Commit | Line | Data |
---|---|---|
b2441318 | 1 | /* SPDX-License-Identifier: GPL-2.0 */ |
31e4c28d VG |
2 | #ifndef _BLK_CGROUP_H |
3 | #define _BLK_CGROUP_H | |
4 | /* | |
5 | * Common Block IO controller cgroup interface | |
6 | * | |
7 | * Based on ideas and code from CFQ, CFS and BFQ: | |
8 | * Copyright (C) 2003 Jens Axboe <axboe@kernel.dk> | |
9 | * | |
10 | * Copyright (C) 2008 Fabio Checconi <fabio@gandalf.sssup.it> | |
11 | * Paolo Valente <paolo.valente@unimore.it> | |
12 | * | |
13 | * Copyright (C) 2009 Vivek Goyal <vgoyal@redhat.com> | |
14 | * Nauman Rafique <nauman@google.com> | |
15 | */ | |
16 | ||
17 | #include <linux/cgroup.h> | |
24bdb8ef | 18 | #include <linux/percpu_counter.h> |
829fdb50 | 19 | #include <linux/seq_file.h> |
a637120e | 20 | #include <linux/radix-tree.h> |
a051661c | 21 | #include <linux/blkdev.h> |
a5049a8a | 22 | #include <linux/atomic.h> |
902ec5b6 | 23 | #include <linux/kthread.h> |
31e4c28d | 24 | |
24bdb8ef TH |
25 | /* percpu_counter batch for blkg_[rw]stats, per-cpu drift doesn't matter */ |
26 | #define BLKG_STAT_CPU_BATCH (INT_MAX / 2) | |
27 | ||
9355aede VG |
28 | /* Max limits for throttle policy */ |
29 | #define THROTL_IOPS_MAX UINT_MAX | |
30 | ||
f48ec1d7 TH |
31 | #ifdef CONFIG_BLK_CGROUP |
32 | ||
edcb0722 TH |
33 | enum blkg_rwstat_type { |
34 | BLKG_RWSTAT_READ, | |
35 | BLKG_RWSTAT_WRITE, | |
36 | BLKG_RWSTAT_SYNC, | |
37 | BLKG_RWSTAT_ASYNC, | |
636620b6 | 38 | BLKG_RWSTAT_DISCARD, |
edcb0722 TH |
39 | |
40 | BLKG_RWSTAT_NR, | |
41 | BLKG_RWSTAT_TOTAL = BLKG_RWSTAT_NR, | |
303a3acb DS |
42 | }; |
43 | ||
a637120e TH |
44 | struct blkcg_gq; |
45 | ||
3c798398 | 46 | struct blkcg { |
36558c8a TH |
47 | struct cgroup_subsys_state css; |
48 | spinlock_t lock; | |
a637120e TH |
49 | |
50 | struct radix_tree_root blkg_tree; | |
55679c8d | 51 | struct blkcg_gq __rcu *blkg_hint; |
36558c8a | 52 | struct hlist_head blkg_list; |
9a9e8a26 | 53 | |
81437648 | 54 | struct blkcg_policy_data *cpd[BLKCG_MAX_POLS]; |
52ebea74 | 55 | |
7876f930 | 56 | struct list_head all_blkcgs_node; |
52ebea74 TH |
57 | #ifdef CONFIG_CGROUP_WRITEBACK |
58 | struct list_head cgwb_list; | |
59 | #endif | |
31e4c28d VG |
60 | }; |
61 | ||
e6269c44 TH |
62 | /* |
63 | * blkg_[rw]stat->aux_cnt is excluded for local stats but included for | |
24bdb8ef TH |
64 | * recursive. Used to carry stats of dead children, and, for blkg_rwstat, |
65 | * to carry result values from read and sum operations. | |
e6269c44 | 66 | */ |
edcb0722 | 67 | struct blkg_stat { |
24bdb8ef | 68 | struct percpu_counter cpu_cnt; |
e6269c44 | 69 | atomic64_t aux_cnt; |
edcb0722 TH |
70 | }; |
71 | ||
72 | struct blkg_rwstat { | |
24bdb8ef | 73 | struct percpu_counter cpu_cnt[BLKG_RWSTAT_NR]; |
e6269c44 | 74 | atomic64_t aux_cnt[BLKG_RWSTAT_NR]; |
edcb0722 TH |
75 | }; |
76 | ||
f95a04af TH |
77 | /* |
78 | * A blkcg_gq (blkg) is association between a block cgroup (blkcg) and a | |
79 | * request_queue (q). This is used by blkcg policies which need to track | |
80 | * information per blkcg - q pair. | |
81 | * | |
001bea73 TH |
82 | * There can be multiple active blkcg policies and each blkg:policy pair is |
83 | * represented by a blkg_policy_data which is allocated and freed by each | |
84 | * policy's pd_alloc/free_fn() methods. A policy can allocate private data | |
85 | * area by allocating larger data structure which embeds blkg_policy_data | |
86 | * at the beginning. | |
f95a04af | 87 | */ |
0381411e | 88 | struct blkg_policy_data { |
b276a876 | 89 | /* the blkg and policy id this per-policy data belongs to */ |
3c798398 | 90 | struct blkcg_gq *blkg; |
b276a876 | 91 | int plid; |
4c699480 | 92 | bool offline; |
0381411e TH |
93 | }; |
94 | ||
e48453c3 | 95 | /* |
e4a9bde9 TH |
96 | * Policies that need to keep per-blkcg data which is independent from any |
97 | * request_queue associated to it should implement cpd_alloc/free_fn() | |
98 | * methods. A policy can allocate private data area by allocating larger | |
99 | * data structure which embeds blkcg_policy_data at the beginning. | |
100 | * cpd_init() is invoked to let each policy handle per-blkcg data. | |
e48453c3 AA |
101 | */ |
102 | struct blkcg_policy_data { | |
81437648 TH |
103 | /* the blkcg and policy id this per-policy data belongs to */ |
104 | struct blkcg *blkcg; | |
e48453c3 | 105 | int plid; |
e48453c3 AA |
106 | }; |
107 | ||
3c798398 TH |
108 | /* association between a blk cgroup and a request queue */ |
109 | struct blkcg_gq { | |
c875f4d0 | 110 | /* Pointer to the associated request_queue */ |
36558c8a TH |
111 | struct request_queue *q; |
112 | struct list_head q_node; | |
113 | struct hlist_node blkcg_node; | |
3c798398 | 114 | struct blkcg *blkcg; |
3c547865 | 115 | |
ce7acfea TH |
116 | /* |
117 | * Each blkg gets congested separately and the congestion state is | |
118 | * propagated to the matching bdi_writeback_congested. | |
119 | */ | |
120 | struct bdi_writeback_congested *wb_congested; | |
121 | ||
3c547865 TH |
122 | /* all non-root blkcg_gq's are guaranteed to have access to parent */ |
123 | struct blkcg_gq *parent; | |
124 | ||
a051661c TH |
125 | /* request allocation list for this blkcg-q pair */ |
126 | struct request_list rl; | |
3c547865 | 127 | |
1adaf3dd | 128 | /* reference count */ |
a5049a8a | 129 | atomic_t refcnt; |
22084190 | 130 | |
f427d909 TH |
131 | /* is this blkg online? protected by both blkcg and q locks */ |
132 | bool online; | |
133 | ||
77ea7338 TH |
134 | struct blkg_rwstat stat_bytes; |
135 | struct blkg_rwstat stat_ios; | |
136 | ||
36558c8a | 137 | struct blkg_policy_data *pd[BLKCG_MAX_POLS]; |
1adaf3dd | 138 | |
36558c8a | 139 | struct rcu_head rcu_head; |
d09d8df3 JB |
140 | |
141 | atomic_t use_delay; | |
142 | atomic64_t delay_nsec; | |
143 | atomic64_t delay_start; | |
144 | u64 last_delay; | |
145 | int last_use; | |
31e4c28d VG |
146 | }; |
147 | ||
e4a9bde9 | 148 | typedef struct blkcg_policy_data *(blkcg_pol_alloc_cpd_fn)(gfp_t gfp); |
81437648 | 149 | typedef void (blkcg_pol_init_cpd_fn)(struct blkcg_policy_data *cpd); |
e4a9bde9 | 150 | typedef void (blkcg_pol_free_cpd_fn)(struct blkcg_policy_data *cpd); |
69d7fde5 | 151 | typedef void (blkcg_pol_bind_cpd_fn)(struct blkcg_policy_data *cpd); |
001bea73 | 152 | typedef struct blkg_policy_data *(blkcg_pol_alloc_pd_fn)(gfp_t gfp, int node); |
a9520cd6 TH |
153 | typedef void (blkcg_pol_init_pd_fn)(struct blkg_policy_data *pd); |
154 | typedef void (blkcg_pol_online_pd_fn)(struct blkg_policy_data *pd); | |
155 | typedef void (blkcg_pol_offline_pd_fn)(struct blkg_policy_data *pd); | |
001bea73 | 156 | typedef void (blkcg_pol_free_pd_fn)(struct blkg_policy_data *pd); |
a9520cd6 | 157 | typedef void (blkcg_pol_reset_pd_stats_fn)(struct blkg_policy_data *pd); |
903d23f0 JB |
158 | typedef size_t (blkcg_pol_stat_pd_fn)(struct blkg_policy_data *pd, char *buf, |
159 | size_t size); | |
3e252066 | 160 | |
3c798398 | 161 | struct blkcg_policy { |
36558c8a | 162 | int plid; |
36558c8a | 163 | /* cgroup files for the policy */ |
2ee867dc | 164 | struct cftype *dfl_cftypes; |
880f50e2 | 165 | struct cftype *legacy_cftypes; |
f9fcc2d3 TH |
166 | |
167 | /* operations */ | |
e4a9bde9 | 168 | blkcg_pol_alloc_cpd_fn *cpd_alloc_fn; |
e48453c3 | 169 | blkcg_pol_init_cpd_fn *cpd_init_fn; |
e4a9bde9 | 170 | blkcg_pol_free_cpd_fn *cpd_free_fn; |
69d7fde5 | 171 | blkcg_pol_bind_cpd_fn *cpd_bind_fn; |
e4a9bde9 | 172 | |
001bea73 | 173 | blkcg_pol_alloc_pd_fn *pd_alloc_fn; |
f9fcc2d3 | 174 | blkcg_pol_init_pd_fn *pd_init_fn; |
f427d909 TH |
175 | blkcg_pol_online_pd_fn *pd_online_fn; |
176 | blkcg_pol_offline_pd_fn *pd_offline_fn; | |
001bea73 | 177 | blkcg_pol_free_pd_fn *pd_free_fn; |
f9fcc2d3 | 178 | blkcg_pol_reset_pd_stats_fn *pd_reset_stats_fn; |
903d23f0 | 179 | blkcg_pol_stat_pd_fn *pd_stat_fn; |
3e252066 VG |
180 | }; |
181 | ||
3c798398 | 182 | extern struct blkcg blkcg_root; |
496d5e75 | 183 | extern struct cgroup_subsys_state * const blkcg_root_css; |
36558c8a | 184 | |
24f29046 TH |
185 | struct blkcg_gq *blkg_lookup_slowpath(struct blkcg *blkcg, |
186 | struct request_queue *q, bool update_hint); | |
3c798398 TH |
187 | struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg, |
188 | struct request_queue *q); | |
36558c8a TH |
189 | int blkcg_init_queue(struct request_queue *q); |
190 | void blkcg_drain_queue(struct request_queue *q); | |
191 | void blkcg_exit_queue(struct request_queue *q); | |
5efd6113 | 192 | |
3e252066 | 193 | /* Blkio controller policy registration */ |
d5bf0291 | 194 | int blkcg_policy_register(struct blkcg_policy *pol); |
3c798398 | 195 | void blkcg_policy_unregister(struct blkcg_policy *pol); |
36558c8a | 196 | int blkcg_activate_policy(struct request_queue *q, |
3c798398 | 197 | const struct blkcg_policy *pol); |
36558c8a | 198 | void blkcg_deactivate_policy(struct request_queue *q, |
3c798398 | 199 | const struct blkcg_policy *pol); |
3e252066 | 200 | |
dd165eb3 | 201 | const char *blkg_dev_name(struct blkcg_gq *blkg); |
3c798398 | 202 | void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg, |
f95a04af TH |
203 | u64 (*prfill)(struct seq_file *, |
204 | struct blkg_policy_data *, int), | |
3c798398 | 205 | const struct blkcg_policy *pol, int data, |
ec399347 | 206 | bool show_total); |
f95a04af TH |
207 | u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v); |
208 | u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd, | |
829fdb50 | 209 | const struct blkg_rwstat *rwstat); |
f95a04af TH |
210 | u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off); |
211 | u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd, | |
212 | int off); | |
77ea7338 TH |
213 | int blkg_print_stat_bytes(struct seq_file *sf, void *v); |
214 | int blkg_print_stat_ios(struct seq_file *sf, void *v); | |
215 | int blkg_print_stat_bytes_recursive(struct seq_file *sf, void *v); | |
216 | int blkg_print_stat_ios_recursive(struct seq_file *sf, void *v); | |
829fdb50 | 217 | |
f12c74ca TH |
218 | u64 blkg_stat_recursive_sum(struct blkcg_gq *blkg, |
219 | struct blkcg_policy *pol, int off); | |
220 | struct blkg_rwstat blkg_rwstat_recursive_sum(struct blkcg_gq *blkg, | |
221 | struct blkcg_policy *pol, int off); | |
16b3de66 | 222 | |
829fdb50 | 223 | struct blkg_conf_ctx { |
36558c8a | 224 | struct gendisk *disk; |
3c798398 | 225 | struct blkcg_gq *blkg; |
36aa9e5f | 226 | char *body; |
829fdb50 TH |
227 | }; |
228 | ||
3c798398 | 229 | int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol, |
36aa9e5f | 230 | char *input, struct blkg_conf_ctx *ctx); |
829fdb50 TH |
231 | void blkg_conf_finish(struct blkg_conf_ctx *ctx); |
232 | ||
233 | ||
a7c6d554 TH |
234 | static inline struct blkcg *css_to_blkcg(struct cgroup_subsys_state *css) |
235 | { | |
236 | return css ? container_of(css, struct blkcg, css) : NULL; | |
237 | } | |
238 | ||
b1208b56 TH |
239 | static inline struct blkcg *bio_blkcg(struct bio *bio) |
240 | { | |
902ec5b6 SL |
241 | struct cgroup_subsys_state *css; |
242 | ||
b1208b56 | 243 | if (bio && bio->bi_css) |
a7c6d554 | 244 | return css_to_blkcg(bio->bi_css); |
902ec5b6 SL |
245 | css = kthread_blkcg(); |
246 | if (css) | |
247 | return css_to_blkcg(css); | |
248 | return css_to_blkcg(task_css(current, io_cgrp_id)); | |
fd383c2d TH |
249 | } |
250 | ||
d09d8df3 JB |
251 | static inline bool blk_cgroup_congested(void) |
252 | { | |
253 | struct cgroup_subsys_state *css; | |
254 | bool ret = false; | |
255 | ||
256 | rcu_read_lock(); | |
257 | css = kthread_blkcg(); | |
258 | if (!css) | |
259 | css = task_css(current, io_cgrp_id); | |
260 | while (css) { | |
261 | if (atomic_read(&css->cgroup->congestion_count)) { | |
262 | ret = true; | |
263 | break; | |
264 | } | |
265 | css = css->parent; | |
266 | } | |
267 | rcu_read_unlock(); | |
268 | return ret; | |
269 | } | |
270 | ||
c7c98fd3 JB |
271 | /** |
272 | * bio_issue_as_root_blkg - see if this bio needs to be issued as root blkg | |
273 | * @return: true if this bio needs to be submitted with the root blkg context. | |
274 | * | |
275 | * In order to avoid priority inversions we sometimes need to issue a bio as if | |
276 | * it were attached to the root blkg, and then backcharge to the actual owning | |
277 | * blkg. The idea is we do bio_blkcg() to look up the actual context for the | |
278 | * bio and attach the appropriate blkg to the bio. Then we call this helper and | |
279 | * if it is true run with the root blkg for that queue and then do any | |
280 | * backcharging to the originating cgroup once the io is complete. | |
281 | */ | |
282 | static inline bool bio_issue_as_root_blkg(struct bio *bio) | |
283 | { | |
0d1e0c7c | 284 | return (bio->bi_opf & (REQ_META | REQ_SWAP)) != 0; |
c7c98fd3 JB |
285 | } |
286 | ||
3c547865 TH |
287 | /** |
288 | * blkcg_parent - get the parent of a blkcg | |
289 | * @blkcg: blkcg of interest | |
290 | * | |
291 | * Return the parent blkcg of @blkcg. Can be called anytime. | |
292 | */ | |
293 | static inline struct blkcg *blkcg_parent(struct blkcg *blkcg) | |
294 | { | |
5c9d535b | 295 | return css_to_blkcg(blkcg->css.parent); |
3c547865 TH |
296 | } |
297 | ||
24f29046 TH |
298 | /** |
299 | * __blkg_lookup - internal version of blkg_lookup() | |
300 | * @blkcg: blkcg of interest | |
301 | * @q: request_queue of interest | |
302 | * @update_hint: whether to update lookup hint with the result or not | |
303 | * | |
304 | * This is internal version and shouldn't be used by policy | |
305 | * implementations. Looks up blkgs for the @blkcg - @q pair regardless of | |
306 | * @q's bypass state. If @update_hint is %true, the caller should be | |
307 | * holding @q->queue_lock and lookup hint is updated on success. | |
308 | */ | |
309 | static inline struct blkcg_gq *__blkg_lookup(struct blkcg *blkcg, | |
310 | struct request_queue *q, | |
311 | bool update_hint) | |
312 | { | |
313 | struct blkcg_gq *blkg; | |
314 | ||
85b6bc9d TH |
315 | if (blkcg == &blkcg_root) |
316 | return q->root_blkg; | |
317 | ||
24f29046 TH |
318 | blkg = rcu_dereference(blkcg->blkg_hint); |
319 | if (blkg && blkg->q == q) | |
320 | return blkg; | |
321 | ||
322 | return blkg_lookup_slowpath(blkcg, q, update_hint); | |
323 | } | |
324 | ||
325 | /** | |
326 | * blkg_lookup - lookup blkg for the specified blkcg - q pair | |
327 | * @blkcg: blkcg of interest | |
328 | * @q: request_queue of interest | |
329 | * | |
330 | * Lookup blkg for the @blkcg - @q pair. This function should be called | |
331 | * under RCU read lock and is guaranteed to return %NULL if @q is bypassing | |
332 | * - see blk_queue_bypass_start() for details. | |
333 | */ | |
334 | static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, | |
335 | struct request_queue *q) | |
336 | { | |
337 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
338 | ||
339 | if (unlikely(blk_queue_bypass(q))) | |
340 | return NULL; | |
341 | return __blkg_lookup(blkcg, q, false); | |
342 | } | |
343 | ||
6bad9b21 | 344 | /** |
b86d865c | 345 | * blk_queue_root_blkg - return blkg for the (blkcg_root, @q) pair |
6bad9b21 BVA |
346 | * @q: request_queue of interest |
347 | * | |
348 | * Lookup blkg for @q at the root level. See also blkg_lookup(). | |
349 | */ | |
b86d865c | 350 | static inline struct blkcg_gq *blk_queue_root_blkg(struct request_queue *q) |
6bad9b21 | 351 | { |
b86d865c | 352 | return q->root_blkg; |
6bad9b21 BVA |
353 | } |
354 | ||
0381411e TH |
355 | /** |
356 | * blkg_to_pdata - get policy private data | |
357 | * @blkg: blkg of interest | |
358 | * @pol: policy of interest | |
359 | * | |
360 | * Return pointer to private data associated with the @blkg-@pol pair. | |
361 | */ | |
f95a04af TH |
362 | static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg, |
363 | struct blkcg_policy *pol) | |
0381411e | 364 | { |
f95a04af | 365 | return blkg ? blkg->pd[pol->plid] : NULL; |
0381411e TH |
366 | } |
367 | ||
e48453c3 AA |
368 | static inline struct blkcg_policy_data *blkcg_to_cpd(struct blkcg *blkcg, |
369 | struct blkcg_policy *pol) | |
370 | { | |
81437648 | 371 | return blkcg ? blkcg->cpd[pol->plid] : NULL; |
e48453c3 AA |
372 | } |
373 | ||
0381411e TH |
374 | /** |
375 | * pdata_to_blkg - get blkg associated with policy private data | |
f95a04af | 376 | * @pd: policy private data of interest |
0381411e | 377 | * |
f95a04af | 378 | * @pd is policy private data. Determine the blkg it's associated with. |
0381411e | 379 | */ |
f95a04af | 380 | static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd) |
0381411e | 381 | { |
f95a04af | 382 | return pd ? pd->blkg : NULL; |
0381411e TH |
383 | } |
384 | ||
81437648 TH |
385 | static inline struct blkcg *cpd_to_blkcg(struct blkcg_policy_data *cpd) |
386 | { | |
387 | return cpd ? cpd->blkcg : NULL; | |
388 | } | |
389 | ||
54e7ed12 TH |
390 | /** |
391 | * blkg_path - format cgroup path of blkg | |
392 | * @blkg: blkg of interest | |
393 | * @buf: target buffer | |
394 | * @buflen: target buffer length | |
395 | * | |
396 | * Format the path of the cgroup of @blkg into @buf. | |
397 | */ | |
3c798398 | 398 | static inline int blkg_path(struct blkcg_gq *blkg, char *buf, int buflen) |
afc24d49 | 399 | { |
4c737b41 | 400 | return cgroup_path(blkg->blkcg->css.cgroup, buf, buflen); |
afc24d49 VG |
401 | } |
402 | ||
1adaf3dd TH |
403 | /** |
404 | * blkg_get - get a blkg reference | |
405 | * @blkg: blkg to get | |
406 | * | |
a5049a8a | 407 | * The caller should be holding an existing reference. |
1adaf3dd | 408 | */ |
3c798398 | 409 | static inline void blkg_get(struct blkcg_gq *blkg) |
1adaf3dd | 410 | { |
a5049a8a TH |
411 | WARN_ON_ONCE(atomic_read(&blkg->refcnt) <= 0); |
412 | atomic_inc(&blkg->refcnt); | |
1adaf3dd TH |
413 | } |
414 | ||
d09d8df3 JB |
415 | /** |
416 | * blkg_try_get - try and get a blkg reference | |
417 | * @blkg: blkg to get | |
418 | * | |
419 | * This is for use when doing an RCU lookup of the blkg. We may be in the midst | |
420 | * of freeing this blkg, so we can only use it if the refcnt is not zero. | |
421 | */ | |
422 | static inline struct blkcg_gq *blkg_try_get(struct blkcg_gq *blkg) | |
423 | { | |
424 | if (atomic_inc_not_zero(&blkg->refcnt)) | |
425 | return blkg; | |
426 | return NULL; | |
427 | } | |
428 | ||
429 | ||
2a4fd070 | 430 | void __blkg_release_rcu(struct rcu_head *rcu); |
1adaf3dd TH |
431 | |
432 | /** | |
433 | * blkg_put - put a blkg reference | |
434 | * @blkg: blkg to put | |
1adaf3dd | 435 | */ |
3c798398 | 436 | static inline void blkg_put(struct blkcg_gq *blkg) |
1adaf3dd | 437 | { |
a5049a8a TH |
438 | WARN_ON_ONCE(atomic_read(&blkg->refcnt) <= 0); |
439 | if (atomic_dec_and_test(&blkg->refcnt)) | |
2a4fd070 | 440 | call_rcu(&blkg->rcu_head, __blkg_release_rcu); |
1adaf3dd TH |
441 | } |
442 | ||
dd4a4ffc TH |
443 | /** |
444 | * blkg_for_each_descendant_pre - pre-order walk of a blkg's descendants | |
445 | * @d_blkg: loop cursor pointing to the current descendant | |
492eb21b | 446 | * @pos_css: used for iteration |
dd4a4ffc TH |
447 | * @p_blkg: target blkg to walk descendants of |
448 | * | |
449 | * Walk @c_blkg through the descendants of @p_blkg. Must be used with RCU | |
450 | * read locked. If called under either blkcg or queue lock, the iteration | |
451 | * is guaranteed to include all and only online blkgs. The caller may | |
492eb21b | 452 | * update @pos_css by calling css_rightmost_descendant() to skip subtree. |
bd8815a6 | 453 | * @p_blkg is included in the iteration and the first node to be visited. |
dd4a4ffc | 454 | */ |
492eb21b TH |
455 | #define blkg_for_each_descendant_pre(d_blkg, pos_css, p_blkg) \ |
456 | css_for_each_descendant_pre((pos_css), &(p_blkg)->blkcg->css) \ | |
457 | if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \ | |
dd4a4ffc TH |
458 | (p_blkg)->q, false))) |
459 | ||
aa539cb3 TH |
460 | /** |
461 | * blkg_for_each_descendant_post - post-order walk of a blkg's descendants | |
462 | * @d_blkg: loop cursor pointing to the current descendant | |
492eb21b | 463 | * @pos_css: used for iteration |
aa539cb3 TH |
464 | * @p_blkg: target blkg to walk descendants of |
465 | * | |
466 | * Similar to blkg_for_each_descendant_pre() but performs post-order | |
bd8815a6 TH |
467 | * traversal instead. Synchronization rules are the same. @p_blkg is |
468 | * included in the iteration and the last node to be visited. | |
aa539cb3 | 469 | */ |
492eb21b TH |
470 | #define blkg_for_each_descendant_post(d_blkg, pos_css, p_blkg) \ |
471 | css_for_each_descendant_post((pos_css), &(p_blkg)->blkcg->css) \ | |
472 | if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \ | |
aa539cb3 TH |
473 | (p_blkg)->q, false))) |
474 | ||
a051661c TH |
475 | /** |
476 | * blk_get_rl - get request_list to use | |
477 | * @q: request_queue of interest | |
478 | * @bio: bio which will be attached to the allocated request (may be %NULL) | |
479 | * | |
480 | * The caller wants to allocate a request from @q to use for @bio. Find | |
481 | * the request_list to use and obtain a reference on it. Should be called | |
482 | * under queue_lock. This function is guaranteed to return non-%NULL | |
483 | * request_list. | |
484 | */ | |
485 | static inline struct request_list *blk_get_rl(struct request_queue *q, | |
486 | struct bio *bio) | |
487 | { | |
488 | struct blkcg *blkcg; | |
489 | struct blkcg_gq *blkg; | |
490 | ||
491 | rcu_read_lock(); | |
492 | ||
493 | blkcg = bio_blkcg(bio); | |
494 | ||
495 | /* bypass blkg lookup and use @q->root_rl directly for root */ | |
496 | if (blkcg == &blkcg_root) | |
497 | goto root_rl; | |
498 | ||
499 | /* | |
500 | * Try to use blkg->rl. blkg lookup may fail under memory pressure | |
501 | * or if either the blkcg or queue is going away. Fall back to | |
502 | * root_rl in such cases. | |
503 | */ | |
ae118896 TH |
504 | blkg = blkg_lookup(blkcg, q); |
505 | if (unlikely(!blkg)) | |
a051661c TH |
506 | goto root_rl; |
507 | ||
508 | blkg_get(blkg); | |
509 | rcu_read_unlock(); | |
510 | return &blkg->rl; | |
511 | root_rl: | |
512 | rcu_read_unlock(); | |
513 | return &q->root_rl; | |
514 | } | |
515 | ||
516 | /** | |
517 | * blk_put_rl - put request_list | |
518 | * @rl: request_list to put | |
519 | * | |
520 | * Put the reference acquired by blk_get_rl(). Should be called under | |
521 | * queue_lock. | |
522 | */ | |
523 | static inline void blk_put_rl(struct request_list *rl) | |
524 | { | |
401efbf8 | 525 | if (rl->blkg->blkcg != &blkcg_root) |
a051661c TH |
526 | blkg_put(rl->blkg); |
527 | } | |
528 | ||
529 | /** | |
530 | * blk_rq_set_rl - associate a request with a request_list | |
531 | * @rq: request of interest | |
532 | * @rl: target request_list | |
533 | * | |
534 | * Associate @rq with @rl so that accounting and freeing can know the | |
535 | * request_list @rq came from. | |
536 | */ | |
537 | static inline void blk_rq_set_rl(struct request *rq, struct request_list *rl) | |
538 | { | |
539 | rq->rl = rl; | |
540 | } | |
541 | ||
542 | /** | |
543 | * blk_rq_rl - return the request_list a request came from | |
544 | * @rq: request of interest | |
545 | * | |
546 | * Return the request_list @rq is allocated from. | |
547 | */ | |
548 | static inline struct request_list *blk_rq_rl(struct request *rq) | |
549 | { | |
550 | return rq->rl; | |
551 | } | |
552 | ||
553 | struct request_list *__blk_queue_next_rl(struct request_list *rl, | |
554 | struct request_queue *q); | |
555 | /** | |
556 | * blk_queue_for_each_rl - iterate through all request_lists of a request_queue | |
557 | * | |
558 | * Should be used under queue_lock. | |
559 | */ | |
560 | #define blk_queue_for_each_rl(rl, q) \ | |
561 | for ((rl) = &(q)->root_rl; (rl); (rl) = __blk_queue_next_rl((rl), (q))) | |
562 | ||
24bdb8ef | 563 | static inline int blkg_stat_init(struct blkg_stat *stat, gfp_t gfp) |
90d3839b | 564 | { |
24bdb8ef TH |
565 | int ret; |
566 | ||
567 | ret = percpu_counter_init(&stat->cpu_cnt, 0, gfp); | |
568 | if (ret) | |
569 | return ret; | |
570 | ||
e6269c44 | 571 | atomic64_set(&stat->aux_cnt, 0); |
24bdb8ef TH |
572 | return 0; |
573 | } | |
574 | ||
575 | static inline void blkg_stat_exit(struct blkg_stat *stat) | |
576 | { | |
577 | percpu_counter_destroy(&stat->cpu_cnt); | |
90d3839b PZ |
578 | } |
579 | ||
edcb0722 TH |
580 | /** |
581 | * blkg_stat_add - add a value to a blkg_stat | |
582 | * @stat: target blkg_stat | |
583 | * @val: value to add | |
584 | * | |
24bdb8ef TH |
585 | * Add @val to @stat. The caller must ensure that IRQ on the same CPU |
586 | * don't re-enter this function for the same counter. | |
edcb0722 TH |
587 | */ |
588 | static inline void blkg_stat_add(struct blkg_stat *stat, uint64_t val) | |
589 | { | |
104b4e51 | 590 | percpu_counter_add_batch(&stat->cpu_cnt, val, BLKG_STAT_CPU_BATCH); |
edcb0722 TH |
591 | } |
592 | ||
593 | /** | |
594 | * blkg_stat_read - read the current value of a blkg_stat | |
595 | * @stat: blkg_stat to read | |
edcb0722 TH |
596 | */ |
597 | static inline uint64_t blkg_stat_read(struct blkg_stat *stat) | |
598 | { | |
24bdb8ef | 599 | return percpu_counter_sum_positive(&stat->cpu_cnt); |
edcb0722 TH |
600 | } |
601 | ||
602 | /** | |
603 | * blkg_stat_reset - reset a blkg_stat | |
604 | * @stat: blkg_stat to reset | |
605 | */ | |
606 | static inline void blkg_stat_reset(struct blkg_stat *stat) | |
607 | { | |
24bdb8ef | 608 | percpu_counter_set(&stat->cpu_cnt, 0); |
e6269c44 | 609 | atomic64_set(&stat->aux_cnt, 0); |
edcb0722 TH |
610 | } |
611 | ||
16b3de66 | 612 | /** |
e6269c44 | 613 | * blkg_stat_add_aux - add a blkg_stat into another's aux count |
16b3de66 TH |
614 | * @to: the destination blkg_stat |
615 | * @from: the source | |
616 | * | |
e6269c44 | 617 | * Add @from's count including the aux one to @to's aux count. |
16b3de66 | 618 | */ |
e6269c44 TH |
619 | static inline void blkg_stat_add_aux(struct blkg_stat *to, |
620 | struct blkg_stat *from) | |
16b3de66 | 621 | { |
e6269c44 TH |
622 | atomic64_add(blkg_stat_read(from) + atomic64_read(&from->aux_cnt), |
623 | &to->aux_cnt); | |
16b3de66 TH |
624 | } |
625 | ||
24bdb8ef | 626 | static inline int blkg_rwstat_init(struct blkg_rwstat *rwstat, gfp_t gfp) |
90d3839b | 627 | { |
24bdb8ef TH |
628 | int i, ret; |
629 | ||
630 | for (i = 0; i < BLKG_RWSTAT_NR; i++) { | |
631 | ret = percpu_counter_init(&rwstat->cpu_cnt[i], 0, gfp); | |
632 | if (ret) { | |
633 | while (--i >= 0) | |
634 | percpu_counter_destroy(&rwstat->cpu_cnt[i]); | |
635 | return ret; | |
636 | } | |
637 | atomic64_set(&rwstat->aux_cnt[i], 0); | |
638 | } | |
639 | return 0; | |
640 | } | |
e6269c44 | 641 | |
24bdb8ef TH |
642 | static inline void blkg_rwstat_exit(struct blkg_rwstat *rwstat) |
643 | { | |
644 | int i; | |
e6269c44 TH |
645 | |
646 | for (i = 0; i < BLKG_RWSTAT_NR; i++) | |
24bdb8ef | 647 | percpu_counter_destroy(&rwstat->cpu_cnt[i]); |
90d3839b PZ |
648 | } |
649 | ||
edcb0722 TH |
650 | /** |
651 | * blkg_rwstat_add - add a value to a blkg_rwstat | |
652 | * @rwstat: target blkg_rwstat | |
ef295ecf | 653 | * @op: REQ_OP and flags |
edcb0722 TH |
654 | * @val: value to add |
655 | * | |
656 | * Add @val to @rwstat. The counters are chosen according to @rw. The | |
657 | * caller is responsible for synchronizing calls to this function. | |
658 | */ | |
659 | static inline void blkg_rwstat_add(struct blkg_rwstat *rwstat, | |
ef295ecf | 660 | unsigned int op, uint64_t val) |
edcb0722 | 661 | { |
24bdb8ef | 662 | struct percpu_counter *cnt; |
edcb0722 | 663 | |
636620b6 TH |
664 | if (op_is_discard(op)) |
665 | cnt = &rwstat->cpu_cnt[BLKG_RWSTAT_DISCARD]; | |
666 | else if (op_is_write(op)) | |
24bdb8ef | 667 | cnt = &rwstat->cpu_cnt[BLKG_RWSTAT_WRITE]; |
edcb0722 | 668 | else |
24bdb8ef TH |
669 | cnt = &rwstat->cpu_cnt[BLKG_RWSTAT_READ]; |
670 | ||
104b4e51 | 671 | percpu_counter_add_batch(cnt, val, BLKG_STAT_CPU_BATCH); |
24bdb8ef | 672 | |
d71d9ae1 | 673 | if (op_is_sync(op)) |
24bdb8ef | 674 | cnt = &rwstat->cpu_cnt[BLKG_RWSTAT_SYNC]; |
edcb0722 | 675 | else |
24bdb8ef | 676 | cnt = &rwstat->cpu_cnt[BLKG_RWSTAT_ASYNC]; |
edcb0722 | 677 | |
104b4e51 | 678 | percpu_counter_add_batch(cnt, val, BLKG_STAT_CPU_BATCH); |
edcb0722 TH |
679 | } |
680 | ||
681 | /** | |
682 | * blkg_rwstat_read - read the current values of a blkg_rwstat | |
683 | * @rwstat: blkg_rwstat to read | |
684 | * | |
24bdb8ef | 685 | * Read the current snapshot of @rwstat and return it in the aux counts. |
edcb0722 | 686 | */ |
c94bed89 | 687 | static inline struct blkg_rwstat blkg_rwstat_read(struct blkg_rwstat *rwstat) |
edcb0722 | 688 | { |
24bdb8ef TH |
689 | struct blkg_rwstat result; |
690 | int i; | |
edcb0722 | 691 | |
24bdb8ef TH |
692 | for (i = 0; i < BLKG_RWSTAT_NR; i++) |
693 | atomic64_set(&result.aux_cnt[i], | |
694 | percpu_counter_sum_positive(&rwstat->cpu_cnt[i])); | |
695 | return result; | |
edcb0722 TH |
696 | } |
697 | ||
698 | /** | |
4d5e80a7 | 699 | * blkg_rwstat_total - read the total count of a blkg_rwstat |
edcb0722 TH |
700 | * @rwstat: blkg_rwstat to read |
701 | * | |
702 | * Return the total count of @rwstat regardless of the IO direction. This | |
703 | * function can be called without synchronization and takes care of u64 | |
704 | * atomicity. | |
705 | */ | |
4d5e80a7 | 706 | static inline uint64_t blkg_rwstat_total(struct blkg_rwstat *rwstat) |
edcb0722 TH |
707 | { |
708 | struct blkg_rwstat tmp = blkg_rwstat_read(rwstat); | |
709 | ||
24bdb8ef TH |
710 | return atomic64_read(&tmp.aux_cnt[BLKG_RWSTAT_READ]) + |
711 | atomic64_read(&tmp.aux_cnt[BLKG_RWSTAT_WRITE]); | |
edcb0722 TH |
712 | } |
713 | ||
714 | /** | |
715 | * blkg_rwstat_reset - reset a blkg_rwstat | |
716 | * @rwstat: blkg_rwstat to reset | |
717 | */ | |
718 | static inline void blkg_rwstat_reset(struct blkg_rwstat *rwstat) | |
719 | { | |
e6269c44 TH |
720 | int i; |
721 | ||
24bdb8ef TH |
722 | for (i = 0; i < BLKG_RWSTAT_NR; i++) { |
723 | percpu_counter_set(&rwstat->cpu_cnt[i], 0); | |
e6269c44 | 724 | atomic64_set(&rwstat->aux_cnt[i], 0); |
24bdb8ef | 725 | } |
edcb0722 TH |
726 | } |
727 | ||
16b3de66 | 728 | /** |
e6269c44 | 729 | * blkg_rwstat_add_aux - add a blkg_rwstat into another's aux count |
16b3de66 TH |
730 | * @to: the destination blkg_rwstat |
731 | * @from: the source | |
732 | * | |
e6269c44 | 733 | * Add @from's count including the aux one to @to's aux count. |
16b3de66 | 734 | */ |
e6269c44 TH |
735 | static inline void blkg_rwstat_add_aux(struct blkg_rwstat *to, |
736 | struct blkg_rwstat *from) | |
16b3de66 | 737 | { |
ddc21231 | 738 | u64 sum[BLKG_RWSTAT_NR]; |
16b3de66 TH |
739 | int i; |
740 | ||
16b3de66 | 741 | for (i = 0; i < BLKG_RWSTAT_NR; i++) |
ddc21231 AB |
742 | sum[i] = percpu_counter_sum_positive(&from->cpu_cnt[i]); |
743 | ||
744 | for (i = 0; i < BLKG_RWSTAT_NR; i++) | |
745 | atomic64_add(sum[i] + atomic64_read(&from->aux_cnt[i]), | |
e6269c44 | 746 | &to->aux_cnt[i]); |
16b3de66 TH |
747 | } |
748 | ||
ae118896 TH |
749 | #ifdef CONFIG_BLK_DEV_THROTTLING |
750 | extern bool blk_throtl_bio(struct request_queue *q, struct blkcg_gq *blkg, | |
751 | struct bio *bio); | |
752 | #else | |
753 | static inline bool blk_throtl_bio(struct request_queue *q, struct blkcg_gq *blkg, | |
754 | struct bio *bio) { return false; } | |
755 | #endif | |
756 | ||
757 | static inline bool blkcg_bio_issue_check(struct request_queue *q, | |
758 | struct bio *bio) | |
759 | { | |
760 | struct blkcg *blkcg; | |
761 | struct blkcg_gq *blkg; | |
762 | bool throtl = false; | |
763 | ||
764 | rcu_read_lock(); | |
765 | blkcg = bio_blkcg(bio); | |
766 | ||
007cc56b SL |
767 | /* associate blkcg if bio hasn't attached one */ |
768 | bio_associate_blkcg(bio, &blkcg->css); | |
769 | ||
ae118896 TH |
770 | blkg = blkg_lookup(blkcg, q); |
771 | if (unlikely(!blkg)) { | |
772 | spin_lock_irq(q->queue_lock); | |
773 | blkg = blkg_lookup_create(blkcg, q); | |
774 | if (IS_ERR(blkg)) | |
775 | blkg = NULL; | |
776 | spin_unlock_irq(q->queue_lock); | |
777 | } | |
778 | ||
779 | throtl = blk_throtl_bio(q, blkg, bio); | |
780 | ||
77ea7338 TH |
781 | if (!throtl) { |
782 | blkg = blkg ?: q->root_blkg; | |
c454edc2 JB |
783 | /* |
784 | * If the bio is flagged with BIO_QUEUE_ENTERED it means this | |
785 | * is a split bio and we would have already accounted for the | |
786 | * size of the bio. | |
787 | */ | |
788 | if (!bio_flagged(bio, BIO_QUEUE_ENTERED)) | |
789 | blkg_rwstat_add(&blkg->stat_bytes, bio->bi_opf, | |
790 | bio->bi_iter.bi_size); | |
ef295ecf | 791 | blkg_rwstat_add(&blkg->stat_ios, bio->bi_opf, 1); |
77ea7338 TH |
792 | } |
793 | ||
ae118896 TH |
794 | rcu_read_unlock(); |
795 | return !throtl; | |
796 | } | |
797 | ||
d09d8df3 JB |
798 | static inline void blkcg_use_delay(struct blkcg_gq *blkg) |
799 | { | |
800 | if (atomic_add_return(1, &blkg->use_delay) == 1) | |
801 | atomic_inc(&blkg->blkcg->css.cgroup->congestion_count); | |
802 | } | |
803 | ||
804 | static inline int blkcg_unuse_delay(struct blkcg_gq *blkg) | |
805 | { | |
806 | int old = atomic_read(&blkg->use_delay); | |
807 | ||
808 | if (old == 0) | |
809 | return 0; | |
810 | ||
811 | /* | |
812 | * We do this song and dance because we can race with somebody else | |
813 | * adding or removing delay. If we just did an atomic_dec we'd end up | |
814 | * negative and we'd already be in trouble. We need to subtract 1 and | |
815 | * then check to see if we were the last delay so we can drop the | |
816 | * congestion count on the cgroup. | |
817 | */ | |
818 | while (old) { | |
819 | int cur = atomic_cmpxchg(&blkg->use_delay, old, old - 1); | |
820 | if (cur == old) | |
821 | break; | |
822 | old = cur; | |
823 | } | |
824 | ||
825 | if (old == 0) | |
826 | return 0; | |
827 | if (old == 1) | |
828 | atomic_dec(&blkg->blkcg->css.cgroup->congestion_count); | |
829 | return 1; | |
830 | } | |
831 | ||
832 | static inline void blkcg_clear_delay(struct blkcg_gq *blkg) | |
833 | { | |
834 | int old = atomic_read(&blkg->use_delay); | |
835 | if (!old) | |
836 | return; | |
837 | /* We only want 1 person clearing the congestion count for this blkg. */ | |
838 | while (old) { | |
839 | int cur = atomic_cmpxchg(&blkg->use_delay, old, 0); | |
840 | if (cur == old) { | |
841 | atomic_dec(&blkg->blkcg->css.cgroup->congestion_count); | |
842 | break; | |
843 | } | |
844 | old = cur; | |
845 | } | |
846 | } | |
847 | ||
848 | void blkcg_add_delay(struct blkcg_gq *blkg, u64 now, u64 delta); | |
849 | void blkcg_schedule_throttle(struct request_queue *q, bool use_memdelay); | |
850 | void blkcg_maybe_throttle_current(void); | |
36558c8a TH |
851 | #else /* CONFIG_BLK_CGROUP */ |
852 | ||
efa7d1c7 TH |
853 | struct blkcg { |
854 | }; | |
2f5ea477 | 855 | |
f95a04af TH |
856 | struct blkg_policy_data { |
857 | }; | |
858 | ||
e48453c3 AA |
859 | struct blkcg_policy_data { |
860 | }; | |
861 | ||
3c798398 | 862 | struct blkcg_gq { |
2f5ea477 JA |
863 | }; |
864 | ||
3c798398 | 865 | struct blkcg_policy { |
3e252066 VG |
866 | }; |
867 | ||
496d5e75 TH |
868 | #define blkcg_root_css ((struct cgroup_subsys_state *)ERR_PTR(-EINVAL)) |
869 | ||
d09d8df3 JB |
870 | static inline void blkcg_maybe_throttle_current(void) { } |
871 | static inline bool blk_cgroup_congested(void) { return false; } | |
872 | ||
efa7d1c7 TH |
873 | #ifdef CONFIG_BLOCK |
874 | ||
d09d8df3 JB |
875 | static inline void blkcg_schedule_throttle(struct request_queue *q, bool use_memdelay) { } |
876 | ||
3c798398 | 877 | static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, void *key) { return NULL; } |
b86d865c BVA |
878 | static inline struct blkcg_gq *blk_queue_root_blkg(struct request_queue *q) |
879 | { return NULL; } | |
5efd6113 TH |
880 | static inline int blkcg_init_queue(struct request_queue *q) { return 0; } |
881 | static inline void blkcg_drain_queue(struct request_queue *q) { } | |
882 | static inline void blkcg_exit_queue(struct request_queue *q) { } | |
d5bf0291 | 883 | static inline int blkcg_policy_register(struct blkcg_policy *pol) { return 0; } |
3c798398 | 884 | static inline void blkcg_policy_unregister(struct blkcg_policy *pol) { } |
a2b1693b | 885 | static inline int blkcg_activate_policy(struct request_queue *q, |
3c798398 | 886 | const struct blkcg_policy *pol) { return 0; } |
a2b1693b | 887 | static inline void blkcg_deactivate_policy(struct request_queue *q, |
3c798398 TH |
888 | const struct blkcg_policy *pol) { } |
889 | ||
b1208b56 | 890 | static inline struct blkcg *bio_blkcg(struct bio *bio) { return NULL; } |
a051661c | 891 | |
f95a04af TH |
892 | static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg, |
893 | struct blkcg_policy *pol) { return NULL; } | |
894 | static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd) { return NULL; } | |
3c798398 TH |
895 | static inline char *blkg_path(struct blkcg_gq *blkg) { return NULL; } |
896 | static inline void blkg_get(struct blkcg_gq *blkg) { } | |
897 | static inline void blkg_put(struct blkcg_gq *blkg) { } | |
afc24d49 | 898 | |
a051661c TH |
899 | static inline struct request_list *blk_get_rl(struct request_queue *q, |
900 | struct bio *bio) { return &q->root_rl; } | |
901 | static inline void blk_put_rl(struct request_list *rl) { } | |
902 | static inline void blk_rq_set_rl(struct request *rq, struct request_list *rl) { } | |
903 | static inline struct request_list *blk_rq_rl(struct request *rq) { return &rq->q->root_rl; } | |
904 | ||
ae118896 TH |
905 | static inline bool blkcg_bio_issue_check(struct request_queue *q, |
906 | struct bio *bio) { return true; } | |
907 | ||
a051661c TH |
908 | #define blk_queue_for_each_rl(rl, q) \ |
909 | for ((rl) = &(q)->root_rl; (rl); (rl) = NULL) | |
910 | ||
efa7d1c7 | 911 | #endif /* CONFIG_BLOCK */ |
36558c8a TH |
912 | #endif /* CONFIG_BLK_CGROUP */ |
913 | #endif /* _BLK_CGROUP_H */ |