blob: 4630ce8f9425a606968807bcf4c2d2f4596ccfbd [file] [log] [blame]
Vivek Goyal31e4c282009-12-03 12:59:42 -05001#ifndef _BLK_CGROUP_H
2#define _BLK_CGROUP_H
3/*
4 * Common Block IO controller cgroup interface
5 *
6 * Based on ideas and code from CFQ, CFS and BFQ:
7 * Copyright (C) 2003 Jens Axboe <axboe@kernel.dk>
8 *
9 * Copyright (C) 2008 Fabio Checconi <fabio@gandalf.sssup.it>
10 * Paolo Valente <paolo.valente@unimore.it>
11 *
12 * Copyright (C) 2009 Vivek Goyal <vgoyal@redhat.com>
13 * Nauman Rafique <nauman@google.com>
14 */
15
16#include <linux/cgroup.h>
Tejun Heo24bdb8e2015-08-18 14:55:22 -070017#include <linux/percpu_counter.h>
Tejun Heo829fdb52012-04-01 14:38:43 -070018#include <linux/seq_file.h>
Tejun Heoa6371202012-04-19 16:29:24 -070019#include <linux/radix-tree.h>
Tejun Heoa0516612012-06-26 15:05:44 -070020#include <linux/blkdev.h>
Tejun Heoa5049a82014-06-19 17:42:57 -040021#include <linux/atomic.h>
Vivek Goyal31e4c282009-12-03 12:59:42 -050022
Tejun Heo24bdb8e2015-08-18 14:55:22 -070023/* percpu_counter batch for blkg_[rw]stats, per-cpu drift doesn't matter */
24#define BLKG_STAT_CPU_BATCH (INT_MAX / 2)
25
Vivek Goyal9355aed2010-10-01 21:16:41 +020026/* Max limits for throttle policy */
27#define THROTL_IOPS_MAX UINT_MAX
28
Tejun Heof48ec1d2012-04-13 13:11:25 -070029#ifdef CONFIG_BLK_CGROUP
30
Tejun Heoedcb0722012-04-01 14:38:42 -070031enum blkg_rwstat_type {
32 BLKG_RWSTAT_READ,
33 BLKG_RWSTAT_WRITE,
34 BLKG_RWSTAT_SYNC,
35 BLKG_RWSTAT_ASYNC,
36
37 BLKG_RWSTAT_NR,
38 BLKG_RWSTAT_TOTAL = BLKG_RWSTAT_NR,
Divyesh Shah303a3ac2010-04-01 15:01:24 -070039};
40
Tejun Heoa6371202012-04-19 16:29:24 -070041struct blkcg_gq;
42
Tejun Heo3c798392012-04-16 13:57:25 -070043struct blkcg {
Tejun Heo36558c82012-04-16 13:57:24 -070044 struct cgroup_subsys_state css;
45 spinlock_t lock;
Tejun Heoa6371202012-04-19 16:29:24 -070046
47 struct radix_tree_root blkg_tree;
48 struct blkcg_gq *blkg_hint;
Tejun Heo36558c82012-04-16 13:57:24 -070049 struct hlist_head blkg_list;
Tejun Heo9a9e8a22012-03-19 15:10:56 -070050
Tejun Heo81437642015-08-18 14:55:15 -070051 struct blkcg_policy_data *cpd[BLKCG_MAX_POLS];
Tejun Heo52ebea72015-05-22 17:13:37 -040052
Tejun Heo7876f932015-07-09 16:39:49 -040053 struct list_head all_blkcgs_node;
Tejun Heo52ebea72015-05-22 17:13:37 -040054#ifdef CONFIG_CGROUP_WRITEBACK
55 struct list_head cgwb_list;
56#endif
Vivek Goyal31e4c282009-12-03 12:59:42 -050057};
58
Tejun Heoe6269c42015-08-18 14:55:21 -070059/*
60 * blkg_[rw]stat->aux_cnt is excluded for local stats but included for
Tejun Heo24bdb8e2015-08-18 14:55:22 -070061 * recursive. Used to carry stats of dead children, and, for blkg_rwstat,
62 * to carry result values from read and sum operations.
Tejun Heoe6269c42015-08-18 14:55:21 -070063 */
Tejun Heoedcb0722012-04-01 14:38:42 -070064struct blkg_stat {
Tejun Heo24bdb8e2015-08-18 14:55:22 -070065 struct percpu_counter cpu_cnt;
Tejun Heoe6269c42015-08-18 14:55:21 -070066 atomic64_t aux_cnt;
Tejun Heoedcb0722012-04-01 14:38:42 -070067};
68
69struct blkg_rwstat {
Tejun Heo24bdb8e2015-08-18 14:55:22 -070070 struct percpu_counter cpu_cnt[BLKG_RWSTAT_NR];
Tejun Heoe6269c42015-08-18 14:55:21 -070071 atomic64_t aux_cnt[BLKG_RWSTAT_NR];
Tejun Heoedcb0722012-04-01 14:38:42 -070072};
73
Tejun Heof95a04a2012-04-16 13:57:26 -070074/*
75 * A blkcg_gq (blkg) is association between a block cgroup (blkcg) and a
76 * request_queue (q). This is used by blkcg policies which need to track
77 * information per blkcg - q pair.
78 *
Tejun Heo001bea72015-08-18 14:55:11 -070079 * There can be multiple active blkcg policies and each blkg:policy pair is
80 * represented by a blkg_policy_data which is allocated and freed by each
81 * policy's pd_alloc/free_fn() methods. A policy can allocate private data
82 * area by allocating larger data structure which embeds blkg_policy_data
83 * at the beginning.
Tejun Heof95a04a2012-04-16 13:57:26 -070084 */
Tejun Heo03814112012-03-05 13:15:14 -080085struct blkg_policy_data {
Tejun Heob276a872013-01-09 08:05:12 -080086 /* the blkg and policy id this per-policy data belongs to */
Tejun Heo3c798392012-04-16 13:57:25 -070087 struct blkcg_gq *blkg;
Tejun Heob276a872013-01-09 08:05:12 -080088 int plid;
Tejun Heo03814112012-03-05 13:15:14 -080089};
90
Arianna Avanzinie48453c2015-06-05 23:38:42 +020091/*
Tejun Heoe4a9bde2015-08-18 14:55:16 -070092 * Policies that need to keep per-blkcg data which is independent from any
93 * request_queue associated to it should implement cpd_alloc/free_fn()
94 * methods. A policy can allocate private data area by allocating larger
95 * data structure which embeds blkcg_policy_data at the beginning.
96 * cpd_init() is invoked to let each policy handle per-blkcg data.
Arianna Avanzinie48453c2015-06-05 23:38:42 +020097 */
98struct blkcg_policy_data {
Tejun Heo81437642015-08-18 14:55:15 -070099 /* the blkcg and policy id this per-policy data belongs to */
100 struct blkcg *blkcg;
Arianna Avanzinie48453c2015-06-05 23:38:42 +0200101 int plid;
Arianna Avanzinie48453c2015-06-05 23:38:42 +0200102};
103
Tejun Heo3c798392012-04-16 13:57:25 -0700104/* association between a blk cgroup and a request queue */
105struct blkcg_gq {
Tejun Heoc875f4d2012-03-05 13:15:22 -0800106 /* Pointer to the associated request_queue */
Tejun Heo36558c82012-04-16 13:57:24 -0700107 struct request_queue *q;
108 struct list_head q_node;
109 struct hlist_node blkcg_node;
Tejun Heo3c798392012-04-16 13:57:25 -0700110 struct blkcg *blkcg;
Tejun Heo3c547862013-01-09 08:05:10 -0800111
Tejun Heoce7acfe2015-05-22 17:13:38 -0400112 /*
113 * Each blkg gets congested separately and the congestion state is
114 * propagated to the matching bdi_writeback_congested.
115 */
116 struct bdi_writeback_congested *wb_congested;
117
Tejun Heo3c547862013-01-09 08:05:10 -0800118 /* all non-root blkcg_gq's are guaranteed to have access to parent */
119 struct blkcg_gq *parent;
120
Tejun Heoa0516612012-06-26 15:05:44 -0700121 /* request allocation list for this blkcg-q pair */
122 struct request_list rl;
Tejun Heo3c547862013-01-09 08:05:10 -0800123
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800124 /* reference count */
Tejun Heoa5049a82014-06-19 17:42:57 -0400125 atomic_t refcnt;
Vivek Goyal22084192009-12-03 12:59:49 -0500126
Tejun Heof427d902013-01-09 08:05:12 -0800127 /* is this blkg online? protected by both blkcg and q locks */
128 bool online;
129
Tejun Heo36558c82012-04-16 13:57:24 -0700130 struct blkg_policy_data *pd[BLKCG_MAX_POLS];
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800131
Tejun Heo36558c82012-04-16 13:57:24 -0700132 struct rcu_head rcu_head;
Vivek Goyal31e4c282009-12-03 12:59:42 -0500133};
134
Tejun Heoe4a9bde2015-08-18 14:55:16 -0700135typedef struct blkcg_policy_data *(blkcg_pol_alloc_cpd_fn)(gfp_t gfp);
Tejun Heo81437642015-08-18 14:55:15 -0700136typedef void (blkcg_pol_init_cpd_fn)(struct blkcg_policy_data *cpd);
Tejun Heoe4a9bde2015-08-18 14:55:16 -0700137typedef void (blkcg_pol_free_cpd_fn)(struct blkcg_policy_data *cpd);
Tejun Heo001bea72015-08-18 14:55:11 -0700138typedef struct blkg_policy_data *(blkcg_pol_alloc_pd_fn)(gfp_t gfp, int node);
Tejun Heoa9520cd2015-08-18 14:55:14 -0700139typedef void (blkcg_pol_init_pd_fn)(struct blkg_policy_data *pd);
140typedef void (blkcg_pol_online_pd_fn)(struct blkg_policy_data *pd);
141typedef void (blkcg_pol_offline_pd_fn)(struct blkg_policy_data *pd);
Tejun Heo001bea72015-08-18 14:55:11 -0700142typedef void (blkcg_pol_free_pd_fn)(struct blkg_policy_data *pd);
Tejun Heoa9520cd2015-08-18 14:55:14 -0700143typedef void (blkcg_pol_reset_pd_stats_fn)(struct blkg_policy_data *pd);
Vivek Goyal3e252062009-12-04 10:36:42 -0500144
Tejun Heo3c798392012-04-16 13:57:25 -0700145struct blkcg_policy {
Tejun Heo36558c82012-04-16 13:57:24 -0700146 int plid;
Tejun Heo36558c82012-04-16 13:57:24 -0700147 /* cgroup files for the policy */
148 struct cftype *cftypes;
Tejun Heof9fcc2d2012-04-16 13:57:27 -0700149
150 /* operations */
Tejun Heoe4a9bde2015-08-18 14:55:16 -0700151 blkcg_pol_alloc_cpd_fn *cpd_alloc_fn;
Arianna Avanzinie48453c2015-06-05 23:38:42 +0200152 blkcg_pol_init_cpd_fn *cpd_init_fn;
Tejun Heoe4a9bde2015-08-18 14:55:16 -0700153 blkcg_pol_free_cpd_fn *cpd_free_fn;
154
Tejun Heo001bea72015-08-18 14:55:11 -0700155 blkcg_pol_alloc_pd_fn *pd_alloc_fn;
Tejun Heof9fcc2d2012-04-16 13:57:27 -0700156 blkcg_pol_init_pd_fn *pd_init_fn;
Tejun Heof427d902013-01-09 08:05:12 -0800157 blkcg_pol_online_pd_fn *pd_online_fn;
158 blkcg_pol_offline_pd_fn *pd_offline_fn;
Tejun Heo001bea72015-08-18 14:55:11 -0700159 blkcg_pol_free_pd_fn *pd_free_fn;
Tejun Heof9fcc2d2012-04-16 13:57:27 -0700160 blkcg_pol_reset_pd_stats_fn *pd_reset_stats_fn;
Vivek Goyal3e252062009-12-04 10:36:42 -0500161};
162
Tejun Heo3c798392012-04-16 13:57:25 -0700163extern struct blkcg blkcg_root;
Tejun Heo496d5e72015-05-22 17:13:21 -0400164extern struct cgroup_subsys_state * const blkcg_root_css;
Tejun Heo36558c82012-04-16 13:57:24 -0700165
Tejun Heo24f29042015-08-18 14:55:17 -0700166struct blkcg_gq *blkg_lookup_slowpath(struct blkcg *blkcg,
167 struct request_queue *q, bool update_hint);
Tejun Heo3c798392012-04-16 13:57:25 -0700168struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg,
169 struct request_queue *q);
Tejun Heo36558c82012-04-16 13:57:24 -0700170int blkcg_init_queue(struct request_queue *q);
171void blkcg_drain_queue(struct request_queue *q);
172void blkcg_exit_queue(struct request_queue *q);
Tejun Heo5efd6112012-03-05 13:15:12 -0800173
Vivek Goyal3e252062009-12-04 10:36:42 -0500174/* Blkio controller policy registration */
Jens Axboed5bf0292014-06-22 16:31:56 -0600175int blkcg_policy_register(struct blkcg_policy *pol);
Tejun Heo3c798392012-04-16 13:57:25 -0700176void blkcg_policy_unregister(struct blkcg_policy *pol);
Tejun Heo36558c82012-04-16 13:57:24 -0700177int blkcg_activate_policy(struct request_queue *q,
Tejun Heo3c798392012-04-16 13:57:25 -0700178 const struct blkcg_policy *pol);
Tejun Heo36558c82012-04-16 13:57:24 -0700179void blkcg_deactivate_policy(struct request_queue *q,
Tejun Heo3c798392012-04-16 13:57:25 -0700180 const struct blkcg_policy *pol);
Vivek Goyal3e252062009-12-04 10:36:42 -0500181
Tejun Heo3c798392012-04-16 13:57:25 -0700182void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg,
Tejun Heof95a04a2012-04-16 13:57:26 -0700183 u64 (*prfill)(struct seq_file *,
184 struct blkg_policy_data *, int),
Tejun Heo3c798392012-04-16 13:57:25 -0700185 const struct blkcg_policy *pol, int data,
Tejun Heoec399342012-04-13 13:11:27 -0700186 bool show_total);
Tejun Heof95a04a2012-04-16 13:57:26 -0700187u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v);
188u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
Tejun Heo829fdb52012-04-01 14:38:43 -0700189 const struct blkg_rwstat *rwstat);
Tejun Heof95a04a2012-04-16 13:57:26 -0700190u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off);
191u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
192 int off);
Tejun Heo829fdb52012-04-01 14:38:43 -0700193
Tejun Heof12c74c2015-08-18 14:55:23 -0700194u64 blkg_stat_recursive_sum(struct blkcg_gq *blkg,
195 struct blkcg_policy *pol, int off);
196struct blkg_rwstat blkg_rwstat_recursive_sum(struct blkcg_gq *blkg,
197 struct blkcg_policy *pol, int off);
Tejun Heo16b3de62013-01-09 08:05:12 -0800198
Tejun Heo829fdb52012-04-01 14:38:43 -0700199struct blkg_conf_ctx {
Tejun Heo36558c82012-04-16 13:57:24 -0700200 struct gendisk *disk;
Tejun Heo3c798392012-04-16 13:57:25 -0700201 struct blkcg_gq *blkg;
Tejun Heo36558c82012-04-16 13:57:24 -0700202 u64 v;
Tejun Heo829fdb52012-04-01 14:38:43 -0700203};
204
Tejun Heo3c798392012-04-16 13:57:25 -0700205int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
206 const char *input, struct blkg_conf_ctx *ctx);
Tejun Heo829fdb52012-04-01 14:38:43 -0700207void blkg_conf_finish(struct blkg_conf_ctx *ctx);
208
209
Tejun Heoa7c6d552013-08-08 20:11:23 -0400210static inline struct blkcg *css_to_blkcg(struct cgroup_subsys_state *css)
211{
212 return css ? container_of(css, struct blkcg, css) : NULL;
213}
214
Tejun Heob1208b52012-06-04 20:40:57 -0700215static inline struct blkcg *task_blkcg(struct task_struct *tsk)
216{
Tejun Heo073219e2014-02-08 10:36:58 -0500217 return css_to_blkcg(task_css(tsk, blkio_cgrp_id));
Tejun Heob1208b52012-06-04 20:40:57 -0700218}
219
220static inline struct blkcg *bio_blkcg(struct bio *bio)
221{
222 if (bio && bio->bi_css)
Tejun Heoa7c6d552013-08-08 20:11:23 -0400223 return css_to_blkcg(bio->bi_css);
Tejun Heob1208b52012-06-04 20:40:57 -0700224 return task_blkcg(current);
225}
226
Tejun Heofd383c22015-05-22 17:13:23 -0400227static inline struct cgroup_subsys_state *
228task_get_blkcg_css(struct task_struct *task)
229{
230 return task_get_css(task, blkio_cgrp_id);
231}
232
Tejun Heo03814112012-03-05 13:15:14 -0800233/**
Tejun Heo3c547862013-01-09 08:05:10 -0800234 * blkcg_parent - get the parent of a blkcg
235 * @blkcg: blkcg of interest
236 *
237 * Return the parent blkcg of @blkcg. Can be called anytime.
238 */
239static inline struct blkcg *blkcg_parent(struct blkcg *blkcg)
240{
Tejun Heo5c9d5352014-05-16 13:22:48 -0400241 return css_to_blkcg(blkcg->css.parent);
Tejun Heo3c547862013-01-09 08:05:10 -0800242}
243
244/**
Tejun Heo24f29042015-08-18 14:55:17 -0700245 * __blkg_lookup - internal version of blkg_lookup()
246 * @blkcg: blkcg of interest
247 * @q: request_queue of interest
248 * @update_hint: whether to update lookup hint with the result or not
249 *
250 * This is internal version and shouldn't be used by policy
251 * implementations. Looks up blkgs for the @blkcg - @q pair regardless of
252 * @q's bypass state. If @update_hint is %true, the caller should be
253 * holding @q->queue_lock and lookup hint is updated on success.
254 */
255static inline struct blkcg_gq *__blkg_lookup(struct blkcg *blkcg,
256 struct request_queue *q,
257 bool update_hint)
258{
259 struct blkcg_gq *blkg;
260
Tejun Heo85b6bc92015-08-18 14:55:18 -0700261 if (blkcg == &blkcg_root)
262 return q->root_blkg;
263
Tejun Heo24f29042015-08-18 14:55:17 -0700264 blkg = rcu_dereference(blkcg->blkg_hint);
265 if (blkg && blkg->q == q)
266 return blkg;
267
268 return blkg_lookup_slowpath(blkcg, q, update_hint);
269}
270
271/**
272 * blkg_lookup - lookup blkg for the specified blkcg - q pair
273 * @blkcg: blkcg of interest
274 * @q: request_queue of interest
275 *
276 * Lookup blkg for the @blkcg - @q pair. This function should be called
277 * under RCU read lock and is guaranteed to return %NULL if @q is bypassing
278 * - see blk_queue_bypass_start() for details.
279 */
280static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg,
281 struct request_queue *q)
282{
283 WARN_ON_ONCE(!rcu_read_lock_held());
284
285 if (unlikely(blk_queue_bypass(q)))
286 return NULL;
287 return __blkg_lookup(blkcg, q, false);
288}
289
290/**
Tejun Heo03814112012-03-05 13:15:14 -0800291 * blkg_to_pdata - get policy private data
292 * @blkg: blkg of interest
293 * @pol: policy of interest
294 *
295 * Return pointer to private data associated with the @blkg-@pol pair.
296 */
Tejun Heof95a04a2012-04-16 13:57:26 -0700297static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
298 struct blkcg_policy *pol)
Tejun Heo03814112012-03-05 13:15:14 -0800299{
Tejun Heof95a04a2012-04-16 13:57:26 -0700300 return blkg ? blkg->pd[pol->plid] : NULL;
Tejun Heo03814112012-03-05 13:15:14 -0800301}
302
Arianna Avanzinie48453c2015-06-05 23:38:42 +0200303static inline struct blkcg_policy_data *blkcg_to_cpd(struct blkcg *blkcg,
304 struct blkcg_policy *pol)
305{
Tejun Heo81437642015-08-18 14:55:15 -0700306 return blkcg ? blkcg->cpd[pol->plid] : NULL;
Arianna Avanzinie48453c2015-06-05 23:38:42 +0200307}
308
Tejun Heo03814112012-03-05 13:15:14 -0800309/**
310 * pdata_to_blkg - get blkg associated with policy private data
Tejun Heof95a04a2012-04-16 13:57:26 -0700311 * @pd: policy private data of interest
Tejun Heo03814112012-03-05 13:15:14 -0800312 *
Tejun Heof95a04a2012-04-16 13:57:26 -0700313 * @pd is policy private data. Determine the blkg it's associated with.
Tejun Heo03814112012-03-05 13:15:14 -0800314 */
Tejun Heof95a04a2012-04-16 13:57:26 -0700315static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd)
Tejun Heo03814112012-03-05 13:15:14 -0800316{
Tejun Heof95a04a2012-04-16 13:57:26 -0700317 return pd ? pd->blkg : NULL;
Tejun Heo03814112012-03-05 13:15:14 -0800318}
319
Tejun Heo81437642015-08-18 14:55:15 -0700320static inline struct blkcg *cpd_to_blkcg(struct blkcg_policy_data *cpd)
321{
322 return cpd ? cpd->blkcg : NULL;
323}
324
Tejun Heo54e7ed12012-04-16 13:57:23 -0700325/**
326 * blkg_path - format cgroup path of blkg
327 * @blkg: blkg of interest
328 * @buf: target buffer
329 * @buflen: target buffer length
330 *
331 * Format the path of the cgroup of @blkg into @buf.
332 */
Tejun Heo3c798392012-04-16 13:57:25 -0700333static inline int blkg_path(struct blkcg_gq *blkg, char *buf, int buflen)
Vivek Goyalafc24d42010-04-26 19:27:56 +0200334{
Tejun Heoe61734c2014-02-12 09:29:50 -0500335 char *p;
Tejun Heo54e7ed12012-04-16 13:57:23 -0700336
Tejun Heoe61734c2014-02-12 09:29:50 -0500337 p = cgroup_path(blkg->blkcg->css.cgroup, buf, buflen);
338 if (!p) {
Tejun Heo54e7ed12012-04-16 13:57:23 -0700339 strncpy(buf, "<unavailable>", buflen);
Tejun Heoe61734c2014-02-12 09:29:50 -0500340 return -ENAMETOOLONG;
341 }
342
343 memmove(buf, p, buf + buflen - p);
344 return 0;
Vivek Goyalafc24d42010-04-26 19:27:56 +0200345}
346
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800347/**
348 * blkg_get - get a blkg reference
349 * @blkg: blkg to get
350 *
Tejun Heoa5049a82014-06-19 17:42:57 -0400351 * The caller should be holding an existing reference.
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800352 */
Tejun Heo3c798392012-04-16 13:57:25 -0700353static inline void blkg_get(struct blkcg_gq *blkg)
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800354{
Tejun Heoa5049a82014-06-19 17:42:57 -0400355 WARN_ON_ONCE(atomic_read(&blkg->refcnt) <= 0);
356 atomic_inc(&blkg->refcnt);
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800357}
358
Tejun Heo2a4fd072013-05-14 13:52:31 -0700359void __blkg_release_rcu(struct rcu_head *rcu);
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800360
361/**
362 * blkg_put - put a blkg reference
363 * @blkg: blkg to put
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800364 */
Tejun Heo3c798392012-04-16 13:57:25 -0700365static inline void blkg_put(struct blkcg_gq *blkg)
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800366{
Tejun Heoa5049a82014-06-19 17:42:57 -0400367 WARN_ON_ONCE(atomic_read(&blkg->refcnt) <= 0);
368 if (atomic_dec_and_test(&blkg->refcnt))
Tejun Heo2a4fd072013-05-14 13:52:31 -0700369 call_rcu(&blkg->rcu_head, __blkg_release_rcu);
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800370}
371
Tejun Heodd4a4ff2013-05-14 13:52:30 -0700372/**
373 * blkg_for_each_descendant_pre - pre-order walk of a blkg's descendants
374 * @d_blkg: loop cursor pointing to the current descendant
Tejun Heo492eb212013-08-08 20:11:25 -0400375 * @pos_css: used for iteration
Tejun Heodd4a4ff2013-05-14 13:52:30 -0700376 * @p_blkg: target blkg to walk descendants of
377 *
378 * Walk @c_blkg through the descendants of @p_blkg. Must be used with RCU
379 * read locked. If called under either blkcg or queue lock, the iteration
380 * is guaranteed to include all and only online blkgs. The caller may
Tejun Heo492eb212013-08-08 20:11:25 -0400381 * update @pos_css by calling css_rightmost_descendant() to skip subtree.
Tejun Heobd8815a2013-08-08 20:11:27 -0400382 * @p_blkg is included in the iteration and the first node to be visited.
Tejun Heodd4a4ff2013-05-14 13:52:30 -0700383 */
Tejun Heo492eb212013-08-08 20:11:25 -0400384#define blkg_for_each_descendant_pre(d_blkg, pos_css, p_blkg) \
385 css_for_each_descendant_pre((pos_css), &(p_blkg)->blkcg->css) \
386 if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \
Tejun Heodd4a4ff2013-05-14 13:52:30 -0700387 (p_blkg)->q, false)))
388
Tejun Heoedcb0722012-04-01 14:38:42 -0700389/**
Tejun Heoaa539cb2013-05-14 13:52:31 -0700390 * blkg_for_each_descendant_post - post-order walk of a blkg's descendants
391 * @d_blkg: loop cursor pointing to the current descendant
Tejun Heo492eb212013-08-08 20:11:25 -0400392 * @pos_css: used for iteration
Tejun Heoaa539cb2013-05-14 13:52:31 -0700393 * @p_blkg: target blkg to walk descendants of
394 *
395 * Similar to blkg_for_each_descendant_pre() but performs post-order
Tejun Heobd8815a2013-08-08 20:11:27 -0400396 * traversal instead. Synchronization rules are the same. @p_blkg is
397 * included in the iteration and the last node to be visited.
Tejun Heoaa539cb2013-05-14 13:52:31 -0700398 */
Tejun Heo492eb212013-08-08 20:11:25 -0400399#define blkg_for_each_descendant_post(d_blkg, pos_css, p_blkg) \
400 css_for_each_descendant_post((pos_css), &(p_blkg)->blkcg->css) \
401 if (((d_blkg) = __blkg_lookup(css_to_blkcg(pos_css), \
Tejun Heoaa539cb2013-05-14 13:52:31 -0700402 (p_blkg)->q, false)))
403
404/**
Tejun Heoa0516612012-06-26 15:05:44 -0700405 * blk_get_rl - get request_list to use
406 * @q: request_queue of interest
407 * @bio: bio which will be attached to the allocated request (may be %NULL)
408 *
409 * The caller wants to allocate a request from @q to use for @bio. Find
410 * the request_list to use and obtain a reference on it. Should be called
411 * under queue_lock. This function is guaranteed to return non-%NULL
412 * request_list.
413 */
414static inline struct request_list *blk_get_rl(struct request_queue *q,
415 struct bio *bio)
416{
417 struct blkcg *blkcg;
418 struct blkcg_gq *blkg;
419
420 rcu_read_lock();
421
422 blkcg = bio_blkcg(bio);
423
424 /* bypass blkg lookup and use @q->root_rl directly for root */
425 if (blkcg == &blkcg_root)
426 goto root_rl;
427
428 /*
429 * Try to use blkg->rl. blkg lookup may fail under memory pressure
430 * or if either the blkcg or queue is going away. Fall back to
431 * root_rl in such cases.
432 */
Tejun Heoae118892015-08-18 14:55:20 -0700433 blkg = blkg_lookup(blkcg, q);
434 if (unlikely(!blkg))
Tejun Heoa0516612012-06-26 15:05:44 -0700435 goto root_rl;
436
437 blkg_get(blkg);
438 rcu_read_unlock();
439 return &blkg->rl;
440root_rl:
441 rcu_read_unlock();
442 return &q->root_rl;
443}
444
445/**
446 * blk_put_rl - put request_list
447 * @rl: request_list to put
448 *
449 * Put the reference acquired by blk_get_rl(). Should be called under
450 * queue_lock.
451 */
452static inline void blk_put_rl(struct request_list *rl)
453{
Tejun Heo401efbf2015-08-18 14:55:06 -0700454 if (rl->blkg->blkcg != &blkcg_root)
Tejun Heoa0516612012-06-26 15:05:44 -0700455 blkg_put(rl->blkg);
456}
457
458/**
459 * blk_rq_set_rl - associate a request with a request_list
460 * @rq: request of interest
461 * @rl: target request_list
462 *
463 * Associate @rq with @rl so that accounting and freeing can know the
464 * request_list @rq came from.
465 */
466static inline void blk_rq_set_rl(struct request *rq, struct request_list *rl)
467{
468 rq->rl = rl;
469}
470
471/**
472 * blk_rq_rl - return the request_list a request came from
473 * @rq: request of interest
474 *
475 * Return the request_list @rq is allocated from.
476 */
477static inline struct request_list *blk_rq_rl(struct request *rq)
478{
479 return rq->rl;
480}
481
482struct request_list *__blk_queue_next_rl(struct request_list *rl,
483 struct request_queue *q);
484/**
485 * blk_queue_for_each_rl - iterate through all request_lists of a request_queue
486 *
487 * Should be used under queue_lock.
488 */
489#define blk_queue_for_each_rl(rl, q) \
490 for ((rl) = &(q)->root_rl; (rl); (rl) = __blk_queue_next_rl((rl), (q)))
491
Tejun Heo24bdb8e2015-08-18 14:55:22 -0700492static inline int blkg_stat_init(struct blkg_stat *stat, gfp_t gfp)
Peter Zijlstra90d38392013-11-12 19:42:14 -0800493{
Tejun Heo24bdb8e2015-08-18 14:55:22 -0700494 int ret;
495
496 ret = percpu_counter_init(&stat->cpu_cnt, 0, gfp);
497 if (ret)
498 return ret;
499
Tejun Heoe6269c42015-08-18 14:55:21 -0700500 atomic64_set(&stat->aux_cnt, 0);
Tejun Heo24bdb8e2015-08-18 14:55:22 -0700501 return 0;
502}
503
504static inline void blkg_stat_exit(struct blkg_stat *stat)
505{
506 percpu_counter_destroy(&stat->cpu_cnt);
Peter Zijlstra90d38392013-11-12 19:42:14 -0800507}
508
Tejun Heoa0516612012-06-26 15:05:44 -0700509/**
Tejun Heoedcb0722012-04-01 14:38:42 -0700510 * blkg_stat_add - add a value to a blkg_stat
511 * @stat: target blkg_stat
512 * @val: value to add
513 *
Tejun Heo24bdb8e2015-08-18 14:55:22 -0700514 * Add @val to @stat. The caller must ensure that IRQ on the same CPU
515 * don't re-enter this function for the same counter.
Tejun Heoedcb0722012-04-01 14:38:42 -0700516 */
517static inline void blkg_stat_add(struct blkg_stat *stat, uint64_t val)
518{
Tejun Heo24bdb8e2015-08-18 14:55:22 -0700519 __percpu_counter_add(&stat->cpu_cnt, val, BLKG_STAT_CPU_BATCH);
Tejun Heoedcb0722012-04-01 14:38:42 -0700520}
521
522/**
523 * blkg_stat_read - read the current value of a blkg_stat
524 * @stat: blkg_stat to read
Tejun Heoedcb0722012-04-01 14:38:42 -0700525 */
526static inline uint64_t blkg_stat_read(struct blkg_stat *stat)
527{
Tejun Heo24bdb8e2015-08-18 14:55:22 -0700528 return percpu_counter_sum_positive(&stat->cpu_cnt);
Tejun Heoedcb0722012-04-01 14:38:42 -0700529}
530
531/**
532 * blkg_stat_reset - reset a blkg_stat
533 * @stat: blkg_stat to reset
534 */
535static inline void blkg_stat_reset(struct blkg_stat *stat)
536{
Tejun Heo24bdb8e2015-08-18 14:55:22 -0700537 percpu_counter_set(&stat->cpu_cnt, 0);
Tejun Heoe6269c42015-08-18 14:55:21 -0700538 atomic64_set(&stat->aux_cnt, 0);
Tejun Heoedcb0722012-04-01 14:38:42 -0700539}
540
541/**
Tejun Heoe6269c42015-08-18 14:55:21 -0700542 * blkg_stat_add_aux - add a blkg_stat into another's aux count
Tejun Heo16b3de62013-01-09 08:05:12 -0800543 * @to: the destination blkg_stat
544 * @from: the source
545 *
Tejun Heoe6269c42015-08-18 14:55:21 -0700546 * Add @from's count including the aux one to @to's aux count.
Tejun Heo16b3de62013-01-09 08:05:12 -0800547 */
Tejun Heoe6269c42015-08-18 14:55:21 -0700548static inline void blkg_stat_add_aux(struct blkg_stat *to,
549 struct blkg_stat *from)
Tejun Heo16b3de62013-01-09 08:05:12 -0800550{
Tejun Heoe6269c42015-08-18 14:55:21 -0700551 atomic64_add(blkg_stat_read(from) + atomic64_read(&from->aux_cnt),
552 &to->aux_cnt);
Tejun Heo16b3de62013-01-09 08:05:12 -0800553}
554
Tejun Heo24bdb8e2015-08-18 14:55:22 -0700555static inline int blkg_rwstat_init(struct blkg_rwstat *rwstat, gfp_t gfp)
556{
557 int i, ret;
558
559 for (i = 0; i < BLKG_RWSTAT_NR; i++) {
560 ret = percpu_counter_init(&rwstat->cpu_cnt[i], 0, gfp);
561 if (ret) {
562 while (--i >= 0)
563 percpu_counter_destroy(&rwstat->cpu_cnt[i]);
564 return ret;
565 }
566 atomic64_set(&rwstat->aux_cnt[i], 0);
567 }
568 return 0;
569}
570
571static inline void blkg_rwstat_exit(struct blkg_rwstat *rwstat)
Peter Zijlstra90d38392013-11-12 19:42:14 -0800572{
Tejun Heoe6269c42015-08-18 14:55:21 -0700573 int i;
574
Tejun Heoe6269c42015-08-18 14:55:21 -0700575 for (i = 0; i < BLKG_RWSTAT_NR; i++)
Tejun Heo24bdb8e2015-08-18 14:55:22 -0700576 percpu_counter_destroy(&rwstat->cpu_cnt[i]);
Peter Zijlstra90d38392013-11-12 19:42:14 -0800577}
578
Tejun Heo16b3de62013-01-09 08:05:12 -0800579/**
Tejun Heoedcb0722012-04-01 14:38:42 -0700580 * blkg_rwstat_add - add a value to a blkg_rwstat
581 * @rwstat: target blkg_rwstat
582 * @rw: mask of REQ_{WRITE|SYNC}
583 * @val: value to add
584 *
585 * Add @val to @rwstat. The counters are chosen according to @rw. The
586 * caller is responsible for synchronizing calls to this function.
587 */
588static inline void blkg_rwstat_add(struct blkg_rwstat *rwstat,
589 int rw, uint64_t val)
590{
Tejun Heo24bdb8e2015-08-18 14:55:22 -0700591 struct percpu_counter *cnt;
Tejun Heoedcb0722012-04-01 14:38:42 -0700592
593 if (rw & REQ_WRITE)
Tejun Heo24bdb8e2015-08-18 14:55:22 -0700594 cnt = &rwstat->cpu_cnt[BLKG_RWSTAT_WRITE];
Tejun Heoedcb0722012-04-01 14:38:42 -0700595 else
Tejun Heo24bdb8e2015-08-18 14:55:22 -0700596 cnt = &rwstat->cpu_cnt[BLKG_RWSTAT_READ];
Tejun Heoedcb0722012-04-01 14:38:42 -0700597
Tejun Heo24bdb8e2015-08-18 14:55:22 -0700598 __percpu_counter_add(cnt, val, BLKG_STAT_CPU_BATCH);
599
600 if (rw & REQ_SYNC)
601 cnt = &rwstat->cpu_cnt[BLKG_RWSTAT_SYNC];
602 else
603 cnt = &rwstat->cpu_cnt[BLKG_RWSTAT_ASYNC];
604
605 __percpu_counter_add(cnt, val, BLKG_STAT_CPU_BATCH);
Tejun Heoedcb0722012-04-01 14:38:42 -0700606}
607
608/**
609 * blkg_rwstat_read - read the current values of a blkg_rwstat
610 * @rwstat: blkg_rwstat to read
611 *
Tejun Heo24bdb8e2015-08-18 14:55:22 -0700612 * Read the current snapshot of @rwstat and return it in the aux counts.
Tejun Heoedcb0722012-04-01 14:38:42 -0700613 */
Tejun Heoc94bed892012-04-16 13:57:22 -0700614static inline struct blkg_rwstat blkg_rwstat_read(struct blkg_rwstat *rwstat)
Tejun Heoedcb0722012-04-01 14:38:42 -0700615{
Tejun Heo24bdb8e2015-08-18 14:55:22 -0700616 struct blkg_rwstat result;
617 int i;
Tejun Heoedcb0722012-04-01 14:38:42 -0700618
Tejun Heo24bdb8e2015-08-18 14:55:22 -0700619 for (i = 0; i < BLKG_RWSTAT_NR; i++)
620 atomic64_set(&result.aux_cnt[i],
621 percpu_counter_sum_positive(&rwstat->cpu_cnt[i]));
622 return result;
Tejun Heoedcb0722012-04-01 14:38:42 -0700623}
624
625/**
Tejun Heo4d5e80a2013-01-09 08:05:12 -0800626 * blkg_rwstat_total - read the total count of a blkg_rwstat
Tejun Heoedcb0722012-04-01 14:38:42 -0700627 * @rwstat: blkg_rwstat to read
628 *
629 * Return the total count of @rwstat regardless of the IO direction. This
630 * function can be called without synchronization and takes care of u64
631 * atomicity.
632 */
Tejun Heo4d5e80a2013-01-09 08:05:12 -0800633static inline uint64_t blkg_rwstat_total(struct blkg_rwstat *rwstat)
Tejun Heoedcb0722012-04-01 14:38:42 -0700634{
635 struct blkg_rwstat tmp = blkg_rwstat_read(rwstat);
636
Tejun Heo24bdb8e2015-08-18 14:55:22 -0700637 return atomic64_read(&tmp.aux_cnt[BLKG_RWSTAT_READ]) +
638 atomic64_read(&tmp.aux_cnt[BLKG_RWSTAT_WRITE]);
Tejun Heoedcb0722012-04-01 14:38:42 -0700639}
640
641/**
642 * blkg_rwstat_reset - reset a blkg_rwstat
643 * @rwstat: blkg_rwstat to reset
644 */
645static inline void blkg_rwstat_reset(struct blkg_rwstat *rwstat)
646{
Tejun Heoe6269c42015-08-18 14:55:21 -0700647 int i;
648
Tejun Heo24bdb8e2015-08-18 14:55:22 -0700649 for (i = 0; i < BLKG_RWSTAT_NR; i++) {
650 percpu_counter_set(&rwstat->cpu_cnt[i], 0);
Tejun Heoe6269c42015-08-18 14:55:21 -0700651 atomic64_set(&rwstat->aux_cnt[i], 0);
Tejun Heo24bdb8e2015-08-18 14:55:22 -0700652 }
Tejun Heoedcb0722012-04-01 14:38:42 -0700653}
654
Tejun Heo16b3de62013-01-09 08:05:12 -0800655/**
Tejun Heoe6269c42015-08-18 14:55:21 -0700656 * blkg_rwstat_add_aux - add a blkg_rwstat into another's aux count
Tejun Heo16b3de62013-01-09 08:05:12 -0800657 * @to: the destination blkg_rwstat
658 * @from: the source
659 *
Tejun Heoe6269c42015-08-18 14:55:21 -0700660 * Add @from's count including the aux one to @to's aux count.
Tejun Heo16b3de62013-01-09 08:05:12 -0800661 */
Tejun Heoe6269c42015-08-18 14:55:21 -0700662static inline void blkg_rwstat_add_aux(struct blkg_rwstat *to,
663 struct blkg_rwstat *from)
Tejun Heo16b3de62013-01-09 08:05:12 -0800664{
665 struct blkg_rwstat v = blkg_rwstat_read(from);
666 int i;
667
Tejun Heo16b3de62013-01-09 08:05:12 -0800668 for (i = 0; i < BLKG_RWSTAT_NR; i++)
Tejun Heo24bdb8e2015-08-18 14:55:22 -0700669 atomic64_add(atomic64_read(&v.aux_cnt[i]) +
670 atomic64_read(&from->aux_cnt[i]),
Tejun Heoe6269c42015-08-18 14:55:21 -0700671 &to->aux_cnt[i]);
Tejun Heo16b3de62013-01-09 08:05:12 -0800672}
673
Tejun Heoae118892015-08-18 14:55:20 -0700674#ifdef CONFIG_BLK_DEV_THROTTLING
675extern bool blk_throtl_bio(struct request_queue *q, struct blkcg_gq *blkg,
676 struct bio *bio);
677#else
678static inline bool blk_throtl_bio(struct request_queue *q, struct blkcg_gq *blkg,
679 struct bio *bio) { return false; }
680#endif
681
682static inline bool blkcg_bio_issue_check(struct request_queue *q,
683 struct bio *bio)
684{
685 struct blkcg *blkcg;
686 struct blkcg_gq *blkg;
687 bool throtl = false;
688
689 rcu_read_lock();
690 blkcg = bio_blkcg(bio);
691
692 blkg = blkg_lookup(blkcg, q);
693 if (unlikely(!blkg)) {
694 spin_lock_irq(q->queue_lock);
695 blkg = blkg_lookup_create(blkcg, q);
696 if (IS_ERR(blkg))
697 blkg = NULL;
698 spin_unlock_irq(q->queue_lock);
699 }
700
701 throtl = blk_throtl_bio(q, blkg, bio);
702
703 rcu_read_unlock();
704 return !throtl;
705}
706
Tejun Heo36558c82012-04-16 13:57:24 -0700707#else /* CONFIG_BLK_CGROUP */
708
Tejun Heoefa7d1c2015-05-22 17:13:18 -0400709struct blkcg {
710};
Jens Axboe2f5ea472009-12-03 21:06:43 +0100711
Tejun Heof95a04a2012-04-16 13:57:26 -0700712struct blkg_policy_data {
713};
714
Arianna Avanzinie48453c2015-06-05 23:38:42 +0200715struct blkcg_policy_data {
716};
717
Tejun Heo3c798392012-04-16 13:57:25 -0700718struct blkcg_gq {
Jens Axboe2f5ea472009-12-03 21:06:43 +0100719};
720
Tejun Heo3c798392012-04-16 13:57:25 -0700721struct blkcg_policy {
Vivek Goyal3e252062009-12-04 10:36:42 -0500722};
723
Tejun Heo496d5e72015-05-22 17:13:21 -0400724#define blkcg_root_css ((struct cgroup_subsys_state *)ERR_PTR(-EINVAL))
725
Tejun Heofd383c22015-05-22 17:13:23 -0400726static inline struct cgroup_subsys_state *
727task_get_blkcg_css(struct task_struct *task)
728{
729 return NULL;
730}
731
Tejun Heoefa7d1c2015-05-22 17:13:18 -0400732#ifdef CONFIG_BLOCK
733
Tejun Heo3c798392012-04-16 13:57:25 -0700734static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, void *key) { return NULL; }
Tejun Heo5efd6112012-03-05 13:15:12 -0800735static inline int blkcg_init_queue(struct request_queue *q) { return 0; }
736static inline void blkcg_drain_queue(struct request_queue *q) { }
737static inline void blkcg_exit_queue(struct request_queue *q) { }
Jens Axboed5bf0292014-06-22 16:31:56 -0600738static inline int blkcg_policy_register(struct blkcg_policy *pol) { return 0; }
Tejun Heo3c798392012-04-16 13:57:25 -0700739static inline void blkcg_policy_unregister(struct blkcg_policy *pol) { }
Tejun Heoa2b16932012-04-13 13:11:33 -0700740static inline int blkcg_activate_policy(struct request_queue *q,
Tejun Heo3c798392012-04-16 13:57:25 -0700741 const struct blkcg_policy *pol) { return 0; }
Tejun Heoa2b16932012-04-13 13:11:33 -0700742static inline void blkcg_deactivate_policy(struct request_queue *q,
Tejun Heo3c798392012-04-16 13:57:25 -0700743 const struct blkcg_policy *pol) { }
Vivek Goyal3e252062009-12-04 10:36:42 -0500744
Tejun Heob1208b52012-06-04 20:40:57 -0700745static inline struct blkcg *bio_blkcg(struct bio *bio) { return NULL; }
Tejun Heoa0516612012-06-26 15:05:44 -0700746
Tejun Heof95a04a2012-04-16 13:57:26 -0700747static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
748 struct blkcg_policy *pol) { return NULL; }
749static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd) { return NULL; }
Tejun Heo3c798392012-04-16 13:57:25 -0700750static inline char *blkg_path(struct blkcg_gq *blkg) { return NULL; }
751static inline void blkg_get(struct blkcg_gq *blkg) { }
752static inline void blkg_put(struct blkcg_gq *blkg) { }
Vivek Goyalafc24d42010-04-26 19:27:56 +0200753
Tejun Heoa0516612012-06-26 15:05:44 -0700754static inline struct request_list *blk_get_rl(struct request_queue *q,
755 struct bio *bio) { return &q->root_rl; }
756static inline void blk_put_rl(struct request_list *rl) { }
757static inline void blk_rq_set_rl(struct request *rq, struct request_list *rl) { }
758static inline struct request_list *blk_rq_rl(struct request *rq) { return &rq->q->root_rl; }
759
Tejun Heoae118892015-08-18 14:55:20 -0700760static inline bool blkcg_bio_issue_check(struct request_queue *q,
761 struct bio *bio) { return true; }
762
Tejun Heoa0516612012-06-26 15:05:44 -0700763#define blk_queue_for_each_rl(rl, q) \
764 for ((rl) = &(q)->root_rl; (rl); (rl) = NULL)
765
Tejun Heoefa7d1c2015-05-22 17:13:18 -0400766#endif /* CONFIG_BLOCK */
Tejun Heo36558c82012-04-16 13:57:24 -0700767#endif /* CONFIG_BLK_CGROUP */
768#endif /* _BLK_CGROUP_H */