]>
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" |
556910e3 | 34 | #include "blk-ioprio.h" |
3e252066 | 35 | |
838f13bf TH |
36 | /* |
37 | * blkcg_pol_mutex protects blkcg_policy[] and policy [de]activation. | |
38 | * blkcg_pol_register_mutex nests outside of it and synchronizes entire | |
39 | * policy [un]register operations including cgroup file additions / | |
40 | * removals. Putting cgroup file registration outside blkcg_pol_mutex | |
41 | * allows grabbing it from cgroup callbacks. | |
42 | */ | |
43 | static DEFINE_MUTEX(blkcg_pol_register_mutex); | |
bc0d6501 | 44 | static DEFINE_MUTEX(blkcg_pol_mutex); |
923adde1 | 45 | |
e48453c3 | 46 | struct blkcg blkcg_root; |
3c798398 | 47 | EXPORT_SYMBOL_GPL(blkcg_root); |
9d6a986c | 48 | |
496d5e75 | 49 | struct cgroup_subsys_state * const blkcg_root_css = &blkcg_root.css; |
9b0eb69b | 50 | EXPORT_SYMBOL_GPL(blkcg_root_css); |
496d5e75 | 51 | |
3c798398 | 52 | static struct blkcg_policy *blkcg_policy[BLKCG_MAX_POLS]; |
035d10b2 | 53 | |
7876f930 TH |
54 | static LIST_HEAD(all_blkcgs); /* protected by blkcg_pol_mutex */ |
55 | ||
07b0fdec | 56 | bool blkcg_debug_stats = false; |
d3f77dfd | 57 | static struct workqueue_struct *blkcg_punt_bio_wq; |
903d23f0 | 58 | |
a731763f YK |
59 | #define BLKG_DESTROY_BATCH_SIZE 64 |
60 | ||
a2b1693b | 61 | static bool blkcg_policy_enabled(struct request_queue *q, |
3c798398 | 62 | const struct blkcg_policy *pol) |
a2b1693b TH |
63 | { |
64 | return pol && test_bit(pol->plid, q->blkcg_pols); | |
65 | } | |
66 | ||
0381411e TH |
67 | /** |
68 | * blkg_free - free a blkg | |
69 | * @blkg: blkg to free | |
70 | * | |
71 | * Free @blkg which may be partially allocated. | |
72 | */ | |
3c798398 | 73 | static void blkg_free(struct blkcg_gq *blkg) |
0381411e | 74 | { |
e8989fae | 75 | int i; |
549d3aa8 TH |
76 | |
77 | if (!blkg) | |
78 | return; | |
79 | ||
db613670 | 80 | for (i = 0; i < BLKCG_MAX_POLS; i++) |
001bea73 TH |
81 | if (blkg->pd[i]) |
82 | blkcg_policy[i]->pd_free_fn(blkg->pd[i]); | |
e8989fae | 83 | |
f7331648 | 84 | free_percpu(blkg->iostat_cpu); |
ef069b97 | 85 | percpu_ref_exit(&blkg->refcnt); |
549d3aa8 | 86 | kfree(blkg); |
0381411e TH |
87 | } |
88 | ||
7fcf2b03 DZ |
89 | static void __blkg_release(struct rcu_head *rcu) |
90 | { | |
91 | struct blkcg_gq *blkg = container_of(rcu, struct blkcg_gq, rcu_head); | |
92 | ||
d3f77dfd TH |
93 | WARN_ON(!bio_list_empty(&blkg->async_bios)); |
94 | ||
7fcf2b03 DZ |
95 | /* release the blkcg and parent blkg refs this blkg has been holding */ |
96 | css_put(&blkg->blkcg->css); | |
97 | if (blkg->parent) | |
98 | blkg_put(blkg->parent); | |
7fcf2b03 DZ |
99 | blkg_free(blkg); |
100 | } | |
101 | ||
102 | /* | |
103 | * A group is RCU protected, but having an rcu lock does not mean that one | |
104 | * can access all the fields of blkg and assume these are valid. For | |
105 | * example, don't try to follow throtl_data and request queue links. | |
106 | * | |
107 | * Having a reference to blkg under an rcu allows accesses to only values | |
108 | * local to groups like group stats and group rate limits. | |
109 | */ | |
110 | static void blkg_release(struct percpu_ref *ref) | |
111 | { | |
112 | struct blkcg_gq *blkg = container_of(ref, struct blkcg_gq, refcnt); | |
113 | ||
114 | call_rcu(&blkg->rcu_head, __blkg_release); | |
115 | } | |
116 | ||
d3f77dfd TH |
117 | static void blkg_async_bio_workfn(struct work_struct *work) |
118 | { | |
119 | struct blkcg_gq *blkg = container_of(work, struct blkcg_gq, | |
120 | async_bio_work); | |
121 | struct bio_list bios = BIO_EMPTY_LIST; | |
122 | struct bio *bio; | |
192f1c6b XT |
123 | struct blk_plug plug; |
124 | bool need_plug = false; | |
d3f77dfd TH |
125 | |
126 | /* as long as there are pending bios, @blkg can't go away */ | |
127 | spin_lock_bh(&blkg->async_bio_lock); | |
128 | bio_list_merge(&bios, &blkg->async_bios); | |
129 | bio_list_init(&blkg->async_bios); | |
130 | spin_unlock_bh(&blkg->async_bio_lock); | |
131 | ||
192f1c6b XT |
132 | /* start plug only when bio_list contains at least 2 bios */ |
133 | if (bios.head && bios.head->bi_next) { | |
134 | need_plug = true; | |
135 | blk_start_plug(&plug); | |
136 | } | |
d3f77dfd TH |
137 | while ((bio = bio_list_pop(&bios))) |
138 | submit_bio(bio); | |
192f1c6b XT |
139 | if (need_plug) |
140 | blk_finish_plug(&plug); | |
d3f77dfd TH |
141 | } |
142 | ||
0381411e TH |
143 | /** |
144 | * blkg_alloc - allocate a blkg | |
145 | * @blkcg: block cgroup the new blkg is associated with | |
146 | * @q: request_queue the new blkg is associated with | |
15974993 | 147 | * @gfp_mask: allocation mask to use |
0381411e | 148 | * |
e8989fae | 149 | * Allocate a new blkg assocating @blkcg and @q. |
0381411e | 150 | */ |
15974993 TH |
151 | static struct blkcg_gq *blkg_alloc(struct blkcg *blkcg, struct request_queue *q, |
152 | gfp_t gfp_mask) | |
0381411e | 153 | { |
3c798398 | 154 | struct blkcg_gq *blkg; |
f7331648 | 155 | int i, cpu; |
0381411e TH |
156 | |
157 | /* alloc and init base part */ | |
15974993 | 158 | blkg = kzalloc_node(sizeof(*blkg), gfp_mask, q->node); |
0381411e TH |
159 | if (!blkg) |
160 | return NULL; | |
161 | ||
ef069b97 TH |
162 | if (percpu_ref_init(&blkg->refcnt, blkg_release, 0, gfp_mask)) |
163 | goto err_free; | |
164 | ||
f7331648 TH |
165 | blkg->iostat_cpu = alloc_percpu_gfp(struct blkg_iostat_set, gfp_mask); |
166 | if (!blkg->iostat_cpu) | |
77ea7338 TH |
167 | goto err_free; |
168 | ||
c875f4d0 | 169 | blkg->q = q; |
e8989fae | 170 | INIT_LIST_HEAD(&blkg->q_node); |
d3f77dfd TH |
171 | spin_lock_init(&blkg->async_bio_lock); |
172 | bio_list_init(&blkg->async_bios); | |
173 | INIT_WORK(&blkg->async_bio_work, blkg_async_bio_workfn); | |
0381411e | 174 | blkg->blkcg = blkcg; |
0381411e | 175 | |
f7331648 TH |
176 | u64_stats_init(&blkg->iostat.sync); |
177 | for_each_possible_cpu(cpu) | |
178 | u64_stats_init(&per_cpu_ptr(blkg->iostat_cpu, cpu)->sync); | |
179 | ||
8bd435b3 | 180 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
3c798398 | 181 | struct blkcg_policy *pol = blkcg_policy[i]; |
e8989fae | 182 | struct blkg_policy_data *pd; |
0381411e | 183 | |
a2b1693b | 184 | if (!blkcg_policy_enabled(q, pol)) |
e8989fae TH |
185 | continue; |
186 | ||
187 | /* alloc per-policy data and attach it to blkg */ | |
cf09a8ee | 188 | pd = pol->pd_alloc_fn(gfp_mask, q, blkcg); |
a051661c TH |
189 | if (!pd) |
190 | goto err_free; | |
549d3aa8 | 191 | |
e8989fae TH |
192 | blkg->pd[i] = pd; |
193 | pd->blkg = blkg; | |
b276a876 | 194 | pd->plid = i; |
e8989fae TH |
195 | } |
196 | ||
0381411e | 197 | return blkg; |
a051661c TH |
198 | |
199 | err_free: | |
200 | blkg_free(blkg); | |
201 | return NULL; | |
0381411e TH |
202 | } |
203 | ||
24f29046 TH |
204 | struct blkcg_gq *blkg_lookup_slowpath(struct blkcg *blkcg, |
205 | struct request_queue *q, bool update_hint) | |
80fd9979 | 206 | { |
3c798398 | 207 | struct blkcg_gq *blkg; |
80fd9979 | 208 | |
a637120e | 209 | /* |
86cde6b6 TH |
210 | * Hint didn't match. Look up from the radix tree. Note that the |
211 | * hint can only be updated under queue_lock as otherwise @blkg | |
212 | * could have already been removed from blkg_tree. The caller is | |
213 | * responsible for grabbing queue_lock if @update_hint. | |
a637120e TH |
214 | */ |
215 | blkg = radix_tree_lookup(&blkcg->blkg_tree, q->id); | |
86cde6b6 TH |
216 | if (blkg && blkg->q == q) { |
217 | if (update_hint) { | |
0d945c1f | 218 | lockdep_assert_held(&q->queue_lock); |
86cde6b6 TH |
219 | rcu_assign_pointer(blkcg->blkg_hint, blkg); |
220 | } | |
a637120e | 221 | return blkg; |
86cde6b6 | 222 | } |
a637120e | 223 | |
80fd9979 TH |
224 | return NULL; |
225 | } | |
ae118896 | 226 | EXPORT_SYMBOL_GPL(blkg_lookup_slowpath); |
80fd9979 | 227 | |
15974993 | 228 | /* |
d708f0d5 JA |
229 | * If @new_blkg is %NULL, this function tries to allocate a new one as |
230 | * necessary using %GFP_NOWAIT. @new_blkg is always consumed on return. | |
15974993 | 231 | */ |
86cde6b6 | 232 | static struct blkcg_gq *blkg_create(struct blkcg *blkcg, |
d708f0d5 JA |
233 | struct request_queue *q, |
234 | struct blkcg_gq *new_blkg) | |
5624a4e4 | 235 | { |
d708f0d5 | 236 | struct blkcg_gq *blkg; |
f427d909 | 237 | int i, ret; |
5624a4e4 | 238 | |
cd1604fa | 239 | WARN_ON_ONCE(!rcu_read_lock_held()); |
0d945c1f | 240 | lockdep_assert_held(&q->queue_lock); |
cd1604fa | 241 | |
0273ac34 DZ |
242 | /* request_queue is dying, do not create/recreate a blkg */ |
243 | if (blk_queue_dying(q)) { | |
244 | ret = -ENODEV; | |
245 | goto err_free_blkg; | |
246 | } | |
247 | ||
7ee9c562 | 248 | /* blkg holds a reference to blkcg */ |
ec903c0c | 249 | if (!css_tryget_online(&blkcg->css)) { |
20386ce0 | 250 | ret = -ENODEV; |
93e6d5d8 | 251 | goto err_free_blkg; |
15974993 | 252 | } |
cd1604fa | 253 | |
d708f0d5 JA |
254 | /* allocate */ |
255 | if (!new_blkg) { | |
256 | new_blkg = blkg_alloc(blkcg, q, GFP_NOWAIT | __GFP_NOWARN); | |
257 | if (unlikely(!new_blkg)) { | |
258 | ret = -ENOMEM; | |
8c911f3d | 259 | goto err_put_css; |
15974993 TH |
260 | } |
261 | } | |
d708f0d5 | 262 | blkg = new_blkg; |
cd1604fa | 263 | |
db613670 | 264 | /* link parent */ |
3c547865 TH |
265 | if (blkcg_parent(blkcg)) { |
266 | blkg->parent = __blkg_lookup(blkcg_parent(blkcg), q, false); | |
267 | if (WARN_ON_ONCE(!blkg->parent)) { | |
20386ce0 | 268 | ret = -ENODEV; |
8c911f3d | 269 | goto err_put_css; |
3c547865 TH |
270 | } |
271 | blkg_get(blkg->parent); | |
272 | } | |
273 | ||
db613670 TH |
274 | /* invoke per-policy init */ |
275 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
276 | struct blkcg_policy *pol = blkcg_policy[i]; | |
277 | ||
278 | if (blkg->pd[i] && pol->pd_init_fn) | |
a9520cd6 | 279 | pol->pd_init_fn(blkg->pd[i]); |
db613670 TH |
280 | } |
281 | ||
282 | /* insert */ | |
cd1604fa | 283 | spin_lock(&blkcg->lock); |
a637120e TH |
284 | ret = radix_tree_insert(&blkcg->blkg_tree, q->id, blkg); |
285 | if (likely(!ret)) { | |
286 | hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list); | |
287 | list_add(&blkg->q_node, &q->blkg_list); | |
f427d909 TH |
288 | |
289 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
290 | struct blkcg_policy *pol = blkcg_policy[i]; | |
291 | ||
292 | if (blkg->pd[i] && pol->pd_online_fn) | |
a9520cd6 | 293 | pol->pd_online_fn(blkg->pd[i]); |
f427d909 | 294 | } |
a637120e | 295 | } |
f427d909 | 296 | blkg->online = true; |
cd1604fa | 297 | spin_unlock(&blkcg->lock); |
496fb780 | 298 | |
ec13b1d6 | 299 | if (!ret) |
a637120e | 300 | return blkg; |
15974993 | 301 | |
3c547865 TH |
302 | /* @blkg failed fully initialized, use the usual release path */ |
303 | blkg_put(blkg); | |
304 | return ERR_PTR(ret); | |
305 | ||
d708f0d5 | 306 | err_put_css: |
496fb780 | 307 | css_put(&blkcg->css); |
93e6d5d8 | 308 | err_free_blkg: |
d708f0d5 | 309 | blkg_free(new_blkg); |
93e6d5d8 | 310 | return ERR_PTR(ret); |
31e4c28d | 311 | } |
3c96cb32 | 312 | |
86cde6b6 | 313 | /** |
8c546287 | 314 | * blkg_lookup_create - lookup blkg, try to create one if not there |
86cde6b6 TH |
315 | * @blkcg: blkcg of interest |
316 | * @q: request_queue of interest | |
317 | * | |
318 | * Lookup blkg for the @blkcg - @q pair. If it doesn't exist, try to | |
3c547865 TH |
319 | * create one. blkg creation is performed recursively from blkcg_root such |
320 | * that all non-root blkg's have access to the parent blkg. This function | |
8c546287 | 321 | * should be called under RCU read lock and takes @q->queue_lock. |
86cde6b6 | 322 | * |
beea9da0 DZ |
323 | * Returns the blkg or the closest blkg if blkg_create() fails as it walks |
324 | * down from root. | |
86cde6b6 | 325 | */ |
8c546287 CH |
326 | static struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg, |
327 | struct request_queue *q) | |
3c96cb32 | 328 | { |
86cde6b6 | 329 | struct blkcg_gq *blkg; |
8c546287 | 330 | unsigned long flags; |
86cde6b6 TH |
331 | |
332 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
86cde6b6 | 333 | |
8c546287 | 334 | blkg = blkg_lookup(blkcg, q); |
86cde6b6 TH |
335 | if (blkg) |
336 | return blkg; | |
337 | ||
8c546287 CH |
338 | spin_lock_irqsave(&q->queue_lock, flags); |
339 | blkg = __blkg_lookup(blkcg, q, true); | |
340 | if (blkg) | |
341 | goto found; | |
342 | ||
3c547865 TH |
343 | /* |
344 | * Create blkgs walking down from blkcg_root to @blkcg, so that all | |
beea9da0 DZ |
345 | * non-root blkgs have access to their parents. Returns the closest |
346 | * blkg to the intended blkg should blkg_create() fail. | |
3c547865 TH |
347 | */ |
348 | while (true) { | |
349 | struct blkcg *pos = blkcg; | |
350 | struct blkcg *parent = blkcg_parent(blkcg); | |
beea9da0 DZ |
351 | struct blkcg_gq *ret_blkg = q->root_blkg; |
352 | ||
353 | while (parent) { | |
354 | blkg = __blkg_lookup(parent, q, false); | |
355 | if (blkg) { | |
356 | /* remember closest blkg */ | |
357 | ret_blkg = blkg; | |
358 | break; | |
359 | } | |
3c547865 TH |
360 | pos = parent; |
361 | parent = blkcg_parent(parent); | |
362 | } | |
363 | ||
d708f0d5 | 364 | blkg = blkg_create(pos, q, NULL); |
8c546287 CH |
365 | if (IS_ERR(blkg)) { |
366 | blkg = ret_blkg; | |
367 | break; | |
368 | } | |
beea9da0 | 369 | if (pos == blkcg) |
8c546287 | 370 | break; |
b978962a DZ |
371 | } |
372 | ||
8c546287 CH |
373 | found: |
374 | spin_unlock_irqrestore(&q->queue_lock, flags); | |
b978962a DZ |
375 | return blkg; |
376 | } | |
377 | ||
3c798398 | 378 | static void blkg_destroy(struct blkcg_gq *blkg) |
03aa264a | 379 | { |
3c798398 | 380 | struct blkcg *blkcg = blkg->blkcg; |
6b065462 | 381 | int i; |
03aa264a | 382 | |
0d945c1f | 383 | lockdep_assert_held(&blkg->q->queue_lock); |
9f13ef67 | 384 | lockdep_assert_held(&blkcg->lock); |
03aa264a TH |
385 | |
386 | /* Something wrong if we are trying to remove same group twice */ | |
e8989fae | 387 | WARN_ON_ONCE(list_empty(&blkg->q_node)); |
9f13ef67 | 388 | WARN_ON_ONCE(hlist_unhashed(&blkg->blkcg_node)); |
a637120e | 389 | |
6b065462 DZF |
390 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
391 | struct blkcg_policy *pol = blkcg_policy[i]; | |
392 | ||
393 | if (blkg->pd[i] && pol->pd_offline_fn) | |
394 | pol->pd_offline_fn(blkg->pd[i]); | |
395 | } | |
396 | ||
f427d909 TH |
397 | blkg->online = false; |
398 | ||
a637120e | 399 | radix_tree_delete(&blkcg->blkg_tree, blkg->q->id); |
e8989fae | 400 | list_del_init(&blkg->q_node); |
9f13ef67 | 401 | hlist_del_init_rcu(&blkg->blkcg_node); |
03aa264a | 402 | |
a637120e TH |
403 | /* |
404 | * Both setting lookup hint to and clearing it from @blkg are done | |
405 | * under queue_lock. If it's not pointing to @blkg now, it never | |
406 | * will. Hint assignment itself can race safely. | |
407 | */ | |
ec6c676a | 408 | if (rcu_access_pointer(blkcg->blkg_hint) == blkg) |
a637120e TH |
409 | rcu_assign_pointer(blkcg->blkg_hint, NULL); |
410 | ||
03aa264a TH |
411 | /* |
412 | * Put the reference taken at the time of creation so that when all | |
413 | * queues are gone, group can be destroyed. | |
414 | */ | |
7fcf2b03 | 415 | percpu_ref_kill(&blkg->refcnt); |
03aa264a TH |
416 | } |
417 | ||
9f13ef67 TH |
418 | /** |
419 | * blkg_destroy_all - destroy all blkgs associated with a request_queue | |
420 | * @q: request_queue of interest | |
9f13ef67 | 421 | * |
3c96cb32 | 422 | * Destroy all blkgs associated with @q. |
9f13ef67 | 423 | */ |
3c96cb32 | 424 | static void blkg_destroy_all(struct request_queue *q) |
72e06c25 | 425 | { |
3c798398 | 426 | struct blkcg_gq *blkg, *n; |
a731763f | 427 | int count = BLKG_DESTROY_BATCH_SIZE; |
72e06c25 | 428 | |
a731763f | 429 | restart: |
0d945c1f | 430 | spin_lock_irq(&q->queue_lock); |
9f13ef67 | 431 | list_for_each_entry_safe(blkg, n, &q->blkg_list, q_node) { |
3c798398 | 432 | struct blkcg *blkcg = blkg->blkcg; |
72e06c25 | 433 | |
9f13ef67 TH |
434 | spin_lock(&blkcg->lock); |
435 | blkg_destroy(blkg); | |
436 | spin_unlock(&blkcg->lock); | |
a731763f YK |
437 | |
438 | /* | |
439 | * in order to avoid holding the spin lock for too long, release | |
440 | * it when a batch of blkgs are destroyed. | |
441 | */ | |
442 | if (!(--count)) { | |
443 | count = BLKG_DESTROY_BATCH_SIZE; | |
444 | spin_unlock_irq(&q->queue_lock); | |
445 | cond_resched(); | |
446 | goto restart; | |
447 | } | |
72e06c25 | 448 | } |
6fe810bd TH |
449 | |
450 | q->root_blkg = NULL; | |
0d945c1f | 451 | spin_unlock_irq(&q->queue_lock); |
72e06c25 TH |
452 | } |
453 | ||
182446d0 TH |
454 | static int blkcg_reset_stats(struct cgroup_subsys_state *css, |
455 | struct cftype *cftype, u64 val) | |
303a3acb | 456 | { |
182446d0 | 457 | struct blkcg *blkcg = css_to_blkcg(css); |
3c798398 | 458 | struct blkcg_gq *blkg; |
f7331648 | 459 | int i, cpu; |
303a3acb | 460 | |
838f13bf | 461 | mutex_lock(&blkcg_pol_mutex); |
303a3acb | 462 | spin_lock_irq(&blkcg->lock); |
997a026c TH |
463 | |
464 | /* | |
465 | * Note that stat reset is racy - it doesn't synchronize against | |
466 | * stat updates. This is a debug feature which shouldn't exist | |
467 | * anyway. If you get hit by a race, retry. | |
468 | */ | |
b67bfe0d | 469 | hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) { |
f7331648 TH |
470 | for_each_possible_cpu(cpu) { |
471 | struct blkg_iostat_set *bis = | |
472 | per_cpu_ptr(blkg->iostat_cpu, cpu); | |
473 | memset(bis, 0, sizeof(*bis)); | |
474 | } | |
475 | memset(&blkg->iostat, 0, sizeof(blkg->iostat)); | |
77ea7338 | 476 | |
8bd435b3 | 477 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
3c798398 | 478 | struct blkcg_policy *pol = blkcg_policy[i]; |
549d3aa8 | 479 | |
a9520cd6 TH |
480 | if (blkg->pd[i] && pol->pd_reset_stats_fn) |
481 | pol->pd_reset_stats_fn(blkg->pd[i]); | |
bc0d6501 | 482 | } |
303a3acb | 483 | } |
f0bdc8cd | 484 | |
303a3acb | 485 | spin_unlock_irq(&blkcg->lock); |
bc0d6501 | 486 | mutex_unlock(&blkcg_pol_mutex); |
303a3acb DS |
487 | return 0; |
488 | } | |
489 | ||
dd165eb3 | 490 | const char *blkg_dev_name(struct blkcg_gq *blkg) |
303a3acb | 491 | { |
d3d32e69 | 492 | /* some drivers (floppy) instantiate a queue w/o disk registered */ |
dc3b17cc | 493 | if (blkg->q->backing_dev_info->dev) |
d51cfc53 | 494 | return bdi_dev_name(blkg->q->backing_dev_info); |
d3d32e69 | 495 | return NULL; |
303a3acb DS |
496 | } |
497 | ||
d3d32e69 TH |
498 | /** |
499 | * blkcg_print_blkgs - helper for printing per-blkg data | |
500 | * @sf: seq_file to print to | |
501 | * @blkcg: blkcg of interest | |
502 | * @prfill: fill function to print out a blkg | |
503 | * @pol: policy in question | |
504 | * @data: data to be passed to @prfill | |
505 | * @show_total: to print out sum of prfill return values or not | |
506 | * | |
507 | * This function invokes @prfill on each blkg of @blkcg if pd for the | |
508 | * policy specified by @pol exists. @prfill is invoked with @sf, the | |
810ecfa7 TH |
509 | * policy data and @data and the matching queue lock held. If @show_total |
510 | * is %true, the sum of the return values from @prfill is printed with | |
511 | * "Total" label at the end. | |
d3d32e69 TH |
512 | * |
513 | * This is to be used to construct print functions for | |
514 | * cftype->read_seq_string method. | |
515 | */ | |
3c798398 | 516 | void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg, |
f95a04af TH |
517 | u64 (*prfill)(struct seq_file *, |
518 | struct blkg_policy_data *, int), | |
3c798398 | 519 | const struct blkcg_policy *pol, int data, |
ec399347 | 520 | bool show_total) |
5624a4e4 | 521 | { |
3c798398 | 522 | struct blkcg_gq *blkg; |
d3d32e69 | 523 | u64 total = 0; |
5624a4e4 | 524 | |
810ecfa7 | 525 | rcu_read_lock(); |
ee89f812 | 526 | hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) { |
0d945c1f | 527 | spin_lock_irq(&blkg->q->queue_lock); |
a2b1693b | 528 | if (blkcg_policy_enabled(blkg->q, pol)) |
f95a04af | 529 | total += prfill(sf, blkg->pd[pol->plid], data); |
0d945c1f | 530 | spin_unlock_irq(&blkg->q->queue_lock); |
810ecfa7 TH |
531 | } |
532 | rcu_read_unlock(); | |
d3d32e69 TH |
533 | |
534 | if (show_total) | |
535 | seq_printf(sf, "Total %llu\n", (unsigned long long)total); | |
536 | } | |
829fdb50 | 537 | EXPORT_SYMBOL_GPL(blkcg_print_blkgs); |
d3d32e69 TH |
538 | |
539 | /** | |
540 | * __blkg_prfill_u64 - prfill helper for a single u64 value | |
541 | * @sf: seq_file to print to | |
f95a04af | 542 | * @pd: policy private data of interest |
d3d32e69 TH |
543 | * @v: value to print |
544 | * | |
f95a04af | 545 | * Print @v to @sf for the device assocaited with @pd. |
d3d32e69 | 546 | */ |
f95a04af | 547 | u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v) |
d3d32e69 | 548 | { |
f95a04af | 549 | const char *dname = blkg_dev_name(pd->blkg); |
d3d32e69 TH |
550 | |
551 | if (!dname) | |
552 | return 0; | |
553 | ||
554 | seq_printf(sf, "%s %llu\n", dname, (unsigned long long)v); | |
555 | return v; | |
556 | } | |
829fdb50 | 557 | EXPORT_SYMBOL_GPL(__blkg_prfill_u64); |
d3d32e69 | 558 | |
457e490f TE |
559 | /* Performs queue bypass and policy enabled checks then looks up blkg. */ |
560 | static struct blkcg_gq *blkg_lookup_check(struct blkcg *blkcg, | |
561 | const struct blkcg_policy *pol, | |
562 | struct request_queue *q) | |
563 | { | |
564 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
0d945c1f | 565 | lockdep_assert_held(&q->queue_lock); |
457e490f TE |
566 | |
567 | if (!blkcg_policy_enabled(q, pol)) | |
568 | return ERR_PTR(-EOPNOTSUPP); | |
457e490f TE |
569 | return __blkg_lookup(blkcg, q, true /* update_hint */); |
570 | } | |
571 | ||
015d254c | 572 | /** |
22ae8ce8 | 573 | * blkcg_conf_open_bdev - parse and open bdev for per-blkg config update |
015d254c TH |
574 | * @inputp: input string pointer |
575 | * | |
576 | * Parse the device node prefix part, MAJ:MIN, of per-blkg config update | |
22ae8ce8 | 577 | * from @input and get and return the matching bdev. *@inputp is |
015d254c TH |
578 | * updated to point past the device node prefix. Returns an ERR_PTR() |
579 | * value on error. | |
580 | * | |
581 | * Use this function iff blkg_conf_prep() can't be used for some reason. | |
582 | */ | |
22ae8ce8 | 583 | struct block_device *blkcg_conf_open_bdev(char **inputp) |
015d254c TH |
584 | { |
585 | char *input = *inputp; | |
586 | unsigned int major, minor; | |
22ae8ce8 CH |
587 | struct block_device *bdev; |
588 | int key_len; | |
015d254c TH |
589 | |
590 | if (sscanf(input, "%u:%u%n", &major, &minor, &key_len) != 2) | |
591 | return ERR_PTR(-EINVAL); | |
592 | ||
593 | input += key_len; | |
594 | if (!isspace(*input)) | |
595 | return ERR_PTR(-EINVAL); | |
596 | input = skip_spaces(input); | |
597 | ||
22ae8ce8 CH |
598 | bdev = blkdev_get_no_open(MKDEV(major, minor)); |
599 | if (!bdev) | |
015d254c | 600 | return ERR_PTR(-ENODEV); |
22ae8ce8 CH |
601 | if (bdev_is_partition(bdev)) { |
602 | blkdev_put_no_open(bdev); | |
015d254c TH |
603 | return ERR_PTR(-ENODEV); |
604 | } | |
605 | ||
606 | *inputp = input; | |
22ae8ce8 | 607 | return bdev; |
015d254c TH |
608 | } |
609 | ||
3a8b31d3 TH |
610 | /** |
611 | * blkg_conf_prep - parse and prepare for per-blkg config update | |
612 | * @blkcg: target block cgroup | |
da8b0662 | 613 | * @pol: target policy |
3a8b31d3 TH |
614 | * @input: input string |
615 | * @ctx: blkg_conf_ctx to be filled | |
616 | * | |
617 | * Parse per-blkg config update from @input and initialize @ctx with the | |
36aa9e5f TH |
618 | * result. @ctx->blkg points to the blkg to be updated and @ctx->body the |
619 | * part of @input following MAJ:MIN. This function returns with RCU read | |
620 | * lock and queue lock held and must be paired with blkg_conf_finish(). | |
3a8b31d3 | 621 | */ |
3c798398 | 622 | int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol, |
36aa9e5f | 623 | char *input, struct blkg_conf_ctx *ctx) |
22ae8ce8 | 624 | __acquires(rcu) __acquires(&bdev->bd_disk->queue->queue_lock) |
34d0f179 | 625 | { |
22ae8ce8 | 626 | struct block_device *bdev; |
457e490f | 627 | struct request_queue *q; |
3c798398 | 628 | struct blkcg_gq *blkg; |
015d254c | 629 | int ret; |
36aa9e5f | 630 | |
22ae8ce8 CH |
631 | bdev = blkcg_conf_open_bdev(&input); |
632 | if (IS_ERR(bdev)) | |
633 | return PTR_ERR(bdev); | |
e56da7e2 | 634 | |
22ae8ce8 | 635 | q = bdev->bd_disk->queue; |
da8b0662 | 636 | |
457e490f | 637 | rcu_read_lock(); |
0d945c1f | 638 | spin_lock_irq(&q->queue_lock); |
e56da7e2 | 639 | |
457e490f | 640 | blkg = blkg_lookup_check(blkcg, pol, q); |
4bfd482e TH |
641 | if (IS_ERR(blkg)) { |
642 | ret = PTR_ERR(blkg); | |
457e490f TE |
643 | goto fail_unlock; |
644 | } | |
645 | ||
646 | if (blkg) | |
647 | goto success; | |
648 | ||
649 | /* | |
650 | * Create blkgs walking down from blkcg_root to @blkcg, so that all | |
651 | * non-root blkgs have access to their parents. | |
652 | */ | |
653 | while (true) { | |
654 | struct blkcg *pos = blkcg; | |
655 | struct blkcg *parent; | |
656 | struct blkcg_gq *new_blkg; | |
657 | ||
658 | parent = blkcg_parent(blkcg); | |
659 | while (parent && !__blkg_lookup(parent, q, false)) { | |
660 | pos = parent; | |
661 | parent = blkcg_parent(parent); | |
662 | } | |
663 | ||
664 | /* Drop locks to do new blkg allocation with GFP_KERNEL. */ | |
0d945c1f | 665 | spin_unlock_irq(&q->queue_lock); |
3a8b31d3 | 666 | rcu_read_unlock(); |
457e490f TE |
667 | |
668 | new_blkg = blkg_alloc(pos, q, GFP_KERNEL); | |
669 | if (unlikely(!new_blkg)) { | |
670 | ret = -ENOMEM; | |
671 | goto fail; | |
7702e8f4 | 672 | } |
3a8b31d3 | 673 | |
f255c19b GKB |
674 | if (radix_tree_preload(GFP_KERNEL)) { |
675 | blkg_free(new_blkg); | |
676 | ret = -ENOMEM; | |
677 | goto fail; | |
678 | } | |
679 | ||
457e490f | 680 | rcu_read_lock(); |
0d945c1f | 681 | spin_lock_irq(&q->queue_lock); |
457e490f TE |
682 | |
683 | blkg = blkg_lookup_check(pos, pol, q); | |
684 | if (IS_ERR(blkg)) { | |
685 | ret = PTR_ERR(blkg); | |
52abfcbd | 686 | blkg_free(new_blkg); |
f255c19b | 687 | goto fail_preloaded; |
457e490f TE |
688 | } |
689 | ||
690 | if (blkg) { | |
691 | blkg_free(new_blkg); | |
692 | } else { | |
693 | blkg = blkg_create(pos, q, new_blkg); | |
98d669b4 | 694 | if (IS_ERR(blkg)) { |
457e490f | 695 | ret = PTR_ERR(blkg); |
f255c19b | 696 | goto fail_preloaded; |
457e490f TE |
697 | } |
698 | } | |
699 | ||
f255c19b GKB |
700 | radix_tree_preload_end(); |
701 | ||
457e490f TE |
702 | if (pos == blkcg) |
703 | goto success; | |
704 | } | |
705 | success: | |
22ae8ce8 | 706 | ctx->bdev = bdev; |
3a8b31d3 | 707 | ctx->blkg = blkg; |
015d254c | 708 | ctx->body = input; |
726fa694 | 709 | return 0; |
457e490f | 710 | |
f255c19b GKB |
711 | fail_preloaded: |
712 | radix_tree_preload_end(); | |
457e490f | 713 | fail_unlock: |
0d945c1f | 714 | spin_unlock_irq(&q->queue_lock); |
457e490f TE |
715 | rcu_read_unlock(); |
716 | fail: | |
22ae8ce8 | 717 | blkdev_put_no_open(bdev); |
457e490f TE |
718 | /* |
719 | * If queue was bypassing, we should retry. Do so after a | |
720 | * short msleep(). It isn't strictly necessary but queue | |
721 | * can be bypassing for some time and it's always nice to | |
722 | * avoid busy looping. | |
723 | */ | |
724 | if (ret == -EBUSY) { | |
725 | msleep(10); | |
726 | ret = restart_syscall(); | |
727 | } | |
728 | return ret; | |
34d0f179 | 729 | } |
89f3b6d6 | 730 | EXPORT_SYMBOL_GPL(blkg_conf_prep); |
34d0f179 | 731 | |
3a8b31d3 TH |
732 | /** |
733 | * blkg_conf_finish - finish up per-blkg config update | |
734 | * @ctx: blkg_conf_ctx intiailized by blkg_conf_prep() | |
735 | * | |
736 | * Finish up after per-blkg config update. This function must be paired | |
737 | * with blkg_conf_prep(). | |
738 | */ | |
829fdb50 | 739 | void blkg_conf_finish(struct blkg_conf_ctx *ctx) |
22ae8ce8 | 740 | __releases(&ctx->bdev->bd_disk->queue->queue_lock) __releases(rcu) |
34d0f179 | 741 | { |
22ae8ce8 | 742 | spin_unlock_irq(&ctx->bdev->bd_disk->queue->queue_lock); |
3a8b31d3 | 743 | rcu_read_unlock(); |
22ae8ce8 | 744 | blkdev_put_no_open(ctx->bdev); |
34d0f179 | 745 | } |
89f3b6d6 | 746 | EXPORT_SYMBOL_GPL(blkg_conf_finish); |
34d0f179 | 747 | |
cd1fc4b9 BB |
748 | static void blkg_iostat_set(struct blkg_iostat *dst, struct blkg_iostat *src) |
749 | { | |
750 | int i; | |
751 | ||
752 | for (i = 0; i < BLKG_IOSTAT_NR; i++) { | |
753 | dst->bytes[i] = src->bytes[i]; | |
754 | dst->ios[i] = src->ios[i]; | |
755 | } | |
756 | } | |
757 | ||
758 | static void blkg_iostat_add(struct blkg_iostat *dst, struct blkg_iostat *src) | |
759 | { | |
760 | int i; | |
761 | ||
762 | for (i = 0; i < BLKG_IOSTAT_NR; i++) { | |
763 | dst->bytes[i] += src->bytes[i]; | |
764 | dst->ios[i] += src->ios[i]; | |
765 | } | |
766 | } | |
767 | ||
768 | static void blkg_iostat_sub(struct blkg_iostat *dst, struct blkg_iostat *src) | |
769 | { | |
770 | int i; | |
771 | ||
772 | for (i = 0; i < BLKG_IOSTAT_NR; i++) { | |
773 | dst->bytes[i] -= src->bytes[i]; | |
774 | dst->ios[i] -= src->ios[i]; | |
775 | } | |
776 | } | |
777 | ||
778 | static void blkcg_rstat_flush(struct cgroup_subsys_state *css, int cpu) | |
779 | { | |
780 | struct blkcg *blkcg = css_to_blkcg(css); | |
781 | struct blkcg_gq *blkg; | |
782 | ||
dc26532a JW |
783 | /* Root-level stats are sourced from system-wide IO stats */ |
784 | if (!cgroup_parent(css->cgroup)) | |
785 | return; | |
786 | ||
cd1fc4b9 BB |
787 | rcu_read_lock(); |
788 | ||
789 | hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) { | |
790 | struct blkcg_gq *parent = blkg->parent; | |
791 | struct blkg_iostat_set *bisc = per_cpu_ptr(blkg->iostat_cpu, cpu); | |
792 | struct blkg_iostat cur, delta; | |
793 | unsigned int seq; | |
794 | ||
795 | /* fetch the current per-cpu values */ | |
796 | do { | |
797 | seq = u64_stats_fetch_begin(&bisc->sync); | |
798 | blkg_iostat_set(&cur, &bisc->cur); | |
799 | } while (u64_stats_fetch_retry(&bisc->sync, seq)); | |
800 | ||
801 | /* propagate percpu delta to global */ | |
802 | u64_stats_update_begin(&blkg->iostat.sync); | |
803 | blkg_iostat_set(&delta, &cur); | |
804 | blkg_iostat_sub(&delta, &bisc->last); | |
805 | blkg_iostat_add(&blkg->iostat.cur, &delta); | |
806 | blkg_iostat_add(&bisc->last, &delta); | |
807 | u64_stats_update_end(&blkg->iostat.sync); | |
808 | ||
dc26532a JW |
809 | /* propagate global delta to parent (unless that's root) */ |
810 | if (parent && parent->parent) { | |
cd1fc4b9 BB |
811 | u64_stats_update_begin(&parent->iostat.sync); |
812 | blkg_iostat_set(&delta, &blkg->iostat.cur); | |
813 | blkg_iostat_sub(&delta, &blkg->iostat.last); | |
814 | blkg_iostat_add(&parent->iostat.cur, &delta); | |
815 | blkg_iostat_add(&blkg->iostat.last, &delta); | |
816 | u64_stats_update_end(&parent->iostat.sync); | |
817 | } | |
818 | } | |
819 | ||
820 | rcu_read_unlock(); | |
821 | } | |
822 | ||
ef45fe47 | 823 | /* |
dc26532a JW |
824 | * We source root cgroup stats from the system-wide stats to avoid |
825 | * tracking the same information twice and incurring overhead when no | |
826 | * cgroups are defined. For that reason, cgroup_rstat_flush in | |
827 | * blkcg_print_stat does not actually fill out the iostat in the root | |
828 | * cgroup's blkcg_gq. | |
ef45fe47 BB |
829 | * |
830 | * However, we would like to re-use the printing code between the root and | |
831 | * non-root cgroups to the extent possible. For that reason, we simulate | |
832 | * flushing the root cgroup's stats by explicitly filling in the iostat | |
833 | * with disk level statistics. | |
834 | */ | |
835 | static void blkcg_fill_root_iostats(void) | |
836 | { | |
837 | struct class_dev_iter iter; | |
838 | struct device *dev; | |
839 | ||
840 | class_dev_iter_init(&iter, &block_class, NULL, &disk_type); | |
841 | while ((dev = class_dev_iter_next(&iter))) { | |
0d02129e CH |
842 | struct block_device *bdev = dev_to_bdev(dev); |
843 | struct blkcg_gq *blkg = | |
844 | blk_queue_root_blkg(bdev->bd_disk->queue); | |
ef45fe47 BB |
845 | struct blkg_iostat tmp; |
846 | int cpu; | |
847 | ||
848 | memset(&tmp, 0, sizeof(tmp)); | |
849 | for_each_possible_cpu(cpu) { | |
850 | struct disk_stats *cpu_dkstats; | |
851 | ||
0d02129e | 852 | cpu_dkstats = per_cpu_ptr(bdev->bd_stats, cpu); |
ef45fe47 BB |
853 | tmp.ios[BLKG_IOSTAT_READ] += |
854 | cpu_dkstats->ios[STAT_READ]; | |
855 | tmp.ios[BLKG_IOSTAT_WRITE] += | |
856 | cpu_dkstats->ios[STAT_WRITE]; | |
857 | tmp.ios[BLKG_IOSTAT_DISCARD] += | |
858 | cpu_dkstats->ios[STAT_DISCARD]; | |
859 | // convert sectors to bytes | |
860 | tmp.bytes[BLKG_IOSTAT_READ] += | |
861 | cpu_dkstats->sectors[STAT_READ] << 9; | |
862 | tmp.bytes[BLKG_IOSTAT_WRITE] += | |
863 | cpu_dkstats->sectors[STAT_WRITE] << 9; | |
864 | tmp.bytes[BLKG_IOSTAT_DISCARD] += | |
865 | cpu_dkstats->sectors[STAT_DISCARD] << 9; | |
866 | ||
867 | u64_stats_update_begin(&blkg->iostat.sync); | |
868 | blkg_iostat_set(&blkg->iostat.cur, &tmp); | |
869 | u64_stats_update_end(&blkg->iostat.sync); | |
870 | } | |
871 | } | |
872 | } | |
873 | ||
2ee867dc TH |
874 | static int blkcg_print_stat(struct seq_file *sf, void *v) |
875 | { | |
876 | struct blkcg *blkcg = css_to_blkcg(seq_css(sf)); | |
877 | struct blkcg_gq *blkg; | |
878 | ||
ef45fe47 BB |
879 | if (!seq_css(sf)->parent) |
880 | blkcg_fill_root_iostats(); | |
881 | else | |
882 | cgroup_rstat_flush(blkcg->css.cgroup); | |
883 | ||
2ee867dc TH |
884 | rcu_read_lock(); |
885 | ||
886 | hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) { | |
f7331648 | 887 | struct blkg_iostat_set *bis = &blkg->iostat; |
2ee867dc | 888 | const char *dname; |
903d23f0 | 889 | char *buf; |
636620b6 | 890 | u64 rbytes, wbytes, rios, wios, dbytes, dios; |
903d23f0 JB |
891 | size_t size = seq_get_buf(sf, &buf), off = 0; |
892 | int i; | |
893 | bool has_stats = false; | |
f7331648 | 894 | unsigned seq; |
2ee867dc | 895 | |
b0814361 TH |
896 | spin_lock_irq(&blkg->q->queue_lock); |
897 | ||
898 | if (!blkg->online) | |
899 | goto skip; | |
900 | ||
2ee867dc TH |
901 | dname = blkg_dev_name(blkg); |
902 | if (!dname) | |
b0814361 | 903 | goto skip; |
2ee867dc | 904 | |
903d23f0 JB |
905 | /* |
906 | * Hooray string manipulation, count is the size written NOT | |
907 | * INCLUDING THE \0, so size is now count+1 less than what we | |
908 | * had before, but we want to start writing the next bit from | |
909 | * the \0 so we only add count to buf. | |
910 | */ | |
911 | off += scnprintf(buf+off, size-off, "%s ", dname); | |
912 | ||
f7331648 TH |
913 | do { |
914 | seq = u64_stats_fetch_begin(&bis->sync); | |
2ee867dc | 915 | |
f7331648 TH |
916 | rbytes = bis->cur.bytes[BLKG_IOSTAT_READ]; |
917 | wbytes = bis->cur.bytes[BLKG_IOSTAT_WRITE]; | |
918 | dbytes = bis->cur.bytes[BLKG_IOSTAT_DISCARD]; | |
919 | rios = bis->cur.ios[BLKG_IOSTAT_READ]; | |
920 | wios = bis->cur.ios[BLKG_IOSTAT_WRITE]; | |
921 | dios = bis->cur.ios[BLKG_IOSTAT_DISCARD]; | |
922 | } while (u64_stats_fetch_retry(&bis->sync, seq)); | |
2ee867dc | 923 | |
903d23f0 JB |
924 | if (rbytes || wbytes || rios || wios) { |
925 | has_stats = true; | |
926 | off += scnprintf(buf+off, size-off, | |
636620b6 TH |
927 | "rbytes=%llu wbytes=%llu rios=%llu wios=%llu dbytes=%llu dios=%llu", |
928 | rbytes, wbytes, rios, wios, | |
929 | dbytes, dios); | |
903d23f0 JB |
930 | } |
931 | ||
07b0fdec | 932 | if (blkcg_debug_stats && atomic_read(&blkg->use_delay)) { |
d09d8df3 JB |
933 | has_stats = true; |
934 | off += scnprintf(buf+off, size-off, | |
935 | " use_delay=%d delay_nsec=%llu", | |
936 | atomic_read(&blkg->use_delay), | |
937 | (unsigned long long)atomic64_read(&blkg->delay_nsec)); | |
938 | } | |
939 | ||
903d23f0 JB |
940 | for (i = 0; i < BLKCG_MAX_POLS; i++) { |
941 | struct blkcg_policy *pol = blkcg_policy[i]; | |
942 | size_t written; | |
943 | ||
944 | if (!blkg->pd[i] || !pol->pd_stat_fn) | |
945 | continue; | |
946 | ||
947 | written = pol->pd_stat_fn(blkg->pd[i], buf+off, size-off); | |
948 | if (written) | |
949 | has_stats = true; | |
950 | off += written; | |
951 | } | |
07b0fdec | 952 | |
903d23f0 | 953 | if (has_stats) { |
f539da82 TH |
954 | if (off < size - 1) { |
955 | off += scnprintf(buf+off, size-off, "\n"); | |
956 | seq_commit(sf, off); | |
957 | } else { | |
958 | seq_commit(sf, -1); | |
959 | } | |
903d23f0 | 960 | } |
b0814361 TH |
961 | skip: |
962 | spin_unlock_irq(&blkg->q->queue_lock); | |
2ee867dc TH |
963 | } |
964 | ||
965 | rcu_read_unlock(); | |
966 | return 0; | |
967 | } | |
968 | ||
e1f3b941 | 969 | static struct cftype blkcg_files[] = { |
2ee867dc TH |
970 | { |
971 | .name = "stat", | |
972 | .seq_show = blkcg_print_stat, | |
973 | }, | |
974 | { } /* terminate */ | |
975 | }; | |
976 | ||
e1f3b941 | 977 | static struct cftype blkcg_legacy_files[] = { |
84c124da DS |
978 | { |
979 | .name = "reset_stats", | |
3c798398 | 980 | .write_u64 = blkcg_reset_stats, |
22084190 | 981 | }, |
4baf6e33 | 982 | { } /* terminate */ |
31e4c28d VG |
983 | }; |
984 | ||
59b57717 DZF |
985 | /* |
986 | * blkcg destruction is a three-stage process. | |
987 | * | |
988 | * 1. Destruction starts. The blkcg_css_offline() callback is invoked | |
989 | * which offlines writeback. Here we tie the next stage of blkg destruction | |
990 | * to the completion of writeback associated with the blkcg. This lets us | |
991 | * avoid punting potentially large amounts of outstanding writeback to root | |
992 | * while maintaining any ongoing policies. The next stage is triggered when | |
993 | * the nr_cgwbs count goes to zero. | |
994 | * | |
995 | * 2. When the nr_cgwbs count goes to zero, blkcg_destroy_blkgs() is called | |
996 | * and handles the destruction of blkgs. Here the css reference held by | |
997 | * the blkg is put back eventually allowing blkcg_css_free() to be called. | |
998 | * This work may occur in cgwb_release_workfn() on the cgwb_release | |
999 | * workqueue. Any submitted ios that fail to get the blkg ref will be | |
1000 | * punted to the root_blkg. | |
1001 | * | |
1002 | * 3. Once the blkcg ref count goes to zero, blkcg_css_free() is called. | |
1003 | * This finally frees the blkcg. | |
1004 | */ | |
1005 | ||
9f13ef67 | 1006 | /** |
92fb9748 | 1007 | * blkcg_css_offline - cgroup css_offline callback |
eb95419b | 1008 | * @css: css of interest |
9f13ef67 | 1009 | * |
59b57717 DZF |
1010 | * This function is called when @css is about to go away. Here the cgwbs are |
1011 | * offlined first and only once writeback associated with the blkcg has | |
1012 | * finished do we start step 2 (see above). | |
9f13ef67 | 1013 | */ |
eb95419b | 1014 | static void blkcg_css_offline(struct cgroup_subsys_state *css) |
31e4c28d | 1015 | { |
eb95419b | 1016 | struct blkcg *blkcg = css_to_blkcg(css); |
b1c35769 | 1017 | |
59b57717 DZF |
1018 | /* this prevents anyone from attaching or migrating to this blkcg */ |
1019 | wb_blkcg_offline(blkcg); | |
1020 | ||
d866dbf6 TH |
1021 | /* put the base online pin allowing step 2 to be triggered */ |
1022 | blkcg_unpin_online(blkcg); | |
59b57717 DZF |
1023 | } |
1024 | ||
1025 | /** | |
1026 | * blkcg_destroy_blkgs - responsible for shooting down blkgs | |
1027 | * @blkcg: blkcg of interest | |
1028 | * | |
1029 | * blkgs should be removed while holding both q and blkcg locks. As blkcg lock | |
1030 | * is nested inside q lock, this function performs reverse double lock dancing. | |
1031 | * Destroying the blkgs releases the reference held on the blkcg's css allowing | |
1032 | * blkcg_css_free to eventually be called. | |
1033 | * | |
1034 | * This is the blkcg counterpart of ioc_release_fn(). | |
1035 | */ | |
1036 | void blkcg_destroy_blkgs(struct blkcg *blkcg) | |
1037 | { | |
6c635cae BW |
1038 | might_sleep(); |
1039 | ||
9f13ef67 | 1040 | spin_lock_irq(&blkcg->lock); |
7ee9c562 | 1041 | |
4c699480 JQ |
1042 | while (!hlist_empty(&blkcg->blkg_list)) { |
1043 | struct blkcg_gq *blkg = hlist_entry(blkcg->blkg_list.first, | |
6b065462 | 1044 | struct blkcg_gq, blkcg_node); |
4c699480 JQ |
1045 | struct request_queue *q = blkg->q; |
1046 | ||
6c635cae BW |
1047 | if (need_resched() || !spin_trylock(&q->queue_lock)) { |
1048 | /* | |
1049 | * Given that the system can accumulate a huge number | |
1050 | * of blkgs in pathological cases, check to see if we | |
1051 | * need to rescheduling to avoid softlockup. | |
1052 | */ | |
4c699480 | 1053 | spin_unlock_irq(&blkcg->lock); |
6c635cae | 1054 | cond_resched(); |
4c699480 | 1055 | spin_lock_irq(&blkcg->lock); |
6c635cae | 1056 | continue; |
4c699480 | 1057 | } |
6c635cae BW |
1058 | |
1059 | blkg_destroy(blkg); | |
1060 | spin_unlock(&q->queue_lock); | |
4c699480 | 1061 | } |
6b065462 | 1062 | |
4c699480 JQ |
1063 | spin_unlock_irq(&blkcg->lock); |
1064 | } | |
1065 | ||
eb95419b | 1066 | static void blkcg_css_free(struct cgroup_subsys_state *css) |
7ee9c562 | 1067 | { |
eb95419b | 1068 | struct blkcg *blkcg = css_to_blkcg(css); |
bc915e61 | 1069 | int i; |
7ee9c562 | 1070 | |
7876f930 | 1071 | mutex_lock(&blkcg_pol_mutex); |
e4a9bde9 | 1072 | |
7876f930 | 1073 | list_del(&blkcg->all_blkcgs_node); |
7876f930 | 1074 | |
bc915e61 | 1075 | for (i = 0; i < BLKCG_MAX_POLS; i++) |
e4a9bde9 TH |
1076 | if (blkcg->cpd[i]) |
1077 | blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]); | |
1078 | ||
1079 | mutex_unlock(&blkcg_pol_mutex); | |
1080 | ||
bc915e61 | 1081 | kfree(blkcg); |
31e4c28d VG |
1082 | } |
1083 | ||
eb95419b TH |
1084 | static struct cgroup_subsys_state * |
1085 | blkcg_css_alloc(struct cgroup_subsys_state *parent_css) | |
31e4c28d | 1086 | { |
3c798398 | 1087 | struct blkcg *blkcg; |
e48453c3 AA |
1088 | struct cgroup_subsys_state *ret; |
1089 | int i; | |
31e4c28d | 1090 | |
7876f930 TH |
1091 | mutex_lock(&blkcg_pol_mutex); |
1092 | ||
eb95419b | 1093 | if (!parent_css) { |
3c798398 | 1094 | blkcg = &blkcg_root; |
bc915e61 TH |
1095 | } else { |
1096 | blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL); | |
1097 | if (!blkcg) { | |
1098 | ret = ERR_PTR(-ENOMEM); | |
4c18c9e9 | 1099 | goto unlock; |
bc915e61 | 1100 | } |
e48453c3 AA |
1101 | } |
1102 | ||
1103 | for (i = 0; i < BLKCG_MAX_POLS ; i++) { | |
1104 | struct blkcg_policy *pol = blkcg_policy[i]; | |
1105 | struct blkcg_policy_data *cpd; | |
1106 | ||
1107 | /* | |
1108 | * If the policy hasn't been attached yet, wait for it | |
1109 | * to be attached before doing anything else. Otherwise, | |
1110 | * check if the policy requires any specific per-cgroup | |
1111 | * data: if it does, allocate and initialize it. | |
1112 | */ | |
e4a9bde9 | 1113 | if (!pol || !pol->cpd_alloc_fn) |
e48453c3 AA |
1114 | continue; |
1115 | ||
e4a9bde9 | 1116 | cpd = pol->cpd_alloc_fn(GFP_KERNEL); |
e48453c3 AA |
1117 | if (!cpd) { |
1118 | ret = ERR_PTR(-ENOMEM); | |
1119 | goto free_pd_blkcg; | |
1120 | } | |
81437648 TH |
1121 | blkcg->cpd[i] = cpd; |
1122 | cpd->blkcg = blkcg; | |
e48453c3 | 1123 | cpd->plid = i; |
e4a9bde9 TH |
1124 | if (pol->cpd_init_fn) |
1125 | pol->cpd_init_fn(cpd); | |
e48453c3 | 1126 | } |
31e4c28d | 1127 | |
31e4c28d | 1128 | spin_lock_init(&blkcg->lock); |
d866dbf6 | 1129 | refcount_set(&blkcg->online_pin, 1); |
e00f4f4d | 1130 | INIT_RADIX_TREE(&blkcg->blkg_tree, GFP_NOWAIT | __GFP_NOWARN); |
31e4c28d | 1131 | INIT_HLIST_HEAD(&blkcg->blkg_list); |
52ebea74 TH |
1132 | #ifdef CONFIG_CGROUP_WRITEBACK |
1133 | INIT_LIST_HEAD(&blkcg->cgwb_list); | |
1134 | #endif | |
7876f930 TH |
1135 | list_add_tail(&blkcg->all_blkcgs_node, &all_blkcgs); |
1136 | ||
1137 | mutex_unlock(&blkcg_pol_mutex); | |
31e4c28d | 1138 | return &blkcg->css; |
e48453c3 AA |
1139 | |
1140 | free_pd_blkcg: | |
1141 | for (i--; i >= 0; i--) | |
e4a9bde9 TH |
1142 | if (blkcg->cpd[i]) |
1143 | blkcg_policy[i]->cpd_free_fn(blkcg->cpd[i]); | |
4c18c9e9 | 1144 | |
1145 | if (blkcg != &blkcg_root) | |
1146 | kfree(blkcg); | |
1147 | unlock: | |
7876f930 | 1148 | mutex_unlock(&blkcg_pol_mutex); |
e48453c3 | 1149 | return ret; |
31e4c28d VG |
1150 | } |
1151 | ||
4308a434 TH |
1152 | static int blkcg_css_online(struct cgroup_subsys_state *css) |
1153 | { | |
1154 | struct blkcg *blkcg = css_to_blkcg(css); | |
1155 | struct blkcg *parent = blkcg_parent(blkcg); | |
1156 | ||
1157 | /* | |
1158 | * blkcg_pin_online() is used to delay blkcg offline so that blkgs | |
1159 | * don't go offline while cgwbs are still active on them. Pin the | |
1160 | * parent so that offline always happens towards the root. | |
1161 | */ | |
1162 | if (parent) | |
1163 | blkcg_pin_online(parent); | |
1164 | return 0; | |
1165 | } | |
1166 | ||
5efd6113 TH |
1167 | /** |
1168 | * blkcg_init_queue - initialize blkcg part of request queue | |
1169 | * @q: request_queue to initialize | |
1170 | * | |
c62b37d9 | 1171 | * Called from blk_alloc_queue(). Responsible for initializing blkcg |
5efd6113 TH |
1172 | * part of new request_queue @q. |
1173 | * | |
1174 | * RETURNS: | |
1175 | * 0 on success, -errno on failure. | |
1176 | */ | |
1177 | int blkcg_init_queue(struct request_queue *q) | |
1178 | { | |
d708f0d5 JA |
1179 | struct blkcg_gq *new_blkg, *blkg; |
1180 | bool preloaded; | |
ec13b1d6 TH |
1181 | int ret; |
1182 | ||
d708f0d5 JA |
1183 | new_blkg = blkg_alloc(&blkcg_root, q, GFP_KERNEL); |
1184 | if (!new_blkg) | |
1185 | return -ENOMEM; | |
1186 | ||
1187 | preloaded = !radix_tree_preload(GFP_KERNEL); | |
1188 | ||
bea54883 | 1189 | /* Make sure the root blkg exists. */ |
ec13b1d6 | 1190 | rcu_read_lock(); |
0d945c1f | 1191 | spin_lock_irq(&q->queue_lock); |
d708f0d5 | 1192 | blkg = blkg_create(&blkcg_root, q, new_blkg); |
901932a3 JB |
1193 | if (IS_ERR(blkg)) |
1194 | goto err_unlock; | |
1195 | q->root_blkg = blkg; | |
0d945c1f | 1196 | spin_unlock_irq(&q->queue_lock); |
ec13b1d6 TH |
1197 | rcu_read_unlock(); |
1198 | ||
d708f0d5 JA |
1199 | if (preloaded) |
1200 | radix_tree_preload_end(); | |
1201 | ||
19688d7f | 1202 | ret = blk_iolatency_init(q); |
04be60b5 CH |
1203 | if (ret) |
1204 | goto err_destroy_all; | |
d7067512 | 1205 | |
556910e3 BVA |
1206 | ret = blk_ioprio_init(q); |
1207 | if (ret) | |
1208 | goto err_destroy_all; | |
1209 | ||
19688d7f BVA |
1210 | ret = blk_throtl_init(q); |
1211 | if (ret) | |
04be60b5 | 1212 | goto err_destroy_all; |
19688d7f | 1213 | |
04be60b5 | 1214 | return 0; |
901932a3 | 1215 | |
04be60b5 | 1216 | err_destroy_all: |
04be60b5 | 1217 | blkg_destroy_all(q); |
04be60b5 | 1218 | return ret; |
901932a3 | 1219 | err_unlock: |
0d945c1f | 1220 | spin_unlock_irq(&q->queue_lock); |
901932a3 JB |
1221 | rcu_read_unlock(); |
1222 | if (preloaded) | |
1223 | radix_tree_preload_end(); | |
1224 | return PTR_ERR(blkg); | |
5efd6113 TH |
1225 | } |
1226 | ||
5efd6113 TH |
1227 | /** |
1228 | * blkcg_exit_queue - exit and release blkcg part of request_queue | |
1229 | * @q: request_queue being released | |
1230 | * | |
7585d508 | 1231 | * Called from blk_exit_queue(). Responsible for exiting blkcg part. |
5efd6113 TH |
1232 | */ |
1233 | void blkcg_exit_queue(struct request_queue *q) | |
1234 | { | |
3c96cb32 | 1235 | blkg_destroy_all(q); |
5efd6113 TH |
1236 | blk_throtl_exit(q); |
1237 | } | |
1238 | ||
69d7fde5 TH |
1239 | static void blkcg_bind(struct cgroup_subsys_state *root_css) |
1240 | { | |
1241 | int i; | |
1242 | ||
1243 | mutex_lock(&blkcg_pol_mutex); | |
1244 | ||
1245 | for (i = 0; i < BLKCG_MAX_POLS; i++) { | |
1246 | struct blkcg_policy *pol = blkcg_policy[i]; | |
1247 | struct blkcg *blkcg; | |
1248 | ||
1249 | if (!pol || !pol->cpd_bind_fn) | |
1250 | continue; | |
1251 | ||
1252 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) | |
1253 | if (blkcg->cpd[pol->plid]) | |
1254 | pol->cpd_bind_fn(blkcg->cpd[pol->plid]); | |
1255 | } | |
1256 | mutex_unlock(&blkcg_pol_mutex); | |
1257 | } | |
1258 | ||
d09d8df3 JB |
1259 | static void blkcg_exit(struct task_struct *tsk) |
1260 | { | |
1261 | if (tsk->throttle_queue) | |
1262 | blk_put_queue(tsk->throttle_queue); | |
1263 | tsk->throttle_queue = NULL; | |
1264 | } | |
1265 | ||
c165b3e3 | 1266 | struct cgroup_subsys io_cgrp_subsys = { |
92fb9748 | 1267 | .css_alloc = blkcg_css_alloc, |
4308a434 | 1268 | .css_online = blkcg_css_online, |
92fb9748 TH |
1269 | .css_offline = blkcg_css_offline, |
1270 | .css_free = blkcg_css_free, | |
f7331648 | 1271 | .css_rstat_flush = blkcg_rstat_flush, |
69d7fde5 | 1272 | .bind = blkcg_bind, |
2ee867dc | 1273 | .dfl_cftypes = blkcg_files, |
880f50e2 | 1274 | .legacy_cftypes = blkcg_legacy_files, |
c165b3e3 | 1275 | .legacy_name = "blkio", |
d09d8df3 | 1276 | .exit = blkcg_exit, |
1ced953b TH |
1277 | #ifdef CONFIG_MEMCG |
1278 | /* | |
1279 | * This ensures that, if available, memcg is automatically enabled | |
1280 | * together on the default hierarchy so that the owner cgroup can | |
1281 | * be retrieved from writeback pages. | |
1282 | */ | |
1283 | .depends_on = 1 << memory_cgrp_id, | |
1284 | #endif | |
676f7c8f | 1285 | }; |
c165b3e3 | 1286 | EXPORT_SYMBOL_GPL(io_cgrp_subsys); |
676f7c8f | 1287 | |
a2b1693b TH |
1288 | /** |
1289 | * blkcg_activate_policy - activate a blkcg policy on a request_queue | |
1290 | * @q: request_queue of interest | |
1291 | * @pol: blkcg policy to activate | |
1292 | * | |
1293 | * Activate @pol on @q. Requires %GFP_KERNEL context. @q goes through | |
1294 | * bypass mode to populate its blkgs with policy_data for @pol. | |
1295 | * | |
1296 | * Activation happens with @q bypassed, so nobody would be accessing blkgs | |
1297 | * from IO path. Update of each blkg is protected by both queue and blkcg | |
1298 | * locks so that holding either lock and testing blkcg_policy_enabled() is | |
1299 | * always enough for dereferencing policy data. | |
1300 | * | |
1301 | * The caller is responsible for synchronizing [de]activations and policy | |
1302 | * [un]registerations. Returns 0 on success, -errno on failure. | |
1303 | */ | |
1304 | int blkcg_activate_policy(struct request_queue *q, | |
3c798398 | 1305 | const struct blkcg_policy *pol) |
a2b1693b | 1306 | { |
4c55f4f9 | 1307 | struct blkg_policy_data *pd_prealloc = NULL; |
9d179b86 | 1308 | struct blkcg_gq *blkg, *pinned_blkg = NULL; |
4c55f4f9 | 1309 | int ret; |
a2b1693b TH |
1310 | |
1311 | if (blkcg_policy_enabled(q, pol)) | |
1312 | return 0; | |
1313 | ||
344e9ffc | 1314 | if (queue_is_mq(q)) |
bd166ef1 | 1315 | blk_mq_freeze_queue(q); |
9d179b86 | 1316 | retry: |
0d945c1f | 1317 | spin_lock_irq(&q->queue_lock); |
a2b1693b | 1318 | |
9d179b86 | 1319 | /* blkg_list is pushed at the head, reverse walk to allocate parents first */ |
71c81407 | 1320 | list_for_each_entry_reverse(blkg, &q->blkg_list, q_node) { |
4c55f4f9 TH |
1321 | struct blkg_policy_data *pd; |
1322 | ||
1323 | if (blkg->pd[pol->plid]) | |
1324 | continue; | |
a2b1693b | 1325 | |
9d179b86 TH |
1326 | /* If prealloc matches, use it; otherwise try GFP_NOWAIT */ |
1327 | if (blkg == pinned_blkg) { | |
1328 | pd = pd_prealloc; | |
1329 | pd_prealloc = NULL; | |
1330 | } else { | |
1331 | pd = pol->pd_alloc_fn(GFP_NOWAIT | __GFP_NOWARN, q, | |
1332 | blkg->blkcg); | |
1333 | } | |
1334 | ||
4c55f4f9 | 1335 | if (!pd) { |
9d179b86 TH |
1336 | /* |
1337 | * GFP_NOWAIT failed. Free the existing one and | |
1338 | * prealloc for @blkg w/ GFP_KERNEL. | |
1339 | */ | |
1340 | if (pinned_blkg) | |
1341 | blkg_put(pinned_blkg); | |
1342 | blkg_get(blkg); | |
1343 | pinned_blkg = blkg; | |
1344 | ||
0d945c1f | 1345 | spin_unlock_irq(&q->queue_lock); |
9d179b86 TH |
1346 | |
1347 | if (pd_prealloc) | |
1348 | pol->pd_free_fn(pd_prealloc); | |
1349 | pd_prealloc = pol->pd_alloc_fn(GFP_KERNEL, q, | |
1350 | blkg->blkcg); | |
1351 | if (pd_prealloc) | |
1352 | goto retry; | |
1353 | else | |
1354 | goto enomem; | |
4c55f4f9 | 1355 | } |
a2b1693b TH |
1356 | |
1357 | blkg->pd[pol->plid] = pd; | |
1358 | pd->blkg = blkg; | |
b276a876 | 1359 | pd->plid = pol->plid; |
a2b1693b TH |
1360 | } |
1361 | ||
9d179b86 TH |
1362 | /* all allocated, init in the same order */ |
1363 | if (pol->pd_init_fn) | |
1364 | list_for_each_entry_reverse(blkg, &q->blkg_list, q_node) | |
1365 | pol->pd_init_fn(blkg->pd[pol->plid]); | |
1366 | ||
a2b1693b TH |
1367 | __set_bit(pol->plid, q->blkcg_pols); |
1368 | ret = 0; | |
4c55f4f9 | 1369 | |
0d945c1f | 1370 | spin_unlock_irq(&q->queue_lock); |
9d179b86 | 1371 | out: |
344e9ffc | 1372 | if (queue_is_mq(q)) |
bd166ef1 | 1373 | blk_mq_unfreeze_queue(q); |
9d179b86 TH |
1374 | if (pinned_blkg) |
1375 | blkg_put(pinned_blkg); | |
001bea73 TH |
1376 | if (pd_prealloc) |
1377 | pol->pd_free_fn(pd_prealloc); | |
a2b1693b | 1378 | return ret; |
9d179b86 TH |
1379 | |
1380 | enomem: | |
1381 | /* alloc failed, nothing's initialized yet, free everything */ | |
1382 | spin_lock_irq(&q->queue_lock); | |
1383 | list_for_each_entry(blkg, &q->blkg_list, q_node) { | |
1384 | if (blkg->pd[pol->plid]) { | |
1385 | pol->pd_free_fn(blkg->pd[pol->plid]); | |
1386 | blkg->pd[pol->plid] = NULL; | |
1387 | } | |
1388 | } | |
1389 | spin_unlock_irq(&q->queue_lock); | |
1390 | ret = -ENOMEM; | |
1391 | goto out; | |
a2b1693b TH |
1392 | } |
1393 | EXPORT_SYMBOL_GPL(blkcg_activate_policy); | |
1394 | ||
1395 | /** | |
1396 | * blkcg_deactivate_policy - deactivate a blkcg policy on a request_queue | |
1397 | * @q: request_queue of interest | |
1398 | * @pol: blkcg policy to deactivate | |
1399 | * | |
1400 | * Deactivate @pol on @q. Follows the same synchronization rules as | |
1401 | * blkcg_activate_policy(). | |
1402 | */ | |
1403 | void blkcg_deactivate_policy(struct request_queue *q, | |
3c798398 | 1404 | const struct blkcg_policy *pol) |
a2b1693b | 1405 | { |
3c798398 | 1406 | struct blkcg_gq *blkg; |
a2b1693b TH |
1407 | |
1408 | if (!blkcg_policy_enabled(q, pol)) | |
1409 | return; | |
1410 | ||
344e9ffc | 1411 | if (queue_is_mq(q)) |
bd166ef1 | 1412 | blk_mq_freeze_queue(q); |
bd166ef1 | 1413 | |
0d945c1f | 1414 | spin_lock_irq(&q->queue_lock); |
a2b1693b TH |
1415 | |
1416 | __clear_bit(pol->plid, q->blkcg_pols); | |
1417 | ||
1418 | list_for_each_entry(blkg, &q->blkg_list, q_node) { | |
001bea73 | 1419 | if (blkg->pd[pol->plid]) { |
6b065462 | 1420 | if (pol->pd_offline_fn) |
a9520cd6 | 1421 | pol->pd_offline_fn(blkg->pd[pol->plid]); |
001bea73 TH |
1422 | pol->pd_free_fn(blkg->pd[pol->plid]); |
1423 | blkg->pd[pol->plid] = NULL; | |
1424 | } | |
a2b1693b TH |
1425 | } |
1426 | ||
0d945c1f | 1427 | spin_unlock_irq(&q->queue_lock); |
bd166ef1 | 1428 | |
344e9ffc | 1429 | if (queue_is_mq(q)) |
bd166ef1 | 1430 | blk_mq_unfreeze_queue(q); |
a2b1693b TH |
1431 | } |
1432 | EXPORT_SYMBOL_GPL(blkcg_deactivate_policy); | |
1433 | ||
8bd435b3 | 1434 | /** |
3c798398 TH |
1435 | * blkcg_policy_register - register a blkcg policy |
1436 | * @pol: blkcg policy to register | |
8bd435b3 | 1437 | * |
3c798398 TH |
1438 | * Register @pol with blkcg core. Might sleep and @pol may be modified on |
1439 | * successful registration. Returns 0 on success and -errno on failure. | |
8bd435b3 | 1440 | */ |
d5bf0291 | 1441 | int blkcg_policy_register(struct blkcg_policy *pol) |
3e252066 | 1442 | { |
06b285bd | 1443 | struct blkcg *blkcg; |
8bd435b3 | 1444 | int i, ret; |
e8989fae | 1445 | |
838f13bf | 1446 | mutex_lock(&blkcg_pol_register_mutex); |
bc0d6501 TH |
1447 | mutex_lock(&blkcg_pol_mutex); |
1448 | ||
8bd435b3 TH |
1449 | /* find an empty slot */ |
1450 | ret = -ENOSPC; | |
1451 | for (i = 0; i < BLKCG_MAX_POLS; i++) | |
3c798398 | 1452 | if (!blkcg_policy[i]) |
8bd435b3 | 1453 | break; |
01c5f85a JA |
1454 | if (i >= BLKCG_MAX_POLS) { |
1455 | pr_warn("blkcg_policy_register: BLKCG_MAX_POLS too small\n"); | |
838f13bf | 1456 | goto err_unlock; |
01c5f85a | 1457 | } |
035d10b2 | 1458 | |
e8401073 | 1459 | /* Make sure cpd/pd_alloc_fn and cpd/pd_free_fn in pairs */ |
1460 | if ((!pol->cpd_alloc_fn ^ !pol->cpd_free_fn) || | |
1461 | (!pol->pd_alloc_fn ^ !pol->pd_free_fn)) | |
1462 | goto err_unlock; | |
1463 | ||
06b285bd | 1464 | /* register @pol */ |
3c798398 | 1465 | pol->plid = i; |
06b285bd TH |
1466 | blkcg_policy[pol->plid] = pol; |
1467 | ||
1468 | /* allocate and install cpd's */ | |
e4a9bde9 | 1469 | if (pol->cpd_alloc_fn) { |
06b285bd TH |
1470 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) { |
1471 | struct blkcg_policy_data *cpd; | |
1472 | ||
e4a9bde9 | 1473 | cpd = pol->cpd_alloc_fn(GFP_KERNEL); |
bbb427e3 | 1474 | if (!cpd) |
06b285bd | 1475 | goto err_free_cpds; |
06b285bd | 1476 | |
81437648 TH |
1477 | blkcg->cpd[pol->plid] = cpd; |
1478 | cpd->blkcg = blkcg; | |
06b285bd | 1479 | cpd->plid = pol->plid; |
86a5bba5 TH |
1480 | if (pol->cpd_init_fn) |
1481 | pol->cpd_init_fn(cpd); | |
06b285bd TH |
1482 | } |
1483 | } | |
1484 | ||
838f13bf | 1485 | mutex_unlock(&blkcg_pol_mutex); |
8bd435b3 | 1486 | |
8bd435b3 | 1487 | /* everything is in place, add intf files for the new policy */ |
2ee867dc TH |
1488 | if (pol->dfl_cftypes) |
1489 | WARN_ON(cgroup_add_dfl_cftypes(&io_cgrp_subsys, | |
1490 | pol->dfl_cftypes)); | |
880f50e2 | 1491 | if (pol->legacy_cftypes) |
c165b3e3 | 1492 | WARN_ON(cgroup_add_legacy_cftypes(&io_cgrp_subsys, |
880f50e2 | 1493 | pol->legacy_cftypes)); |
838f13bf TH |
1494 | mutex_unlock(&blkcg_pol_register_mutex); |
1495 | return 0; | |
1496 | ||
06b285bd | 1497 | err_free_cpds: |
58a9edce | 1498 | if (pol->cpd_free_fn) { |
06b285bd | 1499 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) { |
e4a9bde9 TH |
1500 | if (blkcg->cpd[pol->plid]) { |
1501 | pol->cpd_free_fn(blkcg->cpd[pol->plid]); | |
1502 | blkcg->cpd[pol->plid] = NULL; | |
1503 | } | |
06b285bd TH |
1504 | } |
1505 | } | |
1506 | blkcg_policy[pol->plid] = NULL; | |
838f13bf | 1507 | err_unlock: |
bc0d6501 | 1508 | mutex_unlock(&blkcg_pol_mutex); |
838f13bf | 1509 | mutex_unlock(&blkcg_pol_register_mutex); |
8bd435b3 | 1510 | return ret; |
3e252066 | 1511 | } |
3c798398 | 1512 | EXPORT_SYMBOL_GPL(blkcg_policy_register); |
3e252066 | 1513 | |
8bd435b3 | 1514 | /** |
3c798398 TH |
1515 | * blkcg_policy_unregister - unregister a blkcg policy |
1516 | * @pol: blkcg policy to unregister | |
8bd435b3 | 1517 | * |
3c798398 | 1518 | * Undo blkcg_policy_register(@pol). Might sleep. |
8bd435b3 | 1519 | */ |
3c798398 | 1520 | void blkcg_policy_unregister(struct blkcg_policy *pol) |
3e252066 | 1521 | { |
06b285bd TH |
1522 | struct blkcg *blkcg; |
1523 | ||
838f13bf | 1524 | mutex_lock(&blkcg_pol_register_mutex); |
bc0d6501 | 1525 | |
3c798398 | 1526 | if (WARN_ON(blkcg_policy[pol->plid] != pol)) |
8bd435b3 TH |
1527 | goto out_unlock; |
1528 | ||
1529 | /* kill the intf files first */ | |
2ee867dc TH |
1530 | if (pol->dfl_cftypes) |
1531 | cgroup_rm_cftypes(pol->dfl_cftypes); | |
880f50e2 TH |
1532 | if (pol->legacy_cftypes) |
1533 | cgroup_rm_cftypes(pol->legacy_cftypes); | |
44ea53de | 1534 | |
06b285bd | 1535 | /* remove cpds and unregister */ |
838f13bf | 1536 | mutex_lock(&blkcg_pol_mutex); |
06b285bd | 1537 | |
58a9edce | 1538 | if (pol->cpd_free_fn) { |
06b285bd | 1539 | list_for_each_entry(blkcg, &all_blkcgs, all_blkcgs_node) { |
e4a9bde9 TH |
1540 | if (blkcg->cpd[pol->plid]) { |
1541 | pol->cpd_free_fn(blkcg->cpd[pol->plid]); | |
1542 | blkcg->cpd[pol->plid] = NULL; | |
1543 | } | |
06b285bd TH |
1544 | } |
1545 | } | |
3c798398 | 1546 | blkcg_policy[pol->plid] = NULL; |
06b285bd | 1547 | |
bc0d6501 | 1548 | mutex_unlock(&blkcg_pol_mutex); |
838f13bf TH |
1549 | out_unlock: |
1550 | mutex_unlock(&blkcg_pol_register_mutex); | |
3e252066 | 1551 | } |
3c798398 | 1552 | EXPORT_SYMBOL_GPL(blkcg_policy_unregister); |
903d23f0 | 1553 | |
d3f77dfd TH |
1554 | bool __blkcg_punt_bio_submit(struct bio *bio) |
1555 | { | |
1556 | struct blkcg_gq *blkg = bio->bi_blkg; | |
1557 | ||
1558 | /* consume the flag first */ | |
1559 | bio->bi_opf &= ~REQ_CGROUP_PUNT; | |
1560 | ||
1561 | /* never bounce for the root cgroup */ | |
1562 | if (!blkg->parent) | |
1563 | return false; | |
1564 | ||
1565 | spin_lock_bh(&blkg->async_bio_lock); | |
1566 | bio_list_add(&blkg->async_bios, bio); | |
1567 | spin_unlock_bh(&blkg->async_bio_lock); | |
1568 | ||
1569 | queue_work(blkcg_punt_bio_wq, &blkg->async_bio_work); | |
1570 | return true; | |
1571 | } | |
1572 | ||
d09d8df3 JB |
1573 | /* |
1574 | * Scale the accumulated delay based on how long it has been since we updated | |
1575 | * the delay. We only call this when we are adding delay, in case it's been a | |
1576 | * while since we added delay, and when we are checking to see if we need to | |
1577 | * delay a task, to account for any delays that may have occurred. | |
1578 | */ | |
1579 | static void blkcg_scale_delay(struct blkcg_gq *blkg, u64 now) | |
1580 | { | |
1581 | u64 old = atomic64_read(&blkg->delay_start); | |
1582 | ||
54c52e10 TH |
1583 | /* negative use_delay means no scaling, see blkcg_set_delay() */ |
1584 | if (atomic_read(&blkg->use_delay) < 0) | |
1585 | return; | |
1586 | ||
d09d8df3 JB |
1587 | /* |
1588 | * We only want to scale down every second. The idea here is that we | |
1589 | * want to delay people for min(delay_nsec, NSEC_PER_SEC) in a certain | |
1590 | * time window. We only want to throttle tasks for recent delay that | |
1591 | * has occurred, in 1 second time windows since that's the maximum | |
1592 | * things can be throttled. We save the current delay window in | |
1593 | * blkg->last_delay so we know what amount is still left to be charged | |
1594 | * to the blkg from this point onward. blkg->last_use keeps track of | |
1595 | * the use_delay counter. The idea is if we're unthrottling the blkg we | |
1596 | * are ok with whatever is happening now, and we can take away more of | |
1597 | * the accumulated delay as we've already throttled enough that | |
1598 | * everybody is happy with their IO latencies. | |
1599 | */ | |
1600 | if (time_before64(old + NSEC_PER_SEC, now) && | |
1601 | atomic64_cmpxchg(&blkg->delay_start, old, now) == old) { | |
1602 | u64 cur = atomic64_read(&blkg->delay_nsec); | |
1603 | u64 sub = min_t(u64, blkg->last_delay, now - old); | |
1604 | int cur_use = atomic_read(&blkg->use_delay); | |
1605 | ||
1606 | /* | |
1607 | * We've been unthrottled, subtract a larger chunk of our | |
1608 | * accumulated delay. | |
1609 | */ | |
1610 | if (cur_use < blkg->last_use) | |
1611 | sub = max_t(u64, sub, blkg->last_delay >> 1); | |
1612 | ||
1613 | /* | |
1614 | * This shouldn't happen, but handle it anyway. Our delay_nsec | |
1615 | * should only ever be growing except here where we subtract out | |
1616 | * min(last_delay, 1 second), but lord knows bugs happen and I'd | |
1617 | * rather not end up with negative numbers. | |
1618 | */ | |
1619 | if (unlikely(cur < sub)) { | |
1620 | atomic64_set(&blkg->delay_nsec, 0); | |
1621 | blkg->last_delay = 0; | |
1622 | } else { | |
1623 | atomic64_sub(sub, &blkg->delay_nsec); | |
1624 | blkg->last_delay = cur - sub; | |
1625 | } | |
1626 | blkg->last_use = cur_use; | |
1627 | } | |
1628 | } | |
1629 | ||
1630 | /* | |
1631 | * This is called when we want to actually walk up the hierarchy and check to | |
1632 | * see if we need to throttle, and then actually throttle if there is some | |
1633 | * accumulated delay. This should only be called upon return to user space so | |
1634 | * we're not holding some lock that would induce a priority inversion. | |
1635 | */ | |
1636 | static void blkcg_maybe_throttle_blkg(struct blkcg_gq *blkg, bool use_memdelay) | |
1637 | { | |
fd112c74 | 1638 | unsigned long pflags; |
5160a5a5 | 1639 | bool clamp; |
d09d8df3 JB |
1640 | u64 now = ktime_to_ns(ktime_get()); |
1641 | u64 exp; | |
1642 | u64 delay_nsec = 0; | |
1643 | int tok; | |
1644 | ||
1645 | while (blkg->parent) { | |
5160a5a5 TH |
1646 | int use_delay = atomic_read(&blkg->use_delay); |
1647 | ||
1648 | if (use_delay) { | |
1649 | u64 this_delay; | |
1650 | ||
d09d8df3 | 1651 | blkcg_scale_delay(blkg, now); |
5160a5a5 TH |
1652 | this_delay = atomic64_read(&blkg->delay_nsec); |
1653 | if (this_delay > delay_nsec) { | |
1654 | delay_nsec = this_delay; | |
1655 | clamp = use_delay > 0; | |
1656 | } | |
d09d8df3 JB |
1657 | } |
1658 | blkg = blkg->parent; | |
1659 | } | |
1660 | ||
1661 | if (!delay_nsec) | |
1662 | return; | |
1663 | ||
1664 | /* | |
1665 | * Let's not sleep for all eternity if we've amassed a huge delay. | |
1666 | * Swapping or metadata IO can accumulate 10's of seconds worth of | |
1667 | * delay, and we want userspace to be able to do _something_ so cap the | |
5160a5a5 TH |
1668 | * delays at 0.25s. If there's 10's of seconds worth of delay then the |
1669 | * tasks will be delayed for 0.25 second for every syscall. If | |
1670 | * blkcg_set_delay() was used as indicated by negative use_delay, the | |
1671 | * caller is responsible for regulating the range. | |
d09d8df3 | 1672 | */ |
5160a5a5 TH |
1673 | if (clamp) |
1674 | delay_nsec = min_t(u64, delay_nsec, 250 * NSEC_PER_MSEC); | |
d09d8df3 | 1675 | |
fd112c74 JB |
1676 | if (use_memdelay) |
1677 | psi_memstall_enter(&pflags); | |
d09d8df3 JB |
1678 | |
1679 | exp = ktime_add_ns(now, delay_nsec); | |
1680 | tok = io_schedule_prepare(); | |
1681 | do { | |
1682 | __set_current_state(TASK_KILLABLE); | |
1683 | if (!schedule_hrtimeout(&exp, HRTIMER_MODE_ABS)) | |
1684 | break; | |
1685 | } while (!fatal_signal_pending(current)); | |
1686 | io_schedule_finish(tok); | |
fd112c74 JB |
1687 | |
1688 | if (use_memdelay) | |
1689 | psi_memstall_leave(&pflags); | |
d09d8df3 JB |
1690 | } |
1691 | ||
1692 | /** | |
1693 | * blkcg_maybe_throttle_current - throttle the current task if it has been marked | |
1694 | * | |
1695 | * This is only called if we've been marked with set_notify_resume(). Obviously | |
1696 | * we can be set_notify_resume() for reasons other than blkcg throttling, so we | |
1697 | * check to see if current->throttle_queue is set and if not this doesn't do | |
1698 | * anything. This should only ever be called by the resume code, it's not meant | |
1699 | * to be called by people willy-nilly as it will actually do the work to | |
1700 | * throttle the task if it is setup for throttling. | |
1701 | */ | |
1702 | void blkcg_maybe_throttle_current(void) | |
1703 | { | |
1704 | struct request_queue *q = current->throttle_queue; | |
1705 | struct cgroup_subsys_state *css; | |
1706 | struct blkcg *blkcg; | |
1707 | struct blkcg_gq *blkg; | |
1708 | bool use_memdelay = current->use_memdelay; | |
1709 | ||
1710 | if (!q) | |
1711 | return; | |
1712 | ||
1713 | current->throttle_queue = NULL; | |
1714 | current->use_memdelay = false; | |
1715 | ||
1716 | rcu_read_lock(); | |
1717 | css = kthread_blkcg(); | |
1718 | if (css) | |
1719 | blkcg = css_to_blkcg(css); | |
1720 | else | |
1721 | blkcg = css_to_blkcg(task_css(current, io_cgrp_id)); | |
1722 | ||
1723 | if (!blkcg) | |
1724 | goto out; | |
1725 | blkg = blkg_lookup(blkcg, q); | |
1726 | if (!blkg) | |
1727 | goto out; | |
7754f669 | 1728 | if (!blkg_tryget(blkg)) |
d09d8df3 JB |
1729 | goto out; |
1730 | rcu_read_unlock(); | |
d09d8df3 JB |
1731 | |
1732 | blkcg_maybe_throttle_blkg(blkg, use_memdelay); | |
1733 | blkg_put(blkg); | |
cc7ecc25 | 1734 | blk_put_queue(q); |
d09d8df3 JB |
1735 | return; |
1736 | out: | |
1737 | rcu_read_unlock(); | |
1738 | blk_put_queue(q); | |
1739 | } | |
d09d8df3 JB |
1740 | |
1741 | /** | |
1742 | * blkcg_schedule_throttle - this task needs to check for throttling | |
537d71b3 BVA |
1743 | * @q: the request queue IO was submitted on |
1744 | * @use_memdelay: do we charge this to memory delay for PSI | |
d09d8df3 JB |
1745 | * |
1746 | * This is called by the IO controller when we know there's delay accumulated | |
1747 | * for the blkg for this task. We do not pass the blkg because there are places | |
1748 | * we call this that may not have that information, the swapping code for | |
1749 | * instance will only have a request_queue at that point. This set's the | |
1750 | * notify_resume for the task to check and see if it requires throttling before | |
1751 | * returning to user space. | |
1752 | * | |
1753 | * We will only schedule once per syscall. You can call this over and over | |
1754 | * again and it will only do the check once upon return to user space, and only | |
1755 | * throttle once. If the task needs to be throttled again it'll need to be | |
1756 | * re-set at the next time we see the task. | |
1757 | */ | |
1758 | void blkcg_schedule_throttle(struct request_queue *q, bool use_memdelay) | |
1759 | { | |
1760 | if (unlikely(current->flags & PF_KTHREAD)) | |
1761 | return; | |
1762 | ||
49d1822b CX |
1763 | if (current->throttle_queue != q) { |
1764 | if (!blk_get_queue(q)) | |
1765 | return; | |
1766 | ||
1767 | if (current->throttle_queue) | |
1768 | blk_put_queue(current->throttle_queue); | |
1769 | current->throttle_queue = q; | |
1770 | } | |
d09d8df3 | 1771 | |
d09d8df3 JB |
1772 | if (use_memdelay) |
1773 | current->use_memdelay = use_memdelay; | |
1774 | set_notify_resume(current); | |
1775 | } | |
d09d8df3 JB |
1776 | |
1777 | /** | |
1778 | * blkcg_add_delay - add delay to this blkg | |
537d71b3 BVA |
1779 | * @blkg: blkg of interest |
1780 | * @now: the current time in nanoseconds | |
1781 | * @delta: how many nanoseconds of delay to add | |
d09d8df3 JB |
1782 | * |
1783 | * Charge @delta to the blkg's current delay accumulation. This is used to | |
1784 | * throttle tasks if an IO controller thinks we need more throttling. | |
1785 | */ | |
1786 | void blkcg_add_delay(struct blkcg_gq *blkg, u64 now, u64 delta) | |
1787 | { | |
54c52e10 TH |
1788 | if (WARN_ON_ONCE(atomic_read(&blkg->use_delay) < 0)) |
1789 | return; | |
d09d8df3 JB |
1790 | blkcg_scale_delay(blkg, now); |
1791 | atomic64_add(delta, &blkg->delay_nsec); | |
1792 | } | |
d09d8df3 | 1793 | |
28fc591f CH |
1794 | /** |
1795 | * blkg_tryget_closest - try and get a blkg ref on the closet blkg | |
13c7863d CH |
1796 | * @bio: target bio |
1797 | * @css: target css | |
28fc591f | 1798 | * |
13c7863d CH |
1799 | * As the failure mode here is to walk up the blkg tree, this ensure that the |
1800 | * blkg->parent pointers are always valid. This returns the blkg that it ended | |
1801 | * up taking a reference on or %NULL if no reference was taken. | |
28fc591f | 1802 | */ |
13c7863d CH |
1803 | static inline struct blkcg_gq *blkg_tryget_closest(struct bio *bio, |
1804 | struct cgroup_subsys_state *css) | |
28fc591f | 1805 | { |
13c7863d | 1806 | struct blkcg_gq *blkg, *ret_blkg = NULL; |
28fc591f | 1807 | |
13c7863d | 1808 | rcu_read_lock(); |
309dca30 CH |
1809 | blkg = blkg_lookup_create(css_to_blkcg(css), |
1810 | bio->bi_bdev->bd_disk->queue); | |
28fc591f CH |
1811 | while (blkg) { |
1812 | if (blkg_tryget(blkg)) { | |
1813 | ret_blkg = blkg; | |
1814 | break; | |
1815 | } | |
1816 | blkg = blkg->parent; | |
1817 | } | |
13c7863d | 1818 | rcu_read_unlock(); |
28fc591f CH |
1819 | |
1820 | return ret_blkg; | |
1821 | } | |
1822 | ||
1823 | /** | |
1824 | * bio_associate_blkg_from_css - associate a bio with a specified css | |
1825 | * @bio: target bio | |
1826 | * @css: target css | |
1827 | * | |
1828 | * Associate @bio with the blkg found by combining the css's blkg and the | |
1829 | * request_queue of the @bio. An association failure is handled by walking up | |
1830 | * the blkg tree. Therefore, the blkg associated can be anything between @blkg | |
1831 | * and q->root_blkg. This situation only happens when a cgroup is dying and | |
1832 | * then the remaining bios will spill to the closest alive blkg. | |
1833 | * | |
1834 | * A reference will be taken on the blkg and will be released when @bio is | |
1835 | * freed. | |
1836 | */ | |
1837 | void bio_associate_blkg_from_css(struct bio *bio, | |
1838 | struct cgroup_subsys_state *css) | |
1839 | { | |
28fc591f CH |
1840 | if (bio->bi_blkg) |
1841 | blkg_put(bio->bi_blkg); | |
1842 | ||
a5b97526 | 1843 | if (css && css->parent) { |
13c7863d | 1844 | bio->bi_blkg = blkg_tryget_closest(bio, css); |
a5b97526 | 1845 | } else { |
309dca30 CH |
1846 | blkg_get(bio->bi_bdev->bd_disk->queue->root_blkg); |
1847 | bio->bi_blkg = bio->bi_bdev->bd_disk->queue->root_blkg; | |
a5b97526 | 1848 | } |
28fc591f CH |
1849 | } |
1850 | EXPORT_SYMBOL_GPL(bio_associate_blkg_from_css); | |
1851 | ||
1852 | /** | |
1853 | * bio_associate_blkg - associate a bio with a blkg | |
1854 | * @bio: target bio | |
1855 | * | |
1856 | * Associate @bio with the blkg found from the bio's css and request_queue. | |
1857 | * If one is not found, bio_lookup_blkg() creates the blkg. If a blkg is | |
1858 | * already associated, the css is reused and association redone as the | |
1859 | * request_queue may have changed. | |
1860 | */ | |
1861 | void bio_associate_blkg(struct bio *bio) | |
1862 | { | |
1863 | struct cgroup_subsys_state *css; | |
1864 | ||
1865 | rcu_read_lock(); | |
1866 | ||
1867 | if (bio->bi_blkg) | |
1868 | css = &bio_blkcg(bio)->css; | |
1869 | else | |
1870 | css = blkcg_css(); | |
1871 | ||
1872 | bio_associate_blkg_from_css(bio, css); | |
1873 | ||
1874 | rcu_read_unlock(); | |
1875 | } | |
1876 | EXPORT_SYMBOL_GPL(bio_associate_blkg); | |
1877 | ||
1878 | /** | |
1879 | * bio_clone_blkg_association - clone blkg association from src to dst bio | |
1880 | * @dst: destination bio | |
1881 | * @src: source bio | |
1882 | */ | |
1883 | void bio_clone_blkg_association(struct bio *dst, struct bio *src) | |
1884 | { | |
1885 | if (src->bi_blkg) { | |
1886 | if (dst->bi_blkg) | |
1887 | blkg_put(dst->bi_blkg); | |
1888 | blkg_get(src->bi_blkg); | |
1889 | dst->bi_blkg = src->bi_blkg; | |
1890 | } | |
1891 | } | |
1892 | EXPORT_SYMBOL_GPL(bio_clone_blkg_association); | |
1893 | ||
db18a53e CH |
1894 | static int blk_cgroup_io_type(struct bio *bio) |
1895 | { | |
1896 | if (op_is_discard(bio->bi_opf)) | |
1897 | return BLKG_IOSTAT_DISCARD; | |
1898 | if (op_is_write(bio->bi_opf)) | |
1899 | return BLKG_IOSTAT_WRITE; | |
1900 | return BLKG_IOSTAT_READ; | |
1901 | } | |
1902 | ||
1903 | void blk_cgroup_bio_start(struct bio *bio) | |
1904 | { | |
1905 | int rwd = blk_cgroup_io_type(bio), cpu; | |
1906 | struct blkg_iostat_set *bis; | |
1907 | ||
1908 | cpu = get_cpu(); | |
1909 | bis = per_cpu_ptr(bio->bi_blkg->iostat_cpu, cpu); | |
1910 | u64_stats_update_begin(&bis->sync); | |
1911 | ||
1912 | /* | |
1913 | * If the bio is flagged with BIO_CGROUP_ACCT it means this is a split | |
1914 | * bio and we would have already accounted for the size of the bio. | |
1915 | */ | |
1916 | if (!bio_flagged(bio, BIO_CGROUP_ACCT)) { | |
1917 | bio_set_flag(bio, BIO_CGROUP_ACCT); | |
0b8cc25d | 1918 | bis->cur.bytes[rwd] += bio->bi_iter.bi_size; |
db18a53e CH |
1919 | } |
1920 | bis->cur.ios[rwd]++; | |
1921 | ||
1922 | u64_stats_update_end(&bis->sync); | |
1923 | if (cgroup_subsys_on_dfl(io_cgrp_subsys)) | |
1924 | cgroup_rstat_updated(bio->bi_blkg->blkcg->css.cgroup, cpu); | |
1925 | put_cpu(); | |
1926 | } | |
1927 | ||
d3f77dfd TH |
1928 | static int __init blkcg_init(void) |
1929 | { | |
1930 | blkcg_punt_bio_wq = alloc_workqueue("blkcg_punt_bio", | |
1931 | WQ_MEM_RECLAIM | WQ_FREEZABLE | | |
1932 | WQ_UNBOUND | WQ_SYSFS, 0); | |
1933 | if (!blkcg_punt_bio_wq) | |
1934 | return -ENOMEM; | |
1935 | return 0; | |
1936 | } | |
1937 | subsys_initcall(blkcg_init); | |
1938 | ||
903d23f0 JB |
1939 | module_param(blkcg_debug_stats, bool, 0644); |
1940 | MODULE_PARM_DESC(blkcg_debug_stats, "True if you want debug stats, false if not"); |