1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _BLK_CGROUP_PRIVATE_H
3 #define _BLK_CGROUP_PRIVATE_H
4 /*
5 * block cgroup private header
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/blk-cgroup.h>
18 #include <linux/cgroup.h>
19 #include <linux/kthread.h>
20 #include <linux/blk-mq.h>
21 #include <linux/llist.h>
22 #include "blk.h"
23 #include <linux/android_vendor.h>
24
25 struct blkcg_gq;
26 struct blkg_policy_data;
27
28
29 /* percpu_counter batch for blkg_[rw]stats, per-cpu drift doesn't matter */
30 #define BLKG_STAT_CPU_BATCH (INT_MAX / 2)
31
32 #ifdef CONFIG_BLK_CGROUP
33
34 enum blkg_iostat_type {
35 BLKG_IOSTAT_READ,
36 BLKG_IOSTAT_WRITE,
37 BLKG_IOSTAT_DISCARD,
38
39 BLKG_IOSTAT_NR,
40 };
41
42 struct blkg_iostat {
43 u64 bytes[BLKG_IOSTAT_NR];
44 u64 ios[BLKG_IOSTAT_NR];
45 };
46
47 struct blkg_iostat_set {
48 struct u64_stats_sync sync;
49 struct blkcg_gq *blkg;
50 struct llist_node lnode;
51 int lqueued; /* queued in llist */
52 struct blkg_iostat cur;
53 struct blkg_iostat last;
54 };
55
56 /* association between a blk cgroup and a request queue */
57 struct blkcg_gq {
58 /* Pointer to the associated request_queue */
59 struct request_queue *q;
60 struct list_head q_node;
61 struct hlist_node blkcg_node;
62 struct blkcg *blkcg;
63
64 /* all non-root blkcg_gq's are guaranteed to have access to parent */
65 struct blkcg_gq *parent;
66
67 /* reference count */
68 struct percpu_ref refcnt;
69
70 /* is this blkg online? protected by both blkcg and q locks */
71 bool online;
72
73 struct blkg_iostat_set __percpu *iostat_cpu;
74 struct blkg_iostat_set iostat;
75
76 struct blkg_policy_data *pd[BLKCG_MAX_POLS];
77 #ifdef CONFIG_BLK_CGROUP_PUNT_BIO
78 spinlock_t async_bio_lock;
79 struct bio_list async_bios;
80 #endif
81 union {
82 struct work_struct async_bio_work;
83 struct work_struct free_work;
84 };
85
86 atomic_t use_delay;
87 atomic64_t delay_nsec;
88 atomic64_t delay_start;
89 u64 last_delay;
90 int last_use;
91
92 struct rcu_head rcu_head;
93
94 ANDROID_OEM_DATA(1);
95 };
96
97 struct blkcg {
98 struct cgroup_subsys_state css;
99 spinlock_t lock;
100 refcount_t online_pin;
101 /* If there is block congestion on this cgroup. */
102 atomic_t congestion_count;
103
104 struct radix_tree_root blkg_tree;
105 struct blkcg_gq __rcu *blkg_hint;
106 struct hlist_head blkg_list;
107
108 struct blkcg_policy_data *cpd[BLKCG_MAX_POLS];
109
110 struct list_head all_blkcgs_node;
111
112 /*
113 * List of updated percpu blkg_iostat_set's since the last flush.
114 */
115 struct llist_head __percpu *lhead;
116
117 #ifdef CONFIG_BLK_CGROUP_FC_APPID
118 char fc_app_id[FC_APPID_LEN];
119 #endif
120 #ifdef CONFIG_CGROUP_WRITEBACK
121 struct list_head cgwb_list;
122 #endif
123
124 ANDROID_OEM_DATA(1);
125 };
126
css_to_blkcg(struct cgroup_subsys_state * css)127 static inline struct blkcg *css_to_blkcg(struct cgroup_subsys_state *css)
128 {
129 return css ? container_of(css, struct blkcg, css) : NULL;
130 }
131
132 /*
133 * A blkcg_gq (blkg) is association between a block cgroup (blkcg) and a
134 * request_queue (q). This is used by blkcg policies which need to track
135 * information per blkcg - q pair.
136 *
137 * There can be multiple active blkcg policies and each blkg:policy pair is
138 * represented by a blkg_policy_data which is allocated and freed by each
139 * policy's pd_alloc/free_fn() methods. A policy can allocate private data
140 * area by allocating larger data structure which embeds blkg_policy_data
141 * at the beginning.
142 */
143 struct blkg_policy_data {
144 /* the blkg and policy id this per-policy data belongs to */
145 struct blkcg_gq *blkg;
146 int plid;
147 bool online;
148 };
149
150 /*
151 * Policies that need to keep per-blkcg data which is independent from any
152 * request_queue associated to it should implement cpd_alloc/free_fn()
153 * methods. A policy can allocate private data area by allocating larger
154 * data structure which embeds blkcg_policy_data at the beginning.
155 * cpd_init() is invoked to let each policy handle per-blkcg data.
156 */
157 struct blkcg_policy_data {
158 /* the blkcg and policy id this per-policy data belongs to */
159 struct blkcg *blkcg;
160 int plid;
161 };
162
163 typedef struct blkcg_policy_data *(blkcg_pol_alloc_cpd_fn)(gfp_t gfp);
164 typedef void (blkcg_pol_init_cpd_fn)(struct blkcg_policy_data *cpd);
165 typedef void (blkcg_pol_free_cpd_fn)(struct blkcg_policy_data *cpd);
166 typedef void (blkcg_pol_bind_cpd_fn)(struct blkcg_policy_data *cpd);
167 typedef struct blkg_policy_data *(blkcg_pol_alloc_pd_fn)(struct gendisk *disk,
168 struct blkcg *blkcg, gfp_t gfp);
169 typedef void (blkcg_pol_init_pd_fn)(struct blkg_policy_data *pd);
170 typedef void (blkcg_pol_online_pd_fn)(struct blkg_policy_data *pd);
171 typedef void (blkcg_pol_offline_pd_fn)(struct blkg_policy_data *pd);
172 typedef void (blkcg_pol_free_pd_fn)(struct blkg_policy_data *pd);
173 typedef void (blkcg_pol_reset_pd_stats_fn)(struct blkg_policy_data *pd);
174 typedef void (blkcg_pol_stat_pd_fn)(struct blkg_policy_data *pd,
175 struct seq_file *s);
176
177 struct blkcg_policy {
178 int plid;
179 /* cgroup files for the policy */
180 struct cftype *dfl_cftypes;
181 struct cftype *legacy_cftypes;
182
183 /* operations */
184 blkcg_pol_alloc_cpd_fn *cpd_alloc_fn;
185 blkcg_pol_free_cpd_fn *cpd_free_fn;
186
187 blkcg_pol_alloc_pd_fn *pd_alloc_fn;
188 blkcg_pol_init_pd_fn *pd_init_fn;
189 blkcg_pol_online_pd_fn *pd_online_fn;
190 blkcg_pol_offline_pd_fn *pd_offline_fn;
191 blkcg_pol_free_pd_fn *pd_free_fn;
192 blkcg_pol_reset_pd_stats_fn *pd_reset_stats_fn;
193 blkcg_pol_stat_pd_fn *pd_stat_fn;
194 };
195
196 extern struct blkcg blkcg_root;
197 extern bool blkcg_debug_stats;
198
199 void blkg_init_queue(struct request_queue *q);
200 int blkcg_init_disk(struct gendisk *disk);
201 void blkcg_exit_disk(struct gendisk *disk);
202
203 /* Blkio controller policy registration */
204 int blkcg_policy_register(struct blkcg_policy *pol);
205 void blkcg_policy_unregister(struct blkcg_policy *pol);
206 int blkcg_activate_policy(struct gendisk *disk, const struct blkcg_policy *pol);
207 void blkcg_deactivate_policy(struct gendisk *disk,
208 const struct blkcg_policy *pol);
209
210 const char *blkg_dev_name(struct blkcg_gq *blkg);
211 void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg,
212 u64 (*prfill)(struct seq_file *,
213 struct blkg_policy_data *, int),
214 const struct blkcg_policy *pol, int data,
215 bool show_total);
216 u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v);
217
218 struct blkg_conf_ctx {
219 char *input;
220 char *body;
221 struct block_device *bdev;
222 struct blkcg_gq *blkg;
223 };
224
225 void blkg_conf_init(struct blkg_conf_ctx *ctx, char *input);
226 int blkg_conf_open_bdev(struct blkg_conf_ctx *ctx);
227 int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
228 struct blkg_conf_ctx *ctx);
229 void blkg_conf_exit(struct blkg_conf_ctx *ctx);
230
231 /**
232 * bio_issue_as_root_blkg - see if this bio needs to be issued as root blkg
233 * @return: true if this bio needs to be submitted with the root blkg context.
234 *
235 * In order to avoid priority inversions we sometimes need to issue a bio as if
236 * it were attached to the root blkg, and then backcharge to the actual owning
237 * blkg. The idea is we do bio_blkcg_css() to look up the actual context for
238 * the bio and attach the appropriate blkg to the bio. Then we call this helper
239 * and if it is true run with the root blkg for that queue and then do any
240 * backcharging to the originating cgroup once the io is complete.
241 */
bio_issue_as_root_blkg(struct bio * bio)242 static inline bool bio_issue_as_root_blkg(struct bio *bio)
243 {
244 return (bio->bi_opf & (REQ_META | REQ_SWAP)) != 0;
245 }
246
247 /**
248 * blkg_lookup - lookup blkg for the specified blkcg - q pair
249 * @blkcg: blkcg of interest
250 * @q: request_queue of interest
251 *
252 * Lookup blkg for the @blkcg - @q pair.
253
254 * Must be called in a RCU critical section.
255 */
blkg_lookup(struct blkcg * blkcg,struct request_queue * q)256 static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg,
257 struct request_queue *q)
258 {
259 struct blkcg_gq *blkg;
260
261 if (blkcg == &blkcg_root)
262 return q->root_blkg;
263
264 blkg = rcu_dereference_check(blkcg->blkg_hint,
265 lockdep_is_held(&q->queue_lock));
266 if (blkg && blkg->q == q)
267 return blkg;
268
269 blkg = radix_tree_lookup(&blkcg->blkg_tree, q->id);
270 if (blkg && blkg->q != q)
271 blkg = NULL;
272 return blkg;
273 }
274
275 /**
276 * blkg_to_pdata - get policy private data
277 * @blkg: blkg of interest
278 * @pol: policy of interest
279 *
280 * Return pointer to private data associated with the @blkg-@pol pair.
281 */
blkg_to_pd(struct blkcg_gq * blkg,struct blkcg_policy * pol)282 static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
283 struct blkcg_policy *pol)
284 {
285 return blkg ? blkg->pd[pol->plid] : NULL;
286 }
287
blkcg_to_cpd(struct blkcg * blkcg,struct blkcg_policy * pol)288 static inline struct blkcg_policy_data *blkcg_to_cpd(struct blkcg *blkcg,
289 struct blkcg_policy *pol)
290 {
291 return blkcg ? blkcg->cpd[pol->plid] : NULL;
292 }
293
294 /**
295 * pdata_to_blkg - get blkg associated with policy private data
296 * @pd: policy private data of interest
297 *
298 * @pd is policy private data. Determine the blkg it's associated with.
299 */
pd_to_blkg(struct blkg_policy_data * pd)300 static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd)
301 {
302 return pd ? pd->blkg : NULL;
303 }
304
cpd_to_blkcg(struct blkcg_policy_data * cpd)305 static inline struct blkcg *cpd_to_blkcg(struct blkcg_policy_data *cpd)
306 {
307 return cpd ? cpd->blkcg : NULL;
308 }
309
310 /**
311 * blkg_get - get a blkg reference
312 * @blkg: blkg to get
313 *
314 * The caller should be holding an existing reference.
315 */
blkg_get(struct blkcg_gq * blkg)316 static inline void blkg_get(struct blkcg_gq *blkg)
317 {
318 percpu_ref_get(&blkg->refcnt);
319 }
320
321 /**
322 * blkg_tryget - try and get a blkg reference
323 * @blkg: blkg to get
324 *
325 * This is for use when doing an RCU lookup of the blkg. We may be in the midst
326 * of freeing this blkg, so we can only use it if the refcnt is not zero.
327 */
blkg_tryget(struct blkcg_gq * blkg)328 static inline bool blkg_tryget(struct blkcg_gq *blkg)
329 {
330 return blkg && percpu_ref_tryget(&blkg->refcnt);
331 }
332
333 /**
334 * blkg_put - put a blkg reference
335 * @blkg: blkg to put
336 */
blkg_put(struct blkcg_gq * blkg)337 static inline void blkg_put(struct blkcg_gq *blkg)
338 {
339 percpu_ref_put(&blkg->refcnt);
340 }
341
342 /**
343 * blkg_for_each_descendant_pre - pre-order walk of a blkg's descendants
344 * @d_blkg: loop cursor pointing to the current descendant
345 * @pos_css: used for iteration
346 * @p_blkg: target blkg to walk descendants of
347 *
348 * Walk @c_blkg through the descendants of @p_blkg. Must be used with RCU
349 * read locked. If called under either blkcg or queue lock, the iteration
350 * is guaranteed to include all and only online blkgs. The caller may
351 * update @pos_css by calling css_rightmost_descendant() to skip subtree.
352 * @p_blkg is included in the iteration and the first node to be visited.
353 */
354 #define blkg_for_each_descendant_pre(d_blkg, pos_css, p_blkg) \
355 css_for_each_descendant_pre((pos_css), &(p_blkg)->blkcg->css) \
356 if (((d_blkg) = blkg_lookup(css_to_blkcg(pos_css), \
357 (p_blkg)->q)))
358
359 /**
360 * blkg_for_each_descendant_post - post-order walk of a blkg's descendants
361 * @d_blkg: loop cursor pointing to the current descendant
362 * @pos_css: used for iteration
363 * @p_blkg: target blkg to walk descendants of
364 *
365 * Similar to blkg_for_each_descendant_pre() but performs post-order
366 * traversal instead. Synchronization rules are the same. @p_blkg is
367 * included in the iteration and the last node to be visited.
368 */
369 #define blkg_for_each_descendant_post(d_blkg, pos_css, p_blkg) \
370 css_for_each_descendant_post((pos_css), &(p_blkg)->blkcg->css) \
371 if (((d_blkg) = blkg_lookup(css_to_blkcg(pos_css), \
372 (p_blkg)->q)))
373
blkcg_bio_issue_init(struct bio * bio)374 static inline void blkcg_bio_issue_init(struct bio *bio)
375 {
376 bio_issue_init(&bio->bi_issue, bio_sectors(bio));
377 }
378
blkcg_use_delay(struct blkcg_gq * blkg)379 static inline void blkcg_use_delay(struct blkcg_gq *blkg)
380 {
381 if (WARN_ON_ONCE(atomic_read(&blkg->use_delay) < 0))
382 return;
383 if (atomic_add_return(1, &blkg->use_delay) == 1)
384 atomic_inc(&blkg->blkcg->congestion_count);
385 }
386
blkcg_unuse_delay(struct blkcg_gq * blkg)387 static inline int blkcg_unuse_delay(struct blkcg_gq *blkg)
388 {
389 int old = atomic_read(&blkg->use_delay);
390
391 if (WARN_ON_ONCE(old < 0))
392 return 0;
393 if (old == 0)
394 return 0;
395
396 /*
397 * We do this song and dance because we can race with somebody else
398 * adding or removing delay. If we just did an atomic_dec we'd end up
399 * negative and we'd already be in trouble. We need to subtract 1 and
400 * then check to see if we were the last delay so we can drop the
401 * congestion count on the cgroup.
402 */
403 while (old && !atomic_try_cmpxchg(&blkg->use_delay, &old, old - 1))
404 ;
405
406 if (old == 0)
407 return 0;
408 if (old == 1)
409 atomic_dec(&blkg->blkcg->congestion_count);
410 return 1;
411 }
412
413 /**
414 * blkcg_set_delay - Enable allocator delay mechanism with the specified delay amount
415 * @blkg: target blkg
416 * @delay: delay duration in nsecs
417 *
418 * When enabled with this function, the delay is not decayed and must be
419 * explicitly cleared with blkcg_clear_delay(). Must not be mixed with
420 * blkcg_[un]use_delay() and blkcg_add_delay() usages.
421 */
blkcg_set_delay(struct blkcg_gq * blkg,u64 delay)422 static inline void blkcg_set_delay(struct blkcg_gq *blkg, u64 delay)
423 {
424 int old = atomic_read(&blkg->use_delay);
425
426 /* We only want 1 person setting the congestion count for this blkg. */
427 if (!old && atomic_try_cmpxchg(&blkg->use_delay, &old, -1))
428 atomic_inc(&blkg->blkcg->congestion_count);
429
430 atomic64_set(&blkg->delay_nsec, delay);
431 }
432
433 /**
434 * blkcg_clear_delay - Disable allocator delay mechanism
435 * @blkg: target blkg
436 *
437 * Disable use_delay mechanism. See blkcg_set_delay().
438 */
blkcg_clear_delay(struct blkcg_gq * blkg)439 static inline void blkcg_clear_delay(struct blkcg_gq *blkg)
440 {
441 int old = atomic_read(&blkg->use_delay);
442
443 /* We only want 1 person clearing the congestion count for this blkg. */
444 if (old && atomic_try_cmpxchg(&blkg->use_delay, &old, 0))
445 atomic_dec(&blkg->blkcg->congestion_count);
446 }
447
448 /**
449 * blk_cgroup_mergeable - Determine whether to allow or disallow merges
450 * @rq: request to merge into
451 * @bio: bio to merge
452 *
453 * @bio and @rq should belong to the same cgroup and their issue_as_root should
454 * match. The latter is necessary as we don't want to throttle e.g. a metadata
455 * update because it happens to be next to a regular IO.
456 */
blk_cgroup_mergeable(struct request * rq,struct bio * bio)457 static inline bool blk_cgroup_mergeable(struct request *rq, struct bio *bio)
458 {
459 return rq->bio->bi_blkg == bio->bi_blkg &&
460 bio_issue_as_root_blkg(rq->bio) == bio_issue_as_root_blkg(bio);
461 }
462
463 void blk_cgroup_bio_start(struct bio *bio);
464 void blkcg_add_delay(struct blkcg_gq *blkg, u64 now, u64 delta);
465 #else /* CONFIG_BLK_CGROUP */
466
467 struct blkg_policy_data {
468 };
469
470 struct blkcg_policy_data {
471 };
472
473 struct blkcg_policy {
474 };
475
476 struct blkcg {
477 };
478
blkg_lookup(struct blkcg * blkcg,void * key)479 static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, void *key) { return NULL; }
blkg_init_queue(struct request_queue * q)480 static inline void blkg_init_queue(struct request_queue *q) { }
blkcg_init_disk(struct gendisk * disk)481 static inline int blkcg_init_disk(struct gendisk *disk) { return 0; }
blkcg_exit_disk(struct gendisk * disk)482 static inline void blkcg_exit_disk(struct gendisk *disk) { }
blkcg_policy_register(struct blkcg_policy * pol)483 static inline int blkcg_policy_register(struct blkcg_policy *pol) { return 0; }
blkcg_policy_unregister(struct blkcg_policy * pol)484 static inline void blkcg_policy_unregister(struct blkcg_policy *pol) { }
blkcg_activate_policy(struct gendisk * disk,const struct blkcg_policy * pol)485 static inline int blkcg_activate_policy(struct gendisk *disk,
486 const struct blkcg_policy *pol) { return 0; }
blkcg_deactivate_policy(struct gendisk * disk,const struct blkcg_policy * pol)487 static inline void blkcg_deactivate_policy(struct gendisk *disk,
488 const struct blkcg_policy *pol) { }
489
blkg_to_pd(struct blkcg_gq * blkg,struct blkcg_policy * pol)490 static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
491 struct blkcg_policy *pol) { return NULL; }
pd_to_blkg(struct blkg_policy_data * pd)492 static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd) { return NULL; }
blkg_get(struct blkcg_gq * blkg)493 static inline void blkg_get(struct blkcg_gq *blkg) { }
blkg_put(struct blkcg_gq * blkg)494 static inline void blkg_put(struct blkcg_gq *blkg) { }
blkcg_bio_issue_init(struct bio * bio)495 static inline void blkcg_bio_issue_init(struct bio *bio) { }
blk_cgroup_bio_start(struct bio * bio)496 static inline void blk_cgroup_bio_start(struct bio *bio) { }
blk_cgroup_mergeable(struct request * rq,struct bio * bio)497 static inline bool blk_cgroup_mergeable(struct request *rq, struct bio *bio) { return true; }
498
499 #define blk_queue_for_each_rl(rl, q) \
500 for ((rl) = &(q)->root_rl; (rl); (rl) = NULL)
501
502 #endif /* CONFIG_BLK_CGROUP */
503
504 #endif /* _BLK_CGROUP_PRIVATE_H */
505