERNEL. | |
1355 | */ | |
1356 | if (pinned_blkg) | |
1357 | blkg_put(pinned_blkg); | |
1358 | blkg_get(blkg); | |
1359 | pinned_blkg = blkg; | |
1360 | ||
0d945c1f | 1361 | spin_unlock_irq(&q->queue_lock); |
9d179b86 TH |
1362 | |
1363 | if (pd_prealloc) | |
1364 | pol->pd_free_fn(pd_prealloc); | |
1365 | pd_prealloc = pol->pd_alloc_fn(GFP_KERNEL, q, | |
1366 | blkg->blkcg); | |
1367 | if (pd_prealloc) | |
1368 | goto retry; | |
1369 | else | |
1370 | goto enomem; | |
4c55f4f9 | 1371 | } |
a2b1693b TH |
1372 | |
1373 | blkg->pd[pol->plid] = pd; | |
1374 | pd->blkg = blkg; | |
b276a876 | 1375 | pd->plid = pol->plid; |
a2b1693b TH |
1376 | } |
1377 | ||
9d179b86 TH |
1378 | /* all allocated, init in the same order */ |
1379 | if (pol->pd_init_fn) | |
1380 | list_for_each_entry_reverse(blkg, &q->blkg_list, q_node) | |
1381 | pol->pd_init_fn(blkg->pd[pol->plid]); | |
1382 | ||
a2b1693b TH |
1383 | __set_bit(pol->plid, q->blkcg_pols); |
1384 | ret = 0; | |
4c55f4f9 | 1385 | |
0d945c1f | 1386 | spin_unlock_irq(&q->queue_lock); |
9d179b86 | 1387 | out: |
344e9ffc | 1388 | if (queue_is_mq(q)) |
bd166ef1 | 1389 | blk_mq_unfreeze_queue(q); |
9d179b86 TH |
1390 | if (pinned_blkg) |
1391 | blkg_put(pinned_blkg); | |
001bea73 TH |
1392 | if (pd_prealloc) |
1393 | pol->pd_free_fn(pd_prealloc); | |
a2b1693b | 1394 | return ret; |
9d179b86 TH |
1395 | |
1396 | enomem: | |
1397 | /* alloc failed, nothing's initialized yet, free everything */ | |
1398 | spin_lock_irq(&q->queue_lock); | |
1399 | list_for_each_entry(blkg, &q->blkg_list, q_node) { | |
858560b2 LJ |
1400 | struct blkcg *blkcg = blkg->blkcg; |
1401 | ||
1402 | spin_lock(&blkcg->lock); | |
9d179b86 TH |
1403 | if (blkg->pd[pol->plid]) { |
1404 | pol->pd_free_fn(blkg->pd[pol->plid]); | |
1405 | blkg->pd[pol->plid] = NULL; | |
1406 | } | |
858560b2 | 1407 | spin_unlock(&blkcg->lock); |
9d179b86 TH |
1408 | } |
1409 | spin_unlock_irq(&q->queue_lock); | |
1410 | ret = -ENOMEM; | |
1411 | goto out; | |
a2b1693b TH |
1412 | } |
1413 | EXPORT_SYMBOL_GPL(blkcg_activate_policy); | |
1414 | ||
1415 | /** | |
1416 | * blkcg_deactivate_policy - deactivate a blkcg policy on a request_queue | |
1417 | * @q: request_queue of interest | |
1418 | * @pol: blkcg policy to deactivate | |
1419 | * | |
1420 | * Deactivate @pol on @q. Follows the same synchronization rules as | |
1421 | * blkcg_activate_policy(). | |
1422 | */ | |
1423 | void blkcg_deactivate_policy(struct request_queue *q, | |
3c798398 | 1424 | const struct blkcg_policy *pol) |
a2b1693b | 1425 | { |
3c798398 | 1426 | struct blkcg_gq *blkg; |
a2b1693b TH |
1427 | |
1428 | if (!blkcg_policy_enabled(q, pol)) | |
1429 | return; | |
1430 | ||
344e9ffc | 1431 | if (queue_is_mq(q)) |
bd166ef1 | 1432 | blk_mq_freeze_queue(q); |
bd166ef1 | 1433 | |
0d945c1f | 1434 | spin_lock_irq(&q->queue_lock); |
a2b1693b TH |
1435 | |
1436 | __clear_bit(pol->plid, q->blkcg_pols); | |
1437 | ||
1438 | list_for_each_entry(blkg, &q->blkg_list, q_node) { | |
858560b2 LJ |
1439 | struct blkcg *blkcg = blkg->blkcg; |
1440 | ||
1441 | spin_lock(&blkcg->lock); | |
001bea73 | 1442 | if (blkg->pd[pol->plid]) { |
6b065462 | 1443 | if (pol->pd_offline_fn) |
a9520cd6 | 1444 | pol->pd_offline_fn(blkg->pd[pol->plid]); |
001bea73 TH |
1445 | pol->pd_free_fn(blkg->pd[pol->plid]); |
1446 | blkg->pd[pol->plid] = NULL; | |
1447 | } | |
858560b2 | 1448 | spin_unlock(&blkcg->lock); |
a2b1693b TH |
1449 | } |
1450 | ||
0d945c1f | 1451 | spin_unlock_irq(&q->queue_lock); |
bd166ef1 | 1452 | |
344e9ffc | 1453 | if (queue_is_mq(q)) |
bd166ef1 | 1454 | blk_mq_unfreeze_queue(q); |
a2b1693b TH |
1455 | } |
1456 | EXPORT_SYMBOL_GPL(blkcg_deactivate_policy); | |
1457 | ||
8bd435b3 | 1458 | /** |
3c798398 TH |
1459 | * blkcg_policy_register - register a blkcg policy |
1460 | * @pol: blkcg policy to register | |
8bd435b3 | 1461 | * |
3c798398 TH |
1462 | * Register @pol with blkcg core. Might sleep and @pol may be modified on |
1463 | * successful registration. Returns 0 on success and -errno on failure. | |
8bd435b3 | 1464 | */ |
d5bf0291 | 1465 | int blkcg_policy_register(struct blkcg_policy *pol) |
3e252066 | 1466 | { |
06b285bd | 1467 | struct blkcg *blkcg; |
8bd435b3 | 1468 | int i, ret; |
e8989fae | 1469 | |
838f13bf | 1470 | mutex_lock(&blkcg_pol_register_mutex); |
bc0d6501 TH |
1471 | mutex_lock(&blkcg_pol_mutex); |
1472 | ||
8bd435b3 TH |
1473 | /* find an empty slot */ |
1474 | ret = -ENOSPC; | |
1475 | for (i = 0; i < BLKCG_MAX_POLS; i++) | |
3c798398 | 1476 | if (!blkcg_policy[i]) |
8bd435b3 | 1477 | break; |
01c5f85a JA |
1478 | if (i >= BLKCG_MAX_POLS) { |
1479 | pr_warn("blkcg_policy_register: BLKCG_MAX_POLS too small\n"); | |
838f13bf | 1480 | goto err_unlock; |
01c5f85a | 1481 | } |
035d10b2 | 1482 | |
e8401073 | 1483 | /* Make sure cpd/pd_alloc_fn and cpd/pd_free_fn in pairs */ |
1484 | if ((!pol->cpd_alloc_fn ^ !pol->cpd_free_fn) || | |
1485 | (!pol->pd_alloc_fn ^ !pol->pd_free_fn)) | |
1486 | goto err_unlock; | |
1487 | ||
06b285bd | 1488 | /* register @pol */ |
3c798398 | 1489 | pol->plid = i; |
06b285bd TH |
1490 | blkcg_policy[pol->plid] = pol; |
1491 | ||
1492 | /* allocate and install cpd's */ | |
e4a9bde9 | 1493 | if (pol->cpd_alloc_fn) { |
06b285bd TH |
1494 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) { |
1495 | struct blkcg_policy_data *cpd; | |
1496 | ||
e4a9bde9 | 1497 | cpd = pol->cpd_alloc_fn(GFP_KERNEL); |
bbb427e3 | 1498 | if (!cpd) |
06b285bd | 1499 | goto err_free_cpds; |
06b285bd | 1500 | |
81437648 TH |
1501 | blkcg->cpd[pol->plid] = cpd; |
1502 | cpd->blkcg = blkcg; | |
06b285bd | 1503 | cpd->plid = pol->plid; |
86a5bba5 TH |
1504 | if (pol->cpd_init_fn) |
1505 | pol->cpd_init_fn(cpd); | |
06b285bd TH |
1506 | } |
1507 | } | |
1508 | ||
838f13bf | 1509 | mutex_unlock(&blkcg_pol_mutex); |
8bd435b3 | 1510 | |
8bd435b3 | 1511 | /* everything is in place, add intf files for the new policy */ |
2ee867dc TH |
1512 | if (pol->dfl_cftypes) |
1513 | WARN_ON(cgroup_add_dfl_cftypes(&io_cgrp_subsys, | |
1514 | pol->dfl_cftypes)); | |
880f50e2 | 1515 | if (pol->legacy_cftypes) |
c165b3e3 | 1516 | WARN_ON(cgroup_add_legacy_cftypes(&io_cgrp_subsys, |
880f50e2 | 1517 | pol->legacy_cftypes)); |
838f13bf TH |
1518 | mutex_unlock(&blkcg_pol_register_mutex); |
1519 | return 0; | |
1520 | ||
06b285bd | 1521 | err_free_cpds: |
58a9edce | 1522 | if (pol->cpd_free_fn) { |
06b285bd | 1523 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) { |
e4a9bde9 TH |
1524 | if (blkcg->cpd[pol->plid]) { |
1525 | pol->cpd_free_fn(blkcg->cpd[pol->plid]); | |
1526 | blkcg->cpd[pol->plid] = NULL; | |
1527 | } | |
06b285bd TH |
1528 | } |
1529 | } | |
1530 | blkcg_policy[pol->plid] = NULL; | |
838f13bf | 1531 | err_unlock: |
bc0d6501 | 1532 | mutex_unlock(&blkcg_pol_mutex); |
838f13bf | 1533 | mutex_unlock(&blkcg_pol_register_mutex); |
8bd435b3 | 1534 | return ret; |
3e252066 | 1535 | } |
3c798398 | 1536 | EXPORT_SYMBOL_GPL(blkcg_policy_register); |
3e252066 | 1537 | |
8bd435b3 | 1538 | /** |
3c798398 TH |
1539 | * blkcg_policy_unregister - unregister a blkcg policy |
1540 | * @pol: blkcg policy to unregister | |
8bd435b3 | 1541 | * |
3c798398 | 1542 | * Undo blkcg_policy_register(@pol). Might sleep. |
8bd435b3 | 1543 | */ |
3c798398 | 1544 | void blkcg_policy_unregister(struct blkcg_policy *pol) |
3e252066 | 1545 | { |
06b285bd TH |
1546 | struct blkcg *blkcg; |
1547 | ||
838f13bf | 1548 | mutex_lock(&blkcg_pol_register_mutex); |
bc0d6501 | 1549 | |
3c798398 | 1550 | if (WARN_ON(blkcg_policy[pol->plid] != pol)) |
8bd435b3 TH |
1551 | goto out_unlock; |
1552 | ||
1553 | /* kill the intf files first */ | |
2ee867dc TH |
1554 | if (pol->dfl_cftypes) |
1555 | cgroup_rm_cftypes(pol->dfl_cftypes); | |
880f50e2 TH |
1556 | if (pol->legacy_cftypes) |
1557 | cgroup_rm_cftypes(pol->legacy_cftypes); | |
44ea53de | 1558 | |
06b285bd | 1559 | /* remove cpds and unregister */ |
838f13bf | 1560 | mutex_lock(&blkcg_pol_mutex); |
06b285bd | 1561 | |
58a9edce | 1562 | if (pol->cpd_free_fn) { |
06b285bd | 1563 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) { |
e4a9bde9 TH |
1564 | if (blkcg->cpd[pol->plid]) { |
1565 | pol->cpd_free_fn(blkcg->cpd[pol->plid]); | |
1566 | blkcg->cpd[pol->plid] = NULL; | |
1567 | } | |
06b285bd TH |
1568 | } |
1569 | } | |
3c798398 | 1570 | blkcg_policy[pol->plid] = NULL; |
06b285bd | 1571 | |
bc0d6501 | 1572 | mutex_unlock(&blkcg_pol_mutex); |
838f13bf TH |
1573 | out_unlock: |
1574 | mutex_unlock(&blkcg_pol_register_mutex); | |
3e252066 | 1575 | } |
3c798398 | 1576 | EXPORT_SYMBOL_GPL(blkcg_policy_unregister); |
903d23f0 | 1577 | |
d3f77dfd TH |
1578 | bool __blkcg_punt_bio_submit(struct bio *bio) |
1579 | { | |
1580 | struct blkcg_gq *blkg = bio->bi_blkg; | |
1581 | ||
1582 | /* consume the flag first */ | |
1583 | bio->bi_opf &= ~REQ_CGROUP_PUNT; | |
1584 | ||
1585 | /* never bounce for the root cgroup */ | |
1586 | if (!blkg->parent) | |
1587 | return false; | |
1588 | ||
1589 | spin_lock_bh(&blkg->async_bio_lock); | |
1590 | bio_list_add(&blkg->async_bios, bio); | |
1591 | spin_unlock_bh(&blkg->async_bio_lock); | |
1592 | ||
1593 | queue_work(blkcg_punt_bio_wq, &blkg->async_bio_work); | |
1594 | return true; | |
1595 | } | |
1596 | ||
d09d8df3 JB |
1597 | /* |
1598 | * Scale the accumulated delay based on how long it has been since we updated | |
1599 | * the delay. We only call this when we are adding delay, in case it's been a | |
1600 | * while since we added delay, and when we are checking to see if we need to | |
1601 | * delay a task, to account for any delays that may have occurred. | |
1602 | */ | |
1603 | static void blkcg_scale_delay(struct blkcg_gq *blkg, u64 now) | |
1604 | { | |
1605 | u64 old = atomic64_read(&blkg->delay_start); | |
1606 | ||
54c52e10 TH |
1607 | /* negative use_delay means no scaling, see blkcg_set_delay() */ |
1608 | if (atomic_read(&blkg->use_delay) < 0) | |
1609 | return; | |
1610 | ||
d09d8df3 JB |
1611 | /* |
1612 | * We only want to scale down every second. The idea here is that we | |
1613 | * want to delay people for min(delay_nsec, NSEC_PER_SEC) in a certain | |
1614 | * time window. We only want to throttle tasks for recent delay that | |
1615 | * has occurred, in 1 second time windows since that's the maximum | |
1616 | * things can be throttled. We save the current delay window in | |
1617 | * blkg->last_delay so we know what amount is still left to be charged | |
1618 | * to the blkg from this point onward. blkg->last_use keeps track of | |
1619 | * the use_delay counter. The idea is if we're unthrottling the blkg we | |
1620 | * are ok with whatever is happening now, and we can take away more of | |
1621 | * the accumulated delay as we've already throttled enough that | |
1622 | * everybody is happy with their IO latencies. | |
1623 | */ | |
1624 | if (time_before64(old + NSEC_PER_SEC, now) && | |
1625 | atomic64_cmpxchg(&blkg->delay_start, old, now) == old) { | |
1626 | u64 cur = atomic64_read(&blkg->delay_nsec); | |
1627 | u64 sub = min_t(u64, blkg->last_delay, now - old); | |
1628 | int cur_use = atomic_read(&blkg->use_delay); | |
1629 | ||
1630 | /* | |
1631 | * We've been unthrottled, subtract a larger chunk of our | |
1632 | * accumulated delay. | |
1633 | */ | |
1634 | if (cur_use < blkg->last_use) | |
1635 | sub = max_t(u64, sub, blkg->last_delay >> 1); | |
1636 | ||
1637 | /* | |
1638 | * This shouldn't happen, but handle it anyway. Our delay_nsec | |
1639 | * should only ever be growing except here where we subtract out | |
1640 | * min(last_delay, 1 second), but lord knows bugs happen and I'd | |
1641 | * rather not end up with negative numbers. | |
1642 | */ | |
1643 | if (unlikely(cur < sub)) { | |
1644 | atomic64_set(&blkg->delay_nsec, 0); | |
1645 | blkg->last_delay = 0; | |
1646 | } else { | |
1647 | atomic64_sub(sub, &blkg->delay_nsec); | |
1648 | blkg->last_delay = cur - sub; | |
1649 | } | |
1650 | blkg->last_use = cur_use; | |
1651 | } | |
1652 | } | |
1653 | ||
1654 | /* | |
1655 | * This is called when we want to actually walk up the hierarchy and check to | |
1656 | * see if we need to throttle, and then actually throttle if there is some | |
1657 | * accumulated delay. This should only be called upon return to user space so | |
1658 | * we're not holding some lock that would induce a priority inversion. | |
1659 | */ | |
1660 | static void blkcg_maybe_throttle_blkg(struct blkcg_gq *blkg, bool use_memdelay) | |
1661 | { | |
fd112c74 | 1662 | unsigned long pflags; |
5160a5a5 | 1663 | bool clamp; |
d09d8df3 JB |
1664 | u64 now = ktime_to_ns(ktime_get()); |
1665 | u64 exp; | |
1666 | u64 delay_nsec = 0; | |
1667 | int tok; | |
1668 | ||
1669 | while (blkg->parent) { | |
5160a5a5 TH |
1670 | int use_delay = atomic_read(&blkg->use_delay); |
1671 | ||
1672 | if (use_delay) { | |
1673 | u64 this_delay; | |
1674 | ||
d09d8df3 | 1675 | blkcg_scale_delay(blkg, now); |
5160a5a5 TH |
1676 | this_delay = atomic64_read(&blkg->delay_nsec); |
1677 | if (this_delay > delay_nsec) { | |
1678 | delay_nsec = this_delay; | |
1679 | clamp = use_delay > 0; | |
1680 | } | |
d09d8df3 JB |
1681 | } |
1682 | blkg = blkg->parent; | |
1683 | } | |
1684 | ||
1685 | if (!delay_nsec) | |
1686 | return; | |
1687 | ||
1688 | /* | |
1689 | * Let's not sleep for all eternity if we've amassed a huge delay. | |
1690 | * Swapping or metadata IO can accumulate 10's of seconds worth of | |
1691 | * delay, and we want userspace to be able to do _something_ so cap the | |
5160a5a5 TH |
1692 | * delays at 0.25s. If there's 10's of seconds worth of delay then the |
1693 | * tasks will be delayed for 0.25 second for every syscall. If | |
1694 | * blkcg_set_delay() was used as indicated by negative use_delay, the | |
1695 | * caller is responsible for regulating the range. | |
d09d8df3 | 1696 | */ |
5160a5a5 TH |
1697 | if (clamp) |
1698 | delay_nsec = min_t(u64, delay_nsec, 250 * NSEC_PER_MSEC); | |
d09d8df3 | 1699 | |
fd112c74 JB |
1700 | if (use_memdelay) |
1701 | psi_memstall_enter(&pflags); | |
d09d8df3 JB |
1702 | |
1703 | exp = ktime_add_ns(now, delay_nsec); | |
1704 | tok = io_schedule_prepare(); | |
1705 | do { | |
1706 | __set_current_state(TASK_KILLABLE); | |
1707 | if (!schedule_hrtimeout(&exp, HRTIMER_MODE_ABS)) | |
1708 | break; | |
1709 | } while (!fatal_signal_pending(current)); | |
1710 | io_schedule_finish(tok); | |
fd112c74 JB |
1711 | |
1712 | if (use_memdelay) | |
1713 | psi_memstall_leave(&pflags); | |
d09d8df3 JB |
1714 | } |
1715 | ||
1716 | /** | |
1717 | * blkcg_maybe_throttle_current - throttle the current task if it has been marked | |
1718 | * | |
1719 | * This is only called if we've been marked with set_notify_resume(). Obviously | |
1720 | * we can be set_notify_resume() for reasons other than blkcg throttling, so we | |
1721 | * check to see if current->throttle_queue is set and if not this doesn't do | |
1722 | * anything. This should only ever be called by the resume code, it's not meant | |
1723 | * to be called by people willy-nilly as it will actually do the work to | |
1724 | * throttle the task if it is setup for throttling. | |
1725 | */ | |
1726 | void blkcg_maybe_throttle_current(void) | |
1727 | { | |
1728 | struct request_queue *q = current->throttle_queue; | |
1729 | struct cgroup_subsys_state *css; | |
1730 | struct blkcg *blkcg; | |
1731 | struct blkcg_gq *blkg; | |
1732 | bool use_memdelay = current->use_memdelay; | |
1733 | ||
1734 | if (!q) | |
1735 | return; | |
1736 | ||
1737 | current->throttle_queue = NULL; | |
1738 | current->use_memdelay = false; | |
1739 | ||
1740 | rcu_read_lock(); | |
1741 | css = kthread_blkcg(); | |
1742 | if (css) | |
1743 | blkcg = css_to_blkcg(css); | |
1744 | else | |
1745 | blkcg = css_to_blkcg(task_css(current, io_cgrp_id)); | |
1746 | ||
1747 | if (!blkcg) | |
1748 | goto out; | |
1749 | blkg = blkg_lookup(blkcg, q); | |
1750 | if (!blkg) | |
1751 | goto out; | |
7754f669 | 1752 | if (!blkg_tryget(blkg)) |
d09d8df3 JB |
1753 | goto out; |
1754 | rcu_read_unlock(); | |
d09d8df3 JB |
1755 | |
1756 | blkcg_maybe_throttle_blkg(blkg, use_memdelay); | |
1757 | blkg_put(blkg); | |
cc7ecc25 | 1758 | blk_put_queue(q); |
d09d8df3 JB |
1759 | return; |
1760 | out: | |
1761 | rcu_read_unlock(); | |
1762 | blk_put_queue(q); | |
1763 | } | |
d09d8df3 JB |
1764 | |
1765 | /** | |
1766 | * blkcg_schedule_throttle - this task needs to check for throttling | |
537d71b3 BVA |
1767 | * @q: the request queue IO was submitted on |
1768 | * @use_memdelay: do we charge this to memory delay for PSI | |
d09d8df3 JB |
1769 | * |
1770 | * This is called by the IO controller when we know there's delay accumulated | |
1771 | * for the blkg for this task. We do not pass the blkg because there are places | |
1772 | * we call this that may not have that information, the swapping code for | |
1773 | * instance will only have a request_queue at that point. This set's the | |
1774 | * notify_resume for the task to check and see if it requires throttling before | |
1775 | * returning to user space. | |
1776 | * | |
1777 | * We will only schedule once per syscall. You can call this over and over | |
1778 | * again and it will only do the check once upon return to user space, and only | |
1779 | * throttle once. If the task needs to be throttled again it'll need to be | |
1780 | * re-set at the next time we see the task. | |
1781 | */ | |
1782 | void blkcg_schedule_throttle(struct request_queue *q, bool use_memdelay) | |
1783 | { | |
1784 | if (unlikely(current->flags & PF_KTHREAD)) | |
1785 | return; | |
1786 | ||
49d1822b CX |
1787 | if (current->throttle_queue != q) { |
1788 | if (!blk_get_queue(q)) | |
1789 | return; | |
1790 | ||
1791 | if (current->throttle_queue) | |
1792 | blk_put_queue(current->throttle_queue); | |
1793 | current->throttle_queue = q; | |
1794 | } | |
d09d8df3 | 1795 | |
d09d8df3 JB |
1796 | if (use_memdelay) |
1797 | current->use_memdelay = use_memdelay; | |
1798 | set_notify_resume(current); | |
1799 | } | |
d09d8df3 JB |
1800 | |
1801 | /** | |
1802 | * blkcg_add_delay - add delay to this blkg | |
537d71b3 BVA |
1803 | * @blkg: blkg of interest |
1804 | * @now: the current time in nanoseconds | |
1805 | * @delta: how many nanoseconds of delay to add | |
d09d8df3 JB |
1806 | * |
1807 | * Charge @delta to the blkg's current delay accumulation. This is used to | |
1808 | * throttle tasks if an IO controller thinks we need more throttling. | |
1809 | */ | |
1810 | void blkcg_add_delay(struct blkcg_gq *blkg, u64 now, u64 delta) | |
1811 | { | |
54c52e10 TH |
1812 | if (WARN_ON_ONCE(atomic_read(&blkg->use_delay) < 0)) |
1813 | return; | |
d09d8df3 JB |
1814 | blkcg_scale_delay(blkg, now); |
1815 | atomic64_add(delta, &blkg->delay_nsec); | |
1816 | } | |
d09d8df3 | 1817 | |
28fc591f CH |
1818 | /** |
1819 | * blkg_tryget_closest - try and get a blkg ref on the closet blkg | |
13c7863d CH |
1820 | * @bio: target bio |
1821 | * @css: target css | |
28fc591f | 1822 | * |
13c7863d CH |
1823 | * As the failure mode here is to walk up the blkg tree, this ensure that the |
1824 | * blkg->parent pointers are always valid. This returns the blkg that it ended | |
1825 | * up taking a reference on or %NULL if no reference was taken. | |
28fc591f | 1826 | */ |
13c7863d CH |
1827 | static inline struct blkcg_gq *blkg_tryget_closest(struct bio *bio, |
1828 | struct cgroup_subsys_state *css) | |
28fc591f | 1829 | { |
13c7863d | 1830 | struct blkcg_gq *blkg, *ret_blkg = NULL; |
28fc591f | 1831 | |
13c7863d | 1832 | rcu_read_lock(); |
309dca30 | 1833 | blkg = blkg_lookup_create(css_to_blkcg(css), |
ed6cddef | 1834 | bdev_get_queue(bio->bi_bdev)); |
28fc591f CH |
1835 | while (blkg) { |
1836 | if (blkg_tryget(blkg)) { | |
1837 | ret_blkg = blkg; | |
1838 | break; | |
1839 | } | |
1840 | blkg = blkg->parent; | |
1841 | } | |
13c7863d | 1842 | rcu_read_unlock(); |
28fc591f CH |
1843 | |
1844 | return ret_blkg; | |
1845 | } | |
1846 | ||
1847 | /** | |
1848 | * bio_associate_blkg_from_css - associate a bio with a specified css | |
1849 | * @bio: target bio | |
1850 | * @css: target css | |
1851 | * | |
1852 | * Associate @bio with the blkg found by combining the css's blkg and the | |
1853 | * request_queue of the @bio. An association failure is handled by walking up | |
1854 | * the blkg tree. Therefore, the blkg associated can be anything between @blkg | |
1855 | * and q->root_blkg. This situation only happens when a cgroup is dying and | |
1856 | * then the remaining bios will spill to the closest alive blkg. | |
1857 | * | |
1858 | * A reference will be taken on the blkg and will be released when @bio is | |
1859 | * freed. | |
1860 | */ | |
1861 | void bio_associate_blkg_from_css(struct bio *bio, | |
1862 | struct cgroup_subsys_state *css) | |
1863 | { | |
28fc591f CH |
1864 | if (bio->bi_blkg) |
1865 | blkg_put(bio->bi_blkg); | |
1866 | ||
a5b97526 | 1867 | if (css && css->parent) { |
13c7863d | 1868 | bio->bi_blkg = blkg_tryget_closest(bio, css); |
a5b97526 | 1869 | } else { |
ed6cddef PB |
1870 | blkg_get(bdev_get_queue(bio->bi_bdev)->root_blkg); |
1871 | bio->bi_blkg = bdev_get_queue(bio->bi_bdev)->root_blkg; | |
a5b97526 | 1872 | } |
28fc591f CH |
1873 | } |
1874 | EXPORT_SYMBOL_GPL(bio_associate_blkg_from_css); | |
1875 | ||
1876 | /** | |
1877 | * bio_associate_blkg - associate a bio with a blkg | |
1878 | * @bio: target bio | |
1879 | * | |
1880 | * Associate @bio with the blkg found from the bio's css and request_queue. | |
1881 | * If one is not found, bio_lookup_blkg() creates the blkg. If a blkg is | |
1882 | * already associated, the css is reused and association redone as the | |
1883 | * request_queue may have changed. | |
1884 | */ | |
1885 | void bio_associate_blkg(struct bio *bio) | |
1886 | { | |
1887 | struct cgroup_subsys_state *css; | |
1888 | ||
1889 | rcu_read_lock(); | |
1890 | ||
1891 | if (bio->bi_blkg) | |
1892 | css = &bio_blkcg(bio)->css; | |
1893 | else | |
1894 | css = blkcg_css(); | |
1895 | ||
1896 | bio_associate_blkg_from_css(bio, css); | |
1897 | ||
1898 | rcu_read_unlock(); | |
1899 | } | |
1900 | EXPORT_SYMBOL_GPL(bio_associate_blkg); | |
1901 | ||
1902 | /** | |
1903 | * bio_clone_blkg_association - clone blkg association from src to dst bio | |
1904 | * @dst: destination bio | |
1905 | * @src: source bio | |
1906 | */ | |
1907 | void bio_clone_blkg_association(struct bio *dst, struct bio *src) | |
1908 | { | |
1909 | if (src->bi_blkg) { | |
1910 | if (dst->bi_blkg) | |
1911 | blkg_put(dst->bi_blkg); | |
1912 | blkg_get(src->bi_blkg); | |
1913 | dst->bi_blkg = src->bi_blkg; | |
1914 | } | |
1915 | } | |
1916 | EXPORT_SYMBOL_GPL(bio_clone_blkg_association); | |
1917 | ||
db18a53e CH |
1918 | static int blk_cgroup_io_type(struct bio *bio) |
1919 | { | |
1920 | if (op_is_discard(bio->bi_opf)) | |
1921 | return BLKG_IOSTAT_DISCARD; | |
1922 | if (op_is_write(bio->bi_opf)) | |
1923 | return BLKG_IOSTAT_WRITE; | |
1924 | return BLKG_IOSTAT_READ; | |
1925 | } | |
1926 | ||
1927 | void blk_cgroup_bio_start(struct bio *bio) | |
1928 | { | |
1929 | int rwd = blk_cgroup_io_type(bio), cpu; | |
1930 | struct blkg_iostat_set *bis; | |
3c08b093 | 1931 | unsigned long flags; |
db18a53e CH |
1932 | |
1933 | cpu = get_cpu(); | |
1934 | bis = per_cpu_ptr(bio->bi_blkg->iostat_cpu, cpu); | |
3c08b093 | 1935 | flags = u64_stats_update_begin_irqsave(&bis->sync); |
db18a53e CH |
1936 | |
1937 | /* | |
1938 | * If the bio is flagged with BIO_CGROUP_ACCT it means this is a split | |
1939 | * bio and we would have already accounted for the size of the bio. | |
1940 | */ | |
1941 | if (!bio_flagged(bio, BIO_CGROUP_ACCT)) { | |
1942 | bio_set_flag(bio, BIO_CGROUP_ACCT); | |
0b8cc25d | 1943 | bis->cur.bytes[rwd] += bio->bi_iter.bi_size; |
db18a53e CH |
1944 | } |
1945 | bis->cur.ios[rwd]++; | |
1946 | ||
3c08b093 | 1947 | u64_stats_update_end_irqrestore(&bis->sync, flags); |
db18a53e CH |
1948 | if (cgroup_subsys_on_dfl(io_cgrp_subsys)) |
1949 | cgroup_rstat_updated(bio->bi_blkg->blkcg->css.cgroup, cpu); | |
1950 | put_cpu(); | |
1951 | } | |
1952 | ||
d3f77dfd TH |
1953 | static int __init blkcg_init(void) |
1954 | { | |
1955 | blkcg_punt_bio_wq = alloc_workqueue("blkcg_punt_bio", | |
1956 | WQ_MEM_RECLAIM | WQ_FREEZABLE | | |
1957 | WQ_UNBOUND | WQ_SYSFS, 0); | |
1958 | if (!blkcg_punt_bio_wq) | |
1959 | return -ENOMEM; | |
1960 | return 0; | |
1961 | } | |
1962 | subsys_initcall(blkcg_init); | |
1963 | ||
903d23f0 JB |
1964 | module_param(blkcg_debug_stats, bool, 0644); |
1965 | MODULE_PARM_DESC(blkcg_debug_stats, "True if you want debug stats, false if not"); |