]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blame - include/linux/blk-cgroup.h
KVM: arm64: Fix PMU probe ordering
[mirror_ubuntu-jammy-kernel.git] / include / linux / blk-cgroup.h
CommitLineData
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>
f7331648 18#include <linux/percpu.h>
24bdb8ef 19#include <linux/percpu_counter.h>
f7331648 20#include <linux/u64_stats_sync.h>
829fdb50 21#include <linux/seq_file.h>
a637120e 22#include <linux/radix-tree.h>
a051661c 23#include <linux/blkdev.h>
a5049a8a 24#include <linux/atomic.h>
902ec5b6 25#include <linux/kthread.h>
5cdf2e3f 26#include <linux/fs.h>
31e4c28d 27
24bdb8ef
TH
28/* percpu_counter batch for blkg_[rw]stats, per-cpu drift doesn't matter */
29#define BLKG_STAT_CPU_BATCH (INT_MAX / 2)
30
9355aede
VG
31/* Max limits for throttle policy */
32#define THROTL_IOPS_MAX UINT_MAX
d2bcbeab
MK
33#define FC_APPID_LEN 129
34
9355aede 35
f48ec1d7
TH
36#ifdef CONFIG_BLK_CGROUP
37
f7331648
TH
38enum blkg_iostat_type {
39 BLKG_IOSTAT_READ,
40 BLKG_IOSTAT_WRITE,
41 BLKG_IOSTAT_DISCARD,
42
43 BLKG_IOSTAT_NR,
44};
45
a637120e
TH
46struct blkcg_gq;
47
3c798398 48struct blkcg {
36558c8a
TH
49 struct cgroup_subsys_state css;
50 spinlock_t lock;
d866dbf6 51 refcount_t online_pin;
a637120e
TH
52
53 struct radix_tree_root blkg_tree;
55679c8d 54 struct blkcg_gq __rcu *blkg_hint;
36558c8a 55 struct hlist_head blkg_list;
9a9e8a26 56
81437648 57 struct blkcg_policy_data *cpd[BLKCG_MAX_POLS];
52ebea74 58
7876f930 59 struct list_head all_blkcgs_node;
d2bcbeab
MK
60#ifdef CONFIG_BLK_CGROUP_FC_APPID
61 char fc_app_id[FC_APPID_LEN];
62#endif
52ebea74
TH
63#ifdef CONFIG_CGROUP_WRITEBACK
64 struct list_head cgwb_list;
65#endif
31e4c28d
VG
66};
67
f7331648
TH
68struct blkg_iostat {
69 u64 bytes[BLKG_IOSTAT_NR];
70 u64 ios[BLKG_IOSTAT_NR];
71};
72
73struct blkg_iostat_set {
74 struct u64_stats_sync sync;
75 struct blkg_iostat cur;
76 struct blkg_iostat last;
77};
78
f95a04af
TH
79/*
80 * A blkcg_gq (blkg) is association between a block cgroup (blkcg) and a
81 * request_queue (q). This is used by blkcg policies which need to track
82 * information per blkcg - q pair.
83 *
001bea73
TH
84 * There can be multiple active blkcg policies and each blkg:policy pair is
85 * represented by a blkg_policy_data which is allocated and freed by each
86 * policy's pd_alloc/free_fn() methods. A policy can allocate private data
87 * area by allocating larger data structure which embeds blkg_policy_data
88 * at the beginning.
f95a04af 89 */
0381411e 90struct blkg_policy_data {
b276a876 91 /* the blkg and policy id this per-policy data belongs to */
3c798398 92 struct blkcg_gq *blkg;
b276a876 93 int plid;
0381411e
TH
94};
95
e48453c3 96/*
e4a9bde9
TH
97 * Policies that need to keep per-blkcg data which is independent from any
98 * request_queue associated to it should implement cpd_alloc/free_fn()
99 * methods. A policy can allocate private data area by allocating larger
100 * data structure which embeds blkcg_policy_data at the beginning.
101 * cpd_init() is invoked to let each policy handle per-blkcg data.
e48453c3
AA
102 */
103struct blkcg_policy_data {
81437648
TH
104 /* the blkcg and policy id this per-policy data belongs to */
105 struct blkcg *blkcg;
e48453c3 106 int plid;
e48453c3
AA
107};
108
3c798398
TH
109/* association between a blk cgroup and a request queue */
110struct blkcg_gq {
c875f4d0 111 /* Pointer to the associated request_queue */
36558c8a
TH
112 struct request_queue *q;
113 struct list_head q_node;
114 struct hlist_node blkcg_node;
3c798398 115 struct blkcg *blkcg;
3c547865
TH
116
117 /* all non-root blkcg_gq's are guaranteed to have access to parent */
118 struct blkcg_gq *parent;
119
1adaf3dd 120 /* reference count */
7fcf2b03 121 struct percpu_ref refcnt;
22084190 122
f427d909
TH
123 /* is this blkg online? protected by both blkcg and q locks */
124 bool online;
125
f7331648
TH
126 struct blkg_iostat_set __percpu *iostat_cpu;
127 struct blkg_iostat_set iostat;
77ea7338 128
36558c8a 129 struct blkg_policy_data *pd[BLKCG_MAX_POLS];
1adaf3dd 130
d3f77dfd
TH
131 spinlock_t async_bio_lock;
132 struct bio_list async_bios;
133 struct work_struct async_bio_work;
d09d8df3
JB
134
135 atomic_t use_delay;
136 atomic64_t delay_nsec;
137 atomic64_t delay_start;
138 u64 last_delay;
139 int last_use;
d3f77dfd
TH
140
141 struct rcu_head rcu_head;
31e4c28d
VG
142};
143
e4a9bde9 144typedef struct blkcg_policy_data *(blkcg_pol_alloc_cpd_fn)(gfp_t gfp);
81437648 145typedef void (blkcg_pol_init_cpd_fn)(struct blkcg_policy_data *cpd);
e4a9bde9 146typedef void (blkcg_pol_free_cpd_fn)(struct blkcg_policy_data *cpd);
69d7fde5 147typedef void (blkcg_pol_bind_cpd_fn)(struct blkcg_policy_data *cpd);
cf09a8ee
TH
148typedef struct blkg_policy_data *(blkcg_pol_alloc_pd_fn)(gfp_t gfp,
149 struct request_queue *q, struct blkcg *blkcg);
a9520cd6
TH
150typedef void (blkcg_pol_init_pd_fn)(struct blkg_policy_data *pd);
151typedef void (blkcg_pol_online_pd_fn)(struct blkg_policy_data *pd);
152typedef void (blkcg_pol_offline_pd_fn)(struct blkg_policy_data *pd);
001bea73 153typedef void (blkcg_pol_free_pd_fn)(struct blkg_policy_data *pd);
a9520cd6 154typedef void (blkcg_pol_reset_pd_stats_fn)(struct blkg_policy_data *pd);
252c651a
CH
155typedef bool (blkcg_pol_stat_pd_fn)(struct blkg_policy_data *pd,
156 struct seq_file *s);
3e252066 157
3c798398 158struct blkcg_policy {
36558c8a 159 int plid;
36558c8a 160 /* cgroup files for the policy */
2ee867dc 161 struct cftype *dfl_cftypes;
880f50e2 162 struct cftype *legacy_cftypes;
f9fcc2d3
TH
163
164 /* operations */
e4a9bde9 165 blkcg_pol_alloc_cpd_fn *cpd_alloc_fn;
e48453c3 166 blkcg_pol_init_cpd_fn *cpd_init_fn;
e4a9bde9 167 blkcg_pol_free_cpd_fn *cpd_free_fn;
69d7fde5 168 blkcg_pol_bind_cpd_fn *cpd_bind_fn;
e4a9bde9 169
001bea73 170 blkcg_pol_alloc_pd_fn *pd_alloc_fn;
f9fcc2d3 171 blkcg_pol_init_pd_fn *pd_init_fn;
f427d909
TH
172 blkcg_pol_online_pd_fn *pd_online_fn;
173 blkcg_pol_offline_pd_fn *pd_offline_fn;
001bea73 174 blkcg_pol_free_pd_fn *pd_free_fn;
f9fcc2d3 175 blkcg_pol_reset_pd_stats_fn *pd_reset_stats_fn;
903d23f0 176 blkcg_pol_stat_pd_fn *pd_stat_fn;
3e252066
VG
177};
178
3c798398 179extern struct blkcg blkcg_root;
496d5e75 180extern struct cgroup_subsys_state * const blkcg_root_css;
07b0fdec 181extern bool blkcg_debug_stats;
36558c8a 182
24f29046
TH
183struct blkcg_gq *blkg_lookup_slowpath(struct blkcg *blkcg,
184 struct request_queue *q, bool update_hint);
36558c8a 185int blkcg_init_queue(struct request_queue *q);
36558c8a 186void blkcg_exit_queue(struct request_queue *q);
5efd6113 187
3e252066 188/* Blkio controller policy registration */
d5bf0291 189int blkcg_policy_register(struct blkcg_policy *pol);
3c798398 190void blkcg_policy_unregister(struct blkcg_policy *pol);
36558c8a 191int blkcg_activate_policy(struct request_queue *q,
3c798398 192 const struct blkcg_policy *pol);
36558c8a 193void blkcg_deactivate_policy(struct request_queue *q,
3c798398 194 const struct blkcg_policy *pol);
3e252066 195
dd165eb3 196const char *blkg_dev_name(struct blkcg_gq *blkg);
3c798398 197void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg,
f95a04af
TH
198 u64 (*prfill)(struct seq_file *,
199 struct blkg_policy_data *, int),
3c798398 200 const struct blkcg_policy *pol, int data,
ec399347 201 bool show_total);
f95a04af 202u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v);
16b3de66 203
829fdb50 204struct blkg_conf_ctx {
22ae8ce8 205 struct block_device *bdev;
3c798398 206 struct blkcg_gq *blkg;
36aa9e5f 207 char *body;
829fdb50
TH
208};
209
22ae8ce8 210struct block_device *blkcg_conf_open_bdev(char **inputp);
3c798398 211int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
36aa9e5f 212 char *input, struct blkg_conf_ctx *ctx);
829fdb50
TH
213void blkg_conf_finish(struct blkg_conf_ctx *ctx);
214
0fe061b9
DZ
215/**
216 * blkcg_css - find the current css
217 *
218 * Find the css associated with either the kthread or the current task.
219 * This may return a dying css, so it is up to the caller to use tryget logic
220 * to confirm it is alive and well.
221 */
222static inline struct cgroup_subsys_state *blkcg_css(void)
223{
224 struct cgroup_subsys_state *css;
225
226 css = kthread_blkcg();
227 if (css)
228 return css;
229 return task_css(current, io_cgrp_id);
230}
231
a7c6d554
TH
232static inline struct blkcg *css_to_blkcg(struct cgroup_subsys_state *css)
233{
234 return css ? container_of(css, struct blkcg, css) : NULL;
235}
236
0fe061b9
DZ
237/**
238 * __bio_blkcg - internal, inconsistent version to get blkcg
239 *
240 * DO NOT USE.
241 * This function is inconsistent and consequently is dangerous to use. The
242 * first part of the function returns a blkcg where a reference is owned by the
243 * bio. This means it does not need to be rcu protected as it cannot go away
244 * with the bio owning a reference to it. However, the latter potentially gets
245 * it from task_css(). This can race against task migration and the cgroup
246 * dying. It is also semantically different as it must be called rcu protected
247 * and is susceptible to failure when trying to get a reference to it.
248 * Therefore, it is not ok to assume that *_get() will always succeed on the
249 * blkcg returned here.
250 */
251static inline struct blkcg *__bio_blkcg(struct bio *bio)
27e6fa99 252{
db6638d7
DZ
253 if (bio && bio->bi_blkg)
254 return bio->bi_blkg->blkcg;
0fe061b9
DZ
255 return css_to_blkcg(blkcg_css());
256}
b5f2954d 257
0fe061b9
DZ
258/**
259 * bio_blkcg - grab the blkcg associated with a bio
260 * @bio: target bio
261 *
262 * This returns the blkcg associated with a bio, %NULL if not associated.
263 * Callers are expected to either handle %NULL or know association has been
264 * done prior to calling this.
265 */
266static inline struct blkcg *bio_blkcg(struct bio *bio)
267{
db6638d7
DZ
268 if (bio && bio->bi_blkg)
269 return bio->bi_blkg->blkcg;
0fe061b9 270 return NULL;
fd383c2d
TH
271}
272
d09d8df3
JB
273static inline bool blk_cgroup_congested(void)
274{
275 struct cgroup_subsys_state *css;
276 bool ret = false;
277
278 rcu_read_lock();
279 css = kthread_blkcg();
280 if (!css)
281 css = task_css(current, io_cgrp_id);
282 while (css) {
283 if (atomic_read(&css->cgroup->congestion_count)) {
284 ret = true;
285 break;
286 }
287 css = css->parent;
288 }
289 rcu_read_unlock();
290 return ret;
291}
292
c7c98fd3
JB
293/**
294 * bio_issue_as_root_blkg - see if this bio needs to be issued as root blkg
295 * @return: true if this bio needs to be submitted with the root blkg context.
296 *
297 * In order to avoid priority inversions we sometimes need to issue a bio as if
298 * it were attached to the root blkg, and then backcharge to the actual owning
299 * blkg. The idea is we do bio_blkcg() to look up the actual context for the
300 * bio and attach the appropriate blkg to the bio. Then we call this helper and
301 * if it is true run with the root blkg for that queue and then do any
302 * backcharging to the originating cgroup once the io is complete.
303 */
304static inline bool bio_issue_as_root_blkg(struct bio *bio)
305{
0d1e0c7c 306 return (bio->bi_opf & (REQ_META | REQ_SWAP)) != 0;
c7c98fd3
JB
307}
308
3c547865
TH
309/**
310 * blkcg_parent - get the parent of a blkcg
311 * @blkcg: blkcg of interest
312 *
313 * Return the parent blkcg of @blkcg. Can be called anytime.
314 */
315static inline struct blkcg *blkcg_parent(struct blkcg *blkcg)
316{
5c9d535b 317 return css_to_blkcg(blkcg->css.parent);
3c547865
TH
318}
319
24f29046
TH
320/**
321 * __blkg_lookup - internal version of blkg_lookup()
322 * @blkcg: blkcg of interest
323 * @q: request_queue of interest
324 * @update_hint: whether to update lookup hint with the result or not
325 *
326 * This is internal version and shouldn't be used by policy
327 * implementations. Looks up blkgs for the @blkcg - @q pair regardless of
328 * @q's bypass state. If @update_hint is %true, the caller should be
329 * holding @q->queue_lock and lookup hint is updated on success.
330 */
331static inline struct blkcg_gq *__blkg_lookup(struct blkcg *blkcg,
332 struct request_queue *q,
333 bool update_hint)
334{
335 struct blkcg_gq *blkg;
336
85b6bc9d
TH
337 if (blkcg == &blkcg_root)
338 return q->root_blkg;
339
24f29046
TH
340 blkg = rcu_dereference(blkcg->blkg_hint);
341 if (blkg && blkg->q == q)
342 return blkg;
343
344 return blkg_lookup_slowpath(blkcg, q, update_hint);
345}
346
347/**
348 * blkg_lookup - lookup blkg for the specified blkcg - q pair
349 * @blkcg: blkcg of interest
350 * @q: request_queue of interest
351 *
352 * Lookup blkg for the @blkcg - @q pair. This function should be called
012d4a65 353 * under RCU read lock.
24f29046
TH
354 */
355static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg,
356 struct request_queue *q)
357{
358 WARN_ON_ONCE(!rcu_read_lock_held());
24f29046
TH
359 return __blkg_lookup(blkcg, q, false);
360}
361
6bad9b21 362/**
b86d865c 363 * blk_queue_root_blkg - return blkg for the (blkcg_root, @q) pair
6bad9b21
BVA
364 * @q: request_queue of interest
365 *
366 * Lookup blkg for @q at the root level. See also blkg_lookup().
367 */
b86d865c 368static inline struct blkcg_gq *blk_queue_root_blkg(struct request_queue *q)
6bad9b21 369{
b86d865c 370 return q->root_blkg;
6bad9b21
BVA
371}
372
0381411e
TH
373/**
374 * blkg_to_pdata - get policy private data
375 * @blkg: blkg of interest
376 * @pol: policy of interest
377 *
378 * Return pointer to private data associated with the @blkg-@pol pair.
379 */
f95a04af
TH
380static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
381 struct blkcg_policy *pol)
0381411e 382{
f95a04af 383 return blkg ? blkg->pd[pol->plid] : NULL;
0381411e
TH
384}
385
e48453c3
AA
386static inline struct blkcg_policy_data *blkcg_to_cpd(struct blkcg *blkcg,
387 struct blkcg_policy *pol)
388{
81437648 389 return blkcg ? blkcg->cpd[pol->plid] : NULL;
e48453c3
AA
390}
391
0381411e
TH
392/**
393 * pdata_to_blkg - get blkg associated with policy private data
f95a04af 394 * @pd: policy private data of interest
0381411e 395 *
f95a04af 396 * @pd is policy private data. Determine the blkg it's associated with.
0381411e 397 */
f95a04af 398static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd)
0381411e 399{
f95a04af 400 return pd ? pd->blkg : NULL;
0381411e
TH
401}
402
81437648
TH
403static inline struct blkcg *cpd_to_blkcg(struct blkcg_policy_data *cpd)
404{
405 return cpd ? cpd->blkcg : NULL;
406}
407
59b57717
DZF
408extern void blkcg_destroy_blkgs(struct blkcg *blkcg);
409
59b57717 410/**
d866dbf6 411 * blkcg_pin_online - pin online state
59b57717
DZF
412 * @blkcg: blkcg of interest
413 *
d866dbf6
TH
414 * While pinned, a blkcg is kept online. This is primarily used to
415 * impedance-match blkg and cgwb lifetimes so that blkg doesn't go offline
416 * while an associated cgwb is still active.
59b57717 417 */
d866dbf6 418static inline void blkcg_pin_online(struct blkcg *blkcg)
59b57717 419{
d866dbf6 420 refcount_inc(&blkcg->online_pin);
59b57717
DZF
421}
422
423/**
d866dbf6 424 * blkcg_unpin_online - unpin online state
59b57717
DZF
425 * @blkcg: blkcg of interest
426 *
d866dbf6
TH
427 * This is primarily used to impedance-match blkg and cgwb lifetimes so
428 * that blkg doesn't go offline while an associated cgwb is still active.
429 * When this count goes to zero, all active cgwbs have finished so the
59b57717 430 * blkcg can continue destruction by calling blkcg_destroy_blkgs().
59b57717 431 */
d866dbf6 432static inline void blkcg_unpin_online(struct blkcg *blkcg)
59b57717 433{
4308a434
TH
434 do {
435 if (!refcount_dec_and_test(&blkcg->online_pin))
436 break;
59b57717 437 blkcg_destroy_blkgs(blkcg);
4308a434
TH
438 blkcg = blkcg_parent(blkcg);
439 } while (blkcg);
59b57717
DZF
440}
441
54e7ed12
TH
442/**
443 * blkg_path - format cgroup path of blkg
444 * @blkg: blkg of interest
445 * @buf: target buffer
446 * @buflen: target buffer length
447 *
448 * Format the path of the cgroup of @blkg into @buf.
449 */
3c798398 450static inline int blkg_path(struct blkcg_gq *blkg, char *buf, int buflen)
afc24d49 451{
4c737b41 452 return cgroup_path(blkg->blkcg->css.cgroup, buf, buflen);
afc24d49
VG
453}
454
1adaf3dd
TH
455/**
456 * blkg_get - get a blkg reference
457 * @blkg: blkg to get
458 *
a5049a8a 459 * The caller should be holding an existing reference.
1adaf3dd 460 */
3c798398 461static inline void blkg_get(struct blkcg_gq *blkg)
1adaf3dd 462{
7fcf2b03 463 percpu_ref_get(&blkg->refcnt);
1adaf3dd
TH
464}
465
d09d8df3 466/**
7754f669 467 * blkg_tryget - try and get a blkg reference
d09d8df3
JB
468 * @blkg: blkg to get
469 *
470 * This is for use when doing an RCU lookup of the blkg. We may be in the midst
471 * of freeing this blkg, so we can only use it if the refcnt is not zero.
472 */
7754f669 473static inline bool blkg_tryget(struct blkcg_gq *blkg)
d09d8df3 474{
6ab21879 475 return blkg && percpu_ref_tryget(&blkg->refcnt);
d09d8df3
JB
476}
477
1adaf3dd
TH
478/**
479 * blkg_put - put a blkg reference
480 * @blkg: blkg to put
1adaf3dd 481 */
3c798398 482static inline void blkg_put(struct blkcg_gq *blkg)
1adaf3dd 483{
7fcf2b03 484 percpu_ref_put(&blkg->refcnt);
1adaf3dd
TH
485}
486
dd4a4ffc
TH
487/**
488 * blkg_for_each_descendant_pre - pre-order walk of a blkg's descendants
489 * @d_blkg: loop cursor pointing to the current descendant
492eb21b 490 * @pos_css: used for iteration
dd4a4ffc
TH
491 * @p_blkg: target blkg to walk descendants of
492 *
493 * Walk @c_blkg through the descendants of @p_blkg. Must be used with RCU
494 * read locked. If called under either blkcg or queue lock, the iteration
495 * is guaranteed to include all and only online blkgs. The caller may
492eb21b 496 * update @pos_css by calling css_rightmost_descendant() to skip subtree.
bd8815a6 497 * @p_blkg is included in the iteration and the first node to be visited.
dd4a4ffc 498 */
492eb21b
TH
499#define blkg_for_each_descendant_pre(d_blkg, pos_css, p_blkg) \
500 css_for_each_descendant_pre((pos_css), &(p_blkg)->blkcg->css) \
501 if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \
dd4a4ffc
TH
502 (p_blkg)->q, false)))
503
aa539cb3
TH
504/**
505 * blkg_for_each_descendant_post - post-order walk of a blkg's descendants
506 * @d_blkg: loop cursor pointing to the current descendant
492eb21b 507 * @pos_css: used for iteration
aa539cb3
TH
508 * @p_blkg: target blkg to walk descendants of
509 *
510 * Similar to blkg_for_each_descendant_pre() but performs post-order
bd8815a6
TH
511 * traversal instead. Synchronization rules are the same. @p_blkg is
512 * included in the iteration and the last node to be visited.
aa539cb3 513 */
492eb21b
TH
514#define blkg_for_each_descendant_post(d_blkg, pos_css, p_blkg) \
515 css_for_each_descendant_post((pos_css), &(p_blkg)->blkcg->css) \
516 if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \
aa539cb3
TH
517 (p_blkg)->q, false)))
518
d3f77dfd
TH
519bool __blkcg_punt_bio_submit(struct bio *bio);
520
521static inline bool blkcg_punt_bio_submit(struct bio *bio)
522{
523 if (bio->bi_opf & REQ_CGROUP_PUNT)
524 return __blkcg_punt_bio_submit(bio);
525 else
526 return false;
527}
e439bedf
DZ
528
529static inline void blkcg_bio_issue_init(struct bio *bio)
530{
531 bio_issue_init(&bio->bi_issue, bio_sectors(bio));
532}
533
d09d8df3
JB
534static inline void blkcg_use_delay(struct blkcg_gq *blkg)
535{
54c52e10
TH
536 if (WARN_ON_ONCE(atomic_read(&blkg->use_delay) < 0))
537 return;
d09d8df3
JB
538 if (atomic_add_return(1, &blkg->use_delay) == 1)
539 atomic_inc(&blkg->blkcg->css.cgroup->congestion_count);
540}
541
542static inline int blkcg_unuse_delay(struct blkcg_gq *blkg)
543{
544 int old = atomic_read(&blkg->use_delay);
545
54c52e10
TH
546 if (WARN_ON_ONCE(old < 0))
547 return 0;
d09d8df3
JB
548 if (old == 0)
549 return 0;
550
551 /*
552 * We do this song and dance because we can race with somebody else
553 * adding or removing delay. If we just did an atomic_dec we'd end up
554 * negative and we'd already be in trouble. We need to subtract 1 and
555 * then check to see if we were the last delay so we can drop the
556 * congestion count on the cgroup.
557 */
558 while (old) {
559 int cur = atomic_cmpxchg(&blkg->use_delay, old, old - 1);
560 if (cur == old)
561 break;
562 old = cur;
563 }
564
565 if (old == 0)
566 return 0;
567 if (old == 1)
568 atomic_dec(&blkg->blkcg->css.cgroup->congestion_count);
569 return 1;
570}
571
54c52e10
TH
572/**
573 * blkcg_set_delay - Enable allocator delay mechanism with the specified delay amount
574 * @blkg: target blkg
575 * @delay: delay duration in nsecs
576 *
577 * When enabled with this function, the delay is not decayed and must be
578 * explicitly cleared with blkcg_clear_delay(). Must not be mixed with
579 * blkcg_[un]use_delay() and blkcg_add_delay() usages.
580 */
581static inline void blkcg_set_delay(struct blkcg_gq *blkg, u64 delay)
582{
583 int old = atomic_read(&blkg->use_delay);
584
585 /* We only want 1 person setting the congestion count for this blkg. */
586 if (!old && atomic_cmpxchg(&blkg->use_delay, old, -1) == old)
587 atomic_inc(&blkg->blkcg->css.cgroup->congestion_count);
588
589 atomic64_set(&blkg->delay_nsec, delay);
590}
591
592/**
593 * blkcg_clear_delay - Disable allocator delay mechanism
594 * @blkg: target blkg
595 *
596 * Disable use_delay mechanism. See blkcg_set_delay().
597 */
d09d8df3
JB
598static inline void blkcg_clear_delay(struct blkcg_gq *blkg)
599{
600 int old = atomic_read(&blkg->use_delay);
54c52e10 601
d09d8df3 602 /* We only want 1 person clearing the congestion count for this blkg. */
54c52e10
TH
603 if (old && atomic_cmpxchg(&blkg->use_delay, old, 0) == old)
604 atomic_dec(&blkg->blkcg->css.cgroup->congestion_count);
d09d8df3
JB
605}
606
db18a53e 607void blk_cgroup_bio_start(struct bio *bio);
d09d8df3
JB
608void blkcg_add_delay(struct blkcg_gq *blkg, u64 now, u64 delta);
609void blkcg_schedule_throttle(struct request_queue *q, bool use_memdelay);
610void blkcg_maybe_throttle_current(void);
36558c8a
TH
611#else /* CONFIG_BLK_CGROUP */
612
efa7d1c7
TH
613struct blkcg {
614};
2f5ea477 615
f95a04af
TH
616struct blkg_policy_data {
617};
618
e48453c3
AA
619struct blkcg_policy_data {
620};
621
3c798398 622struct blkcg_gq {
2f5ea477
JA
623};
624
3c798398 625struct blkcg_policy {
3e252066
VG
626};
627
496d5e75
TH
628#define blkcg_root_css ((struct cgroup_subsys_state *)ERR_PTR(-EINVAL))
629
d09d8df3
JB
630static inline void blkcg_maybe_throttle_current(void) { }
631static inline bool blk_cgroup_congested(void) { return false; }
632
efa7d1c7
TH
633#ifdef CONFIG_BLOCK
634
d09d8df3
JB
635static inline void blkcg_schedule_throttle(struct request_queue *q, bool use_memdelay) { }
636
3c798398 637static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, void *key) { return NULL; }
b86d865c
BVA
638static inline struct blkcg_gq *blk_queue_root_blkg(struct request_queue *q)
639{ return NULL; }
5efd6113 640static inline int blkcg_init_queue(struct request_queue *q) { return 0; }
5efd6113 641static inline void blkcg_exit_queue(struct request_queue *q) { }
d5bf0291 642static inline int blkcg_policy_register(struct blkcg_policy *pol) { return 0; }
3c798398 643static inline void blkcg_policy_unregister(struct blkcg_policy *pol) { }
a2b1693b 644static inline int blkcg_activate_policy(struct request_queue *q,
3c798398 645 const struct blkcg_policy *pol) { return 0; }
a2b1693b 646static inline void blkcg_deactivate_policy(struct request_queue *q,
3c798398
TH
647 const struct blkcg_policy *pol) { }
648
0fe061b9 649static inline struct blkcg *__bio_blkcg(struct bio *bio) { return NULL; }
b1208b56 650static inline struct blkcg *bio_blkcg(struct bio *bio) { return NULL; }
a051661c 651
f95a04af
TH
652static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
653 struct blkcg_policy *pol) { return NULL; }
654static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd) { return NULL; }
3c798398
TH
655static inline char *blkg_path(struct blkcg_gq *blkg) { return NULL; }
656static inline void blkg_get(struct blkcg_gq *blkg) { }
657static inline void blkg_put(struct blkcg_gq *blkg) { }
afc24d49 658
d3f77dfd 659static inline bool blkcg_punt_bio_submit(struct bio *bio) { return false; }
e439bedf 660static inline void blkcg_bio_issue_init(struct bio *bio) { }
db18a53e 661static inline void blk_cgroup_bio_start(struct bio *bio) { }
ae118896 662
a051661c
TH
663#define blk_queue_for_each_rl(rl, q) \
664 for ((rl) = &(q)->root_rl; (rl); (rl) = NULL)
665
efa7d1c7 666#endif /* CONFIG_BLOCK */
36558c8a 667#endif /* CONFIG_BLK_CGROUP */
d2bcbeab
MK
668
669#ifdef CONFIG_BLK_CGROUP_FC_APPID
670/*
671 * Sets the fc_app_id field associted to blkcg
672 * @app_id: application identifier
673 * @cgrp_id: cgroup id
674 * @app_id_len: size of application identifier
675 */
676static inline int blkcg_set_fc_appid(char *app_id, u64 cgrp_id, size_t app_id_len)
677{
678 struct cgroup *cgrp;
679 struct cgroup_subsys_state *css;
680 struct blkcg *blkcg;
681 int ret = 0;
682
683 if (app_id_len > FC_APPID_LEN)
684 return -EINVAL;
685
686 cgrp = cgroup_get_from_id(cgrp_id);
687 if (!cgrp)
688 return -ENOENT;
689 css = cgroup_get_e_css(cgrp, &io_cgrp_subsys);
690 if (!css) {
691 ret = -ENOENT;
692 goto out_cgrp_put;
693 }
694 blkcg = css_to_blkcg(css);
695 /*
696 * There is a slight race condition on setting the appid.
697 * Worst case an I/O may not find the right id.
698 * This is no different from the I/O we let pass while obtaining
699 * the vmid from the fabric.
700 * Adding the overhead of a lock is not necessary.
701 */
702 strlcpy(blkcg->fc_app_id, app_id, app_id_len);
703 css_put(css);
704out_cgrp_put:
705 cgroup_put(cgrp);
706 return ret;
707}
708
709/**
710 * blkcg_get_fc_appid - get the fc app identifier associated with a bio
711 * @bio: target bio
712 *
713 * On success return the fc_app_id, on failure return NULL
714 */
715static inline char *blkcg_get_fc_appid(struct bio *bio)
716{
717 if (bio && bio->bi_blkg &&
718 (bio->bi_blkg->blkcg->fc_app_id[0] != '\0'))
719 return bio->bi_blkg->blkcg->fc_app_id;
720 return NULL;
721}
722#else
723static inline int blkcg_set_fc_appid(char *buf, u64 id, size_t len) { return -EINVAL; }
724static inline char *blkcg_get_fc_appid(struct bio *bio) { return NULL; }
725#endif /*CONFIG_BLK_CGROUP_FC_APPID*/
36558c8a 726#endif /* _BLK_CGROUP_H */